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  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  */
  24
  25 #include <linux/capability.h>
  26 #include <linux/slab.h>
  27 #include <linux/msg.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/init.h>
  30 #include <linux/proc_fs.h>
  31 #include <linux/list.h>
  32 #include <linux/security.h>
  33 #include <linux/sched.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/audit.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/mutex.h>
  38 #include <linux/nsproxy.h>
  39
  40 #include <asm/current.h>
  41 #include <asm/uaccess.h>
  42 #include "util.h"
  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         struct msg_msg          *volatile 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 init_msg_ids;
  73
  74 #define msg_ids(ns)     (*((ns)->ids[IPC_MSG_IDS]))
  75
  76 #define msg_lock(ns, id)        ((struct msg_queue*)ipc_lock(&msg_ids(ns), id))
  77 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  78 #define msg_rmid(ns, id)        ((struct msg_queue*)ipc_rmid(&msg_ids(ns), id))
  79 #define msg_checkid(ns, msq, msgid)     \
  80         ipc_checkid(&msg_ids(ns), &msq->q_perm, msgid)
  81 #define msg_buildid(ns, id, seq) \
  82         ipc_buildid(&msg_ids(ns), id, seq)
  83
  84 static void freeque (struct ipc_namespace *ns, struct msg_queue *msq, int id);
  85 static int newque (struct ipc_namespace *ns, key_t key, int msgflg);
  86 #ifdef CONFIG_PROC_FS
  87 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  88 #endif
  89
  90 static void __ipc_init __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  91 {
  92         ns->ids[IPC_MSG_IDS] = ids;
  93         ns->msg_ctlmax = MSGMAX;
  94         ns->msg_ctlmnb = MSGMNB;
  95         ns->msg_ctlmni = MSGMNI;
  96         ipc_init_ids(ids, ns->msg_ctlmni);
  97 }
  98
  99 #ifdef CONFIG_IPC_NS
 100 int msg_init_ns(struct ipc_namespace *ns)
 101 {
 102         struct ipc_ids *ids;
 103
 104         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 105         if (ids == NULL)
 106                 return -ENOMEM;
 107
 108         __msg_init_ns(ns, ids);
 109         return 0;
 110 }
 111
 112 void msg_exit_ns(struct ipc_namespace *ns)
 113 {
 114         int i;
 115         struct msg_queue *msq;
 116
 117         mutex_lock(&msg_ids(ns).mutex);
 118         for (i = 0; i <= msg_ids(ns).max_id; i++) {
 119                 msq = msg_lock(ns, i);
 120                 if (msq == NULL)
 121                         continue;
 122
 123                 freeque(ns, msq, i);
 124         }
 125         mutex_unlock(&msg_ids(ns).mutex);
 126
 127         ipc_fini_ids(ns->ids[IPC_MSG_IDS]);
 128         kfree(ns->ids[IPC_MSG_IDS]);
 129         ns->ids[IPC_MSG_IDS] = NULL;
 130 }
 131 #endif
 132
 133 void __init msg_init(void)
 134 {
 135         __msg_init_ns(&init_ipc_ns, &init_msg_ids);
 136         ipc_init_proc_interface("sysvipc/msg",
 137                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 138                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 139 }
 140
 141 static int newque (struct ipc_namespace *ns, key_t key, int msgflg)
 142 {
 143         struct msg_queue *msq;
 144         int id, retval;
 145
 146         msq = ipc_rcu_alloc(sizeof(*msq));
 147         if (!msq)
 148                 return -ENOMEM;
 149
 150         msq->q_perm.mode = msgflg & S_IRWXUGO;
 151         msq->q_perm.key = key;
 152
 153         msq->q_perm.security = NULL;
 154         retval = security_msg_queue_alloc(msq);
 155         if (retval) {
 156                 ipc_rcu_putref(msq);
 157                 return retval;
 158         }
 159
 160         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 161         if (id == -1) {
 162                 security_msg_queue_free(msq);
 163                 ipc_rcu_putref(msq);
 164                 return -ENOSPC;
 165         }
 166
 167         msq->q_id = msg_buildid(ns, id, msq->q_perm.seq);
 168         msq->q_stime = msq->q_rtime = 0;
 169         msq->q_ctime = get_seconds();
 170         msq->q_cbytes = msq->q_qnum = 0;
 171         msq->q_qbytes = ns->msg_ctlmnb;
