sysv_msg.c revision 163606
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 163606 2006-10-22 11:52:19Z 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/mutex.h> 63#include <sys/module.h> 64#include <sys/msg.h> 65#include <sys/syscall.h> 66#include <sys/syscallsubr.h> 67#include <sys/sysent.h> 68#include <sys/sysctl.h> 69#include <sys/malloc.h> 70#include <sys/jail.h> 71 72#include <security/mac/mac_framework.h> 73 74static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 75 76static void msginit(void); 77static int msgunload(void); 78static int sysvmsg_modload(struct module *, int, void *); 79 80#ifdef MSG_DEBUG 81#define DPRINTF(a) printf a 82#else 83#define DPRINTF(a) 84#endif 85 86static void msg_freehdr(struct msg *msghdr); 87 88/* XXX casting to (sy_call_t *) is bogus, as usual. */ 89static sy_call_t *msgcalls[] = { 90 (sy_call_t *)msgctl, (sy_call_t *)msgget, 91 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 92}; 93 94#ifndef MSGSSZ 95#define MSGSSZ 8 /* Each segment must be 2^N long */ 96#endif 97#ifndef MSGSEG 98#define MSGSEG 2048 /* must be less than 32767 */ 99#endif 100#define MSGMAX (MSGSSZ*MSGSEG) 101#ifndef MSGMNB 102#define MSGMNB 2048 /* max # of bytes in a queue */ 103#endif 104#ifndef MSGMNI 105#define MSGMNI 40 106#endif 107#ifndef MSGTQL 108#define MSGTQL 40 109#endif 110 111/* 112 * Based on the configuration parameters described in an SVR2 (yes, two) 113 * config(1m) man page. 114 * 115 * Each message is broken up and stored in segments that are msgssz bytes 116 * long. For efficiency reasons, this should be a power of two. Also, 117 * it doesn't make sense if it is less than 8 or greater than about 256. 118 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 119 * two between 8 and 1024 inclusive (and panic's if it isn't). 120 */ 121struct msginfo msginfo = { 122 MSGMAX, /* max chars in a message */ 123 MSGMNI, /* # of message queue identifiers */ 124 MSGMNB, /* max chars in a queue */ 125 MSGTQL, /* max messages in system */ 126 MSGSSZ, /* size of a message segment */ 127 /* (must be small power of 2 greater than 4) */ 128 MSGSEG /* number of message segments */ 129}; 130 131/* 132 * macros to convert between msqid_ds's and msqid's. 133 * (specific to this implementation) 134 */ 135#define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 136#define MSQID_IX(id) ((id) & 0xffff) 137#define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 138 139/* 140 * The rest of this file is specific to this particular implementation. 141 */ 142 143struct msgmap { 144 short next; /* next segment in buffer */ 145 /* -1 -> available */ 146 /* 0..(MSGSEG-1) -> index of next segment */ 147}; 148 149#define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 150 151static int nfree_msgmaps; /* # of free map entries */ 152static short free_msgmaps; /* head of linked list of free map entries */ 153static struct msg *free_msghdrs;/* list of free msg headers */ 154static char *msgpool; /* MSGMAX byte long msg buffer pool */ 155static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 156static struct msg *msghdrs; /* MSGTQL msg headers */ 157static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 158static struct mtx msq_mtx; /* global mutex for message queues. */ 159 160static void 161msginit() 162{ 163 register int i; 164 165 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 166 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 167 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 168 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 169 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 170 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 171 172 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 173 if (msgpool == NULL) 174 panic("msgpool is NULL"); 175 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 176 if (msgmaps == NULL) 177 panic("msgmaps is NULL"); 178 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 179 if (msghdrs == NULL) 180 panic("msghdrs is NULL"); 181 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 182 M_WAITOK); 183 if (msqids == NULL) 184 panic("msqids is NULL"); 185 186 /* 187 * msginfo.msgssz should be a power of two for efficiency reasons. 