sys/kern/sysv_shm.c

   1 /*      $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $      */
   2 /*-
   3  * SPDX-License-Identifier: BSD-4-Clause AND BSD-2-Clause-FreeBSD
   4  *
   5  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. All advertising materials mentioning features or use of this software
  16  *    must display the following acknowledgement:
  17  *      This product includes software developed by Adam Glass and Charles
  18  *      Hannum.
  19  * 4. The names of the authors may not be used to endorse or promote products
  20  *    derived from this software without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  25  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33 /*-
  34  * Copyright (c) 2003-2005 McAfee, Inc.
  35  * Copyright (c) 2016-2017 Robert N. M. Watson
  36  * All rights reserved.
  37  *
  38  * This software was developed for the FreeBSD Project in part by McAfee
  39  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
  40  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
  41  * program.
  42  *
  43  * Portions of this software were developed by BAE Systems, the University of
  44  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
  45  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
  46  * Computing (TC) research program.
  47  *
  48  * Redistribution and use in source and binary forms, with or without
  49  * modification, are permitted provided that the following conditions
  50  * are met:
  51  * 1. Redistributions of source code must retain the above copyright
  52  *    notice, this list of conditions and the following disclaimer.
  53  * 2. Redistributions in binary form must reproduce the above copyright
  54  *    notice, this list of conditions and the following disclaimer in the
  55  *    documentation and/or other materials provided with the distribution.
  56  *
  57  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  67  * SUCH DAMAGE.
  68  */
  69
  70 #include <sys/cdefs.h>
  71 __FBSDID("$FreeBSD$");
  72
  73 #include "opt_compat.h"
  74 #include "opt_sysvipc.h"
  75
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/kernel.h>
  79 #include <sys/limits.h>
  80 #include <sys/lock.h>
  81 #include <sys/sysctl.h>
  82 #include <sys/shm.h>
  83 #include <sys/proc.h>
  84 #include <sys/malloc.h>
  85 #include <sys/mman.h>
  86 #include <sys/module.h>
  87 #include <sys/mutex.h>
  88 #include <sys/racct.h>
  89 #include <sys/resourcevar.h>
  90 #include <sys/rwlock.h>
  91 #include <sys/stat.h>
  92 #include <sys/syscall.h>
  93 #include <sys/syscallsubr.h>
  94 #include <sys/sysent.h>
  95 #include <sys/sysproto.h>
  96 #include <sys/jail.h>
  97
  98 #include <security/audit/audit.h>
  99 #include <security/mac/mac_framework.h>
 100
 101 #include <vm/vm.h>
 102 #include <vm/vm_param.h>
 103 #include <vm/pmap.h>
 104 #include <vm/vm_object.h>
 105 #include <vm/vm_map.h>
 106 #include <vm/vm_page.h>
 107 #include <vm/vm_pager.h>
 108
 109 FEATURE(sysv_shm, "System V shared memory segments support");
 110
 111 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
 112
 113 static int shmget_allocate_segment(struct thread *td,
 114     struct shmget_args *uap, int mode);
 115 static int shmget_existing(struct thread *td, struct shmget_args *uap,
 116     int mode, int segnum);
 117
 118 #define SHMSEG_FREE             0x0200
 119 #define SHMSEG_REMOVED          0x0400
 120 #define SHMSEG_ALLOCATED        0x0800
 121
 122 static int shm_last_free, shm_nused, shmalloced;
 123 vm_size_t shm_committed;
 124 static struct shmid_kernel *shmsegs;
 125 static unsigned shm_prison_slot;
 126
 127 struct shmmap_state {
 128         vm_offset_t va;
 129         int shmid;
 130 };
 131
 132 static void shm_deallocate_segment(struct shmid_kernel *);
 133 static int shm_find_segment_by_key(struct prison *, key_t);
 134 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
 135 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
 136 static void shmrealloc(void);
 137 static int shminit(void);
 138 static int sysvshm_modload(struct module *, int, void *);
 139 static int shmunload(void);
 140 static void shmexit_myhook(struct vmspace *vm);
 141 static void shmfork_myhook(struct proc *p1, struct proc *p2);
 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       131072  /* 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     "Current number of shared memory segments allocated");
 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 #if defined(AUDIT) || defined(MAC)
 336         struct shmid_kernel *shmsegptr;
 337 #endif
 338 #ifdef MAC
 339         int error;
 340 #endif
 341         int i;
 342
 343         SYSVSHM_ASSERT_LOCKED();
 344         if (shm_find_prison(td->td_ucred) == NULL)
 345                 return (ENOSYS);
 346         shmmap_s = p->p_vmspace->vm_shm;
 347         if (shmmap_s == NULL)
 348                 return (EINVAL);
 349         AUDIT_ARG_SVIPC_ID(shmmap_s->shmid);
 350         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 351                 if (shmmap_s->shmid != -1 &&
 352                     shmmap_s->va == (vm_offset_t)shmaddr) {
 353                         break;
 354                 }
 355         }
 356         if (i == shminfo.shmseg)
 357                 return (EINVAL);
 358 #if (defined(AUDIT) && defined(KDTRACE_HOOKS)) || defined(MAC)
 359         shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
 360 #endif
 361 #ifdef MAC
 362         error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
