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("$FreeBSD$"); 52 53 #include "opt_sysvipc.h" 54 55 #include <sys/param.h> 56 #include <sys/systm.h> 57 #include <sys/sysproto.h> 58 #include <sys/kernel.h> 59 #include <sys/priv.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 74 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 75 76 static void msginit(void); 77 static int msgunload(void); 78 static 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 86 static void msg_freehdr(struct msg *msghdr); 87 88 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 89 static 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 */ 121 struct 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 */ 142 143 struct 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? */ 150 151 static int nfree_msgmaps; /* # of free map entries */ 152 static short free_msgmaps; /* head of linked list of free map entries */ 153 static struct msg *free_msghdrs;/* list of free msg headers */ 154 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 155 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 156 static struct msg *msghdrs; /* MSGTQL msg headers */ 157 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 158 static struct mtx msq_mtx; /* global mutex for message queues. */ 159 160 static void 161 msginit() 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_sysvmsg_init(&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_sysvmsq_init(&msqids[i]); 240 #endif 241 } 242 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 243 } 244 245 static int 246 msgunload() 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_sysvmsg_destroy(&msghdrs[i]); 272 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 273 mac_sysvmsq_destroy(&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 } 282 283 284 static int 285 sysvmsg_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 } 304 305 static moduledata_t sysvmsg_mod = { 306 "sysvmsg", 307 &sysvmsg_modload, 308 NULL 309 }; 310 311 SYSCALL_MODULE_HELPER(msgsys); 312 SYSCALL_MODULE_HELPER(msgctl); 313 SYSCALL_MODULE_HELPER(msgget); 314 SYSCALL_MODULE_HELPER(msgsnd); 315 SYSCALL_MODULE_HELPER(msgrcv); 316 317 DECLARE_MODULE(sysvmsg, sysvmsg_mod, 318 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 319 MODULE_VERSION(sysvmsg, 1); 320 321 /* 322 * Entry point for all MSG calls. 323 */ 324 int 325 msgsys(td, uap) 326 struct thread *td; 327 /* XXX actually varargs. */ 328 struct msgsys_args /* { 329 int which; 330 int a2; 331 int a3; 332 int a4; 333 int a5; 334 int a6; 335 } */ *uap; 336 { 337 int error; 338 339 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 340 return (ENOSYS); 341 if (uap->which < 0 || 342 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 343 return (EINVAL); 344 error = (*msgcalls[uap->which])(td, &uap->a2); 345 return (error); 346 } 347 348 static void 349 msg_freehdr(msghdr) 350 struct msg *msghdr; 351 { 352 while (msghdr->msg_ts > 0) { 353 short next; 354 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 355 panic("msghdr->msg_spot out of range"); 356 next = msgmaps[msghdr->msg_spot].next; 357 msgmaps[msghdr->msg_spot].next = free_msgmaps; 358 free_msgmaps = msghdr->msg_spot; 359 nfree_msgmaps++; 360 msghdr->msg_spot = next; 361 if (msghdr->msg_ts >= msginfo.msgssz) 362 msghdr->msg_ts -= msginfo.msgssz; 363 else 364 msghdr->msg_ts = 0; 365 } 366 if (msghdr->msg_spot != -1) 367 panic("msghdr->msg_spot != -1"); 368 msghdr->msg_next = free_msghdrs; 369 free_msghdrs = msghdr; 370 #ifdef MAC 371 mac_sysvmsg_cleanup(msghdr); 372 #endif 373 } 374 375 #ifndef _SYS_SYSPROTO_H_ 376 struct msgctl_args { 377 int msqid; 378 int cmd; 379 struct msqid_ds *buf; 380 }; 381 #endif 382 int 383 