sys/kern/sysv_shm.c

   1 /* $FreeBSD: src/sys/kern/sysv_shm.c,v 1.45.2.6 2002/10/22 20:45:03 fjoe Exp $ */
   2 /* $DragonFly: src/sys/kern/sysv_shm.c,v 1.18 2006/09/05 00:55:45 dillon Exp $ */
   3 /*      $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $      */
   4
   5 /*
   6  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by Adam Glass and Charles
  19  *      Hannum.
  20  * 4. The names of the authors may not be used to endorse or promote products
  21  *    derived from this software without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  26  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include "opt_compat.h"
  36 #include "opt_sysvipc.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/sysproto.h>
  41 #include <sys/kernel.h>
  42 #include <sys/sysctl.h>
  43 #include <sys/shm.h>
  44 #include <sys/proc.h>
  45 #include <sys/malloc.h>
  46 #include <sys/mman.h>
  47 #include <sys/stat.h>
  48 #include <sys/sysent.h>
  49 #include <sys/jail.h>
  50
  51 #include <vm/vm.h>
  52 #include <vm/vm_param.h>
  53 #include <sys/lock.h>
  54 #include <vm/pmap.h>
  55 #include <vm/vm_object.h>
  56 #include <vm/vm_map.h>
  57 #include <vm/vm_page.h>
  58 #include <vm/vm_pager.h>
  59
  60 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
  61
  62 struct oshmctl_args;
  63 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
  64
  65 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
  66 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
  67
  68 /* XXX casting to (sy_call_t *) is bogus, as usual. */
  69 static sy_call_t *shmcalls[] = {
  70         (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
  71         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
  72         (sy_call_t *)sys_shmctl
  73 };
  74
  75 #define SHMSEG_FREE             0x0200
  76 #define SHMSEG_REMOVED          0x0400
  77 #define SHMSEG_ALLOCATED        0x0800
  78 #define SHMSEG_WANTED           0x1000
  79
  80 static int shm_last_free, shm_nused, shm_committed, shmalloced;
  81 static struct shmid_ds  *shmsegs;
  82
  83 struct shm_handle {
  84         /* vm_offset_t kva; */
  85         vm_object_t shm_object;
  86 };
  87
  88 struct shmmap_state {
  89         vm_offset_t va;
  90         int shmid;
  91 };
  92
  93 static void shm_deallocate_segment (struct shmid_ds *);
  94 static int shm_find_segment_by_key (key_t);
  95 static struct shmid_ds *shm_find_segment_by_shmid (int);
  96 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
  97 static void shmrealloc (void);
  98 static void shminit (void *);
  99
 100 /*
 101  * Tuneable values
 102  */
 103 #ifndef SHMMAXPGS
 104 #define SHMMAXPGS       8192    /* note: sysv shared memory is swap backed */
 105 #endif
 106 #ifndef SHMMAX
 107 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 108 #endif
 109 #ifndef SHMMIN
 110 #define SHMMIN  1
 111 #endif
 112 #ifndef SHMMNI
 113 #define SHMMNI  192
 114 #endif
 115 #ifndef SHMSEG
 116 #define SHMSEG  128
 117 #endif
 118 #ifndef SHMALL
 119 #define SHMALL  (SHMMAXPGS)
 120 #endif
 121
 122 struct  shminfo shminfo = {
 123         SHMMAX,
 124         SHMMIN,
 125         SHMMNI,
 126         SHMSEG,
 127         SHMALL
 128 };
 129
 130 static int shm_use_phys;
 131
 132 TUNABLE_INT("kern.ipc.shmmin", &shminfo.shmmin);
 133 TUNABLE_INT("kern.ipc.shmmni", &shminfo.shmmni);
 134 TUNABLE_INT("kern.ipc.shmseg", &shminfo.shmseg);
 135 TUNABLE_INT("kern.ipc.shmmaxpgs", &shminfo.shmall);
 136 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
 137
 138 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
 139 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
 140 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
 141 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
 142 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
 143 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
 144
 145 static int
 146 shm_find_segment_by_key(key)
 147         key_t key;
 148 {
 149         int i;
 150
 151         for (i = 0; i < shmalloced; i++)
 152                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 153                     shmsegs[i].shm_perm.key == key)
 154                         return i;
 155         return -1;
 156 }
 157
 158 static struct shmid_ds *
 159 shm_find_segment_by_shmid(shmid)
 160         int shmid;
 161 {
 162         int segnum;
 163         struct shmid_ds *shmseg;
 164
 165         segnum = IPCID_TO_IX(shmid);
 166         if (segnum < 0 || segnum >= shmalloced)
 167                 return NULL;
