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 * Copyright (c) 2019 The FreeBSD Foundation
38 * Portions of this software were developed by BFF Storage Systems, LLC under
39 * sponsorship from the FreeBSD Foundation.
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
50 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
66 #include <sys/param.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
72 #include <sys/filio.h>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
76 #include <sys/limits.h>
78 #include <sys/rwlock.h>
81 #include <sys/mount.h>
82 #include <sys/vnode.h>
83 #include <sys/namei.h>
84 #include <sys/extattr.h>
86 #include <sys/unistd.h>
87 #include <sys/filedesc.h>
89 #include <sys/fcntl.h>
90 #include <sys/dirent.h>
93 #include <sys/sysctl.h>
94 #include <sys/vmmeter.h>
97 #include <vm/vm_extern.h>
99 #include <vm/vm_map.h>
100 #include <vm/vm_page.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_pager.h>
104 #include <vm/vnode_pager.h>
105 #include <vm/vm_object.h>
108 #include "fuse_file.h"
109 #include "fuse_internal.h"
110 #include "fuse_ipc.h"
111 #include "fuse_node.h"
114 #include <sys/priv.h>
116 /* Maximum number of hardlinks to a single FUSE file */
117 #define FUSE_LINK_MAX UINT32_MAX
119 SDT_PROVIDER_DECLARE(fusefs);
122 * arg0: verbosity. Higher numbers give more verbose messages
123 * arg1: Textual message
125 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
128 static vop_access_t fuse_vnop_access;
129 static vop_advlock_t fuse_vnop_advlock;
130 static vop_allocate_t fuse_vnop_allocate;
131 static vop_bmap_t fuse_vnop_bmap;
132 static vop_close_t fuse_fifo_close;
133 static vop_close_t fuse_vnop_close;
134 static vop_copy_file_range_t fuse_vnop_copy_file_range;
135 static vop_create_t fuse_vnop_create;
136 static vop_deallocate_t fuse_vnop_deallocate;
137 static vop_deleteextattr_t fuse_vnop_deleteextattr;
138 static vop_fdatasync_t fuse_vnop_fdatasync;
139 static vop_fsync_t fuse_vnop_fsync;
140 static vop_getattr_t fuse_vnop_getattr;
141 static vop_getextattr_t fuse_vnop_getextattr;
142 static vop_inactive_t fuse_vnop_inactive;
143 static vop_ioctl_t fuse_vnop_ioctl;
144 static vop_link_t fuse_vnop_link;
145 static vop_listextattr_t fuse_vnop_listextattr;
146 static vop_lookup_t fuse_vnop_lookup;
147 static vop_mkdir_t fuse_vnop_mkdir;
148 static vop_mknod_t fuse_vnop_mknod;
149 static vop_open_t fuse_vnop_open;
150 static vop_pathconf_t fuse_vnop_pathconf;
151 static vop_read_t fuse_vnop_read;
152 static vop_readdir_t fuse_vnop_readdir;
153 static vop_readlink_t fuse_vnop_readlink;
154 static vop_reclaim_t fuse_vnop_reclaim;
155 static vop_remove_t fuse_vnop_remove;
156 static vop_rename_t fuse_vnop_rename;
157 static vop_rmdir_t fuse_vnop_rmdir;
158 static vop_setattr_t fuse_vnop_setattr;
159 static vop_setextattr_t fuse_vnop_setextattr;
160 static vop_strategy_t fuse_vnop_strategy;
161 static vop_symlink_t fuse_vnop_symlink;
162 static vop_write_t fuse_vnop_write;
163 static vop_getpages_t fuse_vnop_getpages;
164 static vop_print_t fuse_vnop_print;
165 static vop_vptofh_t fuse_vnop_vptofh;
167 struct vop_vector fuse_fifoops = {
168 .vop_default = &fifo_specops,
169 .vop_access = fuse_vnop_access,
170 .vop_close = fuse_fifo_close,
171 .vop_fsync = fuse_vnop_fsync,
172 .vop_getattr = fuse_vnop_getattr,
173 .vop_inactive = fuse_vnop_inactive,
174 .vop_pathconf = fuse_vnop_pathconf,
175 .vop_print = fuse_vnop_print,
176 .vop_read = VOP_PANIC,
177 .vop_reclaim = fuse_vnop_reclaim,
178 .vop_setattr = fuse_vnop_setattr,
179 .vop_write = VOP_PANIC,
180 .vop_vptofh = fuse_vnop_vptofh,
182 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
184 struct vop_vector fuse_vnops = {
185 .vop_allocate = fuse_vnop_allocate,
186 .vop_default = &default_vnodeops,
187 .vop_access = fuse_vnop_access,
188 .vop_advlock = fuse_vnop_advlock,
189 .vop_bmap = fuse_vnop_bmap,
190 .vop_close = fuse_vnop_close,
191 .vop_copy_file_range = fuse_vnop_copy_file_range,
192 .vop_create = fuse_vnop_create,
193 .vop_deallocate = fuse_vnop_deallocate,
194 .vop_deleteextattr = fuse_vnop_deleteextattr,
195 .vop_fsync = fuse_vnop_fsync,
196 .vop_fdatasync = fuse_vnop_fdatasync,
197 .vop_getattr = fuse_vnop_getattr,
198 .vop_getextattr = fuse_vnop_getextattr,
199 .vop_inactive = fuse_vnop_inactive,
200 .vop_ioctl = fuse_vnop_ioctl,
201 .vop_link = fuse_vnop_link,
202 .vop_listextattr = fuse_vnop_listextattr,
203 .vop_lookup = fuse_vnop_lookup,
204 .vop_mkdir = fuse_vnop_mkdir,
205 .vop_mknod = fuse_vnop_mknod,
206 .vop_open = fuse_vnop_open,
207 .vop_pathconf = fuse_vnop_pathconf,
209 * TODO: implement vop_poll after upgrading to protocol 7.21.
210 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
211 * 7.21, which adds the ability for the client to choose which poll
212 * events it wants, and for a client to deregister a file handle
214 .vop_read = fuse_vnop_read,
215 .vop_readdir = fuse_vnop_readdir,
216 .vop_readlink = fuse_vnop_readlink,
217 .vop_reclaim = fuse_vnop_reclaim,
218 .vop_remove = fuse_vnop_remove,
219 .vop_rename = fuse_vnop_rename,
220 .vop_rmdir = fuse_vnop_rmdir,
221 .vop_setattr = fuse_vnop_setattr,
222 .vop_setextattr = fuse_vnop_setextattr,
223 .vop_strategy = fuse_vnop_strategy,
224 .vop_symlink = fuse_vnop_symlink,
225 .vop_write = fuse_vnop_write,
226 .vop_getpages = fuse_vnop_getpages,
227 .vop_print = fuse_vnop_print,
228 .vop_vptofh = fuse_vnop_vptofh,
230 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
232 uma_zone_t fuse_pbuf_zone;
234 /* Check permission for extattr operations, much like extattr_check_cred */
236 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
237 struct thread *td, accmode_t accmode)
239 struct mount *mp = vnode_mount(vp);
240 struct fuse_data *data = fuse_get_mpdata(mp);
241 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
244 * Kernel-invoked always succeeds.
250 * Do not allow privileged processes in jail to directly manipulate
254 case EXTATTR_NAMESPACE_SYSTEM:
255 if (default_permissions) {
256 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
259 case EXTATTR_NAMESPACE_USER:
260 if (default_permissions) {
261 return (fuse_internal_access(vp, accmode, td, cred));
269 /* Get a filehandle for a directory */
271 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
272 struct ucred *cred, pid_t pid)
274 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
276 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
279 /* Send FUSE_FLUSH for this vnode */
281 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
283 struct fuse_flush_in *ffi;
284 struct fuse_filehandle *fufh;
285 struct fuse_dispatcher fdi;
286 struct thread *td = curthread;
287 struct mount *mp = vnode_mount(vp);
290 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
293 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
297 fdisp_init(&fdi, sizeof(*ffi));
298 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
300 ffi->fh = fufh->fh_id;
302 * If the file has a POSIX lock then we're supposed to set lock_owner.
