sys/kern/sysv_shm.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-4-Clause AND BSD-2-Clause-FreeBSD
   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 __FBSDID("$FreeBSD$");
  73
  74 #include "opt_sysvipc.h"
  75
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/abi_compat.h>
  79 #include <sys/kernel.h>
  80 #include <sys/limits.h>
  81 #include <sys/lock.h>
  82 #include <sys/sysctl.h>
  83 #include <sys/shm.h>
  84 #include <sys/proc.h>
  85 #include <sys/malloc.h>
  86 #include <sys/mman.h>
  87 #include <sys/module.h>
  88 #include <sys/mutex.h>
  89 #include <sys/racct.h>
  90 #include <sys/resourcevar.h>
  91 #include <sys/rwlock.h>
  92 #include <sys/stat.h>
  93 #include <sys/syscall.h>
  94 #include <sys/syscallsubr.h>
  95 #include <sys/sysent.h>
  96 #include <sys/sysproto.h>
  97 #include <sys/jail.h>
  98
  99 #include <security/audit/audit.h>
 100 #include <security/mac/mac_framework.h>
 101
 102 #include <vm/vm.h>
 103 #include <vm/vm_param.h>
 104 #include <vm/pmap.h>
 105 #include <vm/vm_object.h>
 106 #include <vm/vm_map.h>
 107 #include <vm/vm_page.h>
 108 #include <vm/vm_pager.h>
 109
 110 FEATURE(sysv_shm, "System V shared memory segments support");
 111
 112 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
 113
 114 #define SHMSEG_FREE             0x0200
 115 #define SHMSEG_REMOVED          0x0400
 116 #define SHMSEG_ALLOCATED        0x0800
 117
 118 static int shm_last_free, shm_nused, shmalloced;
 119 vm_size_t shm_committed;
 120 static struct shmid_kernel *shmsegs;
 121 static unsigned shm_prison_slot;
 122
 123 struct shmmap_state {
 124         vm_offset_t va;
 125         int shmid;
 126 };
 127
 128 static void shm_deallocate_segment(struct shmid_kernel *);
 129 static int shm_find_segment_by_key(struct prison *, key_t);
 130 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
 131 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
 132 static int shmget_allocate_segment(struct thread *td, key_t key, size_t size,
 133     int mode);
 134 static int shmget_existing(struct thread *td, size_t size, int shmflg,
 135     int mode, int segnum);
 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 #ifndef SYSVSHM
 141 static void shmexit_myhook(struct vmspace *vm);
 142 static void shmfork_myhook(struct proc *p1, struct proc *p2);
 143 #endif
 144 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
 145 static void shm_remove(struct shmid_kernel *, int);
 146 static struct prison *shm_find_prison(struct ucred *);
 147 static int shm_prison_cansee(struct prison *, struct shmid_kernel *);
 148 static int shm_prison_check(void *, void *);
 149 static int shm_prison_set(void *, void *);
 150 static int shm_prison_get(void *, void *);
 151 static int shm_prison_remove(void *, void *);
 152 static void shm_prison_cleanup(struct prison *);
 153
 154 /*
 155  * Tuneable values.
 156  */
 157 #ifndef SHMMAXPGS
 158 #define SHMMAXPGS       131072  /* Note: sysv shared memory is swap backed. */
 159 #endif
 160 #ifndef SHMMAX
 161 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 162 #endif
 163 #ifndef SHMMIN
 164 #define SHMMIN  1
 165 #endif
 166 #ifndef SHMMNI
 167 #define SHMMNI  192
 168 #endif
 169 #ifndef SHMSEG
 170 #define SHMSEG  128
 171 #endif
 172 #ifndef SHMALL
 173 #define SHMALL  (SHMMAXPGS)
 174 #endif
 175
 176 struct  shminfo shminfo = {
 177         .shmmax = SHMMAX,
 178         .shmmin = SHMMIN,
 179         .shmmni = SHMMNI,
 180         .shmseg = SHMSEG,
 181         .shmall = SHMALL
 182 };
 183
 184 static int shm_use_phys;
 185 static int shm_allow_removed = 1;
 186
 187 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
 188     "Maximum shared memory segment size");
 189 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
 190     "Minimum shared memory segment size");
 191 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
 192     "Number of shared memory identifiers");
 193 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
 194     "Number of segments per process");
 195 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
 196     "Maximum number of pages available for shared memory");
 197 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
 198     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
 199 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
 200     &shm_allow_removed, 0,
 201     "Enable/Disable attachment to attached segments marked for removal");
 202 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
 203     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
 204     "Array of struct shmid_kernel for each potential shared memory segment");
 205
 206 static struct sx sysvshmsx;
 207 #define SYSVSHM_LOCK()          sx_xlock(&sysvshmsx)
 208 #define SYSVSHM_UNLOCK()        sx_xunlock(&sysvshmsx)
 209 #define SYSVSHM_ASSERT_LOCKED() sx_assert(&sysvshmsx, SA_XLOCKED)
 210
 211 static int
 212 shm_find_segment_by_key(struct prison *pr, key_t key)
 213 {
 214         int i;
 215
 216         for (i = 0; i < shmalloced; i++)
 217                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
 218                     shmsegs[i].cred != NULL &&
 219                     shmsegs[i].cred->cr_prison == pr &&
 220                     shmsegs[i].u.shm_perm.key == key)
 221                         return (i);
 222         return (-1);
 223 }
 224
 225 /*
 226  * Finds segment either by shmid if is_shmid is true, or by segnum if
 227  * is_shmid is false.