188 * It is also pretty silly if msginfo.msgssz is less than 8 189 * or greater than about 256 so ... 190 */ 191 192 i = 8; 193 while (i < 1024 && i != msginfo.msgssz) 194 i <<= 1; 195 if (i != msginfo.msgssz) { 196 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 197 msginfo.msgssz)); 198 panic("msginfo.msgssz not a small power of 2"); 199 } 200 201 if (msginfo.msgseg > 32767) { 202 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 203 panic("msginfo.msgseg > 32767"); 204 } 205 206 if (msgmaps == NULL) 207 panic("msgmaps is NULL"); 208 209 for (i = 0; i < msginfo.msgseg; i++) { 210 if (i > 0) 211 msgmaps[i-1].next = i; 212 msgmaps[i].next = -1; /* implies entry is available */ 213 } 214 free_msgmaps = 0; 215 nfree_msgmaps = msginfo.msgseg; 216 217 if (msghdrs == NULL) 218 panic("msghdrs is NULL"); 219 220 for (i = 0; i < msginfo.msgtql; i++) { 221 msghdrs[i].msg_type = 0; 222 if (i > 0) 223 msghdrs[i-1].msg_next = &msghdrs[i]; 224 msghdrs[i].msg_next = NULL; 225#ifdef MAC 226 mac_init_sysv_msgmsg(&msghdrs[i]); 227#endif 228 } 229 free_msghdrs = &msghdrs[0]; 230 231 if (msqids == NULL) 232 panic("msqids is NULL"); 233 234 for (i = 0; i < msginfo.msgmni; i++) { 235 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 236 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 237 msqids[i].u.msg_perm.mode = 0; 238#ifdef MAC 239 mac_init_sysv_msgqueue(&msqids[i]); 240#endif 241 } 242 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 243} 244 245static int 246msgunload() 247{ 248 struct msqid_kernel *msqkptr; 249 int msqid; 250#ifdef MAC 251 int i; 252#endif 253 254 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 255 /* 256 * Look for an unallocated and unlocked msqid_ds. 257 * msqid_ds's can be locked by msgsnd or msgrcv while 258 * they are copying the message in/out. We can't 259 * re-use the entry until they release it. 260 */ 261 msqkptr = &msqids[msqid]; 262 if (msqkptr->u.msg_qbytes != 0 || 263 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 264 break; 265 } 266 if (msqid != msginfo.msgmni) 267 return (EBUSY); 268 269#ifdef MAC 270 for (i = 0; i < msginfo.msgtql; i++) 271 mac_destroy_sysv_msgmsg(&msghdrs[i]); 272 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 273 mac_destroy_sysv_msgqueue(&msqids[msqid]); 274#endif 275 free(msgpool, M_MSG); 276 free(msgmaps, M_MSG); 277 free(msghdrs, M_MSG); 278 free(msqids, M_MSG); 279 mtx_destroy(&msq_mtx); 280 return (0); 281} 282 283 284static int 285sysvmsg_modload(struct module *module, int cmd, void *arg) 286{ 287 int error = 0; 288 289 switch (cmd) { 290 case MOD_LOAD: 291 msginit(); 292 break; 293 case MOD_UNLOAD: 294 error = msgunload(); 295 break; 296 case MOD_SHUTDOWN: 297 break; 298 default: 299 error = EINVAL; 300 break; 301 } 302 return (error); 303} 304 305static moduledata_t sysvmsg_mod = { 306 "sysvmsg", 307 &sysvmsg_modload, 308 NULL 309}; 310 311SYSCALL_MODULE_HELPER(msgsys); 312SYSCALL_MODULE_HELPER(msgctl); 313SYSCALL_MODULE_HELPER(msgget); 314SYSCALL_MODULE_HELPER(msgsnd); 315SYSCALL_MODULE_HELPER(msgrcv); 316 317DECLARE_MODULE(sysvmsg, sysvmsg_mod, 318 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 319MODULE_VERSION(sysvmsg, 1); 320 321/* 322 * Entry point for all MSG calls 323 * 324 * MPSAFE 325 */ 326int 327msgsys(td, uap) 328 struct thread *td; 