sysv_msg.c revision 140614
1/*- 2 * Implementation of SVID messages 3 * 4 * Author: Daniel Boulet 5 * 6 * Copyright 1993 Daniel Boulet and RTMX Inc. 7 * 8 * This system call was implemented by Daniel Boulet under contract from RTMX. 9 * 10 * Redistribution and use in source forms, with and without modification, 11 * are permitted provided that this entire comment appears intact. 12 * 13 * Redistribution in binary form may occur without any restrictions. 14 * Obviously, it would be nice if you gave credit where credit is due 15 * but requiring it would be too onerous. 16 * 17 * This software is provided ``AS IS'' without any warranties of any kind. 18 */ 19/*- 20 * Copyright (c) 2003-2005 McAfee, Inc. 21 * All rights reserved. 22 * 23 * This software was developed for the FreeBSD Project in part by McAfee 24 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR 25 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research 26 * program. 27 * 28 * Redistribution and use in source and binary forms, with or without 29 * modification, are permitted provided that the following conditions 30 * are met: 31 * 1. Redistributions of source code must retain the above copyright 32 * notice, this list of conditions and the following disclaimer. 33 * 2. Redistributions in binary form must reproduce the above copyright 34 * notice, this list of conditions and the following disclaimer in the 35 * documentation and/or other materials provided with the distribution. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50#include <sys/cdefs.h> 51__FBSDID("$FreeBSD: head/sys/kern/sysv_msg.c 140614 2005-01-22 18:51:43Z rwatson $"); 52 53#include "opt_sysvipc.h" 54#include "opt_mac.h" 55 56#include <sys/param.h> 57#include <sys/systm.h> 58#include <sys/sysproto.h> 59#include <sys/kernel.h> 60#include <sys/proc.h> 61#include <sys/lock.h> 62#include <sys/mac.h> 63#include <sys/mutex.h> 64#include <sys/module.h> 65#include <sys/msg.h> 66#include <sys/syscall.h> 67#include <sys/sysent.h> 68#include <sys/sysctl.h> 69#include <sys/malloc.h> 70#include <sys/jail.h> 71 72static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 73 74static void msginit(void); 75static int msgunload(void); 76static int sysvmsg_modload(struct module *, int, void *); 77 78#ifdef MSG_DEBUG 79#define DPRINTF(a) printf a 80#else 81#define DPRINTF(a) 82#endif 83#ifdef MAC_DEBUG 84#define MPRINTF(a) printf a 85#else 86#define MPRINTF(a) 87#endif 88 89static void msg_freehdr(struct msg *msghdr); 90 91/* XXX casting to (sy_call_t *) is bogus, as usual. */ 92static sy_call_t *msgcalls[] = { 93 (sy_call_t *)msgctl, (sy_call_t *)msgget, 94 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 95}; 96 97#ifndef MSGSSZ 98#define MSGSSZ 8 /* Each segment must be 2^N long */ 99#endif 100#ifndef MSGSEG 101#define MSGSEG 2048 /* must be less than 32767 */ 102#endif 103#define MSGMAX (MSGSSZ*MSGSEG) 104#ifndef MSGMNB 105#define MSGMNB 2048 /* max # of bytes in a queue */ 106#endif 107#ifndef MSGMNI 108#define MSGMNI 40 109#endif 110#ifndef MSGTQL 111#define MSGTQL 40 112#endif 113 114/* 115 * Based on the configuration parameters described in an SVR2 (yes, two) 116 * config(1m) man page. 117 * 118 * Each message is broken up and stored in segments that are msgssz bytes 119 * long. For efficiency reasons, this should be a power of two. Also, 120 * it doesn't make sense if it is less than 8 or greater than about 256. 121 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 122 * two between 8 and 1024 inclusive (and panic's if it isn't). 123 */ 124struct msginfo msginfo = { 125 MSGMAX, /* max chars in a message */ 126 MSGMNI, /* # of message queue identifiers */ 127 MSGMNB, /* max chars in a queue */ 128 MSGTQL, /* max messages in system */ 129 MSGSSZ, /* size of a message segment */ 130 /* (must be small power of 2 greater than 4) */ 131 MSGSEG /* number of message segments */ 132}; 133 134/* 135 * macros to convert between msqid_ds's and msqid's. 136 * (specific to this implementation) 137 */ 138#define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 139#define MSQID_IX(id) ((id) & 0xffff) 140#define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 141 142/* 143 * The rest of this file is specific to this particular implementation. 144 */ 145 146struct msgmap { 147 short next; /* next segment in buffer */ 148 /* -1 -> available */ 149 /* 0..