 228  */
 229 static struct shmid_kernel *
 230 shm_find_segment(struct prison *rpr, int arg, bool is_shmid)
 231 {
 232         struct shmid_kernel *shmseg;
 233         int segnum;
 234
 235         segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
 236         if (segnum < 0 || segnum >= shmalloced)
 237                 return (NULL);
 238         shmseg = &shmsegs[segnum];
 239         if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 240             (!shm_allow_removed &&
 241             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 242             (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)) ||
 243             shm_prison_cansee(rpr, shmseg) != 0)
 244                 return (NULL);
 245         return (shmseg);
 246 }
 247
 248 static void
 249 shm_deallocate_segment(struct shmid_kernel *shmseg)
 250 {
 251         vm_size_t size;
 252
 253         SYSVSHM_ASSERT_LOCKED();
 254
 255         vm_object_deallocate(shmseg->object);
 256         shmseg->object = NULL;
 257         size = round_page(shmseg->u.shm_segsz);
 258         shm_committed -= btoc(size);
 259         shm_nused--;
 260         shmseg->u.shm_perm.mode = SHMSEG_FREE;
 261 #ifdef MAC
 262         mac_sysvshm_cleanup(shmseg);
 263 #endif
 264         racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
 265         racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
 266         crfree(shmseg->cred);
 267         shmseg->cred = NULL;
 268 }
 269
 270 static int
 271 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 272 {
 273         struct shmid_kernel *shmseg;
 274         int segnum, result;
 275         vm_size_t size;
 276
 277         SYSVSHM_ASSERT_LOCKED();
 278         segnum = IPCID_TO_IX(shmmap_s->shmid);
 279         KASSERT(segnum >= 0 && segnum < shmalloced,
 280             ("segnum %d shmalloced %d", segnum, shmalloced));
 281
 282         shmseg = &shmsegs[segnum];
 283         size = round_page(shmseg->u.shm_segsz);
 284         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 285         if (result != KERN_SUCCESS)
 286                 return (EINVAL);
 287         shmmap_s->shmid = -1;
 288         shmseg->u.shm_dtime = time_second;
 289         if (--shmseg->u.shm_nattch == 0 &&
 290             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
 291                 shm_deallocate_segment(shmseg);
 292                 shm_last_free = segnum;
 293         }
 294         return (0);
 295 }
 296
 297 static void
 298 shm_remove(struct shmid_kernel *shmseg, int segnum)
 299 {
 300
 301         shmseg->u.shm_perm.key = IPC_PRIVATE;
 302         shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
 303         if (shmseg->u.shm_nattch == 0) {
 304                 shm_deallocate_segment(shmseg);
 305                 shm_last_free = segnum;
 306         }
 307 }
 308
 309 static struct prison *
 310 shm_find_prison(struct ucred *cred)
 311 {
 312         struct prison *pr, *rpr;
 313
 314         pr = cred->cr_prison;
 315         prison_lock(pr);
 316         rpr = osd_jail_get(pr, shm_prison_slot);
 317         prison_unlock(pr);
 318         return rpr;
 319 }
 320
 321 static int
 322 shm_prison_cansee(struct prison *rpr, struct shmid_kernel *shmseg)
 323 {
 324
 325         if (shmseg->cred == NULL ||
 326             !(rpr == shmseg->cred->cr_prison ||
 327               prison_ischild(rpr, shmseg->cred->cr_prison)))
 328                 return (EINVAL);
 329         return (0);
 330 }
 331
 332 static int
 333 kern_shmdt_locked(struct thread *td, const void *shmaddr)
 334 {
 335         struct proc *p = td->td_proc;
 336         struct shmmap_state *shmmap_s;
 337 #ifdef MAC
 338         int error;
 339 #endif
 340         int i;
 341
 342         SYSVSHM_ASSERT_LOCKED();
 343         if (shm_find_prison(td->td_ucred) == NULL)
 344                 return (ENOSYS);
 345         shmmap_s = p->p_vmspace->vm_shm;
 346         if (shmmap_s == NULL)
 347                 return (EINVAL);
 348         AUDIT_ARG_SVIPC_ID(shmmap_s->shmid);
 349         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 350                 if (shmmap_s->shmid != -1 &&
 351                     shmmap_s->va == (vm_offset_t)shmaddr) {
 352                         break;
 353                 }
 354         }
 355         if (i == shminfo.shmseg)
 356                 return (EINVAL);
 357 #ifdef MAC
 358         error = mac_sysvshm_check_shmdt(td->td_ucred,
 359             &shmsegs[IPCID_TO_IX(shmmap_s->shmid)]);
 360         if (error != 0)
 361                 return (error);
 362 #endif
 363         return (shm_delete_mapping(p->p_vmspace, shmmap_s));
 364 }
 365
 366 #ifndef _SYS_SYSPROTO_H_
 367 struct shmdt_args {
 368         const void *shmaddr;
 369 };
 370 #endif
 371 int
 372 sys_shmdt(struct thread *td, struct shmdt_args *uap)
 373 {
 374         int error;
 375
 376         SYSVSHM_LOCK();
 377         error = kern_shmdt_locked(td, uap->shmaddr);
 378         SYSVSHM_UNLOCK();
 379         return (error);
 380 }
 381
 382 static int
 383 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
 384     int shmflg)
 385 {
 386         struct prison *rpr;
 387         struct proc *p = td->td_proc;
 388         struct shmid_kernel *shmseg;
 389         struct shmmap_state *shmmap_s;
 390         vm_offset_t attach_va;
 391         vm_prot_t prot;
 392         vm_size_t size;
 393         int cow, error, find_space, i, rv;
 394
 395         AUDIT_ARG_SVIPC_ID(shmid);
 396         AUDIT_ARG_VALUE(shmflg);
 397
 398         SYSVSHM_ASSERT_LOCKED();
 399         rpr = shm_find_prison(td->td_ucred);
 400         if (rpr == NULL)
 401                 return (ENOSYS);
 402         shmmap_s = p->p_vmspace->vm_shm;
 403         if (shmmap_s == NULL) {
 404                 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
 405                     M_SHM, M_WAITOK);
 406                 for (i = 0; i < shminfo.shmseg; i++)
 407                         shmmap_s[i].shmid = -1;
 408                 KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
 409                 p->p_vmspace->vm_shm = shmmap_s;
 410         }
 411         shmseg = shm_find_segment(rpr, shmid, true);
 412         if (shmseg == NULL)
 413                 return (EINVAL);
 414         error = ipcperm(td, &shmseg->u.shm_perm,
 415             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 416         if (error != 0)
 417                 return (error);
 418 #ifdef MAC
 419         error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
 420         if (error != 0)
 421                 return (error);
 422 #endif
 423         for (i = 0; i < shminfo.shmseg; i++) {
 424                 if (shmmap_s->shmid == -1)
 425                         break;
 426                 shmmap_s++;
 427         }
 428         if (i >= shminfo.shmseg)
 429                 return (EMFILE);
 430         size = round_page(shmseg->u.shm_segsz);
 431         prot = VM_PROT_READ;
 432         cow = MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL;
 433         if ((shmflg & SHM_RDONLY) == 0)
 434                 prot |= VM_PROT_WRITE;
 435         if (shmaddr != NULL) {
 436                 if ((shmflg & SHM_RND) != 0)
 437                         attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
 438                 else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
 439                         attach_va = (vm_offset_t)shmaddr;
 440                 else
 441                         return (EINVAL);
 442                 if ((shmflg & SHM_REMAP) != 0)
 443                         cow |= MAP_REMAP;
 444                 find_space = VMFS_NO_SPACE;
 445         } else {
 446                 /*
 447                  * This is just a hint to vm_map_find() about where to
 448                  * put it.
