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 #include "opt_mac.h" 55 56 #include <sys/param.h> 57 #include <sys/systm.h> 58 #include <sys/sysproto.h> 59 #include <sys/kernel.h> 60 #include <sys/proc.h> 61 #include <sys/lock.h> 62 #include <sys/mac.h> 63 #include <sys/mutex.h> 64 #include <sys/module.h> 65 #include <sys/msg.h> 66 #include <sys/syscall.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysent.h> 69 #include <sys/sysctl.h> 70 #include <sys/malloc.h> 71 #include <sys/jail.h> 72 73 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 74 75 static void msginit(void); 76 static int msgunload(void); 77 static int sysvmsg_modload(struct module *, int, void *); 78 79 #ifdef MSG_DEBUG 80 #define DPRINTF(a) printf a 81 #else 82 #define DPRINTF(a) 83 #endif 84 #ifdef MAC_DEBUG 85 #define MPRINTF(a) printf a 86 #else 87 #define MPRINTF(a) 88 #endif 89 90 static void msg_freehdr(struct msg *msghdr); 91 92 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 93 static sy_call_t *msgcalls[] = { 94 (sy_call_t *)msgctl, (sy_call_t *)msgget, 95 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 96 }; 97 98 #ifndef MSGSSZ 99 #define MSGSSZ 8 /* Each segment must be 2^N long */ 100 #endif 101 #ifndef MSGSEG 102 #define MSGSEG 2048 /* must be less than 32767 */ 103 #endif 104 #define MSGMAX (MSGSSZ*MSGSEG) 105 #ifndef MSGMNB 106 #define MSGMNB 2048 /* max # of bytes in a queue */ 107 #endif 108 #ifndef MSGMNI 109 #define MSGMNI 40 110 #endif 111 #ifndef MSGTQL 112 #define MSGTQL 40 113 #endif 114 115 /* 116 * Based on the configuration parameters described in an SVR2 (yes, two) 117 * config(1m) man page. 118 * 119 * Each message is broken up and stored in segments that are msgssz bytes 120 * long. For efficiency reasons, this should be a power of two. Also, 121 * it doesn't make sense if it is less than 8 or greater than about 256. 122 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 123 * two between 8 and 1024 inclusive (and panic's if it isn't). 124 */ 125 struct msginfo msginfo = { 126 MSGMAX, /* max chars in a message */ 127 MSGMNI, /* # of message queue identifiers */ 128 MSGMNB, /* max chars in a queue */ 129 MSGTQL, /* max messages in system */ 130 MSGSSZ, /* size of a message segment */ 131 /* (must be small power of 2 greater than 4) */ 132 MSGSEG /* number of message segments */ 133 }; 134 135 /* 136 * macros to convert between msqid_ds's and msqid's. 137 * (specific to this implementation) 138 */ 139 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 140 #define MSQID_IX(id) ((id) & 0xffff) 141 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 142 143 /* 144 * The rest of this file is specific to this particular implementation. 145 */ 146 147 struct msgmap { 148 short next; /* next segment in buffer */ 149 /* -1 -> available */ 150 /* 0..(MSGSEG-1) -> index of next segment */ 151 }; 152 153 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 154 155 static int nfree_msgmaps; /* # of free map entries */ 156 static short free_msgmaps; /* head of linked list of free map entries */ 157 static struct msg *free_msghdrs;/* list of free msg headers */ 158 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 159 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 160 static struct msg *msghdrs; /* MSGTQL msg headers */ 161 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 162 static struct mtx msq_mtx; /* global mutex for message queues. */ 163 164 static void 165 msginit() 166 { 167 register int i; 168 169 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 170 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 171 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 172 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 173 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 174 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 175 176 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 177 if (msgpool == NULL) 178 panic("msgpool is NULL"); 179 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 180 if (msgmaps == NULL) 181 panic("msgmaps is NULL"); 182 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 183 if (msghdrs == NULL) 184 panic("msghdrs is NULL"); 185 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 186 M_WAITOK); 187 if (msqids == NULL) 188 panic("msqids is NULL"); 189 190 /* 191 * msginfo.msgssz should be a power of two for efficiency reasons. 