 168         shmseg = &shmsegs[segnum];
 169         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
 170             != SHMSEG_ALLOCATED ||
 171             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
 172                 return NULL;
 173         return shmseg;
 174 }
 175
 176 static void
 177 shm_deallocate_segment(shmseg)
 178         struct shmid_ds *shmseg;
 179 {
 180         struct shm_handle *shm_handle;
 181         size_t size;
 182
 183         shm_handle = shmseg->shm_internal;
 184         vm_object_deallocate(shm_handle->shm_object);
 185         kfree((caddr_t)shm_handle, M_SHM);
 186         shmseg->shm_internal = NULL;
 187         size = round_page(shmseg->shm_segsz);
 188         shm_committed -= btoc(size);
 189         shm_nused--;
 190         shmseg->shm_perm.mode = SHMSEG_FREE;
 191 }
 192
 193 static int
 194 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 195 {
 196         struct shmid_ds *shmseg;
 197         int segnum, result;
 198         size_t size;
 199
 200         segnum = IPCID_TO_IX(shmmap_s->shmid);
 201         shmseg = &shmsegs[segnum];
 202         size = round_page(shmseg->shm_segsz);
 203         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 204         if (result != KERN_SUCCESS)
 205                 return EINVAL;
 206         shmmap_s->shmid = -1;
 207         shmseg->shm_dtime = time_second;
 208         if ((--shmseg->shm_nattch <= 0) &&
 209             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 210                 shm_deallocate_segment(shmseg);
 211                 shm_last_free = segnum;
 212         }
 213         return 0;
 214 }
 215
 216 int
 217 sys_shmdt(struct shmdt_args *uap)
 218 {
 219         struct proc *p = curproc;
 220         struct shmmap_state *shmmap_s;
 221         int i;
 222
 223         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 224                 return (ENOSYS);
 225
 226         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 227         if (shmmap_s == NULL)
 228             return EINVAL;
 229         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
 230                 if (shmmap_s->shmid != -1 &&
 231                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 232                         break;
 233         if (i == shminfo.shmseg)
 234                 return EINVAL;
 235         return shm_delete_mapping(p->p_vmspace, shmmap_s);
 236 }
 237
 238 int
 239 sys_shmat(struct shmat_args *uap)
 240 {
 241         struct proc *p = curproc;
 242         int error, i, flags;
 243         struct shmid_ds *shmseg;
 244         struct shmmap_state *shmmap_s = NULL;
 245         struct shm_handle *shm_handle;
 246         vm_offset_t attach_va;
 247         vm_prot_t prot;
 248         vm_size_t size;
 249         int rv;
 250
 251         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 252                 return (ENOSYS);
 253
 254         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 255         if (shmmap_s == NULL) {
 256                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 257                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 258                 for (i = 0; i < shminfo.shmseg; i++)
 259                         shmmap_s[i].shmid = -1;
 260                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 261         }
 262         shmseg = shm_find_segment_by_shmid(uap->shmid);
 263         if (shmseg == NULL)
 264                 return EINVAL;
 265         error = ipcperm(p, &shmseg->shm_perm,
 266             (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 267         if (error)
 268                 return error;
 269         for (i = 0; i < shminfo.shmseg; i++) {
 270                 if (shmmap_s->shmid == -1)
 271                         break;
 272                 shmmap_s++;
 273         }
 274         if (i >= shminfo.shmseg)
 275                 return EMFILE;
 276         size = round_page(shmseg->shm_segsz);
 277 #ifdef VM_PROT_READ_IS_EXEC
 278         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 279 #else
 280         prot = VM_PROT_READ;
 281 #endif
 282         if ((uap->shmflg & SHM_RDONLY) == 0)
 283                 prot |= VM_PROT_WRITE;
 284         flags = MAP_ANON | MAP_SHARED;
 285         if (uap->shmaddr) {
 286                 flags |= MAP_FIXED;
 287                 if (uap->shmflg & SHM_RND)
 288                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
 289                 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
