sysv_shm.c revision 194832
1/* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */ 2/*- 3 * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by Adam Glass and Charles 16 * Hannum. 17 * 4. The names of the authors may not be used to endorse or promote products 18 * derived from this software without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR 21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 23 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, 24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 29 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 30 */ 31/*- 32 * Copyright (c) 2003-2005 McAfee, Inc. 33 * All rights reserved. 34 * 35 * This software was developed for the FreeBSD Project in part by McAfee 36 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR 37 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research 38 * program. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 */ 61 62#include <sys/cdefs.h> 63__FBSDID("$FreeBSD: head/sys/kern/sysv_shm.c 194832 2009-06-24 13:35:38Z jhb $"); 64 65#include "opt_compat.h" 66#include "opt_sysvipc.h" 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/kernel.h> 71#include <sys/lock.h> 72#include <sys/sysctl.h> 73#include <sys/shm.h> 74#include <sys/proc.h> 75#include <sys/malloc.h> 76#include <sys/mman.h> 77#include <sys/module.h> 78#include <sys/mutex.h> 79#include <sys/resourcevar.h> 80#include <sys/stat.h> 81#include <sys/syscall.h> 82#include <sys/syscallsubr.h> 83#include <sys/sysent.h> 84#include <sys/sysproto.h> 85#include <sys/jail.h> 86 87#include <security/mac/mac_framework.h> 88 89#include <vm/vm.h> 90#include <vm/vm_param.h> 91#include <vm/pmap.h> 92#include <vm/vm_object.h> 93#include <vm/vm_map.h> 94#include <vm/vm_page.h> 95#include <vm/vm_pager.h> 96 97static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments"); 98 99#if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) 100struct oshmctl_args; 101static int oshmctl(struct thread *td, struct oshmctl_args *uap); 102#endif 103 104static int shmget_allocate_segment(struct thread *td, 105 struct shmget_args *uap, int mode); 106static int shmget_existing(struct thread *td, struct shmget_args *uap, 107 int mode, int segnum); 108 109#if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) 110/* XXX casting to (sy_call_t *) is bogus, as usual. */ 111static sy_call_t *shmcalls[] = { 112 (sy_call_t *)shmat, (sy_call_t *)oshmctl, 113 (sy_call_t *)shmdt, (sy_call_t *)shmget, 114 (sy_call_t *)shmctl 115}; 116#endif 117 118#define SHMSEG_FREE 0x0200 119#define SHMSEG_REMOVED 0x0400 120#define SHMSEG_ALLOCATED 0x0800 121#define SHMSEG_WANTED 0x1000 122 123static int shm_last_free, shm_nused, shmalloced; 124vm_size_t shm_committed; 125static struct shmid_kernel *shmsegs; 126 127struct shmmap_state { 128 vm_offset_t va; 129 int shmid; 130}; 131 132static void shm_deallocate_segment(struct shmid_kernel *); 133static int shm_find_segment_by_key(key_t); 134static struct shmid_kernel *shm_find_segment_by_shmid(int); 135static struct shmid_kernel *shm_find_segment_by_shmidx(int); 136static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *); 137static void shmrealloc(void); 138static void shminit(void); 139static int sysvshm_modload(struct module *, int, void *); 140static int shmunload(void); 141static void shmexit_myhook(struct vmspace *vm); 142static void shmfork_myhook(struct proc *p1, struct proc *p2); 143static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS); 144 145/* 146 * Tuneable values. 