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 /*      $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $      */
   3
   4 /*
   5  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. All advertising materials mentioning features or use of this software
  16  *    must display the following acknowledgement:
  17  *      This product includes software developed by Adam Glass and Charles
  18  *      Hannum.
  19  * 4. The names of the authors may not be used to endorse or promote products
  20  *    derived from this software without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  25  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include "opt_compat.h"
  35 #include "opt_sysvipc.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/sysproto.h>
  40 #include <sys/kernel.h>
  41 #include <sys/sysctl.h>
  42 #include <sys/shm.h>
  43 #include <sys/proc.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mman.h>
  46 #include <sys/stat.h>
  47 #include <sys/sysent.h>
  48 #include <sys/jail.h>
  49
  50 #include <sys/mplock2.h>
  51
  52 #include <vm/vm.h>
  53 #include <vm/vm_param.h>
  54 #include <sys/lock.h>
  55 #include <vm/pmap.h>
  56 #include <vm/vm_object.h>
  57 #include <vm/vm_map.h>
  58 #include <vm/vm_page.h>
  59 #include <vm/vm_pager.h>
  60
  61 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
  62
  63 struct oshmctl_args;
  64 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
  65
  66 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
  67 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
  68
  69 /* XXX casting to (sy_call_t *) is bogus, as usual. */
  70 static sy_call_t *shmcalls[] = {
  71         (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
  72         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
  73         (sy_call_t *)sys_shmctl
  74 };
  75
  76 #define SHMSEG_FREE             0x0200
  77 #define SHMSEG_REMOVED          0x0400
  78 #define SHMSEG_ALLOCATED        0x0800
  79 #define SHMSEG_WANTED           0x1000
  80
  81 static int shm_last_free, shm_committed, shmalloced;
  82 int shm_nused;
  83 static struct shmid_ds  *shmsegs;
  84
  85 struct shm_handle {
  86         /* vm_offset_t kva; */
  87         vm_object_t shm_object;
  88 };
  89
  90 struct shmmap_state {
  91         vm_offset_t va;
  92         int shmid;
  93 };
  94
  95 static void shm_deallocate_segment (struct shmid_ds *);
  96 static int shm_find_segment_by_key (key_t);
  97 static struct shmid_ds *shm_find_segment_by_shmid (int);
  98 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
  99 static void shmrealloc (void);
 100 static void shminit (void *);
 101
 102 /*
 103  * Tuneable values
 104  */
 105 #ifndef SHMMAXPGS
 106 #define SHMMAXPGS       8192    /* note: sysv shared memory is swap backed */
 107 #endif
 108 #ifndef SHMMAX
 109 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 110 #endif
 111 #ifndef SHMMIN
 112 #define SHMMIN  1
 113 #endif
 114 #ifndef SHMMNI
 115 #define SHMMNI  192
 116 #endif
 117 #ifndef SHMSEG
 118 #define SHMSEG  128
 119 #endif
 120 #ifndef SHMALL
 121 #define SHMALL  (SHMMAXPGS)
 122 #endif
 123
 124 struct  shminfo shminfo = {
 125         SHMMAX,
 126         SHMMIN,
 127         SHMMNI,
 128         SHMSEG,
 129         SHMALL
 130 };
 131
 132 static int shm_use_phys;
 133
 134 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
 135 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
 136 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
 137 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
 138 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
 139
 140 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 141     "Max shared memory segment size");
 142 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 143     "Min shared memory segment size");
 144 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
 145     "Max number of shared memory identifiers");
 146 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
 147     "Max shared memory segments per process");
 148 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 149     "Max pages of shared memory");
 150 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
 151     "Use phys pager allocation instead of swap pager allocation");
 152
 153 static int
 154 shm_find_segment_by_key(key_t key)
 155 {
 156         int i;
 157
 158         for (i = 0; i < shmalloced; i++)
 159                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 160                     shmsegs[i].shm_perm.key == key)
 161                         return i;
 162         return -1;
 163 }
 164
 165 static struct shmid_ds *
 166 shm_find_segment_by_shmid(int shmid)
 167 {
 168         int segnum;
 169         struct shmid_ds *shmseg;
 170
 171         segnum = IPCID_TO_IX(shmid);
 172         if (segnum < 0 || segnum >= shmalloced)
 173                 return NULL;
 174         shmseg = &shmsegs[segnum];
 175         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
 176             != SHMSEG_ALLOCATED ||
 177             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
 178                 return NULL;
 179         return shmseg;
 180 }
 181
 182 static void
 183 shm_deallocate_segment(struct shmid_ds *shmseg)
