xref: /xnu-8019.80.24/bsd/vfs/vfs_vnops.c (revision a325d9c4a84054e40bbe985afedcb50ab80993ea)
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/tty.h>
89 /* Temporary workaround for ubc.h until <rdar://4714366 is resolved */
90 #define ubc_setcred ubc_setcred_deprecated
91 #include <sys/ubc.h>
92 #undef ubc_setcred
93 int     ubc_setcred(struct vnode *, struct proc *);
94 #include <sys/conf.h>
95 #include <sys/disk.h>
96 #include <sys/fsevents.h>
97 #include <sys/kdebug.h>
98 #include <sys/xattr.h>
99 #include <sys/ubc_internal.h>
100 #include <sys/uio_internal.h>
101 #include <sys/resourcevar.h>
102 #include <sys/signalvar.h>
103 
104 #include <vm/vm_kern.h>
105 #include <vm/vm_map.h>
106 
107 #include <miscfs/specfs/specdev.h>
108 #include <miscfs/fifofs/fifo.h>
109 
110 #if CONFIG_MACF
111 #include <security/mac_framework.h>
112 #endif
113 
114 #include <IOKit/IOBSD.h>
115 #include <libkern/section_keywords.h>
116 
117 static int vn_closefile(struct fileglob *fp, vfs_context_t ctx);
118 static int vn_ioctl(struct fileproc *fp, u_long com, caddr_t data,
119     vfs_context_t ctx);
120 static int vn_read(struct fileproc *fp, struct uio *uio, int flags,
121     vfs_context_t ctx);
122 static int vn_write(struct fileproc *fp, struct uio *uio, int flags,
123     vfs_context_t ctx);
124 static int vn_select( struct fileproc *fp, int which, void * wql,
125     vfs_context_t ctx);
126 static int vn_kqfilter(struct fileproc *fp, struct knote *kn,
127     struct kevent_qos_s *kev);
128 static void filt_vndetach(struct knote *kn);
129 static int filt_vnode(struct knote *kn, long hint);
130 static int filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev,
131     vnode_t vp, long hint);
132 static int vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx);
133 
134 const struct fileops vnops = {
135 	.fo_type     = DTYPE_VNODE,
136 	.fo_read     = vn_read,
137 	.fo_write    = vn_write,
138 	.fo_ioctl    = vn_ioctl,
139 	.fo_select   = vn_select,
140 	.fo_close    = vn_closefile,
141 	.fo_drain    = fo_no_drain,
142 	.fo_kqfilter = vn_kqfilter,
143 };
144 
145 static int filt_vntouch(struct knote *kn, struct kevent_qos_s *kev);
146 static int filt_vnprocess(struct knote *kn, struct kevent_qos_s*kev);
147 
148 SECURITY_READ_ONLY_EARLY(struct  filterops) vnode_filtops = {
149 	.f_isfd = 1,
150 	.f_attach = NULL,
151 	.f_detach = filt_vndetach,
152 	.f_event = filt_vnode,
153 	.f_touch = filt_vntouch,
154 	.f_process = filt_vnprocess,
155 };
156 
157 /*
158  * Common code for vnode open operations.
159  * Check permissions, and call the VNOP_OPEN or VNOP_CREATE routine.
160  *
161  * XXX the profusion of interfaces here is probably a bad thing.
162  */
163 int
vn_open(struct nameidata * ndp,int fmode,int cmode)164 vn_open(struct nameidata *ndp, int fmode, int cmode)
165 {
166 	return vn_open_modflags(ndp, &fmode, cmode);
167 }
168 
169 int
vn_open_modflags(struct nameidata * ndp,int * fmodep,int cmode)170 vn_open_modflags(struct nameidata *ndp, int *fmodep, int cmode)
171 {
172 	int error;
173 	struct vnode_attr *vap;
174 
175 	vap = kalloc_type(struct vnode_attr, Z_WAITOK);
176 
177 	VATTR_INIT(vap);
178 	VATTR_SET(vap, va_mode, (mode_t)cmode);
179 
180 	error = vn_open_auth(ndp, fmodep, vap);
181 
182 	kfree_type(struct vnode_attr, vap);
183 
184 	return error;
185 }
186 
187 static int
vn_open_auth_finish(vnode_t vp,int fmode,vfs_context_t ctx)188 vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx)
189 {
190 	int error;
191 
192 	if ((error = vnode_ref_ext(vp, fmode, 0)) != 0) {
193 		goto bad;
194 	}
195 
196 	/* Call out to allow 3rd party notification of open.
197 	 * Ignore result of kauth_authorize_fileop call.
198 	 */
199 #if CONFIG_MACF
200 	mac_vnode_notify_open(ctx, vp, fmode);
201 #endif
202 	kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
203 	    (uintptr_t)vp, 0);
204 
205 	return 0;
206 
207 bad:
208 	return error;
209 }
210 
211 /*
212  * May do nameidone() to allow safely adding an FSEvent.  Cue off of ni_dvp to
213  * determine whether that has happened.
214  */
215 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)216 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 {
218 	uint32_t status = 0;
219 	vnode_t dvp = ndp->ni_dvp;
220 	int batched;
221 	int error;
222 	vnode_t vp;
223 
224 	batched = vnode_compound_open_available(ndp->ni_dvp);
225 	*did_open = FALSE;
226 
227 	VATTR_SET(vap, va_type, VREG);
228 	if (fmode & O_EXCL) {
229 		vap->va_vaflags |= VA_EXCLUSIVE;
230 	}
231 
232 #if NAMEDRSRCFORK
233 	if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK) {
234 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
235 			goto out;
236 		}
237 		if ((error = vnode_makenamedstream(dvp, &ndp->ni_vp, XATTR_RESOURCEFORK_NAME, 0, ctx)) != 0) {
238 			goto out;
239 		}
240 		*did_create = TRUE;
241 	} else {
242 #endif
243 	if (!batched) {
244 		if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0) {
245 			goto out;
246 		}
247 	}
248 
249 	error = vn_create(dvp, &ndp->ni_vp, ndp, vap, VN_CREATE_DOOPEN, fmode, &status, ctx);
250 	if (error != 0) {
251 		if (batched) {
252 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? TRUE : FALSE;
253 		} else {
254 			*did_create = FALSE;
255 		}
256 
257 		if (error == EKEEPLOOKING) {
258 			if (*did_create) {
259 				panic("EKEEPLOOKING, but we did a create?");
260 			}
261 			if (!batched) {
262 				panic("EKEEPLOOKING from filesystem that doesn't support compound vnops?");
263 			}
264 			if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
265 				panic("EKEEPLOOKING, but continue flag not set?");
266 			}
267 
268 			/*
269 			 * Do NOT drop the dvp: we need everything to continue the lookup.
270 			 */
271 			return error;
272 		}
273 	} else {
274 		if (batched) {
275 			*did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? 1 : 0;
276 			*did_open = TRUE;
277 		} else {
278 			*did_create = TRUE;
279 		}
280 	}
281 #if NAMEDRSRCFORK
282 }
283 #endif
284 
285 	vp = ndp->ni_vp;
286 
287 	if (*did_create) {
288 		int     update_flags = 0;
289 
290 		// Make sure the name & parent pointers are hooked up
291 		if (vp->v_name == NULL) {
292 			update_flags |= VNODE_UPDATE_NAME;
293 		}
294 		if (vp->v_parent == NULLVP) {
295 			update_flags |= VNODE_UPDATE_PARENT;
296 		}
297 
298 		if (update_flags) {
299 			vnode_update_identity(vp, dvp, ndp->ni_cnd.cn_nameptr, ndp->ni_cnd.cn_namelen, ndp->ni_cnd.cn_hash, update_flags);
300 		}
301 
302 		vnode_put(dvp);
303 		ndp->ni_dvp = NULLVP;
304 
305 #if CONFIG_FSE
306 		if (need_fsevent(FSE_CREATE_FILE, vp)) {
307 			add_fsevent(FSE_CREATE_FILE, ctx,
308 			    FSE_ARG_VNODE, vp,
309 			    FSE_ARG_DONE);
310 		}
311 #endif
312 	}
313 out:
314 	if (ndp->ni_dvp != NULLVP) {
315 		vnode_put(dvp);
316 		ndp->ni_dvp = NULLVP;
317 	}
318 
319 	return error;
320 }
321 
322 /*
323  * This is the number of times we'll loop in vn_open_auth without explicitly
324  * yielding the CPU when we determine we have to retry.
325  */
326 #define RETRY_NO_YIELD_COUNT    5
327 
328 /*
329  * Open a file with authorization, updating the contents of the structures
330  * pointed to by ndp, fmodep, and vap as necessary to perform the requested
331  * operation.  This function is used for both opens of existing files, and
332  * creation of new files.
333  *
334  * Parameters:	ndp			The nami data pointer describing the
335  *					file
336  *		fmodep			A pointer to an int containg the mode
337  *					information to be used for the open
338  *		vap			A pointer to the vnode attribute
339  *					descriptor to be used for the open
340  *
341  * Indirect:	*			Contents of the data structures pointed
342  *					to by the parameters are modified as
343  *					necessary to the requested operation.
344  *
345  * Returns:	0			Success
346  *		!0			errno value
347  *
348  * Notes:	The kauth_filesec_t in 'vap', if any, is in host byte order.
349  *
350  *		The contents of '*ndp' will be modified, based on the other
351  *		arguments to this function, and to return file and directory
352  *		data necessary to satisfy the requested operation.
353  *
354  *		If the file does not exist and we are creating it, then the
355  *		O_TRUNC flag will be cleared in '*fmodep' to indicate to the
356  *		caller that the file was not truncated.
