1 /* $NetBSD: sysv_msg.c,v 1.52 2007/11/04 13:09:32 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2006, 2007 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Implementation of SVID messages 42 * 43 * Author: Daniel Boulet 44 * 45 * Copyright 1993 Daniel Boulet and RTMX Inc. 46 * 47 * This system call was implemented by Daniel Boulet under contract from RTMX. 48 * 49 * Redistribution and use in source forms, with and without modification, 50 * are permitted provided that this entire comment appears intact. 51 * 52 * Redistribution in binary form may occur without any restrictions. 53 * Obviously, it would be nice if you gave credit where credit is due 54 * but requiring it would be too onerous. 55 * 56 * This software is provided ``AS IS'' without any warranties of any kind. 57 */ 58 59 #include <sys/cdefs.h> 60 __KERNEL_RCSID(0, "$NetBSD: sysv_msg.c,v 1.52 2007/11/04 13:09:32 yamt Exp $"); 61 62 #define SYSVMSG 63 64 #include <sys/param.h> 65 #include <sys/kernel.h> 66 #include <sys/msg.h> 67 #include <sys/sysctl.h> 68 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */ 69 #include <sys/syscallargs.h> 70 #include <sys/kauth.h> 71 72 #define MSG_DEBUG 73 #undef MSG_DEBUG_OK 74 75 #ifdef MSG_DEBUG_OK 76 #define MSG_PRINTF(a) printf a 77 #else 78 #define MSG_PRINTF(a) 79 #endif 80 81 static int nfree_msgmaps; /* # of free map entries */ 82 static short free_msgmaps; /* head of linked list of free map entries */ 83 static struct __msg *free_msghdrs; /* list of free msg headers */ 84 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 85 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 86 static struct __msg *msghdrs; /* MSGTQL msg headers */ 87 88 kmsq_t *msqs; /* MSGMNI msqid_ds struct's */ 89 kmutex_t msgmutex; /* subsystem lock */ 90 91 static u_int msg_waiters = 0; /* total number of msgrcv waiters */ 92 static bool msg_realloc_state; 93 static kcondvar_t msg_realloc_cv; 94 95 static void msg_freehdr(struct __msg *); 96 97 void 98 msginit(void) 99 { 100 int i, sz; 101 vaddr_t v; 102 103 /* 104 * msginfo.msgssz should be a power of two for efficiency reasons. 105 * It is also pretty silly if msginfo.msgssz is less than 8 106 * or greater than about 256 so ... 107 */ 108 109 i = 8; 110 while (i < 1024 && i != msginfo.msgssz) 111 i <<= 1; 112 if (i != msginfo.msgssz) { 113 panic("msginfo.msgssz = %d, not a small power of 2", 114 msginfo.msgssz); 115 } 116 117 if (msginfo.msgseg > 32767) { 118 panic("msginfo.msgseg = %d > 32767", msginfo.msgseg); 119 } 120 121 /* Allocate the wired memory for our structures */ 122 sz = ALIGN(msginfo.msgmax) + 123 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 124 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 125 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 126 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 127 UVM_KMF_WIRED|UVM_KMF_ZERO); 128 if (v == 0) 129 panic("sysv_msg: cannot allocate memory"); 130 msgpool = (void *)v; 131 msgmaps = (void *)(ALIGN(msgpool) + msginfo.msgmax); 132 msghdrs = (void *)(ALIGN(msgmaps) + 133 msginfo.msgseg * sizeof(struct msgmap)); 134 msqs = (void *)(ALIGN(msghdrs) + 135 msginfo.msgtql * sizeof(struct __msg)); 136 137 for (i = 0; i < (msginfo.msgseg - 1); i++) 138 msgmaps[i].next = i + 1; 139 msgmaps[msginfo.msgseg - 1].next = -1; 140 141 free_msgmaps = 0; 142 nfree_msgmaps = msginfo.msgseg; 143 144 for (i = 0; i < (msginfo.msgtql - 1); i++) { 145 msghdrs[i].msg_type = 0; 146 msghdrs[i].msg_next = &msghdrs[i + 1]; 147 } 148 i = msginfo.msgtql - 1; 149 msghdrs[i].msg_type = 0; 150 msghdrs[i].msg_next = NULL; 151 free_msghdrs = &msghdrs[0]; 152 153 for (i = 0; i < msginfo.msgmni; i++) { 154 cv_init(&msqs[i].msq_cv, "msgwait"); 155 /* Implies entry is available */ 156 msqs[i].msq_u.msg_qbytes = 0; 157 /* Reset to a known value */ 158 msqs[i].msq_u.msg_perm._seq = 0; 159 } 160 161 mutex_init(&msgmutex, MUTEX_DEFAULT, IPL_NONE); 162 cv_init(&msg_realloc_cv, "msgrealc"); 163 msg_realloc_state = false; 164 } 165 166 static int 167 msgrealloc(int newmsgmni, int newmsgseg) 168 { 169 struct msgmap *new_msgmaps; 170 struct __msg *new_msghdrs, *new_free_msghdrs; 171 