sysv_shm.c revision 30354
1/* $Id: sysv_shm.c,v 1.30 1997/10/11 18:31:25 phk Exp $ */ 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_rlimit.h" 35 36#include <sys/param.h> 37#include <sys/systm.h> 38#include <sys/sysproto.h> 39#include <sys/kernel.h> 40#include <sys/shm.h> 41#include <sys/proc.h> 42#include <sys/malloc.h> 43#include <sys/mman.h> 44#include <sys/stat.h> 45#include <sys/sysent.h> 46 47#include <vm/vm.h> 48#include <vm/vm_param.h> 49#include <vm/vm_prot.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_pager.h> 55#include <vm/vm_inherit.h> 56 57#ifndef _SYS_SYSPROTO_H_ 58struct shmat_args; 59extern int shmat __P((struct proc *p, struct shmat_args *uap, int *retval)); 60struct shmctl_args; 61extern int shmctl __P((struct proc *p, struct shmctl_args *uap, int *retval)); 62struct shmdt_args; 63extern int shmdt __P((struct proc *p, struct shmdt_args *uap, int *retval)); 64struct shmget_args; 65extern int shmget __P((struct proc *p, struct shmget_args *uap, int *retval)); 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, int *retval)); 75static int shmget_allocate_segment __P((struct proc *p, struct shmget_args *uap, int mode, int *retval)); 76static int shmget_existing __P((struct proc *p, struct shmget_args *uap, int mode, int segnum, int *retval)); 77 78/* XXX casting to (sy_call_t *) is bogus, as usual. */ 79sy_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.tv_sec; 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, retval) 189 struct proc *p; 190 struct shmdt_args *uap; 191 int *retval; 192{ 193 struct shmmap_state *shmmap_s; 194 int i; 195 196 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 197 if (shmmap_s == NULL) 198 return EINVAL; 199 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 200 if (shmmap_s->shmid != -1 && 201 shmmap_s->va == (vm_offset_t)uap->shmaddr) 202 break; 203 if (i == shminfo.shmseg) 204 return EINVAL; 205 return shm_delete_mapping(p, shmmap_s); 206} 207 208#ifndef _SYS_SYSPROTO_H_ 209struct shmat_args { 210 int shmid; 211 void *shmaddr; 212 int shmflg; 213}; 214#endif 215 216int 217shmat(p, uap, retval) 218 struct proc *p; 219 struct shmat_args *uap; 220 int *retval; 221{ 222 int error, i, flags; 223 struct ucred *cred = p->p_ucred; 224 struct shmid_ds *shmseg; 225 struct shmmap_state *shmmap_s = NULL; 226 struct shm_handle *shm_handle; 227 vm_offset_t attach_va; 228 vm_prot_t prot; 229 vm_size_t size; 230 int rv; 231 232 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 233 if (shmmap_s == NULL) { 234 size = shminfo.shmseg * sizeof(struct shmmap_state); 235 shmmap_s = malloc(size, M_SHM, M_WAITOK); 236 for (i = 0; i < shminfo.shmseg; i++) 237 shmmap_s[i].shmid = -1; 238 p->p_vmspace->vm_shm = (caddr_t)shmmap_s; 239 } 240 shmseg = shm_find_segment_by_shmid(uap->shmid); 241 if (shmseg == NULL) 242 return EINVAL; 243 error = ipcperm(cred, &shmseg->shm_perm, 244 (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W); 245 if (error) 246 return error; 247 for (i = 0; i < shminfo.shmseg; i++) { 248 if (shmmap_s->shmid == -1) 249 break; 250 shmmap_s++; 251 } 252 if (i >= shminfo.shmseg) 253 return EMFILE; 254 size = round_page(shmseg->shm_segsz); 255 prot = VM_PROT_READ; 256 if ((uap->shmflg & SHM_RDONLY) == 0) 257 prot |= VM_PROT_WRITE; 258 