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/stat.h>
  83 #include <sys/syscall.h>
  84 #include <sys/syscallsubr.h>
  85 #include <sys/sysent.h>
  86 #include <sys/sysproto.h>
  87 #include <sys/jail.h>
  88
  89 #include <security/mac/mac_framework.h>
  90
  91 #include <vm/vm.h>
  92 #include <vm/vm_param.h>
  93 #include <vm/pmap.h>
  94 #include <vm/vm_object.h>
  95 #include <vm/vm_map.h>
  96 #include <vm/vm_page.h>
  97 #include <vm/vm_pager.h>
  98
  99 FEATURE(sysv_shm, "System V shared memory segments support");
 100
 101 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
 102
 103 static int shmget_allocate_segment(struct thread *td,
 104     struct shmget_args *uap, int mode);
 105 static int shmget_existing(struct thread *td, struct shmget_args *uap,
 106     int mode, int segnum);
 107
 108 #define SHMSEG_FREE             0x0200
 109 #define SHMSEG_REMOVED          0x0400
 110 #define SHMSEG_ALLOCATED        0x0800
 111 #define SHMSEG_WANTED           0x1000
 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_by_shmid(int);
 125 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
 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
 135 /*
 136  * Tuneable values.
 137  */
 138 #ifndef SHMMAXPGS
 139 #define SHMMAXPGS       131072  /* Note: sysv shared memory is swap backed. */
 140 #endif
 141 #ifndef SHMMAX
 142 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 143 #endif
 144 #ifndef SHMMIN
 145 #define SHMMIN  1
 146 #endif
 147 #ifndef SHMMNI
 148 #define SHMMNI  192
 149 #endif
 150 #ifndef SHMSEG
 151 #define SHMSEG  128
 152 #endif
 153 #ifndef SHMALL
 154 #define SHMALL  (SHMMAXPGS)
 155 #endif
 156
 157 struct  shminfo shminfo = {
 158         SHMMAX,
 159         SHMMIN,
 160         SHMMNI,
 161         SHMSEG,
 162         SHMALL
 163 };
 164
 165 static int shm_use_phys;
 166 static int shm_allow_removed;
 167
 168 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 169     "Maximum shared memory segment size");
 170 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 171     "Minimum shared memory segment size");
 172 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
 173     "Number of shared memory identifiers");
 174 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
 175     "Number of segments per process");
 176 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 177     "Maximum number of pages available for shared memory");
 178 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
 179     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
 180 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
 181     &shm_allow_removed, 0,
 182     "Enable/Disable attachment to attached segments marked for removal");
 183 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD,
 184     NULL, 0, sysctl_shmsegs, "",
 185     "Current number of shared memory segments allocated");
 186
 187 static int
 188 shm_find_segment_by_key(key)
 189         key_t key;
 190 {
 191         int i;
 192
 193         for (i = 0; i < shmalloced; i++)
 194                 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
 195                     shmsegs[i].u.shm_perm.key == key)
 196                         return (i);
 197         return (-1);
 198 }
 199
 200 static struct shmid_kernel *
 201 shm_find_segment_by_shmid(int shmid)
 202 {
 203         int segnum;
 204         struct shmid_kernel *shmseg;
 205
 206         segnum = IPCID_TO_IX(shmid);
 207         if (segnum < 0 || segnum >= shmalloced)
 208                 return (NULL);
 209         shmseg = &shmsegs[segnum];
 210         if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 211             (!shm_allow_removed &&
 212              (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 213             shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
 214                 return (NULL);
 215         return (shmseg);
 216 }
 217
 218 static struct shmid_kernel *
 219 shm_find_segment_by_shmidx(int segnum)
 220 {
 221         struct shmid_kernel *shmseg;
 222
 223         if (segnum < 0 || segnum >= shmalloced)
 224                 return (NULL);
 225         shmseg = &shmsegs[segnum];
 226         if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 227             (!shm_allow_removed &&
 228              (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
 229                 return (NULL);
 230         return (shmseg);
 231 }
 232
 233 static void
 234 shm_deallocate_segment(shmseg)
 235         struct shmid_kernel *shmseg;
 236 {
 237         vm_size_t size;
 238
 239         GIANT_REQUIRED;
 240
 241         vm_object_deallocate(shmseg->object);
 242         shmseg->object = NULL;
 243         size = round_page(shmseg->u.shm_segsz);
 244         shm_committed -= btoc(size);
 245         shm_nused--;
 246         shmseg->u.shm_perm.mode = SHMSEG_FREE;
 247 #ifdef MAC
 248         mac_sysvshm_cleanup(shmseg);
 249 #endif
 250         racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
 251         racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
 252         crfree(shmseg->cred);
 253         shmseg->cred = NULL;
 254 }
 255
 256 static int
 257 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 258 {
 259         struct shmid_kernel *shmseg;
 260         int segnum, result;
 261         vm_size_t size;
 262
 263         GIANT_REQUIRED;
 264
 265         segnum = IPCID_TO_IX(shmmap_s->shmid);
 266         shmseg = &shmsegs[segnum];
 267         size = round_page(shmseg->u.shm_segsz);
