sysv_msg.c revision 166829
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 166829 2007-02-19 13:23:45Z 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/priv.h> 61#include <sys/proc.h> 62#include <sys/lock.h> 63#include <sys/mutex.h> 64#include <sys/module.h> 65#include <sys/msg.h> 66#include <sys/syscall.h> 67#include <sys/syscallsubr.h> 68#include <sys/sysent.h> 69#include <sys/sysctl.h> 70#include <sys/malloc.h> 71#include <sys/jail.h> 72 73#include <security/mac/mac_framework.h> 74 75static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 76 77static void msginit(void); 78static int msgunload(void); 79static int sysvmsg_modload(struct module *, int, void *); 80 81#ifdef MSG_DEBUG 82#define DPRINTF(a) printf a 83#else 84#define DPRINTF(a) 85#endif 86 87static void msg_freehdr(struct msg *msghdr); 88 89/* XXX casting to (sy_call_t *) is bogus, as usual. */ 90static sy_call_t *msgcalls[] = { 91 (sy_call_t *)msgctl, (sy_call_t *)msgget, 92 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 93}; 94 95#ifndef MSGSSZ 96#define MSGSSZ 8 /* Each segment must be 2^N long */ 97#endif 98#ifndef MSGSEG 99#define MSGSEG 2048 /* must be less than 32767 */ 100#endif 101#define MSGMAX (MSGSSZ*MSGSEG) 102#ifndef MSGMNB 103#define MSGMNB 2048 /* max # of bytes in a queue */ 104#endif 105#ifndef MSGMNI 106#define MSGMNI 40 107#endif 108#ifndef MSGTQL 109#define MSGTQL 40 110#endif 111 112/* 113 * Based on the configuration parameters described in an SVR2 (yes, two) 114 * config(1m) man page. 115 * 116 * Each message is broken up and stored in segments that are msgssz bytes 117 * long. For efficiency reasons, this should be a power of two. Also, 118 * it doesn't make sense if it is less than 8 or greater than about 256. 119 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 120 * two between 8 and 1024 inclusive (and panic's if it isn't). 121 */ 122struct msginfo msginfo = { 123 MSGMAX, /* max chars in a message */ 124 MSGMNI, /* # of message queue identifiers */ 125 MSGMNB, /* max chars in a queue */ 126 MSGTQL, /* max messages in system */ 127 MSGSSZ, /* size of a message segment */ 128 /* (must be small power of 2 greater than 4) */ 129 MSGSEG /* number of message segments */ 130}; 131 132/* 133 * macros to convert between msqid_ds's and msqid's. 134 * (specific to this implementation) 135 */ 136#define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 137#define MSQID_IX(id) ((id) & 0xffff) 138#define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 139 140/* 141 * The rest of this file is specific to this particular implementation. 142 */ 143 144struct msgmap { 145 short next; /* next segment in buffer */ 146 /* -1 -> available */ 147 /* 0..(MSGSEG-1) -> index of next segment */ 148}; 149 150#define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 151 152static int nfree_msgmaps; /* # of free map entries */ 153static short free_msgmaps; /* head of linked list of free map entries */ 154static struct msg *free_msghdrs;/* list of free msg headers */ 155static char *msgpool; /* MSGMAX byte long msg buffer pool */ 156static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 157static struct msg *msghdrs; /* MSGTQL msg headers */ 158static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 159static struct mtx msq_mtx; /* global mutex for message queues. */ 160 161static void 162msginit() 163{ 164 register int i; 165 166 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 167 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 168 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 169 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 170 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 171 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 172 173 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 174 if (msgpool == NULL) 175 panic("msgpool is NULL"); 176 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 177 if (msgmaps == NULL) 178 panic("msgmaps is NULL"); 179 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 180 if (msghdrs == NULL) 181 panic("msghdrs is NULL"); 182 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 183 M_WAITOK); 184 if (msqids == NULL) 185 panic("msqids is NULL"); 186 187 /* 188 * msginfo.msgssz should be a power of two for efficiency reasons. 