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/mm.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/list.h>
  33 #include <linux/security.h>
  34 #include <linux/sched.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/audit.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/rwsem.h>
  39 #include <linux/nsproxy.h>
  40 #include <linux/ipc_namespace.h>
  41
  42 #include <asm/current.h>
  43 #include <asm/uaccess.h>
  44 #include "util.h"
  45
  46 /*
  47  * one msg_receiver structure for each sleeping receiver:
  48  */
  49 struct msg_receiver {
  50         struct list_head        r_list;
  51         struct task_struct      *r_tsk;
  52
  53         int                     r_mode;
  54         long                    r_msgtype;
  55         long                    r_maxsize;
  56
  57         struct msg_msg          *volatile r_msg;
  58 };
  59
  60 /* one msg_sender for each sleeping sender */
  61 struct msg_sender {
  62         struct list_head        list;
  63         struct task_struct      *tsk;
  64 };
  65
  66 #define SEARCH_ANY              1
  67 #define SEARCH_EQUAL            2
  68 #define SEARCH_NOTEQUAL         3
  69 #define SEARCH_LESSEQUAL        4
  70
  71 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  72
  73 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  74
  75 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  76 static int newque(struct ipc_namespace *, struct ipc_params *);
  77 #ifdef CONFIG_PROC_FS
  78 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  79 #endif
  80
  81 /*
  82  * Scale msgmni with the available lowmem size: the memory dedicated to msg
  83  * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  84  * Also take into account the number of nsproxies created so far.
  85  * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  86  */
  87 static void recompute_msgmni(struct ipc_namespace *ns)
  88 {
  89         struct sysinfo i;
  90         unsigned long allowed;
  91         int nb_ns;
  92
  93         si_meminfo(&i);
  94         allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  95                 / MSGMNB;
  96         nb_ns = atomic_read(&nr_ipc_ns);
  97         allowed /= nb_ns;
  98
  99         if (allowed < MSGMNI) {
 100                 ns->msg_ctlmni = MSGMNI;
 101                 goto out_callback;
 102         }
 103
 104         if (allowed > IPCMNI / nb_ns) {
 105                 ns->msg_ctlmni = IPCMNI / nb_ns;
 106                 goto out_callback;
 107         }
 108
 109         ns->msg_ctlmni = allowed;
 110
 111 out_callback:
 112
 113         printk(KERN_INFO "msgmni has been set to %d for ipc namespace %p\n",
 114                 ns->msg_ctlmni, ns);
 115 }
 116
 117 void msg_init_ns(struct ipc_namespace *ns)
 118 {
 119         ns->msg_ctlmax = MSGMAX;
 120         ns->msg_ctlmnb = MSGMNB;
 121
 122         recompute_msgmni(ns);
 123
 124         atomic_set(&ns->msg_bytes, 0);
 125         atomic_set(&ns->msg_hdrs, 0);
 126         ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
 127 }
 128
 129 #ifdef CONFIG_IPC_NS
 130 void msg_exit_ns(struct ipc_namespace *ns)
 131 {
 132         free_ipcs(ns, &msg_ids(ns), freeque);
 133 }
 134 #endif
 135
 136 void __init msg_init(void)
 137 {
 138         msg_init_ns(&init_ipc_ns);
 139         ipc_init_proc_interface("sysvipc/msg",
 140                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 141                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 142 }
 143
 144 /*
 145  * This routine is called in the paths where the rw_mutex is held to protect
 146  * access to the idr tree.
 147  */
 148 static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
 149                                                 int id)
 150 {
 151         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
 152
 153         if (IS_ERR(ipcp))
 154                 return (struct msg_queue *)ipcp;
 155
 156         return container_of(ipcp, struct msg_queue, q_perm);
 157 }
 158
 159 /*
 160  * msg_lock_(check_) routines are called in the paths where the rw_mutex
 161  * is not held.