357  *
358  *		If the file exists and the O_EXCL flag was not specified, then
359  *		the O_CREAT flag will be cleared in '*fmodep' to indicate to
360  *		the caller that the existing file was merely opened rather
361  *		than created.
362  *
363  *		The contents of '*vap' will be modified as necessary to
364  *		complete the operation, including setting of supported
365  *		attribute, clearing of fields containing unsupported attributes
366  *		in the request, if the request proceeds without them, etc..
367  *
368  * XXX:		This function is too complicated in actings on its arguments
369  *
370  * XXX:		We should enummerate the possible errno values here, and where
371  *		in the code they originated.
372  */
373 int
vn_open_auth(struct nameidata * ndp,int * fmodep,struct vnode_attr * vap)374 vn_open_auth(struct nameidata *ndp, int *fmodep, struct vnode_attr *vap)
375 {
376 	struct vnode *vp;
377 	struct vnode *dvp;
378 	vfs_context_t ctx = ndp->ni_cnd.cn_context;
379 	int error;
380 	int fmode;
381 	uint32_t origcnflags;
382 	boolean_t did_create;
383 	boolean_t did_open;
384 	boolean_t need_vnop_open;
385 	boolean_t batched;
386 	boolean_t ref_failed;
387 	int nretries = 0;
388 
389 again:
390 	vp = NULL;
391 	dvp = NULL;
392 	batched = FALSE;
393 	did_create = FALSE;
394 	need_vnop_open = TRUE;
395 	ref_failed = FALSE;
396 	fmode = *fmodep;
397 	origcnflags = ndp->ni_cnd.cn_flags;
398 
399 	// If raw encrypted mode is requested, handle that here
400 	if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)
401 	    && ISSET(vap->va_dataprotect_flags, VA_DP_RAWENCRYPTED)) {
402 		fmode |= FENCRYPTED;
403 	}
404 
405 	if ((fmode & O_NOFOLLOW_ANY) && (fmode & (O_SYMLINK | O_NOFOLLOW))) {
406 		error = EINVAL;
407 		goto out;
408 	}
409 
410 	/*
411 	 * O_CREAT
412 	 */
413 	if (fmode & O_CREAT) {
414 		if ((fmode & O_DIRECTORY)) {
415 			error = EINVAL;
416 			goto out;
417 		}
418 		ndp->ni_cnd.cn_nameiop = CREATE;
419 #if CONFIG_TRIGGERS
420 		ndp->ni_op = OP_LINK;
421 #endif
422 		/* Inherit USEDVP, vnode_open() supported flags only */
423 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
424 		ndp->ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF | AUDITVNPATH1;
425 		ndp->ni_flag = NAMEI_COMPOUNDOPEN;
426 #if NAMEDRSRCFORK
427 		/* open calls are allowed for resource forks. */
428 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
429 #endif
430 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0 && (origcnflags & FOLLOW) != 0) {
431 			ndp->ni_cnd.cn_flags |= FOLLOW;
432 		}
433 		if (fmode & O_NOFOLLOW_ANY) {
434 			/* will return ELOOP on the first symlink to be hit */
435 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
436 		}
437 
438 continue_create_lookup:
439 		if ((error = namei(ndp))) {
440 			goto out;
441 		}
442 
443 		dvp = ndp->ni_dvp;
444 		vp = ndp->ni_vp;
445 
446 		batched = vnode_compound_open_available(dvp);
447 
448 		/* not found, create */
449 		if (vp == NULL) {
450 			/* must have attributes for a new file */
451 			if (vap == NULL) {
452 				vnode_put(dvp);
453 				error = EINVAL;
454 				goto out;
455 			}
456 			/*
457 			 * Attempt a create.   For a system supporting compound VNOPs, we may
458 			 * find an existing file or create one; in either case, we will already
459 			 * have the file open and no VNOP_OPEN() will be needed.
460 			 */
461 			error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
462 
463 			dvp = ndp->ni_dvp;
464 			vp = ndp->ni_vp;
465 
466 			/*
467 			 * Detected a node that the filesystem couldn't handle.  Don't call
468 			 * nameidone() yet, because we need that path buffer.
469 			 */
470 			if (error == EKEEPLOOKING) {
471 				if (!batched) {
472 					panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
473 				}
474 				goto continue_create_lookup;
475 			}
476 
477 			nameidone(ndp);
478 			if (dvp) {
479 				panic("Shouldn't have a dvp here.");
480 			}
481 
482 			if (error) {
483 				/*
484 				 * Check for a create race.
485 				 */
486 				if ((error == EEXIST) && !(fmode & O_EXCL)) {
487 					if (vp) {
488 						vnode_put(vp);
489 					}
490 					goto again;
491 				}
492 				goto bad;
493 			}
494 
495 			need_vnop_open = !did_open;
496 		} else {
497 			if (fmode & O_EXCL) {
498 				error = EEXIST;
499 			}
500 
501 			/*
502 			 * We have a vnode.  Use compound open if available
503 			 * or else fall through to "traditional" path.  Note: can't
504 			 * do a compound open for root, because the parent belongs
505 			 * to a different FS.
506 			 */
507 			if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
508 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
509 
510 				if (error == 0) {
511 					vp = ndp->ni_vp;
512 					need_vnop_open = FALSE;
513 				} else if (error == EKEEPLOOKING) {
514 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
515 						panic("EKEEPLOOKING, but continue flag not set?");
516 					}
517 					goto continue_create_lookup;
518 				}
519 			}
520 			nameidone(ndp);
521 			vnode_put(dvp);
522 			ndp->ni_dvp = NULLVP;
523 
524 			if (error) {
525 				goto bad;
526 			}
527 
528 			fmode &= ~O_CREAT;
529 
530 			/* Fall through */
531 		}
532 	} else {
533 		/*
534 		 * Not O_CREAT
535 		 */
536 		ndp->ni_cnd.cn_nameiop = LOOKUP;
537 		/* Inherit USEDVP, vnode_open() supported flags only */
538 		ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
539 		ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1 | WANTPARENT;
540 #if NAMEDRSRCFORK
541 		/* open calls are allowed for resource forks. */
542 		ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
543 #endif
544 		if (fmode & FENCRYPTED) {
545 			ndp->ni_cnd.cn_flags |= CN_RAW_ENCRYPTED | CN_SKIPNAMECACHE;
546 		}
547 		ndp->ni_flag = NAMEI_COMPOUNDOPEN;
548 
549 		/* preserve NOFOLLOW from vnode_open() */
550 		if (fmode & O_NOFOLLOW || fmode & O_SYMLINK || (origcnflags & FOLLOW) == 0) {
551 			ndp->ni_cnd.cn_flags &= ~FOLLOW;
552 		}
553 		if (fmode & O_NOFOLLOW_ANY) {
554 			/* will return ELOOP on the first symlink to be hit */
555 			ndp->ni_flag |= NAMEI_NOFOLLOW_ANY;
556 		}
557 
558 		/* Do a lookup, possibly going directly to filesystem for compound operation */
559 		do {
560 			if ((error = namei(ndp))) {
561 				goto out;
562 			}
563 			vp = ndp->ni_vp;
564 			dvp = ndp->ni_dvp;
565 
566 			/* Check for batched lookup-open */
567 			batched = vnode_compound_open_available(dvp);
568 			if (batched && ((vp == NULLVP) || (vnode_mount(dvp) == vnode_mount(vp)))) {
569 				error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
570 				vp = ndp->ni_vp;
571 				if (error == 0) {
572 					need_vnop_open = FALSE;
573 				} else if (error == EKEEPLOOKING) {
574 					if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
575 						panic("EKEEPLOOKING, but continue flag not set?");
576 					}
577 				}
578 			}
579 		} while (error == EKEEPLOOKING);
580 
581 		nameidone(ndp);
582 		vnode_put(dvp);
583 		ndp->ni_dvp = NULLVP;
584 
585 		if (error) {
586 			goto bad;
587 		}
588 	}
589 
590 	/*
591 	 * By this point, nameidone() is called, dvp iocount is dropped,
592 	 * and dvp pointer is cleared.
593 	 */
594 	if (ndp->ni_dvp != NULLVP) {
595 		panic("Haven't cleaned up adequately in vn_open_auth()");
596 	}
597 
598 	/*
599 	 * Expect to use this code for filesystems without compound VNOPs, for the root
600 	 * of a filesystem, which can't be "looked up" in the sense of VNOP_LOOKUP(),
601 	 * and for shadow files, which do not live on the same filesystems as their "parents."
602 	 */
603 	if (need_vnop_open) {
604 		if (batched && !vnode_isvroot(vp) && !vnode_isnamedstream(vp)) {
605 			panic("Why am I trying to use VNOP_OPEN() on anything other than the root or a named stream?");
606 		}
607 
608 		if (!did_create) {
609 			error = vn_authorize_open_existing(vp, &ndp->ni_cnd, fmode, ctx, NULL);
610 			if (error) {
611 				goto bad;
612 			}
613 		}
614 
615 		if (VATTR_IS_ACTIVE(vap, va_dataprotect_flags)
616 		    && ISSET(vap->va_dataprotect_flags, VA_DP_RAWUNENCRYPTED)) {
617 			/* Don't allow unencrypted io request from user space unless entitled */
618 			boolean_t entitled = FALSE;
619 #if !SECURE_KERNEL
620 			entitled = IOCurrentTaskHasEntitlement("com.apple.private.security.file-unencrypt-access");
621 #endif
622 			if (!entitled) {
623 				error = EPERM;
624 				goto bad;
625 			}
626 			fmode |= FUNENCRYPTED;
627 		}
628 
629 		error = VNOP_OPEN(vp, fmode, ctx);
630 		if (error) {
631 			goto bad;
632 		}
633 		need_vnop_open = FALSE;
634 	}
635 
636 	// if the vnode is tagged VOPENEVT and the current process
637 	// has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
638 	// flag to the open mode so that this open won't count against
639 	// the vnode when carbon delete() does a vnode_isinuse() to see
640 	// if a file is currently in use.  this allows spotlight
641 	// importers to not interfere with carbon apps that depend on
642 	// the no-delete-if-busy semantics of carbon delete().
