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/msg.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/init.h>
  29 #include <linux/mm.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/rwsem.h>
  38 #include <linux/nsproxy.h>
  39 #include <linux/ipc_namespace.h>
  40
  41 #include <asm/current.h>
  42 #include <asm/uaccess.h>
  43 #include "util.h"
  44
  45 /*
  46  * one msg_receiver structure for each sleeping receiver:
  47  */
  48 struct msg_receiver {
  49         struct list_head        r_list;
  50         struct task_struct      *r_tsk;
  51
  52         int                     r_mode;
  53         long                    r_msgtype;
  54         long                    r_maxsize;
  55
  56         struct msg_msg          *volatile r_msg;
  57 };
  58
  59 /* one msg_sender for each sleeping sender */
  60 struct msg_sender {
  61         struct list_head        list;
  62         struct task_struct      *tsk;
  63 };
  64
  65 #define SEARCH_ANY              1
  66 #define SEARCH_EQUAL            2
  67 #define SEARCH_NOTEQUAL         3
  68 #define SEARCH_LESSEQUAL        4
  69
  70 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  71
  72 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  73
  74 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  75 static int newque(struct ipc_namespace *, struct ipc_params *);
  76 #ifdef CONFIG_PROC_FS
  77 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  78 #endif
  79
  80 /*
  81  * Scale msgmni with the available lowmem size: the memory dedicated to msg
  82  * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  83  * Also take into account the number of nsproxies created so far.
  84  * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  85  */
  86 void recompute_msgmni(struct ipc_namespace *ns)
  87 {
  88         struct sysinfo i;
  89         unsigned long allowed;
  90         int nb_ns;
  91
  92         si_meminfo(&i);
  93         allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  94                 / MSGMNB;
  95         nb_ns = atomic_read(&nr_ipc_ns);
  96         allowed /= nb_ns;
  97
  98         if (allowed < MSGMNI) {
  99                 ns->msg_ctlmni = MSGMNI;
 100                 return;
 101         }
 102
 103         if (allowed > IPCMNI / nb_ns) {
 104                 ns->msg_ctlmni = IPCMNI / nb_ns;
 105                 return;
 106         }
 107
 108         ns->msg_ctlmni = allowed;
 109 }
 110
 111 void msg_init_ns(struct ipc_namespace *ns)
 112 {
 113         ns->msg_ctlmax = MSGMAX;
 114         ns->msg_ctlmnb = MSGMNB;
 115
 116         recompute_msgmni(ns);
 117
 118         atomic_set(&ns->msg_bytes, 0);
 119         atomic_set(&ns->msg_hdrs, 0);
 120         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 121 }
 122
 123 #ifdef CONFIG_IPC_NS
 124 void msg_exit_ns(struct ipc_namespace *ns)
 125 {
 126         free_ipcs(ns, &msg_ids(ns), freeque);
 127         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
 128 }
 129 #endif
 130
 131 void __init msg_init(void)
 132 {
 133         msg_init_ns(&init_ipc_ns);
 134
 135         printk(KERN_INFO "msgmni has been set to %d\n",
 136                 init_ipc_ns.msg_ctlmni);
 137
 138         ipc_init_proc_interface("sysvipc/msg",
 139                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 140                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 141 }
 142
 143 /*
 144  * msg_lock_(check_) routines are called in the paths where the rw_mutex
 145  * is not held.
