xref: /xnu-11417.140.69/bsd/kern/sysv_shm.c (revision 43a90889846e00bfb5cf1d255cdc0a701a1e05a4)
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_xnu.h>
95 #include <vm/vm_protos.h>
96 #include <vm/vm_memory_entry_xnu.h>
97 #include <vm/vm_kern_xnu.h>
98 
99 #include <kern/locks.h>
100 #include <os/overflow.h>
101 
102 /* Uncomment this line to see MAC debugging output. */
103 /* #define MAC_DEBUG */
104 #if CONFIG_MACF_DEBUG
105 #define MPRINTF(a)      printf a
106 #else
107 #define MPRINTF(a)
108 #endif
109 
110 #if SYSV_SHM
111 static int shminit(void);
112 
113 static LCK_GRP_DECLARE(sysv_shm_subsys_lck_grp, "sysv_shm_subsys_lock");
114 static LCK_MTX_DECLARE(sysv_shm_subsys_mutex, &sysv_shm_subsys_lck_grp);
115 
116 #define SYSV_SHM_SUBSYS_LOCK() lck_mtx_lock(&sysv_shm_subsys_mutex)
117 #define SYSV_SHM_SUBSYS_UNLOCK() lck_mtx_unlock(&sysv_shm_subsys_mutex)
118 
119 static int oshmctl(void *p, void *uap, void *retval);
120 static int shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, int * retval);
121 static int shmget_existing(struct shmget_args *uap, int mode, int segnum, int  * retval);
122 static void shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out);
123 static void shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out);
124 
125 /* XXX casting to (sy_call_t *) is bogus, as usual. */
126 static sy_call_t* const shmcalls[] = {
127 	(sy_call_t *)shmat, (sy_call_t *)oshmctl,
128 	(sy_call_t *)shmdt, (sy_call_t *)shmget,
129 	(sy_call_t *)shmctl
130 };
131 
132 #define SHMSEG_FREE             0x0200
133 #define SHMSEG_REMOVED          0x0400
134 #define SHMSEG_ALLOCATED        0x0800
135 #define SHMSEG_WANTED           0x1000
136 
137 static int shm_last_free, shm_nused, shm_committed;
138 struct shmid_kernel     *shmsegs;       /* 64 bit version */
139 static int shm_inited = 0;
140 
141 /*
142  * Since anonymous memory chunks are limited to ANON_MAX_SIZE bytes,
143  * we have to keep a list of chunks when we want to handle a shared memory
144  * segment bigger than ANON_MAX_SIZE.
145  * Each chunk points to a VM named entry of up to ANON_MAX_SIZE bytes
146  * of anonymous memory.
147  */
148 struct shm_handle {
149 	void * shm_object;                      /* named entry for this chunk*/
150 	memory_object_size_t shm_handle_size;   /* size of this chunk */
151 	struct shm_handle *shm_handle_next;     /* next chunk */
152 };
153 
154 struct shmmap_state {
155 	mach_vm_address_t va;           /* user address */
156 	int shmid;                      /* segment id */
157 };
158 
159 static void shm_deallocate_segment(struct shmid_kernel *);
160 static int shm_find_segment_by_key(key_t);
161 static struct shmid_kernel *shm_find_segment_by_shmid(int);
162 static int shm_delete_mapping(struct proc *, struct shmmap_state *, int);
163 
164 #ifdef __APPLE_API_PRIVATE
165 #define DEFAULT_SHMMAX  (4 * 1024 * 1024)
166 #define DEFAULT_SHMMIN  1
167 #define DEFAULT_SHMMNI  32
168 #define DEFAULT_SHMSEG  8
169 #define DEFAULT_SHMALL  1024
170 
171 struct shminfo shminfo = {
172 	.shmmax = DEFAULT_SHMMAX,
173 	.shmmin = DEFAULT_SHMMIN,
174 	.shmmni = DEFAULT_SHMMNI,
175 	.shmseg = DEFAULT_SHMSEG,
176 	.shmall = DEFAULT_SHMALL
177 };
178 
179 #define SHMID_IS_VALID(x) ((x) >= 0)
180 #define SHMID_UNALLOCATED (-1)
181 #define SHMID_SENTINEL    (-2)
182 
183 #endif /* __APPLE_API_PRIVATE */
184 
185 static __inline__ time_t
sysv_shmtime(void)186 sysv_shmtime(void)
187 {
188 	struct timeval  tv;
189 	microtime(&tv);
190 	return tv.tv_sec;
191 }
192 
193 /*
194  * This conversion is safe, since if we are converting for a 32 bit process,
195  * then it's value of (struct shmid_ds)->shm_segsz will never exceed 4G.
196  *
197  * NOTE: Source and target may *NOT* overlap! (target is smaller)
198  */
199 static void
shmid_ds_64to32(struct user_shmid_ds * in,struct user32_shmid_ds * out)200 shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out)
201 {
202 	out->shm_perm = in->shm_perm;
203 	out->shm_segsz = in->shm_segsz;
204 	out->shm_lpid = in->shm_lpid;
205 	out->shm_cpid = in->shm_cpid;
206 	out->shm_nattch = in->shm_nattch;
207 	out->shm_atime = in->shm_atime;
208 	out->shm_dtime = in->shm_dtime;
209 	out->shm_ctime = in->shm_ctime;
210 	out->shm_internal = CAST_DOWN_EXPLICIT(int, in->shm_internal);
211 }
212 
213 /*
214  * NOTE: Source and target may are permitted to overlap! (source is smaller);
215  * this works because we copy fields in order from the end of the struct to
216  * the beginning.