643 	//
644 	if (!did_create && (vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
645 		fmode |= O_EVTONLY;
646 	}
647 
648 	/*
649 	 * Grab reference, etc.
650 	 */
651 	error = vn_open_auth_finish(vp, fmode, ctx);
652 	if (error) {
653 		ref_failed = TRUE;
654 		goto bad;
655 	}
656 
657 	/* Compound VNOP open is responsible for doing the truncate */
658 	if (batched || did_create) {
659 		fmode &= ~O_TRUNC;
660 	}
661 
662 	*fmodep = fmode;
663 	return 0;
664 
665 bad:
666 	/* Opened either explicitly or by a batched create */
667 	if (!need_vnop_open) {
668 		VNOP_CLOSE(vp, fmode, ctx);
669 	}
670 
671 	ndp->ni_vp = NULL;
672 	if (vp) {
673 #if NAMEDRSRCFORK
674 		/* Aggressively recycle shadow files if we error'd out during open() */
675 		if ((vnode_isnamedstream(vp)) &&
676 		    (vp->v_parent != NULLVP) &&
677 		    (vnode_isshadow(vp))) {
678 			vnode_recycle(vp);
679 		}
680 #endif
681 		vnode_put(vp);
682 		/*
683 		 * Check for a race against unlink.  We had a vnode
684 		 * but according to vnode_authorize or VNOP_OPEN it
685 		 * no longer exists.
686 		 *
687 		 * EREDRIVEOPEN: means that we were hit by the tty allocation race.
688 		 */
689 		if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN) || ref_failed) {
690 			/*
691 			 * We'll retry here but it may be possible that we get
692 			 * into a retry "spin" inside the kernel and not allow
693 			 * threads, which need to run in order for the retry
694 			 * loop to end, to run. An example is an open of a
695 			 * terminal which is getting revoked and we spin here
696 			 * without yielding becasue namei and VNOP_OPEN are
697 			 * successful but vnode_ref fails. The revoke needs
698 			 * threads with an iocount to run but if spin here we
699 			 * may possibly be blcoking other threads from running.
700 			 *
701 			 * We start yielding the CPU after some number of
702 			 * retries for increasing durations. Note that this is
703 			 * still a loop without an exit condition.
704 			 */
705 			nretries += 1;
706 			if (nretries > RETRY_NO_YIELD_COUNT) {
707 				/* Every hz/100 secs is 10 msecs ... */
708 				tsleep(&nretries, PVFS, "vn_open_auth_retry",
709 				    MIN((nretries * (hz / 100)), hz));
710 			}
711 			goto again;
712 		}
713 	}
714 
715 out:
716 	return error;
717 }
718 
719 #if vn_access_DEPRECATED
720 /*
721  * Authorize an action against a vnode.  This has been the canonical way to
722  * ensure that the credential/process/etc. referenced by a vfs_context
723  * is granted the rights called out in 'mode' against the vnode 'vp'.
724  *
725  * Unfortunately, the use of VREAD/VWRITE/VEXEC makes it very difficult
726  * to add support for more rights.  As such, this interface will be deprecated
727  * and callers will use vnode_authorize instead.
728  */
729 int
vn_access(vnode_t vp,int mode,vfs_context_t context)730 vn_access(vnode_t vp, int mode, vfs_context_t context)
731 {
732 	kauth_action_t  action;
733 
734 	action = 0;
735 	if (mode & VREAD) {
736 		action |= KAUTH_VNODE_READ_DATA;
737 	}
738 	if (mode & VWRITE) {
739 		action |= KAUTH_VNODE_WRITE_DATA;
740 	}
741 	if (mode & VEXEC) {
742 		action |= KAUTH_VNODE_EXECUTE;
743 	}
744 
745 	return vnode_authorize(vp, NULL, action, context);
746 }
747 #endif  /* vn_access_DEPRECATED */
748 
749 /*
750  * Vnode close call
751  */
752 int
vn_close(struct vnode * vp,int flags,vfs_context_t ctx)753 vn_close(struct vnode *vp, int flags, vfs_context_t ctx)
754 {
755 	int error;
756 	int flusherror = 0;
757 
758 #if NAMEDRSRCFORK
759 	/* Sync data from resource fork shadow file if needed. */
760 	if ((vp->v_flag & VISNAMEDSTREAM) &&
761 	    (vp->v_parent != NULLVP) &&
762 	    vnode_isshadow(vp)) {
763 		if (flags & FWASWRITTEN) {
764 			flusherror = vnode_flushnamedstream(vp->v_parent, vp, ctx);
765 		}
766 	}
767 #endif
768 	/*
769 	 * If vnode @vp belongs to a chardev or a blkdev then it is handled
770 	 * specially.  We first drop its user reference count @vp->v_usecount
771 	 * before calling VNOP_CLOSE().  This was done historically to ensure
772 	 * that the last close of a special device vnode performed some
773 	 * conditional cleanups.  Now we still need to drop this reference here
774 	 * to ensure that devfsspec_close() can check if the vnode is still in
775 	 * use.
776 	 */
777 	if (vnode_isspec(vp)) {
778 		(void)vnode_rele_ext(vp, flags, 0);
779 	}
780 
781 	/*
782 	 * On HFS, we flush when the last writer closes.  We do this
783 	 * because resource fork vnodes hold a reference on data fork
784 	 * vnodes and that will prevent them from getting VNOP_INACTIVE
785 	 * which will delay when we flush cached data.  In future, we
786 	 * might find it beneficial to do this for all file systems.
787 	 * Note that it's OK to access v_writecount without the lock
788 	 * in this context.
789 	 */
790 	if (vp->v_tag == VT_HFS && (flags & FWRITE) && vp->v_writecount == 1) {
791 		VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
792 	}
793 
794 	error = VNOP_CLOSE(vp, flags, ctx);
795 
796 #if CONFIG_FSE
797 	if (flags & FWASWRITTEN) {
798 		if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
799 			add_fsevent(FSE_CONTENT_MODIFIED, ctx,
800 			    FSE_ARG_VNODE, vp,
801 			    FSE_ARG_DONE);
802 		}
803 	}
804 #endif
805 
806 	if (!vnode_isspec(vp)) {
807 		(void)vnode_rele_ext(vp, flags, 0);
808 	}
809 
810 	if (flusherror) {
811 		error = flusherror;
812 	}
813 	return error;
814 }
815 
816 static int
vn_read_swapfile(struct vnode * vp,uio_t uio)817 vn_read_swapfile(
818 	struct vnode    *vp,
819 	uio_t           uio)
820 {
821 	int     error;
822 	off_t   swap_count, this_count;
823 	off_t   file_end, read_end;
824 	off_t   prev_resid;
825 	char    *my_swap_page;
826 
827 	/*
828 	 * Reading from a swap file will get you zeroes.
829 	 */
830 
831 	my_swap_page = NULL;
832 	error = 0;
833 	swap_count = uio_resid(uio);
834 
835 	file_end = ubc_getsize(vp);
836 	read_end = uio->uio_offset + uio_resid(uio);
837 	if (uio->uio_offset >= file_end) {
838 		/* uio starts after end of file: nothing to read */
839 		swap_count = 0;
840 	} else if (read_end > file_end) {
841 		/* uio extends beyond end of file: stop before that */
842 		swap_count -= (read_end - file_end);
843 	}
844 
845 	while (swap_count > 0) {
846 		if (my_swap_page == NULL) {
847 			my_swap_page = kalloc_data(PAGE_SIZE, Z_WAITOK | Z_ZERO);
848 			/* add an end-of-line to keep line counters happy */
849 			my_swap_page[PAGE_SIZE - 1] = '\n';
850 		}
851 		this_count = swap_count;
852 		if (this_count > PAGE_SIZE) {
853 			this_count = PAGE_SIZE;
854 		}
855 
856 		prev_resid = uio_resid(uio);
857 		error = uiomove((caddr_t) my_swap_page,
858 		    (int)this_count,
859 		    uio);
860 		if (error) {
861 			break;
862 		}
863 		swap_count -= (prev_resid - uio_resid(uio));
864 	}
865 	kfree_data(my_swap_page, PAGE_SIZE);
866 
867 	return error;
868 }
869 /*
870  * Package up an I/O request on a vnode into a uio and do it.
