xref: /xnu-11215.1.10/bsd/vfs/vfs_vnops.c (revision 8d741a5de7ff4191bf97d57b9f54c2f6d4a15585)
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;
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 			return EINVAL;
404 		}
405 		// If raw encrypted mode is requested, handle that here
406 		if (ISSET(vap->va_dataprotect_flags, VA_DP_RAWENCRYPTED)) {
407 			fmode |= FENCRYPTED;
408 		}
409 	}
410 
411 	if ((fmode & O_NOFOLLOW_ANY) && (fmode & O_NOFOLLOW)) {
412 		error = EINVAL;
413 		goto out;
414 	}
415 
416 	/*
417 	 * O_CREAT
418 	 */
419 	if (fmode & O_CREAT) {
420 		if ((fmode & O_DIRECTORY)) {
421 			error = EINVAL;
422 			goto out;
423 		}
424 		ndp->ni_cnd.cn_nameiop = CREATE;
425 #if CONFIG_TRIGGERS
426 		ndp->ni_op = OP_LINK;
427 #endif
428 		/* Inherit USEDVP, vnode_open() supported flags only */
429 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
430 		ndp->ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF | AUDITVNPATH1;
431 		/* Inherit NAMEI_ROOTDIR flag only */
432 		ndp->ni_flag &= NAMEI_ROOTDIR;
433 		ndp->ni_flag |= NAMEI_COMPOUNDOPEN;
434 #if NAMEDRSRCFORK
435 		/* open calls are allowed for resource forks. */
436 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
437 #endif
438 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0 && (origcnflags & FOLLOW) != 0) {
439 			ndp->ni_cnd.cn_flags |= FOLLOW;
440 		}
441 		if (fmode & O_NOFOLLOW_ANY) {
442 			/* will return ELOOP on the first symlink to be hit */
443 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
444 		}
445 
446 continue_create_lookup:
447 		if ((error = namei(ndp))) {
448 			goto out;
449 		}
450 
451 		dvp = ndp->ni_dvp;
452 		vp = ndp->ni_vp;
453 
454 		batched = vnode_compound_open_available(dvp);
455 
456 		/* not found, create */
457 		if (vp == NULL) {
458 			/* must have attributes for a new file */
459 			if (vap == NULL) {
460 				vnode_put(dvp);
461 				error = EINVAL;
462 				goto out;
463 			}
464 			/*
465 			 * Attempt a create.   For a system supporting compound VNOPs, we may
466 			 * find an existing file or create one; in either case, we will already
467 			 * have the file open and no VNOP_OPEN() will be needed.
468 			 */
469 			error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
470 
471 			dvp = ndp->ni_dvp;
472 			vp = ndp->ni_vp;
473 
474 			/*
475 			 * Detected a node that the filesystem couldn't handle.  Don't call
476 			 * nameidone() yet, because we need that path buffer.
477 			 */
478 			if (error == EKEEPLOOKING) {
479 				if (!batched) {
480 					panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
481 				}
482 				goto continue_create_lookup;
483 			}
484 
485 			nameidone(ndp);
486 			if (dvp) {
487 				panic("Shouldn't have a dvp here.");
488 			}
489 
490 			if (error) {
491 				/*
492 				 * Check for a create race.
493 				 */
494 				if ((error == EEXIST) && !(fmode & O_EXCL)) {
495 					if (vp) {
496 						vnode_put(vp);
497 					}
498 					goto again;
499 				}
500 				goto bad;
501 			}
502 
503 			need_vnop_open = !did_open;
504 		} else {
505 			if (fmode & O_EXCL) {
506 				error = EEXIST;
507 			}
508 
509 			/*
510 			 * We have a vnode.  Use compound open if available
511 			 * or else fall through to "traditional" path.  Note: can't
512 			 * do a compound open for root, because the parent belongs
513 			 * to a different FS.
514 			 */
515 			if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
516 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
517 
518 				if (error == 0) {
519 					vp = ndp->ni_vp;
520 					need_vnop_open = FALSE;
521 				} else if (error == EKEEPLOOKING) {
522 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
523 						panic("EKEEPLOOKING, but continue flag not set?");
524 					}
525 					goto continue_create_lookup;
526 				}
527 			}
528 			nameidone(ndp);
529 			vnode_put(dvp);
530 			ndp->ni_dvp = NULLVP;
531 
532 			if (error) {
533 				goto bad;
534 			}
535 
536 			fmode &= ~O_CREAT;
537 
538 			/* Fall through */
539 		}
540 	} else {
541 		/*
542 		 * Not O_CREAT
543 		 */
544 		ndp->ni_cnd.cn_nameiop = LOOKUP;
545 		/* Inherit USEDVP, vnode_open() supported flags only */
546 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
547 		ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1 | WANTPARENT;
548 #if NAMEDRSRCFORK
549 		/* open calls are allowed for resource forks. */
550 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
551 #endif
552 		if (fmode & FENCRYPTED) {
553 			ndp->ni_cnd.cn_flags |= CN_RAW_ENCRYPTED | CN_SKIPNAMECACHE;
554 		}
555 		/* Inherit NAMEI_ROOTDIR flag only */
556 		ndp->ni_flag &= NAMEI_ROOTDIR;
557 		ndp->ni_flag |= NAMEI_COMPOUNDOPEN;
558 
559 		/* preserve NOFOLLOW from vnode_open() */
560 		if (fmode & O_NOFOLLOW || fmode & O_SYMLINK || (origcnflags & FOLLOW) == 0) {
561 			ndp->ni_cnd.cn_flags &= ~FOLLOW;
562 		}
563 		if (fmode & O_NOFOLLOW_ANY) {
564 			/* will return ELOOP on the first symlink to be hit */
565 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
566 		}
567 
568 		/* Do a lookup, possibly going directly to filesystem for compound operation */
569 		do {
570 			if ((error = namei(ndp))) {
571 				goto out;
572 			}
573 			vp = ndp->ni_vp;
574 			dvp = ndp->ni_dvp;
575 
576 			/* Check for batched lookup-open */
577 			batched = vnode_compound_open_available(dvp);
578 			if (batched && ((vp == NULLVP) || (vnode_mount(dvp) == vnode_mount(vp)))) {
579 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
580 				vp = ndp->ni_vp;
581 				if (error == 0) {
582 					need_vnop_open = FALSE;
583 				} else if (error == EKEEPLOOKING) {
584 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
585 						panic("EKEEPLOOKING, but continue flag not set?");
586 					}
587 				} else if ((fmode & (FREAD | FWRITE | FEXEC)) == FEXEC) {
588 					/*
589 					 * Some file systems fail in vnop_open call with absense of
590 					 * both FREAD and FWRITE access modes. Retry the vnop_open
591 					 * call again with FREAD access mode added.
592 					 */
593 					error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0,
594 					    fmode | FREAD, NULL, NULL, ctx);
595 					if (error == 0) {
596 						vp = ndp->ni_vp;
597 						need_vnop_open = FALSE;
598 					}
599 				}
600 			}
601 		} while (error == EKEEPLOOKING);
602 
603 		nameidone(ndp);
604 		vnode_put(dvp);
605 		ndp->ni_dvp = NULLVP;
606 
607 		if (error) {
608 			goto bad;
609 		}
610 	}
611 
612 	/*
613 	 * By this point, nameidone() is called, dvp iocount is dropped,
614 	 * and dvp pointer is cleared.
615 	 */
616 	if (ndp->ni_dvp != NULLVP) {
617 		panic("Haven't cleaned up adequately in vn_open_auth()");
618 	}
619 
620 	/*
621 	 * Expect to use this code for filesystems without compound VNOPs, for the root
622 	 * of a filesystem, which can't be "looked up" in the sense of VNOP_LOOKUP(),
623 	 * and for shadow files, which do not live on the same filesystems as their "parents."
624 	 */
625 	if (need_vnop_open) {
626 		if (batched && !vnode_isvroot(vp) && !vnode_isnamedstream(vp)) {
627 			panic("Why am I trying to use VNOP_OPEN() on anything other than the root or a named stream?");
628 		}
629 
630 		if (!did_create) {
631 			error = vn_authorize_open_existing(vp, &ndp->ni_cnd, fmode, ctx, NULL);
632 			if (error) {
633 				goto bad;
634 			}
635 		}
636 
637 		if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)) {
638 			if (ISSET(vap->va_dataprotect_flags, VA_DP_RAWUNENCRYPTED)) {
639 				/* Don't allow unencrypted io request from user space unless entitled */
640 				boolean_t entitled = FALSE;
641 #if !SECURE_KERNEL
642 				entitled = IOCurrentTaskHasEntitlement("com.apple.private.security.file-unencrypt-access");
643 #endif /* SECURE_KERNEL */
644 				if (!entitled) {
645 					error = EPERM;
646 					goto bad;
647 				}
648 				fmode |= FUNENCRYPTED;
649 			}
650 
651 			if (ISSET(vap->va_dataprotect_flags, VA_DP_AUTHENTICATE)) {
652 				fsioc_auth_fs_t afs = { .authvp = authvp };
653 
654 				error = VNOP_IOCTL(vp, FSIOC_AUTH_FS, (caddr_t)&afs, 0, ctx);
655 				if (error) {
656 					goto bad;
657 				}
658 			}
659 		}
660 
661 		error = VNOP_OPEN(vp, fmode, ctx);
662 		if (error) {
663 			/*
664 			 * Some file systems fail in vnop_open call with absense of both
665 			 * FREAD and FWRITE access modes. Retry the vnop_open call again
666 			 * with FREAD access mode added.