 162  */
 163 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
 164 {
 165         struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
 166
 167         if (IS_ERR(ipcp))
 168                 return (struct msg_queue *)ipcp;
 169
 170         return container_of(ipcp, struct msg_queue, q_perm);
 171 }
 172
 173 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
 174                                                 int id)
 175 {
 176         struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
 177
 178         if (IS_ERR(ipcp))
 179                 return (struct msg_queue *)ipcp;
 180
 181         return container_of(ipcp, struct msg_queue, q_perm);
 182 }
 183
 184 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 185 {
 186         ipc_rmid(&msg_ids(ns), &s->q_perm);
 187 }
 188
 189 /**
 190  * newque - Create a new msg queue
 191  * @ns: namespace
 192  * @params: ptr to the structure that contains the key and msgflg
 193  *
 194  * Called with msg_ids.rw_mutex held (writer)
 195  */
 196 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 197 {
 198         struct msg_queue *msq;
 199         int id, retval;
 200         key_t key = params->key;
 201         int msgflg = params->flg;
 202
 203         msq = ipc_rcu_alloc(sizeof(*msq));
 204         if (!msq)
 205                 return -ENOMEM;
 206
 207         msq->q_perm.mode = msgflg & S_IRWXUGO;
 208         msq->q_perm.key = key;
 209
 210         msq->q_perm.security = NULL;
 211         retval = security_msg_queue_alloc(msq);
 212         if (retval) {
 213                 ipc_rcu_putref(msq);
 214                 return retval;
 215         }
 216
 217         /*
 218          * ipc_addid() locks msq
 219          */
 220         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 221         if (id < 0) {
 222                 security_msg_queue_free(msq);
 223                 ipc_rcu_putref(msq);
 224                 return id;
 225         }
 226
 227         msq->q_stime = msq->q_rtime = 0;
 228         msq->q_ctime = get_seconds();
 229         msq->q_cbytes = msq->q_qnum = 0;
 230         msq->q_qbytes = ns->msg_ctlmnb;
 231         msq->q_lspid = msq->q_lrpid = 0;
 232         INIT_LIST_HEAD(&msq->q_messages);
 233         INIT_LIST_HEAD(&msq->q_receivers);
 234         INIT_LIST_HEAD(&msq->q_senders);
 235
 236         msg_unlock(msq);
 237
 238         return msq->q_perm.id;
 239 }
 240
 241 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
 242 {
 243         mss->tsk = current;
 244         current->state = TASK_INTERRUPTIBLE;
 245         list_add_tail(&mss->list, &msq->q_senders);
 246 }
 247
 248 static inline void ss_del(struct msg_sender *mss)
 249 {
 250         if (mss->list.next != NULL)
 251                 list_del(&mss->list);
 252 }
 253
 254 static void ss_wakeup(struct list_head *h, int kill)
 255 {
 256         struct list_head *tmp;
 257
 258         tmp = h->next;
 259         while (tmp != h) {
 260                 struct msg_sender *mss;
 261
 262                 mss = list_entry(tmp, struct msg_sender, list);
 263                 tmp = tmp->next;
 264                 if (kill)
 265                         mss->list.next = NULL;
 266                 wake_up_process(mss->tsk);
 267         }
 268 }
 269
 270 static void expunge_all(struct msg_queue *msq, int res)
 271 {
 272         struct list_head *tmp;
 273
 274         tmp = msq->q_receivers.next;
 275         while (tmp != &msq->q_receivers) {
 276                 struct msg_receiver *msr;
 277
 278                 msr = list_entry(tmp, struct msg_receiver, r_list);
 279                 tmp = tmp->next;
 280                 msr->r_msg = NULL;
 281                 wake_up_process(msr->r_tsk);
 282                 smp_mb();
 283                 msr->r_msg = ERR_PTR(res);
 284         }
 285 }
 286
 287 /*
 288  * freeque() wakes up waiters on the sender and receiver waiting queue,
 289  * removes the message queue from message queue ID IDR, and cleans up all the
 290  * messages associated with this queue.
 291  *
 292  * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
 293  * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
 294  */
 295 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 296 {
 297         struct list_head *tmp;
 298         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 299
 300         expunge_all(msq, -EIDRM);
 301         ss_wakeup(&msq->q_senders, 1);
 302         msg_rmid(ns, msq);
 303         msg_unlock(msq);
 304
 305         tmp = msq->q_messages.next;
 306         while (tmp != &msq->q_messages) {
 307                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 308
 309                 tmp = tmp->next;
 310                 atomic_dec(&ns->msg_hdrs);
 311                 free_msg(msg);
 312         }
 313         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 314         security_msg_queue_free(msq);
 315         ipc_rcu_putref(msq);
 316 }
 317
 318 /*
 319  * Called with msg_ids.rw_mutex and ipcp locked.
