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 struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
 170 {
 171         struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
 172
 173         if (IS_ERR(ipcp))
 174                 return ERR_CAST(ipcp);
 175
 176         return container_of(ipcp, struct msg_queue, q_perm);
 177 }
 178
 179 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
 180                                                         int id)
 181 {
 182         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
 183
 184         if (IS_ERR(ipcp))
 185                 return ERR_CAST(ipcp);
 186
 187         return container_of(ipcp, struct msg_queue, q_perm);
 188 }
 189
 190 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 191 {
 192         ipc_rmid(&msg_ids(ns), &s->q_perm);
 193 }
 194
 195 /**
 196  * newque - Create a new msg queue
 197  * @ns: namespace
 198  * @params: ptr to the structure that contains the key and msgflg
 199  *
 200  * Called with msg_ids.rw_mutex held (writer)
 201  */
 202 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 203 {
 204         struct msg_queue *msq;
 205         int id, retval;
 206         key_t key = params->key;
 207         int msgflg = params->flg;
 208
 209         msq = ipc_rcu_alloc(sizeof(*msq));
 210         if (!msq)
 211                 return -ENOMEM;
 212
 213         msq->q_perm.mode = msgflg & S_IRWXUGO;
 214         msq->q_perm.key = key;
 215
 216         msq->q_perm.security = NULL;
 217         retval = security_msg_queue_alloc(msq);
 218         if (retval) {
 219                 ipc_rcu_putref(msq);
 220                 return retval;
 221         }
 222
 223         /* ipc_addid() locks msq upon success. */
 224         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 225         if (id < 0) {
 226                 security_msg_queue_free(msq);
 227                 ipc_rcu_putref(msq);
 228                 return id;
 229         }
 230
 231         msq->q_stime = msq->q_rtime = 0;
 232         msq->q_ctime = get_seconds();
 233         msq->q_cbytes = msq->q_qnum = 0;
 234         msq->q_qbytes = ns->msg_ctlmnb;
 235         msq->q_lspid = msq->q_lrpid = 0;
 236         INIT_LIST_HEAD(&msq->q_messages);
 237         INIT_LIST_HEAD(&msq->q_receivers);
 238         INIT_LIST_HEAD(&msq->q_senders);
 239
 240         ipc_unlock_object(&msq->q_perm);
 241         rcu_read_unlock();
 242
 243         return msq->q_perm.id;
 244 }
 245
 246 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 247 {
 248         mss->tsk = current;
 249         current->state = TASK_INTERRUPTIBLE;
 250         list_add_tail(&mss->list, &msq->q_senders);
 251 }
 252
 253 static inline void ss_del(struct msg_sender *mss)
 254 {
 255         if (mss->list.next != NULL)
 256                 list_del(&mss->list);
 257 }
 258
 259 static void ss_wakeup(struct list_head *h, int kill)
 260 {
 261         struct msg_sender *mss, *t;
 262
 263         list_for_each_entry_safe(mss, t, h, list) {
 264                 if (kill)
 265                         mss->list.next = NULL;
 266                 wake_up_process(mss->tsk);
 267         }
 268 }
 269
 270 static void expunge_all(struct msg_queue *msq, int res)
 271 {
 272         struct msg_receiver *msr, *t;
 273
 274         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 275                 msr->r_msg = NULL;
 276                 wake_up_process(msr->r_tsk);
 277                 smp_mb();
 278                 msr->r_msg = ERR_PTR(res);
 279         }
 280 }
 281
 282 /*
 283  * freeque() wakes up waiters on the sender and receiver waiting queue,
 284  * removes the message queue from message queue ID IDR, and cleans up all the
 285  * messages associated with this queue.