 172         msq->q_lspid = msq->q_lrpid = 0;
 173         INIT_LIST_HEAD(&msq->q_messages);
 174         INIT_LIST_HEAD(&msq->q_receivers);
 175         INIT_LIST_HEAD(&msq->q_senders);
 176         msg_unlock(msq);
 177
 178         return msq->q_id;
 179 }
 180
 181 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 182 {
 183         mss->tsk = current;
 184         current->state = TASK_INTERRUPTIBLE;
 185         list_add_tail(&mss->list, &msq->q_senders);
 186 }
 187
 188 static inline void ss_del(struct msg_sender *mss)
 189 {
 190         if (mss->list.next != NULL)
 191                 list_del(&mss->list);
 192 }
 193
 194 static void ss_wakeup(struct list_head *h, int kill)
 195 {
 196         struct list_head *tmp;
 197
 198         tmp = h->next;
 199         while (tmp != h) {
 200                 struct msg_sender *mss;
 201
 202                 mss = list_entry(tmp, struct msg_sender, list);
 203                 tmp = tmp->next;
 204                 if (kill)
 205                         mss->list.next = NULL;
 206                 wake_up_process(mss->tsk);
 207         }
 208 }
 209
 210 static void expunge_all(struct msg_queue *msq, int res)
 211 {
 212         struct list_head *tmp;
 213
 214         tmp = msq->q_receivers.next;
 215         while (tmp != &msq->q_receivers) {
 216                 struct msg_receiver *msr;
 217
 218                 msr = list_entry(tmp, struct msg_receiver, r_list);
 219                 tmp = tmp->next;
 220                 msr->r_msg = NULL;
 221                 wake_up_process(msr->r_tsk);
 222                 smp_mb();
 223                 msr->r_msg = ERR_PTR(res);
 224         }
 225 }
 226
 227 /*
 228  * freeque() wakes up waiters on the sender and receiver waiting queue,
 229  * removes the message queue from message queue ID
 230  * array, and cleans up all the messages associated with this queue.
 231  *
 232  * msg_ids.mutex and the spinlock for this message queue is hold
 233  * before freeque() is called. msg_ids.mutex remains locked on exit.
 234  */
 235 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq, int id)
 236 {
 237         struct list_head *tmp;
 238
 239         expunge_all(msq, -EIDRM);
 240         ss_wakeup(&msq->q_senders, 1);
 241         msq = msg_rmid(ns, id);
 242         msg_unlock(msq);
 243
 244         tmp = msq->q_messages.next;
 245         while (tmp != &msq->q_messages) {
 246                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 247
 248                 tmp = tmp->next;
 249                 atomic_dec(&msg_hdrs);
 250                 free_msg(msg);
 251         }
 252         atomic_sub(msq->q_cbytes, &msg_bytes);
 253         security_msg_queue_free(msq);
 254         ipc_rcu_putref(msq);
 255 }
 256
 257 asmlinkage long sys_msgget(key_t key, int msgflg)
 258 {
 259         struct msg_queue *msq;
 260         int id, ret = -EPERM;
 261         struct ipc_namespace *ns;
 262
 263         ns = current->nsproxy->ipc_ns;
 264
 265         mutex_lock(&msg_ids(ns).mutex);
 266         if (key == IPC_PRIVATE)
 267                 ret = newque(ns, key, msgflg);
 268         else if ((id = ipc_findkey(&msg_ids(ns), key)) == -1) { /* key not used */
 269                 if (!(msgflg & IPC_CREAT))
 270                         ret = -ENOENT;
 271                 else
 272                         ret = newque(ns, key, msgflg);
 273         } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
 274                 ret = -EEXIST;
 275         } else {
 276                 msq = msg_lock(ns, id);
 277                 BUG_ON(msq == NULL);
 278                 if (ipcperms(&msq->q_perm, msgflg))
 279                         ret = -EACCES;
 280                 else {
 281                         int qid = msg_buildid(ns, id, msq->q_perm.seq);
 282
 283                         ret = security_msg_queue_associate(msq, msgflg);
 284                         if (!ret)
 285                                 ret = qid;
 286                 }
 287                 msg_unlock(msq);
 288         }
 289         mutex_unlock(&msg_ids(ns).mutex);
 290
 291         return ret;
 292 }
 293
