1 /* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */
2 /*-
3 * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by Adam Glass and Charles
16 * Hannum.
17 * 4. The names of the authors may not be used to endorse or promote products
18 * derived from this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 */
31 /*-
32 * Copyright (c) 2003-2005 McAfee, Inc.
33 * All rights reserved.
34 *
35 * This software was developed for the FreeBSD Project in part by McAfee
36 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
37 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
38 * program.
39 *
40 * Redistribution and use in source and binary forms, with or without
41 * modification, are permitted provided that the following conditions
42 * are met:
43 * 1. Redistributions of source code must retain the above copyright
44 * notice, this list of conditions and the following disclaimer.
45 * 2. Redistributions in binary form must reproduce the above copyright
46 * notice, this list of conditions and the following disclaimer in the
47 * documentation and/or other materials provided with the distribution.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * SUCH DAMAGE.
60 */
61
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64
65 #include "opt_compat.h"
66 #include "opt_sysvipc.h"
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/abi_compat.h>
71 #include <sys/kernel.h>
72 #include <sys/limits.h>
73 #include <sys/lock.h>
74 #include <sys/sysctl.h>
75 #include <sys/shm.h>
76 #include <sys/proc.h>
77 #include <sys/malloc.h>
78 #include <sys/mman.h>
79 #include <sys/module.h>
80 #include <sys/mutex.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
84 #include <sys/stat.h>
85 #include <sys/syscall.h>
86 #include <sys/syscallsubr.h>
87 #include <sys/sysent.h>
88 #include <sys/sysproto.h>
89 #include <sys/jail.h>
90
91 #include <security/mac/mac_framework.h>
92
93 #include <vm/vm.h>
94 #include <vm/vm_param.h>
95 #include <vm/pmap.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_map.h>
98 #include <vm/vm_page.h>
99 #include <vm/vm_pager.h>
100
101 FEATURE(sysv_shm, "System V shared memory segments support");
102
103 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
104
105 #define SHMSEG_FREE 0x0200
106 #define SHMSEG_REMOVED 0x0400
107 #define SHMSEG_ALLOCATED 0x0800
108
109 static int shm_last_free, shm_nused, shmalloced;
110 vm_size_t shm_committed;
111 static struct shmid_kernel *shmsegs;
112 static unsigned shm_prison_slot;
113
114 struct shmmap_state {
115 vm_offset_t va;
116 int shmid;
117 };
118
119 static void shm_deallocate_segment(struct shmid_kernel *);
120 static int shm_find_segment_by_key(struct prison *, key_t);
121 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
122 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
123 static int shmget_allocate_segment(struct thread *td, key_t key, size_t size,
124 int mode);
125 static int shmget_existing(struct thread *td, size_t size, int shmflg,
126 int mode, int segnum);
127 static void shmrealloc(void);
128 static int shminit(void);
129 static int sysvshm_modload(struct module *, int, void *);
130 static int shmunload(void);
131 static void shmexit_myhook(struct vmspace *vm);
132 static void shmfork_myhook(struct proc *p1, struct proc *p2);
133 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
134 static void shm_remove(struct shmid_kernel *, int);
135 static struct prison *shm_find_prison(struct ucred *);
136 static int shm_prison_cansee(struct prison *, struct shmid_kernel *);
137 static int shm_prison_check(void *, void *);
138 static int shm_prison_set(void *, void *);
139 static int shm_prison_get(void *, void *);
140 static int shm_prison_remove(void *, void *);
141 static void shm_prison_cleanup(struct prison *);
142
143 /*
144 * Tuneable values.
145 */
146 #ifndef SHMMAXPGS
147 #define SHMMAXPGS 131072 /* Note: sysv shared memory is swap backed. */
148 #endif
149 #ifndef SHMMAX
150 #define SHMMAX (SHMMAXPGS*PAGE_SIZE)
151 #endif
152 #ifndef SHMMIN
153 #define SHMMIN 1
154 #endif
155 #ifndef SHMMNI
156 #define SHMMNI 192
157 #endif
158 #ifndef SHMSEG
159 #define SHMSEG 128
160 #endif
161 #ifndef SHMALL
162 #define SHMALL (SHMMAXPGS)
163 #endif
164
165 struct shminfo shminfo = {
166 .shmmax = SHMMAX,
167 .shmmin = SHMMIN,
168 .shmmni = SHMMNI,
169 .shmseg = SHMSEG,
170 .shmall = SHMALL
171 };
172
173 static int shm_use_phys;
174 static int shm_allow_removed = 1;
175
176 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
177 "Maximum shared memory segment size");
178 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
179 "Minimum shared memory segment size");
180 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
181 "Number of shared memory identifiers");
182 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
183 "Number of segments per process");
184 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
185 "Maximum number of pages available for shared memory");
186 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
187 &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
188 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
189 &shm_allow_removed, 0,
190 "Enable/Disable attachment to attached segments marked for removal");
191 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
192 CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
193 "Array of struct shmid_kernel for each potential shared memory segment");
194
195 static struct sx sysvshmsx;
196 #define SYSVSHM_LOCK() sx_xlock(&sysvshmsx)
197 #define SYSVSHM_UNLOCK() sx_xunlock(&sysvshmsx)
198 #define SYSVSHM_ASSERT_LOCKED() sx_assert(&sysvshmsx, SA_XLOCKED)
199
200 static int
shm_find_segment_by_key(struct prison * pr,key_t key)201 shm_find_segment_by_key(struct prison *pr, key_t key)
202 {
203 int i;
204
205 for (i = 0; i < shmalloced; i++)
206 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
207 shmsegs[i].cred != NULL &&
208 shmsegs[i].cred->cr_prison == pr &&
209 shmsegs[i].u.shm_perm.key == key)
210 return (i);
211 return (-1);
212 }
213
214 /*
215 * Finds segment either by shmid if is_shmid is true, or by segnum if
216 * is_shmid is false.
