xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 567e6250c003eeb251b4bc8dbe60d2adabab2988)
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
93 #include <sys/exterrvar.h>
94 
95 #include <vm/vm.h>
96 #include <vm/vm_extern.h>
97 #include <vm/pmap.h>
98 #include <vm/vm_map.h>
99 #include <vm/vm_page.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_pager.h>
103 #include <vm/vnode_pager.h>
104 #include <vm/vm_object.h>
105 
106 #include "fuse.h"
107 #include "fuse_file.h"
108 #include "fuse_internal.h"
109 #include "fuse_ipc.h"
110 #include "fuse_node.h"
111 #include "fuse_io.h"
112 
113 #include <sys/priv.h>
114 
115 /* Maximum number of hardlinks to a single FUSE file */
116 #define FUSE_LINK_MAX                      UINT32_MAX
117 
118 SDT_PROVIDER_DECLARE(fusefs);
119 /*
120  * Fuse trace probe:
121  * arg0: verbosity.  Higher numbers give more verbose messages
122  * arg1: Textual message
123  */
124 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
125 
126 /* vnode ops */
127 static vop_access_t fuse_vnop_access;
128 static vop_advlock_t fuse_vnop_advlock;
129 static vop_allocate_t fuse_vnop_allocate;
130 static vop_bmap_t fuse_vnop_bmap;
131 static vop_close_t fuse_fifo_close;
132 static vop_close_t fuse_vnop_close;
133 static vop_copy_file_range_t fuse_vnop_copy_file_range;
134 static vop_create_t fuse_vnop_create;
135 static vop_deallocate_t fuse_vnop_deallocate;
136 static vop_deleteextattr_t fuse_vnop_deleteextattr;
137 static vop_fdatasync_t fuse_vnop_fdatasync;
138 static vop_fsync_t fuse_vnop_fsync;
139 static vop_getattr_t fuse_vnop_getattr;
140 static vop_getextattr_t fuse_vnop_getextattr;
141 static vop_inactive_t fuse_vnop_inactive;
142 static vop_ioctl_t fuse_vnop_ioctl;
143 static vop_link_t fuse_vnop_link;
144 static vop_listextattr_t fuse_vnop_listextattr;
145 static vop_lookup_t fuse_vnop_lookup;
146 static vop_mkdir_t fuse_vnop_mkdir;
147 static vop_mknod_t fuse_vnop_mknod;
148 static vop_open_t fuse_vnop_open;
149 static vop_pathconf_t fuse_vnop_pathconf;
150 static vop_read_t fuse_vnop_read;
151 static vop_readdir_t fuse_vnop_readdir;
152 static vop_readlink_t fuse_vnop_readlink;
153 static vop_reclaim_t fuse_vnop_reclaim;
154 static vop_remove_t fuse_vnop_remove;
155 static vop_rename_t fuse_vnop_rename;
156 static vop_rmdir_t fuse_vnop_rmdir;
157 static vop_setattr_t fuse_vnop_setattr;
158 static vop_setextattr_t fuse_vnop_setextattr;
159 static vop_strategy_t fuse_vnop_strategy;
160 static vop_symlink_t fuse_vnop_symlink;
161 static vop_write_t fuse_vnop_write;
162 static vop_getpages_t fuse_vnop_getpages;
163 static vop_print_t fuse_vnop_print;
164 static vop_vptofh_t fuse_vnop_vptofh;
165 
166 struct vop_vector fuse_fifoops = {
167 	.vop_default =		&fifo_specops,
168 	.vop_access =		fuse_vnop_access,
169 	.vop_close =		fuse_fifo_close,
170 	.vop_fsync =		fuse_vnop_fsync,
171 	.vop_getattr =		fuse_vnop_getattr,
172 	.vop_inactive =		fuse_vnop_inactive,
173 	.vop_pathconf =		fuse_vnop_pathconf,
174 	.vop_print =		fuse_vnop_print,
175 	.vop_read =		VOP_PANIC,
176 	.vop_reclaim =		fuse_vnop_reclaim,
177 	.vop_setattr =		fuse_vnop_setattr,
178 	.vop_write =		VOP_PANIC,
179 	.vop_vptofh =		fuse_vnop_vptofh,
180 };
181 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
182 
183 struct vop_vector fuse_vnops = {
184 	.vop_allocate =	fuse_vnop_allocate,
185 	.vop_default = &default_vnodeops,
186 	.vop_access = fuse_vnop_access,
187 	.vop_advlock = fuse_vnop_advlock,
188 	.vop_bmap = fuse_vnop_bmap,
189 	.vop_close = fuse_vnop_close,
190 	.vop_copy_file_range = fuse_vnop_copy_file_range,
191 	.vop_create = fuse_vnop_create,
192 	.vop_deallocate = fuse_vnop_deallocate,
193 	.vop_deleteextattr = fuse_vnop_deleteextattr,
194 	.vop_fsync = fuse_vnop_fsync,
195 	.vop_fdatasync = fuse_vnop_fdatasync,
196 	.vop_getattr = fuse_vnop_getattr,
197 	.vop_getextattr = fuse_vnop_getextattr,
198 	.vop_inactive = fuse_vnop_inactive,
199 	.vop_ioctl = fuse_vnop_ioctl,
200 	.vop_link = fuse_vnop_link,
201 	.vop_listextattr = fuse_vnop_listextattr,
202 	.vop_lookup = fuse_vnop_lookup,
203 	.vop_mkdir = fuse_vnop_mkdir,
204 	.vop_mknod = fuse_vnop_mknod,
205 	.vop_open = fuse_vnop_open,
206 	.vop_pathconf = fuse_vnop_pathconf,
207 	/*
208 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
209 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
210 	 * 7.21, which adds the ability for the client to choose which poll
211 	 * events it wants, and for a client to deregister a file handle
212 	 */
213 	.vop_read = fuse_vnop_read,
214 	.vop_readdir = fuse_vnop_readdir,
215 	.vop_readlink = fuse_vnop_readlink,
216 	.vop_reclaim = fuse_vnop_reclaim,
217 	.vop_remove = fuse_vnop_remove,
218 	.vop_rename = fuse_vnop_rename,
219 	.vop_rmdir = fuse_vnop_rmdir,
220 	.vop_setattr = fuse_vnop_setattr,
221 	.vop_setextattr = fuse_vnop_setextattr,
222 	.vop_strategy = fuse_vnop_strategy,
223 	.vop_symlink = fuse_vnop_symlink,
224 	.vop_write = fuse_vnop_write,
225 	.vop_getpages = fuse_vnop_getpages,
226 	.vop_print = fuse_vnop_print,
227 	.vop_vptofh = fuse_vnop_vptofh,
228 };
229 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
230 
231 /* Check permission for extattr operations, much like extattr_check_cred */
232 static int
fuse_extattr_check_cred(struct vnode * vp,int ns,struct ucred * cred,struct thread * td,accmode_t accmode)233 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
234 	struct thread *td, accmode_t accmode)
235 {
236 	struct mount *mp = vnode_mount(vp);
237 	struct fuse_data *data = fuse_get_mpdata(mp);
238 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
239 
240 	/*
241 	 * Kernel-invoked always succeeds.
242 	 */
243 	if (cred == NOCRED)
244 		return (0);
245 
246 	/*
247 	 * Do not allow privileged processes in jail to directly manipulate
248 	 * system attributes.
249 	 */
250 	switch (ns) {
251 	case EXTATTR_NAMESPACE_SYSTEM:
252 		if (default_permissions) {
253 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
254 		}
255 		return (0);
256 	case EXTATTR_NAMESPACE_USER:
257 		if (default_permissions) {
258 			return (fuse_internal_access(vp, accmode, td, cred));
259 		}
260 		return (0);
261 	default:
262 		return (EPERM);
263 	}
264 }
265 
266 /* Get a filehandle for a directory */
267 static int
fuse_filehandle_get_dir(struct vnode * vp,struct fuse_filehandle ** fufhp,struct ucred * cred,pid_t pid)268 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
269 	struct ucred *cred, pid_t pid)
270 {
271 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
272 		return 0;
273 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
274 }
275 
276 /* Send FUSE_FLUSH for this vnode */
277 static int
fuse_flush(struct vnode * vp,struct ucred * cred,pid_t pid,int fflag)278 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
279 {
280 	struct fuse_flush_in *ffi;
281 	struct fuse_filehandle *fufh;
282 	struct fuse_dispatcher fdi;
283 	struct thread *td = curthread;
284 	struct mount *mp = vnode_mount(vp);
285 	int err;
286 
287 	if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
288 		return 0;
289 
290 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
291 	if (err)
292 		return err;
293 
294 	if (fufh->fuse_open_flags & FOPEN_NOFLUSH &&
295 	    (!fsess_opt_writeback(vnode_mount(vp))))
296 		return (0);
297 
298 	fdisp_init(&fdi, sizeof(*ffi));
299 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
300 	ffi = fdi.indata;
301 	ffi->fh = fufh->fh_id;
302 	/*
303 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
304 	 * If not, then lock_owner is undefined.  So we may as well always set
305 	 * it.
306 	 */
307 	ffi->lock_owner = td->td_proc->p_pid;
308 
309 	err = fdisp_wait_answ(&fdi);
310 	if (err == ENOSYS) {
311 		fsess_set_notimpl(mp, FUSE_FLUSH);
312 		err = 0;
313 	}
314 	fdisp_destroy(&fdi);
315 	return err;
316 }
317 
318 /* Close wrapper for fifos.  */
319 static int
fuse_fifo_close(struct vop_close_args * ap)320 fuse_fifo_close(struct vop_close_args *ap)
321 {
322 	return (fifo_specops.vop_close(ap));
323 }
324 
325 /* Invalidate a range of cached data, whether dirty of not */
326 static int
fuse_inval_buf_range(struct vnode * vp,off_t filesize,off_t start,off_t end)327 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
328 {
329 	struct buf *bp;
330 	daddr_t left_lbn, end_lbn, right_lbn;
331 	off_t new_filesize;
332 	int iosize, left_on, right_on, right_blksize;
333 
334 	iosize = fuse_iosize(vp);
335 	left_lbn = start / iosize;
336 	end_lbn = howmany(end, iosize);
337 	left_on = start & (iosize - 1);
338 	if (left_on != 0) {
339 		bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
340 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
341 			/*
342 			 * Flush the dirty buffer, because we don't have a
343 			 * byte-granular way to record which parts of the
344 			 * buffer are valid.
345 			 */
346 			bwrite(bp);
347 			if (bp->b_error)
348 				return (bp->b_error);
349 		} else {
350 			brelse(bp);
351 		}
352 	}
353 	right_on = end & (iosize - 1);
354 	if (right_on != 0) {
355 		right_lbn = end / iosize;
356 		new_filesize = MAX(filesize, end);
357 		right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
358 		bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
359 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
360 			/*
361 			 * Flush the dirty buffer, because we don't have a
362 			 * byte-granular way to record which parts of the
363 			 * buffer are valid.
