xref: /xnu-10063.121.3/bsd/kern/sysv_shm.c (revision 2c2f96dc2b9a4408a43d3150ae9c105355ca3daa)
1 /*
2  * Copyright (c) 2000-2019 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 /*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
29 
30 /*
31  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. All advertising materials mentioning features or use of this software
42  *    must display the following acknowledgement:
43  *	This product includes software developed by Adam Glass and Charles
44  *	Hannum.
45  * 4. The names of the authors may not be used to endorse or promote products
46  *    derived from this software without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
49  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
50  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
51  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
52  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
53  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
54  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
55  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
56  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
57  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
58  */
59 /*
60  * NOTICE: This file was modified by McAfee Research in 2004 to introduce
61  * support for mandatory and extensible security protections.  This notice
62  * is included in support of clause 2.2 (b) of the Apple Public License,
63  * Version 2.0.
64  * Copyright (c) 2005-2006 SPARTA, Inc.
65  */
66 
67 
68 #include <sys/appleapiopts.h>
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.h>
72 #include <sys/shm_internal.h>
73 #include <sys/proc_internal.h>
74 #include <sys/kauth.h>
75 #include <sys/malloc.h>
76 #include <sys/mman.h>
77 #include <sys/stat.h>
78 #include <sys/sysctl.h>
79 #include <sys/ipcs.h>
80 #include <sys/sysent.h>
81 #include <sys/sysproto.h>
82 #if CONFIG_MACF
83 #include <security/mac_framework.h>
84 #endif
85 
86 #include <security/audit/audit.h>
87 
88 #include <mach/mach_types.h>
89 #include <mach/vm_inherit.h>
90 #include <mach/vm_map.h>
91 
92 #include <mach/mach_vm.h>
93 
94 #include <vm/vm_map.h>
95 #include <vm/vm_protos.h>
96 #include <vm/vm_kern.h>
97 
98 #include <kern/locks.h>
99 #include <os/overflow.h>
100 
101 /* Uncomment this line to see MAC debugging output. */
102 /* #define MAC_DEBUG */
103 #if CONFIG_MACF_DEBUG
104 #define MPRINTF(a)      printf a
105 #else
106 #define MPRINTF(a)
107 #endif
108 
109 #if SYSV_SHM
110 static int shminit(void);
111 
112 static LCK_GRP_DECLARE(sysv_shm_subsys_lck_grp, "sysv_shm_subsys_lock");
113 static LCK_MTX_DECLARE(sysv_shm_subsys_mutex, &sysv_shm_subsys_lck_grp);
114 
115 #define SYSV_SHM_SUBSYS_LOCK() lck_mtx_lock(&sysv_shm_subsys_mutex)
116 #define SYSV_SHM_SUBSYS_UNLOCK() lck_mtx_unlock(&sysv_shm_subsys_mutex)
117 
118 static int oshmctl(void *p, void *uap, void *retval);
119 static int shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, int * retval);
120 static int shmget_existing(struct shmget_args *uap, int mode, int segnum, int  * retval);
121 static void shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out);
122 static void shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out);
123 
124 /* XXX casting to (sy_call_t *) is bogus, as usual. */
125 static sy_call_t* const shmcalls[] = {
126 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
127 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
128 	(sy_call_t *)shmctl
129 };
130 
131 #define SHMSEG_FREE             0x0200
132 #define SHMSEG_REMOVED          0x0400
133 #define SHMSEG_ALLOCATED        0x0800
134 #define SHMSEG_WANTED           0x1000
135 
136 static int shm_last_free, shm_nused, shm_committed;
137 struct shmid_kernel     *shmsegs;       /* 64 bit version */
138 static int shm_inited = 0;
139 
140 /*
141  * Since anonymous memory chunks are limited to ANON_MAX_SIZE bytes,
142  * we have to keep a list of chunks when we want to handle a shared memory
143  * segment bigger than ANON_MAX_SIZE.
144  * Each chunk points to a VM named entry of up to ANON_MAX_SIZE bytes
145  * of anonymous memory.
146  */
147 struct shm_handle {
148 	void * shm_object;                      /* named entry for this chunk*/
149 	memory_object_size_t shm_handle_size;   /* size of this chunk */
150 	struct shm_handle *shm_handle_next;     /* next chunk */
151 };
152 
153 struct shmmap_state {
154 	mach_vm_address_t va;           /* user address */
155 	int shmid;                      /* segment id */
156 };
157 
158 static void shm_deallocate_segment(struct shmid_kernel *);
159 static int shm_find_segment_by_key(key_t);
160 static struct shmid_kernel *shm_find_segment_by_shmid(int);
161 static int shm_delete_mapping(struct proc *, struct shmmap_state *, int);
162 
163 #ifdef __APPLE_API_PRIVATE
164 #define DEFAULT_SHMMAX  (4 * 1024 * 1024)
165 #define DEFAULT_SHMMIN  1
166 #define DEFAULT_SHMMNI  32
167 #define DEFAULT_SHMSEG  8
168 #define DEFAULT_SHMALL  1024
169 
170 struct shminfo shminfo = {
171 	.shmmax = DEFAULT_SHMMAX,
172 	.shmmin = DEFAULT_SHMMIN,
173 	.shmmni = DEFAULT_SHMMNI,
174 	.shmseg = DEFAULT_SHMSEG,
175 	.shmall = DEFAULT_SHMALL
176 };
177 
178 #define SHMID_IS_VALID(x) ((x) >= 0)
179 #define SHMID_UNALLOCATED (-1)
180 #define SHMID_SENTINEL    (-2)
181 
182 #endif /* __APPLE_API_PRIVATE */
183 
184 static __inline__ time_t
sysv_shmtime(void)185 sysv_shmtime(void)
186 {
187 	struct timeval  tv;
188 	microtime(&tv);
189 	return tv.tv_sec;
190 }
191 
192 /*
193  * This conversion is safe, since if we are converting for a 32 bit process,
194  * then it's value of (struct shmid_ds)->shm_segsz will never exceed 4G.
