xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 9e682c426093a3afc0f609a1f04048b0beb021ec)
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 (mp == NULL || mp != vnode_mount(outvp))
881 		return (EXTERROR(ENOSYS, "Mount points do not match"));
882 
883 	if (incred->cr_uid != outcred->cr_uid)
884 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
885 		    "support different credentials for infd and outfd"));
886 
887 	if (incred->cr_gid != outcred->cr_gid)
888 		return (EXTERROR(ENOSYS, "FUSE_COPY_FILE_RANGE does not "
889 		    "support different credentials for infd and outfd"));
890 
891 	/* Caller busied mp, mnt_data can be safely accessed. */
892 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
893 		return (EXTERROR(ENOSYS, "This daemon does not "
894 		    "implement COPY_FILE_RANGE"));
895 
896 	if (ap->a_fsizetd == NULL)
897 		td = curthread;
898 	else
899 		td = ap->a_fsizetd;
900 	pid = td->td_proc->p_pid;
901 
902 	vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE);
903 	if (invp->v_data == NULL || outvp->v_data == NULL) {
904 		err = EXTERROR(EBADF, "vnode got reclaimed");
905 		goto unlock;
906 	}
907 
908 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
909 	if (err)
910 		goto unlock;
911 
912 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
913 	if (err)
914 		goto unlock;
915 
916 	io.uio_resid = *ap->a_lenp;
917 	if (ap->a_fsizetd) {
918 		io.uio_offset = *ap->a_outoffp;
919 		err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
920 		if (err != 0)
921 			goto unlock;
922 	}
923 
924 	err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
925 	if (err)
926 		goto unlock;
927 
928 	vnode_pager_clean_sync(invp);
929 	err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
930 		*ap->a_outoffp + io.uio_resid);
931 	if (err)
932 		goto unlock;
933 
934 	fdisp_init(&fdi, sizeof(*fcfri));
935 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
936 	fcfri = fdi.indata;
937 	fcfri->fh_in = infufh->fh_id;
938 	fcfri->off_in = *ap->a_inoffp;
939 	fcfri->nodeid_out = VTOI(outvp);
940 	fcfri->fh_out = outfufh->fh_id;
941 	fcfri->off_out = *ap->a_outoffp;
942 	fcfri->len = io.uio_resid;
943 	fcfri->flags = 0;
944 
945 	err = fdisp_wait_answ(&fdi);
946 	if (err == 0) {
947 		fwo = fdi.answ;
948 		*ap->a_lenp = fwo->size;
949 		*ap->a_inoffp += fwo->size;
950 		*ap->a_outoffp += fwo->size;
951 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
952 		if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
953                         fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
954 			getnanouptime(&outfvdat->last_local_modify);
955 		}
956 		fuse_vnode_update(invp, FN_ATIMECHANGE);
957 		fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
958 	}
959 	fdisp_destroy(&fdi);
960 
961 unlock:
962 	if (invp != outvp)
963 		VOP_UNLOCK(invp);
964 	VOP_UNLOCK(outvp);
965 
966 	if (err == ENOSYS)
967 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
968 
969 	/*
970 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
971 	 * local.
972 	 */
973 	return (err);
974 }
975 
976 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)977 fdisp_make_mknod_for_fallback(
978 	struct fuse_dispatcher *fdip,
979 	struct componentname *cnp,
980 	struct vnode *dvp,
981 	uint64_t parentnid,
982 	struct thread *td,
983 	struct ucred *cred,
984 	mode_t mode,
985 	enum fuse_opcode *op)
986 {
987 	struct fuse_mknod_in *fmni;
988 
989 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
990 	*op = FUSE_MKNOD;
991 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
992 	fmni = fdip->indata;
993 	fmni->mode = mode;
994 	fmni->rdev = 0;
995 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
996 	    cnp->cn_namelen);
997 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
998 }
999 /*
1000     struct vnop_create_args {
1001 	struct vnode *a_dvp;
1002 	struct vnode **a_vpp;
1003 	struct componentname *a_cnp;
1004 	struct vattr *a_vap;
1005     };
1006 */
1007 static int
fuse_vnop_create(struct vop_create_args * ap)1008 fuse_vnop_create(struct vop_create_args *ap)
1009 {
1010 	struct vnode *dvp = ap->a_dvp;
1011 	struct vnode **vpp = ap->a_vpp;
1012 	struct componentname *cnp = ap->a_cnp;
1013 	struct vattr *vap = ap->a_vap;
1014 	struct thread *td = curthread;
1015 	struct ucred *cred = cnp->cn_cred;
1016 
1017 	struct fuse_data *data;
1018 	struct fuse_create_in *fci;
1019 	struct fuse_entry_out *feo;
1020 	struct fuse_open_out *foo;
1021 	struct fuse_dispatcher fdi, fdi2;
1022 	struct fuse_dispatcher *fdip = &fdi;
1023 	struct fuse_dispatcher *fdip2 = NULL;
1024 
1025 	int err;
1026 
1027 	struct mount *mp = vnode_mount(dvp);
1028 	data = fuse_get_mpdata(mp);
1029 	uint64_t parentnid = VTOFUD(dvp)->nid;
1030 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
1031 	enum fuse_opcode op;
1032 	int flags;
1033 
1034 	if (fuse_isdeadfs(dvp))
1035 		return (EXTERROR(ENXIO, "This FUSE session is about "
1036 		    "to be closed"));
1037 
1038 	/* FUSE expects sockets to be created with FUSE_MKNOD */
1039 	if (vap->va_type == VSOCK)
1040 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
1041 
1042 	/*
1043 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
1044 	 * writable mode makes sense, and we might as well include readability
1045 	 * too.
1046 	 */
1047 	flags = O_RDWR;
1048 
1049 	bzero(&fdi, sizeof(fdi));
1050 
1051 	if (vap->va_type != VREG)
1052 		return (EXTERROR(EINVAL, "Only regular files can be created"));
1053 
1054 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
1055 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
1056 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1057 			cred, mode, &op);
1058 	} else {
1059 		/* Use FUSE_CREATE */
1060 		size_t insize;
1061 
1062 		op = FUSE_CREATE;
1063 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1064 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1065 		fci = fdip->indata;
1066 		fci->mode = mode;
1067 		fci->flags = O_CREAT | flags;
1068 		if (fuse_libabi_geq(data, 7, 12)) {
1069 			insize = sizeof(*fci);
1070 			fci->umask = td->td_proc->p_pd->pd_cmask;
1071 		} else {
1072 			insize = sizeof(struct fuse_open_in);
1073 		}
1074 
1075 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1076 		    cnp->cn_namelen);
1077 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1078 	}
1079 
1080 	err = fdisp_wait_answ(fdip);
1081 
1082 	if (err) {
1083 		if (err == ENOSYS && op == FUSE_CREATE) {
1084 			fsess_set_notimpl(mp, FUSE_CREATE);
1085 			fdisp_destroy(fdip);
1086 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1087 				parentnid, td, cred, mode, &op);
1088 			err = fdisp_wait_answ(fdip);
1089 		}
1090 		if (err)
1091 			goto out;
1092 	}
1093 
1094 	feo = fdip->answ;
1095 
1096 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1097 		goto out;
1098 	}
1099 
1100 	if (op == FUSE_CREATE) {
1101 		if (fuse_libabi_geq(data, 7, 9))
1102 			foo = (struct fuse_open_out*)(feo + 1);
1103 		else
1104 			foo = (struct fuse_open_out*)((char*)feo +
1105 				FUSE_COMPAT_ENTRY_OUT_SIZE);
1106 	} else {
1107 		/* Issue a separate FUSE_OPEN */
1108 		struct fuse_open_in *foi;
1109 
1110 		fdip2 = &fdi2;
1111 		fdisp_init(fdip2, sizeof(*foi));
1112 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1113 			cred);
1114 		foi = fdip2->indata;
1115 		foi->flags = flags;
1116 		err = fdisp_wait_answ(fdip2);
1117 		if (err)
1118 			goto out;
1119 		foo = fdip2->answ;
1120 	}
1121 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1122 	if (err) {
1123 		struct fuse_release_in *fri;
1124 		uint64_t nodeid = feo->nodeid;
1125 		uint64_t fh_id = foo->fh;
1126 
1127 		fdisp_destroy(fdip);
1128 		fdisp_init(fdip, sizeof(*fri));
1129 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1130 		fri = fdip->indata;
1131 		fri->fh = fh_id;
1132 		fri->flags = flags;
1133 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1134 		fuse_insert_message(fdip->tick, false);
1135 		goto out;
1136 	}
1137 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1138 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1139 		feo->attr_valid_nsec, NULL, true);
1140 
1141 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1142 	fuse_vnode_open(*vpp, foo->open_flags, td);
1143 	/*
1144 	 * Purge the parent's attribute cache because the daemon should've
1145 	 * updated its mtime and ctime
1146 	 */
1147 	fuse_vnode_clear_attr_cache(dvp);
1148 	cache_purge_negative(dvp);
1149 
1150 out:
1151 	if (fdip2)
1152 		fdisp_destroy(fdip2);
1153 	fdisp_destroy(fdip);
1154 	return err;
1155 }
1156 
1157 /*
1158     struct vnop_fdatasync_args {
1159 	struct vop_generic_args a_gen;
1160 	struct vnode * a_vp;
1161 	struct thread * a_td;
1162     };
1163 */
1164 static int
fuse_vnop_fdatasync(struct vop_fdatasync_args * ap)1165 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1166 {
1167 	struct vnode *vp = ap->a_vp;
1168 	struct thread *td = ap->a_td;
1169 	int waitfor = MNT_WAIT;
1170 
1171 	int err = 0;
1172 
1173 	if (fuse_isdeadfs(vp)) {
1174 		return 0;
1175 	}
1176 	if ((err = vop_stdfdatasync_buf(ap)))
1177 		return err;
1178 
1179 	return fuse_internal_fsync(vp, td, waitfor, true);
1180 }
1181 
1182 /*
1183     struct vnop_fsync_args {
1184 	struct vop_generic_args a_gen;
1185 	struct vnode * a_vp;
1186 	int  a_waitfor;
1187 	struct thread * a_td;
1188     };
1189 */
1190 static int
fuse_vnop_fsync(struct vop_fsync_args * ap)1191 fuse_vnop_fsync(struct vop_fsync_args *ap)
1192 {
1193 	struct vnode *vp = ap->a_vp;
1194 	struct thread *td = ap->a_td;
1195 	int waitfor = ap->a_waitfor;
1196 	int err = 0;
1197 
1198 	if (fuse_isdeadfs(vp)) {
1199 		return 0;
1200 	}
1201 	if ((err = vop_stdfsync(ap)))
1202 		return err;
1203 
1204 	return fuse_internal_fsync(vp, td, waitfor, false);
1205 }
1206 
1207 /*
1208     struct vnop_getattr_args {
1209 	struct vnode *a_vp;
1210 	struct vattr *a_vap;
1211 	struct ucred *a_cred;
1212 	struct thread *a_td;
1213     };
1214 */
1215 static int
fuse_vnop_getattr(struct vop_getattr_args * ap)1216 fuse_vnop_getattr(struct vop_getattr_args *ap)
1217 {
1218 	struct vnode *vp = ap->a_vp;
1219 	struct vattr *vap = ap->a_vap;
1220 	struct ucred *cred = ap->a_cred;
1221 	struct thread *td = curthread;
1222 	int err = 0;
1223 
1224 	err = fuse_internal_getattr(vp, vap, cred, td);
1225 	if (err == ENOTCONN && vnode_isvroot(vp)) {
1226 		/*
1227 		 * We want to seem a legitimate fs even if the daemon is dead,
1228 		 * so that, eg., we can still do path based unmounting after
1229 		 * the daemon dies.