329 /* XXX actually varargs. */ 330 struct msgsys_args /* { 331 int which; 332 int a2; 333 int a3; 334 int a4; 335 int a5; 336 int a6; 337 } */ *uap; 338{ 339 int error; 340 341 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 342 return (ENOSYS); 343 if (uap->which < 0 || 344 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 345 return (EINVAL); 346 error = (*msgcalls[uap->which])(td, &uap->a2); 347 return (error); 348} 349 350static void 351msg_freehdr(msghdr) 352 struct msg *msghdr; 353{ 354 while (msghdr->msg_ts > 0) { 355 short next; 356 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 357 panic("msghdr->msg_spot out of range"); 358 next = msgmaps[msghdr->msg_spot].next; 359 msgmaps[msghdr->msg_spot].next = free_msgmaps; 360 free_msgmaps = msghdr->msg_spot; 361 nfree_msgmaps++; 362 msghdr->msg_spot = next; 363 if (msghdr->msg_ts >= msginfo.msgssz) 364 msghdr->msg_ts -= msginfo.msgssz; 365 else 366 msghdr->msg_ts = 0; 367 } 368 if (msghdr->msg_spot != -1) 369 panic("msghdr->msg_spot != -1"); 370 msghdr->msg_next = free_msghdrs; 371 free_msghdrs = msghdr; 372#ifdef MAC 373 mac_cleanup_sysv_msgmsg(msghdr); 374#endif 375} 376 377#ifndef _SYS_SYSPROTO_H_ 378struct msgctl_args { 379 int msqid; 380 int cmd; 381 struct msqid_ds *buf; 382}; 383#endif 384 385/* 386 * MPSAFE 387 */ 388int 389msgctl(td, uap) 390 struct thread *td; 391 register struct msgctl_args *uap; 392{ 393 int msqid = uap->msqid; 394 int cmd = uap->cmd; 395 struct msqid_ds msqbuf; 396 int error; 397 398 DPRINTF(("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, uap->buf)); 399 if (cmd == IPC_SET && 400 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 401 return (error); 402 error = kern_msgctl(td, msqid, cmd, &msqbuf); 403 if (cmd == IPC_STAT && error == 0) 404 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 405 return (error); 406} 407 408int 409kern_msgctl(td, msqid, cmd, msqbuf) 410 struct thread *td; 411 int msqid; 412 int cmd; 413 struct msqid_ds *msqbuf; 414{ 415 int rval, error, msqix; 416 register struct msqid_kernel *msqkptr; 417 418 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 419 return (ENOSYS); 420 421 msqix = IPCID_TO_IX(msqid); 422 423 if (msqix < 0 || msqix >= msginfo.msgmni) { 424 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 425 msginfo.msgmni)); 426 return (EINVAL); 427 } 428 429 msqkptr = &msqids[msqix]; 430 431 mtx_lock(&msq_mtx); 432 if (msqkptr->u.msg_qbytes == 0) { 433 DPRINTF(("no such msqid\n")); 434 error = EINVAL; 435 goto done2; 436 } 437 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 438 DPRINTF(("wrong sequence number\n")); 439 error = EINVAL; 440 goto done2; 441 } 442#ifdef MAC 443 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd); 444 if (error != 0) 445 goto done2; 446#endif 447 448 error = 0; 449 rval = 0; 450 451 switch (cmd) { 452 453 case IPC_RMID: 454 { 455 struct msg *msghdr; 456 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 457 goto done2; 458 459#ifdef MAC 460 /* 461 * Check that the thread has MAC access permissions to 462 * individual msghdrs. Note: We need to do this in a 463 * separate loop because the actual loop alters the 464 * msq/msghdr info as it progresses, and there is no going 465 * back if half the way through we discover that the 466 * thread cannot free a certain msghdr. The msq will get 467 * into an inconsistent state. 