ipc/msg.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/ipc/msg.c
   4  * Copyright (C) 1992 Krishna Balasubramanian
   5  *
   6  * Removed all the remaining kerneld mess
   7  * Catch the -EFAULT stuff properly
   8  * Use GFP_KERNEL for messages as in 1.2
   9  * Fixed up the unchecked user space derefs
  10  * Copyright (C) 1998 Alan Cox & Andi Kleen
  11  *
  12  * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  13  *
  14  * mostly rewritten, threaded and wake-one semantics added
  15  * MSGMAX limit removed, sysctl's added
  16  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  17  *
  18  * support for audit of ipc object properties and permission changes
  19  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  20  *
  21  * namespaces support
  22  * OpenVZ, SWsoft Inc.
  23  * Pavel Emelianov <xemul@openvz.org>
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/msg.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/init.h>
  30 #include <linux/mm.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/list.h>
  33 #include <linux/security.h>
  34 #include <linux/sched/wake_q.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/audit.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/rwsem.h>
  39 #include <linux/nsproxy.h>
  40 #include <linux/ipc_namespace.h>
  41
  42 #include <asm/current.h>
  43 #include <linux/uaccess.h>
  44 #include "util.h"
  45
  46 /* one msq_queue structure for each present queue on the system */
  47 struct msg_queue {
  48         struct kern_ipc_perm q_perm;
  49         time64_t q_stime;               /* last msgsnd time */
  50         time64_t q_rtime;               /* last msgrcv time */
  51         time64_t q_ctime;               /* last change time */
  52         unsigned long q_cbytes;         /* current number of bytes on queue */
  53         unsigned long q_qnum;           /* number of messages in queue */
  54         unsigned long q_qbytes;         /* max number of bytes on queue */
  55         struct pid *q_lspid;            /* pid of last msgsnd */
  56         struct pid *q_lrpid;            /* last receive pid */
  57
  58         struct list_head q_messages;
  59         struct list_head q_receivers;
  60         struct list_head q_senders;
  61 } __randomize_layout;
  62
  63 /* one msg_receiver structure for each sleeping receiver */
  64 struct msg_receiver {
  65         struct list_head        r_list;
  66         struct task_struct      *r_tsk;
  67
  68         int                     r_mode;
  69         long                    r_msgtype;
  70         long                    r_maxsize;
  71
  72         struct msg_msg          *r_msg;
  73 };
  74
  75 /* one msg_sender for each sleeping sender */
  76 struct msg_sender {
  77         struct list_head        list;
  78         struct task_struct      *tsk;
  79         size_t                  msgsz;
  80 };
  81
  82 #define SEARCH_ANY              1
  83 #define SEARCH_EQUAL            2
  84 #define SEARCH_NOTEQUAL         3
  85 #define SEARCH_LESSEQUAL        4
  86 #define SEARCH_NUMBER           5
  87
  88 #define msg_ids(ns)     ((ns)->ids[IPC_MSG_IDS])
  89
  90 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
  91 {
  92         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
  93
  94         if (IS_ERR(ipcp))
  95                 return ERR_CAST(ipcp);
  96
  97         return container_of(ipcp, struct msg_queue, q_perm);
  98 }
  99
 100 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
 101                                                         int id)
 102 {
 103         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
 104
 105         if (IS_ERR(ipcp))
 106                 return ERR_CAST(ipcp);
 107
 108         return container_of(ipcp, struct msg_queue, q_perm);
 109 }
 110
 111 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
 112 {
 113         ipc_rmid(&msg_ids(ns), &s->q_perm);
 114 }
 115
 116 static void msg_rcu_free(struct rcu_head *head)
 117 {
 118         struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
 119         struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
 120
 121         security_msg_queue_free(&msq->q_perm);
 122         kvfree(msq);
 123 }
 124
 125 /**
 126  * newque - Create a new msg queue
 127  * @ns: namespace
 128  * @params: ptr to the structure that contains the key and msgflg
 129  *
 130  * Called with msg_ids.rwsem held (writer)
 131  */
 132 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
 133 {
 134         struct msg_queue *msq;
 135         int retval;
 136         key_t key = params->key;
 137         int msgflg = params->flg;
 138
 139         msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
 140         if (unlikely(!msq))
 141                 return -ENOMEM;
 142
 143         msq->q_perm.mode = msgflg & S_IRWXUGO;
 144         msq->q_perm.key = key;
 145
 146         msq->q_perm.security = NULL;
 147         retval = security_msg_queue_alloc(&msq->q_perm);
 148         if (retval) {
 149                 kvfree(msq);
 150                 return retval;
 151         }
 152
 153         msq->q_stime = msq->q_rtime = 0;
 154         msq->q_ctime = ktime_get_real_seconds();
 155         msq->q_cbytes = msq->q_qnum = 0;
 156         msq->q_qbytes = ns->msg_ctlmnb;
 157         msq->q_lspid = msq->q_lrpid = NULL;
 158         INIT_LIST_HEAD(&msq->q_messages);