303 * If not, then lock_owner is undefined. So we may as well always set
306 ffi->lock_owner = td->td_proc->p_pid;
308 err = fdisp_wait_answ(&fdi);
310 fsess_set_notimpl(mp, FUSE_FLUSH);
317 /* Close wrapper for fifos. */
319 fuse_fifo_close(struct vop_close_args *ap)
321 return (fifo_specops.vop_close(ap));
324 /* Invalidate a range of cached data, whether dirty of not */
326 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
329 daddr_t left_lbn, end_lbn, right_lbn;
331 int iosize, left_on, right_on, right_blksize;
333 iosize = fuse_iosize(vp);
334 left_lbn = start / iosize;
335 end_lbn = howmany(end, iosize);
336 left_on = start & (iosize - 1);
338 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
339 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
341 * Flush the dirty buffer, because we don't have a
342 * byte-granular way to record which parts of the
347 return (bp->b_error);
352 right_on = end & (iosize - 1);
354 right_lbn = end / iosize;
355 new_filesize = MAX(filesize, end);
356 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
357 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
358 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
360 * Flush the dirty buffer, because we don't have a
361 * byte-granular way to record which parts of the
366 return (bp->b_error);
372 v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
377 /* Send FUSE_LSEEK for this node */
379 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
380 pid_t pid, off_t *offp, int whence)
382 struct fuse_dispatcher fdi;
383 struct fuse_filehandle *fufh;
384 struct fuse_lseek_in *flsi;
385 struct fuse_lseek_out *flso;
386 struct mount *mp = vnode_mount(vp);
389 ASSERT_VOP_LOCKED(vp, __func__);
391 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
394 fdisp_init(&fdi, sizeof(*flsi));
395 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
397 flsi->fh = fufh->fh_id;
398 flsi->offset = *offp;
399 flsi->whence = whence;
400 err = fdisp_wait_answ(&fdi);
402 fsess_set_notimpl(mp, FUSE_LSEEK);
403 } else if (err == 0) {
404 fsess_set_impl(mp, FUSE_LSEEK);
406 *offp = flso->offset;
414 struct vnop_access_args {
416 #if VOP_ACCESS_TAKES_ACCMODE_T
421 struct ucred *a_cred;
426 fuse_vnop_access(struct vop_access_args *ap)
428 struct vnode *vp = ap->a_vp;
429 int accmode = ap->a_accmode;
430 struct ucred *cred = ap->a_cred;
432 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
436 if (fuse_isdeadfs(vp)) {
437 if (vnode_isvroot(vp)) {
442 if (!(data->dataflags & FSESS_INITED)) {
443 if (vnode_isvroot(vp)) {
444 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
445 (fuse_match_cred(data->daemoncred, cred) == 0)) {
451 if (vnode_islnk(vp)) {
455 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
460 * struct vop_advlock_args {
461 * struct vop_generic_args a_gen;
462 * struct vnode *a_vp;
465 * struct flock *a_fl;
470 fuse_vnop_advlock(struct vop_advlock_args *ap)
472 struct vnode *vp = ap->a_vp;
473 struct flock *fl = ap->a_fl;
474 struct thread *td = curthread;
475 struct ucred *cred = td->td_ucred;
476 pid_t pid = td->td_proc->p_pid;
477 struct fuse_filehandle *fufh;
478 struct fuse_dispatcher fdi;
479 struct fuse_lk_in *fli;
480 struct fuse_lk_out *flo;
483 int flags = ap->a_flags;
485 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
487 if (fuse_isdeadfs(vp)) {
508 if (!(dataflags & FSESS_POSIX_LOCKS))
509 return vop_stdadvlock(ap);
510 /* FUSE doesn't properly support flock until protocol 7.17 */
512 return vop_stdadvlock(ap);
514 vn_lock(vp, LK_SHARED | LK_RETRY);
516 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
520 fdisp_init(&fdi, sizeof(*fli));
522 fdisp_make_vp(&fdi, op, vp, td, cred);
524 fli->fh = fufh->fh_id;
525 fli->owner = td->td_proc->p_pid;
526 fli->lk.start = fl->l_start;
528 fli->lk.end = fl->l_start + fl->l_len - 1;
530 fli->lk.end = INT64_MAX;
531 fli->lk.type = fl->l_type;
532 fli->lk.pid = td->td_proc->p_pid;
534 err = fdisp_wait_answ(&fdi);
537 if (err == 0 && op == FUSE_GETLK) {
539 fl->l_type = flo->lk.type;
540 if (flo->lk.type != F_UNLCK) {
541 fl->l_pid = flo->lk.pid;
542 fl->l_start = flo->lk.start;
543 if (flo->lk.end == INT64_MAX)
546 fl->l_len = flo->lk.end - flo->lk.start + 1;
547 fl->l_start = flo->lk.start;
557 fuse_vnop_allocate(struct vop_allocate_args *ap)
559 struct vnode *vp = ap->a_vp;
560 off_t *len = ap->a_len;
561 off_t *offset = ap->a_offset;
562 struct ucred *cred = ap->a_cred;
563 struct fuse_filehandle *fufh;
564 struct mount *mp = vnode_mount(vp);
565 struct fuse_dispatcher fdi;
566 struct fuse_fallocate_in *ffi;
568 pid_t pid = curthread->td_proc->p_pid;
569 struct fuse_vnode_data *fvdat = VTOFUD(vp);
573 if (fuse_isdeadfs(vp))
576 switch (vp->v_type) {
581 if (vfs_isrdonly(mp))
588 if (vfs_isrdonly(mp))
591 if (fsess_not_impl(mp, FUSE_FALLOCATE))
594 io.uio_offset = *offset;
596 err = vn_rlimit_fsize(vp, &io, curthread);
600 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
604 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
606 err = fuse_vnode_size(vp, &filesize, cred, curthread);
609 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
611 fdisp_init(&fdi, sizeof(*ffi));
612 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
614 ffi->fh = fufh->fh_id;
615 ffi->offset = *offset;
618 err = fdisp_wait_answ(&fdi);
621 fsess_set_notimpl(mp, FUSE_FALLOCATE);
623 } else if (err == EOPNOTSUPP) {
625 * The file system server does not support FUSE_FALLOCATE with
626 * the supplied mode for this particular file.
632 fuse_vnode_undirty_cached_timestamps(vp, false);
633 fuse_internal_clear_suid_on_write(vp, cred, curthread);
634 if (*offset > fvdat->cached_attrs.va_size) {
635 fuse_vnode_setsize(vp, *offset, false);
636 getnanouptime(&fvdat->last_local_modify);
646 struct bufobj **a_bop;
652 fuse_vnop_bmap(struct vop_bmap_args *ap)
654 struct vnode *vp = ap->a_vp;
655 struct bufobj **bo = ap->a_bop;
656 struct thread *td = curthread;
658 struct fuse_dispatcher fdi;
659 struct fuse_bmap_in *fbi;
660 struct fuse_bmap_out *fbo;
661 struct fuse_data *data;
662 struct fuse_vnode_data *fvdat = VTOFUD(vp);
665 daddr_t lbn = ap->a_bn;
666 daddr_t *pbn = ap->a_bnp;
667 int *runp = ap->a_runp;
668 int *runb = ap->a_runb;
672 if (fuse_isdeadfs(vp)) {
676 mp = vnode_mount(vp);
677 data = fuse_get_mpdata(mp);
678 biosize = fuse_iosize(vp);
679 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
680 data->max_readahead_blocks);
686 * The FUSE_BMAP operation does not include the runp and runb
687 * variables, so we must guess. Report nonzero contiguous runs so
688 * cluster_read will combine adjacent reads. It's worthwhile to reduce
689 * upcalls even if we don't know the true physical layout of the file.
