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 struct oshmctl_args;
  61 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
  62
  63 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
  64 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
  65
  66 /* XXX casting to (sy_call_t *) is bogus, as usual. */
  67 static sy_call_t *shmcalls[] = {
  68         (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
  69         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
  70         (sy_call_t *)sys_shmctl
  71 };
  72
  73 #define SHMSEG_FREE             0x0200
  74 #define SHMSEG_REMOVED          0x0400
  75 #define SHMSEG_ALLOCATED        0x0800
  76 #define SHMSEG_WANTED           0x1000
  77
  78 static int shm_last_free, shm_committed, shmalloced;
  79 int shm_nused;
  80 static struct shmid_ds  *shmsegs;
  81
  82 struct shm_handle {
  83         /* vm_offset_t kva; */
  84         vm_object_t shm_object;
  85 };
  86
  87 struct shmmap_state {
  88         vm_offset_t va;
  89         int shmid;
  90 };
  91
  92 static void shm_deallocate_segment (struct shmid_ds *);
  93 static int shm_find_segment_by_key (key_t);
  94 static struct shmid_ds *shm_find_segment_by_shmid (int);
  95 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
  96 static void shmrealloc (void);
  97 static void shminit (void *);
  98
  99 /*
 100  * Tuneable values
 101  */
 102 #ifndef SHMMIN
 103 #define SHMMIN  1
 104 #endif
 105 #ifndef SHMMNI
 106 #define SHMMNI  512
 107 #endif
 108 #ifndef SHMSEG
 109 #define SHMSEG  1024
 110 #endif
 111
 112 struct  shminfo shminfo = {
 113         0,
 114         SHMMIN,
 115         SHMMNI,
 116         SHMSEG,
 117         0
 118 };
 119
 120 static int shm_use_phys;
 121
 122 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
 123 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
 124 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
 125 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
 126 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
 127
 128 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
 129     "Max shared memory segment size");
 130 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
 131     "Min shared memory segment size");
 132 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
 133     "Max number of shared memory identifiers");
 134 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
 135     "Max shared memory segments per process");
 136 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
 137     "Max pages of shared memory");
 138 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
 139     "Use phys pager allocation instead of swap pager allocation");
 140
 141 static int
 142 shm_find_segment_by_key(key_t key)
 143 {
 144         int i;
 145
 146         for (i = 0; i < shmalloced; i++) {
 147                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 148                     shmsegs[i].shm_perm.key == key)
 149                         return i;
 150         }
 151         return -1;
 152 }
 153
 154 static struct shmid_ds *
 155 shm_find_segment_by_shmid(int shmid)
 156 {
 157         int segnum;
 158         struct shmid_ds *shmseg;
 159
 160         segnum = IPCID_TO_IX(shmid);
 161         if (segnum < 0 || segnum >= shmalloced)
 162                 return NULL;
 163         shmseg = &shmsegs[segnum];
 164         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
 165             != SHMSEG_ALLOCATED ||
 166             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
 167                 return NULL;
 168         }
 169         return shmseg;
 170 }
 171
 172 static void
 173 shm_deallocate_segment(struct shmid_ds *shmseg)
 174 {
 175         struct shm_handle *shm_handle;
 176         size_t size;
 177
 178         shm_handle = shmseg->shm_internal;
 179         vm_object_deallocate(shm_handle->shm_object);
 180         kfree((caddr_t)shm_handle, M_SHM);
 181         shmseg->shm_internal = NULL;
 182         size = round_page(shmseg->shm_segsz);
 183         shm_committed -= btoc(size);
 184         shm_nused--;
 185         shmseg->shm_perm.mode = SHMSEG_FREE;
 186 }
 187
 188 static int
 189 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 190 {
 191         struct shmid_ds *shmseg;
 192         int segnum, result;
 193         size_t size;
 194
 195         segnum = IPCID_TO_IX(shmmap_s->shmid);
 196         shmseg = &shmsegs[segnum];
 197         size = round_page(shmseg->shm_segsz);
