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/mutex.h>
  38 #include <linux/nsproxy.h>
  39
  40 #include <asm/current.h>
  41 #include <asm/uaccess.h>
  42 #include "util.h"
  43
  44 /*
  45  * one msg_receiver structure for each sleeping receiver:
  46  */
  47 struct msg_receiver {
  48         struct list_head        r_list;
  49         struct task_struct      *r_tsk;
  50
  51         int                     r_mode;
  52         long                    r_msgtype;
  53         long                    r_maxsize;
  54
  55         struct msg_msg          *volatile r_msg;
  56 };
  57
  58 /* one msg_sender for each sleeping sender */
  59 struct msg_sender {
  60         struct list_head        list;
  61         struct task_struct      *tsk;
  62 };
  63
  64 #define SEARCH_ANY              1
  65 #define SEARCH_EQUAL            2
  66 #define SEARCH_NOTEQUAL         3
  67 #define SEARCH_LESSEQUAL        4
  68
  69 static atomic_t msg_bytes =     ATOMIC_INIT(0);
  70 static atomic_t msg_hdrs =      ATOMIC_INIT(0);
  71
  72 static struct ipc_ids init_msg_ids;
  73
  74 #define msg_ids(ns)     (*((ns)->ids[IPC_MSG_IDS]))
  75
  76 #define msg_lock(ns, id)        ((struct msg_queue*)ipc_lock(&msg_ids(ns), id))
  77 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  78 #define msg_checkid(ns, msq, msgid)     \
  79         ipc_checkid(&msg_ids(ns), &msq->q_perm, msgid)
  80 #define msg_buildid(ns, id, seq) \
  81         ipc_buildid(&msg_ids(ns), id, seq)
  82
  83 static void freeque(struct ipc_namespace *, struct msg_queue *);
  84 static int newque(struct ipc_namespace *, struct ipc_params *);
  85 #ifdef CONFIG_PROC_FS
  86 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  87 #endif
  88
  89 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  90 {
  91         ns->ids[IPC_MSG_IDS] = ids;
  92         ns->msg_ctlmax = MSGMAX;
  93         ns->msg_ctlmnb = MSGMNB;
  94         ns->msg_ctlmni = MSGMNI;
  95         ipc_init_ids(ids);
  96 }
  97
  98 int msg_init_ns(struct ipc_namespace *ns)
  99 {
 100         struct ipc_ids *ids;
 101
 102         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 103         if (ids == NULL)
 104                 return -ENOMEM;
 105
 106         __msg_init_ns(ns, ids);
 107         return 0;
 108 }
 109
 110 void msg_exit_ns(struct ipc_namespace *ns)
 111 {
 112         struct msg_queue *msq;
 113         int next_id;
 114         int total, in_use;
 115
 116         mutex_lock(&msg_ids(ns).mutex);
 117
 118         in_use = msg_ids(ns).in_use;
 119
 120         for (total = 0, next_id = 0; total < in_use; next_id++) {
 121                 msq = idr_find(&msg_ids(ns).ipcs_idr, next_id);
 122                 if (msq == NULL)
 123                         continue;
 124                 ipc_lock_by_ptr(&msq->q_perm);
 125                 freeque(ns, msq);
 126                 total++;
 127         }
 128         mutex_unlock(&msg_ids(ns).mutex);
 129
 130         kfree(ns->ids[IPC_MSG_IDS]);
 131         ns->ids[IPC_MSG_IDS] = NULL;
 132 }
 133
 134 void __init msg_init(void)
 135 {
 136         __msg_init_ns(&init_ipc_ns, &init_msg_ids);
 137         ipc_init_proc_interface("sysvipc/msg",
 138                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 139                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 140 }
 141
 142 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 143 {
 144         ipc_rmid(&msg_ids(ns), &s->q_perm);
 145 }
 146
 147 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 148 {
 149         struct msg_queue *msq;
 150         int id, retval;
 151         key_t key = params->key;
 152         int msgflg = params->flg;
 153
 154         msq = ipc_rcu_alloc(sizeof(*msq));
 155         if (!msq)
 156                 return -ENOMEM;
 157
 158         msq->q_perm.mode = msgflg & S_IRWXUGO;
 159         msq->q_perm.key = key;
 160
 161         msq->q_perm.security = NULL;
 162         retval = security_msg_queue_alloc(msq);
 163         if (retval) {
 164                 ipc_rcu_putref(msq);
 165                 return retval;
