ipc/msg.c

   1 /*
   2  * linux/ipc/msg.c
   3  * Copyright (C) 1992 Krishna Balasubramanian
   4  *
   5  * Removed all the remaining kerneld mess
   6  * Catch the -EFAULT stuff properly
   7  * Use GFP_KERNEL for messages as in 1.2
   8  * Fixed up the unchecked user space derefs
   9  * Copyright (C) 1998 Alan Cox & Andi Kleen
  10  *
  11  * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  12  *
  13  * mostly rewritten, threaded and wake-one semantics added
  14  * MSGMAX limit removed, sysctl's added
  15  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  */
  24
  25 #include <linux/capability.h>
  26 #include <linux/slab.h>
  27 #include <linux/msg.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/init.h>
  30 #include <linux/proc_fs.h>
  31 #include <linux/list.h>
  32 #include <linux/security.h>
  33 #include <linux/sched.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/audit.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/rwsem.h>
  38 #include <linux/nsproxy.h>
  39 #include <linux/ipc_namespace.h>
  40
  41 #include <asm/current.h>
  42 #include <asm/uaccess.h>
  43 #include "util.h"
  44
  45 /*
  46  * one msg_receiver structure for each sleeping receiver:
  47  */
  48 struct msg_receiver {
  49         struct list_head        r_list;
  50         struct task_struct      *r_tsk;
  51
  52         int                     r_mode;
  53         long                    r_msgtype;
  54         long                    r_maxsize;
  55
  56         struct msg_msg          *volatile r_msg;
  57 };
  58
  59 /* one msg_sender for each sleeping sender */
  60 struct msg_sender {
  61         struct list_head        list;
  62         struct task_struct      *tsk;
  63 };
  64
  65 #define SEARCH_ANY              1
  66 #define SEARCH_EQUAL            2
  67 #define SEARCH_NOTEQUAL         3
  68 #define SEARCH_LESSEQUAL        4
  69
  70 static struct ipc_ids init_msg_ids;
  71
  72 #define msg_ids(ns)     (*((ns)->ids[IPC_MSG_IDS]))
  73
  74 #define msg_unlock(msq)         ipc_unlock(&(msq)->q_perm)
  75 #define msg_buildid(id, seq)    ipc_buildid(id, seq)
  76
  77 static void freeque(struct ipc_namespace *, struct msg_queue *);
  78 static int newque(struct ipc_namespace *, struct ipc_params *);
  79 #ifdef CONFIG_PROC_FS
  80 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  81 #endif
  82
  83 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  84 {
  85         ns->ids[IPC_MSG_IDS] = ids;
  86         ns->msg_ctlmax = MSGMAX;
  87         ns->msg_ctlmnb = MSGMNB;
  88         ns->msg_ctlmni = MSGMNI;
  89         atomic_set(&ns->msg_bytes, 0);
  90         atomic_set(&ns->msg_hdrs, 0);
  91         ipc_init_ids(ids);
  92 }
  93
  94 #ifdef CONFIG_IPC_NS
  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         struct kern_ipc_perm *perm;
 111         int next_id;
 112         int total, in_use;
 113
 114         down_write(&msg_ids(ns).rw_mutex);
 115
 116         in_use = msg_ids(ns).in_use;
 117
 118         for (total = 0, next_id = 0; total < in_use; next_id++) {
 119                 perm = idr_find(&msg_ids(ns).ipcs_idr, next_id);
 120                 if (perm == NULL)
 121                         continue;
 122                 ipc_lock_by_ptr(perm);
 123                 msq = container_of(perm, struct msg_queue, q_perm);
 124                 freeque(ns, msq);
 125                 total++;
 126         }
 127
 128         up_write(&msg_ids(ns).rw_mutex);
 129
 130         kfree(ns->ids[IPC_MSG_IDS]);
 131         ns->ids[IPC_MSG_IDS] = NULL;
 132 }
 133 #endif
 134
 135 void __init msg_init(void)
 136 {
 137         __msg_init_ns(&init_ipc_ns, &init_msg_ids);
 138         ipc_init_proc_interface("sysvipc/msg",
 139                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
 140                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
 141 }
 142
 143 /*
 144  * This routine is called in the paths where the rw_mutex is held to protect
 145  * access to the idr tree.
