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         size_t                  msgsz;
  62 };
  63
  64 #define SEARCH_ANY              1
  65 #define SEARCH_EQUAL            2
  66 #define SEARCH_NOTEQUAL         3
  67 #define SEARCH_LESSEQUAL        4
  68 #define SEARCH_NUMBER           5
  69
  70 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  71
  72 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
  73 {
  74         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
  75
  76         if (IS_ERR(ipcp))
  77                 return ERR_CAST(ipcp);
  78
  79         return container_of(ipcp, struct msg_queue, q_perm);
  80 }
  81
  82 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
  83                                                         int id)
  84 {
  85         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
  86
  87         if (IS_ERR(ipcp))
  88                 return ERR_CAST(ipcp);
  89
  90         return container_of(ipcp, struct msg_queue, q_perm);
  91 }
  92
  93 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
  94 {
  95         ipc_rmid(&msg_ids(ns), &s->q_perm);
  96 }
  97
  98 static void msg_rcu_free(struct rcu_head *head)
  99 {
 100         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
 101         struct msg_queue *msq = ipc_rcu_to_struct(p);
 102
 103         security_msg_queue_free(msq);
 104         ipc_rcu_free(head);
 105 }
 106
 107 /**
 108  * newque - Create a new msg queue
 109  * @ns: namespace
 110  * @params: ptr to the structure that contains the key and msgflg
 111  *
 112  * Called with msg_ids.rwsem held (writer)
 113  */
 114 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 115 {
 116         struct msg_queue *msq;
 117         int id, retval;
 118         key_t key = params->key;
 119         int msgflg = params->flg;
 120
 121         msq = ipc_rcu_alloc(sizeof(*msq));
 122         if (!msq)
 123                 return -ENOMEM;
 124
 125         msq->q_perm.mode = msgflg & S_IRWXUGO;
 126         msq->q_perm.key = key;
 127
 128         msq->q_perm.security = NULL;
 129         retval = security_msg_queue_alloc(msq);
 130         if (retval) {
 131                 ipc_rcu_putref(msq, ipc_rcu_free);
 132                 return retval;
 133         }
 134
 135         msq->q_stime = msq->q_rtime = 0;
 136         msq->q_ctime = get_seconds();
 137         msq->q_cbytes = msq->q_qnum = 0;
 138         msq->q_qbytes = ns->msg_ctlmnb;
 139         msq->q_lspid = msq->q_lrpid = 0;
 140         INIT_LIST_HEAD(&msq->q_messages);
 141         INIT_LIST_HEAD(&msq->q_receivers);
 142         INIT_LIST_HEAD(&msq->q_senders);
 143
 144         /* ipc_addid() locks msq upon success. */
 145         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 146         if (id < 0) {
 147                 ipc_rcu_putref(msq, msg_rcu_free);
 148                 return id;
 149         }
 150
 151         ipc_unlock_object(&msq->q_perm);
 152         rcu_read_unlock();
 153
 154         return msq->q_perm.id;
 155 }
 156
 157 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
 158 {
 159         return msgsz + msq->q_cbytes <= msq->q_qbytes &&
 160                 1 + msq->q_qnum <= msq->q_qbytes;
 161 }
 162
 163 static inline void ss_add(struct msg_queue *msq,
 164                           struct msg_sender *mss, size_t msgsz)
 165 {
 166         mss->tsk = current;
 167         mss->msgsz = msgsz;
 168         __set_current_state(TASK_INTERRUPTIBLE);
 169         list_add_tail(&mss->list, &msq->q_senders);
 170 }
 171
 172 static inline void ss_del(struct msg_sender *mss)
 173 {
 174         if (mss->list.next)
 175                 list_del(&mss->list);
 176 }
 177
 178 static void ss_wakeup(struct msg_queue *msq,
 179                       struct wake_q_head *wake_q, bool kill)
 180 {
 181         struct msg_sender *mss, *t;
 182         struct task_struct *stop_tsk = NULL;
 183         struct list_head *h = &msq->q_senders;
 184
 185         list_for_each_entry_safe(mss, t, h, list) {
 186                 if (kill)
 187                         mss->list.next = NULL;
 188
 189                 /*
 190                  * Stop at the first task we don't wakeup,
 191                  * we've already iterated the original
 192                  * sender queue.
