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 <linux/uaccess.h>
  43 #include "util.h"
  44
  45 /* one msg_receiver structure for each sleeping receiver */
  46 struct msg_receiver {
  47         struct list_head        r_list;
  48         struct task_struct      *r_tsk;
  49
  50         int                     r_mode;
  51         long                    r_msgtype;
  52         long                    r_maxsize;
  53
  54         struct msg_msg          *r_msg;
  55 };
  56
  57 /* one msg_sender for each sleeping sender */
  58 struct msg_sender {
  59         struct list_head        list;
  60         struct task_struct      *tsk;
  61 };
  62
  63 #define SEARCH_ANY              1
  64 #define SEARCH_EQUAL            2
  65 #define SEARCH_NOTEQUAL         3
  66 #define SEARCH_LESSEQUAL        4
  67 #define SEARCH_NUMBER           5
  68
  69 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  70
  71 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
  72 {
  73         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
  74
  75         if (IS_ERR(ipcp))
  76                 return ERR_CAST(ipcp);
  77
  78         return container_of(ipcp, struct msg_queue, q_perm);
  79 }
  80
  81 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
  82                                                         int id)
  83 {
  84         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
  85
  86         if (IS_ERR(ipcp))
  87                 return ERR_CAST(ipcp);
  88
  89         return container_of(ipcp, struct msg_queue, q_perm);
  90 }
  91
  92 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
  93 {
  94         ipc_rmid(&msg_ids(ns), &s->q_perm);
  95 }
  96
  97 static void msg_rcu_free(struct rcu_head *head)
  98 {
  99         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
 100         struct msg_queue *msq = ipc_rcu_to_struct(p);
 101
 102         security_msg_queue_free(msq);
 103         ipc_rcu_free(head);
 104 }
 105
 106 /**
 107  * newque - Create a new msg queue
 108  * @ns: namespace
 109  * @params: ptr to the structure that contains the key and msgflg
 110  *
 111  * Called with msg_ids.rwsem held (writer)
 112  */
 113 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 114 {
 115         struct msg_queue *msq;
 116         int id, retval;
 117         key_t key = params->key;
 118         int msgflg = params->flg;
 119
 120         msq = ipc_rcu_alloc(sizeof(*msq));
 121         if (!msq)
 122                 return -ENOMEM;
 123
 124         msq->q_perm.mode = msgflg & S_IRWXUGO;
 125         msq->q_perm.key = key;
 126
 127         msq->q_perm.security = NULL;
 128         retval = security_msg_queue_alloc(msq);
 129         if (retval) {
 130                 ipc_rcu_putref(msq, ipc_rcu_free);
 131                 return retval;
 132         }
 133
 134         msq->q_stime = msq->q_rtime = 0;
 135         msq->q_ctime = get_seconds();
 136         msq->q_cbytes = msq->q_qnum = 0;
 137         msq->q_qbytes = ns->msg_ctlmnb;
 138         msq->q_lspid = msq->q_lrpid = 0;
 139         INIT_LIST_HEAD(&msq->q_messages);
 140         INIT_LIST_HEAD(&msq->q_receivers);
 141         INIT_LIST_HEAD(&msq->q_senders);
 142
 143         /* ipc_addid() locks msq upon success. */
 144         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 145         if (id < 0) {
 146                 ipc_rcu_putref(msq, msg_rcu_free);
 147                 return id;
 148         }
 149
 150         ipc_unlock_object(&msq->q_perm);
 151         rcu_read_unlock();
 152
 153         return msq->q_perm.id;
 154 }
 155
 156 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 157 {
 158         mss->tsk = current;
 159         __set_current_state(TASK_INTERRUPTIBLE);
 160         list_add_tail(&mss->list, &msq->q_senders);
 161 }
 162
 163 static inline void ss_del(struct msg_sender *mss)
 164 {
 165         if (mss->list.next != NULL)
 166                 list_del(&mss->list);
 167 }
 168
 169 static void ss_wakeup(struct list_head *h, int kill)
 170 {
 171         struct msg_sender *mss, *t;
 172
 173         list_for_each_entry_safe(mss, t, h, list) {
 174                 if (kill)
 175                         mss->list.next = NULL;
 176                 wake_up_process(mss->tsk);
 177         }
 178 }
 179
 180 static void expunge_all(struct msg_queue *msq, int res,
 181                         struct wake_q_head *wake_q)
 182 {
 183         struct msg_receiver *msr, *t;
 184
 185         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 186                 wake_q_add(wake_q, msr->r_tsk);
 187                 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
 188         }
 189 }
 190
 191 /*
 192  * freeque() wakes up waiters on the sender and receiver waiting queue,
 193  * removes the message queue from message queue ID IDR, and cleans up all the
 194  * messages associated with this queue.
