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
 117 struct shmmap_state {
 118         vm_offset_t va;
 119         int shmid;
 120 };
 121
 122 static void shm_deallocate_segment(struct shmid_kernel *);
 123 static int shm_find_segment_by_key(key_t);
 124 static struct shmid_kernel *shm_find_segment(int, bool);
 125 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
 126 static void shmrealloc(void);
 127 static int shminit(void);
 128 static int sysvshm_modload(struct module *, int, void *);
 129 static int shmunload(void);
 130 static void shmexit_myhook(struct vmspace *vm);
 131 static void shmfork_myhook(struct proc *p1, struct proc *p2);
 132 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
 133
 134 /*
 135  * Tuneable values.
 136  */
 137 #ifndef SHMMAXPGS
 138 #define SHMMAXPGS       131072  /* Note: sysv shared memory is swap backed. */
 139 #endif
 140 #ifndef SHMMAX
 141 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 142 #endif
 143 #ifndef SHMMIN
 144 #define SHMMIN  1
 145 #endif
 146 #ifndef SHMMNI
 147 #define SHMMNI  192
 148 #endif
 149 #ifndef SHMSEG
 150 #define SHMSEG  128
 151 #endif
 152 #ifndef SHMALL
 153 #define SHMALL  (SHMMAXPGS)
 154 #endif
 155
 156 struct  shminfo shminfo = {
 157         SHMMAX,
 158         SHMMIN,
 159         SHMMNI,
 160         SHMSEG,
 161         SHMALL
 162 };
 163
 164 static int shm_use_phys;
 165 static int shm_allow_removed;
 166
 167 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 168     "Maximum shared memory segment size");
 169 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 170     "Minimum shared memory segment size");
 171 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
 172     "Number of shared memory identifiers");
 173 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
 174     "Number of segments per process");
 175 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 176     "Maximum number of pages available for shared memory");
 177 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
 178     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
 179 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
 180     &shm_allow_removed, 0,
 181     "Enable/Disable attachment to attached segments marked for removal");
 182 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
 183     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
 184     "Current number of shared memory segments allocated");
 185
 186 static struct sx sysvshmsx;
 187 #define SYSVSHM_LOCK()          sx_xlock(&sysvshmsx)
 188 #define SYSVSHM_UNLOCK()        sx_xunlock(&sysvshmsx)
 189 #define SYSVSHM_ASSERT_LOCKED() sx_assert(&sysvshmsx, SA_XLOCKED)
 190
 191 static int
 192 shm_find_segment_by_key(key_t key)
 193 {
 194         int i;
 195
 196         for (i = 0; i < shmalloced; i++)
 197                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
 198                     shmsegs[i].u.shm_perm.key == key)
 199                         return (i);
 200         return (-1);
 201 }
 202
 203 /*
 204  * Finds segment either by shmid if is_shmid is true, or by segnum if
 205  * is_shmid is false.
 206  */
 207 static struct shmid_kernel *
 208 shm_find_segment(int arg, bool is_shmid)
 209 {
 210         struct shmid_kernel *shmseg;
 211         int segnum;
 212
 213         segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
 214         if (segnum < 0 || segnum >= shmalloced)
 215                 return (NULL);
 216         shmseg = &shmsegs[segnum];
 217         if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 218             (!shm_allow_removed &&
 219              (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 220             (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)))
 221                 return (NULL);
 222         return (shmseg);
 223 }
 224
 225 static void
 226 shm_deallocate_segment(struct shmid_kernel *shmseg)
 227 {
 228         vm_size_t size;
 229
 230         SYSVSHM_ASSERT_LOCKED();
 231
 232         vm_object_deallocate(shmseg->object);
 233         shmseg->object = NULL;
 234         size = round_page(shmseg->u.shm_segsz);
 235         shm_committed -= btoc(size);
 236         shm_nused--;
 237         shmseg->u.shm_perm.mode = SHMSEG_FREE;
 238 #ifdef MAC
 239         mac_sysvshm_cleanup(shmseg);
 240 #endif
 241         racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
 242         racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
 243         crfree(shmseg->cred);
 244         shmseg->cred = NULL;
 245 }
 246
 247 static int
 248 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 249 {
 250         struct shmid_kernel *shmseg;
 251         int segnum, result;
 252         vm_size_t size;
 253
 254         SYSVSHM_ASSERT_LOCKED();
 255         segnum = IPCID_TO_IX(shmmap_s->shmid);
 256         KASSERT(segnum >= 0 && segnum < shmalloced,
 257             ("segnum %d shmalloced %d", segnum, shmalloced));
 258
 259         shmseg = &shmsegs[segnum];
 260         size = round_page(shmseg->u.shm_segsz);
 261         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 262         if (result != KERN_SUCCESS)
 263                 return (EINVAL);
 264         shmmap_s->shmid = -1;
 265         shmseg->u.shm_dtime = time_second;
 266         if ((--shmseg->u.shm_nattch <= 0) &&
 267             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
 268                 shm_deallocate_segment(shmseg);
 269                 shm_last_free = segnum;
 270         }
 271         return (0);
 272 }
 273
 274 static int
 275 kern_shmdt_locked(struct thread *td, const void *shmaddr)
 276 {
 277         struct proc *p = td->td_proc;
 278         struct shmmap_state *shmmap_s;
 279 #ifdef MAC
 280         struct shmid_kernel *shmsegptr;
 281 #endif
 282         int error, i;
 283
 284         SYSVSHM_ASSERT_LOCKED();
 285         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 286                 return (ENOSYS);