 286  *
 287  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 288  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 289  */
 290 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 291 {
 292         struct msg_msg *msg, *t;
 293         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 294
 295         expunge_all(msq, -EIDRM);
 296         ss_wakeup(&msq->q_senders, 1);
 297         msg_rmid(ns, msq);
 298         msg_unlock(msq);
 299
 300         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 301                 atomic_dec(&ns->msg_hdrs);
 302                 free_msg(msg);
 303         }
 304         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 305         security_msg_queue_free(msq);
 306         ipc_rcu_putref(msq);
 307 }
 308
 309 /*
 310  * Called with msg_ids.rw_mutex and ipcp locked.
 311  */
 312 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 313 {
 314         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 315
 316         return security_msg_queue_associate(msq, msgflg);
 317 }
 318
 319 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 320 {
 321         struct ipc_namespace *ns;
 322         struct ipc_ops msg_ops;
 323         struct ipc_params msg_params;
 324
 325         ns = current->nsproxy->ipc_ns;
 326
 327         msg_ops.getnew = newque;
 328         msg_ops.associate = msg_security;
 329         msg_ops.more_checks = NULL;
 330
 331         msg_params.key = key;
 332         msg_params.flg = msgflg;
 333
 334         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 335 }
 336
 337 static inline unsigned long
 338 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 339 {
 340         switch(version) {
 341         case IPC_64:
 342                 return copy_to_user(buf, in, sizeof(*in));
 343         case IPC_OLD:
 344         {
 345                 struct msqid_ds out;
 346
 347                 memset(&out, 0, sizeof(out));
 348
 349                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 350
 351                 out.msg_stime           = in->msg_stime;
 352                 out.msg_rtime           = in->msg_rtime;
 353                 out.msg_ctime           = in->msg_ctime;
 354
 355                 if (in->msg_cbytes > USHRT_MAX)
 356                         out.msg_cbytes  = USHRT_MAX;
 357                 else
 358                         out.msg_cbytes  = in->msg_cbytes;
 359                 out.msg_lcbytes         = in->msg_cbytes;
 360
 361                 if (in->msg_qnum > USHRT_MAX)
 362                         out.msg_qnum    = USHRT_MAX;
 363                 else
 364                         out.msg_qnum    = in->msg_qnum;
 365
 366                 if (in->msg_qbytes > USHRT_MAX)
 367                         out.msg_qbytes  = USHRT_MAX;
 368                 else
 369                         out.msg_qbytes  = in->msg_qbytes;
 370                 out.msg_lqbytes         = in->msg_qbytes;
 371
 372                 out.msg_lspid           = in->msg_lspid;
 373                 out.msg_lrpid           = in->msg_lrpid;
 374
 375                 return copy_to_user(buf, &out, sizeof(out));
 376         }
 377         default:
 378                 return -EINVAL;
 379         }
 380 }
 381
 382 static inline unsigned long
 383 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 384 {
 385         switch(version) {
 386         case IPC_64:
 387                 if (copy_from_user(out, buf, sizeof(*out)))
 388                         return -EFAULT;
 389                 return 0;
 390         case IPC_OLD:
 391         {
 392                 struct msqid_ds tbuf_old;
 393
 394                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 395                         return -EFAULT;
 396
 397                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 398                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 399                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 400
 401                 if (tbuf_old.msg_qbytes == 0)
 402                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 403                 else
 404                         out->msg_qbytes = tbuf_old.msg_qbytes;
 405
 406                 return 0;
 407         }
 408         default:
 409                 return -EINVAL;
 410         }
 411 }
 412
 413 /*
 414  * This function handles some msgctl commands which require the rw_mutex
 415  * to be held in write mode.