192 * It is also pretty silly if msginfo.msgssz is less than 8 193 * or greater than about 256 so ... 194 */ 195 196 i = 8; 197 while (i < 1024 && i != msginfo.msgssz) 198 i <<= 1; 199 if (i != msginfo.msgssz) { 200 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 201 msginfo.msgssz)); 202 panic("msginfo.msgssz not a small power of 2"); 203 } 204 205 if (msginfo.msgseg > 32767) { 206 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 207 panic("msginfo.msgseg > 32767"); 208 } 209 210 if (msgmaps == NULL) 211 panic("msgmaps is NULL"); 212 213 for (i = 0; i < msginfo.msgseg; i++) { 214 if (i > 0) 215 msgmaps[i-1].next = i; 216 msgmaps[i].next = -1; /* implies entry is available */ 217 } 218 free_msgmaps = 0; 219 nfree_msgmaps = msginfo.msgseg; 220 221 if (msghdrs == NULL) 222 panic("msghdrs is NULL"); 223 224 for (i = 0; i < msginfo.msgtql; i++) { 225 msghdrs[i].msg_type = 0; 226 if (i > 0) 227 msghdrs[i-1].msg_next = &msghdrs[i]; 228 msghdrs[i].msg_next = NULL; 229 #ifdef MAC 230 mac_init_sysv_msgmsg(&msghdrs[i]); 231 #endif 232 } 233 free_msghdrs = &msghdrs[0]; 234 235 if (msqids == NULL) 236 panic("msqids is NULL"); 237 238 for (i = 0; i < msginfo.msgmni; i++) { 239 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 240 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 241 msqids[i].u.msg_perm.mode = 0; 242 #ifdef MAC 243 mac_init_sysv_msgqueue(&msqids[i]); 244 #endif 245 } 246 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 247 } 248 249 static int 250 msgunload() 251 { 252 struct msqid_kernel *msqkptr; 253 int msqid; 254 #ifdef MAC 255 int i; 256 #endif 257 258 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 259 /* 260 * Look for an unallocated and unlocked msqid_ds. 261 * msqid_ds's can be locked by msgsnd or msgrcv while 262 * they are copying the message in/out. We can't 263 * re-use the entry until they release it. 264 */ 265 msqkptr = &msqids[msqid]; 266 if (msqkptr->u.msg_qbytes != 0 || 267 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 268 break; 269 } 270 if (msqid != msginfo.msgmni) 271 return (EBUSY); 272 273 #ifdef MAC 274 for (i = 0; i < msginfo.msgtql; i++) 275 mac_destroy_sysv_msgmsg(&msghdrs[i]); 276 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 277 mac_destroy_sysv_msgqueue(&msqids[msqid]); 278 #endif 279 free(msgpool, M_MSG); 280 free(msgmaps, M_MSG); 281 free(msghdrs, M_MSG); 282 free(msqids, M_MSG); 283 mtx_destroy(&msq_mtx); 284 return (0); 285 } 286 287 288 static int 289 sysvmsg_modload(struct module *module, int cmd, void *arg) 290 { 291 int error = 0; 292 293 switch (cmd) { 294 case MOD_LOAD: 295 msginit(); 296 break; 297 case MOD_UNLOAD: 298 error = msgunload(); 299 break; 300 case MOD_SHUTDOWN: 301 break; 302 default: 303 error = EINVAL; 304 break; 305 } 306 return (error); 307 } 308 309 static moduledata_t sysvmsg_mod = { 310 "sysvmsg", 311 &sysvmsg_modload, 312 NULL 313 }; 314 315 SYSCALL_MODULE_HELPER(msgsys); 316 SYSCALL_MODULE_HELPER(msgctl); 317 SYSCALL_MODULE_HELPER(msgget); 318 SYSCALL_MODULE_HELPER(msgsnd); 319 SYSCALL_MODULE_HELPER(msgrcv); 320 321 DECLARE_MODULE(sysvmsg, sysvmsg_mod, 322 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 323 MODULE_VERSION(sysvmsg, 1); 324 325 /* 326 * Entry point for all MSG calls 327 * 328 * MPSAFE 329 */ 330 int 331 msgsys(td, uap) 332 struct thread *td; 333 /* XXX actually varargs. */ 334 struct msgsys_args /* { 335 int which; 336 int a2; 337 int a3; 338 int a4; 339 int a5; 340 int a6; 341 } */ *uap; 342 { 343 int error; 344 345 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 346 return (ENOSYS); 347 if (uap->which < 0 || 348 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 349 return (EINVAL); 350 error = (*msgcalls[uap->which])(td, &uap->a2); 351 return (error); 352 } 353 354 static void 355 msg_freehdr(msghdr) 356 struct msg *msghdr; 357 { 358 while (msghdr->msg_ts > 0) { 359 short next; 360 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 361 panic("msghdr->msg_spot out of range"); 