 363         if (error != 0)
 364                 return (error);
 365 #endif
 366         return (shm_delete_mapping(p->p_vmspace, shmmap_s));
 367 }
 368
 369 #ifndef _SYS_SYSPROTO_H_
 370 struct shmdt_args {
 371         const void *shmaddr;
 372 };
 373 #endif
 374 int
 375 sys_shmdt(struct thread *td, struct shmdt_args *uap)
 376 {
 377         int error;
 378
 379         SYSVSHM_LOCK();
 380         error = kern_shmdt_locked(td, uap->shmaddr);
 381         SYSVSHM_UNLOCK();
 382         return (error);
 383 }
 384
 385 static int
 386 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
 387     int shmflg)
 388 {
 389         struct prison *rpr;
 390         struct proc *p = td->td_proc;
 391         struct shmid_kernel *shmseg;
 392         struct shmmap_state *shmmap_s;
 393         vm_offset_t attach_va;
 394         vm_prot_t prot;
 395         vm_size_t size;
 396         int error, i, rv;
 397
 398         AUDIT_ARG_SVIPC_ID(shmid);
 399         AUDIT_ARG_VALUE(shmflg);
 400
 401         SYSVSHM_ASSERT_LOCKED();
 402         rpr = shm_find_prison(td->td_ucred);
 403         if (rpr == NULL)
 404                 return (ENOSYS);
 405         shmmap_s = p->p_vmspace->vm_shm;
 406         if (shmmap_s == NULL) {
 407                 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
 408                     M_SHM, M_WAITOK);
 409                 for (i = 0; i < shminfo.shmseg; i++)
 410                         shmmap_s[i].shmid = -1;
 411                 KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
 412                 p->p_vmspace->vm_shm = shmmap_s;
 413         }
 414         shmseg = shm_find_segment(rpr, shmid, true);
 415         if (shmseg == NULL)
 416                 return (EINVAL);
 417         error = ipcperm(td, &shmseg->u.shm_perm,
 418             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 419         if (error != 0)
 420                 return (error);
 421 #ifdef MAC
 422         error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
 423         if (error != 0)
 424                 return (error);
 425 #endif
 426         for (i = 0; i < shminfo.shmseg; i++) {
 427                 if (shmmap_s->shmid == -1)
 428                         break;
 429                 shmmap_s++;
 430         }
 431         if (i >= shminfo.shmseg)
 432                 return (EMFILE);
 433         size = round_page(shmseg->u.shm_segsz);
 434         prot = VM_PROT_READ;
 435         if ((shmflg & SHM_RDONLY) == 0)
 436                 prot |= VM_PROT_WRITE;
 437         if (shmaddr != NULL) {
 438                 if ((shmflg & SHM_RND) != 0)
 439                         attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
 440                 else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
 441                         attach_va = (vm_offset_t)shmaddr;
 442                 else
 443                         return (EINVAL);
 444         } else {
 445                 /*
 446                  * This is just a hint to vm_map_find() about where to
 447                  * put it.
 448                  */
 449                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
 450                     lim_max(td, RLIMIT_DATA));
 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, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
 456             prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
 457         if (rv != KERN_SUCCESS) {
 458                 vm_object_deallocate(shmseg->object);
 459                 return (ENOMEM);
 460         }
 461
 462         shmmap_s->va = attach_va;
 463         shmmap_s->shmid = shmid;
 464         shmseg->u.shm_lpid = p->p_pid;
 465         shmseg->u.shm_atime = time_second;
 466         shmseg->u.shm_nattch++;
 467         td->td_retval[0] = attach_va;
 468         return (error);
 469 }
 470
 471 int
 472 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
 473 {
 474         int error;
 475
 476         SYSVSHM_LOCK();
 477         error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
 478         SYSVSHM_UNLOCK();
 479         return (error);
 480 }
 481
 482 #ifndef _SYS_SYSPROTO_H_
 483 struct shmat_args {
 484         int shmid;
 485         const void *shmaddr;
 486         int shmflg;
 487 };
 488 #endif
 489 int
 490 sys_shmat(struct thread *td, struct shmat_args *uap)
 491 {
 492
 493         return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
 494 }
 495
 496 static int
 497 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
 498     size_t *bufsz)
 499 {
 500         struct prison *rpr;
 501         struct shmid_kernel *shmseg;
 502         struct shmid_ds *shmidp;
 503         struct shm_info shm_info;
 504         int error;
 505
 506         SYSVSHM_ASSERT_LOCKED();
 507
 508         rpr = shm_find_prison(td->td_ucred);
 509         if (rpr == NULL)
 510                 return (ENOSYS);
 511
 512         AUDIT_ARG_SVIPC_ID(shmid);
 513         AUDIT_ARG_SVIPC_CMD(cmd);
 514
 515         switch (cmd) {
 516         /*
 517          * It is possible that kern_shmctl is being called from the Linux ABI
 518          * layer, in which case, we will need to implement IPC_INFO.  It should
 519          * be noted that other shmctl calls will be funneled through here for
 520          * Linix binaries as well.
 521          *
 522          * NB: The Linux ABI layer will convert this data to structure(s) more
 523          * consistent with the Linux ABI.
