sys/kern/sysv_shm.c

   1 /*
   2  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions
   6  * are met:
   7  * 1. Redistributions of source code must retain the above copyright
   8  *    notice, this list of conditions and the following disclaimer.
   9  * 2. Redistributions in binary form must reproduce the above copyright
  10  *    notice, this list of conditions and the following disclaimer in the
  11  *    documentation and/or other materials provided with the distribution.
  12  * 3. All advertising materials mentioning features or use of this software
  13  *    must display the following acknowledgement:
  14  *      This product includes software developed by Adam Glass and Charles
  15  *      Hannum.
  16  * 4. The names of the authors may not be used to endorse or promote products
  17  *    derived from this software without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  22  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29  */
  30
  31 #include "opt_compat.h"
  32 #include "opt_sysvipc.h"
  33
  34 #include <sys/param.h>
  35 #include <sys/systm.h>
  36 #include <sys/sysproto.h>
  37 #include <sys/kernel.h>
  38 #include <sys/sysctl.h>
  39 #include <sys/shm.h>
  40 #include <sys/proc.h>
  41 #include <sys/malloc.h>
  42 #include <sys/mman.h>
  43 #include <sys/stat.h>
  44 #include <sys/sysent.h>
  45 #include <sys/jail.h>
  46
  47 #include <sys/mplock2.h>
  48
  49 #include <vm/vm.h>
  50 #include <vm/vm_param.h>
  51 #include <sys/lock.h>
  52 #include <vm/pmap.h>
  53 #include <vm/vm_object.h>
  54 #include <vm/vm_map.h>
  55 #include <vm/vm_page.h>
  56 #include <vm/vm_pager.h>
  57
  58 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
  59
  60 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
  61 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
  62
  63 #define SHMSEG_FREE             0x0200
  64 #define SHMSEG_REMOVED          0x0400
  65 #define SHMSEG_ALLOCATED        0x0800
  66 #define SHMSEG_WANTED           0x1000
  67
  68 static int shm_last_free, shm_committed, shmalloced;
  69 int shm_nused;
  70 static struct shmid_ds  *shmsegs;
  71
  72 struct shm_handle {
  73         /* vm_offset_t kva; */
  74         vm_object_t shm_object;
  75 };
  76
  77 struct shmmap_state {
  78         vm_offset_t va;
  79         int shmid;
  80 };
  81
  82 static void shm_deallocate_segment (struct shmid_ds *);
  83 static int shm_find_segment_by_key (key_t);
  84 static struct shmid_ds *shm_find_segment_by_shmid (int);
  85 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
  86 static void shmrealloc (void);
  87 static void shminit (void *);
  88
  89 /*
  90  * Tuneable values
  91  */
  92 #ifndef SHMMIN
  93 #define SHMMIN  1
  94 #endif
  95 #ifndef SHMMNI
  96 #define SHMMNI  512
  97 #endif
  98 #ifndef SHMSEG
  99 #define SHMSEG  1024
 100 #endif
 101
 102 struct  shminfo shminfo = {
 103         0,
 104         SHMMIN,
 105         SHMMNI,
 106         SHMSEG,
 107         0
 108 };
 109
 110 static int shm_allow_removed;
 111 static int shm_use_phys = 1;
 112
 113 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
 114 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
 115 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
 116 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
 117 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
 118
 119 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 120     "Max shared memory segment size");
 121 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 122     "Min shared memory segment size");
 123 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
 124     "Max number of shared memory identifiers");
 125 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
 126     "Max shared memory segments per process");
 127 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 128     "Max pages of shared memory");
 129 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
 130     "Use phys pager allocation instead of swap pager allocation");
 131 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
 132     &shm_allow_removed, 0,
 133     "Enable/Disable attachment to attached segments marked for removal");
 134
 135 static int
 136 shm_find_segment_by_key(key_t key)
 137 {
 138         int i;
 139
 140         for (i = 0; i < shmalloced; i++) {
 141                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 142                     shmsegs[i].shm_perm.key == key)
 143                         return i;
 144         }
 145         return -1;
 146 }
 147
 148 static struct shmid_ds *
 149 shm_find_segment_by_shmid(int shmid)
 150 {
 151         int segnum;
 152         struct shmid_ds *shmseg;
 153
 154         segnum = IPCID_TO_IX(shmid);
 155         if (segnum < 0 || segnum >= shmalloced)
 156                 return NULL;
 157         shmseg = &shmsegs[segnum];
 158         if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 159             (!shm_allow_removed &&
 160             (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 161             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
 162                 return NULL;
 163         }
 164         return shmseg;
 165 }
 166
 167 static void
 168 shm_deallocate_segment(struct shmid_ds *shmseg)