 287         shmmap_s = p->p_vmspace->vm_shm;
 288         if (shmmap_s == NULL)
 289                 return (EINVAL);
 290         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 291                 if (shmmap_s->shmid != -1 &&
 292                     shmmap_s->va == (vm_offset_t)shmaddr) {
 293                         break;
 294                 }
 295         }
 296         if (i == shminfo.shmseg)
 297                 return (EINVAL);
 298 #ifdef MAC
 299         shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
 300         error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
 301         if (error != 0)
 302                 return (error);
 303 #endif
 304         error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 305         return (error);
 306 }
 307
 308 #ifndef _SYS_SYSPROTO_H_
 309 struct shmdt_args {
 310         const void *shmaddr;
 311 };
 312 #endif
 313 int
 314 sys_shmdt(struct thread *td, struct shmdt_args *uap)
 315 {
 316         int error;
 317
 318         SYSVSHM_LOCK();
 319         error = kern_shmdt_locked(td, uap->shmaddr);
 320         SYSVSHM_UNLOCK();
 321         return (error);
 322 }
 323
 324 static int
 325 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
 326     int shmflg)
 327 {
 328         struct proc *p = td->td_proc;
 329         struct shmid_kernel *shmseg;
 330         struct shmmap_state *shmmap_s;
 331         vm_offset_t attach_va;
 332         vm_prot_t prot;
 333         vm_size_t size;
 334         int error, i, rv;
 335
 336         SYSVSHM_ASSERT_LOCKED();
 337         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 338                 return (ENOSYS);
 339         shmmap_s = p->p_vmspace->vm_shm;
 340         if (shmmap_s == NULL) {
 341                 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
 342                     M_SHM, M_WAITOK);
 343                 for (i = 0; i < shminfo.shmseg; i++)
 344                         shmmap_s[i].shmid = -1;
 345                 KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
 346                 p->p_vmspace->vm_shm = shmmap_s;
 347         }
 348         shmseg = shm_find_segment(shmid, true);
 349         if (shmseg == NULL)
 350                 return (EINVAL);
 351         error = ipcperm(td, &shmseg->u.shm_perm,
 352             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 353         if (error != 0)
 354                 return (error);
 355 #ifdef MAC
 356         error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
 357         if (error != 0)
 358                 return (error);
 359 #endif
 360         for (i = 0; i < shminfo.shmseg; i++) {
 361                 if (shmmap_s->shmid == -1)
 362                         break;
 363                 shmmap_s++;
 364         }
 365         if (i >= shminfo.shmseg)
 366                 return (EMFILE);
 367         size = round_page(shmseg->u.shm_segsz);
 368         prot = VM_PROT_READ;
 369         if ((shmflg & SHM_RDONLY) == 0)
 370                 prot |= VM_PROT_WRITE;
 371         if (shmaddr != NULL) {
 372                 if ((shmflg & SHM_RND) != 0)
 373                         attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
 374                 else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
 375                         attach_va = (vm_offset_t)shmaddr;
 376                 else
 377                         return (EINVAL);
 378         } else {
 379                 /*
 380                  * This is just a hint to vm_map_find() about where to
 381                  * put it.
 382                  */
 383                 PROC_LOCK(p);
 384                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
 385                     lim_max(p, RLIMIT_DATA));
 386                 PROC_UNLOCK(p);
 387         }
 388
 389         vm_object_reference(shmseg->object);
 390         rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object,
 391             0, &attach_va, size, 0, shmaddr != NULL ? VMFS_NO_SPACE :
 392             VMFS_OPTIMAL_SPACE, prot, prot, MAP_INHERIT_SHARE);
 393         if (rv != KERN_SUCCESS) {
 394                 vm_object_deallocate(shmseg->object);
 395                 return (ENOMEM);
 396         }
 397
 398         shmmap_s->va = attach_va;
 399         shmmap_s->shmid = shmid;
 400         shmseg->u.shm_lpid = p->p_pid;
 401         shmseg->u.shm_atime = time_second;
 402         shmseg->u.shm_nattch++;
 403         td->td_retval[0] = attach_va;
 404         return (error);
 405 }
 406
 407 int
 408 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
 409 {
 410         int error;
 411
 412         SYSVSHM_LOCK();
 413         error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
 414         SYSVSHM_UNLOCK();
 415         return (error);
 416 }
 417
 418 #ifndef _SYS_SYSPROTO_H_
 419 struct shmat_args {
 420         int shmid;
 421         const void *shmaddr;
 422         int shmflg;
 423 };
 424 #endif
 425 int
 426 sys_shmat(struct thread *td, struct shmat_args *uap)
 427 {
 428
 429         return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
 430 }
 431
 432 static int
 433 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
 434     size_t *bufsz)
 435 {
 436         struct shmid_kernel *shmseg;
 437         struct shmid_ds *shmidp;
 438         struct shm_info shm_info;
 439         int error;
 440
 441         SYSVSHM_ASSERT_LOCKED();
 442
 443         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 444                 return (ENOSYS);
 445
 446         error = 0;
 447         switch (cmd) {
 448         /*
 449          * It is possible that kern_shmctl is being called from the Linux ABI
 450          * layer, in which case, we will need to implement IPC_INFO.  It should
 451          * be noted that other shmctl calls will be funneled through here for
 452          * Linix binaries as well.
