xref: /xnu-12377.41.6/bsd/vfs/vfs_vnops.c (revision bbb1b6f9e71b8cdde6e5cd6f4841f207dee3d828)
1 /*
2  * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29 /*
30  * Copyright (c) 1982, 1986, 1989, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  * (c) UNIX System Laboratories, Inc.
33  * All or some portions of this file are derived from material licensed
34  * to the University of California by American Telephone and Telegraph
35  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
36  * the permission of UNIX System Laboratories, Inc.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_vnops.c	8.14 (Berkeley) 6/15/95
67  *
68  */
69 /*
70  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
71  * support for mandatory and extensible security protections.  This notice
72  * is included in support of clause 2.2 (b) of the Apple Public License,
73  * Version 2.0.
74  */
75 
76 #include <sys/param.h>
77 #include <sys/types.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
80 #include <sys/file_internal.h>
81 #include <sys/stat.h>
82 #include <sys/proc_internal.h>
83 #include <sys/kauth.h>
84 #include <sys/mount_internal.h>
85 #include <sys/namei.h>
86 #include <sys/vnode_internal.h>
87 #include <sys/ioctl.h>
88 #include <sys/fsctl.h>
89 #include <sys/tty.h>
90 #include <sys/ubc.h>
91 #include <sys/conf.h>
92 #include <sys/disk.h>
93 #include <sys/fsevents.h>
94 #include <sys/kdebug.h>
95 #include <sys/xattr.h>
96 #include <sys/ubc_internal.h>
97 #include <sys/uio_internal.h>
98 #include <sys/resourcevar.h>
99 #include <sys/signalvar.h>
100 
101 #include <vm/vm_kern.h>
102 #include <vm/vm_map.h>
103 
104 #include <miscfs/specfs/specdev.h>
105 #include <miscfs/fifofs/fifo.h>
106 
107 #if CONFIG_MACF
108 #include <security/mac_framework.h>
109 #endif
110 
111 #include <IOKit/IOBSD.h>
112 #include <libkern/section_keywords.h>
113 
114 static int vn_closefile(struct fileglob *fp, vfs_context_t ctx);
115 static int vn_ioctl(struct fileproc *fp, u_long com, caddr_t data,
116     vfs_context_t ctx);
117 static int vn_read(struct fileproc *fp, struct uio *uio, int flags,
118     vfs_context_t ctx);
119 static int vn_write(struct fileproc *fp, struct uio *uio, int flags,
120     vfs_context_t ctx);
121 static int vn_select( struct fileproc *fp, int which, void * wql,
122     vfs_context_t ctx);
123 static int vn_kqfilter(struct fileproc *fp, struct knote *kn,
124     struct kevent_qos_s *kev);
125 static void filt_vndetach(struct knote *kn);
126 static int filt_vnode(struct knote *kn, long hint);
127 static int filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev,
128     vnode_t vp, long hint);
129 static int vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx);
130 
131 const struct fileops vnops = {
132 	.fo_type     = DTYPE_VNODE,
133 	.fo_read     = vn_read,
134 	.fo_write    = vn_write,
135 	.fo_ioctl    = vn_ioctl,
136 	.fo_select   = vn_select,
137 	.fo_close    = vn_closefile,
138 	.fo_drain    = fo_no_drain,
139 	.fo_kqfilter = vn_kqfilter,
140 };
141 
142 static int filt_vntouch(struct knote *kn, struct kevent_qos_s *kev);
143 static int filt_vnprocess(struct knote *kn, struct kevent_qos_s*kev);
144 
145 SECURITY_READ_ONLY_EARLY(struct  filterops) vnode_filtops = {
146 	.f_isfd = 1,
147 	.f_attach = NULL,
148 	.f_detach = filt_vndetach,
149 	.f_event = filt_vnode,
150 	.f_touch = filt_vntouch,
151 	.f_process = filt_vnprocess,
152 };
153 
154 /*
155  * Common code for vnode open operations.
156  * Check permissions, and call the VNOP_OPEN or VNOP_CREATE routine.
157  *
158  * XXX the profusion of interfaces here is probably a bad thing.
159  */
160 int
vn_open(struct nameidata * ndp,int fmode,int cmode)161 vn_open(struct nameidata *ndp, int fmode, int cmode)
162 {
163 	return vn_open_modflags(ndp, &fmode, cmode);
164 }
165 
166 int
vn_open_modflags(struct nameidata * ndp,int * fmodep,int cmode)167 vn_open_modflags(struct nameidata *ndp, int *fmodep, int cmode)
168 {
169 	int error;
170 	struct vnode_attr *vap;
171 
172 	vap = kalloc_type(struct vnode_attr, Z_WAITOK);
173 
174 	VATTR_INIT(vap);
175 	VATTR_SET(vap, va_mode, (mode_t)cmode);
176 
177 	error = vn_open_auth(ndp, fmodep, vap, NULLVP);
178 
179 	kfree_type(struct vnode_attr, vap);
180 
181 	return error;
182 }
183 
184 static int
vn_open_auth_finish(vnode_t vp,int fmode,vfs_context_t ctx)185 vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx)
186 {
187 	int error;
188 
189 	if ((error = vnode_ref_ext(vp, fmode, 0)) != 0) {
190 		goto bad;
191 	}
192 
193 	/* Call out to allow 3rd party notification of open.
194 	 * Ignore result of kauth_authorize_fileop call.
195 	 */
196 #if CONFIG_MACF
197 	mac_vnode_notify_open(ctx, vp, fmode);
198 #endif
199 	kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
200 	    (uintptr_t)vp, 0);
201 
202 	return 0;
203 
204 bad:
205 	return error;
206 }
207 
208 /*
209  * May do nameidone() to allow safely adding an FSEvent.  Cue off of ni_dvp to
210  * determine whether that has happened.
211  */
212 static int
vn_open_auth_do_create(struct nameidata * ndp,struct vnode_attr * vap,int fmode,boolean_t * did_create,boolean_t * did_open,vfs_context_t ctx)213 vn_open_auth_do_create(struct nameidata *ndp, struct vnode_attr *vap, int fmode, boolean_t *did_create, boolean_t *did_open, vfs_context_t ctx)
214 {
215 	uint32_t status = 0;
216 	vnode_t dvp = ndp->ni_dvp;
217 	int batched;
218 	int error;
219 	vnode_t vp;
220 
221 	batched = vnode_compound_open_available(ndp->ni_dvp);
222 	*did_open = FALSE;
223 
224 	VATTR_SET(vap, va_type, VREG);
225 	if (fmode & O_EXCL) {
226 		vap->va_vaflags |= VA_EXCLUSIVE;
227 	}
228 
229 #if NAMEDRSRCFORK
230 	if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK) {
231 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
232 			goto out;
233 		}
234 		if ((error = vnode_makenamedstream(dvp, &ndp->ni_vp, XATTR_RESOURCEFORK_NAME, 0, ctx)) != 0) {
235 			goto out;
236 		}
237 		*did_create = TRUE;
238 	} else {
239 #endif
240 	if (!batched) {
241 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
242 			goto out;
243 		}
244 	}
245 
246 	error = vn_create(dvp, &ndp->ni_vp, ndp, vap, VN_CREATE_DOOPEN, fmode, &status, ctx);
247 	if (error != 0) {
248 		if (batched) {
249 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? TRUE : FALSE;
250 		} else {
251 			*did_create = FALSE;
252 		}
253 
254 		if (error == EKEEPLOOKING) {
255 			if (*did_create) {
256 				panic("EKEEPLOOKING, but we did a create?");
257 			}
258 			if (!batched) {
259 				panic("EKEEPLOOKING from filesystem that doesn't support compound vnops?");
260 			}
261 			if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
262 				panic("EKEEPLOOKING, but continue flag not set?");
263 			}
264 
265 			/*
266 			 * Do NOT drop the dvp: we need everything to continue the lookup.
267 			 */
268 			return error;
269 		}
270 	} else {
271 		if (batched) {
272 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? 1 : 0;
273 			*did_open = TRUE;
274 		} else {
275 			*did_create = TRUE;
276 		}
277 	}
278 #if NAMEDRSRCFORK
279 }
280 #endif
281 
282 	vp = ndp->ni_vp;
283 
284 	if (*did_create) {
285 		int     update_flags = 0;
286 
287 		// Make sure the name & parent pointers are hooked up
288 		if (vp->v_name == NULL) {
289 			update_flags |= VNODE_UPDATE_NAME;
290 		}
291 		if (vp->v_parent == NULLVP) {
292 			update_flags |= VNODE_UPDATE_PARENT;
293 		}
294 
295 		if (update_flags) {
296 			vnode_update_identity(vp, dvp, ndp->ni_cnd.cn_nameptr, ndp->ni_cnd.cn_namelen, ndp->ni_cnd.cn_hash, update_flags);
297 		}
298 
299 		vnode_put(dvp);
300 		ndp->ni_dvp = NULLVP;
301 
302 #if CONFIG_FSE
303 		if (need_fsevent(FSE_CREATE_FILE, vp)) {
304 			add_fsevent(FSE_CREATE_FILE, ctx,
305 			    FSE_ARG_VNODE, vp,
306 			    FSE_ARG_DONE);
307 		}
308 #endif
309 	}
310 out:
311 	if (ndp->ni_dvp != NULLVP) {
312 		vnode_put(dvp);
313 		ndp->ni_dvp = NULLVP;
314 	}
315 
316 	return error;
317 }
318 
319 /*
320  * This is the number of times we'll loop in vn_open_auth without explicitly
321  * yielding the CPU when we determine we have to retry.
322  */
323 #define RETRY_NO_YIELD_COUNT    5
324 
325 /*
326  * Open a file with authorization, updating the contents of the structures
327  * pointed to by ndp, fmodep, and vap as necessary to perform the requested
328  * operation.  This function is used for both opens of existing files, and
329  * creation of new files.
330  *
331  * Parameters:	ndp			The nami data pointer describing the
332  *					file
333  *		fmodep			A pointer to an int containg the mode
334  *					information to be used for the open
335  *		vap			A pointer to the vnode attribute
336  *					descriptor to be used for the open
337  *		authvp		If non-null and VA_DP_AUTHENTICATE is set in vap,
338  *					have a supporting filesystem verify that the file
339  *					to be opened is on the same volume as authvp and
340  *					that authvp is on an authenticated volume
341  *
342  * Indirect:	*			Contents of the data structures pointed
343  *					to by the parameters are modified as
344  *					necessary to the requested operation.
345  *
346  * Returns:	0			Success
347  *		!0			errno value
348  *
349  * Notes:	The kauth_filesec_t in 'vap', if any, is in host byte order.
350  *
351  *		The contents of '*ndp' will be modified, based on the other
352  *		arguments to this function, and to return file and directory
353  *		data necessary to satisfy the requested operation.
354  *
355  *		If the file does not exist and we are creating it, then the
356  *		O_TRUNC flag will be cleared in '*fmodep' to indicate to the
357  *		caller that the file was not truncated.
