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 <vm/vm.h>
  47 #include <vm/vm_param.h>
  48 #include <sys/lock.h>
  49 #include <vm/pmap.h>
  50 #include <vm/vm_object.h>
  51 #include <vm/vm_map.h>
  52 #include <vm/vm_page.h>
  53 #include <vm/vm_pager.h>
  54
  55 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
  56
  57 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
  58 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
  59
  60 #define SHMSEG_FREE             0x0200
  61 #define SHMSEG_REMOVED          0x0400
  62 #define SHMSEG_ALLOCATED        0x0800
  63 #define SHMSEG_WANTED           0x1000
  64
  65 static int shm_last_free, shm_committed, shmalloced;
  66 int shm_nused;
  67 static struct shmid_ds  *shmsegs;
  68 static struct lwkt_token shm_token = LWKT_TOKEN_INITIALIZER(shm_token);
  69
  70 struct shm_handle {
  71         /* vm_offset_t kva; */
  72         vm_object_t shm_object;
  73 };
  74
  75 struct shmmap_state {
  76         vm_offset_t va;
  77         int shmid;
  78 };
  79
  80 static void shm_deallocate_segment (struct shmid_ds *);
  81 static int shm_find_segment_by_key (key_t);
  82 static struct shmid_ds *shm_find_segment_by_shmid (int);
  83 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
  84 static void shmrealloc (void);
  85 static void shminit (void *);
  86
  87 /*
  88  * Tuneable values
  89  */
  90 #ifndef SHMMIN
  91 #define SHMMIN  1
  92 #endif
  93 #ifndef SHMMNI
  94 #define SHMMNI  512
  95 #endif
  96 #ifndef SHMSEG
  97 #define SHMSEG  1024
  98 #endif
  99
 100 struct  shminfo shminfo = {
 101         0,
 102         SHMMIN,
 103         SHMMNI,
 104         SHMSEG,
 105         0
 106 };
 107
 108 static int shm_allow_removed;
 109 static int shm_use_phys = 1;
 110
 111 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
 112 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
 113 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
 114 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
 115 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
 116
 117 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 118     "Max shared memory segment size");
 119 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 120     "Min shared memory segment size");
 121 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
 122     "Max number of shared memory identifiers");
 123 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
 124     "Max shared memory segments per process");
 125 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 126     "Max pages of shared memory");
 127 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
 128     "Use phys pager allocation instead of swap pager allocation");
 129 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
 130     &shm_allow_removed, 0,
 131     "Enable/Disable attachment to attached segments marked for removal");
 132
 133 static int
 134 shm_find_segment_by_key(key_t key)
 135 {
 136         int i;
 137
 138         for (i = 0; i < shmalloced; i++) {
 139                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 140                     shmsegs[i].shm_perm.key == key)
 141                         return i;
 142         }
 143         return -1;
 144 }
 145
 146 static struct shmid_ds *
 147 shm_find_segment_by_shmid(int shmid)
 148 {
 149         int segnum;
 150         struct shmid_ds *shmseg;
 151
 152         segnum = IPCID_TO_IX(shmid);
 153         if (segnum < 0 || segnum >= shmalloced)
 154                 return NULL;
 155         shmseg = &shmsegs[segnum];
 156         if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 157             (!shm_allow_removed &&
 158             (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 159             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
 160                 return NULL;
 161         }
 162         return shmseg;
 163 }
 164
 165 static void
 166 shm_deallocate_segment(struct shmid_ds *shmseg)
 167 {
 168         struct shm_handle *shm_handle;
 169         size_t size;
 170
 171         shm_handle = shmseg->shm_internal;
 172         vm_object_deallocate(shm_handle->shm_object);
 173         kfree((caddr_t)shm_handle, M_SHM);
 174         shmseg->shm_internal = NULL;
 175         size = round_page(shmseg->shm_segsz);
 176         shm_committed -= btoc(size);
 177         shm_nused--;
 178         shmseg->shm_perm.mode = SHMSEG_FREE;
 179 }
 180
 181 static int
 182 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 183 {
 184         struct shmid_ds *shmseg;
 185         int segnum, result;
 186         size_t size;
 187
 188         segnum = IPCID_TO_IX(shmmap_s->shmid);
 189         shmseg = &shmsegs[segnum];
 190         size = round_page(shmseg->shm_segsz);
 191         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 192         if (result != KERN_SUCCESS)
 193                 return EINVAL;
 194         shmmap_s->shmid = -1;
 195         shmseg->shm_dtime = time_second;
 196         if ((--shmseg->shm_nattch <= 0) &&
 197             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 198                 shm_deallocate_segment(shmseg);
 199                 shm_last_free = segnum;
 200         }
 201         return 0;
 202 }
 203
 204 /*
 205  * MPALMOSTSAFE
 206  */
 207 int
 208 sys_shmdt(struct shmdt_args *uap)
 209 {
 210         struct thread *td = curthread;
 211         struct proc *p = td->td_proc;
 212         struct shmmap_state *shmmap_s;
 213         long i;
 214         int error;
 215
 216         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 217                 return (ENOSYS);
 218
 219         lwkt_gettoken(&shm_token);
