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 /*
  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 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  74 static int newque(struct ipc_namespace *, struct ipc_params *);
  75 #ifdef CONFIG_PROC_FS
  76 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  77 #endif
  78
  79 /*
  80  * Scale msgmni with the available lowmem size: the memory dedicated to msg
  81  * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  82  * Also take into account the number of nsproxies created so far.
  83  * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  84  */
  85 void recompute_msgmni(struct ipc_namespace *ns)
  86 {
  87         struct sysinfo i;
  88         unsigned long allowed;
  89         int nb_ns;
  90
  91         si_meminfo(&i);
  92         allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  93                 / MSGMNB;
  94         nb_ns = atomic_read(&nr_ipc_ns);
  95         allowed /= nb_ns;
  96
  97         if (allowed < MSGMNI) {
  98                 ns->msg_ctlmni = MSGMNI;
  99                 return;
 100         }
 101
 102         if (allowed > IPCMNI / nb_ns) {
 103                 ns->msg_ctlmni = IPCMNI / nb_ns;
 104                 return;
 105         }
 106
 107         ns->msg_ctlmni = allowed;
 108 }
 109
 110 void msg_init_ns(struct ipc_namespace *ns)
 111 {
 112         ns->msg_ctlmax = MSGMAX;
 113         ns->msg_ctlmnb = MSGMNB;
 114
 115         recompute_msgmni(ns);
 116
 117         atomic_set(&ns->msg_bytes, 0);
 118         atomic_set(&ns->msg_hdrs, 0);
 119         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 120 }
 121
 122 #ifdef CONFIG_IPC_NS
 123 void msg_exit_ns(struct ipc_namespace *ns)
 124 {
 125         free_ipcs(ns, &msg_ids(ns), freeque);
 126         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
 127 }
 128 #endif
 129
 130 void __init msg_init(void)
 131 {
 132         msg_init_ns(&init_ipc_ns);
 133
 134         printk(KERN_INFO "msgmni has been set to %d\n",
 135                 init_ipc_ns.msg_ctlmni);
 136
 137         ipc_init_proc_interface("sysvipc/msg",
 138                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 139                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 140 }
 141
 142 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
 143 {
 144         struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
 145
 146         if (IS_ERR(ipcp))
 147                 return ERR_CAST(ipcp);
 148
 149         return container_of(ipcp, struct msg_queue, q_perm);
 150 }
 151
 152 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
 153                                                         int id)
 154 {
 155         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
 156
 157         if (IS_ERR(ipcp))
 158                 return ERR_CAST(ipcp);
 159
 160         return container_of(ipcp, struct msg_queue, q_perm);
 161 }
 162
 163 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 164 {
 165         ipc_rmid(&msg_ids(ns), &s->q_perm);
 166 }
 167
 168 static void msg_rcu_free(struct rcu_head *head)
 169 {
 170         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
 171         struct msg_queue *msq = ipc_rcu_to_struct(p);
 172
 173         security_msg_queue_free(msq);
 174         ipc_rcu_free(head);
 175 }
 176
 177 /**
 178  * newque - Create a new msg queue
 179  * @ns: namespace
 180  * @params: ptr to the structure that contains the key and msgflg
 181  *
 182  * Called with msg_ids.rwsem held (writer)
 183  */
 184 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 185 {
 186         struct msg_queue *msq;
 187         int id, retval;
 188         key_t key = params->key;
 189         int msgflg = params->flg;