667 			 */
668 			if ((fmode & (FREAD | FWRITE | FEXEC)) == FEXEC) {
669 				error = VNOP_OPEN(vp, fmode | FREAD, ctx);
670 			}
671 			if (error) {
672 				goto bad;
673 			}
674 		}
675 		need_vnop_open = FALSE;
676 	}
677 
678 	// if the vnode is tagged VOPENEVT and the current process
679 	// has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
680 	// flag to the open mode so that this open won't count against
681 	// the vnode when carbon delete() does a vnode_isinuse() to see
682 	// if a file is currently in use.  this allows spotlight
683 	// importers to not interfere with carbon apps that depend on
684 	// the no-delete-if-busy semantics of carbon delete().
685 	//
686 	if (!did_create && (vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
687 		fmode |= O_EVTONLY;
688 	}
689 
690 	/*
691 	 * Grab reference, etc.
692 	 */
693 	error = vn_open_auth_finish(vp, fmode, ctx);
694 	if (error) {
695 		ref_failed = TRUE;
696 		goto bad;
697 	}
698 
699 	/* Compound VNOP open is responsible for doing the truncate */
700 	if (batched || did_create) {
701 		fmode &= ~O_TRUNC;
702 	}
703 
704 	*fmodep = fmode;
705 	return 0;
706 
707 bad:
708 	/* Opened either explicitly or by a batched create */
709 	if (!need_vnop_open) {
710 		VNOP_CLOSE(vp, fmode, ctx);
711 	}
712 
713 	ndp->ni_vp = NULL;
714 	if (vp) {
715 #if NAMEDRSRCFORK
716 		/* Aggressively recycle shadow files if we error'd out during open() */
717 		if ((vnode_isnamedstream(vp)) &&
718 		    (vp->v_parent != NULLVP) &&
719 		    (vnode_isshadow(vp))) {
720 			vnode_recycle(vp);
721 		}
722 #endif
723 		vnode_put(vp);
724 		/*
725 		 * Check for a race against unlink.  We had a vnode
726 		 * but according to vnode_authorize or VNOP_OPEN it
727 		 * no longer exists.
728 		 *
729 		 * EREDRIVEOPEN: means that we were hit by the tty allocation race or NFSv4 open/create race.
730 		 */
731 		if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN) || ref_failed) {
732 			/*
733 			 * We'll retry here but it may be possible that we get
734 			 * into a retry "spin" inside the kernel and not allow
735 			 * threads, which need to run in order for the retry
736 			 * loop to end, to run. An example is an open of a
737 			 * terminal which is getting revoked and we spin here
738 			 * without yielding becasue namei and VNOP_OPEN are
739 			 * successful but vnode_ref fails. The revoke needs
740 			 * threads with an iocount to run but if spin here we
741 			 * may possibly be blcoking other threads from running.
742 			 *
743 			 * We start yielding the CPU after some number of
744 			 * retries for increasing durations. Note that this is
745 			 * still a loop without an exit condition.
746 			 */
747 			nretries += 1;
748 			if (nretries > RETRY_NO_YIELD_COUNT) {
749 				/* Every hz/100 secs is 10 msecs ... */
750 				tsleep(&nretries, PVFS, "vn_open_auth_retry",
751 				    MIN((nretries * (hz / 100)), hz));
752 			}
753 			goto again;
754 		}
755 	}
756 
757 out:
758 	return error;
759 }
760 
761 #if vn_access_DEPRECATED
762 /*
763  * Authorize an action against a vnode.  This has been the canonical way to
764  * ensure that the credential/process/etc. referenced by a vfs_context
765  * is granted the rights called out in 'mode' against the vnode 'vp'.
766  *
767  * Unfortunately, the use of VREAD/VWRITE/VEXEC makes it very difficult
768  * to add support for more rights.  As such, this interface will be deprecated
769  * and callers will use vnode_authorize instead.
770  */
771 int
vn_access(vnode_t vp,int mode,vfs_context_t context)772 vn_access(vnode_t vp, int mode, vfs_context_t context)
773 {
774 	kauth_action_t  action;
775 
776 	action = 0;
777 	if (mode & VREAD) {
778 		action |= KAUTH_VNODE_READ_DATA;
779 	}
780 	if (mode & VWRITE) {
781 		action |= KAUTH_VNODE_WRITE_DATA;
782 	}
783 	if (mode & VEXEC) {
784 		action |= KAUTH_VNODE_EXECUTE;
785 	}
786 
787 	return vnode_authorize(vp, NULL, action, context);
788 }
789 #endif  /* vn_access_DEPRECATED */
790 
791 /*
792  * Vnode close call
793  */
794 int
vn_close(struct vnode * vp,int flags,vfs_context_t ctx)795 vn_close(struct vnode *vp, int flags, vfs_context_t ctx)
796 {
797 	int error;
798 	int flusherror = 0;
799 
800 #if NAMEDRSRCFORK
801 	/* Sync data from resource fork shadow file if needed. */
802 	if ((vp->v_flag & VISNAMEDSTREAM) &&
803 	    (vp->v_parent != NULLVP) &&
804 	    vnode_isshadow(vp)) {
805 		if (flags & FWASWRITTEN) {
806 			flusherror = vnode_flushnamedstream(vp->v_parent, vp, ctx);
807 		}
808 	}
809 #endif
810 	/*
811 	 * If vnode @vp belongs to a chardev or a blkdev then it is handled
812 	 * specially.  We first drop its user reference count @vp->v_usecount
813 	 * before calling VNOP_CLOSE().  This was done historically to ensure
814 	 * that the last close of a special device vnode performed some
815 	 * conditional cleanups.  Now we still need to drop this reference here
816 	 * to ensure that devfsspec_close() can check if the vnode is still in
817 	 * use.
818 	 */
819 	if (vnode_isspec(vp)) {
820 		(void)vnode_rele_ext(vp, flags, 0);
821 	}
822 
823 	/*
824 	 * On HFS, we flush when the last writer closes.  We do this
825 	 * because resource fork vnodes hold a reference on data fork
826 	 * vnodes and that will prevent them from getting VNOP_INACTIVE
827 	 * which will delay when we flush cached data.  In future, we
828 	 * might find it beneficial to do this for all file systems.
829 	 * Note that it's OK to access v_writecount without the lock
830 	 * in this context.
831 	 */
832 	if (vp->v_tag == VT_HFS && (flags & FWRITE) && vp->v_writecount == 1) {
833 		VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
834 	}
835 
836 	error = VNOP_CLOSE(vp, flags, ctx);
837 
838 #if CONFIG_FSE
839 	if (flags & FWASWRITTEN) {
840 		if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
841 			add_fsevent(FSE_CONTENT_MODIFIED, ctx,
842 			    FSE_ARG_VNODE, vp,
843 			    FSE_ARG_DONE);
844 		}
845 	}
846 #endif
847 
848 	if (!vnode_isspec(vp)) {
849 		(void)vnode_rele_ext(vp, flags, 0);
850 	}
851 
852 	if (flusherror) {
853 		error = flusherror;
854 	}
855 	return error;
856 }
857 
858 static int
vn_read_swapfile(struct vnode * vp,uio_t uio)859 vn_read_swapfile(
860 	struct vnode    *vp,
861 	uio_t           uio)
862 {
863 	int     error;
864 	off_t   swap_count, this_count;
865 	off_t   file_end, read_end;
866 	off_t   prev_resid;
867 	char    *my_swap_page;
868 
869 	/*
870 	 * Reading from a swap file will get you zeroes.
871 	 */
872 
873 	my_swap_page = NULL;
874 	error = 0;
875 	swap_count = uio_resid(uio);
876 
877 	file_end = ubc_getsize(vp);
878 	read_end = uio->uio_offset + uio_resid(uio);
879 	if (uio->uio_offset >= file_end) {
880 		/* uio starts after end of file: nothing to read */
881 		swap_count = 0;
882 	} else if (read_end > file_end) {
883 		/* uio extends beyond end of file: stop before that */
884 		swap_count -= (read_end - file_end);
885 	}
886 
887 	while (swap_count > 0) {
888 		if (my_swap_page == NULL) {
889 			my_swap_page = kalloc_data(PAGE_SIZE, Z_WAITOK | Z_ZERO);
890 			/* add an end-of-line to keep line counters happy */
891 			my_swap_page[PAGE_SIZE - 1] = '\n';
892 		}
893 		this_count = swap_count;
894 		if (this_count > PAGE_SIZE) {
895 			this_count = PAGE_SIZE;
896 		}
897 
898 		prev_resid = uio_resid(uio);
899 		error = uiomove((caddr_t) my_swap_page,
900 		    (int)this_count,
901 		    uio);
902 		if (error) {
903 			break;
904 		}
905 		swap_count -= (prev_resid - uio_resid(uio));
906 	}
907 	kfree_data(my_swap_page, PAGE_SIZE);
908 
909 	return error;
910 }
911 /*
912  * Package up an I/O request on a vnode into a uio and do it.