 416  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 417  */
 418 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 419                        struct msqid_ds __user *buf, int version)
 420 {
 421         struct kern_ipc_perm *ipcp;
 422         struct msqid64_ds uninitialized_var(msqid64);
 423         struct msg_queue *msq;
 424         int err;
 425
 426         if (cmd == IPC_SET) {
 427                 if (copy_msqid_from_user(&msqid64, buf, version))
 428                         return -EFAULT;
 429         }
 430
 431         down_write(&msg_ids(ns).rw_mutex);
 432         rcu_read_lock();
 433
 434         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 435                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 436         if (IS_ERR(ipcp)) {
 437                 err = PTR_ERR(ipcp);
 438                 goto out_unlock1;
 439         }
 440
 441         msq = container_of(ipcp, struct msg_queue, q_perm);
 442
 443         err = security_msg_queue_msgctl(msq, cmd);
 444         if (err)
 445                 goto out_unlock1;
 446
 447         switch (cmd) {
 448         case IPC_RMID:
 449                 ipc_lock_object(&msq->q_perm);
 450                 /* freeque unlocks the ipc object and rcu */
 451                 freeque(ns, ipcp);
 452                 goto out_up;
 453         case IPC_SET:
 454                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 455                     !capable(CAP_SYS_RESOURCE)) {
 456                         err = -EPERM;
 457                         goto out_unlock1;
 458                 }
 459
 460                 ipc_lock_object(&msq->q_perm);
 461                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 462                 if (err)
 463                         goto out_unlock0;
 464
 465                 msq->q_qbytes = msqid64.msg_qbytes;
 466
 467                 msq->q_ctime = get_seconds();
 468                 /* sleeping receivers might be excluded by
 469                  * stricter permissions.
 470                  */
 471                 expunge_all(msq, -EAGAIN);
 472                 /* sleeping senders might be able to send
 473                  * due to a larger queue size.
 474                  */
 475                 ss_wakeup(&msq->q_senders, 0);
 476                 break;
 477         default:
 478                 err = -EINVAL;
 479                 goto out_unlock1;
 480         }
 481
 482 out_unlock0:
 483         ipc_unlock_object(&msq->q_perm);
 484 out_unlock1:
 485         rcu_read_unlock();
 486 out_up:
 487         up_write(&msg_ids(ns).rw_mutex);
 488         return err;
 489 }
 490
 491 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
 492                          int cmd, int version, void __user *buf)
 493 {
 494         int err;
 495         struct msg_queue *msq;
 496
 497         switch (cmd) {
 498         case IPC_INFO:
 499         case MSG_INFO:
 500         {
 501                 struct msginfo msginfo;
 502                 int max_id;
 503
 504                 if (!buf)
 505                         return -EFAULT;
 506
 507                 /*
 508                  * We must not return kernel stack data.
 509                  * due to padding, it's not enough
 510                  * to set all member fields.
 511                  */
 512                 err = security_msg_queue_msgctl(NULL, cmd);
 513                 if (err)
 514                         return err;
 515
 516                 memset(&msginfo, 0, sizeof(msginfo));
 517                 msginfo.msgmni = ns->msg_ctlmni;
 518                 msginfo.msgmax = ns->msg_ctlmax;
 519                 msginfo.msgmnb = ns->msg_ctlmnb;
 520                 msginfo.msgssz = MSGSSZ;
 521                 msginfo.msgseg = MSGSEG;
 522                 down_read(&msg_ids(ns).rw_mutex);
 523                 if (cmd == MSG_INFO) {
 524                         msginfo.msgpool = msg_ids(ns).in_use;
 525                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 526                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 527                 } else {
 528                         msginfo.msgmap = MSGMAP;
 529                         msginfo.msgpool = MSGPOOL;
 530                         msginfo.msgtql = MSGTQL;
 531                 }
 532                 max_id = ipc_get_maxid(&msg_ids(ns));
 533                 up_read(&msg_ids(ns).rw_mutex);
 534                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 535                         return -EFAULT;
 536                 return (max_id < 0) ? 0 : max_id;
 537         }
 538
 539         case MSG_STAT:
 540         case IPC_STAT:
 541         {
 542                 struct msqid64_ds tbuf;
 543                 int success_return;
 544
