linux/net/socket.c
<<
>>
Prefs
   1/*
   2 * NET          An implementation of the SOCKET network access protocol.
   3 *
   4 * Version:     @(#)socket.c    1.1.93  18/02/95
   5 *
   6 * Authors:     Orest Zborowski, <obz@Kodak.COM>
   7 *              Ross Biro
   8 *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
   9 *
  10 * Fixes:
  11 *              Anonymous       :       NOTSOCK/BADF cleanup. Error fix in
  12 *                                      shutdown()
  13 *              Alan Cox        :       verify_area() fixes
  14 *              Alan Cox        :       Removed DDI
  15 *              Jonathan Kamens :       SOCK_DGRAM reconnect bug
  16 *              Alan Cox        :       Moved a load of checks to the very
  17 *                                      top level.
  18 *              Alan Cox        :       Move address structures to/from user
  19 *                                      mode above the protocol layers.
  20 *              Rob Janssen     :       Allow 0 length sends.
  21 *              Alan Cox        :       Asynchronous I/O support (cribbed from the
  22 *                                      tty drivers).
  23 *              Niibe Yutaka    :       Asynchronous I/O for writes (4.4BSD style)
  24 *              Jeff Uphoff     :       Made max number of sockets command-line
  25 *                                      configurable.
  26 *              Matti Aarnio    :       Made the number of sockets dynamic,
  27 *                                      to be allocated when needed, and mr.
  28 *                                      Uphoff's max is used as max to be
  29 *                                      allowed to allocate.
  30 *              Linus           :       Argh. removed all the socket allocation
  31 *                                      altogether: it's in the inode now.
  32 *              Alan Cox        :       Made sock_alloc()/sock_release() public
  33 *                                      for NetROM and future kernel nfsd type
  34 *                                      stuff.
  35 *              Alan Cox        :       sendmsg/recvmsg basics.
  36 *              Tom Dyas        :       Export net symbols.
  37 *              Marcin Dalecki  :       Fixed problems with CONFIG_NET="n".
  38 *              Alan Cox        :       Added thread locking to sys_* calls
  39 *                                      for sockets. May have errors at the
  40 *                                      moment.
  41 *              Kevin Buhr      :       Fixed the dumb errors in the above.
  42 *              Andi Kleen      :       Some small cleanups, optimizations,
  43 *                                      and fixed a copy_from_user() bug.
  44 *              Tigran Aivazian :       sys_send(args) calls sys_sendto(args, NULL, 0)
  45 *              Tigran Aivazian :       Made listen(2) backlog sanity checks
  46 *                                      protocol-independent
  47 *
  48 *
  49 *              This program is free software; you can redistribute it and/or
  50 *              modify it under the terms of the GNU General Public License
  51 *              as published by the Free Software Foundation; either version
  52 *              2 of the License, or (at your option) any later version.
  53 *
  54 *
  55 *      This module is effectively the top level interface to the BSD socket
  56 *      paradigm.
  57 *
  58 *      Based upon Swansea University Computer Society NET3.039
  59 */
  60
  61#include <linux/mm.h>
  62#include <linux/socket.h>
  63#include <linux/file.h>
  64#include <linux/net.h>
  65#include <linux/interrupt.h>
  66#include <linux/thread_info.h>
  67#include <linux/rcupdate.h>
  68#include <linux/netdevice.h>
  69#include <linux/proc_fs.h>
  70#include <linux/seq_file.h>
  71#include <linux/mutex.h>
  72#include <linux/wanrouter.h>
  73#include <linux/if_bridge.h>
  74#include <linux/if_frad.h>
  75#include <linux/if_vlan.h>
  76#include <linux/init.h>
  77#include <linux/poll.h>
  78#include <linux/cache.h>
  79#include <linux/module.h>
  80#include <linux/highmem.h>
  81#include <linux/mount.h>
  82#include <linux/security.h>
  83#include <linux/syscalls.h>
  84#include <linux/compat.h>
  85#include <linux/kmod.h>
  86#include <linux/audit.h>
  87#include <linux/wireless.h>
  88#include <linux/nsproxy.h>
  89#include <linux/magic.h>
  90#include <linux/slab.h>
  91
  92#include <asm/uaccess.h>
  93#include <asm/unistd.h>
  94
  95#include <net/compat.h>
  96#include <net/wext.h>
  97#include <net/cls_cgroup.h>
  98
  99#include <net/sock.h>
 100#include <linux/netfilter.h>
 101
 102#include <linux/if_tun.h>
 103#include <linux/ipv6_route.h>
 104#include <linux/route.h>
 105#include <linux/sockios.h>
 106#include <linux/atalk.h>
 107
 108static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
 109static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
 110                         unsigned long nr_segs, loff_t pos);
 111static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
 112                          unsigned long nr_segs, loff_t pos);
 113static int sock_mmap(struct file *file, struct vm_area_struct *vma);
 114
 115static int sock_close(struct inode *inode, struct file *file);
 116static unsigned int sock_poll(struct file *file,
 117                              struct poll_table_struct *wait);
 118static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 119#ifdef CONFIG_COMPAT
 120static long compat_sock_ioctl(struct file *file,
 121                              unsigned int cmd, unsigned long arg);
 122#endif
 123static int sock_fasync(int fd, struct file *filp, int on);
 124static ssize_t sock_sendpage(struct file *file, struct page *page,
 125                             int offset, size_t size, loff_t *ppos, int more);
 126static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 127                                struct pipe_inode_info *pipe, size_t len,
 128                                unsigned int flags);
 129
 130/*
 131 *      Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
 132 *      in the operation structures but are done directly via the socketcall() multiplexor.
 133 */
 134
 135static const struct file_operations socket_file_ops = {
 136        .owner =        THIS_MODULE,
 137        .llseek =       no_llseek,
 138        .aio_read =     sock_aio_read,
 139        .aio_write =    sock_aio_write,
 140        .poll =         sock_poll,
 141        .unlocked_ioctl = sock_ioctl,
 142#ifdef CONFIG_COMPAT
 143        .compat_ioctl = compat_sock_ioctl,
 144#endif
 145        .mmap =         sock_mmap,
 146        .open =         sock_no_open,   /* special open code to disallow open via /proc */
 147        .release =      sock_close,
 148        .fasync =       sock_fasync,
 149        .sendpage =     sock_sendpage,
 150        .splice_write = generic_splice_sendpage,
 151        .splice_read =  sock_splice_read,
 152};
 153
 154/*
 155 *      The protocol list. Each protocol is registered in here.
 156 */
 157
 158static DEFINE_SPINLOCK(net_family_lock);
 159static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
 160
 161/*
 162 *      Statistics counters of the socket lists
 163 */
 164
 165static DEFINE_PER_CPU(int, sockets_in_use);
 166
 167/*
 168 * Support routines.
 169 * Move socket addresses back and forth across the kernel/user
 170 * divide and look after the messy bits.
 171 */
 172
 173/**
 174 *      move_addr_to_kernel     -       copy a socket address into kernel space
 175 *      @uaddr: Address in user space
 176 *      @kaddr: Address in kernel space
 177 *      @ulen: Length in user space
 178 *
 179 *      The address is copied into kernel space. If the provided address is
 180 *      too long an error code of -EINVAL is returned. If the copy gives
 181 *      invalid addresses -EFAULT is returned. On a success 0 is returned.
 182 */
 183
 184int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
 185{
 186        if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
 187                return -EINVAL;
 188        if (ulen == 0)
 189                return 0;
 190        if (copy_from_user(kaddr, uaddr, ulen))
 191                return -EFAULT;
 192        return audit_sockaddr(ulen, kaddr);
 193}
 194
 195/**
 196 *      move_addr_to_user       -       copy an address to user space
 197 *      @kaddr: kernel space address
 198 *      @klen: length of address in kernel
 199 *      @uaddr: user space address
 200 *      @ulen: pointer to user length field
 201 *
 202 *      The value pointed to by ulen on entry is the buffer length available.
 203 *      This is overwritten with the buffer space used. -EINVAL is returned
 204 *      if an overlong buffer is specified or a negative buffer size. -EFAULT
 205 *      is returned if either the buffer or the length field are not
 206 *      accessible.
 207 *      After copying the data up to the limit the user specifies, the true
 208 *      length of the data is written over the length limit the user
 209 *      specified. Zero is returned for a success.
 210 */
 211
 212static int move_addr_to_user(struct sockaddr *kaddr, int klen,
 213                             void __user *uaddr, int __user *ulen)
 214{
 215        int err;
 216        int len;
 217
 218        err = get_user(len, ulen);
 219        if (err)
 220                return err;
 221        if (len > klen)
 222                len = klen;
 223        if (len < 0 || len > sizeof(struct sockaddr_storage))
 224                return -EINVAL;
 225        if (len) {
 226                if (audit_sockaddr(klen, kaddr))
 227                        return -ENOMEM;
 228                if (copy_to_user(uaddr, kaddr, len))
 229                        return -EFAULT;
 230        }
 231        /*
 232         *      "fromlen shall refer to the value before truncation.."
