sys/kern/sysv_shm.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-4-Clause AND BSD-2-Clause
   3  *
   4  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 3. All advertising materials mentioning features or use of this software
  15  *    must display the following acknowledgement:
  16  *      This product includes software developed by Adam Glass and Charles
  17  *      Hannum.
  18  * 4. The names of the authors may not be used to endorse or promote products
  19  *    derived from this software without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  24  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  *
  32  * $NetBSD: sysv_shm.c,v 1.39 1997/10/07 10:02:03 drochner Exp $
  33  */
  34 /*-
  35  * Copyright (c) 2003-2005 McAfee, Inc.
  36  * Copyright (c) 2016-2017 Robert N. M. Watson
  37  * All rights reserved.
  38  *
  39  * This software was developed for the FreeBSD Project in part by McAfee
  40  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
  41  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
  42  * program.
  43  *
  44  * Portions of this software were developed by BAE Systems, the University of
  45  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
  46  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
  47  * Computing (TC) research program.
  48  *
  49  * Redistribution and use in source and binary forms, with or without
  50  * modification, are permitted provided that the following conditions
  51  * are met:
  52  * 1. Redistributions of source code must retain the above copyright
  53  *    notice, this list of conditions and the following disclaimer.
  54  * 2. Redistributions in binary form must reproduce the above copyright
  55  *    notice, this list of conditions and the following disclaimer in the
  56  *    documentation and/or other materials provided with the distribution.
  57  *
  58  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  68  * SUCH DAMAGE.
  69  */
  70
  71 #include <sys/cdefs.h>
  72 #include "opt_sysvipc.h"
  73
  74 #include <sys/param.h>
  75 #include <sys/systm.h>
  76 #include <sys/abi_compat.h>
  77 #include <sys/kernel.h>
  78 #include <sys/limits.h>
  79 #include <sys/lock.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/shm.h>
  82 #include <sys/proc.h>
  83 #include <sys/malloc.h>
  84 #include <sys/mman.h>
  85 #include <sys/module.h>
  86 #include <sys/mutex.h>
  87 #include <sys/racct.h>
  88 #include <sys/resourcevar.h>
  89 #include <sys/rwlock.h>
  90 #include <sys/stat.h>
  91 #include <sys/syscall.h>
  92 #include <sys/syscallsubr.h>
  93 #include <sys/sysent.h>
  94 #include <sys/sysproto.h>
  95 #include <sys/jail.h>
  96
  97 #include <security/audit/audit.h>
  98 #include <security/mac/mac_framework.h>
  99
 100 #include <vm/vm.h>
 101 #include <vm/vm_param.h>
 102 #include <vm/pmap.h>
 103 #include <vm/vm_object.h>
 104 #include <vm/vm_map.h>
 105 #include <vm/vm_page.h>
 106 #include <vm/vm_pager.h>
 107
 108 FEATURE(sysv_shm, "System V shared memory segments support");
 109
 110 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
 111
 112 #define SHMSEG_FREE             0x0200
 113 #define SHMSEG_REMOVED          0x0400
 114 #define SHMSEG_ALLOCATED        0x0800
 115
 116 static int shm_last_free, shm_nused, shmalloced;
 117 vm_size_t shm_committed;
 118 static struct shmid_kernel *shmsegs;
 119 static unsigned shm_prison_slot;
 120
 121 struct shmmap_state {
 122         vm_offset_t va;
 123         int shmid;
 124 };
 125
 126 static void shm_deallocate_segment(struct shmid_kernel *);
 127 static int shm_find_segment_by_key(struct prison *, key_t);
 128 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
 129 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
 130 static int shmget_allocate_segment(struct thread *td, key_t key, size_t size,
 131     int mode);
 132 static int shmget_existing(struct thread *td, size_t size, int shmflg,
 133     int mode, int segnum);
 134 static void shmrealloc(void);
 135 static int shminit(void);
 136 static int sysvshm_modload(struct module *, int, void *);
 137 static int shmunload(void);
 138 #ifndef SYSVSHM
 139 static void shmexit_myhook(struct vmspace *vm);
 140 static void shmfork_myhook(struct proc *p1, struct proc *p2);
 141 #endif
 142 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
 143 static void shm_remove(struct shmid_kernel *, int);
 144 static struct prison *shm_find_prison(struct ucred *);
 145 static int shm_prison_cansee(struct prison *, struct shmid_kernel *);
 146 static int shm_prison_check(void *, void *);
 147 static int shm_prison_set(void *, void *);
 148 static int shm_prison_get(void *, void *);
 149 static int shm_prison_remove(void *, void *);
 150 static void shm_prison_cleanup(struct prison *);
 151
 152 /*
 153  * Tuneable values.
 154  */
 155 #ifndef SHMMAXPGS
 156 #define SHMMAXPGS       131072ul /* Note: sysv shared memory is swap backed. */
 157 #endif
 158 #ifndef SHMMAX
 159 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 160 #endif
 161 #ifndef SHMMIN
 162 #define SHMMIN  1
 163 #endif
 164 #ifndef SHMMNI
 165 #define SHMMNI  192
 166 #endif
 167 #ifndef SHMSEG
 168 #define SHMSEG  128
 169 #endif
 170 #ifndef SHMALL
 171 #define SHMALL  (SHMMAXPGS)
 172 #endif
 173
 174 struct  shminfo shminfo = {
 175         .shmmax = SHMMAX,
 176         .shmmin = SHMMIN,
 177         .shmmni = SHMMNI,
 178         .shmseg = SHMSEG,
 179         .shmall = SHMALL
 180 };
 181
 182 static int shm_use_phys;
 183 static int shm_allow_removed = 1;
 184
 185 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
 186     "Maximum shared memory segment size");
 187 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
 188     "Minimum shared memory segment size");
 189 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
 190     "Number of shared memory identifiers");
 191 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
 192     "Number of segments per process");
 193 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
 194     "Maximum number of pages available for shared memory");
 195 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
 196     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
 197 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
 198     &shm_allow_removed, 0,
 199     "Enable/Disable attachment to attached segments marked for removal");
 200 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
 201     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
 202     "Array of struct shmid_kernel for each potential shared memory segment");
 203
 204 static struct sx sysvshmsx;
 205 #define SYSVSHM_LOCK()          sx_xlock(&sysvshmsx)
 206 #define SYSVSHM_UNLOCK()        sx_xunlock(&sysvshmsx)
 207 #define SYSVSHM_ASSERT_LOCKED() sx_assert(&sysvshmsx, SA_XLOCKED)
 208
 209 static int
 210 shm_find_segment_by_key(struct prison *pr, key_t key)
 211 {
 212         int i;
 213
 214         for (i = 0; i < shmalloced; i++)
 215                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
 216                     shmsegs[i].cred != NULL &&
 217                     shmsegs[i].cred->cr_prison == pr &&
 218                     shmsegs[i].u.shm_perm.key == key)
 219                         return (i);
 220         return (-1);
 221 }
 222
 223 /*
 224  * Finds segment either by shmid if is_shmid is true, or by segnum if
 225  * is_shmid is false.
