2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following disclaimer
15 * in the documentation and/or other materials provided with the
17 * * Neither the name of Google Inc. nor the names of its
18 * contributors may be used to endorse or promote products derived from
19 * this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
45 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
61 #include <sys/param.h>
62 #include <sys/module.h>
63 #include <sys/systm.h>
64 #include <sys/errno.h>
65 #include <sys/kernel.h>
68 #include <sys/malloc.h>
69 #include <sys/queue.h>
71 #include <sys/rwlock.h>
74 #include <sys/mount.h>
75 #include <sys/vnode.h>
76 #include <sys/namei.h>
77 #include <sys/extattr.h>
79 #include <sys/unistd.h>
80 #include <sys/filedesc.h>
82 #include <sys/fcntl.h>
83 #include <sys/dirent.h>
86 #include <sys/sysctl.h>
87 #include <sys/vmmeter.h>
90 #include <vm/vm_extern.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_param.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/vm_object.h>
101 #include "fuse_file.h"
102 #include "fuse_internal.h"
103 #include "fuse_ipc.h"
104 #include "fuse_node.h"
107 #include <sys/priv.h>
109 /* Maximum number of hardlinks to a single FUSE file */
110 #define FUSE_LINK_MAX UINT32_MAX
112 SDT_PROVIDER_DECLARE(fuse);
115 * arg0: verbosity. Higher numbers give more verbose messages
116 * arg1: Textual message
118 SDT_PROBE_DEFINE2(fuse, , vnops, trace, "int", "char*");
121 static vop_access_t fuse_vnop_access;
122 static vop_close_t fuse_vnop_close;
123 static vop_create_t fuse_vnop_create;
124 static vop_deleteextattr_t fuse_vnop_deleteextattr;
125 static vop_fdatasync_t fuse_vnop_fdatasync;
126 static vop_fsync_t fuse_vnop_fsync;
127 static vop_getattr_t fuse_vnop_getattr;
128 static vop_getextattr_t fuse_vnop_getextattr;
129 static vop_inactive_t fuse_vnop_inactive;
130 static vop_link_t fuse_vnop_link;
131 static vop_listextattr_t fuse_vnop_listextattr;
132 static vop_lookup_t fuse_vnop_lookup;
133 static vop_mkdir_t fuse_vnop_mkdir;
134 static vop_mknod_t fuse_vnop_mknod;
135 static vop_open_t fuse_vnop_open;
136 static vop_pathconf_t fuse_vnop_pathconf;
137 static vop_read_t fuse_vnop_read;
138 static vop_readdir_t fuse_vnop_readdir;
139 static vop_readlink_t fuse_vnop_readlink;
140 static vop_reclaim_t fuse_vnop_reclaim;
141 static vop_remove_t fuse_vnop_remove;
142 static vop_rename_t fuse_vnop_rename;
143 static vop_rmdir_t fuse_vnop_rmdir;
144 static vop_setattr_t fuse_vnop_setattr;
145 static vop_setextattr_t fuse_vnop_setextattr;
146 static vop_strategy_t fuse_vnop_strategy;
147 static vop_symlink_t fuse_vnop_symlink;
148 static vop_write_t fuse_vnop_write;
149 static vop_getpages_t fuse_vnop_getpages;
150 static vop_putpages_t fuse_vnop_putpages;
151 static vop_print_t fuse_vnop_print;
153 struct vop_vector fuse_vnops = {
154 .vop_default = &default_vnodeops,
155 .vop_access = fuse_vnop_access,
156 .vop_close = fuse_vnop_close,
157 .vop_create = fuse_vnop_create,
158 .vop_deleteextattr = fuse_vnop_deleteextattr,
159 .vop_fsync = fuse_vnop_fsync,
160 .vop_fdatasync = fuse_vnop_fdatasync,
161 .vop_getattr = fuse_vnop_getattr,
162 .vop_getextattr = fuse_vnop_getextattr,
163 .vop_inactive = fuse_vnop_inactive,
164 .vop_link = fuse_vnop_link,
165 .vop_listextattr = fuse_vnop_listextattr,
166 .vop_lookup = fuse_vnop_lookup,
167 .vop_mkdir = fuse_vnop_mkdir,
168 .vop_mknod = fuse_vnop_mknod,
169 .vop_open = fuse_vnop_open,
170 .vop_pathconf = fuse_vnop_pathconf,
171 .vop_read = fuse_vnop_read,
172 .vop_readdir = fuse_vnop_readdir,
173 .vop_readlink = fuse_vnop_readlink,
174 .vop_reclaim = fuse_vnop_reclaim,
175 .vop_remove = fuse_vnop_remove,
176 .vop_rename = fuse_vnop_rename,
177 .vop_rmdir = fuse_vnop_rmdir,
178 .vop_setattr = fuse_vnop_setattr,
179 .vop_setextattr = fuse_vnop_setextattr,
180 .vop_strategy = fuse_vnop_strategy,
181 .vop_symlink = fuse_vnop_symlink,
182 .vop_write = fuse_vnop_write,
183 .vop_getpages = fuse_vnop_getpages,
184 .vop_putpages = fuse_vnop_putpages,
185 .vop_print = fuse_vnop_print,
188 static u_long fuse_lookup_cache_hits = 0;
190 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
191 &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup");
193 static u_long fuse_lookup_cache_misses = 0;
195 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
196 &fuse_lookup_cache_misses, 0, "number of cache misses in lookup");
199 * XXX: This feature is highly experimental and can bring to instabilities,
200 * needs revisiting before to be enabled by default.
202 static int fuse_reclaim_revoked = 0;
204 SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
205 &fuse_reclaim_revoked, 0, "");
207 uma_zone_t fuse_pbuf_zone;
209 #define fuse_vm_page_lock(m) vm_page_lock((m));
210 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
211 #define fuse_vm_page_lock_queues() ((void)0)
212 #define fuse_vm_page_unlock_queues() ((void)0)
214 /* Check permission for extattr operations, much like extattr_check_cred */
216 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
217 struct thread *td, accmode_t accmode)
219 struct mount *mp = vnode_mount(vp);
220 struct fuse_data *data = fuse_get_mpdata(mp);
223 * Kernel-invoked always succeeds.
