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