 159         INIT_LIST_HEAD(&msq->q_receivers);
 160         INIT_LIST_HEAD(&msq->q_senders);
 161
 162         /* ipc_addid() locks msq upon success. */
 163         retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
 164         if (retval < 0) {
 165                 call_rcu(&msq->q_perm.rcu, msg_rcu_free);
 166                 return retval;
 167         }
 168
 169         ipc_unlock_object(&msq->q_perm);
 170         rcu_read_unlock();
 171
 172         return msq->q_perm.id;
 173 }
 174
 175 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
 176 {
 177         return msgsz + msq->q_cbytes <= msq->q_qbytes &&
 178                 1 + msq->q_qnum <= msq->q_qbytes;
 179 }
 180
 181 static inline void ss_add(struct msg_queue *msq,
 182                           struct msg_sender *mss, size_t msgsz)
 183 {
 184         mss->tsk = current;
 185         mss->msgsz = msgsz;
 186         __set_current_state(TASK_INTERRUPTIBLE);
 187         list_add_tail(&mss->list, &msq->q_senders);
 188 }
 189
 190 static inline void ss_del(struct msg_sender *mss)
 191 {
 192         if (mss->list.next)
 193                 list_del(&mss->list);
 194 }
 195
 196 static void ss_wakeup(struct msg_queue *msq,
 197                       struct wake_q_head *wake_q, bool kill)
 198 {
 199         struct msg_sender *mss, *t;
 200         struct task_struct *stop_tsk = NULL;
 201         struct list_head *h = &msq->q_senders;
 202
 203         list_for_each_entry_safe(mss, t, h, list) {
 204                 if (kill)
 205                         mss->list.next = NULL;
 206
 207                 /*
 208                  * Stop at the first task we don't wakeup,
 209                  * we've already iterated the original
 210                  * sender queue.
 211                  */
 212                 else if (stop_tsk == mss->tsk)
 213                         break;
 214                 /*
 215                  * We are not in an EIDRM scenario here, therefore
 216                  * verify that we really need to wakeup the task.
 217                  * To maintain current semantics and wakeup order,
 218                  * move the sender to the tail on behalf of the
 219                  * blocked task.
 220                  */
 221                 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
 222                         if (!stop_tsk)
 223                                 stop_tsk = mss->tsk;
 224
 225                         list_move_tail(&mss->list, &msq->q_senders);
 226                         continue;
 227                 }
 228
 229                 wake_q_add(wake_q, mss->tsk);
 230         }
 231 }
 232
 233 static void expunge_all(struct msg_queue *msq, int res,
 234                         struct wake_q_head *wake_q)
 235 {
 236         struct msg_receiver *msr, *t;
 237
 238         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 239                 wake_q_add(wake_q, msr->r_tsk);
 240                 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
 241         }
 242 }
 243
 244 /*
 245  * freeque() wakes up waiters on the sender and receiver waiting queue,
 246  * removes the message queue from message queue ID IDR, and cleans up all the
 247  * messages associated with this queue.
 248  *
 249  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
 250  * before freeque() is called. msg_ids.rwsem remains locked on exit.
 251  */
 252 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 253 {
 254         struct msg_msg *msg, *t;
 255         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
 256         DEFINE_WAKE_Q(wake_q);
 257
 258         expunge_all(msq, -EIDRM, &wake_q);
 259         ss_wakeup(msq, &wake_q, true);
 260         msg_rmid(ns, msq);
 261         ipc_unlock_object(&msq->q_perm);
 262         wake_up_q(&wake_q);
 263         rcu_read_unlock();
 264
 265         list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
 266                 atomic_dec(&ns->msg_hdrs);
 267                 free_msg(msg);
 268         }
 269         atomic_sub(msq->q_cbytes, &ns->msg_bytes);
 270         ipc_update_pid(&msq->q_lspid, NULL);
 271         ipc_update_pid(&msq->q_lrpid, NULL);
 272         ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
 273 }
 274
 275 long ksys_msgget(key_t key, int msgflg)
 276 {
 277         struct ipc_namespace *ns;
 278         static const struct ipc_ops msg_ops = {
 279                 .getnew = newque,
 280                 .associate = security_msg_queue_associate,
 281         };
 282         struct ipc_params msg_params;
 283
 284         ns = current->nsproxy->ipc_ns;
 285
 286         msg_params.key = key;
 287         msg_params.flg = msgflg;
 288
 289         return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
 290 }
 291
 292 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
 293 {
 294         return ksys_msgget(key, msgflg);
 295 }
 296
 297 static inline unsigned long
 298 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
 299 {
 300         switch (version) {
 301         case IPC_64:
 302                 return copy_to_user(buf, in, sizeof(*in));
 303         case IPC_OLD:
 304         {
 305                 struct msqid_ds out;
 306
 307                 memset(&out, 0, sizeof(out));
 308
 309                 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
 310
 311                 out.msg_stime           = in->msg_stime;
