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 1.21 2008/01/06 16:55:51 swildner 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_t key)
 147 {
 148         int i;
 149
 150         for (i = 0; i < shmalloced; i++)
 151                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 152                     shmsegs[i].shm_perm.key == key)
 153                         return i;
 154         return -1;
 155 }
 156
 157 static struct shmid_ds *
 158 shm_find_segment_by_shmid(int shmid)
 159 {
 160         int segnum;
 161         struct shmid_ds *shmseg;
 162
 163         segnum = IPCID_TO_IX(shmid);
 164         if (segnum < 0 || segnum >= shmalloced)
 165                 return NULL;
 166         shmseg = &shmsegs[segnum];
 167         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
 168             != SHMSEG_ALLOCATED ||
 169             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
 170                 return NULL;
 171         return shmseg;
 172 }
 173
 174 static void
 175 shm_deallocate_segment(struct shmid_ds *shmseg)
 176 {
 177         struct shm_handle *shm_handle;
 178         size_t size;
 179
 180         shm_handle = shmseg->shm_internal;
 181         vm_object_deallocate(shm_handle->shm_object);
 182         kfree((caddr_t)shm_handle, M_SHM);
 183         shmseg->shm_internal = NULL;
 184         size = round_page(shmseg->shm_segsz);
 185         shm_committed -= btoc(size);
 186         shm_nused--;
 187         shmseg->shm_perm.mode = SHMSEG_FREE;
 188 }
 189
 190 static int
 191 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 192 {
 193         struct shmid_ds *shmseg;
 194         int segnum, result;
 195         size_t size;
 196
 197         segnum = IPCID_TO_IX(shmmap_s->shmid);
 198         shmseg = &shmsegs[segnum];
 199         size = round_page(shmseg->shm_segsz);
 200         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 201         if (result != KERN_SUCCESS)
 202                 return EINVAL;
 203         shmmap_s->shmid = -1;
 204         shmseg->shm_dtime = time_second;
 205         if ((--shmseg->shm_nattch <= 0) &&
 206             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 207                 shm_deallocate_segment(shmseg);
 208                 shm_last_free = segnum;
 209         }
 210         return 0;
 211 }
 212
 213 /*
 214  * MPALMOSTSAFE
 215  */
 216 int
 217 sys_shmdt(struct shmdt_args *uap)
 218 {
 219         struct thread *td = curthread;
 220         struct proc *p = td->td_proc;
 221         struct shmmap_state *shmmap_s;
 222         int i;
 223         int error;
 224
 225         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 226                 return (ENOSYS);
 227
 228         get_mplock();
 229         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 230         if (shmmap_s == NULL) {
 231                 error = EINVAL;
 232                 goto done;
 233         }
 234         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 235                 if (shmmap_s->shmid != -1 &&
 236                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 237                         break;
 238         }
 239         if (i == shminfo.shmseg)
 240                 error = EINVAL;
 241         else
 242                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 243 done:
 244         rel_mplock();
 245         return (error);
 246 }
 247
 248 /*
 249  * MPALMOSTSAFE
 250  */
 251 int
 252 sys_shmat(struct shmat_args *uap)
 253 {
 254         struct thread *td = curthread;
 255         struct proc *p = td->td_proc;
 256         int error, i, flags;
 257         struct shmid_ds *shmseg;
 258         struct shmmap_state *shmmap_s = NULL;
 259         struct shm_handle *shm_handle;
 260         vm_offset_t attach_va;
 261         vm_prot_t prot;
 262         vm_size_t size;
 263         int rv;
 264
 265         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 266                 return (ENOSYS);
 267
 268         get_mplock();
 269 again:
 270         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 271         if (shmmap_s == NULL) {
 272                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 273                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 274                 for (i = 0; i < shminfo.shmseg; i++)
 275                         shmmap_s[i].shmid = -1;
 276                 if (p->p_vmspace->vm_shm != NULL) {
 277                         kfree(shmmap_s, M_SHM);
 278                         goto again;
 279                 }
 280                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 281         }
 282         shmseg = shm_find_segment_by_shmid(uap->shmid);
 283         if (shmseg == NULL) {
 284                 error = EINVAL;
 285                 goto done;
 286         }
 287         error = ipcperm(p, &shmseg->shm_perm,
 288                         (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 289         if (error)
 290                 goto done;
 291         for (i = 0; i < shminfo.shmseg; i++) {
 292                 if (shmmap_s->shmid == -1)
 293                         break;
 294                 shmmap_s++;
 295         }
 296         if (i >= shminfo.shmseg) {
 297                 error = EMFILE;
 298                 goto done;
 299         }
 300         size = round_page(shmseg->shm_segsz);
 301 #ifdef VM_PROT_READ_IS_EXEC
 302         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 303 #else
 304         prot = VM_PROT_READ;
 305 #endif
 306         if ((uap->shmflg & SHM_RDONLY) == 0)
 307                 prot |= VM_PROT_WRITE;
 308         flags = MAP_ANON | MAP_SHARED;
 309         if (uap->shmaddr) {
 310                 flags |= MAP_FIXED;
 311                 if (uap->shmflg & SHM_RND) {
 312                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
 313                 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
 314                         attach_va = (vm_offset_t)uap->shmaddr;
 315                 } else {
 316                         error = EINVAL;
 317                         goto done;
 318                 }
 319         } else {
 320                 /*
 321                  * This is just a hint to vm_map_find() about where to put it.
