ipc/msg.c

   1 /*
   2  * linux/ipc/msg.c
   3  * Copyright (C) 1992 Krishna Balasubramanian
   4  *
   5  * Removed all the remaining kerneld mess
   6  * Catch the -EFAULT stuff properly
   7  * Use GFP_KERNEL for messages as in 1.2
   8  * Fixed up the unchecked user space derefs
   9  * Copyright (C) 1998 Alan Cox & Andi Kleen
  10  *
  11  * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  12  *
  13  * mostly rewritten, threaded and wake-one semantics added
  14  * MSGMAX limit removed, sysctl's added
  15  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  */
  24
  25 #include <linux/capability.h>
  26 #include <linux/msg.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/init.h>
  29 #include <linux/mm.h>
  30 #include <linux/proc_fs.h>
  31 #include <linux/list.h>
  32 #include <linux/security.h>
  33 #include <linux/sched.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/audit.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/rwsem.h>
  38 #include <linux/nsproxy.h>
  39 #include <linux/ipc_namespace.h>
  40
  41 #include <asm/current.h>
  42 #include <asm/uaccess.h>
  43 #include "util.h"
  44
  45 /*
  46  * one msg_receiver structure for each sleeping receiver:
  47  */
  48 struct msg_receiver {
  49         struct list_head        r_list;
  50         struct task_struct      *r_tsk;
  51
  52         int                     r_mode;
  53         long                    r_msgtype;
  54         long                    r_maxsize;
  55
  56         struct msg_msg          *volatile r_msg;
  57 };
  58
  59 /* one msg_sender for each sleeping sender */
  60 struct msg_sender {
  61         struct list_head        list;
  62         struct task_struct      *tsk;
  63 };
  64
  65 #define SEARCH_ANY              1
  66 #define SEARCH_EQUAL            2
  67 #define SEARCH_NOTEQUAL         3
  68 #define SEARCH_LESSEQUAL        4
  69 #define SEARCH_NUMBER           5
  70
  71 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  72
  73 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  74
  75 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  76 static int newque(struct ipc_namespace *, struct ipc_params *);
  77 #ifdef CONFIG_PROC_FS
  78 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  79 #endif
  80
  81 /*
  82  * Scale msgmni with the available lowmem size: the memory dedicated to msg
  83  * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  84  * Also take into account the number of nsproxies created so far.
  85  * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  86  */
  87 void recompute_msgmni(struct ipc_namespace *ns)
  88 {
  89         struct sysinfo i;
  90         unsigned long allowed;
  91         int nb_ns;
  92
  93         si_meminfo(&i);
  94         allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  95                 / MSGMNB;
  96         nb_ns = atomic_read(&nr_ipc_ns);
  97         allowed /= nb_ns;
  98
  99         if (allowed < MSGMNI) {
 100                 ns->msg_ctlmni = MSGMNI;
 101                 return;
 102         }
 103
 104         if (allowed > IPCMNI / nb_ns) {
 105                 ns->msg_ctlmni = IPCMNI / nb_ns;
 106                 return;
 107         }
 108
 109         ns->msg_ctlmni = allowed;
 110 }
 111
 112 void msg_init_ns(struct ipc_namespace *ns)
 113 {
 114         ns->msg_ctlmax = MSGMAX;
 115         ns->msg_ctlmnb = MSGMNB;
 116
 117         recompute_msgmni(ns);
 118
 119         atomic_set(&ns->msg_bytes, 0);
 120         atomic_set(&ns->msg_hdrs, 0);
 121         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 122 }
 123
 124 #ifdef CONFIG_IPC_NS
 125 void msg_exit_ns(struct ipc_namespace *ns)
 126 {
 127         free_ipcs(ns, &msg_ids(ns), freeque);
 128         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
 129 }
 130 #endif
 131
 132 void __init msg_init(void)
 133 {
 134         msg_init_ns(&init_ipc_ns);
 135
 136         printk(KERN_INFO "msgmni has been set to %d\n",
 137                 init_ipc_ns.msg_ctlmni);
 138
 139         ipc_init_proc_interface("sysvipc/msg",
 140                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 141                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 142 }
 143
 144 /*
 145  * msg_lock_(check_) routines are called in the paths where the rw_mutex
 146  * is not held.