 190
 191         msq = ipc_rcu_alloc(sizeof(*msq));
 192         if (!msq)
 193                 return -ENOMEM;
 194
 195         msq->q_perm.mode = msgflg & S_IRWXUGO;
 196         msq->q_perm.key = key;
 197
 198         msq->q_perm.security = NULL;
 199         retval = security_msg_queue_alloc(msq);
 200         if (retval) {
 201                 ipc_rcu_putref(msq, ipc_rcu_free);
 202                 return retval;
 203         }
 204
 205         /* ipc_addid() locks msq upon success. */
 206         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 207         if (id < 0) {
 208                 ipc_rcu_putref(msq, msg_rcu_free);
 209                 return id;
 210         }
 211
 212         msq->q_stime = msq->q_rtime = 0;
 213         msq->q_ctime = get_seconds();
 214         msq->q_cbytes = msq->q_qnum = 0;
 215         msq->q_qbytes = ns->msg_ctlmnb;
 216         msq->q_lspid = msq->q_lrpid = 0;
 217         INIT_LIST_HEAD(&msq->q_messages);
 218         INIT_LIST_HEAD(&msq->q_receivers);
 219         INIT_LIST_HEAD(&msq->q_senders);
 220
 221         ipc_unlock_object(&msq->q_perm);
 222         rcu_read_unlock();
 223
 224         return msq->q_perm.id;
 225 }
 226
 227 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 228 {
 229         mss->tsk = current;
 230         __set_current_state(TASK_INTERRUPTIBLE);
 231         list_add_tail(&mss->list, &msq->q_senders);
 232 }
 233
 234 static inline void ss_del(struct msg_sender *mss)
 235 {
 236         if (mss->list.next != NULL)
 237                 list_del(&mss->list);
 238 }
 239
 240 static void ss_wakeup(struct list_head *h, int kill)
 241 {
 242         struct msg_sender *mss, *t;
 243
 244         list_for_each_entry_safe(mss, t, h, list) {
 245                 if (kill)
 246                         mss->list.next = NULL;
 247                 wake_up_process(mss->tsk);
 248         }
 249 }
 250
 251 static void expunge_all(struct msg_queue *msq, int res)
 252 {
 253         struct msg_receiver *msr, *t;
 254
 255         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 256                 msr->r_msg = NULL; /* initialize expunge ordering */
 257                 wake_up_process(msr->r_tsk);
 258                 /*
 259                  * Ensure that the wakeup is visible before setting r_msg as
 260                  * the receiving end depends on it: either spinning on a nil,
 261                  * or dealing with -EAGAIN cases. See lockless receive part 1
 262                  * and 2 in do_msgrcv().
 263                  */
 264                 smp_mb();
 265                 msr->r_msg = ERR_PTR(res);
 266         }
 267 }
 268
 269 /*
 270  * freeque() wakes up waiters on the sender and receiver waiting queue,
 271  * removes the message queue from message queue ID IDR, and cleans up all the
 272  * messages associated with this queue.
 273  *
 274  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
 275  * before freeque() is called. msg_ids.rwsem remains locked on exit.
 276  */
 277 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 278 {
 279         struct msg_msg *msg, *t;
 280         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 281
 282         expunge_all(msq, -EIDRM);
 283         ss_wakeup(&msq->q_senders, 1);
 284         msg_rmid(ns, msq);
 285         ipc_unlock_object(&msq->q_perm);
 286         rcu_read_unlock();
 287
 288         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 289                 atomic_dec(&ns->msg_hdrs);
 290                 free_msg(msg);
 291         }
 292         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 293         ipc_rcu_putref(msq, msg_rcu_free);
 294 }
 295
 296 /*
 297  * Called with msg_ids.rwsem and ipcp locked.
