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_compat.h" 54 #include "opt_sysvipc.h" 55 56 #include <sys/param.h> 57 #include <sys/systm.h> 58 #include <sys/sysproto.h> 59 #include <sys/kernel.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/lock.h> 63 #include <sys/mutex.h> 64 #include <sys/module.h> 65 #include <sys/mount.h> 66 #include <sys/msg.h> 67 #include <sys/racct.h> 68 #include <sys/sx.h> 69 #include <sys/syscall.h> 70 #include <sys/syscallsubr.h> 71 #include <sys/sysent.h> 72 #include <sys/sysctl.h> 73 #include <sys/malloc.h> 74 #include <sys/jail.h> 75 76 #include <security/audit/audit.h> 77 #include <security/mac/mac_framework.h> 78 79 FEATURE(sysv_msg, "System V message queues support"); 80 81 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 82 83 static int msginit(void); 84 static int msgunload(void); 85 static int sysvmsg_modload(struct module *, int, void *); 86 static void msq_remove(struct msqid_kernel *); 87 static struct prison *msg_find_prison(struct ucred *); 88 static int msq_prison_cansee(struct prison *, struct msqid_kernel *); 89 static int msg_prison_check(void *, void *); 90 static int msg_prison_set(void *, void *); 91 static int msg_prison_get(void *, void *); 92 static int msg_prison_remove(void *, void *); 93 static void msg_prison_cleanup(struct prison *); 94 95 96 #ifdef MSG_DEBUG 97 #define DPRINTF(a) printf a 98 #else 99 #define DPRINTF(a) (void)0 100 #endif 101 102 static void msg_freehdr(struct msg *msghdr); 103 104 #ifndef MSGSSZ 105 #define MSGSSZ 8 /* Each segment must be 2^N long */ 106 #endif 107 #ifndef MSGSEG 108 #define MSGSEG 2048 /* must be less than 32767 */ 109 #endif 110 #define MSGMAX (MSGSSZ*MSGSEG) 111 #ifndef MSGMNB 112 #define MSGMNB 2048 /* max # of bytes in a queue */ 113 #endif 114 #ifndef MSGMNI 115 #define MSGMNI 40 116 #endif 117 #ifndef MSGTQL 118 #define MSGTQL 40 119 #endif 120 121 /* 122 * Based on the configuration parameters described in an SVR2 (yes, two) 123 * config(1m) man page. 124 * 125 * Each message is broken up and stored in segments that are msgssz bytes 126 * long. For efficiency reasons, this should be a power of two. Also, 127 * it doesn't make sense if it is less than 8 or greater than about 256. 128 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 129 * two between 8 and 1024 inclusive (and panic's if it isn't). 130 */ 131 struct msginfo msginfo = { 132 MSGMAX, /* max chars in a message */ 133 MSGMNI, /* # of message queue identifiers */ 134 MSGMNB, /* max chars in a queue */ 135 MSGTQL, /* max messages in system */ 136 MSGSSZ, /* size of a message segment */ 137 /* (must be small power of 2 greater than 4) */ 138 MSGSEG /* number of message segments */ 139 }; 140 141 /* 142 * macros to convert between msqid_ds's and msqid's. 143 * (specific to this implementation) 144 */ 145 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 146 #define MSQID_IX(id) ((id) & 0xffff) 147 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 148 149 /* 150 * The rest of this file is specific to this particular implementation. 151 */ 152 153 struct msgmap { 154 short next; /* next segment in buffer */ 155 /* -1 -> available */ 156 /* 0..(MSGSEG-1) -> index of next segment */ 157 }; 158 159 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 160 161 static int nfree_msgmaps; /* # of free map entries */ 162 static short free_msgmaps; /* head of linked list of free map entries */ 163 static struct msg *free_msghdrs;/* list of free msg headers */ 164 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 165 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 166 static struct msg *msghdrs; /* MSGTQL msg headers */ 167 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 168 static struct mtx msq_mtx; /* global mutex for message queues. */ 169 static unsigned msg_prison_slot;/* prison OSD slot */ 170 171 static struct syscall_helper_data msg_syscalls[] = { 172 SYSCALL_INIT_HELPER(msgctl), 173 SYSCALL_INIT_HELPER(msgget), 174 SYSCALL_INIT_HELPER(msgsnd), 175 SYSCALL_INIT_HELPER(msgrcv), 176 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 177 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 178 SYSCALL_INIT_HELPER(msgsys), 179 SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl), 180 #endif 181 SYSCALL_INIT_LAST 182 }; 183 184 #ifdef COMPAT_FREEBSD32 185 #include <compat/freebsd32/freebsd32.h> 186 #include <compat/freebsd32/freebsd32_ipc.h> 187 #include <compat/freebsd32/freebsd32_proto.h> 188 #include <compat/freebsd32/freebsd32_signal.h> 189 #include <compat/freebsd32/freebsd32_syscall.h> 190 #include <compat/freebsd32/freebsd32_util.h> 191 192 static struct syscall_helper_data msg32_syscalls[] = { 193 SYSCALL32_INIT_HELPER(freebsd32_msgctl), 194 SYSCALL32_INIT_HELPER(freebsd32_msgsnd), 195 SYSCALL32_INIT_HELPER(freebsd32_msgrcv), 196 SYSCALL32_INIT_HELPER_COMPAT(msgget), 197 SYSCALL32_INIT_HELPER(freebsd32_msgsys), 198 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 199 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 200 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl), 201 #endif 202 SYSCALL_INIT_LAST 203 }; 204 #endif 205 206 static int 207 msginit() 208 { 209 struct prison *pr; 210 void **rsv; 211 int i, error; 212 osd_method_t methods[PR_MAXMETHOD] = { 213 [PR_METHOD_CHECK] = msg_prison_check, 214 [PR_METHOD_SET] = msg_prison_set, 215 [PR_METHOD_GET] = msg_prison_get, 216 [PR_METHOD_REMOVE] = msg_prison_remove, 217 }; 218 219 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 220 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 221 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 222 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 223 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 224 M_WAITOK); 225 226 /* 227 * msginfo.msgssz should be a power of two for efficiency reasons. 228 * It is also pretty silly if msginfo.msgssz is less than 8 229 * or greater than about 256 so ... 