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 * Copyright (c) 2016-2017 Robert N. M. Watson 22 * All rights reserved. 23 * 24 * This software was developed for the FreeBSD Project in part by McAfee 25 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR 26 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research 27 * program. 28 * 29 * Portions of this software were developed by BAE Systems, the University of 30 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL 31 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent 32 * Computing (TC) research program. 33 * 34 * Redistribution and use in source and binary forms, with or without 35 * modification, are permitted provided that the following conditions 36 * are met: 37 * 1. Redistributions of source code must retain the above copyright 38 * notice, this list of conditions and the following disclaimer. 39 * 2. Redistributions in binary form must reproduce the above copyright 40 * notice, this list of conditions and the following disclaimer in the 41 * documentation and/or other materials provided with the distribution. 42 * 43 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 46 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 53 * SUCH DAMAGE. 54 */ 55 56 #include <sys/cdefs.h> 57 __FBSDID("$FreeBSD$"); 58 59 #include "opt_compat.h" 60 #include "opt_sysvipc.h" 61 62 #include <sys/param.h> 63 #include <sys/systm.h> 64 #include <sys/sysproto.h> 65 #include <sys/kernel.h> 66 #include <sys/priv.h> 67 #include <sys/proc.h> 68 #include <sys/lock.h> 69 #include <sys/mutex.h> 70 #include <sys/module.h> 71 #include <sys/mount.h> 72 #include <sys/msg.h> 73 #include <sys/racct.h> 74 #include <sys/sx.h> 75 #include <sys/syscall.h> 76 #include <sys/syscallsubr.h> 77 #include <sys/sysent.h> 78 #include <sys/sysctl.h> 79 #include <sys/malloc.h> 80 #include <sys/jail.h> 81 82 #include <security/audit/audit.h> 83 #include <security/mac/mac_framework.h> 84 85 FEATURE(sysv_msg, "System V message queues support"); 86 87 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 88 89 static int msginit(void); 90 static int msgunload(void); 91 static int sysvmsg_modload(struct module *, int, void *); 92 static void msq_remove(struct msqid_kernel *); 93 static struct prison *msg_find_prison(struct ucred *); 94 static int msq_prison_cansee(struct prison *, struct msqid_kernel *); 95 static int msg_prison_check(void *, void *); 96 static int msg_prison_set(void *, void *); 97 static int msg_prison_get(void *, void *); 98 static int msg_prison_remove(void *, void *); 99 static void msg_prison_cleanup(struct prison *); 100 101 102 #ifdef MSG_DEBUG 103 #define DPRINTF(a) printf a 104 #else 105 #define DPRINTF(a) (void)0 106 #endif 107 108 static void msg_freehdr(struct msg *msghdr); 109 110 #ifndef MSGSSZ 111 #define MSGSSZ 8 /* Each segment must be 2^N long */ 112 #endif 113 #ifndef MSGSEG 114 #define MSGSEG 2048 /* must be less than 32767 */ 115 #endif 116 #define MSGMAX (MSGSSZ*MSGSEG) 117 #ifndef MSGMNB 118 #define MSGMNB 2048 /* max # of bytes in a queue */ 119 #endif 120 #ifndef MSGMNI 121 #define MSGMNI 40 122 #endif 123 #ifndef MSGTQL 124 #define MSGTQL 40 125 #endif 126 127 /* 128 * Based on the configuration parameters described in an SVR2 (yes, two) 129 * config(1m) man page. 130 * 131 * Each message is broken up and stored in segments that are msgssz bytes 132 * long. For efficiency reasons, this should be a power of two. Also, 133 * it doesn't make sense if it is less than 8 or greater than about 256. 134 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 135 * two between 8 and 1024 inclusive (and panic's if it isn't). 136 */ 137 struct msginfo msginfo = { 138 MSGMAX, /* max chars in a message */ 139 MSGMNI, /* # of message queue identifiers */ 140 MSGMNB, /* max chars in a queue */ 141 MSGTQL, /* max messages in system */ 142 MSGSSZ, /* size of a message segment */ 143 /* (must be small power of 2 greater than 4) */ 144 MSGSEG /* number of message segments */ 145 }; 146 147 /* 148 * macros to convert between msqid_ds's and msqid's. 149 * (specific to this implementation) 150 */ 151 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 152 #define MSQID_IX(id) ((id) & 0xffff) 153 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 154 155 /* 156 * The rest of this file is specific to this particular implementation. 157 */ 158 159 struct msgmap { 160 short next; /* next segment in buffer */ 161 /* -1 -> available */ 162 /* 0..(MSGSEG-1) -> index of next segment */ 163 }; 164 165 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 166 167 static int nfree_msgmaps; /* # of free map entries */ 168 static short free_msgmaps; /* head of linked list of free map entries */ 169 static struct msg *free_msghdrs;/* list of free msg headers */ 170 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 171 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 172 static struct msg *msghdrs; /* MSGTQL msg headers */ 173 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 174 static struct mtx msq_mtx; /* global mutex for message queues. */ 175 static unsigned msg_prison_slot;/* prison OSD slot */ 176 177 static struct syscall_helper_data msg_syscalls[] = { 178 SYSCALL_INIT_HELPER(msgctl), 179 SYSCALL_INIT_HELPER(msgget), 180 SYSCALL_INIT_HELPER(msgsnd), 181 SYSCALL_INIT_HELPER(msgrcv), 182 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 183 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 184 SYSCALL_INIT_HELPER(msgsys), 185 SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl), 186 #endif 187 SYSCALL_INIT_LAST 188 }; 189 190 #ifdef COMPAT_FREEBSD32 191 #include <compat/freebsd32/freebsd32.h> 192 #include <compat/freebsd32/freebsd32_ipc.h> 193 #include <compat/freebsd32/freebsd32_proto.h> 194 #include <compat/freebsd32/freebsd32_signal.h> 195 #include <compat/freebsd32/freebsd32_syscall.h> 196 #include <compat/freebsd32/freebsd32_util.h> 197 198 static struct syscall_helper_data msg32_syscalls[] = { 199 SYSCALL32_INIT_HELPER(freebsd32_msgctl), 200 SYSCALL32_INIT_HELPER(freebsd32_msgsnd), 201 SYSCALL32_INIT_HELPER(freebsd32_msgrcv), 202 SYSCALL32_INIT_HELPER_COMPAT(msgget), 203 SYSCALL32_INIT_HELPER(freebsd32_msgsys), 204 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 205 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 206 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl), 207 #endif 208 SYSCALL_INIT_LAST 209 }; 210 #endif 211 212 static int 213 msginit() 214 { 215 struct prison *pr; 216 void **rsv; 217 int i, error; 218 osd_method_t methods[PR_MAXMETHOD] = { 219 [PR_METHOD_CHECK] = msg_prison_check, 220 [PR_METHOD_SET] = msg_prison_set, 221 [PR_METHOD_GET] = msg_prison_get, 222 [PR_METHOD_REMOVE] = msg_prison_remove, 223 }; 224 225 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 226 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 227 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 228 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 229 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 230 M_WAITOK); 231 232 /* 233 * msginfo.msgssz should be a power of two for efficiency reasons. 