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