1 /* 2 * linux/ipc/msg.c 3 * Copyright (C) 1992 Krishna Balasubramanian 4 * 5 * Removed all the remaining kerneld mess 6 * Catch the -EFAULT stuff properly 7 * Use GFP_KERNEL for messages as in 1.2 8 * Fixed up the unchecked user space derefs 9 * Copyright (C) 1998 Alan Cox & Andi Kleen 10 * 11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com> 12 * 13 * mostly rewritten, threaded and wake-one semantics added 14 * MSGMAX limit removed, sysctl's added 15 * (c) 1999 Manfred Spraul <manfred@colorfullife.com> 16 */ 17 18 #include <linux/capability.h> 19 #include <linux/config.h> 20 #include <linux/slab.h> 21 #include <linux/msg.h> 22 #include <linux/spinlock.h> 23 #include <linux/init.h> 24 #include <linux/proc_fs.h> 25 #include <linux/list.h> 26 #include <linux/security.h> 27 #include <linux/sched.h> 28 #include <linux/syscalls.h> 29 #include <linux/audit.h> 30 #include <linux/seq_file.h> 31 #include <linux/mutex.h> 32 33 #include <asm/current.h> 34 #include <asm/uaccess.h> 35 #include "util.h" 36 37 /* sysctl: */ 38 int msg_ctlmax = MSGMAX; 39 int msg_ctlmnb = MSGMNB; 40 int msg_ctlmni = MSGMNI; 41 42 /* one msg_receiver structure for each sleeping receiver */ 43 struct msg_receiver { 44 struct list_head r_list; 45 struct task_struct* r_tsk; 46 47 int r_mode; 48 long r_msgtype; 49 long r_maxsize; 50 51 struct msg_msg* volatile r_msg; 52 }; 53 54 /* one msg_sender for each sleeping sender */ 55 struct msg_sender { 56 struct list_head list; 57 struct task_struct* tsk; 58 }; 59 60 #define SEARCH_ANY 1 61 #define SEARCH_EQUAL 2 62 #define SEARCH_NOTEQUAL 3 63 #define SEARCH_LESSEQUAL 4 64 65 static atomic_t msg_bytes = ATOMIC_INIT(0); 66 static atomic_t msg_hdrs = ATOMIC_INIT(0); 67 68 static struct ipc_ids msg_ids; 69 70 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id)) 71 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm) 72 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id)) 73 #define msg_checkid(msq, msgid) \ 74 ipc_checkid(&msg_ids,&msq->q_perm,msgid) 75 #define msg_buildid(id, seq) \ 76 ipc_buildid(&msg_ids, id, seq) 77 78 static void freeque (struct msg_queue *msq, int id); 79 static int newque (key_t key, int msgflg); 80 #ifdef CONFIG_PROC_FS 81 static int sysvipc_msg_proc_show(struct seq_file *s, void *it); 82 #endif 83 84 void __init msg_init (void) 85 { 86 ipc_init_ids(&msg_ids,msg_ctlmni); 87 ipc_init_proc_interface("sysvipc/msg", 88 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n", 89 &msg_ids, 90 sysvipc_msg_proc_show); 91 } 92 93 static int newque (key_t key, int msgflg) 94 { 95 int id; 96 int retval; 97 struct msg_queue *msq; 98 99 msq = ipc_rcu_alloc(sizeof(*msq)); 100 if (!msq) 101 return -ENOMEM; 102 103 msq->q_perm.mode = (msgflg & S_IRWXUGO); 104 msq->q_perm.key = key; 105 106 msq->q_perm.security = NULL; 107 retval = security_msg_queue_alloc(msq); 108 if (retval) { 109 ipc_rcu_putref(msq); 110 return retval; 111 } 112 113 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni); 114 if(id == -1) { 115 security_msg_queue_free(msq); 116 ipc_rcu_putref(msq); 117 return -ENOSPC; 118 } 119 120 msq->q_id = msg_buildid(id,msq->q_perm.seq); 121 msq->q_stime = msq->q_rtime = 0; 122 msq->q_ctime = get_seconds(); 123 msq->q_cbytes = msq->q_qnum = 0; 124 msq->q_qbytes = msg_ctlmnb; 125 msq->q_lspid = msq->q_lrpid = 0; 126 INIT_LIST_HEAD(&msq->q_messages); 127 INIT_LIST_HEAD(&msq->q_receivers); 