 268         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 269         if (result != KERN_SUCCESS)
 270                 return (EINVAL);
 271         shmmap_s->shmid = -1;
 272         shmseg->u.shm_dtime = time_second;
 273         if ((--shmseg->u.shm_nattch <= 0) &&
 274             (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
 275                 shm_deallocate_segment(shmseg);
 276                 shm_last_free = segnum;
 277         }
 278         return (0);
 279 }
 280
 281 #ifndef _SYS_SYSPROTO_H_
 282 struct shmdt_args {
 283         const void *shmaddr;
 284 };
 285 #endif
 286 int
 287 sys_shmdt(td, uap)
 288         struct thread *td;
 289         struct shmdt_args *uap;
 290 {
 291         struct proc *p = td->td_proc;
 292         struct shmmap_state *shmmap_s;
 293 #ifdef MAC
 294         struct shmid_kernel *shmsegptr;
 295 #endif
 296         int i;
 297         int error = 0;
 298
 299         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 300                 return (ENOSYS);
 301         mtx_lock(&Giant);
 302         shmmap_s = p->p_vmspace->vm_shm;
 303         if (shmmap_s == NULL) {
 304                 error = EINVAL;
 305                 goto done2;
 306         }
 307         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 308                 if (shmmap_s->shmid != -1 &&
 309                     shmmap_s->va == (vm_offset_t)uap->shmaddr) {
 310                         break;
 311                 }
 312         }
 313         if (i == shminfo.shmseg) {
 314                 error = EINVAL;
 315                 goto done2;
 316         }
 317 #ifdef MAC
 318         shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
 319         error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
 320         if (error != 0)
 321                 goto done2;
 322 #endif
 323         error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 324 done2:
 325         mtx_unlock(&Giant);
 326         return (error);
 327 }
 328
 329 #ifndef _SYS_SYSPROTO_H_
 330 struct shmat_args {
 331         int shmid;
 332         const void *shmaddr;
 333         int shmflg;
 334 };
 335 #endif
 336 int
 337 kern_shmat(td, shmid, shmaddr, shmflg)
 338         struct thread *td;
 339         int shmid;
 340         const void *shmaddr;
 341         int shmflg;
 342 {
 343         struct proc *p = td->td_proc;
 344         int i, flags;
 345         struct shmid_kernel *shmseg;
 346         struct shmmap_state *shmmap_s = NULL;
 347         vm_offset_t attach_va;
 348         vm_prot_t prot;
 349         vm_size_t size;
 350         int rv;
 351         int error = 0;
 352
 353         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 354                 return (ENOSYS);
 355         mtx_lock(&Giant);
 356         shmmap_s = p->p_vmspace->vm_shm;
 357         if (shmmap_s == NULL) {
 358                 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
 359                     M_SHM, M_WAITOK);
 360
 361                 /*
 362                  * If malloc() above sleeps, the Giant lock is
 363                  * temporarily dropped, which allows another thread to
 364                  * allocate shmmap_state and set vm_shm.  Recheck
 365                  * vm_shm and free the new shmmap_state if another one
 366                  * is already allocated.
 367                  */
 368                 if (p->p_vmspace->vm_shm != NULL) {
 369                         free(shmmap_s, M_SHM);
 370                         shmmap_s = p->p_vmspace->vm_shm;
 371                 } else {
 372                         for (i = 0; i < shminfo.shmseg; i++)
 373                                 shmmap_s[i].shmid = -1;
 374                         p->p_vmspace->vm_shm = shmmap_s;
 375                 }
 376         }
 377         shmseg = shm_find_segment_by_shmid(shmid);
 378         if (shmseg == NULL) {
 379                 error = EINVAL;
 380                 goto done2;
 381         }
 382         error = ipcperm(td, &shmseg->u.shm_perm,
 383             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 384         if (error)
 385                 goto done2;
 386 #ifdef MAC
 387         error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
 388         if (error != 0)
 389                 goto done2;
 390 #endif
 391         for (i = 0; i < shminfo.shmseg; i++) {
 392                 if (shmmap_s->shmid == -1)
 393                         break;
 394                 shmmap_s++;
 395         }
 396         if (i >= shminfo.shmseg) {
 397                 error = EMFILE;
 398                 goto done2;
 399         }
 400         size = round_page(shmseg->u.shm_segsz);
 401         prot = VM_PROT_READ;
 402         if ((shmflg & SHM_RDONLY) == 0)
 403                 prot |= VM_PROT_WRITE;
 404         flags = MAP_ANON | MAP_SHARED;
 405         if (shmaddr) {
 406                 flags |= MAP_FIXED;
 407                 if (shmflg & SHM_RND) {
 408                         attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
 409                 } else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
 410                         attach_va = (vm_offset_t)shmaddr;
 411                 } else {
 412                         error = EINVAL;
 413                         goto done2;
 414                 }
 415         } else {
 416                 /*
 417                  * This is just a hint to vm_map_find() about where to
 418                  * put it.