691 * FUSE file systems may opt out of read clustering in two ways:
692 * * mounting with -onoclusterr
693 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
696 *runb = MIN(lbn, maxrun);
697 if (runp != NULL && maxrun == 0)
699 else if (runp != NULL) {
701 * If the file's size is cached, use that value to calculate
702 * runp, even if the cache is expired. runp is only advisory,
703 * and the risk of getting it wrong is not worth the cost of
706 if (fvdat->cached_attrs.va_size != VNOVAL)
707 fsize = fvdat->cached_attrs.va_size;
709 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
711 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
717 if (fsess_maybe_impl(mp, FUSE_BMAP)) {
718 fdisp_init(&fdi, sizeof(*fbi));
719 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
722 fbi->blocksize = biosize;
723 error = fdisp_wait_answ(&fdi);
724 if (error == ENOSYS) {
726 fsess_set_notimpl(mp, FUSE_BMAP);
730 if (error == 0 && pbn != NULL)
737 /* If the daemon doesn't support BMAP, make up a sensible default */
739 *pbn = lbn * btodb(biosize);
744 struct vop_close_args {
747 struct ucred *a_cred;
752 fuse_vnop_close(struct vop_close_args *ap)
754 struct vnode *vp = ap->a_vp;
755 struct ucred *cred = ap->a_cred;
756 int fflag = ap->a_fflag;
757 struct thread *td = ap->a_td;
758 pid_t pid = td->td_proc->p_pid;
759 struct fuse_vnode_data *fvdat = VTOFUD(vp);
762 if (fuse_isdeadfs(vp))
766 if (fflag & IO_NDELAY)
769 err = fuse_flush(vp, cred, pid, fflag);
770 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE)) {
774 vap.va_atime = fvdat->cached_attrs.va_atime;
775 err = fuse_internal_setattr(vp, &vap, td, NULL);
777 /* TODO: close the file handle, if we're sure it's no longer used */
778 if ((fvdat->flag & FN_SIZECHANGE) != 0) {
779 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
785 struct vop_copy_file_range_args {
786 struct vop_generic_args a_gen;
787 struct vnode *a_invp;
789 struct vnode *a_outvp;
792 unsigned int a_flags;
793 struct ucred *a_incred;
794 struct ucred *a_outcred;
795 struct thread *a_fsizetd;
799 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
801 struct vnode *invp = ap->a_invp;
802 struct vnode *outvp = ap->a_outvp;
803 struct mount *mp = vnode_mount(invp);
804 struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
805 struct fuse_dispatcher fdi;
806 struct fuse_filehandle *infufh, *outfufh;
807 struct fuse_copy_file_range_in *fcfri;
808 struct ucred *incred = ap->a_incred;
809 struct ucred *outcred = ap->a_outcred;
810 struct fuse_write_out *fwo;
818 if (mp != vnode_mount(outvp))
821 if (incred->cr_uid != outcred->cr_uid)
824 if (incred->cr_groups[0] != outcred->cr_groups[0])
827 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
830 if (ap->a_fsizetd == NULL)
834 pid = td->td_proc->p_pid;
836 /* Lock both vnodes, avoiding risk of deadlock. */
838 err = vn_lock(outvp, LK_EXCLUSIVE);
842 err = vn_lock(invp, LK_SHARED | LK_NOWAIT);
846 err = vn_lock(invp, LK_SHARED);
854 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
858 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
862 io.uio_resid = *ap->a_lenp;
864 io.uio_offset = *ap->a_outoffp;
865 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
870 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
874 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
875 *ap->a_outoffp + io.uio_resid);
879 fdisp_init(&fdi, sizeof(*fcfri));
880 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
882 fcfri->fh_in = infufh->fh_id;
883 fcfri->off_in = *ap->a_inoffp;
884 fcfri->nodeid_out = VTOI(outvp);
885 fcfri->fh_out = outfufh->fh_id;
886 fcfri->off_out = *ap->a_outoffp;
887 fcfri->len = io.uio_resid;
890 err = fdisp_wait_answ(&fdi);
893 *ap->a_lenp = fwo->size;
894 *ap->a_inoffp += fwo->size;
895 *ap->a_outoffp += fwo->size;
896 fuse_internal_clear_suid_on_write(outvp, outcred, td);
897 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
898 fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
899 getnanouptime(&outfvdat->last_local_modify);
901 fuse_vnode_update(invp, FN_ATIMECHANGE);
902 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
912 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
914 err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
915 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
916 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
920 * No need to call vn_rlimit_fsizex_res before return, since the uio is
927 fdisp_make_mknod_for_fallback(
928 struct fuse_dispatcher *fdip,
929 struct componentname *cnp,
935 enum fuse_opcode *op)
937 struct fuse_mknod_in *fmni;
939 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
941 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
945 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
947 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
950 struct vnop_create_args {
952 struct vnode **a_vpp;
953 struct componentname *a_cnp;
958 fuse_vnop_create(struct vop_create_args *ap)
960 struct vnode *dvp = ap->a_dvp;
961 struct vnode **vpp = ap->a_vpp;
962 struct componentname *cnp = ap->a_cnp;
963 struct vattr *vap = ap->a_vap;
964 struct thread *td = curthread;
965 struct ucred *cred = cnp->cn_cred;
967 struct fuse_data *data;
968 struct fuse_create_in *fci;
969 struct fuse_entry_out *feo;
970 struct fuse_open_out *foo;
971 struct fuse_dispatcher fdi, fdi2;
972 struct fuse_dispatcher *fdip = &fdi;
973 struct fuse_dispatcher *fdip2 = NULL;
977 struct mount *mp = vnode_mount(dvp);
978 data = fuse_get_mpdata(mp);
979 uint64_t parentnid = VTOFUD(dvp)->nid;
980 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
984 if (fuse_isdeadfs(dvp))
987 /* FUSE expects sockets to be created with FUSE_MKNOD */
988 if (vap->va_type == VSOCK)
989 return fuse_internal_mknod(dvp, vpp, cnp, vap);
992 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
993 * writable mode makes sense, and we might as well include readability
998 bzero(&fdi, sizeof(fdi));
1000 if (vap->va_type != VREG)
1003 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1004 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
1005 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1008 /* Use FUSE_CREATE */
1012 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1013 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1016 fci->flags = O_CREAT | flags;
1017 if (fuse_libabi_geq(data, 7, 12)) {
1018 insize = sizeof(*fci);
1019 fci->umask = td->td_proc->p_pd->pd_cmask;
1021 insize = sizeof(struct fuse_open_in);
1024 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1026 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1029 err = fdisp_wait_answ(fdip);
1032 if (err == ENOSYS && op == FUSE_CREATE) {
1033 fsess_set_notimpl(mp, FUSE_CREATE);
1034 fdisp_destroy(fdip);
1035 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1036 parentnid, td, cred, mode, &op);
1037 err = fdisp_wait_answ(fdip);
1045 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1049 if (op == FUSE_CREATE) {
1050 if (fuse_libabi_geq(data, 7, 9))
1051 foo = (struct fuse_open_out*)(feo + 1);
1053 foo = (struct fuse_open_out*)((char*)feo +
1054 FUSE_COMPAT_ENTRY_OUT_SIZE);
1056 /* Issue a separate FUSE_OPEN */
1057 struct fuse_open_in *foi;
1060 fdisp_init(fdip2, sizeof(*foi));
1061 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1063 foi = fdip2->indata;
1065 err = fdisp_wait_answ(fdip2);
1070 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1072 struct fuse_release_in *fri;
1073 uint64_t nodeid = feo->nodeid;
1074 uint64_t fh_id = foo->fh;
1076 fdisp_init(fdip, sizeof(*fri));
1077 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1081 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1082 fuse_insert_message(fdip->tick, false);
1085 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1086 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1087 feo->attr_valid_nsec, NULL, true);
1089 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1090 fuse_vnode_open(*vpp, foo->open_flags, td);
1092 * Purge the parent's attribute cache because the daemon should've
1093 * updated its mtime and ctime