 298  */
 299 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 300 {
 301         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 302
 303         return security_msg_queue_associate(msq, msgflg);
 304 }
 305
 306 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 307 {
 308         struct ipc_namespace *ns;
 309         static const struct ipc_ops msg_ops = {
 310                 .getnew = newque,
 311                 .associate = msg_security,
 312         };
 313         struct ipc_params msg_params;
 314
 315         ns = current->nsproxy->ipc_ns;
 316
 317         msg_params.key = key;
 318         msg_params.flg = msgflg;
 319
 320         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 321 }
 322
 323 static inline unsigned long
 324 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 325 {
 326         switch (version) {
 327         case IPC_64:
 328                 return copy_to_user(buf, in, sizeof(*in));
 329         case IPC_OLD:
 330         {
 331                 struct msqid_ds out;
 332
 333                 memset(&out, 0, sizeof(out));
 334
 335                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 336
 337                 out.msg_stime           = in->msg_stime;
 338                 out.msg_rtime           = in->msg_rtime;
 339                 out.msg_ctime           = in->msg_ctime;
 340
 341                 if (in->msg_cbytes > USHRT_MAX)
 342                         out.msg_cbytes  = USHRT_MAX;
 343                 else
 344                         out.msg_cbytes  = in->msg_cbytes;
 345                 out.msg_lcbytes         = in->msg_cbytes;
 346
 347                 if (in->msg_qnum > USHRT_MAX)
 348                         out.msg_qnum    = USHRT_MAX;
 349                 else
 350                         out.msg_qnum    = in->msg_qnum;
 351
 352                 if (in->msg_qbytes > USHRT_MAX)
 353                         out.msg_qbytes  = USHRT_MAX;
 354                 else
 355                         out.msg_qbytes  = in->msg_qbytes;
 356                 out.msg_lqbytes         = in->msg_qbytes;
 357
 358                 out.msg_lspid           = in->msg_lspid;
 359                 out.msg_lrpid           = in->msg_lrpid;
 360
 361                 return copy_to_user(buf, &out, sizeof(out));
 362         }
 363         default:
 364                 return -EINVAL;
 365         }
 366 }
 367
 368 static inline unsigned long
 369 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 370 {
 371         switch (version) {
 372         case IPC_64:
 373                 if (copy_from_user(out, buf, sizeof(*out)))
 374                         return -EFAULT;
 375                 return 0;
 376         case IPC_OLD:
 377         {
 378                 struct msqid_ds tbuf_old;
 379
 380                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 381                         return -EFAULT;
 382
 383                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 384                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 385                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 386
 387                 if (tbuf_old.msg_qbytes == 0)
 388                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 389                 else
 390                         out->msg_qbytes = tbuf_old.msg_qbytes;
 391
 392                 return 0;
 393         }
 394         default:
 395                 return -EINVAL;
 396         }
 397 }
 398
 399 /*
 400  * This function handles some msgctl commands which require the rwsem
 401  * to be held in write mode.
 402  * NOTE: no locks must be held, the rwsem is taken inside this function.
 403  */
 404 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 405                        struct msqid_ds __user *buf, int version)
 406 {
 407         struct kern_ipc_perm *ipcp;
 408         struct msqid64_ds uninitialized_var(msqid64);
 409         struct msg_queue *msq;
 410         int err;
 411
 412         if (cmd == IPC_SET) {
 413                 if (copy_msqid_from_user(&msqid64, buf, version))
 414                         return -EFAULT;
 415         }
 416
 417         down_write(&msg_ids(ns).rwsem);
 418         rcu_read_lock();
 419
 420         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 421                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 422         if (IS_ERR(ipcp)) {
 423                 err = PTR_ERR(ipcp);
 424                 goto out_unlock1;
 425         }
 426
 427         msq = container_of(ipcp, struct msg_queue, q_perm);
 428
 429         err = security_msg_queue_msgctl(msq, cmd);
 430         if (err)
 431                 goto out_unlock1;
 432
 433         switch (cmd) {
 434         case IPC_RMID:
 435                 ipc_lock_object(&msq->q_perm);
 436                 /* freeque unlocks the ipc object and rcu */
 437                 freeque(ns, ipcp);
 438                 goto out_up;
 439         case IPC_SET:
 440                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 441                     !capable(CAP_SYS_RESOURCE)) {
 442                         err = -EPERM;
 443                         goto out_unlock1;
 444                 }
 445
 446                 ipc_lock_object(&msq->q_perm);
 447                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 448                 if (err)
 449                         goto out_unlock0;
 450
 451                 msq->q_qbytes = msqid64.msg_qbytes;
 452
 453                 msq->q_ctime = get_seconds();
 454                 /* sleeping receivers might be excluded by
 455                  * stricter permissions.