 524          */
 525         case IPC_INFO:
 526                 memcpy(buf, &shminfo, sizeof(shminfo));
 527                 if (bufsz)
 528                         *bufsz = sizeof(shminfo);
 529                 td->td_retval[0] = shmalloced;
 530                 return (0);
 531         case SHM_INFO: {
 532                 shm_info.used_ids = shm_nused;
 533                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 534                 shm_info.shm_tot = 0;   /*XXX where to get from ? */
 535                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 536                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 537                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 538                 memcpy(buf, &shm_info, sizeof(shm_info));
 539                 if (bufsz != NULL)
 540                         *bufsz = sizeof(shm_info);
 541                 td->td_retval[0] = shmalloced;
 542                 return (0);
 543         }
 544         }
 545         shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
 546         if (shmseg == NULL)
 547                 return (EINVAL);
 548 #ifdef MAC
 549         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
 550         if (error != 0)
 551                 return (error);
 552 #endif
 553         switch (cmd) {
 554         case SHM_STAT:
 555         case IPC_STAT:
 556                 shmidp = (struct shmid_ds *)buf;
 557                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 558                 if (error != 0)
 559                         return (error);
 560                 memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
 561                 if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
 562                         shmidp->shm_perm.key = IPC_PRIVATE;
 563                 if (bufsz != NULL)
 564                         *bufsz = sizeof(struct shmid_ds);
 565                 if (cmd == SHM_STAT) {
 566                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
 567                             shmseg->u.shm_perm);
 568                 }
 569                 break;
 570         case IPC_SET:
 571                 shmidp = (struct shmid_ds *)buf;
 572                 AUDIT_ARG_SVIPC_PERM(&shmidp->shm_perm);
 573                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 574                 if (error != 0)
 575                         return (error);
 576                 shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
 577                 shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
 578                 shmseg->u.shm_perm.mode =
 579                     (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
 580                     (shmidp->shm_perm.mode & ACCESSPERMS);
 581                 shmseg->u.shm_ctime = time_second;
 582                 break;
 583         case IPC_RMID:
 584                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 585                 if (error != 0)
 586                         return (error);
 587                 shm_remove(shmseg, IPCID_TO_IX(shmid));
 588                 break;
 589 #if 0
 590         case SHM_LOCK:
 591         case SHM_UNLOCK:
 592 #endif
 593         default:
 594                 error = EINVAL;
 595                 break;
 596         }
 597         return (error);
 598 }
 599
 600 int
 601 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
 602 {
 603         int error;
 604
 605         SYSVSHM_LOCK();
 606         error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
 607         SYSVSHM_UNLOCK();
 608         return (error);
 609 }
 610
 611
 612 #ifndef _SYS_SYSPROTO_H_
 613 struct shmctl_args {
 614         int shmid;
 615         int cmd;
 616         struct shmid_ds *buf;
 617 };
 618 #endif
 619 int
 620 sys_shmctl(struct thread *td, struct shmctl_args *uap)
 621 {
 622         int error;
 623         struct shmid_ds buf;
 624         size_t bufsz;
 625
 626         /*
 627          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
 628          * Linux binaries.  If we see the call come through the FreeBSD ABI,
 629          * return an error back to the user since we do not to support this.
 630          */
 631         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
 632             uap->cmd == SHM_STAT)
 633                 return (EINVAL);
 634
 635         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
 636         if (uap->cmd == IPC_SET) {
 637                 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
 638                         goto done;
 639         }
 640
 641         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
 642         if (error)
 643                 goto done;
 644
 645         /* Cases in which we need to copyout */
 646         switch (uap->cmd) {
 647         case IPC_STAT:
 648                 error = copyout(&buf, uap->buf, bufsz);
 649                 break;
 650         }
 651
 652 done:
 653         if (error) {
 654                 /* Invalidate the return value */
 655                 td->td_retval[0] = -1;
 656         }
 657         return (error);
 658 }
 659
 660
 661 static int
 662 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
 663     int segnum)
 664 {
 665         struct shmid_kernel *shmseg;
 666 #ifdef MAC
 667         int error;
 668 #endif
 669
 670         SYSVSHM_ASSERT_LOCKED();
 671         KASSERT(segnum >= 0 && segnum < shmalloced,
 672             ("segnum %d shmalloced %d", segnum, shmalloced));
 673         shmseg = &shmsegs[segnum];
 674         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 675                 return (EEXIST);
 676 #ifdef MAC
 677         error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
 678         if (error != 0)
 679                 return (error);
 680 #endif
 681         if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
 682                 return (EINVAL);
 683         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 684         return (0);
 685 }
 686
 687 static int
 688 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
 689 {
 690         struct ucred *cred = td->td_ucred;
 691         struct shmid_kernel *shmseg;
 692         vm_object_t shm_object;
 693         int i, segnum;
 694         size_t size;
 695
 696         SYSVSHM_ASSERT_LOCKED();
 697
 698         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 699                 return (EINVAL);
 700         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 701                 return (ENOSPC);
 702         size = round_page(uap->size);
 703         if (shm_committed + btoc(size) > shminfo.shmall)
 704                 return (ENOMEM);
 705         if (shm_last_free < 0) {
 706                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 707                 for (i = 0; i < shmalloced; i++)
 708                         if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
 709                                 break;
 710                 if (i == shmalloced)
 711                         return (ENOSPC);
 712                 segnum = i;
 713         } else  {
 714                 segnum = shm_last_free;
 715                 shm_last_free = -1;
 716         }
 717         KASSERT(segnum >= 0 && segnum < shmalloced,
 718             ("segnum %d shmalloced %d", segnum, shmalloced));
 719         shmseg = &shmsegs[segnum];
 720 #ifdef RACCT
 721         if (racct_enable) {
 722                 PROC_LOCK(td->td_proc);
 723                 if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
 724                         PROC_UNLOCK(td->td_proc);
 725                         return (ENOSPC);
 726                 }
 727                 if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
 728                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 729                         PROC_UNLOCK(td->td_proc);
 730                         return (ENOMEM);
 731                 }
 732                 PROC_UNLOCK(td->td_proc);
 733         }
 734 #endif
 735
 736         /*
 737          * We make sure that we have allocated a pager before we need
 738          * to.