 169 {
 170         struct shm_handle *shm_handle;
 171         size_t size;
 172
 173         shm_handle = shmseg->shm_internal;
 174         vm_object_deallocate(shm_handle->shm_object);
 175         kfree((caddr_t)shm_handle, M_SHM);
 176         shmseg->shm_internal = NULL;
 177         size = round_page(shmseg->shm_segsz);
 178         shm_committed -= btoc(size);
 179         shm_nused--;
 180         shmseg->shm_perm.mode = SHMSEG_FREE;
 181 }
 182
 183 static int
 184 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 185 {
 186         struct shmid_ds *shmseg;
 187         int segnum, result;
 188         size_t size;
 189
 190         segnum = IPCID_TO_IX(shmmap_s->shmid);
 191         shmseg = &shmsegs[segnum];
 192         size = round_page(shmseg->shm_segsz);
 193         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 194         if (result != KERN_SUCCESS)
 195                 return EINVAL;
 196         shmmap_s->shmid = -1;
 197         shmseg->shm_dtime = time_second;
 198         if ((--shmseg->shm_nattch <= 0) &&
 199             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 200                 shm_deallocate_segment(shmseg);
 201                 shm_last_free = segnum;
 202         }
 203         return 0;
 204 }
 205
 206 /*
 207  * MPALMOSTSAFE
 208  */
 209 int
 210 sys_shmdt(struct shmdt_args *uap)
 211 {
 212         struct thread *td = curthread;
 213         struct proc *p = td->td_proc;
 214         struct shmmap_state *shmmap_s;
 215         long i;
 216         int error;
 217
 218         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 219                 return (ENOSYS);
 220
 221         get_mplock();
 222         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 223         if (shmmap_s == NULL) {
 224                 error = EINVAL;
 225                 goto done;
 226         }
 227         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 228                 if (shmmap_s->shmid != -1 &&
 229                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 230                         break;
 231         }
 232         if (i == shminfo.shmseg)
 233                 error = EINVAL;
 234         else
 235                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 236 done:
 237         rel_mplock();
 238         return (error);
 239 }
 240
 241 /*
 242  * MPALMOSTSAFE
 243  */
 244 int
 245 sys_shmat(struct shmat_args *uap)
 246 {
 247         struct thread *td = curthread;
 248         struct proc *p = td->td_proc;
 249         int error, flags;
 250         long i;
 251         struct shmid_ds *shmseg;
 252         struct shmmap_state *shmmap_s = NULL;
 253         struct shm_handle *shm_handle;
 254         vm_offset_t attach_va;
 255         vm_prot_t prot;
 256         vm_size_t size;
 257         vm_size_t align;
 258         int rv;
 259
 260         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 261                 return (ENOSYS);
 262
 263         get_mplock();
 264 again:
 265         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 266         if (shmmap_s == NULL) {
 267                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 268                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 269                 for (i = 0; i < shminfo.shmseg; i++)
 270                         shmmap_s[i].shmid = -1;
 271                 if (p->p_vmspace->vm_shm != NULL) {
 272                         kfree(shmmap_s, M_SHM);
 273                         goto again;
 274                 }
 275                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 276         }
 277         shmseg = shm_find_segment_by_shmid(uap->shmid);
 278         if (shmseg == NULL) {
 279                 error = EINVAL;
 280                 goto done;
 281         }
 282         error = ipcperm(p, &shmseg->shm_perm,
 283                         (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 284         if (error)
 285                 goto done;
 286         for (i = 0; i < shminfo.shmseg; i++) {
 287                 if (shmmap_s->shmid == -1)
 288                         break;
 289                 shmmap_s++;
 290         }
 291         if (i >= shminfo.shmseg) {
 292                 error = EMFILE;
 293                 goto done;
 294         }
 295         size = round_page(shmseg->shm_segsz);
 296 #ifdef VM_PROT_READ_IS_EXEC
 297         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 298 #else
 299         prot = VM_PROT_READ;
 300 #endif
 301         if ((uap->shmflg & SHM_RDONLY) == 0)
 302                 prot |= VM_PROT_WRITE;
 303         flags = MAP_ANON | MAP_SHARED;
 304         if (uap->shmaddr) {
 305                 flags |= MAP_FIXED;
 306                 if (uap->shmflg & SHM_RND) {
 307                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
 308                 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
 309                         attach_va = (vm_offset_t)uap->shmaddr;
 310                 } else {
 311                         error = EINVAL;
 312                         goto done;
 313                 }
 314         } else {
 315                 /*
 316                  * This is just a hint to vm_map_find() about where to put it.