(MSGSEG-1) -> index of next segment */ 150}; 151 152#define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 153 154static int nfree_msgmaps; /* # of free map entries */ 155static short free_msgmaps; /* head of linked list of free map entries */ 156static struct msg *free_msghdrs;/* list of free msg headers */ 157static char *msgpool; /* MSGMAX byte long msg buffer pool */ 158static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 159static struct msg *msghdrs; /* MSGTQL msg headers */ 160static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 161static struct mtx msq_mtx; /* global mutex for message queues. */ 162 163static void 164msginit() 165{ 166 register int i; 167 168 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 169 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 170 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 171 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 172 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 173 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 174 175 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 176 if (msgpool == NULL) 177 panic("msgpool is NULL"); 178 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 179 if (msgmaps == NULL) 180 panic("msgmaps is NULL"); 181 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 182 if (msghdrs == NULL) 183 panic("msghdrs is NULL"); 184 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 185 M_WAITOK); 186 if (msqids == NULL) 187 panic("msqids is NULL"); 188 189 /* 190 * msginfo.msgssz should be a power of two for efficiency reasons. 191 * It is also pretty silly if msginfo.msgssz is less than 8 192 * or greater than about 256 so ... 193 */ 194 195 i = 8; 196 while (i < 1024 && i != msginfo.msgssz) 197 i <<= 1; 198 if (i != msginfo.msgssz) { 199 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 200 msginfo.msgssz)); 201 panic("msginfo.msgssz not a small power of 2"); 202 } 203 204 if (msginfo.msgseg > 32767) { 205 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 206 panic("msginfo.msgseg > 32767"); 207 } 208 209 if (msgmaps == NULL) 210 panic("msgmaps is NULL"); 211 212 for (i = 0; i < msginfo.msgseg; i++) { 213 if (i > 0) 214 msgmaps[i-1].next = i; 215 msgmaps[i].next = -1; /* implies entry is available */ 216 } 217 free_msgmaps = 0; 218 nfree_msgmaps = msginfo.msgseg; 219 220 if (msghdrs == NULL) 221 panic("msghdrs is NULL"); 222 223 for (i = 0; i < msginfo.msgtql; i++) { 224 msghdrs[i].msg_type = 0; 225 if (i > 0) 226 msghdrs[i-1].msg_next = &msghdrs[i]; 227 msghdrs[i].msg_next = NULL; 228#ifdef MAC 229 mac_init_sysv_msgmsg(&msghdrs[i]); 230#endif 231 } 232 free_msghdrs = &msghdrs[0]; 233 234 if (msqids == NULL) 235 panic("msqids is NULL"); 236 237 for (i = 0; i < msginfo.msgmni; i++) { 238 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 239 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 240 msqids[i].u.msg_perm.mode = 0; 241#ifdef MAC 242 mac_init_sysv_msgqueue(&msqids[i]); 243#endif 244 } 245 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 246} 247 248static int 249msgunload() 250{ 251 struct msqid_kernel *msqkptr; 252 int msqid; 253#ifdef MAC 254 int i; 255#endif 256 257 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 258 /* 259 * Look for an unallocated and unlocked msqid_ds. 260 * msqid_ds's can be locked by msgsnd or msgrcv while 261 * they are copying the message in/out. We can't 262 * re-use the entry until they release it. 263 */ 264 msqkptr = &msqids[msqid]; 265 if (msqkptr->u.msg_qbytes != 0 || 266 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 267 break; 268 } 269 if (msqid != msginfo.msgmni) 270 return (EBUSY); 271 272#ifdef MAC 273 for (i = 0; i < msginfo.msgtql; i++) 274 mac_destroy_sysv_msgmsg(&msghdrs[i]); 275 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 276 mac_destroy_sysv_msgqueue(&msqids[msqid]); 277#endif 278 free(msgpool, M_MSG); 279 free(msgmaps, M_MSG); 280 free(msghdrs, M_MSG); 281 free(msqids, M_MSG); 282 mtx_destroy(&msq_mtx); 283 return (0); 284} 285 286 287static int 288sysvmsg_modload(struct module *module, int cmd, void *arg) 289{ 290 int error = 0; 291 292 switch (cmd) { 293 case MOD_LOAD: 294 msginit(); 295 break; 296 case MOD_UNLOAD: 297 error = msgunload(); 298 break; 299 case MOD_SHUTDOWN: 300 break; 301 default: 