913  */
914 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)915 vn_rdwr(
916 	enum uio_rw rw,
917 	struct vnode *vp,
918 	caddr_t base,
919 	int len,
920 	off_t offset,
921 	enum uio_seg segflg,
922 	int ioflg,
923 	kauth_cred_t cred,
924 	int *aresid,
925 	proc_t p)
926 {
927 	int64_t resid;
928 	int result;
929 
930 	if (len < 0) {
931 		return EINVAL;
932 	}
933 
934 	result = vn_rdwr_64(rw,
935 	    vp,
936 	    (uint64_t)(uintptr_t)base,
937 	    (int64_t)len,
938 	    offset,
939 	    segflg,
940 	    ioflg,
941 	    cred,
942 	    &resid,
943 	    p);
944 
945 	/* "resid" should be bounded above by "len," which is an int */
946 	if (aresid != NULL) {
947 		*aresid = (int)resid;
948 	}
949 
950 	return result;
951 }
952 
953 
954 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)955 vn_rdwr_64(
956 	enum uio_rw rw,
957 	struct vnode *vp,
958 	uint64_t base,
959 	int64_t len,
960 	off_t offset,
961 	enum uio_seg segflg,
962 	int ioflg,
963 	kauth_cred_t cred,
964 	int64_t *aresid,
965 	proc_t p)
966 {
967 	uio_t auio;
968 	int spacetype;
969 	struct vfs_context context;
970 	int error = 0;
971 	UIO_STACKBUF(uio_buf, 1);
972 
973 	context.vc_thread = current_thread();
974 	context.vc_ucred = cred;
975 
976 	if (UIO_SEG_IS_USER_SPACE(segflg)) {
977 		spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
978 	} else {
979 		spacetype = UIO_SYSSPACE;
980 	}
981 
982 	if (len < 0) {
983 		return EINVAL;
984 	}
985 
986 	auio = uio_createwithbuffer(1, offset, spacetype, rw,
987 	    &uio_buf[0], sizeof(uio_buf));
988 	uio_addiov(auio, CAST_USER_ADDR_T(base), (user_size_t)len);
989 
990 #if CONFIG_MACF
991 	/* XXXMAC
992 	 *      IO_NOAUTH should be re-examined.
993 	 *	Likely that mediation should be performed in caller.
994 	 */
995 	if ((ioflg & IO_NOAUTH) == 0) {
996 		/* passed cred is fp->f_cred */
997 		if (rw == UIO_READ) {
998 			error = mac_vnode_check_read(&context, cred, vp);
999 		} else {
1000 			error = mac_vnode_check_write(&context, cred, vp);
1001 		}
1002 	}
1003 #endif
1004 
1005 	if (error == 0) {
1006 		if (rw == UIO_READ) {
1007 			if (vnode_isswap(vp) && ((ioflg & IO_SWAP_DISPATCH) == 0)) {
1008 				error = vn_read_swapfile(vp, auio);
1009 			} else {
1010 				error = VNOP_READ(vp, auio, ioflg, &context);
1011 			}
1012 		} else {
1013 			error = VNOP_WRITE(vp, auio, ioflg, &context);
1014 		}
1015 	}
1016 
1017 	if (aresid) {
1018 		*aresid = uio_resid(auio);
1019 		assert(*aresid <= len);
1020 	} else if (uio_resid(auio) && error == 0) {
1021 		error = EIO;
1022 	}
1023 	return error;
1024 }
1025 
1026 void
vn_offset_lock(struct fileglob * fg)1027 vn_offset_lock(struct fileglob *fg)
1028 {
1029 	lck_mtx_lock_spin(&fg->fg_lock);
1030 	while (fg->fg_lflags & FG_OFF_LOCKED) {
1031 		fg->fg_lflags |= FG_OFF_LOCKWANT;
1032 		msleep(&fg->fg_lflags, &fg->fg_lock, PVFS | PSPIN,
1033 		    "fg_offset_lock_wait", 0);
1034 	}
1035 	fg->fg_lflags |= FG_OFF_LOCKED;
1036 	lck_mtx_unlock(&fg->fg_lock);
1037 }
1038 
1039 void
vn_offset_unlock(struct fileglob * fg)1040 vn_offset_unlock(struct fileglob *fg)
1041 {
1042 	int lock_wanted = 0;
1043 
1044 	lck_mtx_lock_spin(&fg->fg_lock);
1045 	if (fg->fg_lflags & FG_OFF_LOCKWANT) {
1046 		lock_wanted = 1;
1047 	}
1048 	fg->fg_lflags &= ~(FG_OFF_LOCKED | FG_OFF_LOCKWANT);
1049 	lck_mtx_unlock(&fg->fg_lock);
1050 	if (lock_wanted) {
1051 		wakeup(&fg->fg_lflags);
1052 	}
1053 }
1054 
1055 static int
vn_read_common(vnode_t vp,struct uio * uio,int fflag,vfs_context_t ctx)1056 vn_read_common(vnode_t vp, struct uio *uio, int fflag, vfs_context_t ctx)
1057 {
1058 	int error;
1059 	int ioflag;
1060 	off_t read_offset;
1061 	user_ssize_t  read_len;
1062 	user_ssize_t  adjusted_read_len;
1063 	user_ssize_t  clippedsize;
1064 
1065 	/* Caller has already validated read_len. */
1066 	read_len = uio_resid(uio);
1067 	assert(read_len >= 0 && read_len <= INT_MAX);
1068 
1069 	adjusted_read_len = read_len;
1070 	clippedsize = 0;
1071 
1072 #if CONFIG_MACF
1073 	error = mac_vnode_check_read(ctx, vfs_context_ucred(ctx), vp);
1074 	if (error) {
1075 		return error;
1076 	}
1077 #endif
1078 
1079 	/* This signals to VNOP handlers that this read came from a file table read */
1080 	ioflag = IO_SYSCALL_DISPATCH;
1081 
1082 	if (fflag & FNONBLOCK) {
1083 		ioflag |= IO_NDELAY;
1084 	}
1085 	if ((fflag & FNOCACHE) || vnode_isnocache(vp)) {
1086 		ioflag |= IO_NOCACHE;
1087 	}
1088 	if (fflag & FENCRYPTED) {
1089 		ioflag |= IO_ENCRYPTED;
1090 	}
1091 	if (fflag & FUNENCRYPTED) {
1092 		ioflag |= IO_SKIP_ENCRYPTION;
1093 	}
1094 	if (fflag & O_EVTONLY) {
1095 		ioflag |= IO_EVTONLY;
1096 	}
1097 	if (fflag & FNORDAHEAD) {
1098 		ioflag |= IO_RAOFF;
1099 	}
1100 
1101 	read_offset = uio_offset(uio);
1102 	/* POSIX allows negative offsets for character devices. */
1103 	if ((read_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1104 		error = EINVAL;
1105 		goto error_out;
1106 	}
1107 
1108 	if (read_offset == INT64_MAX) {
1109 		/* can't read any more */
1110 		error = 0;
1111 		goto error_out;
1112 	}
1113 
1114 	/*
1115 	 * If offset + len will cause overflow, reduce the len to a value
1116 	 * (adjusted_read_len) where it won't
1117 	 */
1118 	if ((read_offset >= 0) && (INT64_MAX - read_offset) < read_len) {
1119 		/*
1120 		 * 0                                          read_offset  INT64_MAX
1121 		 * |-----------------------------------------------|----------|~~~
1122 		 *                                                 <--read_len-->
1123 		 *                                                 <-adjusted->
1124 		 */
1125 		adjusted_read_len = (user_ssize_t)(INT64_MAX - read_offset);
1126 	}
1127 
1128 	if (adjusted_read_len < read_len) {
1129 		uio_setresid(uio, adjusted_read_len);
1130 		clippedsize = read_len - adjusted_read_len;
1131 	}
1132 
1133 	if (vnode_isswap(vp) && !(IO_SKIP_ENCRYPTION & ioflag)) {
1134 		/* special case for swap files */
1135 		error = vn_read_swapfile(vp, uio);
1136 	} else {
1137 		error = VNOP_READ(vp, uio, ioflag, ctx);
1138 	}
1139 
1140 	if (clippedsize) {
1141 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1142 	}
1143 
1144 error_out:
1145 	return error;
1146 }
1147 
1148 /*
1149  * File table vnode read routine.
