xref: /xnu-8792.61.2/bsd/vfs/vfs_vnops.c (revision 42e220869062b56f8d7d0726fd4c88954f87902c)
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 /* Temporary workaround for ubc.h until <rdar://4714366 is resolved */
91 #define ubc_setcred ubc_setcred_deprecated
92 #include <sys/ubc.h>
93 #undef ubc_setcred
94 int     ubc_setcred(struct vnode *, struct proc *);
95 #include <sys/conf.h>
96 #include <sys/disk.h>
97 #include <sys/fsevents.h>
98 #include <sys/kdebug.h>
99 #include <sys/xattr.h>
100 #include <sys/ubc_internal.h>
101 #include <sys/uio_internal.h>
102 #include <sys/resourcevar.h>
103 #include <sys/signalvar.h>
104 
105 #include <vm/vm_kern.h>
106 #include <vm/vm_map.h>
107 
108 #include <miscfs/specfs/specdev.h>
109 #include <miscfs/fifofs/fifo.h>
110 
111 #if CONFIG_MACF
112 #include <security/mac_framework.h>
113 #endif
114 
115 #include <IOKit/IOBSD.h>
116 #include <libkern/section_keywords.h>
117 
118 static int vn_closefile(struct fileglob *fp, vfs_context_t ctx);
119 static int vn_ioctl(struct fileproc *fp, u_long com, caddr_t data,
120     vfs_context_t ctx);
121 static int vn_read(struct fileproc *fp, struct uio *uio, int flags,
122     vfs_context_t ctx);
123 static int vn_write(struct fileproc *fp, struct uio *uio, int flags,
124     vfs_context_t ctx);
125 static int vn_select( struct fileproc *fp, int which, void * wql,
126     vfs_context_t ctx);
127 static int vn_kqfilter(struct fileproc *fp, struct knote *kn,
128     struct kevent_qos_s *kev);
129 static void filt_vndetach(struct knote *kn);
130 static int filt_vnode(struct knote *kn, long hint);
131 static int filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev,
132     vnode_t vp, long hint);
133 static int vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx);
134 
135 const struct fileops vnops = {
136 	.fo_type     = DTYPE_VNODE,
137 	.fo_read     = vn_read,
138 	.fo_write    = vn_write,
139 	.fo_ioctl    = vn_ioctl,
140 	.fo_select   = vn_select,
141 	.fo_close    = vn_closefile,
142 	.fo_drain    = fo_no_drain,
143 	.fo_kqfilter = vn_kqfilter,
144 };
145 
146 static int filt_vntouch(struct knote *kn, struct kevent_qos_s *kev);
147 static int filt_vnprocess(struct knote *kn, struct kevent_qos_s*kev);
148 
149 SECURITY_READ_ONLY_EARLY(struct  filterops) vnode_filtops = {
150 	.f_isfd = 1,
151 	.f_attach = NULL,
152 	.f_detach = filt_vndetach,
153 	.f_event = filt_vnode,
154 	.f_touch = filt_vntouch,
155 	.f_process = filt_vnprocess,
156 };
157 
158 /*
159  * Common code for vnode open operations.
160  * Check permissions, and call the VNOP_OPEN or VNOP_CREATE routine.
161  *
162  * XXX the profusion of interfaces here is probably a bad thing.
163  */
164 int
vn_open(struct nameidata * ndp,int fmode,int cmode)165 vn_open(struct nameidata *ndp, int fmode, int cmode)
166 {
167 	return vn_open_modflags(ndp, &fmode, cmode);
168 }
169 
170 int
vn_open_modflags(struct nameidata * ndp,int * fmodep,int cmode)171 vn_open_modflags(struct nameidata *ndp, int *fmodep, int cmode)
172 {
173 	int error;
174 	struct vnode_attr *vap;
175 
176 	vap = kalloc_type(struct vnode_attr, Z_WAITOK);
177 
178 	VATTR_INIT(vap);
179 	VATTR_SET(vap, va_mode, (mode_t)cmode);
180 
181 	error = vn_open_auth(ndp, fmodep, vap, NULLVP);
182 
183 	kfree_type(struct vnode_attr, vap);
184 
185 	return error;
186 }
187 
188 static int
vn_open_auth_finish(vnode_t vp,int fmode,vfs_context_t ctx)189 vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx)
190 {
191 	int error;
192 
193 	if ((error = vnode_ref_ext(vp, fmode, 0)) != 0) {
194 		goto bad;
195 	}
196 
197 	/* Call out to allow 3rd party notification of open.
198 	 * Ignore result of kauth_authorize_fileop call.
199 	 */
200 #if CONFIG_MACF
201 	mac_vnode_notify_open(ctx, vp, fmode);
202 #endif
203 	kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
204 	    (uintptr_t)vp, 0);
205 
206 	return 0;
207 
208 bad:
209 	return error;
210 }
211 
212 /*
213  * May do nameidone() to allow safely adding an FSEvent.  Cue off of ni_dvp to
214  * determine whether that has happened.
215  */
216 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)217 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)
218 {
219 	uint32_t status = 0;
220 	vnode_t dvp = ndp->ni_dvp;
221 	int batched;
222 	int error;
223 	vnode_t vp;
224 
225 	batched = vnode_compound_open_available(ndp->ni_dvp);
226 	*did_open = FALSE;
227 
228 	VATTR_SET(vap, va_type, VREG);
229 	if (fmode & O_EXCL) {
230 		vap->va_vaflags |= VA_EXCLUSIVE;
231 	}
232 
233 #if NAMEDRSRCFORK
234 	if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK) {
235 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
236 			goto out;
237 		}
238 		if ((error = vnode_makenamedstream(dvp, &ndp->ni_vp, XATTR_RESOURCEFORK_NAME, 0, ctx)) != 0) {
239 			goto out;
240 		}
241 		*did_create = TRUE;
242 	} else {
243 #endif
244 	if (!batched) {
245 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
246 			goto out;
247 		}
248 	}
249 
250 	error = vn_create(dvp, &ndp->ni_vp, ndp, vap, VN_CREATE_DOOPEN, fmode, &status, ctx);
251 	if (error != 0) {
252 		if (batched) {
253 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? TRUE : FALSE;
254 		} else {
255 			*did_create = FALSE;
256 		}
257 
258 		if (error == EKEEPLOOKING) {
259 			if (*did_create) {
260 				panic("EKEEPLOOKING, but we did a create?");
261 			}
262 			if (!batched) {
263 				panic("EKEEPLOOKING from filesystem that doesn't support compound vnops?");
264 			}
265 			if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
266 				panic("EKEEPLOOKING, but continue flag not set?");
267 			}
268 
269 			/*
270 			 * Do NOT drop the dvp: we need everything to continue the lookup.