 233         *                      1003.1g
 234         */
 235        return __put_user(klen, ulen);
 236}
 237
 238static struct kmem_cache *sock_inode_cachep __read_mostly;
 239
 240static struct inode *sock_alloc_inode(struct super_block *sb)
 241{
 242        struct socket_alloc *ei;
 243        struct socket_wq *wq;
 244
 245        ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
 246        if (!ei)
 247                return NULL;
 248        wq = kmalloc(sizeof(*wq), GFP_KERNEL);
 249        if (!wq) {
 250                kmem_cache_free(sock_inode_cachep, ei);
 251                return NULL;
 252        }
 253        init_waitqueue_head(&wq->wait);
 254        wq->fasync_list = NULL;
 255        RCU_INIT_POINTER(ei->socket.wq, wq);
 256
 257        ei->socket.state = SS_UNCONNECTED;
 258        ei->socket.flags = 0;
 259        ei->socket.ops = NULL;
 260        ei->socket.sk = NULL;
 261        ei->socket.file = NULL;
 262
 263        return &ei->vfs_inode;
 264}
 265
 266
 267
 268static void wq_free_rcu(struct rcu_head *head)
 269{
 270        struct socket_wq *wq = container_of(head, struct socket_wq, rcu);
 271
 272        kfree(wq);
 273}
 274
 275static void sock_destroy_inode(struct inode *inode)
 276{
 277        struct socket_alloc *ei;
 278        struct socket_wq *wq;
 279
 280        ei = container_of(inode, struct socket_alloc, vfs_inode);
 281        wq = rcu_dereference_protected(ei->socket.wq, 1);
 282        call_rcu(&wq->rcu, wq_free_rcu);
 283        kmem_cache_free(sock_inode_cachep, ei);
 284}
 285
 286static void init_once(void *foo)
 287{
 288        struct socket_alloc *ei = (struct socket_alloc *)foo;
 289
 290        inode_init_once(&ei->vfs_inode);
 291}
 292
 293static int init_inodecache(void)
 294{
 295        sock_inode_cachep = kmem_cache_create("sock_inode_cache",
 296                                              sizeof(struct socket_alloc),
 297                                              0,
 298                                              (SLAB_HWCACHE_ALIGN |
 299                                               SLAB_RECLAIM_ACCOUNT |
 300                                               SLAB_MEM_SPREAD),
 301                                              init_once);
 302        if (sock_inode_cachep == NULL)
 303                return -ENOMEM;
 304        return 0;
 305}
 306
 307static const struct super_operations sockfs_ops = {
 308        .alloc_inode    = sock_alloc_inode,
 309        .destroy_inode  = sock_destroy_inode,
 310        .statfs         = simple_statfs,
 311};
 312
 313/*
 314 * sockfs_dname() is called from d_path().
 315 */
 316static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
 317{
 318        return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
 319                                dentry->d_inode->i_ino);
 320}
 321
 322static const struct dentry_operations sockfs_dentry_operations = {
 323        .d_dname  = sockfs_dname,
 324};
 325
 326static struct dentry *sockfs_mount(struct file_system_type *fs_type,
 327                         int flags, const char *dev_name, void *data)
 328{
 329        return mount_pseudo(fs_type, "socket:", &sockfs_ops,
 330                &sockfs_dentry_operations, SOCKFS_MAGIC);
 331}
 332
 333static struct vfsmount *sock_mnt __read_mostly;
 334
 335static struct file_system_type sock_fs_type = {
 336        .name =         "sockfs",
 337        .mount =        sockfs_mount,
 338        .kill_sb =      kill_anon_super,
 339};
 340
 341/*
 342 *      Obtains the first available file descriptor and sets it up for use.
 343 *
 344 *      These functions create file structures and maps them to fd space
 345 *      of the current process. On success it returns file descriptor
 346 *      and file struct implicitly stored in sock->file.
 347 *      Note that another thread may close file descriptor before we return
 348 *      from this function. We use the fact that now we do not refer
 349 *      to socket after mapping. If one day we will need it, this
 350 *      function will increment ref. count on file by 1.
 351 *
 352 *      In any case returned fd MAY BE not valid!
 353 *      This race condition is unavoidable
 354 *      with shared fd spaces, we cannot solve it inside kernel,
 355 *      but we take care of internal coherence yet.
 356 */
 357
 358static int sock_alloc_file(struct socket *sock, struct file **f, int flags)
 359{
 360        struct qstr name = { .name = "" };
 361        struct path path;
 362        struct file *file;
 363        int fd;
 364
 365        fd = get_unused_fd_flags(flags);
 366        if (unlikely(fd < 0))
 367                return fd;
 368
 369        path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
 370        if (unlikely(!path.dentry)) {
 371                put_unused_fd(fd);
 372                return -ENOMEM;
 373        }
 374        path.mnt = mntget(sock_mnt);
 375
 376        d_instantiate(path.dentry, SOCK_INODE(sock));
 377        SOCK_INODE(sock)->i_fop = &socket_file_ops;
 378
 379        file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
 380                  &socket_file_ops);
 381        if (unlikely(!file)) {
 382                /* drop dentry, keep inode */
 383                ihold(path.dentry->d_inode);
 384                path_put(&path);
 385                put_unused_fd(fd);
 386                return -ENFILE;
 387        }
 388
 389        sock->file = file;
 390        file->f_flags = O_RDWR | (flags & O_NONBLOCK);
 391        file->f_pos = 0;
 392        file->private_data = sock;
 393
 394        *f = file;
 395        return fd;
 396}
 397
 398int sock_map_fd(struct socket *sock, int flags)
 399{
 400        struct file *newfile;
 401        int fd = sock_alloc_file(sock, &newfile, flags);
 402
 403        if (likely(fd >= 0))
 404                fd_install(fd, newfile);
 405
 406        return fd;
 407}
 408EXPORT_SYMBOL(sock_map_fd);
 409
 410static struct socket *sock_from_file(struct file *file, int *err)
 411{
 412        if (file->f_op == &socket_file_ops)
 413                return file->private_data;      /* set in sock_map_fd */
 414
 415        *err = -ENOTSOCK;
 416        return NULL;
 417}
 418
 419/**
 420 *      sockfd_lookup - Go from a file number to its socket slot
 421 *      @fd: file handle
 422 *      @err: pointer to an error code return
 423 *
 424 *      The file handle passed in is locked and the socket it is bound
 425 *      too is returned. If an error occurs the err pointer is overwritten
 426 *      with a negative errno code and NULL is returned. The function checks
 427 *      for both invalid handles and passing a handle which is not a socket.
 428 *
 429 *      On a success the socket object pointer is returned.
 430 */
 431
 432struct socket *sockfd_lookup(int fd, int *err)
 433{
 434        struct file *file;
 435        struct socket *sock;
 436
 437        file = fget(fd);
 438        if (!file) {
 439                *err = -EBADF;
 440                return NULL;
 441        }
 442
 443        sock = sock_from_file(file, err);
 444        if (!sock)
 445                fput(file);
 446        return sock;
 447}
 448EXPORT_SYMBOL(sockfd_lookup);
 449
 450static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
 451{
 452        struct file *file;
 453        struct socket *sock;
 454
 455        *err = -EBADF;
 456        file = fget_light(fd, fput_needed);
 457        if (file) {
 458                sock = sock_from_file(file, err);
 459                if (sock)
 460                        return sock;
 461                fput_light(file, *fput_needed);
 462        }
 463        return NULL;
 464}
 465
 466/**
 467 *      sock_alloc      -       allocate a socket
 468 *
 469 *      Allocate a new inode and socket object. The two are bound together
 470 *      and initialised. The socket is then returned. If we are out of inodes
 471 *      NULL is returned.
 472 */
 473
 474static struct socket *sock_alloc(void)
 475{
 476        struct inode *inode;
 477        struct socket *sock;
 478
 479        inode = new_inode(sock_mnt->mnt_sb);
 480        if (!inode)
 481                return NULL;
 482
 483        sock = SOCKET_I(inode);
 484
 485        kmemcheck_annotate_bitfield(sock, type);
 486        inode->i_ino = get_next_ino();
 487        inode->i_mode = S_IFSOCK | S_IRWXUGO;
 488        inode->i_uid = current_fsuid();
 489        inode->i_gid = current_fsgid();
 490
 491        percpu_add(sockets_in_use, 1);
 492        return sock;
 493}
 494
 495/*
 496 *      In theory you can't get an open on this inode, but /proc provides
 497 *      a back door. Remember to keep it shut otherwise you'll let the
 498 *      creepy crawlies in.
 499 */
 500
 501static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
 502{
 503        return -ENXIO;
 504}
 505
 506const struct file_operations bad_sock_fops = {
 507        .owner = THIS_MODULE,
 508        .open = sock_no_open,
 509        .llseek = noop_llseek,
 510};
 511
 512/**
 513 *      sock_release    -       close a socket
 514 *      @sock: socket to close
 515 *
 516 *      The socket is released from the protocol stack if it has a release
 517 *      callback, and the inode is then released if the socket is bound to
 518 *      an inode not a file.