128 INIT_LIST_HEAD(&msq->q_senders); 129 msg_unlock(msq); 130 131 return msq->q_id; 132 } 133 134 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss) 135 { 136 mss->tsk=current; 137 current->state=TASK_INTERRUPTIBLE; 138 list_add_tail(&mss->list,&msq->q_senders); 139 } 140 141 static inline void ss_del(struct msg_sender* mss) 142 { 143 if(mss->list.next != NULL) 144 list_del(&mss->list); 145 } 146 147 static void ss_wakeup(struct list_head* h, int kill) 148 { 149 struct list_head *tmp; 150 151 tmp = h->next; 152 while (tmp != h) { 153 struct msg_sender* mss; 154 155 mss = list_entry(tmp,struct msg_sender,list); 156 tmp = tmp->next; 157 if(kill) 158 mss->list.next=NULL; 159 wake_up_process(mss->tsk); 160 } 161 } 162 163 static void expunge_all(struct msg_queue* msq, int res) 164 { 165 struct list_head *tmp; 166 167 tmp = msq->q_receivers.next; 168 while (tmp != &msq->q_receivers) { 169 struct msg_receiver* msr; 170 171 msr = list_entry(tmp,struct msg_receiver,r_list); 172 tmp = tmp->next; 173 msr->r_msg = NULL; 174 wake_up_process(msr->r_tsk); 175 smp_mb(); 176 msr->r_msg = ERR_PTR(res); 177 } 178 } 179 /* 180 * freeque() wakes up waiters on the sender and receiver waiting queue, 181 * removes the message queue from message queue ID 182 * array, and cleans up all the messages associated with this queue. 183 * 184 * msg_ids.mutex and the spinlock for this message queue is hold 185 * before freeque() is called. msg_ids.mutex remains locked on exit. 186 */ 187 static void freeque (struct msg_queue *msq, int id) 188 { 189 struct list_head *tmp; 190 191 expunge_all(msq,-EIDRM); 192 ss_wakeup(&msq->q_senders,1); 193 msq = msg_rmid(id); 194 msg_unlock(msq); 195 196 tmp = msq->q_messages.next; 197 while(tmp != &msq->q_messages) { 198 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list); 199 tmp = tmp->next; 200 atomic_dec(&msg_hdrs); 201 free_msg(msg); 202 } 203 atomic_sub(msq->q_cbytes, &msg_bytes); 204 security_msg_queue_free(msq); 205 ipc_rcu_putref(msq); 206 } 207 208 asmlinkage long sys_msgget (key_t key, int msgflg) 209 { 210 int id, ret = -EPERM; 211 struct msg_queue *msq; 212 213 mutex_lock(&msg_ids.mutex); 214 if (key == IPC_PRIVATE) 215 ret = newque(key, msgflg); 216 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */ 217 if (!(msgflg & IPC_CREAT)) 218 ret = -ENOENT; 219 else 220 ret = newque(key, msgflg); 221 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) { 222 ret = -EEXIST; 223 } else { 224 msq = msg_lock(id); 225 BUG_ON(msq==NULL); 226 if (ipcperms(&msq->q_perm, msgflg)) 227 ret = -EACCES; 228 else { 229 int qid = msg_buildid(id, msq->q_perm.seq); 230 ret = security_msg_queue_associate(msq, msgflg); 231 if (!ret) 232 ret = qid; 233 } 234 msg_unlock(msq); 235 } 236 mutex_unlock(&msg_ids.mutex); 237 return ret; 238 } 239 240 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version) 241 { 242 switch(version) { 243 case IPC_64: 244 return copy_to_user (buf, in, sizeof(*in)); 245 case IPC_OLD: 246 { 247 struct msqid_ds out; 248 249 memset(&out,0,sizeof(out)); 250 251 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm); 252 253 out.msg_stime = in->msg_stime; 254 out.msg_rtime = in->msg_rtime; 255 out.msg_ctime = in->msg_ctime; 256 257 if(in->msg_cbytes > USHRT_MAX) 258 out.msg_cbytes = USHRT_MAX; 259 else 260 out.msg_cbytes = in->msg_cbytes; 261 out.msg_lcbytes = in->msg_cbytes; 262 263 if(in->msg_qnum > USHRT_MAX) 264 out.msg_qnum = USHRT_MAX; 265 else 266 out.msg_qnum = in->msg_qnum; 267 268 if(in->msg_qbytes > USHRT_MAX) 269 out.msg_qbytes = USHRT_MAX; 270 else 271 out.msg_qbytes = in->msg_qbytes; 272 out.msg_lqbytes = in->msg_qbytes; 273 274 out.msg_lspid = in->msg_lspid; 275 out.msg_lrpid = in->msg_lrpid; 276 277 return copy_to_user (buf, &out, sizeof(out)); 278 } 279 default: 280 return -EINVAL; 281 } 282 } 283 284 struct msq_setbuf { 285 unsigned long qbytes; 286 uid_t uid; 287 gid_t gid; 288 mode_t mode; 289 }; 290 291 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version) 292 { 293 switch(version) { 294 case IPC_64: 295 { 296 struct msqid64_ds tbuf; 297 298 if (copy_from_user (&tbuf, buf, sizeof (tbuf))) 299 return -EFAULT; 300 301 out->qbytes = tbuf.msg_qbytes; 302 out->uid = tbuf.msg_perm.uid; 303 out->gid = tbuf.msg_perm.gid; 304 out->mode = tbuf.msg_perm.mode; 305 306 return 0; 307 } 308 case IPC_OLD: 309 { 310 struct msqid_ds tbuf_old; 311 312 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old))) 313 return -EFAULT; 314 315 out->uid = tbuf_old.msg_perm.uid; 316 out->gid = tbuf_old.msg_perm.gid; 317 out->mode = tbuf_old.msg_perm.mode; 318 319 if(tbuf_old.msg_qbytes == 0) 320 out->qbytes = tbuf_old.msg_lqbytes; 321 else 322 out->qbytes = tbuf_old.msg_qbytes; 323 324 return 0; 325 } 326 default: 327 return -EINVAL; 328 } 329 } 330 331 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf) 332 { 333 int err, version; 334 struct msg_queue *msq; 335 struct msq_setbuf setbuf; 336 struct kern_ipc_perm *ipcp; 337 338 if (msqid < 0 || cmd < 0) 339 return -EINVAL; 340 341 version = ipc_parse_version(&cmd); 342 343 switch (cmd) { 344 case IPC_INFO: 345 case MSG_INFO: 346 { 347 struct msginfo msginfo; 348 int max_id; 349 if (!buf) 350 return -EFAULT; 351 /* We must not return kernel stack data. 352 * due to padding, it's not enough 353 * to set all member fields. 354 */ 355 356 err = security_msg_queue_msgctl(NULL, cmd); 357 if (err) 358 return err; 359 360 memset(&msginfo,0,sizeof(msginfo)); 361 msginfo.msgmni = msg_ctlmni; 362 msginfo.msgmax = msg_ctlmax; 363 msginfo.msgmnb = msg_ctlmnb; 364 msginfo.msgssz = MSGSSZ; 365 msginfo.msgseg = MSGSEG; 366 mutex_lock(&msg_ids.mutex); 367 if (cmd == MSG_INFO) { 368 msginfo.msgpool = msg_ids.in_use; 369 msginfo.msgmap = atomic_read(&msg_hdrs); 370 msginfo.msgtql = atomic_read(&msg_bytes); 371 } else { 372 msginfo.msgmap = MSGMAP; 373 msginfo.msgpool = MSGPOOL; 374 msginfo.msgtql = MSGTQL; 375 } 376 max_id = msg_ids.max_id; 377 mutex_unlock(&msg_ids.mutex); 378 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo))) 379 return -EFAULT; 380 return (max_id < 0) ? 0: max_id; 381 } 382 case MSG_STAT: 383 case IPC_STAT: 384 { 385 struct msqid64_ds tbuf; 386 int success_return; 387 if (!buf) 388 return -EFAULT; 389 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size) 390 return -EINVAL; 391 392 memset(&tbuf,0,sizeof(tbuf)); 393 394 msq = msg_lock(msqid); 395 if (msq == NULL) 396 return -EINVAL; 397 398 if(cmd == MSG_STAT) { 399 success_return = msg_buildid(msqid, msq->q_perm.seq); 400 } else { 401 err = -EIDRM; 402 if (msg_checkid(msq,msqid)) 403 goto out_unlock; 404 success_return = 0; 405 } 406 err = -EACCES; 407 if (ipcperms (&msq->q_perm, S_IRUGO)) 408 goto out_unlock; 409 410 err = security_msg_queue_msgctl(msq, cmd); 411 if (err) 412 goto out_unlock; 413 