1095 fuse_vnode_clear_attr_cache(dvp);
1096 cache_purge_negative(dvp);
1100 fdisp_destroy(fdip2);
1101 fdisp_destroy(fdip);
1106 struct vnop_fdatasync_args {
1107 struct vop_generic_args a_gen;
1108 struct vnode * a_vp;
1109 struct thread * a_td;
1113 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1115 struct vnode *vp = ap->a_vp;
1116 struct thread *td = ap->a_td;
1117 int waitfor = MNT_WAIT;
1121 if (fuse_isdeadfs(vp)) {
1124 if ((err = vop_stdfdatasync_buf(ap)))
1127 return fuse_internal_fsync(vp, td, waitfor, true);
1131 struct vnop_fsync_args {
1132 struct vop_generic_args a_gen;
1133 struct vnode * a_vp;
1135 struct thread * a_td;
1139 fuse_vnop_fsync(struct vop_fsync_args *ap)
1141 struct vnode *vp = ap->a_vp;
1142 struct thread *td = ap->a_td;
1143 int waitfor = ap->a_waitfor;
1146 if (fuse_isdeadfs(vp)) {
1149 if ((err = vop_stdfsync(ap)))
1152 return fuse_internal_fsync(vp, td, waitfor, false);
1156 struct vnop_getattr_args {
1158 struct vattr *a_vap;
1159 struct ucred *a_cred;
1160 struct thread *a_td;
1164 fuse_vnop_getattr(struct vop_getattr_args *ap)
1166 struct vnode *vp = ap->a_vp;
1167 struct vattr *vap = ap->a_vap;
1168 struct ucred *cred = ap->a_cred;
1169 struct thread *td = curthread;
1174 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
1176 /* Note that we are not bailing out on a dead file system just yet. */
1178 if (!(dataflags & FSESS_INITED)) {
1179 if (!vnode_isvroot(vp)) {
1180 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
1187 err = fuse_internal_getattr(vp, vap, cred, td);
1188 if (err == ENOTCONN && vnode_isvroot(vp)) {
1189 /* see comment in fuse_vfsop_statfs() */
1196 bzero(vap, sizeof(*vap));
1197 vap->va_type = vnode_vtype(vp);
1203 struct vnop_inactive_args {
1208 fuse_vnop_inactive(struct vop_inactive_args *ap)
1210 struct vnode *vp = ap->a_vp;
1211 struct thread *td = curthread;
1213 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1214 struct fuse_filehandle *fufh, *fufh_tmp;
1218 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1219 if (need_flush && vp->v_type == VREG) {
1220 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1221 fuse_vnode_savesize(vp, NULL, 0);
1223 if ((fvdat->flag & FN_REVOKED) != 0)
1224 fuse_io_invalbuf(vp, td);
1226 fuse_io_flushbuf(vp, MNT_WAIT, td);
1229 fuse_filehandle_close(vp, fufh, td, NULL);
1232 if ((fvdat->flag & FN_REVOKED) != 0)
1239 struct vnop_ioctl_args {
1244 struct ucred *a_cred;
1245 struct thread *a_td;
1249 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1251 struct vnode *vp = ap->a_vp;
1252 struct mount *mp = vnode_mount(vp);
1253 struct ucred *cred = ap->a_cred;
1255 pid_t pid = ap->a_td->td_proc->p_pid;
1258 switch (ap->a_command) {
1261 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1262 if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1266 if (ap->a_command == FIOSEEKDATA)
1271 vn_lock(vp, LK_SHARED | LK_RETRY);
1272 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1276 if (fsess_not_impl(mp, FUSE_LSEEK))
1277 err = vop_stdioctl(ap);
1280 /* TODO: implement FUSE_IOCTL */
1289 struct vnop_link_args {
1290 struct vnode *a_tdvp;
1292 struct componentname *a_cnp;
1296 fuse_vnop_link(struct vop_link_args *ap)
1298 struct vnode *vp = ap->a_vp;
1299 struct vnode *tdvp = ap->a_tdvp;
1300 struct componentname *cnp = ap->a_cnp;
1302 struct vattr *vap = VTOVA(vp);
1304 struct fuse_dispatcher fdi;
1305 struct fuse_entry_out *feo;
1306 struct fuse_link_in fli;
1310 if (fuse_isdeadfs(vp)) {
1313 if (vnode_mount(tdvp) != vnode_mount(vp)) {
1318 * This is a seatbelt check to protect naive userspace filesystems from
1319 * themselves and the limitations of the FUSE IPC protocol. If a
1320 * filesystem does not allow attribute caching, assume it is capable of
1321 * validating that nlink does not overflow.
1323 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1325 fli.oldnodeid = VTOI(vp);
1327 fdisp_init(&fdi, 0);
1328 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1329 FUSE_LINK, &fli, sizeof(fli), &fdi);
1330 if ((err = fdisp_wait_answ(&fdi))) {
1335 if (fli.oldnodeid != feo->nodeid) {
1336 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1337 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE,
1338 "Assigned wrong inode for a hard link.");
1339 fuse_vnode_clear_attr_cache(vp);
1340 fuse_vnode_clear_attr_cache(tdvp);
1345 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1348 * Purge the parent's attribute cache because the daemon
1349 * should've updated its mtime and ctime
1351 fuse_vnode_clear_attr_cache(tdvp);
1352 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1353 feo->attr_valid_nsec, NULL, true);
1356 fdisp_destroy(&fdi);
1360 struct fuse_lookup_alloc_arg {
1361 struct fuse_entry_out *feo;
1362 struct componentname *cnp;
1367 /* Callback for vn_get_ino */
1369 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1371 struct fuse_lookup_alloc_arg *flaa = arg;
1373 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1377 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1378 "int", "struct timespec*", "struct timespec*");
1380 struct vnop_lookup_args {
1381 struct vnodeop_desc *a_desc;
1382 struct vnode *a_dvp;
1383 struct vnode **a_vpp;
1384 struct componentname *a_cnp;
1388 fuse_vnop_lookup(struct vop_lookup_args *ap)
1390 struct vnode *dvp = ap->a_dvp;
1391 struct vnode **vpp = ap->a_vpp;
1392 struct componentname *cnp = ap->a_cnp;
1393 struct thread *td = curthread;
1394 struct ucred *cred = cnp->cn_cred;
1395 struct timespec now;
1397 int nameiop = cnp->cn_nameiop;
1398 int flags = cnp->cn_flags;
1399 int islastcn = flags & ISLASTCN;
1400 struct mount *mp = vnode_mount(dvp);
1401 struct fuse_data *data = fuse_get_mpdata(mp);
1402 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1407 struct vnode *vp = NULL;
1409 struct fuse_dispatcher fdi;
1410 bool did_lookup = false;
1411 struct fuse_entry_out *feo = NULL;
1412 enum vtype vtyp; /* vnode type of target */
1416 if (fuse_isdeadfs(dvp)) {
1420 if (!vnode_isdir(dvp))
1423 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1426 if ((cnp->cn_flags & NOEXECCHECK) != 0)
1427 cnp->cn_flags &= ~NOEXECCHECK;
1428 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1431 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1432 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT))
1434 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1436 * Since the file system doesn't support ".." lookups,
1437 * we have no way to find this entry.
1441 nid = VTOFUD(dvp)->parent_nid;
1444 /* .. is obviously a directory */
1446 } else if (is_dot) {
1448 /* . is obviously a directory */
1451 struct timespec timeout;
1452 int ncpticks; /* here to accommodate for API contract */
1454 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1455 getnanouptime(&now);
1456 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1458 case -1: /* positive match */
1459 if (timespeccmp(&timeout, &now, >)) {
1460 counter_u64_add(fuse_lookup_cache_hits, 1);
1463 counter_u64_add(fuse_lookup_cache_misses, 1);
1465 &VTOFUD(*vpp)->entry_cache_timeout);
1476 case 0: /* no match in cache */
1477 counter_u64_add(fuse_lookup_cache_misses, 1);
1480 case ENOENT: /* negative match */
1481 if (timespeccmp(&timeout, &now, <=)) {
1483 cache_purge_negative(dvp);
1491 fdisp_init(&fdi, cnp->cn_namelen + 1);
1492 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1494 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1495 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1496 lookup_err = fdisp_wait_answ(&fdi);
1500 /* lookup call succeeded */
1501 feo = (struct fuse_entry_out *)fdi.answ;
1504 /* zero nodeid means ENOENT and cache it */
1505 struct timespec timeout;
1507 fdi.answ_stat = ENOENT;
1508 lookup_err = ENOENT;
1509 if (cnp->cn_flags & MAKEENTRY) {
1510 fuse_validity_2_timespec(feo, &timeout);
1511 /* Use the same entry_time for .. as for
1512 * the file itself. That doesn't honor
1513 * exactly what the fuse server tells
1514 * us, but to do otherwise would require
1515 * another cache lookup at this point.