 545                 if (!buf)
 546                         return -EFAULT;
 547
 548                 if (cmd == MSG_STAT) {
 549                         msq = msg_lock(ns, msqid);
 550                         if (IS_ERR(msq))
 551                                 return PTR_ERR(msq);
 552                         success_return = msq->q_perm.id;
 553                 } else {
 554                         msq = msg_lock_check(ns, msqid);
 555                         if (IS_ERR(msq))
 556                                 return PTR_ERR(msq);
 557                         success_return = 0;
 558                 }
 559                 err = -EACCES;
 560                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 561                         goto out_unlock;
 562
 563                 err = security_msg_queue_msgctl(msq, cmd);
 564                 if (err)
 565                         goto out_unlock;
 566
 567                 memset(&tbuf, 0, sizeof(tbuf));
 568
 569                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 570                 tbuf.msg_stime  = msq->q_stime;
 571                 tbuf.msg_rtime  = msq->q_rtime;
 572                 tbuf.msg_ctime  = msq->q_ctime;
 573                 tbuf.msg_cbytes = msq->q_cbytes;
 574                 tbuf.msg_qnum   = msq->q_qnum;
 575                 tbuf.msg_qbytes = msq->q_qbytes;
 576                 tbuf.msg_lspid  = msq->q_lspid;
 577                 tbuf.msg_lrpid  = msq->q_lrpid;
 578                 msg_unlock(msq);
 579                 if (copy_msqid_to_user(buf, &tbuf, version))
 580                         return -EFAULT;
 581                 return success_return;
 582         }
 583
 584         default:
 585                 return -EINVAL;
 586         }
 587
 588         return err;
 589 out_unlock:
 590         msg_unlock(msq);
 591         return err;
 592 }
 593
 594 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 595 {
 596         int version;
 597         struct ipc_namespace *ns;
 598
 599         if (msqid < 0 || cmd < 0)
 600                 return -EINVAL;
 601
 602         version = ipc_parse_version(&cmd);
 603         ns = current->nsproxy->ipc_ns;
 604
 605         switch (cmd) {
 606         case IPC_INFO:
 607         case MSG_INFO:
 608         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 609         case IPC_STAT:
 610                 return msgctl_nolock(ns, msqid, cmd, version, buf);
 611         case IPC_SET:
 612         case IPC_RMID:
 613                 return msgctl_down(ns, msqid, cmd, buf, version);
 614         default:
 615                 return  -EINVAL;
 616         }
 617 }
 618
 619 static int testmsg(struct msg_msg *msg, long type, int mode)
 620 {
 621         switch(mode)
 622         {
 623                 case SEARCH_ANY:
 624                 case SEARCH_NUMBER:
 625                         return 1;
 626                 case SEARCH_LESSEQUAL:
 627                         if (msg->m_type <=type)
 628                                 return 1;
 629                         break;
 630                 case SEARCH_EQUAL:
 631                         if (msg->m_type == type)
 632                                 return 1;
 633                         break;
 634                 case SEARCH_NOTEQUAL:
 635                         if (msg->m_type != type)
 636                                 return 1;
 637                         break;
 638         }
 639         return 0;
 640 }
 641
 642 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 643 {
 644         struct msg_receiver *msr, *t;
 645
 646         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 647                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 648                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 649                                                msr->r_msgtype, msr->r_mode)) {
 650
 651                         list_del(&msr->r_list);
 652                         if (msr->r_maxsize < msg->m_ts) {
 653                                 msr->r_msg = NULL;
 654                                 wake_up_process(msr->r_tsk);
 655                                 smp_mb();
 656                                 msr->r_msg = ERR_PTR(-E2BIG);
 657                         } else {
 658                                 msr->r_msg = NULL;
 659                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 660                                 msq->q_rtime = get_seconds();
 661                                 wake_up_process(msr->r_tsk);
 662                                 smp_mb();
 663                                 msr->r_msg = msg;
 664
 665                                 return 1;
 666                         }
 667                 }
 668         }
 669         return 0;