217 */
218 static struct shmid_kernel *
shm_find_segment(struct prison * rpr,int arg,bool is_shmid)219 shm_find_segment(struct prison *rpr, int arg, bool is_shmid)
220 {
221 struct shmid_kernel *shmseg;
222 int segnum;
223
224 segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
225 if (segnum < 0 || segnum >= shmalloced)
226 return (NULL);
227 shmseg = &shmsegs[segnum];
228 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
229 (!shm_allow_removed &&
230 (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
231 (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)) ||
232 shm_prison_cansee(rpr, shmseg) != 0)
233 return (NULL);
234 return (shmseg);
235 }
236
237 static void
shm_deallocate_segment(struct shmid_kernel * shmseg)238 shm_deallocate_segment(struct shmid_kernel *shmseg)
239 {
240 vm_size_t size;
241
242 SYSVSHM_ASSERT_LOCKED();
243
244 vm_object_deallocate(shmseg->object);
245 shmseg->object = NULL;
246 size = round_page(shmseg->u.shm_segsz);
247 shm_committed -= btoc(size);
248 shm_nused--;
249 shmseg->u.shm_perm.mode = SHMSEG_FREE;
250 #ifdef MAC
251 mac_sysvshm_cleanup(shmseg);
252 #endif
253 racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
254 racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
255 crfree(shmseg->cred);
256 shmseg->cred = NULL;
257 }
258
259 static int
shm_delete_mapping(struct vmspace * vm,struct shmmap_state * shmmap_s)260 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
261 {
262 struct shmid_kernel *shmseg;
263 int segnum, result;
264 vm_size_t size;
265
266 SYSVSHM_ASSERT_LOCKED();
267 segnum = IPCID_TO_IX(shmmap_s->shmid);
268 KASSERT(segnum >= 0 && segnum < shmalloced,
269 ("segnum %d shmalloced %d", segnum, shmalloced));
270
271 shmseg = &shmsegs[segnum];
272 size = round_page(shmseg->u.shm_segsz);
273 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
274 if (result != KERN_SUCCESS)
275 return (EINVAL);
276 shmmap_s->shmid = -1;
277 shmseg->u.shm_dtime = time_second;
278 if ((--shmseg->u.shm_nattch <= 0) &&
279 (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
280 shm_deallocate_segment(shmseg);
281 shm_last_free = segnum;
282 }
283 return (0);
284 }
285
286 static void
shm_remove(struct shmid_kernel * shmseg,int segnum)287 shm_remove(struct shmid_kernel *shmseg, int segnum)
288 {
289
290 shmseg->u.shm_perm.key = IPC_PRIVATE;
291 shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
292 if (shmseg->u.shm_nattch <= 0) {
293 shm_deallocate_segment(shmseg);
294 shm_last_free = segnum;
295 }
296 }
297
298 static struct prison *
shm_find_prison(struct ucred * cred)299 shm_find_prison(struct ucred *cred)
300 {
301 struct prison *pr, *rpr;
302
303 pr = cred->cr_prison;
304 prison_lock(pr);
305 rpr = osd_jail_get(pr, shm_prison_slot);
306 prison_unlock(pr);
307 return rpr;
308 }
309
310 static int
shm_prison_cansee(struct prison * rpr,struct shmid_kernel * shmseg)311 shm_prison_cansee(struct prison *rpr, struct shmid_kernel *shmseg)
312 {
313
314 if (shmseg->cred == NULL ||
315 !(rpr == shmseg->cred->cr_prison ||
316 prison_ischild(rpr, shmseg->cred->cr_prison)))
317 return (EINVAL);
318 return (0);
319 }
320
321 static int
kern_shmdt_locked(struct thread * td,const void * shmaddr)322 kern_shmdt_locked(struct thread *td, const void *shmaddr)
323 {
324 struct proc *p = td->td_proc;
325 struct shmmap_state *shmmap_s;
326 #ifdef MAC
327 struct shmid_kernel *shmsegptr;
328 int error;
329 #endif
330 int i;
331
332 SYSVSHM_ASSERT_LOCKED();
333 if (shm_find_prison(td->td_ucred) == NULL)
334 return (ENOSYS);
335 shmmap_s = p->p_vmspace->vm_shm;
336 if (shmmap_s == NULL)
337 return (EINVAL);
338 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
339 if (shmmap_s->shmid != -1 &&
340 shmmap_s->va == (vm_offset_t)shmaddr) {
341 break;
342 }
343 }
344 if (i == shminfo.shmseg)
345 return (EINVAL);
346 #ifdef MAC
347 shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
348 error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
349 if (error != 0)
350 return (error);
351 #endif
352 return (shm_delete_mapping(p->p_vmspace, shmmap_s));
353 }
354
355 #ifndef _SYS_SYSPROTO_H_
356 struct shmdt_args {
357 const void *shmaddr;
358 };
359 #endif
360 int
sys_shmdt(struct thread * td,struct shmdt_args * uap)361 sys_shmdt(struct thread *td, struct shmdt_args *uap)
362 {
363 int error;
364
365 SYSVSHM_LOCK();
366 error = kern_shmdt_locked(td, uap->shmaddr);
367 SYSVSHM_UNLOCK();
368 return (error);
369 }
370
371 static int
kern_shmat_locked(struct thread * td,int shmid,const void * shmaddr,int shmflg)372 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
373 int shmflg)
374 {
375 struct prison *rpr;
376 struct proc *p = td->td_proc;
377 struct shmid_kernel *shmseg;
378 struct shmmap_state *shmmap_s;
379 vm_offset_t attach_va;
380 vm_prot_t prot;
381 vm_size_t size;
382 int cow, error, find_space, i, rv;
383
384 SYSVSHM_ASSERT_LOCKED();
385 rpr = shm_find_prison(td->td_ucred);
386 if (rpr == NULL)
387 return (ENOSYS);
388 shmmap_s = p->p_vmspace->vm_shm;
389 if (shmmap_s == NULL) {
390 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
391 M_SHM, M_WAITOK);
392 for (i = 0; i < shminfo.shmseg; i++)
393 shmmap_s[i].shmid = -1;
394 KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
395 p->p_vmspace->vm_shm = shmmap_s;
396 }
397 shmseg = shm_find_segment(rpr, shmid, true);
398 if (shmseg == NULL)
399 return (EINVAL);
400 error = ipcperm(td, &shmseg->u.shm_perm,
401 (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
402 if (error != 0)
403 return (error);
404 #ifdef MAC
405 error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
406 if (error != 0)
407 return (error);
408 #endif
409 for (i = 0; i < shminfo.shmseg; i++) {
410 if (shmmap_s->shmid == -1)
411 break;
412 shmmap_s++;
413 }
414 if (i >= shminfo.shmseg)
415 return (EMFILE);
416 size = round_page(shmseg->u.shm_segsz);
417 prot = VM_PROT_READ;
418 cow = MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL;
419 if ((shmflg & SHM_RDONLY) == 0)
420 prot |= VM_PROT_WRITE;
421 if (shmaddr != NULL) {
422 if ((shmflg & SHM_RND) != 0)
423 attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
424 else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
425 attach_va = (vm_offset_t)shmaddr;
426 else
427 return (EINVAL);
428 if ((shmflg & SHM_REMAP) != 0)
429 cow |= MAP_REMAP;
430 find_space = VMFS_NO_SPACE;
431 } else {
432 /*
433 * This is just a hint to vm_map_find() about where to
434 * put it.