 419                  */
 420                 PROC_LOCK(p);
 421                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
 422                     lim_max(p, RLIMIT_DATA));
 423                 PROC_UNLOCK(p);
 424         }
 425
 426         vm_object_reference(shmseg->object);
 427         rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object,
 428             0, &attach_va, size, (flags & MAP_FIXED) ? VMFS_NO_SPACE :
 429             VMFS_OPTIMAL_SPACE, prot, prot, MAP_INHERIT_SHARE);
 430         if (rv != KERN_SUCCESS) {
 431                 vm_object_deallocate(shmseg->object);
 432                 error = ENOMEM;
 433                 goto done2;
 434         }
 435
 436         shmmap_s->va = attach_va;
 437         shmmap_s->shmid = shmid;
 438         shmseg->u.shm_lpid = p->p_pid;
 439         shmseg->u.shm_atime = time_second;
 440         shmseg->u.shm_nattch++;
 441         td->td_retval[0] = attach_va;
 442 done2:
 443         mtx_unlock(&Giant);
 444         return (error);
 445 }
 446
 447 int
 448 sys_shmat(td, uap)
 449         struct thread *td;
 450         struct shmat_args *uap;
 451 {
 452         return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
 453 }
 454
 455 int
 456 kern_shmctl(td, shmid, cmd, buf, bufsz)
 457         struct thread *td;
 458         int shmid;
 459         int cmd;
 460         void *buf;
 461         size_t *bufsz;
 462 {
 463         int error = 0;
 464         struct shmid_kernel *shmseg;
 465
 466         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 467                 return (ENOSYS);
 468
 469         mtx_lock(&Giant);
 470         switch (cmd) {
 471         /*
 472          * It is possible that kern_shmctl is being called from the Linux ABI
 473          * layer, in which case, we will need to implement IPC_INFO.  It should
 474          * be noted that other shmctl calls will be funneled through here for
 475          * Linix binaries as well.
 476          *
 477          * NB: The Linux ABI layer will convert this data to structure(s) more
 478          * consistent with the Linux ABI.
 479          */
 480         case IPC_INFO:
 481                 memcpy(buf, &shminfo, sizeof(shminfo));
 482                 if (bufsz)
 483                         *bufsz = sizeof(shminfo);
 484                 td->td_retval[0] = shmalloced;
 485                 goto done2;
 486         case SHM_INFO: {
 487                 struct shm_info shm_info;
 488                 shm_info.used_ids = shm_nused;
 489                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 490                 shm_info.shm_tot = 0;   /*XXX where to get from ? */
 491                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 492                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 493                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 494                 memcpy(buf, &shm_info, sizeof(shm_info));
 495                 if (bufsz)
 496                         *bufsz = sizeof(shm_info);
 497                 td->td_retval[0] = shmalloced;
 498                 goto done2;
 499         }
 500         }
 501         if (cmd == SHM_STAT)
 502                 shmseg = shm_find_segment_by_shmidx(shmid);
 503         else
 504                 shmseg = shm_find_segment_by_shmid(shmid);
 505         if (shmseg == NULL) {
 506                 error = EINVAL;
 507                 goto done2;
 508         }
 509 #ifdef MAC
 510         error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
 511         if (error != 0)
 512                 goto done2;
 513 #endif
 514         switch (cmd) {
 515         case SHM_STAT:
 516         case IPC_STAT:
 517                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
 518                 if (error)
 519                         goto done2;
 520                 memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
 521                 if (bufsz)
 522                         *bufsz = sizeof(struct shmid_ds);
 523                 if (cmd == SHM_STAT)
 524                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
 525                 break;
 526         case IPC_SET: {
 527                 struct shmid_ds *shmid;
 528
 529                 shmid = (struct shmid_ds *)buf;
 530                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 531                 if (error)
 532                         goto done2;
 533                 shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
 534                 shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
 535                 shmseg->u.shm_perm.mode =
 536                     (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
 537                     (shmid->shm_perm.mode & ACCESSPERMS);
 538                 shmseg->u.shm_ctime = time_second;
 539                 break;
 540         }
 541         case IPC_RMID:
 542                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
 543                 if (error)
 544                         goto done2;
 545                 shmseg->u.shm_perm.key = IPC_PRIVATE;
 546                 shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
 547                 if (shmseg->u.shm_nattch <= 0) {
 548                         shm_deallocate_segment(shmseg);
 549                         shm_last_free = IPCID_TO_IX(shmid);
 550                 }
 551                 break;
 552 #if 0
 553         case SHM_LOCK:
 554         case SHM_UNLOCK:
 555 #endif
 556         default:
 557                 error = EINVAL;
 558                 break;
 559         }
 560 done2:
 561         mtx_unlock(&Giant);
 562         return (error);
 563 }
 564
 565 #ifndef _SYS_SYSPROTO_H_
 566 struct shmctl_args {
 567         int shmid;
 568         int cmd;
 569         struct shmid_ds *buf;
 570 };
 571 #endif
 572 int
 573 sys_shmctl(td, uap)
 574         struct thread *td;
 575         struct shmctl_args *uap;
 576 {
 577         int error = 0;
 578         struct shmid_ds buf;
 579         size_t bufsz;
 580
 581         /*
 582          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
 583          * Linux binaries.  If we see the call come through the FreeBSD ABI,
 584          * return an error back to the user since we do not to support this.