 198         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 199         if (result != KERN_SUCCESS)
 200                 return EINVAL;
 201         shmmap_s->shmid = -1;
 202         shmseg->shm_dtime = time_second;
 203         if ((--shmseg->shm_nattch <= 0) &&
 204             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 205                 shm_deallocate_segment(shmseg);
 206                 shm_last_free = segnum;
 207         }
 208         return 0;
 209 }
 210
 211 /*
 212  * MPALMOSTSAFE
 213  */
 214 int
 215 sys_shmdt(struct shmdt_args *uap)
 216 {
 217         struct thread *td = curthread;
 218         struct proc *p = td->td_proc;
 219         struct shmmap_state *shmmap_s;
 220         long i;
 221         int error;
 222
 223         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 224                 return (ENOSYS);
 225
 226         get_mplock();
 227         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 228         if (shmmap_s == NULL) {
 229                 error = EINVAL;
 230                 goto done;
 231         }
 232         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 233                 if (shmmap_s->shmid != -1 &&
 234                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
 235                         break;
 236         }
 237         if (i == shminfo.shmseg)
 238                 error = EINVAL;
 239         else
 240                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 241 done:
 242         rel_mplock();
 243         return (error);
 244 }
 245
 246 /*
 247  * MPALMOSTSAFE
 248  */
 249 int
 250 sys_shmat(struct shmat_args *uap)
 251 {
 252         struct thread *td = curthread;
 253         struct proc *p = td->td_proc;
 254         int error, flags;
 255         long i;
 256         struct shmid_ds *shmseg;
 257         struct shmmap_state *shmmap_s = NULL;
 258         struct shm_handle *shm_handle;
 259         vm_offset_t attach_va;
 260         vm_prot_t prot;
 261         vm_size_t size;
 262         int rv;
 263
 264         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 265                 return (ENOSYS);
 266
 267         get_mplock();
 268 again:
 269         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
 270         if (shmmap_s == NULL) {
 271                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 272                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 273                 for (i = 0; i < shminfo.shmseg; i++)
 274                         shmmap_s[i].shmid = -1;
 275                 if (p->p_vmspace->vm_shm != NULL) {
 276                         kfree(shmmap_s, M_SHM);
 277                         goto again;
 278                 }
 279                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 280         }
 281         shmseg = shm_find_segment_by_shmid(uap->shmid);
 282         if (shmseg == NULL) {
 283                 error = EINVAL;
 284                 goto done;
 285         }
 286         error = ipcperm(p, &shmseg->shm_perm,
 287                         (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 288         if (error)
 289                 goto done;
 290         for (i = 0; i < shminfo.shmseg; i++) {
 291                 if (shmmap_s->shmid == -1)
 292                         break;
 293                 shmmap_s++;
 294         }
 295         if (i >= shminfo.shmseg) {
 296                 error = EMFILE;
 297                 goto done;
 298         }
 299         size = round_page(shmseg->shm_segsz);
 300 #ifdef VM_PROT_READ_IS_EXEC
 301         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 302 #else
 303         prot = VM_PROT_READ;
 304 #endif
 305         if ((uap->shmflg & SHM_RDONLY) == 0)
 306                 prot |= VM_PROT_WRITE;
 307         flags = MAP_ANON | MAP_SHARED;
 308         if (uap->shmaddr) {
 309                 flags |= MAP_FIXED;
 310                 if (uap->shmflg & SHM_RND) {
 311                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
 312                 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
 313                         attach_va = (vm_offset_t)uap->shmaddr;
 314                 } else {
 315                         error = EINVAL;
 316                         goto done;
 317                 }
 318         } else {
 319                 /*
 320                  * This is just a hint to vm_map_find() about where to put it.
