ipc/msg.c

   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  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  */
  20
  21 #include <linux/capability.h>
  22 #include <linux/config.h>
  23 #include <linux/slab.h>
  24 #include <linux/msg.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/init.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/list.h>
  29 #include <linux/security.h>
  30 #include <linux/sched.h>
  31 #include <linux/syscalls.h>
  32 #include <linux/audit.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/mutex.h>
  35
  36 #include <asm/current.h>
  37 #include <asm/uaccess.h>
  38 #include "util.h"
  39
  40 /* sysctl: */
  41 int msg_ctlmax = MSGMAX;
  42 int msg_ctlmnb = MSGMNB;
  43 int msg_ctlmni = MSGMNI;
  44
  45 /* one msg_receiver structure for each sleeping receiver */
  46 struct msg_receiver {
  47         struct list_head r_list;
  48         struct task_struct* r_tsk;
  49
  50         int r_mode;
  51         long r_msgtype;
  52         long r_maxsize;
  53
  54         struct msg_msg* volatile r_msg;
  55 };
  56
  57 /* one msg_sender for each sleeping sender */
  58 struct msg_sender {
  59         struct list_head list;
  60         struct task_struct* tsk;
  61 };
  62
  63 #define SEARCH_ANY              1
  64 #define SEARCH_EQUAL            2
  65 #define SEARCH_NOTEQUAL         3
  66 #define SEARCH_LESSEQUAL        4
  67
  68 static atomic_t msg_bytes = ATOMIC_INIT(0);
  69 static atomic_t msg_hdrs = ATOMIC_INIT(0);
  70
  71 static struct ipc_ids msg_ids;
  72
  73 #define msg_lock(id)    ((struct msg_queue*)ipc_lock(&msg_ids,id))
  74 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
  75 #define msg_rmid(id)    ((struct msg_queue*)ipc_rmid(&msg_ids,id))
  76 #define msg_checkid(msq, msgid) \
  77         ipc_checkid(&msg_ids,&msq->q_perm,msgid)
  78 #define msg_buildid(id, seq) \
  79         ipc_buildid(&msg_ids, id, seq)
  80
  81 static void freeque (struct msg_queue *msq, int id);
  82 static int newque (key_t key, int msgflg);
  83 #ifdef CONFIG_PROC_FS
  84 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  85 #endif
  86
  87 void __init msg_init (void)
  88 {
  89         ipc_init_ids(&msg_ids,msg_ctlmni);
  90         ipc_init_proc_interface("sysvipc/msg",
  91                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
  92                                 &msg_ids,
  93                                 sysvipc_msg_proc_show);
  94 }
  95
  96 static int newque (key_t key, int msgflg)
  97 {
  98         int id;
  99         int retval;
 100         struct msg_queue *msq;
 101
 102         msq  = ipc_rcu_alloc(sizeof(*msq));
 103         if (!msq)
 104                 return -ENOMEM;
 105
 106         msq->q_perm.mode = (msgflg & S_IRWXUGO);
 107         msq->q_perm.key = key;
 108
 109         msq->q_perm.security = NULL;
 110         retval = security_msg_queue_alloc(msq);
 111         if (retval) {
 112                 ipc_rcu_putref(msq);
 113                 return retval;
 114         }
 115
 116         id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
 117         if(id == -1) {
 118                 security_msg_queue_free(msq);
 119                 ipc_rcu_putref(msq);
 120                 return -ENOSPC;
 121         }
 122
 123         msq->q_id = msg_buildid(id,msq->q_perm.seq);
 124         msq->q_stime = msq->q_rtime = 0;
 125         msq->q_ctime = get_seconds();
 126         msq->q_cbytes = msq->q_qnum = 0;
 127         msq->q_qbytes = msg_ctlmnb;
 128         msq->q_lspid = msq->q_lrpid = 0;
 129         INIT_LIST_HEAD(&msq->q_messages);
 130         INIT_LIST_HEAD(&msq->q_receivers);
 131         INIT_LIST_HEAD(&msq->q_senders);
 132         msg_unlock(msq);
 133
 134         return msq->q_id;
 135 }
 136
 137 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