 312                 out.msg_rtime           = in->msg_rtime;
 313                 out.msg_ctime           = in->msg_ctime;
 314
 315                 if (in->msg_cbytes > USHRT_MAX)
 316                         out.msg_cbytes  = USHRT_MAX;
 317                 else
 318                         out.msg_cbytes  = in->msg_cbytes;
 319                 out.msg_lcbytes         = in->msg_cbytes;
 320
 321                 if (in->msg_qnum > USHRT_MAX)
 322                         out.msg_qnum    = USHRT_MAX;
 323                 else
 324                         out.msg_qnum    = in->msg_qnum;
 325
 326                 if (in->msg_qbytes > USHRT_MAX)
 327                         out.msg_qbytes  = USHRT_MAX;
 328                 else
 329                         out.msg_qbytes  = in->msg_qbytes;
 330                 out.msg_lqbytes         = in->msg_qbytes;
 331
 332                 out.msg_lspid           = in->msg_lspid;
 333                 out.msg_lrpid           = in->msg_lrpid;
 334
 335                 return copy_to_user(buf, &out, sizeof(out));
 336         }
 337         default:
 338                 return -EINVAL;
 339         }
 340 }
 341
 342 static inline unsigned long
 343 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
 344 {
 345         switch (version) {
 346         case IPC_64:
 347                 if (copy_from_user(out, buf, sizeof(*out)))
 348                         return -EFAULT;
 349                 return 0;
 350         case IPC_OLD:
 351         {
 352                 struct msqid_ds tbuf_old;
 353
 354                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 355                         return -EFAULT;
 356
 357                 out->msg_perm.uid       = tbuf_old.msg_perm.uid;
 358                 out->msg_perm.gid       = tbuf_old.msg_perm.gid;
 359                 out->msg_perm.mode      = tbuf_old.msg_perm.mode;
 360
 361                 if (tbuf_old.msg_qbytes == 0)
 362                         out->msg_qbytes = tbuf_old.msg_lqbytes;
 363                 else
 364                         out->msg_qbytes = tbuf_old.msg_qbytes;
 365
 366                 return 0;
 367         }
 368         default:
 369                 return -EINVAL;
 370         }
 371 }
 372
 373 /*
 374  * This function handles some msgctl commands which require the rwsem
 375  * to be held in write mode.
 376  * NOTE: no locks must be held, the rwsem is taken inside this function.
 377  */
 378 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
 379                         struct msqid64_ds *msqid64)
 380 {
 381         struct kern_ipc_perm *ipcp;
 382         struct msg_queue *msq;
 383         int err;
 384
 385         down_write(&msg_ids(ns).rwsem);
 386         rcu_read_lock();
 387
 388         ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
 389                                       &msqid64->msg_perm, msqid64->msg_qbytes);
 390         if (IS_ERR(ipcp)) {
 391                 err = PTR_ERR(ipcp);
 392                 goto out_unlock1;
 393         }
 394
 395         msq = container_of(ipcp, struct msg_queue, q_perm);
 396
 397         err = security_msg_queue_msgctl(&msq->q_perm, cmd);
 398         if (err)
 399                 goto out_unlock1;
 400
 401         switch (cmd) {
 402         case IPC_RMID:
 403                 ipc_lock_object(&msq->q_perm);
 404                 /* freeque unlocks the ipc object and rcu */
 405                 freeque(ns, ipcp);
 406                 goto out_up;
 407         case IPC_SET:
 408         {
 409                 DEFINE_WAKE_Q(wake_q);
 410
 411                 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
 412                     !capable(CAP_SYS_RESOURCE)) {
 413                         err = -EPERM;
 414                         goto out_unlock1;
 415                 }
 416
 417                 ipc_lock_object(&msq->q_perm);
 418                 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
 419                 if (err)
 420                         goto out_unlock0;
 421
 422                 msq->q_qbytes = msqid64->msg_qbytes;
 423
 424                 msq->q_ctime = ktime_get_real_seconds();
 425                 /*
 426                  * Sleeping receivers might be excluded by
 427                  * stricter permissions.
 428                  */
 429                 expunge_all(msq, -EAGAIN, &wake_q);
 430                 /*
 431                  * Sleeping senders might be able to send
 432                  * due to a larger queue size.
 433                  */
 434                 ss_wakeup(msq, &wake_q, false);
 435                 ipc_unlock_object(&msq->q_perm);
 436                 wake_up_q(&wake_q);
 437
 438                 goto out_unlock1;
 439         }
 440         default:
 441                 err = -EINVAL;
 442                 goto out_unlock1;
 443         }
 444
 445 out_unlock0:
 446         ipc_unlock_object(&msq->q_perm);
 447 out_unlock1:
 448         rcu_read_unlock();
 449 out_up:
 450         up_write(&msg_ids(ns).rwsem);
 451         return err;
 452 }
 453
 454 static int msgctl_info(struct ipc_namespace *ns, int msqid,
 455                          int cmd, struct msginfo *msginfo)
 456 {
 457         int err;
 458         int max_id;
 459
 460         /*
 461          * We must not return kernel stack data.
 462          * due to padding, it's not enough
 463          * to set all member fields.