 321                  */
 322                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
 323         }
 324
 325         shm_handle = shmseg->shm_internal;
 326         vm_object_hold(shm_handle->shm_object);
 327         vm_object_reference_locked(shm_handle->shm_object);
 328         rv = vm_map_find(&p->p_vmspace->vm_map,
 329                          shm_handle->shm_object, 0,
 330                          &attach_va,
 331                          size, PAGE_SIZE,
 332                          ((flags & MAP_FIXED) ? 0 : 1),
 333                          VM_MAPTYPE_NORMAL,
 334                          prot, prot,
 335                          0);
 336         vm_object_drop(shm_handle->shm_object);
 337         if (rv != KERN_SUCCESS) {
 338                 vm_object_deallocate(shm_handle->shm_object);
 339                 error = ENOMEM;
 340                 goto done;
 341         }
 342         vm_map_inherit(&p->p_vmspace->vm_map,
 343                        attach_va, attach_va + size, VM_INHERIT_SHARE);
 344
 345         KKASSERT(shmmap_s->shmid == -1);
 346         shmmap_s->va = attach_va;
 347         shmmap_s->shmid = uap->shmid;
 348         shmseg->shm_lpid = p->p_pid;
 349         shmseg->shm_atime = time_second;
 350         shmseg->shm_nattch++;
 351         uap->sysmsg_resultp = (void *)attach_va;
 352         error = 0;
 353 done:
 354         rel_mplock();
 355         return error;
 356 }
 357
 358 struct oshmid_ds {
 359         struct  ipc_perm shm_perm;      /* operation perms */
 360         int     shm_segsz;              /* size of segment (bytes) */
 361         ushort  shm_cpid;               /* pid, creator */
 362         ushort  shm_lpid;               /* pid, last operation */
 363         short   shm_nattch;             /* no. of current attaches */
 364         time_t  shm_atime;              /* last attach time */
 365         time_t  shm_dtime;              /* last detach time */
 366         time_t  shm_ctime;              /* last change time */
 367         void    *shm_handle;            /* internal handle for shm segment */
 368 };
 369
 370 struct oshmctl_args {
 371         struct sysmsg sysmsg;
 372         int shmid;
 373         int cmd;
 374         struct oshmid_ds *ubuf;
 375 };
 376
 377 /*
 378  * MPALMOSTSAFE
 379  */
 380 static int
 381 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
 382 {
 383 #ifdef COMPAT_43
 384         struct thread *td = curthread;
 385         struct shmid_ds *shmseg;
 386         struct oshmid_ds outbuf;
 387         int error;
 388
 389         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 390                 return (ENOSYS);
 391
 392         get_mplock();
 393         shmseg = shm_find_segment_by_shmid(uap->shmid);
 394         if (shmseg == NULL) {
 395                 error = EINVAL;
 396                 goto done;
 397         }
 398
 399         switch (uap->cmd) {
 400         case IPC_STAT:
 401                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 402                 if (error)
 403                         break;
 404                 outbuf.shm_perm = shmseg->shm_perm;
 405                 outbuf.shm_segsz = shmseg->shm_segsz;
 406                 outbuf.shm_cpid = shmseg->shm_cpid;
 407                 outbuf.shm_lpid = shmseg->shm_lpid;
 408                 outbuf.shm_nattch = shmseg->shm_nattch;
 409                 outbuf.shm_atime = shmseg->shm_atime;
 410                 outbuf.shm_dtime = shmseg->shm_dtime;
 411                 outbuf.shm_ctime = shmseg->shm_ctime;
 412                 outbuf.shm_handle = shmseg->shm_internal;
 413                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
 414                 break;
 415         default:
 416                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 417                 error = sys_shmctl((struct shmctl_args *)uap);
 418         }
 419 done:
 420         rel_mplock();
 421         return error;
 422 #else
 423         return EINVAL;
 424 #endif
 425 }
 426
 427 /*
 428  * MPALMOSTSAFE
 429  */
 430 int
 431 sys_shmctl(struct shmctl_args *uap)
 432 {
 433         struct thread *td = curthread;
 434         struct proc *p = td->td_proc;
 435         int error;
 436         struct shmid_ds inbuf;
 437         struct shmid_ds *shmseg;
 438
 439         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 440                 return (ENOSYS);
 441
 442         get_mplock();
 443         shmseg = shm_find_segment_by_shmid(uap->shmid);
 444         if (shmseg == NULL) {
 445                 error = EINVAL;
 446                 goto done;
 447         }
 448
 449         switch (uap->cmd) {
 450         case IPC_STAT:
 451                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
 452                 if (error == 0)
 453                         error = copyout(shmseg, uap->buf, sizeof(inbuf));
 454                 break;
 455         case IPC_SET:
 456                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 457                 if (error == 0)
 458                         error = copyin(uap->buf, &inbuf, sizeof(inbuf));
 459                 if (error == 0) {
 460                         shmseg->shm_perm.uid = inbuf.shm_perm.uid;
 461                         shmseg->shm_perm.gid = inbuf.shm_perm.gid;
 462                         shmseg->shm_perm.mode =
 463                             (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 464                             (inbuf.shm_perm.mode & ACCESSPERMS);
 465                         shmseg->shm_ctime = time_second;
 466                 }
 467                 break;
 468         case IPC_RMID:
 469                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
 470                 if (error == 0) {
 471                         shmseg->shm_perm.key = IPC_PRIVATE;
 472                         shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 473                         if (shmseg->shm_nattch <= 0) {
 474                                 shm_deallocate_segment(shmseg);
 475                                 shm_last_free = IPCID_TO_IX(uap->shmid);
 476                         }
 477                 }
 478                 break;
 479 #if 0
 480         case SHM_LOCK:
 481         case SHM_UNLOCK:
 482 #endif
 483         default:
 484                 error = EINVAL;
 485                 break;
 486         }
 487 done:
 488         rel_mplock();
 489         return error;
 490 }
 491
 492 static int
 493 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
 494 {
 495         struct shmid_ds *shmseg;
 496         int error;
 497
 498         shmseg = &shmsegs[segnum];
 499         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 500                 /*
 501                  * This segment is in the process of being allocated.  Wait
 502                  * until it's done, and look the key up again (in case the
 503                  * allocation failed or it was freed).
