xref: /src/sys/fs/fuse/fuse_vnops.c (revision 7e68af7ce2c1b892954df415774fe59fd2f1b62f)
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)330 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
331 {
332 	struct buf *bp;
333 	daddr_t left_lbn, end_lbn, right_lbn;
334 	off_t new_filesize;
335 	int iosize, left_on, right_on, right_blksize;
336 
337 	iosize = fuse_iosize(vp);
338 	left_lbn = start / iosize;
339 	end_lbn = howmany(end, iosize);
340 	left_on = start & (iosize - 1);
341 	if (left_on != 0) {
342 		bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
343 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
344 			/*
345 			 * Flush the dirty buffer, because we don't have a
346 			 * byte-granular way to record which parts of the
347 			 * buffer are valid.
348 			 */
349 			bwrite(bp);
350 			if (bp->b_error)
351 				return (bp->b_error);
352 		} else {
353 			brelse(bp);
354 		}
355 	}
356 	right_on = end & (iosize - 1);
357 	if (right_on != 0) {
358 		right_lbn = end / iosize;
359 		new_filesize = MAX(filesize, end);
360 		right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
361 		bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
362 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
363 			/*
364 			 * Flush the dirty buffer, because we don't have a
365 			 * byte-granular way to record which parts of the
366 			 * buffer are valid.
367 			 */
368 			bwrite(bp);
369 			if (bp->b_error)
370 				return (bp->b_error);
371 		} else {
372 			brelse(bp);
373 		}
374 	}
375 
376 	v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
377 	return (0);
378 }
379 
380 /* Send FUSE_IOCTL for this node */
381 static int
fuse_vnop_do_ioctl(struct vnode * vp,u_long cmd,void * arg,int fflag,struct ucred * cred,struct thread * td)382 fuse_vnop_do_ioctl(struct vnode *vp, u_long cmd, void *arg, int fflag,
383 	struct ucred *cred, struct thread *td)
384 {
385 	struct fuse_dispatcher fdi;
386 	struct fuse_ioctl_in *fii;
387 	struct fuse_ioctl_out *fio;
388 	struct fuse_filehandle *fufh;
389 	uint32_t flags = 0;
390 	uint32_t insize = 0;
391 	uint32_t outsize = 0;
392 	int err;
393 
394 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, td->td_proc->p_pid);
395 	if (err != 0)
396 		return (err);
397 
398 	if (vnode_isdir(vp)) {
399 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
400 
401 		if (!fuse_libabi_geq(data, 7, 18))
402 			return (ENOTTY);
403 		flags |= FUSE_IOCTL_DIR;
404 	}
405 #ifdef __LP64__
406 #ifdef COMPAT_FREEBSD32
407 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32))
408 		flags |= FUSE_IOCTL_32BIT;
409 #endif
410 #else /* !defined(__LP64__) */
411 	flags |= FUSE_IOCTL_32BIT;
412 #endif
413 
414 	if ((cmd & IOC_OUT) != 0)
415 		outsize = IOCPARM_LEN(cmd);
416 	/* _IOWINT() sets IOC_VOID */
417 	if ((cmd & (IOC_VOID | IOC_IN)) != 0)
418 		insize = IOCPARM_LEN(cmd);
419 
420 	fdisp_init(&fdi, sizeof(*fii) + insize);
421 	fdisp_make_vp(&fdi, FUSE_IOCTL, vp, td, cred);
422 	fii = fdi.indata;
423 	fii->fh = fufh->fh_id;
424 	fii->flags = flags;
425 	fii->cmd = cmd;
426 	fii->arg = (uintptr_t)arg;
427 	fii->in_size = insize;
428 	fii->out_size = outsize;
429 	if (insize > 0)
430 		memcpy((char *)fii + sizeof(*fii), arg, insize);
431 
432 	err = fdisp_wait_answ(&fdi);
433 	if (err != 0) {
434 		if (err == ENOSYS)
435 			err = ENOTTY;
436 		goto out;
437 	}
438 
439 	fio = fdi.answ;
440 	if (fdi.iosize > sizeof(*fio)) {
441 		size_t realoutsize = fdi.iosize - sizeof(*fio);
442 
443 		if (realoutsize > outsize) {
444 			err = EIO;
445 			goto out;
446 		}
447 		memcpy(arg, (char *)fio + sizeof(*fio), realoutsize);
448 	}
449 	if (fio->result > 0)
450 		td->td_retval[0] = fio->result;
451 	else
452 		err = -fio->result;
453 
454 out:
455 	fdisp_destroy(&fdi);
456 	return (err);
457 }
458 
459 /* Send FUSE_LSEEK for this node */
460 static int
fuse_vnop_do_lseek(struct vnode * vp,struct thread * td,struct ucred * cred,pid_t pid,off_t * offp,int whence)461 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
462 	pid_t pid, off_t *offp, int whence)
463 {
464 	struct fuse_dispatcher fdi;
465 	struct fuse_filehandle *fufh;
466 	struct fuse_lseek_in *flsi;
467 	struct fuse_lseek_out *flso;
468 	struct mount *mp = vnode_mount(vp);
469 	int err;
470 
471 	ASSERT_VOP_LOCKED(vp, __func__);
472 
473 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
474 	if (err)
475 		return (err);
476 	fdisp_init(&fdi, sizeof(*flsi));
477 	fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
478 	flsi = fdi.indata;
479 	flsi->fh = fufh->fh_id;
480 	flsi->offset = *offp;
481 	flsi->whence = whence;
482 	err = fdisp_wait_answ(&fdi);
483 	if (err == ENOSYS) {
484 		fsess_set_notimpl(mp, FUSE_LSEEK);
485 	} else if (err == ENXIO) {
486 		/* Note: ENXIO means "no more hole/data regions until EOF" */
487 		fsess_set_impl(mp, FUSE_LSEEK);
488 	} else if (err == 0) {
489 		fsess_set_impl(mp, FUSE_LSEEK);
490 		flso = fdi.answ;
491 		*offp = flso->offset;
492 	}
493 	fdisp_destroy(&fdi);
494 
495 	return (err);
496 }
497 
498 /*
499     struct vnop_access_args {
500 	struct vnode *a_vp;
501 #if VOP_ACCESS_TAKES_ACCMODE_T
502 	accmode_t a_accmode;
503 #else
504 	int a_mode;
505 #endif
506 	struct ucred *a_cred;
507 	struct thread *a_td;
508     };
509 */
510 static int
fuse_vnop_access(struct vop_access_args * ap)511 fuse_vnop_access(struct vop_access_args *ap)
512 {
513 	struct vnode *vp = ap->a_vp;
514 	int accmode = ap->a_accmode;
515 	struct ucred *cred = ap->a_cred;
516 
517 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
518 
519 	int err;
520 
521 	if (fuse_isdeadfs(vp)) {
522 		if (vnode_isvroot(vp)) {
523 			return 0;
524 		}
525 		return (EXTERROR(ENXIO, "This FUSE session is about "
526 		    "to be closed"));
527 	}
528 	if (!(data->dataflags & FSESS_INITED)) {
529 		if (vnode_isvroot(vp)) {
530 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
531 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
532 				return 0;
533 			}
534 		}
535 		return (EXTERROR(EBADF, "Access denied until FUSE session "
536 		    "is initialized"));
537 	}
538 	if (vnode_islnk(vp)) {
539 		return 0;
540 	}
541 
542 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
543 	return err;
544 }
545 
546 /*
547  * struct vop_advlock_args {
548  *	struct vop_generic_args a_gen;
549  *	struct vnode *a_vp;
550  *	void *a_id;
551  *	int a_op;
552  *	struct flock *a_fl;
553  *	int a_flags;
554  * }
555  */
556 static int
fuse_vnop_advlock(struct vop_advlock_args * ap)557 fuse_vnop_advlock(struct vop_advlock_args *ap)
558 {
559 	struct vnode *vp = ap->a_vp;
560 	struct flock *fl = ap->a_fl;
561 	struct thread *td = curthread;
562 	struct ucred *cred = td->td_ucred;
563 	pid_t pid = td->td_proc->p_pid;
564 	struct fuse_filehandle *fufh;
565 	struct fuse_dispatcher fdi;
566 	struct fuse_lk_in *fli;
567 	struct fuse_lk_out *flo;
568 	struct vattr vattr;
569 	enum fuse_opcode op;
570 	off_t size, start;
571 	int dataflags, err;
572 	int flags = ap->a_flags;
573 
574 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
575 
576 	if (fuse_isdeadfs(vp)) {
577 		return (EXTERROR(ENXIO, "This FUSE session is about "
578 		    "to be closed"));
579 	}
580 
581 	switch(ap->a_op) {
582 	case F_GETLK:
583 		op = FUSE_GETLK;
584 		break;
585 	case F_SETLK:
586 		if (flags & F_WAIT)
587 			op = FUSE_SETLKW;
588 		else
589 			op = FUSE_SETLK;
590 		break;
591 	case F_UNLCK:
592 		op = FUSE_SETLK;
593 		break;
594 	default:
595 		return (EXTERROR(EINVAL, "Unsupported lock flags"));
596 	}
597 
598 	if (!(dataflags & FSESS_POSIX_LOCKS))
599 		return vop_stdadvlock(ap);
600 	/* FUSE doesn't properly support flock until protocol 7.17 */
601 	if (flags & F_FLOCK)
602 		return vop_stdadvlock(ap);
603 
604 	vn_lock(vp, LK_SHARED | LK_RETRY);
605 
606 	switch (fl->l_whence) {
607 	case SEEK_SET:
608 	case SEEK_CUR:
609 		/*
610 		 * Caller is responsible for adding any necessary offset
611 		 * when SEEK_CUR is used.
612 		 */
613 		start = fl->l_start;
614 		break;
615 
616 	case SEEK_END:
617 		err = fuse_internal_getattr(vp, &vattr, cred, td);
618 		if (err)
619 			goto out;
620 		size = vattr.va_size;
621 		if (size > OFF_MAX ||
622 		    (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
623 			err = EXTERROR(EOVERFLOW, "Offset is too large");
624 			goto out;
625 		}
626 		start = size + fl->l_start;
627 		break;
628 
629 	default:
630 		return (EXTERROR(EINVAL, "Unsupported offset type"));
631 	}
632 
633 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
634 	if (err)
635 		goto out;
636 
637 	fdisp_init(&fdi, sizeof(*fli));
638 
639 	fdisp_make_vp(&fdi, op, vp, td, cred);
640 	fli = fdi.indata;
641 	fli->fh = fufh->fh_id;
642 	fli->owner = td->td_proc->p_pid;
643 	fli->lk.start = start;
644 	if (fl->l_len != 0)
645 		fli->lk.end = start + fl->l_len - 1;
646 	else
647 		fli->lk.end = INT64_MAX;
648 	fli->lk.type = fl->l_type;
649 	fli->lk.pid = td->td_proc->p_pid;
650 
651 	err = fdisp_wait_answ(&fdi);
652 	fdisp_destroy(&fdi);
653 
654 	if (err == 0 && op == FUSE_GETLK) {
655 		flo = fdi.answ;
656 		fl->l_type = flo->lk.type;
657 		fl->l_whence = SEEK_SET;
658 		if (flo->lk.type != F_UNLCK) {
659 			fl->l_pid = flo->lk.pid;
660 			fl->l_start = flo->lk.start;
661 			if (flo->lk.end == INT64_MAX)
662 				fl->l_len = 0;
663 			else
664 				fl->l_len = flo->lk.end - flo->lk.start + 1;
665 			fl->l_start = flo->lk.start;
666 		}
667 	}
668 
669 out:
670 	VOP_UNLOCK(vp);
671 	return err;
672 }
673 
674 static int
fuse_vnop_allocate(struct vop_allocate_args * ap)675 fuse_vnop_allocate(struct vop_allocate_args *ap)
676 {
677 	struct vnode *vp = ap->a_vp;
678 	off_t *len = ap->a_len;
679 	off_t *offset = ap->a_offset;
680 	struct ucred *cred = ap->a_cred;
681 	struct fuse_filehandle *fufh;
682 	struct mount *mp = vnode_mount(vp);
683 	struct fuse_dispatcher fdi;
684 	struct fuse_fallocate_in *ffi;
685 	struct uio io;
686 	pid_t pid = curthread->td_proc->p_pid;
687 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
688 	off_t filesize;
689 	int err;
690 
691 	if (fuse_isdeadfs(vp))
692 		return (EXTERROR(ENXIO, "This FUSE session is about "
693 		    "to be closed"));
694 
695 	switch (vp->v_type) {
696 	case VFIFO:
697 		return (ESPIPE);
698 	case VLNK:
699 	case VREG:
700 		break;
701 	default:
702 		return (ENODEV);
703 	}
704 
705 	if (vfs_isrdonly(mp))
706 		return (EROFS);
707 
708 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
709 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
710 		    "FUSE_FALLOCATE"));
711 
712 	ASSERT_CACHED_ATTRS_LOCKED(vp);
713 
714 	io.uio_offset = *offset;
715 	io.uio_resid = *len;
716 	err = vn_rlimit_fsize(vp, &io, curthread);
717 	if (err)
718 		return (err);
719 
720 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
721 	if (err)
722 		return (err);
723 
724 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
725 
726 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
727 	if (err)
728 		return (err);
729 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
730 
731 	fdisp_init(&fdi, sizeof(*ffi));
732 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
733 	ffi = fdi.indata;
734 	ffi->fh = fufh->fh_id;
735 	ffi->offset = *offset;
736 	ffi->length = *len;
737 	ffi->mode = 0;
738 	err = fdisp_wait_answ(&fdi);
739 
740 	if (err == ENOSYS) {
741 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
742 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
743 		    "FUSE_ALLOCATE");
744 	} else if (err == EOPNOTSUPP) {
745 		/*
746 		 * The file system server does not support FUSE_FALLOCATE with
747 		 * the supplied mode for this particular file.