234 * It is also pretty silly if msginfo.msgssz is less than 8 235 * or greater than about 256 so ... 236 */ 237 238 i = 8; 239 while (i < 1024 && i != msginfo.msgssz) 240 i <<= 1; 241 if (i != msginfo.msgssz) { 242 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 243 msginfo.msgssz)); 244 panic("msginfo.msgssz not a small power of 2"); 245 } 246 247 if (msginfo.msgseg > 32767) { 248 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 249 panic("msginfo.msgseg > 32767"); 250 } 251 252 for (i = 0; i < msginfo.msgseg; i++) { 253 if (i > 0) 254 msgmaps[i-1].next = i; 255 msgmaps[i].next = -1; /* implies entry is available */ 256 } 257 free_msgmaps = 0; 258 nfree_msgmaps = msginfo.msgseg; 259 260 for (i = 0; i < msginfo.msgtql; i++) { 261 msghdrs[i].msg_type = 0; 262 if (i > 0) 263 msghdrs[i-1].msg_next = &msghdrs[i]; 264 msghdrs[i].msg_next = NULL; 265 #ifdef MAC 266 mac_sysvmsg_init(&msghdrs[i]); 267 #endif 268 } 269 free_msghdrs = &msghdrs[0]; 270 271 for (i = 0; i < msginfo.msgmni; i++) { 272 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 273 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 274 msqids[i].u.msg_perm.mode = 0; 275 #ifdef MAC 276 mac_sysvmsq_init(&msqids[i]); 277 #endif 278 } 279 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 280 281 /* Set current prisons according to their allow.sysvipc. */ 282 msg_prison_slot = osd_jail_register(NULL, methods); 283 rsv = osd_reserve(msg_prison_slot); 284 prison_lock(&prison0); 285 (void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0); 286 prison_unlock(&prison0); 287 rsv = NULL; 288 sx_slock(&allprison_lock); 289 TAILQ_FOREACH(pr, &allprison, pr_list) { 290 if (rsv == NULL) 291 rsv = osd_reserve(msg_prison_slot); 292 prison_lock(pr); 293 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) { 294 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 295 &prison0); 296 rsv = NULL; 297 } 298 prison_unlock(pr); 299 } 300 if (rsv != NULL) 301 osd_free_reserved(rsv); 302 sx_sunlock(&allprison_lock); 303 304 error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD); 305 if (error != 0) 306 return (error); 307 #ifdef COMPAT_FREEBSD32 308 error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD); 309 if (error != 0) 310 return (error); 311 #endif 312 return (0); 313 } 314 315 static int 316 msgunload() 317 { 318 struct msqid_kernel *msqkptr; 319 int msqid; 320 #ifdef MAC 321 int i; 322 #endif 323 324 syscall_helper_unregister(msg_syscalls); 325 #ifdef COMPAT_FREEBSD32 326 syscall32_helper_unregister(msg32_syscalls); 327 #endif 328 329 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 330 msqkptr = &msqids[msqid]; 331 if (msqkptr->u.msg_qbytes != 0 || 332 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 333 break; 334 } 335 if (msqid != msginfo.msgmni) 336 return (EBUSY); 337 338 if (msg_prison_slot != 0) 339 osd_jail_deregister(msg_prison_slot); 340 #ifdef MAC 341 for (i = 0; i < msginfo.msgtql; i++) 342 mac_sysvmsg_destroy(&msghdrs[i]); 343 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 344 mac_sysvmsq_destroy(&msqids[msqid]); 345 #endif 346 free(msgpool, M_MSG); 347 free(msgmaps, M_MSG); 348 free(msghdrs, M_MSG); 349 free(msqids, M_MSG); 350 mtx_destroy(&msq_mtx); 351 return (0); 352 } 353 354 355 static int 356 sysvmsg_modload(struct module *module, int cmd, void *arg) 357 { 358 int error = 0; 359 360 switch (cmd) { 361 case MOD_LOAD: 362 error = msginit(); 363 if (error != 0) 364 msgunload(); 365 break; 366 case MOD_UNLOAD: 367 error = msgunload(); 368 break; 369 case MOD_SHUTDOWN: 370 break; 371 default: 372 error = EINVAL; 373 break; 374 } 375 return (error); 376 } 377 378 static moduledata_t sysvmsg_mod = { 379 "sysvmsg", 380 &sysvmsg_modload, 381 NULL 382 }; 383 384 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 385 MODULE_VERSION(sysvmsg, 1); 386 387 static void 388 msg_freehdr(msghdr) 389 struct msg *msghdr; 390 { 391 while (msghdr->msg_ts > 0) { 392 short next; 393 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 394 panic("msghdr->msg_spot out of range"); 395 next = msgmaps[msghdr->msg_spot].next; 396 msgmaps[msghdr->msg_spot].next = free_msgmaps; 397 free_msgmaps = msghdr->msg_spot; 398 nfree_msgmaps++; 399 msghdr->msg_spot = next; 400 if (msghdr->msg_ts >= msginfo.msgssz) 401 msghdr->msg_ts -= msginfo.msgssz; 402 else 403 msghdr->msg_ts = 0; 404 } 405 if (msghdr->msg_spot != -1) 406 panic("msghdr->msg_spot != -1"); 407 msghdr->msg_next = free_msghdrs; 408 free_msghdrs = msghdr; 409 #ifdef MAC 410 mac_sysvmsg_cleanup(msghdr); 411 #endif 412 } 413 414 static void 415 msq_remove(struct msqid_kernel *msqkptr) 416 { 417 struct msg *msghdr; 418 419 racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1); 420 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum); 421 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes); 422 crfree(msqkptr->cred); 423 msqkptr->cred = NULL; 424 425 /* Free the message headers */ 426 msghdr = msqkptr->u.msg_first; 427 while (msghdr != NULL) { 428 struct msg *msghdr_tmp; 429 430 /* Free the segments of each message */ 431 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 432 msqkptr->u.msg_qnum--; 433 msghdr_tmp = msghdr; 434 msghdr = msghdr->msg_next; 435 msg_freehdr(msghdr_tmp); 436 } 437 438 if (msqkptr->u.msg_cbytes != 0) 439 panic("msg_cbytes is screwed up"); 440 if (msqkptr->u.msg_qnum != 0) 441 panic("msg_qnum is screwed up"); 442 443 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 444 445 #ifdef MAC 446 mac_sysvmsq_cleanup(msqkptr); 447 #endif 448 449 wakeup(msqkptr); 450 } 451 452 static struct prison * 453 msg_find_prison(struct ucred *cred) 454 { 455 struct prison *pr, *rpr; 456 457 pr = cred->cr_prison; 458 prison_lock(pr); 459 rpr = osd_jail_get(pr, msg_prison_slot); 460 prison_unlock(pr); 461 return rpr; 462 } 463 464 static int 465 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr) 466 { 467 468 if (msqkptr->cred == NULL || 469 !