char *old_msgpool, *new_msgpool; 172 kmsq_t *new_msqs; 173 vaddr_t v; 174 int i, sz, msqid, newmsgmax, new_nfree_msgmaps; 175 short new_free_msgmaps; 176 177 if (newmsgmni < 1 || newmsgseg < 1) 178 return EINVAL; 179 180 /* Allocate the wired memory for our structures */ 181 newmsgmax = msginfo.msgssz * newmsgseg; 182 sz = ALIGN(newmsgmax) + 183 ALIGN(newmsgseg * sizeof(struct msgmap)) + 184 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 185 ALIGN(newmsgmni * sizeof(kmsq_t)); 186 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 187 UVM_KMF_WIRED|UVM_KMF_ZERO); 188 if (v == 0) 189 return ENOMEM; 190 191 mutex_enter(&msgmutex); 192 if (msg_realloc_state) { 193 mutex_exit(&msgmutex); 194 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 195 return EBUSY; 196 } 197 msg_realloc_state = true; 198 if (msg_waiters) { 199 /* 200 * Mark reallocation state, wake-up all waiters, 201 * and wait while they will all exit. 202 */ 203 for (i = 0; i < msginfo.msgmni; i++) 204 cv_broadcast(&msqs[i].msq_cv); 205 while (msg_waiters) 206 cv_wait(&msg_realloc_cv, &msgmutex); 207 } 208 old_msgpool = msgpool; 209 210 /* We cannot reallocate less memory than we use */ 211 i = 0; 212 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 213 struct msqid_ds *mptr; 214 kmsq_t *msq; 215 216 msq = &msqs[msqid]; 217 mptr = &msq->msq_u; 218 if (mptr->msg_qbytes || (mptr->msg_perm.mode & MSG_LOCKED)) 219 i = msqid; 220 } 221 if (i >= newmsgmni || (msginfo.msgseg - nfree_msgmaps) > newmsgseg) { 222 mutex_exit(&msgmutex); 223 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 224 return EBUSY; 225 } 226 227 new_msgpool = (void *)v; 228 new_msgmaps = (void *)(ALIGN(new_msgpool) + newmsgmax); 229 new_msghdrs = (void *)(ALIGN(new_msgmaps) + 230 newmsgseg * sizeof(struct msgmap)); 231 new_msqs = (void *)(ALIGN(new_msghdrs) + 232 msginfo.msgtql * sizeof(struct __msg)); 233 234 /* Initialize the structures */ 235 for (i = 0; i < (newmsgseg - 1); i++) 236 new_msgmaps[i].next = i + 1; 237 new_msgmaps[newmsgseg - 1].next = -1; 238 new_free_msgmaps = 0; 239 new_nfree_msgmaps = newmsgseg; 240 241 for (i = 0; i < (msginfo.msgtql - 1); i++) { 242 new_msghdrs[i].msg_type = 0; 243 new_msghdrs[i].msg_next = &new_msghdrs[i + 1]; 244 } 245 i = msginfo.msgtql - 1; 246 new_msghdrs[i].msg_type = 0; 247 new_msghdrs[i].msg_next = NULL; 248 new_free_msghdrs = &new_msghdrs[0]; 249 250 for (i = 0; i < newmsgmni; i++) { 251 new_msqs[i].msq_u.msg_qbytes = 0; 252 new_msqs[i].msq_u.msg_perm._seq = 0; 253 cv_init(&new_msqs[i].msq_cv, "msgwait"); 254 } 255 256 /* 257 * Copy all message queue identifiers, mesage headers and buffer 258 * pools to the new memory location. 259 */ 260 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 261 struct __msg *nmsghdr, *msghdr, *pmsghdr; 262 struct msqid_ds *nmptr, *mptr; 263 kmsq_t *nmsq, *msq; 264 265 msq = &msqs[msqid]; 266 mptr = &msq->msq_u; 267 268 if (mptr->msg_qbytes == 0 && 269 (mptr->msg_perm.mode & MSG_LOCKED) == 0) 270 continue; 271 272 nmsq = &new_msqs[msqid]; 273 nmptr = &nmsq->msq_u; 274 memcpy(nmptr, mptr, sizeof(struct msqid_ds)); 275 276 /* 277 * Go through the message headers, and and copy each 278 * one by taking the new ones, and thus defragmenting. 279 */ 280 nmsghdr = pmsghdr = NULL; 281 msghdr = mptr->_msg_first; 282 while (msghdr) { 283 short nnext = 0, next; 284 u_short msgsz, segcnt; 285 286 /* Take an entry from the new list of free msghdrs */ 287 nmsghdr = new_free_msghdrs; 288 KASSERT(nmsghdr != NULL); 289 new_free_msghdrs = nmsghdr->msg_next; 290 291 nmsghdr->msg_next = NULL; 292 if (pmsghdr) { 293 pmsghdr->msg_next = nmsghdr; 294 } else { 295 nmptr->_msg_first = nmsghdr; 296 pmsghdr = nmsghdr; 297 } 298 nmsghdr->msg_ts = msghdr->msg_ts; 299 nmsghdr->msg_spot = -1; 300 301 /* Compute the amount of segments and reserve them */ 302 msgsz = msghdr->msg_ts; 303 segcnt = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 304 if (segcnt == 0) 305 continue; 306 while (segcnt--) { 307 nnext = new_free_msgmaps; 308 new_free_msgmaps = new_msgmaps[nnext].next; 309 new_nfree_msgmaps--; 310 new_msgmaps[nnext].next = nmsghdr->msg_spot; 311 nmsghdr->msg_spot = nnext; 312 } 313 314 /* Copy all segments */ 315 KASSERT(nnext == nmsghdr->msg_spot); 316 next = msghdr->msg_spot; 317 while (msgsz > 0) { 318 size_t tlen; 319 320 if (msgsz >= msginfo.msgssz) { 321 tlen = msginfo.msgssz; 322 msgsz -= msginfo.msgssz; 323 } else { 324 tlen = msgsz; 325 msgsz = 0; 326 } 327 328 /* Copy the message buffer */ 329 memcpy(&new_msgpool[nnext * msginfo.msgssz], 330 &msgpool[next * msginfo.msgssz], tlen); 331 332 /* Next entry of the map */ 333 nnext = msgmaps[nnext].next; 334 next = msgmaps[next].next; 335 } 336 337 /* Next message header */ 338 msghdr = msghdr->msg_next; 339 } 340 nmptr->_msg_last = nmsghdr; 341 } 342 KASSERT((msginfo.msgseg - nfree_msgmaps) == 343 (newmsgseg - new_nfree_msgmaps)); 344 345 sz = ALIGN(msginfo.msgmax) + 346 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 347 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 348 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 349 350 for (i = 0; i < msginfo.msgmni; i++) 351 cv_destroy(&msqs[i].msq_cv); 352 353 /* Set the pointers and update the new values */ 354 msgpool = new_msgpool; 355 msgmaps = new_msgmaps; 356 msghdrs = new_msghdrs; 357 msqs = new_msqs; 358 359 free_msghdrs = new_free_msghdrs; 360 free_msgmaps = new_free_msgmaps; 361 nfree_msgmaps = new_nfree_msgmaps; 362 msginfo.msgmni = newmsgmni; 363 msginfo.msgseg = newmsgseg; 364 msginfo.msgmax = newmsgmax; 365 366 /* Reallocation completed - notify all waiters, if any */ 367 msg_realloc_state = false; 368 cv_broadcast(&msg_realloc_cv); 369 mutex_exit(&msgmutex); 370 371 uvm_km_free(kernel_map, (vaddr_t)old_msgpool, sz, UVM_KMF_WIRED); 372 return 0; 373 } 374 375 static void 376 msg_freehdr(struct __msg *msghdr) 377 { 378 379 KASSERT(mutex_owned(&msgmutex)); 380 381 while (msghdr->msg_ts > 0) { 382 short next; 383 KASSERT(msghdr->msg_spot >= 0); 384 KASSERT(msghdr->msg_spot < msginfo.msgseg); 385 386 next = msgmaps[msghdr->msg_spot].next; 387 msgmaps[msghdr->msg_spot].next = free_msgmaps; 388 free_msgmaps = msghdr->msg_spot; 389 nfree_msgmaps++; 390 msghdr->msg_spot = next; 391 if (msghdr->msg_ts >= msginfo.msgssz) 392 msghdr->msg_ts -= msginfo.msgssz; 393 else 394 msghdr->msg_ts = 0; 395 } 396 KASSERT(msghdr->msg_spot == -1); 397 msghdr->msg_next = free_msghdrs; 398 free_msghdrs = msghdr; 399 } 400 401 int 402 sys___msgctl13(struct lwp *l, void *v, register_t *retval) 403 { 404 struct sys___msgctl13_args /* { 405 syscallarg(int) msqid; 406 syscallarg(int) cmd; 407 syscallarg(struct msqid_ds *) buf; 408 } */ *uap = v; 409 struct msqid_ds msqbuf; 410 int cmd, error; 411 412 cmd = SCARG(uap, cmd); 413 414 if (cmd == IPC_SET) { 415 error = copyin(SCARG(uap, buf), &msqbuf, sizeof(msqbuf)); 416 if (error) 417 return (error); 418 } 419 420 error = msgctl1(l, SCARG(uap, msqid), cmd, 421 (cmd == IPC_SET || cmd == IPC_STAT) ? &msqbuf : NULL); 422 423 if (error == 0 && cmd == IPC_STAT) 424 error = copyout(&msqbuf, SCARG(uap, buf), sizeof(msqbuf)); 425 426 return (error); 427 } 428 429 int 430 msgctl1(struct lwp *l, int msqid, int cmd, struct msqid_ds *msqbuf) 431 { 432 kauth_cred_t cred = l->l_cred; 433 struct msqid_ds *msqptr; 434 kmsq_t *msq; 435 int error = 0, ix; 436 437 MSG_PRINTF(("call to msgctl1(%d, %d)\n", msqid, cmd)); 438 439 ix = IPCID_TO_IX(msqid); 440 441 mutex_enter(&msgmutex); 442 443 if (ix < 0 || ix >= msginfo.msgmni) { 444 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", ix, 445 msginfo.msgmni)); 446 error = EINVAL; 447 goto unlock; 448 } 449 450 msq = &msqs[ix]; 451 msqptr = &msq->msq_u; 452 453 if (msqptr->msg_qbytes == 0) { 454 MSG_PRINTF(("no such msqid\n")); 455 error = EINVAL; 456 goto unlock; 457 } 458 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqid)) { 459 MSG_PRINTF(("wrong sequence number\n")); 460 error = EINVAL; 461 goto unlock; 462 } 463 464 switch (cmd) { 465 case IPC_RMID: 466 { 467 struct __msg *msghdr; 468 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M)) != 0) 469 break; 470 /* Free the message headers */ 471 msghdr = msqptr->_msg_first; 472 while (msghdr != NULL) { 473 struct __msg *msghdr_tmp; 474 475 /* Free the segments of each message */ 476 