 449                  */
 450                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
 451                     lim_max(td, RLIMIT_DATA));
 452                 find_space = VMFS_OPTIMAL_SPACE;
 453         }
 454
 455         vm_object_reference(shmseg->object);
 456         rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
 457             size, 0, find_space, prot, prot, cow);
 458         if (rv != KERN_SUCCESS) {
 459                 vm_object_deallocate(shmseg->object);
 460                 return (ENOMEM);
 461         }
 462
 463         shmmap_s->va = attach_va;
 464         shmmap_s->shmid = shmid;
 465         shmseg->u.shm_lpid = p->p_pid;
 466         shmseg->u.shm_atime = time_second;
 467         shmseg->u.shm_nattch++;
 468         td->td_retval[0] = attach_va;
 469         return (error);
 470 }
 471
 472 int
 473 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
 474 {
 475         int error;
 476
 477         SYSVSHM_LOCK();
 478         error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
 479         SYSVSHM_UNLOCK();
 480         return (error);
 481 }
 482
 483 #ifndef _SYS_SYSPROTO_H_
 484 struct shmat_args {
 485         int shmid;
 486         const void *shmaddr;
 487         int shmflg;
 488 };
 489 #endif
 490 int
 491 sys_shmat(struct thread *td, struct shmat_args *uap)
 492 {
 493
 494         return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
 495 }
 496
 497 static int
 498 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
 499     size_t *bufsz)
 500 {
 501         struct prison *rpr;
 502         struct shmid_kernel *shmseg;
 503         struct shmid_ds *shmidp;
 504         struct shm_info shm_info;
 505         int error;
 506
 507         SYSVSHM_ASSERT_LOCKED();
 508
 509         rpr = shm_find_prison(td->td_ucred);
 510         if (rpr == NULL)
 511                 return (ENOSYS);
 512
 513         AUDIT_ARG_SVIPC_ID(shmid);
 514         AUDIT_ARG_SVIPC_CMD(cmd);
 515
 516         switch (cmd) {
 517         /*
 518          * It is possible that kern_shmctl is being called from the Linux ABI
 519          * layer, in which case, we will need to implement IPC_INFO.  It should
 520          * be noted that other shmctl calls will be funneled through here for
 521          * Linix binaries as well.
 522          *
 523          * NB: The Linux ABI layer will convert this data to structure(s) more
 524          * consistent with the Linux ABI.
 525          */
 526         case IPC_INFO:
 527                 memcpy(buf, &shminfo, sizeof(shminfo));
 528                 if (bufsz)
 529                         *bufsz = sizeof(shminfo);
 530                 td->td_retval[0] = shmalloced;
 531                 return (0);
 532         case SHM_INFO: {
 533                 shm_info.used_ids = shm_nused;
 534                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 535                 shm_info.shm_tot = 0;   /*XXX where to get from ? */
 536                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 537                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 538                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 539                 memcpy(buf, &shm_info, sizeof(shm_info));
 540                 if (bufsz != NULL)
 541                         *bufsz = sizeof(shm_info);
 542                 td->td_retval[0] = shmalloced;
 543                 return (0);
 544         }
 545         }
 546         shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
 547         if (shmseg == NULL)
 548                 return (EINVAL);
 549 #ifdef MAC
 550         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
 551         if (error != 0)
 552                 return (error);
 553 #endif
 554         switch (cmd) {
 555         case SHM_STAT:
 556         case IPC_STAT:
 557                 shmidp = (struct shmid_ds *)buf;
 558                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 559                 if (error != 0)
 560                         return (error);
 561                 memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
 562                 if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
 563                         shmidp->shm_perm.key = IPC_PRIVATE;
 564                 if (bufsz != NULL)
 565                         *bufsz = sizeof(struct shmid_ds);
 566                 if (cmd == SHM_STAT) {
 567                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
 568                             shmseg->u.shm_perm);
 569                 }
 570                 break;
 571         case IPC_SET:
 572                 shmidp = (struct shmid_ds *)buf;
 573                 AUDIT_ARG_SVIPC_PERM(&shmidp->shm_perm);
 574                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 575                 if (error != 0)
 576                         return (error);
 577                 shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
 578                 shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
 579                 shmseg->u.shm_perm.mode =
 580                     (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
 581                     (shmidp->shm_perm.mode & ACCESSPERMS);
 582                 shmseg->u.shm_ctime = time_second;
 583                 break;
 584         case IPC_RMID:
 585                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 586                 if (error != 0)
 587                         return (error);
 588                 shm_remove(shmseg, IPCID_TO_IX(shmid));
 589                 break;
 590 #if 0
 591         case SHM_LOCK:
 592         case SHM_UNLOCK:
 593 #endif
 594         default:
 595                 error = EINVAL;
 596                 break;
 597         }
 598         return (error);
 599 }
 600
 601 int
 602 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
 603 {
 604         int error;
 605
 606         SYSVSHM_LOCK();
 607         error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
 608         SYSVSHM_UNLOCK();
 609         return (error);
 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 static int
 661 shmget_existing(struct thread *td, size_t size, int shmflg, int mode,
 662     int segnum)
 663 {
 664         struct shmid_kernel *shmseg;
 665 #ifdef MAC
 666         int error;
 667 #endif
 668
 669         SYSVSHM_ASSERT_LOCKED();
 670         KASSERT(segnum >= 0 && segnum < shmalloced,
 671             ("segnum %d shmalloced %d", segnum, shmalloced));
 672         shmseg = &shmsegs[segnum];
 673         if ((shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 674                 return (EEXIST);
 675 #ifdef MAC
 676         error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, shmflg);
 677         if (error != 0)
 678                 return (error);
 679 #endif
 680         if (size != 0 && size > shmseg->u.shm_segsz)
 681                 return (EINVAL);
 682         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 683         return (0);
 684 }
 685
 686 static int
 687 shmget_allocate_segment(struct thread *td, key_t key, size_t size, int mode)
 688 {
 689         struct ucred *cred = td->td_ucred;
 690         struct shmid_kernel *shmseg;
 691         vm_object_t shm_object;
 692         int i, segnum;
 693
 694         SYSVSHM_ASSERT_LOCKED();
 695
 696         if (size < shminfo.shmmin || size > shminfo.shmmax)
 697                 return (EINVAL);
 698         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 699                 return (ENOSPC);
 700         size = round_page(size);
 701         if (shm_committed + btoc(size) > shminfo.shmall)
 702                 return (ENOMEM);
 703         if (shm_last_free < 0) {
 704                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 705                 for (i = 0; i < shmalloced; i++)
 706                         if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
 707                                 break;
 708                 if (i == shmalloced)
 709                         return (ENOSPC);
 710                 segnum = i;
 711         } else  {
 712                 segnum = shm_last_free;
 713                 shm_last_free = -1;
 714         }
 715         KASSERT(segnum >= 0 && segnum < shmalloced,
 716             ("segnum %d shmalloced %d", segnum, shmalloced));
 717         shmseg = &shmsegs[segnum];
 718 #ifdef RACCT
 719         if (racct_enable) {
 720                 PROC_LOCK(td->td_proc);
 721                 if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
 722                         PROC_UNLOCK(td->td_proc);
 723                         return (ENOSPC);
 724                 }
 725                 if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
 726                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 727                         PROC_UNLOCK(td->td_proc);
 728                         return (ENOMEM);
 729                 }
 730                 PROC_UNLOCK(td->td_proc);
 731         }
 732 #endif
 733
 734         /*
 735          * We make sure that we have allocated a pager before we need
 736          * to.