217  */
218 static void
shmid_ds_32to64(struct user32_shmid_ds * in,struct user_shmid_ds * out)219 shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out)
220 {
221 	out->shm_internal = in->shm_internal;
222 	out->shm_ctime = in->shm_ctime;
223 	out->shm_dtime = in->shm_dtime;
224 	out->shm_atime = in->shm_atime;
225 	out->shm_nattch = in->shm_nattch;
226 	out->shm_cpid = in->shm_cpid;
227 	out->shm_lpid = in->shm_lpid;
228 	out->shm_segsz = in->shm_segsz;
229 	out->shm_perm = in->shm_perm;
230 }
231 
232 
233 static int
shm_find_segment_by_key(key_t key)234 shm_find_segment_by_key(key_t key)
235 {
236 	int i;
237 
238 	for (i = 0; i < shminfo.shmmni; i++) {
239 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
240 		    shmsegs[i].u.shm_perm._key == key) {
241 			return i;
242 		}
243 	}
244 	return -1;
245 }
246 
247 static struct shmid_kernel *
shm_find_segment_by_shmid(int shmid)248 shm_find_segment_by_shmid(int shmid)
249 {
250 	int segnum;
251 	struct shmid_kernel *shmseg;
252 
253 	segnum = IPCID_TO_IX(shmid);
254 	if (segnum < 0 || segnum >= shminfo.shmmni) {
255 		return NULL;
256 	}
257 	shmseg = &shmsegs[segnum];
258 	if ((shmseg->u.shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
259 	    != SHMSEG_ALLOCATED ||
260 	    shmseg->u.shm_perm._seq != IPCID_TO_SEQ(shmid)) {
261 		return NULL;
262 	}
263 	return shmseg;
264 }
265 
266 static void
shm_deallocate_segment(struct shmid_kernel * shmseg)267 shm_deallocate_segment(struct shmid_kernel *shmseg)
268 {
269 	struct shm_handle *shm_handle, *shm_handle_next;
270 	mach_vm_size_t size;
271 
272 	for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal); /* tunnel */
273 	    shm_handle != NULL;
274 	    shm_handle = shm_handle_next) {
275 		shm_handle_next = shm_handle->shm_handle_next;
276 		mach_memory_entry_port_release(shm_handle->shm_object);
277 		kfree_type(struct shm_handle, shm_handle);
278 	}
279 	shmseg->u.shm_internal = USER_ADDR_NULL;                /* tunnel */
280 	size = vm_map_round_page(shmseg->u.shm_segsz,
281 	    vm_map_page_mask(current_map()));
282 	shm_committed -= btoc(size);
283 	shm_nused--;
284 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
285 #if CONFIG_MACF
286 	/* Reset the MAC label */
287 	mac_sysvshm_label_recycle(shmseg);
288 #endif
289 }
290 
291 static int
shm_delete_mapping(__unused struct proc * p,struct shmmap_state * shmmap_s,int deallocate)292 shm_delete_mapping(__unused struct proc *p, struct shmmap_state *shmmap_s,
293     int deallocate)
294 {
295 	struct shmid_kernel *shmseg;
296 	int segnum, result;
297 	mach_vm_size_t size;
298 
299 	segnum = IPCID_TO_IX(shmmap_s->shmid);
300 	shmseg = &shmsegs[segnum];
301 	size = vm_map_round_page(shmseg->u.shm_segsz,
302 	    vm_map_page_mask(current_map())); /* XXX done for us? */
303 	if (deallocate) {
304 		result = mach_vm_deallocate(current_map(), shmmap_s->va, size);
305 		if (result != KERN_SUCCESS) {
306 			return EINVAL;
307 		}
308 	}
309 	shmmap_s->shmid = SHMID_UNALLOCATED;
310 	shmseg->u.shm_dtime = sysv_shmtime();
311 	if ((--shmseg->u.shm_nattch <= 0) &&
312 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
313 		shm_deallocate_segment(shmseg);
314 		shm_last_free = segnum;
315 	}
316 	return 0;
317 }
318 
319 int
shmdt(struct proc * p,struct shmdt_args * uap,int32_t * retval)320 shmdt(struct proc *p, struct shmdt_args *uap, int32_t *retval)
321 {
322 #if CONFIG_MACF
323 	struct shmid_kernel *shmsegptr;
324 #endif
325 	struct shmmap_state *shmmap_s;
326 	int i;
327 	int shmdtret = 0;
328 
329 	AUDIT_ARG(svipc_addr, uap->shmaddr);
330 
331 	SYSV_SHM_SUBSYS_LOCK();
332 
333 	if ((shmdtret = shminit())) {
334 		goto shmdt_out;
335 	}
336 
337 	shmmap_s = (struct shmmap_state *)p->vm_shm;
338 	if (shmmap_s == NULL) {
339 		shmdtret = EINVAL;
340 		goto shmdt_out;
341 	}
342 
343 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
344 		if (SHMID_IS_VALID(shmmap_s->shmid) &&
345 		    shmmap_s->va == (mach_vm_offset_t)uap->shmaddr) {
346 			break;
347 		}
348 	}
349 
350 	if (!SHMID_IS_VALID(shmmap_s->shmid)) {
351 		shmdtret = EINVAL;
352 		goto shmdt_out;
353 	}
354 
355 #if CONFIG_MACF
356 	/*
357 	 * XXX: It might be useful to move this into the shm_delete_mapping
358 	 * function
359 	 */
360 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