 166         }
 167
 168         /*
 169          * ipc_addid() locks msq
 170          */
 171         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 172         if (id == -1) {
 173                 security_msg_queue_free(msq);
 174                 ipc_rcu_putref(msq);
 175                 return -ENOSPC;
 176         }
 177
 178         msq->q_perm.id = msg_buildid(ns, id, msq->q_perm.seq);
 179         msq->q_stime = msq->q_rtime = 0;
 180         msq->q_ctime = get_seconds();
 181         msq->q_cbytes = msq->q_qnum = 0;
 182         msq->q_qbytes = ns->msg_ctlmnb;
 183         msq->q_lspid = msq->q_lrpid = 0;
 184         INIT_LIST_HEAD(&msq->q_messages);
 185         INIT_LIST_HEAD(&msq->q_receivers);
 186         INIT_LIST_HEAD(&msq->q_senders);
 187
 188         msg_unlock(msq);
 189
 190         return msq->q_perm.id;
 191 }
 192
 193 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 194 {
 195         mss->tsk = current;
 196         current->state = TASK_INTERRUPTIBLE;
 197         list_add_tail(&mss->list, &msq->q_senders);
 198 }
 199
 200 static inline void ss_del(struct msg_sender *mss)
 201 {
 202         if (mss->list.next != NULL)
 203                 list_del(&mss->list);
 204 }
 205
 206 static void ss_wakeup(struct list_head *h, int kill)
 207 {
 208         struct list_head *tmp;
 209
 210         tmp = h->next;
 211         while (tmp != h) {
 212                 struct msg_sender *mss;
 213
 214                 mss = list_entry(tmp, struct msg_sender, list);
 215                 tmp = tmp->next;
 216                 if (kill)
 217                         mss->list.next = NULL;
 218                 wake_up_process(mss->tsk);
 219         }
 220 }
 221
 222 static void expunge_all(struct msg_queue *msq, int res)
 223 {
 224         struct list_head *tmp;
 225
 226         tmp = msq->q_receivers.next;
 227         while (tmp != &msq->q_receivers) {
 228                 struct msg_receiver *msr;
 229
 230                 msr = list_entry(tmp, struct msg_receiver, r_list);
 231                 tmp = tmp->next;
 232                 msr->r_msg = NULL;
 233                 wake_up_process(msr->r_tsk);
 234                 smp_mb();
 235                 msr->r_msg = ERR_PTR(res);
 236         }
 237 }
 238
 239 /*
 240  * freeque() wakes up waiters on the sender and receiver waiting queue,
 241  * removes the message queue from message queue ID
 242  * IDR, and cleans up all the messages associated with this queue.
 243  *
 244  * msg_ids.mutex and the spinlock for this message queue are held
 245  * before freeque() is called. msg_ids.mutex remains locked on exit.
 246  */
 247 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
 248 {
 249         struct list_head *tmp;
 250
 251         expunge_all(msq, -EIDRM);
 252         ss_wakeup(&msq->q_senders, 1);
 253         msg_rmid(ns, msq);
 254         msg_unlock(msq);
 255
 256         tmp = msq->q_messages.next;
 257         while (tmp != &msq->q_messages) {
 258                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 259
 260                 tmp = tmp->next;
 261                 atomic_dec(&msg_hdrs);
 262                 free_msg(msg);
 263         }
 264         atomic_sub(msq->q_cbytes, &msg_bytes);
 265         security_msg_queue_free(msq);
 266         ipc_rcu_putref(msq);
 267 }
 268
 269 static inline int msg_security(void *msq, int msgflg)
 270 {
 271         return security_msg_queue_associate((struct msg_queue *) msq, msgflg);
 272 }
 273
 274 asmlinkage long sys_msgget(key_t key, int msgflg)
 275 {
 276         struct ipc_namespace *ns;
 277         struct ipc_ops msg_ops;
 278         struct ipc_params msg_params;
 279
 280         ns = current->nsproxy->ipc_ns;
 281
 282         msg_ops.getnew = newque;
 283         msg_ops.associate = msg_security;
 284         msg_ops.more_checks = NULL;
 285
 286         msg_params.key = key;
 287         msg_params.flg = msgflg;
 288
 289         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 290 }
 291
 292 static inline unsigned long
 293 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 294 {