 585          */
 586         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
 587             uap->cmd == SHM_STAT)
 588                 return (EINVAL);
 589
 590         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
 591         if (uap->cmd == IPC_SET) {
 592                 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
 593                         goto done;
 594         }
 595
 596         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
 597         if (error)
 598                 goto done;
 599
 600         /* Cases in which we need to copyout */
 601         switch (uap->cmd) {
 602         case IPC_STAT:
 603                 error = copyout(&buf, uap->buf, bufsz);
 604                 break;
 605         }
 606
 607 done:
 608         if (error) {
 609                 /* Invalidate the return value */
 610                 td->td_retval[0] = -1;
 611         }
 612         return (error);
 613 }
 614
 615
 616 static int
 617 shmget_existing(td, uap, mode, segnum)
 618         struct thread *td;
 619         struct shmget_args *uap;
 620         int mode;
 621         int segnum;
 622 {
 623         struct shmid_kernel *shmseg;
 624         int error;
 625
 626         shmseg = &shmsegs[segnum];
 627         if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
 628                 /*
 629                  * This segment is in the process of being allocated.  Wait
 630                  * until it's done, and look the key up again (in case the
 631                  * allocation failed or it was freed).
 632                  */
 633                 shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
 634                 error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
 635                 if (error)
 636                         return (error);
 637                 return (EAGAIN);
 638         }
 639         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 640                 return (EEXIST);
 641 #ifdef MAC
 642         error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
 643         if (error != 0)
 644                 return (error);
 645 #endif
 646         if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
 647                 return (EINVAL);
 648         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 649         return (0);
 650 }
 651
 652 static int
 653 shmget_allocate_segment(td, uap, mode)
 654         struct thread *td;
 655         struct shmget_args *uap;
 656         int mode;
 657 {
 658         int i, segnum, shmid;
 659         size_t size;
 660         struct ucred *cred = td->td_ucred;
 661         struct shmid_kernel *shmseg;
 662         vm_object_t shm_object;
 663
 664         GIANT_REQUIRED;
 665
 666         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 667                 return (EINVAL);
 668         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 669                 return (ENOSPC);
 670         size = round_page(uap->size);
 671         if (shm_committed + btoc(size) > shminfo.shmall)
 672                 return (ENOMEM);
 673         if (shm_last_free < 0) {
 674                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 675                 for (i = 0; i < shmalloced; i++)
 676                         if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
 677                                 break;
 678                 if (i == shmalloced)
 679                         return (ENOSPC);
 680                 segnum = i;
 681         } else  {
 682                 segnum = shm_last_free;
 683                 shm_last_free = -1;
 684         }
 685         shmseg = &shmsegs[segnum];
 686 #ifdef RACCT
 687         PROC_LOCK(td->td_proc);
 688         if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
 689                 PROC_UNLOCK(td->td_proc);
 690                 return (ENOSPC);
 691         }
 692         if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
 693                 racct_sub(td->td_proc, RACCT_NSHM, 1);
 694                 PROC_UNLOCK(td->td_proc);
 695                 return (ENOMEM);
 696         }
 697         PROC_UNLOCK(td->td_proc);
 698 #endif
 699         /*
 700          * In case we sleep in malloc(), mark the segment present but deleted
 701          * so that noone else tries to create the same key.
 702          */
 703         shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 704         shmseg->u.shm_perm.key = uap->key;
 705         shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
 706         shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
 707
 708         /*
 709          * We make sure that we have allocated a pager before we need
 710          * to.
 711          */
 712         shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
 713             0, size, VM_PROT_DEFAULT, 0, cred);
 714         if (shm_object == NULL) {
 715 #ifdef RACCT
 716                 PROC_LOCK(td->td_proc);
 717                 racct_sub(td->td_proc, RACCT_NSHM, 1);
 718                 racct_sub(td->td_proc, RACCT_SHMSIZE, size);
 719                 PROC_UNLOCK(td->td_proc);
 720 #endif
 721                 return (ENOMEM);
 722         }
 723         VM_OBJECT_LOCK(shm_object);
 724         vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
 725         vm_object_set_flag(shm_object, OBJ_NOSPLIT);
 726         VM_OBJECT_UNLOCK(shm_object);
 727
 728         shmseg->object = shm_object;
 729         shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
 730         shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
 731         shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
 732             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 733         shmseg->cred = crhold(cred);
 734         shmseg->u.shm_segsz = uap->size;
 735         shmseg->u.shm_cpid = td->td_proc->p_pid;
 736         shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
 737         shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
 738 #ifdef MAC
 739         mac_sysvshm_create(cred, shmseg);
 740 #endif
 741         shmseg->u.shm_ctime = time_second;
 742         shm_committed += btoc(size);
 743         shm_nused++;
 744         if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
 745                 /*
 746                  * Somebody else wanted this key while we were asleep.  Wake
 747                  * them up now.
