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