 147  */
 148 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
 149 {
 150         struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
 151
 152         if (IS_ERR(ipcp))
 153                 return (struct msg_queue *)ipcp;
 154
 155         return container_of(ipcp, struct msg_queue, q_perm);
 156 }
 157
 158 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
 159                                                 int id)
 160 {
 161         struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
 162
 163         if (IS_ERR(ipcp))
 164                 return (struct msg_queue *)ipcp;
 165
 166         return container_of(ipcp, struct msg_queue, q_perm);
 167 }
 168
 169 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
 170 {
 171         struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
 172
 173         if (IS_ERR(ipcp))
 174                 return ERR_CAST(ipcp);
 175
 176         return container_of(ipcp, struct msg_queue, q_perm);
 177 }
 178
 179 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
 180                                                         int id)
 181 {
 182         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
 183
 184         if (IS_ERR(ipcp))
 185                 return ERR_CAST(ipcp);
 186
 187         return container_of(ipcp, struct msg_queue, q_perm);
 188 }
 189
 190 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 191 {
 192         ipc_rmid(&msg_ids(ns), &s->q_perm);
 193 }
 194
 195 /**
 196  * newque - Create a new msg queue
 197  * @ns: namespace
 198  * @params: ptr to the structure that contains the key and msgflg
 199  *
 200  * Called with msg_ids.rw_mutex held (writer)
 201  */
 202 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 203 {
 204         struct msg_queue *msq;
 205         int id, retval;
 206         key_t key = params->key;
 207         int msgflg = params->flg;
 208
 209         msq = ipc_rcu_alloc(sizeof(*msq));
 210         if (!msq)
 211                 return -ENOMEM;
 212
 213         msq->q_perm.mode = msgflg & S_IRWXUGO;
 214         msq->q_perm.key = key;
 215
 216         msq->q_perm.security = NULL;
 217         retval = security_msg_queue_alloc(msq);
 218         if (retval) {
 219                 ipc_rcu_putref(msq);
 220                 return retval;
 221         }
 222
 223         /* ipc_addid() locks msq upon success. */
 224         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 225         if (id < 0) {
 226                 security_msg_queue_free(msq);
 227                 ipc_rcu_putref(msq);
 228                 return id;
 229         }
 230
 231         msq->q_stime = msq->q_rtime = 0;
 232         msq->q_ctime = get_seconds();
 233         msq->q_cbytes = msq->q_qnum = 0;
 234         msq->q_qbytes = ns->msg_ctlmnb;
 235         msq->q_lspid = msq->q_lrpid = 0;
 236         INIT_LIST_HEAD(&msq->q_messages);
 237         INIT_LIST_HEAD(&msq->q_receivers);
 238         INIT_LIST_HEAD(&msq->q_senders);
 239
 240         ipc_unlock_object(&msq->q_perm);
 241         rcu_read_unlock();
 242
 243         return msq->q_perm.id;
 244 }
 245
 246 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 247 {
 248         mss->tsk = current;
 249         current->state = TASK_INTERRUPTIBLE;
 250         list_add_tail(&mss->list, &msq->q_senders);
 251 }
 252
 253 static inline void ss_del(struct msg_sender *mss)
 254 {
 255         if (mss->list.next != NULL)
 256                 list_del(&mss->list);
 257 }
 258
 259 static void ss_wakeup(struct list_head *h, int kill)
 260 {
 261         struct msg_sender *mss, *t;
 262
 263         list_for_each_entry_safe(mss, t, h, list) {
 264                 if (kill)
 265                         mss->list.next = NULL;
 266                 wake_up_process(mss->tsk);
 267         }
 268 }
 269
 270 static void expunge_all(struct msg_queue *msq, int res)
 271 {
 272         struct msg_receiver *msr, *t;
 273
 274         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 275                 msr->r_msg = NULL;
 276                 wake_up_process(msr->r_tsk);
 277                 smp_mb();
 278                 msr->r_msg = ERR_PTR(res);
 279         }
 280 }
 281
 282 /*
 283  * freeque() wakes up waiters on the sender and receiver waiting queue,
 284  * removes the message queue from message queue ID IDR, and cleans up all the
 285  * messages associated with this queue.