 220         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 221         if (shmmap_s == NULL) {
 222                 error = EINVAL;
 223                 goto done;
 224         }
 225         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 226                 if (shmmap_s->shmid != -1 &&
 227                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 228                         break;
 229         }
 230         if (i == shminfo.shmseg)
 231                 error = EINVAL;
 232         else
 233                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 234 done:
 235         lwkt_reltoken(&shm_token);
 236
 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         lwkt_gettoken(&shm_token);
 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         lwkt_reltoken(&shm_token);
 361
 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         lwkt_gettoken(&shm_token);
 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         lwkt_reltoken(&shm_token);
 427
 428         return error;
 429 }
 430
 431 static int
 432 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
 433 {
 434         struct shmid_ds *shmseg;
 435         int error;
 436
 437         shmseg = &shmsegs[segnum];
 438         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 439                 /*
 440                  * This segment is in the process of being allocated.  Wait
 441                  * until it's done, and look the key up again (in case the
 442                  * allocation failed or it was freed).
 443                  */
 444                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 445                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 446                 if (error)
 447                         return error;
 448                 return EAGAIN;
 449         }
 450         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 451                 return EEXIST;
 452         error = ipcperm(p, &shmseg->shm_perm, mode);
 453         if (error)
 454                 return error;
 455         if (uap->size && uap->size > shmseg->shm_segsz)
 456                 return EINVAL;
 457         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 458         return 0;
 459 }
 460
 461 static int
 462 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
 463 {
 464         int i, segnum, shmid;
 465         size_t size;
 466         struct ucred *cred = p->p_ucred;
 467         struct shmid_ds *shmseg;
 468         struct shm_handle *shm_handle;
 469
 470         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 471                 return EINVAL;
 472         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 473                 return ENOSPC;
 474         size = round_page(uap->size);
 475         if (shm_committed + btoc(size) > shminfo.shmall)
 476                 return ENOMEM;
 477         if (shm_last_free < 0) {
 478                 shmrealloc();   /* maybe expand the shmsegs[] array */
 479                 for (i = 0; i < shmalloced; i++) {
 480                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 481                                 break;
 482                 }
 483                 if (i == shmalloced)
 484                         return ENOSPC;
 485                 segnum = i;
 486         } else  {
 487                 segnum = shm_last_free;
 488                 shm_last_free = -1;
 489         }
 490         shmseg = &shmsegs[segnum];
 491         /*
 492          * In case we sleep in malloc(), mark the segment present but deleted
 493          * so that noone else tries to create the same key.
 494          */
 495         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 496         shmseg->shm_perm.key = uap->key;
 497         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 498         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 499         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 500
 501         /*
 502          * We make sure that we have allocated a pager before we need
 503          * to.
 504          */
 505         if (shm_use_phys) {
 506                 shm_handle->shm_object =
 507                    phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 508         } else {
 509                 shm_handle->shm_object =
 510                    swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 511         }
 512         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 513         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 514
 515         shmseg->shm_internal = shm_handle;
 516         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 517         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 518         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 519             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 520         shmseg->shm_segsz = uap->size;
 521         shmseg->shm_cpid = p->p_pid;
 522         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 523         shmseg->shm_atime = shmseg->shm_dtime = 0;
 524         shmseg->shm_ctime = time_second;
 525         shm_committed += btoc(size);
 526         shm_nused++;
 527
 528         /*
 529          * If a physical mapping is desired and we have a ton of free pages
 530          * we pre-allocate the pages here in order to avoid on-the-fly
 531          * allocation later.  This has a big effect on database warm-up
 532          * times since DFly supports concurrent page faults coming from the
 533          * same VM object for pages which already exist.
 534          *
 535          * This can hang the kernel for a while so only do it if shm_use_phys
 536          * is set to 2 or higher.