 320  */
 321 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 322 {
 323         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 324
 325         return security_msg_queue_associate(msq, msgflg);
 326 }
 327
 328 asmlinkage long sys_msgget(key_t key, int msgflg)
 329 {
 330         struct ipc_namespace *ns;
 331         struct ipc_ops msg_ops;
 332         struct ipc_params msg_params;
 333
 334         ns = current->nsproxy->ipc_ns;
 335
 336         msg_ops.getnew = newque;
 337         msg_ops.associate = msg_security;
 338         msg_ops.more_checks = NULL;
 339
 340         msg_params.key = key;
 341         msg_params.flg = msgflg;
 342
 343         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 344 }
 345
 346 static inline unsigned long
 347 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 348 {
 349         switch(version) {
 350         case IPC_64:
 351                 return copy_to_user(buf, in, sizeof(*in));
 352         case IPC_OLD:
 353         {
 354                 struct msqid_ds out;
 355
 356                 memset(&out, 0, sizeof(out));
 357
 358                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 359
 360                 out.msg_stime           = in->msg_stime;
 361                 out.msg_rtime           = in->msg_rtime;
 362                 out.msg_ctime           = in->msg_ctime;
 363
 364                 if (in->msg_cbytes > USHRT_MAX)
 365                         out.msg_cbytes  = USHRT_MAX;
 366                 else
 367                         out.msg_cbytes  = in->msg_cbytes;
 368                 out.msg_lcbytes         = in->msg_cbytes;
 369
 370                 if (in->msg_qnum > USHRT_MAX)
 371                         out.msg_qnum    = USHRT_MAX;
 372                 else
 373                         out.msg_qnum    = in->msg_qnum;
 374
 375                 if (in->msg_qbytes > USHRT_MAX)
 376                         out.msg_qbytes  = USHRT_MAX;
 377                 else
 378                         out.msg_qbytes  = in->msg_qbytes;
 379                 out.msg_lqbytes         = in->msg_qbytes;
 380
 381                 out.msg_lspid           = in->msg_lspid;
 382                 out.msg_lrpid           = in->msg_lrpid;
 383
 384                 return copy_to_user(buf, &out, sizeof(out));
 385         }
 386         default:
 387                 return -EINVAL;
 388         }
 389 }
 390
 391 struct msq_setbuf {
 392         unsigned long   qbytes;
 393         uid_t           uid;
 394         gid_t           gid;
 395         mode_t          mode;
 396 };
 397
 398 static inline unsigned long
 399 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 400 {
 401         switch(version) {
 402         case IPC_64:
 403         {
 404                 struct msqid64_ds tbuf;
 405
 406                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 407                         return -EFAULT;
 408
 409                 out->qbytes             = tbuf.msg_qbytes;
 410                 out->uid                = tbuf.msg_perm.uid;
 411                 out->gid                = tbuf.msg_perm.gid;
 412                 out->mode               = tbuf.msg_perm.mode;
 413
 414                 return 0;
 415         }
 416         case IPC_OLD:
 417         {
 418                 struct msqid_ds tbuf_old;
 419
 420                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 421                         return -EFAULT;
 422
 423                 out->uid                = tbuf_old.msg_perm.uid;
 424                 out->gid                = tbuf_old.msg_perm.gid;
 425                 out->mode               = tbuf_old.msg_perm.mode;
 426
 427                 if (tbuf_old.msg_qbytes == 0)
 428                         out->qbytes     = tbuf_old.msg_lqbytes;
 429                 else
 430                         out->qbytes     = tbuf_old.msg_qbytes;
 431
 432                 return 0;
 433         }
 434         default:
 435                 return -EINVAL;
 436         }
 437 }
 438
 439 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 440 {
 441         struct kern_ipc_perm *ipcp;
 442         struct msq_setbuf uninitialized_var(setbuf);
 443         struct msg_queue *msq;
 444         int err, version;
 445         struct ipc_namespace *ns;
 446
 447         if (msqid < 0 || cmd < 0)
 448                 return -EINVAL;
 449
 450         version = ipc_parse_version(&cmd);
 451         ns = current->nsproxy->ipc_ns;
 452
 453         switch (cmd) {
 454         case IPC_INFO:
 455         case MSG_INFO:
 456         {
 457                 struct msginfo msginfo;
 458                 int max_id;
 459
 460                 if (!buf)
 461                         return -EFAULT;
 462                 /*
 463                  * We must not return kernel stack data.