358  *
359  *		If the file exists and the O_EXCL flag was not specified, then
360  *		the O_CREAT flag will be cleared in '*fmodep' to indicate to
361  *		the caller that the existing file was merely opened rather
362  *		than created.
363  *
364  *		The contents of '*vap' will be modified as necessary to
365  *		complete the operation, including setting of supported
366  *		attribute, clearing of fields containing unsupported attributes
367  *		in the request, if the request proceeds without them, etc..
368  *
369  * XXX:		This function is too complicated in actings on its arguments
370  *
371  * XXX:		We should enummerate the possible errno values here, and where
372  *		in the code they originated.
373  */
374 int
vn_open_auth(struct nameidata * ndp,int * fmodep,struct vnode_attr * vap,vnode_t authvp)375 vn_open_auth(struct nameidata *ndp, int *fmodep, struct vnode_attr *vap, vnode_t authvp)
376 {
377 	struct vnode *vp;
378 	struct vnode *dvp;
379 	vfs_context_t ctx = ndp->ni_cnd.cn_context;
380 	int error;
381 	int fmode;
382 	uint32_t origcnflags;
383 	boolean_t did_create;
384 	boolean_t did_open;
385 	boolean_t need_vnop_open;
386 	boolean_t batched;
387 	boolean_t ref_failed;
388 	int nretries = 0, max_retries = 10;
389 
390 again:
391 	vp = NULL;
392 	dvp = NULL;
393 	batched = FALSE;
394 	did_create = FALSE;
395 	need_vnop_open = TRUE;
396 	ref_failed = FALSE;
397 	fmode = *fmodep;
398 	origcnflags = ndp->ni_cnd.cn_flags;
399 
400 	if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)) {
401 		if ((authvp != NULLVP)
402 		    && !ISSET(vap->va_dataprotect_flags, VA_DP_AUTHENTICATE)) {
403 			error = EINVAL;
404 			goto out;
405 		}
406 		// If raw encrypted mode is requested, handle that here
407 		if (ISSET(vap->va_dataprotect_flags, VA_DP_RAWENCRYPTED)) {
408 			fmode |= FENCRYPTED;
409 		}
410 	}
411 
412 	if ((fmode & O_NOFOLLOW_ANY) && (fmode & O_NOFOLLOW)) {
413 		error = EINVAL;
414 		goto out;
415 	}
416 
417 	/*
418 	 * We only call nameidone when exiting the function because we may
419 	 * end up needing the pathname buffer in the component structure.
420 	 */
421 
422 	/*
423 	 * O_CREAT
424 	 */
425 	if (fmode & O_CREAT) {
426 		if ((fmode & O_DIRECTORY)) {
427 			error = EINVAL;
428 			goto out;
429 		}
430 		ndp->ni_cnd.cn_nameiop = CREATE;
431 #if CONFIG_TRIGGERS
432 		ndp->ni_op = OP_LINK;
433 #endif
434 		/* Inherit USEDVP, vnode_open() supported flags only */
435 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
436 		ndp->ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF | AUDITVNPATH1;
437 		/* Inherit NAMEI_ROOTDIR flag only */
438 		ndp->ni_flag &= NAMEI_ROOTDIR;
439 		ndp->ni_flag |= NAMEI_COMPOUNDOPEN;
440 #if NAMEDRSRCFORK
441 		/* open calls are allowed for resource forks. */
442 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
443 #endif
444 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0 && (origcnflags & FOLLOW) != 0 && (fmode & O_SYMLINK) == 0) {
445 			ndp->ni_cnd.cn_flags |= FOLLOW;
446 		}
447 		if (fmode & O_NOFOLLOW_ANY) {
448 			/* will return ELOOP on the first symlink to be hit */
449 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
450 		}
451 		if (fmode & O_RESOLVE_BENEATH) {
452 			/* will return ENOTCAPABLE if path resolution escapes the starting directory */
453 			ndp->ni_flag |= NAMEI_RESOLVE_BENEATH;
454 		}
455 		if (fmode & O_UNIQUE) {
456 			/* will return ENOTCAPABLE if target vnode has multiple links */
457 			ndp->ni_flag |= NAMEI_UNIQUE;
458 		}
459 
460 continue_create_lookup:
461 		if ((error = namei(ndp))) {
462 			goto out;
463 		}
464 
465 		dvp = ndp->ni_dvp;
466 		vp = ndp->ni_vp;
467 
468 		batched = vnode_compound_open_available(dvp);
469 
470 		/* not found, create */
471 		if (vp == NULL) {
472 			/* must have attributes for a new file */
473 			if (vap == NULL) {
474 				vnode_put(dvp);
475 				error = EINVAL;
476 				goto out;
477 			}
478 			/*
479 			 * Attempt a create.   For a system supporting compound VNOPs, we may
480 			 * find an existing file or create one; in either case, we will already
481 			 * have the file open and no VNOP_OPEN() will be needed.
482 			 */
483 			error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
484 
485 			dvp = ndp->ni_dvp;
486 			vp = ndp->ni_vp;
487 
488 			/* Detected a node that the filesystem couldn't handle */
489 			if (error == EKEEPLOOKING) {
490 				if (!batched) {
491 					panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
492 				}
493 				goto continue_create_lookup;
494 			}
495 
496 			if (dvp) {
497 				panic("Shouldn't have a dvp here.");
498 			}
499 
500 			if (error) {
501 				/*
502 				 * Check for a create race.
503 				 */
504 				if ((error == EEXIST) && !(fmode & O_EXCL)) {
505 					if (vp) {
506 						vnode_put(vp);
507 					}
508 					nameidone(ndp);
509 					goto again;
510 				}
511 				goto bad;
512 			}
513 
514 			need_vnop_open = !did_open;
515 		} else {
516 			if (fmode & O_EXCL) {
517 				error = EEXIST;
518 			}
519 
520 			/*
521 			 * We have a vnode.  Use compound open if available
522 			 * or else fall through to "traditional" path.  Note: can't
523 			 * do a compound open for root, because the parent belongs
524 			 * to a different FS.
525 			 */
526 			if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
527 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
528 
529 				if (error == 0) {
530 					vp = ndp->ni_vp;
531 					need_vnop_open = FALSE;
532 				} else if (error == EKEEPLOOKING) {
533 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
534 						panic("EKEEPLOOKING, but continue flag not set?");
535 					}
536 					goto continue_create_lookup;
537 				}
538 			}
539 			vnode_put(dvp);
540 			ndp->ni_dvp = NULLVP;
541 
542 			if (error) {
543 				goto bad;
544 			}
545 
546 			fmode &= ~O_CREAT;
547 
548 			/* Fall through */
549 		}
550 	} else {
551 		/*
552 		 * Not O_CREAT
553 		 */
554 		ndp->ni_cnd.cn_nameiop = LOOKUP;
555 		/* Inherit USEDVP, vnode_open() supported flags only */
556 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
557 		ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1 | WANTPARENT;
558 #if NAMEDRSRCFORK
559 		/* open calls are allowed for resource forks. */
560 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
561 #endif
562 		if (fmode & FENCRYPTED) {
563 			ndp->ni_cnd.cn_flags |= CN_RAW_ENCRYPTED | CN_SKIPNAMECACHE;
564 		}
565 		/* Inherit NAMEI_ROOTDIR flag only */
566 		ndp->ni_flag &= NAMEI_ROOTDIR;
567 		ndp->ni_flag |= NAMEI_COMPOUNDOPEN;
568 
569 		/* preserve NOFOLLOW from vnode_open() */
570 		if (fmode & O_NOFOLLOW || fmode & O_SYMLINK || (origcnflags & FOLLOW) == 0) {
571 			ndp->ni_cnd.cn_flags &= ~FOLLOW;
572 		}
573 		if (fmode & O_NOFOLLOW_ANY) {
574 			/* will return ELOOP on the first symlink to be hit */
575 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
576 		}
577 		if (fmode & O_RESOLVE_BENEATH) {
578 			/* will return ENOTCAPABLE if path resolution escapes the starting directory */
579 			ndp->ni_flag |= NAMEI_RESOLVE_BENEATH;
580 		}
581 		if (fmode & O_UNIQUE) {
582 			/* will return ENOTCAPABLE if target vnode has multiple links */
583 			ndp->ni_flag |= NAMEI_UNIQUE;
584 		}
585 
586 		/* Do a lookup, possibly going directly to filesystem for compound operation */
587 		do {
588 			if ((error = namei(ndp))) {
589 				goto out;
590 			}
591 			vp = ndp->ni_vp;
592 			dvp = ndp->ni_dvp;
593 
594 			/* Check for batched lookup-open */
595 			batched = vnode_compound_open_available(dvp);
596 			if (batched && ((vp == NULLVP) || (vnode_mount(dvp) == vnode_mount(vp)))) {
597 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
598 				vp = ndp->ni_vp;
599 				if (error == 0) {
600 					need_vnop_open = FALSE;
601 				} else if (error == EKEEPLOOKING) {
602 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
603 						panic("EKEEPLOOKING, but continue flag not set?");
604 					}
605 				} else if ((fmode & (FREAD | FWRITE | FEXEC)) == FEXEC) {
606 					/*
607 					 * Some file systems fail in vnop_open call with absense of
608 					 * both FREAD and FWRITE access modes. Retry the vnop_open
609 					 * call again with FREAD access mode added.
610 					 */
611 					error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0,
612 					    fmode | FREAD, NULL, NULL, ctx);
613 					if (error == 0) {
614 						vp = ndp->ni_vp;
615 						need_vnop_open = FALSE;
616 					}
617 				}
618 			}
619 		} while (error == EKEEPLOOKING);
620 
621 		vnode_put(dvp);
622 		ndp->ni_dvp = NULLVP;
623 
624 		if (error) {
625 			goto bad;
626 		}
627 	}
628 
629 	/*
630 	 * By this point, dvp iocount is dropped and dvp pointer is cleared.
631 	 */
632 	if (ndp->ni_dvp != NULLVP) {
633 		panic("Haven't cleaned up adequately in vn_open_auth()");
634 	}
635 
636 	/* Verify that the vnode returned from namei() has the expected fileid and fsid */
637 	if (VATTR_IS_ACTIVE(vap, va_flags) && ISSET(vap->va_flags, VA_VAFILEID)) {
638 		vnode_t tmp_vp;
639 
640 		if (!VATTR_IS_ACTIVE(vap, va_fsid64) || !VATTR_IS_ACTIVE(vap, va_fileid)) {
641 			error = EINVAL;
642 			goto bad;
643 		}
644 
645 		error = vnode_getfromid(vap->va_fsid64.val[0], vap->va_fileid, ctx, FSOPT_ISREALFSID, &tmp_vp);
646 		if (error) {
647 			goto bad;
648 		}
649 
650 		if (tmp_vp != vp) {
651 			vnode_put(tmp_vp);
652 			error = ERECYCLE;
653 			goto bad;
654 		}
655 		vnode_put(tmp_vp);
656 	}
657 
658 	/*
659 	 * Expect to use this code for filesystems without compound VNOPs, for the root
660 	 * of a filesystem, which can't be "looked up" in the sense of VNOP_LOOKUP(),
661 	 * and for shadow files, which do not live on the same filesystems as their "parents."