1517 struct timespec *dtsp = NULL;
1518 cache_enter_time(dvp, *vpp, cnp,
1522 vtyp = IFTOVT(feo->attr.mode);
1524 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1525 fdisp_destroy(&fdi);
1529 /* lookup_err, if non-zero, must be ENOENT at this point */
1532 /* Entry not found */
1533 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1534 if (default_permissions)
1535 err = fuse_internal_access(dvp, VWRITE, td,
1546 /* Entry was found */
1547 if (flags & ISDOTDOT) {
1548 struct fuse_lookup_alloc_arg flaa;
1554 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1557 } else if (nid == VTOI(dvp)) {
1562 fuse_warn(fuse_get_mpdata(mp),
1563 FSESS_WARN_ILLEGAL_INODE,
1564 "Assigned same inode to both parent and "
1570 struct fuse_vnode_data *fvdat;
1572 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1580 if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1582 * Attributes from the server are definitely
1583 * newer than the last attributes we sent to
1584 * the server, so cache them.
1586 fuse_internal_cache_attrs(*vpp, &feo->attr,
1587 feo->attr_valid, feo->attr_valid_nsec,
1590 fuse_validity_2_bintime(feo->entry_valid,
1591 feo->entry_valid_nsec,
1592 &fvdat->entry_cache_timeout);
1594 if ((nameiop == DELETE || nameiop == RENAME) &&
1595 islastcn && default_permissions)
1597 struct vattr dvattr;
1599 err = fuse_internal_access(dvp, VWRITE, td,
1604 * if the parent's sticky bit is set, check
1605 * whether we're allowed to remove the file.
1606 * Need to figure out the vnode locking to make
1609 fuse_internal_getattr(dvp, &dvattr, cred, td);
1610 if ((dvattr.va_mode & S_ISTXT) &&
1611 fuse_internal_access(dvp, VADMIN, td,
1613 fuse_internal_access(*vpp, VADMIN, td,
1623 if (vp != NULL && dvp != vp)
1625 else if (vp != NULL)
1630 fdisp_destroy(&fdi);
1636 struct vnop_mkdir_args {
1637 struct vnode *a_dvp;
1638 struct vnode **a_vpp;
1639 struct componentname *a_cnp;
1640 struct vattr *a_vap;
1644 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1646 struct vnode *dvp = ap->a_dvp;
1647 struct vnode **vpp = ap->a_vpp;
1648 struct componentname *cnp = ap->a_cnp;
1649 struct vattr *vap = ap->a_vap;
1651 struct fuse_mkdir_in fmdi;
1653 if (fuse_isdeadfs(dvp)) {
1656 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1657 fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1659 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1660 sizeof(fmdi), VDIR));
1664 struct vnop_mknod_args {
1665 struct vnode *a_dvp;
1666 struct vnode **a_vpp;
1667 struct componentname *a_cnp;
1668 struct vattr *a_vap;
1672 fuse_vnop_mknod(struct vop_mknod_args *ap)
1675 struct vnode *dvp = ap->a_dvp;
1676 struct vnode **vpp = ap->a_vpp;
1677 struct componentname *cnp = ap->a_cnp;
1678 struct vattr *vap = ap->a_vap;
1680 if (fuse_isdeadfs(dvp))
1683 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1687 struct vop_open_args {
1690 struct ucred *a_cred;
1691 struct thread *a_td;
1692 int a_fdidx; / struct file *a_fp;
1696 fuse_vnop_open(struct vop_open_args *ap)
1698 struct vnode *vp = ap->a_vp;
1699 int a_mode = ap->a_mode;
1700 struct thread *td = ap->a_td;
1701 struct ucred *cred = ap->a_cred;
1702 pid_t pid = td->td_proc->p_pid;
1704 if (fuse_isdeadfs(vp))
1706 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1707 return (EOPNOTSUPP);
1708 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1711 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1712 fuse_vnode_open(vp, 0, td);
1716 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1720 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1722 struct vnode *vp = ap->a_vp;
1725 switch (ap->a_name) {
1726 case _PC_FILESIZEBITS:
1730 *ap->a_retval = NAME_MAX;
1733 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1735 case _PC_SYMLINK_MAX:
1736 *ap->a_retval = MAXPATHLEN;
1741 case _PC_MIN_HOLE_SIZE:
1743 * The FUSE protocol provides no mechanism for a server to
1744 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead,
1745 * return EINVAL if the server does not support FUSE_LSEEK, or
1748 mp = vnode_mount(vp);
1749 if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1750 !fsess_not_impl(mp, FUSE_LSEEK)) {
1753 /* Issue a FUSE_LSEEK to find out if it's implemented */
1754 fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred,
1755 curthread->td_proc->p_pid, &offset, SEEK_DATA);
1758 if (fsess_is_impl(mp, FUSE_LSEEK)) {
1763 * Probably FUSE_LSEEK is not implemented. It might
1764 * be, if the FUSE_LSEEK above returned an error like
1765 * EACCES, but in that case we can't tell, so it's
1766 * safest to report EINVAL anyway.
1771 return (vop_stdpathconf(ap));
1775 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1776 "struct uio*", "struct ucred*");
1778 struct vnop_read_args {
1782 struct ucred *a_cred;
1786 fuse_vnop_read(struct vop_read_args *ap)
1788 struct vnode *vp = ap->a_vp;
1789 struct uio *uio = ap->a_uio;
1790 int ioflag = ap->a_ioflag;
1791 struct ucred *cred = ap->a_cred;
1792 pid_t pid = curthread->td_proc->p_pid;
1793 struct fuse_filehandle *fufh;
1795 bool closefufh = false, directio;
1797 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1799 if (fuse_isdeadfs(vp)) {
1803 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1804 ioflag |= IO_DIRECT;
1807 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1808 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1810 * nfsd will do I/O without first doing VOP_OPEN. We
1811 * must implicitly open the file here
1813 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1817 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1822 * Ideally, when the daemon asks for direct io at open time, the
1823 * standard file flag should be set according to this, so that would
1824 * just change the default mode, which later on could be changed via
1826 * But this doesn't work, the O_DIRECT flag gets cleared at some point
1827 * (don't know where). So to make any use of the Fuse direct_io option,
1828 * we hardwire it into the file's private data (similarly to Linux,
1831 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1833 fuse_vnode_update(vp, FN_ATIMECHANGE);
1835 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1836 err = fuse_read_directbackend(vp, uio, cred, fufh);
1838 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1839 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1843 fuse_filehandle_close(vp, fufh, curthread, cred);
1849 struct vnop_readdir_args {
1852 struct ucred *a_cred;
1855 uint64_t **a_cookies;
1859 fuse_vnop_readdir(struct vop_readdir_args *ap)
1861 struct vnode *vp = ap->a_vp;
1862 struct uio *uio = ap->a_uio;
1863 struct ucred *cred = ap->a_cred;
1864 struct fuse_filehandle *fufh = NULL;
1865 struct mount *mp = vnode_mount(vp);
1866 struct fuse_iov cookediov;
1871 bool closefufh = false;
1872 pid_t pid = curthread->td_proc->p_pid;
1876 if (fuse_isdeadfs(vp)) {
1879 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1880 (uio_resid(uio) < sizeof(struct dirent))) {
1884 tresid = uio->uio_resid;
1885 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1886 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1887 KASSERT(fuse_get_mpdata(mp)->dataflags
1888 & FSESS_NO_OPENDIR_SUPPORT,
1889 ("FUSE file systems that don't set "
1890 "FUSE_NO_OPENDIR_SUPPORT should not be exported"));
1892 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1893 * must implicitly open the directory here.