 456                  */
 457                 expunge_all(msq, -EAGAIN);
 458                 /* sleeping senders might be able to send
 459                  * due to a larger queue size.
 460                  */
 461                 ss_wakeup(&msq->q_senders, 0);
 462                 break;
 463         default:
 464                 err = -EINVAL;
 465                 goto out_unlock1;
 466         }
 467
 468 out_unlock0:
 469         ipc_unlock_object(&msq->q_perm);
 470 out_unlock1:
 471         rcu_read_unlock();
 472 out_up:
 473         up_write(&msg_ids(ns).rwsem);
 474         return err;
 475 }
 476
 477 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
 478                          int cmd, int version, void __user *buf)
 479 {
 480         int err;
 481         struct msg_queue *msq;
 482
 483         switch (cmd) {
 484         case IPC_INFO:
 485         case MSG_INFO:
 486         {
 487                 struct msginfo msginfo;
 488                 int max_id;
 489
 490                 if (!buf)
 491                         return -EFAULT;
 492
 493                 /*
 494                  * We must not return kernel stack data.
 495                  * due to padding, it's not enough
 496                  * to set all member fields.
 497                  */
 498                 err = security_msg_queue_msgctl(NULL, cmd);
 499                 if (err)
 500                         return err;
 501
 502                 memset(&msginfo, 0, sizeof(msginfo));
 503                 msginfo.msgmni = ns->msg_ctlmni;
 504                 msginfo.msgmax = ns->msg_ctlmax;
 505                 msginfo.msgmnb = ns->msg_ctlmnb;
 506                 msginfo.msgssz = MSGSSZ;
 507                 msginfo.msgseg = MSGSEG;
 508                 down_read(&msg_ids(ns).rwsem);
 509                 if (cmd == MSG_INFO) {
 510                         msginfo.msgpool = msg_ids(ns).in_use;
 511                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 512                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 513                 } else {
 514                         msginfo.msgmap = MSGMAP;
 515                         msginfo.msgpool = MSGPOOL;
 516                         msginfo.msgtql = MSGTQL;
 517                 }
 518                 max_id = ipc_get_maxid(&msg_ids(ns));
 519                 up_read(&msg_ids(ns).rwsem);
 520                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 521                         return -EFAULT;
 522                 return (max_id < 0) ? 0 : max_id;
 523         }
 524
 525         case MSG_STAT:
 526         case IPC_STAT:
 527         {
 528                 struct msqid64_ds tbuf;
 529                 int success_return;
 530
 531                 if (!buf)
 532                         return -EFAULT;
 533
 534                 memset(&tbuf, 0, sizeof(tbuf));
 535
 536                 rcu_read_lock();
 537                 if (cmd == MSG_STAT) {
 538                         msq = msq_obtain_object(ns, msqid);
 539                         if (IS_ERR(msq)) {
 540                                 err = PTR_ERR(msq);
 541                                 goto out_unlock;
 542                         }
 543                         success_return = msq->q_perm.id;
 544                 } else {
 545                         msq = msq_obtain_object_check(ns, msqid);
 546                         if (IS_ERR(msq)) {
 547                                 err = PTR_ERR(msq);
 548                                 goto out_unlock;
 549                         }
 550                         success_return = 0;
 551                 }
 552
 553                 err = -EACCES;
 554                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 555                         goto out_unlock;
 556
 557                 err = security_msg_queue_msgctl(msq, cmd);
 558                 if (err)
