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 #define SEARCH_NUMBER           5
  70
  71 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  72
  73 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  74
  75 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  76 static int newque(struct ipc_namespace *, struct ipc_params *);
  77 #ifdef CONFIG_PROC_FS
  78 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  79 #endif
  80
  81 /*
  82  * Scale msgmni with the available lowmem size: the memory dedicated to msg
  83  * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  84  * Also take into account the number of nsproxies created so far.
  85  * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  86  */
  87 void recompute_msgmni(struct ipc_namespace *ns)
  88 {
  89         struct sysinfo i;
  90         unsigned long allowed;
  91         int nb_ns;
  92
  93         si_meminfo(&i);
  94         allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  95                 / MSGMNB;
  96         nb_ns = atomic_read(&nr_ipc_ns);
  97         allowed /= nb_ns;
  98
  99         if (allowed < MSGMNI) {
 100                 ns->msg_ctlmni = MSGMNI;
 101                 return;
 102         }
 103
 104         if (allowed > IPCMNI / nb_ns) {
 105                 ns->msg_ctlmni = IPCMNI / nb_ns;
 106                 return;
 107         }
 108
 109         ns->msg_ctlmni = allowed;
 110 }
 111
 112 void msg_init_ns(struct ipc_namespace *ns)
 113 {
 114         ns->msg_ctlmax = MSGMAX;
 115         ns->msg_ctlmnb = MSGMNB;
 116
 117         recompute_msgmni(ns);
 118
 119         atomic_set(&ns->msg_bytes, 0);
 120         atomic_set(&ns->msg_hdrs, 0);
 121         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 122 }
 123
 124 #ifdef CONFIG_IPC_NS
 125 void msg_exit_ns(struct ipc_namespace *ns)
 126 {
 127         free_ipcs(ns, &msg_ids(ns), freeque);
 128         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
 129 }
 130 #endif
 131
 132 void __init msg_init(void)
 133 {
 134         msg_init_ns(&init_ipc_ns);
 135
 136         printk(KERN_INFO "msgmni has been set to %d\n",
 137                 init_ipc_ns.msg_ctlmni);
 138
 139         ipc_init_proc_interface("sysvipc/msg",
 140                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 141                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 142 }
 143
 144 /*
 145  * msg_lock_(check_) routines are called in the paths where the rw_mutex
 146  * is not held.
 147  */
 148 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
 149 {
 150         struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
 151
 152         if (IS_ERR(ipcp))
 153                 return (struct msg_queue *)ipcp;
 154
 155         return container_of(ipcp, struct msg_queue, q_perm);
 156 }
 157
 158 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
 159                                                 int id)
 160 {
 161         struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
 162
 163         if (IS_ERR(ipcp))
 164                 return (struct msg_queue *)ipcp;
 165
 166         return container_of(ipcp, struct msg_queue, q_perm);
 167 }
 168
 169 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 170 {
 171         ipc_rmid(&msg_ids(ns), &s->q_perm);
 172 }
 173
 174 /**
 175  * newque - Create a new msg queue
 176  * @ns: namespace
 177  * @params: ptr to the structure that contains the key and msgflg
 178  *
 179  * Called with msg_ids.rw_mutex held (writer)
 180  */
 181 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 182 {
 183         struct msg_queue *msq;
 184         int id, retval;
 185         key_t key = params->key;
 186         int msgflg = params->flg;
 187
 188         msq = ipc_rcu_alloc(sizeof(*msq));
 189         if (!msq)
 190                 return -ENOMEM;
 191
 192         msq->q_perm.mode = msgflg & S_IRWXUGO;
 193         msq->q_perm.key = key;
 194
 195         msq->q_perm.security = NULL;
 196         retval = security_msg_queue_alloc(msq);
 197         if (retval) {
 198                 ipc_rcu_putref(msq);
 199                 return retval;
 200         }
 201
 202         /*
 203          * ipc_addid() locks msq
 204          */
 205         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 206         if (id < 0) {
 207                 security_msg_queue_free(msq);
 208                 ipc_rcu_putref(msq);
 209                 return id;
 210         }
 211
 212         msq->q_stime = msq->q_rtime = 0;
 213         msq->q_ctime = get_seconds();