748 		 */
749 		err = EXTERROR(EOPNOTSUPP, "This file can't be pre-allocated");
750 	} else if (!err) {
751 		*offset += *len;
752 		*len = 0;
753 		fuse_vnode_undirty_cached_timestamps(vp, false);
754 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
755 		if (*offset > fvdat->cached_attrs.va_size) {
756 			fuse_vnode_setsize(vp, *offset, false);
757 			getnanouptime(&fvdat->last_local_modify);
758 		}
759 	}
760 
761 	fdisp_destroy(&fdi);
762 	return (err);
763 }
764 
765 /* {
766 	struct vnode *a_vp;
767 	daddr_t a_bn;
768 	struct bufobj **a_bop;
769 	daddr_t *a_bnp;
770 	int *a_runp;
771 	int *a_runb;
772 } */
773 static int
fuse_vnop_bmap(struct vop_bmap_args * ap)774 fuse_vnop_bmap(struct vop_bmap_args *ap)
775 {
776 	struct vnode *vp = ap->a_vp;
777 	struct bufobj **bo = ap->a_bop;
778 	struct thread *td = curthread;
779 	struct mount *mp;
780 	struct fuse_dispatcher fdi;
781 	struct fuse_bmap_in *fbi;
782 	struct fuse_bmap_out *fbo;
783 	struct fuse_data *data;
784 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
785 	uint64_t biosize;
786 	off_t fsize = VNOVAL;
787 	daddr_t lbn = ap->a_bn;
788 	daddr_t *pbn = ap->a_bnp;
789 	int *runp = ap->a_runp;
790 	int *runb = ap->a_runb;
791 	int error = 0;
792 	int maxrun;
793 
794 	if (fuse_isdeadfs(vp)) {
795 		return (EXTERROR(ENXIO, "This FUSE session is about "
796 		    "to be closed"));
797 	}
798 
799 	mp = vnode_mount(vp);
800 	data = fuse_get_mpdata(mp);
801 	biosize = fuse_iosize(vp);
802 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
803 		data->max_readahead_blocks);
804 
805 	if (bo != NULL)
806 		*bo = &vp->v_bufobj;
807 
808 	/*
809 	 * The FUSE_BMAP operation does not include the runp and runb
810 	 * variables, so we must guess.  Report nonzero contiguous runs so
811 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
812 	 * upcalls even if we don't know the true physical layout of the file.
813 	 *
814 	 * FUSE file systems may opt out of read clustering in two ways:
815 	 * * mounting with -onoclusterr
816 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
817 	 */
818 	if (runb != NULL)
819 		*runb = MIN(lbn, maxrun);
820 	if (runp != NULL && maxrun == 0)
821 		*runp = 0;
822 	else if (runp != NULL) {
823 		/*
824 		 * If the file's size is cached, use that value to calculate
825 		 * runp, even if the cache is expired.  runp is only advisory,
826 		 * and the risk of getting it wrong is not worth the cost of
827 		 * another upcall.
828 		 */
829 		CACHED_ATTR_LOCK(vp);
830 		fsize = fvdat->cached_attrs.va_size;
831 		CACHED_ATTR_UNLOCK(vp);
832 		if (fsize == VNOVAL)
833 			error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
834 		if (error == 0)
835 			*runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
836 				    maxrun);
837 		else
838 			*runp = 0;
839 	}
840 
841 	if (fsess_maybe_impl(mp, FUSE_BMAP)) {
842 		fdisp_init(&fdi, sizeof(*fbi));
843 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
844 		fbi = fdi.indata;
845 		fbi->block = lbn;
846 		fbi->blocksize = biosize;
847 		error = fdisp_wait_answ(&fdi);
848 		if (error == ENOSYS) {
849 			fdisp_destroy(&fdi);
850 			fsess_set_notimpl(mp, FUSE_BMAP);
851 			error = 0;
852 		} else {
853 			fbo = fdi.answ;
854 			if (error == 0 && pbn != NULL)
855 				*pbn = fbo->block;
856 			fdisp_destroy(&fdi);
857 			return error;
858 		}
859 	}
860 
861 	/* If the daemon doesn't support BMAP, make up a sensible default */
862 	if (pbn != NULL)
863 		*pbn = lbn * btodb(biosize);
864 	return (error);
865 }
866 
867 /*
868     struct vop_close_args {
869 	struct vnode *a_vp;
870 	int  a_fflag;
871 	struct ucred *a_cred;
872 	struct thread *a_td;
873     };
874 */
875 static int
fuse_vnop_close(struct vop_close_args * ap)876 fuse_vnop_close(struct vop_close_args *ap)
877 {
878 	struct vnode *vp = ap->a_vp;
879 	struct mount *mp = vnode_mount(vp);
880 	struct ucred *cred = ap->a_cred;
881 	int fflag = ap->a_fflag;
882 	struct thread *td;
883 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
884 	pid_t pid;
885 	int err = 0;
886 
887 	/* NB: a_td will be NULL from some async kernel contexts */
888 	td = ap->a_td ? ap->a_td : curthread;
889 	pid = td->td_proc->p_pid;
890 
891 	if (fuse_isdeadfs(vp))
892 		return 0;
893 	if (vnode_isdir(vp))
894 		return 0;
895 	if (fflag & IO_NDELAY)
896 		return 0;
897 
898 	if (cred == NULL)
899 		cred = td->td_ucred;
900 
901 	err = fuse_flush(vp, cred, pid, fflag);
902 	ASSERT_CACHED_ATTRS_LOCKED(vp);	/* For fvdat->flag */
903 	if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) {
904 		struct vattr vap;
905 		struct fuse_data *data;
906 		int dataflags;
907 		int access_e = 0;
908 
909 		data = fuse_get_mpdata(mp);
910 		dataflags = data->dataflags;
911 		if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
912 			struct vattr va;
913 
914 			fuse_internal_getattr(vp, &va, cred, td);
915 			access_e = vaccess(vp->v_type, va.va_mode, va.va_uid,
916 			    va.va_gid, VWRITE, cred);
917 		}
918 		if (access_e == 0) {
919 			VATTR_NULL(&vap);
920 			ASSERT_CACHED_ATTRS_LOCKED(vp);
921 			vap.va_atime = fvdat->cached_attrs.va_atime;
922 			/*
923 			 * Ignore errors setting when setting atime.  That
924 			 * should not cause close(2) to fail.
925 			 */
926 			fuse_internal_setattr(vp, &vap, td, NULL);
927 		}
928 	}
929 	/* TODO: close the file handle, if we're sure it's no longer used */
930 	if ((fvdat->flag & FN_SIZECHANGE) != 0) {
931 		fuse_vnode_savesize(vp, cred, pid);
932 	}
933 	return err;
934 }
935 
936 /*
937    struct vop_copy_file_range_args {
938 	struct vop_generic_args a_gen;
939 	struct vnode *a_invp;
940 	off_t *a_inoffp;
941 	struct vnode *a_outvp;
942 	off_t *a_outoffp;
943 	size_t *a_lenp;
944 	unsigned int a_flags;
945 	struct ucred *a_incred;
946 	struct ucred *a_outcred;
947 	struct thread *a_fsizetd;
948 }
949  */
950 static int
fuse_vnop_copy_file_range(struct vop_copy_file_range_args * ap)951 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
952 {
953 	struct vnode *invp = ap->a_invp;
954 	struct vnode *outvp = ap->a_outvp;
955 	struct mount *mp = vnode_mount(invp);
956 	struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
957 	struct fuse_dispatcher fdi;
958 	struct fuse_filehandle *infufh, *outfufh;
959 	struct fuse_copy_file_range_in *fcfri;
960 	struct ucred *incred = ap->a_incred;
961 	struct ucred *outcred = ap->a_outcred;
962 	struct fuse_write_out *fwo;
963 	struct thread *td;
964 	struct uio io;
965 	off_t outfilesize;
966 	ssize_t r = 0;
967 	pid_t pid;
968 	int err;
969 
970 	if ((ap->a_flags & COPY_FILE_RANGE_CLONE) != 0)
971 		return (EXTERROR(ENOSYS, "Cannot clone"));
972 
973 	if (mp == NULL || mp != vnode_mount(outvp))
974 		return (EXTERROR(ENOSYS, "Mount points do not match"));
975 
976 	if (incred->cr_uid != outcred->cr_uid)
977 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
978 		    "support different credentials for infd and outfd"));
979 
980 	if (incred->cr_gid != outcred->cr_gid)
981 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
982 		    "support different credentials for infd and outfd"));
983 
984 	/* Caller busied mp, mnt_data can be safely accessed. */
985 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
986 		return (EXTERROR(ENOSYS, "This daemon does not "
987 		    "implement COPY_FILE_RANGE"));
988 
989 	if (ap->a_fsizetd == NULL)
990 		td = curthread;
991 	else
992 		td = ap->a_fsizetd;
993 	pid = td->td_proc->p_pid;
994 
995 	vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE);
996 	if (invp->v_data == NULL || outvp->v_data == NULL) {
997 		err = EXTERROR(EBADF, "vnode got reclaimed");
998 		goto unlock;
999 	}
1000 
1001 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
1002 	if (err)
1003 		goto unlock;
1004 
1005 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
1006 	if (err)
1007 		goto unlock;
1008 
1009 	io.uio_resid = *ap->a_lenp;
1010 	if (ap->a_fsizetd) {
1011 		io.uio_offset = *ap->a_outoffp;
1012 		err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
1013 		if (err != 0)
1014 			goto unlock;
1015 	}
1016 
1017 	err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
1018 	if (err)
1019 		goto unlock;
1020 
1021 	vnode_pager_clean_sync(invp);
1022 	err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
1023 		*ap->a_outoffp + io.uio_resid);
1024 	if (err)
1025 		goto unlock;
1026 
1027 	fdisp_init(&fdi, sizeof(*fcfri));
1028 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
1029 	fcfri = fdi.indata;
1030 	fcfri->fh_in = infufh->fh_id;
1031 	fcfri->off_in = *ap->a_inoffp;
1032 	fcfri->nodeid_out = VTOI(outvp);
1033 	fcfri->fh_out = outfufh->fh_id;
1034 	fcfri->off_out = *ap->a_outoffp;
1035 	fcfri->len = io.uio_resid;
1036 	fcfri->flags = 0;
1037 
1038 	err = fdisp_wait_answ(&fdi);
1039 	if (err == 0) {
1040 		fwo = fdi.answ;
1041 		*ap->a_lenp = fwo->size;
1042 		*ap->a_inoffp += fwo->size;
1043 		*ap->a_outoffp += fwo->size;
1044 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
1045 		ASSERT_CACHED_ATTRS_LOCKED(outvp);
1046 		if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
1047 			fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
1048 			getnanouptime(&outfvdat->last_local_modify);
1049 		}
1050 		fuse_vnode_update(invp, FN_ATIMECHANGE);
1051 		fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
1052 	}
1053 	fdisp_destroy(&fdi);
1054 
1055 unlock:
1056 	if (invp != outvp)
1057 		VOP_UNLOCK(invp);
1058 	VOP_UNLOCK(outvp);
1059 
1060 	if (err == ENOSYS)
1061 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
1062 
1063 	/*
1064 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
1065 	 * local.