302 error = EINVAL; 303 break; 304 } 305 return (error); 306} 307 308static moduledata_t sysvmsg_mod = { 309 "sysvmsg", 310 &sysvmsg_modload, 311 NULL 312}; 313 314SYSCALL_MODULE_HELPER(msgsys); 315SYSCALL_MODULE_HELPER(msgctl); 316SYSCALL_MODULE_HELPER(msgget); 317SYSCALL_MODULE_HELPER(msgsnd); 318SYSCALL_MODULE_HELPER(msgrcv); 319 320DECLARE_MODULE(sysvmsg, sysvmsg_mod, 321 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 322MODULE_VERSION(sysvmsg, 1); 323 324/* 325 * Entry point for all MSG calls 326 * 327 * MPSAFE 328 */ 329int 330msgsys(td, uap) 331 struct thread *td; 332 /* XXX actually varargs. */ 333 struct msgsys_args /* { 334 int which; 335 int a2; 336 int a3; 337 int a4; 338 int a5; 339 int a6; 340 } */ *uap; 341{ 342 int error; 343 344 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 345 return (ENOSYS); 346 if (uap->which < 0 || 347 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 348 return (EINVAL); 349 error = (*msgcalls[uap->which])(td, &uap->a2); 350 return (error); 351} 352 353static void 354msg_freehdr(msghdr) 355 struct msg *msghdr; 356{ 357 while (msghdr->msg_ts > 0) { 358 short next; 359 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 360 panic("msghdr->msg_spot out of range"); 361 next = msgmaps[msghdr->msg_spot].next; 362 msgmaps[msghdr->msg_spot].next = free_msgmaps; 363 free_msgmaps = msghdr->msg_spot; 364 nfree_msgmaps++; 365 msghdr->msg_spot = next; 366 if (msghdr->msg_ts >= msginfo.msgssz) 367 msghdr->msg_ts -= msginfo.msgssz; 368 else 369 msghdr->msg_ts = 0; 370 } 371 if (msghdr->msg_spot != -1) 372 panic("msghdr->msg_spot != -1"); 373 msghdr->msg_next = free_msghdrs; 374 free_msghdrs = msghdr; 375#ifdef MAC 376 mac_cleanup_sysv_msgmsg(msghdr); 377#endif 378} 379 380#ifndef _SYS_SYSPROTO_H_ 381struct msgctl_args { 382 int msqid; 383 int cmd; 384 struct msqid_ds *buf; 385}; 386#endif 387 388/* 389 * MPSAFE 390 */ 391int 392msgctl(td, uap) 393 struct thread *td; 394 register struct msgctl_args *uap; 395{ 396 int msqid = uap->msqid; 397 int cmd = uap->cmd; 398 struct msqid_ds *user_msqptr = uap->buf; 399 int rval, error; 400 struct msqid_ds msqbuf; 401 register struct msqid_kernel *msqkptr; 402 403 DPRINTF(("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr)); 404 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 405 return (ENOSYS); 406 407 msqid = IPCID_TO_IX(msqid); 408 409 if (msqid < 0 || msqid >= msginfo.msgmni) { 410 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 411 msginfo.msgmni)); 412 return (EINVAL); 413 } 414 if (cmd == IPC_SET && 415 (error = copyin(user_msqptr, &msqbuf, sizeof(msqbuf))) != 0) 416 return (error); 417 418 msqkptr = &msqids[msqid]; 419 420 mtx_lock(&msq_mtx); 421 if (msqkptr->u.msg_qbytes == 0) { 422 DPRINTF(("no such msqid\n")); 423 error = EINVAL; 424 goto done2; 425 } 426 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 427 DPRINTF(("wrong sequence number\n")); 428 error = EINVAL; 429 goto done2; 430 } 431#ifdef MAC 432 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd); 433 if (error != 0) { 434 MPRINTF(("mac_check_sysv_msqctl returned %d\n", error)); 435 goto done2; 436 } 437#endif 438 439 error = 0; 440 rval = 0; 441 442 switch (cmd) { 443 444 case IPC_RMID: 445 { 446 struct msg *msghdr; 447 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 448 goto done2; 449 450#ifdef MAC 451 /* 452 * Check that the thread has MAC access permissions to 453 * individual msghdrs. Note: We need to do this in a 454 * separate loop because the actual loop alters the 455 * msq/msghdr info as it progresses, and there is no going 456 * back if half the way through we discover that the 457 * thread cannot free a certain msghdr. The msq will get 458 * into an inconsistent state. 