230 */ 231 232 i = 8; 233 while (i < 1024 && i != msginfo.msgssz) 234 i <<= 1; 235 if (i != msginfo.msgssz) { 236 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 237 msginfo.msgssz)); 238 panic("msginfo.msgssz not a small power of 2"); 239 } 240 241 if (msginfo.msgseg > 32767) { 242 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 243 panic("msginfo.msgseg > 32767"); 244 } 245 246 for (i = 0; i < msginfo.msgseg; i++) { 247 if (i > 0) 248 msgmaps[i-1].next = i; 249 msgmaps[i].next = -1; /* implies entry is available */ 250 } 251 free_msgmaps = 0; 252 nfree_msgmaps = msginfo.msgseg; 253 254 for (i = 0; i < msginfo.msgtql; i++) { 255 msghdrs[i].msg_type = 0; 256 if (i > 0) 257 msghdrs[i-1].msg_next = &msghdrs[i]; 258 msghdrs[i].msg_next = NULL; 259 #ifdef MAC 260 mac_sysvmsg_init(&msghdrs[i]); 261 #endif 262 } 263 free_msghdrs = &msghdrs[0]; 264 265 for (i = 0; i < msginfo.msgmni; i++) { 266 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 267 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 268 msqids[i].u.msg_perm.mode = 0; 269 #ifdef MAC 270 mac_sysvmsq_init(&msqids[i]); 271 #endif 272 } 273 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 274 275 /* Set current prisons according to their allow.sysvipc. */ 276 msg_prison_slot = osd_jail_register(NULL, methods); 277 rsv = osd_reserve(msg_prison_slot); 278 prison_lock(&prison0); 279 (void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0); 280 prison_unlock(&prison0); 281 rsv = NULL; 282 sx_slock(&allprison_lock); 283 TAILQ_FOREACH(pr, &allprison, pr_list) { 284 if (rsv == NULL) 285 rsv = osd_reserve(msg_prison_slot); 286 prison_lock(pr); 287 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) { 288 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 289 &prison0); 290 rsv = NULL; 291 } 292 prison_unlock(pr); 293 } 294 if (rsv != NULL) 295 osd_free_reserved(rsv); 296 sx_sunlock(&allprison_lock); 297 298 error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD); 299 if (error != 0) 300 return (error); 301 #ifdef COMPAT_FREEBSD32 302 error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD); 303 if (error != 0) 304 return (error); 305 #endif 306 return (0); 307 } 308 309 static int 310 msgunload() 311 { 312 struct msqid_kernel *msqkptr; 313 int msqid; 314 #ifdef MAC 315 int i; 316 #endif 317 318 syscall_helper_unregister(msg_syscalls); 319 #ifdef COMPAT_FREEBSD32 320 syscall32_helper_unregister(msg32_syscalls); 321 #endif 322 323 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 324 msqkptr = &msqids[msqid]; 325 if (msqkptr->u.msg_qbytes != 0 || 326 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 327 break; 328 } 329 if (msqid != msginfo.msgmni) 330 return (EBUSY); 331 332 if (msg_prison_slot != 0) 333 osd_jail_deregister(msg_prison_slot); 334 #ifdef MAC 335 for (i = 0; i < msginfo.msgtql; i++) 336 mac_sysvmsg_destroy(&msghdrs[i]); 337 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 338 mac_sysvmsq_destroy(&msqids[msqid]); 339 #endif 340 free(msgpool, M_MSG); 341 free(msgmaps, M_MSG); 342 free(msghdrs, M_MSG); 343 free(msqids, M_MSG); 344 mtx_destroy(&msq_mtx); 345 return (0); 346 } 347 348 349 static int 350 sysvmsg_modload(struct module *module, int cmd, void *arg) 351 { 352 int error = 0; 353 354 switch (cmd) { 355 case MOD_LOAD: 356 error = msginit(); 357 if (error != 0) 358 msgunload(); 359 break; 360 case MOD_UNLOAD: 361 error = msgunload(); 362 break; 363 case MOD_SHUTDOWN: 364 break; 365 default: 366 error = EINVAL; 367 break; 368 } 369 return (error); 370 } 371 372 static moduledata_t sysvmsg_mod = { 373 "sysvmsg", 374 &sysvmsg_modload, 375 NULL 376 }; 377 378 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 379 MODULE_VERSION(sysvmsg, 1); 380 381 static void 382 msg_freehdr(msghdr) 383 struct msg *msghdr; 384 { 385 while (msghdr->msg_ts > 0) { 386 short next; 387 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 388 panic("msghdr->msg_spot out of range"); 389 next = msgmaps[msghdr->msg_spot].next; 390 msgmaps[msghdr->msg_spot].next = free_msgmaps; 391 free_msgmaps = msghdr->msg_spot; 392 nfree_msgmaps++; 393 msghdr->msg_spot = next; 394 if (msghdr->msg_ts >= msginfo.msgssz) 395 msghdr->msg_ts -= msginfo.msgssz; 396 else 397 msghdr->msg_ts = 0; 398 } 399 if (msghdr->msg_spot != -1) 400 panic("msghdr->msg_spot != -1"); 401 msghdr->msg_next = free_msghdrs; 402 free_msghdrs = msghdr; 403 #ifdef MAC 404 mac_sysvmsg_cleanup(msghdr); 405 #endif 406 } 407 408 static void 409 msq_remove(struct msqid_kernel *msqkptr) 410 { 411 struct msg *msghdr; 412 413 racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1); 414 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum); 415 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes); 416 crfree(msqkptr->cred); 417 msqkptr->cred = NULL; 418 419 /* Free the message headers */ 420 msghdr = msqkptr->u.msg_first; 421 while (msghdr != NULL) { 422 struct msg *msghdr_tmp; 423 424 /* Free the segments of each message */ 425 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 426 msqkptr->u.msg_qnum--; 427 msghdr_tmp = msghdr; 428 msghdr = msghdr->msg_next; 429 msg_freehdr(msghdr_tmp); 430 } 431 432 if (msqkptr->u.msg_cbytes != 0) 433 panic("msg_cbytes is screwed up"); 434 if (msqkptr->u.msg_qnum != 0) 435 panic("msg_qnum is screwed up"); 436 437 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 438 439 #ifdef MAC 440 mac_sysvmsq_cleanup(msqkptr); 441 #endif 442 443 wakeup(msqkptr); 444 } 445 446 static struct prison * 447 msg_find_prison(struct ucred *cred) 448 { 449 struct prison *pr, *rpr; 450 451 pr = cred->cr_prison; 452 prison_lock(pr); 453 rpr = osd_jail_get(pr, msg_prison_slot); 454 prison_unlock(pr); 455 return rpr; 456 } 457 458 static int 459 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr) 460 { 461 462 if (msqkptr->cred == NULL || 463 !