147 */ 148#ifndef SHMMAXPGS 149#define SHMMAXPGS 8192 /* Note: sysv shared memory is swap backed. */ 150#endif 151#ifndef SHMMAX 152#define SHMMAX (SHMMAXPGS*PAGE_SIZE) 153#endif 154#ifndef SHMMIN 155#define SHMMIN 1 156#endif 157#ifndef SHMMNI 158#define SHMMNI 192 159#endif 160#ifndef SHMSEG 161#define SHMSEG 128 162#endif 163#ifndef SHMALL 164#define SHMALL (SHMMAXPGS) 165#endif 166 167struct shminfo shminfo = { 168 SHMMAX, 169 SHMMIN, 170 SHMMNI, 171 SHMSEG, 172 SHMALL 173}; 174 175static int shm_use_phys; 176static int shm_allow_removed; 177 178SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, 179 "Maximum shared memory segment size"); 180SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, 181 "Minimum shared memory segment size"); 182SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0, 183 "Number of shared memory identifiers"); 184SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0, 185 "Number of segments per process"); 186SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, 187 "Maximum number of pages available for shared memory"); 188SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, 189 &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core"); 190SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW, 191 &shm_allow_removed, 0, 192 "Enable/Disable attachment to attached segments marked for removal"); 193SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD, 194 NULL, 0, sysctl_shmsegs, "", 195 "Current number of shared memory segments allocated"); 196 197static int 198shm_find_segment_by_key(key) 199 key_t key; 200{ 201 int i; 202 203 for (i = 0; i < shmalloced; i++) 204 if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) && 205 shmsegs[i].u.shm_perm.key == key) 206 return (i); 207 return (-1); 208} 209 210static struct shmid_kernel * 211shm_find_segment_by_shmid(int shmid) 212{ 213 int segnum; 214 struct shmid_kernel *shmseg; 215 216 segnum = IPCID_TO_IX(shmid); 217 if (segnum < 0 || segnum >= shmalloced) 218 return (NULL); 219 shmseg = &shmsegs[segnum]; 220 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 || 221 (!shm_allow_removed && 222 (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) || 223 shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid)) 224 return (NULL); 225 return (shmseg); 226} 227 228static struct shmid_kernel * 229shm_find_segment_by_shmidx(int segnum) 230{ 231 struct shmid_kernel *shmseg; 232 233 if (segnum < 0 || segnum >= shmalloced) 234 return (NULL); 235 shmseg = &shmsegs[segnum]; 236 if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 || 237 (!shm_allow_removed && 238 (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0)) 239 return (NULL); 240 return (shmseg); 241} 242 243static void 244shm_deallocate_segment(shmseg) 245 struct shmid_kernel *shmseg; 246{ 247 vm_size_t size; 248 249 GIANT_REQUIRED; 250 251 vm_object_deallocate(shmseg->u.shm_internal); 252 shmseg->u.shm_internal = NULL; 253 size = round_page(shmseg->shm_bsegsz); 254 shm_committed -= btoc(size); 255 shm_nused--; 256 shmseg->u.shm_perm.mode = SHMSEG_FREE; 257#ifdef MAC 258 mac_sysvshm_cleanup(shmseg); 259#endif 260} 261 262static int 263shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s) 264{ 265 struct shmid_kernel *shmseg; 266 int segnum, result; 267 vm_size_t size; 268 269 GIANT_REQUIRED; 270 271 segnum = IPCID_TO_IX(shmmap_s->shmid); 272 shmseg = &shmsegs[segnum]; 273 size = round_page(shmseg->shm_bsegsz); 274 result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size); 275 if (result != KERN_SUCCESS) 276 return (EINVAL); 277 shmmap_s->shmid = -1; 278 shmseg->u.shm_dtime = time_second; 279 if ((--shmseg->u.shm_nattch <= 0) && 280 (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) { 281 shm_deallocate_segment(shmseg); 282 shm_last_free = segnum; 283 } 284 return (0); 285} 286 287#ifndef _SYS_SYSPROTO_H_ 288struct shmdt_args { 289 const void *shmaddr; 290}; 291#endif 292int 293shmdt(td, uap) 294 struct thread *td; 295 struct shmdt_args *uap; 296{ 297 struct proc *p = td->td_proc; 298 struct shmmap_state *shmmap_s; 299#ifdef MAC 300 struct shmid_kernel *shmsegptr; 301#endif 302 int i; 303 int error = 0; 304 305 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 306 return (ENOSYS); 307 mtx_lock(&Giant); 308 shmmap_s = p->p_vmspace->vm_shm; 309 if (shmmap_s == NULL) { 310 error = EINVAL; 311 goto done2; 312 } 313 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) { 314 if (shmmap_s->shmid != -1 && 315 shmmap_s->va == (vm_offset_t)uap->shmaddr) { 316 break; 317 } 318 } 319 if (i == shminfo.shmseg) { 320 error = EINVAL; 321 goto done2; 322 } 323#ifdef MAC 324 shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)]; 325 error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr); 326 if (error != 0) 327 goto done2; 328#endif 329 error = shm_delete_mapping(p->p_vmspace, shmmap_s); 330done2: 331 mtx_unlock(&Giant); 332 return (error); 333} 334 335#ifndef _SYS_SYSPROTO_H_ 336struct shmat_args { 337 int shmid; 338 const void *shmaddr; 339 int shmflg; 340}; 341#endif 342int 343kern_shmat(td, shmid, shmaddr, shmflg) 344 struct thread *td; 345 int shmid; 346 const void *shmaddr; 347 int shmflg; 348{ 349 struct proc *p = td->td_proc; 350 int i, flags; 351 struct shmid_kernel *shmseg; 352 struct shmmap_state *shmmap_s = NULL; 353 vm_offset_t attach_va; 354 vm_prot_t prot; 355 vm_size_t size; 356 int rv; 357 int error = 0; 358 359 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 360 return (ENOSYS); 361 mtx_lock(&Giant); 362 shmmap_s = p->p_vmspace->vm_shm; 363 if (shmmap_s == NULL) { 364 shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state), 365 M_SHM, M_WAITOK); 366 for (i = 0; i < shminfo.shmseg; i++) 367 shmmap_s[i].shmid = -1; 368 p->p_vmspace->vm_shm = shmmap_s; 369 } 370 shmseg = shm_find_segment_by_shmid(shmid); 371 if (shmseg == NULL) { 372 error = EINVAL; 373 goto done2; 374 } 375 error = ipcperm(td, &shmseg->u.shm_perm, 376 (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W); 377 if (error) 378 goto done2; 379#ifdef MAC 380 error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg); 381 if (error != 0) 382 goto done2; 383#endif 384 for (i = 0; i < shminfo.shmseg; i++) { 385 if (shmmap_s->shmid == -1) 386 break; 387 shmmap_s++; 388 } 389 if (i >= shminfo.shmseg) { 390 error = EMFILE; 391 goto done2; 392 } 393 size = round_page(shmseg->shm_bsegsz); 394 prot = VM_PROT_READ; 395 if ((shmflg & SHM_RDONLY) == 0) 396 prot |= VM_PROT_WRITE; 397 flags = MAP_ANON | MAP_SHARED; 398 if (shmaddr) { 399 flags |= MAP_FIXED; 400 if (shmflg & SHM_RND) { 401 attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1); 402 } else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) { 403 attach_va = (vm_offset_t)shmaddr; 404 } else { 405 error = EINVAL; 406 goto done2; 407 } 408 } else { 409 /* 410 * This is just a hint to vm_map_find() about where to 411 * put it. 412 */ 413 PROC_LOCK(p); 414 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr + 415 lim_max(p, RLIMIT_DATA)); 416 PROC_UNLOCK(p); 417 } 418 419 vm_object_reference(shmseg->u.shm_internal); 420 rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->u.shm_internal, 421 0, &attach_va, size, (flags & MAP_FIXED) ? VMFS_NO_SPACE : 422 VMFS_ANY_SPACE, prot, prot, 0); 423 if (rv != KERN_SUCCESS) { 424 vm_object_deallocate(shmseg->u.shm_internal); 425 error = ENOMEM; 426 goto done2; 427 } 428 vm_map_inherit(&p->p_vmspace->vm_map, 429 attach_va, attach_va + size, VM_INHERIT_SHARE); 430 431 shmmap_s->va = attach_va; 432 shmmap_s->shmid = shmid; 433 shmseg->u.shm_lpid = p->p_pid; 434 shmseg->u.shm_atime = time_second; 435 shmseg->u.shm_nattch++; 436 td->td_retval[0] = attach_va; 437done2: 438 mtx_unlock(&Giant); 439 return (error); 440} 441 442int 443shmat(td, uap) 444 struct thread *td; 445 struct shmat_args *uap; 446{ 447 