271 			 */
272 			return error;
273 		}
274 	} else {
275 		if (batched) {
276 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? 1 : 0;
277 			*did_open = TRUE;
278 		} else {
279 			*did_create = TRUE;
280 		}
281 	}
282 #if NAMEDRSRCFORK
283 }
284 #endif
285 
286 	vp = ndp->ni_vp;
287 
288 	if (*did_create) {
289 		int     update_flags = 0;
290 
291 		// Make sure the name & parent pointers are hooked up
292 		if (vp->v_name == NULL) {
293 			update_flags |= VNODE_UPDATE_NAME;
294 		}
295 		if (vp->v_parent == NULLVP) {
296 			update_flags |= VNODE_UPDATE_PARENT;
297 		}
298 
299 		if (update_flags) {
300 			vnode_update_identity(vp, dvp, ndp->ni_cnd.cn_nameptr, ndp->ni_cnd.cn_namelen, ndp->ni_cnd.cn_hash, update_flags);
301 		}
302 
303 		vnode_put(dvp);
304 		ndp->ni_dvp = NULLVP;
305 
306 #if CONFIG_FSE
307 		if (need_fsevent(FSE_CREATE_FILE, vp)) {
308 			add_fsevent(FSE_CREATE_FILE, ctx,
309 			    FSE_ARG_VNODE, vp,
310 			    FSE_ARG_DONE);
311 		}
312 #endif
313 	}
314 out:
315 	if (ndp->ni_dvp != NULLVP) {
316 		vnode_put(dvp);
317 		ndp->ni_dvp = NULLVP;
318 	}
319 
320 	return error;
321 }
322 
323 /*
324  * This is the number of times we'll loop in vn_open_auth without explicitly
325  * yielding the CPU when we determine we have to retry.
326  */
327 #define RETRY_NO_YIELD_COUNT    5
328 
329 /*
330  * Open a file with authorization, updating the contents of the structures
331  * pointed to by ndp, fmodep, and vap as necessary to perform the requested
332  * operation.  This function is used for both opens of existing files, and
333  * creation of new files.
334  *
335  * Parameters:	ndp			The nami data pointer describing the
336  *					file
337  *		fmodep			A pointer to an int containg the mode
338  *					information to be used for the open
339  *		vap			A pointer to the vnode attribute
340  *					descriptor to be used for the open
341  *		authvp		If non-null and VA_DP_AUTHENTICATE is set in vap,
342  *					have a supporting filesystem verify that the file
343  *					to be opened is on the same volume as authvp and
344  *					that authvp is on an authenticated volume
345  *
346  * Indirect:	*			Contents of the data structures pointed
347  *					to by the parameters are modified as
348  *					necessary to the requested operation.
349  *
350  * Returns:	0			Success
351  *		!0			errno value
352  *
353  * Notes:	The kauth_filesec_t in 'vap', if any, is in host byte order.
354  *
355  *		The contents of '*ndp' will be modified, based on the other
356  *		arguments to this function, and to return file and directory
357  *		data necessary to satisfy the requested operation.
358  *
359  *		If the file does not exist and we are creating it, then the
360  *		O_TRUNC flag will be cleared in '*fmodep' to indicate to the
361  *		caller that the file was not truncated.
362  *
363  *		If the file exists and the O_EXCL flag was not specified, then
364  *		the O_CREAT flag will be cleared in '*fmodep' to indicate to
365  *		the caller that the existing file was merely opened rather
366  *		than created.
367  *
368  *		The contents of '*vap' will be modified as necessary to
369  *		complete the operation, including setting of supported
370  *		attribute, clearing of fields containing unsupported attributes
371  *		in the request, if the request proceeds without them, etc..
372  *
373  * XXX:		This function is too complicated in actings on its arguments
374  *
375  * XXX:		We should enummerate the possible errno values here, and where
376  *		in the code they originated.
377  */
378 int
vn_open_auth(struct nameidata * ndp,int * fmodep,struct vnode_attr * vap,vnode_t authvp)379 vn_open_auth(struct nameidata *ndp, int *fmodep, struct vnode_attr *vap, vnode_t authvp)
380 {
381 	struct vnode *vp;
382 	struct vnode *dvp;
383 	vfs_context_t ctx = ndp->ni_cnd.cn_context;
384 	int error;
385 	int fmode;
386 	uint32_t origcnflags;
387 	boolean_t did_create;
388 	boolean_t did_open;
389 	boolean_t need_vnop_open;
390 	boolean_t batched;
391 	boolean_t ref_failed;
392 	int nretries = 0;
393 
394 again:
395 	vp = NULL;
396 	dvp = NULL;
397 	batched = FALSE;
398 	did_create = FALSE;
399 	need_vnop_open = TRUE;
400 	ref_failed = FALSE;
401 	fmode = *fmodep;
402 	origcnflags = ndp->ni_cnd.cn_flags;
403 
404 	if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)) {
405 		if ((authvp != NULLVP)
406 		    && !ISSET(vap->va_dataprotect_flags, VA_DP_AUTHENTICATE)) {
407 			return EINVAL;
408 		}
409 		// If raw encrypted mode is requested, handle that here
410 		if (ISSET(vap->va_dataprotect_flags, VA_DP_RAWENCRYPTED)) {
411 			fmode |= FENCRYPTED;
412 		}
413 	}
414 
415 	if ((fmode & O_NOFOLLOW_ANY) && (fmode & (O_SYMLINK | O_NOFOLLOW))) {
416 		error = EINVAL;
417 		goto out;
418 	}
419 
420 	/*
421 	 * O_CREAT
422 	 */
423 	if (fmode & O_CREAT) {
424 		if ((fmode & O_DIRECTORY)) {
425 			error = EINVAL;
426 			goto out;
427 		}
428 		ndp->ni_cnd.cn_nameiop = CREATE;
429 #if CONFIG_TRIGGERS
430 		ndp->ni_op = OP_LINK;
431 #endif
432 		/* Inherit USEDVP, vnode_open() supported flags only */
433 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
434 		ndp->ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF | AUDITVNPATH1;
435 		ndp->ni_flag = NAMEI_COMPOUNDOPEN;
436 #if NAMEDRSRCFORK
437 		/* open calls are allowed for resource forks. */
438 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
439 #endif
440 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0 && (origcnflags & FOLLOW) != 0) {
441 			ndp->ni_cnd.cn_flags |= FOLLOW;
442 		}
443 		if (fmode & O_NOFOLLOW_ANY) {
444 			/* will return ELOOP on the first symlink to be hit */
445 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
446 		}
447 
448 continue_create_lookup:
449 		if ((error = namei(ndp))) {
450 			goto out;
451 		}
452 
453 		dvp = ndp->ni_dvp;
454 		vp = ndp->ni_vp;
455 
456 		batched = vnode_compound_open_available(dvp);
457 
458 		/* not found, create */
459 		if (vp == NULL) {
460 			/* must have attributes for a new file */
461 			if (vap == NULL) {
462 				vnode_put(dvp);
463 				error = EINVAL;
464 				goto out;
465 			}
466 			/*
467 			 * Attempt a create.   For a system supporting compound VNOPs, we may
468 			 * find an existing file or create one; in either case, we will already
469 			 * have the file open and no VNOP_OPEN() will be needed.