(rpr == msqkptr->cred->cr_prison || 464 prison_ischild(rpr, msqkptr->cred->cr_prison))) 465 return (EINVAL); 466 return (0); 467 } 468 469 #ifndef _SYS_SYSPROTO_H_ 470 struct msgctl_args { 471 int msqid; 472 int cmd; 473 struct msqid_ds *buf; 474 }; 475 #endif 476 int 477 sys_msgctl(td, uap) 478 struct thread *td; 479 register struct msgctl_args *uap; 480 { 481 int msqid = uap->msqid; 482 int cmd = uap->cmd; 483 struct msqid_ds msqbuf; 484 int error; 485 486 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 487 if (cmd == IPC_SET && 488 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 489 return (error); 490 error = kern_msgctl(td, msqid, cmd, &msqbuf); 491 if (cmd == IPC_STAT && error == 0) 492 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 493 return (error); 494 } 495 496 int 497 kern_msgctl(td, msqid, cmd, msqbuf) 498 struct thread *td; 499 int msqid; 500 int cmd; 501 struct msqid_ds *msqbuf; 502 { 503 int rval, error, msqix; 504 register struct msqid_kernel *msqkptr; 505 struct prison *rpr; 506 507 rpr = msg_find_prison(td->td_ucred); 508 if (rpr == NULL) 509 return (ENOSYS); 510 511 msqix = IPCID_TO_IX(msqid); 512 513 if (msqix < 0 || msqix >= msginfo.msgmni) { 514 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 515 msginfo.msgmni)); 516 return (EINVAL); 517 } 518 519 msqkptr = &msqids[msqix]; 520 521 mtx_lock(&msq_mtx); 522 if (msqkptr->u.msg_qbytes == 0) { 523 DPRINTF(("no such msqid\n")); 524 error = EINVAL; 525 goto done2; 526 } 527 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 528 DPRINTF(("wrong sequence number\n")); 529 error = EINVAL; 530 goto done2; 531 } 532 533 error = msq_prison_cansee(rpr, msqkptr); 534 if (error != 0) { 535 DPRINTF(("requester can't see prison\n")); 536 goto done2; 537 } 538 539 #ifdef MAC 540 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 541 if (error != 0) 542 goto done2; 543 #endif 544 545 error = 0; 546 rval = 0; 547 548 switch (cmd) { 549 550 case IPC_RMID: 551 { 552 #ifdef MAC 553 struct msg *msghdr; 554 #endif 555 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 556 goto done2; 557 558 #ifdef MAC 559 /* 560 * Check that the thread has MAC access permissions to 561 * individual msghdrs. Note: We need to do this in a 562 * separate loop because the actual loop alters the 563 * msq/msghdr info as it progresses, and there is no going 564 * back if half the way through we discover that the 565 * thread cannot free a certain msghdr. The msq will get 566 * into an inconsistent state. 567 */ 568 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 569 msghdr = msghdr->msg_next) { 570 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 571 if (error != 0) 572 goto done2; 573 } 574 #endif 575 576 msq_remove(msqkptr); 577 } 578 579 break; 580 581 case IPC_SET: 582 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 583 goto done2; 584 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 585 error = priv_check(td, PRIV_IPC_MSGSIZE); 586 if (error) 587 goto done2; 588 } 589 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 590 DPRINTF(("can't increase msg_qbytes beyond %d" 591 "(truncating)\n", msginfo.msgmnb)); 592 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 593 } 594 if (msqbuf->msg_qbytes == 0) { 595 DPRINTF(("can't reduce msg_qbytes to 0\n")); 596 error = EINVAL; /* non-standard errno! */ 597 goto done2; 598 } 599 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 600 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 601 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 602 (msqbuf->msg_perm.mode & 0777); 603 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 604 msqkptr->u.msg_ctime = time_second; 605 break; 606 607 case IPC_STAT: 608 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 609 DPRINTF(("requester doesn't have read access\n")); 610 goto done2; 611 } 612 *msqbuf = msqkptr->u; 613 if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison) 614 msqbuf->msg_perm.key = IPC_PRIVATE; 615 break; 616 617 default: 618 DPRINTF(("invalid command %d\n", cmd)); 619 error = EINVAL; 620 goto done2; 621 } 622 623 if (error == 0) 624 td->td_retval[0] = rval; 625 done2: 626 mtx_unlock(&msq_mtx); 627 return (error); 628 } 629 630 #ifndef _SYS_SYSPROTO_H_ 631 struct msgget_args { 632 key_t key; 633 int msgflg; 634 }; 635 #endif 636 637 int 638 sys_msgget(td, uap) 639 struct thread *td; 640 register struct msgget_args *uap; 641 { 642 int msqid, error = 0; 643 int key = uap->key; 644 int msgflg = uap->msgflg; 645 struct ucred *cred = td->td_ucred; 646 register struct msqid_kernel *msqkptr = NULL; 647 648 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 649 650 if (msg_find_prison(cred) == NULL) 651 return (ENOSYS); 652 653 mtx_lock(&msq_mtx); 654 if (key != IPC_PRIVATE) { 655 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 656 msqkptr = &msqids[msqid]; 657 if (msqkptr->u.msg_qbytes != 0 && 658 msqkptr->cred != NULL && 659 msqkptr->cred->cr_prison == cred->cr_prison && 660 msqkptr->u.msg_perm.key == key) 661 break; 662 } 663 if (msqid < msginfo.msgmni) { 664 DPRINTF(("found public key\n")); 665 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 666 DPRINTF(("not exclusive\n")); 667 error = EEXIST; 668 goto done2; 669 } 670 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 671 msgflg & 0700))) { 672 DPRINTF(("requester doesn't have 0%o access\n", 673 msgflg & 0700)); 674 goto done2; 675 } 676 #ifdef MAC 677 error = mac_sysvmsq_check_msqget(cred, msqkptr); 678 if (error != 0) 679 goto done2; 680 #endif 681 goto found; 682 } 683 } 684 685 DPRINTF(("need to allocate the msqid_ds\n")); 686 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 687 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 688 /* 689 * Look for an unallocated and unlocked msqid_ds. 690 * msqid_ds's can be locked by msgsnd or msgrcv while 691 * they are copying the message in/out. We can't 692 * re-use the entry until they release it. 693 */ 694 msqkptr = &msqids[msqid]; 695 if (msqkptr->u.msg_qbytes == 0 && 696 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 697 break; 698 } 699 if (msqid == msginfo.msgmni) { 700 DPRINTF(("no more msqid_ds's available\n")); 701 error = ENOSPC; 702 goto done2; 703 } 704 #ifdef RACCT 705 if (racct_enable) { 706 PROC_LOCK(td->td_proc); 707 error = racct_add(td->td_proc, RACCT_NMSGQ, 1); 708 PROC_UNLOCK(td->td_proc); 709 if (error != 0) { 710 error = ENOSPC; 711 goto done2; 712 } 713 } 714 #endif 715 DPRINTF(("msqid %d is available\n", msqid)); 716 msqkptr->u.msg_perm.key = key; 717 msqkptr->u.msg_perm.cuid = cred->cr_uid; 718 msqkptr->u.msg_perm.uid = cred->cr_uid; 719 msqkptr->u.msg_perm.cgid = cred->cr_gid; 720 msqkptr->u.msg_perm.gid = cred->cr_gid; 721 msqkptr->u.msg_perm.mode = (msgflg & 0777); 722 msqkptr->cred = crhold(cred); 723 /* Make sure that the returned msqid is unique */ 724 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 725 msqkptr->u.msg_first = NULL; 726 msqkptr->u.msg_last = NULL; 727 msqkptr->u.msg_cbytes = 0; 728 msqkptr->u.msg_qnum = 0; 729 