195  *
196  * NOTE: Source and target may *NOT* overlap! (target is smaller)
197  */
198 static void
shmid_ds_64to32(struct user_shmid_ds * in,struct user32_shmid_ds * out)199 shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out)
200 {
201 	out->shm_perm = in->shm_perm;
202 	out->shm_segsz = in->shm_segsz;
203 	out->shm_lpid = in->shm_lpid;
204 	out->shm_cpid = in->shm_cpid;
205 	out->shm_nattch = in->shm_nattch;
206 	out->shm_atime = in->shm_atime;
207 	out->shm_dtime = in->shm_dtime;
208 	out->shm_ctime = in->shm_ctime;
209 	out->shm_internal = CAST_DOWN_EXPLICIT(int, in->shm_internal);
210 }
211 
212 /*
213  * NOTE: Source and target may are permitted to overlap! (source is smaller);
214  * this works because we copy fields in order from the end of the struct to
215  * the beginning.
216  */
217 static void
shmid_ds_32to64(struct user32_shmid_ds * in,struct user_shmid_ds * out)218 shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out)
219 {
220 	out->shm_internal = in->shm_internal;
221 	out->shm_ctime = in->shm_ctime;
222 	out->shm_dtime = in->shm_dtime;
223 	out->shm_atime = in->shm_atime;
224 	out->shm_nattch = in->shm_nattch;
225 	out->shm_cpid = in->shm_cpid;
226 	out->shm_lpid = in->shm_lpid;
227 	out->shm_segsz = in->shm_segsz;
228 	out->shm_perm = in->shm_perm;
229 }
230 
231 
232 static int
shm_find_segment_by_key(key_t key)233 shm_find_segment_by_key(key_t key)
234 {
235 	int i;
236 
237 	for (i = 0; i < shminfo.shmmni; i++) {
238 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
239 		    shmsegs[i].u.shm_perm._key == key) {
240 			return i;
241 		}
242 	}
243 	return -1;
244 }
245 
246 static struct shmid_kernel *
shm_find_segment_by_shmid(int shmid)247 shm_find_segment_by_shmid(int shmid)
248 {
249 	int segnum;
250 	struct shmid_kernel *shmseg;
251 
252 	segnum = IPCID_TO_IX(shmid);
253 	if (segnum < 0 || segnum >= shminfo.shmmni) {
254 		return NULL;
255 	}
256 	shmseg = &shmsegs[segnum];
257 	if ((shmseg->u.shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
258 	    != SHMSEG_ALLOCATED ||
259 	    shmseg->u.shm_perm._seq != IPCID_TO_SEQ(shmid)) {
260 		return NULL;
261 	}
262 	return shmseg;
263 }
264 
265 static void
shm_deallocate_segment(struct shmid_kernel * shmseg)266 shm_deallocate_segment(struct shmid_kernel *shmseg)
267 {
268 	struct shm_handle *shm_handle, *shm_handle_next;
269 	mach_vm_size_t size;
270 
271 	for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal); /* tunnel */
272 	    shm_handle != NULL;
273 	    shm_handle = shm_handle_next) {
274 		shm_handle_next = shm_handle->shm_handle_next;
275 		mach_memory_entry_port_release(shm_handle->shm_object);
276 		kfree_type(struct shm_handle, shm_handle);
277 	}
278 	shmseg->u.shm_internal = USER_ADDR_NULL;                /* tunnel */
279 	size = vm_map_round_page(shmseg->u.shm_segsz,
280 	    vm_map_page_mask(current_map()));
281 	shm_committed -= btoc(size);
282 	shm_nused--;
283 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
284 #if CONFIG_MACF
285 	/* Reset the MAC label */
286 	mac_sysvshm_label_recycle(shmseg);
287 #endif
288 }
289 
290 static int
shm_delete_mapping(__unused struct proc * p,struct shmmap_state * shmmap_s,int deallocate)291 shm_delete_mapping(__unused struct proc *p, struct shmmap_state *shmmap_s,
292     int deallocate)
293 {
294 	struct shmid_kernel *shmseg;
295 	int segnum, result;
296 	mach_vm_size_t size;
297 
298 	segnum = IPCID_TO_IX(shmmap_s->shmid);
299 	shmseg = &shmsegs[segnum];
300 	size = vm_map_round_page(shmseg->u.shm_segsz,
301 	    vm_map_page_mask(current_map())); /* XXX done for us? */
302 	if (deallocate) {
303 		result = mach_vm_deallocate(current_map(), shmmap_s->va, size);
304 		if (result != KERN_SUCCESS) {
305 			return EINVAL;
306 		}
307 	}
308 	shmmap_s->shmid = SHMID_UNALLOCATED;
309 	shmseg->u.shm_dtime = sysv_shmtime();
310 	if ((--shmseg->u.shm_nattch <= 0) &&
311 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
312 		shm_deallocate_segment(shmseg);
313 		shm_last_free = segnum;
314 	}
315 	return 0;
316 }
317 
318 int
shmdt(struct proc * p,struct shmdt_args * uap,int32_t * retval)319 shmdt(struct proc *p, struct shmdt_args *uap, int32_t *retval)
320 {
321 #if CONFIG_MACF
322 	struct shmid_kernel *shmsegptr;
323 #endif
324 	struct shmmap_state *shmmap_s;
325 	int i;
326 	int shmdtret = 0;
327 
328 	AUDIT_ARG(svipc_addr, uap->shmaddr);
329 
330 	SYSV_SHM_SUBSYS_LOCK();
331 
332 	if ((shmdtret = shminit())) {
333 		goto shmdt_out;
334 	}
335 
336 	shmmap_s = (struct shmmap_state *)p->vm_shm;
337 	if (shmmap_s == NULL) {
338 		shmdtret = EINVAL;
339 		goto shmdt_out;
340 	}
341 
342 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
343 		if (SHMID_IS_VALID(shmmap_s->shmid) &&