229 * Do not allow privileged processes in jail to directly manipulate
233 case EXTATTR_NAMESPACE_SYSTEM:
234 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
235 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
238 case EXTATTR_NAMESPACE_USER:
239 return (fuse_internal_access(vp, accmode, td, cred));
245 /* Get a filehandle for a directory */
247 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
248 struct ucred *cred, pid_t pid)
250 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
252 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
255 /* Send FUSE_FLUSH for this vnode */
257 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
259 struct fuse_flush_in *ffi;
260 struct fuse_filehandle *fufh;
261 struct fuse_dispatcher fdi;
262 struct thread *td = curthread;
263 struct mount *mp = vnode_mount(vp);
266 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
269 err = fuse_filehandle_get(vp, fflag, &fufh, cred, pid);
273 fdisp_init(&fdi, sizeof(*ffi));
274 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
276 ffi->fh = fufh->fh_id;
278 err = fdisp_wait_answ(&fdi);
280 fsess_set_notimpl(mp, FUSE_FLUSH);
288 struct vnop_access_args {
290 #if VOP_ACCESS_TAKES_ACCMODE_T
295 struct ucred *a_cred;
300 fuse_vnop_access(struct vop_access_args *ap)
302 struct vnode *vp = ap->a_vp;
303 int accmode = ap->a_accmode;
304 struct ucred *cred = ap->a_cred;
306 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
310 if (fuse_isdeadfs(vp)) {
311 if (vnode_isvroot(vp)) {
316 if (!(data->dataflags & FSESS_INITED)) {
317 if (vnode_isvroot(vp)) {
318 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
319 (fuse_match_cred(data->daemoncred, cred) == 0)) {
325 if (vnode_islnk(vp)) {
329 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
334 struct vop_close_args {
337 struct ucred *a_cred;
342 fuse_vnop_close(struct vop_close_args *ap)
344 struct vnode *vp = ap->a_vp;
345 struct ucred *cred = ap->a_cred;
346 int fflag = ap->a_fflag;
347 struct thread *td = ap->a_td;
348 pid_t pid = td->td_proc->p_pid;
351 if (fuse_isdeadfs(vp))
355 if (fflag & IO_NDELAY)
358 err = fuse_flush(vp, cred, pid, fflag);
359 /* TODO: close the file handle, if we're sure it's no longer used */
360 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
361 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
367 fdisp_make_mknod_for_fallback(
368 struct fuse_dispatcher *fdip,
369 struct componentname *cnp,
375 enum fuse_opcode *op)
377 struct fuse_mknod_in *fmni;
379 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
381 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
385 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
387 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
390 struct vnop_create_args {
392 struct vnode **a_vpp;
393 struct componentname *a_cnp;
398 fuse_vnop_create(struct vop_create_args *ap)
400 struct vnode *dvp = ap->a_dvp;
401 struct vnode **vpp = ap->a_vpp;
402 struct componentname *cnp = ap->a_cnp;
403 struct vattr *vap = ap->a_vap;
404 struct thread *td = cnp->cn_thread;
405 struct ucred *cred = cnp->cn_cred;
407 struct fuse_open_in *foi;
408 struct fuse_entry_out *feo;
409 struct fuse_open_out *foo;
410 struct fuse_dispatcher fdi, fdi2;
411 struct fuse_dispatcher *fdip = &fdi;
412 struct fuse_dispatcher *fdip2 = NULL;
416 struct mount *mp = vnode_mount(dvp);
417 uint64_t parentnid = VTOFUD(dvp)->nid;
418 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
423 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
424 * writable mode makes sense, and we might as well include readability
429 if (fuse_isdeadfs(dvp)) {
432 bzero(&fdi, sizeof(fdi));
434 if ((vap->va_type != VREG))
437 if (!fsess_isimpl(mp, FUSE_CREATE)) {
438 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
439 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
442 /* Use FUSE_CREATE */
444 fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
445 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
448 foi->flags = O_CREAT | flags;
449 memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
451 ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
454 err = fdisp_wait_answ(fdip);
457 if (err == ENOSYS && op == FUSE_CREATE) {
458 fsess_set_notimpl(mp, FUSE_CREATE);
459 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
460 parentnid, td, cred, mode, &op);
461 err = fdisp_wait_answ(fdip);
469 if ((err = fuse_internal_checkentry(feo, VREG))) {
473 if (op == FUSE_CREATE) {
474 foo = (struct fuse_open_out*)(feo + 1);
476 /* Issue a separate FUSE_OPEN */
478 fdisp_init(fdip2, sizeof(*foi));
479 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
484 err = fdisp_wait_answ(fdip2);
489 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, VREG);
491 struct fuse_release_in *fri;
492 uint64_t nodeid = feo->nodeid;
493 uint64_t fh_id = foo->fh;
495 fdisp_init(fdip, sizeof(*fri));
496 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
500 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
501 fuse_insert_message(fdip->tick);
504 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
505 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
506 feo->attr_valid_nsec, NULL);
508 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
509 fuse_vnode_open(*vpp, foo->open_flags, td);
510 cache_purge_negative(dvp);
514 fdisp_destroy(fdip2);
520 struct vnop_fdatasync_args {
521 struct vop_generic_args a_gen;
523 struct thread * a_td;
527 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
529 struct vnode *vp = ap->a_vp;
530 struct thread *td = ap->a_td;
531 int waitfor = MNT_WAIT;
535 if (fuse_isdeadfs(vp)) {
538 if ((err = vop_stdfdatasync_buf(ap)))
541 return fuse_internal_fsync(vp, td, waitfor, true);
545 struct vnop_fsync_args {
546 struct vop_generic_args a_gen;
549 struct thread * a_td;
553 fuse_vnop_fsync(struct vop_fsync_args *ap)