flags = MAP_ANON | MAP_SHARED; 259 if (uap->shmaddr) { 260 flags |= MAP_FIXED; 261 if (uap->shmflg & SHM_RND) 262 attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1); 263 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) 264 attach_va = (vm_offset_t)uap->shmaddr; 265 else 266 return EINVAL; 267 } else { 268 /* This is just a hint to vm_map_find() about where to put it. */ 269 attach_va = round_page(p->p_vmspace->vm_taddr + MAXTSIZ + MAXDSIZ); 270 } 271 272 shm_handle = shmseg->shm_internal; 273 vm_object_reference(shm_handle->shm_object); 274 rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object, 275 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0); 276 if (rv != KERN_SUCCESS) { 277 return ENOMEM; 278 } 279 vm_map_inherit(&p->p_vmspace->vm_map, 280 attach_va, attach_va + size, VM_INHERIT_SHARE); 281 282 shmmap_s->va = attach_va; 283 shmmap_s->shmid = uap->shmid; 284 shmseg->shm_lpid = p->p_pid; 285 shmseg->shm_atime = time.tv_sec; 286 shmseg->shm_nattch++; 287 *retval = attach_va; 288 return 0; 289} 290 291struct oshmid_ds { 292 struct ipc_perm shm_perm; /* operation perms */ 293 int shm_segsz; /* size of segment (bytes) */ 294 ushort shm_cpid; /* pid, creator */ 295 ushort shm_lpid; /* pid, last operation */ 296 short shm_nattch; /* no. of current attaches */ 297 time_t shm_atime; /* last attach time */ 298 time_t shm_dtime; /* last detach time */ 299 time_t shm_ctime; /* last change time */ 300 void *shm_handle; /* internal handle for shm segment */ 301}; 302 303struct oshmctl_args { 304 int shmid; 305 int cmd; 306 struct oshmid_ds *ubuf; 307}; 308 309static int 310oshmctl(p, uap, retval) 311 struct proc *p; 312 struct oshmctl_args *uap; 313 int *retval; 314{ 315#ifdef COMPAT_43 316 int error; 317 struct ucred *cred = p->p_ucred; 318 struct shmid_ds *shmseg; 319 struct oshmid_ds outbuf; 320 321 shmseg = shm_find_segment_by_shmid(uap->shmid); 322 if (shmseg == NULL) 323 return EINVAL; 324 switch (uap->cmd) { 325 case IPC_STAT: 326 error = ipcperm(cred, &shmseg->shm_perm, IPC_R); 327 if (error) 328 return error; 329 outbuf.shm_perm = shmseg->shm_perm; 330 outbuf.shm_segsz = shmseg->shm_segsz; 331 outbuf.shm_cpid = shmseg->shm_cpid; 332 outbuf.shm_lpid = shmseg->shm_lpid; 333 outbuf.shm_nattch = shmseg->shm_nattch; 334 outbuf.shm_atime = shmseg->shm_atime; 335 outbuf.shm_dtime = shmseg->shm_dtime; 336 outbuf.shm_ctime = shmseg->shm_ctime; 337 outbuf.shm_handle = shmseg->shm_internal; 338 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf)); 339 if (error) 340 return error; 341 break; 342 default: 343 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 344 return ((sy_call_t *)shmctl)(p, uap, retval); 345 } 346 return 0; 347#else 348 return EINVAL; 349#endif 350} 351 352#ifndef _SYS_SYSPROTO_H_ 353struct shmctl_args { 354 int shmid; 355 int cmd; 356 struct shmid_ds *buf; 357}; 358#endif 359 360int 361shmctl(p, uap, retval) 362 struct proc *p; 363 struct shmctl_args *uap; 364 int *retval; 365{ 366 int error; 367 struct ucred *cred = p->p_ucred; 368 struct shmid_ds inbuf; 369 struct shmid_ds *shmseg; 370 371 shmseg = shm_find_segment_by_shmid(uap->shmid); 372 if (shmseg == NULL) 373 return EINVAL; 374 switch (uap->cmd) { 375 case IPC_STAT: 376 error = ipcperm(cred, &shmseg->shm_perm, IPC_R); 