 294 static inline unsigned long
 295 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 296 {
 297         switch(version) {
 298         case IPC_64:
 299                 return copy_to_user(buf, in, sizeof(*in));
 300         case IPC_OLD:
 301         {
 302                 struct msqid_ds out;
 303
 304                 memset(&out, 0, sizeof(out));
 305
 306                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 307
 308                 out.msg_stime           = in->msg_stime;
 309                 out.msg_rtime           = in->msg_rtime;
 310                 out.msg_ctime           = in->msg_ctime;
 311
 312                 if (in->msg_cbytes > USHRT_MAX)
 313                         out.msg_cbytes  = USHRT_MAX;
 314                 else
 315                         out.msg_cbytes  = in->msg_cbytes;
 316                 out.msg_lcbytes         = in->msg_cbytes;
 317
 318                 if (in->msg_qnum > USHRT_MAX)
 319                         out.msg_qnum    = USHRT_MAX;
 320                 else
 321                         out.msg_qnum    = in->msg_qnum;
 322
 323                 if (in->msg_qbytes > USHRT_MAX)
 324                         out.msg_qbytes  = USHRT_MAX;
 325                 else
 326                         out.msg_qbytes  = in->msg_qbytes;
 327                 out.msg_lqbytes         = in->msg_qbytes;
 328
 329                 out.msg_lspid           = in->msg_lspid;
 330                 out.msg_lrpid           = in->msg_lrpid;
 331
 332                 return copy_to_user(buf, &out, sizeof(out));
 333         }
 334         default:
 335                 return -EINVAL;
 336         }
 337 }
 338
 339 struct msq_setbuf {
 340         unsigned long   qbytes;
 341         uid_t           uid;
 342         gid_t           gid;
 343         mode_t          mode;
 344 };
 345
 346 static inline unsigned long
 347 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 348 {
 349         switch(version) {
 350         case IPC_64:
 351         {
 352                 struct msqid64_ds tbuf;
 353
 354                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 355                         return -EFAULT;
 356
 357                 out->qbytes             = tbuf.msg_qbytes;
 358                 out->uid                = tbuf.msg_perm.uid;
 359                 out->gid                = tbuf.msg_perm.gid;
 360                 out->mode               = tbuf.msg_perm.mode;
 361
 362                 return 0;
 363         }
 364         case IPC_OLD:
 365         {
 366                 struct msqid_ds tbuf_old;
 367
 368                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 369                         return -EFAULT;
 370
 371                 out->uid                = tbuf_old.msg_perm.uid;
 372                 out->gid                = tbuf_old.msg_perm.gid;
 373                 out->mode               = tbuf_old.msg_perm.mode;
 374
 375                 if (tbuf_old.msg_qbytes == 0)
 376                         out->qbytes     = tbuf_old.msg_lqbytes;
 377                 else
 378                         out->qbytes     = tbuf_old.msg_qbytes;
 379
 380                 return 0;
 381         }
 382         default:
 383                 return -EINVAL;
 384         }
 385 }
 386
 387 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 388 {
 389         struct kern_ipc_perm *ipcp;
 390         struct msq_setbuf setbuf;
 391         struct msg_queue *msq;
 392         int err, version;
 393         struct ipc_namespace *ns;
 394
 395         if (msqid < 0 || cmd < 0)
 396                 return -EINVAL;
 397
 398         version = ipc_parse_version(&cmd);
 399         ns = current->nsproxy->ipc_ns;
 400
 401         switch (cmd) {
 402         case IPC_INFO:
 403         case MSG_INFO:
 404         {
 405                 struct msginfo msginfo;
 406                 int max_id;
 407
 408                 if (!buf)
 409                         return -EFAULT;
 410                 /*
 411                  * We must not return kernel stack data.
 412                  * due to padding, it's not enough
 413                  * to set all member fields.