 214         msq->q_cbytes = msq->q_qnum = 0;
 215         msq->q_qbytes = ns->msg_ctlmnb;
 216         msq->q_lspid = msq->q_lrpid = 0;
 217         INIT_LIST_HEAD(&msq->q_messages);
 218         INIT_LIST_HEAD(&msq->q_receivers);
 219         INIT_LIST_HEAD(&msq->q_senders);
 220
 221         msg_unlock(msq);
 222
 223         return msq->q_perm.id;
 224 }
 225
 226 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 227 {
 228         mss->tsk = current;
 229         current->state = TASK_INTERRUPTIBLE;
 230         list_add_tail(&mss->list, &msq->q_senders);
 231 }
 232
 233 static inline void ss_del(struct msg_sender *mss)
 234 {
 235         if (mss->list.next != NULL)
 236                 list_del(&mss->list);
 237 }
 238
 239 static void ss_wakeup(struct list_head *h, int kill)
 240 {
 241         struct msg_sender *mss, *t;
 242
 243         list_for_each_entry_safe(mss, t, h, list) {
 244                 if (kill)
 245                         mss->list.next = NULL;
 246                 wake_up_process(mss->tsk);
 247         }
 248 }
 249
 250 static void expunge_all(struct msg_queue *msq, int res)
 251 {
 252         struct msg_receiver *msr, *t;
 253
 254         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 255                 msr->r_msg = NULL;
 256                 wake_up_process(msr->r_tsk);
 257                 smp_mb();
 258                 msr->r_msg = ERR_PTR(res);
 259         }
 260 }
 261
 262 /*
 263  * freeque() wakes up waiters on the sender and receiver waiting queue,
 264  * removes the message queue from message queue ID IDR, and cleans up all the
 265  * messages associated with this queue.
 266  *
 267  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 268  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 269  */
 270 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 271 {
 272         struct msg_msg *msg, *t;
 273         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 274
 275         expunge_all(msq, -EIDRM);
 276         ss_wakeup(&msq->q_senders, 1);
 277         msg_rmid(ns, msq);
 278         msg_unlock(msq);
 279
 280         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 281                 atomic_dec(&ns->msg_hdrs);
 282                 free_msg(msg);
 283         }
 284         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 285         security_msg_queue_free(msq);
 286         ipc_rcu_putref(msq);
 287 }
 288
 289 /*
 290  * Called with msg_ids.rw_mutex and ipcp locked.
 291  */
 292 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 293 {
 294         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 295
 296         return security_msg_queue_associate(msq, msgflg);
 297 }
 298
 299 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 300 {
 301         struct ipc_namespace *ns;
 302         struct ipc_ops msg_ops;
 303         struct ipc_params msg_params;
 304
 305         ns = current->nsproxy->ipc_ns;
 306
 307         msg_ops.getnew = newque;
 308         msg_ops.associate = msg_security;
 309         msg_ops.more_checks = NULL;
 310
 311         msg_params.key = key;
 312         msg_params.flg = msgflg;
 313
 314         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 315 }
 316
 317 static inline unsigned long
 318 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 319 {
 320         switch(version) {
 321         case IPC_64:
 322                 return copy_to_user(buf, in, sizeof(*in));
 323         case IPC_OLD:
 324         {
 325                 struct msqid_ds out;
 326
 327                 memset(&out, 0, sizeof(out));
 328
 329                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 330
 331                 out.msg_stime           = in->msg_stime;
 332                 out.msg_rtime           = in->msg_rtime;
 333                 out.msg_ctime           = in->msg_ctime;
 334
 335                 if (in->msg_cbytes > USHRT_MAX)
 336                         out.msg_cbytes  = USHRT_MAX;
 337                 else
 338                         out.msg_cbytes  = in->msg_cbytes;
 339                 out.msg_lcbytes         = in->msg_cbytes;
 340
 341                 if (in->msg_qnum > USHRT_MAX)
 342                         out.msg_qnum    = USHRT_MAX;
 343                 else
 344                         out.msg_qnum    = in->msg_qnum;
 345
 346                 if (in->msg_qbytes > USHRT_MAX)
 347                         out.msg_qbytes  = USHRT_MAX;
 348                 else
 349                         out.msg_qbytes  = in->msg_qbytes;
 350                 out.msg_lqbytes         = in->msg_qbytes;