 453          *
 454          * NB: The Linux ABI layer will convert this data to structure(s) more
 455          * consistent with the Linux ABI.
 456          */
 457         case IPC_INFO:
 458                 memcpy(buf, &shminfo, sizeof(shminfo));
 459                 if (bufsz)
 460                         *bufsz = sizeof(shminfo);
 461                 td->td_retval[0] = shmalloced;
 462                 return (0);
 463         case SHM_INFO: {
 464                 shm_info.used_ids = shm_nused;
 465                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 466                 shm_info.shm_tot = 0;   /*XXX where to get from ? */
 467                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 468                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 469                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 470                 memcpy(buf, &shm_info, sizeof(shm_info));
 471                 if (bufsz != NULL)
 472                         *bufsz = sizeof(shm_info);
 473                 td->td_retval[0] = shmalloced;
 474                 return (0);
 475         }
 476         }
 477         shmseg = shm_find_segment(shmid, cmd != SHM_STAT);
 478         if (shmseg == NULL)
 479                 return (EINVAL);
 480 #ifdef MAC
 481         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
 482         if (error != 0)
 483                 return (error);
 484 #endif
 485         switch (cmd) {
 486         case SHM_STAT:
 487         case IPC_STAT:
 488                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 489                 if (error != 0)
 490                         return (error);
 491                 memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
 492                 if (bufsz != NULL)
 493                         *bufsz = sizeof(struct shmid_ds);
 494                 if (cmd == SHM_STAT) {
 495                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
 496                             shmseg->u.shm_perm);
 497                 }
 498                 break;
 499         case IPC_SET:
 500                 shmidp = (struct shmid_ds *)buf;
 501                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 502                 if (error != 0)
 503                         return (error);
 504                 shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
 505                 shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
 506                 shmseg->u.shm_perm.mode =
 507                     (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
 508                     (shmidp->shm_perm.mode & ACCESSPERMS);
 509                 shmseg->u.shm_ctime = time_second;
 510                 break;
 511         case IPC_RMID:
 512                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 513                 if (error != 0)
 514                         return (error);
 515                 shmseg->u.shm_perm.key = IPC_PRIVATE;
 516                 shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
 517                 if (shmseg->u.shm_nattch <= 0) {
 518                         shm_deallocate_segment(shmseg);
 519                         shm_last_free = IPCID_TO_IX(shmid);
 520                 }
 521                 break;
 522 #if 0
 523         case SHM_LOCK:
 524         case SHM_UNLOCK:
 525 #endif
 526         default:
 527                 error = EINVAL;
 528                 break;
 529         }
 530         return (error);
 531 }
 532
 533 int
 534 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
 535 {
 536         int error;
 537
 538         SYSVSHM_LOCK();
 539         error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
 540         SYSVSHM_UNLOCK();
 541         return (error);
 542 }
 543
 544
 545 #ifndef _SYS_SYSPROTO_H_
 546 struct shmctl_args {
 547         int shmid;
 548         int cmd;
 549         struct shmid_ds *buf;
 550 };
 551 #endif
 552 int
 553 sys_shmctl(struct thread *td, struct shmctl_args *uap)
 554 {
 555         int error = 0;
 556         struct shmid_ds buf;
 557         size_t bufsz;
 558
 559         /*
 560          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
 561          * Linux binaries.  If we see the call come through the FreeBSD ABI,
 562          * return an error back to the user since we do not to support this.