 737          */
 738         shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
 739             0, size, VM_PROT_DEFAULT, 0, cred);
 740         if (shm_object == NULL) {
 741 #ifdef RACCT
 742                 if (racct_enable) {
 743                         PROC_LOCK(td->td_proc);
 744                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 745                         racct_sub(td->td_proc, RACCT_SHMSIZE, size);
 746                         PROC_UNLOCK(td->td_proc);
 747                 }
 748 #endif
 749                 return (ENOMEM);
 750         }
 751
 752         shmseg->object = shm_object;
 753         shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
 754         shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
 755         shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 756         shmseg->u.shm_perm.key = key;
 757         shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
 758         shmseg->cred = crhold(cred);
 759         shmseg->u.shm_segsz = size;
 760         shmseg->u.shm_cpid = td->td_proc->p_pid;
 761         shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
 762         shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
 763 #ifdef MAC
 764         mac_sysvshm_create(cred, shmseg);
 765 #endif
 766         shmseg->u.shm_ctime = time_second;
 767         shm_committed += btoc(size);
 768         shm_nused++;
 769         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 770
 771         return (0);
 772 }
 773
 774 #ifndef _SYS_SYSPROTO_H_
 775 struct shmget_args {
 776         key_t key;
 777         size_t size;
 778         int shmflg;
 779 };
 780 #endif
 781 int
 782 sys_shmget(struct thread *td, struct shmget_args *uap)
 783 {
 784         int segnum, mode;
 785         int error;
 786
 787         if (shm_find_prison(td->td_ucred) == NULL)
 788                 return (ENOSYS);
 789         mode = uap->shmflg & ACCESSPERMS;
 790         SYSVSHM_LOCK();
 791         if (uap->key == IPC_PRIVATE) {
 792                 error = shmget_allocate_segment(td, uap->key, uap->size, mode);
 793         } else {
 794                 segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
 795                     uap->key);
 796                 if (segnum >= 0)
 797                         error = shmget_existing(td, uap->size, uap->shmflg,
 798                             mode, segnum);
 799                 else if ((uap->shmflg & IPC_CREAT) == 0)
 800                         error = ENOENT;
 801                 else
 802                         error = shmget_allocate_segment(td, uap->key,
 803                             uap->size, mode);
 804         }
 805         SYSVSHM_UNLOCK();
 806         return (error);
 807 }
 808
 809 #ifdef SYSVSHM
 810 void
 811 shmfork(struct proc *p1, struct proc *p2)
 812 #else
 813 static void
 814 shmfork_myhook(struct proc *p1, struct proc *p2)
 815 #endif
 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 #ifdef SYSVSHM
 839 void
 840 shmexit(struct vmspace *vm)
 841 #else
 842 static void
 843 shmexit_myhook(struct vmspace *vm)
 844 #endif
 845 {
 846         struct shmmap_state *base, *shm;
 847         int i;
 848
 849         base = vm->vm_shm;
 850         if (base != NULL) {
 851                 vm->vm_shm = NULL;
 852                 SYSVSHM_LOCK();
 853                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 854                         if (shm->shmid != -1)
 855                                 shm_delete_mapping(vm, shm);
 856                 }
 857                 SYSVSHM_UNLOCK();
 858                 free(base, M_SHM);
 859         }
 860 }
 861
 862 static void
 863 shmrealloc(void)
 864 {
 865         struct shmid_kernel *newsegs;
 866         int i;
 867
 868         SYSVSHM_ASSERT_LOCKED();
 869
 870         if (shmalloced >= shminfo.shmmni)
 871                 return;
 872
 873         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM,
 874             M_WAITOK | M_ZERO);
 875         for (i = 0; i < shmalloced; i++)
 876                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 877         for (; i < shminfo.shmmni; i++) {
 878                 newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 879                 newsegs[i].u.shm_perm.seq = 0;
 880 #ifdef MAC
 881                 mac_sysvshm_init(&newsegs[i]);
 882 #endif
 883         }
 884         free(shmsegs, M_SHM);
 885         shmsegs = newsegs;
 886         shmalloced = shminfo.shmmni;
 887 }
 888
 889 static struct syscall_helper_data shm_syscalls[] = {
 890         SYSCALL_INIT_HELPER(shmat),
 891         SYSCALL_INIT_HELPER(shmctl),
 892         SYSCALL_INIT_HELPER(shmdt),
 893         SYSCALL_INIT_HELPER(shmget),
 894 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 895     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 896         SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
 897 #endif
 898 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 899         SYSCALL_INIT_HELPER(shmsys),
 900 #endif
 901         SYSCALL_INIT_LAST
 902 };
 903
 904 #ifdef COMPAT_FREEBSD32
 905 #include <compat/freebsd32/freebsd32.h>
 906 #include <compat/freebsd32/freebsd32_ipc.h>
 907 #include <compat/freebsd32/freebsd32_proto.h>
 908 #include <compat/freebsd32/freebsd32_signal.h>
 909 #include <compat/freebsd32/freebsd32_syscall.h>
 910 #include <compat/freebsd32/freebsd32_util.h>
 911
 912 static struct syscall_helper_data shm32_syscalls[] = {
 913         SYSCALL32_INIT_HELPER_COMPAT(shmat),
 914         SYSCALL32_INIT_HELPER_COMPAT(shmdt),
 915         SYSCALL32_INIT_HELPER_COMPAT(shmget),
 916         SYSCALL32_INIT_HELPER(freebsd32_shmsys),
 917         SYSCALL32_INIT_HELPER(freebsd32_shmctl),
 918 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 919     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 920         SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
 921 #endif
 922         SYSCALL_INIT_LAST
 923 };
 924 #endif
 925
 926 static int
 927 shminit(void)
 928 {
 929         struct prison *pr;
 930         void **rsv;
 931         int i, error;
 932         osd_method_t methods[PR_MAXMETHOD] = {
 933             [PR_METHOD_CHECK] =         shm_prison_check,
 934             [PR_METHOD_SET] =           shm_prison_set,
 935             [PR_METHOD_GET] =           shm_prison_get,
 936             [PR_METHOD_REMOVE] =        shm_prison_remove,
 937         };
 938
 939 #ifndef BURN_BRIDGES
 940         if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