1150  */
1151 static int
vn_read(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1152 vn_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1153 {
1154 	vnode_t vp;
1155 	user_ssize_t read_len;
1156 	int error;
1157 	bool offset_locked;
1158 
1159 	read_len = uio_resid(uio);
1160 	if (read_len < 0 || read_len > INT_MAX) {
1161 		return EINVAL;
1162 	}
1163 
1164 	if ((flags & FOF_OFFSET) == 0) {
1165 		vn_offset_lock(fp->fp_glob);
1166 		offset_locked = true;
1167 	} else {
1168 		offset_locked = false;
1169 	}
1170 	vp = (struct vnode *)fp_get_data(fp);
1171 	if ((error = vnode_getwithref(vp))) {
1172 		if (offset_locked) {
1173 			vn_offset_unlock(fp->fp_glob);
1174 		}
1175 		return error;
1176 	}
1177 
1178 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1179 		vn_offset_unlock(fp->fp_glob);
1180 		offset_locked = false;
1181 	}
1182 
1183 	if ((flags & FOF_OFFSET) == 0) {
1184 		uio_setoffset(uio, fp->fp_glob->fg_offset);
1185 	}
1186 
1187 	error = vn_read_common(vp, uio, fp->fp_glob->fg_flag, ctx);
1188 
1189 	if ((flags & FOF_OFFSET) == 0) {
1190 		fp->fp_glob->fg_offset += read_len - uio_resid(uio);
1191 	}
1192 
1193 	if (offset_locked) {
1194 		vn_offset_unlock(fp->fp_glob);
1195 		offset_locked = false;
1196 	}
1197 
1198 	vnode_put(vp);
1199 	return error;
1200 }
1201 
1202 /*
1203  * File table vnode write routine.
1204  */
1205 static int
vn_write(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1206 vn_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1207 {
1208 	struct vnode *vp;
1209 	int error, ioflag;
1210 	off_t write_offset;
1211 	off_t write_end_offset;
1212 	user_ssize_t write_len;
1213 	user_ssize_t adjusted_write_len;
1214 	user_ssize_t clippedsize;
1215 	bool offset_locked;
1216 	proc_t p = vfs_context_proc(ctx);
1217 	rlim_t rlim_cur_fsize = p ? proc_limitgetcur(p, RLIMIT_FSIZE) : 0;
1218 
1219 	write_len = uio_resid(uio);
1220 	if (write_len < 0 || write_len > INT_MAX) {
1221 		return EINVAL;
1222 	}
1223 	adjusted_write_len = write_len;
1224 	clippedsize = 0;
1225 
1226 	if ((flags & FOF_OFFSET) == 0) {
1227 		vn_offset_lock(fp->fp_glob);
1228 		offset_locked = true;
1229 	} else {
1230 		offset_locked = false;
1231 	}
1232 
1233 	vp = (struct vnode *)fp_get_data(fp);
1234 	if ((error = vnode_getwithref(vp))) {
1235 		if (offset_locked) {
1236 			vn_offset_unlock(fp->fp_glob);
1237 		}
1238 		return error;
1239 	}
1240 
1241 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1242 		vn_offset_unlock(fp->fp_glob);
1243 		offset_locked = false;
1244 	}
1245 
1246 #if CONFIG_MACF
1247 	error = mac_vnode_check_write(ctx, vfs_context_ucred(ctx), vp);
1248 	if (error) {
1249 		(void)vnode_put(vp);
1250 		if (offset_locked) {
1251 			vn_offset_unlock(fp->fp_glob);
1252 		}
1253 		return error;
1254 	}
1255 #endif
1256 
1257 	/*
1258 	 * IO_SYSCALL_DISPATCH signals to VNOP handlers that this write came from
1259 	 * a file table write
1260 	 */
1261 	ioflag = (IO_UNIT | IO_SYSCALL_DISPATCH);
1262 
1263 	if (vp->v_type == VREG && (fp->fp_glob->fg_flag & O_APPEND)) {
1264 		ioflag |= IO_APPEND;
1265 	}
1266 	if (fp->fp_glob->fg_flag & FNONBLOCK) {
1267 		ioflag |= IO_NDELAY;
1268 	}
1269 	assert(!(fp->fp_glob->fg_flag & O_NOCTTY) || (fp->fp_glob->fg_flag & FNOCACHE));
1270 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1271 		ioflag |= IO_NOCACHE;
1272 		/*
1273 		 * O_NOCTTY is repurposed to indicate the preference for relaxed
1274 		 * alignment and size restrictions. O_NOCTTY has no meaning beyond
1275 		 * the open of the fd and is used only for ttys and the use
1276 		 * is mutually exclusive with nocache io for regular files.
1277 		 */
1278 		if (fp->fp_glob->fg_flag & O_NOCTTY || vfs_context_allow_fs_blksize_nocache_write(ctx)) {
1279 			ioflag |= IO_NOCACHE_SWRITE;
1280 		}
1281 	}
1282 	if (fp->fp_glob->fg_flag & FNODIRECT) {
1283 		ioflag |= IO_NODIRECT;
1284 	}
1285 	if (fp->fp_glob->fg_flag & FSINGLE_WRITER) {
1286 		ioflag |= IO_SINGLE_WRITER;
1287 	}
1288 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1289 		ioflag |= IO_EVTONLY;
1290 	}
1291 
1292 	/*
1293 	 * Treat synchronous mounts and O_FSYNC on the fd as equivalent.
1294 	 *
1295 	 * XXX We treat O_DSYNC as O_FSYNC for now, since we can not delay
1296 	 * XXX the non-essential metadata without some additional VFS work;
1297 	 * XXX the intent at this point is to plumb the interface for it.
1298 	 */
1299 	if ((fp->fp_glob->fg_flag & (O_FSYNC | O_DSYNC)) ||
1300 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) {
1301 		ioflag |= IO_SYNC;
1302 	}
1303 
1304 	if ((flags & FOF_OFFSET) == 0) {
1305 		write_offset = fp->fp_glob->fg_offset;
1306 		uio_setoffset(uio, write_offset);
1307 	} else {
1308 		/* for pwrite, append should  be ignored */
1309 		ioflag &= ~IO_APPEND;
1310 		write_offset = uio_offset(uio);
1311 		/* POSIX allows negative offsets for character devices. */
1312 		if ((write_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1313 			error = EINVAL;
1314 			goto error_out;
1315 		}
1316 	}
1317 
1318 	if (write_offset == INT64_MAX) {
1319 		/* writes are not possible */
1320 		error = EFBIG;
1321 		goto error_out;
1322 	}
1323 
1324 	/*
1325 	 * write_len is the original write length that was requested.
1326 	 * We may however need to reduce that becasue of two reasons
1327 	 *
1328 	 * 1) If write_offset + write_len will exceed OFF_T_MAX (i.e. INT64_MAX)
1329 	 *    and/or
1330 	 * 2) If write_offset + write_len will exceed the administrative
1331 	 *    limit for the maximum file size.
1332 	 *
1333 	 * In both cases the write will be denied if we can't write even a single
1334 	 * byte otherwise it will be "clipped" (i.e. a short write).
1335 	 */
1336 
1337 	/*
1338 	 * If offset + len will cause overflow, reduce the len
1339 	 * to a value (adjusted_write_len) where it won't
1340 	 */
1341 	if ((write_offset >= 0) && (INT64_MAX - write_offset) < write_len) {
1342 		/*
1343 		 * 0                                          write_offset  INT64_MAX
1344 		 * |-----------------------------------------------|----------|~~~
1345 		 *                                                 <--write_len-->
1346 		 *                                                 <-adjusted->
1347 		 */
1348 		adjusted_write_len = (user_ssize_t)(INT64_MAX - write_offset);
1349 	}
1350 
1351 	/* write_end_offset will always be [0, INT64_MAX] */
1352 	write_end_offset = write_offset + adjusted_write_len;
1353 
1354 	if (p && (vp->v_type == VREG) &&
1355 	    (rlim_cur_fsize != RLIM_INFINITY) &&
1356 	    (rlim_cur_fsize <= INT64_MAX) &&
1357 	    (write_end_offset > (off_t)rlim_cur_fsize)) {
1358 		/*
1359 		 * If the requested residual would cause us to go past the
1360 		 * administrative limit, then we need to adjust the residual
1361 		 * down to cause fewer bytes than requested to be written.  If
1362 		 * we can't do that (e.g. the residual is already 1 byte),
1363 		 * then we fail the write with EFBIG.
1364 		 */
1365 		if (write_offset >= (off_t)rlim_cur_fsize) {
1366 			/*
1367 			 * 0                  rlim_fsize  write_offset  write_end  INT64_MAX
1368 			 * |------------------------|----------|-------------|--------|
1369 			 *                                     <--write_len-->
1370 			 *
1371 			 *                write not permitted
1372 			 */
1373 			psignal(p, SIGXFSZ);
1374 			error = EFBIG;
1375 			goto error_out;
1376 		}
1377 
1378 		/*
1379 		 * 0                  write_offset rlim_fsize  write_end  INT64_MAX
1380 		 * |------------------------|-----------|---------|------------|
1381 		 *                          <------write_len------>
1382 		 *                          <-adjusted-->
1383 		 */
1384 		adjusted_write_len =  (user_ssize_t)((off_t)rlim_cur_fsize - write_offset);
1385 		assert((adjusted_write_len > 0) && (adjusted_write_len < write_len));
1386 	}
1387 
1388 	if (adjusted_write_len < write_len) {
1389 		uio_setresid(uio, adjusted_write_len);
1390 		clippedsize = write_len - adjusted_write_len;
1391 	}
1392 
1393 	error = VNOP_WRITE(vp, uio, ioflag, ctx);
1394 
1395 	/*
1396 	 * If we had to reduce the size of write requested either because
1397 	 * of rlimit or because it would have exceeded
1398 	 * maximum file size, we have to add that back to the residual so
1399 	 * it correctly reflects what we did in this function.
