xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision d91c459a93e5f70c1d3ad3d504bcf64babed8600)
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 #define EXTERR_CATEGORY EXTERR_CAT_FUSE_VNOPS
93 #include <sys/exterrvar.h>
94 #include <sys/sysent.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_delayed_setsize_t fuse_vnop_delayed_setsize;
138 static vop_deleteextattr_t fuse_vnop_deleteextattr;
139 static vop_fdatasync_t fuse_vnop_fdatasync;
140 static vop_fsync_t fuse_vnop_fsync;
141 static vop_getattr_t fuse_vnop_getattr;
142 static vop_getextattr_t fuse_vnop_getextattr;
143 static vop_inactive_t fuse_vnop_inactive;
144 static vop_ioctl_t fuse_vnop_ioctl;
145 static vop_link_t fuse_vnop_link;
146 static vop_listextattr_t fuse_vnop_listextattr;
147 static vop_lookup_t fuse_vnop_lookup;
148 static vop_mkdir_t fuse_vnop_mkdir;
149 static vop_mknod_t fuse_vnop_mknod;
150 static vop_open_t fuse_vnop_open;
151 static vop_pathconf_t fuse_vnop_pathconf;
152 static vop_read_t fuse_vnop_read;
153 static vop_readdir_t fuse_vnop_readdir;
154 static vop_readlink_t fuse_vnop_readlink;
155 static vop_reclaim_t fuse_vnop_reclaim;
156 static vop_remove_t fuse_vnop_remove;
157 static vop_rename_t fuse_vnop_rename;
158 static vop_rmdir_t fuse_vnop_rmdir;
159 static vop_setattr_t fuse_vnop_setattr;
160 static vop_setextattr_t fuse_vnop_setextattr;
161 static vop_strategy_t fuse_vnop_strategy;
162 static vop_symlink_t fuse_vnop_symlink;
163 static vop_write_t fuse_vnop_write;
164 static vop_getpages_t fuse_vnop_getpages;
165 static vop_print_t fuse_vnop_print;
166 static vop_vptofh_t fuse_vnop_vptofh;
167 
168 struct vop_vector fuse_fifoops = {
169 	.vop_default =		&fifo_specops,
170 	.vop_access =		fuse_vnop_access,
171 	.vop_close =		fuse_fifo_close,
172 	.vop_fsync =		fuse_vnop_fsync,
173 	.vop_getattr =		fuse_vnop_getattr,
174 	.vop_inactive =		fuse_vnop_inactive,
175 	.vop_pathconf =		fuse_vnop_pathconf,
176 	.vop_print =		fuse_vnop_print,
177 	.vop_read =		VOP_PANIC,
178 	.vop_reclaim =		fuse_vnop_reclaim,
179 	.vop_setattr =		fuse_vnop_setattr,
180 	.vop_write =		VOP_PANIC,
181 	.vop_vptofh =		fuse_vnop_vptofh,
182 };
183 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
184 
185 struct vop_vector fuse_vnops = {
186 	.vop_allocate =	fuse_vnop_allocate,
187 	.vop_default = &default_vnodeops,
188 	.vop_access = fuse_vnop_access,
189 	.vop_advlock = fuse_vnop_advlock,
190 	.vop_bmap = fuse_vnop_bmap,
191 	.vop_close = fuse_vnop_close,
192 	.vop_copy_file_range = fuse_vnop_copy_file_range,
193 	.vop_create = fuse_vnop_create,
194 	.vop_deallocate = fuse_vnop_deallocate,
195 	.vop_delayed_setsize = fuse_vnop_delayed_setsize,
196 	.vop_deleteextattr = fuse_vnop_deleteextattr,
197 	.vop_fsync = fuse_vnop_fsync,
198 	.vop_fdatasync = fuse_vnop_fdatasync,
199 	.vop_getattr = fuse_vnop_getattr,
200 	.vop_getextattr = fuse_vnop_getextattr,
201 	.vop_inactive = fuse_vnop_inactive,
202 	.vop_ioctl = fuse_vnop_ioctl,
203 	.vop_link = fuse_vnop_link,
204 	.vop_listextattr = fuse_vnop_listextattr,
205 	.vop_lookup = fuse_vnop_lookup,
206 	.vop_mkdir = fuse_vnop_mkdir,
207 	.vop_mknod = fuse_vnop_mknod,
208 	.vop_open = fuse_vnop_open,
209 	.vop_pathconf = fuse_vnop_pathconf,
210 	/*
211 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
212 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
213 	 * 7.21, which adds the ability for the client to choose which poll
214 	 * events it wants, and for a client to deregister a file handle
215 	 */
216 	.vop_read = fuse_vnop_read,
217 	.vop_readdir = fuse_vnop_readdir,
218 	.vop_readlink = fuse_vnop_readlink,
219 	.vop_reclaim = fuse_vnop_reclaim,
220 	.vop_remove = fuse_vnop_remove,
221 	.vop_rename = fuse_vnop_rename,
222 	.vop_rmdir = fuse_vnop_rmdir,
223 	.vop_setattr = fuse_vnop_setattr,
224 	.vop_setextattr = fuse_vnop_setextattr,
225 	.vop_strategy = fuse_vnop_strategy,
226 	.vop_symlink = fuse_vnop_symlink,
227 	.vop_write = fuse_vnop_write,
228 	.vop_getpages = fuse_vnop_getpages,
229 	.vop_print = fuse_vnop_print,
230 	.vop_vptofh = fuse_vnop_vptofh,
231 };
232 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
233 
234 /* Check permission for extattr operations, much like extattr_check_cred */
235 static int
fuse_extattr_check_cred(struct vnode * vp,int ns,struct ucred * cred,struct thread * td,accmode_t accmode)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
fuse_filehandle_get_dir(struct vnode * vp,struct fuse_filehandle ** fufhp,struct ucred * cred,pid_t pid)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
fuse_flush(struct vnode * vp,struct ucred * cred,pid_t pid,int fflag)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(mp, FUSE_FLUSH))
291 		return 0;
292 
293 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
294 	if (err)
295 		return err;
296 
297 	if (fufh->fuse_open_flags & FOPEN_NOFLUSH &&
298 	    (!fsess_opt_writeback(mp)))
299 		return (0);
300 
301 	fdisp_init(&fdi, sizeof(*ffi));
302 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
303 	ffi = fdi.indata;
304 	ffi->fh = fufh->fh_id;
305 	/*
306 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
307 	 * If not, then lock_owner is undefined.  So we may as well always set
308 	 * it.
309 	 */
310 	ffi->lock_owner = td->td_proc->p_pid;
311 
312 	err = fdisp_wait_answ(&fdi);
313 	if (err == ENOSYS) {
314 		fsess_set_notimpl(mp, FUSE_FLUSH);
315 		err = 0;
316 	}
317 	fdisp_destroy(&fdi);
318 	return err;
319 }
320 
321 /* Close wrapper for fifos.  */
322 static int
fuse_fifo_close(struct vop_close_args * ap)323 fuse_fifo_close(struct vop_close_args *ap)
324 {
325 	return (fifo_specops.vop_close(ap));
326 }
327 
328 /* Invalidate a range of cached data, whether dirty of not */
329 static int
fuse_inval_buf_range(struct vnode * vp,off_t filesize,off_t start,off_t end,int slpflag)330 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end,
331 	int slpflag)
332 {
333 	struct buf *bp;
334 	daddr_t left_lbn, end_lbn, right_lbn;
335 	off_t new_filesize;
336 	int iosize, left_on, right_on, right_blksize;
337 
338 	iosize = fuse_iosize(vp);
339 	left_lbn = start / iosize;
340 	end_lbn = howmany(end, iosize);
341 	left_on = start & (iosize - 1);
342 	if (left_on != 0) {
343 		bp = getblk(vp, left_lbn, iosize, slpflag, 0, 0);
344 		if (!bp)
345 			return (EINTR);
346 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
347 			/*
348 			 * Flush the dirty buffer, because we don't have a
349 			 * byte-granular way to record which parts of the
350 			 * buffer are valid.
351 			 */
352 			bwrite(bp);
353 			if (bp->b_error)
354 				return (bp->b_error);
355 		} else {
356 			brelse(bp);
357 		}
358 	}
359 	right_on = end & (iosize - 1);
360 	if (right_on != 0) {
361 		right_lbn = end / iosize;
362 		new_filesize = MAX(filesize, end);
363 		right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
364 		bp = getblk(vp, right_lbn, right_blksize, slpflag, 0, 0);
365 		if (!bp)
366 			return (EINTR);
367 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
368 			/*
369 			 * Flush the dirty buffer, because we don't have a
370 			 * byte-granular way to record which parts of the
371 			 * buffer are valid.
372 			 */
373 			bwrite(bp);
374 			if (bp->b_error)
375 				return (bp->b_error);
376 		} else {
377 			brelse(bp);
378 		}
379 	}
380 
381 	v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
382 	return (0);
383 }
384 
385 /* Send FUSE_IOCTL for this node */
386 static int
fuse_vnop_do_ioctl(struct vnode * vp,u_long cmd,void * arg,int fflag,struct ucred * cred,struct thread * td)387 fuse_vnop_do_ioctl(struct vnode *vp, u_long cmd, void *arg, int fflag,
388 	struct ucred *cred, struct thread *td)
389 {
390 	struct fuse_dispatcher fdi;
391 	struct fuse_ioctl_in *fii;
392 	struct fuse_ioctl_out *fio;
393 	struct fuse_filehandle *fufh;
394 	uint32_t flags = 0;
395 	uint32_t insize = 0;
396 	uint32_t outsize = 0;
397 	int err;
398 
399 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, td->td_proc->p_pid);
400 	if (err != 0)
401 		return (err);
402 
403 	if (vnode_isdir(vp)) {
404 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
405 
406 		if (!fuse_libabi_geq(data, 7, 18))
407 			return (ENOTTY);
408 		flags |= FUSE_IOCTL_DIR;
409 	}
410 #ifdef __LP64__
411 #ifdef COMPAT_FREEBSD32
412 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32))
413 		flags |= FUSE_IOCTL_32BIT;
414 #endif
415 #else /* !defined(__LP64__) */
416 	flags |= FUSE_IOCTL_32BIT;
417 #endif
418 
419 	if ((cmd & IOC_OUT) != 0)
420 		outsize = IOCPARM_LEN(cmd);
421 	/* _IOWINT() sets IOC_VOID */
422 	if ((cmd & (IOC_VOID | IOC_IN)) != 0)
423 		insize = IOCPARM_LEN(cmd);
424 
425 	fdisp_init(&fdi, sizeof(*fii) + insize);
426 	fdisp_make_vp(&fdi, FUSE_IOCTL, vp, td, cred);
427 	fii = fdi.indata;
428 	fii->fh = fufh->fh_id;
429 	fii->flags = flags;
430 	fii->cmd = cmd;
431 	fii->arg = (uintptr_t)arg;
432 	fii->in_size = insize;
433 	fii->out_size = outsize;
434 	if (insize > 0)
435 		memcpy((char *)fii + sizeof(*fii), arg, insize);
436 
437 	err = fdisp_wait_answ(&fdi);
438 	if (err != 0) {
439 		if (err == ENOSYS)
440 			err = ENOTTY;
441 		goto out;
442 	}
443 
444 	fio = fdi.answ;
445 	if (fdi.iosize > sizeof(*fio)) {
446 		size_t realoutsize = fdi.iosize - sizeof(*fio);
447 
448 		if (realoutsize > outsize) {
449 			err = EIO;
450 			goto out;
451 		}
452 		memcpy(arg, (char *)fio + sizeof(*fio), realoutsize);
453 	}
454 	if (fio->result > 0)
455 		td->td_retval[0] = fio->result;
456 	else
457 		err = -fio->result;
458 
459 out:
460 	fdisp_destroy(&fdi);
461 	return (err);
462 }
463 
464 /* Send FUSE_LSEEK for this node */
465 static int
fuse_vnop_do_lseek(struct vnode * vp,struct thread * td,struct ucred * cred,pid_t pid,off_t * offp,int whence)466 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
467 	pid_t pid, off_t *offp, int whence)
468 {
469 	struct fuse_dispatcher fdi;
470 	struct fuse_filehandle *fufh;
471 	struct fuse_lseek_in *flsi;
472 	struct fuse_lseek_out *flso;
473 	struct mount *mp = vnode_mount(vp);
474 	int err;
475 
476 	ASSERT_VOP_LOCKED(vp, __func__);
477 
478 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
479 	if (err)
480 		return (err);
481 	fdisp_init(&fdi, sizeof(*flsi));
482 	fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
483 	flsi = fdi.indata;
484 	flsi->fh = fufh->fh_id;
485 	flsi->offset = *offp;
486 	flsi->whence = whence;
487 	err = fdisp_wait_answ(&fdi);
488 	if (err == ENOSYS) {
489 		fsess_set_notimpl(mp, FUSE_LSEEK);
490 	} else if (err == ENXIO) {
491 		/* Note: ENXIO means "no more hole/data regions until EOF" */
492 		fsess_set_impl(mp, FUSE_LSEEK);
493 	} else if (err == 0) {
494 		fsess_set_impl(mp, FUSE_LSEEK);
495 		flso = fdi.answ;
496 		*offp = flso->offset;
497 	}
498 	fdisp_destroy(&fdi);
499 
500 	return (err);
501 }
502 
503 /*
504     struct vnop_access_args {
505 	struct vnode *a_vp;
506 #if VOP_ACCESS_TAKES_ACCMODE_T
507 	accmode_t a_accmode;
508 #else
509 	int a_mode;
510 #endif
511 	struct ucred *a_cred;
512 	struct thread *a_td;
513     };
514 */
515 static int
fuse_vnop_access(struct vop_access_args * ap)516 fuse_vnop_access(struct vop_access_args *ap)
517 {
518 	struct vnode *vp = ap->a_vp;
519 	int accmode = ap->a_accmode;
520 	struct ucred *cred = ap->a_cred;
521 
522 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
523 
524 	int err;
525 
526 	if (fuse_isdeadfs(vp)) {
527 		if (vnode_isvroot(vp)) {
528 			return 0;
529 		}
530 		return (EXTERROR(ENXIO, "This FUSE session is about "
531 		    "to be closed"));
532 	}
533 	if (!(data->dataflags & FSESS_INITED)) {
534 		if (vnode_isvroot(vp)) {
535 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
536 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
537 				return 0;
538 			}
539 		}
540 		return (EXTERROR(EBADF, "Access denied until FUSE session "
541 		    "is initialized"));
542 	}
543 	if (vnode_islnk(vp)) {
544 		return 0;
545 	}
546 
547 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
548 	return err;
549 }
550 
551 /*
552  * struct vop_advlock_args {
553  *	struct vop_generic_args a_gen;
554  *	struct vnode *a_vp;
555  *	void *a_id;
556  *	int a_op;
557  *	struct flock *a_fl;
558  *	int a_flags;
559  * }
560  */
561 static int
fuse_vnop_advlock(struct vop_advlock_args * ap)562 fuse_vnop_advlock(struct vop_advlock_args *ap)
563 {
564 	struct vnode *vp = ap->a_vp;
565 	struct flock *fl = ap->a_fl;
566 	struct thread *td = curthread;
567 	struct ucred *cred = td->td_ucred;
568 	pid_t pid = td->td_proc->p_pid;
569 	struct fuse_filehandle *fufh;
570 	struct fuse_dispatcher fdi;
571 	struct fuse_lk_in *fli;
572 	struct fuse_lk_out *flo;
573 	struct vattr vattr;
574 	enum fuse_opcode op;
575 	off_t size, start;
576 	int dataflags, err;
577 	int flags = ap->a_flags;
578 
579 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
580 
581 	if (fuse_isdeadfs(vp)) {
582 		return (EXTERROR(ENXIO, "This FUSE session is about "
583 		    "to be closed"));
584 	}
585 
586 	switch(ap->a_op) {
587 	case F_GETLK:
588 		op = FUSE_GETLK;
589 		break;
590 	case F_SETLK:
591 		if (flags & F_WAIT)
592 			op = FUSE_SETLKW;
593 		else
594 			op = FUSE_SETLK;
595 		break;
596 	case F_UNLCK:
597 		op = FUSE_SETLK;
598 		break;
599 	default:
600 		return (EXTERROR(EINVAL, "Unsupported lock flags"));
601 	}
602 
603 	if (!(dataflags & FSESS_POSIX_LOCKS))
604 		return vop_stdadvlock(ap);
605 	/* FUSE doesn't properly support flock until protocol 7.17 */
606 	if (flags & F_FLOCK)
607 		return vop_stdadvlock(ap);
608 
609 	vn_lock(vp, LK_SHARED | LK_RETRY);
610 
611 	switch (fl->l_whence) {
612 	case SEEK_SET:
613 	case SEEK_CUR:
614 		/*
615 		 * Caller is responsible for adding any necessary offset
616 		 * when SEEK_CUR is used.