msqptr->_msg_cbytes -= msghdr->msg_ts; 477 msqptr->msg_qnum--; 478 msghdr_tmp = msghdr; 479 msghdr = msghdr->msg_next; 480 msg_freehdr(msghdr_tmp); 481 } 482 KASSERT(msqptr->_msg_cbytes == 0); 483 KASSERT(msqptr->msg_qnum == 0); 484 485 /* Mark it as free */ 486 msqptr->msg_qbytes = 0; 487 cv_broadcast(&msq->msq_cv); 488 } 489 break; 490 491 case IPC_SET: 492 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M))) 493 break; 494 if (msqbuf->msg_qbytes > msqptr->msg_qbytes && 495 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 496 NULL) != 0) { 497 error = EPERM; 498 break; 499 } 500 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 501 MSG_PRINTF(("can't increase msg_qbytes beyond %d " 502 "(truncating)\n", msginfo.msgmnb)); 503 /* silently restrict qbytes to system limit */ 504 msqbuf->msg_qbytes = msginfo.msgmnb; 505 } 506 if (msqbuf->msg_qbytes == 0) { 507 MSG_PRINTF(("can't reduce msg_qbytes to 0\n")); 508 error = EINVAL; /* XXX non-standard errno! */ 509 break; 510 } 511 msqptr->msg_perm.uid = msqbuf->msg_perm.uid; 512 msqptr->msg_perm.gid = msqbuf->msg_perm.gid; 513 msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) | 514 (msqbuf->msg_perm.mode & 0777); 515 msqptr->msg_qbytes = msqbuf->msg_qbytes; 516 msqptr->msg_ctime = time_second; 517 break; 518 519 case IPC_STAT: 520 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 521 MSG_PRINTF(("requester doesn't have read access\n")); 522 break; 523 } 524 memcpy(msqbuf, msqptr, sizeof(struct msqid_ds)); 525 break; 526 527 default: 528 MSG_PRINTF(("invalid command %d\n", cmd)); 529 error = EINVAL; 530 break; 531 } 532 533 unlock: 534 mutex_exit(&msgmutex); 535 return (error); 536 } 537 538 int 539 sys_msgget(struct lwp *l, void *v, register_t *retval) 540 { 541 struct sys_msgget_args /* { 542 syscallarg(key_t) key; 543 syscallarg(int) msgflg; 544 } */ *uap = v; 545 int msqid, error = 0; 546 int key = SCARG(uap, key); 547 int msgflg = SCARG(uap, msgflg); 548 kauth_cred_t cred = l->l_cred; 549 struct msqid_ds *msqptr = NULL; 550 kmsq_t *msq; 551 552 mutex_enter(&msgmutex); 553 554 MSG_PRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 555 556 if (key != IPC_PRIVATE) { 557 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 558 msq = &msqs[msqid]; 559 msqptr = &msq->msq_u; 560 if (msqptr->msg_qbytes != 0 && 561 msqptr->msg_perm._key == key) 562 break; 563 } 564 if (msqid < msginfo.msgmni) { 565 MSG_PRINTF(("found public key\n")); 566 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 567 MSG_PRINTF(("not exclusive\n")); 568 error = EEXIST; 569 goto unlock; 570 } 571 if ((error = ipcperm(cred, &msqptr->msg_perm, 572 msgflg & 0700 ))) { 573 MSG_PRINTF(("requester doesn't have 0%o access\n", 574 msgflg & 0700)); 575 goto unlock; 576 } 577 goto found; 578 } 579 } 580 581 MSG_PRINTF(("need to allocate the msqid_ds\n")); 582 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 583 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 584 /* 585 * Look for an unallocated and unlocked msqid_ds. 586 * msqid_ds's can be locked by msgsnd or msgrcv while 587 * they are copying the message in/out. We can't 588 * re-use the entry until they release it. 589 */ 590 msq = &msqs[msqid]; 591 msqptr = &msq->msq_u; 592 if (msqptr->msg_qbytes == 0 && 593 (msqptr->msg_perm.mode & MSG_LOCKED) == 0) 594 break; 595 } 596 if (msqid == msginfo.msgmni) { 597 MSG_PRINTF(("no more msqid_ds's available\n")); 598 error = ENOSPC; 599 goto unlock; 600 } 601 MSG_PRINTF(("msqid %d is available\n", msqid)); 602 msqptr->msg_perm._key = key; 603 msqptr->msg_perm.cuid = kauth_cred_geteuid(cred); 604 msqptr->msg_perm.uid = kauth_cred_geteuid(cred); 605 msqptr->msg_perm.cgid = kauth_cred_getegid(cred); 606 msqptr->msg_perm.gid = kauth_cred_getegid(cred); 607 msqptr->msg_perm.mode = (msgflg & 0777); 608 /* Make sure that the returned msqid is unique */ 609 msqptr->msg_perm._seq++; 610 msqptr->_msg_first = NULL; 611 msqptr->_msg_last = NULL; 612 msqptr->_msg_cbytes = 0; 613 msqptr->msg_qnum = 0; 614 msqptr->msg_qbytes = msginfo.msgmnb; 615 msqptr->msg_lspid = 0; 616 msqptr->msg_lrpid = 0; 617 msqptr->msg_stime = 0; 618 msqptr->msg_rtime = 0; 619 msqptr->msg_ctime = time_second; 620 } else { 621 MSG_PRINTF(("didn't find it and wasn't asked to create it\n")); 622 error = ENOENT; 623 goto unlock; 624 } 625 626 found: 627 /* Construct the unique msqid */ 628 *retval = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm); 629 630 unlock: 631 mutex_exit(&msgmutex); 632 return (error); 633 } 634 635 int 636 sys_msgsnd(struct lwp *l, void *v, register_t *retval) 637 { 638 struct sys_msgsnd_args /* { 639 syscallarg(int) msqid; 640 syscallarg(const void *) msgp; 641 syscallarg(size_t) msgsz; 642 syscallarg(int) msgflg; 643 } */ *uap = v; 644 645 return msgsnd1(l, SCARG(uap, msqid), SCARG(uap, msgp), 646 SCARG(uap, msgsz), SCARG(uap, msgflg), sizeof(long), copyin); 647 } 648 649 int 650 msgsnd1(struct lwp *l, int msqidr, const char *user_msgp, size_t msgsz, 651 int msgflg, size_t typesz, copyin_t fetch_type) 652 { 653 int segs_needed, error = 0, msqid; 654 kauth_cred_t cred = l->l_cred; 655 struct msqid_ds *msqptr; 656 struct __msg *msghdr; 657 kmsq_t *msq; 658 short next; 659 660 MSG_PRINTF(("call to msgsnd(%d, %p, %lld, %d)\n", msqid, user_msgp, 661 (long long)msgsz, msgflg)); 662 663 msqid = IPCID_TO_IX(msqidr); 664 665 mutex_enter(&msgmutex); 666 if (msg_realloc_state) { 667 /* In case of reallocation, we will wait for completion */ 668 while (msg_realloc_state) 669 cv_wait(&msg_realloc_cv, &msgmutex); 670 } 671 672 if (msqid < 0 || msqid >= msginfo.msgmni) { 673 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 674 msginfo.msgmni)); 675 error = EINVAL; 676 goto unlock; 677 } 678 679 msq = &msqs[msqid]; 680 msqptr = &msq->msq_u; 681 682 if (msqptr->msg_qbytes == 0) { 683 MSG_PRINTF(("no such message queue id\n")); 684 error = EINVAL; 685 goto unlock; 686 } 687 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 688 MSG_PRINTF(("wrong sequence number\n")); 689 error = EINVAL; 690 goto unlock; 691 } 692 693 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_W))) { 694 MSG_PRINTF(("requester doesn't have write access\n")); 695 goto unlock; 696 } 697 698 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 699 MSG_PRINTF(("msgsz=%lld, msgssz=%d, segs_needed=%d\n", 700 (long long)msgsz, msginfo.msgssz, segs_needed)); 701 for (;;) { 702 int need_more_resources = 0; 703 704 /* 705 * check msgsz [cannot be negative since it is unsigned] 706 * (inside this loop in case msg_qbytes changes while we sleep) 707 */ 708 709 if (msgsz > msqptr->msg_qbytes) { 710 MSG_PRINTF(("msgsz > msqptr->msg_qbytes\n")); 711 error = EINVAL; 712 goto unlock; 713 } 714 715 if (msqptr->msg_perm.mode & MSG_LOCKED) { 716 MSG_PRINTF(("msqid is locked\n")); 717 need_more_resources = 1; 718 } 719 if (msgsz + msqptr->_msg_cbytes > msqptr->msg_qbytes) { 720 MSG_PRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 721 need_more_resources = 1; 722 } 723 if (segs_needed > nfree_msgmaps) { 724 MSG_PRINTF(("segs_needed > nfree_msgmaps\n")); 725 need_more_resources = 1; 726 } 727 if (free_msghdrs == NULL) { 728 MSG_PRINTF(("no more msghdrs\n")); 729 need_more_resources = 1; 730 } 731 732 if (need_more_resources) { 733 int we_own_it; 734 735 if ((msgflg & IPC_NOWAIT) != 0) { 736 MSG_PRINTF(("need more resources but caller " 737 "doesn't want to wait\n")); 738 error = EAGAIN; 739 goto unlock; 740 } 741 742 if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { 743 MSG_PRINTF(("we don't own the msqid_ds\n")); 744 we_own_it = 0; 745 } else { 746 /* Force later arrivals to wait for our 747 request */ 748 MSG_PRINTF(("we own the msqid_ds\n")); 749 msqptr->msg_perm.mode |= MSG_LOCKED; 750 we_own_it = 1; 751 } 752 753 msg_waiters++; 754 MSG_PRINTF(("goodnight\n")); 755 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 756 MSG_PRINTF(("good morning, error=%d\n", error)); 757 msg_waiters--; 758 759 /* Notify reallocator, in case of such state */ 760 if (msg_realloc_state) 761 cv_broadcast(&msg_realloc_cv); 762 763 if (we_own_it) 764 msqptr->msg_perm.mode &= ~MSG_LOCKED; 765 if (error || msg_realloc_state) { 766 MSG_PRINTF(("msgsnd: interrupted system " 767 "call\n")); 768 error = EINTR; 769 goto unlock; 770 } 771 772 /* 773 * Make sure that