 286  *
 287  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 288  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 289  */
 290 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 291 {
 292         struct msg_msg *msg, *t;
 293         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 294
 295         expunge_all(msq, -EIDRM);
 296         ss_wakeup(&msq->q_senders, 1);
 297         msg_rmid(ns, msq);
 298         msg_unlock(msq);
 299
 300         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 301                 atomic_dec(&ns->msg_hdrs);
 302                 free_msg(msg);
 303         }
 304         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 305         security_msg_queue_free(msq);
 306         ipc_rcu_putref(msq);
 307 }
 308
 309 /*
 310  * Called with msg_ids.rw_mutex and ipcp locked.
 311  */
 312 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 313 {
 314         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 315
 316         return security_msg_queue_associate(msq, msgflg);
 317 }
 318
 319 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 320 {
 321         struct ipc_namespace *ns;
 322         struct ipc_ops msg_ops;
 323         struct ipc_params msg_params;
 324
 325         ns = current->nsproxy->ipc_ns;
 326
 327         msg_ops.getnew = newque;
 328         msg_ops.associate = msg_security;
 329         msg_ops.more_checks = NULL;
 330
 331         msg_params.key = key;
 332         msg_params.flg = msgflg;
 333
 334         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 335 }
 336
 337 static inline unsigned long
 338 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 339 {
 340         switch(version) {
 341         case IPC_64:
 342                 return copy_to_user(buf, in, sizeof(*in));
 343         case IPC_OLD:
 344         {
 345                 struct msqid_ds out;
 346
 347                 memset(&out, 0, sizeof(out));
 348
 349                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 350
 351                 out.msg_stime           = in->msg_stime;
 352                 out.msg_rtime           = in->msg_rtime;
 353                 out.msg_ctime           = in->msg_ctime;
 354
 355                 if (in->msg_cbytes > USHRT_MAX)
 356                         out.msg_cbytes  = USHRT_MAX;
 357                 else
 358                         out.msg_cbytes  = in->msg_cbytes;
 359                 out.msg_lcbytes         = in->msg_cbytes;
 360
 361                 if (in->msg_qnum > USHRT_MAX)
 362                         out.msg_qnum    = USHRT_MAX;
 363                 else
 364                         out.msg_qnum    = in->msg_qnum;
 365
 366                 if (in->msg_qbytes > USHRT_MAX)
 367                         out.msg_qbytes  = USHRT_MAX;
 368                 else
 369                         out.msg_qbytes  = in->msg_qbytes;
 370                 out.msg_lqbytes         = in->msg_qbytes;
 371
 372                 out.msg_lspid           = in->msg_lspid;
 373                 out.msg_lrpid           = in->msg_lrpid;
 374
 375                 return copy_to_user(buf, &out, sizeof(out));
 376         }
 377         default:
 378                 return -EINVAL;
 379         }
 380 }
 381
 382 static inline unsigned long
 383 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 384 {
 385         switch(version) {
 386         case IPC_64:
 387                 if (copy_from_user(out, buf, sizeof(*out)))
 388                         return -EFAULT;
 389                 return 0;
 390         case IPC_OLD:
 391         {
 392                 struct msqid_ds tbuf_old;
 393
 394                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 395                         return -EFAULT;
 396
 397                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 398                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 399                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 400
 401                 if (tbuf_old.msg_qbytes == 0)
 402                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 403                 else
 404                         out->msg_qbytes = tbuf_old.msg_qbytes;
 405
 406                 return 0;
 407         }
 408         default:
 409                 return -EINVAL;
 410         }
 411 }
 412
 413 /*
 414  * This function handles some msgctl commands which require the rw_mutex
 415  * to be held in write mode.
