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/cdefs.h>
64 __FBSDID("$FreeBSD$");
65
66 #include <sys/param.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
71 #include <sys/conf.h>
72 #include <sys/filio.h>
73 #include <sys/uio.h>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
76 #include <sys/limits.h>
77 #include <sys/lock.h>
78 #include <sys/rwlock.h>
79 #include <sys/sx.h>
80 #include <sys/proc.h>
81 #include <sys/mount.h>
82 #include <sys/vnode.h>
83 #include <sys/namei.h>
84 #include <sys/extattr.h>
85 #include <sys/stat.h>
86 #include <sys/unistd.h>
87 #include <sys/filedesc.h>
88 #include <sys/file.h>
89 #include <sys/fcntl.h>
90 #include <sys/dirent.h>
91 #include <sys/bio.h>
92 #include <sys/buf.h>
93 #include <sys/sysctl.h>
94 #include <sys/vmmeter.h>
95
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98 #include <vm/pmap.h>
99 #include <vm/vm_map.h>
100 #include <vm/vm_page.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_pager.h>
104 #include <vm/vnode_pager.h>
105 #include <vm/vm_object.h>
106
107 #include "fuse.h"
108 #include "fuse_file.h"
109 #include "fuse_internal.h"
110 #include "fuse_ipc.h"
111 #include "fuse_node.h"
112 #include "fuse_io.h"
113
114 #include <sys/priv.h>
115
116 /* Maximum number of hardlinks to a single FUSE file */
117 #define FUSE_LINK_MAX UINT32_MAX
118
119 SDT_PROVIDER_DECLARE(fusefs);
120 /*
121 * Fuse trace probe:
122 * arg0: verbosity. Higher numbers give more verbose messages
123 * arg1: Textual message
124 */
125 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
126
127 /* vnode ops */
128 static vop_access_t fuse_vnop_access;
129 static vop_advlock_t fuse_vnop_advlock;
130 static vop_allocate_t fuse_vnop_allocate;
131 static vop_bmap_t fuse_vnop_bmap;
132 static vop_close_t fuse_fifo_close;
133 static vop_close_t fuse_vnop_close;
134 static vop_copy_file_range_t fuse_vnop_copy_file_range;
135 static vop_create_t fuse_vnop_create;
136 static vop_deallocate_t fuse_vnop_deallocate;
137 static vop_deleteextattr_t fuse_vnop_deleteextattr;
138 static vop_fdatasync_t fuse_vnop_fdatasync;
139 static vop_fsync_t fuse_vnop_fsync;
140 static vop_getattr_t fuse_vnop_getattr;
141 static vop_getextattr_t fuse_vnop_getextattr;
142 static vop_inactive_t fuse_vnop_inactive;
143 static vop_ioctl_t fuse_vnop_ioctl;
144 static vop_link_t fuse_vnop_link;
145 static vop_listextattr_t fuse_vnop_listextattr;
146 static vop_lookup_t fuse_vnop_lookup;
147 static vop_mkdir_t fuse_vnop_mkdir;
148 static vop_mknod_t fuse_vnop_mknod;
149 static vop_open_t fuse_vnop_open;
150 static vop_pathconf_t fuse_vnop_pathconf;
151 static vop_read_t fuse_vnop_read;
152 static vop_readdir_t fuse_vnop_readdir;
153 static vop_readlink_t fuse_vnop_readlink;
154 static vop_reclaim_t fuse_vnop_reclaim;
155 static vop_remove_t fuse_vnop_remove;
156 static vop_rename_t fuse_vnop_rename;
157 static vop_rmdir_t fuse_vnop_rmdir;
158 static vop_setattr_t fuse_vnop_setattr;
159 static vop_setextattr_t fuse_vnop_setextattr;
160 static vop_strategy_t fuse_vnop_strategy;
161 static vop_symlink_t fuse_vnop_symlink;
162 static vop_write_t fuse_vnop_write;
163 static vop_getpages_t fuse_vnop_getpages;
164 static vop_print_t fuse_vnop_print;
165 static vop_vptofh_t fuse_vnop_vptofh;
166
167 struct vop_vector fuse_fifoops = {
168 .vop_default = &fifo_specops,
169 .vop_access = fuse_vnop_access,
170 .vop_close = fuse_fifo_close,
171 .vop_fsync = fuse_vnop_fsync,
172 .vop_getattr = fuse_vnop_getattr,
173 .vop_inactive = fuse_vnop_inactive,
174 .vop_pathconf = fuse_vnop_pathconf,
175 .vop_print = fuse_vnop_print,
176 .vop_read = VOP_PANIC,
177 .vop_reclaim = fuse_vnop_reclaim,
178 .vop_setattr = fuse_vnop_setattr,
179 .vop_write = VOP_PANIC,
180 .vop_vptofh = fuse_vnop_vptofh,
181 };
182 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
183
184 struct vop_vector fuse_vnops = {
185 .vop_allocate = fuse_vnop_allocate,
186 .vop_default = &default_vnodeops,
187 .vop_access = fuse_vnop_access,
188 .vop_advlock = fuse_vnop_advlock,
189 .vop_bmap = fuse_vnop_bmap,
190 .vop_close = fuse_vnop_close,
191 .vop_copy_file_range = fuse_vnop_copy_file_range,
192 .vop_create = fuse_vnop_create,
193 .vop_deallocate = fuse_vnop_deallocate,
194 .vop_deleteextattr = fuse_vnop_deleteextattr,
195 .vop_fsync = fuse_vnop_fsync,
196 .vop_fdatasync = fuse_vnop_fdatasync,
197 .vop_getattr = fuse_vnop_getattr,
198 .vop_getextattr = fuse_vnop_getextattr,
199 .vop_inactive = fuse_vnop_inactive,
200 .vop_ioctl = fuse_vnop_ioctl,
201 .vop_link = fuse_vnop_link,
202 .vop_listextattr = fuse_vnop_listextattr,
203 .vop_lookup = fuse_vnop_lookup,
204 .vop_mkdir = fuse_vnop_mkdir,
205 .vop_mknod = fuse_vnop_mknod,
206 .vop_open = fuse_vnop_open,
207 .vop_pathconf = fuse_vnop_pathconf,
208 /*
209 * TODO: implement vop_poll after upgrading to protocol 7.21.
210 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
211 * 7.21, which adds the ability for the client to choose which poll
212 * events it wants, and for a client to deregister a file handle
213 */
214 .vop_read = fuse_vnop_read,
215 .vop_readdir = fuse_vnop_readdir,
216 .vop_readlink = fuse_vnop_readlink,
217 .vop_reclaim = fuse_vnop_reclaim,
218 .vop_remove = fuse_vnop_remove,
219 .vop_rename = fuse_vnop_rename,
220 .vop_rmdir = fuse_vnop_rmdir,
221 .vop_setattr = fuse_vnop_setattr,
222 .vop_setextattr = fuse_vnop_setextattr,
223 .vop_strategy = fuse_vnop_strategy,
224 .vop_symlink = fuse_vnop_symlink,
225 .vop_write = fuse_vnop_write,
226 .vop_getpages = fuse_vnop_getpages,
227 .vop_print = fuse_vnop_print,
228 .vop_vptofh = fuse_vnop_vptofh,
229 };
230 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
231
232 uma_zone_t fuse_pbuf_zone;
233
234 /* Check permission for extattr operations, much like extattr_check_cred */
235 static int
236 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
237 struct thread *td, accmode_t accmode)
238 {
239 struct mount *mp = vnode_mount(vp);
240 struct fuse_data *data = fuse_get_mpdata(mp);
241 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
242
243 /*
244 * Kernel-invoked always succeeds.
245 */
246 if (cred == NOCRED)
247 return (0);
248
249 /*
250 * Do not allow privileged processes in jail to directly manipulate
251 * system attributes.
252 */
253 switch (ns) {
254 case EXTATTR_NAMESPACE_SYSTEM:
255 if (default_permissions) {
256 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
257 }
258 return (0);
259 case EXTATTR_NAMESPACE_USER:
260 if (default_permissions) {
261 return (fuse_internal_access(vp, accmode, td, cred));
262 }
263 return (0);
264 default:
265 return (EPERM);
266 }
267 }
268
269 /* Get a filehandle for a directory */
270 static int
271 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
272 struct ucred *cred, pid_t pid)
273 {
274 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
275 return 0;
276 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
277 }
278
279 /* Send FUSE_FLUSH for this vnode */
280 static int
281 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
282 {
283 struct fuse_flush_in *ffi;
284 struct fuse_filehandle *fufh;
285 struct fuse_dispatcher fdi;
286 struct thread *td = curthread;
287 struct mount *mp = vnode_mount(vp);
288 int err;
289
290 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
291 return 0;
292
293 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
294 if (err)
295 return err;
296
297 fdisp_init(&fdi, sizeof(*ffi));
298 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
299 ffi = fdi.indata;
300 ffi->fh = fufh->fh_id;
301 /*
302 * If the file has a POSIX lock then we're supposed to set lock_owner.
303 * If not, then lock_owner is undefined. So we may as well always set
304 * it.
305 */
306 ffi->lock_owner = td->td_proc->p_pid;
307
308 err = fdisp_wait_answ(&fdi);
309 if (err == ENOSYS) {
310 fsess_set_notimpl(mp, FUSE_FLUSH);
311 err = 0;
312 }
313 fdisp_destroy(&fdi);
314 return err;
315 }
316
317 /* Close wrapper for fifos. */
318 static int
319 fuse_fifo_close(struct vop_close_args *ap)
320 {
321 return (fifo_specops.vop_close(ap));
322 }
323
324 /* Invalidate a range of cached data, whether dirty of not */
325 static int
326 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
327 {
328 struct buf *bp;
329 daddr_t left_lbn, end_lbn, right_lbn;
330 off_t new_filesize;
331 int iosize, left_on, right_on, right_blksize;
332
333 iosize = fuse_iosize(vp);
334 left_lbn = start / iosize;
335 end_lbn = howmany(end, iosize);
336 left_on = start & (iosize - 1);
337 if (left_on != 0) {
338 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
339 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
340 /*
341 * Flush the dirty buffer, because we don't have a
342 * byte-granular way to record which parts of the
343 * buffer are valid.
344 */
345 bwrite(bp);
346 if (bp->b_error)
347 return (bp->b_error);
348 } else {
349 brelse(bp);
350 }
351 }
352 right_on = end & (iosize - 1);
353 if (right_on != 0) {
354 right_lbn = end / iosize;
355 new_filesize = MAX(filesize, end);
356 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
357 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
358 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
359 /*
360 * Flush the dirty buffer, because we don't have a
361 * byte-granular way to record which parts of the
362 * buffer are valid.