return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg); 448} 449 450#if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) 451struct oshmid_ds { 452 struct ipc_perm shm_perm; /* operation perms */ 453 int shm_segsz; /* size of segment (bytes) */ 454 u_short shm_cpid; /* pid, creator */ 455 u_short shm_lpid; /* pid, last operation */ 456 short shm_nattch; /* no. of current attaches */ 457 time_t shm_atime; /* last attach time */ 458 time_t shm_dtime; /* last detach time */ 459 time_t shm_ctime; /* last change time */ 460 void *shm_handle; /* internal handle for shm segment */ 461}; 462 463struct oshmctl_args { 464 int shmid; 465 int cmd; 466 struct oshmid_ds *ubuf; 467}; 468static int 469oshmctl(td, uap) 470 struct thread *td; 471 struct oshmctl_args *uap; 472{ 473#ifdef COMPAT_43 474 int error = 0; 475 struct shmid_kernel *shmseg; 476 struct oshmid_ds outbuf; 477 478 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 479 return (ENOSYS); 480 mtx_lock(&Giant); 481 shmseg = shm_find_segment_by_shmid(uap->shmid); 482 if (shmseg == NULL) { 483 error = EINVAL; 484 goto done2; 485 } 486 switch (uap->cmd) { 487 case IPC_STAT: 488 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R); 489 if (error) 490 goto done2; 491#ifdef MAC 492 error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd); 493 if (error != 0) 494 goto done2; 495#endif 496 outbuf.shm_perm = shmseg->u.shm_perm; 497 outbuf.shm_segsz = shmseg->u.shm_segsz; 498 outbuf.shm_cpid = shmseg->u.shm_cpid; 499 outbuf.shm_lpid = shmseg->u.shm_lpid; 500 outbuf.shm_nattch = shmseg->u.shm_nattch; 501 outbuf.shm_atime = shmseg->u.shm_atime; 502 outbuf.shm_dtime = shmseg->u.shm_dtime; 503 outbuf.shm_ctime = shmseg->u.shm_ctime; 504 outbuf.shm_handle = shmseg->u.shm_internal; 505 error = copyout(&outbuf, uap->ubuf, sizeof(outbuf)); 506 if (error) 507 goto done2; 508 break; 509 default: 510 error = shmctl(td, (struct shmctl_args *)uap); 511 break; 512 } 513done2: 514 mtx_unlock(&Giant); 515 return (error); 516#else 517 return (EINVAL); 518#endif 519} 520#endif 521 522int 523kern_shmctl(td, shmid, cmd, buf, bufsz) 524 struct thread *td; 525 int shmid; 526 int cmd; 527 void *buf; 528 size_t *bufsz; 529{ 530 int error = 0; 531 struct shmid_kernel *shmseg; 532 533 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 534 return (ENOSYS); 535 536 mtx_lock(&Giant); 537 switch (cmd) { 538 /* 539 * It is possible that kern_shmctl is being called from the Linux ABI 540 * layer, in which case, we will need to implement IPC_INFO. It should 541 * be noted that other shmctl calls will be funneled through here for 542 * Linix binaries as well. 543 * 544 * NB: The Linux ABI layer will convert this data to structure(s) more 545 * consistent with the Linux ABI. 546 */ 547 case IPC_INFO: 548 memcpy(buf, &shminfo, sizeof(shminfo)); 549 if (bufsz) 550 *bufsz = sizeof(shminfo); 551 td->td_retval[0] = shmalloced; 552 goto done2; 553 case SHM_INFO: { 554 struct shm_info shm_info; 555 shm_info.used_ids = shm_nused; 556 shm_info.shm_rss = 0; /*XXX where to get from ? */ 557 shm_info.shm_tot = 0; /*XXX where to get from ? */ 558 shm_info.shm_swp = 0; /*XXX where to get from ? */ 559 shm_info.swap_attempts = 0; /*XXX where to get from ? */ 560 shm_info.swap_successes = 0; /*XXX where to get from ? */ 561 memcpy(buf, &shm_info, sizeof(shm_info)); 562 if (bufsz) 563 *bufsz = sizeof(shm_info); 564 td->td_retval[0] = shmalloced; 565 goto done2; 566 } 567 } 568 if (cmd == SHM_STAT) 569 shmseg = shm_find_segment_by_shmidx(shmid); 570 else 571 shmseg = shm_find_segment_by_shmid(shmid); 572 if (shmseg == NULL) { 573 error = EINVAL; 574 goto done2; 575 } 576#ifdef MAC 577 error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd); 578 if (error != 0) 579 goto done2; 580#endif 581 switch (cmd) { 582 case SHM_STAT: 