1230 		 */
1231 		err = 0;
1232 		bzero(vap, sizeof(*vap));
1233 		vap->va_type = vnode_vtype(vp);
1234 	}
1235 	return err;
1236 }
1237 
1238 /*
1239     struct vnop_inactive_args {
1240 	struct vnode *a_vp;
1241     };
1242 */
1243 static int
fuse_vnop_inactive(struct vop_inactive_args * ap)1244 fuse_vnop_inactive(struct vop_inactive_args *ap)
1245 {
1246 	struct vnode *vp = ap->a_vp;
1247 	struct thread *td = curthread;
1248 
1249 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1250 	struct fuse_filehandle *fufh, *fufh_tmp;
1251 
1252 	int need_flush = 1;
1253 
1254 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1255 		if (need_flush && vp->v_type == VREG) {
1256 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1257 				fuse_vnode_savesize(vp, NULL, 0);
1258 			}
1259 			if ((fvdat->flag & FN_REVOKED) != 0)
1260 				fuse_io_invalbuf(vp, td);
1261 			else
1262 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1263 			need_flush = 0;
1264 		}
1265 		fuse_filehandle_close(vp, fufh, td, NULL);
1266 	}
1267 
1268 	if ((fvdat->flag & FN_REVOKED) != 0)
1269 		vrecycle(vp);
1270 
1271 	return 0;
1272 }
1273 
1274 /*
1275     struct vnop_ioctl_args {
1276 	struct vnode *a_vp;
1277 	u_long a_command;
1278 	caddr_t a_data;
1279 	int a_fflag;
1280 	struct ucred *a_cred;
1281 	struct thread *a_td;
1282     };
1283 */
1284 static int
fuse_vnop_ioctl(struct vop_ioctl_args * ap)1285 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1286 {
1287 	struct vnode *vp = ap->a_vp;
1288 	struct mount *mp = vnode_mount(vp);
1289 	struct ucred *cred = ap->a_cred;
1290 	off_t *offp;
1291 	pid_t pid = ap->a_td->td_proc->p_pid;
1292 	int err;
1293 
1294 	switch (ap->a_command) {
1295 	case FIOSEEKDATA:
1296 	case FIOSEEKHOLE:
1297 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1298 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1299 			int whence;
1300 
1301 			offp = ap->a_data;
1302 			if (ap->a_command == FIOSEEKDATA)
1303 				whence = SEEK_DATA;
1304 			else
1305 				whence = SEEK_HOLE;
1306 
1307 			vn_lock(vp, LK_SHARED | LK_RETRY);
1308 			err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1309 			    whence);
1310 			VOP_UNLOCK(vp);
1311 		}
1312 		if (fsess_not_impl(mp, FUSE_LSEEK))
1313 			err = vop_stdioctl(ap);
1314 		break;
1315 	default:
1316 		/* TODO: implement FUSE_IOCTL */
1317 		err = ENOTTY;
1318 		break;
1319 	}
1320 	return (err);
1321 }
1322 
1323 
1324 /*
1325     struct vnop_link_args {
1326 	struct vnode *a_tdvp;
1327 	struct vnode *a_vp;
1328 	struct componentname *a_cnp;
1329     };
1330 */
1331 static int
fuse_vnop_link(struct vop_link_args * ap)1332 fuse_vnop_link(struct vop_link_args *ap)
1333 {
1334 	struct vnode *vp = ap->a_vp;
1335 	struct vnode *tdvp = ap->a_tdvp;
1336 	struct componentname *cnp = ap->a_cnp;
1337 
1338 	struct vattr *vap = VTOVA(vp);
1339 
1340 	struct fuse_dispatcher fdi;
1341 	struct fuse_entry_out *feo;
1342 	struct fuse_link_in fli;
1343 
1344 	int err;
1345 
1346 	if (fuse_isdeadfs(vp)) {
1347 		return (EXTERROR(ENXIO, "This FUSE session is about "
1348 		    "to be closed"));
1349 	}
1350 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1351 		return (EXDEV);
1352 	}
1353 
1354 	/*
1355 	 * This is a seatbelt check to protect naive userspace filesystems from
1356 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1357 	 * filesystem does not allow attribute caching, assume it is capable of
1358 	 * validating that nlink does not overflow.
1359 	 */
1360 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1361 		return (EMLINK);
1362 	fli.oldnodeid = VTOI(vp);
1363 
1364 	fdisp_init(&fdi, 0);
1365 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1366 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1367 	if ((err = fdisp_wait_answ(&fdi))) {
1368 		goto out;
1369 	}
1370 	feo = fdi.answ;
1371 
1372 	if (fli.oldnodeid != feo->nodeid) {
1373 		static const char exterr[] = "Server assigned wrong inode "
1374 		    "for a hard link.";
1375 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
1376 		fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, exterr);
1377 		fuse_vnode_clear_attr_cache(vp);
1378 		fuse_vnode_clear_attr_cache(tdvp);
1379 		err = EXTERROR(EIO, exterr);
1380 		goto out;
1381 	}
1382 
1383 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1384 	if (!err) {
1385 		/*
1386 		 * Purge the parent's attribute cache because the daemon
1387 		 * should've updated its mtime and ctime
1388 		 */
1389 		fuse_vnode_clear_attr_cache(tdvp);
1390 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1391 			feo->attr_valid_nsec, NULL, true);
1392 	}
1393 out:
1394 	fdisp_destroy(&fdi);
1395 	return err;
1396 }
1397 
1398 struct fuse_lookup_alloc_arg {
1399 	struct fuse_entry_out *feo;
1400 	struct componentname *cnp;
1401 	uint64_t nid;
1402 	__enum_uint8(vtype) vtyp;
1403 };
1404 
1405 /* Callback for vn_get_ino */
1406 static int
fuse_lookup_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)1407 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1408 {
1409 	struct fuse_lookup_alloc_arg *flaa = arg;
1410 
1411 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1412 		flaa->vtyp);
1413 }
1414 
1415 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1416 	"int", "struct timespec*", "struct timespec*");
1417 /*
1418     struct vnop_lookup_args {
1419 	struct vnodeop_desc *a_desc;
1420 	struct vnode *a_dvp;
1421 	struct vnode **a_vpp;
1422 	struct componentname *a_cnp;
1423     };
1424 */
1425 int
fuse_vnop_lookup(struct vop_lookup_args * ap)1426 fuse_vnop_lookup(struct vop_lookup_args *ap)
1427 {
1428 	struct vnode *dvp = ap->a_dvp;
1429 	struct vnode **vpp = ap->a_vpp;
1430 	struct componentname *cnp = ap->a_cnp;
1431 	struct thread *td = curthread;
1432 	struct ucred *cred = cnp->cn_cred;
1433 	struct timespec now;
1434 
1435 	int nameiop = cnp->cn_nameiop;
1436 	bool isdotdot = cnp->cn_flags & ISDOTDOT;
1437 	bool islastcn = cnp->cn_flags & ISLASTCN;
1438 	struct mount *mp = vnode_mount(dvp);
1439 	struct fuse_data *data = fuse_get_mpdata(mp);
1440 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1441 	bool is_dot;
1442 
1443 	int err = 0;
1444 	int lookup_err = 0;
1445 	struct vnode *vp = NULL;
1446 
1447 	struct fuse_dispatcher fdi;
1448 	bool did_lookup = false;
1449 	struct fuse_entry_out *feo = NULL;
1450 	__enum_uint8(vtype) vtyp;	/* vnode type of target */
1451 
1452 	uint64_t nid;
1453 
1454 	if (fuse_isdeadfs(dvp)) {
1455 		*vpp = NULL;
1456 		return (EXTERROR(ENXIO, "This FUSE session is about "
1457 		    "to be closed"));
1458 	}
1459 	if (!vnode_isdir(dvp))
1460 		return ENOTDIR;
1461 
1462 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1463 		return EROFS;
1464 
1465 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1466 		cnp->cn_flags &= ~NOEXECCHECK;
1467 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1468 		return err;
1469 
1470 	is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.';
1471 	if (isdotdot && !(data->dataflags & FSESS_EXPORT_SUPPORT)) {
1472 		if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1473 			/*
1474 			 * Since the file system doesn't support ".." lookups,
1475 			 * we have no way to find this entry.