555 struct vnode *vp = ap->a_vp;
556 struct thread *td = ap->a_td;
557 int waitfor = ap->a_waitfor;
560 if (fuse_isdeadfs(vp)) {
563 if ((err = vop_stdfsync(ap)))
566 return fuse_internal_fsync(vp, td, waitfor, false);
570 struct vnop_getattr_args {
573 struct ucred *a_cred;
578 fuse_vnop_getattr(struct vop_getattr_args *ap)
580 struct vnode *vp = ap->a_vp;
581 struct vattr *vap = ap->a_vap;
582 struct ucred *cred = ap->a_cred;
583 struct thread *td = curthread;
588 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
590 /* Note that we are not bailing out on a dead file system just yet. */
592 if (!(dataflags & FSESS_INITED)) {
593 if (!vnode_isvroot(vp)) {
594 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
601 err = fuse_internal_getattr(vp, vap, cred, td);
602 if (err == ENOTCONN && vnode_isvroot(vp)) {
603 /* see comment in fuse_vfsop_statfs() */
610 bzero(vap, sizeof(*vap));
611 vap->va_type = vnode_vtype(vp);
617 struct vnop_inactive_args {
623 fuse_vnop_inactive(struct vop_inactive_args *ap)
625 struct vnode *vp = ap->a_vp;
626 struct thread *td = ap->a_td;
628 struct fuse_vnode_data *fvdat = VTOFUD(vp);
629 struct fuse_filehandle *fufh, *fufh_tmp;
633 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
634 if (need_flush && vp->v_type == VREG) {
635 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
636 fuse_vnode_savesize(vp, NULL, 0);
638 if (fuse_data_cache_invalidate ||
639 (fvdat->flag & FN_REVOKED) != 0)
640 fuse_io_invalbuf(vp, td);
642 fuse_io_flushbuf(vp, MNT_WAIT, td);
645 fuse_filehandle_close(vp, fufh, td, NULL);
648 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
655 struct vnop_link_args {
656 struct vnode *a_tdvp;
658 struct componentname *a_cnp;
662 fuse_vnop_link(struct vop_link_args *ap)
664 struct vnode *vp = ap->a_vp;
665 struct vnode *tdvp = ap->a_tdvp;
666 struct componentname *cnp = ap->a_cnp;
668 struct vattr *vap = VTOVA(vp);
670 struct fuse_dispatcher fdi;
671 struct fuse_entry_out *feo;
672 struct fuse_link_in fli;
676 if (fuse_isdeadfs(vp)) {
679 if (vnode_mount(tdvp) != vnode_mount(vp)) {
684 * This is a seatbelt check to protect naive userspace filesystems from
685 * themselves and the limitations of the FUSE IPC protocol. If a
686 * filesystem does not allow attribute caching, assume it is capable of
687 * validating that nlink does not overflow.
689 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
691 fli.oldnodeid = VTOI(vp);
694 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
695 FUSE_LINK, &fli, sizeof(fli), &fdi);
696 if ((err = fdisp_wait_answ(&fdi))) {
701 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
707 struct fuse_lookup_alloc_arg {
708 struct fuse_entry_out *feo;
709 struct componentname *cnp;
714 /* Callback for vn_get_ino */
716 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
718 struct fuse_lookup_alloc_arg *flaa = arg;
720 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
724 SDT_PROBE_DEFINE3(fuse, , vnops, cache_lookup,
725 "int", "struct timespec*", "struct timespec*");
727 struct vnop_lookup_args {
728 struct vnodeop_desc *a_desc;
730 struct vnode **a_vpp;
731 struct componentname *a_cnp;
735 fuse_vnop_lookup(struct vop_lookup_args *ap)
737 struct vnode *dvp = ap->a_dvp;
738 struct vnode **vpp = ap->a_vpp;
739 struct componentname *cnp = ap->a_cnp;
740 struct thread *td = cnp->cn_thread;
741 struct ucred *cred = cnp->cn_cred;
743 int nameiop = cnp->cn_nameiop;
744 int flags = cnp->cn_flags;
745 int wantparent = flags & (LOCKPARENT | WANTPARENT);
746 int islastcn = flags & ISLASTCN;
747 struct mount *mp = vnode_mount(dvp);
751 struct vnode *vp = NULL;
753 struct fuse_dispatcher fdi;
754 bool did_lookup = false;
755 struct fuse_entry_out *feo = NULL;
756 enum vtype vtyp; /* vnode type of target */
757 off_t filesize; /* filesize of target */
761 if (fuse_isdeadfs(dvp)) {
765 if (!vnode_isdir(dvp))
768 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
771 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
774 if (flags & ISDOTDOT) {
775 nid = VTOFUD(dvp)->parent_nid;
778 /* .. is obviously a directory */
781 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
783 /* . is obviously a directory */
787 struct timespec now, timeout;
789 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
791 SDT_PROBE3(fuse, , vnops, cache_lookup, err, &timeout, &now);
793 case -1: /* positive match */
794 if (timespeccmp(&timeout, &now, >)) {
795 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
798 atomic_add_acq_long(&fuse_lookup_cache_misses,
800 fuse_internal_vnode_disappear(*vpp);
810 case 0: /* no match in cache */
811 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
814 case ENOENT: /* negative match */
816 if (timespeccmp(&timeout, &now, <=)) {
818 cache_purge_negative(dvp);
827 fdisp_init(&fdi, cnp->cn_namelen + 1);
828 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
830 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
831 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
832 lookup_err = fdisp_wait_answ(&fdi);
836 /* lookup call succeeded */
837 nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
838 feo = (struct fuse_entry_out *)fdi.answ;
840 /* zero nodeid means ENOENT and cache it */
841 struct timespec timeout;
843 fdi.answ_stat = ENOENT;
845 if (cnp->cn_flags & MAKEENTRY) {
846 fuse_validity_2_timespec(feo, &timeout);
847 cache_enter_time(dvp, *vpp, cnp,
850 } else if (nid == FUSE_ROOT_ID) {
853 vtyp = IFTOVT(feo->attr.mode);
854 filesize = feo->attr.size;
856 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
861 /* lookup_err, if non-zero, must be ENOENT at this point */
864 /* Entry not found */
865 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
866 err = fuse_internal_access(dvp, VWRITE, td, cred);
871 * Set the SAVENAME flag to hold onto the
872 * pathname for use later in VOP_CREATE or VOP_RENAME.