363 */
364 bwrite(bp);
365 if (bp->b_error)
366 return (bp->b_error);
367 } else {
368 brelse(bp);
369 }
370 }
371
372 v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
373 return (0);
374 }
375
376
377 /* Send FUSE_LSEEK for this node */
378 static int
379 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
380 pid_t pid, off_t *offp, int whence)
381 {
382 struct fuse_dispatcher fdi;
383 struct fuse_filehandle *fufh;
384 struct fuse_lseek_in *flsi;
385 struct fuse_lseek_out *flso;
386 struct mount *mp = vnode_mount(vp);
387 int err;
388
389 ASSERT_VOP_LOCKED(vp, __func__);
390
391 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
392 if (err)
393 return (err);
394 fdisp_init(&fdi, sizeof(*flsi));
395 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
396 flsi = fdi.indata;
397 flsi->fh = fufh->fh_id;
398 flsi->offset = *offp;
399 flsi->whence = whence;
400 err = fdisp_wait_answ(&fdi);
401 if (err == ENOSYS) {
402 fsess_set_notimpl(mp, FUSE_LSEEK);
403 } else if (err == 0) {
404 fsess_set_impl(mp, FUSE_LSEEK);
405 flso = fdi.answ;
406 *offp = flso->offset;
407 }
408 fdisp_destroy(&fdi);
409
410 return (err);
411 }
412
413 /*
414 struct vnop_access_args {
415 struct vnode *a_vp;
416 #if VOP_ACCESS_TAKES_ACCMODE_T
417 accmode_t a_accmode;
418 #else
419 int a_mode;
420 #endif
421 struct ucred *a_cred;
422 struct thread *a_td;
423 };
424 */
425 static int
426 fuse_vnop_access(struct vop_access_args *ap)
427 {
428 struct vnode *vp = ap->a_vp;
429 int accmode = ap->a_accmode;
430 struct ucred *cred = ap->a_cred;
431
432 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
433
434 int err;
435
436 if (fuse_isdeadfs(vp)) {
437 if (vnode_isvroot(vp)) {
438 return 0;
439 }
440 return ENXIO;
441 }
442 if (!(data->dataflags & FSESS_INITED)) {
443 if (vnode_isvroot(vp)) {
444 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
445 (fuse_match_cred(data->daemoncred, cred) == 0)) {
446 return 0;
447 }
448 }
449 return EBADF;
450 }
451 if (vnode_islnk(vp)) {
452 return 0;
453 }
454
455 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
456 return err;
457 }
458
459 /*
460 * struct vop_advlock_args {
461 * struct vop_generic_args a_gen;
462 * struct vnode *a_vp;
463 * void *a_id;
464 * int a_op;
465 * struct flock *a_fl;
466 * int a_flags;
467 * }
468 */
469 static int
470 fuse_vnop_advlock(struct vop_advlock_args *ap)
471 {
472 struct vnode *vp = ap->a_vp;
473 struct flock *fl = ap->a_fl;
474 struct thread *td = curthread;
475 struct ucred *cred = td->td_ucred;
476 pid_t pid = td->td_proc->p_pid;
477 struct fuse_filehandle *fufh;
478 struct fuse_dispatcher fdi;
479 struct fuse_lk_in *fli;
480 struct fuse_lk_out *flo;
481 struct vattr vattr;
482 enum fuse_opcode op;
483 off_t size, start;
484 int dataflags, err;
485 int flags = ap->a_flags;
486
487 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
488
489 if (fuse_isdeadfs(vp)) {
490 return ENXIO;
491 }
492
493 switch(ap->a_op) {
494 case F_GETLK:
495 op = FUSE_GETLK;
496 break;
497 case F_SETLK:
498 if (flags & F_WAIT)
499 op = FUSE_SETLKW;
500 else
501 op = FUSE_SETLK;
502 break;
503 case F_UNLCK:
504 op = FUSE_SETLK;
505 break;
506 default:
507 return EINVAL;
508 }
509
510 if (!(dataflags & FSESS_POSIX_LOCKS))
511 return vop_stdadvlock(ap);
512 /* FUSE doesn't properly support flock until protocol 7.17 */
513 if (flags & F_FLOCK)
514 return vop_stdadvlock(ap);
515
516 vn_lock(vp, LK_SHARED | LK_RETRY);
517
518 switch (fl->l_whence) {
519 case SEEK_SET:
520 case SEEK_CUR:
521 /*
522 * Caller is responsible for adding any necessary offset
523 * when SEEK_CUR is used.
524 */
525 start = fl->l_start;
526 break;
527
528 case SEEK_END:
529 err = fuse_internal_getattr(vp, &vattr, cred, td);
530 if (err)
531 goto out;
532 size = vattr.va_size;
533 if (size > OFF_MAX ||
534 (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
535 err = EOVERFLOW;
536 goto out;
537 }
538 start = size + fl->l_start;
539 break;
540
541 default:
542 return (EINVAL);
543 }
544
545 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
546 if (err)
547 goto out;
548
549 fdisp_init(&fdi, sizeof(*fli));
550
551 fdisp_make_vp(&fdi, op, vp, td, cred);
552 fli = fdi.indata;
553 fli->fh = fufh->fh_id;
554 fli->owner = td->td_proc->p_pid;
555 fli->lk.start = start;
556 if (fl->l_len != 0)
557 fli->lk.end = start + fl->l_len - 1;
558 else
559 fli->lk.end = INT64_MAX;
560 fli->lk.type = fl->l_type;
561 fli->lk.pid = td->td_proc->p_pid;
562
563 err = fdisp_wait_answ(&fdi);
564 fdisp_destroy(&fdi);
565
566 if (err == 0 && op == FUSE_GETLK) {
567 flo = fdi.answ;
568 fl->l_type = flo->lk.type;
569 fl->l_whence = SEEK_SET;
570 if (flo->lk.type != F_UNLCK) {
571 fl->l_pid = flo->lk.pid;
572 fl->l_start = flo->lk.start;
573 if (flo->lk.end == INT64_MAX)
574 fl->l_len = 0;
575 else
576 fl->l_len = flo->lk.end - flo->lk.start + 1;
577 fl->l_start = flo->lk.start;
578 }
579 }
580
581 out:
582 VOP_UNLOCK(vp);
583 return err;
584 }
585
586 static int
587 fuse_vnop_allocate(struct vop_allocate_args *ap)
588 {
589 struct vnode *vp = ap->a_vp;
590 off_t *len = ap->a_len;
591 off_t *offset = ap->a_offset;
592 struct ucred *cred = ap->a_cred;
593 struct fuse_filehandle *fufh;
594 struct mount *mp = vnode_mount(vp);
595 struct fuse_dispatcher fdi;
596 struct fuse_fallocate_in *ffi;
597 struct uio io;
598 pid_t pid = curthread->td_proc->p_pid;
599 struct fuse_vnode_data *fvdat = VTOFUD(vp);
600 off_t filesize;
601 int err;
602
603 if (fuse_isdeadfs(vp))
604 return (ENXIO);
605
606 switch (vp->v_type) {
607 case VFIFO:
608 return (ESPIPE);
609 case VLNK:
610 case VREG:
611 if (vfs_isrdonly(mp))
612 return (EROFS);
613 break;
614 default:
615 return (ENODEV);
616 }
617
618 if (vfs_isrdonly(mp))
619 return (EROFS);
620
621 if (fsess_not_impl(mp, FUSE_FALLOCATE))
622 return (EINVAL);
623
624 io.uio_offset = *offset;
625 io.uio_resid = *len;
626 err = vn_rlimit_fsize(vp, &io, curthread);
627 if (err)
628 return (err);
629
630 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
631 if (err)
632 return (err);
633
634 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
635
636 err = fuse_vnode_size(vp, &filesize, cred, curthread);
637 if (err)
638 return (err);
639 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
640
641 fdisp_init(&fdi, sizeof(*ffi));
642 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
643 ffi = fdi.indata;
644 ffi->fh = fufh->fh_id;
645 ffi->offset = *offset;
646 ffi->length = *len;
647 ffi->mode = 0;
648 err = fdisp_wait_answ(&fdi);
649
650 if (err == ENOSYS) {
651 fsess_set_notimpl(mp, FUSE_FALLOCATE);
652 err = EINVAL;
653 } else if (err == EOPNOTSUPP) {
654 /*
655 * The file system server does not support FUSE_FALLOCATE with
656 * the supplied mode for this particular file.
657 */
658 err = EINVAL;
659 } else if (!err) {
660 *offset += *len;
661 *len = 0;
662 fuse_vnode_undirty_cached_timestamps(vp, false);
663 fuse_internal_clear_suid_on_write(vp, cred, curthread);
664 if (*offset > fvdat->cached_attrs.va_size) {
665 fuse_vnode_setsize(vp, *offset, false);
666 getnanouptime(&fvdat->last_local_modify);
667 }
668 }
669
670 return (err);
671 }
672
673 /* {
674 struct vnode *a_vp;
675 daddr_t a_bn;
676 struct bufobj **a_bop;
677 daddr_t *a_bnp;
678 int *a_runp;
679 int *a_runb;
680 } */
681 static int
682 fuse_vnop_bmap(struct vop_bmap_args *ap)
683 {
684 struct vnode *vp = ap->a_vp;
685 struct bufobj **bo = ap->a_bop;
686 struct thread *td = curthread;
687 struct mount *mp;
688 struct fuse_dispatcher fdi;
689 struct fuse_bmap_in *fbi;
690 struct fuse_bmap_out *fbo;
691 struct fuse_data *data;
692 struct fuse_vnode_data *fvdat = VTOFUD(vp);
693 uint64_t biosize;
694 off_t fsize;
695 daddr_t lbn = ap->a_bn;
696 daddr_t *pbn = ap->a_bnp;
697 int *runp = ap->a_runp;
698 int *runb = ap->a_runb;
699 int error = 0;
700 int maxrun;
701
702 if (fuse_isdeadfs(vp)) {
703 return ENXIO;
704 }
705
706 mp = vnode_mount(vp);
707 data = fuse_get_mpdata(mp);
708 biosize = fuse_iosize(vp);
709 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
710 data->max_readahead_blocks);
711
712 if (bo != NULL)
713 *bo = &vp->v_bufobj;
714
715 /*
716 * The FUSE_BMAP operation does not include the runp and runb
717 * variables, so we must guess. Report nonzero contiguous runs so
718 * cluster_read will combine adjacent reads. It's worthwhile to reduce
719 * upcalls even if we don't know the true physical layout of the file.
720 *
721 * FUSE file systems may opt out of read clustering in two ways:
722 * * mounting with -onoclusterr
723 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
724 */
725 if (runb != NULL)
726 *runb = MIN(lbn, maxrun);
727 if (runp != NULL && maxrun == 0)
728 *runp = 0;
729 else if (runp != NULL) {
730 /*
731 * If the file's size is cached, use that value to calculate
732 * runp, even if the cache is expired. runp is only advisory,
733 * and the risk of getting it wrong is not worth the cost of
734 * another upcall.