 146  */
 147 static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
 148                                                 int id)
 149 {
 150         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
 151
 152         if (IS_ERR(ipcp))
 153                 return (struct msg_queue *)ipcp;
 154
 155         return container_of(ipcp, struct msg_queue, q_perm);
 156 }
 157
 158 /*
 159  * msg_lock_(check_) routines are called in the paths where the rw_mutex
 160  * is not held.
 161  */
 162 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
 163 {
 164         struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
 165
 166         if (IS_ERR(ipcp))
 167                 return (struct msg_queue *)ipcp;
 168
 169         return container_of(ipcp, struct msg_queue, q_perm);
 170 }
 171
 172 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
 173                                                 int id)
 174 {
 175         struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
 176
 177         if (IS_ERR(ipcp))
 178                 return (struct msg_queue *)ipcp;
 179
 180         return container_of(ipcp, struct msg_queue, q_perm);
 181 }
 182
 183 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 184 {
 185         ipc_rmid(&msg_ids(ns), &s->q_perm);
 186 }
 187
 188 /**
 189  * newque - Create a new msg queue
 190  * @ns: namespace
 191  * @params: ptr to the structure that contains the key and msgflg
 192  *
 193  * Called with msg_ids.rw_mutex held (writer)
 194  */
 195 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 196 {
 197         struct msg_queue *msq;
 198         int id, retval;
 199         key_t key = params->key;
 200         int msgflg = params->flg;
 201
 202         msq = ipc_rcu_alloc(sizeof(*msq));
 203         if (!msq)
 204                 return -ENOMEM;
 205
 206         msq->q_perm.mode = msgflg & S_IRWXUGO;
 207         msq->q_perm.key = key;
 208
 209         msq->q_perm.security = NULL;
 210         retval = security_msg_queue_alloc(msq);
 211         if (retval) {
 212                 ipc_rcu_putref(msq);
 213                 return retval;
 214         }
 215
 216         /*
 217          * ipc_addid() locks msq
 218          */
 219         id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 220         if (id < 0) {
 221                 security_msg_queue_free(msq);
 222                 ipc_rcu_putref(msq);
 223                 return id;
 224         }
 225
 226         msq->q_perm.id = msg_buildid(id, msq->q_perm.seq);
 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 msg_queue *msq)
 296 {
 297         struct list_head *tmp;
 298
 299         expunge_all(msq, -EIDRM);
 300         ss_wakeup(&msq->q_senders, 1);
 301         msg_rmid(ns, msq);
 302         msg_unlock(msq);
 303
 304         tmp = msq->q_messages.next;
 305         while (tmp != &msq->q_messages) {
 306                 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
 307
 308                 tmp = tmp->next;
 309                 atomic_dec(&ns->msg_hdrs);
 310                 free_msg(msg);
 311         }
 312         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 313         security_msg_queue_free(msq);
 314         ipc_rcu_putref(msq);
 315 }
 316
 317 /*
 318  * Called with msg_ids.rw_mutex and ipcp locked.