 317                  */
 318                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr +
 319                                        maxtsiz + maxdsiz);
 320         }
 321
 322         /*
 323          * Handle alignment.  For large memory maps it is possible
 324          * that the MMU can optimize the page table so align anything
 325          * that is a multiple of SEG_SIZE to SEG_SIZE.
 326          */
 327         if ((flags & MAP_FIXED) == 0 && (size & SEG_MASK) == 0)
 328                 align = SEG_SIZE;
 329         else
 330                 align = PAGE_SIZE;
 331
 332         shm_handle = shmseg->shm_internal;
 333         vm_object_hold(shm_handle->shm_object);
 334         vm_object_chain_wait(shm_handle->shm_object, 0);
 335         vm_object_reference_locked(shm_handle->shm_object);
 336         rv = vm_map_find(&p->p_vmspace->vm_map,
 337                          shm_handle->shm_object, NULL,
 338                          0, &attach_va, size,
 339                          align,
 340                          ((flags & MAP_FIXED) ? 0 : 1),
 341                          VM_MAPTYPE_NORMAL,
 342                          prot, prot, 0);
 343         vm_object_drop(shm_handle->shm_object);
 344         if (rv != KERN_SUCCESS) {
 345                 vm_object_deallocate(shm_handle->shm_object);
 346                 error = ENOMEM;
 347                 goto done;
 348         }
 349         vm_map_inherit(&p->p_vmspace->vm_map,
 350                        attach_va, attach_va + size, VM_INHERIT_SHARE);
 351
 352         KKASSERT(shmmap_s->shmid == -1);
 353         shmmap_s->va = attach_va;
 354         shmmap_s->shmid = uap->shmid;
 355         shmseg->shm_lpid = p->p_pid;
 356         shmseg->shm_atime = time_second;
 357         shmseg->shm_nattch++;
 358         uap->sysmsg_resultp = (void *)attach_va;
 359         error = 0;
 360 done:
 361         rel_mplock();
 362         return error;
 363 }
 364
 365 /*
 366  * MPALMOSTSAFE
 367  */
 368 int
 369 sys_shmctl(struct shmctl_args *uap)
 370 {
 371         struct thread *td = curthread;
 372         struct proc *p = td->td_proc;
 373         int error;
 374         struct shmid_ds inbuf;
 375         struct shmid_ds *shmseg;
 376
 377         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 378                 return (ENOSYS);
 379
 380         get_mplock();
 381         shmseg = shm_find_segment_by_shmid(uap->shmid);
 382         if (shmseg == NULL) {
 383                 error = EINVAL;
 384                 goto done;
 385         }
 386
 387         switch (uap->cmd) {
 388         case IPC_STAT:
 389                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 390                 if (error == 0)
 391                         error = copyout(shmseg, uap->buf, sizeof(inbuf));
 392                 break;
 393         case IPC_SET:
 394                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 395                 if (error == 0)
 396                         error = copyin(uap->buf, &inbuf, sizeof(inbuf));
 397                 if (error == 0) {
 398                         shmseg->shm_perm.uid = inbuf.shm_perm.uid;
 399                         shmseg->shm_perm.gid = inbuf.shm_perm.gid;
 400                         shmseg->shm_perm.mode =
 401                             (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 402                             (inbuf.shm_perm.mode & ACCESSPERMS);
 403                         shmseg->shm_ctime = time_second;
 404                 }
 405                 break;
 406         case IPC_RMID:
 407                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 408                 if (error == 0) {
 409                         shmseg->shm_perm.key = IPC_PRIVATE;
 410                         shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 411                         if (shmseg->shm_nattch <= 0) {
 412                                 shm_deallocate_segment(shmseg);
 413                                 shm_last_free = IPCID_TO_IX(uap->shmid);
 414                         }
 415                 }
 416                 break;
 417 #if 0
 418         case SHM_LOCK:
 419         case SHM_UNLOCK:
 420 #endif
 421         default:
 422                 error = EINVAL;
 423                 break;
 424         }
 425 done:
 426         rel_mplock();
 427         return error;
 428 }
 429
 430 static int
 431 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
 432 {
 433         struct shmid_ds *shmseg;
 434         int error;
 435
 436         shmseg = &shmsegs[segnum];
 437         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 438                 /*
 439                  * This segment is in the process of being allocated.  Wait
 440                  * until it's done, and look the key up again (in case the
 441                  * allocation failed or it was freed).