1066 	 */
1067 	return (err);
1068 }
1069 
1070 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)1071 fdisp_make_mknod_for_fallback(
1072 	struct fuse_dispatcher *fdip,
1073 	struct componentname *cnp,
1074 	struct vnode *dvp,
1075 	uint64_t parentnid,
1076 	struct thread *td,
1077 	struct ucred *cred,
1078 	mode_t mode,
1079 	enum fuse_opcode *op)
1080 {
1081 	struct fuse_mknod_in *fmni;
1082 
1083 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
1084 	*op = FUSE_MKNOD;
1085 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
1086 	fmni = fdip->indata;
1087 	fmni->mode = mode;
1088 	fmni->rdev = 0;
1089 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
1090 	    cnp->cn_namelen);
1091 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
1092 }
1093 /*
1094     struct vnop_create_args {
1095 	struct vnode *a_dvp;
1096 	struct vnode **a_vpp;
1097 	struct componentname *a_cnp;
1098 	struct vattr *a_vap;
1099     };
1100 */
1101 static int
fuse_vnop_create(struct vop_create_args * ap)1102 fuse_vnop_create(struct vop_create_args *ap)
1103 {
1104 	struct vnode *dvp = ap->a_dvp;
1105 	struct vnode **vpp = ap->a_vpp;
1106 	struct componentname *cnp = ap->a_cnp;
1107 	struct vattr *vap = ap->a_vap;
1108 	struct thread *td = curthread;
1109 	struct ucred *cred = cnp->cn_cred;
1110 
1111 	struct fuse_data *data;
1112 	struct fuse_create_in *fci;
1113 	struct fuse_entry_out *feo;
1114 	struct fuse_open_out *foo;
1115 	struct fuse_dispatcher fdi, fdi2;
1116 	struct fuse_dispatcher *fdip = &fdi;
1117 	struct fuse_dispatcher *fdip2 = NULL;
1118 
1119 	int err;
1120 
1121 	struct mount *mp = vnode_mount(dvp);
1122 	data = fuse_get_mpdata(mp);
1123 	uint64_t parentnid = VTOFUD(dvp)->nid;
1124 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1125 	enum fuse_opcode op;
1126 	int flags;
1127 
1128 	if (fuse_isdeadfs(dvp))
1129 		return (EXTERROR(ENXIO, "This FUSE session is about "
1130 		    "to be closed"));
1131 
1132 	/* FUSE expects sockets to be created with FUSE_MKNOD */
1133 	if (vap->va_type == VSOCK)
1134 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
1135 
1136 	/*
1137 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
1138 	 * writable mode makes sense, and we might as well include readability
1139 	 * too.
1140 	 */
1141 	flags = O_RDWR;
1142 
1143 	bzero(&fdi, sizeof(fdi));
1144 
1145 	if (vap->va_type != VREG)
1146 		return (EXTERROR(EINVAL, "Only regular files can be created"));
1147 
1148 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1149 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
1150 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1151 			cred, mode, &op);
1152 	} else {
1153 		/* Use FUSE_CREATE */
1154 		size_t insize;
1155 
1156 		op = FUSE_CREATE;
1157 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1158 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1159 		fci = fdip->indata;
1160 		fci->mode = mode;
1161 		fci->flags = O_CREAT | flags;
1162 		if (fuse_libabi_geq(data, 7, 12)) {
1163 			insize = sizeof(*fci);
1164 			fci->umask = td->td_proc->p_pd->pd_cmask;
1165 		} else {
1166 			insize = sizeof(struct fuse_open_in);
1167 		}
1168 
1169 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1170 		    cnp->cn_namelen);
1171 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1172 	}
1173 
1174 	err = fdisp_wait_answ(fdip);
1175 
1176 	if (err) {
1177 		if (err == ENOSYS && op == FUSE_CREATE) {
1178 			fsess_set_notimpl(mp, FUSE_CREATE);
1179 			fdisp_destroy(fdip);
1180 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1181 				parentnid, td, cred, mode, &op);
1182 			err = fdisp_wait_answ(fdip);
1183 		}
1184 		if (err)
1185 			goto out;
1186 	}
1187 
1188 	feo = fdip->answ;
1189 
1190 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1191 		goto out;
1192 	}
1193 
1194 	if (op == FUSE_CREATE) {
1195 		if (fuse_libabi_geq(data, 7, 9))
1196 			foo = (struct fuse_open_out*)(feo + 1);
1197 		else
1198 			foo = (struct fuse_open_out*)((char*)feo +
1199 				FUSE_COMPAT_ENTRY_OUT_SIZE);
1200 	} else {
1201 		/* Issue a separate FUSE_OPEN */
1202 		struct fuse_open_in *foi;
1203 
1204 		fdip2 = &fdi2;
1205 		fdisp_init(fdip2, sizeof(*foi));
1206 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1207 			cred);
1208 		foi = fdip2->indata;
1209 		foi->flags = flags;
1210 		err = fdisp_wait_answ(fdip2);
1211 		if (err)
1212 			goto out;
1213 		foo = fdip2->answ;
1214 	}
1215 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1216 	if (err) {
1217 		struct fuse_release_in *fri;
1218 		uint64_t nodeid = feo->nodeid;
1219 		uint64_t fh_id = foo->fh;
1220 
1221 		fdisp_destroy(fdip);
1222 		fdisp_init(fdip, sizeof(*fri));
1223 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1224 		fri = fdip->indata;
1225 		fri->fh = fh_id;
1226 		fri->flags = flags;
1227 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1228 		fuse_insert_message(fdip->tick, false);
1229 		goto out;
1230 	}
1231 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1232 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1233 		feo->attr_valid_nsec, NULL, true);
1234 
1235 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1236 	fuse_vnode_open(*vpp, foo->open_flags, td);
1237 	/*
1238 	 * Purge the parent's attribute cache because the daemon should've
1239 	 * updated its mtime and ctime
1240 	 */
1241 	fuse_vnode_clear_attr_cache(dvp);
1242 	cache_purge_negative(dvp);
1243 
1244 out:
1245 	if (fdip2)
1246 		fdisp_destroy(fdip2);
1247 	fdisp_destroy(fdip);
1248 	return err;
1249 }
1250 
1251 /*
1252     struct vnop_fdatasync_args {
1253 	struct vop_generic_args a_gen;
1254 	struct vnode * a_vp;
1255 	struct thread * a_td;
1256     };
1257 */
1258 static int
fuse_vnop_fdatasync(struct vop_fdatasync_args * ap)1259 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1260 {
1261 	struct vnode *vp = ap->a_vp;
1262 	struct thread *td = ap->a_td;
1263 	int waitfor = MNT_WAIT;
1264 
1265 	int err = 0;
1266 
1267 	if (fuse_isdeadfs(vp)) {
1268 		return 0;
1269 	}
1270 	if ((err = vop_stdfdatasync_buf(ap)))
1271 		return err;
1272 
1273 	return fuse_internal_fsync(vp, td, waitfor, true);
1274 }
1275 
1276 /*
1277     struct vnop_fsync_args {
1278 	struct vop_generic_args a_gen;
1279 	struct vnode * a_vp;
1280 	int  a_waitfor;
1281 	struct thread * a_td;
1282     };
1283 */
1284 static int
fuse_vnop_fsync(struct vop_fsync_args * ap)1285 fuse_vnop_fsync(struct vop_fsync_args *ap)
1286 {
1287 	struct vnode *vp = ap->a_vp;
1288 	struct thread *td = ap->a_td;
1289 	int waitfor = ap->a_waitfor;
1290 	int err = 0;
1291 
1292 	if (fuse_isdeadfs(vp)) {
1293 		return 0;
1294 	}
1295 	if ((err = vop_stdfsync(ap)))
1296 		return err;
1297 
1298 	return fuse_internal_fsync(vp, td, waitfor, false);
1299 }
1300 
1301 /*
1302     struct vnop_getattr_args {
1303 	struct vnode *a_vp;
1304 	struct vattr *a_vap;
1305 	struct ucred *a_cred;
1306 	struct thread *a_td;
1307     };
1308 */
1309 static int
fuse_vnop_getattr(struct vop_getattr_args * ap)1310 fuse_vnop_getattr(struct vop_getattr_args *ap)
1311 {
1312 	struct vnode *vp = ap->a_vp;
1313 	struct vattr *vap = ap->a_vap;
1314 	struct ucred *cred = ap->a_cred;
1315 	struct thread *td = curthread;
1316 	int err = 0;
1317 
1318 	err = fuse_internal_getattr(vp, vap, cred, td);
1319 	if (err == ENOTCONN && vnode_isvroot(vp)) {
1320 		/*
1321 		 * We want to seem a legitimate fs even if the daemon is dead,
1322 		 * so that, eg., we can still do path based unmounting after
1323 		 * the daemon dies.
1324 		 */
1325 		err = 0;
1326 		bzero(vap, sizeof(*vap));
1327 		vap->va_type = vnode_vtype(vp);
1328 	}
1329 	return err;
1330 }
1331 
1332 /*
1333     struct vnop_inactive_args {
1334 	struct vnode *a_vp;
1335     };
1336 */
1337 static int
fuse_vnop_inactive(struct vop_inactive_args * ap)1338 fuse_vnop_inactive(struct vop_inactive_args *ap)
1339 {
1340 	struct vnode *vp = ap->a_vp;
1341 	struct thread *td = curthread;
1342 
1343 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1344 	struct fuse_filehandle *fufh, *fufh_tmp;
1345 
1346 	int need_flush = 1;
1347 
1348 	ASSERT_CACHED_ATTRS_LOCKED(vp);	/* For fvdat->flag */
1349 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1350 		if (need_flush && vp->v_type == VREG) {
1351 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1352 				fuse_vnode_savesize(vp, NULL, 0);
1353 			}
1354 			if ((fvdat->flag & FN_REVOKED) != 0)
1355 				fuse_io_invalbuf(vp, td);
1356 			else
1357 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1358 			need_flush = 0;
1359 		}
1360 		fuse_filehandle_close(vp, fufh, td, NULL);
1361 	}
1362 
1363 	if ((fvdat->flag & FN_REVOKED) != 0)
1364 		vrecycle(vp);
1365 
1366 	return 0;
1367 }
1368 
1369 /*
1370     struct vnop_ioctl_args {
1371 	struct vnode *a_vp;
1372 	u_long a_command;
1373 	caddr_t a_data;
1374 	int a_fflag;
1375 	struct ucred *a_cred;
1376 	struct thread *a_td;
1377     };
1378 */
1379 static int
fuse_vnop_ioctl(struct vop_ioctl_args * ap)1380 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1381 {
1382 	struct vnode *vp = ap->a_vp;
1383 	struct mount *mp = vnode_mount(vp);
1384 	struct ucred *cred = ap->a_cred;
1385 	struct thread *td = ap->a_td;
1386 	int err;
1387 
1388 	if (fuse_isdeadfs(vp)) {
1389 		return (ENXIO);
1390 	}
1391 
1392 	switch (ap->a_command) {
1393 	case FIOSEEKDATA:
1394 	case FIOSEEKHOLE:
1395 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1396 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1397 			off_t *offp = ap->a_data;
1398 			pid_t pid = td->td_proc->p_pid;
1399 			int whence;
1400 
1401 			if (ap->a_command == FIOSEEKDATA)
1402 				whence = SEEK_DATA;
1403 			else
1404 				whence = SEEK_HOLE;
1405 
1406 			vn_lock(vp, LK_SHARED | LK_RETRY);
1407 			err = fuse_vnop_do_lseek(vp, td, cred, pid, offp,
1408 			    whence);
1409 			VOP_UNLOCK(vp);
1410 		}
1411 		if (fsess_not_impl(mp, FUSE_LSEEK))
1412 			err = vop_stdioctl(ap);
1413 		break;
1414 	default:
1415 		err = fuse_vnop_do_ioctl(vp, ap->a_command, ap->a_data,
1416 		    ap->a_fflag, cred, td);
1417 		break;
1418 	}
1419 	return (err);
1420 }
1421 
1422 
1423 /*
1424     struct vnop_link_args {
1425 	struct vnode *a_tdvp;
1426 	struct vnode *a_vp;
1427 	struct componentname *a_cnp;
1428     };
1429 */
1430 static int
fuse_vnop_link(struct vop_link_args * ap)1431 fuse_vnop_link(struct vop_link_args *ap)
1432 {
1433 	struct vnode *vp = ap->a_vp;
1434 	struct vnode *tdvp = ap->a_tdvp;
1435 	struct componentname *cnp = ap->a_cnp;
1436 
1437 	struct vattr *vap = VTOVA(vp);
1438 
1439 	struct fuse_dispatcher fdi;
1440 	struct fuse_entry_out *feo;
1441 	struct fuse_link_in fli;
1442 
1443 	int err;
1444 
1445 	if (fuse_isdeadfs(vp)) {
1446 		return (EXTERROR(ENXIO, "This FUSE session is about "
1447 		    "to be closed"));
1448 	}
1449 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1450 		return (EXDEV);
1451 	}
1452 
1453 	/*
1454 	 * This is a seatbelt check to protect naive userspace filesystems from
1455 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1456 	 * filesystem does not allow attribute caching, assume it is capable of
1457 	 * validating that nlink does not overflow.