1476 			 */
1477 			return (EXTERROR(ESTALE, "This server does not support "
1478 			    "'..' lookups"));
1479 		}
1480 		nid = VTOFUD(dvp)->parent_nid;
1481 		if (nid == 0)
1482 			return ENOENT;
1483 		/* .. is obviously a directory */
1484 		vtyp = VDIR;
1485 	} else if (is_dot) {
1486 		nid = VTOI(dvp);
1487 		/* . is obviously a directory */
1488 		vtyp = VDIR;
1489 	} else {
1490 		struct timespec timeout;
1491 		int ncpticks; /* here to accommodate for API contract */
1492 
1493 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1494 		getnanouptime(&now);
1495 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1496 		switch (err) {
1497 		case -1:		/* positive match */
1498 			if (timespeccmp(&timeout, &now, >)) {
1499 				counter_u64_add(fuse_lookup_cache_hits, 1);
1500 			} else {
1501 				/* Cache timeout */
1502 				counter_u64_add(fuse_lookup_cache_misses, 1);
1503 				bintime_clear(
1504 					&VTOFUD(*vpp)->entry_cache_timeout);
1505 				cache_purge(*vpp);
1506 				if (dvp != *vpp)
1507 					vput(*vpp);
1508 				else
1509 					vrele(*vpp);
1510 				*vpp = NULL;
1511 				break;
1512 			}
1513 			return 0;
1514 
1515 		case 0:		/* no match in cache */
1516 			counter_u64_add(fuse_lookup_cache_misses, 1);
1517 			break;
1518 
1519 		case ENOENT:		/* negative match */
1520 			if (timespeccmp(&timeout, &now, <=)) {
1521 				/* Cache timeout */
1522 				cache_purge_negative(dvp);
1523 				break;
1524 			}
1525 			/* fall through */
1526 		default:
1527 			return err;
1528 		}
1529 
1530 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1531 		fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1532 
1533 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1534 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1535 		lookup_err = fdisp_wait_answ(&fdi);
1536 		did_lookup = true;
1537 
1538 		if (!lookup_err) {
1539 			/* lookup call succeeded */
1540 			feo = (struct fuse_entry_out *)fdi.answ;
1541 			nid = feo->nodeid;
1542 			if (nid == 0) {
1543 				/* zero nodeid means ENOENT and cache it */
1544 				struct timespec timeout;
1545 
1546 				fdi.answ_stat = ENOENT;
1547 				lookup_err = ENOENT;
1548 				if (cnp->cn_flags & MAKEENTRY) {
1549 					fuse_validity_2_timespec(feo, &timeout);
1550 					/* Use the same entry_time for .. as for
1551 					 * the file itself.  That doesn't honor
1552 					 * exactly what the fuse server tells
1553 					 * us, but to do otherwise would require
1554 					 * another cache lookup at this point.
1555 					 */
1556 					struct timespec *dtsp = NULL;
1557 					cache_enter_time(dvp, *vpp, cnp,
1558 						&timeout, dtsp);
1559 				}
1560 			}
1561 			vtyp = IFTOVT(feo->attr.mode);
1562 		}
1563 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1564 			fdisp_destroy(&fdi);
1565 			return lookup_err;
1566 		}
1567 	}
1568 	/* lookup_err, if non-zero, must be ENOENT at this point */
1569 
1570 	if (lookup_err) {
1571 		/* Entry not found */
1572 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1573 			if (default_permissions)
1574 				err = fuse_internal_access(dvp, VWRITE, td,
1575 				    cred);
1576 			else
1577 				err = 0;
1578 			if (!err) {
1579 				err = EJUSTRETURN;
1580 			}
1581 		} else {
1582 			err = ENOENT;
1583 		}
1584 	} else {
1585 		/* Entry was found */
1586 		if (isdotdot) {
1587 			struct fuse_lookup_alloc_arg flaa;
1588 
1589 			flaa.nid = nid;
1590 			flaa.feo = feo;
1591 			flaa.cnp = cnp;
1592 			flaa.vtyp = vtyp;
1593 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1594 				&vp);
1595 			*vpp = vp;
1596 		} else if (nid == VTOI(dvp)) {
1597 			if (is_dot) {
1598 				vref(dvp);
1599 				*vpp = dvp;
1600 			} else {
1601 				static const char exterr[] = "Server assigned "
1602 				    "same inode to both parent and child.";
1603 				fuse_warn(fuse_get_mpdata(mp),
1604 				    FSESS_WARN_ILLEGAL_INODE, exterr);
1605 				err = EXTERROR(EIO, exterr);
1606 			}
1607 
1608 		} else {
1609 			struct fuse_vnode_data *fvdat;
1610 
1611 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1612 			    &vp, cnp, vtyp);
1613 			if (err)
1614 				goto out;
1615 			*vpp = vp;
1616 			fvdat = VTOFUD(vp);
1617 
1618 			MPASS(feo != NULL);
1619 			if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1620 				/*
1621 				 * Attributes from the server are definitely
1622 				 * newer than the last attributes we sent to
1623 				 * the server, so cache them.
1624 				 */
1625 				fuse_internal_cache_attrs(*vpp, &feo->attr,
1626 					feo->attr_valid, feo->attr_valid_nsec,
1627 					NULL, true);
1628 			}
1629 			fuse_validity_2_bintime(feo->entry_valid,
1630 				feo->entry_valid_nsec,
1631 				&fvdat->entry_cache_timeout);
1632 
1633 			if ((nameiop == DELETE || nameiop == RENAME) &&
1634 				islastcn && default_permissions)
1635 			{
1636 				struct vattr dvattr;
1637 
1638 				err = fuse_internal_access(dvp, VWRITE, td,
1639 					cred);
1640 				if (err != 0)
1641 					goto out;
1642 				/*
1643 				 * if the parent's sticky bit is set, check
1644 				 * whether we're allowed to remove the file.
1645 				 * Need to figure out the vnode locking to make
1646 				 * this work.
1647 				 */
1648 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1649 				if ((dvattr.va_mode & S_ISTXT) &&
1650 					fuse_internal_access(dvp, VADMIN, td,
1651 						cred) &&
1652 					fuse_internal_access(*vpp, VADMIN, td,
1653 						cred)) {
1654 					err = EPERM;
1655 					goto out;
1656 				}
1657 			}
1658 		}
1659 	}
1660 out:
1661 	if (err) {
1662 		if (vp != NULL && dvp != vp)
1663 			vput(vp);
1664 		else if (vp != NULL)
1665 			vrele(vp);
1666 		*vpp = NULL;
1667 	}
1668 	if (did_lookup)
1669 		fdisp_destroy(&fdi);
1670 
1671 	return err;
1672 }
1673 
1674 /*
1675     struct vnop_mkdir_args {
1676 	struct vnode *a_dvp;
1677 	struct vnode **a_vpp;
1678 	struct componentname *a_cnp;
1679 	struct vattr *a_vap;
1680     };
1681 */
1682 static int
fuse_vnop_mkdir(struct vop_mkdir_args * ap)1683 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1684 {
1685 	struct vnode *dvp = ap->a_dvp;
1686 	struct vnode **vpp = ap->a_vpp;
1687 	struct componentname *cnp = ap->a_cnp;
1688 	struct vattr *vap = ap->a_vap;
1689 
1690 	struct fuse_mkdir_in fmdi;
1691 
1692 	if (fuse_isdeadfs(dvp)) {
1693 		return (EXTERROR(ENXIO, "This FUSE session is about "
1694 		    "to be closed"));
1695 	}
1696 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1697 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1698 
1699 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1700 	    sizeof(fmdi), VDIR));
1701 }
1702 
1703 /*
1704     struct vnop_mknod_args {
1705 	struct vnode *a_dvp;
1706 	struct vnode **a_vpp;
1707 	struct componentname *a_cnp;
1708 	struct vattr *a_vap;
1709     };
1710 */
1711 static int
fuse_vnop_mknod(struct vop_mknod_args * ap)1712 fuse_vnop_mknod(struct vop_mknod_args *ap)
1713 {
1714 
1715 	struct vnode *dvp = ap->a_dvp;
1716 	struct vnode **vpp = ap->a_vpp;
1717 	struct componentname *cnp = ap->a_cnp;
1718 	struct vattr *vap = ap->a_vap;
1719 
1720 	if (fuse_isdeadfs(dvp))
1721 		return (EXTERROR(ENXIO, "This FUSE session is about "
1722 		    "to be closed"));
1723 
1724 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1725 }
1726 
1727 /*
1728     struct vop_open_args {
1729 	struct vnode *a_vp;
1730 	int  a_mode;
1731 	struct ucred *a_cred;
1732 	struct thread *a_td;
1733 	int a_fdidx; / struct file *a_fp;
1734     };
1735 */
1736 static int
fuse_vnop_open(struct vop_open_args * ap)1737 fuse_vnop_open(struct vop_open_args *ap)
1738 {
1739 	struct vnode *vp = ap->a_vp;
1740 	int a_mode = ap->a_mode;
1741 	struct thread *td = ap->a_td;
1742 	struct ucred *cred = ap->a_cred;
1743 	pid_t pid = td->td_proc->p_pid;
1744 
1745 	if (fuse_isdeadfs(vp))
1746 		return (EXTERROR(ENXIO, "This FUSE session is about "
1747 		    "to be closed"));
1748 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1749 		return (EXTERROR(EOPNOTSUPP, "Unsupported vnode type",
1750 		    vp->v_type));
1751 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1752 		return (EXTERROR(EINVAL, "Illegal mode", a_mode));
1753 
1754 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1755 		fuse_vnode_open(vp, 0, td);
1756 		return 0;
1757 	}
1758 
1759 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1760 }
1761 
1762 static int
fuse_vnop_pathconf(struct vop_pathconf_args * ap)1763 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1764 {
1765 	struct vnode *vp = ap->a_vp;
1766 	struct mount *mp;
1767 	struct fuse_filehandle *fufh;
1768 	int err;
1769 	bool closefufh = false;
1770 
1771 	switch (ap->a_name) {
1772 	case _PC_FILESIZEBITS:
1773 		*ap->a_retval = 64;
1774 		return (0);
1775 	case _PC_NAME_MAX:
1776 		*ap->a_retval = NAME_MAX;
1777 		return (0);
1778 	case _PC_LINK_MAX:
1779 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1780 		return (0);
1781 	case _PC_SYMLINK_MAX:
1782 		*ap->a_retval = MAXPATHLEN;
1783 		return (0);
1784 	case _PC_NO_TRUNC:
1785 		*ap->a_retval = 1;
1786 		return (0);
1787 	case _PC_MIN_HOLE_SIZE:
1788 		/*
1789 		 * The FUSE protocol provides no mechanism for a server to
1790 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1791 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1792 		 * 1 if it does.
