xref: /xnu-11215.81.4/bsd/vfs/vfs_vnops.c (revision d4514f0bc1d3f944c22d92e68b646ac3fb40d452)
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 			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) {
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 
452 continue_create_lookup:
453 		if ((error = namei(ndp))) {
454 			goto out;
455 		}
456 
457 		dvp = ndp->ni_dvp;
458 		vp = ndp->ni_vp;
459 
460 		batched = vnode_compound_open_available(dvp);
461 
462 		/* not found, create */
463 		if (vp == NULL) {
464 			/* must have attributes for a new file */
465 			if (vap == NULL) {
466 				vnode_put(dvp);
467 				error = EINVAL;
468 				goto out;
469 			}
470 			/*
471 			 * Attempt a create.   For a system supporting compound VNOPs, we may
472 			 * find an existing file or create one; in either case, we will already
473 			 * have the file open and no VNOP_OPEN() will be needed.
474 			 */
475 			error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
476 
477 			dvp = ndp->ni_dvp;
478 			vp = ndp->ni_vp;
479 
480 			/* Detected a node that the filesystem couldn't handle */
481 			if (error == EKEEPLOOKING) {
482 				if (!batched) {
483 					panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
484 				}
485 				goto continue_create_lookup;
486 			}
487 
488 			if (dvp) {
489 				panic("Shouldn't have a dvp here.");
490 			}
491 
492 			if (error) {
493 				/*
494 				 * Check for a create race.
495 				 */
496 				if ((error == EEXIST) && !(fmode & O_EXCL)) {
497 					if (vp) {
498 						vnode_put(vp);
499 					}
500 					nameidone(ndp);
501 					goto again;
502 				}
503 				goto bad;
504 			}
505 
506 			need_vnop_open = !did_open;
507 		} else {
508 			if (fmode & O_EXCL) {
509 				error = EEXIST;
510 			}
511 
512 			/*
513 			 * We have a vnode.  Use compound open if available
514 			 * or else fall through to "traditional" path.  Note: can't
515 			 * do a compound open for root, because the parent belongs
516 			 * to a different FS.
517 			 */
518 			if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
519 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
520 
521 				if (error == 0) {
522 					vp = ndp->ni_vp;
523 					need_vnop_open = FALSE;
524 				} else if (error == EKEEPLOOKING) {
525 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
526 						panic("EKEEPLOOKING, but continue flag not set?");
527 					}
528 					goto continue_create_lookup;
529 				}
530 			}
531 			vnode_put(dvp);
532 			ndp->ni_dvp = NULLVP;
533 
534 			if (error) {
535 				goto bad;
536 			}
537 
538 			fmode &= ~O_CREAT;
539 
540 			/* Fall through */
541 		}
542 	} else {
543 		/*
544 		 * Not O_CREAT
545 		 */
546 		ndp->ni_cnd.cn_nameiop = LOOKUP;
547 		/* Inherit USEDVP, vnode_open() supported flags only */
548 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
549 		ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1 | WANTPARENT;
550 #if NAMEDRSRCFORK
551 		/* open calls are allowed for resource forks. */
552 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
553 #endif
554 		if (fmode & FENCRYPTED) {
555 			ndp->ni_cnd.cn_flags |= CN_RAW_ENCRYPTED | CN_SKIPNAMECACHE;
556 		}
557 		/* Inherit NAMEI_ROOTDIR flag only */
558 		ndp->ni_flag &= NAMEI_ROOTDIR;
559 		ndp->ni_flag |= NAMEI_COMPOUNDOPEN;
560 
561 		/* preserve NOFOLLOW from vnode_open() */
562 		if (fmode & O_NOFOLLOW || fmode & O_SYMLINK || (origcnflags & FOLLOW) == 0) {
563 			ndp->ni_cnd.cn_flags &= ~FOLLOW;
564 		}
565 		if (fmode & O_NOFOLLOW_ANY) {
566 			/* will return ELOOP on the first symlink to be hit */
567 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
568 		}
569 
570 		/* Do a lookup, possibly going directly to filesystem for compound operation */
571 		do {
572 			if ((error = namei(ndp))) {
573 				goto out;
574 			}
575 			vp = ndp->ni_vp;
576 			dvp = ndp->ni_dvp;
577 
578 			/* Check for batched lookup-open */
579 			batched = vnode_compound_open_available(dvp);
580 			if (batched && ((vp == NULLVP) || (vnode_mount(dvp) == vnode_mount(vp)))) {
581 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
582 				vp = ndp->ni_vp;
583 				if (error == 0) {
584 					need_vnop_open = FALSE;
585 				} else if (error == EKEEPLOOKING) {
586 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
587 						panic("EKEEPLOOKING, but continue flag not set?");
588 					}
589 				} else if ((fmode & (FREAD | FWRITE | FEXEC)) == FEXEC) {
590 					/*
591 					 * Some file systems fail in vnop_open call with absense of
592 					 * both FREAD and FWRITE access modes. Retry the vnop_open
593 					 * call again with FREAD access mode added.
594 					 */
595 					error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0,
596 					    fmode | FREAD, NULL, NULL, ctx);
597 					if (error == 0) {
598 						vp = ndp->ni_vp;
599 						need_vnop_open = FALSE;
600 					}
601 				}
602 			}
603 		} while (error == EKEEPLOOKING);
604 
605 		vnode_put(dvp);
606 		ndp->ni_dvp = NULLVP;
607 
608 		if (error) {
609 			goto bad;
610 		}
611 	}
612 
613 	/*
614 	 * By this point, dvp iocount is dropped 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 	nameidone(ndp);
706 	return 0;
707 
708 bad:
709 	/* Opened either explicitly or by a batched create */
710 	if (!need_vnop_open) {
711 		VNOP_CLOSE(vp, fmode, ctx);
712 	}
713 
714 	ndp->ni_vp = NULL;
715 	if (vp) {
716 #if NAMEDRSRCFORK
717 		/* Aggressively recycle shadow files if we error'd out during open() */
718 		if ((vnode_isnamedstream(vp)) &&
719 		    (vp->v_parent != NULLVP) &&
720 		    (vnode_isshadow(vp))) {
721 			vnode_recycle(vp);
722 		}
723 #endif
724 		vnode_put(vp);
725 		/*
726 		 * Check for a race against unlink.  We had a vnode
727 		 * but according to vnode_authorize or VNOP_OPEN it
728 		 * no longer exists.
729 		 *
730 		 * EREDRIVEOPEN: means that we were hit by the tty allocation race or NFSv4 open/create race.
731 		 */
732 		if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN) || ref_failed) {
733 			/*
734 			 * We'll retry here but it may be possible that we get
735 			 * into a retry "spin" inside the kernel and not allow
736 			 * threads, which need to run in order for the retry
737 			 * loop to end, to run. An example is an open of a
738 			 * terminal which is getting revoked and we spin here
739 			 * without yielding becasue namei and VNOP_OPEN are
740 			 * successful but vnode_ref fails. The revoke needs
741 			 * threads with an iocount to run but if spin here we
742 			 * may possibly be blcoking other threads from running.
743 			 *
744 			 * We start yielding the CPU after some number of
745 			 * retries for increasing durations. Note that this is
746 			 * still a loop without an exit condition.
747 			 */
748 			nretries += 1;
749 			if (nretries > RETRY_NO_YIELD_COUNT) {
750 				/* Every hz/100 secs is 10 msecs ... */
751 				tsleep(&nretries, PVFS, "vn_open_auth_retry",
752 				    MIN((nretries * (hz / 100)), hz));
753 			}
754 			nameidone(ndp);
755 			goto again;
756 		}
757 	}
758 
759 out:
760 	nameidone(ndp);
761 	return error;
762 }
763 
764 #if vn_access_DEPRECATED
765 /*
766  * Authorize an action against a vnode.  This has been the canonical way to
767  * ensure that the credential/process/etc. referenced by a vfs_context
768  * is granted the rights called out in 'mode' against the vnode 'vp'.
769  *
770  * Unfortunately, the use of VREAD/VWRITE/VEXEC makes it very difficult
771  * to add support for more rights.  As such, this interface will be deprecated
772  * and callers will use vnode_authorize instead.
