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