the msq queue still exists 774 */ 775 776 if (msqptr->msg_qbytes == 0) { 777 MSG_PRINTF(("msqid deleted\n")); 778 error = EIDRM; 779 goto unlock; 780 } 781 } else { 782 MSG_PRINTF(("got all the resources that we need\n")); 783 break; 784 } 785 } 786 787 /* 788 * We have the resources that we need. 789 * Make sure! 790 */ 791 792 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 793 KASSERT(segs_needed <= nfree_msgmaps); 794 KASSERT(msgsz + msqptr->_msg_cbytes <= msqptr->msg_qbytes); 795 KASSERT(free_msghdrs != NULL); 796 797 /* 798 * Re-lock the msqid_ds in case we page-fault when copying in the 799 * message 800 */ 801 802 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 803 msqptr->msg_perm.mode |= MSG_LOCKED; 804 805 /* 806 * Allocate a message header 807 */ 808 809 msghdr = free_msghdrs; 810 free_msghdrs = msghdr->msg_next; 811 msghdr->msg_spot = -1; 812 msghdr->msg_ts = msgsz; 813 814 /* 815 * Allocate space for the message 816 */ 817 818 while (segs_needed > 0) { 819 KASSERT(nfree_msgmaps > 0); 820 KASSERT(free_msgmaps != -1); 821 KASSERT(free_msgmaps < msginfo.msgseg); 822 823 next = free_msgmaps; 824 MSG_PRINTF(("allocating segment %d to message\n", next)); 825 free_msgmaps = msgmaps[next].next; 826 nfree_msgmaps--; 827 msgmaps[next].next = msghdr->msg_spot; 828 msghdr->msg_spot = next; 829 segs_needed--; 830 } 831 832 /* 833 * Copy in the message type 834 */ 835 mutex_exit(&msgmutex); 836 error = (*fetch_type)(user_msgp, &msghdr->msg_type, typesz); 837 mutex_enter(&msgmutex); 838 if (error != 0) { 839 MSG_PRINTF(("error %d copying the message type\n", error)); 840 msg_freehdr(msghdr); 841 msqptr->msg_perm.mode &= ~MSG_LOCKED; 842 cv_broadcast(&msq->msq_cv); 843 goto unlock; 844 } 845 user_msgp += typesz; 846 847 /* 848 * Validate the message type 849 */ 850 851 if (msghdr->msg_type < 1) { 852 msg_freehdr(msghdr); 853 msqptr->msg_perm.mode &= ~MSG_LOCKED; 854 cv_broadcast(&msq->msq_cv); 855 MSG_PRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 856 goto unlock; 857 } 858 859 /* 860 * Copy in the message body 861 */ 862 863 next = msghdr->msg_spot; 864 while (msgsz > 0) { 865 size_t tlen; 866 KASSERT(next > -1); 867 KASSERT(next < msginfo.msgseg); 868 869 if (msgsz > msginfo.msgssz) 870 tlen = msginfo.msgssz; 871 else 872 tlen = msgsz; 873 mutex_exit(&msgmutex); 874 error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen); 875 mutex_enter(&msgmutex); 876 if (error != 0) { 877 MSG_PRINTF(("error %d copying in message segment\n", 878 error)); 879 msg_freehdr(msghdr); 880 msqptr->msg_perm.mode &= ~MSG_LOCKED; 881 cv_broadcast(&msq->msq_cv); 882 goto unlock; 883 } 884 msgsz -= tlen; 885 user_msgp += tlen; 886 next = msgmaps[next].next; 887 } 888 KASSERT(next == -1); 889 890 /* 891 * We've got the message. Unlock the msqid_ds. 892 */ 893 894 msqptr->msg_perm.mode &= ~MSG_LOCKED; 895 896 /* 897 * Make sure that the msqid_ds is still allocated. 898 */ 899 900 if (msqptr->msg_qbytes == 0) { 901 msg_freehdr(msghdr); 902 cv_broadcast(&msq->msq_cv); 903 error = EIDRM; 904 goto unlock; 905 } 906 907 /* 908 * Put the message into the queue 909 */ 910 911 if (msqptr->_msg_first == NULL) { 912 msqptr->_msg_first = msghdr; 913 msqptr->_msg_last = msghdr; 914 } else { 915 msqptr->_msg_last->msg_next = msghdr; 916 msqptr->_msg_last = msghdr; 917 } 918 msqptr->_msg_last->msg_next = NULL; 919 920 msqptr->_msg_cbytes += msghdr->msg_ts; 921 msqptr->msg_qnum++; 922 msqptr->msg_lspid = l->l_proc->p_pid; 923 msqptr->msg_stime = time_second; 924 925 cv_broadcast(&msq->msq_cv); 926 927 unlock: 928 mutex_exit(&msgmutex); 929 return error; 930 } 931 932 int 933 sys_msgrcv(struct lwp *l, void *v, register_t *retval) 934 { 935 struct sys_msgrcv_args /* { 936 syscallarg(int) msqid; 937 syscallarg(void *) msgp; 938 syscallarg(size_t) msgsz; 939 syscallarg(long) msgtyp; 940 syscallarg(int) msgflg; 941 } */ *uap = v; 942 943 return msgrcv1(l, SCARG(uap, msqid), SCARG(uap, msgp), 944 SCARG(uap, msgsz), SCARG(uap, msgtyp), SCARG(uap, msgflg), 945 sizeof(long), copyout, retval); 946 } 947 948 int 949 msgrcv1(struct lwp *l, int msqidr, char *user_msgp, size_t msgsz, long msgtyp, 