773  */
774 int
vn_access(vnode_t vp,int mode,vfs_context_t context)775 vn_access(vnode_t vp, int mode, vfs_context_t context)
776 {
777 	kauth_action_t  action;
778 
779 	action = 0;
780 	if (mode & VREAD) {
781 		action |= KAUTH_VNODE_READ_DATA;
782 	}
783 	if (mode & VWRITE) {
784 		action |= KAUTH_VNODE_WRITE_DATA;
785 	}
786 	if (mode & VEXEC) {
787 		action |= KAUTH_VNODE_EXECUTE;
788 	}
789 
790 	return vnode_authorize(vp, NULL, action, context);
791 }
792 #endif  /* vn_access_DEPRECATED */
793 
794 /*
795  * Vnode close call
796  */
797 int
vn_close(struct vnode * vp,int flags,vfs_context_t ctx)798 vn_close(struct vnode *vp, int flags, vfs_context_t ctx)
799 {
800 	int error;
801 	int flusherror = 0;
802 
803 #if NAMEDRSRCFORK
804 	/* Sync data from resource fork shadow file if needed. */
805 	if ((vp->v_flag & VISNAMEDSTREAM) &&
806 	    (vp->v_parent != NULLVP) &&
807 	    vnode_isshadow(vp)) {
808 		if (flags & FWASWRITTEN) {
809 			flusherror = vnode_flushnamedstream(vp->v_parent, vp, ctx);
810 		}
811 	}
812 #endif
813 	/*
814 	 * If vnode @vp belongs to a chardev or a blkdev then it is handled
815 	 * specially.  We first drop its user reference count @vp->v_usecount
816 	 * before calling VNOP_CLOSE().  This was done historically to ensure
817 	 * that the last close of a special device vnode performed some
818 	 * conditional cleanups.  Now we still need to drop this reference here
819 	 * to ensure that devfsspec_close() can check if the vnode is still in
820 	 * use.
821 	 */
822 	if (vnode_isspec(vp)) {
823 		(void)vnode_rele_ext(vp, flags, 0);
824 	}
825 
826 	/*
827 	 * On HFS, we flush when the last writer closes.  We do this
828 	 * because resource fork vnodes hold a reference on data fork
829 	 * vnodes and that will prevent them from getting VNOP_INACTIVE
830 	 * which will delay when we flush cached data.  In future, we
831 	 * might find it beneficial to do this for all file systems.
832 	 * Note that it's OK to access v_writecount without the lock
833 	 * in this context.
834 	 */
835 	if (vp->v_tag == VT_HFS && (flags & FWRITE) && vp->v_writecount == 1) {
836 		VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
837 	}
838 
839 	error = VNOP_CLOSE(vp, flags, ctx);
840 
841 #if CONFIG_FSE
842 	if (flags & FWASWRITTEN) {
843 		if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
844 			add_fsevent(FSE_CONTENT_MODIFIED, ctx,
845 			    FSE_ARG_VNODE, vp,
846 			    FSE_ARG_DONE);
847 		}
848 	}
849 #endif
850 
851 	if (!vnode_isspec(vp)) {
852 		(void)vnode_rele_ext(vp, flags, 0);
853 	}
854 
855 	if (flusherror) {
856 		error = flusherror;
857 	}
858 	return error;
859 }
860 
861 static int
vn_read_swapfile(struct vnode * vp,uio_t uio)862 vn_read_swapfile(
863 	struct vnode    *vp,
864 	uio_t           uio)
865 {
866 	int     error;
867 	off_t   swap_count, this_count;
868 	off_t   file_end, read_end;
869 	off_t   prev_resid;
870 	char    *my_swap_page;
871 
872 	/*
873 	 * Reading from a swap file will get you zeroes.
874 	 */
875 
876 	my_swap_page = NULL;
877 	error = 0;
878 	swap_count = uio_resid(uio);
879 
880 	file_end = ubc_getsize(vp);
881 	read_end = uio->uio_offset + uio_resid(uio);
882 	if (uio->uio_offset >= file_end) {
883 		/* uio starts after end of file: nothing to read */
884 		swap_count = 0;
885 	} else if (read_end > file_end) {
886 		/* uio extends beyond end of file: stop before that */
887 		swap_count -= (read_end - file_end);
888 	}
889 
890 	while (swap_count > 0) {
891 		if (my_swap_page == NULL) {
892 			my_swap_page = kalloc_data(PAGE_SIZE, Z_WAITOK | Z_ZERO);
893 			/* add an end-of-line to keep line counters happy */
894 			my_swap_page[PAGE_SIZE - 1] = '\n';
895 		}
896 		this_count = swap_count;
897 		if (this_count > PAGE_SIZE) {
898 			this_count = PAGE_SIZE;
899 		}
900 
901 		prev_resid = uio_resid(uio);
902 		error = uiomove((caddr_t) my_swap_page,
903 		    (int)this_count,
904 		    uio);
905 		if (error) {
906 			break;
907 		}
908 		swap_count -= (prev_resid - uio_resid(uio));
909 	}
910 	kfree_data(my_swap_page, PAGE_SIZE);
911 
912 	return error;
913 }
914 /*
915  * Package up an I/O request on a vnode into a uio and do it.
916  */
917 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)918 vn_rdwr(
919 	enum uio_rw rw,
920 	struct vnode *vp,
921 	caddr_t base,
922 	int len,
923 	off_t offset,
924 	enum uio_seg segflg,
925 	int ioflg,
926 	kauth_cred_t cred,
927 	int *aresid,
928 	proc_t p)
929 {
930 	int64_t resid;
931 	int result;
932 
933 	if (len < 0) {
934 		return EINVAL;
935 	}
936 
937 	result = vn_rdwr_64(rw,
938 	    vp,
939 	    (uint64_t)(uintptr_t)base,
940 	    (int64_t)len,
941 	    offset,
942 	    segflg,
943 	    ioflg,
944 	    cred,
945 	    &resid,
946 	    p);
947 
948 	/* "resid" should be bounded above by "len," which is an int */
949 	if (aresid != NULL) {
950 		*aresid = (int)resid;
951 	}
952 
953 	return result;
954 }
955 
956 
957 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)958 vn_rdwr_64(
959 	enum uio_rw rw,
960 	struct vnode *vp,
961 	uint64_t base,
962 	int64_t len,
963 	off_t offset,
964 	enum uio_seg segflg,
965 	int ioflg,
966 	kauth_cred_t cred,
967 	int64_t *aresid,
968 	proc_t p)
969 {
970 	uio_t auio;
971 	int spacetype;
972 	struct vfs_context context;
973 	int error = 0;
974 	UIO_STACKBUF(uio_buf, 1);
975 
976 	context.vc_thread = current_thread();
977 	context.vc_ucred = cred;
978 
979 	if (UIO_SEG_IS_USER_SPACE(segflg)) {
980 		spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
981 	} else {
982 		spacetype = UIO_SYSSPACE;
983 	}
984 
985 	if (len < 0) {
986 		return EINVAL;
987 	}
988 
989 	auio = uio_createwithbuffer(1, offset, spacetype, rw,
990 	    &uio_buf[0], sizeof(uio_buf));
991 	uio_addiov(auio, CAST_USER_ADDR_T(base), (user_size_t)len);
992 
993 #if CONFIG_MACF
994 	/* XXXMAC
995 	 *      IO_NOAUTH should be re-examined.
996 	 *	Likely that mediation should be performed in caller.