1400 	 */
1401 	if (clippedsize) {
1402 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1403 	}
1404 
1405 	if ((flags & FOF_OFFSET) == 0) {
1406 		if (ioflag & IO_APPEND) {
1407 			fp->fp_glob->fg_offset = uio_offset(uio);
1408 		} else {
1409 			fp->fp_glob->fg_offset += (write_len - uio_resid(uio));
1410 		}
1411 		if (offset_locked) {
1412 			vn_offset_unlock(fp->fp_glob);
1413 			offset_locked = false;
1414 		}
1415 	}
1416 
1417 	/*
1418 	 * Set the credentials on successful writes
1419 	 */
1420 	if ((error == 0) && (vp->v_tag == VT_NFS) && (UBCINFOEXISTS(vp))) {
1421 		ubc_setcred(vp, vfs_context_ucred(ctx));
1422 	}
1423 
1424 #if CONFIG_FILE_LEASES
1425 	/*
1426 	 * On success, break the parent dir lease as the file's attributes (size
1427 	 * and/or mtime) have changed. Best attempt to break lease, just drop the
1428 	 * the error upon failure as there is no point to return error when the
1429 	 * write has completed successfully.
1430 	 */
1431 	if (__probable(error == 0)) {
1432 		vnode_breakdirlease(vp, true, O_WRONLY);
1433 	}
1434 #endif /* CONFIG_FILE_LEASES */
1435 
1436 	(void)vnode_put(vp);
1437 	return error;
1438 
1439 error_out:
1440 	if (offset_locked) {
1441 		vn_offset_unlock(fp->fp_glob);
1442 	}
1443 	(void)vnode_put(vp);
1444 	return error;
1445 }
1446 
1447 /*
1448  * File table vnode stat routine.
1449  *
1450  * Returns:	0			Success
1451  *		EBADF
1452  *		ENOMEM
1453  *	vnode_getattr:???
1454  */
1455 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)1456 vn_stat_noauth(struct vnode *vp, void *sbptr, kauth_filesec_t *xsec, int isstat64,
1457     int needsrealdev, vfs_context_t ctx, struct ucred *file_cred)
1458 {
1459 	struct vnode_attr va;
1460 	int error;
1461 	u_short mode;
1462 	kauth_filesec_t fsec;
1463 	struct stat *sb = (struct stat *)0;     /* warning avoidance ; protected by isstat64 */
1464 	struct stat64 * sb64 = (struct stat64 *)0;  /* warning avoidance ; protected by isstat64 */
1465 
1466 	if (isstat64 != 0) {
1467 		sb64 = (struct stat64 *)sbptr;
1468 	} else {
1469 		sb = (struct stat *)sbptr;
1470 	}
1471 	memset(&va, 0, sizeof(va));
1472 	VATTR_INIT(&va);
1473 	VATTR_WANTED(&va, va_fsid);
1474 	VATTR_WANTED(&va, va_fileid);
1475 	VATTR_WANTED(&va, va_mode);
1476 	VATTR_WANTED(&va, va_type);
1477 	VATTR_WANTED(&va, va_nlink);
1478 	VATTR_WANTED(&va, va_uid);
1479 	VATTR_WANTED(&va, va_gid);
1480 	VATTR_WANTED(&va, va_rdev);
1481 	VATTR_WANTED(&va, va_data_size);
1482 	VATTR_WANTED(&va, va_access_time);
1483 	VATTR_WANTED(&va, va_modify_time);
1484 	VATTR_WANTED(&va, va_change_time);
1485 	VATTR_WANTED(&va, va_create_time);
1486 	VATTR_WANTED(&va, va_flags);
1487 	VATTR_WANTED(&va, va_gen);
1488 	VATTR_WANTED(&va, va_iosize);
1489 	/* lower layers will synthesise va_total_alloc from va_data_size if required */
1490 	VATTR_WANTED(&va, va_total_alloc);
1491 	if (xsec != NULL) {
1492 		VATTR_WANTED(&va, va_uuuid);
1493 		VATTR_WANTED(&va, va_guuid);
1494 		VATTR_WANTED(&va, va_acl);
1495 	}
1496 	if (needsrealdev) {
1497 		va.va_vaflags = VA_REALFSID;
1498 	}
1499 	error = vnode_getattr(vp, &va, ctx);
1500 	if (error) {
1501 		goto out;
1502 	}
1503 #if CONFIG_MACF
1504 	/*
1505 	 * Give MAC polices a chance to reject or filter the attributes
1506 	 * returned by the filesystem.  Note that MAC policies are consulted
1507 	 * *after* calling the filesystem because filesystems can return more
1508 	 * attributes than were requested so policies wouldn't be authoritative
1509 	 * is consulted beforehand.  This also gives policies an opportunity
1510 	 * to change the values of attributes retrieved.
1511 	 */
1512 	error = mac_vnode_check_getattr(ctx, file_cred, vp, &va);
1513 	if (error) {
1514 		goto out;
1515 	}
1516 #endif
1517 	/*
1518 	 * Copy from vattr table
1519 	 */
1520 	if (isstat64 != 0) {
1521 		sb64->st_dev = va.va_fsid;
1522 		sb64->st_ino = (ino64_t)va.va_fileid;
1523 	} else {
1524 		sb->st_dev = va.va_fsid;
1525 		sb->st_ino = (ino_t)va.va_fileid;
1526 	}
1527 	mode = va.va_mode;
1528 	switch (vp->v_type) {
1529 	case VREG:
1530 		mode |= S_IFREG;
1531 		break;
1532 	case VDIR:
1533 		mode |= S_IFDIR;
1534 		break;
1535 	case VBLK:
1536 		mode |= S_IFBLK;
1537 		break;
1538 	case VCHR:
1539 		mode |= S_IFCHR;
1540 		break;
1541 	case VLNK:
1542 		mode |= S_IFLNK;
1543 		break;
1544 	case VSOCK:
1545 		mode |= S_IFSOCK;
1546 		break;
1547 	case VFIFO:
1548 		mode |= S_IFIFO;
1549 		break;
1550 	default:
1551 		error = EBADF;
1552 		goto out;
1553 	}
1554 	;
1555 	if (isstat64 != 0) {
1556 		sb64->st_mode = mode;
1557 		sb64->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1558 		sb64->st_uid = va.va_uid;
1559 		sb64->st_gid = va.va_gid;
1560 		sb64->st_rdev = va.va_rdev;
1561 		sb64->st_size = va.va_data_size;
1562 		sb64->st_atimespec = va.va_access_time;
1563 		sb64->st_mtimespec = va.va_modify_time;
1564 		sb64->st_ctimespec = va.va_change_time;
1565 		if (VATTR_IS_SUPPORTED(&va, va_create_time)) {
1566 			sb64->st_birthtimespec =  va.va_create_time;
1567 		} else {
1568 			sb64->st_birthtimespec.tv_sec = sb64->st_birthtimespec.tv_nsec = 0;
1569 		}
1570 		sb64->st_blksize = va.va_iosize;
1571 		sb64->st_flags = va.va_flags;
1572 		sb64->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1573 	} else {
1574 		sb->st_mode = mode;
1575 		sb->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1576 		sb->st_uid = va.va_uid;
1577 		sb->st_gid = va.va_gid;
1578 		sb->st_rdev = va.va_rdev;
1579 		sb->st_size = va.va_data_size;
1580 		sb->st_atimespec = va.va_access_time;
1581 		sb->st_mtimespec = va.va_modify_time;
1582 		sb->st_ctimespec = va.va_change_time;
1583 		sb->st_blksize = va.va_iosize;
1584 		sb->st_flags = va.va_flags;
1585 		sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1586 	}
1587 
1588 	/* if we're interested in extended security data and we got an ACL */
1589 	if (xsec != NULL) {
1590 		if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
1591 		    !VATTR_IS_SUPPORTED(&va, va_uuuid) &&
1592 		    !VATTR_IS_SUPPORTED(&va, va_guuid)) {
1593 			*xsec = KAUTH_FILESEC_NONE;
1594 		} else {
1595 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1596 				fsec = kauth_filesec_alloc(va.va_acl->acl_entrycount);
1597 			} else {
1598 				fsec = kauth_filesec_alloc(0);
1599 			}
1600 			if (fsec == NULL) {
1601 				error = ENOMEM;
1602 				goto out;
1603 			}
1604 			fsec->fsec_magic = KAUTH_FILESEC_MAGIC;
1605 			if (VATTR_IS_SUPPORTED(&va, va_uuuid)) {
1606 				fsec->fsec_owner = va.va_uuuid;
1607 			} else {
1608 				fsec->fsec_owner = kauth_null_guid;
1609 			}
1610 			if (VATTR_IS_SUPPORTED(&va, va_guuid)) {
1611 				fsec->fsec_group = va.va_guuid;
1612 			} else {
1613 				fsec->fsec_group = kauth_null_guid;
1614 			}
1615 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1616 				__nochk_bcopy(va.va_acl, &(fsec->fsec_acl), KAUTH_ACL_COPYSIZE(va.va_acl));
1617 			} else {
1618 				fsec->fsec_acl.acl_entrycount = KAUTH_FILESEC_NOACL;
1619 			}
1620 			*xsec = fsec;
1621 		}
1622 	}
1623 
1624 	/* Do not give the generation number out to unpriviledged users */
1625 	if (va.va_gen && !vfs_context_issuser(ctx)) {
1626 		if (isstat64 != 0) {
1627 			sb64->st_gen = 0;
1628 		} else {
1629 			sb->st_gen = 0;
1630 		}
1631 	} else {
1632 		if (isstat64 != 0) {
1633 			sb64->st_gen = va.va_gen;
1634 		} else {
1635 			sb->st_gen = va.va_gen;
1636 		}
1637 	}
1638 
1639 	error = 0;
1640 out:
1641 	if (VATTR_IS_SUPPORTED(&va, va_acl) && va.va_acl != NULL) {
1642 		kauth_acl_free(va.va_acl);
1643 	}
1644 	return error;
1645 }
1646 
1647 int
vn_stat(struct vnode * vp,void * sb,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx)1648 vn_stat(struct vnode *vp, void *sb, kauth_filesec_t *xsec, int isstat64, int needsrealdev, vfs_context_t ctx)
1649 {
1650 	int error;
1651 
1652 #if CONFIG_MACF
1653 	error = mac_vnode_check_stat(ctx, NOCRED, vp);
1654 	if (error) {
1655 		return error;
1656 	}
1657 #endif
1658 
1659 	/* authorize */
1660 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_ATTRIBUTES | KAUTH_VNODE_READ_SECURITY, ctx)) != 0) {
1661 		return error;
1662 	}
1663 
1664 	/* actual stat */
1665 	return vn_stat_noauth(vp, sb, xsec, isstat64, needsrealdev, ctx, NOCRED);
1666 }
1667 
1668 
1669 /*
1670  * File table vnode ioctl routine.