874 cnp->cn_flags |= SAVENAME;
884 /* Entry was found */
885 if ((nameiop == DELETE || nameiop == RENAME) && islastcn) {
886 err = fuse_internal_access(dvp, VWRITE, td, cred);
890 * TODO: if the parent's sticky bit is set, check
891 * whether we're allowed to remove the file.
892 * Need to figure out the vnode locking to make this
897 fuse_internal_getattr(dvp, &dvattr, cred, td);
898 if ((dvattr.va_mode & S_ISTXT) &&
899 fuse_internal_access(dvp, VADMIN, td, cred) &&
900 fuse_internal_access(*vpp, VADMIN, td, cred)) {
908 * If deleting, and at end of pathname, return parameters
909 * which can be used to remove file. If the wantparent flag
910 * isn't set, we return only the directory, otherwise we go on
911 * and lock the inode, being careful with ".".
913 if (nameiop == DELETE && islastcn) {
914 if (nid == VTOI(dvp)) {
918 err = fuse_vnode_get(dvp->v_mount, feo, nid,
919 dvp, &vp, cnp, vtyp);
926 * Save the name for use in VOP_RMDIR and VOP_REMOVE
929 cnp->cn_flags |= SAVENAME;
934 * If rewriting (RENAME), return the inode and the
935 * information required to rewrite the present directory
936 * Must get inode of directory entry to verify it's a
937 * regular file, or empty directory.
939 if (nameiop == RENAME && wantparent && islastcn) {
943 if (nid == VTOI(dvp)) {
947 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
954 * Save the name for use in VOP_RENAME later.
956 cnp->cn_flags |= SAVENAME;
960 if (flags & ISDOTDOT) {
961 struct fuse_lookup_alloc_arg flaa;
967 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
972 } else if (nid == VTOI(dvp)) {
976 struct fuse_vnode_data *fvdat;
978 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
983 fuse_vnode_setparent(vp, dvp);
986 * In the case where we are looking up a FUSE node
987 * represented by an existing cached vnode, and the
988 * true size reported by FUSE_LOOKUP doesn't match
989 * the vnode's cached size, fix the vnode cache to
990 * match the real object size.
992 * This can occur via FUSE distributed filesystems,
993 * irregular files, etc.
996 if (vnode_isreg(vp) &&
997 filesize != fvdat->filesize) {
999 * The FN_SIZECHANGE flag reflects a dirty
1000 * append. If userspace lets us know our cache
1001 * is invalid, that write was lost. (Dirty
1002 * writes that do not cause append are also
1003 * lost, but we don't detect them here.)
1005 * XXX: Maybe disable WB caching on this mount.
1007 if (fvdat->flag & FN_SIZECHANGE)
1008 printf("%s: WB cache incoherent on "
1010 vnode_mount(vp)->mnt_stat.f_mntonname);
1012 (void)fuse_vnode_setsize(vp, cred, filesize);
1013 fvdat->flag &= ~FN_SIZECHANGE;
1019 fuse_internal_cache_attrs(*vpp, &feo->attr,
1020 feo->attr_valid, feo->attr_valid_nsec, NULL);
1026 /* No lookup error; need to clean up. */
1028 if (err) { /* Found inode; exit with no vnode. */
1030 fuse_internal_forget_send(vnode_mount(dvp), td,
1034 fdisp_destroy(&fdi);
1039 fdisp_destroy(&fdi);
1045 struct vnop_mkdir_args {
1046 struct vnode *a_dvp;
1047 struct vnode **a_vpp;
1048 struct componentname *a_cnp;
1049 struct vattr *a_vap;
1053 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1055 struct vnode *dvp = ap->a_dvp;
1056 struct vnode **vpp = ap->a_vpp;
1057 struct componentname *cnp = ap->a_cnp;
1058 struct vattr *vap = ap->a_vap;
1060 struct fuse_mkdir_in fmdi;
1062 if (fuse_isdeadfs(dvp)) {
1065 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1067 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1068 sizeof(fmdi), VDIR));
1072 struct vnop_mknod_args {
1073 struct vnode *a_dvp;
1074 struct vnode **a_vpp;
1075 struct componentname *a_cnp;
1076 struct vattr *a_vap;
1080 fuse_vnop_mknod(struct vop_mknod_args *ap)
1083 struct vnode *dvp = ap->a_dvp;
1084 struct vnode **vpp = ap->a_vpp;
1085 struct componentname *cnp = ap->a_cnp;
1086 struct vattr *vap = ap->a_vap;
1087 struct fuse_mknod_in fmni;
1089 if (fuse_isdeadfs(dvp))
1092 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1093 fmni.rdev = vap->va_rdev;
1094 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
1095 sizeof(fmni), vap->va_type));
1099 struct vnop_open_args {
1102 struct ucred *a_cred;
1103 struct thread *a_td;
1104 int a_fdidx; / struct file *a_fp;
1108 fuse_vnop_open(struct vop_open_args *ap)
1110 struct vnode *vp = ap->a_vp;
1111 int a_mode = ap->a_mode;
1112 struct thread *td = ap->a_td;
1113 struct ucred *cred = ap->a_cred;
1114 pid_t pid = td->td_proc->p_pid;
1115 struct fuse_vnode_data *fvdat;
1117 if (fuse_isdeadfs(vp))