 138 {
 139         mss->tsk=current;
 140         current->state=TASK_INTERRUPTIBLE;
 141         list_add_tail(&mss->list,&msq->q_senders);
 142 }
 143
 144 static inline void ss_del(struct msg_sender* mss)
 145 {
 146         if(mss->list.next != NULL)
 147                 list_del(&mss->list);
 148 }
 149
 150 static void ss_wakeup(struct list_head* h, int kill)
 151 {
 152         struct list_head *tmp;
 153
 154         tmp = h->next;
 155         while (tmp != h) {
 156                 struct msg_sender* mss;
 157
 158                 mss = list_entry(tmp,struct msg_sender,list);
 159                 tmp = tmp->next;
 160                 if(kill)
 161                         mss->list.next=NULL;
 162                 wake_up_process(mss->tsk);
 163         }
 164 }
 165
 166 static void expunge_all(struct msg_queue* msq, int res)
 167 {
 168         struct list_head *tmp;
 169
 170         tmp = msq->q_receivers.next;
 171         while (tmp != &msq->q_receivers) {
 172                 struct msg_receiver* msr;
 173
 174                 msr = list_entry(tmp,struct msg_receiver,r_list);
 175                 tmp = tmp->next;
 176                 msr->r_msg = NULL;
 177                 wake_up_process(msr->r_tsk);
 178                 smp_mb();
 179                 msr->r_msg = ERR_PTR(res);
 180         }
 181 }
 182 /*
 183  * freeque() wakes up waiters on the sender and receiver waiting queue,
 184  * removes the message queue from message queue ID
 185  * array, and cleans up all the messages associated with this queue.
 186  *
 187  * msg_ids.mutex and the spinlock for this message queue is hold
 188  * before freeque() is called. msg_ids.mutex remains locked on exit.
 189  */
 190 static void freeque (struct msg_queue *msq, int id)
 191 {
 192         struct list_head *tmp;
 193
 194         expunge_all(msq,-EIDRM);
 195         ss_wakeup(&msq->q_senders,1);
 196         msq = msg_rmid(id);
 197         msg_unlock(msq);
 198
 199         tmp = msq->q_messages.next;
 200         while(tmp != &msq->q_messages) {
 201                 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
 202                 tmp = tmp->next;
 203                 atomic_dec(&msg_hdrs);
 204                 free_msg(msg);
 205         }
 206         atomic_sub(msq->q_cbytes, &msg_bytes);
 207         security_msg_queue_free(msq);
 208         ipc_rcu_putref(msq);
 209 }
 210
 211 asmlinkage long sys_msgget (key_t key, int msgflg)
 212 {
 213         int id, ret = -EPERM;
 214         struct msg_queue *msq;
 215
 216         mutex_lock(&msg_ids.mutex);
 217         if (key == IPC_PRIVATE)
 218                 ret = newque(key, msgflg);
 219         else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
 220                 if (!(msgflg & IPC_CREAT))
 221                         ret = -ENOENT;
 222                 else
 223                         ret = newque(key, msgflg);
 224         } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
 225                 ret = -EEXIST;
 226         } else {
 227                 msq = msg_lock(id);
 228                 BUG_ON(msq==NULL);
 229                 if (ipcperms(&msq->q_perm, msgflg))
 230                         ret = -EACCES;
 231                 else {
 232                         int qid = msg_buildid(id, msq->q_perm.seq);
 233                         ret = security_msg_queue_associate(msq, msgflg);
 234                         if (!ret)
 235                                 ret = qid;
 236                 }
 237                 msg_unlock(msq);
 238         }
 239         mutex_unlock(&msg_ids.mutex);
 240         return ret;
 241 }
 242
 243 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 244 {
 245         switch(version) {
 246         case IPC_64:
 247                 return copy_to_user (buf, in, sizeof(*in));
