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/slab.h>
  23 #include <linux/msg.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/init.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/list.h>
  28 #include <linux/security.h>
  29 #include <linux/sched.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/audit.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/mutex.h>
  34
  35 #include <asm/current.h>
  36 #include <asm/uaccess.h>
  37 #include "util.h"
  38
  39 /* sysctl: */
  40 int msg_ctlmax = MSGMAX;
  41 int msg_ctlmnb = MSGMNB;
  42 int msg_ctlmni = MSGMNI;
  43
  44 /*
  45  * one msg_receiver structure for each sleeping receiver:
  46  */
  47 struct msg_receiver {
  48         struct list_head        r_list;
  49         struct task_struct      *r_tsk;
  50
  51         int                     r_mode;
  52         long                    r_msgtype;
  53         long                    r_maxsize;
  54
  55         volatile struct msg_msg *r_msg;
  56 };
  57
  58 /* one msg_sender for each sleeping sender */
  59 struct msg_sender {
  60         struct list_head        list;
  61         struct task_struct      *tsk;
  62 };
  63
  64 #define SEARCH_ANY              1
  65 #define SEARCH_EQUAL            2
  66 #define SEARCH_NOTEQUAL         3
  67 #define SEARCH_LESSEQUAL        4
  68
  69 static atomic_t msg_bytes =     ATOMIC_INIT(0);
  70 static atomic_t msg_hdrs =      ATOMIC_INIT(0);
  71
  72 static struct ipc_ids msg_ids;
  73
  74 #define msg_lock(id)            ((struct msg_queue *)ipc_lock(&msg_ids, id))
  75 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  76 #define msg_rmid(id)            ((struct msg_queue *)ipc_rmid(&msg_ids, id))
  77 #define msg_checkid(msq, msgid) ipc_checkid(&msg_ids, &msq->q_perm, msgid)
  78 #define msg_buildid(id, seq)    ipc_buildid(&msg_ids, id, seq)
  79
  80 static void freeque(struct msg_queue *msq, int id);
  81 static int newque(key_t key, int msgflg);
  82 #ifdef CONFIG_PROC_FS
  83 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  84 #endif
  85
  86 void __init msg_init(void)
  87 {
  88         ipc_init_ids(&msg_ids, msg_ctlmni);
  89         ipc_init_proc_interface("sysvipc/msg",
  90                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
  91                                 &msg_ids,
  92                                 sysvipc_msg_proc_show);
  93 }
  94
  95 static int newque(key_t key, int msgflg)
  96 {
  97         struct msg_queue *msq;
  98         int id, retval;
  99
 100         msq = ipc_rcu_alloc(sizeof(*msq));
 101         if (!msq)
 102                 return -ENOMEM;
 103
 104         msq->q_perm.mode = msgflg & S_IRWXUGO;
 105         msq->q_perm.key = key;
 106
 107         msq->q_perm.security = NULL;
 108         retval = security_msg_queue_alloc(msq);
 109         if (retval) {
 110                 ipc_rcu_putref(msq);
 111                 return retval;
 112         }
 113
 114         id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
 115         if (id == -1) {
 116                 security_msg_queue_free(msq);
 117                 ipc_rcu_putref(msq);
 118                 return -ENOSPC;
 119         }
 120
 121         msq->q_id = msg_buildid(id, msq->q_perm.seq);
 122         msq->q_stime = msq->q_rtime = 0;
 123         msq->q_ctime = get_seconds();
 124         msq->q_cbytes = msq->q_qnum = 0;
 125         msq->q_qbytes = msg_ctlmnb;
 126         msq->q_lspid = msq->q_lrpid = 0;
 127         INIT_LIST_HEAD(&msq->q_messages);
 128         INIT_LIST_HEAD(&msq->q_receivers);
 129         INIT_LIST_HEAD(&msq->q_senders);
 130         msg_unlock(msq);
 131
 132         return msq->q_id;
 133 }
 134
 135 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 136 {
 137         mss->tsk = current;
 138         current->state = TASK_INTERRUPTIBLE;
 139         list_add_tail(&mss->list, &msq->q_senders);
 140 }
 141
 142 static inline void ss_del(struct msg_sender *mss)
 143 {
 144         if (mss->list.next != NULL)