364 			 */
365 			bwrite(bp);
366 			if (bp->b_error)
367 				return (bp->b_error);
368 		} else {
369 			brelse(bp);
370 		}
371 	}
372 
373 	v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
374 	return (0);
375 }
376 
377 
378 /* Send FUSE_LSEEK for this node */
379 static int
fuse_vnop_do_lseek(struct vnode * vp,struct thread * td,struct ucred * cred,pid_t pid,off_t * offp,int whence)380 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
381 	pid_t pid, off_t *offp, int whence)
382 {
383 	struct fuse_dispatcher fdi;
384 	struct fuse_filehandle *fufh;
385 	struct fuse_lseek_in *flsi;
386 	struct fuse_lseek_out *flso;
387 	struct mount *mp = vnode_mount(vp);
388 	int err;
389 
390 	ASSERT_VOP_LOCKED(vp, __func__);
391 
392 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
393 	if (err)
394 		return (err);
395 	fdisp_init(&fdi, sizeof(*flsi));
396 	fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
397 	flsi = fdi.indata;
398 	flsi->fh = fufh->fh_id;
399 	flsi->offset = *offp;
400 	flsi->whence = whence;
401 	err = fdisp_wait_answ(&fdi);
402 	if (err == ENOSYS) {
403 		fsess_set_notimpl(mp, FUSE_LSEEK);
404 	} else if (err == ENXIO) {
405 		/* Note: ENXIO means "no more hole/data regions until EOF" */
406 		fsess_set_impl(mp, FUSE_LSEEK);
407 	} else if (err == 0) {
408 		fsess_set_impl(mp, FUSE_LSEEK);
409 		flso = fdi.answ;
410 		*offp = flso->offset;
411 	}
412 	fdisp_destroy(&fdi);
413 
414 	return (err);
415 }
416 
417 /*
418     struct vnop_access_args {
419 	struct vnode *a_vp;
420 #if VOP_ACCESS_TAKES_ACCMODE_T
421 	accmode_t a_accmode;
422 #else
423 	int a_mode;
424 #endif
425 	struct ucred *a_cred;
426 	struct thread *a_td;
427     };
428 */
429 static int
fuse_vnop_access(struct vop_access_args * ap)430 fuse_vnop_access(struct vop_access_args *ap)
431 {
432 	struct vnode *vp = ap->a_vp;
433 	int accmode = ap->a_accmode;
434 	struct ucred *cred = ap->a_cred;
435 
436 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
437 
438 	int err;
439 
440 	if (fuse_isdeadfs(vp)) {
441 		if (vnode_isvroot(vp)) {
442 			return 0;
443 		}
444 		return (EXTERROR(ENXIO, "This FUSE session is about "
445 		    "to be closed"));
446 	}
447 	if (!(data->dataflags & FSESS_INITED)) {
448 		if (vnode_isvroot(vp)) {
449 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
450 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
451 				return 0;
452 			}
453 		}
454 		return (EXTERROR(EBADF, "Access denied until FUSE session "
455 		    "is initialized"));
456 	}
457 	if (vnode_islnk(vp)) {
458 		return 0;
459 	}
460 
461 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
462 	return err;
463 }
464 
465 /*
466  * struct vop_advlock_args {
467  *	struct vop_generic_args a_gen;
468  *	struct vnode *a_vp;
469  *	void *a_id;
470  *	int a_op;
471  *	struct flock *a_fl;
472  *	int a_flags;
473  * }
474  */
475 static int
fuse_vnop_advlock(struct vop_advlock_args * ap)476 fuse_vnop_advlock(struct vop_advlock_args *ap)
477 {
478 	struct vnode *vp = ap->a_vp;
479 	struct flock *fl = ap->a_fl;
480 	struct thread *td = curthread;
481 	struct ucred *cred = td->td_ucred;
482 	pid_t pid = td->td_proc->p_pid;
483 	struct fuse_filehandle *fufh;
484 	struct fuse_dispatcher fdi;
485 	struct fuse_lk_in *fli;
486 	struct fuse_lk_out *flo;
487 	struct vattr vattr;
488 	enum fuse_opcode op;
489 	off_t size, start;
490 	int dataflags, err;
491 	int flags = ap->a_flags;
492 
493 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
494 
495 	if (fuse_isdeadfs(vp)) {
496 		return (EXTERROR(ENXIO, "This FUSE session is about "
497 		    "to be closed"));
498 	}
499 
500 	switch(ap->a_op) {
501 	case F_GETLK:
502 		op = FUSE_GETLK;
503 		break;
504 	case F_SETLK:
505 		if (flags & F_WAIT)
506 			op = FUSE_SETLKW;
507 		else
508 			op = FUSE_SETLK;
509 		break;
510 	case F_UNLCK:
511 		op = FUSE_SETLK;
512 		break;
513 	default:
514 		return (EXTERROR(EINVAL, "Unsupported lock flags"));
515 	}
516 
517 	if (!(dataflags & FSESS_POSIX_LOCKS))
518 		return vop_stdadvlock(ap);
519 	/* FUSE doesn't properly support flock until protocol 7.17 */
520 	if (flags & F_FLOCK)
521 		return vop_stdadvlock(ap);
522 
523 	vn_lock(vp, LK_SHARED | LK_RETRY);
524 
525 	switch (fl->l_whence) {
526 	case SEEK_SET:
527 	case SEEK_CUR:
528 		/*
529 		 * Caller is responsible for adding any necessary offset
530 		 * when SEEK_CUR is used.
531 		 */
532 		start = fl->l_start;
533 		break;
534 
535 	case SEEK_END:
536 		err = fuse_internal_getattr(vp, &vattr, cred, td);
537 		if (err)
538 			goto out;
539 		size = vattr.va_size;
540 		if (size > OFF_MAX ||
541 		    (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
542 			err = EXTERROR(EOVERFLOW, "Offset is too large");
543 			goto out;
544 		}
545 		start = size + fl->l_start;
546 		break;
547 
548 	default:
549 		return (EXTERROR(EINVAL, "Unsupported offset type"));
550 	}
551 
552 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
553 	if (err)
554 		goto out;
555 
556 	fdisp_init(&fdi, sizeof(*fli));
557 
558 	fdisp_make_vp(&fdi, op, vp, td, cred);
559 	fli = fdi.indata;
560 	fli->fh = fufh->fh_id;
561 	fli->owner = td->td_proc->p_pid;
562 	fli->lk.start = start;
563 	if (fl->l_len != 0)
564 		fli->lk.end = start + fl->l_len - 1;
565 	else
566 		fli->lk.end = INT64_MAX;
567 	fli->lk.type = fl->l_type;
568 	fli->lk.pid = td->td_proc->p_pid;
569 
570 	err = fdisp_wait_answ(&fdi);
571 	fdisp_destroy(&fdi);
572 
573 	if (err == 0 && op == FUSE_GETLK) {
574 		flo = fdi.answ;
575 		fl->l_type = flo->lk.type;
576 		fl->l_whence = SEEK_SET;
577 		if (flo->lk.type != F_UNLCK) {
578 			fl->l_pid = flo->lk.pid;
579 			fl->l_start = flo->lk.start;
580 			if (flo->lk.end == INT64_MAX)
581 				fl->l_len = 0;
582 			else
583 				fl->l_len = flo->lk.end - flo->lk.start + 1;
584 			fl->l_start = flo->lk.start;
585 		}
586 	}
587 
588 out:
589 	VOP_UNLOCK(vp);
590 	return err;
591 }
592 
593 static int
fuse_vnop_allocate(struct vop_allocate_args * ap)594 fuse_vnop_allocate(struct vop_allocate_args *ap)
595 {
596 	struct vnode *vp = ap->a_vp;
597 	off_t *len = ap->a_len;
598 	off_t *offset = ap->a_offset;
599 	struct ucred *cred = ap->a_cred;
600 	struct fuse_filehandle *fufh;
601 	struct mount *mp = vnode_mount(vp);
602 	struct fuse_dispatcher fdi;
603 	struct fuse_fallocate_in *ffi;
604 	struct uio io;
605 	pid_t pid = curthread->td_proc->p_pid;
606 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
607 	off_t filesize;
608 	int err;
609 
610 	if (fuse_isdeadfs(vp))
611 		return (EXTERROR(ENXIO, "This FUSE session is about "
612 		    "to be closed"));
613 
614 	switch (vp->v_type) {
615 	case VFIFO:
616 		return (ESPIPE);
617 	case VLNK:
618 	case VREG:
619 		break;
620 	default:
621 		return (ENODEV);
622 	}
623 
624 	if (vfs_isrdonly(mp))
625 		return (EROFS);
626 
627 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
628 		return (EXTERROR(EINVAL, "This server does not implement "
629 		    "FUSE_FALLOCATE"));
630 
631 	io.uio_offset = *offset;
632 	io.uio_resid = *len;
633 	err = vn_rlimit_fsize(vp, &io, curthread);
634 	if (err)
635 		return (err);
636 
637 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
638 	if (err)
639 		return (err);
640 
641 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
642 
643 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
644 	if (err)
645 		return (err);
646 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
647 
648 	fdisp_init(&fdi, sizeof(*ffi));
649 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
650 	ffi = fdi.indata;
651 	ffi->fh = fufh->fh_id;
652 	ffi->offset = *offset;
653 	ffi->length = *len;
654 	ffi->mode = 0;
655 	err = fdisp_wait_answ(&fdi);
656 
657 	if (err == ENOSYS) {
658 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
659 		err = EXTERROR(EINVAL, "This server does not implement "
660 		    "FUSE_ALLOCATE");
661 	} else if (err == EOPNOTSUPP) {
662 		/*
663 		 * The file system server does not support FUSE_FALLOCATE with
664 		 * the supplied mode for this particular file.
665 		 */
666 		err = EXTERROR(EINVAL, "This file can't be pre-allocated");
667 	} else if (!err) {
668 		*offset += *len;
669 		*len = 0;
670 		fuse_vnode_undirty_cached_timestamps(vp, false);
671 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
672 		if (*offset > fvdat->cached_attrs.va_size) {
673 			fuse_vnode_setsize(vp, *offset, false);
674 			getnanouptime(&fvdat->last_local_modify);
675 		}
676 	}
677 
678 	fdisp_destroy(&fdi);
679 	return (err);
680 }
681 
682 /* {
683 	struct vnode *a_vp;
684 	daddr_t a_bn;
685 	struct bufobj **a_bop;
686 	daddr_t *a_bnp;
687 	int *a_runp;
688 	int *a_runb;
689 } */
690 static int
fuse_vnop_bmap(struct vop_bmap_args * ap)691 fuse_vnop_bmap(struct vop_bmap_args *ap)
692 {
693 	struct vnode *vp = ap->a_vp;
694 	struct bufobj **bo = ap->a_bop;
695 	struct thread *td = curthread;
696 	struct mount *mp;
697 	struct fuse_dispatcher fdi;
698 	struct fuse_bmap_in *fbi;
699 	struct fuse_bmap_out *fbo;
700 	struct fuse_data *data;
701 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
702 	uint64_t biosize;
703 	off_t fsize;
704 	daddr_t lbn = ap->a_bn;
705 	daddr_t *pbn = ap->a_bnp;
706 	int *runp = ap->a_runp;
707 	int *runb = ap->a_runb;
708 	int error = 0;
709 	int maxrun;
710 
711 	if (fuse_isdeadfs(vp)) {
712 		return (EXTERROR(ENXIO, "This FUSE session is about "
713 		    "to be closed"));
714 	}
715 
716 	mp = vnode_mount(vp);
717 	data = fuse_get_mpdata(mp);
718 	biosize = fuse_iosize(vp);
719 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
720 		data->max_readahead_blocks);
721 
722 	if (bo != NULL)
723 		*bo = &vp->v_bufobj;
724 
725 	/*
726 	 * The FUSE_BMAP operation does not include the runp and runb
727 	 * variables, so we must guess.  Report nonzero contiguous runs so
728 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
729 	 * upcalls even if we don't know the true physical layout of the file.
730 	 *
731 	 * FUSE file systems may opt out of read clustering in two ways:
732 	 * * mounting with -onoclusterr
733 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
734 	 */
735 	if (runb != NULL)
736 		*runb = MIN(lbn, maxrun);
737 	if (runp != NULL && maxrun == 0)
738 		*runp = 0;
739 	else if (runp != NULL) {
740 		/*
741 		 * If the file's size is cached, use that value to calculate
742 		 * runp, even if the cache is expired.  runp is only advisory,
743 		 * and the risk of getting it wrong is not worth the cost of
744 		 * another upcall.
745 		 */
746 		if (fvdat->cached_attrs.va_size != VNOVAL)
747 			fsize = fvdat->cached_attrs.va_size;
748 		else
749 			error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
750 		if (error == 0)
751 			*runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
752 				    maxrun);
753 		else
754 			*runp = 0;
755 	}
756 
757 	if (fsess_maybe_impl(mp, FUSE_BMAP)) {
758 		fdisp_init(&fdi, sizeof(*fbi));
759 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
760 		fbi = fdi.indata;
761 		fbi->block = lbn;
762 		fbi->blocksize = biosize;
763 		error = fdisp_wait_answ(&fdi);
764 		if (error == ENOSYS) {
765 			fdisp_destroy(&fdi);
766 			fsess_set_notimpl(mp, FUSE_BMAP);
767 			error = 0;
768 		} else {
769 			fbo = fdi.answ;
770 			if (error == 0 && pbn != NULL)
771 				*pbn = fbo->block;
772 			fdisp_destroy(&fdi);
773 			return error;
774 		}
775 	}
776 
777 	/* If the daemon doesn't support BMAP, make up a sensible default */
778 	if (pbn != NULL)
779 		*pbn = lbn * btodb(biosize);
780 	return (error);
781 }
782 
783 /*
784     struct vop_close_args {
785 	struct vnode *a_vp;
786 	int  a_fflag;
787 	struct ucred *a_cred;
788 	struct thread *a_td;
789     };
790 */
791 static int
fuse_vnop_close(struct vop_close_args * ap)792 fuse_vnop_close(struct vop_close_args *ap)
793 {
794 	struct vnode *vp = ap->a_vp;
795 	struct mount *mp = vnode_mount(vp);
796 	struct ucred *cred = ap->a_cred;
797 	int fflag = ap->a_fflag;
798 	struct thread *td = ap->a_td;
799 	pid_t pid = td->td_proc->p_pid;
800 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
801 	int err = 0;
802 
803 	if (fuse_isdeadfs(vp))
804 		return 0;
805 	if (vnode_isdir(vp))
806 		return 0;
807 	if (fflag & IO_NDELAY)
808 		return 0;
809 
810 	if (cred == NULL)
811 		cred = td->td_ucred;
812 
813 	err = fuse_flush(vp, cred, pid, fflag);
814 	if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) {
815 		struct vattr vap;
816 		struct fuse_data *data;
817 		int dataflags;
818 		int access_e = 0;
819 
820 		data = fuse_get_mpdata(mp);
821 		dataflags = data->dataflags;
822 		if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
823 			struct vattr va;
824 
825 			fuse_internal_getattr(vp, &va, cred, td);
826 			access_e = vaccess(vp->v_type, va.va_mode, va.va_uid,
827 			    va.va_gid, VWRITE, cred);
828 		}
829 		if (access_e == 0) {
830 			VATTR_NULL(&vap);
831 			vap.va_atime = fvdat->cached_attrs.va_atime;
832 			/*
833 			 * Ignore errors setting when setting atime.  That
834 			 * should not cause close(2) to fail.