1119 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1120 return (EOPNOTSUPP);
1121 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1126 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1127 fuse_vnode_open(vp, 0, td);
1131 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1135 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1138 switch (ap->a_name) {
1139 case _PC_FILESIZEBITS:
1143 *ap->a_retval = NAME_MAX;
1146 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1148 case _PC_SYMLINK_MAX:
1149 *ap->a_retval = MAXPATHLEN;
1155 return (vop_stdpathconf(ap));
1160 struct vnop_read_args {
1164 struct ucred *a_cred;
1168 fuse_vnop_read(struct vop_read_args *ap)
1170 struct vnode *vp = ap->a_vp;
1171 struct uio *uio = ap->a_uio;
1172 int ioflag = ap->a_ioflag;
1173 struct ucred *cred = ap->a_cred;
1174 pid_t pid = curthread->td_proc->p_pid;
1176 if (fuse_isdeadfs(vp)) {
1180 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1181 ioflag |= IO_DIRECT;
1184 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1188 struct vnop_readdir_args {
1191 struct ucred *a_cred;
1198 fuse_vnop_readdir(struct vop_readdir_args *ap)
1200 struct vnode *vp = ap->a_vp;
1201 struct uio *uio = ap->a_uio;
1202 struct ucred *cred = ap->a_cred;
1203 struct fuse_filehandle *fufh = NULL;
1204 struct fuse_iov cookediov;
1206 pid_t pid = curthread->td_proc->p_pid;
1208 if (fuse_isdeadfs(vp)) {
1211 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1212 (uio_resid(uio) < sizeof(struct dirent))) {
1216 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1219 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1220 fiov_init(&cookediov, DIRCOOKEDSIZE);
1222 err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1224 fiov_teardown(&cookediov);
1230 struct vnop_readlink_args {
1233 struct ucred *a_cred;
1237 fuse_vnop_readlink(struct vop_readlink_args *ap)
1239 struct vnode *vp = ap->a_vp;
1240 struct uio *uio = ap->a_uio;
1241 struct ucred *cred = ap->a_cred;
1243 struct fuse_dispatcher fdi;
1246 if (fuse_isdeadfs(vp)) {
1249 if (!vnode_islnk(vp)) {
1252 fdisp_init(&fdi, 0);
1253 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1257 if (((char *)fdi.answ)[0] == '/' &&
1258 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1259 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1261 err = uiomove(mpth, strlen(mpth), uio);
1264 err = uiomove(fdi.answ, fdi.iosize, uio);
1267 fdisp_destroy(&fdi);
1272 struct vnop_reclaim_args {
1274 struct thread *a_td;
1278 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1280 struct vnode *vp = ap->a_vp;
1281 struct thread *td = ap->a_td;
1282 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1283 struct fuse_filehandle *fufh, *fufh_tmp;
1286 panic("FUSE: no vnode data during recycling");
1288 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1289 printf("FUSE: vnode being reclaimed with open fufh "
1290 "(type=%#x)", fufh->fufh_type);
1291 fuse_filehandle_close(vp, fufh, td, NULL);
1294 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1295 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1298 fuse_vnode_setparent(vp, NULL);
1300 vfs_hash_remove(vp);
1301 vnode_destroy_vobject(vp);
1302 fuse_vnode_destroy(vp);
1308 struct vnop_remove_args {
1309 struct vnode *a_dvp;
1311 struct componentname *a_cnp;
1315 fuse_vnop_remove(struct vop_remove_args *ap)
1317 struct vnode *dvp = ap->a_dvp;
1318 struct vnode *vp = ap->a_vp;
1319 struct componentname *cnp = ap->a_cnp;
1323 if (fuse_isdeadfs(vp)) {
1326 if (vnode_isdir(vp)) {
1331 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1334 fuse_internal_vnode_disappear(vp);
1339 struct vnop_rename_args {
1340 struct vnode *a_fdvp;
1341 struct vnode *a_fvp;
1342 struct componentname *a_fcnp;
1343 struct vnode *a_tdvp;
1344 struct vnode *a_tvp;
1345 struct componentname *a_tcnp;
1349 fuse_vnop_rename(struct vop_rename_args *ap)
1351 struct vnode *fdvp = ap->a_fdvp;
1352 struct vnode *fvp = ap->a_fvp;
1353 struct componentname *fcnp = ap->a_fcnp;
1354 struct vnode *tdvp = ap->a_tdvp;
1355 struct vnode *tvp = ap->a_tvp;
1356 struct componentname *tcnp = ap->a_tcnp;
1357 struct fuse_data *data;
1361 if (fuse_isdeadfs(fdvp)) {
1364 if (fvp->v_mount != tdvp->v_mount ||
1365 (tvp && fvp->v_mount != tvp->v_mount)) {
1366 SDT_PROBE2(fuse, , vnops, trace, 1, "cross-device rename");
1373 * FUSE library is expected to check if target directory is not
1374 * under the source directory in the file system tree.
1375 * Linux performs this check at VFS level.