435 */
436 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
437 lim_max(td, RLIMIT_DATA));
438 find_space = VMFS_OPTIMAL_SPACE;
439 }
440
441 vm_object_reference(shmseg->object);
442 rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
443 size, 0, find_space, prot, prot, cow);
444 if (rv != KERN_SUCCESS) {
445 vm_object_deallocate(shmseg->object);
446 return (ENOMEM);
447 }
448
449 shmmap_s->va = attach_va;
450 shmmap_s->shmid = shmid;
451 shmseg->u.shm_lpid = p->p_pid;
452 shmseg->u.shm_atime = time_second;
453 shmseg->u.shm_nattch++;
454 td->td_retval[0] = attach_va;
455 return (error);
456 }
457
458 int
kern_shmat(struct thread * td,int shmid,const void * shmaddr,int shmflg)459 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
460 {
461 int error;
462
463 SYSVSHM_LOCK();
464 error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
465 SYSVSHM_UNLOCK();
466 return (error);
467 }
468
469 #ifndef _SYS_SYSPROTO_H_
470 struct shmat_args {
471 int shmid;
472 const void *shmaddr;
473 int shmflg;
474 };
475 #endif
476 int
sys_shmat(struct thread * td,struct shmat_args * uap)477 sys_shmat(struct thread *td, struct shmat_args *uap)
478 {
479
480 return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
481 }
482
483 static int
kern_shmctl_locked(struct thread * td,int shmid,int cmd,void * buf,size_t * bufsz)484 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
485 size_t *bufsz)
486 {
487 struct prison *rpr;
488 struct shmid_kernel *shmseg;
489 struct shmid_ds *shmidp;
490 struct shm_info shm_info;
491 int error;
492
493 SYSVSHM_ASSERT_LOCKED();
494
495 rpr = shm_find_prison(td->td_ucred);
496 if (rpr == NULL)
497 return (ENOSYS);
498
499 switch (cmd) {
500 /*
501 * It is possible that kern_shmctl is being called from the Linux ABI
502 * layer, in which case, we will need to implement IPC_INFO. It should
503 * be noted that other shmctl calls will be funneled through here for
504 * Linix binaries as well.
505 *
506 * NB: The Linux ABI layer will convert this data to structure(s) more
507 * consistent with the Linux ABI.
508 */
509 case IPC_INFO:
510 memcpy(buf, &shminfo, sizeof(shminfo));
511 if (bufsz)
512 *bufsz = sizeof(shminfo);
513 td->td_retval[0] = shmalloced;
514 return (0);
515 case SHM_INFO: {
516 shm_info.used_ids = shm_nused;
517 shm_info.shm_rss = 0; /*XXX where to get from ? */
518 shm_info.shm_tot = 0; /*XXX where to get from ? */
519 shm_info.shm_swp = 0; /*XXX where to get from ? */
520 shm_info.swap_attempts = 0; /*XXX where to get from ? */
521 shm_info.swap_successes = 0; /*XXX where to get from ? */
522 memcpy(buf, &shm_info, sizeof(shm_info));
523 if (bufsz != NULL)
524 *bufsz = sizeof(shm_info);
525 td->td_retval[0] = shmalloced;
526 return (0);
527 }
528 }
529 shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
530 if (shmseg == NULL)
531 return (EINVAL);
532 #ifdef MAC
533 error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
534 if (error != 0)
535 return (error);
536 #endif
537 switch (cmd) {
538 case SHM_STAT:
539 case IPC_STAT:
540 shmidp = (struct shmid_ds *)buf;
541 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
542 if (error != 0)
543 return (error);
544 memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
545 if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
546 shmidp->shm_perm.key = IPC_PRIVATE;
547 if (bufsz != NULL)
548 *bufsz = sizeof(struct shmid_ds);
549 if (cmd == SHM_STAT) {
550 td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
551 shmseg->u.shm_perm);
552 }
553 break;
554 case IPC_SET:
555 shmidp = (struct shmid_ds *)buf;
556 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
557 if (error != 0)
558 return (error);
559 shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
560 shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
561 shmseg->u.shm_perm.mode =
562 (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
563 (shmidp->shm_perm.mode & ACCESSPERMS);
564 shmseg->u.shm_ctime = time_second;
565 break;
566 case IPC_RMID:
567 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
568 if (error != 0)
569 return (error);
570 shm_remove(shmseg, IPCID_TO_IX(shmid));
571 break;
572 #if 0
573 case SHM_LOCK:
574 case SHM_UNLOCK:
575 #endif
576 default:
577 error = EINVAL;
578 break;
579 }
580 return (error);
581 }
582
583 int
kern_shmctl(struct thread * td,int shmid,int cmd,void * buf,size_t * bufsz)584 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
585 {
586 int error;
587
588 SYSVSHM_LOCK();
589 error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
590 SYSVSHM_UNLOCK();
591 return (error);
592 }
593
594
595 #ifndef _SYS_SYSPROTO_H_
596 struct shmctl_args {
597 int shmid;
598 int cmd;
599 struct shmid_ds *buf;
600 };
601 #endif
602 int
sys_shmctl(struct thread * td,struct shmctl_args * uap)603 sys_shmctl(struct thread *td, struct shmctl_args *uap)
604 {
605 int error;
606 struct shmid_ds buf;
607 size_t bufsz;
608
609 /*
610 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
611 * Linux binaries. If we see the call come through the FreeBSD ABI,
612 * return an error back to the user since we do not to support this.