 248         case IPC_OLD:
 249             {
 250                 struct msqid_ds out;
 251
 252                 memset(&out,0,sizeof(out));
 253
 254                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 255
 256                 out.msg_stime           = in->msg_stime;
 257                 out.msg_rtime           = in->msg_rtime;
 258                 out.msg_ctime           = in->msg_ctime;
 259
 260                 if(in->msg_cbytes > USHRT_MAX)
 261                         out.msg_cbytes  = USHRT_MAX;
 262                 else
 263                         out.msg_cbytes  = in->msg_cbytes;
 264                 out.msg_lcbytes         = in->msg_cbytes;
 265
 266                 if(in->msg_qnum > USHRT_MAX)
 267                         out.msg_qnum    = USHRT_MAX;
 268                 else
 269                         out.msg_qnum    = in->msg_qnum;
 270
 271                 if(in->msg_qbytes > USHRT_MAX)
 272                         out.msg_qbytes  = USHRT_MAX;
 273                 else
 274                         out.msg_qbytes  = in->msg_qbytes;
 275                 out.msg_lqbytes         = in->msg_qbytes;
 276
 277                 out.msg_lspid           = in->msg_lspid;
 278                 out.msg_lrpid           = in->msg_lrpid;
 279
 280                 return copy_to_user (buf, &out, sizeof(out));
 281             }
 282         default:
 283                 return -EINVAL;
 284         }
 285 }
 286
 287 struct msq_setbuf {
 288         unsigned long   qbytes;
 289         uid_t           uid;
 290         gid_t           gid;
 291         mode_t          mode;
 292 };
 293
 294 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 295 {
 296         switch(version) {
 297         case IPC_64:
 298             {
 299                 struct msqid64_ds tbuf;
 300
 301                 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
 302                         return -EFAULT;
 303
 304                 out->qbytes             = tbuf.msg_qbytes;
 305                 out->uid                = tbuf.msg_perm.uid;
 306                 out->gid                = tbuf.msg_perm.gid;
 307                 out->mode               = tbuf.msg_perm.mode;
 308
 309                 return 0;
 310             }
 311         case IPC_OLD:
 312             {
 313                 struct msqid_ds tbuf_old;
 314
 315                 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
 316                         return -EFAULT;
 317
 318                 out->uid                = tbuf_old.msg_perm.uid;
 319                 out->gid                = tbuf_old.msg_perm.gid;
 320                 out->mode               = tbuf_old.msg_perm.mode;
 321
 322                 if(tbuf_old.msg_qbytes == 0)
 323                         out->qbytes     = tbuf_old.msg_lqbytes;
 324                 else
 325                         out->qbytes     = tbuf_old.msg_qbytes;
 326
 327                 return 0;
 328             }
 329         default:
 330                 return -EINVAL;
 331         }
 332 }
 333
 334 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
 335 {
 336         int err, version;
 337         struct msg_queue *msq;
 338         struct msq_setbuf setbuf;
 339         struct kern_ipc_perm *ipcp;
 340
 341         if (msqid < 0 || cmd < 0)
 342                 return -EINVAL;
 343
 344         version = ipc_parse_version(&cmd);
 345
 346         switch (cmd) {
 347         case IPC_INFO:
 348         case MSG_INFO:
 349         {
 350                 struct msginfo msginfo;
 351                 int max_id;
 352                 if (!buf)
 353                         return -EFAULT;
 354                 /* We must not return kernel stack data.
 355                  * due to padding, it's not enough
 356                  * to set all member fields.