835 			 */
836 			fuse_internal_setattr(vp, &vap, td, NULL);
837 		}
838 	}
839 	/* TODO: close the file handle, if we're sure it's no longer used */
840 	if ((fvdat->flag & FN_SIZECHANGE) != 0) {
841 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
842 	}
843 	return err;
844 }
845 
846 /*
847    struct vop_copy_file_range_args {
848 	struct vop_generic_args a_gen;
849 	struct vnode *a_invp;
850 	off_t *a_inoffp;
851 	struct vnode *a_outvp;
852 	off_t *a_outoffp;
853 	size_t *a_lenp;
854 	unsigned int a_flags;
855 	struct ucred *a_incred;
856 	struct ucred *a_outcred;
857 	struct thread *a_fsizetd;
858 }
859  */
860 static int
fuse_vnop_copy_file_range(struct vop_copy_file_range_args * ap)861 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
862 {
863 	struct vnode *invp = ap->a_invp;
864 	struct vnode *outvp = ap->a_outvp;
865 	struct mount *mp = vnode_mount(invp);
866 	struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
867 	struct fuse_dispatcher fdi;
868 	struct fuse_filehandle *infufh, *outfufh;
869 	struct fuse_copy_file_range_in *fcfri;
870 	struct ucred *incred = ap->a_incred;
871 	struct ucred *outcred = ap->a_outcred;
872 	struct fuse_write_out *fwo;
873 	struct thread *td;
874 	struct uio io;
875 	off_t outfilesize;
876 	ssize_t r = 0;
877 	pid_t pid;
878 	int err;
879 
880 	if ((ap->a_flags & COPY_FILE_RANGE_CLONE) != 0)
881 		return (EXTERROR(ENOSYS, "Cannot clone"));
882 
883 	if (mp == NULL || mp != vnode_mount(outvp))
884 		return (EXTERROR(ENOSYS, "Mount points do not match"));
885 
886 	if (incred->cr_uid != outcred->cr_uid)
887 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
888 		    "support different credentials for infd and outfd"));
889 
890 	if (incred->cr_gid != outcred->cr_gid)
891 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
892 		    "support different credentials for infd and outfd"));
893 
894 	/* Caller busied mp, mnt_data can be safely accessed. */
895 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
896 		return (EXTERROR(ENOSYS, "This daemon does not "
897 		    "implement COPY_FILE_RANGE"));
898 
899 	if (ap->a_fsizetd == NULL)
900 		td = curthread;
901 	else
902 		td = ap->a_fsizetd;
903 	pid = td->td_proc->p_pid;
904 
905 	vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE);
906 	if (invp->v_data == NULL || outvp->v_data == NULL) {
907 		err = EXTERROR(EBADF, "vnode got reclaimed");
908 		goto unlock;
909 	}
910 
911 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
912 	if (err)
913 		goto unlock;
914 
915 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
916 	if (err)
917 		goto unlock;
918 
919 	io.uio_resid = *ap->a_lenp;
920 	if (ap->a_fsizetd) {
921 		io.uio_offset = *ap->a_outoffp;
922 		err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
923 		if (err != 0)
924 			goto unlock;
925 	}
926 
927 	err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
928 	if (err)
929 		goto unlock;
930 
931 	vnode_pager_clean_sync(invp);
932 	err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
933 		*ap->a_outoffp + io.uio_resid);
934 	if (err)
935 		goto unlock;
936 
937 	fdisp_init(&fdi, sizeof(*fcfri));
938 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
939 	fcfri = fdi.indata;
940 	fcfri->fh_in = infufh->fh_id;
941 	fcfri->off_in = *ap->a_inoffp;
942 	fcfri->nodeid_out = VTOI(outvp);
943 	fcfri->fh_out = outfufh->fh_id;
944 	fcfri->off_out = *ap->a_outoffp;
945 	fcfri->len = io.uio_resid;
946 	fcfri->flags = 0;
947 
948 	err = fdisp_wait_answ(&fdi);
949 	if (err == 0) {
950 		fwo = fdi.answ;
951 		*ap->a_lenp = fwo->size;
952 		*ap->a_inoffp += fwo->size;
953 		*ap->a_outoffp += fwo->size;
954 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
955 		if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
956                         fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
957 			getnanouptime(&outfvdat->last_local_modify);
958 		}
959 		fuse_vnode_update(invp, FN_ATIMECHANGE);
960 		fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
961 	}
962 	fdisp_destroy(&fdi);
963 
964 unlock:
965 	if (invp != outvp)
966 		VOP_UNLOCK(invp);
967 	VOP_UNLOCK(outvp);
968 
969 	if (err == ENOSYS)
970 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
971 
972 	/*
973 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
974 	 * local.
975 	 */
976 	return (err);
977 }
978 
979 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)980 fdisp_make_mknod_for_fallback(
981 	struct fuse_dispatcher *fdip,
982 	struct componentname *cnp,
983 	struct vnode *dvp,
984 	uint64_t parentnid,
985 	struct thread *td,
986 	struct ucred *cred,
987 	mode_t mode,
988 	enum fuse_opcode *op)
989 {
990 	struct fuse_mknod_in *fmni;
991 
992 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
993 	*op = FUSE_MKNOD;
994 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
995 	fmni = fdip->indata;
996 	fmni->mode = mode;
997 	fmni->rdev = 0;
998 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
999 	    cnp->cn_namelen);
1000 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
1001 }
1002 /*
1003     struct vnop_create_args {
1004 	struct vnode *a_dvp;
1005 	struct vnode **a_vpp;
1006 	struct componentname *a_cnp;
1007 	struct vattr *a_vap;
1008     };
1009 */
1010 static int
fuse_vnop_create(struct vop_create_args * ap)1011 fuse_vnop_create(struct vop_create_args *ap)
1012 {
1013 	struct vnode *dvp = ap->a_dvp;
1014 	struct vnode **vpp = ap->a_vpp;
1015 	struct componentname *cnp = ap->a_cnp;
1016 	struct vattr *vap = ap->a_vap;
1017 	struct thread *td = curthread;
1018 	struct ucred *cred = cnp->cn_cred;
1019 
1020 	struct fuse_data *data;
1021 	struct fuse_create_in *fci;
1022 	struct fuse_entry_out *feo;
1023 	struct fuse_open_out *foo;
1024 	struct fuse_dispatcher fdi, fdi2;
1025 	struct fuse_dispatcher *fdip = &fdi;
1026 	struct fuse_dispatcher *fdip2 = NULL;
1027 
1028 	int err;
1029 
1030 	struct mount *mp = vnode_mount(dvp);
1031 	data = fuse_get_mpdata(mp);
1032 	uint64_t parentnid = VTOFUD(dvp)->nid;
1033 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1034 	enum fuse_opcode op;
1035 	int flags;
1036 
1037 	if (fuse_isdeadfs(dvp))
1038 		return (EXTERROR(ENXIO, "This FUSE session is about "
1039 		    "to be closed"));
1040 
1041 	/* FUSE expects sockets to be created with FUSE_MKNOD */
1042 	if (vap->va_type == VSOCK)
1043 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
1044 
1045 	/*
1046 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
1047 	 * writable mode makes sense, and we might as well include readability
1048 	 * too.
1049 	 */
1050 	flags = O_RDWR;
1051 
1052 	bzero(&fdi, sizeof(fdi));
1053 
1054 	if (vap->va_type != VREG)
1055 		return (EXTERROR(EINVAL, "Only regular files can be created"));
1056 
1057 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1058 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
1059 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1060 			cred, mode, &op);
1061 	} else {
1062 		/* Use FUSE_CREATE */
1063 		size_t insize;
1064 
1065 		op = FUSE_CREATE;
1066 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1067 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1068 		fci = fdip->indata;
1069 		fci->mode = mode;
1070 		fci->flags = O_CREAT | flags;
1071 		if (fuse_libabi_geq(data, 7, 12)) {
1072 			insize = sizeof(*fci);
1073 			fci->umask = td->td_proc->p_pd->pd_cmask;
1074 		} else {
1075 			insize = sizeof(struct fuse_open_in);
1076 		}
1077 
1078 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1079 		    cnp->cn_namelen);
1080 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1081 	}
1082 
1083 	err = fdisp_wait_answ(fdip);
1084 
1085 	if (err) {
1086 		if (err == ENOSYS && op == FUSE_CREATE) {
1087 			fsess_set_notimpl(mp, FUSE_CREATE);
1088 			fdisp_destroy(fdip);
1089 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1090 				parentnid, td, cred, mode, &op);
1091 			err = fdisp_wait_answ(fdip);
1092 		}
1093 		if (err)
1094 			goto out;
1095 	}
1096 
1097 	feo = fdip->answ;
1098 
1099 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1100 		goto out;
1101 	}
1102 
1103 	if (op == FUSE_CREATE) {
1104 		if (fuse_libabi_geq(data, 7, 9))
1105 			foo = (struct fuse_open_out*)(feo + 1);
1106 		else
1107 			foo = (struct fuse_open_out*)((char*)feo +
1108 				FUSE_COMPAT_ENTRY_OUT_SIZE);
1109 	} else {
1110 		/* Issue a separate FUSE_OPEN */
1111 		struct fuse_open_in *foi;
1112 
1113 		fdip2 = &fdi2;
1114 		fdisp_init(fdip2, sizeof(*foi));
1115 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1116 			cred);
1117 		foi = fdip2->indata;
1118 		foi->flags = flags;
1119 		err = fdisp_wait_answ(fdip2);
1120 		if (err)
1121 			goto out;
1122 		foo = fdip2->answ;
1123 	}
1124 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1125 	if (err) {
1126 		struct fuse_release_in *fri;
1127 		uint64_t nodeid = feo->nodeid;
1128 		uint64_t fh_id = foo->fh;
1129 
1130 		fdisp_destroy(fdip);
1131 		fdisp_init(fdip, sizeof(*fri));
1132 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1133 		fri = fdip->indata;
1134 		fri->fh = fh_id;
1135 		fri->flags = flags;
1136 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1137 		fuse_insert_message(fdip->tick, false);
1138 		goto out;
1139 	}
1140 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1141 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1142 		feo->attr_valid_nsec, NULL, true);
1143 
1144 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1145 	fuse_vnode_open(*vpp, foo->open_flags, td);
1146 	/*
1147 	 * Purge the parent's attribute cache because the daemon should've
1148 	 * updated its mtime and ctime
1149 	 */
1150 	fuse_vnode_clear_attr_cache(dvp);
1151 	cache_purge_negative(dvp);
1152 
1153 out:
1154 	if (fdip2)
1155 		fdisp_destroy(fdip2);
1156 	fdisp_destroy(fdip);
1157 	return err;
1158 }
1159 
1160 /*
1161     struct vnop_fdatasync_args {
1162 	struct vop_generic_args a_gen;
1163 	struct vnode * a_vp;
1164 	struct thread * a_td;
1165     };
1166 */
1167 static int
fuse_vnop_fdatasync(struct vop_fdatasync_args * ap)1168 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1169 {
1170 	struct vnode *vp = ap->a_vp;
1171 	struct thread *td = ap->a_td;
1172 	int waitfor = MNT_WAIT;
1173 
1174 	int err = 0;
1175 
1176 	if (fuse_isdeadfs(vp)) {
1177 		return 0;
1178 	}
1179 	if ((err = vop_stdfdatasync_buf(ap)))
1180 		return err;
1181 
1182 	return fuse_internal_fsync(vp, td, waitfor, true);
1183 }
1184 
1185 /*
1186     struct vnop_fsync_args {
1187 	struct vop_generic_args a_gen;
1188 	struct vnode * a_vp;
1189 	int  a_waitfor;
1190 	struct thread * a_td;
1191     };
1192 */
1193 static int
fuse_vnop_fsync(struct vop_fsync_args * ap)1194 fuse_vnop_fsync(struct vop_fsync_args *ap)
1195 {
1196 	struct vnode *vp = ap->a_vp;
1197 	struct thread *td = ap->a_td;
1198 	int waitfor = ap->a_waitfor;
1199 	int err = 0;
1200 
1201 	if (fuse_isdeadfs(vp)) {
1202 		return 0;
1203 	}
1204 	if ((err = vop_stdfsync(ap)))
1205 		return err;
1206 
1207 	return fuse_internal_fsync(vp, td, waitfor, false);
1208 }
1209 
1210 /*
1211     struct vnop_getattr_args {
1212 	struct vnode *a_vp;
1213 	struct vattr *a_vap;
1214 	struct ucred *a_cred;
1215 	struct thread *a_td;
1216     };
1217 */
1218 static int
fuse_vnop_getattr(struct vop_getattr_args * ap)1219 fuse_vnop_getattr(struct vop_getattr_args *ap)
1220 {
1221 	struct vnode *vp = ap->a_vp;
1222 	struct vattr *vap = ap->a_vap;
1223 	struct ucred *cred = ap->a_cred;
1224 	struct thread *td = curthread;
1225 	int err = 0;
1226 
1227 	err = fuse_internal_getattr(vp, vap, cred, td);
1228 	if (err == ENOTCONN && vnode_isvroot(vp)) {
1229 		/*
1230 		 * We want to seem a legitimate fs even if the daemon is dead,
1231 		 * so that, eg., we can still do path based unmounting after
1232 		 * the daemon dies.
