sysv_shm.c revision 57884
1/* $FreeBSD: head/sys/kern/sysv_shm.c 57884 2000-03-10 09:11:24Z alc $ */ 2/* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */ 3 4/* 5 * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by Adam Glass and Charles 18 * Hannum. 19 * 4. The names of the authors may not be used to endorse or promote products 20 * derived from this software without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34#include "opt_compat.h" 35#include "opt_rlimit.h" 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/sysproto.h> 40#include <sys/kernel.h> 41#include <sys/shm.h> 42#include <sys/proc.h> 43#include <sys/malloc.h> 44#include <sys/mman.h> 45#include <sys/stat.h> 46#include <sys/sysent.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#ifndef _SYS_SYSPROTO_H_ 58struct shmat_args; 59extern int shmat __P((struct proc *p, struct shmat_args *uap)); 60struct shmctl_args; 61extern int shmctl __P((struct proc *p, struct shmctl_args *uap)); 62struct shmdt_args; 63extern int shmdt __P((struct proc *p, struct shmdt_args *uap)); 64struct shmget_args; 65extern int shmget __P((struct proc *p, struct shmget_args *uap)); 66#endif 67 68static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments"); 69 70static void shminit __P((void *)); 71SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL) 72 73struct oshmctl_args; 74static int oshmctl __P((struct proc *p, struct oshmctl_args *uap)); 75static int shmget_allocate_segment __P((struct proc *p, struct shmget_args *uap, int mode)); 76static int shmget_existing __P((struct proc *p, struct shmget_args *uap, int mode, int segnum)); 77 78/* XXX casting to (sy_call_t *) is bogus, as usual. */ 79static sy_call_t *shmcalls[] = { 80 (sy_call_t *)shmat, (sy_call_t *)oshmctl, 81 (sy_call_t *)shmdt, (sy_call_t *)shmget, 82 (sy_call_t *)shmctl 83}; 84 85#define SHMSEG_FREE 0x0200 86#define SHMSEG_REMOVED 0x0400 87#define SHMSEG_ALLOCATED 0x0800 88#define SHMSEG_WANTED 0x1000 89 90static int shm_last_free, shm_nused, shm_committed; 91struct shmid_ds *shmsegs; 92 93struct shm_handle { 94 /* vm_offset_t kva; */ 95 vm_object_t shm_object; 96}; 97 98struct shmmap_state { 99 vm_offset_t va; 100 int shmid; 101}; 102 103static void shm_deallocate_segment __P((struct shmid_ds *)); 104static int shm_find_segment_by_key __P((key_t)); 105static struct shmid_ds *shm_find_segment_by_shmid __P((int)); 106static int shm_delete_mapping __P((struct proc *, struct shmmap_state *)); 107 108static int 109shm_find_segment_by_key(key) 110 key_t key; 111{ 112 int i; 113 114 for (i = 0; i < shminfo.shmmni; i++) 115 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) && 116 shmsegs[i].shm_perm.key == key) 117 return i; 118 return -1; 119} 120 121static struct shmid_ds * 122shm_find_segment_by_shmid(shmid) 123 int shmid; 124{ 125 int segnum; 126 struct shmid_ds *shmseg; 127 128 segnum = IPCID_TO_IX(shmid); 129 if (segnum < 0 || segnum >= shminfo.shmmni) 130 return NULL; 131 shmseg = &shmsegs[segnum]; 132 if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED)) 133 != SHMSEG_ALLOCATED || 134 shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) 135 return NULL; 136 return shmseg; 137} 138 139static void 140shm_deallocate_segment(shmseg) 141 struct shmid_ds *shmseg; 142{ 143 struct shm_handle *shm_handle; 144 size_t size; 145 146 shm_handle = shmseg->shm_internal; 147 vm_object_deallocate(shm_handle->shm_object); 148 free((caddr_t)shm_handle, M_SHM); 149 shmseg->shm_internal = NULL; 150 size = round_page(shmseg->shm_segsz); 151 shm_committed -= btoc(size); 152 shm_nused--; 153 shmseg->shm_perm.mode = SHMSEG_FREE; 154} 155 156static int 157shm_delete_mapping(p, shmmap_s) 158 struct proc *p; 159 struct shmmap_state *shmmap_s; 160{ 161 struct shmid_ds *shmseg; 162 int segnum, result; 163 size_t size; 164 165 segnum = IPCID_TO_IX(shmmap_s->shmid); 166 shmseg = &shmsegs[segnum]; 