583 case IPC_STAT: 584 error = ipcperm(td, &shmseg->u.shm_perm, IPC_R); 585 if (error) 586 goto done2; 587 memcpy(buf, &shmseg->u, sizeof(struct shmid_ds)); 588 if (bufsz) 589 *bufsz = sizeof(struct shmid_ds); 590 if (cmd == SHM_STAT) 591 td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm); 592 break; 593 case IPC_SET: { 594 struct shmid_ds *shmid; 595 596 shmid = (struct shmid_ds *)buf; 597 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M); 598 if (error) 599 goto done2; 600 shmseg->u.shm_perm.uid = shmid->shm_perm.uid; 601 shmseg->u.shm_perm.gid = shmid->shm_perm.gid; 602 shmseg->u.shm_perm.mode = 603 (shmseg->u.shm_perm.mode & ~ACCESSPERMS) | 604 (shmid->shm_perm.mode & ACCESSPERMS); 605 shmseg->u.shm_ctime = time_second; 606 break; 607 } 608 case IPC_RMID: 609 error = ipcperm(td, &shmseg->u.shm_perm, IPC_M); 610 if (error) 611 goto done2; 612 shmseg->u.shm_perm.key = IPC_PRIVATE; 613 shmseg->u.shm_perm.mode |= SHMSEG_REMOVED; 614 if (shmseg->u.shm_nattch <= 0) { 615 shm_deallocate_segment(shmseg); 616 shm_last_free = IPCID_TO_IX(shmid); 617 } 618 break; 619#if 0 620 case SHM_LOCK: 621 case SHM_UNLOCK: 622#endif 623 default: 624 error = EINVAL; 625 break; 626 } 627done2: 628 mtx_unlock(&Giant); 629 return (error); 630} 631 632#ifndef _SYS_SYSPROTO_H_ 633struct shmctl_args { 634 int shmid; 635 int cmd; 636 struct shmid_ds *buf; 637}; 638#endif 639int 640shmctl(td, uap) 641 struct thread *td; 642 struct shmctl_args *uap; 643{ 644 int error = 0; 645 struct shmid_ds buf; 646 size_t bufsz; 647 648 /* 649 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support 650 * Linux binaries. If we see the call come through the FreeBSD ABI, 651 * return an error back to the user since we do not to support this. 652 */ 653 if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO || 654 uap->cmd == SHM_STAT) 655 return (EINVAL); 656 657 /* IPC_SET needs to copyin the buffer before calling kern_shmctl */ 658 if (uap->cmd == IPC_SET) { 659 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds)))) 660 goto done; 661 } 662 663 error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz); 664 if (error) 665 goto done; 666 667 /* Cases in which we need to copyout */ 668 switch (uap->cmd) { 669 case IPC_STAT: 670 error = copyout(&buf, uap->buf, bufsz); 671 break; 672 } 673 674done: 675 if (error) { 676 /* Invalidate the return value */ 677 td->td_retval[0] = -1; 678 } 679 return (error); 680} 681 682 683static int 684shmget_existing(td, uap, mode, segnum) 685 struct thread *td; 686 struct shmget_args *uap; 687 int mode; 688 int segnum; 689{ 690 struct shmid_kernel *shmseg; 691 int error; 692 693 shmseg = &shmsegs[segnum]; 694 if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) { 695 /* 696 * This segment is in the process of being allocated. Wait 697 * until it's done, and look the key up again (in case the 698 * allocation failed or it was freed). 699 */ 700 shmseg->u.shm_perm.mode |= SHMSEG_WANTED; 701 error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0); 702 if (error) 703 return (error); 704 return (EAGAIN); 705 } 706 if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) 707 return (EEXIST); 708#ifdef MAC 709 error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg); 710 if (error != 0) 711 return (error); 712#endif 713 if (uap->size != 0 && uap->size > shmseg->shm_bsegsz) 714 return (EINVAL); 715 td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm); 716 return (0); 717} 718 719static int 720shmget_allocate_segment(td, uap, mode) 721 struct thread *td; 722 struct shmget_args *uap; 723 int mode; 724{ 725 int i, segnum, shmid; 726 size_t size; 727 struct ucred *cred = td->td_ucred; 728 struct shmid_kernel *shmseg; 729 vm_object_t shm_object; 730 731 GIANT_REQUIRED; 732 733 