 226  */
 227 static struct shmid_kernel *
 228 shm_find_segment(struct prison *rpr, int arg, bool is_shmid)
 229 {
 230         struct shmid_kernel *shmseg;
 231         int segnum;
 232
 233         segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
 234         if (segnum < 0 || segnum >= shmalloced)
 235                 return (NULL);
 236         shmseg = &shmsegs[segnum];
 237         if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 238             (!shm_allow_removed &&
 239             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 240             (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)) ||
 241             shm_prison_cansee(rpr, shmseg) != 0)
 242                 return (NULL);
 243         return (shmseg);
 244 }
 245
 246 static void
 247 shm_deallocate_segment(struct shmid_kernel *shmseg)
 248 {
 249         vm_size_t size;
 250
 251         SYSVSHM_ASSERT_LOCKED();
 252
 253         vm_object_deallocate(shmseg->object);
 254         shmseg->object = NULL;
 255         size = round_page(shmseg->u.shm_segsz);
 256         shm_committed -= btoc(size);
 257         shm_nused--;
 258         shmseg->u.shm_perm.mode = SHMSEG_FREE;
 259 #ifdef MAC
 260         mac_sysvshm_cleanup(shmseg);
 261 #endif
 262         racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
 263         racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
 264         crfree(shmseg->cred);
 265         shmseg->cred = NULL;
 266 }
 267
 268 static int
 269 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 270 {
 271         struct shmid_kernel *shmseg;
 272         int segnum, result;
 273         vm_size_t size;
 274
 275         SYSVSHM_ASSERT_LOCKED();
 276         segnum = IPCID_TO_IX(shmmap_s->shmid);
 277         KASSERT(segnum >= 0 && segnum < shmalloced,
 278             ("segnum %d shmalloced %d", segnum, shmalloced));
 279
 280         shmseg = &shmsegs[segnum];
 281         size = round_page(shmseg->u.shm_segsz);
 282         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 283         if (result != KERN_SUCCESS)
 284                 return (EINVAL);
 285         shmmap_s->shmid = -1;
 286         shmseg->u.shm_dtime = time_second;
 287         if (--shmseg->u.shm_nattch == 0 &&
 288             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
 289                 shm_deallocate_segment(shmseg);
 290                 shm_last_free = segnum;
 291         }
 292         return (0);
 293 }
 294
 295 static void
 296 shm_remove(struct shmid_kernel *shmseg, int segnum)
 297 {
 298
 299         shmseg->u.shm_perm.key = IPC_PRIVATE;
 300         shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
 301         if (shmseg->u.shm_nattch == 0) {
 302                 shm_deallocate_segment(shmseg);
 303                 shm_last_free = segnum;
 304         }
 305 }
 306
 307 static struct prison *
 308 shm_find_prison(struct ucred *cred)
 309 {
 310         struct prison *pr, *rpr;
 311
 312         pr = cred->cr_prison;
 313         prison_lock(pr);
 314         rpr = osd_jail_get(pr, shm_prison_slot);
 315         prison_unlock(pr);
 316         return rpr;
 317 }
 318
 319 static int
 320 shm_prison_cansee(struct prison *rpr, struct shmid_kernel *shmseg)
 321 {
 322
 323         if (shmseg->cred == NULL ||
 324             !(rpr == shmseg->cred->cr_prison ||
 325               prison_ischild(rpr, shmseg->cred->cr_prison)))
 326                 return (EINVAL);
 327         return (0);
 328 }
 329
 330 static int
 331 kern_shmdt_locked(struct thread *td, const void *shmaddr)
 332 {
 333         struct proc *p = td->td_proc;
 334         struct shmmap_state *shmmap_s;
 335 #ifdef MAC
 336         int error;
 337 #endif
 338         int i;
 339
 340         SYSVSHM_ASSERT_LOCKED();
 341         if (shm_find_prison(td->td_ucred) == NULL)
 342                 return (ENOSYS);
 343         shmmap_s = p->p_vmspace->vm_shm;
 344         if (shmmap_s == NULL)
 345                 return (EINVAL);
 346         AUDIT_ARG_SVIPC_ID(shmmap_s->shmid);
 347         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 348                 if (shmmap_s->shmid != -1 &&
 349                     shmmap_s->va == (vm_offset_t)shmaddr) {
 350                         break;
 351                 }
 352         }
 353         if (i == shminfo.shmseg)
 354                 return (EINVAL);
 355 #ifdef MAC
 356         error = mac_sysvshm_check_shmdt(td->td_ucred,
 357             &shmsegs[IPCID_TO_IX(shmmap_s->shmid)]);
 358         if (error != 0)
 359                 return (error);
 360 #endif
 361         return (shm_delete_mapping(p->p_vmspace, shmmap_s));
 362 }
 363
 364 #ifndef _SYS_SYSPROTO_H_
 365 struct shmdt_args {
 366         const void *shmaddr;
 367 };
 368 #endif
 369 int
 370 sys_shmdt(struct thread *td, struct shmdt_args *uap)
 371 {
 372         int error;
 373
 374         SYSVSHM_LOCK();
 375         error = kern_shmdt_locked(td, uap->shmaddr);
 376         SYSVSHM_UNLOCK();
 377         return (error);
 378 }
 379
 380 static int
 381 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
 382     int shmflg)
 383 {
 384         struct prison *rpr;
 385         struct proc *p = td->td_proc;
 386         struct shmid_kernel *shmseg;
 387         struct shmmap_state *shmmap_s;
 388         vm_offset_t attach_va;
 389         vm_prot_t prot;
 390         vm_size_t size;
 391         int cow, error, find_space, i, rv;
 392
 393         AUDIT_ARG_SVIPC_ID(shmid);
 394         AUDIT_ARG_VALUE(shmflg);
 395
 396         SYSVSHM_ASSERT_LOCKED();
 397         rpr = shm_find_prison(td->td_ucred);
 398         if (rpr == NULL)
 399                 return (ENOSYS);
 400         shmmap_s = p->p_vmspace->vm_shm;
 401         if (shmmap_s == NULL) {
 402                 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
 403                     M_SHM, M_WAITOK);
 404                 for (i = 0; i < shminfo.shmseg; i++)
 405                         shmmap_s[i].shmid = -1;
 406                 KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
 407                 p->p_vmspace->vm_shm = shmmap_s;
 408         }
 409         shmseg = shm_find_segment(rpr, shmid, true);
 410         if (shmseg == NULL)
 411                 return (EINVAL);
 412         error = ipcperm(td, &shmseg->u.shm_perm,
 413             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 414         if (error != 0)
 415                 return (error);
 416 #ifdef MAC
 417         error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
 418         if (error != 0)
 419                 return (error);
 420 #endif
 421         for (i = 0; i < shminfo.shmseg; i++) {
 422                 if (shmmap_s->shmid == -1)
 423                         break;
 424                 shmmap_s++;
 425         }
 426         if (i >= shminfo.shmseg)
 427                 return (EMFILE);
 428         size = round_page(shmseg->u.shm_segsz);
 429         prot = VM_PROT_READ;
 430         cow = MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL;
 431         if ((shmflg & SHM_RDONLY) == 0)
 432                 prot |= VM_PROT_WRITE;
 433         if (shmaddr != NULL) {
 434                 if ((shmflg & SHM_RND) != 0)
 435                         attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
 436                 else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
 437                         attach_va = (vm_offset_t)shmaddr;
 438                 else
 439                         return (EINVAL);
 440                 if ((shmflg & SHM_REMAP) != 0)
 441                         cow |= MAP_REMAP;
 442                 find_space = VMFS_NO_SPACE;
 443         } else {
 444                 /*
 445                  * This is just a hint to vm_map_find() about where to
 446                  * put it.