 563          */
 564         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
 565             uap->cmd == SHM_STAT)
 566                 return (EINVAL);
 567
 568         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
 569         if (uap->cmd == IPC_SET) {
 570                 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
 571                         goto done;
 572         }
 573
 574         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
 575         if (error)
 576                 goto done;
 577
 578         /* Cases in which we need to copyout */
 579         switch (uap->cmd) {
 580         case IPC_STAT:
 581                 error = copyout(&buf, uap->buf, bufsz);
 582                 break;
 583         }
 584
 585 done:
 586         if (error) {
 587                 /* Invalidate the return value */
 588                 td->td_retval[0] = -1;
 589         }
 590         return (error);
 591 }
 592
 593
 594 static int
 595 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
 596     int segnum)
 597 {
 598         struct shmid_kernel *shmseg;
 599 #ifdef MAC
 600         int error;
 601 #endif
 602
 603         SYSVSHM_ASSERT_LOCKED();
 604         KASSERT(segnum >= 0 && segnum < shmalloced,
 605             ("segnum %d shmalloced %d", segnum, shmalloced));
 606         shmseg = &shmsegs[segnum];
 607         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 608                 return (EEXIST);
 609 #ifdef MAC
 610         error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
 611         if (error != 0)
 612                 return (error);
 613 #endif
 614         if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
 615                 return (EINVAL);
 616         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 617         return (0);
 618 }
 619
 620 static int
 621 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
 622 {
 623         struct ucred *cred = td->td_ucred;
 624         struct shmid_kernel *shmseg;
 625         vm_object_t shm_object;
 626         int i, segnum;
 627         size_t size;
 628
 629         SYSVSHM_ASSERT_LOCKED();
 630
 631         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 632                 return (EINVAL);
 633         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 634                 return (ENOSPC);
 635         size = round_page(uap->size);
 636         if (shm_committed + btoc(size) > shminfo.shmall)
 637                 return (ENOMEM);
 638         if (shm_last_free < 0) {
 639                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 640                 for (i = 0; i < shmalloced; i++)
 641                         if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
 642                                 break;
 643                 if (i == shmalloced)
 644                         return (ENOSPC);
 645                 segnum = i;
 646         } else  {
 647                 segnum = shm_last_free;
 648                 shm_last_free = -1;
 649         }
 650         KASSERT(segnum >= 0 && segnum < shmalloced,
 651             ("segnum %d shmalloced %d", segnum, shmalloced));
 652         shmseg = &shmsegs[segnum];
 653 #ifdef RACCT
 654         if (racct_enable) {
 655                 PROC_LOCK(td->td_proc);
 656                 if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
 657                         PROC_UNLOCK(td->td_proc);
 658                         return (ENOSPC);
 659                 }
 660                 if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
 661                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 662                         PROC_UNLOCK(td->td_proc);
 663                         return (ENOMEM);
 664                 }
 665                 PROC_UNLOCK(td->td_proc);
 666         }
 667 #endif
 668
 669         /*
 670          * We make sure that we have allocated a pager before we need
 671          * to.
 672          */
 673         shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
 674             0, size, VM_PROT_DEFAULT, 0, cred);
 675         if (shm_object == NULL) {
 676 #ifdef RACCT
 677                 if (racct_enable) {
 678                         PROC_LOCK(td->td_proc);
 679                         racct_sub(td->td_proc, RACCT_NSHM, 1);
 680                         racct_sub(td->td_proc, RACCT_SHMSIZE, size);
 681                         PROC_UNLOCK(td->td_proc);
 682                 }
 683 #endif
 684                 return (ENOMEM);
 685         }
 686         shm_object->pg_color = 0;
 687         VM_OBJECT_WLOCK(shm_object);
 688         vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
 689         vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
 690         VM_OBJECT_WUNLOCK(shm_object);
 691
 692         shmseg->object = shm_object;
 693         shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
 694         shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
 695         shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 696         shmseg->u.shm_perm.key = uap->key;
 697         shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
 698         shmseg->cred = crhold(cred);
 699         shmseg->u.shm_segsz = uap->size;
 700         shmseg->u.shm_cpid = td->td_proc->p_pid;
 701         shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
 702         shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
 703 #ifdef MAC
 704         mac_sysvshm_create(cred, shmseg);
 705 #endif
 706         shmseg->u.shm_ctime = time_second;
 707         shm_committed += btoc(size);
 708         shm_nused++;
 709         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 710
 711         return (0);
 712 }
 713
 714 #ifndef _SYS_SYSPROTO_H_
 715 struct shmget_args {
 716         key_t key;
 717         size_t size;
 718         int shmflg;
 719 };
 720 #endif
 721 int
 722 sys_shmget(struct thread *td, struct shmget_args *uap)
 723 {
 724         int segnum, mode;
 725         int error;
 726
 727         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 728                 return (ENOSYS);
 729         mode = uap->shmflg & ACCESSPERMS;
 730         SYSVSHM_LOCK();
 731         if (uap->key == IPC_PRIVATE) {
 732                 error = shmget_allocate_segment(td, uap, mode);
 733         } else {
 734                 segnum = shm_find_segment_by_key(uap->key);
 735                 if (segnum >= 0)
 736                         error = shmget_existing(td, uap, mode, segnum);
 737                 else if ((uap->shmflg & IPC_CREAT) == 0)
 738                         error = ENOENT;
 739                 else