414 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm); 415 tbuf.msg_stime = msq->q_stime; 416 tbuf.msg_rtime = msq->q_rtime; 417 tbuf.msg_ctime = msq->q_ctime; 418 tbuf.msg_cbytes = msq->q_cbytes; 419 tbuf.msg_qnum = msq->q_qnum; 420 tbuf.msg_qbytes = msq->q_qbytes; 421 tbuf.msg_lspid = msq->q_lspid; 422 tbuf.msg_lrpid = msq->q_lrpid; 423 msg_unlock(msq); 424 if (copy_msqid_to_user(buf, &tbuf, version)) 425 return -EFAULT; 426 return success_return; 427 } 428 case IPC_SET: 429 if (!buf) 430 return -EFAULT; 431 if (copy_msqid_from_user (&setbuf, buf, version)) 432 return -EFAULT; 433 break; 434 case IPC_RMID: 435 break; 436 default: 437 return -EINVAL; 438 } 439 440 mutex_lock(&msg_ids.mutex); 441 msq = msg_lock(msqid); 442 err=-EINVAL; 443 if (msq == NULL) 444 goto out_up; 445 446 err = -EIDRM; 447 if (msg_checkid(msq,msqid)) 448 goto out_unlock_up; 449 ipcp = &msq->q_perm; 450 err = -EPERM; 451 if (current->euid != ipcp->cuid && 452 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) 453 /* We _could_ check for CAP_CHOWN above, but we don't */ 454 goto out_unlock_up; 455 456 err = security_msg_queue_msgctl(msq, cmd); 457 if (err) 458 goto out_unlock_up; 459 460 switch (cmd) { 461 case IPC_SET: 462 { 463 if ((err = audit_ipc_perms(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode, ipcp))) 464 goto out_unlock_up; 465 466 err = -EPERM; 467 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE)) 468 goto out_unlock_up; 469 470 msq->q_qbytes = setbuf.qbytes; 471 472 ipcp->uid = setbuf.uid; 473 ipcp->gid = setbuf.gid; 474 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) | 475 (S_IRWXUGO & setbuf.mode); 476 msq->q_ctime = get_seconds(); 477 /* sleeping receivers might be excluded by 478 * stricter permissions. 479 */ 480 expunge_all(msq,-EAGAIN); 481 /* sleeping senders might be able to send 482 * due to a larger queue size. 483 */ 484 ss_wakeup(&msq->q_senders,0); 485 msg_unlock(msq); 486 break; 487 } 488 case IPC_RMID: 489 freeque (msq, msqid); 490 break; 491 } 492 err = 0; 493 out_up: 494 mutex_unlock(&msg_ids.mutex); 495 return err; 496 out_unlock_up: 497 msg_unlock(msq); 498 goto out_up; 499 out_unlock: 500 msg_unlock(msq); 501 return err; 502 } 503 504 static int testmsg(struct msg_msg* msg,long type,int mode) 505 { 506 switch(mode) 507 { 508 case SEARCH_ANY: 509 return 1; 510 case SEARCH_LESSEQUAL: 511 if(msg->m_type <=type) 512 return 1; 513 break; 514 case SEARCH_EQUAL: 515 if(msg->m_type == type) 516 return 1; 517 break; 518 case SEARCH_NOTEQUAL: 519 if(msg->m_type != type) 520 return 1; 521 break; 522 } 523 return 0; 524 } 525 526 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg) 527 { 528 struct list_head* tmp; 529 530 tmp = msq->q_receivers.next; 531 while (tmp != &msq->q_receivers) { 532 struct msg_receiver* msr; 533 msr = list_entry(tmp,struct msg_receiver,r_list); 534 tmp = tmp->next; 535 if(testmsg(msg,msr->r_msgtype,msr->r_mode) && 536 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) { 537 list_del(&msr->r_list); 538 if(msr->r_maxsize < msg->m_ts) { 539 msr->r_msg = NULL; 540 wake_up_process(msr->r_tsk); 541 smp_mb(); 542 msr->r_msg = ERR_PTR(-E2BIG); 543 } else { 544 msr->r_msg = NULL; 545 msq->q_lrpid = msr->r_tsk->pid; 546 msq->q_rtime = get_seconds(); 547 wake_up_process(msr->r_tsk); 548 smp_mb(); 549 msr->r_msg = msg; 550 return 1; 551 } 552 } 553 } 554 return 0; 555 } 556 557 