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