997 	 */
998 	if ((ioflg & IO_NOAUTH) == 0) {
999 		/* passed cred is fp->f_cred */
1000 		if (rw == UIO_READ) {
1001 			error = mac_vnode_check_read(&context, cred, vp);
1002 		} else {
1003 			error = mac_vnode_check_write(&context, cred, vp);
1004 		}
1005 	}
1006 #endif
1007 
1008 	if (error == 0) {
1009 		if (rw == UIO_READ) {
1010 			if (vnode_isswap(vp) && ((ioflg & IO_SWAP_DISPATCH) == 0)) {
1011 				error = vn_read_swapfile(vp, auio);
1012 			} else {
1013 				error = VNOP_READ(vp, auio, ioflg, &context);
1014 			}
1015 		} else {
1016 			error = VNOP_WRITE(vp, auio, ioflg, &context);
1017 		}
1018 	}
1019 
1020 	if (aresid) {
1021 		*aresid = uio_resid(auio);
1022 		assert(*aresid <= len);
1023 	} else if (uio_resid(auio) && error == 0) {
1024 		error = EIO;
1025 	}
1026 	return error;
1027 }
1028 
1029 void
vn_offset_lock(struct fileglob * fg)1030 vn_offset_lock(struct fileglob *fg)
1031 {
1032 	lck_mtx_lock_spin(&fg->fg_lock);
1033 	while (fg->fg_lflags & FG_OFF_LOCKED) {
1034 		fg->fg_lflags |= FG_OFF_LOCKWANT;
1035 		msleep(&fg->fg_lflags, &fg->fg_lock, PVFS | PSPIN,
1036 		    "fg_offset_lock_wait", 0);
1037 	}
1038 	fg->fg_lflags |= FG_OFF_LOCKED;
1039 	lck_mtx_unlock(&fg->fg_lock);
1040 }
1041 
1042 void
vn_offset_unlock(struct fileglob * fg)1043 vn_offset_unlock(struct fileglob *fg)
1044 {
1045 	int lock_wanted = 0;
1046 
1047 	lck_mtx_lock_spin(&fg->fg_lock);
1048 	if (fg->fg_lflags & FG_OFF_LOCKWANT) {
1049 		lock_wanted = 1;
1050 	}
1051 	fg->fg_lflags &= ~(FG_OFF_LOCKED | FG_OFF_LOCKWANT);
1052 	lck_mtx_unlock(&fg->fg_lock);
1053 	if (lock_wanted) {
1054 		wakeup(&fg->fg_lflags);
1055 	}
1056 }
1057 
1058 static int
vn_read_common(vnode_t vp,struct uio * uio,int fflag,vfs_context_t ctx)1059 vn_read_common(vnode_t vp, struct uio *uio, int fflag, vfs_context_t ctx)
1060 {
1061 	int error;
1062 	int ioflag;
1063 	off_t read_offset;
1064 	user_ssize_t  read_len;
1065 	user_ssize_t  adjusted_read_len;
1066 	user_ssize_t  clippedsize;
1067 
1068 	/* Caller has already validated read_len. */
1069 	read_len = uio_resid(uio);
1070 	assert(read_len >= 0 && read_len <= INT_MAX);
1071 
1072 	adjusted_read_len = read_len;
1073 	clippedsize = 0;
1074 
1075 #if CONFIG_MACF
1076 	error = mac_vnode_check_read(ctx, vfs_context_ucred(ctx), vp);
1077 	if (error) {
1078 		return error;
1079 	}
1080 #endif
1081 
1082 	/* This signals to VNOP handlers that this read came from a file table read */
1083 	ioflag = IO_SYSCALL_DISPATCH;
1084 
1085 	if (fflag & FNONBLOCK) {
1086 		ioflag |= IO_NDELAY;
1087 	}
1088 	if ((fflag & FNOCACHE) || vnode_isnocache(vp)) {
1089 		ioflag |= IO_NOCACHE;
1090 	}
1091 	if (fflag & FENCRYPTED) {
1092 		ioflag |= IO_ENCRYPTED;
1093 	}
1094 	if (fflag & FUNENCRYPTED) {
1095 		ioflag |= IO_SKIP_ENCRYPTION;
1096 	}
1097 	if (fflag & O_EVTONLY) {
1098 		ioflag |= IO_EVTONLY;
1099 	}
1100 	if (fflag & FNORDAHEAD) {
1101 		ioflag |= IO_RAOFF;
1102 	}
1103 
1104 	read_offset = uio_offset(uio);
1105 	/* POSIX allows negative offsets for character devices. */
1106 	if ((read_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1107 		error = EINVAL;
1108 		goto error_out;
1109 	}
1110 
1111 	if (read_offset == INT64_MAX) {
1112 		/* can't read any more */
1113 		error = 0;
1114 		goto error_out;
1115 	}
1116 
1117 	/*
1118 	 * If offset + len will cause overflow, reduce the len to a value
1119 	 * (adjusted_read_len) where it won't
1120 	 */
1121 	if ((read_offset >= 0) && (INT64_MAX - read_offset) < read_len) {
1122 		/*
1123 		 * 0                                          read_offset  INT64_MAX
1124 		 * |-----------------------------------------------|----------|~~~
1125 		 *                                                 <--read_len-->
1126 		 *                                                 <-adjusted->
1127 		 */
1128 		adjusted_read_len = (user_ssize_t)(INT64_MAX - read_offset);
1129 	}
1130 
1131 	if (adjusted_read_len < read_len) {
1132 		uio_setresid(uio, adjusted_read_len);
1133 		clippedsize = read_len - adjusted_read_len;
1134 	}
1135 
1136 	if (vnode_isswap(vp) && !(IO_SKIP_ENCRYPTION & ioflag)) {
1137 		/* special case for swap files */
1138 		error = vn_read_swapfile(vp, uio);
1139 	} else {
1140 		error = VNOP_READ(vp, uio, ioflag, ctx);
1141 	}
1142 
1143 	if (clippedsize) {
1144 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1145 	}
1146 
1147 error_out:
1148 	return error;
1149 }
1150 
1151 /*
1152  * File table vnode read routine.
1153  */
1154 static int
vn_read(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1155 vn_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1156 {
1157 	vnode_t vp;
1158 	user_ssize_t read_len;
1159 	int error;
1160 	bool offset_locked;
1161 
1162 	read_len = uio_resid(uio);
1163 	if (read_len < 0 || read_len > INT_MAX) {
1164 		return EINVAL;
1165 	}
1166 
1167 	if ((flags & FOF_OFFSET) == 0) {
1168 		vn_offset_lock(fp->fp_glob);
1169 		offset_locked = true;
1170 	} else {
1171 		offset_locked = false;
1172 	}
1173 	vp = (struct vnode *)fp_get_data(fp);
1174 	if ((error = vnode_getwithref(vp))) {
1175 		if (offset_locked) {
1176 			vn_offset_unlock(fp->fp_glob);
1177 		}
1178 		return error;
1179 	}
1180 
1181 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1182 		vn_offset_unlock(fp->fp_glob);
1183 		offset_locked = false;
1184 	}
1185 
1186 	if ((flags & FOF_OFFSET) == 0) {
1187 		uio_setoffset(uio, fp->fp_glob->fg_offset);
1188 	}
1189 
1190 	error = vn_read_common(vp, uio, fp->fp_glob->fg_flag, ctx);
1191 
1192 	if ((flags & FOF_OFFSET) == 0) {
1193 		fp->fp_glob->fg_offset += read_len - uio_resid(uio);
1194 	}
1195 
1196 	if (offset_locked) {
1197 		vn_offset_unlock(fp->fp_glob);
1198 		offset_locked = false;
1199 	}
1200 
1201 	vnode_put(vp);
1202 	return error;
1203 }
1204 
1205 /*
1206  * File table vnode write routine.
1207  */
1208 static int
vn_write(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1209 vn_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1210 {
1211 	struct vnode *vp;
1212 	int error, ioflag;
1213 	off_t write_offset;
1214 	off_t write_end_offset;
1215 	user_ssize_t write_len;
1216 	user_ssize_t adjusted_write_len;
1217 	user_ssize_t clippedsize;
1218 	bool offset_locked;
1219 	proc_t p = vfs_context_proc(ctx);
1220 	rlim_t rlim_cur_fsize = p ? proc_limitgetcur(p, RLIMIT_FSIZE) : 0;
1221 
1222 	write_len = uio_resid(uio);
1223 	if (write_len < 0 || write_len > INT_MAX) {
1224 		return EINVAL;
1225 	}
1226 	adjusted_write_len = write_len;
1227 	clippedsize = 0;
1228 
1229 	if ((flags & FOF_OFFSET) == 0) {
1230 		vn_offset_lock(fp->fp_glob);
1231 		offset_locked = true;
1232 	} else {
1233 		offset_locked = false;
1234 	}
1235 
1236 	vp = (struct vnode *)fp_get_data(fp);
1237 	if ((error = vnode_getwithref(vp))) {
1238 		if (offset_locked) {
1239 			vn_offset_unlock(fp->fp_glob);
1240 		}
1241 		return error;
1242 	}
1243 
1244 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1245 		vn_offset_unlock(fp->fp_glob);
1246 		offset_locked = false;
1247 	}
1248 
1249 #if CONFIG_MACF
1250 	error = mac_vnode_check_write(ctx, vfs_context_ucred(ctx), vp);
1251 	if (error) {
1252 		(void)vnode_put(vp);
1253 		if (offset_locked) {
1254 			vn_offset_unlock(fp->fp_glob);
1255 		}
1256 		return error;
1257 	}
1258 #endif
1259 
1260 	/*
1261 	 * IO_SYSCALL_DISPATCH signals to VNOP handlers that this write came from
1262 	 * a file table write
1263 	 */
1264 	ioflag = (IO_UNIT | IO_SYSCALL_DISPATCH);
1265 
1266 	if (vp->v_type == VREG && (fp->fp_glob->fg_flag & O_APPEND)) {
1267 		ioflag |= IO_APPEND;
1268 	}
1269 	if (fp->fp_glob->fg_flag & FNONBLOCK) {
1270 		ioflag |= IO_NDELAY;
1271 	}
1272 	assert(!(fp->fp_glob->fg_flag & O_NOCTTY) || (fp->fp_glob->fg_flag & FNOCACHE));
1273 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1274 		ioflag |= IO_NOCACHE;
1275 		/*
1276 		 * O_NOCTTY is repurposed to indicate the preference for relaxed
1277 		 * alignment and size restrictions. O_NOCTTY has no meaning beyond
1278 		 * the open of the fd and is used only for ttys and the use
1279 		 * is mutually exclusive with nocache io for regular files.