 146  */
 147 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
 148 {
 149         struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
 150
 151         if (IS_ERR(ipcp))
 152                 return (struct msg_queue *)ipcp;
 153
 154         return container_of(ipcp, struct msg_queue, q_perm);
 155 }
 156
 157 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
 158                                                 int id)
 159 {
 160         struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
 161
 162         if (IS_ERR(ipcp))
 163                 return (struct msg_queue *)ipcp;
 164
 165         return container_of(ipcp, struct msg_queue, q_perm);
 166 }
 167
 168 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 169 {
 170         ipc_rmid(&msg_ids(ns), &s->q_perm);
 171 }
 172
 173 /**
 174  * newque - Create a new msg queue
 175  * @ns: namespace
 176  * @params: ptr to the structure that contains the key and msgflg
 177  *
 178  * Called with msg_ids.rw_mutex held (writer)
 179  */
 180 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 181 {
 182         struct msg_queue *msq;
 183         int id, retval;
 184         key_t key = params->key;
 185         int msgflg = params->flg;
 186
 187         msq = ipc_rcu_alloc(sizeof(*msq));
 188         if (!msq)
 189                 return -ENOMEM;
 190
 191         msq->q_perm.mode = msgflg & S_IRWXUGO;
 192         msq->q_perm.key = key;
 193
 194         msq->q_perm.security = NULL;
 195         retval = security_msg_queue_alloc(msq);
 196         if (retval) {
 197                 ipc_rcu_putref(msq);
 198                 return retval;
 199         }
 200
 201         /*
 202          * ipc_addid() locks msq
 203          */
 204         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 205         if (id < 0) {
 206                 security_msg_queue_free(msq);
 207                 ipc_rcu_putref(msq);
 208                 return id;
 209         }
 210
 211         msq->q_stime = msq->q_rtime = 0;
 212         msq->q_ctime = get_seconds();
 213         msq->q_cbytes = msq->q_qnum = 0;
 214         msq->q_qbytes = ns->msg_ctlmnb;
 215         msq->q_lspid = msq->q_lrpid = 0;
 216         INIT_LIST_HEAD(&msq->q_messages);
 217         INIT_LIST_HEAD(&msq->q_receivers);
 218         INIT_LIST_HEAD(&msq->q_senders);
 219
 220         msg_unlock(msq);
 221
 222         return msq->q_perm.id;
 223 }
 224
 225 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 226 {
 227         mss->tsk = current;
 228         current->state = TASK_INTERRUPTIBLE;
 229         list_add_tail(&mss->list, &msq->q_senders);
 230 }
 231
 232 static inline void ss_del(struct msg_sender *mss)
 233 {
 234         if (mss->list.next != NULL)
 235                 list_del(&mss->list);
 236 }
 237
 238 static void ss_wakeup(struct list_head *h, int kill)
 239 {
 240         struct list_head *tmp;
 241
 242         tmp = h->next;
 243         while (tmp != h) {
 244                 struct msg_sender *mss;
 245
 246                 mss = list_entry(tmp, struct msg_sender, list);
 247                 tmp = tmp->next;
 248                 if (kill)
 249                         mss->list.next = NULL;
 250                 wake_up_process(mss->tsk);
 251         }
 252 }
 253
 254 static void expunge_all(struct msg_queue *msq, int res)
 255 {
 256         struct list_head *tmp;
 257
 258         tmp = msq->q_receivers.next;
 259         while (tmp != &msq->q_receivers) {
 260                 struct msg_receiver *msr;
 261
 262                 msr = list_entry(tmp, struct msg_receiver, r_list);
 263                 tmp = tmp->next;
 264                 msr->r_msg = NULL;
 265                 wake_up_process(msr->r_tsk);
 266                 smp_mb();
 267                 msr->r_msg = ERR_PTR(res);
 268         }
 269 }
 270
 271 /*
 272  * freeque() wakes up waiters on the sender and receiver waiting queue,
 273  * removes the message queue from message queue ID IDR, and cleans up all the
 274  * messages associated with this queue.
 275  *
 276  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 277  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 278  */
 279 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 280 {
 281         struct list_head *tmp;
 282         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 283
 284         expunge_all(msq, -EIDRM);
 285         ss_wakeup(&msq->q_senders, 1);
 286         msg_rmid(ns, msq);
 287         msg_unlock(msq);
 288
 289         tmp = msq->q_messages.next;
 290         while (tmp != &msq->q_messages) {
 291                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 292
 293                 tmp = tmp->next;
 294                 atomic_dec(&ns->msg_hdrs);
 295                 free_msg(msg);
 296         }
 297         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 298         security_msg_queue_free(msq);
 299         ipc_rcu_putref(msq);
 300 }
 301
 302 /*
 303  * Called with msg_ids.rw_mutex and ipcp locked.