 740                         error = shmget_allocate_segment(td, uap, mode);
 741         }
 742         SYSVSHM_UNLOCK();
 743         return (error);
 744 }
 745
 746 static void
 747 shmfork_myhook(struct proc *p1, struct proc *p2)
 748 {
 749         struct shmmap_state *shmmap_s;
 750         size_t size;
 751         int i;
 752
 753         SYSVSHM_LOCK();
 754         size = shminfo.shmseg * sizeof(struct shmmap_state);
 755         shmmap_s = malloc(size, M_SHM, M_WAITOK);
 756         bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
 757         p2->p_vmspace->vm_shm = shmmap_s;
 758         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 759                 if (shmmap_s->shmid != -1) {
 760                         KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
 761                             IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
 762                             ("segnum %d shmalloced %d",
 763                             IPCID_TO_IX(shmmap_s->shmid), shmalloced));
 764                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
 765                 }
 766         }
 767         SYSVSHM_UNLOCK();
 768 }
 769
 770 static void
 771 shmexit_myhook(struct vmspace *vm)
 772 {
 773         struct shmmap_state *base, *shm;
 774         int i;
 775
 776         base = vm->vm_shm;
 777         if (base != NULL) {
 778                 vm->vm_shm = NULL;
 779                 SYSVSHM_LOCK();
 780                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 781                         if (shm->shmid != -1)
 782                                 shm_delete_mapping(vm, shm);
 783                 }
 784                 SYSVSHM_UNLOCK();
 785                 free(base, M_SHM);
 786         }
 787 }
 788
 789 static void
 790 shmrealloc(void)
 791 {
 792         struct shmid_kernel *newsegs;
 793         int i;
 794
 795         SYSVSHM_ASSERT_LOCKED();
 796
 797         if (shmalloced >= shminfo.shmmni)
 798                 return;
 799
 800         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 801         for (i = 0; i < shmalloced; i++)
 802                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 803         for (; i < shminfo.shmmni; i++) {
 804                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 805                 shmsegs[i].u.shm_perm.seq = 0;
 806 #ifdef MAC
 807                 mac_sysvshm_init(&shmsegs[i]);
 808 #endif
 809         }
 810         free(shmsegs, M_SHM);
 811         shmsegs = newsegs;
 812         shmalloced = shminfo.shmmni;
 813 }
 814
 815 static struct syscall_helper_data shm_syscalls[] = {
 816         SYSCALL_INIT_HELPER(shmat),
 817         SYSCALL_INIT_HELPER(shmctl),
 818         SYSCALL_INIT_HELPER(shmdt),
 819         SYSCALL_INIT_HELPER(shmget),
 820 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 821     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 822         SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
 823 #endif
 824 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 825         SYSCALL_INIT_HELPER(shmsys),
 826 #endif
 827         SYSCALL_INIT_LAST
 828 };
 829
 830 #ifdef COMPAT_FREEBSD32
 831 #include <compat/freebsd32/freebsd32.h>
 832 #include <compat/freebsd32/freebsd32_ipc.h>
 833 #include <compat/freebsd32/freebsd32_proto.h>
 834 #include <compat/freebsd32/freebsd32_signal.h>
 835 #include <compat/freebsd32/freebsd32_syscall.h>
 836 #include <compat/freebsd32/freebsd32_util.h>
 837
 838 static struct syscall_helper_data shm32_syscalls[] = {
 839         SYSCALL32_INIT_HELPER_COMPAT(shmat),
 840         SYSCALL32_INIT_HELPER_COMPAT(shmdt),
 841         SYSCALL32_INIT_HELPER_COMPAT(shmget),
 842         SYSCALL32_INIT_HELPER(freebsd32_shmsys),
 843         SYSCALL32_INIT_HELPER(freebsd32_shmctl),
 844 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 845     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 846         SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
 847 #endif
 848         SYSCALL_INIT_LAST
 849 };
 850 #endif
 851
 852 static int
 853 shminit(void)
 854 {
 855         int i, error;
 856
 857 #ifndef BURN_BRIDGES
 858         if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
 859                 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
 860 #endif
 861         TUNABLE_ULONG_FETCH("kern.ipc.shmall", &shminfo.shmall);
 862         if (!TUNABLE_ULONG_FETCH("kern.ipc.shmmax", &shminfo.shmmax)) {
 863                 /* Initialize shmmax dealing with possible overflow. */
 864                 for (i = PAGE_SIZE; i > 0; i--) {
 865                         shminfo.shmmax = shminfo.shmall * i;
 866                         if (shminfo.shmmax >= shminfo.shmall)
 867                                 break;
 868                 }
 869         }
 870         TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
 871         TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
 872         TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
 873         TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
 874
 875         shmalloced = shminfo.shmmni;
 876         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 877         for (i = 0; i < shmalloced; i++) {
 878                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 879                 shmsegs[i].u.shm_perm.seq = 0;
 880 #ifdef MAC
 881                 mac_sysvshm_init(&shmsegs[i]);
 882 #endif
 883         }
 884         shm_last_free = 0;
 885         shm_nused = 0;
 886         shm_committed = 0;
 887         sx_init(&sysvshmsx, "sysvshmsx");
 888         shmexit_hook = &shmexit_myhook;
 889         shmfork_hook = &shmfork_myhook;
 890
 891         error = syscall_helper_register(shm_syscalls);
 892         if (error != 0)
 893                 return (error);
 894 #ifdef COMPAT_FREEBSD32
 895         error = syscall32_helper_register(shm32_syscalls);
 896         if (error != 0)
 897                 return (error);
 898 #endif
 899         return (0);
 900 }
 901
 902 static int
 903 shmunload(void)
 904 {
 905         int i;
 906
 907         if (shm_nused > 0)
 908                 return (EBUSY);
 909
 910 #ifdef COMPAT_FREEBSD32
 911         syscall32_helper_unregister(shm32_syscalls);
 912 #endif
 913         syscall_helper_unregister(shm_syscalls);
 914
 915         for (i = 0; i < shmalloced; i++) {
 916 #ifdef MAC
 917                 mac_sysvshm_destroy(&shmsegs[i]);
 918 #endif
 919                 /*
 920                  * Objects might be still mapped into the processes
 921                  * address spaces.  Actual free would happen on the
 922                  * last mapping destruction.