 447                  */
 448                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
 449                     lim_max(td, RLIMIT_DATA));
 450                 find_space = VMFS_OPTIMAL_SPACE;
 451         }
 452
 453         vm_object_reference(shmseg->object);
 454         rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
 455             size, 0, find_space, prot, prot, cow);
 456         if (rv != KERN_SUCCESS) {
 457                 vm_object_deallocate(shmseg->object);
 458                 return (ENOMEM);
 459         }
 460
 461         shmmap_s->va = attach_va;
 462         shmmap_s->shmid = shmid;
 463         shmseg->u.shm_lpid = p->p_pid;
 464         shmseg->u.shm_atime = time_second;
 465         shmseg->u.shm_nattch++;
 466         td->td_retval[0] = attach_va;
 467         return (error);
 468 }
 469
 470 int
 471 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
 472 {
 473         int error;
 474
 475         SYSVSHM_LOCK();
 476         error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
 477         SYSVSHM_UNLOCK();
 478         return (error);
 479 }
 480
 481 #ifndef _SYS_SYSPROTO_H_
 482 struct shmat_args {
 483         int shmid;
 484         const void *shmaddr;
 485         int shmflg;
 486 };
 487 #endif
 488 int
 489 sys_shmat(struct thread *td, struct shmat_args *uap)
 490 {
 491
 492         return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
 493 }
 494
 495 static int
 496 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
 497     size_t *bufsz)
 498 {
 499         struct prison *rpr;
 500         struct shmid_kernel *shmseg;
 501         struct shmid_ds *shmidp;
 502         struct shm_info shm_info;
 503         int error;
 504
 505         SYSVSHM_ASSERT_LOCKED();
 506
 507         rpr = shm_find_prison(td->td_ucred);
 508         if (rpr == NULL)
 509                 return (ENOSYS);
 510
 511         AUDIT_ARG_SVIPC_ID(shmid);
 512         AUDIT_ARG_SVIPC_CMD(cmd);
 513
 514         switch (cmd) {
 515         /*
 516          * It is possible that kern_shmctl is being called from the Linux ABI
 517          * layer, in which case, we will need to implement IPC_INFO.  It should
 518          * be noted that other shmctl calls will be funneled through here for
 519          * Linix binaries as well.
 520          *
 521          * NB: The Linux ABI layer will convert this data to structure(s) more
 522          * consistent with the Linux ABI.
 523          */
 524         case IPC_INFO:
 525                 memcpy(buf, &shminfo, sizeof(shminfo));
 526                 if (bufsz)
 527                         *bufsz = sizeof(shminfo);
 528                 td->td_retval[0] = shmalloced;
 529                 return (0);
 530         case SHM_INFO: {
 531                 shm_info.used_ids = shm_nused;
 532                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 533                 shm_info.shm_tot = 0;   /*XXX where to get from ? */
 534                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 535                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 536                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 537                 memcpy(buf, &shm_info, sizeof(shm_info));
 538                 if (bufsz != NULL)
 539                         *bufsz = sizeof(shm_info);
 540                 td->td_retval[0] = shmalloced;
 541                 return (0);
 542         }
 543         }
 544         shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
 545         if (shmseg == NULL)
 546                 return (EINVAL);
 547 #ifdef MAC
 548         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
 549         if (error != 0)
 550                 return (error);
 551 #endif
 552         switch (cmd) {
 553         case SHM_STAT:
 554         case IPC_STAT:
 555                 shmidp = (struct shmid_ds *)buf;
 556                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 557                 if (error != 0)
 558                         return (error);
 559                 memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
 560                 if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
 561                         shmidp->shm_perm.key = IPC_PRIVATE;
 562                 if (bufsz != NULL)
 563                         *bufsz = sizeof(struct shmid_ds);
 564                 if (cmd == SHM_STAT) {
 565                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
 566                             shmseg->u.shm_perm);
 567                 }
 568                 break;
 569         case IPC_SET:
 570                 shmidp = (struct shmid_ds *)buf;
 571                 AUDIT_ARG_SVIPC_PERM(&shmidp->shm_perm);
 572                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 573                 if (error != 0)
 574                         return (error);
 575                 shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
 576                 shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
 577                 shmseg->u.shm_perm.mode =
 578                     (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
 579                     (shmidp->shm_perm.mode & ACCESSPERMS);
 580                 shmseg->u.shm_ctime = time_second;
 581                 break;
 582         case IPC_RMID:
 583                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 584                 if (error != 0)
 585                         return (error);
 586                 shm_remove(shmseg, IPCID_TO_IX(shmid));
 587                 break;
 588 #if 0
 589         case SHM_LOCK:
 590         case SHM_UNLOCK:
 591 #endif
 592         default:
 593                 error = EINVAL;
 594                 break;
 595         }
 596         return (error);
 597 }
 598
 599 int
 600 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
 601 {
 602         int error;
 603
 604         SYSVSHM_LOCK();
 605         error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
 606         SYSVSHM_UNLOCK();
 607         return (error);
 608 }
 609
 610 #ifndef _SYS_SYSPROTO_H_
 611 struct shmctl_args {
 612         int shmid;
 613         int cmd;
 614         struct shmid_ds *buf;
 615 };
 616 #endif
 617 int
 618 sys_shmctl(struct thread *td, struct shmctl_args *uap)
 619 {
 620         int error;
 621         struct shmid_ds buf;
 622         size_t bufsz;
 623
 624         /*
 625          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
 626          * Linux binaries.  If we see the call come through the FreeBSD ABI,
 627          * return an error back to the user since we do not to support this.
