1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
9 * met:
10 *
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
16 * distribution.
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.
20 *
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.
32 *
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
35 *
36 * Copyright (c) 2019 The FreeBSD Foundation
37 *
38 * Portions of this software were developed by BFF Storage Systems, LLC under
39 * sponsorship from the FreeBSD Foundation.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
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.
49 *
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
60 * SUCH DAMAGE.
61 */
62
63 #include <sys/param.h>
64 #include <sys/module.h>
65 #include <sys/systm.h>
66 #include <sys/errno.h>
67 #include <sys/kernel.h>
68 #include <sys/conf.h>
69 #include <sys/filio.h>
70 #include <sys/uio.h>
71 #include <sys/malloc.h>
72 #include <sys/queue.h>
73 #include <sys/limits.h>
74 #include <sys/lock.h>
75 #include <sys/rwlock.h>
76 #include <sys/sx.h>
77 #include <sys/proc.h>
78 #include <sys/mount.h>
79 #include <sys/vnode.h>
80 #include <sys/namei.h>
81 #include <sys/extattr.h>
82 #include <sys/stat.h>
83 #include <sys/unistd.h>
84 #include <sys/filedesc.h>
85 #include <sys/file.h>
86 #include <sys/fcntl.h>
87 #include <sys/dirent.h>
88 #include <sys/bio.h>
89 #include <sys/buf.h>
90 #include <sys/sysctl.h>
91 #include <sys/vmmeter.h>
92
93 #include <vm/vm.h>
94 #include <vm/vm_extern.h>
95 #include <vm/pmap.h>
96 #include <vm/vm_map.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_object.h>
100 #include <vm/vm_pager.h>
101 #include <vm/vnode_pager.h>
102 #include <vm/vm_object.h>
103
104 #include "fuse.h"
105 #include "fuse_file.h"
106 #include "fuse_internal.h"
107 #include "fuse_ipc.h"
108 #include "fuse_node.h"
109 #include "fuse_io.h"
110
111 #include <sys/priv.h>
112
113 /* Maximum number of hardlinks to a single FUSE file */
114 #define FUSE_LINK_MAX UINT32_MAX
115
116 SDT_PROVIDER_DECLARE(fusefs);
117 /*
118 * Fuse trace probe:
119 * arg0: verbosity. Higher numbers give more verbose messages
120 * arg1: Textual message
121 */
122 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
123
124 /* vnode ops */
125 static vop_access_t fuse_vnop_access;
126 static vop_advlock_t fuse_vnop_advlock;
127 static vop_allocate_t fuse_vnop_allocate;
128 static vop_bmap_t fuse_vnop_bmap;
129 static vop_close_t fuse_fifo_close;
130 static vop_close_t fuse_vnop_close;
131 static vop_copy_file_range_t fuse_vnop_copy_file_range;
132 static vop_create_t fuse_vnop_create;
133 static vop_deallocate_t fuse_vnop_deallocate;
134 static vop_deleteextattr_t fuse_vnop_deleteextattr;
135 static vop_fdatasync_t fuse_vnop_fdatasync;
136 static vop_fsync_t fuse_vnop_fsync;
137 static vop_getattr_t fuse_vnop_getattr;
138 static vop_getextattr_t fuse_vnop_getextattr;
139 static vop_inactive_t fuse_vnop_inactive;
140 static vop_ioctl_t fuse_vnop_ioctl;
141 static vop_link_t fuse_vnop_link;
142 static vop_listextattr_t fuse_vnop_listextattr;
143 static vop_lookup_t fuse_vnop_lookup;
144 static vop_mkdir_t fuse_vnop_mkdir;
145 static vop_mknod_t fuse_vnop_mknod;
146 static vop_open_t fuse_vnop_open;
147 static vop_pathconf_t fuse_vnop_pathconf;
148 static vop_read_t fuse_vnop_read;
149 static vop_readdir_t fuse_vnop_readdir;
150 static vop_readlink_t fuse_vnop_readlink;
151 static vop_reclaim_t fuse_vnop_reclaim;
152 static vop_remove_t fuse_vnop_remove;
153 static vop_rename_t fuse_vnop_rename;
154 static vop_rmdir_t fuse_vnop_rmdir;
155 static vop_setattr_t fuse_vnop_setattr;
156 static vop_setextattr_t fuse_vnop_setextattr;
157 static vop_strategy_t fuse_vnop_strategy;
158 static vop_symlink_t fuse_vnop_symlink;
159 static vop_write_t fuse_vnop_write;
160 static vop_getpages_t fuse_vnop_getpages;
161 static vop_print_t fuse_vnop_print;
162 static vop_vptofh_t fuse_vnop_vptofh;
163
164 struct vop_vector fuse_fifoops = {
165 .vop_default = &fifo_specops,
166 .vop_access = fuse_vnop_access,
167 .vop_close = fuse_fifo_close,
168 .vop_fsync = fuse_vnop_fsync,
169 .vop_getattr = fuse_vnop_getattr,
170 .vop_inactive = fuse_vnop_inactive,
171 .vop_pathconf = fuse_vnop_pathconf,
172 .vop_print = fuse_vnop_print,
173 .vop_read = VOP_PANIC,
174 .vop_reclaim = fuse_vnop_reclaim,
175 .vop_setattr = fuse_vnop_setattr,
176 .vop_write = VOP_PANIC,
177 .vop_vptofh = fuse_vnop_vptofh,
178 };
179 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
180
181 struct vop_vector fuse_vnops = {
182 .vop_allocate = fuse_vnop_allocate,
183 .vop_default = &default_vnodeops,
184 .vop_access = fuse_vnop_access,
185 .vop_advlock = fuse_vnop_advlock,
186 .vop_bmap = fuse_vnop_bmap,
187 .vop_close = fuse_vnop_close,
188 .vop_copy_file_range = fuse_vnop_copy_file_range,
189 .vop_create = fuse_vnop_create,
190 .vop_deallocate = fuse_vnop_deallocate,
191 .vop_deleteextattr = fuse_vnop_deleteextattr,
192 .vop_fsync = fuse_vnop_fsync,
193 .vop_fdatasync = fuse_vnop_fdatasync,
194 .vop_getattr = fuse_vnop_getattr,
195 .vop_getextattr = fuse_vnop_getextattr,
196 .vop_inactive = fuse_vnop_inactive,
197 .vop_ioctl = fuse_vnop_ioctl,
198 .vop_link = fuse_vnop_link,
199 .vop_listextattr = fuse_vnop_listextattr,
200 .vop_lookup = fuse_vnop_lookup,
201 .vop_mkdir = fuse_vnop_mkdir,
202 .vop_mknod = fuse_vnop_mknod,
203 .vop_open = fuse_vnop_open,
204 .vop_pathconf = fuse_vnop_pathconf,
205 /*
206 * TODO: implement vop_poll after upgrading to protocol 7.21.
207 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
208 * 7.21, which adds the ability for the client to choose which poll
209 * events it wants, and for a client to deregister a file handle
210 */
211 .vop_read = fuse_vnop_read,
212 .vop_readdir = fuse_vnop_readdir,
213 .vop_readlink = fuse_vnop_readlink,
214 .vop_reclaim = fuse_vnop_reclaim,
215 .vop_remove = fuse_vnop_remove,
216 .vop_rename = fuse_vnop_rename,
217 .vop_rmdir = fuse_vnop_rmdir,
218 .vop_setattr = fuse_vnop_setattr,
219 .vop_setextattr = fuse_vnop_setextattr,
220 .vop_strategy = fuse_vnop_strategy,
221 .vop_symlink = fuse_vnop_symlink,
222 .vop_write = fuse_vnop_write,
223 .vop_getpages = fuse_vnop_getpages,
224 .vop_print = fuse_vnop_print,
225 .vop_vptofh = fuse_vnop_vptofh,
226 };
227 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
228
229 /* Check permission for extattr operations, much like extattr_check_cred */
230 static int
fuse_extattr_check_cred(struct vnode * vp,int ns,struct ucred * cred,struct thread * td,accmode_t accmode)231 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
232 struct thread *td, accmode_t accmode)
233 {
234 struct mount *mp = vnode_mount(vp);
235 struct fuse_data *data = fuse_get_mpdata(mp);
236 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
237
238 /*
239 * Kernel-invoked always succeeds.
240 */
241 if (cred == NOCRED)
242 return (0);
243
244 /*
245 * Do not allow privileged processes in jail to directly manipulate
246 * system attributes.
247 */
248 switch (ns) {
249 case EXTATTR_NAMESPACE_SYSTEM:
250 if (default_permissions) {
251 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
252 }
253 return (0);
254 case EXTATTR_NAMESPACE_USER:
255 if (default_permissions) {
256 return (fuse_internal_access(vp, accmode, td, cred));
257 }
258 return (0);
259 default:
260 return (EPERM);
261 }
262 }
263
264 /* Get a filehandle for a directory */
265 static int
fuse_filehandle_get_dir(struct vnode * vp,struct fuse_filehandle ** fufhp,struct ucred * cred,pid_t pid)266 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
267 struct ucred *cred, pid_t pid)
268 {
269 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
270 return 0;
271 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
272 }
273
274 /* Send FUSE_FLUSH for this vnode */
275 static int
fuse_flush(struct vnode * vp,struct ucred * cred,pid_t pid,int fflag)276 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
277 {
278 struct fuse_flush_in *ffi;
279 struct fuse_filehandle *fufh;
280 struct fuse_dispatcher fdi;
281 struct thread *td = curthread;
282 struct mount *mp = vnode_mount(vp);
283 int err;
284
285 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
286 return 0;
287
288 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
289 if (err)
290 return err;
291
292 if (fufh->fuse_open_flags & FOPEN_NOFLUSH &&
293 (!fsess_opt_writeback(vnode_mount(vp))))
294 return (0);
295
296 fdisp_init(&fdi, sizeof(*ffi));
297 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
298 ffi = fdi.indata;
299 ffi->fh = fufh->fh_id;
300 /*
301 * If the file has a POSIX lock then we're supposed to set lock_owner.
302 * If not, then lock_owner is undefined. So we may as well always set
303 * it.
304 */
305 ffi->lock_owner = td->td_proc->p_pid;
306
307 err = fdisp_wait_answ(&fdi);
308 if (err == ENOSYS) {
309 fsess_set_notimpl(mp, FUSE_FLUSH);
310 err = 0;
311 }
312 fdisp_destroy(&fdi);
313 return err;
314 }
315
316 /* Close wrapper for fifos. */
317 static int
fuse_fifo_close(struct vop_close_args * ap)318 fuse_fifo_close(struct vop_close_args *ap)
319 {
320 return (fifo_specops.vop_close(ap));
321 }
322
323 /* Invalidate a range of cached data, whether dirty of not */
324 static int
fuse_inval_buf_range(struct vnode * vp,off_t filesize,off_t start,off_t end)325 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
326 {
327 struct buf *bp;
328 daddr_t left_lbn, end_lbn, right_lbn;
329 off_t new_filesize;
330 int iosize, left_on, right_on, right_blksize;
331
332 iosize = fuse_iosize(vp);
333 left_lbn = start / iosize;
334 end_lbn = howmany(end, iosize);
335 left_on = start & (iosize - 1);
336 if (left_on != 0) {
337 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
338 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
339 /*
340 * Flush the dirty buffer, because we don't have a
341 * byte-granular way to record which parts of the
342 * buffer are valid.
343 */
344 bwrite(bp);
345 if (bp->b_error)
346 return (bp->b_error);
347 } else {
348 brelse(bp);
349 }
350 }
351 right_on = end & (iosize - 1);
352 if (right_on != 0) {
353 right_lbn = end / iosize;
354 new_filesize = MAX(filesize, end);
355 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
356 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
357 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
358 /*
359 * Flush the dirty buffer, because we don't have a
360 * byte-granular way to record which parts of the
361 * buffer are valid.