 319  */
 320 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
 321 {
 322         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 323
 324         return security_msg_queue_associate(msq, msgflg);
 325 }
 326
 327 asmlinkage long sys_msgget(key_t key, int msgflg)
 328 {
 329         struct ipc_namespace *ns;
 330         struct ipc_ops msg_ops;
 331         struct ipc_params msg_params;
 332
 333         ns = current->nsproxy->ipc_ns;
 334
 335         msg_ops.getnew = newque;
 336         msg_ops.associate = msg_security;
 337         msg_ops.more_checks = NULL;
 338
 339         msg_params.key = key;
 340         msg_params.flg = msgflg;
 341
 342         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 343 }
 344
 345 static inline unsigned long
 346 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 347 {
 348         switch(version) {
 349         case IPC_64:
 350                 return copy_to_user(buf, in, sizeof(*in));
 351         case IPC_OLD:
 352         {
 353                 struct msqid_ds out;
 354
 355                 memset(&out, 0, sizeof(out));
 356
 357                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 358
 359                 out.msg_stime           = in->msg_stime;
 360                 out.msg_rtime           = in->msg_rtime;
 361                 out.msg_ctime           = in->msg_ctime;
 362
 363                 if (in->msg_cbytes > USHRT_MAX)
 364                         out.msg_cbytes  = USHRT_MAX;
 365                 else
 366                         out.msg_cbytes  = in->msg_cbytes;
 367                 out.msg_lcbytes         = in->msg_cbytes;
 368
 369                 if (in->msg_qnum > USHRT_MAX)
 370                         out.msg_qnum    = USHRT_MAX;
 371                 else
 372                         out.msg_qnum    = in->msg_qnum;
 373
 374                 if (in->msg_qbytes > USHRT_MAX)
 375                         out.msg_qbytes  = USHRT_MAX;
 376                 else
 377                         out.msg_qbytes  = in->msg_qbytes;
 378                 out.msg_lqbytes         = in->msg_qbytes;
 379
 380                 out.msg_lspid           = in->msg_lspid;
 381                 out.msg_lrpid           = in->msg_lrpid;
 382
 383                 return copy_to_user(buf, &out, sizeof(out));
 384         }
 385         default:
 386                 return -EINVAL;
 387         }
 388 }
 389
 390 struct msq_setbuf {
 391         unsigned long   qbytes;
 392         uid_t           uid;
 393         gid_t           gid;
 394         mode_t          mode;
 395 };
 396
 397 static inline unsigned long
 398 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
 399 {
 400         switch(version) {
 401         case IPC_64:
 402         {
 403                 struct msqid64_ds tbuf;
 404
 405                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 406                         return -EFAULT;
 407
 408                 out->qbytes             = tbuf.msg_qbytes;
 409                 out->uid                = tbuf.msg_perm.uid;
 410                 out->gid                = tbuf.msg_perm.gid;
 411                 out->mode               = tbuf.msg_perm.mode;
 412
 413                 return 0;
 414         }
 415         case IPC_OLD:
 416         {
 417                 struct msqid_ds tbuf_old;
 418
 419                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 420                         return -EFAULT;
 421
 422                 out->uid                = tbuf_old.msg_perm.uid;
 423                 out->gid                = tbuf_old.msg_perm.gid;
 424                 out->mode               = tbuf_old.msg_perm.mode;
 425
 426                 if (tbuf_old.msg_qbytes == 0)
 427                         out->qbytes     = tbuf_old.msg_lqbytes;
 428                 else
 429                         out->qbytes     = tbuf_old.msg_qbytes;
 430
 431                 return 0;
 432         }
 433         default:
 434                 return -EINVAL;
 435         }
 436 }
 437
 438 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 439 {
 440         struct kern_ipc_perm *ipcp;
 441         struct msq_setbuf uninitialized_var(setbuf);
 442         struct msg_queue *msq;
 443         int err, version;
 444         struct ipc_namespace *ns;
 445
 446         if (msqid < 0 || cmd < 0)
 447                 return -EINVAL;
 448
 449         version = ipc_parse_version(&cmd);
 450         ns = current->nsproxy->ipc_ns;
 451
 452         switch (cmd) {
 453         case IPC_INFO:
 454         case MSG_INFO:
 455         {
 456                 struct msginfo msginfo;
 457                 int max_id;
 458
 459                 if (!buf)
 460                         return -EFAULT;
 461                 /*
 462                  * We must not return kernel stack data.
 463                  * due to padding, it's not enough
 464                  * to set all member fields.