 195  *
 196  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
 197  * before freeque() is called. msg_ids.rwsem remains locked on exit.
 198  */
 199 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 200 {
 201         struct msg_msg *msg, *t;
 202         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 203         WAKE_Q(wake_q);
 204
 205         expunge_all(msq, -EIDRM, &wake_q);
 206         ss_wakeup(&msq->q_senders, 1);
 207         msg_rmid(ns, msq);
 208         ipc_unlock_object(&msq->q_perm);
 209         wake_up_q(&wake_q);
 210         rcu_read_unlock();
 211
 212         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 213                 atomic_dec(&ns->msg_hdrs);
 214                 free_msg(msg);
 215         }
 216         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 217         ipc_rcu_putref(msq, msg_rcu_free);
 218 }
 219
 220 /*
 221  * Called with msg_ids.rwsem and ipcp locked.
 222  */
 223 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 224 {
 225         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 226
 227         return security_msg_queue_associate(msq, msgflg);
 228 }
 229
 230 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 231 {
 232         struct ipc_namespace *ns;
 233         static const struct ipc_ops msg_ops = {
 234                 .getnew = newque,
 235                 .associate = msg_security,
 236         };
 237         struct ipc_params msg_params;
 238
 239         ns = current->nsproxy->ipc_ns;
 240
 241         msg_params.key = key;
 242         msg_params.flg = msgflg;
 243
 244         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 245 }
 246
 247 static inline unsigned long
 248 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 249 {
 250         switch (version) {
 251         case IPC_64:
 252                 return copy_to_user(buf, in, sizeof(*in));
 253         case IPC_OLD:
 254         {
 255                 struct msqid_ds out;
 256
 257                 memset(&out, 0, sizeof(out));
 258
 259                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 260
 261                 out.msg_stime           = in->msg_stime;
 262                 out.msg_rtime           = in->msg_rtime;
 263                 out.msg_ctime           = in->msg_ctime;
 264
 265                 if (in->msg_cbytes > USHRT_MAX)
 266                         out.msg_cbytes  = USHRT_MAX;
 267                 else
 268                         out.msg_cbytes  = in->msg_cbytes;
 269                 out.msg_lcbytes         = in->msg_cbytes;
 270
 271                 if (in->msg_qnum > USHRT_MAX)
 272                         out.msg_qnum    = USHRT_MAX;
 273                 else
 274                         out.msg_qnum    = in->msg_qnum;
 275
 276                 if (in->msg_qbytes > USHRT_MAX)
 277                         out.msg_qbytes  = USHRT_MAX;
 278                 else
 279                         out.msg_qbytes  = in->msg_qbytes;
 280                 out.msg_lqbytes         = in->msg_qbytes;
 281
 282                 out.msg_lspid           = in->msg_lspid;
 283                 out.msg_lrpid           = in->msg_lrpid;
 284
 285                 return copy_to_user(buf, &out, sizeof(out));
 286         }
 287         default:
 288                 return -EINVAL;
 289         }
 290 }
 291
 292 static inline unsigned long
 293 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 294 {
 295         switch (version) {
 296         case IPC_64:
 297                 if (copy_from_user(out, buf, sizeof(*out)))
 298                         return -EFAULT;
 299                 return 0;
 300         case IPC_OLD:
 301         {
 302                 struct msqid_ds tbuf_old;
 303
 304                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 305                         return -EFAULT;
 306
 307                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 308                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 309                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 310
 311                 if (tbuf_old.msg_qbytes == 0)
 312                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 313                 else
 314                         out->msg_qbytes = tbuf_old.msg_qbytes;
 315
 316                 return 0;
 317         }
 318         default:
 319                 return -EINVAL;
 320         }
 321 }
 322
 323 /*
 324  * This function handles some msgctl commands which require the rwsem
 325  * to be held in write mode.
 326  * NOTE: no locks must be held, the rwsem is taken inside this function.