362 */
363 bwrite(bp);
364 if (bp->b_error)
365 return (bp->b_error);
366 } else {
367 brelse(bp);
368 }
369 }
370
371 v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
372 return (0);
373 }
374
375
376 /* Send FUSE_LSEEK for this node */
377 static int
fuse_vnop_do_lseek(struct vnode * vp,struct thread * td,struct ucred * cred,pid_t pid,off_t * offp,int whence)378 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
379 pid_t pid, off_t *offp, int whence)
380 {
381 struct fuse_dispatcher fdi;
382 struct fuse_filehandle *fufh;
383 struct fuse_lseek_in *flsi;
384 struct fuse_lseek_out *flso;
385 struct mount *mp = vnode_mount(vp);
386 int err;
387
388 ASSERT_VOP_LOCKED(vp, __func__);
389
390 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
391 if (err)
392 return (err);
393 fdisp_init(&fdi, sizeof(*flsi));
394 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
395 flsi = fdi.indata;
396 flsi->fh = fufh->fh_id;
397 flsi->offset = *offp;
398 flsi->whence = whence;
399 err = fdisp_wait_answ(&fdi);
400 if (err == ENOSYS) {
401 fsess_set_notimpl(mp, FUSE_LSEEK);
402 } else if (err == ENXIO) {
403 /* Note: ENXIO means "no more hole/data regions until EOF" */
404 fsess_set_impl(mp, FUSE_LSEEK);
405 } else if (err == 0) {
406 fsess_set_impl(mp, FUSE_LSEEK);
407 flso = fdi.answ;
408 *offp = flso->offset;
409 }
410 fdisp_destroy(&fdi);
411
412 return (err);
413 }
414
415 /*
416 struct vnop_access_args {
417 struct vnode *a_vp;
418 #if VOP_ACCESS_TAKES_ACCMODE_T
419 accmode_t a_accmode;
420 #else
421 int a_mode;
422 #endif
423 struct ucred *a_cred;
424 struct thread *a_td;
425 };
426 */
427 static int
fuse_vnop_access(struct vop_access_args * ap)428 fuse_vnop_access(struct vop_access_args *ap)
429 {
430 struct vnode *vp = ap->a_vp;
431 int accmode = ap->a_accmode;
432 struct ucred *cred = ap->a_cred;
433
434 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
435
436 int err;
437
438 if (fuse_isdeadfs(vp)) {
439 if (vnode_isvroot(vp)) {
440 return 0;
441 }
442 return ENXIO;
443 }
444 if (!(data->dataflags & FSESS_INITED)) {
445 if (vnode_isvroot(vp)) {
446 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
447 (fuse_match_cred(data->daemoncred, cred) == 0)) {
448 return 0;
449 }
450 }
451 return EBADF;
452 }
453 if (vnode_islnk(vp)) {
454 return 0;
455 }
456
457 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
458 return err;
459 }
460
461 /*
462 * struct vop_advlock_args {
463 * struct vop_generic_args a_gen;
464 * struct vnode *a_vp;
465 * void *a_id;
466 * int a_op;
467 * struct flock *a_fl;
468 * int a_flags;
469 * }
470 */
471 static int
fuse_vnop_advlock(struct vop_advlock_args * ap)472 fuse_vnop_advlock(struct vop_advlock_args *ap)
473 {
474 struct vnode *vp = ap->a_vp;
475 struct flock *fl = ap->a_fl;
476 struct thread *td = curthread;
477 struct ucred *cred = td->td_ucred;
478 pid_t pid = td->td_proc->p_pid;
479 struct fuse_filehandle *fufh;
480 struct fuse_dispatcher fdi;
481 struct fuse_lk_in *fli;
482 struct fuse_lk_out *flo;
483 struct vattr vattr;
484 enum fuse_opcode op;
485 off_t size, start;
486 int dataflags, err;
487 int flags = ap->a_flags;
488
489 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
490
491 if (fuse_isdeadfs(vp)) {
492 return ENXIO;
493 }
494
495 switch(ap->a_op) {
496 case F_GETLK:
497 op = FUSE_GETLK;
498 break;
499 case F_SETLK:
500 if (flags & F_WAIT)
501 op = FUSE_SETLKW;
502 else
503 op = FUSE_SETLK;
504 break;
505 case F_UNLCK:
506 op = FUSE_SETLK;
507 break;
508 default:
509 return EINVAL;
510 }
511
512 if (!(dataflags & FSESS_POSIX_LOCKS))
513 return vop_stdadvlock(ap);
514 /* FUSE doesn't properly support flock until protocol 7.17 */
515 if (flags & F_FLOCK)
516 return vop_stdadvlock(ap);
517
518 vn_lock(vp, LK_SHARED | LK_RETRY);
519
520 switch (fl->l_whence) {
521 case SEEK_SET:
522 case SEEK_CUR:
523 /*
524 * Caller is responsible for adding any necessary offset
525 * when SEEK_CUR is used.
526 */
527 start = fl->l_start;
528 break;
529
530 case SEEK_END:
531 err = fuse_internal_getattr(vp, &vattr, cred, td);
532 if (err)
533 goto out;
534 size = vattr.va_size;
535 if (size > OFF_MAX ||
536 (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
537 err = EOVERFLOW;
538 goto out;
539 }
540 start = size + fl->l_start;
541 break;
542
543 default:
544 return (EINVAL);
545 }
546
547 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
548 if (err)
549 goto out;
550
551 fdisp_init(&fdi, sizeof(*fli));
552
553 fdisp_make_vp(&fdi, op, vp, td, cred);
554 fli = fdi.indata;
555 fli->fh = fufh->fh_id;
556 fli->owner = td->td_proc->p_pid;
557 fli->lk.start = start;
558 if (fl->l_len != 0)
559 fli->lk.end = start + fl->l_len - 1;
560 else
561 fli->lk.end = INT64_MAX;
562 fli->lk.type = fl->l_type;
563 fli->lk.pid = td->td_proc->p_pid;
564
565 err = fdisp_wait_answ(&fdi);
566 fdisp_destroy(&fdi);
567
568 if (err == 0 && op == FUSE_GETLK) {
569 flo = fdi.answ;
570 fl->l_type = flo->lk.type;
571 fl->l_whence = SEEK_SET;
572 if (flo->lk.type != F_UNLCK) {
573 fl->l_pid = flo->lk.pid;
574 fl->l_start = flo->lk.start;
575 if (flo->lk.end == INT64_MAX)
576 fl->l_len = 0;
577 else
578 fl->l_len = flo->lk.end - flo->lk.start + 1;
579 fl->l_start = flo->lk.start;
580 }
581 }
582
583 out:
584 VOP_UNLOCK(vp);
585 return err;
586 }
587
588 static int
fuse_vnop_allocate(struct vop_allocate_args * ap)589 fuse_vnop_allocate(struct vop_allocate_args *ap)
590 {
591 struct vnode *vp = ap->a_vp;
592 off_t *len = ap->a_len;
593 off_t *offset = ap->a_offset;
594 struct ucred *cred = ap->a_cred;
595 struct fuse_filehandle *fufh;
596 struct mount *mp = vnode_mount(vp);
597 struct fuse_dispatcher fdi;
598 struct fuse_fallocate_in *ffi;
599 struct uio io;
600 pid_t pid = curthread->td_proc->p_pid;
601 struct fuse_vnode_data *fvdat = VTOFUD(vp);
602 off_t filesize;
603 int err;
604
605 if (fuse_isdeadfs(vp))
606 return (ENXIO);
607
608 switch (vp->v_type) {
609 case VFIFO:
610 return (ESPIPE);
611 case VLNK:
612 case VREG:
613 if (vfs_isrdonly(mp))
614 return (EROFS);
615 break;
616 default:
617 return (ENODEV);
618 }
619
620 if (vfs_isrdonly(mp))
621 return (EROFS);
622
623 if (fsess_not_impl(mp, FUSE_FALLOCATE))
624 return (EINVAL);
625
626 io.uio_offset = *offset;
627 io.uio_resid = *len;
628 err = vn_rlimit_fsize(vp, &io, curthread);
629 if (err)
630 return (err);
631
632 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
633 if (err)
634 return (err);
635
636 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
637
638 err = fuse_vnode_size(vp, &filesize, cred, curthread);
639 if (err)
640 return (err);
641 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
642
643 fdisp_init(&fdi, sizeof(*ffi));
644 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
645 ffi = fdi.indata;
646 ffi->fh = fufh->fh_id;
647 ffi->offset = *offset;
648 ffi->length = *len;
649 ffi->mode = 0;
650 err = fdisp_wait_answ(&fdi);
651
652 if (err == ENOSYS) {
653 fsess_set_notimpl(mp, FUSE_FALLOCATE);
654 err = EINVAL;
655 } else if (err == EOPNOTSUPP) {
656 /*
657 * The file system server does not support FUSE_FALLOCATE with
658 * the supplied mode for this particular file.
659 */
660 err = EINVAL;
661 } else if (!err) {
662 *offset += *len;
663 *len = 0;
664 fuse_vnode_undirty_cached_timestamps(vp, false);
665 fuse_internal_clear_suid_on_write(vp, cred, curthread);
666 if (*offset > fvdat->cached_attrs.va_size) {
667 fuse_vnode_setsize(vp, *offset, false);
668 getnanouptime(&fvdat->last_local_modify);
669 }
670 }
671
672 fdisp_destroy(&fdi);
673 return (err);
674 }
675
676 /* {
677 struct vnode *a_vp;
678 daddr_t a_bn;
679 struct bufobj **a_bop;
680 daddr_t *a_bnp;
681 int *a_runp;
682 int *a_runb;
683 } */
684 static int
fuse_vnop_bmap(struct vop_bmap_args * ap)685 fuse_vnop_bmap(struct vop_bmap_args *ap)
686 {
687 struct vnode *vp = ap->a_vp;
688 struct bufobj **bo = ap->a_bop;
689 struct thread *td = curthread;
690 struct mount *mp;
691 struct fuse_dispatcher fdi;
692 struct fuse_bmap_in *fbi;
693 struct fuse_bmap_out *fbo;
694 struct fuse_data *data;
695 struct fuse_vnode_data *fvdat = VTOFUD(vp);
696 uint64_t biosize;
697 off_t fsize;
698 daddr_t lbn = ap->a_bn;
699 daddr_t *pbn = ap->a_bnp;
700 int *runp = ap->a_runp;
701 int *runb = ap->a_runb;
702 int error = 0;
703 int maxrun;
704
705 if (fuse_isdeadfs(vp)) {
706 return ENXIO;
707 }
708
709 mp = vnode_mount(vp);
710 data = fuse_get_mpdata(mp);
711 biosize = fuse_iosize(vp);
712 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
713 data->max_readahead_blocks);
714
715 if (bo != NULL)
716 *bo = &vp->v_bufobj;
717
718 /*
719 * The FUSE_BMAP operation does not include the runp and runb
720 * variables, so we must guess. Report nonzero contiguous runs so
721 * cluster_read will combine adjacent reads. It's worthwhile to reduce
722 * upcalls even if we don't know the true physical layout of the file.
723 *
724 * FUSE file systems may opt out of read clustering in two ways:
725 * * mounting with -onoclusterr
726 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
727 */
728 if (runb != NULL)
729 *runb = MIN(lbn, maxrun);
730 if (runp != NULL && maxrun == 0)
731 *runp = 0;
732 else if (runp != NULL) {
733 /*
734 * If the file's size is cached, use that value to calculate
735 * runp, even if the cache is expired. runp is only advisory,
736 * and the risk of getting it wrong is not worth the cost of
737 * another upcall.
738 */
739 if (fvdat->cached_attrs.va_size != VNOVAL)
740 fsize = fvdat->cached_attrs.va_size;
741 else
742 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
743 if (error == 0)
744 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
745 maxrun);
746 else
747 *runp = 0;
748 }
749
750 if (fsess_maybe_impl(mp, FUSE_BMAP)) {
751 fdisp_init(&fdi, sizeof(*fbi));
752 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
753 fbi = fdi.indata;
754 fbi->block = lbn;
755 fbi->blocksize = biosize;
756 error = fdisp_wait_answ(&fdi);
757 if (error == ENOSYS) {
758 fdisp_destroy(&fdi);
759 fsess_set_notimpl(mp, FUSE_BMAP);
760 error = 0;
761 } else {
762 fbo = fdi.answ;
763 if (error == 0 && pbn != NULL)
764 *pbn = fbo->block;
765 fdisp_destroy(&fdi);
766 return error;
767 }
768 }
769
770 /* If the daemon doesn't support BMAP, make up a sensible default */
771 if (pbn != NULL)
772 *pbn = lbn * btodb(biosize);
773 return (error);
774 }
775
776 /*
777 struct vop_close_args {
778 struct vnode *a_vp;
779 int a_fflag;
780 struct ucred *a_cred;
781 struct thread *a_td;
782 };
783 */
784 static int
fuse_vnop_close(struct vop_close_args * ap)785 fuse_vnop_close(struct vop_close_args *ap)
786 {
787 struct vnode *vp = ap->a_vp;
788 struct mount *mp = vnode_mount(vp);
789 struct ucred *cred = ap->a_cred;
790 int fflag = ap->a_fflag;
791 struct thread *td = ap->a_td;
792 pid_t pid = td->td_proc->p_pid;
793 struct fuse_vnode_data *fvdat = VTOFUD(vp);
794 int err = 0;
795
796 if (fuse_isdeadfs(vp))
797 return 0;
798 if (vnode_isdir(vp))
799 return 0;
800 if (fflag & IO_NDELAY)
801 return 0;
802
803 if (cred == NULL)
804 cred = td->td_ucred;
805
806 err = fuse_flush(vp, cred, pid, fflag);
807 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) {
808 struct vattr vap;
809 struct fuse_data *data;
810 int dataflags;
811 int access_e = 0;
812
813 data = fuse_get_mpdata(mp);
814 dataflags = data->dataflags;
815 if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
816 struct vattr va;
817
818 fuse_internal_getattr(vp, &va, cred, td);
819 access_e = vaccess(vp->v_type, va.va_mode, va.va_uid,
820 va.va_gid, VWRITE, cred);
821 }
822 if (access_e == 0) {
823 VATTR_NULL(&vap);
824 vap.va_atime = fvdat->cached_attrs.va_atime;
825 /*
826 * Ignore errors setting when setting atime. That
827 * should not cause close(2) to fail.