735 */
736 if (fvdat->cached_attrs.va_size != VNOVAL)
737 fsize = fvdat->cached_attrs.va_size;
738 else
739 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
740 if (error == 0)
741 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
742 maxrun);
743 else
744 *runp = 0;
745 }
746
747 if (fsess_maybe_impl(mp, FUSE_BMAP)) {
748 fdisp_init(&fdi, sizeof(*fbi));
749 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
750 fbi = fdi.indata;
751 fbi->block = lbn;
752 fbi->blocksize = biosize;
753 error = fdisp_wait_answ(&fdi);
754 if (error == ENOSYS) {
755 fdisp_destroy(&fdi);
756 fsess_set_notimpl(mp, FUSE_BMAP);
757 error = 0;
758 } else {
759 fbo = fdi.answ;
760 if (error == 0 && pbn != NULL)
761 *pbn = fbo->block;
762 fdisp_destroy(&fdi);
763 return error;
764 }
765 }
766
767 /* If the daemon doesn't support BMAP, make up a sensible default */
768 if (pbn != NULL)
769 *pbn = lbn * btodb(biosize);
770 return (error);
771 }
772
773 /*
774 struct vop_close_args {
775 struct vnode *a_vp;
776 int a_fflag;
777 struct ucred *a_cred;
778 struct thread *a_td;
779 };
780 */
781 static int
782 fuse_vnop_close(struct vop_close_args *ap)
783 {
784 struct vnode *vp = ap->a_vp;
785 struct ucred *cred = ap->a_cred;
786 int fflag = ap->a_fflag;
787 struct thread *td = ap->a_td;
788 pid_t pid = td->td_proc->p_pid;
789 struct fuse_vnode_data *fvdat = VTOFUD(vp);
790 int err = 0;
791
792 if (fuse_isdeadfs(vp))
793 return 0;
794 if (vnode_isdir(vp))
795 return 0;
796 if (fflag & IO_NDELAY)
797 return 0;
798
799 err = fuse_flush(vp, cred, pid, fflag);
800 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE)) {
801 struct vattr vap;
802
803 VATTR_NULL(&vap);
804 vap.va_atime = fvdat->cached_attrs.va_atime;
805 err = fuse_internal_setattr(vp, &vap, td, NULL);
806 }
807 /* TODO: close the file handle, if we're sure it's no longer used */
808 if ((fvdat->flag & FN_SIZECHANGE) != 0) {
809 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
810 }
811 return err;
812 }
813
814 /*
815 struct vop_copy_file_range_args {
816 struct vop_generic_args a_gen;
817 struct vnode *a_invp;
818 off_t *a_inoffp;
819 struct vnode *a_outvp;
820 off_t *a_outoffp;
821 size_t *a_lenp;
822 unsigned int a_flags;
823 struct ucred *a_incred;
824 struct ucred *a_outcred;
825 struct thread *a_fsizetd;
826 }
827 */
828 static int
829 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
830 {
831 struct vnode *invp = ap->a_invp;
832 struct vnode *outvp = ap->a_outvp;
833 struct mount *mp = vnode_mount(invp);
834 struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
835 struct fuse_dispatcher fdi;
836 struct fuse_filehandle *infufh, *outfufh;
837 struct fuse_copy_file_range_in *fcfri;
838 struct ucred *incred = ap->a_incred;
839 struct ucred *outcred = ap->a_outcred;
840 struct fuse_write_out *fwo;
841 struct thread *td;
842 struct uio io;
843 off_t outfilesize;
844 ssize_t r = 0;
845 pid_t pid;
846 int err;
847
848 if (mp != vnode_mount(outvp))
849 goto fallback;
850
851 if (incred->cr_uid != outcred->cr_uid)
852 goto fallback;
853
854 if (incred->cr_groups[0] != outcred->cr_groups[0])
855 goto fallback;
856
857 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
858 goto fallback;
859
860 if (ap->a_fsizetd == NULL)
861 td = curthread;
862 else
863 td = ap->a_fsizetd;
864 pid = td->td_proc->p_pid;
865
866 /* Lock both vnodes, avoiding risk of deadlock. */
867 do {
868 err = vn_lock(outvp, LK_EXCLUSIVE);
869 if (invp == outvp)
870 break;
871 if (err == 0) {
872 err = vn_lock(invp, LK_SHARED | LK_NOWAIT);
873 if (err == 0)
874 break;
875 VOP_UNLOCK(outvp);
876 err = vn_lock(invp, LK_SHARED);
877 if (err == 0)
878 VOP_UNLOCK(invp);
879 }
880 } while (err == 0);
881 if (err != 0)
882 return (err);
883
884 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
885 if (err)
886 goto unlock;
887
888 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
889 if (err)
890 goto unlock;
891
892 io.uio_resid = *ap->a_lenp;
893 if (ap->a_fsizetd) {
894 io.uio_offset = *ap->a_outoffp;
895 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
896 if (err != 0)
897 goto unlock;
898 }
899
900 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
901 if (err)
902 goto unlock;
903
904 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
905 *ap->a_outoffp + io.uio_resid);
906 if (err)
907 goto unlock;
908
909 fdisp_init(&fdi, sizeof(*fcfri));
910 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
911 fcfri = fdi.indata;
912 fcfri->fh_in = infufh->fh_id;
913 fcfri->off_in = *ap->a_inoffp;
914 fcfri->nodeid_out = VTOI(outvp);
915 fcfri->fh_out = outfufh->fh_id;
916 fcfri->off_out = *ap->a_outoffp;
917 fcfri->len = io.uio_resid;
918 fcfri->flags = 0;
919
920 err = fdisp_wait_answ(&fdi);
921 if (err == 0) {
922 fwo = fdi.answ;
923 *ap->a_lenp = fwo->size;
924 *ap->a_inoffp += fwo->size;
925 *ap->a_outoffp += fwo->size;
926 fuse_internal_clear_suid_on_write(outvp, outcred, td);
927 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
928 fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
929 getnanouptime(&outfvdat->last_local_modify);
930 }
931 fuse_vnode_update(invp, FN_ATIMECHANGE);
932 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
933 }
934 fdisp_destroy(&fdi);
935
936 unlock:
937 if (invp != outvp)
938 VOP_UNLOCK(invp);
939 VOP_UNLOCK(outvp);
940
941 if (err == ENOSYS) {
942 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
943 fallback:
944 err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
945 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
946 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
947 }
948
949 /*
950 * No need to call vn_rlimit_fsizex_res before return, since the uio is
951 * local.
952 */
953 return (err);
954 }
955
956 static void
957 fdisp_make_mknod_for_fallback(
958 struct fuse_dispatcher *fdip,
959 struct componentname *cnp,
960 struct vnode *dvp,
961 uint64_t parentnid,
962 struct thread *td,
963 struct ucred *cred,
964 mode_t mode,
965 enum fuse_opcode *op)
966 {
967 struct fuse_mknod_in *fmni;
968
969 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
970 *op = FUSE_MKNOD;
971 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
972 fmni = fdip->indata;
973 fmni->mode = mode;
974 fmni->rdev = 0;
975 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
976 cnp->cn_namelen);
977 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
978 }
979 /*
980 struct vnop_create_args {
981 struct vnode *a_dvp;
982 struct vnode **a_vpp;
983 struct componentname *a_cnp;
984 struct vattr *a_vap;
985 };
986 */
987 static int
988 fuse_vnop_create(struct vop_create_args *ap)
989 {
990 struct vnode *dvp = ap->a_dvp;
991 struct vnode **vpp = ap->a_vpp;
992 struct componentname *cnp = ap->a_cnp;
993 struct vattr *vap = ap->a_vap;
994 struct thread *td = curthread;
995 struct ucred *cred = cnp->cn_cred;
996
997 struct fuse_data *data;
998 struct fuse_create_in *fci;
999 struct fuse_entry_out *feo;
1000 struct fuse_open_out *foo;
1001 struct fuse_dispatcher fdi, fdi2;
1002 struct fuse_dispatcher *fdip = &fdi;
1003 struct fuse_dispatcher *fdip2 = NULL;
1004
1005 int err;
1006
1007 struct mount *mp = vnode_mount(dvp);
1008 data = fuse_get_mpdata(mp);
1009 uint64_t parentnid = VTOFUD(dvp)->nid;
1010 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1011 enum fuse_opcode op;
1012 int flags;
1013
1014 if (fuse_isdeadfs(dvp))
1015 return ENXIO;
1016
1017 /* FUSE expects sockets to be created with FUSE_MKNOD */
1018 if (vap->va_type == VSOCK)
1019 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1020
1021 /*
1022 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
1023 * writable mode makes sense, and we might as well include readability
1024 * too.
1025 */
1026 flags = O_RDWR;
1027
1028 bzero(&fdi, sizeof(fdi));
1029
1030 if (vap->va_type != VREG)
1031 return (EINVAL);
1032
1033 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1034 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
1035 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1036 cred, mode, &op);
1037 } else {
1038 /* Use FUSE_CREATE */
1039 size_t insize;
1040
1041 op = FUSE_CREATE;
1042 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1043 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1044 fci = fdip->indata;
1045 fci->mode = mode;
1046 fci->flags = O_CREAT | flags;
1047 if (fuse_libabi_geq(data, 7, 12)) {
1048 insize = sizeof(*fci);
1049 fci->umask = td->td_proc->p_pd->pd_cmask;
1050 } else {
1051 insize = sizeof(struct fuse_open_in);
1052 }
1053
1054 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1055 cnp->cn_namelen);
1056 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1057 }
1058
1059 err = fdisp_wait_answ(fdip);
1060
1061 if (err) {
1062 if (err == ENOSYS && op == FUSE_CREATE) {
1063 fsess_set_notimpl(mp, FUSE_CREATE);
1064 fdisp_destroy(fdip);
1065 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1066 parentnid, td, cred, mode, &op);
1067 err = fdisp_wait_answ(fdip);
1068 }
1069 if (err)
1070 goto out;
1071 }
1072
1073 feo = fdip->answ;
1074
1075 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1076 goto out;
1077 }
1078
1079 if (op == FUSE_CREATE) {
1080 if (fuse_libabi_geq(data, 7, 9))
1081 foo = (struct fuse_open_out*)(feo + 1);
1082 else
1083 foo = (struct fuse_open_out*)((char*)feo +
1084 FUSE_COMPAT_ENTRY_OUT_SIZE);
1085 } else {
1086 /* Issue a separate FUSE_OPEN */
1087 struct fuse_open_in *foi;
1088
1089 fdip2 = &fdi2;
1090 fdisp_init(fdip2, sizeof(*foi));
1091 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1092 cred);
1093 foi = fdip2->indata;
1094 foi->flags = flags;
1095 err = fdisp_wait_answ(fdip2);
1096 if (err)
1097 goto out;
1098 foo = fdip2->answ;
1099 }
1100 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1101 if (err) {
1102 struct fuse_release_in *fri;
1103 uint64_t nodeid = feo->nodeid;
1104 uint64_t fh_id = foo->fh;
1105
1106 fdisp_init(fdip, sizeof(*fri));
1107 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1108 fri = fdip->indata;
1109 fri->fh = fh_id;
1110 fri->flags = flags;
1111 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1112 fuse_insert_message(fdip->tick, false);
1113 goto out;
1114 }
1115 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1116 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1117 feo->attr_valid_nsec, NULL, true);
1118
1119 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1120 fuse_vnode_open(*vpp, foo->open_flags, td);
1121 /*
1122 * Purge the parent's attribute cache because the daemon should've
1123 * updated its mtime and ctime
1124 */
1125 fuse_vnode_clear_attr_cache(dvp);
1126 cache_purge_negative(dvp);
1127
1128 out:
1129 if (fdip2)
1130 fdisp_destroy(fdip2);
1131 fdisp_destroy(fdip);
1132 return err;
1133 }
1134
1135 /*
1136 struct vnop_fdatasync_args {
1137 struct vop_generic_args a_gen;
1138 struct vnode * a_vp;
1139 struct thread * a_td;
1140 };
1141 */
1142 static int
1143 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1144 {
1145 struct vnode *vp = ap->a_vp;
1146 struct thread *td = ap->a_td;
1147 int waitfor = MNT_WAIT;
1148
1149 int err = 0;
1150
1151 if (fuse_isdeadfs(vp)) {
1152 return 0;
1153 }
1154 if ((err = vop_stdfdatasync_buf(ap)))
1155 return err;
1156
1157 return fuse_internal_fsync(vp, td, waitfor, true);
1158 }
1159
1160 /*
1161 struct vnop_fsync_args {
1162 struct vop_generic_args a_gen;
1163 struct vnode * a_vp;
1164 int a_waitfor;
1165 struct thread * a_td;
1166 };
1167 */
1168 static int
1169 fuse_vnop_fsync(struct vop_fsync_args *ap)
