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);
540 if (fuse_isdeadfs(dvp))
543 /* FUSE expects sockets to be created with FUSE_MKNOD */
544 if (vap->va_type == VSOCK)
545 return fuse_internal_mknod(dvp, vpp, cnp, vap);
548 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
549 * writable mode makes sense, and we might as well include readability
554 bzero(&fdi, sizeof(fdi));
556 if (vap->va_type != VREG)
559 if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
560 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
561 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
564 /* Use FUSE_CREATE */
566 fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
567 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
570 foi->flags = O_CREAT | flags;
571 memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
573 ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
576 err = fdisp_wait_answ(fdip);
579 if (err == ENOSYS && op == FUSE_CREATE) {
580 fsess_set_notimpl(mp, FUSE_CREATE);
581 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
582 parentnid, td, cred, mode, &op);
583 err = fdisp_wait_answ(fdip);
591 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
595 if (op == FUSE_CREATE) {
596 foo = (struct fuse_open_out*)(feo + 1);
598 /* Issue a separate FUSE_OPEN */
600 fdisp_init(fdip2, sizeof(*foi));
601 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
606 err = fdisp_wait_answ(fdip2);
611 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
613 struct fuse_release_in *fri;
614 uint64_t nodeid = feo->nodeid;
615 uint64_t fh_id = foo->fh;
617 fdisp_init(fdip, sizeof(*fri));
618 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
622 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
623 fuse_insert_message(fdip->tick, false);
626 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
627 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
628 feo->attr_valid_nsec, NULL);
630 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
631 fuse_vnode_open(*vpp, foo->open_flags, td);
633 * Purge the parent's attribute cache because the daemon should've
634 * updated its mtime and ctime
636 fuse_vnode_clear_attr_cache(dvp);
637 cache_purge_negative(dvp);
641 fdisp_destroy(fdip2);
647 struct vnop_fdatasync_args {
648 struct vop_generic_args a_gen;
650 struct thread * a_td;
654 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
656 struct vnode *vp = ap->a_vp;
657 struct thread *td = ap->a_td;
658 int waitfor = MNT_WAIT;
662 if (fuse_isdeadfs(vp)) {
665 if ((err = vop_stdfdatasync_buf(ap)))
668 return fuse_internal_fsync(vp, td, waitfor, true);
672 struct vnop_fsync_args {
673 struct vop_generic_args a_gen;
676 struct thread * a_td;
680 fuse_vnop_fsync(struct vop_fsync_args *ap)
682 struct vnode *vp = ap->a_vp;
683 struct thread *td = ap->a_td;
684 int waitfor = ap->a_waitfor;
687 if (fuse_isdeadfs(vp)) {
690 if ((err = vop_stdfsync(ap)))
693 return fuse_internal_fsync(vp, td, waitfor, false);
697 struct vnop_getattr_args {
700 struct ucred *a_cred;
705 fuse_vnop_getattr(struct vop_getattr_args *ap)
707 struct vnode *vp = ap->a_vp;
708 struct vattr *vap = ap->a_vap;
709 struct ucred *cred = ap->a_cred;
710 struct thread *td = curthread;
715 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
717 /* Note that we are not bailing out on a dead file system just yet. */
719 if (!(dataflags & FSESS_INITED)) {
720 if (!vnode_isvroot(vp)) {
721 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
728 err = fuse_internal_getattr(vp, vap, cred, td);
729 if (err == ENOTCONN && vnode_isvroot(vp)) {
730 /* see comment in fuse_vfsop_statfs() */
737 bzero(vap, sizeof(*vap));
738 vap->va_type = vnode_vtype(vp);
744 struct vnop_inactive_args {
750 fuse_vnop_inactive(struct vop_inactive_args *ap)
752 struct vnode *vp = ap->a_vp;
753 struct thread *td = ap->a_td;
755 struct fuse_vnode_data *fvdat = VTOFUD(vp);
756 struct fuse_filehandle *fufh, *fufh_tmp;
760 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
761 if (need_flush && vp->v_type == VREG) {
762 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
763 fuse_vnode_savesize(vp, NULL, 0);
765 if ((fvdat->flag & FN_REVOKED) != 0)
766 fuse_io_invalbuf(vp, td);
768 fuse_io_flushbuf(vp, MNT_WAIT, td);
771 fuse_filehandle_close(vp, fufh, td, NULL);
774 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
781 struct vnop_link_args {
782 struct vnode *a_tdvp;
784 struct componentname *a_cnp;
788 fuse_vnop_link(struct vop_link_args *ap)
790 struct vnode *vp = ap->a_vp;
791 struct vnode *tdvp = ap->a_tdvp;
792 struct componentname *cnp = ap->a_cnp;
794 struct vattr *vap = VTOVA(vp);
796 struct fuse_dispatcher fdi;
797 struct fuse_entry_out *feo;
798 struct fuse_link_in fli;
802 if (fuse_isdeadfs(vp)) {
805 if (vnode_mount(tdvp) != vnode_mount(vp)) {
810 * This is a seatbelt check to protect naive userspace filesystems from
811 * themselves and the limitations of the FUSE IPC protocol. If a
812 * filesystem does not allow attribute caching, assume it is capable of
813 * validating that nlink does not overflow.