 351
 352                 out.msg_lspid           = in->msg_lspid;
 353                 out.msg_lrpid           = in->msg_lrpid;
 354
 355                 return copy_to_user(buf, &out, sizeof(out));
 356         }
 357         default:
 358                 return -EINVAL;
 359         }
 360 }
 361
 362 static inline unsigned long
 363 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 364 {
 365         switch(version) {
 366         case IPC_64:
 367                 if (copy_from_user(out, buf, sizeof(*out)))
 368                         return -EFAULT;
 369                 return 0;
 370         case IPC_OLD:
 371         {
 372                 struct msqid_ds tbuf_old;
 373
 374                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 375                         return -EFAULT;
 376
 377                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 378                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 379                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 380
 381                 if (tbuf_old.msg_qbytes == 0)
 382                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 383                 else
 384                         out->msg_qbytes = tbuf_old.msg_qbytes;
 385
 386                 return 0;
 387         }
 388         default:
 389                 return -EINVAL;
 390         }
 391 }
 392
 393 /*
 394  * This function handles some msgctl commands which require the rw_mutex
 395  * to be held in write mode.
 396  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 397  */
 398 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 399                        struct msqid_ds __user *buf, int version)
 400 {
 401         struct kern_ipc_perm *ipcp;
 402         struct msqid64_ds uninitialized_var(msqid64);
 403         struct msg_queue *msq;
 404         int err;
 405
 406         if (cmd == IPC_SET) {
 407                 if (copy_msqid_from_user(&msqid64, buf, version))
 408                         return -EFAULT;
 409         }
 410
 411         ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
 412                                &msqid64.msg_perm, msqid64.msg_qbytes);
 413         if (IS_ERR(ipcp))
 414                 return PTR_ERR(ipcp);
 415
 416         msq = container_of(ipcp, struct msg_queue, q_perm);
 417
 418         err = security_msg_queue_msgctl(msq, cmd);
 419         if (err)
 420                 goto out_unlock;
 421
 422         switch (cmd) {
 423         case IPC_RMID:
 424                 freeque(ns, ipcp);
 425                 goto out_up;
 426         case IPC_SET:
 427                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 428                     !capable(CAP_SYS_RESOURCE)) {
 429                         err = -EPERM;
 430                         goto out_unlock;
 431                 }
 432
 433                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 434                 if (err)
 435                         goto out_unlock;
 436
 437                 msq->q_qbytes = msqid64.msg_qbytes;
 438
 439                 msq->q_ctime = get_seconds();
 440                 /* sleeping receivers might be excluded by
 441                  * stricter permissions.
 442                  */
 443                 expunge_all(msq, -EAGAIN);
 444                 /* sleeping senders might be able to send
 445                  * due to a larger queue size.
 446                  */
 447                 ss_wakeup(&msq->q_senders, 0);
 448                 break;
 449         default:
 450                 err = -EINVAL;
 451         }
 452 out_unlock:
 453         msg_unlock(msq);
 454 out_up:
 455         up_write(&msg_ids(ns).rw_mutex);
 456         return err;
 457 }
 458
 459 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 460 {
 461         struct msg_queue *msq;
 462         int err, version;
 463         struct ipc_namespace *ns;
 464
 465         if (msqid < 0 || cmd < 0)
 466                 return -EINVAL;
 467
 468         version = ipc_parse_version(&cmd);
 469         ns = current->nsproxy->ipc_ns;
 470
 471         switch (cmd) {
 472         case IPC_INFO:
 473         case MSG_INFO:
 474         {
 475                 struct msginfo msginfo;
 476                 int max_id;
 477
 478                 if (!buf)
 479                         return -EFAULT;
 480                 /*
 481                  * We must not return kernel stack data.
 482                  * due to padding, it's not enough
 483                  * to set all member fields.