msgctl(td, uap) 384 struct thread *td; 385 register struct msgctl_args *uap; 386 { 387 int msqid = uap->msqid; 388 int cmd = uap->cmd; 389 struct msqid_ds msqbuf; 390 int error; 391 392 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 393 if (cmd == IPC_SET && 394 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 395 return (error); 396 error = kern_msgctl(td, msqid, cmd, &msqbuf); 397 if (cmd == IPC_STAT && error == 0) 398 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 399 return (error); 400 } 401 402 int 403 kern_msgctl(td, msqid, cmd, msqbuf) 404 struct thread *td; 405 int msqid; 406 int cmd; 407 struct msqid_ds *msqbuf; 408 { 409 int rval, error, msqix; 410 register struct msqid_kernel *msqkptr; 411 412 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 413 return (ENOSYS); 414 415 msqix = IPCID_TO_IX(msqid); 416 417 if (msqix < 0 || msqix >= msginfo.msgmni) { 418 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 419 msginfo.msgmni)); 420 return (EINVAL); 421 } 422 423 msqkptr = &msqids[msqix]; 424 425 mtx_lock(&msq_mtx); 426 if (msqkptr->u.msg_qbytes == 0) { 427 DPRINTF(("no such msqid\n")); 428 error = EINVAL; 429 goto done2; 430 } 431 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 432 DPRINTF(("wrong sequence number\n")); 433 error = EINVAL; 434 goto done2; 435 } 436 #ifdef MAC 437 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 438 if (error != 0) 439 goto done2; 440 #endif 441 442 error = 0; 443 rval = 0; 444 445 switch (cmd) { 446 447 case IPC_RMID: 448 { 449 struct msg *msghdr; 450 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 451 goto done2; 452 453 #ifdef MAC 454 /* 455 * Check that the thread has MAC access permissions to 456 * individual msghdrs. Note: We need to do this in a 457 * separate loop because the actual loop alters the 458 * msq/msghdr info as it progresses, and there is no going 459 * back if half the way through we discover that the 460 * thread cannot free a certain msghdr. The msq will get 461 * into an inconsistent state. 462 */ 463 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 464 msghdr = msghdr->msg_next) { 465 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 466 if (error != 0) 467 goto done2; 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_sysvmsq_cleanup(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 = priv_check(td, PRIV_IPC_MSGSIZE); 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 *msqbuf = msqkptr->u; 532 break; 533 534 default: 535 DPRINTF(("invalid command %d\n", cmd)); 536 error = EINVAL; 537 goto done2; 538 } 539 540 if (error == 0) 541 td->td_retval[0] = rval; 542 done2: 543 mtx_unlock(&msq_mtx); 544 return (error); 545 } 546 547 #ifndef _SYS_SYSPROTO_H_ 548 struct msgget_args { 549 key_t key; 550 int msgflg; 551 }; 552 #endif 553 int 554 msgget(td, uap) 555 struct thread *td; 556 register struct msgget_args *uap; 557 { 558 int msqid, error = 0; 559 int key = uap->key; 560 int msgflg = uap->msgflg; 561 struct ucred *cred = td->td_ucred; 562 register struct msqid_kernel *msqkptr = NULL; 563 564 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 565 566 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 567 return (ENOSYS); 568 569 mtx_lock(&msq_mtx); 570 if (key != IPC_PRIVATE) { 571 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 572 msqkptr = &msqids[msqid]; 573 if (msqkptr->u.msg_qbytes != 0 && 574 msqkptr->u.msg_perm.key == key) 575 break; 576 } 577 if (msqid < msginfo.msgmni) { 578 DPRINTF(("found public key\n")); 579 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 580 DPRINTF(("not exclusive\n")); 581 error = EEXIST; 582 goto done2; 583 } 584 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 585 msgflg & 0700))) { 586 