1170 {
1171 struct vnode *vp = ap->a_vp;
1172 struct thread *td = ap->a_td;
1173 int waitfor = ap->a_waitfor;
1174 int err = 0;
1175
1176 if (fuse_isdeadfs(vp)) {
1177 return 0;
1178 }
1179 if ((err = vop_stdfsync(ap)))
1180 return err;
1181
1182 return fuse_internal_fsync(vp, td, waitfor, false);
1183 }
1184
1185 /*
1186 struct vnop_getattr_args {
1187 struct vnode *a_vp;
1188 struct vattr *a_vap;
1189 struct ucred *a_cred;
1190 struct thread *a_td;
1191 };
1192 */
1193 static int
1194 fuse_vnop_getattr(struct vop_getattr_args *ap)
1195 {
1196 struct vnode *vp = ap->a_vp;
1197 struct vattr *vap = ap->a_vap;
1198 struct ucred *cred = ap->a_cred;
1199 struct thread *td = curthread;
1200
1201 int err = 0;
1202 int dataflags;
1203
1204 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
1205
1206 /* Note that we are not bailing out on a dead file system just yet. */
1207
1208 if (!(dataflags & FSESS_INITED)) {
1209 if (!vnode_isvroot(vp)) {
1210 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
1211 err = ENOTCONN;
1212 return err;
1213 } else {
1214 goto fake;
1215 }
1216 }
1217 err = fuse_internal_getattr(vp, vap, cred, td);
1218 if (err == ENOTCONN && vnode_isvroot(vp)) {
1219 /* see comment in fuse_vfsop_statfs() */
1220 goto fake;
1221 } else {
1222 return err;
1223 }
1224
1225 fake:
1226 bzero(vap, sizeof(*vap));
1227 vap->va_type = vnode_vtype(vp);
1228
1229 return 0;
1230 }
1231
1232 /*
1233 struct vnop_inactive_args {
1234 struct vnode *a_vp;
1235 };
1236 */
1237 static int
1238 fuse_vnop_inactive(struct vop_inactive_args *ap)
1239 {
1240 struct vnode *vp = ap->a_vp;
1241 struct thread *td = curthread;
1242
1243 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1244 struct fuse_filehandle *fufh, *fufh_tmp;
1245
1246 int need_flush = 1;
1247
1248 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1249 if (need_flush && vp->v_type == VREG) {
1250 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1251 fuse_vnode_savesize(vp, NULL, 0);
1252 }
1253 if ((fvdat->flag & FN_REVOKED) != 0)
1254 fuse_io_invalbuf(vp, td);
1255 else
1256 fuse_io_flushbuf(vp, MNT_WAIT, td);
1257 need_flush = 0;
1258 }
1259 fuse_filehandle_close(vp, fufh, td, NULL);
1260 }
1261
1262 if ((fvdat->flag & FN_REVOKED) != 0)
1263 vrecycle(vp);
1264
1265 return 0;
1266 }
1267
1268 /*
1269 struct vnop_ioctl_args {
1270 struct vnode *a_vp;
1271 u_long a_command;
1272 caddr_t a_data;
1273 int a_fflag;
1274 struct ucred *a_cred;
1275 struct thread *a_td;
1276 };
1277 */
1278 static int
1279 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1280 {
1281 struct vnode *vp = ap->a_vp;
1282 struct mount *mp = vnode_mount(vp);
1283 struct ucred *cred = ap->a_cred;
1284 off_t *offp;
1285 pid_t pid = ap->a_td->td_proc->p_pid;
1286 int err;
1287
1288 switch (ap->a_command) {
1289 case FIOSEEKDATA:
1290 case FIOSEEKHOLE:
1291 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1292 if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1293 int whence;
1294
1295 offp = ap->a_data;
1296 if (ap->a_command == FIOSEEKDATA)
1297 whence = SEEK_DATA;
1298 else
1299 whence = SEEK_HOLE;
1300
1301 vn_lock(vp, LK_SHARED | LK_RETRY);
1302 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1303 whence);
1304 VOP_UNLOCK(vp);
1305 }
1306 if (fsess_not_impl(mp, FUSE_LSEEK))
1307 err = vop_stdioctl(ap);
1308 break;
1309 default:
1310 /* TODO: implement FUSE_IOCTL */
1311 err = ENOTTY;
1312 break;
1313 }
1314 return (err);
1315 }
1316
1317
1318 /*
1319 struct vnop_link_args {
1320 struct vnode *a_tdvp;
1321 struct vnode *a_vp;
1322 struct componentname *a_cnp;
1323 };
1324 */
1325 static int
1326 fuse_vnop_link(struct vop_link_args *ap)
1327 {
1328 struct vnode *vp = ap->a_vp;
1329 struct vnode *tdvp = ap->a_tdvp;
1330 struct componentname *cnp = ap->a_cnp;
1331
1332 struct vattr *vap = VTOVA(vp);
1333
1334 struct fuse_dispatcher fdi;
1335 struct fuse_entry_out *feo;
1336 struct fuse_link_in fli;
1337
1338 int err;
1339
1340 if (fuse_isdeadfs(vp)) {
1341 return ENXIO;
1342 }
1343 if (vnode_mount(tdvp) != vnode_mount(vp)) {
1344 return EXDEV;
1345 }
1346
1347 /*
1348 * This is a seatbelt check to protect naive userspace filesystems from
1349 * themselves and the limitations of the FUSE IPC protocol. If a
1350 * filesystem does not allow attribute caching, assume it is capable of
1351 * validating that nlink does not overflow.
1352 */
1353 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1354 return EMLINK;
1355 fli.oldnodeid = VTOI(vp);
1356
1357 fdisp_init(&fdi, 0);
1358 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1359 FUSE_LINK, &fli, sizeof(fli), &fdi);
1360 if ((err = fdisp_wait_answ(&fdi))) {
1361 goto out;
1362 }
1363 feo = fdi.answ;
1364
1365 if (fli.oldnodeid != feo->nodeid) {
1366 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1367 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE,
1368 "Assigned wrong inode for a hard link.");
1369 fuse_vnode_clear_attr_cache(vp);
1370 fuse_vnode_clear_attr_cache(tdvp);
1371 err = EIO;
1372 goto out;
1373 }
1374
1375 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1376 if (!err) {
1377 /*
1378 * Purge the parent's attribute cache because the daemon
1379 * should've updated its mtime and ctime
1380 */
1381 fuse_vnode_clear_attr_cache(tdvp);
1382 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1383 feo->attr_valid_nsec, NULL, true);
1384 }
1385 out:
1386 fdisp_destroy(&fdi);
1387 return err;
1388 }
1389
1390 struct fuse_lookup_alloc_arg {
1391 struct fuse_entry_out *feo;
1392 struct componentname *cnp;
1393 uint64_t nid;
1394 enum vtype vtyp;
1395 };
1396
1397 /* Callback for vn_get_ino */
1398 static int
1399 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1400 {
1401 struct fuse_lookup_alloc_arg *flaa = arg;
1402
1403 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1404 flaa->vtyp);
1405 }
1406
1407 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1408 "int", "struct timespec*", "struct timespec*");
1409 /*
1410 struct vnop_lookup_args {
1411 struct vnodeop_desc *a_desc;
1412 struct vnode *a_dvp;
1413 struct vnode **a_vpp;
1414 struct componentname *a_cnp;
1415 };
1416 */
1417 int
1418 fuse_vnop_lookup(struct vop_lookup_args *ap)
1419 {
1420 struct vnode *dvp = ap->a_dvp;
1421 struct vnode **vpp = ap->a_vpp;
1422 struct componentname *cnp = ap->a_cnp;
1423 struct thread *td = curthread;
1424 struct ucred *cred = cnp->cn_cred;
1425 struct timespec now;
1426
1427 int nameiop = cnp->cn_nameiop;
1428 int flags = cnp->cn_flags;
1429 int islastcn = flags & ISLASTCN;
1430 struct mount *mp = vnode_mount(dvp);
1431 struct fuse_data *data = fuse_get_mpdata(mp);
1432 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1433 bool is_dot;
1434
1435 int err = 0;
1436 int lookup_err = 0;
1437 struct vnode *vp = NULL;
1438
1439 struct fuse_dispatcher fdi;
1440 bool did_lookup = false;
1441 struct fuse_entry_out *feo = NULL;
1442 enum vtype vtyp; /* vnode type of target */
1443
1444 uint64_t nid;
1445
1446 if (fuse_isdeadfs(dvp)) {
1447 *vpp = NULL;
1448 return ENXIO;
1449 }
1450 if (!vnode_isdir(dvp))
1451 return ENOTDIR;
1452
1453 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1454 return EROFS;
1455
1456 if ((cnp->cn_flags & NOEXECCHECK) != 0)
1457 cnp->cn_flags &= ~NOEXECCHECK;
1458 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1459 return err;
1460
1461 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1462 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT))
1463 {
1464 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1465 /*
1466 * Since the file system doesn't support ".." lookups,
1467 * we have no way to find this entry.
1468 */
1469 return ESTALE;
1470 }
1471 nid = VTOFUD(dvp)->parent_nid;
1472 if (nid == 0)
1473 return ENOENT;
1474 /* .. is obviously a directory */
1475 vtyp = VDIR;
1476 } else if (is_dot) {
1477 nid = VTOI(dvp);
1478 /* . is obviously a directory */
1479 vtyp = VDIR;
1480 } else {
1481 struct timespec timeout;
1482 int ncpticks; /* here to accommodate for API contract */
1483
1484 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1485 getnanouptime(&now);
1486 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1487 switch (err) {
1488 case -1: /* positive match */
1489 if (timespeccmp(&timeout, &now, >)) {
1490 counter_u64_add(fuse_lookup_cache_hits, 1);
1491 } else {
1492 /* Cache timeout */
1493 counter_u64_add(fuse_lookup_cache_misses, 1);
1494 bintime_clear(
1495 &VTOFUD(*vpp)->entry_cache_timeout);
1496 cache_purge(*vpp);
1497 if (dvp != *vpp)
1498 vput(*vpp);
1499 else
1500 vrele(*vpp);
1501 *vpp = NULL;
1502 break;
1503 }
1504 return 0;
1505
1506 case 0: /* no match in cache */
1507 counter_u64_add(fuse_lookup_cache_misses, 1);
1508 break;
1509
1510 case ENOENT: /* negative match */
1511 if (timespeccmp(&timeout, &now, <=)) {
1512 /* Cache timeout */
1513 cache_purge_negative(dvp);
1514 break;
1515 }
1516 /* fall through */
1517 default:
1518 return err;
1519 }
1520
1521 fdisp_init(&fdi, cnp->cn_namelen + 1);
1522 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1523
1524 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1525 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1526 lookup_err = fdisp_wait_answ(&fdi);
1527 did_lookup = true;
1528
1529 if (!lookup_err) {
1530 /* lookup call succeeded */
1531 feo = (struct fuse_entry_out *)fdi.answ;
1532 nid = feo->nodeid;
1533 if (nid == 0) {
1534 /* zero nodeid means ENOENT and cache it */
1535 struct timespec timeout;
1536
1537 fdi.answ_stat = ENOENT;
1538 lookup_err = ENOENT;
1539 if (cnp->cn_flags & MAKEENTRY) {
1540 fuse_validity_2_timespec(feo, &timeout);
1541 /* Use the same entry_time for .. as for
1542 * the file itself. That doesn't honor
1543 * exactly what the fuse server tells
1544 * us, but to do otherwise would require
1545 * another cache lookup at this point.
1546 */
1547 struct timespec *dtsp = NULL;
1548 cache_enter_time(dvp, *vpp, cnp,
1549 &timeout, dtsp);
1550 }
1551 }
1552 vtyp = IFTOVT(feo->attr.mode);
1553 }
1554 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1555 fdisp_destroy(&fdi);
1556 return lookup_err;
1557 }
1558 }
1559 /* lookup_err, if non-zero, must be ENOENT at this point */
1560
1561 if (lookup_err) {
1562 /* Entry not found */
1563 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1564 if (default_permissions)
1565 err = fuse_internal_access(dvp, VWRITE, td,
1566 cred);
1567 else
1568 err = 0;
1569 if (!err) {
1570 err = EJUSTRETURN;
1571 }
1572 } else {
1573 err = ENOENT;
1574 }
1575 } else {
1576 /* Entry was found */
1577 if (flags & ISDOTDOT) {
1578 struct fuse_lookup_alloc_arg flaa;
1579
1580 flaa.nid = nid;
1581 flaa.feo = feo;
1582 flaa.cnp = cnp;
1583 flaa.vtyp = vtyp;
1584 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1585 &vp);
1586 *vpp = vp;
1587 } else if (nid == VTOI(dvp)) {
1588 if (is_dot) {
1589 vref(dvp);
1590 *vpp = dvp;
1591 } else {
1592 fuse_warn(fuse_get_mpdata(mp),
1593 FSESS_WARN_ILLEGAL_INODE,
1594 "Assigned same inode to both parent and "
1595 "child.");
1596 err = EIO;
1597 }
1598
1599 } else {
1600 struct fuse_vnode_data *fvdat;
1601
1602 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1603 &vp, cnp, vtyp);
1604 if (err)
1605 goto out;
1606 *vpp = vp;
1607 fvdat = VTOFUD(vp);
1608
1609 MPASS(feo != NULL);
1610 if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1611 /*
1612 * Attributes from the server are definitely
1613 * newer than the last attributes we sent to
1614 * the server, so cache them.