 295         switch(version) {
 296         case IPC_64:
 297                 return copy_to_user(buf, in, sizeof(*in));
 298         case IPC_OLD:
 299         {
 300                 struct msqid_ds out;
 301
 302                 memset(&out, 0, sizeof(out));
 303
 304                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 305
 306                 out.msg_stime           = in->msg_stime;
 307                 out.msg_rtime           = in->msg_rtime;
 308                 out.msg_ctime           = in->msg_ctime;
 309
 310                 if (in->msg_cbytes > USHRT_MAX)
 311                         out.msg_cbytes  = USHRT_MAX;
 312                 else
 313                         out.msg_cbytes  = in->msg_cbytes;
 314                 out.msg_lcbytes         = in->msg_cbytes;
 315
 316                 if (in->msg_qnum > USHRT_MAX)
 317                         out.msg_qnum    = USHRT_MAX;
 318                 else
 319                         out.msg_qnum    = in->msg_qnum;
 320
 321                 if (in->msg_qbytes > USHRT_MAX)
 322                         out.msg_qbytes  = USHRT_MAX;
 323                 else
 324                         out.msg_qbytes  = in->msg_qbytes;
 325                 out.msg_lqbytes         = in->msg_qbytes;
 326
 327                 out.msg_lspid           = in->msg_lspid;
 328                 out.msg_lrpid           = in->msg_lrpid;
 329
 330                 return copy_to_user(buf, &out, sizeof(out));
 331         }
 332         default:
 333                 return -EINVAL;
 334         }
 335 }
 336
 337 struct msq_setbuf {
 338         unsigned long   qbytes;
 339         uid_t           uid;
 340         gid_t           gid;
 341         mode_t          mode;
 342 };
 343
 344 static inline unsigned long
 345 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 346 {
 347         switch(version) {
 348         case IPC_64:
 349         {
 350                 struct msqid64_ds tbuf;
 351
 352                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 353                         return -EFAULT;
 354
 355                 out->qbytes             = tbuf.msg_qbytes;
 356                 out->uid                = tbuf.msg_perm.uid;
 357                 out->gid                = tbuf.msg_perm.gid;
 358                 out->mode               = tbuf.msg_perm.mode;
 359
 360                 return 0;
 361         }
 362         case IPC_OLD:
 363         {
 364                 struct msqid_ds tbuf_old;
 365
 366                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 367                         return -EFAULT;
 368
 369                 out->uid                = tbuf_old.msg_perm.uid;
 370                 out->gid                = tbuf_old.msg_perm.gid;
 371                 out->mode               = tbuf_old.msg_perm.mode;
 372
 373                 if (tbuf_old.msg_qbytes == 0)
 374                         out->qbytes     = tbuf_old.msg_lqbytes;
 375                 else
 376                         out->qbytes     = tbuf_old.msg_qbytes;
 377
 378                 return 0;
 379         }
 380         default:
 381                 return -EINVAL;
 382         }
 383 }
 384
 385 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 386 {
 387         struct kern_ipc_perm *ipcp;
 388         struct msq_setbuf uninitialized_var(setbuf);
 389         struct msg_queue *msq;
 390         int err, version;
 391         struct ipc_namespace *ns;
 392
 393         if (msqid < 0 || cmd < 0)
 394                 return -EINVAL;
 395
 396         version = ipc_parse_version(&cmd);
 397         ns = current->nsproxy->ipc_ns;
 398
 399         switch (cmd) {
 400         case IPC_INFO:
 401         case MSG_INFO:
 402         {
 403                 struct msginfo msginfo;
 404                 int max_id;
 405
 406                 if (!buf)
 407                         return -EFAULT;
 408                 /*
 409                  * We must not return kernel stack data.
 410                  * due to padding, it's not enough
 411                  * to set all member fields.