DPRINTF(("requester doesn't have 0%o access\n", 587 msgflg & 0700)); 588 goto done2; 589 } 590 #ifdef MAC 591 error = mac_sysvmsq_check_msqget(cred, msqkptr); 592 if (error != 0) 593 goto done2; 594 #endif 595 goto found; 596 } 597 } 598 599 DPRINTF(("need to allocate the msqid_ds\n")); 600 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 601 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 602 /* 603 * Look for an unallocated and unlocked msqid_ds. 604 * msqid_ds's can be locked by msgsnd or msgrcv while 605 * they are copying the message in/out. We can't 606 * re-use the entry until they release it. 607 */ 608 msqkptr = &msqids[msqid]; 609 if (msqkptr->u.msg_qbytes == 0 && 610 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 611 break; 612 } 613 if (msqid == msginfo.msgmni) { 614 DPRINTF(("no more msqid_ds's available\n")); 615 error = ENOSPC; 616 goto done2; 617 } 618 DPRINTF(("msqid %d is available\n", msqid)); 619 msqkptr->u.msg_perm.key = key; 620 msqkptr->u.msg_perm.cuid = cred->cr_uid; 621 msqkptr->u.msg_perm.uid = cred->cr_uid; 622 msqkptr->u.msg_perm.cgid = cred->cr_gid; 623 msqkptr->u.msg_perm.gid = cred->cr_gid; 624 msqkptr->u.msg_perm.mode = (msgflg & 0777); 625 /* Make sure that the returned msqid is unique */ 626 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 627 msqkptr->u.msg_first = NULL; 628 msqkptr->u.msg_last = NULL; 629 msqkptr->u.msg_cbytes = 0; 630 msqkptr->u.msg_qnum = 0; 631 msqkptr->u.msg_qbytes = msginfo.msgmnb; 632 msqkptr->u.msg_lspid = 0; 633 msqkptr->u.msg_lrpid = 0; 634 msqkptr->u.msg_stime = 0; 635 msqkptr->u.msg_rtime = 0; 636 msqkptr->u.msg_ctime = time_second; 637 #ifdef MAC 638 mac_sysvmsq_create(cred, msqkptr); 639 #endif 640 } else { 641 DPRINTF(("didn't find it and wasn't asked to create it\n")); 642 error = ENOENT; 643 goto done2; 644 } 645 646 found: 647 /* Construct the unique msqid */ 648 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 649 done2: 650 mtx_unlock(&msq_mtx); 651 return (error); 652 } 653 654 #ifndef _SYS_SYSPROTO_H_ 655 struct msgsnd_args { 656 int msqid; 657 const void *msgp; 658 size_t msgsz; 659 int msgflg; 660 }; 661 #endif 662 int 663 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 664 struct thread *td; 665 int msqid; 666 const void *msgp; /* XXX msgp is actually mtext. */ 667 size_t msgsz; 668 int msgflg; 669 long mtype; 670 { 671 int msqix, segs_needed, error = 0; 672 register struct msqid_kernel *msqkptr; 673 register struct msg *msghdr; 674 short next; 675 676 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 677 return (ENOSYS); 678 679 mtx_lock(&msq_mtx); 680 msqix = IPCID_TO_IX(msqid); 681 682 if (msqix < 0 || msqix >= msginfo.msgmni) { 683 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 684 msginfo.msgmni)); 685 error = EINVAL; 686 goto done2; 687 } 688 689 msqkptr = &msqids[msqix]; 690 if (msqkptr->u.msg_qbytes == 0) { 691 DPRINTF(("no such message queue id\n")); 692 error = EINVAL; 693 goto done2; 694 } 695 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 696 DPRINTF(("wrong sequence number\n")); 697 error = EINVAL; 698 goto done2; 699 } 700 701 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 702 DPRINTF(("requester doesn't have write access\n")); 703 goto done2; 704 } 705 706 #ifdef MAC 707 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 708 if (error != 0) 709 goto done2; 710 #endif 711 712 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 713 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 714 msginfo.msgssz, segs_needed)); 715 for (;;) { 716 int need_more_resources = 0; 717 718 /* 719 * check msgsz 720 * (inside this loop in case msg_qbytes changes while we sleep) 721 */ 722 723 if (msgsz > msqkptr->u.msg_qbytes) { 724 