662 	 */
663 	if (need_vnop_open) {
664 		if (batched && !vnode_isvroot(vp) && !vnode_isnamedstream(vp)) {
665 			panic("Why am I trying to use VNOP_OPEN() on anything other than the root or a named stream?");
666 		}
667 
668 		if (!did_create) {
669 			error = vn_authorize_open_existing(vp, &ndp->ni_cnd, fmode, ctx, NULL);
670 			if (error) {
671 				goto bad;
672 			}
673 		}
674 
675 		if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)) {
676 			if (ISSET(vap->va_dataprotect_flags, VA_DP_RAWUNENCRYPTED)) {
677 				/* Don't allow unencrypted io request from user space unless entitled */
678 				boolean_t entitled = FALSE;
679 #if !SECURE_KERNEL
680 				entitled = IOCurrentTaskHasEntitlement("com.apple.private.security.file-unencrypt-access");
681 #endif /* SECURE_KERNEL */
682 				if (!entitled) {
683 					error = EPERM;
684 					goto bad;
685 				}
686 				fmode |= FUNENCRYPTED;
687 			}
688 
689 			if (ISSET(vap->va_dataprotect_flags, VA_DP_AUTHENTICATE)) {
690 				fsioc_auth_fs_t afs = { .authvp = authvp };
691 
692 				error = VNOP_IOCTL(vp, FSIOC_AUTH_FS, (caddr_t)&afs, 0, ctx);
693 				if (error) {
694 					goto bad;
695 				}
696 			}
697 		}
698 
699 		error = VNOP_OPEN(vp, fmode, ctx);
700 		if (error) {
701 			/*
702 			 * Some file systems fail in vnop_open call with absense of both
703 			 * FREAD and FWRITE access modes. Retry the vnop_open call again
704 			 * with FREAD access mode added.
705 			 */
706 			if ((fmode & (FREAD | FWRITE | FEXEC)) == FEXEC) {
707 				error = VNOP_OPEN(vp, fmode | FREAD, ctx);
708 			}
709 			if (error) {
710 				goto bad;
711 			}
712 		}
713 		need_vnop_open = FALSE;
714 	}
715 
716 	// if the vnode is tagged VOPENEVT and the current process
717 	// has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
718 	// flag to the open mode so that this open won't count against
719 	// the vnode when carbon delete() does a vnode_isinuse() to see
720 	// if a file is currently in use.  this allows spotlight
721 	// importers to not interfere with carbon apps that depend on
722 	// the no-delete-if-busy semantics of carbon delete().
723 	//
724 	if (!did_create && (vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
725 		fmode |= O_EVTONLY;
726 	}
727 
728 	/*
729 	 * Grab reference, etc.
730 	 */
731 	error = vn_open_auth_finish(vp, fmode, ctx);
732 	if (error) {
733 		ref_failed = TRUE;
734 		goto bad;
735 	}
736 
737 	/* Compound VNOP open is responsible for doing the truncate */
738 	if (batched || did_create) {
739 		fmode &= ~O_TRUNC;
740 	}
741 
742 	*fmodep = fmode;
743 	nameidone(ndp);
744 	return 0;
745 
746 bad:
747 	/* Opened either explicitly or by a batched create */
748 	if (!need_vnop_open) {
749 		VNOP_CLOSE(vp, fmode, ctx);
750 	}
751 
752 	ndp->ni_vp = NULL;
753 	if (vp) {
754 #if NAMEDRSRCFORK
755 		/* Aggressively recycle shadow files if we error'd out during open() */
756 		if ((vnode_isnamedstream(vp)) &&
757 		    (vp->v_parent != NULLVP) &&
758 		    (vnode_isshadow(vp))) {
759 			vnode_recycle(vp);
760 		}
761 #endif
762 		vnode_put(vp);
763 		/*
764 		 * Check for a race against unlink.  We had a vnode
765 		 * but according to vnode_authorize or VNOP_OPEN it
766 		 * no longer exists.
767 		 *
768 		 * EREDRIVEOPEN: means that we were hit by the tty allocation race or NFSv4 open/create race.
769 		 */
770 		if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN) || ref_failed) {
771 			/*
772 			 * We'll retry here but it may be possible that we get
773 			 * into a retry "spin" inside the kernel and not allow
774 			 * threads, which need to run in order for the retry
775 			 * loop to end, to run. An example is an open of a
776 			 * terminal which is getting revoked and we spin here
777 			 * without yielding becasue namei and VNOP_OPEN are
778 			 * successful but vnode_ref fails. The revoke needs
779 			 * threads with an iocount to run but if spin here we
780 			 * may possibly be blcoking other threads from running.
781 			 *
782 			 * We start yielding the CPU after some number of
783 			 * retries for increasing durations.
784 			 */
785 			nretries += 1;
786 			if (nretries > max_retries) {
787 				printf("%s: reached max_retries error %d need_vnop_open %d "
788 				    "fmode 0x%x ref_failed %d\n",
789 				    __func__, error, need_vnop_open, *fmodep, ref_failed);
790 				goto out;
791 			}
792 			if (nretries > RETRY_NO_YIELD_COUNT) {
793 				struct timespec to;
794 
795 				to.tv_sec = 0;
796 				to.tv_nsec = nretries * 10 * NSEC_PER_MSEC;
797 				msleep(&nretries, (lck_mtx_t *)0, PVFS, "vn_open_auth_retry", &to);
798 			}
799 			nameidone(ndp);
800 			goto again;
801 		}
802 	}
803 
804 out:
805 	nameidone(ndp);
806 	return error;
807 }
808 
809 #if vn_access_DEPRECATED
810 /*
811  * Authorize an action against a vnode.  This has been the canonical way to
812  * ensure that the credential/process/etc. referenced by a vfs_context
813  * is granted the rights called out in 'mode' against the vnode 'vp'.
814  *
815  * Unfortunately, the use of VREAD/VWRITE/VEXEC makes it very difficult
816  * to add support for more rights.  As such, this interface will be deprecated
817  * and callers will use vnode_authorize instead.
818  */
819 int
vn_access(vnode_t vp,int mode,vfs_context_t context)820 vn_access(vnode_t vp, int mode, vfs_context_t context)
821 {
822 	kauth_action_t  action;
823 
824 	action = 0;
825 	if (mode & VREAD) {
826 		action |= KAUTH_VNODE_READ_DATA;
827 	}
828 	if (mode & VWRITE) {
829 		action |= KAUTH_VNODE_WRITE_DATA;
830 	}
831 	if (mode & VEXEC) {
832 		action |= KAUTH_VNODE_EXECUTE;
833 	}
834 
835 	return vnode_authorize(vp, NULL, action, context);
836 }
837 #endif  /* vn_access_DEPRECATED */
838 
839 /*
840  * Vnode close call
841  */
842 int
vn_close(struct vnode * vp,int flags,vfs_context_t ctx)843 vn_close(struct vnode *vp, int flags, vfs_context_t ctx)
844 {
845 	int error;
846 	int flusherror = 0;
847 
848 #if NAMEDRSRCFORK
849 	/* Sync data from resource fork shadow file if needed. */
850 	if ((vp->v_flag & VISNAMEDSTREAM) &&
851 	    (vp->v_parent != NULLVP) &&
852 	    vnode_isshadow(vp)) {
853 		if (flags & FWASWRITTEN) {
854 			flusherror = vnode_flushnamedstream(vp->v_parent, vp, ctx);
855 		}
856 	}
857 #endif
858 	/*
859 	 * If vnode @vp belongs to a chardev or a blkdev then it is handled
860 	 * specially.  We first drop its user reference count @vp->v_usecount
861 	 * before calling VNOP_CLOSE().  This was done historically to ensure
862 	 * that the last close of a special device vnode performed some
863 	 * conditional cleanups.  Now we still need to drop this reference here
864 	 * to ensure that devfsspec_close() can check if the vnode is still in
865 	 * use.
866 	 */
867 	if (vnode_isspec(vp)) {
868 		(void)vnode_rele_ext(vp, flags, 0);
869 	}
870 
871 	/*
872 	 * On HFS, we flush when the last writer closes.  We do this
873 	 * because resource fork vnodes hold a reference on data fork
874 	 * vnodes and that will prevent them from getting VNOP_INACTIVE
875 	 * which will delay when we flush cached data.  In future, we
876 	 * might find it beneficial to do this for all file systems.
877 	 * Note that it's OK to access v_writecount without the lock
878 	 * in this context.
879 	 */
880 	if (vp->v_tag == VT_HFS && (flags & FWRITE) && vp->v_writecount == 1) {
881 		VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
882 	}
883 
884 	error = VNOP_CLOSE(vp, flags, ctx);
885 
886 #if CONFIG_FSE
887 	if (flags & FWASWRITTEN) {
888 		if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
889 			add_fsevent(FSE_CONTENT_MODIFIED, ctx,
890 			    FSE_ARG_VNODE, vp,
891 			    FSE_ARG_DONE);
892 		}
893 	}
894 #endif
895 
896 	if (!vnode_isspec(vp)) {
897 		(void)vnode_rele_ext(vp, flags, 0);
898 	}
899 
900 	if (flusherror) {
901 		error = flusherror;
902 	}
903 	return error;
904 }
905 
906 static int
vn_read_swapfile(struct vnode * vp,uio_t uio)907 vn_read_swapfile(
908 	struct vnode    *vp,
909 	uio_t           uio)
910 {
911 	int     error;
912 	off_t   swap_count, this_count;
913 	off_t   file_end, read_end;
914 	off_t   prev_resid;
915 	char    *my_swap_page;
916 
917 	/*
918 	 * Reading from a swap file will get you zeroes.
919 	 */
920 
921 	my_swap_page = NULL;
922 	error = 0;
923 	swap_count = uio_resid(uio);
924 
925 	file_end = ubc_getsize(vp);
926 	read_end = uio->uio_offset + uio_resid(uio);
927 	if (uio->uio_offset >= file_end) {
928 		/* uio starts after end of file: nothing to read */
929 		swap_count = 0;
930 	} else if (read_end > file_end) {
931 		/* uio extends beyond end of file: stop before that */
932 		swap_count -= (read_end - file_end);
933 	}
934 
935 	while (swap_count > 0) {
936 		if (my_swap_page == NULL) {
937 			my_swap_page = kalloc_data(PAGE_SIZE, Z_WAITOK | Z_ZERO);
938 			/* add an end-of-line to keep line counters happy */
939 			my_swap_page[PAGE_SIZE - 1] = '\n';
940 		}
941 		this_count = swap_count;
942 		if (this_count > PAGE_SIZE) {
943 			this_count = PAGE_SIZE;
944 		}
945 
946 		prev_resid = uio_resid(uio);
947 		error = uiomove((caddr_t) my_swap_page,
948 		    (int)this_count,
949 		    uio);
950 		if (error) {
951 			break;
952 		}
953 		swap_count -= (prev_resid - uio_resid(uio));
954 	}
955 	kfree_data(my_swap_page, PAGE_SIZE);
956 
957 	return error;
958 }
959 /*
960  * Package up an I/O request on a vnode into a uio and do it.