 416  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 417  */
 418 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 419                        struct msqid_ds __user *buf, int version)
 420 {
 421         struct kern_ipc_perm *ipcp;
 422         struct msqid64_ds uninitialized_var(msqid64);
 423         struct msg_queue *msq;
 424         int err;
 425
 426         if (cmd == IPC_SET) {
 427                 if (copy_msqid_from_user(&msqid64, buf, version))
 428                         return -EFAULT;
 429         }
 430
 431         down_write(&msg_ids(ns).rw_mutex);
 432         rcu_read_lock();
 433
 434         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 435                                       &msqid64.msg_perm, msqid64.msg_qbytes);
 436         if (IS_ERR(ipcp)) {
 437                 err = PTR_ERR(ipcp);
 438                 goto out_unlock1;
 439         }
 440
 441         msq = container_of(ipcp, struct msg_queue, q_perm);
 442
 443         err = security_msg_queue_msgctl(msq, cmd);
 444         if (err)
 445                 goto out_unlock1;
 446
 447         switch (cmd) {
 448         case IPC_RMID:
 449                 ipc_lock_object(&msq->q_perm);
 450                 /* freeque unlocks the ipc object and rcu */
 451                 freeque(ns, ipcp);
 452                 goto out_up;
 453         case IPC_SET:
 454                 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
 455                     !capable(CAP_SYS_RESOURCE)) {
 456                         err = -EPERM;
 457                         goto out_unlock1;
 458                 }
 459
 460                 ipc_lock_object(&msq->q_perm);
 461                 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
 462                 if (err)
 463                         goto out_unlock0;
 464
 465                 msq->q_qbytes = msqid64.msg_qbytes;
 466
 467                 msq->q_ctime = get_seconds();
 468                 /* sleeping receivers might be excluded by
 469                  * stricter permissions.
 470                  */
 471                 expunge_all(msq, -EAGAIN);
 472                 /* sleeping senders might be able to send
 473                  * due to a larger queue size.
 474                  */
 475                 ss_wakeup(&msq->q_senders, 0);
 476                 break;
 477         default:
 478                 err = -EINVAL;
 479                 goto out_unlock1;
 480         }
 481
 482 out_unlock0:
 483         ipc_unlock_object(&msq->q_perm);
 484 out_unlock1:
 485         rcu_read_unlock();
 486 out_up:
 487         up_write(&msg_ids(ns).rw_mutex);
 488         return err;
 489 }
 490
 491 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
 492                          int cmd, int version, void __user *buf)
 493 {
 494         int err;
 495         struct msg_queue *msq;
 496
 497         switch (cmd) {
 498         case IPC_INFO:
 499         case MSG_INFO:
 500         {
 501                 struct msginfo msginfo;
 502                 int max_id;
 503
 504                 if (!buf)
 505                         return -EFAULT;
 506
 507                 /*
 508                  * We must not return kernel stack data.
 509                  * due to padding, it's not enough
 510                  * to set all member fields.
 511                  */
 512                 err = security_msg_queue_msgctl(NULL, cmd);
 513                 if (err)
 514                         return err;
 515
 516                 memset(&msginfo, 0, sizeof(msginfo));
 517                 msginfo.msgmni = ns->msg_ctlmni;
 518                 msginfo.msgmax = ns->msg_ctlmax;
 519                 msginfo.msgmnb = ns->msg_ctlmnb;
 520                 msginfo.msgssz = MSGSSZ;
 521                 msginfo.msgseg = MSGSEG;
 522                 down_read(&msg_ids(ns).rw_mutex);
 523                 if (cmd == MSG_INFO) {
 524                         msginfo.msgpool = msg_ids(ns).in_use;
 525                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 526                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 527                 } else {
 528                         msginfo.msgmap = MSGMAP;
 529                         msginfo.msgpool = MSGPOOL;
 530                         msginfo.msgtql = MSGTQL;
 531                 }
 532                 max_id = ipc_get_maxid(&msg_ids(ns));
 533                 up_read(&msg_ids(ns).rw_mutex);
 534                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 535                         return -EFAULT;
 536                 return (max_id < 0) ? 0 : max_id;
 537         }
 538
 539         case MSG_STAT:
 540         case IPC_STAT:
 541         {
 542                 struct msqid64_ds tbuf;
 543                 int success_return;
 544
 545                 if (!buf)
 546                         return -EFAULT;
 547
 548                 memset(&tbuf, 0, sizeof(tbuf));
 549
 550                 rcu_read_lock();