1671  */
1672 static int
vn_ioctl(struct fileproc * fp,u_long com,caddr_t data,vfs_context_t ctx)1673 vn_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
1674 {
1675 	struct vnode *vp = (struct vnode *)fp_get_data(fp);
1676 	off_t file_size;
1677 	int error;
1678 	struct vnode *ttyvp;
1679 
1680 	if ((error = vnode_getwithref(vp))) {
1681 		return error;
1682 	}
1683 
1684 #if CONFIG_MACF
1685 	error = mac_vnode_check_ioctl(ctx, vp, com);
1686 	if (error) {
1687 		goto out;
1688 	}
1689 #endif
1690 
1691 	switch (vp->v_type) {
1692 	case VREG:
1693 	case VDIR:
1694 		if (com == FIONREAD) {
1695 			off_t temp_nbytes;
1696 			if ((error = vnode_size(vp, &file_size, ctx)) != 0) {
1697 				goto out;
1698 			}
1699 			temp_nbytes = file_size - fp->fp_glob->fg_offset;
1700 			if (temp_nbytes > INT_MAX) {
1701 				*(int *)data = INT_MAX;
1702 			} else if (temp_nbytes < 0) {
1703 				*(int *)data = 0;
1704 			} else {
1705 				*(int *)data = (int)temp_nbytes;
1706 			}
1707 			goto out;
1708 		}
1709 		if (com == FIONBIO || com == FIOASYNC) {        /* XXX */
1710 			goto out;
1711 		}
1712 		OS_FALLTHROUGH;
1713 
1714 	default:
1715 		error = ENOTTY;
1716 		goto out;
1717 
1718 	case VFIFO:
1719 	case VCHR:
1720 	case VBLK:
1721 
1722 		if (com == TIOCREVOKE || com == TIOCREVOKECLEAR) {
1723 			error = ENOTTY;
1724 			goto out;
1725 		}
1726 
1727 		/* Should not be able to set block size from user space */
1728 		if (com == DKIOCSETBLOCKSIZE) {
1729 			error = EPERM;
1730 			goto out;
1731 		}
1732 
1733 		/* Should not be able to check if filesystem is authenticated from user space */
1734 		if (com == FSIOC_AUTH_FS) {
1735 			error = ENOTTY;
1736 			goto out;
1737 		}
1738 
1739 		if (com == FIODTYPE) {
1740 			if (vp->v_type == VBLK) {
1741 				if (major(vp->v_rdev) >= nblkdev) {
1742 					error = ENXIO;
1743 					goto out;
1744 				}
1745 				*(int *)data = bdevsw[major(vp->v_rdev)].d_type;
1746 			} else if (vp->v_type == VCHR) {
1747 				if (major(vp->v_rdev) >= nchrdev) {
1748 					error = ENXIO;
1749 					goto out;
1750 				}
1751 				*(int *)data = cdevsw[major(vp->v_rdev)].d_type;
1752 			} else {
1753 				error = ENOTTY;
1754 				goto out;
1755 			}
1756 			goto out;
1757 		}
1758 		error = VNOP_IOCTL(vp, com, data, fp->fp_glob->fg_flag, ctx);
1759 
1760 		if (error == 0 && com == TIOCSCTTY) {
1761 			struct session *sessp;
1762 			struct pgrp *pg;
1763 
1764 			pg = proc_pgrp(vfs_context_proc(ctx), &sessp);
1765 
1766 			session_lock(sessp);
1767 			ttyvp = sessp->s_ttyvp;
1768 			sessp->s_ttyvp = vp;
1769 			sessp->s_ttyvid = vnode_vid(vp);
1770 			session_unlock(sessp);
1771 
1772 			pgrp_rele(pg);
1773 		}
1774 	}
1775 out:
1776 	(void)vnode_put(vp);
1777 	return error;
1778 }
1779 
1780 /*
1781  * File table vnode select routine.
1782  */
1783 static int
vn_select(struct fileproc * fp,int which,void * wql,__unused vfs_context_t ctx)1784 vn_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx)
1785 {
1786 	int error;
1787 	struct vnode * vp = (struct vnode *)fp_get_data(fp);
1788 	struct vfs_context context;
1789 
1790 	if ((error = vnode_getwithref(vp)) == 0) {
1791 		context.vc_thread = current_thread();
1792 		context.vc_ucred = fp->fp_glob->fg_cred;
1793 
1794 #if CONFIG_MACF
1795 		/*
1796 		 * XXX We should use a per thread credential here; minimally,
1797 		 * XXX the process credential should have a persistent
1798 		 * XXX reference on it before being passed in here.
1799 		 */
1800 		error = mac_vnode_check_select(ctx, vp, which);
1801 		if (error == 0)
1802 #endif
1803 		error = VNOP_SELECT(vp, which, fp->fp_glob->fg_flag, wql, ctx);
1804 
1805 		(void)vnode_put(vp);
1806 	}
1807 	return error;
1808 }
1809 
1810 /*
1811  * File table vnode close routine.
1812  */
1813 static int
vn_closefile(struct fileglob * fg,vfs_context_t ctx)1814 vn_closefile(struct fileglob *fg, vfs_context_t ctx)
1815 {
1816 	struct vnode *vp = fg_get_data(fg);
1817 	int error;
1818 
1819 	if ((error = vnode_getwithref(vp)) == 0) {
1820 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE &&
1821 		    ((fg->fg_flag & FWASLOCKED) != 0 ||
1822 		    (fg->fg_lflags & FG_HAS_OFDLOCK) != 0)) {
1823 			struct flock lf = {
1824 				.l_whence = SEEK_SET,
1825 				.l_start = 0,
1826 				.l_len = 0,
1827 				.l_type = F_UNLCK
1828 			};
1829 
1830 			if ((fg->fg_flag & FWASLOCKED) != 0) {
1831 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1832 				    F_UNLCK, &lf, F_FLOCK, ctx, NULL);
1833 			}
1834 
1835 			if ((fg->fg_lflags & FG_HAS_OFDLOCK) != 0) {
1836 				(void) VNOP_ADVLOCK(vp, ofd_to_id(fg),
1837 				    F_UNLCK, &lf, F_OFD_LOCK, ctx, NULL);
1838 			}
1839 		}
1840 #if CONFIG_FILE_LEASES
1841 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE && !LIST_EMPTY(&vp->v_leases)) {
1842 			/* Expected open count doesn't matter for release. */
1843 			(void)vnode_setlease(vp, fg, F_UNLCK, 0, ctx);
1844 		}
1845 #endif
1846 		error = vn_close(vp, fg->fg_flag, ctx);
1847 		(void) vnode_put(vp);
1848 	}
1849 	return error;
1850 }
1851 
1852 /*
1853  * Returns:	0			Success
1854  *	VNOP_PATHCONF:???