470 			 */
471 			error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
472 
473 			dvp = ndp->ni_dvp;
474 			vp = ndp->ni_vp;
475 
476 			/*
477 			 * Detected a node that the filesystem couldn't handle.  Don't call
478 			 * nameidone() yet, because we need that path buffer.
479 			 */
480 			if (error == EKEEPLOOKING) {
481 				if (!batched) {
482 					panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
483 				}
484 				goto continue_create_lookup;
485 			}
486 
487 			nameidone(ndp);
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 					goto again;
501 				}
502 				goto bad;
503 			}
504 
505 			need_vnop_open = !did_open;
506 		} else {
507 			if (fmode & O_EXCL) {
508 				error = EEXIST;
509 			}
510 
511 			/*
512 			 * We have a vnode.  Use compound open if available
513 			 * or else fall through to "traditional" path.  Note: can't
514 			 * do a compound open for root, because the parent belongs
515 			 * to a different FS.
516 			 */
517 			if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
518 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
519 
520 				if (error == 0) {
521 					vp = ndp->ni_vp;
522 					need_vnop_open = FALSE;
523 				} else if (error == EKEEPLOOKING) {
524 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
525 						panic("EKEEPLOOKING, but continue flag not set?");
526 					}
527 					goto continue_create_lookup;
528 				}
529 			}
530 			nameidone(ndp);
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 		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.
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_t uio_buf[UIO_SIZEOF(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 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1270 		ioflag |= IO_NOCACHE;
1271 	}
1272 	if (fp->fp_glob->fg_flag & FNODIRECT) {
1273 		ioflag |= IO_NODIRECT;
1274 	}
1275 	if (fp->fp_glob->fg_flag & FSINGLE_WRITER) {
1276 		ioflag |= IO_SINGLE_WRITER;
1277 	}
1278 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1279 		ioflag |= IO_EVTONLY;
1280 	}
1281 
1282 	/*
1283 	 * Treat synchronous mounts and O_FSYNC on the fd as equivalent.
1284 	 *
1285 	 * XXX We treat O_DSYNC as O_FSYNC for now, since we can not delay
1286 	 * XXX the non-essential metadata without some additional VFS work;
1287 	 * XXX the intent at this point is to plumb the interface for it.
1288 	 */
1289 	if ((fp->fp_glob->fg_flag & (O_FSYNC | O_DSYNC)) ||
1290 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) {
1291 		ioflag |= IO_SYNC;
1292 	}
1293 
1294 	if ((flags & FOF_OFFSET) == 0) {
1295 		write_offset = fp->fp_glob->fg_offset;
1296 		uio_setoffset(uio, write_offset);
1297 	} else {
1298 		/* for pwrite, append should  be ignored */
1299 		ioflag &= ~IO_APPEND;
1300 		write_offset = uio_offset(uio);
1301 		/* POSIX allows negative offsets for character devices. */
1302 		if ((write_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1303 			error = EINVAL;
1304 			goto error_out;
1305 		}
1306 	}
1307 
1308 	if (write_offset == INT64_MAX) {
1309 		/* writes are not possible */
1310 		error = EFBIG;
1311 		goto error_out;
1312 	}
1313 
1314 	/*
1315 	 * write_len is the original write length that was requested.
1316 	 * We may however need to reduce that becasue of two reasons
1317 	 *
1318 	 * 1) If write_offset + write_len will exceed OFF_T_MAX (i.e. INT64_MAX)
1319 	 *    and/or
1320 	 * 2) If write_offset + write_len will exceed the administrative
1321 	 *    limit for the maximum file size.
1322 	 *
1323 	 * In both cases the write will be denied if we can't write even a single
1324 	 * byte otherwise it will be "clipped" (i.e. a short write).
1325 	 */
1326 
1327 	/*
1328 	 * If offset + len will cause overflow, reduce the len
1329 	 * to a value (adjusted_write_len) where it won't
1330 	 */
1331 	if ((write_offset >= 0) && (INT64_MAX - write_offset) < write_len) {
1332 		/*
1333 		 * 0                                          write_offset  INT64_MAX
1334 		 * |-----------------------------------------------|----------|~~~
1335 		 *                                                 <--write_len-->
1336 		 *                                                 <-adjusted->
1337 		 */
1338 		adjusted_write_len = (user_ssize_t)(INT64_MAX - write_offset);
1339 	}
1340 
1341 	/* write_end_offset will always be [0, INT64_MAX] */
1342 	write_end_offset = write_offset + adjusted_write_len;
1343 
1344 	if (p && (vp->v_type == VREG) &&
1345 	    (rlim_cur_fsize != RLIM_INFINITY) &&
1346 	    (rlim_cur_fsize <= INT64_MAX) &&
1347 	    (write_end_offset > (off_t)rlim_cur_fsize)) {
1348 		/*
1349 		 * If the requested residual would cause us to go past the
1350 		 * administrative limit, then we need to adjust the residual
1351 		 * down to cause fewer bytes than requested to be written.  If
1352 		 * we can't do that (e.g. the residual is already 1 byte),
1353 		 * then we fail the write with EFBIG.
1354 		 */
1355 		if (write_offset >= (off_t)rlim_cur_fsize) {
1356 			/*
1357 			 * 0                  rlim_fsize  write_offset  write_end  INT64_MAX
1358 			 * |------------------------|----------|-------------|--------|
1359 			 *                                     <--write_len-->
1360 			 *
1361 			 *                write not permitted
1362 			 */
1363 			psignal(p, SIGXFSZ);
1364 			error = EFBIG;
1365 			goto error_out;
1366 		}
1367 
1368 		/*
1369 		 * 0                  write_offset rlim_fsize  write_end  INT64_MAX
1370 		 * |------------------------|-----------|---------|------------|
1371 		 *                          <------write_len------>
1372 		 *                          <-adjusted-->
1373 		 */
1374 		adjusted_write_len =  (user_ssize_t)((off_t)rlim_cur_fsize - write_offset);
1375 		assert((adjusted_write_len > 0) && (adjusted_write_len < write_len));
1376 	}
1377 
1378 	if (adjusted_write_len < write_len) {
1379 		uio_setresid(uio, adjusted_write_len);
1380 		clippedsize = write_len - adjusted_write_len;
1381 	}
1382 
1383 	error = VNOP_WRITE(vp, uio, ioflag, ctx);
1384 
1385 	/*
1386 	 * If we had to reduce the size of write requested either because
1387 	 * of rlimit or because it would have exceeded
1388 	 * maximum file size, we have to add that back to the residual so
1389 	 * it correctly reflects what we did in this function.
1390 	 */
1391 	if (clippedsize) {
1392 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1393 	}
1394 
1395 	if ((flags & FOF_OFFSET) == 0) {
1396 		if (ioflag & IO_APPEND) {
1397 			fp->fp_glob->fg_offset = uio_offset(uio);
1398 		} else {
1399 			fp->fp_glob->fg_offset += (write_len - uio_resid(uio));
1400 		}
1401 		if (offset_locked) {
1402 			vn_offset_unlock(fp->fp_glob);
1403 			offset_locked = false;
1404 		}
1405 	}
1406 
1407 	/*
1408 	 * Set the credentials on successful writes
1409 	 */
1410 	if ((error == 0) && (vp->v_tag == VT_NFS) && (UBCINFOEXISTS(vp))) {
1411 		/*
1412 		 * When called from aio subsystem, we only have the proc from
1413 		 * which to get the credential, at this point, so use that
1414 		 * instead.  This means aio functions are incompatible with
1415 		 * per-thread credentials (aio operations are proxied).  We
1416 		 * can't easily correct the aio vs. settid race in this case
1417 		 * anyway, so we disallow it.