1458 	 */
1459 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1460 		return (EMLINK);
1461 	fli.oldnodeid = VTOI(vp);
1462 
1463 	fdisp_init(&fdi, 0);
1464 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1465 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1466 	if ((err = fdisp_wait_answ(&fdi))) {
1467 		goto out;
1468 	}
1469 	feo = fdi.answ;
1470 
1471 	if (fli.oldnodeid != feo->nodeid) {
1472 		static const char exterr[] = "Server assigned wrong inode "
1473 		    "for a hard link.";
1474 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1475 		fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, exterr);
1476 		fuse_vnode_clear_attr_cache(vp);
1477 		fuse_vnode_clear_attr_cache(tdvp);
1478 		err = EXTERROR(EIO, exterr);
1479 		goto out;
1480 	}
1481 
1482 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1483 	if (!err) {
1484 		/*
1485 		 * Purge the parent's attribute cache because the daemon
1486 		 * should've updated its mtime and ctime
1487 		 */
1488 		fuse_vnode_clear_attr_cache(tdvp);
1489 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1490 			feo->attr_valid_nsec, NULL, true);
1491 	}
1492 out:
1493 	fdisp_destroy(&fdi);
1494 	return err;
1495 }
1496 
1497 struct fuse_lookup_alloc_arg {
1498 	struct fuse_entry_out *feo;
1499 	struct componentname *cnp;
1500 	uint64_t nid;
1501 	__enum_uint8(vtype) vtyp;
1502 };
1503 
1504 /* Callback for vn_get_ino */
1505 static int
fuse_lookup_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)1506 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1507 {
1508 	struct fuse_lookup_alloc_arg *flaa = arg;
1509 
1510 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1511 		flaa->vtyp);
1512 }
1513 
1514 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1515 	"int", "struct timespec*", "struct timespec*");
1516 /*
1517     struct vnop_lookup_args {
1518 	struct vnodeop_desc *a_desc;
1519 	struct vnode *a_dvp;
1520 	struct vnode **a_vpp;
1521 	struct componentname *a_cnp;
1522     };
1523 */
1524 int
fuse_vnop_lookup(struct vop_lookup_args * ap)1525 fuse_vnop_lookup(struct vop_lookup_args *ap)
1526 {
1527 	struct vnode *dvp = ap->a_dvp;
1528 	struct vnode **vpp = ap->a_vpp;
1529 	struct componentname *cnp = ap->a_cnp;
1530 	struct thread *td = curthread;
1531 	struct ucred *cred = cnp->cn_cred;
1532 	struct timespec now;
1533 
1534 	int nameiop = cnp->cn_nameiop;
1535 	bool isdotdot = cnp->cn_flags & ISDOTDOT;
1536 	bool islastcn = cnp->cn_flags & ISLASTCN;
1537 	struct mount *mp = vnode_mount(dvp);
1538 	struct fuse_data *data = fuse_get_mpdata(mp);
1539 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1540 	bool is_dot;
1541 
1542 	int err = 0;
1543 	int lookup_err = 0;
1544 	struct vnode *vp = NULL;
1545 
1546 	struct fuse_dispatcher fdi;
1547 	bool did_lookup = false;
1548 	struct fuse_entry_out *feo = NULL;
1549 	__enum_uint8(vtype) vtyp;	/* vnode type of target */
1550 
1551 	uint64_t nid;
1552 
1553 	if (fuse_isdeadfs(dvp)) {
1554 		*vpp = NULL;
1555 		return (EXTERROR(ENXIO, "This FUSE session is about "
1556 		    "to be closed"));
1557 	}
1558 	if (!vnode_isdir(dvp))
1559 		return ENOTDIR;
1560 
1561 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1562 		return EROFS;
1563 
1564 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1565 		cnp->cn_flags &= ~NOEXECCHECK;
1566 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1567 		return err;
1568 
1569 	ASSERT_CACHED_ATTRS_LOCKED(dvp);	/* For flag */
1570 	is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1571 	if (isdotdot && !(data->dataflags & FSESS_EXPORT_SUPPORT)) {
1572 		if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1573 			/*
1574 			 * Since the file system doesn't support ".." lookups,
1575 			 * we have no way to find this entry.
1576 			 */
1577 			return (EXTERROR(ESTALE, "This server does not support "
1578 			    "'..' lookups"));
1579 		}
1580 		nid = VTOFUD(dvp)->parent_nid;
1581 		if (nid == 0)
1582 			return ENOENT;
1583 		/* .. is obviously a directory */
1584 		vtyp = VDIR;
1585 	} else if (is_dot) {
1586 		nid = VTOI(dvp);
1587 		/* . is obviously a directory */
1588 		vtyp = VDIR;
1589 	} else {
1590 		struct timespec timeout;
1591 		int ncpticks; /* here to accommodate for API contract */
1592 
1593 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1594 		getnanouptime(&now);
1595 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1596 		switch (err) {
1597 		case -1:		/* positive match */
1598 			if (timespeccmp(&timeout, &now, >)) {
1599 				counter_u64_add(fuse_lookup_cache_hits, 1);
1600 			} else {
1601 				/* Cache timeout */
1602 				counter_u64_add(fuse_lookup_cache_misses, 1);
1603 				bintime_clear(
1604 					&VTOFUD(*vpp)->entry_cache_timeout);
1605 				cache_purge(*vpp);
1606 				if (dvp != *vpp)
1607 					vput(*vpp);
1608 				else
1609 					vrele(*vpp);
1610 				*vpp = NULL;
1611 				break;
1612 			}
1613 			return 0;
1614 
1615 		case 0:		/* no match in cache */
1616 			counter_u64_add(fuse_lookup_cache_misses, 1);
1617 			break;
1618 
1619 		case ENOENT:		/* negative match */
1620 			if (timespeccmp(&timeout, &now, <=)) {
1621 				/* Cache timeout */
1622 				cache_purge_negative(dvp);
1623 				break;
1624 			}
1625 			/* fall through */
1626 		default:
1627 			return err;
1628 		}
1629 
1630 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1631 		fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1632 
1633 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1634 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1635 		lookup_err = fdisp_wait_answ(&fdi);
1636 		did_lookup = true;
1637 
1638 		if (!lookup_err) {
1639 			/* lookup call succeeded */
1640 			feo = (struct fuse_entry_out *)fdi.answ;
1641 			nid = feo->nodeid;
1642 			if (nid == 0) {
1643 				/* zero nodeid means ENOENT and cache it */
1644 				struct timespec timeout;
1645 
1646 				fdi.answ_stat = ENOENT;
1647 				lookup_err = ENOENT;
1648 				if (cnp->cn_flags & MAKEENTRY) {
1649 					fuse_validity_2_timespec(feo, &timeout);
1650 					/* Use the same entry_time for .. as for
1651 					 * the file itself.  That doesn't honor
1652 					 * exactly what the fuse server tells
1653 					 * us, but to do otherwise would require
1654 					 * another cache lookup at this point.
1655 					 */
1656 					struct timespec *dtsp = NULL;
1657 					cache_enter_time(dvp, *vpp, cnp,
1658 						&timeout, dtsp);
1659 				}
1660 			}
1661 			vtyp = IFTOVT(feo->attr.mode);
1662 		}
1663 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1664 			fdisp_destroy(&fdi);
1665 			return lookup_err;
1666 		}
1667 	}
1668 	/* lookup_err, if non-zero, must be ENOENT at this point */
1669 
1670 	if (lookup_err) {
1671 		/* Entry not found */
1672 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1673 			if (default_permissions)
1674 				err = fuse_internal_access(dvp, VWRITE, td,
1675 				    cred);
1676 			else
1677 				err = 0;
1678 			if (!err) {
1679 				err = EJUSTRETURN;
1680 			}
1681 		} else {
1682 			err = ENOENT;
1683 		}
1684 	} else {
1685 		/* Entry was found */
1686 		if (isdotdot) {
1687 			struct fuse_lookup_alloc_arg flaa;
1688 
1689 			flaa.nid = nid;
1690 			flaa.feo = feo;
1691 			flaa.cnp = cnp;
1692 			flaa.vtyp = vtyp;
1693 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1694 				&vp);
1695 			*vpp = vp;
1696 		} else if (nid == VTOI(dvp)) {
1697 			if (is_dot) {
1698 				vref(dvp);
1699 				*vpp = dvp;
1700 			} else {
1701 				static const char exterr[] = "Server assigned "
1702 				    "same inode to both parent and child.";
1703 				fuse_warn(fuse_get_mpdata(mp),
1704 				    FSESS_WARN_ILLEGAL_INODE, exterr);
1705 				err = EXTERROR(EIO, exterr);
1706 			}
1707 
1708 		} else {
1709 			struct fuse_vnode_data *fvdat;
1710 
1711 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1712 			    &vp, cnp, vtyp);
1713 			if (err)
1714 				goto out;
1715 			*vpp = vp;
1716 			fvdat = VTOFUD(vp);
1717 
1718 			MPASS(feo != NULL);
1719 			if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1720 				/*
1721 				 * Attributes from the server are definitely
1722 				 * newer than the last attributes we sent to
1723 				 * the server, so cache them.
1724 				 */
1725 				fuse_internal_cache_attrs(*vpp, &feo->attr,
1726 					feo->attr_valid, feo->attr_valid_nsec,
1727 					NULL, true);
1728 			}
1729 			fuse_validity_2_bintime(feo->entry_valid,
1730 				feo->entry_valid_nsec,
1731 				&fvdat->entry_cache_timeout);
1732 
1733 			if ((nameiop == DELETE || nameiop == RENAME) &&
1734 				islastcn && default_permissions)
1735 			{
1736 				struct vattr dvattr;
1737 
1738 				err = fuse_internal_access(dvp, VWRITE, td,
1739 					cred);
1740 				if (err != 0)
1741 					goto out;
1742 				/*
1743 				 * if the parent's sticky bit is set, check
1744 				 * whether we're allowed to remove the file.
1745 				 * Need to figure out the vnode locking to make
1746 				 * this work.
1747 				 */
1748 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1749 				if ((dvattr.va_mode & S_ISTXT) &&
1750 					fuse_internal_access(dvp, VADMIN, td,
1751 						cred) &&
1752 					fuse_internal_access(*vpp, VADMIN, td,
1753 						cred)) {
1754 					err = EPERM;
1755 					goto out;
1756 				}
1757 			}
1758 		}
1759 	}
1760 out:
1761 	if (err) {
1762 		if (vp != NULL && dvp != vp)
1763 			vput(vp);
1764 		else if (vp != NULL)
1765 			vrele(vp);
1766 		*vpp = NULL;
1767 	}
1768 	if (did_lookup)
1769 		fdisp_destroy(&fdi);
1770 
1771 	return err;
1772 }
1773 
1774 /*
1775     struct vnop_mkdir_args {
1776 	struct vnode *a_dvp;
1777 	struct vnode **a_vpp;
1778 	struct componentname *a_cnp;
1779 	struct vattr *a_vap;
1780     };
1781 */
1782 static int
fuse_vnop_mkdir(struct vop_mkdir_args * ap)1783 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1784 {
1785 	struct vnode *dvp = ap->a_dvp;
1786 	struct vnode **vpp = ap->a_vpp;
1787 	struct componentname *cnp = ap->a_cnp;
1788 	struct vattr *vap = ap->a_vap;
1789 
1790 	struct fuse_mkdir_in fmdi;
1791 
1792 	if (fuse_isdeadfs(dvp)) {
1793 		return (EXTERROR(ENXIO, "This FUSE session is about "
1794 		    "to be closed"));
1795 	}
1796 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1797 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1798 
1799 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1800 	    sizeof(fmdi), VDIR));
1801 }
1802 
1803 /*
1804     struct vnop_mknod_args {
1805 	struct vnode *a_dvp;
1806 	struct vnode **a_vpp;
1807 	struct componentname *a_cnp;
1808 	struct vattr *a_vap;
1809     };
1810 */
1811 static int
fuse_vnop_mknod(struct vop_mknod_args * ap)1812 fuse_vnop_mknod(struct vop_mknod_args *ap)
1813 {
1814 
1815 	struct vnode *dvp = ap->a_dvp;
1816 	struct vnode **vpp = ap->a_vpp;
1817 	struct componentname *cnp = ap->a_cnp;
1818 	struct vattr *vap = ap->a_vap;
1819 
1820 	if (fuse_isdeadfs(dvp))
1821 		return (EXTERROR(ENXIO, "This FUSE session is about "
1822 		    "to be closed"));
1823 
1824 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1825 }
1826 
1827 /*
1828     struct vop_open_args {
1829 	struct vnode *a_vp;
1830 	int  a_mode;
1831 	struct ucred *a_cred;
1832 	struct thread *a_td;
1833 	int a_fdidx; / struct file *a_fp;
1834     };
1835 */
1836 static int
fuse_vnop_open(struct vop_open_args * ap)1837 fuse_vnop_open(struct vop_open_args *ap)
1838 {
1839 	struct vnode *vp = ap->a_vp;
1840 	int a_mode = ap->a_mode;
1841 	struct thread *td = ap->a_td;
1842 	struct ucred *cred = ap->a_cred;
1843 	pid_t pid = td->td_proc->p_pid;
1844 
1845 	if (fuse_isdeadfs(vp))
1846 		return (EXTERROR(ENXIO, "This FUSE session is about "
1847 		    "to be closed"));
1848 	if (VN_ISDEV(vp) || vp->v_type == VFIFO)
1849 		return (EXTERROR(EOPNOTSUPP, "Unsupported vnode type",
1850 		    vp->v_type));
1851 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1852 		return (EXTERROR(EINVAL, "Illegal mode", a_mode));
1853 
1854 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1855 		fuse_vnode_open(vp, 0, td);
1856 		return 0;
1857 	}
1858 
1859 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1860 }
1861 
1862 static int
fuse_vnop_pathconf(struct vop_pathconf_args * ap)1863 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1864 {
1865 	struct vnode *vp = ap->a_vp;
1866 	struct mount *mp;
1867 	struct fuse_filehandle *fufh;
1868 	int err;
1869 	bool closefufh = false;
1870 
1871 	switch (ap->a_name) {
1872 	case _PC_FILESIZEBITS:
1873 		*ap->a_retval = 64;
1874 		return (0);
1875 	case _PC_NAME_MAX:
1876 		*ap->a_retval = NAME_MAX;
1877 		return (0);
1878 	case _PC_LINK_MAX:
1879 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1880 		return (0);
1881 	case _PC_SYMLINK_MAX:
1882 		*ap->a_retval = MAXPATHLEN;
1883 		return (0);
1884 	case _PC_NO_TRUNC:
1885 		*ap->a_retval = 1;
1886 		return (0);
1887 	case _PC_MIN_HOLE_SIZE:
1888 		/*
1889 		 * The FUSE protocol provides no mechanism for a server to
1890 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1891 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1892 		 * 1 if it does.