 748                  */
 749                 shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
 750                 wakeup(shmseg);
 751         }
 752         td->td_retval[0] = shmid;
 753         return (0);
 754 }
 755
 756 #ifndef _SYS_SYSPROTO_H_
 757 struct shmget_args {
 758         key_t key;
 759         size_t size;
 760         int shmflg;
 761 };
 762 #endif
 763 int
 764 sys_shmget(td, uap)
 765         struct thread *td;
 766         struct shmget_args *uap;
 767 {
 768         int segnum, mode;
 769         int error;
 770
 771         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
 772                 return (ENOSYS);
 773         mtx_lock(&Giant);
 774         mode = uap->shmflg & ACCESSPERMS;
 775         if (uap->key != IPC_PRIVATE) {
 776         again:
 777                 segnum = shm_find_segment_by_key(uap->key);
 778                 if (segnum >= 0) {
 779                         error = shmget_existing(td, uap, mode, segnum);
 780                         if (error == EAGAIN)
 781                                 goto again;
 782                         goto done2;
 783                 }
 784                 if ((uap->shmflg & IPC_CREAT) == 0) {
 785                         error = ENOENT;
 786                         goto done2;
 787                 }
 788         }
 789         error = shmget_allocate_segment(td, uap, mode);
 790 done2:
 791         mtx_unlock(&Giant);
 792         return (error);
 793 }
 794
 795 static void
 796 shmfork_myhook(p1, p2)
 797         struct proc *p1, *p2;
 798 {
 799         struct shmmap_state *shmmap_s;
 800         size_t size;
 801         int i;
 802
 803         mtx_lock(&Giant);
 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                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
 811         mtx_unlock(&Giant);
 812 }
 813
 814 static void
 815 shmexit_myhook(struct vmspace *vm)
 816 {
 817         struct shmmap_state *base, *shm;
 818         int i;
 819
 820         if ((base = vm->vm_shm) != NULL) {
 821                 vm->vm_shm = NULL;
 822                 mtx_lock(&Giant);
 823                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 824                         if (shm->shmid != -1)
 825                                 shm_delete_mapping(vm, shm);
 826                 }
 827                 mtx_unlock(&Giant);
 828                 free(base, M_SHM);
 829         }
 830 }
 831
 832 static void
 833 shmrealloc(void)
 834 {
 835         int i;
 836         struct shmid_kernel *newsegs;
 837
 838         if (shmalloced >= shminfo.shmmni)
 839                 return;
 840
 841         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 842         for (i = 0; i < shmalloced; i++)
 843                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 844         for (; i < shminfo.shmmni; i++) {
 845                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 846                 shmsegs[i].u.shm_perm.seq = 0;
 847 #ifdef MAC
 848                 mac_sysvshm_init(&shmsegs[i]);
 849 #endif
 850         }
 851         free(shmsegs, M_SHM);
 852         shmsegs = newsegs;
 853         shmalloced = shminfo.shmmni;
 854 }
 855
 856 static struct syscall_helper_data shm_syscalls[] = {
 857         SYSCALL_INIT_HELPER(shmat),
 858         SYSCALL_INIT_HELPER(shmctl),
 859         SYSCALL_INIT_HELPER(shmdt),
 860         SYSCALL_INIT_HELPER(shmget),
 861 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 862     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 863         SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
 864 #endif
 865 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 866         SYSCALL_INIT_HELPER(shmsys),
 867 #endif
 868         SYSCALL_INIT_LAST
 869 };
 870
 871 #ifdef COMPAT_FREEBSD32
 872 #include <compat/freebsd32/freebsd32.h>
 873 #include <compat/freebsd32/freebsd32_ipc.h>
 874 #include <compat/freebsd32/freebsd32_proto.h>
 875 #include <compat/freebsd32/freebsd32_signal.h>
 876 #include <compat/freebsd32/freebsd32_syscall.h>
 877 #include <compat/freebsd32/freebsd32_util.h>
 878
 879 static struct syscall_helper_data shm32_syscalls[] = {
 880         SYSCALL32_INIT_HELPER_COMPAT(shmat),
 881         SYSCALL32_INIT_HELPER_COMPAT(shmdt),
 882         SYSCALL32_INIT_HELPER_COMPAT(shmget),
 883         SYSCALL32_INIT_HELPER(freebsd32_shmsys),
 884         SYSCALL32_INIT_HELPER(freebsd32_shmctl),
 885 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
 886     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
 887         SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
 888 #endif
 889         SYSCALL_INIT_LAST
 890 };
 891 #endif
 892
 893 static int
 894 shminit()
 895 {
 896         int i, error;
 897
 898 #ifndef BURN_BRIDGES
 899         if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
 900                 printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
 901 #endif
 902         TUNABLE_ULONG_FETCH("kern.ipc.shmall", &shminfo.shmall);
 903
 904         /* Initialize shmmax dealing with possible overflow. */
 905         for (i = PAGE_SIZE; i > 0; i--) {
 906                 shminfo.shmmax = shminfo.shmall * i;
 907                 if (shminfo.shmmax >= shminfo.shmall)
 908                         break;
 909         }
 910
 911         TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
 912         TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
 913         TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
 914         TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
 915
 916         shmalloced = shminfo.shmmni;
 917         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 918         for (i = 0; i < shmalloced; i++) {
 919                 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
 920                 shmsegs[i].u.shm_perm.seq = 0;
 921 #ifdef MAC
 922                 mac_sysvshm_init(&shmsegs[i]);
 923 #endif
 924         }
 925         shm_last_free = 0;
 926         shm_nused = 0;
 927         shm_committed = 0;
 928         shmexit_hook = &shmexit_myhook;
 929         shmfork_hook = &shmfork_myhook;
 930
 931         error = syscall_helper_register(shm_syscalls);
 932         if (error != 0)
 933                 return (error);
 934 #ifdef COMPAT_FREEBSD32
 935         error = syscall32_helper_register(shm32_syscalls);
 936         if (error != 0)
 937                 return (error);
 938 #endif
 939         return (0);
 940 }
 941
 942 static int
 943 shmunload()
 944 {
 945         int i;
 946
 947         if (shm_nused > 0)
 948                 return (EBUSY);
 949
 950 #ifdef COMPAT_FREEBSD32
 951         syscall32_helper_unregister(shm32_syscalls);
 952 #endif
 953         syscall_helper_unregister(shm_syscalls);
 954
 955         for (i = 0; i < shmalloced; i++) {
 956 #ifdef MAC
 957                 mac_sysvshm_destroy(&shmsegs[i]);
 958 #endif
 959                 /*
 960                  * Objects might be still mapped into the processes
 961                  * address spaces.  Actual free would happen on the
 962                  * last mapping destruction.