1418 		 */
1419 		if ((flags & FOF_PCRED) == 0) {
1420 			ubc_setthreadcred(vp, p, current_thread());
1421 		} else {
1422 			ubc_setcred(vp, p);
1423 		}
1424 	}
1425 
1426 #if CONFIG_FILE_LEASES
1427 	/*
1428 	 * On success, break the parent dir lease as the file's attributes (size
1429 	 * and/or mtime) have changed. Best attempt to break lease, just drop the
1430 	 * the error upon failure as there is no point to return error when the
1431 	 * write has completed successfully.
1432 	 */
1433 	if (__probable(error == 0)) {
1434 		vnode_breakdirlease(vp, true, O_WRONLY);
1435 	}
1436 #endif /* CONFIG_FILE_LEASES */
1437 
1438 	(void)vnode_put(vp);
1439 	return error;
1440 
1441 error_out:
1442 	if (offset_locked) {
1443 		vn_offset_unlock(fp->fp_glob);
1444 	}
1445 	(void)vnode_put(vp);
1446 	return error;
1447 }
1448 
1449 /*
1450  * File table vnode stat routine.
1451  *
1452  * Returns:	0			Success
1453  *		EBADF
1454  *		ENOMEM
1455  *	vnode_getattr:???
1456  */
1457 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)1458 vn_stat_noauth(struct vnode *vp, void *sbptr, kauth_filesec_t *xsec, int isstat64,
1459     int needsrealdev, vfs_context_t ctx, struct ucred *file_cred)
1460 {
1461 	struct vnode_attr va;
1462 	int error;
1463 	u_short mode;
1464 	kauth_filesec_t fsec;
1465 	struct stat *sb = (struct stat *)0;     /* warning avoidance ; protected by isstat64 */
1466 	struct stat64 * sb64 = (struct stat64 *)0;  /* warning avoidance ; protected by isstat64 */
1467 
1468 	if (isstat64 != 0) {
1469 		sb64 = (struct stat64 *)sbptr;
1470 	} else {
1471 		sb = (struct stat *)sbptr;
1472 	}
1473 	memset(&va, 0, sizeof(va));
1474 	VATTR_INIT(&va);
1475 	VATTR_WANTED(&va, va_fsid);
1476 	VATTR_WANTED(&va, va_fileid);
1477 	VATTR_WANTED(&va, va_mode);
1478 	VATTR_WANTED(&va, va_type);
1479 	VATTR_WANTED(&va, va_nlink);
1480 	VATTR_WANTED(&va, va_uid);
1481 	VATTR_WANTED(&va, va_gid);
1482 	VATTR_WANTED(&va, va_rdev);
1483 	VATTR_WANTED(&va, va_data_size);
1484 	VATTR_WANTED(&va, va_access_time);
1485 	VATTR_WANTED(&va, va_modify_time);
1486 	VATTR_WANTED(&va, va_change_time);
1487 	VATTR_WANTED(&va, va_create_time);
1488 	VATTR_WANTED(&va, va_flags);
1489 	VATTR_WANTED(&va, va_gen);
1490 	VATTR_WANTED(&va, va_iosize);
1491 	/* lower layers will synthesise va_total_alloc from va_data_size if required */
1492 	VATTR_WANTED(&va, va_total_alloc);
1493 	if (xsec != NULL) {
1494 		VATTR_WANTED(&va, va_uuuid);
1495 		VATTR_WANTED(&va, va_guuid);
1496 		VATTR_WANTED(&va, va_acl);
1497 	}
1498 	if (needsrealdev) {
1499 		va.va_vaflags = VA_REALFSID;
1500 	}
1501 	error = vnode_getattr(vp, &va, ctx);
1502 	if (error) {
1503 		goto out;
1504 	}
1505 #if CONFIG_MACF
1506 	/*
1507 	 * Give MAC polices a chance to reject or filter the attributes
1508 	 * returned by the filesystem.  Note that MAC policies are consulted
1509 	 * *after* calling the filesystem because filesystems can return more
1510 	 * attributes than were requested so policies wouldn't be authoritative
1511 	 * is consulted beforehand.  This also gives policies an opportunity
1512 	 * to change the values of attributes retrieved.
1513 	 */
1514 	error = mac_vnode_check_getattr(ctx, file_cred, vp, &va);
1515 	if (error) {
1516 		goto out;
1517 	}
1518 #endif
1519 	/*
1520 	 * Copy from vattr table
1521 	 */
1522 	if (isstat64 != 0) {
1523 		sb64->st_dev = va.va_fsid;
1524 		sb64->st_ino = (ino64_t)va.va_fileid;
1525 	} else {
1526 		sb->st_dev = va.va_fsid;
1527 		sb->st_ino = (ino_t)va.va_fileid;
1528 	}
1529 	mode = va.va_mode;
1530 	switch (vp->v_type) {
1531 	case VREG:
1532 		mode |= S_IFREG;
1533 		break;
1534 	case VDIR:
1535 		mode |= S_IFDIR;
1536 		break;
1537 	case VBLK:
1538 		mode |= S_IFBLK;
1539 		break;
1540 	case VCHR:
1541 		mode |= S_IFCHR;
1542 		break;
1543 	case VLNK:
1544 		mode |= S_IFLNK;
1545 		break;
1546 	case VSOCK:
1547 		mode |= S_IFSOCK;
1548 		break;
1549 	case VFIFO:
1550 		mode |= S_IFIFO;
1551 		break;
1552 	default:
1553 		error = EBADF;
1554 		goto out;
1555 	}
1556 	;
1557 	if (isstat64 != 0) {
1558 		sb64->st_mode = mode;
1559 		sb64->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1560 		sb64->st_uid = va.va_uid;
1561 		sb64->st_gid = va.va_gid;
1562 		sb64->st_rdev = va.va_rdev;
1563 		sb64->st_size = va.va_data_size;
1564 		sb64->st_atimespec = va.va_access_time;
1565 		sb64->st_mtimespec = va.va_modify_time;
1566 		sb64->st_ctimespec = va.va_change_time;
1567 		if (VATTR_IS_SUPPORTED(&va, va_create_time)) {
1568 			sb64->st_birthtimespec =  va.va_create_time;
1569 		} else {
1570 			sb64->st_birthtimespec.tv_sec = sb64->st_birthtimespec.tv_nsec = 0;
1571 		}
1572 		sb64->st_blksize = va.va_iosize;
1573 		sb64->st_flags = va.va_flags;
1574 		sb64->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1575 	} else {
1576 		sb->st_mode = mode;
1577 		sb->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1578 		sb->st_uid = va.va_uid;
1579 		sb->st_gid = va.va_gid;
1580 		sb->st_rdev = va.va_rdev;
1581 		sb->st_size = va.va_data_size;
1582 		sb->st_atimespec = va.va_access_time;
1583 		sb->st_mtimespec = va.va_modify_time;
1584 		sb->st_ctimespec = va.va_change_time;
1585 		sb->st_blksize = va.va_iosize;
1586 		sb->st_flags = va.va_flags;
1587 		sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1588 	}