344 		    shmmap_s->va == (mach_vm_offset_t)uap->shmaddr) {
345 			break;
346 		}
347 	}
348 
349 	if (!SHMID_IS_VALID(shmmap_s->shmid)) {
350 		shmdtret = EINVAL;
351 		goto shmdt_out;
352 	}
353 
354 #if CONFIG_MACF
355 	/*
356 	 * XXX: It might be useful to move this into the shm_delete_mapping
357 	 * function
358 	 */
359 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
360 	shmdtret = mac_sysvshm_check_shmdt(kauth_cred_get(), shmsegptr);
361 	if (shmdtret) {
362 		goto shmdt_out;
363 	}
364 #endif
365 	i = shm_delete_mapping(p, shmmap_s, 1);
366 
367 	if (i == 0) {
368 		*retval = 0;
369 	}
370 	shmdtret = i;
371 shmdt_out:
372 	SYSV_SHM_SUBSYS_UNLOCK();
373 	return shmdtret;
374 }
375 
376 int
shmat(struct proc * p,struct shmat_args * uap,user_addr_t * retval)377 shmat(struct proc *p, struct shmat_args *uap, user_addr_t *retval)
378 {
379 	int error, flags;
380 	struct shmid_kernel     *shmseg;
381 	struct shmmap_state     *shmmap_s = NULL;
382 	struct shm_handle       *shm_handle;
383 	mach_vm_address_t       attach_va;      /* attach address in/out */
384 	mach_vm_address_t       shmlba;
385 	mach_vm_size_t          map_size;       /* size of map entry */
386 	mach_vm_size_t          mapped_size;
387 	vm_prot_t               prot;
388 	kern_return_t           rv;
389 	int                     shmat_ret;
390 	vm_map_kernel_flags_t   vmk_flags;
391 
392 	shmat_ret = 0;
393 
394 	AUDIT_ARG(svipc_id, uap->shmid);
395 	AUDIT_ARG(svipc_addr, uap->shmaddr);
396 
397 	SYSV_SHM_SUBSYS_LOCK();
398 
399 	if ((shmat_ret = shminit())) {
400 		goto shmat_out;
401 	}
402 
403 	shmmap_s = (struct shmmap_state *)p->vm_shm;
404 	if (shmmap_s == NULL) {
405 		/* lazily allocate the shm map */
406 
407 		int nsegs = shminfo.shmseg;
408 		if (nsegs <= 0) {
409 			shmat_ret = EMFILE;
410 			goto shmat_out;
411 		}
412 
413 		/* +1 for the sentinel */
414 		shmmap_s = kalloc_type(struct shmmap_state, nsegs + 1, Z_WAITOK);
415 		if (shmmap_s == NULL) {
416 			shmat_ret = ENOMEM;
417 			goto shmat_out;
418 		}
419 
420 		/* initialize the entries */
421 		for (int i = 0; i < nsegs; i++) {
422 			shmmap_s[i].shmid = SHMID_UNALLOCATED;
423 		}
424 		shmmap_s[nsegs].shmid = SHMID_SENTINEL;
425 
426 		p->vm_shm = (caddr_t)shmmap_s;
427 	}
428 
429 	shmseg = shm_find_segment_by_shmid(uap->shmid);
430 	if (shmseg == NULL) {
431 		shmat_ret = EINVAL;
432 		goto shmat_out;
433 	}
434 
435 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
436 	error = ipcperm(kauth_cred_get(), &shmseg->u.shm_perm,
437 	    (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R | IPC_W);
438 	if (error) {
439 		shmat_ret = error;
440 		goto shmat_out;
441 	}
442 
443 #if CONFIG_MACF
444 	error = mac_sysvshm_check_shmat(kauth_cred_get(), shmseg, uap->shmflg);
445 	if (error) {
446 		shmat_ret = error;
447 		goto shmat_out;
448 	}
449 #endif
450 
451 	/* find a free shmid */
452 	while (SHMID_IS_VALID(shmmap_s->shmid)) {
453 		shmmap_s++;
454 	}
455 	if (shmmap_s->shmid != SHMID_UNALLOCATED) {
456 		/* no free shmids */
457 		shmat_ret = EMFILE;
458 		goto shmat_out;
459 	}
460 
461 	map_size = vm_map_round_page(shmseg->u.shm_segsz,
462 	    vm_map_page_mask(current_map()));
463 	prot = VM_PROT_READ;
464 	if ((uap->shmflg & SHM_RDONLY) == 0) {
465 		prot |= VM_PROT_WRITE;
466 	}
467 	flags = MAP_ANON | MAP_SHARED;
468 	if (uap->shmaddr) {
469 		flags |= MAP_FIXED;
470 	}
471 
472 	attach_va = (mach_vm_address_t)uap->shmaddr;
473 	shmlba = vm_map_page_size(current_map()); /* XXX instead of SHMLBA */
474 	if (uap->shmflg & SHM_RND) {
475 		attach_va &= ~(shmlba - 1);
476 	} else if ((attach_va & (shmlba - 1)) != 0) {
477 		shmat_ret = EINVAL;
478 		goto shmat_out;
479 	}
480 
481 	if (flags & MAP_FIXED) {
482 		vmk_flags = VM_MAP_KERNEL_FLAGS_FIXED();
483 	} else {
484 		vmk_flags = VM_MAP_KERNEL_FLAGS_ANYWHERE();
485 	}
486 
487 	mapped_size = 0;
488 
489 	/* first reserve enough space... */
490 	rv = mach_vm_map_kernel(current_map(),
491 	    &attach_va,
492 	    map_size,
493 	    0,
494 	    vmk_flags,
495 	    IPC_PORT_NULL,
496 	    0,
497 	    FALSE,
498 	    VM_PROT_NONE,
499 	    VM_PROT_NONE,
500 	    VM_INHERIT_NONE);
501 	if (rv != KERN_SUCCESS) {
502 		goto out;
503 	}
504 
505 	shmmap_s->va = attach_va;
506 
507 	/* ... then map the shared memory over the reserved space */
508 	for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal);/* tunnel */
509 	    shm_handle != NULL;
510 	    shm_handle = shm_handle->shm_handle_next) {
511 		vm_map_size_t chunk_size;
512 
513 		assert(mapped_size < map_size);
514 		chunk_size = shm_handle->shm_handle_size;
515 		if (chunk_size > map_size - mapped_size) {
516 			/*
517 			 * Partial mapping of last chunk due to
518 			 * page size mismatch.