377 if (error) 378 return error; 379 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf)); 380 if (error) 381 return error; 382 break; 383 case IPC_SET: 384 error = ipcperm(cred, &shmseg->shm_perm, IPC_M); 385 if (error) 386 return error; 387 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf)); 388 if (error) 389 return error; 390 shmseg->shm_perm.uid = inbuf.shm_perm.uid; 391 shmseg->shm_perm.gid = inbuf.shm_perm.gid; 392 shmseg->shm_perm.mode = 393 (shmseg->shm_perm.mode & ~ACCESSPERMS) | 394 (inbuf.shm_perm.mode & ACCESSPERMS); 395 shmseg->shm_ctime = time.tv_sec; 396 break; 397 case IPC_RMID: 398 error = ipcperm(cred, &shmseg->shm_perm, IPC_M); 399 if (error) 400 return error; 401 shmseg->shm_perm.key = IPC_PRIVATE; 402 shmseg->shm_perm.mode |= SHMSEG_REMOVED; 403 if (shmseg->shm_nattch <= 0) { 404 shm_deallocate_segment(shmseg); 405 shm_last_free = IPCID_TO_IX(uap->shmid); 406 } 407 break; 408#if 0 409 case SHM_LOCK: 410 case SHM_UNLOCK: 411#endif 412 default: 413 return EINVAL; 414 } 415 return 0; 416} 417 418#ifndef _SYS_SYSPROTO_H_ 419struct shmget_args { 420 key_t key; 421 size_t size; 422 int shmflg; 423}; 424#endif 425 426static int 427shmget_existing(p, uap, mode, segnum, retval) 428 struct proc *p; 429 struct shmget_args *uap; 430 int mode; 431 int segnum; 432 int *retval; 433{ 434 struct shmid_ds *shmseg; 435 struct ucred *cred = p->p_ucred; 436 int error; 437 438 shmseg = &shmsegs[segnum]; 439 if (shmseg->shm_perm.mode & SHMSEG_REMOVED) { 440 /* 441 * This segment is in the process of being allocated. Wait 442 * until it's done, and look the key up again (in case the 443 * allocation failed or it was freed). 444 */ 445 shmseg->shm_perm.mode |= SHMSEG_WANTED; 446 error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0); 447 if (error) 448 return error; 449 return EAGAIN; 450 } 451 error = ipcperm(cred, &shmseg->shm_perm, mode); 452 if (error) 453 return error; 454 if (uap->size && uap->size > shmseg->shm_segsz) 455 return EINVAL; 456 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) 457 return EEXIST; 458 *retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); 459 return 0; 460} 461 462static int 463shmget_allocate_segment(p, uap, mode, retval) 464 struct proc *p; 465 struct shmget_args *uap; 466 int mode; 467 int *retval; 468{ 469 int i, segnum, shmid, size; 470 struct ucred *cred = p->p_ucred; 471 struct shmid_ds *shmseg; 472 struct shm_handle *shm_handle; 473 474 if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax) 475 return EINVAL; 476 if (shm_nused >= shminfo.shmmni) /* any shmids left? */ 477 return ENOSPC; 478 size = round_page(uap->size); 479 if (shm_committed + btoc(size) > shminfo.shmall) 480 return ENOMEM; 481 if (shm_last_free < 0) { 482 for (i = 0; i < shminfo.shmmni; i++) 483 if (shmsegs[i].shm_perm.mode & SHMSEG_FREE) 484 break; 485 if (i == shminfo.shmmni) 486 panic("shmseg free count inconsistent"); 487 segnum = i; 488 } else { 489 segnum = shm_last_free; 490 shm_last_free = -1; 491 } 492 shmseg = &shmsegs[segnum]; 493 /* 494 * In case we sleep in malloc(), mark the segment present but deleted 495 * so that noone else tries to create the same key. 496 */ 497 shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED; 498 shmseg->shm_perm.key = uap->key; 