 519 */
 520
 521void sock_release(struct socket *sock)
 522{
 523        if (sock->ops) {
 524                struct module *owner = sock->ops->owner;
 525
 526                sock->ops->release(sock);
 527                sock->ops = NULL;
 528                module_put(owner);
 529        }
 530
 531        if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
 532                printk(KERN_ERR "sock_release: fasync list not empty!\n");
 533
 534        percpu_sub(sockets_in_use, 1);
 535        if (!sock->file) {
 536                iput(SOCK_INODE(sock));
 537                return;
 538        }
 539        sock->file = NULL;
 540}
 541EXPORT_SYMBOL(sock_release);
 542
 543int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags)
 544{
 545        *tx_flags = 0;
 546        if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 547                *tx_flags |= SKBTX_HW_TSTAMP;
 548        if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 549                *tx_flags |= SKBTX_SW_TSTAMP;
 550        return 0;
 551}
 552EXPORT_SYMBOL(sock_tx_timestamp);
 553
 554static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
 555                                 struct msghdr *msg, size_t size)
 556{
 557        struct sock_iocb *si = kiocb_to_siocb(iocb);
 558        int err;
 559
 560        sock_update_classid(sock->sk);
 561
 562        si->sock = sock;
 563        si->scm = NULL;
 564        si->msg = msg;
 565        si->size = size;
 566
 567        err = security_socket_sendmsg(sock, msg, size);
 568        if (err)
 569                return err;
 570
 571        return sock->ops->sendmsg(iocb, sock, msg, size);
 572}
 573
 574int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 575{
 576        struct kiocb iocb;
 577        struct sock_iocb siocb;
 578        int ret;
 579
 580        init_sync_kiocb(&iocb, NULL);
 581        iocb.private = &siocb;
 582        ret = __sock_sendmsg(&iocb, sock, msg, size);
 583        if (-EIOCBQUEUED == ret)
 584                ret = wait_on_sync_kiocb(&iocb);
 585        return ret;
 586}
 587EXPORT_SYMBOL(sock_sendmsg);
 588
 589int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
 590                   struct kvec *vec, size_t num, size_t size)
 591{
 592        mm_segment_t oldfs = get_fs();
 593        int result;
 594
 595        set_fs(KERNEL_DS);
 596        /*
 597         * the following is safe, since for compiler definitions of kvec and
 598         * iovec are identical, yielding the same in-core layout and alignment
 599         */
 600        msg->msg_iov = (struct iovec *)vec;
 601        msg->msg_iovlen = num;
 602        result = sock_sendmsg(sock, msg, size);
 603        set_fs(oldfs);
 604        return result;
 605}
 606EXPORT_SYMBOL(kernel_sendmsg);
 607
 608static int ktime2ts(ktime_t kt, struct timespec *ts)
 609{
 610        if (kt.tv64) {
 611                *ts = ktime_to_timespec(kt);
 612                return 1;
 613        } else {
 614                return 0;
 615        }
 616}
 617
 618/*
 619 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
 620 */
 621void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 622        struct sk_buff *skb)
 623{
 624        int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
 625        struct timespec ts[3];
 626        int empty = 1;
 627        struct skb_shared_hwtstamps *shhwtstamps =
 628                skb_hwtstamps(skb);
 629
 630        /* Race occurred between timestamp enabling and packet
 631           receiving.  Fill in the current time for now. */
 632        if (need_software_tstamp && skb->tstamp.tv64 == 0)
 633                __net_timestamp(skb);
 634
 635        if (need_software_tstamp) {
 636                if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
 637                        struct timeval tv;
 638                        skb_get_timestamp(skb, &tv);
 639                        put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 640                                 sizeof(tv), &tv);
 641                } else {
 642                        skb_get_timestampns(skb, &ts[0]);
 643                        put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
 644                                 sizeof(ts[0]), &ts[0]);
 645                }
 646        }
 647
 648
 649        memset(ts, 0, sizeof(ts));
 650        if (skb->tstamp.tv64 &&
 651            sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
 652                skb_get_timestampns(skb, ts + 0);
 653                empty = 0;
 654        }
 655        if (shhwtstamps) {
 656                if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
 657                    ktime2ts(shhwtstamps->syststamp, ts + 1))
 658                        empty = 0;
 659                if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
 660                    ktime2ts(shhwtstamps->hwtstamp, ts + 2))
 661                        empty = 0;
 662        }
 663        if (!empty)
 664                put_cmsg(msg, SOL_SOCKET,
 665                         SCM_TIMESTAMPING, sizeof(ts), &ts);
 666}
 667EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
 668
 669static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
 670                                   struct sk_buff *skb)
 671{
 672        if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
 673                put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
 674                        sizeof(__u32), &skb->dropcount);
 675}
 676
 677void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
 678        struct sk_buff *skb)
 679{
 680        sock_recv_timestamp(msg, sk, skb);
 681        sock_recv_drops(msg, sk, skb);
 682}
 683EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
 684
 685static inline int __sock_recvmsg_nosec(struct kiocb *iocb, struct socket *sock,
 686                                       struct msghdr *msg, size_t size, int flags)
 687{
 688        struct sock_iocb *si = kiocb_to_siocb(iocb);
 689
 690        sock_update_classid(sock->sk);
 691
 692        si->sock = sock;
 693        si->scm = NULL;
 694        si->msg = msg;
 695        si->size = size;
 696        si->flags = flags;
 697
 698        return sock->ops->recvmsg(iocb, sock, msg, size, flags);
 699}
 700
 701static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
 702                                 struct msghdr *msg, size_t size, int flags)
 703{
 704        int err = security_socket_recvmsg(sock, msg, size, flags);
 705
 706        return err ?: __sock_recvmsg_nosec(iocb, sock, msg, size, flags);
 707}
 708
 709int sock_recvmsg(struct socket *sock, struct msghdr *msg,
 710                 size_t size, int flags)
 711{
 712        struct kiocb iocb;
 713        struct sock_iocb siocb;
 714        int ret;
 715
 716        init_sync_kiocb(&iocb, NULL);
 717        iocb.private = &siocb;
 718        ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
 719        if (-EIOCBQUEUED == ret)
 720                ret = wait_on_sync_kiocb(&iocb);
 721        return ret;
 722}
 723EXPORT_SYMBOL(sock_recvmsg);
 724
 725static int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
 726                              size_t size, int flags)
 727{
 728        struct kiocb iocb;
 729        struct sock_iocb siocb;
 730        int ret;
 731
 732        init_sync_kiocb(&iocb, NULL);
 733        iocb.private = &siocb;
 734        ret = __sock_recvmsg_nosec(&iocb, sock, msg, size, flags);
 735        if (-EIOCBQUEUED == ret)
 736                ret = wait_on_sync_kiocb(&iocb);
 737        return ret;
 738}
 739
 740/**
 741 * kernel_recvmsg - Receive a message from a socket (kernel space)
 742 * @sock:       The socket to receive the message from
 743 * @msg:        Received message
 744 * @vec:        Input s/g array for message data
 745 * @num:        Size of input s/g array
 746 * @size:       Number of bytes to read
 747 * @flags:      Message flags (MSG_DONTWAIT, etc...)
 748 *
 749 * On return the msg structure contains the scatter/gather array passed in the
 750 * vec argument. The array is modified so that it consists of the unfilled
 751 * portion of the original array.
 752 *
 753 * The returned value is the total number of bytes received, or an error.
 754 */
 755int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
 756                   struct kvec *vec, size_t num, size_t size, int flags)
 757{
 758        mm_segment_t oldfs = get_fs();
 759        int result;
 760
 761        set_fs(KERNEL_DS);
 762        /*
 763         * the following is safe, since for compiler definitions of kvec and
 764         * iovec are identical, yielding the same in-core layout and alignment
 765         */
 766        msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
 767        result = sock_recvmsg(sock, msg, size, flags);
 768        set_fs(oldfs);
 769        return result;
 770}
 771EXPORT_SYMBOL(kernel_recvmsg);
 772
 773static void sock_aio_dtor(struct kiocb *iocb)
 774{
 775        kfree(iocb->private);
 776}
 777
 778static ssize_t sock_sendpage(struct file *file, struct page *page,
 779                             int offset, size_t size, loff_t *ppos, int more)
 780{
 781        struct socket *sock;
 782        int flags;
 783
 784        sock = file->private_data;
 785
 786        flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
 787        if (more)
 788                flags |= MSG_MORE;
 789
 790        return kernel_sendpage(sock, page, offset, size, flags);
 791}
 792
 793static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 794                                struct pipe_inode_info *pipe, size_t len,
 795                                unsigned int flags)
 796{
 797        struct socket *sock = file->private_data;
 798
 799        if (unlikely(!sock->ops->splice_read))
 800                return -EINVAL;
 801
 802        sock_update_classid(sock->sk);
 803
 804        return sock->ops->splice_read(sock, ppos, pipe, len, flags);
 805}
 806
 807static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
 808                                         struct sock_iocb *siocb)
 809{
 810        if (!is_sync_kiocb(iocb)) {
 811                siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
 812                if (!siocb)
 813                        return NULL;
 814                iocb->ki_dtor = sock_aio_dtor;
 815        }
 816
 817        siocb->kiocb = iocb;
 818        iocb->private = siocb;
 819        return siocb;
 820}
 821
 822static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
 823                struct file *file, const struct iovec *iov,
 824                unsigned long nr_segs)
 825{
 826        struct socket *sock = file->private_data;
 827        size_t size = 0;
 828        int i;
 829
 830        for (i = 0; i < nr_segs; i++)
 831                size += iov[i].iov_len;
 832
 833        msg->msg_name = NULL;
 834        msg->msg_namelen = 0;
 835        msg->msg_control = NULL;
 836        msg->msg_controllen = 0;
 837        msg->msg_iov = (struct iovec *)iov;
 838        msg->msg_iovlen = nr_segs;
 839        msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 840
 841        return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
 842}
 843
 844static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
 845                                unsigned long nr_segs, loff_t pos)
 846{
 847        struct sock_iocb siocb, *x;
 848
 849        if (pos != 0)
 850                return -ESPIPE;
 851
 852        if (iocb->ki_left == 0) /* Match SYS5 behaviour */
 853                return 0;
 854
 855
 856        x = alloc_sock_iocb(iocb, &siocb);
 857        if (!x)
 858                return -ENOMEM;
 859        return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
 860}
 861
 862static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
 863                        struct file *file, const struct iovec *iov,
 864                        unsigned long nr_segs)
 865{
 866        struct socket *sock = file->private_data;
 867        size_t size = 0;
 868        int i;
 869
 870        for (i = 0; i < nr_segs; i++)
 871                size += iov[i].iov_len;
 872
 873        msg->msg_name = NULL;
 874        msg->msg_namelen = 0;
 875        msg->msg_control = NULL;
 876        msg->msg_controllen = 0;
 877        msg->msg_iov = (struct iovec *)iov;
 878        msg->msg_iovlen = nr_segs;
 879        msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 880        if (sock->type == SOCK_SEQPACKET)
 881                msg->msg_flags |= MSG_EOR;
 882
 883        return __sock_sendmsg(iocb, sock, msg, size);
 884}
 885
 886static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
 887                          unsigned long nr_segs, loff_t pos)
 888{
 889        struct sock_iocb siocb, *x;
 890
 891        if (pos != 0)
 892                return -ESPIPE;
 893
 894        x = alloc_sock_iocb(iocb, &siocb);
 895        if (!x)
 896                return -ENOMEM;
 897
 898        return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
 899}
 900
 901/*
 902 * Atomic setting of ioctl hooks to avoid race
 903 * with module unload.