815 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
817 fli.oldnodeid = VTOI(vp);
820 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
821 FUSE_LINK, &fli, sizeof(fli), &fdi);
822 if ((err = fdisp_wait_answ(&fdi))) {
827 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
830 * Purge the parent's attribute cache because the daemon
831 * should've updated its mtime and ctime
833 fuse_vnode_clear_attr_cache(tdvp);
834 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
835 feo->attr_valid_nsec, NULL);
842 struct fuse_lookup_alloc_arg {
843 struct fuse_entry_out *feo;
844 struct componentname *cnp;
849 /* Callback for vn_get_ino */
851 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
853 struct fuse_lookup_alloc_arg *flaa = arg;
855 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
859 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
860 "int", "struct timespec*", "struct timespec*");
862 struct vnop_lookup_args {
863 struct vnodeop_desc *a_desc;
865 struct vnode **a_vpp;
866 struct componentname *a_cnp;
870 fuse_vnop_lookup(struct vop_lookup_args *ap)
872 struct vnode *dvp = ap->a_dvp;
873 struct vnode **vpp = ap->a_vpp;
874 struct componentname *cnp = ap->a_cnp;
875 struct thread *td = cnp->cn_thread;
876 struct ucred *cred = cnp->cn_cred;
878 int nameiop = cnp->cn_nameiop;
879 int flags = cnp->cn_flags;
880 int wantparent = flags & (LOCKPARENT | WANTPARENT);
881 int islastcn = flags & ISLASTCN;
882 struct mount *mp = vnode_mount(dvp);
886 struct vnode *vp = NULL;
888 struct fuse_dispatcher fdi;
889 bool did_lookup = false;
890 struct fuse_entry_out *feo = NULL;
891 enum vtype vtyp; /* vnode type of target */
892 off_t filesize; /* filesize of target */
896 if (fuse_isdeadfs(dvp)) {
900 if (!vnode_isdir(dvp))
903 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
906 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
909 if (flags & ISDOTDOT) {
910 nid = VTOFUD(dvp)->parent_nid;
913 /* .. is obviously a directory */
916 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
918 /* . is obviously a directory */
922 struct timespec now, timeout;
924 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
926 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
928 case -1: /* positive match */
929 if (timespeccmp(&timeout, &now, >)) {
930 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
933 atomic_add_acq_long(&fuse_lookup_cache_misses,
945 case 0: /* no match in cache */
946 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
949 case ENOENT: /* negative match */
951 if (timespeccmp(&timeout, &now, <=)) {
953 cache_purge_negative(dvp);
962 fdisp_init(&fdi, cnp->cn_namelen + 1);
963 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
965 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
966 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
967 lookup_err = fdisp_wait_answ(&fdi);
971 /* lookup call succeeded */
972 nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
973 feo = (struct fuse_entry_out *)fdi.answ;
975 /* zero nodeid means ENOENT and cache it */
976 struct timespec timeout;
978 fdi.answ_stat = ENOENT;
980 if (cnp->cn_flags & MAKEENTRY) {
981 fuse_validity_2_timespec(feo, &timeout);
982 cache_enter_time(dvp, *vpp, cnp,
985 } else if (nid == FUSE_ROOT_ID) {
988 vtyp = IFTOVT(feo->attr.mode);
989 filesize = feo->attr.size;
991 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
996 /* lookup_err, if non-zero, must be ENOENT at this point */
999 /* Entry not found */
1000 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1001 err = fuse_internal_access(dvp, VWRITE, td, cred);
1004 * Set the SAVENAME flag to hold onto the
1005 * pathname for use later in VOP_CREATE or
1008 cnp->cn_flags |= SAVENAME;
1016 /* Entry was found */
1017 if (flags & ISDOTDOT) {
1018 struct fuse_lookup_alloc_arg flaa;
1024 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1027 } else if (nid == VTOI(dvp)) {
1031 struct fuse_vnode_data *fvdat;
1033 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1039 fuse_vnode_setparent(vp, dvp);
1042 * In the case where we are looking up a FUSE node
1043 * represented by an existing cached vnode, and the
1044 * true size reported by FUSE_LOOKUP doesn't match
1045 * the vnode's cached size, fix the vnode cache to
1046 * match the real object size.
1049 * * following attribute cache expiration, or
1050 * * due a bug in the daemon, or
1051 * * the first time that we looked up the file.