468 */ 469 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 470 msghdr = msghdr->msg_next) { 471 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr); 472 if (error != 0) 473 goto done2; 474 } 475#endif 476 477 /* Free the message headers */ 478 msghdr = msqkptr->u.msg_first; 479 while (msghdr != NULL) { 480 struct msg *msghdr_tmp; 481 482 /* Free the segments of each message */ 483 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 484 msqkptr->u.msg_qnum--; 485 msghdr_tmp = msghdr; 486 msghdr = msghdr->msg_next; 487 msg_freehdr(msghdr_tmp); 488 } 489 490 if (msqkptr->u.msg_cbytes != 0) 491 panic("msg_cbytes is screwed up"); 492 if (msqkptr->u.msg_qnum != 0) 493 panic("msg_qnum is screwed up"); 494 495 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 496 497#ifdef MAC 498 mac_cleanup_sysv_msgqueue(msqkptr); 499#endif 500 501 wakeup(msqkptr); 502 } 503 504 break; 505 506 case IPC_SET: 507 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 508 goto done2; 509 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 510 error = suser(td); 511 if (error) 512 goto done2; 513 } 514 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 515 DPRINTF(("can't increase msg_qbytes beyond %d" 516 "(truncating)\n", msginfo.msgmnb)); 517 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 518 } 519 if (msqbuf->msg_qbytes == 0) { 520 DPRINTF(("can't reduce msg_qbytes to 0\n")); 521 error = EINVAL; /* non-standard errno! */ 522 goto done2; 523 } 524 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 525 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 526 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 527 (msqbuf->msg_perm.mode & 0777); 528 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 529 msqkptr->u.msg_ctime = time_second; 530 break; 531 532 case IPC_STAT: 533 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 534 DPRINTF(("requester doesn't have read access\n")); 535 goto done2; 536 } 537 *msqbuf = msqkptr->u; 538 break; 539 540 default: 541 DPRINTF(("invalid command %d\n", cmd)); 542 error = EINVAL; 543 goto done2; 544 } 545 546 if (error == 0) 547 td->td_retval[0] = rval; 548done2: 549 mtx_unlock(&msq_mtx); 550 return (error); 551} 552 553#ifndef _SYS_SYSPROTO_H_ 554struct msgget_args { 555 key_t key; 556 int msgflg; 557}; 558#endif 559 560/* 561 * MPSAFE 562 */ 563int 564msgget(td, uap) 565 struct thread *td; 566 register struct msgget_args *uap; 567{ 568 int msqid, error = 0; 569 int key = uap->key; 570 int msgflg = uap->msgflg; 571 struct ucred *cred = td->td_ucred; 572 register struct msqid_kernel *msqkptr = NULL; 573 574 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 575 576 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 577 return (ENOSYS); 578 579 mtx_lock(&msq_mtx); 580 if (key != IPC_PRIVATE) { 581 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 582 msqkptr = &msqids[msqid]; 583 if (msqkptr->u.msg_qbytes != 0 && 584 msqkptr->u.msg_perm.key == key) 585 break; 586 } 587 if (msqid < msginfo.msgmni) { 588 DPRINTF(("found public key\n")); 589 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 590 DPRINTF(("not exclusive\n")); 591 error = EEXIST; 592 goto done2; 593 } 594 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 595 msgflg & 0700))) { 596 DPRINTF(("requester doesn't have 0%o access\n", 597 msgflg & 0700)); 598 goto done2; 599 } 600#ifdef MAC 601 error = mac_check_sysv_msqget(cred, msqkptr); 602 if (error != 0) 603 goto done2; 604#endif 605 goto found; 606 } 607 } 608 609 DPRINTF(("need to allocate the msqid_ds\n")); 610 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 611 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 612 /* 613 * Look for an unallocated and unlocked msqid_ds. 614 * msqid_ds's can be locked by msgsnd or msgrcv while 615 * they are copying the message in/out. We can't 616 * re-use the entry until they release it. 617 */ 618 msqkptr = &msqids[msqid]; 619 if (msqkptr->u.msg_qbytes == 0 && 620 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 621 break; 622 } 623 if (msqid == msginfo.msgmni) { 624 DPRINTF(("no more msqid_ds's available\n")); 625 error = ENOSPC; 626 goto done2; 627 } 