if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax) 734 return (EINVAL); 735 if (shm_nused >= shminfo.shmmni) /* Any shmids left? */ 736 return (ENOSPC); 737 size = round_page(uap->size); 738 if (shm_committed + btoc(size) > shminfo.shmall) 739 return (ENOMEM); 740 if (shm_last_free < 0) { 741 shmrealloc(); /* Maybe expand the shmsegs[] array. */ 742 for (i = 0; i < shmalloced; i++) 743 if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE) 744 break; 745 if (i == shmalloced) 746 return (ENOSPC); 747 segnum = i; 748 } else { 749 segnum = shm_last_free; 750 shm_last_free = -1; 751 } 752 shmseg = &shmsegs[segnum]; 753 /* 754 * In case we sleep in malloc(), mark the segment present but deleted 755 * so that noone else tries to create the same key. 756 */ 757 shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED; 758 shmseg->u.shm_perm.key = uap->key; 759 shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff; 760 shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm); 761 762 /* 763 * We make sure that we have allocated a pager before we need 764 * to. 765 */ 766 shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP, 767 0, size, VM_PROT_DEFAULT, 0, cred); 768 if (shm_object == NULL) 769 return (ENOMEM); 770 VM_OBJECT_LOCK(shm_object); 771 vm_object_clear_flag(shm_object, OBJ_ONEMAPPING); 772 vm_object_set_flag(shm_object, OBJ_NOSPLIT); 773 VM_OBJECT_UNLOCK(shm_object); 774 775 shmseg->u.shm_internal = shm_object; 776 shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid; 777 shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid; 778 shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) | 779 (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; 780 shmseg->u.shm_segsz = uap->size; 781 shmseg->shm_bsegsz = uap->size; 782 shmseg->u.shm_cpid = td->td_proc->p_pid; 783 shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0; 784 shmseg->u.shm_atime = shmseg->u.shm_dtime = 0; 785#ifdef MAC 786 mac_sysvshm_create(cred, shmseg); 787#endif 788 shmseg->u.shm_ctime = time_second; 789 shm_committed += btoc(size); 790 shm_nused++; 791 if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) { 792 /* 793 * Somebody else wanted this key while we were asleep. Wake 794 * them up now. 795 */ 796 shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED; 797 wakeup(shmseg); 798 } 799 td->td_retval[0] = shmid; 800 return (0); 801} 802 803#ifndef _SYS_SYSPROTO_H_ 804struct shmget_args { 805 key_t key; 806 size_t size; 807 int shmflg; 808}; 809#endif 810int 811shmget(td, uap) 812 struct thread *td; 813 struct shmget_args *uap; 814{ 815 int segnum, mode; 816 int error; 817 818 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 819 return (ENOSYS); 820 mtx_lock(&Giant); 821 mode = uap->shmflg & ACCESSPERMS; 822 if (uap->key != IPC_PRIVATE) { 823 again: 824 segnum = shm_find_segment_by_key(uap->key); 825 if (segnum >= 0) { 826 error = shmget_existing(td, uap, mode, segnum); 827 if (error == EAGAIN) 828 goto again; 829 goto done2; 830 } 831 if ((uap->shmflg & IPC_CREAT) == 0) { 832 error = ENOENT; 833 goto done2; 834 } 835 } 836 error = shmget_allocate_segment(td, uap, mode); 837done2: 838 mtx_unlock(&Giant); 839 return (error); 840} 841 842int 843shmsys(td, uap) 844 struct thread *td; 845 /* XXX actually varargs. */ 846 struct shmsys_args /* { 847 int which; 848 int a2; 849 int a3; 850 int a4; 851 } */ *uap; 852{ 853#if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) 854 int error; 855 856 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 857 return (ENOSYS); 858 if (uap->which < 0 || 859 uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0])) 860 return (EINVAL); 861 mtx_lock(&Giant); 862 error = (*shmcalls[uap->which])(td, &uap->a2); 863 mtx_unlock(&Giant); 864 return (error); 865#else 866 return (nosys(td, NULL)); 