961  */
962 int
vn_rdwr(enum uio_rw rw,struct vnode * vp,caddr_t base,int len,off_t offset,enum uio_seg segflg,int ioflg,kauth_cred_t cred,int * aresid,proc_t p)963 vn_rdwr(
964 	enum uio_rw rw,
965 	struct vnode *vp,
966 	caddr_t base,
967 	int len,
968 	off_t offset,
969 	enum uio_seg segflg,
970 	int ioflg,
971 	kauth_cred_t cred,
972 	int *aresid,
973 	proc_t p)
974 {
975 	int64_t resid;
976 	int result;
977 
978 	if (len < 0) {
979 		return EINVAL;
980 	}
981 
982 	result = vn_rdwr_64(rw,
983 	    vp,
984 	    (uint64_t)(uintptr_t)base,
985 	    (int64_t)len,
986 	    offset,
987 	    segflg,
988 	    ioflg,
989 	    cred,
990 	    &resid,
991 	    p);
992 
993 	/* "resid" should be bounded above by "len," which is an int */
994 	if (aresid != NULL) {
995 		*aresid = (int)resid;
996 	}
997 
998 	return result;
999 }
1000 
1001 
1002 int
vn_rdwr_64(enum uio_rw rw,struct vnode * vp,uint64_t base,int64_t len,off_t offset,enum uio_seg segflg,int ioflg,kauth_cred_t cred,int64_t * aresid,proc_t p)1003 vn_rdwr_64(
1004 	enum uio_rw rw,
1005 	struct vnode *vp,
1006 	uint64_t base,
1007 	int64_t len,
1008 	off_t offset,
1009 	enum uio_seg segflg,
1010 	int ioflg,
1011 	kauth_cred_t cred,
1012 	int64_t *aresid,
1013 	proc_t p)
1014 {
1015 	uio_t auio;
1016 	int spacetype;
1017 	struct vfs_context context;
1018 	int error = 0;
1019 	UIO_STACKBUF(uio_buf, 1);
1020 
1021 	context.vc_thread = current_thread();
1022 	context.vc_ucred = cred;
1023 
1024 	if (UIO_SEG_IS_USER_SPACE(segflg)) {
1025 		spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
1026 	} else {
1027 		spacetype = UIO_SYSSPACE;
1028 	}
1029 
1030 	if (len < 0) {
1031 		return EINVAL;
1032 	}
1033 
1034 	auio = uio_createwithbuffer(1, offset, spacetype, rw,
1035 	    &uio_buf[0], sizeof(uio_buf));
1036 	uio_addiov(auio, CAST_USER_ADDR_T(base), (user_size_t)len);
1037 
1038 #if CONFIG_MACF
1039 	/* XXXMAC
1040 	 *      IO_NOAUTH should be re-examined.
1041 	 *	Likely that mediation should be performed in caller.
1042 	 */
1043 	if ((ioflg & IO_NOAUTH) == 0) {
1044 		/* passed cred is fp->f_cred */
1045 		if (rw == UIO_READ) {
1046 			error = mac_vnode_check_read(&context, cred, vp);
1047 		} else {
1048 			error = mac_vnode_check_write(&context, cred, vp);
1049 		}
1050 	}
1051 #endif
1052 
1053 	if (error == 0) {
1054 		if (rw == UIO_READ) {
1055 			if (vnode_isswap(vp) && ((ioflg & IO_SWAP_DISPATCH) == 0)) {
1056 				error = vn_read_swapfile(vp, auio);
1057 			} else {
1058 				error = VNOP_READ(vp, auio, ioflg, &context);
1059 			}
1060 		} else {
1061 			error = VNOP_WRITE(vp, auio, ioflg, &context);
1062 		}
1063 	}
1064 
1065 	if (aresid) {
1066 		*aresid = uio_resid(auio);
1067 		assert(*aresid <= len);
1068 	} else if (uio_resid(auio) && error == 0) {
1069 		error = EIO;
1070 	}
1071 	return error;
1072 }
1073 
1074 void
vn_offset_lock(struct fileglob * fg)1075 vn_offset_lock(struct fileglob *fg)
1076 {
1077 	lck_mtx_lock_spin(&fg->fg_lock);
1078 	while (fg->fg_lflags & FG_OFF_LOCKED) {
1079 		fg->fg_lflags |= FG_OFF_LOCKWANT;
1080 		msleep(&fg->fg_lflags, &fg->fg_lock, PVFS | PSPIN,
1081 		    "fg_offset_lock_wait", 0);
1082 	}
1083 	fg->fg_lflags |= FG_OFF_LOCKED;
1084 	lck_mtx_unlock(&fg->fg_lock);
1085 }
1086 
1087 void
vn_offset_unlock(struct fileglob * fg)1088 vn_offset_unlock(struct fileglob *fg)
1089 {
1090 	int lock_wanted = 0;
1091 
1092 	lck_mtx_lock_spin(&fg->fg_lock);
1093 	if (fg->fg_lflags & FG_OFF_LOCKWANT) {
1094 		lock_wanted = 1;
1095 	}
1096 	fg->fg_lflags &= ~(FG_OFF_LOCKED | FG_OFF_LOCKWANT);
1097 	lck_mtx_unlock(&fg->fg_lock);
1098 	if (lock_wanted) {
1099 		wakeup(&fg->fg_lflags);
1100 	}
1101 }
1102 
1103 static int
vn_read_common(vnode_t vp,struct uio * uio,int fflag,vfs_context_t ctx)1104 vn_read_common(vnode_t vp, struct uio *uio, int fflag, vfs_context_t ctx)
1105 {
1106 	int error;
1107 	int ioflag;
1108 	off_t read_offset;
1109 	user_ssize_t  read_len;
1110 	user_ssize_t  adjusted_read_len;
1111 	user_ssize_t  clippedsize;
1112 
1113 	/* Caller has already validated read_len. */
1114 	read_len = uio_resid(uio);
1115 	assert(read_len >= 0 && read_len <= INT_MAX);
1116 
1117 	adjusted_read_len = read_len;
1118 	clippedsize = 0;
1119 
1120 #if CONFIG_MACF
1121 	error = mac_vnode_check_read(ctx, vfs_context_ucred(ctx), vp);
1122 	if (error) {
1123 		return error;
1124 	}
1125 #endif
1126 
1127 	/* This signals to VNOP handlers that this read came from a file table read */
1128 	ioflag = IO_SYSCALL_DISPATCH;
1129 
1130 	if (fflag & FNONBLOCK) {
1131 		ioflag |= IO_NDELAY;
1132 	}
1133 	if ((fflag & FNOCACHE) || vnode_isnocache(vp)) {
1134 		ioflag |= IO_NOCACHE;
1135 	}
1136 	if (fflag & FENCRYPTED) {
1137 		ioflag |= IO_ENCRYPTED;
1138 	}
1139 	if (fflag & FUNENCRYPTED) {
1140 		ioflag |= IO_SKIP_ENCRYPTION;
1141 	}
1142 	if (fflag & O_EVTONLY) {
1143 		ioflag |= IO_EVTONLY;
1144 	}
1145 	if (fflag & FNORDAHEAD) {
1146 		ioflag |= IO_RAOFF;
1147 	}
1148 
1149 	read_offset = uio_offset(uio);
1150 	/* POSIX allows negative offsets for character devices. */
1151 	if ((read_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1152 		error = EINVAL;
1153 		goto error_out;
1154 	}
1155 
1156 	if (read_offset == INT64_MAX) {
1157 		/* can't read any more */
1158 		error = 0;
1159 		goto error_out;
1160 	}
1161 
1162 	/*
1163 	 * If offset + len will cause overflow, reduce the len to a value
1164 	 * (adjusted_read_len) where it won't
1165 	 */
1166 	if ((read_offset >= 0) && (INT64_MAX - read_offset) < read_len) {
1167 		/*
1168 		 * 0                                          read_offset  INT64_MAX
1169 		 * |-----------------------------------------------|----------|~~~
1170 		 *                                                 <--read_len-->
1171 		 *                                                 <-adjusted->
1172 		 */
1173 		adjusted_read_len = (user_ssize_t)(INT64_MAX - read_offset);
1174 	}
1175 
1176 	if (adjusted_read_len < read_len) {
1177 		uio_setresid(uio, adjusted_read_len);
1178 		clippedsize = read_len - adjusted_read_len;
1179 	}
1180 
1181 	if (vnode_isswap(vp) && !(IO_SKIP_ENCRYPTION & ioflag)) {
1182 		/* special case for swap files */
1183 		error = vn_read_swapfile(vp, uio);
1184 	} else {
1185 		error = VNOP_READ(vp, uio, ioflag, ctx);
1186 	}
1187 
1188 	if (clippedsize) {
1189 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1190 	}
1191 
1192 error_out:
1193 	return error;
1194 }
1195 
1196 /*
1197  * File table vnode read routine.
1198  */
1199 static int
vn_read(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1200 vn_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1201 {
1202 	vnode_t vp;
1203 	user_ssize_t read_len;
1204 	int error;
1205 	bool offset_locked;
1206 
1207 	read_len = uio_resid(uio);
1208 	if (read_len < 0 || read_len > INT_MAX) {
1209 		return EINVAL;
1210 	}
1211 
1212 	if ((flags & FOF_OFFSET) == 0) {
1213 		vn_offset_lock(fp->fp_glob);
1214 		offset_locked = true;
1215 	} else {
1216 		offset_locked = false;
1217 	}
1218 	vp = (struct vnode *)fp_get_data(fp);
1219 	if ((error = vnode_getwithref(vp))) {
1220 		if (offset_locked) {
1221 			vn_offset_unlock(fp->fp_glob);
1222 		}
1223 		return error;
1224 	}
1225 
1226 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1227 		vn_offset_unlock(fp->fp_glob);
1228 		offset_locked = false;
1229 	}
1230 
1231 	if ((flags & FOF_OFFSET) == 0) {
1232 		uio_setoffset(uio, fp->fp_glob->fg_offset);
1233 	}
1234 
1235 	error = vn_read_common(vp, uio, fp->fp_glob->fg_flag, ctx);
1236 
1237 	if ((flags & FOF_OFFSET) == 0) {
1238 		fp->fp_glob->fg_offset += read_len - uio_resid(uio);
1239 	}
1240 
1241 	if (offset_locked) {
1242 		vn_offset_unlock(fp->fp_glob);
1243 		offset_locked = false;
1244 	}
1245 
1246 	vnode_put(vp);
1247 	return error;
1248 }
1249 
1250 /*
1251  * File table vnode write routine.