1054 if (vnode_isreg(vp) &&
1055 filesize != fvdat->cached_attrs.va_size) {
1057 * The FN_SIZECHANGE flag reflects a dirty
1058 * append. If userspace lets us know our cache
1059 * is invalid, that write was lost. (Dirty
1060 * writes that do not cause append are also
1061 * lost, but we don't detect them here.)
1063 * XXX: Maybe disable WB caching on this mount.
1065 if (fvdat->flag & FN_SIZECHANGE)
1066 printf("%s: WB cache incoherent on "
1068 vnode_mount(vp)->mnt_stat.f_mntonname);
1070 (void)fuse_vnode_setsize(vp, cred, filesize);
1071 fvdat->flag &= ~FN_SIZECHANGE;
1075 fuse_internal_cache_attrs(*vpp, &feo->attr,
1076 feo->attr_valid, feo->attr_valid_nsec, NULL);
1078 if ((nameiop == DELETE || nameiop == RENAME) &&
1081 struct vattr dvattr;
1083 err = fuse_internal_access(dvp, VWRITE, td,
1088 * if the parent's sticky bit is set, check
1089 * whether we're allowed to remove the file.
1090 * Need to figure out the vnode locking to make
1093 fuse_internal_getattr(dvp, &dvattr, cred, td);
1094 if ((dvattr.va_mode & S_ISTXT) &&
1095 fuse_internal_access(dvp, VADMIN, td,
1097 fuse_internal_access(*vpp, VADMIN, td,
1105 (nameiop == DELETE) ||
1106 (nameiop == RENAME && wantparent))) {
1107 cnp->cn_flags |= SAVENAME;
1114 if (vp != NULL && dvp != vp)
1116 else if (vp != NULL)
1121 fdisp_destroy(&fdi);
1127 struct vnop_mkdir_args {
1128 struct vnode *a_dvp;
1129 struct vnode **a_vpp;
1130 struct componentname *a_cnp;
1131 struct vattr *a_vap;
1135 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1137 struct vnode *dvp = ap->a_dvp;
1138 struct vnode **vpp = ap->a_vpp;
1139 struct componentname *cnp = ap->a_cnp;
1140 struct vattr *vap = ap->a_vap;
1142 struct fuse_mkdir_in fmdi;
1144 if (fuse_isdeadfs(dvp)) {
1147 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1149 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1150 sizeof(fmdi), VDIR));
1154 struct vnop_mknod_args {
1155 struct vnode *a_dvp;
1156 struct vnode **a_vpp;
1157 struct componentname *a_cnp;
1158 struct vattr *a_vap;
1162 fuse_vnop_mknod(struct vop_mknod_args *ap)
1165 struct vnode *dvp = ap->a_dvp;
1166 struct vnode **vpp = ap->a_vpp;
1167 struct componentname *cnp = ap->a_cnp;
1168 struct vattr *vap = ap->a_vap;
1170 if (fuse_isdeadfs(dvp))
1173 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1177 struct vnop_open_args {
1180 struct ucred *a_cred;
1181 struct thread *a_td;
1182 int a_fdidx; / struct file *a_fp;
1186 fuse_vnop_open(struct vop_open_args *ap)
1188 struct vnode *vp = ap->a_vp;
1189 int a_mode = ap->a_mode;
1190 struct thread *td = ap->a_td;
1191 struct ucred *cred = ap->a_cred;
1192 pid_t pid = td->td_proc->p_pid;
1193 struct fuse_vnode_data *fvdat;
1195 if (fuse_isdeadfs(vp))
1197 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1198 return (EOPNOTSUPP);
1199 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1204 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1205 fuse_vnode_open(vp, 0, td);
1209 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1213 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1216 switch (ap->a_name) {
1217 case _PC_FILESIZEBITS:
1221 *ap->a_retval = NAME_MAX;
1224 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1226 case _PC_SYMLINK_MAX:
1227 *ap->a_retval = MAXPATHLEN;
1233 return (vop_stdpathconf(ap));
1238 struct vnop_read_args {
1242 struct ucred *a_cred;
1246 fuse_vnop_read(struct vop_read_args *ap)
1248 struct vnode *vp = ap->a_vp;
1249 struct uio *uio = ap->a_uio;
1250 int ioflag = ap->a_ioflag;
1251 struct ucred *cred = ap->a_cred;
1252 pid_t pid = curthread->td_proc->p_pid;
1254 if (fuse_isdeadfs(vp)) {
1258 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1259 ioflag |= IO_DIRECT;
1262 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1266 struct vnop_readdir_args {
