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