1793 		 */
1794 		mp = vnode_mount(vp);
1795 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1796 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1797 			off_t offset = 0;
1798 
1799 			/*
1800 			 * Issue a FUSE_LSEEK to find out if it's supported.
1801 			 * Use SEEK_DATA instead of SEEK_HOLE, because the
1802 			 * latter generally requires sequential scans of file
1803 			 * metadata, which can be slow.
1804 			 */
1805 			err = fuse_vnop_do_lseek(vp, curthread,
1806 			    curthread->td_ucred, curthread->td_proc->p_pid,
1807 			    &offset, SEEK_DATA);
1808 			if (err == EBADF) {
1809 				/*
1810 				 * pathconf() doesn't necessarily open the
1811 				 * file.  So we may need to do it here.
1812 				 */
1813 				err = fuse_filehandle_open(vp, FREAD, &fufh,
1814 				    curthread, curthread->td_ucred);
1815 				if (err == 0) {
1816 					closefufh = true;
1817 					err = fuse_vnop_do_lseek(vp, curthread,
1818 					    curthread->td_ucred,
1819 					    curthread->td_proc->p_pid, &offset,
1820 					    SEEK_DATA);
1821 				}
1822 				if (closefufh)
1823 					fuse_filehandle_close(vp, fufh,
1824 					    curthread, curthread->td_ucred);
1825 			}
1826 
1827 		}
1828 
1829 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1830 			*ap->a_retval = 1;
1831 			return (0);
1832 		} else if (fsess_not_impl(mp, FUSE_LSEEK)) {
1833 			/* FUSE_LSEEK is not implemented */
1834 			return (EXTERROR(EINVAL, "This server does not "
1835 			    "implement FUSE_LSEEK"));
1836 		} else {
1837 			return (err);
1838 		}
1839 	default:
1840 		return (vop_stdpathconf(ap));
1841 	}
1842 }
1843 
1844 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1845     "struct uio*", "struct ucred*");
1846 /*
1847     struct vnop_read_args {
1848 	struct vnode *a_vp;
1849 	struct uio *a_uio;
1850 	int  a_ioflag;
1851 	struct ucred *a_cred;
1852     };
1853 */
1854 static int
fuse_vnop_read(struct vop_read_args * ap)1855 fuse_vnop_read(struct vop_read_args *ap)
1856 {
1857 	struct vnode *vp = ap->a_vp;
1858 	struct uio *uio = ap->a_uio;
1859 	int ioflag = ap->a_ioflag;
1860 	struct ucred *cred = ap->a_cred;
1861 	pid_t pid = curthread->td_proc->p_pid;
1862 	struct fuse_filehandle *fufh;
1863 	int err;
1864 	bool closefufh = false, directio;
1865 
1866 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1867 
1868 	if (fuse_isdeadfs(vp)) {
1869 		return (EXTERROR(ENXIO, "This FUSE session is about "
1870 		    "to be closed"));
1871 	}
1872 
1873 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1874 		ioflag |= IO_DIRECT;
1875 	}
1876 
1877 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1878 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1879 		/*
1880 		 * nfsd will do I/O without first doing VOP_OPEN.  We
1881 		 * must implicitly open the file here
1882 		 */
1883 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1884 		closefufh = true;
1885 	}
1886 	if (err) {
1887 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1888 		return err;
1889 	}
1890 
1891 	/*
1892          * Ideally, when the daemon asks for direct io at open time, the
1893          * standard file flag should be set according to this, so that would
1894          * just change the default mode, which later on could be changed via
1895          * fcntl(2).
1896          * But this doesn't work, the O_DIRECT flag gets cleared at some point
1897          * (don't know where). So to make any use of the Fuse direct_io option,
1898          * we hardwire it into the file's private data (similarly to Linux,
1899          * btw.).
1900          */
1901 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1902 
1903 	fuse_vnode_update(vp, FN_ATIMECHANGE);
1904 	if (directio) {
1905 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1906 		err = fuse_read_directbackend(vp, uio, cred, fufh);
1907 	} else {
1908 		SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1909 		err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1910 	}
1911 
1912 	if (closefufh)
1913 		fuse_filehandle_close(vp, fufh, curthread, cred);
1914 
1915 	return (err);
1916 }
1917 
1918 /*
1919     struct vnop_readdir_args {
1920 	struct vnode *a_vp;
1921 	struct uio *a_uio;
1922 	struct ucred *a_cred;
1923 	int *a_eofflag;
1924 	int *a_ncookies;
1925 	uint64_t **a_cookies;
1926     };
1927 */
1928 static int
fuse_vnop_readdir(struct vop_readdir_args * ap)1929 fuse_vnop_readdir(struct vop_readdir_args *ap)
1930 {
1931 	struct vnode *vp = ap->a_vp;
1932 	struct uio *uio = ap->a_uio;
1933 	struct ucred *cred = ap->a_cred;
1934 	struct fuse_filehandle *fufh = NULL;
1935 	struct mount *mp = vnode_mount(vp);
1936 	struct fuse_iov cookediov;
1937 	int err = 0;
1938 	uint64_t *cookies;
1939 	ssize_t tresid;
1940 	int ncookies;
1941 	bool closefufh = false;
1942 	pid_t pid = curthread->td_proc->p_pid;
1943 
1944 	if (ap->a_eofflag)
1945 		*ap->a_eofflag = 0;
1946 	if (fuse_isdeadfs(vp)) {
1947 		return (EXTERROR(ENXIO, "This FUSE session is about "
1948 		    "to be closed"));
1949 	}
1950 	if (uio_resid(uio) < sizeof(struct dirent))
1951 		return (EXTERROR(EINVAL, "Buffer is too small"));
1952 
1953 	tresid = uio->uio_resid;
1954 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1955 	if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1956 		KASSERT(!fsess_is_impl(mp, FUSE_OPENDIR),
1957 			("FUSE file systems that implement "
1958 			 "FUSE_OPENDIR should not be exported"));
1959 		/*
1960 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1961 		 * must implicitly open the directory here.
1962 		 */
1963 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1964 		closefufh = true;
1965 	}
1966 	if (err)
1967 		return (err);
1968 	if (ap->a_ncookies != NULL) {
1969 		ncookies = uio->uio_resid /
1970 			(offsetof(struct dirent, d_name) + 4) + 1;
1971 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1972 		*ap->a_ncookies = ncookies;
1973 		*ap->a_cookies = cookies;
1974 	} else {
1975 		ncookies = 0;
1976 		cookies = NULL;
1977 	}
1978 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1979 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1980 
1981 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1982 		&ncookies, cookies);
1983 
1984 	fiov_teardown(&cookediov);
1985 	if (closefufh)
1986 		fuse_filehandle_close(vp, fufh, curthread, cred);
1987 
1988 	if (ap->a_ncookies != NULL) {
1989 		if (err == 0) {
1990 			*ap->a_ncookies -= ncookies;
1991 		} else {
1992 			free(*ap->a_cookies, M_TEMP);
1993 			*ap->a_ncookies = 0;
1994 			*ap->a_cookies = NULL;
1995 		}
1996 	}
1997 	if (err == 0 && tresid == uio->uio_resid)
1998 		*ap->a_eofflag = 1;
1999 
2000 	return err;
2001 }
2002 
2003 /*
2004     struct vnop_readlink_args {
2005 	struct vnode *a_vp;
2006 	struct uio *a_uio;
2007 	struct ucred *a_cred;
2008     };
2009 */
2010 static int
fuse_vnop_readlink(struct vop_readlink_args * ap)2011 fuse_vnop_readlink(struct vop_readlink_args *ap)
2012 {
2013 	struct vnode *vp = ap->a_vp;
2014 	struct uio *uio = ap->a_uio;
2015 	struct ucred *cred = ap->a_cred;
2016 
2017 	struct fuse_dispatcher fdi;
2018 	int err;
2019 
2020 	if (fuse_isdeadfs(vp)) {
2021 		return (EXTERROR(ENXIO, "This FUSE session is about "
2022 		    "to be closed"));
2023 	}
2024 	if (!vnode_islnk(vp)) {
2025 		return EINVAL;
2026 	}
2027 	fdisp_init(&fdi, 0);
2028 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
2029 	if (err) {
2030 		goto out;
2031 	}
2032 	if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) {
2033 		static const char exterr[] = "Server returned an embedded NUL "
2034 		    "from FUSE_READLINK.";
2035 		struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
2036 		fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, exterr);
2037 		err = EXTERROR(EIO, exterr);
2038 		goto out;
2039 	}
2040 	if (((char *)fdi.answ)[0] == '/' &&
2041 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
2042 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
2043 
2044 		err = uiomove(mpth, strlen(mpth), uio);
2045 	}
2046 	if (!err) {
2047 		err = uiomove(fdi.answ, fdi.iosize, uio);
2048 	}
2049 out:
2050 	fdisp_destroy(&fdi);
2051 	return err;
2052 }
2053 
2054 /*
2055     struct vnop_reclaim_args {
2056 	struct vnode *a_vp;
2057     };
2058 */
2059 static int
fuse_vnop_reclaim(struct vop_reclaim_args * ap)2060 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
2061 {
2062 	struct vnode *vp = ap->a_vp;
2063 	struct thread *td = curthread;
2064 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2065 	struct fuse_filehandle *fufh, *fufh_tmp;
2066 
2067 	if (!fvdat) {
2068 		panic("FUSE: no vnode data during recycling");
2069 	}
2070 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
2071 		printf("FUSE: vnode being reclaimed with open fufh "
2072 			"(type=%#x)", fufh->fufh_type);
2073 		fuse_filehandle_close(vp, fufh, td, NULL);
2074 	}
2075 
2076 	if (VTOI(vp) == 1) {
2077 		/*
2078 		 * Don't send FUSE_FORGET for the root inode, because
2079 		 * we never send FUSE_LOOKUP for it (see
2080 		 * fuse_vfsop_root) and we don't want the server to see
2081 		 * mismatched lookup counts.