 184 {
 185         struct shm_handle *shm_handle;
 186         size_t size;
 187
 188         shm_handle = shmseg->shm_internal;
 189         vm_object_deallocate(shm_handle->shm_object);
 190         kfree((caddr_t)shm_handle, M_SHM);
 191         shmseg->shm_internal = NULL;
 192         size = round_page(shmseg->shm_segsz);
 193         shm_committed -= btoc(size);
 194         shm_nused--;
 195         shmseg->shm_perm.mode = SHMSEG_FREE;
 196 }
 197
 198 static int
 199 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 200 {
 201         struct shmid_ds *shmseg;
 202         int segnum, result;
 203         size_t size;
 204
 205         segnum = IPCID_TO_IX(shmmap_s->shmid);
 206         shmseg = &shmsegs[segnum];
 207         size = round_page(shmseg->shm_segsz);
 208         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 209         if (result != KERN_SUCCESS)
 210                 return EINVAL;
 211         shmmap_s->shmid = -1;
 212         shmseg->shm_dtime = time_second;
 213         if ((--shmseg->shm_nattch <= 0) &&
 214             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 215                 shm_deallocate_segment(shmseg);
 216                 shm_last_free = segnum;
 217         }
 218         return 0;
 219 }
 220
 221 /*
 222  * MPALMOSTSAFE
 223  */
 224 int
 225 sys_shmdt(struct shmdt_args *uap)
 226 {
 227         struct thread *td = curthread;
 228         struct proc *p = td->td_proc;
 229         struct shmmap_state *shmmap_s;
 230         long i;
 231         int error;
 232
 233         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 234                 return (ENOSYS);
 235
 236         get_mplock();
 237         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 238         if (shmmap_s == NULL) {
 239                 error = EINVAL;
 240                 goto done;
 241         }
 242         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 243                 if (shmmap_s->shmid != -1 &&
 244                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 245                         break;
 246         }
 247         if (i == shminfo.shmseg)
 248                 error = EINVAL;
 249         else
 250                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 251 done:
 252         rel_mplock();
 253         return (error);
 254 }
 255
 256 /*
 257  * MPALMOSTSAFE
 258  */
 259 int
 260 sys_shmat(struct shmat_args *uap)
 261 {
 262         struct thread *td = curthread;
 263         struct proc *p = td->td_proc;
 264         int error, flags;
 265         long i;
 266         struct shmid_ds *shmseg;
 267         struct shmmap_state *shmmap_s = NULL;
 268         struct shm_handle *shm_handle;
 269         vm_offset_t attach_va;
 270         vm_prot_t prot;
 271         vm_size_t size;
 272         int rv;
 273
 274         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 275                 return (ENOSYS);
 276
 277         get_mplock();
 278 again:
 279         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 280         if (shmmap_s == NULL) {
 281                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 282                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 283                 for (i = 0; i < shminfo.shmseg; i++)
 284                         shmmap_s[i].shmid = -1;
 285                 if (p->p_vmspace->vm_shm != NULL) {
 286                         kfree(shmmap_s, M_SHM);
 287                         goto again;
 288                 }
 289                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 290         }
 291         shmseg = shm_find_segment_by_shmid(uap->shmid);
 292         if (shmseg == NULL) {
 293                 error = EINVAL;
 294                 goto done;
 295         }
 296         error = ipcperm(p, &shmseg->shm_perm,
 297                         (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 298         if (error)
 299                 goto done;
 300         for (i = 0; i < shminfo.shmseg; i++) {
 301                 if (shmmap_s->shmid == -1)
 302                         break;
 303                 shmmap_s++;
 304         }
 305         if (i >= shminfo.shmseg) {
 306                 error = EMFILE;
 307                 goto done;
 308         }
 309         size = round_page(shmseg->shm_segsz);
 310 #ifdef VM_PROT_READ_IS_EXEC
 311         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 312 #else
 313         prot = VM_PROT_READ;
 314 #endif
 315         if ((uap->shmflg & SHM_RDONLY) == 0)
 316                 prot |= VM_PROT_WRITE;
 317         flags = MAP_ANON | MAP_SHARED;
 318         if (uap->shmaddr) {
 319                 flags |= MAP_FIXED;
 320                 if (uap->shmflg & SHM_RND) {
 321                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
 322                 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
 323                         attach_va = (vm_offset_t)uap->shmaddr;
 324                 } else {
 325                         error = EINVAL;
 326                         goto done;
 327                 }
 328         } else {
 329                 /*
 330                  * This is just a hint to vm_map_find() about where to put it.