1233 		 */
1234 		err = 0;
1235 		bzero(vap, sizeof(*vap));
1236 		vap->va_type = vnode_vtype(vp);
1237 	}
1238 	return err;
1239 }
1240 
1241 /*
1242     struct vnop_inactive_args {
1243 	struct vnode *a_vp;
1244     };
1245 */
1246 static int
fuse_vnop_inactive(struct vop_inactive_args * ap)1247 fuse_vnop_inactive(struct vop_inactive_args *ap)
1248 {
1249 	struct vnode *vp = ap->a_vp;
1250 	struct thread *td = curthread;
1251 
1252 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1253 	struct fuse_filehandle *fufh, *fufh_tmp;
1254 
1255 	int need_flush = 1;
1256 
1257 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1258 		if (need_flush && vp->v_type == VREG) {
1259 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1260 				fuse_vnode_savesize(vp, NULL, 0);
1261 			}
1262 			if ((fvdat->flag & FN_REVOKED) != 0)
1263 				fuse_io_invalbuf(vp, td);
1264 			else
1265 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1266 			need_flush = 0;
1267 		}
1268 		fuse_filehandle_close(vp, fufh, td, NULL);
1269 	}
1270 
1271 	if ((fvdat->flag & FN_REVOKED) != 0)
1272 		vrecycle(vp);
1273 
1274 	return 0;
1275 }
1276 
1277 /*
1278     struct vnop_ioctl_args {
1279 	struct vnode *a_vp;
1280 	u_long a_command;
1281 	caddr_t a_data;
1282 	int a_fflag;
1283 	struct ucred *a_cred;
1284 	struct thread *a_td;
1285     };
1286 */
1287 static int
fuse_vnop_ioctl(struct vop_ioctl_args * ap)1288 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1289 {
1290 	struct vnode *vp = ap->a_vp;
1291 	struct mount *mp = vnode_mount(vp);
1292 	struct ucred *cred = ap->a_cred;
1293 	off_t *offp;
1294 	pid_t pid = ap->a_td->td_proc->p_pid;
1295 	int err;
1296 
1297 	switch (ap->a_command) {
1298 	case FIOSEEKDATA:
1299 	case FIOSEEKHOLE:
1300 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1301 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1302 			int whence;
1303 
1304 			offp = ap->a_data;
1305 			if (ap->a_command == FIOSEEKDATA)
1306 				whence = SEEK_DATA;
1307 			else
1308 				whence = SEEK_HOLE;
1309 
1310 			vn_lock(vp, LK_SHARED | LK_RETRY);
1311 			err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1312 			    whence);
1313 			VOP_UNLOCK(vp);
1314 		}
1315 		if (fsess_not_impl(mp, FUSE_LSEEK))
1316 			err = vop_stdioctl(ap);
1317 		break;
1318 	default:
1319 		/* TODO: implement FUSE_IOCTL */
1320 		err = ENOTTY;
1321 		break;
1322 	}
1323 	return (err);
1324 }
1325 
1326 
1327 /*
1328     struct vnop_link_args {
1329 	struct vnode *a_tdvp;
1330 	struct vnode *a_vp;
1331 	struct componentname *a_cnp;
1332     };
1333 */
1334 static int
fuse_vnop_link(struct vop_link_args * ap)1335 fuse_vnop_link(struct vop_link_args *ap)
1336 {
1337 	struct vnode *vp = ap->a_vp;
1338 	struct vnode *tdvp = ap->a_tdvp;
1339 	struct componentname *cnp = ap->a_cnp;
1340 
1341 	struct vattr *vap = VTOVA(vp);
1342 
1343 	struct fuse_dispatcher fdi;
1344 	struct fuse_entry_out *feo;
1345 	struct fuse_link_in fli;
1346 
1347 	int err;
1348 
1349 	if (fuse_isdeadfs(vp)) {
1350 		return (EXTERROR(ENXIO, "This FUSE session is about "
1351 		    "to be closed"));
1352 	}
1353 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1354 		return (EXDEV);
1355 	}
1356 
1357 	/*
1358 	 * This is a seatbelt check to protect naive userspace filesystems from
1359 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1360 	 * filesystem does not allow attribute caching, assume it is capable of
1361 	 * validating that nlink does not overflow.
1362 	 */
1363 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1364 		return (EMLINK);
1365 	fli.oldnodeid = VTOI(vp);
1366 
1367 	fdisp_init(&fdi, 0);
1368 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1369 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1370 	if ((err = fdisp_wait_answ(&fdi))) {
1371 		goto out;
1372 	}
1373 	feo = fdi.answ;
1374 
1375 	if (fli.oldnodeid != feo->nodeid) {
1376 		static const char exterr[] = "Server assigned wrong inode "
1377 		    "for a hard link.";
1378 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1379 		fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, exterr);
1380 		fuse_vnode_clear_attr_cache(vp);
1381 		fuse_vnode_clear_attr_cache(tdvp);
1382 		err = EXTERROR(EIO, exterr);
1383 		goto out;
1384 	}
1385 
1386 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1387 	if (!err) {
1388 		/*
1389 		 * Purge the parent's attribute cache because the daemon
1390 		 * should've updated its mtime and ctime
1391 		 */
1392 		fuse_vnode_clear_attr_cache(tdvp);
1393 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1394 			feo->attr_valid_nsec, NULL, true);
1395 	}
1396 out:
1397 	fdisp_destroy(&fdi);
1398 	return err;
1399 }
1400 
1401 struct fuse_lookup_alloc_arg {
1402 	struct fuse_entry_out *feo;
1403 	struct componentname *cnp;
1404 	uint64_t nid;
1405 	__enum_uint8(vtype) vtyp;
1406 };
1407 
1408 /* Callback for vn_get_ino */
1409 static int
fuse_lookup_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)1410 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1411 {
1412 	struct fuse_lookup_alloc_arg *flaa = arg;
1413 
1414 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1415 		flaa->vtyp);
1416 }
1417 
1418 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1419 	"int", "struct timespec*", "struct timespec*");
1420 /*
1421     struct vnop_lookup_args {
1422 	struct vnodeop_desc *a_desc;
1423 	struct vnode *a_dvp;
1424 	struct vnode **a_vpp;
1425 	struct componentname *a_cnp;
1426     };
1427 */
1428 int
fuse_vnop_lookup(struct vop_lookup_args * ap)1429 fuse_vnop_lookup(struct vop_lookup_args *ap)
1430 {
1431 	struct vnode *dvp = ap->a_dvp;
1432 	struct vnode **vpp = ap->a_vpp;
1433 	struct componentname *cnp = ap->a_cnp;
1434 	struct thread *td = curthread;
1435 	struct ucred *cred = cnp->cn_cred;
1436 	struct timespec now;
1437 
1438 	int nameiop = cnp->cn_nameiop;
1439 	bool isdotdot = cnp->cn_flags & ISDOTDOT;
1440 	bool islastcn = cnp->cn_flags & ISLASTCN;
1441 	struct mount *mp = vnode_mount(dvp);
1442 	struct fuse_data *data = fuse_get_mpdata(mp);
1443 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1444 	bool is_dot;
1445 
1446 	int err = 0;
1447 	int lookup_err = 0;
1448 	struct vnode *vp = NULL;
1449 
1450 	struct fuse_dispatcher fdi;
1451 	bool did_lookup = false;
1452 	struct fuse_entry_out *feo = NULL;
1453 	__enum_uint8(vtype) vtyp;	/* vnode type of target */
1454 
1455 	uint64_t nid;
1456 
1457 	if (fuse_isdeadfs(dvp)) {
1458 		*vpp = NULL;
1459 		return (EXTERROR(ENXIO, "This FUSE session is about "
1460 		    "to be closed"));
1461 	}
1462 	if (!vnode_isdir(dvp))
1463 		return ENOTDIR;
1464 
1465 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1466 		return EROFS;
1467 
1468 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1469 		cnp->cn_flags &= ~NOEXECCHECK;
1470 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1471 		return err;
1472 
1473 	is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1474 	if (isdotdot && !(data->dataflags & FSESS_EXPORT_SUPPORT)) {
1475 		if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1476 			/*
1477 			 * Since the file system doesn't support ".." lookups,
1478 			 * we have no way to find this entry.
1479 			 */
1480 			return (EXTERROR(ESTALE, "This server does not support "
1481 			    "'..' lookups"));
1482 		}
1483 		nid = VTOFUD(dvp)->parent_nid;
1484 		if (nid == 0)
1485 			return ENOENT;
1486 		/* .. is obviously a directory */
1487 		vtyp = VDIR;
1488 	} else if (is_dot) {
1489 		nid = VTOI(dvp);
1490 		/* . is obviously a directory */
1491 		vtyp = VDIR;
1492 	} else {
1493 		struct timespec timeout;
1494 		int ncpticks; /* here to accommodate for API contract */
1495 
1496 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1497 		getnanouptime(&now);
1498 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1499 		switch (err) {
1500 		case -1:		/* positive match */
1501 			if (timespeccmp(&timeout, &now, >)) {
1502 				counter_u64_add(fuse_lookup_cache_hits, 1);
1503 			} else {
1504 				/* Cache timeout */
1505 				counter_u64_add(fuse_lookup_cache_misses, 1);
1506 				bintime_clear(
1507 					&VTOFUD(*vpp)->entry_cache_timeout);
1508 				cache_purge(*vpp);
1509 				if (dvp != *vpp)
1510 					vput(*vpp);
1511 				else
1512 					vrele(*vpp);
1513 				*vpp = NULL;
1514 				break;
1515 			}
1516 			return 0;
1517 
1518 		case 0:		/* no match in cache */
1519 			counter_u64_add(fuse_lookup_cache_misses, 1);
1520 			break;
1521 
1522 		case ENOENT:		/* negative match */
1523 			if (timespeccmp(&timeout, &now, <=)) {
1524 				/* Cache timeout */
1525 				cache_purge_negative(dvp);
1526 				break;
1527 			}
1528 			/* fall through */
1529 		default:
1530 			return err;
1531 		}
1532 
1533 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1534 		fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1535 
1536 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1537 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1538 		lookup_err = fdisp_wait_answ(&fdi);
1539 		did_lookup = true;
1540 
1541 		if (!lookup_err) {
1542 			/* lookup call succeeded */
1543 			feo = (struct fuse_entry_out *)fdi.answ;
1544 			nid = feo->nodeid;
1545 			if (nid == 0) {
1546 				/* zero nodeid means ENOENT and cache it */
1547 				struct timespec timeout;
1548 
1549 				fdi.answ_stat = ENOENT;
1550 				lookup_err = ENOENT;
1551 				if (cnp->cn_flags & MAKEENTRY) {
1552 					fuse_validity_2_timespec(feo, &timeout);
1553 					/* Use the same entry_time for .. as for
1554 					 * the file itself.  That doesn't honor
1555 					 * exactly what the fuse server tells
1556 					 * us, but to do otherwise would require
1557 					 * another cache lookup at this point.
1558 					 */
1559 					struct timespec *dtsp = NULL;
1560 					cache_enter_time(dvp, *vpp, cnp,
1561 						&timeout, dtsp);
1562 				}
1563 			}
1564 			vtyp = IFTOVT(feo->attr.mode);
1565 		}
1566 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1567 			fdisp_destroy(&fdi);
1568 			return lookup_err;
1569 		}
1570 	}
1571 	/* lookup_err, if non-zero, must be ENOENT at this point */
1572 
1573 	if (lookup_err) {
1574 		/* Entry not found */
1575 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1576 			if (default_permissions)
1577 				err = fuse_internal_access(dvp, VWRITE, td,
1578 				    cred);
1579 			else
1580 				err = 0;
1581 			if (!err) {
1582 				err = EJUSTRETURN;
1583 			}
1584 		} else {
1585 			err = ENOENT;
1586 		}
1587 	} else {
1588 		/* Entry was found */
1589 		if (isdotdot) {
1590 			struct fuse_lookup_alloc_arg flaa;
1591 
1592 			flaa.nid = nid;
1593 			flaa.feo = feo;
1594 			flaa.cnp = cnp;
1595 			flaa.vtyp = vtyp;
1596 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1597 				&vp);
1598 			*vpp = vp;
1599 		} else if (nid == VTOI(dvp)) {
1600 			if (is_dot) {
1601 				vref(dvp);
1602 				*vpp = dvp;
1603 			} else {
1604 				static const char exterr[] = "Server assigned "
1605 				    "same inode to both parent and child.";
1606 				fuse_warn(fuse_get_mpdata(mp),
1607 				    FSESS_WARN_ILLEGAL_INODE, exterr);
1608 				err = EXTERROR(EIO, exterr);
1609 			}
1610 
1611 		} else {
1612 			struct fuse_vnode_data *fvdat;
1613 
1614 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1615 			    &vp, cnp, vtyp);
1616 			if (err)
1617 				goto out;
1618 			*vpp = vp;
1619 			fvdat = VTOFUD(vp);
1620 
1621 			MPASS(feo != NULL);
1622 			if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1623 				/*
1624 				 * Attributes from the server are definitely
1625 				 * newer than the last attributes we sent to
1626 				 * the server, so cache them.