1269 struct ucred *a_cred;
1276 fuse_vnop_readdir(struct vop_readdir_args *ap)
1278 struct vnode *vp = ap->a_vp;
1279 struct uio *uio = ap->a_uio;
1280 struct ucred *cred = ap->a_cred;
1281 struct fuse_filehandle *fufh = NULL;
1282 struct fuse_iov cookediov;
1284 pid_t pid = curthread->td_proc->p_pid;
1286 if (fuse_isdeadfs(vp)) {
1289 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1290 (uio_resid(uio) < sizeof(struct dirent))) {
1294 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1297 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1298 fiov_init(&cookediov, DIRCOOKEDSIZE);
1300 err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1302 fiov_teardown(&cookediov);
1308 struct vnop_readlink_args {
1311 struct ucred *a_cred;
1315 fuse_vnop_readlink(struct vop_readlink_args *ap)
1317 struct vnode *vp = ap->a_vp;
1318 struct uio *uio = ap->a_uio;
1319 struct ucred *cred = ap->a_cred;
1321 struct fuse_dispatcher fdi;
1324 if (fuse_isdeadfs(vp)) {
1327 if (!vnode_islnk(vp)) {
1330 fdisp_init(&fdi, 0);
1331 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1335 if (((char *)fdi.answ)[0] == '/' &&
1336 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1337 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1339 err = uiomove(mpth, strlen(mpth), uio);
1342 err = uiomove(fdi.answ, fdi.iosize, uio);
1345 fdisp_destroy(&fdi);
1350 struct vnop_reclaim_args {
1352 struct thread *a_td;
1356 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1358 struct vnode *vp = ap->a_vp;
1359 struct thread *td = ap->a_td;
1360 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1361 struct fuse_filehandle *fufh, *fufh_tmp;
1364 panic("FUSE: no vnode data during recycling");
1366 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1367 printf("FUSE: vnode being reclaimed with open fufh "
1368 "(type=%#x)", fufh->fufh_type);
1369 fuse_filehandle_close(vp, fufh, td, NULL);
1372 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1373 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1376 fuse_vnode_setparent(vp, NULL);
1378 vfs_hash_remove(vp);
1379 vnode_destroy_vobject(vp);
1380 fuse_vnode_destroy(vp);
1386 struct vnop_remove_args {
1387 struct vnode *a_dvp;
1389 struct componentname *a_cnp;
1393 fuse_vnop_remove(struct vop_remove_args *ap)
1395 struct vnode *dvp = ap->a_dvp;
1396 struct vnode *vp = ap->a_vp;
1397 struct componentname *cnp = ap->a_cnp;
1401 if (fuse_isdeadfs(vp)) {
1404 if (vnode_isdir(vp)) {
1409 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1412 fuse_internal_vnode_disappear(vp);
1414 * Purge the parent's attribute cache because the daemon
1415 * should've updated its mtime and ctime
1417 fuse_vnode_clear_attr_cache(dvp);
1423 struct vnop_rename_args {
1424 struct vnode *a_fdvp;
1425 struct vnode *a_fvp;
1426 struct componentname *a_fcnp;
1427 struct vnode *a_tdvp;
1428 struct vnode *a_tvp;
1429 struct componentname *a_tcnp;
1433 fuse_vnop_rename(struct vop_rename_args *ap)
1435 struct vnode *fdvp = ap->a_fdvp;
1436 struct vnode *fvp = ap->a_fvp;
1437 struct componentname *fcnp = ap->a_fcnp;
1438 struct vnode *tdvp = ap->a_tdvp;
1439 struct vnode *tvp = ap->a_tvp;
1440 struct componentname *tcnp = ap->a_tcnp;
1441 struct fuse_data *data;
1442 bool newparent = fdvp != tdvp;
1443 bool isdir = fvp->v_type == VDIR;
1446 if (fuse_isdeadfs(fdvp)) {
1449 if (fvp->v_mount != tdvp->v_mount ||
1450 (tvp && fvp->v_mount != tvp->v_mount)) {
1451 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1458 * FUSE library is expected to check if target directory is not
1459 * under the source directory in the file system tree.
1460 * Linux performs this check at VFS level.
1463 * If source is a directory, and it will get a new parent, user must
1464 * have write permission to it, so ".." can be modified.