 739          */
 740         shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
 741             0, size, VM_PROT_DEFAULT, 0, cred);
 742         if (shm_object == NULL) {
 743 #ifdef RACCT
 744                 if (racct_enable) {
 745                         PROC_LOCK(td->td_proc);
 746                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 747                         racct_sub(td->td_proc, RACCT_SHMSIZE, size);
 748                         PROC_UNLOCK(td->td_proc);
 749                 }
 750 #endif
 751                 return (ENOMEM);
 752         }
 753         shm_object->pg_color = 0;
 754         VM_OBJECT_WLOCK(shm_object);
 755         vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
 756         vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
 757         VM_OBJECT_WUNLOCK(shm_object);
 758
 759         shmseg->object = shm_object;
 760         shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
 761         shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
 762         shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 763         shmseg->u.shm_perm.key = uap->key;
 764         shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
 765         shmseg->cred = crhold(cred);
 766         shmseg->u.shm_segsz = uap->size;
 767         shmseg->u.shm_cpid = td->td_proc->p_pid;
 768         shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
 769         shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
 770 #ifdef MAC
 771         mac_sysvshm_create(cred, shmseg);
 772 #endif
 773         shmseg->u.shm_ctime = time_second;
 774         shm_committed += btoc(size);
 775         shm_nused++;
 776         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 777
 778         return (0);
 779 }
 780
 781 #ifndef _SYS_SYSPROTO_H_
 782 struct shmget_args {
 783         key_t key;
 784         size_t size;
 785         int shmflg;
 786 };
 787 #endif
 788 int
 789 sys_shmget(struct thread *td, struct shmget_args *uap)
 790 {
 791         int segnum, mode;
 792         int error;
 793
 794         if (shm_find_prison(td->td_ucred) == NULL)
 795                 return (ENOSYS);
 796         mode = uap->shmflg & ACCESSPERMS;
 797         SYSVSHM_LOCK();
 798         if (uap->key == IPC_PRIVATE) {
 799                 error = shmget_allocate_segment(td, uap, mode);
 800         } else {
 801                 segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
 802                     uap->key);
 803                 if (segnum >= 0)
 804                         error = shmget_existing(td, uap, mode, segnum);
 805                 else if ((uap->shmflg & IPC_CREAT) == 0)
 806                         error = ENOENT;
 807                 else
 808                         error = shmget_allocate_segment(td, uap, mode);
 809         }
 810         SYSVSHM_UNLOCK();
 811         return (error);
 812 }
 813
 814 static void
 815 shmfork_myhook(struct proc *p1, struct proc *p2)
 816 {
 817         struct shmmap_state *shmmap_s;
 818         size_t size;
 819         int i;
 820
 821         SYSVSHM_LOCK();
 822         size = shminfo.shmseg * sizeof(struct shmmap_state);
 823         shmmap_s = malloc(size, M_SHM, M_WAITOK);
 824         bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
 825         p2->p_vmspace->vm_shm = shmmap_s;
 826         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 827                 if (shmmap_s->shmid != -1) {
 828                         KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
 829                             IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
 830                             ("segnum %d shmalloced %d",
 831                             IPCID_TO_IX(shmmap_s->shmid), shmalloced));
 832                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
 833                 }
 834         }
 835         SYSVSHM_UNLOCK();
 836 }
 837
 838 static void
 839 shmexit_myhook(struct vmspace *vm)
 840 {
 841         struct shmmap_state *base, *shm;
 842         int i;
 843
 844         base = vm->vm_shm;
 845         if (base != NULL) {
 846                 vm->vm_shm = NULL;
 847                 SYSVSHM_LOCK();
 848                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 849                         if (shm->shmid != -1)
 850                                 shm_delete_mapping(vm, shm);
 851                 }
 852                 SYSVSHM_UNLOCK();
 853                 free(base, M_SHM);
 854         }
 855 }
 856
 857 static void
 858 shmrealloc(void)
 859 {
 860         struct shmid_kernel *newsegs;
 861         int i;
 862
 863         SYSVSHM_ASSERT_LOCKED();
 864
 865         if (shmalloced >= shminfo.shmmni)
 866                 return;
 867
 868         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 869         for (i = 0; i < shmalloced; i++)
 870                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 871         for (; i < shminfo.shmmni; i++) {
 872                 newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 873                 newsegs[i].u.shm_perm.seq = 0;
 874 #ifdef MAC
 875                 mac_sysvshm_init(&newsegs[i]);
 876 #endif
 877         }
 878         free(shmsegs, M_SHM);
 879         shmsegs = newsegs;
 880         shmalloced = shminfo.shmmni;
 881 }
 882
 883 static struct syscall_helper_data shm_syscalls[] = {
 884         SYSCALL_INIT_HELPER(shmat),
 885         SYSCALL_INIT_HELPER(shmctl),
 886         SYSCALL_INIT_HELPER(shmdt),
 887         SYSCALL_INIT_HELPER(shmget),
 888 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 889     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 890         SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
 891 #endif
 892 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 893         SYSCALL_INIT_HELPER(shmsys),
 894 #endif
 895         SYSCALL_INIT_LAST
 896 };
 897
 898 #ifdef COMPAT_FREEBSD32
 899 #include <compat/freebsd32/freebsd32.h>
 900 #include <compat/freebsd32/freebsd32_ipc.h>
 901 #include <compat/freebsd32/freebsd32_proto.h>
 902 #include <compat/freebsd32/freebsd32_signal.h>
 903 #include <compat/freebsd32/freebsd32_syscall.h>
 904 #include <compat/freebsd32/freebsd32_util.h>
 905
 906 static struct syscall_helper_data shm32_syscalls[] = {
 907         SYSCALL32_INIT_HELPER_COMPAT(shmat),
 908         SYSCALL32_INIT_HELPER_COMPAT(shmdt),
 909         SYSCALL32_INIT_HELPER_COMPAT(shmget),
 910         SYSCALL32_INIT_HELPER(freebsd32_shmsys),
 911         SYSCALL32_INIT_HELPER(freebsd32_shmctl),
 912 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 913     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 914         SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
 915 #endif
 916         SYSCALL_INIT_LAST
 917 };
 918 #endif
 919
 920 static int
 921 shminit(void)
 922 {
 923         struct prison *pr;
 924         void **rsv;