617 		 */
618 		start = fl->l_start;
619 		break;
620 
621 	case SEEK_END:
622 		err = fuse_internal_getattr(vp, &vattr, cred, td);
623 		if (err)
624 			goto out;
625 		size = vattr.va_size;
626 		if (size > OFF_MAX ||
627 		    (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
628 			err = EXTERROR(EOVERFLOW, "Offset is too large");
629 			goto out;
630 		}
631 		start = size + fl->l_start;
632 		break;
633 
634 	default:
635 		return (EXTERROR(EINVAL, "Unsupported offset type"));
636 	}
637 
638 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
639 	if (err)
640 		goto out;
641 
642 	fdisp_init(&fdi, sizeof(*fli));
643 
644 	fdisp_make_vp(&fdi, op, vp, td, cred);
645 	fli = fdi.indata;
646 	fli->fh = fufh->fh_id;
647 	fli->owner = td->td_proc->p_pid;
648 	fli->lk.start = start;
649 	if (fl->l_len != 0)
650 		fli->lk.end = start + fl->l_len - 1;
651 	else
652 		fli->lk.end = INT64_MAX;
653 	fli->lk.type = fl->l_type;
654 	fli->lk.pid = td->td_proc->p_pid;
655 
656 	err = fdisp_wait_answ(&fdi);
657 	fdisp_destroy(&fdi);
658 
659 	if (err == 0 && op == FUSE_GETLK) {
660 		flo = fdi.answ;
661 		fl->l_type = flo->lk.type;
662 		fl->l_whence = SEEK_SET;
663 		if (flo->lk.type != F_UNLCK) {
664 			fl->l_pid = flo->lk.pid;
665 			fl->l_start = flo->lk.start;
666 			if (flo->lk.end == INT64_MAX)
667 				fl->l_len = 0;
668 			else
669 				fl->l_len = flo->lk.end - flo->lk.start + 1;
670 			fl->l_start = flo->lk.start;
671 		}
672 	}
673 
674 out:
675 	VOP_UNLOCK(vp);
676 	return err;
677 }
678 
679 static int
fuse_vnop_allocate(struct vop_allocate_args * ap)680 fuse_vnop_allocate(struct vop_allocate_args *ap)
681 {
682 	struct vnode *vp = ap->a_vp;
683 	off_t *len = ap->a_len;
684 	off_t *offset = ap->a_offset;
685 	struct ucred *cred = ap->a_cred;
686 	struct fuse_filehandle *fufh;
687 	struct mount *mp = vnode_mount(vp);
688 	struct fuse_dispatcher fdi;
689 	struct fuse_fallocate_in *ffi;
690 	struct uio io;
691 	pid_t pid = curthread->td_proc->p_pid;
692 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
693 	off_t filesize;
694 	int err;
695 
696 	if (fuse_isdeadfs(vp))
697 		return (EXTERROR(ENXIO, "This FUSE session is about "
698 		    "to be closed"));
699 
700 	switch (vp->v_type) {
701 	case VFIFO:
702 		return (ESPIPE);
703 	case VLNK:
704 	case VREG:
705 		break;
706 	default:
707 		return (ENODEV);
708 	}
709 
710 	if (vfs_isrdonly(mp))
711 		return (EROFS);
712 
713 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
714 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
715 		    "FUSE_FALLOCATE"));
716 
717 	ASSERT_CACHED_ATTRS_LOCKED(vp);
718 
719 	io.uio_offset = *offset;
720 	io.uio_resid = *len;
721 	err = vn_rlimit_fsize(vp, &io, curthread);
722 	if (err)
723 		return (err);
724 
725 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
726 	if (err)
727 		return (err);
728 
729 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
730 
731 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
732 	if (err)
733 		return (err);
734 	err = fuse_inval_buf_range(vp, filesize, *offset, *offset + *len,
735 	    PCATCH);
736 	if (err)
737 		return (err);
738 
739 	fdisp_init(&fdi, sizeof(*ffi));
740 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
741 	ffi = fdi.indata;
742 	ffi->fh = fufh->fh_id;
743 	ffi->offset = *offset;
744 	ffi->length = *len;
745 	ffi->mode = 0;
746 	err = fdisp_wait_answ(&fdi);
747 
748 	if (err == ENOSYS) {
749 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
750 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
751 		    "FUSE_ALLOCATE");
752 	} else if (err == EOPNOTSUPP) {
753 		/*
754 		 * The file system server does not support FUSE_FALLOCATE with
755 		 * the supplied mode for this particular file.
756 		 */
757 		err = EXTERROR(EOPNOTSUPP, "This file can't be pre-allocated");
758 	} else if (!err) {
759 		*offset += *len;
760 		*len = 0;
761 		fuse_vnode_undirty_cached_timestamps(vp, false);
762 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
763 		if (*offset > fvdat->cached_attrs.va_size) {
764 			fuse_vnode_setsize(vp, *offset, false);
765 			getnanouptime(&fvdat->last_local_modify);
766 		}
767 	}
768 
769 	fdisp_destroy(&fdi);
770 	return (err);
771 }
772 
773 /* {
774 	struct vnode *a_vp;
775 	daddr_t a_bn;
776 	struct bufobj **a_bop;
777 	daddr_t *a_bnp;
778 	int *a_runp;
779 	int *a_runb;
780 } */
781 static int
fuse_vnop_bmap(struct vop_bmap_args * ap)782 fuse_vnop_bmap(struct vop_bmap_args *ap)
783 {
784 	struct vnode *vp = ap->a_vp;
785 	struct bufobj **bo = ap->a_bop;
786 	struct thread *td = curthread;
787 	struct mount *mp;
788 	struct fuse_dispatcher fdi;
789 	struct fuse_bmap_in *fbi;
790 	struct fuse_bmap_out *fbo;
791 	struct fuse_data *data;
792 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
793 	uint64_t biosize;
794 	off_t fsize = VNOVAL;
795 	daddr_t lbn = ap->a_bn;
796 	daddr_t *pbn = ap->a_bnp;
797 	int *runp = ap->a_runp;
798 	int *runb = ap->a_runb;
799 	int error = 0;
800 	int maxrun;
801 
802 	if (fuse_isdeadfs(vp)) {
803 		return (EXTERROR(ENXIO, "This FUSE session is about "
804 		    "to be closed"));
805 	}
806 
807 	mp = vnode_mount(vp);
808 	data = fuse_get_mpdata(mp);
809 	biosize = fuse_iosize(vp);
810 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
811 		data->max_readahead_blocks);
812 
813 	if (bo != NULL)
814 		*bo = &vp->v_bufobj;
815 
816 	/*
817 	 * The FUSE_BMAP operation does not include the runp and runb
818 	 * variables, so we must guess.  Report nonzero contiguous runs so
819 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
820 	 * upcalls even if we don't know the true physical layout of the file.
821 	 *
822 	 * FUSE file systems may opt out of read clustering in two ways:
823 	 * * mounting with -onoclusterr
824 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
825 	 */
826 	if (runb != NULL)
827 		*runb = MIN(lbn, maxrun);
828 	if (runp != NULL && maxrun == 0)
829 		*runp = 0;
830 	else if (runp != NULL) {
831 		/*
832 		 * If the file's size is cached, use that value to calculate
833 		 * runp, even if the cache is expired.  runp is only advisory,
834 		 * and the risk of getting it wrong is not worth the cost of
835 		 * another upcall.
836 		 */
837 		CACHED_ATTR_LOCK(vp);
838 		fsize = fvdat->cached_attrs.va_size;
839 		CACHED_ATTR_UNLOCK(vp);
840 		if (fsize == VNOVAL)
841 			error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
842 		if (error == 0)
843 			*runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
844 				    maxrun);
845 		else
846 			*runp = 0;
847 	}
848 
849 	if (fsess_maybe_impl(mp, FUSE_BMAP)) {
850 		fdisp_init(&fdi, sizeof(*fbi));
851 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
852 		fbi = fdi.indata;
853 		fbi->block = lbn;
854 		fbi->blocksize = biosize;
855 		error = fdisp_wait_answ(&fdi);
856 		if (error == ENOSYS) {
857 			fdisp_destroy(&fdi);
858 			fsess_set_notimpl(mp, FUSE_BMAP);
859 			error = 0;
860 		} else {
861 			fbo = fdi.answ;
862 			if (error == 0 && pbn != NULL)
863 				*pbn = fbo->block;
864 			fdisp_destroy(&fdi);
865 			return error;
866 		}
867 	}
868 
869 	/* If the daemon doesn't support BMAP, make up a sensible default */
870 	if (pbn != NULL)
871 		*pbn = lbn * btodb(biosize);
872 	return (error);
873 }
874 
875 /*
876     struct vop_close_args {
877 	struct vnode *a_vp;
878 	int  a_fflag;
879 	struct ucred *a_cred;
880 	struct thread *a_td;
881     };
882 */
883 static int
fuse_vnop_close(struct vop_close_args * ap)884 fuse_vnop_close(struct vop_close_args *ap)
885 {
886 	struct vnode *vp = ap->a_vp;
887 	struct mount *mp = vnode_mount(vp);
888 	struct ucred *cred = ap->a_cred;
889 	int fflag = ap->a_fflag;
890 	struct thread *td;
891 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
892 	pid_t pid;
893 	int err = 0;
894 
895 	/* NB: a_td will be NULL from some async kernel contexts */
896 	td = ap->a_td ? ap->a_td : curthread;
897 	pid = td->td_proc->p_pid;
898 
899 	if (fuse_isdeadfs(vp))
900 		return 0;
901 	if (vnode_isdir(vp))
902 		return 0;
903 	if (fflag & IO_NDELAY)
904 		return 0;
905 
906 	if (cred == NULL)
907 		cred = td->td_ucred;
908 
909 	err = fuse_flush(vp, cred, pid, fflag);
910 	ASSERT_CACHED_ATTRS_LOCKED(vp);	/* For fvdat->flag */
911 	if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) {
912 		struct vattr vap;
913 		struct fuse_data *data;
914 		int dataflags;
915 		int access_e = 0;
916 
917 		data = fuse_get_mpdata(mp);
918 		dataflags = data->dataflags;
919 		if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
920 			struct vattr va;
921 
922 			fuse_internal_getattr(vp, &va, cred, td);
923 			access_e = vaccess(vp->v_type, va.va_mode, va.va_uid,
924 			    va.va_gid, VWRITE, cred);
925 		}
926 		if (access_e == 0) {
927 			VATTR_NULL(&vap);
928 			ASSERT_CACHED_ATTRS_LOCKED(vp);
929 			vap.va_atime = fvdat->cached_attrs.va_atime;
930 			/*
931 			 * Ignore errors setting when setting atime.  That
932 			 * should not cause close(2) to fail.