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 725 error = EINVAL; 726 goto done2; 727 } 728 729 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 730 DPRINTF(("msqid is locked\n")); 731 need_more_resources = 1; 732 } 733 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 734 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 735 need_more_resources = 1; 736 } 737 if (segs_needed > nfree_msgmaps) { 738 DPRINTF(("segs_needed > nfree_msgmaps\n")); 739 need_more_resources = 1; 740 } 741 if (free_msghdrs == NULL) { 742 DPRINTF(("no more msghdrs\n")); 743 need_more_resources = 1; 744 } 745 746 if (need_more_resources) { 747 int we_own_it; 748 749 if ((msgflg & IPC_NOWAIT) != 0) { 750 DPRINTF(("need more resources but caller " 751 "doesn't want to wait\n")); 752 error = EAGAIN; 753 goto done2; 754 } 755 756 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 757 DPRINTF(("we don't own the msqid_ds\n")); 758 we_own_it = 0; 759 } else { 760 /* Force later arrivals to wait for our 761 request */ 762 DPRINTF(("we own the msqid_ds\n")); 763 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 764 we_own_it = 1; 765 } 766 DPRINTF(("msgsnd: goodnight\n")); 767 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 768 "msgsnd", hz); 769 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 770 if (we_own_it) 771 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 772 if (error == EWOULDBLOCK) { 773 DPRINTF(("msgsnd: timed out\n")); 774 continue; 775 } 776 if (error != 0) { 777 DPRINTF(("msgsnd: interrupted system call\n")); 778 error = EINTR; 779 goto done2; 780 } 781 782 /* 783 * Make sure that the msq queue still exists 784 */ 785 786 if (msqkptr->u.msg_qbytes == 0) { 787 DPRINTF(("msqid deleted\n")); 788 error = EIDRM; 789 goto done2; 790 } 791 792 } else { 793 DPRINTF(("got all the resources that we need\n")); 794 break; 795 } 796 } 797 798 /* 799 * We have the resources that we need. 800 * Make sure! 801 */ 802 803 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 804 panic("msg_perm.mode & MSG_LOCKED"); 805 if (segs_needed > nfree_msgmaps) 806 panic("segs_needed > nfree_msgmaps"); 807 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 808 panic("msgsz + msg_cbytes > msg_qbytes"); 809 if (free_msghdrs == NULL) 810 panic("no more msghdrs"); 811 812 /* 813 * Re-lock the msqid_ds in case we page-fault when copying in the 814 * message 815 */ 816 817 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 818 panic("msqid_ds is already locked"); 819 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 820 821 /* 822 * Allocate a message header 823 */ 824 825 msghdr = free_msghdrs; 826 free_msghdrs = msghdr->msg_next; 827 msghdr->msg_spot = -1; 828 msghdr->msg_ts = msgsz; 829 msghdr->msg_type = mtype; 830 #ifdef MAC 831 /* 832 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 833 * immediately? Or, should it be checked just before the msg is 834 * enqueued in the msgq (as it is done now)? 835 */ 836 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 837 #endif 838 839 /* 840 * Allocate space for the message 841 */ 842 843 while (segs_needed > 0) { 844 if (nfree_msgmaps <= 0) 845 panic("not enough msgmaps"); 846 if (free_msgmaps == -1) 847 panic("nil free_msgmaps"); 848 next = free_msgmaps; 849 if (next <= -1) 850 panic("next too low #1"); 851 if (next >= msginfo.msgseg) 852 panic("next out of range #1"); 853 DPRINTF(("allocating segment %d to message\n", next)); 854 free_msgmaps = msgmaps[next].next; 855 nfree_msgmaps--; 856 msgmaps[next].next = msghdr->msg_spot; 857 msghdr->msg_spot = next; 858 segs_needed--; 859 } 860 861 /* 862 * Validate the message type 863 */ 864 865 if (msghdr->msg_type < 1) { 866 msg_freehdr(msghdr); 867 