 145                 list_del(&mss->list);
 146 }
 147
 148 static void ss_wakeup(struct list_head *h, int kill)
 149 {
 150         struct list_head *tmp;
 151
 152         tmp = h->next;
 153         while (tmp != h) {
 154                 struct msg_sender *mss;
 155
 156                 mss = list_entry(tmp, struct msg_sender, list);
 157                 tmp = tmp->next;
 158                 if (kill)
 159                         mss->list.next = NULL;
 160                 wake_up_process(mss->tsk);
 161         }
 162 }
 163
 164 static void expunge_all(struct msg_queue *msq, int res)
 165 {
 166         struct list_head *tmp;
 167
 168         tmp = msq->q_receivers.next;
 169         while (tmp != &msq->q_receivers) {
 170                 struct msg_receiver *msr;
 171
 172                 msr = list_entry(tmp, struct msg_receiver, r_list);
 173                 tmp = tmp->next;
 174                 msr->r_msg = NULL;
 175                 wake_up_process(msr->r_tsk);
 176                 smp_mb();
 177                 msr->r_msg = ERR_PTR(res);
 178         }
 179 }
 180
 181 /*
 182  * freeque() wakes up waiters on the sender and receiver waiting queue,
 183  * removes the message queue from message queue ID
 184  * array, and cleans up all the messages associated with this queue.
 185  *
 186  * msg_ids.mutex and the spinlock for this message queue is hold
 187  * before freeque() is called. msg_ids.mutex remains locked on exit.
 188  */
 189 static void freeque(struct msg_queue *msq, int id)
 190 {
 191         struct list_head *tmp;
 192
 193         expunge_all(msq, -EIDRM);
 194         ss_wakeup(&msq->q_senders, 1);
 195         msq = msg_rmid(id);
 196         msg_unlock(msq);
 197
 198         tmp = msq->q_messages.next;
 199         while (tmp != &msq->q_messages) {
 200                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 201
 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         struct msg_queue *msq;
 214         int id, ret = -EPERM;
 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
 234                         ret = security_msg_queue_associate(msq, msgflg);
 235                         if (!ret)
 236                                 ret = qid;
 237                 }
 238                 msg_unlock(msq);
 239         }
 240         mutex_unlock(&msg_ids.mutex);
 241
 242         return ret;
 243 }
 244
 245 static inline unsigned long
 246 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 247 {
 248         switch(version) {
 249         case IPC_64:
 250                 return copy_to_user(buf, in, sizeof(*in));
 251         case IPC_OLD:
 252         {
 253                 struct msqid_ds out;
 254
 255                 memset(&out, 0, sizeof(out));
 256
 257                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 258
 259                 out.msg_stime           = in->msg_stime;
 260                 out.msg_rtime           = in->msg_rtime;
 261                 out.msg_ctime           = in->msg_ctime;
 262
 263                 if (in->msg_cbytes > USHRT_MAX)
 264                         out.msg_cbytes  = USHRT_MAX;
 265                 else
 266                         out.msg_cbytes  = in->msg_cbytes;
 267                 out.msg_lcbytes         = in->msg_cbytes;
 268
 269                 if (in->msg_qnum > USHRT_MAX)
 270                         out.msg_qnum    = USHRT_MAX;
 271                 else
 272                         out.msg_qnum    = in->msg_qnum;
 273
 274                 if (in->msg_qbytes > USHRT_MAX)
 275                         out.msg_qbytes  = USHRT_MAX;
 276                 else
 277                         out.msg_qbytes  = in->msg_qbytes;
 278                 out.msg_lqbytes         = in->msg_qbytes;
 279
 280                 out.msg_lspid           = in->msg_lspid;
 281                 out.msg_lrpid           = in->msg_lrpid;
 282
 283                 return copy_to_user(buf, &out, sizeof(out));
 284         }
 285         default:
 286                 return -EINVAL;
 287         }
 288 }
 289
 290 struct msq_setbuf {
 291         unsigned long   qbytes;
 292         uid_t           uid;
 293         gid_t           gid;
 294         mode_t          mode;
 295 };
 296
 297 static inline unsigned long
 298 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 299 {
 300         switch(version) {
 301         case IPC_64:
 302         {
 303                 struct msqid64_ds tbuf;
 304
 305                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 306                         return -EFAULT;
 307
 308                 out->qbytes             = tbuf.msg_qbytes;
 309                 out->uid                = tbuf.msg_perm.uid;
 310                 out->gid                = tbuf.msg_perm.gid;
 311                 out->mode               = tbuf.msg_perm.mode;
 312
 313                 return 0;
 314         }
 315         case IPC_OLD:
 316         {
 317                 struct msqid_ds tbuf_old;
 318
 319                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 320                         return -EFAULT;
 321
 322                 out->uid                = tbuf_old.msg_perm.uid;
 323                 out->gid                = tbuf_old.msg_perm.gid;
 324                 out->mode               = tbuf_old.msg_perm.mode;
 325
 326                 if (tbuf_old.msg_qbytes == 0)
 327                         out->qbytes     = tbuf_old.msg_lqbytes;
 328                 else
 329                         out->qbytes     = tbuf_old.msg_qbytes;
 330
 331                 return 0;
 332         }
 333         default:
 334                 return -EINVAL;
 335         }
 336 }
 337
 338 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 339 {
 340         struct kern_ipc_perm *ipcp;
 341         struct msq_setbuf setbuf;
 342         struct msg_queue *msq;
 343         int err, version;
 344
 345         if (msqid < 0 || cmd < 0)
 346                 return -EINVAL;
 347
 348         version = ipc_parse_version(&cmd);
 349
 350         switch (cmd) {
 351         case IPC_INFO:
 352         case MSG_INFO:
 353         {
 354                 struct msginfo msginfo;
 355                 int max_id;
 356
 357                 if (!buf)
 358                         return -EFAULT;
 359                 /*
 360                  * We must not return kernel stack data.
 361                  * due to padding, it's not enough
 362                  * to set all member fields.
 363                  */
 364                 err = security_msg_queue_msgctl(NULL, cmd);
 365                 if (err)
 366                         return err;
 367
 368                 memset(&msginfo, 0, sizeof(msginfo));
 369                 msginfo.msgmni = msg_ctlmni;
 370                 msginfo.msgmax = msg_ctlmax;
 371                 msginfo.msgmnb = msg_ctlmnb;
 372                 msginfo.msgssz = MSGSSZ;
 373                 msginfo.msgseg = MSGSEG;
 374                 mutex_lock(&msg_ids.mutex);
 375                 if (cmd == MSG_INFO) {
 376                         msginfo.msgpool = msg_ids.in_use;
 377                         msginfo.msgmap = atomic_read(&msg_hdrs);
 378                         msginfo.msgtql = atomic_read(&msg_bytes);
 379                 } else {
 380                         msginfo.msgmap = MSGMAP;
 381                         msginfo.msgpool = MSGPOOL;
 382                         msginfo.msgtql = MSGTQL;
 383                 }
 384                 max_id = msg_ids.max_id;
 385                 mutex_unlock(&msg_ids.mutex);
 386                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 387                         return -EFAULT;
 388                 return (max_id < 0) ? 0 : max_id;
 389         }
 390         case MSG_STAT:
 391         case IPC_STAT:
 392         {
 393                 struct msqid64_ds tbuf;
 394                 int success_return;
 395
 396                 if (!buf)
 397                         return -EFAULT;