 941                 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
 942 #endif
 943         if (shminfo.shmmax == SHMMAX) {
 944                 /* Initialize shmmax dealing with possible overflow. */
 945                 for (i = PAGE_SIZE; i != 0; i--) {
 946                         shminfo.shmmax = shminfo.shmall * i;
 947                         if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
 948                                 break;
 949                 }
 950         }
 951         shmalloced = shminfo.shmmni;
 952         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM,
 953             M_WAITOK|M_ZERO);
 954         for (i = 0; i < shmalloced; i++) {
 955                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 956                 shmsegs[i].u.shm_perm.seq = 0;
 957 #ifdef MAC
 958                 mac_sysvshm_init(&shmsegs[i]);
 959 #endif
 960         }
 961         shm_last_free = 0;
 962         shm_nused = 0;
 963         shm_committed = 0;
 964         sx_init(&sysvshmsx, "sysvshmsx");
 965 #ifndef SYSVSHM
 966         shmexit_hook = &shmexit_myhook;
 967         shmfork_hook = &shmfork_myhook;
 968 #endif
 969
 970         /* Set current prisons according to their allow.sysvipc. */
 971         shm_prison_slot = osd_jail_register(NULL, methods);
 972         rsv = osd_reserve(shm_prison_slot);
 973         prison_lock(&prison0);
 974         (void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
 975         prison_unlock(&prison0);
 976         rsv = NULL;
 977         sx_slock(&allprison_lock);
 978         TAILQ_FOREACH(pr, &allprison, pr_list) {
 979                 if (rsv == NULL)
 980                         rsv = osd_reserve(shm_prison_slot);
 981                 prison_lock(pr);
 982                 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
 983                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
 984                             &prison0);
 985                         rsv = NULL;
 986                 }
 987                 prison_unlock(pr);
 988         }
 989         if (rsv != NULL)
 990                 osd_free_reserved(rsv);
 991         sx_sunlock(&allprison_lock);
 992
 993         error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
 994         if (error != 0)
 995                 return (error);
 996 #ifdef COMPAT_FREEBSD32
 997         error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
 998         if (error != 0)
 999                 return (error);
1000 #endif
1001         return (0);
1002 }
1003
1004 static int
1005 shmunload(void)
1006 {
1007         int i;
1008
1009         if (shm_nused > 0)
1010                 return (EBUSY);
1011
1012 #ifdef COMPAT_FREEBSD32
1013         syscall32_helper_unregister(shm32_syscalls);
1014 #endif
1015         syscall_helper_unregister(shm_syscalls);
1016         if (shm_prison_slot != 0)
1017                 osd_jail_deregister(shm_prison_slot);
1018
1019         for (i = 0; i < shmalloced; i++) {
1020 #ifdef MAC
1021                 mac_sysvshm_destroy(&shmsegs[i]);
1022 #endif
1023                 /*
1024                  * Objects might be still mapped into the processes
1025                  * address spaces.  Actual free would happen on the
1026                  * last mapping destruction.
1027                  */
1028                 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1029                         vm_object_deallocate(shmsegs[i].object);
1030         }
1031         free(shmsegs, M_SHM);
1032 #ifndef SYSVSHM
1033         shmexit_hook = NULL;
1034         shmfork_hook = NULL;
1035 #endif
1036         sx_destroy(&sysvshmsx);
1037         return (0);
1038 }
1039
1040 static int
1041 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1042 {
1043         struct shmid_kernel tshmseg;
1044 #ifdef COMPAT_FREEBSD32
1045         struct shmid_kernel32 tshmseg32;
1046 #endif
1047         struct prison *pr, *rpr;
1048         void *outaddr;
1049         size_t outsize;
1050         int error, i;
1051
1052         SYSVSHM_LOCK();
1053         pr = req->td->td_ucred->cr_prison;
1054         rpr = shm_find_prison(req->td->td_ucred);
1055         error = 0;
1056         for (i = 0; i < shmalloced; i++) {
1057                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1058                     rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1059                         bzero(&tshmseg, sizeof(tshmseg));
1060                         tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1061                 } else {
1062                         tshmseg = shmsegs[i];
1063                         if (tshmseg.cred->cr_prison != pr)
1064                                 tshmseg.u.shm_perm.key = IPC_PRIVATE;
1065                 }
1066 #ifdef COMPAT_FREEBSD32
1067                 if (SV_CURPROC_FLAG(SV_ILP32)) {
1068                         bzero(&tshmseg32, sizeof(tshmseg32));
1069                         freebsd32_ipcperm_out(&tshmseg.u.shm_perm,
1070                             &tshmseg32.u.shm_perm);
1071                         CP(tshmseg, tshmseg32, u.shm_segsz);
1072                         CP(tshmseg, tshmseg32, u.shm_lpid);
1073                         CP(tshmseg, tshmseg32, u.shm_cpid);
1074                         CP(tshmseg, tshmseg32, u.shm_nattch);
1075                         CP(tshmseg, tshmseg32, u.shm_atime);
1076                         CP(tshmseg, tshmseg32, u.shm_dtime);
1077                         CP(tshmseg, tshmseg32, u.shm_ctime);
1078                         /* Don't copy object, label, or cred */
1079                         outaddr = &tshmseg32;
1080                         outsize = sizeof(tshmseg32);
1081                 } else
1082 #endif
1083                 {
1084                         tshmseg.object = NULL;
1085                         tshmseg.label = NULL;
1086                         tshmseg.cred = NULL;
1087                         outaddr = &tshmseg;
1088                         outsize = sizeof(tshmseg);
1089                 }
1090                 error = SYSCTL_OUT(req, outaddr, outsize);
1091                 if (error != 0)
1092                         break;
1093         }
1094         SYSVSHM_UNLOCK();
1095         return (error);
1096 }
1097
1098 static int
1099 shm_prison_check(void *obj, void *data)
1100 {
1101         struct prison *pr = obj;
1102         struct prison *prpr;
1103         struct vfsoptlist *opts = data;
1104         int error, jsys;
1105
1106         /*
1107          * sysvshm is a jailsys integer.