 414                  */
 415                 err = security_msg_queue_msgctl(NULL, cmd);
 416                 if (err)
 417                         return err;
 418
 419                 memset(&msginfo, 0, sizeof(msginfo));
 420                 msginfo.msgmni = ns->msg_ctlmni;
 421                 msginfo.msgmax = ns->msg_ctlmax;
 422                 msginfo.msgmnb = ns->msg_ctlmnb;
 423                 msginfo.msgssz = MSGSSZ;
 424                 msginfo.msgseg = MSGSEG;
 425                 mutex_lock(&msg_ids(ns).mutex);
 426                 if (cmd == MSG_INFO) {
 427                         msginfo.msgpool = msg_ids(ns).in_use;
 428                         msginfo.msgmap = atomic_read(&msg_hdrs);
 429                         msginfo.msgtql = atomic_read(&msg_bytes);
 430                 } else {
 431                         msginfo.msgmap = MSGMAP;
 432                         msginfo.msgpool = MSGPOOL;
 433                         msginfo.msgtql = MSGTQL;
 434                 }
 435                 max_id = msg_ids(ns).max_id;
 436                 mutex_unlock(&msg_ids(ns).mutex);
 437                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 438                         return -EFAULT;
 439                 return (max_id < 0) ? 0 : max_id;
 440         }
 441         case MSG_STAT:
 442         case IPC_STAT:
 443         {
 444                 struct msqid64_ds tbuf;
 445                 int success_return;
 446
 447                 if (!buf)
 448                         return -EFAULT;
 449                 if (cmd == MSG_STAT && msqid >= msg_ids(ns).entries->size)
 450                         return -EINVAL;
 451
 452                 memset(&tbuf, 0, sizeof(tbuf));
 453
 454                 msq = msg_lock(ns, msqid);
 455                 if (msq == NULL)
 456                         return -EINVAL;
 457
 458                 if (cmd == MSG_STAT) {
 459                         success_return = msg_buildid(ns, msqid, msq->q_perm.seq);
 460                 } else {
 461                         err = -EIDRM;
 462                         if (msg_checkid(ns, msq, msqid))
 463                                 goto out_unlock;
 464                         success_return = 0;
 465                 }
 466                 err = -EACCES;
 467                 if (ipcperms(&msq->q_perm, S_IRUGO))
 468                         goto out_unlock;
 469
 470                 err = security_msg_queue_msgctl(msq, cmd);
 471                 if (err)
 472                         goto out_unlock;
 473
 474                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 475                 tbuf.msg_stime  = msq->q_stime;
 476                 tbuf.msg_rtime  = msq->q_rtime;
 477                 tbuf.msg_ctime  = msq->q_ctime;
 478                 tbuf.msg_cbytes = msq->q_cbytes;
 479                 tbuf.msg_qnum   = msq->q_qnum;
 480                 tbuf.msg_qbytes = msq->q_qbytes;
 481                 tbuf.msg_lspid  = msq->q_lspid;
 482                 tbuf.msg_lrpid  = msq->q_lrpid;
 483                 msg_unlock(msq);
 484                 if (copy_msqid_to_user(buf, &tbuf, version))
 485                         return -EFAULT;
 486                 return success_return;
 487         }
 488         case IPC_SET:
 489                 if (!buf)
 490                         return -EFAULT;
 491                 if (copy_msqid_from_user(&setbuf, buf, version))
 492                         return -EFAULT;
 493                 break;
 494         case IPC_RMID:
 495                 break;
 496         default:
 497                 return  -EINVAL;
 498         }
 499
 500         mutex_lock(&msg_ids(ns).mutex);
 501         msq = msg_lock(ns, msqid);
 502         err = -EINVAL;
 503         if (msq == NULL)
 504                 goto out_up;
 505
 506         err = -EIDRM;
 507         if (msg_checkid(ns, msq, msqid))
 508                 goto out_unlock_up;
 509         ipcp = &msq->q_perm;
 510
 511         err = audit_ipc_obj(ipcp);
 512         if (err)
 513                 goto out_unlock_up;
 514         if (cmd==IPC_SET) {
 515                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
 516                                          setbuf.mode);
 517                 if (err)
 518                         goto out_unlock_up;
 519         }
 520
 521         err = -EPERM;
 522         if (current->euid != ipcp->cuid &&
 523             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 524                 /* We _could_ check for CAP_CHOWN above, but we don't */
 525                 goto out_unlock_up;
 526
 527         err = security_msg_queue_msgctl(msq, cmd);
 528         if (err)
 529                 goto out_unlock_up;
 530
 531         switch (cmd) {
 532         case IPC_SET:
 533         {
 534                 err = -EPERM;
 535                 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 536                         goto out_unlock_up;
 537
 538                 msq->q_qbytes = setbuf.qbytes;
 539
 540                 ipcp->uid = setbuf.uid;
 541                 ipcp->gid = setbuf.gid;
 542                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 543                              (S_IRWXUGO & setbuf.mode);
 544                 msq->q_ctime = get_seconds();
 545                 /* sleeping receivers might be excluded by
 546                  * stricter permissions.