1252  */
1253 static int
vn_write(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1254 vn_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1255 {
1256 	struct vnode *vp;
1257 	int error, ioflag;
1258 	off_t write_offset;
1259 	off_t write_end_offset;
1260 	user_ssize_t write_len;
1261 	user_ssize_t adjusted_write_len;
1262 	user_ssize_t clippedsize;
1263 	bool offset_locked;
1264 	proc_t p = vfs_context_proc(ctx);
1265 	rlim_t rlim_cur_fsize = p ? proc_limitgetcur(p, RLIMIT_FSIZE) : 0;
1266 
1267 	write_len = uio_resid(uio);
1268 	if (write_len < 0 || write_len > INT_MAX) {
1269 		return EINVAL;
1270 	}
1271 	adjusted_write_len = write_len;
1272 	clippedsize = 0;
1273 
1274 	if ((flags & FOF_OFFSET) == 0) {
1275 		vn_offset_lock(fp->fp_glob);
1276 		offset_locked = true;
1277 	} else {
1278 		offset_locked = false;
1279 	}
1280 
1281 	vp = (struct vnode *)fp_get_data(fp);
1282 	if ((error = vnode_getwithref(vp))) {
1283 		if (offset_locked) {
1284 			vn_offset_unlock(fp->fp_glob);
1285 		}
1286 		return error;
1287 	}
1288 
1289 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1290 		vn_offset_unlock(fp->fp_glob);
1291 		offset_locked = false;
1292 	}
1293 
1294 #if CONFIG_MACF
1295 	error = mac_vnode_check_write(ctx, vfs_context_ucred(ctx), vp);
1296 	if (error) {
1297 		(void)vnode_put(vp);
1298 		if (offset_locked) {
1299 			vn_offset_unlock(fp->fp_glob);
1300 		}
1301 		return error;
1302 	}
1303 #endif
1304 
1305 	/*
1306 	 * IO_SYSCALL_DISPATCH signals to VNOP handlers that this write came from
1307 	 * a file table write
1308 	 */
1309 	ioflag = (IO_UNIT | IO_SYSCALL_DISPATCH);
1310 
1311 	if (vp->v_type == VREG && (fp->fp_glob->fg_flag & O_APPEND)) {
1312 		ioflag |= IO_APPEND;
1313 	}
1314 	if (fp->fp_glob->fg_flag & FNONBLOCK) {
1315 		ioflag |= IO_NDELAY;
1316 	}
1317 	assert(!(fp->fp_glob->fg_flag & O_NOCTTY) || (fp->fp_glob->fg_flag & FNOCACHE));
1318 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1319 		ioflag |= IO_NOCACHE;
1320 		/*
1321 		 * O_NOCTTY is repurposed to indicate the preference for relaxed
1322 		 * alignment and size restrictions. O_NOCTTY has no meaning beyond
1323 		 * the open of the fd and is used only for ttys and the use
1324 		 * is mutually exclusive with nocache io for regular files.
1325 		 */
1326 		if (fp->fp_glob->fg_flag & O_NOCTTY || vfs_context_allow_fs_blksize_nocache_write(ctx)) {
1327 			ioflag |= IO_NOCACHE_SWRITE;
1328 		}
1329 	}
1330 	if (fp->fp_glob->fg_flag & FNODIRECT) {
1331 		ioflag |= IO_NODIRECT;
1332 	}
1333 	if (fp->fp_glob->fg_flag & FSINGLE_WRITER) {
1334 		ioflag |= IO_SINGLE_WRITER;
1335 	}
1336 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1337 		ioflag |= IO_EVTONLY;
1338 	}
1339 
1340 	/*
1341 	 * Treat synchronous mounts and O_FSYNC on the fd as equivalent.
1342 	 *
1343 	 * XXX We treat O_DSYNC as O_FSYNC for now, since we can not delay
1344 	 * XXX the non-essential metadata without some additional VFS work;
1345 	 * XXX the intent at this point is to plumb the interface for it.
1346 	 */
1347 	if ((fp->fp_glob->fg_flag & (O_FSYNC | O_DSYNC)) ||
1348 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) {
1349 		ioflag |= IO_SYNC;
1350 	}
1351 
1352 	if ((flags & FOF_OFFSET) == 0) {
1353 		write_offset = fp->fp_glob->fg_offset;
1354 		uio_setoffset(uio, write_offset);
1355 	} else {
1356 		/* for pwrite, append should  be ignored */
1357 		ioflag &= ~IO_APPEND;
1358 		write_offset = uio_offset(uio);
1359 		/* POSIX allows negative offsets for character devices. */
1360 		if ((write_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1361 			error = EINVAL;
1362 			goto error_out;
1363 		}
1364 	}
1365 
1366 	if (write_offset == INT64_MAX) {
1367 		/* writes are not possible */
1368 		error = EFBIG;
1369 		goto error_out;
1370 	}
1371 
1372 	/*
1373 	 * write_len is the original write length that was requested.
1374 	 * We may however need to reduce that becasue of two reasons
1375 	 *
1376 	 * 1) If write_offset + write_len will exceed OFF_T_MAX (i.e. INT64_MAX)
1377 	 *    and/or
1378 	 * 2) If write_offset + write_len will exceed the administrative
1379 	 *    limit for the maximum file size.
1380 	 *
1381 	 * In both cases the write will be denied if we can't write even a single
1382 	 * byte otherwise it will be "clipped" (i.e. a short write).
1383 	 */
1384 
1385 	/*
1386 	 * If offset + len will cause overflow, reduce the len
1387 	 * to a value (adjusted_write_len) where it won't
1388 	 */
1389 	if ((write_offset >= 0) && (INT64_MAX - write_offset) < write_len) {
1390 		/*
1391 		 * 0                                          write_offset  INT64_MAX
1392 		 * |-----------------------------------------------|----------|~~~
1393 		 *                                                 <--write_len-->
1394 		 *                                                 <-adjusted->
1395 		 */
1396 		adjusted_write_len = (user_ssize_t)(INT64_MAX - write_offset);
1397 	}
1398 
1399 	/* write_end_offset will always be [0, INT64_MAX] */
1400 	write_end_offset = write_offset + adjusted_write_len;
1401 
1402 	if (p && (vp->v_type == VREG) &&
1403 	    (rlim_cur_fsize != RLIM_INFINITY) &&
1404 	    (rlim_cur_fsize <= INT64_MAX) &&
1405 	    (write_end_offset > (off_t)rlim_cur_fsize)) {
1406 		/*
1407 		 * If the requested residual would cause us to go past the
1408 		 * administrative limit, then we need to adjust the residual
1409 		 * down to cause fewer bytes than requested to be written.  If
1410 		 * we can't do that (e.g. the residual is already 1 byte),
1411 		 * then we fail the write with EFBIG.
1412 		 */
1413 		if (write_offset >= (off_t)rlim_cur_fsize) {
1414 			/*
1415 			 * 0                  rlim_fsize  write_offset  write_end  INT64_MAX
1416 			 * |------------------------|----------|-------------|--------|
1417 			 *                                     <--write_len-->
1418 			 *
1419 			 *                write not permitted
1420 			 */
1421 			psignal(p, SIGXFSZ);
1422 			error = EFBIG;
1423 			goto error_out;
1424 		}
1425 
1426 		/*
1427 		 * 0                  write_offset rlim_fsize  write_end  INT64_MAX
1428 		 * |------------------------|-----------|---------|------------|
1429 		 *                          <------write_len------>
1430 		 *                          <-adjusted-->
1431 		 */
1432 		adjusted_write_len =  (user_ssize_t)((off_t)rlim_cur_fsize - write_offset);
1433 		assert((adjusted_write_len > 0) && (adjusted_write_len < write_len));
1434 	}
1435 
1436 	if (adjusted_write_len < write_len) {
1437 		uio_setresid(uio, adjusted_write_len);
1438 		clippedsize = write_len - adjusted_write_len;
1439 	}
1440 
1441 	error = VNOP_WRITE(vp, uio, ioflag, ctx);
1442 
1443 	/*
1444 	 * If we had to reduce the size of write requested either because
1445 	 * of rlimit or because it would have exceeded
1446 	 * maximum file size, we have to add that back to the residual so
1447 	 * it correctly reflects what we did in this function.
1448 	 */
1449 	if (clippedsize) {
1450 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1451 	}
1452 
1453 	if ((flags & FOF_OFFSET) == 0) {
1454 		if (ioflag & IO_APPEND) {
1455 			fp->fp_glob->fg_offset = uio_offset(uio);
1456 		} else {
1457 			fp->fp_glob->fg_offset += (write_len - uio_resid(uio));
1458 		}
1459 		if (offset_locked) {
1460 			vn_offset_unlock(fp->fp_glob);
1461 			offset_locked = false;
1462 		}
1463 	}
1464 
1465 	/*
1466 	 * Set the credentials on successful writes
1467 	 */
1468 	if ((error == 0) && (vp->v_tag == VT_NFS) && (UBCINFOEXISTS(vp))) {
1469 		ubc_setcred(vp, vfs_context_ucred(ctx));
1470 	}
1471 
1472 #if CONFIG_FILE_LEASES
1473 	/*
1474 	 * On success, break the parent dir lease as the file's attributes (size
1475 	 * and/or mtime) have changed. Best attempt to break lease, just drop the
1476 	 * the error upon failure as there is no point to return error when the
1477 	 * write has completed successfully.
1478 	 */
1479 	if (__probable(error == 0)) {
1480 		vnode_breakdirlease(vp, true, O_WRONLY);
1481 	}
1482 #endif /* CONFIG_FILE_LEASES */
1483 
1484 	(void)vnode_put(vp);
1485 	return error;
1486 
1487 error_out:
1488 	if (offset_locked) {
1489 		vn_offset_unlock(fp->fp_glob);
1490 	}
1491 	(void)vnode_put(vp);
1492 	return error;
1493 }
1494 
1495 /*
1496  * File table vnode stat routine.
1497  *
1498  * Returns:	0			Success
1499  *		EBADF
1500  *		ENOMEM
1501  *	vnode_getattr:???