613 */
614 if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
615 uap->cmd == SHM_STAT)
616 return (EINVAL);
617
618 /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
619 if (uap->cmd == IPC_SET) {
620 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
621 goto done;
622 }
623
624 error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
625 if (error)
626 goto done;
627
628 /* Cases in which we need to copyout */
629 switch (uap->cmd) {
630 case IPC_STAT:
631 error = copyout(&buf, uap->buf, bufsz);
632 break;
633 }
634
635 done:
636 if (error) {
637 /* Invalidate the return value */
638 td->td_retval[0] = -1;
639 }
640 return (error);
641 }
642
643
644 static int
shmget_existing(struct thread * td,size_t size,int shmflg,int mode,int segnum)645 shmget_existing(struct thread *td, size_t size, int shmflg, int mode,
646 int segnum)
647 {
648 struct shmid_kernel *shmseg;
649 #ifdef MAC
650 int error;
651 #endif
652
653 SYSVSHM_ASSERT_LOCKED();
654 KASSERT(segnum >= 0 && segnum < shmalloced,
655 ("segnum %d shmalloced %d", segnum, shmalloced));
656 shmseg = &shmsegs[segnum];
657 if ((shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
658 return (EEXIST);
659 #ifdef MAC
660 error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, shmflg);
661 if (error != 0)
662 return (error);
663 #endif
664 if (size != 0 && size > shmseg->u.shm_segsz)
665 return (EINVAL);
666 td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
667 return (0);
668 }
669
670 static int
shmget_allocate_segment(struct thread * td,key_t key,size_t size,int mode)671 shmget_allocate_segment(struct thread *td, key_t key, size_t size, int mode)
672 {
673 struct ucred *cred = td->td_ucred;
674 struct shmid_kernel *shmseg;
675 vm_object_t shm_object;
676 int i, segnum;
677
678 SYSVSHM_ASSERT_LOCKED();
679
680 if (size < shminfo.shmmin || size > shminfo.shmmax)
681 return (EINVAL);
682 if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
683 return (ENOSPC);
684 size = round_page(size);
685 if (shm_committed + btoc(size) > shminfo.shmall)
686 return (ENOMEM);
687 if (shm_last_free < 0) {
688 shmrealloc(); /* Maybe expand the shmsegs[] array. */
689 for (i = 0; i < shmalloced; i++)
690 if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
691 break;
692 if (i == shmalloced)
693 return (ENOSPC);
694 segnum = i;
695 } else {
696 segnum = shm_last_free;
697 shm_last_free = -1;
698 }
699 KASSERT(segnum >= 0 && segnum < shmalloced,
700 ("segnum %d shmalloced %d", segnum, shmalloced));
701 shmseg = &shmsegs[segnum];
702 #ifdef RACCT
703 if (racct_enable) {
704 PROC_LOCK(td->td_proc);
705 if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
706 PROC_UNLOCK(td->td_proc);
707 return (ENOSPC);
708 }
709 if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
710 racct_sub(td->td_proc, RACCT_NSHM, 1);
711 PROC_UNLOCK(td->td_proc);
712 return (ENOMEM);
713 }
714 PROC_UNLOCK(td->td_proc);
715 }
716 #endif
717
718 /*
719 * We make sure that we have allocated a pager before we need
720 * to.
721 */
722 shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
723 0, size, VM_PROT_DEFAULT, 0, cred);
724 if (shm_object == NULL) {
725 #ifdef RACCT
726 if (racct_enable) {
727 PROC_LOCK(td->td_proc);
728 racct_sub(td->td_proc, RACCT_NSHM, 1);
729 racct_sub(td->td_proc, RACCT_SHMSIZE, size);
730 PROC_UNLOCK(td->td_proc);
731 }
732 #endif
733 return (ENOMEM);
734 }
735 shm_object->pg_color = 0;
736 VM_OBJECT_WLOCK(shm_object);
737 vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
738 vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
739 VM_OBJECT_WUNLOCK(shm_object);
740
741 shmseg->object = shm_object;
742 shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
743 shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
744 shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
745 shmseg->u.shm_perm.key = key;
746 shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
747 shmseg->cred = crhold(cred);
748 shmseg->u.shm_segsz = size;
749 shmseg->u.shm_cpid = td->td_proc->p_pid;
750 shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
751 shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
752 #ifdef MAC
753 mac_sysvshm_create(cred, shmseg);
754 #endif
755 shmseg->u.shm_ctime = time_second;
756 shm_committed += btoc(size);
757 shm_nused++;
758 td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
759
760 return (0);
761 }
762
763 #ifndef _SYS_SYSPROTO_H_
764 struct shmget_args {
765 key_t key;
766 size_t size;
767 int shmflg;
768 };
769 #endif
770 int
sys_shmget(struct thread * td,struct shmget_args * uap)771 sys_shmget(struct thread *td, struct shmget_args *uap)
772 {
773 int segnum, mode;
774 int error;
775
776 if (shm_find_prison(td->td_ucred) == NULL)
777 return (ENOSYS);
778 mode = uap->shmflg & ACCESSPERMS;
779 SYSVSHM_LOCK();
780 if (uap->key == IPC_PRIVATE) {
781 error = shmget_allocate_segment(td, uap->key, uap->size, mode);
782 } else {
783 segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
784 uap->key);
785 if (segnum >= 0)
786 error = shmget_existing(td, uap->size, uap->shmflg,
787 mode, segnum);
788 else if ((uap->shmflg & IPC_CREAT) == 0)
789 error = ENOENT;
790 else
791 error = shmget_allocate_segment(td, uap->key,
792 uap->size, mode);
793 }
794 SYSVSHM_UNLOCK();
795 return (error);
796 }
797
798 static void
shmfork_myhook(struct proc * p1,struct proc * p2)799 shmfork_myhook(struct proc *p1, struct proc *p2)
800 {
801 struct shmmap_state *shmmap_s;
802 size_t size;
803 int i;
804
805 SYSVSHM_LOCK();
806 size = shminfo.shmseg * sizeof(struct shmmap_state);
807 shmmap_s = malloc(size, M_SHM, M_WAITOK);
808 bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
809 p2->p_vmspace->vm_shm = shmmap_s;
810 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
811 if (shmmap_s->shmid != -1) {
812 KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
813 IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
814 ("segnum %d shmalloced %d",
815 IPCID_TO_IX(shmmap_s->shmid), shmalloced));
816 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
817 }
818 }
819 SYSVSHM_UNLOCK();
820 }
821
822 static void
shmexit_myhook(struct vmspace * vm)823 shmexit_myhook(struct vmspace *vm)
824 {
825 struct shmmap_state *base, *shm;
826 int i;
827
828 base = vm->vm_shm;
829 if (base != NULL) {
830 vm->vm_shm = NULL;
831 SYSVSHM_LOCK();
832 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
833 if (shm->shmid != -1)
834 shm_delete_mapping(vm, shm);
835 }
836 SYSVSHM_UNLOCK();
837 free(base, M_SHM);
838 }
839 }
840
841 static void
shmrealloc(void)842 shmrealloc(void)
843 {
844 struct shmid_kernel *newsegs;
845 int i;
846
847 SYSVSHM_ASSERT_LOCKED();
848
849 if (shmalloced >= shminfo.shmmni)
850 return;
851
852 newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM,
853 M_WAITOK | M_ZERO);
