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