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         /* ipc_addid() locks msq upon success. */
 203         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 204         if (id < 0) {
 205                 security_msg_queue_free(msq);
 206                 ipc_rcu_putref(msq);
 207                 return id;
 208         }
 209
 210         msq->q_stime = msq->q_rtime = 0;
 211         msq->q_ctime = get_seconds();
 212         msq->q_cbytes = msq->q_qnum = 0;
 213         msq->q_qbytes = ns->msg_ctlmnb;
 214         msq->q_lspid = msq->q_lrpid = 0;
 215         INIT_LIST_HEAD(&msq->q_messages);
 216         INIT_LIST_HEAD(&msq->q_receivers);
 217         INIT_LIST_HEAD(&msq->q_senders);
 218
 219         ipc_unlock_object(&msq->q_perm);
 220         rcu_read_unlock();
 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 msg_sender *mss, *t;
 241
 242         list_for_each_entry_safe(mss, t, h, list) {
 243                 if (kill)
 244                         mss->list.next = NULL;
 245                 wake_up_process(mss->tsk);
 246         }
 247 }
 248
 249 static void expunge_all(struct msg_queue *msq, int res)
 250 {
 251         struct msg_receiver *msr, *t;
 252
 253         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 254                 msr->r_msg = NULL;
 255                 wake_up_process(msr->r_tsk);
 256                 smp_mb();
 257                 msr->r_msg = ERR_PTR(res);
 258         }
 259 }
 260
 261 /*
 262  * freeque() wakes up waiters on the sender and receiver waiting queue,
 263  * removes the message queue from message queue ID IDR, and cleans up all the
 264  * messages associated with this queue.
 265  *
 266  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 267  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 268  */
 269 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 270 {
 271         struct msg_msg *msg, *t;
 272         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 273
 274         expunge_all(msq, -EIDRM);
 275         ss_wakeup(&msq->q_senders, 1);
 276         msg_rmid(ns, msq);
 277         msg_unlock(msq);
 278
 279         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 280                 atomic_dec(&ns->msg_hdrs);
 281                 free_msg(msg);
 282         }
 283         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 284         security_msg_queue_free(msq);
 285         ipc_rcu_putref(msq);
 286 }
 287
 288 /*
 289  * Called with msg_ids.rw_mutex and ipcp locked.
 290  */
 291 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 292 {
 293         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 294
 295         return security_msg_queue_associate(msq, msgflg);
 296 }
 297
 298 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 299 {
 300         struct ipc_namespace *ns;
 301         struct ipc_ops msg_ops;
 302         struct ipc_params msg_params;
 303
 304         ns = current->nsproxy->ipc_ns;
 305
 306         msg_ops.getnew = newque;
 307         msg_ops.associate = msg_security;
 308         msg_ops.more_checks = NULL;
 309
 310         msg_params.key = key;
 311         msg_params.flg = msgflg;
 312
 313         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 314 }
 315
 316 static inline unsigned long
 317 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 318 {
 319         switch(version) {
 320         case IPC_64:
 321                 return copy_to_user(buf, in, sizeof(*in));
 322         case IPC_OLD:
 323         {
 324                 struct msqid_ds out;
 325
 326                 memset(&out, 0, sizeof(out));
 327
 328                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 329
 330                 out.msg_stime           = in->msg_stime;
 331                 out.msg_rtime           = in->msg_rtime;
 332                 out.msg_ctime           = in->msg_ctime;
 333
 334                 if (in->msg_cbytes > USHRT_MAX)
 335                         out.msg_cbytes  = USHRT_MAX;
 336                 else
 337                         out.msg_cbytes  = in->msg_cbytes;
 338                 out.msg_lcbytes         = in->msg_cbytes;
 339
 340                 if (in->msg_qnum > USHRT_MAX)
 341                         out.msg_qnum    = USHRT_MAX;
 342                 else
 343                         out.msg_qnum    = in->msg_qnum;
 344
 345                 if (in->msg_qbytes > USHRT_MAX)
 346                         out.msg_qbytes  = USHRT_MAX;
 347                 else
 348                         out.msg_qbytes  = in->msg_qbytes;
 349                 out.msg_lqbytes         = in->msg_qbytes;
 350
 351                 out.msg_lspid           = in->msg_lspid;
 352                 out.msg_lrpid           = in->msg_lrpid;
 353
 354                 return copy_to_user(buf, &out, sizeof(out));
 355         }
 356         default:
 357                 return -EINVAL;
 358         }
 359 }
 360
 361 static inline unsigned long
 362 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 363 {
 364         switch(version) {
 365         case IPC_64:
 366                 if (copy_from_user(out, buf, sizeof(*out)))
 367                         return -EFAULT;
 368                 return 0;
 369         case IPC_OLD:
 370         {
 371                 struct msqid_ds tbuf_old;
 372
 373                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 374                         return -EFAULT;
 375
 376                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 377                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 378                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 379
 380                 if (tbuf_old.msg_qbytes == 0)
 381                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 382                 else
 383                         out->msg_qbytes = tbuf_old.msg_qbytes;
 384
 385                 return 0;
 386         }
 387         default:
 388                 return -EINVAL;
 389         }
 390 }
 391
 392 /*
 393  * This function handles some msgctl commands which require the rw_mutex
 394  * to be held in write mode.