 551                 if (cmd == MSG_STAT) {
 552                         msq = msq_obtain_object(ns, msqid);
 553                         if (IS_ERR(msq)) {
 554                                 err = PTR_ERR(msq);
 555                                 goto out_unlock;
 556                         }
 557                         success_return = msq->q_perm.id;
 558                 } else {
 559                         msq = msq_obtain_object_check(ns, msqid);
 560                         if (IS_ERR(msq)) {
 561                                 err = PTR_ERR(msq);
 562                                 goto out_unlock;
 563                         }
 564                         success_return = 0;
 565                 }
 566
 567                 err = -EACCES;
 568                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 569                         goto out_unlock;
 570
 571                 err = security_msg_queue_msgctl(msq, cmd);
 572                 if (err)
 573                         goto out_unlock;
 574
 575                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 576                 tbuf.msg_stime  = msq->q_stime;
 577                 tbuf.msg_rtime  = msq->q_rtime;
 578                 tbuf.msg_ctime  = msq->q_ctime;
 579                 tbuf.msg_cbytes = msq->q_cbytes;
 580                 tbuf.msg_qnum   = msq->q_qnum;
 581                 tbuf.msg_qbytes = msq->q_qbytes;
 582                 tbuf.msg_lspid  = msq->q_lspid;
 583                 tbuf.msg_lrpid  = msq->q_lrpid;
 584                 rcu_read_unlock();
 585
 586                 if (copy_msqid_to_user(buf, &tbuf, version))
 587                         return -EFAULT;
 588                 return success_return;
 589         }
 590
 591         default:
 592                 return -EINVAL;
 593         }
 594
 595         return err;
 596 out_unlock:
 597         rcu_read_unlock();
 598         return err;
 599 }
 600
 601 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 602 {
 603         int version;
 604         struct ipc_namespace *ns;
 605
 606         if (msqid < 0 || cmd < 0)
 607                 return -EINVAL;
 608
 609         version = ipc_parse_version(&cmd);
 610         ns = current->nsproxy->ipc_ns;
 611
 612         switch (cmd) {
 613         case IPC_INFO:
 614         case MSG_INFO:
 615         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 616         case IPC_STAT:
 617                 return msgctl_nolock(ns, msqid, cmd, version, buf);
 618         case IPC_SET:
 619         case IPC_RMID:
 620                 return msgctl_down(ns, msqid, cmd, buf, version);
 621         default:
 622                 return  -EINVAL;
 623         }
 624 }
 625
 626 static int testmsg(struct msg_msg *msg, long type, int mode)
 627 {
 628         switch(mode)
 629         {
 630                 case SEARCH_ANY:
 631                 case SEARCH_NUMBER:
 632                         return 1;
 633                 case SEARCH_LESSEQUAL:
 634                         if (msg->m_type <=type)
 635                                 return 1;
 636                         break;
 637                 case SEARCH_EQUAL:
 638                         if (msg->m_type == type)
 639                                 return 1;
 640                         break;
 641                 case SEARCH_NOTEQUAL:
 642                         if (msg->m_type != type)
 643                                 return 1;
 644                         break;
 645         }
 646         return 0;
 647 }
 648
 649 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 650 {
 651         struct msg_receiver *msr, *t;
 652
 653         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 654                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 655                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 656                                                msr->r_msgtype, msr->r_mode)) {
 657
 658                         list_del(&msr->r_list);
 659                         if (msr->r_maxsize < msg->m_ts) {
 660                                 msr->r_msg = NULL;
 661                                 wake_up_process(msr->r_tsk);
 662                                 smp_mb();
 663                                 msr->r_msg = ERR_PTR(-E2BIG);
 664                         } else {
 665                                 msr->r_msg = NULL;
 666                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 667                                 msq->q_rtime = get_seconds();
 668                                 wake_up_process(msr->r_tsk);
 669                                 smp_mb();
 670                                 msr->r_msg = msg;
 671
 672                                 return 1;
 673                         }
 674                 }
 675         }
 676         return 0;
 677 }
 678
 679 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 680                 size_t msgsz, int msgflg)
 681 {