933 			 */
934 			fuse_internal_setattr(vp, &vap, td, NULL);
935 		}
936 	}
937 	/* TODO: close the file handle, if we're sure it's no longer used */
938 	if ((fvdat->flag & FN_SIZECHANGE) != 0) {
939 		fuse_vnode_savesize(vp, cred, pid);
940 	}
941 	return err;
942 }
943 
944 /*
945    struct vop_copy_file_range_args {
946 	struct vop_generic_args a_gen;
947 	struct vnode *a_invp;
948 	off_t *a_inoffp;
949 	struct vnode *a_outvp;
950 	off_t *a_outoffp;
951 	size_t *a_lenp;
952 	unsigned int a_flags;
953 	struct ucred *a_incred;
954 	struct ucred *a_outcred;
955 	struct thread *a_fsizetd;
956 }
957  */
958 static int
fuse_vnop_copy_file_range(struct vop_copy_file_range_args * ap)959 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
960 {
961 	struct vnode *invp = ap->a_invp;
962 	struct vnode *outvp = ap->a_outvp;
963 	struct mount *mp = vnode_mount(invp);
964 	struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
965 	struct fuse_dispatcher fdi;
966 	struct fuse_filehandle *infufh, *outfufh;
967 	struct fuse_copy_file_range_in *fcfri;
968 	struct ucred *incred = ap->a_incred;
969 	struct ucred *outcred = ap->a_outcred;
970 	struct fuse_write_out *fwo;
971 	struct thread *td;
972 	struct uio io;
973 	off_t outfilesize;
974 	ssize_t r = 0;
975 	pid_t pid;
976 	int err;
977 
978 	if ((ap->a_flags & COPY_FILE_RANGE_CLONE) != 0)
979 		return (EXTERROR(ENOSYS, "Cannot clone"));
980 
981 	if (mp == NULL || mp != vnode_mount(outvp))
982 		return (EXTERROR(ENOSYS, "Mount points do not match"));
983 
984 	if (incred->cr_uid != outcred->cr_uid)
985 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
986 		    "support different credentials for infd and outfd"));
987 
988 	if (incred->cr_gid != outcred->cr_gid)
989 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
990 		    "support different credentials for infd and outfd"));
991 
992 	/* Caller busied mp, mnt_data can be safely accessed. */
993 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
994 		return (EXTERROR(ENOSYS, "This daemon does not "
995 		    "implement COPY_FILE_RANGE"));
996 
997 	if (ap->a_fsizetd == NULL)
998 		td = curthread;
999 	else
1000 		td = ap->a_fsizetd;
1001 	pid = td->td_proc->p_pid;
1002 
1003 	vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE);
1004 	if (invp->v_data == NULL || outvp->v_data == NULL) {
1005 		err = EXTERROR(EBADF, "vnode got reclaimed");
1006 		goto unlock;
1007 	}
1008 
1009 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
1010 	if (err)
1011 		goto unlock;
1012 
1013 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
1014 	if (err)
1015 		goto unlock;
1016 
1017 	io.uio_resid = *ap->a_lenp;
1018 	if (ap->a_fsizetd) {
1019 		io.uio_offset = *ap->a_outoffp;
1020 		err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
1021 		if (err != 0)
1022 			goto unlock;
1023 	}
1024 
1025 	err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
1026 	if (err)
1027 		goto unlock;
1028 
1029 	vnode_pager_clean_sync(invp);
1030 	err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
1031 		*ap->a_outoffp + io.uio_resid, PCATCH);
1032 	if (err)
1033 		goto unlock;
1034 
1035 	fdisp_init(&fdi, sizeof(*fcfri));
1036 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
1037 	fcfri = fdi.indata;
1038 	fcfri->fh_in = infufh->fh_id;
1039 	fcfri->off_in = *ap->a_inoffp;
1040 	fcfri->nodeid_out = VTOI(outvp);
1041 	fcfri->fh_out = outfufh->fh_id;
1042 	fcfri->off_out = *ap->a_outoffp;
1043 	fcfri->len = io.uio_resid;
1044 	fcfri->flags = 0;
1045 
1046 	err = fdisp_wait_answ(&fdi);
1047 	if (err == 0) {
1048 		fwo = fdi.answ;
1049 		*ap->a_lenp = fwo->size;
1050 		*ap->a_inoffp += fwo->size;
1051 		*ap->a_outoffp += fwo->size;
1052 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
1053 		ASSERT_CACHED_ATTRS_LOCKED(outvp);
1054 		if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
1055 			fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
1056 			getnanouptime(&outfvdat->last_local_modify);
1057 		}
1058 		fuse_vnode_update(invp, FN_ATIMECHANGE);
1059 		fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
1060 	}
1061 	fdisp_destroy(&fdi);
1062 
1063 unlock:
1064 	if (invp != outvp)
1065 		VOP_UNLOCK(invp);
1066 	VOP_UNLOCK(outvp);
1067 
1068 	if (err == ENOSYS)
1069 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
1070 
1071 	/*
1072 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
1073 	 * local.
1074 	 */
1075 	return (err);
1076 }
1077 
1078 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)1079 fdisp_make_mknod_for_fallback(
1080 	struct fuse_dispatcher *fdip,
1081 	struct componentname *cnp,
1082 	struct vnode *dvp,
1083 	uint64_t parentnid,
1084 	struct thread *td,
1085 	struct ucred *cred,
1086 	mode_t mode,
1087 	enum fuse_opcode *op)
1088 {
1089 	struct fuse_mknod_in *fmni;
1090 
1091 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
1092 	*op = FUSE_MKNOD;
1093 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
1094 	fmni = fdip->indata;
1095 	fmni->mode = mode;
1096 	fmni->rdev = 0;
1097 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
1098 	    cnp->cn_namelen);
1099 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
1100 }
1101 /*
1102     struct vnop_create_args {
1103 	struct vnode *a_dvp;
1104 	struct vnode **a_vpp;
1105 	struct componentname *a_cnp;
1106 	struct vattr *a_vap;
1107     };
1108 */
1109 static int
fuse_vnop_create(struct vop_create_args * ap)1110 fuse_vnop_create(struct vop_create_args *ap)
1111 {
1112 	struct vnode *dvp = ap->a_dvp;
1113 	struct vnode **vpp = ap->a_vpp;
1114 	struct componentname *cnp = ap->a_cnp;
1115 	struct vattr *vap = ap->a_vap;
1116 	struct thread *td = curthread;
1117 	struct ucred *cred = cnp->cn_cred;
1118 
1119 	struct fuse_data *data;
1120 	struct fuse_create_in *fci;
1121 	struct fuse_entry_out *feo;
1122 	struct fuse_open_out *foo;
1123 	struct fuse_dispatcher fdi, fdi2;
1124 	struct fuse_dispatcher *fdip = &fdi;
1125 	struct fuse_dispatcher *fdip2 = NULL;
1126 
1127 	int err;
1128 
1129 	struct mount *mp = vnode_mount(dvp);
1130 	data = fuse_get_mpdata(mp);
1131 	uint64_t parentnid = VTOFUD(dvp)->nid;
1132 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1133 	enum fuse_opcode op;
1134 	int flags;
1135 
1136 	if (fuse_isdeadfs(dvp))
1137 		return (EXTERROR(ENXIO, "This FUSE session is about "
1138 		    "to be closed"));
1139 
1140 	/* FUSE expects sockets to be created with FUSE_MKNOD */
1141 	if (vap->va_type == VSOCK)
1142 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
1143 
1144 	/*
1145 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
1146 	 * writable mode makes sense, and we might as well include readability
1147 	 * too.
1148 	 */
1149 	flags = O_RDWR;
1150 
1151 	bzero(&fdi, sizeof(fdi));
1152 
1153 	if (vap->va_type != VREG)
1154 		return (EXTERROR(EINVAL, "Only regular files can be created"));
1155 
1156 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1157 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
1158 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1159 			cred, mode, &op);
1160 	} else {
1161 		/* Use FUSE_CREATE */
1162 		size_t insize;
1163 
1164 		op = FUSE_CREATE;
1165 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1166 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1167 		fci = fdip->indata;
1168 		fci->mode = mode;
1169 		fci->flags = O_CREAT | flags;
1170 		if (fuse_libabi_geq(data, 7, 12)) {
1171 			insize = sizeof(*fci);
1172 			fci->umask = td->td_proc->p_pd->pd_cmask;
1173 		} else {
1174 			insize = sizeof(struct fuse_open_in);
1175 		}
1176 
1177 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1178 		    cnp->cn_namelen);
1179 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1180 	}
1181 
1182 	err = fdisp_wait_answ(fdip);
1183 
1184 	if (err) {
1185 		if (err == ENOSYS && op == FUSE_CREATE) {
1186 			fsess_set_notimpl(mp, FUSE_CREATE);
1187 			fdisp_destroy(fdip);
1188 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1189 				parentnid, td, cred, mode, &op);
1190 			err = fdisp_wait_answ(fdip);
1191 		}
1192 		if (err)
1193 			goto out;
1194 	}
1195 
1196 	feo = fdip->answ;
1197 
1198 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1199 		goto out;
1200 	}
1201 
1202 	if (op == FUSE_CREATE) {
1203 		if (fuse_libabi_geq(data, 7, 9))
1204 			foo = (struct fuse_open_out*)(feo + 1);
1205 		else
1206 			foo = (struct fuse_open_out*)((char*)feo +
1207 				FUSE_COMPAT_ENTRY_OUT_SIZE);
1208 	} else {
1209 		/* Issue a separate FUSE_OPEN */
1210 		struct fuse_open_in *foi;
1211 
1212 		fdip2 = &fdi2;
1213 		fdisp_init(fdip2, sizeof(*foi));
1214 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1215 			cred);
1216 		foi = fdip2->indata;
1217 		foi->flags = flags;
1218 		err = fdisp_wait_answ(fdip2);
1219 		if (err)
1220 			goto out;
1221 		foo = fdip2->answ;
1222 	}
1223 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1224 	if (err) {
1225 		struct fuse_release_in *fri;
1226 		uint64_t nodeid = feo->nodeid;
1227 		uint64_t fh_id = foo->fh;
1228 
1229 		fdisp_destroy(fdip);
1230 		fdisp_init(fdip, sizeof(*fri));
1231 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1232 		fri = fdip->indata;
1233 		fri->fh = fh_id;
1234 		fri->flags = flags;
1235 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1236 		fuse_insert_message(fdip->tick, false);
1237 		goto out;
1238 	}
1239 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1240 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1241 		feo->attr_valid_nsec, NULL, true);
1242 
1243 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1244 	fuse_vnode_open(*vpp, foo->open_flags, td);
1245 	/*
1246 	 * Purge the parent's attribute cache because the daemon should've
1247 	 * updated its mtime and ctime
1248 	 */
1249 	fuse_vnode_clear_attr_cache(dvp);
1250 	cache_purge_negative(dvp);
1251 
1252 out:
1253 	if (fdip2)
1254 		fdisp_destroy(fdip2);
1255 	fdisp_destroy(fdip);
1256 	return err;
1257 }
1258 
1259 /*
1260     struct vnop_fdatasync_args {
1261 	struct vop_generic_args a_gen;
1262 	struct vnode * a_vp;
1263 	struct thread * a_td;
1264     };
1265 */
1266 static int
fuse_vnop_fdatasync(struct vop_fdatasync_args * ap)1267 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1268 {
1269 	struct vnode *vp = ap->a_vp;
1270 	struct thread *td = ap->a_td;
1271 	int waitfor = MNT_WAIT;
1272 
1273 	int err = 0;
1274 
1275 	if (fuse_isdeadfs(vp)) {
1276 		return 0;
1277 	}
1278 	if ((err = vop_stdfdatasync_buf(ap)))
1279 		return err;
1280 
1281 	return fuse_internal_fsync(vp, td, waitfor, true);
1282 }
1283 
1284 /*
1285     struct vnop_fsync_args {
1286 	struct vop_generic_args a_gen;
1287 	struct vnode * a_vp;
1288 	int  a_waitfor;
1289 	struct thread * a_td;
1290     };
1291 */
1292 static int
fuse_vnop_fsync(struct vop_fsync_args * ap)1293 fuse_vnop_fsync(struct vop_fsync_args *ap)
1294 {
1295 	struct vnode *vp = ap->a_vp;
1296 	struct thread *td = ap->a_td;
1297 	int waitfor = ap->a_waitfor;
1298 	int err = 0;
1299 
1300 	if (fuse_isdeadfs(vp)) {
1301 		return 0;
1302 	}
1303 	if ((err = vop_stdfsync(ap)))
1304 		return err;
1305 
1306 	return fuse_internal_fsync(vp, td, waitfor, false);
1307 }
1308 
1309 /*
1310     struct vnop_getattr_args {
1311 	struct vnode *a_vp;
1312 	struct vattr *a_vap;
1313 	struct ucred *a_cred;
1314 	struct thread *a_td;
1315     };
1316 */
1317 static int
fuse_vnop_getattr(struct vop_getattr_args * ap)1318 fuse_vnop_getattr(struct vop_getattr_args *ap)
1319 {
1320 	struct vnode *vp = ap->a_vp;
1321 	struct vattr *vap = ap->a_vap;
1322 	struct ucred *cred = ap->a_cred;
1323 	struct thread *td = curthread;
1324 	int err = 0;
1325 
1326 	err = fuse_internal_getattr(vp, vap, cred, td);
1327 	if (err == ENOTCONN && vnode_isvroot(vp)) {
1328 		/*
1329 		 * We want to seem a legitimate fs even if the daemon is dead,
1330 		 * so that, eg., we can still do path based unmounting after
1331 		 * the daemon dies.