 327  */
 328 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 329                        struct msqid_ds __user *buf, int version)
 330 {
 331         struct kern_ipc_perm *ipcp;
 332         struct msqid64_ds uninitialized_var(msqid64);
 333         struct msg_queue *msq;
 334         WAKE_Q(wake_q);
 335         int err;
 336
 337         if (cmd == IPC_SET) {
 338                 if (copy_msqid_from_user(&msqid64, buf, version))
 339                         return -EFAULT;
 340         }
 341
 342         down_write(&msg_ids(ns).rwsem);
 343         rcu_read_lock();
 344
 345         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 346                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 347         if (IS_ERR(ipcp)) {
 348                 err = PTR_ERR(ipcp);
 349                 goto out_unlock1;
 350         }
 351
 352         msq = container_of(ipcp, struct msg_queue, q_perm);
 353
 354         err = security_msg_queue_msgctl(msq, cmd);
 355         if (err)
 356                 goto out_unlock1;
 357
 358         switch (cmd) {
 359         case IPC_RMID:
 360                 ipc_lock_object(&msq->q_perm);
 361                 /* freeque unlocks the ipc object and rcu */
 362                 freeque(ns, ipcp);
 363                 goto out_up;
 364         case IPC_SET:
 365                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 366                     !capable(CAP_SYS_RESOURCE)) {
 367                         err = -EPERM;
 368                         goto out_unlock1;
 369                 }
 370
 371                 ipc_lock_object(&msq->q_perm);
 372                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 373                 if (err)
 374                         goto out_unlock0;
 375
 376                 msq->q_qbytes = msqid64.msg_qbytes;
 377
 378                 msq->q_ctime = get_seconds();
 379                 /* sleeping receivers might be excluded by
 380                  * stricter permissions.
 381                  */
 382                 expunge_all(msq, -EAGAIN, &wake_q);
 383                 /* sleeping senders might be able to send
 384                  * due to a larger queue size.
 385                  */
 386                 ss_wakeup(&msq->q_senders, 0);
 387                 break;
 388         default:
 389                 err = -EINVAL;
 390                 goto out_unlock1;
 391         }
 392
 393 out_unlock0:
 394         ipc_unlock_object(&msq->q_perm);
 395         wake_up_q(&wake_q);
 396 out_unlock1:
 397         rcu_read_unlock();
 398 out_up:
 399         up_write(&msg_ids(ns).rwsem);
 400         return err;
 401 }
 402
 403 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
 404                          int cmd, int version, void __user *buf)
 405 {
 406         int err;
 407         struct msg_queue *msq;
 408
 409         switch (cmd) {
 410         case IPC_INFO:
 411         case MSG_INFO:
 412         {
 413                 struct msginfo msginfo;
 414                 int max_id;
 415
 416                 if (!buf)
 417                         return -EFAULT;
 418
 419                 /*
 420                  * We must not return kernel stack data.
 421                  * due to padding, it's not enough
 422                  * to set all member fields.
 423                  */
 424                 err = security_msg_queue_msgctl(NULL, cmd);
 425                 if (err)
 426                         return err;
 427
 428                 memset(&msginfo, 0, sizeof(msginfo));
 429                 msginfo.msgmni = ns->msg_ctlmni;
 430                 msginfo.msgmax = ns->msg_ctlmax;
 431                 msginfo.msgmnb = ns->msg_ctlmnb;
 432                 msginfo.msgssz = MSGSSZ;
 433                 msginfo.msgseg = MSGSEG;
 434                 down_read(&msg_ids(ns).rwsem);
 435                 if (cmd == MSG_INFO) {
 436                         msginfo.msgpool = msg_ids(ns).in_use;
 437                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 438                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 439                 } else {
 440                         msginfo.msgmap = MSGMAP;
 441                         msginfo.msgpool = MSGPOOL;
 442                         msginfo.msgtql = MSGTQL;
 443                 }
 444                 max_id = ipc_get_maxid(&msg_ids(ns));
 445                 up_read(&msg_ids(ns).rwsem);
 446                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 447                         return -EFAULT;
 448                 return (max_id < 0) ? 0 : max_id;
 449         }
 450
 451         case MSG_STAT:
 452         case IPC_STAT:
 453         {
 454                 struct msqid64_ds tbuf;
 455                 int success_return;
 456
 457                 if (!buf)
 458                         return -EFAULT;
 459
 460                 memset(&tbuf, 0, sizeof(tbuf));
 461
 462                 rcu_read_lock();
 463                 if (cmd == MSG_STAT) {