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 868 wakeup(msqkptr); 869 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 870 error = EINVAL; 871 goto done2; 872 } 873 874 /* 875 * Copy in the message body 876 */ 877 878 next = msghdr->msg_spot; 879 while (msgsz > 0) { 880 size_t tlen; 881 if (msgsz > msginfo.msgssz) 882 tlen = msginfo.msgssz; 883 else 884 tlen = msgsz; 885 if (next <= -1) 886 panic("next too low #2"); 887 if (next >= msginfo.msgseg) 888 panic("next out of range #2"); 889 mtx_unlock(&msq_mtx); 890 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 891 tlen)) != 0) { 892 mtx_lock(&msq_mtx); 893 DPRINTF(("error %d copying in message segment\n", 894 error)); 895 msg_freehdr(msghdr); 896 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 897 wakeup(msqkptr); 898 goto done2; 899 } 900 mtx_lock(&msq_mtx); 901 msgsz -= tlen; 902 msgp = (const char *)msgp + tlen; 903 next = msgmaps[next].next; 904 } 905 if (next != -1) 906 panic("didn't use all the msg segments"); 907 908 /* 909 * We've got the message. Unlock the msqid_ds. 910 */ 911 912 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 913 914 /* 915 * Make sure that the msqid_ds is still allocated. 916 */ 917 918 if (msqkptr->u.msg_qbytes == 0) { 919 msg_freehdr(msghdr); 920 wakeup(msqkptr); 921 error = EIDRM; 922 goto done2; 923 } 924 925 #ifdef MAC 926 /* 927 * Note: Since the task/thread allocates the msghdr and usually 928 * primes it with its own MAC label, for a majority of policies, it 929 * won't be necessary to check whether the msghdr has access 930 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 931 * suffice in that case. However, this hook may be required where 932 * individual policies derive a non-identical label for the msghdr 933 * from the current thread label and may want to check the msghdr 934 * enqueue permissions, along with read/write permissions to the 935 * msgq. 936 */ 937 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 938 if (error != 0) { 939 msg_freehdr(msghdr); 940 wakeup(msqkptr); 941 goto done2; 942 } 943 #endif 944 945 /* 946 * Put the message into the queue 947 */ 948 if (msqkptr->u.msg_first == NULL) { 949 msqkptr->u.msg_first = msghdr; 950 msqkptr->u.msg_last = msghdr; 951 } else { 952 msqkptr->u.msg_last->msg_next = msghdr; 953 msqkptr->u.msg_last = msghdr; 954 } 955 msqkptr->u.msg_last->msg_next = NULL; 956 957 msqkptr->u.msg_cbytes += msghdr->msg_ts; 958 msqkptr->u.msg_qnum++; 959 msqkptr->u.msg_lspid = td->td_proc->p_pid; 960 msqkptr->u.msg_stime = time_second; 961 962 wakeup(msqkptr); 963 td->td_retval[0] = 0; 964 done2: 965 mtx_unlock(&msq_mtx); 966 return (error); 967 } 968 969 int 970 msgsnd(td, uap) 971 struct thread *td; 972 register struct msgsnd_args *uap; 973 { 974 int error; 975 long mtype; 976 977 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 978 uap->msgsz, uap->msgflg)); 979 980 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 981 DPRINTF(("error %d copying the message type\n", error)); 982 return (error); 983 } 984 return (kern_msgsnd(td, uap->msqid, 985 (const char *)uap->msgp + sizeof(mtype), 986 uap->msgsz, uap->msgflg, mtype)); 987 } 988 989 #ifndef _SYS_SYSPROTO_H_ 990 struct msgrcv_args { 991 int msqid; 992 void *msgp; 993 size_t msgsz; 994 long msgtyp; 995 int msgflg; 996 }; 997 #endif 998 int 999 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1000 struct thread *td; 1001 int msqid; 1002 void *msgp; /* XXX msgp is actually mtext. */ 1003 size_t msgsz; 1004 long msgtyp; 1005 int msgflg; 1006 long *mtype; 1007 { 1008 size_t len; 1009 register struct msqid_kernel *msqkptr; 1010 register struct msg *msghdr; 1011 int msqix, error = 0; 1012 short next; 1013 1014 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 