 670 }
 671
 672 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 673                 size_t msgsz, int msgflg)
 674 {
 675         struct msg_queue *msq;
 676         struct msg_msg *msg;
 677         int err;
 678         struct ipc_namespace *ns;
 679
 680         ns = current->nsproxy->ipc_ns;
 681
 682         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 683                 return -EINVAL;
 684         if (mtype < 1)
 685                 return -EINVAL;
 686
 687         msg = load_msg(mtext, msgsz);
 688         if (IS_ERR(msg))
 689                 return PTR_ERR(msg);
 690
 691         msg->m_type = mtype;
 692         msg->m_ts = msgsz;
 693
 694         msq = msg_lock_check(ns, msqid);
 695         if (IS_ERR(msq)) {
 696                 err = PTR_ERR(msq);
 697                 goto out_free;
 698         }
 699
 700         for (;;) {
 701                 struct msg_sender s;
 702
 703                 err = -EACCES;
 704                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 705                         goto out_unlock_free;
 706
 707                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 708                 if (err)
 709                         goto out_unlock_free;
 710
 711                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 712                                 1 + msq->q_qnum <= msq->q_qbytes) {
 713                         break;
 714                 }
 715
 716                 /* queue full, wait: */
 717                 if (msgflg & IPC_NOWAIT) {
 718                         err = -EAGAIN;
 719                         goto out_unlock_free;
 720                 }
 721                 ss_add(msq, &s);
 722
 723                 if (!ipc_rcu_getref(msq)) {
 724                         err = -EIDRM;
 725                         goto out_unlock_free;
 726                 }
 727
 728                 msg_unlock(msq);
 729                 schedule();
 730
 731                 ipc_lock_by_ptr(&msq->q_perm);
 732                 ipc_rcu_putref(msq);
 733                 if (msq->q_perm.deleted) {
 734                         err = -EIDRM;
 735                         goto out_unlock_free;
 736                 }
 737                 ss_del(&s);
 738
 739                 if (signal_pending(current)) {
 740                         err = -ERESTARTNOHAND;
 741                         goto out_unlock_free;
 742                 }
 743         }
 744
 745         msq->q_lspid = task_tgid_vnr(current);
 746         msq->q_stime = get_seconds();
 747
 748         if (!pipelined_send(msq, msg)) {
 749                 /* no one is waiting for this message, enqueue it */
 750                 list_add_tail(&msg->m_list, &msq->q_messages);
 751                 msq->q_cbytes += msgsz;
 752                 msq->q_qnum++;
 753                 atomic_add(msgsz, &ns->msg_bytes);
 754                 atomic_inc(&ns->msg_hdrs);
 755         }
 756
 757         err = 0;
 758         msg = NULL;
 759
 760 out_unlock_free:
 761         msg_unlock(msq);
 762 out_free:
 763         if (msg != NULL)
 764                 free_msg(msg);
 765         return err;
 766 }
 767
 768 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 769                 int, msgflg)
 770 {
 771         long mtype;
 772
 773         if (get_user(mtype, &msgp->mtype))
 774                 return -EFAULT;
 775         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 776 }
 777
 778 static inline int convert_mode(long *msgtyp, int msgflg)
 779 {
 780         if (msgflg & MSG_COPY)
 781                 return SEARCH_NUMBER;
 782         /*
 783          *  find message of correct type.
 784          *  msgtyp = 0 => get first.
 785          *  msgtyp > 0 => get first message of matching type.
 786          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 787          */
 788         if (*msgtyp == 0)
 789                 return SEARCH_ANY;
 790         if (*msgtyp < 0) {
 791                 *msgtyp = -*msgtyp;
 792                 return SEARCH_LESSEQUAL;
 793         }
 794         if (msgflg & MSG_EXCEPT)
 795                 return SEARCH_NOTEQUAL;
 796         return SEARCH_EQUAL;
 797 }
 798
 799 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 800 {
 801         struct msgbuf __user *msgp = dest;
 802         size_t msgsz;
 803
 804         if (put_user(msg->m_type, &msgp->mtype))
 805                 return -EFAULT;
 806
 807         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 808         if (store_msg(msgp->mtext, msg, msgsz))
 809                 return -EFAULT;
 810         return msgsz;
 811 }
 812
 813 #ifdef CONFIG_CHECKPOINT_RESTORE
 814 /*
 815  * This function creates new kernel message structure, large enough to store
 816  * bufsz message bytes.