 628          */
 629         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
 630             uap->cmd == SHM_STAT)
 631                 return (EINVAL);
 632
 633         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
 634         if (uap->cmd == IPC_SET) {
 635                 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
 636                         goto done;
 637         }
 638
 639         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
 640         if (error)
 641                 goto done;
 642
 643         /* Cases in which we need to copyout */
 644         switch (uap->cmd) {
 645         case IPC_STAT:
 646                 error = copyout(&buf, uap->buf, bufsz);
 647                 break;
 648         }
 649
 650 done:
 651         if (error) {
 652                 /* Invalidate the return value */
 653                 td->td_retval[0] = -1;
 654         }
 655         return (error);
 656 }
 657
 658 static int
 659 shmget_existing(struct thread *td, size_t size, int shmflg, int mode,
 660     int segnum)
 661 {
 662         struct shmid_kernel *shmseg;
 663 #ifdef MAC
 664         int error;
 665 #endif
 666
 667         SYSVSHM_ASSERT_LOCKED();
 668         KASSERT(segnum >= 0 && segnum < shmalloced,
 669             ("segnum %d shmalloced %d", segnum, shmalloced));
 670         shmseg = &shmsegs[segnum];
 671         if ((shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 672                 return (EEXIST);
 673 #ifdef MAC
 674         error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, shmflg);
 675         if (error != 0)
 676                 return (error);
 677 #endif
 678         if (size != 0 && size > shmseg->u.shm_segsz)
 679                 return (EINVAL);
 680         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 681         return (0);
 682 }
 683
 684 static int
 685 shmget_allocate_segment(struct thread *td, key_t key, size_t size, int mode)
 686 {
 687         struct ucred *cred = td->td_ucred;
 688         struct shmid_kernel *shmseg;
 689         vm_object_t shm_object;
 690         int i, segnum;
 691
 692         SYSVSHM_ASSERT_LOCKED();
 693
 694         if (size < shminfo.shmmin || size > shminfo.shmmax)
 695                 return (EINVAL);
 696         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 697                 return (ENOSPC);
 698         size = round_page(size);
 699         if (shm_committed + btoc(size) > shminfo.shmall)
 700                 return (ENOMEM);
 701         if (shm_last_free < 0) {
 702                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 703                 for (i = 0; i < shmalloced; i++)
 704                         if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
 705                                 break;
 706                 if (i == shmalloced)
 707                         return (ENOSPC);
 708                 segnum = i;
 709         } else  {
 710                 segnum = shm_last_free;
 711                 shm_last_free = -1;
 712         }
 713         KASSERT(segnum >= 0 && segnum < shmalloced,
 714             ("segnum %d shmalloced %d", segnum, shmalloced));
 715         shmseg = &shmsegs[segnum];
 716 #ifdef RACCT
 717         if (racct_enable) {
 718                 PROC_LOCK(td->td_proc);
 719                 if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
 720                         PROC_UNLOCK(td->td_proc);
 721                         return (ENOSPC);
 722                 }
 723                 if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
 724                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 725                         PROC_UNLOCK(td->td_proc);
 726                         return (ENOMEM);
 727                 }
 728                 PROC_UNLOCK(td->td_proc);
 729         }
 730 #endif
 731
 732         /*
 733          * We make sure that we have allocated a pager before we need
 734          * to.
 735          */
 736         shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
 737             0, size, VM_PROT_DEFAULT, 0, cred);
 738         if (shm_object == NULL) {
 739 #ifdef RACCT
 740                 if (racct_enable) {
 741                         PROC_LOCK(td->td_proc);
 742                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 743                         racct_sub(td->td_proc, RACCT_SHMSIZE, size);
 744                         PROC_UNLOCK(td->td_proc);
 745                 }
 746 #endif
 747                 return (ENOMEM);
 748         }
 749
 750         shmseg->object = shm_object;
 751         shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
 752         shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
 753         shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 754         shmseg->u.shm_perm.key = key;
 755         shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
 756         shmseg->cred = crhold(cred);
 757         shmseg->u.shm_segsz = size;
 758         shmseg->u.shm_cpid = td->td_proc->p_pid;
 759         shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
 760         shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
 761 #ifdef MAC
 762         mac_sysvshm_create(cred, shmseg);
 763 #endif
 764         shmseg->u.shm_ctime = time_second;
 765         shm_committed += btoc(size);
 766         shm_nused++;
 767         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 768
 769         return (0);
 770 }
 771
 772 #ifndef _SYS_SYSPROTO_H_
 773 struct shmget_args {
 774         key_t key;
 775         size_t size;
 776         int shmflg;
 777 };
 778 #endif
 779 int
 780 sys_shmget(struct thread *td, struct shmget_args *uap)
 781 {
 782         int segnum, mode;
 783         int error;
 784
 785         if (shm_find_prison(td->td_ucred) == NULL)
 786                 return (ENOSYS);
 787         mode = uap->shmflg & ACCESSPERMS;
 788         SYSVSHM_LOCK();
 789         if (uap->key == IPC_PRIVATE) {
 790                 error = shmget_allocate_segment(td, uap->key, uap->size, mode);
 791         } else {
 792                 segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
 793                     uap->key);
 794                 if (segnum >= 0)
 795                         error = shmget_existing(td, uap->size, uap->shmflg,
 796                             mode, segnum);
 797                 else if ((uap->shmflg & IPC_CREAT) == 0)
 798                         error = ENOENT;
 799                 else
 800                         error = shmget_allocate_segment(td, uap->key,
 801                             uap->size, mode);
 802         }
 803         SYSVSHM_UNLOCK();
 804         return (error);
 805 }
 806
 807 #ifdef SYSVSHM
 808 void
 809 shmfork(struct proc *p1, struct proc *p2)
 810 #else
 811 static void
 812 shmfork_myhook(struct proc *p1, struct proc *p2)
 813 #endif
 814 {
 815         struct shmmap_state *shmmap_s;
 816         size_t size;
 817         int i;
 818
 819         SYSVSHM_LOCK();
 820         size = shminfo.shmseg * sizeof(struct shmmap_state);
 821         shmmap_s = malloc(size, M_SHM, M_WAITOK);
 822         bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
 823         p2->p_vmspace->vm_shm = shmmap_s;
 824         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 825                 if (shmmap_s->shmid != -1) {