 465                  */
 466                 err = security_msg_queue_msgctl(NULL, cmd);
 467                 if (err)
 468                         return err;
 469
 470                 memset(&msginfo, 0, sizeof(msginfo));
 471                 msginfo.msgmni = ns->msg_ctlmni;
 472                 msginfo.msgmax = ns->msg_ctlmax;
 473                 msginfo.msgmnb = ns->msg_ctlmnb;
 474                 msginfo.msgssz = MSGSSZ;
 475                 msginfo.msgseg = MSGSEG;
 476                 down_read(&msg_ids(ns).rw_mutex);
 477                 if (cmd == MSG_INFO) {
 478                         msginfo.msgpool = msg_ids(ns).in_use;
 479                         msginfo.msgmap = atomic_read(&ns->msg_hdrs);
 480                         msginfo.msgtql = atomic_read(&ns->msg_bytes);
 481                 } else {
 482                         msginfo.msgmap = MSGMAP;
 483                         msginfo.msgpool = MSGPOOL;
 484                         msginfo.msgtql = MSGTQL;
 485                 }
 486                 max_id = ipc_get_maxid(&msg_ids(ns));
 487                 up_read(&msg_ids(ns).rw_mutex);
 488                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 489                         return -EFAULT;
 490                 return (max_id < 0) ? 0 : max_id;
 491         }
 492         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 493         case IPC_STAT:
 494         {
 495                 struct msqid64_ds tbuf;
 496                 int success_return;
 497
 498                 if (!buf)
 499                         return -EFAULT;
 500
 501                 if (cmd == MSG_STAT) {
 502                         msq = msg_lock(ns, msqid);
 503                         if (IS_ERR(msq))
 504                                 return PTR_ERR(msq);
 505                         success_return = msq->q_perm.id;
 506                 } else {
 507                         msq = msg_lock_check(ns, msqid);
 508                         if (IS_ERR(msq))
 509                                 return PTR_ERR(msq);
 510                         success_return = 0;
 511                 }
 512                 err = -EACCES;
 513                 if (ipcperms(&msq->q_perm, S_IRUGO))
 514                         goto out_unlock;
 515
 516                 err = security_msg_queue_msgctl(msq, cmd);
 517                 if (err)
 518                         goto out_unlock;
 519
 520                 memset(&tbuf, 0, sizeof(tbuf));
 521
 522                 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
 523                 tbuf.msg_stime  = msq->q_stime;
 524                 tbuf.msg_rtime  = msq->q_rtime;
 525                 tbuf.msg_ctime  = msq->q_ctime;
 526                 tbuf.msg_cbytes = msq->q_cbytes;
 527                 tbuf.msg_qnum   = msq->q_qnum;
 528                 tbuf.msg_qbytes = msq->q_qbytes;
 529                 tbuf.msg_lspid  = msq->q_lspid;
 530                 tbuf.msg_lrpid  = msq->q_lrpid;
 531                 msg_unlock(msq);
 532                 if (copy_msqid_to_user(buf, &tbuf, version))
 533                         return -EFAULT;
 534                 return success_return;
 535         }
 536         case IPC_SET:
 537                 if (!buf)
 538                         return -EFAULT;
 539                 if (copy_msqid_from_user(&setbuf, buf, version))
 540                         return -EFAULT;
 541                 break;
 542         case IPC_RMID:
 543                 break;
 544         default:
 545                 return  -EINVAL;
 546         }
 547
 548         down_write(&msg_ids(ns).rw_mutex);
 549         msq = msg_lock_check_down(ns, msqid);
 550         if (IS_ERR(msq)) {
 551                 err = PTR_ERR(msq);
 552                 goto out_up;
 553         }
 554
 555         ipcp = &msq->q_perm;
 556
 557         err = audit_ipc_obj(ipcp);
 558         if (err)
 559                 goto out_unlock_up;
 560         if (cmd == IPC_SET) {
 561                 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
 562                                          setbuf.mode);
 563                 if (err)
 564                         goto out_unlock_up;
 565         }
 566
 567         err = -EPERM;
 568         if (current->euid != ipcp->cuid &&
 569             current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
 570                 /* We _could_ check for CAP_CHOWN above, but we don't */
 571                 goto out_unlock_up;
 572
 573         err = security_msg_queue_msgctl(msq, cmd);
 574         if (err)
 575                 goto out_unlock_up;
 576
 577         switch (cmd) {
 578         case IPC_SET:
 579         {
 580                 err = -EPERM;
 581                 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
 582                         goto out_unlock_up;
 583
 584                 msq->q_qbytes = setbuf.qbytes;
 585
 586                 ipcp->uid = setbuf.uid;
 587                 ipcp->gid = setbuf.gid;
 588                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
 589                              (S_IRWXUGO & setbuf.mode);
 590                 msq->q_ctime = get_seconds();
 591                 /* sleeping receivers might be excluded by
 592                  * stricter permissions.
 593                  */
 594                 expunge_all(msq, -EAGAIN);
 595                 /* sleeping senders might be able to send
 596                  * due to a larger queue size.