1377 data = fuse_get_mpdata(vnode_mount(tdvp));
1378 sx_xlock(&data->rename_lock);
1379 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1382 fuse_vnode_setparent(fvp, tdvp);
1384 fuse_vnode_setparent(tvp, NULL);
1386 sx_unlock(&data->rename_lock);
1388 if (tvp != NULL && tvp != fvp) {
1391 if (vnode_isdir(fvp)) {
1392 if ((tvp != NULL) && vnode_isdir(tvp)) {
1413 struct vnop_rmdir_args {
1414 struct vnode *a_dvp;
1416 struct componentname *a_cnp;
1420 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1422 struct vnode *dvp = ap->a_dvp;
1423 struct vnode *vp = ap->a_vp;
1427 if (fuse_isdeadfs(vp)) {
1430 if (VTOFUD(vp) == VTOFUD(dvp)) {
1433 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1436 fuse_internal_vnode_disappear(vp);
1441 struct vnop_setattr_args {
1443 struct vattr *a_vap;
1444 struct ucred *a_cred;
1445 struct thread *a_td;
1449 fuse_vnop_setattr(struct vop_setattr_args *ap)
1451 struct vnode *vp = ap->a_vp;
1452 struct vattr *vap = ap->a_vap;
1453 struct ucred *cred = ap->a_cred;
1454 struct thread *td = curthread;
1455 struct fuse_dispatcher fdi;
1456 struct fuse_setattr_in *fsai;
1457 pid_t pid = td->td_proc->p_pid;
1461 int sizechanged = 0;
1462 uint64_t newsize = 0;
1463 accmode_t accmode = 0;
1465 if (fuse_isdeadfs(vp)) {
1468 fdisp_init(&fdi, sizeof(*fsai));
1469 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1473 if (vap->va_uid != (uid_t)VNOVAL) {
1474 fsai->uid = vap->va_uid;
1475 fsai->valid |= FATTR_UID;
1478 if (vap->va_gid != (gid_t)VNOVAL) {
1479 fsai->gid = vap->va_gid;
1480 fsai->valid |= FATTR_GID;
1483 if (vap->va_size != VNOVAL) {
1485 struct fuse_filehandle *fufh = NULL;
1487 /*Truncate to a new value. */
1488 fsai->size = vap->va_size;
1490 newsize = vap->va_size;
1491 fsai->valid |= FATTR_SIZE;
1494 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1496 fsai->fh = fufh->fh_id;
1497 fsai->valid |= FATTR_FH;
1500 if (vap->va_atime.tv_sec != VNOVAL) {
1501 fsai->atime = vap->va_atime.tv_sec;
1502 fsai->atimensec = vap->va_atime.tv_nsec;
1503 fsai->valid |= FATTR_ATIME;
1506 * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181
1509 if (vap->va_mtime.tv_sec != VNOVAL) {
1510 fsai->mtime = vap->va_mtime.tv_sec;
1511 fsai->mtimensec = vap->va_mtime.tv_nsec;
1512 fsai->valid |= FATTR_MTIME;
1515 * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181
1518 if (vap->va_mode != (mode_t)VNOVAL) {
1519 fsai->mode = vap->va_mode & ALLPERMS;
1520 fsai->valid |= FATTR_MODE;
1526 vtyp = vnode_vtype(vp);
1528 if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
1532 if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
1536 err = fuse_internal_access(vp, accmode, td, cred);
1540 if ((err = fdisp_wait_answ(&fdi)))
1542 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1544 if (vnode_vtype(vp) != vtyp) {
1545 if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1546 SDT_PROBE2(fuse, , vnops, trace, 1, "FUSE: Dang! "
1547 "vnode_vtype is VNON and vtype isn't.");
1550 * STALE vnode, ditch
1552 * The vnode has changed its type "behind our back".
1553 * There's nothing really we can do, so let us just
1554 * force an internal revocation and tell the caller to
1555 * try again, if interested.
1557 fuse_internal_vnode_disappear(vp);
1562 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1563 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1564 fao->attr_valid_nsec, NULL);
1568 fdisp_destroy(&fdi);
1569 if (!err && sizechanged) {
1570 fuse_vnode_setsize(vp, cred, newsize);
1571 VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1577 struct vnop_strategy_args {
1583 fuse_vnop_strategy(struct vop_strategy_args *ap)
1585 struct vnode *vp = ap->a_vp;
1586 struct buf *bp = ap->a_bp;
1588 if (!vp || fuse_isdeadfs(vp)) {
1589 bp->b_ioflags |= BIO_ERROR;
1590 bp->b_error = ENXIO;
1594 if (bp->b_iocmd == BIO_WRITE) {
1597 err = fuse_vnode_refreshsize(vp, NOCRED);
1599 bp->b_ioflags |= BIO_ERROR;
1607 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1608 * fuse_io_strategy sets bp's error fields
1610 (void)fuse_io_strategy(vp, bp);
1617 struct vnop_symlink_args {
1618 struct vnode *a_dvp;
1619 struct vnode **a_vpp;
1620 struct componentname *a_cnp;
1621 struct vattr *a_vap;
1626 fuse_vnop_symlink(struct vop_symlink_args *ap)
1628 struct vnode *dvp = ap->a_dvp;
1629 struct vnode **vpp = ap->a_vpp;
1630 struct componentname *cnp = ap->a_cnp;
1631 const char *target = ap->a_target;
1633 struct fuse_dispatcher fdi;
1638 if (fuse_isdeadfs(dvp)) {
1642 * Unlike the other creator type calls, here we have to create a message
1643 * where the name of the new entry comes first, and the data describing
1644 * the entry comes second.
1645 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1646 * but put together the message manually and just call the core part.
1649 len = strlen(target) + 1;
1650 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1651 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1653 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1654 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1655 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1657 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1658 fdisp_destroy(&fdi);
1663 struct vnop_write_args {
1667 struct ucred *a_cred;
1671 fuse_vnop_write(struct vop_write_args *ap)
1673 struct vnode *vp = ap->a_vp;
1674 struct uio *uio = ap->a_uio;
1675 int ioflag = ap->a_ioflag;
1676 struct ucred *cred = ap->a_cred;
1677 pid_t pid = curthread->td_proc->p_pid;
1680 if (fuse_isdeadfs(vp)) {
1683 err = fuse_vnode_refreshsize(vp, cred);
1687 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1688 ioflag |= IO_DIRECT;
1691 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1694 SDT_PROBE_DEFINE1(fuse, , vnops, vnop_getpages_error, "int");
1696 struct vnop_getpages_args {
1704 fuse_vnop_getpages(struct vop_getpages_args *ap)
1706 int i, error, nextoff, size, toff, count, npages;
1715 pid_t pid = curthread->td_proc->p_pid;
1718 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1719 td = curthread; /* XXX */
1720 cred = curthread->td_ucred; /* XXX */
1722 npages = ap->a_count;
1724 if (!fsess_opt_mmap(vnode_mount(vp))) {
1725 SDT_PROBE2(fuse, , vnops, trace, 1,
1726 "called on non-cacheable vnode??\n");
1727 return (VM_PAGER_ERROR);
1731 * If the last page is partially valid, just return it and allow
1732 * the pager to zero-out the blanks. Partially valid pages can
1733 * only occur at the file EOF.
1735 * XXXGL: is that true for FUSE, which is a local filesystem,
1736 * but still somewhat disconnected from the kernel?
1738 VM_OBJECT_WLOCK(vp->v_object);
1739 if (pages[npages - 1]->valid != 0 && --npages == 0)
1741 VM_OBJECT_WUNLOCK(vp->v_object);
1744 * We use only the kva address for the buffer, but this is extremely
1745 * convenient and fast.