 290                         attach_va = (vm_offset_t)uap->shmaddr;
 291                 else
 292                         return EINVAL;
 293         } else {
 294                 /* This is just a hint to vm_map_find() about where to put it. */
 295                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
 296         }
 297
 298         shm_handle = shmseg->shm_internal;
 299         vm_object_reference(shm_handle->shm_object);
 300         rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
 301                 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
 302         if (rv != KERN_SUCCESS) {
 303                 vm_object_deallocate(shm_handle->shm_object);
 304                 return ENOMEM;
 305         }
 306         vm_map_inherit(&p->p_vmspace->vm_map,
 307                 attach_va, attach_va + size, VM_INHERIT_SHARE);
 308
 309         shmmap_s->va = attach_va;
 310         shmmap_s->shmid = uap->shmid;
 311         shmseg->shm_lpid = p->p_pid;
 312         shmseg->shm_atime = time_second;
 313         shmseg->shm_nattch++;
 314         uap->sysmsg_result = attach_va;
 315         return 0;
 316 }
 317
 318 struct oshmid_ds {
 319         struct  ipc_perm shm_perm;      /* operation perms */
 320         int     shm_segsz;              /* size of segment (bytes) */
 321         ushort  shm_cpid;               /* pid, creator */
 322         ushort  shm_lpid;               /* pid, last operation */
 323         short   shm_nattch;             /* no. of current attaches */
 324         time_t  shm_atime;              /* last attach time */
 325         time_t  shm_dtime;              /* last detach time */
 326         time_t  shm_ctime;              /* last change time */
 327         void    *shm_handle;            /* internal handle for shm segment */
 328 };
 329
 330 struct oshmctl_args {
 331         struct sysmsg sysmsg;
 332         int shmid;
 333         int cmd;
 334         struct oshmid_ds *ubuf;
 335 };
 336
 337 static int
 338 sys_oshmctl(p, uap)
 339         struct proc *p;
 340         struct oshmctl_args *uap;
 341 {
 342 #ifdef COMPAT_43
 343         int error;
 344         struct shmid_ds *shmseg;
 345         struct oshmid_ds outbuf;
 346
 347         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 348                 return (ENOSYS);
 349
 350         shmseg = shm_find_segment_by_shmid(uap->shmid);
 351         if (shmseg == NULL)
 352                 return EINVAL;
 353         switch (uap->cmd) {
 354         case IPC_STAT:
 355                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 356                 if (error)
 357                         return error;
 358                 outbuf.shm_perm = shmseg->shm_perm;
 359                 outbuf.shm_segsz = shmseg->shm_segsz;
 360                 outbuf.shm_cpid = shmseg->shm_cpid;
 361                 outbuf.shm_lpid = shmseg->shm_lpid;
 362                 outbuf.shm_nattch = shmseg->shm_nattch;
 363                 outbuf.shm_atime = shmseg->shm_atime;
 364                 outbuf.shm_dtime = shmseg->shm_dtime;
 365                 outbuf.shm_ctime = shmseg->shm_ctime;
 366                 outbuf.shm_handle = shmseg->shm_internal;
 367                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
 368                 if (error)
 369                         return error;
 370                 break;
 371         default:
 372                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 373                 return (sys_shmctl((struct shmctl_args *)uap));
 374         }
 375         return 0;
 376 #else
 377         return EINVAL;
 378 #endif
 379 }
 380
 381 int
 382 sys_shmctl(struct shmctl_args *uap)
 383 {
 384         struct proc *p = curproc;
 385         int error;
 386         struct shmid_ds inbuf;
 387         struct shmid_ds *shmseg;
 388
 389         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 390                 return (ENOSYS);
 391
 392         shmseg = shm_find_segment_by_shmid(uap->shmid);
 393         if (shmseg == NULL)
 394                 return EINVAL;
 395         switch (uap->cmd) {
 396         case IPC_STAT:
 397                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 398                 if (error)
 399                         return error;
 400                 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf));
 401                 if (error)
 402                         return error;
 403                 break;
 404         case IPC_SET:
 405                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 406                 if (error)
 407                         return error;
 408                 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf));
 409                 if (error)
 410                         return error;
 411                 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
 412                 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
 413                 shmseg->shm_perm.mode =
 414                     (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 415                     (inbuf.shm_perm.mode & ACCESSPERMS);
 416                 shmseg->shm_ctime = time_second;
 417                 break;
 418         case IPC_RMID:
 419                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 420                 if (error)
 421                         return error;
 422                 shmseg->shm_perm.key = IPC_PRIVATE;
 423                 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 424                 if (shmseg->shm_nattch <= 0) {
 425                         shm_deallocate_segment(shmseg);
 426                         shm_last_free = IPCID_TO_IX(uap->shmid);
 427                 }
 428                 break;
 429 #if 0
 430         case SHM_LOCK:
 431         case SHM_UNLOCK:
 432 #endif
 433         default:
 434                 return EINVAL;
 435         }
 436         return 0;
 437 }
 438
 439 static int
 440 shmget_existing(p, uap, mode, segnum)
 441         struct proc *p;
 442         struct shmget_args *uap;
 443         int mode;
 444         int segnum;
 445 {
 446         struct shmid_ds *shmseg;
 447         int error;
 448
 449         shmseg = &shmsegs[segnum];
 450         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 451                 /*
 452                  * This segment is in the process of being allocated.  Wait
 453                  * until it's done, and look the key up again (in case the
 454                  * allocation failed or it was freed).
 455                  */
 456                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 457                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 458                 if (error)
 459                         return error;
 460                 return EAGAIN;
 461         }
 462         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 463                 return EEXIST;
 464         error = ipcperm(p, &shmseg->shm_perm, mode);
 465         if (error)
 466                 return error;
 467         if (uap->size && uap->size > shmseg->shm_segsz)
 468                 return EINVAL;
 469         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 470         return 0;
 471 }
 472
 473 static int
 474 shmget_allocate_segment(p, uap, mode)
 475         struct proc *p;
 476         struct shmget_args *uap;
 477         int mode;
 478 {
 479         int i, segnum, shmid, size;
 480         struct ucred *cred = p->p_ucred;
 481         struct shmid_ds *shmseg;
 482         struct shm_handle *shm_handle;
 483
 484         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 485                 return EINVAL;
 486         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 487                 return ENOSPC;
 488         size = round_page(uap->size);
 489         if (shm_committed + btoc(size) > shminfo.shmall)
 490                 return ENOMEM;
 491         if (shm_last_free < 0) {
 492                 shmrealloc();   /* maybe expand the shmsegs[] array */
 493                 for (i = 0; i < shmalloced; i++)
 494                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 495                                 break;
 496                 if (i == shmalloced)
 497                         return ENOSPC;
 498                 segnum = i;
 499         } else  {
 500                 segnum = shm_last_free;
 501                 shm_last_free = -1;
 502         }
 503         shmseg = &shmsegs[segnum];
 504         /*
 505          * In case we sleep in malloc(), mark the segment present but deleted
 506          * so that noone else tries to create the same key.
 507          */
 508         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 509         shmseg->shm_perm.key = uap->key;
 510         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 511         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 512         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 513
 514         /*
 515          * We make sure that we have allocated a pager before we need
 516          * to.