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg) 558 { 559 struct msg_queue *msq; 560 struct msg_msg *msg; 561 long mtype; 562 int err; 563 564 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0) 565 return -EINVAL; 566 if (get_user(mtype, &msgp->mtype)) 567 return -EFAULT; 568 if (mtype < 1) 569 return -EINVAL; 570 571 msg = load_msg(msgp->mtext, msgsz); 572 if(IS_ERR(msg)) 573 return PTR_ERR(msg); 574 575 msg->m_type = mtype; 576 msg->m_ts = msgsz; 577 578 msq = msg_lock(msqid); 579 err=-EINVAL; 580 if(msq==NULL) 581 goto out_free; 582 583 err= -EIDRM; 584 if (msg_checkid(msq,msqid)) 585 goto out_unlock_free; 586 587 for (;;) { 588 struct msg_sender s; 589 590 err=-EACCES; 591 if (ipcperms(&msq->q_perm, S_IWUGO)) 592 goto out_unlock_free; 593 594 err = security_msg_queue_msgsnd(msq, msg, msgflg); 595 if (err) 596 goto out_unlock_free; 597 598 if(msgsz + msq->q_cbytes <= msq->q_qbytes && 599 1 + msq->q_qnum <= msq->q_qbytes) { 600 break; 601 } 602 603 /* queue full, wait: */ 604 if(msgflg&IPC_NOWAIT) { 605 err=-EAGAIN; 606 goto out_unlock_free; 607 } 608 ss_add(msq, &s); 609 ipc_rcu_getref(msq); 610 msg_unlock(msq); 611 schedule(); 612 613 ipc_lock_by_ptr(&msq->q_perm); 614 ipc_rcu_putref(msq); 615 if (msq->q_perm.deleted) { 616 err = -EIDRM; 617 goto out_unlock_free; 618 } 619 ss_del(&s); 620 621 if (signal_pending(current)) { 622 err=-ERESTARTNOHAND; 623 goto out_unlock_free; 624 } 625 } 626 627 msq->q_lspid = current->tgid; 628 msq->q_stime = get_seconds(); 629 630 if(!pipelined_send(msq,msg)) { 631 /* noone is waiting for this message, enqueue it */ 632 list_add_tail(&msg->m_list,&msq->q_messages); 633 msq->q_cbytes += msgsz; 634 msq->q_qnum++; 635 atomic_add(msgsz,&msg_bytes); 636 atomic_inc(&msg_hdrs); 637 } 638 639 err = 0; 640 msg = NULL; 641 642 out_unlock_free: 643 msg_unlock(msq); 644 out_free: 645 if(msg!=NULL) 646 free_msg(msg); 647 return err; 648 } 649 650 static inline int convert_mode(long* msgtyp, int msgflg) 651 { 652 /* 653 * find message of correct type. 654 * msgtyp = 0 => get first. 655 * msgtyp > 0 => get first message of matching type. 656 * msgtyp < 0 => get message with least type must be < abs(msgtype). 657 */ 658 if(*msgtyp==0) 659 return SEARCH_ANY; 660 if(*msgtyp<0) { 661 *msgtyp=-(*msgtyp); 662 return SEARCH_LESSEQUAL; 663 } 664 if(msgflg & MSG_EXCEPT) 665 return SEARCH_NOTEQUAL; 666 return SEARCH_EQUAL; 667 } 668 669 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz, 670 long msgtyp, int msgflg) 671 { 672 struct msg_queue *msq; 673 struct msg_msg *msg; 674 int mode; 675 676 if (msqid < 0 || (long) msgsz < 0) 677 return -EINVAL; 678 mode = convert_mode(&msgtyp,msgflg); 679 680 msq = msg_lock(msqid); 681 if(msq==NULL) 682 return -EINVAL; 683 684 msg = ERR_PTR(-EIDRM); 685 if (msg_checkid(msq,msqid)) 686 goto out_unlock; 687 688 for (;;) { 689 struct msg_receiver msr_d; 690 struct list_head* tmp; 691 692 msg = ERR_PTR(-EACCES); 693 if (ipcperms (&msq->q_perm, S_IRUGO)) 694 goto out_unlock; 695 696 msg = ERR_PTR(-EAGAIN); 697 tmp = msq->q_messages.next; 698 while (tmp != &msq->q_messages) { 699 struct msg_msg *walk_msg; 700 walk_msg = list_entry(tmp,struct msg_msg,m_list); 701 if(testmsg(walk_msg,msgtyp,mode) && 702 !