 559                         goto out_unlock;
 560
 561                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 562                 tbuf.msg_stime  = msq->q_stime;
 563                 tbuf.msg_rtime  = msq->q_rtime;
 564                 tbuf.msg_ctime  = msq->q_ctime;
 565                 tbuf.msg_cbytes = msq->q_cbytes;
 566                 tbuf.msg_qnum   = msq->q_qnum;
 567                 tbuf.msg_qbytes = msq->q_qbytes;
 568                 tbuf.msg_lspid  = msq->q_lspid;
 569                 tbuf.msg_lrpid  = msq->q_lrpid;
 570                 rcu_read_unlock();
 571
 572                 if (copy_msqid_to_user(buf, &tbuf, version))
 573                         return -EFAULT;
 574                 return success_return;
 575         }
 576
 577         default:
 578                 return -EINVAL;
 579         }
 580
 581         return err;
 582 out_unlock:
 583         rcu_read_unlock();
 584         return err;
 585 }
 586
 587 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 588 {
 589         int version;
 590         struct ipc_namespace *ns;
 591
 592         if (msqid < 0 || cmd < 0)
 593                 return -EINVAL;
 594
 595         version = ipc_parse_version(&cmd);
 596         ns = current->nsproxy->ipc_ns;
 597
 598         switch (cmd) {
 599         case IPC_INFO:
 600         case MSG_INFO:
 601         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 602         case IPC_STAT:
 603                 return msgctl_nolock(ns, msqid, cmd, version, buf);
 604         case IPC_SET:
 605         case IPC_RMID:
 606                 return msgctl_down(ns, msqid, cmd, buf, version);
 607         default:
 608                 return  -EINVAL;
 609         }
 610 }
 611
 612 static int testmsg(struct msg_msg *msg, long type, int mode)
 613 {
 614         switch (mode) {
 615         case SEARCH_ANY:
 616         case SEARCH_NUMBER:
 617                 return 1;
 618         case SEARCH_LESSEQUAL:
 619                 if (msg->m_type <= type)
 620                         return 1;
 621                 break;
 622         case SEARCH_EQUAL:
 623                 if (msg->m_type == type)
 624                         return 1;
 625                 break;
 626         case SEARCH_NOTEQUAL:
 627                 if (msg->m_type != type)
 628                         return 1;
 629                 break;
 630         }
 631         return 0;
 632 }
 633
 634 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 635 {
 636         struct msg_receiver *msr, *t;
 637
 638         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 639                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 640                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 641                                                msr->r_msgtype, msr->r_mode)) {
 642
 643                         list_del(&msr->r_list);
 644                         if (msr->r_maxsize < msg->m_ts) {
 645                                 /* initialize pipelined send ordering */
 646                                 msr->r_msg = NULL;
 647                                 wake_up_process(msr->r_tsk);
 648                                 smp_mb(); /* see barrier comment below */
 649                                 msr->r_msg = ERR_PTR(-E2BIG);
 650                         } else {
 651                                 msr->r_msg = NULL;
 652                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 653                                 msq->q_rtime = get_seconds();
 654                                 wake_up_process(msr->r_tsk);
 655                                 /*
 656                                  * Ensure that the wakeup is visible before
 657                                  * setting r_msg, as the receiving end depends
 658                                  * on it. See lockless receive part 1 and 2 in
 659                                  * do_msgrcv().