 925         int i, error;
 926         osd_method_t methods[PR_MAXMETHOD] = {
 927             [PR_METHOD_CHECK] =         shm_prison_check,
 928             [PR_METHOD_SET] =           shm_prison_set,
 929             [PR_METHOD_GET] =           shm_prison_get,
 930             [PR_METHOD_REMOVE] =        shm_prison_remove,
 931         };
 932
 933 #ifndef BURN_BRIDGES
 934         if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
 935                 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
 936 #endif
 937         if (shminfo.shmmax == SHMMAX) {
 938                 /* Initialize shmmax dealing with possible overflow. */
 939                 for (i = PAGE_SIZE; i != 0; i--) {
 940                         shminfo.shmmax = shminfo.shmall * i;
 941                         if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
 942                                 break;
 943                 }
 944         }
 945         shmalloced = shminfo.shmmni;
 946         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 947         for (i = 0; i < shmalloced; i++) {
 948                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 949                 shmsegs[i].u.shm_perm.seq = 0;
 950 #ifdef MAC
 951                 mac_sysvshm_init(&shmsegs[i]);
 952 #endif
 953         }
 954         shm_last_free = 0;
 955         shm_nused = 0;
 956         shm_committed = 0;
 957         sx_init(&sysvshmsx, "sysvshmsx");
 958         shmexit_hook = &shmexit_myhook;
 959         shmfork_hook = &shmfork_myhook;
 960
 961         /* Set current prisons according to their allow.sysvipc. */
 962         shm_prison_slot = osd_jail_register(NULL, methods);
 963         rsv = osd_reserve(shm_prison_slot);
 964         prison_lock(&prison0);
 965         (void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
 966         prison_unlock(&prison0);
 967         rsv = NULL;
 968         sx_slock(&allprison_lock);
 969         TAILQ_FOREACH(pr, &allprison, pr_list) {
 970                 if (rsv == NULL)
 971                         rsv = osd_reserve(shm_prison_slot);
 972                 prison_lock(pr);
 973                 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
 974                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
 975                             &prison0);
 976                         rsv = NULL;
 977                 }
 978                 prison_unlock(pr);
 979         }
 980         if (rsv != NULL)
 981                 osd_free_reserved(rsv);
 982         sx_sunlock(&allprison_lock);
 983
 984         error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
 985         if (error != 0)
 986                 return (error);
 987 #ifdef COMPAT_FREEBSD32
 988         error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
 989         if (error != 0)
 990                 return (error);
 991 #endif
 992         return (0);
 993 }
 994
 995 static int
 996 shmunload(void)
 997 {
 998         int i;
 999
1000         if (shm_nused > 0)
1001                 return (EBUSY);
1002
1003 #ifdef COMPAT_FREEBSD32
1004         syscall32_helper_unregister(shm32_syscalls);
1005 #endif
1006         syscall_helper_unregister(shm_syscalls);
1007         if (shm_prison_slot != 0)
1008                 osd_jail_deregister(shm_prison_slot);
1009
1010         for (i = 0; i < shmalloced; i++) {
1011 #ifdef MAC
1012                 mac_sysvshm_destroy(&shmsegs[i]);
1013 #endif
1014                 /*
1015                  * Objects might be still mapped into the processes
1016                  * address spaces.  Actual free would happen on the
1017                  * last mapping destruction.
1018                  */
1019                 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1020                         vm_object_deallocate(shmsegs[i].object);
1021         }
1022         free(shmsegs, M_SHM);
1023         shmexit_hook = NULL;
1024         shmfork_hook = NULL;
1025         sx_destroy(&sysvshmsx);
1026         return (0);
1027 }
1028
1029 static int
1030 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1031 {
1032         struct shmid_kernel tshmseg;
1033         struct prison *pr, *rpr;
1034         int error, i;
1035
1036         SYSVSHM_LOCK();
1037         pr = req->td->td_ucred->cr_prison;
1038         rpr = shm_find_prison(req->td->td_ucred);
1039         error = 0;
1040         for (i = 0; i < shmalloced; i++) {
1041                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1042                     rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1043                         bzero(&tshmseg, sizeof(tshmseg));
1044                         tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1045                 } else {
1046                         tshmseg = shmsegs[i];
1047                         if (tshmseg.cred->cr_prison != pr)
1048                                 tshmseg.u.shm_perm.key = IPC_PRIVATE;
1049                 }
1050                 error = SYSCTL_OUT(req, &tshmseg, sizeof(tshmseg));
1051                 if (error != 0)
1052                         break;
1053         }
1054         SYSVSHM_UNLOCK();
1055         return (error);
1056 }
1057
1058 static int
1059 shm_prison_check(void *obj, void *data)
1060 {
1061         struct prison *pr = obj;
1062         struct prison *prpr;
1063         struct vfsoptlist *opts = data;
1064         int error, jsys;
1065
1066         /*
1067          * sysvshm is a jailsys integer.
1068          * It must be "disable" if the parent jail is disabled.
1069          */
1070         error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1071         if (error != ENOENT) {
1072                 if (error != 0)
1073                         return (error);
1074                 switch (jsys) {
1075                 case JAIL_SYS_DISABLE:
1076                         break;
1077                 case JAIL_SYS_NEW:
1078                 case JAIL_SYS_INHERIT:
1079                         prison_lock(pr->pr_parent);
1080                         prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1081                         prison_unlock(pr->pr_parent);
1082                         if (prpr == NULL)
1083                                 return (EPERM);
1084                         break;
1085                 default:
1086                         return (EINVAL);
1087                 }
1088         }
1089
1090         return (0);
1091 }
1092
1093 static int
1094 shm_prison_set(void *obj, void *data)
1095 {
1096         struct prison *pr = obj;
1097         struct prison *tpr, *orpr, *nrpr, *trpr;
1098         struct vfsoptlist *opts = data;
1099         void *rsv;
1100         int jsys, descend;
1101
1102         /*
1103          * sysvshm controls which jail is the root of the associated segments
1104          * (this jail or same as the parent), or if the feature is available
1105          * at all.