 464          */
 465         err = security_msg_queue_msgctl(NULL, cmd);
 466         if (err)
 467                 return err;
 468
 469         memset(msginfo, 0, sizeof(*msginfo));
 470         msginfo->msgmni = ns->msg_ctlmni;
 471         msginfo->msgmax = ns->msg_ctlmax;
 472         msginfo->msgmnb = ns->msg_ctlmnb;
 473         msginfo->msgssz = MSGSSZ;
 474         msginfo->msgseg = MSGSEG;
 475         down_read(&msg_ids(ns).rwsem);
 476         if (cmd == MSG_INFO) {
 477                 msginfo->msgpool = msg_ids(ns).in_use;
 478                 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
 479                 msginfo->msgtql = atomic_read(&ns->msg_bytes);
 480         } else {
 481                 msginfo->msgmap = MSGMAP;
 482                 msginfo->msgpool = MSGPOOL;
 483                 msginfo->msgtql = MSGTQL;
 484         }
 485         max_id = ipc_get_maxid(&msg_ids(ns));
 486         up_read(&msg_ids(ns).rwsem);
 487         return (max_id < 0) ? 0 : max_id;
 488 }
 489
 490 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
 491                          int cmd, struct msqid64_ds *p)
 492 {
 493         struct msg_queue *msq;
 494         int id = 0;
 495         int err;
 496
 497         memset(p, 0, sizeof(*p));
 498
 499         rcu_read_lock();
 500         if (cmd == MSG_STAT) {
 501                 msq = msq_obtain_object(ns, msqid);
 502                 if (IS_ERR(msq)) {
 503                         err = PTR_ERR(msq);
 504                         goto out_unlock;
 505                 }
 506                 id = msq->q_perm.id;
 507         } else {
 508                 msq = msq_obtain_object_check(ns, msqid);
 509                 if (IS_ERR(msq)) {
 510                         err = PTR_ERR(msq);
 511                         goto out_unlock;
 512                 }
 513         }
 514
 515         err = -EACCES;
 516         if (ipcperms(ns, &msq->q_perm, S_IRUGO))
 517                 goto out_unlock;
 518
 519         err = security_msg_queue_msgctl(&msq->q_perm, cmd);
 520         if (err)
 521                 goto out_unlock;
 522
 523         ipc_lock_object(&msq->q_perm);
 524
 525         if (!ipc_valid_object(&msq->q_perm)) {
 526                 ipc_unlock_object(&msq->q_perm);
 527                 err = -EIDRM;
 528                 goto out_unlock;
 529         }
 530
 531         kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
 532         p->msg_stime  = msq->q_stime;
 533         p->msg_rtime  = msq->q_rtime;
 534         p->msg_ctime  = msq->q_ctime;
 535         p->msg_cbytes = msq->q_cbytes;
 536         p->msg_qnum   = msq->q_qnum;
 537         p->msg_qbytes = msq->q_qbytes;
 538         p->msg_lspid  = pid_vnr(msq->q_lspid);
 539         p->msg_lrpid  = pid_vnr(msq->q_lrpid);
 540
 541         ipc_unlock_object(&msq->q_perm);
 542         rcu_read_unlock();
 543         return id;
 544
 545 out_unlock:
 546         rcu_read_unlock();
 547         return err;
 548 }
 549
 550 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
 551 {
 552         int version;
 553         struct ipc_namespace *ns;
 554         struct msqid64_ds msqid64;
 555         int err;
 556
 557         if (msqid < 0 || cmd < 0)
 558                 return -EINVAL;
 559
 560         version = ipc_parse_version(&cmd);
 561         ns = current->nsproxy->ipc_ns;
 562
 563         switch (cmd) {
 564         case IPC_INFO:
 565         case MSG_INFO: {
 566                 struct msginfo msginfo;
 567                 err = msgctl_info(ns, msqid, cmd, &msginfo);
 568                 if (err < 0)
 569                         return err;
 570                 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
 571                         err = -EFAULT;
 572                 return err;
 573         }
 574         case MSG_STAT:  /* msqid is an index rather than a msg queue id */
 575         case IPC_STAT:
 576                 err = msgctl_stat(ns, msqid, cmd, &msqid64);
 577                 if (err < 0)
 578                         return err;
 579                 if (copy_msqid_to_user(buf, &msqid64, version))
 580                         err = -EFAULT;
 581                 return err;
 582         case IPC_SET:
 583                 if (copy_msqid_from_user(&msqid64, buf, version))
 584                         return -EFAULT;
 585                 /* fallthru */
 586         case IPC_RMID:
 587                 return msgctl_down(ns, msqid, cmd, &msqid64);
 588         default:
 589                 return  -EINVAL;
 590         }
 591 }
 592
 593 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
 594 {
 595         return ksys_msgctl(msqid, cmd, buf);
 596 }
 597
 598 #ifdef CONFIG_COMPAT
 599
 600 struct compat_msqid_ds {
 601         struct compat_ipc_perm msg_perm;
 602         compat_uptr_t msg_first;
 603         compat_uptr_t msg_last;
 604         compat_time_t msg_stime;
 605         compat_time_t msg_rtime;
 606         compat_time_t msg_ctime;
 607         compat_ulong_t msg_lcbytes;
 608         compat_ulong_t msg_lqbytes;
 609         unsigned short msg_cbytes;
 610         unsigned short msg_qnum;
 611         unsigned short msg_qbytes;
 612         compat_ipc_pid_t msg_lspid;
 613         compat_ipc_pid_t msg_lrpid;
 614 };
 615
 616 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
 617                                         int version)
 618 {
 619         memset(out, 0, sizeof(*out));
 620         if (version == IPC_64) {
 621                 struct compat_msqid64_ds __user *p = buf;
 622                 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
 623                         return -EFAULT;
 624                 if (get_user(out->msg_qbytes, &p->msg_qbytes))
 625                         return -EFAULT;
 626         } else {
 627                 struct compat_msqid_ds __user *p = buf;
 628                 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