871  */
872 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)873 vn_rdwr(
874 	enum uio_rw rw,
875 	struct vnode *vp,
876 	caddr_t base,
877 	int len,
878 	off_t offset,
879 	enum uio_seg segflg,
880 	int ioflg,
881 	kauth_cred_t cred,
882 	int *aresid,
883 	proc_t p)
884 {
885 	int64_t resid;
886 	int result;
887 
888 	if (len < 0) {
889 		return EINVAL;
890 	}
891 
892 	result = vn_rdwr_64(rw,
893 	    vp,
894 	    (uint64_t)(uintptr_t)base,
895 	    (int64_t)len,
896 	    offset,
897 	    segflg,
898 	    ioflg,
899 	    cred,
900 	    &resid,
901 	    p);
902 
903 	/* "resid" should be bounded above by "len," which is an int */
904 	if (aresid != NULL) {
905 		*aresid = (int)resid;
906 	}
907 
908 	return result;
909 }
910 
911 
912 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)913 vn_rdwr_64(
914 	enum uio_rw rw,
915 	struct vnode *vp,
916 	uint64_t base,
917 	int64_t len,
918 	off_t offset,
919 	enum uio_seg segflg,
920 	int ioflg,
921 	kauth_cred_t cred,
922 	int64_t *aresid,
923 	proc_t p)
924 {
925 	uio_t auio;
926 	int spacetype;
927 	struct vfs_context context;
928 	int error = 0;
929 	uio_stackbuf_t uio_buf[UIO_SIZEOF(1)];
930 
931 	context.vc_thread = current_thread();
932 	context.vc_ucred = cred;
933 
934 	if (UIO_SEG_IS_USER_SPACE(segflg)) {
935 		spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
936 	} else {
937 		spacetype = UIO_SYSSPACE;
938 	}
939 
940 	if (len < 0) {
941 		return EINVAL;
942 	}
943 
944 	auio = uio_createwithbuffer(1, offset, spacetype, rw,
945 	    &uio_buf[0], sizeof(uio_buf));
946 	uio_addiov(auio, CAST_USER_ADDR_T(base), (user_size_t)len);
947 
948 #if CONFIG_MACF
949 	/* XXXMAC
950 	 *      IO_NOAUTH should be re-examined.
951 	 *	Likely that mediation should be performed in caller.
952 	 */
953 	if ((ioflg & IO_NOAUTH) == 0) {
954 		/* passed cred is fp->f_cred */
955 		if (rw == UIO_READ) {
956 			error = mac_vnode_check_read(&context, cred, vp);
957 		} else {
958 			error = mac_vnode_check_write(&context, cred, vp);
959 		}
960 	}
961 #endif
962 
963 	if (error == 0) {
964 		if (rw == UIO_READ) {
965 			if (vnode_isswap(vp) && ((ioflg & IO_SWAP_DISPATCH) == 0)) {
966 				error = vn_read_swapfile(vp, auio);
967 			} else {
968 				error = VNOP_READ(vp, auio, ioflg, &context);
969 			}
970 		} else {
971 			error = VNOP_WRITE(vp, auio, ioflg, &context);
972 		}
973 	}
974 
975 	if (aresid) {
976 		*aresid = uio_resid(auio);
977 		assert(*aresid <= len);
978 	} else if (uio_resid(auio) && error == 0) {
979 		error = EIO;
980 	}
981 	return error;
982 }
983 
984 void
vn_offset_lock(struct fileglob * fg)985 vn_offset_lock(struct fileglob *fg)
986 {
987 	lck_mtx_lock_spin(&fg->fg_lock);
988 	while (fg->fg_lflags & FG_OFF_LOCKED) {
989 		fg->fg_lflags |= FG_OFF_LOCKWANT;
990 		msleep(&fg->fg_lflags, &fg->fg_lock, PVFS | PSPIN,
991 		    "fg_offset_lock_wait", 0);
992 	}
993 	fg->fg_lflags |= FG_OFF_LOCKED;
994 	lck_mtx_unlock(&fg->fg_lock);
995 }
996 
997 void
vn_offset_unlock(struct fileglob * fg)998 vn_offset_unlock(struct fileglob *fg)
999 {
1000 	int lock_wanted = 0;
1001 
1002 	lck_mtx_lock_spin(&fg->fg_lock);
1003 	if (fg->fg_lflags & FG_OFF_LOCKWANT) {
1004 		lock_wanted = 1;
1005 	}
1006 	fg->fg_lflags &= ~(FG_OFF_LOCKED | FG_OFF_LOCKWANT);
1007 	lck_mtx_unlock(&fg->fg_lock);
1008 	if (lock_wanted) {
1009 		wakeup(&fg->fg_lflags);
1010 	}
1011 }
1012 
1013 /*
1014  * File table vnode read routine.
1015  */
1016 static int
vn_read(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1017 vn_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1018 {
1019 	struct vnode *vp;
1020 	int error;
1021 	int ioflag;
1022 	off_t read_offset;
1023 	user_ssize_t  read_len;
1024 	user_ssize_t  adjusted_read_len;
1025 	user_ssize_t  clippedsize;
1026 	bool offset_locked;
1027 
1028 	read_len = uio_resid(uio);
1029 	if (read_len < 0 || read_len > INT_MAX) {
1030 		return EINVAL;
1031 	}
1032 	adjusted_read_len = read_len;
1033 	clippedsize = 0;
1034 
1035 	if ((flags & FOF_OFFSET) == 0) {
1036 		vn_offset_lock(fp->fp_glob);
1037 		offset_locked = true;
1038 	} else {
1039 		offset_locked = false;
1040 	}
1041 	vp = (struct vnode *)fp_get_data(fp);
1042 	if ((error = vnode_getwithref(vp))) {
1043 		if (offset_locked) {
1044 			vn_offset_unlock(fp->fp_glob);
1045 		}
1046 		return error;
1047 	}
1048 
1049 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1050 		vn_offset_unlock(fp->fp_glob);
1051 		offset_locked = false;
1052 	}
1053 
1054 #if CONFIG_MACF
1055 	error = mac_vnode_check_read(ctx, vfs_context_ucred(ctx), vp);
1056 	if (error) {
1057 		(void)vnode_put(vp);
1058 		if (offset_locked) {
1059 			vn_offset_unlock(fp->fp_glob);
1060 		}
1061 		return error;
1062 	}
1063 #endif
1064 
1065 	/* This signals to VNOP handlers that this read came from a file table read */
1066 	ioflag = IO_SYSCALL_DISPATCH;
1067 
1068 	if (fp->fp_glob->fg_flag & FNONBLOCK) {
1069 		ioflag |= IO_NDELAY;
1070 	}
1071 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1072 		ioflag |= IO_NOCACHE;
1073 	}
1074 	if (fp->fp_glob->fg_flag & FENCRYPTED) {
1075 		ioflag |= IO_ENCRYPTED;
1076 	}
1077 	if (fp->fp_glob->fg_flag & FUNENCRYPTED) {
1078 		ioflag |= IO_SKIP_ENCRYPTION;
1079 	}
1080 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1081 		ioflag |= IO_EVTONLY;
1082 	}
1083 	if (fp->fp_glob->fg_flag & FNORDAHEAD) {
1084 		ioflag |= IO_RAOFF;
1085 	}
1086 
1087 	if ((flags & FOF_OFFSET) == 0) {
1088 		read_offset = fp->fp_glob->fg_offset;
1089 		uio_setoffset(uio, read_offset);
1090 	} else {
1091 		read_offset = uio_offset(uio);
1092 		/* POSIX allows negative offsets for character devices. */
1093 		if ((read_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1094 			error = EINVAL;
1095 			goto error_out;
1096 		}
1097 	}
1098 
1099 	if (read_offset == INT64_MAX) {
1100 		/* can't read any more */
1101 		error = 0;
1102 		goto error_out;
1103 	}
1104 
1105 	/*
1106 	 * If offset + len will cause overflow, reduce the len to a value
1107 	 * (adjusted_read_len) where it won't
1108 	 */
1109 	if ((read_offset >= 0) && (INT64_MAX - read_offset) < read_len) {
1110 		/*
1111 		 * 0                                          read_offset  INT64_MAX
1112 		 * |-----------------------------------------------|----------|~~~
1113 		 *                                                 <--read_len-->
1114 		 *                                                 <-adjusted->
1115 		 */
1116 		adjusted_read_len = (user_ssize_t)(INT64_MAX - read_offset);
1117 	}
1118 
1119 	if (adjusted_read_len < read_len) {
1120 		uio_setresid(uio, adjusted_read_len);
1121 		clippedsize = read_len - adjusted_read_len;
1122 	}
1123 
1124 	if (vnode_isswap(vp) && !(IO_SKIP_ENCRYPTION & ioflag)) {
1125 		/* special case for swap files */
1126 		error = vn_read_swapfile(vp, uio);
1127 	} else {
1128 		error = VNOP_READ(vp, uio, ioflag, ctx);
1129 	}
1130 
1131 	if (clippedsize) {
1132 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1133 	}
1134 
1135 	if ((flags & FOF_OFFSET) == 0) {
1136 		fp->fp_glob->fg_offset += read_len - uio_resid(uio);
1137 	}
1138 
1139 error_out:
1140 	if (offset_locked) {
1141 		vn_offset_unlock(fp->fp_glob);
1142 		offset_locked = false;
1143 	}
1144 
1145 	(void)vnode_put(vp);
1146 	return error;
1147 }
1148 
1149 
1150 /*
1151  * File table vnode write routine.