2082 		 */
2083 		struct fuse_data *data;
2084 		struct vnode *vroot;
2085 
2086 		data = fuse_get_mpdata(vnode_mount(vp));
2087 		FUSE_LOCK();
2088 		vroot = data->vroot;
2089 		data->vroot = NULL;
2090 		FUSE_UNLOCK();
2091 		if (vroot)
2092 			vrele(vroot);
2093 	} else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2094 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2095 		    fvdat->nlookup);
2096 	}
2097 	cache_purge(vp);
2098 	vfs_hash_remove(vp);
2099 	fuse_vnode_destroy(vp);
2100 
2101 	return 0;
2102 }
2103 
2104 /*
2105     struct vnop_remove_args {
2106 	struct vnode *a_dvp;
2107 	struct vnode *a_vp;
2108 	struct componentname *a_cnp;
2109     };
2110 */
2111 static int
fuse_vnop_remove(struct vop_remove_args * ap)2112 fuse_vnop_remove(struct vop_remove_args *ap)
2113 {
2114 	struct vnode *dvp = ap->a_dvp;
2115 	struct vnode *vp = ap->a_vp;
2116 	struct componentname *cnp = ap->a_cnp;
2117 
2118 	int err;
2119 
2120 	if (fuse_isdeadfs(vp)) {
2121 		return (EXTERROR(ENXIO, "This FUSE session is about "
2122 		    "to be closed"));
2123 	}
2124 	if (vnode_isdir(vp)) {
2125 		return (EXTERROR(EPERM, "vnode is a directory"));
2126 	}
2127 
2128 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2129 
2130 	return err;
2131 }
2132 
2133 /*
2134     struct vnop_rename_args {
2135 	struct vnode *a_fdvp;
2136 	struct vnode *a_fvp;
2137 	struct componentname *a_fcnp;
2138 	struct vnode *a_tdvp;
2139 	struct vnode *a_tvp;
2140 	struct componentname *a_tcnp;
2141     };
2142 */
2143 static int
fuse_vnop_rename(struct vop_rename_args * ap)2144 fuse_vnop_rename(struct vop_rename_args *ap)
2145 {
2146 	struct vnode *fdvp = ap->a_fdvp;
2147 	struct vnode *fvp = ap->a_fvp;
2148 	struct componentname *fcnp = ap->a_fcnp;
2149 	struct vnode *tdvp = ap->a_tdvp;
2150 	struct vnode *tvp = ap->a_tvp;
2151 	struct componentname *tcnp = ap->a_tcnp;
2152 	struct fuse_data *data;
2153 	bool newparent = fdvp != tdvp;
2154 	bool isdir = fvp->v_type == VDIR;
2155 	int err = 0;
2156 
2157 	if (fuse_isdeadfs(fdvp)) {
2158 		return (EXTERROR(ENXIO, "This FUSE session is about "
2159 		    "to be closed"));
2160 	}
2161 	if (fvp->v_mount != tdvp->v_mount ||
2162 	    (tvp && fvp->v_mount != tvp->v_mount)) {
2163 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2164 		err = EXTERROR(EXDEV, "Cross-device rename");
2165 		goto out;
2166 	}
2167 	cache_purge(fvp);
2168 
2169 	/*
2170 	 * FUSE library is expected to check if target directory is not
2171 	 * under the source directory in the file system tree.
2172 	 * Linux performs this check at VFS level.
2173 	 */
2174 	/*
2175 	 * If source is a directory, and it will get a new parent, user must
2176 	 * have write permission to it, so ".." can be modified.
2177 	 */
2178 	data = fuse_get_mpdata(vnode_mount(tdvp));
2179 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2180 		err = fuse_internal_access(fvp, VWRITE,
2181 			curthread, tcnp->cn_cred);
2182 		if (err)
2183 			goto out;
2184 	}
2185 	sx_xlock(&data->rename_lock);
2186 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2187 	if (err == 0) {
2188 		if (tdvp != fdvp)
2189 			fuse_vnode_setparent(fvp, tdvp);
2190 		if (tvp != NULL)
2191 			fuse_vnode_setparent(tvp, NULL);
2192 	}
2193 	sx_unlock(&data->rename_lock);
2194 
2195 	if (tvp != NULL && tvp != fvp) {
2196 		cache_purge(tvp);
2197 	}
2198 	if (vnode_isdir(fvp)) {
2199 		if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2200 			cache_purge(tdvp);
2201 		}
2202 		cache_purge(fdvp);
2203 	}
2204 out:
2205 	if (tdvp == tvp) {
2206 		vrele(tdvp);
2207 	} else {
2208 		vput(tdvp);
2209 	}
2210 	if (tvp != NULL) {
2211 		vput(tvp);
2212 	}
2213 	vrele(fdvp);
2214 	vrele(fvp);
2215 
2216 	return err;
2217 }
2218 
2219 /*
2220     struct vnop_rmdir_args {
2221 	    struct vnode *a_dvp;
2222 	    struct vnode *a_vp;
2223 	    struct componentname *a_cnp;
2224     } *ap;
2225 */
2226 static int
fuse_vnop_rmdir(struct vop_rmdir_args * ap)2227 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2228 {
2229 	struct vnode *dvp = ap->a_dvp;
2230 	struct vnode *vp = ap->a_vp;
2231 
2232 	int err;
2233 
2234 	if (fuse_isdeadfs(vp)) {
2235 		return (EXTERROR(ENXIO, "This FUSE session is about "
2236 		    "to be closed"));
2237 	}
2238 	if (VTOFUD(vp) == VTOFUD(dvp)) {
2239 		return (EXTERROR(EINVAL, "Directory to be removed "
2240 		    "contains itself"));
2241 	}
2242 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2243 
2244 	return err;
2245 }
2246 
2247 /*
2248     struct vnop_setattr_args {
2249 	struct vnode *a_vp;
2250 	struct vattr *a_vap;
2251 	struct ucred *a_cred;
2252 	struct thread *a_td;
2253     };
2254 */
2255 static int
fuse_vnop_setattr(struct vop_setattr_args * ap)2256 fuse_vnop_setattr(struct vop_setattr_args *ap)
2257 {
2258 	struct vnode *vp = ap->a_vp;
2259 	struct vattr *vap = ap->a_vap;
2260 	struct ucred *cred = ap->a_cred;
2261 	struct thread *td = curthread;
2262 	struct mount *mp;
2263 	struct fuse_data *data;
2264 	struct vattr old_va;
2265 	int dataflags;
2266 	int err = 0, err2;
2267 	accmode_t accmode = 0;
2268 	bool checkperm;
2269 	bool drop_suid = false;
2270 
2271 	mp = vnode_mount(vp);
2272 	data = fuse_get_mpdata(mp);
2273 	dataflags = data->dataflags;
2274 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2275 
2276 	if (fuse_isdeadfs(vp)) {
2277 		return (EXTERROR(ENXIO, "This FUSE session is about "
2278 		    "to be closed"));
2279 	}
2280 
2281 	if (vap->va_uid != (uid_t)VNOVAL) {
2282 		if (checkperm) {
2283 			/* Only root may change a file's owner */
2284 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2285 			if (err) {
2286 				/* As a special case, allow the null chown */
2287 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2288 					td);
2289 				if (err2)
2290 					return (err2);
2291 				if (vap->va_uid != old_va.va_uid)
2292 					return err;
2293 				drop_suid = true;
2294 			}
2295 		}
2296 		accmode |= VADMIN;
2297 	}
2298 	if (vap->va_gid != (gid_t)VNOVAL) {
2299 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2300 			drop_suid = true;
2301 		if (checkperm && !groupmember(vap->va_gid, cred)) {
2302 			/*
2303 			 * Non-root users may only chgrp to one of their own
2304 			 * groups
2305 			 */
2306 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2307 			if (err) {
2308 				/* As a special case, allow the null chgrp */
2309 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2310 					td);
2311 				if (err2)
2312 					return (err2);
2313 				if (vap->va_gid != old_va.va_gid)
2314 					return err;
2315 			}
2316 		}
2317 		accmode |= VADMIN;
2318 	}
2319 	if (vap->va_size != VNOVAL) {
2320 		switch (vp->v_type) {
2321 		case VDIR:
2322 			return (EISDIR);
2323 		case VLNK:
2324 		case VREG:
2325 			if (vfs_isrdonly(mp))
2326 				return (EROFS);
2327 			err = vn_rlimit_trunc(vap->va_size, td);
2328 			if (err)
2329 				return (err);
2330 			break;
2331 		default:
2332 			/*
2333 			 * According to POSIX, the result is unspecified
2334 			 * for file types other than regular files,
2335 			 * directories and shared memory objects.  We
2336 			 * don't support shared memory objects in the file
2337 			 * system, and have dubious support for truncating
2338 			 * symlinks.  Just ignore the request in other cases.