 484                  */
 485                 err = security_msg_queue_msgctl(NULL, cmd);
 486                 if (err)
 487                         return err;
 488
 489                 memset(&msginfo, 0, sizeof(msginfo));
 490                 msginfo.msgmni = ns->msg_ctlmni;
 491                 msginfo.msgmax = ns->msg_ctlmax;
 492                 msginfo.msgmnb = ns->msg_ctlmnb;
 493                 msginfo.msgssz = MSGSSZ;
 494                 msginfo.msgseg = MSGSEG;
 495                 down_read(&msg_ids(ns).rw_mutex);
 496                 if (cmd == MSG_INFO) {
 497                         msginfo.msgpool = msg_ids(ns).in_use;
 498                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 499                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 500                 } else {
 501                         msginfo.msgmap = MSGMAP;
 502                         msginfo.msgpool = MSGPOOL;
 503                         msginfo.msgtql = MSGTQL;
 504                 }
 505                 max_id = ipc_get_maxid(&msg_ids(ns));
 506                 up_read(&msg_ids(ns).rw_mutex);
 507                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 508                         return -EFAULT;
 509                 return (max_id < 0) ? 0 : max_id;
 510         }
 511         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 512         case IPC_STAT:
 513         {
 514                 struct msqid64_ds tbuf;
 515                 int success_return;
 516
 517                 if (!buf)
 518                         return -EFAULT;
 519
 520                 if (cmd == MSG_STAT) {
 521                         msq = msg_lock(ns, msqid);
 522                         if (IS_ERR(msq))
 523                                 return PTR_ERR(msq);
 524                         success_return = msq->q_perm.id;
 525                 } else {
 526                         msq = msg_lock_check(ns, msqid);
 527                         if (IS_ERR(msq))
 528                                 return PTR_ERR(msq);
 529                         success_return = 0;
 530                 }
 531                 err = -EACCES;
 532                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 533                         goto out_unlock;
 534
 535                 err = security_msg_queue_msgctl(msq, cmd);
 536                 if (err)
 537                         goto out_unlock;
 538
 539                 memset(&tbuf, 0, sizeof(tbuf));
 540
 541                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 542                 tbuf.msg_stime  = msq->q_stime;
 543                 tbuf.msg_rtime  = msq->q_rtime;
 544                 tbuf.msg_ctime  = msq->q_ctime;
 545                 tbuf.msg_cbytes = msq->q_cbytes;
 546                 tbuf.msg_qnum   = msq->q_qnum;
 547                 tbuf.msg_qbytes = msq->q_qbytes;
 548                 tbuf.msg_lspid  = msq->q_lspid;
 549                 tbuf.msg_lrpid  = msq->q_lrpid;
 550                 msg_unlock(msq);
 551                 if (copy_msqid_to_user(buf, &tbuf, version))
 552                         return -EFAULT;
 553                 return success_return;
 554         }
 555         case IPC_SET:
 556         case IPC_RMID:
 557                 err = msgctl_down(ns, msqid, cmd, buf, version);
 558                 return err;
 559         default:
 560                 return  -EINVAL;
 561         }
 562
 563 out_unlock:
 564         msg_unlock(msq);
 565         return err;
 566 }
 567
 568 static int testmsg(struct msg_msg *msg, long type, int mode)
 569 {
 570         switch(mode)
 571         {
 572                 case SEARCH_ANY:
 573                 case SEARCH_NUMBER:
 574                         return 1;
 575                 case SEARCH_LESSEQUAL:
 576                         if (msg->m_type <=type)
 577                                 return 1;
 578                         break;
 579                 case SEARCH_EQUAL:
 580                         if (msg->m_type == type)
 581                                 return 1;
 582                         break;
 583                 case SEARCH_NOTEQUAL:
 584                         if (msg->m_type != type)
 585                                 return 1;
 586                         break;
 587         }
 588         return 0;
 589 }
 590
 591 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 592 {
 593         struct msg_receiver *msr, *t;
 594
 595         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 596                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 597                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 598                                                msr->r_msgtype, msr->r_mode)) {