854 for (i = 0; i < shmalloced; i++)
855 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
856 for (; i < shminfo.shmmni; i++) {
857 newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
858 newsegs[i].u.shm_perm.seq = 0;
859 #ifdef MAC
860 mac_sysvshm_init(&newsegs[i]);
861 #endif
862 }
863 free(shmsegs, M_SHM);
864 shmsegs = newsegs;
865 shmalloced = shminfo.shmmni;
866 }
867
868 static struct syscall_helper_data shm_syscalls[] = {
869 SYSCALL_INIT_HELPER(shmat),
870 SYSCALL_INIT_HELPER(shmctl),
871 SYSCALL_INIT_HELPER(shmdt),
872 SYSCALL_INIT_HELPER(shmget),
873 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
874 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
875 SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
876 #endif
877 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
878 SYSCALL_INIT_HELPER(shmsys),
879 #endif
880 SYSCALL_INIT_LAST
881 };
882
883 #ifdef COMPAT_FREEBSD32
884 #include <compat/freebsd32/freebsd32.h>
885 #include <compat/freebsd32/freebsd32_ipc.h>
886 #include <compat/freebsd32/freebsd32_proto.h>
887 #include <compat/freebsd32/freebsd32_signal.h>
888 #include <compat/freebsd32/freebsd32_syscall.h>
889 #include <compat/freebsd32/freebsd32_util.h>
890
891 static struct syscall_helper_data shm32_syscalls[] = {
892 SYSCALL32_INIT_HELPER_COMPAT(shmat),
893 SYSCALL32_INIT_HELPER_COMPAT(shmdt),
894 SYSCALL32_INIT_HELPER_COMPAT(shmget),
895 SYSCALL32_INIT_HELPER(freebsd32_shmsys),
896 SYSCALL32_INIT_HELPER(freebsd32_shmctl),
897 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
898 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
899 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
900 #endif
901 SYSCALL_INIT_LAST
902 };
903 #endif
904
905 static int
shminit(void)906 shminit(void)
907 {
908 struct prison *pr;
909 void **rsv;
910 int i, error;
911 osd_method_t methods[PR_MAXMETHOD] = {
912 [PR_METHOD_CHECK] = shm_prison_check,
913 [PR_METHOD_SET] = shm_prison_set,
914 [PR_METHOD_GET] = shm_prison_get,
915 [PR_METHOD_REMOVE] = shm_prison_remove,
916 };
917
918 #ifndef BURN_BRIDGES
919 if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
920 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
921 #endif
922 if (shminfo.shmmax == SHMMAX) {
923 /* Initialize shmmax dealing with possible overflow. */
924 for (i = PAGE_SIZE; i != 0; i--) {
925 shminfo.shmmax = shminfo.shmall * i;
926 if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
927 break;
928 }
929 }
930 shmalloced = shminfo.shmmni;
931 shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM,
932 M_WAITOK|M_ZERO);
933 for (i = 0; i < shmalloced; i++) {
934 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
935 shmsegs[i].u.shm_perm.seq = 0;
936 #ifdef MAC
937 mac_sysvshm_init(&shmsegs[i]);
938 #endif
939 }
940 shm_last_free = 0;
941 shm_nused = 0;
942 shm_committed = 0;
943 sx_init(&sysvshmsx, "sysvshmsx");
944 shmexit_hook = &shmexit_myhook;
945 shmfork_hook = &shmfork_myhook;
946
947 /* Set current prisons according to their allow.sysvipc. */
948 shm_prison_slot = osd_jail_register(NULL, methods);
949 rsv = osd_reserve(shm_prison_slot);
950 prison_lock(&prison0);
951 (void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
952 prison_unlock(&prison0);
953 rsv = NULL;
954 sx_slock(&allprison_lock);
955 TAILQ_FOREACH(pr, &allprison, pr_list) {
956 if (rsv == NULL)
957 rsv = osd_reserve(shm_prison_slot);
958 prison_lock(pr);
959 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
960 (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
961 &prison0);
962 rsv = NULL;
963 }
964 prison_unlock(pr);
965 }
966 if (rsv != NULL)
967 osd_free_reserved(rsv);
968 sx_sunlock(&allprison_lock);
969
970 error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
971 if (error != 0)
972 return (error);
973 #ifdef COMPAT_FREEBSD32
974 error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
975 if (error != 0)
976 return (error);
977 #endif
978 return (0);
979 }
980
981 static int
shmunload(void)982 shmunload(void)
983 {
984 int i;
985
986 if (shm_nused > 0)
987 return (EBUSY);
988
989 #ifdef COMPAT_FREEBSD32
990 syscall32_helper_unregister(shm32_syscalls);
991 #endif
992 syscall_helper_unregister(shm_syscalls);
993 if (shm_prison_slot != 0)
994 osd_jail_deregister(shm_prison_slot);
995
996 for (i = 0; i < shmalloced; i++) {
997 #ifdef MAC
998 mac_sysvshm_destroy(&shmsegs[i]);
999 #endif
1000 /*
1001 * Objects might be still mapped into the processes
1002 * address spaces. Actual free would happen on the
1003 * last mapping destruction.
1004 */
1005 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1006 vm_object_deallocate(shmsegs[i].object);
1007 }
1008 free(shmsegs, M_SHM);
1009 shmexit_hook = NULL;
1010 shmfork_hook = NULL;
1011 sx_destroy(&sysvshmsx);
1012 return (0);
1013 }
1014
1015 static int
sysctl_shmsegs(SYSCTL_HANDLER_ARGS)1016 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1017 {
1018 struct shmid_kernel tshmseg;
1019 #ifdef COMPAT_FREEBSD32
1020 struct shmid_kernel32 tshmseg32;
1021 #endif
1022 struct prison *pr, *rpr;
1023 void *outaddr;
1024 size_t outsize;
1025 int error, i;
1026
1027 SYSVSHM_LOCK();
1028 pr = req->td->td_ucred->cr_prison;
1029 rpr = shm_find_prison(req->td->td_ucred);
1030 error = 0;
1031 for (i = 0; i < shmalloced; i++) {
1032 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1033 rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1034 bzero(&tshmseg, sizeof(tshmseg));
1035 tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1036 } else {
1037 tshmseg = shmsegs[i];
1038 if (tshmseg.cred->cr_prison != pr)
1039 tshmseg.u.shm_perm.key = IPC_PRIVATE;
1040 }
1041 #ifdef COMPAT_FREEBSD32
1042 if (SV_CURPROC_FLAG(SV_ILP32)) {
1043 bzero(&tshmseg32, sizeof(tshmseg32));
1044 freebsd32_ipcperm_out(&tshmseg.u.shm_perm,
1045 &tshmseg32.u.shm_perm);
1046 CP(tshmseg, tshmseg32, u.shm_segsz);
1047 CP(tshmseg, tshmseg32, u.shm_lpid);
1048 CP(tshmseg, tshmseg32, u.shm_cpid);
1049 CP(tshmseg, tshmseg32, u.shm_nattch);
1050 CP(tshmseg, tshmseg32, u.shm_atime);
1051 CP(tshmseg, tshmseg32, u.shm_dtime);
1052 CP(tshmseg, tshmseg32, u.shm_ctime);
1053 /* Don't copy object, label, or cred */
1054 outaddr = &tshmseg32;
1055 outsize = sizeof(tshmseg32);
1056 } else
1057 #endif
1058 {
1059 tshmseg.object = NULL;
1060 tshmseg.label = NULL;
1061 tshmseg.cred = NULL;
1062 outaddr = &tshmseg;
1063 outsize = sizeof(tshmseg);
1064 }
1065 error = SYSCTL_OUT(req, outaddr, outsize);
1066 if (error != 0)
1067 break;
1068 }
1069 SYSVSHM_UNLOCK();
1070 return (error);
1071 }
1072
1073 static int
shm_prison_check(void * obj,void * data)1074 shm_prison_check(void *obj, void *data)
1075 {
1076 struct prison *pr = obj;
1077 struct prison *prpr;
1078 struct vfsoptlist *opts = data;
1079 int error, jsys;
1080
1081 /*
1082 * sysvshm is a jailsys integer.