1280 		 */
1281 		if (fp->fp_glob->fg_flag & O_NOCTTY || vfs_context_allow_fs_blksize_nocache_write(ctx)) {
1282 			ioflag |= IO_NOCACHE_SWRITE;
1283 		}
1284 	}
1285 	if (fp->fp_glob->fg_flag & FNODIRECT) {
1286 		ioflag |= IO_NODIRECT;
1287 	}
1288 	if (fp->fp_glob->fg_flag & FSINGLE_WRITER) {
1289 		ioflag |= IO_SINGLE_WRITER;
1290 	}
1291 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1292 		ioflag |= IO_EVTONLY;
1293 	}
1294 
1295 	/*
1296 	 * Treat synchronous mounts and O_FSYNC on the fd as equivalent.
1297 	 *
1298 	 * XXX We treat O_DSYNC as O_FSYNC for now, since we can not delay
1299 	 * XXX the non-essential metadata without some additional VFS work;
1300 	 * XXX the intent at this point is to plumb the interface for it.
1301 	 */
1302 	if ((fp->fp_glob->fg_flag & (O_FSYNC | O_DSYNC)) ||
1303 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) {
1304 		ioflag |= IO_SYNC;
1305 	}
1306 
1307 	if ((flags & FOF_OFFSET) == 0) {
1308 		write_offset = fp->fp_glob->fg_offset;
1309 		uio_setoffset(uio, write_offset);
1310 	} else {
1311 		/* for pwrite, append should  be ignored */
1312 		ioflag &= ~IO_APPEND;
1313 		write_offset = uio_offset(uio);
1314 		/* POSIX allows negative offsets for character devices. */
1315 		if ((write_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1316 			error = EINVAL;
1317 			goto error_out;
1318 		}
1319 	}
1320 
1321 	if (write_offset == INT64_MAX) {
1322 		/* writes are not possible */
1323 		error = EFBIG;
1324 		goto error_out;
1325 	}
1326 
1327 	/*
1328 	 * write_len is the original write length that was requested.
1329 	 * We may however need to reduce that becasue of two reasons
1330 	 *
1331 	 * 1) If write_offset + write_len will exceed OFF_T_MAX (i.e. INT64_MAX)
1332 	 *    and/or
1333 	 * 2) If write_offset + write_len will exceed the administrative
1334 	 *    limit for the maximum file size.
1335 	 *
1336 	 * In both cases the write will be denied if we can't write even a single
1337 	 * byte otherwise it will be "clipped" (i.e. a short write).
1338 	 */
1339 
1340 	/*
1341 	 * If offset + len will cause overflow, reduce the len
1342 	 * to a value (adjusted_write_len) where it won't
1343 	 */
1344 	if ((write_offset >= 0) && (INT64_MAX - write_offset) < write_len) {
1345 		/*
1346 		 * 0                                          write_offset  INT64_MAX
1347 		 * |-----------------------------------------------|----------|~~~
1348 		 *                                                 <--write_len-->
1349 		 *                                                 <-adjusted->
1350 		 */
1351 		adjusted_write_len = (user_ssize_t)(INT64_MAX - write_offset);
1352 	}
1353 
1354 	/* write_end_offset will always be [0, INT64_MAX] */
1355 	write_end_offset = write_offset + adjusted_write_len;
1356 
1357 	if (p && (vp->v_type == VREG) &&
1358 	    (rlim_cur_fsize != RLIM_INFINITY) &&
1359 	    (rlim_cur_fsize <= INT64_MAX) &&
1360 	    (write_end_offset > (off_t)rlim_cur_fsize)) {
1361 		/*
1362 		 * If the requested residual would cause us to go past the
1363 		 * administrative limit, then we need to adjust the residual
1364 		 * down to cause fewer bytes than requested to be written.  If
1365 		 * we can't do that (e.g. the residual is already 1 byte),
1366 		 * then we fail the write with EFBIG.
1367 		 */
1368 		if (write_offset >= (off_t)rlim_cur_fsize) {
1369 			/*
1370 			 * 0                  rlim_fsize  write_offset  write_end  INT64_MAX
1371 			 * |------------------------|----------|-------------|--------|
1372 			 *                                     <--write_len-->
1373 			 *
1374 			 *                write not permitted
1375 			 */
1376 			psignal(p, SIGXFSZ);
1377 			error = EFBIG;
1378 			goto error_out;
1379 		}
1380 
1381 		/*
1382 		 * 0                  write_offset rlim_fsize  write_end  INT64_MAX
1383 		 * |------------------------|-----------|---------|------------|
1384 		 *                          <------write_len------>
1385 		 *                          <-adjusted-->
1386 		 */
1387 		adjusted_write_len =  (user_ssize_t)((off_t)rlim_cur_fsize - write_offset);
1388 		assert((adjusted_write_len > 0) && (adjusted_write_len < write_len));
1389 	}
1390 
1391 	if (adjusted_write_len < write_len) {
1392 		uio_setresid(uio, adjusted_write_len);
1393 		clippedsize = write_len - adjusted_write_len;
1394 	}
1395 
1396 	error = VNOP_WRITE(vp, uio, ioflag, ctx);
1397 
1398 	/*
1399 	 * If we had to reduce the size of write requested either because
1400 	 * of rlimit or because it would have exceeded
1401 	 * maximum file size, we have to add that back to the residual so
1402 	 * it correctly reflects what we did in this function.
1403 	 */
1404 	if (clippedsize) {
1405 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1406 	}
1407 
1408 	if ((flags & FOF_OFFSET) == 0) {
1409 		if (ioflag & IO_APPEND) {
1410 			fp->fp_glob->fg_offset = uio_offset(uio);
1411 		} else {
1412 			fp->fp_glob->fg_offset += (write_len - uio_resid(uio));
1413 		}
1414 		if (offset_locked) {
1415 			vn_offset_unlock(fp->fp_glob);
1416 			offset_locked = false;
1417 		}
1418 	}
1419 
1420 	/*
1421 	 * Set the credentials on successful writes
1422 	 */
1423 	if ((error == 0) && (vp->v_tag == VT_NFS) && (UBCINFOEXISTS(vp))) {
1424 		ubc_setcred(vp, vfs_context_ucred(ctx));
1425 	}
1426 
1427 #if CONFIG_FILE_LEASES
1428 	/*
1429 	 * On success, break the parent dir lease as the file's attributes (size
1430 	 * and/or mtime) have changed. Best attempt to break lease, just drop the
1431 	 * the error upon failure as there is no point to return error when the
1432 	 * write has completed successfully.
1433 	 */
1434 	if (__probable(error == 0)) {
1435 		vnode_breakdirlease(vp, true, O_WRONLY);
1436 	}
1437 #endif /* CONFIG_FILE_LEASES */
1438 
1439 	(void)vnode_put(vp);
1440 	return error;
1441 
1442 error_out:
1443 	if (offset_locked) {
1444 		vn_offset_unlock(fp->fp_glob);
1445 	}
1446 	(void)vnode_put(vp);
1447 	return error;
1448 }
1449 
1450 /*
1451  * File table vnode stat routine.
1452  *
1453  * Returns:	0			Success
1454  *		EBADF
1455  *		ENOMEM
1456  *	vnode_getattr:???