361 	shmdtret = mac_sysvshm_check_shmdt(kauth_cred_get(), shmsegptr);
362 	if (shmdtret) {
363 		goto shmdt_out;
364 	}
365 #endif
366 	i = shm_delete_mapping(p, shmmap_s, 1);
367 
368 	if (i == 0) {
369 		*retval = 0;
370 	}
371 	shmdtret = i;
372 shmdt_out:
373 	SYSV_SHM_SUBSYS_UNLOCK();
374 	return shmdtret;
375 }
376 
377 int
shmat(struct proc * p,struct shmat_args * uap,user_addr_t * retval)378 shmat(struct proc *p, struct shmat_args *uap, user_addr_t *retval)
379 {
380 	int error, flags;
381 	struct shmid_kernel     *shmseg;
382 	struct shmmap_state     *shmmap_s = NULL;
383 	struct shm_handle       *shm_handle;
384 	mach_vm_address_t       attach_va;      /* attach address in/out */
385 	mach_vm_address_t       shmlba;
386 	mach_vm_size_t          map_size;       /* size of map entry */
387 	mach_vm_size_t          mapped_size;
388 	vm_prot_t               prot;
389 	kern_return_t           rv;
390 	int                     shmat_ret;
391 	vm_map_kernel_flags_t   vmk_flags;
392 
393 	shmat_ret = 0;
394 
395 	AUDIT_ARG(svipc_id, uap->shmid);
396 	AUDIT_ARG(svipc_addr, uap->shmaddr);
397 
398 	SYSV_SHM_SUBSYS_LOCK();
399 
400 	if ((shmat_ret = shminit())) {
401 		goto shmat_out;
402 	}
403 
404 	shmmap_s = (struct shmmap_state *)p->vm_shm;
405 	if (shmmap_s == NULL) {
406 		/* lazily allocate the shm map */
407 
408 		int nsegs = shminfo.shmseg;
409 		if (nsegs <= 0) {
410 			shmat_ret = EMFILE;
411 			goto shmat_out;
412 		}
413 
414 		/* +1 for the sentinel */
415 		shmmap_s = kalloc_type(struct shmmap_state, nsegs + 1, Z_WAITOK);
416 		if (shmmap_s == NULL) {
417 			shmat_ret = ENOMEM;
418 			goto shmat_out;
419 		}
420 
421 		/* initialize the entries */
422 		for (int i = 0; i < nsegs; i++) {
423 			shmmap_s[i].shmid = SHMID_UNALLOCATED;
424 		}
425 		shmmap_s[nsegs].shmid = SHMID_SENTINEL;
426 
427 		p->vm_shm = (caddr_t)shmmap_s;
428 	}
429 
430 	shmseg = shm_find_segment_by_shmid(uap->shmid);
431 	if (shmseg == NULL) {
432 		shmat_ret = EINVAL;
433 		goto shmat_out;
434 	}
435 
436 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
437 	error = ipcperm(kauth_cred_get(), &shmseg->u.shm_perm,
438 	    (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R | IPC_W);
439 	if (error) {
440 		shmat_ret = error;
441 		goto shmat_out;
442 	}
443 
444 #if CONFIG_MACF
445 	error = mac_sysvshm_check_shmat(kauth_cred_get(), shmseg, uap->shmflg);
446 	if (error) {
447 		shmat_ret = error;
448 		goto shmat_out;
449 	}
450 #endif
451 
452 	/* find a free shmid */
453 	while (SHMID_IS_VALID(shmmap_s->shmid)) {
454 		shmmap_s++;
455 	}
456 	if (shmmap_s->shmid != SHMID_UNALLOCATED) {
457 		/* no free shmids */
458 		shmat_ret = EMFILE;
459 		goto shmat_out;
460 	}
461 
462 	map_size = vm_map_round_page(shmseg->u.shm_segsz,
463 	    vm_map_page_mask(current_map()));
464 	prot = VM_PROT_READ;
465 	if ((uap->shmflg & SHM_RDONLY) == 0) {
466 		prot |= VM_PROT_WRITE;
467 	}
468 	flags = MAP_ANON | MAP_SHARED;
469 	if (uap->shmaddr) {
470 		flags |= MAP_FIXED;
471 	}
472 
473 	attach_va = (mach_vm_address_t)uap->shmaddr;
474 	shmlba = vm_map_page_size(current_map()); /* XXX instead of SHMLBA */
475 	if (uap->shmflg & SHM_RND) {
476 		attach_va &= ~(shmlba - 1);
477 	} else if ((attach_va & (shmlba - 1)) != 0) {
478 		shmat_ret = EINVAL;
479 		goto shmat_out;
480 	}
481 
482 	if (flags & MAP_FIXED) {
483 		vmk_flags = VM_MAP_KERNEL_FLAGS_FIXED();
484 	} else {
485 		vmk_flags = VM_MAP_KERNEL_FLAGS_ANYWHERE();
486 	}
487 
488 	mapped_size = 0;
489 
490 	/* first reserve enough space... */
491 	rv = mach_vm_map_kernel(current_map(),
492 	    &attach_va,
493 	    map_size,
494 	    0,
495 	    vmk_flags,
496 	    IPC_PORT_NULL,
497 	    0,
498 	    FALSE,
499 	    VM_PROT_NONE,
500 	    VM_PROT_NONE,
501 	    VM_INHERIT_NONE);
502 	if (rv != KERN_SUCCESS) {
503 		goto out;
504 	}
505 
506 	shmmap_s->va = attach_va;
507 
508 	/* ... then map the shared memory over the reserved space */
509 	for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal);/* tunnel */
510 	    shm_handle != NULL;
511 	    shm_handle = shm_handle->shm_handle_next) {
512 		vm_map_size_t chunk_size;
513 
514 		assert(mapped_size < map_size);
515 		chunk_size = shm_handle->shm_handle_size;
516 		if (chunk_size > map_size - mapped_size) {
517 			/*
518 			 * Partial mapping of last chunk due to
519 			 * page size mismatch.