1332 		 */
1333 		err = 0;
1334 		bzero(vap, sizeof(*vap));
1335 		vap->va_type = vnode_vtype(vp);
1336 	}
1337 	return err;
1338 }
1339 
1340 /*
1341     struct vnop_inactive_args {
1342 	struct vnode *a_vp;
1343     };
1344 */
1345 static int
fuse_vnop_inactive(struct vop_inactive_args * ap)1346 fuse_vnop_inactive(struct vop_inactive_args *ap)
1347 {
1348 	struct vnode *vp = ap->a_vp;
1349 	struct thread *td = curthread;
1350 
1351 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1352 	struct fuse_filehandle *fufh, *fufh_tmp;
1353 
1354 	int need_flush = 1;
1355 
1356 	ASSERT_CACHED_ATTRS_LOCKED(vp);	/* For fvdat->flag */
1357 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1358 		if (need_flush && vp->v_type == VREG) {
1359 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1360 				fuse_vnode_savesize(vp, NULL, 0);
1361 			}
1362 			if ((fvdat->flag & FN_REVOKED) != 0)
1363 				fuse_io_invalbuf(vp, td);
1364 			else
1365 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1366 			need_flush = 0;
1367 		}
1368 		fuse_filehandle_close(vp, fufh, td, NULL);
1369 	}
1370 
1371 	if ((fvdat->flag & FN_REVOKED) != 0)
1372 		vrecycle(vp);
1373 
1374 	return 0;
1375 }
1376 
1377 /*
1378     struct vnop_ioctl_args {
1379 	struct vnode *a_vp;
1380 	u_long a_command;
1381 	caddr_t a_data;
1382 	int a_fflag;
1383 	struct ucred *a_cred;
1384 	struct thread *a_td;
1385     };
1386 */
1387 static int
fuse_vnop_ioctl(struct vop_ioctl_args * ap)1388 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1389 {
1390 	struct vnode *vp = ap->a_vp;
1391 	struct mount *mp = vnode_mount(vp);
1392 	struct ucred *cred = ap->a_cred;
1393 	struct thread *td = ap->a_td;
1394 	int err;
1395 
1396 	if (fuse_isdeadfs(vp)) {
1397 		return (ENXIO);
1398 	}
1399 
1400 	switch (ap->a_command) {
1401 	case FIOSEEKDATA:
1402 	case FIOSEEKHOLE:
1403 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1404 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1405 			off_t *offp = ap->a_data;
1406 			pid_t pid = td->td_proc->p_pid;
1407 			int whence;
1408 
1409 			if (ap->a_command == FIOSEEKDATA)
1410 				whence = SEEK_DATA;
1411 			else
1412 				whence = SEEK_HOLE;
1413 
1414 			vn_lock(vp, LK_SHARED | LK_RETRY);
1415 			err = fuse_vnop_do_lseek(vp, td, cred, pid, offp,
1416 			    whence);
1417 			VOP_UNLOCK(vp);
1418 		}
1419 		if (fsess_not_impl(mp, FUSE_LSEEK))
1420 			err = vop_stdioctl(ap);
1421 		break;
1422 	default:
1423 		err = fuse_vnop_do_ioctl(vp, ap->a_command, ap->a_data,
1424 		    ap->a_fflag, cred, td);
1425 		break;
1426 	}
1427 	return (err);
1428 }
1429 
1430 
1431 /*
1432     struct vnop_link_args {
1433 	struct vnode *a_tdvp;
1434 	struct vnode *a_vp;
1435 	struct componentname *a_cnp;
1436     };
1437 */
1438 static int
fuse_vnop_link(struct vop_link_args * ap)1439 fuse_vnop_link(struct vop_link_args *ap)
1440 {
1441 	struct vnode *vp = ap->a_vp;
1442 	struct vnode *tdvp = ap->a_tdvp;
1443 	struct componentname *cnp = ap->a_cnp;
1444 
1445 	struct vattr *vap = VTOVA(vp);
1446 
1447 	struct fuse_dispatcher fdi;
1448 	struct fuse_entry_out *feo;
1449 	struct fuse_link_in fli;
1450 
1451 	int err;
1452 
1453 	if (fuse_isdeadfs(vp)) {
1454 		return (EXTERROR(ENXIO, "This FUSE session is about "
1455 		    "to be closed"));
1456 	}
1457 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1458 		return (EXDEV);
1459 	}
1460 
1461 	/*
1462 	 * This is a seatbelt check to protect naive userspace filesystems from
1463 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1464 	 * filesystem does not allow attribute caching, assume it is capable of
1465 	 * validating that nlink does not overflow.
1466 	 */
1467 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1468 		return (EMLINK);
1469 	fli.oldnodeid = VTOI(vp);
1470 
1471 	fdisp_init(&fdi, 0);
1472 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1473 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1474 	if ((err = fdisp_wait_answ(&fdi))) {
1475 		goto out;
1476 	}
1477 	feo = fdi.answ;
1478 
1479 	if (fli.oldnodeid != feo->nodeid) {
1480 		static const char exterr[] = "Server assigned wrong inode "
1481 		    "for a hard link.";
1482 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1483 		fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, exterr);
1484 		fuse_vnode_clear_attr_cache(vp);
1485 		fuse_vnode_clear_attr_cache(tdvp);
1486 		err = EXTERROR(EIO, exterr);
1487 		goto out;
1488 	}
1489 
1490 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1491 	if (!err) {
1492 		/*
1493 		 * Purge the parent's attribute cache because the daemon
1494 		 * should've updated its mtime and ctime
1495 		 */
1496 		fuse_vnode_clear_attr_cache(tdvp);
1497 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1498 			feo->attr_valid_nsec, NULL, true);
1499 	}
1500 out:
1501 	fdisp_destroy(&fdi);
1502 	return err;
1503 }
1504 
1505 struct fuse_lookup_alloc_arg {
1506 	struct fuse_entry_out *feo;
1507 	struct componentname *cnp;
1508 	uint64_t nid;
1509 	__enum_uint8(vtype) vtyp;
1510 };
1511 
1512 /* Callback for vn_get_ino */
1513 static int
fuse_lookup_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)1514 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1515 {
1516 	struct fuse_lookup_alloc_arg *flaa = arg;
1517 
1518 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1519 		flaa->vtyp);
1520 }
1521 
1522 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1523 	"int", "struct timespec*", "struct timespec*");
1524 /*
1525     struct vnop_lookup_args {
1526 	struct vnodeop_desc *a_desc;
1527 	struct vnode *a_dvp;
1528 	struct vnode **a_vpp;
1529 	struct componentname *a_cnp;
1530     };
1531 */
1532 int
fuse_vnop_lookup(struct vop_lookup_args * ap)1533 fuse_vnop_lookup(struct vop_lookup_args *ap)
1534 {
1535 	struct vnode *dvp = ap->a_dvp;
1536 	struct vnode **vpp = ap->a_vpp;
1537 	struct componentname *cnp = ap->a_cnp;
1538 	struct thread *td = curthread;
1539 	struct ucred *cred = cnp->cn_cred;
1540 	struct timespec now;
1541 
1542 	int nameiop = cnp->cn_nameiop;
1543 	bool isdotdot = cnp->cn_flags & ISDOTDOT;
1544 	bool islastcn = cnp->cn_flags & ISLASTCN;
1545 	struct mount *mp = vnode_mount(dvp);
1546 	struct fuse_data *data = fuse_get_mpdata(mp);
1547 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1548 	bool is_dot;
1549 
1550 	int err = 0;
1551 	int lookup_err = 0;
1552 	struct vnode *vp = NULL;
1553 
1554 	struct fuse_dispatcher fdi;
1555 	bool did_lookup = false;
1556 	struct fuse_entry_out *feo = NULL;
1557 	__enum_uint8(vtype) vtyp;	/* vnode type of target */
1558 
1559 	uint64_t nid;
1560 
1561 	if (fuse_isdeadfs(dvp)) {
1562 		*vpp = NULL;
1563 		return (EXTERROR(ENXIO, "This FUSE session is about "
1564 		    "to be closed"));
1565 	}
1566 	if (!vnode_isdir(dvp))
1567 		return ENOTDIR;
1568 
1569 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1570 		return EROFS;
1571 
1572 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1573 		cnp->cn_flags &= ~NOEXECCHECK;
1574 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1575 		return err;
1576 
1577 	ASSERT_CACHED_ATTRS_LOCKED(dvp);	/* For flag */
1578 	is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1579 	if (isdotdot && !(data->dataflags & FSESS_EXPORT_SUPPORT)) {
1580 		if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1581 			/*
1582 			 * Since the file system doesn't support ".." lookups,
1583 			 * we have no way to find this entry.
1584 			 */
1585 			return (EXTERROR(ESTALE, "This server does not support "
1586 			    "'..' lookups"));
1587 		}
1588 		nid = VTOFUD(dvp)->parent_nid;
1589 		if (nid == 0)
1590 			return ENOENT;
1591 		/* .. is obviously a directory */
1592 		vtyp = VDIR;
1593 	} else if (is_dot) {
1594 		nid = VTOI(dvp);
1595 		/* . is obviously a directory */
1596 		vtyp = VDIR;
1597 	} else {
1598 		struct timespec timeout;
1599 		int ncpticks; /* here to accommodate for API contract */
1600 
1601 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1602 		getnanouptime(&now);
1603 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1604 		switch (err) {
1605 		case -1:		/* positive match */
1606 			if (timespeccmp(&timeout, &now, >)) {
1607 				counter_u64_add(fuse_lookup_cache_hits, 1);
1608 			} else {
1609 				/* Cache timeout */
1610 				counter_u64_add(fuse_lookup_cache_misses, 1);
1611 				bintime_clear(
1612 					&VTOFUD(*vpp)->entry_cache_timeout);
1613 				cache_purge(*vpp);
1614 				if (dvp != *vpp)
1615 					vput(*vpp);
1616 				else
1617 					vrele(*vpp);
1618 				*vpp = NULL;
1619 				break;
1620 			}
1621 			return 0;
1622 
1623 		case 0:		/* no match in cache */
1624 			counter_u64_add(fuse_lookup_cache_misses, 1);
1625 			break;
1626 
1627 		case ENOENT:		/* negative match */
1628 			if (timespeccmp(&timeout, &now, <=)) {
1629 				/* Cache timeout */
1630 				cache_purge_negative(dvp);
1631 				break;
1632 			}
1633 			/* fall through */
1634 		default:
1635 			return err;
1636 		}
1637 
1638 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1639 		fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1640 
1641 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1642 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1643 		lookup_err = fdisp_wait_answ(&fdi);
1644 		did_lookup = true;
1645 
1646 		if (!lookup_err) {
1647 			/* lookup call succeeded */
1648 			feo = (struct fuse_entry_out *)fdi.answ;
1649 			nid = feo->nodeid;
1650 			if (nid == 0) {
1651 				/* zero nodeid means ENOENT and cache it */
1652 				struct timespec timeout;
1653 
1654 				fdi.answ_stat = ENOENT;
1655 				lookup_err = ENOENT;
1656 				if (cnp->cn_flags & MAKEENTRY) {
1657 					fuse_validity_2_timespec(feo, &timeout);
1658 					/* Use the same entry_time for .. as for
1659 					 * the file itself.  That doesn't honor
1660 					 * exactly what the fuse server tells
1661 					 * us, but to do otherwise would require
1662 					 * another cache lookup at this point.
1663 					 */
1664 					struct timespec *dtsp = NULL;
1665 					cache_enter_time(dvp, *vpp, cnp,
1666 						&timeout, dtsp);
1667 				}
1668 			}
1669 			vtyp = IFTOVT(feo->attr.mode);
1670 		}
1671 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1672 			fdisp_destroy(&fdi);
1673 			return lookup_err;
1674 		}
1675 	}
1676 	/* lookup_err, if non-zero, must be ENOENT at this point */
1677 
1678 	if (lookup_err) {
1679 		/* Entry not found */
1680 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1681 			if (default_permissions)
1682 				err = fuse_internal_access(dvp, VWRITE, td,
1683 				    cred);
1684 			else
1685 				err = 0;
1686 			if (!err) {
1687 				err = EJUSTRETURN;
1688 			}
1689 		} else {
1690 			err = ENOENT;
1691 		}
1692 	} else {
1693 		/* Entry was found */
1694 		if (isdotdot) {
1695 			struct fuse_lookup_alloc_arg flaa;
1696 
1697 			flaa.nid = nid;
1698 			flaa.feo = feo;
1699 			flaa.cnp = cnp;
1700 			flaa.vtyp = vtyp;
1701 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1702 				&vp);
1703 			*vpp = vp;
1704 		} else if (nid == VTOI(dvp)) {
1705 			if (is_dot) {
1706 				vref(dvp);
1707 				*vpp = dvp;
1708 			} else {
1709 				static const char exterr[] = "Server assigned "
1710 				    "same inode to both parent and child.";
1711 				fuse_warn(fuse_get_mpdata(mp),
1712 				    FSESS_WARN_ILLEGAL_INODE, exterr);
1713 				err = EXTERROR(EIO, exterr);
1714 			}
1715 
1716 		} else {
1717 			struct fuse_vnode_data *fvdat;
1718 
1719 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1720 			    &vp, cnp, vtyp);
1721 			if (err)
1722 				goto out;
1723 			*vpp = vp;
1724 			fvdat = VTOFUD(vp);
1725 
1726 			MPASS(feo != NULL);
1727 			if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1728 				/*
1729 				 * Attributes from the server are definitely
1730 				 * newer than the last attributes we sent to
1731 				 * the server, so cache them.
1732 				 */
1733 				fuse_internal_cache_attrs(*vpp, &feo->attr,
1734 					feo->attr_valid, feo->attr_valid_nsec,
1735 					NULL, true);
1736 			}
1737 			fuse_validity_2_bintime(feo->entry_valid,
1738 				feo->entry_valid_nsec,
1739 				&fvdat->entry_cache_timeout);
1740 
1741 			if ((nameiop == DELETE || nameiop == RENAME) &&
1742 				islastcn && default_permissions)
1743 			{
1744 				struct vattr dvattr;
1745 
1746 				err = fuse_internal_access(dvp, VWRITE, td,
1747 					cred);
1748 				if (err != 0)
1749 					goto out;
1750 				/*
1751 				 * if the parent's sticky bit is set, check
1752 				 * whether we're allowed to remove the file.
1753 				 * Need to figure out the vnode locking to make
1754 				 * this work.