1457  */
1458 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)1459 vn_stat_noauth(struct vnode *vp, void *sbptr, kauth_filesec_t *xsec, int isstat64,
1460     int needsrealdev, vfs_context_t ctx, struct ucred *file_cred)
1461 {
1462 	struct vnode_attr va;
1463 	int error;
1464 	u_short mode;
1465 	kauth_filesec_t fsec;
1466 	struct stat *sb = (struct stat *)0;     /* warning avoidance ; protected by isstat64 */
1467 	struct stat64 * sb64 = (struct stat64 *)0;  /* warning avoidance ; protected by isstat64 */
1468 
1469 	if (isstat64 != 0) {
1470 		sb64 = (struct stat64 *)sbptr;
1471 	} else {
1472 		sb = (struct stat *)sbptr;
1473 	}
1474 	memset(&va, 0, sizeof(va));
1475 	VATTR_INIT(&va);
1476 	VATTR_WANTED(&va, va_fsid);
1477 	VATTR_WANTED(&va, va_fileid);
1478 	VATTR_WANTED(&va, va_mode);
1479 	VATTR_WANTED(&va, va_type);
1480 	VATTR_WANTED(&va, va_nlink);
1481 	VATTR_WANTED(&va, va_uid);
1482 	VATTR_WANTED(&va, va_gid);
1483 	VATTR_WANTED(&va, va_rdev);
1484 	VATTR_WANTED(&va, va_data_size);
1485 	VATTR_WANTED(&va, va_access_time);
1486 	VATTR_WANTED(&va, va_modify_time);
1487 	VATTR_WANTED(&va, va_change_time);
1488 	VATTR_WANTED(&va, va_create_time);
1489 	VATTR_WANTED(&va, va_flags);
1490 	VATTR_WANTED(&va, va_gen);
1491 	VATTR_WANTED(&va, va_iosize);
1492 	/* lower layers will synthesise va_total_alloc from va_data_size if required */
1493 	VATTR_WANTED(&va, va_total_alloc);
1494 	if (xsec != NULL) {
1495 		VATTR_WANTED(&va, va_uuuid);
1496 		VATTR_WANTED(&va, va_guuid);
1497 		VATTR_WANTED(&va, va_acl);
1498 	}
1499 	if (needsrealdev) {
1500 		va.va_vaflags = VA_REALFSID;
1501 	}
1502 	error = vnode_getattr(vp, &va, ctx);
1503 	if (error) {
1504 		goto out;
1505 	}
1506 #if CONFIG_MACF
1507 	/*
1508 	 * Give MAC polices a chance to reject or filter the attributes
1509 	 * returned by the filesystem.  Note that MAC policies are consulted
1510 	 * *after* calling the filesystem because filesystems can return more
1511 	 * attributes than were requested so policies wouldn't be authoritative
1512 	 * is consulted beforehand.  This also gives policies an opportunity
1513 	 * to change the values of attributes retrieved.
1514 	 */
1515 	error = mac_vnode_check_getattr(ctx, file_cred, vp, &va);
1516 	if (error) {
1517 		goto out;
1518 	}
1519 #endif
1520 	/*
1521 	 * Copy from vattr table
1522 	 */
1523 	if (isstat64 != 0) {
1524 		sb64->st_dev = va.va_fsid;
1525 		sb64->st_ino = (ino64_t)va.va_fileid;
1526 	} else {
1527 		sb->st_dev = va.va_fsid;
1528 		sb->st_ino = (ino_t)va.va_fileid;
1529 	}
1530 	mode = va.va_mode;
1531 	switch (vp->v_type) {
1532 	case VREG:
1533 		mode |= S_IFREG;
1534 		break;
1535 	case VDIR:
1536 		mode |= S_IFDIR;
1537 		break;
1538 	case VBLK:
1539 		mode |= S_IFBLK;
1540 		break;
1541 	case VCHR:
1542 		mode |= S_IFCHR;
1543 		break;
1544 	case VLNK:
1545 		mode |= S_IFLNK;
1546 		break;
1547 	case VSOCK:
1548 		mode |= S_IFSOCK;
1549 		break;
1550 	case VFIFO:
1551 		mode |= S_IFIFO;
1552 		break;
1553 	default:
1554 		error = EBADF;
1555 		goto out;
1556 	}
1557 	;
1558 	if (isstat64 != 0) {
1559 		sb64->st_mode = mode;
1560 		sb64->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1561 		sb64->st_uid = va.va_uid;
1562 		sb64->st_gid = va.va_gid;
1563 		sb64->st_rdev = va.va_rdev;
1564 		sb64->st_size = va.va_data_size;
1565 		sb64->st_atimespec = va.va_access_time;
1566 		sb64->st_mtimespec = va.va_modify_time;
1567 		sb64->st_ctimespec = va.va_change_time;
1568 		if (VATTR_IS_SUPPORTED(&va, va_create_time)) {
1569 			sb64->st_birthtimespec =  va.va_create_time;
1570 		} else {
1571 			sb64->st_birthtimespec.tv_sec = sb64->st_birthtimespec.tv_nsec = 0;
1572 		}
1573 		sb64->st_blksize = va.va_iosize;
1574 		sb64->st_flags = va.va_flags;
1575 		sb64->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1576 	} else {
1577 		sb->st_mode = mode;
1578 		sb->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1579 		sb->st_uid = va.va_uid;
1580 		sb->st_gid = va.va_gid;
1581 		sb->st_rdev = va.va_rdev;
1582 		sb->st_size = va.va_data_size;
1583 		sb->st_atimespec = va.va_access_time;
1584 		sb->st_mtimespec = va.va_modify_time;
1585 		sb->st_ctimespec = va.va_change_time;
1586 		sb->st_blksize = va.va_iosize;
1587 		sb->st_flags = va.va_flags;
1588 		sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1589 	}
1590 
1591 	/* if we're interested in extended security data and we got an ACL */
1592 	if (xsec != NULL) {
1593 		if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
1594 		    !VATTR_IS_SUPPORTED(&va, va_uuuid) &&
1595 		    !VATTR_IS_SUPPORTED(&va, va_guuid)) {
1596 			*xsec = KAUTH_FILESEC_NONE;
1597 		} else {
1598 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1599 				fsec = kauth_filesec_alloc(va.va_acl->acl_entrycount);
1600 			} else {
1601 				fsec = kauth_filesec_alloc(0);
1602 			}
1603 			if (fsec == NULL) {
1604 				error = ENOMEM;
1605 				goto out;
1606 			}
1607 			fsec->fsec_magic = KAUTH_FILESEC_MAGIC;
1608 			if (VATTR_IS_SUPPORTED(&va, va_uuuid)) {
1609 				fsec->fsec_owner = va.va_uuuid;
1610 			} else {
1611 				fsec->fsec_owner = kauth_null_guid;
1612 			}
1613 			if (VATTR_IS_SUPPORTED(&va, va_guuid)) {
1614 				fsec->fsec_group = va.va_guuid;
1615 			} else {
1616 				fsec->fsec_group = kauth_null_guid;
1617 			}
1618 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1619 				__nochk_bcopy(va.va_acl, &(fsec->fsec_acl), KAUTH_ACL_COPYSIZE(va.va_acl));
1620 			} else {
1621 				fsec->fsec_acl.acl_entrycount = KAUTH_FILESEC_NOACL;
1622 			}
1623 			*xsec = fsec;
1624 		}
1625 	}
1626 
1627 	/* Do not give the generation number out to unpriviledged users */
1628 	if (va.va_gen && !vfs_context_issuser(ctx)) {
1629 		if (isstat64 != 0) {
1630 			sb64->st_gen = 0;
1631 		} else {
1632 			sb->st_gen = 0;
1633 		}
1634 	} else {
1635 		if (isstat64 != 0) {
1636 			sb64->st_gen = va.va_gen;
1637 		} else {
1638 			sb->st_gen = va.va_gen;
1639 		}
1640 	}
1641 
1642 	error = 0;
1643 out:
1644 	if (VATTR_IS_SUPPORTED(&va, va_acl) && va.va_acl != NULL) {
1645 		kauth_acl_free(va.va_acl);
1646 	}
1647 	return error;
1648 }
1649 
1650 int
vn_stat(struct vnode * vp,void * sb,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx)1651 vn_stat(struct vnode *vp, void *sb, kauth_filesec_t *xsec, int isstat64, int needsrealdev, vfs_context_t ctx)
1652 {
1653 	int error;
1654 
1655 #if CONFIG_MACF
1656 	error = mac_vnode_check_stat(ctx, NOCRED, vp);
1657 	if (error) {
1658 		return error;
1659 	}
1660 #endif
1661 
1662 	/* authorize */
1663 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_ATTRIBUTES | KAUTH_VNODE_READ_SECURITY, ctx)) != 0) {
1664 		return error;
1665 	}
1666 
1667 	/* actual stat */
1668 	return vn_stat_noauth(vp, sb, xsec, isstat64, needsrealdev, ctx, NOCRED);
1669 }
1670 
1671 
1672 /*
1673  * File table vnode ioctl routine.