520 			 */
521 			assert(vm_map_page_shift(current_map()) < PAGE_SHIFT);
522 			assert(shm_handle->shm_handle_next == NULL);
523 			chunk_size = map_size - mapped_size;
524 		}
525 		rv = mach_vm_map_kernel(
526 			current_map(),          /* process map */
527 			&attach_va,             /* attach address */
528 			chunk_size,             /* size to map */
529 			(mach_vm_offset_t)0,    /* alignment mask */
530 			VM_MAP_KERNEL_FLAGS_FIXED(.vmf_overwrite = true),
531 			shm_handle->shm_object,
532 			(mach_vm_offset_t)0,
533 			FALSE,
534 			prot,
535 			prot,
536 			VM_INHERIT_SHARE);
537 		if (rv != KERN_SUCCESS) {
538 			goto out;
539 		}
540 
541 		mapped_size += chunk_size;
542 		attach_va = attach_va + chunk_size;
543 	}
544 
545 	shmmap_s->shmid = uap->shmid;
546 	shmseg->u.shm_lpid = proc_getpid(p);
547 	shmseg->u.shm_atime = sysv_shmtime();
548 	shmseg->u.shm_nattch++;
549 	*retval = shmmap_s->va; /* XXX return -1 on error */
550 	shmat_ret = 0;
551 	goto shmat_out;
552 out:
553 	if (mapped_size > 0) {
554 		(void) mach_vm_deallocate(current_map(),
555 		    shmmap_s->va,
556 		    mapped_size);
557 	}
558 	switch (rv) {
559 	case KERN_INVALID_ADDRESS:
560 	case KERN_NO_SPACE:
561 		shmat_ret = ENOMEM;
562 		break;
563 	case KERN_PROTECTION_FAILURE:
564 		shmat_ret = EACCES;
565 		break;
566 	default:
567 		shmat_ret = EINVAL;
568 		break;
569 	}
570 shmat_out:
571 	SYSV_SHM_SUBSYS_UNLOCK();
572 	return shmat_ret;
573 }
574 
575 static int
oshmctl(__unused void * p,__unused void * uap,__unused void * retval)576 oshmctl(__unused void *p, __unused void *uap, __unused void *retval)
577 {
578 	return EINVAL;
579 }
580 
581 /*
582  * Returns:	0			Success
583  *		EINVAL
584  *	copyout:EFAULT
585  *	copyin:EFAULT
586  *	ipcperm:EPERM
587  *	ipcperm:EACCES
588  */
589 int
shmctl(__unused struct proc * p,struct shmctl_args * uap,int32_t * retval)590 shmctl(__unused struct proc *p, struct shmctl_args *uap, int32_t *retval)
591 {
592 	int error;
593 	kauth_cred_t cred = kauth_cred_get();
594 	struct user_shmid_ds inbuf;
595 	struct shmid_kernel *shmseg;
596 
597 	int shmctl_ret = 0;
598 
599 	AUDIT_ARG(svipc_cmd, uap->cmd);
600 	AUDIT_ARG(svipc_id, uap->shmid);
601 
602 	SYSV_SHM_SUBSYS_LOCK();
603 
604 	if ((shmctl_ret = shminit())) {
605 		goto shmctl_out;
606 	}
607 
608 	shmseg = shm_find_segment_by_shmid(uap->shmid);
609 	if (shmseg == NULL) {
610 		shmctl_ret = EINVAL;
611 		goto shmctl_out;
612 	}
613 
614 	/* XXAUDIT: This is the perms BEFORE any change by this call. This
615 	 * may not be what is desired.
616 	 */
617 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
618 
619 #if CONFIG_MACF
620 	error = mac_sysvshm_check_shmctl(cred, shmseg, uap->cmd);
621 	if (error) {
622 		shmctl_ret = error;
623 		goto shmctl_out;
624 	}
625 #endif
626 	switch (uap->cmd) {
627 	case IPC_STAT:
628 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_R);
629 		if (error) {
630 			shmctl_ret = error;
631 			goto shmctl_out;
632 		}
633 
634 		if (IS_64BIT_PROCESS(p)) {
635 			struct user_shmid_ds shmid_ds = {};
636 			memcpy(&shmid_ds, &shmseg->u, sizeof(struct user_shmid_ds));
637 
638 			/* Clear kernel reserved pointer before copying to user space */
639 			shmid_ds.shm_internal = USER_ADDR_NULL;
640 
641 			error = copyout(&shmid_ds, uap->buf, sizeof(shmid_ds));
642 		} else {
643 			struct user32_shmid_ds shmid_ds32 = {};
644 			shmid_ds_64to32(&shmseg->u, &shmid_ds32);
645 
646 			/* Clear kernel reserved pointer before copying to user space */
647 			shmid_ds32.shm_internal = (user32_addr_t)0;
648 
649 			error = copyout(&shmid_ds32, uap->buf, sizeof(shmid_ds32));
650 		}
651 		if (error) {
652 			shmctl_ret = error;
653 			goto shmctl_out;
654 		}
655 		break;
656 	case IPC_SET:
657 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_M);
658 		if (error) {
659 			shmctl_ret = error;
660 			goto shmctl_out;
661 		}
662 		if (IS_64BIT_PROCESS(p)) {
663 			error = copyin(uap->buf, &inbuf, sizeof(struct user_shmid_ds));
664 		} else {
665 			struct user32_shmid_ds shmid_ds32;
666 			error = copyin(uap->buf, &shmid_ds32, sizeof(shmid_ds32));
667 			/* convert in place; ugly, but safe */
668 			shmid_ds_32to64(&shmid_ds32, &inbuf);
669 		}
670 		if (error) {
671 			shmctl_ret = error;
672 			goto shmctl_out;
673 		}
674 		shmseg->u.shm_perm.uid = inbuf.shm_perm.uid;
675 		shmseg->u.shm_perm.gid = inbuf.shm_perm.gid;
676 		shmseg->u.shm_perm.mode =
677 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
678 		    (inbuf.shm_perm.mode & ACCESSPERMS);
679 		shmseg->u.shm_ctime = sysv_shmtime();
680 		break;
681 	case IPC_RMID:
682 		error = ipcperm(cred, &shmseg->u.shm_perm, IPC_M);
683 		if (error) {
684 			shmctl_ret = error;
685 			goto shmctl_out;
686 		}
687 		shmseg->u.shm_perm._key = IPC_PRIVATE;
688 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
689 		if (shmseg->u.shm_nattch <= 0) {
690 			shm_deallocate_segment(shmseg);
691 			shm_last_free = IPCID_TO_IX(uap->shmid);
692 		}
693 		break;
694 #if 0
695 	case SHM_LOCK:
696 	case SHM_UNLOCK:
697 #endif
698 	default:
699 		shmctl_ret = EINVAL;
700 		goto shmctl_out;
701 	}
702 	*retval = 0;
703 	shmctl_ret = 0;
704 shmctl_out:
705 	SYSV_SHM_SUBSYS_UNLOCK();
706 	return shmctl_ret;
707 }
708 
709 static int
shmget_existing(struct shmget_args * uap,int mode,int segnum,int * retval)710 shmget_existing(struct shmget_args *uap, int mode, int segnum, int *retval)
711 {
712 	struct shmid_kernel *shmseg;
713 	int error = 0;
714 
715 	shmseg = &shmsegs[segnum];
716 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
717 		/*
718 		 * This segment is in the process of being allocated.  Wait
719 		 * until it's done, and look the key up again (in case the
720 		 * allocation failed or it was freed).