189 * It is also pretty silly if msginfo.msgssz is less than 8 190 * or greater than about 256 so ... 191 */ 192 193 i = 8; 194 while (i < 1024 && i != msginfo.msgssz) 195 i <<= 1; 196 if (i != msginfo.msgssz) { 197 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 198 msginfo.msgssz)); 199 panic("msginfo.msgssz not a small power of 2"); 200 } 201 202 if (msginfo.msgseg > 32767) { 203 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 204 panic("msginfo.msgseg > 32767"); 205 } 206 207 if (msgmaps == NULL) 208 panic("msgmaps is NULL"); 209 210 for (i = 0; i < msginfo.msgseg; i++) { 211 if (i > 0) 212 msgmaps[i-1].next = i; 213 msgmaps[i].next = -1; /* implies entry is available */ 214 } 215 free_msgmaps = 0; 216 nfree_msgmaps = msginfo.msgseg; 217 218 if (msghdrs == NULL) 219 panic("msghdrs is NULL"); 220 221 for (i = 0; i < msginfo.msgtql; i++) { 222 msghdrs[i].msg_type = 0; 223 if (i > 0) 224 msghdrs[i-1].msg_next = &msghdrs[i]; 225 msghdrs[i].msg_next = NULL; 226#ifdef MAC 227 mac_init_sysv_msgmsg(&msghdrs[i]); 228#endif 229 } 230 free_msghdrs = &msghdrs[0]; 231 232 if (msqids == NULL) 233 panic("msqids is NULL"); 234 235 for (i = 0; i < msginfo.msgmni; i++) { 236 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 237 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 238 msqids[i].u.msg_perm.mode = 0; 239#ifdef MAC 240 mac_init_sysv_msgqueue(&msqids[i]); 241#endif 242 } 243 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 244} 245 246static int 247msgunload() 248{ 249 struct msqid_kernel *msqkptr; 250 int msqid; 251#ifdef MAC 252 int i; 253#endif 254 255 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 256 /* 257 * Look for an unallocated and unlocked msqid_ds. 258 * msqid_ds's can be locked by msgsnd or msgrcv while 259 * they are copying the message in/out. We can't 260 * re-use the entry until they release it. 261 */ 262 msqkptr = &msqids[msqid]; 263 if (msqkptr->u.msg_qbytes != 0 || 264 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 265 break; 266 } 267 if (msqid != msginfo.msgmni) 268 return (EBUSY); 269 270#ifdef MAC 271 for (i = 0; i < msginfo.msgtql; i++) 272 mac_destroy_sysv_msgmsg(&msghdrs[i]); 273 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 274 mac_destroy_sysv_msgqueue(&msqids[msqid]); 275#endif 276 free(msgpool, M_MSG); 277 free(msgmaps, M_MSG); 278 free(msghdrs, M_MSG); 279 free(msqids, M_MSG); 280 mtx_destroy(&msq_mtx); 281 return (0); 282} 283 284 285static int 286sysvmsg_modload(struct module *module, int cmd, void *arg) 287{ 288 int error = 0; 289 290 switch (cmd) { 291 case MOD_LOAD: 292 msginit(); 293 break; 294 case MOD_UNLOAD: 295 error = msgunload(); 296 break; 297 case MOD_SHUTDOWN: 298 break; 299 default: 300 error = EINVAL; 301 break; 302 } 303 return (error); 304} 305 306static moduledata_t sysvmsg_mod = { 307 "sysvmsg", 308 &sysvmsg_modload, 309 NULL 310}; 311 312SYSCALL_MODULE_HELPER(msgsys); 313SYSCALL_MODULE_HELPER(msgctl); 314SYSCALL_MODULE_HELPER(msgget); 315SYSCALL_MODULE_HELPER(msgsnd); 316SYSCALL_MODULE_HELPER(msgrcv); 317 318DECLARE_MODULE(sysvmsg, sysvmsg_mod, 319 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 320MODULE_VERSION(sysvmsg, 1); 321 322/* 323 * Entry point for all MSG calls 324 * 325 * MPSAFE 326 */ 327int 328msgsys(td, uap) 329 struct thread *td; 330 /* XXX actually varargs. */ 331 struct msgsys_args /* { 332 int which; 333 int a2; 334 int a3; 335 int a4; 336 int a5; 337 int a6; 338 } */ *uap; 339{ 340 int error; 341 342 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 343 return (ENOSYS); 344 if (uap->which < 0 || 345 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 346 return (EINVAL); 347 error = (*msgcalls[uap->which])(td, &uap->a2); 348 return (error); 349} 350 351static void 352msg_freehdr(msghdr) 353 struct msg *msghdr; 354{ 355 while (msghdr->msg_ts > 0) { 356 short next; 357 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 358 panic("msghdr->msg_spot out