828 */
829 fuse_internal_setattr(vp, &vap, td, NULL);
830 }
831 }
832 /* TODO: close the file handle, if we're sure it's no longer used */
833 if ((fvdat->flag & FN_SIZECHANGE) != 0) {
834 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
835 }
836 return err;
837 }
838
839 /*
840 struct vop_copy_file_range_args {
841 struct vop_generic_args a_gen;
842 struct vnode *a_invp;
843 off_t *a_inoffp;
844 struct vnode *a_outvp;
845 off_t *a_outoffp;
846 size_t *a_lenp;
847 unsigned int a_flags;
848 struct ucred *a_incred;
849 struct ucred *a_outcred;
850 struct thread *a_fsizetd;
851 }
852 */
853 static int
fuse_vnop_copy_file_range(struct vop_copy_file_range_args * ap)854 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
855 {
856 struct vnode *invp = ap->a_invp;
857 struct vnode *outvp = ap->a_outvp;
858 struct mount *mp = vnode_mount(invp);
859 struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
860 struct fuse_dispatcher fdi;
861 struct fuse_filehandle *infufh, *outfufh;
862 struct fuse_copy_file_range_in *fcfri;
863 struct ucred *incred = ap->a_incred;
864 struct ucred *outcred = ap->a_outcred;
865 struct fuse_write_out *fwo;
866 struct thread *td;
867 struct uio io;
868 off_t outfilesize;
869 ssize_t r = 0;
870 pid_t pid;
871 int err;
872
873 err = ENOSYS;
874 if (mp == NULL || mp != vnode_mount(outvp))
875 goto fallback;
876
877 if (incred->cr_uid != outcred->cr_uid)
878 goto fallback;
879
880 if (incred->cr_groups[0] != outcred->cr_groups[0])
881 goto fallback;
882
883 /* Caller busied mp, mnt_data can be safely accessed. */
884 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
885 goto fallback;
886
887 if (ap->a_fsizetd == NULL)
888 td = curthread;
889 else
890 td = ap->a_fsizetd;
891 pid = td->td_proc->p_pid;
892
893 vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE);
894 if (invp->v_data == NULL || outvp->v_data == NULL) {
895 err = EBADF;
896 goto unlock;
897 }
898
899 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
900 if (err)
901 goto unlock;
902
903 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
904 if (err)
905 goto unlock;
906
907 io.uio_resid = *ap->a_lenp;
908 if (ap->a_fsizetd) {
909 io.uio_offset = *ap->a_outoffp;
910 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
911 if (err != 0)
912 goto unlock;
913 }
914
915 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
916 if (err)
917 goto unlock;
918
919 vnode_pager_clean_sync(invp);
920 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
921 *ap->a_outoffp + io.uio_resid);
922 if (err)
923 goto unlock;
924
925 fdisp_init(&fdi, sizeof(*fcfri));
926 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
927 fcfri = fdi.indata;
928 fcfri->fh_in = infufh->fh_id;
929 fcfri->off_in = *ap->a_inoffp;
930 fcfri->nodeid_out = VTOI(outvp);
931 fcfri->fh_out = outfufh->fh_id;
932 fcfri->off_out = *ap->a_outoffp;
933 fcfri->len = io.uio_resid;
934 fcfri->flags = 0;
935
936 err = fdisp_wait_answ(&fdi);
937 if (err == 0) {
938 fwo = fdi.answ;
939 *ap->a_lenp = fwo->size;
940 *ap->a_inoffp += fwo->size;
941 *ap->a_outoffp += fwo->size;
942 fuse_internal_clear_suid_on_write(outvp, outcred, td);
943 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
944 fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
945 getnanouptime(&outfvdat->last_local_modify);
946 }
947 fuse_vnode_update(invp, FN_ATIMECHANGE);
948 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
949 }
950 fdisp_destroy(&fdi);
951
952 unlock:
953 if (invp != outvp)
954 VOP_UNLOCK(invp);
955 VOP_UNLOCK(outvp);
956
957 if (err == ENOSYS)
958 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
959 fallback:
960
961 /*
962 * No need to call vn_rlimit_fsizex_res before return, since the uio is
963 * local.
964 */
965 return (err);
966 }
967
968 static void
fdisp_make_mknod_for_fallback(struct fuse_dispatcher * fdip,struct componentname * cnp,struct vnode * dvp,uint64_t parentnid,struct thread * td,struct ucred * cred,mode_t mode,enum fuse_opcode * op)969 fdisp_make_mknod_for_fallback(
970 struct fuse_dispatcher *fdip,
971 struct componentname *cnp,
972 struct vnode *dvp,
973 uint64_t parentnid,
974 struct thread *td,
975 struct ucred *cred,
976 mode_t mode,
977 enum fuse_opcode *op)
978 {
979 struct fuse_mknod_in *fmni;
980
981 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
982 *op = FUSE_MKNOD;
983 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
984 fmni = fdip->indata;
985 fmni->mode = mode;
986 fmni->rdev = 0;
987 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
988 cnp->cn_namelen);
989 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
990 }
991 /*
992 struct vnop_create_args {
993 struct vnode *a_dvp;
994 struct vnode **a_vpp;
995 struct componentname *a_cnp;
996 struct vattr *a_vap;
997 };
998 */
999 static int
fuse_vnop_create(struct vop_create_args * ap)1000 fuse_vnop_create(struct vop_create_args *ap)
1001 {
1002 struct vnode *dvp = ap->a_dvp;
1003 struct vnode **vpp = ap->a_vpp;
1004 struct componentname *cnp = ap->a_cnp;
1005 struct vattr *vap = ap->a_vap;
1006 struct thread *td = curthread;
1007 struct ucred *cred = cnp->cn_cred;
1008
1009 struct fuse_data *data;
1010 struct fuse_create_in *fci;
1011 struct fuse_entry_out *feo;
1012 struct fuse_open_out *foo;
1013 struct fuse_dispatcher fdi, fdi2;
1014 struct fuse_dispatcher *fdip = &fdi;
1015 struct fuse_dispatcher *fdip2 = NULL;
1016
1017 int err;
1018
1019 struct mount *mp = vnode_mount(dvp);
1020 data = fuse_get_mpdata(mp);
1021 uint64_t parentnid = VTOFUD(dvp)->nid;
1022 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1023 enum fuse_opcode op;
1024 int flags;
1025
1026 if (fuse_isdeadfs(dvp))
1027 return ENXIO;
1028
1029 /* FUSE expects sockets to be created with FUSE_MKNOD */
1030 if (vap->va_type == VSOCK)
1031 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1032
1033 /*
1034 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
1035 * writable mode makes sense, and we might as well include readability
1036 * too.
1037 */
1038 flags = O_RDWR;
1039
1040 bzero(&fdi, sizeof(fdi));
1041
1042 if (vap->va_type != VREG)
1043 return (EINVAL);
1044
1045 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1046 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
1047 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1048 cred, mode, &op);
1049 } else {
1050 /* Use FUSE_CREATE */
1051 size_t insize;
1052
1053 op = FUSE_CREATE;
1054 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1055 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1056 fci = fdip->indata;
1057 fci->mode = mode;
1058 fci->flags = O_CREAT | flags;
1059 if (fuse_libabi_geq(data, 7, 12)) {
1060 insize = sizeof(*fci);
1061 fci->umask = td->td_proc->p_pd->pd_cmask;
1062 } else {
1063 insize = sizeof(struct fuse_open_in);
1064 }
1065
1066 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1067 cnp->cn_namelen);
1068 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1069 }
1070
1071 err = fdisp_wait_answ(fdip);
1072
1073 if (err) {
1074 if (err == ENOSYS && op == FUSE_CREATE) {
1075 fsess_set_notimpl(mp, FUSE_CREATE);
1076 fdisp_destroy(fdip);
1077 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1078 parentnid, td, cred, mode, &op);
1079 err = fdisp_wait_answ(fdip);
1080 }
1081 if (err)
1082 goto out;
1083 }
1084
1085 feo = fdip->answ;
1086
1087 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1088 goto out;
1089 }
1090
1091 if (op == FUSE_CREATE) {
1092 if (fuse_libabi_geq(data, 7, 9))
1093 foo = (struct fuse_open_out*)(feo + 1);
1094 else
1095 foo = (struct fuse_open_out*)((char*)feo +
1096 FUSE_COMPAT_ENTRY_OUT_SIZE);
1097 } else {
1098 /* Issue a separate FUSE_OPEN */
1099 struct fuse_open_in *foi;
1100
1101 fdip2 = &fdi2;
1102 fdisp_init(fdip2, sizeof(*foi));
1103 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1104 cred);
1105 foi = fdip2->indata;
1106 foi->flags = flags;
1107 err = fdisp_wait_answ(fdip2);
1108 if (err)
1109 goto out;
1110 foo = fdip2->answ;
1111 }
1112 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1113 if (err) {
1114 struct fuse_release_in *fri;
1115 uint64_t nodeid = feo->nodeid;
1116 uint64_t fh_id = foo->fh;
1117
1118 fdisp_destroy(fdip);
1119 fdisp_init(fdip, sizeof(*fri));
1120 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1121 fri = fdip->indata;
1122 fri->fh = fh_id;
1123 fri->flags = flags;
1124 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1125 fuse_insert_message(fdip->tick, false);
1126 goto out;
1127 }
1128 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1129 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1130 feo->attr_valid_nsec, NULL, true);
1131
1132 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1133 fuse_vnode_open(*vpp, foo->open_flags, td);
1134 /*
1135 * Purge the parent's attribute cache because the daemon should've
1136 * updated its mtime and ctime
1137 */
1138 fuse_vnode_clear_attr_cache(dvp);
1139 cache_purge_negative(dvp);
1140
1141 out:
1142 if (fdip2)
1143 fdisp_destroy(fdip2);
1144 fdisp_destroy(fdip);
1145 return err;
1146 }
1147
1148 /*
1149 struct vnop_fdatasync_args {
1150 struct vop_generic_args a_gen;
1151 struct vnode * a_vp;
1152 struct thread * a_td;
1153 };
1154 */
1155 static int
fuse_vnop_fdatasync(struct vop_fdatasync_args * ap)1156 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1157 {
1158 struct vnode *vp = ap->a_vp;
1159 struct thread *td = ap->a_td;
1160 int waitfor = MNT_WAIT;
1161
1162 int err = 0;
1163
1164 if (fuse_isdeadfs(vp)) {
1165 return 0;
1166 }
1167 if ((err = vop_stdfdatasync_buf(ap)))
1168 return err;
1169
1170 return fuse_internal_fsync(vp, td, waitfor, true);
1171 }
1172
1173 /*
1174 struct vnop_fsync_args {
1175 struct vop_generic_args a_gen;
1176 struct vnode * a_vp;
1177 int a_waitfor;
1178 struct thread * a_td;
1179 };
1180 */
1181 static int
fuse_vnop_fsync(struct vop_fsync_args * ap)1182 fuse_vnop_fsync(struct vop_fsync_args *ap)
1183 {
1184 struct vnode *vp = ap->a_vp;
1185 struct thread *td = ap->a_td;
1186 int waitfor = ap->a_waitfor;
1187 int err = 0;
1188
1189 if (fuse_isdeadfs(vp)) {
1190 return 0;
1191 }
1192 if ((err = vop_stdfsync(ap)))
1193 return err;
1194
1195 return fuse_internal_fsync(vp, td, waitfor, false);
1196 }
1197
1198 /*
1199 struct vnop_getattr_args {
1200 struct vnode *a_vp;
1201 struct vattr *a_vap;
1202 struct ucred *a_cred;
1203 struct thread *a_td;
1204 };
1205 */
1206 static int
fuse_vnop_getattr(struct vop_getattr_args * ap)1207 fuse_vnop_getattr(struct vop_getattr_args *ap)
1208 {
1209 struct vnode *vp = ap->a_vp;
1210 struct vattr *vap = ap->a_vap;
1211 struct ucred *cred = ap->a_cred;
1212 struct thread *td = curthread;
1213
1214 int err = 0;
1215 int dataflags;
1216
1217 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
1218
1219 /* Note that we are not bailing out on a dead file system just yet. */
1220