628 DPRINTF(("msqid %d is available\n", msqid)); 629 msqkptr->u.msg_perm.key = key; 630 msqkptr->u.msg_perm.cuid = cred->cr_uid; 631 msqkptr->u.msg_perm.uid = cred->cr_uid; 632 msqkptr->u.msg_perm.cgid = cred->cr_gid; 633 msqkptr->u.msg_perm.gid = cred->cr_gid; 634 msqkptr->u.msg_perm.mode = (msgflg & 0777); 635 /* Make sure that the returned msqid is unique */ 636 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 637 msqkptr->u.msg_first = NULL; 638 msqkptr->u.msg_last = NULL; 639 msqkptr->u.msg_cbytes = 0; 640 msqkptr->u.msg_qnum = 0; 641 msqkptr->u.msg_qbytes = msginfo.msgmnb; 642 msqkptr->u.msg_lspid = 0; 643 msqkptr->u.msg_lrpid = 0; 644 msqkptr->u.msg_stime = 0; 645 msqkptr->u.msg_rtime = 0; 646 msqkptr->u.msg_ctime = time_second; 647#ifdef MAC 648 mac_create_sysv_msgqueue(cred, msqkptr); 649#endif 650 } else { 651 DPRINTF(("didn't find it and wasn't asked to create it\n")); 652 error = ENOENT; 653 goto done2; 654 } 655 656found: 657 /* Construct the unique msqid */ 658 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 659done2: 660 mtx_unlock(&msq_mtx); 661 return (error); 662} 663 664#ifndef _SYS_SYSPROTO_H_ 665struct msgsnd_args { 666 int msqid; 667 const void *msgp; 668 size_t msgsz; 669 int msgflg; 670}; 671#endif 672 673/* 674 * MPSAFE 675 */ 676int 677msgsnd(td, uap) 678 struct thread *td; 679 register struct msgsnd_args *uap; 680{ 681 int msqid = uap->msqid; 682 const void *user_msgp = uap->msgp; 683 size_t msgsz = uap->msgsz; 684 int msgflg = uap->msgflg; 685 int segs_needed, error = 0; 686 register struct msqid_kernel *msqkptr; 687 register struct msg *msghdr; 688 short next; 689 690 DPRINTF(("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz, 691 msgflg)); 692 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 693 return (ENOSYS); 694 695 mtx_lock(&msq_mtx); 696 msqid = IPCID_TO_IX(msqid); 697 698 if (msqid < 0 || msqid >= msginfo.msgmni) { 699 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 700 msginfo.msgmni)); 701 error = EINVAL; 702 goto done2; 703 } 704 705 msqkptr = &msqids[msqid]; 706 if (msqkptr->u.msg_qbytes == 0) { 707 DPRINTF(("no such message queue id\n")); 708 error = EINVAL; 709 goto done2; 710 } 711 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 712 DPRINTF(("wrong sequence number\n")); 713 error = EINVAL; 714 goto done2; 715 } 716 717 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 718 DPRINTF(("requester doesn't have write access\n")); 719 goto done2; 720 } 721 722#ifdef MAC 723 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr); 724 if (error != 0) 725 goto done2; 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 msg_freehdr(msghdr); 968 wakeup(msqkptr); 969 goto done2; 970 } 971#endif 972 973 /* 974 * Put the message into the queue 975 */ 976 if (msqkptr->u.msg_first == NULL) { 977 msqkptr->u.msg_first = msghdr; 978 msqkptr->u.msg_last = msghdr; 979 } else { 980 msqkptr->u.msg_last->msg_next = msghdr; 981 msqkptr->u.msg_last = msghdr; 982 } 983 msqkptr->u.msg_last->msg_next = NULL; 984 985 msqkptr->u.msg_cbytes += msghdr->msg_ts; 986 msqkptr->u.msg_qnum++; 987 msqkptr->u.msg_lspid = td->td_proc->p_pid; 988 msqkptr->u.msg_stime = time_second; 989 990 wakeup(msqkptr); 991 td->td_retval[0] = 0; 992done2: 993 mtx_unlock(&msq_mtx); 994 return (error); 995} 996 997#ifndef _SYS_SYSPROTO_H_ 998struct msgrcv_args { 999 int msqid; 1000 void *msgp; 1001 size_t msgsz; 1002 long msgtyp; 1003 int msgflg; 1004}; 1005#endif 1006 1007/* 1008 * MPSAFE 1009 */ 1010int 1011msgrcv(td, uap) 1012 struct thread *td; 1013 register struct msgrcv_args *uap; 1014{ 1015 int msqid = uap->msqid; 1016 void *user_msgp = uap->msgp; 1017 size_t msgsz = uap->msgsz; 1018 long msgtyp = uap->msgtyp; 1019 int msgflg = uap->msgflg; 1020 size_t len; 1021 register