1502  */
1503 int
vn_stat_noauth(struct vnode * vp,void * sbptr,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx,struct ucred * file_cred)1504 vn_stat_noauth(struct vnode *vp, void *sbptr, kauth_filesec_t *xsec, int isstat64,
1505     int needsrealdev, vfs_context_t ctx, struct ucred *file_cred)
1506 {
1507 	struct vnode_attr va;
1508 	int error;
1509 	u_short mode;
1510 	kauth_filesec_t fsec;
1511 	struct stat *sb = (struct stat *)0;     /* warning avoidance ; protected by isstat64 */
1512 	struct stat64 * sb64 = (struct stat64 *)0;  /* warning avoidance ; protected by isstat64 */
1513 
1514 	if (isstat64 != 0) {
1515 		sb64 = (struct stat64 *)sbptr;
1516 	} else {
1517 		sb = (struct stat *)sbptr;
1518 	}
1519 	memset(&va, 0, sizeof(va));
1520 	VATTR_INIT(&va);
1521 	VATTR_WANTED(&va, va_fsid);
1522 	VATTR_WANTED(&va, va_fileid);
1523 	VATTR_WANTED(&va, va_mode);
1524 	VATTR_WANTED(&va, va_type);
1525 	VATTR_WANTED(&va, va_nlink);
1526 	VATTR_WANTED(&va, va_uid);
1527 	VATTR_WANTED(&va, va_gid);
1528 	VATTR_WANTED(&va, va_rdev);
1529 	VATTR_WANTED(&va, va_data_size);
1530 	VATTR_WANTED(&va, va_access_time);
1531 	VATTR_WANTED(&va, va_modify_time);
1532 	VATTR_WANTED(&va, va_change_time);
1533 	VATTR_WANTED(&va, va_create_time);
1534 	VATTR_WANTED(&va, va_flags);
1535 	VATTR_WANTED(&va, va_gen);
1536 	VATTR_WANTED(&va, va_iosize);
1537 	/* lower layers will synthesise va_total_alloc from va_data_size if required */
1538 	VATTR_WANTED(&va, va_total_alloc);
1539 	if (xsec != NULL && !vnode_isnamedstream(vp)) {
1540 		VATTR_WANTED(&va, va_uuuid);
1541 		VATTR_WANTED(&va, va_guuid);
1542 		VATTR_WANTED(&va, va_acl);
1543 	}
1544 	if (needsrealdev) {
1545 		va.va_vaflags = VA_REALFSID;
1546 	}
1547 	error = vnode_getattr(vp, &va, ctx);
1548 	if (error) {
1549 		goto out;
1550 	}
1551 #if CONFIG_MACF
1552 	/*
1553 	 * Give MAC polices a chance to reject or filter the attributes
1554 	 * returned by the filesystem.  Note that MAC policies are consulted
1555 	 * *after* calling the filesystem because filesystems can return more
1556 	 * attributes than were requested so policies wouldn't be authoritative
1557 	 * is consulted beforehand.  This also gives policies an opportunity
1558 	 * to change the values of attributes retrieved.
1559 	 */
1560 	error = mac_vnode_check_getattr(ctx, file_cred, vp, &va);
1561 	if (error) {
1562 		goto out;
1563 	}
1564 #endif
1565 	/*
1566 	 * Copy from vattr table
1567 	 */
1568 	if (isstat64 != 0) {
1569 		sb64->st_dev = va.va_fsid;
1570 		sb64->st_ino = (ino64_t)va.va_fileid;
1571 	} else {
1572 		sb->st_dev = va.va_fsid;
1573 		sb->st_ino = (ino_t)va.va_fileid;
1574 	}
1575 	mode = va.va_mode;
1576 	switch (vp->v_type) {
1577 	case VREG:
1578 		mode |= S_IFREG;
1579 		break;
1580 	case VDIR:
1581 		mode |= S_IFDIR;
1582 		break;
1583 	case VBLK:
1584 		mode |= S_IFBLK;
1585 		break;
1586 	case VCHR:
1587 		mode |= S_IFCHR;
1588 		break;
1589 	case VLNK:
1590 		mode |= S_IFLNK;
1591 		break;
1592 	case VSOCK:
1593 		mode |= S_IFSOCK;
1594 		break;
1595 	case VFIFO:
1596 		mode |= S_IFIFO;
1597 		break;
1598 	default:
1599 		error = EBADF;
1600 		goto out;
1601 	}
1602 	;
1603 	if (isstat64 != 0) {
1604 		sb64->st_mode = mode;
1605 		sb64->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1606 		sb64->st_uid = va.va_uid;
1607 		sb64->st_gid = va.va_gid;
1608 		sb64->st_rdev = va.va_rdev;
1609 		sb64->st_size = va.va_data_size;
1610 		sb64->st_atimespec = va.va_access_time;
1611 		sb64->st_mtimespec = va.va_modify_time;
1612 		sb64->st_ctimespec = va.va_change_time;
1613 		if (VATTR_IS_SUPPORTED(&va, va_create_time)) {
1614 			sb64->st_birthtimespec =  va.va_create_time;
1615 		} else {
1616 			sb64->st_birthtimespec.tv_sec = sb64->st_birthtimespec.tv_nsec = 0;
1617 		}
1618 		sb64->st_blksize = va.va_iosize;
1619 		sb64->st_flags = va.va_flags;
1620 		sb64->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1621 	} else {
1622 		sb->st_mode = mode;
1623 		sb->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1624 		sb->st_uid = va.va_uid;
1625 		sb->st_gid = va.va_gid;
1626 		sb->st_rdev = va.va_rdev;
1627 		sb->st_size = va.va_data_size;
1628 		sb->st_atimespec = va.va_access_time;
1629 		sb->st_mtimespec = va.va_modify_time;
1630 		sb->st_ctimespec = va.va_change_time;
1631 		sb->st_blksize = va.va_iosize;
1632 		sb->st_flags = va.va_flags;
1633 		sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1634 	}
1635 
1636 	/* if we're interested in extended security data and we got an ACL */
1637 	if (xsec != NULL) {
1638 		if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
1639 		    !VATTR_IS_SUPPORTED(&va, va_uuuid) &&
1640 		    !VATTR_IS_SUPPORTED(&va, va_guuid)) {
1641 			*xsec = KAUTH_FILESEC_NONE;
1642 		} else {
1643 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1644 				fsec = kauth_filesec_alloc(va.va_acl->acl_entrycount);
1645 			} else {
1646 				fsec = kauth_filesec_alloc(0);
1647 			}
1648 			if (fsec == NULL) {
1649 				error = ENOMEM;
1650 				goto out;
1651 			}
1652 			fsec->fsec_magic = KAUTH_FILESEC_MAGIC;
1653 			if (VATTR_IS_SUPPORTED(&va, va_uuuid)) {
1654 				fsec->fsec_owner = va.va_uuuid;
1655 			} else {
1656 				fsec->fsec_owner = kauth_null_guid;
1657 			}
1658 			if (VATTR_IS_SUPPORTED(&va, va_guuid)) {
1659 				fsec->fsec_group = va.va_guuid;
1660 			} else {
1661 				fsec->fsec_group = kauth_null_guid;
1662 			}
1663 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1664 				__nochk_bcopy(va.va_acl, &(fsec->fsec_acl), KAUTH_ACL_COPYSIZE(va.va_acl));
1665 			} else {
1666 				fsec->fsec_acl.acl_entrycount = KAUTH_FILESEC_NOACL;
1667 			}
1668 			*xsec = fsec;
1669 		}
1670 	}
1671 
1672 	/* Do not give the generation number out to unpriviledged users */
1673 	if (va.va_gen && !vfs_context_issuser(ctx)) {
1674 		if (isstat64 != 0) {
1675 			sb64->st_gen = 0;
1676 		} else {
1677 			sb->st_gen = 0;
1678 		}
1679 	} else {
1680 		if (isstat64 != 0) {
1681 			sb64->st_gen = va.va_gen;
1682 		} else {
1683 			sb->st_gen = va.va_gen;
1684 		}
1685 	}
1686 
1687 	error = 0;
1688 out:
1689 	if (VATTR_IS_SUPPORTED(&va, va_acl) && va.va_acl != NULL) {
1690 		kauth_acl_free(va.va_acl);
1691 	}
1692 	return error;
1693 }
1694 
1695 int
vn_stat(struct vnode * vp,void * sb,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx)1696 vn_stat(struct vnode *vp, void *sb, kauth_filesec_t *xsec, int isstat64, int needsrealdev, vfs_context_t ctx)
1697 {
1698 	int error;
1699 
1700 #if CONFIG_MACF
1701 	error = mac_vnode_check_stat(ctx, NOCRED, vp);
1702 	if (error) {
1703 		return error;
1704 	}
1705 #endif
1706 
1707 	/* authorize */
1708 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_ATTRIBUTES | KAUTH_VNODE_READ_SECURITY, ctx)) != 0) {
1709 		return error;
1710 	}
1711 
1712 	/* actual stat */
1713 	return vn_stat_noauth(vp, sb, xsec, isstat64, needsrealdev, ctx, NOCRED);
1714 }
1715 
1716 
1717 /*
1718  * File table vnode ioctl routine.
1719  */
1720 static int
vn_ioctl(struct fileproc * fp,u_long com,caddr_t data,vfs_context_t ctx)1721 vn_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
1722 {
1723 	struct vnode *vp = (struct vnode *)fp_get_data(fp);
1724 	off_t file_size;
1725 	int error;
1726 	struct vnode *ttyvp;
1727 
1728 	if ((error = vnode_getwithref(vp))) {
1729 		return error;
1730 	}
1731 
1732 #if CONFIG_MACF
1733 	error = mac_vnode_check_ioctl(ctx, vp, com);
1734 	if (error) {
1735 		goto out;
1736 	}
1737 #endif
1738 
1739 	switch (vp->v_type) {
1740 	case VREG:
1741 	case VDIR:
1742 		if (com == FIONREAD) {
1743 			off_t temp_nbytes;
1744 			if ((error = vnode_size(vp, &file_size, ctx)) != 0) {
1745 				goto out;
1746 			}
1747 			temp_nbytes = file_size - fp->fp_glob->fg_offset;
1748 			if (temp_nbytes > INT_MAX) {
1749 				*(int *)data = INT_MAX;
1750 			} else if (temp_nbytes < 0) {
1751 				*(int *)data = 0;
1752 			} else {
1753 				*(int *)data = (int)temp_nbytes;
1754 			}
1755 			goto out;
1756 		}
1757 		if (com == FIONBIO || com == FIOASYNC) {        /* XXX */
1758 			goto out;
1759 		}
1760 		OS_FALLTHROUGH;
1761 
1762 	default:
1763 		error = ENOTTY;
1764 		goto out;
1765 
1766 	case VFIFO:
1767 	case VCHR:
1768 	case VBLK:
1769 
1770 		if (com == TIOCREVOKE || com == TIOCREVOKECLEAR) {
1771 			error = ENOTTY;
1772 			goto out;
1773 		}
1774 
1775 		/* Should not be able to set block size from user space */
1776 		if (com == DKIOCSETBLOCKSIZE) {
1777 			error = EPERM;
1778 			goto out;
1779 		}
1780 
1781 		/* Should not be able to check if filesystem is authenticated from user space */
1782 		if (com == FSIOC_AUTH_FS) {
1783 			error = ENOTTY;
1784 			goto out;
1785 		}
1786 
1787 		if (com == FIODTYPE) {
1788 			if (vp->v_type == VBLK) {
1789 				if (major(vp->v_rdev) >= nblkdev) {
1790 					error = ENXIO;
1791 					goto out;
1792 				}
1793 				*(int *)data = bdevsw[major(vp->v_rdev)].d_type;
1794 			} else if (vp->v_type == VCHR) {
1795 				if (major(vp->v_rdev) >= nchrdev) {
1796 					error = ENXIO;
1797 					goto out;
1798 				}
1799 				*(int *)data = cdevsw[major(vp->v_rdev)].d_type;
1800 			} else {
1801 				error = ENOTTY;
1802 				goto out;
1803 			}
1804 			goto out;
1805 		}
1806 		error = VNOP_IOCTL(vp, com, data, fp->fp_glob->fg_flag, ctx);
1807 
1808 		if (error == 0 && com == TIOCSCTTY) {
1809 			struct session *sessp;
1810 			struct pgrp *pg;
1811 
1812 			pg = proc_pgrp(vfs_context_proc(ctx), &sessp);
1813 
1814 			session_lock(sessp);
1815 			ttyvp = sessp->s_ttyvp;
1816 			sessp->s_ttyvp = vp;
1817 			sessp->s_ttyvid = vnode_vid(vp);
1818 			session_unlock(sessp);
1819 
1820 			pgrp_rele(pg);
1821 		}
1822 	}
1823 out:
1824 	(void)vnode_put(vp);
1825 	return error;
1826 }
1827 
1828 /*
1829  * File table vnode select routine.