 923                  */
 924                 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
 925                         vm_object_deallocate(shmsegs[i].object);
 926         }
 927         free(shmsegs, M_SHM);
 928         shmexit_hook = NULL;
 929         shmfork_hook = NULL;
 930         sx_destroy(&sysvshmsx);
 931         return (0);
 932 }
 933
 934 static int
 935 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
 936 {
 937         int error;
 938
 939         SYSVSHM_LOCK();
 940         error = SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0]));
 941         SYSVSHM_UNLOCK();
 942         return (error);
 943 }
 944
 945 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 946 struct oshmid_ds {
 947         struct  ipc_perm_old shm_perm;  /* operation perms */
 948         int     shm_segsz;              /* size of segment (bytes) */
 949         u_short shm_cpid;               /* pid, creator */
 950         u_short shm_lpid;               /* pid, last operation */
 951         short   shm_nattch;             /* no. of current attaches */
 952         time_t  shm_atime;              /* last attach time */
 953         time_t  shm_dtime;              /* last detach time */
 954         time_t  shm_ctime;              /* last change time */
 955         void    *shm_handle;            /* internal handle for shm segment */
 956 };
 957
 958 struct oshmctl_args {
 959         int shmid;
 960         int cmd;
 961         struct oshmid_ds *ubuf;
 962 };
 963
 964 static int
 965 oshmctl(struct thread *td, struct oshmctl_args *uap)
 966 {
 967 #ifdef COMPAT_43
 968         int error = 0;
 969         struct shmid_kernel *shmseg;
 970         struct oshmid_ds outbuf;
 971
 972         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 973                 return (ENOSYS);
 974         if (uap->cmd != IPC_STAT) {
 975                 return (freebsd7_shmctl(td,
 976                     (struct freebsd7_shmctl_args *)uap));
 977         }
 978         SYSVSHM_LOCK();
 979         shmseg = shm_find_segment(uap->shmid, true);
 980         if (shmseg == NULL) {
 981                 SYSVSHM_UNLOCK();
 982                 return (EINVAL);
 983         }
 984         error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 985         if (error != 0) {
 986                 SYSVSHM_UNLOCK();
 987                 return (error);
 988         }
 989 #ifdef MAC
 990         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
 991         if (error != 0) {
 992                 SYSVSHM_UNLOCK();
 993                 return (error);
 994         }
 995 #endif
 996         ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
 997         outbuf.shm_segsz = shmseg->u.shm_segsz;
 998         outbuf.shm_cpid = shmseg->u.shm_cpid;
 999         outbuf.shm_lpid = shmseg->u.shm_lpid;
1000         outbuf.shm_nattch = shmseg->u.shm_nattch;
1001         outbuf.shm_atime = shmseg->u.shm_atime;
1002         outbuf.shm_dtime = shmseg->u.shm_dtime;
1003         outbuf.shm_ctime = shmseg->u.shm_ctime;
1004         outbuf.shm_handle = shmseg->object;
1005         SYSVSHM_UNLOCK();
1006         error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
1007         return (error);
1008 #else
1009         return (EINVAL);
1010 #endif
1011 }
1012
1013 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1014 static sy_call_t *shmcalls[] = {
1015         (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1016         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1017         (sy_call_t *)freebsd7_shmctl
1018 };
1019
1020 #ifndef _SYS_SYSPROTO_H_
1021 /* XXX actually varargs. */
1022 struct shmsys_args {
1023         int     which;
1024         int     a2;
1025         int     a3;
1026         int     a4;
1027 };
1028 #endif
1029 int
1030 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1031 {
1032         int error;
1033
1034         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1035                 return (ENOSYS);
1036         if (uap->which < 0 || uap->which >= nitems(shmcalls))
1037                 return (EINVAL);
1038         error = (*shmcalls[uap->which])(td, &uap->a2);
1039         return (error);
1040 }
1041
1042 #endif  /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1043
1044 #ifdef COMPAT_FREEBSD32
1045
1046 int
1047 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1048 {
1049
1050 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1051     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1052         switch (uap->which) {
1053         case 0: {       /* shmat */
1054                 struct shmat_args ap;
1055
1056                 ap.shmid = uap->a2;
1057                 ap.shmaddr = PTRIN(uap->a3);
1058                 ap.shmflg = uap->a4;
1059                 return (sysent[SYS_shmat].sy_call(td, &ap));
1060         }
1061         case 2: {       /* shmdt */
1062                 struct shmdt_args ap;
1063
1064                 ap.shmaddr = PTRIN(uap->a2);
1065                 return (sysent[SYS_shmdt].sy_call(td, &ap));
1066         }
1067         case 3: {       /* shmget */
1068                 struct shmget_args ap;
1069
1070                 ap.key = uap->a2;
1071                 ap.size = uap->a3;
1072                 ap.shmflg = uap->a4;
1073                 return (sysent[SYS_shmget].sy_call(td, &ap));
1074         }
1075         case 4: {       /* shmctl */
1076                 struct freebsd7_freebsd32_shmctl_args ap;
1077
1078                 ap.shmid = uap->a2;
1079                 ap.cmd = uap->a3;
1080                 ap.buf = PTRIN(uap->a4);
1081                 return (freebsd7_freebsd32_shmctl(td, &ap));
1082         }
1083         case 1:         /* oshmctl */
1084         default:
1085                 return (EINVAL);
1086         }
1087 #else
1088         return (nosys(td, NULL));
1089 #endif
1090 }