 464                         msq = msq_obtain_object(ns, msqid);
 465                         if (IS_ERR(msq)) {
 466                                 err = PTR_ERR(msq);
 467                                 goto out_unlock;
 468                         }
 469                         success_return = msq->q_perm.id;
 470                 } else {
 471                         msq = msq_obtain_object_check(ns, msqid);
 472                         if (IS_ERR(msq)) {
 473                                 err = PTR_ERR(msq);
 474                                 goto out_unlock;
 475                         }
 476                         success_return = 0;
 477                 }
 478
 479                 err = -EACCES;
 480                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 481                         goto out_unlock;
 482
 483                 err = security_msg_queue_msgctl(msq, cmd);
 484                 if (err)
 485                         goto out_unlock;
 486
 487                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 488                 tbuf.msg_stime  = msq->q_stime;
 489                 tbuf.msg_rtime  = msq->q_rtime;
 490                 tbuf.msg_ctime  = msq->q_ctime;
 491                 tbuf.msg_cbytes = msq->q_cbytes;
 492                 tbuf.msg_qnum   = msq->q_qnum;
 493                 tbuf.msg_qbytes = msq->q_qbytes;
 494                 tbuf.msg_lspid  = msq->q_lspid;
 495                 tbuf.msg_lrpid  = msq->q_lrpid;
 496                 rcu_read_unlock();
 497
 498                 if (copy_msqid_to_user(buf, &tbuf, version))
 499                         return -EFAULT;
 500                 return success_return;
 501         }
 502
 503         default:
 504                 return -EINVAL;
 505         }
 506
 507         return err;
 508 out_unlock:
 509         rcu_read_unlock();
 510         return err;
 511 }
 512
 513 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 514 {
 515         int version;
 516         struct ipc_namespace *ns;
 517
 518         if (msqid < 0 || cmd < 0)
 519                 return -EINVAL;
 520
 521         version = ipc_parse_version(&cmd);
 522         ns = current->nsproxy->ipc_ns;
 523
 524         switch (cmd) {
 525         case IPC_INFO:
 526         case MSG_INFO:
 527         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 528         case IPC_STAT:
 529                 return msgctl_nolock(ns, msqid, cmd, version, buf);
 530         case IPC_SET:
 531         case IPC_RMID:
 532                 return msgctl_down(ns, msqid, cmd, buf, version);
 533         default:
 534                 return  -EINVAL;
 535         }
 536 }
 537
 538 static int testmsg(struct msg_msg *msg, long type, int mode)
 539 {
 540         switch (mode) {
 541         case SEARCH_ANY:
 542         case SEARCH_NUMBER:
 543                 return 1;
 544         case SEARCH_LESSEQUAL:
 545                 if (msg->m_type <= type)
 546                         return 1;
 547                 break;
 548         case SEARCH_EQUAL:
 549                 if (msg->m_type == type)
 550                         return 1;
 551                 break;
 552         case SEARCH_NOTEQUAL:
 553                 if (msg->m_type != type)
 554                         return 1;
 555                 break;
 556         }
 557         return 0;
 558 }
 559
 560 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
 561                                  struct wake_q_head *wake_q)
 562 {
 563         struct msg_receiver *msr, *t;
 564
 565         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 566                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 567                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 568                                                msr->r_msgtype, msr->r_mode)) {
 569
 570                         list_del(&msr->r_list);
 571                         if (msr->r_maxsize < msg->m_ts) {
 572                                 wake_q_add(wake_q, msr->r_tsk);
 573                                 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
 574                         } else {
 575                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 576                                 msq->q_rtime = get_seconds();
 577
 578                                 wake_q_add(wake_q, msr->r_tsk);
 579                                 WRITE_ONCE(msr->r_msg, msg);
 580                                 return 1;
 581                         }
 582                 }
 583         }
 584
 585         return 0;
 586 }
 587
 588 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 589                 size_t msgsz, int msgflg)
 590 {
 591         struct msg_queue *msq;
 592         struct msg_msg *msg;
 593         int err;
 594         struct ipc_namespace *ns;
 595         WAKE_Q(wake_q);
 596
 597         ns = current->nsproxy->ipc_ns;
 598
 599         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 600                 return -EINVAL;