459 */ 460 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 461 msghdr = msghdr->msg_next) { 462 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr); 463 if (error != 0) { 464 MPRINTF(("mac_check_sysv_msgrmid returned %d\n", 465 error)); 466 goto done2; 467 } 468 } 469#endif 470 471 /* Free the message headers */ 472 msghdr = msqkptr->u.msg_first; 473 while (msghdr != NULL) { 474 struct msg *msghdr_tmp; 475 476 /* Free the segments of each message */ 477 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 478 msqkptr->u.msg_qnum--; 479 msghdr_tmp = msghdr; 480 msghdr = msghdr->msg_next; 481 msg_freehdr(msghdr_tmp); 482 } 483 484 if (msqkptr->u.msg_cbytes != 0) 485 panic("msg_cbytes is screwed up"); 486 if (msqkptr->u.msg_qnum != 0) 487 panic("msg_qnum is screwed up"); 488 489 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 490 491#ifdef MAC 492 mac_cleanup_sysv_msgqueue(msqkptr); 493#endif 494 495 wakeup(msqkptr); 496 } 497 498 break; 499 500 case IPC_SET: 501 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 502 goto done2; 503 if (msqbuf.msg_qbytes > msqkptr->u.msg_qbytes) { 504 error = suser(td); 505 if (error) 506 goto done2; 507 } 508 if (msqbuf.msg_qbytes > msginfo.msgmnb) { 509 DPRINTF(("can't increase msg_qbytes beyond %d" 510 "(truncating)\n", msginfo.msgmnb)); 511 msqbuf.msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 512 } 513 if (msqbuf.msg_qbytes == 0) { 514 DPRINTF(("can't reduce msg_qbytes to 0\n")); 515 error = EINVAL; /* non-standard errno! */ 516 goto done2; 517 } 518 msqkptr->u.msg_perm.uid = msqbuf.msg_perm.uid; /* change the owner */ 519 msqkptr->u.msg_perm.gid = msqbuf.msg_perm.gid; /* change the owner */ 520 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 521 (msqbuf.msg_perm.mode & 0777); 522 msqkptr->u.msg_qbytes = msqbuf.msg_qbytes; 523 msqkptr->u.msg_ctime = time_second; 524 break; 525 526 case IPC_STAT: 527 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 528 DPRINTF(("requester doesn't have read access\n")); 529 goto done2; 530 } 531 break; 532 533 default: 534 DPRINTF(("invalid command %d\n", cmd)); 535 error = EINVAL; 536 goto done2; 537 } 538 539 if (error == 0) 540 td->td_retval[0] = rval; 541done2: 542 mtx_unlock(&msq_mtx); 543 if (cmd == IPC_STAT && error == 0) 544 error = copyout(&(msqkptr->u), user_msqptr, sizeof(struct msqid_ds)); 545 return(error); 546} 547 548#ifndef _SYS_SYSPROTO_H_ 549struct msgget_args { 550 key_t key; 551 int msgflg; 552}; 553#endif 554 555/* 556 * MPSAFE 557 */ 558int 559msgget(td, uap) 560 struct thread *td; 561 register struct msgget_args *uap; 562{ 563 int msqid, error = 0; 564 int key = uap->key; 565 int msgflg = uap->msgflg; 566 struct ucred *cred = td->td_ucred; 567 register struct msqid_kernel *msqkptr = NULL; 568 569 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 570 571 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 572 return (ENOSYS); 573 574 mtx_lock(&msq_mtx); 575 if (key != IPC_PRIVATE) { 576 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 577 msqkptr = &msqids[msqid]; 578 if (msqkptr->u.msg_qbytes != 0 && 579 msqkptr->u.msg_perm.key == key) 580 break; 581 } 582 if (msqid < msginfo.msgmni) { 583 DPRINTF(("found public key\n")); 584 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 585 DPRINTF(("not exclusive\n")); 586 error = EEXIST; 587 goto done2; 588 } 589 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 590 msgflg & 0700))) { 591 DPRINTF(("requester doesn't have 0%o access\n", 592 msgflg & 0700)); 593 goto done2; 594 } 595#ifdef MAC 596 error = mac_check_sysv_msqget(cred, msqkptr); 597 if (error != 0) { 598 MPRINTF(("mac_check_sysv_msqget returned %d\n", 599 error)); 600 goto done2; 601 } 602#endif 603 goto found; 604 } 605 } 606 607 DPRINTF(("need to allocate the msqid_ds\n")); 608 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 609 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 610 /* 611 * Look for an unallocated and unlocked msqid_ds. 612 * msqid_ds's can be locked by msgsnd or msgrcv while 613 * they are copying the message in/out. We can't 614 * re-use the entry until they release it. 615 */ 616 msqkptr = &msqids[msqid]; 617 if (msqkptr->u.msg_qbytes == 0 && 618 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 619 break; 620 } 621 if (msqid == msginfo.msgmni) { 622 DPRINTF(("no more msqid_ds's available\n")); 623 error = ENOSPC; 624 goto done2; 625 } 626 DPRINTF(("msqid %d is available\n", msqid)); 627 msqkptr->u.msg_perm.key = key; 628 msqkptr->u.msg_perm.cuid = cred->cr_uid; 629 msqkptr->u.msg_perm.uid = cred->cr_uid; 630 msqkptr->u.msg_perm.cgid = cred->cr_gid; 631 msqkptr->u.msg_perm.gid = cred->cr_gid; 632 msqkptr->u.msg_perm.mode = (msgflg & 0777); 633 /* Make sure that the returned msqid is unique */ 634 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 635 msqkptr->u.msg_first = NULL; 636 msqkptr->u.msg_last = NULL; 637 msqkptr->u.msg_cbytes = 0; 638 msqkptr->u.msg_qnum = 0; 639 msqkptr->u.msg_qbytes = msginfo.msgmnb; 640 msqkptr->u.msg_lspid = 0; 641 msqkptr->u.msg_lrpid = 0; 642 msqkptr->u.msg_stime = 0; 643 msqkptr->u.msg_rtime = 0; 644 msqkptr->u.msg_ctime = time_second; 645#ifdef MAC 646 mac_create_sysv_msgqueue(cred, msqkptr); 647#endif 648 } else { 649 DPRINTF(("didn't find it and wasn't asked to create it\n")); 650 error = ENOENT; 651 goto done2; 652 } 653 654found: 655 /* Construct the unique msqid */ 656 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 657done2: 658 mtx_unlock(&msq_mtx); 659 return (error); 660} 661 662#ifndef _SYS_SYSPROTO_H_ 663struct msgsnd_args { 664 int msqid; 665 const void *msgp; 666 size_t msgsz; 667 int msgflg; 668}; 669#endif 670 671/* 672 * MPSAFE 673 */ 674int 675msgsnd(td, uap) 676 struct thread *td; 677 register struct msgsnd_args *uap; 678{ 679 int msqid = uap->msqid; 680 const void *user_msgp = uap->msgp; 681 size_t msgsz = uap->msgsz; 682 int msgflg = uap->msgflg; 683 int segs_needed, error = 0; 684 register struct msqid_kernel *msqkptr; 685 register struct msg *msghdr; 686 short next; 687 688 DPRINTF(("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz, 689 msgflg)); 690 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 691 return (ENOSYS); 692 693 mtx_lock(&msq_mtx); 694 msqid = IPCID_TO_IX(msqid); 695 696 if (msqid < 0 || msqid >= msginfo.msgmni) { 697 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 698 msginfo.msgmni)); 699 error = EINVAL; 700 goto done2; 701 } 702 703 msqkptr = &msqids[msqid]; 704 if (msqkptr->u.msg_qbytes == 0) { 705 DPRINTF(("no such message queue id\n")); 706 error = EINVAL; 707 goto done2; 708 } 709 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 710 DPRINTF(("wrong sequence number\n")); 711 error = EINVAL; 712 goto done2; 713 } 714 715 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 716 DPRINTF(("requester doesn't have write access\n")); 717 goto done2; 718 } 719 720#ifdef MAC 721 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr); 722 if (error != 0) { 723 MPRINTF(("mac_check_sysv_msqsnd returned %d\n", error)); 724 goto done2; 725 } 726#endif 727 728 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 729 DPRINTF(("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz, 730 segs_needed)); 731 for (;;) { 732 int need_more_resources = 0; 733 734 /* 735 * check msgsz 736 * (inside this loop in case msg_qbytes changes while we sleep) 737 */ 738 739 if (msgsz > msqkptr->u.msg_qbytes) { 740 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 741 error = EINVAL; 742 goto done2; 743 } 744 745 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 746 DPRINTF(("msqid is locked\n")); 747 need_more_resources = 1; 748 } 749 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 750 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 751 need_more_resources = 1; 752 } 753 if (segs_needed > nfree_msgmaps) { 754 DPRINTF(("segs_needed > nfree_msgmaps\n")); 755 need_more_resources = 1; 756 } 757 if (free_msghdrs == NULL) { 758 DPRINTF(("no more msghdrs\n")); 759 need_more_resources = 1; 760 } 761 762 if (need_more_resources) { 763 int we_own_it; 764 765 if ((msgflg & IPC_NOWAIT) != 0) { 766 DPRINTF(("need more resources but caller " 767 "doesn't want to wait\n")); 768 error = EAGAIN; 769 goto done2; 770 } 771 772 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 773 DPRINTF(("we don't own the msqid_ds\n")); 774 we_own_it = 0; 775 } else { 776 /* Force later arrivals to wait for our 777 request */ 778 DPRINTF(("we own the msqid_ds\n")); 779 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 780 we_own_it = 1; 781 } 782 DPRINTF(("goodnight\n")); 783 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 784 "msgwait", 0); 785 DPRINTF(("good morning, error=%d\n", error)); 786 if (we_own_it) 787 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 788 if (error != 0) { 789 DPRINTF(("msgsnd: interrupted system call\n")); 790 error = EINTR; 791 goto done2; 792 } 793 794 /* 795 * Make sure that the msq queue still exists 796 */ 797 798 if (msqkptr->u.msg_qbytes == 0) { 799 DPRINTF(("msqid deleted\n")); 800 error = EIDRM; 801 goto done2; 802 } 803 804 } else { 805 DPRINTF(("got all the resources that we need\n")); 806 break; 807 } 808 } 809 810 /* 811 * We have the resources that we need. 812 * Make sure! 813 */ 814 815 