362 next = msgmaps[msghdr->msg_spot].next; 363 msgmaps[msghdr->msg_spot].next = free_msgmaps; 364 free_msgmaps = msghdr->msg_spot; 365 nfree_msgmaps++; 366 msghdr->msg_spot = next; 367 if (msghdr->msg_ts >= msginfo.msgssz) 368 msghdr->msg_ts -= msginfo.msgssz; 369 else 370 msghdr->msg_ts = 0; 371 } 372 if (msghdr->msg_spot != -1) 373 panic("msghdr->msg_spot != -1"); 374 msghdr->msg_next = free_msghdrs; 375 free_msghdrs = msghdr; 376 #ifdef MAC 377 mac_cleanup_sysv_msgmsg(msghdr); 378 #endif 379 } 380 381 #ifndef _SYS_SYSPROTO_H_ 382 struct msgctl_args { 383 int msqid; 384 int cmd; 385 struct msqid_ds *buf; 386 }; 387 #endif 388 389 /* 390 * MPSAFE 391 */ 392 int 393 msgctl(td, uap) 394 struct thread *td; 395 register struct msgctl_args *uap; 396 { 397 int msqid = uap->msqid; 398 int cmd = uap->cmd; 399 struct msqid_ds msqbuf; 400 int error; 401 402 DPRINTF(("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, uap->buf)); 403 if (cmd == IPC_SET && 404 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 405 return (error); 406 error = kern_msgctl(td, msqid, cmd, &msqbuf); 407 if (cmd == IPC_STAT && error == 0) 408 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 409 return (error); 410 } 411 412 int 413 kern_msgctl(td, msqid, cmd, msqbuf) 414 struct thread *td; 415 int msqid; 416 int cmd; 417 struct msqid_ds *msqbuf; 418 { 419 int rval, error, msqix; 420 register struct msqid_kernel *msqkptr; 421 422 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 423 return (ENOSYS); 424 425 msqix = IPCID_TO_IX(msqid); 426 427 if (msqix < 0 || msqix >= msginfo.msgmni) { 428 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 429 msginfo.msgmni)); 430 return (EINVAL); 431 } 432 433 msqkptr = &msqids[msqix]; 434 435 mtx_lock(&msq_mtx); 436 if (msqkptr->u.msg_qbytes == 0) { 437 DPRINTF(("no such msqid\n")); 438 error = EINVAL; 439 goto done2; 440 } 441 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 442 DPRINTF(("wrong sequence number\n")); 443 error = EINVAL; 444 goto done2; 445 } 446 #ifdef MAC 447 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd); 448 if (error != 0) { 449 MPRINTF(("mac_check_sysv_msqctl returned %d\n", error)); 450 goto done2; 451 } 452 #endif 453 454 error = 0; 455 rval = 0; 456 457 switch (cmd) { 458 459 case IPC_RMID: 460 { 461 struct msg *msghdr; 462 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 463 goto done2; 464 465 #ifdef MAC 466 /* 467 * Check that the thread has MAC access permissions to 468 * individual msghdrs. Note: We need to do this in a 469 * separate loop because the actual loop alters the 470 * msq/msghdr info as it progresses, and there is no going 471 * back if half the way through we discover that the 472 * thread cannot free a certain msghdr. The msq will get 473 * into an inconsistent state. 474 */ 475 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 476 msghdr = msghdr->msg_next) { 477 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr); 478 if (error != 0) { 479 MPRINTF(("mac_check_sysv_msgrmid returned %d\n", 480 error)); 481 goto done2; 482 } 483 } 484 #endif 485 486 /* Free the message headers */ 487 msghdr = msqkptr->u.msg_first; 488 while (msghdr != NULL) { 489 struct msg *msghdr_tmp; 490 491 /* Free the segments of each message */ 492 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 493 msqkptr->u.msg_qnum--; 494 msghdr_tmp = msghdr; 495 msghdr = msghdr->msg_next; 496 msg_freehdr(msghdr_tmp); 497 } 498 499 if (msqkptr->u.msg_cbytes != 0) 500 panic("msg_cbytes is screwed up"); 501 if (msqkptr->u.msg_qnum != 0) 502 panic("msg_qnum is screwed up"); 503 504 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 505 506 #ifdef MAC 507 mac_cleanup_sysv_msgqueue(msqkptr); 508 #endif 509 510 wakeup(msqkptr); 511 } 512 513 break; 514 515 case IPC_SET: 516 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 517 goto done2; 518 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 519 error = suser(td); 520 if (error) 521 goto done2; 522 } 523 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 524 DPRINTF(("can't increase msg_qbytes beyond %d" 525 "(truncating)\n", msginfo.msgmnb)); 526 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 527 } 