721 		 */
722 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
723 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
724 		if (error) {
725 			return error;
726 		}
727 		return EAGAIN;
728 	}
729 
730 	/*
731 	 * The low 9 bits of shmflag are the mode bits being requested, which
732 	 * are the actual mode bits desired on the segment, and not in IPC_R
733 	 * form; therefore it would be incorrect to call ipcperm() to validate
734 	 * them; instead, we AND the existing mode with the requested mode, and
735 	 * verify that it matches the requested mode; otherwise, we fail with
736 	 * EACCES (access denied).
737 	 */
738 	if ((shmseg->u.shm_perm.mode & mode) != mode) {
739 		return EACCES;
740 	}
741 
742 #if CONFIG_MACF
743 	error = mac_sysvshm_check_shmget(kauth_cred_get(), shmseg, uap->shmflg);
744 	if (error) {
745 		return error;
746 	}
747 #endif
748 
749 	if (uap->size && uap->size > shmseg->u.shm_segsz) {
750 		return EINVAL;
751 	}
752 
753 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) {
754 		return EEXIST;
755 	}
756 
757 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
758 	return 0;
759 }
760 
761 static int
shmget_allocate_segment(struct proc * p,struct shmget_args * uap,int mode,int * retval)762 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode,
763     int *retval)
764 {
765 	int i, segnum, shmid;
766 	kauth_cred_t cred = kauth_cred_get();
767 	struct shmid_kernel *shmseg;
768 	struct shm_handle *shm_handle;
769 	kern_return_t kret;
770 	mach_vm_size_t total_size, size = 0, alloc_size;
771 	void * mem_object;
772 	struct shm_handle *shm_handle_next, **shm_handle_next_p;
773 
774 	if (uap->size <= 0 ||
775 	    uap->size < (user_size_t)shminfo.shmmin ||
776 	    uap->size > (user_size_t)shminfo.shmmax) {
777 		return EINVAL;
778 	}
779 	if (shm_nused >= shminfo.shmmni) { /* any shmids left? */
780 		return ENOSPC;
781 	}
782 	if (mach_vm_round_page_overflow(uap->size, &total_size)) {
783 		return EINVAL;
784 	}
785 	if ((user_ssize_t)(shm_committed + btoc(total_size)) > shminfo.shmall) {
786 		return ENOMEM;
787 	}
788 	if (shm_last_free < 0) {
789 		for (i = 0; i < shminfo.shmmni; i++) {
790 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE) {
791 				break;
792 			}
793 		}
794 		if (i == shminfo.shmmni) {
795 			panic("shmseg free count inconsistent");
796 		}
797 		segnum = i;
798 	} else {
799 		segnum = shm_last_free;
800 		shm_last_free = -1;
801 	}
802 	shmseg = &shmsegs[segnum];
803 
804 	/*
805 	 * In case we sleep in malloc(), mark the segment present but deleted
806 	 * so that noone else tries to create the same key.
807 	 * XXX but we don't release the global lock !?
808 	 */
809 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
810 	shmseg->u.shm_perm._key = uap->key;
811 	shmseg->u.shm_perm._seq = (shmseg->u.shm_perm._seq + 1) & 0x7fff;
812 
813 	shm_handle_next_p = NULL;
814 	for (alloc_size = 0;
815 	    alloc_size < total_size;
816 	    alloc_size += size) {
817 		size = MIN(total_size - alloc_size, ANON_MAX_SIZE);
818 		kret = mach_make_memory_entry_64(
819 			VM_MAP_NULL,
820 			(memory_object_size_t *) &size,
821 			(memory_object_offset_t) 0,
822 			MAP_MEM_NAMED_CREATE | VM_PROT_DEFAULT,
823 			(ipc_port_t *) &mem_object, 0);
824 		if (kret != KERN_SUCCESS) {
825 			goto out;
826 		}
827 
828 		shm_handle = kalloc_type(struct shm_handle, Z_WAITOK | Z_NOFAIL);
829 		shm_handle->shm_object = mem_object;
830 		shm_handle->shm_handle_size = size;
831 		shm_handle->shm_handle_next = NULL;
832 		if (shm_handle_next_p == NULL) {
833 			shmseg->u.shm_internal = CAST_USER_ADDR_T(shm_handle);/* tunnel */
834 		} else {
835 			*shm_handle_next_p = shm_handle;
836 		}
837 		shm_handle_next_p = &shm_handle->shm_handle_next;
838 	}
839 
840 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
841 
842 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = kauth_cred_getuid(cred);
843 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = kauth_cred_getgid(cred);
844 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
845 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
846 	shmseg->u.shm_segsz = uap->size;
847 	shmseg->u.shm_cpid = proc_getpid(p);
848 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
849 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
850 #if CONFIG_MACF
851 	mac_sysvshm_label_associate(cred, shmseg);
852 #endif
853 	shmseg->u.shm_ctime = sysv_shmtime();
854 	shm_committed += btoc(size);
855 	shm_nused++;
856 	AUDIT_ARG(svipc_perm, &shmseg->u.shm_perm);
857 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
858 		/*
859 		 * Somebody else wanted this key while we were asleep.  Wake
860 		 * them up now.