 464                  * due to padding, it's not enough
 465                  * to set all member fields.
 466                  */
 467                 err = security_msg_queue_msgctl(NULL, cmd);
 468                 if (err)
 469                         return err;
 470
 471                 memset(&msginfo, 0, sizeof(msginfo));
 472                 msginfo.msgmni = ns->msg_ctlmni;
 473                 msginfo.msgmax = ns->msg_ctlmax;
 474                 msginfo.msgmnb = ns->msg_ctlmnb;
 475                 msginfo.msgssz = MSGSSZ;
 476                 msginfo.msgseg = MSGSEG;
 477                 down_read(&msg_ids(ns).rw_mutex);
 478                 if (cmd == MSG_INFO) {
 479                         msginfo.msgpool = msg_ids(ns).in_use;
 480                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 481                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 482                 } else {
 483                         msginfo.msgmap = MSGMAP;
 484                         msginfo.msgpool = MSGPOOL;
 485                         msginfo.msgtql = MSGTQL;
 486                 }
 487                 max_id = ipc_get_maxid(&msg_ids(ns));
 488                 up_read(&msg_ids(ns).rw_mutex);
 489                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 490                         return -EFAULT;
 491                 return (max_id < 0) ? 0 : max_id;
 492         }
 493         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 494         case IPC_STAT:
 495         {
 496                 struct msqid64_ds tbuf;
 497                 int success_return;
 498
 499                 if (!buf)
 500                         return -EFAULT;
 501
 502                 if (cmd == MSG_STAT) {
 503                         msq = msg_lock(ns, msqid);
 504                         if (IS_ERR(msq))
 505                                 return PTR_ERR(msq);
 506                         success_return = msq->q_perm.id;
 507                 } else {
 508                         msq = msg_lock_check(ns, msqid);
 509                         if (IS_ERR(msq))
 510                                 return PTR_ERR(msq);
 511                         success_return = 0;
 512                 }
 513                 err = -EACCES;
 514                 if (ipcperms(&msq->q_perm, S_IRUGO))
 515                         goto out_unlock;
 516
 517                 err = security_msg_queue_msgctl(msq, cmd);
 518                 if (err)
 519                         goto out_unlock;
 520
 521                 memset(&tbuf, 0, sizeof(tbuf));
 522
 523                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 524                 tbuf.msg_stime  = msq->q_stime;
 525                 tbuf.msg_rtime  = msq->q_rtime;
 526                 tbuf.msg_ctime  = msq->q_ctime;
 527                 tbuf.msg_cbytes = msq->q_cbytes;
 528                 tbuf.msg_qnum   = msq->q_qnum;
 529                 tbuf.msg_qbytes = msq->q_qbytes;
 530                 tbuf.msg_lspid  = msq->q_lspid;
 531                 tbuf.msg_lrpid  = msq->q_lrpid;
 532                 msg_unlock(msq);
 533                 if (copy_msqid_to_user(buf, &tbuf, version))
 534                         return -EFAULT;
 535                 return success_return;
 536         }
 537         case IPC_SET:
 538                 if (!buf)
 539                         return -EFAULT;
 540                 if (copy_msqid_from_user(&setbuf, buf, version))
 541                         return -EFAULT;
 542                 break;
 543         case IPC_RMID:
 544                 break;
 545         default:
 546                 return  -EINVAL;
 547         }
 548
 549         down_write(&msg_ids(ns).rw_mutex);
 550         msq = msg_lock_check_down(ns, msqid);
 551         if (IS_ERR(msq)) {
 552                 err = PTR_ERR(msq);
 553                 goto out_up;
 554         }
 555
 556         ipcp = &msq->q_perm;
 557
 558         err = audit_ipc_obj(ipcp);
 559         if (err)
 560                 goto out_unlock_up;
 561         if (cmd == IPC_SET) {
 562                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
 563                                          setbuf.mode);
 564                 if (err)
 565                         goto out_unlock_up;
 566         }
 567
 568         err = -EPERM;
 569         if (current->euid != ipcp->cuid &&
 570             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 571                 /* We _could_ check for CAP_CHOWN above, but we don't */
 572                 goto out_unlock_up;
 573
 574         err = security_msg_queue_msgctl(msq, cmd);
 575         if (err)
 576                 goto out_unlock_up;
 577
 578         switch (cmd) {
 579         case IPC_SET:
 580         {
 581                 err = -EPERM;
 582                 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 583                         goto out_unlock_up;
 584
 585                 msq->q_qbytes = setbuf.qbytes;
 586
 587                 ipcp->uid = setbuf.uid;
 588                 ipcp->gid = setbuf.gid;
 589                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 590                              (S_IRWXUGO & setbuf.mode);
 591                 msq->q_ctime = get_seconds();
 592                 /* sleeping receivers might be excluded by
 593                  * stricter permissions.