1091
1092 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1093     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1094 int
1095 freebsd7_freebsd32_shmctl(struct thread *td,
1096     struct freebsd7_freebsd32_shmctl_args *uap)
1097 {
1098         int error = 0;
1099         union {
1100                 struct shmid_ds shmid_ds;
1101                 struct shm_info shm_info;
1102                 struct shminfo shminfo;
1103         } u;
1104         union {
1105                 struct shmid_ds32_old shmid_ds32;
1106                 struct shm_info32 shm_info32;
1107                 struct shminfo32 shminfo32;
1108         } u32;
1109         size_t sz;
1110
1111         if (uap->cmd == IPC_SET) {
1112                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1113                     sizeof(u32.shmid_ds32))))
1114                         goto done;
1115                 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1116                     &u.shmid_ds.shm_perm);
1117                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1118                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1119                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1120                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1121                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1122                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1123                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1124         }
1125
1126         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1127         if (error)
1128                 goto done;
1129
1130         /* Cases in which we need to copyout */
1131         switch (uap->cmd) {
1132         case IPC_INFO:
1133                 CP(u.shminfo, u32.shminfo32, shmmax);
1134                 CP(u.shminfo, u32.shminfo32, shmmin);
1135                 CP(u.shminfo, u32.shminfo32, shmmni);
1136                 CP(u.shminfo, u32.shminfo32, shmseg);
1137                 CP(u.shminfo, u32.shminfo32, shmall);
1138                 error = copyout(&u32.shminfo32, uap->buf,
1139                     sizeof(u32.shminfo32));
1140                 break;
1141         case SHM_INFO:
1142                 CP(u.shm_info, u32.shm_info32, used_ids);
1143                 CP(u.shm_info, u32.shm_info32, shm_rss);
1144                 CP(u.shm_info, u32.shm_info32, shm_tot);
1145                 CP(u.shm_info, u32.shm_info32, shm_swp);
1146                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1147                 CP(u.shm_info, u32.shm_info32, swap_successes);
1148                 error = copyout(&u32.shm_info32, uap->buf,
1149                     sizeof(u32.shm_info32));
1150                 break;
1151         case SHM_STAT:
1152         case IPC_STAT:
1153                 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1154                     &u32.shmid_ds32.shm_perm);
1155                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1156                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1157                 else
1158                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1159                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1160                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1161                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1162                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1163                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1164                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1165                 u32.shmid_ds32.shm_internal = 0;
1166                 error = copyout(&u32.shmid_ds32, uap->buf,
1167                     sizeof(u32.shmid_ds32));
1168                 break;
1169         }
1170
1171 done:
1172         if (error) {
1173                 /* Invalidate the return value */
1174                 td->td_retval[0] = -1;
1175         }
1176         return (error);
1177 }
1178 #endif
1179
1180 int
1181 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1182 {
1183         int error = 0;
1184         union {
1185                 struct shmid_ds shmid_ds;
1186                 struct shm_info shm_info;
1187                 struct shminfo shminfo;
1188         } u;
1189         union {
1190                 struct shmid_ds32 shmid_ds32;
1191                 struct shm_info32 shm_info32;
1192                 struct shminfo32 shminfo32;
1193         } u32;
1194         size_t sz;
1195
1196         if (uap->cmd == IPC_SET) {
1197                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1198                     sizeof(u32.shmid_ds32))))
1199                         goto done;
1200                 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1201                     &u.shmid_ds.shm_perm);
1202                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1203                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1204                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1205                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1206                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1207                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1208                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1209         }
1210
1211         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1212         if (error)
1213                 goto done;
1214
1215         /* Cases in which we need to copyout */
1216         switch (uap->cmd) {
1217         case IPC_INFO:
1218                 CP(u.shminfo, u32.shminfo32, shmmax);
1219                 CP(u.shminfo, u32.shminfo32, shmmin);
1220                 CP(u.shminfo, u32.shminfo32, shmmni);
1221                 CP(u.shminfo, u32.shminfo32, shmseg);