528 if (msqbuf->msg_qbytes == 0) { 529 DPRINTF(("can't reduce msg_qbytes to 0\n")); 530 error = EINVAL; /* non-standard errno! */ 531 goto done2; 532 } 533 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 534 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 535 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 536 (msqbuf->msg_perm.mode & 0777); 537 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 538 msqkptr->u.msg_ctime = time_second; 539 break; 540 541 case IPC_STAT: 542 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 543 DPRINTF(("requester doesn't have read access\n")); 544 goto done2; 545 } 546 *msqbuf = msqkptr->u; 547 break; 548 549 default: 550 DPRINTF(("invalid command %d\n", cmd)); 551 error = EINVAL; 552 goto done2; 553 } 554 555 if (error == 0) 556 td->td_retval[0] = rval; 557 done2: 558 mtx_unlock(&msq_mtx); 559 return (error); 560 } 561 562 #ifndef _SYS_SYSPROTO_H_ 563 struct msgget_args { 564 key_t key; 565 int msgflg; 566 }; 567 #endif 568 569 /* 570 * MPSAFE 571 */ 572 int 573 msgget(td, uap) 574 struct thread *td; 575 register struct msgget_args *uap; 576 { 577 int msqid, error = 0; 578 int key = uap->key; 579 int msgflg = uap->msgflg; 580 struct ucred *cred = td->td_ucred; 581 register struct msqid_kernel *msqkptr = NULL; 582 583 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 584 585 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 586 return (ENOSYS); 587 588 mtx_lock(&msq_mtx); 589 if (key != IPC_PRIVATE) { 590 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 591 msqkptr = &msqids[msqid]; 592 if (msqkptr->u.msg_qbytes != 0 && 593 msqkptr->u.msg_perm.key == key) 594 break; 595 } 596 if (msqid < msginfo.msgmni) { 597 DPRINTF(("found public key\n")); 598 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 599 DPRINTF(("not exclusive\n")); 600 error = EEXIST; 601 goto done2; 602 } 603 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 604 msgflg & 0700))) { 605 DPRINTF(("requester doesn't have 0%o access\n", 606 msgflg & 0700)); 607 goto done2; 608 } 609 #ifdef MAC 610 error = mac_check_sysv_msqget(cred, msqkptr); 611 if (error != 0) { 612 MPRINTF(("mac_check_sysv_msqget returned %d\n", 613 error)); 614 goto done2; 615 } 616 #endif 617 goto found; 618 } 619 } 620 621 DPRINTF(("need to allocate the msqid_ds\n")); 622 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 623 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 624 /* 625 * Look for an unallocated and unlocked msqid_ds. 626 * msqid_ds's can be locked by msgsnd or msgrcv while 627 * they are copying the message in/out. We can't 628 * re-use the entry until they release it. 629 */ 630 msqkptr = &msqids[msqid]; 631 if (msqkptr->u.msg_qbytes == 0 && 632 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 633 break; 634 } 635 if (msqid == msginfo.msgmni) { 636 DPRINTF(("no more msqid_ds's available\n")); 637 error = ENOSPC; 638 goto done2; 639 } 640 DPRINTF(("msqid %d is available\n", msqid)); 641 msqkptr->u.msg_perm.key = key; 642 msqkptr->u.msg_perm.cuid = cred->cr_uid; 643 msqkptr->u.msg_perm.uid = cred->cr_uid; 644 msqkptr->u.msg_perm.cgid = cred->cr_gid; 645 msqkptr->u.msg_perm.gid = cred->cr_gid; 646 msqkptr->u.msg_perm.mode = (msgflg & 0777); 647 /* Make sure that the returned msqid is unique */ 648 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 649 msqkptr->u.msg_first = NULL; 650 msqkptr->u.msg_last = NULL; 651 msqkptr->u.msg_cbytes = 0; 652 msqkptr->u.msg_qnum = 0; 653 msqkptr->u.msg_qbytes = msginfo.msgmnb; 654 msqkptr->u.msg_lspid = 0; 655 msqkptr->u.msg_lrpid = 0; 656 msqkptr->u.msg_stime = 0; 657 msqkptr->u.msg_rtime = 0; 658 msqkptr->u.msg_ctime = time_second; 659 #ifdef MAC 660 mac_create_sysv_msgqueue(cred, msqkptr); 661 #endif 662 } else { 663 DPRINTF(("didn't find it and wasn't asked to create it\n")); 664 error = ENOENT; 665 goto done2; 666 } 667 668 found: 669 /* Construct the unique msqid */ 670 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 671 done2: 672 mtx_unlock(&msq_mtx); 673 return (error); 674 } 675 676 #ifndef _SYS_SYSPROTO_H_ 677 struct msgsnd_args { 678 int msqid; 679 const void *msgp; 680 size_t msgsz; 681 int msgflg; 682 }; 683 #endif 684 685 /* 