(rpr == msqkptr->cred->cr_prison || 470 prison_ischild(rpr, msqkptr->cred->cr_prison))) 471 return (EINVAL); 472 return (0); 473 } 474 475 #ifndef _SYS_SYSPROTO_H_ 476 struct msgctl_args { 477 int msqid; 478 int cmd; 479 struct msqid_ds *buf; 480 }; 481 #endif 482 int 483 sys_msgctl(td, uap) 484 struct thread *td; 485 register struct msgctl_args *uap; 486 { 487 int msqid = uap->msqid; 488 int cmd = uap->cmd; 489 struct msqid_ds msqbuf; 490 int error; 491 492 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 493 if (cmd == IPC_SET && 494 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 495 return (error); 496 error = kern_msgctl(td, msqid, cmd, &msqbuf); 497 if (cmd == IPC_STAT && error == 0) 498 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 499 return (error); 500 } 501 502 int 503 kern_msgctl(td, msqid, cmd, msqbuf) 504 struct thread *td; 505 int msqid; 506 int cmd; 507 struct msqid_ds *msqbuf; 508 { 509 int rval, error, msqix; 510 register struct msqid_kernel *msqkptr; 511 struct prison *rpr; 512 513 rpr = msg_find_prison(td->td_ucred); 514 if (rpr == NULL) 515 return (ENOSYS); 516 517 AUDIT_ARG_SVIPC_CMD(cmd); 518 AUDIT_ARG_SVIPC_ID(msqid); 519 msqix = IPCID_TO_IX(msqid); 520 521 if (msqix < 0 || msqix >= msginfo.msgmni) { 522 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 523 msginfo.msgmni)); 524 return (EINVAL); 525 } 526 527 msqkptr = &msqids[msqix]; 528 529 mtx_lock(&msq_mtx); 530 if (msqkptr->u.msg_qbytes == 0) { 531 DPRINTF(("no such msqid\n")); 532 error = EINVAL; 533 goto done2; 534 } 535 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 536 DPRINTF(("wrong sequence number\n")); 537 error = EINVAL; 538 goto done2; 539 } 540 541 error = msq_prison_cansee(rpr, msqkptr); 542 if (error != 0) { 543 DPRINTF(("requester can't see prison\n")); 544 goto done2; 545 } 546 547 #ifdef MAC 548 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 549 if (error != 0) 550 goto done2; 551 #endif 552 553 error = 0; 554 rval = 0; 555 556 switch (cmd) { 557 558 case IPC_RMID: 559 { 560 #ifdef MAC 561 struct msg *msghdr; 562 #endif 563 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 564 goto done2; 565 566 #ifdef MAC 567 /* 568 * Check that the thread has MAC access permissions to 569 * individual msghdrs. Note: We need to do this in a 570 * separate loop because the actual loop alters the 571 * msq/msghdr info as it progresses, and there is no going 572 * back if half the way through we discover that the 573 * thread cannot free a certain msghdr. The msq will get 574 * into an inconsistent state. 575 */ 576 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 577 msghdr = msghdr->msg_next) { 578 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 579 if (error != 0) 580 goto done2; 581 } 582 #endif 583 584 msq_remove(msqkptr); 585 } 586 587 break; 588 589 case IPC_SET: 590 AUDIT_ARG_SVIPC_PERM(&msqbuf->msg_perm); 591 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 592 goto done2; 593 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 594 error = priv_check(td, PRIV_IPC_MSGSIZE); 595 if (error) 596 goto done2; 597 } 598 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 599 DPRINTF(("can't increase msg_qbytes beyond %d" 600 "(truncating)\n", msginfo.msgmnb)); 601 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 602 } 603 if (msqbuf->msg_qbytes == 0) { 604 DPRINTF(("can't reduce msg_qbytes to 0\n")); 605 error = EINVAL; /* non-standard errno! */ 606 goto done2; 607 } 608 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 609 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 610 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 611 (msqbuf->msg_perm.mode & 0777); 612 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 613 msqkptr->u.msg_ctime = time_second; 614 break; 615 616 case IPC_STAT: 617 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 618 DPRINTF(("requester doesn't have read access\n")); 619 goto done2; 620 } 621 *msqbuf = msqkptr->u; 622 if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison) 623 msqbuf->msg_perm.key = IPC_PRIVATE; 624 break; 625 626 default: 627 DPRINTF(("invalid command %d\n", cmd)); 628 error = EINVAL; 629 goto done2; 630 } 631 632 if (error == 0) 633 td->td_retval[0] = rval; 634 done2: 635 mtx_unlock(&msq_mtx); 636 return (error); 637 } 638 639 #ifndef _SYS_SYSPROTO_H_ 640 struct msgget_args { 641 key_t key; 642 int msgflg; 643 }; 644 #endif 645 646 int 647 sys_msgget(td, uap) 648 struct thread *td; 649 register struct msgget_args *uap; 650 { 651 int msqid, error = 0; 652 int key = uap->key; 653 int msgflg = uap->msgflg; 654 struct ucred *cred = td->td_ucred; 655 register struct msqid_kernel *msqkptr = NULL; 656 657 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 658 659 if (msg_find_prison(cred) == NULL) 660 return (ENOSYS); 661 662 mtx_lock(&msq_mtx); 663 if (key != IPC_PRIVATE) { 664 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 665 msqkptr = &msqids[msqid]; 666 if (msqkptr->u.msg_qbytes != 0 && 667 msqkptr->cred != NULL && 668 msqkptr->cred->cr_prison == cred->cr_prison && 669 msqkptr->u.msg_perm.key == key) 670 break; 671 } 672 if (msqid < msginfo.msgmni) { 673 DPRINTF(("found public key\n")); 674 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 675 DPRINTF(("not exclusive\n")); 676 error = EEXIST; 677 goto done2; 678 } 679 AUDIT_ARG_SVIPC_ID(IXSEQ_TO_IPCID(msqid, 680 msqkptr->u.msg_perm)); 681 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 682 msgflg & 0700))) { 683 DPRINTF(("requester doesn't have 0%o access\n", 684 msgflg & 0700)); 685 goto done2; 686 } 687 #ifdef MAC 688 error = mac_sysvmsq_check_msqget(cred, msqkptr); 689 if (error != 0) 690 goto done2; 691 #endif 692 goto found; 693 } 694 } 695 696 DPRINTF(("need to allocate the msqid_ds\n")); 697 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 698 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 699 /* 700 * Look for an unallocated and unlocked msqid_ds. 701 * msqid_ds's can be locked by msgsnd or msgrcv while 702 * they are copying the message in/out. We can't 703 * re-use the entry until they release it. 704 */ 705 msqkptr = &msqids[msqid]; 706 if (msqkptr->u.msg_qbytes == 0 && 707 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 708 break; 709 } 710 if (msqid == msginfo.msgmni) { 711 DPRINTF(("no more msqid_ds's available\n")); 712 error = ENOSPC; 713 goto done2; 714 } 715 #ifdef RACCT 716 if (racct_enable) { 717 PROC_LOCK(td->td_proc); 718 error = racct_add(td->td_proc, RACCT_NMSGQ, 1); 719 PROC_UNLOCK(td->td_proc); 720 if (error != 0) { 721 error = ENOSPC; 722 goto done2; 723 } 724 } 725 #endif 726 DPRINTF(("msqid %d is available\n", msqid)); 727 msqkptr->u.msg_perm.key = key; 728 msqkptr->u.msg_perm.cuid = cred->cr_uid; 729 msqkptr->u.msg_perm.uid = cred->cr_uid; 730 msqkptr->u.msg_perm.cgid = cred->cr_gid; 731 msqkptr->u.msg_perm.gid = cred->cr_gid; 732 msqkptr->u.msg_perm.mode = (msgflg & 0777); 733 msqkptr->cred = crhold(cred); 734 /* Make sure that the returned msqid is unique */ 735 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 736 msqkptr->u.msg_first = NULL; 737 msqkptr->u.msg_last = NULL; 738 msqkptr->u.msg_cbytes = 0; 739 msqkptr->u.msg_qnum = 0; 740 msqkptr->u.msg_qbytes = msginfo.msgmnb; 741 msqkptr->u.msg_lspid = 0; 742 msqkptr->u.msg_lrpid = 0; 743 msqkptr->u.msg_stime = 0; 744 msqkptr->u.msg_rtime = 0; 745 msqkptr->u.msg_ctime = time_second; 746 #ifdef MAC 747 mac_sysvmsq_create(cred, msqkptr); 748 #endif 749 AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm); 750 } else { 751 DPRINTF(("didn't find it and wasn't asked to create it\n")); 752 error = ENOENT; 753 goto done2; 754 } 755 756 found: 757 /* Construct the unique msqid */ 758 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 759 done2: 760 mtx_unlock(&msq_mtx); 761 return (error); 762 } 763 764 #ifndef _SYS_SYSPROTO_H_ 765 struct msgsnd_args { 766 int msqid; 767 const void *msgp; 768 size_t msgsz; 769 int msgflg; 770 }; 771 #endif 772 int 773 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 774 struct thread *td; 775 int msqid; 776 const void *msgp; /* XXX msgp is actually mtext. */ 777 size_t msgsz; 778 int msgflg; 779 long mtype; 780 { 781 int msqix, segs_needed, error = 0; 782 register struct msqid_kernel *msqkptr; 783 register struct msg *msghdr; 784 struct prison *rpr; 785 short next; 786 #ifdef RACCT 787 size_t saved_msgsz; 788 #endif 789 790 rpr = msg_find_prison(td->td_ucred); 791 if (rpr == NULL) 792 return (ENOSYS); 793 794 mtx_lock(&msq_mtx); 795 AUDIT_ARG_SVIPC_ID(msqid); 796 msqix = IPCID_TO_IX(msqid); 797 798 if (msqix < 0 || msqix >= msginfo.msgmni) { 799 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 800 msginfo.msgmni)); 801 error = EINVAL; 802 goto done2; 803 } 804 805 msqkptr = &msqids[msqix]; 806 AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm); 807 if (msqkptr->u.msg_qbytes == 0) { 808 DPRINTF(("no such message queue id\n")); 809 error = EINVAL; 810 goto done2; 811 } 812 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 813 DPRINTF(("wrong sequence number\n")); 814 error = EINVAL; 815 goto done2; 816 } 817 818 if ((error = msq_prison_cansee(rpr, msqkptr))) { 819 DPRINTF(("requester can't see prison\n")); 820 goto done2; 821 } 822 823 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 824 DPRINTF(("requester doesn't have write access\n")); 825 goto done2; 826 } 827 828 #ifdef MAC 829 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 830 if (error != 0) 831 goto done2; 832 #endif 833 834 #ifdef RACCT 835 if (racct_enable) { 836 PROC_LOCK(td->td_proc); 837 if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) { 838 PROC_UNLOCK(td->td_proc); 839 error = EAGAIN; 840 goto done2; 841 } 842 saved_msgsz = msgsz; 843 if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) { 844 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 845 PROC_UNLOCK(td->td_proc); 846 error = EAGAIN; 847 goto done2; 848 } 849 PROC_UNLOCK(td->td_proc); 850 } 851 #endif 852 853 segs_needed = howmany(msgsz, msginfo.msgssz); 854 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 855 msginfo.msgssz, segs_needed)); 856 for (;;) { 857 int need_more_resources = 0; 858 859 /* 860 * check msgsz 861 * (inside this loop in case msg_qbytes changes while we sleep) 862 */ 863 864 if (msgsz > msqkptr->u.msg_qbytes) { 865 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 866 error = EINVAL; 867 goto done3; 868 } 869 870 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 871 DPRINTF(("msqid is locked\n")); 872 need_more_resources = 1; 873 } 874 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 875 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 876 need_more_resources = 1; 877 } 878 if (segs_needed > nfree_msgmaps) { 879 DPRINTF(("segs_needed > nfree_msgmaps\n")); 880 need_more_resources = 1; 881 } 882 if (free_msghdrs == NULL) { 883 DPRINTF(("no more msghdrs\n")); 884 need_more_resources = 1; 885 } 886 887 if (need_more_resources) { 888 int we_own_it; 889 890 if ((msgflg & IPC_NOWAIT) != 0) { 891 DPRINTF(("need more resources but caller " 892 "doesn't want to wait\n")); 893 error = EAGAIN; 894 goto done3; 895 } 896 897 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 898 DPRINTF(("we don't own the msqid_ds\n")); 899 we_own_it = 0; 900 } else { 901 /* Force later arrivals to wait for our 902 request */ 903 DPRINTF(("we own the msqid_ds\n")); 904 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 905 we_own_it = 1; 906 } 907 DPRINTF(("msgsnd: goodnight\n")); 908 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 909 "msgsnd", hz); 910 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 911 if (we_own_it) 912 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 913 if (error == EWOULDBLOCK) { 914 DPRINTF(("msgsnd: timed out\n")); 915 continue; 916 } 917 if (error != 0) { 918 DPRINTF(("msgsnd: interrupted system call\n")); 919 error = EINTR; 920 goto done3; 921 } 922 923 /* 924 * Make sure that the msq queue still exists 925 */ 926 927 if (msqkptr->u.msg_qbytes == 0) { 928 DPRINTF(("msqid deleted\n")); 929 error = EIDRM; 930 goto done3; 931 } 932 933 } else { 934 DPRINTF(("got all the resources that we need\n")); 935 break; 936 } 937 } 938 939 /* 940 * We have the resources that we need. 941 * Make sure! 942 */ 943 944 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 945 panic("msg_perm.mode & MSG_LOCKED"); 946 if (segs_needed > nfree_msgmaps) 947 panic("segs_needed > nfree_msgmaps"); 948 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 949 panic("msgsz + msg_cbytes > msg_qbytes"); 950 if (free_msghdrs == NULL) 951 panic("no more msghdrs"); 952 953 /* 954 * Re-lock the msqid_ds in case we page-fault when copying in the 955 * message 956 */ 957 958 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 959 panic("msqid_ds is already locked"); 960 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 961 962 /* 963 * Allocate a message header 964 */ 965 966 msghdr = free_msghdrs; 967 free_msghdrs = msghdr->msg_next; 968 msghdr->msg_spot = -1; 969 msghdr->msg_ts = msgsz; 970 msghdr->msg_type = mtype; 971 #ifdef MAC 972 /* 973 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 974 * immediately? Or, should it be checked just before the msg is 975 * enqueued in the msgq (as it is done now)? 