1589 
1590 	/* if we're interested in extended security data and we got an ACL */
1591 	if (xsec != NULL) {
1592 		if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
1593 		    !VATTR_IS_SUPPORTED(&va, va_uuuid) &&
1594 		    !VATTR_IS_SUPPORTED(&va, va_guuid)) {
1595 			*xsec = KAUTH_FILESEC_NONE;
1596 		} else {
1597 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1598 				fsec = kauth_filesec_alloc(va.va_acl->acl_entrycount);
1599 			} else {
1600 				fsec = kauth_filesec_alloc(0);
1601 			}
1602 			if (fsec == NULL) {
1603 				error = ENOMEM;
1604 				goto out;
1605 			}
1606 			fsec->fsec_magic = KAUTH_FILESEC_MAGIC;
1607 			if (VATTR_IS_SUPPORTED(&va, va_uuuid)) {
1608 				fsec->fsec_owner = va.va_uuuid;
1609 			} else {
1610 				fsec->fsec_owner = kauth_null_guid;
1611 			}
1612 			if (VATTR_IS_SUPPORTED(&va, va_guuid)) {
1613 				fsec->fsec_group = va.va_guuid;
1614 			} else {
1615 				fsec->fsec_group = kauth_null_guid;
1616 			}
1617 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1618 				__nochk_bcopy(va.va_acl, &(fsec->fsec_acl), KAUTH_ACL_COPYSIZE(va.va_acl));
1619 			} else {
1620 				fsec->fsec_acl.acl_entrycount = KAUTH_FILESEC_NOACL;
1621 			}
1622 			*xsec = fsec;
1623 		}
1624 	}
1625 
1626 	/* Do not give the generation number out to unpriviledged users */
1627 	if (va.va_gen && !vfs_context_issuser(ctx)) {
1628 		if (isstat64 != 0) {
1629 			sb64->st_gen = 0;
1630 		} else {
1631 			sb->st_gen = 0;
1632 		}
1633 	} else {
1634 		if (isstat64 != 0) {
1635 			sb64->st_gen = va.va_gen;
1636 		} else {
1637 			sb->st_gen = va.va_gen;
1638 		}
1639 	}
1640 
1641 	error = 0;
1642 out:
1643 	if (VATTR_IS_SUPPORTED(&va, va_acl) && va.va_acl != NULL) {
1644 		kauth_acl_free(va.va_acl);
1645 	}
1646 	return error;
1647 }
1648 
1649 int
vn_stat(struct vnode * vp,void * sb,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx)1650 vn_stat(struct vnode *vp, void *sb, kauth_filesec_t *xsec, int isstat64, int needsrealdev, vfs_context_t ctx)
1651 {
1652 	int error;
1653 
1654 #if CONFIG_MACF
1655 	error = mac_vnode_check_stat(ctx, NOCRED, vp);
1656 	if (error) {
1657 		return error;
1658 	}
1659 #endif
1660 
1661 	/* authorize */
1662 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_ATTRIBUTES | KAUTH_VNODE_READ_SECURITY, ctx)) != 0) {
1663 		return error;
1664 	}
1665 
1666 	/* actual stat */
1667 	return vn_stat_noauth(vp, sb, xsec, isstat64, needsrealdev, ctx, NOCRED);
1668 }
1669 
1670 
1671 /*
1672  * File table vnode ioctl routine.
1673  */
1674 static int
vn_ioctl(struct fileproc * fp,u_long com,caddr_t data,vfs_context_t ctx)1675 vn_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
1676 {
1677 	struct vnode *vp = (struct vnode *)fp_get_data(fp);
1678 	off_t file_size;
1679 	int error;
1680 	struct vnode *ttyvp;
1681 
1682 	if ((error = vnode_getwithref(vp))) {
1683 		return error;
1684 	}
1685 
1686 #if CONFIG_MACF
1687 	error = mac_vnode_check_ioctl(ctx, vp, com);
1688 	if (error) {
1689 		goto out;
1690 	}
1691 #endif
1692 
1693 	switch (vp->v_type) {
1694 	case VREG:
1695 	case VDIR:
1696 		if (com == FIONREAD) {
1697 			off_t temp_nbytes;
1698 			if ((error = vnode_size(vp, &file_size, ctx)) != 0) {
1699 				goto out;
1700 			}
1701 			temp_nbytes = file_size - fp->fp_glob->fg_offset;
1702 			if (temp_nbytes > INT_MAX) {
1703 				*(int *)data = INT_MAX;
1704 			} else if (temp_nbytes < 0) {
1705 				*(int *)data = 0;
1706 			} else {
1707 				*(int *)data = (int)temp_nbytes;
1708 			}
1709 			goto out;
1710 		}
1711 		if (com == FIONBIO || com == FIOASYNC) {        /* XXX */
1712 			goto out;
1713 		}
1714 		OS_FALLTHROUGH;
1715 
1716 	default:
1717 		error = ENOTTY;
1718 		goto out;
1719 
1720 	case VFIFO:
1721 	case VCHR:
1722 	case VBLK:
1723 
1724 		if (com == TIOCREVOKE || com == TIOCREVOKECLEAR) {
1725 			error = ENOTTY;
1726 			goto out;
1727 		}
1728 
1729 		/* Should not be able to set block size from user space */
1730 		if (com == DKIOCSETBLOCKSIZE) {
1731 			error = EPERM;
1732 			goto out;
1733 		}
1734 
1735 		/* Should not be able to check if filesystem is authenticated from user space */
1736 		if (com == FSIOC_AUTH_FS) {
1737 			error = ENOTTY;
1738 			goto out;
1739 		}
1740 
1741 		if (com == FIODTYPE) {
1742 			if (vp->v_type == VBLK) {
1743 				if (major(vp->v_rdev) >= nblkdev) {
1744 					error = ENXIO;
1745 					goto out;
1746 				}
1747 				*(int *)data = bdevsw[major(vp->v_rdev)].d_type;
1748 			} else if (vp->v_type == VCHR) {
1749 				if (major(vp->v_rdev) >= nchrdev) {
1750 					error = ENXIO;
1751 					goto out;
1752 				}
1753 				*(int *)data = cdevsw[major(vp->v_rdev)].d_type;
1754 			} else {
1755 				error = ENOTTY;
1756 				goto out;
1757 			}
1758 			goto out;
1759 		}
1760 		error = VNOP_IOCTL(vp, com, data, fp->fp_glob->fg_flag, ctx);
1761 
1762 		if (error == 0 && com == TIOCSCTTY) {
1763 			struct session *sessp;
1764 			struct pgrp *pg;
1765 
1766 			pg = proc_pgrp(vfs_context_proc(ctx), &sessp);
1767 
1768 			session_lock(sessp);
1769 			ttyvp = sessp->s_ttyvp;
1770 			sessp->s_ttyvp = vp;
1771 			sessp->s_ttyvid = vnode_vid(vp);
1772 			session_unlock(sessp);
1773 
1774 			pgrp_rele(pg);
1775 		}
1776 	}
1777 out:
1778 	(void)vnode_put(vp);
1779 	return error;
1780 }
1781 
1782 /*
1783  * File table vnode select routine.