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 816 panic("msg_perm.mode & MSG_LOCKED"); 817 if (segs_needed > nfree_msgmaps) 818 panic("segs_needed > nfree_msgmaps"); 819 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 820 panic("msgsz + msg_cbytes > msg_qbytes"); 821 if (free_msghdrs == NULL) 822 panic("no more msghdrs"); 823 824 /* 825 * Re-lock the msqid_ds in case we page-fault when copying in the 826 * message 827 */ 828 829 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 830 panic("msqid_ds is already locked"); 831 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 832 833 /* 834 * Allocate a message header 835 */ 836 837 msghdr = free_msghdrs; 838 free_msghdrs = msghdr->msg_next; 839 msghdr->msg_spot = -1; 840 msghdr->msg_ts = msgsz; 841#ifdef MAC 842 /* 843 * XXXMAC: Should the mac_check_sysv_msgmsq check follow here 844 * immediately? Or, should it be checked just before the msg is 845 * enqueued in the msgq (as it is done now)? 846 */ 847 mac_create_sysv_msgmsg(td->td_ucred, msqkptr, msghdr); 848#endif 849 850 /* 851 * Allocate space for the message 852 */ 853 854 while (segs_needed > 0) { 855 if (nfree_msgmaps <= 0) 856 panic("not enough msgmaps"); 857 if (free_msgmaps == -1) 858 panic("nil free_msgmaps"); 859 next = free_msgmaps; 860 if (next <= -1) 861 panic("next too low #1"); 862 if (next >= msginfo.msgseg) 863 panic("next out of range #1"); 864 DPRINTF(("allocating segment %d to message\n", next)); 865 free_msgmaps = msgmaps[next].next; 866 nfree_msgmaps--; 867 msgmaps[next].next = msghdr->msg_spot; 868 msghdr->msg_spot = next; 869 segs_needed--; 870 } 871 872 /* 873 * Copy in the message type 874 */ 875 876 mtx_unlock(&msq_mtx); 877 if ((error = copyin(user_msgp, &msghdr->msg_type, 878 sizeof(msghdr->msg_type))) != 0) { 879 mtx_lock(&msq_mtx); 880 DPRINTF(("error %d copying the message type\n", error)); 881 msg_freehdr(msghdr); 882 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 883 wakeup(msqkptr); 884 goto done2; 885 } 886 mtx_lock(&msq_mtx); 887 user_msgp = (const char *)user_msgp + sizeof(msghdr->msg_type); 888 889 /* 890 * Validate the message type 891 */ 892 893 if (msghdr->msg_type < 1) { 894 msg_freehdr(msghdr); 895 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 896 wakeup(msqkptr); 897 DPRINTF(("mtype (%d) < 1\n", msghdr->msg_type)); 898 error = EINVAL; 899 goto done2; 900 } 901 902 /* 903 * Copy in the message body 904 */ 905 906 next = msghdr->msg_spot; 907 while (msgsz > 0) { 908 size_t tlen; 909 if (msgsz > msginfo.msgssz) 910 tlen = msginfo.msgssz; 911 else 912 tlen = msgsz; 913 if (next <= -1) 914 panic("next too low #2"); 915 if (next >= msginfo.msgseg) 916 panic("next out of range #2"); 917 mtx_unlock(&msq_mtx); 918 if ((error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], 919 tlen)) != 0) { 920 mtx_lock(&msq_mtx); 921 DPRINTF(("error %d copying in message segment\n", 922 error)); 923 msg_freehdr(msghdr); 924 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 925 wakeup(msqkptr); 926 goto done2; 927 } 928 mtx_lock(&msq_mtx); 929 msgsz -= tlen; 930 user_msgp = (const char *)user_msgp + tlen; 931 next = msgmaps[next].next; 932 } 933 if (next != -1) 934 panic("didn't use all the msg segments"); 935 936 /* 937 * We've got the message. Unlock the msqid_ds. 938 */ 939 940 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 941 942 /* 943 * Make sure that the msqid_ds is still allocated. 944 */ 945 946 if (msqkptr->u.msg_qbytes == 0) { 947 msg_freehdr(msghdr); 948 wakeup(msqkptr); 949 error = EIDRM; 950 goto done2; 951 } 952 953#ifdef MAC 954 /* 955 * Note: Since the task/thread allocates the msghdr and usually 956 * primes it with its own MAC label, for a majority of policies, it 957 * won't be necessary to check whether the msghdr has access 958 * permissions to the msgq. The mac_check_sysv_msqsnd check would 959 * suffice in that case. However, this hook may be required where 960 * individual policies derive a non-identical label for the msghdr 961 * from the current thread label and may want to check the msghdr 962 * enqueue permissions, along with read/write permissions to the 963 * msgq. 964 */ 965 error = mac_check_sysv_msgmsq(td->td_ucred, msghdr, msqkptr); 966 if (error != 0) { 967 MPRINTF(("mac_check_sysv_msqmsq returned %d\n", error)); 968 msg_freehdr(msghdr); 969 wakeup(msqkptr); 