 963                  */
 964                 if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
 965                         vm_object_deallocate(shmsegs[i].object);
 966         }
 967         free(shmsegs, M_SHM);
 968         shmexit_hook = NULL;
 969         shmfork_hook = NULL;
 970         return (0);
 971 }
 972
 973 static int
 974 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
 975 {
 976
 977         return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
 978 }
 979
 980 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
 981 struct oshmid_ds {
 982         struct  ipc_perm_old shm_perm;  /* operation perms */
 983         int     shm_segsz;              /* size of segment (bytes) */
 984         u_short shm_cpid;               /* pid, creator */
 985         u_short shm_lpid;               /* pid, last operation */
 986         short   shm_nattch;             /* no. of current attaches */
 987         time_t  shm_atime;              /* last attach time */
 988         time_t  shm_dtime;              /* last detach time */
 989         time_t  shm_ctime;              /* last change time */
 990         void    *shm_handle;            /* internal handle for shm segment */
 991 };
 992
 993 struct oshmctl_args {
 994         int shmid;
 995         int cmd;
 996         struct oshmid_ds *ubuf;
 997 };
 998
 999 static int
1000 oshmctl(struct thread *td, struct oshmctl_args *uap)
1001 {
1002 #ifdef COMPAT_43
1003         int error = 0;
1004         struct shmid_kernel *shmseg;
1005         struct oshmid_ds outbuf;
1006
1007         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1008                 return (ENOSYS);
1009         mtx_lock(&Giant);
1010         shmseg = shm_find_segment_by_shmid(uap->shmid);
1011         if (shmseg == NULL) {
1012                 error = EINVAL;
1013                 goto done2;
1014         }
1015         switch (uap->cmd) {
1016         case IPC_STAT:
1017                 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1018                 if (error)
1019                         goto done2;
1020 #ifdef MAC
1021                 error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1022                 if (error != 0)
1023                         goto done2;
1024 #endif
1025                 ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1026                 outbuf.shm_segsz = shmseg->u.shm_segsz;
1027                 outbuf.shm_cpid = shmseg->u.shm_cpid;
1028                 outbuf.shm_lpid = shmseg->u.shm_lpid;
1029                 outbuf.shm_nattch = shmseg->u.shm_nattch;
1030                 outbuf.shm_atime = shmseg->u.shm_atime;
1031                 outbuf.shm_dtime = shmseg->u.shm_dtime;
1032                 outbuf.shm_ctime = shmseg->u.shm_ctime;
1033                 outbuf.shm_handle = shmseg->object;
1034                 error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
1035                 if (error)
1036                         goto done2;
1037                 break;
1038         default:
1039                 error = freebsd7_shmctl(td, (struct freebsd7_shmctl_args *)uap);
1040                 break;
1041         }
1042 done2:
1043         mtx_unlock(&Giant);
1044         return (error);
1045 #else
1046         return (EINVAL);
1047 #endif
1048 }
1049
1050 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1051 static sy_call_t *shmcalls[] = {
1052         (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1053         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1054         (sy_call_t *)freebsd7_shmctl
1055 };
1056
1057 int
1058 sys_shmsys(td, uap)
1059         struct thread *td;
1060         /* XXX actually varargs. */
1061         struct shmsys_args /* {
1062                 int     which;
1063                 int     a2;
1064                 int     a3;
1065                 int     a4;
1066         } */ *uap;
1067 {
1068         int error;
1069
1070         if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1071                 return (ENOSYS);
1072         if (uap->which < 0 ||
1073             uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
1074                 return (EINVAL);
1075         mtx_lock(&Giant);
1076         error = (*shmcalls[uap->which])(td, &uap->a2);
1077         mtx_unlock(&Giant);
1078         return (error);
1079 }
1080
1081 #endif  /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1082
1083 #ifdef COMPAT_FREEBSD32
1084
1085 int
1086 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1087 {
1088
1089 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1090     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1091         switch (uap->which) {
1092         case 0: {       /* shmat */
1093                 struct shmat_args ap;
1094
1095                 ap.shmid = uap->a2;
1096                 ap.shmaddr = PTRIN(uap->a3);
1097                 ap.shmflg = uap->a4;
1098                 return (sysent[SYS_shmat].sy_call(td, &ap));
1099         }
1100         case 2: {       /* shmdt */
1101                 struct shmdt_args ap;
1102
1103                 ap.shmaddr = PTRIN(uap->a2);
1104                 return (sysent[SYS_shmdt].sy_call(td, &ap));
1105         }
1106         case 3: {       /* shmget */
1107                 struct shmget_args ap;
1108
1109                 ap.key = uap->a2;
1110                 ap.size = uap->a3;
1111                 ap.shmflg = uap->a4;
1112                 return (sysent[SYS_shmget].sy_call(td, &ap));
1113         }
1114         case 4: {       /* shmctl */
1115                 struct freebsd7_freebsd32_shmctl_args ap;
1116
1117                 ap.shmid = uap->a2;
1118                 ap.cmd = uap->a3;
1119                 ap.buf = PTRIN(uap->a4);
1120                 return (freebsd7_freebsd32_shmctl(td, &ap));
1121         }
1122         case 1:         /* oshmctl */
1123         default:
1124                 return (EINVAL);
1125         }