1108          * It must be "disable" if the parent jail is disabled.
1109          */
1110         error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1111         if (error != ENOENT) {
1112                 if (error != 0)
1113                         return (error);
1114                 switch (jsys) {
1115                 case JAIL_SYS_DISABLE:
1116                         break;
1117                 case JAIL_SYS_NEW:
1118                 case JAIL_SYS_INHERIT:
1119                         prison_lock(pr->pr_parent);
1120                         prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1121                         prison_unlock(pr->pr_parent);
1122                         if (prpr == NULL)
1123                                 return (EPERM);
1124                         break;
1125                 default:
1126                         return (EINVAL);
1127                 }
1128         }
1129
1130         return (0);
1131 }
1132
1133 static int
1134 shm_prison_set(void *obj, void *data)
1135 {
1136         struct prison *pr = obj;
1137         struct prison *tpr, *orpr, *nrpr, *trpr;
1138         struct vfsoptlist *opts = data;
1139         void *rsv;
1140         int jsys, descend;
1141
1142         /*
1143          * sysvshm controls which jail is the root of the associated segments
1144          * (this jail or same as the parent), or if the feature is available
1145          * at all.
1146          */
1147         if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1148                 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1149                     ? JAIL_SYS_INHERIT
1150                     : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1151                     ? JAIL_SYS_DISABLE
1152                     : -1;
1153         if (jsys == JAIL_SYS_DISABLE) {
1154                 prison_lock(pr);
1155                 orpr = osd_jail_get(pr, shm_prison_slot);
1156                 if (orpr != NULL)
1157                         osd_jail_del(pr, shm_prison_slot);
1158                 prison_unlock(pr);
1159                 if (orpr != NULL) {
1160                         if (orpr == pr)
1161                                 shm_prison_cleanup(pr);
1162                         /* Disable all child jails as well. */
1163                         FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1164                                 prison_lock(tpr);
1165                                 trpr = osd_jail_get(tpr, shm_prison_slot);
1166                                 if (trpr != NULL) {
1167                                         osd_jail_del(tpr, shm_prison_slot);
1168                                         prison_unlock(tpr);
1169                                         if (trpr == tpr)
1170                                                 shm_prison_cleanup(tpr);
1171                                 } else {
1172                                         prison_unlock(tpr);
1173                                         descend = 0;
1174                                 }
1175                         }
1176                 }
1177         } else if (jsys != -1) {
1178                 if (jsys == JAIL_SYS_NEW)
1179                         nrpr = pr;
1180                 else {
1181                         prison_lock(pr->pr_parent);
1182                         nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1183                         prison_unlock(pr->pr_parent);
1184                 }
1185                 rsv = osd_reserve(shm_prison_slot);
1186                 prison_lock(pr);
1187                 orpr = osd_jail_get(pr, shm_prison_slot);
1188                 if (orpr != nrpr)
1189                         (void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1190                             nrpr);
1191                 else
1192                         osd_free_reserved(rsv);
1193                 prison_unlock(pr);
1194                 if (orpr != nrpr) {
1195                         if (orpr == pr)
1196                                 shm_prison_cleanup(pr);
1197                         if (orpr != NULL) {
1198                                 /* Change child jails matching the old root, */
1199                                 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1200                                         prison_lock(tpr);
1201                                         trpr = osd_jail_get(tpr,
1202                                             shm_prison_slot);
1203                                         if (trpr == orpr) {
1204                                                 (void)osd_jail_set(tpr,
1205                                                     shm_prison_slot, nrpr);
1206                                                 prison_unlock(tpr);
1207                                                 if (trpr == tpr)
1208                                                         shm_prison_cleanup(tpr);
1209                                         } else {
1210                                                 prison_unlock(tpr);
1211                                                 descend = 0;
1212                                         }
1213                                 }
1214                         }
1215                 }
1216         }
1217
1218         return (0);
1219 }
1220
1221 static int
1222 shm_prison_get(void *obj, void *data)
1223 {
1224         struct prison *pr = obj;
1225         struct prison *rpr;
1226         struct vfsoptlist *opts = data;
1227         int error, jsys;
1228
1229         /* Set sysvshm based on the jail's root prison. */
1230         prison_lock(pr);
1231         rpr = osd_jail_get(pr, shm_prison_slot);
1232         prison_unlock(pr);
1233         jsys = rpr == NULL ? JAIL_SYS_DISABLE
1234             : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1235         error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1236         if (error == ENOENT)
1237                 error = 0;
1238         return (error);
1239 }
1240
1241 static int
1242 shm_prison_remove(void *obj, void *data __unused)
1243 {
1244         struct prison *pr = obj;
1245         struct prison *rpr;
1246
1247         SYSVSHM_LOCK();
1248         prison_lock(pr);
1249         rpr = osd_jail_get(pr, shm_prison_slot);
1250         prison_unlock(pr);
1251         if (rpr == pr)
1252                 shm_prison_cleanup(pr);
1253         SYSVSHM_UNLOCK();
1254         return (0);
1255 }
1256
1257 static void
1258 shm_prison_cleanup(struct prison *pr)
1259 {
1260         struct shmid_kernel *shmseg;
1261         int i;
1262
1263         /* Remove any segments that belong to this jail. */
1264         for (i = 0; i < shmalloced; i++) {
1265                 shmseg = &shmsegs[i];
1266                 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1267                     shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1268                         shm_remove(shmseg, i);
1269                 }
1270         }
1271 }
1272
1273 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1274
1275 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1276 struct oshmid_ds {
1277         struct  ipc_perm_old shm_perm;  /* operation perms */
1278         int     shm_segsz;              /* size of segment (bytes) */
1279         u_short shm_cpid;               /* pid, creator */
1280         u_short shm_lpid;               /* pid, last operation */
1281         short   shm_nattch;             /* no. of current attaches */
1282         time_t  shm_atime;              /* last attach time */
1283         time_t  shm_dtime;              /* last detach time */
1284         time_t  shm_ctime;              /* last change time */
1285         void    *shm_handle;            /* internal handle for shm segment */
1286 };
1287
1288 struct oshmctl_args {
1289         int shmid;
1290         int cmd;
1291         struct oshmid_ds *ubuf;
1292 };
1293
1294 static int
1295 oshmctl(struct thread *td, struct oshmctl_args *uap)