1152  */
1153 static int
vn_write(struct fileproc * fp,struct uio * uio,int flags,vfs_context_t ctx)1154 vn_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
1155 {
1156 	struct vnode *vp;
1157 	int error, ioflag;
1158 	off_t write_offset;
1159 	off_t write_end_offset;
1160 	user_ssize_t write_len;
1161 	user_ssize_t adjusted_write_len;
1162 	user_ssize_t clippedsize;
1163 	bool offset_locked;
1164 	proc_t p = vfs_context_proc(ctx);
1165 	rlim_t rlim_cur_fsize = p ? proc_limitgetcur(p, RLIMIT_FSIZE) : 0;
1166 
1167 	write_len = uio_resid(uio);
1168 	if (write_len < 0 || write_len > INT_MAX) {
1169 		return EINVAL;
1170 	}
1171 	adjusted_write_len = write_len;
1172 	clippedsize = 0;
1173 
1174 	if ((flags & FOF_OFFSET) == 0) {
1175 		vn_offset_lock(fp->fp_glob);
1176 		offset_locked = true;
1177 	} else {
1178 		offset_locked = false;
1179 	}
1180 
1181 	vp = (struct vnode *)fp_get_data(fp);
1182 	if ((error = vnode_getwithref(vp))) {
1183 		if (offset_locked) {
1184 			vn_offset_unlock(fp->fp_glob);
1185 		}
1186 		return error;
1187 	}
1188 
1189 	if (offset_locked && (vnode_vtype(vp) != VREG || vnode_isswap(vp))) {
1190 		vn_offset_unlock(fp->fp_glob);
1191 		offset_locked = false;
1192 	}
1193 
1194 #if CONFIG_MACF
1195 	error = mac_vnode_check_write(ctx, vfs_context_ucred(ctx), vp);
1196 	if (error) {
1197 		(void)vnode_put(vp);
1198 		if (offset_locked) {
1199 			vn_offset_unlock(fp->fp_glob);
1200 		}
1201 		return error;
1202 	}
1203 #endif
1204 
1205 	/*
1206 	 * IO_SYSCALL_DISPATCH signals to VNOP handlers that this write came from
1207 	 * a file table write
1208 	 */
1209 	ioflag = (IO_UNIT | IO_SYSCALL_DISPATCH);
1210 
1211 	if (vp->v_type == VREG && (fp->fp_glob->fg_flag & O_APPEND)) {
1212 		ioflag |= IO_APPEND;
1213 	}
1214 	if (fp->fp_glob->fg_flag & FNONBLOCK) {
1215 		ioflag |= IO_NDELAY;
1216 	}
1217 	if ((fp->fp_glob->fg_flag & FNOCACHE) || vnode_isnocache(vp)) {
1218 		ioflag |= IO_NOCACHE;
1219 	}
1220 	if (fp->fp_glob->fg_flag & FNODIRECT) {
1221 		ioflag |= IO_NODIRECT;
1222 	}
1223 	if (fp->fp_glob->fg_flag & FSINGLE_WRITER) {
1224 		ioflag |= IO_SINGLE_WRITER;
1225 	}
1226 	if (fp->fp_glob->fg_flag & O_EVTONLY) {
1227 		ioflag |= IO_EVTONLY;
1228 	}
1229 
1230 	/*
1231 	 * Treat synchronous mounts and O_FSYNC on the fd as equivalent.
1232 	 *
1233 	 * XXX We treat O_DSYNC as O_FSYNC for now, since we can not delay
1234 	 * XXX the non-essential metadata without some additional VFS work;
1235 	 * XXX the intent at this point is to plumb the interface for it.
1236 	 */
1237 	if ((fp->fp_glob->fg_flag & (O_FSYNC | O_DSYNC)) ||
1238 	    (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) {
1239 		ioflag |= IO_SYNC;
1240 	}
1241 
1242 	if ((flags & FOF_OFFSET) == 0) {
1243 		write_offset = fp->fp_glob->fg_offset;
1244 		uio_setoffset(uio, write_offset);
1245 	} else {
1246 		/* for pwrite, append should  be ignored */
1247 		ioflag &= ~IO_APPEND;
1248 		write_offset = uio_offset(uio);
1249 		/* POSIX allows negative offsets for character devices. */
1250 		if ((write_offset < 0) && (vnode_vtype(vp) != VCHR)) {
1251 			error = EINVAL;
1252 			goto error_out;
1253 		}
1254 	}
1255 
1256 	if (write_offset == INT64_MAX) {
1257 		/* writes are not possible */
1258 		error = EFBIG;
1259 		goto error_out;
1260 	}
1261 
1262 	/*
1263 	 * write_len is the original write length that was requested.
1264 	 * We may however need to reduce that becasue of two reasons
1265 	 *
1266 	 * 1) If write_offset + write_len will exceed OFF_T_MAX (i.e. INT64_MAX)
1267 	 *    and/or
1268 	 * 2) If write_offset + write_len will exceed the administrative
1269 	 *    limit for the maximum file size.
1270 	 *
1271 	 * In both cases the write will be denied if we can't write even a single
1272 	 * byte otherwise it will be "clipped" (i.e. a short write).
1273 	 */
1274 
1275 	/*
1276 	 * If offset + len will cause overflow, reduce the len
1277 	 * to a value (adjusted_write_len) where it won't
1278 	 */
1279 	if ((write_offset >= 0) && (INT64_MAX - write_offset) < write_len) {
1280 		/*
1281 		 * 0                                          write_offset  INT64_MAX
1282 		 * |-----------------------------------------------|----------|~~~
1283 		 *                                                 <--write_len-->
1284 		 *                                                 <-adjusted->
1285 		 */
1286 		adjusted_write_len = (user_ssize_t)(INT64_MAX - write_offset);
1287 	}
1288 
1289 	/* write_end_offset will always be [0, INT64_MAX] */
1290 	write_end_offset = write_offset + adjusted_write_len;
1291 
1292 	if (p && (vp->v_type == VREG) &&
1293 	    (rlim_cur_fsize != RLIM_INFINITY) &&
1294 	    (rlim_cur_fsize <= INT64_MAX) &&
1295 	    (write_end_offset > (off_t)rlim_cur_fsize)) {
1296 		/*
1297 		 * If the requested residual would cause us to go past the
1298 		 * administrative limit, then we need to adjust the residual
1299 		 * down to cause fewer bytes than requested to be written.  If
1300 		 * we can't do that (e.g. the residual is already 1 byte),
1301 		 * then we fail the write with EFBIG.
1302 		 */
1303 		if (write_offset >= (off_t)rlim_cur_fsize) {
1304 			/*
1305 			 * 0                  rlim_fsize  write_offset  write_end  INT64_MAX
1306 			 * |------------------------|----------|-------------|--------|
1307 			 *                                     <--write_len-->
1308 			 *
1309 			 *                write not permitted
1310 			 */
1311 			psignal(p, SIGXFSZ);
1312 			error = EFBIG;
1313 			goto error_out;
1314 		}
1315 
1316 		/*
1317 		 * 0                  write_offset rlim_fsize  write_end  INT64_MAX
1318 		 * |------------------------|-----------|---------|------------|
1319 		 *                          <------write_len------>
1320 		 *                          <-adjusted-->
1321 		 */
1322 		adjusted_write_len =  (user_ssize_t)((off_t)rlim_cur_fsize - write_offset);
1323 		assert((adjusted_write_len > 0) && (adjusted_write_len < write_len));
1324 	}
1325 
1326 	if (adjusted_write_len < write_len) {
1327 		uio_setresid(uio, adjusted_write_len);
1328 		clippedsize = write_len - adjusted_write_len;
1329 	}
1330 
1331 	error = VNOP_WRITE(vp, uio, ioflag, ctx);
1332 
1333 	/*
1334 	 * If we had to reduce the size of write requested either because
1335 	 * of rlimit or because it would have exceeded
1336 	 * maximum file size, we have to add that back to the residual so
1337 	 * it correctly reflects what we did in this function.
1338 	 */
1339 	if (clippedsize) {
1340 		uio_setresid(uio, (uio_resid(uio) + clippedsize));
1341 	}
1342 
1343 	if ((flags & FOF_OFFSET) == 0) {
1344 		if (ioflag & IO_APPEND) {
1345 			fp->fp_glob->fg_offset = uio_offset(uio);
1346 		} else {
1347 			fp->fp_glob->fg_offset += (write_len - uio_resid(uio));
1348 		}
1349 		if (offset_locked) {
1350 			vn_offset_unlock(fp->fp_glob);
1351 			offset_locked = false;
1352 		}
1353 	}
1354 
1355 	/*
1356 	 * Set the credentials on successful writes
1357 	 */
1358 	if ((error == 0) && (vp->v_tag == VT_NFS) && (UBCINFOEXISTS(vp))) {
1359 		/*
1360 		 * When called from aio subsystem, we only have the proc from
1361 		 * which to get the credential, at this point, so use that
1362 		 * instead.  This means aio functions are incompatible with
1363 		 * per-thread credentials (aio operations are proxied).  We
1364 		 * can't easily correct the aio vs. settid race in this case
1365 		 * anyway, so we disallow it.
1366 		 */
1367 		if ((flags & FOF_PCRED) == 0) {
1368 			ubc_setthreadcred(vp, p, current_thread());
1369 		} else {
1370 			ubc_setcred(vp, p);
1371 		}
1372 	}
1373 	(void)vnode_put(vp);
1374 	return error;
1375 
1376 error_out:
1377 	if (offset_locked) {
1378 		vn_offset_unlock(fp->fp_glob);
1379 	}
1380 	(void)vnode_put(vp);
1381 	return error;
1382 }
1383 
1384 /*
1385  * File table vnode stat routine.
1386  *
1387  * Returns:	0			Success
1388  *		EBADF
1389  *		ENOMEM
1390  *	vnode_getattr:???
