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 /* Check permission for extattr operations, much like extattr_check_cred */
234 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
235 struct thread *td, accmode_t accmode)
237 struct mount *mp = vnode_mount(vp);
238 struct fuse_data *data = fuse_get_mpdata(mp);
239 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
242 * Kernel-invoked always succeeds.
248 * Do not allow privileged processes in jail to directly manipulate
252 case EXTATTR_NAMESPACE_SYSTEM:
253 if (default_permissions) {
254 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
257 case EXTATTR_NAMESPACE_USER:
258 if (default_permissions) {
259 return (fuse_internal_access(vp, accmode, td, cred));
267 /* Get a filehandle for a directory */
269 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
270 struct ucred *cred, pid_t pid)
272 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
274 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
277 /* Send FUSE_FLUSH for this vnode */
279 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
281 struct fuse_flush_in *ffi;
282 struct fuse_filehandle *fufh;
283 struct fuse_dispatcher fdi;
284 struct thread *td = curthread;
285 struct mount *mp = vnode_mount(vp);
288 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
291 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
295 fdisp_init(&fdi, sizeof(*ffi));
296 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
298 ffi->fh = fufh->fh_id;
300 * If the file has a POSIX lock then we're supposed to set lock_owner.
301 * If not, then lock_owner is undefined. So we may as well always set
304 ffi->lock_owner = td->td_proc->p_pid;
306 err = fdisp_wait_answ(&fdi);
308 fsess_set_notimpl(mp, FUSE_FLUSH);
315 /* Close wrapper for fifos. */
317 fuse_fifo_close(struct vop_close_args *ap)
319 return (fifo_specops.vop_close(ap));
322 /* Invalidate a range of cached data, whether dirty of not */
324 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
327 daddr_t left_lbn, end_lbn, right_lbn;
329 int iosize, left_on, right_on, right_blksize;
331 iosize = fuse_iosize(vp);
332 left_lbn = start / iosize;
333 end_lbn = howmany(end, iosize);
334 left_on = start & (iosize - 1);
336 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
337 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
339 * Flush the dirty buffer, because we don't have a
340 * byte-granular way to record which parts of the
345 return (bp->b_error);
350 right_on = end & (iosize - 1);
352 right_lbn = end / iosize;
353 new_filesize = MAX(filesize, end);
354 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
355 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
356 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
358 * Flush the dirty buffer, because we don't have a
359 * byte-granular way to record which parts of the
364 return (bp->b_error);
370 v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
375 /* Send FUSE_LSEEK for this node */
377 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
378 pid_t pid, off_t *offp, int whence)
380 struct fuse_dispatcher fdi;
381 struct fuse_filehandle *fufh;
382 struct fuse_lseek_in *flsi;
383 struct fuse_lseek_out *flso;
384 struct mount *mp = vnode_mount(vp);
387 ASSERT_VOP_LOCKED(vp, __func__);
389 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
392 fdisp_init(&fdi, sizeof(*flsi));
393 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
395 flsi->fh = fufh->fh_id;
396 flsi->offset = *offp;
397 flsi->whence = whence;
398 err = fdisp_wait_answ(&fdi);
400 fsess_set_notimpl(mp, FUSE_LSEEK);
401 } else if (err == 0) {
402 fsess_set_impl(mp, FUSE_LSEEK);
404 *offp = flso->offset;
412 struct vnop_access_args {
414 #if VOP_ACCESS_TAKES_ACCMODE_T
419 struct ucred *a_cred;
424 fuse_vnop_access(struct vop_access_args *ap)
426 struct vnode *vp = ap->a_vp;
427 int accmode = ap->a_accmode;
428 struct ucred *cred = ap->a_cred;
430 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
434 if (fuse_isdeadfs(vp)) {
435 if (vnode_isvroot(vp)) {
440 if (!(data->dataflags & FSESS_INITED)) {
441 if (vnode_isvroot(vp)) {
442 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
443 (fuse_match_cred(data->daemoncred, cred) == 0)) {
449 if (vnode_islnk(vp)) {
453 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
458 * struct vop_advlock_args {
459 * struct vop_generic_args a_gen;
460 * struct vnode *a_vp;
463 * struct flock *a_fl;
468 fuse_vnop_advlock(struct vop_advlock_args *ap)
470 struct vnode *vp = ap->a_vp;
471 struct flock *fl = ap->a_fl;
472 struct thread *td = curthread;
473 struct ucred *cred = td->td_ucred;
474 pid_t pid = td->td_proc->p_pid;
475 struct fuse_filehandle *fufh;
476 struct fuse_dispatcher fdi;
477 struct fuse_lk_in *fli;
478 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 switch (fl->l_whence) {
520 * Caller is responsible for adding any necessary offset
521 * when SEEK_CUR is used.
527 err = fuse_internal_getattr(vp, &vattr, cred, td);
530 size = vattr.va_size;
531 if (size > OFF_MAX ||
532 (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
536 start = size + fl->l_start;
543 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
547 fdisp_init(&fdi, sizeof(*fli));
549 fdisp_make_vp(&fdi, op, vp, td, cred);
551 fli->fh = fufh->fh_id;
552 fli->owner = td->td_proc->p_pid;
553 fli->lk.start = start;
555 fli->lk.end = start + fl->l_len - 1;
557 fli->lk.end = INT64_MAX;
558 fli->lk.type = fl->l_type;
559 fli->lk.pid = td->td_proc->p_pid;
561 err = fdisp_wait_answ(&fdi);
564 if (err == 0 && op == FUSE_GETLK) {
566 fl->l_type = flo->lk.type;
567 fl->l_whence = SEEK_SET;
568 if (flo->lk.type != F_UNLCK) {
569 fl->l_pid = flo->lk.pid;
570 fl->l_start = flo->lk.start;
571 if (flo->lk.end == INT64_MAX)
574 fl->l_len = flo->lk.end - flo->lk.start + 1;
575 fl->l_start = flo->lk.start;
585 fuse_vnop_allocate(struct vop_allocate_args *ap)
587 struct vnode *vp = ap->a_vp;
588 off_t *len = ap->a_len;
589 off_t *offset = ap->a_offset;
590 struct ucred *cred = ap->a_cred;
591 struct fuse_filehandle *fufh;
592 struct mount *mp = vnode_mount(vp);
593 struct fuse_dispatcher fdi;
594 struct fuse_fallocate_in *ffi;
596 pid_t pid = curthread->td_proc->p_pid;
597 struct fuse_vnode_data *fvdat = VTOFUD(vp);
601 if (fuse_isdeadfs(vp))
604 switch (vp->v_type) {
609 if (vfs_isrdonly(mp))
616 if (vfs_isrdonly(mp))
619 if (fsess_not_impl(mp, FUSE_FALLOCATE))
622 io.uio_offset = *offset;
624 err = vn_rlimit_fsize(vp, &io, curthread);
628 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
632 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
634 err = fuse_vnode_size(vp, &filesize, cred, curthread);
637 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
639 fdisp_init(&fdi, sizeof(*ffi));
640 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
642 ffi->fh = fufh->fh_id;
643 ffi->offset = *offset;
646 err = fdisp_wait_answ(&fdi);
649 fsess_set_notimpl(mp, FUSE_FALLOCATE);
651 } else if (err == EOPNOTSUPP) {
653 * The file system server does not support FUSE_FALLOCATE with
654 * the supplied mode for this particular file.