1755 				 */
1756 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1757 				if ((dvattr.va_mode & S_ISTXT) &&
1758 					fuse_internal_access(dvp, VADMIN, td,
1759 						cred) &&
1760 					fuse_internal_access(*vpp, VADMIN, td,
1761 						cred)) {
1762 					err = EPERM;
1763 					goto out;
1764 				}
1765 			}
1766 		}
1767 	}
1768 out:
1769 	if (err) {
1770 		if (vp != NULL && dvp != vp)
1771 			vput(vp);
1772 		else if (vp != NULL)
1773 			vrele(vp);
1774 		*vpp = NULL;
1775 	}
1776 	if (did_lookup)
1777 		fdisp_destroy(&fdi);
1778 
1779 	return err;
1780 }
1781 
1782 /*
1783     struct vnop_mkdir_args {
1784 	struct vnode *a_dvp;
1785 	struct vnode **a_vpp;
1786 	struct componentname *a_cnp;
1787 	struct vattr *a_vap;
1788     };
1789 */
1790 static int
fuse_vnop_mkdir(struct vop_mkdir_args * ap)1791 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1792 {
1793 	struct vnode *dvp = ap->a_dvp;
1794 	struct vnode **vpp = ap->a_vpp;
1795 	struct componentname *cnp = ap->a_cnp;
1796 	struct vattr *vap = ap->a_vap;
1797 
1798 	struct fuse_mkdir_in fmdi;
1799 
1800 	if (fuse_isdeadfs(dvp)) {
1801 		return (EXTERROR(ENXIO, "This FUSE session is about "
1802 		    "to be closed"));
1803 	}
1804 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1805 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1806 
1807 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1808 	    sizeof(fmdi), VDIR));
1809 }
1810 
1811 /*
1812     struct vnop_mknod_args {
1813 	struct vnode *a_dvp;
1814 	struct vnode **a_vpp;
1815 	struct componentname *a_cnp;
1816 	struct vattr *a_vap;
1817     };
1818 */
1819 static int
fuse_vnop_mknod(struct vop_mknod_args * ap)1820 fuse_vnop_mknod(struct vop_mknod_args *ap)
1821 {
1822 
1823 	struct vnode *dvp = ap->a_dvp;
1824 	struct vnode **vpp = ap->a_vpp;
1825 	struct componentname *cnp = ap->a_cnp;
1826 	struct vattr *vap = ap->a_vap;
1827 
1828 	if (fuse_isdeadfs(dvp))
1829 		return (EXTERROR(ENXIO, "This FUSE session is about "
1830 		    "to be closed"));
1831 
1832 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1833 }
1834 
1835 /*
1836     struct vop_open_args {
1837 	struct vnode *a_vp;
1838 	int  a_mode;
1839 	struct ucred *a_cred;
1840 	struct thread *a_td;
1841 	int a_fdidx; / struct file *a_fp;
1842     };
1843 */
1844 static int
fuse_vnop_open(struct vop_open_args * ap)1845 fuse_vnop_open(struct vop_open_args *ap)
1846 {
1847 	struct vnode *vp = ap->a_vp;
1848 	int a_mode = ap->a_mode;
1849 	struct thread *td = ap->a_td;
1850 	struct ucred *cred = ap->a_cred;
1851 	pid_t pid = td->td_proc->p_pid;
1852 
1853 	if (fuse_isdeadfs(vp))
1854 		return (EXTERROR(ENXIO, "This FUSE session is about "
1855 		    "to be closed"));
1856 	if (VN_ISDEV(vp) || vp->v_type == VFIFO)
1857 		return (EXTERROR(EOPNOTSUPP, "Unsupported vnode type",
1858 		    vp->v_type));
1859 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1860 		return (EXTERROR(EINVAL, "Illegal mode", a_mode));
1861 
1862 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1863 		fuse_vnode_open(vp, 0, td);
1864 		return 0;
1865 	}
1866 
1867 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1868 }
1869 
1870 static int
fuse_vnop_pathconf(struct vop_pathconf_args * ap)1871 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1872 {
1873 	struct vnode *vp = ap->a_vp;
1874 	struct mount *mp;
1875 	struct fuse_filehandle *fufh;
1876 	int err;
1877 	bool closefufh = false;
1878 
1879 	switch (ap->a_name) {
1880 	case _PC_FILESIZEBITS:
1881 		*ap->a_retval = 64;
1882 		return (0);
1883 	case _PC_NAME_MAX:
1884 		*ap->a_retval = NAME_MAX;
1885 		return (0);
1886 	case _PC_LINK_MAX:
1887 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1888 		return (0);
1889 	case _PC_SYMLINK_MAX:
1890 		*ap->a_retval = MAXPATHLEN;
1891 		return (0);
1892 	case _PC_NO_TRUNC:
1893 		*ap->a_retval = 1;
1894 		return (0);
1895 	case _PC_MIN_HOLE_SIZE:
1896 		/*
1897 		 * The FUSE protocol provides no mechanism for a server to
1898 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1899 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1900 		 * 1 if it does.
1901 		 */
1902 		mp = vnode_mount(vp);
1903 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1904 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1905 			off_t offset = 0;
1906 
1907 			/*
1908 			 * Issue a FUSE_LSEEK to find out if it's supported.
1909 			 * Use SEEK_DATA instead of SEEK_HOLE, because the
1910 			 * latter generally requires sequential scans of file
1911 			 * metadata, which can be slow.
1912 			 */
1913 			err = fuse_vnop_do_lseek(vp, curthread,
1914 			    curthread->td_ucred, curthread->td_proc->p_pid,
1915 			    &offset, SEEK_DATA);
1916 			if (err == EBADF) {
1917 				/*
1918 				 * pathconf() doesn't necessarily open the
1919 				 * file.  So we may need to do it here.
1920 				 */
1921 				err = fuse_filehandle_open(vp, FREAD, &fufh,
1922 				    curthread, curthread->td_ucred);
1923 				if (err == 0) {
1924 					closefufh = true;
1925 					err = fuse_vnop_do_lseek(vp, curthread,
1926 					    curthread->td_ucred,
1927 					    curthread->td_proc->p_pid, &offset,
1928 					    SEEK_DATA);
1929 				}
1930 				if (closefufh)
1931 					fuse_filehandle_close(vp, fufh,
1932 					    curthread, curthread->td_ucred);
1933 			}
1934 
1935 		}
1936 
1937 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1938 			*ap->a_retval = 1;
1939 			return (0);
1940 		} else if (fsess_not_impl(mp, FUSE_LSEEK)) {
1941 			/* FUSE_LSEEK is not implemented */
1942 			return (EXTERROR(EINVAL, "This server does not "
1943 			    "implement FUSE_LSEEK"));
1944 		} else {
1945 			return (err);
1946 		}
1947 	default:
1948 		return (vop_stdpathconf(ap));
1949 	}
1950 }
1951 
1952 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1953     "struct uio*", "struct ucred*");
1954 /*
1955     struct vnop_read_args {
1956 	struct vnode *a_vp;
1957 	struct uio *a_uio;
1958 	int  a_ioflag;
1959 	struct ucred *a_cred;
1960     };
1961 */
1962 static int
fuse_vnop_read(struct vop_read_args * ap)1963 fuse_vnop_read(struct vop_read_args *ap)
1964 {
1965 	struct vnode *vp = ap->a_vp;
1966 	struct uio *uio = ap->a_uio;
1967 	int ioflag = ap->a_ioflag;
1968 	struct ucred *cred = ap->a_cred;
1969 	pid_t pid = curthread->td_proc->p_pid;
1970 	struct fuse_filehandle *fufh;
1971 	int err;
1972 	bool closefufh = false, directio;
1973 
1974 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1975 
1976 	if (fuse_isdeadfs(vp)) {
1977 		return (EXTERROR(ENXIO, "This FUSE session is about "
1978 		    "to be closed"));
1979 	}
1980 
1981 	/*
1982 	 * XXX Check this flag without the lock.  See
1983 	 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293088
1984 	 */
1985 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1986 		ioflag |= IO_DIRECT;
1987 	}
1988 
1989 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1990 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1991 		/*
1992 		 * nfsd will do I/O without first doing VOP_OPEN.  We
1993 		 * must implicitly open the file here
1994 		 */
1995 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1996 		closefufh = true;
1997 	}
1998 	if (err) {
1999 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2000 		return err;
2001 	}
2002 
2003 	/*
2004          * Ideally, when the daemon asks for direct io at open time, the
2005          * standard file flag should be set according to this, so that would
2006          * just change the default mode, which later on could be changed via
2007          * fcntl(2).
2008          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2009          * (don't know where). So to make any use of the Fuse direct_io option,
2010          * we hardwire it into the file's private data (similarly to Linux,
2011          * btw.).
2012          */
2013 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2014 
2015 	fuse_vnode_update(vp, FN_ATIMECHANGE);
2016 	if (directio) {
2017 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
2018 		err = fuse_read_directbackend(vp, uio, cred, fufh);
2019 	} else {
2020 		SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
2021 		err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
2022 	}
2023 
2024 	if (closefufh)
2025 		fuse_filehandle_close(vp, fufh, curthread, cred);
2026 
2027 	return (err);
2028 }
2029 
2030 /*
2031     struct vnop_readdir_args {
2032 	struct vnode *a_vp;
2033 	struct uio *a_uio;
2034 	struct ucred *a_cred;
2035 	int *a_eofflag;
2036 	int *a_ncookies;
2037 	uint64_t **a_cookies;
2038     };
2039 */
2040 static int
fuse_vnop_readdir(struct vop_readdir_args * ap)2041 fuse_vnop_readdir(struct vop_readdir_args *ap)
2042 {
2043 	struct vnode *vp = ap->a_vp;
2044 	struct uio *uio = ap->a_uio;
2045 	struct ucred *cred = ap->a_cred;
2046 	struct fuse_filehandle *fufh = NULL;
2047 	struct mount *mp = vnode_mount(vp);
2048 	struct fuse_iov cookediov;
2049 	int err = 0;
2050 	uint64_t *cookies;
2051 	ssize_t tresid;
2052 	int ncookies;
2053 	bool closefufh = false;
2054 	pid_t pid = curthread->td_proc->p_pid;
2055 
2056 	if (ap->a_eofflag)
2057 		*ap->a_eofflag = 0;
2058 	if (fuse_isdeadfs(vp)) {
2059 		return (EXTERROR(ENXIO, "This FUSE session is about "
2060 		    "to be closed"));
2061 	}
2062 	if (uio_resid(uio) < sizeof(struct dirent))
2063 		return (EXTERROR(EINVAL, "Buffer is too small"));
2064 
2065 	tresid = uio->uio_resid;
2066 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
2067 	if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
2068 		KASSERT(!fsess_is_impl(mp, FUSE_OPENDIR),
2069 			("FUSE file systems that implement "
2070 			 "FUSE_OPENDIR should not be exported"));
2071 		/*
2072 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
2073 		 * must implicitly open the directory here.
2074 		 */
2075 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
2076 		closefufh = true;
2077 	}
2078 	if (err)
2079 		return (err);
2080 	if (ap->a_ncookies != NULL) {
2081 		ncookies = uio->uio_resid /
2082 			(offsetof(struct dirent, d_name) + 4) + 1;
2083 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
2084 		*ap->a_ncookies = ncookies;
2085 		*ap->a_cookies = cookies;
2086 	} else {
2087 		ncookies = 0;
2088 		cookies = NULL;
2089 	}
2090 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
2091 	fiov_init(&cookediov, DIRCOOKEDSIZE);
2092 
2093 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
2094 		&ncookies, cookies);
2095 
2096 	fiov_teardown(&cookediov);
2097 	if (closefufh)
2098 		fuse_filehandle_close(vp, fufh, curthread, cred);
2099 
2100 	if (ap->a_ncookies != NULL) {
2101 		if (err == 0) {
2102 			*ap->a_ncookies -= ncookies;
2103 		} else {
2104 			free(*ap->a_cookies, M_TEMP);
2105 			*ap->a_ncookies = 0;
2106 			*ap->a_cookies = NULL;
2107 		}
2108 	}
2109 	if (err == 0 && tresid == uio->uio_resid)
2110 		*ap->a_eofflag = 1;
2111 
2112 	return err;
2113 }
2114 
2115 /*
2116     struct vnop_readlink_args {
2117 	struct vnode *a_vp;
2118 	struct uio *a_uio;
2119 	struct ucred *a_cred;
2120     };
2121 */
2122 static int
fuse_vnop_readlink(struct vop_readlink_args * ap)2123 fuse_vnop_readlink(struct vop_readlink_args *ap)
2124 {
2125 	struct vnode *vp = ap->a_vp;
2126 	struct uio *uio = ap->a_uio;
2127 	struct ucred *cred = ap->a_cred;
2128 
2129 	struct fuse_dispatcher fdi;
2130 	int err;
2131 
2132 	if (fuse_isdeadfs(vp)) {
2133 		return (EXTERROR(ENXIO, "This FUSE session is about "
2134 		    "to be closed"));
2135 	}
2136 	if (!vnode_islnk(vp)) {
2137 		return EINVAL;
2138 	}
2139 	fdisp_init(&fdi, 0);
2140 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
2141 	if (err) {
2142 		goto out;
2143 	}
2144 	if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) {
2145 		static const char exterr[] = "Server returned an embedded NUL "
2146 		    "from FUSE_READLINK.";
2147 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
2148 		fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, exterr);
2149 		err = EXTERROR(EIO, exterr);
2150 		goto out;
2151 	}
2152 	if (((char *)fdi.answ)[0] == '/' &&
2153 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
2154 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
2155 
2156 		err = uiomove(mpth, strlen(mpth), uio);
2157 	}
2158 	if (!err) {
2159 		err = uiomove(fdi.answ, fdi.iosize, uio);
2160 	}
2161 out:
2162 	fdisp_destroy(&fdi);
2163 	return err;
2164 }
2165 
2166 /*
2167     struct vnop_reclaim_args {
2168 	struct vnode *a_vp;
2169     };
2170 */
2171 static int
fuse_vnop_reclaim(struct vop_reclaim_args * ap)2172 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2173 {
2174 	struct vnode *vp = ap->a_vp;
2175 	struct thread *td = curthread;
2176 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2177 	struct fuse_filehandle *fufh, *fufh_tmp;
2178 
2179 	if (!fvdat) {
2180 		panic("FUSE: no vnode data during recycling");
2181 	}
2182 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2183 		printf("FUSE: vnode being reclaimed with open fufh "
2184 			"(type=%#x)", fufh->fufh_type);
2185 		fuse_filehandle_close(vp, fufh, td, NULL);
2186 	}
2187 
2188 	if (VTOI(vp) == 1) {
2189 		/*
2190 		 * Don't send FUSE_FORGET for the root inode, because
2191 		 * we never send FUSE_LOOKUP for it (see
2192 		 * fuse_vfsop_root) and we don't want the server to see
2193 		 * mismatched lookup counts.