 357                  */
 358
 359                 err = security_msg_queue_msgctl(NULL, cmd);
 360                 if (err)
 361                         return err;
 362
 363                 memset(&msginfo,0,sizeof(msginfo));
 364                 msginfo.msgmni = msg_ctlmni;
 365                 msginfo.msgmax = msg_ctlmax;
 366                 msginfo.msgmnb = msg_ctlmnb;
 367                 msginfo.msgssz = MSGSSZ;
 368                 msginfo.msgseg = MSGSEG;
 369                 mutex_lock(&msg_ids.mutex);
 370                 if (cmd == MSG_INFO) {
 371                         msginfo.msgpool = msg_ids.in_use;
 372                         msginfo.msgmap = atomic_read(&msg_hdrs);
 373                         msginfo.msgtql = atomic_read(&msg_bytes);
 374                 } else {
 375                         msginfo.msgmap = MSGMAP;
 376                         msginfo.msgpool = MSGPOOL;
 377                         msginfo.msgtql = MSGTQL;
 378                 }
 379                 max_id = msg_ids.max_id;
 380                 mutex_unlock(&msg_ids.mutex);
 381                 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
 382                         return -EFAULT;
 383                 return (max_id < 0) ? 0: max_id;
 384         }
 385         case MSG_STAT:
 386         case IPC_STAT:
 387         {
 388                 struct msqid64_ds tbuf;
 389                 int success_return;
 390                 if (!buf)
 391                         return -EFAULT;
 392                 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size)
 393                         return -EINVAL;
 394
 395                 memset(&tbuf,0,sizeof(tbuf));
 396
 397                 msq = msg_lock(msqid);
 398                 if (msq == NULL)
 399                         return -EINVAL;
 400
 401                 if(cmd == MSG_STAT) {
 402                         success_return = msg_buildid(msqid, msq->q_perm.seq);
 403                 } else {
 404                         err = -EIDRM;
 405                         if (msg_checkid(msq,msqid))
 406                                 goto out_unlock;
 407                         success_return = 0;
 408                 }
 409                 err = -EACCES;
 410                 if (ipcperms (&msq->q_perm, S_IRUGO))
 411                         goto out_unlock;
 412
 413                 err = security_msg_queue_msgctl(msq, cmd);
 414                 if (err)
 415                         goto out_unlock;
 416
 417                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 418                 tbuf.msg_stime  = msq->q_stime;
 419                 tbuf.msg_rtime  = msq->q_rtime;
 420                 tbuf.msg_ctime  = msq->q_ctime;
 421                 tbuf.msg_cbytes = msq->q_cbytes;
 422                 tbuf.msg_qnum   = msq->q_qnum;
 423                 tbuf.msg_qbytes = msq->q_qbytes;
 424                 tbuf.msg_lspid  = msq->q_lspid;
 425                 tbuf.msg_lrpid  = msq->q_lrpid;
 426                 msg_unlock(msq);
 427                 if (copy_msqid_to_user(buf, &tbuf, version))
 428                         return -EFAULT;
 429                 return success_return;
 430         }
 431         case IPC_SET:
 432                 if (!buf)
 433                         return -EFAULT;
 434                 if (copy_msqid_from_user (&setbuf, buf, version))
 435                         return -EFAULT;
 436                 break;
 437         case IPC_RMID:
 438                 break;
 439         default:
 440                 return  -EINVAL;
 441         }
 442
 443         mutex_lock(&msg_ids.mutex);
 444         msq = msg_lock(msqid);
 445         err=-EINVAL;
 446         if (msq == NULL)
 447                 goto out_up;
 448
 449         err = -EIDRM;
 450         if (msg_checkid(msq,msqid))
 451                 goto out_unlock_up;
 452         ipcp = &msq->q_perm;
 453
 454         err = audit_ipc_obj(ipcp);
 455         if (err)
 456                 goto out_unlock_up;
 457
 458         err = -EPERM;
 459         if (current->euid != ipcp->cuid &&
 460             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 461             /* We _could_ check for CAP_CHOWN above, but we don't */
 462                 goto out_unlock_up;
 463
 464         err = security_msg_queue_msgctl(msq, cmd);
 465         if (err)
 466                 goto out_unlock_up;
 467
 468         switch (cmd) {
 469         case IPC_SET:
 470         {
 471                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode, ipcp);
 472                 if (err)
 473                         goto out_unlock_up;
 474
 475                 err = -EPERM;
 476                 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 477                         goto out_unlock_up;
 478
 479                 msq->q_qbytes = setbuf.qbytes;
 480
 481                 ipcp->uid = setbuf.uid;
 482                 ipcp->gid = setbuf.gid;
 483                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 484                         (S_IRWXUGO & setbuf.mode);
 485                 msq->q_ctime = get_seconds();
 486                 /* sleeping receivers might be excluded by
 487                  * stricter permissions.