 597                  */
 598                 ss_wakeup(&msq->q_senders, 0);
 599                 msg_unlock(msq);
 600                 break;
 601         }
 602         case IPC_RMID:
 603                 freeque(ns, msq);
 604                 break;
 605         }
 606         err = 0;
 607 out_up:
 608         up_write(&msg_ids(ns).rw_mutex);
 609         return err;
 610 out_unlock_up:
 611         msg_unlock(msq);
 612         goto out_up;
 613 out_unlock:
 614         msg_unlock(msq);
 615         return err;
 616 }
 617
 618 static int testmsg(struct msg_msg *msg, long type, int mode)
 619 {
 620         switch(mode)
 621         {
 622                 case SEARCH_ANY:
 623                         return 1;
 624                 case SEARCH_LESSEQUAL:
 625                         if (msg->m_type <=type)
 626                                 return 1;
 627                         break;
 628                 case SEARCH_EQUAL:
 629                         if (msg->m_type == type)
 630                                 return 1;
 631                         break;
 632                 case SEARCH_NOTEQUAL:
 633                         if (msg->m_type != type)
 634                                 return 1;
 635                         break;
 636         }
 637         return 0;
 638 }
 639
 640 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
 641 {
 642         struct list_head *tmp;
 643
 644         tmp = msq->q_receivers.next;
 645         while (tmp != &msq->q_receivers) {
 646                 struct msg_receiver *msr;
 647
 648                 msr = list_entry(tmp, struct msg_receiver, r_list);
 649                 tmp = tmp->next;
 650                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 651                     !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
 652                                                msr->r_msgtype, msr->r_mode)) {
 653
 654                         list_del(&msr->r_list);
 655                         if (msr->r_maxsize < msg->m_ts) {
 656                                 msr->r_msg = NULL;
 657                                 wake_up_process(msr->r_tsk);
 658                                 smp_mb();
 659                                 msr->r_msg = ERR_PTR(-E2BIG);
 660                         } else {
 661                                 msr->r_msg = NULL;
 662                                 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
 663                                 msq->q_rtime = get_seconds();
 664                                 wake_up_process(msr->r_tsk);
 665                                 smp_mb();
 666                                 msr->r_msg = msg;
 667
 668                                 return 1;
 669                         }
 670                 }
 671         }
 672         return 0;
 673 }
 674
 675 long do_msgsnd(int msqid, long mtype, void __user *mtext,
 676                 size_t msgsz, int msgflg)
 677 {
 678         struct msg_queue *msq;
 679         struct msg_msg *msg;
 680         int err;
 681         struct ipc_namespace *ns;
 682
 683         ns = current->nsproxy->ipc_ns;
 684
 685         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 686                 return -EINVAL;
 687         if (mtype < 1)
 688                 return -EINVAL;
 689
 690         msg = load_msg(mtext, msgsz);
 691         if (IS_ERR(msg))
 692                 return PTR_ERR(msg);
 693
 694         msg->m_type = mtype;
 695         msg->m_ts = msgsz;
 696
 697         msq = msg_lock_check(ns, msqid);
 698         if (IS_ERR(msq)) {
 699                 err = PTR_ERR(msq);
 700                 goto out_free;
 701         }
 702
 703         for (;;) {
 704                 struct msg_sender s;
 705
 706                 err = -EACCES;
 707                 if (ipcperms(&msq->q_perm, S_IWUGO))
 708                         goto out_unlock_free;
 709
 710                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
 711                 if (err)
 712                         goto out_unlock_free;
 713
 714                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
 715                                 1 + msq->q_qnum <= msq->q_qbytes) {
 716                         break;
 717                 }
 718
 719                 /* queue full, wait: */
 720                 if (msgflg & IPC_NOWAIT) {
 721                         err = -EAGAIN;
 722                         goto out_unlock_free;
 723                 }
 724                 ss_add(msq, &s);
 725                 ipc_rcu_getref(msq);
 726                 msg_unlock(msq);
 727                 schedule();
 728
 729                 ipc_lock_by_ptr(&msq->q_perm);
 730                 ipc_rcu_putref(msq);
 731                 if (msq->q_perm.deleted) {
 732                         err = -EIDRM;
 733                         goto out_unlock_free;
 734                 }
 735                 ss_del(&s);
 736
 737                 if (signal_pending(current)) {
 738                         err = -ERESTARTNOHAND;
 739                         goto out_unlock_free;
 740                 }
 741         }
 742
 743         msq->q_lspid = task_tgid_vnr(current);
 744         msq->q_stime = get_seconds();
 745
 746         if (!pipelined_send(msq, msg)) {
 747                 /* noone is waiting for this message, enqueue it */
 748                 list_add_tail(&msg->m_list, &msq->q_messages);
 749                 msq->q_cbytes += msgsz;
 750                 msq->q_qnum++;
 751                 atomic_add(msgsz, &ns->msg_bytes);
 752                 atomic_inc(&ns->msg_hdrs);
 753         }
 754
 755         err = 0;
 756         msg = NULL;
 757
 758 out_unlock_free:
 759         msg_unlock(msq);
 760 out_free:
 761         if (msg != NULL)
 762                 free_msg(msg);
 763         return err;
 764 }
 765
 766 asmlinkage long
 767 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
 768 {
 769         long mtype;
 770
 771         if (get_user(mtype, &msgp->mtype))
 772                 return -EFAULT;
 773         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 774 }
 775
 776 static inline int convert_mode(long *msgtyp, int msgflg)
 777 {
 778         /*
 779          *  find message of correct type.