 504                  */
 505                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 506                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
 507                 if (error)
 508                         return error;
 509                 return EAGAIN;
 510         }
 511         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 512                 return EEXIST;
 513         error = ipcperm(p, &shmseg->shm_perm, mode);
 514         if (error)
 515                 return error;
 516         if (uap->size && uap->size > shmseg->shm_segsz)
 517                 return EINVAL;
 518         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 519         return 0;
 520 }
 521
 522 static int
 523 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
 524 {
 525         int i, segnum, shmid;
 526         size_t size;
 527         struct ucred *cred = p->p_ucred;
 528         struct shmid_ds *shmseg;
 529         struct shm_handle *shm_handle;
 530
 531         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
 532                 return EINVAL;
 533         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
 534                 return ENOSPC;
 535         size = round_page(uap->size);
 536         if (shm_committed + btoc(size) > shminfo.shmall)
 537                 return ENOMEM;
 538         if (shm_last_free < 0) {
 539                 shmrealloc();   /* maybe expand the shmsegs[] array */
 540                 for (i = 0; i < shmalloced; i++) {
 541                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 542                                 break;
 543                 }
 544                 if (i == shmalloced)
 545                         return ENOSPC;
 546                 segnum = i;
 547         } else  {
 548                 segnum = shm_last_free;
 549                 shm_last_free = -1;
 550         }
 551         shmseg = &shmsegs[segnum];
 552         /*
 553          * In case we sleep in malloc(), mark the segment present but deleted
 554          * so that noone else tries to create the same key.
 555          */
 556         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 557         shmseg->shm_perm.key = uap->key;
 558         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 559         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 560         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 561
 562         /*
 563          * We make sure that we have allocated a pager before we need
 564          * to.
 565          */
 566         if (shm_use_phys) {
 567                 shm_handle->shm_object =
 568                    phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 569         } else {
 570                 shm_handle->shm_object =
 571                    swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
 572         }
 573         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 574         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 575
 576         shmseg->shm_internal = shm_handle;
 577         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 578         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 579         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 580             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 581         shmseg->shm_segsz = uap->size;
 582         shmseg->shm_cpid = p->p_pid;
 583         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 584         shmseg->shm_atime = shmseg->shm_dtime = 0;
 585         shmseg->shm_ctime = time_second;
 586         shm_committed += btoc(size);
 587         shm_nused++;
 588
 589         /*
 590          * If a physical mapping is desired and we have a ton of free pages
 591          * we pre-allocate the pages here in order to avoid on-the-fly
 592          * allocation later.  This has a big effect on database warm-up
 593          * times since DFly supports concurrent page faults coming from the
 594          * same VM object for pages which already exist.
 595          *
 596          * This can hang the kernel for a while so only do it if shm_use_phys
 597          * is set to 2 or higher.
 598          */
 599         if (shm_use_phys > 1) {
 600                 vm_pindex_t pi, pmax;
 601                 vm_page_t m;
 602
 603                 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
 604                 vm_object_hold(shm_handle->shm_object);
 605                 if (pmax > vmstats.v_free_count)
 606                         pmax = vmstats.v_free_count;
 607                 for (pi = 0; pi < pmax; ++pi) {
 608                         m = vm_page_grab(shm_handle->shm_object, pi,
 609                                          VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
 610                                          VM_ALLOC_ZERO);
 611                         if (m == NULL)
 612                                 break;
 613                         vm_pager_get_page(shm_handle->shm_object, &m, 1);
 614                         vm_page_activate(m);
 615                         vm_page_wakeup(m);
 616                         lwkt_yield();
 617                 }
 618                 vm_object_drop(shm_handle->shm_object);
 619         }
 620
 621         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 622                 /*
 623                  * Somebody else wanted this key while we were asleep.  Wake
 624                  * them up now.