970 goto done2; 971 } 972#endif 973 974 /* 975 * Put the message into the queue 976 */ 977 if (msqkptr->u.msg_first == NULL) { 978 msqkptr->u.msg_first = msghdr; 979 msqkptr->u.msg_last = msghdr; 980 } else { 981 msqkptr->u.msg_last->msg_next = msghdr; 982 msqkptr->u.msg_last = msghdr; 983 } 984 msqkptr->u.msg_last->msg_next = NULL; 985 986 msqkptr->u.msg_cbytes += msghdr->msg_ts; 987 msqkptr->u.msg_qnum++; 988 msqkptr->u.msg_lspid = td->td_proc->p_pid; 989 msqkptr->u.msg_stime = time_second; 990 991 wakeup(msqkptr); 992 td->td_retval[0] = 0; 993done2: 994 mtx_unlock(&msq_mtx); 995 return (error); 996} 997 998#ifndef _SYS_SYSPROTO_H_ 999struct msgrcv_args { 1000 int msqid; 1001 void *msgp; 1002 size_t msgsz; 1003 long msgtyp; 1004 int msgflg; 1005}; 1006#endif 1007 1008/* 1009 * MPSAFE 1010 */ 1011int 1012msgrcv(td, uap) 1013 struct thread *td; 1014 register struct msgrcv_args *uap; 1015{ 1016 int msqid = uap->msqid; 1017 void *user_msgp = uap->msgp; 1018 size_t msgsz = uap->msgsz; 1019 long msgtyp = uap->msgtyp; 1020 int msgflg = uap->msgflg; 1021 size_t len; 1022 register struct msqid_kernel *msqkptr; 1023 register struct msg *msghdr; 1024 int error = 0; 1025 short next; 1026 1027 DPRINTF(("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp, 1028 msgsz, msgtyp, msgflg)); 1029 1030 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 1031 return (ENOSYS); 1032 1033 msqid = IPCID_TO_IX(msqid); 1034 1035 if (msqid < 0 || msqid >= msginfo.msgmni) { 1036 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 1037 msginfo.msgmni)); 1038 return (EINVAL); 1039 } 1040 1041 msqkptr = &msqids[msqid]; 1042 mtx_lock(&msq_mtx); 1043 if (msqkptr->u.msg_qbytes == 0) { 1044 DPRINTF(("no such message queue id\n")); 1045 error = EINVAL; 1046 goto done2; 1047 } 1048 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1049 DPRINTF(("wrong sequence number\n")); 1050 error = EINVAL; 1051 goto done2; 1052 } 1053 1054 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1055 DPRINTF(("requester doesn't have read access\n")); 1056 goto done2; 1057 } 1058 1059#ifdef MAC 1060 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr); 1061 if (error != 0) { 1062 MPRINTF(("mac_check_sysv_msqrcv returned %d\n", error)); 1063 goto done2; 1064 } 1065#endif 1066 1067 msghdr = NULL; 1068 while (msghdr == NULL) { 1069 if (msgtyp == 0) { 1070 msghdr = msqkptr->u.msg_first; 1071 if (msghdr != NULL) { 1072 if (msgsz < msghdr->msg_ts && 1073 (msgflg & MSG_NOERROR) == 0) { 1074 DPRINTF(("first message on the queue " 1075 "is too big (want %d, got %d)\n", 1076 msgsz, msghdr->msg_ts)); 1077 error = E2BIG; 1078 goto done2; 1079 } 1080#ifdef MAC 1081 error = mac_check_sysv_msgrcv(td->td_ucred, 1082 msghdr); 1083 if (error != 0) { 1084 MPRINTF(("mac_check_sysv_msgrcv " 1085 "returned %d\n", error)); 1086 goto done2; 1087 } 1088#endif 1089 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1090 msqkptr->u.msg_first = NULL; 1091 msqkptr->u.msg_last = NULL; 1092 } else { 1093 msqkptr->u.msg_first = msghdr->msg_next; 1094 if (msqkptr->u.msg_first == NULL) 1095 panic("msg_first/last screwed up #1"); 1096 } 1097 } 1098 } else { 1099 struct msg *previous; 1100 struct msg **prev; 1101 1102 previous = NULL; 1103 prev = &(msqkptr->u.msg_first); 1104 while ((msghdr = *prev) != NULL) { 1105 /* 1106 * Is this message's type an exact match or is 1107 * this message's type less than or equal to 1108 * the absolute value of a negative msgtyp? 1109 * Note that the second half of this test can 1110 * NEVER be true if msgtyp is positive since 1111 * msg_type is always positive! 1112 */ 1113 1114 if (msgtyp == msghdr->msg_type || 1115 msghdr->msg_type <= -msgtyp) { 1116 DPRINTF(("found message type %d, " 1117 "requested %d\n", 1118 msghdr->msg_type, msgtyp)); 1119 if (msgsz < msghdr->msg_ts && 1120 (msgflg & MSG_NOERROR) == 0) { 1121 DPRINTF(("requested message " 1122 "on the queue is too big " 1123 "(want %d, got %d)\n", 1124 msgsz, msghdr->msg_ts)); 1125 error = E2BIG; 1126 goto done2; 1127 } 1128#ifdef MAC 1129 error = mac_check_sysv_msgrcv( 1130 td->td_ucred, msghdr); 1131 if (error != 0) { 1132 MPRINTF(("mac_check_sysv_" 1133 "msgrcv returned %d\n", 1134 error)); 1135 goto done2; 1136 } 1137#endif 1138 *prev = msghdr->msg_next; 1139 if (msghdr == msqkptr->u.msg_last) { 1140 if (previous == NULL) { 1141 if (prev != 