499 shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff; 500 shm_handle = (struct shm_handle *) 501 malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK); 502 shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); 503 504 /* 505 * We make sure that we have allocated a pager before we need 506 * to. 507 */ 508 shm_handle->shm_object = 509 vm_pager_allocate(OBJT_SWAP, 0, OFF_TO_IDX(size), 510 VM_PROT_DEFAULT, 0); 511 shmseg->shm_internal = shm_handle; 512 shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid; 513 shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid; 514 shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) | 515 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; 516 shmseg->shm_segsz = uap->size; 517 shmseg->shm_cpid = p->p_pid; 518 shmseg->shm_lpid = shmseg->shm_nattch = 0; 519 shmseg->shm_atime = shmseg->shm_dtime = 0; 520 shmseg->shm_ctime = time.tv_sec; 521 shm_committed += btoc(size); 522 shm_nused++; 523 if (shmseg->shm_perm.mode & SHMSEG_WANTED) { 524 /* 525 * Somebody else wanted this key while we were asleep. Wake 526 * them up now. 527 */ 528 shmseg->shm_perm.mode &= ~SHMSEG_WANTED; 529 wakeup((caddr_t)shmseg); 530 } 531 *retval = shmid; 532 return 0; 533} 534 535int 536shmget(p, uap, retval) 537 struct proc *p; 538 struct shmget_args *uap; 539 int *retval; 540{ 541 int segnum, mode, error; 542 543 mode = uap->shmflg & ACCESSPERMS; 544 if (uap->key != IPC_PRIVATE) { 545 again: 546 segnum = shm_find_segment_by_key(uap->key); 547 if (segnum >= 0) { 548 error = shmget_existing(p, uap, mode, segnum, retval); 549 if (error == EAGAIN) 550 goto again; 551 return error; 552 } 553 if ((uap->shmflg & IPC_CREAT) == 0) 554 return ENOENT; 555 } 556 return shmget_allocate_segment(p, uap, mode, retval); 557} 558 559int 560shmsys(p, uap, retval) 561 struct proc *p; 562 /* XXX actually varargs. */ 563 struct shmsys_args /* { 564 u_int which; 565 int a2; 566 int a3; 567 int a4; 568 } */ *uap; 569 int *retval; 570{ 571 572 if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0])) 573 return EINVAL; 574 return ((*shmcalls[uap->which])(p, &uap->a2, retval)); 575} 576 577void 578shmfork(p1, p2) 579 struct proc *p1, *p2; 580{ 581 struct shmmap_state *shmmap_s; 582 size_t size; 583 int i; 584 585 size = shminfo.shmseg * sizeof(struct shmmap_state); 586 shmmap_s = malloc(size, M_SHM, M_WAITOK); 587 bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size); 588 p2->p_vmspace->vm_shm = (caddr_t)shmmap_s; 589 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 590 if (shmmap_s->shmid != -1) 591 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++; 592} 593 594void 595shmexit(p) 596 struct proc *p; 597{ 598 struct shmmap_state *shmmap_s; 599 int i; 600 601 shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm; 602 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 603 if (shmmap_s->shmid != -1) 604 shm_delete_mapping(p, shmmap_s); 605 free((caddr_t)p->p_vmspace->vm_shm, M_SHM); 606 p->p_vmspace->vm_shm = NULL; 607} 608 609void 610shminit(dummy) 611 void *dummy; 612{ 613 int i; 614 for (i = 0; i < shminfo.shmmni; i++) { 615 shmsegs[i].shm_perm.mode = SHMSEG_FREE; 616 shmsegs[i].shm_perm.seq = 0; 617 } 618 shm_last_free = 0; 619 shm_nused = 0; 620 shm_committed = 0; 621} 622