660 fuse_vnode_undirty_cached_timestamps(vp, false);
661 fuse_internal_clear_suid_on_write(vp, cred, curthread);
662 if (*offset > fvdat->cached_attrs.va_size) {
663 fuse_vnode_setsize(vp, *offset, false);
664 getnanouptime(&fvdat->last_local_modify);
675 struct bufobj **a_bop;
681 fuse_vnop_bmap(struct vop_bmap_args *ap)
683 struct vnode *vp = ap->a_vp;
684 struct bufobj **bo = ap->a_bop;
685 struct thread *td = curthread;
687 struct fuse_dispatcher fdi;
688 struct fuse_bmap_in *fbi;
689 struct fuse_bmap_out *fbo;
690 struct fuse_data *data;
691 struct fuse_vnode_data *fvdat = VTOFUD(vp);
694 daddr_t lbn = ap->a_bn;
695 daddr_t *pbn = ap->a_bnp;
696 int *runp = ap->a_runp;
697 int *runb = ap->a_runb;
701 if (fuse_isdeadfs(vp)) {
705 mp = vnode_mount(vp);
706 data = fuse_get_mpdata(mp);
707 biosize = fuse_iosize(vp);
708 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
709 data->max_readahead_blocks);
715 * The FUSE_BMAP operation does not include the runp and runb
716 * variables, so we must guess. Report nonzero contiguous runs so
717 * cluster_read will combine adjacent reads. It's worthwhile to reduce
718 * upcalls even if we don't know the true physical layout of the file.
720 * FUSE file systems may opt out of read clustering in two ways:
721 * * mounting with -onoclusterr
722 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
725 *runb = MIN(lbn, maxrun);
726 if (runp != NULL && maxrun == 0)
728 else if (runp != NULL) {
730 * If the file's size is cached, use that value to calculate
731 * runp, even if the cache is expired. runp is only advisory,
732 * and the risk of getting it wrong is not worth the cost of
735 if (fvdat->cached_attrs.va_size != VNOVAL)
736 fsize = fvdat->cached_attrs.va_size;
738 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
740 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
746 if (fsess_maybe_impl(mp, FUSE_BMAP)) {
747 fdisp_init(&fdi, sizeof(*fbi));
748 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
751 fbi->blocksize = biosize;
752 error = fdisp_wait_answ(&fdi);
753 if (error == ENOSYS) {
755 fsess_set_notimpl(mp, FUSE_BMAP);
759 if (error == 0 && pbn != NULL)
766 /* If the daemon doesn't support BMAP, make up a sensible default */
768 *pbn = lbn * btodb(biosize);
773 struct vop_close_args {
776 struct ucred *a_cred;
781 fuse_vnop_close(struct vop_close_args *ap)
783 struct vnode *vp = ap->a_vp;
784 struct ucred *cred = ap->a_cred;
785 int fflag = ap->a_fflag;
786 struct thread *td = ap->a_td;
787 pid_t pid = td->td_proc->p_pid;
788 struct fuse_vnode_data *fvdat = VTOFUD(vp);
791 if (fuse_isdeadfs(vp))
795 if (fflag & IO_NDELAY)
798 err = fuse_flush(vp, cred, pid, fflag);
799 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE)) {
803 vap.va_atime = fvdat->cached_attrs.va_atime;
804 err = fuse_internal_setattr(vp, &vap, td, NULL);
806 /* TODO: close the file handle, if we're sure it's no longer used */
807 if ((fvdat->flag & FN_SIZECHANGE) != 0) {
808 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
814 struct vop_copy_file_range_args {
815 struct vop_generic_args a_gen;
816 struct vnode *a_invp;
818 struct vnode *a_outvp;
821 unsigned int a_flags;
822 struct ucred *a_incred;
823 struct ucred *a_outcred;
824 struct thread *a_fsizetd;
828 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
830 struct vnode *invp = ap->a_invp;
831 struct vnode *outvp = ap->a_outvp;
832 struct mount *mp = vnode_mount(invp);
833 struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
834 struct fuse_dispatcher fdi;
835 struct fuse_filehandle *infufh, *outfufh;
836 struct fuse_copy_file_range_in *fcfri;
837 struct ucred *incred = ap->a_incred;
838 struct ucred *outcred = ap->a_outcred;
839 struct fuse_write_out *fwo;
847 if (mp != vnode_mount(outvp))
850 if (incred->cr_uid != outcred->cr_uid)
853 if (incred->cr_groups[0] != outcred->cr_groups[0])
856 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
859 if (ap->a_fsizetd == NULL)
863 pid = td->td_proc->p_pid;
865 /* Lock both vnodes, avoiding risk of deadlock. */
867 err = vn_lock(outvp, LK_EXCLUSIVE);
871 err = vn_lock(invp, LK_SHARED | LK_NOWAIT);
875 err = vn_lock(invp, LK_SHARED);
883 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
887 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
891 io.uio_resid = *ap->a_lenp;
893 io.uio_offset = *ap->a_outoffp;
894 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
899 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
903 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
904 *ap->a_outoffp + io.uio_resid);
908 fdisp_init(&fdi, sizeof(*fcfri));
909 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
911 fcfri->fh_in = infufh->fh_id;
912 fcfri->off_in = *ap->a_inoffp;
913 fcfri->nodeid_out = VTOI(outvp);
914 fcfri->fh_out = outfufh->fh_id;
915 fcfri->off_out = *ap->a_outoffp;
916 fcfri->len = io.uio_resid;
919 err = fdisp_wait_answ(&fdi);
922 *ap->a_lenp = fwo->size;
923 *ap->a_inoffp += fwo->size;
924 *ap->a_outoffp += fwo->size;
925 fuse_internal_clear_suid_on_write(outvp, outcred, td);
926 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
927 fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
928 getnanouptime(&outfvdat->last_local_modify);
930 fuse_vnode_update(invp, FN_ATIMECHANGE);
931 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
941 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
943 err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
944 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
945 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
949 * No need to call vn_rlimit_fsizex_res before return, since the uio is
956 fdisp_make_mknod_for_fallback(
957 struct fuse_dispatcher *fdip,
958 struct componentname *cnp,
964 enum fuse_opcode *op)
966 struct fuse_mknod_in *fmni;
968 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
970 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
974 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
976 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
979 struct vnop_create_args {
981 struct vnode **a_vpp;
982 struct componentname *a_cnp;
987 fuse_vnop_create(struct vop_create_args *ap)
989 struct vnode *dvp = ap->a_dvp;
990 struct vnode **vpp = ap->a_vpp;
991 struct componentname *cnp = ap->a_cnp;
992 struct vattr *vap = ap->a_vap;
993 struct thread *td = curthread;
994 struct ucred *cred = cnp->cn_cred;
996 struct fuse_data *data;
997 struct fuse_create_in *fci;
998 struct fuse_entry_out *feo;
999 struct fuse_open_out *foo;
1000 struct fuse_dispatcher fdi, fdi2;
1001 struct fuse_dispatcher *fdip = &fdi;