950 int msgflg, size_t typesz, copyout_t put_type, register_t *retval) 951 { 952 size_t len; 953 kauth_cred_t cred = l->l_cred; 954 struct msqid_ds *msqptr; 955 struct __msg *msghdr; 956 int error = 0, msqid; 957 kmsq_t *msq; 958 short next; 959 960 MSG_PRINTF(("call to msgrcv(%d, %p, %lld, %ld, %d)\n", msqid, 961 user_msgp, (long long)msgsz, msgtyp, msgflg)); 962 963 msqid = IPCID_TO_IX(msqidr); 964 965 mutex_enter(&msgmutex); 966 if (msg_realloc_state) { 967 /* In case of reallocation, we will wait for completion */ 968 while (msg_realloc_state) 969 cv_wait(&msg_realloc_cv, &msgmutex); 970 } 971 972 if (msqid < 0 || msqid >= msginfo.msgmni) { 973 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 974 msginfo.msgmni)); 975 error = EINVAL; 976 goto unlock; 977 } 978 979 msq = &msqs[msqid]; 980 msqptr = &msq->msq_u; 981 982 if (msqptr->msg_qbytes == 0) { 983 MSG_PRINTF(("no such message queue id\n")); 984 error = EINVAL; 985 goto unlock; 986 } 987 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 988 MSG_PRINTF(("wrong sequence number\n")); 989 error = EINVAL; 990 goto unlock; 991 } 992 993 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 994 MSG_PRINTF(("requester doesn't have read access\n")); 995 goto unlock; 996 } 997 998 msghdr = NULL; 999 while (msghdr == NULL) { 1000 if (msgtyp == 0) { 1001 msghdr = msqptr->_msg_first; 1002 if (msghdr != NULL) { 1003 if (msgsz < msghdr->msg_ts && 1004 (msgflg & MSG_NOERROR) == 0) { 1005 MSG_PRINTF(("first msg on the queue " 1006 "is too big (want %lld, got %d)\n", 1007 (long long)msgsz, msghdr->msg_ts)); 1008 error = E2BIG; 1009 goto unlock; 1010 } 1011 if (msqptr->_msg_first == msqptr->_msg_last) { 1012 msqptr->_msg_first = NULL; 1013 msqptr->_msg_last = NULL; 1014 } else { 1015 msqptr->_msg_first = msghdr->msg_next; 1016 KASSERT(msqptr->_msg_first != NULL); 1017 } 1018 } 1019 } else { 1020 struct __msg *previous; 1021 struct __msg **prev; 1022 1023 for (previous = NULL, prev = &msqptr->_msg_first; 1024 (msghdr = *prev) != NULL; 1025 previous = msghdr, prev = &msghdr->msg_next) { 1026 /* 1027 * Is this message's type an exact match or is 1028 * this message's type less than or equal to 1029 * the absolute value of a negative msgtyp? 1030 * Note that the second half of this test can 1031 * NEVER be true if msgtyp is positive since 1032 * msg_type is always positive! 1033 */ 1034 1035 if (msgtyp != msghdr->msg_type && 1036 msghdr->msg_type > -msgtyp) 1037 continue; 1038 1039 MSG_PRINTF(("found message type %ld, requested %ld\n", 1040 msghdr->msg_type, msgtyp)); 1041 if (msgsz < msghdr->msg_ts && 1042 (msgflg & MSG_NOERROR) == 0) { 1043 MSG_PRINTF(("requested message on the queue " 1044 "is too big (want %lld, got %d)\n", 1045 (long long)msgsz, msghdr->msg_ts)); 1046 error = E2BIG; 1047 goto unlock; 1048 } 1049 *prev = msghdr->msg_next; 1050 if (msghdr != msqptr->_msg_last) 1051 break; 1052 if (previous == NULL) { 1053 KASSERT(prev == &msqptr->_msg_first); 1054 msqptr->_msg_first = NULL; 1055 msqptr->_msg_last = NULL; 1056 } else { 1057 KASSERT(prev != &msqptr->_msg_first); 1058 msqptr->_msg_last = previous; 1059 } 1060 break; 1061 } 1062 } 1063 1064 /* 1065 * We've either extracted the msghdr for the appropriate 1066 * message or there isn't one. 1067 * If there is one then bail out of this loop. 1068 */ 1069 if (msghdr != NULL) 1070 break; 1071 1072 /* 1073 * Hmph! No message found. Does the user want to wait? 1074 */ 1075 1076 if ((msgflg & IPC_NOWAIT) != 0) { 1077 MSG_PRINTF(("no appropriate message found (msgtyp=%ld)\n", 1078 msgtyp)); 1079 error = ENOMSG; 1080 goto unlock; 1081 } 1082 1083 /* 1084 * Wait for something to happen 1085 */ 1086 1087 msg_waiters++; 1088 MSG_PRINTF(("msgrcv: goodnight\n")); 1089 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 1090 MSG_PRINTF(("msgrcv: good morning (error=%d)\n", error)); 1091 msg_waiters--; 1092 1093 /* Notify reallocator, in case of such state */ 1094 if (msg_realloc_state) 1095 cv_broadcast(&msg_realloc_cv); 1096 1097 if (error || msg_realloc_state) { 1098 MSG_PRINTF(("msgsnd: interrupted system call\n")); 1099 error = EINTR; 1100 goto unlock; 1101 } 1102 1103 /* 1104 * Make sure that the msq queue still exists 1105 */ 1106 1107 if (msqptr->msg_qbytes == 0 || 1108 msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 1109 MSG_PRINTF(("msqid deleted\n")); 1110 error = EIDRM; 1111 goto unlock; 1112 } 1113 } 1114 1115 /* 1116 * Return the message to the user. 1117 * 1118 * First, do the bookkeeping (before we risk being interrupted). 