1893 		 */
1894 		mp = vnode_mount(vp);
1895 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1896 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1897 			off_t offset = 0;
1898 
1899 			/*
1900 			 * Issue a FUSE_LSEEK to find out if it's supported.
1901 			 * Use SEEK_DATA instead of SEEK_HOLE, because the
1902 			 * latter generally requires sequential scans of file
1903 			 * metadata, which can be slow.
1904 			 */
1905 			err = fuse_vnop_do_lseek(vp, curthread,
1906 			    curthread->td_ucred, curthread->td_proc->p_pid,
1907 			    &offset, SEEK_DATA);
1908 			if (err == EBADF) {
1909 				/*
1910 				 * pathconf() doesn't necessarily open the
1911 				 * file.  So we may need to do it here.
1912 				 */
1913 				err = fuse_filehandle_open(vp, FREAD, &fufh,
1914 				    curthread, curthread->td_ucred);
1915 				if (err == 0) {
1916 					closefufh = true;
1917 					err = fuse_vnop_do_lseek(vp, curthread,
1918 					    curthread->td_ucred,
1919 					    curthread->td_proc->p_pid, &offset,
1920 					    SEEK_DATA);
1921 				}
1922 				if (closefufh)
1923 					fuse_filehandle_close(vp, fufh,
1924 					    curthread, curthread->td_ucred);
1925 			}
1926 
1927 		}
1928 
1929 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1930 			*ap->a_retval = 1;
1931 			return (0);
1932 		} else if (fsess_not_impl(mp, FUSE_LSEEK)) {
1933 			/* FUSE_LSEEK is not implemented */
1934 			return (EXTERROR(EINVAL, "This server does not "
1935 			    "implement FUSE_LSEEK"));
1936 		} else {
1937 			return (err);
1938 		}
1939 	default:
1940 		return (vop_stdpathconf(ap));
1941 	}
1942 }
1943 
1944 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1945     "struct uio*", "struct ucred*");
1946 /*
1947     struct vnop_read_args {
1948 	struct vnode *a_vp;
1949 	struct uio *a_uio;
1950 	int  a_ioflag;
1951 	struct ucred *a_cred;
1952     };
1953 */
1954 static int
fuse_vnop_read(struct vop_read_args * ap)1955 fuse_vnop_read(struct vop_read_args *ap)
1956 {
1957 	struct vnode *vp = ap->a_vp;
1958 	struct uio *uio = ap->a_uio;
1959 	int ioflag = ap->a_ioflag;
1960 	struct ucred *cred = ap->a_cred;
1961 	pid_t pid = curthread->td_proc->p_pid;
1962 	struct fuse_filehandle *fufh;
1963 	int err;
1964 	bool closefufh = false, directio;
1965 
1966 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1967 
1968 	if (fuse_isdeadfs(vp)) {
1969 		return (EXTERROR(ENXIO, "This FUSE session is about "
1970 		    "to be closed"));
1971 	}
1972 
1973 	/*
1974 	 * XXX Check this flag without the lock.  See
1975 	 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293088
1976 	 */
1977 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1978 		ioflag |= IO_DIRECT;
1979 	}
1980 
1981 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1982 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1983 		/*
1984 		 * nfsd will do I/O without first doing VOP_OPEN.  We
1985 		 * must implicitly open the file here
1986 		 */
1987 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1988 		closefufh = true;
1989 	}
1990 	if (err) {
1991 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1992 		return err;
1993 	}
1994 
1995 	/*
1996          * Ideally, when the daemon asks for direct io at open time, the
1997          * standard file flag should be set according to this, so that would
1998          * just change the default mode, which later on could be changed via
1999          * fcntl(2).
2000          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2001          * (don't know where). So to make any use of the Fuse direct_io option,
2002          * we hardwire it into the file's private data (similarly to Linux,
2003          * btw.).
2004          */
2005 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2006 
2007 	fuse_vnode_update(vp, FN_ATIMECHANGE);
2008 	if (directio) {
2009 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
2010 		err = fuse_read_directbackend(vp, uio, cred, fufh);
2011 	} else {
2012 		SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
2013 		err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
2014 	}
2015 
2016 	if (closefufh)
2017 		fuse_filehandle_close(vp, fufh, curthread, cred);
2018 
2019 	return (err);
2020 }
2021 
2022 /*
2023     struct vnop_readdir_args {
2024 	struct vnode *a_vp;
2025 	struct uio *a_uio;
2026 	struct ucred *a_cred;
2027 	int *a_eofflag;
2028 	int *a_ncookies;
2029 	uint64_t **a_cookies;
2030     };
2031 */
2032 static int
fuse_vnop_readdir(struct vop_readdir_args * ap)2033 fuse_vnop_readdir(struct vop_readdir_args *ap)
2034 {
2035 	struct vnode *vp = ap->a_vp;
2036 	struct uio *uio = ap->a_uio;
2037 	struct ucred *cred = ap->a_cred;
2038 	struct fuse_filehandle *fufh = NULL;
2039 	struct mount *mp = vnode_mount(vp);
2040 	struct fuse_iov cookediov;
2041 	int err = 0;
2042 	uint64_t *cookies;
2043 	ssize_t tresid;
2044 	int ncookies;
2045 	bool closefufh = false;
2046 	pid_t pid = curthread->td_proc->p_pid;
2047 
2048 	if (ap->a_eofflag)
2049 		*ap->a_eofflag = 0;
2050 	if (fuse_isdeadfs(vp)) {
2051 		return (EXTERROR(ENXIO, "This FUSE session is about "
2052 		    "to be closed"));
2053 	}
2054 	if (uio_resid(uio) < sizeof(struct dirent))
2055 		return (EXTERROR(EINVAL, "Buffer is too small"));
2056 
2057 	tresid = uio->uio_resid;
2058 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
2059 	if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
2060 		KASSERT(!fsess_is_impl(mp, FUSE_OPENDIR),
2061 			("FUSE file systems that implement "
2062 			 "FUSE_OPENDIR should not be exported"));
2063 		/*
2064 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
2065 		 * must implicitly open the directory here.
2066 		 */
2067 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
2068 		closefufh = true;
2069 	}
2070 	if (err)
2071 		return (err);
2072 	if (ap->a_ncookies != NULL) {
2073 		ncookies = uio->uio_resid /
2074 			(offsetof(struct dirent, d_name) + 4) + 1;
2075 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
2076 		*ap->a_ncookies = ncookies;
2077 		*ap->a_cookies = cookies;
2078 	} else {
2079 		ncookies = 0;
2080 		cookies = NULL;
2081 	}
2082 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
2083 	fiov_init(&cookediov, DIRCOOKEDSIZE);
2084 
2085 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
2086 		&ncookies, cookies);
2087 
2088 	fiov_teardown(&cookediov);
2089 	if (closefufh)
2090 		fuse_filehandle_close(vp, fufh, curthread, cred);
2091 
2092 	if (ap->a_ncookies != NULL) {
2093 		if (err == 0) {
2094 			*ap->a_ncookies -= ncookies;
2095 		} else {
2096 			free(*ap->a_cookies, M_TEMP);
2097 			*ap->a_ncookies = 0;
2098 			*ap->a_cookies = NULL;
2099 		}
2100 	}
2101 	if (err == 0 && tresid == uio->uio_resid)
2102 		*ap->a_eofflag = 1;
2103 
2104 	return err;
2105 }
2106 
2107 /*
2108     struct vnop_readlink_args {
2109 	struct vnode *a_vp;
2110 	struct uio *a_uio;
2111 	struct ucred *a_cred;
2112     };
2113 */
2114 static int
fuse_vnop_readlink(struct vop_readlink_args * ap)2115 fuse_vnop_readlink(struct vop_readlink_args *ap)
2116 {
2117 	struct vnode *vp = ap->a_vp;
2118 	struct uio *uio = ap->a_uio;
2119 	struct ucred *cred = ap->a_cred;
2120 
2121 	struct fuse_dispatcher fdi;
2122 	int err;
2123 
2124 	if (fuse_isdeadfs(vp)) {
2125 		return (EXTERROR(ENXIO, "This FUSE session is about "
2126 		    "to be closed"));
2127 	}
2128 	if (!vnode_islnk(vp)) {
2129 		return EINVAL;
2130 	}
2131 	fdisp_init(&fdi, 0);
2132 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
2133 	if (err) {
2134 		goto out;
2135 	}
2136 	if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) {
2137 		static const char exterr[] = "Server returned an embedded NUL "
2138 		    "from FUSE_READLINK.";
2139 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
2140 		fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, exterr);
2141 		err = EXTERROR(EIO, exterr);
2142 		goto out;
2143 	}
2144 	if (((char *)fdi.answ)[0] == '/' &&
2145 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
2146 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
2147 
2148 		err = uiomove(mpth, strlen(mpth), uio);
2149 	}
2150 	if (!err) {
2151 		err = uiomove(fdi.answ, fdi.iosize, uio);
2152 	}
2153 out:
2154 	fdisp_destroy(&fdi);
2155 	return err;
2156 }
2157 
2158 /*
2159     struct vnop_reclaim_args {
2160 	struct vnode *a_vp;
2161     };
2162 */
2163 static int
fuse_vnop_reclaim(struct vop_reclaim_args * ap)2164 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2165 {
2166 	struct vnode *vp = ap->a_vp;
2167 	struct thread *td = curthread;
2168 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2169 	struct fuse_filehandle *fufh, *fufh_tmp;
2170 
2171 	if (!fvdat) {
2172 		panic("FUSE: no vnode data during recycling");
2173 	}
2174 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2175 		printf("FUSE: vnode being reclaimed with open fufh "
2176 			"(type=%#x)", fufh->fufh_type);
2177 		fuse_filehandle_close(vp, fufh, td, NULL);
2178 	}
2179 
2180 	if (VTOI(vp) == 1) {
2181 		/*
2182 		 * Don't send FUSE_FORGET for the root inode, because
2183 		 * we never send FUSE_LOOKUP for it (see
2184 		 * fuse_vfsop_root) and we don't want the server to see
2185 		 * mismatched lookup counts.