1002 struct fuse_dispatcher *fdip2 = NULL;
1006 struct mount *mp = vnode_mount(dvp);
1007 data = fuse_get_mpdata(mp);
1008 uint64_t parentnid = VTOFUD(dvp)->nid;
1009 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1010 enum fuse_opcode op;
1013 if (fuse_isdeadfs(dvp))
1016 /* FUSE expects sockets to be created with FUSE_MKNOD */
1017 if (vap->va_type == VSOCK)
1018 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1021 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
1022 * writable mode makes sense, and we might as well include readability
1027 bzero(&fdi, sizeof(fdi));
1029 if (vap->va_type != VREG)
1032 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1033 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
1034 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1037 /* Use FUSE_CREATE */
1041 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1042 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1045 fci->flags = O_CREAT | flags;
1046 if (fuse_libabi_geq(data, 7, 12)) {
1047 insize = sizeof(*fci);
1048 fci->umask = td->td_proc->p_pd->pd_cmask;
1050 insize = sizeof(struct fuse_open_in);
1053 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1055 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1058 err = fdisp_wait_answ(fdip);
1061 if (err == ENOSYS && op == FUSE_CREATE) {
1062 fsess_set_notimpl(mp, FUSE_CREATE);
1063 fdisp_destroy(fdip);
1064 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1065 parentnid, td, cred, mode, &op);
1066 err = fdisp_wait_answ(fdip);
1074 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1078 if (op == FUSE_CREATE) {
1079 if (fuse_libabi_geq(data, 7, 9))
1080 foo = (struct fuse_open_out*)(feo + 1);
1082 foo = (struct fuse_open_out*)((char*)feo +
1083 FUSE_COMPAT_ENTRY_OUT_SIZE);
1085 /* Issue a separate FUSE_OPEN */
1086 struct fuse_open_in *foi;
1089 fdisp_init(fdip2, sizeof(*foi));
1090 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1092 foi = fdip2->indata;
1094 err = fdisp_wait_answ(fdip2);
1099 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1101 struct fuse_release_in *fri;
1102 uint64_t nodeid = feo->nodeid;
1103 uint64_t fh_id = foo->fh;
1105 fdisp_destroy(fdip);
1106 fdisp_init(fdip, sizeof(*fri));
1107 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1111 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1112 fuse_insert_message(fdip->tick, false);
1115 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1116 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1117 feo->attr_valid_nsec, NULL, true);
1119 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1120 fuse_vnode_open(*vpp, foo->open_flags, td);
1122 * Purge the parent's attribute cache because the daemon should've
1123 * updated its mtime and ctime
1125 fuse_vnode_clear_attr_cache(dvp);
1126 cache_purge_negative(dvp);
1130 fdisp_destroy(fdip2);
1131 fdisp_destroy(fdip);
1136 struct vnop_fdatasync_args {
1137 struct vop_generic_args a_gen;
1138 struct vnode * a_vp;
1139 struct thread * a_td;
1143 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1145 struct vnode *vp = ap->a_vp;
1146 struct thread *td = ap->a_td;
1147 int waitfor = MNT_WAIT;
1151 if (fuse_isdeadfs(vp)) {
1154 if ((err = vop_stdfdatasync_buf(ap)))
1157 return fuse_internal_fsync(vp, td, waitfor, true);
1161 struct vnop_fsync_args {
1162 struct vop_generic_args a_gen;
1163 struct vnode * a_vp;
1165 struct thread * a_td;
1169 fuse_vnop_fsync(struct vop_fsync_args *ap)
1171 struct vnode *vp = ap->a_vp;
1172 struct thread *td = ap->a_td;
1173 int waitfor = ap->a_waitfor;
1176 if (fuse_isdeadfs(vp)) {
1179 if ((err = vop_stdfsync(ap)))
1182 return fuse_internal_fsync(vp, td, waitfor, false);
1186 struct vnop_getattr_args {
1188 struct vattr *a_vap;
1189 struct ucred *a_cred;
1190 struct thread *a_td;
1194 fuse_vnop_getattr(struct vop_getattr_args *ap)
1196 struct vnode *vp = ap->a_vp;
1197 struct vattr *vap = ap->a_vap;
1198 struct ucred *cred = ap->a_cred;
1199 struct thread *td = curthread;
1204 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
1206 /* Note that we are not bailing out on a dead file system just yet. */
1208 if (!(dataflags & FSESS_INITED)) {
1209 if (!vnode_isvroot(vp)) {
1210 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
1217 err = fuse_internal_getattr(vp, vap, cred, td);
1218 if (err == ENOTCONN && vnode_isvroot(vp)) {
1219 /* see comment in fuse_vfsop_statfs() */
1226 bzero(vap, sizeof(*vap));
1227 vap->va_type = vnode_vtype(vp);
1233 struct vnop_inactive_args {
1238 fuse_vnop_inactive(struct vop_inactive_args *ap)
1240 struct vnode *vp = ap->a_vp;
1241 struct thread *td = curthread;
1243 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1244 struct fuse_filehandle *fufh, *fufh_tmp;
1248 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1249 if (need_flush && vp->v_type == VREG) {
1250 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1251 fuse_vnode_savesize(vp, NULL, 0);
1253 if ((fvdat->flag & FN_REVOKED) != 0)
1254 fuse_io_invalbuf(vp, td);
1256 fuse_io_flushbuf(vp, MNT_WAIT, td);
1259 fuse_filehandle_close(vp, fufh, td, NULL);
1262 if ((fvdat->flag & FN_REVOKED) != 0)
1269 struct vnop_ioctl_args {
1274 struct ucred *a_cred;
1275 struct thread *a_td;
1279 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1281 struct vnode *vp = ap->a_vp;
1282 struct mount *mp = vnode_mount(vp);
1283 struct ucred *cred = ap->a_cred;
1285 pid_t pid = ap->a_td->td_proc->p_pid;
1288 switch (ap->a_command) {
1291 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1292 if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1296 if (ap->a_command == FIOSEEKDATA)
1301 vn_lock(vp, LK_SHARED | LK_RETRY);
1302 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1306 if (fsess_not_impl(mp, FUSE_LSEEK))
1307 err = vop_stdioctl(ap);
1310 /* TODO: implement FUSE_IOCTL */
1319 struct vnop_link_args {
1320 struct vnode *a_tdvp;
1322 struct componentname *a_cnp;
1326 fuse_vnop_link(struct vop_link_args *ap)
1328 struct vnode *vp = ap->a_vp;
1329 struct vnode *tdvp = ap->a_tdvp;
1330 struct componentname *cnp = ap->a_cnp;
1332 struct vattr *vap = VTOVA(vp);
1334 struct fuse_dispatcher fdi;
1335 struct fuse_entry_out *feo;
1336 struct fuse_link_in fli;
1340 if (fuse_isdeadfs(vp)) {
1343 if (vnode_mount(tdvp) != vnode_mount(vp)) {
1348 * This is a seatbelt check to protect naive userspace filesystems from
1349 * themselves and the limitations of the FUSE IPC protocol. If a
1350 * filesystem does not allow attribute caching, assume it is capable of
1351 * validating that nlink does not overflow.