1126 #else
1127         return (nosys(td, NULL));
1128 #endif
1129 }
1130
1131 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1132     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1133 int
1134 freebsd7_freebsd32_shmctl(struct thread *td,
1135     struct freebsd7_freebsd32_shmctl_args *uap)
1136 {
1137         int error = 0;
1138         union {
1139                 struct shmid_ds shmid_ds;
1140                 struct shm_info shm_info;
1141                 struct shminfo shminfo;
1142         } u;
1143         union {
1144                 struct shmid_ds32_old shmid_ds32;
1145                 struct shm_info32 shm_info32;
1146                 struct shminfo32 shminfo32;
1147         } u32;
1148         size_t sz;
1149
1150         if (uap->cmd == IPC_SET) {
1151                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1152                     sizeof(u32.shmid_ds32))))
1153                         goto done;
1154                 freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1155                     &u.shmid_ds.shm_perm);
1156                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1157                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1158                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1159                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1160                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1161                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1162                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1163         }
1164
1165         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1166         if (error)
1167                 goto done;
1168
1169         /* Cases in which we need to copyout */
1170         switch (uap->cmd) {
1171         case IPC_INFO:
1172                 CP(u.shminfo, u32.shminfo32, shmmax);
1173                 CP(u.shminfo, u32.shminfo32, shmmin);
1174                 CP(u.shminfo, u32.shminfo32, shmmni);
1175                 CP(u.shminfo, u32.shminfo32, shmseg);
1176                 CP(u.shminfo, u32.shminfo32, shmall);
1177                 error = copyout(&u32.shminfo32, uap->buf,
1178                     sizeof(u32.shminfo32));
1179                 break;
1180         case SHM_INFO:
1181                 CP(u.shm_info, u32.shm_info32, used_ids);
1182                 CP(u.shm_info, u32.shm_info32, shm_rss);
1183                 CP(u.shm_info, u32.shm_info32, shm_tot);
1184                 CP(u.shm_info, u32.shm_info32, shm_swp);
1185                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1186                 CP(u.shm_info, u32.shm_info32, swap_successes);
1187                 error = copyout(&u32.shm_info32, uap->buf,
1188                     sizeof(u32.shm_info32));
1189                 break;
1190         case SHM_STAT:
1191         case IPC_STAT:
1192                 freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1193                     &u32.shmid_ds32.shm_perm);
1194                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1195                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1196                 else
1197                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1198                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1199                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1200                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1201                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1202                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1203                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1204                 u32.shmid_ds32.shm_internal = 0;
1205                 error = copyout(&u32.shmid_ds32, uap->buf,
1206                     sizeof(u32.shmid_ds32));
1207                 break;
1208         }
1209
1210 done:
1211         if (error) {
1212                 /* Invalidate the return value */
1213                 td->td_retval[0] = -1;
1214         }
1215         return (error);
1216 }
1217 #endif
1218
1219 int
1220 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1221 {
1222         int error = 0;
1223         union {
1224                 struct shmid_ds shmid_ds;
1225                 struct shm_info shm_info;
1226                 struct shminfo shminfo;
1227         } u;
1228         union {
1229                 struct shmid_ds32 shmid_ds32;
1230                 struct shm_info32 shm_info32;
1231                 struct shminfo32 shminfo32;
1232         } u32;
1233         size_t sz;
1234
1235         if (uap->cmd == IPC_SET) {
1236                 if ((error = copyin(uap->buf, &u32.shmid_ds32,
1237                     sizeof(u32.shmid_ds32))))
1238                         goto done;
1239                 freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1240                     &u.shmid_ds.shm_perm);
1241                 CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1242                 CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1243                 CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1244                 CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1245                 CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1246                 CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1247                 CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1248         }
1249
1250         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1251         if (error)
1252                 goto done;
1253
1254         /* Cases in which we need to copyout */
1255         switch (uap->cmd) {
1256         case IPC_INFO:
1257                 CP(u.shminfo, u32.shminfo32, shmmax);
1258                 CP(u.shminfo, u32.shminfo32, shmmin);
1259                 CP(u.shminfo, u32.shminfo32, shmmni);
1260                 CP(u.shminfo, u32.shminfo32, shmseg);