 682         struct msg_queue *msq;
 683         struct msg_msg *msg;
 684         int err;
 685         struct ipc_namespace *ns;
 686
 687         ns = current->nsproxy->ipc_ns;
 688
 689         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 690                 return -EINVAL;
 691         if (mtype < 1)
 692                 return -EINVAL;
 693
 694         msg = load_msg(mtext, msgsz);
 695         if (IS_ERR(msg))
 696                 return PTR_ERR(msg);
 697
 698         msg->m_type = mtype;
 699         msg->m_ts = msgsz;
 700
 701         rcu_read_lock();
 702         msq = msq_obtain_object_check(ns, msqid);
 703         if (IS_ERR(msq)) {
 704                 err = PTR_ERR(msq);
 705                 goto out_unlock1;
 706         }
 707
 708         for (;;) {
 709                 struct msg_sender s;
 710
 711                 err = -EACCES;
 712                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 713                         goto out_unlock1;
 714
 715                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 716                 if (err)
 717                         goto out_unlock1;
 718
 719                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 720                                 1 + msq->q_qnum <= msq->q_qbytes) {
 721                         break;
 722                 }
 723
 724                 /* queue full, wait: */
 725                 if (msgflg & IPC_NOWAIT) {
 726                         err = -EAGAIN;
 727                         goto out_unlock1;
 728                 }
 729
 730                 ipc_lock_object(&msq->q_perm);
 731                 ss_add(msq, &s);
 732
 733                 if (!ipc_rcu_getref(msq)) {
 734                         err = -EIDRM;
 735                         goto out_unlock0;
 736                 }
 737
 738                 ipc_unlock_object(&msq->q_perm);
 739                 rcu_read_unlock();
 740                 schedule();
 741
 742                 rcu_read_lock();
 743                 ipc_lock_object(&msq->q_perm);
 744
 745                 ipc_rcu_putref(msq);
 746                 if (msq->q_perm.deleted) {
 747                         err = -EIDRM;
 748                         goto out_unlock0;
 749                 }
 750
 751                 ss_del(&s);
 752
 753                 if (signal_pending(current)) {
 754                         err = -ERESTARTNOHAND;
 755                         goto out_unlock0;
 756                 }
 757
 758                 ipc_unlock_object(&msq->q_perm);
 759         }
 760
 761         ipc_lock_object(&msq->q_perm);
 762         msq->q_lspid = task_tgid_vnr(current);
 763         msq->q_stime = get_seconds();
 764
 765         if (!pipelined_send(msq, msg)) {
 766                 /* no one is waiting for this message, enqueue it */
 767                 list_add_tail(&msg->m_list, &msq->q_messages);
 768                 msq->q_cbytes += msgsz;
 769                 msq->q_qnum++;
 770                 atomic_add(msgsz, &ns->msg_bytes);
 771                 atomic_inc(&ns->msg_hdrs);
 772         }
 773
 774         err = 0;
 775         msg = NULL;
 776
 777 out_unlock0:
 778         ipc_unlock_object(&msq->q_perm);
 779 out_unlock1:
 780         rcu_read_unlock();
 781         if (msg != NULL)
 782                 free_msg(msg);
 783         return err;
 784 }
 785
 786 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 787                 int, msgflg)
 788 {
 789         long mtype;
 790
 791         if (get_user(mtype, &msgp->mtype))
 792                 return -EFAULT;
 793         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 794 }
 795
 796 static inline int convert_mode(long *msgtyp, int msgflg)
 797 {
 798         if (msgflg & MSG_COPY)
 799                 return SEARCH_NUMBER;
 800         /*
 801          *  find message of correct type.
 802          *  msgtyp = 0 => get first.
 803          *  msgtyp > 0 => get first message of matching type.
 804          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 805          */
 806         if (*msgtyp == 0)
 807                 return SEARCH_ANY;
 808         if (*msgtyp < 0) {
 809                 *msgtyp = -*msgtyp;
 810                 return SEARCH_LESSEQUAL;
 811         }
 812         if (msgflg & MSG_EXCEPT)
 813                 return SEARCH_NOTEQUAL;
 814         return SEARCH_EQUAL;
 815 }
 816
 817 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 818 {
 819         struct msgbuf __user *msgp = dest;
 820         size_t msgsz;
 821
 822         if (put_user(msg->m_type, &msgp->mtype))
 823                 return -EFAULT;
 824
 825         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 826         if (store_msg(msgp->mtext, msg, msgsz))
 827                 return -EFAULT;
 828         return msgsz;
 829 }
 830
 831 #ifdef CONFIG_CHECKPOINT_RESTORE
 832 /*
 833  * This function creates new kernel message structure, large enough to store
 834  * bufsz message bytes.