msqkptr->u.msg_qbytes = msginfo.msgmnb; 730 msqkptr->u.msg_lspid = 0; 731 msqkptr->u.msg_lrpid = 0; 732 msqkptr->u.msg_stime = 0; 733 msqkptr->u.msg_rtime = 0; 734 msqkptr->u.msg_ctime = time_second; 735 #ifdef MAC 736 mac_sysvmsq_create(cred, msqkptr); 737 #endif 738 } else { 739 DPRINTF(("didn't find it and wasn't asked to create it\n")); 740 error = ENOENT; 741 goto done2; 742 } 743 744 found: 745 /* Construct the unique msqid */ 746 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 747 done2: 748 mtx_unlock(&msq_mtx); 749 return (error); 750 } 751 752 #ifndef _SYS_SYSPROTO_H_ 753 struct msgsnd_args { 754 int msqid; 755 const void *msgp; 756 size_t msgsz; 757 int msgflg; 758 }; 759 #endif 760 int 761 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 762 struct thread *td; 763 int msqid; 764 const void *msgp; /* XXX msgp is actually mtext. */ 765 size_t msgsz; 766 int msgflg; 767 long mtype; 768 { 769 int msqix, segs_needed, error = 0; 770 register struct msqid_kernel *msqkptr; 771 register struct msg *msghdr; 772 struct prison *rpr; 773 short next; 774 #ifdef RACCT 775 size_t saved_msgsz; 776 #endif 777 778 rpr = msg_find_prison(td->td_ucred); 779 if (rpr == NULL) 780 return (ENOSYS); 781 782 mtx_lock(&msq_mtx); 783 msqix = IPCID_TO_IX(msqid); 784 785 if (msqix < 0 || msqix >= msginfo.msgmni) { 786 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 787 msginfo.msgmni)); 788 error = EINVAL; 789 goto done2; 790 } 791 792 msqkptr = &msqids[msqix]; 793 if (msqkptr->u.msg_qbytes == 0) { 794 DPRINTF(("no such message queue id\n")); 795 error = EINVAL; 796 goto done2; 797 } 798 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 799 DPRINTF(("wrong sequence number\n")); 800 error = EINVAL; 801 goto done2; 802 } 803 804 if ((error = msq_prison_cansee(rpr, msqkptr))) { 805 DPRINTF(("requester can't see prison\n")); 806 goto done2; 807 } 808 809 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 810 DPRINTF(("requester doesn't have write access\n")); 811 goto done2; 812 } 813 814 #ifdef MAC 815 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 816 if (error != 0) 817 goto done2; 818 #endif 819 820 #ifdef RACCT 821 if (racct_enable) { 822 PROC_LOCK(td->td_proc); 823 if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) { 824 PROC_UNLOCK(td->td_proc); 825 error = EAGAIN; 826 goto done2; 827 } 828 saved_msgsz = msgsz; 829 if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) { 830 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 831 PROC_UNLOCK(td->td_proc); 832 error = EAGAIN; 833 goto done2; 834 } 835 PROC_UNLOCK(td->td_proc); 836 } 837 #endif 838 839 segs_needed = howmany(msgsz, msginfo.msgssz); 840 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 841 msginfo.msgssz, segs_needed)); 842 for (;;) { 843 int need_more_resources = 0; 844 845 /* 846 * check msgsz 847 * (inside this loop in case msg_qbytes changes while we sleep) 848 */ 849 850 if (msgsz > msqkptr->u.msg_qbytes) { 851 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 852 error = EINVAL; 853 goto done3; 854 } 855 856 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 857 DPRINTF(("msqid is locked\n")); 858 need_more_resources = 1; 859 } 860 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 861 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 862 need_more_resources = 1; 863 } 864 if (segs_needed > nfree_msgmaps) { 865 DPRINTF(("segs_needed > nfree_msgmaps\n")); 866 need_more_resources = 1; 867 } 868 if (free_msghdrs == NULL) { 869 DPRINTF(("no more msghdrs\n")); 870 need_more_resources = 1; 871 } 872 873 if (need_more_resources) { 874 int we_own_it; 875 876 if ((msgflg & IPC_NOWAIT) != 0) { 877 DPRINTF(("need more resources but caller " 878 "doesn't want to wait\n")); 879 error = EAGAIN; 880 goto done3; 881 } 882 883 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 884 DPRINTF(("we don't own the msqid_ds\n")); 885 we_own_it = 0; 886 } else { 887 /* Force later arrivals to wait for our 888 request */ 889 DPRINTF(("we own the msqid_ds\n")); 890 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 891 we_own_it = 1; 892 } 893 DPRINTF(("msgsnd: goodnight\n")); 894 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 895 "msgsnd", hz); 896 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 897 if (we_own_it) 898 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 899 if (error == EWOULDBLOCK) { 900 DPRINTF(("msgsnd: timed out\n")); 901 continue; 902 } 903 if (error != 0) { 904 DPRINTF(("msgsnd: interrupted system call\n")); 905 error = EINTR; 906 goto done3; 907 } 908 909 /* 910 * Make sure that the msq queue still exists 911 */ 912 913 if (msqkptr->u.msg_qbytes == 0) { 914 DPRINTF(("msqid deleted\n")); 915 error = EIDRM; 916 goto done3; 917 } 918 919 } else { 920 DPRINTF(("got all the resources that we need\n")); 921 break; 922 } 923 } 924 925 /* 926 * We have the resources that we need. 927 * Make sure! 928 */ 929 930 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 931 panic("msg_perm.mode & MSG_LOCKED"); 932 if (segs_needed > nfree_msgmaps) 933 panic("segs_needed > nfree_msgmaps"); 934 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 935 panic("msgsz + msg_cbytes > msg_qbytes"); 936 if (free_msghdrs == NULL) 937 panic("no more msghdrs"); 938 939 /* 940 * Re-lock the msqid_ds in case we page-fault when copying in the 941 * message 942 */ 943 944 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 945 panic("msqid_ds is already locked"); 946 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 947 948 /* 949 * Allocate a message header 950 */ 951 952 msghdr = free_msghdrs; 953 free_msghdrs = msghdr->msg_next; 954 msghdr->msg_spot = -1; 955 msghdr->msg_ts = msgsz; 956 msghdr->msg_type = mtype; 957 #ifdef MAC 958 /* 959 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 960 * immediately? Or, should it be checked just before the msg is 961 * enqueued in the msgq (as it is done now)? 