1353 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1355 fli.oldnodeid = VTOI(vp);
1357 fdisp_init(&fdi, 0);
1358 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1359 FUSE_LINK, &fli, sizeof(fli), &fdi);
1360 if ((err = fdisp_wait_answ(&fdi))) {
1365 if (fli.oldnodeid != feo->nodeid) {
1366 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1367 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE,
1368 "Assigned wrong inode for a hard link.");
1369 fuse_vnode_clear_attr_cache(vp);
1370 fuse_vnode_clear_attr_cache(tdvp);
1375 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1378 * Purge the parent's attribute cache because the daemon
1379 * should've updated its mtime and ctime
1381 fuse_vnode_clear_attr_cache(tdvp);
1382 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1383 feo->attr_valid_nsec, NULL, true);
1386 fdisp_destroy(&fdi);
1390 struct fuse_lookup_alloc_arg {
1391 struct fuse_entry_out *feo;
1392 struct componentname *cnp;
1397 /* Callback for vn_get_ino */
1399 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1401 struct fuse_lookup_alloc_arg *flaa = arg;
1403 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1407 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1408 "int", "struct timespec*", "struct timespec*");
1410 struct vnop_lookup_args {
1411 struct vnodeop_desc *a_desc;
1412 struct vnode *a_dvp;
1413 struct vnode **a_vpp;
1414 struct componentname *a_cnp;
1418 fuse_vnop_lookup(struct vop_lookup_args *ap)
1420 struct vnode *dvp = ap->a_dvp;
1421 struct vnode **vpp = ap->a_vpp;
1422 struct componentname *cnp = ap->a_cnp;
1423 struct thread *td = curthread;
1424 struct ucred *cred = cnp->cn_cred;
1425 struct timespec now;
1427 int nameiop = cnp->cn_nameiop;
1428 int flags = cnp->cn_flags;
1429 int islastcn = flags & ISLASTCN;
1430 struct mount *mp = vnode_mount(dvp);
1431 struct fuse_data *data = fuse_get_mpdata(mp);
1432 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1437 struct vnode *vp = NULL;
1439 struct fuse_dispatcher fdi;
1440 bool did_lookup = false;
1441 struct fuse_entry_out *feo = NULL;
1442 enum vtype vtyp; /* vnode type of target */
1446 if (fuse_isdeadfs(dvp)) {
1450 if (!vnode_isdir(dvp))
1453 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1456 if ((cnp->cn_flags & NOEXECCHECK) != 0)
1457 cnp->cn_flags &= ~NOEXECCHECK;
1458 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1461 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1462 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT))
1464 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1466 * Since the file system doesn't support ".." lookups,
1467 * we have no way to find this entry.
1471 nid = VTOFUD(dvp)->parent_nid;
1474 /* .. is obviously a directory */
1476 } else if (is_dot) {
1478 /* . is obviously a directory */
1481 struct timespec timeout;
1482 int ncpticks; /* here to accommodate for API contract */
1484 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1485 getnanouptime(&now);
1486 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1488 case -1: /* positive match */
1489 if (timespeccmp(&timeout, &now, >)) {
1490 counter_u64_add(fuse_lookup_cache_hits, 1);
1493 counter_u64_add(fuse_lookup_cache_misses, 1);
1495 &VTOFUD(*vpp)->entry_cache_timeout);
1506 case 0: /* no match in cache */
1507 counter_u64_add(fuse_lookup_cache_misses, 1);
1510 case ENOENT: /* negative match */
1511 if (timespeccmp(&timeout, &now, <=)) {
1513 cache_purge_negative(dvp);
1521 fdisp_init(&fdi, cnp->cn_namelen + 1);
1522 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1524 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1525 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1526 lookup_err = fdisp_wait_answ(&fdi);
1530 /* lookup call succeeded */
1531 feo = (struct fuse_entry_out *)fdi.answ;
1534 /* zero nodeid means ENOENT and cache it */
1535 struct timespec timeout;
1537 fdi.answ_stat = ENOENT;
1538 lookup_err = ENOENT;
1539 if (cnp->cn_flags & MAKEENTRY) {
1540 fuse_validity_2_timespec(feo, &timeout);
1541 /* Use the same entry_time for .. as for
1542 * the file itself. That doesn't honor
1543 * exactly what the fuse server tells
1544 * us, but to do otherwise would require
1545 * another cache lookup at this point.
1547 struct timespec *dtsp = NULL;
1548 cache_enter_time(dvp, *vpp, cnp,
1552 vtyp = IFTOVT(feo->attr.mode);
1554 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1555 fdisp_destroy(&fdi);
1559 /* lookup_err, if non-zero, must be ENOENT at this point */
1562 /* Entry not found */
1563 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1564 if (default_permissions)
1565 err = fuse_internal_access(dvp, VWRITE, td,
1576 /* Entry was found */
1577 if (flags & ISDOTDOT) {
1578 struct fuse_lookup_alloc_arg flaa;
1584 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1587 } else if (nid == VTOI(dvp)) {
1592 fuse_warn(fuse_get_mpdata(mp),
1593 FSESS_WARN_ILLEGAL_INODE,
1594 "Assigned same inode to both parent and "
1600 struct fuse_vnode_data *fvdat;
1602 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1610 if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1612 * Attributes from the server are definitely
1613 * newer than the last attributes we sent to
1614 * the server, so cache them.
1616 fuse_internal_cache_attrs(*vpp, &feo->attr,
1617 feo->attr_valid, feo->attr_valid_nsec,
1620 fuse_validity_2_bintime(feo->entry_valid,
1621 feo->entry_valid_nsec,
1622 &fvdat->entry_cache_timeout);
1624 if ((nameiop == DELETE || nameiop == RENAME) &&
1625 islastcn && default_permissions)
1627 struct vattr dvattr;
1629 err = fuse_internal_access(dvp, VWRITE, td,
1634 * if the parent's sticky bit is set, check
1635 * whether we're allowed to remove the file.
1636 * Need to figure out the vnode locking to make
1639 fuse_internal_getattr(dvp, &dvattr, cred, td);
1640 if ((dvattr.va_mode & S_ISTXT) &&
1641 fuse_internal_access(dvp, VADMIN, td,
1643 fuse_internal_access(*vpp, VADMIN, td,
1653 if (vp != NULL && dvp != vp)
1655 else if (vp != NULL)
1660 fdisp_destroy(&fdi);
1666 struct vnop_mkdir_args {
1667 struct vnode *a_dvp;
1668 struct vnode **a_vpp;
1669 struct componentname *a_cnp;
1670 struct vattr *a_vap;
1674 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1676 struct vnode *dvp = ap->a_dvp;
1677 struct vnode **vpp = ap->a_vpp;
1678 struct componentname *cnp = ap->a_cnp;
1679 struct vattr *vap = ap->a_vap;
1681 struct fuse_mkdir_in fmdi;
1683 if (fuse_isdeadfs(dvp)) {
1686 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1687 fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1689 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1690 sizeof(fmdi), VDIR));
1694 struct vnop_mknod_args {
1695 struct vnode *a_dvp;
1696 struct vnode **a_vpp;
1697 struct componentname *a_cnp;
1698 struct vattr *a_vap;
1702 fuse_vnop_mknod(struct vop_mknod_args *ap)
1705 struct vnode *dvp = ap->a_dvp;
1706 struct vnode **vpp = ap->a_vpp;
1707 struct componentname *cnp = ap->a_cnp;
1708 struct vattr *vap = ap->a_vap;
1710 if (fuse_isdeadfs(dvp))
1713 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1717 struct vop_open_args {
1720 struct ucred *a_cred;
1721 struct thread *a_td;
1722 int a_fdidx; / struct file *a_fp;
1726 fuse_vnop_open(struct vop_open_args *ap)
1728 struct vnode *vp = ap->a_vp;
1729 int a_mode = ap->a_mode;
1730 struct thread *td = ap->a_td;
1731 struct ucred *cred = ap->a_cred;
1732 pid_t pid = td->td_proc->p_pid;
1734 if (fuse_isdeadfs(vp))
1736 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1737 return (EOPNOTSUPP);
1738 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1741 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1742 fuse_vnode_open(vp, 0, td);
1746 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1750 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1752 struct vnode *vp = ap->a_vp;
1755 switch (ap->a_name) {
1756 case _PC_FILESIZEBITS:
1760 *ap->a_retval = NAME_MAX;
1763 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1765 case _PC_SYMLINK_MAX:
1766 *ap->a_retval = MAXPATHLEN;
1771 case _PC_MIN_HOLE_SIZE:
1773 * The FUSE protocol provides no mechanism for a server to
1774 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead,
1775 * return EINVAL if the server does not support FUSE_LSEEK, or
1778 mp = vnode_mount(vp);
1779 if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1780 !fsess_not_impl(mp, FUSE_LSEEK)) {
1783 /* Issue a FUSE_LSEEK to find out if it's implemented */
1784 fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred,
1785 curthread->td_proc->p_pid, &offset, SEEK_DATA);
1788 if (fsess_is_impl(mp, FUSE_LSEEK)) {
1793 * Probably FUSE_LSEEK is not implemented. It might
1794 * be, if the FUSE_LSEEK above returned an error like
1795 * EACCES, but in that case we can't tell, so it's
1796 * safest to report EINVAL anyway.