 395  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 396  */
 397 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 398                        struct msqid_ds __user *buf, int version)
 399 {
 400         struct kern_ipc_perm *ipcp;
 401         struct msqid64_ds uninitialized_var(msqid64);
 402         struct msg_queue *msq;
 403         int err;
 404
 405         if (cmd == IPC_SET) {
 406                 if (copy_msqid_from_user(&msqid64, buf, version))
 407                         return -EFAULT;
 408         }
 409
 410         down_write(&msg_ids(ns).rw_mutex);
 411         rcu_read_lock();
 412
 413         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 414                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 415         if (IS_ERR(ipcp)) {
 416                 err = PTR_ERR(ipcp);
 417                 goto out_unlock1;
 418         }
 419
 420         msq = container_of(ipcp, struct msg_queue, q_perm);
 421
 422         err = security_msg_queue_msgctl(msq, cmd);
 423         if (err)
 424                 goto out_unlock1;
 425
 426         switch (cmd) {
 427         case IPC_RMID:
 428                 ipc_lock_object(&msq->q_perm);
 429                 /* freeque unlocks the ipc object and rcu */
 430                 freeque(ns, ipcp);
 431                 goto out_up;
 432         case IPC_SET:
 433                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 434                     !capable(CAP_SYS_RESOURCE)) {
 435                         err = -EPERM;
 436                         goto out_unlock1;
 437                 }
 438
 439                 ipc_lock_object(&msq->q_perm);
 440                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 441                 if (err)
 442                         goto out_unlock0;
 443
 444                 msq->q_qbytes = msqid64.msg_qbytes;
 445
 446                 msq->q_ctime = get_seconds();
 447                 /* sleeping receivers might be excluded by
 448                  * stricter permissions.
 449                  */
 450                 expunge_all(msq, -EAGAIN);
 451                 /* sleeping senders might be able to send
 452                  * due to a larger queue size.
 453                  */
 454                 ss_wakeup(&msq->q_senders, 0);
 455                 break;
 456         default:
 457                 err = -EINVAL;
 458                 goto out_unlock1;
 459         }
 460
 461 out_unlock0:
 462         ipc_unlock_object(&msq->q_perm);
 463 out_unlock1:
 464         rcu_read_unlock();
 465 out_up:
 466         up_write(&msg_ids(ns).rw_mutex);
 467         return err;
 468 }
 469
 470 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 471 {
 472         struct msg_queue *msq;
 473         int err, version;
 474         struct ipc_namespace *ns;
 475
 476         if (msqid < 0 || cmd < 0)
 477                 return -EINVAL;
 478
 479         version = ipc_parse_version(&cmd);
 480         ns = current->nsproxy->ipc_ns;
 481
 482         switch (cmd) {
 483         case IPC_INFO:
 484         case MSG_INFO:
 485         {
 486                 struct msginfo msginfo;
 487                 int max_id;
 488
 489                 if (!buf)
 490                         return -EFAULT;
 491                 /*
 492                  * We must not return kernel stack data.
 493                  * due to padding, it's not enough
 494                  * to set all member fields.