 835  */
 836 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 837 {
 838         struct msg_msg *copy;
 839
 840         /*
 841          * Create dummy message to copy real message to.
 842          */
 843         copy = load_msg(buf, bufsz);
 844         if (!IS_ERR(copy))
 845                 copy->m_ts = bufsz;
 846         return copy;
 847 }
 848
 849 static inline void free_copy(struct msg_msg *copy)
 850 {
 851         if (copy)
 852                 free_msg(copy);
 853 }
 854 #else
 855 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 856 {
 857         return ERR_PTR(-ENOSYS);
 858 }
 859
 860 static inline void free_copy(struct msg_msg *copy)
 861 {
 862 }
 863 #endif
 864
 865 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 866 {
 867         struct msg_msg *msg, *found = NULL;
 868         long count = 0;
 869
 870         list_for_each_entry(msg, &msq->q_messages, m_list) {
 871                 if (testmsg(msg, *msgtyp, mode) &&
 872                     !security_msg_queue_msgrcv(msq, msg, current,
 873                                                *msgtyp, mode)) {
 874                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
 875                                 *msgtyp = msg->m_type - 1;
 876                                 found = msg;
 877                         } else if (mode == SEARCH_NUMBER) {
 878                                 if (*msgtyp == count)
 879                                         return msg;
 880                         } else
 881                                 return msg;
 882                         count++;
 883                 }
 884         }
 885
 886         return found ?: ERR_PTR(-EAGAIN);
 887 }
 888
 889 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
 890                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
 891 {
 892         int mode;
 893         struct msg_queue *msq;
 894         struct ipc_namespace *ns;
 895         struct msg_msg *msg, *copy = NULL;
 896
 897         ns = current->nsproxy->ipc_ns;
 898
 899         if (msqid < 0 || (long) bufsz < 0)
 900                 return -EINVAL;
 901
 902         if (msgflg & MSG_COPY) {
 903                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
 904                 if (IS_ERR(copy))
 905                         return PTR_ERR(copy);
 906         }
 907         mode = convert_mode(&msgtyp, msgflg);
 908
 909         rcu_read_lock();
 910         msq = msq_obtain_object_check(ns, msqid);
 911         if (IS_ERR(msq)) {
 912                 rcu_read_unlock();
 913                 free_copy(copy);
 914                 return PTR_ERR(msq);
 915         }
 916
 917         for (;;) {
 918                 struct msg_receiver msr_d;
 919
 920                 msg = ERR_PTR(-EACCES);
 921                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 922                         goto out_unlock1;
 923
 924                 ipc_lock_object(&msq->q_perm);
 925                 msg = find_msg(msq, &msgtyp, mode);
 926                 if (!IS_ERR(msg)) {
 927                         /*
 928                          * Found a suitable message.
 929                          * Unlink it from the queue.
 930                          */
 931                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 932                                 msg = ERR_PTR(-E2BIG);
 933                                 goto out_unlock0;
 934                         }
 935                         /*
 936                          * If we are copying, then do not unlink message and do
 937                          * not update queue parameters.