519 			 */
520 			assert(vm_map_page_shift(current_map()) < PAGE_SHIFT);
521 			assert(shm_handle->shm_handle_next == NULL);
522 			chunk_size = map_size - mapped_size;
523 		}
524 		rv = vm_map_enter_mem_object(
525 			current_map(),          /* process map */
526 			&attach_va,             /* attach address */
527 			chunk_size,             /* size to map */
528 			(mach_vm_offset_t)0,    /* alignment mask */
529 			VM_MAP_KERNEL_FLAGS_FIXED(.vmf_overwrite = true),
530 			shm_handle->shm_object,
531 			(mach_vm_offset_t)0,
532 			FALSE,
533 			prot,
534 			prot,
535 			VM_INHERIT_SHARE);
536 		if (rv != KERN_SUCCESS) {
537 			goto out;
538 		}
539 
540 		mapped_size += chunk_size;
541 		attach_va = attach_va + chunk_size;
542 	}
543 
544 	shmmap_s->shmid = uap->shmid;
545 	shmseg->u.shm_lpid = proc_getpid(p);
546 	shmseg->u.shm_atime = sysv_shmtime();
547 	shmseg->u.shm_nattch++;
548 	*retval = shmmap_s->va; /* XXX return -1 on error */
549 	shmat_ret = 0;
550 	goto shmat_out;
551 out:
552 	if (mapped_size > 0) {
553 		(void) mach_vm_deallocate(current_map(),
554 		    shmmap_s->va,
555 		    mapped_size);
556 	}
557 	switch (rv) {
558 	case KERN_INVALID_ADDRESS:
559 	case KERN_NO_SPACE:
560 		shmat_ret = ENOMEM;
561 		break;
562 	case KERN_PROTECTION_FAILURE:
563 		shmat_ret = EACCES;
564 		break;
565 	default:
566 		shmat_ret = EINVAL;
567 		break;
568 	}
569 shmat_out:
570 	SYSV_SHM_SUBSYS_UNLOCK();
571 	return shmat_ret;
572 }
573 
574 static int
oshmctl(__unused void * p,__unused void * uap,__unused void * retval)575 oshmctl(__unused void *p, __unused void *uap, __unused void *retval)
576 {
577 	return EINVAL;
578 }
579 
580 /*
581  * Returns:	0			Success
582  *		EINVAL
583  *	copyout:EFAULT
584  *	copyin:EFAULT
585  *	ipcperm:EPERM
586  *	ipcperm:EACCES
587  */
588 int
shmctl(__unused struct proc * p,struct shmctl_args * uap,int32_t * retval)589 shmctl(__unused struct proc *p, struct shmctl_args *uap, int32_t *retval)
590 {
591 	int error;
592 	kauth_cred_t cred = kauth_cred_get();
593 	struct user_shmid_ds inbuf;
594 	struct shmid_kernel *shmseg;
595 
596 	int shmctl_ret = 0;
597 
598 	AUDIT_ARG(svipc_cmd, uap->cmd);
599 	AUDIT_ARG(svipc_id, uap->shmid);
600 
601 	SYSV_SHM_SUBSYS_LOCK();
602 
603 	if ((shmctl_ret = shminit())) {
604 		goto shmctl_out;
605 	}
606 
607 	shmseg = shm_find_segment_by_shmid(uap->shmid);
608 	if (shmseg == NULL) {
609 		shmctl_ret = EINVAL;
610 		goto shmctl_out;
611 	}
612 
613 	/* XXAUDIT: This is the perms BEFORE any change by this call. This
614 	 * may not be what is desired.
615 	 */
616 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
617 
618 #if CONFIG_MACF
619 	error = mac_sysvshm_check_shmctl(cred, shmseg, uap->cmd);
620 	if (error) {
621 		shmctl_ret = error;
622 		goto shmctl_out;
623 	}
624 #endif
625 	switch (uap->cmd) {
626 	case IPC_STAT:
627 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_R);
628 		if (error) {
629 			shmctl_ret = error;
630 			goto shmctl_out;
631 		}
632 
633 		if (IS_64BIT_PROCESS(p)) {
634 			struct user_shmid_ds shmid_ds = {};
635 			memcpy(&shmid_ds, &shmseg->u, sizeof(struct user_shmid_ds));
636 
637 			/* Clear kernel reserved pointer before copying to user space */
638 			shmid_ds.shm_internal = USER_ADDR_NULL;
639 
640 			error = copyout(&shmid_ds, uap->buf, sizeof(shmid_ds));
641 		} else {
642 			struct user32_shmid_ds shmid_ds32 = {};
643 			shmid_ds_64to32(&shmseg->u, &shmid_ds32);
644 
645 			/* Clear kernel reserved pointer before copying to user space */
646 			shmid_ds32.shm_internal = (user32_addr_t)0;
647 
648 			error = copyout(&shmid_ds32, uap->buf, sizeof(shmid_ds32));
649 		}
650 		if (error) {
651 			shmctl_ret = error;
652 			goto shmctl_out;
653 		}
654 		break;
655 	case IPC_SET:
656 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_M);
657 		if (error) {
658 			shmctl_ret = error;
659 			goto shmctl_out;
660 		}
661 		if (IS_64BIT_PROCESS(p)) {
662 			error = copyin(uap->buf, &inbuf, sizeof(struct user_shmid_ds));
663 		} else {
664 			struct user32_shmid_ds shmid_ds32;
665 			error = copyin(uap->buf, &shmid_ds32, sizeof(shmid_ds32));
666 			/* convert in place; ugly, but safe */
667 			shmid_ds_32to64(&shmid_ds32, &inbuf);
668 		}
669 		if (error) {
670 			shmctl_ret = error;
671 			goto shmctl_out;
672 		}
673 		shmseg->u.shm_perm.uid = inbuf.shm_perm.uid;
674 		shmseg->u.shm_perm.gid = inbuf.shm_perm.gid;
675 		shmseg->u.shm_perm.mode =
676 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
677 		    (inbuf.shm_perm.mode & ACCESSPERMS);
678 		shmseg->u.shm_ctime = sysv_shmtime();
679 		break;
680 	case IPC_RMID:
681 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_M);
682 		if (error) {
683 			shmctl_ret = error;
684 			goto shmctl_out;
685 		}
686 		shmseg->u.shm_perm._key = IPC_PRIVATE;
687 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
688 		if (shmseg->u.shm_nattch <= 0) {
689 			shm_deallocate_segment(shmseg);
690 			shm_last_free = IPCID_TO_IX(uap->shmid);
691 		}
692 		break;
693 #if 0
694 	case SHM_LOCK:
695 	case SHM_UNLOCK:
696 #endif
697 	default:
698 		shmctl_ret = EINVAL;
699 		goto shmctl_out;
700 	}
701 	*retval = 0;
702 	shmctl_ret = 0;
703 shmctl_out:
704 	SYSV_SHM_SUBSYS_UNLOCK();
705 	return shmctl_ret;
706 }
707 
708 static int
shmget_existing(struct shmget_args * uap,int mode,int segnum,int * retval)709 shmget_existing(struct shmget_args *uap, int mode, int segnum, int *retval)
710 {
711 	struct shmid_kernel *shmseg;
712 	int error = 0;
713 
714 	shmseg = &shmsegs[segnum];
715 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
716 		/*
717 		 * This segment is in the process of being allocated.  Wait
718 		 * until it's done, and look the key up again (in case the
719 		 * allocation failed or it was freed).