 442                  */
 443                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 444                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 445                 if (error)
 446                         return error;
 447                 return EAGAIN;
 448         }
 449         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 450                 return EEXIST;
 451         error = ipcperm(p, &shmseg->shm_perm, mode);
 452         if (error)
 453                 return error;
 454         if (uap->size && uap->size > shmseg->shm_segsz)
 455                 return EINVAL;
 456         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 457         return 0;
 458 }
 459
 460 static int
 461 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
 462 {
 463         int i, segnum, shmid;
 464         size_t size;
 465         struct ucred *cred = p->p_ucred;
 466         struct shmid_ds *shmseg;
 467         struct shm_handle *shm_handle;
 468
 469         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 470                 return EINVAL;
 471         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 472                 return ENOSPC;
 473         size = round_page(uap->size);
 474         if (shm_committed + btoc(size) > shminfo.shmall)
 475                 return ENOMEM;
 476         if (shm_last_free < 0) {
 477                 shmrealloc();   /* maybe expand the shmsegs[] array */
 478                 for (i = 0; i < shmalloced; i++) {
 479                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 480                                 break;
 481                 }
 482                 if (i == shmalloced)
 483                         return ENOSPC;
 484                 segnum = i;
 485         } else  {
 486                 segnum = shm_last_free;
 487                 shm_last_free = -1;
 488         }
 489         shmseg = &shmsegs[segnum];
 490         /*
 491          * In case we sleep in malloc(), mark the segment present but deleted
 492          * so that noone else tries to create the same key.
 493          */
 494         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 495         shmseg->shm_perm.key = uap->key;
 496         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 497         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 498         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 499
 500         /*
 501          * We make sure that we have allocated a pager before we need
 502          * to.
 503          */
 504         if (shm_use_phys) {
 505                 shm_handle->shm_object =
 506                    phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 507         } else {
 508                 shm_handle->shm_object =
 509                    swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 510         }
 511         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 512         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 513
 514         shmseg->shm_internal = shm_handle;
 515         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 516         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 517         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 518             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 519         shmseg->shm_segsz = uap->size;
 520         shmseg->shm_cpid = p->p_pid;
 521         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 522         shmseg->shm_atime = shmseg->shm_dtime = 0;
 523         shmseg->shm_ctime = time_second;
 524         shm_committed += btoc(size);
 525         shm_nused++;
 526
 527         /*
 528          * If a physical mapping is desired and we have a ton of free pages
 529          * we pre-allocate the pages here in order to avoid on-the-fly
 530          * allocation later.  This has a big effect on database warm-up
 531          * times since DFly supports concurrent page faults coming from the
 532          * same VM object for pages which already exist.
 533          *
 534          * This can hang the kernel for a while so only do it if shm_use_phys
 535          * is set to 2 or higher.