 660                                  */
 661                                 smp_mb();
 662                                 msr->r_msg = msg;
 663
 664                                 return 1;
 665                         }
 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         rcu_read_lock();
 695         msq = msq_obtain_object_check(ns, msqid);
 696         if (IS_ERR(msq)) {
 697                 err = PTR_ERR(msq);
 698                 goto out_unlock1;
 699         }
 700
 701         ipc_lock_object(&msq->q_perm);
 702
 703         for (;;) {
 704                 struct msg_sender s;
 705
 706                 err = -EACCES;
 707                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 708                         goto out_unlock0;
 709
 710                 /* raced with RMID? */
 711                 if (!ipc_valid_object(&msq->q_perm)) {
 712                         err = -EIDRM;
 713                         goto out_unlock0;
 714                 }
 715
 716                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 717                 if (err)
 718                         goto out_unlock0;
 719
 720                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 721                                 1 + msq->q_qnum <= msq->q_qbytes) {
 722                         break;
 723                 }
 724
 725                 /* queue full, wait: */
 726                 if (msgflg & IPC_NOWAIT) {
 727                         err = -EAGAIN;
 728                         goto out_unlock0;
 729                 }
 730
 731                 /* enqueue the sender and prepare to block */
 732                 ss_add(msq, &s);
 733
 734                 if (!ipc_rcu_getref(msq)) {
 735                         err = -EIDRM;
 736                         goto out_unlock0;
 737                 }
 738
 739                 ipc_unlock_object(&msq->q_perm);
 740                 rcu_read_unlock();
 741                 schedule();
 742
 743                 rcu_read_lock();
 744                 ipc_lock_object(&msq->q_perm);
 745
 746                 ipc_rcu_putref(msq, ipc_rcu_free);
 747                 /* raced with RMID? */
 748                 if (!ipc_valid_object(&msq->q_perm)) {
 749                         err = -EIDRM;
 750                         goto out_unlock0;
 751                 }
 752
 753                 ss_del(&s);
 754
 755                 if (signal_pending(current)) {
 756                         err = -ERESTARTNOHAND;
 757                         goto out_unlock0;
 758                 }
 759
 760         }
 761         msq->q_lspid = task_tgid_vnr(current);
 762         msq->q_stime = get_seconds();
 763
 764         if (!pipelined_send(msq, msg)) {
 765                 /* no one is waiting for this message, enqueue it */
 766                 list_add_tail(&msg->m_list, &msq->q_messages);
 767                 msq->q_cbytes += msgsz;
 768                 msq->q_qnum++;
 769                 atomic_add(msgsz, &ns->msg_bytes);
 770                 atomic_inc(&ns->msg_hdrs);
 771         }
 772
 773         err = 0;
 774         msg = NULL;
 775
 776 out_unlock0:
 777         ipc_unlock_object(&msq->q_perm);
 778 out_unlock1:
 779         rcu_read_unlock();
 780         if (msg != NULL)
 781                 free_msg(msg);
 782         return err;
 783 }
 784
 785 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 786                 int, msgflg)
 787 {
 788         long mtype;
 789
 790         if (get_user(mtype, &msgp->mtype))
 791                 return -EFAULT;
 792         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 793 }
 794
 795 static inline int convert_mode(long *msgtyp, int msgflg)
 796 {
 797         if (msgflg & MSG_COPY)
 798                 return SEARCH_NUMBER;
 799         /*
 800          *  find message of correct type.
 801          *  msgtyp = 0 => get first.
 802          *  msgtyp > 0 => get first message of matching type.
 803          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 804          */
 805         if (*msgtyp == 0)
 806                 return SEARCH_ANY;
 807         if (*msgtyp < 0) {
 808                 *msgtyp = -*msgtyp;
 809                 return SEARCH_LESSEQUAL;
 810         }
 811         if (msgflg & MSG_EXCEPT)
 812                 return SEARCH_NOTEQUAL;
 813         return SEARCH_EQUAL;
 814 }
 815
 816 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 817 {
 818         struct msgbuf __user *msgp = dest;
 819         size_t msgsz;
 820
 821         if (put_user(msg->m_type, &msgp->mtype))
 822                 return -EFAULT;
 823
 824         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 825         if (store_msg(msgp->mtext, msg, msgsz))
 826                 return -EFAULT;
 827         return msgsz;
 828 }
 829
 830 #ifdef CONFIG_CHECKPOINT_RESTORE
 831 /*
 832  * This function creates new kernel message structure, large enough to store
 833  * bufsz message bytes.
 834  */
 835 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 836 {
 837         struct msg_msg *copy;
 838
 839         /*
 840          * Create dummy message to copy real message to.
