1 /* $NetBSD: sysv_msg.c,v 1.62 2011/07/30 06:19:02 uebayasi 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 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Implementation of SVID messages 35 * 36 * Author: Daniel Boulet 37 * 38 * Copyright 1993 Daniel Boulet and RTMX Inc. 39 * 40 * This system call was implemented by Daniel Boulet under contract from RTMX. 41 * 42 * Redistribution and use in source forms, with and without modification, 43 * are permitted provided that this entire comment appears intact. 44 * 45 * Redistribution in binary form may occur without any restrictions. 46 * Obviously, it would be nice if you gave credit where credit is due 47 * but requiring it would be too onerous. 48 * 49 * This software is provided ``AS IS'' without any warranties of any kind. 50 */ 51 52 #include <sys/cdefs.h> 53 __KERNEL_RCSID(0, "$NetBSD: sysv_msg.c,v 1.62 2011/07/30 06:19:02 uebayasi Exp $"); 54 55 #define SYSVMSG 56 57 #include <sys/param.h> 58 #include <sys/kernel.h> 59 #include <sys/msg.h> 60 #include <sys/sysctl.h> 61 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */ 62 #include <sys/syscallargs.h> 63 #include <sys/kauth.h> 64 65 #define MSG_DEBUG 66 #undef MSG_DEBUG_OK 67 68 #ifdef MSG_DEBUG_OK 69 #define MSG_PRINTF(a) printf a 70 #else 71 #define MSG_PRINTF(a) 72 #endif 73 74 static int nfree_msgmaps; /* # of free map entries */ 75 static short free_msgmaps; /* head of linked list of free map entries */ 76 static struct __msg *free_msghdrs; /* list of free msg headers */ 77 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 78 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 79 static struct __msg *msghdrs; /* MSGTQL msg headers */ 80 81 kmsq_t *msqs; /* MSGMNI msqid_ds struct's */ 82 kmutex_t msgmutex; /* subsystem lock */ 83 84 static u_int msg_waiters = 0; /* total number of msgrcv waiters */ 85 static bool msg_realloc_state; 86 static kcondvar_t msg_realloc_cv; 87 88 static void msg_freehdr(struct __msg *); 89 90 void 91 msginit(void) 92 { 93 int i, sz; 94 vaddr_t v; 95 96 /* 97 * msginfo.msgssz should be a power of two for efficiency reasons. 98 * It is also pretty silly if msginfo.msgssz is less than 8 99 * or greater than about 256 so ... 100 */ 101 102 i = 8; 103 while (i < 1024 && i != msginfo.msgssz) 104 i <<= 1; 105 if (i != msginfo.msgssz) { 106 panic("msginfo.msgssz = %d, not a small power of 2", 107 msginfo.msgssz); 108 } 109 110 if (msginfo.msgseg > 32767) { 111 panic("msginfo.msgseg = %d > 32767", msginfo.msgseg); 112 } 113 114 /* Allocate the wired memory for our structures */ 115 sz = ALIGN(msginfo.msgmax) + 116 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 117 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 118 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 119 sz = round_page(sz); 120 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 121 if (v == 0) 122 panic("sysv_msg: cannot allocate memory"); 123 msgpool = (void *)v; 124 msgmaps = (void *)((uintptr_t)msgpool + ALIGN(msginfo.msgmax)); 125 msghdrs = (void *)((uintptr_t)msgmaps + 126 ALIGN(msginfo.msgseg * sizeof(struct msgmap))); 127 msqs = (void *)((uintptr_t)msghdrs + 128 ALIGN(msginfo.msgtql * sizeof(struct __msg))); 129 130 for (i = 0; i < (msginfo.msgseg - 1); i++) 131 msgmaps[i].next = i + 1; 132 msgmaps[msginfo.msgseg - 1].next = -1; 133 134 free_msgmaps = 0; 135 nfree_msgmaps = msginfo.msgseg; 136 137 for (i = 0; i < (msginfo.msgtql - 1); i++) { 138 msghdrs[i].msg_type = 0; 139 msghdrs[i].msg_next = &msghdrs[i + 1]; 140 } 141 i = msginfo.msgtql - 1; 142 msghdrs[i].msg_type = 0; 143 msghdrs[i].msg_next = NULL; 144 free_msghdrs = &msghdrs[0]; 145 146 for (i = 0; i < msginfo.msgmni; i++) { 147 cv_init(&msqs[i].msq_cv, "msgwait"); 148 /* Implies entry is available */ 149 msqs[i].msq_u.msg_qbytes = 0; 150 /* Reset to a known value */ 151 msqs[i].msq_u.msg_perm._seq = 0; 152 } 153 154 mutex_init(&msgmutex, MUTEX_DEFAULT, IPL_NONE); 155 cv_init(&msg_realloc_cv, "msgrealc"); 156 msg_realloc_state = false; 157 } 158 159 static int 160 msgrealloc(int newmsgmni, int newmsgseg) 161 { 162 struct msgmap *new_msgmaps; 163 struct __msg *new_msghdrs, *new_free_msghdrs; 164 char *old_msgpool, *new_msgpool; 