2339 			 */
2340 			return (0);
2341 		}
2342 		/* Don't set accmode.  Permission to trunc is checked upstack */
2343 	}
2344 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2345 		if (vap->va_vaflags & VA_UTIMES_NULL)
2346 			accmode |= VWRITE;
2347 		else
2348 			accmode |= VADMIN;
2349 	}
2350 	if (drop_suid) {
2351 		if (vap->va_mode != (mode_t)VNOVAL)
2352 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2353 		else {
2354 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2355 			if (err)
2356 				return (err);
2357 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2358 		}
2359 	}
2360 	if (vap->va_mode != (mode_t)VNOVAL) {
2361 		/* Only root may set the sticky bit on non-directories */
2362 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2363 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2364 			return EFTYPE;
2365 		if (checkperm && (vap->va_mode & S_ISGID)) {
2366 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2367 			if (err)
2368 				return (err);
2369 			if (!groupmember(old_va.va_gid, cred)) {
2370 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2371 				if (err)
2372 					return (err);
2373 			}
2374 		}
2375 		accmode |= VADMIN;
2376 	}
2377 
2378 	if (vfs_isrdonly(mp))
2379 		return EROFS;
2380 
2381 	if (checkperm) {
2382 		err = fuse_internal_access(vp, accmode, td, cred);
2383 	} else {
2384 		err = 0;
2385 	}
2386 	if (err)
2387 		return err;
2388 	else
2389 		return fuse_internal_setattr(vp, vap, td, cred);
2390 }
2391 
2392 /*
2393     struct vnop_strategy_args {
2394 	struct vnode *a_vp;
2395 	struct buf *a_bp;
2396     };
2397 */
2398 static int
fuse_vnop_strategy(struct vop_strategy_args * ap)2399 fuse_vnop_strategy(struct vop_strategy_args *ap)
2400 {
2401 	struct vnode *vp = ap->a_vp;
2402 	struct buf *bp = ap->a_bp;
2403 
2404 	if (!vp || fuse_isdeadfs(vp)) {
2405 		bp->b_ioflags |= BIO_ERROR;
2406 		bp->b_error = ENXIO;
2407 		bufdone(bp);
2408 		return 0;
2409 	}
2410 
2411 	/*
2412 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2413 	 * fuse_io_strategy sets bp's error fields
2414 	 */
2415 	(void)fuse_io_strategy(vp, bp);
2416 
2417 	return 0;
2418 }
2419 
2420 /*
2421     struct vnop_symlink_args {
2422 	struct vnode *a_dvp;
2423 	struct vnode **a_vpp;
2424 	struct componentname *a_cnp;
2425 	struct vattr *a_vap;
2426 	char *a_target;
2427     };
2428 */
2429 static int
fuse_vnop_symlink(struct vop_symlink_args * ap)2430 fuse_vnop_symlink(struct vop_symlink_args *ap)
2431 {
2432 	struct vnode *dvp = ap->a_dvp;
2433 	struct vnode **vpp = ap->a_vpp;
2434 	struct componentname *cnp = ap->a_cnp;
2435 	const char *target = ap->a_target;
2436 
2437 	struct fuse_dispatcher fdi;
2438 
2439 	int err;
2440 	size_t len;
2441 
2442 	if (fuse_isdeadfs(dvp)) {
2443 		return (EXTERROR(ENXIO, "This FUSE session is about "
2444 		    "to be closed"));
2445 	}
2446 	/*
2447 	 * Unlike the other creator type calls, here we have to create a message
2448 	 * where the name of the new entry comes first, and the data describing
2449 	 * the entry comes second.
2450 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2451 	 * but put together the message manually and just call the core part.
2452 	 */
2453 
2454 	len = strlen(target) + 1;
2455 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2456 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2457 
2458 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2459 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2460 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2461 
2462 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2463 	fdisp_destroy(&fdi);
2464 	return err;
2465 }
2466 
2467 /*
2468     struct vnop_write_args {
2469 	struct vnode *a_vp;
2470 	struct uio *a_uio;
2471 	int  a_ioflag;
2472 	struct ucred *a_cred;
2473     };
2474 */
2475 static int
fuse_vnop_write(struct vop_write_args * ap)2476 fuse_vnop_write(struct vop_write_args *ap)
2477 {
2478 	struct vnode *vp = ap->a_vp;
2479 	struct uio *uio = ap->a_uio;
2480 	int ioflag = ap->a_ioflag;
2481 	struct ucred *cred = ap->a_cred;
2482 	pid_t pid = curthread->td_proc->p_pid;
2483 	struct fuse_filehandle *fufh;
2484 	int err;
2485 	bool closefufh = false, directio;
2486 
2487 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2488 
2489 	if (fuse_isdeadfs(vp)) {
2490 		return (EXTERROR(ENXIO, "This FUSE session is about "
2491 		    "to be closed"));
2492 	}
2493 
2494 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2495 		ioflag |= IO_DIRECT;
2496 	}
2497 
2498 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2499 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2500 		/*
2501 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2502 		 * must implicitly open the file here
2503 		 */
2504 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2505 		closefufh = true;
2506 	}
2507 	if (err) {
2508 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2509 		return err;
2510 	}
2511 
2512 	/*
2513          * Ideally, when the daemon asks for direct io at open time, the
2514          * standard file flag should be set according to this, so that would
2515          * just change the default mode, which later on could be changed via
2516          * fcntl(2).
2517          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2518          * (don't know where). So to make any use of the Fuse direct_io option,
2519          * we hardwire it into the file's private data (similarly to Linux,
2520          * btw.).
2521          */
2522 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2523 
2524 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2525 	if (directio) {
2526 		off_t start, end, filesize;
2527 		bool pages = (ioflag & IO_VMIO) != 0;
2528 
2529 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2530 
2531 		err = fuse_vnode_size(vp, &filesize, cred, curthread);
2532 		if (err)
2533 			goto out;
2534 
2535 		start = uio->uio_offset;
2536 		end = start + uio->uio_resid;
2537 		if (!pages) {
2538 			err = fuse_inval_buf_range(vp, filesize, start,
2539 			    end);
2540 			if (err)
2541 				goto out;
2542 		}
2543 		err = fuse_write_directbackend(vp, uio, cred, fufh,
2544 			filesize, ioflag, pages);
2545 	} else {
2546 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2547 			"buffered write of vnode");
2548 		if (!fsess_opt_writeback(vnode_mount(vp)))
2549 			ioflag |= IO_SYNC;
2550 		err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2551 	}
2552 	fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2553 
2554 out:
2555 	if (closefufh)
2556 		fuse_filehandle_close(vp, fufh, curthread, cred);
2557 
2558 	return (err);
2559 }
2560 
2561 static daddr_t
fuse_gbp_getblkno(struct vnode * vp,vm_ooffset_t off)2562 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2563 {
2564 	const int biosize = fuse_iosize(vp);
2565 
2566 	return (off / biosize);
2567 }
2568 
2569 static int
fuse_gbp_getblksz(struct vnode * vp,daddr_t lbn,long * blksz)2570 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2571 {
2572 	off_t filesize;
2573 	int err;
2574 	const int biosize = fuse_iosize(vp);
2575 
2576 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2577 	if (err) {
2578 		/* This will turn into a SIGBUS */
2579 		return (EIO);
2580 	} else if ((off_t)lbn * biosize >= filesize) {
2581 		*blksz = 0;
2582 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2583 		*blksz = filesize - (off_t)lbn *biosize;
2584 	} else {
2585 		*blksz = biosize;
2586 	}
2587 	return (0);
2588 }
2589 
2590 /*
2591     struct vnop_getpages_args {
2592 	struct vnode *a_vp;
2593 	vm_page_t *a_m;
2594 	int a_count;
2595 	int a_reqpage;
2596     };
2597 */
2598 static int
fuse_vnop_getpages(struct vop_getpages_args * ap)2599 fuse_vnop_getpages(struct vop_getpages_args *ap)
2600 {
2601 	struct vnode *vp = ap->a_vp;
2602 
2603 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2604 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2605 			"called on non-cacheable vnode??\n");
2606 		return (VM_PAGER_ERROR);
2607 	}
2608 
2609 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2610 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2611 }
2612 
2613 static const char extattr_namespace_separator = '.';
2614 
2615 /*
2616     struct vop_getextattr_args {
2617 	struct vop_generic_args a_gen;
2618 	struct vnode *a_vp;
2619 	int a_attrnamespace;
2620 	const char *a_name;
2621 	struct uio *a_uio;
2622 	size_t *a_size;
2623 	struct ucred *a_cred;
2624 	struct thread *a_td;
2625     };
2626 */
2627 static int
fuse_vnop_getextattr(struct vop_getextattr_args * ap)2628 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2629 {
2630 	struct vnode *vp = ap->a_vp;
2631 	struct uio *uio = ap->a_uio;
2632 	struct fuse_dispatcher fdi;
2633 	struct fuse_getxattr_in *get_xattr_in;
2634 	struct fuse_getxattr_out *get_xattr_out;
2635 	struct mount *mp = vnode_mount(vp);
2636 	struct thread *td = ap->a_td;
2637 	struct ucred *cred = ap->a_cred;
2638 	char *prefix;
2639 	char *attr_str;
2640 	size_t len;
2641 	int err;
2642 
2643 	if (fuse_isdeadfs(vp))
2644 		return (EXTERROR(ENXIO, "This FUSE session is about "
2645 		    "to be closed"));
2646 
2647 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2648 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2649 		    "extended attributes"));
2650 
2651 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2652 	if (err)
2653 		return err;
2654 
2655 	/* Default to looking for user attributes. */
2656 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2657 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2658 	else
2659 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2660 
2661 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2662 	    strlen(ap->a_name) + 1;
2663 
2664 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2665 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2666 
2667 	get_xattr_in = fdi.indata;
2668 	/*
2669 	 * Check to see whether we're querying the available size or
2670 	 * issuing the actual request.  If we pass in 0, we get back struct
2671 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2672 	 * that much data, without the struct fuse_getxattr_out header.