1083 * It must be "disable" if the parent jail is disabled.
1084 */
1085 error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1086 if (error != ENOENT) {
1087 if (error != 0)
1088 return (error);
1089 switch (jsys) {
1090 case JAIL_SYS_DISABLE:
1091 break;
1092 case JAIL_SYS_NEW:
1093 case JAIL_SYS_INHERIT:
1094 prison_lock(pr->pr_parent);
1095 prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1096 prison_unlock(pr->pr_parent);
1097 if (prpr == NULL)
1098 return (EPERM);
1099 break;
1100 default:
1101 return (EINVAL);
1102 }
1103 }
1104
1105 return (0);
1106 }
1107
1108 static int
shm_prison_set(void * obj,void * data)1109 shm_prison_set(void *obj, void *data)
1110 {
1111 struct prison *pr = obj;
1112 struct prison *tpr, *orpr, *nrpr, *trpr;
1113 struct vfsoptlist *opts = data;
1114 void *rsv;
1115 int jsys, descend;
1116
1117 /*
1118 * sysvshm controls which jail is the root of the associated segments
1119 * (this jail or same as the parent), or if the feature is available
1120 * at all.
1121 */
1122 if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1123 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1124 ? JAIL_SYS_INHERIT
1125 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1126 ? JAIL_SYS_DISABLE
1127 : -1;
1128 if (jsys == JAIL_SYS_DISABLE) {
1129 prison_lock(pr);
1130 orpr = osd_jail_get(pr, shm_prison_slot);
1131 if (orpr != NULL)
1132 osd_jail_del(pr, shm_prison_slot);
1133 prison_unlock(pr);
1134 if (orpr != NULL) {
1135 if (orpr == pr)
1136 shm_prison_cleanup(pr);
1137 /* Disable all child jails as well. */
1138 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1139 prison_lock(tpr);
1140 trpr = osd_jail_get(tpr, shm_prison_slot);
1141 if (trpr != NULL) {
1142 osd_jail_del(tpr, shm_prison_slot);
1143 prison_unlock(tpr);
1144 if (trpr == tpr)
1145 shm_prison_cleanup(tpr);
1146 } else {
1147 prison_unlock(tpr);
1148 descend = 0;
1149 }
1150 }
1151 }
1152 } else if (jsys != -1) {
1153 if (jsys == JAIL_SYS_NEW)
1154 nrpr = pr;
1155 else {
1156 prison_lock(pr->pr_parent);
1157 nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1158 prison_unlock(pr->pr_parent);
1159 }
1160 rsv = osd_reserve(shm_prison_slot);
1161 prison_lock(pr);
1162 orpr = osd_jail_get(pr, shm_prison_slot);
1163 if (orpr != nrpr)
1164 (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1165 nrpr);
1166 else
1167 osd_free_reserved(rsv);
1168 prison_unlock(pr);
1169 if (orpr != nrpr) {
1170 if (orpr == pr)
1171 shm_prison_cleanup(pr);
1172 if (orpr != NULL) {
1173 /* Change child jails matching the old root, */
1174 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1175 prison_lock(tpr);
1176 trpr = osd_jail_get(tpr,
1177 shm_prison_slot);
1178 if (trpr == orpr) {
1179 (void)osd_jail_set(tpr,
1180 shm_prison_slot, nrpr);
1181 prison_unlock(tpr);
1182 if (trpr == tpr)
1183 shm_prison_cleanup(tpr);
1184 } else {
1185 prison_unlock(tpr);
1186 descend = 0;
1187 }
1188 }
1189 }
1190 }
1191 }
1192
1193 return (0);
1194 }
1195
1196 static int
shm_prison_get(void * obj,void * data)1197 shm_prison_get(void *obj, void *data)
1198 {
1199 struct prison *pr = obj;
1200 struct prison *rpr;
1201 struct vfsoptlist *opts = data;
1202 int error, jsys;
1203
1204 /* Set sysvshm based on the jail's root prison. */
1205 prison_lock(pr);
1206 rpr = osd_jail_get(pr, shm_prison_slot);
1207 prison_unlock(pr);
1208 jsys = rpr == NULL ? JAIL_SYS_DISABLE
1209 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1210 error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1211 if (error == ENOENT)
1212 error = 0;
1213 return (error);
1214 }
1215
1216 static int
shm_prison_remove(void * obj,void * data __unused)1217 shm_prison_remove(void *obj, void *data __unused)
1218 {
1219 struct prison *pr = obj;
1220 struct prison *rpr;
1221
1222 SYSVSHM_LOCK();
1223 prison_lock(pr);
1224 rpr = osd_jail_get(pr, shm_prison_slot);
1225 prison_unlock(pr);
1226 if (rpr == pr)
1227 shm_prison_cleanup(pr);
1228 SYSVSHM_UNLOCK();
1229 return (0);
1230 }
1231
1232 static void
shm_prison_cleanup(struct prison * pr)1233 shm_prison_cleanup(struct prison *pr)
1234 {
1235 struct shmid_kernel *shmseg;
1236 int i;
1237
1238 /* Remove any segments that belong to this jail. */
1239 for (i = 0; i < shmalloced; i++) {
1240 shmseg = &shmsegs[i];
1241 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1242 shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1243 shm_remove(shmseg, i);
1244 }
1245 }
1246 }
1247
1248 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1249
1250 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1251 struct oshmid_ds {
1252 struct ipc_perm_old shm_perm; /* operation perms */
1253 int shm_segsz; /* size of segment (bytes) */
1254 u_short shm_cpid; /* pid, creator */
1255 u_short shm_lpid; /* pid, last operation */
1256 short shm_nattch; /* no. of current attaches */
1257 time_t shm_atime; /* last attach time */
1258 time_t shm_dtime; /* last detach time */
1259 time_t shm_ctime; /* last change time */
1260 void *shm_handle; /* internal handle for shm segment */
1261 };
1262
1263 struct oshmctl_args {
1264 int shmid;
1265 int cmd;
1266 struct oshmid_ds *ubuf;
1267 };
1268
1269 static int
oshmctl(struct thread * td,struct oshmctl_args * uap)1270 oshmctl(struct thread *td, struct oshmctl_args *uap)
1271 {
1272 #ifdef COMPAT_43
1273 int error = 0;
1274 struct prison *rpr;
1275 struct shmid_kernel *shmseg;
1276 struct oshmid_ds outbuf;
1277
1278 rpr = shm_find_prison(td->td_ucred);
1279 if (rpr == NULL)
1280 return (ENOSYS);
1281 if (uap->cmd != IPC_STAT) {
1282 return (freebsd7_shmctl(td,
1283 (struct freebsd7_shmctl_args *)uap));
1284 }
1285 SYSVSHM_LOCK();
1286 shmseg = shm_find_segment(rpr, uap->shmid, true);
1287 if (shmseg == NULL) {
1288 SYSVSHM_UNLOCK();
1289 return (EINVAL);
1290 }
1291 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1292 if (error != 0) {