1627 				 */
1628 				fuse_internal_cache_attrs(*vpp, &feo->attr,
1629 					feo->attr_valid, feo->attr_valid_nsec,
1630 					NULL, true);
1631 			}
1632 			fuse_validity_2_bintime(feo->entry_valid,
1633 				feo->entry_valid_nsec,
1634 				&fvdat->entry_cache_timeout);
1635 
1636 			if ((nameiop == DELETE || nameiop == RENAME) &&
1637 				islastcn && default_permissions)
1638 			{
1639 				struct vattr dvattr;
1640 
1641 				err = fuse_internal_access(dvp, VWRITE, td,
1642 					cred);
1643 				if (err != 0)
1644 					goto out;
1645 				/*
1646 				 * if the parent's sticky bit is set, check
1647 				 * whether we're allowed to remove the file.
1648 				 * Need to figure out the vnode locking to make
1649 				 * this work.
1650 				 */
1651 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1652 				if ((dvattr.va_mode & S_ISTXT) &&
1653 					fuse_internal_access(dvp, VADMIN, td,
1654 						cred) &&
1655 					fuse_internal_access(*vpp, VADMIN, td,
1656 						cred)) {
1657 					err = EPERM;
1658 					goto out;
1659 				}
1660 			}
1661 		}
1662 	}
1663 out:
1664 	if (err) {
1665 		if (vp != NULL && dvp != vp)
1666 			vput(vp);
1667 		else if (vp != NULL)
1668 			vrele(vp);
1669 		*vpp = NULL;
1670 	}
1671 	if (did_lookup)
1672 		fdisp_destroy(&fdi);
1673 
1674 	return err;
1675 }
1676 
1677 /*
1678     struct vnop_mkdir_args {
1679 	struct vnode *a_dvp;
1680 	struct vnode **a_vpp;
1681 	struct componentname *a_cnp;
1682 	struct vattr *a_vap;
1683     };
1684 */
1685 static int
fuse_vnop_mkdir(struct vop_mkdir_args * ap)1686 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1687 {
1688 	struct vnode *dvp = ap->a_dvp;
1689 	struct vnode **vpp = ap->a_vpp;
1690 	struct componentname *cnp = ap->a_cnp;
1691 	struct vattr *vap = ap->a_vap;
1692 
1693 	struct fuse_mkdir_in fmdi;
1694 
1695 	if (fuse_isdeadfs(dvp)) {
1696 		return (EXTERROR(ENXIO, "This FUSE session is about "
1697 		    "to be closed"));
1698 	}
1699 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1700 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1701 
1702 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1703 	    sizeof(fmdi), VDIR));
1704 }
1705 
1706 /*
1707     struct vnop_mknod_args {
1708 	struct vnode *a_dvp;
1709 	struct vnode **a_vpp;
1710 	struct componentname *a_cnp;
1711 	struct vattr *a_vap;
1712     };
1713 */
1714 static int
fuse_vnop_mknod(struct vop_mknod_args * ap)1715 fuse_vnop_mknod(struct vop_mknod_args *ap)
1716 {
1717 
1718 	struct vnode *dvp = ap->a_dvp;
1719 	struct vnode **vpp = ap->a_vpp;
1720 	struct componentname *cnp = ap->a_cnp;
1721 	struct vattr *vap = ap->a_vap;
1722 
1723 	if (fuse_isdeadfs(dvp))
1724 		return (EXTERROR(ENXIO, "This FUSE session is about "
1725 		    "to be closed"));
1726 
1727 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1728 }
1729 
1730 /*
1731     struct vop_open_args {
1732 	struct vnode *a_vp;
1733 	int  a_mode;
1734 	struct ucred *a_cred;
1735 	struct thread *a_td;
1736 	int a_fdidx; / struct file *a_fp;
1737     };
1738 */
1739 static int
fuse_vnop_open(struct vop_open_args * ap)1740 fuse_vnop_open(struct vop_open_args *ap)
1741 {
1742 	struct vnode *vp = ap->a_vp;
1743 	int a_mode = ap->a_mode;
1744 	struct thread *td = ap->a_td;
1745 	struct ucred *cred = ap->a_cred;
1746 	pid_t pid = td->td_proc->p_pid;
1747 
1748 	if (fuse_isdeadfs(vp))
1749 		return (EXTERROR(ENXIO, "This FUSE session is about "
1750 		    "to be closed"));
1751 	if (VN_ISDEV(vp) || vp->v_type == VFIFO)
1752 		return (EXTERROR(EOPNOTSUPP, "Unsupported vnode type",
1753 		    vp->v_type));
1754 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1755 		return (EXTERROR(EINVAL, "Illegal mode", a_mode));
1756 
1757 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1758 		fuse_vnode_open(vp, 0, td);
1759 		return 0;
1760 	}
1761 
1762 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1763 }
1764 
1765 static int
fuse_vnop_pathconf(struct vop_pathconf_args * ap)1766 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1767 {
1768 	struct vnode *vp = ap->a_vp;
1769 	struct mount *mp;
1770 	struct fuse_filehandle *fufh;
1771 	int err;
1772 	bool closefufh = false;
1773 
1774 	switch (ap->a_name) {
1775 	case _PC_FILESIZEBITS:
1776 		*ap->a_retval = 64;
1777 		return (0);
1778 	case _PC_NAME_MAX:
1779 		*ap->a_retval = NAME_MAX;
1780 		return (0);
1781 	case _PC_LINK_MAX:
1782 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1783 		return (0);
1784 	case _PC_SYMLINK_MAX:
1785 		*ap->a_retval = MAXPATHLEN;
1786 		return (0);
1787 	case _PC_NO_TRUNC:
1788 		*ap->a_retval = 1;
1789 		return (0);
1790 	case _PC_MIN_HOLE_SIZE:
1791 		/*
1792 		 * The FUSE protocol provides no mechanism for a server to
1793 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1794 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1795 		 * 1 if it does.
1796 		 */
1797 		mp = vnode_mount(vp);
1798 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1799 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1800 			off_t offset = 0;
1801 
1802 			/*
1803 			 * Issue a FUSE_LSEEK to find out if it's supported.
1804 			 * Use SEEK_DATA instead of SEEK_HOLE, because the
1805 			 * latter generally requires sequential scans of file
1806 			 * metadata, which can be slow.
1807 			 */
1808 			err = fuse_vnop_do_lseek(vp, curthread,
1809 			    curthread->td_ucred, curthread->td_proc->p_pid,
1810 			    &offset, SEEK_DATA);
1811 			if (err == EBADF) {
1812 				/*
1813 				 * pathconf() doesn't necessarily open the
1814 				 * file.  So we may need to do it here.
1815 				 */
1816 				err = fuse_filehandle_open(vp, FREAD, &fufh,
1817 				    curthread, curthread->td_ucred);
1818 				if (err == 0) {
1819 					closefufh = true;
1820 					err = fuse_vnop_do_lseek(vp, curthread,
1821 					    curthread->td_ucred,
1822 					    curthread->td_proc->p_pid, &offset,
1823 					    SEEK_DATA);
1824 				}
1825 				if (closefufh)
1826 					fuse_filehandle_close(vp, fufh,
1827 					    curthread, curthread->td_ucred);
1828 			}
1829 
1830 		}
1831 
1832 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1833 			*ap->a_retval = 1;
1834 			return (0);
1835 		} else if (fsess_not_impl(mp, FUSE_LSEEK)) {
1836 			/* FUSE_LSEEK is not implemented */
1837 			return (EXTERROR(EINVAL, "This server does not "
1838 			    "implement FUSE_LSEEK"));
1839 		} else {
1840 			return (err);
1841 		}
1842 	default:
1843 		return (vop_stdpathconf(ap));
1844 	}
1845 }
1846 
1847 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1848     "struct uio*", "struct ucred*");
1849 /*
1850     struct vnop_read_args {
1851 	struct vnode *a_vp;
1852 	struct uio *a_uio;
1853 	int  a_ioflag;
1854 	struct ucred *a_cred;
1855     };
1856 */
1857 static int
fuse_vnop_read(struct vop_read_args * ap)1858 fuse_vnop_read(struct vop_read_args *ap)
1859 {
1860 	struct vnode *vp = ap->a_vp;
1861 	struct uio *uio = ap->a_uio;
1862 	int ioflag = ap->a_ioflag;
1863 	struct ucred *cred = ap->a_cred;
1864 	pid_t pid = curthread->td_proc->p_pid;
1865 	struct fuse_filehandle *fufh;
1866 	int err;
1867 	bool closefufh = false, directio;
1868 
1869 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1870 
1871 	if (fuse_isdeadfs(vp)) {
1872 		return (EXTERROR(ENXIO, "This FUSE session is about "
1873 		    "to be closed"));
1874 	}
1875 
1876 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1877 		ioflag |= IO_DIRECT;
1878 	}
1879 
1880 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1881 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1882 		/*
1883 		 * nfsd will do I/O without first doing VOP_OPEN.  We
1884 		 * must implicitly open the file here
1885 		 */
1886 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1887 		closefufh = true;
1888 	}
1889 	if (err) {
1890 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1891 		return err;
1892 	}
1893 
1894 	/*
1895          * Ideally, when the daemon asks for direct io at open time, the
1896          * standard file flag should be set according to this, so that would
1897          * just change the default mode, which later on could be changed via
1898          * fcntl(2).
1899          * But this doesn't work, the O_DIRECT flag gets cleared at some point
1900          * (don't know where). So to make any use of the Fuse direct_io option,
1901          * we hardwire it into the file's private data (similarly to Linux,
1902          * btw.).
1903          */
1904 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1905 
1906 	fuse_vnode_update(vp, FN_ATIMECHANGE);
1907 	if (directio) {
1908 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1909 		err = fuse_read_directbackend(vp, uio, cred, fufh);
1910 	} else {
1911 		SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1912 		err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1913 	}
1914 
1915 	if (closefufh)
1916 		fuse_filehandle_close(vp, fufh, curthread, cred);
1917 
1918 	return (err);
1919 }
1920 
1921 /*
1922     struct vnop_readdir_args {
1923 	struct vnode *a_vp;
1924 	struct uio *a_uio;
1925 	struct ucred *a_cred;
1926 	int *a_eofflag;
1927 	int *a_ncookies;
1928 	uint64_t **a_cookies;
1929     };
1930 */
1931 static int
fuse_vnop_readdir(struct vop_readdir_args * ap)1932 fuse_vnop_readdir(struct vop_readdir_args *ap)
1933 {
1934 	struct vnode *vp = ap->a_vp;
1935 	struct uio *uio = ap->a_uio;
1936 	struct ucred *cred = ap->a_cred;
1937 	struct fuse_filehandle *fufh = NULL;
1938 	struct mount *mp = vnode_mount(vp);
1939 	struct fuse_iov cookediov;
1940 	int err = 0;
1941 	uint64_t *cookies;
1942 	ssize_t tresid;
1943 	int ncookies;
1944 	bool closefufh = false;
1945 	pid_t pid = curthread->td_proc->p_pid;
1946 
1947 	if (ap->a_eofflag)
1948 		*ap->a_eofflag = 0;
1949 	if (fuse_isdeadfs(vp)) {
1950 		return (EXTERROR(ENXIO, "This FUSE session is about "
1951 		    "to be closed"));
1952 	}
1953 	if (uio_resid(uio) < sizeof(struct dirent))
1954 		return (EXTERROR(EINVAL, "Buffer is too small"));
1955 
1956 	tresid = uio->uio_resid;
1957 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1958 	if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1959 		KASSERT(!fsess_is_impl(mp, FUSE_OPENDIR),
1960 			("FUSE file systems that implement "
1961 			 "FUSE_OPENDIR should not be exported"));
1962 		/*
1963 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1964 		 * must implicitly open the directory here.