1855  */
1856 int
vn_pathconf(vnode_t vp,int name,int32_t * retval,vfs_context_t ctx)1857 vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
1858 {
1859 	int     error = 0;
1860 	struct vfs_attr vfa;
1861 
1862 	switch (name) {
1863 	case _PC_EXTENDED_SECURITY_NP:
1864 		*retval = vfs_extendedsecurity(vnode_mount(vp)) ? 1 : 0;
1865 		break;
1866 	case _PC_AUTH_OPAQUE_NP:
1867 		*retval = vfs_authopaque(vnode_mount(vp));
1868 		break;
1869 	case _PC_2_SYMLINKS:
1870 		*retval = 1;    /* XXX NOTSUP on MSDOS, etc. */
1871 		break;
1872 	case _PC_ALLOC_SIZE_MIN:
1873 		*retval = 1;    /* XXX lie: 1 byte */
1874 		break;
1875 	case _PC_ASYNC_IO:      /* unistd.h: _POSIX_ASYNCHRONUS_IO */
1876 		*retval = 1;    /* [AIO] option is supported */
1877 		break;
1878 	case _PC_PRIO_IO:       /* unistd.h: _POSIX_PRIORITIZED_IO */
1879 		*retval = 0;    /* [PIO] option is not supported */
1880 		break;
1881 	case _PC_REC_INCR_XFER_SIZE:
1882 		*retval = 4096; /* XXX go from MIN to MAX 4K at a time */
1883 		break;
1884 	case _PC_REC_MIN_XFER_SIZE:
1885 		*retval = 4096; /* XXX recommend 4K minimum reads/writes */
1886 		break;
1887 	case _PC_REC_MAX_XFER_SIZE:
1888 		*retval = 65536; /* XXX recommend 64K maximum reads/writes */
1889 		break;
1890 	case _PC_REC_XFER_ALIGN:
1891 		*retval = 4096; /* XXX recommend page aligned buffers */
1892 		break;
1893 	case _PC_SYMLINK_MAX:
1894 		*retval = 255;  /* Minimum acceptable POSIX value */
1895 		break;
1896 	case _PC_SYNC_IO:       /* unistd.h: _POSIX_SYNCHRONIZED_IO */
1897 		*retval = 0;    /* [SIO] option is not supported */
1898 		break;
1899 	case _PC_XATTR_SIZE_BITS:
1900 		/* The number of bits used to store maximum extended
1901 		 * attribute size in bytes.  For example, if the maximum
1902 		 * attribute size supported by a file system is 128K, the
1903 		 * value returned will be 18.  However a value 18 can mean
1904 		 * that the maximum attribute size can be anywhere from
1905 		 * (256KB - 1) to 128KB.  As a special case, the resource
1906 		 * fork can have much larger size, and some file system
1907 		 * specific extended attributes can have smaller and preset
1908 		 * size; for example, Finder Info is always 32 bytes.
1909 		 */
1910 		memset(&vfa, 0, sizeof(vfa));
1911 		VFSATTR_INIT(&vfa);
1912 		VFSATTR_WANTED(&vfa, f_capabilities);
1913 		if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
1914 		    (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
1915 		    (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
1916 		    (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
1917 			/* Supports native extended attributes */
1918 			error = VNOP_PATHCONF(vp, name, retval, ctx);
1919 		} else {
1920 			/* Number of bits used to represent the maximum size of
1921 			 * extended attribute stored in an Apple Double file.
1922 			 */
1923 			*retval = AD_XATTR_SIZE_BITS;
1924 		}
1925 		break;
1926 	default:
1927 		error = VNOP_PATHCONF(vp, name, retval, ctx);
1928 		break;
1929 	}
1930 
1931 	return error;
1932 }
1933 
1934 static int
vn_kqfilter(struct fileproc * fp,struct knote * kn,struct kevent_qos_s * kev)1935 vn_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
1936 {
1937 	vfs_context_t ctx = vfs_context_current();
1938 	struct vnode *vp;
1939 	int error = 0;
1940 	int result = 0;
1941 
1942 	vp = (struct vnode *)fp_get_data(fp);
1943 
1944 	/*
1945 	 * Don't attach a knote to a dead vnode.
1946 	 */
1947 	if ((error = vget_internal(vp, 0, VNODE_NODEAD)) == 0) {
1948 		switch (kn->kn_filter) {
1949 		case EVFILT_READ:
1950 		case EVFILT_WRITE:
1951 			if (vnode_isfifo(vp)) {
1952 				/* We'll only watch FIFOs that use our fifofs */
1953 				if (!(vp->v_fifoinfo && vp->v_fifoinfo->fi_readsock)) {
1954 					error = ENOTSUP;
1955 				}
1956 			} else if (!vnode_isreg(vp)) {
1957 				if (vnode_ischr(vp)) {
1958 					result = spec_kqfilter(vp, kn, kev);
1959 					if ((kn->kn_flags & EV_ERROR) == 0) {
1960 						/* claimed by a special device */
1961 						vnode_put(vp);
1962 						return result;
1963 					}
1964 				}
1965 				error = EINVAL;
1966 			}
1967 			break;
1968 		case EVFILT_VNODE:
1969 			break;
1970 		default:
1971 			error = EINVAL;
1972 		}
1973 
1974 		if (error == 0) {
1975 #if CONFIG_MACF
1976 			error = mac_vnode_check_kqfilter(ctx, fp->fp_glob->fg_cred, kn, vp);
1977 			if (error) {
1978 				vnode_put(vp);
1979 				goto out;
1980 			}
1981 #endif
1982 
1983 			kn->kn_filtid = EVFILTID_VN;
1984 			knote_kn_hook_set_raw(kn, (void *)vp);
1985 			vnode_hold(vp);
1986 
1987 			vnode_lock(vp);
1988 			KNOTE_ATTACH(&vp->v_knotes, kn);
1989 			result = filt_vnode_common(kn, NULL, vp, 0);
1990 			vnode_unlock(vp);
1991 
1992 			/*
1993 			 * Ask the filesystem to provide remove notifications,
1994 			 * but ignore failure
1995 			 */
1996 			VNOP_MONITOR(vp, 0, VNODE_MONITOR_BEGIN, (void*) kn, ctx);
1997 		}
1998 
1999 		vnode_put(vp);
2000 	}
2001 
2002 out:
2003 	if (error) {
2004 		knote_set_error(kn, error);
2005 	}
2006 
2007 	return result;
2008 }
2009 
2010 static void
filt_vndetach(struct knote * kn)2011 filt_vndetach(struct knote *kn)
2012 {
2013 	vfs_context_t ctx = vfs_context_current();
2014 	struct vnode *vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2015 	uint32_t vid = vnode_vid(vp);
2016 	if (vnode_getwithvid(vp, vid)) {
2017 		vnode_drop(vp);
2018 		return;
2019 	}
2020 	vnode_drop(vp);
2021 
2022 	vnode_lock(vp);
2023 	KNOTE_DETACH(&vp->v_knotes, kn);
2024 	vnode_unlock(vp);
2025 
2026 	/*
2027 	 * Tell a (generally networked) filesystem that we're no longer watching
2028 	 * If the FS wants to track contexts, it should still be using the one from
2029 	 * the VNODE_MONITOR_BEGIN.
2030 	 */
2031 	VNOP_MONITOR(vp, 0, VNODE_MONITOR_END, (void*)kn, ctx);
2032 	vnode_put(vp);
2033 }
2034 
2035 
2036 /*
2037  * Used for EVFILT_READ
2038  *
2039  * Takes only VFIFO or VREG. vnode is locked.  We handle the "poll" case
2040  * differently than the regular case for VREG files.  If not in poll(),
2041  * then we need to know current fileproc offset for VREG.
2042  */
2043 static int64_t
vnode_readable_data_count(vnode_t vp,off_t current_offset,int ispoll)2044 vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
2045 {
2046 	if (vnode_isfifo(vp)) {
2047 #if FIFO
2048 		int cnt;
2049 		int err = fifo_charcount(vp, &cnt);
2050 		if (err == 0) {
2051 			return (int64_t)cnt;
2052 		} else
2053 #endif
2054 		{
2055 			return 0;
2056 		}
2057 	} else if (vnode_isreg(vp)) {
2058 		if (ispoll) {
2059 			return 1;
2060 		}
2061 
2062 		off_t amount;
2063 		amount = vp->v_un.vu_ubcinfo->ui_size - current_offset;
2064 		if (amount > INT64_MAX) {
2065 			return INT64_MAX;
2066 		} else if (amount < INT64_MIN) {
2067 			return INT64_MIN;
2068 		} else {
2069 			return (int64_t)amount;
2070 		}
2071 	} else {
2072 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2073 		return 0;
2074 	}
2075 }
2076 
2077 /*
2078  * Used for EVFILT_WRITE.