 826                         KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
 827                             IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
 828                             ("segnum %d shmalloced %d",
 829                             IPCID_TO_IX(shmmap_s->shmid), shmalloced));
 830                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
 831                 }
 832         }
 833         SYSVSHM_UNLOCK();
 834 }
 835
 836 #ifdef SYSVSHM
 837 void
 838 shmexit(struct vmspace *vm)
 839 #else
 840 static void
 841 shmexit_myhook(struct vmspace *vm)
 842 #endif
 843 {
 844         struct shmmap_state *base, *shm;
 845         int i;
 846
 847         base = vm->vm_shm;
 848         if (base != NULL) {
 849                 vm->vm_shm = NULL;
 850                 SYSVSHM_LOCK();
 851                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 852                         if (shm->shmid != -1)
 853                                 shm_delete_mapping(vm, shm);
 854                 }
 855                 SYSVSHM_UNLOCK();
 856                 free(base, M_SHM);
 857         }
 858 }
 859
 860 static void
 861 shmrealloc(void)
 862 {
 863         struct shmid_kernel *newsegs;
 864         int i;
 865
 866         SYSVSHM_ASSERT_LOCKED();
 867
 868         if (shmalloced >= shminfo.shmmni)
 869                 return;
 870
 871         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM,
 872             M_WAITOK | M_ZERO);
 873         for (i = 0; i < shmalloced; i++)
 874                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 875         for (; i < shminfo.shmmni; i++) {
 876                 newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 877                 newsegs[i].u.shm_perm.seq = 0;
 878 #ifdef MAC
 879                 mac_sysvshm_init(&newsegs[i]);
 880 #endif
 881         }
 882         free(shmsegs, M_SHM);
 883         shmsegs = newsegs;
 884         shmalloced = shminfo.shmmni;
 885 }
 886
 887 static struct syscall_helper_data shm_syscalls[] = {
 888         SYSCALL_INIT_HELPER(shmat),
 889         SYSCALL_INIT_HELPER(shmctl),
 890         SYSCALL_INIT_HELPER(shmdt),
 891         SYSCALL_INIT_HELPER(shmget),
 892 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 893     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 894         SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
 895 #endif
 896 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 897         SYSCALL_INIT_HELPER(shmsys),
 898 #endif
 899         SYSCALL_INIT_LAST
 900 };
 901
 902 #ifdef COMPAT_FREEBSD32
 903 #include <compat/freebsd32/freebsd32.h>
 904 #include <compat/freebsd32/freebsd32_ipc.h>
 905 #include <compat/freebsd32/freebsd32_proto.h>
 906 #include <compat/freebsd32/freebsd32_signal.h>
 907 #include <compat/freebsd32/freebsd32_syscall.h>
 908 #include <compat/freebsd32/freebsd32_util.h>
 909
 910 static struct syscall_helper_data shm32_syscalls[] = {
 911         SYSCALL32_INIT_HELPER_COMPAT(shmat),
 912         SYSCALL32_INIT_HELPER_COMPAT(shmdt),
 913         SYSCALL32_INIT_HELPER_COMPAT(shmget),
 914         SYSCALL32_INIT_HELPER(freebsd32_shmsys),
 915         SYSCALL32_INIT_HELPER(freebsd32_shmctl),
 916 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 917     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 918         SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
 919 #endif
 920         SYSCALL_INIT_LAST
 921 };
 922 #endif
 923
 924 static int
 925 shminit(void)
 926 {
 927         struct prison *pr;
 928         void **rsv;
 929         int i, error;
 930         osd_method_t methods[PR_MAXMETHOD] = {
 931             [PR_METHOD_CHECK] =         shm_prison_check,
 932             [PR_METHOD_SET] =           shm_prison_set,
 933             [PR_METHOD_GET] =           shm_prison_get,
 934             [PR_METHOD_REMOVE] =        shm_prison_remove,
 935         };
 936
 937 #ifndef BURN_BRIDGES
 938         if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
 939                 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
 940 #endif
 941         if (shminfo.shmmax == SHMMAX) {
 942                 /* Initialize shmmax dealing with possible overflow. */
 943                 for (i = PAGE_SIZE; i != 0; i--) {
 944                         shminfo.shmmax = shminfo.shmall * i;
 945                         if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
 946                                 break;
 947                 }
 948         }
 949         shmalloced = shminfo.shmmni;
 950         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM,
 951             M_WAITOK|M_ZERO);
 952         for (i = 0; i < shmalloced; i++) {
 953                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 954                 shmsegs[i].u.shm_perm.seq = 0;
 955 #ifdef MAC
 956                 mac_sysvshm_init(&shmsegs[i]);
 957 #endif
 958         }
 959         shm_last_free = 0;
 960         shm_nused = 0;
 961         shm_committed = 0;
 962         sx_init(&sysvshmsx, "sysvshmsx");
 963 #ifndef SYSVSHM
 964         shmexit_hook = &shmexit_myhook;
 965         shmfork_hook = &shmfork_myhook;
 966 #endif
 967
 968         /* Set current prisons according to their allow.sysvipc. */
 969         shm_prison_slot = osd_jail_register(NULL, methods);
 970         rsv = osd_reserve(shm_prison_slot);
 971         prison_lock(&prison0);
 972         (void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
 973         prison_unlock(&prison0);
 974         rsv = NULL;
 975         sx_slock(&allprison_lock);
 976         TAILQ_FOREACH(pr, &allprison, pr_list) {
 977                 if (rsv == NULL)
 978                         rsv = osd_reserve(shm_prison_slot);
 979                 prison_lock(pr);
 980                 if (pr->pr_allow & PR_ALLOW_SYSVIPC) {
 981                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
 982                             &prison0);
 983                         rsv = NULL;
 984                 }
 985                 prison_unlock(pr);
 986         }
 987         if (rsv != NULL)
 988                 osd_free_reserved(rsv);
 989         sx_sunlock(&allprison_lock);
 990
 991         error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
 992         if (error != 0)
 993                 return (error);
 994 #ifdef COMPAT_FREEBSD32
 995         error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
 996         if (error != 0)
 997                 return (error);
 998 #endif
 999         return (0);
1000 }
1001
1002 static int
1003 shmunload(void)
1004 {
1005         int i;
1006
1007         if (shm_nused > 0)
1008                 return (EBUSY);
1009
1010 #ifdef COMPAT_FREEBSD32
1011         syscall32_helper_unregister(shm32_syscalls);
1012 #endif
1013         syscall_helper_unregister(shm_syscalls);
1014         if (shm_prison_slot != 0)
1015                 osd_jail_deregister(shm_prison_slot);
1016
1017         for (i = 0; i < shmalloced; i++) {
1018 #ifdef MAC
1019                 mac_sysvshm_destroy(&shmsegs[i]);
1020 #endif
1021                 /*
1022                  * Objects might be still mapped into the processes
1023                  * address spaces.  Actual free would happen on the
1024                  * last mapping destruction.