 193                  */
 194                 else if (stop_tsk == mss->tsk)
 195                         break;
 196                 /*
 197                  * We are not in an EIDRM scenario here, therefore
 198                  * verify that we really need to wakeup the task.
 199                  * To maintain current semantics and wakeup order,
 200                  * move the sender to the tail on behalf of the
 201                  * blocked task.
 202                  */
 203                 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
 204                         if (!stop_tsk)
 205                                 stop_tsk = mss->tsk;
 206
 207                         list_move_tail(&mss->list, &msq->q_senders);
 208                         continue;
 209                 }
 210
 211                 wake_q_add(wake_q, mss->tsk);
 212         }
 213 }
 214
 215 static void expunge_all(struct msg_queue *msq, int res,
 216                         struct wake_q_head *wake_q)
 217 {
 218         struct msg_receiver *msr, *t;
 219
 220         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 221                 wake_q_add(wake_q, msr->r_tsk);
 222                 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
 223         }
 224 }
 225
 226 /*
 227  * freeque() wakes up waiters on the sender and receiver waiting queue,
 228  * removes the message queue from message queue ID IDR, and cleans up all the
 229  * messages associated with this queue.
 230  *
 231  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
 232  * before freeque() is called. msg_ids.rwsem remains locked on exit.
 233  */
 234 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 235 {
 236         struct msg_msg *msg, *t;
 237         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 238         WAKE_Q(wake_q);
 239
 240         expunge_all(msq, -EIDRM, &wake_q);
 241         ss_wakeup(msq, &wake_q, true);
 242         msg_rmid(ns, msq);
 243         ipc_unlock_object(&msq->q_perm);
 244         wake_up_q(&wake_q);
 245         rcu_read_unlock();
 246
 247         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 248                 atomic_dec(&ns->msg_hdrs);
 249                 free_msg(msg);
 250         }
 251         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 252         ipc_rcu_putref(msq, msg_rcu_free);
 253 }
 254
 255 /*
 256  * Called with msg_ids.rwsem and ipcp locked.
 257  */
 258 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 259 {
 260         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 261
 262         return security_msg_queue_associate(msq, msgflg);
 263 }
 264
 265 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 266 {
 267         struct ipc_namespace *ns;
 268         static const struct ipc_ops msg_ops = {
 269                 .getnew = newque,
 270                 .associate = msg_security,
 271         };
 272         struct ipc_params msg_params;
 273
 274         ns = current->nsproxy->ipc_ns;
 275
 276         msg_params.key = key;
 277         msg_params.flg = msgflg;
 278
 279         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 280 }
 281
 282 static inline unsigned long
 283 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 284 {
 285         switch (version) {
 286         case IPC_64:
 287                 return copy_to_user(buf, in, sizeof(*in));
 288         case IPC_OLD:
 289         {
 290                 struct msqid_ds out;
 291
 292                 memset(&out, 0, sizeof(out));
 293
 294                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 295
 296                 out.msg_stime           = in->msg_stime;
 297                 out.msg_rtime           = in->msg_rtime;
 298                 out.msg_ctime           = in->msg_ctime;
 299
 300                 if (in->msg_cbytes > USHRT_MAX)
 301                         out.msg_cbytes  = USHRT_MAX;
 302                 else
 303                         out.msg_cbytes  = in->msg_cbytes;
 304                 out.msg_lcbytes         = in->msg_cbytes;
 305
 306                 if (in->msg_qnum > USHRT_MAX)
 307                         out.msg_qnum    = USHRT_MAX;
 308                 else