1674  */
1675 static int
vn_ioctl(struct fileproc * fp,u_long com,caddr_t data,vfs_context_t ctx)1676 vn_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
1677 {
1678 	struct vnode *vp = (struct vnode *)fp_get_data(fp);
1679 	off_t file_size;
1680 	int error;
1681 	struct vnode *ttyvp;
1682 
1683 	if ((error = vnode_getwithref(vp))) {
1684 		return error;
1685 	}
1686 
1687 #if CONFIG_MACF
1688 	error = mac_vnode_check_ioctl(ctx, vp, com);
1689 	if (error) {
1690 		goto out;
1691 	}
1692 #endif
1693 
1694 	switch (vp->v_type) {
1695 	case VREG:
1696 	case VDIR:
1697 		if (com == FIONREAD) {
1698 			off_t temp_nbytes;
1699 			if ((error = vnode_size(vp, &file_size, ctx)) != 0) {
1700 				goto out;
1701 			}
1702 			temp_nbytes = file_size - fp->fp_glob->fg_offset;
1703 			if (temp_nbytes > INT_MAX) {
1704 				*(int *)data = INT_MAX;
1705 			} else if (temp_nbytes < 0) {
1706 				*(int *)data = 0;
1707 			} else {
1708 				*(int *)data = (int)temp_nbytes;
1709 			}
1710 			goto out;
1711 		}
1712 		if (com == FIONBIO || com == FIOASYNC) {        /* XXX */
1713 			goto out;
1714 		}
1715 		OS_FALLTHROUGH;
1716 
1717 	default:
1718 		error = ENOTTY;
1719 		goto out;
1720 
1721 	case VFIFO:
1722 	case VCHR:
1723 	case VBLK:
1724 
1725 		if (com == TIOCREVOKE || com == TIOCREVOKECLEAR) {
1726 			error = ENOTTY;
1727 			goto out;
1728 		}
1729 
1730 		/* Should not be able to set block size from user space */
1731 		if (com == DKIOCSETBLOCKSIZE) {
1732 			error = EPERM;
1733 			goto out;
1734 		}
1735 
1736 		/* Should not be able to check if filesystem is authenticated from user space */
1737 		if (com == FSIOC_AUTH_FS) {
1738 			error = ENOTTY;
1739 			goto out;
1740 		}
1741 
1742 		if (com == FIODTYPE) {
1743 			if (vp->v_type == VBLK) {
1744 				if (major(vp->v_rdev) >= nblkdev) {
1745 					error = ENXIO;
1746 					goto out;
1747 				}
1748 				*(int *)data = bdevsw[major(vp->v_rdev)].d_type;
1749 			} else if (vp->v_type == VCHR) {
1750 				if (major(vp->v_rdev) >= nchrdev) {
1751 					error = ENXIO;
1752 					goto out;
1753 				}
1754 				*(int *)data = cdevsw[major(vp->v_rdev)].d_type;
1755 			} else {
1756 				error = ENOTTY;
1757 				goto out;
1758 			}
1759 			goto out;
1760 		}
1761 		error = VNOP_IOCTL(vp, com, data, fp->fp_glob->fg_flag, ctx);
1762 
1763 		if (error == 0 && com == TIOCSCTTY) {
1764 			struct session *sessp;
1765 			struct pgrp *pg;
1766 
1767 			pg = proc_pgrp(vfs_context_proc(ctx), &sessp);
1768 
1769 			session_lock(sessp);
1770 			ttyvp = sessp->s_ttyvp;
1771 			sessp->s_ttyvp = vp;
1772 			sessp->s_ttyvid = vnode_vid(vp);
1773 			session_unlock(sessp);
1774 
1775 			pgrp_rele(pg);
1776 		}
1777 	}
1778 out:
1779 	(void)vnode_put(vp);
1780 	return error;
1781 }
1782 
1783 /*
1784  * File table vnode select routine.
1785  */
1786 static int
vn_select(struct fileproc * fp,int which,void * wql,__unused vfs_context_t ctx)1787 vn_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx)
1788 {
1789 	int error;
1790 	struct vnode * vp = (struct vnode *)fp_get_data(fp);
1791 	struct vfs_context context;
1792 
1793 	if ((error = vnode_getwithref(vp)) == 0) {
1794 		context.vc_thread = current_thread();
1795 		context.vc_ucred = fp->fp_glob->fg_cred;
1796 
1797 #if CONFIG_MACF
1798 		/*
1799 		 * XXX We should use a per thread credential here; minimally,
1800 		 * XXX the process credential should have a persistent
1801 		 * XXX reference on it before being passed in here.
1802 		 */
1803 		error = mac_vnode_check_select(ctx, vp, which);
1804 		if (error == 0)
1805 #endif
1806 		error = VNOP_SELECT(vp, which, fp->fp_glob->fg_flag, wql, ctx);
1807 
1808 		(void)vnode_put(vp);
1809 	}
1810 	return error;
1811 }
1812 
1813 /*
1814  * File table vnode close routine.
1815  */
1816 static int
vn_closefile(struct fileglob * fg,vfs_context_t ctx)1817 vn_closefile(struct fileglob *fg, vfs_context_t ctx)
1818 {
1819 	struct vnode *vp = fg_get_data(fg);
1820 	int error;
1821 
1822 	if ((error = vnode_getwithref(vp)) == 0) {
1823 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE &&
1824 		    ((fg->fg_flag & FWASLOCKED) != 0 ||
1825 		    (fg->fg_lflags & FG_HAS_OFDLOCK) != 0)) {
1826 			struct flock lf = {
1827 				.l_whence = SEEK_SET,
1828 				.l_start = 0,
1829 				.l_len = 0,
1830 				.l_type = F_UNLCK
1831 			};
1832 
1833 			if ((fg->fg_flag & FWASLOCKED) != 0) {
1834 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1835 				    F_UNLCK, &lf, F_FLOCK, ctx, NULL);
1836 			}
1837 
1838 			if ((fg->fg_lflags & FG_HAS_OFDLOCK) != 0) {
1839 				(void) VNOP_ADVLOCK(vp, ofd_to_id(fg),
1840 				    F_UNLCK, &lf, F_OFD_LOCK, ctx, NULL);
1841 			}
1842 		}
1843 #if CONFIG_FILE_LEASES
1844 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE && !LIST_EMPTY(&vp->v_leases)) {
1845 			/* Expected open count doesn't matter for release. */
1846 			(void)vnode_setlease(vp, fg, F_UNLCK, 0, ctx);
1847 		}
1848 #endif
1849 		error = vn_close(vp, fg->fg_flag, ctx);
1850 		(void) vnode_put(vp);
1851 	}
1852 	return error;
1853 }
1854 
1855 /*
1856  * Returns:	0			Success
1857  *	VNOP_PATHCONF:???