1391  */
1392 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)1393 vn_stat_noauth(struct vnode *vp, void *sbptr, kauth_filesec_t *xsec, int isstat64,
1394     int needsrealdev, vfs_context_t ctx, struct ucred *file_cred)
1395 {
1396 	struct vnode_attr va;
1397 	int error;
1398 	u_short mode;
1399 	kauth_filesec_t fsec;
1400 	struct stat *sb = (struct stat *)0;     /* warning avoidance ; protected by isstat64 */
1401 	struct stat64 * sb64 = (struct stat64 *)0;  /* warning avoidance ; protected by isstat64 */
1402 
1403 	if (isstat64 != 0) {
1404 		sb64 = (struct stat64 *)sbptr;
1405 	} else {
1406 		sb = (struct stat *)sbptr;
1407 	}
1408 	memset(&va, 0, sizeof(va));
1409 	VATTR_INIT(&va);
1410 	VATTR_WANTED(&va, va_fsid);
1411 	VATTR_WANTED(&va, va_fileid);
1412 	VATTR_WANTED(&va, va_mode);
1413 	VATTR_WANTED(&va, va_type);
1414 	VATTR_WANTED(&va, va_nlink);
1415 	VATTR_WANTED(&va, va_uid);
1416 	VATTR_WANTED(&va, va_gid);
1417 	VATTR_WANTED(&va, va_rdev);
1418 	VATTR_WANTED(&va, va_data_size);
1419 	VATTR_WANTED(&va, va_access_time);
1420 	VATTR_WANTED(&va, va_modify_time);
1421 	VATTR_WANTED(&va, va_change_time);
1422 	VATTR_WANTED(&va, va_create_time);
1423 	VATTR_WANTED(&va, va_flags);
1424 	VATTR_WANTED(&va, va_gen);
1425 	VATTR_WANTED(&va, va_iosize);
1426 	/* lower layers will synthesise va_total_alloc from va_data_size if required */
1427 	VATTR_WANTED(&va, va_total_alloc);
1428 	if (xsec != NULL) {
1429 		VATTR_WANTED(&va, va_uuuid);
1430 		VATTR_WANTED(&va, va_guuid);
1431 		VATTR_WANTED(&va, va_acl);
1432 	}
1433 	if (needsrealdev) {
1434 		va.va_vaflags = VA_REALFSID;
1435 	}
1436 	error = vnode_getattr(vp, &va, ctx);
1437 	if (error) {
1438 		goto out;
1439 	}
1440 #if CONFIG_MACF
1441 	/*
1442 	 * Give MAC polices a chance to reject or filter the attributes
1443 	 * returned by the filesystem.  Note that MAC policies are consulted
1444 	 * *after* calling the filesystem because filesystems can return more
1445 	 * attributes than were requested so policies wouldn't be authoritative
1446 	 * is consulted beforehand.  This also gives policies an opportunity
1447 	 * to change the values of attributes retrieved.
1448 	 */
1449 	error = mac_vnode_check_getattr(ctx, file_cred, vp, &va);
1450 	if (error) {
1451 		goto out;
1452 	}
1453 #endif
1454 	/*
1455 	 * Copy from vattr table
1456 	 */
1457 	if (isstat64 != 0) {
1458 		sb64->st_dev = va.va_fsid;
1459 		sb64->st_ino = (ino64_t)va.va_fileid;
1460 	} else {
1461 		sb->st_dev = va.va_fsid;
1462 		sb->st_ino = (ino_t)va.va_fileid;
1463 	}
1464 	mode = va.va_mode;
1465 	switch (vp->v_type) {
1466 	case VREG:
1467 		mode |= S_IFREG;
1468 		break;
1469 	case VDIR:
1470 		mode |= S_IFDIR;
1471 		break;
1472 	case VBLK:
1473 		mode |= S_IFBLK;
1474 		break;
1475 	case VCHR:
1476 		mode |= S_IFCHR;
1477 		break;
1478 	case VLNK:
1479 		mode |= S_IFLNK;
1480 		break;
1481 	case VSOCK:
1482 		mode |= S_IFSOCK;
1483 		break;
1484 	case VFIFO:
1485 		mode |= S_IFIFO;
1486 		break;
1487 	default:
1488 		error = EBADF;
1489 		goto out;
1490 	}
1491 	;
1492 	if (isstat64 != 0) {
1493 		sb64->st_mode = mode;
1494 		sb64->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1495 		sb64->st_uid = va.va_uid;
1496 		sb64->st_gid = va.va_gid;
1497 		sb64->st_rdev = va.va_rdev;
1498 		sb64->st_size = va.va_data_size;
1499 		sb64->st_atimespec = va.va_access_time;
1500 		sb64->st_mtimespec = va.va_modify_time;
1501 		sb64->st_ctimespec = va.va_change_time;
1502 		if (VATTR_IS_SUPPORTED(&va, va_create_time)) {
1503 			sb64->st_birthtimespec =  va.va_create_time;
1504 		} else {
1505 			sb64->st_birthtimespec.tv_sec = sb64->st_birthtimespec.tv_nsec = 0;
1506 		}
1507 		sb64->st_blksize = va.va_iosize;
1508 		sb64->st_flags = va.va_flags;
1509 		sb64->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1510 	} else {
1511 		sb->st_mode = mode;
1512 		sb->st_nlink = VATTR_IS_SUPPORTED(&va, va_nlink) ? va.va_nlink > UINT16_MAX ? UINT16_MAX : (u_int16_t)va.va_nlink : 1;
1513 		sb->st_uid = va.va_uid;
1514 		sb->st_gid = va.va_gid;
1515 		sb->st_rdev = va.va_rdev;
1516 		sb->st_size = va.va_data_size;
1517 		sb->st_atimespec = va.va_access_time;
1518 		sb->st_mtimespec = va.va_modify_time;
1519 		sb->st_ctimespec = va.va_change_time;
1520 		sb->st_blksize = va.va_iosize;
1521 		sb->st_flags = va.va_flags;
1522 		sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
1523 	}
1524 
1525 	/* if we're interested in extended security data and we got an ACL */
1526 	if (xsec != NULL) {
1527 		if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
1528 		    !VATTR_IS_SUPPORTED(&va, va_uuuid) &&
1529 		    !VATTR_IS_SUPPORTED(&va, va_guuid)) {
1530 			*xsec = KAUTH_FILESEC_NONE;
1531 		} else {
1532 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1533 				fsec = kauth_filesec_alloc(va.va_acl->acl_entrycount);
1534 			} else {
1535 				fsec = kauth_filesec_alloc(0);
1536 			}
1537 			if (fsec == NULL) {
1538 				error = ENOMEM;
1539 				goto out;
1540 			}
1541 			fsec->fsec_magic = KAUTH_FILESEC_MAGIC;
1542 			if (VATTR_IS_SUPPORTED(&va, va_uuuid)) {
1543 				fsec->fsec_owner = va.va_uuuid;
1544 			} else {
1545 				fsec->fsec_owner = kauth_null_guid;
1546 			}
1547 			if (VATTR_IS_SUPPORTED(&va, va_guuid)) {
1548 				fsec->fsec_group = va.va_guuid;
1549 			} else {
1550 				fsec->fsec_group = kauth_null_guid;
1551 			}
1552 			if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) {
1553 				__nochk_bcopy(va.va_acl, &(fsec->fsec_acl), KAUTH_ACL_COPYSIZE(va.va_acl));
1554 			} else {
1555 				fsec->fsec_acl.acl_entrycount = KAUTH_FILESEC_NOACL;
1556 			}
1557 			*xsec = fsec;
1558 		}
1559 	}
1560 
1561 	/* Do not give the generation number out to unpriviledged users */
1562 	if (va.va_gen && !vfs_context_issuser(ctx)) {
1563 		if (isstat64 != 0) {
1564 			sb64->st_gen = 0;
1565 		} else {
1566 			sb->st_gen = 0;
1567 		}
1568 	} else {
1569 		if (isstat64 != 0) {
1570 			sb64->st_gen = va.va_gen;
1571 		} else {
1572 			sb->st_gen = va.va_gen;
1573 		}
1574 	}
1575 
1576 	error = 0;
1577 out:
1578 	if (VATTR_IS_SUPPORTED(&va, va_acl) && va.va_acl != NULL) {
1579 		kauth_acl_free(va.va_acl);
1580 	}
1581 	return error;
1582 }
1583 
1584 int
vn_stat(struct vnode * vp,void * sb,kauth_filesec_t * xsec,int isstat64,int needsrealdev,vfs_context_t ctx)1585 vn_stat(struct vnode *vp, void *sb, kauth_filesec_t *xsec, int isstat64, int needsrealdev, vfs_context_t ctx)
1586 {
1587 	int error;
1588 
1589 #if CONFIG_MACF
1590 	error = mac_vnode_check_stat(ctx, NOCRED, vp);
1591 	if (error) {
1592 		return error;
1593 	}
1594 #endif
1595 
1596 	/* authorize */
1597 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_ATTRIBUTES | KAUTH_VNODE_READ_SECURITY, ctx)) != 0) {
1598 		return error;
1599 	}
1600 
1601 	/* actual stat */
1602 	return vn_stat_noauth(vp, sb, xsec, isstat64, needsrealdev, ctx, NOCRED);
1603 }
1604 
1605 
1606 /*
1607  * File table vnode ioctl routine.