 309                         out.msg_qnum    = in->msg_qnum;
 310
 311                 if (in->msg_qbytes > USHRT_MAX)
 312                         out.msg_qbytes  = USHRT_MAX;
 313                 else
 314                         out.msg_qbytes  = in->msg_qbytes;
 315                 out.msg_lqbytes         = in->msg_qbytes;
 316
 317                 out.msg_lspid           = in->msg_lspid;
 318                 out.msg_lrpid           = in->msg_lrpid;
 319
 320                 return copy_to_user(buf, &out, sizeof(out));
 321         }
 322         default:
 323                 return -EINVAL;
 324         }
 325 }
 326
 327 static inline unsigned long
 328 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 329 {
 330         switch (version) {
 331         case IPC_64:
 332                 if (copy_from_user(out, buf, sizeof(*out)))
 333                         return -EFAULT;
 334                 return 0;
 335         case IPC_OLD:
 336         {
 337                 struct msqid_ds tbuf_old;
 338
 339                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 340                         return -EFAULT;
 341
 342                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 343                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 344                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 345
 346                 if (tbuf_old.msg_qbytes == 0)
 347                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 348                 else
 349                         out->msg_qbytes = tbuf_old.msg_qbytes;
 350
 351                 return 0;
 352         }
 353         default:
 354                 return -EINVAL;
 355         }
 356 }
 357
 358 /*
 359  * This function handles some msgctl commands which require the rwsem
 360  * to be held in write mode.
 361  * NOTE: no locks must be held, the rwsem is taken inside this function.
 362  */
 363 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 364                        struct msqid_ds __user *buf, int version)
 365 {
 366         struct kern_ipc_perm *ipcp;
 367         struct msqid64_ds uninitialized_var(msqid64);
 368         struct msg_queue *msq;
 369         int err;
 370
 371         if (cmd == IPC_SET) {
 372                 if (copy_msqid_from_user(&msqid64, buf, version))
 373                         return -EFAULT;
 374         }
 375
 376         down_write(&msg_ids(ns).rwsem);
 377         rcu_read_lock();
 378
 379         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 380                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 381         if (IS_ERR(ipcp)) {
 382                 err = PTR_ERR(ipcp);
 383                 goto out_unlock1;
 384         }
 385
 386         msq = container_of(ipcp, struct msg_queue, q_perm);
 387
 388         err = security_msg_queue_msgctl(msq, cmd);
 389         if (err)
 390                 goto out_unlock1;
 391
 392         switch (cmd) {
 393         case IPC_RMID:
 394                 ipc_lock_object(&msq->q_perm);
 395                 /* freeque unlocks the ipc object and rcu */
 396                 freeque(ns, ipcp);
 397                 goto out_up;
 398         case IPC_SET:
 399         {
 400                 WAKE_Q(wake_q);
 401
 402                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 403                     !capable(CAP_SYS_RESOURCE)) {
 404                         err = -EPERM;
 405                         goto out_unlock1;
 406                 }
 407
 408                 ipc_lock_object(&msq->q_perm);
 409                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 410                 if (err)
 411                         goto out_unlock0;
 412
 413                 msq->q_qbytes = msqid64.msg_qbytes;
 414
 415                 msq->q_ctime = get_seconds();
 416                 /*
 417                  * Sleeping receivers might be excluded by
 418                  * stricter permissions.
 419                  */
 420                 expunge_all(msq, -EAGAIN, &wake_q);
 421                 /*
 422                  * Sleeping senders might be able to send
 423                  * due to a larger queue size.