 547                  */
 548                 expunge_all(msq, -EAGAIN);
 549                 /* sleeping senders might be able to send
 550                  * due to a larger queue size.
 551                  */
 552                 ss_wakeup(&msq->q_senders, 0);
 553                 msg_unlock(msq);
 554                 break;
 555         }
 556         case IPC_RMID:
 557                 freeque(ns, msq, msqid);
 558                 break;
 559         }
 560         err = 0;
 561 out_up:
 562         mutex_unlock(&msg_ids(ns).mutex);
 563         return err;
 564 out_unlock_up:
 565         msg_unlock(msq);
 566         goto out_up;
 567 out_unlock:
 568         msg_unlock(msq);
 569         return err;
 570 }
 571
 572 static int testmsg(struct msg_msg *msg, long type, int mode)
 573 {
 574         switch(mode)
 575         {
 576                 case SEARCH_ANY:
 577                         return 1;
 578                 case SEARCH_LESSEQUAL:
 579                         if (msg->m_type <=type)
 580                                 return 1;
 581                         break;
 582                 case SEARCH_EQUAL:
 583                         if (msg->m_type == type)
 584                                 return 1;
 585                         break;
 586                 case SEARCH_NOTEQUAL:
 587                         if (msg->m_type != type)
 588                                 return 1;
 589                         break;
 590         }
 591         return 0;
 592 }
 593
 594 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 595 {
 596         struct list_head *tmp;
 597
 598         tmp = msq->q_receivers.next;
 599         while (tmp != &msq->q_receivers) {
 600                 struct msg_receiver *msr;
 601
 602                 msr = list_entry(tmp, struct msg_receiver, r_list);
 603                 tmp = tmp->next;
 604                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 605                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 606                                                msr->r_msgtype, msr->r_mode)) {
 607
 608                         list_del(&msr->r_list);
 609 #if 0   // mask by Victor Yu. 03-02-2007
 610                         if (msr->r_maxsize < msg->m_ts) {
 611                                 msr->r_msg = NULL;
 612                                 wake_up_process(msr->r_tsk);
 613                                 smp_mb();
 614                                 msr->r_msg = ERR_PTR(-E2BIG);
 615                         } else {
 616 #endif
 617                                 msr->r_msg = NULL;
 618                                 msq->q_lrpid = msr->r_tsk->pid;
 619                                 msq->q_rtime = get_seconds();
 620                                 wake_up_process(msr->r_tsk);
 621                                 smp_mb();
 622                                 msr->r_msg = msg;
 623
 624                                 return 1;
 625 #if 0   // mask by Victor Yu. 03-02-2007
 626                         }
 627 #endif
 628                 }
 629         }
 630         return 0;
 631 }
 632
 633 asmlinkage long
 634 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 635 {
 636         struct msg_queue *msq;
 637         struct msg_msg *msg;
 638         long mtype;
 639         int err;
 640         struct ipc_namespace *ns;
 641
 642         ns = current->nsproxy->ipc_ns;
 643
 644         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 645                 return -EINVAL;
 646         if (get_user(mtype, &msgp->mtype))
 647                 return -EFAULT;
 648         if (mtype < 1)
 649                 return -EINVAL;
 650
 651         msg = load_msg(msgp->mtext, msgsz);
 652         if (IS_ERR(msg))
 653                 return PTR_ERR(msg);
 654
 655         msg->m_type = mtype;
 656         msg->m_ts = msgsz;
 657
 658         msq = msg_lock(ns, msqid);
 659         err = -EINVAL;
 660         if (msq == NULL)
 661                 goto out_free;
 662
 663         err= -EIDRM;
 664         if (msg_checkid(ns, msq, msqid))
 665                 goto out_unlock_free;
 666
 667         for (;;) {
 668                 struct msg_sender s;
 669
 670                 err = -EACCES;
 671                 if (ipcperms(&msq->q_perm, S_IWUGO))
 672                         goto out_unlock_free;
 673
 674                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 675                 if (err)
 676                         goto out_unlock_free;
 677
 678                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 679                                 1 + msq->q_qnum <= msq->q_qbytes) {