 495                  */
 496                 err = security_msg_queue_msgctl(NULL, cmd);
 497                 if (err)
 498                         return err;
 499
 500                 memset(&msginfo, 0, sizeof(msginfo));
 501                 msginfo.msgmni = ns->msg_ctlmni;
 502                 msginfo.msgmax = ns->msg_ctlmax;
 503                 msginfo.msgmnb = ns->msg_ctlmnb;
 504                 msginfo.msgssz = MSGSSZ;
 505                 msginfo.msgseg = MSGSEG;
 506                 down_read(&msg_ids(ns).rw_mutex);
 507                 if (cmd == MSG_INFO) {
 508                         msginfo.msgpool = msg_ids(ns).in_use;
 509                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 510                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 511                 } else {
 512                         msginfo.msgmap = MSGMAP;
 513                         msginfo.msgpool = MSGPOOL;
 514                         msginfo.msgtql = MSGTQL;
 515                 }
 516                 max_id = ipc_get_maxid(&msg_ids(ns));
 517                 up_read(&msg_ids(ns).rw_mutex);
 518                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 519                         return -EFAULT;
 520                 return (max_id < 0) ? 0 : max_id;
 521         }
 522         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 523         case IPC_STAT:
 524         {
 525                 struct msqid64_ds tbuf;
 526                 int success_return;
 527
 528                 if (!buf)
 529                         return -EFAULT;
 530
 531                 if (cmd == MSG_STAT) {
 532                         msq = msg_lock(ns, msqid);
 533                         if (IS_ERR(msq))
 534                                 return PTR_ERR(msq);
 535                         success_return = msq->q_perm.id;
 536                 } else {
 537                         msq = msg_lock_check(ns, msqid);
 538                         if (IS_ERR(msq))
 539                                 return PTR_ERR(msq);
 540                         success_return = 0;
 541                 }
 542                 err = -EACCES;
 543                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 544                         goto out_unlock;
 545
 546                 err = security_msg_queue_msgctl(msq, cmd);
 547                 if (err)
 548                         goto out_unlock;
 549
 550                 memset(&tbuf, 0, sizeof(tbuf));
 551
 552                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 553                 tbuf.msg_stime  = msq->q_stime;
 554                 tbuf.msg_rtime  = msq->q_rtime;
 555                 tbuf.msg_ctime  = msq->q_ctime;
 556                 tbuf.msg_cbytes = msq->q_cbytes;
 557                 tbuf.msg_qnum   = msq->q_qnum;
 558                 tbuf.msg_qbytes = msq->q_qbytes;
 559                 tbuf.msg_lspid  = msq->q_lspid;
 560                 tbuf.msg_lrpid  = msq->q_lrpid;
 561                 msg_unlock(msq);
 562                 if (copy_msqid_to_user(buf, &tbuf, version))
 563                         return -EFAULT;
 564                 return success_return;
 565         }
 566         case IPC_SET:
 567         case IPC_RMID:
 568                 err = msgctl_down(ns, msqid, cmd, buf, version);
 569                 return err;
 570         default:
 571                 return  -EINVAL;
 572         }
 573
 574 out_unlock:
 575         msg_unlock(msq);
 576         return err;
 577 }
 578
 579 static int testmsg(struct msg_msg *msg, long type, int mode)
 580 {
 581         switch(mode)
 582         {
 583                 case SEARCH_ANY:
 584                 case SEARCH_NUMBER:
 585                         return 1;
 586                 case SEARCH_LESSEQUAL:
 587                         if (msg->m_type <=type)
 588                                 return 1;
 589                         break;
 590                 case SEARCH_EQUAL:
 591                         if (msg->m_type == type)
 592                                 return 1;
 593                         break;
 594                 case SEARCH_NOTEQUAL:
 595                         if (msg->m_type != type)
 596                                 return 1;
 597                         break;
 598         }
 599         return 0;
 600 }
 601
 602 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 603 {
 604         struct msg_receiver *msr, *t;
 605
 606         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 607                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 608                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 609                                                msr->r_msgtype, msr->r_mode)) {
