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