1221 if (!(dataflags & FSESS_INITED)) {
1222 if (!vnode_isvroot(vp)) {
1223 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
1224 err = ENOTCONN;
1225 return err;
1226 } else {
1227 goto fake;
1228 }
1229 }
1230 err = fuse_internal_getattr(vp, vap, cred, td);
1231 if (err == ENOTCONN && vnode_isvroot(vp)) {
1232 /* see comment in fuse_vfsop_statfs() */
1233 goto fake;
1234 } else {
1235 return err;
1236 }
1237
1238 fake:
1239 bzero(vap, sizeof(*vap));
1240 vap->va_type = vnode_vtype(vp);
1241
1242 return 0;
1243 }
1244
1245 /*
1246 struct vnop_inactive_args {
1247 struct vnode *a_vp;
1248 };
1249 */
1250 static int
fuse_vnop_inactive(struct vop_inactive_args * ap)1251 fuse_vnop_inactive(struct vop_inactive_args *ap)
1252 {
1253 struct vnode *vp = ap->a_vp;
1254 struct thread *td = curthread;
1255
1256 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1257 struct fuse_filehandle *fufh, *fufh_tmp;
1258
1259 int need_flush = 1;
1260
1261 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1262 if (need_flush && vp->v_type == VREG) {
1263 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1264 fuse_vnode_savesize(vp, NULL, 0);
1265 }
1266 if ((fvdat->flag & FN_REVOKED) != 0)
1267 fuse_io_invalbuf(vp, td);
1268 else
1269 fuse_io_flushbuf(vp, MNT_WAIT, td);
1270 need_flush = 0;
1271 }
1272 fuse_filehandle_close(vp, fufh, td, NULL);
1273 }
1274
1275 if ((fvdat->flag & FN_REVOKED) != 0)
1276 vrecycle(vp);
1277
1278 return 0;
1279 }
1280
1281 /*
1282 struct vnop_ioctl_args {
1283 struct vnode *a_vp;
1284 u_long a_command;
1285 caddr_t a_data;
1286 int a_fflag;
1287 struct ucred *a_cred;
1288 struct thread *a_td;
1289 };
1290 */
1291 static int
fuse_vnop_ioctl(struct vop_ioctl_args * ap)1292 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1293 {
1294 struct vnode *vp = ap->a_vp;
1295 struct mount *mp = vnode_mount(vp);
1296 struct ucred *cred = ap->a_cred;
1297 off_t *offp;
1298 pid_t pid = ap->a_td->td_proc->p_pid;
1299 int err;
1300
1301 switch (ap->a_command) {
1302 case FIOSEEKDATA:
1303 case FIOSEEKHOLE:
1304 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1305 if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1306 int whence;
1307
1308 offp = ap->a_data;
1309 if (ap->a_command == FIOSEEKDATA)
1310 whence = SEEK_DATA;
1311 else
1312 whence = SEEK_HOLE;
1313
1314 vn_lock(vp, LK_SHARED | LK_RETRY);
1315 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1316 whence);
1317 VOP_UNLOCK(vp);
1318 }
1319 if (fsess_not_impl(mp, FUSE_LSEEK))
1320 err = vop_stdioctl(ap);
1321 break;
1322 default:
1323 /* TODO: implement FUSE_IOCTL */
1324 err = ENOTTY;
1325 break;
1326 }
1327 return (err);
1328 }
1329
1330
1331 /*
1332 struct vnop_link_args {
1333 struct vnode *a_tdvp;
1334 struct vnode *a_vp;
1335 struct componentname *a_cnp;
1336 };
1337 */
1338 static int
fuse_vnop_link(struct vop_link_args * ap)1339 fuse_vnop_link(struct vop_link_args *ap)
1340 {
1341 struct vnode *vp = ap->a_vp;
1342 struct vnode *tdvp = ap->a_tdvp;
1343 struct componentname *cnp = ap->a_cnp;
1344
1345 struct vattr *vap = VTOVA(vp);
1346
1347 struct fuse_dispatcher fdi;
1348 struct fuse_entry_out *feo;
1349 struct fuse_link_in fli;
1350
1351 int err;
1352
1353 if (fuse_isdeadfs(vp)) {
1354 return ENXIO;
1355 }
1356 if (vnode_mount(tdvp) != vnode_mount(vp)) {
1357 return EXDEV;
1358 }
1359
1360 /*
1361 * This is a seatbelt check to protect naive userspace filesystems from
1362 * themselves and the limitations of the FUSE IPC protocol. If a
1363 * filesystem does not allow attribute caching, assume it is capable of
1364 * validating that nlink does not overflow.
1365 */
1366 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1367 return EMLINK;
1368 fli.oldnodeid = VTOI(vp);
1369
1370 fdisp_init(&fdi, 0);
1371 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1372 FUSE_LINK, &fli, sizeof(fli), &fdi);
1373 if ((err = fdisp_wait_answ(&fdi))) {
1374 goto out;
1375 }
1376 feo = fdi.answ;
1377
1378 if (fli.oldnodeid != feo->nodeid) {
1379 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1380 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE,
1381 "Assigned wrong inode for a hard link.");
1382 fuse_vnode_clear_attr_cache(vp);
1383 fuse_vnode_clear_attr_cache(tdvp);
1384 err = EIO;
1385 goto out;
1386 }
1387
1388 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1389 if (!err) {
1390 /*
1391 * Purge the parent's attribute cache because the daemon
1392 * should've updated its mtime and ctime
1393 */
1394 fuse_vnode_clear_attr_cache(tdvp);
1395 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1396 feo->attr_valid_nsec, NULL, true);
1397 }
1398 out:
1399 fdisp_destroy(&fdi);
1400 return err;
1401 }
1402
1403 struct fuse_lookup_alloc_arg {
1404 struct fuse_entry_out *feo;
1405 struct componentname *cnp;
1406 uint64_t nid;
1407 __enum_uint8(vtype) vtyp;
1408 };
1409
1410 /* Callback for vn_get_ino */
1411 static int
fuse_lookup_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)1412 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1413 {
1414 struct fuse_lookup_alloc_arg *flaa = arg;
1415
1416 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1417 flaa->vtyp);
1418 }
1419
1420 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1421 "int", "struct timespec*", "struct timespec*");
1422 /*
1423 struct vnop_lookup_args {
1424 struct vnodeop_desc *a_desc;
1425 struct vnode *a_dvp;
1426 struct vnode **a_vpp;
1427 struct componentname *a_cnp;
1428 };
1429 */
1430 int
fuse_vnop_lookup(struct vop_lookup_args * ap)1431 fuse_vnop_lookup(struct vop_lookup_args *ap)
1432 {
1433 struct vnode *dvp = ap->a_dvp;
1434 struct vnode **vpp = ap->a_vpp;
1435 struct componentname *cnp = ap->a_cnp;
1436 struct thread *td = curthread;
1437 struct ucred *cred = cnp->cn_cred;
1438 struct timespec now;
1439
1440 int nameiop = cnp->cn_nameiop;
1441 bool isdotdot = cnp->cn_flags & ISDOTDOT;
1442 bool islastcn = cnp->cn_flags & ISLASTCN;
1443 struct mount *mp = vnode_mount(dvp);
1444 struct fuse_data *data = fuse_get_mpdata(mp);
1445 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1446 bool is_dot;
1447
1448 int err = 0;
1449 int lookup_err = 0;
1450 struct vnode *vp = NULL;
1451
1452 struct fuse_dispatcher fdi;
1453 bool did_lookup = false;
1454 struct fuse_entry_out *feo = NULL;
1455 __enum_uint8(vtype) vtyp; /* vnode type of target */
1456
1457 uint64_t nid;
1458
1459 if (fuse_isdeadfs(dvp)) {
1460 *vpp = NULL;
1461 return ENXIO;
1462 }
1463 if (!vnode_isdir(dvp))
1464 return ENOTDIR;
1465
1466 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1467 return EROFS;
1468
1469 if ((cnp->cn_flags & NOEXECCHECK) != 0)
1470 cnp->cn_flags &= ~NOEXECCHECK;
1471 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1472 return err;
1473
1474 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1475 if (isdotdot && !(data->dataflags & FSESS_EXPORT_SUPPORT)) {
1476 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1477 /*
1478 * Since the file system doesn't support ".." lookups,
1479 * we have no way to find this entry.
1480 */
1481 return ESTALE;
1482 }
1483 nid = VTOFUD(dvp)->parent_nid;
1484 if (nid == 0)
1485 return ENOENT;
1486 /* .. is obviously a directory */
1487 vtyp = VDIR;
1488 } else if (is_dot) {
1489 nid = VTOI(dvp);
1490 /* . is obviously a directory */
1491 vtyp = VDIR;
1492 } else {
1493 struct timespec timeout;
1494 int ncpticks; /* here to accommodate for API contract */
1495
1496 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1497 getnanouptime(&now);
1498 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1499 switch (err) {
1500 case -1: /* positive match */
1501 if (timespeccmp(&timeout, &now, >)) {
1502 counter_u64_add(fuse_lookup_cache_hits, 1);
1503 } else {
1504 /* Cache timeout */
1505 counter_u64_add(fuse_lookup_cache_misses, 1);
1506 bintime_clear(
1507 &VTOFUD(*vpp)->entry_cache_timeout);
1508 cache_purge(*vpp);
1509 if (dvp != *vpp)
1510 vput(*vpp);
1511 else
1512 vrele(*vpp);
1513 *vpp = NULL;
1514 break;
1515 }
1516 return 0;
1517
1518 case 0: /* no match in cache */
1519 counter_u64_add(fuse_lookup_cache_misses, 1);
1520 break;
1521
1522 case ENOENT: /* negative match */
1523 if (timespeccmp(&timeout, &now, <=)) {
1524 /* Cache timeout */
1525 cache_purge_negative(dvp);
1526 break;
1527 }
1528 /* fall through */
1529 default:
1530 return err;
1531 }
1532
1533 fdisp_init(&fdi, cnp->cn_namelen + 1);
1534 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1535
1536 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1537 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1538 lookup_err = fdisp_wait_answ(&fdi);
1539 did_lookup = true;
1540
1541 if (!lookup_err) {
1542 /* lookup call succeeded */
1543 feo = (struct fuse_entry_out *)fdi.answ;
1544 nid = feo->nodeid;
1545 if (nid == 0) {
1546 /* zero nodeid means ENOENT and cache it */
1547 struct timespec timeout;
1548
1549 fdi.answ_stat = ENOENT;
1550 lookup_err = ENOENT;
1551 if (cnp->cn_flags & MAKEENTRY) {
1552 fuse_validity_2_timespec(feo, &timeout);
1553 /* Use the same entry_time for .. as for
1554 * the file itself. That doesn't honor
1555 * exactly what the fuse server tells
1556 * us, but to do otherwise would require
1557 * another cache lookup at this point.
1558 */
1559 struct timespec *dtsp = NULL;
1560 cache_enter_time(dvp, *vpp, cnp,
1561 &timeout, dtsp);
1562 }
1563 }
1564 vtyp = IFTOVT(feo->attr.mode);
1565 }
1566 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1567 fdisp_destroy(&fdi);
1568 return lookup_err;
1569 }
1570 }
1571 /* lookup_err, if non-zero, must be ENOENT at this point */
1572
1573 if (lookup_err) {
1574 /* Entry not found */
1575 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1576 if (default_permissions)
1577 err = fuse_internal_access(dvp, VWRITE, td,
1578 cred);
1579 else
1580 err = 0;
1581 if (!err) {
1582 err = EJUSTRETURN;
1583 }
1584 } else {
1585 err = ENOENT;
1586 }
1587 } else {
1588 /* Entry was found */
1589 if (isdotdot) {
1590 struct fuse_lookup_alloc_arg flaa;
1591
1592 flaa.nid = nid;
1593 flaa.feo = feo;
1594 flaa.cnp = cnp;
1595 flaa.vtyp = vtyp;
1596 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1597 &vp);
1598 *vpp = vp;
1599 } else if (nid == VTOI(dvp)) {
1600 if (is_dot) {
1601 vref(dvp);
1602 *vpp = dvp;
1603 } else {
1604 fuse_warn(fuse_get_mpdata(mp),
1605 FSESS_WARN_ILLEGAL_INODE,
1606 "Assigned same inode to both parent and "
1607 "child.");
1608 err = EIO;
1609 }
1610
1611 } else {
1612 struct fuse_vnode_data *fvdat;
1613
1614 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1615 &vp, cnp, vtyp);
1616 if (err)
1617 goto out;
1618 *vpp = vp;
1619 fvdat = VTOFUD(vp);
1620
1621 MPASS(feo != NULL);
1622 if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1623 /*
1624 * Attributes from the server are definitely
1625 * newer than the last attributes we sent to
1626 * the server, so cache them.