2186 		 */
2187 		struct fuse_data *data;
2188 		struct vnode *vroot;
2189 
2190 		data = fuse_get_mpdata(vnode_mount(vp));
2191 		FUSE_LOCK();
2192 		vroot = data->vroot;
2193 		data->vroot = NULL;
2194 		FUSE_UNLOCK();
2195 		if (vroot)
2196 			vrele(vroot);
2197 	} else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2198 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2199 		    fvdat->nlookup);
2200 	}
2201 	cache_purge(vp);
2202 	vfs_hash_remove(vp);
2203 	fuse_vnode_destroy(vp);
2204 
2205 	return 0;
2206 }
2207 
2208 /*
2209     struct vnop_remove_args {
2210 	struct vnode *a_dvp;
2211 	struct vnode *a_vp;
2212 	struct componentname *a_cnp;
2213     };
2214 */
2215 static int
fuse_vnop_remove(struct vop_remove_args * ap)2216 fuse_vnop_remove(struct vop_remove_args *ap)
2217 {
2218 	struct vnode *dvp = ap->a_dvp;
2219 	struct vnode *vp = ap->a_vp;
2220 	struct componentname *cnp = ap->a_cnp;
2221 
2222 	int err;
2223 
2224 	if (fuse_isdeadfs(vp)) {
2225 		return (EXTERROR(ENXIO, "This FUSE session is about "
2226 		    "to be closed"));
2227 	}
2228 	if (vnode_isdir(vp)) {
2229 		return (EXTERROR(EPERM, "vnode is a directory"));
2230 	}
2231 
2232 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2233 
2234 	return err;
2235 }
2236 
2237 SDT_PROBE_DEFINE4(fusefs, , vnops, erelookup, "struct vnode*",
2238 	"struct vnode*", "struct vnode*", "struct vnode*");
2239 /*
2240     struct vnop_rename_args {
2241 	struct vnode *a_fdvp;
2242 	struct vnode *a_fvp;
2243 	struct componentname *a_fcnp;
2244 	struct vnode *a_tdvp;
2245 	struct vnode *a_tvp;
2246 	struct componentname *a_tcnp;
2247     };
2248 */
2249 static int
fuse_vnop_rename(struct vop_rename_args * ap)2250 fuse_vnop_rename(struct vop_rename_args *ap)
2251 {
2252 	struct vnode *fdvp = ap->a_fdvp;
2253 	struct vnode *fvp = ap->a_fvp;
2254 	struct componentname *fcnp = ap->a_fcnp;
2255 	struct vnode *tdvp = ap->a_tdvp;
2256 	struct vnode *tvp = ap->a_tvp;
2257 	struct componentname *tcnp = ap->a_tcnp;
2258 	struct fuse_data *data;
2259 	bool newparent = fdvp != tdvp;
2260 	bool isdir = fvp->v_type == VDIR;
2261 	int locktype;
2262 	int err = 0;
2263 
2264 	if (fuse_isdeadfs(fdvp)) {
2265 		return (EXTERROR(ENXIO, "This FUSE session is about "
2266 		    "to be closed"));
2267 	}
2268 	if (fvp->v_mount != tdvp->v_mount ||
2269 	    (tvp && fvp->v_mount != tvp->v_mount)) {
2270 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2271 		err = EXTERROR(EXDEV, "Cross-device rename");
2272 		goto out;
2273 	}
2274 	if (ap->a_flags != 0) {
2275 		err = EOPNOTSUPP;
2276 		goto out;
2277 	}
2278 	cache_purge(fvp);
2279 
2280 	/*
2281 	 * FUSE library is expected to check if target directory is not
2282 	 * under the source directory in the file system tree.
2283 	 * Linux performs this check at VFS level.
2284 	 */
2285 	/*
2286 	 * If source is a directory, and it will get a new parent, user must
2287 	 * have write permission to it, so ".." can be modified.
2288 	 */
2289 	data = fuse_get_mpdata(vnode_mount(tdvp));
2290 
2291 	if (tdvp != fdvp)
2292 		locktype = LK_EXCLUSIVE; /* for fuse_vnode_setparent */
2293 	else
2294 		locktype = LK_SHARED;
2295 
2296 	/*
2297 	 * Must use LK_NOWAIT to prevent LORs between fvp and tdvp or
2298 	 * tvp
2299 	 */
2300 	if (vn_lock(fvp, locktype | LK_NOWAIT) != 0) {
2301 		/*
2302 		 * Can't release tdvp or tvp to try avoiding the LOR.
2303 		 * Must return instead.
2304 		 */
2305 		SDT_PROBE4(fusefs, , vnops, erelookup, fdvp, fvp, tdvp,
2306 			tvp);
2307 		err = ERELOOKUP;
2308 		goto out;
2309 	}
2310 
2311 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2312 		err = fuse_internal_access(fvp, VWRITE,
2313 			curthread, tcnp->cn_cred);
2314 		if (err)
2315 			goto unlock;
2316 	}
2317 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2318 	if (err == 0) {
2319 		if (tdvp != fdvp)
2320 			fuse_vnode_setparent(fvp, tdvp);
2321 		if (tvp != NULL)
2322 			fuse_vnode_setparent(tvp, NULL);
2323 	}
2324 
2325 	if (tvp != NULL && tvp != fvp) {
2326 		cache_purge(tvp);
2327 	}
2328 	if (vnode_isdir(fvp)) {
2329 		if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2330 			cache_purge(tdvp);
2331 		}
2332 		cache_purge(fdvp);
2333 	}
2334 unlock:
2335 	VOP_UNLOCK(fvp);
2336 out:
2337 	if (tdvp == tvp) {
2338 		vrele(tdvp);
2339 	} else {
2340 		vput(tdvp);
2341 	}
2342 	if (tvp != NULL) {
2343 		vput(tvp);
2344 	}
2345 	vrele(fdvp);
2346 	vrele(fvp);
2347 
2348 	return err;
2349 }
2350 
2351 /*
2352     struct vnop_rmdir_args {
2353 	    struct vnode *a_dvp;
2354 	    struct vnode *a_vp;
2355 	    struct componentname *a_cnp;
2356     } *ap;
2357 */
2358 static int
fuse_vnop_rmdir(struct vop_rmdir_args * ap)2359 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2360 {
2361 	struct vnode *dvp = ap->a_dvp;
2362 	struct vnode *vp = ap->a_vp;
2363 
2364 	int err;
2365 
2366 	if (fuse_isdeadfs(vp)) {
2367 		return (EXTERROR(ENXIO, "This FUSE session is about "
2368 		    "to be closed"));
2369 	}
2370 	if (VTOFUD(vp) == VTOFUD(dvp)) {
2371 		return (EXTERROR(EINVAL, "Directory to be removed "
2372 		    "contains itself"));
2373 	}
2374 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2375 
2376 	return err;
2377 }
2378 
2379 /*
2380     struct vnop_setattr_args {
2381 	struct vnode *a_vp;
2382 	struct vattr *a_vap;
2383 	struct ucred *a_cred;
2384 	struct thread *a_td;
2385     };
2386 */
2387 static int
fuse_vnop_setattr(struct vop_setattr_args * ap)2388 fuse_vnop_setattr(struct vop_setattr_args *ap)
2389 {
2390 	struct vnode *vp = ap->a_vp;
2391 	struct vattr *vap = ap->a_vap;
2392 	struct ucred *cred = ap->a_cred;
2393 	struct thread *td = curthread;
2394 	struct mount *mp;
2395 	struct fuse_data *data;
2396 	struct vattr old_va;
2397 	int dataflags;
2398 	int err = 0, err2;
2399 	accmode_t accmode = 0;
2400 	bool checkperm;
2401 	bool drop_suid = false;
2402 
2403 	mp = vnode_mount(vp);
2404 	data = fuse_get_mpdata(mp);
2405 	dataflags = data->dataflags;
2406 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2407 
2408 	if (fuse_isdeadfs(vp)) {
2409 		return (EXTERROR(ENXIO, "This FUSE session is about "
2410 		    "to be closed"));
2411 	}
2412 
2413 	if (vap->va_uid != (uid_t)VNOVAL) {
2414 		if (checkperm) {
2415 			/* Only root may change a file's owner */
2416 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2417 			if (err) {
2418 				/* As a special case, allow the null chown */
2419 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2420 					td);
2421 				if (err2)
2422 					return (err2);
2423 				if (vap->va_uid != old_va.va_uid)
2424 					return err;
2425 				drop_suid = true;
2426 			}
2427 		}
2428 		accmode |= VADMIN;
2429 	}
2430 	if (vap->va_gid != (gid_t)VNOVAL) {
2431 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2432 			drop_suid = true;
2433 		if (checkperm && !groupmember(vap->va_gid, cred)) {
2434 			/*
2435 			 * Non-root users may only chgrp to one of their own
2436 			 * groups
2437 			 */
2438 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2439 			if (err) {
2440 				/* As a special case, allow the null chgrp */
2441 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2442 					td);
2443 				if (err2)
2444 					return (err2);
2445 				if (vap->va_gid != old_va.va_gid)
2446 					return err;
2447 			}
2448 		}
2449 		accmode |= VADMIN;
2450 	}
2451 	if (vap->va_size != VNOVAL) {
2452 		switch (vp->v_type) {
2453 		case VDIR:
2454 			return (EISDIR);
2455 		case VLNK:
2456 		case VREG:
2457 			if (vfs_isrdonly(mp))
2458 				return (EROFS);
2459 			err = vn_rlimit_trunc(vap->va_size, td);
2460 			if (err)
2461 				return (err);
2462 			break;
2463 		default:
2464 			/*
2465 			 * According to POSIX, the result is unspecified
2466 			 * for file types other than regular files,
2467 			 * directories and shared memory objects.  We
2468 			 * don't support shared memory objects in the file
2469 			 * system, and have dubious support for truncating
2470 			 * symlinks.  Just ignore the request in other cases.
2471 			 */
2472 			return (0);
2473 		}
2474 		/* Don't set accmode.  Permission to trunc is checked upstack */
2475 	}
2476 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2477 		if (vap->va_vaflags & VA_UTIMES_NULL)
2478 			accmode |= VWRITE;
2479 		else
2480 			accmode |= VADMIN;
2481 	}
2482 	if (drop_suid) {
2483 		if (vap->va_mode != (mode_t)VNOVAL)
2484 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2485 		else {
2486 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2487 			if (err)
2488 				return (err);
2489 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2490 		}
2491 	}
2492 	if (vap->va_mode != (mode_t)VNOVAL) {
2493 		/* Only root may set the sticky bit on non-directories */
2494 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2495 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2496 			return EFTYPE;
2497 		if (checkperm && (vap->va_mode & S_ISGID)) {
2498 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2499 			if (err)
2500 				return (err);
2501 			if (!groupmember(old_va.va_gid, cred)) {
2502 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2503 				if (err)
2504 					return (err);
2505 			}
2506 		}
2507 		accmode |= VADMIN;
2508 	}
2509 
2510 	if (vfs_isrdonly(mp))
2511 		return EROFS;
2512 
2513 	if (checkperm) {
2514 		err = fuse_internal_access(vp, accmode, td, cred);
2515 	} else {
2516 		err = 0;
2517 	}
2518 	if (err)
2519 		return err;
2520 	else
2521 		return fuse_internal_setattr(vp, vap, td, cred);
2522 }
2523 
2524 /*
2525     struct vnop_strategy_args {
2526 	struct vnode *a_vp;
2527 	struct buf *a_bp;
2528     };
2529 */
2530 static int
fuse_vnop_strategy(struct vop_strategy_args * ap)2531 fuse_vnop_strategy(struct vop_strategy_args *ap)
2532 {
2533 	struct vnode *vp = ap->a_vp;
2534 	struct buf *bp = ap->a_bp;
2535 
2536 	if (!vp || fuse_isdeadfs(vp)) {
2537 		bp->b_ioflags |= BIO_ERROR;
2538 		bp->b_error = ENXIO;
2539 		bufdone(bp);
2540 		return 0;
2541 	}
2542 
2543 	/*
2544 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2545 	 * fuse_io_strategy sets bp's error fields
2546 	 */
2547 	(void)fuse_io_strategy(vp, bp);
2548 
2549 	return 0;
2550 }
2551 
2552 /*
2553     struct vnop_symlink_args {
2554 	struct vnode *a_dvp;
2555 	struct vnode **a_vpp;
2556 	struct componentname *a_cnp;
2557 	struct vattr *a_vap;
2558 	char *a_target;
2559     };
2560 */
2561 static int
fuse_vnop_symlink(struct vop_symlink_args * ap)2562 fuse_vnop_symlink(struct vop_symlink_args *ap)
2563 {
2564 	struct vnode *dvp = ap->a_dvp;
2565 	struct vnode **vpp = ap->a_vpp;
2566 	struct componentname *cnp = ap->a_cnp;
2567 	const char *target = ap->a_target;
2568 
2569 	struct fuse_dispatcher fdi;
2570 
2571 	int err;
2572 	size_t len;
2573 
2574 	if (fuse_isdeadfs(dvp)) {
2575 		return (EXTERROR(ENXIO, "This FUSE session is about "
2576 		    "to be closed"));
2577 	}
2578 	/*
2579 	 * Unlike the other creator type calls, here we have to create a message
2580 	 * where the name of the new entry comes first, and the data describing
2581 	 * the entry comes second.
2582 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2583 	 * but put together the message manually and just call the core part.
2584 	 */
2585 
2586 	len = strlen(target) + 1;
2587 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2588 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2589 
2590 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2591 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2592 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2593 
2594 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2595 	fdisp_destroy(&fdi);
2596 	return err;
2597 }
2598 
2599 /*
2600     struct vnop_write_args {
2601 	struct vnode *a_vp;
2602 	struct uio *a_uio;
2603 	int  a_ioflag;
2604 	struct ucred *a_cred;
2605     };
2606 */
2607 static int
fuse_vnop_write(struct vop_write_args * ap)2608 fuse_vnop_write(struct vop_write_args *ap)
2609 {
2610 	struct vnode *vp = ap->a_vp;
2611 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2612 	struct uio *uio = ap->a_uio;
2613 	int ioflag = ap->a_ioflag;
2614 	struct ucred *cred = ap->a_cred;
2615 	pid_t pid = curthread->td_proc->p_pid;
2616 	struct fuse_filehandle *fufh;
2617 	int err;
2618 	bool closefufh = false, directio;
2619 
2620 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2621 
2622 	if (fuse_isdeadfs(vp)) {
2623 		return (EXTERROR(ENXIO, "This FUSE session is about "
2624 		    "to be closed"));
2625 	}
2626 
2627 	/*
2628 	 * XXX Check this flag without the lock.  See
2629 	 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293088
2630 	 */
2631 	if (fvdat->flag & FN_DIRECTIO)
2632 		ioflag |= IO_DIRECT;
2633 
2634 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2635 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2636 		/*
2637 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2638 		 * must implicitly open the file here
2639 		 */
2640 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2641 		closefufh = true;
2642 	}
2643 	if (err) {
2644 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2645 		return err;
2646 	}
2647 
2648 	/*
2649          * Ideally, when the daemon asks for direct io at open time, the
2650          * standard file flag should be set according to this, so that would
2651          * just change the default mode, which later on could be changed via
2652          * fcntl(2).