1296 {
1297 #ifdef COMPAT_43
1298         int error = 0;
1299         struct prison *rpr;
1300         struct shmid_kernel *shmseg;
1301         struct oshmid_ds outbuf;
1302
1303         rpr = shm_find_prison(td->td_ucred);
1304         if (rpr == NULL)
1305                 return (ENOSYS);
1306         if (uap->cmd != IPC_STAT) {
1307                 return (freebsd7_shmctl(td,
1308                     (struct freebsd7_shmctl_args *)uap));
1309         }
1310         SYSVSHM_LOCK();
1311         shmseg = shm_find_segment(rpr, uap->shmid, true);
1312         if (shmseg == NULL) {
1313                 SYSVSHM_UNLOCK();
1314                 return (EINVAL);
1315         }
1316         error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1317         if (error != 0) {
1318                 SYSVSHM_UNLOCK();
1319                 return (error);
1320         }
1321 #ifdef MAC
1322         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1323         if (error != 0) {
1324                 SYSVSHM_UNLOCK();
1325                 return (error);
1326         }
1327 #endif
1328         ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1329         outbuf.shm_segsz = shmseg->u.shm_segsz;
1330         outbuf.shm_cpid = shmseg->u.shm_cpid;
1331         outbuf.shm_lpid = shmseg->u.shm_lpid;
1332         outbuf.shm_nattch = shmseg->u.shm_nattch;
1333         outbuf.shm_atime = shmseg->u.shm_atime;
1334         outbuf.shm_dtime = shmseg->u.shm_dtime;
1335         outbuf.shm_ctime = shmseg->u.shm_ctime;
1336         outbuf.shm_handle = shmseg->object;
1337         SYSVSHM_UNLOCK();
1338         return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1339 #else
1340         return (EINVAL);
1341 #endif
1342 }
1343
1344 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1345 static sy_call_t *shmcalls[] = {
1346         (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1347         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1348         (sy_call_t *)freebsd7_shmctl
1349 };
1350
1351 #ifndef _SYS_SYSPROTO_H_
1352 /* XXX actually varargs. */
1353 struct shmsys_args {
1354         int     which;
1355         int     a2;
1356         int     a3;
1357         int     a4;
1358 };
1359 #endif
1360 int
1361 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1362 {
1363
1364         AUDIT_ARG_SVIPC_WHICH(uap->which);
1365         if (uap->which < 0 || uap->which >= nitems(shmcalls))
1366                 return (EINVAL);
1367         return ((*shmcalls[uap->which])(td, &uap->a2));
1368 }
1369
1370 #endif  /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1371
1372 #ifdef COMPAT_FREEBSD32
1373
1374 int
1375 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1376 {
1377
1378 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1379     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1380         AUDIT_ARG_SVIPC_WHICH(uap->which);
1381         switch (uap->which) {
1382         case 0: {       /* shmat */
1383                 struct shmat_args ap;
1384
1385                 ap.shmid = uap->a2;
1386                 ap.shmaddr = PTRIN(uap->a3);
1387                 ap.shmflg = uap->a4;
1388                 return (sysent[SYS_shmat].sy_call(td, &ap));
1389         }
1390         case 2: {       /* shmdt */
1391                 struct shmdt_args ap;
1392
1393                 ap.shmaddr = PTRIN(uap->a2);
1394                 return (sysent[SYS_shmdt].sy_call(td, &ap));
1395         }
1396         case 3: {       /* shmget */
1397                 struct shmget_args ap;
1398
1399                 ap.key = uap->a2;
1400                 ap.size = uap->a3;
1401                 ap.shmflg = uap->a4;
1402                 return (sysent[SYS_shmget].sy_call(td, &ap));
1403         }
1404         case 4: {       /* shmctl */
1405                 struct freebsd7_freebsd32_shmctl_args ap;
1406
1407                 ap.shmid = uap->a2;
1408                 ap.cmd = uap->a3;
1409                 ap.buf = PTRIN(uap->a4);
1410                 return (freebsd7_freebsd32_shmctl(td, &ap));
1411         }
1412         case 1:         /* oshmctl */
1413         default:
1414                 return (EINVAL);
1415         }
1416 #else
1417         return (nosys(td, NULL));
1418 #endif
1419 }
1420
1421 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1422     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1423 int
1424 freebsd7_freebsd32_shmctl(struct thread *td,
1425     struct freebsd7_freebsd32_shmctl_args *uap)
1426 {
1427         int error;
1428         union {
1429                 struct shmid_ds shmid_ds;
1430                 struct shm_info shm_info;
1431                 struct shminfo shminfo;
1432         } u;
1433         union {
1434                 struct shmid_ds32_old shmid_ds32;
1435                 struct shm_info32 shm_info32;
1436                 struct shminfo32 shminfo32;
1437         } u32;
1438         size_t sz;
1439
1440         if (uap->cmd == IPC_SET) {
1441                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1442                     sizeof(u32.shmid_ds32))))
1443                         goto done;
1444                 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1445                     &u.shmid_ds.shm_perm);
1446                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1447                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1448                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1449                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1450                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1451                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1452                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1453         }
1454
1455         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1456         if (error)
1457                 goto done;
1458
1459         /* Cases in which we need to copyout */
1460         switch (uap->cmd) {
1461         case IPC_INFO:
1462                 CP(u.shminfo, u32.shminfo32, shmmax);
1463                 CP(u.shminfo, u32.shminfo32, shmmin);
1464                 CP(u.shminfo, u32.shminfo32, shmmni);
1465                 CP(u.shminfo, u32.shminfo32, shmseg);
1466                 CP(u.shminfo, u32.shminfo32, shmall);
1467                 error = copyout(&u32.shminfo32, uap->buf,
1468                     sizeof(u32.shminfo32));
1469                 break;
1470         case SHM_INFO:
1471                 CP(u.shm_info, u32.shm_info32, used_ids);
1472                 CP(u.shm_info, u32.shm_info32, shm_rss);
1473                 CP(u.shm_info, u32.shm_info32, shm_tot);
1474                 CP(u.shm_info, u32.shm_info32, shm_swp);
1475                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1476                 CP(u.shm_info, u32.shm_info32, swap_successes);
1477                 error = copyout(&u32.shm_info32, uap->buf,
1478                     sizeof(u32.shm_info32));
1479                 break;
1480         case SHM_STAT:
1481         case IPC_STAT:
1482                 memset(&u32.shmid_ds32, 0, sizeof(u32.shmid_ds32));
1483                 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1484                     &u32.shmid_ds32.shm_perm);
1485                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1486                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1487                 else
1488                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1489                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1490                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1491                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1492                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1493                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1494                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1495                 u32.shmid_ds32.shm_internal = 0;