of range"); 359 next = msgmaps[msghdr->msg_spot].next; 360 msgmaps[msghdr->msg_spot].next = free_msgmaps; 361 free_msgmaps = msghdr->msg_spot; 362 nfree_msgmaps++; 363 msghdr->msg_spot = next; 364 if (msghdr->msg_ts >= msginfo.msgssz) 365 msghdr->msg_ts -= msginfo.msgssz; 366 else 367 msghdr->msg_ts = 0; 368 } 369 if (msghdr->msg_spot != -1) 370 panic("msghdr->msg_spot != -1"); 371 msghdr->msg_next = free_msghdrs; 372 free_msghdrs = msghdr; 373#ifdef MAC 374 mac_cleanup_sysv_msgmsg(msghdr); 375#endif 376} 377 378#ifndef _SYS_SYSPROTO_H_ 379struct msgctl_args { 380 int msqid; 381 int cmd; 382 struct msqid_ds *buf; 383}; 384#endif 385 386/* 387 * MPSAFE 388 */ 389int 390msgctl(td, uap) 391 struct thread *td; 392 register struct msgctl_args *uap; 393{ 394 int msqid = uap->msqid; 395 int cmd = uap->cmd; 396 struct msqid_ds msqbuf; 397 int error; 398 399 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 400 if (cmd == IPC_SET && 401 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 402 return (error); 403 error = kern_msgctl(td, msqid, cmd, &msqbuf); 404 if (cmd == IPC_STAT && error == 0) 405 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 406 return (error); 407} 408 409int 410kern_msgctl(td, msqid, cmd, msqbuf) 411 struct thread *td; 412 int msqid; 413 int cmd; 414 struct msqid_ds *msqbuf; 415{ 416 int rval, error, msqix; 417 register struct msqid_kernel *msqkptr; 418 419 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 420 return (ENOSYS); 421 422 msqix = IPCID_TO_IX(msqid); 423 424 if (msqix < 0 || msqix >= msginfo.msgmni) { 425 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 426 msginfo.msgmni)); 427 return (EINVAL); 428 } 429 430 msqkptr = &msqids[msqix]; 431 432 mtx_lock(&msq_mtx); 433 if (msqkptr->u.msg_qbytes == 0) { 434 DPRINTF(("no such msqid\n")); 435 error = EINVAL; 436 goto done2; 437 } 438 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 439 DPRINTF(("wrong sequence number\n")); 440 error = EINVAL; 441 goto done2; 442 } 443#ifdef MAC 444 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd); 445 if (error != 0) 446 goto done2; 447#endif 448 449 error = 0; 450 rval = 0; 451 452 switch (cmd) { 453 454 case IPC_RMID: 455 { 456 struct msg *msghdr; 457 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 458 goto done2; 459 460#ifdef MAC 461 /* 462 * Check that the thread has MAC access permissions to 463 * individual msghdrs. Note: We need to do this in a 464 * separate loop because the actual loop alters the 465 * msq/msghdr info as it progresses, and there is no going 466 * back if half the way through we discover that the 467 * thread cannot free a certain msghdr. The msq will get 468 * into an inconsistent state. 469 */ 470 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 471 msghdr = msghdr->msg_next) { 472 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr); 473 if (error != 0) 474 goto done2; 475 } 476#endif 477 478 /* Free the message headers */ 479 msghdr = msqkptr->u.msg_first; 480 while (msghdr != NULL) { 481 struct msg *msghdr_tmp; 482 483 /* Free the segments of each message */ 484 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 485 msqkptr->u.msg_qnum--; 486 msghdr_tmp = msghdr; 487 msghdr = msghdr->msg_next; 488 msg_freehdr(msghdr_tmp); 489 } 490 491 if (msqkptr->u.msg_cbytes != 0) 492 panic("msg_cbytes is screwed up"); 493 if (msqkptr->u.msg_qnum != 0) 494 panic("msg_qnum is screwed up"); 495 496 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 497 498#ifdef MAC 499 mac_cleanup_sysv_msgqueue(msqkptr); 500#endif 501 502 wakeup(msqkptr); 503 } 504 505 break; 506 507 case IPC_SET: 508 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 509 goto done2; 510 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 511 error = priv_check_cred(td->td_ucred, 512 PRIV_IPC_MSGSIZE, SUSER_ALLOWJAIL); 513 if (error) 514 goto done2; 515 } 516 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 517 DPRINTF(("can't increase msg_qbytes beyond %d" 518 "(truncating)\n", msginfo.msgmnb)); 519 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 520 } 521 if (msqbuf->msg_qbytes == 0) { 522 