 601         if (mtype < 1)
 602                 return -EINVAL;
 603
 604         msg = load_msg(mtext, msgsz);
 605         if (IS_ERR(msg))
 606                 return PTR_ERR(msg);
 607
 608         msg->m_type = mtype;
 609         msg->m_ts = msgsz;
 610
 611         rcu_read_lock();
 612         msq = msq_obtain_object_check(ns, msqid);
 613         if (IS_ERR(msq)) {
 614                 err = PTR_ERR(msq);
 615                 goto out_unlock1;
 616         }
 617
 618         ipc_lock_object(&msq->q_perm);
 619
 620         for (;;) {
 621                 struct msg_sender s;
 622
 623                 err = -EACCES;
 624                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 625                         goto out_unlock0;
 626
 627                 /* raced with RMID? */
 628                 if (!ipc_valid_object(&msq->q_perm)) {
 629                         err = -EIDRM;
 630                         goto out_unlock0;
 631                 }
 632
 633                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 634                 if (err)
 635                         goto out_unlock0;
 636
 637                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 638                                 1 + msq->q_qnum <= msq->q_qbytes) {
 639                         break;
 640                 }
 641
 642                 /* queue full, wait: */
 643                 if (msgflg & IPC_NOWAIT) {
 644                         err = -EAGAIN;
 645                         goto out_unlock0;
 646                 }
 647
 648                 /* enqueue the sender and prepare to block */
 649                 ss_add(msq, &s);
 650
 651                 if (!ipc_rcu_getref(msq)) {
 652                         err = -EIDRM;
 653                         goto out_unlock0;
 654                 }
 655
 656                 ipc_unlock_object(&msq->q_perm);
 657                 rcu_read_unlock();
 658                 schedule();
 659
 660                 rcu_read_lock();
 661                 ipc_lock_object(&msq->q_perm);
 662
 663                 ipc_rcu_putref(msq, msg_rcu_free);
 664                 /* raced with RMID? */
 665                 if (!ipc_valid_object(&msq->q_perm)) {
 666                         err = -EIDRM;
 667                         goto out_unlock0;
 668                 }
 669                 ss_del(&s);
 670
 671                 if (signal_pending(current)) {
 672                         err = -ERESTARTNOHAND;
 673                         goto out_unlock0;
 674                 }
 675
 676         }
 677         msq->q_lspid = task_tgid_vnr(current);
 678         msq->q_stime = get_seconds();
 679
 680         if (!pipelined_send(msq, msg, &wake_q)) {
 681                 /* no one is waiting for this message, enqueue it */
 682                 list_add_tail(&msg->m_list, &msq->q_messages);
 683                 msq->q_cbytes += msgsz;
 684                 msq->q_qnum++;
 685                 atomic_add(msgsz, &ns->msg_bytes);
 686                 atomic_inc(&ns->msg_hdrs);
 687         }
 688
 689         err = 0;
 690         msg = NULL;
 691
 692 out_unlock0:
 693         ipc_unlock_object(&msq->q_perm);
 694         wake_up_q(&wake_q);
 695 out_unlock1:
 696         rcu_read_unlock();
 697         if (msg != NULL)
 698                 free_msg(msg);
 699         return err;
 700 }
 701
 702 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 703                 int, msgflg)
 704 {
 705         long mtype;
 706
 707         if (get_user(mtype, &msgp->mtype))
 708                 return -EFAULT;
 709         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 710 }
 711
 712 static inline int convert_mode(long *msgtyp, int msgflg)
 713 {
 714         if (msgflg & MSG_COPY)
 715                 return SEARCH_NUMBER;
 716         /*
 717          *  find message of correct type.
 718          *  msgtyp = 0 => get first.
 719          *  msgtyp > 0 => get first message of matching type.
 720          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 721          */
 722         if (*msgtyp == 0)
 723                 return SEARCH_ANY;
 724         if (*msgtyp < 0) {
 725                 *msgtyp = -*msgtyp;
 726                 return SEARCH_LESSEQUAL;
 727         }
 728         if (msgflg & MSG_EXCEPT)
 729                 return SEARCH_NOTEQUAL;
 730         return SEARCH_EQUAL;
 731 }
 732
 733 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 734 {
 735         struct msgbuf __user *msgp = dest;
 736         size_t msgsz;
 737
 738         if (put_user(msg->m_type, &msgp->mtype))
 739                 return -EFAULT;
 740
 741         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 742         if (store_msg(msgp->mtext, msg, msgsz))
 743                 return -EFAULT;
 744         return msgsz;
 745 }
 746
 747 #ifdef CONFIG_CHECKPOINT_RESTORE
 748 /*
 749  * This function creates new kernel message structure, large enough to store
 750  * bufsz message bytes.