 841          */
 842         copy = load_msg(buf, bufsz);
 843         if (!IS_ERR(copy))
 844                 copy->m_ts = bufsz;
 845         return copy;
 846 }
 847
 848 static inline void free_copy(struct msg_msg *copy)
 849 {
 850         if (copy)
 851                 free_msg(copy);
 852 }
 853 #else
 854 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 855 {
 856         return ERR_PTR(-ENOSYS);
 857 }
 858
 859 static inline void free_copy(struct msg_msg *copy)
 860 {
 861 }
 862 #endif
 863
 864 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 865 {
 866         struct msg_msg *msg, *found = NULL;
 867         long count = 0;
 868
 869         list_for_each_entry(msg, &msq->q_messages, m_list) {
 870                 if (testmsg(msg, *msgtyp, mode) &&
 871                     !security_msg_queue_msgrcv(msq, msg, current,
 872                                                *msgtyp, mode)) {
 873                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 874                                 *msgtyp = msg->m_type - 1;
 875                                 found = msg;
 876                         } else if (mode == SEARCH_NUMBER) {
 877                                 if (*msgtyp == count)
 878                                         return msg;
 879                         } else
 880                                 return msg;
 881                         count++;
 882                 }
 883         }
 884
 885         return found ?: ERR_PTR(-EAGAIN);
 886 }
 887
 888 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
 889                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 890 {
 891         int mode;
 892         struct msg_queue *msq;
 893         struct ipc_namespace *ns;
 894         struct msg_msg *msg, *copy = NULL;
 895
 896         ns = current->nsproxy->ipc_ns;
 897
 898         if (msqid < 0 || (long) bufsz < 0)
 899                 return -EINVAL;
 900
 901         if (msgflg & MSG_COPY) {
 902                 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
 903                         return -EINVAL;
 904                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 905                 if (IS_ERR(copy))
 906                         return PTR_ERR(copy);
 907         }
 908         mode = convert_mode(&msgtyp, msgflg);
 909
 910         rcu_read_lock();
 911         msq = msq_obtain_object_check(ns, msqid);
 912         if (IS_ERR(msq)) {
 913                 rcu_read_unlock();
 914                 free_copy(copy);
 915                 return PTR_ERR(msq);
 916         }
 917
 918         for (;;) {
 919                 struct msg_receiver msr_d;
 920
 921                 msg = ERR_PTR(-EACCES);
 922                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 923                         goto out_unlock1;
 924
 925                 ipc_lock_object(&msq->q_perm);
 926
 927                 /* raced with RMID? */
 928                 if (!ipc_valid_object(&msq->q_perm)) {
 929                         msg = ERR_PTR(-EIDRM);
 930                         goto out_unlock0;
 931                 }
 932
 933                 msg = find_msg(msq, &msgtyp, mode);
 934                 if (!IS_ERR(msg)) {
 935                         /*
 936                          * Found a suitable message.
 937                          * Unlink it from the queue.
 938                          */
 939                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 940                                 msg = ERR_PTR(-E2BIG);
 941                                 goto out_unlock0;
 942                         }
 943                         /*
 944                          * If we are copying, then do not unlink message and do
 945                          * not update queue parameters.
 946                          */
 947                         if (msgflg & MSG_COPY) {
 948                                 msg = copy_msg(msg, copy);
 949                                 goto out_unlock0;
 950                         }
 951
 952                         list_del(&msg->m_list);
 953                         msq->q_qnum--;
 954                         msq->q_rtime = get_seconds();
 955                         msq->q_lrpid = task_tgid_vnr(current);
 956                         msq->q_cbytes -= msg->m_ts;
 957                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 958                         atomic_dec(&ns->msg_hdrs);
 959                         ss_wakeup(&msq->q_senders, 0);
 960
 961                         goto out_unlock0;
 962                 }
 963
 964                 /* No message waiting. Wait for a message */
 965                 if (msgflg & IPC_NOWAIT) {
 966                         msg = ERR_PTR(-ENOMSG);
 967                         goto out_unlock0;
 968                 }
 969
 970                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 971                 msr_d.r_tsk = current;
 972                 msr_d.r_msgtype = msgtyp;
 973                 msr_d.r_mode = mode;
 974                 if (msgflg & MSG_NOERROR)
 975                         msr_d.r_maxsize = INT_MAX;
 976                 else
 977                         msr_d.r_maxsize = bufsz;
 978                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 979                 __set_current_state(TASK_INTERRUPTIBLE);
 980
 981                 ipc_unlock_object(&msq->q_perm);
 982                 rcu_read_unlock();
 983                 schedule();
 984
 985                 /* Lockless receive, part 1:
 986                  * Disable preemption.  We don't hold a reference to the queue
 987                  * and getting a reference would defeat the idea of a lockless
 988                  * operation, thus the code relies on rcu to guarantee the
 989                  * existence of msq:
 990                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 991                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 992                  * rcu_read_lock() prevents preemption between reading r_msg
 993                  * and acquiring the q_perm.lock in ipc_lock_object().