1747 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1749 kva = (vm_offset_t)bp->b_data;
1750 pmap_qenter(kva, pages, npages);
1751 VM_CNT_INC(v_vnodein);
1752 VM_CNT_ADD(v_vnodepgsin, npages);
1754 count = npages << PAGE_SHIFT;
1755 iov.iov_base = (caddr_t)kva;
1756 iov.iov_len = count;
1759 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1760 uio.uio_resid = count;
1761 uio.uio_segflg = UIO_SYSSPACE;
1762 uio.uio_rw = UIO_READ;
1765 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid);
1766 pmap_qremove(kva, npages);
1768 uma_zfree(fuse_pbuf_zone, bp);
1770 if (error && (uio.uio_resid == count)) {
1771 SDT_PROBE1(fuse, , vnops, vnop_getpages_error, error);
1772 return VM_PAGER_ERROR;
1775 * Calculate the number of bytes read and validate only that number
1776 * of bytes. Note that due to pending writes, size may be 0. This
1777 * does not mean that the remaining data is invalid!
1780 size = count - uio.uio_resid;
1781 VM_OBJECT_WLOCK(vp->v_object);
1782 fuse_vm_page_lock_queues();
1783 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1786 nextoff = toff + PAGE_SIZE;
1789 if (nextoff <= size) {
1791 * Read operation filled an entire page
1793 m->valid = VM_PAGE_BITS_ALL;
1794 KASSERT(m->dirty == 0,
1795 ("fuse_getpages: page %p is dirty", m));
1796 } else if (size > toff) {
1798 * Read operation filled a partial page.
1801 vm_page_set_valid_range(m, 0, size - toff);
1802 KASSERT(m->dirty == 0,
1803 ("fuse_getpages: page %p is dirty", m));
1806 * Read operation was short. If no error occurred
1807 * we may have hit a zero-fill section. We simply
1808 * leave valid set to 0.
1813 fuse_vm_page_unlock_queues();
1815 VM_OBJECT_WUNLOCK(vp->v_object);
1820 return (VM_PAGER_OK);
1824 struct vnop_putpages_args {
1830 vm_ooffset_t a_offset;
1834 fuse_vnop_putpages(struct vop_putpages_args *ap)
1840 int i, error, npages, count;
1848 pid_t pid = curthread->td_proc->p_pid;
1851 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1852 fsize = vp->v_object->un_pager.vnp.vnp_size;
1853 td = curthread; /* XXX */
1854 cred = curthread->td_ucred; /* XXX */
1856 count = ap->a_count;
1857 rtvals = ap->a_rtvals;
1858 npages = btoc(count);
1859 offset = IDX_TO_OFF(pages[0]->pindex);
1861 if (!fsess_opt_mmap(vnode_mount(vp))) {
1862 SDT_PROBE2(fuse, , vnops, trace, 1,
1863 "called on non-cacheable vnode??\n");
1865 for (i = 0; i < npages; i++)
1866 rtvals[i] = VM_PAGER_AGAIN;
1869 * When putting pages, do not extend file past EOF.
1872 if (offset + count > fsize) {
1873 count = fsize - offset;
1878 * We use only the kva address for the buffer, but this is extremely
1879 * convenient and fast.
1881 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1883 kva = (vm_offset_t)bp->b_data;
1884 pmap_qenter(kva, pages, npages);
1885 VM_CNT_INC(v_vnodeout);
1886 VM_CNT_ADD(v_vnodepgsout, count);
1888 iov.iov_base = (caddr_t)kva;
1889 iov.iov_len = count;
1892 uio.uio_offset = offset;
1893 uio.uio_resid = count;
1894 uio.uio_segflg = UIO_SYSSPACE;
1895 uio.uio_rw = UIO_WRITE;
1898 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid);
1900 pmap_qremove(kva, npages);
1901 uma_zfree(fuse_pbuf_zone, bp);
1904 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1906 for (i = 0; i < nwritten; i++) {
1907 rtvals[i] = VM_PAGER_OK;
1908 VM_OBJECT_WLOCK(pages[i]->object);
1909 vm_page_undirty(pages[i]);
1910 VM_OBJECT_WUNLOCK(pages[i]->object);
1916 static const char extattr_namespace_separator = '.';
1919 struct vop_getextattr_args {
1920 struct vop_generic_args a_gen;
1922 int a_attrnamespace;
1926 struct ucred *a_cred;
1927 struct thread *a_td;
1931 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
1933 struct vnode *vp = ap->a_vp;
1934 struct uio *uio = ap->a_uio;
1935 struct fuse_dispatcher fdi;
1936 struct fuse_getxattr_in *get_xattr_in;
1937 struct fuse_getxattr_out *get_xattr_out;
1938 struct mount *mp = vnode_mount(vp);
1939 struct thread *td = ap->a_td;
1940 struct ucred *cred = ap->a_cred;
1946 if (fuse_isdeadfs(vp))
1949 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
1953 /* Default to looking for user attributes. */
1954 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
1955 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
1957 prefix = EXTATTR_NAMESPACE_USER_STRING;
1959 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
1960 strlen(ap->a_name) + 1;
1962 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
1963 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
1965 get_xattr_in = fdi.indata;
1967 * Check to see whether we're querying the available size or
1968 * issuing the actual request. If we pass in 0, we get back struct
1969 * fuse_getxattr_out. If we pass in a non-zero size, we get back
1970 * that much data, without the struct fuse_getxattr_out header.