 322                  */
 323                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
 324         }
 325
 326         shm_handle = shmseg->shm_internal;
 327         vm_object_reference(shm_handle->shm_object);
 328         rv = vm_map_find(&p->p_vmspace->vm_map,
 329                          shm_handle->shm_object, 0,
 330                          &attach_va, size,
 331                          ((flags & MAP_FIXED) ? 0 : 1),
 332                          VM_MAPTYPE_NORMAL,
 333                          prot, prot,
 334                          0);
 335         if (rv != KERN_SUCCESS) {
 336                 vm_object_deallocate(shm_handle->shm_object);
 337                 error = ENOMEM;
 338                 goto done;
 339         }
 340         vm_map_inherit(&p->p_vmspace->vm_map,
 341                        attach_va, attach_va + size, VM_INHERIT_SHARE);
 342
 343         KKASSERT(shmmap_s->shmid == -1);
 344         shmmap_s->va = attach_va;
 345         shmmap_s->shmid = uap->shmid;
 346         shmseg->shm_lpid = p->p_pid;
 347         shmseg->shm_atime = time_second;
 348         shmseg->shm_nattch++;
 349         uap->sysmsg_result = attach_va;
 350         error = 0;
 351 done:
 352         rel_mplock();
 353         return error;
 354 }
 355
 356 struct oshmid_ds {
 357         struct  ipc_perm shm_perm;      /* operation perms */
 358         int     shm_segsz;              /* size of segment (bytes) */
 359         ushort  shm_cpid;               /* pid, creator */
 360         ushort  shm_lpid;               /* pid, last operation */
 361         short   shm_nattch;             /* no. of current attaches */
 362         time_t  shm_atime;              /* last attach time */
 363         time_t  shm_dtime;              /* last detach time */
 364         time_t  shm_ctime;              /* last change time */
 365         void    *shm_handle;            /* internal handle for shm segment */
 366 };
 367
 368 struct oshmctl_args {
 369         struct sysmsg sysmsg;
 370         int shmid;
 371         int cmd;
 372         struct oshmid_ds *ubuf;
 373 };
 374
 375 /*
 376  * MPALMOSTSAFE
 377  */
 378 static int
 379 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
 380 {
 381 #ifdef COMPAT_43
 382         struct thread *td = curthread;
 383         struct shmid_ds *shmseg;
 384         struct oshmid_ds outbuf;
 385         int error;
 386
 387         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 388                 return (ENOSYS);
 389
 390         get_mplock();
 391         shmseg = shm_find_segment_by_shmid(uap->shmid);
 392         if (shmseg == NULL) {
 393                 error = EINVAL;
 394                 goto done;
 395         }
 396
 397         switch (uap->cmd) {
 398         case IPC_STAT:
 399                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 400                 if (error)
 401                         break;
 402                 outbuf.shm_perm = shmseg->shm_perm;
 403                 outbuf.shm_segsz = shmseg->shm_segsz;
 404                 outbuf.shm_cpid = shmseg->shm_cpid;
 405                 outbuf.shm_lpid = shmseg->shm_lpid;
 406                 outbuf.shm_nattch = shmseg->shm_nattch;
 407                 outbuf.shm_atime = shmseg->shm_atime;
 408                 outbuf.shm_dtime = shmseg->shm_dtime;
 409                 outbuf.shm_ctime = shmseg->shm_ctime;
 410                 outbuf.shm_handle = shmseg->shm_internal;
 411                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
 412                 break;
 413         default:
 414                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 415                 error = sys_shmctl((struct shmctl_args *)uap);
 416         }
 417 done:
 418         rel_mplock();
 419         return error;
 420 #else
 421         return EINVAL;
 422 #endif
 423 }
 424
 425 /*
 426  * MPALMOSTSAFE
 427  */
 428 int
 429 sys_shmctl(struct shmctl_args *uap)
 430 {
 431         struct thread *td = curthread;
 432         struct proc *p = td->td_proc;
 433         int error;
 434         struct shmid_ds inbuf;
 435         struct shmid_ds *shmseg;
 436
 437         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 438                 return (ENOSYS);
 439
 440         get_mplock();
 441         shmseg = shm_find_segment_by_shmid(uap->shmid);
 442         if (shmseg == NULL) {
 443                 error = EINVAL;
 444                 goto done;
 445         }
 446
 447         switch (uap->cmd) {
 448         case IPC_STAT:
 449                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 450                 if (error == 0)