 599
 600                         list_del(&msr->r_list);
 601                         if (msr->r_maxsize < msg->m_ts) {
 602                                 msr->r_msg = NULL;
 603                                 wake_up_process(msr->r_tsk);
 604                                 smp_mb();
 605                                 msr->r_msg = ERR_PTR(-E2BIG);
 606                         } else {
 607                                 msr->r_msg = NULL;
 608                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 609                                 msq->q_rtime = get_seconds();
 610                                 wake_up_process(msr->r_tsk);
 611                                 smp_mb();
 612                                 msr->r_msg = msg;
 613
 614                                 return 1;
 615                         }
 616                 }
 617         }
 618         return 0;
 619 }
 620
 621 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 622                 size_t msgsz, int msgflg)
 623 {
 624         struct msg_queue *msq;
 625         struct msg_msg *msg;
 626         int err;
 627         struct ipc_namespace *ns;
 628
 629         ns = current->nsproxy->ipc_ns;
 630
 631         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 632                 return -EINVAL;
 633         if (mtype < 1)
 634                 return -EINVAL;
 635
 636         msg = load_msg(mtext, msgsz);
 637         if (IS_ERR(msg))
 638                 return PTR_ERR(msg);
 639
 640         msg->m_type = mtype;
 641         msg->m_ts = msgsz;
 642
 643         msq = msg_lock_check(ns, msqid);
 644         if (IS_ERR(msq)) {
 645                 err = PTR_ERR(msq);
 646                 goto out_free;
 647         }
 648
 649         for (;;) {
 650                 struct msg_sender s;
 651
 652                 err = -EACCES;
 653                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 654                         goto out_unlock_free;
 655
 656                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 657                 if (err)
 658                         goto out_unlock_free;
 659
 660                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 661                                 1 + msq->q_qnum <= msq->q_qbytes) {
 662                         break;
 663                 }
 664
 665                 /* queue full, wait: */
 666                 if (msgflg & IPC_NOWAIT) {
 667                         err = -EAGAIN;
 668                         goto out_unlock_free;
 669                 }
 670                 ss_add(msq, &s);
 671
 672                 if (!ipc_rcu_getref(msq)) {
 673                         err = -EIDRM;
 674                         goto out_unlock_free;
 675                 }
 676
 677                 msg_unlock(msq);
 678                 schedule();
 679
 680                 ipc_lock_by_ptr(&msq->q_perm);
 681                 ipc_rcu_putref(msq);
 682                 if (msq->q_perm.deleted) {
 683                         err = -EIDRM;
 684                         goto out_unlock_free;
 685                 }
 686                 ss_del(&s);
 687
 688                 if (signal_pending(current)) {
 689                         err = -ERESTARTNOHAND;
 690                         goto out_unlock_free;
 691                 }
 692         }
 693
 694         msq->q_lspid = task_tgid_vnr(current);
 695         msq->q_stime = get_seconds();
 696
 697         if (!pipelined_send(msq, msg)) {
 698                 /* no one is waiting for this message, enqueue it */
 699                 list_add_tail(&msg->m_list, &msq->q_messages);
 700                 msq->q_cbytes += msgsz;
 701                 msq->q_qnum++;
 702                 atomic_add(msgsz, &ns->msg_bytes);
 703                 atomic_inc(&ns->msg_hdrs);
 704         }
 705
 706         err = 0;
 707         msg = NULL;
 708
 709 out_unlock_free:
 710         msg_unlock(msq);
 711 out_free:
 712         if (msg != NULL)
 713                 free_msg(msg);
 714         return err;
 715 }
 716
 717 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 718                 int, msgflg)
 719 {
 720         long mtype;
 721
 722         if (get_user(mtype, &msgp->mtype))
 723                 return -EFAULT;
 724         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 725 }
 726
 727 static inline int convert_mode(long *msgtyp, int msgflg)
 728 {
 729         if (msgflg & MSG_COPY)
 730                 return SEARCH_NUMBER;
 731         /*
 732          *  find message of correct type.
 733          *  msgtyp = 0 => get first.
 734          *  msgtyp > 0 => get first message of matching type.