1615 */
1616 fuse_internal_cache_attrs(*vpp, &feo->attr,
1617 feo->attr_valid, feo->attr_valid_nsec,
1618 NULL, true);
1619 }
1620 fuse_validity_2_bintime(feo->entry_valid,
1621 feo->entry_valid_nsec,
1622 &fvdat->entry_cache_timeout);
1623
1624 if ((nameiop == DELETE || nameiop == RENAME) &&
1625 islastcn && default_permissions)
1626 {
1627 struct vattr dvattr;
1628
1629 err = fuse_internal_access(dvp, VWRITE, td,
1630 cred);
1631 if (err != 0)
1632 goto out;
1633 /*
1634 * if the parent's sticky bit is set, check
1635 * whether we're allowed to remove the file.
1636 * Need to figure out the vnode locking to make
1637 * this work.
1638 */
1639 fuse_internal_getattr(dvp, &dvattr, cred, td);
1640 if ((dvattr.va_mode & S_ISTXT) &&
1641 fuse_internal_access(dvp, VADMIN, td,
1642 cred) &&
1643 fuse_internal_access(*vpp, VADMIN, td,
1644 cred)) {
1645 err = EPERM;
1646 goto out;
1647 }
1648 }
1649 }
1650 }
1651 out:
1652 if (err) {
1653 if (vp != NULL && dvp != vp)
1654 vput(vp);
1655 else if (vp != NULL)
1656 vrele(vp);
1657 *vpp = NULL;
1658 }
1659 if (did_lookup)
1660 fdisp_destroy(&fdi);
1661
1662 return err;
1663 }
1664
1665 /*
1666 struct vnop_mkdir_args {
1667 struct vnode *a_dvp;
1668 struct vnode **a_vpp;
1669 struct componentname *a_cnp;
1670 struct vattr *a_vap;
1671 };
1672 */
1673 static int
1674 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1675 {
1676 struct vnode *dvp = ap->a_dvp;
1677 struct vnode **vpp = ap->a_vpp;
1678 struct componentname *cnp = ap->a_cnp;
1679 struct vattr *vap = ap->a_vap;
1680
1681 struct fuse_mkdir_in fmdi;
1682
1683 if (fuse_isdeadfs(dvp)) {
1684 return ENXIO;
1685 }
1686 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1687 fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1688
1689 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1690 sizeof(fmdi), VDIR));
1691 }
1692
1693 /*
1694 struct vnop_mknod_args {
1695 struct vnode *a_dvp;
1696 struct vnode **a_vpp;
1697 struct componentname *a_cnp;
1698 struct vattr *a_vap;
1699 };
1700 */
1701 static int
1702 fuse_vnop_mknod(struct vop_mknod_args *ap)
1703 {
1704
1705 struct vnode *dvp = ap->a_dvp;
1706 struct vnode **vpp = ap->a_vpp;
1707 struct componentname *cnp = ap->a_cnp;
1708 struct vattr *vap = ap->a_vap;
1709
1710 if (fuse_isdeadfs(dvp))
1711 return ENXIO;
1712
1713 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1714 }
1715
1716 /*
1717 struct vop_open_args {
1718 struct vnode *a_vp;
1719 int a_mode;
1720 struct ucred *a_cred;
1721 struct thread *a_td;
1722 int a_fdidx; / struct file *a_fp;
1723 };
1724 */
1725 static int
1726 fuse_vnop_open(struct vop_open_args *ap)
1727 {
1728 struct vnode *vp = ap->a_vp;
1729 int a_mode = ap->a_mode;
1730 struct thread *td = ap->a_td;
1731 struct ucred *cred = ap->a_cred;
1732 pid_t pid = td->td_proc->p_pid;
1733
1734 if (fuse_isdeadfs(vp))
1735 return ENXIO;
1736 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1737 return (EOPNOTSUPP);
1738 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1739 return EINVAL;
1740
1741 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1742 fuse_vnode_open(vp, 0, td);
1743 return 0;
1744 }
1745
1746 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1747 }
1748
1749 static int
1750 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1751 {
1752 struct vnode *vp = ap->a_vp;
1753 struct mount *mp;
1754
1755 switch (ap->a_name) {
1756 case _PC_FILESIZEBITS:
1757 *ap->a_retval = 64;
1758 return (0);
1759 case _PC_NAME_MAX:
1760 *ap->a_retval = NAME_MAX;
1761 return (0);
1762 case _PC_LINK_MAX:
1763 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1764 return (0);
1765 case _PC_SYMLINK_MAX:
1766 *ap->a_retval = MAXPATHLEN;
1767 return (0);
1768 case _PC_NO_TRUNC:
1769 *ap->a_retval = 1;
1770 return (0);
1771 case _PC_MIN_HOLE_SIZE:
1772 /*
1773 * The FUSE protocol provides no mechanism for a server to
1774 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead,
1775 * return EINVAL if the server does not support FUSE_LSEEK, or
1776 * 1 if it does.
1777 */
1778 mp = vnode_mount(vp);
1779 if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1780 !fsess_not_impl(mp, FUSE_LSEEK)) {
1781 off_t offset = 0;
1782
1783 /* Issue a FUSE_LSEEK to find out if it's implemented */
1784 fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred,
1785 curthread->td_proc->p_pid, &offset, SEEK_DATA);
1786 }
1787
1788 if (fsess_is_impl(mp, FUSE_LSEEK)) {
1789 *ap->a_retval = 1;
1790 return (0);
1791 } else {
1792 /*
1793 * Probably FUSE_LSEEK is not implemented. It might
1794 * be, if the FUSE_LSEEK above returned an error like
1795 * EACCES, but in that case we can't tell, so it's
1796 * safest to report EINVAL anyway.
1797 */
1798 return (EINVAL);
1799 }
1800 default:
1801 return (vop_stdpathconf(ap));
1802 }
1803 }
1804
1805 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1806 "struct uio*", "struct ucred*");
1807 /*
1808 struct vnop_read_args {
1809 struct vnode *a_vp;
1810 struct uio *a_uio;
1811 int a_ioflag;
1812 struct ucred *a_cred;
1813 };
1814 */
1815 static int
1816 fuse_vnop_read(struct vop_read_args *ap)
1817 {
1818 struct vnode *vp = ap->a_vp;
1819 struct uio *uio = ap->a_uio;
1820 int ioflag = ap->a_ioflag;
1821 struct ucred *cred = ap->a_cred;
1822 pid_t pid = curthread->td_proc->p_pid;
1823 struct fuse_filehandle *fufh;
1824 int err;
1825 bool closefufh = false, directio;
1826
1827 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1828
1829 if (fuse_isdeadfs(vp)) {
1830 return ENXIO;
1831 }
1832
1833 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1834 ioflag |= IO_DIRECT;
1835 }
1836
1837 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1838 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1839 /*
1840 * nfsd will do I/O without first doing VOP_OPEN. We
1841 * must implicitly open the file here
1842 */
1843 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1844 closefufh = true;
1845 }
1846 if (err) {
1847 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1848 return err;
1849 }
1850
1851 /*
1852 * Ideally, when the daemon asks for direct io at open time, the
1853 * standard file flag should be set according to this, so that would
1854 * just change the default mode, which later on could be changed via
1855 * fcntl(2).
1856 * But this doesn't work, the O_DIRECT flag gets cleared at some point
1857 * (don't know where). So to make any use of the Fuse direct_io option,
1858 * we hardwire it into the file's private data (similarly to Linux,
1859 * btw.).
1860 */
1861 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1862
1863 fuse_vnode_update(vp, FN_ATIMECHANGE);
1864 if (directio) {
1865 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1866 err = fuse_read_directbackend(vp, uio, cred, fufh);
1867 } else {
1868 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1869 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1870 }
1871
1872 if (closefufh)
1873 fuse_filehandle_close(vp, fufh, curthread, cred);
1874
1875 return (err);
1876 }
1877
1878 /*
1879 struct vnop_readdir_args {
1880 struct vnode *a_vp;
1881 struct uio *a_uio;
1882 struct ucred *a_cred;
1883 int *a_eofflag;
1884 int *a_ncookies;
1885 uint64_t **a_cookies;
1886 };
1887 */
1888 static int
1889 fuse_vnop_readdir(struct vop_readdir_args *ap)
1890 {
1891 struct vnode *vp = ap->a_vp;
1892 struct uio *uio = ap->a_uio;
1893 struct ucred *cred = ap->a_cred;
1894 struct fuse_filehandle *fufh = NULL;
1895 struct mount *mp = vnode_mount(vp);
1896 struct fuse_iov cookediov;
1897 int err = 0;
1898 uint64_t *cookies;
1899 ssize_t tresid;
1900 int ncookies;
1901 bool closefufh = false;
1902 pid_t pid = curthread->td_proc->p_pid;
1903
1904 if (ap->a_eofflag)
1905 *ap->a_eofflag = 0;
1906 if (fuse_isdeadfs(vp)) {
1907 return ENXIO;
1908 }
1909 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1910 (uio_resid(uio) < sizeof(struct dirent))) {
1911 return EINVAL;
1912 }
1913
1914 tresid = uio->uio_resid;
1915 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1916 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1917 KASSERT(fuse_get_mpdata(mp)->dataflags
1918 & FSESS_NO_OPENDIR_SUPPORT,
1919 ("FUSE file systems that don't set "
1920 "FUSE_NO_OPENDIR_SUPPORT should not be exported"));
1921 /*
1922 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1923 * must implicitly open the directory here.
1924 */
1925 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1926 closefufh = true;
1927 }
1928 if (err)
1929 return (err);
1930 if (ap->a_ncookies != NULL) {
1931 ncookies = uio->uio_resid /
1932 (offsetof(struct dirent, d_name) + 4) + 1;
1933 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1934 *ap->a_ncookies = ncookies;
1935 *ap->a_cookies = cookies;
1936 } else {
1937 ncookies = 0;
1938 cookies = NULL;
1939 }
1940 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1941 fiov_init(&cookediov, DIRCOOKEDSIZE);
1942
1943 err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1944 &ncookies, cookies);
1945
1946 fiov_teardown(&cookediov);
1947 if (closefufh)
1948 fuse_filehandle_close(vp, fufh, curthread, cred);
1949
1950 if (ap->a_ncookies != NULL) {
1951 if (err == 0) {
1952 *ap->a_ncookies -= ncookies;
1953 } else {
1954 free(*ap->a_cookies, M_TEMP);
1955 *ap->a_ncookies = 0;
1956 *ap->a_cookies = NULL;
1957 }
1958 }
1959 if (err == 0 && tresid == uio->uio_resid)
1960 *ap->a_eofflag = 1;
1961
1962 return err;
1963 }
1964
1965 /*
1966 struct vnop_readlink_args {
1967 struct vnode *a_vp;
1968 struct uio *a_uio;
1969 struct ucred *a_cred;
1970 };
1971 */
1972 static int
1973 fuse_vnop_readlink(struct vop_readlink_args *ap)
1974 {
1975 struct vnode *vp = ap->a_vp;
1976 struct uio *uio = ap->a_uio;
1977 struct ucred *cred = ap->a_cred;
1978
1979 struct fuse_dispatcher fdi;
1980 int err;
1981
1982 if (fuse_isdeadfs(vp)) {
1983 return ENXIO;
1984 }
1985 if (!vnode_islnk(vp)) {
1986 return EINVAL;
1987 }
1988 fdisp_init(&fdi, 0);
1989 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1990 if (err) {
1991 goto out;
1992 }
1993 if (((char *)fdi.answ)[0] == '/' &&
1994 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1995 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1996
1997 err = uiomove(mpth, strlen(mpth), uio);
1998 }
1999 if (!err) {
2000 err = uiomove(fdi.answ, fdi.iosize, uio);
2001 }
2002 out:
2003 fdisp_destroy(&fdi);
2004 return err;
2005 }
2006
2007 /*
2008 struct vnop_reclaim_args {
2009 struct vnode *a_vp;
2010 };
2011 */
2012 static int
2013 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2014 {
2015 struct vnode *vp = ap->a_vp;
2016 struct thread *td = curthread;
2017 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2018 struct fuse_filehandle *fufh, *fufh_tmp;
2019
2020 if (!fvdat) {
2021 panic("FUSE: no vnode data during recycling");
2022 }
2023 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2024 printf("FUSE: vnode being reclaimed with open fufh "
2025 "(type=%#x)", fufh->fufh_type);
2026 fuse_filehandle_close(vp, fufh, td, NULL);
2027 }
2028
2029 if (VTOI(vp) == 1) {
2030 /*
2031 * Don't send FUSE_FORGET for the root inode, because
2032 * we never send FUSE_LOOKUP for it (see
2033 * fuse_vfsop_root) and we don't want the server to see
2034 * mismatched lookup counts.