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) { 703 msg = walk_msg; 704 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) { 705 msg=walk_msg; 706 msgtyp=walk_msg->m_type-1; 707 } else { 708 msg=walk_msg; 709 break; 710 } 711 } 712 tmp = tmp->next; 713 } 714 if(!IS_ERR(msg)) { 715 /* Found a suitable message. Unlink it from the queue. */ 716 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) { 717 msg = ERR_PTR(-E2BIG); 718 goto out_unlock; 719 } 720 list_del(&msg->m_list); 721 msq->q_qnum--; 722 msq->q_rtime = get_seconds(); 723 msq->q_lrpid = current->tgid; 724 msq->q_cbytes -= msg->m_ts; 725 atomic_sub(msg->m_ts,&msg_bytes); 726 atomic_dec(&msg_hdrs); 727 ss_wakeup(&msq->q_senders,0); 728 msg_unlock(msq); 729 break; 730 } 731 /* No message waiting. Wait for a message */ 732 if (msgflg & IPC_NOWAIT) { 733 msg = ERR_PTR(-ENOMSG); 734 goto out_unlock; 735 } 736 list_add_tail(&msr_d.r_list,&msq->q_receivers); 737 msr_d.r_tsk = current; 738 msr_d.r_msgtype = msgtyp; 739 msr_d.r_mode = mode; 740 if(msgflg & MSG_NOERROR) 741 msr_d.r_maxsize = INT_MAX; 742 else 743 msr_d.r_maxsize = msgsz; 744 msr_d.r_msg = ERR_PTR(-EAGAIN); 745 current->state = TASK_INTERRUPTIBLE; 746 msg_unlock(msq); 747 748 schedule(); 749 750 /* Lockless receive, part 1: 751 * Disable preemption. We don't hold a reference to the queue 752 * and getting a reference would defeat the idea of a lockless 753 * operation, thus the code relies on rcu to guarantee the 754 * existance of msq: 755 * Prior to destruction, expunge_all(-EIRDM) changes r_msg. 756 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed. 757 * rcu_read_lock() prevents preemption between reading r_msg 758 * and the spin_lock() inside ipc_lock_by_ptr(). 759 */ 760 rcu_read_lock(); 761 762 /* Lockless receive, part 2: 763 * Wait until pipelined_send or expunge_all are outside of 764 * wake_up_process(). There is a race with exit(), see 765 * ipc/mqueue.c for the details. 766 */ 767 msg = (struct msg_msg*) msr_d.r_msg; 768 while (msg == NULL) { 769 cpu_relax(); 770 msg = (struct msg_msg*) msr_d.r_msg; 771 } 772 773 /* Lockless receive, part 3: 774 * If there is a message or an error then accept it without 775 * locking. 776 */ 777 if(msg != ERR_PTR(-EAGAIN)) { 778 rcu_read_unlock(); 779 break; 780 } 781 782 /* Lockless receive, part 3: 783 * Acquire the queue spinlock. 784 */ 785 ipc_lock_by_ptr(&msq->q_perm); 786 rcu_read_unlock(); 787 788 /* Lockless receive, part 4: 789 * Repeat test after acquiring the spinlock. 790 */ 791 msg = (struct msg_msg*)msr_d.r_msg; 792 if(msg != ERR_PTR(-EAGAIN)) 793 goto out_unlock; 794 795 list_del(&msr_d.r_list); 796 if (signal_pending(current)) { 797 msg = ERR_PTR(-ERESTARTNOHAND); 798 out_unlock: 799 msg_unlock(msq); 800 break; 801 } 802 } 803 if (IS_ERR(msg)) 804 return PTR_ERR(msg); 805 806 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz; 807 if (put_user (msg->m_type, &msgp->mtype) || 808 store_msg(msgp->mtext, msg, msgsz)) { 809 msgsz = -EFAULT; 810 } 811 free_msg(msg); 812 return msgsz; 813 } 814 815 #ifdef CONFIG_PROC_FS 816 static int sysvipc_msg_proc_show(struct seq_file *s, void *it) 817 { 818 struct msg_queue *msq = it; 819 820 return seq_printf(s, 821 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n", 822 msq->q_perm.key, 823 msq->q_id, 824 msq->q_perm.mode, 825 msq->q_cbytes, 826 msq->q_qnum, 827 msq->q_lspid, 828 msq->q_lrpid, 829 msq->q_perm.uid, 830 msq->q_perm.gid, 831 msq->q_perm.cuid, 832 msq->q_perm.cgid, 833 msq->q_stime, 834 msq->q_rtime, 835 msq->q_ctime); 836 } 837 #endif 838