1858  */
1859 int
vn_pathconf(vnode_t vp,int name,int32_t * retval,vfs_context_t ctx)1860 vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
1861 {
1862 	int     error = 0;
1863 	struct vfs_attr vfa;
1864 
1865 	switch (name) {
1866 	case _PC_EXTENDED_SECURITY_NP:
1867 		*retval = vfs_extendedsecurity(vnode_mount(vp)) ? 1 : 0;
1868 		break;
1869 	case _PC_AUTH_OPAQUE_NP:
1870 		*retval = vfs_authopaque(vnode_mount(vp));
1871 		break;
1872 	case _PC_2_SYMLINKS:
1873 		*retval = 1;    /* XXX NOTSUP on MSDOS, etc. */
1874 		break;
1875 	case _PC_ALLOC_SIZE_MIN:
1876 		*retval = 1;    /* XXX lie: 1 byte */
1877 		break;
1878 	case _PC_ASYNC_IO:      /* unistd.h: _POSIX_ASYNCHRONUS_IO */
1879 		*retval = 1;    /* [AIO] option is supported */
1880 		break;
1881 	case _PC_PRIO_IO:       /* unistd.h: _POSIX_PRIORITIZED_IO */
1882 		*retval = 0;    /* [PIO] option is not supported */
1883 		break;
1884 	case _PC_REC_INCR_XFER_SIZE:
1885 		*retval = 4096; /* XXX go from MIN to MAX 4K at a time */
1886 		break;
1887 	case _PC_REC_MIN_XFER_SIZE:
1888 		*retval = 4096; /* XXX recommend 4K minimum reads/writes */
1889 		break;
1890 	case _PC_REC_MAX_XFER_SIZE:
1891 		*retval = 65536; /* XXX recommend 64K maximum reads/writes */
1892 		break;
1893 	case _PC_REC_XFER_ALIGN:
1894 		*retval = 4096; /* XXX recommend page aligned buffers */
1895 		break;
1896 	case _PC_SYMLINK_MAX:
1897 		*retval = 255;  /* Minimum acceptable POSIX value */
1898 		break;
1899 	case _PC_SYNC_IO:       /* unistd.h: _POSIX_SYNCHRONIZED_IO */
1900 		*retval = 0;    /* [SIO] option is not supported */
1901 		break;
1902 	case _PC_XATTR_SIZE_BITS:
1903 		/* The number of bits used to store maximum extended
1904 		 * attribute size in bytes.  For example, if the maximum
1905 		 * attribute size supported by a file system is 128K, the
1906 		 * value returned will be 18.  However a value 18 can mean
1907 		 * that the maximum attribute size can be anywhere from
1908 		 * (256KB - 1) to 128KB.  As a special case, the resource
1909 		 * fork can have much larger size, and some file system
1910 		 * specific extended attributes can have smaller and preset
1911 		 * size; for example, Finder Info is always 32 bytes.
1912 		 */
1913 		memset(&vfa, 0, sizeof(vfa));
1914 		VFSATTR_INIT(&vfa);
1915 		VFSATTR_WANTED(&vfa, f_capabilities);
1916 		if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
1917 		    (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
1918 		    (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
1919 		    (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
1920 			/* Supports native extended attributes */
1921 			error = VNOP_PATHCONF(vp, name, retval, ctx);
1922 		} else {
1923 			/* Number of bits used to represent the maximum size of
1924 			 * extended attribute stored in an Apple Double file.
1925 			 */
1926 			*retval = AD_XATTR_SIZE_BITS;
1927 		}
1928 		break;
1929 	default:
1930 		error = VNOP_PATHCONF(vp, name, retval, ctx);
1931 		break;
1932 	}
1933 
1934 	return error;
1935 }
1936 
1937 static int
vn_kqfilter(struct fileproc * fp,struct knote * kn,struct kevent_qos_s * kev)1938 vn_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
1939 {
1940 	vfs_context_t ctx = vfs_context_current();
1941 	struct vnode *vp;
1942 	int error = 0;
1943 	int result = 0;
1944 
1945 	vp = (struct vnode *)fp_get_data(fp);
1946 
1947 	/*
1948 	 * Don't attach a knote to a dead vnode.
1949 	 */
1950 	if ((error = vget_internal(vp, 0, VNODE_NODEAD)) == 0) {
1951 		switch (kn->kn_filter) {
1952 		case EVFILT_READ:
1953 		case EVFILT_WRITE:
1954 			if (vnode_isfifo(vp)) {
1955 				/* We'll only watch FIFOs that use our fifofs */
1956 				if (!(vp->v_fifoinfo && vp->v_fifoinfo->fi_readsock)) {
1957 					error = ENOTSUP;
1958 				}
1959 			} else if (!vnode_isreg(vp)) {
1960 				if (vnode_ischr(vp)) {
1961 					result = spec_kqfilter(vp, kn, kev);
1962 					if ((kn->kn_flags & EV_ERROR) == 0) {
1963 						/* claimed by a special device */
1964 						vnode_put(vp);
1965 						return result;
1966 					}
1967 				}
1968 				error = EINVAL;
1969 			}
1970 			break;
1971 		case EVFILT_VNODE:
1972 			break;
1973 		default:
1974 			error = EINVAL;
1975 		}
1976 
1977 		if (error == 0) {
1978 #if CONFIG_MACF
1979 			error = mac_vnode_check_kqfilter(ctx, fp->fp_glob->fg_cred, kn, vp);
1980 			if (error) {
1981 				vnode_put(vp);
1982 				goto out;
1983 			}
1984 #endif
1985 
1986 			kn->kn_filtid = EVFILTID_VN;
1987 			knote_kn_hook_set_raw(kn, (void *)vp);
1988 			vnode_hold(vp);
1989 
1990 			vnode_lock(vp);
1991 			KNOTE_ATTACH(&vp->v_knotes, kn);
1992 			result = filt_vnode_common(kn, NULL, vp, 0);
1993 			vnode_unlock(vp);
1994 
1995 			/*
1996 			 * Ask the filesystem to provide remove notifications,
1997 			 * but ignore failure
1998 			 */
1999 			VNOP_MONITOR(vp, 0, VNODE_MONITOR_BEGIN, (void*) kn, ctx);
2000 		}
2001 
2002 		vnode_put(vp);
2003 	}
2004 
2005 out:
2006 	if (error) {
2007 		knote_set_error(kn, error);
2008 	}
2009 
2010 	return result;
2011 }
2012 
2013 static void
filt_vndetach(struct knote * kn)2014 filt_vndetach(struct knote *kn)
2015 {
2016 	vfs_context_t ctx = vfs_context_current();
2017 	struct vnode *vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2018 	uint32_t vid = vnode_vid(vp);
2019 	if (vnode_getwithvid(vp, vid)) {
2020 		vnode_drop(vp);
2021 		return;
2022 	}
2023 	vnode_drop(vp);
2024 
2025 	vnode_lock(vp);
2026 	KNOTE_DETACH(&vp->v_knotes, kn);
2027 	vnode_unlock(vp);
2028 
2029 	/*
2030 	 * Tell a (generally networked) filesystem that we're no longer watching
2031 	 * If the FS wants to track contexts, it should still be using the one from
2032 	 * the VNODE_MONITOR_BEGIN.
2033 	 */
2034 	VNOP_MONITOR(vp, 0, VNODE_MONITOR_END, (void*)kn, ctx);
2035 	vnode_put(vp);
2036 }
2037 
2038 
2039 /*
2040  * Used for EVFILT_READ
2041  *
2042  * Takes only VFIFO or VREG. vnode is locked.  We handle the "poll" case
2043  * differently than the regular case for VREG files.  If not in poll(),
2044  * then we need to know current fileproc offset for VREG.
2045  */
2046 static int64_t
vnode_readable_data_count(vnode_t vp,off_t current_offset,int ispoll)2047 vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
2048 {
2049 	if (vnode_isfifo(vp)) {
2050 #if FIFO
2051 		int cnt;
2052 		int err = fifo_charcount(vp, &cnt);
2053 		if (err == 0) {
2054 			return (int64_t)cnt;
2055 		} else
2056 #endif
2057 		{
2058 			return 0;
2059 		}
2060 	} else if (vnode_isreg(vp)) {
2061 		if (ispoll) {
2062 			return 1;
2063 		}
2064 
2065 		off_t amount;
2066 		amount = vp->v_un.vu_ubcinfo->ui_size - current_offset;
2067 		if (amount > INT64_MAX) {
2068 			return INT64_MAX;
2069 		} else if (amount < INT64_MIN) {
2070 			return INT64_MIN;
2071 		} else {
2072 			return (int64_t)amount;
2073 		}
2074 	} else {
2075 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2076 		return 0;
2077 	}
2078 }
2079 
2080 /*
2081  * Used for EVFILT_WRITE.
2082  *
2083  * For regular vnodes, we can always write (1).  For named pipes,
2084  * see how much space there is in the buffer.  Nothing else is covered.
2085  */
2086 static intptr_t
vnode_writable_space_count(vnode_t vp)2087 vnode_writable_space_count(vnode_t vp)
2088 {
2089 	if (vnode_isfifo(vp)) {
2090 #if FIFO
2091 		long spc;
2092 		int err = fifo_freespace(vp, &spc);
2093 		if (err == 0) {
2094 			return (intptr_t)spc;
2095 		} else
2096 #endif
2097 		{
2098 			return (intptr_t)0;
2099 		}
2100 	} else if (vnode_isreg(vp)) {
2101 		return (intptr_t)1;
2102 	} else {
2103 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2104 		return 0;
2105 	}
2106 }
2107 
2108 /*
2109  * Determine whether this knote should be active
2110  *
2111  * This is kind of subtle.