 412                  */
 413                 err = security_msg_queue_msgctl(NULL, cmd);
 414                 if (err)
 415                         return err;
 416
 417                 memset(&msginfo, 0, sizeof(msginfo));
 418                 msginfo.msgmni = ns->msg_ctlmni;
 419                 msginfo.msgmax = ns->msg_ctlmax;
 420                 msginfo.msgmnb = ns->msg_ctlmnb;
 421                 msginfo.msgssz = MSGSSZ;
 422                 msginfo.msgseg = MSGSEG;
 423                 mutex_lock(&msg_ids(ns).mutex);
 424                 if (cmd == MSG_INFO) {
 425                         msginfo.msgpool = msg_ids(ns).in_use;
 426                         msginfo.msgmap = atomic_read(&msg_hdrs);
 427                         msginfo.msgtql = atomic_read(&msg_bytes);
 428                 } else {
 429                         msginfo.msgmap = MSGMAP;
 430                         msginfo.msgpool = MSGPOOL;
 431                         msginfo.msgtql = MSGTQL;
 432                 }
 433                 max_id = ipc_get_maxid(&msg_ids(ns));
 434                 mutex_unlock(&msg_ids(ns).mutex);
 435                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 436                         return -EFAULT;
 437                 return (max_id < 0) ? 0 : max_id;
 438         }
 439         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 440         case IPC_STAT:
 441         {
 442                 struct msqid64_ds tbuf;
 443                 int success_return;
 444
 445                 if (!buf)
 446                         return -EFAULT;
 447
 448                 memset(&tbuf, 0, sizeof(tbuf));
 449
 450                 msq = msg_lock(ns, msqid);
 451                 if (msq == NULL)
 452                         return -EINVAL;
 453
 454                 if (cmd == MSG_STAT) {
 455                         success_return = msq->q_perm.id;
 456                 } else {
 457                         err = -EIDRM;
 458                         if (msg_checkid(ns, msq, msqid))
 459                                 goto out_unlock;
 460                         success_return = 0;
 461                 }
 462                 err = -EACCES;
 463                 if (ipcperms(&msq->q_perm, S_IRUGO))
 464                         goto out_unlock;
 465
 466                 err = security_msg_queue_msgctl(msq, cmd);
 467                 if (err)
 468                         goto out_unlock;
 469
 470                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 471                 tbuf.msg_stime  = msq->q_stime;
 472                 tbuf.msg_rtime  = msq->q_rtime;
 473                 tbuf.msg_ctime  = msq->q_ctime;
 474                 tbuf.msg_cbytes = msq->q_cbytes;
 475                 tbuf.msg_qnum   = msq->q_qnum;
 476                 tbuf.msg_qbytes = msq->q_qbytes;
 477                 tbuf.msg_lspid  = msq->q_lspid;
 478                 tbuf.msg_lrpid  = msq->q_lrpid;
 479                 msg_unlock(msq);
 480                 if (copy_msqid_to_user(buf, &tbuf, version))
 481                         return -EFAULT;
 482                 return success_return;
 483         }
 484         case IPC_SET:
 485                 if (!buf)
 486                         return -EFAULT;
 487                 if (copy_msqid_from_user(&setbuf, buf, version))
 488                         return -EFAULT;
 489                 break;
 490         case IPC_RMID:
 491                 break;
 492         default:
 493                 return  -EINVAL;
 494         }
 495
 496         mutex_lock(&msg_ids(ns).mutex);
 497         msq = msg_lock(ns, msqid);
 498         err = -EINVAL;
 499         if (msq == NULL)
 500                 goto out_up;
 501
 502         err = -EIDRM;
 503         if (msg_checkid(ns, msq, msqid))
 504                 goto out_unlock_up;
 505         ipcp = &msq->q_perm;
 506
 507         err = audit_ipc_obj(ipcp);
 508         if (err)
 509                 goto out_unlock_up;
 510         if (cmd == IPC_SET) {
 511                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
 512                                          setbuf.mode);
 513                 if (err)
 514                         goto out_unlock_up;
 515         }
 516
 517         err = -EPERM;
 518         if (current->euid != ipcp->cuid &&
 519             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 520                 /* We _could_ check for CAP_CHOWN above, but we don't */
 521                 goto out_unlock_up;
 522
 523         err = security_msg_queue_msgctl(msq, cmd);
 524         if (err)
 525                 goto out_unlock_up;
 526
 527         switch (cmd) {
 528         case IPC_SET:
 529         {
 530                 err = -EPERM;
 531                 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 532                         goto out_unlock_up;
 533
 534                 msq->q_qbytes = setbuf.qbytes;
 535
 536                 ipcp->uid = setbuf.uid;
 537                 ipcp->gid = setbuf.gid;
 538                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 539                              (S_IRWXUGO & setbuf.mode);
 540                 msq->q_ctime = get_seconds();
 541                 /* sleeping receivers might be excluded by
 542                  * stricter permissions.