1106          */
1107         if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1108                 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1109                     ? JAIL_SYS_INHERIT
1110                     : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1111                     ? JAIL_SYS_DISABLE
1112                     : -1;
1113         if (jsys == JAIL_SYS_DISABLE) {
1114                 prison_lock(pr);
1115                 orpr = osd_jail_get(pr, shm_prison_slot);
1116                 if (orpr != NULL)
1117                         osd_jail_del(pr, shm_prison_slot);
1118                 prison_unlock(pr);
1119                 if (orpr != NULL) {
1120                         if (orpr == pr)
1121                                 shm_prison_cleanup(pr);
1122                         /* Disable all child jails as well. */
1123                         FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1124                                 prison_lock(tpr);
1125                                 trpr = osd_jail_get(tpr, shm_prison_slot);
1126                                 if (trpr != NULL) {
1127                                         osd_jail_del(tpr, shm_prison_slot);
1128                                         prison_unlock(tpr);
1129                                         if (trpr == tpr)
1130                                                 shm_prison_cleanup(tpr);
1131                                 } else {
1132                                         prison_unlock(tpr);
1133                                         descend = 0;
1134                                 }
1135                         }
1136                 }
1137         } else if (jsys != -1) {
1138                 if (jsys == JAIL_SYS_NEW)
1139                         nrpr = pr;
1140                 else {
1141                         prison_lock(pr->pr_parent);
1142                         nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1143                         prison_unlock(pr->pr_parent);
1144                 }
1145                 rsv = osd_reserve(shm_prison_slot);
1146                 prison_lock(pr);
1147                 orpr = osd_jail_get(pr, shm_prison_slot);
1148                 if (orpr != nrpr)
1149                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1150                             nrpr);
1151                 else
1152                         osd_free_reserved(rsv);
1153                 prison_unlock(pr);
1154                 if (orpr != nrpr) {
1155                         if (orpr == pr)
1156                                 shm_prison_cleanup(pr);
1157                         if (orpr != NULL) {
1158                                 /* Change child jails matching the old root, */
1159                                 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1160                                         prison_lock(tpr);
1161                                         trpr = osd_jail_get(tpr,
1162                                             shm_prison_slot);
1163                                         if (trpr == orpr) {
1164                                                 (void)osd_jail_set(tpr,
1165                                                     shm_prison_slot, nrpr);
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                 }
1176         }
1177
1178         return (0);
1179 }
1180
1181 static int
1182 shm_prison_get(void *obj, void *data)
1183 {
1184         struct prison *pr = obj;
1185         struct prison *rpr;
1186         struct vfsoptlist *opts = data;
1187         int error, jsys;
1188
1189         /* Set sysvshm based on the jail's root prison. */
1190         prison_lock(pr);
1191         rpr = osd_jail_get(pr, shm_prison_slot);
1192         prison_unlock(pr);
1193         jsys = rpr == NULL ? JAIL_SYS_DISABLE
1194             : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1195         error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1196         if (error == ENOENT)
1197                 error = 0;
1198         return (error);
1199 }
1200
1201 static int
1202 shm_prison_remove(void *obj, void *data __unused)
1203 {
1204         struct prison *pr = obj;
1205         struct prison *rpr;
1206
1207         SYSVSHM_LOCK();
1208         prison_lock(pr);
1209         rpr = osd_jail_get(pr, shm_prison_slot);
1210         prison_unlock(pr);
1211         if (rpr == pr)
1212                 shm_prison_cleanup(pr);
1213         SYSVSHM_UNLOCK();
1214         return (0);
1215 }
1216
1217 static void
1218 shm_prison_cleanup(struct prison *pr)
1219 {
1220         struct shmid_kernel *shmseg;
1221         int i;
1222
1223         /* Remove any segments that belong to this jail. */
1224         for (i = 0; i < shmalloced; i++) {
1225                 shmseg = &shmsegs[i];
1226                 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1227                     shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1228                         shm_remove(shmseg, i);
1229                 }
1230         }
1231 }
1232
1233 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1234
1235 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1236 struct oshmid_ds {
1237         struct  ipc_perm_old shm_perm;  /* operation perms */
1238         int     shm_segsz;              /* size of segment (bytes) */
1239         u_short shm_cpid;               /* pid, creator */
1240         u_short shm_lpid;               /* pid, last operation */
1241         short   shm_nattch;             /* no. of current attaches */
1242         time_t  shm_atime;              /* last attach time */
1243         time_t  shm_dtime;              /* last detach time */
1244         time_t  shm_ctime;              /* last change time */
1245         void    *shm_handle;            /* internal handle for shm segment */
1246 };
1247
1248 struct oshmctl_args {
1249         int shmid;
1250         int cmd;
1251         struct oshmid_ds *ubuf;
1252 };
1253
1254 static int
1255 oshmctl(struct thread *td, struct oshmctl_args *uap)
1256 {
1257 #ifdef COMPAT_43
1258         int error = 0;
1259         struct prison *rpr;
1260         struct shmid_kernel *shmseg;
1261         struct oshmid_ds outbuf;
1262
1263         rpr = shm_find_prison(td->td_ucred);
1264         if (rpr == NULL)
1265                 return (ENOSYS);
1266         if (uap->cmd != IPC_STAT) {
1267                 return (freebsd7_shmctl(td,
1268                     (struct freebsd7_shmctl_args *)uap));
1269         }
1270         SYSVSHM_LOCK();
1271         shmseg = shm_find_segment(rpr, uap->shmid, true);
1272         if (shmseg == NULL) {
1273                 SYSVSHM_UNLOCK();
1274                 return (EINVAL);
1275         }
1276         error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1277         if (error != 0) {
1278                 SYSVSHM_UNLOCK();
1279                 return (error);
1280         }
1281 #ifdef MAC
1282         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1283         if (error != 0) {
1284                 SYSVSHM_UNLOCK();
1285                 return (error);
1286         }
1287 #endif
1288         ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1289         outbuf.shm_segsz = shmseg->u.shm_segsz;
1290         outbuf.shm_cpid = shmseg->u.shm_cpid;