1119 */ 1120 1121 msqptr->_msg_cbytes -= msghdr->msg_ts; 1122 msqptr->msg_qnum--; 1123 msqptr->msg_lrpid = l->l_proc->p_pid; 1124 msqptr->msg_rtime = time_second; 1125 1126 /* 1127 * Make msgsz the actual amount that we'll be returning. 1128 * Note that this effectively truncates the message if it is too long 1129 * (since msgsz is never increased). 1130 */ 1131 1132 MSG_PRINTF(("found a message, msgsz=%lld, msg_ts=%d\n", 1133 (long long)msgsz, msghdr->msg_ts)); 1134 if (msgsz > msghdr->msg_ts) 1135 msgsz = msghdr->msg_ts; 1136 1137 /* 1138 * Return the type to the user. 1139 */ 1140 mutex_exit(&msgmutex); 1141 error = (*put_type)(&msghdr->msg_type, user_msgp, typesz); 1142 mutex_enter(&msgmutex); 1143 if (error != 0) { 1144 MSG_PRINTF(("error (%d) copying out message type\n", error)); 1145 msg_freehdr(msghdr); 1146 cv_broadcast(&msq->msq_cv); 1147 goto unlock; 1148 } 1149 user_msgp += typesz; 1150 1151 /* 1152 * Return the segments to the user 1153 */ 1154 1155 next = msghdr->msg_spot; 1156 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1157 size_t tlen; 1158 KASSERT(next > -1); 1159 KASSERT(next < msginfo.msgseg); 1160 1161 if (msgsz - len > msginfo.msgssz) 1162 tlen = msginfo.msgssz; 1163 else 1164 tlen = msgsz - len; 1165 mutex_exit(&msgmutex); 1166 error = (*put_type)(&msgpool[next * msginfo.msgssz], 1167 user_msgp, tlen); 1168 mutex_enter(&msgmutex); 1169 if (error != 0) { 1170 MSG_PRINTF(("error (%d) copying out message segment\n", 1171 error)); 1172 msg_freehdr(msghdr); 1173 cv_broadcast(&msq->msq_cv); 1174 goto unlock; 1175 } 1176 user_msgp += tlen; 1177 next = msgmaps[next].next; 1178 } 1179 1180 /* 1181 * Done, return the actual number of bytes copied out. 1182 */ 1183 1184 msg_freehdr(msghdr); 1185 cv_broadcast(&msq->msq_cv); 1186 *retval = msgsz; 1187 1188 unlock: 1189 mutex_exit(&msgmutex); 1190 return error; 1191 } 1192 1193 /* 1194 * Sysctl initialization and nodes. 1195 */ 1196 1197 static int 1198 sysctl_ipc_msgmni(SYSCTLFN_ARGS) 1199 { 1200 int newsize, error; 1201 struct sysctlnode node; 1202 node = *rnode; 1203 node.sysctl_data = &newsize; 1204 1205 newsize = msginfo.msgmni; 1206 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1207 if (error || newp == NULL) 1208 return error; 1209 1210 return msgrealloc(newsize, msginfo.msgseg); 1211 } 1212 1213 static int 1214 sysctl_ipc_msgseg(SYSCTLFN_ARGS) 1215 { 1216 int newsize, error; 1217 struct sysctlnode node; 1218 node = *rnode; 1219 node.sysctl_data = &newsize; 1220 1221 newsize = msginfo.msgseg; 1222 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1223 if (error || newp == NULL) 1224 return error; 1225 1226 return msgrealloc(msginfo.msgmni, newsize); 1227 } 1228 1229 SYSCTL_SETUP(sysctl_ipc_msg_setup, "sysctl kern.ipc subtree setup") 1230 { 1231 const struct sysctlnode *node = NULL; 1232 1233 sysctl_createv(clog, 0, NULL, NULL, 1234 CTLFLAG_PERMANENT, 1235 CTLTYPE_NODE, "kern", NULL, 1236 NULL, 0, NULL, 0, 1237 CTL_KERN, CTL_EOL); 1238 sysctl_createv(clog, 0, NULL, &node, 1239 CTLFLAG_PERMANENT, 1240 CTLTYPE_NODE, "ipc", 1241 SYSCTL_DESCR("SysV IPC options"), 1242 NULL, 0, NULL, 0, 1243 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1244 1245 if (node == NULL) 1246 return; 1247 1248 sysctl_createv(clog, 0, &node, NULL, 1249 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1250 CTLTYPE_INT, "msgmni", 1251 SYSCTL_DESCR("Max number of message queue identifiers"), 1252 sysctl_ipc_msgmni, 0, &msginfo.msgmni, 0, 1253 CTL_CREATE, CTL_EOL); 1254 sysctl_createv(clog, 0, &node, NULL, 1255 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1256 CTLTYPE_INT, "msgseg", 1257 SYSCTL_DESCR("Max number of number of message segments"), 1258 sysctl_ipc_msgseg, 0, &msginfo.msgseg, 0, 1259 CTL_CREATE, CTL_EOL); 1260 } 1261