1466 data = fuse_get_mpdata(vnode_mount(tdvp));
1467 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1468 err = fuse_internal_access(fvp, VWRITE,
1469 tcnp->cn_thread, tcnp->cn_cred);
1473 sx_xlock(&data->rename_lock);
1474 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1477 fuse_vnode_setparent(fvp, tdvp);
1479 fuse_vnode_setparent(tvp, NULL);
1481 sx_unlock(&data->rename_lock);
1483 if (tvp != NULL && tvp != fvp) {
1486 if (vnode_isdir(fvp)) {
1487 if ((tvp != NULL) && vnode_isdir(tvp)) {
1508 struct vnop_rmdir_args {
1509 struct vnode *a_dvp;
1511 struct componentname *a_cnp;
1515 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1517 struct vnode *dvp = ap->a_dvp;
1518 struct vnode *vp = ap->a_vp;
1522 if (fuse_isdeadfs(vp)) {
1525 if (VTOFUD(vp) == VTOFUD(dvp)) {
1528 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1531 fuse_internal_vnode_disappear(vp);
1533 * Purge the parent's attribute cache because the daemon
1534 * should've updated its mtime and ctime
1536 fuse_vnode_clear_attr_cache(dvp);
1542 struct vnop_setattr_args {
1544 struct vattr *a_vap;
1545 struct ucred *a_cred;
1546 struct thread *a_td;
1550 fuse_vnop_setattr(struct vop_setattr_args *ap)
1552 struct vnode *vp = ap->a_vp;
1553 struct vattr *vap = ap->a_vap;
1554 struct ucred *cred = ap->a_cred;
1555 struct thread *td = curthread;
1557 struct fuse_data *data;
1558 struct vattr old_va;
1561 accmode_t accmode = 0;
1563 bool drop_suid = false;
1566 mp = vnode_mount(vp);
1567 data = fuse_get_mpdata(mp);
1568 dataflags = data->dataflags;
1569 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1570 if (cred->cr_ngroups > 0)
1571 cr_gid = cred->cr_groups[0];
1575 if (fuse_isdeadfs(vp)) {
1579 if (vap->va_uid != (uid_t)VNOVAL) {
1581 /* Only root may change a file's owner */
1582 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1584 /* As a special case, allow the null chown */
1585 err2 = fuse_internal_getattr(vp, &old_va, cred,
1589 if (vap->va_uid != old_va.va_uid)
1599 if (vap->va_gid != (gid_t)VNOVAL) {
1600 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1602 if (checkperm && !groupmember(vap->va_gid, cred))
1605 * Non-root users may only chgrp to one of their own
1608 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1610 /* As a special case, allow the null chgrp */
1611 err2 = fuse_internal_getattr(vp, &old_va, cred,
1615 if (vap->va_gid != old_va.va_gid)
1623 if (vap->va_size != VNOVAL) {
1624 switch (vp->v_type) {
1629 if (vfs_isrdonly(mp))
1634 * According to POSIX, the result is unspecified
1635 * for file types other than regular files,
1636 * directories and shared memory objects. We
1637 * don't support shared memory objects in the file
1638 * system, and have dubious support for truncating
1639 * symlinks. Just ignore the request in other cases.
1643 /* Don't set accmode. Permission to trunc is checked upstack */
1645 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1646 if (vap->va_vaflags & VA_UTIMES_NULL)
1652 if (vap->va_mode != (mode_t)VNOVAL)
1653 vap->va_mode &= ~(S_ISUID | S_ISGID);
1655 err = fuse_internal_getattr(vp, &old_va, cred, td);
1658 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1661 if (vap->va_mode != (mode_t)VNOVAL) {
1662 /* Only root may set the sticky bit on non-directories */
1663 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1664 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1666 if (checkperm && (vap->va_mode & S_ISGID)) {
1667 err = fuse_internal_getattr(vp, &old_va, cred, td);
1670 if (!groupmember(old_va.va_gid, cred)) {
1671 err = priv_check_cred(cred, PRIV_VFS_SETGID);
1679 if (vfs_isrdonly(mp))
1682 err = fuse_internal_access(vp, accmode, td, cred);
1686 return fuse_internal_setattr(vp, vap, td, cred);
1690 struct vnop_strategy_args {
1696 fuse_vnop_strategy(struct vop_strategy_args *ap)
1698 struct vnode *vp = ap->a_vp;
1699 struct buf *bp = ap->a_bp;
1701 if (!vp || fuse_isdeadfs(vp)) {
1702 bp->b_ioflags |= BIO_ERROR;
1703 bp->b_error = ENXIO;
1709 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1710 * fuse_io_strategy sets bp's error fields
1712 (void)fuse_io_strategy(vp, bp);
1719 struct vnop_symlink_args {
1720 struct vnode *a_dvp;
1721 struct vnode **a_vpp;
1722 struct componentname *a_cnp;
1723 struct vattr *a_vap;
1728 fuse_vnop_symlink(struct vop_symlink_args *ap)
1730 struct vnode *dvp = ap->a_dvp;
1731 struct vnode **vpp = ap->a_vpp;
1732 struct componentname *cnp = ap->a_cnp;
1733 const char *target = ap->a_target;
1735 struct fuse_dispatcher fdi;
1740 if (fuse_isdeadfs(dvp)) {
1744 * Unlike the other creator type calls, here we have to create a message
1745 * where the name of the new entry comes first, and the data describing
1746 * the entry comes second.
1747 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1748 * but put together the message manually and just call the core part.