1496                 error = copyout(&u32.shmid_ds32, uap->buf,
1497                     sizeof(u32.shmid_ds32));
1498                 break;
1499         }
1500
1501 done:
1502         if (error) {
1503                 /* Invalidate the return value */
1504                 td->td_retval[0] = -1;
1505         }
1506         return (error);
1507 }
1508 #endif
1509
1510 int
1511 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1512 {
1513         int error;
1514         union {
1515                 struct shmid_ds shmid_ds;
1516                 struct shm_info shm_info;
1517                 struct shminfo shminfo;
1518         } u;
1519         union {
1520                 struct shmid_ds32 shmid_ds32;
1521                 struct shm_info32 shm_info32;
1522                 struct shminfo32 shminfo32;
1523         } u32;
1524         size_t sz;
1525
1526         if (uap->cmd == IPC_SET) {
1527                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1528                     sizeof(u32.shmid_ds32))))
1529                         goto done;
1530                 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1531                     &u.shmid_ds.shm_perm);
1532                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1533                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1534                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1535                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1536                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1537                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1538                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1539         }
1540
1541         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1542         if (error)
1543                 goto done;
1544
1545         /* Cases in which we need to copyout */
1546         switch (uap->cmd) {
1547         case IPC_INFO:
1548                 CP(u.shminfo, u32.shminfo32, shmmax);
1549                 CP(u.shminfo, u32.shminfo32, shmmin);
1550                 CP(u.shminfo, u32.shminfo32, shmmni);
1551                 CP(u.shminfo, u32.shminfo32, shmseg);
1552                 CP(u.shminfo, u32.shminfo32, shmall);
1553                 error = copyout(&u32.shminfo32, uap->buf,
1554                     sizeof(u32.shminfo32));
1555                 break;
1556         case SHM_INFO:
1557                 CP(u.shm_info, u32.shm_info32, used_ids);
1558                 CP(u.shm_info, u32.shm_info32, shm_rss);
1559                 CP(u.shm_info, u32.shm_info32, shm_tot);
1560                 CP(u.shm_info, u32.shm_info32, shm_swp);
1561                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1562                 CP(u.shm_info, u32.shm_info32, swap_successes);
1563                 error = copyout(&u32.shm_info32, uap->buf,
1564                     sizeof(u32.shm_info32));
1565                 break;
1566         case SHM_STAT:
1567         case IPC_STAT:
1568                 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1569                     &u32.shmid_ds32.shm_perm);
1570                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1571                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1572                 else
1573                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1574                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1575                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1576                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1577                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1578                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1579                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1580                 error = copyout(&u32.shmid_ds32, uap->buf,
1581                     sizeof(u32.shmid_ds32));
1582                 break;
1583         }
1584
1585 done:
1586         if (error) {
1587                 /* Invalidate the return value */
1588                 td->td_retval[0] = -1;
1589         }
1590         return (error);
1591 }
1592 #endif
1593
1594 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1595     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1596
1597 #ifndef _SYS_SYSPROTO_H_
1598 struct freebsd7_shmctl_args {
1599         int shmid;
1600         int cmd;
1601         struct shmid_ds_old *buf;
1602 };
1603 #endif
1604 int
1605 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1606 {
1607         int error;
1608         struct shmid_ds_old old;
1609         struct shmid_ds buf;
1610         size_t bufsz;
1611
1612         /*
1613          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1614          * Linux binaries.  If we see the call come through the FreeBSD ABI,
1615          * return an error back to the user since we do not to support this.
1616          */
1617         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1618             uap->cmd == SHM_STAT)
1619                 return (EINVAL);
1620
1621         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1622         if (uap->cmd == IPC_SET) {
1623                 if ((error = copyin(uap->buf, &old, sizeof(old))))
1624                         goto done;
1625                 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1626                 CP(old, buf, shm_segsz);
1627                 CP(old, buf, shm_lpid);
1628                 CP(old, buf, shm_cpid);
1629                 CP(old, buf, shm_nattch);
1630                 CP(old, buf, shm_atime);
1631                 CP(old, buf, shm_dtime);
1632                 CP(old, buf, shm_ctime);
1633         }
1634
1635         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1636         if (error)
1637                 goto done;
1638
1639         /* Cases in which we need to copyout */
1640         switch (uap->cmd) {
1641         case IPC_STAT:
1642                 memset(&old, 0, sizeof(old));
1643                 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1644                 if (buf.shm_segsz > INT_MAX)
1645                         old.shm_segsz = INT_MAX;
1646                 else
1647                         CP(buf, old, shm_segsz);
1648                 CP(buf, old, shm_lpid);
1649                 CP(buf, old, shm_cpid);
1650                 if (buf.shm_nattch > SHRT_MAX)
1651                         old.shm_nattch = SHRT_MAX;
1652                 else
1653                         CP(buf, old, shm_nattch);
1654                 CP(buf, old, shm_atime);
1655                 CP(buf, old, shm_dtime);
1656                 CP(buf, old, shm_ctime);
1657                 old.shm_internal = NULL;
1658                 error = copyout(&old, uap->buf, sizeof(old));
1659                 break;
1660         }
1661
1662 done:
1663         if (error) {
1664                 /* Invalidate the return value */
1665                 td->td_retval[0] = -1;
1666         }
1667         return (error);
1668 }
1669
1670 #endif  /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1671            COMPAT_FREEBSD7 */
1672
1673 static int
1674 sysvshm_modload(struct module *module, int cmd, void *arg)
1675 {
1676         int error = 0;
1677
1678         switch (cmd) {
1679         case MOD_LOAD:
1680                 error = shminit();
1681                 if (error != 0)
1682                         shmunload();
1683                 break;
1684         case MOD_UNLOAD:
1685                 error = shmunload();
1686                 break;
1687         case MOD_SHUTDOWN:
1688                 break;
1689         default:
1690                 error = EINVAL;
1691                 break;
1692         }
1693         return (error);
1694 }
1695
1696 static moduledata_t sysvshm_mod = {
1697         "sysvshm",
1698         &sysvshm_modload,
1699         NULL
1700 };
1701
1702 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1703 MODULE_VERSION(sysvshm, 1);