 543                  */
 544                 expunge_all(msq, -EAGAIN);
 545                 /* sleeping senders might be able to send
 546                  * due to a larger queue size.
 547                  */
 548                 ss_wakeup(&msq->q_senders, 0);
 549                 msg_unlock(msq);
 550                 break;
 551         }
 552         case IPC_RMID:
 553                 freeque(ns, msq);
 554                 break;
 555         }
 556         err = 0;
 557 out_up:
 558         mutex_unlock(&msg_ids(ns).mutex);
 559         return err;
 560 out_unlock_up:
 561         msg_unlock(msq);
 562         goto out_up;
 563 out_unlock:
 564         msg_unlock(msq);
 565         return err;
 566 }
 567
 568 static int testmsg(struct msg_msg *msg, long type, int mode)
 569 {
 570         switch(mode)
 571         {
 572                 case SEARCH_ANY:
 573                         return 1;
 574                 case SEARCH_LESSEQUAL:
 575                         if (msg->m_type <=type)
 576                                 return 1;
 577                         break;
 578                 case SEARCH_EQUAL:
 579                         if (msg->m_type == type)
 580                                 return 1;
 581                         break;
 582                 case SEARCH_NOTEQUAL:
 583                         if (msg->m_type != type)
 584                                 return 1;
 585                         break;
 586         }
 587         return 0;
 588 }
 589
 590 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 591 {
 592         struct list_head *tmp;
 593
 594         tmp = msq->q_receivers.next;
 595         while (tmp != &msq->q_receivers) {
 596                 struct msg_receiver *msr;
 597
 598                 msr = list_entry(tmp, struct msg_receiver, r_list);
 599                 tmp = tmp->next;
 600                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 601                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 602                                                msr->r_msgtype, msr->r_mode)) {
 603
 604                         list_del(&msr->r_list);
 605                         if (msr->r_maxsize < msg->m_ts) {
 606                                 msr->r_msg = NULL;
 607                                 wake_up_process(msr->r_tsk);
 608                                 smp_mb();
 609                                 msr->r_msg = ERR_PTR(-E2BIG);
 610                         } else {
 611                                 msr->r_msg = NULL;
 612                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 613                                 msq->q_rtime = get_seconds();
 614                                 wake_up_process(msr->r_tsk);
 615                                 smp_mb();
 616                                 msr->r_msg = msg;
 617
 618                                 return 1;
 619                         }
 620                 }
 621         }
 622         return 0;
 623 }
 624
 625 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 626                 size_t msgsz, int msgflg)
 627 {
 628         struct msg_queue *msq;
 629         struct msg_msg *msg;
 630         int err;
 631         struct ipc_namespace *ns;
 632
 633         ns = current->nsproxy->ipc_ns;
 634
 635         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 636                 return -EINVAL;
 637         if (mtype < 1)
 638                 return -EINVAL;
 639
 640         msg = load_msg(mtext, msgsz);
 641         if (IS_ERR(msg))
 642                 return PTR_ERR(msg);
 643
 644         msg->m_type = mtype;
 645         msg->m_ts = msgsz;
 646
 647         msq = msg_lock(ns, msqid);
 648         err = -EINVAL;
 649         if (msq == NULL)
 650                 goto out_free;
 651
 652         err= -EIDRM;
 653         if (msg_checkid(ns, msq, msqid))
 654                 goto out_unlock_free;
 655
 656         for (;;) {
 657                 struct msg_sender s;
 658
 659                 err = -EACCES;
 660                 if (ipcperms(&msq->q_perm, S_IWUGO))
 661                         goto out_unlock_free;
 662
 663                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 664                 if (err)
 665                         goto out_unlock_free;
 666
 667                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 668                                 1 + msq->q_qnum <= msq->q_qbytes) {
 669                         break;
 670                 }
 671
 672                 /* queue full, wait: */
 673                 if (msgflg & IPC_NOWAIT) {
 674                         err = -EAGAIN;