1293 SYSVSHM_UNLOCK();
1294 return (error);
1295 }
1296 #ifdef MAC
1297 error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1298 if (error != 0) {
1299 SYSVSHM_UNLOCK();
1300 return (error);
1301 }
1302 #endif
1303 ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1304 outbuf.shm_segsz = shmseg->u.shm_segsz;
1305 outbuf.shm_cpid = shmseg->u.shm_cpid;
1306 outbuf.shm_lpid = shmseg->u.shm_lpid;
1307 outbuf.shm_nattch = shmseg->u.shm_nattch;
1308 outbuf.shm_atime = shmseg->u.shm_atime;
1309 outbuf.shm_dtime = shmseg->u.shm_dtime;
1310 outbuf.shm_ctime = shmseg->u.shm_ctime;
1311 outbuf.shm_handle = shmseg->object;
1312 SYSVSHM_UNLOCK();
1313 return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1314 #else
1315 return (EINVAL);
1316 #endif
1317 }
1318
1319 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1320 static sy_call_t *shmcalls[] = {
1321 (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1322 (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1323 (sy_call_t *)freebsd7_shmctl
1324 };
1325
1326 #ifndef _SYS_SYSPROTO_H_
1327 /* XXX actually varargs. */
1328 struct shmsys_args {
1329 int which;
1330 int a2;
1331 int a3;
1332 int a4;
1333 };
1334 #endif
1335 int
sys_shmsys(struct thread * td,struct shmsys_args * uap)1336 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1337 {
1338
1339 if (uap->which < 0 || uap->which >= nitems(shmcalls))
1340 return (EINVAL);
1341 return ((*shmcalls[uap->which])(td, &uap->a2));
1342 }
1343
1344 #endif /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1345
1346 #ifdef COMPAT_FREEBSD32
1347
1348 int
freebsd32_shmsys(struct thread * td,struct freebsd32_shmsys_args * uap)1349 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1350 {
1351
1352 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1353 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1354 switch (uap->which) {
1355 case 0: { /* shmat */
1356 struct shmat_args ap;
1357
1358 ap.shmid = uap->a2;
1359 ap.shmaddr = PTRIN(uap->a3);
1360 ap.shmflg = uap->a4;
1361 return (sysent[SYS_shmat].sy_call(td, &ap));
1362 }
1363 case 2: { /* shmdt */
1364 struct shmdt_args ap;
1365
1366 ap.shmaddr = PTRIN(uap->a2);
1367 return (sysent[SYS_shmdt].sy_call(td, &ap));
1368 }
1369 case 3: { /* shmget */
1370 struct shmget_args ap;
1371
1372 ap.key = uap->a2;
1373 ap.size = uap->a3;
1374 ap.shmflg = uap->a4;
1375 return (sysent[SYS_shmget].sy_call(td, &ap));
1376 }
1377 case 4: { /* shmctl */
1378 struct freebsd7_freebsd32_shmctl_args ap;
1379
1380 ap.shmid = uap->a2;
1381 ap.cmd = uap->a3;
1382 ap.buf = PTRIN(uap->a4);
1383 return (freebsd7_freebsd32_shmctl(td, &ap));
1384 }
1385 case 1: /* oshmctl */
1386 default:
1387 return (EINVAL);
1388 }
1389 #else
1390 return (nosys(td, NULL));
1391 #endif
1392 }
1393
1394 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1395 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1396 int
freebsd7_freebsd32_shmctl(struct thread * td,struct freebsd7_freebsd32_shmctl_args * uap)1397 freebsd7_freebsd32_shmctl(struct thread *td,
1398 struct freebsd7_freebsd32_shmctl_args *uap)
1399 {
1400 int error;
1401 union {
1402 struct shmid_ds shmid_ds;
1403 struct shm_info shm_info;
1404 struct shminfo shminfo;
1405 } u;
1406 union {
1407 struct shmid_ds32_old shmid_ds32;
1408 struct shm_info32 shm_info32;
1409 struct shminfo32 shminfo32;
1410 } u32;
1411 size_t sz;
1412
1413 if (uap->cmd == IPC_SET) {
1414 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1415 sizeof(u32.shmid_ds32))))
1416 goto done;
1417 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1418 &u.shmid_ds.shm_perm);
1419 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1420 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1421 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1422 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1423 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1424 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1425 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1426 }
1427
1428 error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1429 if (error)
1430 goto done;
1431
1432 /* Cases in which we need to copyout */
1433 switch (uap->cmd) {
1434 case IPC_INFO:
1435 CP(u.shminfo, u32.shminfo32, shmmax);
1436 CP(u.shminfo, u32.shminfo32, shmmin);
1437 CP(u.shminfo, u32.shminfo32, shmmni);
1438 CP(u.shminfo, u32.shminfo32, shmseg);
1439 CP(u.shminfo, u32.shminfo32, shmall);
1440 error = copyout(&u32.shminfo32, uap->buf,
1441 sizeof(u32.shminfo32));
1442 break;
1443 case SHM_INFO:
1444 CP(u.shm_info, u32.shm_info32, used_ids);
1445 CP(u.shm_info, u32.shm_info32, shm_rss);
1446 CP(u.shm_info, u32.shm_info32, shm_tot);
1447 CP(u.shm_info, u32.shm_info32, shm_swp);
1448 CP(u.shm_info, u32.shm_info32, swap_attempts);
1449 CP(u.shm_info, u32.shm_info32, swap_successes);
1450 error = copyout(&u32.shm_info32, uap->buf,
1451 sizeof(u32.shm_info32));
1452 break;
1453 case SHM_STAT:
1454 case IPC_STAT:
1455 memset(&u32.shmid_ds32, 0, sizeof(u32.shmid_ds32));
1456 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1457 &u32.shmid_ds32.shm_perm);
1458 if (u.shmid_ds.shm_segsz > INT32_MAX)
1459 u32.shmid_ds32.shm_segsz = INT32_MAX;
1460 else
1461 CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1462 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1463 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1464 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1465 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1466 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1467 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1468 u32.shmid_ds32.shm_internal = 0;
1469 error = copyout(&u32.shmid_ds32, uap->buf,
1470 sizeof(u32.shmid_ds32));
1471 break;
1472 }
1473
1474 done:
1475 if (error) {
1476 /* Invalidate the return value */