861 		 */
862 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
863 		wakeup((caddr_t)shmseg);
864 	}
865 	*retval = shmid;
866 	AUDIT_ARG(svipc_id, shmid);
867 	return 0;
868 out:
869 	if (kret != KERN_SUCCESS) {
870 		for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal); /* tunnel */
871 		    shm_handle != NULL;
872 		    shm_handle = shm_handle_next) {
873 			shm_handle_next = shm_handle->shm_handle_next;
874 			mach_memory_entry_port_release(shm_handle->shm_object);
875 			kfree_type(struct shm_handle, shm_handle);
876 		}
877 		shmseg->u.shm_internal = USER_ADDR_NULL; /* tunnel */
878 	}
879 
880 	switch (kret) {
881 	case KERN_INVALID_ADDRESS:
882 	case KERN_NO_SPACE:
883 		return ENOMEM;
884 	case KERN_PROTECTION_FAILURE:
885 		return EACCES;
886 	default:
887 		return EINVAL;
888 	}
889 }
890 
891 int
shmget(struct proc * p,struct shmget_args * uap,int32_t * retval)892 shmget(struct proc *p, struct shmget_args *uap, int32_t *retval)
893 {
894 	int segnum, mode, error;
895 	int shmget_ret = 0;
896 
897 	/* Auditing is actually done in shmget_allocate_segment() */
898 
899 	SYSV_SHM_SUBSYS_LOCK();
900 
901 	if ((shmget_ret = shminit())) {
902 		goto shmget_out;
903 	}
904 
905 	mode = uap->shmflg & ACCESSPERMS;
906 	if (uap->key != IPC_PRIVATE) {
907 again:
908 		segnum = shm_find_segment_by_key(uap->key);
909 		if (segnum >= 0) {
910 			error = shmget_existing(uap, mode, segnum, retval);
911 			if (error == EAGAIN) {
912 				goto again;
913 			}
914 			shmget_ret = error;
915 			goto shmget_out;
916 		}
917 		if ((uap->shmflg & IPC_CREAT) == 0) {
918 			shmget_ret = ENOENT;
919 			goto shmget_out;
920 		}
921 	}
922 	shmget_ret = shmget_allocate_segment(p, uap, mode, retval);
923 shmget_out:
924 	SYSV_SHM_SUBSYS_UNLOCK();
925 	return shmget_ret;
926 }
927 
928 /*
929  * shmsys
930  *
931  * Entry point for all SHM calls: shmat, oshmctl, shmdt, shmget, shmctl
932  *
933  * Parameters:	p	Process requesting the call
934  *              uap	User argument descriptor (see below)
935  *              retval	Return value of the selected shm call
936  *
937  * Indirect parameters:	uap->which	msg call to invoke (index in array of shm calls)
938  *                      uap->a2		User argument descriptor
939  *
940  * Returns:	0	Success
941  *              !0	Not success
942  *
943  * Implicit returns: retval     Return value of the selected shm call
944  *
945  * DEPRECATED:  This interface should not be used to call the other SHM
946  *              functions (shmat, oshmctl, shmdt, shmget, shmctl). The correct
947  *              usage is to call the other SHM functions directly.
948  */
949 int
shmsys(struct proc * p,struct shmsys_args * uap,int32_t * retval)950 shmsys(struct proc *p, struct shmsys_args *uap, int32_t *retval)
951 {
952 	/* The routine that we are dispatching already does this */
953 
954 	if (uap->which >= sizeof(shmcalls) / sizeof(shmcalls[0])) {
955 		return EINVAL;
956 	}
957 	return (*shmcalls[uap->which])(p, &uap->a2, retval);
958 }
959 
960 /*
961  * Return 0 on success, 1 on failure.
962  */
963 int
shmfork(struct proc * p1,struct proc * p2)964 shmfork(struct proc *p1, struct proc *p2)
965 {
966 	struct shmmap_state *shmmap_s;
967 	int nsegs = 0;
968 	int ret = 0;
969 
970 	SYSV_SHM_SUBSYS_LOCK();
971 
972 	if (shminit()) {
973 		ret = 1;
974 		goto shmfork_out;
975 	}
976 
977 	struct shmmap_state *src = (struct shmmap_state *)p1->vm_shm;
978 	assert(src);
979 
980 	/* count number of shmid entries in src */
981 	for (struct shmmap_state *s = src; s->shmid != SHMID_SENTINEL; s++) {
982 		nsegs++;
983 	}
984 
985 	shmmap_s = kalloc_type(struct shmmap_state, nsegs + 1, Z_WAITOK);
986 	if (shmmap_s == NULL) {
987 		ret = 1;
988 		goto shmfork_out;
989 	}
990 
991 	bcopy(src, (caddr_t)shmmap_s, (nsegs + 1) * sizeof(struct shmmap_state));
992 	p2->vm_shm = (caddr_t)shmmap_s;
993 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
994 		if (SHMID_IS_VALID(shmmap_s->shmid)) {
995 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
996 		}
997 	}
998 
999 shmfork_out:
1000 	SYSV_SHM_SUBSYS_UNLOCK();
1001 	return ret;
1002 }
1003 
1004 static void
shmcleanup(struct proc * p,int deallocate)1005 shmcleanup(struct proc *p, int deallocate)
1006 {
1007 	struct shmmap_state *shmmap_s;
1008 	int nsegs = 0;
1009 
1010 	SYSV_SHM_SUBSYS_LOCK();
1011 
1012 	shmmap_s = (struct shmmap_state *)p->vm_shm;
1013 	for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) {
1014 		nsegs++;
1015 		if (SHMID_IS_VALID(shmmap_s->shmid)) {
1016 			/*
1017 			 * XXX: Should the MAC framework enforce
1018 			 * check here as well.
1019 			 */
1020 			shm_delete_mapping(p, shmmap_s, deallocate);
1021 		}
1022 	}
1023 
1024 	kfree_type(struct shmmap_state, nsegs + 1, p->vm_shm);
1025 	SYSV_SHM_SUBSYS_UNLOCK();
1026 }
1027 
1028 void
shmexit(struct proc * p)1029 shmexit(struct proc *p)
1030 {
1031 	shmcleanup(p, 1);
1032 }
1033 
1034 /*
1035  * shmexec() is like shmexit(), only it doesn't delete the mappings,
1036  * since the old address space has already been destroyed and the new
1037  * one instantiated.  Instead, it just does the housekeeping work we
1038  * need to do to keep the System V shared memory subsystem sane.