 680                         break;
 681                 }
 682
 683                 /* queue full, wait: */
 684                 if (msgflg & IPC_NOWAIT) {
 685                         err = -EAGAIN;
 686                         goto out_unlock_free;
 687                 }
 688                 ss_add(msq, &s);
 689                 ipc_rcu_getref(msq);
 690                 msg_unlock(msq);
 691                 schedule();
 692
 693                 ipc_lock_by_ptr(&msq->q_perm);
 694                 ipc_rcu_putref(msq);
 695                 if (msq->q_perm.deleted) {
 696                         err = -EIDRM;
 697                         goto out_unlock_free;
 698                 }
 699                 ss_del(&s);
 700
 701                 if (signal_pending(current)) {
 702                         err = -ERESTARTNOHAND;
 703                         goto out_unlock_free;
 704                 }
 705         }
 706
 707         msq->q_lspid = current->tgid;
 708         msq->q_stime = get_seconds();
 709
 710         if (!pipelined_send(msq, msg)) {
 711                 /* noone is waiting for this message, enqueue it */
 712                 list_add_tail(&msg->m_list, &msq->q_messages);
 713                 msq->q_cbytes += msgsz;
 714                 msq->q_qnum++;
 715                 atomic_add(msgsz, &msg_bytes);
 716                 atomic_inc(&msg_hdrs);
 717         }
 718
 719         err = 0;
 720         msg = NULL;
 721
 722 out_unlock_free:
 723         msg_unlock(msq);
 724 out_free:
 725         if (msg != NULL)
 726                 free_msg(msg);
 727         return err;
 728 }
 729
 730 static inline int convert_mode(long *msgtyp, int msgflg)
 731 {
 732         /*
 733          *  find message of correct type.
 734          *  msgtyp = 0 => get first.
 735          *  msgtyp > 0 => get first message of matching type.
 736          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 737          */
 738         if (*msgtyp == 0)
 739                 return SEARCH_ANY;
 740         if (*msgtyp < 0) {
 741                 *msgtyp = -*msgtyp;
 742                 return SEARCH_LESSEQUAL;
 743         }
 744         if (msgflg & MSG_EXCEPT)
 745                 return SEARCH_NOTEQUAL;
 746         return SEARCH_EQUAL;
 747 }
 748
 749 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 750                            long msgtyp, int msgflg)
 751 {
 752         struct msg_queue *msq;
 753         struct msg_msg *msg;
 754         int mode;
 755         struct ipc_namespace *ns;
 756
 757         if (msqid < 0 || (long) msgsz < 0)
 758                 return -EINVAL;
 759         mode = convert_mode(&msgtyp, msgflg);
 760         ns = current->nsproxy->ipc_ns;
 761
 762         msq = msg_lock(ns, msqid);
 763         if (msq == NULL)
 764                 return -EINVAL;
 765
 766         msg = ERR_PTR(-EIDRM);
 767         if (msg_checkid(ns, msq, msqid))
 768                 goto out_unlock;
 769
 770         for (;;) {
 771                 struct msg_receiver msr_d;
 772                 struct list_head *tmp;
 773
 774                 msg = ERR_PTR(-EACCES);
 775                 if (ipcperms(&msq->q_perm, S_IRUGO))
 776                         goto out_unlock;
 777
 778                 msg = ERR_PTR(-EAGAIN);
 779                 tmp = msq->q_messages.next;
 780                 while (tmp != &msq->q_messages) {
 781                         struct msg_msg *walk_msg;
 782
 783                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 784                         if (testmsg(walk_msg, msgtyp, mode) &&
 785                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 786                                                        msgtyp, mode)) {
 787
 788                                 msg = walk_msg;
 789                                 if (mode == SEARCH_LESSEQUAL &&
 790                                                 walk_msg->m_type != 1) {
 791                                         msg = walk_msg;
 792                                         msgtyp = walk_msg->m_type - 1;
 793                                 } else {
 794                                         msg = walk_msg;
 795                                         break;
 796                                 }
 797                         }
 798                         tmp = tmp->next;
 799                 }
 800                 if (!IS_ERR(msg)) {
 801                         /*
 802                          * Found a suitable message.
 803                          * Unlink it from the queue.