1608  */
1609 static int
vn_ioctl(struct fileproc * fp,u_long com,caddr_t data,vfs_context_t ctx)1610 vn_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
1611 {
1612 	struct vnode *vp = (struct vnode *)fp_get_data(fp);
1613 	off_t file_size;
1614 	int error;
1615 	struct vnode *ttyvp;
1616 
1617 	if ((error = vnode_getwithref(vp))) {
1618 		return error;
1619 	}
1620 
1621 #if CONFIG_MACF
1622 	error = mac_vnode_check_ioctl(ctx, vp, com);
1623 	if (error) {
1624 		goto out;
1625 	}
1626 #endif
1627 
1628 	switch (vp->v_type) {
1629 	case VREG:
1630 	case VDIR:
1631 		if (com == FIONREAD) {
1632 			off_t temp_nbytes;
1633 			if ((error = vnode_size(vp, &file_size, ctx)) != 0) {
1634 				goto out;
1635 			}
1636 			temp_nbytes = file_size - fp->fp_glob->fg_offset;
1637 			if (temp_nbytes > INT_MAX) {
1638 				*(int *)data = INT_MAX;
1639 			} else if (temp_nbytes < 0) {
1640 				*(int *)data = 0;
1641 			} else {
1642 				*(int *)data = (int)temp_nbytes;
1643 			}
1644 			goto out;
1645 		}
1646 		if (com == FIONBIO || com == FIOASYNC) {        /* XXX */
1647 			goto out;
1648 		}
1649 		OS_FALLTHROUGH;
1650 
1651 	default:
1652 		error = ENOTTY;
1653 		goto out;
1654 
1655 	case VFIFO:
1656 	case VCHR:
1657 	case VBLK:
1658 
1659 		if (com == TIOCREVOKE || com == TIOCREVOKECLEAR) {
1660 			error = ENOTTY;
1661 			goto out;
1662 		}
1663 
1664 		/* Should not be able to set block size from user space */
1665 		if (com == DKIOCSETBLOCKSIZE) {
1666 			error = EPERM;
1667 			goto out;
1668 		}
1669 
1670 		if (com == FIODTYPE) {
1671 			if (vp->v_type == VBLK) {
1672 				if (major(vp->v_rdev) >= nblkdev) {
1673 					error = ENXIO;
1674 					goto out;
1675 				}
1676 				*(int *)data = bdevsw[major(vp->v_rdev)].d_type;
1677 			} else if (vp->v_type == VCHR) {
1678 				if (major(vp->v_rdev) >= nchrdev) {
1679 					error = ENXIO;
1680 					goto out;
1681 				}
1682 				*(int *)data = cdevsw[major(vp->v_rdev)].d_type;
1683 			} else {
1684 				error = ENOTTY;
1685 				goto out;
1686 			}
1687 			goto out;
1688 		}
1689 		error = VNOP_IOCTL(vp, com, data, fp->fp_glob->fg_flag, ctx);
1690 
1691 		if (error == 0 && com == TIOCSCTTY) {
1692 			struct session *sessp;
1693 			struct pgrp *pg;
1694 
1695 			pg = proc_pgrp(vfs_context_proc(ctx), &sessp);
1696 
1697 			session_lock(sessp);
1698 			ttyvp = sessp->s_ttyvp;
1699 			sessp->s_ttyvp = vp;
1700 			sessp->s_ttyvid = vnode_vid(vp);
1701 			session_unlock(sessp);
1702 
1703 			pgrp_rele(pg);
1704 		}
1705 	}
1706 out:
1707 	(void)vnode_put(vp);
1708 	return error;
1709 }
1710 
1711 /*
1712  * File table vnode select routine.
1713  */
1714 static int
vn_select(struct fileproc * fp,int which,void * wql,__unused vfs_context_t ctx)1715 vn_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx)
1716 {
1717 	int error;
1718 	struct vnode * vp = (struct vnode *)fp_get_data(fp);
1719 	struct vfs_context context;
1720 
1721 	if ((error = vnode_getwithref(vp)) == 0) {
1722 		context.vc_thread = current_thread();
1723 		context.vc_ucred = fp->fp_glob->fg_cred;
1724 
1725 #if CONFIG_MACF
1726 		/*
1727 		 * XXX We should use a per thread credential here; minimally,
1728 		 * XXX the process credential should have a persistent
1729 		 * XXX reference on it before being passed in here.
1730 		 */
1731 		error = mac_vnode_check_select(ctx, vp, which);
1732 		if (error == 0)
1733 #endif
1734 		error = VNOP_SELECT(vp, which, fp->fp_glob->fg_flag, wql, ctx);
1735 
1736 		(void)vnode_put(vp);
1737 	}
1738 	return error;
1739 }
1740 
1741 /*
1742  * File table vnode close routine.
1743  */
1744 static int
vn_closefile(struct fileglob * fg,vfs_context_t ctx)1745 vn_closefile(struct fileglob *fg, vfs_context_t ctx)
1746 {
1747 	struct vnode *vp = fg_get_data(fg);
1748 	int error;
1749 
1750 	if ((error = vnode_getwithref(vp)) == 0) {
1751 		if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE &&
1752 		    ((fg->fg_flag & FWASLOCKED) != 0 ||
1753 		    (fg->fg_lflags & FG_HAS_OFDLOCK) != 0)) {
1754 			struct flock lf = {
1755 				.l_whence = SEEK_SET,
1756 				.l_start = 0,
1757 				.l_len = 0,
1758 				.l_type = F_UNLCK
1759 			};
1760 
1761 			if ((fg->fg_flag & FWASLOCKED) != 0) {
1762 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1763 				    F_UNLCK, &lf, F_FLOCK, ctx, NULL);
1764 			}
1765 
1766 			if ((fg->fg_lflags & FG_HAS_OFDLOCK) != 0) {
1767 				(void) VNOP_ADVLOCK(vp, (caddr_t)fg,
1768 				    F_UNLCK, &lf, F_OFD_LOCK, ctx, NULL);
1769 			}
1770 		}
1771 		error = vn_close(vp, fg->fg_flag, ctx);
1772 		(void) vnode_put(vp);
1773 	}
1774 	return error;
1775 }
1776 
1777 /*
1778  * Returns:	0			Success
1779  *	VNOP_PATHCONF:???
1780  */
1781 int
vn_pathconf(vnode_t vp,int name,int32_t * retval,vfs_context_t ctx)1782 vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
1783 {
1784 	int     error = 0;
1785 	struct vfs_attr vfa;
1786 
1787 	switch (name) {
1788 	case _PC_EXTENDED_SECURITY_NP:
1789 		*retval = vfs_extendedsecurity(vnode_mount(vp)) ? 1 : 0;
1790 		break;
1791 	case _PC_AUTH_OPAQUE_NP:
1792 		*retval = vfs_authopaque(vnode_mount(vp));
1793 		break;
1794 	case _PC_2_SYMLINKS:
1795 		*retval = 1;    /* XXX NOTSUP on MSDOS, etc. */
1796 		break;
1797 	case _PC_ALLOC_SIZE_MIN:
1798 		*retval = 1;    /* XXX lie: 1 byte */
1799 		break;
1800 	case _PC_ASYNC_IO:      /* unistd.h: _POSIX_ASYNCHRONUS_IO */
1801 		*retval = 1;    /* [AIO] option is supported */
1802 		break;
1803 	case _PC_PRIO_IO:       /* unistd.h: _POSIX_PRIORITIZED_IO */
1804 		*retval = 0;    /* [PIO] option is not supported */
1805 		break;
1806 	case _PC_REC_INCR_XFER_SIZE:
1807 		*retval = 4096; /* XXX go from MIN to MAX 4K at a time */
1808 		break;
1809 	case _PC_REC_MIN_XFER_SIZE:
1810 		*retval = 4096; /* XXX recommend 4K minimum reads/writes */
1811 		break;
1812 	case _PC_REC_MAX_XFER_SIZE:
1813 		*retval = 65536; /* XXX recommend 64K maximum reads/writes */
1814 		break;
1815 	case _PC_REC_XFER_ALIGN:
1816 		*retval = 4096; /* XXX recommend page aligned buffers */
1817 		break;
1818 	case _PC_SYMLINK_MAX:
1819 		*retval = 255;  /* Minimum acceptable POSIX value */
1820 		break;
1821 	case _PC_SYNC_IO:       /* unistd.h: _POSIX_SYNCHRONIZED_IO */
1822 		*retval = 0;    /* [SIO] option is not supported */
1823 		break;
1824 	case _PC_XATTR_SIZE_BITS:
1825 		/* The number of bits used to store maximum extended
1826 		 * attribute size in bytes.  For example, if the maximum
1827 		 * attribute size supported by a file system is 128K, the
1828 		 * value returned will be 18.  However a value 18 can mean
1829 		 * that the maximum attribute size can be anywhere from
1830 		 * (256KB - 1) to 128KB.  As a special case, the resource
1831 		 * fork can have much larger size, and some file system
1832 		 * specific extended attributes can have smaller and preset
1833 		 * size; for example, Finder Info is always 32 bytes.
1834 		 */
1835 		memset(&vfa, 0, sizeof(vfa));
1836 		VFSATTR_INIT(&vfa);
1837 		VFSATTR_WANTED(&vfa, f_capabilities);
1838 		if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
1839 		    (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
1840 		    (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
1841 		    (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
1842 			/* Supports native extended attributes */
1843 			error = VNOP_PATHCONF(vp, name, retval, ctx);
1844 		} else {
1845 			/* Number of bits used to represent the maximum size of
1846 			 * extended attribute stored in an Apple Double file.
1847 			 */
1848 			*retval = AD_XATTR_SIZE_BITS;
1849 		}
1850 		break;
1851 	default:
1852 		error = VNOP_PATHCONF(vp, name, retval, ctx);
1853 		break;
1854 	}
1855 
1856 	return error;
1857 }
1858 
1859 static int
vn_kqfilter(struct fileproc * fp,struct knote * kn,struct kevent_qos_s * kev)1860 vn_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
1861 {
1862 	vfs_context_t ctx = vfs_context_current();
1863 	struct vnode *vp;
1864 	int error = 0;
1865 	int result = 0;
1866 
1867 	vp = (struct vnode *)fp_get_data(fp);
1868 
1869 	/*
1870 	 * Don't attach a knote to a dead vnode.