1801 return (vop_stdpathconf(ap));
1805 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1806 "struct uio*", "struct ucred*");
1808 struct vnop_read_args {
1812 struct ucred *a_cred;
1816 fuse_vnop_read(struct vop_read_args *ap)
1818 struct vnode *vp = ap->a_vp;
1819 struct uio *uio = ap->a_uio;
1820 int ioflag = ap->a_ioflag;
1821 struct ucred *cred = ap->a_cred;
1822 pid_t pid = curthread->td_proc->p_pid;
1823 struct fuse_filehandle *fufh;
1825 bool closefufh = false, directio;
1827 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1829 if (fuse_isdeadfs(vp)) {
1833 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1834 ioflag |= IO_DIRECT;
1837 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1838 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1840 * nfsd will do I/O without first doing VOP_OPEN. We
1841 * must implicitly open the file here
1843 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1847 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1852 * Ideally, when the daemon asks for direct io at open time, the
1853 * standard file flag should be set according to this, so that would
1854 * just change the default mode, which later on could be changed via
1856 * But this doesn't work, the O_DIRECT flag gets cleared at some point
1857 * (don't know where). So to make any use of the Fuse direct_io option,
1858 * we hardwire it into the file's private data (similarly to Linux,
1861 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1863 fuse_vnode_update(vp, FN_ATIMECHANGE);
1865 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1866 err = fuse_read_directbackend(vp, uio, cred, fufh);
1868 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1869 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1873 fuse_filehandle_close(vp, fufh, curthread, cred);
1879 struct vnop_readdir_args {
1882 struct ucred *a_cred;
1885 uint64_t **a_cookies;
1889 fuse_vnop_readdir(struct vop_readdir_args *ap)
1891 struct vnode *vp = ap->a_vp;
1892 struct uio *uio = ap->a_uio;
1893 struct ucred *cred = ap->a_cred;
1894 struct fuse_filehandle *fufh = NULL;
1895 struct mount *mp = vnode_mount(vp);
1896 struct fuse_iov cookediov;
1901 bool closefufh = false;
1902 pid_t pid = curthread->td_proc->p_pid;
1906 if (fuse_isdeadfs(vp)) {
1909 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1910 (uio_resid(uio) < sizeof(struct dirent))) {
1914 tresid = uio->uio_resid;
1915 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1916 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1917 KASSERT(fuse_get_mpdata(mp)->dataflags
1918 & FSESS_NO_OPENDIR_SUPPORT,
1919 ("FUSE file systems that don't set "
1920 "FUSE_NO_OPENDIR_SUPPORT should not be exported"));
1922 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1923 * must implicitly open the directory here.
1925 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1930 if (ap->a_ncookies != NULL) {
1931 ncookies = uio->uio_resid /
1932 (offsetof(struct dirent, d_name) + 4) + 1;
1933 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1934 *ap->a_ncookies = ncookies;
1935 *ap->a_cookies = cookies;
1940 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1941 fiov_init(&cookediov, DIRCOOKEDSIZE);
1943 err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1944 &ncookies, cookies);
1946 fiov_teardown(&cookediov);
1948 fuse_filehandle_close(vp, fufh, curthread, cred);
1950 if (ap->a_ncookies != NULL) {
1952 *ap->a_ncookies -= ncookies;
1954 free(*ap->a_cookies, M_TEMP);
1955 *ap->a_ncookies = 0;
1956 *ap->a_cookies = NULL;
1959 if (err == 0 && tresid == uio->uio_resid)
1966 struct vnop_readlink_args {
1969 struct ucred *a_cred;
1973 fuse_vnop_readlink(struct vop_readlink_args *ap)
1975 struct vnode *vp = ap->a_vp;
1976 struct uio *uio = ap->a_uio;
1977 struct ucred *cred = ap->a_cred;
1979 struct fuse_dispatcher fdi;
1982 if (fuse_isdeadfs(vp)) {
1985 if (!vnode_islnk(vp)) {
1988 fdisp_init(&fdi, 0);
1989 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1993 if (((char *)fdi.answ)[0] == '/' &&
1994 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1995 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1997 err = uiomove(mpth, strlen(mpth), uio);
2000 err = uiomove(fdi.answ, fdi.iosize, uio);
2003 fdisp_destroy(&fdi);
2008 struct vnop_reclaim_args {
2013 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2015 struct vnode *vp = ap->a_vp;
2016 struct thread *td = curthread;
2017 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2018 struct fuse_filehandle *fufh, *fufh_tmp;
2021 panic("FUSE: no vnode data during recycling");
2023 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2024 printf("FUSE: vnode being reclaimed with open fufh "
2025 "(type=%#x)", fufh->fufh_type);
2026 fuse_filehandle_close(vp, fufh, td, NULL);
2029 if (VTOI(vp) == 1) {
2031 * Don't send FUSE_FORGET for the root inode, because
2032 * we never send FUSE_LOOKUP for it (see
2033 * fuse_vfsop_root) and we don't want the server to see
2034 * mismatched lookup counts.
2036 struct fuse_data *data;
2037 struct vnode *vroot;
2039 data = fuse_get_mpdata(vnode_mount(vp));
2041 vroot = data->vroot;
2046 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2047 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2051 vfs_hash_remove(vp);
2052 fuse_vnode_destroy(vp);
2058 struct vnop_remove_args {
2059 struct vnode *a_dvp;
2061 struct componentname *a_cnp;
2065 fuse_vnop_remove(struct vop_remove_args *ap)
2067 struct vnode *dvp = ap->a_dvp;
2068 struct vnode *vp = ap->a_vp;
2069 struct componentname *cnp = ap->a_cnp;
2073 if (fuse_isdeadfs(vp)) {
2076 if (vnode_isdir(vp)) {
2080 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2086 struct vnop_rename_args {
2087 struct vnode *a_fdvp;
2088 struct vnode *a_fvp;
2089 struct componentname *a_fcnp;
2090 struct vnode *a_tdvp;
2091 struct vnode *a_tvp;
2092 struct componentname *a_tcnp;
2096 fuse_vnop_rename(struct vop_rename_args *ap)
2098 struct vnode *fdvp = ap->a_fdvp;
2099 struct vnode *fvp = ap->a_fvp;
2100 struct componentname *fcnp = ap->a_fcnp;
2101 struct vnode *tdvp = ap->a_tdvp;
2102 struct vnode *tvp = ap->a_tvp;
2103 struct componentname *tcnp = ap->a_tcnp;
2104 struct fuse_data *data;
2105 bool newparent = fdvp != tdvp;
2106 bool isdir = fvp->v_type == VDIR;
2109 if (fuse_isdeadfs(fdvp)) {
2112 if (fvp->v_mount != tdvp->v_mount ||
2113 (tvp && fvp->v_mount != tvp->v_mount)) {
2114 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2121 * FUSE library is expected to check if target directory is not
2122 * under the source directory in the file system tree.