1025                  */
1026                 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1027                         vm_object_deallocate(shmsegs[i].object);
1028         }
1029         free(shmsegs, M_SHM);
1030 #ifndef SYSVSHM
1031         shmexit_hook = NULL;
1032         shmfork_hook = NULL;
1033 #endif
1034         sx_destroy(&sysvshmsx);
1035         return (0);
1036 }
1037
1038 static int
1039 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1040 {
1041         struct shmid_kernel tshmseg;
1042 #ifdef COMPAT_FREEBSD32
1043         struct shmid_kernel32 tshmseg32;
1044 #endif
1045         struct prison *pr, *rpr;
1046         void *outaddr;
1047         size_t outsize;
1048         int error, i;
1049
1050         SYSVSHM_LOCK();
1051         pr = req->td->td_ucred->cr_prison;
1052         rpr = shm_find_prison(req->td->td_ucred);
1053         error = 0;
1054         for (i = 0; i < shmalloced; i++) {
1055                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1056                     rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1057                         bzero(&tshmseg, sizeof(tshmseg));
1058                         tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1059                 } else {
1060                         tshmseg = shmsegs[i];
1061                         if (tshmseg.cred->cr_prison != pr)
1062                                 tshmseg.u.shm_perm.key = IPC_PRIVATE;
1063                 }
1064 #ifdef COMPAT_FREEBSD32
1065                 if (SV_CURPROC_FLAG(SV_ILP32)) {
1066                         bzero(&tshmseg32, sizeof(tshmseg32));
1067                         freebsd32_ipcperm_out(&tshmseg.u.shm_perm,
1068                             &tshmseg32.u.shm_perm);
1069                         CP(tshmseg, tshmseg32, u.shm_segsz);
1070                         CP(tshmseg, tshmseg32, u.shm_lpid);
1071                         CP(tshmseg, tshmseg32, u.shm_cpid);
1072                         CP(tshmseg, tshmseg32, u.shm_nattch);
1073                         CP(tshmseg, tshmseg32, u.shm_atime);
1074                         CP(tshmseg, tshmseg32, u.shm_dtime);
1075                         CP(tshmseg, tshmseg32, u.shm_ctime);
1076                         /* Don't copy object, label, or cred */
1077                         outaddr = &tshmseg32;
1078                         outsize = sizeof(tshmseg32);
1079                 } else
1080 #endif
1081                 {
1082                         tshmseg.object = NULL;
1083                         tshmseg.label = NULL;
1084                         tshmseg.cred = NULL;
1085                         outaddr = &tshmseg;
1086                         outsize = sizeof(tshmseg);
1087                 }
1088                 error = SYSCTL_OUT(req, outaddr, outsize);
1089                 if (error != 0)
1090                         break;
1091         }
1092         SYSVSHM_UNLOCK();
1093         return (error);
1094 }
1095
1096 static int
1097 shm_prison_check(void *obj, void *data)
1098 {
1099         struct prison *pr = obj;
1100         struct prison *prpr;
1101         struct vfsoptlist *opts = data;
1102         int error, jsys;
1103
1104         /*
1105          * sysvshm is a jailsys integer.
1106          * It must be "disable" if the parent jail is disabled.
1107          */
1108         error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1109         if (error != ENOENT) {
1110                 if (error != 0)
1111                         return (error);
1112                 switch (jsys) {
1113                 case JAIL_SYS_DISABLE:
1114                         break;
1115                 case JAIL_SYS_NEW:
1116                 case JAIL_SYS_INHERIT:
1117                         prison_lock(pr->pr_parent);
1118                         prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1119                         prison_unlock(pr->pr_parent);
1120                         if (prpr == NULL)
1121                                 return (EPERM);
1122                         break;
1123                 default:
1124                         return (EINVAL);
1125                 }
1126         }
1127
1128         return (0);
1129 }
1130
1131 static int
1132 shm_prison_set(void *obj, void *data)
1133 {
1134         struct prison *pr = obj;
1135         struct prison *tpr, *orpr, *nrpr, *trpr;
1136         struct vfsoptlist *opts = data;
1137         void *rsv;
1138         int jsys, descend;
1139
1140         /*
1141          * sysvshm controls which jail is the root of the associated segments
1142          * (this jail or same as the parent), or if the feature is available
1143          * at all.
1144          */
1145         if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1146                 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1147                     ? JAIL_SYS_INHERIT
1148                     : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1149                     ? JAIL_SYS_DISABLE
1150                     : -1;
1151         if (jsys == JAIL_SYS_DISABLE) {
1152                 prison_lock(pr);
1153                 orpr = osd_jail_get(pr, shm_prison_slot);
1154                 if (orpr != NULL)
1155                         osd_jail_del(pr, shm_prison_slot);
1156                 prison_unlock(pr);
1157                 if (orpr != NULL) {
1158                         if (orpr == pr)
1159                                 shm_prison_cleanup(pr);
1160                         /* Disable all child jails as well. */
1161                         FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1162                                 prison_lock(tpr);
1163                                 trpr = osd_jail_get(tpr, shm_prison_slot);
1164                                 if (trpr != NULL) {
1165                                         osd_jail_del(tpr, shm_prison_slot);
1166                                         prison_unlock(tpr);
1167                                         if (trpr == tpr)
1168                                                 shm_prison_cleanup(tpr);
1169                                 } else {
1170                                         prison_unlock(tpr);
1171                                         descend = 0;
1172                                 }
1173                         }
1174                 }
1175         } else if (jsys != -1) {
1176                 if (jsys == JAIL_SYS_NEW)
1177                         nrpr = pr;
1178                 else {
1179                         prison_lock(pr->pr_parent);
1180                         nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1181                         prison_unlock(pr->pr_parent);
1182                 }
1183                 rsv = osd_reserve(shm_prison_slot);
1184                 prison_lock(pr);
1185                 orpr = osd_jail_get(pr, shm_prison_slot);
1186                 if (orpr != nrpr)
1187                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1188                             nrpr);
1189                 else
1190                         osd_free_reserved(rsv);
1191                 prison_unlock(pr);
1192                 if (orpr != nrpr) {
1193                         if (orpr == pr)
1194                                 shm_prison_cleanup(pr);
1195                         if (orpr != NULL) {
1196                                 /* Change child jails matching the old root, */
1197                                 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1198                                         prison_lock(tpr);
1199                                         trpr = osd_jail_get(tpr,
1200                                             shm_prison_slot);
1201                                         if (trpr == orpr) {
1202                                                 (void)osd_jail_set(tpr,
1203                                                     shm_prison_slot, nrpr);
1204                                                 prison_unlock(tpr);
1205                                                 if (trpr == tpr)
1206                                                         shm_prison_cleanup(tpr);
1207                                         } else {
1208                                                 prison_unlock(tpr);
1209                                                 descend = 0;
1210                                         }
1211                                 }
1212                         }
1213                 }
1214         }
1215
1216         return (0);
1217 }
1218
1219 static int
1220 shm_prison_get(void *obj, void *data)
1221 {
1222         struct prison *pr = obj;
1223         struct prison *rpr;
1224         struct vfsoptlist *opts = data;
1225         int error, jsys;
1226
1227         /* Set sysvshm based on the jail's root prison. */
1228         prison_lock(pr);
1229         rpr = osd_jail_get(pr, shm_prison_slot);
1230         prison_unlock(pr);
1231         jsys = rpr == NULL ? JAIL_SYS_DISABLE