 304  */
 305 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 306 {
 307         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 308
 309         return security_msg_queue_associate(msq, msgflg);
 310 }
 311
 312 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 313 {
 314         struct ipc_namespace *ns;
 315         struct ipc_ops msg_ops;
 316         struct ipc_params msg_params;
 317
 318         ns = current->nsproxy->ipc_ns;
 319
 320         msg_ops.getnew = newque;
 321         msg_ops.associate = msg_security;
 322         msg_ops.more_checks = NULL;
 323
 324         msg_params.key = key;
 325         msg_params.flg = msgflg;
 326
 327         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 328 }
 329
 330 static inline unsigned long
 331 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 332 {
 333         switch(version) {
 334         case IPC_64:
 335                 return copy_to_user(buf, in, sizeof(*in));
 336         case IPC_OLD:
 337         {
 338                 struct msqid_ds out;
 339
 340                 memset(&out, 0, sizeof(out));
 341
 342                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 343
 344                 out.msg_stime           = in->msg_stime;
 345                 out.msg_rtime           = in->msg_rtime;
 346                 out.msg_ctime           = in->msg_ctime;
 347
 348                 if (in->msg_cbytes > USHRT_MAX)
 349                         out.msg_cbytes  = USHRT_MAX;
 350                 else
 351                         out.msg_cbytes  = in->msg_cbytes;
 352                 out.msg_lcbytes         = in->msg_cbytes;
 353
 354                 if (in->msg_qnum > USHRT_MAX)
 355                         out.msg_qnum    = USHRT_MAX;
 356                 else
 357                         out.msg_qnum    = in->msg_qnum;
 358
 359                 if (in->msg_qbytes > USHRT_MAX)
 360                         out.msg_qbytes  = USHRT_MAX;
 361                 else
 362                         out.msg_qbytes  = in->msg_qbytes;
 363                 out.msg_lqbytes         = in->msg_qbytes;
 364
 365                 out.msg_lspid           = in->msg_lspid;
 366                 out.msg_lrpid           = in->msg_lrpid;
 367
 368                 return copy_to_user(buf, &out, sizeof(out));
 369         }
 370         default:
 371                 return -EINVAL;
 372         }
 373 }
 374
 375 static inline unsigned long
 376 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 377 {
 378         switch(version) {
 379         case IPC_64:
 380                 if (copy_from_user(out, buf, sizeof(*out)))
 381                         return -EFAULT;
 382                 return 0;
 383         case IPC_OLD:
 384         {
 385                 struct msqid_ds tbuf_old;
 386
 387                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 388                         return -EFAULT;
 389
 390                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 391                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 392                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 393
 394                 if (tbuf_old.msg_qbytes == 0)
 395                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 396                 else
 397                         out->msg_qbytes = tbuf_old.msg_qbytes;
 398
 399                 return 0;
 400         }
 401         default:
 402                 return -EINVAL;
 403         }
 404 }
 405
 406 /*
 407  * This function handles some msgctl commands which require the rw_mutex
 408  * to be held in write mode.
 409  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 410  */
 411 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 412                        struct msqid_ds __user *buf, int version)
 413 {
 414         struct kern_ipc_perm *ipcp;
 415         struct msqid64_ds uninitialized_var(msqid64);
 416         struct msg_queue *msq;
 417         int err;
 418
 419         if (cmd == IPC_SET) {
 420                 if (copy_msqid_from_user(&msqid64, buf, version))
 421                         return -EFAULT;
 422         }
 423
 424         ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
 425                                &msqid64.msg_perm, msqid64.msg_qbytes);
 426         if (IS_ERR(ipcp))
 427                 return PTR_ERR(ipcp);
 428
 429         msq = container_of(ipcp, struct msg_queue, q_perm);
 430
 431         err = security_msg_queue_msgctl(msq, cmd);
 432         if (err)
 433                 goto out_unlock;
 434
 435         switch (cmd) {
 436         case IPC_RMID:
 437                 freeque(ns, ipcp);
 438                 goto out_up;
 439         case IPC_SET:
 440                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 441                     !capable(CAP_SYS_RESOURCE)) {
 442                         err = -EPERM;
 443                         goto out_unlock;
 444                 }
 445
 446                 msq->q_qbytes = msqid64.msg_qbytes;
 447
 448                 ipc_update_perm(&msqid64.msg_perm, ipcp);
 449                 msq->q_ctime = get_seconds();
 450                 /* sleeping receivers might be excluded by
 451                  * stricter permissions.