2194 		 */
2195 		struct fuse_data *data;
2196 		struct vnode *vroot;
2197 
2198 		data = fuse_get_mpdata(vnode_mount(vp));
2199 		FUSE_LOCK();
2200 		vroot = data->vroot;
2201 		data->vroot = NULL;
2202 		FUSE_UNLOCK();
2203 		if (vroot)
2204 			vrele(vroot);
2205 	} else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2206 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2207 		    fvdat->nlookup);
2208 	}
2209 	cache_purge(vp);
2210 	vfs_hash_remove(vp);
2211 	fuse_vnode_destroy(vp);
2212 
2213 	return 0;
2214 }
2215 
2216 /*
2217     struct vnop_remove_args {
2218 	struct vnode *a_dvp;
2219 	struct vnode *a_vp;
2220 	struct componentname *a_cnp;
2221     };
2222 */
2223 static int
fuse_vnop_remove(struct vop_remove_args * ap)2224 fuse_vnop_remove(struct vop_remove_args *ap)
2225 {
2226 	struct vnode *dvp = ap->a_dvp;
2227 	struct vnode *vp = ap->a_vp;
2228 	struct componentname *cnp = ap->a_cnp;
2229 
2230 	int err;
2231 
2232 	if (fuse_isdeadfs(vp)) {
2233 		return (EXTERROR(ENXIO, "This FUSE session is about "
2234 		    "to be closed"));
2235 	}
2236 	if (vnode_isdir(vp)) {
2237 		return (EXTERROR(EPERM, "vnode is a directory"));
2238 	}
2239 
2240 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2241 
2242 	return err;
2243 }
2244 
2245 SDT_PROBE_DEFINE4(fusefs, , vnops, erelookup, "struct vnode*",
2246 	"struct vnode*", "struct vnode*", "struct vnode*");
2247 /*
2248     struct vnop_rename_args {
2249 	struct vnode *a_fdvp;
2250 	struct vnode *a_fvp;
2251 	struct componentname *a_fcnp;
2252 	struct vnode *a_tdvp;
2253 	struct vnode *a_tvp;
2254 	struct componentname *a_tcnp;
2255     };
2256 */
2257 static int
fuse_vnop_rename(struct vop_rename_args * ap)2258 fuse_vnop_rename(struct vop_rename_args *ap)
2259 {
2260 	struct vnode *fdvp = ap->a_fdvp;
2261 	struct vnode *fvp = ap->a_fvp;
2262 	struct componentname *fcnp = ap->a_fcnp;
2263 	struct vnode *tdvp = ap->a_tdvp;
2264 	struct vnode *tvp = ap->a_tvp;
2265 	struct componentname *tcnp = ap->a_tcnp;
2266 	struct fuse_data *data;
2267 	bool newparent = fdvp != tdvp;
2268 	bool isdir = fvp->v_type == VDIR;
2269 	int locktype;
2270 	int err = 0;
2271 
2272 	if (fuse_isdeadfs(fdvp)) {
2273 		return (EXTERROR(ENXIO, "This FUSE session is about "
2274 		    "to be closed"));
2275 	}
2276 	if (fvp->v_mount != tdvp->v_mount ||
2277 	    (tvp && fvp->v_mount != tvp->v_mount)) {
2278 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2279 		err = EXTERROR(EXDEV, "Cross-device rename");
2280 		goto out;
2281 	}
2282 	if (ap->a_flags != 0) {
2283 		err = EOPNOTSUPP;
2284 		goto out;
2285 	}
2286 	cache_purge(fvp);
2287 
2288 	/*
2289 	 * FUSE library is expected to check if target directory is not
2290 	 * under the source directory in the file system tree.
2291 	 * Linux performs this check at VFS level.
2292 	 */
2293 	/*
2294 	 * If source is a directory, and it will get a new parent, user must
2295 	 * have write permission to it, so ".." can be modified.
2296 	 */
2297 	data = fuse_get_mpdata(vnode_mount(tdvp));
2298 
2299 	if (tdvp != fdvp)
2300 		locktype = LK_EXCLUSIVE; /* for fuse_vnode_setparent */
2301 	else
2302 		locktype = LK_SHARED;
2303 
2304 	/*
2305 	 * Must use LK_NOWAIT to prevent LORs between fvp and tdvp or
2306 	 * tvp
2307 	 */
2308 	if (vn_lock(fvp, locktype | LK_NOWAIT) != 0) {
2309 		/*
2310 		 * Can't release tdvp or tvp to try avoiding the LOR.
2311 		 * Must return instead.
2312 		 */
2313 		SDT_PROBE4(fusefs, , vnops, erelookup, fdvp, fvp, tdvp,
2314 			tvp);
2315 		err = ERELOOKUP;
2316 		goto out;
2317 	}
2318 
2319 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2320 		err = fuse_internal_access(fvp, VWRITE,
2321 			curthread, tcnp->cn_cred);
2322 		if (err)
2323 			goto unlock;
2324 	}
2325 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2326 	if (err == 0) {
2327 		if (tdvp != fdvp)
2328 			fuse_vnode_setparent(fvp, tdvp);
2329 		if (tvp != NULL)
2330 			fuse_vnode_setparent(tvp, NULL);
2331 	}
2332 
2333 	if (tvp != NULL && tvp != fvp) {
2334 		cache_purge(tvp);
2335 	}
2336 	if (vnode_isdir(fvp)) {
2337 		if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2338 			cache_purge(tdvp);
2339 		}
2340 		cache_purge(fdvp);
2341 	}
2342 unlock:
2343 	VOP_UNLOCK(fvp);
2344 out:
2345 	if (tdvp == tvp) {
2346 		vrele(tdvp);
2347 	} else {
2348 		vput(tdvp);
2349 	}
2350 	if (tvp != NULL) {
2351 		vput(tvp);
2352 	}
2353 	vrele(fdvp);
2354 	vrele(fvp);
2355 
2356 	return err;
2357 }
2358 
2359 /*
2360     struct vnop_rmdir_args {
2361 	    struct vnode *a_dvp;
2362 	    struct vnode *a_vp;
2363 	    struct componentname *a_cnp;
2364     } *ap;
2365 */
2366 static int
fuse_vnop_rmdir(struct vop_rmdir_args * ap)2367 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2368 {
2369 	struct vnode *dvp = ap->a_dvp;
2370 	struct vnode *vp = ap->a_vp;
2371 
2372 	int err;
2373 
2374 	if (fuse_isdeadfs(vp)) {
2375 		return (EXTERROR(ENXIO, "This FUSE session is about "
2376 		    "to be closed"));
2377 	}
2378 	if (VTOFUD(vp) == VTOFUD(dvp)) {
2379 		return (EXTERROR(EINVAL, "Directory to be removed "
2380 		    "contains itself"));
2381 	}
2382 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2383 
2384 	return err;
2385 }
2386 
2387 /*
2388     struct vnop_setattr_args {
2389 	struct vnode *a_vp;
2390 	struct vattr *a_vap;
2391 	struct ucred *a_cred;
2392 	struct thread *a_td;
2393     };
2394 */
2395 static int
fuse_vnop_setattr(struct vop_setattr_args * ap)2396 fuse_vnop_setattr(struct vop_setattr_args *ap)
2397 {
2398 	struct vnode *vp = ap->a_vp;
2399 	struct vattr *vap = ap->a_vap;
2400 	struct ucred *cred = ap->a_cred;
2401 	struct thread *td = curthread;
2402 	struct mount *mp;
2403 	struct fuse_data *data;
2404 	struct vattr old_va;
2405 	int dataflags;
2406 	int err = 0, err2;
2407 	accmode_t accmode = 0;
2408 	bool checkperm;
2409 	bool drop_suid = false;
2410 
2411 	mp = vnode_mount(vp);
2412 	data = fuse_get_mpdata(mp);
2413 	dataflags = data->dataflags;
2414 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2415 
2416 	if (fuse_isdeadfs(vp)) {
2417 		return (EXTERROR(ENXIO, "This FUSE session is about "
2418 		    "to be closed"));
2419 	}
2420 
2421 	if (vap->va_uid != (uid_t)VNOVAL) {
2422 		if (checkperm) {
2423 			/* Only root may change a file's owner */
2424 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2425 			if (err) {
2426 				/* As a special case, allow the null chown */
2427 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2428 					td);
2429 				if (err2)
2430 					return (err2);
2431 				if (vap->va_uid != old_va.va_uid)
2432 					return err;
2433 				drop_suid = true;
2434 			}
2435 		}
2436 		accmode |= VADMIN;
2437 	}
2438 	if (vap->va_gid != (gid_t)VNOVAL) {
2439 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2440 			drop_suid = true;
2441 		if (checkperm && !groupmember(vap->va_gid, cred)) {
2442 			/*
2443 			 * Non-root users may only chgrp to one of their own
2444 			 * groups
2445 			 */
2446 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2447 			if (err) {
2448 				/* As a special case, allow the null chgrp */
2449 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2450 					td);
2451 				if (err2)
2452 					return (err2);
2453 				if (vap->va_gid != old_va.va_gid)
2454 					return err;
2455 			}
2456 		}
2457 		accmode |= VADMIN;
2458 	}
2459 	if (vap->va_size != VNOVAL) {
2460 		switch (vp->v_type) {
2461 		case VDIR:
2462 			return (EISDIR);
2463 		case VLNK:
2464 		case VREG:
2465 			if (vfs_isrdonly(mp))
2466 				return (EROFS);
2467 			err = vn_rlimit_trunc(vap->va_size, td);
2468 			if (err)
2469 				return (err);
2470 			break;
2471 		default:
2472 			/*
2473 			 * According to POSIX, the result is unspecified
2474 			 * for file types other than regular files,
2475 			 * directories and shared memory objects.  We
2476 			 * don't support shared memory objects in the file
2477 			 * system, and have dubious support for truncating
2478 			 * symlinks.  Just ignore the request in other cases.
2479 			 */
2480 			return (0);
2481 		}
2482 		/* Don't set accmode.  Permission to trunc is checked upstack */
2483 	}
2484 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2485 		if (vap->va_vaflags & VA_UTIMES_NULL)
2486 			accmode |= VWRITE;
2487 		else
2488 			accmode |= VADMIN;
2489 	}
2490 	if (drop_suid) {
2491 		if (vap->va_mode != (mode_t)VNOVAL)
2492 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2493 		else {
2494 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2495 			if (err)
2496 				return (err);
2497 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2498 		}
2499 	}
2500 	if (vap->va_mode != (mode_t)VNOVAL) {
2501 		/* Only root may set the sticky bit on non-directories */
2502 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2503 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2504 			return EFTYPE;
2505 		if (checkperm && (vap->va_mode & S_ISGID)) {
2506 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2507 			if (err)
2508 				return (err);
2509 			if (!groupmember(old_va.va_gid, cred)) {
2510 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2511 				if (err)
2512 					return (err);
2513 			}
2514 		}
2515 		accmode |= VADMIN;
2516 	}
2517 
2518 	if (vfs_isrdonly(mp))
2519 		return EROFS;
2520 
2521 	if (checkperm) {
2522 		err = fuse_internal_access(vp, accmode, td, cred);
2523 	} else {
2524 		err = 0;
2525 	}
2526 	if (err)
2527 		return err;
2528 	else
2529 		return fuse_internal_setattr(vp, vap, td, cred);
2530 }
2531 
2532 /*
2533     struct vnop_strategy_args {
2534 	struct vnode *a_vp;
2535 	struct buf *a_bp;
2536     };
2537 */
2538 static int
fuse_vnop_strategy(struct vop_strategy_args * ap)2539 fuse_vnop_strategy(struct vop_strategy_args *ap)
2540 {
2541 	struct vnode *vp = ap->a_vp;
2542 	struct buf *bp = ap->a_bp;
2543 
2544 	if (!vp || fuse_isdeadfs(vp)) {
2545 		bp->b_ioflags |= BIO_ERROR;
2546 		bp->b_error = ENXIO;
2547 		bufdone(bp);
2548 		return 0;
2549 	}
2550 
2551 	/*
2552 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2553 	 * fuse_io_strategy sets bp's error fields
2554 	 */
2555 	(void)fuse_io_strategy(vp, bp);
2556 
2557 	return 0;
2558 }
2559 
2560 /*
2561     struct vnop_symlink_args {
2562 	struct vnode *a_dvp;
2563 	struct vnode **a_vpp;
2564 	struct componentname *a_cnp;
2565 	struct vattr *a_vap;
2566 	char *a_target;
2567     };
2568 */
2569 static int
fuse_vnop_symlink(struct vop_symlink_args * ap)2570 fuse_vnop_symlink(struct vop_symlink_args *ap)
2571 {
2572 	struct vnode *dvp = ap->a_dvp;
2573 	struct vnode **vpp = ap->a_vpp;
2574 	struct componentname *cnp = ap->a_cnp;
2575 	const char *target = ap->a_target;
2576 
2577 	struct fuse_dispatcher fdi;
2578 
2579 	int err;
2580 	size_t len;
2581 
2582 	if (fuse_isdeadfs(dvp)) {
2583 		return (EXTERROR(ENXIO, "This FUSE session is about "
2584 		    "to be closed"));
2585 	}
2586 	/*
2587 	 * Unlike the other creator type calls, here we have to create a message
2588 	 * where the name of the new entry comes first, and the data describing
2589 	 * the entry comes second.