1232             : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1233         error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1234         if (error == ENOENT)
1235                 error = 0;
1236         return (error);
1237 }
1238
1239 static int
1240 shm_prison_remove(void *obj, void *data __unused)
1241 {
1242         struct prison *pr = obj;
1243         struct prison *rpr;
1244
1245         SYSVSHM_LOCK();
1246         prison_lock(pr);
1247         rpr = osd_jail_get(pr, shm_prison_slot);
1248         prison_unlock(pr);
1249         if (rpr == pr)
1250                 shm_prison_cleanup(pr);
1251         SYSVSHM_UNLOCK();
1252         return (0);
1253 }
1254
1255 static void
1256 shm_prison_cleanup(struct prison *pr)
1257 {
1258         struct shmid_kernel *shmseg;
1259         int i;
1260
1261         /* Remove any segments that belong to this jail. */
1262         for (i = 0; i < shmalloced; i++) {
1263                 shmseg = &shmsegs[i];
1264                 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1265                     shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1266                         shm_remove(shmseg, i);
1267                 }
1268         }
1269 }
1270
1271 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1272
1273 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1274 struct oshmid_ds {
1275         struct  ipc_perm_old shm_perm;  /* operation perms */
1276         int     shm_segsz;              /* size of segment (bytes) */
1277         u_short shm_cpid;               /* pid, creator */
1278         u_short shm_lpid;               /* pid, last operation */
1279         short   shm_nattch;             /* no. of current attaches */
1280         time_t  shm_atime;              /* last attach time */
1281         time_t  shm_dtime;              /* last detach time */
1282         time_t  shm_ctime;              /* last change time */
1283         void    *shm_handle;            /* internal handle for shm segment */
1284 };
1285
1286 struct oshmctl_args {
1287         int shmid;
1288         int cmd;
1289         struct oshmid_ds *ubuf;
1290 };
1291
1292 static int
1293 oshmctl(struct thread *td, struct oshmctl_args *uap)
1294 {
1295 #ifdef COMPAT_43
1296         int error = 0;
1297         struct prison *rpr;
1298         struct shmid_kernel *shmseg;
1299         struct oshmid_ds outbuf;
1300
1301         rpr = shm_find_prison(td->td_ucred);
1302         if (rpr == NULL)
1303                 return (ENOSYS);
1304         if (uap->cmd != IPC_STAT) {
1305                 return (freebsd7_shmctl(td,
1306                     (struct freebsd7_shmctl_args *)uap));
1307         }
1308         SYSVSHM_LOCK();
1309         shmseg = shm_find_segment(rpr, uap->shmid, true);
1310         if (shmseg == NULL) {
1311                 SYSVSHM_UNLOCK();
1312                 return (EINVAL);
1313         }
1314         error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1315         if (error != 0) {
1316                 SYSVSHM_UNLOCK();
1317                 return (error);
1318         }
1319 #ifdef MAC
1320         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1321         if (error != 0) {
1322                 SYSVSHM_UNLOCK();
1323                 return (error);
1324         }
1325 #endif
1326         ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1327         outbuf.shm_segsz = shmseg->u.shm_segsz;
1328         outbuf.shm_cpid = shmseg->u.shm_cpid;
1329         outbuf.shm_lpid = shmseg->u.shm_lpid;
1330         outbuf.shm_nattch = shmseg->u.shm_nattch;
1331         outbuf.shm_atime = shmseg->u.shm_atime;
1332         outbuf.shm_dtime = shmseg->u.shm_dtime;
1333         outbuf.shm_ctime = shmseg->u.shm_ctime;
1334         outbuf.shm_handle = shmseg->object;
1335         SYSVSHM_UNLOCK();
1336         return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1337 #else
1338         return (EINVAL);
1339 #endif
1340 }
1341
1342 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1343 static sy_call_t *shmcalls[] = {
1344         (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1345         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1346         (sy_call_t *)freebsd7_shmctl
1347 };
1348
1349 #ifndef _SYS_SYSPROTO_H_
1350 /* XXX actually varargs. */
1351 struct shmsys_args {
1352         int     which;
1353         int     a2;
1354         int     a3;
1355         int     a4;
1356 };
1357 #endif
1358 int
1359 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1360 {
1361
1362         AUDIT_ARG_SVIPC_WHICH(uap->which);
1363         if (uap->which < 0 || uap->which >= nitems(shmcalls))
1364                 return (EINVAL);
1365         return ((*shmcalls[uap->which])(td, &uap->a2));
1366 }
1367
1368 #endif  /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1369
1370 #ifdef COMPAT_FREEBSD32
1371
1372 int
1373 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1374 {
1375
1376 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1377     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1378         AUDIT_ARG_SVIPC_WHICH(uap->which);
1379         switch (uap->which) {
1380         case 0: {       /* shmat */
1381                 struct shmat_args ap;
1382
1383                 ap.shmid = uap->a2;
1384                 ap.shmaddr = PTRIN(uap->a3);
1385                 ap.shmflg = uap->a4;
1386                 return (sysent[SYS_shmat].sy_call(td, &ap));
1387         }
1388         case 2: {       /* shmdt */
1389                 struct shmdt_args ap;
1390
1391                 ap.shmaddr = PTRIN(uap->a2);
1392                 return (sysent[SYS_shmdt].sy_call(td, &ap));
1393         }
1394         case 3: {       /* shmget */
1395                 struct shmget_args ap;
1396
1397                 ap.key = uap->a2;
1398                 ap.size = uap->a3;
1399                 ap.shmflg = uap->a4;
1400                 return (sysent[SYS_shmget].sy_call(td, &ap));
1401         }
1402         case 4: {       /* shmctl */
1403                 struct freebsd7_freebsd32_shmctl_args ap;
1404
1405                 ap.shmid = uap->a2;
1406                 ap.cmd = uap->a3;
1407                 ap.buf = PTRIN(uap->a4);
1408                 return (freebsd7_freebsd32_shmctl(td, &ap));
1409         }
1410         case 1:         /* oshmctl */
1411         default:
1412                 return (EINVAL);
1413         }
1414 #else
1415         return (nosys(td, NULL));
1416 #endif
1417 }
1418
1419 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1420     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1421 int
1422 freebsd7_freebsd32_shmctl(struct thread *td,
1423     struct freebsd7_freebsd32_shmctl_args *uap)
1424 {
1425         int error;
1426         union {
1427                 struct shmid_ds shmid_ds;
1428                 struct shm_info shm_info;
1429                 struct shminfo shminfo;
1430         } u;
1431         union {
1432                 struct shmid_ds_old32 shmid_ds32;
1433                 struct shm_info32 shm_info32;
1434                 struct shminfo32 shminfo32;
1435         } u32;
1436         size_t sz;
1437
1438         if (uap->cmd == IPC_SET) {
1439                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1440                     sizeof(u32.shmid_ds32))))
1441                         goto done;
1442                 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1443                     &u.shmid_ds.shm_perm);
1444                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1445                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1446                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1447                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1448                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1449                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1450                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1451         }
1452
1453         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1454         if (error)
1455                 goto done;
1456
1457         /* Cases in which we need to copyout */
1458         switch (uap->cmd) {
1459         case IPC_INFO:
1460                 CP(u.shminfo, u32.shminfo32, shmmax);
1461                 CP(u.shminfo, u32.shminfo32, shmmin);
1462                 CP(u.shminfo, u32.shminfo32, shmmni);
1463                 CP(u.shminfo, u32.shminfo32, shmseg);
1464                 CP(u.shminfo, u32.shminfo32, shmall);
1465                 error = copyout(&u32.shminfo32, uap->buf,
1466                     sizeof(u32.shminfo32));
1467                 break;
1468         case SHM_INFO:
1469                 CP(u.shm_info, u32.shm_info32, used_ids);
1470                 CP(u.shm_info, u32.shm_info32, shm_rss);
1471                 CP(u.shm_info, u32.shm_info32, shm_tot);
1472                 CP(u.shm_info, u32.shm_info32, shm_swp);
1473                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1474                 CP(u.shm_info, u32.shm_info32, swap_successes);
1475                 error = copyout(&u32.shm_info32, uap->buf,