1039  */
1040 __private_extern__ void
shmexec(struct proc * p)1041 shmexec(struct proc *p)
1042 {
1043 	shmcleanup(p, 0);
1044 }
1045 
1046 int
shminit(void)1047 shminit(void)
1048 {
1049 	size_t sz;
1050 	int i;
1051 
1052 	if (!shm_inited) {
1053 		/*
1054 		 * we store internally 64 bit, since if we didn't, we would
1055 		 * be unable to represent a segment size in excess of 32 bits
1056 		 * with the (struct shmid_ds)->shm_segsz field; also, POSIX
1057 		 * dictates this filed be a size_t, which is 64 bits when
1058 		 * running 64 bit binaries.
1059 		 */
1060 		if (os_mul_overflow(shminfo.shmmni, sizeof(struct shmid_kernel), &sz)) {
1061 			return ENOMEM;
1062 		}
1063 
1064 		shmsegs = zalloc_permanent(sz, ZALIGN_PTR);
1065 		if (shmsegs == NULL) {
1066 			return ENOMEM;
1067 		}
1068 		for (i = 0; i < shminfo.shmmni; i++) {
1069 			shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
1070 			shmsegs[i].u.shm_perm._seq = 0;
1071 #if CONFIG_MACF
1072 			mac_sysvshm_label_init(&shmsegs[i]);
1073 #endif
1074 		}
1075 		shm_last_free = 0;
1076 		shm_nused = 0;
1077 		shm_committed = 0;
1078 		shm_inited = 1;
1079 	}
1080 
1081 	return 0;
1082 }
1083 
1084 /* (struct sysctl_oid *oidp, void *arg1, int arg2, \
1085  *       struct sysctl_req *req) */
1086 static int
sysctl_shminfo(__unused struct sysctl_oid * oidp,void * arg1,__unused int arg2,struct sysctl_req * req)1087 sysctl_shminfo(__unused struct sysctl_oid *oidp, void *arg1,
1088     __unused int arg2, struct sysctl_req *req)
1089 {
1090 	int error = 0;
1091 	int sysctl_shminfo_ret = 0;
1092 	int64_t saved_shmmax;
1093 	int64_t saved_shmmin;
1094 	int64_t saved_shmseg;
1095 	int64_t saved_shmmni;
1096 	int64_t saved_shmall;
1097 
1098 	error = SYSCTL_OUT(req, arg1, sizeof(int64_t));
1099 	if (error || req->newptr == USER_ADDR_NULL) {
1100 		return error;
1101 	}
1102 
1103 	SYSV_SHM_SUBSYS_LOCK();
1104 
1105 	/* shmmni can not be changed after SysV SHM has been initialized */
1106 	if (shm_inited && arg1 == &shminfo.shmmni) {
1107 		sysctl_shminfo_ret = EPERM;
1108 		goto sysctl_shminfo_out;
1109 	}
1110 	saved_shmmax = shminfo.shmmax;
1111 	saved_shmmin = shminfo.shmmin;
1112 	saved_shmseg = shminfo.shmseg;
1113 	saved_shmmni = shminfo.shmmni;
1114 	saved_shmall = shminfo.shmall;
1115 
1116 	if ((error = SYSCTL_IN(req, arg1, sizeof(int64_t))) != 0) {
1117 		sysctl_shminfo_ret = error;
1118 		goto sysctl_shminfo_out;
1119 	}
1120 
1121 	if (arg1 == &shminfo.shmmax) {
1122 		/* shmmax needs to be page-aligned */
1123 		if (shminfo.shmmax & PAGE_MASK_64 || shminfo.shmmax < 0) {
1124 			shminfo.shmmax = saved_shmmax;
1125 			sysctl_shminfo_ret = EINVAL;
1126 			goto sysctl_shminfo_out;
1127 		}
1128 	} else if (arg1 == &shminfo.shmmin) {
1129 		if (shminfo.shmmin < 0) {
1130 			shminfo.shmmin = saved_shmmin;
1131 			sysctl_shminfo_ret = EINVAL;
1132 			goto sysctl_shminfo_out;
1133 		}
1134 	} else if (arg1 == &shminfo.shmseg) {
1135 		/* add a sanity check - 20847256 */
1136 		if (shminfo.shmseg > INT32_MAX || shminfo.shmseg < 0) {
1137 			shminfo.shmseg = saved_shmseg;
1138 			sysctl_shminfo_ret = EINVAL;
1139 			goto sysctl_shminfo_out;
1140 		}
1141 	} else if (arg1 == &shminfo.shmmni) {
1142 		/* add a sanity check - 20847256 */
1143 		if (shminfo.shmmni > INT32_MAX || shminfo.shmmni < 0) {
1144 			shminfo.shmmni = saved_shmmni;
1145 			sysctl_shminfo_ret = EINVAL;
1146 			goto sysctl_shminfo_out;
1147 		}
1148 	} else if (arg1 == &shminfo.shmall) {
1149 		/* add a sanity check - 20847256 */
1150 		if (shminfo.shmall > INT32_MAX || shminfo.shmall < 0) {
1151 			shminfo.shmall = saved_shmall;
1152 			sysctl_shminfo_ret = EINVAL;
1153 			goto sysctl_shminfo_out;
1154 		}
1155 	}
1156 	sysctl_shminfo_ret = 0;
1157 sysctl_shminfo_out:
1158 	SYSV_SHM_SUBSYS_UNLOCK();
1159 	return sysctl_shminfo_ret;
1160 }
1161 
1162 static int
IPCS_shm_sysctl(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)1163 IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1,
1164     __unused int arg2, struct sysctl_req *req)
1165 {
1166 	int error;
1167 	int cursor;
1168 	union {
1169 		struct user32_IPCS_command u32;
1170 		struct user_IPCS_command u64;
1171 	} ipcs = { };
1172 	struct user32_shmid_ds shmid_ds32 = { }; /* post conversion, 32 bit version */
1173 	struct user_shmid_ds   shmid_ds = { };   /* 64 bit version */
1174 	void *shmid_dsp;
1175 	size_t ipcs_sz = sizeof(struct user_IPCS_command);
1176 	size_t shmid_ds_sz = sizeof(struct user_shmid_ds);