165 kmsq_t *new_msqs; 166 vaddr_t v; 167 int i, sz, msqid, newmsgmax, new_nfree_msgmaps; 168 short new_free_msgmaps; 169 170 if (newmsgmni < 1 || newmsgseg < 1) 171 return EINVAL; 172 173 /* Allocate the wired memory for our structures */ 174 newmsgmax = msginfo.msgssz * newmsgseg; 175 sz = ALIGN(newmsgmax) + 176 ALIGN(newmsgseg * sizeof(struct msgmap)) + 177 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 178 ALIGN(newmsgmni * sizeof(kmsq_t)); 179 sz = round_page(sz); 180 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 181 if (v == 0) 182 return ENOMEM; 183 184 mutex_enter(&msgmutex); 185 if (msg_realloc_state) { 186 mutex_exit(&msgmutex); 187 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 188 return EBUSY; 189 } 190 msg_realloc_state = true; 191 if (msg_waiters) { 192 /* 193 * Mark reallocation state, wake-up all waiters, 194 * and wait while they will all exit. 195 */ 196 for (i = 0; i < msginfo.msgmni; i++) 197 cv_broadcast(&msqs[i].msq_cv); 198 while (msg_waiters) 199 cv_wait(&msg_realloc_cv, &msgmutex); 200 } 201 old_msgpool = msgpool; 202 203 /* We cannot reallocate less memory than we use */ 204 i = 0; 205 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 206 struct msqid_ds *mptr; 207 kmsq_t *msq; 208 209 msq = &msqs[msqid]; 210 mptr = &msq->msq_u; 211 if (mptr->msg_qbytes || (mptr->msg_perm.mode & MSG_LOCKED)) 212 i = msqid; 213 } 214 if (i >= newmsgmni || (msginfo.msgseg - nfree_msgmaps) > newmsgseg) { 215 mutex_exit(&msgmutex); 216 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 217 return EBUSY; 218 } 219 220 new_msgpool = (void *)v; 221 new_msgmaps = (void *)((uintptr_t)new_msgpool + ALIGN(newmsgmax)); 222 new_msghdrs = (void *)((uintptr_t)new_msgmaps + 223 ALIGN(newmsgseg * sizeof(struct msgmap))); 224 new_msqs = (void *)((uintptr_t)new_msghdrs + 225 ALIGN(msginfo.msgtql * sizeof(struct __msg))); 226 227 /* Initialize the structures */ 228 for (i = 0; i < (newmsgseg - 1); i++) 229 new_msgmaps[i].next = i + 1; 230 new_msgmaps[newmsgseg - 1].next = -1; 231 new_free_msgmaps = 0; 232 new_nfree_msgmaps = newmsgseg; 233 234 for (i = 0; i < (msginfo.msgtql - 1); i++) { 235 new_msghdrs[i].msg_type = 0; 236 new_msghdrs[i].msg_next = &new_msghdrs[i + 1]; 237 } 238 i = msginfo.msgtql - 1; 239 new_msghdrs[i].msg_type = 0; 240 new_msghdrs[i].msg_next = NULL; 241 new_free_msghdrs = &new_msghdrs[0]; 242 243 for (i = 0; i < newmsgmni; i++) { 244 new_msqs[i].msq_u.msg_qbytes = 0; 245 new_msqs[i].msq_u.msg_perm._seq = 0; 246 cv_init(&new_msqs[i].msq_cv, "msgwait"); 247 } 248 249 /* 250 * Copy all message queue identifiers, mesage headers and buffer 251 * pools to the new memory location. 252 */ 253 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 254 struct __msg *nmsghdr, *msghdr, *pmsghdr; 255 struct msqid_ds *nmptr, *mptr; 256 kmsq_t *nmsq, *msq; 257 258 msq = &msqs[msqid]; 259 mptr = &msq->msq_u; 260 261 if (mptr->msg_qbytes == 0 && 262 (mptr->msg_perm.mode & MSG_LOCKED) == 0) 263 continue; 264 265 nmsq = &new_msqs[msqid]; 266 nmptr = &nmsq->msq_u; 267 memcpy(nmptr, mptr, sizeof(struct msqid_ds)); 268 269 /* 270 * Go through the message headers, and and copy each 271 * one by taking the new ones, and thus defragmenting. 272 */ 273 nmsghdr = pmsghdr = NULL; 274 msghdr = mptr->_msg_first; 275 while (msghdr) { 276 short nnext = 0, next; 277 u_short msgsz, segcnt; 278 279 /* Take an entry from the new list of free msghdrs */ 280 nmsghdr = new_free_msghdrs; 281 KASSERT(nmsghdr != NULL); 282 new_free_msghdrs = nmsghdr->msg_next; 283 284 nmsghdr->msg_next = NULL; 285 if (pmsghdr) { 286 pmsghdr->msg_next = nmsghdr; 287 } else { 288 nmptr->_msg_first = nmsghdr; 289 pmsghdr = nmsghdr; 290 } 291 nmsghdr->msg_ts = msghdr->msg_ts; 292 nmsghdr->msg_spot = -1; 293 294 /* Compute the amount of segments and reserve them */ 295 msgsz = msghdr->msg_ts; 296 segcnt = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 297 if (segcnt == 0) 298 continue; 299 while (segcnt--) { 300 nnext = new_free_msgmaps; 301 new_free_msgmaps = new_msgmaps[nnext].next; 302 new_nfree_msgmaps--; 303 new_msgmaps[nnext].next = nmsghdr->msg_spot; 304 nmsghdr->msg_spot = nnext; 305 } 306 307 /* Copy all segments */ 308 KASSERT(nnext == nmsghdr->msg_spot); 309 next = msghdr->msg_spot; 310 while (msgsz > 0) { 311 size_t tlen; 312 313 if (msgsz >= msginfo.msgssz) { 314 tlen = msginfo.msgssz; 315 msgsz -= msginfo.msgssz; 316 } else { 317 tlen = msgsz; 318 msgsz = 0; 319 } 320 321 /* Copy the message buffer */ 322 memcpy(&new_msgpool[nnext * msginfo.msgssz], 323 &msgpool[next * msginfo.msgssz], tlen); 324 325 /* Next entry of the map */ 326 nnext = msgmaps[nnext].next; 327 next = msgmaps[next].next; 328 } 329 330 /* Next message header */ 331 msghdr = msghdr->msg_next; 332 } 333 nmptr->_msg_last = nmsghdr; 334 } 335 KASSERT((msginfo.msgseg - nfree_msgmaps) == 336 (newmsgseg - new_nfree_msgmaps)); 337 338 sz = ALIGN(msginfo.msgmax) + 339 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 340 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 341 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 342 sz = round_page(sz); 343 344 for (i = 0; i < msginfo.msgmni; i++) 345 cv_destroy(&msqs[i].msq_cv); 346 347 /* Set the pointers and update the new values */ 348 msgpool = new_msgpool; 349 msgmaps = new_msgmaps; 350 msghdrs = new_msghdrs; 351 msqs = new_msqs; 352 353 free_msghdrs = new_free_msghdrs; 354 free_msgmaps = new_free_msgmaps; 355 nfree_msgmaps = new_nfree_msgmaps; 356 msginfo.msgmni = newmsgmni; 357 msginfo.msgseg = newmsgseg; 358 msginfo.msgmax = newmsgmax; 359 360 /* Reallocation completed - notify all waiters, if any */ 361 msg_realloc_state = false; 362 cv_broadcast(&msg_realloc_cv); 363 mutex_exit(&msgmutex); 364 365 uvm_km_free(kernel_map, (vaddr_t)old_msgpool, sz, UVM_KMF_WIRED); 366 return 0; 367 } 368 369 static void 370 msg_freehdr(struct __msg *msghdr) 371 { 372 373 KASSERT(mutex_owned(&msgmutex)); 374 375 while (msghdr->msg_ts > 0) { 376 short next; 377 KASSERT(msghdr->msg_spot >= 0); 378 KASSERT(msghdr->msg_spot < msginfo.msgseg); 379 380 next = msgmaps[msghdr->msg_spot].next; 381 msgmaps[msghdr->msg_spot].next = free_msgmaps; 382 free_msgmaps = msghdr->msg_spot; 383 nfree_msgmaps++; 384 msghdr->msg_spot = next; 385 if (msghdr->msg_ts >= msginfo.msgssz) 386 msghdr->msg_ts -= msginfo.msgssz; 387 else 388 msghdr->msg_ts = 0; 389 } 390 KASSERT(msghdr->msg_spot == -1); 391 msghdr->msg_next = free_msghdrs; 392 free_msghdrs = msghdr; 393 } 394 395 int 396 sys___msgctl50(struct lwp *l, const struct sys___msgctl50_args *uap, 397 register_t *retval) 398 { 399 /* { 400 syscallarg(int) msqid; 401 syscallarg(int) cmd; 402 syscallarg(struct msqid_ds *) buf; 403 } */ 404 struct msqid_ds msqbuf; 405 int cmd, error; 406 407 cmd = SCARG(uap, cmd); 408 409 if (cmd == IPC_SET) { 410 error = copyin(SCARG(uap, buf), &msqbuf, sizeof(msqbuf)); 411 if (error) 412 return (error); 413 } 414 415 error = msgctl1(l, SCARG(uap, msqid), cmd, 416 (cmd == IPC_SET || cmd == IPC_STAT) ? &msqbuf : NULL); 417 418 if (error == 0 && cmd == IPC_STAT) 419 error = copyout(&msqbuf, SCARG(uap, buf), sizeof(msqbuf)); 420 421 return (error); 422 } 423 424 int 425 msgctl1(struct lwp *l, int msqid, int cmd, struct msqid_ds *msqbuf) 426 { 427 kauth_cred_t cred = l->l_cred; 428 struct msqid_ds *msqptr; 429 kmsq_t *msq; 430 int error = 0, ix; 431 432 MSG_PRINTF(("call to msgctl1(%d, %d)\n", msqid, cmd)); 433 434 ix = IPCID_TO_IX(msqid); 435 436 mutex_enter(&msgmutex); 437 438 if (ix < 0 || ix >= msginfo.msgmni) { 439 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", ix, 440 msginfo.msgmni)); 441 error = EINVAL; 442 goto unlock; 443 } 444 445 msq = &msqs[ix]; 446 msqptr = &msq->msq_u; 447 448 if (msqptr->msg_qbytes == 0) { 449 MSG_PRINTF(("no such msqid\n")); 450 error = EINVAL; 451 goto unlock; 452 } 453 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqid)) { 454 MSG_PRINTF(("wrong sequence number\n")); 455 error = EINVAL; 456 goto unlock; 457 } 458 459 switch (cmd) { 460 case IPC_RMID: 461 { 462 struct __msg *msghdr; 463 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M)) != 0) 464 break; 465 /* Free the message headers */ 466 msghdr = msqptr->_msg_first; 467 while (msghdr != NULL) { 468 struct __msg *msghdr_tmp; 469 470 /* Free the segments of each message */ 471 msqptr->_msg_cbytes -= msghdr->msg_ts; 472 msqptr->msg_qnum--; 473 msghdr_tmp = msghdr; 474 msghdr = msghdr->msg_next; 475 msg_freehdr(msghdr_tmp); 476 } 477 KASSERT(msqptr->_msg_cbytes == 0); 478 KASSERT(msqptr->msg_qnum == 0); 479 480 /* Mark it as free */ 481 msqptr->msg_qbytes = 0; 482 cv_broadcast(&msq->msq_cv); 483 } 484 break; 485 486 case IPC_SET: 487 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M))) 488 break; 489 if (msqbuf->msg_qbytes > msqptr->msg_qbytes && 490 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 491 NULL) != 0) { 492 error = EPERM; 493 break; 494 } 495 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 496 MSG_PRINTF(("can't increase msg_qbytes beyond %d " 497 "(truncating)\n", msginfo.msgmnb)); 498 /* silently restrict qbytes to system limit */ 499 msqbuf->msg_qbytes = msginfo.msgmnb; 500 } 501 if (msqbuf->msg_qbytes == 0) { 502 MSG_PRINTF(("can't reduce msg_qbytes to 0\n")); 503 error = EINVAL; /* XXX non-standard errno! */ 504 break; 505 } 506 msqptr->msg_perm.uid = msqbuf->msg_perm.uid; 507 msqptr->msg_perm.gid = msqbuf->msg_perm.gid; 508 msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) | 509 (msqbuf->msg_perm.mode & 0777); 510 msqptr->msg_qbytes = msqbuf->msg_qbytes; 511 msqptr->msg_ctime = time_second; 512 break; 513 514 case IPC_STAT: 515 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 516 MSG_PRINTF(("requester doesn't have read access\n")); 517 break; 518 } 519 memcpy(msqbuf, msqptr, sizeof(struct msqid_ds)); 520 break; 521 522 default: 523 MSG_PRINTF(("invalid command %d\n", cmd)); 524 error = EINVAL; 525 break; 526 } 527 528 unlock: 529 mutex_exit(&msgmutex); 530 return (error); 531 } 532 533 int 534 sys_msgget(struct lwp *l, const struct sys_msgget_args *uap, register_t *retval) 535 { 536 /* { 537 syscallarg(key_t) key; 538 syscallarg(int) msgflg; 539 } */ 540 int msqid, error = 0; 541 int key = SCARG(uap, key); 542 int msgflg = SCARG(uap, msgflg); 543 kauth_cred_t cred = l->l_cred; 544 struct msqid_ds *msqptr = NULL; 545 kmsq_t *msq; 546 547 mutex_enter(&msgmutex); 548 549 MSG_PRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 550 551 if (key != IPC_PRIVATE) { 552 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 553 msq = &msqs[msqid]; 554 msqptr = &msq->msq_u; 555 if (msqptr->msg_qbytes != 0 && 556 msqptr->msg_perm._key == key) 557 break; 558 } 559 if (msqid < msginfo.msgmni) { 560 MSG_PRINTF(("found public key\n")); 561 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 562 MSG_PRINTF(("not exclusive\n")); 563 error = EEXIST; 564 goto unlock; 565 } 566 if ((error = ipcperm(cred, &msqptr->msg_perm, 567 msgflg & 0700 ))) { 568 MSG_PRINTF(("requester doesn't have 0%o access\n", 569 msgflg & 0700)); 570 goto unlock; 571 } 572 goto found; 573 } 574 } 575 576 MSG_PRINTF(("need to allocate the msqid_ds\n")); 577 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 578 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 579 /* 580 * Look for an unallocated and unlocked msqid_ds. 581 * msqid_ds's can be locked by msgsnd or msgrcv while 582 * they are copying the message in/out. We can't 583 * re-use the entry until they release it. 584 */ 585 msq = &msqs[msqid]; 586 msqptr = &msq->msq_u; 587 if (msqptr->msg_qbytes == 0 && 588 (msqptr->msg_perm.mode & MSG_LOCKED) == 0) 589 break; 590 } 591 if (msqid == msginfo.msgmni) { 592 MSG_PRINTF(("no more msqid_ds's available\n")); 593 error = ENOSPC; 594 goto unlock; 595 } 596 MSG_PRINTF(("msqid %d is available\n", msqid)); 597 msqptr->msg_perm._key = key; 598 msqptr->msg_perm.cuid = kauth_cred_geteuid(cred); 599 msqptr->msg_perm.uid = kauth_cred_geteuid(cred); 600 msqptr->msg_perm.cgid = kauth_cred_getegid(cred); 601 msqptr->msg_perm.gid = kauth_cred_getegid(cred); 602 msqptr->msg_perm.mode = (msgflg & 0777); 603 /* Make sure that the returned msqid is unique */ 604 msqptr->msg_perm._seq++; 605 msqptr->_msg_first = NULL; 606 msqptr->_msg_last = NULL; 607 msqptr->_msg_cbytes = 0; 608 msqptr->msg_qnum = 0; 609 msqptr->msg_qbytes = msginfo.msgmnb; 610 msqptr->msg_lspid = 0; 611 msqptr->msg_lrpid = 0; 612 msqptr->msg_stime = 0; 613 msqptr->msg_rtime = 0; 614 msqptr->msg_ctime = time_second; 615 } else { 616 MSG_PRINTF(("didn't find it and wasn't asked to create it\n")); 617 error = ENOENT; 618 goto unlock; 619 } 620 621 found: 622 /* Construct the unique msqid */ 623 *retval = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm); 624 625 unlock: 626 mutex_exit(&msgmutex); 627 return (error); 628 } 629 630 int 631 sys_msgsnd(struct lwp *l, const struct sys_msgsnd_args *uap, register_t *retval) 632 { 633 /* { 634 syscallarg(int) msqid; 635 syscallarg(const void *) msgp; 636 syscallarg(size_t) msgsz; 637 syscallarg(int) msgflg; 638 } */ 639 640 return msgsnd1(l, SCARG(uap, msqid), SCARG(uap, msgp), 641 SCARG(uap, msgsz), SCARG(uap, msgflg), sizeof(long), copyin); 642 } 643 644 int 645 msgsnd1(struct lwp *l, int msqidr, const char *user_msgp, size_t msgsz, 646 int msgflg, size_t typesz, copyin_t fetch_type) 647 { 648 int segs_needed, error = 0, msqid; 649 kauth_cred_t cred = l->l_cred; 650 struct msqid_ds *msqptr; 651 struct __msg *msghdr; 652 kmsq_t *msq; 653 short next; 654 655 MSG_PRINTF(("call to msgsnd(%d, %p, %lld, %d)\n", msqid, user_msgp, 656 (long long)msgsz, msgflg)); 657 658 if ((ssize_t)msgsz < 0) 659 return EINVAL; 660 661 restart: 662 msqid = IPCID_TO_IX(msqidr); 663 664 mutex_enter(&msgmutex); 665 /* In case of reallocation, we will wait for completion */ 666 while (__predict_false(msg_realloc_state)) 667 cv_wait(&msg_realloc_cv, &msgmutex); 668 669 if (msqid < 0 || msqid >= msginfo.msgmni) { 670 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 671 msginfo.msgmni)); 672 error = EINVAL; 673 goto unlock; 674 } 675 676 msq = &msqs[msqid]; 677 msqptr = &msq->msq_u; 678 679 if (msqptr->msg_qbytes == 0) { 680 MSG_PRINTF(("no such message queue id\n")); 681 error = EINVAL; 682 goto unlock; 683 } 684 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 685 MSG_PRINTF(("wrong sequence number\n")); 686 error = EINVAL; 687 goto unlock; 688 } 689 690 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_W))) { 691 MSG_PRINTF(("requester doesn't have write access\n")); 692 goto unlock; 693 } 694 695 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 696 MSG_PRINTF(("msgsz=%lld, msgssz=%d, segs_needed=%d\n", 697 (long long)msgsz, msginfo.msgssz, segs_needed)); 698 for (;;) { 699 int need_more_resources = 0; 700 701 /* 702 * check msgsz [cannot be negative since it is unsigned] 703 * (inside this loop in case msg_qbytes changes while we sleep) 704 */ 705 706 if (msgsz > msqptr->msg_qbytes) { 707 MSG_PRINTF(("msgsz > msqptr->msg_qbytes\n")); 708 error = EINVAL; 709 goto unlock; 710 } 711 712 if (msqptr->msg_perm.mode & MSG_LOCKED) { 713 MSG_PRINTF(("msqid is locked\n")); 714 need_more_resources = 1; 715 } 716 if (msgsz + msqptr->_msg_cbytes > msqptr->msg_qbytes) { 717 MSG_PRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 718 need_more_resources = 1; 719 } 720 if (segs_needed > nfree_msgmaps) { 721 MSG_PRINTF(("segs_needed > nfree_msgmaps\n")); 722 need_more_resources = 1; 723 } 724 if (free_msghdrs == NULL) { 725 MSG_PRINTF(("no more msghdrs\n")); 726 need_more_resources = 1; 727 } 728 729 if (need_more_resources) { 730 int we_own_it; 731 732 if ((msgflg & IPC_NOWAIT) != 0) { 733 MSG_PRINTF(("need more resources but caller " 734 "doesn't want to wait\n")); 735 error = EAGAIN; 736 goto unlock; 737 } 738 739 if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { 740 MSG_PRINTF(("we don't own the msqid_ds\n")); 741 we_own_it = 0; 742 } else { 743 /* Force later arrivals to wait for our 744 request */ 745 MSG_PRINTF(("we own the msqid_ds\n")); 746 msqptr->msg_perm.mode |= MSG_LOCKED; 747 we_own_it = 1; 748 } 749 750 msg_waiters++; 751 MSG_PRINTF(("goodnight\n")); 752 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 753 MSG_PRINTF(("good morning, error=%d\n", error)); 754 msg_waiters--; 755 756 if (we_own_it) 757 msqptr->msg_perm.mode &= ~MSG_LOCKED; 758 759 /* 760 * In case of such state, notify reallocator and 761 * restart the call. 762 */ 763 if (msg_realloc_state) { 764 cv_broadcast(&msg_realloc_cv); 765 mutex_exit(&msgmutex); 766 goto restart; 767 } 768 769 if (error != 0) { 770 MSG_PRINTF(("msgsnd: interrupted system " 771 "call\n")); 772 error = EINTR; 773 goto unlock; 774 } 775 776 /* 777 * Make sure that the msq queue still exists 778 */ 779 780 if (msqptr->msg_qbytes == 0) { 781 MSG_PRINTF(("msqid