167 size = round_page(shmseg->shm_segsz); 168 result = vm_map_remove(&p->p_vmspace->vm_map, shmmap_s->va, shmmap_s->va + size); 169 if (result != KERN_SUCCESS) 170 return EINVAL; 171 shmmap_s->shmid = -1; 172 shmseg->shm_dtime = time_second; 173 if ((--shmseg->shm_nattch <= 0) && 174 (shmseg->shm_perm.mode & SHMSEG_REMOVED)) { 175 shm_deallocate_segment(shmseg); 176 shm_last_free = segnum; 177 } 178 return 0; 179} 180 181#ifndef _SYS_SYSPROTO_H_ 182struct shmdt_args { 183 void *shmaddr; 184}; 185#endif 186 187int 188shmdt(p, uap) 189 struct proc *p; 190 struct shmdt_args *uap; 191{ 192 struct shmmap_state *shmmap_s; 193 int i; 194 195 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 196 if (shmmap_s == NULL) 197 return EINVAL; 198 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 199 if (shmmap_s->shmid != -1 && 200 shmmap_s->va == (vm_offset_t)uap->shmaddr) 201 break; 202 if (i == shminfo.shmseg) 203 return EINVAL; 204 return shm_delete_mapping(p, shmmap_s); 205} 206 207#ifndef _SYS_SYSPROTO_H_ 208struct shmat_args { 209 int shmid; 210 void *shmaddr; 211 int shmflg; 212}; 213#endif 214 215int 216shmat(p, uap) 217 struct proc *p; 218 struct shmat_args *uap; 219{ 220 int error, i, flags; 221 struct shmid_ds *shmseg; 222 struct shmmap_state *shmmap_s = NULL; 223 struct shm_handle *shm_handle; 224 vm_offset_t attach_va; 225 vm_prot_t prot; 226 vm_size_t size; 227 int rv; 228 229 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 230 if (shmmap_s == NULL) { 231 size = shminfo.shmseg * sizeof(struct shmmap_state); 232 shmmap_s = malloc(size, M_SHM, M_WAITOK); 233 for (i = 0; i < shminfo.shmseg; i++) 234 shmmap_s[i].shmid = -1; 235 p->p_vmspace->vm_shm = (caddr_t)shmmap_s; 236 } 237 shmseg = shm_find_segment_by_shmid(uap->shmid); 238 if (shmseg == NULL) 239 return EINVAL; 240 error = ipcperm(p, &shmseg->shm_perm, 241 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W); 242 if (error) 243 return error; 244 for (i = 0; i < shminfo.shmseg; i++) { 245 if (shmmap_s->shmid == -1) 246 break; 247 shmmap_s++; 248 } 249 if (i >= shminfo.shmseg) 250 return EMFILE; 251 size = round_page(shmseg->shm_segsz); 252#ifdef VM_PROT_READ_IS_EXEC 253 prot = VM_PROT_READ | VM_PROT_EXECUTE; 254#else 255 prot = VM_PROT_READ; 256#endif 257 if ((uap->shmflg & SHM_RDONLY) == 0) 258 prot |= VM_PROT_WRITE; 259 flags = MAP_ANON | MAP_SHARED; 260 if (uap->shmaddr) { 261 flags |= MAP_FIXED; 262 if (uap->shmflg & SHM_RND) 263 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1); 264 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) 265 attach_va = (vm_offset_t)uap->shmaddr; 266 else 267 return EINVAL; 268 } else { 269 /* This is just a hint to vm_map_find() about where to put it. */ 270 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + MAXTSIZ + MAXDSIZ); 271 } 272 273 shm_handle = shmseg->shm_internal; 274 vm_object_reference(shm_handle->shm_object); 275 rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object, 276 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0); 277 if (rv != KERN_SUCCESS) { 278 return ENOMEM; 279 } 280 vm_map_inherit(&p->p_vmspace->vm_map, 281 attach_va, attach_va + size, VM_INHERIT_SHARE); 282 283 shmmap_s->va = attach_va; 284 shmmap_s->shmid = uap->shmid; 285 shmseg->shm_lpid = p->p_pid; 286 shmseg->shm_atime = time_second; 287 shmseg->shm_nattch++; 288 p->p_retval[0] = attach_va; 289 return 0; 290} 291 292struct oshmid_ds { 293 struct ipc_perm shm_perm; /* operation perms */ 294 int shm_segsz; /* size of segment (bytes) */ 295 ushort shm_cpid; /* pid, creator */ 296 ushort shm_lpid; /* pid, last operation */ 297 short shm_nattch; /* no. of current attaches */ 298 time_t shm_atime; /* last attach time */ 299 time_t shm_dtime; /* last detach time */ 300 time_t shm_ctime; /* last change time */ 301 void *shm_handle; /* internal handle for shm segment */ 302}; 303 304struct oshmctl_args { 305 int shmid; 306 int cmd; 307 struct oshmid_ds *ubuf; 308}; 309 310static int 311oshmctl(p, uap) 312 struct proc *p; 313 struct oshmctl_args *uap; 314{ 315#ifdef COMPAT_43 316 int error; 317 struct shmid_ds *shmseg; 318 struct oshmid_ds outbuf; 319 320 shmseg = shm_find_segment_by_shmid(uap->shmid); 321 if (shmseg == NULL) 322 return EINVAL; 323 switch (uap->cmd) { 324 case IPC_STAT: 325 error = ipcperm(p, &shmseg->shm_perm, IPC_R); 326 if (error) 327 return error; 328 outbuf.shm_perm = shmseg->shm_perm; 329 outbuf.shm_segsz = shmseg->shm_segsz; 330 outbuf.shm_cpid = shmseg->shm_cpid; 331 outbuf.shm_lpid = shmseg->shm_lpid; 332 outbuf.shm_nattch = shmseg->shm_nattch; 333 outbuf.shm_atime = shmseg->shm_atime; 334 outbuf.shm_dtime = shmseg->shm_dtime; 335 outbuf.shm_ctime = shmseg->shm_ctime; 336 outbuf.shm_handle = shmseg->shm_internal; 337 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf)); 338 if (error) 339 return error; 340 break; 341 default: 342 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 343 return ((sy_call_t *)shmctl)(p, uap); 344 } 345 return 0; 346#else 347 return EINVAL; 348#endif 349} 350 351#ifndef _SYS_SYSPROTO_H_ 352struct shmctl_args { 353 int shmid; 354 int cmd; 355 struct shmid_ds *buf; 356}; 357#endif 358 359int 360shmctl(p, uap) 361 struct proc *p; 362 struct shmctl_args *uap; 363{ 364 int error; 365 struct shmid_ds inbuf; 366 struct shmid_ds *shmseg; 367 368 shmseg = shm_find_segment_by_shmid(uap->shmid); 369 if (shmseg == NULL) 370 return EINVAL; 371 switch (uap->cmd) { 372 case IPC_STAT: 373 error = ipcperm(p, &shmseg->shm_perm, IPC_R); 374 if (error) 375 return error; 376 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf)); 377 if (error) 378 return error; 379 break; 380 case IPC_SET: 381 error = ipcperm(p, &shmseg->shm_perm, IPC_M); 382 if (error) 383 return error; 384 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf)); 385 if (error) 386 return error; 387 shmseg->shm_perm.uid = inbuf.shm_perm.uid; 388 shmseg->shm_perm.gid = inbuf.shm_perm.gid; 389 shmseg->shm_perm.mode = 390 (shmseg->shm_perm.mode & ~ACCESSPERMS) | 391 (inbuf.shm_perm.mode & ACCESSPERMS); 392 shmseg->shm_ctime = time_second; 393 break; 394 case IPC_RMID: 395 error = ipcperm(p, &shmseg->shm_perm, IPC_M); 396 if (error) 397 return error; 398 shmseg->shm_perm.key = IPC_PRIVATE; 399 shmseg->shm_perm.mode |= SHMSEG_REMOVED; 400 if (shmseg->shm_nattch <= 0) { 401 shm_deallocate_segment(shmseg); 402 shm_last_free = IPCID_TO_IX(uap->shmid); 403 } 404 break; 405#if 0 406 case SHM_LOCK: 407 case SHM_UNLOCK: 408#endif 409 default: 410 return EINVAL; 411 } 412 return 0; 413} 414 415#ifndef _SYS_SYSPROTO_H_ 416struct shmget_args { 417 key_t key; 418 size_t size; 419 int shmflg; 420}; 421#endif 422 423static int 424shmget_existing(p, uap, mode, segnum) 425 struct proc *p; 426 struct shmget_args *uap; 427 int mode; 428 int segnum; 429{ 430 struct shmid_ds *shmseg; 431 int error; 432 433 shmseg = &shmsegs[segnum]; 434 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) { 435 /* 436 * This segment is in the process of being allocated. Wait 437 * until it's done, and look the key up again (in case the 438 * allocation failed or it was freed). 439 */ 440 shmseg->shm_perm.mode |= SHMSEG_WANTED; 441 error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0); 442 if (error) 443 return error; 444 return EAGAIN; 445 } 446 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) 447 return EEXIST; 448 error = ipcperm(p, &shmseg->shm_perm, mode); 449 if (error) 450 return error; 451 if (uap->size && uap->size > shmseg->shm_segsz) 452 return EINVAL; 453 p->p_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); 454 return 0; 455} 456 457static int 458shmget_allocate_segment(p, uap, mode) 459 struct proc *p; 460 struct shmget_args *uap; 461 int mode; 462{ 463 int i, segnum, shmid, 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 