DPRINTF(("can't reduce msg_qbytes to 0\n")); 523 error = EINVAL; /* non-standard errno! */ 524 goto done2; 525 } 526 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 527 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 528 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 529 (msqbuf->msg_perm.mode & 0777); 530 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 531 msqkptr->u.msg_ctime = time_second; 532 break; 533 534 case IPC_STAT: 535 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 536 DPRINTF(("requester doesn't have read access\n")); 537 goto done2; 538 } 539 *msqbuf = msqkptr->u; 540 break; 541 542 default: 543 DPRINTF(("invalid command %d\n", cmd)); 544 error = EINVAL; 545 goto done2; 546 } 547 548 if (error == 0) 549 td->td_retval[0] = rval; 550done2: 551 mtx_unlock(&msq_mtx); 552 return (error); 553} 554 555#ifndef _SYS_SYSPROTO_H_ 556struct msgget_args { 557 key_t key; 558 int msgflg; 559}; 560#endif 561 562/* 563 * MPSAFE 564 */ 565int 566msgget(td, uap) 567 struct thread *td; 568 register struct msgget_args *uap; 569{ 570 int msqid, error = 0; 571 int key = uap->key; 572 int msgflg = uap->msgflg; 573 struct ucred *cred = td->td_ucred; 574 register struct msqid_kernel *msqkptr = NULL; 575 576 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 577 578 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 579 return (ENOSYS); 580 581 mtx_lock(&msq_mtx); 582 if (key != IPC_PRIVATE) { 583 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 584 msqkptr = &msqids[msqid]; 585 if (msqkptr->u.msg_qbytes != 0 && 586 msqkptr->u.msg_perm.key == key) 587 break; 588 } 589 if (msqid < msginfo.msgmni) { 590 DPRINTF(("found public key\n")); 591 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 592 DPRINTF(("not exclusive\n")); 593 error = EEXIST; 594 goto done2; 595 } 596 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 597 msgflg & 0700))) { 598 DPRINTF(("requester doesn't have 0%o access\n", 599 msgflg & 0700)); 600 goto done2; 601 } 602#ifdef MAC 603 error = mac_check_sysv_msqget(cred, msqkptr); 604 if (error != 0) 605 goto done2; 606#endif 607 goto found; 608 } 609 } 610 611 DPRINTF(("need to allocate the msqid_ds\n")); 612 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 613 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 614 /* 615 * Look for an unallocated and unlocked msqid_ds. 616 * msqid_ds's can be locked by msgsnd or msgrcv while 617 * they are copying the message in/out. We can't 618 * re-use the entry until they release it. 619 */ 620 msqkptr = &msqids[msqid]; 621 if (msqkptr->u.msg_qbytes == 0 && 622 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 623 break; 624 } 625 if (msqid == msginfo.msgmni) { 626 DPRINTF(("no more msqid_ds's available\n")); 627 error = ENOSPC; 628 goto done2; 629 } 630 DPRINTF(("msqid %d is available\n", msqid)); 631 msqkptr->u.msg_perm.key = key; 632 msqkptr->u.msg_perm.cuid = cred->cr_uid; 633 msqkptr->u.msg_perm.uid = cred->cr_uid; 634 msqkptr->u.msg_perm.cgid = cred->cr_gid; 635 msqkptr->u.msg_perm.gid = cred->cr_gid; 636 msqkptr->u.msg_perm.mode = (msgflg & 0777); 637 /* Make sure that the returned msqid is unique */ 638 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 639 msqkptr->u.msg_first = NULL; 640 msqkptr->u.msg_last = NULL; 641 msqkptr->u.msg_cbytes = 0; 642 msqkptr->u.msg_qnum = 0; 643 msqkptr->u.msg_qbytes = msginfo.msgmnb; 644 msqkptr->u.msg_lspid = 0; 645 msqkptr->u.msg_lrpid = 0; 646 msqkptr->u.msg_stime = 0; 647 msqkptr->u.msg_rtime = 0; 648 msqkptr->u.msg_ctime = time_second; 649#ifdef MAC 650 mac_create_sysv_msgqueue(cred, msqkptr); 651#endif 652 } else { 653 DPRINTF(("didn't find it and wasn't asked to create it\n")); 654 error = ENOENT; 655 goto done2; 656 } 657 658found: 659 /* Construct the unique msqid */ 660 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 661done2: 662 mtx_unlock(&msq_mtx); 663 return (error); 664} 665 666#ifndef _SYS_SYSPROTO_H_ 667struct msgsnd_args { 668 int msqid; 669 const void *msgp; 670 size_t msgsz; 671 int msgflg; 672}; 673#endif 674 675int 676kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 677 