deleted\n")); 782 error = EIDRM; 783 goto unlock; 784 } 785 } else { 786 MSG_PRINTF(("got all the resources that we need\n")); 787 break; 788 } 789 } 790 791 /* 792 * We have the resources that we need. 793 * Make sure! 794 */ 795 796 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 797 KASSERT(segs_needed <= nfree_msgmaps); 798 KASSERT(msgsz + msqptr->_msg_cbytes <= msqptr->msg_qbytes); 799 KASSERT(free_msghdrs != NULL); 800 801 /* 802 * Re-lock the msqid_ds in case we page-fault when copying in the 803 * message 804 */ 805 806 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 807 msqptr->msg_perm.mode |= MSG_LOCKED; 808 809 /* 810 * Allocate a message header 811 */ 812 813 msghdr = free_msghdrs; 814 free_msghdrs = msghdr->msg_next; 815 msghdr->msg_spot = -1; 816 msghdr->msg_ts = msgsz; 817 818 /* 819 * Allocate space for the message 820 */ 821 822 while (segs_needed > 0) { 823 KASSERT(nfree_msgmaps > 0); 824 KASSERT(free_msgmaps != -1); 825 KASSERT(free_msgmaps < msginfo.msgseg); 826 827 next = free_msgmaps; 828 MSG_PRINTF(("allocating segment %d to message\n", next)); 829 free_msgmaps = msgmaps[next].next; 830 nfree_msgmaps--; 831 msgmaps[next].next = msghdr->msg_spot; 832 msghdr->msg_spot = next; 833 segs_needed--; 834 } 835 836 /* 837 * Copy in the message type 838 */ 839 mutex_exit(&msgmutex); 840 error = (*fetch_type)(user_msgp, &msghdr->msg_type, typesz); 841 mutex_enter(&msgmutex); 842 if (error != 0) { 843 MSG_PRINTF(("error %d copying the message type\n", error)); 844 msg_freehdr(msghdr); 845 msqptr->msg_perm.mode &= ~MSG_LOCKED; 846 cv_broadcast(&msq->msq_cv); 847 goto unlock; 848 } 849 user_msgp += typesz; 850 851 /* 852 * Validate the message type 853 */ 854 855 if (msghdr->msg_type < 1) { 856 msg_freehdr(msghdr); 857 msqptr->msg_perm.mode &= ~MSG_LOCKED; 858 cv_broadcast(&msq->msq_cv); 859 MSG_PRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 860 error = EINVAL; 861 goto unlock; 862 } 863 864 /* 865 * Copy in the message body 866 */ 867 868 next = msghdr->msg_spot; 869 while (msgsz > 0) { 870 size_t tlen; 871 KASSERT(next > -1); 872 KASSERT(next < msginfo.msgseg); 873 874 if (msgsz > msginfo.msgssz) 875 tlen = msginfo.msgssz; 876 else 877 tlen = msgsz; 878 mutex_exit(&msgmutex); 879 error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen); 880 mutex_enter(&msgmutex); 881 if (error != 0) { 882 MSG_PRINTF(("error %d copying in message segment\n", 883 error)); 884 msg_freehdr(msghdr); 885 msqptr->msg_perm.mode &= ~MSG_LOCKED; 886 cv_broadcast(&msq->msq_cv); 887 goto unlock; 888 } 889 msgsz -= tlen; 890 user_msgp += tlen; 891 next = msgmaps[next].next; 892 } 893 KASSERT(next == -1); 894 895 /* 896 * We've got the message. Unlock the msqid_ds. 897 */ 898 899 msqptr->msg_perm.mode &= ~MSG_LOCKED; 900 901 /* 902 * Make sure that the msqid_ds is still allocated. 903 */ 904 905 if (msqptr->msg_qbytes == 0) { 906 msg_freehdr(msghdr); 907 cv_broadcast(&msq->msq_cv); 908 error = EIDRM; 909 goto unlock; 910 } 911 912 /* 913 * Put the message into the queue 914 */ 915 916 if (msqptr->_msg_first == NULL) { 917 msqptr->_msg_first = msghdr; 918 msqptr->_msg_last = msghdr; 919 } else { 920 msqptr->_msg_last->msg_next = msghdr; 921 msqptr->_msg_last = msghdr; 922 } 923 msqptr->_msg_last->msg_next = NULL; 924 925 msqptr->_msg_cbytes += msghdr->msg_ts; 926 msqptr->msg_qnum++; 927 msqptr->msg_lspid = l->l_proc->p_pid; 928 msqptr->msg_stime = time_second; 929 930 cv_broadcast(&msq->msq_cv); 931 932 unlock: 933 mutex_exit(&msgmutex); 934 return error; 935 } 936 937 int 938 sys_msgrcv(struct lwp *l, const struct sys_msgrcv_args *uap, register_t *retval) 939 { 940 /* { 941 syscallarg(int) msqid; 942 syscallarg(void *) msgp; 943 syscallarg(size_t) msgsz; 944 syscallarg(long) msgtyp; 945 syscallarg(int) msgflg; 946 } */ 947 948 return msgrcv1(l, SCARG(uap, msqid), SCARG(uap, msgp), 949 SCARG(uap, msgsz), SCARG(uap, msgtyp), SCARG(uap, msgflg), 950 sizeof(long), copyout, retval); 951 } 952 953 int 954 msgrcv1(struct lwp *l, int msqidr, char *user_msgp, size_t msgsz, long msgtyp, 955 int msgflg, size_t typesz, copyout_t put_type, register_t *retval) 956 { 957 size_t len; 958 kauth_cred_t cred = l->l_cred; 959 struct msqid_ds *msqptr; 960 struct __msg *msghdr; 961 int error = 0, msqid; 962 kmsq_t *msq; 963 short next; 964 965 MSG_PRINTF(("call to msgrcv(%d, %p, %lld, %ld, %d)\n", msqid, 966 user_msgp, (long long)msgsz, msgtyp, msgflg)); 967 968 if ((ssize_t)msgsz < 0) 969 return EINVAL; 970 971 restart: 972 msqid = IPCID_TO_IX(msqidr); 973 974 mutex_enter(&msgmutex); 975 /* In case of reallocation, we will wait for completion */ 976 while (__predict_false(msg_realloc_state)) 977 cv_wait(&msg_realloc_cv, &msgmutex); 978 979 if (msqid < 0 || msqid >= msginfo.msgmni) { 980 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 981 msginfo.msgmni)); 982 error = EINVAL; 983 goto unlock; 984 } 985 986 msq = &msqs[msqid]; 987 msqptr = &msq->msq_u; 988 989 if (msqptr->msg_qbytes == 0) { 990 MSG_PRINTF(("no such message queue id\n")); 991 error = EINVAL; 992 goto unlock; 993 } 994 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 995 MSG_PRINTF(("wrong sequence number\n")); 996 error = EINVAL; 997 goto unlock; 998 } 999 1000 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 1001 MSG_PRINTF(("requester doesn't have read access\n")); 1002 goto unlock; 1003 } 1004 1005 msghdr = NULL; 1006 while (msghdr == NULL) { 1007 if (msgtyp == 0) { 1008 msghdr = msqptr->_msg_first; 1009 if (msghdr != NULL) { 1010 if (msgsz < msghdr->msg_ts && 1011 (msgflg & MSG_NOERROR) == 0) { 1012 MSG_PRINTF(("first msg on the queue " 1013 "is too big (want %lld, got %d)\n", 1014 (long long)msgsz, msghdr->msg_ts)); 1015 error = E2BIG; 1016 goto unlock; 1017 } 1018 if (msqptr->_msg_first == msqptr->_msg_last) { 1019 msqptr->_msg_first = NULL; 1020 msqptr->_msg_last = NULL; 1021 } else { 1022 msqptr->_msg_first = msghdr->msg_next; 1023 KASSERT(msqptr->_msg_first != NULL); 1024 } 1025 } 1026 } else { 1027 struct __msg *previous; 1028 struct __msg **prev; 1029 1030 for (previous = NULL, prev = &msqptr->_msg_first; 1031 (msghdr = *prev) != NULL; 1032 previous = msghdr, prev = &msghdr->msg_next) { 1033 /* 1034 * Is this message's type an exact match or is 1035 * this message's type less than or equal to 1036 * the absolute value of a negative msgtyp? 1037 * Note that the second half of this test can 1038 * NEVER be true if msgtyp is positive since 1039 * msg_type is always positive! 1040 */ 1041 1042 if (msgtyp != msghdr->msg_type && 1043 msghdr->msg_type > -msgtyp) 1044 continue; 1045 1046 MSG_PRINTF(("found message type %ld, requested %ld\n", 1047 msghdr->msg_type, msgtyp)); 1048 if (msgsz < msghdr->msg_ts && 1049 (msgflg & MSG_NOERROR) == 0) { 1050 MSG_PRINTF(("requested message on the queue " 1051 "is too big (want %lld, got %d)\n", 1052 (long long)msgsz, msghdr->msg_ts)); 1053 error = E2BIG; 1054 goto unlock; 1055 } 1056 *prev = msghdr->msg_next; 1057 if (msghdr != msqptr->_msg_last) 1058 break; 1059 if (previous == NULL) { 1060 KASSERT(prev == &msqptr->_msg_first); 1061 msqptr->_msg_first = NULL; 1062 msqptr->_msg_last = NULL; 1063 } else { 1064 KASSERT(prev != &msqptr->_msg_first); 1065 msqptr->_msg_last = previous; 1066 } 1067 break; 1068 } 1069 } 1070 1071 /* 1072 * We've either extracted the msghdr for the appropriate 1073 * message or there isn't one. 1074 * If there is one then bail out of this loop. 1075 */ 1076 if (msghdr != NULL) 1077 break; 1078 1079 /* 1080 * Hmph! No message found. Does the user want to wait? 1081 */ 1082 1083 if ((msgflg & IPC_NOWAIT) != 0) { 1084 MSG_PRINTF(("no appropriate message found (msgtyp=%ld)\n", 1085 msgtyp)); 1086 error = ENOMSG; 1087 goto unlock; 1088 } 1089 1090 /* 1091 * Wait for something to happen 1092 */ 1093 1094 msg_waiters++; 1095 MSG_PRINTF(("msgrcv: goodnight\n")); 1096 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 1097 MSG_PRINTF(("msgrcv: good morning (error=%d)\n", error)); 1098 msg_waiters--; 1099 1100 /* 1101 * In case of such state, notify reallocator and 1102 * restart the call. 1103 */ 1104 if (msg_realloc_state) { 1105 cv_broadcast(&msg_realloc_cv); 1106 mutex_exit(&msgmutex); 1107 goto restart; 1108 } 1109 1110 if (error != 0) { 1111 MSG_PRINTF(("msgsnd: interrupted system call\n")); 1112 error = EINTR; 1113 goto unlock; 1114 } 1115 1116 /* 1117 * Make sure that the msq queue still exists 1118 */ 1119 1120 if (msqptr->msg_qbytes == 0 || 1121 msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 1122 MSG_PRINTF(("msqid deleted\n")); 1123 error = EIDRM; 1124 goto unlock; 1125 } 1126 } 1127 1128 /* 1129 * Return the message to the user. 1130 * 1131 * First, do the bookkeeping (before we risk being interrupted). 