1784  */
1785 static int
vn_select(struct fileproc * fp,int which,void * wql,__unused vfs_context_t ctx)1786 vn_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx)
1787 {
1788 	int error;
1789 	struct vnode * vp = (struct vnode *)fp_get_data(fp);
1790 	struct vfs_context context;
1791 
1792 	if ((error = vnode_getwithref(vp)) == 0) {
1793 		context.vc_thread = current_thread();
1794 		context.vc_ucred = fp->fp_glob->fg_cred;
1795 
1796 #if CONFIG_MACF
1797 		/*
1798 		 * XXX We should use a per thread credential here; minimally,
1799 		 * XXX the process credential should have a persistent
1800 		 * XXX reference on it before being passed in here.
1801 		 */
1802 		error = mac_vnode_check_select(ctx, vp, which);
1803 		if (error == 0)
1804 #endif
1805 		error = VNOP_SELECT(vp, which, fp->fp_glob->fg_flag, wql, ctx);
1806 
1807 		(void)vnode_put(vp);
1808 	}
1809 	return error;
1810 }
1811 
1812 /*
1813  * File table vnode close routine.
1814  */
1815 static int
vn_closefile(struct fileglob * fg,vfs_context_t ctx)1816 vn_closefile(struct fileglob *fg, vfs_context_t ctx)
1817 {
1818 	struct vnode *vp = fg_get_data(fg);
1819 	int error;
1820 
1821 	if ((error = vnode_getwithref(vp)) == 0) {
1822 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE &&
1823 		    ((fg->fg_flag & FWASLOCKED) != 0 ||
1824 		    (fg->fg_lflags & FG_HAS_OFDLOCK) != 0)) {
1825 			struct flock lf = {
1826 				.l_whence = SEEK_SET,
1827 				.l_start = 0,
1828 				.l_len = 0,
1829 				.l_type = F_UNLCK
1830 			};
1831 
1832 			if ((fg->fg_flag & FWASLOCKED) != 0) {
1833 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1834 				    F_UNLCK, &lf, F_FLOCK, ctx, NULL);
1835 			}
1836 
1837 			if ((fg->fg_lflags & FG_HAS_OFDLOCK) != 0) {
1838 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1839 				    F_UNLCK, &lf, F_OFD_LOCK, ctx, NULL);
1840 			}
1841 		}
1842 #if CONFIG_FILE_LEASES
1843 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE && !LIST_EMPTY(&vp->v_leases)) {
1844 			/* Expected open count doesn't matter for release. */
1845 			(void)vnode_setlease(vp, fg, F_UNLCK, 0, ctx);
1846 		}
1847 #endif
1848 		error = vn_close(vp, fg->fg_flag, ctx);
1849 		(void) vnode_put(vp);
1850 	}
1851 	return error;
1852 }
1853 
1854 /*
1855  * Returns:	0			Success
1856  *	VNOP_PATHCONF:???
1857  */
1858 int
vn_pathconf(vnode_t vp,int name,int32_t * retval,vfs_context_t ctx)1859 vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
1860 {
1861 	int     error = 0;
1862 	struct vfs_attr vfa;
1863 
1864 	switch (name) {
1865 	case _PC_EXTENDED_SECURITY_NP:
1866 		*retval = vfs_extendedsecurity(vnode_mount(vp)) ? 1 : 0;
1867 		break;
1868 	case _PC_AUTH_OPAQUE_NP:
1869 		*retval = vfs_authopaque(vnode_mount(vp));
1870 		break;
1871 	case _PC_2_SYMLINKS:
1872 		*retval = 1;    /* XXX NOTSUP on MSDOS, etc. */
1873 		break;
1874 	case _PC_ALLOC_SIZE_MIN:
1875 		*retval = 1;    /* XXX lie: 1 byte */
1876 		break;
1877 	case _PC_ASYNC_IO:      /* unistd.h: _POSIX_ASYNCHRONUS_IO */
1878 		*retval = 1;    /* [AIO] option is supported */
1879 		break;
1880 	case _PC_PRIO_IO:       /* unistd.h: _POSIX_PRIORITIZED_IO */
1881 		*retval = 0;    /* [PIO] option is not supported */
1882 		break;
1883 	case _PC_REC_INCR_XFER_SIZE:
1884 		*retval = 4096; /* XXX go from MIN to MAX 4K at a time */
1885 		break;
1886 	case _PC_REC_MIN_XFER_SIZE:
1887 		*retval = 4096; /* XXX recommend 4K minimum reads/writes */
1888 		break;
1889 	case _PC_REC_MAX_XFER_SIZE:
1890 		*retval = 65536; /* XXX recommend 64K maximum reads/writes */
1891 		break;
1892 	case _PC_REC_XFER_ALIGN:
1893 		*retval = 4096; /* XXX recommend page aligned buffers */
1894 		break;
1895 	case _PC_SYMLINK_MAX:
1896 		*retval = 255;  /* Minimum acceptable POSIX value */
1897 		break;
1898 	case _PC_SYNC_IO:       /* unistd.h: _POSIX_SYNCHRONIZED_IO */
1899 		*retval = 0;    /* [SIO] option is not supported */
1900 		break;
1901 	case _PC_XATTR_SIZE_BITS:
1902 		/* The number of bits used to store maximum extended
1903 		 * attribute size in bytes.  For example, if the maximum
1904 		 * attribute size supported by a file system is 128K, the
1905 		 * value returned will be 18.  However a value 18 can mean
1906 		 * that the maximum attribute size can be anywhere from
1907 		 * (256KB - 1) to 128KB.  As a special case, the resource
1908 		 * fork can have much larger size, and some file system
1909 		 * specific extended attributes can have smaller and preset
1910 		 * size; for example, Finder Info is always 32 bytes.
1911 		 */
1912 		memset(&vfa, 0, sizeof(vfa));
1913 		VFSATTR_INIT(&vfa);
1914 		VFSATTR_WANTED(&vfa, f_capabilities);
1915 		if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
1916 		    (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
1917 		    (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
1918 		    (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
1919 			/* Supports native extended attributes */
1920 			error = VNOP_PATHCONF(vp, name, retval, ctx);
1921 		} else {
1922 			/* Number of bits used to represent the maximum size of
1923 			 * extended attribute stored in an Apple Double file.