1627 */
1628 fuse_internal_cache_attrs(*vpp, &feo->attr,
1629 feo->attr_valid, feo->attr_valid_nsec,
1630 NULL, true);
1631 }
1632 fuse_validity_2_bintime(feo->entry_valid,
1633 feo->entry_valid_nsec,
1634 &fvdat->entry_cache_timeout);
1635
1636 if ((nameiop == DELETE || nameiop == RENAME) &&
1637 islastcn && default_permissions)
1638 {
1639 struct vattr dvattr;
1640
1641 err = fuse_internal_access(dvp, VWRITE, td,
1642 cred);
1643 if (err != 0)
1644 goto out;
1645 /*
1646 * if the parent's sticky bit is set, check
1647 * whether we're allowed to remove the file.
1648 * Need to figure out the vnode locking to make
1649 * this work.
1650 */
1651 fuse_internal_getattr(dvp, &dvattr, cred, td);
1652 if ((dvattr.va_mode & S_ISTXT) &&
1653 fuse_internal_access(dvp, VADMIN, td,
1654 cred) &&
1655 fuse_internal_access(*vpp, VADMIN, td,
1656 cred)) {
1657 err = EPERM;
1658 goto out;
1659 }
1660 }
1661 }
1662 }
1663 out:
1664 if (err) {
1665 if (vp != NULL && dvp != vp)
1666 vput(vp);
1667 else if (vp != NULL)
1668 vrele(vp);
1669 *vpp = NULL;
1670 }
1671 if (did_lookup)
1672 fdisp_destroy(&fdi);
1673
1674 return err;
1675 }
1676
1677 /*
1678 struct vnop_mkdir_args {
1679 struct vnode *a_dvp;
1680 struct vnode **a_vpp;
1681 struct componentname *a_cnp;
1682 struct vattr *a_vap;
1683 };
1684 */
1685 static int
fuse_vnop_mkdir(struct vop_mkdir_args * ap)1686 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1687 {
1688 struct vnode *dvp = ap->a_dvp;
1689 struct vnode **vpp = ap->a_vpp;
1690 struct componentname *cnp = ap->a_cnp;
1691 struct vattr *vap = ap->a_vap;
1692
1693 struct fuse_mkdir_in fmdi;
1694
1695 if (fuse_isdeadfs(dvp)) {
1696 return ENXIO;
1697 }
1698 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1699 fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1700
1701 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1702 sizeof(fmdi), VDIR));
1703 }
1704
1705 /*
1706 struct vnop_mknod_args {
1707 struct vnode *a_dvp;
1708 struct vnode **a_vpp;
1709 struct componentname *a_cnp;
1710 struct vattr *a_vap;
1711 };
1712 */
1713 static int
fuse_vnop_mknod(struct vop_mknod_args * ap)1714 fuse_vnop_mknod(struct vop_mknod_args *ap)
1715 {
1716
1717 struct vnode *dvp = ap->a_dvp;
1718 struct vnode **vpp = ap->a_vpp;
1719 struct componentname *cnp = ap->a_cnp;
1720 struct vattr *vap = ap->a_vap;
1721
1722 if (fuse_isdeadfs(dvp))
1723 return ENXIO;
1724
1725 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1726 }
1727
1728 /*
1729 struct vop_open_args {
1730 struct vnode *a_vp;
1731 int a_mode;
1732 struct ucred *a_cred;
1733 struct thread *a_td;
1734 int a_fdidx; / struct file *a_fp;
1735 };
1736 */
1737 static int
fuse_vnop_open(struct vop_open_args * ap)1738 fuse_vnop_open(struct vop_open_args *ap)
1739 {
1740 struct vnode *vp = ap->a_vp;
1741 int a_mode = ap->a_mode;
1742 struct thread *td = ap->a_td;
1743 struct ucred *cred = ap->a_cred;
1744 pid_t pid = td->td_proc->p_pid;
1745
1746 if (fuse_isdeadfs(vp))
1747 return ENXIO;
1748 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1749 return (EOPNOTSUPP);
1750 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1751 return EINVAL;
1752
1753 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1754 fuse_vnode_open(vp, 0, td);
1755 return 0;
1756 }
1757
1758 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1759 }
1760
1761 static int
fuse_vnop_pathconf(struct vop_pathconf_args * ap)1762 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1763 {
1764 struct vnode *vp = ap->a_vp;
1765 struct mount *mp;
1766 struct fuse_filehandle *fufh;
1767 int err;
1768 bool closefufh = false;
1769
1770 switch (ap->a_name) {
1771 case _PC_FILESIZEBITS:
1772 *ap->a_retval = 64;
1773 return (0);
1774 case _PC_NAME_MAX:
1775 *ap->a_retval = NAME_MAX;
1776 return (0);
1777 case _PC_LINK_MAX:
1778 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1779 return (0);
1780 case _PC_SYMLINK_MAX:
1781 *ap->a_retval = MAXPATHLEN;
1782 return (0);
1783 case _PC_NO_TRUNC:
1784 *ap->a_retval = 1;
1785 return (0);
1786 case _PC_MIN_HOLE_SIZE:
1787 /*
1788 * The FUSE protocol provides no mechanism for a server to
1789 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead,
1790 * return EINVAL if the server does not support FUSE_LSEEK, or
1791 * 1 if it does.
1792 */
1793 mp = vnode_mount(vp);
1794 if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1795 !fsess_not_impl(mp, FUSE_LSEEK)) {
1796 off_t offset = 0;
1797
1798 /*
1799 * Issue a FUSE_LSEEK to find out if it's supported.
1800 * Use SEEK_DATA instead of SEEK_HOLE, because the
1801 * latter generally requires sequential scans of file
1802 * metadata, which can be slow.
1803 */
1804 err = fuse_vnop_do_lseek(vp, curthread,
1805 curthread->td_ucred, curthread->td_proc->p_pid,
1806 &offset, SEEK_DATA);
1807 if (err == EBADF) {
1808 /*
1809 * pathconf() doesn't necessarily open the
1810 * file. So we may need to do it here.
1811 */
1812 err = fuse_filehandle_open(vp, FREAD, &fufh,
1813 curthread, curthread->td_ucred);
1814 if (err == 0) {
1815 closefufh = true;
1816 err = fuse_vnop_do_lseek(vp, curthread,
1817 curthread->td_ucred,
1818 curthread->td_proc->p_pid, &offset,
1819 SEEK_DATA);
1820 }
1821 if (closefufh)
1822 fuse_filehandle_close(vp, fufh,
1823 curthread, curthread->td_ucred);
1824 }
1825
1826 }
1827
1828 if (fsess_is_impl(mp, FUSE_LSEEK)) {
1829 *ap->a_retval = 1;
1830 return (0);
1831 } else if (fsess_not_impl(mp, FUSE_LSEEK)) {
1832 /* FUSE_LSEEK is not implemented */
1833 return (EINVAL);
1834 } else {
1835 return (err);
1836 }
1837 default:
1838 return (vop_stdpathconf(ap));
1839 }
1840 }
1841
1842 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1843 "struct uio*", "struct ucred*");
1844 /*
1845 struct vnop_read_args {
1846 struct vnode *a_vp;
1847 struct uio *a_uio;
1848 int a_ioflag;
1849 struct ucred *a_cred;
1850 };
1851 */
1852 static int
fuse_vnop_read(struct vop_read_args * ap)1853 fuse_vnop_read(struct vop_read_args *ap)
1854 {
1855 struct vnode *vp = ap->a_vp;
1856 struct uio *uio = ap->a_uio;
1857 int ioflag = ap->a_ioflag;
1858 struct ucred *cred = ap->a_cred;
1859 pid_t pid = curthread->td_proc->p_pid;
1860 struct fuse_filehandle *fufh;
1861 int err;
1862 bool closefufh = false, directio;
1863
1864 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1865
1866 if (fuse_isdeadfs(vp)) {
1867 return ENXIO;
1868 }
1869
1870 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1871 ioflag |= IO_DIRECT;
1872 }
1873
1874 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1875 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1876 /*
1877 * nfsd will do I/O without first doing VOP_OPEN. We
1878 * must implicitly open the file here
1879 */
1880 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1881 closefufh = true;
1882 }
1883 if (err) {
1884 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1885 return err;
1886 }
1887
1888 /*
1889 * Ideally, when the daemon asks for direct io at open time, the
1890 * standard file flag should be set according to this, so that would
1891 * just change the default mode, which later on could be changed via
1892 * fcntl(2).
1893 * But this doesn't work, the O_DIRECT flag gets cleared at some point
1894 * (don't know where). So to make any use of the Fuse direct_io option,
1895 * we hardwire it into the file's private data (similarly to Linux,
1896 * btw.).
1897 */
1898 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1899
1900 fuse_vnode_update(vp, FN_ATIMECHANGE);
1901 if (directio) {
1902 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1903 err = fuse_read_directbackend(vp, uio, cred, fufh);
1904 } else {
1905 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1906 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1907 }
1908
1909 if (closefufh)
1910 fuse_filehandle_close(vp, fufh, curthread, cred);
1911
1912 return (err);
1913 }
1914
1915 /*
1916 struct vnop_readdir_args {
1917 struct vnode *a_vp;
1918 struct uio *a_uio;
1919 struct ucred *a_cred;
1920 int *a_eofflag;
1921 int *a_ncookies;
1922 uint64_t **a_cookies;
1923 };
1924 */
1925 static int
fuse_vnop_readdir(struct vop_readdir_args * ap)1926 fuse_vnop_readdir(struct vop_readdir_args *ap)
1927 {
1928 struct vnode *vp = ap->a_vp;
1929 struct uio *uio = ap->a_uio;
1930 struct ucred *cred = ap->a_cred;
1931 struct fuse_filehandle *fufh = NULL;
1932 struct mount *mp = vnode_mount(vp);
1933 struct fuse_iov cookediov;
1934 int err = 0;
1935 uint64_t *cookies;
1936 ssize_t tresid;
1937 int ncookies;
1938 bool closefufh = false;
1939 pid_t pid = curthread->td_proc->p_pid;
1940
1941 if (ap->a_eofflag)
1942 *ap->a_eofflag = 0;
1943 if (fuse_isdeadfs(vp)) {
1944 return ENXIO;
1945 }
1946 if (uio_resid(uio) < sizeof(struct dirent))
1947 return EINVAL;
1948
1949 tresid = uio->uio_resid;
1950 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1951 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1952 KASSERT(!fsess_is_impl(mp, FUSE_OPENDIR),
1953 ("FUSE file systems that implement "
1954 "FUSE_OPENDIR should not be exported"));
1955 /*
1956 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1957 * must implicitly open the directory here.