1965 		 */
1966 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1967 		closefufh = true;
1968 	}
1969 	if (err)
1970 		return (err);
1971 	if (ap->a_ncookies != NULL) {
1972 		ncookies = uio->uio_resid /
1973 			(offsetof(struct dirent, d_name) + 4) + 1;
1974 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1975 		*ap->a_ncookies = ncookies;
1976 		*ap->a_cookies = cookies;
1977 	} else {
1978 		ncookies = 0;
1979 		cookies = NULL;
1980 	}
1981 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1982 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1983 
1984 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1985 		&ncookies, cookies);
1986 
1987 	fiov_teardown(&cookediov);
1988 	if (closefufh)
1989 		fuse_filehandle_close(vp, fufh, curthread, cred);
1990 
1991 	if (ap->a_ncookies != NULL) {
1992 		if (err == 0) {
1993 			*ap->a_ncookies -= ncookies;
1994 		} else {
1995 			free(*ap->a_cookies, M_TEMP);
1996 			*ap->a_ncookies = 0;
1997 			*ap->a_cookies = NULL;
1998 		}
1999 	}
2000 	if (err == 0 && tresid == uio->uio_resid)
2001 		*ap->a_eofflag = 1;
2002 
2003 	return err;
2004 }
2005 
2006 /*
2007     struct vnop_readlink_args {
2008 	struct vnode *a_vp;
2009 	struct uio *a_uio;
2010 	struct ucred *a_cred;
2011     };
2012 */
2013 static int
fuse_vnop_readlink(struct vop_readlink_args * ap)2014 fuse_vnop_readlink(struct vop_readlink_args *ap)
2015 {
2016 	struct vnode *vp = ap->a_vp;
2017 	struct uio *uio = ap->a_uio;
2018 	struct ucred *cred = ap->a_cred;
2019 
2020 	struct fuse_dispatcher fdi;
2021 	int err;
2022 
2023 	if (fuse_isdeadfs(vp)) {
2024 		return (EXTERROR(ENXIO, "This FUSE session is about "
2025 		    "to be closed"));
2026 	}
2027 	if (!vnode_islnk(vp)) {
2028 		return EINVAL;
2029 	}
2030 	fdisp_init(&fdi, 0);
2031 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
2032 	if (err) {
2033 		goto out;
2034 	}
2035 	if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) {
2036 		static const char exterr[] = "Server returned an embedded NUL "
2037 		    "from FUSE_READLINK.";
2038 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
2039 		fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, exterr);
2040 		err = EXTERROR(EIO, exterr);
2041 		goto out;
2042 	}
2043 	if (((char *)fdi.answ)[0] == '/' &&
2044 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
2045 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
2046 
2047 		err = uiomove(mpth, strlen(mpth), uio);
2048 	}
2049 	if (!err) {
2050 		err = uiomove(fdi.answ, fdi.iosize, uio);
2051 	}
2052 out:
2053 	fdisp_destroy(&fdi);
2054 	return err;
2055 }
2056 
2057 /*
2058     struct vnop_reclaim_args {
2059 	struct vnode *a_vp;
2060     };
2061 */
2062 static int
fuse_vnop_reclaim(struct vop_reclaim_args * ap)2063 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2064 {
2065 	struct vnode *vp = ap->a_vp;
2066 	struct thread *td = curthread;
2067 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2068 	struct fuse_filehandle *fufh, *fufh_tmp;
2069 
2070 	if (!fvdat) {
2071 		panic("FUSE: no vnode data during recycling");
2072 	}
2073 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2074 		printf("FUSE: vnode being reclaimed with open fufh "
2075 			"(type=%#x)", fufh->fufh_type);
2076 		fuse_filehandle_close(vp, fufh, td, NULL);
2077 	}
2078 
2079 	if (VTOI(vp) == 1) {
2080 		/*
2081 		 * Don't send FUSE_FORGET for the root inode, because
2082 		 * we never send FUSE_LOOKUP for it (see
2083 		 * fuse_vfsop_root) and we don't want the server to see
2084 		 * mismatched lookup counts.
2085 		 */
2086 		struct fuse_data *data;
2087 		struct vnode *vroot;
2088 
2089 		data = fuse_get_mpdata(vnode_mount(vp));
2090 		FUSE_LOCK();
2091 		vroot = data->vroot;
2092 		data->vroot = NULL;
2093 		FUSE_UNLOCK();
2094 		if (vroot)
2095 			vrele(vroot);
2096 	} else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2097 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2098 		    fvdat->nlookup);
2099 	}
2100 	cache_purge(vp);
2101 	vfs_hash_remove(vp);
2102 	fuse_vnode_destroy(vp);
2103 
2104 	return 0;
2105 }
2106 
2107 /*
2108     struct vnop_remove_args {
2109 	struct vnode *a_dvp;
2110 	struct vnode *a_vp;
2111 	struct componentname *a_cnp;
2112     };
2113 */
2114 static int
fuse_vnop_remove(struct vop_remove_args * ap)2115 fuse_vnop_remove(struct vop_remove_args *ap)
2116 {
2117 	struct vnode *dvp = ap->a_dvp;
2118 	struct vnode *vp = ap->a_vp;
2119 	struct componentname *cnp = ap->a_cnp;
2120 
2121 	int err;
2122 
2123 	if (fuse_isdeadfs(vp)) {
2124 		return (EXTERROR(ENXIO, "This FUSE session is about "
2125 		    "to be closed"));
2126 	}
2127 	if (vnode_isdir(vp)) {
2128 		return (EXTERROR(EPERM, "vnode is a directory"));
2129 	}
2130 
2131 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2132 
2133 	return err;
2134 }
2135 
2136 /*
2137     struct vnop_rename_args {
2138 	struct vnode *a_fdvp;
2139 	struct vnode *a_fvp;
2140 	struct componentname *a_fcnp;
2141 	struct vnode *a_tdvp;
2142 	struct vnode *a_tvp;
2143 	struct componentname *a_tcnp;
2144     };
2145 */
2146 static int
fuse_vnop_rename(struct vop_rename_args * ap)2147 fuse_vnop_rename(struct vop_rename_args *ap)
2148 {
2149 	struct vnode *fdvp = ap->a_fdvp;
2150 	struct vnode *fvp = ap->a_fvp;
2151 	struct componentname *fcnp = ap->a_fcnp;
2152 	struct vnode *tdvp = ap->a_tdvp;
2153 	struct vnode *tvp = ap->a_tvp;
2154 	struct componentname *tcnp = ap->a_tcnp;
2155 	struct fuse_data *data;
2156 	bool newparent = fdvp != tdvp;
2157 	bool isdir = fvp->v_type == VDIR;
2158 	int err = 0;
2159 
2160 	if (fuse_isdeadfs(fdvp)) {
2161 		return (EXTERROR(ENXIO, "This FUSE session is about "
2162 		    "to be closed"));
2163 	}
2164 	if (fvp->v_mount != tdvp->v_mount ||
2165 	    (tvp && fvp->v_mount != tvp->v_mount)) {
2166 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2167 		err = EXTERROR(EXDEV, "Cross-device rename");
2168 		goto out;
2169 	}
2170 	cache_purge(fvp);
2171 
2172 	/*
2173 	 * FUSE library is expected to check if target directory is not
2174 	 * under the source directory in the file system tree.
2175 	 * Linux performs this check at VFS level.
2176 	 */
2177 	/*
2178 	 * If source is a directory, and it will get a new parent, user must
2179 	 * have write permission to it, so ".." can be modified.
2180 	 */
2181 	data = fuse_get_mpdata(vnode_mount(tdvp));
2182 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2183 		err = fuse_internal_access(fvp, VWRITE,
2184 			curthread, tcnp->cn_cred);
2185 		if (err)
2186 			goto out;
2187 	}
2188 	sx_xlock(&data->rename_lock);
2189 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2190 	if (err == 0) {
2191 		if (tdvp != fdvp)
2192 			fuse_vnode_setparent(fvp, tdvp);
2193 		if (tvp != NULL)
2194 			fuse_vnode_setparent(tvp, NULL);
2195 	}
2196 	sx_unlock(&data->rename_lock);
2197 
2198 	if (tvp != NULL && tvp != fvp) {
2199 		cache_purge(tvp);
2200 	}
2201 	if (vnode_isdir(fvp)) {
2202 		if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2203 			cache_purge(tdvp);
2204 		}
2205 		cache_purge(fdvp);
2206 	}
2207 out:
2208 	if (tdvp == tvp) {
2209 		vrele(tdvp);
2210 	} else {
2211 		vput(tdvp);
2212 	}
2213 	if (tvp != NULL) {
2214 		vput(tvp);
2215 	}
2216 	vrele(fdvp);
2217 	vrele(fvp);
2218 
2219 	return err;
2220 }
2221 
2222 /*
2223     struct vnop_rmdir_args {
2224 	    struct vnode *a_dvp;
2225 	    struct vnode *a_vp;
2226 	    struct componentname *a_cnp;
2227     } *ap;
2228 */
2229 static int
fuse_vnop_rmdir(struct vop_rmdir_args * ap)2230 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2231 {
2232 	struct vnode *dvp = ap->a_dvp;
2233 	struct vnode *vp = ap->a_vp;
2234 
2235 	int err;
2236 
2237 	if (fuse_isdeadfs(vp)) {
2238 		return (EXTERROR(ENXIO, "This FUSE session is about "
2239 		    "to be closed"));
2240 	}
2241 	if (VTOFUD(vp) == VTOFUD(dvp)) {
2242 		return (EXTERROR(EINVAL, "Directory to be removed "
2243 		    "contains itself"));
2244 	}
2245 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2246 
2247 	return err;
2248 }
2249 
2250 /*
2251     struct vnop_setattr_args {
2252 	struct vnode *a_vp;
2253 	struct vattr *a_vap;
2254 	struct ucred *a_cred;
2255 	struct thread *a_td;
2256     };
2257 */
2258 static int
fuse_vnop_setattr(struct vop_setattr_args * ap)2259 fuse_vnop_setattr(struct vop_setattr_args *ap)
2260 {
2261 	struct vnode *vp = ap->a_vp;
2262 	struct vattr *vap = ap->a_vap;
2263 	struct ucred *cred = ap->a_cred;
2264 	struct thread *td = curthread;
2265 	struct mount *mp;
2266 	struct fuse_data *data;
2267 	struct vattr old_va;
2268 	int dataflags;
2269 	int err = 0, err2;
2270 	accmode_t accmode = 0;
2271 	bool checkperm;
2272 	bool drop_suid = false;
2273 
2274 	mp = vnode_mount(vp);
2275 	data = fuse_get_mpdata(mp);
2276 	dataflags = data->dataflags;
2277 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2278 
2279 	if (fuse_isdeadfs(vp)) {
2280 		return (EXTERROR(ENXIO, "This FUSE session is about "
2281 		    "to be closed"));
2282 	}
2283 
2284 	if (vap->va_uid != (uid_t)VNOVAL) {
2285 		if (checkperm) {
2286 			/* Only root may change a file's owner */
2287 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2288 			if (err) {
2289 				/* As a special case, allow the null chown */
2290 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2291 					td);
2292 				if (err2)
2293 					return (err2);
2294 				if (vap->va_uid != old_va.va_uid)
2295 					return err;
2296 				drop_suid = true;
2297 			}
2298 		}
2299 		accmode |= VADMIN;
2300 	}
2301 	if (vap->va_gid != (gid_t)VNOVAL) {
2302 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2303 			drop_suid = true;
2304 		if (checkperm && !groupmember(vap->va_gid, cred)) {
2305 			/*
2306 			 * Non-root users may only chgrp to one of their own
2307 			 * groups
2308 			 */
2309 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2310 			if (err) {
2311 				/* As a special case, allow the null chgrp */
2312 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2313 					td);
2314 				if (err2)
2315 					return (err2);
2316 				if (vap->va_gid != old_va.va_gid)
2317 					return err;
2318 			}
2319 		}
2320 		accmode |= VADMIN;
2321 	}
2322 	if (vap->va_size != VNOVAL) {
2323 		switch (vp->v_type) {
2324 		case VDIR:
2325 			return (EISDIR);
2326 		case VLNK:
2327 		case VREG:
2328 			if (vfs_isrdonly(mp))
2329 				return (EROFS);
2330 			err = vn_rlimit_trunc(vap->va_size, td);
2331 			if (err)
2332 				return (err);
2333 			break;
2334 		default:
2335 			/*
2336 			 * According to POSIX, the result is unspecified
2337 			 * for file types other than regular files,
2338 			 * directories and shared memory objects.  We
2339 			 * don't support shared memory objects in the file
2340 			 * system, and have dubious support for truncating
2341 			 * symlinks.  Just ignore the request in other cases.
2342 			 */
2343 			return (0);
2344 		}
2345 		/* Don't set accmode.  Permission to trunc is checked upstack */
2346 	}
2347 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2348 		if (vap->va_vaflags & VA_UTIMES_NULL)
2349 			accmode |= VWRITE;
2350 		else
2351 			accmode |= VADMIN;
2352 	}
2353 	if (drop_suid) {
2354 		if (vap->va_mode != (mode_t)VNOVAL)
2355 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2356 		else {
2357 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2358 			if (err)
2359 				return (err);
2360 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2361 		}
2362 	}
2363 	if (vap->va_mode != (mode_t)VNOVAL) {
2364 		/* Only root may set the sticky bit on non-directories */
2365 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2366 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2367 			return EFTYPE;
2368 		if (checkperm && (vap->va_mode & S_ISGID)) {
2369 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2370 			if (err)
2371 				return (err);
2372 			if (!groupmember(old_va.va_gid, cred)) {
2373 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2374 				if (err)
2375 					return (err);
2376 			}
2377 		}
2378 		accmode |= VADMIN;
2379 	}
2380 
2381 	if (vfs_isrdonly(mp))
2382 		return EROFS;
2383 
2384 	if (checkperm) {
2385 		err = fuse_internal_access(vp, accmode, td, cred);
2386 	} else {
2387 		err = 0;
2388 	}
2389 	if (err)
2390 		return err;
2391 	else
2392 		return fuse_internal_setattr(vp, vap, td, cred);
2393 }
2394 
2395 /*
2396     struct vnop_strategy_args {
2397 	struct vnode *a_vp;
2398 	struct buf *a_bp;
2399     };
2400 */
2401 static int
fuse_vnop_strategy(struct vop_strategy_args * ap)2402 fuse_vnop_strategy(struct vop_strategy_args *ap)
2403 {
2404 	struct vnode *vp = ap->a_vp;
2405 	struct buf *bp = ap->a_bp;
2406 
2407 	if (!vp || fuse_isdeadfs(vp)) {
2408 		bp->b_ioflags |= BIO_ERROR;
2409 		bp->b_error = ENXIO;
2410 		bufdone(bp);
2411 		return 0;
2412 	}
2413 
2414 	/*
2415 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2416 	 * fuse_io_strategy sets bp's error fields
2417 	 */
2418 	(void)fuse_io_strategy(vp, bp);
2419 
2420 	return 0;
2421 }
2422 
2423 /*
2424     struct vnop_symlink_args {
2425 	struct vnode *a_dvp;
2426 	struct vnode **a_vpp;
2427 	struct componentname *a_cnp;
2428 	struct vattr *a_vap;
2429 	char *a_target;
2430     };
2431 */
2432 static int
fuse_vnop_symlink(struct vop_symlink_args * ap)2433 fuse_vnop_symlink(struct vop_symlink_args *ap)
2434 {
2435 	struct vnode *dvp = ap->a_dvp;
2436 	struct vnode **vpp = ap->a_vpp;
2437 	struct componentname *cnp = ap->a_cnp;
2438 	const char *target = ap->a_target;
2439 
2440 	struct fuse_dispatcher fdi;
2441 
2442 	int err;
2443 	size_t len;
2444 
2445 	if (fuse_isdeadfs(dvp)) {
2446 		return (EXTERROR(ENXIO, "This FUSE session is about "
2447 		    "to be closed"));
2448 	}
2449 	/*
2450 	 * Unlike the other creator type calls, here we have to create a message
2451 	 * where the name of the new entry comes first, and the data describing
2452 	 * the entry comes second.