2590 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2591 	 * but put together the message manually and just call the core part.
2592 	 */
2593 
2594 	len = strlen(target) + 1;
2595 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2596 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2597 
2598 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2599 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2600 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2601 
2602 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2603 	fdisp_destroy(&fdi);
2604 	return err;
2605 }
2606 
2607 /*
2608     struct vnop_write_args {
2609 	struct vnode *a_vp;
2610 	struct uio *a_uio;
2611 	int  a_ioflag;
2612 	struct ucred *a_cred;
2613     };
2614 */
2615 static int
fuse_vnop_write(struct vop_write_args * ap)2616 fuse_vnop_write(struct vop_write_args *ap)
2617 {
2618 	struct vnode *vp = ap->a_vp;
2619 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2620 	struct uio *uio = ap->a_uio;
2621 	int ioflag = ap->a_ioflag;
2622 	struct ucred *cred = ap->a_cred;
2623 	pid_t pid = curthread->td_proc->p_pid;
2624 	struct fuse_filehandle *fufh;
2625 	int err;
2626 	bool closefufh = false, directio;
2627 
2628 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2629 
2630 	if (fuse_isdeadfs(vp)) {
2631 		return (EXTERROR(ENXIO, "This FUSE session is about "
2632 		    "to be closed"));
2633 	}
2634 
2635 	/*
2636 	 * XXX Check this flag without the lock.  See
2637 	 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293088
2638 	 */
2639 	if (fvdat->flag & FN_DIRECTIO)
2640 		ioflag |= IO_DIRECT;
2641 
2642 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2643 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2644 		/*
2645 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2646 		 * must implicitly open the file here
2647 		 */
2648 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2649 		closefufh = true;
2650 	}
2651 	if (err) {
2652 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2653 		return err;
2654 	}
2655 
2656 	/*
2657          * Ideally, when the daemon asks for direct io at open time, the
2658          * standard file flag should be set according to this, so that would
2659          * just change the default mode, which later on could be changed via
2660          * fcntl(2).
2661          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2662          * (don't know where). So to make any use of the Fuse direct_io option,
2663          * we hardwire it into the file's private data (similarly to Linux,
2664          * btw.).
2665          */
2666 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2667 
2668 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2669 	if (directio) {
2670 		off_t start, end, filesize;
2671 		bool pages = (ioflag & IO_VMIO) != 0;
2672 
2673 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2674 
2675 		err = fuse_vnode_size(vp, &filesize, cred, curthread);
2676 		if (err)
2677 			goto out;
2678 
2679 		start = uio->uio_offset;
2680 		end = start + uio->uio_resid;
2681 		if (!pages) {
2682 			err = fuse_inval_buf_range(vp, filesize, start,
2683 			    end, PCATCH);
2684 			if (err)
2685 				goto out;
2686 		}
2687 		err = fuse_write_directbackend(vp, uio, cred, fufh,
2688 			filesize, ioflag, pages);
2689 	} else {
2690 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2691 			"buffered write of vnode");
2692 		if (!fsess_opt_writeback(vnode_mount(vp)))
2693 			ioflag |= IO_SYNC;
2694 		err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2695 	}
2696 	fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2697 
2698 out:
2699 	if (closefufh)
2700 		fuse_filehandle_close(vp, fufh, curthread, cred);
2701 
2702 	return (err);
2703 }
2704 
2705 static daddr_t
fuse_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)2706 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2707 {
2708 	const int biosize = fuse_iosize(vp);
2709 
2710 	return (off / biosize);
2711 }
2712 
2713 static int
fuse_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * blksz)2714 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2715 {
2716 	off_t filesize;
2717 	int err;
2718 	const int biosize = fuse_iosize(vp);
2719 
2720 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2721 	if (err) {
2722 		/* This will turn into a SIGBUS */
2723 		return (EIO);
2724 	} else if ((off_t)lbn * biosize >= filesize) {
2725 		*blksz = 0;
2726 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2727 		*blksz = filesize - (off_t)lbn *biosize;
2728 	} else {
2729 		*blksz = biosize;
2730 	}
2731 	return (0);
2732 }
2733 
2734 /*
2735     struct vnop_getpages_args {
2736 	struct vnode *a_vp;
2737 	vm_page_t *a_m;
2738 	int a_count;
2739 	int a_reqpage;
2740     };
2741 */
2742 static int
fuse_vnop_getpages(struct vop_getpages_args * ap)2743 fuse_vnop_getpages(struct vop_getpages_args *ap)
2744 {
2745 	struct vnode *vp = ap->a_vp;
2746 
2747 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2748 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2749 			"called on non-cacheable vnode??\n");
2750 		return (VM_PAGER_ERROR);
2751 	}
2752 
2753 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2754 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2755 }
2756 
2757 static const char extattr_namespace_separator = '.';
2758 
2759 /*
2760     struct vop_getextattr_args {
2761 	struct vop_generic_args a_gen;
2762 	struct vnode *a_vp;
2763 	int a_attrnamespace;
2764 	const char *a_name;
2765 	struct uio *a_uio;
2766 	size_t *a_size;
2767 	struct ucred *a_cred;
2768 	struct thread *a_td;
2769     };
2770 */
2771 static int
fuse_vnop_getextattr(struct vop_getextattr_args * ap)2772 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2773 {
2774 	struct vnode *vp = ap->a_vp;
2775 	struct uio *uio = ap->a_uio;
2776 	struct fuse_dispatcher fdi;
2777 	struct fuse_getxattr_in *get_xattr_in;
2778 	struct fuse_getxattr_out *get_xattr_out;
2779 	struct mount *mp = vnode_mount(vp);
2780 	struct thread *td = ap->a_td;
2781 	struct ucred *cred = ap->a_cred;
2782 	char *prefix;
2783 	char *attr_str;
2784 	size_t len;
2785 	int err;
2786 
2787 	if (fuse_isdeadfs(vp))
2788 		return (EXTERROR(ENXIO, "This FUSE session is about "
2789 		    "to be closed"));
2790 
2791 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2792 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2793 		    "extended attributes"));
2794 
2795 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2796 	if (err)
2797 		return err;
2798 
2799 	/* Default to looking for user attributes. */
2800 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2801 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2802 	else
2803 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2804 
2805 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2806 	    strlen(ap->a_name) + 1;
2807 
2808 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2809 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2810 
2811 	get_xattr_in = fdi.indata;
2812 	/*
2813 	 * Check to see whether we're querying the available size or
2814 	 * issuing the actual request.  If we pass in 0, we get back struct
2815 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2816 	 * that much data, without the struct fuse_getxattr_out header.
2817 	 */
2818 	if (uio == NULL)
2819 		get_xattr_in->size = 0;
2820 	else
2821 		get_xattr_in->size = uio->uio_resid;
2822 
2823 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2824 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2825 	    ap->a_name);
2826 
2827 	err = fdisp_wait_answ(&fdi);
2828 	if (err != 0) {
2829 		if (err == ENOSYS) {
2830 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2831 			err = (EXTERROR(EOPNOTSUPP, "This server does not "
2832 			    "implement extended attributes"));
2833 		}
2834 		goto out;
2835 	}
2836 
2837 	get_xattr_out = fdi.answ;
2838 
2839 	if (ap->a_size != NULL)
2840 		*ap->a_size = get_xattr_out->size;
2841 
2842 	if (uio != NULL)
2843 		err = uiomove(fdi.answ, fdi.iosize, uio);
2844 
2845 out:
2846 	fdisp_destroy(&fdi);
2847 	return (err);
2848 }
2849 
2850 /*
2851     struct vop_setextattr_args {
2852 	struct vop_generic_args a_gen;
2853 	struct vnode *a_vp;
2854 	int a_attrnamespace;
2855 	const char *a_name;
2856 	struct uio *a_uio;
2857 	struct ucred *a_cred;
2858 	struct thread *a_td;
2859     };
2860 */
2861 static int
fuse_vnop_setextattr(struct vop_setextattr_args * ap)2862 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2863 {
2864 	struct vnode *vp = ap->a_vp;
2865 	struct uio *uio = ap->a_uio;
2866 	struct fuse_dispatcher fdi;
2867 	struct fuse_setxattr_in *set_xattr_in;
2868 	struct mount *mp = vnode_mount(vp);
2869 	struct thread *td = ap->a_td;
2870 	struct ucred *cred = ap->a_cred;
2871 	size_t struct_size = FUSE_COMPAT_SETXATTR_IN_SIZE;
2872 	char *prefix;
2873 	size_t len;
2874 	char *attr_str;
2875 	int err;
2876 
2877 	if (fuse_isdeadfs(vp))
2878 		return (EXTERROR(ENXIO, "This FUSE session is about "
2879 		    "to be closed"));
2880 
2881 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2882 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2883 		    "setting extended attributes"));
2884 
2885 	if (vfs_isrdonly(mp))
2886 		return EROFS;
2887 
2888 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2889 	if (ap->a_uio == NULL) {
2890 		/*
2891 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2892 		 * return EOPNOTSUPP.
2893 		 */
2894 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2895 			return (EXTERROR(EOPNOTSUPP, "This server does not "
2896 			    "implement removing extended attributes"));
2897 		else
2898 			return (EXTERROR(EINVAL, "DELETEEXTATTR should be used "
2899 			    "to remove extattrs"));
2900 	}
2901 
2902 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2903 		VWRITE);
2904 	if (err)
2905 		return err;
2906 
2907 	/* Default to looking for user attributes. */
2908 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2909 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2910 	else
2911 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2912 
2913 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2914 	    strlen(ap->a_name) + 1;
2915 
2916 	/* older FUSE servers  use a smaller fuse_setxattr_in struct*/
2917 	if (fuse_get_mpdata(mp)->dataflags & FSESS_SETXATTR_EXT)
2918 		struct_size = sizeof(*set_xattr_in);
2919 
2920 	fdisp_init(&fdi, len + struct_size + uio->uio_resid);
2921 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2922 
2923 	set_xattr_in = fdi.indata;
2924 	set_xattr_in->size = uio->uio_resid;
2925 
2926 	if (fuse_get_mpdata(mp)->dataflags & FSESS_SETXATTR_EXT) {
2927 		set_xattr_in->setxattr_flags = 0;
2928 		set_xattr_in->padding = 0;
2929 	}
2930 
2931 	attr_str = (char *)fdi.indata + struct_size;
2932 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2933 	    ap->a_name);
2934 
2935 	err = uiomove((char *)fdi.indata + struct_size + len,
2936 	    uio->uio_resid, uio);
2937 	if (err != 0) {
2938 		goto out;
2939 	}
2940 
2941 	err = fdisp_wait_answ(&fdi);
2942 
2943 	if (err == ENOSYS) {
2944 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2945 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
2946 		    "setting extended attributes");
2947 	}
2948 	if (err == ERESTART) {
2949 		/* Can't restart after calling uiomove */
2950 		err = EINTR;
2951 	}
2952 
2953 out:
2954 	fdisp_destroy(&fdi);
2955 	return (err);
2956 }
2957 
2958 /*
2959  * The Linux / FUSE extended attribute list is simply a collection of
2960  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2961  * byte length followed by a non-NUL terminated string.  So, this allows
2962  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2963  * Linux attribute names are reported with the namespace as a prefix (e.g.
2964  * "user.attribute_name"), but in FreeBSD they are reported without the
2965  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2966  *
2967  * user.attr_name1\0user.attr_name2\0
2968  *
2969  * to:
2970  *
2971  * <num>attr_name1<num>attr_name2
2972  *
2973  * Where "<num>" is a single byte number of characters in the attribute name.
2974  *
2975  * Args:
2976  * prefix - exattr namespace prefix string
2977  * list, list_len - input list with namespace prefixes
2978  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2979  */
2980 static int
fuse_xattrlist_convert(struct fuse_data * data,char * prefix,const char * list,int list_len,char * bsd_list,int * bsd_list_len)2981 fuse_xattrlist_convert(struct fuse_data *data, char *prefix, const char *list,
2982     int list_len, char *bsd_list, int *bsd_list_len)
2983 {
2984 	int len, pos, dist_to_next, prefix_len;
2985 
2986 	pos = 0;
2987 	*bsd_list_len = 0;
2988 	prefix_len = strlen(prefix);
2989 
2990 	while (pos < list_len && list[pos] != '\0') {
2991 		dist_to_next = strnlen(&list[pos], list_len - pos - 1) + 1;
2992 		if (list[pos + dist_to_next - 1] != '\0') {
2993 			fuse_warn(data, FSESS_WARN_LSEXTATTR_NUL,
2994 				"The FUSE server returned a non nul-terminated "
2995 				"LISTXATTR response.");
2996 			return (EXTERROR(EIO,
2997 				"The FUSE server returned a malformed list"));
2998 		}
2999 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
3000 		    list[pos + prefix_len] == extattr_namespace_separator) {
3001 			len = dist_to_next -
3002 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
3003 			if (len >= EXTATTR_MAXNAMELEN)
3004 				return (ENAMETOOLONG);
3005 
3006 			bsd_list[*bsd_list_len] = len;
3007 			memcpy(&bsd_list[*bsd_list_len + 1],
3008 			    &list[pos + prefix_len +
3009 			    sizeof(extattr_namespace_separator)], len);
3010 
3011 			*bsd_list_len += len + 1;
3012 		}
3013 
3014 		pos += dist_to_next;
3015 	}
3016 
3017 	return (0);
3018 }
3019 
3020 /*
3021  * List extended attributes
3022  *
3023  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
3024  * has a number of differences compared to its FreeBSD equivalent,
3025  * extattr_list_file:
3026  *
3027  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
3028  *   whereas listxattr(2) only returns attributes for a single namespace
3029  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
3030  * - If the provided buffer is not large enough to hold the result,
3031  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
3032  *   return as many results as will fit.