 610
 611                         list_del(&msr->r_list);
 612                         if (msr->r_maxsize < msg->m_ts) {
 613                                 msr->r_msg = NULL;
 614                                 wake_up_process(msr->r_tsk);
 615                                 smp_mb();
 616                                 msr->r_msg = ERR_PTR(-E2BIG);
 617                         } else {
 618                                 msr->r_msg = NULL;
 619                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 620                                 msq->q_rtime = get_seconds();
 621                                 wake_up_process(msr->r_tsk);
 622                                 smp_mb();
 623                                 msr->r_msg = msg;
 624
 625                                 return 1;
 626                         }
 627                 }
 628         }
 629         return 0;
 630 }
 631
 632 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 633                 size_t msgsz, int msgflg)
 634 {
 635         struct msg_queue *msq;
 636         struct msg_msg *msg;
 637         int err;
 638         struct ipc_namespace *ns;
 639
 640         ns = current->nsproxy->ipc_ns;
 641
 642         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 643                 return -EINVAL;
 644         if (mtype < 1)
 645                 return -EINVAL;
 646
 647         msg = load_msg(mtext, msgsz);
 648         if (IS_ERR(msg))
 649                 return PTR_ERR(msg);
 650
 651         msg->m_type = mtype;
 652         msg->m_ts = msgsz;
 653
 654         msq = msg_lock_check(ns, msqid);
 655         if (IS_ERR(msq)) {
 656                 err = PTR_ERR(msq);
 657                 goto out_free;
 658         }
 659
 660         for (;;) {
 661                 struct msg_sender s;
 662
 663                 err = -EACCES;
 664                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 665                         goto out_unlock_free;
 666
 667                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 668                 if (err)
 669                         goto out_unlock_free;
 670
 671                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 672                                 1 + msq->q_qnum <= msq->q_qbytes) {
 673                         break;
 674                 }
 675
 676                 /* queue full, wait: */
 677                 if (msgflg & IPC_NOWAIT) {
 678                         err = -EAGAIN;
 679                         goto out_unlock_free;
 680                 }
 681                 ss_add(msq, &s);
 682
 683                 if (!ipc_rcu_getref(msq)) {
 684                         err = -EIDRM;
 685                         goto out_unlock_free;
 686                 }
 687
 688                 msg_unlock(msq);
 689                 schedule();
 690
 691                 ipc_lock_by_ptr(&msq->q_perm);
 692                 ipc_rcu_putref(msq);
 693                 if (msq->q_perm.deleted) {
 694                         err = -EIDRM;
 695                         goto out_unlock_free;
 696                 }
 697                 ss_del(&s);
 698
 699                 if (signal_pending(current)) {
 700                         err = -ERESTARTNOHAND;
 701                         goto out_unlock_free;
 702                 }
 703         }
 704
 705         msq->q_lspid = task_tgid_vnr(current);
 706         msq->q_stime = get_seconds();
 707
 708         if (!pipelined_send(msq, msg)) {
 709                 /* no one is waiting for this message, enqueue it */
 710                 list_add_tail(&msg->m_list, &msq->q_messages);
 711                 msq->q_cbytes += msgsz;
 712                 msq->q_qnum++;
 713                 atomic_add(msgsz, &ns->msg_bytes);
 714                 atomic_inc(&ns->msg_hdrs);
 715         }
 716
 717         err = 0;
 718         msg = NULL;
 719
 720 out_unlock_free:
 721         msg_unlock(msq);
 722 out_free:
 723         if (msg != NULL)
 724                 free_msg(msg);
 725         return err;
 726 }
 727
 728 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 729                 int, msgflg)
 730 {
 731         long mtype;
 732
 733         if (get_user(mtype, &msgp->mtype))
 734                 return -EFAULT;
 735         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 736 }
 737
 738 static inline int convert_mode(long *msgtyp, int msgflg)
 739 {
 740         if (msgflg & MSG_COPY)
 741                 return SEARCH_NUMBER;
 742         /*
 743          *  find message of correct type.
 744          *  msgtyp = 0 => get first.
 745          *  msgtyp > 0 => get first message of matching type.