962 */ 963 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 964 #endif 965 966 /* 967 * Allocate space for the message 968 */ 969 970 while (segs_needed > 0) { 971 if (nfree_msgmaps <= 0) 972 panic("not enough msgmaps"); 973 if (free_msgmaps == -1) 974 panic("nil free_msgmaps"); 975 next = free_msgmaps; 976 if (next <= -1) 977 panic("next too low #1"); 978 if (next >= msginfo.msgseg) 979 panic("next out of range #1"); 980 DPRINTF(("allocating segment %d to message\n", next)); 981 free_msgmaps = msgmaps[next].next; 982 nfree_msgmaps--; 983 msgmaps[next].next = msghdr->msg_spot; 984 msghdr->msg_spot = next; 985 segs_needed--; 986 } 987 988 /* 989 * Validate the message type 990 */ 991 992 if (msghdr->msg_type < 1) { 993 msg_freehdr(msghdr); 994 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 995 wakeup(msqkptr); 996 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 997 error = EINVAL; 998 goto done3; 999 } 1000 1001 /* 1002 * Copy in the message body 1003 */ 1004 1005 next = msghdr->msg_spot; 1006 while (msgsz > 0) { 1007 size_t tlen; 1008 if (msgsz > msginfo.msgssz) 1009 tlen = msginfo.msgssz; 1010 else 1011 tlen = msgsz; 1012 if (next <= -1) 1013 panic("next too low #2"); 1014 if (next >= msginfo.msgseg) 1015 panic("next out of range #2"); 1016 mtx_unlock(&msq_mtx); 1017 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 1018 tlen)) != 0) { 1019 mtx_lock(&msq_mtx); 1020 DPRINTF(("error %d copying in message segment\n", 1021 error)); 1022 msg_freehdr(msghdr); 1023 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1024 wakeup(msqkptr); 1025 goto done3; 1026 } 1027 mtx_lock(&msq_mtx); 1028 msgsz -= tlen; 1029 msgp = (const char *)msgp + tlen; 1030 next = msgmaps[next].next; 1031 } 1032 if (next != -1) 1033 panic("didn't use all the msg segments"); 1034 1035 /* 1036 * We've got the message. Unlock the msqid_ds. 1037 */ 1038 1039 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1040 1041 /* 1042 * Make sure that the msqid_ds is still allocated. 1043 */ 1044 1045 if (msqkptr->u.msg_qbytes == 0) { 1046 msg_freehdr(msghdr); 1047 wakeup(msqkptr); 1048 error = EIDRM; 1049 goto done3; 1050 } 1051 1052 #ifdef MAC 1053 /* 1054 * Note: Since the task/thread allocates the msghdr and usually 1055 * primes it with its own MAC label, for a majority of policies, it 1056 * won't be necessary to check whether the msghdr has access 1057 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 1058 * suffice in that case. However, this hook may be required where 1059 * individual policies derive a non-identical label for the msghdr 1060 * from the current thread label and may want to check the msghdr 1061 * enqueue permissions, along with read/write permissions to the 1062 * msgq. 1063 */ 1064 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 1065 if (error != 0) { 1066 msg_freehdr(msghdr); 1067 wakeup(msqkptr); 1068 goto done3; 1069 } 1070 #endif 1071 1072 /* 1073 * Put the message into the queue 1074 */ 1075 if (msqkptr->u.msg_first == NULL) { 1076 msqkptr->u.msg_first = msghdr; 1077 msqkptr->u.msg_last = msghdr; 1078 } else { 1079 msqkptr->u.msg_last->msg_next = msghdr; 1080 msqkptr->u.msg_last = msghdr; 1081 } 1082 msqkptr->u.msg_last->msg_next = NULL; 1083 1084 msqkptr->u.msg_cbytes += msghdr->msg_ts; 1085 msqkptr->u.msg_qnum++; 1086 msqkptr->u.msg_lspid = td->td_proc->p_pid; 1087 msqkptr->u.msg_stime = time_second; 1088 1089 wakeup(msqkptr); 1090 td->td_retval[0] = 0; 1091 done3: 1092 #ifdef RACCT 1093 if (racct_enable && error != 0) { 1094 PROC_LOCK(td->td_proc); 1095 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 1096 racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz); 1097 PROC_UNLOCK(td->td_proc); 1098 } 1099 #endif 1100 done2: 1101 mtx_unlock(&msq_mtx); 1102 return (error); 1103 } 1104 1105 int 1106 sys_msgsnd(td, uap) 1107 struct thread *td; 1108 register struct msgsnd_args *uap; 1109 { 1110 int error; 1111 long mtype; 1112 1113 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 1114 uap->msgsz, uap->msgflg)); 1115 1116 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 1117 DPRINTF(("error %d copying the message type\n", error)); 1118 return (error); 1119 } 1120 return (kern_msgsnd(td, uap->msqid, 1121 (const char *)uap->msgp + sizeof(mtype), 1122 uap->msgsz, uap->msgflg, mtype)); 1123 } 1124 1125 #ifndef _SYS_SYSPROTO_H_ 1126 struct msgrcv_args { 1127 int msqid; 1128 void *msgp; 1129 size_t msgsz; 1130 long msgtyp; 1131 int msgflg; 1132 }; 1133 #endif 1134 int 1135 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1136 struct thread *td; 1137 int msqid; 1138 void *msgp; /* XXX msgp is actually mtext. */ 1139 size_t msgsz; 1140 long msgtyp; 1141 int msgflg; 1142 long *mtype; 1143 { 1144 size_t len; 1145 register struct msqid_kernel *msqkptr; 1146 register struct msg *msghdr; 1147 struct prison *rpr; 1148 int msqix, error = 0; 1149 short next; 1150 1151 rpr = msg_find_prison(td->td_ucred); 1152 if (rpr == NULL) 1153 return (ENOSYS); 1154 1155 msqix = IPCID_TO_IX(msqid); 1156 1157 if (msqix < 0 || msqix >= msginfo.msgmni) { 1158 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1159 msginfo.msgmni)); 1160 return (EINVAL); 1161 } 1162 1163 msqkptr = &msqids[msqix]; 1164 mtx_lock(&msq_mtx); 1165 if (msqkptr->u.msg_qbytes == 0) { 1166 DPRINTF(("no such message queue id\n")); 1167 error = EINVAL; 1168 goto done2; 1169 } 1170 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1171 DPRINTF(("wrong sequence number\n")); 1172 error = EINVAL; 1173 goto done2; 1174 } 1175 1176 if ((error = msq_prison_cansee(rpr, msqkptr))) { 1177 DPRINTF(("requester can't see prison\n")); 1178 goto done2; 1179 } 1180 1181 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1182 DPRINTF(("requester doesn't have read access\n")); 1183 goto done2; 1184 } 1185 1186 #ifdef MAC 1187 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1188 if (error != 0) 1189 goto done2; 1190 #endif 1191 1192 msghdr = NULL; 1193 while (msghdr == NULL) { 1194 if (msgtyp == 0) { 1195 msghdr = msqkptr->u.msg_first; 1196 if (msghdr != NULL) { 1197 if (msgsz < msghdr->msg_ts && 1198 (msgflg & MSG_NOERROR) == 0) { 1199 DPRINTF(("first message on the queue " 1200 "is too big (want %zu, got %d)\n", 1201 msgsz, msghdr->msg_ts)); 1202 error = E2BIG; 1203 goto done2; 1204 } 1205 #ifdef MAC 1206 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1207 msghdr); 1208 if (error != 0) 1209 goto done2; 1210 #endif 1211 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1212 msqkptr->u.msg_first = NULL; 1213 msqkptr->u.msg_last = NULL; 1214 } else { 1215 msqkptr->u.msg_first = msghdr->msg_next; 1216 if (msqkptr->u.msg_first == NULL) 1217 panic("msg_first/last screwed up #1"); 1218 } 1219 } 1220 } else { 1221 struct msg *previous; 1222 struct msg **prev; 1223 1224 previous = NULL; 1225 prev = &(msqkptr->u.msg_first); 1226 while ((msghdr = *prev) != NULL) { 1227 /* 1228 * Is this message's type an exact match or is 1229 * this message's type less than or equal to 1230 * the absolute value of a negative msgtyp? 