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