 938                          */
 939                         if (msgflg & MSG_COPY) {
 940                                 msg = copy_msg(msg, copy);
 941                                 goto out_unlock0;
 942                         }
 943
 944                         list_del(&msg->m_list);
 945                         msq->q_qnum--;
 946                         msq->q_rtime = get_seconds();
 947                         msq->q_lrpid = task_tgid_vnr(current);
 948                         msq->q_cbytes -= msg->m_ts;
 949                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 950                         atomic_dec(&ns->msg_hdrs);
 951                         ss_wakeup(&msq->q_senders, 0);
 952
 953                         goto out_unlock0;
 954                 }
 955
 956                 /* No message waiting. Wait for a message */
 957                 if (msgflg & IPC_NOWAIT) {
 958                         msg = ERR_PTR(-ENOMSG);
 959                         goto out_unlock0;
 960                 }
 961
 962                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 963                 msr_d.r_tsk = current;
 964                 msr_d.r_msgtype = msgtyp;
 965                 msr_d.r_mode = mode;
 966                 if (msgflg & MSG_NOERROR)
 967                         msr_d.r_maxsize = INT_MAX;
 968                 else
 969                         msr_d.r_maxsize = bufsz;
 970                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 971                 current->state = TASK_INTERRUPTIBLE;
 972
 973                 ipc_unlock_object(&msq->q_perm);
 974                 rcu_read_unlock();
 975                 schedule();
 976
 977                 /* Lockless receive, part 1:
 978                  * Disable preemption.  We don't hold a reference to the queue
 979                  * and getting a reference would defeat the idea of a lockless
 980                  * operation, thus the code relies on rcu to guarantee the
 981                  * existence of msq:
 982                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 983                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 984                  * rcu_read_lock() prevents preemption between reading r_msg
 985                  * and acquiring the q_perm.lock in ipc_lock_object().
 986                  */
 987                 rcu_read_lock();
 988
 989                 /* Lockless receive, part 2:
 990                  * Wait until pipelined_send or expunge_all are outside of
 991                  * wake_up_process(). There is a race with exit(), see
 992                  * ipc/mqueue.c for the details.
 993                  */
 994                 msg = (struct msg_msg*)msr_d.r_msg;
 995                 while (msg == NULL) {
 996                         cpu_relax();
 997                         msg = (struct msg_msg *)msr_d.r_msg;
 998                 }
 999
1000                 /* Lockless receive, part 3:
1001                  * If there is a message or an error then accept it without
1002                  * locking.
1003                  */
1004                 if (msg != ERR_PTR(-EAGAIN))
1005                         goto out_unlock1;
1006
1007                 /* Lockless receive, part 3:
1008                  * Acquire the queue spinlock.
1009                  */
1010                 ipc_lock_object(&msq->q_perm);
1011
1012                 /* Lockless receive, part 4:
1013                  * Repeat test after acquiring the spinlock.
1014                  */
1015                 msg = (struct msg_msg*)msr_d.r_msg;
1016                 if (msg != ERR_PTR(-EAGAIN))
1017                         goto out_unlock0;
1018
1019                 list_del(&msr_d.r_list);
1020                 if (signal_pending(current)) {
1021                         msg = ERR_PTR(-ERESTARTNOHAND);
1022                         goto out_unlock0;
1023                 }
1024
1025                 ipc_unlock_object(&msq->q_perm);
1026         }
1027
1028 out_unlock0:
1029         ipc_unlock_object(&msq->q_perm);
1030 out_unlock1:
1031         rcu_read_unlock();
1032         if (IS_ERR(msg)) {
1033                 free_copy(copy);
1034                 return PTR_ERR(msg);
1035         }
1036
1037         bufsz = msg_handler(buf, msg, bufsz);
1038         free_msg(msg);
1039
1040         return bufsz;
1041 }
1042
1043 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1044                 long, msgtyp, int, msgflg)
1045 {
1046         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1047 }
1048
1049 #ifdef CONFIG_PROC_FS
1050 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1051 {
1052         struct user_namespace *user_ns = seq_user_ns(s);
1053         struct msg_queue *msq = it;
1054
1055         return seq_printf(s,
1056                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1057                         msq->q_perm.key,
1058                         msq->q_perm.id,
1059                         msq->q_perm.mode,
1060                         msq->q_cbytes,
1061                         msq->q_qnum,
1062                         msq->q_lspid,
1063                         msq->q_lrpid,
1064                         from_kuid_munged(user_ns, msq->q_perm.uid),
1065                         from_kgid_munged(user_ns, msq->q_perm.gid),
1066                         from_kuid_munged(user_ns, msq->q_perm.cuid),
1067                         from_kgid_munged(user_ns, msq->q_perm.cgid),
1068                         msq->q_stime,
1069                         msq->q_rtime,
1070                         msq->q_ctime);
1071 }
1072 #endif