1958 */
1959 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1960 closefufh = true;
1961 }
1962 if (err)
1963 return (err);
1964 if (ap->a_ncookies != NULL) {
1965 ncookies = uio->uio_resid /
1966 (offsetof(struct dirent, d_name) + 4) + 1;
1967 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1968 *ap->a_ncookies = ncookies;
1969 *ap->a_cookies = cookies;
1970 } else {
1971 ncookies = 0;
1972 cookies = NULL;
1973 }
1974 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1975 fiov_init(&cookediov, DIRCOOKEDSIZE);
1976
1977 err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1978 &ncookies, cookies);
1979
1980 fiov_teardown(&cookediov);
1981 if (closefufh)
1982 fuse_filehandle_close(vp, fufh, curthread, cred);
1983
1984 if (ap->a_ncookies != NULL) {
1985 if (err == 0) {
1986 *ap->a_ncookies -= ncookies;
1987 } else {
1988 free(*ap->a_cookies, M_TEMP);
1989 *ap->a_ncookies = 0;
1990 *ap->a_cookies = NULL;
1991 }
1992 }
1993 if (err == 0 && tresid == uio->uio_resid)
1994 *ap->a_eofflag = 1;
1995
1996 return err;
1997 }
1998
1999 /*
2000 struct vnop_readlink_args {
2001 struct vnode *a_vp;
2002 struct uio *a_uio;
2003 struct ucred *a_cred;
2004 };
2005 */
2006 static int
fuse_vnop_readlink(struct vop_readlink_args * ap)2007 fuse_vnop_readlink(struct vop_readlink_args *ap)
2008 {
2009 struct vnode *vp = ap->a_vp;
2010 struct uio *uio = ap->a_uio;
2011 struct ucred *cred = ap->a_cred;
2012
2013 struct fuse_dispatcher fdi;
2014 int err;
2015
2016 if (fuse_isdeadfs(vp)) {
2017 return ENXIO;
2018 }
2019 if (!vnode_islnk(vp)) {
2020 return EINVAL;
2021 }
2022 fdisp_init(&fdi, 0);
2023 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
2024 if (err) {
2025 goto out;
2026 }
2027 if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) {
2028 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
2029 fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL,
2030 "Returned an embedded NUL from FUSE_READLINK.");
2031 err = EIO;
2032 goto out;
2033 }
2034 if (((char *)fdi.answ)[0] == '/' &&
2035 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
2036 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
2037
2038 err = uiomove(mpth, strlen(mpth), uio);
2039 }
2040 if (!err) {
2041 err = uiomove(fdi.answ, fdi.iosize, uio);
2042 }
2043 out:
2044 fdisp_destroy(&fdi);
2045 return err;
2046 }
2047
2048 /*
2049 struct vnop_reclaim_args {
2050 struct vnode *a_vp;
2051 };
2052 */
2053 static int
fuse_vnop_reclaim(struct vop_reclaim_args * ap)2054 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2055 {
2056 struct vnode *vp = ap->a_vp;
2057 struct thread *td = curthread;
2058 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2059 struct fuse_filehandle *fufh, *fufh_tmp;
2060
2061 if (!fvdat) {
2062 panic("FUSE: no vnode data during recycling");
2063 }
2064 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2065 printf("FUSE: vnode being reclaimed with open fufh "
2066 "(type=%#x)", fufh->fufh_type);
2067 fuse_filehandle_close(vp, fufh, td, NULL);
2068 }
2069
2070 if (VTOI(vp) == 1) {
2071 /*
2072 * Don't send FUSE_FORGET for the root inode, because
2073 * we never send FUSE_LOOKUP for it (see
2074 * fuse_vfsop_root) and we don't want the server to see
2075 * mismatched lookup counts.
2076 */
2077 struct fuse_data *data;
2078 struct vnode *vroot;
2079
2080 data = fuse_get_mpdata(vnode_mount(vp));
2081 FUSE_LOCK();
2082 vroot = data->vroot;
2083 data->vroot = NULL;
2084 FUSE_UNLOCK();
2085 if (vroot)
2086 vrele(vroot);
2087 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2088 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2089 fvdat->nlookup);
2090 }
2091 cache_purge(vp);
2092 vfs_hash_remove(vp);
2093 fuse_vnode_destroy(vp);
2094
2095 return 0;
2096 }
2097
2098 /*
2099 struct vnop_remove_args {
2100 struct vnode *a_dvp;
2101 struct vnode *a_vp;
2102 struct componentname *a_cnp;
2103 };
2104 */
2105 static int
fuse_vnop_remove(struct vop_remove_args * ap)2106 fuse_vnop_remove(struct vop_remove_args *ap)
2107 {
2108 struct vnode *dvp = ap->a_dvp;
2109 struct vnode *vp = ap->a_vp;
2110 struct componentname *cnp = ap->a_cnp;
2111
2112 int err;
2113
2114 if (fuse_isdeadfs(vp)) {
2115 return ENXIO;
2116 }
2117 if (vnode_isdir(vp)) {
2118 return EPERM;
2119 }
2120
2121 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2122
2123 return err;
2124 }
2125
2126 /*
2127 struct vnop_rename_args {
2128 struct vnode *a_fdvp;
2129 struct vnode *a_fvp;
2130 struct componentname *a_fcnp;
2131 struct vnode *a_tdvp;
2132 struct vnode *a_tvp;
2133 struct componentname *a_tcnp;
2134 };
2135 */
2136 static int
fuse_vnop_rename(struct vop_rename_args * ap)2137 fuse_vnop_rename(struct vop_rename_args *ap)
2138 {
2139 struct vnode *fdvp = ap->a_fdvp;
2140 struct vnode *fvp = ap->a_fvp;
2141 struct componentname *fcnp = ap->a_fcnp;
2142 struct vnode *tdvp = ap->a_tdvp;
2143 struct vnode *tvp = ap->a_tvp;
2144 struct componentname *tcnp = ap->a_tcnp;
2145 struct fuse_data *data;
2146 bool newparent = fdvp != tdvp;
2147 bool isdir = fvp->v_type == VDIR;
2148 int err = 0;
2149
2150 if (fuse_isdeadfs(fdvp)) {
2151 return ENXIO;
2152 }
2153 if (fvp->v_mount != tdvp->v_mount ||
2154 (tvp && fvp->v_mount != tvp->v_mount)) {
2155 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2156 err = EXDEV;
2157 goto out;
2158 }
2159 cache_purge(fvp);
2160
2161 /*
2162 * FUSE library is expected to check if target directory is not
2163 * under the source directory in the file system tree.
2164 * Linux performs this check at VFS level.
2165 */
2166 /*
2167 * If source is a directory, and it will get a new parent, user must
2168 * have write permission to it, so ".." can be modified.
2169 */
2170 data = fuse_get_mpdata(vnode_mount(tdvp));
2171 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2172 err = fuse_internal_access(fvp, VWRITE,
2173 curthread, tcnp->cn_cred);
2174 if (err)
2175 goto out;
2176 }
2177 sx_xlock(&data->rename_lock);
2178 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2179 if (err == 0) {
2180 if (tdvp != fdvp)
2181 fuse_vnode_setparent(fvp, tdvp);
2182 if (tvp != NULL)
2183 fuse_vnode_setparent(tvp, NULL);
2184 }
2185 sx_unlock(&data->rename_lock);
2186
2187 if (tvp != NULL && tvp != fvp) {
2188 cache_purge(tvp);
2189 }
2190 if (vnode_isdir(fvp)) {
2191 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2192 cache_purge(tdvp);
2193 }
2194 cache_purge(fdvp);
2195 }
2196 out:
2197 if (tdvp == tvp) {
2198 vrele(tdvp);
2199 } else {
2200 vput(tdvp);
2201 }
2202 if (tvp != NULL) {
2203 vput(tvp);
2204 }
2205 vrele(fdvp);
2206 vrele(fvp);
2207
2208 return err;
2209 }
2210
2211 /*
2212 struct vnop_rmdir_args {
2213 struct vnode *a_dvp;
2214 struct vnode *a_vp;
2215 struct componentname *a_cnp;
2216 } *ap;
2217 */
2218 static int
fuse_vnop_rmdir(struct vop_rmdir_args * ap)2219 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2220 {
2221 struct vnode *dvp = ap->a_dvp;
2222 struct vnode *vp = ap->a_vp;
2223
2224 int err;
2225
2226 if (fuse_isdeadfs(vp)) {
2227 return ENXIO;
2228 }
2229 if (VTOFUD(vp) == VTOFUD(dvp)) {
2230 return EINVAL;
2231 }
2232 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2233
2234 return err;
2235 }
2236
2237 /*
2238 struct vnop_setattr_args {
2239 struct vnode *a_vp;
2240 struct vattr *a_vap;
2241 struct ucred *a_cred;
2242 struct thread *a_td;
2243 };
2244 */
2245 static int
fuse_vnop_setattr(struct vop_setattr_args * ap)2246 fuse_vnop_setattr(struct vop_setattr_args *ap)
2247 {
2248 struct vnode *vp = ap->a_vp;
2249 struct vattr *vap = ap->a_vap;
2250 struct ucred *cred = ap->a_cred;
2251 struct thread *td = curthread;
2252 struct mount *mp;
2253 struct fuse_data *data;
2254 struct vattr old_va;
2255 int dataflags;
2256 int err = 0, err2;
2257 accmode_t accmode = 0;
2258 bool checkperm;
2259 bool drop_suid = false;
2260
2261 mp = vnode_mount(vp);
2262 data = fuse_get_mpdata(mp);
2263 dataflags = data->dataflags;
2264 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2265
2266 if (fuse_isdeadfs(vp)) {
2267 return ENXIO;
2268 }
2269
2270 if (vap->va_uid != (uid_t)VNOVAL) {
2271 if (checkperm) {
2272 /* Only root may change a file's owner */
2273 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2274 if (err) {
2275 /* As a special case, allow the null chown */
2276 err2 = fuse_internal_getattr(vp, &old_va, cred,
2277 td);
2278 if (err2)
2279 return (err2);
2280 if (vap->va_uid != old_va.va_uid)
2281 return err;
2282 drop_suid = true;
2283 }
2284 }
2285 accmode |= VADMIN;
2286 }
2287 if (vap->va_gid != (gid_t)VNOVAL) {
2288 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2289 drop_suid = true;
2290 if (checkperm && !groupmember(vap->va_gid, cred)) {
2291 /*
2292 * Non-root users may only chgrp to one of their own
2293 * groups
2294 */
2295 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2296 if (err) {
2297 /* As a special case, allow the null chgrp */
2298 err2 = fuse_internal_getattr(vp, &old_va, cred,
2299 td);
2300 if (err2)
2301 return (err2);
2302 if (vap->va_gid != old_va.va_gid)
2303 return err;
2304 }
2305 }
2306 accmode |= VADMIN;
2307 }
2308 if (vap->va_size != VNOVAL) {
2309 switch (vp->v_type) {
2310 case VDIR:
2311 return (EISDIR);
2312 case VLNK:
2313 case VREG:
2314 if (vfs_isrdonly(mp))
2315 return (EROFS);
2316 err = vn_rlimit_trunc(vap->va_size, td);
2317 if (err)
2318 return (err);
2319 break;
2320 default:
2321 /*
2322 * According to POSIX, the result is unspecified
2323 * for file types other than regular files,
2324 * directories and shared memory objects. We
2325 * don't support shared memory objects in the file
2326 * system, and have dubious support for truncating
2327 * symlinks. Just ignore the request in other cases.
2328 */
2329 return (0);
2330 }
2331 /* Don't set accmode. Permission to trunc is checked upstack */
2332 }
2333 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2334 if (vap->va_vaflags & VA_UTIMES_NULL)
2335 accmode |= VWRITE;
2336 else
2337 accmode |= VADMIN;
2338 }
2339 if (drop_suid) {
2340 if (vap->va_mode != (mode_t)VNOVAL)
2341 vap->va_mode &= ~(S_ISUID | S_ISGID);
2342 else {
2343 err = fuse_internal_getattr(vp, &old_va, cred, td);
2344 if (err)
2345 return (err);
2346 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2347 }
2348 }
2349 if (vap->va_mode != (mode_t)VNOVAL) {
2350 /* Only root may set the sticky bit on non-directories */
2351 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2352 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2353 return EFTYPE;
2354 if (checkperm && (vap->va_mode & S_ISGID)) {
2355 err = fuse_internal_getattr(vp, &old_va, cred, td);
2356 if (err)
2357 return (err);
2358 if (!groupmember(old_va.va_gid, cred)) {
2359 err = priv_check_cred(cred, PRIV_VFS_SETGID);
2360 if (err)
2361 return (err);
2362 }
2363 }
2364 accmode |= VADMIN;
2365 }
2366
2367 if (vfs_isrdonly(mp))
2368 return EROFS;
2369
2370 if (checkperm) {
2371 err = fuse_internal_access(vp, accmode, td, cred);
2372 } else {
2373 err = 0;
2374 }
2375 if (err)
2376 return err;
2377 else
2378 return fuse_internal_setattr(vp, vap, td, cred);
2379 }
2380
2381 /*
2382 struct vnop_strategy_args {
2383 struct vnode *a_vp;
2384 struct buf *a_bp;
2385 };
2386 */
2387 static int
fuse_vnop_strategy(struct vop_strategy_args * ap)2388 fuse_vnop_strategy(struct vop_strategy_args *ap)
2389 {
2390 struct vnode *vp = ap->a_vp;
2391 struct buf *bp = ap->a_bp;
2392
2393 if (!vp || fuse_isdeadfs(vp)) {
2394 bp->b_ioflags |= BIO_ERROR;
2395 bp->b_error = ENXIO;
2396 bufdone(bp);
2397 return 0;
2398 }
2399
2400 /*
2401 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2402 * fuse_io_strategy sets bp's error fields
2403 */
2404 (void)fuse_io_strategy(vp, bp);
2405
2406 return 0;
2407 }
2408
2409 /*
2410 struct vnop_symlink_args {
2411 struct vnode *a_dvp;
2412 struct vnode **a_vpp;
2413 struct componentname *a_cnp;
2414 struct vattr *a_vap;
2415 char *a_target;
2416 };
2417 */
2418 static int
fuse_vnop_symlink(struct vop_symlink_args * ap)2419 fuse_vnop_symlink(struct vop_symlink_args *ap)
2420 {
2421 struct vnode *dvp = ap->a_dvp;
2422 struct vnode **vpp = ap->a_vpp;
2423 struct componentname *cnp = ap->a_cnp;
2424 const char *target = ap->a_target;
2425
2426 struct fuse_dispatcher fdi;
2427
2428 int err;
2429 size_t len;
2430
2431 if (fuse_isdeadfs(dvp)) {
2432 return ENXIO;
2433 }
2434 /*
2435 * Unlike the other creator type calls, here we have to create a message
2436 * where the name of the new entry comes first, and the data describing
2437 * the entry comes second.