1895 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1900 if (ap->a_ncookies != NULL) {
1901 ncookies = uio->uio_resid /
1902 (offsetof(struct dirent, d_name) + 4) + 1;
1903 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1904 *ap->a_ncookies = ncookies;
1905 *ap->a_cookies = cookies;
1910 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1911 fiov_init(&cookediov, DIRCOOKEDSIZE);
1913 err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1914 &ncookies, cookies);
1916 fiov_teardown(&cookediov);
1918 fuse_filehandle_close(vp, fufh, curthread, cred);
1920 if (ap->a_ncookies != NULL) {
1922 *ap->a_ncookies -= ncookies;
1924 free(*ap->a_cookies, M_TEMP);
1925 *ap->a_ncookies = 0;
1926 *ap->a_cookies = NULL;
1929 if (err == 0 && tresid == uio->uio_resid)
1936 struct vnop_readlink_args {
1939 struct ucred *a_cred;
1943 fuse_vnop_readlink(struct vop_readlink_args *ap)
1945 struct vnode *vp = ap->a_vp;
1946 struct uio *uio = ap->a_uio;
1947 struct ucred *cred = ap->a_cred;
1949 struct fuse_dispatcher fdi;
1952 if (fuse_isdeadfs(vp)) {
1955 if (!vnode_islnk(vp)) {
1958 fdisp_init(&fdi, 0);
1959 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1963 if (((char *)fdi.answ)[0] == '/' &&
1964 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1965 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1967 err = uiomove(mpth, strlen(mpth), uio);
1970 err = uiomove(fdi.answ, fdi.iosize, uio);
1973 fdisp_destroy(&fdi);
1978 struct vnop_reclaim_args {
1983 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1985 struct vnode *vp = ap->a_vp;
1986 struct thread *td = curthread;
1987 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1988 struct fuse_filehandle *fufh, *fufh_tmp;
1991 panic("FUSE: no vnode data during recycling");
1993 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1994 printf("FUSE: vnode being reclaimed with open fufh "
1995 "(type=%#x)", fufh->fufh_type);
1996 fuse_filehandle_close(vp, fufh, td, NULL);
1999 if (VTOI(vp) == 1) {
2001 * Don't send FUSE_FORGET for the root inode, because
2002 * we never send FUSE_LOOKUP for it (see
2003 * fuse_vfsop_root) and we don't want the server to see
2004 * mismatched lookup counts.
2006 struct fuse_data *data;
2007 struct vnode *vroot;
2009 data = fuse_get_mpdata(vnode_mount(vp));
2011 vroot = data->vroot;
2016 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2017 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2021 vfs_hash_remove(vp);
2022 fuse_vnode_destroy(vp);
2028 struct vnop_remove_args {
2029 struct vnode *a_dvp;
2031 struct componentname *a_cnp;
2035 fuse_vnop_remove(struct vop_remove_args *ap)
2037 struct vnode *dvp = ap->a_dvp;
2038 struct vnode *vp = ap->a_vp;
2039 struct componentname *cnp = ap->a_cnp;
2043 if (fuse_isdeadfs(vp)) {
2046 if (vnode_isdir(vp)) {
2050 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2056 struct vnop_rename_args {
2057 struct vnode *a_fdvp;
2058 struct vnode *a_fvp;
2059 struct componentname *a_fcnp;
2060 struct vnode *a_tdvp;
2061 struct vnode *a_tvp;
2062 struct componentname *a_tcnp;
2066 fuse_vnop_rename(struct vop_rename_args *ap)
2068 struct vnode *fdvp = ap->a_fdvp;
2069 struct vnode *fvp = ap->a_fvp;
2070 struct componentname *fcnp = ap->a_fcnp;
2071 struct vnode *tdvp = ap->a_tdvp;
2072 struct vnode *tvp = ap->a_tvp;
2073 struct componentname *tcnp = ap->a_tcnp;
2074 struct fuse_data *data;
2075 bool newparent = fdvp != tdvp;
2076 bool isdir = fvp->v_type == VDIR;
2079 if (fuse_isdeadfs(fdvp)) {
2082 if (fvp->v_mount != tdvp->v_mount ||
2083 (tvp && fvp->v_mount != tvp->v_mount)) {
2084 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2091 * FUSE library is expected to check if target directory is not
2092 * under the source directory in the file system tree.
2093 * Linux performs this check at VFS level.
2096 * If source is a directory, and it will get a new parent, user must
2097 * have write permission to it, so ".." can be modified.
2099 data = fuse_get_mpdata(vnode_mount(tdvp));
2100 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2101 err = fuse_internal_access(fvp, VWRITE,
2102 curthread, tcnp->cn_cred);
2106 sx_xlock(&data->rename_lock);
2107 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2110 fuse_vnode_setparent(fvp, tdvp);
2112 fuse_vnode_setparent(tvp, NULL);
2114 sx_unlock(&data->rename_lock);
2116 if (tvp != NULL && tvp != fvp) {
2119 if (vnode_isdir(fvp)) {
2120 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2141 struct vnop_rmdir_args {
2142 struct vnode *a_dvp;
2144 struct componentname *a_cnp;
2148 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2150 struct vnode *dvp = ap->a_dvp;
2151 struct vnode *vp = ap->a_vp;
2155 if (fuse_isdeadfs(vp)) {
2158 if (VTOFUD(vp) == VTOFUD(dvp)) {
2161 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2167 struct vnop_setattr_args {
2169 struct vattr *a_vap;
2170 struct ucred *a_cred;
2171 struct thread *a_td;
2175 fuse_vnop_setattr(struct vop_setattr_args *ap)
2177 struct vnode *vp = ap->a_vp;
2178 struct vattr *vap = ap->a_vap;
2179 struct ucred *cred = ap->a_cred;
2180 struct thread *td = curthread;
2182 struct fuse_data *data;
2183 struct vattr old_va;
2186 accmode_t accmode = 0;
2188 bool drop_suid = false;
2191 mp = vnode_mount(vp);
2192 data = fuse_get_mpdata(mp);
2193 dataflags = data->dataflags;
2194 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2195 if (cred->cr_ngroups > 0)
2196 cr_gid = cred->cr_groups[0];
2200 if (fuse_isdeadfs(vp)) {
2204 if (vap->va_uid != (uid_t)VNOVAL) {
2206 /* Only root may change a file's owner */
2207 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2209 /* As a special case, allow the null chown */
2210 err2 = fuse_internal_getattr(vp, &old_va, cred,
2214 if (vap->va_uid != old_va.va_uid)
2224 if (vap->va_gid != (gid_t)VNOVAL) {
2225 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2227 if (checkperm && !groupmember(vap->va_gid, cred))
2230 * Non-root users may only chgrp to one of their own
2233 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2235 /* As a special case, allow the null chgrp */
2236 err2 = fuse_internal_getattr(vp, &old_va, cred,
2240 if (vap->va_gid != old_va.va_gid)
2248 if (vap->va_size != VNOVAL) {
2249 switch (vp->v_type) {
2254 if (vfs_isrdonly(mp))
2256 err = vn_rlimit_trunc(vap->va_size, td);
2262 * According to POSIX, the result is unspecified
2263 * for file types other than regular files,
2264 * directories and shared memory objects. We
2265 * don't support shared memory objects in the file
2266 * system, and have dubious support for truncating
2267 * symlinks. Just ignore the request in other cases.
2271 /* Don't set accmode. Permission to trunc is checked upstack */
2273 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2274 if (vap->va_vaflags & VA_UTIMES_NULL)
2280 if (vap->va_mode != (mode_t)VNOVAL)
2281 vap->va_mode &= ~(S_ISUID | S_ISGID);
2283 err = fuse_internal_getattr(vp, &old_va, cred, td);
2286 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2289 if (vap->va_mode != (mode_t)VNOVAL) {
2290 /* Only root may set the sticky bit on non-directories */
2291 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2292 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2294 if (checkperm && (vap->va_mode & S_ISGID)) {
2295 err = fuse_internal_getattr(vp, &old_va, cred, td);
2298 if (!groupmember(old_va.va_gid, cred)) {
2299 err = priv_check_cred(cred, PRIV_VFS_SETGID);
2307 if (vfs_isrdonly(mp))
2311 err = fuse_internal_access(vp, accmode, td, cred);
2318 return fuse_internal_setattr(vp, vap, td, cred);
2322 struct vnop_strategy_args {
2328 fuse_vnop_strategy(struct vop_strategy_args *ap)
2330 struct vnode *vp = ap->a_vp;
2331 struct buf *bp = ap->a_bp;
2333 if (!vp || fuse_isdeadfs(vp)) {
2334 bp->b_ioflags |= BIO_ERROR;
2335 bp->b_error = ENXIO;
2341 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2342 * fuse_io_strategy sets bp's error fields
2344 (void)fuse_io_strategy(vp, bp);
2350 struct vnop_symlink_args {
2351 struct vnode *a_dvp;
2352 struct vnode **a_vpp;
2353 struct componentname *a_cnp;
2354 struct vattr *a_vap;
2359 fuse_vnop_symlink(struct vop_symlink_args *ap)
2361 struct vnode *dvp = ap->a_dvp;
2362 struct vnode **vpp = ap->a_vpp;
2363 struct componentname *cnp = ap->a_cnp;
2364 const char *target = ap->a_target;
2366 struct fuse_dispatcher fdi;
2371 if (fuse_isdeadfs(dvp)) {
2375 * Unlike the other creator type calls, here we have to create a message
2376 * where the name of the new entry comes first, and the data describing
2377 * the entry comes second.