struct thread *td; 678 int msqid; 679 const void *msgp; /* XXX msgp is actually mtext. */ 680 size_t msgsz; 681 int msgflg; 682 long mtype; 683{ 684 int msqix, segs_needed, error = 0; 685 register struct msqid_kernel *msqkptr; 686 register struct msg *msghdr; 687 short next; 688 689 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 690 return (ENOSYS); 691 692 mtx_lock(&msq_mtx); 693 msqix = IPCID_TO_IX(msqid); 694 695 if (msqix < 0 || msqix >= msginfo.msgmni) { 696 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 697 msginfo.msgmni)); 698 error = EINVAL; 699 goto done2; 700 } 701 702 msqkptr = &msqids[msqix]; 703 if (msqkptr->u.msg_qbytes == 0) { 704 DPRINTF(("no such message queue id\n")); 705 error = EINVAL; 706 goto done2; 707 } 708 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 709 DPRINTF(("wrong sequence number\n")); 710 error = EINVAL; 711 goto done2; 712 } 713 714 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 715 DPRINTF(("requester doesn't have write access\n")); 716 goto done2; 717 } 718 719#ifdef MAC 720 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr); 721 if (error != 0) 722 goto done2; 723#endif 724 725 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 726 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 727 msginfo.msgssz, segs_needed)); 728 for (;;) { 729 int need_more_resources = 0; 730 731 /* 732 * check msgsz 733 * (inside this loop in case msg_qbytes changes while we sleep) 734 */ 735 736 if (msgsz > msqkptr->u.msg_qbytes) { 737 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 738 error = EINVAL; 739 goto done2; 740 } 741 742 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 743 DPRINTF(("msqid is locked\n")); 744 need_more_resources = 1; 745 } 746 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 747 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 748 need_more_resources = 1; 749 } 750 if (segs_needed > nfree_msgmaps) { 751 DPRINTF(("segs_needed > nfree_msgmaps\n")); 752 need_more_resources = 1; 753 } 754 if (free_msghdrs == NULL) { 755 DPRINTF(("no more msghdrs\n")); 756 need_more_resources = 1; 757 } 758 759 if (need_more_resources) { 760 int we_own_it; 761 762 if ((msgflg & IPC_NOWAIT) != 0) { 763 DPRINTF(("need more resources but caller " 764 "doesn't want to wait\n")); 765 error = EAGAIN; 766 goto done2; 767 } 768 769 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 770 DPRINTF(("we don't own the msqid_ds\n")); 771 we_own_it = 0; 772 } else { 773 /* Force later arrivals to wait for our 774 request */ 775 DPRINTF(("we own the msqid_ds\n")); 776 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 777 we_own_it = 1; 778 } 779 DPRINTF(("msgsnd: goodnight\n")); 780 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 781 "msgsnd", hz); 782 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 783 if (we_own_it) 784 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 785 if (error == EWOULDBLOCK) { 786 DPRINTF(("msgsnd: timed out\n")); 787 continue; 788 } 789 if (error != 0) { 790 DPRINTF(("msgsnd: interrupted system call\n")); 791 error = EINTR; 792 goto done2; 793 } 794 795 /* 796 * Make sure that the msq queue still exists 797 */ 798 799 if (msqkptr->u.msg_qbytes == 0) { 800 DPRINTF(("msqid deleted\n")); 801 error = EIDRM; 802 goto done2; 803 } 804 805 } else { 806 DPRINTF(("got all the resources that we need\n")); 807 break; 808 } 809 } 810 811 /* 812 * We have the resources that we need. 813 * Make sure! 814 */ 815 816 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 817 panic("msg_perm.mode & MSG_LOCKED"); 818 if (segs_needed > nfree_msgmaps) 819 panic("segs_needed > nfree_msgmaps"); 820 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 821 panic("msgsz + msg_cbytes > msg_qbytes"); 822 if (free_msghdrs == NULL) 823 panic("no more msghdrs"); 824 825 /* 826 * Re-lock the msqid_ds in case we page-fault when copying in the 827 * message 828 */ 829 830 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 831 panic("msqid_ds is already locked"); 832 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 833 834 /* 835 * Allocate a message header 836 */ 837 838 msghdr = free_msghdrs; 839 free_msghdrs = msghdr->msg_next; 840 msghdr->msg_spot = -1; 841 msghdr->msg_ts = msgsz; 842 msghdr->msg_type = mtype; 843#ifdef MAC 844 /* 845 * XXXMAC: Should the mac_check_sysv_msgmsq check follow here 846 * immediately? Or, should it be checked just before the msg is 847 * enqueued in the msgq (as it is done now)? 