2123 * Linux performs this check at VFS level.
2126 * If source is a directory, and it will get a new parent, user must
2127 * have write permission to it, so ".." can be modified.
2129 data = fuse_get_mpdata(vnode_mount(tdvp));
2130 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2131 err = fuse_internal_access(fvp, VWRITE,
2132 curthread, tcnp->cn_cred);
2136 sx_xlock(&data->rename_lock);
2137 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2140 fuse_vnode_setparent(fvp, tdvp);
2142 fuse_vnode_setparent(tvp, NULL);
2144 sx_unlock(&data->rename_lock);
2146 if (tvp != NULL && tvp != fvp) {
2149 if (vnode_isdir(fvp)) {
2150 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2171 struct vnop_rmdir_args {
2172 struct vnode *a_dvp;
2174 struct componentname *a_cnp;
2178 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2180 struct vnode *dvp = ap->a_dvp;
2181 struct vnode *vp = ap->a_vp;
2185 if (fuse_isdeadfs(vp)) {
2188 if (VTOFUD(vp) == VTOFUD(dvp)) {
2191 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2197 struct vnop_setattr_args {
2199 struct vattr *a_vap;
2200 struct ucred *a_cred;
2201 struct thread *a_td;
2205 fuse_vnop_setattr(struct vop_setattr_args *ap)
2207 struct vnode *vp = ap->a_vp;
2208 struct vattr *vap = ap->a_vap;
2209 struct ucred *cred = ap->a_cred;
2210 struct thread *td = curthread;
2212 struct fuse_data *data;
2213 struct vattr old_va;
2216 accmode_t accmode = 0;
2218 bool drop_suid = false;
2220 mp = vnode_mount(vp);
2221 data = fuse_get_mpdata(mp);
2222 dataflags = data->dataflags;
2223 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2225 if (fuse_isdeadfs(vp)) {
2229 if (vap->va_uid != (uid_t)VNOVAL) {
2231 /* Only root may change a file's owner */
2232 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2234 /* As a special case, allow the null chown */
2235 err2 = fuse_internal_getattr(vp, &old_va, cred,
2239 if (vap->va_uid != old_va.va_uid)
2249 if (vap->va_gid != (gid_t)VNOVAL) {
2250 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2252 if (checkperm && !groupmember(vap->va_gid, cred))
2255 * Non-root users may only chgrp to one of their own
2258 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2260 /* As a special case, allow the null chgrp */
2261 err2 = fuse_internal_getattr(vp, &old_va, cred,
2265 if (vap->va_gid != old_va.va_gid)
2273 if (vap->va_size != VNOVAL) {
2274 switch (vp->v_type) {
2279 if (vfs_isrdonly(mp))
2281 err = vn_rlimit_trunc(vap->va_size, td);
2287 * According to POSIX, the result is unspecified
2288 * for file types other than regular files,
2289 * directories and shared memory objects. We
2290 * don't support shared memory objects in the file
2291 * system, and have dubious support for truncating
2292 * symlinks. Just ignore the request in other cases.
2296 /* Don't set accmode. Permission to trunc is checked upstack */
2298 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2299 if (vap->va_vaflags & VA_UTIMES_NULL)
2305 if (vap->va_mode != (mode_t)VNOVAL)
2306 vap->va_mode &= ~(S_ISUID | S_ISGID);
2308 err = fuse_internal_getattr(vp, &old_va, cred, td);
2311 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2314 if (vap->va_mode != (mode_t)VNOVAL) {
2315 /* Only root may set the sticky bit on non-directories */
2316 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2317 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2319 if (checkperm && (vap->va_mode & S_ISGID)) {
2320 err = fuse_internal_getattr(vp, &old_va, cred, td);
2323 if (!groupmember(old_va.va_gid, cred)) {
2324 err = priv_check_cred(cred, PRIV_VFS_SETGID);
2332 if (vfs_isrdonly(mp))
2336 err = fuse_internal_access(vp, accmode, td, cred);
2343 return fuse_internal_setattr(vp, vap, td, cred);
2347 struct vnop_strategy_args {
2353 fuse_vnop_strategy(struct vop_strategy_args *ap)
2355 struct vnode *vp = ap->a_vp;
2356 struct buf *bp = ap->a_bp;
2358 if (!vp || fuse_isdeadfs(vp)) {
2359 bp->b_ioflags |= BIO_ERROR;
2360 bp->b_error = ENXIO;
2366 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2367 * fuse_io_strategy sets bp's error fields
2369 (void)fuse_io_strategy(vp, bp);
2375 struct vnop_symlink_args {
2376 struct vnode *a_dvp;
2377 struct vnode **a_vpp;
2378 struct componentname *a_cnp;
2379 struct vattr *a_vap;
2384 fuse_vnop_symlink(struct vop_symlink_args *ap)
2386 struct vnode *dvp = ap->a_dvp;
2387 struct vnode **vpp = ap->a_vpp;
2388 struct componentname *cnp = ap->a_cnp;
2389 const char *target = ap->a_target;
2391 struct fuse_dispatcher fdi;
2396 if (fuse_isdeadfs(dvp)) {
2400 * Unlike the other creator type calls, here we have to create a message
2401 * where the name of the new entry comes first, and the data describing
2402 * the entry comes second.
2403 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2404 * but put together the message manually and just call the core part.