 904 */
 905
 906static DEFINE_MUTEX(br_ioctl_mutex);
 907static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
 908
 909void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
 910{
 911        mutex_lock(&br_ioctl_mutex);
 912        br_ioctl_hook = hook;
 913        mutex_unlock(&br_ioctl_mutex);
 914}
 915EXPORT_SYMBOL(brioctl_set);
 916
 917static DEFINE_MUTEX(vlan_ioctl_mutex);
 918static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
 919
 920void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
 921{
 922        mutex_lock(&vlan_ioctl_mutex);
 923        vlan_ioctl_hook = hook;
 924        mutex_unlock(&vlan_ioctl_mutex);
 925}
 926EXPORT_SYMBOL(vlan_ioctl_set);
 927
 928static DEFINE_MUTEX(dlci_ioctl_mutex);
 929static int (*dlci_ioctl_hook) (unsigned int, void __user *);
 930
 931void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
 932{
 933        mutex_lock(&dlci_ioctl_mutex);
 934        dlci_ioctl_hook = hook;
 935        mutex_unlock(&dlci_ioctl_mutex);
 936}
 937EXPORT_SYMBOL(dlci_ioctl_set);
 938
 939static long sock_do_ioctl(struct net *net, struct socket *sock,
 940                                 unsigned int cmd, unsigned long arg)
 941{
 942        int err;
 943        void __user *argp = (void __user *)arg;
 944
 945        err = sock->ops->ioctl(sock, cmd, arg);
 946
 947        /*
 948         * If this ioctl is unknown try to hand it down
 949         * to the NIC driver.
 950         */
 951        if (err == -ENOIOCTLCMD)
 952                err = dev_ioctl(net, cmd, argp);
 953
 954        return err;
 955}
 956
 957/*
 958 *      With an ioctl, arg may well be a user mode pointer, but we don't know
 959 *      what to do with it - that's up to the protocol still.
 960 */
 961
 962static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 963{
 964        struct socket *sock;
 965        struct sock *sk;
 966        void __user *argp = (void __user *)arg;
 967        int pid, err;
 968        struct net *net;
 969
 970        sock = file->private_data;
 971        sk = sock->sk;
 972        net = sock_net(sk);
 973        if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
 974                err = dev_ioctl(net, cmd, argp);
 975        } else
 976#ifdef CONFIG_WEXT_CORE
 977        if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
 978                err = dev_ioctl(net, cmd, argp);
 979        } else
 980#endif
 981                switch (cmd) {
 982                case FIOSETOWN:
 983                case SIOCSPGRP:
 984                        err = -EFAULT;
 985                        if (get_user(pid, (int __user *)argp))
 986                                break;
 987                        err = f_setown(sock->file, pid, 1);
 988                        break;
 989                case FIOGETOWN:
 990                case SIOCGPGRP:
 991                        err = put_user(f_getown(sock->file),
 992                                       (int __user *)argp);
 993                        break;
 994                case SIOCGIFBR:
 995                case SIOCSIFBR:
 996                case SIOCBRADDBR:
 997                case SIOCBRDELBR:
 998                        err = -ENOPKG;
 999                        if (!br_ioctl_hook)
1000                                request_module("bridge");
1001
1002                        mutex_lock(&br_ioctl_mutex);
1003                        if (br_ioctl_hook)
1004                                err = br_ioctl_hook(net, cmd, argp);
1005                        mutex_unlock(&br_ioctl_mutex);
1006                        break;
1007                case SIOCGIFVLAN:
1008                case SIOCSIFVLAN:
1009                        err = -ENOPKG;
1010                        if (!vlan_ioctl_hook)
1011                                request_module("8021q");
1012
1013                        mutex_lock(&vlan_ioctl_mutex);
1014                        if (vlan_ioctl_hook)
1015                                err = vlan_ioctl_hook(net, argp);
1016                        mutex_unlock(&vlan_ioctl_mutex);
1017                        break;
1018                case SIOCADDDLCI:
1019                case SIOCDELDLCI:
1020                        err = -ENOPKG;
1021                        if (!dlci_ioctl_hook)
1022                                request_module("dlci");
1023
1024                        mutex_lock(&dlci_ioctl_mutex);
1025                        if (dlci_ioctl_hook)
1026                                err = dlci_ioctl_hook(cmd, argp);
1027                        mutex_unlock(&dlci_ioctl_mutex);
1028                        break;
1029                default:
1030                        err = sock_do_ioctl(net, sock, cmd, arg);
1031                        break;
1032                }
1033        return err;
1034}
1035
1036int sock_create_lite(int family, int type, int protocol, struct socket **res)
1037{
1038        int err;
1039        struct socket *sock = NULL;
1040
1041        err = security_socket_create(family, type, protocol, 1);
1042        if (err)
1043                goto out;
1044
1045        sock = sock_alloc();
1046        if (!sock) {
1047                err = -ENOMEM;
1048                goto out;
1049        }
1050
1051        sock->type = type;
1052        err = security_socket_post_create(sock, family, type, protocol, 1);
1053        if (err)
1054                goto out_release;
1055
1056out:
1057        *res = sock;
1058        return err;
1059out_release:
1060        sock_release(sock);
1061        sock = NULL;
1062        goto out;
1063}
1064EXPORT_SYMBOL(sock_create_lite);
1065
1066/* No kernel lock held - perfect */
1067static unsigned int sock_poll(struct file *file, poll_table *wait)
1068{
1069        struct socket *sock;
1070
1071        /*
1072         *      We can't return errors to poll, so it's either yes or no.
1073         */
1074        sock = file->private_data;
1075        return sock->ops->poll(file, sock, wait);
1076}
1077
1078static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1079{
1080        struct socket *sock = file->private_data;
1081
1082        return sock->ops->mmap(file, sock, vma);
1083}
1084
1085static int sock_close(struct inode *inode, struct file *filp)
1086{
1087        /*
1088         *      It was possible the inode is NULL we were
1089         *      closing an unfinished socket.
1090         */
1091
1092        if (!inode) {
1093                printk(KERN_DEBUG "sock_close: NULL inode\n");
1094                return 0;
1095        }
1096        sock_release(SOCKET_I(inode));
1097        return 0;
1098}
1099
1100/*
1101 *      Update the socket async list
1102 *
1103 *      Fasync_list locking strategy.
1104 *
1105 *      1. fasync_list is modified only under process context socket lock
1106 *         i.e. under semaphore.
1107 *      2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1108 *         or under socket lock
1109 */
1110
1111static int sock_fasync(int fd, struct file *filp, int on)
1112{
1113        struct socket *sock = filp->private_data;
1114        struct sock *sk = sock->sk;
1115        struct socket_wq *wq;
1116
1117        if (sk == NULL)
1118                return -EINVAL;
1119
1120        lock_sock(sk);
1121        wq = rcu_dereference_protected(sock->wq, sock_owned_by_user(sk));
1122        fasync_helper(fd, filp, on, &wq->fasync_list);
1123
1124        if (!wq->fasync_list)
1125                sock_reset_flag(sk, SOCK_FASYNC);
1126        else
1127                sock_set_flag(sk, SOCK_FASYNC);
1128
1129        release_sock(sk);
1130        return 0;
1131}
1132
1133/* This function may be called only under socket lock or callback_lock or rcu_lock */
1134
1135int sock_wake_async(struct socket *sock, int how, int band)
1136{
1137        struct socket_wq *wq;
1138
1139        if (!sock)
1140                return -1;
1141        rcu_read_lock();
1142        wq = rcu_dereference(sock->wq);
1143        if (!wq || !wq->fasync_list) {
1144                rcu_read_unlock();
1145                return -1;
1146        }
1147        switch (how) {
1148        case SOCK_WAKE_WAITD:
1149                if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1150                        break;
1151                goto call_kill;
1152        case SOCK_WAKE_SPACE:
1153                if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1154                        break;
1155                /* fall through */
1156        case SOCK_WAKE_IO:
1157call_kill:
1158                kill_fasync(&wq->fasync_list, SIGIO, band);
1159                break;
1160        case SOCK_WAKE_URG:
1161                kill_fasync(&wq->fasync_list, SIGURG, band);
1162        }
1163        rcu_read_unlock();
1164        return 0;
1165}
1166EXPORT_SYMBOL(sock_wake_async);
1167
1168int __sock_create(struct net *net, int family, int type, int protocol,
1169                         struct socket **res, int kern)
1170{
1171        int err;
1172        struct socket *sock;
1173        const struct net_proto_family *pf;
1174
1175        /*
1176         *      Check protocol is in range
1177         */
1178        if (family < 0 || family >= NPROTO)
1179                return -EAFNOSUPPORT;
1180        if (type < 0 || type >= SOCK_MAX)
1181                return -EINVAL;
1182
1183        /* Compatibility.