2112  *      --First, notice if the vnode has been revoked: in so, override hint
2113  *      --EVFILT_READ knotes are checked no matter what the hint is
2114  *      --Other knotes activate based on hint.
2115  *      --If hint is revoke, set special flags and activate
2116  */
2117 static int
filt_vnode_common(struct knote * kn,struct kevent_qos_s * kev,vnode_t vp,long hint)2118 filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev, vnode_t vp, long hint)
2119 {
2120 	int activate = 0;
2121 	int64_t data = 0;
2122 
2123 	lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
2124 
2125 	/* Special handling for vnodes that are in recycle or already gone */
2126 	if (NOTE_REVOKE == hint) {
2127 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2128 		activate = 1;
2129 
2130 		if ((kn->kn_filter == EVFILT_VNODE) && (kn->kn_sfflags & NOTE_REVOKE)) {
2131 			kn->kn_fflags |= NOTE_REVOKE;
2132 		}
2133 	} else {
2134 		switch (kn->kn_filter) {
2135 		case EVFILT_READ:
2136 			data = vnode_readable_data_count(vp, kn->kn_fp->fp_glob->fg_offset, (kn->kn_flags & EV_POLL));
2137 			activate = (data != 0);
2138 			break;
2139 		case EVFILT_WRITE:
2140 			data = vnode_writable_space_count(vp);
2141 			activate = (data != 0);
2142 			break;
2143 		case EVFILT_VNODE:
2144 			/* Check events this note matches against the hint */
2145 			if (kn->kn_sfflags & hint) {
2146 				kn->kn_fflags |= (uint32_t)hint; /* Set which event occurred */
2147 			}
2148 			activate = (kn->kn_fflags != 0);
2149 			break;
2150 		default:
2151 			panic("Invalid knote filter on a vnode!");
2152 		}
2153 	}
2154 
2155 	if (kev && activate) {
2156 		knote_fill_kevent(kn, kev, data);
2157 	}
2158 
2159 	return activate;
2160 }
2161 
2162 static int
filt_vnode(struct knote * kn,long hint)2163 filt_vnode(struct knote *kn, long hint)
2164 {
2165 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2166 
2167 	return filt_vnode_common(kn, NULL, vp, hint);
2168 }
2169 
2170 static int
filt_vntouch(struct knote * kn,struct kevent_qos_s * kev)2171 filt_vntouch(struct knote *kn, struct kevent_qos_s *kev)
2172 {
2173 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2174 	uint32_t vid = vnode_vid(vp);
2175 	int activate;
2176 	int hint = 0;
2177 
2178 	vnode_lock(vp);
2179 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2180 		/* is recycled */
2181 		hint = NOTE_REVOKE;
2182 	}
2183 
2184 	/* accept new input fflags mask */
2185 	kn->kn_sfflags = kev->fflags;
2186 
2187 	activate = filt_vnode_common(kn, NULL, vp, hint);
2188 
2189 	if (hint == 0) {
2190 		vnode_put_locked(vp);
2191 	}
2192 	vnode_unlock(vp);
2193 
2194 	return activate;
2195 }
2196 
2197 static int
filt_vnprocess(struct knote * kn,struct kevent_qos_s * kev)2198 filt_vnprocess(struct knote *kn, struct kevent_qos_s *kev)
2199 {
2200 	vnode_t vp = (struct vnode *)knote_kn_hook_get_raw(kn);
2201 	uint32_t vid = vnode_vid(vp);
2202 	int activate;
2203 	int hint = 0;
2204 
2205 	vnode_lock(vp);
2206 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2207 		/* Is recycled */
2208 		hint = NOTE_REVOKE;
2209 	}
2210 	activate = filt_vnode_common(kn, kev, vp, hint);
2211 
2212 	/* Definitely need to unlock, may need to put */
2213 	if (hint == 0) {
2214 		vnode_put_locked(vp);
2215 	}
2216 	vnode_unlock(vp);
2217 
2218 	return activate;
2219 }
2220 
2221 struct vniodesc {
2222 	vnode_t         vnio_vnode;     /* associated vnode */
2223 	int             vnio_fflags;    /* cached fileglob flags */
2224 };
2225 
2226 errno_t
vnio_openfd(int fd,vniodesc_t * vniop)2227 vnio_openfd(int fd, vniodesc_t *vniop)
2228 {
2229 	proc_t p = current_proc();
2230 	struct fileproc *fp = NULL;
2231 	vniodesc_t vnio;
2232 	vnode_t vp;
2233 	int error;
2234 	bool need_drop = false;
2235 
2236 	vnio = kalloc_type(struct vniodesc, Z_WAITOK);
2237 
2238 	error = fp_getfvp(p, fd, &fp, &vp);
2239 	if (error) {
2240 		goto out;
2241 	}
2242 	need_drop = true;
2243 
2244 	if ((fp->fp_glob->fg_flag & (FREAD | FWRITE)) == 0) {
2245 		error = EBADF;
2246 		goto out;
2247 	}
2248 
2249 	if (vnode_isswap(vp)) {
2250 		error = EPERM;
2251 		goto out;
2252 	}
2253 
2254 	if (vp->v_type != VREG) {
2255 		error = EFTYPE;
2256 		goto out;
2257 	}
2258 
2259 	error = vnode_getwithref(vp);
2260 	if (error) {
2261 		goto out;
2262 	}
2263 
2264 	vnio->vnio_vnode = vp;
2265 	vnio->vnio_fflags = fp->fp_glob->fg_flag;
2266 
2267 	(void)fp_drop(p, fd, fp, 0);
2268 	need_drop = false;
2269 
2270 	error = vnode_ref_ext(vp, vnio->vnio_fflags, 0);
2271 	if (error == 0) {
2272 		*vniop = vnio;
2273 		vnio = NULL;
2274 	}
2275 
2276 	vnode_put(vp);
2277 
2278 out:
2279 	if (need_drop) {
2280 		fp_drop(p, fd, fp, 0);
2281 	}
2282 	if (vnio != NULL) {
2283 		kfree_type(struct vniodesc, vnio);
2284 	}
2285 	return error;
2286 }
2287 
2288 errno_t
vnio_close(vniodesc_t vnio)2289 vnio_close(vniodesc_t vnio)
2290 {
2291 	int error;
2292 
2293 	/*
2294 	 * The vniodesc is always destroyed, because the close
2295 	 * always "succeeds".  We just return whatever error
2296 	 * might have been encountered while doing so.
2297 	 */
2298 
2299 	error = vnode_getwithref(vnio->vnio_vnode);
2300 	if (error == 0) {
2301 		error = vnode_close(vnio->vnio_vnode, vnio->vnio_fflags, NULL);
2302 	}
2303 
2304 	kfree_type(struct vniodesc, vnio);
2305 
2306 	return error;
2307 }
2308 
2309 errno_t
vnio_read(vniodesc_t vnio,uio_t uio)2310 vnio_read(vniodesc_t vnio, uio_t uio)
2311 {
2312 	vnode_t vp = vnio->vnio_vnode;
2313 	user_ssize_t read_len;
2314 	int error;
2315 
2316 	read_len = uio_resid(uio);
2317 	if (read_len < 0 || read_len > INT_MAX) {
2318 		return EINVAL;
2319 	}
2320 
2321 	error = vnode_getwithref(vp);
2322 	if (error == 0) {
2323 		error = vn_read_common(vp, uio, vnio->vnio_fflags,
2324 		    vfs_context_current());
2325 		vnode_put(vp);
2326 	}
2327 
2328 	return error;
2329 }
2330 
2331 vnode_t
vnio_vnode(vniodesc_t vnio)2332 vnio_vnode(vniodesc_t vnio)
2333 {
2334 	return vnio->vnio_vnode;
2335 }
2336 
2337 int
vnode_isauthfs(vnode_t vp)2338 vnode_isauthfs(vnode_t vp)
2339 {
2340 	fsioc_auth_fs_t afs = { .authvp = NULL };
2341 	int error;
2342 
2343 	error = VNOP_IOCTL(vp, FSIOC_AUTH_FS, (caddr_t)&afs, 0,
2344 	    vfs_context_current());
2345 
2346 	return error == 0 ? 1 : 0;
2347 }
2348