 735          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 736          */
 737         if (*msgtyp == 0)
 738                 return SEARCH_ANY;
 739         if (*msgtyp < 0) {
 740                 *msgtyp = -*msgtyp;
 741                 return SEARCH_LESSEQUAL;
 742         }
 743         if (msgflg & MSG_EXCEPT)
 744                 return SEARCH_NOTEQUAL;
 745         return SEARCH_EQUAL;
 746 }
 747
 748 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 749 {
 750         struct msgbuf __user *msgp = dest;
 751         size_t msgsz;
 752
 753         if (put_user(msg->m_type, &msgp->mtype))
 754                 return -EFAULT;
 755
 756         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 757         if (store_msg(msgp->mtext, msg, msgsz))
 758                 return -EFAULT;
 759         return msgsz;
 760 }
 761
 762 #ifdef CONFIG_CHECKPOINT_RESTORE
 763 /*
 764  * This function creates new kernel message structure, large enough to store
 765  * bufsz message bytes.
 766  */
 767 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 768 {
 769         struct msg_msg *copy;
 770
 771         /*
 772          * Create dummy message to copy real message to.
 773          */
 774         copy = load_msg(buf, bufsz);
 775         if (!IS_ERR(copy))
 776                 copy->m_ts = bufsz;
 777         return copy;
 778 }
 779
 780 static inline void free_copy(struct msg_msg *copy)
 781 {
 782         if (copy)
 783                 free_msg(copy);
 784 }
 785 #else
 786 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 787 {
 788         return ERR_PTR(-ENOSYS);
 789 }
 790
 791 static inline void free_copy(struct msg_msg *copy)
 792 {
 793 }
 794 #endif
 795
 796 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 797 {
 798         struct msg_msg *msg, *found = NULL;
 799         long count = 0;
 800
 801         list_for_each_entry(msg, &msq->q_messages, m_list) {
 802                 if (testmsg(msg, *msgtyp, mode) &&
 803                     !security_msg_queue_msgrcv(msq, msg, current,
 804                                                *msgtyp, mode)) {
 805                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 806                                 *msgtyp = msg->m_type - 1;
 807                                 found = msg;
 808                         } else if (mode == SEARCH_NUMBER) {
 809                                 if (*msgtyp == count)
 810                                         return msg;
 811                         } else
 812                                 return msg;
 813                         count++;
 814                 }
 815         }
 816
 817         return found ?: ERR_PTR(-EAGAIN);
 818 }
 819
 820
 821 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
 822                int msgflg,
 823                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 824 {
 825         struct msg_queue *msq;
 826         struct msg_msg *msg;
 827         int mode;
 828         struct ipc_namespace *ns;
 829         struct msg_msg *copy = NULL;
 830
 831         ns = current->nsproxy->ipc_ns;
 832
 833         if (msqid < 0 || (long) bufsz < 0)
 834                 return -EINVAL;
 835         if (msgflg & MSG_COPY) {
 836                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 837                 if (IS_ERR(copy))
 838                         return PTR_ERR(copy);
 839         }
 840         mode = convert_mode(&msgtyp, msgflg);
 841
 842         msq = msg_lock_check(ns, msqid);
 843         if (IS_ERR(msq)) {
 844                 free_copy(copy);
 845                 return PTR_ERR(msq);
 846         }
 847
 848         for (;;) {
 849                 struct msg_receiver msr_d;
 850
 851                 msg = ERR_PTR(-EACCES);
 852                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 853                         goto out_unlock;
 854
 855                 msg = find_msg(msq, &msgtyp, mode);
 856
 857                 if (!IS_ERR(msg)) {
 858                         /*
 859                          * Found a suitable message.
 860                          * Unlink it from the queue.
 861                          */
 862                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 863                                 msg = ERR_PTR(-E2BIG);
 864                                 goto out_unlock;
 865                         }
 866                         /*
 867                          * If we are copying, then do not unlink message and do
 868                          * not update queue parameters.