1184
1185           This uglymoron is moved from INET layer to here to avoid
1186           deadlock in module load.
1187         */
1188        if (family == PF_INET && type == SOCK_PACKET) {
1189                static int warned;
1190                if (!warned) {
1191                        warned = 1;
1192                        printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1193                               current->comm);
1194                }
1195                family = PF_PACKET;
1196        }
1197
1198        err = security_socket_create(family, type, protocol, kern);
1199        if (err)
1200                return err;
1201
1202        /*
1203         *      Allocate the socket and allow the family to set things up. if
1204         *      the protocol is 0, the family is instructed to select an appropriate
1205         *      default.
1206         */
1207        sock = sock_alloc();
1208        if (!sock) {
1209                if (net_ratelimit())
1210                        printk(KERN_WARNING "socket: no more sockets\n");
1211                return -ENFILE; /* Not exactly a match, but its the
1212                                   closest posix thing */
1213        }
1214
1215        sock->type = type;
1216
1217#ifdef CONFIG_MODULES
1218        /* Attempt to load a protocol module if the find failed.
1219         *
1220         * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1221         * requested real, full-featured networking support upon configuration.
1222         * Otherwise module support will break!
1223         */
1224        if (rcu_access_pointer(net_families[family]) == NULL)
1225                request_module("net-pf-%d", family);
1226#endif
1227
1228        rcu_read_lock();
1229        pf = rcu_dereference(net_families[family]);
1230        err = -EAFNOSUPPORT;
1231        if (!pf)
1232                goto out_release;
1233
1234        /*
1235         * We will call the ->create function, that possibly is in a loadable
1236         * module, so we have to bump that loadable module refcnt first.
1237         */
1238        if (!try_module_get(pf->owner))
1239                goto out_release;
1240
1241        /* Now protected by module ref count */
1242        rcu_read_unlock();
1243
1244        err = pf->create(net, sock, protocol, kern);
1245        if (err < 0)
1246                goto out_module_put;
1247
1248        /*
1249         * Now to bump the refcnt of the [loadable] module that owns this
1250         * socket at sock_release time we decrement its refcnt.
1251         */
1252        if (!try_module_get(sock->ops->owner))
1253                goto out_module_busy;
1254
1255        /*
1256         * Now that we're done with the ->create function, the [loadable]
1257         * module can have its refcnt decremented
1258         */
1259        module_put(pf->owner);
1260        err = security_socket_post_create(sock, family, type, protocol, kern);
1261        if (err)
1262                goto out_sock_release;
1263        *res = sock;
1264
1265        return 0;
1266
1267out_module_busy:
1268        err = -EAFNOSUPPORT;
1269out_module_put:
1270        sock->ops = NULL;
1271        module_put(pf->owner);
1272out_sock_release:
1273        sock_release(sock);
1274        return err;
1275
1276out_release:
1277        rcu_read_unlock();
1278        goto out_sock_release;
1279}
1280EXPORT_SYMBOL(__sock_create);
1281
1282int sock_create(int family, int type, int protocol, struct socket **res)
1283{
1284        return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1285}
1286EXPORT_SYMBOL(sock_create);
1287
1288int sock_create_kern(int family, int type, int protocol, struct socket **res)
1289{
1290        return __sock_create(&init_net, family, type, protocol, res, 1);
1291}
1292EXPORT_SYMBOL(sock_create_kern);
1293
1294SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1295{
1296        int retval;
1297        struct socket *sock;
1298        int flags;
1299
1300        /* Check the SOCK_* constants for consistency.  */
1301        BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1302        BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1303        BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1304        BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1305
1306        flags = type & ~SOCK_TYPE_MASK;
1307        if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1308                return -EINVAL;
1309        type &= SOCK_TYPE_MASK;
1310
1311        if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1312                flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1313
1314        retval = sock_create(family, type, protocol, &sock);
1315        if (retval < 0)
1316                goto out;
1317
1318        retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1319        if (retval < 0)
1320                goto out_release;
1321
1322out:
1323        /* It may be already another descriptor 8) Not kernel problem. */
1324        return retval;
1325
1326out_release:
1327        sock_release(sock);
1328        return retval;
1329}
1330
1331/*
1332 *      Create a pair of connected sockets.
1333 */
1334
1335SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1336                int __user *, usockvec)
1337{
1338        struct socket *sock1, *sock2;
1339        int fd1, fd2, err;
1340        struct file *newfile1, *newfile2;
1341        int flags;
1342
1343        flags = type & ~SOCK_TYPE_MASK;
1344        if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1345                return -EINVAL;
1346        type &= SOCK_TYPE_MASK;
1347
1348        if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1349                flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1350
1351        /*
1352         * Obtain the first socket and check if the underlying protocol
1353         * supports the socketpair call.
1354         */
1355
1356        err = sock_create(family, type, protocol, &sock1);
1357        if (err < 0)
1358                goto out;
1359
1360        err = sock_create(family, type, protocol, &sock2);
1361        if (err < 0)
1362                goto out_release_1;
1363
1364        err = sock1->ops->socketpair(sock1, sock2);
1365        if (err < 0)
1366                goto out_release_both;
1367
1368        fd1 = sock_alloc_file(sock1, &newfile1, flags);
1369        if (unlikely(fd1 < 0)) {
1370                err = fd1;
1371                goto out_release_both;
1372        }
1373
1374        fd2 = sock_alloc_file(sock2, &newfile2, flags);
1375        if (unlikely(fd2 < 0)) {
1376                err = fd2;
1377                fput(newfile1);
1378                put_unused_fd(fd1);
1379                sock_release(sock2);
1380                goto out;
1381        }
1382
1383        audit_fd_pair(fd1, fd2);
1384        fd_install(fd1, newfile1);
1385        fd_install(fd2, newfile2);
1386        /* fd1 and fd2 may be already another descriptors.
1387         * Not kernel problem.
1388         */
1389
1390        err = put_user(fd1, &usockvec[0]);
1391        if (!err)
1392                err = put_user(fd2, &usockvec[1]);
1393        if (!err)
1394                return 0;
1395
1396        sys_close(fd2);
1397        sys_close(fd1);
1398        return err;
1399
1400out_release_both:
1401        sock_release(sock2);
1402out_release_1:
1403        sock_release(sock1);
1404out:
1405        return err;
1406}
1407
1408/*
1409 *      Bind a name to a socket. Nothing much to do here since it's
1410 *      the protocol's responsibility to handle the local address.
1411 *
1412 *      We move the socket address to kernel space before we call
1413 *      the protocol layer (having also checked the address is ok).
1414 */
1415
1416SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1417{
1418        struct socket *sock;
1419        struct sockaddr_storage address;
1420        int err, fput_needed;
1421
1422        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1423        if (sock) {
1424                err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1425                if (err >= 0) {
1426                        err = security_socket_bind(sock,
1427                                                   (struct sockaddr *)&address,
1428                                                   addrlen);
1429                        if (!err)
1430                                err = sock->ops->bind(sock,
1431                                                      (struct sockaddr *)
1432                                                      &address, addrlen);
1433                }
1434                fput_light(sock->file, fput_needed);
1435        }
1436        return err;
1437}
1438
1439/*
1440 *      Perform a listen. Basically, we allow the protocol to do anything
1441 *      necessary for a listen, and if that works, we mark the socket as
1442 *      ready for listening.
1443 */
1444
1445SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1446{
1447        struct socket *sock;
1448        int err, fput_needed;
1449        int somaxconn;
1450
1451        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1452        if (sock) {
1453                somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1454                if ((unsigned)backlog > somaxconn)
1455                        backlog = somaxconn;
1456
1457                err = security_socket_listen(sock, backlog);
1458                if (!err)
1459                        err = sock->ops->listen(sock, backlog);
1460
1461                fput_light(sock->file, fput_needed);
1462        }
1463        return err;
1464}
1465
1466/*
1467 *      For accept, we attempt to create a new socket, set up the link
1468 *      with the client, wake up the client, then return the new
1469 *      connected fd. We collect the address of the connector in kernel
1470 *      space and move it to user at the very end. This is unclean because
1471 *      we open the socket then return an error.
1472 *
1473 *      1003.1g adds the ability to recvmsg() to query connection pending
1474 *      status to recvmsg. We need to add that support in a way thats
1475 *      clean when we restucture accept also.
1476 */
1477
1478SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1479                int __user *, upeer_addrlen, int, flags)
1480{
1481        struct socket *sock, *newsock;
1482        struct file *newfile;
1483        int err, len, newfd, fput_needed;
1484        struct sockaddr_storage address;
1485
1486        if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1487                return -EINVAL;
1488
1489        if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1490                flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1491
1492        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1493        if (!sock)
1494                goto out;
1495
1496        err = -ENFILE;
1497        newsock = sock_alloc();
1498        if (!newsock)
1499                goto out_put;
1500
1501        newsock->type = sock->type;
1502        newsock->ops = sock->ops;
1503
1504        /*
1505         * We don't need try_module_get here, as the listening socket (sock)
1506         * has the protocol module (sock->ops->owner) held.