 452                  */
 453                 expunge_all(msq, -EAGAIN);
 454                 /* sleeping senders might be able to send
 455                  * due to a larger queue size.
 456                  */
 457                 ss_wakeup(&msq->q_senders, 0);
 458                 break;
 459         default:
 460                 err = -EINVAL;
 461         }
 462 out_unlock:
 463         msg_unlock(msq);
 464 out_up:
 465         up_write(&msg_ids(ns).rw_mutex);
 466         return err;
 467 }
 468
 469 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 470 {
 471         struct msg_queue *msq;
 472         int err, version;
 473         struct ipc_namespace *ns;
 474
 475         if (msqid < 0 || cmd < 0)
 476                 return -EINVAL;
 477
 478         version = ipc_parse_version(&cmd);
 479         ns = current->nsproxy->ipc_ns;
 480
 481         switch (cmd) {
 482         case IPC_INFO:
 483         case MSG_INFO:
 484         {
 485                 struct msginfo msginfo;
 486                 int max_id;
 487
 488                 if (!buf)
 489                         return -EFAULT;
 490                 /*
 491                  * We must not return kernel stack data.
 492                  * due to padding, it's not enough
 493                  * to set all member fields.
 494                  */
 495                 err = security_msg_queue_msgctl(NULL, cmd);
 496                 if (err)
 497                         return err;
 498
 499                 memset(&msginfo, 0, sizeof(msginfo));
 500                 msginfo.msgmni = ns->msg_ctlmni;
 501                 msginfo.msgmax = ns->msg_ctlmax;
 502                 msginfo.msgmnb = ns->msg_ctlmnb;
 503                 msginfo.msgssz = MSGSSZ;
 504                 msginfo.msgseg = MSGSEG;
 505                 down_read(&msg_ids(ns).rw_mutex);
 506                 if (cmd == MSG_INFO) {
 507                         msginfo.msgpool = msg_ids(ns).in_use;
 508                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 509                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 510                 } else {
 511                         msginfo.msgmap = MSGMAP;
 512                         msginfo.msgpool = MSGPOOL;
 513                         msginfo.msgtql = MSGTQL;
 514                 }
 515                 max_id = ipc_get_maxid(&msg_ids(ns));
 516                 up_read(&msg_ids(ns).rw_mutex);
 517                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 518                         return -EFAULT;
 519                 return (max_id < 0) ? 0 : max_id;
 520         }
 521         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 522         case IPC_STAT:
 523         {
 524                 struct msqid64_ds tbuf;
 525                 int success_return;
 526
 527                 if (!buf)
 528                         return -EFAULT;
 529
 530                 if (cmd == MSG_STAT) {
 531                         msq = msg_lock(ns, msqid);
 532                         if (IS_ERR(msq))
 533                                 return PTR_ERR(msq);
 534                         success_return = msq->q_perm.id;
 535                 } else {
 536                         msq = msg_lock_check(ns, msqid);
 537                         if (IS_ERR(msq))
 538                                 return PTR_ERR(msq);
 539                         success_return = 0;
 540                 }
 541                 err = -EACCES;
 542                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 543                         goto out_unlock;
 544
 545                 err = security_msg_queue_msgctl(msq, cmd);
 546                 if (err)
 547                         goto out_unlock;
 548
 549                 memset(&tbuf, 0, sizeof(tbuf));
 550
 551                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 552                 tbuf.msg_stime  = msq->q_stime;
 553                 tbuf.msg_rtime  = msq->q_rtime;
 554                 tbuf.msg_ctime  = msq->q_ctime;
 555                 tbuf.msg_cbytes = msq->q_cbytes;
 556                 tbuf.msg_qnum   = msq->q_qnum;
 557                 tbuf.msg_qbytes = msq->q_qbytes;
 558                 tbuf.msg_lspid  = msq->q_lspid;
 559                 tbuf.msg_lrpid  = msq->q_lrpid;
 560                 msg_unlock(msq);
 561                 if (copy_msqid_to_user(buf, &tbuf, version))