2453 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2454 	 * but put together the message manually and just call the core part.
2455 	 */
2456 
2457 	len = strlen(target) + 1;
2458 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2459 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2460 
2461 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2462 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2463 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2464 
2465 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2466 	fdisp_destroy(&fdi);
2467 	return err;
2468 }
2469 
2470 /*
2471     struct vnop_write_args {
2472 	struct vnode *a_vp;
2473 	struct uio *a_uio;
2474 	int  a_ioflag;
2475 	struct ucred *a_cred;
2476     };
2477 */
2478 static int
fuse_vnop_write(struct vop_write_args * ap)2479 fuse_vnop_write(struct vop_write_args *ap)
2480 {
2481 	struct vnode *vp = ap->a_vp;
2482 	struct uio *uio = ap->a_uio;
2483 	int ioflag = ap->a_ioflag;
2484 	struct ucred *cred = ap->a_cred;
2485 	pid_t pid = curthread->td_proc->p_pid;
2486 	struct fuse_filehandle *fufh;
2487 	int err;
2488 	bool closefufh = false, directio;
2489 
2490 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2491 
2492 	if (fuse_isdeadfs(vp)) {
2493 		return (EXTERROR(ENXIO, "This FUSE session is about "
2494 		    "to be closed"));
2495 	}
2496 
2497 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2498 		ioflag |= IO_DIRECT;
2499 	}
2500 
2501 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2502 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2503 		/*
2504 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2505 		 * must implicitly open the file here
2506 		 */
2507 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2508 		closefufh = true;
2509 	}
2510 	if (err) {
2511 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2512 		return err;
2513 	}
2514 
2515 	/*
2516          * Ideally, when the daemon asks for direct io at open time, the
2517          * standard file flag should be set according to this, so that would
2518          * just change the default mode, which later on could be changed via
2519          * fcntl(2).
2520          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2521          * (don't know where). So to make any use of the Fuse direct_io option,
2522          * we hardwire it into the file's private data (similarly to Linux,
2523          * btw.).
2524          */
2525 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2526 
2527 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2528 	if (directio) {
2529 		off_t start, end, filesize;
2530 		bool pages = (ioflag & IO_VMIO) != 0;
2531 
2532 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2533 
2534 		err = fuse_vnode_size(vp, &filesize, cred, curthread);
2535 		if (err)
2536 			goto out;
2537 
2538 		start = uio->uio_offset;
2539 		end = start + uio->uio_resid;
2540 		if (!pages) {
2541 			err = fuse_inval_buf_range(vp, filesize, start,
2542 			    end);
2543 			if (err)
2544 				goto out;
2545 		}
2546 		err = fuse_write_directbackend(vp, uio, cred, fufh,
2547 			filesize, ioflag, pages);
2548 	} else {
2549 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2550 			"buffered write of vnode");
2551 		if (!fsess_opt_writeback(vnode_mount(vp)))
2552 			ioflag |= IO_SYNC;
2553 		err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2554 	}
2555 	fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2556 
2557 out:
2558 	if (closefufh)
2559 		fuse_filehandle_close(vp, fufh, curthread, cred);
2560 
2561 	return (err);
2562 }
2563 
2564 static daddr_t
fuse_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)2565 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2566 {
2567 	const int biosize = fuse_iosize(vp);
2568 
2569 	return (off / biosize);
2570 }
2571 
2572 static int
fuse_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * blksz)2573 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2574 {
2575 	off_t filesize;
2576 	int err;
2577 	const int biosize = fuse_iosize(vp);
2578 
2579 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2580 	if (err) {
2581 		/* This will turn into a SIGBUS */
2582 		return (EIO);
2583 	} else if ((off_t)lbn * biosize >= filesize) {
2584 		*blksz = 0;
2585 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2586 		*blksz = filesize - (off_t)lbn *biosize;
2587 	} else {
2588 		*blksz = biosize;
2589 	}
2590 	return (0);
2591 }
2592 
2593 /*
2594     struct vnop_getpages_args {
2595 	struct vnode *a_vp;
2596 	vm_page_t *a_m;
2597 	int a_count;
2598 	int a_reqpage;
2599     };
2600 */
2601 static int
fuse_vnop_getpages(struct vop_getpages_args * ap)2602 fuse_vnop_getpages(struct vop_getpages_args *ap)
2603 {
2604 	struct vnode *vp = ap->a_vp;
2605 
2606 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2607 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2608 			"called on non-cacheable vnode??\n");
2609 		return (VM_PAGER_ERROR);
2610 	}
2611 
2612 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2613 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2614 }
2615 
2616 static const char extattr_namespace_separator = '.';
2617 
2618 /*
2619     struct vop_getextattr_args {
2620 	struct vop_generic_args a_gen;
2621 	struct vnode *a_vp;
2622 	int a_attrnamespace;
2623 	const char *a_name;
2624 	struct uio *a_uio;
2625 	size_t *a_size;
2626 	struct ucred *a_cred;
2627 	struct thread *a_td;
2628     };
2629 */
2630 static int
fuse_vnop_getextattr(struct vop_getextattr_args * ap)2631 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2632 {
2633 	struct vnode *vp = ap->a_vp;
2634 	struct uio *uio = ap->a_uio;
2635 	struct fuse_dispatcher fdi;
2636 	struct fuse_getxattr_in *get_xattr_in;
2637 	struct fuse_getxattr_out *get_xattr_out;
2638 	struct mount *mp = vnode_mount(vp);
2639 	struct thread *td = ap->a_td;
2640 	struct ucred *cred = ap->a_cred;
2641 	char *prefix;
2642 	char *attr_str;
2643 	size_t len;
2644 	int err;
2645 
2646 	if (fuse_isdeadfs(vp))
2647 		return (EXTERROR(ENXIO, "This FUSE session is about "
2648 		    "to be closed"));
2649 
2650 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2651 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2652 		    "extended attributes"));
2653 
2654 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2655 	if (err)
2656 		return err;
2657 
2658 	/* Default to looking for user attributes. */
2659 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2660 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2661 	else
2662 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2663 
2664 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2665 	    strlen(ap->a_name) + 1;
2666 
2667 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2668 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2669 
2670 	get_xattr_in = fdi.indata;
2671 	/*
2672 	 * Check to see whether we're querying the available size or
2673 	 * issuing the actual request.  If we pass in 0, we get back struct
2674 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2675 	 * that much data, without the struct fuse_getxattr_out header.
2676 	 */
2677 	if (uio == NULL)
2678 		get_xattr_in->size = 0;
2679 	else
2680 		get_xattr_in->size = uio->uio_resid;
2681 
2682 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2683 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2684 	    ap->a_name);
2685 
2686 	err = fdisp_wait_answ(&fdi);
2687 	if (err != 0) {
2688 		if (err == ENOSYS) {
2689 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2690 			err = (EXTERROR(EOPNOTSUPP, "This server does not "
2691 			    "implement extended attributes"));
2692 		}
2693 		goto out;
2694 	}
2695 
2696 	get_xattr_out = fdi.answ;
2697 
2698 	if (ap->a_size != NULL)
2699 		*ap->a_size = get_xattr_out->size;
2700 
2701 	if (uio != NULL)
2702 		err = uiomove(fdi.answ, fdi.iosize, uio);
2703 
2704 out:
2705 	fdisp_destroy(&fdi);
2706 	return (err);
2707 }
2708 
2709 /*
2710     struct vop_setextattr_args {
2711 	struct vop_generic_args a_gen;
2712 	struct vnode *a_vp;
2713 	int a_attrnamespace;
2714 	const char *a_name;
2715 	struct uio *a_uio;
2716 	struct ucred *a_cred;
2717 	struct thread *a_td;
2718     };
2719 */
2720 static int
fuse_vnop_setextattr(struct vop_setextattr_args * ap)2721 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2722 {
2723 	struct vnode *vp = ap->a_vp;
2724 	struct uio *uio = ap->a_uio;
2725 	struct fuse_dispatcher fdi;
2726 	struct fuse_setxattr_in *set_xattr_in;
2727 	struct mount *mp = vnode_mount(vp);
2728 	struct thread *td = ap->a_td;
2729 	struct ucred *cred = ap->a_cred;
2730 	size_t struct_size = FUSE_COMPAT_SETXATTR_IN_SIZE;
2731 	char *prefix;
2732 	size_t len;
2733 	char *attr_str;
2734 	int err;
2735 
2736 	if (fuse_isdeadfs(vp))
2737 		return (EXTERROR(ENXIO, "This FUSE session is about "
2738 		    "to be closed"));
2739 
2740 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2741 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2742 		    "setting extended attributes"));
2743 
2744 	if (vfs_isrdonly(mp))
2745 		return EROFS;
2746 
2747 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2748 	if (ap->a_uio == NULL) {
2749 		/*
2750 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2751 		 * return EOPNOTSUPP.
2752 		 */
2753 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2754 			return (EXTERROR(EOPNOTSUPP, "This server does not "
2755 			    "implement removing extended attributess"));
2756 		else
2757 			return (EXTERROR(EINVAL, "DELETEEXTATTR should be used "
2758 			    "to remove extattrs"));
2759 	}
2760 
2761 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2762 		VWRITE);
2763 	if (err)
2764 		return err;
2765 
2766 	/* Default to looking for user attributes. */
2767 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2768 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2769 	else
2770 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2771 
2772 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2773 	    strlen(ap->a_name) + 1;
2774 
2775 	/* older FUSE servers  use a smaller fuse_setxattr_in struct*/
2776 	if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33))
2777 		struct_size = sizeof(*set_xattr_in);
2778 
2779 	fdisp_init(&fdi, len + struct_size + uio->uio_resid);
2780 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2781 
2782 	set_xattr_in = fdi.indata;
2783 	set_xattr_in->size = uio->uio_resid;
2784 
2785 	if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33)) {
2786 		set_xattr_in->setxattr_flags = 0;
2787 		set_xattr_in->padding = 0;
2788 	}
2789 
2790 	attr_str = (char *)fdi.indata + struct_size;
2791 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2792 	    ap->a_name);
2793 
2794 	err = uiomove((char *)fdi.indata + struct_size + len,
2795 	    uio->uio_resid, uio);
2796 	if (err != 0) {
2797 		goto out;
2798 	}
2799 
2800 	err = fdisp_wait_answ(&fdi);
2801 
2802 	if (err == ENOSYS) {
2803 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2804 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
2805 		    "setting extended attributes");
2806 	}
2807 	if (err == ERESTART) {
2808 		/* Can't restart after calling uiomove */
2809 		err = EINTR;
2810 	}
2811 
2812 out:
2813 	fdisp_destroy(&fdi);
2814 	return (err);
2815 }
2816 
2817 /*
2818  * The Linux / FUSE extended attribute list is simply a collection of
2819  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2820  * byte length followed by a non-NUL terminated string.  So, this allows
2821  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2822  * Linux attribute names are reported with the namespace as a prefix (e.g.
2823  * "user.attribute_name"), but in FreeBSD they are reported without the
2824  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2825  *
2826  * user.attr_name1\0user.attr_name2\0
2827  *
2828  * to:
2829  *
2830  * <num>attr_name1<num>attr_name2
2831  *
2832  * Where "<num>" is a single byte number of characters in the attribute name.
2833  *
2834  * Args:
2835  * prefix - exattr namespace prefix string
2836  * list, list_len - input list with namespace prefixes
2837  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2838  */
2839 static int
fuse_xattrlist_convert(char * prefix,const char * list,int list_len,char * bsd_list,int * bsd_list_len)2840 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2841     char *bsd_list, int *bsd_list_len)
2842 {
2843 	int len, pos, dist_to_next, prefix_len;
2844 
2845 	pos = 0;
2846 	*bsd_list_len = 0;
2847 	prefix_len = strlen(prefix);
2848 
2849 	while (pos < list_len && list[pos] != '\0') {
2850 		dist_to_next = strlen(&list[pos]) + 1;
2851 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2852 		    list[pos + prefix_len] == extattr_namespace_separator) {
2853 			len = dist_to_next -
2854 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2855 			if (len >= EXTATTR_MAXNAMELEN)
2856 				return (ENAMETOOLONG);
2857 
2858 			bsd_list[*bsd_list_len] = len;
2859 			memcpy(&bsd_list[*bsd_list_len + 1],
2860 			    &list[pos + prefix_len +
2861 			    sizeof(extattr_namespace_separator)], len);
2862 
2863 			*bsd_list_len += len + 1;
2864 		}
2865 
2866 		pos += dist_to_next;
2867 	}
2868 
2869 	return (0);
2870 }
2871 
2872 /*
2873  * List extended attributes
2874  *
2875  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2876  * has a number of differences compared to its FreeBSD equivalent,
2877  * extattr_list_file:
2878  *
2879  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2880  *   whereas listxattr(2) only returns attributes for a single namespace
2881  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2882  * - If the provided buffer is not large enough to hold the result,
2883  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2884  *   return as many results as will fit.