1507         */
1508        __module_get(newsock->ops->owner);
1509
1510        newfd = sock_alloc_file(newsock, &newfile, flags);
1511        if (unlikely(newfd < 0)) {
1512                err = newfd;
1513                sock_release(newsock);
1514                goto out_put;
1515        }
1516
1517        err = security_socket_accept(sock, newsock);
1518        if (err)
1519                goto out_fd;
1520
1521        err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1522        if (err < 0)
1523                goto out_fd;
1524
1525        if (upeer_sockaddr) {
1526                if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1527                                          &len, 2) < 0) {
1528                        err = -ECONNABORTED;
1529                        goto out_fd;
1530                }
1531                err = move_addr_to_user((struct sockaddr *)&address,
1532                                        len, upeer_sockaddr, upeer_addrlen);
1533                if (err < 0)
1534                        goto out_fd;
1535        }
1536
1537        /* File flags are not inherited via accept() unlike another OSes. */
1538
1539        fd_install(newfd, newfile);
1540        err = newfd;
1541
1542out_put:
1543        fput_light(sock->file, fput_needed);
1544out:
1545        return err;
1546out_fd:
1547        fput(newfile);
1548        put_unused_fd(newfd);
1549        goto out_put;
1550}
1551
1552SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1553                int __user *, upeer_addrlen)
1554{
1555        return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1556}
1557
1558/*
1559 *      Attempt to connect to a socket with the server address.  The address
1560 *      is in user space so we verify it is OK and move it to kernel space.
1561 *
1562 *      For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1563 *      break bindings
1564 *
1565 *      NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1566 *      other SEQPACKET protocols that take time to connect() as it doesn't
1567 *      include the -EINPROGRESS status for such sockets.
1568 */
1569
1570SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1571                int, addrlen)
1572{
1573        struct socket *sock;
1574        struct sockaddr_storage address;
1575        int err, fput_needed;
1576
1577        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1578        if (!sock)
1579                goto out;
1580        err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1581        if (err < 0)
1582                goto out_put;
1583
1584        err =
1585            security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1586        if (err)
1587                goto out_put;
1588
1589        err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1590                                 sock->file->f_flags);
1591out_put:
1592        fput_light(sock->file, fput_needed);
1593out:
1594        return err;
1595}
1596
1597/*
1598 *      Get the local address ('name') of a socket object. Move the obtained
1599 *      name to user space.
1600 */
1601
1602SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1603                int __user *, usockaddr_len)
1604{
1605        struct socket *sock;
1606        struct sockaddr_storage address;
1607        int len, err, fput_needed;
1608
1609        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1610        if (!sock)
1611                goto out;
1612
1613        err = security_socket_getsockname(sock);
1614        if (err)
1615                goto out_put;
1616
1617        err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1618        if (err)
1619                goto out_put;
1620        err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1621
1622out_put:
1623        fput_light(sock->file, fput_needed);
1624out:
1625        return err;
1626}
1627
1628/*
1629 *      Get the remote address ('name') of a socket object. Move the obtained
1630 *      name to user space.
1631 */
1632
1633SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1634                int __user *, usockaddr_len)
1635{
1636        struct socket *sock;
1637        struct sockaddr_storage address;
1638        int len, err, fput_needed;
1639
1640        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1641        if (sock != NULL) {
1642                err = security_socket_getpeername(sock);
1643                if (err) {
1644                        fput_light(sock->file, fput_needed);
1645                        return err;
1646                }
1647
1648                err =
1649                    sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1650                                       1);
1651                if (!err)
1652                        err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1653                                                usockaddr_len);
1654                fput_light(sock->file, fput_needed);
1655        }
1656        return err;
1657}
1658
1659/*
1660 *      Send a datagram to a given address. We move the address into kernel
1661 *      space and check the user space data area is readable before invoking
1662 *      the protocol.
1663 */
1664
1665SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1666                unsigned, flags, struct sockaddr __user *, addr,
1667                int, addr_len)
1668{
1669        struct socket *sock;
1670        struct sockaddr_storage address;
1671        int err;
1672        struct msghdr msg;
1673        struct iovec iov;
1674        int fput_needed;
1675
1676        if (len > INT_MAX)
1677                len = INT_MAX;
1678        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1679        if (!sock)
1680                goto out;
1681
1682        iov.iov_base = buff;
1683        iov.iov_len = len;
1684        msg.msg_name = NULL;
1685        msg.msg_iov = &iov;
1686        msg.msg_iovlen = 1;
1687        msg.msg_control = NULL;
1688        msg.msg_controllen = 0;
1689        msg.msg_namelen = 0;
1690        if (addr) {
1691                err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1692                if (err < 0)
1693                        goto out_put;
1694                msg.msg_name = (struct sockaddr *)&address;
1695                msg.msg_namelen = addr_len;
1696        }
1697        if (sock->file->f_flags & O_NONBLOCK)
1698                flags |= MSG_DONTWAIT;
1699        msg.msg_flags = flags;
1700        err = sock_sendmsg(sock, &msg, len);
1701
1702out_put:
1703        fput_light(sock->file, fput_needed);
1704out:
1705        return err;
1706}
1707
1708/*
1709 *      Send a datagram down a socket.
1710 */
1711
1712SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1713                unsigned, flags)
1714{
1715        return sys_sendto(fd, buff, len, flags, NULL, 0);
1716}
1717
1718/*
1719 *      Receive a frame from the socket and optionally record the address of the
1720 *      sender. We verify the buffers are writable and if needed move the
1721 *      sender address from kernel to user space.
1722 */
1723
1724SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1725                unsigned, flags, struct sockaddr __user *, addr,
1726                int __user *, addr_len)
1727{
1728        struct socket *sock;
1729        struct iovec iov;
1730        struct msghdr msg;
1731        struct sockaddr_storage address;
1732        int err, err2;
1733        int fput_needed;
1734
1735        if (size > INT_MAX)
1736                size = INT_MAX;
1737        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1738        if (!sock)
1739                goto out;
1740
1741        msg.msg_control = NULL;
1742        msg.msg_controllen = 0;
1743        msg.msg_iovlen = 1;
1744        msg.msg_iov = &iov;
1745        iov.iov_len = size;
1746        iov.iov_base = ubuf;
1747        msg.msg_name = (struct sockaddr *)&address;
1748        msg.msg_namelen = sizeof(address);
1749        if (sock->file->f_flags & O_NONBLOCK)
1750                flags |= MSG_DONTWAIT;
1751        err = sock_recvmsg(sock, &msg, size, flags);
1752
1753        if (err >= 0 && addr != NULL) {
1754                err2 = move_addr_to_user((struct sockaddr *)&address,
1755                                         msg.msg_namelen, addr, addr_len);
1756                if (err2 < 0)
1757                        err = err2;
1758        }
1759
1760        fput_light(sock->file, fput_needed);
1761out:
1762        return err;
1763}
1764
1765/*
1766 *      Receive a datagram from a socket.
1767 */
1768
1769asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1770                         unsigned flags)
1771{
1772        return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1773}
1774
1775/*
1776 *      Set a socket option. Because we don't know the option lengths we have
1777 *      to pass the user mode parameter for the protocols to sort out.
1778 */
1779
1780SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1781                char __user *, optval, int, optlen)
1782{
1783        int err, fput_needed;
1784        struct socket *sock;
1785
1786        if (optlen < 0)
1787                return -EINVAL;
1788
1789        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1790        if (sock != NULL) {
1791                err = security_socket_setsockopt(sock, level, optname);
1792                if (err)
1793                        goto out_put;
1794
1795                if (level == SOL_SOCKET)
1796                        err =
1797                            sock_setsockopt(sock, level, optname, optval,
1798                                            optlen);
1799                else
1800                        err =
1801                            sock->ops->setsockopt(sock, level, optname, optval,
1802                                                  optlen);
1803out_put:
1804                fput_light(sock->file, fput_needed);
1805        }
1806        return err;
1807}
1808
1809/*
1810 *      Get a socket option. Because we don't know the option lengths we have
1811 *      to pass a user mode parameter for the protocols to sort out.
1812 */
1813
1814SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1815                char __user *, optval, int __user *, optlen)
1816{
1817        int err, fput_needed;
1818        struct socket *sock;
1819
1820        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1821        if (sock != NULL) {
1822                err = security_socket_getsockopt(sock, level, optname);
1823                if (err)
1824                        goto out_put;
1825
1826                if (level == SOL_SOCKET)
1827                        err =
1828                            sock_getsockopt(sock, level, optname, optval,
1829                                            optlen);
1830                else
1831                        err =
1832                            sock->ops->getsockopt(sock, level, optname, optval,
1833                                                  optlen);
1834out_put:
1835                fput_light(sock->file, fput_needed);
1836        }
1837        return err;
1838}
1839
1840/*
1841 *      Shutdown a socket.
1842 */
1843
1844SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1845{
1846        int err, fput_needed;
1847        struct socket *sock;
1848
1849        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1850        if (sock != NULL) {
1851                err = security_socket_shutdown(sock, how);
1852                if (!err)
1853                        err = sock->ops->shutdown(sock, how);
1854                fput_light(sock->file, fput_needed);
1855        }
1856        return err;
1857}
1858
1859/* A couple of helpful macros for getting the address of the 32/64 bit
1860 * fields which are the same type (int / unsigned) on our platforms.