 537          */
 538         if (shm_use_phys > 1) {
 539                 vm_pindex_t pi, pmax;
 540                 vm_page_t m;
 541
 542                 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
 543                 vm_object_hold(shm_handle->shm_object);
 544                 if (pmax > vmstats.v_free_count)
 545                         pmax = vmstats.v_free_count;
 546                 for (pi = 0; pi < pmax; ++pi) {
 547                         m = vm_page_grab(shm_handle->shm_object, pi,
 548                                          VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
 549                                          VM_ALLOC_ZERO);
 550                         if (m == NULL)
 551                                 break;
 552                         vm_pager_get_page(shm_handle->shm_object, &m, 1);
 553                         vm_page_activate(m);
 554                         vm_page_wakeup(m);
 555                         lwkt_yield();
 556                 }
 557                 vm_object_drop(shm_handle->shm_object);
 558         }
 559
 560         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 561                 /*
 562                  * Somebody else wanted this key while we were asleep.  Wake
 563                  * them up now.
 564                  */
 565                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 566                 wakeup((caddr_t)shmseg);
 567         }
 568         uap->sysmsg_result = shmid;
 569         return 0;
 570 }
 571
 572 /*
 573  * MPALMOSTSAFE
 574  */
 575 int
 576 sys_shmget(struct shmget_args *uap)
 577 {
 578         struct thread *td = curthread;
 579         struct proc *p = td->td_proc;
 580         int segnum, mode, error;
 581
 582         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 583                 return (ENOSYS);
 584
 585         mode = uap->shmflg & ACCESSPERMS;
 586
 587         lwkt_gettoken(&shm_token);
 588
 589         if (uap->key != IPC_PRIVATE) {
 590         again:
 591                 segnum = shm_find_segment_by_key(uap->key);
 592                 if (segnum >= 0) {
 593                         error = shmget_existing(p, uap, mode, segnum);
 594                         if (error == EAGAIN)
 595                                 goto again;
 596                         goto done;
 597                 }
 598                 if ((uap->shmflg & IPC_CREAT) == 0) {
 599                         error = ENOENT;
 600                         goto done;
 601                 }
 602         }
 603         error = shmget_allocate_segment(p, uap, mode);
 604 done:
 605         lwkt_reltoken(&shm_token);
 606
 607         return (error);
 608 }
 609
 610 void
 611 shmfork(struct proc *p1, struct proc *p2)
 612 {
 613         struct shmmap_state *shmmap_s;
 614         size_t size;
 615         int i;
 616
 617         lwkt_gettoken(&shm_token);
 618         size = shminfo.shmseg * sizeof(struct shmmap_state);
 619         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 620         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 621         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 622         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 623                 if (shmmap_s->shmid != -1)
 624                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 625         }
 626         lwkt_reltoken(&shm_token);
 627 }
 628
 629 void
 630 shmexit(struct vmspace *vm)
 631 {
 632         struct shmmap_state *base, *shm;
 633         int i;
 634
 635         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 636                 vm->vm_shm = NULL;
 637                 lwkt_gettoken(&shm_token);
 638                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 639                         if (shm->shmid != -1)
 640                                 shm_delete_mapping(vm, shm);
 641                 }
 642                 kfree(base, M_SHM);
 643                 lwkt_reltoken(&shm_token);
 644         }
 645 }
 646
 647 static void
 648 shmrealloc(void)
 649 {
 650         int i;
 651         struct shmid_ds *newsegs;
 652
 653         if (shmalloced >= shminfo.shmmni)
 654                 return;
 655
 656         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 657         for (i = 0; i < shmalloced; i++)
 658                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 659         for (; i < shminfo.shmmni; i++) {
 660                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 661                 shmsegs[i].shm_perm.seq = 0;
 662         }
 663         kfree(shmsegs, M_SHM);
 664         shmsegs = newsegs;
 665         shmalloced = shminfo.shmmni;
 666 }
 667
 668 static void
 669 shminit(void *dummy)
 670 {
 671         int i;
 672
 673         /*
 674          * If not overridden by a tunable set the maximum shm to
 675          * 2/3 of main memory.
 676          */
 677         if (shminfo.shmall == 0)
 678                 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
 679
 680         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 681         shmalloced = shminfo.shmmni;
 682         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 683         for (i = 0; i < shmalloced; i++) {
 684                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 685                 shmsegs[i].shm_perm.seq = 0;
 686         }
 687         shm_last_free = 0;
 688         shm_nused = 0;
 689         shm_committed = 0;
 690 }
 691 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);