 594                  */
 595                 expunge_all(msq, -EAGAIN);
 596                 /* sleeping senders might be able to send
 597                  * due to a larger queue size.
 598                  */
 599                 ss_wakeup(&msq->q_senders, 0);
 600                 msg_unlock(msq);
 601                 break;
 602         }
 603         case IPC_RMID:
 604                 freeque(ns, &msq->q_perm);
 605                 break;
 606         }
 607         err = 0;
 608 out_up:
 609         up_write(&msg_ids(ns).rw_mutex);
 610         return err;
 611 out_unlock_up:
 612         msg_unlock(msq);
 613         goto out_up;
 614 out_unlock:
 615         msg_unlock(msq);
 616         return err;
 617 }
 618
 619 static int testmsg(struct msg_msg *msg, long type, int mode)
 620 {
 621         switch(mode)
 622         {
 623                 case SEARCH_ANY:
 624                         return 1;
 625                 case SEARCH_LESSEQUAL:
 626                         if (msg->m_type <=type)
 627                                 return 1;
 628                         break;
 629                 case SEARCH_EQUAL:
 630                         if (msg->m_type == type)
 631                                 return 1;
 632                         break;
 633                 case SEARCH_NOTEQUAL:
 634                         if (msg->m_type != type)
 635                                 return 1;
 636                         break;
 637         }
 638         return 0;
 639 }
 640
 641 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 642 {
 643         struct list_head *tmp;
 644
 645         tmp = msq->q_receivers.next;
 646         while (tmp != &msq->q_receivers) {
 647                 struct msg_receiver *msr;
 648
 649                 msr = list_entry(tmp, struct msg_receiver, r_list);
 650                 tmp = tmp->next;
 651                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 652                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 653                                                msr->r_msgtype, msr->r_mode)) {
 654
 655                         list_del(&msr->r_list);
 656                         if (msr->r_maxsize < msg->m_ts) {
 657                                 msr->r_msg = NULL;
 658                                 wake_up_process(msr->r_tsk);
 659                                 smp_mb();
 660                                 msr->r_msg = ERR_PTR(-E2BIG);
 661                         } else {
 662                                 msr->r_msg = NULL;
 663                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 664                                 msq->q_rtime = get_seconds();
 665                                 wake_up_process(msr->r_tsk);
 666                                 smp_mb();
 667                                 msr->r_msg = msg;
 668
 669                                 return 1;
 670                         }
 671                 }
 672         }
 673         return 0;
 674 }
 675
 676 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 677                 size_t msgsz, int msgflg)
 678 {
 679         struct msg_queue *msq;
 680         struct msg_msg *msg;
 681         int err;
 682         struct ipc_namespace *ns;
 683
 684         ns = current->nsproxy->ipc_ns;
 685
 686         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 687                 return -EINVAL;
 688         if (mtype < 1)
 689                 return -EINVAL;
 690
 691         msg = load_msg(mtext, msgsz);
 692         if (IS_ERR(msg))
 693                 return PTR_ERR(msg);
 694
 695         msg->m_type = mtype;
 696         msg->m_ts = msgsz;
 697
 698         msq = msg_lock_check(ns, msqid);
 699         if (IS_ERR(msq)) {
 700                 err = PTR_ERR(msq);
 701                 goto out_free;
 702         }
 703
 704         for (;;) {
 705                 struct msg_sender s;
 706
 707                 err = -EACCES;
 708                 if (ipcperms(&msq->q_perm, S_IWUGO))
 709                         goto out_unlock_free;
 710
 711                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 712                 if (err)
 713                         goto out_unlock_free;
 714
 715                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 716                                 1 + msq->q_qnum <= msq->q_qbytes) {
 717                         break;
 718                 }
 719
 720                 /* queue full, wait: */
 721                 if (msgflg & IPC_NOWAIT) {
 722                         err = -EAGAIN;