1924 			 */
1925 			*retval = AD_XATTR_SIZE_BITS;
1926 		}
1927 		break;
1928 	default:
1929 		error = VNOP_PATHCONF(vp, name, retval, ctx);
1930 		break;
1931 	}
1932 
1933 	return error;
1934 }
1935 
1936 static int
vn_kqfilter(struct fileproc * fp,struct knote * kn,struct kevent_qos_s * kev)1937 vn_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
1938 {
1939 	vfs_context_t ctx = vfs_context_current();
1940 	struct vnode *vp;
1941 	int error = 0;
1942 	int result = 0;
1943 
1944 	vp = (struct vnode *)fp_get_data(fp);
1945 
1946 	/*
1947 	 * Don't attach a knote to a dead vnode.
1948 	 */
1949 	if ((error = vget_internal(vp, 0, VNODE_NODEAD)) == 0) {
1950 		switch (kn->kn_filter) {
1951 		case EVFILT_READ:
1952 		case EVFILT_WRITE:
1953 			if (vnode_isfifo(vp)) {
1954 				/* We'll only watch FIFOs that use our fifofs */
1955 				if (!(vp->v_fifoinfo && vp->v_fifoinfo->fi_readsock)) {
1956 					error = ENOTSUP;
1957 				}
1958 			} else if (!vnode_isreg(vp)) {
1959 				if (vnode_ischr(vp)) {
1960 					result = spec_kqfilter(vp, kn, kev);
1961 					if ((kn->kn_flags & EV_ERROR) == 0) {
1962 						/* claimed by a special device */
1963 						vnode_put(vp);
1964 						return result;
1965 					}
1966 				}
1967 				error = EINVAL;
1968 			}
1969 			break;
1970 		case EVFILT_VNODE:
1971 			break;
1972 		default:
1973 			error = EINVAL;
1974 		}
1975 
1976 		if (error == 0) {
1977 #if CONFIG_MACF
1978 			error = mac_vnode_check_kqfilter(ctx, fp->fp_glob->fg_cred, kn, vp);
1979 			if (error) {
1980 				vnode_put(vp);
1981 				goto out;
1982 			}
1983 #endif
1984 
1985 			kn->kn_hook = (void*)vp;
1986 			kn->kn_filtid = EVFILTID_VN;
1987 			vnode_hold(vp);
1988 
1989 			vnode_lock(vp);
1990 			KNOTE_ATTACH(&vp->v_knotes, kn);
1991 			result = filt_vnode_common(kn, NULL, vp, 0);
1992 			vnode_unlock(vp);
1993 
1994 			/*
1995 			 * Ask the filesystem to provide remove notifications,
1996 			 * but ignore failure
1997 			 */
1998 			VNOP_MONITOR(vp, 0, VNODE_MONITOR_BEGIN, (void*) kn, ctx);
1999 		}
2000 
2001 		vnode_put(vp);
2002 	}
2003 
2004 out:
2005 	if (error) {
2006 		knote_set_error(kn, error);
2007 	}
2008 
2009 	return result;
2010 }
2011 
2012 static void
filt_vndetach(struct knote * kn)2013 filt_vndetach(struct knote *kn)
2014 {
2015 	vfs_context_t ctx = vfs_context_current();
2016 	struct vnode *vp = (struct vnode *)kn->kn_hook;
2017 	uint32_t vid = vnode_vid(vp);
2018 	if (vnode_getwithvid(vp, vid)) {
2019 		vnode_drop(vp);
2020 		return;
2021 	}
2022 	vnode_drop(vp);
2023 
2024 	vnode_lock(vp);
2025 	KNOTE_DETACH(&vp->v_knotes, kn);
2026 	vnode_unlock(vp);
2027 
2028 	/*
2029 	 * Tell a (generally networked) filesystem that we're no longer watching
2030 	 * If the FS wants to track contexts, it should still be using the one from
2031 	 * the VNODE_MONITOR_BEGIN.
2032 	 */
2033 	VNOP_MONITOR(vp, 0, VNODE_MONITOR_END, (void*)kn, ctx);
2034 	vnode_put(vp);
2035 }
2036 
2037 
2038 /*
2039  * Used for EVFILT_READ
2040  *
2041  * Takes only VFIFO or VREG. vnode is locked.  We handle the "poll" case
2042  * differently than the regular case for VREG files.  If not in poll(),
2043  * then we need to know current fileproc offset for VREG.
2044  */
2045 static int64_t
vnode_readable_data_count(vnode_t vp,off_t current_offset,int ispoll)2046 vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
2047 {
2048 	if (vnode_isfifo(vp)) {
2049 #if FIFO
2050 		int cnt;
2051 		int err = fifo_charcount(vp, &cnt);
2052 		if (err == 0) {
2053 			return (int64_t)cnt;
2054 		} else
2055 #endif
2056 		{
2057 			return 0;
2058 		}
2059 	} else if (vnode_isreg(vp)) {
2060 		if (ispoll) {
2061 			return 1;
2062 		}
2063 
2064 		off_t amount;
2065 		amount = vp->v_un.vu_ubcinfo->ui_size - current_offset;
2066 		if (amount > INT64_MAX) {
2067 			return INT64_MAX;
2068 		} else if (amount < INT64_MIN) {
2069 			return INT64_MIN;
2070 		} else {
2071 			return (int64_t)amount;
2072 		}
2073 	} else {
2074 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2075 		return 0;
2076 	}
2077 }
2078 
2079 /*
2080  * Used for EVFILT_WRITE.
2081  *
2082  * For regular vnodes, we can always write (1).  For named pipes,
2083  * see how much space there is in the buffer.  Nothing else is covered.
2084  */
2085 static intptr_t
vnode_writable_space_count(vnode_t vp)2086 vnode_writable_space_count(vnode_t vp)
2087 {
2088 	if (vnode_isfifo(vp)) {
2089 #if FIFO
2090 		long spc;
2091 		int err = fifo_freespace(vp, &spc);
2092 		if (err == 0) {
2093 			return (intptr_t)spc;
2094 		} else
2095 #endif
2096 		{
2097 			return (intptr_t)0;
2098 		}
2099 	} else if (vnode_isreg(vp)) {
2100 		return (intptr_t)1;
2101 	} else {
2102 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2103 		return 0;
2104 	}
2105 }
2106 
2107 /*
2108  * Determine whether this knote should be active
2109  *
2110  * This is kind of subtle.
2111  *      --First, notice if the vnode has been revoked: in so, override hint
2112  *      --EVFILT_READ knotes are checked no matter what the hint is
2113  *      --Other knotes activate based on hint.