1231 * Note that the second half of this test can 1232 * NEVER be true if msgtyp is positive since 1233 * msg_type is always positive! 1234 */ 1235 1236 if (msgtyp == msghdr->msg_type || 1237 msghdr->msg_type <= -msgtyp) { 1238 DPRINTF(("found message type %ld, " 1239 "requested %ld\n", 1240 msghdr->msg_type, msgtyp)); 1241 if (msgsz < msghdr->msg_ts && 1242 (msgflg & MSG_NOERROR) == 0) { 1243 DPRINTF(("requested message " 1244 "on the queue is too big " 1245 "(want %zu, got %hu)\n", 1246 msgsz, msghdr->msg_ts)); 1247 error = E2BIG; 1248 goto done2; 1249 } 1250 #ifdef MAC 1251 error = mac_sysvmsq_check_msgrcv( 1252 td->td_ucred, msghdr); 1253 if (error != 0) 1254 goto done2; 1255 #endif 1256 *prev = msghdr->msg_next; 1257 if (msghdr == msqkptr->u.msg_last) { 1258 if (previous == NULL) { 1259 if (prev != 1260 &msqkptr->u.msg_first) 1261 panic("msg_first/last screwed up #2"); 1262 msqkptr->u.msg_first = 1263 NULL; 1264 msqkptr->u.msg_last = 1265 NULL; 1266 } else { 1267 if (prev == 1268 &msqkptr->u.msg_first) 1269 panic("msg_first/last screwed up #3"); 1270 msqkptr->u.msg_last = 1271 previous; 1272 } 1273 } 1274 break; 1275 } 1276 previous = msghdr; 1277 prev = &(msghdr->msg_next); 1278 } 1279 } 1280 1281 /* 1282 * We've either extracted the msghdr for the appropriate 1283 * message or there isn't one. 1284 * If there is one then bail out of this loop. 1285 */ 1286 1287 if (msghdr != NULL) 1288 break; 1289 1290 /* 1291 * Hmph! No message found. Does the user want to wait? 1292 */ 1293 1294 if ((msgflg & IPC_NOWAIT) != 0) { 1295 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1296 msgtyp)); 1297 /* The SVID says to return ENOMSG. */ 1298 error = ENOMSG; 1299 goto done2; 1300 } 1301 1302 /* 1303 * Wait for something to happen 1304 */ 1305 1306 DPRINTF(("msgrcv: goodnight\n")); 1307 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1308 "msgrcv", 0); 1309 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1310 1311 if (error != 0) { 1312 DPRINTF(("msgrcv: interrupted system call\n")); 1313 error = EINTR; 1314 goto done2; 1315 } 1316 1317 /* 1318 * Make sure that the msq queue still exists 1319 */ 1320 1321 if (msqkptr->u.msg_qbytes == 0 || 1322 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1323 DPRINTF(("msqid deleted\n")); 1324 error = EIDRM; 1325 goto done2; 1326 } 1327 } 1328 1329 /* 1330 * Return the message to the user. 1331 * 1332 * First, do the bookkeeping (before we risk being interrupted). 1333 */ 1334 1335 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1336 msqkptr->u.msg_qnum--; 1337 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1338 msqkptr->u.msg_rtime = time_second; 1339 1340 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1); 1341 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts); 1342 1343 /* 1344 * Make msgsz the actual amount that we'll be returning. 1345 * Note that this effectively truncates the message if it is too long 1346 * (since msgsz is never increased). 1347 */ 1348 1349 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1350 msghdr->msg_ts)); 1351 if (msgsz > msghdr->msg_ts) 1352 msgsz = msghdr->msg_ts; 1353 *mtype = msghdr->msg_type; 1354 1355 /* 1356 * Return the segments to the user 1357 */ 1358 1359 next = msghdr->msg_spot; 1360 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1361 size_t tlen; 1362 1363 if (msgsz - len > msginfo.msgssz) 1364 tlen = msginfo.msgssz; 1365 else 1366 tlen = msgsz - len; 1367 if (next <= -1) 1368 panic("next too low #3"); 1369 if (next >= msginfo.msgseg) 1370 panic("next out of range #3"); 1371 mtx_unlock(&msq_mtx); 1372 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1373 mtx_lock(&msq_mtx); 1374 if (error != 0) { 1375 DPRINTF(("error (%d) copying out message segment\n", 1376 error)); 1377 msg_freehdr(msghdr); 1378 wakeup(msqkptr); 1379 goto done2; 1380 } 1381 msgp = (char *)msgp + tlen; 1382 next = msgmaps[next].next; 1383 } 1384 1385 /* 1386 * Done, return the actual number of bytes copied out. 1387 */ 1388 1389 msg_freehdr(msghdr); 1390 wakeup(msqkptr); 1391 td->td_retval[0] = msgsz; 1392 done2: 1393 mtx_unlock(&msq_mtx); 1394 return (error); 1395 } 1396 1397 int 1398 sys_msgrcv(td, uap) 1399 struct thread *td; 1400 register struct msgrcv_args *uap; 1401 { 1402 int error; 1403 long mtype; 1404 1405 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1406 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1407 1408 if ((error = kern_msgrcv(td, uap->msqid, 1409 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1410 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1411 return (error); 1412 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1413 DPRINTF(("error %d copying the message type\n", error)); 1414 return (error); 1415 } 1416 1417 static int 1418 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1419 { 1420 struct msqid_kernel tmsqk; 1421 struct prison *pr, *rpr; 1422 int error, i; 1423 1424 pr = req->td->td_ucred->cr_prison; 1425 rpr = msg_find_prison(req->td->td_ucred); 1426 error = 0; 1427 for (i = 0; i < msginfo.msgmni; i++) { 1428 mtx_lock(&msq_mtx); 1429 if (msqids[i].u.msg_qbytes == 0 || rpr == NULL || 1430 msq_prison_cansee(rpr, &msqids[i]) != 0) 1431 bzero(&tmsqk, sizeof(tmsqk)); 1432 else { 1433 tmsqk = msqids[i]; 1434 if (tmsqk.cred->cr_prison != pr) 1435 tmsqk.u.msg_perm.key = IPC_PRIVATE; 1436 } 1437 mtx_unlock(&msq_mtx); 1438 error = SYSCTL_OUT(req, &tmsqk, sizeof(tmsqk)); 1439 if (error != 0) 1440 break; 1441 } 1442 return (error); 1443 } 1444 1445 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1446 "Maximum message size"); 1447 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1448 "Number of message queue identifiers"); 1449 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1450 "Maximum number of bytes in a queue"); 1451 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1452 "Maximum number of messages in the system"); 1453 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1454 "Size of a message segment"); 1455 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1456 "Number of message segments"); 1457 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, 1458 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1459 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1460 1461 static int 1462 msg_prison_check(void *obj, void *data) 1463 { 1464 struct prison *pr = obj; 1465 struct prison *prpr; 1466 struct vfsoptlist *opts = data; 1467 int error, jsys; 1468 1469 /* 1470 * sysvmsg is a jailsys integer. 