720 		 */
721 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
722 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
723 		if (error) {
724 			return error;
725 		}
726 		return EAGAIN;
727 	}
728 
729 	/*
730 	 * The low 9 bits of shmflag are the mode bits being requested, which
731 	 * are the actual mode bits desired on the segment, and not in IPC_R
732 	 * form; therefore it would be incorrect to call ipcperm() to validate
733 	 * them; instead, we AND the existing mode with the requested mode, and
734 	 * verify that it matches the requested mode; otherwise, we fail with
735 	 * EACCES (access denied).
736 	 */
737 	if ((shmseg->u.shm_perm.mode & mode) != mode) {
738 		return EACCES;
739 	}
740 
741 #if CONFIG_MACF
742 	error = mac_sysvshm_check_shmget(kauth_cred_get(), shmseg, uap->shmflg);
743 	if (error) {
744 		return error;
745 	}
746 #endif
747 
748 	if (uap->size && uap->size > shmseg->u.shm_segsz) {
749 		return EINVAL;
750 	}
751 
752 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) {
753 		return EEXIST;
754 	}
755 
756 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
757 	return 0;
758 }
759 
760 static int
shmget_allocate_segment(struct proc * p,struct shmget_args * uap,int mode,int * retval)761 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode,
762     int *retval)
763 {
764 	int i, segnum, shmid;
765 	kauth_cred_t cred = kauth_cred_get();
766 	struct shmid_kernel *shmseg;
767 	struct shm_handle *shm_handle;
768 	kern_return_t kret;
769 	mach_vm_size_t total_size, size = 0, alloc_size;
770 	void * mem_object;
771 	struct shm_handle *shm_handle_next, **shm_handle_next_p;
772 
773 	if (uap->size <= 0 ||
774 	    uap->size < (user_size_t)shminfo.shmmin ||
775 	    uap->size > (user_size_t)shminfo.shmmax) {
776 		return EINVAL;
777 	}
778 	if (shm_nused >= shminfo.shmmni) { /* any shmids left? */
779 		return ENOSPC;
780 	}
781 	if (mach_vm_round_page_overflow(uap->size, &total_size)) {
782 		return EINVAL;
783 	}
784 	if ((user_ssize_t)(shm_committed + btoc(total_size)) > shminfo.shmall) {
785 		return ENOMEM;
786 	}
787 	if (shm_last_free < 0) {
788 		for (i = 0; i < shminfo.shmmni; i++) {
789 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE) {
790 				break;
791 			}
792 		}
793 		if (i == shminfo.shmmni) {
794 			panic("shmseg free count inconsistent");
795 		}
796 		segnum = i;
797 	} else {
798 		segnum = shm_last_free;
799 		shm_last_free = -1;
800 	}
801 	shmseg = &shmsegs[segnum];
802 
803 	/*
804 	 * In case we sleep in malloc(), mark the segment present but deleted
805 	 * so that noone else tries to create the same key.
806 	 * XXX but we don't release the global lock !?
807 	 */
808 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
809 	shmseg->u.shm_perm._key = uap->key;
810 	shmseg->u.shm_perm._seq = (shmseg->u.shm_perm._seq + 1) & 0x7fff;
811 
812 	shm_handle_next_p = NULL;
813 	for (alloc_size = 0;
814 	    alloc_size < total_size;
815 	    alloc_size += size) {
816 		size = MIN(total_size - alloc_size, ANON_MAX_SIZE);
817 		kret = mach_make_memory_entry_64(
818 			VM_MAP_NULL,
819 			(memory_object_size_t *) &size,
820 			(memory_object_offset_t) 0,
821 			MAP_MEM_NAMED_CREATE | VM_PROT_DEFAULT,
822 			(ipc_port_t *) &mem_object, 0);
823 		if (kret != KERN_SUCCESS) {
824 			goto out;
825 		}
826 
827 		shm_handle = kalloc_type(struct shm_handle, Z_WAITOK | Z_NOFAIL);
828 		shm_handle->shm_object = mem_object;
829 		shm_handle->shm_handle_size = size;
830 		shm_handle->shm_handle_next = NULL;
831 		if (shm_handle_next_p == NULL) {
832 			shmseg->u.shm_internal = CAST_USER_ADDR_T(shm_handle);/* tunnel */
833 		} else {
834 			*shm_handle_next_p = shm_handle;
835 		}
836 		shm_handle_next_p = &shm_handle->shm_handle_next;
837 	}
838 
839 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
840 
841 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = kauth_cred_getuid(cred);
842 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = kauth_cred_getgid(cred);
843 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
844 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
845 	shmseg->u.shm_segsz = uap->size;
846 	shmseg->u.shm_cpid = proc_getpid(p);
847 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
848 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
849 #if CONFIG_MACF
850 	mac_sysvshm_label_associate(cred, shmseg);
851 #endif
852 	shmseg->u.shm_ctime = sysv_shmtime();
853 	shm_committed += btoc(size);
854 	shm_nused++;
855 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
856 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
857 		/*
858 		 * Somebody else wanted this key while we were asleep.  Wake
859 		 * them up now.