1861 */
1862#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1863#define COMPAT_NAMELEN(msg)     COMPAT_MSG(msg, msg_namelen)
1864#define COMPAT_FLAGS(msg)       COMPAT_MSG(msg, msg_flags)
1865
1866/*
1867 *      BSD sendmsg interface
1868 */
1869
1870SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1871{
1872        struct compat_msghdr __user *msg_compat =
1873            (struct compat_msghdr __user *)msg;
1874        struct socket *sock;
1875        struct sockaddr_storage address;
1876        struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1877        unsigned char ctl[sizeof(struct cmsghdr) + 20]
1878            __attribute__ ((aligned(sizeof(__kernel_size_t))));
1879        /* 20 is size of ipv6_pktinfo */
1880        unsigned char *ctl_buf = ctl;
1881        struct msghdr msg_sys;
1882        int err, ctl_len, iov_size, total_len;
1883        int fput_needed;
1884
1885        err = -EFAULT;
1886        if (MSG_CMSG_COMPAT & flags) {
1887                if (get_compat_msghdr(&msg_sys, msg_compat))
1888                        return -EFAULT;
1889        } else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1890                return -EFAULT;
1891
1892        sock = sockfd_lookup_light(fd, &err, &fput_needed);
1893        if (!sock)
1894                goto out;
1895
1896        /* do not move before msg_sys is valid */
1897        err = -EMSGSIZE;
1898        if (msg_sys.msg_iovlen > UIO_MAXIOV)
1899                goto out_put;
1900
1901        /* Check whether to allocate the iovec area */
1902        err = -ENOMEM;
1903        iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1904        if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1905                iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1906                if (!iov)
1907                        goto out_put;
1908        }
1909
1910        /* This will also move the address data into kernel space */
1911        if (MSG_CMSG_COMPAT & flags) {
1912                err = verify_compat_iovec(&msg_sys, iov,
1913                                          (struct sockaddr *)&address,
1914                                          VERIFY_READ);
1915        } else
1916                err = verify_iovec(&msg_sys, iov,
1917                                   (struct sockaddr *)&address,
1918                                   VERIFY_READ);
1919        if (err < 0)
1920                goto out_freeiov;
1921        total_len = err;
1922
1923        err = -ENOBUFS;
1924
1925        if (msg_sys.msg_controllen > INT_MAX)
1926                goto out_freeiov;
1927        ctl_len = msg_sys.msg_controllen;
1928        if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1929                err =
1930                    cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1931                                                     sizeof(ctl));
1932                if (err)
1933                        goto out_freeiov;
1934                ctl_buf = msg_sys.msg_control;
1935                ctl_len = msg_sys.msg_controllen;
1936        } else if (ctl_len) {
1937                if (ctl_len > sizeof(ctl)) {
1938                        ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1939                        if (ctl_buf == NULL)
1940                                goto out_freeiov;
1941                }
1942                err = -EFAULT;
1943                /*
1944                 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1945                 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1946                 * checking falls down on this.
1947                 */
1948                if (copy_from_user(ctl_buf,
1949                                   (void __user __force *)msg_sys.msg_control,
1950                                   ctl_len))
1951                        goto out_freectl;
1952                msg_sys.msg_control = ctl_buf;
1953        }
1954        msg_sys.msg_flags = flags;
1955
1956        if (sock->file->f_flags & O_NONBLOCK)
1957                msg_sys.msg_flags |= MSG_DONTWAIT;
1958        err = sock_sendmsg(sock, &msg_sys, total_len);
1959
1960out_freectl:
1961        if (ctl_buf != ctl)
1962                sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1963out_freeiov:
1964        if (iov != iovstack)
1965                sock_kfree_s(sock->sk, iov, iov_size);
1966out_put:
1967        fput_light(sock->file, fput_needed);
1968out:
1969        return err;
1970}
1971
1972static int __sys_recvmsg(struct socket *sock, struct msghdr __user *msg,
1973                         struct msghdr *msg_sys, unsigned flags, int nosec)
1974{
1975        struct compat_msghdr __user *msg_compat =
1976            (struct compat_msghdr __user *)msg;
1977        struct iovec iovstack[UIO_FASTIOV];
1978        struct iovec *iov = iovstack;
1979        unsigned long cmsg_ptr;
1980        int err, iov_size, total_len, len;
1981
1982        /* kernel mode address */
1983        struct sockaddr_storage addr;
1984
1985        /* user mode address pointers */
1986        struct sockaddr __user *uaddr;
1987        int __user *uaddr_len;
1988
1989        if (MSG_CMSG_COMPAT & flags) {
1990                if (get_compat_msghdr(msg_sys, msg_compat))
1991                        return -EFAULT;
1992        } else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
1993                return -EFAULT;
1994
1995        err = -EMSGSIZE;
1996        if (msg_sys->msg_iovlen > UIO_MAXIOV)
1997                goto out;
1998
1999        /* Check whether to allocate the iovec area */
2000        err = -ENOMEM;
2001        iov_size = msg_sys->msg_iovlen * sizeof(struct iovec);
2002        if (msg_sys->msg_iovlen > UIO_FASTIOV) {
2003                iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2004                if (!iov)
2005                        goto out;
2006        }
2007
2008        /*
2009         *      Save the user-mode address (verify_iovec will change the
2010         *      kernel msghdr to use the kernel address space)
2011         */
2012
2013        uaddr = (__force void __user *)msg_sys->msg_name;
2014        uaddr_len = COMPAT_NAMELEN(msg);
2015        if (MSG_CMSG_COMPAT & flags) {
2016                err = verify_compat_iovec(msg_sys, iov,
2017                                          (struct sockaddr *)&addr,
2018                                          VERIFY_WRITE);
2019        } else
2020                err = verify_iovec(msg_sys, iov,
2021                                   (struct sockaddr *)&addr,
2022                                   VERIFY_WRITE);
2023        if (err < 0)
2024                goto out_freeiov;
2025        total_len = err;
2026
2027        cmsg_ptr = (unsigned long)msg_sys->msg_control;
2028        msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2029
2030        if (sock->file->f_flags & O_NONBLOCK)
2031                flags |= MSG_DONTWAIT;
2032        err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys,
2033                                                          total_len, flags);
2034        if (err < 0)
2035                goto out_freeiov;
2036        len = err;
2037
2038        if (uaddr != NULL) {
2039                err = move_addr_to_user((struct sockaddr *)&addr,
2040                                        msg_sys->msg_namelen, uaddr,
2041                                        uaddr_len);
2042                if (err < 0)
2043                        goto out_freeiov;
2044        }
2045        err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2046                         COMPAT_FLAGS(msg));
2047        if (err)
2048                goto out_freeiov;
2049        if (MSG_CMSG_COMPAT & flags)
2050                err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2051                                 &msg_compat->msg_controllen);
2052        else
2053                err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2054                                 &msg->msg_controllen);
2055        if (err)
2056                goto out_freeiov;
2057        err = len;
2058
2059out_freeiov:
2060        if (iov != iovstack)
2061                sock_kfree_s(sock->sk, iov, iov_size);
2062out:
2063        return err;
2064}
2065
2066/*
2067 *      BSD recvmsg interface
2068 */
2069
2070SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
2071                unsigned int, flags)
2072{
2073        int fput_needed, err;
2074        struct msghdr msg_sys;
2075        struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2076
2077        if (!sock)
2078                goto out;
2079
2080        err = __sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2081
2082        fput_light(sock->file, fput_needed);
2083out:
2084        return err;
2085}
2086
2087/*
2088 *     Linux recvmmsg interface
2089 */
2090
2091int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2092                   unsigned int flags, struct timespec *timeout)
2093{
2094        int fput_needed, err, datagrams;
2095        struct socket *sock;
2096        struct mmsghdr __user *entry;
2097        struct compat_mmsghdr __user *compat_entry;
2098        struct msghdr msg_sys;
2099        struct timespec end_time;
2100
2101        if (timeout &&
2102            poll_select_set_timeout(&end_time, timeout->tv_sec,
2103                                    timeout->tv_nsec))
2104                return -EINVAL;
2105
2106        datagrams = 0;
2107
2108        sock = sockfd_lookup_light(fd, &err, &fput_needed);
2109        if (!sock)
2110                return err;
2111
2112        err = sock_error(sock->sk);
2113        if (err)
2114                goto out_put;
2115
2116        entry = mmsg;
2117        compat_entry = (struct compat_mmsghdr __user *)mmsg;
2118
2119        while (datagrams < vlen) {
2120                /*
2121                 * No need to ask LSM for more than the first datagram.
2122                 */
2123                if (MSG_CMSG_COMPAT & flags) {
2124                        err = __sys_recvmsg(sock, (struct msghdr __user *)compat_entry,
2125                                            &msg_sys, flags, datagrams);
2126                        if (err < 0)
2127                                break;
2128                        err = __put_user(err, &compat_entry->msg_len);
2129                        ++compat_entry;
2130                } else {
2131                        err = __sys_recvmsg(sock, (struct msghdr __user *)entry,
2132                                            &msg_sys, flags, datagrams);
2133                        if (err < 0)
2134                                break;
2135                        err = put_user(err, &entry->msg_len);
2136                        ++entry;
2137                }
2138
2139                if (err)
2140                        break;
2141                ++datagrams;
2142
2143                /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2144                if (flags & MSG_WAITFORONE)
2145                        flags |= MSG_DONTWAIT;
2146
2147                if (timeout) {
2148                        ktime_get_ts(timeout);
2149                        *timeout = timespec_sub(end_time, *timeout);
2150                        if (timeout->tv_sec < 0) {
2151                                timeout->tv_sec = timeout->tv_nsec = 0;
2152                                break;
2153                        }
2154
2155                        /* Timeout, return less than vlen datagrams */
2156                        if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2157                                break;
2158                }
2159
2160                /* Out of band data, return right away */
2161                if (msg_sys.msg_flags & MSG_OOB)
2162                        break;
2163        }
2164
2165out_put:
2166        fput_light(sock->file, fput_needed);
2167
2168        if (err == 0)
2169                return datagrams;
2170
2171        if (datagrams != 0) {
2172                /*
2173                 * We may return less entries than requested (vlen) if the
2174                 * sock is non block and there aren't enough datagrams...