1471 * It must be "disable" if the parent jail is disabled. 1472 */ 1473 error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1474 if (error != ENOENT) { 1475 if (error != 0) 1476 return (error); 1477 switch (jsys) { 1478 case JAIL_SYS_DISABLE: 1479 break; 1480 case JAIL_SYS_NEW: 1481 case JAIL_SYS_INHERIT: 1482 prison_lock(pr->pr_parent); 1483 prpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1484 prison_unlock(pr->pr_parent); 1485 if (prpr == NULL) 1486 return (EPERM); 1487 break; 1488 default: 1489 return (EINVAL); 1490 } 1491 } 1492 1493 return (0); 1494 } 1495 1496 static int 1497 msg_prison_set(void *obj, void *data) 1498 { 1499 struct prison *pr = obj; 1500 struct prison *tpr, *orpr, *nrpr, *trpr; 1501 struct vfsoptlist *opts = data; 1502 void *rsv; 1503 int jsys, descend; 1504 1505 /* 1506 * sysvmsg controls which jail is the root of the associated msgs (this 1507 * jail or same as the parent), or if the feature is available at all. 1508 */ 1509 if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT) 1510 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0) 1511 ? JAIL_SYS_INHERIT 1512 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0) 1513 ? JAIL_SYS_DISABLE 1514 : -1; 1515 if (jsys == JAIL_SYS_DISABLE) { 1516 prison_lock(pr); 1517 orpr = osd_jail_get(pr, msg_prison_slot); 1518 if (orpr != NULL) 1519 osd_jail_del(pr, msg_prison_slot); 1520 prison_unlock(pr); 1521 if (orpr != NULL) { 1522 if (orpr == pr) 1523 msg_prison_cleanup(pr); 1524 /* Disable all child jails as well. */ 1525 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1526 prison_lock(tpr); 1527 trpr = osd_jail_get(tpr, msg_prison_slot); 1528 if (trpr != NULL) { 1529 osd_jail_del(tpr, msg_prison_slot); 1530 prison_unlock(tpr); 1531 if (trpr == tpr) 1532 msg_prison_cleanup(tpr); 1533 } else { 1534 prison_unlock(tpr); 1535 descend = 0; 1536 } 1537 } 1538 } 1539 } else if (jsys != -1) { 1540 if (jsys == JAIL_SYS_NEW) 1541 nrpr = pr; 1542 else { 1543 prison_lock(pr->pr_parent); 1544 nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1545 prison_unlock(pr->pr_parent); 1546 } 1547 rsv = osd_reserve(msg_prison_slot); 1548 prison_lock(pr); 1549 orpr = osd_jail_get(pr, msg_prison_slot); 1550 if (orpr != nrpr) 1551 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 1552 nrpr); 1553 else 1554 osd_free_reserved(rsv); 1555 prison_unlock(pr); 1556 if (orpr != nrpr) { 1557 if (orpr == pr) 1558 msg_prison_cleanup(pr); 1559 if (orpr != NULL) { 1560 /* Change child jails matching the old root, */ 1561 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1562 prison_lock(tpr); 1563 trpr = osd_jail_get(tpr, 1564 msg_prison_slot); 1565 if (trpr == orpr) { 1566 (void)osd_jail_set(tpr, 1567 msg_prison_slot, nrpr); 1568 prison_unlock(tpr); 1569 if (trpr == tpr) 1570 msg_prison_cleanup(tpr); 1571 } else { 1572 prison_unlock(tpr); 1573 descend = 0; 1574 } 1575 } 1576 } 1577 } 1578 } 1579 1580 return (0); 1581 } 1582 1583 static int 1584 msg_prison_get(void *obj, void *data) 1585 { 1586 struct prison *pr = obj; 1587 struct prison *rpr; 1588 struct vfsoptlist *opts = data; 1589 int error, jsys; 1590 1591 /* Set sysvmsg based on the jail's root prison. */ 1592 prison_lock(pr); 1593 rpr = osd_jail_get(pr, msg_prison_slot); 1594 prison_unlock(pr); 1595 jsys = rpr == NULL ? JAIL_SYS_DISABLE 1596 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; 1597 error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1598 if (error == ENOENT) 1599 error = 0; 1600 return (error); 1601 } 1602 1603 static int 1604 msg_prison_remove(void *obj, void *data __unused) 1605 { 1606 struct prison *pr = obj; 1607 struct prison *rpr; 1608 1609 prison_lock(pr); 1610 rpr = osd_jail_get(pr, msg_prison_slot); 1611 prison_unlock(pr); 1612 if (rpr == pr) 1613 msg_prison_cleanup(pr); 1614 return (0); 1615 } 1616 1617 static void 1618 msg_prison_cleanup(struct prison *pr) 1619 { 1620 struct msqid_kernel *msqkptr; 1621 int i; 1622 1623 /* Remove any msqs that belong to this jail. */ 1624 mtx_lock(&msq_mtx); 1625 for (i = 0; i < msginfo.msgmni; i++) { 1626 msqkptr = &msqids[i]; 1627 if (msqkptr->u.msg_qbytes != 0 && 1628 msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr) 1629 msq_remove(msqkptr); 1630 } 1631 mtx_unlock(&msq_mtx); 1632 } 1633 1634 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues"); 1635 1636 #ifdef COMPAT_FREEBSD32 1637 int 1638 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap) 1639 { 1640 1641 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1642 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1643 AUDIT_ARG_SVIPC_WHICH(uap->which); 1644 switch (uap->which) { 1645 case 0: 1646 return (freebsd7_freebsd32_msgctl(td, 1647 (struct freebsd7_freebsd32_msgctl_args *)&uap->a2)); 1648 case 2: 1649 return (freebsd32_msgsnd(td, 1650 (struct freebsd32_msgsnd_args *)&uap->a2)); 1651 case 3: 1652 return (freebsd32_msgrcv(td, 1653 (struct freebsd32_msgrcv_args *)&uap->a2)); 1654 default: 1655 return (sys_msgsys(td, (struct msgsys_args *)uap)); 1656 } 1657 #else 1658 return (nosys(td, NULL)); 1659 #endif 1660 } 1661 1662 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1663 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1664 int 1665 freebsd7_freebsd32_msgctl(struct thread *td, 1666 struct freebsd7_freebsd32_msgctl_args *uap) 1667 { 1668 struct msqid_ds msqbuf; 1669 struct msqid_ds32_old msqbuf32; 1670 int error; 1671 1672 if (uap->cmd == IPC_SET) { 1673 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1674 if (error) 1675 return (error); 1676 freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1677 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1678 