 398                 if (cmd == MSG_STAT && msqid >= msg_ids.entries->size)
 399                         return -EINVAL;
 400
 401                 memset(&tbuf, 0, sizeof(tbuf));
 402
 403                 msq = msg_lock(msqid);
 404                 if (msq == NULL)
 405                         return -EINVAL;
 406
 407                 if (cmd == MSG_STAT) {
 408                         success_return = msg_buildid(msqid, msq->q_perm.seq);
 409                 } else {
 410                         err = -EIDRM;
 411                         if (msg_checkid(msq, msqid))
 412                                 goto out_unlock;
 413                         success_return = 0;
 414                 }
 415                 err = -EACCES;
 416                 if (ipcperms(&msq->q_perm, S_IRUGO))
 417                         goto out_unlock;
 418
 419                 err = security_msg_queue_msgctl(msq, cmd);
 420                 if (err)
 421                         goto out_unlock;
 422
 423                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 424                 tbuf.msg_stime  = msq->q_stime;
 425                 tbuf.msg_rtime  = msq->q_rtime;
 426                 tbuf.msg_ctime  = msq->q_ctime;
 427                 tbuf.msg_cbytes = msq->q_cbytes;
 428                 tbuf.msg_qnum   = msq->q_qnum;
 429                 tbuf.msg_qbytes = msq->q_qbytes;
 430                 tbuf.msg_lspid  = msq->q_lspid;
 431                 tbuf.msg_lrpid  = msq->q_lrpid;
 432                 msg_unlock(msq);
 433                 if (copy_msqid_to_user(buf, &tbuf, version))
 434                         return -EFAULT;
 435                 return success_return;
 436         }
 437         case IPC_SET:
 438                 if (!buf)
 439                         return -EFAULT;
 440                 if (copy_msqid_from_user(&setbuf, buf, version))
 441                         return -EFAULT;
 442                 break;
 443         case IPC_RMID:
 444                 break;
 445         default:
 446                 return  -EINVAL;
 447         }
 448
 449         mutex_lock(&msg_ids.mutex);
 450         msq = msg_lock(msqid);
 451         err = -EINVAL;
 452         if (msq == NULL)
 453                 goto out_up;
 454
 455         err = -EIDRM;
 456         if (msg_checkid(msq, msqid))
 457                 goto out_unlock_up;
 458         ipcp = &msq->q_perm;
 459
 460         err = audit_ipc_obj(ipcp);
 461         if (err)
 462                 goto out_unlock_up;
 463         if (cmd==IPC_SET) {
 464                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
 465                                          setbuf.mode);
 466                 if (err)
 467                         goto out_unlock_up;
 468         }
 469
 470         err = -EPERM;
 471         if (current->euid != ipcp->cuid &&
 472             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 473                 /* We _could_ check for CAP_CHOWN above, but we don't */
 474                 goto out_unlock_up;
 475
 476         err = security_msg_queue_msgctl(msq, cmd);
 477         if (err)
 478                 goto out_unlock_up;
 479
 480         switch (cmd) {
 481         case IPC_SET:
 482         {
 483                 err = -EPERM;
 484                 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 485                         goto out_unlock_up;
 486
 487                 msq->q_qbytes = setbuf.qbytes;
 488
 489                 ipcp->uid = setbuf.uid;
 490                 ipcp->gid = setbuf.gid;
 491                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 492                              (S_IRWXUGO & setbuf.mode);
 493                 msq->q_ctime = get_seconds();
 494                 /* sleeping receivers might be excluded by
 495                  * stricter permissions.
 496                  */
 497                 expunge_all(msq, -EAGAIN);
 498                 /* sleeping senders might be able to send
 499                  * due to a larger queue size.