 817  */
 818 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 819 {
 820         struct msg_msg *copy;
 821
 822         /*
 823          * Create dummy message to copy real message to.
 824          */
 825         copy = load_msg(buf, bufsz);
 826         if (!IS_ERR(copy))
 827                 copy->m_ts = bufsz;
 828         return copy;
 829 }
 830
 831 static inline void free_copy(struct msg_msg *copy)
 832 {
 833         if (copy)
 834                 free_msg(copy);
 835 }
 836 #else
 837 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 838 {
 839         return ERR_PTR(-ENOSYS);
 840 }
 841
 842 static inline void free_copy(struct msg_msg *copy)
 843 {
 844 }
 845 #endif
 846
 847 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 848 {
 849         struct msg_msg *msg;
 850         long count = 0;
 851
 852         list_for_each_entry(msg, &msq->q_messages, m_list) {
 853                 if (testmsg(msg, *msgtyp, mode) &&
 854                     !security_msg_queue_msgrcv(msq, msg, current,
 855                                                *msgtyp, mode)) {
 856                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 857                                 *msgtyp = msg->m_type - 1;
 858                         } else if (mode == SEARCH_NUMBER) {
 859                                 if (*msgtyp == count)
 860                                         return msg;
 861                         } else
 862                                 return msg;
 863                         count++;
 864                 }
 865         }
 866
 867         return ERR_PTR(-EAGAIN);
 868 }
 869
 870
 871 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
 872                int msgflg,
 873                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 874 {
 875         struct msg_queue *msq;
 876         struct msg_msg *msg;
 877         int mode;
 878         struct ipc_namespace *ns;
 879         struct msg_msg *copy = NULL;
 880
 881         ns = current->nsproxy->ipc_ns;
 882
 883         if (msqid < 0 || (long) bufsz < 0)
 884                 return -EINVAL;
 885         if (msgflg & MSG_COPY) {
 886                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 887                 if (IS_ERR(copy))
 888                         return PTR_ERR(copy);
 889         }
 890         mode = convert_mode(&msgtyp, msgflg);
 891
 892         msq = msg_lock_check(ns, msqid);
 893         if (IS_ERR(msq)) {
 894                 free_copy(copy);
 895                 return PTR_ERR(msq);
 896         }
 897
 898         for (;;) {
 899                 struct msg_receiver msr_d;
 900
 901                 msg = ERR_PTR(-EACCES);
 902                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 903                         goto out_unlock;
 904
 905                 msg = find_msg(msq, &msgtyp, mode);
 906
 907                 if (!IS_ERR(msg)) {
 908                         /*
 909                          * Found a suitable message.
 910                          * Unlink it from the queue.
 911                          */
 912                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 913                                 msg = ERR_PTR(-E2BIG);
 914                                 goto out_unlock;
 915                         }
 916                         /*
 917                          * If we are copying, then do not unlink message and do
 918                          * not update queue parameters.