1291         outbuf.shm_lpid = shmseg->u.shm_lpid;
1292         outbuf.shm_nattch = shmseg->u.shm_nattch;
1293         outbuf.shm_atime = shmseg->u.shm_atime;
1294         outbuf.shm_dtime = shmseg->u.shm_dtime;
1295         outbuf.shm_ctime = shmseg->u.shm_ctime;
1296         outbuf.shm_handle = shmseg->object;
1297         SYSVSHM_UNLOCK();
1298         return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1299 #else
1300         return (EINVAL);
1301 #endif
1302 }
1303
1304 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1305 static sy_call_t *shmcalls[] = {
1306         (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1307         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1308         (sy_call_t *)freebsd7_shmctl
1309 };
1310
1311 #ifndef _SYS_SYSPROTO_H_
1312 /* XXX actually varargs. */
1313 struct shmsys_args {
1314         int     which;
1315         int     a2;
1316         int     a3;
1317         int     a4;
1318 };
1319 #endif
1320 int
1321 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1322 {
1323
1324         AUDIT_ARG_SVIPC_WHICH(uap->which);
1325         if (uap->which < 0 || uap->which >= nitems(shmcalls))
1326                 return (EINVAL);
1327         return ((*shmcalls[uap->which])(td, &uap->a2));
1328 }
1329
1330 #endif  /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1331
1332 #ifdef COMPAT_FREEBSD32
1333
1334 int
1335 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1336 {
1337
1338 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1339     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1340         AUDIT_ARG_SVIPC_WHICH(uap->which);
1341         switch (uap->which) {
1342         case 0: {       /* shmat */
1343                 struct shmat_args ap;
1344
1345                 ap.shmid = uap->a2;
1346                 ap.shmaddr = PTRIN(uap->a3);
1347                 ap.shmflg = uap->a4;
1348                 return (sysent[SYS_shmat].sy_call(td, &ap));
1349         }
1350         case 2: {       /* shmdt */
1351                 struct shmdt_args ap;
1352
1353                 ap.shmaddr = PTRIN(uap->a2);
1354                 return (sysent[SYS_shmdt].sy_call(td, &ap));
1355         }
1356         case 3: {       /* shmget */
1357                 struct shmget_args ap;
1358
1359                 ap.key = uap->a2;
1360                 ap.size = uap->a3;
1361                 ap.shmflg = uap->a4;
1362                 return (sysent[SYS_shmget].sy_call(td, &ap));
1363         }
1364         case 4: {       /* shmctl */
1365                 struct freebsd7_freebsd32_shmctl_args ap;
1366
1367                 ap.shmid = uap->a2;
1368                 ap.cmd = uap->a3;
1369                 ap.buf = PTRIN(uap->a4);
1370                 return (freebsd7_freebsd32_shmctl(td, &ap));
1371         }
1372         case 1:         /* oshmctl */
1373         default:
1374                 return (EINVAL);
1375         }
1376 #else
1377         return (nosys(td, NULL));
1378 #endif
1379 }
1380
1381 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1382     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1383 int
1384 freebsd7_freebsd32_shmctl(struct thread *td,
1385     struct freebsd7_freebsd32_shmctl_args *uap)
1386 {
1387         int error;
1388         union {
1389                 struct shmid_ds shmid_ds;
1390                 struct shm_info shm_info;
1391                 struct shminfo shminfo;
1392         } u;
1393         union {
1394                 struct shmid_ds32_old shmid_ds32;
1395                 struct shm_info32 shm_info32;
1396                 struct shminfo32 shminfo32;
1397         } u32;
1398         size_t sz;
1399
1400         if (uap->cmd == IPC_SET) {
1401                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1402                     sizeof(u32.shmid_ds32))))
1403                         goto done;
1404                 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1405                     &u.shmid_ds.shm_perm);
1406                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1407                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1408                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1409                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1410                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1411                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1412                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1413         }
1414
1415         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1416         if (error)
1417                 goto done;
1418
1419         /* Cases in which we need to copyout */
1420         switch (uap->cmd) {
1421         case IPC_INFO:
1422                 CP(u.shminfo, u32.shminfo32, shmmax);
1423                 CP(u.shminfo, u32.shminfo32, shmmin);
1424                 CP(u.shminfo, u32.shminfo32, shmmni);
1425                 CP(u.shminfo, u32.shminfo32, shmseg);
1426                 CP(u.shminfo, u32.shminfo32, shmall);
1427                 error = copyout(&u32.shminfo32, uap->buf,
1428                     sizeof(u32.shminfo32));
1429                 break;
1430         case SHM_INFO:
1431                 CP(u.shm_info, u32.shm_info32, used_ids);
1432                 CP(u.shm_info, u32.shm_info32, shm_rss);
1433                 CP(u.shm_info, u32.shm_info32, shm_tot);
1434                 CP(u.shm_info, u32.shm_info32, shm_swp);
1435                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1436                 CP(u.shm_info, u32.shm_info32, swap_successes);
1437                 error = copyout(&u32.shm_info32, uap->buf,
1438                     sizeof(u32.shm_info32));
1439                 break;
1440         case SHM_STAT:
1441         case IPC_STAT:
1442                 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1443                     &u32.shmid_ds32.shm_perm);
1444                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1445                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1446                 else
1447                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1448                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1449                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1450                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1451                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1452                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1453                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1454                 u32.shmid_ds32.shm_internal = 0;
1455                 error = copyout(&u32.shmid_ds32, uap->buf,
1456                     sizeof(u32.shmid_ds32));
1457                 break;
1458         }
1459
1460 done:
1461         if (error) {
1462                 /* Invalidate the return value */
1463                 td->td_retval[0] = -1;
1464         }
1465         return (error);
1466 }
1467 #endif
1468
1469 int
1470 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1471 {
1472         int error;
1473         union {
1474                 struct shmid_ds shmid_ds;
1475                 struct shm_info shm_info;
1476                 struct shminfo shminfo;
1477         } u;
1478         union {
1479                 struct shmid_ds32 shmid_ds32;