2653          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2654          * (don't know where). So to make any use of the Fuse direct_io option,
2655          * we hardwire it into the file's private data (similarly to Linux,
2656          * btw.).
2657          */
2658 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2659 
2660 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2661 	if (directio) {
2662 		off_t start, end, filesize;
2663 		bool pages = (ioflag & IO_VMIO) != 0;
2664 
2665 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2666 
2667 		err = fuse_vnode_size(vp, &filesize, cred, curthread);
2668 		if (err)
2669 			goto out;
2670 
2671 		start = uio->uio_offset;
2672 		end = start + uio->uio_resid;
2673 		if (!pages) {
2674 			err = fuse_inval_buf_range(vp, filesize, start,
2675 			    end);
2676 			if (err)
2677 				goto out;
2678 		}
2679 		err = fuse_write_directbackend(vp, uio, cred, fufh,
2680 			filesize, ioflag, pages);
2681 	} else {
2682 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2683 			"buffered write of vnode");
2684 		if (!fsess_opt_writeback(vnode_mount(vp)))
2685 			ioflag |= IO_SYNC;
2686 		err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2687 	}
2688 	fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2689 
2690 out:
2691 	if (closefufh)
2692 		fuse_filehandle_close(vp, fufh, curthread, cred);
2693 
2694 	return (err);
2695 }
2696 
2697 static daddr_t
fuse_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)2698 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2699 {
2700 	const int biosize = fuse_iosize(vp);
2701 
2702 	return (off / biosize);
2703 }
2704 
2705 static int
fuse_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * blksz)2706 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2707 {
2708 	off_t filesize;
2709 	int err;
2710 	const int biosize = fuse_iosize(vp);
2711 
2712 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2713 	if (err) {
2714 		/* This will turn into a SIGBUS */
2715 		return (EIO);
2716 	} else if ((off_t)lbn * biosize >= filesize) {
2717 		*blksz = 0;
2718 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2719 		*blksz = filesize - (off_t)lbn *biosize;
2720 	} else {
2721 		*blksz = biosize;
2722 	}
2723 	return (0);
2724 }
2725 
2726 /*
2727     struct vnop_getpages_args {
2728 	struct vnode *a_vp;
2729 	vm_page_t *a_m;
2730 	int a_count;
2731 	int a_reqpage;
2732     };
2733 */
2734 static int
fuse_vnop_getpages(struct vop_getpages_args * ap)2735 fuse_vnop_getpages(struct vop_getpages_args *ap)
2736 {
2737 	struct vnode *vp = ap->a_vp;
2738 
2739 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2740 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2741 			"called on non-cacheable vnode??\n");
2742 		return (VM_PAGER_ERROR);
2743 	}
2744 
2745 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2746 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2747 }
2748 
2749 static const char extattr_namespace_separator = '.';
2750 
2751 /*
2752     struct vop_getextattr_args {
2753 	struct vop_generic_args a_gen;
2754 	struct vnode *a_vp;
2755 	int a_attrnamespace;
2756 	const char *a_name;
2757 	struct uio *a_uio;
2758 	size_t *a_size;
2759 	struct ucred *a_cred;
2760 	struct thread *a_td;
2761     };
2762 */
2763 static int
fuse_vnop_getextattr(struct vop_getextattr_args * ap)2764 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2765 {
2766 	struct vnode *vp = ap->a_vp;
2767 	struct uio *uio = ap->a_uio;
2768 	struct fuse_dispatcher fdi;
2769 	struct fuse_getxattr_in *get_xattr_in;
2770 	struct fuse_getxattr_out *get_xattr_out;
2771 	struct mount *mp = vnode_mount(vp);
2772 	struct thread *td = ap->a_td;
2773 	struct ucred *cred = ap->a_cred;
2774 	char *prefix;
2775 	char *attr_str;
2776 	size_t len;
2777 	int err;
2778 
2779 	if (fuse_isdeadfs(vp))
2780 		return (EXTERROR(ENXIO, "This FUSE session is about "
2781 		    "to be closed"));
2782 
2783 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2784 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2785 		    "extended attributes"));
2786 
2787 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2788 	if (err)
2789 		return err;
2790 
2791 	/* Default to looking for user attributes. */
2792 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2793 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2794 	else
2795 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2796 
2797 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2798 	    strlen(ap->a_name) + 1;
2799 
2800 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2801 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2802 
2803 	get_xattr_in = fdi.indata;
2804 	/*
2805 	 * Check to see whether we're querying the available size or
2806 	 * issuing the actual request.  If we pass in 0, we get back struct
2807 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2808 	 * that much data, without the struct fuse_getxattr_out header.
2809 	 */
2810 	if (uio == NULL)
2811 		get_xattr_in->size = 0;
2812 	else
2813 		get_xattr_in->size = uio->uio_resid;
2814 
2815 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2816 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2817 	    ap->a_name);
2818 
2819 	err = fdisp_wait_answ(&fdi);
2820 	if (err != 0) {
2821 		if (err == ENOSYS) {
2822 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2823 			err = (EXTERROR(EOPNOTSUPP, "This server does not "
2824 			    "implement extended attributes"));
2825 		}
2826 		goto out;
2827 	}
2828 
2829 	get_xattr_out = fdi.answ;
2830 
2831 	if (ap->a_size != NULL)
2832 		*ap->a_size = get_xattr_out->size;
2833 
2834 	if (uio != NULL)
2835 		err = uiomove(fdi.answ, fdi.iosize, uio);
2836 
2837 out:
2838 	fdisp_destroy(&fdi);
2839 	return (err);
2840 }
2841 
2842 /*
2843     struct vop_setextattr_args {
2844 	struct vop_generic_args a_gen;
2845 	struct vnode *a_vp;
2846 	int a_attrnamespace;
2847 	const char *a_name;
2848 	struct uio *a_uio;
2849 	struct ucred *a_cred;
2850 	struct thread *a_td;
2851     };
2852 */
2853 static int
fuse_vnop_setextattr(struct vop_setextattr_args * ap)2854 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2855 {
2856 	struct vnode *vp = ap->a_vp;
2857 	struct uio *uio = ap->a_uio;
2858 	struct fuse_dispatcher fdi;
2859 	struct fuse_setxattr_in *set_xattr_in;
2860 	struct mount *mp = vnode_mount(vp);
2861 	struct thread *td = ap->a_td;
2862 	struct ucred *cred = ap->a_cred;
2863 	size_t struct_size = FUSE_COMPAT_SETXATTR_IN_SIZE;
2864 	char *prefix;
2865 	size_t len;
2866 	char *attr_str;
2867 	int err;
2868 
2869 	if (fuse_isdeadfs(vp))
2870 		return (EXTERROR(ENXIO, "This FUSE session is about "
2871 		    "to be closed"));
2872 
2873 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2874 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2875 		    "setting extended attributes"));
2876 
2877 	if (vfs_isrdonly(mp))
2878 		return EROFS;
2879 
2880 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2881 	if (ap->a_uio == NULL) {
2882 		/*
2883 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2884 		 * return EOPNOTSUPP.
2885 		 */
2886 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2887 			return (EXTERROR(EOPNOTSUPP, "This server does not "
2888 			    "implement removing extended attributes"));
2889 		else
2890 			return (EXTERROR(EINVAL, "DELETEEXTATTR should be used "
2891 			    "to remove extattrs"));
2892 	}
2893 
2894 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2895 		VWRITE);
2896 	if (err)
2897 		return err;
2898 
2899 	/* Default to looking for user attributes. */
2900 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2901 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2902 	else
2903 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2904 
2905 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2906 	    strlen(ap->a_name) + 1;
2907 
2908 	/* older FUSE servers  use a smaller fuse_setxattr_in struct*/
2909 	if (fuse_get_mpdata(mp)->dataflags & FSESS_SETXATTR_EXT)
2910 		struct_size = sizeof(*set_xattr_in);
2911 
2912 	fdisp_init(&fdi, len + struct_size + uio->uio_resid);
2913 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2914 
2915 	set_xattr_in = fdi.indata;
2916 	set_xattr_in->size = uio->uio_resid;
2917 
2918 	if (fuse_get_mpdata(mp)->dataflags & FSESS_SETXATTR_EXT) {
2919 		set_xattr_in->setxattr_flags = 0;
2920 		set_xattr_in->padding = 0;
2921 	}
2922 
2923 	attr_str = (char *)fdi.indata + struct_size;
2924 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2925 	    ap->a_name);
2926 
2927 	err = uiomove((char *)fdi.indata + struct_size + len,
2928 	    uio->uio_resid, uio);
2929 	if (err != 0) {
2930 		goto out;
2931 	}
2932 
2933 	err = fdisp_wait_answ(&fdi);
2934 
2935 	if (err == ENOSYS) {
2936 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2937 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
2938 		    "setting extended attributes");
2939 	}
2940 	if (err == ERESTART) {
2941 		/* Can't restart after calling uiomove */
2942 		err = EINTR;
2943 	}
2944 
2945 out:
2946 	fdisp_destroy(&fdi);
2947 	return (err);
2948 }
2949 
2950 /*
2951  * The Linux / FUSE extended attribute list is simply a collection of
2952  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2953  * byte length followed by a non-NUL terminated string.  So, this allows
2954  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2955  * Linux attribute names are reported with the namespace as a prefix (e.g.
2956  * "user.attribute_name"), but in FreeBSD they are reported without the
2957  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2958  *
2959  * user.attr_name1\0user.attr_name2\0
2960  *
2961  * to:
2962  *
2963  * <num>attr_name1<num>attr_name2
2964  *
2965  * Where "<num>" is a single byte number of characters in the attribute name.
2966  *
2967  * Args:
2968  * prefix - exattr namespace prefix string
2969  * list, list_len - input list with namespace prefixes
2970  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2971  */
2972 static int
fuse_xattrlist_convert(char * prefix,const char * list,int list_len,char * bsd_list,int * bsd_list_len)2973 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2974     char *bsd_list, int *bsd_list_len)
2975 {
2976 	int len, pos, dist_to_next, prefix_len;
2977 
2978 	pos = 0;
2979 	*bsd_list_len = 0;
2980 	prefix_len = strlen(prefix);
2981 
2982 	while (pos < list_len && list[pos] != '\0') {
2983 		dist_to_next = strlen(&list[pos]) + 1;
2984 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2985 		    list[pos + prefix_len] == extattr_namespace_separator) {
2986 			len = dist_to_next -
2987 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2988 			if (len >= EXTATTR_MAXNAMELEN)
2989 				return (ENAMETOOLONG);
2990 
2991 			bsd_list[*bsd_list_len] = len;
2992 			memcpy(&bsd_list[*bsd_list_len + 1],
2993 			    &list[pos + prefix_len +
2994 			    sizeof(extattr_namespace_separator)], len);
2995 
2996 			*bsd_list_len += len + 1;
2997 		}
2998 
2999 		pos += dist_to_next;
3000 	}
3001 
3002 	return (0);
3003 }
3004 
3005 /*
3006  * List extended attributes
3007  *
3008  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
3009  * has a number of differences compared to its FreeBSD equivalent,
3010  * extattr_list_file:
3011  *
3012  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
3013  *   whereas listxattr(2) only returns attributes for a single namespace
3014  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
3015  * - If the provided buffer is not large enough to hold the result,
3016  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
3017  *   return as many results as will fit.
3018  */
3019 /*
3020     struct vop_listextattr_args {
3021 	struct vop_generic_args a_gen;
3022 	struct vnode *a_vp;
3023 	int a_attrnamespace;
3024 	struct uio *a_uio;
3025 	size_t *a_size;
3026 	struct ucred *a_cred;
3027 	struct thread *a_td;
3028     };
3029 */
3030 static int
fuse_vnop_listextattr(struct vop_listextattr_args * ap)3031 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
3032 {
3033 	struct vnode *vp = ap->a_vp;
3034 	struct uio *uio = ap->a_uio;
3035 	struct fuse_dispatcher fdi;
3036 	struct fuse_listxattr_in *list_xattr_in;
3037 	struct fuse_listxattr_out *list_xattr_out;
3038 	struct mount *mp = vnode_mount(vp);
3039 	struct thread *td = ap->a_td;
3040 	struct ucred *cred = ap->a_cred;
3041 	char *prefix;
3042 	char *bsd_list = NULL;
3043 	char *linux_list;
3044 	int bsd_list_len;
3045 	int linux_list_len;
3046 	int err;
3047 
3048 	if (fuse_isdeadfs(vp))
3049 		return (EXTERROR(ENXIO, "This FUSE session is about "
3050 		    "to be closed"));
3051 
3052 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
3053 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3054 		    "extended attributes"));
3055 
3056 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
3057 	if (err)
3058 		return err;
3059 
3060 	/*
3061 	 * Add space for a NUL and the period separator if enabled.