2438 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2439 * but put together the message manually and just call the core part.
2440 */
2441
2442 len = strlen(target) + 1;
2443 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2444 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2445
2446 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2447 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2448 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2449
2450 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2451 fdisp_destroy(&fdi);
2452 return err;
2453 }
2454
2455 /*
2456 struct vnop_write_args {
2457 struct vnode *a_vp;
2458 struct uio *a_uio;
2459 int a_ioflag;
2460 struct ucred *a_cred;
2461 };
2462 */
2463 static int
fuse_vnop_write(struct vop_write_args * ap)2464 fuse_vnop_write(struct vop_write_args *ap)
2465 {
2466 struct vnode *vp = ap->a_vp;
2467 struct uio *uio = ap->a_uio;
2468 int ioflag = ap->a_ioflag;
2469 struct ucred *cred = ap->a_cred;
2470 pid_t pid = curthread->td_proc->p_pid;
2471 struct fuse_filehandle *fufh;
2472 int err;
2473 bool closefufh = false, directio;
2474
2475 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2476
2477 if (fuse_isdeadfs(vp)) {
2478 return ENXIO;
2479 }
2480
2481 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2482 ioflag |= IO_DIRECT;
2483 }
2484
2485 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2486 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2487 /*
2488 * nfsd will do I/O without first doing VOP_OPEN. We
2489 * must implicitly open the file here
2490 */
2491 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2492 closefufh = true;
2493 }
2494 if (err) {
2495 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2496 return err;
2497 }
2498
2499 /*
2500 * Ideally, when the daemon asks for direct io at open time, the
2501 * standard file flag should be set according to this, so that would
2502 * just change the default mode, which later on could be changed via
2503 * fcntl(2).
2504 * But this doesn't work, the O_DIRECT flag gets cleared at some point
2505 * (don't know where). So to make any use of the Fuse direct_io option,
2506 * we hardwire it into the file's private data (similarly to Linux,
2507 * btw.).
2508 */
2509 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2510
2511 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2512 if (directio) {
2513 off_t start, end, filesize;
2514 bool pages = (ioflag & IO_VMIO) != 0;
2515
2516 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2517
2518 err = fuse_vnode_size(vp, &filesize, cred, curthread);
2519 if (err)
2520 goto out;
2521
2522 start = uio->uio_offset;
2523 end = start + uio->uio_resid;
2524 if (!pages) {
2525 err = fuse_inval_buf_range(vp, filesize, start,
2526 end);
2527 if (err)
2528 goto out;
2529 }
2530 err = fuse_write_directbackend(vp, uio, cred, fufh,
2531 filesize, ioflag, pages);
2532 } else {
2533 SDT_PROBE2(fusefs, , vnops, trace, 1,
2534 "buffered write of vnode");
2535 if (!fsess_opt_writeback(vnode_mount(vp)))
2536 ioflag |= IO_SYNC;
2537 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2538 }
2539 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2540
2541 out:
2542 if (closefufh)
2543 fuse_filehandle_close(vp, fufh, curthread, cred);
2544
2545 return (err);
2546 }
2547
2548 static daddr_t
fuse_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)2549 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2550 {
2551 const int biosize = fuse_iosize(vp);
2552
2553 return (off / biosize);
2554 }
2555
2556 static int
fuse_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * blksz)2557 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2558 {
2559 off_t filesize;
2560 int err;
2561 const int biosize = fuse_iosize(vp);
2562
2563 err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2564 if (err) {
2565 /* This will turn into a SIGBUS */
2566 return (EIO);
2567 } else if ((off_t)lbn * biosize >= filesize) {
2568 *blksz = 0;
2569 } else if ((off_t)(lbn + 1) * biosize > filesize) {
2570 *blksz = filesize - (off_t)lbn *biosize;
2571 } else {
2572 *blksz = biosize;
2573 }
2574 return (0);
2575 }
2576
2577 /*
2578 struct vnop_getpages_args {
2579 struct vnode *a_vp;
2580 vm_page_t *a_m;
2581 int a_count;
2582 int a_reqpage;
2583 };
2584 */
2585 static int
fuse_vnop_getpages(struct vop_getpages_args * ap)2586 fuse_vnop_getpages(struct vop_getpages_args *ap)
2587 {
2588 struct vnode *vp = ap->a_vp;
2589
2590 if (!fsess_opt_mmap(vnode_mount(vp))) {
2591 SDT_PROBE2(fusefs, , vnops, trace, 1,
2592 "called on non-cacheable vnode??\n");
2593 return (VM_PAGER_ERROR);
2594 }
2595
2596 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2597 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2598 }
2599
2600 static const char extattr_namespace_separator = '.';
2601
2602 /*
2603 struct vop_getextattr_args {
2604 struct vop_generic_args a_gen;
2605 struct vnode *a_vp;
2606 int a_attrnamespace;
2607 const char *a_name;
2608 struct uio *a_uio;
2609 size_t *a_size;
2610 struct ucred *a_cred;
2611 struct thread *a_td;
2612 };
2613 */
2614 static int
fuse_vnop_getextattr(struct vop_getextattr_args * ap)2615 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2616 {
2617 struct vnode *vp = ap->a_vp;
2618 struct uio *uio = ap->a_uio;
2619 struct fuse_dispatcher fdi;
2620 struct fuse_getxattr_in *get_xattr_in;
2621 struct fuse_getxattr_out *get_xattr_out;
2622 struct mount *mp = vnode_mount(vp);
2623 struct thread *td = ap->a_td;
2624 struct ucred *cred = ap->a_cred;
2625 char *prefix;
2626 char *attr_str;
2627 size_t len;
2628 int err;
2629
2630 if (fuse_isdeadfs(vp))
2631 return (ENXIO);
2632
2633 if (fsess_not_impl(mp, FUSE_GETXATTR))
2634 return EOPNOTSUPP;
2635
2636 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2637 if (err)
2638 return err;
2639
2640 /* Default to looking for user attributes. */
2641 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2642 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2643 else
2644 prefix = EXTATTR_NAMESPACE_USER_STRING;
2645
2646 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2647 strlen(ap->a_name) + 1;
2648
2649 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2650 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2651
2652 get_xattr_in = fdi.indata;
2653 /*
2654 * Check to see whether we're querying the available size or
2655 * issuing the actual request. If we pass in 0, we get back struct
2656 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2657 * that much data, without the struct fuse_getxattr_out header.
2658 */
2659 if (uio == NULL)
2660 get_xattr_in->size = 0;
2661 else
2662 get_xattr_in->size = uio->uio_resid;
2663
2664 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2665 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2666 ap->a_name);
2667
2668 err = fdisp_wait_answ(&fdi);
2669 if (err != 0) {
2670 if (err == ENOSYS) {
2671 fsess_set_notimpl(mp, FUSE_GETXATTR);
2672 err = EOPNOTSUPP;
2673 }
2674 goto out;
2675 }
2676
2677 get_xattr_out = fdi.answ;
2678
2679 if (ap->a_size != NULL)
2680 *ap->a_size = get_xattr_out->size;
2681
2682 if (uio != NULL)
2683 err = uiomove(fdi.answ, fdi.iosize, uio);
2684
2685 out:
2686 fdisp_destroy(&fdi);
2687 return (err);
2688 }
2689
2690 /*
2691 struct vop_setextattr_args {
2692 struct vop_generic_args a_gen;
2693 struct vnode *a_vp;
2694 int a_attrnamespace;
2695 const char *a_name;
2696 struct uio *a_uio;
2697 struct ucred *a_cred;
2698 struct thread *a_td;
2699 };
2700 */
2701 static int
fuse_vnop_setextattr(struct vop_setextattr_args * ap)2702 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2703 {
2704 struct vnode *vp = ap->a_vp;
2705 struct uio *uio = ap->a_uio;
2706 struct fuse_dispatcher fdi;
2707 struct fuse_setxattr_in *set_xattr_in;
2708 struct mount *mp = vnode_mount(vp);
2709 struct thread *td = ap->a_td;
2710 struct ucred *cred = ap->a_cred;
2711 size_t struct_size = FUSE_COMPAT_SETXATTR_IN_SIZE;
2712 char *prefix;
2713 size_t len;
2714 char *attr_str;
2715 int err;
2716
2717 if (fuse_isdeadfs(vp))
2718 return (ENXIO);
2719
2720 if (fsess_not_impl(mp, FUSE_SETXATTR))
2721 return EOPNOTSUPP;
2722
2723 if (vfs_isrdonly(mp))
2724 return EROFS;
2725
2726 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2727 if (ap->a_uio == NULL) {
2728 /*
2729 * If we got here as fallback from VOP_DELETEEXTATTR, then
2730 * return EOPNOTSUPP.
2731 */
2732 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2733 return (EOPNOTSUPP);
2734 else
2735 return (EINVAL);
2736 }
2737
2738 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2739 VWRITE);
2740 if (err)
2741 return err;
2742
2743 /* Default to looking for user attributes. */
2744 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2745 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2746 else
2747 prefix = EXTATTR_NAMESPACE_USER_STRING;
2748
2749 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2750 strlen(ap->a_name) + 1;
2751
2752 /* older FUSE servers use a smaller fuse_setxattr_in struct*/
2753 if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33))
2754 struct_size = sizeof(*set_xattr_in);
2755
2756 fdisp_init(&fdi, len + struct_size + uio->uio_resid);
2757 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2758
2759 set_xattr_in = fdi.indata;
2760 set_xattr_in->size = uio->uio_resid;
2761
2762 if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33)) {
2763 set_xattr_in->setxattr_flags = 0;
2764 set_xattr_in->padding = 0;
2765 }
2766
2767 attr_str = (char *)fdi.indata + struct_size;
2768 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2769 ap->a_name);
2770
2771 err = uiomove((char *)fdi.indata + struct_size + len,
2772 uio->uio_resid, uio);
2773 if (err != 0) {
2774 goto out;
2775 }
2776
2777 err = fdisp_wait_answ(&fdi);
2778
2779 if (err == ENOSYS) {
2780 fsess_set_notimpl(mp, FUSE_SETXATTR);
2781 err = EOPNOTSUPP;
2782 }
2783 if (err == ERESTART) {
2784 /* Can't restart after calling uiomove */
2785 err = EINTR;
2786 }
2787
2788 out:
2789 fdisp_destroy(&fdi);
2790 return (err);
2791 }
2792
2793 /*
2794 * The Linux / FUSE extended attribute list is simply a collection of
2795 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2796 * byte length followed by a non-NUL terminated string. So, this allows
2797 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2798 * Linux attribute names are reported with the namespace as a prefix (e.g.
2799 * "user.attribute_name"), but in FreeBSD they are reported without the
2800 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2801 *
2802 * user.attr_name1\0user.attr_name2\0
2803 *
2804 * to:
2805 *
2806 * <num>attr_name1<num>attr_name2
2807 *
2808 * Where "<num>" is a single byte number of characters in the attribute name.
2809 *
2810 * Args:
2811 * prefix - exattr namespace prefix string
2812 * list, list_len - input list with namespace prefixes
2813 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2814 */
2815 static int
fuse_xattrlist_convert(char * prefix,const char * list,int list_len,char * bsd_list,int * bsd_list_len)2816 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2817 char *bsd_list, int *bsd_list_len)
2818 {
2819 int len, pos, dist_to_next, prefix_len;
2820
2821 pos = 0;
2822 *bsd_list_len = 0;
2823 prefix_len = strlen(prefix);
2824
2825 while (pos < list_len && list[pos] != '\0') {
2826 dist_to_next = strlen(&list[pos]) + 1;
2827 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2828 list[pos + prefix_len] == extattr_namespace_separator) {
2829 len = dist_to_next -
2830 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2831 if (len >= EXTATTR_MAXNAMELEN)
2832 return (ENAMETOOLONG);
2833
2834 bsd_list[*bsd_list_len] = len;
2835 memcpy(&bsd_list[*bsd_list_len + 1],
2836 &list[pos + prefix_len +
2837 sizeof(extattr_namespace_separator)], len);
2838
2839 *bsd_list_len += len + 1;
2840 }
2841
2842 pos += dist_to_next;
2843 }
2844
2845 return (0);
2846 }
2847
2848 /*
2849 * List extended attributes
2850 *
2851 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2852 * has a number of differences compared to its FreeBSD equivalent,
2853 * extattr_list_file:
2854 *
2855 * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2856 * whereas listxattr(2) only returns attributes for a single namespace
2857 * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2858 * - If the provided buffer is not large enough to hold the result,
2859 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2860 * return as many results as will fit.