 675                         goto out_unlock_free;
 676                 }
 677                 ss_add(msq, &s);
 678                 ipc_rcu_getref(msq);
 679                 msg_unlock(msq);
 680                 schedule();
 681
 682                 ipc_lock_by_ptr(&msq->q_perm);
 683                 ipc_rcu_putref(msq);
 684                 if (msq->q_perm.deleted) {
 685                         err = -EIDRM;
 686                         goto out_unlock_free;
 687                 }
 688                 ss_del(&s);
 689
 690                 if (signal_pending(current)) {
 691                         err = -ERESTARTNOHAND;
 692                         goto out_unlock_free;
 693                 }
 694         }
 695
 696         msq->q_lspid = task_tgid_vnr(current);
 697         msq->q_stime = get_seconds();
 698
 699         if (!pipelined_send(msq, msg)) {
 700                 /* noone is waiting for this message, enqueue it */
 701                 list_add_tail(&msg->m_list, &msq->q_messages);
 702                 msq->q_cbytes += msgsz;
 703                 msq->q_qnum++;
 704                 atomic_add(msgsz, &msg_bytes);
 705                 atomic_inc(&msg_hdrs);
 706         }
 707
 708         err = 0;
 709         msg = NULL;
 710
 711 out_unlock_free:
 712         msg_unlock(msq);
 713 out_free:
 714         if (msg != NULL)
 715                 free_msg(msg);
 716         return err;
 717 }
 718
 719 asmlinkage long
 720 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 721 {
 722         long mtype;
 723
 724         if (get_user(mtype, &msgp->mtype))
 725                 return -EFAULT;
 726         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 727 }
 728
 729 static inline int convert_mode(long *msgtyp, int msgflg)
 730 {
 731         /*
 732          *  find message of correct type.
 733          *  msgtyp = 0 => get first.
 734          *  msgtyp > 0 => get first message of matching type.
 735          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 736          */
 737         if (*msgtyp == 0)
 738                 return SEARCH_ANY;
 739         if (*msgtyp < 0) {
 740                 *msgtyp = -*msgtyp;
 741                 return SEARCH_LESSEQUAL;
 742         }
 743         if (msgflg & MSG_EXCEPT)
 744                 return SEARCH_NOTEQUAL;
 745         return SEARCH_EQUAL;
 746 }
 747
 748 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
 749                 size_t msgsz, long msgtyp, int msgflg)
 750 {
 751         struct msg_queue *msq;
 752         struct msg_msg *msg;
 753         int mode;
 754         struct ipc_namespace *ns;
 755
 756         if (msqid < 0 || (long) msgsz < 0)
 757                 return -EINVAL;
 758         mode = convert_mode(&msgtyp, msgflg);
 759         ns = current->nsproxy->ipc_ns;
 760
 761         msq = msg_lock(ns, msqid);
 762         if (msq == NULL)
 763                 return -EINVAL;
 764
 765         msg = ERR_PTR(-EIDRM);
 766         if (msg_checkid(ns, msq, msqid))
 767                 goto out_unlock;
 768
 769         for (;;) {
 770                 struct msg_receiver msr_d;
 771                 struct list_head *tmp;
 772
 773                 msg = ERR_PTR(-EACCES);
 774                 if (ipcperms(&msq->q_perm, S_IRUGO))
 775                         goto out_unlock;
 776
 777                 msg = ERR_PTR(-EAGAIN);
 778                 tmp = msq->q_messages.next;
 779                 while (tmp != &msq->q_messages) {
 780                         struct msg_msg *walk_msg;
 781
 782                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 783                         if (testmsg(walk_msg, msgtyp, mode) &&
 784                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 785                                                        msgtyp, mode)) {
 786
 787                                 msg = walk_msg;
 788                                 if (mode == SEARCH_LESSEQUAL &&
 789                                                 walk_msg->m_type != 1) {
 790                                         msg = walk_msg;
 791                                         msgtyp = walk_msg->m_type - 1;
 792                                 } else {
 793                                         msg = walk_msg;
 794                                         break;
 795                                 }
 796                         }
 797                         tmp = tmp->next;
 798                 }
 799                 if (!IS_ERR(msg)) {
 800                         /*
 801                          * Found a suitable message.
 802                          * Unlink it from the queue.