 804                          */
 805                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 806                                 msg = ERR_PTR(-E2BIG);
 807                                 goto out_unlock;
 808                         }
 809                         list_del(&msg->m_list);
 810                         msq->q_qnum--;
 811                         msq->q_rtime = get_seconds();
 812                         msq->q_lrpid = current->tgid;
 813                         msq->q_cbytes -= msg->m_ts;
 814                         atomic_sub(msg->m_ts, &msg_bytes);
 815                         atomic_dec(&msg_hdrs);
 816                         ss_wakeup(&msq->q_senders, 0);
 817                         msg_unlock(msq);
 818                         break;
 819                 }
 820                 /* No message waiting. Wait for a message */
 821                 if (msgflg & IPC_NOWAIT) {
 822                         msg = ERR_PTR(-ENOMSG);
 823                         goto out_unlock;
 824                 }
 825                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 826                 msr_d.r_tsk = current;
 827                 msr_d.r_msgtype = msgtyp;
 828                 msr_d.r_mode = mode;
 829                 if (msgflg & MSG_NOERROR)
 830                         msr_d.r_maxsize = INT_MAX;
 831                 else
 832                         msr_d.r_maxsize = msgsz;
 833                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 834                 current->state = TASK_INTERRUPTIBLE;
 835                 msg_unlock(msq);
 836
 837                 schedule();
 838
 839                 /* Lockless receive, part 1:
 840                  * Disable preemption.  We don't hold a reference to the queue
 841                  * and getting a reference would defeat the idea of a lockless
 842                  * operation, thus the code relies on rcu to guarantee the
 843                  * existance of msq:
 844                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 845                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 846                  * rcu_read_lock() prevents preemption between reading r_msg
 847                  * and the spin_lock() inside ipc_lock_by_ptr().
 848                  */
 849                 rcu_read_lock();
 850
 851                 /* Lockless receive, part 2:
 852                  * Wait until pipelined_send or expunge_all are outside of
 853                  * wake_up_process(). There is a race with exit(), see
 854                  * ipc/mqueue.c for the details.
 855                  */
 856                 msg = (struct msg_msg*)msr_d.r_msg;
 857                 while (msg == NULL) {
 858                         cpu_relax();
 859                         msg = (struct msg_msg *)msr_d.r_msg;
 860                 }
 861
 862                 /* Lockless receive, part 3:
 863                  * If there is a message or an error then accept it without
 864                  * locking.
 865                  */
 866                 if (msg != ERR_PTR(-EAGAIN)) {
 867                         rcu_read_unlock();
 868                         break;
 869                 }
 870
 871                 /* Lockless receive, part 3:
 872                  * Acquire the queue spinlock.
 873                  */
 874                 ipc_lock_by_ptr(&msq->q_perm);
 875                 rcu_read_unlock();
 876
 877                 /* Lockless receive, part 4:
 878                  * Repeat test after acquiring the spinlock.
 879                  */
 880                 msg = (struct msg_msg*)msr_d.r_msg;
 881                 if (msg != ERR_PTR(-EAGAIN))
 882                         goto out_unlock;
 883
 884                 list_del(&msr_d.r_list);
 885                 if (signal_pending(current)) {
 886                         msg = ERR_PTR(-ERESTARTNOHAND);
 887 out_unlock:
 888                         msg_unlock(msq);
 889                         break;
 890                 }
 891         }
 892         if (IS_ERR(msg))
 893                 return PTR_ERR(msg);
 894
 895         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 896         if (put_user (msg->m_type, &msgp->mtype) ||
 897             store_msg(msgp->mtext, msg, msgsz)) {
 898                 msgsz = -EFAULT;
 899         }
 900         free_msg(msg);
 901
 902         return msgsz;
 903 }
 904
 905 #ifdef CONFIG_PROC_FS
 906 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 907 {
 908         struct msg_queue *msq = it;
 909
 910         return seq_printf(s,
 911                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 912                         msq->q_perm.key,
 913                         msq->q_id,
 914                         msq->q_perm.mode,
 915                         msq->q_cbytes,
 916                         msq->q_qnum,
 917                         msq->q_lspid,
 918                         msq->q_lrpid,
 919                         msq->q_perm.uid,
 920                         msq->q_perm.gid,
 921                         msq->q_perm.cuid,
 922                         msq->q_perm.cgid,
 923                         msq->q_stime,
 924                         msq->q_rtime,
 925                         msq->q_ctime);
 926 }
 927 #endif