 625                  */
 626                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 627                 wakeup((caddr_t)shmseg);
 628         }
 629         uap->sysmsg_result = shmid;
 630         return 0;
 631 }
 632
 633 /*
 634  * MPALMOSTSAFE
 635  */
 636 int
 637 sys_shmget(struct shmget_args *uap)
 638 {
 639         struct thread *td = curthread;
 640         struct proc *p = td->td_proc;
 641         int segnum, mode, error;
 642
 643         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 644                 return (ENOSYS);
 645
 646         mode = uap->shmflg & ACCESSPERMS;
 647         get_mplock();
 648
 649         if (uap->key != IPC_PRIVATE) {
 650         again:
 651                 segnum = shm_find_segment_by_key(uap->key);
 652                 if (segnum >= 0) {
 653                         error = shmget_existing(p, uap, mode, segnum);
 654                         if (error == EAGAIN)
 655                                 goto again;
 656                         goto done;
 657                 }
 658                 if ((uap->shmflg & IPC_CREAT) == 0) {
 659                         error = ENOENT;
 660                         goto done;
 661                 }
 662         }
 663         error = shmget_allocate_segment(p, uap, mode);
 664 done:
 665         rel_mplock();
 666         return (error);
 667 }
 668
 669 /*
 670  * shmsys_args(int which, int a2, ...) (VARARGS)
 671  *
 672  * MPALMOSTSAFE
 673  */
 674 int
 675 sys_shmsys(struct shmsys_args *uap)
 676 {
 677         struct thread *td = curthread;
 678         unsigned int which = (unsigned int)uap->which;
 679         int error;
 680
 681         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 682                 return (ENOSYS);
 683
 684         if (which >= NELEM(shmcalls))
 685                 return EINVAL;
 686         get_mplock();
 687         bcopy(&uap->a2, &uap->which,
 688                 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
 689         error = ((*shmcalls[which])(uap));
 690         rel_mplock();
 691
 692         return(error);
 693 }
 694
 695 void
 696 shmfork(struct proc *p1, struct proc *p2)
 697 {
 698         struct shmmap_state *shmmap_s;
 699         size_t size;
 700         int i;
 701
 702         get_mplock();
 703         size = shminfo.shmseg * sizeof(struct shmmap_state);
 704         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
 705         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
 706         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
 707         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 708                 if (shmmap_s->shmid != -1)
 709                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 710         }
 711         rel_mplock();
 712 }
 713
 714 void
 715 shmexit(struct vmspace *vm)
 716 {
 717         struct shmmap_state *base, *shm;
 718         int i;
 719
 720         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
 721                 vm->vm_shm = NULL;
 722                 get_mplock();
 723                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 724                         if (shm->shmid != -1)
 725                                 shm_delete_mapping(vm, shm);
 726                 }
 727                 kfree(base, M_SHM);
 728                 rel_mplock();
 729         }
 730 }
 731
 732 static void
 733 shmrealloc(void)
 734 {
 735         int i;
 736         struct shmid_ds *newsegs;
 737
 738         if (shmalloced >= shminfo.shmmni)
 739                 return;
 740
 741         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 742         for (i = 0; i < shmalloced; i++)
 743                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 744         for (; i < shminfo.shmmni; i++) {
 745                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 746                 shmsegs[i].shm_perm.seq = 0;
 747         }
 748         kfree(shmsegs, M_SHM);
 749         shmsegs = newsegs;
 750         shmalloced = shminfo.shmmni;
 751 }
 752
 753 static void
 754 shminit(void *dummy)
 755 {
 756         int i;
 757
 758         /*
 759          * If not overridden by a tunable set the maximum shm to
 760          * 2/3 of main memory.
 761          */
 762         if (shminfo.shmall == 0)
 763                 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
 764
 765         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
 766         shmalloced = shminfo.shmmni;
 767         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 768         for (i = 0; i < shmalloced; i++) {
 769                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 770                 shmsegs[i].shm_perm.seq = 0;
 771         }
 772         shm_last_free = 0;
 773         shm_nused = 0;
 774         shm_committed = 0;
 775 }
 776 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);