686 * MPSAFE 687 */ 688 int 689 msgsnd(td, uap) 690 struct thread *td; 691 register struct msgsnd_args *uap; 692 { 693 int msqid = uap->msqid; 694 const void *user_msgp = uap->msgp; 695 size_t msgsz = uap->msgsz; 696 int msgflg = uap->msgflg; 697 int segs_needed, error = 0; 698 register struct msqid_kernel *msqkptr; 699 register struct msg *msghdr; 700 short next; 701 702 DPRINTF(("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz, 703 msgflg)); 704 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 705 return (ENOSYS); 706 707 mtx_lock(&msq_mtx); 708 msqid = IPCID_TO_IX(msqid); 709 710 if (msqid < 0 || msqid >= msginfo.msgmni) { 711 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 712 msginfo.msgmni)); 713 error = EINVAL; 714 goto done2; 715 } 716 717 msqkptr = &msqids[msqid]; 718 if (msqkptr->u.msg_qbytes == 0) { 719 DPRINTF(("no such message queue id\n")); 720 error = EINVAL; 721 goto done2; 722 } 723 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 724 DPRINTF(("wrong sequence number\n")); 725 error = EINVAL; 726 goto done2; 727 } 728 729 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 730 DPRINTF(("requester doesn't have write access\n")); 731 goto done2; 732 } 733 734 #ifdef MAC 735 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr); 736 if (error != 0) { 737 MPRINTF(("mac_check_sysv_msqsnd returned %d\n", error)); 738 goto done2; 739 } 740 #endif 741 742 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 743 DPRINTF(("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz, 744 segs_needed)); 745 for (;;) { 746 int need_more_resources = 0; 747 748 /* 749 * check msgsz 750 * (inside this loop in case msg_qbytes changes while we sleep) 751 */ 752 753 if (msgsz > msqkptr->u.msg_qbytes) { 754 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 755 error = EINVAL; 756 goto done2; 757 } 758 759 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 760 DPRINTF(("msqid is locked\n")); 761 need_more_resources = 1; 762 } 763 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 764 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 765 need_more_resources = 1; 766 } 767 if (segs_needed > nfree_msgmaps) { 768 DPRINTF(("segs_needed > nfree_msgmaps\n")); 769 need_more_resources = 1; 770 } 771 if (free_msghdrs == NULL) { 772 DPRINTF(("no more msghdrs\n")); 773 need_more_resources = 1; 774 } 775 776 if (need_more_resources) { 777 int we_own_it; 778 779 if ((msgflg & IPC_NOWAIT) != 0) { 780 DPRINTF(("need more resources but caller " 781 "doesn't want to wait\n")); 782 error = EAGAIN; 783 goto done2; 784 } 785 786 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 787 DPRINTF(("we don't own the msqid_ds\n")); 788 we_own_it = 0; 789 } else { 790 /* Force later arrivals to wait for our 791 request */ 792 DPRINTF(("we own the msqid_ds\n")); 793 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 794 we_own_it = 1; 795 } 796 DPRINTF(("goodnight\n")); 797 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 798 "msgwait", 0); 799 DPRINTF(("good morning, error=%d\n", error)); 800 if (we_own_it) 801 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 802 if (error != 0) { 803 DPRINTF(("msgsnd: interrupted system call\n")); 804 error = EINTR; 805 goto done2; 806 } 807 808 /* 809 * Make sure that the msq queue still exists 810 */ 811 812 if (msqkptr->u.msg_qbytes == 0) { 813 DPRINTF(("msqid deleted\n")); 814 error = EIDRM; 815 goto done2; 816 } 817 818 } else { 819 DPRINTF(("got all the resources that we need\n")); 820 break; 821 } 822 } 823 824 /* 825 * We have the resources that we need. 826 * Make sure! 827 */ 828 829 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 830 panic("msg_perm.mode & MSG_LOCKED"); 831 if (segs_needed > nfree_msgmaps) 832 panic("segs_needed > nfree_msgmaps"); 833 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 834 panic("msgsz + msg_cbytes > msg_qbytes"); 835 if (free_msghdrs == NULL) 836 panic("no more msghdrs"); 837 838 /* 839 * Re-lock the msqid_ds in case we page-fault when copying in the 840 * message 841 */ 842 843 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 844 panic("msqid_ds is already locked"); 845 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 846 847 /* 848 * Allocate a message header 849 */ 850 851 msghdr = free_msghdrs; 852 free_msghdrs = msghdr->msg_next; 853 msghdr->msg_spot = -1; 854 msghdr->msg_ts = msgsz; 855 #ifdef MAC 856 /* 857 * XXXMAC: Should the mac_check_sysv_msgmsq check follow here 858 * immediately? Or, should it be checked just before the msg is 859 * enqueued in the msgq (as it is done now)? 