 629                         return -EFAULT;
 630                 if (get_user(out->msg_qbytes, &p->msg_qbytes))
 631                         return -EFAULT;
 632         }
 633         return 0;
 634 }
 635
 636 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
 637                                         int version)
 638 {
 639         if (version == IPC_64) {
 640                 struct compat_msqid64_ds v;
 641                 memset(&v, 0, sizeof(v));
 642                 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
 643                 v.msg_stime = in->msg_stime;
 644                 v.msg_rtime = in->msg_rtime;
 645                 v.msg_ctime = in->msg_ctime;
 646                 v.msg_cbytes = in->msg_cbytes;
 647                 v.msg_qnum = in->msg_qnum;
 648                 v.msg_qbytes = in->msg_qbytes;
 649                 v.msg_lspid = in->msg_lspid;
 650                 v.msg_lrpid = in->msg_lrpid;
 651                 return copy_to_user(buf, &v, sizeof(v));
 652         } else {
 653                 struct compat_msqid_ds v;
 654                 memset(&v, 0, sizeof(v));
 655                 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
 656                 v.msg_stime = in->msg_stime;
 657                 v.msg_rtime = in->msg_rtime;
 658                 v.msg_ctime = in->msg_ctime;
 659                 v.msg_cbytes = in->msg_cbytes;
 660                 v.msg_qnum = in->msg_qnum;
 661                 v.msg_qbytes = in->msg_qbytes;
 662                 v.msg_lspid = in->msg_lspid;
 663                 v.msg_lrpid = in->msg_lrpid;
 664                 return copy_to_user(buf, &v, sizeof(v));
 665         }
 666 }
 667
 668 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
 669 {
 670         struct ipc_namespace *ns;
 671         int err;
 672         struct msqid64_ds msqid64;
 673         int version = compat_ipc_parse_version(&cmd);
 674
 675         ns = current->nsproxy->ipc_ns;
 676
 677         if (msqid < 0 || cmd < 0)
 678                 return -EINVAL;
 679
 680         switch (cmd & (~IPC_64)) {
 681         case IPC_INFO:
 682         case MSG_INFO: {
 683                 struct msginfo msginfo;
 684                 err = msgctl_info(ns, msqid, cmd, &msginfo);
 685                 if (err < 0)
 686                         return err;
 687                 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
 688                         err = -EFAULT;
 689                 return err;
 690         }
 691         case IPC_STAT:
 692         case MSG_STAT:
 693                 err = msgctl_stat(ns, msqid, cmd, &msqid64);
 694                 if (err < 0)
 695                         return err;
 696                 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
 697                         err = -EFAULT;
 698                 return err;
 699         case IPC_SET:
 700                 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
 701                         return -EFAULT;
 702                 /* fallthru */
 703         case IPC_RMID:
 704                 return msgctl_down(ns, msqid, cmd, &msqid64);
 705         default:
 706                 return -EINVAL;
 707         }
 708 }
 709
 710 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
 711 {
 712         return compat_ksys_msgctl(msqid, cmd, uptr);
 713 }
 714 #endif
 715
 716 static int testmsg(struct msg_msg *msg, long type, int mode)
 717 {
 718         switch (mode) {
 719         case SEARCH_ANY:
 720         case SEARCH_NUMBER:
 721                 return 1;
 722         case SEARCH_LESSEQUAL:
 723                 if (msg->m_type <= type)
 724                         return 1;
 725                 break;
 726         case SEARCH_EQUAL:
 727                 if (msg->m_type == type)
 728                         return 1;
 729                 break;
 730         case SEARCH_NOTEQUAL:
 731                 if (msg->m_type != type)
 732                         return 1;
 733                 break;
 734         }
 735         return 0;
 736 }
 737
 738 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
 739                                  struct wake_q_head *wake_q)
 740 {
 741         struct msg_receiver *msr, *t;
 742
 743         list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
 744                 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
 745                     !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
 746                                                msr->r_msgtype, msr->r_mode)) {
 747
 748                         list_del(&msr->r_list);
 749                         if (msr->r_maxsize < msg->m_ts) {
 750                                 wake_q_add(wake_q, msr->r_tsk);
 751                                 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
 752                         } else {
 753                                 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
 754                                 msq->q_rtime = get_seconds();
 755
 756                                 wake_q_add(wake_q, msr->r_tsk);
 757                                 WRITE_ONCE(msr->r_msg, msg);
 758                                 return 1;
 759                         }
 760                 }
 761         }
 762
 763         return 0;
 764 }
 765
 766 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
 767                 size_t msgsz, int msgflg)
 768 {
 769         struct msg_queue *msq;
 770         struct msg_msg *msg;
 771         int err;
 772         struct ipc_namespace *ns;
 773         DEFINE_WAKE_Q(wake_q);
 774
 775         ns = current->nsproxy->ipc_ns;
 776
 777         if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
 778                 return -EINVAL;
 779         if (mtype < 1)
 780                 return -EINVAL;
 781
 782         msg = load_msg(mtext, msgsz);
 783         if (IS_ERR(msg))