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1679 CP(msqbuf32, msqbuf, msg_cbytes); 1680 CP(msqbuf32, msqbuf, msg_qnum); 1681 CP(msqbuf32, msqbuf, msg_qbytes); 1682 CP(msqbuf32, msqbuf, msg_lspid); 1683 CP(msqbuf32, msqbuf, msg_lrpid); 1684 CP(msqbuf32, msqbuf, msg_stime); 1685 CP(msqbuf32, msqbuf, msg_rtime); 1686 CP(msqbuf32, msqbuf, msg_ctime); 1687 } 1688 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1689 if (error) 1690 return (error); 1691 if (uap->cmd == IPC_STAT) { 1692 bzero(&msqbuf32, sizeof(msqbuf32)); 1693 freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1694 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1695 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1696 CP(msqbuf, msqbuf32, msg_cbytes); 1697 CP(msqbuf, msqbuf32, msg_qnum); 1698 CP(msqbuf, msqbuf32, msg_qbytes); 1699 CP(msqbuf, msqbuf32, msg_lspid); 1700 CP(msqbuf, msqbuf32, msg_lrpid); 1701 CP(msqbuf, msqbuf32, msg_stime); 1702 CP(msqbuf, msqbuf32, msg_rtime); 1703 CP(msqbuf, msqbuf32, msg_ctime); 1704 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1705 } 1706 return (error); 1707 } 1708 #endif 1709 1710 int 1711 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap) 1712 { 1713 struct msqid_ds msqbuf; 1714 struct msqid_ds32 msqbuf32; 1715 int error; 1716 1717 if (uap->cmd == IPC_SET) { 1718 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1719 if (error) 1720 return (error); 1721 freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1722 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1723 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1724 CP(msqbuf32, msqbuf, msg_cbytes); 1725 CP(msqbuf32, msqbuf, msg_qnum); 1726 CP(msqbuf32, msqbuf, msg_qbytes); 1727 CP(msqbuf32, msqbuf, msg_lspid); 1728 CP(msqbuf32, msqbuf, msg_lrpid); 1729 CP(msqbuf32, msqbuf, msg_stime); 1730 CP(msqbuf32, msqbuf, msg_rtime); 1731 CP(msqbuf32, msqbuf, msg_ctime); 1732 } 1733 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1734 if (error) 1735 return (error); 1736 if (uap->cmd == IPC_STAT) { 1737 freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1738 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1739 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1740 CP(msqbuf, msqbuf32, msg_cbytes); 1741 CP(msqbuf, msqbuf32, msg_qnum); 1742 CP(msqbuf, msqbuf32, msg_qbytes); 1743 CP(msqbuf, msqbuf32, msg_lspid); 1744 CP(msqbuf, msqbuf32, msg_lrpid); 1745 CP(msqbuf, msqbuf32, msg_stime); 1746 CP(msqbuf, msqbuf32, msg_rtime); 1747 CP(msqbuf, msqbuf32, msg_ctime); 1748 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1749 } 1750 return (error); 1751 } 1752 1753 int 1754 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap) 1755 { 1756 const void *msgp; 1757 long mtype; 1758 int32_t mtype32; 1759 int error; 1760 1761 msgp = PTRIN(uap->msgp); 1762 if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0) 1763 return (error); 1764 mtype = mtype32; 1765 return (kern_msgsnd(td, uap->msqid, 1766 (const char *)msgp + sizeof(mtype32), 1767 uap->msgsz, uap->msgflg, mtype)); 1768 } 1769 1770 int 1771 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap) 1772 { 1773 void *msgp; 1774 long mtype; 1775 int32_t mtype32; 1776 int error; 1777 1778 msgp = PTRIN(uap->msgp); 1779 if ((error = kern_msgrcv(td, uap->msqid, 1780 (char *)msgp + sizeof(mtype32), uap->msgsz, 1781 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1782 return (error); 1783 mtype32 = (int32_t)mtype; 1784 return (copyout(&mtype32, msgp, sizeof(mtype32))); 1785 } 1786 #endif 1787 1788 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1789 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1790 1791 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 1792 static sy_call_t *msgcalls[] = { 1793 (sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget, 1794 (sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv 1795 }; 1796 1797 /* 1798 * Entry point for all MSG calls. 1799 */ 1800 int 1801 sys_msgsys(td, uap) 1802 struct thread *td; 1803 /* XXX actually varargs. */ 1804 struct msgsys_args /* { 1805 int which; 1806 int a2; 1807 int a3; 1808 int a4; 1809 int a5; 1810 int a6; 1811 } */ *uap; 1812 { 1813 int error; 1814 1815 AUDIT_ARG_SVIPC_WHICH(uap->which); 1816 if (uap->which < 0 || uap->which >= nitems(msgcalls)) 1817 return (EINVAL); 1818 error = (*msgcalls[uap->which])(td, &uap->a2); 1819 return (error); 1820 } 1821 1822 #ifndef CP 1823 #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0) 1824 #endif 1825 1826 #ifndef _SYS_SYSPROTO_H_ 1827 struct freebsd7_msgctl_args { 1828 int msqid; 1829 int cmd; 1830 struct msqid_ds_old *buf; 1831 }; 1832 #endif 1833 int 1834 freebsd7_msgctl(td, uap) 1835 struct thread *td; 1836 struct freebsd7_msgctl_args *uap; 1837 { 1838 struct msqid_ds_old msqold; 1839 struct msqid_ds msqbuf; 1840 int error; 1841 1842 DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd, 1843 uap->buf)); 1844 if (uap->cmd == IPC_SET) { 1845 error = copyin(uap->buf, &msqold, sizeof(msqold)); 1846 if (error) 1847 return (error); 1848 ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm); 1849 CP(msqold, msqbuf, msg_first); 1850 CP(msqold, msqbuf, msg_last); 1851 CP(msqold, msqbuf, msg_cbytes); 1852 CP(msqold, msqbuf, msg_qnum); 1853 CP(msqold, msqbuf, msg_qbytes); 1854 CP(msqold, msqbuf, msg_lspid); 1855 CP(msqold, msqbuf, msg_lrpid); 1856 CP(msqold, msqbuf, msg_stime); 1857 CP(msqold, msqbuf, msg_rtime); 1858 CP(msqold, msqbuf, msg_ctime); 1859 } 1860 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1861 if (error) 1862 return (error); 1863 if (uap->cmd == IPC_STAT) { 1864 bzero(&msqold, sizeof(msqold)); 1865 ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm); 1866 CP(msqbuf, msqold, msg_first); 1867 CP(msqbuf, msqold, msg_last); 1868 CP(msqbuf, msqold, msg_cbytes); 1869 CP(msqbuf, msqold, msg_qnum); 1870 CP(msqbuf, msqold, msg_qbytes); 1871 CP(msqbuf, msqold, msg_lspid); 1872 CP(msqbuf, msqold, msg_lrpid); 1873 CP(msqbuf, msqold, msg_stime); 1874 CP(msqbuf, msqold, msg_rtime); 1875 CP(msqbuf, msqold, msg_ctime); 1876 error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old)); 1877 } 1878 return (error); 1879 } 1880 1881 #undef CP 1882 1883 #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 || 1884 COMPAT_FREEBSD7 */ 1885