860 */ 861 mac_create_sysv_msgmsg(td->td_ucred, msqkptr, msghdr); 862 #endif 863 864 /* 865 * Allocate space for the message 866 */ 867 868 while (segs_needed > 0) { 869 if (nfree_msgmaps <= 0) 870 panic("not enough msgmaps"); 871 if (free_msgmaps == -1) 872 panic("nil free_msgmaps"); 873 next = free_msgmaps; 874 if (next <= -1) 875 panic("next too low #1"); 876 if (next >= msginfo.msgseg) 877 panic("next out of range #1"); 878 DPRINTF(("allocating segment %d to message\n", next)); 879 free_msgmaps = msgmaps[next].next; 880 nfree_msgmaps--; 881 msgmaps[next].next = msghdr->msg_spot; 882 msghdr->msg_spot = next; 883 segs_needed--; 884 } 885 886 /* 887 * Copy in the message type 888 */ 889 890 mtx_unlock(&msq_mtx); 891 if ((error = copyin(user_msgp, &msghdr->msg_type, 892 sizeof(msghdr->msg_type))) != 0) { 893 mtx_lock(&msq_mtx); 894 DPRINTF(("error %d copying the message type\n", 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 user_msgp = (const char *)user_msgp + sizeof(msghdr->msg_type); 902 903 /* 904 * Validate the message type 905 */ 906 907 if (msghdr->msg_type < 1) { 908 msg_freehdr(msghdr); 909 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 910 wakeup(msqkptr); 911 DPRINTF(("mtype (%d) < 1\n", msghdr->msg_type)); 912 error = EINVAL; 913 goto done2; 914 } 915 916 /* 917 * Copy in the message body 918 */ 919 920 next = msghdr->msg_spot; 921 while (msgsz > 0) { 922 size_t tlen; 923 if (msgsz > msginfo.msgssz) 924 tlen = msginfo.msgssz; 925 else 926 tlen = msgsz; 927 if (next <= -1) 928 panic("next too low #2"); 929 if (next >= msginfo.msgseg) 930 panic("next out of range #2"); 931 mtx_unlock(&msq_mtx); 932 if ((error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], 933 tlen)) != 0) { 934 mtx_lock(&msq_mtx); 935 DPRINTF(("error %d copying in message segment\n", 936 error)); 937 msg_freehdr(msghdr); 938 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 939 wakeup(msqkptr); 940 goto done2; 941 } 942 mtx_lock(&msq_mtx); 943 msgsz -= tlen; 944 user_msgp = (const char *)user_msgp + tlen; 945 next = msgmaps[next].next; 946 } 947 if (next != -1) 948 panic("didn't use all the msg segments"); 949 950 /* 951 * We've got the message. Unlock the msqid_ds. 952 */ 953 954 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 955 956 /* 957 * Make sure that the msqid_ds is still allocated. 958 */ 959 960 if (msqkptr->u.msg_qbytes == 0) { 961 msg_freehdr(msghdr); 962 wakeup(msqkptr); 963 error = EIDRM; 964 goto done2; 965 } 966 967 #ifdef MAC 968 /* 969 * Note: Since the task/thread allocates the msghdr and usually 970 * primes it with its own MAC label, for a majority of policies, it 971 * won't be necessary to check whether the msghdr has access 972 * permissions to the msgq. The mac_check_sysv_msqsnd check would 973 * suffice in that case. However, this hook may be required where 974 * individual policies derive a non-identical label for the msghdr 975 * from the current thread label and may want to check the msghdr 976 * enqueue permissions, along with read/write permissions to the 977 * msgq. 978 */ 979 error = mac_check_sysv_msgmsq(td->td_ucred, msghdr, msqkptr); 980 if (error != 0) { 981 MPRINTF(("mac_check_sysv_msqmsq returned %d\n", error)); 982 msg_freehdr(msghdr); 983 wakeup(msqkptr); 984 goto done2; 985 } 986 #endif 987 988 /* 989 * Put the message into the queue 990 */ 991 if (msqkptr->u.msg_first == NULL) { 992 msqkptr->u.msg_first = msghdr; 993 msqkptr->u.msg_last = msghdr; 994 } else { 995 msqkptr->u.msg_last->msg_next = msghdr; 996 msqkptr->u.msg_last = msghdr; 997 } 998 msqkptr->u.msg_last->msg_next = NULL; 999 1000 msqkptr->u.msg_cbytes += msghdr->msg_ts; 1001 msqkptr->u.msg_qnum++; 1002 msqkptr->u.msg_lspid = td->td_proc->p_pid; 1003 msqkptr->u.msg_stime = time_second; 1004 1005 wakeup(msqkptr); 1006 