1177 	struct proc *p = current_proc();
1178 
1179 	SYSV_SHM_SUBSYS_LOCK();
1180 
1181 	if ((error = shminit())) {
1182 		goto ipcs_shm_sysctl_out;
1183 	}
1184 
1185 	if (!IS_64BIT_PROCESS(p)) {
1186 		ipcs_sz = sizeof(struct user32_IPCS_command);
1187 		shmid_ds_sz = sizeof(struct user32_shmid_ds);
1188 	}
1189 
1190 	/* Copy in the command structure */
1191 	if ((error = SYSCTL_IN(req, &ipcs, ipcs_sz)) != 0) {
1192 		goto ipcs_shm_sysctl_out;
1193 	}
1194 
1195 	if (!IS_64BIT_PROCESS(p)) {     /* convert in place */
1196 		ipcs.u64.ipcs_data = CAST_USER_ADDR_T(ipcs.u32.ipcs_data);
1197 	}
1198 
1199 	/* Let us version this interface... */
1200 	if (ipcs.u64.ipcs_magic != IPCS_MAGIC) {
1201 		error = EINVAL;
1202 		goto ipcs_shm_sysctl_out;
1203 	}
1204 
1205 	switch (ipcs.u64.ipcs_op) {
1206 	case IPCS_SHM_CONF:     /* Obtain global configuration data */
1207 		if (ipcs.u64.ipcs_datalen != sizeof(struct shminfo)) {
1208 			if (ipcs.u64.ipcs_cursor != 0) { /* fwd. compat. */
1209 				error = ENOMEM;
1210 				break;
1211 			}
1212 			error = ERANGE;
1213 			break;
1214 		}
1215 		error = copyout(&shminfo, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
1216 		break;
1217 
1218 	case IPCS_SHM_ITER:     /* Iterate over existing segments */
1219 		cursor = ipcs.u64.ipcs_cursor;
1220 		if (cursor < 0 || cursor >= shminfo.shmmni) {
1221 			error = ERANGE;
1222 			break;
1223 		}
1224 		if (ipcs.u64.ipcs_datalen != (int)shmid_ds_sz) {
1225 			error = EINVAL;
1226 			break;
1227 		}
1228 		for (; cursor < shminfo.shmmni; cursor++) {
1229 			if (shmsegs[cursor].u.shm_perm.mode & SHMSEG_ALLOCATED) {
1230 				break;
1231 			}
1232 			continue;
1233 		}
1234 		if (cursor == shminfo.shmmni) {
1235 			error = ENOENT;
1236 			break;
1237 		}
1238 
1239 		shmid_dsp = &shmsegs[cursor];   /* default: 64 bit */
1240 
1241 		/*
1242 		 * If necessary, convert the 64 bit kernel segment
1243 		 * descriptor to a 32 bit user one.
1244 		 */
1245 		if (!IS_64BIT_PROCESS(p)) {
1246 			shmid_ds_64to32(shmid_dsp, &shmid_ds32);
1247 
1248 			/* Clear kernel reserved pointer before copying to user space */
1249 			shmid_ds32.shm_internal = (user32_addr_t)0;
1250 
1251 			shmid_dsp = &shmid_ds32;
1252 		} else {
1253 			memcpy(&shmid_ds, shmid_dsp, sizeof(shmid_ds));
1254 
1255 			/* Clear kernel reserved pointer before copying to user space */
1256 			shmid_ds.shm_internal = USER_ADDR_NULL;
1257 
1258 			shmid_dsp = &shmid_ds;
1259 		}
1260 		error = copyout(shmid_dsp, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen);
1261 		if (!error) {
1262 			/* update cursor */
1263 			ipcs.u64.ipcs_cursor = cursor + 1;
1264 
1265 			if (!IS_64BIT_PROCESS(p)) { /* convert in place */
1266 				ipcs.u32.ipcs_data = CAST_DOWN_EXPLICIT(user32_addr_t, ipcs.u64.ipcs_data);
1267 			}
1268 
1269 			error = SYSCTL_OUT(req, &ipcs, ipcs_sz);
1270 		}
1271 		break;
1272 
1273 	default:
1274 		error = EINVAL;
1275 		break;
1276 	}
1277 ipcs_shm_sysctl_out:
1278 	SYSV_SHM_SUBSYS_UNLOCK();
1279 	return error;
1280 }
1281 
1282 SYSCTL_NODE(_kern, KERN_SYSV, sysv, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSV");
1283 
1284 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmax, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1285     &shminfo.shmmax, 0, &sysctl_shminfo, "Q", "shmmax");
1286 
1287 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmin, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1288     &shminfo.shmmin, 0, &sysctl_shminfo, "Q", "shmmin");
1289 
1290 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmni, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1291     &shminfo.shmmni, 0, &sysctl_shminfo, "Q", "shmmni");
1292 
1293 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmseg, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1294     &shminfo.shmseg, 0, &sysctl_shminfo, "Q", "shmseg");
1295 
1296 SYSCTL_PROC(_kern_sysv, OID_AUTO, shmall, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
1297     &shminfo.shmall, 0, &sysctl_shminfo, "Q", "shmall");
1298 
1299 SYSCTL_NODE(_kern_sysv, OID_AUTO, ipcs, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSVIPCS");
1300 
1301 SYSCTL_PROC(_kern_sysv_ipcs, OID_AUTO, shm, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED,
1302     0, 0, IPCS_shm_sysctl,
1303     "S,IPCS_shm_command",
1304     "ipcs shm command interface");
1305 #endif /* SYSV_SHM */
1306 
1307 /* DSEP Review Done pl-20051108-v02 @2743,@2908,@2913,@3009 */
1308