 451                         error = copyout(shmseg, uap->buf, sizeof(inbuf));
 452                 break;
 453         case IPC_SET:
 454                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 455                 if (error == 0)
 456                         error = copyin(uap->buf, &inbuf, sizeof(inbuf));
 457                 if (error == 0) {
 458                         shmseg->shm_perm.uid = inbuf.shm_perm.uid;
 459                         shmseg->shm_perm.gid = inbuf.shm_perm.gid;
 460                         shmseg->shm_perm.mode =
 461                             (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 462                             (inbuf.shm_perm.mode & ACCESSPERMS);
 463                         shmseg->shm_ctime = time_second;
 464                 }
 465                 break;
 466         case IPC_RMID:
 467                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 468                 if (error == 0) {
 469                         shmseg->shm_perm.key = IPC_PRIVATE;
 470                         shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 471                         if (shmseg->shm_nattch <= 0) {
 472                                 shm_deallocate_segment(shmseg);
 473                                 shm_last_free = IPCID_TO_IX(uap->shmid);
 474                         }
 475                 }
 476                 break;
 477 #if 0
 478         case SHM_LOCK:
 479         case SHM_UNLOCK:
 480 #endif
 481         default:
 482                 error = EINVAL;
 483                 break;
 484         }
 485 done:
 486         rel_mplock();
 487         return error;
 488 }
 489
 490 static int
 491 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
 492 {
 493         struct shmid_ds *shmseg;
 494         int error;
 495
 496         shmseg = &shmsegs[segnum];
 497         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 498                 /*
 499                  * This segment is in the process of being allocated.  Wait
 500                  * until it's done, and look the key up again (in case the
 501                  * allocation failed or it was freed).
 502                  */
 503                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 504                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 505                 if (error)
 506                         return error;
 507                 return EAGAIN;
 508         }
 509         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 510                 return EEXIST;
 511         error = ipcperm(p, &shmseg->shm_perm, mode);
 512         if (error)
 513                 return error;
 514         if (uap->size && uap->size > shmseg->shm_segsz)
 515                 return EINVAL;
 516         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 517         return 0;
 518 }
 519
 520 static int
 521 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
 522 {
 523         int i, segnum, shmid, size;
 524         struct ucred *cred = p->p_ucred;
 525         struct shmid_ds *shmseg;
 526         struct shm_handle *shm_handle;
 527
 528         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 529                 return EINVAL;
 530         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 531                 return ENOSPC;
 532         size = round_page(uap->size);
 533         if (shm_committed + btoc(size) > shminfo.shmall)
 534                 return ENOMEM;
 535         if (shm_last_free < 0) {
 536                 shmrealloc();   /* maybe expand the shmsegs[] array */
 537                 for (i = 0; i < shmalloced; i++)
 538                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 539                                 break;
 540                 if (i == shmalloced)
 541                         return ENOSPC;
 542                 segnum = i;
 543         } else  {
 544                 segnum = shm_last_free;
 545                 shm_last_free = -1;
 546         }
 547         shmseg = &shmsegs[segnum];
 548         /*
 549          * In case we sleep in malloc(), mark the segment present but deleted
 550          * so that noone else tries to create the same key.
 551          */
 552         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 553         shmseg->shm_perm.key = uap->key;
 554         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 555         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 556         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 557
 558         /*
 559          * We make sure that we have allocated a pager before we need
 560          * to.