848 */ 849 mac_create_sysv_msgmsg(td->td_ucred, msqkptr, msghdr); 850#endif 851 852 /* 853 * Allocate space for the message 854 */ 855 856 while (segs_needed > 0) { 857 if (nfree_msgmaps <= 0) 858 panic("not enough msgmaps"); 859 if (free_msgmaps == -1) 860 panic("nil free_msgmaps"); 861 next = free_msgmaps; 862 if (next <= -1) 863 panic("next too low #1"); 864 if (next >= msginfo.msgseg) 865 panic("next out of range #1"); 866 DPRINTF(("allocating segment %d to message\n", next)); 867 free_msgmaps = msgmaps[next].next; 868 nfree_msgmaps--; 869 msgmaps[next].next = msghdr->msg_spot; 870 msghdr->msg_spot = next; 871 segs_needed--; 872 } 873 874 /* 875 * Validate the message type 876 */ 877 878 if (msghdr->msg_type < 1) { 879 msg_freehdr(msghdr); 880 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 881 wakeup(msqkptr); 882 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 883 error = EINVAL; 884 goto done2; 885 } 886 887 /* 888 * Copy in the message body 889 */ 890 891 next = msghdr->msg_spot; 892 while (msgsz > 0) { 893 size_t tlen; 894 if (msgsz > msginfo.msgssz) 895 tlen = msginfo.msgssz; 896 else 897 tlen = msgsz; 898 if (next <= -1) 899 panic("next too low #2"); 900 if (next >= msginfo.msgseg) 901 panic("next out of range #2"); 902 mtx_unlock(&msq_mtx); 903 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 904 tlen)) != 0) { 905 mtx_lock(&msq_mtx); 906 DPRINTF(("error %d copying in message segment\n", 907 error)); 908 msg_freehdr(msghdr); 909 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 910 wakeup(msqkptr); 911 goto done2; 912 } 913 mtx_lock(&msq_mtx); 914 msgsz -= tlen; 915 msgp = (const char *)msgp + tlen; 916 next = msgmaps[next].next; 917 } 918 if (next != -1) 919 panic("didn't use all the msg segments"); 920 921 /* 922 * We've got the message. Unlock the msqid_ds. 923 */ 924 925 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 926 927 /* 928 * Make sure that the msqid_ds is still allocated. 929 */ 930 931 if (msqkptr->u.msg_qbytes == 0) { 932 msg_freehdr(msghdr); 933 wakeup(msqkptr); 934 error = EIDRM; 935 goto done2; 936 } 937 938#ifdef MAC 939 /* 940 * Note: Since the task/thread allocates the msghdr and usually 941 * primes it with its own MAC label, for a majority of policies, it 942 * won't be necessary to check whether the msghdr has access 943 * permissions to the msgq. The mac_check_sysv_msqsnd check would 944 * suffice in that case. However, this hook may be required where 945 * individual policies derive a non-identical label for the msghdr 946 * from the current thread label and may want to check the msghdr 947 * enqueue permissions, along with read/write permissions to the 948 * msgq. 949 */ 950 error = mac_check_sysv_msgmsq(td->td_ucred, msghdr, msqkptr); 951 if (error != 0) { 952 msg_freehdr(msghdr); 953 wakeup(msqkptr); 954 goto done2; 955 } 956#endif 957 958 /* 959 * Put the message into the queue 960 */ 961 if (msqkptr->u.msg_first == NULL) { 962 msqkptr->u.msg_first = msghdr; 963 msqkptr->u.msg_last = msghdr; 964 } else { 965 msqkptr->u.msg_last->msg_next = msghdr; 966 msqkptr->u.msg_last = msghdr; 967 } 968 msqkptr->u.msg_last->msg_next = NULL; 969 970 msqkptr->u.msg_cbytes += msghdr->msg_ts; 971 msqkptr->u.msg_qnum++; 972 msqkptr->u.msg_lspid = td->td_proc->p_pid; 973 msqkptr->u.msg_stime = time_second; 974 975 wakeup(msqkptr); 976 td->td_retval[0] = 0; 977done2: 978 mtx_unlock(&msq_mtx); 979 return (error); 980} 981 982/* 983 * MPSAFE 984 */ 985int 986msgsnd(td, uap) 987 struct thread *td; 988 register struct msgsnd_args *uap; 989{ 990 int error; 991 long mtype; 992 993 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 994 uap->msgsz, uap->msgflg)); 995 996 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 997 DPRINTF(("error %d