2079  *
2080  * For regular vnodes, we can always write (1).  For named pipes,
2081  * see how much space there is in the buffer.  Nothing else is covered.
2082  */
2083 static intptr_t
vnode_writable_space_count(vnode_t vp)2084 vnode_writable_space_count(vnode_t vp)
2085 {
2086 	if (vnode_isfifo(vp)) {
2087 #if FIFO
2088 		long spc;
2089 		int err = fifo_freespace(vp, &spc);
2090 		if (err == 0) {
2091 			return (intptr_t)spc;
2092 		} else
2093 #endif
2094 		{
2095 			return (intptr_t)0;
2096 		}
2097 	} else if (vnode_isreg(vp)) {
2098 		return (intptr_t)1;
2099 	} else {
2100 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2101 		return 0;
2102 	}
2103 }
2104 
2105 /*
2106  * Determine whether this knote should be active
2107  *
2108  * This is kind of subtle.
2109  *      --First, notice if the vnode has been revoked: in so, override hint
2110  *      --EVFILT_READ knotes are checked no matter what the hint is
2111  *      --Other knotes activate based on hint.
2112  *      --If hint is revoke, set special flags and activate
2113  */
2114 static int
filt_vnode_common(struct knote * kn,struct kevent_qos_s * kev,vnode_t vp,long hint)2115 filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev, vnode_t vp, long hint)
2116 {
2117 	int activate = 0;
2118 	int64_t data = 0;
2119 
2120 	lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
2121 
2122 	/* Special handling for vnodes that are in recycle or already gone */
2123 	if (NOTE_REVOKE == hint) {
2124 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2125 		activate = 1;
2126 
2127 		if ((kn->kn_filter == EVFILT_VNODE) && (kn->kn_sfflags & NOTE_REVOKE)) {
2128 			kn->kn_fflags |= NOTE_REVOKE;
2129 		}
2130 	} else {
2131 		switch (kn->kn_filter) {
2132 		case EVFILT_READ:
2133 			data = vnode_readable_data_count(vp, kn->kn_fp->fp_glob->fg_offset, (kn->kn_flags & EV_POLL));
2134 			activate = (data != 0);
2135 			break;
2136 		case EVFILT_WRITE:
2137 			data = vnode_writable_space_count(vp);
2138 			activate = (data != 0);
2139 			break;
2140 		case EVFILT_VNODE:
2141 			/* Check events this note matches against the hint */
2142 			if (kn->kn_sfflags & hint) {
2143 				kn->kn_fflags |= (uint32_t)hint; /* Set which event occurred */
2144 			}
2145 			activate = (kn->kn_fflags != 0);
2146 			break;
2147 		default:
2148 			panic("Invalid knote filter on a vnode!");
2149 		}
2150 	}
2151 
2152 	if (kev && activate) {
2153 		knote_fill_kevent(kn, kev, data);
2154 	}
2155 
2156 	return activate;
2157 }
2158 
2159 static int
filt_vnode(struct knote * kn,long hint)2160 filt_vnode(struct knote *kn, long hint)
2161 {
2162 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2163 
2164 	return filt_vnode_common(kn, NULL, vp, hint);
2165 }
2166 
2167 static int
filt_vntouch(struct knote * kn,struct kevent_qos_s * kev)2168 filt_vntouch(struct knote *kn, struct kevent_qos_s *kev)
2169 {
2170 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2171 	uint32_t vid = vnode_vid(vp);
2172 	int activate;
2173 	int hint = 0;
2174 
2175 	vnode_lock(vp);
2176 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2177 		/* is recycled */
2178 		hint = NOTE_REVOKE;
2179 	}
2180 
2181 	/* accept new input fflags mask */
2182 	kn->kn_sfflags = kev->fflags;
2183 
2184 	activate = filt_vnode_common(kn, NULL, vp, hint);
2185 
2186 	if (hint == 0) {
2187 		vnode_put_locked(vp);
2188 	}
2189 	vnode_unlock(vp);
2190 
2191 	return activate;
2192 }
2193 
2194 static int
filt_vnprocess(struct knote * kn,struct kevent_qos_s * kev)2195 filt_vnprocess(struct knote *kn, struct kevent_qos_s *kev)
2196 {
2197 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2198 	uint32_t vid = vnode_vid(vp);
2199 	int activate;
2200 	int hint = 0;
2201 
2202 	vnode_lock(vp);
2203 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2204 		/* Is recycled */
2205 		hint = NOTE_REVOKE;
2206 	}
2207 	activate = filt_vnode_common(kn, kev, vp, hint);
2208 
2209 	/* Definitely need to unlock, may need to put */
2210 	if (hint == 0) {
2211 		vnode_put_locked(vp);
2212 	}
2213 	vnode_unlock(vp);
2214 
2215 	return activate;
2216 }
2217 
2218 struct vniodesc {
2219 	vnode_t         vnio_vnode;     /* associated vnode */
2220 	int             vnio_fflags;    /* cached fileglob flags */
2221 };
2222 
2223 errno_t
vnio_openfd(int fd,vniodesc_t * vniop)2224 vnio_openfd(int fd, vniodesc_t *vniop)
2225 {
2226 	proc_t p = current_proc();
2227 	struct fileproc *fp = NULL;
2228 	vniodesc_t vnio;
2229 	vnode_t vp;
2230 	int error;
2231 	bool need_drop = false;
2232 
2233 	vnio = kalloc_type(struct vniodesc, Z_WAITOK);
2234 
2235 	error = fp_getfvp(p, fd, &fp, &vp);
2236 	if (error) {
2237 		goto out;
2238 	}
2239 	need_drop = true;
2240 
2241 	if ((fp->fp_glob->fg_flag & (FREAD | FWRITE)) == 0) {
2242 		error = EBADF;
2243 		goto out;
2244 	}
2245 
2246 	if (vnode_isswap(vp)) {
2247 		error = EPERM;
2248 		goto out;
2249 	}
2250 
2251 	if (vp->v_type != VREG) {
2252 		error = EFTYPE;
2253 		goto out;
2254 	}
2255 
2256 	error = vnode_getwithref(vp);
2257 	if (error) {
2258 		goto out;
2259 	}
2260 
2261 	vnio->vnio_vnode = vp;
2262 	vnio->vnio_fflags = fp->fp_glob->fg_flag;
2263 
2264 	(void)fp_drop(p, fd, fp, 0);
2265 	need_drop = false;
2266 
2267 	error = vnode_ref_ext(vp, vnio->vnio_fflags, 0);
2268 	if (error == 0) {
2269 		*vniop = vnio;
2270 		vnio = NULL;
2271 	}
2272 
2273 	vnode_put(vp);
2274 
2275 out:
2276 	if (need_drop) {
2277 		fp_drop(p, fd, fp, 0);
2278 	}
2279 	if (vnio != NULL) {
2280 		kfree_type(struct vniodesc, vnio);
2281 	}
2282 	return error;
2283 }
2284 
2285 errno_t
vnio_close(vniodesc_t vnio)2286 vnio_close(vniodesc_t vnio)
2287 {
2288 	int error;
2289 
2290 	/*
2291 	 * The vniodesc is always destroyed, because the close
2292 	 * always "succeeds".  We just return whatever error
2293 	 * might have been encountered while doing so.
2294 	 */
2295 
2296 	error = vnode_getwithref(vnio->vnio_vnode);
2297 	if (error == 0) {
2298 		error = vnode_close(vnio->vnio_vnode, vnio->vnio_fflags, NULL);
2299 	}
2300 
2301 	kfree_type(struct vniodesc, vnio);
2302 
2303 	return error;
2304 }
2305 
2306 errno_t
vnio_read(vniodesc_t vnio,uio_t uio)2307 vnio_read(vniodesc_t vnio, uio_t uio)
2308 {
2309 	vnode_t vp = vnio->vnio_vnode;
2310 	user_ssize_t read_len;
2311 	int error;
2312 
2313 	read_len = uio_resid(uio);
2314 	if (read_len < 0 || read_len > INT_MAX) {
2315 		return EINVAL;
2316 	}
2317 
2318 	error = vnode_getwithref(vp);
2319 	if (error == 0) {
2320 		error = vn_read_common(vp, uio, vnio->vnio_fflags,
2321 		    vfs_context_current());
2322 		vnode_put(vp);
2323 	}
2324 
2325 	return error;
2326 }
2327 
2328 vnode_t
vnio_vnode(vniodesc_t vnio)2329 vnio_vnode(vniodesc_t vnio)
2330 {
2331 	return vnio->vnio_vnode;
2332 }
2333 
2334 int
vnode_isauthfs(vnode_t vp)2335 vnode_isauthfs(vnode_t vp)
2336 {
2337 	fsioc_auth_fs_t afs = { .authvp = NULL };
2338 	int error;
2339 
2340 	error = VNOP_IOCTL(vp, FSIOC_AUTH_FS, (caddr_t)&afs, 0,
2341 	    vfs_context_current());
2342 
2343 	return error == 0 ? 1 : 0;
2344 }
2345