 561          */
 562         if (shm_use_phys) {
 563                 shm_handle->shm_object =
 564                     vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
 565         } else {
 566                 shm_handle->shm_object =
 567                     vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
 568         }
 569         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 570         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 571
 572         shmseg->shm_internal = shm_handle;
 573         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 574         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 575         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 576             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 577         shmseg->shm_segsz = uap->size;
 578         shmseg->shm_cpid = p->p_pid;
 579         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 580         shmseg->shm_atime = shmseg->shm_dtime = 0;
 581         shmseg->shm_ctime = time_second;
 582         shm_committed += btoc(size);
 583         shm_nused++;
 584         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 585                 /*
 586                  * Somebody else wanted this key while we were asleep.  Wake
 587                  * them up now.
 588                  */
 589                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 590                 wakeup((caddr_t)shmseg);
 591         }
 592         uap->sysmsg_result = shmid;
 593         return 0;
 594 }
 595
 596 /*
 597  * MPALMOSTSAFE
 598  */
 599 int
 600 sys_shmget(struct shmget_args *uap)
 601 {
 602         struct thread *td = curthread;
 603         struct proc *p = td->td_proc;
 604         int segnum, mode, error;
 605
 606         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 607                 return (ENOSYS);
 608
 609         mode = uap->shmflg & ACCESSPERMS;
 610         get_mplock();
 611
 612         if (uap->key != IPC_PRIVATE) {
 613         again:
 614                 segnum = shm_find_segment_by_key(uap->key);
 615                 if (segnum >= 0) {
 616                         error = shmget_existing(p, uap, mode, segnum);
 617                         if (error == EAGAIN)
 618                                 goto again;
 619                         goto done;
 620                 }
 621                 if ((uap->shmflg & IPC_CREAT) == 0) {
 622                         error = ENOENT;
 623                         goto done;
 624                 }
 625         }
 626         error = shmget_allocate_segment(p, uap, mode);
 627 done:
 628         rel_mplock();
 629         return (error);
 630 }
 631
 632 /*
 633  * shmsys_args(int which, int a2, ...) (VARARGS)
 634  *
 635  * MPALMOSTSAFE
 636  */
 637 int
 638 sys_shmsys(struct shmsys_args *uap)
 639 {
 640         struct thread *td = curthread;
 641         unsigned int which = (unsigned int)uap->which;
 642         int error;
 643
 644         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 645                 return (ENOSYS);
 646
 647         if (which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
 648                 return EINVAL;
 649         get_mplock();
 650         bcopy(&uap->a2, &uap->which,
 651                 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
 652         error = ((*shmcalls[which])(uap));
 653         rel_mplock();
 654
 655         return(error);
 656 }
 657
 658 void
 659 shmfork(struct proc *p1, struct proc *p2)
 660 {
 661         struct shmmap_state *shmmap_s;
 662         size_t size;
 663         int i;
 664
 665         size = shminfo.shmseg * sizeof(struct shmmap_state);
 666         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 667         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 668         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 669         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
 670                 if (shmmap_s->shmid != -1)
 671                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 672 }
 673
 674 void
 675 shmexit(struct vmspace *vm)
 676 {
 677         struct shmmap_state *base, *shm;
 678         int i;
 679
 680         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 681                 vm->vm_shm = NULL;
 682                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 683                         if (shm->shmid != -1)
 684                                 shm_delete_mapping(vm, shm);
 685                 }
 686                 kfree(base, M_SHM);
 687         }
 688 }
 689
 690 static void
 691 shmrealloc(void)
 692 {
 693         int i;
 694         struct shmid_ds *newsegs;
 695
 696         if (shmalloced >= shminfo.shmmni)
 697                 return;
 698
 699         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 700         for (i = 0; i < shmalloced; i++)
 701                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 702         for (; i < shminfo.shmmni; i++) {
 703                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 704                 shmsegs[i].shm_perm.seq = 0;
 705         }
 706         kfree(shmsegs, M_SHM);
 707         shmsegs = newsegs;
 708         shmalloced = shminfo.shmmni;
 709 }
 710
 711 static void
 712 shminit(void *dummy)
 713 {
 714         int i;
 715
 716         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 717         shmalloced = shminfo.shmmni;
 718         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 719         for (i = 0; i < shmalloced; i++) {
 720                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 721                 shmsegs[i].shm_perm.seq = 0;
 722         }
 723         shm_last_free = 0;
 724         shm_nused = 0;
 725         shm_committed = 0;
 726 }
 727 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);