1477 td->td_retval[0] = -1;
1478 }
1479 return (error);
1480 }
1481 #endif
1482
1483 int
freebsd32_shmctl(struct thread * td,struct freebsd32_shmctl_args * uap)1484 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1485 {
1486 int error;
1487 union {
1488 struct shmid_ds shmid_ds;
1489 struct shm_info shm_info;
1490 struct shminfo shminfo;
1491 } u;
1492 union {
1493 struct shmid_ds32 shmid_ds32;
1494 struct shm_info32 shm_info32;
1495 struct shminfo32 shminfo32;
1496 } u32;
1497 size_t sz;
1498
1499 if (uap->cmd == IPC_SET) {
1500 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1501 sizeof(u32.shmid_ds32))))
1502 goto done;
1503 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1504 &u.shmid_ds.shm_perm);
1505 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1506 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1507 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1508 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1509 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1510 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1511 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1512 }
1513
1514 error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1515 if (error)
1516 goto done;
1517
1518 /* Cases in which we need to copyout */
1519 switch (uap->cmd) {
1520 case IPC_INFO:
1521 CP(u.shminfo, u32.shminfo32, shmmax);
1522 CP(u.shminfo, u32.shminfo32, shmmin);
1523 CP(u.shminfo, u32.shminfo32, shmmni);
1524 CP(u.shminfo, u32.shminfo32, shmseg);
1525 CP(u.shminfo, u32.shminfo32, shmall);
1526 error = copyout(&u32.shminfo32, uap->buf,
1527 sizeof(u32.shminfo32));
1528 break;
1529 case SHM_INFO:
1530 CP(u.shm_info, u32.shm_info32, used_ids);
1531 CP(u.shm_info, u32.shm_info32, shm_rss);
1532 CP(u.shm_info, u32.shm_info32, shm_tot);
1533 CP(u.shm_info, u32.shm_info32, shm_swp);
1534 CP(u.shm_info, u32.shm_info32, swap_attempts);
1535 CP(u.shm_info, u32.shm_info32, swap_successes);
1536 error = copyout(&u32.shm_info32, uap->buf,
1537 sizeof(u32.shm_info32));
1538 break;
1539 case SHM_STAT:
1540 case IPC_STAT:
1541 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1542 &u32.shmid_ds32.shm_perm);
1543 if (u.shmid_ds.shm_segsz > INT32_MAX)
1544 u32.shmid_ds32.shm_segsz = INT32_MAX;
1545 else
1546 CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1547 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1548 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1549 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1550 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1551 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1552 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1553 error = copyout(&u32.shmid_ds32, uap->buf,
1554 sizeof(u32.shmid_ds32));
1555 break;
1556 }
1557
1558 done:
1559 if (error) {
1560 /* Invalidate the return value */
1561 td->td_retval[0] = -1;
1562 }
1563 return (error);
1564 }
1565 #endif
1566
1567 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1568 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1569
1570 #ifndef _SYS_SYSPROTO_H_
1571 struct freebsd7_shmctl_args {
1572 int shmid;
1573 int cmd;
1574 struct shmid_ds_old *buf;
1575 };
1576 #endif
1577 int
freebsd7_shmctl(struct thread * td,struct freebsd7_shmctl_args * uap)1578 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1579 {
1580 int error;
1581 struct shmid_ds_old old;
1582 struct shmid_ds buf;
1583 size_t bufsz;
1584
1585 /*
1586 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1587 * Linux binaries. If we see the call come through the FreeBSD ABI,
1588 * return an error back to the user since we do not to support this.
1589 */
1590 if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1591 uap->cmd == SHM_STAT)
1592 return (EINVAL);
1593
1594 /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1595 if (uap->cmd == IPC_SET) {
1596 if ((error = copyin(uap->buf, &old, sizeof(old))))
1597 goto done;
1598 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1599 CP(old, buf, shm_segsz);
1600 CP(old, buf, shm_lpid);
1601 CP(old, buf, shm_cpid);
1602 CP(old, buf, shm_nattch);
1603 CP(old, buf, shm_atime);
1604 CP(old, buf, shm_dtime);
1605 CP(old, buf, shm_ctime);
1606 }
1607
1608 error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1609 if (error)
1610 goto done;
1611
1612 /* Cases in which we need to copyout */
1613 switch (uap->cmd) {
1614 case IPC_STAT:
1615 memset(&old, 0, sizeof(old));
1616 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1617 if (buf.shm_segsz > INT_MAX)
1618 old.shm_segsz = INT_MAX;
1619 else
1620 CP(buf, old, shm_segsz);
1621 CP(buf, old, shm_lpid);
1622 CP(buf, old, shm_cpid);
1623 if (buf.shm_nattch > SHRT_MAX)
1624 old.shm_nattch = SHRT_MAX;
1625 else
1626 CP(buf, old, shm_nattch);
1627 CP(buf, old, shm_atime);
1628 CP(buf, old, shm_dtime);
1629 CP(buf, old, shm_ctime);
1630 old.shm_internal = NULL;
1631 error = copyout(&old, uap->buf, sizeof(old));
1632 break;
1633 }
1634
1635 done:
1636 if (error) {
1637 /* Invalidate the return value */
1638 td->td_retval[0] = -1;
1639 }
1640 return (error);
1641 }
1642
1643 #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1644 COMPAT_FREEBSD7 */
1645
1646 static int
sysvshm_modload(struct module * module,int cmd,void * arg)1647 sysvshm_modload(struct module *module, int cmd, void *arg)
1648 {
1649 int error = 0;
1650
1651 switch (cmd) {
1652 case MOD_LOAD:
1653 error = shminit();
1654 if (error != 0)
1655 shmunload();
1656 break;
1657 case MOD_UNLOAD:
1658 error = shmunload();
1659 break;
1660 case MOD_SHUTDOWN:
1661 break;
1662 default:
1663 error = EINVAL;
1664 break;
1665 }
1666 return (error);
1667 }
1668
1669 static moduledata_t sysvshm_mod = {
1670 "sysvshm",
1671 &sysvshm_modload,
1672 NULL
1673 };
1674
1675 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1676 MODULE_VERSION(sysvshm, 1);
1677