copying the message type\n", error)); 998 return (error); 999 } 1000 return (kern_msgsnd(td, uap->msqid, 1001 (const char *)uap->msgp + sizeof(mtype), 1002 uap->msgsz, uap->msgflg, mtype)); 1003} 1004 1005#ifndef _SYS_SYSPROTO_H_ 1006struct msgrcv_args { 1007 int msqid; 1008 void *msgp; 1009 size_t msgsz; 1010 long msgtyp; 1011 int msgflg; 1012}; 1013#endif 1014 1015int 1016kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1017 struct thread *td; 1018 int msqid; 1019 void *msgp; /* XXX msgp is actually mtext. */ 1020 size_t msgsz; 1021 long msgtyp; 1022 int msgflg; 1023 long *mtype; 1024{ 1025 size_t len; 1026 register struct msqid_kernel *msqkptr; 1027 register struct msg *msghdr; 1028 int msqix, error = 0; 1029 short next; 1030 1031 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 1032 return (ENOSYS); 1033 1034 msqix = IPCID_TO_IX(msqid); 1035 1036 if (msqix < 0 || msqix >= msginfo.msgmni) { 1037 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1038 msginfo.msgmni)); 1039 return (EINVAL); 1040 } 1041 1042 msqkptr = &msqids[msqix]; 1043 mtx_lock(&msq_mtx); 1044 if (msqkptr->u.msg_qbytes == 0) { 1045 DPRINTF(("no such message queue id\n")); 1046 error = EINVAL; 1047 goto done2; 1048 } 1049 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1050 DPRINTF(("wrong sequence number\n")); 1051 error = EINVAL; 1052 goto done2; 1053 } 1054 1055 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1056 DPRINTF(("requester doesn't have read access\n")); 1057 goto done2; 1058 } 1059 1060#ifdef MAC 1061 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr); 1062 if (error != 0) 1063 goto done2; 1064#endif 1065 1066 msghdr = NULL; 1067 while (msghdr == NULL) { 1068 if (msgtyp == 0) { 1069 msghdr = msqkptr->u.msg_first; 1070 if (msghdr != NULL) { 1071 if (msgsz < msghdr->msg_ts && 1072 (msgflg & MSG_NOERROR) == 0) { 1073 DPRINTF(("first message on the queue " 1074 "is too big (want %zu, got %d)\n", 1075 msgsz, msghdr->msg_ts)); 1076 error = E2BIG; 1077 goto done2; 1078 } 1079#ifdef MAC 1080 error = mac_check_sysv_msgrcv(td->td_ucred, 1081 msghdr); 1082 if (error != 0) 1083 goto done2; 1084#endif 1085 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1086 msqkptr->u.msg_first = NULL; 1087 msqkptr->u.msg_last = NULL; 1088 } else { 1089 msqkptr->u.msg_first = msghdr->msg_next; 1090 if (msqkptr->u.msg_first == NULL) 1091 panic("msg_first/last screwed up #1"); 1092 } 1093 } 1094 } else { 1095 struct msg *previous; 1096 struct msg **prev; 1097 1098 previous = NULL; 1099 prev = &(msqkptr->u.msg_first); 1100 while ((msghdr = *prev) != NULL) { 1101 /* 1102 * Is this message's type an exact match or is 1103 * this message's type less than or equal to 1104 * the absolute value of a negative msgtyp? 1105 * Note that the second half of this test can 1106 * NEVER be true if msgtyp is positive since 1107 * msg_type is always positive! 1108 */ 1109 1110 if (msgtyp == msghdr->msg_type || 1111 msghdr->msg_type <= -msgtyp) { 1112 DPRINTF(("found message type %ld, " 1113 "requested %ld\n", 1114 msghdr->msg_type, msgtyp)); 1115 if (msgsz < msghdr->msg_ts && 1116 (msgflg & MSG_NOERROR) == 0) { 1117 DPRINTF(("requested message " 1118 "on the queue is too big " 1119 "(want %zu, got %hu)\n", 1120 msgsz, msghdr->msg_ts)); 1121 error = E2BIG; 1122 goto done2; 1123 } 1124#ifdef MAC 1125 error = mac_check_sysv_msgrcv( 1126 td->td_ucred, msghdr); 1127 if (error != 0) 1128 goto done2; 1129#endif 1130 *prev = msghdr->msg_next; 1131 if (msghdr == msqkptr->u.msg_last) { 1132 if (previous == NULL) { 1133 if (prev != 1134 &msqkptr->u.msg_first) 1135 panic("msg_first/last screwed up #2"); 1136 msqkptr->u.msg_first = 1137 NULL; 1138 msqkptr->u.msg_last = 1139 NULL; 1140 } else { 1141 if (prev == 1142 &msqkptr->u.msg_first) 1143 panic("msg_first/last screwed up #3"); 1144 msqkptr->u.msg_last = 1145 previous; 1146 } 1147 } 1148 break; 1149 } 1150 previous = msghdr; 1151 prev = &(msghdr->msg_next); 1152 } 1153 } 1154 1155 /* 1156 * We've either extracted the msghdr for the appropriate 1157 * message or there isn't one. 1158 * If there is one then bail out of this loop. 1159 */ 1160 