1830  */
1831 static int
vn_select(struct fileproc * fp,int which,void * wql,__unused vfs_context_t ctx)1832 vn_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx)
1833 {
1834 	int error;
1835 	struct vnode * vp = (struct vnode *)fp_get_data(fp);
1836 	struct vfs_context context;
1837 
1838 	if ((error = vnode_getwithref(vp)) == 0) {
1839 		context.vc_thread = current_thread();
1840 		context.vc_ucred = fp->fp_glob->fg_cred;
1841 
1842 #if CONFIG_MACF
1843 		/*
1844 		 * XXX We should use a per thread credential here; minimally,
1845 		 * XXX the process credential should have a persistent
1846 		 * XXX reference on it before being passed in here.
1847 		 */
1848 		error = mac_vnode_check_select(ctx, vp, which);
1849 		if (error == 0)
1850 #endif
1851 		error = VNOP_SELECT(vp, which, fp->fp_glob->fg_flag, wql, ctx);
1852 
1853 		(void)vnode_put(vp);
1854 	}
1855 	return error;
1856 }
1857 
1858 /*
1859  * File table vnode close routine.
1860  */
1861 static int
vn_closefile(struct fileglob * fg,vfs_context_t ctx)1862 vn_closefile(struct fileglob *fg, vfs_context_t ctx)
1863 {
1864 	struct vnode *vp = fg_get_data(fg);
1865 	int error;
1866 
1867 	if ((error = vnode_getwithref(vp)) == 0) {
1868 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE &&
1869 		    ((fg->fg_flag & FWASLOCKED) != 0 ||
1870 		    (fg->fg_lflags & FG_HAS_OFDLOCK) != 0)) {
1871 			struct flock lf = {
1872 				.l_whence = SEEK_SET,
1873 				.l_start = 0,
1874 				.l_len = 0,
1875 				.l_type = F_UNLCK
1876 			};
1877 
1878 			if ((fg->fg_flag & FWASLOCKED) != 0) {
1879 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1880 				    F_UNLCK, &lf, F_FLOCK, ctx, NULL);
1881 			}
1882 
1883 			if ((fg->fg_lflags & FG_HAS_OFDLOCK) != 0) {
1884 				(void) VNOP_ADVLOCK(vp, ofd_to_id(fg),
1885 				    F_UNLCK, &lf, F_OFD_LOCK, ctx, NULL);
1886 			}
1887 		}
1888 #if CONFIG_FILE_LEASES
1889 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE && !LIST_EMPTY(&vp->v_leases)) {
1890 			/* Expected open count doesn't matter for release. */
1891 			(void)vnode_setlease(vp, fg, F_UNLCK, 0, ctx);
1892 		}
1893 #endif
1894 		error = vn_close(vp, fg->fg_flag, ctx);
1895 		(void) vnode_put(vp);
1896 	}
1897 	return error;
1898 }
1899 
1900 /*
1901  * Returns:	0			Success
1902  *	VNOP_PATHCONF:???
1903  */
1904 int
vn_pathconf(vnode_t vp,int name,int32_t * retval,vfs_context_t ctx)1905 vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
1906 {
1907 	int     error = 0;
1908 	struct vfs_attr vfa;
1909 
1910 	switch (name) {
1911 	case _PC_EXTENDED_SECURITY_NP:
1912 		*retval = vfs_extendedsecurity(vnode_mount(vp)) ? 1 : 0;
1913 		break;
1914 	case _PC_AUTH_OPAQUE_NP:
1915 		*retval = vfs_authopaque(vnode_mount(vp));
1916 		break;
1917 	case _PC_2_SYMLINKS:
1918 		*retval = 1;    /* XXX NOTSUP on MSDOS, etc. */
1919 		break;
1920 	case _PC_ALLOC_SIZE_MIN:
1921 		*retval = 1;    /* XXX lie: 1 byte */
1922 		break;
1923 	case _PC_ASYNC_IO:      /* unistd.h: _POSIX_ASYNCHRONUS_IO */
1924 		*retval = 1;    /* [AIO] option is supported */
1925 		break;
1926 	case _PC_PRIO_IO:       /* unistd.h: _POSIX_PRIORITIZED_IO */
1927 		*retval = 0;    /* [PIO] option is not supported */
1928 		break;
1929 	case _PC_REC_INCR_XFER_SIZE:
1930 		*retval = 4096; /* XXX go from MIN to MAX 4K at a time */
1931 		break;
1932 	case _PC_REC_MIN_XFER_SIZE:
1933 		*retval = 4096; /* XXX recommend 4K minimum reads/writes */
1934 		break;
1935 	case _PC_REC_MAX_XFER_SIZE:
1936 		*retval = 65536; /* XXX recommend 64K maximum reads/writes */
1937 		break;
1938 	case _PC_REC_XFER_ALIGN:
1939 		*retval = 4096; /* XXX recommend page aligned buffers */
1940 		break;
1941 	case _PC_SYMLINK_MAX:
1942 		*retval = 255;  /* Minimum acceptable POSIX value */
1943 		break;
1944 	case _PC_SYNC_IO:       /* unistd.h: _POSIX_SYNCHRONIZED_IO */
1945 		*retval = 0;    /* [SIO] option is not supported */
1946 		break;
1947 	case _PC_XATTR_SIZE_BITS:
1948 		/* The number of bits used to store maximum extended
1949 		 * attribute size in bytes.  For example, if the maximum
1950 		 * attribute size supported by a file system is 128K, the
1951 		 * value returned will be 18.  However a value 18 can mean
1952 		 * that the maximum attribute size can be anywhere from
1953 		 * (256KB - 1) to 128KB.  As a special case, the resource
1954 		 * fork can have much larger size, and some file system
1955 		 * specific extended attributes can have smaller and preset
1956 		 * size; for example, Finder Info is always 32 bytes.
1957 		 */
1958 		memset(&vfa, 0, sizeof(vfa));
1959 		VFSATTR_INIT(&vfa);
1960 		VFSATTR_WANTED(&vfa, f_capabilities);
1961 		if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
1962 		    (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
1963 		    (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
1964 		    (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
1965 			/* Supports native extended attributes */
1966 			error = VNOP_PATHCONF(vp, name, retval, ctx);
1967 		} else {
1968 			/* Number of bits used to represent the maximum size of
1969 			 * extended attribute stored in an Apple Double file.
1970 			 */
1971 			*retval = AD_XATTR_SIZE_BITS;
1972 		}
1973 		break;
1974 	default:
1975 		error = VNOP_PATHCONF(vp, name, retval, ctx);
1976 		break;
1977 	}
1978 
1979 	return error;
1980 }
1981 
1982 static int
vn_kqfilter(struct fileproc * fp,struct knote * kn,struct kevent_qos_s * kev)1983 vn_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
1984 {
1985 	vfs_context_t ctx = vfs_context_current();
1986 	struct vnode *vp;
1987 	int error = 0;
1988 	int result = 0;
1989 
1990 	vp = (struct vnode *)fp_get_data(fp);
1991 
1992 	/*
1993 	 * Don't attach a knote to a dead vnode.
1994 	 */
1995 	if ((error = vget_internal(vp, 0, VNODE_NODEAD)) == 0) {
1996 		switch (kn->kn_filter) {
1997 		case EVFILT_READ:
1998 		case EVFILT_WRITE:
1999 			if (vnode_isfifo(vp)) {
2000 				/* We'll only watch FIFOs that use our fifofs */
2001 				if (!(vp->v_fifoinfo && vp->v_fifoinfo->fi_readsock)) {
2002 					error = ENOTSUP;
2003 				}
2004 			} else if (!vnode_isreg(vp)) {
2005 				if (vnode_ischr(vp)) {
2006 					result = spec_kqfilter(vp, kn, kev);
2007 					if ((kn->kn_flags & EV_ERROR) == 0) {
2008 						/* claimed by a special device */
2009 						vnode_put(vp);
2010 						return result;
2011 					}
2012 				}
2013 				error = EINVAL;
2014 			}
2015 			break;
2016 		case EVFILT_VNODE:
2017 			break;
2018 		default:
2019 			error = EINVAL;
2020 		}
2021 
2022 		if (error == 0) {
2023 #if CONFIG_MACF
2024 			error = mac_vnode_check_kqfilter(ctx, fp->fp_glob->fg_cred, kn, vp);
2025 			if (error) {
2026 				vnode_put(vp);
2027 				goto out;
2028 			}
2029 #endif
2030 
2031 			kn->kn_filtid = EVFILTID_VN;
2032 			knote_kn_hook_set_raw(kn, (void *)vp);
2033 			vnode_hold(vp);
2034 
2035 			vnode_lock(vp);
2036 			KNOTE_ATTACH(&vp->v_knotes, kn);
2037 			result = filt_vnode_common(kn, NULL, vp, 0);
2038 			vnode_unlock(vp);
2039 
2040 			/*
2041 			 * Ask the filesystem to provide remove notifications,
2042 			 * but ignore failure
2043 			 */
2044 			VNOP_MONITOR(vp, 0, VNODE_MONITOR_BEGIN, (void*) kn, ctx);
2045 		}
2046 
2047 		vnode_put(vp);
2048 	}
2049 
2050 out:
2051 	if (error) {
2052 		knote_set_error(kn, error);
2053 	}
2054 
2055 	return result;
2056 }
2057 
2058 static void
filt_vndetach(struct knote * kn)2059 filt_vndetach(struct knote *kn)
2060 {
2061 	vfs_context_t ctx = vfs_context_current();
2062 	struct vnode *vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2063 	uint32_t vid = vnode_vid(vp);
2064 	if (vnode_getwithvid(vp, vid)) {
2065 		vnode_drop(vp);
2066 		return;
2067 	}
2068 	vnode_drop(vp);
2069 
2070 	vnode_lock(vp);
2071 	KNOTE_DETACH(&vp->v_knotes, kn);
2072 	vnode_unlock(vp);
2073 
2074 	/*
2075 	 * Tell a (generally networked) filesystem that we're no longer watching
2076 	 * If the FS wants to track contexts, it should still be using the one from
2077 	 * the VNODE_MONITOR_BEGIN.