for (i = 0; i < shminfo.shmmni; i++) 477 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE) 478 break; 479 if (i == shminfo.shmmni) 480 panic("shmseg free count inconsistent"); 481 segnum = i; 482 } else { 483 segnum = shm_last_free; 484 shm_last_free = -1; 485 } 486 shmseg = &shmsegs[segnum]; 487 /* 488 * In case we sleep in malloc(), mark the segment present but deleted 489 * so that noone else tries to create the same key. 490 */ 491 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED; 492 shmseg->shm_perm.key = uap->key; 493 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff; 494 shm_handle = (struct shm_handle *) 495 malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK); 496 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); 497 498 /* 499 * We make sure that we have allocated a pager before we need 500 * to. 501 */ 502 shm_handle->shm_object = 503 vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0); 504 vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING); 505 vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT); 506 507 shmseg->shm_internal = shm_handle; 508 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid; 509 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid; 510 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) | 511 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; 512 shmseg->shm_segsz = uap->size; 513 shmseg->shm_cpid = p->p_pid; 514 shmseg->shm_lpid = shmseg->shm_nattch = 0; 515 shmseg->shm_atime = shmseg->shm_dtime = 0; 516 shmseg->shm_ctime = time_second; 517 shm_committed += btoc(size); 518 shm_nused++; 519 if (shmseg->shm_perm.mode & SHMSEG_WANTED) { 520 /* 521 * Somebody else wanted this key while we were asleep. Wake 522 * them up now. 523 */ 524 shmseg->shm_perm.mode &= ~SHMSEG_WANTED; 525 wakeup((caddr_t)shmseg); 526 } 527 p->p_retval[0] = shmid; 528 return 0; 529} 530 531int 532shmget(p, uap) 533 struct proc *p; 534 struct shmget_args *uap; 535{ 536 int segnum, mode, error; 537 538 mode = uap->shmflg & ACCESSPERMS; 539 if (uap->key != IPC_PRIVATE) { 540 again: 541 segnum = shm_find_segment_by_key(uap->key); 542 if (segnum >= 0) { 543 error = shmget_existing(p, uap, mode, segnum); 544 if (error == EAGAIN) 545 goto again; 546 return error; 547 } 548 if ((uap->shmflg & IPC_CREAT) == 0) 549 return ENOENT; 550 } 551 return shmget_allocate_segment(p, uap, mode); 552} 553 554int 555shmsys(p, uap) 556 struct proc *p; 557 /* XXX actually varargs. */ 558 struct shmsys_args /* { 559 u_int which; 560 int a2; 561 int a3; 562 int a4; 563 } */ *uap; 564{ 565 566 if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0])) 567 return EINVAL; 568 return ((*shmcalls[uap->which])(p, &uap->a2)); 569} 570 571void 572shmfork(p1, p2) 573 struct proc *p1, *p2; 574{ 575 struct shmmap_state *shmmap_s; 576 size_t size; 577 int i; 578 579 size = shminfo.shmseg * sizeof(struct shmmap_state); 580 shmmap_s = malloc(size, M_SHM, M_WAITOK); 581 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size); 582 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s; 583 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 584 if (shmmap_s->shmid != -1) 585 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++; 586} 587 588void 589shmexit(p) 590 struct proc *p; 591{ 592 struct shmmap_state *shmmap_s; 593 int i; 594 595 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 596 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 597 if (shmmap_s->shmid != -1) 598 shm_delete_mapping(p, shmmap_s); 599 free((caddr_t)p->p_vmspace->vm_shm, M_SHM); 600 p->p_vmspace->vm_shm = NULL; 601} 602 603void 604shminit(dummy) 605 void *dummy; 606{ 607 int i; 608 for (i = 0; i < shminfo.shmmni; i++) { 609 shmsegs[i].shm_perm.mode = SHMSEG_FREE; 610 shmsegs[i].shm_perm.seq = 0; 611 } 612 shm_last_free = 0; 613 shm_nused = 0; 614 shm_committed = 0; 615} 616