 424                  */
 425                 ss_wakeup(msq, &wake_q, false);
 426                 ipc_unlock_object(&msq->q_perm);
 427                 wake_up_q(&wake_q);
 428
 429                 goto out_unlock1;
 430         }
 431         default:
 432                 err = -EINVAL;
 433                 goto out_unlock1;
 434         }
 435
 436 out_unlock0:
 437         ipc_unlock_object(&msq->q_perm);
 438 out_unlock1:
 439         rcu_read_unlock();
 440 out_up:
 441         up_write(&msg_ids(ns).rwsem);
 442         return err;
 443 }
 444
 445 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
 446                          int cmd, int version, void __user *buf)
 447 {
 448         int err;
 449         struct msg_queue *msq;
 450
 451         switch (cmd) {
 452         case IPC_INFO:
 453         case MSG_INFO:
 454         {
 455                 struct msginfo msginfo;
 456                 int max_id;
 457
 458                 if (!buf)
 459                         return -EFAULT;
 460
 461                 /*
 462                  * We must not return kernel stack data.
 463                  * due to padding, it's not enough
 464                  * to set all member fields.
 465                  */
 466                 err = security_msg_queue_msgctl(NULL, cmd);
 467                 if (err)
 468                         return err;
 469
 470                 memset(&msginfo, 0, sizeof(msginfo));
 471                 msginfo.msgmni = ns->msg_ctlmni;
 472                 msginfo.msgmax = ns->msg_ctlmax;
 473                 msginfo.msgmnb = ns->msg_ctlmnb;
 474                 msginfo.msgssz = MSGSSZ;
 475                 msginfo.msgseg = MSGSEG;
 476                 down_read(&msg_ids(ns).rwsem);
 477                 if (cmd == MSG_INFO) {
 478                         msginfo.msgpool = msg_ids(ns).in_use;
 479                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 480                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 481                 } else {
 482                         msginfo.msgmap = MSGMAP;
 483                         msginfo.msgpool = MSGPOOL;
 484                         msginfo.msgtql = MSGTQL;
 485                 }
 486                 max_id = ipc_get_maxid(&msg_ids(ns));
 487                 up_read(&msg_ids(ns).rwsem);
 488                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 489                         return -EFAULT;
 490                 return (max_id < 0) ? 0 : max_id;
 491         }
 492
 493         case MSG_STAT:
 494         case IPC_STAT:
 495         {
 496                 struct msqid64_ds tbuf;
 497                 int success_return;
 498
 499                 if (!buf)
 500                         return -EFAULT;
 501
 502                 memset(&tbuf, 0, sizeof(tbuf));
 503
 504                 rcu_read_lock();
 505                 if (cmd == MSG_STAT) {
 506                         msq = msq_obtain_object(ns, msqid);
 507                         if (IS_ERR(msq)) {
 508                                 err = PTR_ERR(msq);
 509                                 goto out_unlock;
 510                         }
 511                         success_return = msq->q_perm.id;
 512                 } else {
 513                         msq = msq_obtain_object_check(ns, msqid);
 514                         if (IS_ERR(msq)) {
 515                                 err = PTR_ERR(msq);
 516                                 goto out_unlock;
 517                         }
 518                         success_return = 0;
 519                 }
 520
 521                 err = -EACCES;
 522                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 523                         goto out_unlock;
 524
 525                 err = security_msg_queue_msgctl(msq, cmd);
 526                 if (err)
 527                         goto out_unlock;
 528
 529                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 530                 tbuf.msg_stime  = msq->q_stime;
 531                 tbuf.msg_rtime  = msq->q_rtime;
 532                 tbuf.msg_ctime  = msq->q_ctime;
 533                 tbuf.msg_cbytes = msq->q_cbytes;