 488                  */
 489                 expunge_all(msq,-EAGAIN);
 490                 /* sleeping senders might be able to send
 491                  * due to a larger queue size.
 492                  */
 493                 ss_wakeup(&msq->q_senders,0);
 494                 msg_unlock(msq);
 495                 break;
 496         }
 497         case IPC_RMID:
 498                 freeque (msq, msqid);
 499                 break;
 500         }
 501         err = 0;
 502 out_up:
 503         mutex_unlock(&msg_ids.mutex);
 504         return err;
 505 out_unlock_up:
 506         msg_unlock(msq);
 507         goto out_up;
 508 out_unlock:
 509         msg_unlock(msq);
 510         return err;
 511 }
 512
 513 static int testmsg(struct msg_msg* msg,long type,int mode)
 514 {
 515         switch(mode)
 516         {
 517                 case SEARCH_ANY:
 518                         return 1;
 519                 case SEARCH_LESSEQUAL:
 520                         if(msg->m_type <=type)
 521                                 return 1;
 522                         break;
 523                 case SEARCH_EQUAL:
 524                         if(msg->m_type == type)
 525                                 return 1;
 526                         break;
 527                 case SEARCH_NOTEQUAL:
 528                         if(msg->m_type != type)
 529                                 return 1;
 530                         break;
 531         }
 532         return 0;
 533 }
 534
 535 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
 536 {
 537         struct list_head* tmp;
 538
 539         tmp = msq->q_receivers.next;
 540         while (tmp != &msq->q_receivers) {
 541                 struct msg_receiver* msr;
 542                 msr = list_entry(tmp,struct msg_receiver,r_list);
 543                 tmp = tmp->next;
 544                 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
 545                    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
 546                         list_del(&msr->r_list);
 547                         if(msr->r_maxsize < msg->m_ts) {
 548                                 msr->r_msg = NULL;
 549                                 wake_up_process(msr->r_tsk);
 550                                 smp_mb();
 551                                 msr->r_msg = ERR_PTR(-E2BIG);
 552                         } else {
 553                                 msr->r_msg = NULL;
 554                                 msq->q_lrpid = msr->r_tsk->pid;
 555                                 msq->q_rtime = get_seconds();
 556                                 wake_up_process(msr->r_tsk);
 557                                 smp_mb();
 558                                 msr->r_msg = msg;
 559                                 return 1;
 560                         }
 561                 }
 562         }
 563         return 0;
 564 }
 565
 566 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 567 {
 568         struct msg_queue *msq;
 569         struct msg_msg *msg;
 570         long mtype;
 571         int err;
 572
 573         if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 574                 return -EINVAL;
 575         if (get_user(mtype, &msgp->mtype))
 576                 return -EFAULT;
 577         if (mtype < 1)
 578                 return -EINVAL;
 579
 580         msg = load_msg(msgp->mtext, msgsz);
 581         if(IS_ERR(msg))
 582                 return PTR_ERR(msg);
 583
 584         msg->m_type = mtype;
 585         msg->m_ts = msgsz;
 586
 587         msq = msg_lock(msqid);
 588         err=-EINVAL;
 589         if(msq==NULL)
 590                 goto out_free;
 591
 592         err= -EIDRM;
 593         if (msg_checkid(msq,msqid))
 594                 goto out_unlock_free;
 595
 596         for (;;) {
 597                 struct msg_sender s;
 598
 599                 err=-EACCES;
 600                 if (ipcperms(&msq->q_perm, S_IWUGO))
 601                         goto out_unlock_free;
 602
 603                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 604                 if (err)
 605                         goto out_unlock_free;
 606
 607                 if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
 608                                 1 + msq->q_qnum <= msq->q_qbytes) {
 609                         break;
 610                 }
 611
 612                 /* queue full, wait: */
 613                 if(msgflg&IPC_NOWAIT) {
 614                         err=-EAGAIN;
 615                         goto out_unlock_free;
 616                 }
 617                 ss_add(msq, &s);