 746          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 747          */
 748         if (*msgtyp == 0)
 749                 return SEARCH_ANY;
 750         if (*msgtyp < 0) {
 751                 *msgtyp = -*msgtyp;
 752                 return SEARCH_LESSEQUAL;
 753         }
 754         if (msgflg & MSG_EXCEPT)
 755                 return SEARCH_NOTEQUAL;
 756         return SEARCH_EQUAL;
 757 }
 758
 759 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 760 {
 761         struct msgbuf __user *msgp = dest;
 762         size_t msgsz;
 763
 764         if (put_user(msg->m_type, &msgp->mtype))
 765                 return -EFAULT;
 766
 767         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 768         if (store_msg(msgp->mtext, msg, msgsz))
 769                 return -EFAULT;
 770         return msgsz;
 771 }
 772
 773 #ifdef CONFIG_CHECKPOINT_RESTORE
 774 /*
 775  * This function creates new kernel message structure, large enough to store
 776  * bufsz message bytes.
 777  */
 778 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 779 {
 780         struct msg_msg *copy;
 781
 782         /*
 783          * Create dummy message to copy real message to.
 784          */
 785         copy = load_msg(buf, bufsz);
 786         if (!IS_ERR(copy))
 787                 copy->m_ts = bufsz;
 788         return copy;
 789 }
 790
 791 static inline void free_copy(struct msg_msg *copy)
 792 {
 793         if (copy)
 794                 free_msg(copy);
 795 }
 796 #else
 797 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 798 {
 799         return ERR_PTR(-ENOSYS);
 800 }
 801
 802 static inline void free_copy(struct msg_msg *copy)
 803 {
 804 }
 805 #endif
 806
 807 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 808 {
 809         struct msg_msg *msg;
 810         long count = 0;
 811
 812         list_for_each_entry(msg, &msq->q_messages, m_list) {
 813                 if (testmsg(msg, *msgtyp, mode) &&
 814                     !security_msg_queue_msgrcv(msq, msg, current,
 815                                                *msgtyp, mode)) {
 816                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 817                                 *msgtyp = msg->m_type - 1;
 818                         } else if (mode == SEARCH_NUMBER) {
 819                                 if (*msgtyp == count)
 820                                         return msg;
 821                         } else
 822                                 return msg;
 823                         count++;
 824                 }
 825         }
 826
 827         return ERR_PTR(-EAGAIN);
 828 }
 829
 830
 831 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
 832                int msgflg,
 833                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 834 {
 835         struct msg_queue *msq;
 836         struct msg_msg *msg;
 837         int mode;
 838         struct ipc_namespace *ns;
 839         struct msg_msg *copy = NULL;
 840
 841         ns = current->nsproxy->ipc_ns;
 842
 843         if (msqid < 0 || (long) bufsz < 0)
 844                 return -EINVAL;
 845         if (msgflg & MSG_COPY) {
 846                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 847                 if (IS_ERR(copy))
 848                         return PTR_ERR(copy);
 849         }
 850         mode = convert_mode(&msgtyp, msgflg);
 851
 852         msq = msg_lock_check(ns, msqid);
 853         if (IS_ERR(msq)) {
 854                 free_copy(copy);
 855                 return PTR_ERR(msq);
 856         }
 857
 858         for (;;) {
 859                 struct msg_receiver msr_d;
 860
 861                 msg = ERR_PTR(-EACCES);
 862                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 863                         goto out_unlock;
 864
 865                 msg = find_msg(msq, &msgtyp, mode);
 866
 867                 if (!IS_ERR(msg)) {
 868                         /*
 869                          * Found a suitable message.
 870                          * Unlink it from the queue.
 871                          */
 872                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 873                                 msg = ERR_PTR(-E2BIG);
 874                                 goto out_unlock;
 875                         }
 876                         /*
 877                          * If we are copying, then do not unlink message and do
 878                          * not update queue parameters.