3062 	 * Default to looking for user attributes.
3063 	 */
3064 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3065 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3066 	else
3067 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3068 
3069 	fdisp_init(&fdi, sizeof(*list_xattr_in));
3070 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
3071 
3072 	/*
3073 	 * Retrieve Linux / FUSE compatible list size.
3074 	 */
3075 	list_xattr_in = fdi.indata;
3076 	list_xattr_in->size = 0;
3077 
3078 	err = fdisp_wait_answ(&fdi);
3079 	if (err != 0) {
3080 		if (err == ENOSYS) {
3081 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
3082 			err = EXTERROR(EOPNOTSUPP, "This server does not "
3083 			    "implement extended attributes");
3084 		}
3085 		goto out;
3086 	}
3087 
3088 	list_xattr_out = fdi.answ;
3089 	linux_list_len = list_xattr_out->size;
3090 	if (linux_list_len == 0) {
3091 		if (ap->a_size != NULL)
3092 			*ap->a_size = linux_list_len;
3093 		goto out;
3094 	}
3095 
3096 	/*
3097 	 * Retrieve Linux / FUSE compatible list values.
3098 	 */
3099 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
3100 	list_xattr_in = fdi.indata;
3101 	list_xattr_in->size = linux_list_len;
3102 
3103 	err = fdisp_wait_answ(&fdi);
3104 	if (err == ERANGE) {
3105 		/*
3106 		 * Race detected.  The attribute list must've grown since the
3107 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
3108 		 * to userland so we can process signals, if necessary, before
3109 		 * restarting.
3110 		 */
3111 		err = ERESTART;
3112 		goto out;
3113 	} else if (err != 0)
3114 		goto out;
3115 
3116 	linux_list = fdi.answ;
3117 	/* FUSE doesn't allow the server to return more data than requested */
3118 	if (fdi.iosize > linux_list_len) {
3119 		struct fuse_data *data = fuse_get_mpdata(mp);
3120 
3121 		fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
3122 			"server returned "
3123 			"more extended attribute data than requested; "
3124 			"should've returned ERANGE instead.");
3125 	} else {
3126 		/* But returning less data is fine */
3127 		linux_list_len = fdi.iosize;
3128 	}
3129 
3130 	/*
3131 	 * Retrieve the BSD compatible list values.
3132 	 * The Linux / FUSE attribute list format isn't the same
3133 	 * as FreeBSD's format. So we need to transform it into
3134 	 * FreeBSD's format before giving it to the user.
3135 	 */
3136 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
3137 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
3138 	    bsd_list, &bsd_list_len);
3139 	if (err != 0)
3140 		goto out;
3141 
3142 	if (ap->a_size != NULL)
3143 		*ap->a_size = bsd_list_len;
3144 
3145 	if (uio != NULL)
3146 		err = uiomove(bsd_list, bsd_list_len, uio);
3147 
3148 out:
3149 	free(bsd_list, M_TEMP);
3150 	fdisp_destroy(&fdi);
3151 	return (err);
3152 }
3153 
3154 /*
3155     struct vop_deallocate_args {
3156 	struct vop_generic_args a_gen;
3157 	struct vnode *a_vp;
3158 	off_t *a_offset;
3159 	off_t *a_len;
3160 	int a_flags;
3161 	int a_ioflag;
3162         struct ucred *a_cred;
3163     };
3164 */
3165 static int
fuse_vnop_deallocate(struct vop_deallocate_args * ap)3166 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
3167 {
3168 	struct vnode *vp = ap->a_vp;
3169 	struct mount *mp = vnode_mount(vp);
3170 	struct fuse_filehandle *fufh;
3171 	struct fuse_dispatcher fdi;
3172 	struct fuse_fallocate_in *ffi;
3173 	struct ucred *cred = ap->a_cred;
3174 	pid_t pid = curthread->td_proc->p_pid;
3175 	off_t *len = ap->a_len;
3176 	off_t *offset = ap->a_offset;
3177 	int ioflag = ap->a_ioflag;
3178 	off_t filesize;
3179 	int err;
3180 	bool closefufh = false;
3181 
3182 	if (fuse_isdeadfs(vp))
3183 		return (EXTERROR(ENXIO, "This FUSE session is about "
3184 		    "to be closed"));
3185 
3186 	if (vfs_isrdonly(mp))
3187 		return (EROFS);
3188 
3189 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
3190 		goto fallback;
3191 
3192 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
3193 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
3194 		/*
3195 		 * nfsd will do I/O without first doing VOP_OPEN.  We
3196 		 * must implicitly open the file here
3197 		 */
3198 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
3199 		closefufh = true;
3200 	}
3201 	if (err)
3202 		return (err);
3203 
3204 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3205 
3206 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
3207 	if (err)
3208 		goto out;
3209 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3210 
3211 	fdisp_init(&fdi, sizeof(*ffi));
3212 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3213 	ffi = fdi.indata;
3214 	ffi->fh = fufh->fh_id;
3215 	ffi->offset = *offset;
3216 	ffi->length = *len;
3217 	/*
3218 	 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3219 	 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3220 	 */
3221 	ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3222 	err = fdisp_wait_answ(&fdi);
3223 
3224 	if (err == ENOSYS) {
3225 		fdisp_destroy(&fdi);
3226 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
3227 		goto fallback;
3228 	} else if (err == EOPNOTSUPP) {
3229 		/*
3230 		 * The file system server does not support FUSE_FALLOCATE with
3231 		 * the supplied mode for this particular file.
3232 		 */
3233 		fdisp_destroy(&fdi);
3234 		goto fallback;
3235 	} else if (!err) {
3236 		/*
3237 		 * Clip the returned offset to EoF.  Do it here rather than
3238 		 * before FUSE_FALLOCATE just in case the kernel's cached file
3239 		 * size is out of date.  Unfortunately, FUSE does not return
3240 		 * any information about filesize from that operation.
3241 		 */
3242 		*offset = MIN(*offset + *len, filesize);
3243 		*len = 0;
3244 		fuse_vnode_undirty_cached_timestamps(vp, false);
3245 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
3246 
3247 		if (ioflag & IO_SYNC)
3248 			err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3249 			    false);
3250 	}
3251 
3252 	fdisp_destroy(&fdi);
3253 out:
3254 	if (closefufh)
3255 		fuse_filehandle_close(vp, fufh, curthread, cred);
3256 
3257 	return (err);
3258 
3259 fallback:
3260 	if (closefufh)
3261 		fuse_filehandle_close(vp, fufh, curthread, cred);
3262 
3263 	return (vop_stddeallocate(ap));
3264 }
3265 
3266 /*
3267    struct vop_delayed_setsize_args {
3268 	struct vop_generic_args a_gen;
3269 	struct vnode *a_vp;
3270   };
3271  */
3272 static int
fuse_vnop_delayed_setsize(struct vop_delayed_setsize_args * ap)3273 fuse_vnop_delayed_setsize(struct vop_delayed_setsize_args *ap)
3274 {
3275 	struct vnode *vp = ap->a_vp;
3276 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3277 	bool shrink = (fvdat->flag & FN_DELAYED_TRUNCATE) != 0;
3278 	int err;
3279 
3280 	if (!fvdat)
3281 		return (0);
3282 
3283 	err = fuse_vnode_setsize_immediate(vp, shrink);
3284 	fvdat->flag &= ~FN_DELAYED_TRUNCATE;
3285 
3286 	return (err);
3287 }
3288 
3289 /*
3290     struct vop_deleteextattr_args {
3291 	struct vop_generic_args a_gen;
3292 	struct vnode *a_vp;
3293 	int a_attrnamespace;
3294 	const char *a_name;
3295 	struct ucred *a_cred;
3296 	struct thread *a_td;
3297     };
3298 */
3299 static int
fuse_vnop_deleteextattr(struct vop_deleteextattr_args * ap)3300 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3301 {
3302 	struct vnode *vp = ap->a_vp;
3303 	struct fuse_dispatcher fdi;
3304 	struct mount *mp = vnode_mount(vp);
3305 	struct thread *td = ap->a_td;
3306 	struct ucred *cred = ap->a_cred;
3307 	char *prefix;
3308 	size_t len;
3309 	char *attr_str;
3310 	int err;
3311 
3312 	if (fuse_isdeadfs(vp))
3313 		return (EXTERROR(ENXIO, "This FUSE session is about "
3314 		    "to be closed"));
3315 
3316 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3317 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3318 		    "removing extended attributes"));
3319 
3320 	if (vfs_isrdonly(mp))
3321 		return EROFS;
3322 
3323 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3324 		VWRITE);
3325 	if (err)
3326 		return err;
3327 
3328 	/* Default to looking for user attributes. */
3329 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3330 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3331 	else
3332 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3333 
3334 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3335 	    strlen(ap->a_name) + 1;
3336 
3337 	fdisp_init(&fdi, len);
3338 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3339 
3340 	attr_str = fdi.indata;
3341 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3342 	    ap->a_name);
3343 
3344 	err = fdisp_wait_answ(&fdi);
3345 	if (err == ENOSYS) {
3346 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3347 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
3348 		    "removing extended attributes");
3349 	}
3350 
3351 	fdisp_destroy(&fdi);
3352 	return (err);
3353 }
3354 
3355 /*
3356     struct vnop_print_args {
3357 	struct vnode *a_vp;
3358     };
3359 */
3360 static int
fuse_vnop_print(struct vop_print_args * ap)3361 fuse_vnop_print(struct vop_print_args *ap)
3362 {
3363 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3364 
3365 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3366 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3367 	    (uintmax_t)fvdat->nlookup,
3368 	    fvdat->flag);
3369 
3370 	return 0;
3371 }
3372 
3373 /*
3374  * Get an NFS filehandle for a FUSE file.
3375  *
3376  * This will only work for FUSE file systems that guarantee the uniqueness of
3377  * nodeid:generation, which most don't.
3378  */
3379 /*
3380 vop_vptofh {
3381 	IN struct vnode *a_vp;
3382 	IN struct fid *a_fhp;
3383 };
3384 */
3385 static int
fuse_vnop_vptofh(struct vop_vptofh_args * ap)3386 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3387 {
3388 	struct vnode *vp = ap->a_vp;
3389 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
3390 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3391 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3392 		"FUSE fid type is too big");
3393 	struct mount *mp = vnode_mount(vp);
3394 	struct fuse_data *data = fuse_get_mpdata(mp);
3395 	struct vattr va;
3396 	int err;
3397 
3398 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3399 		/* NFS requires lookups for "." and ".." */
3400 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3401 			"VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3402 		return (EXTERROR(EOPNOTSUPP, "This server is "
3403 		    "missing FUSE_EXPORT_SUPPORT"));
3404 	}
3405 	if ((mp->mnt_flag & MNT_EXPORTED) &&
3406 		fsess_is_impl(mp, FUSE_OPENDIR))
3407 	{
3408 		/*
3409 		 * NFS is stateless, so nfsd must reopen a directory on every
3410 		 * call to VOP_READDIR, passing in the d_off field from the
3411 		 * final dirent of the previous invocation.  But if the server
3412 		 * implements FUSE_OPENDIR, the FUSE protocol does not
3413 		 * guarantee that d_off will be valid after a directory is
3414 		 * closed and reopened.  So prohibit exporting FUSE file
3415 		 * systems that implement FUSE_OPENDIR.
3416 		 *
3417 		 * But userspace NFS servers don't have this problem.
3418                  */
3419 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3420 			"VOP_VPTOFH with FUSE_OPENDIR");
3421 		return (EXTERROR(EOPNOTSUPP, "This server implements "
3422 		    "FUSE_OPENDIR so is not compatible with getfh"));
3423 	}
3424 
3425 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3426 	if (err)
3427 		return err;
3428 
3429 	/*ip = VTOI(ap->a_vp);*/
3430 	/*ufhp = (struct ufid *)ap->a_fhp;*/
3431 	fhp->len = sizeof(struct fuse_fid);
3432 	fhp->nid = fvdat->nid;
3433 	if (fvdat->generation <= UINT32_MAX)
3434 		fhp->gen = fvdat->generation;
3435 	else
3436 		return (EXTERROR(EOVERFLOW, "inode generation "
3437 		    "number overflow"));
3438 	return (0);
3439 }
3440