1015 return (ENOSYS); 1016 1017 msqix = IPCID_TO_IX(msqid); 1018 1019 if (msqix < 0 || msqix >= msginfo.msgmni) { 1020 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1021 msginfo.msgmni)); 1022 return (EINVAL); 1023 } 1024 1025 msqkptr = &msqids[msqix]; 1026 mtx_lock(&msq_mtx); 1027 if (msqkptr->u.msg_qbytes == 0) { 1028 DPRINTF(("no such message queue id\n")); 1029 error = EINVAL; 1030 goto done2; 1031 } 1032 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1033 DPRINTF(("wrong sequence number\n")); 1034 error = EINVAL; 1035 goto done2; 1036 } 1037 1038 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1039 DPRINTF(("requester doesn't have read access\n")); 1040 goto done2; 1041 } 1042 1043 #ifdef MAC 1044 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1045 if (error != 0) 1046 goto done2; 1047 #endif 1048 1049 msghdr = NULL; 1050 while (msghdr == NULL) { 1051 if (msgtyp == 0) { 1052 msghdr = msqkptr->u.msg_first; 1053 if (msghdr != NULL) { 1054 if (msgsz < msghdr->msg_ts && 1055 (msgflg & MSG_NOERROR) == 0) { 1056 DPRINTF(("first message on the queue " 1057 "is too big (want %zu, got %d)\n", 1058 msgsz, msghdr->msg_ts)); 1059 error = E2BIG; 1060 goto done2; 1061 } 1062 #ifdef MAC 1063 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1064 msghdr); 1065 if (error != 0) 1066 goto done2; 1067 #endif 1068 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1069 msqkptr->u.msg_first = NULL; 1070 msqkptr->u.msg_last = NULL; 1071 } else { 1072 msqkptr->u.msg_first = msghdr->msg_next; 1073 if (msqkptr->u.msg_first == NULL) 1074 panic("msg_first/last screwed up #1"); 1075 } 1076 } 1077 } else { 1078 struct msg *previous; 1079 struct msg **prev; 1080 1081 previous = NULL; 1082 prev = &(msqkptr->u.msg_first); 1083 while ((msghdr = *prev) != NULL) { 1084 /* 1085 * Is this message's type an exact match or is 1086 * this message's type less than or equal to 1087 * the absolute value of a negative msgtyp? 1088 * Note that the second half of this test can 1089 * NEVER be true if msgtyp is positive since 1090 * msg_type is always positive! 1091 */ 1092 1093 if (msgtyp == msghdr->msg_type || 1094 msghdr->msg_type <= -msgtyp) { 1095 DPRINTF(("found message type %ld, " 1096 "requested %ld\n", 1097 msghdr->msg_type, msgtyp)); 1098 if (msgsz < msghdr->msg_ts && 1099 (msgflg & MSG_NOERROR) == 0) { 1100 DPRINTF(("requested message " 1101 "on the queue is too big " 1102 "(want %zu, got %hu)\n", 1103 msgsz, msghdr->msg_ts)); 1104 error = E2BIG; 1105 goto done2; 1106 } 1107 #ifdef MAC 1108 error = mac_sysvmsq_check_msgrcv( 1109 td->td_ucred, msghdr); 1110 if (error != 0) 1111 goto done2; 1112 #endif 1113 *prev = msghdr->msg_next; 1114 if (msghdr == msqkptr->u.msg_last) { 1115 if (previous == NULL) { 1116 if (prev != 1117 &msqkptr->u.msg_first) 1118 panic("msg_first/last screwed up #2"); 1119 msqkptr->u.msg_first = 1120 NULL; 1121 msqkptr->u.msg_last = 1122 NULL; 1123 } else { 1124 if (prev == 1125 &msqkptr->u.msg_first) 1126 panic("msg_first/last screwed up #3"); 1127 msqkptr->u.msg_last = 1128 previous; 1129 } 1130 } 1131 break; 1132 } 1133 previous = msghdr; 1134 prev = &(msghdr->msg_next); 1135 } 1136 } 1137 1138 /* 1139 * We've either extracted the msghdr for the appropriate 1140 * message or there isn't one. 1141 * If there is one then bail out of this loop. 1142 */ 1143 1144 if (msghdr != NULL) 1145 break; 1146 1147 /* 1148 * Hmph! No message found. Does the user want to wait? 1149 */ 1150 1151 if ((msgflg & IPC_NOWAIT) != 0) { 1152 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1153 msgtyp)); 1154 /* The SVID says to return ENOMSG. */ 1155 error = ENOMSG; 1156 goto done2; 1157 } 1158 1159 /* 1160 * Wait for something to happen 1161 */ 1162 1163 