2078 	 */
2079 	VNOP_MONITOR(vp, 0, VNODE_MONITOR_END, (void*)kn, ctx);
2080 	vnode_put(vp);
2081 }
2082 
2083 
2084 /*
2085  * Used for EVFILT_READ
2086  *
2087  * Takes only VFIFO or VREG. vnode is locked.  We handle the "poll" case
2088  * differently than the regular case for VREG files.  If not in poll(),
2089  * then we need to know current fileproc offset for VREG.
2090  */
2091 static int64_t
vnode_readable_data_count(vnode_t vp,off_t current_offset,int ispoll)2092 vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
2093 {
2094 	if (vnode_isfifo(vp)) {
2095 #if FIFO
2096 		int cnt;
2097 		int err = fifo_charcount(vp, &cnt);
2098 		if (err == 0) {
2099 			return (int64_t)cnt;
2100 		} else
2101 #endif
2102 		{
2103 			return 0;
2104 		}
2105 	} else if (vnode_isreg(vp)) {
2106 		if (ispoll) {
2107 			return 1;
2108 		}
2109 
2110 		off_t amount;
2111 		amount = vp->v_un.vu_ubcinfo->ui_size - current_offset;
2112 		if (amount > INT64_MAX) {
2113 			return INT64_MAX;
2114 		} else if (amount < INT64_MIN) {
2115 			return INT64_MIN;
2116 		} else {
2117 			return (int64_t)amount;
2118 		}
2119 	} else {
2120 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2121 		return 0;
2122 	}
2123 }
2124 
2125 /*
2126  * Used for EVFILT_WRITE.
2127  *
2128  * For regular vnodes, we can always write (1).  For named pipes,
2129  * see how much space there is in the buffer.  Nothing else is covered.
2130  */
2131 static intptr_t
vnode_writable_space_count(vnode_t vp)2132 vnode_writable_space_count(vnode_t vp)
2133 {
2134 	if (vnode_isfifo(vp)) {
2135 #if FIFO
2136 		long spc;
2137 		int err = fifo_freespace(vp, &spc);
2138 		if (err == 0) {
2139 			return (intptr_t)spc;
2140 		} else
2141 #endif
2142 		{
2143 			return (intptr_t)0;
2144 		}
2145 	} else if (vnode_isreg(vp)) {
2146 		return (intptr_t)1;
2147 	} else {
2148 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2149 		return 0;
2150 	}
2151 }
2152 
2153 /*
2154  * Determine whether this knote should be active
2155  *
2156  * This is kind of subtle.
2157  *      --First, notice if the vnode has been revoked: in so, override hint
2158  *      --EVFILT_READ knotes are checked no matter what the hint is
2159  *      --Other knotes activate based on hint.
2160  *      --If hint is revoke, set special flags and activate
2161  */
2162 static int
filt_vnode_common(struct knote * kn,struct kevent_qos_s * kev,vnode_t vp,long hint)2163 filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev, vnode_t vp, long hint)
2164 {
2165 	int activate = 0;
2166 	int64_t data = 0;
2167 
2168 	lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
2169 
2170 	/* Special handling for vnodes that are in recycle or already gone */
2171 	if (NOTE_REVOKE == hint) {
2172 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2173 		activate = 1;
2174 
2175 		if ((kn->kn_filter == EVFILT_VNODE) && (kn->kn_sfflags & NOTE_REVOKE)) {
2176 			kn->kn_fflags |= NOTE_REVOKE;
2177 		}
2178 	} else {
2179 		switch (kn->kn_filter) {
2180 		case EVFILT_READ:
2181 			data = vnode_readable_data_count(vp, kn->kn_fp->fp_glob->fg_offset, (kn->kn_flags & EV_POLL));
2182 			activate = (data != 0);
2183 			break;
2184 		case EVFILT_WRITE:
2185 			data = vnode_writable_space_count(vp);
2186 			activate = (data != 0);
2187 			break;
2188 		case EVFILT_VNODE:
2189 			/* Check events this note matches against the hint */
2190 			if (kn->kn_sfflags & hint) {
2191 				kn->kn_fflags |= (uint32_t)hint; /* Set which event occurred */
2192 			}
2193 			activate = (kn->kn_fflags != 0);
2194 			break;
2195 		default:
2196 			panic("Invalid knote filter on a vnode!");
2197 		}
2198 	}
2199 
2200 	if (kev && activate) {
2201 		knote_fill_kevent(kn, kev, data);
2202 	}
2203 
2204 	return activate;
2205 }
2206 
2207 static int
filt_vnode(struct knote * kn,long hint)2208 filt_vnode(struct knote *kn, long hint)
2209 {
2210 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2211 
2212 	return filt_vnode_common(kn, NULL, vp, hint);
2213 }
2214 
2215 static int
filt_vntouch(struct knote * kn,struct kevent_qos_s * kev)2216 filt_vntouch(struct knote *kn, struct kevent_qos_s *kev)
2217 {
2218 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2219 	uint32_t vid = vnode_vid(vp);
2220 	int activate;
2221 	int hint = 0;
2222 
2223 	vnode_lock(vp);
2224 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2225 		/* is recycled */
2226 		hint = NOTE_REVOKE;
2227 	}
2228 
2229 	/* accept new input fflags mask */
2230 	kn->kn_sfflags = kev->fflags;
2231 
2232 	activate = filt_vnode_common(kn, NULL, vp, hint);
2233 
2234 	if (hint == 0) {
2235 		vnode_put_locked(vp);
2236 	}
2237 	vnode_unlock(vp);
2238 
2239 	return activate;
2240 }
2241 
2242 static int
filt_vnprocess(struct knote * kn,struct kevent_qos_s * kev)2243 filt_vnprocess(struct knote *kn, struct kevent_qos_s *kev)
2244 {
2245 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2246 	uint32_t vid = vnode_vid(vp);
2247 	int activate;
2248 	int hint = 0;
2249 
2250 	vnode_lock(vp);
2251 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2252 		/* Is recycled */
2253 		hint = NOTE_REVOKE;
2254 	}
2255 	activate = filt_vnode_common(kn, kev, vp, hint);
2256 
2257 	/* Definitely need to unlock, may need to put */
2258 	if (hint == 0) {
2259 		vnode_put_locked(vp);
2260 	}
2261 	vnode_unlock(vp);
2262 
2263 	return activate;
2264 }
2265 
2266 struct vniodesc {
2267 	vnode_t         vnio_vnode;     /* associated vnode */
2268 	int             vnio_fflags;    /* cached fileglob flags */
2269 };
2270 
2271 errno_t
vnio_openfd(int fd,vniodesc_t * vniop)2272 vnio_openfd(int fd, vniodesc_t *vniop)
2273 {
2274 	proc_t p = current_proc();
2275 	struct fileproc *fp = NULL;
2276 	vniodesc_t vnio;
2277 	vnode_t vp;
2278 	int error;
2279 	bool need_drop = false;
2280 
2281 	vnio = kalloc_type(struct vniodesc, Z_WAITOK);
2282 
2283 	error = fp_getfvp(p, fd, &fp, &vp);
2284 	if (error) {
2285 		goto out;
2286 	}
2287 	need_drop = true;
2288 
2289 	if ((fp->fp_glob->fg_flag & (FREAD | FWRITE)) == 0) {
2290 		error = EBADF;
2291 		goto out;
2292 	}
2293 
2294 	if (vnode_isswap(vp)) {
2295 		error = EPERM;
2296 		goto out;
2297 	}
2298 
2299 	if (vp->v_type != VREG) {
2300 		error = EFTYPE;
2301 		goto out;
2302 	}
2303 
2304 	error = vnode_getwithref(vp);
2305 	if (error) {
2306 		goto out;
2307 	}
2308 
2309 	vnio->vnio_vnode = vp;
2310 	vnio->vnio_fflags = fp->fp_glob->fg_flag;
2311 
2312 	(void)fp_drop(p, fd, fp, 0);
2313 	need_drop = false;
2314 
2315 	error = vnode_ref_ext(vp, vnio->vnio_fflags, 0);
2316 	if (error == 0) {
2317 		*vniop = vnio;
2318 		vnio = NULL;
2319 	}
2320 
2321 	vnode_put(vp);
2322 
2323 out:
2324 	if (need_drop) {
2325 		fp_drop(p, fd, fp, 0);
2326 	}
2327 	if (vnio != NULL) {
2328 		kfree_type(struct vniodesc, vnio);
2329 	}
2330 	return error;
2331 }
2332 
2333 errno_t
vnio_close(vniodesc_t vnio)2334 vnio_close(vniodesc_t vnio)
2335 {
2336 	int error;
2337 
2338 	/*
2339 	 * The vniodesc is always destroyed, because the close
2340 	 * always "succeeds".  We just return whatever error
2341 	 * might have been encountered while doing so.
2342 	 */
2343 
2344 	error = vnode_getwithref(vnio->vnio_vnode);
2345 	if (error == 0) {
2346 		error = vnode_close(vnio->vnio_vnode, vnio->vnio_fflags, NULL);
2347 	}
2348 
2349 	kfree_type(struct vniodesc, vnio);
2350 
2351 	return error;
2352 }
2353 
2354 errno_t
vnio_read(vniodesc_t vnio,uio_t uio)2355 vnio_read(vniodesc_t vnio, uio_t uio)
2356 {
2357 	vnode_t vp = vnio->vnio_vnode;
2358 	user_ssize_t read_len;
2359 	int error;
2360 
2361 	read_len = uio_resid(uio);
2362 	if (read_len < 0 || read_len > INT_MAX) {
2363 		return EINVAL;
2364 	}
2365 
2366 	error = vnode_getwithref(vp);
2367 	if (error == 0) {
2368 		error = vn_read_common(vp, uio, vnio->vnio_fflags,
2369 		    vfs_context_current());
2370 		vnode_put(vp);
2371 	}
2372 
2373 	return error;
2374 }
2375 
2376 vnode_t
vnio_vnode(vniodesc_t vnio)2377 vnio_vnode(vniodesc_t vnio)
2378 {
2379 	return vnio->vnio_vnode;
2380 }
2381 
2382 int
vnode_isauthfs(vnode_t vp)2383 vnode_isauthfs(vnode_t vp)
2384 {
2385 	fsioc_auth_fs_t afs = { .authvp = NULL };
2386 	int error;
2387 
2388 	error = VNOP_IOCTL(vp, FSIOC_AUTH_FS, (caddr_t)&afs, 0,
2389 	    vfs_context_current());
2390 
2391 	return error == 0 ? 1 : 0;
2392 }
2393