 500                  */
 501                 ss_wakeup(&msq->q_senders, 0);
 502                 msg_unlock(msq);
 503                 break;
 504         }
 505         case IPC_RMID:
 506                 freeque(msq, msqid);
 507                 break;
 508         }
 509         err = 0;
 510 out_up:
 511         mutex_unlock(&msg_ids.mutex);
 512         return err;
 513 out_unlock_up:
 514         msg_unlock(msq);
 515         goto out_up;
 516 out_unlock:
 517         msg_unlock(msq);
 518         return err;
 519 }
 520
 521 static int testmsg(struct msg_msg *msg, long type, int mode)
 522 {
 523         switch(mode)
 524         {
 525                 case SEARCH_ANY:
 526                         return 1;
 527                 case SEARCH_LESSEQUAL:
 528                         if (msg->m_type <=type)
 529                                 return 1;
 530                         break;
 531                 case SEARCH_EQUAL:
 532                         if (msg->m_type == type)
 533                                 return 1;
 534                         break;
 535                 case SEARCH_NOTEQUAL:
 536                         if (msg->m_type != type)
 537                                 return 1;
 538                         break;
 539         }
 540         return 0;
 541 }
 542
 543 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 544 {
 545         struct list_head *tmp;
 546
 547         tmp = msq->q_receivers.next;
 548         while (tmp != &msq->q_receivers) {
 549                 struct msg_receiver *msr;
 550
 551                 msr = list_entry(tmp, struct msg_receiver, r_list);
 552                 tmp = tmp->next;
 553                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 554                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 555                                                msr->r_msgtype, msr->r_mode)) {
 556
 557                         list_del(&msr->r_list);
 558                         if (msr->r_maxsize < msg->m_ts) {
 559                                 msr->r_msg = NULL;
 560                                 wake_up_process(msr->r_tsk);
 561                                 smp_mb();
 562                                 msr->r_msg = ERR_PTR(-E2BIG);
 563                         } else {
 564                                 msr->r_msg = NULL;
 565                                 msq->q_lrpid = msr->r_tsk->pid;
 566                                 msq->q_rtime = get_seconds();
 567                                 wake_up_process(msr->r_tsk);
 568                                 smp_mb();
 569                                 msr->r_msg = msg;
 570
 571                                 return 1;
 572                         }
 573                 }
 574         }
 575         return 0;
 576 }
 577
 578 asmlinkage long
 579 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 580 {
 581         struct msg_queue *msq;
 582         struct msg_msg *msg;
 583         long mtype;
 584         int err;
 585
 586         if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 587                 return -EINVAL;
 588         if (get_user(mtype, &msgp->mtype))
 589                 return -EFAULT;
 590         if (mtype < 1)
 591                 return -EINVAL;
 592
 593         msg = load_msg(msgp->mtext, msgsz);
 594         if (IS_ERR(msg))
 595                 return PTR_ERR(msg);
 596
 597         msg->m_type = mtype;
 598         msg->m_ts = msgsz;
 599
 600         msq = msg_lock(msqid);
 601         err = -EINVAL;
 602         if (msq == NULL)
 603                 goto out_free;
 604
 605         err= -EIDRM;
 606         if (msg_checkid(msq, msqid))
 607                 goto out_unlock_free;
 608
 609         for (;;) {
 610                 struct msg_sender s;
 611
 612                 err = -EACCES;
 613                 if (ipcperms(&msq->q_perm, S_IWUGO))
 614                         goto out_unlock_free;
 615
 616                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 617                 if (err)
 618                         goto out_unlock_free;
 619
 620                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 621                                 1 + msq->q_qnum <= msq->q_qbytes) {
 622                         break;
 623                 }
 624
 625                 /* queue full, wait: */
 626                 if (msgflg & IPC_NOWAIT) {
 627                         err = -EAGAIN;
 628                         goto out_unlock_free;
 629                 }
 630                 ss_add(msq, &s);
 631                 ipc_rcu_getref(msq);
 632                 msg_unlock(msq);
 633                 schedule();
 634
 635                 ipc_lock_by_ptr(&msq->q_perm);
 636                 ipc_rcu_putref(msq);
 637                 if (msq->q_perm.deleted) {
 638                         err = -EIDRM;
 639                         goto out_unlock_free;
 640                 }
 641                 ss_del(&s);
 642
 643                 if (signal_pending(current)) {
 644                         err = -ERESTARTNOHAND;
 645                         goto out_unlock_free;
 646                 }
 647         }
 648
 649         msq->q_lspid = current->tgid;
 650         msq->q_stime = get_seconds();
 651
 652         if (!pipelined_send(msq, msg)) {
 653                 /* noone is waiting for this message, enqueue it */
 654                 list_add_tail(&msg->m_list, &msq->q_messages);
 655                 msq->q_cbytes += msgsz;
 656                 msq->q_qnum++;
 657                 atomic_add(msgsz, &msg_bytes);
 658                 atomic_inc(&msg_hdrs);
 659         }
 660
 661         err = 0;
 662         msg = NULL;
 663
 664 out_unlock_free:
 665         msg_unlock(msq);
 666 out_free:
 667         if (msg != NULL)
 668                 free_msg(msg);
 669         return err;
 670 }
 671
 672 static inline int convert_mode(long *msgtyp, int msgflg)
 673 {
 674         /*
 675          *  find message of correct type.