td->td_retval[0] = 0; 1007 done2: 1008 mtx_unlock(&msq_mtx); 1009 return (error); 1010 } 1011 1012 #ifndef _SYS_SYSPROTO_H_ 1013 struct msgrcv_args { 1014 int msqid; 1015 void *msgp; 1016 size_t msgsz; 1017 long msgtyp; 1018 int msgflg; 1019 }; 1020 #endif 1021 1022 /* 1023 * MPSAFE 1024 */ 1025 int 1026 msgrcv(td, uap) 1027 struct thread *td; 1028 register struct msgrcv_args *uap; 1029 { 1030 int msqid = uap->msqid; 1031 void *user_msgp = uap->msgp; 1032 size_t msgsz = uap->msgsz; 1033 long msgtyp = uap->msgtyp; 1034 int msgflg = uap->msgflg; 1035 size_t len; 1036 register struct msqid_kernel *msqkptr; 1037 register struct msg *msghdr; 1038 int error = 0; 1039 short next; 1040 1041 DPRINTF(("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp, 1042 msgsz, msgtyp, msgflg)); 1043 1044 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 1045 return (ENOSYS); 1046 1047 msqid = IPCID_TO_IX(msqid); 1048 1049 if (msqid < 0 || msqid >= msginfo.msgmni) { 1050 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 1051 msginfo.msgmni)); 1052 return (EINVAL); 1053 } 1054 1055 msqkptr = &msqids[msqid]; 1056 mtx_lock(&msq_mtx); 1057 if (msqkptr->u.msg_qbytes == 0) { 1058 DPRINTF(("no such message queue id\n")); 1059 error = EINVAL; 1060 goto done2; 1061 } 1062 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1063 DPRINTF(("wrong sequence number\n")); 1064 error = EINVAL; 1065 goto done2; 1066 } 1067 1068 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1069 DPRINTF(("requester doesn't have read access\n")); 1070 goto done2; 1071 } 1072 1073 #ifdef MAC 1074 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr); 1075 if (error != 0) { 1076 MPRINTF(("mac_check_sysv_msqrcv returned %d\n", error)); 1077 goto done2; 1078 } 1079 #endif 1080 1081 msghdr = NULL; 1082 while (msghdr == NULL) { 1083 if (msgtyp == 0) { 1084 msghdr = msqkptr->u.msg_first; 1085 if (msghdr != NULL) { 1086 if (msgsz < msghdr->msg_ts && 1087 (msgflg & MSG_NOERROR) == 0) { 1088 DPRINTF(("first message on the queue " 1089 "is too big (want %d, got %d)\n", 1090 msgsz, msghdr->msg_ts)); 1091 error = E2BIG; 1092 goto done2; 1093 } 1094 #ifdef MAC 1095 error = mac_check_sysv_msgrcv(td->td_ucred, 1096 msghdr); 1097 if (error != 0) { 1098 MPRINTF(("mac_check_sysv_msgrcv " 1099 "returned %d\n", error)); 1100 goto done2; 1101 } 1102 #endif 1103 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1104 msqkptr->u.msg_first = NULL; 1105 msqkptr->u.msg_last = NULL; 1106 } else { 1107 msqkptr->u.msg_first = msghdr->msg_next; 1108 if (msqkptr->u.msg_first == NULL) 1109 panic("msg_first/last screwed up #1"); 1110 } 1111 } 1112 } else { 1113 struct msg *previous; 1114 struct msg **prev; 1115 1116 previous = NULL; 1117 prev = &(msqkptr->u.msg_first); 1118 while ((msghdr = *prev) != NULL) { 1119 /* 1120 * Is this message's type an exact match or is 1121 * this message's type less than or equal to 1122 * the absolute value of a negative msgtyp? 1123 * Note that the second half of this test can 1124 * NEVER be true if msgtyp is positive since 1125 * msg_type is always positive! 1126 */ 1127 1128 if (msgtyp == msghdr->msg_type || 1129 msghdr->msg_type <= -msgtyp) { 1130 DPRINTF(("found message type %d, " 1131 "requested %d\n", 1132 msghdr->msg_type, msgtyp)); 1133 if (msgsz < msghdr->msg_ts && 1134 (msgflg & MSG_NOERROR) == 0) { 1135 DPRINTF(("requested message " 1136 "on the queue is too big " 1137 "(want %d, got %d)\n", 1138 msgsz, msghdr->msg_ts)); 1139 error = E2BIG; 1140 goto done2; 1141 } 1142 #ifdef MAC 1143 error = mac_check_sysv_msgrcv( 1144 td->td_ucred, msghdr); 1145 if (error != 0) { 1146 MPRINTF(("mac_check_sysv_" 1147 "msgrcv returned %d\n", 1148 error)); 1149 goto done2; 1150 } 1151 #endif 1152 *prev = msghdr->msg_next; 1153 if (msghdr == msqkptr->u.msg_last) { 1154 if (previous == NULL) { 1155 if (prev != 1156 &msqkptr->u.msg_first) 1157 panic("msg_first/last screwed up #2"); 1158 msqkptr->u.msg_first = 1159 NULL; 1160 msqkptr->u.msg_last = 1161 NULL; 1162 } else { 1163 if (prev == 1164 &msqkptr->u.msg_first) 1165 panic("msg_first/last screwed up #3"); 1166 msqkptr->u.msg_last = 1167 previous; 1168 } 1169 } 1170 break; 1171 } 1172 previous = msghdr; 1173 prev = &(msghdr->msg_next); 1174 } 1175 } 1176 1177 /* 1178 * We've either extracted the msghdr for the appropriate 1179 * message or there isn't one. 1180 * If there is one then bail out of this loop. 1181 */ 1182 1183 if (msghdr != NULL) 1184 break; 1185 1186 /* 1187 * Hmph! No message found. Does the user want to wait? 1188 */ 1189 1190 if ((msgflg & IPC_NOWAIT) != 0) { 1191 DPRINTF(("no appropriate message found (msgtyp=%d)\n", 1192 msgtyp)); 1193 /* The SVID says to return ENOMSG. */ 1194 error = ENOMSG; 1195 goto done2; 1196 } 1197 1198 /* 1199 * Wait for something to happen 1200 */ 1201 1202 DPRINTF(("msgrcv: goodnight\n")); 1203 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1204 "msgwait", 0); 1205 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1206 1207 if (error != 0) { 1208 DPRINTF(("msgsnd: interrupted system call\n")); 1209 error = EINTR; 1210 goto done2; 1211 } 1212 1213 /* 1214 * Make sure that the msq queue still exists 1215 */ 1216 1217 if (msqkptr->u.msg_qbytes == 0 || 1218 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { 1219 DPRINTF(("msqid deleted\n")); 1220 error = EIDRM; 1221 goto done2; 1222 } 1223 } 1224 1225 /* 1226 * Return the message to the user. 1227 * 1228 * First, do the bookkeeping (before we risk being interrupted). 1229 */ 1230 1231 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1232 msqkptr->u.msg_qnum--; 1233 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1234 msqkptr->u.msg_rtime = time_second; 1235 1236 /* 1237 * Make msgsz the actual amount that we'll be returning. 1238 * Note that this effectively truncates the message if it is too long 1239 * (since msgsz is never increased). 1240 */ 1241 1242 DPRINTF(("found a message, msgsz=%d, msg_ts=%d\n", msgsz, 1243 msghdr->msg_ts)); 1244 if (msgsz > msghdr->msg_ts) 1245 msgsz = msghdr->msg_ts; 1246 1247 /* 1248 * Return the type to the user. 1249 */ 1250 1251 mtx_unlock(&msq_mtx); 1252 error = copyout(&(msghdr->msg_type), user_msgp, 1253 sizeof(msghdr->msg_type)); 1254 mtx_lock(&msq_mtx); 1255 if (error != 0) { 1256 DPRINTF(("error (%d) copying out message type\n", error)); 1257 msg_freehdr(msghdr); 1258 wakeup(msqkptr); 1259 goto done2; 1260 } 1261 user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type); 1262 1263 /* 1264 * Return the segments to the user 1265 */ 1266 1267 next = msghdr->msg_spot; 1268 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1269 size_t tlen; 1270 1271 if (msgsz - len > msginfo.msgssz) 1272 tlen = msginfo.msgssz; 1273 else 1274 tlen = msgsz - len; 1275 if (next <= -1) 1276 panic("next too low #3"); 1277 if (next >= msginfo.msgseg) 1278 panic("next out of range #3"); 1279 mtx_unlock(&msq_mtx); 1280 error = copyout(&msgpool[next * msginfo.msgssz], 1281 user_msgp, tlen); 1282 mtx_lock(&msq_mtx); 1283 if (error != 0) { 1284 DPRINTF(("error (%d) copying out message segment\n", 1285 error)); 1286 msg_freehdr(msghdr); 1287 wakeup(msqkptr); 1288 goto done2; 1289 } 1290 user_msgp = (char *)user_msgp + tlen; 1291 next = msgmaps[next].next; 1292 } 1293 1294 /* 1295 * Done, return the actual number of bytes copied out. 1296 */ 1297 1298 msg_freehdr(msghdr); 1299 wakeup(msqkptr); 1300 td->td_retval[0] = msgsz; 1301 done2: 1302 mtx_unlock(&msq_mtx); 1303 return (error); 1304 } 1305 1306 static int 1307 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1308 { 1309 1310 return (SYSCTL_OUT(req, msqids, 1311 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1312 } 1313 1314 SYSCTL_DECL(_kern_ipc); 1315 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1316 "Maximum message size"); 1317 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1318 "Number of message queue identifiers"); 1319 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1320 "Maximum number of bytes in a queue"); 1321 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1322 "Maximum number of messages in the system"); 1323 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1324 "Size of a message segment"); 1325 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1326 "Number of message segments"); 1327 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1328 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1329