1161 if (msghdr != NULL) 1162 break; 1163 1164 /* 1165 * Hmph! No message found. Does the user want to wait? 1166 */ 1167 1168 if ((msgflg & IPC_NOWAIT) != 0) { 1169 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1170 msgtyp)); 1171 /* The SVID says to return ENOMSG. */ 1172 error = ENOMSG; 1173 goto done2; 1174 } 1175 1176 /* 1177 * Wait for something to happen 1178 */ 1179 1180 DPRINTF(("msgrcv: goodnight\n")); 1181 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1182 "msgrcv", 0); 1183 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1184 1185 if (error != 0) { 1186 DPRINTF(("msgrcv: interrupted system call\n")); 1187 error = EINTR; 1188 goto done2; 1189 } 1190 1191 /* 1192 * Make sure that the msq queue still exists 1193 */ 1194 1195 if (msqkptr->u.msg_qbytes == 0 || 1196 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1197 DPRINTF(("msqid deleted\n")); 1198 error = EIDRM; 1199 goto done2; 1200 } 1201 } 1202 1203 /* 1204 * Return the message to the user. 1205 * 1206 * First, do the bookkeeping (before we risk being interrupted). 1207 */ 1208 1209 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1210 msqkptr->u.msg_qnum--; 1211 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1212 msqkptr->u.msg_rtime = time_second; 1213 1214 /* 1215 * Make msgsz the actual amount that we'll be returning. 1216 * Note that this effectively truncates the message if it is too long 1217 * (since msgsz is never increased). 1218 */ 1219 1220 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1221 msghdr->msg_ts)); 1222 if (msgsz > msghdr->msg_ts) 1223 msgsz = msghdr->msg_ts; 1224 *mtype = msghdr->msg_type; 1225 1226 /* 1227 * Return the segments to the user 1228 */ 1229 1230 next = msghdr->msg_spot; 1231 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1232 size_t tlen; 1233 1234 if (msgsz - len > msginfo.msgssz) 1235 tlen = msginfo.msgssz; 1236 else 1237 tlen = msgsz - len; 1238 if (next <= -1) 1239 panic("next too low #3"); 1240 if (next >= msginfo.msgseg) 1241 panic("next out of range #3"); 1242 mtx_unlock(&msq_mtx); 1243 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1244 mtx_lock(&msq_mtx); 1245 if (error != 0) { 1246 DPRINTF(("error (%d) copying out message segment\n", 1247 error)); 1248 msg_freehdr(msghdr); 1249 wakeup(msqkptr); 1250 goto done2; 1251 } 1252 msgp = (char *)msgp + tlen; 1253 next = msgmaps[next].next; 1254 } 1255 1256 /* 1257 * Done, return the actual number of bytes copied out. 1258 */ 1259 1260 msg_freehdr(msghdr); 1261 wakeup(msqkptr); 1262 td->td_retval[0] = msgsz; 1263done2: 1264 mtx_unlock(&msq_mtx); 1265 return (error); 1266} 1267 1268/* 1269 * MPSAFE 1270 */ 1271int 1272msgrcv(td, uap) 1273 struct thread *td; 1274 register struct msgrcv_args *uap; 1275{ 1276 int error; 1277 long mtype; 1278 1279 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1280 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1281 1282 if ((error = kern_msgrcv(td, uap->msqid, 1283 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1284 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1285 return (error); 1286 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1287 DPRINTF(("error %d copying the message type\n", error)); 1288 return (error); 1289} 1290 1291static int 1292sysctl_msqids(SYSCTL_HANDLER_ARGS) 1293{ 1294 1295 return (SYSCTL_OUT(req, msqids, 1296 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1297} 1298 1299SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1300 "Maximum message size"); 1301SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1302 "Number of message queue identifiers"); 1303SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1304 "Maximum number of bytes in a queue"); 1305SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1306 "Maximum number of messages in the system"); 1307SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1308 "Size of a message segment"); 1309SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1310 "Number of message segments"); 1311SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1312 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1313