 534                 tbuf.msg_qnum   = msq->q_qnum;
 535                 tbuf.msg_qbytes = msq->q_qbytes;
 536                 tbuf.msg_lspid  = msq->q_lspid;
 537                 tbuf.msg_lrpid  = msq->q_lrpid;
 538                 rcu_read_unlock();
 539
 540                 if (copy_msqid_to_user(buf, &tbuf, version))
 541                         return -EFAULT;
 542                 return success_return;
 543         }
 544
 545         default:
 546                 return -EINVAL;
 547         }
 548
 549         return err;
 550 out_unlock:
 551         rcu_read_unlock();
 552         return err;
 553 }
 554
 555 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 556 {
 557         int version;
 558         struct ipc_namespace *ns;
 559
 560         if (msqid < 0 || cmd < 0)
 561                 return -EINVAL;
 562
 563         version = ipc_parse_version(&cmd);
 564         ns = current->nsproxy->ipc_ns;
 565
 566         switch (cmd) {
 567         case IPC_INFO:
 568         case MSG_INFO:
 569         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 570         case IPC_STAT:
 571                 return msgctl_nolock(ns, msqid, cmd, version, buf);
 572         case IPC_SET:
 573         case IPC_RMID:
 574                 return msgctl_down(ns, msqid, cmd, buf, version);
 575         default:
 576                 return  -EINVAL;
 577         }
 578 }
 579
 580 static int testmsg(struct msg_msg *msg, long type, int mode)
 581 {
 582         switch (mode) {
 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                                  struct wake_q_head *wake_q)
 604 {
 605         struct msg_receiver *msr, *t;
 606
 607         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 608                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 609                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 610                                                msr->r_msgtype, msr->r_mode)) {
 611
 612                         list_del(&msr->r_list);
 613                         if (msr->r_maxsize < msg->m_ts) {
 614                                 wake_q_add(wake_q, msr->r_tsk);
 615                                 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
 616                         } else {
 617                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 618                                 msq->q_rtime = get_seconds();
 619
 620                                 wake_q_add(wake_q, msr->r_tsk);
 621                                 WRITE_ONCE(msr->r_msg, msg);
 622                                 return 1;
 623                         }
 624                 }
 625         }
 626
 627         return 0;
 628 }
 629
 630 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 631                 size_t msgsz, int msgflg)
 632 {
 633         struct msg_queue *msq;
 634         struct msg_msg *msg;
 635         int err;
 636         struct ipc_namespace *ns;
 637         WAKE_Q(wake_q);
 638
 639         ns = current->nsproxy->ipc_ns;
 640
 641         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 642                 return -EINVAL;
 643         if (mtype < 1)
 644                 return -EINVAL;
 645
 646         msg = load_msg(mtext, msgsz);
 647         if (IS_ERR(msg))
 648                 return PTR_ERR(msg);
 649
 650         msg->m_type = mtype;
 651         msg->m_ts = msgsz;
 652
 653         rcu_read_lock();
 654         msq = msq_obtain_object_check(ns, msqid);
 655         if (IS_ERR(msq)) {
 656                 err = PTR_ERR(msq);
 657                 goto out_unlock1;
 658         }
 659
 660         ipc_lock_object(&msq->q_perm);
 661
 662         for (;;) {
 663                 struct msg_sender s;
 664
 665                 err = -EACCES;
 666                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 667                         goto out_unlock0;
 668
 669                 /* raced with RMID? */
 670                 if (!ipc_valid_object(&msq->q_perm)) {
 671                         err = -EIDRM;
 672                         goto out_unlock0;
 673                 }
 674
 675                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 676                 if (err)