struct msqid_kernel *msqkptr; 1022 register struct msg *msghdr; 1023 int error = 0; 1024 short next; 1025 1026 DPRINTF(("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp, 1027 msgsz, msgtyp, msgflg)); 1028 1029 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 1030 return (ENOSYS); 1031 1032 msqid = IPCID_TO_IX(msqid); 1033 1034 if (msqid < 0 || msqid >= msginfo.msgmni) { 1035 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 1036 msginfo.msgmni)); 1037 return (EINVAL); 1038 } 1039 1040 msqkptr = &msqids[msqid]; 1041 mtx_lock(&msq_mtx); 1042 if (msqkptr->u.msg_qbytes == 0) { 1043 DPRINTF(("no such message queue id\n")); 1044 error = EINVAL; 1045 goto done2; 1046 } 1047 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1048 DPRINTF(("wrong sequence number\n")); 1049 error = EINVAL; 1050 goto done2; 1051 } 1052 1053 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1054 DPRINTF(("requester doesn't have read access\n")); 1055 goto done2; 1056 } 1057 1058#ifdef MAC 1059 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr); 1060 if (error != 0) 1061 goto done2; 1062#endif 1063 1064 msghdr = NULL; 1065 while (msghdr == NULL) { 1066 if (msgtyp == 0) { 1067 msghdr = msqkptr->u.msg_first; 1068 if (msghdr != NULL) { 1069 if (msgsz < msghdr->msg_ts && 1070 (msgflg & MSG_NOERROR) == 0) { 1071 DPRINTF(("first message on the queue " 1072 "is too big (want %d, got %d)\n", 1073 msgsz, msghdr->msg_ts)); 1074 error = E2BIG; 1075 goto done2; 1076 } 1077#ifdef MAC 1078 error = mac_check_sysv_msgrcv(td->td_ucred, 1079 msghdr); 1080 if (error != 0) 1081 goto done2; 1082#endif 1083 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1084 msqkptr->u.msg_first = NULL; 1085 msqkptr->u.msg_last = NULL; 1086 } else { 1087 msqkptr->u.msg_first = msghdr->msg_next; 1088 if (msqkptr->u.msg_first == NULL) 1089 panic("msg_first/last screwed up #1"); 1090 } 1091 } 1092 } else { 1093 struct msg *previous; 1094 struct msg **prev; 1095 1096 previous = NULL; 1097 prev = &(msqkptr->u.msg_first); 1098 while ((msghdr = *prev) != NULL) { 1099 /* 1100 * Is this message's type an exact match or is 1101 * this message's type less than or equal to 1102 * the absolute value of a negative msgtyp? 1103 * Note that the second half of this test can 1104 * NEVER be true if msgtyp is positive since 1105 * msg_type is always positive! 1106 */ 1107 1108 if (msgtyp == msghdr->msg_type || 1109 msghdr->msg_type <= -msgtyp) { 1110 DPRINTF(("found message type %d, " 1111 "requested %d\n", 1112 msghdr->msg_type, msgtyp)); 1113 if (msgsz < msghdr->msg_ts && 1114 (msgflg & MSG_NOERROR) == 0) { 1115 DPRINTF(("requested message " 1116 "on the queue is too big " 1117 "(want %d, got %d)\n", 1118 msgsz, msghdr->msg_ts)); 1119 error = E2BIG; 1120 goto done2; 1121 } 1122#ifdef MAC 1123 error = mac_check_sysv_msgrcv( 1124 td->td_ucred, msghdr); 1125 if (error != 0) 1126 goto done2; 1127#endif 1128 *prev = msghdr->msg_next; 1129 if (msghdr == msqkptr->u.msg_last) { 1130 if (previous == NULL) { 1131 if (prev != 1132 &msqkptr->u.msg_first) 1133 panic("msg_first/last screwed up #2"); 1134 msqkptr->u.msg_first = 1135 NULL; 1136 msqkptr->u.msg_last = 1137 NULL; 1138 } else { 1139 if (prev == 1140 &msqkptr->u.msg_first) 1141 panic("msg_first/last screwed up #3"); 1142 msqkptr->u.msg_last = 1143 previous; 1144 } 1145 } 1146 break; 1147 } 1148 previous = msghdr; 1149 prev = &(msghdr->msg_next); 1150 } 1151 } 1152 1153 /* 1154 * We've either extracted the msghdr for the appropriate 1155 * message or there isn't one. 1156 * If there is one then bail out of this loop. 1157 */ 1158 1159 if (msghdr != NULL) 1160 break; 1161 1162 /* 1163 * Hmph! No message found. Does the user want to wait? 1164 */ 1165 1166 if ((msgflg & IPC_NOWAIT) != 0) { 1167 DPRINTF(("no