 562                         return -EFAULT;
 563                 return success_return;
 564         }
 565         case IPC_SET:
 566         case IPC_RMID:
 567                 err = msgctl_down(ns, msqid, cmd, buf, version);
 568                 return err;
 569         default:
 570                 return  -EINVAL;
 571         }
 572
 573 out_unlock:
 574         msg_unlock(msq);
 575         return err;
 576 }
 577
 578 static int testmsg(struct msg_msg *msg, long type, int mode)
 579 {
 580         switch(mode)
 581         {
 582                 case SEARCH_ANY:
 583                         return 1;
 584                 case SEARCH_LESSEQUAL:
 585                         if (msg->m_type <=type)
 586                                 return 1;
 587                         break;
 588                 case SEARCH_EQUAL:
 589                         if (msg->m_type == type)
 590                                 return 1;
 591                         break;
 592                 case SEARCH_NOTEQUAL:
 593                         if (msg->m_type != type)
 594                                 return 1;
 595                         break;
 596         }
 597         return 0;
 598 }
 599
 600 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 601 {
 602         struct list_head *tmp;
 603
 604         tmp = msq->q_receivers.next;
 605         while (tmp != &msq->q_receivers) {
 606                 struct msg_receiver *msr;
 607
 608                 msr = list_entry(tmp, struct msg_receiver, r_list);
 609                 tmp = tmp->next;
 610                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 611                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 612                                                msr->r_msgtype, msr->r_mode)) {
 613
 614                         list_del(&msr->r_list);
 615                         if (msr->r_maxsize < msg->m_ts) {
 616                                 msr->r_msg = NULL;
 617                                 wake_up_process(msr->r_tsk);
 618                                 smp_mb();
 619                                 msr->r_msg = ERR_PTR(-E2BIG);
 620                         } else {
 621                                 msr->r_msg = NULL;
 622                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 623                                 msq->q_rtime = get_seconds();
 624                                 wake_up_process(msr->r_tsk);
 625                                 smp_mb();
 626                                 msr->r_msg = msg;
 627
 628                                 return 1;
 629                         }
 630                 }
 631         }
 632         return 0;
 633 }
 634
 635 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 636                 size_t msgsz, int msgflg)
 637 {
 638         struct msg_queue *msq;
 639         struct msg_msg *msg;
 640         int err;
 641         struct ipc_namespace *ns;
 642
 643         ns = current->nsproxy->ipc_ns;
 644
 645         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 646                 return -EINVAL;
 647         if (mtype < 1)
 648                 return -EINVAL;
 649
 650         msg = load_msg(mtext, msgsz);
 651         if (IS_ERR(msg))
 652                 return PTR_ERR(msg);
 653
 654         msg->m_type = mtype;
 655         msg->m_ts = msgsz;
 656
 657         msq = msg_lock_check(ns, msqid);
 658         if (IS_ERR(msq)) {
 659                 err = PTR_ERR(msq);
 660                 goto out_free;
 661         }
 662
 663         for (;;) {
 664                 struct msg_sender s;
 665
 666                 err = -EACCES;
 667                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 668                         goto out_unlock_free;
 669
 670                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 671                 if (err)
 672                         goto out_unlock_free;
 673
 674                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 675                                 1 + msq->q_qnum <= msq->q_qbytes) {
 676                         break;
 677                 }
 678
 679                 /* queue full, wait: */
 680                 if (msgflg & IPC_NOWAIT) {
 681                         err = -EAGAIN;
 682                         goto out_unlock_free;
 683                 }
 684                 ss_add(msq, &s);
 685                 ipc_rcu_getref(msq);