 677                         goto out_unlock0;
 678
 679                 if (msg_fits_inqueue(msq, msgsz))
 680                         break;
 681
 682                 /* queue full, wait: */
 683                 if (msgflg & IPC_NOWAIT) {
 684                         err = -EAGAIN;
 685                         goto out_unlock0;
 686                 }
 687
 688                 /* enqueue the sender and prepare to block */
 689                 ss_add(msq, &s, msgsz);
 690
 691                 if (!ipc_rcu_getref(msq)) {
 692                         err = -EIDRM;
 693                         goto out_unlock0;
 694                 }
 695
 696                 ipc_unlock_object(&msq->q_perm);
 697                 rcu_read_unlock();
 698                 schedule();
 699
 700                 rcu_read_lock();
 701                 ipc_lock_object(&msq->q_perm);
 702
 703                 ipc_rcu_putref(msq, msg_rcu_free);
 704                 /* raced with RMID? */
 705                 if (!ipc_valid_object(&msq->q_perm)) {
 706                         err = -EIDRM;
 707                         goto out_unlock0;
 708                 }
 709                 ss_del(&s);
 710
 711                 if (signal_pending(current)) {
 712                         err = -ERESTARTNOHAND;
 713                         goto out_unlock0;
 714                 }
 715
 716         }
 717
 718         msq->q_lspid = task_tgid_vnr(current);
 719         msq->q_stime = get_seconds();
 720
 721         if (!pipelined_send(msq, msg, &wake_q)) {
 722                 /* no one is waiting for this message, enqueue it */
 723                 list_add_tail(&msg->m_list, &msq->q_messages);
 724                 msq->q_cbytes += msgsz;
 725                 msq->q_qnum++;
 726                 atomic_add(msgsz, &ns->msg_bytes);
 727                 atomic_inc(&ns->msg_hdrs);
 728         }
 729
 730         err = 0;
 731         msg = NULL;
 732
 733 out_unlock0:
 734         ipc_unlock_object(&msq->q_perm);
 735         wake_up_q(&wake_q);
 736 out_unlock1:
 737         rcu_read_unlock();
 738         if (msg != NULL)
 739                 free_msg(msg);
 740         return err;
 741 }
 742
 743 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 744                 int, msgflg)
 745 {
 746         long mtype;
 747
 748         if (get_user(mtype, &msgp->mtype))
 749                 return -EFAULT;
 750         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 751 }
 752
 753 static inline int convert_mode(long *msgtyp, int msgflg)
 754 {
 755         if (msgflg & MSG_COPY)
 756                 return SEARCH_NUMBER;
 757         /*
 758          *  find message of correct type.
 759          *  msgtyp = 0 => get first.
 760          *  msgtyp > 0 => get first message of matching type.
 761          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 762          */
 763         if (*msgtyp == 0)
 764                 return SEARCH_ANY;
 765         if (*msgtyp < 0) {
 766                 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
 767                         *msgtyp = LONG_MAX;
 768                 else
 769                         *msgtyp = -*msgtyp;
 770                 return SEARCH_LESSEQUAL;
 771         }
 772         if (msgflg & MSG_EXCEPT)
 773                 return SEARCH_NOTEQUAL;
 774         return SEARCH_EQUAL;
 775 }
 776
 777 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 778 {
 779         struct msgbuf __user *msgp = dest;
 780         size_t msgsz;
 781
 782         if (put_user(msg->m_type, &msgp->mtype))
 783                 return -EFAULT;
 784
 785         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 786         if (store_msg(msgp->mtext, msg, msgsz))
 787                 return -EFAULT;
 788         return msgsz;
 789 }
 790
 791 #ifdef CONFIG_CHECKPOINT_RESTORE
 792 /*
 793  * This function creates new kernel message structure, large enough to store
 794  * bufsz message bytes.
 795  */
 796 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 797 {
 798         struct msg_msg *copy;
 799
 800         /*
 801          * Create dummy message to copy real message to.