860 		 */
861 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
862 		wakeup((caddr_t)shmseg);
863 	}
864 	*retval = shmid;
865 	AUDIT_ARG(svipc_id, shmid);
866 	return 0;
867 out:
868 	if (kret != KERN_SUCCESS) {
869 		for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal); /* tunnel */
870 		    shm_handle != NULL;
871 		    shm_handle = shm_handle_next) {
872 			shm_handle_next = shm_handle->shm_handle_next;
873 			mach_memory_entry_port_release(shm_handle->shm_object);
874 			kfree_type(struct shm_handle, shm_handle);
875 		}
876 		shmseg->u.shm_internal = USER_ADDR_NULL; /* tunnel */
877 	}
878 
879 	switch (kret) {
880 	case KERN_INVALID_ADDRESS:
881 	case KERN_NO_SPACE:
882 		return ENOMEM;
883 	case KERN_PROTECTION_FAILURE:
884 		return EACCES;
885 	default:
886 		return EINVAL;
887 	}
888 }
889 
890 int
shmget(struct proc * p,struct shmget_args * uap,int32_t * retval)891 shmget(struct proc *p, struct shmget_args *uap, int32_t *retval)
892 {
893 	int segnum, mode, error;
894 	int shmget_ret = 0;
895 
896 	/* Auditing is actually done in shmget_allocate_segment() */
897 
898 	SYSV_SHM_SUBSYS_LOCK();
899 
900 	if ((shmget_ret = shminit())) {
901 		goto shmget_out;
902 	}
903 
904 	mode = uap->shmflg & ACCESSPERMS;
905 	if (uap->key != IPC_PRIVATE) {
906 again:
907 		segnum = shm_find_segment_by_key(uap->key);
908 		if (segnum >= 0) {
909 			error = shmget_existing(uap, mode, segnum, retval);
910 			if (error == EAGAIN) {
911 				goto again;
912 			}
913 			shmget_ret = error;
914 			goto shmget_out;
915 		}
916 		if ((uap->shmflg & IPC_CREAT) == 0) {
917 			shmget_ret = ENOENT;
918 			goto shmget_out;
919 		}
920 	}
921 	shmget_ret = shmget_allocate_segment(p, uap, mode, retval);
922 shmget_out:
923 	SYSV_SHM_SUBSYS_UNLOCK();
924 	return shmget_ret;
925 }
926 
927 /*
928  * shmsys
929  *
930  * Entry point for all SHM calls: shmat, oshmctl, shmdt, shmget, shmctl
931  *
932  * Parameters:	p	Process requesting the call
933  *              uap	User argument descriptor (see below)
934  *              retval	Return value of the selected shm call
935  *
936  * Indirect parameters:	uap->which	msg call to invoke (index in array of shm calls)
937  *                      uap->a2		User argument descriptor
938  *
939  * Returns:	0	Success
940  *              !0	Not success
941  *
942  * Implicit returns: retval     Return value of the selected shm call
943  *
944  * DEPRECATED:  This interface should not be used to call the other SHM
945  *              functions (shmat, oshmctl, shmdt, shmget, shmctl). The correct
946  *              usage is to call the other SHM functions directly.
947  */
948 int
shmsys(struct proc * p,struct shmsys_args * uap,int32_t * retval)949 shmsys(struct proc *p, struct shmsys_args *uap, int32_t *retval)
950 {
951 	/* The routine that we are dispatching already does this */
952 
953 	if (uap->which >= sizeof(shmcalls) / sizeof(shmcalls[0])) {
954 		return EINVAL;
955 	}
956 	return (*shmcalls[uap->which])(p, &uap->a2, retval);
957 }
958 
959 /*
960  * Return 0 on success, 1 on failure.
961  */
962 int
shmfork(struct proc * p1,struct proc * p2)963 shmfork(struct proc *p1, struct proc *p2)
964 {
965 	struct shmmap_state *shmmap_s;
966 	int nsegs = 0;
967 	int ret = 0;
968 
969 	SYSV_SHM_SUBSYS_LOCK();
970 
971 	if (shminit()) {
972 		ret = 1;
973 		goto shmfork_out;
974 	}
975 
976 	struct shmmap_state *src = (struct shmmap_state *)p1->vm_shm;
977 	assert(src);
978 
979 	/* count number of shmid entries in src */
980 	for (struct shmmap_state *s = src; s->shmid != SHMID_SENTINEL; s++) {
981 		nsegs++;
982 	}
983 
984 	shmmap_s = kalloc_type(struct shmmap_state, nsegs + 1, Z_WAITOK);
985 	if (shmmap_s == NULL) {
986 		ret = 1;
987 		goto shmfork_out;
988 	}
989 
990 	bcopy(src, (caddr_t)shmmap_s, (nsegs + 1) * sizeof(struct shmmap_state));
991 	p2->vm_shm = (caddr_t)shmmap_s;
992 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
993 		if (SHMID_IS_VALID(shmmap_s->shmid)) {
994 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
995 		}
996 	}
997 
998 shmfork_out:
999 	SYSV_SHM_SUBSYS_UNLOCK();
1000 	return ret;
1001 }
1002 
1003 static void
shmcleanup(struct proc * p,int deallocate)1004 shmcleanup(struct proc *p, int deallocate)
1005 {
1006 	struct shmmap_state *shmmap_s;
1007 	int nsegs = 0;
1008 
1009 	SYSV_SHM_SUBSYS_LOCK();
1010 
1011 	shmmap_s = (struct shmmap_state *)p->vm_shm;
1012 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
1013 		nsegs++;
1014 		if (SHMID_IS_VALID(shmmap_s->shmid)) {
1015 			/*
1016 			 * XXX: Should the MAC framework enforce
1017 			 * check here as well.