1973 get_xattr_in->size = 0;
1975 get_xattr_in->size = uio->uio_resid;
1977 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
1978 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
1981 err = fdisp_wait_answ(&fdi);
1984 fsess_set_notimpl(mp, FUSE_GETXATTR);
1988 get_xattr_out = fdi.answ;
1990 if (ap->a_size != NULL)
1991 *ap->a_size = get_xattr_out->size;
1994 err = uiomove(fdi.answ, fdi.iosize, uio);
1997 fdisp_destroy(&fdi);
2002 struct vop_setextattr_args {
2003 struct vop_generic_args a_gen;
2005 int a_attrnamespace;
2008 struct ucred *a_cred;
2009 struct thread *a_td;
2013 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2015 struct vnode *vp = ap->a_vp;
2016 struct uio *uio = ap->a_uio;
2017 struct fuse_dispatcher fdi;
2018 struct fuse_setxattr_in *set_xattr_in;
2019 struct mount *mp = vnode_mount(vp);
2020 struct thread *td = ap->a_td;
2021 struct ucred *cred = ap->a_cred;
2027 if (fuse_isdeadfs(vp))
2030 if (vfs_isrdonly(mp))
2033 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2034 if (ap->a_uio == NULL)
2037 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2042 /* Default to looking for user attributes. */
2043 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2044 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2046 prefix = EXTATTR_NAMESPACE_USER_STRING;
2048 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2049 strlen(ap->a_name) + 1;
2051 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2052 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2054 set_xattr_in = fdi.indata;
2055 set_xattr_in->size = uio->uio_resid;
2057 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2058 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2061 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2062 uio->uio_resid, uio);
2067 err = fdisp_wait_answ(&fdi);
2071 fsess_set_notimpl(mp, FUSE_SETXATTR);
2076 fdisp_destroy(&fdi);
2081 * The Linux / FUSE extended attribute list is simply a collection of
2082 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2083 * byte length followed by a non-NUL terminated string. So, this allows
2084 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2085 * Linux attribute names are reported with the namespace as a prefix (e.g.
2086 * "user.attribute_name"), but in FreeBSD they are reported without the
2087 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2089 * user.attr_name1\0user.attr_name2\0
2093 * <num>attr_name1<num>attr_name2
2095 * Where "<num>" is a single byte number of characters in the attribute name.
2098 * prefix - exattr namespace prefix string
2099 * list, list_len - input list with namespace prefixes
2100 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2103 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2104 char *bsd_list, int *bsd_list_len)
2106 int len, pos, dist_to_next, prefix_len;
2110 prefix_len = strlen(prefix);
2112 while (pos < list_len && list[pos] != '\0') {
2113 dist_to_next = strlen(&list[pos]) + 1;
2114 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2115 list[pos + prefix_len] == extattr_namespace_separator) {
2116 len = dist_to_next -
2117 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2118 if (len >= EXTATTR_MAXNAMELEN)
2119 return (ENAMETOOLONG);
2121 bsd_list[*bsd_list_len] = len;
2122 memcpy(&bsd_list[*bsd_list_len + 1],
2123 &list[pos + prefix_len +
2124 sizeof(extattr_namespace_separator)], len);
2126 *bsd_list_len += len + 1;
2129 pos += dist_to_next;
2136 struct vop_listextattr_args {
2137 struct vop_generic_args a_gen;
2139 int a_attrnamespace;
2142 struct ucred *a_cred;
2143 struct thread *a_td;
2147 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2149 struct vnode *vp = ap->a_vp;
2150 struct uio *uio = ap->a_uio;
2151 struct fuse_dispatcher fdi;
2152 struct fuse_listxattr_in *list_xattr_in;
2153 struct fuse_listxattr_out *list_xattr_out;
2154 struct mount *mp = vnode_mount(vp);
2155 struct thread *td = ap->a_td;
2156 struct ucred *cred = ap->a_cred;
2160 char *bsd_list = NULL;
2166 if (fuse_isdeadfs(vp))
2169 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2174 * Add space for a NUL and the period separator if enabled.
2175 * Default to looking for user attributes.
2177 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2178 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2180 prefix = EXTATTR_NAMESPACE_USER_STRING;
2182 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2184 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2185 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2188 * Retrieve Linux / FUSE compatible list size.
2190 list_xattr_in = fdi.indata;
2191 list_xattr_in->size = 0;
2192 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2193 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2195 err = fdisp_wait_answ(&fdi);
2198 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2202 list_xattr_out = fdi.answ;
2203 linux_list_len = list_xattr_out->size;
2204 if (linux_list_len == 0) {
2205 if (ap->a_size != NULL)
2206 *ap->a_size = linux_list_len;
2211 * Retrieve Linux / FUSE compatible list values.
2213 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2214 list_xattr_in = fdi.indata;
2215 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2216 list_xattr_in->flags = 0;
2217 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2218 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2220 err = fdisp_wait_answ(&fdi);
2224 linux_list = fdi.answ;
2225 linux_list_len = fdi.iosize;
2228 * Retrieve the BSD compatible list values.
2229 * The Linux / FUSE attribute list format isn't the same
2230 * as FreeBSD's format. So we need to transform it into
2231 * FreeBSD's format before giving it to the user.
2233 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2234 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2235 bsd_list, &bsd_list_len);
2239 if (ap->a_size != NULL)
2240 *ap->a_size = bsd_list_len;
2243 err = uiomove(bsd_list, bsd_list_len, uio);
2246 free(bsd_list, M_TEMP);
2247 fdisp_destroy(&fdi);
2252 struct vop_deleteextattr_args {
2253 struct vop_generic_args a_gen;
2255 int a_attrnamespace;
2257 struct ucred *a_cred;
2258 struct thread *a_td;
2262 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2264 struct vnode *vp = ap->a_vp;
2265 struct fuse_dispatcher fdi;
2266 struct mount *mp = vnode_mount(vp);
2267 struct thread *td = ap->a_td;
2268 struct ucred *cred = ap->a_cred;
2274 if (fuse_isdeadfs(vp))
2277 if (vfs_isrdonly(mp))
2280 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2285 /* Default to looking for user attributes. */
2286 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2287 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2289 prefix = EXTATTR_NAMESPACE_USER_STRING;
2291 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2292 strlen(ap->a_name) + 1;
2294 fdisp_init(&fdi, len);
2295 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2297 attr_str = fdi.indata;
2298 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2301 err = fdisp_wait_answ(&fdi);
2304 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2307 fdisp_destroy(&fdi);
2312 struct vnop_print_args {
2317 fuse_vnop_print(struct vop_print_args *ap)
2319 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2321 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2322 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2323 (uintmax_t)fvdat->nlookup,