1480                 struct shm_info32 shm_info32;
1481                 struct shminfo32 shminfo32;
1482         } u32;
1483         size_t sz;
1484
1485         if (uap->cmd == IPC_SET) {
1486                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1487                     sizeof(u32.shmid_ds32))))
1488                         goto done;
1489                 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1490                     &u.shmid_ds.shm_perm);
1491                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1492                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1493                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1494                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1495                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1496                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1497                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1498         }
1499
1500         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1501         if (error)
1502                 goto done;
1503
1504         /* Cases in which we need to copyout */
1505         switch (uap->cmd) {
1506         case IPC_INFO:
1507                 CP(u.shminfo, u32.shminfo32, shmmax);
1508                 CP(u.shminfo, u32.shminfo32, shmmin);
1509                 CP(u.shminfo, u32.shminfo32, shmmni);
1510                 CP(u.shminfo, u32.shminfo32, shmseg);
1511                 CP(u.shminfo, u32.shminfo32, shmall);
1512                 error = copyout(&u32.shminfo32, uap->buf,
1513                     sizeof(u32.shminfo32));
1514                 break;
1515         case SHM_INFO:
1516                 CP(u.shm_info, u32.shm_info32, used_ids);
1517                 CP(u.shm_info, u32.shm_info32, shm_rss);
1518                 CP(u.shm_info, u32.shm_info32, shm_tot);
1519                 CP(u.shm_info, u32.shm_info32, shm_swp);
1520                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1521                 CP(u.shm_info, u32.shm_info32, swap_successes);
1522                 error = copyout(&u32.shm_info32, uap->buf,
1523                     sizeof(u32.shm_info32));
1524                 break;
1525         case SHM_STAT:
1526         case IPC_STAT:
1527                 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1528                     &u32.shmid_ds32.shm_perm);
1529                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1530                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1531                 else
1532                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1533                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1534                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1535                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1536                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1537                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1538                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1539                 error = copyout(&u32.shmid_ds32, uap->buf,
1540                     sizeof(u32.shmid_ds32));
1541                 break;
1542         }
1543
1544 done:
1545         if (error) {
1546                 /* Invalidate the return value */
1547                 td->td_retval[0] = -1;
1548         }
1549         return (error);
1550 }
1551 #endif
1552
1553 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1554     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1555
1556 #ifndef CP
1557 #define CP(src, dst, fld)       do { (dst).fld = (src).fld; } while (0)
1558 #endif
1559
1560 #ifndef _SYS_SYSPROTO_H_
1561 struct freebsd7_shmctl_args {
1562         int shmid;
1563         int cmd;
1564         struct shmid_ds_old *buf;
1565 };
1566 #endif
1567 int
1568 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1569 {
1570         int error;
1571         struct shmid_ds_old old;
1572         struct shmid_ds buf;
1573         size_t bufsz;
1574
1575         /*
1576          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1577          * Linux binaries.  If we see the call come through the FreeBSD ABI,
1578          * return an error back to the user since we do not to support this.
1579          */
1580         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1581             uap->cmd == SHM_STAT)
1582                 return (EINVAL);
1583
1584         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1585         if (uap->cmd == IPC_SET) {
1586                 if ((error = copyin(uap->buf, &old, sizeof(old))))
1587                         goto done;
1588                 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1589                 CP(old, buf, shm_segsz);
1590                 CP(old, buf, shm_lpid);
1591                 CP(old, buf, shm_cpid);
1592                 CP(old, buf, shm_nattch);
1593                 CP(old, buf, shm_atime);
1594                 CP(old, buf, shm_dtime);
1595                 CP(old, buf, shm_ctime);
1596         }
1597
1598         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1599         if (error)
1600                 goto done;
1601
1602         /* Cases in which we need to copyout */
1603         switch (uap->cmd) {
1604         case IPC_STAT:
1605                 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1606                 if (buf.shm_segsz > INT_MAX)
1607                         old.shm_segsz = INT_MAX;
1608                 else
1609                         CP(buf, old, shm_segsz);
1610                 CP(buf, old, shm_lpid);
1611                 CP(buf, old, shm_cpid);
1612                 if (buf.shm_nattch > SHRT_MAX)
1613                         old.shm_nattch = SHRT_MAX;
1614                 else
1615                         CP(buf, old, shm_nattch);
1616                 CP(buf, old, shm_atime);
1617                 CP(buf, old, shm_dtime);
1618                 CP(buf, old, shm_ctime);
1619                 old.shm_internal = NULL;
1620                 error = copyout(&old, uap->buf, sizeof(old));
1621                 break;
1622         }
1623
1624 done:
1625         if (error) {
1626                 /* Invalidate the return value */
1627                 td->td_retval[0] = -1;
1628         }
1629         return (error);
1630 }
1631
1632 #endif  /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1633            COMPAT_FREEBSD7 */
1634
1635 static int
1636 sysvshm_modload(struct module *module, int cmd, void *arg)
1637 {
1638         int error = 0;
1639
1640         switch (cmd) {
1641         case MOD_LOAD:
1642                 error = shminit();
1643                 if (error != 0)
1644                         shmunload();
1645                 break;
1646         case MOD_UNLOAD:
1647                 error = shmunload();
1648                 break;
1649         case MOD_SHUTDOWN:
1650                 break;
1651         default:
1652                 error = EINVAL;
1653                 break;
1654         }
1655         return (error);
1656 }
1657
1658 static moduledata_t sysvshm_mod = {
1659         "sysvshm",
1660         &sysvshm_modload,
1661         NULL
1662 };
1663
1664 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1665 MODULE_VERSION(sysvshm, 1);