1751 len = strlen(target) + 1;
1752 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1753 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1755 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1756 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1757 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1759 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1760 fdisp_destroy(&fdi);
1765 struct vnop_write_args {
1769 struct ucred *a_cred;
1773 fuse_vnop_write(struct vop_write_args *ap)
1775 struct vnode *vp = ap->a_vp;
1776 struct uio *uio = ap->a_uio;
1777 int ioflag = ap->a_ioflag;
1778 struct ucred *cred = ap->a_cred;
1779 pid_t pid = curthread->td_proc->p_pid;
1781 if (fuse_isdeadfs(vp)) {
1785 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1786 ioflag |= IO_DIRECT;
1789 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1792 SDT_PROBE_DEFINE1(fusefs, , vnops, vnop_getpages_error, "int");
1794 struct vnop_getpages_args {
1802 fuse_vnop_getpages(struct vop_getpages_args *ap)
1804 int i, error, nextoff, size, toff, count, npages;
1813 pid_t pid = curthread->td_proc->p_pid;
1816 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1817 td = curthread; /* XXX */
1818 cred = curthread->td_ucred; /* XXX */
1820 npages = ap->a_count;
1822 if (!fsess_opt_mmap(vnode_mount(vp))) {
1823 SDT_PROBE2(fusefs, , vnops, trace, 1,
1824 "called on non-cacheable vnode??\n");
1825 return (VM_PAGER_ERROR);
1829 * If the last page is partially valid, just return it and allow
1830 * the pager to zero-out the blanks. Partially valid pages can
1831 * only occur at the file EOF.
1833 * XXXGL: is that true for FUSE, which is a local filesystem,
1834 * but still somewhat disconnected from the kernel?
1836 VM_OBJECT_WLOCK(vp->v_object);
1837 if (pages[npages - 1]->valid != 0 && --npages == 0)
1839 VM_OBJECT_WUNLOCK(vp->v_object);
1842 * We use only the kva address for the buffer, but this is extremely
1843 * convenient and fast.
1845 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1847 kva = (vm_offset_t)bp->b_data;
1848 pmap_qenter(kva, pages, npages);
1849 VM_CNT_INC(v_vnodein);
1850 VM_CNT_ADD(v_vnodepgsin, npages);
1852 count = npages << PAGE_SHIFT;
1853 iov.iov_base = (caddr_t)kva;
1854 iov.iov_len = count;
1857 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1858 uio.uio_resid = count;
1859 uio.uio_segflg = UIO_SYSSPACE;
1860 uio.uio_rw = UIO_READ;
1863 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1864 pmap_qremove(kva, npages);
1866 uma_zfree(fuse_pbuf_zone, bp);
1868 if (error && (uio.uio_resid == count)) {
1869 SDT_PROBE1(fusefs, , vnops, vnop_getpages_error, error);
1870 return VM_PAGER_ERROR;
1873 * Calculate the number of bytes read and validate only that number
1874 * of bytes. Note that due to pending writes, size may be 0. This
1875 * does not mean that the remaining data is invalid!
1878 size = count - uio.uio_resid;
1879 VM_OBJECT_WLOCK(vp->v_object);
1880 fuse_vm_page_lock_queues();
1881 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1884 nextoff = toff + PAGE_SIZE;
1887 if (nextoff <= size) {
1889 * Read operation filled an entire page
1891 m->valid = VM_PAGE_BITS_ALL;
1892 KASSERT(m->dirty == 0,
1893 ("fuse_getpages: page %p is dirty", m));
1894 } else if (size > toff) {
1896 * Read operation filled a partial page.
1899 vm_page_set_valid_range(m, 0, size - toff);
1900 KASSERT(m->dirty == 0,
1901 ("fuse_getpages: page %p is dirty", m));
1904 * Read operation was short. If no error occurred
1905 * we may have hit a zero-fill section. We simply
1906 * leave valid set to 0.
1911 fuse_vm_page_unlock_queues();
1913 VM_OBJECT_WUNLOCK(vp->v_object);
1918 return (VM_PAGER_OK);
1922 struct vnop_putpages_args {
1928 vm_ooffset_t a_offset;
1932 fuse_vnop_putpages(struct vop_putpages_args *ap)
1938 int i, error, npages, count;
1946 pid_t pid = curthread->td_proc->p_pid;
1949 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1950 fsize = vp->v_object->un_pager.vnp.vnp_size;
1951 td = curthread; /* XXX */
1952 cred = curthread->td_ucred; /* XXX */
1954 count = ap->a_count;
1955 rtvals = ap->a_rtvals;
1956 npages = btoc(count);
1957 offset = IDX_TO_OFF(pages[0]->pindex);
1959 if (!fsess_opt_mmap(vnode_mount(vp))) {
1960 SDT_PROBE2(fusefs, , vnops, trace, 1,
1961 "called on non-cacheable vnode??\n");
1963 for (i = 0; i < npages; i++)
1964 rtvals[i] = VM_PAGER_AGAIN;
1967 * When putting pages, do not extend file past EOF.