 784                 return PTR_ERR(msg);
 785
 786         msg->m_type = mtype;
 787         msg->m_ts = msgsz;
 788
 789         rcu_read_lock();
 790         msq = msq_obtain_object_check(ns, msqid);
 791         if (IS_ERR(msq)) {
 792                 err = PTR_ERR(msq);
 793                 goto out_unlock1;
 794         }
 795
 796         ipc_lock_object(&msq->q_perm);
 797
 798         for (;;) {
 799                 struct msg_sender s;
 800
 801                 err = -EACCES;
 802                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
 803                         goto out_unlock0;
 804
 805                 /* raced with RMID? */
 806                 if (!ipc_valid_object(&msq->q_perm)) {
 807                         err = -EIDRM;
 808                         goto out_unlock0;
 809                 }
 810
 811                 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
 812                 if (err)
 813                         goto out_unlock0;
 814
 815                 if (msg_fits_inqueue(msq, msgsz))
 816                         break;
 817
 818                 /* queue full, wait: */
 819                 if (msgflg & IPC_NOWAIT) {
 820                         err = -EAGAIN;
 821                         goto out_unlock0;
 822                 }
 823
 824                 /* enqueue the sender and prepare to block */
 825                 ss_add(msq, &s, msgsz);
 826
 827                 if (!ipc_rcu_getref(&msq->q_perm)) {
 828                         err = -EIDRM;
 829                         goto out_unlock0;
 830                 }
 831
 832                 ipc_unlock_object(&msq->q_perm);
 833                 rcu_read_unlock();
 834                 schedule();
 835
 836                 rcu_read_lock();
 837                 ipc_lock_object(&msq->q_perm);
 838
 839                 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
 840                 /* raced with RMID? */
 841                 if (!ipc_valid_object(&msq->q_perm)) {
 842                         err = -EIDRM;
 843                         goto out_unlock0;
 844                 }
 845                 ss_del(&s);
 846
 847                 if (signal_pending(current)) {
 848                         err = -ERESTARTNOHAND;
 849                         goto out_unlock0;
 850                 }
 851
 852         }
 853
 854         ipc_update_pid(&msq->q_lspid, task_tgid(current));
 855         msq->q_stime = get_seconds();
 856
 857         if (!pipelined_send(msq, msg, &wake_q)) {
 858                 /* no one is waiting for this message, enqueue it */
 859                 list_add_tail(&msg->m_list, &msq->q_messages);
 860                 msq->q_cbytes += msgsz;
 861                 msq->q_qnum++;
 862                 atomic_add(msgsz, &ns->msg_bytes);
 863                 atomic_inc(&ns->msg_hdrs);
 864         }
 865
 866         err = 0;
 867         msg = NULL;
 868
 869 out_unlock0:
 870         ipc_unlock_object(&msq->q_perm);
 871         wake_up_q(&wake_q);
 872 out_unlock1:
 873         rcu_read_unlock();
 874         if (msg != NULL)
 875                 free_msg(msg);
 876         return err;
 877 }
 878
 879 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
 880                  int msgflg)
 881 {
 882         long mtype;
 883
 884         if (get_user(mtype, &msgp->mtype))
 885                 return -EFAULT;
 886         return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
 887 }
 888
 889 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
 890                 int, msgflg)
 891 {
 892         return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
 893 }
 894
 895 #ifdef CONFIG_COMPAT
 896
 897 struct compat_msgbuf {
 898         compat_long_t mtype;
 899         char mtext[1];
 900 };
 901
 902 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
 903                        compat_ssize_t msgsz, int msgflg)
 904 {
 905         struct compat_msgbuf __user *up = compat_ptr(msgp);
 906         compat_long_t mtype;
 907
 908         if (get_user(mtype, &up->mtype))
 909                 return -EFAULT;
 910         return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
 911 }
 912
 913 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
 914                        compat_ssize_t, msgsz, int, msgflg)
 915 {
 916         return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
 917 }
 918 #endif
 919
 920 static inline int convert_mode(long *msgtyp, int msgflg)
 921 {
 922         if (msgflg & MSG_COPY)
 923                 return SEARCH_NUMBER;
 924         /*
 925          *  find message of correct type.
 926          *  msgtyp = 0 => get first.
 927          *  msgtyp > 0 => get first message of matching type.
 928          *  msgtyp < 0 => get message with least type must be < abs(msgtype).
 929          */
 930         if (*msgtyp == 0)
 931                 return SEARCH_ANY;
 932         if (*msgtyp < 0) {
 933                 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
 934                         *msgtyp = LONG_MAX;
 935                 else
 936                         *msgtyp = -*msgtyp;
 937                 return SEARCH_LESSEQUAL;
 938         }
 939         if (msgflg & MSG_EXCEPT)
 940                 return SEARCH_NOTEQUAL;
 941         return SEARCH_EQUAL;
 942 }
 943
 944 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
 945 {
 946         struct msgbuf __user *msgp = dest;
 947         size_t msgsz;
 948
 949         if (put_user(msg->m_type, &msgp->mtype))
 950                 return -EFAULT;
 951
 952         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
 953         if (store_msg(msgp->mtext, msg, msgsz))
 954                 return -EFAULT;
 955         return msgsz;
 956 }
 957
 958 #ifdef CONFIG_CHECKPOINT_RESTORE
 959 /*
 960  * This function creates new kernel message structure, large enough to store
 961  * bufsz message bytes.