 751  */
 752 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 753 {
 754         struct msg_msg *copy;
 755
 756         /*
 757          * Create dummy message to copy real message to.
 758          */
 759         copy = load_msg(buf, bufsz);
 760         if (!IS_ERR(copy))
 761                 copy->m_ts = bufsz;
 762         return copy;
 763 }
 764
 765 static inline void free_copy(struct msg_msg *copy)
 766 {
 767         if (copy)
 768                 free_msg(copy);
 769 }
 770 #else
 771 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 772 {
 773         return ERR_PTR(-ENOSYS);
 774 }
 775
 776 static inline void free_copy(struct msg_msg *copy)
 777 {
 778 }
 779 #endif
 780
 781 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 782 {
 783         struct msg_msg *msg, *found = NULL;
 784         long count = 0;
 785
 786         list_for_each_entry(msg, &msq->q_messages, m_list) {
 787                 if (testmsg(msg, *msgtyp, mode) &&
 788                     !security_msg_queue_msgrcv(msq, msg, current,
 789                                                *msgtyp, mode)) {
 790                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 791                                 *msgtyp = msg->m_type - 1;
 792                                 found = msg;
 793                         } else if (mode == SEARCH_NUMBER) {
 794                                 if (*msgtyp == count)
 795                                         return msg;
 796                         } else
 797                                 return msg;
 798                         count++;
 799                 }
 800         }
 801
 802         return found ?: ERR_PTR(-EAGAIN);
 803 }
 804
 805 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
 806                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 807 {
 808         int mode;
 809         struct msg_queue *msq;
 810         struct ipc_namespace *ns;
 811         struct msg_msg *msg, *copy = NULL;
 812
 813         ns = current->nsproxy->ipc_ns;
 814
 815         if (msqid < 0 || (long) bufsz < 0)
 816                 return -EINVAL;
 817
 818         if (msgflg & MSG_COPY) {
 819                 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
 820                         return -EINVAL;
 821                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 822                 if (IS_ERR(copy))
 823                         return PTR_ERR(copy);
 824         }
 825         mode = convert_mode(&msgtyp, msgflg);
 826
 827         rcu_read_lock();
 828         msq = msq_obtain_object_check(ns, msqid);
 829         if (IS_ERR(msq)) {
 830                 rcu_read_unlock();
 831                 free_copy(copy);
 832                 return PTR_ERR(msq);
 833         }
 834
 835         for (;;) {
 836                 struct msg_receiver msr_d;
 837
 838                 msg = ERR_PTR(-EACCES);
 839                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 840                         goto out_unlock1;
 841
 842                 ipc_lock_object(&msq->q_perm);
 843
 844                 /* raced with RMID? */
 845                 if (!ipc_valid_object(&msq->q_perm)) {
 846                         msg = ERR_PTR(-EIDRM);
 847                         goto out_unlock0;
 848                 }
 849
 850                 msg = find_msg(msq, &msgtyp, mode);
 851                 if (!IS_ERR(msg)) {
 852                         /*
 853                          * Found a suitable message.
 854                          * Unlink it from the queue.
 855                          */
 856                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 857                                 msg = ERR_PTR(-E2BIG);
 858                                 goto out_unlock0;
 859                         }
 860                         /*
 861                          * If we are copying, then do not unlink message and do
 862                          * not update queue parameters.