 686                 msg_unlock(msq);
 687                 schedule();
 688
 689                 ipc_lock_by_ptr(&msq->q_perm);
 690                 ipc_rcu_putref(msq);
 691                 if (msq->q_perm.deleted) {
 692                         err = -EIDRM;
 693                         goto out_unlock_free;
 694                 }
 695                 ss_del(&s);
 696
 697                 if (signal_pending(current)) {
 698                         err = -ERESTARTNOHAND;
 699                         goto out_unlock_free;
 700                 }
 701         }
 702
 703         msq->q_lspid = task_tgid_vnr(current);
 704         msq->q_stime = get_seconds();
 705
 706         if (!pipelined_send(msq, msg)) {
 707                 /* no one is waiting for this message, enqueue it */
 708                 list_add_tail(&msg->m_list, &msq->q_messages);
 709                 msq->q_cbytes += msgsz;
 710                 msq->q_qnum++;
 711                 atomic_add(msgsz, &ns->msg_bytes);
 712                 atomic_inc(&ns->msg_hdrs);
 713         }
 714
 715         err = 0;
 716         msg = NULL;
 717
 718 out_unlock_free:
 719         msg_unlock(msq);
 720 out_free:
 721         if (msg != NULL)
 722                 free_msg(msg);
 723         return err;
 724 }
 725
 726 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 727                 int, msgflg)
 728 {
 729         long mtype;
 730
 731         if (get_user(mtype, &msgp->mtype))
 732                 return -EFAULT;
 733         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 734 }
 735
 736 static inline int convert_mode(long *msgtyp, int msgflg)
 737 {
 738         /*
 739          *  find message of correct type.
 740          *  msgtyp = 0 => get first.
 741          *  msgtyp > 0 => get first message of matching type.
 742          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 743          */
 744         if (*msgtyp == 0)
 745                 return SEARCH_ANY;
 746         if (*msgtyp < 0) {
 747                 *msgtyp = -*msgtyp;
 748                 return SEARCH_LESSEQUAL;
 749         }
 750         if (msgflg & MSG_EXCEPT)
 751                 return SEARCH_NOTEQUAL;
 752         return SEARCH_EQUAL;
 753 }
 754
 755 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
 756                 size_t msgsz, long msgtyp, int msgflg)
 757 {
 758         struct msg_queue *msq;
 759         struct msg_msg *msg;
 760         int mode;
 761         struct ipc_namespace *ns;
 762
 763         if (msqid < 0 || (long) msgsz < 0)
 764                 return -EINVAL;
 765         mode = convert_mode(&msgtyp, msgflg);
 766         ns = current->nsproxy->ipc_ns;
 767
 768         msq = msg_lock_check(ns, msqid);
 769         if (IS_ERR(msq))
 770                 return PTR_ERR(msq);
 771
 772         for (;;) {
 773                 struct msg_receiver msr_d;
 774                 struct list_head *tmp;
 775
 776                 msg = ERR_PTR(-EACCES);
 777                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 778                         goto out_unlock;
 779
 780                 msg = ERR_PTR(-EAGAIN);
 781                 tmp = msq->q_messages.next;
 782                 while (tmp != &msq->q_messages) {
 783                         struct msg_msg *walk_msg;
 784
 785                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 786                         if (testmsg(walk_msg, msgtyp, mode) &&
 787                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 788                                                        msgtyp, mode)) {
 789
 790                                 msg = walk_msg;
 791                                 if (mode == SEARCH_LESSEQUAL &&
 792                                                 walk_msg->m_type != 1) {
 793                                         msg = walk_msg;
 794                                         msgtyp = walk_msg->m_type - 1;
 795                                 } else {
 796                                         msg = walk_msg;
 797                                         break;
 798                                 }
 799                         }
 800                         tmp = tmp->next;
 801                 }
 802                 if (!IS_ERR(msg)) {
 803                         /*
 804                          * Found a suitable message.
 805                          * Unlink it from the queue.