DPRINTF(("msgrcv: goodnight\n")); 1164 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1165 "msgrcv", 0); 1166 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1167 1168 if (error != 0) { 1169 DPRINTF(("msgrcv: interrupted system call\n")); 1170 error = EINTR; 1171 goto done2; 1172 } 1173 1174 /* 1175 * Make sure that the msq queue still exists 1176 */ 1177 1178 if (msqkptr->u.msg_qbytes == 0 || 1179 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1180 DPRINTF(("msqid deleted\n")); 1181 error = EIDRM; 1182 goto done2; 1183 } 1184 } 1185 1186 /* 1187 * Return the message to the user. 1188 * 1189 * First, do the bookkeeping (before we risk being interrupted). 1190 */ 1191 1192 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1193 msqkptr->u.msg_qnum--; 1194 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1195 msqkptr->u.msg_rtime = time_second; 1196 1197 /* 1198 * Make msgsz the actual amount that we'll be returning. 1199 * Note that this effectively truncates the message if it is too long 1200 * (since msgsz is never increased). 1201 */ 1202 1203 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1204 msghdr->msg_ts)); 1205 if (msgsz > msghdr->msg_ts) 1206 msgsz = msghdr->msg_ts; 1207 *mtype = msghdr->msg_type; 1208 1209 /* 1210 * Return the segments to the user 1211 */ 1212 1213 next = msghdr->msg_spot; 1214 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1215 size_t tlen; 1216 1217 if (msgsz - len > msginfo.msgssz) 1218 tlen = msginfo.msgssz; 1219 else 1220 tlen = msgsz - len; 1221 if (next <= -1) 1222 panic("next too low #3"); 1223 if (next >= msginfo.msgseg) 1224 panic("next out of range #3"); 1225 mtx_unlock(&msq_mtx); 1226 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1227 mtx_lock(&msq_mtx); 1228 if (error != 0) { 1229 DPRINTF(("error (%d) copying out message segment\n", 1230 error)); 1231 msg_freehdr(msghdr); 1232 wakeup(msqkptr); 1233 goto done2; 1234 } 1235 msgp = (char *)msgp + tlen; 1236 next = msgmaps[next].next; 1237 } 1238 1239 /* 1240 * Done, return the actual number of bytes copied out. 1241 */ 1242 1243 msg_freehdr(msghdr); 1244 wakeup(msqkptr); 1245 td->td_retval[0] = msgsz; 1246 done2: 1247 mtx_unlock(&msq_mtx); 1248 return (error); 1249 } 1250 1251 int 1252 msgrcv(td, uap) 1253 struct thread *td; 1254 register struct msgrcv_args *uap; 1255 { 1256 int error; 1257 long mtype; 1258 1259 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1260 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1261 1262 if ((error = kern_msgrcv(td, uap->msqid, 1263 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1264 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1265 return (error); 1266 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1267 DPRINTF(("error %d copying the message type\n", error)); 1268 return (error); 1269 } 1270 1271 static int 1272 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1273 { 1274 1275 return (SYSCTL_OUT(req, msqids, 1276 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1277 } 1278 1279 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1280 "Maximum message size"); 1281 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1282 "Number of message queue identifiers"); 1283 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1284 "Maximum number of bytes in a queue"); 1285 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1286 "Maximum number of messages in the system"); 1287 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1288 "Size of a message segment"); 1289 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1290 "Number of message segments"); 1291 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1292 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1293