 863                          */
 864                         if (msgflg & MSG_COPY) {
 865                                 msg = copy_msg(msg, copy);
 866                                 goto out_unlock0;
 867                         }
 868
 869                         list_del(&msg->m_list);
 870                         msq->q_qnum--;
 871                         msq->q_rtime = get_seconds();
 872                         msq->q_lrpid = task_tgid_vnr(current);
 873                         msq->q_cbytes -= msg->m_ts;
 874                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 875                         atomic_dec(&ns->msg_hdrs);
 876                         ss_wakeup(&msq->q_senders, 0);
 877
 878                         goto out_unlock0;
 879                 }
 880
 881                 /* No message waiting. Wait for a message */
 882                 if (msgflg & IPC_NOWAIT) {
 883                         msg = ERR_PTR(-ENOMSG);
 884                         goto out_unlock0;
 885                 }
 886
 887                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 888                 msr_d.r_tsk = current;
 889                 msr_d.r_msgtype = msgtyp;
 890                 msr_d.r_mode = mode;
 891                 if (msgflg & MSG_NOERROR)
 892                         msr_d.r_maxsize = INT_MAX;
 893                 else
 894                         msr_d.r_maxsize = bufsz;
 895                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 896                 __set_current_state(TASK_INTERRUPTIBLE);
 897
 898                 ipc_unlock_object(&msq->q_perm);
 899                 rcu_read_unlock();
 900                 schedule();
 901
 902                 /*
 903                  * Lockless receive, part 1:
 904                  * We don't hold a reference to the queue and getting a
 905                  * reference would defeat the idea of a lockless operation,
 906                  * thus the code relies on rcu to guarantee the existence of
 907                  * msq:
 908                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 909                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 910                  */
 911                 rcu_read_lock();
 912
 913                 /*
 914                  * Lockless receive, part 2:
 915                  * The work in pipelined_send() and expunge_all():
 916                  * - Set pointer to message
 917                  * - Queue the receiver task for later wakeup
 918                  * - Wake up the process after the lock is dropped.
 919                  *
 920                  * Should the process wake up before this wakeup (due to a
 921                  * signal) it will either see the message and continue ...
 922                  */
 923                 msg = READ_ONCE(msr_d.r_msg);
 924                 if (msg != ERR_PTR(-EAGAIN))
 925                         goto out_unlock1;
 926
 927                  /*
 928                   * ... or see -EAGAIN, acquire the lock to check the message
 929                   * again.
 930                   */
 931                 ipc_lock_object(&msq->q_perm);
 932
 933                 msg = msr_d.r_msg;
 934                 if (msg != ERR_PTR(-EAGAIN))
 935                         goto out_unlock0;
 936
 937                 list_del(&msr_d.r_list);
 938                 if (signal_pending(current)) {
 939                         msg = ERR_PTR(-ERESTARTNOHAND);
 940                         goto out_unlock0;
 941                 }
 942
 943                 ipc_unlock_object(&msq->q_perm);
 944         }
 945
 946 out_unlock0:
 947         ipc_unlock_object(&msq->q_perm);
 948 out_unlock1:
 949         rcu_read_unlock();
 950         if (IS_ERR(msg)) {
 951                 free_copy(copy);
 952                 return PTR_ERR(msg);
 953         }
 954
 955         bufsz = msg_handler(buf, msg, bufsz);
 956         free_msg(msg);
 957
 958         return bufsz;
 959 }
 960
 961 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 962                 long, msgtyp, int, msgflg)
 963 {
 964         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
 965 }
 966
 967
 968 void msg_init_ns(struct ipc_namespace *ns)
 969 {
 970         ns->msg_ctlmax = MSGMAX;
 971         ns->msg_ctlmnb = MSGMNB;
 972         ns->msg_ctlmni = MSGMNI;
 973
 974         atomic_set(&ns->msg_bytes, 0);
 975         atomic_set(&ns->msg_hdrs, 0);
 976         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 977 }
 978
 979 #ifdef CONFIG_IPC_NS
 980 void msg_exit_ns(struct ipc_namespace *ns)
 981 {
 982         free_ipcs(ns, &msg_ids(ns), freeque);
 983         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
 984 }
 985 #endif
 986
 987 #ifdef CONFIG_PROC_FS
 988 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 989 {
 990         struct user_namespace *user_ns = seq_user_ns(s);
 991         struct msg_queue *msq = it;
 992
 993         seq_printf(s,
 994                    "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 995                    msq->q_perm.key,
 996                    msq->q_perm.id,
 997                    msq->q_perm.mode,
 998                    msq->q_cbytes,
 999                    msq->q_qnum,
1000                    msq->q_lspid,
1001                    msq->q_lrpid,
1002                    from_kuid_munged(user_ns, msq->q_perm.uid),
1003                    from_kgid_munged(user_ns, msq->q_perm.gid),
1004                    from_kuid_munged(user_ns, msq->q_perm.cuid),
1005                    from_kgid_munged(user_ns, msq->q_perm.cgid),
1006                    msq->q_stime,
1007                    msq->q_rtime,
1008                    msq->q_ctime);
1009
1010         return 0;
1011 }
1012 #endif
1013
1014 void __init msg_init(void)
1015 {
1016         msg_init_ns(&init_ipc_ns);
1017
1018         ipc_init_proc_interface("sysvipc/msg",
1019                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1020                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
1021 }