 869                          */
 870                         if (msgflg & MSG_COPY) {
 871                                 msg = copy_msg(msg, copy);
 872                                 goto out_unlock;
 873                         }
 874                         list_del(&msg->m_list);
 875                         msq->q_qnum--;
 876                         msq->q_rtime = get_seconds();
 877                         msq->q_lrpid = task_tgid_vnr(current);
 878                         msq->q_cbytes -= msg->m_ts;
 879                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 880                         atomic_dec(&ns->msg_hdrs);
 881                         ss_wakeup(&msq->q_senders, 0);
 882                         msg_unlock(msq);
 883                         break;
 884                 }
 885                 /* No message waiting. Wait for a message */
 886                 if (msgflg & IPC_NOWAIT) {
 887                         msg = ERR_PTR(-ENOMSG);
 888                         goto out_unlock;
 889                 }
 890                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 891                 msr_d.r_tsk = current;
 892                 msr_d.r_msgtype = msgtyp;
 893                 msr_d.r_mode = mode;
 894                 if (msgflg & MSG_NOERROR)
 895                         msr_d.r_maxsize = INT_MAX;
 896                 else
 897                         msr_d.r_maxsize = bufsz;
 898                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 899                 current->state = TASK_INTERRUPTIBLE;
 900                 msg_unlock(msq);
 901
 902                 schedule();
 903
 904                 /* Lockless receive, part 1:
 905                  * Disable preemption.  We don't hold a reference to the queue
 906                  * and getting a reference would defeat the idea of a lockless
 907                  * operation, thus the code relies on rcu to guarantee the
 908                  * existence of msq:
 909                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 910                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 911                  * rcu_read_lock() prevents preemption between reading r_msg
 912                  * and the spin_lock() inside ipc_lock_by_ptr().
 913                  */
 914                 rcu_read_lock();
 915
 916                 /* Lockless receive, part 2:
 917                  * Wait until pipelined_send or expunge_all are outside of
 918                  * wake_up_process(). There is a race with exit(), see
 919                  * ipc/mqueue.c for the details.
 920                  */
 921                 msg = (struct msg_msg*)msr_d.r_msg;
 922                 while (msg == NULL) {
 923                         cpu_relax();
 924                         msg = (struct msg_msg *)msr_d.r_msg;
 925                 }
 926
 927                 /* Lockless receive, part 3:
 928                  * If there is a message or an error then accept it without
 929                  * locking.
 930                  */
 931                 if (msg != ERR_PTR(-EAGAIN)) {
 932                         rcu_read_unlock();
 933                         break;
 934                 }
 935
 936                 /* Lockless receive, part 3:
 937                  * Acquire the queue spinlock.
 938                  */
 939                 ipc_lock_by_ptr(&msq->q_perm);
 940                 rcu_read_unlock();
 941
 942                 /* Lockless receive, part 4:
 943                  * Repeat test after acquiring the spinlock.
 944                  */
 945                 msg = (struct msg_msg*)msr_d.r_msg;
 946                 if (msg != ERR_PTR(-EAGAIN))
 947                         goto out_unlock;
 948
 949                 list_del(&msr_d.r_list);
 950                 if (signal_pending(current)) {
 951                         msg = ERR_PTR(-ERESTARTNOHAND);
 952 out_unlock:
 953                         msg_unlock(msq);
 954                         break;
 955                 }
 956         }
 957         if (IS_ERR(msg)) {
 958                 free_copy(copy);
 959                 return PTR_ERR(msg);
 960         }
 961
 962         bufsz = msg_handler(buf, msg, bufsz);
 963         free_msg(msg);
 964
 965         return bufsz;
 966 }
 967
 968 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 969                 long, msgtyp, int, msgflg)
 970 {
 971         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
 972 }
 973
 974 #ifdef CONFIG_PROC_FS
 975 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 976 {
 977         struct user_namespace *user_ns = seq_user_ns(s);
 978         struct msg_queue *msq = it;
 979
 980         return seq_printf(s,
 981                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 982                         msq->q_perm.key,
 983                         msq->q_perm.id,
 984                         msq->q_perm.mode,
 985                         msq->q_cbytes,
 986                         msq->q_qnum,
 987                         msq->q_lspid,
 988                         msq->q_lrpid,
 989                         from_kuid_munged(user_ns, msq->q_perm.uid),
 990                         from_kgid_munged(user_ns, msq->q_perm.gid),
 991                         from_kuid_munged(user_ns, msq->q_perm.cuid),
 992                         from_kgid_munged(user_ns, msq->q_perm.cgid),
 993                         msq->q_stime,
 994                         msq->q_rtime,
 995                         msq->q_ctime);
 996 }
 997 #endif