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