 331                  */
 332                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
 333         }
 334
 335         shm_handle = shmseg->shm_internal;
 336         vm_object_hold(shm_handle->shm_object);
 337         vm_object_reference_locked(shm_handle->shm_object);
 338         rv = vm_map_find(&p->p_vmspace->vm_map,
 339                          shm_handle->shm_object, 0,
 340                          &attach_va,
 341                          size, PAGE_SIZE,
 342                          ((flags & MAP_FIXED) ? 0 : 1),
 343                          VM_MAPTYPE_NORMAL,
 344                          prot, prot,
 345                          0);
 346         vm_object_drop(shm_handle->shm_object);
 347         if (rv != KERN_SUCCESS) {
 348                 vm_object_deallocate(shm_handle->shm_object);
 349                 error = ENOMEM;
 350                 goto done;
 351         }
 352         vm_map_inherit(&p->p_vmspace->vm_map,
 353                        attach_va, attach_va + size, VM_INHERIT_SHARE);
 354
 355         KKASSERT(shmmap_s->shmid == -1);
 356         shmmap_s->va = attach_va;
 357         shmmap_s->shmid = uap->shmid;
 358         shmseg->shm_lpid = p->p_pid;
 359         shmseg->shm_atime = time_second;
 360         shmseg->shm_nattch++;
 361         uap->sysmsg_resultp = (void *)attach_va;
 362         error = 0;
 363 done:
 364         rel_mplock();
 365         return error;
 366 }
 367
 368 struct oshmid_ds {
 369         struct  ipc_perm shm_perm;      /* operation perms */
 370         int     shm_segsz;              /* size of segment (bytes) */
 371         ushort  shm_cpid;               /* pid, creator */
 372         ushort  shm_lpid;               /* pid, last operation */
 373         short   shm_nattch;             /* no. of current attaches */
 374         time_t  shm_atime;              /* last attach time */
 375         time_t  shm_dtime;              /* last detach time */
 376         time_t  shm_ctime;              /* last change time */
 377         void    *shm_handle;            /* internal handle for shm segment */
 378 };
 379
 380 struct oshmctl_args {
 381         struct sysmsg sysmsg;
 382         int shmid;
 383         int cmd;
 384         struct oshmid_ds *ubuf;
 385 };
 386
 387 /*
 388  * MPALMOSTSAFE
 389  */
 390 static int
 391 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
 392 {
 393 #ifdef COMPAT_43
 394         struct thread *td = curthread;
 395         struct shmid_ds *shmseg;
 396         struct oshmid_ds outbuf;
 397         int error;
 398
 399         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 400                 return (ENOSYS);
 401
 402         get_mplock();
 403         shmseg = shm_find_segment_by_shmid(uap->shmid);
 404         if (shmseg == NULL) {
 405                 error = EINVAL;
 406                 goto done;
 407         }
 408
 409         switch (uap->cmd) {
 410         case IPC_STAT:
 411                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 412                 if (error)
 413                         break;
 414                 outbuf.shm_perm = shmseg->shm_perm;
 415                 outbuf.shm_segsz = shmseg->shm_segsz;
 416                 outbuf.shm_cpid = shmseg->shm_cpid;
 417                 outbuf.shm_lpid = shmseg->shm_lpid;
 418                 outbuf.shm_nattch = shmseg->shm_nattch;
 419                 outbuf.shm_atime = shmseg->shm_atime;
 420                 outbuf.shm_dtime = shmseg->shm_dtime;
 421                 outbuf.shm_ctime = shmseg->shm_ctime;
 422                 outbuf.shm_handle = shmseg->shm_internal;
 423                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
 424                 break;
 425         default:
 426                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 427                 error = sys_shmctl((struct shmctl_args *)uap);
 428         }
 429 done:
 430         rel_mplock();
 431         return error;
 432 #else
 433         return EINVAL;
 434 #endif
 435 }
 436
 437 /*
 438  * MPALMOSTSAFE
 439  */
 440 int
 441 sys_shmctl(struct shmctl_args *uap)
 442 {
 443         struct thread *td = curthread;
 444         struct proc *p = td->td_proc;
 445         int error;
 446         struct shmid_ds inbuf;
 447         struct shmid_ds *shmseg;
 448
 449         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 450                 return (ENOSYS);
 451
 452         get_mplock();
 453         shmseg = shm_find_segment_by_shmid(uap->shmid);
 454         if (shmseg == NULL) {
 455                 error = EINVAL;
 456                 goto done;
 457         }
 458
 459         switch (uap->cmd) {
 460         case IPC_STAT:
 461                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 462                 if (error == 0)
 463                         error = copyout(shmseg, uap->buf, sizeof(inbuf));
 464                 break;
 465         case IPC_SET:
 466                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 467                 if (error == 0)
 468                         error = copyin(uap->buf, &inbuf, sizeof(inbuf));
 469                 if (error == 0) {
 470                         shmseg->shm_perm.uid = inbuf.shm_perm.uid;
 471                         shmseg->shm_perm.gid = inbuf.shm_perm.gid;
 472                         shmseg->shm_perm.mode =
 473                             (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 474                             (inbuf.shm_perm.mode & ACCESSPERMS);
 475                         shmseg->shm_ctime = time_second;
 476                 }
 477                 break;
 478         case IPC_RMID:
 479                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 480                 if (error == 0) {
 481                         shmseg->shm_perm.key = IPC_PRIVATE;
 482                         shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 483                         if (shmseg->shm_nattch <= 0) {
 484                                 shm_deallocate_segment(shmseg);
 485                                 shm_last_free = IPCID_TO_IX(uap->shmid);
 486                         }
 487                 }
 488                 break;
 489 #if 0
 490         case SHM_LOCK:
 491         case SHM_UNLOCK:
 492 #endif
 493         default:
 494                 error = EINVAL;
 495                 break;
 496         }
 497 done:
 498         rel_mplock();
 499         return error;
 500 }
 501
 502 static int
 503 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
 504 {
 505         struct shmid_ds *shmseg;
 506         int error;
 507
 508         shmseg = &shmsegs[segnum];
 509         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 510                 /*
 511                  * This segment is in the process of being allocated.  Wait
 512                  * until it's done, and look the key up again (in case the
 513                  * allocation failed or it was freed).
 514                  */
 515                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 516                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 517                 if (error)
 518                         return error;
 519                 return EAGAIN;
 520         }
 521         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 522                 return EEXIST;
 523         error = ipcperm(p, &shmseg->shm_perm, mode);
 524         if (error)
 525                 return error;
 526         if (uap->size && uap->size > shmseg->shm_segsz)
 527                 return EINVAL;
 528         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 529         return 0;
 530 }
 531
 532 static int
 533 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
 534 {
 535         int i, segnum, shmid;
 536         size_t size;
 537         struct ucred *cred = p->p_ucred;
 538         struct shmid_ds *shmseg;
 539         struct shm_handle *shm_handle;
 540
 541         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 542                 return EINVAL;
 543         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 544                 return ENOSPC;
 545         size = round_page(uap->size);
 546         if (shm_committed + btoc(size) > shminfo.shmall)
 547                 return ENOMEM;
 548         if (shm_last_free < 0) {
 549                 shmrealloc();   /* maybe expand the shmsegs[] array */
 550                 for (i = 0; i < shmalloced; i++)
 551                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 552                                 break;
 553                 if (i == shmalloced)
 554                         return ENOSPC;
 555                 segnum = i;
 556         } else  {
 557                 segnum = shm_last_free;
 558                 shm_last_free = -1;
 559         }
 560         shmseg = &shmsegs[segnum];
 561         /*
 562          * In case we sleep in malloc(), mark the segment present but deleted
 563          * so that noone else tries to create the same key.
 564          */
 565         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 566         shmseg->shm_perm.key = uap->key;
 567         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 568         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 569         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 570
 571         /*
 572          * We make sure that we have allocated a pager before we need
 573          * to.