 879                          */
 880                         if (msgflg & MSG_COPY) {
 881                                 msg = copy_msg(msg, copy);
 882                                 goto out_unlock;
 883                         }
 884                         list_del(&msg->m_list);
 885                         msq->q_qnum--;
 886                         msq->q_rtime = get_seconds();
 887                         msq->q_lrpid = task_tgid_vnr(current);
 888                         msq->q_cbytes -= msg->m_ts;
 889                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 890                         atomic_dec(&ns->msg_hdrs);
 891                         ss_wakeup(&msq->q_senders, 0);
 892                         msg_unlock(msq);
 893                         break;
 894                 }
 895                 /* No message waiting. Wait for a message */
 896                 if (msgflg & IPC_NOWAIT) {
 897                         msg = ERR_PTR(-ENOMSG);
 898                         goto out_unlock;
 899                 }
 900                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 901                 msr_d.r_tsk = current;
 902                 msr_d.r_msgtype = msgtyp;
 903                 msr_d.r_mode = mode;
 904                 if (msgflg & MSG_NOERROR)
 905                         msr_d.r_maxsize = INT_MAX;
 906                 else
 907                         msr_d.r_maxsize = bufsz;
 908                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 909                 current->state = TASK_INTERRUPTIBLE;
 910                 msg_unlock(msq);
 911
 912                 schedule();
 913
 914                 /* Lockless receive, part 1:
 915                  * Disable preemption.  We don't hold a reference to the queue
 916                  * and getting a reference would defeat the idea of a lockless
 917                  * operation, thus the code relies on rcu to guarantee the
 918                  * existence of msq:
 919                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 920                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 921                  * rcu_read_lock() prevents preemption between reading r_msg
 922                  * and the spin_lock() inside ipc_lock_by_ptr().
 923                  */
 924                 rcu_read_lock();
 925
 926                 /* Lockless receive, part 2:
 927                  * Wait until pipelined_send or expunge_all are outside of
 928                  * wake_up_process(). There is a race with exit(), see
 929                  * ipc/mqueue.c for the details.
 930                  */
 931                 msg = (struct msg_msg*)msr_d.r_msg;
 932                 while (msg == NULL) {
 933                         cpu_relax();
 934                         msg = (struct msg_msg *)msr_d.r_msg;
 935                 }
 936
 937                 /* Lockless receive, part 3:
 938                  * If there is a message or an error then accept it without
 939                  * locking.
 940                  */
 941                 if (msg != ERR_PTR(-EAGAIN)) {
 942                         rcu_read_unlock();
 943                         break;
 944                 }
 945
 946                 /* Lockless receive, part 3:
 947                  * Acquire the queue spinlock.
 948                  */
 949                 ipc_lock_by_ptr(&msq->q_perm);
 950                 rcu_read_unlock();
 951
 952                 /* Lockless receive, part 4:
 953                  * Repeat test after acquiring the spinlock.
 954                  */
 955                 msg = (struct msg_msg*)msr_d.r_msg;
 956                 if (msg != ERR_PTR(-EAGAIN))
 957                         goto out_unlock;
 958
 959                 list_del(&msr_d.r_list);
 960                 if (signal_pending(current)) {
 961                         msg = ERR_PTR(-ERESTARTNOHAND);
 962 out_unlock:
 963                         msg_unlock(msq);
 964                         break;
 965                 }
 966         }
 967         if (IS_ERR(msg)) {
 968                 free_copy(copy);
 969                 return PTR_ERR(msg);
 970         }
 971
 972         bufsz = msg_handler(buf, msg, bufsz);
 973         free_msg(msg);
 974
 975         return bufsz;
 976 }
 977
 978 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 979                 long, msgtyp, int, msgflg)
 980 {
 981         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
 982 }
 983
 984 #ifdef CONFIG_PROC_FS
 985 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 986 {
 987         struct user_namespace *user_ns = seq_user_ns(s);
 988         struct msg_queue *msq = it;
 989
 990         return seq_printf(s,
 991                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 992                         msq->q_perm.key,
 993                         msq->q_perm.id,
 994                         msq->q_perm.mode,
 995                         msq->q_cbytes,
 996                         msq->q_qnum,
 997                         msq->q_lspid,
 998                         msq->q_lrpid,
 999                         from_kuid_munged(user_ns, msq->q_perm.uid),
1000                         from_kgid_munged(user_ns, msq->q_perm.gid),
1001                         from_kuid_munged(user_ns, msq->q_perm.cuid),
1002                         from_kgid_munged(user_ns, msq->q_perm.cgid),
1003                         msq->q_stime,
1004                         msq->q_rtime,
1005                         msq->q_ctime);
1006 }
1007 #endif