2407 len = strlen(target) + 1;
2408 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2409 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2411 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2412 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2413 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2415 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2416 fdisp_destroy(&fdi);
2421 struct vnop_write_args {
2425 struct ucred *a_cred;
2429 fuse_vnop_write(struct vop_write_args *ap)
2431 struct vnode *vp = ap->a_vp;
2432 struct uio *uio = ap->a_uio;
2433 int ioflag = ap->a_ioflag;
2434 struct ucred *cred = ap->a_cred;
2435 pid_t pid = curthread->td_proc->p_pid;
2436 struct fuse_filehandle *fufh;
2438 bool closefufh = false, directio;
2440 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2442 if (fuse_isdeadfs(vp)) {
2446 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2447 ioflag |= IO_DIRECT;
2450 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2451 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2453 * nfsd will do I/O without first doing VOP_OPEN. We
2454 * must implicitly open the file here
2456 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2460 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2465 * Ideally, when the daemon asks for direct io at open time, the
2466 * standard file flag should be set according to this, so that would
2467 * just change the default mode, which later on could be changed via
2469 * But this doesn't work, the O_DIRECT flag gets cleared at some point
2470 * (don't know where). So to make any use of the Fuse direct_io option,
2471 * we hardwire it into the file's private data (similarly to Linux,
2474 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2476 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2478 off_t start, end, filesize;
2479 bool pages = (ioflag & IO_VMIO) != 0;
2481 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2483 err = fuse_vnode_size(vp, &filesize, cred, curthread);
2487 start = uio->uio_offset;
2488 end = start + uio->uio_resid;
2490 err = fuse_inval_buf_range(vp, filesize, start,
2495 err = fuse_write_directbackend(vp, uio, cred, fufh,
2496 filesize, ioflag, pages);
2498 SDT_PROBE2(fusefs, , vnops, trace, 1,
2499 "buffered write of vnode");
2500 if (!fsess_opt_writeback(vnode_mount(vp)))
2502 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2504 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2508 fuse_filehandle_close(vp, fufh, curthread, cred);
2514 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2516 const int biosize = fuse_iosize(vp);
2518 return (off / biosize);
2522 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2526 const int biosize = fuse_iosize(vp);
2528 err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2530 /* This will turn into a SIGBUS */
2532 } else if ((off_t)lbn * biosize >= filesize) {
2534 } else if ((off_t)(lbn + 1) * biosize > filesize) {
2535 *blksz = filesize - (off_t)lbn *biosize;
2543 struct vnop_getpages_args {
2551 fuse_vnop_getpages(struct vop_getpages_args *ap)
2553 struct vnode *vp = ap->a_vp;
2555 if (!fsess_opt_mmap(vnode_mount(vp))) {
2556 SDT_PROBE2(fusefs, , vnops, trace, 1,
2557 "called on non-cacheable vnode??\n");
2558 return (VM_PAGER_ERROR);
2561 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2562 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2565 static const char extattr_namespace_separator = '.';
2568 struct vop_getextattr_args {
2569 struct vop_generic_args a_gen;
2571 int a_attrnamespace;
2575 struct ucred *a_cred;
2576 struct thread *a_td;
2580 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2582 struct vnode *vp = ap->a_vp;
2583 struct uio *uio = ap->a_uio;
2584 struct fuse_dispatcher fdi;
2585 struct fuse_getxattr_in *get_xattr_in;
2586 struct fuse_getxattr_out *get_xattr_out;
2587 struct mount *mp = vnode_mount(vp);
2588 struct thread *td = ap->a_td;
2589 struct ucred *cred = ap->a_cred;
2595 if (fuse_isdeadfs(vp))
2598 if (fsess_not_impl(mp, FUSE_GETXATTR))
2601 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2605 /* Default to looking for user attributes. */
2606 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2607 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2609 prefix = EXTATTR_NAMESPACE_USER_STRING;
2611 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2612 strlen(ap->a_name) + 1;
2614 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2615 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2617 get_xattr_in = fdi.indata;
2619 * Check to see whether we're querying the available size or
2620 * issuing the actual request. If we pass in 0, we get back struct
2621 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2622 * that much data, without the struct fuse_getxattr_out header.
2625 get_xattr_in->size = 0;
2627 get_xattr_in->size = uio->uio_resid;
2629 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2630 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2633 err = fdisp_wait_answ(&fdi);
2635 if (err == ENOSYS) {
2636 fsess_set_notimpl(mp, FUSE_GETXATTR);
2642 get_xattr_out = fdi.answ;
2644 if (ap->a_size != NULL)
2645 *ap->a_size = get_xattr_out->size;
2648 err = uiomove(fdi.answ, fdi.iosize, uio);
2651 fdisp_destroy(&fdi);
2656 struct vop_setextattr_args {
2657 struct vop_generic_args a_gen;
2659 int a_attrnamespace;
2662 struct ucred *a_cred;
2663 struct thread *a_td;
2667 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2669 struct vnode *vp = ap->a_vp;
2670 struct uio *uio = ap->a_uio;
2671 struct fuse_dispatcher fdi;
2672 struct fuse_setxattr_in *set_xattr_in;
2673 struct mount *mp = vnode_mount(vp);
2674 struct thread *td = ap->a_td;
2675 struct ucred *cred = ap->a_cred;
2681 if (fuse_isdeadfs(vp))
2684 if (fsess_not_impl(mp, FUSE_SETXATTR))
2687 if (vfs_isrdonly(mp))
2690 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2691 if (ap->a_uio == NULL) {
2693 * If we got here as fallback from VOP_DELETEEXTATTR, then
2694 * return EOPNOTSUPP.
2696 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2697 return (EOPNOTSUPP);
2702 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2707 /* Default to looking for user attributes. */
2708 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2709 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2711 prefix = EXTATTR_NAMESPACE_USER_STRING;
2713 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2714 strlen(ap->a_name) + 1;
2716 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2717 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2719 set_xattr_in = fdi.indata;
2720 set_xattr_in->size = uio->uio_resid;
2722 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2723 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2726 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2727 uio->uio_resid, uio);
2732 err = fdisp_wait_answ(&fdi);
2734 if (err == ENOSYS) {
2735 fsess_set_notimpl(mp, FUSE_SETXATTR);
2738 if (err == ERESTART) {
2739 /* Can't restart after calling uiomove */
2744 fdisp_destroy(&fdi);
2749 * The Linux / FUSE extended attribute list is simply a collection of
2750 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2751 * byte length followed by a non-NUL terminated string. So, this allows
2752 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2753 * Linux attribute names are reported with the namespace as a prefix (e.g.
2754 * "user.attribute_name"), but in FreeBSD they are reported without the
2755 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2757 * user.attr_name1\0user.attr_name2\0
2761 * <num>attr_name1<num>attr_name2
2763 * Where "<num>" is a single byte number of characters in the attribute name.
2766 * prefix - exattr namespace prefix string
2767 * list, list_len - input list with namespace prefixes
2768 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2771 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2772 char *bsd_list, int *bsd_list_len)
2774 int len, pos, dist_to_next, prefix_len;
2778 prefix_len = strlen(prefix);
2780 while (pos < list_len && list[pos] != '\0') {
2781 dist_to_next = strlen(&list[pos]) + 1;
2782 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2783 list[pos + prefix_len] == extattr_namespace_separator) {
2784 len = dist_to_next -
2785 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2786 if (len >= EXTATTR_MAXNAMELEN)
2787 return (ENAMETOOLONG);
2789 bsd_list[*bsd_list_len] = len;
2790 memcpy(&bsd_list[*bsd_list_len + 1],
2791 &list[pos + prefix_len +
2792 sizeof(extattr_namespace_separator)], len);
2794 *bsd_list_len += len + 1;
2797 pos += dist_to_next;
2804 * List extended attributes
2806 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2807 * has a number of differences compared to its FreeBSD equivalent,
2808 * extattr_list_file:
2810 * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2811 * whereas listxattr(2) only returns attributes for a single namespace
2812 * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2813 * - If the provided buffer is not large enough to hold the result,
2814 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2815 * return as many results as will fit.
2818 struct vop_listextattr_args {
2819 struct vop_generic_args a_gen;
2821 int a_attrnamespace;
2824 struct ucred *a_cred;
2825 struct thread *a_td;
2829 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2831 struct vnode *vp = ap->a_vp;
2832 struct uio *uio = ap->a_uio;
2833 struct fuse_dispatcher fdi;
2834 struct fuse_listxattr_in *list_xattr_in;
2835 struct fuse_listxattr_out *list_xattr_out;
2836 struct mount *mp = vnode_mount(vp);
2837 struct thread *td = ap->a_td;
2838 struct ucred *cred = ap->a_cred;
2840 char *bsd_list = NULL;
2846 if (fuse_isdeadfs(vp))
2849 if (fsess_not_impl(mp, FUSE_LISTXATTR))
2852 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2857 * Add space for a NUL and the period separator if enabled.