 994                  */
 995                 rcu_read_lock();
 996
 997                 /* Lockless receive, part 2:
 998                  * Wait until pipelined_send or expunge_all are outside of
 999                  * wake_up_process(). There is a race with exit(), see
1000                  * ipc/mqueue.c for the details.
1001                  */
1002                 msg = (struct msg_msg *)msr_d.r_msg;
1003                 while (msg == NULL) {
1004                         cpu_relax();
1005                         msg = (struct msg_msg *)msr_d.r_msg;
1006                 }
1007
1008                 /* Lockless receive, part 3:
1009                  * If there is a message or an error then accept it without
1010                  * locking.
1011                  */
1012                 if (msg != ERR_PTR(-EAGAIN))
1013                         goto out_unlock1;
1014
1015                 /* Lockless receive, part 3:
1016                  * Acquire the queue spinlock.
1017                  */
1018                 ipc_lock_object(&msq->q_perm);
1019
1020                 /* Lockless receive, part 4:
1021                  * Repeat test after acquiring the spinlock.
1022                  */
1023                 msg = (struct msg_msg *)msr_d.r_msg;
1024                 if (msg != ERR_PTR(-EAGAIN))
1025                         goto out_unlock0;
1026
1027                 list_del(&msr_d.r_list);
1028                 if (signal_pending(current)) {
1029                         msg = ERR_PTR(-ERESTARTNOHAND);
1030                         goto out_unlock0;
1031                 }
1032
1033                 ipc_unlock_object(&msq->q_perm);
1034         }
1035
1036 out_unlock0:
1037         ipc_unlock_object(&msq->q_perm);
1038 out_unlock1:
1039         rcu_read_unlock();
1040         if (IS_ERR(msg)) {
1041                 free_copy(copy);
1042                 return PTR_ERR(msg);
1043         }
1044
1045         bufsz = msg_handler(buf, msg, bufsz);
1046         free_msg(msg);
1047
1048         return bufsz;
1049 }
1050
1051 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1052                 long, msgtyp, int, msgflg)
1053 {
1054         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1055 }
1056
1057 #ifdef CONFIG_PROC_FS
1058 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1059 {
1060         struct user_namespace *user_ns = seq_user_ns(s);
1061         struct msg_queue *msq = it;
1062
1063         return seq_printf(s,
1064                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1065                         msq->q_perm.key,
1066                         msq->q_perm.id,
1067                         msq->q_perm.mode,
1068                         msq->q_cbytes,
1069                         msq->q_qnum,
1070                         msq->q_lspid,
1071                         msq->q_lrpid,
1072                         from_kuid_munged(user_ns, msq->q_perm.uid),
1073                         from_kgid_munged(user_ns, msq->q_perm.gid),
1074                         from_kuid_munged(user_ns, msq->q_perm.cuid),
1075                         from_kgid_munged(user_ns, msq->q_perm.cgid),
1076                         msq->q_stime,
1077                         msq->q_rtime,
1078                         msq->q_ctime);
1079 }
1080 #endif