1018 			 */
1019 			shm_delete_mapping(p, shmmap_s, deallocate);
1020 		}
1021 	}
1022 
1023 	kfree_type(struct shmmap_state, nsegs + 1, p->vm_shm);
1024 	SYSV_SHM_SUBSYS_UNLOCK();
1025 }
1026 
1027 void
shmexit(struct proc * p)1028 shmexit(struct proc *p)
1029 {
1030 	shmcleanup(p, 1);
1031 }
1032 
1033 /*
1034  * shmexec() is like shmexit(), only it doesn't delete the mappings,
1035  * since the old address space has already been destroyed and the new
1036  * one instantiated.  Instead, it just does the housekeeping work we
1037  * need to do to keep the System V shared memory subsystem sane.
1038  */
1039 __private_extern__ void
shmexec(struct proc * p)1040 shmexec(struct proc *p)
1041 {
1042 	shmcleanup(p, 0);
1043 }
1044 
1045 int
shminit(void)1046 shminit(void)
1047 {
1048 	size_t sz;
1049 	int i;
1050 
1051 	if (!shm_inited) {
1052 		/*
1053 		 * we store internally 64 bit, since if we didn't, we would
1054 		 * be unable to represent a segment size in excess of 32 bits
1055 		 * with the (struct shmid_ds)->shm_segsz field; also, POSIX
1056 		 * dictates this filed be a size_t, which is 64 bits when
1057 		 * running 64 bit binaries.
1058 		 */
1059 		if (os_mul_overflow(shminfo.shmmni, sizeof(struct shmid_kernel), &sz)) {
1060 			return ENOMEM;
1061 		}
1062 
1063 		shmsegs = zalloc_permanent(sz, ZALIGN_PTR);
1064 		if (shmsegs == NULL) {
1065 			return ENOMEM;
1066 		}
1067 		for (i = 0; i < shminfo.shmmni; i++) {
1068 			shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
1069 			shmsegs[i].u.shm_perm._seq = 0;
1070 #if CONFIG_MACF
1071 			mac_sysvshm_label_init(&shmsegs[i]);
1072 #endif
1073 		}
1074 		shm_last_free = 0;
1075 		shm_nused = 0;
1076 		shm_committed = 0;
1077 		shm_inited = 1;
1078 	}
1079 
1080 	return 0;
1081 }
1082 
1083 /* (struct sysctl_oid *oidp, void *arg1, int arg2, \
1084  *       struct sysctl_req *req) */
1085 static int
sysctl_shminfo(__unused struct sysctl_oid * oidp,void * arg1,__unused int arg2,struct sysctl_req * req)1086 sysctl_shminfo(__unused struct sysctl_oid *oidp, void *arg1,
1087     __unused int arg2, struct sysctl_req *req)
1088 {
1089 	int error = 0;
1090 	int sysctl_shminfo_ret = 0;
1091 	int64_t saved_shmmax;
1092 	int64_t saved_shmmin;
1093 	int64_t saved_shmseg;
1094 	int64_t saved_shmmni;
1095 	int64_t saved_shmall;
1096 
1097 	error = SYSCTL_OUT(req, arg1, sizeof(int64_t));
1098 	if (error || req->newptr == USER_ADDR_NULL) {
1099 		return error;
1100 	}
1101 
1102 	SYSV_SHM_SUBSYS_LOCK();
1103 
1104 	/* shmmni can not be changed after SysV SHM has been initialized */
1105 	if (shm_inited && arg1 == &shminfo.shmmni) {
1106 		sysctl_shminfo_ret = EPERM;
1107 		goto sysctl_shminfo_out;
1108 	}
1109 	saved_shmmax = shminfo.shmmax;
1110 	saved_shmmin = shminfo.shmmin;
1111 	saved_shmseg = shminfo.shmseg;
1112 	saved_shmmni = shminfo.shmmni;
1113 	saved_shmall = shminfo.shmall;
1114 
1115 	if ((error = SYSCTL_IN(req, arg1, sizeof(int64_t))) != 0) {
1116 		sysctl_shminfo_ret = error;
1117 		goto sysctl_shminfo_out;
1118 	}
1119 
1120 	if (arg1 == &shminfo.shmmax) {
1121 		/* shmmax needs to be page-aligned */
1122 		if (shminfo.shmmax & PAGE_MASK_64 || shminfo.shmmax < 0) {
1123 			shminfo.shmmax = saved_shmmax;
1124 			sysctl_shminfo_ret = EINVAL;
1125 			goto sysctl_shminfo_out;
1126 		}
1127 	} else if (arg1 == &shminfo.shmmin) {
1128 		if (shminfo.shmmin < 0) {
1129 			shminfo.shmmin = saved_shmmin;
1130 			sysctl_shminfo_ret = EINVAL;
1131 			goto sysctl_shminfo_out;
1132 		}
1133 	} else if (arg1 == &shminfo.shmseg) {
1134 		/* add a sanity check - 20847256 */
1135 		if (shminfo.shmseg > INT32_MAX || shminfo.shmseg < 0) {
1136 			shminfo.shmseg = saved_shmseg;
1137 			sysctl_shminfo_ret = EINVAL;
1138 			goto sysctl_shminfo_out;
1139 		}
1140 	} else if (arg1 == &shminfo.shmmni) {
1141 		/* add a sanity check - 20847256 */
1142 		if (shminfo.shmmni > INT32_MAX || shminfo.shmmni < 0) {
1143 			shminfo.shmmni = saved_shmmni;
1144 			sysctl_shminfo_ret = EINVAL;
1145 			goto sysctl_shminfo_out;
1146 		}
1147 	} else if (arg1 == &shminfo.shmall) {
1148 		/* add a sanity check - 20847256 */
1149 		if (shminfo.shmall > INT32_MAX || shminfo.shmall < 0) {
1150 			shminfo.shmall = saved_shmall;
1151 			sysctl_shminfo_ret = EINVAL;
1152 			goto sysctl_shminfo_out;
1153 		}
1154 	}
1155 	sysctl_shminfo_ret = 0;
1156 sysctl_shminfo_out:
1157 	SYSV_SHM_SUBSYS_UNLOCK();
1158 	return sysctl_shminfo_ret;
1159 }
1160 
1161 static int
IPCS_shm_sysctl(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)1162 IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1,
1163     __unused int arg2, struct sysctl_req *req)
1164 {
1165 	int error;
1166 	int cursor;
1167 	union {
1168 		struct user32_IPCS_command u32;
1169 		struct user_IPCS_command u64;