 919                          */
 920                         if (msgflg & MSG_COPY) {
 921                                 msg = copy_msg(msg, copy);
 922                                 goto out_unlock;
 923                         }
 924                         list_del(&msg->m_list);
 925                         msq->q_qnum--;
 926                         msq->q_rtime = get_seconds();
 927                         msq->q_lrpid = task_tgid_vnr(current);
 928                         msq->q_cbytes -= msg->m_ts;
 929                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 930                         atomic_dec(&ns->msg_hdrs);
 931                         ss_wakeup(&msq->q_senders, 0);
 932                         msg_unlock(msq);
 933                         break;
 934                 }
 935                 /* No message waiting. Wait for a message */
 936                 if (msgflg & IPC_NOWAIT) {
 937                         msg = ERR_PTR(-ENOMSG);
 938                         goto out_unlock;
 939                 }
 940                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 941                 msr_d.r_tsk = current;
 942                 msr_d.r_msgtype = msgtyp;
 943                 msr_d.r_mode = mode;
 944                 if (msgflg & MSG_NOERROR)
 945                         msr_d.r_maxsize = INT_MAX;
 946                 else
 947                         msr_d.r_maxsize = bufsz;
 948                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 949                 current->state = TASK_INTERRUPTIBLE;
 950                 msg_unlock(msq);
 951
 952                 schedule();
 953
 954                 /* Lockless receive, part 1:
 955                  * Disable preemption.  We don't hold a reference to the queue
 956                  * and getting a reference would defeat the idea of a lockless
 957                  * operation, thus the code relies on rcu to guarantee the
 958                  * existence of msq:
 959                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 960                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 961                  * rcu_read_lock() prevents preemption between reading r_msg
 962                  * and the spin_lock() inside ipc_lock_by_ptr().
 963                  */
 964                 rcu_read_lock();
 965
 966                 /* Lockless receive, part 2:
 967                  * Wait until pipelined_send or expunge_all are outside of
 968                  * wake_up_process(). There is a race with exit(), see
 969                  * ipc/mqueue.c for the details.
 970                  */
 971                 msg = (struct msg_msg*)msr_d.r_msg;
 972                 while (msg == NULL) {
 973                         cpu_relax();
 974                         msg = (struct msg_msg *)msr_d.r_msg;
 975                 }
 976
 977                 /* Lockless receive, part 3:
 978                  * If there is a message or an error then accept it without
 979                  * locking.
 980                  */
 981                 if (msg != ERR_PTR(-EAGAIN)) {
 982                         rcu_read_unlock();
 983                         break;
 984                 }
 985
 986                 /* Lockless receive, part 3:
 987                  * Acquire the queue spinlock.
 988                  */
 989                 ipc_lock_by_ptr(&msq->q_perm);
 990                 rcu_read_unlock();
 991
 992                 /* Lockless receive, part 4:
 993                  * Repeat test after acquiring the spinlock.
 994                  */
 995                 msg = (struct msg_msg*)msr_d.r_msg;
 996                 if (msg != ERR_PTR(-EAGAIN))
 997                         goto out_unlock;
 998
 999                 list_del(&msr_d.r_list);
1000                 if (signal_pending(current)) {
1001                         msg = ERR_PTR(-ERESTARTNOHAND);
1002 out_unlock:
1003                         msg_unlock(msq);
1004                         break;
1005                 }
1006         }
1007         if (IS_ERR(msg)) {
1008                 free_copy(copy);
1009                 return PTR_ERR(msg);
1010         }
1011
1012         bufsz = msg_handler(buf, msg, bufsz);
1013         free_msg(msg);
1014
1015         return bufsz;
1016 }
1017
1018 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1019                 long, msgtyp, int, msgflg)
1020 {
1021         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1022 }
1023
1024 #ifdef CONFIG_PROC_FS
1025 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1026 {
1027         struct user_namespace *user_ns = seq_user_ns(s);
1028         struct msg_queue *msq = it;
1029
1030         return seq_printf(s,
1031                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1032                         msq->q_perm.key,
1033                         msq->q_perm.id,
1034                         msq->q_perm.mode,
1035                         msq->q_cbytes,
1036                         msq->q_qnum,
1037                         msq->q_lspid,
1038                         msq->q_lrpid,
1039                         from_kuid_munged(user_ns, msq->q_perm.uid),
1040                         from_kgid_munged(user_ns, msq->q_perm.gid),
1041                         from_kuid_munged(user_ns, msq->q_perm.cuid),
1042                         from_kgid_munged(user_ns, msq->q_perm.cgid),
1043                         msq->q_stime,
1044                         msq->q_rtime,
1045                         msq->q_ctime);
1046 }
1047 #endif