 574          */
 575         if (shm_use_phys) {
 576                 shm_handle->shm_object =
 577                    phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 578         } else {
 579                 shm_handle->shm_object =
 580                    swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 581         }
 582         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 583         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 584
 585         shmseg->shm_internal = shm_handle;
 586         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 587         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 588         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 589             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 590         shmseg->shm_segsz = uap->size;
 591         shmseg->shm_cpid = p->p_pid;
 592         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 593         shmseg->shm_atime = shmseg->shm_dtime = 0;
 594         shmseg->shm_ctime = time_second;
 595         shm_committed += btoc(size);
 596         shm_nused++;
 597         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 598                 /*
 599                  * Somebody else wanted this key while we were asleep.  Wake
 600                  * them up now.
 601                  */
 602                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 603                 wakeup((caddr_t)shmseg);
 604         }
 605         uap->sysmsg_result = shmid;
 606         return 0;
 607 }
 608
 609 /*
 610  * MPALMOSTSAFE
 611  */
 612 int
 613 sys_shmget(struct shmget_args *uap)
 614 {
 615         struct thread *td = curthread;
 616         struct proc *p = td->td_proc;
 617         int segnum, mode, error;
 618
 619         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 620                 return (ENOSYS);
 621
 622         mode = uap->shmflg & ACCESSPERMS;
 623         get_mplock();
 624
 625         if (uap->key != IPC_PRIVATE) {
 626         again:
 627                 segnum = shm_find_segment_by_key(uap->key);
 628                 if (segnum >= 0) {
 629                         error = shmget_existing(p, uap, mode, segnum);
 630                         if (error == EAGAIN)
 631                                 goto again;
 632                         goto done;
 633                 }
 634                 if ((uap->shmflg & IPC_CREAT) == 0) {
 635                         error = ENOENT;
 636                         goto done;
 637                 }
 638         }
 639         error = shmget_allocate_segment(p, uap, mode);
 640 done:
 641         rel_mplock();
 642         return (error);
 643 }
 644
 645 /*
 646  * shmsys_args(int which, int a2, ...) (VARARGS)
 647  *
 648  * MPALMOSTSAFE
 649  */
 650 int
 651 sys_shmsys(struct shmsys_args *uap)
 652 {
 653         struct thread *td = curthread;
 654         unsigned int which = (unsigned int)uap->which;
 655         int error;
 656
 657         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 658                 return (ENOSYS);
 659
 660         if (which >= NELEM(shmcalls))
 661                 return EINVAL;
 662         get_mplock();
 663         bcopy(&uap->a2, &uap->which,
 664                 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
 665         error = ((*shmcalls[which])(uap));
 666         rel_mplock();
 667
 668         return(error);
 669 }
 670
 671 void
 672 shmfork(struct proc *p1, struct proc *p2)
 673 {
 674         struct shmmap_state *shmmap_s;
 675         size_t size;
 676         int i;
 677
 678         size = shminfo.shmseg * sizeof(struct shmmap_state);
 679         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 680         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 681         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 682         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
 683                 if (shmmap_s->shmid != -1)
 684                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 685 }
 686
 687 void
 688 shmexit(struct vmspace *vm)
 689 {
 690         struct shmmap_state *base, *shm;
 691         int i;
 692
 693         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 694                 vm->vm_shm = NULL;
 695                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 696                         if (shm->shmid != -1)
 697                                 shm_delete_mapping(vm, shm);
 698                 }
 699                 kfree(base, M_SHM);
 700         }
 701 }
 702
 703 static void
 704 shmrealloc(void)
 705 {
 706         int i;
 707         struct shmid_ds *newsegs;
 708
 709         if (shmalloced >= shminfo.shmmni)
 710                 return;
 711
 712         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 713         for (i = 0; i < shmalloced; i++)
 714                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 715         for (; i < shminfo.shmmni; i++) {
 716                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 717                 shmsegs[i].shm_perm.seq = 0;
 718         }
 719         kfree(shmsegs, M_SHM);
 720         shmsegs = newsegs;
 721         shmalloced = shminfo.shmmni;
 722 }
 723
 724 static void
 725 shminit(void *dummy)
 726 {
 727         int i;
 728
 729         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 730         shmalloced = shminfo.shmmni;
 731         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 732         for (i = 0; i < shmalloced; i++) {
 733                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 734                 shmsegs[i].shm_perm.seq = 0;
 735         }
 736         shm_last_free = 0;
 737         shm_nused = 0;
 738         shm_committed = 0;
 739 }
 740 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);