 536          */
 537         if (shm_use_phys > 1) {
 538                 vm_pindex_t pi, pmax;
 539                 vm_page_t m;
 540
 541                 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
 542                 vm_object_hold(shm_handle->shm_object);
 543                 if (pmax > vmstats.v_free_count)
 544                         pmax = vmstats.v_free_count;
 545                 for (pi = 0; pi < pmax; ++pi) {
 546                         m = vm_page_grab(shm_handle->shm_object, pi,
 547                                          VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
 548                                          VM_ALLOC_ZERO);
 549                         if (m == NULL)
 550                                 break;
 551                         vm_pager_get_page(shm_handle->shm_object, &m, 1);
 552                         vm_page_activate(m);
 553                         vm_page_wakeup(m);
 554                         lwkt_yield();
 555                 }
 556                 vm_object_drop(shm_handle->shm_object);
 557         }
 558
 559         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 560                 /*
 561                  * Somebody else wanted this key while we were asleep.  Wake
 562                  * them up now.
 563                  */
 564                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 565                 wakeup((caddr_t)shmseg);
 566         }
 567         uap->sysmsg_result = shmid;
 568         return 0;
 569 }
 570
 571 /*
 572  * MPALMOSTSAFE
 573  */
 574 int
 575 sys_shmget(struct shmget_args *uap)
 576 {
 577         struct thread *td = curthread;
 578         struct proc *p = td->td_proc;
 579         int segnum, mode, error;
 580
 581         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 582                 return (ENOSYS);
 583
 584         mode = uap->shmflg & ACCESSPERMS;
 585         get_mplock();
 586
 587         if (uap->key != IPC_PRIVATE) {
 588         again:
 589                 segnum = shm_find_segment_by_key(uap->key);
 590                 if (segnum >= 0) {
 591                         error = shmget_existing(p, uap, mode, segnum);
 592                         if (error == EAGAIN)
 593                                 goto again;
 594                         goto done;
 595                 }
 596                 if ((uap->shmflg & IPC_CREAT) == 0) {
 597                         error = ENOENT;
 598                         goto done;
 599                 }
 600         }
 601         error = shmget_allocate_segment(p, uap, mode);
 602 done:
 603         rel_mplock();
 604         return (error);
 605 }
 606
 607 void
 608 shmfork(struct proc *p1, struct proc *p2)
 609 {
 610         struct shmmap_state *shmmap_s;
 611         size_t size;
 612         int i;
 613
 614         get_mplock();
 615         size = shminfo.shmseg * sizeof(struct shmmap_state);
 616         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 617         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 618         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 619         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 620                 if (shmmap_s->shmid != -1)
 621                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 622         }
 623         rel_mplock();
 624 }
 625
 626 void
 627 shmexit(struct vmspace *vm)
 628 {
 629         struct shmmap_state *base, *shm;
 630         int i;
 631
 632         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 633                 vm->vm_shm = NULL;
 634                 get_mplock();
 635                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 636                         if (shm->shmid != -1)
 637                                 shm_delete_mapping(vm, shm);
 638                 }
 639                 kfree(base, M_SHM);
 640                 rel_mplock();
 641         }
 642 }
 643
 644 static void
 645 shmrealloc(void)
 646 {
 647         int i;
 648         struct shmid_ds *newsegs;
 649
 650         if (shmalloced >= shminfo.shmmni)
 651                 return;
 652
 653         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 654         for (i = 0; i < shmalloced; i++)
 655                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 656         for (; i < shminfo.shmmni; i++) {
 657                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 658                 shmsegs[i].shm_perm.seq = 0;
 659         }
 660         kfree(shmsegs, M_SHM);
 661         shmsegs = newsegs;
 662         shmalloced = shminfo.shmmni;
 663 }
 664
 665 static void
 666 shminit(void *dummy)
 667 {
 668         int i;
 669
 670         /*
 671          * If not overridden by a tunable set the maximum shm to
 672          * 2/3 of main memory.
 673          */
 674         if (shminfo.shmall == 0)
 675                 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
 676
 677         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 678         shmalloced = shminfo.shmmni;
 679         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 680         for (i = 0; i < shmalloced; i++) {
 681                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 682                 shmsegs[i].shm_perm.seq = 0;
 683         }
 684         shm_last_free = 0;
 685         shm_nused = 0;
 686         shm_committed = 0;
 687 }
 688 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);