1970 if (offset + count > fsize) {
1971 count = fsize - offset;
1976 * We use only the kva address for the buffer, but this is extremely
1977 * convenient and fast.
1979 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1981 kva = (vm_offset_t)bp->b_data;
1982 pmap_qenter(kva, pages, npages);
1983 VM_CNT_INC(v_vnodeout);
1984 VM_CNT_ADD(v_vnodepgsout, count);
1986 iov.iov_base = (caddr_t)kva;
1987 iov.iov_len = count;
1990 uio.uio_offset = offset;
1991 uio.uio_resid = count;
1992 uio.uio_segflg = UIO_SYSSPACE;
1993 uio.uio_rw = UIO_WRITE;
1996 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1998 pmap_qremove(kva, npages);
1999 uma_zfree(fuse_pbuf_zone, bp);
2002 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
2004 for (i = 0; i < nwritten; i++) {
2005 rtvals[i] = VM_PAGER_OK;
2006 VM_OBJECT_WLOCK(pages[i]->object);
2007 vm_page_undirty(pages[i]);
2008 VM_OBJECT_WUNLOCK(pages[i]->object);
2014 static const char extattr_namespace_separator = '.';
2017 struct vop_getextattr_args {
2018 struct vop_generic_args a_gen;
2020 int a_attrnamespace;
2024 struct ucred *a_cred;
2025 struct thread *a_td;
2029 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2031 struct vnode *vp = ap->a_vp;
2032 struct uio *uio = ap->a_uio;
2033 struct fuse_dispatcher fdi;
2034 struct fuse_getxattr_in *get_xattr_in;
2035 struct fuse_getxattr_out *get_xattr_out;
2036 struct mount *mp = vnode_mount(vp);
2037 struct thread *td = ap->a_td;
2038 struct ucred *cred = ap->a_cred;
2044 if (fuse_isdeadfs(vp))
2047 if (!fsess_isimpl(mp, FUSE_GETXATTR))
2050 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2054 /* Default to looking for user attributes. */
2055 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2056 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2058 prefix = EXTATTR_NAMESPACE_USER_STRING;
2060 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2061 strlen(ap->a_name) + 1;
2063 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2064 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2066 get_xattr_in = fdi.indata;
2068 * Check to see whether we're querying the available size or
2069 * issuing the actual request. If we pass in 0, we get back struct
2070 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2071 * that much data, without the struct fuse_getxattr_out header.
2074 get_xattr_in->size = 0;
2076 get_xattr_in->size = uio->uio_resid;
2078 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2079 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2082 err = fdisp_wait_answ(&fdi);
2084 if (err == ENOSYS) {
2085 fsess_set_notimpl(mp, FUSE_GETXATTR);
2091 get_xattr_out = fdi.answ;
2093 if (ap->a_size != NULL)
2094 *ap->a_size = get_xattr_out->size;
2097 err = uiomove(fdi.answ, fdi.iosize, uio);
2100 fdisp_destroy(&fdi);
2105 struct vop_setextattr_args {
2106 struct vop_generic_args a_gen;
2108 int a_attrnamespace;
2111 struct ucred *a_cred;
2112 struct thread *a_td;
2116 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2118 struct vnode *vp = ap->a_vp;
2119 struct uio *uio = ap->a_uio;
2120 struct fuse_dispatcher fdi;
2121 struct fuse_setxattr_in *set_xattr_in;
2122 struct mount *mp = vnode_mount(vp);
2123 struct thread *td = ap->a_td;
2124 struct ucred *cred = ap->a_cred;
2130 if (fuse_isdeadfs(vp))
2133 if (!fsess_isimpl(mp, FUSE_SETXATTR))
2136 if (vfs_isrdonly(mp))
2139 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2140 if (ap->a_uio == NULL) {
2142 * If we got here as fallback from VOP_DELETEEXTATTR, then
2143 * return EOPNOTSUPP.
2145 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2146 return (EOPNOTSUPP);
2151 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2156 /* Default to looking for user attributes. */
2157 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2158 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2160 prefix = EXTATTR_NAMESPACE_USER_STRING;
2162 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2163 strlen(ap->a_name) + 1;
2165 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2166 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2168 set_xattr_in = fdi.indata;
2169 set_xattr_in->size = uio->uio_resid;
2171 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2172 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2175 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2176 uio->uio_resid, uio);
2181 err = fdisp_wait_answ(&fdi);
2183 if (err == ENOSYS) {
2184 fsess_set_notimpl(mp, FUSE_SETXATTR);
2187 if (err == ERESTART) {
2188 /* Can't restart after calling uiomove */
2193 fdisp_destroy(&fdi);
2198 * The Linux / FUSE extended attribute list is simply a collection of
2199 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2200 * byte length followed by a non-NUL terminated string. So, this allows
2201 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2202 * Linux attribute names are reported with the namespace as a prefix (e.g.