2378 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2379 * but put together the message manually and just call the core part.
2382 len = strlen(target) + 1;
2383 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2384 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2386 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2387 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2388 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2390 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2391 fdisp_destroy(&fdi);
2396 struct vnop_write_args {
2400 struct ucred *a_cred;
2404 fuse_vnop_write(struct vop_write_args *ap)
2406 struct vnode *vp = ap->a_vp;
2407 struct uio *uio = ap->a_uio;
2408 int ioflag = ap->a_ioflag;
2409 struct ucred *cred = ap->a_cred;
2410 pid_t pid = curthread->td_proc->p_pid;
2411 struct fuse_filehandle *fufh;
2413 bool closefufh = false, directio;
2415 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2417 if (fuse_isdeadfs(vp)) {
2421 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2422 ioflag |= IO_DIRECT;
2425 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2426 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2428 * nfsd will do I/O without first doing VOP_OPEN. We
2429 * must implicitly open the file here
2431 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2435 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2440 * Ideally, when the daemon asks for direct io at open time, the
2441 * standard file flag should be set according to this, so that would
2442 * just change the default mode, which later on could be changed via
2444 * But this doesn't work, the O_DIRECT flag gets cleared at some point
2445 * (don't know where). So to make any use of the Fuse direct_io option,
2446 * we hardwire it into the file's private data (similarly to Linux,
2449 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2451 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2453 off_t start, end, filesize;
2454 bool pages = (ioflag & IO_VMIO) != 0;
2456 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2458 err = fuse_vnode_size(vp, &filesize, cred, curthread);
2462 start = uio->uio_offset;
2463 end = start + uio->uio_resid;
2465 err = fuse_inval_buf_range(vp, filesize, start,
2470 err = fuse_write_directbackend(vp, uio, cred, fufh,
2471 filesize, ioflag, pages);
2473 SDT_PROBE2(fusefs, , vnops, trace, 1,
2474 "buffered write of vnode");
2475 if (!fsess_opt_writeback(vnode_mount(vp)))
2477 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2479 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2483 fuse_filehandle_close(vp, fufh, curthread, cred);
2489 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2491 const int biosize = fuse_iosize(vp);
2493 return (off / biosize);
2497 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2501 const int biosize = fuse_iosize(vp);
2503 err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2505 /* This will turn into a SIGBUS */
2507 } else if ((off_t)lbn * biosize >= filesize) {
2509 } else if ((off_t)(lbn + 1) * biosize > filesize) {
2510 *blksz = filesize - (off_t)lbn *biosize;
2518 struct vnop_getpages_args {
2526 fuse_vnop_getpages(struct vop_getpages_args *ap)
2528 struct vnode *vp = ap->a_vp;
2530 if (!fsess_opt_mmap(vnode_mount(vp))) {
2531 SDT_PROBE2(fusefs, , vnops, trace, 1,
2532 "called on non-cacheable vnode??\n");
2533 return (VM_PAGER_ERROR);
2536 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2537 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2540 static const char extattr_namespace_separator = '.';
2543 struct vop_getextattr_args {
2544 struct vop_generic_args a_gen;
2546 int a_attrnamespace;
2550 struct ucred *a_cred;
2551 struct thread *a_td;
2555 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2557 struct vnode *vp = ap->a_vp;
2558 struct uio *uio = ap->a_uio;
2559 struct fuse_dispatcher fdi;
2560 struct fuse_getxattr_in *get_xattr_in;
2561 struct fuse_getxattr_out *get_xattr_out;
2562 struct mount *mp = vnode_mount(vp);
2563 struct thread *td = ap->a_td;
2564 struct ucred *cred = ap->a_cred;
2570 if (fuse_isdeadfs(vp))
2573 if (fsess_not_impl(mp, FUSE_GETXATTR))
2576 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2580 /* Default to looking for user attributes. */
2581 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2582 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2584 prefix = EXTATTR_NAMESPACE_USER_STRING;
2586 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2587 strlen(ap->a_name) + 1;
2589 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2590 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2592 get_xattr_in = fdi.indata;
2594 * Check to see whether we're querying the available size or
2595 * issuing the actual request. If we pass in 0, we get back struct
2596 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2597 * that much data, without the struct fuse_getxattr_out header.
2600 get_xattr_in->size = 0;
2602 get_xattr_in->size = uio->uio_resid;
2604 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2605 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2608 err = fdisp_wait_answ(&fdi);
2610 if (err == ENOSYS) {
2611 fsess_set_notimpl(mp, FUSE_GETXATTR);
2617 get_xattr_out = fdi.answ;
2619 if (ap->a_size != NULL)
2620 *ap->a_size = get_xattr_out->size;
2623 err = uiomove(fdi.answ, fdi.iosize, uio);
2626 fdisp_destroy(&fdi);
2631 struct vop_setextattr_args {
2632 struct vop_generic_args a_gen;
2634 int a_attrnamespace;
2637 struct ucred *a_cred;
2638 struct thread *a_td;
2642 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2644 struct vnode *vp = ap->a_vp;
2645 struct uio *uio = ap->a_uio;
2646 struct fuse_dispatcher fdi;
2647 struct fuse_setxattr_in *set_xattr_in;
2648 struct mount *mp = vnode_mount(vp);
2649 struct thread *td = ap->a_td;
2650 struct ucred *cred = ap->a_cred;
2656 if (fuse_isdeadfs(vp))
2659 if (fsess_not_impl(mp, FUSE_SETXATTR))
2662 if (vfs_isrdonly(mp))
2665 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2666 if (ap->a_uio == NULL) {
2668 * If we got here as fallback from VOP_DELETEEXTATTR, then
2669 * return EOPNOTSUPP.
2671 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2672 return (EOPNOTSUPP);
2677 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2682 /* Default to looking for user attributes. */
2683 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2684 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2686 prefix = EXTATTR_NAMESPACE_USER_STRING;
2688 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2689 strlen(ap->a_name) + 1;
2691 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2692 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2694 set_xattr_in = fdi.indata;
2695 set_xattr_in->size = uio->uio_resid;
2697 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2698 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2701 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2702 uio->uio_resid, uio);
2707 err = fdisp_wait_answ(&fdi);
2709 if (err == ENOSYS) {
2710 fsess_set_notimpl(mp, FUSE_SETXATTR);
2713 if (err == ERESTART) {
2714 /* Can't restart after calling uiomove */
2719 fdisp_destroy(&fdi);
2724 * The Linux / FUSE extended attribute list is simply a collection of
2725 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2726 * byte length followed by a non-NUL terminated string. So, this allows
2727 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2728 * Linux attribute names are reported with the namespace as a prefix (e.g.
2729 * "user.attribute_name"), but in FreeBSD they are reported without the
2730 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2732 * user.attr_name1\0user.attr_name2\0
2736 * <num>attr_name1<num>attr_name2
2738 * Where "<num>" is a single byte number of characters in the attribute name.
2741 * prefix - exattr namespace prefix string
2742 * list, list_len - input list with namespace prefixes
2743 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2746 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2747 char *bsd_list, int *bsd_list_len)
2749 int len, pos, dist_to_next, prefix_len;
2753 prefix_len = strlen(prefix);
2755 while (pos < list_len && list[pos] != '\0') {
2756 dist_to_next = strlen(&list[pos]) + 1;
2757 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2758 list[pos + prefix_len] == extattr_namespace_separator) {
2759 len = dist_to_next -
2760 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2761 if (len >= EXTATTR_MAXNAMELEN)
2762 return (ENAMETOOLONG);
2764 bsd_list[*bsd_list_len] = len;
2765 memcpy(&bsd_list[*bsd_list_len + 1],
2766 &list[pos + prefix_len +
2767 sizeof(extattr_namespace_separator)], len);
2769 *bsd_list_len += len + 1;
2772 pos += dist_to_next;
2779 * List extended attributes
2781 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2782 * has a number of differences compared to its FreeBSD equivalent,
2783 * extattr_list_file:
2785 * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2786 * whereas listxattr(2) only returns attributes for a single namespace
2787 * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2788 * - If the provided buffer is not large enough to hold the result,
2789 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2790 * return as many results as will fit.