2673 	 */
2674 	if (uio == NULL)
2675 		get_xattr_in->size = 0;
2676 	else
2677 		get_xattr_in->size = uio->uio_resid;
2678 
2679 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2680 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2681 	    ap->a_name);
2682 
2683 	err = fdisp_wait_answ(&fdi);
2684 	if (err != 0) {
2685 		if (err == ENOSYS) {
2686 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2687 			err = (EXTERROR(EOPNOTSUPP, "This server does not "
2688 			    "implement extended attributes"));
2689 		}
2690 		goto out;
2691 	}
2692 
2693 	get_xattr_out = fdi.answ;
2694 
2695 	if (ap->a_size != NULL)
2696 		*ap->a_size = get_xattr_out->size;
2697 
2698 	if (uio != NULL)
2699 		err = uiomove(fdi.answ, fdi.iosize, uio);
2700 
2701 out:
2702 	fdisp_destroy(&fdi);
2703 	return (err);
2704 }
2705 
2706 /*
2707     struct vop_setextattr_args {
2708 	struct vop_generic_args a_gen;
2709 	struct vnode *a_vp;
2710 	int a_attrnamespace;
2711 	const char *a_name;
2712 	struct uio *a_uio;
2713 	struct ucred *a_cred;
2714 	struct thread *a_td;
2715     };
2716 */
2717 static int
fuse_vnop_setextattr(struct vop_setextattr_args * ap)2718 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2719 {
2720 	struct vnode *vp = ap->a_vp;
2721 	struct uio *uio = ap->a_uio;
2722 	struct fuse_dispatcher fdi;
2723 	struct fuse_setxattr_in *set_xattr_in;
2724 	struct mount *mp = vnode_mount(vp);
2725 	struct thread *td = ap->a_td;
2726 	struct ucred *cred = ap->a_cred;
2727 	size_t struct_size = FUSE_COMPAT_SETXATTR_IN_SIZE;
2728 	char *prefix;
2729 	size_t len;
2730 	char *attr_str;
2731 	int err;
2732 
2733 	if (fuse_isdeadfs(vp))
2734 		return (EXTERROR(ENXIO, "This FUSE session is about "
2735 		    "to be closed"));
2736 
2737 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2738 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2739 		    "setting extended attributes"));
2740 
2741 	if (vfs_isrdonly(mp))
2742 		return EROFS;
2743 
2744 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2745 	if (ap->a_uio == NULL) {
2746 		/*
2747 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2748 		 * return EOPNOTSUPP.
2749 		 */
2750 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2751 			return (EXTERROR(EOPNOTSUPP, "This server does not "
2752 			    "implement removing extended attributess"));
2753 		else
2754 			return (EXTERROR(EINVAL, "DELETEEXTATTR should be used "
2755 			    "to remove extattrs"));
2756 	}
2757 
2758 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2759 		VWRITE);
2760 	if (err)
2761 		return err;
2762 
2763 	/* Default to looking for user attributes. */
2764 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2765 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2766 	else
2767 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2768 
2769 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2770 	    strlen(ap->a_name) + 1;
2771 
2772 	/* older FUSE servers  use a smaller fuse_setxattr_in struct*/
2773 	if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33))
2774 		struct_size = sizeof(*set_xattr_in);
2775 
2776 	fdisp_init(&fdi, len + struct_size + uio->uio_resid);
2777 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2778 
2779 	set_xattr_in = fdi.indata;
2780 	set_xattr_in->size = uio->uio_resid;
2781 
2782 	if (fuse_libabi_geq(fuse_get_mpdata(mp), 7, 33)) {
2783 		set_xattr_in->setxattr_flags = 0;
2784 		set_xattr_in->padding = 0;
2785 	}
2786 
2787 	attr_str = (char *)fdi.indata + struct_size;
2788 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2789 	    ap->a_name);
2790 
2791 	err = uiomove((char *)fdi.indata + struct_size + len,
2792 	    uio->uio_resid, uio);
2793 	if (err != 0) {
2794 		goto out;
2795 	}
2796 
2797 	err = fdisp_wait_answ(&fdi);
2798 
2799 	if (err == ENOSYS) {
2800 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2801 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
2802 		    "setting extended attributes");
2803 	}
2804 	if (err == ERESTART) {
2805 		/* Can't restart after calling uiomove */
2806 		err = EINTR;
2807 	}
2808 
2809 out:
2810 	fdisp_destroy(&fdi);
2811 	return (err);
2812 }
2813 
2814 /*
2815  * The Linux / FUSE extended attribute list is simply a collection of
2816  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2817  * byte length followed by a non-NUL terminated string.  So, this allows
2818  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2819  * Linux attribute names are reported with the namespace as a prefix (e.g.
2820  * "user.attribute_name"), but in FreeBSD they are reported without the
2821  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2822  *
2823  * user.attr_name1\0user.attr_name2\0
2824  *
2825  * to:
2826  *
2827  * <num>attr_name1<num>attr_name2
2828  *
2829  * Where "<num>" is a single byte number of characters in the attribute name.
2830  *
2831  * Args:
2832  * prefix - exattr namespace prefix string
2833  * list, list_len - input list with namespace prefixes
2834  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2835  */
2836 static int
fuse_xattrlist_convert(char * prefix,const char * list,int list_len,char * bsd_list,int * bsd_list_len)2837 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2838     char *bsd_list, int *bsd_list_len)
2839 {
2840 	int len, pos, dist_to_next, prefix_len;
2841 
2842 	pos = 0;
2843 	*bsd_list_len = 0;
2844 	prefix_len = strlen(prefix);
2845 
2846 	while (pos < list_len && list[pos] != '\0') {
2847 		dist_to_next = strlen(&list[pos]) + 1;
2848 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2849 		    list[pos + prefix_len] == extattr_namespace_separator) {
2850 			len = dist_to_next -
2851 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2852 			if (len >= EXTATTR_MAXNAMELEN)
2853 				return (ENAMETOOLONG);
2854 
2855 			bsd_list[*bsd_list_len] = len;
2856 			memcpy(&bsd_list[*bsd_list_len + 1],
2857 			    &list[pos + prefix_len +
2858 			    sizeof(extattr_namespace_separator)], len);
2859 
2860 			*bsd_list_len += len + 1;
2861 		}
2862 
2863 		pos += dist_to_next;
2864 	}
2865 
2866 	return (0);
2867 }
2868 
2869 /*
2870  * List extended attributes
2871  *
2872  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2873  * has a number of differences compared to its FreeBSD equivalent,
2874  * extattr_list_file:
2875  *
2876  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2877  *   whereas listxattr(2) only returns attributes for a single namespace
2878  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2879  * - If the provided buffer is not large enough to hold the result,
2880  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2881  *   return as many results as will fit.
2882  */
2883 /*
2884     struct vop_listextattr_args {
2885 	struct vop_generic_args a_gen;
2886 	struct vnode *a_vp;
2887 	int a_attrnamespace;
2888 	struct uio *a_uio;
2889 	size_t *a_size;
2890 	struct ucred *a_cred;
2891 	struct thread *a_td;
2892     };
2893 */
2894 static int
fuse_vnop_listextattr(struct vop_listextattr_args * ap)2895 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2896 {
2897 	struct vnode *vp = ap->a_vp;
2898 	struct uio *uio = ap->a_uio;
2899 	struct fuse_dispatcher fdi;
2900 	struct fuse_listxattr_in *list_xattr_in;
2901 	struct fuse_listxattr_out *list_xattr_out;
2902 	struct mount *mp = vnode_mount(vp);
2903 	struct thread *td = ap->a_td;
2904 	struct ucred *cred = ap->a_cred;
2905 	char *prefix;
2906 	char *bsd_list = NULL;
2907 	char *linux_list;
2908 	int bsd_list_len;
2909 	int linux_list_len;
2910 	int err;
2911 
2912 	if (fuse_isdeadfs(vp))
2913 		return (EXTERROR(ENXIO, "This FUSE session is about "
2914 		    "to be closed"));
2915 
2916 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
2917 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
2918 		    "extended attributes"));
2919 
2920 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2921 	if (err)
2922 		return err;
2923 
2924 	/*
2925 	 * Add space for a NUL and the period separator if enabled.
2926 	 * Default to looking for user attributes.
2927 	 */
2928 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2929 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2930 	else
2931 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2932 
2933 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2934 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2935 
2936 	/*
2937 	 * Retrieve Linux / FUSE compatible list size.
2938 	 */
2939 	list_xattr_in = fdi.indata;
2940 	list_xattr_in->size = 0;
2941 
2942 	err = fdisp_wait_answ(&fdi);
2943 	if (err != 0) {
2944 		if (err == ENOSYS) {
2945 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2946 			err = EXTERROR(EOPNOTSUPP, "This server does not "
2947 			    "implement extended attributes");
2948 		}
2949 		goto out;
2950 	}
2951 
2952 	list_xattr_out = fdi.answ;
2953 	linux_list_len = list_xattr_out->size;
2954 	if (linux_list_len == 0) {
2955 		if (ap->a_size != NULL)
2956 			*ap->a_size = linux_list_len;
2957 		goto out;
2958 	}
2959 
2960 	/*
2961 	 * Retrieve Linux / FUSE compatible list values.
2962 	 */
2963 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2964 	list_xattr_in = fdi.indata;
2965 	list_xattr_in->size = linux_list_len;
2966 
2967 	err = fdisp_wait_answ(&fdi);
2968 	if (err == ERANGE) {
2969 		/*
2970 		 * Race detected.  The attribute list must've grown since the
2971 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2972 		 * to userland so we can process signals, if necessary, before
2973 		 * restarting.