 962  */
 963 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 964 {
 965         struct msg_msg *copy;
 966
 967         /*
 968          * Create dummy message to copy real message to.
 969          */
 970         copy = load_msg(buf, bufsz);
 971         if (!IS_ERR(copy))
 972                 copy->m_ts = bufsz;
 973         return copy;
 974 }
 975
 976 static inline void free_copy(struct msg_msg *copy)
 977 {
 978         if (copy)
 979                 free_msg(copy);
 980 }
 981 #else
 982 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
 983 {
 984         return ERR_PTR(-ENOSYS);
 985 }
 986
 987 static inline void free_copy(struct msg_msg *copy)
 988 {
 989 }
 990 #endif
 991
 992 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
 993 {
 994         struct msg_msg *msg, *found = NULL;
 995         long count = 0;
 996
 997         list_for_each_entry(msg, &msq->q_messages, m_list) {
 998                 if (testmsg(msg, *msgtyp, mode) &&
 999                     !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1000                                                *msgtyp, mode)) {
1001                         if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1002                                 *msgtyp = msg->m_type - 1;
1003                                 found = msg;
1004                         } else if (mode == SEARCH_NUMBER) {
1005                                 if (*msgtyp == count)
1006                                         return msg;
1007                         } else
1008                                 return msg;
1009                         count++;
1010                 }
1011         }
1012
1013         return found ?: ERR_PTR(-EAGAIN);
1014 }
1015
1016 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1017                long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1018 {
1019         int mode;
1020         struct msg_queue *msq;
1021         struct ipc_namespace *ns;
1022         struct msg_msg *msg, *copy = NULL;
1023         DEFINE_WAKE_Q(wake_q);
1024
1025         ns = current->nsproxy->ipc_ns;
1026
1027         if (msqid < 0 || (long) bufsz < 0)
1028                 return -EINVAL;
1029
1030         if (msgflg & MSG_COPY) {
1031                 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1032                         return -EINVAL;
1033                 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1034                 if (IS_ERR(copy))
1035                         return PTR_ERR(copy);
1036         }
1037         mode = convert_mode(&msgtyp, msgflg);
1038
1039         rcu_read_lock();
1040         msq = msq_obtain_object_check(ns, msqid);
1041         if (IS_ERR(msq)) {
1042                 rcu_read_unlock();
1043                 free_copy(copy);
1044                 return PTR_ERR(msq);
1045         }
1046
1047         for (;;) {
1048                 struct msg_receiver msr_d;
1049
1050                 msg = ERR_PTR(-EACCES);
1051                 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1052                         goto out_unlock1;
1053
1054                 ipc_lock_object(&msq->q_perm);
1055
1056                 /* raced with RMID? */
1057                 if (!ipc_valid_object(&msq->q_perm)) {
1058                         msg = ERR_PTR(-EIDRM);
1059                         goto out_unlock0;
1060                 }
1061
1062                 msg = find_msg(msq, &msgtyp, mode);
1063                 if (!IS_ERR(msg)) {
1064                         /*
1065                          * Found a suitable message.
1066                          * Unlink it from the queue.
1067                          */
1068                         if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1069                                 msg = ERR_PTR(-E2BIG);
1070                                 goto out_unlock0;
1071                         }
1072                         /*
1073                          * If we are copying, then do not unlink message and do
1074                          * not update queue parameters.
1075                          */
1076                         if (msgflg & MSG_COPY) {
1077                                 msg = copy_msg(msg, copy);
1078                                 goto out_unlock0;
1079                         }
1080
1081                         list_del(&msg->m_list);
1082                         msq->q_qnum--;
1083                         msq->q_rtime = get_seconds();
1084                         ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1085                         msq->q_cbytes -= msg->m_ts;
1086                         atomic_sub(msg->m_ts, &ns->msg_bytes);
1087                         atomic_dec(&ns->msg_hdrs);
1088                         ss_wakeup(msq, &wake_q, false);
1089
1090                         goto out_unlock0;
1091                 }
1092
1093                 /* No message waiting. Wait for a message */
1094                 if (msgflg & IPC_NOWAIT) {
1095                         msg = ERR_PTR(-ENOMSG);
1096                         goto out_unlock0;
1097                 }
1098
1099                 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1100                 msr_d.r_tsk = current;
1101                 msr_d.r_msgtype = msgtyp;
1102                 msr_d.r_mode = mode;
1103                 if (msgflg & MSG_NOERROR)
1104                         msr_d.r_maxsize = INT_MAX;
1105                 else
1106                         msr_d.r_maxsize = bufsz;
1107                 msr_d.r_msg = ERR_PTR(-EAGAIN);
1108                 __set_current_state(TASK_INTERRUPTIBLE);
1109
1110                 ipc_unlock_object(&msq->q_perm);
1111                 rcu_read_unlock();
1112                 schedule();
1113
1114                 /*
1115                  * Lockless receive, part 1:
1116                  * We don't hold a reference to the queue and getting a
1117                  * reference would defeat the idea of a lockless operation,
1118                  * thus the code relies on rcu to guarantee the existence of
1119                  * msq:
1120                  * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1121                  * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1122                  */
1123                 rcu_read_lock();
1124
1125                 /*
1126                  * Lockless receive, part 2:
1127                  * The work in pipelined_send() and expunge_all():
1128                  * - Set pointer to message
1129                  * - Queue the receiver task for later wakeup
1130                  * - Wake up the process after the lock is dropped.