2035 */
2036 struct fuse_data *data;
2037 struct vnode *vroot;
2038
2039 data = fuse_get_mpdata(vnode_mount(vp));
2040 FUSE_LOCK();
2041 vroot = data->vroot;
2042 data->vroot = NULL;
2043 FUSE_UNLOCK();
2044 if (vroot)
2045 vrele(vroot);
2046 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2047 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2048 fvdat->nlookup);
2049 }
2050 cache_purge(vp);
2051 vfs_hash_remove(vp);
2052 fuse_vnode_destroy(vp);
2053
2054 return 0;
2055 }
2056
2057 /*
2058 struct vnop_remove_args {
2059 struct vnode *a_dvp;
2060 struct vnode *a_vp;
2061 struct componentname *a_cnp;
2062 };
2063 */
2064 static int
2065 fuse_vnop_remove(struct vop_remove_args *ap)
2066 {
2067 struct vnode *dvp = ap->a_dvp;
2068 struct vnode *vp = ap->a_vp;
2069 struct componentname *cnp = ap->a_cnp;
2070
2071 int err;
2072
2073 if (fuse_isdeadfs(vp)) {
2074 return ENXIO;
2075 }
2076 if (vnode_isdir(vp)) {
2077 return EPERM;
2078 }
2079
2080 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2081
2082 return err;
2083 }
2084
2085 /*
2086 struct vnop_rename_args {
2087 struct vnode *a_fdvp;
2088 struct vnode *a_fvp;
2089 struct componentname *a_fcnp;
2090 struct vnode *a_tdvp;
2091 struct vnode *a_tvp;
2092 struct componentname *a_tcnp;
2093 };
2094 */
2095 static int
2096 fuse_vnop_rename(struct vop_rename_args *ap)
2097 {
2098 struct vnode *fdvp = ap->a_fdvp;
2099 struct vnode *fvp = ap->a_fvp;
2100 struct componentname *fcnp = ap->a_fcnp;
2101 struct vnode *tdvp = ap->a_tdvp;
2102 struct vnode *tvp = ap->a_tvp;
2103 struct componentname *tcnp = ap->a_tcnp;
2104 struct fuse_data *data;
2105 bool newparent = fdvp != tdvp;
2106 bool isdir = fvp->v_type == VDIR;
2107 int err = 0;
2108
2109 if (fuse_isdeadfs(fdvp)) {
2110 return ENXIO;
2111 }
2112 if (fvp->v_mount != tdvp->v_mount ||
2113 (tvp && fvp->v_mount != tvp->v_mount)) {
2114 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2115 err = EXDEV;
2116 goto out;
2117 }
2118 cache_purge(fvp);
2119
2120 /*
2121 * FUSE library is expected to check if target directory is not
2122 * under the source directory in the file system tree.
2123 * Linux performs this check at VFS level.
2124 */
2125 /*
2126 * If source is a directory, and it will get a new parent, user must
2127 * have write permission to it, so ".." can be modified.
2128 */
2129 data = fuse_get_mpdata(vnode_mount(tdvp));
2130 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2131 err = fuse_internal_access(fvp, VWRITE,
2132 curthread, tcnp->cn_cred);
2133 if (err)
2134 goto out;
2135 }
2136 sx_xlock(&data->rename_lock);
2137 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2138 if (err == 0) {
2139 if (tdvp != fdvp)
2140 fuse_vnode_setparent(fvp, tdvp);
2141 if (tvp != NULL)
2142 fuse_vnode_setparent(tvp, NULL);
2143 }
2144 sx_unlock(&data->rename_lock);
2145
2146 if (tvp != NULL && tvp != fvp) {
2147 cache_purge(tvp);
2148 }
2149 if (vnode_isdir(fvp)) {
2150 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2151 cache_purge(tdvp);
2152 }
2153 cache_purge(fdvp);
2154 }
2155 out:
2156 if (tdvp == tvp) {
2157 vrele(tdvp);
2158 } else {
2159 vput(tdvp);
2160 }
2161 if (tvp != NULL) {
2162 vput(tvp);
2163 }
2164 vrele(fdvp);
2165 vrele(fvp);
2166
2167 return err;
2168 }
2169
2170 /*
2171 struct vnop_rmdir_args {
2172 struct vnode *a_dvp;
2173 struct vnode *a_vp;
2174 struct componentname *a_cnp;
2175 } *ap;
2176 */
2177 static int
2178 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2179 {
2180 struct vnode *dvp = ap->a_dvp;
2181 struct vnode *vp = ap->a_vp;
2182
2183 int err;
2184
2185 if (fuse_isdeadfs(vp)) {
2186 return ENXIO;
2187 }
2188 if (VTOFUD(vp) == VTOFUD(dvp)) {
2189 return EINVAL;
2190 }
2191 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2192
2193 return err;
2194 }
2195
2196 /*
2197 struct vnop_setattr_args {
2198 struct vnode *a_vp;
2199 struct vattr *a_vap;
2200 struct ucred *a_cred;
2201 struct thread *a_td;
2202 };
2203 */
2204 static int
2205 fuse_vnop_setattr(struct vop_setattr_args *ap)
2206 {
2207 struct vnode *vp = ap->a_vp;
2208 struct vattr *vap = ap->a_vap;
2209 struct ucred *cred = ap->a_cred;
2210 struct thread *td = curthread;
2211 struct mount *mp;
2212 struct fuse_data *data;
2213 struct vattr old_va;
2214 int dataflags;
2215 int err = 0, err2;
2216 accmode_t accmode = 0;
2217 bool checkperm;
2218 bool drop_suid = false;
2219 gid_t cr_gid;
2220
2221 mp = vnode_mount(vp);
2222 data = fuse_get_mpdata(mp);
2223 dataflags = data->dataflags;
2224 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2225 if (cred->cr_ngroups > 0)
2226 cr_gid = cred->cr_groups[0];
2227 else
2228 cr_gid = 0;
2229
2230 if (fuse_isdeadfs(vp)) {
2231 return ENXIO;
2232 }
2233
2234 if (vap->va_uid != (uid_t)VNOVAL) {
2235 if (checkperm) {
2236 /* Only root may change a file's owner */
2237 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2238 if (err) {
2239 /* As a special case, allow the null chown */
2240 err2 = fuse_internal_getattr(vp, &old_va, cred,
2241 td);
2242 if (err2)
2243 return (err2);
2244 if (vap->va_uid != old_va.va_uid)
2245 return err;
2246 else
2247 accmode |= VADMIN;
2248 drop_suid = true;
2249 } else
2250 accmode |= VADMIN;
2251 } else
2252 accmode |= VADMIN;
2253 }
2254 if (vap->va_gid != (gid_t)VNOVAL) {
2255 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2256 drop_suid = true;
2257 if (checkperm && !groupmember(vap->va_gid, cred))
2258 {
2259 /*
2260 * Non-root users may only chgrp to one of their own
2261 * groups
2262 */
2263 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2264 if (err) {
2265 /* As a special case, allow the null chgrp */
2266 err2 = fuse_internal_getattr(vp, &old_va, cred,
2267 td);
2268 if (err2)
2269 return (err2);
2270 if (vap->va_gid != old_va.va_gid)
2271 return err;
2272 accmode |= VADMIN;
2273 } else
2274 accmode |= VADMIN;
2275 } else
2276 accmode |= VADMIN;
2277 }
2278 if (vap->va_size != VNOVAL) {
2279 switch (vp->v_type) {
2280 case VDIR:
2281 return (EISDIR);
2282 case VLNK:
2283 case VREG:
2284 if (vfs_isrdonly(mp))
2285 return (EROFS);
2286 err = vn_rlimit_trunc(vap->va_size, td);
2287 if (err)
2288 return (err);
2289 break;
2290 default:
2291 /*
2292 * According to POSIX, the result is unspecified
2293 * for file types other than regular files,
2294 * directories and shared memory objects. We
2295 * don't support shared memory objects in the file
2296 * system, and have dubious support for truncating
2297 * symlinks. Just ignore the request in other cases.
2298 */
2299 return (0);
2300 }
2301 /* Don't set accmode. Permission to trunc is checked upstack */
2302 }
2303 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2304 if (vap->va_vaflags & VA_UTIMES_NULL)
2305 accmode |= VWRITE;
2306 else
2307 accmode |= VADMIN;
2308 }
2309 if (drop_suid) {
2310 if (vap->va_mode != (mode_t)VNOVAL)
2311 vap->va_mode &= ~(S_ISUID | S_ISGID);
2312 else {
2313 err = fuse_internal_getattr(vp, &old_va, cred, td);
2314 if (err)
2315 return (err);
2316 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2317 }
2318 }
2319 if (vap->va_mode != (mode_t)VNOVAL) {
2320 /* Only root may set the sticky bit on non-directories */
2321 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2322 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2323 return EFTYPE;
2324 if (checkperm && (vap->va_mode & S_ISGID)) {
2325 err = fuse_internal_getattr(vp, &old_va, cred, td);
2326 if (err)
2327 return (err);
2328 if (!groupmember(old_va.va_gid, cred)) {
2329 err = priv_check_cred(cred, PRIV_VFS_SETGID);
2330 if (err)
2331 return (err);
2332 }
2333 }
2334 accmode |= VADMIN;
2335 }
2336
2337 if (vfs_isrdonly(mp))
2338 return EROFS;
2339
2340 if (checkperm) {
2341 err = fuse_internal_access(vp, accmode, td, cred);
2342 } else {
2343 err = 0;
2344 }
2345 if (err)
2346 return err;
2347 else
2348 return fuse_internal_setattr(vp, vap, td, cred);
2349 }
2350
2351 /*
2352 struct vnop_strategy_args {
2353 struct vnode *a_vp;
2354 struct buf *a_bp;
2355 };
2356 */
2357 static int
2358 fuse_vnop_strategy(struct vop_strategy_args *ap)
2359 {
2360 struct vnode *vp = ap->a_vp;
2361 struct buf *bp = ap->a_bp;
2362
2363 if (!vp || fuse_isdeadfs(vp)) {
2364 bp->b_ioflags |= BIO_ERROR;
2365 bp->b_error = ENXIO;
2366 bufdone(bp);
2367 return 0;
2368 }
2369
2370 /*
2371 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2372 * fuse_io_strategy sets bp's error fields
2373 */
2374 (void)fuse_io_strategy(vp, bp);
2375
2376 return 0;
2377 }
2378
2379 /*
2380 struct vnop_symlink_args {
2381 struct vnode *a_dvp;
2382 struct vnode **a_vpp;
2383 struct componentname *a_cnp;
2384 struct vattr *a_vap;
2385 char *a_target;
2386 };
2387 */
2388 static int
2389 fuse_vnop_symlink(struct vop_symlink_args *ap)
2390 {
2391 struct vnode *dvp = ap->a_dvp;
2392 struct vnode **vpp = ap->a_vpp;
2393 struct componentname *cnp = ap->a_cnp;
2394 const char *target = ap->a_target;
2395
2396 struct fuse_dispatcher fdi;
2397
2398 int err;
2399 size_t len;
2400
2401 if (fuse_isdeadfs(dvp)) {
2402 return ENXIO;
2403 }
2404 /*
2405 * Unlike the other creator type calls, here we have to create a message
2406 * where the name of the new entry comes first, and the data describing
2407 * the entry comes second.