 802          */
 803         copy = load_msg(buf, bufsz);
 804         if (!IS_ERR(copy))
 805                 copy->m_ts = bufsz;
 806         return copy;
 807 }
 808
 809 static inline void free_copy(struct msg_msg *copy)
 810 {
 811         if (copy)
 812                 free_msg(copy);
 813 }
 814 #else
 815 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 816 {
 817         return ERR_PTR(-ENOSYS);
 818 }
 819
 820 static inline void free_copy(struct msg_msg *copy)
 821 {
 822 }
 823 #endif
 824
 825 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 826 {
 827         struct msg_msg *msg, *found = NULL;
 828         long count = 0;
 829
 830         list_for_each_entry(msg, &msq->q_messages, m_list) {
 831                 if (testmsg(msg, *msgtyp, mode) &&
 832                     !security_msg_queue_msgrcv(msq, msg, current,
 833                                                *msgtyp, mode)) {
 834                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 835                                 *msgtyp = msg->m_type - 1;
 836                                 found = msg;
 837                         } else if (mode == SEARCH_NUMBER) {
 838                                 if (*msgtyp == count)
 839                                         return msg;
 840                         } else
 841                                 return msg;
 842                         count++;
 843                 }
 844         }
 845
 846         return found ?: ERR_PTR(-EAGAIN);
 847 }
 848
 849 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
 850                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 851 {
 852         int mode;
 853         struct msg_queue *msq;
 854         struct ipc_namespace *ns;
 855         struct msg_msg *msg, *copy = NULL;
 856         WAKE_Q(wake_q);
 857
 858         ns = current->nsproxy->ipc_ns;
 859
 860         if (msqid < 0 || (long) bufsz < 0)
 861                 return -EINVAL;
 862
 863         if (msgflg & MSG_COPY) {
 864                 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
 865                         return -EINVAL;
 866                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 867                 if (IS_ERR(copy))
 868                         return PTR_ERR(copy);
 869         }
 870         mode = convert_mode(&msgtyp, msgflg);
 871
 872         rcu_read_lock();
 873         msq = msq_obtain_object_check(ns, msqid);
 874         if (IS_ERR(msq)) {
 875                 rcu_read_unlock();
 876                 free_copy(copy);
 877                 return PTR_ERR(msq);
 878         }
 879
 880         for (;;) {
 881                 struct msg_receiver msr_d;
 882
 883                 msg = ERR_PTR(-EACCES);
 884                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 885                         goto out_unlock1;
 886
 887                 ipc_lock_object(&msq->q_perm);
 888
 889                 /* raced with RMID? */
 890                 if (!ipc_valid_object(&msq->q_perm)) {
 891                         msg = ERR_PTR(-EIDRM);
 892                         goto out_unlock0;
 893                 }
 894
 895                 msg = find_msg(msq, &msgtyp, mode);
 896                 if (!IS_ERR(msg)) {
 897                         /*
 898                          * Found a suitable message.
 899                          * Unlink it from the queue.
 900                          */
 901                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 902                                 msg = ERR_PTR(-E2BIG);
 903                                 goto out_unlock0;
 904                         }
 905                         /*
 906                          * If we are copying, then do not unlink message and do
 907                          * not update queue parameters.
 908                          */
 909                         if (msgflg & MSG_COPY) {
 910                                 msg = copy_msg(msg, copy);
 911                                 goto out_unlock0;
 912                         }
 913
 914                         list_del(&msg->m_list);
 915                         msq->q_qnum--;
 916                         msq->q_rtime = get_seconds();
 917                         msq->q_lrpid = task_tgid_vnr(current);
 918                         msq->q_cbytes -= msg->m_ts;
 919                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 920                         atomic_dec(&ns->msg_hdrs);
 921                         ss_wakeup(msq, &wake_q, false);
 922
 923                         goto out_unlock0;
 924                 }
 925
 926                 /* No message waiting. Wait for a message */
 927                 if (msgflg & IPC_NOWAIT) {
 928                         msg = ERR_PTR(-ENOMSG);
 929                         goto out_unlock0;
 930                 }
 931
 932                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 933                 msr_d.r_tsk = current;
 934                 msr_d.r_msgtype = msgtyp;
 935                 msr_d.r_mode = mode;
 936                 if (msgflg & MSG_NOERROR)
 937                         msr_d.r_maxsize = INT_MAX;
 938                 else
 939                         msr_d.r_maxsize = bufsz;
 940                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 941                 __set_current_state(TASK_INTERRUPTIBLE);
 942
 943                 ipc_unlock_object(&msq->q_perm);
 944                 rcu_read_unlock();
 945                 schedule();
 946
 947                 /*
 948                  * Lockless receive, part 1:
 949                  * We don't hold a reference to the queue and getting a
 950                  * reference would defeat the idea of a lockless operation,
 951                  * thus the code relies on rcu to guarantee the existence of
 952                  * msq:
 953                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 954                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 955                  */
 956                 rcu_read_lock();
 957
 958                 /*
 959                  * Lockless receive, part 2:
 960                  * The work in pipelined_send() and expunge_all():
 961                  * - Set pointer to message
 962                  * - Queue the receiver task for later wakeup
 963                  * - Wake up the process after the lock is dropped.