 618                 ipc_rcu_getref(msq);
 619                 msg_unlock(msq);
 620                 schedule();
 621
 622                 ipc_lock_by_ptr(&msq->q_perm);
 623                 ipc_rcu_putref(msq);
 624                 if (msq->q_perm.deleted) {
 625                         err = -EIDRM;
 626                         goto out_unlock_free;
 627                 }
 628                 ss_del(&s);
 629
 630                 if (signal_pending(current)) {
 631                         err=-ERESTARTNOHAND;
 632                         goto out_unlock_free;
 633                 }
 634         }
 635
 636         msq->q_lspid = current->tgid;
 637         msq->q_stime = get_seconds();
 638
 639         if(!pipelined_send(msq,msg)) {
 640                 /* noone is waiting for this message, enqueue it */
 641                 list_add_tail(&msg->m_list,&msq->q_messages);
 642                 msq->q_cbytes += msgsz;
 643                 msq->q_qnum++;
 644                 atomic_add(msgsz,&msg_bytes);
 645                 atomic_inc(&msg_hdrs);
 646         }
 647
 648         err = 0;
 649         msg = NULL;
 650
 651 out_unlock_free:
 652         msg_unlock(msq);
 653 out_free:
 654         if(msg!=NULL)
 655                 free_msg(msg);
 656         return err;
 657 }
 658
 659 static inline int convert_mode(long* msgtyp, int msgflg)
 660 {
 661         /*
 662          *  find message of correct type.
 663          *  msgtyp = 0 => get first.
 664          *  msgtyp > 0 => get first message of matching type.
 665          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 666          */
 667         if(*msgtyp==0)
 668                 return SEARCH_ANY;
 669         if(*msgtyp<0) {
 670                 *msgtyp=-(*msgtyp);
 671                 return SEARCH_LESSEQUAL;
 672         }
 673         if(msgflg & MSG_EXCEPT)
 674                 return SEARCH_NOTEQUAL;
 675         return SEARCH_EQUAL;
 676 }
 677
 678 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
 679                             long msgtyp, int msgflg)
 680 {
 681         struct msg_queue *msq;
 682         struct msg_msg *msg;
 683         int mode;
 684
 685         if (msqid < 0 || (long) msgsz < 0)
 686                 return -EINVAL;
 687         mode = convert_mode(&msgtyp,msgflg);
 688
 689         msq = msg_lock(msqid);
 690         if(msq==NULL)
 691                 return -EINVAL;
 692
 693         msg = ERR_PTR(-EIDRM);
 694         if (msg_checkid(msq,msqid))
 695                 goto out_unlock;
 696
 697         for (;;) {
 698                 struct msg_receiver msr_d;
 699                 struct list_head* tmp;
 700
 701                 msg = ERR_PTR(-EACCES);
 702                 if (ipcperms (&msq->q_perm, S_IRUGO))
 703                         goto out_unlock;
 704
 705                 msg = ERR_PTR(-EAGAIN);
 706                 tmp = msq->q_messages.next;
 707                 while (tmp != &msq->q_messages) {
 708                         struct msg_msg *walk_msg;
 709                         walk_msg = list_entry(tmp,struct msg_msg,m_list);
 710                         if(testmsg(walk_msg,msgtyp,mode) &&
 711                            !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
 712                                 msg = walk_msg;
 713                                 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
 714                                         msg=walk_msg;
 715                                         msgtyp=walk_msg->m_type-1;
 716                                 } else {
 717                                         msg=walk_msg;
 718                                         break;
 719                                 }
 720                         }
 721                         tmp = tmp->next;
 722                 }
 723                 if(!IS_ERR(msg)) {
 724                         /* Found a suitable message. Unlink it from the queue. */
 725                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 726                                 msg = ERR_PTR(-E2BIG);
 727                                 goto out_unlock;
 728                         }
 729                         list_del(&msg->m_list);
 730                         msq->q_qnum--;
 731                         msq->q_rtime = get_seconds();
 732                         msq->q_lrpid = current->tgid;
 733                         msq->q_cbytes -= msg->m_ts;
 734                         atomic_sub(msg->m_ts,&msg_bytes);
 735                         atomic_dec(&msg_hdrs);
 736                         ss_wakeup(&msq->q_senders,0);