1261                 CP(u.shminfo, u32.shminfo32, shmall);
1262                 error = copyout(&u32.shminfo32, uap->buf,
1263                     sizeof(u32.shminfo32));
1264                 break;
1265         case SHM_INFO:
1266                 CP(u.shm_info, u32.shm_info32, used_ids);
1267                 CP(u.shm_info, u32.shm_info32, shm_rss);
1268                 CP(u.shm_info, u32.shm_info32, shm_tot);
1269                 CP(u.shm_info, u32.shm_info32, shm_swp);
1270                 CP(u.shm_info, u32.shm_info32, swap_attempts);
1271                 CP(u.shm_info, u32.shm_info32, swap_successes);
1272                 error = copyout(&u32.shm_info32, uap->buf,
1273                     sizeof(u32.shm_info32));
1274                 break;
1275         case SHM_STAT:
1276         case IPC_STAT:
1277                 freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1278                     &u32.shmid_ds32.shm_perm);
1279                 if (u.shmid_ds.shm_segsz > INT32_MAX)
1280                         u32.shmid_ds32.shm_segsz = INT32_MAX;
1281                 else
1282                         CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1283                 CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1284                 CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1285                 CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1286                 CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1287                 CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1288                 CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1289                 error = copyout(&u32.shmid_ds32, uap->buf,
1290                     sizeof(u32.shmid_ds32));
1291                 break;
1292         }
1293
1294 done:
1295         if (error) {
1296                 /* Invalidate the return value */
1297                 td->td_retval[0] = -1;
1298         }
1299         return (error);
1300 }
1301 #endif
1302
1303 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1304     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1305
1306 #ifndef CP
1307 #define CP(src, dst, fld)       do { (dst).fld = (src).fld; } while (0)
1308 #endif
1309
1310 #ifndef _SYS_SYSPROTO_H_
1311 struct freebsd7_shmctl_args {
1312         int shmid;
1313         int cmd;
1314         struct shmid_ds_old *buf;
1315 };
1316 #endif
1317 int
1318 freebsd7_shmctl(td, uap)
1319         struct thread *td;
1320         struct freebsd7_shmctl_args *uap;
1321 {
1322         int error = 0;
1323         struct shmid_ds_old old;
1324         struct shmid_ds buf;
1325         size_t bufsz;
1326
1327         /*
1328          * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1329          * Linux binaries.  If we see the call come through the FreeBSD ABI,
1330          * return an error back to the user since we do not to support this.
1331          */
1332         if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1333             uap->cmd == SHM_STAT)
1334                 return (EINVAL);
1335
1336         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1337         if (uap->cmd == IPC_SET) {
1338                 if ((error = copyin(uap->buf, &old, sizeof(old))))
1339                         goto done;
1340                 ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1341                 CP(old, buf, shm_segsz);
1342                 CP(old, buf, shm_lpid);
1343                 CP(old, buf, shm_cpid);
1344                 CP(old, buf, shm_nattch);
1345                 CP(old, buf, shm_atime);
1346                 CP(old, buf, shm_dtime);
1347                 CP(old, buf, shm_ctime);
1348         }
1349
1350         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1351         if (error)
1352                 goto done;
1353
1354         /* Cases in which we need to copyout */
1355         switch (uap->cmd) {
1356         case IPC_STAT:
1357                 ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1358                 if (buf.shm_segsz > INT_MAX)
1359                         old.shm_segsz = INT_MAX;
1360                 else
1361                         CP(buf, old, shm_segsz);
1362                 CP(buf, old, shm_lpid);
1363                 CP(buf, old, shm_cpid);
1364                 if (buf.shm_nattch > SHRT_MAX)
1365                         old.shm_nattch = SHRT_MAX;
1366                 else
1367                         CP(buf, old, shm_nattch);
1368                 CP(buf, old, shm_atime);
1369                 CP(buf, old, shm_dtime);
1370                 CP(buf, old, shm_ctime);
1371                 old.shm_internal = NULL;
1372                 error = copyout(&old, uap->buf, sizeof(old));
1373                 break;
1374         }
1375
1376 done:
1377         if (error) {
1378                 /* Invalidate the return value */
1379                 td->td_retval[0] = -1;
1380         }
1381         return (error);
1382 }
1383
1384 #endif  /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1385            COMPAT_FREEBSD7 */
1386
1387 static int
1388 sysvshm_modload(struct module *module, int cmd, void *arg)
1389 {
1390         int error = 0;
1391
1392         switch (cmd) {
1393         case MOD_LOAD:
1394                 error = shminit();
1395                 if (error != 0)
1396                         shmunload();
1397                 break;
1398         case MOD_UNLOAD:
1399                 error = shmunload();
1400                 break;
1401         case MOD_SHUTDOWN:
1402                 break;
1403         default:
1404                 error = EINVAL;
1405                 break;
1406         }
1407         return (error);
1408 }
1409
1410 static moduledata_t sysvshm_mod = {
1411         "sysvshm",
1412         &sysvshm_modload,
1413         NULL
1414 };
1415
1416 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1417 MODULE_VERSION(sysvshm, 1);