1871 	 */
1872 	if ((error = vget_internal(vp, 0, VNODE_NODEAD)) == 0) {
1873 		switch (kn->kn_filter) {
1874 		case EVFILT_READ:
1875 		case EVFILT_WRITE:
1876 			if (vnode_isfifo(vp)) {
1877 				/* We'll only watch FIFOs that use our fifofs */
1878 				if (!(vp->v_fifoinfo && vp->v_fifoinfo->fi_readsock)) {
1879 					error = ENOTSUP;
1880 				}
1881 			} else if (!vnode_isreg(vp)) {
1882 				if (vnode_ischr(vp)) {
1883 					result = spec_kqfilter(vp, kn, kev);
1884 					if ((kn->kn_flags & EV_ERROR) == 0) {
1885 						/* claimed by a special device */
1886 						vnode_put(vp);
1887 						return result;
1888 					}
1889 				}
1890 				error = EINVAL;
1891 			}
1892 			break;
1893 		case EVFILT_VNODE:
1894 			break;
1895 		default:
1896 			error = EINVAL;
1897 		}
1898 
1899 		if (error == 0) {
1900 #if CONFIG_MACF
1901 			error = mac_vnode_check_kqfilter(ctx, fp->fp_glob->fg_cred, kn, vp);
1902 			if (error) {
1903 				vnode_put(vp);
1904 				goto out;
1905 			}
1906 #endif
1907 
1908 			kn->kn_hook = (void*)vp;
1909 			kn->kn_filtid = EVFILTID_VN;
1910 
1911 			vnode_lock(vp);
1912 			KNOTE_ATTACH(&vp->v_knotes, kn);
1913 			result = filt_vnode_common(kn, NULL, vp, 0);
1914 			vnode_unlock(vp);
1915 
1916 			/*
1917 			 * Ask the filesystem to provide remove notifications,
1918 			 * but ignore failure
1919 			 */
1920 			VNOP_MONITOR(vp, 0, VNODE_MONITOR_BEGIN, (void*) kn, ctx);
1921 		}
1922 
1923 		vnode_put(vp);
1924 	}
1925 
1926 out:
1927 	if (error) {
1928 		knote_set_error(kn, error);
1929 	}
1930 
1931 	return result;
1932 }
1933 
1934 static void
filt_vndetach(struct knote * kn)1935 filt_vndetach(struct knote *kn)
1936 {
1937 	vfs_context_t ctx = vfs_context_current();
1938 	struct vnode *vp = (struct vnode *)kn->kn_hook;
1939 	uint32_t vid = vnode_vid(vp);
1940 	if (vnode_getwithvid(vp, vid)) {
1941 		return;
1942 	}
1943 
1944 	vnode_lock(vp);
1945 	KNOTE_DETACH(&vp->v_knotes, kn);
1946 	vnode_unlock(vp);
1947 
1948 	/*
1949 	 * Tell a (generally networked) filesystem that we're no longer watching
1950 	 * If the FS wants to track contexts, it should still be using the one from
1951 	 * the VNODE_MONITOR_BEGIN.
1952 	 */
1953 	VNOP_MONITOR(vp, 0, VNODE_MONITOR_END, (void*)kn, ctx);
1954 	vnode_put(vp);
1955 }
1956 
1957 
1958 /*
1959  * Used for EVFILT_READ
1960  *
1961  * Takes only VFIFO or VREG. vnode is locked.  We handle the "poll" case
1962  * differently than the regular case for VREG files.  If not in poll(),
1963  * then we need to know current fileproc offset for VREG.
1964  */
1965 static int64_t
vnode_readable_data_count(vnode_t vp,off_t current_offset,int ispoll)1966 vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
1967 {
1968 	if (vnode_isfifo(vp)) {
1969 #if FIFO
1970 		int cnt;
1971 		int err = fifo_charcount(vp, &cnt);
1972 		if (err == 0) {
1973 			return (int64_t)cnt;
1974 		} else
1975 #endif
1976 		{
1977 			return 0;
1978 		}
1979 	} else if (vnode_isreg(vp)) {
1980 		if (ispoll) {
1981 			return 1;
1982 		}
1983 
1984 		off_t amount;
1985 		amount = vp->v_un.vu_ubcinfo->ui_size - current_offset;
1986 		if (amount > INT64_MAX) {
1987 			return INT64_MAX;
1988 		} else if (amount < INT64_MIN) {
1989 			return INT64_MIN;
1990 		} else {
1991 			return (int64_t)amount;
1992 		}
1993 	} else {
1994 		panic("Should never have an EVFILT_READ except for reg or fifo.");
1995 		return 0;
1996 	}
1997 }
1998 
1999 /*
2000  * Used for EVFILT_WRITE.
2001  *
2002  * For regular vnodes, we can always write (1).  For named pipes,
2003  * see how much space there is in the buffer.  Nothing else is covered.
2004  */
2005 static intptr_t
vnode_writable_space_count(vnode_t vp)2006 vnode_writable_space_count(vnode_t vp)
2007 {
2008 	if (vnode_isfifo(vp)) {
2009 #if FIFO
2010 		long spc;
2011 		int err = fifo_freespace(vp, &spc);
2012 		if (err == 0) {
2013 			return (intptr_t)spc;
2014 		} else
2015 #endif
2016 		{
2017 			return (intptr_t)0;
2018 		}
2019 	} else if (vnode_isreg(vp)) {
2020 		return (intptr_t)1;
2021 	} else {
2022 		panic("Should never have an EVFILT_READ except for reg or fifo.");
2023 		return 0;
2024 	}
2025 }
2026 
2027 /*
2028  * Determine whether this knote should be active
2029  *
2030  * This is kind of subtle.
2031  *      --First, notice if the vnode has been revoked: in so, override hint
2032  *      --EVFILT_READ knotes are checked no matter what the hint is
2033  *      --Other knotes activate based on hint.
2034  *      --If hint is revoke, set special flags and activate
2035  */
2036 static int
filt_vnode_common(struct knote * kn,struct kevent_qos_s * kev,vnode_t vp,long hint)2037 filt_vnode_common(struct knote *kn, struct kevent_qos_s *kev, vnode_t vp, long hint)
2038 {
2039 	int activate = 0;
2040 	int64_t data = 0;
2041 
2042 	lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
2043 
2044 	/* Special handling for vnodes that are in recycle or already gone */
2045 	if (NOTE_REVOKE == hint) {
2046 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2047 		activate = 1;
2048 
2049 		if ((kn->kn_filter == EVFILT_VNODE) && (kn->kn_sfflags & NOTE_REVOKE)) {
2050 			kn->kn_fflags |= NOTE_REVOKE;
2051 		}
2052 	} else {
2053 		switch (kn->kn_filter) {
2054 		case EVFILT_READ:
2055 			data = vnode_readable_data_count(vp, kn->kn_fp->fp_glob->fg_offset, (kn->kn_flags & EV_POLL));
2056 			activate = (data != 0);
2057 			break;
2058 		case EVFILT_WRITE:
2059 			data = vnode_writable_space_count(vp);
2060 			activate = (data != 0);
2061 			break;
2062 		case EVFILT_VNODE:
2063 			/* Check events this note matches against the hint */
2064 			if (kn->kn_sfflags & hint) {
2065 				kn->kn_fflags |= hint;         /* Set which event occurred */
2066 			}
2067 			activate = (kn->kn_fflags != 0);
2068 			break;
2069 		default:
2070 			panic("Invalid knote filter on a vnode!");
2071 		}
2072 	}
2073 
2074 	if (kev && activate) {
2075 		knote_fill_kevent(kn, kev, data);
2076 	}
2077 
2078 	return activate;
2079 }
2080 
2081 static int
filt_vnode(struct knote * kn,long hint)2082 filt_vnode(struct knote *kn, long hint)
2083 {
2084 	vnode_t vp = (struct vnode *)kn->kn_hook;
2085 
2086 	return filt_vnode_common(kn, NULL, vp, hint);
2087 }
2088 
2089 static int
filt_vntouch(struct knote * kn,struct kevent_qos_s * kev)2090 filt_vntouch(struct knote *kn, struct kevent_qos_s *kev)
2091 {
2092 	vnode_t vp = (struct vnode *)kn->kn_hook;
2093 	uint32_t vid = vnode_vid(vp);
2094 	int activate;
2095 	int hint = 0;
2096 
2097 	vnode_lock(vp);
2098 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2099 		/* is recycled */
2100 		hint = NOTE_REVOKE;
2101 	}
2102 
2103 	/* accept new input fflags mask */
2104 	kn->kn_sfflags = kev->fflags;
2105 
2106 	activate = filt_vnode_common(kn, NULL, vp, hint);
2107 
2108 	if (hint == 0) {
2109 		vnode_put_locked(vp);
2110 	}
2111 	vnode_unlock(vp);
2112 
2113 	return activate;
2114 }
2115 
2116 static int
filt_vnprocess(struct knote * kn,struct kevent_qos_s * kev)2117 filt_vnprocess(struct knote *kn, struct kevent_qos_s *kev)
2118 {
2119 	vnode_t vp = (struct vnode *)kn->kn_hook;
2120 	uint32_t vid = vnode_vid(vp);
2121 	int activate;
2122 	int hint = 0;
2123 
2124 	vnode_lock(vp);
2125 	if (vnode_getiocount(vp, vid, VNODE_NODEAD | VNODE_WITHID) != 0) {
2126 		/* Is recycled */
2127 		hint = NOTE_REVOKE;
2128 	}
2129 	activate = filt_vnode_common(kn, kev, vp, hint);
2130 
2131 	/* Definitely need to unlock, may need to put */
2132 	if (hint == 0) {
2133 		vnode_put_locked(vp);
2134 	}
2135 	vnode_unlock(vp);
2136 
2137 	return activate;
2138 }
2139