1170 	} ipcs = { };
1171 	struct user32_shmid_ds shmid_ds32 = { }; /* post conversion, 32 bit version */
1172 	struct user_shmid_ds   shmid_ds = { };   /* 64 bit version */
1173 	void *shmid_dsp;
1174 	size_t ipcs_sz = sizeof(struct user_IPCS_command);
1175 	size_t shmid_ds_sz = sizeof(struct user_shmid_ds);
1176 	struct proc *p = current_proc();
1177 
1178 	SYSV_SHM_SUBSYS_LOCK();
1179 
1180 	if ((error = shminit())) {
1181 		goto ipcs_shm_sysctl_out;
1182 	}
1183 
1184 	if (!IS_64BIT_PROCESS(p)) {
1185 		ipcs_sz = sizeof(struct user32_IPCS_command);
1186 		shmid_ds_sz = sizeof(struct user32_shmid_ds);
1187 	}
1188 
1189 	/* Copy in the command structure */
1190 	if ((error = SYSCTL_IN(req, &ipcs, ipcs_sz)) != 0) {
1191 		goto ipcs_shm_sysctl_out;
1192 	}
1193 
1194 	if (!IS_64BIT_PROCESS(p)) {     /* convert in place */
1195 		ipcs.u64.ipcs_data = CAST_USER_ADDR_T(ipcs.u32.ipcs_data);
1196 	}
1197 
1198 	/* Let us version this interface... */
1199 	if (ipcs.u64.ipcs_magic != IPCS_MAGIC) {
1200 		error = EINVAL;
1201 		goto ipcs_shm_sysctl_out;
1202 	}
1203 
1204 	switch (ipcs.u64.ipcs_op) {
1205 	case IPCS_SHM_CONF:     /* Obtain global configuration data */
1206 		if (ipcs.u64.ipcs_datalen != sizeof(struct shminfo)) {
1207 			if (ipcs.u64.ipcs_cursor != 0) { /* fwd. compat. */
1208 				error = ENOMEM;
1209 				break;
1210 			}
1211 			error = ERANGE;
1212 			break;
1213 		}
1214 		error = copyout(&shminfo, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
1215 		break;
1216 
1217 	case IPCS_SHM_ITER:     /* Iterate over existing segments */
1218 		cursor = ipcs.u64.ipcs_cursor;
1219 		if (cursor < 0 || cursor >= shminfo.shmmni) {
1220 			error = ERANGE;
1221 			break;
1222 		}
1223 		if (ipcs.u64.ipcs_datalen != (int)shmid_ds_sz) {
1224 			error = EINVAL;
1225 			break;
1226 		}
1227 		for (; cursor < shminfo.shmmni; cursor++) {
1228 			if (shmsegs[cursor].u.shm_perm.mode & SHMSEG_ALLOCATED) {
1229 				break;
1230 			}
1231 			continue;
1232 		}
1233 		if (cursor == shminfo.shmmni) {
1234 			error = ENOENT;
1235 			break;
1236 		}
1237 
1238 		shmid_dsp = &shmsegs[cursor];   /* default: 64 bit */
1239 
1240 		/*
1241 		 * If necessary, convert the 64 bit kernel segment
1242 		 * descriptor to a 32 bit user one.
1243 		 */
1244 		if (!IS_64BIT_PROCESS(p)) {
1245 			shmid_ds_64to32(shmid_dsp, &shmid_ds32);
1246 
1247 			/* Clear kernel reserved pointer before copying to user space */
1248 			shmid_ds32.shm_internal = (user32_addr_t)0;
1249 
1250 			shmid_dsp = &shmid_ds32;
1251 		} else {
1252 			memcpy(&shmid_ds, shmid_dsp, sizeof(shmid_ds));
1253 
1254 			/* Clear kernel reserved pointer before copying to user space */
1255 			shmid_ds.shm_internal = USER_ADDR_NULL;
1256 
1257 			shmid_dsp = &shmid_ds;
1258 		}
1259 		error = copyout(shmid_dsp, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
1260 		if (!error) {
1261 			/* update cursor */
1262 			ipcs.u64.ipcs_cursor = cursor + 1;
1263 
1264 			if (!IS_64BIT_PROCESS(p)) { /* convert in place */
1265 				ipcs.u32.ipcs_data = CAST_DOWN_EXPLICIT(user32_addr_t, ipcs.u64.ipcs_data);
1266 			}
1267 
1268 			error = SYSCTL_OUT(req, &ipcs, ipcs_sz);
1269 		}
1270 		break;
1271 
1272 	default:
1273 		error = EINVAL;
1274 		break;
1275 	}
1276 ipcs_shm_sysctl_out:
1277 	SYSV_SHM_SUBSYS_UNLOCK();
1278 	return error;
1279 }
1280 
1281 SYSCTL_NODE(_kern, KERN_SYSV, sysv, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSV");
1282 
1283 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmax, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1284     &shminfo.shmmax, 0, &sysctl_shminfo, "Q", "shmmax");
1285 
1286 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmin, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1287     &shminfo.shmmin, 0, &sysctl_shminfo, "Q", "shmmin");
1288 
1289 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmni, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1290     &shminfo.shmmni, 0, &sysctl_shminfo, "Q", "shmmni");
1291 
1292 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmseg, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1293     &shminfo.shmseg, 0, &sysctl_shminfo, "Q", "shmseg");
1294 
1295 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmall, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1296     &shminfo.shmall, 0, &sysctl_shminfo, "Q", "shmall");
1297 
1298 SYSCTL_NODE(_kern_sysv, OID_AUTO, ipcs, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSVIPCS");
1299 
1300 SYSCTL_PROC(_kern_sysv_ipcs, OID_AUTO, shm, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED,
1301     0, 0, IPCS_shm_sysctl,
1302     "S,IPCS_shm_command",
1303     "ipcs shm command interface");
1304 #endif /* SYSV_SHM */
1305 
1306 /* DSEP Review Done pl-20051108-v02 @2743,@2908,@2913,@3009 */
1307