976 */ 977 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 978 #endif 979 980 /* 981 * Allocate space for the message 982 */ 983 984 while (segs_needed > 0) { 985 if (nfree_msgmaps <= 0) 986 panic("not enough msgmaps"); 987 if (free_msgmaps == -1) 988 panic("nil free_msgmaps"); 989 next = free_msgmaps; 990 if (next <= -1) 991 panic("next too low #1"); 992 if (next >= msginfo.msgseg) 993 panic("next out of range #1"); 994 DPRINTF(("allocating segment %d to message\n", next)); 995 free_msgmaps = msgmaps[next].next; 996 nfree_msgmaps--; 997 msgmaps[next].next = msghdr->msg_spot; 998 msghdr->msg_spot = next; 999 segs_needed--; 1000 } 1001 1002 /* 1003 * Validate the message type 1004 */ 1005 1006 if (msghdr->msg_type < 1) { 1007 msg_freehdr(msghdr); 1008 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1009 wakeup(msqkptr); 1010 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 1011 error = EINVAL; 1012 goto done3; 1013 } 1014 1015 /* 1016 * Copy in the message body 1017 */ 1018 1019 next = msghdr->msg_spot; 1020 while (msgsz > 0) { 1021 size_t tlen; 1022 if (msgsz > msginfo.msgssz) 1023 tlen = msginfo.msgssz; 1024 else 1025 tlen = msgsz; 1026 if (next <= -1) 1027 panic("next too low #2"); 1028 if (next >= msginfo.msgseg) 1029 panic("next out of range #2"); 1030 mtx_unlock(&msq_mtx); 1031 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 1032 tlen)) != 0) { 1033 mtx_lock(&msq_mtx); 1034 DPRINTF(("error %d copying in message segment\n", 1035 error)); 1036 msg_freehdr(msghdr); 1037 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1038 wakeup(msqkptr); 1039 goto done3; 1040 } 1041 mtx_lock(&msq_mtx); 1042 msgsz -= tlen; 1043 msgp = (const char *)msgp + tlen; 1044 next = msgmaps[next].next; 1045 } 1046 if (next != -1) 1047 panic("didn't use all the msg segments"); 1048 1049 /* 1050 * We've got the message. Unlock the msqid_ds. 1051 */ 1052 1053 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1054 1055 /* 1056 * Make sure that the msqid_ds is still allocated. 1057 */ 1058 1059 if (msqkptr->u.msg_qbytes == 0) { 1060 msg_freehdr(msghdr); 1061 wakeup(msqkptr); 1062 error = EIDRM; 1063 goto done3; 1064 } 1065 1066 #ifdef MAC 1067 /* 1068 * Note: Since the task/thread allocates the msghdr and usually 1069 * primes it with its own MAC label, for a majority of policies, it 1070 * won't be necessary to check whether the msghdr has access 1071 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 1072 * suffice in that case. However, this hook may be required where 1073 * individual policies derive a non-identical label for the msghdr 1074 * from the current thread label and may want to check the msghdr 1075 * enqueue permissions, along with read/write permissions to the 1076 * msgq. 1077 */ 1078 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 1079 if (error != 0) { 1080 msg_freehdr(msghdr); 1081 wakeup(msqkptr); 1082 goto done3; 1083 } 1084 #endif 1085 1086 /* 1087 * Put the message into the queue 1088 */ 1089 if (msqkptr->u.msg_first == NULL) { 1090 msqkptr->u.msg_first = msghdr; 1091 msqkptr->u.msg_last = msghdr; 1092 } else { 1093 msqkptr->u.msg_last->msg_next = msghdr; 1094 msqkptr->u.msg_last = msghdr; 1095 } 1096 msqkptr->u.msg_last->msg_next = NULL; 1097 1098 msqkptr->u.msg_cbytes += msghdr->msg_ts; 1099 msqkptr->u.msg_qnum++; 1100 msqkptr->u.msg_lspid = td->td_proc->p_pid; 1101 msqkptr->u.msg_stime = time_second; 1102 1103 wakeup(msqkptr); 1104 td->td_retval[0] = 0; 1105 done3: 1106 #ifdef RACCT 1107 if (racct_enable && error != 0) { 1108 PROC_LOCK(td->td_proc); 1109 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 1110 racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz); 1111 PROC_UNLOCK(td->td_proc); 1112 } 1113 #endif 1114 done2: 1115 mtx_unlock(&msq_mtx); 1116 return (error); 1117 } 1118 1119 int 1120 sys_msgsnd(td, uap) 1121 struct thread *td; 1122 register struct msgsnd_args *uap; 1123 { 1124 int error; 1125 long mtype; 1126 1127 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 1128 uap->msgsz, uap->msgflg)); 1129 1130 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 1131 DPRINTF(("error %d copying the message type\n", error)); 1132 return (error); 1133 } 1134 return (kern_msgsnd(td, uap->msqid, 1135 (const char *)uap->msgp + sizeof(mtype), 1136 uap->msgsz, uap->msgflg, mtype)); 1137 } 1138 1139 #ifndef _SYS_SYSPROTO_H_ 1140 struct msgrcv_args { 1141 int msqid; 1142 void *msgp; 1143 size_t msgsz; 1144 long msgtyp; 1145 int msgflg; 1146 }; 1147 #endif 1148 int 1149 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1150 struct thread *td; 1151 int msqid; 1152 void *msgp; /* XXX msgp is actually mtext. */ 1153 size_t msgsz; 1154 long msgtyp; 1155 int msgflg; 1156 long *mtype; 1157 { 1158 size_t len; 1159 register struct msqid_kernel *msqkptr; 1160 register struct msg *msghdr; 1161 struct prison *rpr; 1162 int msqix, error = 0; 1163 short next; 1164 1165 rpr = msg_find_prison(td->td_ucred); 1166 if (rpr == NULL) 1167 return (ENOSYS); 1168 1169 AUDIT_ARG_SVIPC_ID(msqid); 1170 msqix = IPCID_TO_IX(msqid); 1171 1172 if (msqix < 0 || msqix >= msginfo.msgmni) { 1173 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1174 msginfo.msgmni)); 1175 return (EINVAL); 1176 } 1177 1178 msqkptr = &msqids[msqix]; 1179 mtx_lock(&msq_mtx); 1180 AUDIT_ARG_SVIPC_PERM(&msqkptr->u.msg_perm); 1181 if (msqkptr->u.msg_qbytes == 0) { 1182 DPRINTF(("no such message queue id\n")); 1183 error = EINVAL; 1184 goto done2; 1185 } 1186 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1187 DPRINTF(("wrong sequence number\n")); 1188 error = EINVAL; 1189 goto done2; 1190 } 1191 1192 if ((error = msq_prison_cansee(rpr, msqkptr))) { 1193 DPRINTF(("requester can't see prison\n")); 1194 goto done2; 1195 } 1196 1197 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1198 DPRINTF(("requester doesn't have read access\n")); 1199 goto done2; 1200 } 1201 1202 #ifdef MAC 1203 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1204 if (error != 0) 1205 goto done2; 1206 #endif 1207 1208 msghdr = NULL; 1209 while (msghdr == NULL) { 1210 if (msgtyp == 0) { 1211 msghdr = msqkptr->u.msg_first; 1212 if (msghdr != NULL) { 1213 if (msgsz < msghdr->msg_ts && 1214 (msgflg & MSG_NOERROR) == 0) { 1215 DPRINTF(("first message on the queue " 1216 "is too big (want %zu, got %d)\n", 1217 msgsz, msghdr->msg_ts)); 1218 error = E2BIG; 1219 goto done2; 1220 } 1221 #ifdef MAC 1222 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1223 msghdr); 1224 if (error != 0) 1225 goto done2; 1226 #endif 1227 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1228 msqkptr->u.msg_first = NULL; 1229 msqkptr->u.msg_last = NULL; 1230 } else { 1231 msqkptr->u.msg_first = msghdr->msg_next; 1232 if (msqkptr->u.msg_first == NULL) 1233 panic("msg_first/last screwed up #1"); 1234 } 1235 } 1236 } else { 1237 struct msg *previous; 1238 struct msg **prev; 1239 1240 previous = NULL; 1241 prev = &(msqkptr->u.msg_first); 1242 while ((msghdr = *prev) != NULL) { 1243 /* 1244 * Is this message's type an exact match or is 1245 * this message's type less than or equal to 1246 * the absolute value of a negative msgtyp? 1247 * Note that the second half of this test can 1248 * NEVER be true if msgtyp is positive since 1249 * msg_type is always positive! 1250 */ 1251 1252 if (msgtyp == msghdr->msg_type || 1253 msghdr->msg_type <= -msgtyp) { 1254 DPRINTF(("found message type %ld, " 1255 "requested %ld\n", 1256 msghdr->msg_type, msgtyp)); 1257 if (msgsz < msghdr->msg_ts && 1258 (msgflg & MSG_NOERROR) == 0) { 1259 DPRINTF(("requested message " 1260 "on the queue is too big " 1261 "(want %zu, got %hu)\n", 1262 msgsz, msghdr->msg_ts)); 1263 error = E2BIG; 1264 goto done2; 1265 } 1266 #ifdef MAC 1267 error = mac_sysvmsq_check_msgrcv( 1268 td->td_ucred, msghdr); 1269 if (error != 0) 1270 goto done2; 1271 #endif 1272 *prev = msghdr->msg_next; 1273 if (msghdr == msqkptr->u.msg_last) { 1274 if (previous == NULL) { 1275 if (prev != 1276 &msqkptr->u.msg_first) 1277 panic("msg_first/last screwed up #2"); 1278 msqkptr->u.msg_first = 1279 NULL; 1280 msqkptr->u.msg_last = 1281 NULL; 1282 } else { 1283 if (prev == 1284 &msqkptr->u.msg_first) 1285 panic("msg_first/last screwed up #3"); 1286 msqkptr->u.msg_last = 1287 previous; 1288 } 1289 } 1290 break; 1291 } 1292 previous = msghdr; 1293 prev = &(msghdr->msg_next); 1294 } 1295 } 1296 1297 /* 1298 * We've either extracted the msghdr for the appropriate 1299 * message or there isn't one. 1300 * If there is one then bail out of this loop. 1301 */ 1302 1303 if (msghdr != NULL) 1304 break; 1305 1306 /* 1307 * Hmph! No message found. Does the user want to wait? 1308 */ 1309 1310 if ((msgflg & IPC_NOWAIT) != 0) { 1311 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1312 msgtyp)); 1313 /* The SVID says to return ENOMSG. */ 1314 error = ENOMSG; 1315 goto done2; 1316 } 1317 1318 /* 1319 * Wait for something to happen 1320 */ 1321 1322 DPRINTF(("msgrcv: goodnight\n")); 1323 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1324 "msgrcv", 0); 1325 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1326 1327 if (error != 0) { 1328 DPRINTF(("msgrcv: interrupted system call\n")); 1329 error = EINTR; 1330 goto done2; 1331 } 1332 1333 /* 1334 * Make sure that the msq queue still exists 1335 */ 1336 1337 if (msqkptr->u.msg_qbytes == 0 || 1338 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1339 DPRINTF(("msqid deleted\n")); 1340 error = EIDRM; 1341 goto done2; 1342 } 1343 } 1344 1345 /* 1346 * Return the message to the user. 1347 * 1348 * First, do the bookkeeping (before we risk being interrupted). 1349 */ 1350 1351 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1352 msqkptr->u.msg_qnum--; 1353 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1354 msqkptr->u.msg_rtime = time_second; 1355 1356 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1); 1357 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts); 1358 1359 /* 1360 * Make msgsz the actual amount that we'll be returning. 1361 * Note that this effectively truncates the message if it is too long 1362 * (since msgsz is never increased). 1363 */ 1364 1365 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1366 msghdr->msg_ts)); 1367 if (msgsz > msghdr->msg_ts) 1368 msgsz = msghdr->msg_ts; 1369 *mtype = msghdr->msg_type; 1370 1371 /* 1372 * Return the segments to the user 1373 */ 1374 1375 next = msghdr->msg_spot; 1376 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1377 size_t tlen; 1378 1379 if (msgsz - len > msginfo.msgssz) 1380 tlen = msginfo.msgssz; 1381 else 1382 tlen = msgsz - len; 1383 if (next <= -1) 1384 panic("next too low #3"); 1385 if (next >= msginfo.msgseg) 1386 panic("next out of range #3"); 1387 mtx_unlock(&msq_mtx); 1388 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1389 mtx_lock(&msq_mtx); 1390 if (error != 0) { 1391 DPRINTF(("error (%d) copying out message segment\n", 1392 error)); 1393 msg_freehdr(msghdr); 1394 wakeup(msqkptr); 1395 goto done2; 1396 } 1397 msgp = (char *)msgp + tlen; 1398 next = msgmaps[next].next; 1399 } 1400 1401 /* 1402 * Done, return the actual number of bytes copied out. 1403 */ 1404 1405 msg_freehdr(msghdr); 1406 wakeup(msqkptr); 1407 td->td_retval[0] = msgsz; 1408 done2: 1409 mtx_unlock(&msq_mtx); 1410 return (error); 1411 } 1412 1413 int 1414 sys_msgrcv(td, uap) 1415 struct thread *td; 1416 register struct msgrcv_args *uap; 1417 { 1418 int error; 1419 long mtype; 1420 1421 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1422 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1423 1424 if ((error = kern_msgrcv(td, uap->msqid, 1425 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1426 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1427 return (error); 1428 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1429 DPRINTF(("error %d copying the message type\n", error)); 1430 return (error); 1431 } 1432 1433 static int 1434 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1435 { 1436 struct msqid_kernel tmsqk; 1437 struct prison *pr, *rpr; 1438 int error, i; 1439 1440 pr = req->td->td_ucred->cr_prison; 1441 rpr = msg_find_prison(req->td->td_ucred); 1442 error = 0; 1443 for (i = 0; i < msginfo.msgmni; i++) { 1444 mtx_lock(&msq_mtx); 1445 if (msqids[i].u.msg_qbytes == 0 || rpr == NULL || 1446 msq_prison_cansee(rpr, &msqids[i]) != 0) 1447 bzero(&tmsqk, sizeof(tmsqk)); 1448 else { 1449 tmsqk = msqids[i]; 1450 if (tmsqk.cred->cr_prison != pr) 1451 tmsqk.u.msg_perm.key = IPC_PRIVATE; 1452 } 1453 mtx_unlock(&msq_mtx); 1454 error = SYSCTL_OUT(req, &tmsqk, sizeof(tmsqk)); 1455 if (error != 0) 1456 break; 1457 } 1458 return (error); 1459 } 1460 1461 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1462 "Maximum message size"); 1463 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1464 "Number of message queue identifiers"); 1465 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1466 "Maximum number of bytes in a queue"); 1467 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1468 "Maximum number of messages in the system"); 1469 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1470 "Size of a message segment"); 1471 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1472 "Number of message segments"); 1473 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, 1474 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1475 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1476 1477 static int 1478 msg_prison_check(void *obj, void *data) 1479 { 1480 struct prison *pr = obj; 1481 struct prison *prpr; 1482 struct vfsoptlist *opts = data; 1483 int error, jsys; 1484 1485 /* 1486 * sysvmsg is a jailsys integer. 1487 * It must be "disable" if the parent jail is disabled. 1488 */ 1489 error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1490 if (error != ENOENT) { 1491 if (error != 0) 1492 return (error); 1493 switch (jsys) { 1494 case JAIL_SYS_DISABLE: 1495 break; 1496 case JAIL_SYS_NEW: 1497 case JAIL_SYS_INHERIT: 1498 prison_lock(pr->pr_parent); 1499 prpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1500 prison_unlock(pr->pr_parent); 1501 if (prpr == NULL) 1502 return (EPERM); 1503 break; 1504 default: 1505 return (EINVAL); 1506 } 1507 } 1508 1509 return (0); 1510 } 1511 1512 static int 1513 msg_prison_set(void *obj, void *data) 1514 { 1515 struct prison *pr = obj; 1516 struct prison *tpr, *orpr, *nrpr, *trpr; 1517 struct vfsoptlist *opts = data; 1518 void *rsv; 1519 int jsys, descend; 1520 1521 /* 1522 * sysvmsg controls which jail is the root of the associated msgs (this 1523 * jail or same as the parent), or if the feature is available at all. 1524 */ 1525 if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT) 1526 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0) 1527 ? JAIL_SYS_INHERIT 1528 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0) 1529 ? JAIL_SYS_DISABLE 1530 : -1; 1531 if (jsys == JAIL_SYS_DISABLE) { 1532 prison_lock(pr); 1533 orpr = osd_jail_get(pr, msg_prison_slot); 1534 if (orpr != NULL) 1535 osd_jail_del(pr, msg_prison_slot); 1536 prison_unlock(pr); 1537 if (orpr != NULL) { 1538 if (orpr == pr) 1539 msg_prison_cleanup(pr); 1540 /* Disable all child jails as well. */ 1541 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1542 prison_lock(tpr); 1543 trpr = osd_jail_get(tpr, msg_prison_slot); 1544 if (trpr != NULL) { 1545 osd_jail_del(tpr, msg_prison_slot); 1546 prison_unlock(tpr); 1547 if (trpr == tpr) 1548 msg_prison_cleanup(tpr); 1549 } else { 1550 prison_unlock(tpr); 1551 descend = 0; 1552 } 1553 } 1554 } 1555 } else if (jsys != -1) { 1556 if (jsys == JAIL_SYS_NEW) 1557 nrpr = pr; 1558 else { 1559 prison_lock(pr->pr_parent); 1560 nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1561 prison_unlock(pr->pr_parent); 1562 } 1563 rsv = osd_reserve(msg_prison_slot); 1564 prison_lock(pr); 1565 orpr = osd_jail_get(pr, msg_prison_slot); 1566 if (orpr != nrpr) 1567 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 1568 nrpr); 1569 else 1570 osd_free_reserved(rsv); 1571 prison_unlock(pr); 1572 if (orpr != nrpr) { 1573 if (orpr == pr) 1574 msg_prison_cleanup(pr); 1575 if (orpr != NULL) { 1576 /* Change child jails matching the old root, */ 1577 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1578 prison_lock(tpr); 1579 trpr = osd_jail_get(tpr, 1580 msg_prison_slot); 1581 if (trpr == orpr) { 1582 (void)osd_jail_set(tpr, 1583 msg_prison_slot, nrpr); 1584 prison_unlock(tpr); 1585 if (trpr == tpr) 1586 msg_prison_cleanup(tpr); 1587 } else { 1588 prison_unlock(tpr); 1589 descend = 0; 1590 } 1591 } 1592 } 1593 } 1594 } 1595 1596 return (0); 1597 } 1598 1599 static int 1600 msg_prison_get(void *obj, void *data) 1601 { 1602 struct prison *pr = obj; 1603 struct prison *rpr; 1604 struct vfsoptlist *opts = data; 1605 int error, jsys; 1606 1607 /* Set sysvmsg based on the jail's root prison. */ 1608 prison_lock(pr); 1609 rpr = osd_jail_get(pr, msg_prison_slot); 1610 prison_unlock(pr); 1611 jsys = rpr == NULL ? JAIL_SYS_DISABLE 1612 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; 1613 error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1614 if (error == ENOENT) 1615 error = 0; 1616 return (error); 1617 } 1618 1619 static int 1620 msg_prison_remove(void *obj, void *data __unused) 1621 { 1622 struct prison *pr = obj; 1623 struct prison *rpr; 1624 1625 prison_lock(pr); 1626 rpr = osd_jail_get(pr, msg_prison_slot); 1627 prison_unlock(pr); 1628 if (rpr == pr) 1629 msg_prison_cleanup(pr); 1630 return (0); 1631 } 1632 1633 static void 1634 msg_prison_cleanup(struct prison *pr) 1635 { 1636 struct msqid_kernel *msqkptr; 1637 int i; 1638 1639 /* Remove any msqs that belong to this jail. */ 1640 mtx_lock(&msq_mtx); 1641 for (i = 0; i < msginfo.msgmni; i++) { 1642 msqkptr = &msqids[i]; 1643 if (msqkptr->u.msg_qbytes != 0 && 1644 msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr) 1645 msq_remove(msqkptr); 1646 } 1647 mtx_unlock(&msq_mtx); 1648 } 1649 1650 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues"); 1651 1652 #ifdef COMPAT_FREEBSD32 1653 int 1654 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap) 1655 { 1656 1657 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1658 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1659 AUDIT_ARG_SVIPC_WHICH(uap->which); 1660 switch (uap->which) { 1661 case 0: 1662 return (freebsd7_freebsd32_msgctl(td, 1663 (struct freebsd7_freebsd32_msgctl_args *)&uap->a2)); 1664 case 2: 1665 return (freebsd32_msgsnd(td, 1666 (struct freebsd32_msgsnd_args *)&uap->a2)); 1667 case 3: 1668 return (freebsd32_msgrcv(td, 1669 (struct freebsd32_msgrcv_args *)&uap->a2)); 1670 default: 1671 return (sys_msgsys(td, (struct msgsys_args *)uap)); 1672 } 1673 #else 1674 return (nosys(td, NULL)); 1675 #endif 1676 } 1677 1678 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1679 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1680 int 1681 freebsd7_freebsd32_msgctl(struct thread *td, 1682 struct freebsd7_freebsd32_msgctl_args *uap) 1683 { 1684 struct msqid_ds msqbuf; 1685 struct msqid_ds32_old msqbuf32; 1686 int error; 1687 1688 if (uap->cmd == IPC_SET) { 1689 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1690 if (error) 1691 return (error); 1692 freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1693 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1694 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1695 CP(msqbuf32, msqbuf, msg_cbytes); 1696 CP(msqbuf32, msqbuf, msg_qnum); 1697 CP(msqbuf32, msqbuf, msg_qbytes); 1698 CP(msqbuf32, msqbuf, msg_lspid); 1699 CP(msqbuf32, msqbuf, msg_lrpid); 1700 CP(msqbuf32, msqbuf, msg_stime); 1701 CP(msqbuf32, msqbuf, msg_rtime); 1702 CP(msqbuf32, msqbuf, msg_ctime); 1703 } 1704 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1705 if (error) 1706 return (error); 1707 if (uap->cmd == IPC_STAT) { 1708 bzero(&msqbuf32, sizeof(msqbuf32)); 1709 freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1710 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1711 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1712 CP(msqbuf, msqbuf32, msg_cbytes); 1713 CP(msqbuf, msqbuf32, msg_qnum); 1714 CP(msqbuf, msqbuf32, msg_qbytes); 1715 CP(msqbuf, msqbuf32, msg_lspid); 1716 CP(msqbuf, msqbuf32, msg_lrpid); 1717 CP(msqbuf, msqbuf32, msg_stime); 1718 CP(msqbuf, msqbuf32, msg_rtime); 1719 CP(msqbuf, msqbuf32, msg_ctime); 1720 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1721 } 1722 return (error); 1723 } 1724 #endif 1725 1726 int 1727 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap) 1728 { 1729 struct msqid_ds msqbuf; 1730 struct msqid_ds32 msqbuf32; 1731 int error; 1732 1733 if (uap->cmd == IPC_SET) { 1734 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1735 if (error) 1736 return (error); 1737 freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1738 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1739 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1740 CP(msqbuf32, msqbuf, msg_cbytes); 1741 CP(msqbuf32, msqbuf, msg_qnum); 1742 CP(msqbuf32, msqbuf, msg_qbytes); 1743 CP(msqbuf32, msqbuf, msg_lspid); 1744 CP(msqbuf32, msqbuf, msg_lrpid); 1745 CP(msqbuf32, msqbuf, msg_stime); 1746 CP(msqbuf32, msqbuf, msg_rtime); 1747 CP(msqbuf32, msqbuf, msg_ctime); 1748 } 1749 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1750 if (error) 1751 return (error); 1752 if (uap->cmd == IPC_STAT) { 1753 freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1754 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1755 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1756 CP(msqbuf, msqbuf32, msg_cbytes); 1757 CP(msqbuf, msqbuf32, msg_qnum); 1758 CP(msqbuf, msqbuf32, msg_qbytes); 1759 CP(msqbuf, msqbuf32, msg_lspid); 1760 CP(msqbuf, msqbuf32, msg_lrpid); 1761 CP(msqbuf, msqbuf32, msg_stime); 1762 CP(msqbuf, msqbuf32, msg_rtime); 1763 CP(msqbuf, msqbuf32, msg_ctime); 1764 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1765 } 1766 return (error); 1767 } 1768 1769 int 1770 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap) 1771 { 1772 const void *msgp; 1773 long mtype; 1774 int32_t mtype32; 1775 int error; 1776 1777 msgp = PTRIN(uap->msgp); 1778 if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0) 1779 return (error); 1780 mtype = mtype32; 1781 return (kern_msgsnd(td, uap->msqid, 1782 (const char *)msgp + sizeof(mtype32), 1783 uap->msgsz, uap->msgflg, mtype)); 1784 } 1785 1786 int 1787 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap) 1788 { 1789 void *msgp; 1790 long mtype; 1791 int32_t mtype32; 1792 int error; 1793 1794 msgp = PTRIN(uap->msgp); 1795 if ((error = kern_msgrcv(td, uap->msqid, 1796 (char *)msgp + sizeof(mtype32), uap->msgsz, 1797 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1798 return (error); 1799 mtype32 = (int32_t)mtype; 1800 return (copyout(&mtype32, msgp, sizeof(mtype32))); 1801 } 1802 #endif 1803 1804 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1805 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1806 1807 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 1808 static sy_call_t *msgcalls[] = { 1809 (sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget, 1810 (sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv 1811 }; 1812 1813 /* 1814 * Entry point for all MSG calls. 1815 */ 1816 int 1817 sys_msgsys(td, uap) 1818 struct thread *td; 1819 /* XXX actually varargs. */ 1820 struct msgsys_args /* { 1821 int which; 1822 int a2; 1823 int a3; 1824 int a4; 1825 int a5; 1826 int a6; 1827 } */ *uap; 1828 { 1829 int error; 1830 1831 AUDIT_ARG_SVIPC_WHICH(uap->which); 1832 if (uap->which < 0 || uap->which >= nitems(msgcalls)) 1833 return (EINVAL); 1834 error = (*msgcalls[uap->which])(td, &uap->a2); 1835 return (error); 1836 } 1837 1838 #ifndef CP 1839 #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0) 1840 #endif 1841 1842 #ifndef _SYS_SYSPROTO_H_ 1843 struct freebsd7_msgctl_args { 1844 int msqid; 1845 int cmd; 1846 struct msqid_ds_old *buf; 1847 }; 1848 #endif 1849 int 1850 freebsd7_msgctl(td, uap) 1851 struct thread *td; 1852 struct freebsd7_msgctl_args *uap; 1853 { 1854 struct msqid_ds_old msqold; 1855 struct msqid_ds msqbuf; 1856 int error; 1857 1858 DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd, 1859 uap->buf)); 1860 if (uap->cmd == IPC_SET) { 1861 error = copyin(uap->buf, &msqold, sizeof(msqold)); 1862 if (error) 1863 return (error); 1864 ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm); 1865 CP(msqold, msqbuf, msg_first); 1866 CP(msqold, msqbuf, msg_last); 1867 CP(msqold, msqbuf, msg_cbytes); 1868 CP(msqold, msqbuf, msg_qnum); 1869 CP(msqold, msqbuf, msg_qbytes); 1870 CP(msqold, msqbuf, msg_lspid); 1871 CP(msqold, msqbuf, msg_lrpid); 1872 CP(msqold, msqbuf, msg_stime); 1873 CP(msqold, msqbuf, msg_rtime); 1874 CP(msqold, msqbuf, msg_ctime); 1875 } 1876 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1877 if (error) 1878 return (error); 1879 if (uap->cmd == IPC_STAT) { 1880 bzero(&msqold, sizeof(msqold)); 1881 ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm); 1882 CP(msqbuf, msqold, msg_first); 1883 CP(msqbuf, msqold, msg_last); 1884 CP(msqbuf, msqold, msg_cbytes); 1885 CP(msqbuf, msqold, msg_qnum); 1886 CP(msqbuf, msqold, msg_qbytes); 1887 CP(msqbuf, msqold, msg_lspid); 1888 CP(msqbuf, msqold, msg_lrpid); 1889 CP(msqbuf, msqold, msg_stime); 1890 CP(msqbuf, msqold, msg_rtime); 1891 CP(msqbuf, msqold, msg_ctime); 1892 error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old)); 1893 } 1894 return (error); 1895 } 1896 1897 #undef CP 1898 1899 #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 || 1900 COMPAT_FREEBSD7 */ 1901