2974 		 */
2975 		err = ERESTART;
2976 		goto out;
2977 	} else if (err != 0)
2978 		goto out;
2979 
2980 	linux_list = fdi.answ;
2981 	/* FUSE doesn't allow the server to return more data than requested */
2982 	if (fdi.iosize > linux_list_len) {
2983 		struct fuse_data *data = fuse_get_mpdata(mp);
2984 
2985 		fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2986 			"server returned "
2987 			"more extended attribute data than requested; "
2988 			"should've returned ERANGE instead.");
2989 	} else {
2990 		/* But returning less data is fine */
2991 		linux_list_len = fdi.iosize;
2992 	}
2993 
2994 	/*
2995 	 * Retrieve the BSD compatible list values.
2996 	 * The Linux / FUSE attribute list format isn't the same
2997 	 * as FreeBSD's format. So we need to transform it into
2998 	 * FreeBSD's format before giving it to the user.
2999 	 */
3000 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
3001 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
3002 	    bsd_list, &bsd_list_len);
3003 	if (err != 0)
3004 		goto out;
3005 
3006 	if (ap->a_size != NULL)
3007 		*ap->a_size = bsd_list_len;
3008 
3009 	if (uio != NULL)
3010 		err = uiomove(bsd_list, bsd_list_len, uio);
3011 
3012 out:
3013 	free(bsd_list, M_TEMP);
3014 	fdisp_destroy(&fdi);
3015 	return (err);
3016 }
3017 
3018 /*
3019     struct vop_deallocate_args {
3020 	struct vop_generic_args a_gen;
3021 	struct vnode *a_vp;
3022 	off_t *a_offset;
3023 	off_t *a_len;
3024 	int a_flags;
3025 	int a_ioflag;
3026         struct ucred *a_cred;
3027     };
3028 */
3029 static int
fuse_vnop_deallocate(struct vop_deallocate_args * ap)3030 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
3031 {
3032 	struct vnode *vp = ap->a_vp;
3033 	struct mount *mp = vnode_mount(vp);
3034 	struct fuse_filehandle *fufh;
3035 	struct fuse_dispatcher fdi;
3036 	struct fuse_fallocate_in *ffi;
3037 	struct ucred *cred = ap->a_cred;
3038 	pid_t pid = curthread->td_proc->p_pid;
3039 	off_t *len = ap->a_len;
3040 	off_t *offset = ap->a_offset;
3041 	int ioflag = ap->a_ioflag;
3042 	off_t filesize;
3043 	int err;
3044 	bool closefufh = false;
3045 
3046 	if (fuse_isdeadfs(vp))
3047 		return (EXTERROR(ENXIO, "This FUSE session is about "
3048 		    "to be closed"));
3049 
3050 	if (vfs_isrdonly(mp))
3051 		return (EROFS);
3052 
3053 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
3054 		goto fallback;
3055 
3056 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
3057 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
3058 		/*
3059 		 * nfsd will do I/O without first doing VOP_OPEN.  We
3060 		 * must implicitly open the file here
3061 		 */
3062 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
3063 		closefufh = true;
3064 	}
3065 	if (err)
3066 		return (err);
3067 
3068 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
3069 
3070 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
3071 	if (err)
3072 		goto out;
3073 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
3074 
3075 	fdisp_init(&fdi, sizeof(*ffi));
3076 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
3077 	ffi = fdi.indata;
3078 	ffi->fh = fufh->fh_id;
3079 	ffi->offset = *offset;
3080 	ffi->length = *len;
3081 	/*
3082 	 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
3083 	 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
3084 	 */
3085 	ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
3086 	err = fdisp_wait_answ(&fdi);
3087 
3088 	if (err == ENOSYS) {
3089 		fdisp_destroy(&fdi);
3090 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
3091 		goto fallback;
3092 	} else if (err == EOPNOTSUPP) {
3093 		/*
3094 		 * The file system server does not support FUSE_FALLOCATE with
3095 		 * the supplied mode for this particular file.
3096 		 */
3097 		fdisp_destroy(&fdi);
3098 		goto fallback;
3099 	} else if (!err) {
3100 		/*
3101 		 * Clip the returned offset to EoF.  Do it here rather than
3102 		 * before FUSE_FALLOCATE just in case the kernel's cached file
3103 		 * size is out of date.  Unfortunately, FUSE does not return
3104 		 * any information about filesize from that operation.
3105 		 */
3106 		*offset = MIN(*offset + *len, filesize);
3107 		*len = 0;
3108 		fuse_vnode_undirty_cached_timestamps(vp, false);
3109 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
3110 
3111 		if (ioflag & IO_SYNC)
3112 			err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3113 			    false);
3114 	}
3115 
3116 	fdisp_destroy(&fdi);
3117 out:
3118 	if (closefufh)
3119 		fuse_filehandle_close(vp, fufh, curthread, cred);
3120 
3121 	return (err);
3122 
3123 fallback:
3124 	if (closefufh)
3125 		fuse_filehandle_close(vp, fufh, curthread, cred);
3126 
3127 	return (vop_stddeallocate(ap));
3128 }
3129 
3130 /*
3131     struct vop_deleteextattr_args {
3132 	struct vop_generic_args a_gen;
3133 	struct vnode *a_vp;
3134 	int a_attrnamespace;
3135 	const char *a_name;
3136 	struct ucred *a_cred;
3137 	struct thread *a_td;
3138     };
3139 */
3140 static int
fuse_vnop_deleteextattr(struct vop_deleteextattr_args * ap)3141 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3142 {
3143 	struct vnode *vp = ap->a_vp;
3144 	struct fuse_dispatcher fdi;
3145 	struct mount *mp = vnode_mount(vp);
3146 	struct thread *td = ap->a_td;
3147 	struct ucred *cred = ap->a_cred;
3148 	char *prefix;
3149 	size_t len;
3150 	char *attr_str;
3151 	int err;
3152 
3153 	if (fuse_isdeadfs(vp))
3154 		return (EXTERROR(ENXIO, "This FUSE session is about "
3155 		    "to be closed"));
3156 
3157 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3158 		return (EXTERROR(EOPNOTSUPP, "This server does not implement "
3159 		    "removing extended attributes"));
3160 
3161 	if (vfs_isrdonly(mp))
3162 		return EROFS;
3163 
3164 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3165 		VWRITE);
3166 	if (err)
3167 		return err;
3168 
3169 	/* Default to looking for user attributes. */
3170 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3171 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3172 	else
3173 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3174 
3175 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3176 	    strlen(ap->a_name) + 1;
3177 
3178 	fdisp_init(&fdi, len);
3179 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3180 
3181 	attr_str = fdi.indata;
3182 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3183 	    ap->a_name);
3184 
3185 	err = fdisp_wait_answ(&fdi);
3186 	if (err == ENOSYS) {
3187 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3188 		err = EXTERROR(EOPNOTSUPP, "This server does not implement "
3189 		    "removing extended attributes");
3190 	}
3191 
3192 	fdisp_destroy(&fdi);
3193 	return (err);
3194 }
3195 
3196 /*
3197     struct vnop_print_args {
3198 	struct vnode *a_vp;
3199     };
3200 */
3201 static int
fuse_vnop_print(struct vop_print_args * ap)3202 fuse_vnop_print(struct vop_print_args *ap)
3203 {
3204 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3205 
3206 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3207 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3208 	    (uintmax_t)fvdat->nlookup,
3209 	    fvdat->flag);
3210 
3211 	return 0;
3212 }
3213 
3214 /*
3215  * Get an NFS filehandle for a FUSE file.
3216  *
3217  * This will only work for FUSE file systems that guarantee the uniqueness of
3218  * nodeid:generation, which most don't.
3219  */
3220 /*
3221 vop_vptofh {
3222 	IN struct vnode *a_vp;
3223 	IN struct fid *a_fhp;
3224 };
3225 */
3226 static int
fuse_vnop_vptofh(struct vop_vptofh_args * ap)3227 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3228 {
3229 	struct vnode *vp = ap->a_vp;
3230 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
3231 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3232 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3233 		"FUSE fid type is too big");
3234 	struct mount *mp = vnode_mount(vp);
3235 	struct fuse_data *data = fuse_get_mpdata(mp);
3236 	struct vattr va;
3237 	int err;
3238 
3239 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3240 		/* NFS requires lookups for "." and ".." */
3241 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3242 			"VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3243 		return (EXTERROR(EOPNOTSUPP, "This server is "
3244 		    "missing FUSE_EXPORT_SUPPORT"));
3245 	}
3246 	if ((mp->mnt_flag & MNT_EXPORTED) &&
3247 		fsess_is_impl(mp, FUSE_OPENDIR))
3248 	{
3249 		/*
3250 		 * NFS is stateless, so nfsd must reopen a directory on every
3251 		 * call to VOP_READDIR, passing in the d_off field from the
3252 		 * final dirent of the previous invocation.  But if the server
3253 		 * implements FUSE_OPENDIR, the FUSE protocol does not
3254 		 * guarantee that d_off will be valid after a directory is
3255 		 * closed and reopened.  So prohibit exporting FUSE file
3256 		 * systems that implement FUSE_OPENDIR.
3257 		 *
3258 		 * But userspace NFS servers don't have this problem.
3259                  */
3260 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3261 			"VOP_VPTOFH with FUSE_OPENDIR");
3262 		return (EXTERROR(EOPNOTSUPP, "This server implements "
3263 		    "FUSE_OPENDIR so is not compatible with getfh"));
3264 	}
3265 
3266 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3267 	if (err)
3268 		return err;
3269 
3270 	/*ip = VTOI(ap->a_vp);*/
3271 	/*ufhp = (struct ufid *)ap->a_fhp;*/
3272 	fhp->len = sizeof(struct fuse_fid);
3273 	fhp->nid = fvdat->nid;
3274 	if (fvdat->generation <= UINT32_MAX)
3275 		fhp->gen = fvdat->generation;
3276 	else
3277 		return (EXTERROR(EOVERFLOW, "inode generation "
3278 		    "number overflow"));
3279 	return (0);
3280 }
3281