2408 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2409 * but put together the message manually and just call the core part.
2410 */
2411
2412 len = strlen(target) + 1;
2413 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2414 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2415
2416 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2417 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2418 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2419
2420 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2421 fdisp_destroy(&fdi);
2422 return err;
2423 }
2424
2425 /*
2426 struct vnop_write_args {
2427 struct vnode *a_vp;
2428 struct uio *a_uio;
2429 int a_ioflag;
2430 struct ucred *a_cred;
2431 };
2432 */
2433 static int
2434 fuse_vnop_write(struct vop_write_args *ap)
2435 {
2436 struct vnode *vp = ap->a_vp;
2437 struct uio *uio = ap->a_uio;
2438 int ioflag = ap->a_ioflag;
2439 struct ucred *cred = ap->a_cred;
2440 pid_t pid = curthread->td_proc->p_pid;
2441 struct fuse_filehandle *fufh;
2442 int err;
2443 bool closefufh = false, directio;
2444
2445 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2446
2447 if (fuse_isdeadfs(vp)) {
2448 return ENXIO;
2449 }
2450
2451 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2452 ioflag |= IO_DIRECT;
2453 }
2454
2455 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2456 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2457 /*
2458 * nfsd will do I/O without first doing VOP_OPEN. We
2459 * must implicitly open the file here
2460 */
2461 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2462 closefufh = true;
2463 }
2464 if (err) {
2465 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2466 return err;
2467 }
2468
2469 /*
2470 * Ideally, when the daemon asks for direct io at open time, the
2471 * standard file flag should be set according to this, so that would
2472 * just change the default mode, which later on could be changed via
2473 * fcntl(2).
2474 * But this doesn't work, the O_DIRECT flag gets cleared at some point
2475 * (don't know where). So to make any use of the Fuse direct_io option,
2476 * we hardwire it into the file's private data (similarly to Linux,
2477 * btw.).
2478 */
2479 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2480
2481 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2482 if (directio) {
2483 off_t start, end, filesize;
2484 bool pages = (ioflag & IO_VMIO) != 0;
2485
2486 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2487
2488 err = fuse_vnode_size(vp, &filesize, cred, curthread);
2489 if (err)
2490 goto out;
2491
2492 start = uio->uio_offset;
2493 end = start + uio->uio_resid;
2494 if (!pages) {
2495 err = fuse_inval_buf_range(vp, filesize, start,
2496 end);
2497 if (err)
2498 goto out;
2499 }
2500 err = fuse_write_directbackend(vp, uio, cred, fufh,
2501 filesize, ioflag, pages);
2502 } else {
2503 SDT_PROBE2(fusefs, , vnops, trace, 1,
2504 "buffered write of vnode");
2505 if (!fsess_opt_writeback(vnode_mount(vp)))
2506 ioflag |= IO_SYNC;
2507 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2508 }
2509 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2510
2511 out:
2512 if (closefufh)
2513 fuse_filehandle_close(vp, fufh, curthread, cred);
2514
2515 return (err);
2516 }
2517
2518 static daddr_t
2519 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2520 {
2521 const int biosize = fuse_iosize(vp);
2522
2523 return (off / biosize);
2524 }
2525
2526 static int
2527 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2528 {
2529 off_t filesize;
2530 int err;
2531 const int biosize = fuse_iosize(vp);
2532
2533 err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2534 if (err) {
2535 /* This will turn into a SIGBUS */
2536 return (EIO);
2537 } else if ((off_t)lbn * biosize >= filesize) {
2538 *blksz = 0;
2539 } else if ((off_t)(lbn + 1) * biosize > filesize) {
2540 *blksz = filesize - (off_t)lbn *biosize;
2541 } else {
2542 *blksz = biosize;
2543 }
2544 return (0);
2545 }
2546
2547 /*
2548 struct vnop_getpages_args {
2549 struct vnode *a_vp;
2550 vm_page_t *a_m;
2551 int a_count;
2552 int a_reqpage;
2553 };
2554 */
2555 static int
2556 fuse_vnop_getpages(struct vop_getpages_args *ap)
2557 {
2558 struct vnode *vp = ap->a_vp;
2559
2560 if (!fsess_opt_mmap(vnode_mount(vp))) {
2561 SDT_PROBE2(fusefs, , vnops, trace, 1,
2562 "called on non-cacheable vnode??\n");
2563 return (VM_PAGER_ERROR);
2564 }
2565
2566 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2567 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2568 }
2569
2570 static const char extattr_namespace_separator = '.';
2571
2572 /*
2573 struct vop_getextattr_args {
2574 struct vop_generic_args a_gen;
2575 struct vnode *a_vp;
2576 int a_attrnamespace;
2577 const char *a_name;
2578 struct uio *a_uio;
2579 size_t *a_size;
2580 struct ucred *a_cred;
2581 struct thread *a_td;
2582 };
2583 */
2584 static int
2585 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2586 {
2587 struct vnode *vp = ap->a_vp;
2588 struct uio *uio = ap->a_uio;
2589 struct fuse_dispatcher fdi;
2590 struct fuse_getxattr_in *get_xattr_in;
2591 struct fuse_getxattr_out *get_xattr_out;
2592 struct mount *mp = vnode_mount(vp);
2593 struct thread *td = ap->a_td;
2594 struct ucred *cred = ap->a_cred;
2595 char *prefix;
2596 char *attr_str;
2597 size_t len;
2598 int err;
2599
2600 if (fuse_isdeadfs(vp))
2601 return (ENXIO);
2602
2603 if (fsess_not_impl(mp, FUSE_GETXATTR))
2604 return EOPNOTSUPP;
2605
2606 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2607 if (err)
2608 return err;
2609
2610 /* Default to looking for user attributes. */
2611 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2612 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2613 else
2614 prefix = EXTATTR_NAMESPACE_USER_STRING;
2615
2616 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2617 strlen(ap->a_name) + 1;
2618
2619 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2620 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2621
2622 get_xattr_in = fdi.indata;
2623 /*
2624 * Check to see whether we're querying the available size or
2625 * issuing the actual request. If we pass in 0, we get back struct
2626 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2627 * that much data, without the struct fuse_getxattr_out header.
2628 */
2629 if (uio == NULL)
2630 get_xattr_in->size = 0;
2631 else
2632 get_xattr_in->size = uio->uio_resid;
2633
2634 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2635 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2636 ap->a_name);
2637
2638 err = fdisp_wait_answ(&fdi);
2639 if (err != 0) {
2640 if (err == ENOSYS) {
2641 fsess_set_notimpl(mp, FUSE_GETXATTR);
2642 err = EOPNOTSUPP;
2643 }
2644 goto out;
2645 }
2646
2647 get_xattr_out = fdi.answ;
2648
2649 if (ap->a_size != NULL)
2650 *ap->a_size = get_xattr_out->size;
2651
2652 if (uio != NULL)
2653 err = uiomove(fdi.answ, fdi.iosize, uio);
2654
2655 out:
2656 fdisp_destroy(&fdi);
2657 return (err);
2658 }
2659
2660 /*
2661 struct vop_setextattr_args {
2662 struct vop_generic_args a_gen;
2663 struct vnode *a_vp;
2664 int a_attrnamespace;
2665 const char *a_name;
2666 struct uio *a_uio;
2667 struct ucred *a_cred;
2668 struct thread *a_td;
2669 };
2670 */
2671 static int
2672 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2673 {
2674 struct vnode *vp = ap->a_vp;
2675 struct uio *uio = ap->a_uio;
2676 struct fuse_dispatcher fdi;
2677 struct fuse_setxattr_in *set_xattr_in;
2678 struct mount *mp = vnode_mount(vp);
2679 struct thread *td = ap->a_td;
2680 struct ucred *cred = ap->a_cred;
2681 char *prefix;
2682 size_t len;
2683 char *attr_str;
2684 int err;
2685
2686 if (fuse_isdeadfs(vp))
2687 return (ENXIO);
2688
2689 if (fsess_not_impl(mp, FUSE_SETXATTR))
2690 return EOPNOTSUPP;
2691
2692 if (vfs_isrdonly(mp))
2693 return EROFS;
2694
2695 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2696 if (ap->a_uio == NULL) {
2697 /*
2698 * If we got here as fallback from VOP_DELETEEXTATTR, then
2699 * return EOPNOTSUPP.
2700 */
2701 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2702 return (EOPNOTSUPP);
2703 else
2704 return (EINVAL);
2705 }
2706
2707 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2708 VWRITE);
2709 if (err)
2710 return err;
2711
2712 /* Default to looking for user attributes. */
2713 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2714 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2715 else
2716 prefix = EXTATTR_NAMESPACE_USER_STRING;
2717
2718 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2719 strlen(ap->a_name) + 1;
2720
2721 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2722 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2723
2724 set_xattr_in = fdi.indata;
2725 set_xattr_in->size = uio->uio_resid;
2726
2727 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2728 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2729 ap->a_name);
2730
2731 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2732 uio->uio_resid, uio);
2733 if (err != 0) {
2734 goto out;
2735 }
2736
2737 err = fdisp_wait_answ(&fdi);
2738
2739 if (err == ENOSYS) {
2740 fsess_set_notimpl(mp, FUSE_SETXATTR);
2741 err = EOPNOTSUPP;
2742 }
2743 if (err == ERESTART) {
2744 /* Can't restart after calling uiomove */
2745 err = EINTR;
2746 }
2747
2748 out:
2749 fdisp_destroy(&fdi);
2750 return (err);
2751 }
2752
2753 /*
2754 * The Linux / FUSE extended attribute list is simply a collection of
2755 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2756 * byte length followed by a non-NUL terminated string. So, this allows
2757 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2758 * Linux attribute names are reported with the namespace as a prefix (e.g.
2759 * "user.attribute_name"), but in FreeBSD they are reported without the
2760 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2761 *
2762 * user.attr_name1\0user.attr_name2\0
2763 *
2764 * to:
2765 *
2766 * <num>attr_name1<num>attr_name2
2767 *
2768 * Where "<num>" is a single byte number of characters in the attribute name.
2769 *
2770 * Args:
2771 * prefix - exattr namespace prefix string
2772 * list, list_len - input list with namespace prefixes
2773 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2774 */
2775 static int
2776 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2777 char *bsd_list, int *bsd_list_len)
2778 {
2779 int len, pos, dist_to_next, prefix_len;
2780
2781 pos = 0;
2782 *bsd_list_len = 0;
2783 prefix_len = strlen(prefix);
2784
2785 while (pos < list_len && list[pos] != '\0') {
2786 dist_to_next = strlen(&list[pos]) + 1;
2787 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2788 list[pos + prefix_len] == extattr_namespace_separator) {
2789 len = dist_to_next -
2790 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2791 if (len >= EXTATTR_MAXNAMELEN)
2792 return (ENAMETOOLONG);
2793
2794 bsd_list[*bsd_list_len] = len;
2795 memcpy(&bsd_list[*bsd_list_len + 1],
2796 &list[pos + prefix_len +
2797 sizeof(extattr_namespace_separator)], len);
2798
2799 *bsd_list_len += len + 1;
2800 }
2801
2802 pos += dist_to_next;
2803 }
2804
2805 return (0);
2806 }
2807
2808 /*
2809 * List extended attributes
2810 *
2811 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2812 * has a number of differences compared to its FreeBSD equivalent,
2813 * extattr_list_file:
2814 *
2815 * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2816 * whereas listxattr(2) only returns attributes for a single namespace
2817 * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2818 * - If the provided buffer is not large enough to hold the result,
2819 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2820 * return as many results as will fit.