2858 * Default to looking for user attributes.
2860 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2861 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2863 prefix = EXTATTR_NAMESPACE_USER_STRING;
2865 fdisp_init(&fdi, sizeof(*list_xattr_in));
2866 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2869 * Retrieve Linux / FUSE compatible list size.
2871 list_xattr_in = fdi.indata;
2872 list_xattr_in->size = 0;
2874 err = fdisp_wait_answ(&fdi);
2876 if (err == ENOSYS) {
2877 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2883 list_xattr_out = fdi.answ;
2884 linux_list_len = list_xattr_out->size;
2885 if (linux_list_len == 0) {
2886 if (ap->a_size != NULL)
2887 *ap->a_size = linux_list_len;
2892 * Retrieve Linux / FUSE compatible list values.
2894 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2895 list_xattr_in = fdi.indata;
2896 list_xattr_in->size = linux_list_len;
2898 err = fdisp_wait_answ(&fdi);
2899 if (err == ERANGE) {
2901 * Race detected. The attribute list must've grown since the
2902 * first FUSE_LISTXATTR call. Start over. Go all the way back
2903 * to userland so we can process signals, if necessary, before
2908 } else if (err != 0)
2911 linux_list = fdi.answ;
2912 /* FUSE doesn't allow the server to return more data than requested */
2913 if (fdi.iosize > linux_list_len) {
2914 struct fuse_data *data = fuse_get_mpdata(mp);
2916 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2918 "more extended attribute data than requested; "
2919 "should've returned ERANGE instead.");
2921 /* But returning less data is fine */
2922 linux_list_len = fdi.iosize;
2926 * Retrieve the BSD compatible list values.
2927 * The Linux / FUSE attribute list format isn't the same
2928 * as FreeBSD's format. So we need to transform it into
2929 * FreeBSD's format before giving it to the user.
2931 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2932 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2933 bsd_list, &bsd_list_len);
2937 if (ap->a_size != NULL)
2938 *ap->a_size = bsd_list_len;
2941 err = uiomove(bsd_list, bsd_list_len, uio);
2944 free(bsd_list, M_TEMP);
2945 fdisp_destroy(&fdi);
2950 struct vop_deallocate_args {
2951 struct vop_generic_args a_gen;
2957 struct ucred *a_cred;
2961 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
2963 struct vnode *vp = ap->a_vp;
2964 struct mount *mp = vnode_mount(vp);
2965 struct fuse_filehandle *fufh;
2966 struct fuse_dispatcher fdi;
2967 struct fuse_fallocate_in *ffi;
2968 struct ucred *cred = ap->a_cred;
2969 pid_t pid = curthread->td_proc->p_pid;
2970 off_t *len = ap->a_len;
2971 off_t *offset = ap->a_offset;
2972 int ioflag = ap->a_ioflag;
2975 bool closefufh = false;
2977 if (fuse_isdeadfs(vp))
2980 if (vfs_isrdonly(mp))
2983 if (fsess_not_impl(mp, FUSE_FALLOCATE))
2986 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2987 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2989 * nfsd will do I/O without first doing VOP_OPEN. We
2990 * must implicitly open the file here
2992 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2998 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3000 err = fuse_vnode_size(vp, &filesize, cred, curthread);
3003 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3005 fdisp_init(&fdi, sizeof(*ffi));
3006 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3008 ffi->fh = fufh->fh_id;
3009 ffi->offset = *offset;
3012 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3013 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3015 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3016 err = fdisp_wait_answ(&fdi);
3018 if (err == ENOSYS) {
3019 fdisp_destroy(&fdi);
3020 fsess_set_notimpl(mp, FUSE_FALLOCATE);
3022 } else if (err == EOPNOTSUPP) {
3024 * The file system server does not support FUSE_FALLOCATE with
3025 * the supplied mode for this particular file.
3027 fdisp_destroy(&fdi);
3031 * Clip the returned offset to EoF. Do it here rather than
3032 * before FUSE_FALLOCATE just in case the kernel's cached file
3033 * size is out of date. Unfortunately, FUSE does not return
3034 * any information about filesize from that operation.
3036 *offset = MIN(*offset + *len, filesize);
3038 fuse_vnode_undirty_cached_timestamps(vp, false);
3039 fuse_internal_clear_suid_on_write(vp, cred, curthread);
3041 if (ioflag & IO_SYNC)
3042 err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3047 fdisp_destroy(&fdi);
3049 fuse_filehandle_close(vp, fufh, curthread, cred);
3055 fuse_filehandle_close(vp, fufh, curthread, cred);
3057 return (vop_stddeallocate(ap));
3061 struct vop_deleteextattr_args {
3062 struct vop_generic_args a_gen;
3064 int a_attrnamespace;
3066 struct ucred *a_cred;
3067 struct thread *a_td;
3071 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3073 struct vnode *vp = ap->a_vp;
3074 struct fuse_dispatcher fdi;
3075 struct mount *mp = vnode_mount(vp);
3076 struct thread *td = ap->a_td;
3077 struct ucred *cred = ap->a_cred;
3083 if (fuse_isdeadfs(vp))
3086 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3089 if (vfs_isrdonly(mp))
3092 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3097 /* Default to looking for user attributes. */
3098 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3099 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3101 prefix = EXTATTR_NAMESPACE_USER_STRING;
3103 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3104 strlen(ap->a_name) + 1;
3106 fdisp_init(&fdi, len);
3107 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3109 attr_str = fdi.indata;
3110 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3113 err = fdisp_wait_answ(&fdi);
3114 if (err == ENOSYS) {
3115 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3119 fdisp_destroy(&fdi);
3124 struct vnop_print_args {
3129 fuse_vnop_print(struct vop_print_args *ap)
3131 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3133 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3134 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3135 (uintmax_t)fvdat->nlookup,
3142 * Get an NFS filehandle for a FUSE file.
3144 * This will only work for FUSE file systems that guarantee the uniqueness of
3145 * nodeid:generation, which most don't.
3149 IN struct vnode *a_vp;
3150 IN struct fid *a_fhp;
3154 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3156 struct vnode *vp = ap->a_vp;
3157 struct fuse_vnode_data *fvdat = VTOFUD(vp);
3158 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3159 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3160 "FUSE fid type is too big");
3161 struct mount *mp = vnode_mount(vp);
3162 struct fuse_data *data = fuse_get_mpdata(mp);
3166 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3167 /* NFS requires lookups for "." and ".." */
3168 SDT_PROBE2(fusefs, , vnops, trace, 1,
3169 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3172 if ((mp->mnt_flag & MNT_EXPORTED) &&
3173 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT))
3176 * NFS is stateless, so nfsd must reopen a directory on every
3177 * call to VOP_READDIR, passing in the d_off field from the
3178 * final dirent of the previous invocation. But without
3179 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not
3180 * guarantee that d_off will be valid after a directory is
3181 * closed and reopened. So prohibit exporting FUSE file
3182 * systems that don't set that flag.
3184 * But userspace NFS servers don't have this problem.
3186 SDT_PROBE2(fusefs, , vnops, trace, 1,
3187 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT");
3191 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3195 /*ip = VTOI(ap->a_vp);*/
3196 /*ufhp = (struct ufid *)ap->a_fhp;*/
3197 fhp->len = sizeof(struct fuse_fid);
3198 fhp->nid = fvdat->nid;
3199 if (fvdat->generation <= UINT32_MAX)
3200 fhp->gen = fvdat->generation;