3033  */
3034 /*
3035     struct vop_listextattr_args {
3036 	struct vop_generic_args a_gen;
3037 	struct vnode *a_vp;
3038 	int a_attrnamespace;
3039 	struct uio *a_uio;
3040 	size_t *a_size;
3041 	struct ucred *a_cred;
3042 	struct thread *a_td;
3043     };
3044 */
3045 static int
fuse_vnop_listextattr(struct vop_listextattr_args * ap)3046 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
3047 {
3048 	struct vnode *vp = ap->a_vp;
3049 	struct uio *uio = ap->a_uio;
3050 	struct fuse_dispatcher fdi;
3051 	struct fuse_listxattr_in *list_xattr_in;
3052 	struct fuse_listxattr_out *list_xattr_out;
3053 	struct mount *mp = vnode_mount(vp);
3054 	struct fuse_data *data = fuse_get_mpdata(mp);
3055 	struct thread *td = ap->a_td;
3056 	struct ucred *cred = ap->a_cred;
3057 	char *prefix;
3058 	char *bsd_list = NULL;
3059 	char *linux_list;
3060 	int bsd_list_len;
3061 	int linux_list_len;
3062 	int err;
3063 
3064 	if (fuse_isdeadfs(vp))
3065 		return (EXTERROR(ENXIO, "This FUSE session is about "
3066 		    "to be closed"));
3067 
3068 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
3069 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3070 		    "extended attributes"));
3071 
3072 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
3073 	if (err)
3074 		return err;
3075 
3076 	/*
3077 	 * Add space for a NUL and the period separator if enabled.
3078 	 * Default to looking for user attributes.
3079 	 */
3080 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3081 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3082 	else
3083 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3084 
3085 	fdisp_init(&fdi, sizeof(*list_xattr_in));
3086 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
3087 
3088 	/*
3089 	 * Retrieve Linux / FUSE compatible list size.
3090 	 */
3091 	list_xattr_in = fdi.indata;
3092 	list_xattr_in->size = 0;
3093 
3094 	err = fdisp_wait_answ(&fdi);
3095 	if (err != 0) {
3096 		if (err == ENOSYS) {
3097 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
3098 			err = EXTERROR(EOPNOTSUPP, "This server does not "
3099 			    "implement extended attributes");
3100 		}
3101 		goto out;
3102 	}
3103 
3104 	list_xattr_out = fdi.answ;
3105 	linux_list_len = list_xattr_out->size;
3106 	if (linux_list_len == 0) {
3107 		if (ap->a_size != NULL)
3108 			*ap->a_size = linux_list_len;
3109 		goto out;
3110 	}
3111 
3112 	/*
3113 	 * Retrieve Linux / FUSE compatible list values.
3114 	 */
3115 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
3116 	list_xattr_in = fdi.indata;
3117 	list_xattr_in->size = linux_list_len;
3118 
3119 	err = fdisp_wait_answ(&fdi);
3120 	if (err == ERANGE) {
3121 		/*
3122 		 * Race detected.  The attribute list must've grown since the
3123 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
3124 		 * to userland so we can process signals, if necessary, before
3125 		 * restarting.
3126 		 */
3127 		err = ERESTART;
3128 		goto out;
3129 	} else if (err != 0)
3130 		goto out;
3131 
3132 	linux_list = fdi.answ;
3133 	/* FUSE doesn't allow the server to return more data than requested */
3134 	if (fdi.iosize > linux_list_len) {
3135 		fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
3136 			"server returned "
3137 			"more extended attribute data than requested; "
3138 			"should've returned ERANGE instead.");
3139 	} else {
3140 		/* But returning less data is fine */
3141 		linux_list_len = fdi.iosize;
3142 	}
3143 
3144 	/*
3145 	 * Retrieve the BSD compatible list values.
3146 	 * The Linux / FUSE attribute list format isn't the same
3147 	 * as FreeBSD's format. So we need to transform it into
3148 	 * FreeBSD's format before giving it to the user.
3149 	 */
3150 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
3151 	err = fuse_xattrlist_convert(data, prefix, linux_list, linux_list_len,
3152 	    bsd_list, &bsd_list_len);
3153 	if (err != 0)
3154 		goto out;
3155 
3156 	if (ap->a_size != NULL)
3157 		*ap->a_size = bsd_list_len;
3158 
3159 	if (uio != NULL)
3160 		err = uiomove(bsd_list, bsd_list_len, uio);
3161 
3162 out:
3163 	free(bsd_list, M_TEMP);
3164 	fdisp_destroy(&fdi);
3165 	return (err);
3166 }
3167 
3168 /*
3169     struct vop_deallocate_args {
3170 	struct vop_generic_args a_gen;
3171 	struct vnode *a_vp;
3172 	off_t *a_offset;
3173 	off_t *a_len;
3174 	int a_flags;
3175 	int a_ioflag;
3176         struct ucred *a_cred;
3177     };
3178 */
3179 static int
fuse_vnop_deallocate(struct vop_deallocate_args * ap)3180 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
3181 {
3182 	struct vnode *vp = ap->a_vp;
3183 	struct mount *mp = vnode_mount(vp);
3184 	struct fuse_filehandle *fufh;
3185 	struct fuse_dispatcher fdi;
3186 	struct fuse_fallocate_in *ffi;
3187 	struct ucred *cred = ap->a_cred;
3188 	pid_t pid = curthread->td_proc->p_pid;
3189 	off_t *len = ap->a_len;
3190 	off_t *offset = ap->a_offset;
3191 	int ioflag = ap->a_ioflag;
3192 	off_t filesize;
3193 	int err;
3194 	bool closefufh = false;
3195 
3196 	if (fuse_isdeadfs(vp))
3197 		return (EXTERROR(ENXIO, "This FUSE session is about "
3198 		    "to be closed"));
3199 
3200 	if (vfs_isrdonly(mp))
3201 		return (EROFS);
3202 
3203 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
3204 		goto fallback;
3205 
3206 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
3207 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
3208 		/*
3209 		 * nfsd will do I/O without first doing VOP_OPEN.  We
3210 		 * must implicitly open the file here
3211 		 */
3212 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
3213 		closefufh = true;
3214 	}
3215 	if (err)
3216 		return (err);
3217 
3218 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3219 
3220 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
3221 	if (err)
3222 		goto out;
3223 	err = fuse_inval_buf_range(vp, filesize, *offset, *offset + *len, 0);
3224 	if (err)
3225 		goto out;
3226 
3227 	fdisp_init(&fdi, sizeof(*ffi));
3228 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3229 	ffi = fdi.indata;
3230 	ffi->fh = fufh->fh_id;
3231 	ffi->offset = *offset;
3232 	ffi->length = *len;
3233 	/*
3234 	 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3235 	 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3236 	 */
3237 	ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3238 	err = fdisp_wait_answ(&fdi);
3239 
3240 	if (err == ENOSYS) {
3241 		fdisp_destroy(&fdi);
3242 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
3243 		goto fallback;
3244 	} else if (err == EOPNOTSUPP) {
3245 		/*
3246 		 * The file system server does not support FUSE_FALLOCATE with
3247 		 * the supplied mode for this particular file.
3248 		 */
3249 		fdisp_destroy(&fdi);
3250 		goto fallback;
3251 	} else if (!err) {
3252 		/*
3253 		 * Clip the returned offset to EoF.  Do it here rather than
3254 		 * before FUSE_FALLOCATE just in case the kernel's cached file
3255 		 * size is out of date.  Unfortunately, FUSE does not return
3256 		 * any information about filesize from that operation.
3257 		 */
3258 		*offset = MIN(*offset + *len, filesize);
3259 		*len = 0;
3260 		fuse_vnode_undirty_cached_timestamps(vp, false);
3261 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
3262 
3263 		if (ioflag & IO_SYNC)
3264 			err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3265 			    false);
3266 	}
3267 
3268 	fdisp_destroy(&fdi);
3269 out:
3270 	if (closefufh)
3271 		fuse_filehandle_close(vp, fufh, curthread, cred);
3272 
3273 	return (err);
3274 
3275 fallback:
3276 	if (closefufh)
3277 		fuse_filehandle_close(vp, fufh, curthread, cred);
3278 
3279 	return (vop_stddeallocate(ap));
3280 }
3281 
3282 /*
3283    struct vop_delayed_setsize_args {
3284 	struct vop_generic_args a_gen;
3285 	struct vnode *a_vp;
3286   };
3287  */
3288 static int
fuse_vnop_delayed_setsize(struct vop_delayed_setsize_args * ap)3289 fuse_vnop_delayed_setsize(struct vop_delayed_setsize_args *ap)
3290 {
3291 	struct vnode *vp = ap->a_vp;
3292 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3293 	bool shrink = (fvdat->flag & FN_DELAYED_TRUNCATE) != 0;
3294 	int err;
3295 
3296 	if (!fvdat)
3297 		return (0);
3298 
3299 	err = fuse_vnode_setsize_immediate(vp, shrink);
3300 	fvdat->flag &= ~FN_DELAYED_TRUNCATE;
3301 
3302 	return (err);
3303 }
3304 
3305 /*
3306     struct vop_deleteextattr_args {
3307 	struct vop_generic_args a_gen;
3308 	struct vnode *a_vp;
3309 	int a_attrnamespace;
3310 	const char *a_name;
3311 	struct ucred *a_cred;
3312 	struct thread *a_td;
3313     };
3314 */
3315 static int
fuse_vnop_deleteextattr(struct vop_deleteextattr_args * ap)3316 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3317 {
3318 	struct vnode *vp = ap->a_vp;
3319 	struct fuse_dispatcher fdi;
3320 	struct mount *mp = vnode_mount(vp);
3321 	struct thread *td = ap->a_td;
3322 	struct ucred *cred = ap->a_cred;
3323 	char *prefix;
3324 	size_t len;
3325 	char *attr_str;
3326 	int err;
3327 
3328 	if (fuse_isdeadfs(vp))
3329 		return (EXTERROR(ENXIO, "This FUSE session is about "
3330 		    "to be closed"));
3331 
3332 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3333 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3334 		    "removing extended attributes"));
3335 
3336 	if (vfs_isrdonly(mp))
3337 		return EROFS;
3338 
3339 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3340 		VWRITE);
3341 	if (err)
3342 		return err;
3343 
3344 	/* Default to looking for user attributes. */
3345 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3346 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3347 	else
3348 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3349 
3350 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3351 	    strlen(ap->a_name) + 1;
3352 
3353 	fdisp_init(&fdi, len);
3354 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3355 
3356 	attr_str = fdi.indata;
3357 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3358 	    ap->a_name);
3359 
3360 	err = fdisp_wait_answ(&fdi);
3361 	if (err == ENOSYS) {
3362 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3363 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
3364 		    "removing extended attributes");
3365 	}
3366 
3367 	fdisp_destroy(&fdi);
3368 	return (err);
3369 }
3370 
3371 /*
3372     struct vnop_print_args {
3373 	struct vnode *a_vp;
3374     };
3375 */
3376 static int
fuse_vnop_print(struct vop_print_args * ap)3377 fuse_vnop_print(struct vop_print_args *ap)
3378 {
3379 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3380 
3381 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3382 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3383 	    (uintmax_t)fvdat->nlookup,
3384 	    fvdat->flag);
3385 
3386 	return 0;
3387 }
3388 
3389 /*
3390  * Get an NFS filehandle for a FUSE file.
3391  *
3392  * This will only work for FUSE file systems that guarantee the uniqueness of
3393  * nodeid:generation, which most don't.
3394  */
3395 /*
3396 vop_vptofh {
3397 	IN struct vnode *a_vp;
3398 	IN struct fid *a_fhp;
3399 };
3400 */
3401 static int
fuse_vnop_vptofh(struct vop_vptofh_args * ap)3402 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3403 {
3404 	struct vnode *vp = ap->a_vp;
3405 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
3406 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3407 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3408 		"FUSE fid type is too big");
3409 	struct mount *mp = vnode_mount(vp);
3410 	struct fuse_data *data = fuse_get_mpdata(mp);
3411 	struct vattr va;
3412 	int err;
3413 
3414 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3415 		/* NFS requires lookups for "." and ".." */
3416 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3417 			"VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3418 		return (EXTERROR(EOPNOTSUPP, "This server is "
3419 		    "missing FUSE_EXPORT_SUPPORT"));
3420 	}
3421 	if ((mp->mnt_flag & MNT_EXPORTED) &&
3422 		fsess_is_impl(mp, FUSE_OPENDIR))
3423 	{
3424 		/*
3425 		 * NFS is stateless, so nfsd must reopen a directory on every
3426 		 * call to VOP_READDIR, passing in the d_off field from the
3427 		 * final dirent of the previous invocation.  But if the server
3428 		 * implements FUSE_OPENDIR, the FUSE protocol does not
3429 		 * guarantee that d_off will be valid after a directory is
3430 		 * closed and reopened.  So prohibit exporting FUSE file
3431 		 * systems that implement FUSE_OPENDIR.
3432 		 *
3433 		 * But userspace NFS servers don't have this problem.
3434                  */
3435 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3436 			"VOP_VPTOFH with FUSE_OPENDIR");
3437 		return (EXTERROR(EOPNOTSUPP, "This server implements "
3438 		    "FUSE_OPENDIR so is not compatible with getfh"));
3439 	}
3440 
3441 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3442 	if (err)
3443 		return err;
3444 
3445 	/*ip = VTOI(ap->a_vp);*/
3446 	/*ufhp = (struct ufid *)ap->a_fhp;*/
3447 	fhp->len = sizeof(struct fuse_fid);
3448 	fhp->nid = fvdat->nid;
3449 	if (fvdat->generation <= UINT32_MAX)
3450 		fhp->gen = fvdat->generation;
3451 	else
3452 		return (EXTERROR(EOVERFLOW, "inode generation "
3453 		    "number overflow"));
3454 	return (0);
3455 }
3456