 780          *  msgtyp = 0 => get first.
 781          *  msgtyp > 0 => get first message of matching type.
 782          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 783          */
 784         if (*msgtyp == 0)
 785                 return SEARCH_ANY;
 786         if (*msgtyp < 0) {
 787                 *msgtyp = -*msgtyp;
 788                 return SEARCH_LESSEQUAL;
 789         }
 790         if (msgflg & MSG_EXCEPT)
 791                 return SEARCH_NOTEQUAL;
 792         return SEARCH_EQUAL;
 793 }
 794
 795 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
 796                 size_t msgsz, long msgtyp, int msgflg)
 797 {
 798         struct msg_queue *msq;
 799         struct msg_msg *msg;
 800         int mode;
 801         struct ipc_namespace *ns;
 802
 803         if (msqid < 0 || (long) msgsz < 0)
 804                 return -EINVAL;
 805         mode = convert_mode(&msgtyp, msgflg);
 806         ns = current->nsproxy->ipc_ns;
 807
 808         msq = msg_lock_check(ns, msqid);
 809         if (IS_ERR(msq))
 810                 return PTR_ERR(msq);
 811
 812         for (;;) {
 813                 struct msg_receiver msr_d;
 814                 struct list_head *tmp;
 815
 816                 msg = ERR_PTR(-EACCES);
 817                 if (ipcperms(&msq->q_perm, S_IRUGO))
 818                         goto out_unlock;
 819
 820                 msg = ERR_PTR(-EAGAIN);
 821                 tmp = msq->q_messages.next;
 822                 while (tmp != &msq->q_messages) {
 823                         struct msg_msg *walk_msg;
 824
 825                         walk_msg = list_entry(tmp, struct msg_msg, m_list);
 826                         if (testmsg(walk_msg, msgtyp, mode) &&
 827                             !security_msg_queue_msgrcv(msq, walk_msg, current,
 828                                                        msgtyp, mode)) {
 829
 830                                 msg = walk_msg;
 831                                 if (mode == SEARCH_LESSEQUAL &&
 832                                                 walk_msg->m_type != 1) {
 833                                         msg = walk_msg;
 834                                         msgtyp = walk_msg->m_type - 1;
 835                                 } else {
 836                                         msg = walk_msg;
 837                                         break;
 838                                 }
 839                         }
 840                         tmp = tmp->next;
 841                 }
 842                 if (!IS_ERR(msg)) {
 843                         /*
 844                          * Found a suitable message.
 845                          * Unlink it from the queue.