 723                         goto out_unlock_free;
 724                 }
 725                 ss_add(msq, &s);
 726                 ipc_rcu_getref(msq);
 727                 msg_unlock(msq);
 728                 schedule();
 729
 730                 ipc_lock_by_ptr(&msq->q_perm);
 731                 ipc_rcu_putref(msq);
 732                 if (msq->q_perm.deleted) {
 733                         err = -EIDRM;
 734                         goto out_unlock_free;
 735                 }
 736                 ss_del(&s);
 737
 738                 if (signal_pending(current)) {
 739                         err = -ERESTARTNOHAND;
 740                         goto out_unlock_free;
 741                 }
 742         }
 743
 744         msq->q_lspid = task_tgid_vnr(current);
 745         msq->q_stime = get_seconds();
 746
 747         if (!pipelined_send(msq, msg)) {
 748                 /* noone is waiting for this message, enqueue it */
 749                 list_add_tail(&msg->m_list, &msq->q_messages);
 750                 msq->q_cbytes += msgsz;
 751                 msq->q_qnum++;
 752                 atomic_add(msgsz, &ns->msg_bytes);
 753                 atomic_inc(&ns->msg_hdrs);
 754         }
 755
 756         err = 0;
 757         msg = NULL;
 758
 759 out_unlock_free:
 760         msg_unlock(msq);
 761 out_free:
 762         if (msg != NULL)
 763                 free_msg(msg);
 764         return err;
 765 }
 766
 767 asmlinkage long
 768 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 769 {
 770         long mtype;
 771
 772         if (get_user(mtype, &msgp->mtype))
 773                 return -EFAULT;
 774         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 775 }
 776
 777 static inline int convert_mode(long *msgtyp, int msgflg)
 778 {
 779         /*
 780          *  find message of correct type.
 781          *  msgtyp = 0 => get first.
 782          *  msgtyp > 0 => get first message of matching type.
 783          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 784          */
 785         if (*msgtyp == 0)
 786                 return SEARCH_ANY;
 787         if (*msgtyp < 0) {
 788                 *msgtyp = -*msgtyp;
 789                 return SEARCH_LESSEQUAL;
 790         }
 791         if (msgflg & MSG_EXCEPT)
 792                 return SEARCH_NOTEQUAL;
 793         return SEARCH_EQUAL;
 794 }
 795
 796 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
 797                 size_t msgsz, long msgtyp, int msgflg)
 798 {
 799         struct msg_queue *msq;
 800         struct msg_msg *msg;
 801         int mode;
 802         struct ipc_namespace *ns;
 803
 804         if (msqid < 0 || (long) msgsz < 0)
 805                 return -EINVAL;
 806         mode = convert_mode(&msgtyp, msgflg);
 807         ns = current->nsproxy->ipc_ns;
 808
 809         msq = msg_lock_check(ns, msqid);
 810         if (IS_ERR(msq))
 811                 return PTR_ERR(msq);
 812
 813         for (;;) {
 814                 struct msg_receiver msr_d;
 815                 struct list_head *tmp;
 816
 817                 msg = ERR_PTR(-EACCES);
 818                 if (ipcperms(&msq->q_perm, S_IRUGO))
 819                         goto out_unlock;
 820
 821                 msg = ERR_PTR(-EAGAIN);
 822                 tmp = msq->q_messages.next;
 823                 while (tmp != &msq->q_messages) {
 824                         struct msg_msg *walk_msg;
 825
 826                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 827                         if (testmsg(walk_msg, msgtyp, mode) &&
 828                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 829                                                        msgtyp, mode)) {
 830
 831                                 msg = walk_msg;
 832                                 if (mode == SEARCH_LESSEQUAL &&
 833                                                 walk_msg->m_type != 1) {
 834                                         msg = walk_msg;
 835                                         msgtyp = walk_msg->m_type - 1;
 836                                 } else {
 837                                         msg = walk_msg;
 838                                         break;
 839                                 }
 840                         }
 841                         tmp = tmp->next;
 842                 }
 843                 if (!IS_ERR(msg)) {
 844                         /*
 845                          * Found a suitable message.
 846                          * Unlink it from the queue.