1142 &msqkptr->u.msg_first) 1143 panic("msg_first/last screwed up #2"); 1144 msqkptr->u.msg_first = 1145 NULL; 1146 msqkptr->u.msg_last = 1147 NULL; 1148 } else { 1149 if (prev == 1150 &msqkptr->u.msg_first) 1151 panic("msg_first/last screwed up #3"); 1152 msqkptr->u.msg_last = 1153 previous; 1154 } 1155 } 1156 break; 1157 } 1158 previous = msghdr; 1159 prev = &(msghdr->msg_next); 1160 } 1161 } 1162 1163 /* 1164 * We've either extracted the msghdr for the appropriate 1165 * message or there isn't one. 1166 * If there is one then bail out of this loop. 1167 */ 1168 1169 if (msghdr != NULL) 1170 break; 1171 1172 /* 1173 * Hmph! No message found. Does the user want to wait? 1174 */ 1175 1176 if ((msgflg & IPC_NOWAIT) != 0) { 1177 DPRINTF(("no appropriate message found (msgtyp=%d)\n", 1178 msgtyp)); 1179 /* The SVID says to return ENOMSG. */ 1180 error = ENOMSG; 1181 goto done2; 1182 } 1183 1184 /* 1185 * Wait for something to happen 1186 */ 1187 1188 DPRINTF(("msgrcv: goodnight\n")); 1189 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1190 "msgwait", 0); 1191 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1192 1193 if (error != 0) { 1194 DPRINTF(("msgsnd: interrupted system call\n")); 1195 error = EINTR; 1196 goto done2; 1197 } 1198 1199 /* 1200 * Make sure that the msq queue still exists 1201 */ 1202 1203 if (msqkptr->u.msg_qbytes == 0 || 1204 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1205 DPRINTF(("msqid deleted\n")); 1206 error = EIDRM; 1207 goto done2; 1208 } 1209 } 1210 1211 /* 1212 * Return the message to the user. 1213 * 1214 * First, do the bookkeeping (before we risk being interrupted). 1215 */ 1216 1217 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1218 msqkptr->u.msg_qnum--; 1219 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1220 msqkptr->u.msg_rtime = time_second; 1221 1222 /* 1223 * Make msgsz the actual amount that we'll be returning. 1224 * Note that this effectively truncates the message if it is too long 1225 * (since msgsz is never increased). 1226 */ 1227 1228 DPRINTF(("found a message, msgsz=%d, msg_ts=%d\n", msgsz, 1229 msghdr->msg_ts)); 1230 if (msgsz > msghdr->msg_ts) 1231 msgsz = msghdr->msg_ts; 1232 1233 /* 1234 * Return the type to the user. 1235 */ 1236 1237 mtx_unlock(&msq_mtx); 1238 error = copyout(&(msghdr->msg_type), user_msgp, 1239 sizeof(msghdr->msg_type)); 1240 mtx_lock(&msq_mtx); 1241 if (error != 0) { 1242 DPRINTF(("error (%d) copying out message type\n", error)); 1243 msg_freehdr(msghdr); 1244 wakeup(msqkptr); 1245 goto done2; 1246 } 1247 user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type); 1248 1249 /* 1250 * Return the segments to the user 1251 */ 1252 1253 next = msghdr->msg_spot; 1254 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1255 size_t tlen; 1256 1257 if (msgsz - len > msginfo.msgssz) 1258 tlen = msginfo.msgssz; 1259 else 1260 tlen = msgsz - len; 1261 if (next <= -1) 1262 panic("next too low #3"); 1263 if (next >= msginfo.msgseg) 1264 panic("next out of range #3"); 1265 mtx_unlock(&msq_mtx); 1266 error = copyout(&msgpool[next * msginfo.msgssz], 1267 user_msgp, tlen); 1268 mtx_lock(&msq_mtx); 1269 if (error != 0) { 1270 DPRINTF(("error (%d) copying out message segment\n", 1271 error)); 1272 msg_freehdr(msghdr); 1273 wakeup(msqkptr); 1274 goto done2; 1275 } 1276 user_msgp = (char *)user_msgp + tlen; 1277 next = msgmaps[next].next; 1278 } 1279 1280 /* 1281 * Done, return the actual number of bytes copied out. 1282 */ 1283 1284 msg_freehdr(msghdr); 1285 wakeup(msqkptr); 1286 td->td_retval[0] = msgsz; 1287done2: 1288 mtx_unlock(&msq_mtx); 1289 return (error); 1290} 1291 1292static int 1293sysctl_msqids(SYSCTL_HANDLER_ARGS) 1294{ 1295 1296 return (SYSCTL_OUT(req, msqids, 1297 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1298} 1299 1300SYSCTL_DECL(_kern_ipc); 1301SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, ""); 1302SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, ""); 1303SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, ""); 1304SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, ""); 1305SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, ""); 1306SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, ""); 1307SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1308 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1309