1131                  *
1132                  * Should the process wake up before this wakeup (due to a
1133                  * signal) it will either see the message and continue ...
1134                  */
1135                 msg = READ_ONCE(msr_d.r_msg);
1136                 if (msg != ERR_PTR(-EAGAIN))
1137                         goto out_unlock1;
1138
1139                  /*
1140                   * ... or see -EAGAIN, acquire the lock to check the message
1141                   * again.
1142                   */
1143                 ipc_lock_object(&msq->q_perm);
1144
1145                 msg = msr_d.r_msg;
1146                 if (msg != ERR_PTR(-EAGAIN))
1147                         goto out_unlock0;
1148
1149                 list_del(&msr_d.r_list);
1150                 if (signal_pending(current)) {
1151                         msg = ERR_PTR(-ERESTARTNOHAND);
1152                         goto out_unlock0;
1153                 }
1154
1155                 ipc_unlock_object(&msq->q_perm);
1156         }
1157
1158 out_unlock0:
1159         ipc_unlock_object(&msq->q_perm);
1160         wake_up_q(&wake_q);
1161 out_unlock1:
1162         rcu_read_unlock();
1163         if (IS_ERR(msg)) {
1164                 free_copy(copy);
1165                 return PTR_ERR(msg);
1166         }
1167
1168         bufsz = msg_handler(buf, msg, bufsz);
1169         free_msg(msg);
1170
1171         return bufsz;
1172 }
1173
1174 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1175                  long msgtyp, int msgflg)
1176 {
1177         return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1178 }
1179
1180 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1181                 long, msgtyp, int, msgflg)
1182 {
1183         return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1184 }
1185
1186 #ifdef CONFIG_COMPAT
1187 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1188 {
1189         struct compat_msgbuf __user *msgp = dest;
1190         size_t msgsz;
1191
1192         if (put_user(msg->m_type, &msgp->mtype))
1193                 return -EFAULT;
1194
1195         msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1196         if (store_msg(msgp->mtext, msg, msgsz))
1197                 return -EFAULT;
1198         return msgsz;
1199 }
1200
1201 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1202                         compat_long_t msgtyp, int msgflg)
1203 {
1204         return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1205                          msgflg, compat_do_msg_fill);
1206 }
1207
1208 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1209                        compat_ssize_t, msgsz, compat_long_t, msgtyp,
1210                        int, msgflg)
1211 {
1212         return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1213 }
1214 #endif
1215
1216 int msg_init_ns(struct ipc_namespace *ns)
1217 {
1218         ns->msg_ctlmax = MSGMAX;
1219         ns->msg_ctlmnb = MSGMNB;
1220         ns->msg_ctlmni = MSGMNI;
1221
1222         atomic_set(&ns->msg_bytes, 0);
1223         atomic_set(&ns->msg_hdrs, 0);
1224         return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1225 }
1226
1227 #ifdef CONFIG_IPC_NS
1228 void msg_exit_ns(struct ipc_namespace *ns)
1229 {
1230         free_ipcs(ns, &msg_ids(ns), freeque);
1231         idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1232         rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1233 }
1234 #endif
1235
1236 #ifdef CONFIG_PROC_FS
1237 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1238 {
1239         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1240         struct user_namespace *user_ns = seq_user_ns(s);
1241         struct kern_ipc_perm *ipcp = it;
1242         struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1243
1244         seq_printf(s,
1245                    "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1246                    msq->q_perm.key,
1247                    msq->q_perm.id,
1248                    msq->q_perm.mode,
1249                    msq->q_cbytes,
1250                    msq->q_qnum,
1251                    pid_nr_ns(msq->q_lspid, pid_ns),
1252                    pid_nr_ns(msq->q_lrpid, pid_ns),
1253                    from_kuid_munged(user_ns, msq->q_perm.uid),
1254                    from_kgid_munged(user_ns, msq->q_perm.gid),
1255                    from_kuid_munged(user_ns, msq->q_perm.cuid),
1256                    from_kgid_munged(user_ns, msq->q_perm.cgid),
1257                    msq->q_stime,
1258                    msq->q_rtime,
1259                    msq->q_ctime);
1260
1261         return 0;
1262 }
1263 #endif
1264
1265 int __init msg_init(void)
1266 {
1267         const int err = msg_init_ns(&init_ipc_ns);
1268
1269         ipc_init_proc_interface("sysvipc/msg",
1270                                 "       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1271                                 IPC_MSG_IDS, sysvipc_msg_proc_show);
1272         return err;
1273 }