1222                 CP(u.shminfo, u32.shminfo32, shmall);
1223                 error = copyout(&u32.shminfo32, uap->buf,
1224                     sizeof(u32.shminfo32));
1225                 break;
1226         case SHM_INFO:
1227                 CP(u.shm_info, u32.shm_info32, used_ids);
1228                 CP(u.shm_info, u32.shm_info32, shm_rss);
1229                 CP(u.shm_info, u32.shm_info32, shm_tot);
1230                 CP(u.shm_info, u32.shm_info32, shm_swp);
1231                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1232                 CP(u.shm_info, u32.shm_info32, swap_successes);
1233                 error = copyout(&u32.shm_info32, uap->buf,
1234                     sizeof(u32.shm_info32));
1235                 break;
1236         case SHM_STAT:
1237         case IPC_STAT:
1238                 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1239                     &u32.shmid_ds32.shm_perm);
1240                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1241                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1242                 else
1243                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1244                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1245                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1246                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1247                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1248                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1249                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1250                 error = copyout(&u32.shmid_ds32, uap->buf,
1251                     sizeof(u32.shmid_ds32));
1252                 break;
1253         }
1254
1255 done:
1256         if (error) {
1257                 /* Invalidate the return value */
1258                 td->td_retval[0] = -1;
1259         }
1260         return (error);
1261 }
1262 #endif
1263
1264 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1265     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1266
1267 #ifndef CP
1268 #define CP(src, dst, fld)       do { (dst).fld = (src).fld; } while (0)
1269 #endif
1270
1271 #ifndef _SYS_SYSPROTO_H_
1272 struct freebsd7_shmctl_args {
1273         int shmid;
1274         int cmd;
1275         struct shmid_ds_old *buf;
1276 };
1277 #endif
1278 int
1279 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1280 {
1281         int error = 0;
1282         struct shmid_ds_old old;
1283         struct shmid_ds buf;
1284         size_t bufsz;
1285
1286         /*
1287          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1288          * Linux binaries.  If we see the call come through the FreeBSD ABI,
1289          * return an error back to the user since we do not to support this.
1290          */
1291         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1292             uap->cmd == SHM_STAT)
1293                 return (EINVAL);
1294
1295         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1296         if (uap->cmd == IPC_SET) {
1297                 if ((error = copyin(uap->buf, &old, sizeof(old))))
1298                         goto done;
1299                 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1300                 CP(old, buf, shm_segsz);
1301                 CP(old, buf, shm_lpid);
1302                 CP(old, buf, shm_cpid);
1303                 CP(old, buf, shm_nattch);
1304                 CP(old, buf, shm_atime);
1305                 CP(old, buf, shm_dtime);
1306                 CP(old, buf, shm_ctime);
1307         }
1308
1309         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1310         if (error)
1311                 goto done;
1312
1313         /* Cases in which we need to copyout */
1314         switch (uap->cmd) {
1315         case IPC_STAT:
1316                 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1317                 if (buf.shm_segsz > INT_MAX)
1318                         old.shm_segsz = INT_MAX;
1319                 else
1320                         CP(buf, old, shm_segsz);
1321                 CP(buf, old, shm_lpid);
1322                 CP(buf, old, shm_cpid);
1323                 if (buf.shm_nattch > SHRT_MAX)
1324                         old.shm_nattch = SHRT_MAX;
1325                 else
1326                         CP(buf, old, shm_nattch);
1327                 CP(buf, old, shm_atime);
1328                 CP(buf, old, shm_dtime);
1329                 CP(buf, old, shm_ctime);
1330                 old.shm_internal = NULL;
1331                 error = copyout(&old, uap->buf, sizeof(old));
1332                 break;
1333         }
1334
1335 done:
1336         if (error) {
1337                 /* Invalidate the return value */
1338                 td->td_retval[0] = -1;
1339         }
1340         return (error);
1341 }
1342
1343 #endif  /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1344            COMPAT_FREEBSD7 */
1345
1346 static int
1347 sysvshm_modload(struct module *module, int cmd, void *arg)
1348 {
1349         int error = 0;
1350
1351         switch (cmd) {
1352         case MOD_LOAD:
1353                 error = shminit();
1354                 if (error != 0)
1355                         shmunload();
1356                 break;
1357         case MOD_UNLOAD:
1358                 error = shmunload();
1359                 break;
1360         case MOD_SHUTDOWN:
1361                 break;
1362         default:
1363                 error = EINVAL;
1364                 break;
1365         }
1366         return (error);
1367 }
1368
1369 static moduledata_t sysvshm_mod = {
1370         "sysvshm",
1371         &sysvshm_modload,
1372         NULL
1373 };
1374
1375 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1376 MODULE_VERSION(sysvshm, 1);