2821 */
2822 /*
2823 struct vop_listextattr_args {
2824 struct vop_generic_args a_gen;
2825 struct vnode *a_vp;
2826 int a_attrnamespace;
2827 struct uio *a_uio;
2828 size_t *a_size;
2829 struct ucred *a_cred;
2830 struct thread *a_td;
2831 };
2832 */
2833 static int
2834 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2835 {
2836 struct vnode *vp = ap->a_vp;
2837 struct uio *uio = ap->a_uio;
2838 struct fuse_dispatcher fdi;
2839 struct fuse_listxattr_in *list_xattr_in;
2840 struct fuse_listxattr_out *list_xattr_out;
2841 struct mount *mp = vnode_mount(vp);
2842 struct thread *td = ap->a_td;
2843 struct ucred *cred = ap->a_cred;
2844 char *prefix;
2845 char *bsd_list = NULL;
2846 char *linux_list;
2847 int bsd_list_len;
2848 int linux_list_len;
2849 int err;
2850
2851 if (fuse_isdeadfs(vp))
2852 return (ENXIO);
2853
2854 if (fsess_not_impl(mp, FUSE_LISTXATTR))
2855 return EOPNOTSUPP;
2856
2857 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2858 if (err)
2859 return err;
2860
2861 /*
2862 * Add space for a NUL and the period separator if enabled.
2863 * Default to looking for user attributes.
2864 */
2865 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2866 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2867 else
2868 prefix = EXTATTR_NAMESPACE_USER_STRING;
2869
2870 fdisp_init(&fdi, sizeof(*list_xattr_in));
2871 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2872
2873 /*
2874 * Retrieve Linux / FUSE compatible list size.
2875 */
2876 list_xattr_in = fdi.indata;
2877 list_xattr_in->size = 0;
2878
2879 err = fdisp_wait_answ(&fdi);
2880 if (err != 0) {
2881 if (err == ENOSYS) {
2882 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2883 err = EOPNOTSUPP;
2884 }
2885 goto out;
2886 }
2887
2888 list_xattr_out = fdi.answ;
2889 linux_list_len = list_xattr_out->size;
2890 if (linux_list_len == 0) {
2891 if (ap->a_size != NULL)
2892 *ap->a_size = linux_list_len;
2893 goto out;
2894 }
2895
2896 /*
2897 * Retrieve Linux / FUSE compatible list values.
2898 */
2899 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2900 list_xattr_in = fdi.indata;
2901 list_xattr_in->size = linux_list_len;
2902
2903 err = fdisp_wait_answ(&fdi);
2904 if (err == ERANGE) {
2905 /*
2906 * Race detected. The attribute list must've grown since the
2907 * first FUSE_LISTXATTR call. Start over. Go all the way back
2908 * to userland so we can process signals, if necessary, before
2909 * restarting.
2910 */
2911 err = ERESTART;
2912 goto out;
2913 } else if (err != 0)
2914 goto out;
2915
2916 linux_list = fdi.answ;
2917 /* FUSE doesn't allow the server to return more data than requested */
2918 if (fdi.iosize > linux_list_len) {
2919 struct fuse_data *data = fuse_get_mpdata(mp);
2920
2921 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2922 "server returned "
2923 "more extended attribute data than requested; "
2924 "should've returned ERANGE instead.");
2925 } else {
2926 /* But returning less data is fine */
2927 linux_list_len = fdi.iosize;
2928 }
2929
2930 /*
2931 * Retrieve the BSD compatible list values.
2932 * The Linux / FUSE attribute list format isn't the same
2933 * as FreeBSD's format. So we need to transform it into
2934 * FreeBSD's format before giving it to the user.
2935 */
2936 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2937 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2938 bsd_list, &bsd_list_len);
2939 if (err != 0)
2940 goto out;
2941
2942 if (ap->a_size != NULL)
2943 *ap->a_size = bsd_list_len;
2944
2945 if (uio != NULL)
2946 err = uiomove(bsd_list, bsd_list_len, uio);
2947
2948 out:
2949 free(bsd_list, M_TEMP);
2950 fdisp_destroy(&fdi);
2951 return (err);
2952 }
2953
2954 /*
2955 struct vop_deallocate_args {
2956 struct vop_generic_args a_gen;
2957 struct vnode *a_vp;
2958 off_t *a_offset;
2959 off_t *a_len;
2960 int a_flags;
2961 int a_ioflag;
2962 struct ucred *a_cred;
2963 };
2964 */
2965 static int
2966 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
2967 {
2968 struct vnode *vp = ap->a_vp;
2969 struct mount *mp = vnode_mount(vp);
2970 struct fuse_filehandle *fufh;
2971 struct fuse_dispatcher fdi;
2972 struct fuse_fallocate_in *ffi;
2973 struct ucred *cred = ap->a_cred;
2974 pid_t pid = curthread->td_proc->p_pid;
2975 off_t *len = ap->a_len;
2976 off_t *offset = ap->a_offset;
2977 int ioflag = ap->a_ioflag;
2978 off_t filesize;
2979 int err;
2980 bool closefufh = false;
2981
2982 if (fuse_isdeadfs(vp))
2983 return (ENXIO);
2984
2985 if (vfs_isrdonly(mp))
2986 return (EROFS);
2987
2988 if (fsess_not_impl(mp, FUSE_FALLOCATE))
2989 goto fallback;
2990
2991 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2992 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2993 /*
2994 * nfsd will do I/O without first doing VOP_OPEN. We
2995 * must implicitly open the file here
2996 */
2997 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2998 closefufh = true;
2999 }
3000 if (err)
3001 return (err);
3002
3003 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3004
3005 err = fuse_vnode_size(vp, &filesize, cred, curthread);
3006 if (err)
3007 goto out;
3008 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3009
3010 fdisp_init(&fdi, sizeof(*ffi));
3011 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3012 ffi = fdi.indata;
3013 ffi->fh = fufh->fh_id;
3014 ffi->offset = *offset;
3015 ffi->length = *len;
3016 /*
3017 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3018 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3019 */
3020 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3021 err = fdisp_wait_answ(&fdi);
3022
3023 if (err == ENOSYS) {
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 goto fallback;
3032 } else if (!err) {
3033 /*
3034 * Clip the returned offset to EoF. Do it here rather than
3035 * before FUSE_FALLOCATE just in case the kernel's cached file
3036 * size is out of date. Unfortunately, FUSE does not return
3037 * any information about filesize from that operation.
3038 */
3039 *offset = MIN(*offset + *len, filesize);
3040 *len = 0;
3041 fuse_vnode_undirty_cached_timestamps(vp, false);
3042 fuse_internal_clear_suid_on_write(vp, cred, curthread);
3043
3044 if (ioflag & IO_SYNC)
3045 err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3046 false);
3047 }
3048
3049 out:
3050 if (closefufh)
3051 fuse_filehandle_close(vp, fufh, curthread, cred);
3052
3053 return (err);
3054
3055 fallback:
3056 if (closefufh)
3057 fuse_filehandle_close(vp, fufh, curthread, cred);
3058
3059 return (vop_stddeallocate(ap));
3060 }
3061
3062 /*
3063 struct vop_deleteextattr_args {
3064 struct vop_generic_args a_gen;
3065 struct vnode *a_vp;
3066 int a_attrnamespace;
3067 const char *a_name;
3068 struct ucred *a_cred;
3069 struct thread *a_td;
3070 };
3071 */
3072 static int
3073 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3074 {
3075 struct vnode *vp = ap->a_vp;
3076 struct fuse_dispatcher fdi;
3077 struct mount *mp = vnode_mount(vp);
3078 struct thread *td = ap->a_td;
3079 struct ucred *cred = ap->a_cred;
3080 char *prefix;
3081 size_t len;
3082 char *attr_str;
3083 int err;
3084
3085 if (fuse_isdeadfs(vp))
3086 return (ENXIO);
3087
3088 if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3089 return EOPNOTSUPP;
3090
3091 if (vfs_isrdonly(mp))
3092 return EROFS;
3093
3094 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3095 VWRITE);
3096 if (err)
3097 return err;
3098
3099 /* Default to looking for user attributes. */
3100 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3101 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3102 else
3103 prefix = EXTATTR_NAMESPACE_USER_STRING;
3104
3105 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3106 strlen(ap->a_name) + 1;
3107
3108 fdisp_init(&fdi, len);
3109 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3110
3111 attr_str = fdi.indata;
3112 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3113 ap->a_name);
3114
3115 err = fdisp_wait_answ(&fdi);
3116 if (err == ENOSYS) {
3117 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3118 err = EOPNOTSUPP;
3119 }
3120
3121 fdisp_destroy(&fdi);
3122 return (err);
3123 }
3124
3125 /*
3126 struct vnop_print_args {
3127 struct vnode *a_vp;
3128 };
3129 */
3130 static int
3131 fuse_vnop_print(struct vop_print_args *ap)
3132 {
3133 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3134
3135 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3136 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3137 (uintmax_t)fvdat->nlookup,
3138 fvdat->flag);
3139
3140 return 0;
3141 }
3142
3143 /*
3144 * Get an NFS filehandle for a FUSE file.
3145 *
3146 * This will only work for FUSE file systems that guarantee the uniqueness of
3147 * nodeid:generation, which most don't.
3148 */
3149 /*
3150 vop_vptofh {
3151 IN struct vnode *a_vp;
3152 IN struct fid *a_fhp;
3153 };
3154 */
3155 static int
3156 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3157 {
3158 struct vnode *vp = ap->a_vp;
3159 struct fuse_vnode_data *fvdat = VTOFUD(vp);
3160 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3161 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3162 "FUSE fid type is too big");
3163 struct mount *mp = vnode_mount(vp);
3164 struct fuse_data *data = fuse_get_mpdata(mp);
3165 struct vattr va;
3166 int err;
3167
3168 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3169 /* NFS requires lookups for "." and ".." */
3170 SDT_PROBE2(fusefs, , vnops, trace, 1,
3171 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3172 return EOPNOTSUPP;
3173 }
3174 if ((mp->mnt_flag & MNT_EXPORTED) &&
3175 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT))
3176 {
3177 /*
3178 * NFS is stateless, so nfsd must reopen a directory on every
3179 * call to VOP_READDIR, passing in the d_off field from the
3180 * final dirent of the previous invocation. But without
3181 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not
3182 * guarantee that d_off will be valid after a directory is
3183 * closed and reopened. So prohibit exporting FUSE file
3184 * systems that don't set that flag.
3185 *
3186 * But userspace NFS servers don't have this problem.
3187 */
3188 SDT_PROBE2(fusefs, , vnops, trace, 1,
3189 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT");
3190 return EOPNOTSUPP;
3191 }
3192
3193 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3194 if (err)
3195 return err;
3196
3197 /*ip = VTOI(ap->a_vp);*/
3198 /*ufhp = (struct ufid *)ap->a_fhp;*/
3199 fhp->len = sizeof(struct fuse_fid);
3200 fhp->nid = fvdat->nid;
3201 if (fvdat->generation <= UINT32_MAX)
3202 fhp->gen = fvdat->generation;
3203 else
3204 return EOVERFLOW;
3205 return (0);
3206 }
Cache object: ae4579af3493381976354dc901c2cb5b
|