 846                          */
 847                         if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
 848                                 msg = ERR_PTR(-E2BIG);
 849                                 goto out_unlock;
 850                         }
 851                         list_del(&msg->m_list);
 852                         msq->q_qnum--;
 853                         msq->q_rtime = get_seconds();
 854                         msq->q_lrpid = task_tgid_vnr(current);
 855                         msq->q_cbytes -= msg->m_ts;
 856                         atomic_sub(msg->m_ts, &ns->msg_bytes);
 857                         atomic_dec(&ns->msg_hdrs);
 858                         ss_wakeup(&msq->q_senders, 0);
 859                         msg_unlock(msq);
 860                         break;
 861                 }
 862                 /* No message waiting. Wait for a message */
 863                 if (msgflg & IPC_NOWAIT) {
 864                         msg = ERR_PTR(-ENOMSG);
 865                         goto out_unlock;
 866                 }
 867                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
 868                 msr_d.r_tsk = current;
 869                 msr_d.r_msgtype = msgtyp;
 870                 msr_d.r_mode = mode;
 871                 if (msgflg & MSG_NOERROR)
 872                         msr_d.r_maxsize = INT_MAX;
 873                 else
 874                         msr_d.r_maxsize = msgsz;
 875                 msr_d.r_msg = ERR_PTR(-EAGAIN);
 876                 current->state = TASK_INTERRUPTIBLE;
 877                 msg_unlock(msq);
 878
 879                 schedule();
 880
 881                 /* Lockless receive, part 1:
 882                  * Disable preemption.  We don't hold a reference to the queue
 883                  * and getting a reference would defeat the idea of a lockless
 884                  * operation, thus the code relies on rcu to guarantee the
 885                  * existance of msq:
 886                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
 887                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
 888                  * rcu_read_lock() prevents preemption between reading r_msg
 889                  * and the spin_lock() inside ipc_lock_by_ptr().
 890                  */
 891                 rcu_read_lock();
 892
 893                 /* Lockless receive, part 2:
 894                  * Wait until pipelined_send or expunge_all are outside of
 895                  * wake_up_process(). There is a race with exit(), see
 896                  * ipc/mqueue.c for the details.
 897                  */
 898                 msg = (struct msg_msg*)msr_d.r_msg;
 899                 while (msg == NULL) {
 900                         cpu_relax();
 901                         msg = (struct msg_msg *)msr_d.r_msg;
 902                 }
 903
 904                 /* Lockless receive, part 3:
 905                  * If there is a message or an error then accept it without
 906                  * locking.
 907                  */
 908                 if (msg != ERR_PTR(-EAGAIN)) {
 909                         rcu_read_unlock();
 910                         break;
 911                 }
 912
 913                 /* Lockless receive, part 3:
 914                  * Acquire the queue spinlock.
 915                  */
 916                 ipc_lock_by_ptr(&msq->q_perm);
 917                 rcu_read_unlock();
 918
 919                 /* Lockless receive, part 4:
 920                  * Repeat test after acquiring the spinlock.
 921                  */
 922                 msg = (struct msg_msg*)msr_d.r_msg;
 923                 if (msg != ERR_PTR(-EAGAIN))
 924                         goto out_unlock;
 925
 926                 list_del(&msr_d.r_list);
 927                 if (signal_pending(current)) {
 928                         msg = ERR_PTR(-ERESTARTNOHAND);
 929 out_unlock:
 930                         msg_unlock(msq);
 931                         break;
 932                 }
 933         }
 934         if (IS_ERR(msg))
 935                 return PTR_ERR(msg);
 936
 937         msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
 938         *pmtype = msg->m_type;
 939         if (store_msg(mtext, msg, msgsz))
 940                 msgsz = -EFAULT;
 941
 942         free_msg(msg);
 943
 944         return msgsz;
 945 }
 946
 947 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 948                            long msgtyp, int msgflg)
 949 {
 950         long err, mtype;
 951
 952         err =  do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
 953         if (err < 0)
 954                 goto out;
 955
 956         if (put_user(mtype, &msgp->mtype))
 957                 err = -EFAULT;
 958 out:
 959         return err;
 960 }
 961
 962 #ifdef CONFIG_PROC_FS
 963 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
 964 {
 965         struct msg_queue *msq = it;
 966
 967         return seq_printf(s,
 968                         "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
 969                         msq->q_perm.key,
 970                         msq->q_perm.id,
 971                         msq->q_perm.mode,
 972                         msq->q_cbytes,
 973                         msq->q_qnum,
 974                         msq->q_lspid,
 975                         msq->q_lrpid,
 976                         msq->q_perm.uid,
 977                         msq->q_perm.gid,
 978                         msq->q_perm.cuid,
 979                         msq->q_perm.cgid,
 980                         msq->q_stime,
 981                         msq->q_rtime,
 982                         msq->q_ctime);
 983 }
 984 #endif