sys/kern/sysv_shm.c

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