2203 * "user.attribute_name"), but in FreeBSD they are reported without the
2204 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2206 * user.attr_name1\0user.attr_name2\0
2210 * <num>attr_name1<num>attr_name2
2212 * Where "<num>" is a single byte number of characters in the attribute name.
2215 * prefix - exattr namespace prefix string
2216 * list, list_len - input list with namespace prefixes
2217 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2220 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2221 char *bsd_list, int *bsd_list_len)
2223 int len, pos, dist_to_next, prefix_len;
2227 prefix_len = strlen(prefix);
2229 while (pos < list_len && list[pos] != '\0') {
2230 dist_to_next = strlen(&list[pos]) + 1;
2231 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2232 list[pos + prefix_len] == extattr_namespace_separator) {
2233 len = dist_to_next -
2234 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2235 if (len >= EXTATTR_MAXNAMELEN)
2236 return (ENAMETOOLONG);
2238 bsd_list[*bsd_list_len] = len;
2239 memcpy(&bsd_list[*bsd_list_len + 1],
2240 &list[pos + prefix_len +
2241 sizeof(extattr_namespace_separator)], len);
2243 *bsd_list_len += len + 1;
2246 pos += dist_to_next;
2253 struct vop_listextattr_args {
2254 struct vop_generic_args a_gen;
2256 int a_attrnamespace;
2259 struct ucred *a_cred;
2260 struct thread *a_td;
2264 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2266 struct vnode *vp = ap->a_vp;
2267 struct uio *uio = ap->a_uio;
2268 struct fuse_dispatcher fdi;
2269 struct fuse_listxattr_in *list_xattr_in;
2270 struct fuse_listxattr_out *list_xattr_out;
2271 struct mount *mp = vnode_mount(vp);
2272 struct thread *td = ap->a_td;
2273 struct ucred *cred = ap->a_cred;
2277 char *bsd_list = NULL;
2283 if (fuse_isdeadfs(vp))
2286 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2289 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2294 * Add space for a NUL and the period separator if enabled.
2295 * Default to looking for user attributes.
2297 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2298 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2300 prefix = EXTATTR_NAMESPACE_USER_STRING;
2302 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2304 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2305 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2308 * Retrieve Linux / FUSE compatible list size.
2310 list_xattr_in = fdi.indata;
2311 list_xattr_in->size = 0;
2312 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2313 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2315 err = fdisp_wait_answ(&fdi);
2317 if (err == ENOSYS) {
2318 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2324 list_xattr_out = fdi.answ;
2325 linux_list_len = list_xattr_out->size;
2326 if (linux_list_len == 0) {
2327 if (ap->a_size != NULL)
2328 *ap->a_size = linux_list_len;
2333 * Retrieve Linux / FUSE compatible list values.
2335 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2336 list_xattr_in = fdi.indata;
2337 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2338 list_xattr_in->flags = 0;
2339 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2340 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2342 err = fdisp_wait_answ(&fdi);
2346 linux_list = fdi.answ;
2347 linux_list_len = fdi.iosize;
2350 * Retrieve the BSD compatible list values.
2351 * The Linux / FUSE attribute list format isn't the same
2352 * as FreeBSD's format. So we need to transform it into
2353 * FreeBSD's format before giving it to the user.
2355 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2356 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2357 bsd_list, &bsd_list_len);
2361 if (ap->a_size != NULL)
2362 *ap->a_size = bsd_list_len;
2365 err = uiomove(bsd_list, bsd_list_len, uio);
2368 free(bsd_list, M_TEMP);
2369 fdisp_destroy(&fdi);
2374 struct vop_deleteextattr_args {
2375 struct vop_generic_args a_gen;
2377 int a_attrnamespace;
2379 struct ucred *a_cred;
2380 struct thread *a_td;
2384 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2386 struct vnode *vp = ap->a_vp;
2387 struct fuse_dispatcher fdi;
2388 struct mount *mp = vnode_mount(vp);
2389 struct thread *td = ap->a_td;
2390 struct ucred *cred = ap->a_cred;
2396 if (fuse_isdeadfs(vp))
2399 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2402 if (vfs_isrdonly(mp))
2405 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2410 /* Default to looking for user attributes. */
2411 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2412 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2414 prefix = EXTATTR_NAMESPACE_USER_STRING;
2416 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2417 strlen(ap->a_name) + 1;
2419 fdisp_init(&fdi, len);
2420 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2422 attr_str = fdi.indata;
2423 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2426 err = fdisp_wait_answ(&fdi);
2427 if (err == ENOSYS) {
2428 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2432 fdisp_destroy(&fdi);
2437 struct vnop_print_args {
2442 fuse_vnop_print(struct vop_print_args *ap)
2444 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2446 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2447 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2448 (uintmax_t)fvdat->nlookup,