appropriate message found (msgtyp=%d)\n", 1168 msgtyp)); 1169 /* The SVID says to return ENOMSG. */ 1170 error = ENOMSG; 1171 goto done2; 1172 } 1173 1174 /* 1175 * Wait for something to happen 1176 */ 1177 1178 DPRINTF(("msgrcv: goodnight\n")); 1179 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1180 "msgwait", 0); 1181 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1182 1183 if (error != 0) { 1184 DPRINTF(("msgsnd: interrupted system call\n")); 1185 error = EINTR; 1186 goto done2; 1187 } 1188 1189 /* 1190 * Make sure that the msq queue still exists 1191 */ 1192 1193 if (msqkptr->u.msg_qbytes == 0 || 1194 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1195 DPRINTF(("msqid deleted\n")); 1196 error = EIDRM; 1197 goto done2; 1198 } 1199 } 1200 1201 /* 1202 * Return the message to the user. 1203 * 1204 * First, do the bookkeeping (before we risk being interrupted). 1205 */ 1206 1207 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1208 msqkptr->u.msg_qnum--; 1209 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1210 msqkptr->u.msg_rtime = time_second; 1211 1212 /* 1213 * Make msgsz the actual amount that we'll be returning. 1214 * Note that this effectively truncates the message if it is too long 1215 * (since msgsz is never increased). 1216 */ 1217 1218 DPRINTF(("found a message, msgsz=%d, msg_ts=%d\n", msgsz, 1219 msghdr->msg_ts)); 1220 if (msgsz > msghdr->msg_ts) 1221 msgsz = msghdr->msg_ts; 1222 1223 /* 1224 * Return the type to the user. 1225 */ 1226 1227 mtx_unlock(&msq_mtx); 1228 error = copyout(&(msghdr->msg_type), user_msgp, 1229 sizeof(msghdr->msg_type)); 1230 mtx_lock(&msq_mtx); 1231 if (error != 0) { 1232 DPRINTF(("error (%d) copying out message type\n", error)); 1233 msg_freehdr(msghdr); 1234 wakeup(msqkptr); 1235 goto done2; 1236 } 1237 user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type); 1238 1239 /* 1240 * Return the segments to the user 1241 */ 1242 1243 next = msghdr->msg_spot; 1244 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1245 size_t tlen; 1246 1247 if (msgsz - len > msginfo.msgssz) 1248 tlen = msginfo.msgssz; 1249 else 1250 tlen = msgsz - len; 1251 if (next <= -1) 1252 panic("next too low #3"); 1253 if (next >= msginfo.msgseg) 1254 panic("next out of range #3"); 1255 mtx_unlock(&msq_mtx); 1256 error = copyout(&msgpool[next * msginfo.msgssz], 1257 user_msgp, tlen); 1258 mtx_lock(&msq_mtx); 1259 if (error != 0) { 1260 DPRINTF(("error (%d) copying out message segment\n", 1261 error)); 1262 msg_freehdr(msghdr); 1263 wakeup(msqkptr); 1264 goto done2; 1265 } 1266 user_msgp = (char *)user_msgp + tlen; 1267 next = msgmaps[next].next; 1268 } 1269 1270 /* 1271 * Done, return the actual number of bytes copied out. 1272 */ 1273 1274 msg_freehdr(msghdr); 1275 wakeup(msqkptr); 1276 td->td_retval[0] = msgsz; 1277done2: 1278 mtx_unlock(&msq_mtx); 1279 return (error); 1280} 1281 1282static int 1283sysctl_msqids(SYSCTL_HANDLER_ARGS) 1284{ 1285 1286 return (SYSCTL_OUT(req, msqids, 1287 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1288} 1289 1290SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1291 "Maximum message size"); 1292SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1293 "Number of message queue identifiers"); 1294SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1295 "Maximum number of bytes in a queue"); 1296SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1297 "Maximum number of messages in the system"); 1298SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1299 "Size of a message segment"); 1300SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1301 "Number of message segments"); 1302SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1303 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1304