 737                         msg_unlock(msq);
 738                         break;
 739                 }
 740                 /* No message waiting. Wait for a message */
 741                 if (msgflg & IPC_NOWAIT) {
 742                         msg = ERR_PTR(-ENOMSG);
 743                         goto out_unlock;
 744                 }
 745                 list_add_tail(&msr_d.r_list,&msq->q_receivers);
 746                 msr_d.r_tsk = current;
 747                 msr_d.r_msgtype = msgtyp;
 748                 msr_d.r_mode = mode;
 749                 if(msgflg & MSG_NOERROR)
 750                         msr_d.r_maxsize = INT_MAX;
 751                  else
 752                         msr_d.r_maxsize = msgsz;
 753                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 754                 current->state = TASK_INTERRUPTIBLE;
 755                 msg_unlock(msq);
 756
 757                 schedule();
 758
 759                 /* Lockless receive, part 1:
 760                  * Disable preemption.  We don't hold a reference to the queue
 761                  * and getting a reference would defeat the idea of a lockless
 762                  * operation, thus the code relies on rcu to guarantee the
 763                  * existance of msq:
 764                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 765                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 766                  * rcu_read_lock() prevents preemption between reading r_msg
 767                  * and the spin_lock() inside ipc_lock_by_ptr().
 768                  */
 769                 rcu_read_lock();
 770
 771                 /* Lockless receive, part 2:
 772                  * Wait until pipelined_send or expunge_all are outside of
 773                  * wake_up_process(). There is a race with exit(), see
 774                  * ipc/mqueue.c for the details.
 775                  */
 776                 msg = (struct msg_msg*) msr_d.r_msg;
 777                 while (msg == NULL) {
 778                         cpu_relax();
 779                         msg = (struct msg_msg*) msr_d.r_msg;
 780                 }
 781
 782                 /* Lockless receive, part 3:
 783                  * If there is a message or an error then accept it without
 784                  * locking.
 785                  */
 786                 if(msg != ERR_PTR(-EAGAIN)) {
 787                         rcu_read_unlock();
 788                         break;
 789                 }
 790
 791                 /* Lockless receive, part 3:
 792                  * Acquire the queue spinlock.
 793                  */
 794                 ipc_lock_by_ptr(&msq->q_perm);
 795                 rcu_read_unlock();
 796
 797                 /* Lockless receive, part 4:
 798                  * Repeat test after acquiring the spinlock.
 799                  */
 800                 msg = (struct msg_msg*)msr_d.r_msg;
 801                 if(msg != ERR_PTR(-EAGAIN))
 802                         goto out_unlock;
 803
 804                 list_del(&msr_d.r_list);
 805                 if (signal_pending(current)) {
 806                         msg = ERR_PTR(-ERESTARTNOHAND);
 807 out_unlock:
 808                         msg_unlock(msq);
 809                         break;
 810                 }
 811         }
 812         if (IS_ERR(msg))
 813                 return PTR_ERR(msg);
 814
 815         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 816         if (put_user (msg->m_type, &msgp->mtype) ||
 817             store_msg(msgp->mtext, msg, msgsz)) {
 818                     msgsz = -EFAULT;
 819         }
 820         free_msg(msg);
 821         return msgsz;
 822 }
 823
 824 #ifdef CONFIG_PROC_FS
 825 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 826 {
 827         struct msg_queue *msq = it;
 828
 829         return seq_printf(s,
 830                           "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 831                           msq->q_perm.key,
 832                           msq->q_id,
 833                           msq->q_perm.mode,
 834                           msq->q_cbytes,
 835                           msq->q_qnum,
 836                           msq->q_lspid,
 837                           msq->q_lrpid,
 838                           msq->q_perm.uid,
 839                           msq->q_perm.gid,
 840                           msq->q_perm.cuid,
 841                           msq->q_perm.cgid,
 842                           msq->q_stime,
 843                           msq->q_rtime,
 844                           msq->q_ctime);
 845 }
 846 #endif