 517          */
 518         if (shm_use_phys) {
 519                 shm_handle->shm_object =
 520                     vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
 521         } else {
 522                 shm_handle->shm_object =
 523                     vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
 524         }
 525         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 526         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 527
 528         shmseg->shm_internal = shm_handle;
 529         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 530         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 531         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 532             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 533         shmseg->shm_segsz = uap->size;
 534         shmseg->shm_cpid = p->p_pid;
 535         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 536         shmseg->shm_atime = shmseg->shm_dtime = 0;
 537         shmseg->shm_ctime = time_second;
 538         shm_committed += btoc(size);
 539         shm_nused++;
 540         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 541                 /*
 542                  * Somebody else wanted this key while we were asleep.  Wake
 543                  * them up now.
 544                  */
 545                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 546                 wakeup((caddr_t)shmseg);
 547         }
 548         uap->sysmsg_result = shmid;
 549         return 0;
 550 }
 551
 552 int
 553 sys_shmget(struct shmget_args *uap)
 554 {
 555         struct proc *p = curproc;
 556         int segnum, mode, error;
 557
 558         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 559                 return (ENOSYS);
 560
 561         mode = uap->shmflg & ACCESSPERMS;
 562         if (uap->key != IPC_PRIVATE) {
 563         again:
 564                 segnum = shm_find_segment_by_key(uap->key);
 565                 if (segnum >= 0) {
 566                         error = shmget_existing(p, uap, mode, segnum);
 567                         if (error == EAGAIN)
 568                                 goto again;
 569                         return error;
 570                 }
 571                 if ((uap->shmflg & IPC_CREAT) == 0)
 572                         return ENOENT;
 573         }
 574         return shmget_allocate_segment(p, uap, mode);
 575 }
 576
 577 /*
 578  *  shmsys_args(int which, int a2, ...) (VARARGS)
 579  */
 580 int
 581 sys_shmsys(struct shmsys_args *uap)
 582 {
 583         struct proc *p = curproc;
 584         unsigned int which = (unsigned int)uap->which;
 585         int error;
 586
 587         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
 588                 return (ENOSYS);
 589
 590         if (which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
 591                 return EINVAL;
 592         bcopy(&uap->a2, &uap->which,
 593                 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
 594         error = ((*shmcalls[which])(uap));
 595         return(error);
 596 }
 597
 598 void
 599 shmfork(p1, p2)
 600         struct proc *p1, *p2;
 601 {
 602         struct shmmap_state *shmmap_s;
 603         size_t size;
 604         int i;
 605
 606         size = shminfo.shmseg * sizeof(struct shmmap_state);
 607         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 608         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 609         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 610         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
 611                 if (shmmap_s->shmid != -1)
 612                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 613 }
 614
 615 void
 616 shmexit(struct vmspace *vm)
 617 {
 618         struct shmmap_state *base, *shm;
 619         int i;
 620
 621         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 622                 vm->vm_shm = NULL;
 623                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 624                         if (shm->shmid != -1)
 625                                 shm_delete_mapping(vm, shm);
 626                 }
 627                 kfree(base, M_SHM);
 628         }
 629 }
 630
 631 static void
 632 shmrealloc(void)
 633 {
 634         int i;
 635         struct shmid_ds *newsegs;
 636
 637         if (shmalloced >= shminfo.shmmni)
 638                 return;
 639
 640         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 641         if (newsegs == NULL)
 642                 return;
 643         for (i = 0; i < shmalloced; i++)
 644                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 645         for (; i < shminfo.shmmni; i++) {
 646                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 647                 shmsegs[i].shm_perm.seq = 0;
 648         }
 649         kfree(shmsegs, M_SHM);
 650         shmsegs = newsegs;
 651         shmalloced = shminfo.shmmni;
 652 }
 653
 654 static void
 655 shminit(dummy)
 656         void *dummy;
 657 {
 658         int i;
 659
 660         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 661         shmalloced = shminfo.shmmni;
 662         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 663         if (shmsegs == NULL)
 664                 panic("cannot allocate initial memory for sysvshm");
 665         for (i = 0; i < shmalloced; i++) {
 666                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 667                 shmsegs[i].shm_perm.seq = 0;
 668         }
 669         shm_last_free = 0;
 670         shm_nused = 0;
 671         shm_committed = 0;
 672 }
 673 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);