2793 struct vop_listextattr_args {
2794 struct vop_generic_args a_gen;
2796 int a_attrnamespace;
2799 struct ucred *a_cred;
2800 struct thread *a_td;
2804 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2806 struct vnode *vp = ap->a_vp;
2807 struct uio *uio = ap->a_uio;
2808 struct fuse_dispatcher fdi;
2809 struct fuse_listxattr_in *list_xattr_in;
2810 struct fuse_listxattr_out *list_xattr_out;
2811 struct mount *mp = vnode_mount(vp);
2812 struct thread *td = ap->a_td;
2813 struct ucred *cred = ap->a_cred;
2815 char *bsd_list = NULL;
2821 if (fuse_isdeadfs(vp))
2824 if (fsess_not_impl(mp, FUSE_LISTXATTR))
2827 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2832 * Add space for a NUL and the period separator if enabled.
2833 * Default to looking for user attributes.
2835 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2836 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2838 prefix = EXTATTR_NAMESPACE_USER_STRING;
2840 fdisp_init(&fdi, sizeof(*list_xattr_in));
2841 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2844 * Retrieve Linux / FUSE compatible list size.
2846 list_xattr_in = fdi.indata;
2847 list_xattr_in->size = 0;
2849 err = fdisp_wait_answ(&fdi);
2851 if (err == ENOSYS) {
2852 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2858 list_xattr_out = fdi.answ;
2859 linux_list_len = list_xattr_out->size;
2860 if (linux_list_len == 0) {
2861 if (ap->a_size != NULL)
2862 *ap->a_size = linux_list_len;
2867 * Retrieve Linux / FUSE compatible list values.
2869 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2870 list_xattr_in = fdi.indata;
2871 list_xattr_in->size = linux_list_len;
2873 err = fdisp_wait_answ(&fdi);
2874 if (err == ERANGE) {
2876 * Race detected. The attribute list must've grown since the
2877 * first FUSE_LISTXATTR call. Start over. Go all the way back
2878 * to userland so we can process signals, if necessary, before
2883 } else if (err != 0)
2886 linux_list = fdi.answ;
2887 /* FUSE doesn't allow the server to return more data than requested */
2888 if (fdi.iosize > linux_list_len) {
2889 struct fuse_data *data = fuse_get_mpdata(mp);
2891 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2893 "more extended attribute data than requested; "
2894 "should've returned ERANGE instead.");
2896 /* But returning less data is fine */
2897 linux_list_len = fdi.iosize;
2901 * Retrieve the BSD compatible list values.
2902 * The Linux / FUSE attribute list format isn't the same
2903 * as FreeBSD's format. So we need to transform it into
2904 * FreeBSD's format before giving it to the user.
2906 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2907 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2908 bsd_list, &bsd_list_len);
2912 if (ap->a_size != NULL)
2913 *ap->a_size = bsd_list_len;
2916 err = uiomove(bsd_list, bsd_list_len, uio);
2919 free(bsd_list, M_TEMP);
2920 fdisp_destroy(&fdi);
2925 struct vop_deallocate_args {
2926 struct vop_generic_args a_gen;
2932 struct ucred *a_cred;
2936 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
2938 struct vnode *vp = ap->a_vp;
2939 struct mount *mp = vnode_mount(vp);
2940 struct fuse_filehandle *fufh;
2941 struct fuse_dispatcher fdi;
2942 struct fuse_fallocate_in *ffi;
2943 struct ucred *cred = ap->a_cred;
2944 pid_t pid = curthread->td_proc->p_pid;
2945 off_t *len = ap->a_len;
2946 off_t *offset = ap->a_offset;
2947 int ioflag = ap->a_ioflag;
2950 bool closefufh = false;
2952 if (fuse_isdeadfs(vp))
2955 if (vfs_isrdonly(mp))
2958 if (fsess_not_impl(mp, FUSE_FALLOCATE))
2961 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2962 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2964 * nfsd will do I/O without first doing VOP_OPEN. We
2965 * must implicitly open the file here
2967 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2973 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2975 err = fuse_vnode_size(vp, &filesize, cred, curthread);
2978 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
2980 fdisp_init(&fdi, sizeof(*ffi));
2981 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
2983 ffi->fh = fufh->fh_id;
2984 ffi->offset = *offset;
2987 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
2988 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
2990 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
2991 err = fdisp_wait_answ(&fdi);
2993 if (err == ENOSYS) {
2994 fsess_set_notimpl(mp, FUSE_FALLOCATE);
2996 } else if (err == EOPNOTSUPP) {
2998 * The file system server does not support FUSE_FALLOCATE with
2999 * the supplied mode for this particular file.
3004 * Clip the returned offset to EoF. Do it here rather than
3005 * before FUSE_FALLOCATE just in case the kernel's cached file
3006 * size is out of date. Unfortunately, FUSE does not return
3007 * any information about filesize from that operation.
3009 *offset = MIN(*offset + *len, filesize);
3011 fuse_vnode_undirty_cached_timestamps(vp, false);
3012 fuse_internal_clear_suid_on_write(vp, cred, curthread);
3014 if (ioflag & IO_SYNC)
3015 err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3021 fuse_filehandle_close(vp, fufh, curthread, cred);
3027 fuse_filehandle_close(vp, fufh, curthread, cred);
3029 return (vop_stddeallocate(ap));
3033 struct vop_deleteextattr_args {
3034 struct vop_generic_args a_gen;
3036 int a_attrnamespace;
3038 struct ucred *a_cred;
3039 struct thread *a_td;
3043 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3045 struct vnode *vp = ap->a_vp;
3046 struct fuse_dispatcher fdi;
3047 struct mount *mp = vnode_mount(vp);
3048 struct thread *td = ap->a_td;
3049 struct ucred *cred = ap->a_cred;
3055 if (fuse_isdeadfs(vp))
3058 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3061 if (vfs_isrdonly(mp))
3064 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3069 /* Default to looking for user attributes. */
3070 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3071 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3073 prefix = EXTATTR_NAMESPACE_USER_STRING;
3075 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3076 strlen(ap->a_name) + 1;
3078 fdisp_init(&fdi, len);
3079 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3081 attr_str = fdi.indata;
3082 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3085 err = fdisp_wait_answ(&fdi);
3086 if (err == ENOSYS) {
3087 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3091 fdisp_destroy(&fdi);
3096 struct vnop_print_args {
3101 fuse_vnop_print(struct vop_print_args *ap)
3103 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3105 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3106 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3107 (uintmax_t)fvdat->nlookup,
3114 * Get an NFS filehandle for a FUSE file.
3116 * This will only work for FUSE file systems that guarantee the uniqueness of
3117 * nodeid:generation, which most don't.
3121 IN struct vnode *a_vp;
3122 IN struct fid *a_fhp;
3126 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3128 struct vnode *vp = ap->a_vp;
3129 struct fuse_vnode_data *fvdat = VTOFUD(vp);
3130 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3131 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3132 "FUSE fid type is too big");
3133 struct mount *mp = vnode_mount(vp);
3134 struct fuse_data *data = fuse_get_mpdata(mp);
3138 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3139 /* NFS requires lookups for "." and ".." */
3140 SDT_PROBE2(fusefs, , vnops, trace, 1,
3141 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3144 if ((mp->mnt_flag & MNT_EXPORTED) &&
3145 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT))
3148 * NFS is stateless, so nfsd must reopen a directory on every
3149 * call to VOP_READDIR, passing in the d_off field from the
3150 * final dirent of the previous invocation. But without
3151 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not
3152 * guarantee that d_off will be valid after a directory is
3153 * closed and reopened. So prohibit exporting FUSE file
3154 * systems that don't set that flag.
3156 * But userspace NFS servers don't have this problem.
3158 SDT_PROBE2(fusefs, , vnops, trace, 1,
3159 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT");
3163 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3167 /*ip = VTOI(ap->a_vp);*/
3168 /*ufhp = (struct ufid *)ap->a_fhp;*/
3169 fhp->len = sizeof(struct fuse_fid);
3170 fhp->nid = fvdat->nid;
3171 if (fvdat->generation <= UINT32_MAX)
3172 fhp->gen = fvdat->generation;