 806                          */
 807                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 808                                 msg = ERR_PTR(-E2BIG);
 809                                 goto out_unlock;
 810                         }
 811                         list_del(&msg->m_list);
 812                         msq->q_qnum--;
 813                         msq->q_rtime = get_seconds();
 814                         msq->q_lrpid = task_tgid_vnr(current);
 815                         msq->q_cbytes -= msg->m_ts;
 816                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 817                         atomic_dec(&ns->msg_hdrs);
 818                         ss_wakeup(&msq->q_senders, 0);
 819                         msg_unlock(msq);
 820                         break;
 821                 }
 822                 /* No message waiting. Wait for a message */
 823                 if (msgflg & IPC_NOWAIT) {
 824                         msg = ERR_PTR(-ENOMSG);
 825                         goto out_unlock;
 826                 }
 827                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 828                 msr_d.r_tsk = current;
 829                 msr_d.r_msgtype = msgtyp;
 830                 msr_d.r_mode = mode;
 831                 if (msgflg & MSG_NOERROR)
 832                         msr_d.r_maxsize = INT_MAX;
 833                 else
 834                         msr_d.r_maxsize = msgsz;
 835                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 836                 current->state = TASK_INTERRUPTIBLE;
 837                 msg_unlock(msq);
 838
 839                 schedule();
 840
 841                 /* Lockless receive, part 1:
 842                  * Disable preemption.  We don't hold a reference to the queue
 843                  * and getting a reference would defeat the idea of a lockless
 844                  * operation, thus the code relies on rcu to guarantee the
 845                  * existence of msq:
 846                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 847                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 848                  * rcu_read_lock() prevents preemption between reading r_msg
 849                  * and the spin_lock() inside ipc_lock_by_ptr().
 850                  */
 851                 rcu_read_lock();
 852
 853                 /* Lockless receive, part 2:
 854                  * Wait until pipelined_send or expunge_all are outside of
 855                  * wake_up_process(). There is a race with exit(), see
 856                  * ipc/mqueue.c for the details.
 857                  */
 858                 msg = (struct msg_msg*)msr_d.r_msg;
 859                 while (msg == NULL) {
 860                         cpu_relax();
 861                         msg = (struct msg_msg *)msr_d.r_msg;
 862                 }
 863
 864                 /* Lockless receive, part 3:
 865                  * If there is a message or an error then accept it without
 866                  * locking.
 867                  */
 868                 if (msg != ERR_PTR(-EAGAIN)) {
 869                         rcu_read_unlock();
 870                         break;
 871                 }
 872
 873                 /* Lockless receive, part 3:
 874                  * Acquire the queue spinlock.
 875                  */
 876                 ipc_lock_by_ptr(&msq->q_perm);
 877                 rcu_read_unlock();
 878
 879                 /* Lockless receive, part 4:
 880                  * Repeat test after acquiring the spinlock.
 881                  */
 882                 msg = (struct msg_msg*)msr_d.r_msg;
 883                 if (msg != ERR_PTR(-EAGAIN))
 884                         goto out_unlock;
 885
 886                 list_del(&msr_d.r_list);
 887                 if (signal_pending(current)) {
 888                         msg = ERR_PTR(-ERESTARTNOHAND);
 889 out_unlock:
 890                         msg_unlock(msq);
 891                         break;
 892                 }
 893         }
 894         if (IS_ERR(msg))
 895                 return PTR_ERR(msg);
 896
 897         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 898         *pmtype = msg->m_type;
 899         if (store_msg(mtext, msg, msgsz))
 900                 msgsz = -EFAULT;
 901
 902         free_msg(msg);
 903
 904         return msgsz;
 905 }
 906
 907 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 908                 long, msgtyp, int, msgflg)
 909 {
 910         long err, mtype;
 911
 912         err =  do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
 913         if (err < 0)
 914                 goto out;
 915
 916         if (put_user(mtype, &msgp->mtype))
 917                 err = -EFAULT;
 918 out:
 919         return err;
 920 }
 921
 922 #ifdef CONFIG_PROC_FS
 923 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 924 {
 925         struct msg_queue *msq = it;
 926
 927         return seq_printf(s,
 928                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 929                         msq->q_perm.key,
 930                         msq->q_perm.id,
 931                         msq->q_perm.mode,
 932                         msq->q_cbytes,
 933                         msq->q_qnum,
 934                         msq->q_lspid,
 935                         msq->q_lrpid,
 936                         msq->q_perm.uid,
 937                         msq->q_perm.gid,
 938                         msq->q_perm.cuid,
 939                         msq->q_perm.cgid,
 940                         msq->q_stime,
 941                         msq->q_rtime,
 942                         msq->q_ctime);
 943 }
 944 #endif