 676          *  msgtyp = 0 => get first.
 677          *  msgtyp > 0 => get first message of matching type.
 678          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 679          */
 680         if (*msgtyp == 0)
 681                 return SEARCH_ANY;
 682         if (*msgtyp < 0) {
 683                 *msgtyp = -*msgtyp;
 684                 return SEARCH_LESSEQUAL;
 685         }
 686         if (msgflg & MSG_EXCEPT)
 687                 return SEARCH_NOTEQUAL;
 688         return SEARCH_EQUAL;
 689 }
 690
 691 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 692                            long msgtyp, int msgflg)
 693 {
 694         struct msg_queue *msq;
 695         struct msg_msg *msg;
 696         int mode;
 697
 698         if (msqid < 0 || (long) msgsz < 0)
 699                 return -EINVAL;
 700         mode = convert_mode(&msgtyp, msgflg);
 701
 702         msq = msg_lock(msqid);
 703         if (msq == NULL)
 704                 return -EINVAL;
 705
 706         msg = ERR_PTR(-EIDRM);
 707         if (msg_checkid(msq, msqid))
 708                 goto out_unlock;
 709
 710         for (;;) {
 711                 struct msg_receiver msr_d;
 712                 struct list_head *tmp;
 713
 714                 msg = ERR_PTR(-EACCES);
 715                 if (ipcperms(&msq->q_perm, S_IRUGO))
 716                         goto out_unlock;
 717
 718                 msg = ERR_PTR(-EAGAIN);
 719                 tmp = msq->q_messages.next;
 720                 while (tmp != &msq->q_messages) {
 721                         struct msg_msg *walk_msg;
 722
 723                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 724                         if (testmsg(walk_msg, msgtyp, mode) &&
 725                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 726                                                        msgtyp, mode)) {
 727
 728                                 msg = walk_msg;
 729                                 if (mode == SEARCH_LESSEQUAL &&
 730                                                 walk_msg->m_type != 1) {
 731                                         msg = walk_msg;
 732                                         msgtyp = walk_msg->m_type - 1;
 733                                 } else {
 734                                         msg = walk_msg;
 735                                         break;
 736                                 }
 737                         }
 738                         tmp = tmp->next;
 739                 }
 740                 if (!IS_ERR(msg)) {
 741                         /*
 742                          * Found a suitable message.
 743                          * Unlink it from the queue.