 803                          */
 804                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 805                                 msg = ERR_PTR(-E2BIG);
 806                                 goto out_unlock;
 807                         }
 808                         list_del(&msg->m_list);
 809                         msq->q_qnum--;
 810                         msq->q_rtime = get_seconds();
 811                         msq->q_lrpid = task_tgid_vnr(current);
 812                         msq->q_cbytes -= msg->m_ts;
 813                         atomic_sub(msg->m_ts, &msg_bytes);
 814                         atomic_dec(&msg_hdrs);
 815                         ss_wakeup(&msq->q_senders, 0);
 816                         msg_unlock(msq);
 817                         break;
 818                 }
 819                 /* No message waiting. Wait for a message */
 820                 if (msgflg & IPC_NOWAIT) {
 821                         msg = ERR_PTR(-ENOMSG);
 822                         goto out_unlock;
 823                 }
 824                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 825                 msr_d.r_tsk = current;
 826                 msr_d.r_msgtype = msgtyp;
 827                 msr_d.r_mode = mode;
 828                 if (msgflg & MSG_NOERROR)
 829                         msr_d.r_maxsize = INT_MAX;
 830                 else
 831                         msr_d.r_maxsize = msgsz;
 832                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 833                 current->state = TASK_INTERRUPTIBLE;
 834                 msg_unlock(msq);
 835
 836                 schedule();
 837
 838                 /* Lockless receive, part 1:
 839                  * Disable preemption.  We don't hold a reference to the queue
 840                  * and getting a reference would defeat the idea of a lockless
 841                  * operation, thus the code relies on rcu to guarantee the
 842                  * existance of msq:
 843                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 844                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 845                  * rcu_read_lock() prevents preemption between reading r_msg
 846                  * and the spin_lock() inside ipc_lock_by_ptr().
 847                  */
 848                 rcu_read_lock();
 849
 850                 /* Lockless receive, part 2:
 851                  * Wait until pipelined_send or expunge_all are outside of
 852                  * wake_up_process(). There is a race with exit(), see
 853                  * ipc/mqueue.c for the details.
 854                  */
 855                 msg = (struct msg_msg*)msr_d.r_msg;
 856                 while (msg == NULL) {
 857                         cpu_relax();
 858                         msg = (struct msg_msg *)msr_d.r_msg;
 859                 }
 860
 861                 /* Lockless receive, part 3:
 862                  * If there is a message or an error then accept it without
 863                  * locking.
 864                  */
 865                 if (msg != ERR_PTR(-EAGAIN)) {
 866                         rcu_read_unlock();
 867                         break;
 868                 }
 869
 870                 /* Lockless receive, part 3:
 871                  * Acquire the queue spinlock.
 872                  */
 873                 ipc_lock_by_ptr(&msq->q_perm);
 874                 rcu_read_unlock();
 875
 876                 /* Lockless receive, part 4:
 877                  * Repeat test after acquiring the spinlock.
 878                  */
 879                 msg = (struct msg_msg*)msr_d.r_msg;
 880                 if (msg != ERR_PTR(-EAGAIN))
 881                         goto out_unlock;
 882
 883                 list_del(&msr_d.r_list);
 884                 if (signal_pending(current)) {
 885                         msg = ERR_PTR(-ERESTARTNOHAND);
 886 out_unlock:
 887                         msg_unlock(msq);
 888                         break;
 889                 }
 890         }
 891         if (IS_ERR(msg))
 892                 return PTR_ERR(msg);
 893
 894         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 895         *pmtype = msg->m_type;
 896         if (store_msg(mtext, msg, msgsz))
 897                 msgsz = -EFAULT;
 898
 899         free_msg(msg);
 900
 901         return msgsz;
 902 }
 903
 904 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 905                            long msgtyp, int msgflg)
 906 {
 907         long err, mtype;
 908
 909         err =  do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
 910         if (err < 0)
 911                 goto out;
 912
 913         if (put_user(mtype, &msgp->mtype))
 914                 err = -EFAULT;
 915 out:
 916         return err;
 917 }
 918
 919 #ifdef CONFIG_PROC_FS
 920 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 921 {
 922         struct msg_queue *msq = it;
 923
 924         return seq_printf(s,
 925                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 926                         msq->q_perm.key,
 927                         msq->q_perm.id,
 928                         msq->q_perm.mode,
 929                         msq->q_cbytes,
 930                         msq->q_qnum,
 931                         msq->q_lspid,
 932                         msq->q_lrpid,
 933                         msq->q_perm.uid,
 934                         msq->q_perm.gid,
 935                         msq->q_perm.cuid,
 936                         msq->q_perm.cgid,
 937                         msq->q_stime,
 938                         msq->q_rtime,
 939                         msq->q_ctime);
 940 }
 941 #endif