2885  */
2886 /*
2887     struct vop_listextattr_args {
2888 	struct vop_generic_args a_gen;
2889 	struct vnode *a_vp;
2890 	int a_attrnamespace;
2891 	struct uio *a_uio;
2892 	size_t *a_size;
2893 	struct ucred *a_cred;
2894 	struct thread *a_td;
2895     };
2896 */
2897 static int
fuse_vnop_listextattr(struct vop_listextattr_args * ap)2898 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2899 {
2900 	struct vnode *vp = ap->a_vp;
2901 	struct uio *uio = ap->a_uio;
2902 	struct fuse_dispatcher fdi;
2903 	struct fuse_listxattr_in *list_xattr_in;
2904 	struct fuse_listxattr_out *list_xattr_out;
2905 	struct mount *mp = vnode_mount(vp);
2906 	struct thread *td = ap->a_td;
2907 	struct ucred *cred = ap->a_cred;
2908 	char *prefix;
2909 	char *bsd_list = NULL;
2910 	char *linux_list;
2911 	int bsd_list_len;
2912 	int linux_list_len;
2913 	int err;
2914 
2915 	if (fuse_isdeadfs(vp))
2916 		return (EXTERROR(ENXIO, "This FUSE session is about "
2917 		    "to be closed"));
2918 
2919 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
2920 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2921 		    "extended attributes"));
2922 
2923 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2924 	if (err)
2925 		return err;
2926 
2927 	/*
2928 	 * Add space for a NUL and the period separator if enabled.
2929 	 * Default to looking for user attributes.
2930 	 */
2931 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2932 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2933 	else
2934 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2935 
2936 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2937 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2938 
2939 	/*
2940 	 * Retrieve Linux / FUSE compatible list size.
2941 	 */
2942 	list_xattr_in = fdi.indata;
2943 	list_xattr_in->size = 0;
2944 
2945 	err = fdisp_wait_answ(&fdi);
2946 	if (err != 0) {
2947 		if (err == ENOSYS) {
2948 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2949 			err = EXTERROR(EOPNOTSUPP, "This server does not "
2950 			    "implement extended attributes");
2951 		}
2952 		goto out;
2953 	}
2954 
2955 	list_xattr_out = fdi.answ;
2956 	linux_list_len = list_xattr_out->size;
2957 	if (linux_list_len == 0) {
2958 		if (ap->a_size != NULL)
2959 			*ap->a_size = linux_list_len;
2960 		goto out;
2961 	}
2962 
2963 	/*
2964 	 * Retrieve Linux / FUSE compatible list values.
2965 	 */
2966 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2967 	list_xattr_in = fdi.indata;
2968 	list_xattr_in->size = linux_list_len;
2969 
2970 	err = fdisp_wait_answ(&fdi);
2971 	if (err == ERANGE) {
2972 		/*
2973 		 * Race detected.  The attribute list must've grown since the
2974 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2975 		 * to userland so we can process signals, if necessary, before
2976 		 * restarting.
2977 		 */
2978 		err = ERESTART;
2979 		goto out;
2980 	} else if (err != 0)
2981 		goto out;
2982 
2983 	linux_list = fdi.answ;
2984 	/* FUSE doesn't allow the server to return more data than requested */
2985 	if (fdi.iosize > linux_list_len) {
2986 		struct fuse_data *data = fuse_get_mpdata(mp);
2987 
2988 		fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2989 			"server returned "
2990 			"more extended attribute data than requested; "
2991 			"should've returned ERANGE instead.");
2992 	} else {
2993 		/* But returning less data is fine */
2994 		linux_list_len = fdi.iosize;
2995 	}
2996 
2997 	/*
2998 	 * Retrieve the BSD compatible list values.
2999 	 * The Linux / FUSE attribute list format isn't the same
3000 	 * as FreeBSD's format. So we need to transform it into
3001 	 * FreeBSD's format before giving it to the user.
3002 	 */
3003 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
3004 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
3005 	    bsd_list, &bsd_list_len);
3006 	if (err != 0)
3007 		goto out;
3008 
3009 	if (ap->a_size != NULL)
3010 		*ap->a_size = bsd_list_len;
3011 
3012 	if (uio != NULL)
3013 		err = uiomove(bsd_list, bsd_list_len, uio);
3014 
3015 out:
3016 	free(bsd_list, M_TEMP);
3017 	fdisp_destroy(&fdi);
3018 	return (err);
3019 }
3020 
3021 /*
3022     struct vop_deallocate_args {
3023 	struct vop_generic_args a_gen;
3024 	struct vnode *a_vp;
3025 	off_t *a_offset;
3026 	off_t *a_len;
3027 	int a_flags;
3028 	int a_ioflag;
3029         struct ucred *a_cred;
3030     };
3031 */
3032 static int
fuse_vnop_deallocate(struct vop_deallocate_args * ap)3033 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
3034 {
3035 	struct vnode *vp = ap->a_vp;
3036 	struct mount *mp = vnode_mount(vp);
3037 	struct fuse_filehandle *fufh;
3038 	struct fuse_dispatcher fdi;
3039 	struct fuse_fallocate_in *ffi;
3040 	struct ucred *cred = ap->a_cred;
3041 	pid_t pid = curthread->td_proc->p_pid;
3042 	off_t *len = ap->a_len;
3043 	off_t *offset = ap->a_offset;
3044 	int ioflag = ap->a_ioflag;
3045 	off_t filesize;
3046 	int err;
3047 	bool closefufh = false;
3048 
3049 	if (fuse_isdeadfs(vp))
3050 		return (EXTERROR(ENXIO, "This FUSE session is about "
3051 		    "to be closed"));
3052 
3053 	if (vfs_isrdonly(mp))
3054 		return (EROFS);
3055 
3056 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
3057 		goto fallback;
3058 
3059 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
3060 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
3061 		/*
3062 		 * nfsd will do I/O without first doing VOP_OPEN.  We
3063 		 * must implicitly open the file here
3064 		 */
3065 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
3066 		closefufh = true;
3067 	}
3068 	if (err)
3069 		return (err);
3070 
3071 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3072 
3073 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
3074 	if (err)
3075 		goto out;
3076 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3077 
3078 	fdisp_init(&fdi, sizeof(*ffi));
3079 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3080 	ffi = fdi.indata;
3081 	ffi->fh = fufh->fh_id;
3082 	ffi->offset = *offset;
3083 	ffi->length = *len;
3084 	/*
3085 	 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3086 	 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3087 	 */
3088 	ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3089 	err = fdisp_wait_answ(&fdi);
3090 
3091 	if (err == ENOSYS) {
3092 		fdisp_destroy(&fdi);
3093 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
3094 		goto fallback;
3095 	} else if (err == EOPNOTSUPP) {
3096 		/*
3097 		 * The file system server does not support FUSE_FALLOCATE with
3098 		 * the supplied mode for this particular file.
3099 		 */
3100 		fdisp_destroy(&fdi);
3101 		goto fallback;
3102 	} else if (!err) {
3103 		/*
3104 		 * Clip the returned offset to EoF.  Do it here rather than
3105 		 * before FUSE_FALLOCATE just in case the kernel's cached file
3106 		 * size is out of date.  Unfortunately, FUSE does not return
3107 		 * any information about filesize from that operation.
3108 		 */
3109 		*offset = MIN(*offset + *len, filesize);
3110 		*len = 0;
3111 		fuse_vnode_undirty_cached_timestamps(vp, false);
3112 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
3113 
3114 		if (ioflag & IO_SYNC)
3115 			err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3116 			    false);
3117 	}
3118 
3119 	fdisp_destroy(&fdi);
3120 out:
3121 	if (closefufh)
3122 		fuse_filehandle_close(vp, fufh, curthread, cred);
3123 
3124 	return (err);
3125 
3126 fallback:
3127 	if (closefufh)
3128 		fuse_filehandle_close(vp, fufh, curthread, cred);
3129 
3130 	return (vop_stddeallocate(ap));
3131 }
3132 
3133 /*
3134     struct vop_deleteextattr_args {
3135 	struct vop_generic_args a_gen;
3136 	struct vnode *a_vp;
3137 	int a_attrnamespace;
3138 	const char *a_name;
3139 	struct ucred *a_cred;
3140 	struct thread *a_td;
3141     };
3142 */
3143 static int
fuse_vnop_deleteextattr(struct vop_deleteextattr_args * ap)3144 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3145 {
3146 	struct vnode *vp = ap->a_vp;
3147 	struct fuse_dispatcher fdi;
3148 	struct mount *mp = vnode_mount(vp);
3149 	struct thread *td = ap->a_td;
3150 	struct ucred *cred = ap->a_cred;
3151 	char *prefix;
3152 	size_t len;
3153 	char *attr_str;
3154 	int err;
3155 
3156 	if (fuse_isdeadfs(vp))
3157 		return (EXTERROR(ENXIO, "This FUSE session is about "
3158 		    "to be closed"));
3159 
3160 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3161 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3162 		    "removing extended attributes"));
3163 
3164 	if (vfs_isrdonly(mp))
3165 		return EROFS;
3166 
3167 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3168 		VWRITE);
3169 	if (err)
3170 		return err;
3171 
3172 	/* Default to looking for user attributes. */
3173 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3174 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3175 	else
3176 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3177 
3178 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3179 	    strlen(ap->a_name) + 1;
3180 
3181 	fdisp_init(&fdi, len);
3182 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3183 
3184 	attr_str = fdi.indata;
3185 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3186 	    ap->a_name);
3187 
3188 	err = fdisp_wait_answ(&fdi);
3189 	if (err == ENOSYS) {
3190 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3191 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
3192 		    "removing extended attributes");
3193 	}
3194 
3195 	fdisp_destroy(&fdi);
3196 	return (err);
3197 }
3198 
3199 /*
3200     struct vnop_print_args {
3201 	struct vnode *a_vp;
3202     };
3203 */
3204 static int
fuse_vnop_print(struct vop_print_args * ap)3205 fuse_vnop_print(struct vop_print_args *ap)
3206 {
3207 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3208 
3209 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3210 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3211 	    (uintmax_t)fvdat->nlookup,
3212 	    fvdat->flag);
3213 
3214 	return 0;
3215 }
3216 
3217 /*
3218  * Get an NFS filehandle for a FUSE file.
3219  *
3220  * This will only work for FUSE file systems that guarantee the uniqueness of
3221  * nodeid:generation, which most don't.
3222  */
3223 /*
3224 vop_vptofh {
3225 	IN struct vnode *a_vp;
3226 	IN struct fid *a_fhp;
3227 };
3228 */
3229 static int
fuse_vnop_vptofh(struct vop_vptofh_args * ap)3230 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3231 {
3232 	struct vnode *vp = ap->a_vp;
3233 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
3234 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3235 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3236 		"FUSE fid type is too big");
3237 	struct mount *mp = vnode_mount(vp);
3238 	struct fuse_data *data = fuse_get_mpdata(mp);
3239 	struct vattr va;
3240 	int err;
3241 
3242 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3243 		/* NFS requires lookups for "." and ".." */
3244 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3245 			"VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3246 		return (EXTERROR(EOPNOTSUPP, "This server is "
3247 		    "missing FUSE_EXPORT_SUPPORT"));
3248 	}
3249 	if ((mp->mnt_flag & MNT_EXPORTED) &&
3250 		fsess_is_impl(mp, FUSE_OPENDIR))
3251 	{
3252 		/*
3253 		 * NFS is stateless, so nfsd must reopen a directory on every
3254 		 * call to VOP_READDIR, passing in the d_off field from the
3255 		 * final dirent of the previous invocation.  But if the server
3256 		 * implements FUSE_OPENDIR, the FUSE protocol does not
3257 		 * guarantee that d_off will be valid after a directory is
3258 		 * closed and reopened.  So prohibit exporting FUSE file
3259 		 * systems that implement FUSE_OPENDIR.
3260 		 *
3261 		 * But userspace NFS servers don't have this problem.
3262                  */
3263 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3264 			"VOP_VPTOFH with FUSE_OPENDIR");
3265 		return (EXTERROR(EOPNOTSUPP, "This server implements "
3266 		    "FUSE_OPENDIR so is not compatible with getfh"));
3267 	}
3268 
3269 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3270 	if (err)
3271 		return err;
3272 
3273 	/*ip = VTOI(ap->a_vp);*/
3274 	/*ufhp = (struct ufid *)ap->a_fhp;*/
3275 	fhp->len = sizeof(struct fuse_fid);
3276 	fhp->nid = fvdat->nid;
3277 	if (fvdat->generation <= UINT32_MAX)
3278 		fhp->gen = fvdat->generation;
3279 	else
3280 		return (EXTERROR(EOVERFLOW, "inode generation "
3281 		    "number overflow"));
3282 	return (0);
3283 }
3284