 847                          */
 848                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 849                                 msg = ERR_PTR(-E2BIG);
 850                                 goto out_unlock;
 851                         }
 852                         list_del(&msg->m_list);
 853                         msq->q_qnum--;
 854                         msq->q_rtime = get_seconds();
 855                         msq->q_lrpid = task_tgid_vnr(current);
 856                         msq->q_cbytes -= msg->m_ts;
 857                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 858                         atomic_dec(&ns->msg_hdrs);
 859                         ss_wakeup(&msq->q_senders, 0);
 860                         msg_unlock(msq);
 861                         break;
 862                 }
 863                 /* No message waiting. Wait for a message */
 864                 if (msgflg & IPC_NOWAIT) {
 865                         msg = ERR_PTR(-ENOMSG);
 866                         goto out_unlock;
 867                 }
 868                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 869                 msr_d.r_tsk = current;
 870                 msr_d.r_msgtype = msgtyp;
 871                 msr_d.r_mode = mode;
 872                 if (msgflg & MSG_NOERROR)
 873                         msr_d.r_maxsize = INT_MAX;
 874                 else
 875                         msr_d.r_maxsize = msgsz;
 876                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 877                 current->state = TASK_INTERRUPTIBLE;
 878                 msg_unlock(msq);
 879
 880                 schedule();
 881
 882                 /* Lockless receive, part 1:
 883                  * Disable preemption.  We don't hold a reference to the queue
 884                  * and getting a reference would defeat the idea of a lockless
 885                  * operation, thus the code relies on rcu to guarantee the
 886                  * existance of msq:
 887                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 888                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 889                  * rcu_read_lock() prevents preemption between reading r_msg
 890                  * and the spin_lock() inside ipc_lock_by_ptr().
 891                  */
 892                 rcu_read_lock();
 893
 894                 /* Lockless receive, part 2:
 895                  * Wait until pipelined_send or expunge_all are outside of
 896                  * wake_up_process(). There is a race with exit(), see
 897                  * ipc/mqueue.c for the details.
 898                  */
 899                 msg = (struct msg_msg*)msr_d.r_msg;
 900                 while (msg == NULL) {
 901                         cpu_relax();
 902                         msg = (struct msg_msg *)msr_d.r_msg;
 903                 }
 904
 905                 /* Lockless receive, part 3:
 906                  * If there is a message or an error then accept it without
 907                  * locking.
 908                  */
 909                 if (msg != ERR_PTR(-EAGAIN)) {
 910                         rcu_read_unlock();
 911                         break;
 912                 }
 913
 914                 /* Lockless receive, part 3:
 915                  * Acquire the queue spinlock.
 916                  */
 917                 ipc_lock_by_ptr(&msq->q_perm);
 918                 rcu_read_unlock();
 919
 920                 /* Lockless receive, part 4:
 921                  * Repeat test after acquiring the spinlock.
 922                  */
 923                 msg = (struct msg_msg*)msr_d.r_msg;
 924                 if (msg != ERR_PTR(-EAGAIN))
 925                         goto out_unlock;
 926
 927                 list_del(&msr_d.r_list);
 928                 if (signal_pending(current)) {
 929                         msg = ERR_PTR(-ERESTARTNOHAND);
 930 out_unlock:
 931                         msg_unlock(msq);
 932                         break;
 933                 }
 934         }
 935         if (IS_ERR(msg))
 936                 return PTR_ERR(msg);
 937
 938         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 939         *pmtype = msg->m_type;
 940         if (store_msg(mtext, msg, msgsz))
 941                 msgsz = -EFAULT;
 942
 943         free_msg(msg);
 944
 945         return msgsz;
 946 }
 947
 948 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 949                            long msgtyp, int msgflg)
 950 {
 951         long err, mtype;
 952
 953         err =  do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
 954         if (err < 0)
 955                 goto out;
 956
 957         if (put_user(mtype, &msgp->mtype))
 958                 err = -EFAULT;
 959 out:
 960         return err;
 961 }
 962
 963 #ifdef CONFIG_PROC_FS
 964 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 965 {
 966         struct msg_queue *msq = it;
 967
 968         return seq_printf(s,
 969                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 970                         msq->q_perm.key,
 971                         msq->q_perm.id,
 972                         msq->q_perm.mode,
 973                         msq->q_cbytes,
 974                         msq->q_qnum,
 975                         msq->q_lspid,
 976                         msq->q_lrpid,
 977                         msq->q_perm.uid,
 978                         msq->q_perm.gid,
 979                         msq->q_perm.cuid,
 980                         msq->q_perm.cgid,
 981                         msq->q_stime,
 982                         msq->q_rtime,
 983                         msq->q_ctime);
 984 }
 985 #endif