2114  *      --If hint is revoke, set special flags and activate
2115  */
2116 static int
filt_vnode_common(struct knote * kn,struct kevent_qos_s * kev,vnode_t vp,long hint)2117 filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev, vnode_t vp, long hint)
2118 {
2119 	int activate = 0;
2120 	int64_t data = 0;
2121 
2122 	lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
2123 
2124 	/* Special handling for vnodes that are in recycle or already gone */
2125 	if (NOTE_REVOKE == hint) {
2126 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2127 		activate = 1;
2128 
2129 		if ((kn->kn_filter == EVFILT_VNODE) && (kn->kn_sfflags & NOTE_REVOKE)) {
2130 			kn->kn_fflags |= NOTE_REVOKE;
2131 		}
2132 	} else {
2133 		switch (kn->kn_filter) {
2134 		case EVFILT_READ:
2135 			data = vnode_readable_data_count(vp, kn->kn_fp->fp_glob->fg_offset, (kn->kn_flags & EV_POLL));
2136 			activate = (data != 0);
2137 			break;
2138 		case EVFILT_WRITE:
2139 			data = vnode_writable_space_count(vp);
2140 			activate = (data != 0);
2141 			break;
2142 		case EVFILT_VNODE:
2143 			/* Check events this note matches against the hint */
2144 			if (kn->kn_sfflags & hint) {
2145 				kn->kn_fflags |= (uint32_t)hint; /* Set which event occurred */
2146 			}
2147 			activate = (kn->kn_fflags != 0);
2148 			break;
2149 		default:
2150 			panic("Invalid knote filter on a vnode!");
2151 		}
2152 	}
2153 
2154 	if (kev && activate) {
2155 		knote_fill_kevent(kn, kev, data);
2156 	}
2157 
2158 	return activate;
2159 }
2160 
2161 static int
filt_vnode(struct knote * kn,long hint)2162 filt_vnode(struct knote *kn, long hint)
2163 {
2164 	vnode_t vp = (struct vnode *)kn->kn_hook;
2165 
2166 	return filt_vnode_common(kn, NULL, vp, hint);
2167 }
2168 
2169 static int
filt_vntouch(struct knote * kn,struct kevent_qos_s * kev)2170 filt_vntouch(struct knote *kn, struct kevent_qos_s *kev)
2171 {
2172 	vnode_t vp = (struct vnode *)kn->kn_hook;
2173 	uint32_t vid = vnode_vid(vp);
2174 	int activate;
2175 	int hint = 0;
2176 
2177 	vnode_lock(vp);
2178 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2179 		/* is recycled */
2180 		hint = NOTE_REVOKE;
2181 	}
2182 
2183 	/* accept new input fflags mask */
2184 	kn->kn_sfflags = kev->fflags;
2185 
2186 	activate = filt_vnode_common(kn, NULL, vp, hint);
2187 
2188 	if (hint == 0) {
2189 		vnode_put_locked(vp);
2190 	}
2191 	vnode_unlock(vp);
2192 
2193 	return activate;
2194 }
2195 
2196 static int
filt_vnprocess(struct knote * kn,struct kevent_qos_s * kev)2197 filt_vnprocess(struct knote *kn, struct kevent_qos_s *kev)
2198 {
2199 	vnode_t vp = (struct vnode *)kn->kn_hook;
2200 	uint32_t vid = vnode_vid(vp);
2201 	int activate;
2202 	int hint = 0;
2203 
2204 	vnode_lock(vp);
2205 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2206 		/* Is recycled */
2207 		hint = NOTE_REVOKE;
2208 	}
2209 	activate = filt_vnode_common(kn, kev, vp, hint);
2210 
2211 	/* Definitely need to unlock, may need to put */
2212 	if (hint == 0) {
2213 		vnode_put_locked(vp);
2214 	}
2215 	vnode_unlock(vp);
2216 
2217 	return activate;
2218 }
2219 
2220 struct vniodesc {
2221 	vnode_t         vnio_vnode;     /* associated vnode */
2222 	int             vnio_fflags;    /* cached fileglob flags */
2223 };
2224 
2225 errno_t
vnio_openfd(int fd,vniodesc_t * vniop)2226 vnio_openfd(int fd, vniodesc_t *vniop)
2227 {
2228 	proc_t p = current_proc();
2229 	struct fileproc *fp = NULL;
2230 	vniodesc_t vnio;
2231 	vnode_t vp;
2232 	int error;
2233 	bool need_drop = false;
2234 
2235 	vnio = kalloc_type(struct vniodesc, Z_WAITOK);
2236 
2237 	error = fp_getfvp(p, fd, &fp, &vp);
2238 	if (error) {
2239 		goto out;
2240 	}
2241 	need_drop = true;
2242 
2243 	if ((fp->fp_glob->fg_flag & (FREAD | FWRITE)) == 0) {
2244 		error = EBADF;
2245 		goto out;
2246 	}
2247 
2248 	if (vnode_isswap(vp)) {
2249 		error = EPERM;
2250 		goto out;
2251 	}
2252 
2253 	if (vp->v_type != VREG) {
2254 		error = EFTYPE;
2255 		goto out;
2256 	}
2257 
2258 	error = vnode_getwithref(vp);
2259 	if (error) {
2260 		goto out;
2261 	}
2262 
2263 	vnio->vnio_vnode = vp;
2264 	vnio->vnio_fflags = fp->fp_glob->fg_flag;
2265 
2266 	(void)fp_drop(p, fd, fp, 0);
2267 	need_drop = false;
2268 
2269 	error = vnode_ref_ext(vp, vnio->vnio_fflags, 0);
2270 	if (error == 0) {
2271 		*vniop = vnio;
2272 		vnio = NULL;
2273 	}
2274 
2275 	vnode_put(vp);
2276 
2277 out:
2278 	if (need_drop) {
2279 		fp_drop(p, fd, fp, 0);
2280 	}
2281 	if (vnio != NULL) {
2282 		kfree_type(struct vniodesc, vnio);
2283 	}
2284 	return error;
2285 }
2286 
2287 errno_t
vnio_close(vniodesc_t vnio)2288 vnio_close(vniodesc_t vnio)
2289 {
2290 	int error;
2291 
2292 	/*
2293 	 * The vniodesc is always destroyed, because the close
2294 	 * always "succeeds".  We just return whatever error
2295 	 * might have been encountered while doing so.
2296 	 */
2297 
2298 	error = vnode_getwithref(vnio->vnio_vnode);
2299 	if (error == 0) {
2300 		error = vnode_close(vnio->vnio_vnode, vnio->vnio_fflags, NULL);
2301 	}
2302 
2303 	kfree_type(struct vniodesc, vnio);
2304 
2305 	return error;
2306 }
2307 
2308 errno_t
vnio_read(vniodesc_t vnio,uio_t uio)2309 vnio_read(vniodesc_t vnio, uio_t uio)
2310 {
2311 	vnode_t vp = vnio->vnio_vnode;
2312 	user_ssize_t read_len;
2313 	int error;
2314 
2315 	read_len = uio_resid(uio);
2316 	if (read_len < 0 || read_len > INT_MAX) {
2317 		return EINVAL;
2318 	}
2319 
2320 	error = vnode_getwithref(vp);
2321 	if (error == 0) {
2322 		error = vn_read_common(vp, uio, vnio->vnio_fflags,
2323 		    vfs_context_current());
2324 		vnode_put(vp);
2325 	}
2326 
2327 	return error;
2328 }
2329 
2330 vnode_t
vnio_vnode(vniodesc_t vnio)2331 vnio_vnode(vniodesc_t vnio)
2332 {
2333 	return vnio->vnio_vnode;
2334 }
2335