1132 */ 1133 1134 msqptr->_msg_cbytes -= msghdr->msg_ts; 1135 msqptr->msg_qnum--; 1136 msqptr->msg_lrpid = l->l_proc->p_pid; 1137 msqptr->msg_rtime = time_second; 1138 1139 /* 1140 * Make msgsz the actual amount that we'll be returning. 1141 * Note that this effectively truncates the message if it is too long 1142 * (since msgsz is never increased). 1143 */ 1144 1145 MSG_PRINTF(("found a message, msgsz=%lld, msg_ts=%d\n", 1146 (long long)msgsz, msghdr->msg_ts)); 1147 if (msgsz > msghdr->msg_ts) 1148 msgsz = msghdr->msg_ts; 1149 1150 /* 1151 * Return the type to the user. 1152 */ 1153 mutex_exit(&msgmutex); 1154 error = (*put_type)(&msghdr->msg_type, user_msgp, typesz); 1155 mutex_enter(&msgmutex); 1156 if (error != 0) { 1157 MSG_PRINTF(("error (%d) copying out message type\n", error)); 1158 msg_freehdr(msghdr); 1159 cv_broadcast(&msq->msq_cv); 1160 goto unlock; 1161 } 1162 user_msgp += typesz; 1163 1164 /* 1165 * Return the segments to the user 1166 */ 1167 1168 next = msghdr->msg_spot; 1169 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1170 size_t tlen; 1171 KASSERT(next > -1); 1172 KASSERT(next < msginfo.msgseg); 1173 1174 if (msgsz - len > msginfo.msgssz) 1175 tlen = msginfo.msgssz; 1176 else 1177 tlen = msgsz - len; 1178 mutex_exit(&msgmutex); 1179 error = copyout(&msgpool[next * msginfo.msgssz], 1180 user_msgp, tlen); 1181 mutex_enter(&msgmutex); 1182 if (error != 0) { 1183 MSG_PRINTF(("error (%d) copying out message segment\n", 1184 error)); 1185 msg_freehdr(msghdr); 1186 cv_broadcast(&msq->msq_cv); 1187 goto unlock; 1188 } 1189 user_msgp += tlen; 1190 next = msgmaps[next].next; 1191 } 1192 1193 /* 1194 * Done, return the actual number of bytes copied out. 1195 */ 1196 1197 msg_freehdr(msghdr); 1198 cv_broadcast(&msq->msq_cv); 1199 *retval = msgsz; 1200 1201 unlock: 1202 mutex_exit(&msgmutex); 1203 return error; 1204 } 1205 1206 /* 1207 * Sysctl initialization and nodes. 1208 */ 1209 1210 static int 1211 sysctl_ipc_msgmni(SYSCTLFN_ARGS) 1212 { 1213 int newsize, error; 1214 struct sysctlnode node; 1215 node = *rnode; 1216 node.sysctl_data = &newsize; 1217 1218 newsize = msginfo.msgmni; 1219 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1220 if (error || newp == NULL) 1221 return error; 1222 1223 sysctl_unlock(); 1224 error = msgrealloc(newsize, msginfo.msgseg); 1225 sysctl_relock(); 1226 return error; 1227 } 1228 1229 static int 1230 sysctl_ipc_msgseg(SYSCTLFN_ARGS) 1231 { 1232 int newsize, error; 1233 struct sysctlnode node; 1234 node = *rnode; 1235 node.sysctl_data = &newsize; 1236 1237 newsize = msginfo.msgseg; 1238 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1239 if (error || newp == NULL) 1240 return error; 1241 1242 sysctl_unlock(); 1243 error = msgrealloc(msginfo.msgmni, newsize); 1244 sysctl_relock(); 1245 return error; 1246 } 1247 1248 SYSCTL_SETUP(sysctl_ipc_msg_setup, "sysctl kern.ipc subtree setup") 1249 { 1250 const struct sysctlnode *node = NULL; 1251 1252 sysctl_createv(clog, 0, NULL, NULL, 1253 CTLFLAG_PERMANENT, 1254 CTLTYPE_NODE, "kern", NULL, 1255 NULL, 0, NULL, 0, 1256 CTL_KERN, CTL_EOL); 1257 sysctl_createv(clog, 0, NULL, &node, 1258 CTLFLAG_PERMANENT, 1259 CTLTYPE_NODE, "ipc", 1260 SYSCTL_DESCR("SysV IPC options"), 1261 NULL, 0, NULL, 0, 1262 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1263 1264 if (node == NULL) 1265 return; 1266 1267 sysctl_createv(clog, 0, &node, NULL, 1268 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1269 CTLTYPE_INT, "msgmni", 1270 SYSCTL_DESCR("Max number of message queue identifiers"), 1271 sysctl_ipc_msgmni, 0, &msginfo.msgmni, 0, 1272 CTL_CREATE, CTL_EOL); 1273 sysctl_createv(clog, 0, &node, NULL, 1274 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1275 CTLTYPE_INT, "msgseg", 1276 SYSCTL_DESCR("Max number of number of message segments"), 1277 sysctl_ipc_msgseg, 0, &msginfo.msgseg, 0, 1278 CTL_CREATE, CTL_EOL); 1279 } 1280