2861 */
2862 /*
2863 struct vop_listextattr_args {
2864 struct vop_generic_args a_gen;
2865 struct vnode *a_vp;
2866 int a_attrnamespace;
2867 struct uio *a_uio;
2868 size_t *a_size;
2869 struct ucred *a_cred;
2870 struct thread *a_td;
2871 };
2872 */
2873 static int
fuse_vnop_listextattr(struct vop_listextattr_args * ap)2874 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2875 {
2876 struct vnode *vp = ap->a_vp;
2877 struct uio *uio = ap->a_uio;
2878 struct fuse_dispatcher fdi;
2879 struct fuse_listxattr_in *list_xattr_in;
2880 struct fuse_listxattr_out *list_xattr_out;
2881 struct mount *mp = vnode_mount(vp);
2882 struct thread *td = ap->a_td;
2883 struct ucred *cred = ap->a_cred;
2884 char *prefix;
2885 char *bsd_list = NULL;
2886 char *linux_list;
2887 int bsd_list_len;
2888 int linux_list_len;
2889 int err;
2890
2891 if (fuse_isdeadfs(vp))
2892 return (ENXIO);
2893
2894 if (fsess_not_impl(mp, FUSE_LISTXATTR))
2895 return EOPNOTSUPP;
2896
2897 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2898 if (err)
2899 return err;
2900
2901 /*
2902 * Add space for a NUL and the period separator if enabled.
2903 * Default to looking for user attributes.
2904 */
2905 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2906 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2907 else
2908 prefix = EXTATTR_NAMESPACE_USER_STRING;
2909
2910 fdisp_init(&fdi, sizeof(*list_xattr_in));
2911 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2912
2913 /*
2914 * Retrieve Linux / FUSE compatible list size.
2915 */
2916 list_xattr_in = fdi.indata;
2917 list_xattr_in->size = 0;
2918
2919 err = fdisp_wait_answ(&fdi);
2920 if (err != 0) {
2921 if (err == ENOSYS) {
2922 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2923 err = EOPNOTSUPP;
2924 }
2925 goto out;
2926 }
2927
2928 list_xattr_out = fdi.answ;
2929 linux_list_len = list_xattr_out->size;
2930 if (linux_list_len == 0) {
2931 if (ap->a_size != NULL)
2932 *ap->a_size = linux_list_len;
2933 goto out;
2934 }
2935
2936 /*
2937 * Retrieve Linux / FUSE compatible list values.
2938 */
2939 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2940 list_xattr_in = fdi.indata;
2941 list_xattr_in->size = linux_list_len;
2942
2943 err = fdisp_wait_answ(&fdi);
2944 if (err == ERANGE) {
2945 /*
2946 * Race detected. The attribute list must've grown since the
2947 * first FUSE_LISTXATTR call. Start over. Go all the way back
2948 * to userland so we can process signals, if necessary, before
2949 * restarting.
2950 */
2951 err = ERESTART;
2952 goto out;
2953 } else if (err != 0)
2954 goto out;
2955
2956 linux_list = fdi.answ;
2957 /* FUSE doesn't allow the server to return more data than requested */
2958 if (fdi.iosize > linux_list_len) {
2959 struct fuse_data *data = fuse_get_mpdata(mp);
2960
2961 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2962 "server returned "
2963 "more extended attribute data than requested; "
2964 "should've returned ERANGE instead.");
2965 } else {
2966 /* But returning less data is fine */
2967 linux_list_len = fdi.iosize;
2968 }
2969
2970 /*
2971 * Retrieve the BSD compatible list values.
2972 * The Linux / FUSE attribute list format isn't the same
2973 * as FreeBSD's format. So we need to transform it into
2974 * FreeBSD's format before giving it to the user.
2975 */
2976 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2977 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2978 bsd_list, &bsd_list_len);
2979 if (err != 0)
2980 goto out;
2981
2982 if (ap->a_size != NULL)
2983 *ap->a_size = bsd_list_len;
2984
2985 if (uio != NULL)
2986 err = uiomove(bsd_list, bsd_list_len, uio);
2987
2988 out:
2989 free(bsd_list, M_TEMP);
2990 fdisp_destroy(&fdi);
2991 return (err);
2992 }
2993
2994 /*
2995 struct vop_deallocate_args {
2996 struct vop_generic_args a_gen;
2997 struct vnode *a_vp;
2998 off_t *a_offset;
2999 off_t *a_len;
3000 int a_flags;
3001 int a_ioflag;
3002 struct ucred *a_cred;
3003 };
3004 */
3005 static int
fuse_vnop_deallocate(struct vop_deallocate_args * ap)3006 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
3007 {
3008 struct vnode *vp = ap->a_vp;
3009 struct mount *mp = vnode_mount(vp);
3010 struct fuse_filehandle *fufh;
3011 struct fuse_dispatcher fdi;
3012 struct fuse_fallocate_in *ffi;
3013 struct ucred *cred = ap->a_cred;
3014 pid_t pid = curthread->td_proc->p_pid;
3015 off_t *len = ap->a_len;
3016 off_t *offset = ap->a_offset;
3017 int ioflag = ap->a_ioflag;
3018 off_t filesize;
3019 int err;
3020 bool closefufh = false;
3021
3022 if (fuse_isdeadfs(vp))
3023 return (ENXIO);
3024
3025 if (vfs_isrdonly(mp))
3026 return (EROFS);
3027
3028 if (fsess_not_impl(mp, FUSE_FALLOCATE))
3029 goto fallback;
3030
3031 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
3032 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
3033 /*
3034 * nfsd will do I/O without first doing VOP_OPEN. We
3035 * must implicitly open the file here
3036 */
3037 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
3038 closefufh = true;
3039 }
3040 if (err)
3041 return (err);
3042
3043 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3044
3045 err = fuse_vnode_size(vp, &filesize, cred, curthread);
3046 if (err)
3047 goto out;
3048 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3049
3050 fdisp_init(&fdi, sizeof(*ffi));
3051 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3052 ffi = fdi.indata;
3053 ffi->fh = fufh->fh_id;
3054 ffi->offset = *offset;
3055 ffi->length = *len;
3056 /*
3057 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3058 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3059 */
3060 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3061 err = fdisp_wait_answ(&fdi);
3062
3063 if (err == ENOSYS) {
3064 fdisp_destroy(&fdi);
3065 fsess_set_notimpl(mp, FUSE_FALLOCATE);
3066 goto fallback;
3067 } else if (err == EOPNOTSUPP) {
3068 /*
3069 * The file system server does not support FUSE_FALLOCATE with
3070 * the supplied mode for this particular file.
3071 */
3072 fdisp_destroy(&fdi);
3073 goto fallback;
3074 } else if (!err) {
3075 /*
3076 * Clip the returned offset to EoF. Do it here rather than
3077 * before FUSE_FALLOCATE just in case the kernel's cached file
3078 * size is out of date. Unfortunately, FUSE does not return
3079 * any information about filesize from that operation.
3080 */
3081 *offset = MIN(*offset + *len, filesize);
3082 *len = 0;
3083 fuse_vnode_undirty_cached_timestamps(vp, false);
3084 fuse_internal_clear_suid_on_write(vp, cred, curthread);
3085
3086 if (ioflag & IO_SYNC)
3087 err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3088 false);
3089 }
3090
3091 fdisp_destroy(&fdi);
3092 out:
3093 if (closefufh)
3094 fuse_filehandle_close(vp, fufh, curthread, cred);
3095
3096 return (err);
3097
3098 fallback:
3099 if (closefufh)
3100 fuse_filehandle_close(vp, fufh, curthread, cred);
3101
3102 return (vop_stddeallocate(ap));
3103 }
3104
3105 /*
3106 struct vop_deleteextattr_args {
3107 struct vop_generic_args a_gen;
3108 struct vnode *a_vp;
3109 int a_attrnamespace;
3110 const char *a_name;
3111 struct ucred *a_cred;
3112 struct thread *a_td;
3113 };
3114 */
3115 static int
fuse_vnop_deleteextattr(struct vop_deleteextattr_args * ap)3116 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3117 {
3118 struct vnode *vp = ap->a_vp;
3119 struct fuse_dispatcher fdi;
3120 struct mount *mp = vnode_mount(vp);
3121 struct thread *td = ap->a_td;
3122 struct ucred *cred = ap->a_cred;
3123 char *prefix;
3124 size_t len;
3125 char *attr_str;
3126 int err;
3127
3128 if (fuse_isdeadfs(vp))
3129 return (ENXIO);
3130
3131 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3132 return EOPNOTSUPP;
3133
3134 if (vfs_isrdonly(mp))
3135 return EROFS;
3136
3137 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3138 VWRITE);
3139 if (err)
3140 return err;
3141
3142 /* Default to looking for user attributes. */
3143 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3144 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3145 else
3146 prefix = EXTATTR_NAMESPACE_USER_STRING;
3147
3148 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3149 strlen(ap->a_name) + 1;
3150
3151 fdisp_init(&fdi, len);
3152 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3153
3154 attr_str = fdi.indata;
3155 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3156 ap->a_name);
3157
3158 err = fdisp_wait_answ(&fdi);
3159 if (err == ENOSYS) {
3160 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3161 err = EOPNOTSUPP;
3162 }
3163
3164 fdisp_destroy(&fdi);
3165 return (err);
3166 }
3167
3168 /*
3169 struct vnop_print_args {
3170 struct vnode *a_vp;
3171 };
3172 */
3173 static int
fuse_vnop_print(struct vop_print_args * ap)3174 fuse_vnop_print(struct vop_print_args *ap)
3175 {
3176 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3177
3178 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3179 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3180 (uintmax_t)fvdat->nlookup,
3181 fvdat->flag);
3182
3183 return 0;
3184 }
3185
3186 /*
3187 * Get an NFS filehandle for a FUSE file.
3188 *
3189 * This will only work for FUSE file systems that guarantee the uniqueness of
3190 * nodeid:generation, which most don't.
3191 */
3192 /*
3193 vop_vptofh {
3194 IN struct vnode *a_vp;
3195 IN struct fid *a_fhp;
3196 };
3197 */
3198 static int
fuse_vnop_vptofh(struct vop_vptofh_args * ap)3199 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3200 {
3201 struct vnode *vp = ap->a_vp;
3202 struct fuse_vnode_data *fvdat = VTOFUD(vp);
3203 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3204 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3205 "FUSE fid type is too big");
3206 struct mount *mp = vnode_mount(vp);
3207 struct fuse_data *data = fuse_get_mpdata(mp);
3208 struct vattr va;
3209 int err;
3210
3211 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3212 /* NFS requires lookups for "." and ".." */
3213 SDT_PROBE2(fusefs, , vnops, trace, 1,
3214 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3215 return EOPNOTSUPP;
3216 }
3217 if ((mp->mnt_flag & MNT_EXPORTED) &&
3218 fsess_is_impl(mp, FUSE_OPENDIR))
3219 {
3220 /*
3221 * NFS is stateless, so nfsd must reopen a directory on every
3222 * call to VOP_READDIR, passing in the d_off field from the
3223 * final dirent of the previous invocation. But if the server
3224 * implements FUSE_OPENDIR, the FUSE protocol does not
3225 * guarantee that d_off will be valid after a directory is
3226 * closed and reopened. So prohibit exporting FUSE file
3227 * systems that implement FUSE_OPENDIR.
3228 *
3229 * But userspace NFS servers don't have this problem.
3230 */
3231 SDT_PROBE2(fusefs, , vnops, trace, 1,
3232 "VOP_VPTOFH with FUSE_OPENDIR");
3233 return EOPNOTSUPP;
3234 }
3235
3236 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3237 if (err)
3238 return err;
3239
3240 /*ip = VTOI(ap->a_vp);*/
3241 /*ufhp = (struct ufid *)ap->a_fhp;*/
3242 fhp->len = sizeof(struct fuse_fid);
3243 fhp->nid = fvdat->nid;
3244 if (fvdat->generation <= UINT32_MAX)
3245 fhp->gen = fvdat->generation;
3246 else
3247 return EOVERFLOW;
3248 return (0);
3249 }
3250