867#endif 868} 869 870static void 871shmfork_myhook(p1, p2) 872 struct proc *p1, *p2; 873{ 874 struct shmmap_state *shmmap_s; 875 size_t size; 876 int i; 877 878 mtx_lock(&Giant); 879 size = shminfo.shmseg * sizeof(struct shmmap_state); 880 shmmap_s = malloc(size, M_SHM, M_WAITOK); 881 bcopy(p1->p_vmspace->vm_shm, shmmap_s, size); 882 p2->p_vmspace->vm_shm = shmmap_s; 883 for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) 884 if (shmmap_s->shmid != -1) 885 shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++; 886 mtx_unlock(&Giant); 887} 888 889static void 890shmexit_myhook(struct vmspace *vm) 891{ 892 struct shmmap_state *base, *shm; 893 int i; 894 895 if ((base = vm->vm_shm) != NULL) { 896 vm->vm_shm = NULL; 897 mtx_lock(&Giant); 898 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) { 899 if (shm->shmid != -1) 900 shm_delete_mapping(vm, shm); 901 } 902 mtx_unlock(&Giant); 903 free(base, M_SHM); 904 } 905} 906 907static void 908shmrealloc(void) 909{ 910 int i; 911 struct shmid_kernel *newsegs; 912 913 if (shmalloced >= shminfo.shmmni) 914 return; 915 916 newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK); 917 if (newsegs == NULL) 918 return; 919 for (i = 0; i < shmalloced; i++) 920 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0])); 921 for (; i < shminfo.shmmni; i++) { 922 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE; 923 shmsegs[i].u.shm_perm.seq = 0; 924#ifdef MAC 925 mac_sysvshm_init(&shmsegs[i]); 926#endif 927 } 928 free(shmsegs, M_SHM); 929 shmsegs = newsegs; 930 shmalloced = shminfo.shmmni; 931} 932 933static void 934shminit() 935{ 936 int i; 937 938 TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall); 939 for (i = PAGE_SIZE; i > 0; i--) { 940 shminfo.shmmax = shminfo.shmall * i; 941 if (shminfo.shmmax >= shminfo.shmall) 942 break; 943 } 944 TUNABLE_ULONG_FETCH("kern.ipc.shmmin", &shminfo.shmmin); 945 TUNABLE_ULONG_FETCH("kern.ipc.shmmni", &shminfo.shmmni); 946 TUNABLE_ULONG_FETCH("kern.ipc.shmseg", &shminfo.shmseg); 947 TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys); 948 949 shmalloced = shminfo.shmmni; 950 shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK); 951 if (shmsegs == NULL) 952 panic("cannot allocate initial memory for sysvshm"); 953 for (i = 0; i < shmalloced; i++) { 954 shmsegs[i].u.shm_perm.mode = SHMSEG_FREE; 955 shmsegs[i].u.shm_perm.seq = 0; 956#ifdef MAC 957 mac_sysvshm_init(&shmsegs[i]); 958#endif 959 } 960 shm_last_free = 0; 961 shm_nused = 0; 962 shm_committed = 0; 963 shmexit_hook = &shmexit_myhook; 964 shmfork_hook = &shmfork_myhook; 965} 966 967static int 968shmunload() 969{ 970#ifdef MAC 971 int i; 972#endif 973 974 if (shm_nused > 0) 975 return (EBUSY); 976 977#ifdef MAC 978 for (i = 0; i < shmalloced; i++) 979 mac_sysvshm_destroy(&shmsegs[i]); 980#endif 981 free(shmsegs, M_SHM); 982 shmexit_hook = NULL; 983 shmfork_hook = NULL; 984 return (0); 985} 986 987static int 988sysctl_shmsegs(SYSCTL_HANDLER_ARGS) 989{ 990 991 return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0]))); 992} 993 994static int 995sysvshm_modload(struct module *module, int cmd, void *arg) 996{ 997 int error = 0; 998 999 switch (cmd) { 1000 case MOD_LOAD: 1001 shminit(); 1002 break; 1003 case MOD_UNLOAD: 1004 error = shmunload(); 1005 break; 1006 case MOD_SHUTDOWN: 1007 break; 1008 default: 1009 error = EINVAL; 1010 break; 1011 } 1012 return (error); 1013} 1014 1015static moduledata_t sysvshm_mod = { 1016 "sysvshm", 1017 &sysvshm_modload, 1018 NULL 1019}; 1020 1021SYSCALL_MODULE_HELPER(shmsys); 1022SYSCALL_MODULE_HELPER(shmat); 1023SYSCALL_MODULE_HELPER(shmctl); 1024SYSCALL_MODULE_HELPER(shmdt); 1025SYSCALL_MODULE_HELPER(shmget); 1026 1027DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST); 1028MODULE_VERSION(sysvshm, 1); 1029