 744                          */
 745                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 746                                 msg = ERR_PTR(-E2BIG);
 747                                 goto out_unlock;
 748                         }
 749                         list_del(&msg->m_list);
 750                         msq->q_qnum--;
 751                         msq->q_rtime = get_seconds();
 752                         msq->q_lrpid = current->tgid;
 753                         msq->q_cbytes -= msg->m_ts;
 754                         atomic_sub(msg->m_ts, &msg_bytes);
 755                         atomic_dec(&msg_hdrs);
 756                         ss_wakeup(&msq->q_senders, 0);
 757                         msg_unlock(msq);
 758                         break;
 759                 }
 760                 /* No message waiting. Wait for a message */
 761                 if (msgflg & IPC_NOWAIT) {
 762                         msg = ERR_PTR(-ENOMSG);
 763                         goto out_unlock;
 764                 }
 765                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 766                 msr_d.r_tsk = current;
 767                 msr_d.r_msgtype = msgtyp;
 768                 msr_d.r_mode = mode;
 769                 if (msgflg & MSG_NOERROR)
 770                         msr_d.r_maxsize = INT_MAX;
 771                 else
 772                         msr_d.r_maxsize = msgsz;
 773                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 774                 current->state = TASK_INTERRUPTIBLE;
 775                 msg_unlock(msq);
 776
 777                 schedule();
 778
 779                 /* Lockless receive, part 1:
 780                  * Disable preemption.  We don't hold a reference to the queue
 781                  * and getting a reference would defeat the idea of a lockless
 782                  * operation, thus the code relies on rcu to guarantee the
 783                  * existance of msq:
 784                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 785                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 786                  * rcu_read_lock() prevents preemption between reading r_msg
 787                  * and the spin_lock() inside ipc_lock_by_ptr().
 788                  */
 789                 rcu_read_lock();
 790
 791                 /* Lockless receive, part 2:
 792                  * Wait until pipelined_send or expunge_all are outside of
 793                  * wake_up_process(). There is a race with exit(), see
 794                  * ipc/mqueue.c for the details.
 795                  */
 796                 msg = (struct msg_msg*)msr_d.r_msg;
 797                 while (msg == NULL) {
 798                         cpu_relax();
 799                         msg = (struct msg_msg *)msr_d.r_msg;
 800                 }
 801
 802                 /* Lockless receive, part 3:
 803                  * If there is a message or an error then accept it without
 804                  * locking.
 805                  */
 806                 if (msg != ERR_PTR(-EAGAIN)) {
 807                         rcu_read_unlock();
 808                         break;
 809                 }
 810
 811                 /* Lockless receive, part 3:
 812                  * Acquire the queue spinlock.
 813                  */
 814                 ipc_lock_by_ptr(&msq->q_perm);
 815                 rcu_read_unlock();
 816
 817                 /* Lockless receive, part 4:
 818                  * Repeat test after acquiring the spinlock.
 819                  */
 820                 msg = (struct msg_msg*)msr_d.r_msg;
 821                 if (msg != ERR_PTR(-EAGAIN))
 822                         goto out_unlock;
 823
 824                 list_del(&msr_d.r_list);
 825                 if (signal_pending(current)) {
 826                         msg = ERR_PTR(-ERESTARTNOHAND);
 827 out_unlock:
 828                         msg_unlock(msq);
 829                         break;
 830                 }
 831         }
 832         if (IS_ERR(msg))
 833                 return PTR_ERR(msg);
 834
 835         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 836         if (put_user (msg->m_type, &msgp->mtype) ||
 837             store_msg(msgp->mtext, msg, msgsz)) {
 838                 msgsz = -EFAULT;
 839         }
 840         free_msg(msg);
 841
 842         return msgsz;
 843 }
 844
 845 #ifdef CONFIG_PROC_FS
 846 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 847 {
 848         struct msg_queue *msq = it;
 849
 850         return seq_printf(s,
 851                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 852                         msq->q_perm.key,
 853                         msq->q_id,
 854                         msq->q_perm.mode,
 855                         msq->q_cbytes,
 856                         msq->q_qnum,
 857                         msq->q_lspid,
 858                         msq->q_lrpid,
 859                         msq->q_perm.uid,
 860                         msq->q_perm.gid,
 861                         msq->q_perm.cuid,
 862                         msq->q_perm.cgid,
 863                         msq->q_stime,
 864                         msq->q_rtime,
 865                         msq->q_ctime);
 866 }
 867 #endif