 964                  *
 965                  * Should the process wake up before this wakeup (due to a
 966                  * signal) it will either see the message and continue ...
 967                  */
 968                 msg = READ_ONCE(msr_d.r_msg);
 969                 if (msg != ERR_PTR(-EAGAIN))
 970                         goto out_unlock1;
 971
 972                  /*
 973                   * ... or see -EAGAIN, acquire the lock to check the message
 974                   * again.
 975                   */
 976                 ipc_lock_object(&msq->q_perm);
 977
 978                 msg = msr_d.r_msg;
 979                 if (msg != ERR_PTR(-EAGAIN))
 980                         goto out_unlock0;
 981
 982                 list_del(&msr_d.r_list);
 983                 if (signal_pending(current)) {
 984                         msg = ERR_PTR(-ERESTARTNOHAND);
 985                         goto out_unlock0;
 986                 }
 987
 988                 ipc_unlock_object(&msq->q_perm);
 989         }
 990
 991 out_unlock0:
 992         ipc_unlock_object(&msq->q_perm);
 993         wake_up_q(&wake_q);
 994 out_unlock1:
 995         rcu_read_unlock();
 996         if (IS_ERR(msg)) {
 997                 free_copy(copy);
 998                 return PTR_ERR(msg);
 999         }
1000
1001         bufsz = msg_handler(buf, msg, bufsz);
1002         free_msg(msg);
1003
1004         return bufsz;
1005 }
1006
1007 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1008                 long, msgtyp, int, msgflg)
1009 {
1010         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1011 }
1012
1013
1014 void msg_init_ns(struct ipc_namespace *ns)
1015 {
1016         ns->msg_ctlmax = MSGMAX;
1017         ns->msg_ctlmnb = MSGMNB;
1018         ns->msg_ctlmni = MSGMNI;
1019
1020         atomic_set(&ns->msg_bytes, 0);
1021         atomic_set(&ns->msg_hdrs, 0);
1022         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1023 }
1024
1025 #ifdef CONFIG_IPC_NS
1026 void msg_exit_ns(struct ipc_namespace *ns)
1027 {
1028         free_ipcs(ns, &msg_ids(ns), freeque);
1029         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1030 }
1031 #endif
1032
1033 #ifdef CONFIG_PROC_FS
1034 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1035 {
1036         struct user_namespace *user_ns = seq_user_ns(s);
1037         struct msg_queue *msq = it;
1038
1039         seq_printf(s,
1040                    "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1041                    msq->q_perm.key,
1042                    msq->q_perm.id,
1043                    msq->q_perm.mode,
1044                    msq->q_cbytes,
1045                    msq->q_qnum,
1046                    msq->q_lspid,
1047                    msq->q_lrpid,
1048                    from_kuid_munged(user_ns, msq->q_perm.uid),
1049                    from_kgid_munged(user_ns, msq->q_perm.gid),
1050                    from_kuid_munged(user_ns, msq->q_perm.cuid),
1051                    from_kgid_munged(user_ns, msq->q_perm.cgid),
1052                    msq->q_stime,
1053                    msq->q_rtime,
1054                    msq->q_ctime);
1055
1056         return 0;
1057 }
1058 #endif
1059
1060 void __init msg_init(void)
1061 {
1062         msg_init_ns(&init_ipc_ns);
1063
1064         ipc_init_proc_interface("sysvipc/msg",
1065                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1066                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
1067 }