1476                     sizeof(u32.shm_info32));
1477                 break;
1478         case SHM_STAT:
1479         case IPC_STAT:
1480                 memset(&u32.shmid_ds32, 0, sizeof(u32.shmid_ds32));
1481                 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1482                     &u32.shmid_ds32.shm_perm);
1483                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1484                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1485                 else
1486                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1487                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1488                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1489                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1490                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1491                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1492                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1493                 u32.shmid_ds32.shm_internal = 0;
1494                 error = copyout(&u32.shmid_ds32, uap->buf,
1495                     sizeof(u32.shmid_ds32));
1496                 break;
1497         }
1498
1499 done:
1500         if (error) {
1501                 /* Invalidate the return value */
1502                 td->td_retval[0] = -1;
1503         }
1504         return (error);
1505 }
1506 #endif
1507
1508 int
1509 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1510 {
1511         int error;
1512         union {
1513                 struct shmid_ds shmid_ds;
1514                 struct shm_info shm_info;
1515                 struct shminfo shminfo;
1516         } u;
1517         union {
1518                 struct shmid_ds32 shmid_ds32;
1519                 struct shm_info32 shm_info32;
1520                 struct shminfo32 shminfo32;
1521         } u32;
1522         size_t sz;
1523
1524         if (uap->cmd == IPC_SET) {
1525                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1526                     sizeof(u32.shmid_ds32))))
1527                         goto done;
1528                 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1529                     &u.shmid_ds.shm_perm);
1530                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1531                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1532                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1533                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1534                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1535                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1536                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1537         }
1538
1539         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1540         if (error)
1541                 goto done;
1542
1543         /* Cases in which we need to copyout */
1544         switch (uap->cmd) {
1545         case IPC_INFO:
1546                 CP(u.shminfo, u32.shminfo32, shmmax);
1547                 CP(u.shminfo, u32.shminfo32, shmmin);
1548                 CP(u.shminfo, u32.shminfo32, shmmni);
1549                 CP(u.shminfo, u32.shminfo32, shmseg);
1550                 CP(u.shminfo, u32.shminfo32, shmall);
1551                 error = copyout(&u32.shminfo32, uap->buf,
1552                     sizeof(u32.shminfo32));
1553                 break;
1554         case SHM_INFO:
1555                 CP(u.shm_info, u32.shm_info32, used_ids);
1556                 CP(u.shm_info, u32.shm_info32, shm_rss);
1557                 CP(u.shm_info, u32.shm_info32, shm_tot);
1558                 CP(u.shm_info, u32.shm_info32, shm_swp);
1559                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1560                 CP(u.shm_info, u32.shm_info32, swap_successes);
1561                 error = copyout(&u32.shm_info32, uap->buf,
1562                     sizeof(u32.shm_info32));
1563                 break;
1564         case SHM_STAT:
1565         case IPC_STAT:
1566                 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1567                     &u32.shmid_ds32.shm_perm);
1568                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1569                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1570                 else
1571                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1572                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1573                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1574                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1575                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1576                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1577                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1578                 error = copyout(&u32.shmid_ds32, uap->buf,
1579                     sizeof(u32.shmid_ds32));
1580                 break;
1581         }
1582
1583 done:
1584         if (error) {
1585                 /* Invalidate the return value */
1586                 td->td_retval[0] = -1;
1587         }
1588         return (error);
1589 }
1590 #endif
1591
1592 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1593     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1594
1595 #ifndef _SYS_SYSPROTO_H_
1596 struct freebsd7_shmctl_args {
1597         int shmid;
1598         int cmd;
1599         struct shmid_ds_old *buf;
1600 };
1601 #endif
1602 int
1603 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1604 {
1605         int error;
1606         struct shmid_ds_old old;
1607         struct shmid_ds buf;
1608         size_t bufsz;
1609
1610         /*
1611          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1612          * Linux binaries.  If we see the call come through the FreeBSD ABI,
1613          * return an error back to the user since we do not to support this.
1614          */
1615         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1616             uap->cmd == SHM_STAT)
1617                 return (EINVAL);
1618
1619         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1620         if (uap->cmd == IPC_SET) {
1621                 if ((error = copyin(uap->buf, &old, sizeof(old))))
1622                         goto done;
1623                 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1624                 CP(old, buf, shm_segsz);
1625                 CP(old, buf, shm_lpid);
1626                 CP(old, buf, shm_cpid);
1627                 CP(old, buf, shm_nattch);
1628                 CP(old, buf, shm_atime);
1629                 CP(old, buf, shm_dtime);
1630                 CP(old, buf, shm_ctime);
1631         }
1632
1633         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1634         if (error)
1635                 goto done;
1636
1637         /* Cases in which we need to copyout */
1638         switch (uap->cmd) {
1639         case IPC_STAT:
1640                 memset(&old, 0, sizeof(old));
1641                 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1642                 if (buf.shm_segsz > INT_MAX)
1643                         old.shm_segsz = INT_MAX;
1644                 else
1645                         CP(buf, old, shm_segsz);
1646                 CP(buf, old, shm_lpid);
1647                 CP(buf, old, shm_cpid);
1648                 if (buf.shm_nattch > SHRT_MAX)
1649                         old.shm_nattch = SHRT_MAX;
1650                 else
1651                         CP(buf, old, shm_nattch);
1652                 CP(buf, old, shm_atime);
1653                 CP(buf, old, shm_dtime);
1654                 CP(buf, old, shm_ctime);
1655                 old.shm_internal = NULL;
1656                 error = copyout(&old, uap->buf, sizeof(old));
1657                 break;
1658         }
1659
1660 done:
1661         if (error) {
1662                 /* Invalidate the return value */
1663                 td->td_retval[0] = -1;
1664         }
1665         return (error);
1666 }
1667
1668 #endif  /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1669            COMPAT_FREEBSD7 */
1670
1671 static int
1672 sysvshm_modload(struct module *module, int cmd, void *arg)
1673 {
1674         int error = 0;
1675
1676         switch (cmd) {
1677         case MOD_LOAD:
1678                 error = shminit();
1679                 if (error != 0)
1680                         shmunload();
1681                 break;
1682         case MOD_UNLOAD:
1683                 error = shmunload();
1684                 break;
1685         case MOD_SHUTDOWN:
1686                 break;
1687         default:
1688                 error = EINVAL;
1689                 break;
1690         }
1691         return (error);
1692 }
1693
1694 static moduledata_t sysvshm_mod = {
1695         "sysvshm",
1696         &sysvshm_modload,
1697         NULL
1698 };
1699
1700 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1701 MODULE_VERSION(sysvshm, 1);