2175                 */
2176                if (err != -EAGAIN) {
2177                        /*
2178                         * ... or  if recvmsg returns an error after we
2179                         * received some datagrams, where we record the
2180                         * error to return on the next call or if the
2181                         * app asks about it using getsockopt(SO_ERROR).
2182                         */
2183                        sock->sk->sk_err = -err;
2184                }
2185
2186                return datagrams;
2187        }
2188
2189        return err;
2190}
2191
2192SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2193                unsigned int, vlen, unsigned int, flags,
2194                struct timespec __user *, timeout)
2195{
2196        int datagrams;
2197        struct timespec timeout_sys;
2198
2199        if (!timeout)
2200                return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2201
2202        if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2203                return -EFAULT;
2204
2205        datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2206
2207        if (datagrams > 0 &&
2208            copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2209                datagrams = -EFAULT;
2210
2211        return datagrams;
2212}
2213
2214#ifdef __ARCH_WANT_SYS_SOCKETCALL
2215/* Argument list sizes for sys_socketcall */
2216#define AL(x) ((x) * sizeof(unsigned long))
2217static const unsigned char nargs[20] = {
2218        AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2219        AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2220        AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2221        AL(4), AL(5)
2222};
2223
2224#undef AL
2225
2226/*
2227 *      System call vectors.
2228 *
2229 *      Argument checking cleaned up. Saved 20% in size.
2230 *  This function doesn't need to set the kernel lock because
2231 *  it is set by the callees.
2232 */
2233
2234SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2235{
2236        unsigned long a[6];
2237        unsigned long a0, a1;
2238        int err;
2239        unsigned int len;
2240
2241        if (call < 1 || call > SYS_RECVMMSG)
2242                return -EINVAL;
2243
2244        len = nargs[call];
2245        if (len > sizeof(a))
2246                return -EINVAL;
2247
2248        /* copy_from_user should be SMP safe. */
2249        if (copy_from_user(a, args, len))
2250                return -EFAULT;
2251
2252        audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2253
2254        a0 = a[0];
2255        a1 = a[1];
2256
2257        switch (call) {
2258        case SYS_SOCKET:
2259                err = sys_socket(a0, a1, a[2]);
2260                break;
2261        case SYS_BIND:
2262                err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2263                break;
2264        case SYS_CONNECT:
2265                err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2266                break;
2267        case SYS_LISTEN:
2268                err = sys_listen(a0, a1);
2269                break;
2270        case SYS_ACCEPT:
2271                err = sys_accept4(a0, (struct sockaddr __user *)a1,
2272                                  (int __user *)a[2], 0);
2273                break;
2274        case SYS_GETSOCKNAME:
2275                err =
2276                    sys_getsockname(a0, (struct sockaddr __user *)a1,
2277                                    (int __user *)a[2]);
2278                break;
2279        case SYS_GETPEERNAME:
2280                err =
2281                    sys_getpeername(a0, (struct sockaddr __user *)a1,
2282                                    (int __user *)a[2]);
2283                break;
2284        case SYS_SOCKETPAIR:
2285                err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2286                break;
2287        case SYS_SEND:
2288                err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2289                break;
2290        case SYS_SENDTO:
2291                err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2292                                 (struct sockaddr __user *)a[4], a[5]);
2293                break;
2294        case SYS_RECV:
2295                err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2296                break;
2297        case SYS_RECVFROM:
2298                err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2299                                   (struct sockaddr __user *)a[4],
2300                                   (int __user *)a[5]);
2301                break;
2302        case SYS_SHUTDOWN:
2303                err = sys_shutdown(a0, a1);
2304                break;
2305        case SYS_SETSOCKOPT:
2306                err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2307                break;
2308        case SYS_GETSOCKOPT:
2309                err =
2310                    sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2311                                   (int __user *)a[4]);
2312                break;
2313        case SYS_SENDMSG:
2314                err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2315                break;
2316        case SYS_RECVMSG:
2317                err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2318                break;
2319        case SYS_RECVMMSG:
2320                err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2321                                   (struct timespec __user *)a[4]);
2322                break;
2323        case SYS_ACCEPT4:
2324                err = sys_accept4(a0, (struct sockaddr __user *)a1,
2325                                  (int __user *)a[2], a[3]);
2326                break;
2327        default:
2328                err = -EINVAL;
2329                break;
2330        }
2331        return err;
2332}
2333
2334#endif                          /* __ARCH_WANT_SYS_SOCKETCALL */
2335
2336/**
2337 *      sock_register - add a socket protocol handler
2338 *      @ops: description of protocol
2339 *
2340 *      This function is called by a protocol handler that wants to
2341 *      advertise its address family, and have it linked into the
2342 *      socket interface. The value ops->family coresponds to the
2343 *      socket system call protocol family.
2344 */
2345int sock_register(const struct net_proto_family *ops)
2346{
2347        int err;
2348
2349        if (ops->family >= NPROTO) {
2350                printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2351                       NPROTO);
2352                return -ENOBUFS;
2353        }
2354
2355        spin_lock(&net_family_lock);
2356        if (rcu_dereference_protected(net_families[ops->family],
2357                                      lockdep_is_held(&net_family_lock)))
2358                err = -EEXIST;
2359        else {
2360                rcu_assign_pointer(net_families[ops->family], ops);
2361                err = 0;
2362        }
2363        spin_unlock(&net_family_lock);
2364
2365        printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2366        return err;
2367}
2368EXPORT_SYMBOL(sock_register);
2369
2370/**
2371 *      sock_unregister - remove a protocol handler
2372 *      @family: protocol family to remove
2373 *
2374 *      This function is called by a protocol handler that wants to
2375 *      remove its address family, and have it unlinked from the
2376 *      new socket creation.
2377 *
2378 *      If protocol handler is a module, then it can use module reference
2379 *      counts to protect against new references. If protocol handler is not
2380 *      a module then it needs to provide its own protection in
2381 *      the ops->create routine.
2382 */
2383void sock_unregister(int family)
2384{
2385        BUG_ON(family < 0 || family >= NPROTO);
2386
2387        spin_lock(&net_family_lock);
2388        rcu_assign_pointer(net_families[family], NULL);
2389        spin_unlock(&net_family_lock);
2390
2391        synchronize_rcu();
2392
2393        printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2394}
2395EXPORT_SYMBOL(sock_unregister);
2396
2397static int __init sock_init(void)
2398{
2399        int err;
2400
2401        /*
2402         *      Initialize sock SLAB cache.
2403         */
2404
2405        sk_init();
2406
2407        /*
2408         *      Initialize skbuff SLAB cache
2409         */
2410        skb_init();
2411
2412        /*
2413         *      Initialize the protocols module.
2414         */
2415
2416        init_inodecache();
2417
2418        err = register_filesystem(&sock_fs_type);
2419        if (err)
2420                goto out_fs;
2421        sock_mnt = kern_mount(&sock_fs_type);
2422        if (IS_ERR(sock_mnt)) {
2423                err = PTR_ERR(sock_mnt);
2424                goto out_mount;
2425        }
2426
2427        /* The real protocol initialization is performed in later initcalls.
2428         */
2429
2430#ifdef CONFIG_NETFILTER
2431        netfilter_init();
2432#endif
2433
2434#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2435        skb_timestamping_init();
2436#endif
2437
2438out:
2439        return err;
2440
2441out_mount:
2442        unregister_filesystem(&sock_fs_type);
2443out_fs:
2444        goto out;
2445}
2446
2447core_initcall(sock_init);       /* early initcall */
2448
2449#ifdef CONFIG_PROC_FS
2450void socket_seq_show(struct seq_file *seq)
2451{
2452        int cpu;
2453        int counter = 0;
2454
2455        for_each_possible_cpu(cpu)
2456            counter += per_cpu(sockets_in_use, cpu);
2457
2458        /* It can be negative, by the way. 8) */
2459        if (counter < 0)
2460                counter = 0;
2461
2462        seq_printf(seq, "sockets: used %d\n", counter);
2463}
2464#endif                          /* CONFIG_PROC_FS */
2465
2466#ifdef CONFIG_COMPAT
2467static int do_siocgstamp(struct net *net, struct socket *sock,
2468                         unsigned int cmd, struct compat_timeval __user *up)
2469{
2470        mm_segment_t old_fs = get_fs();
2471        struct timeval ktv;
2472        int err;
2473
2474        set_fs(KERNEL_DS);
2475        err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2476        set_fs(old_fs);
2477        if (!err) {
2478                err = put_user(ktv.tv_sec, &up->tv_sec);
2479                err |= __put_user(ktv.tv_usec, &up->tv_usec);
2480        }
2481        return err;
2482}
2483
2484static int do_siocgstampns(struct net *net, struct socket *sock,
2485                         unsigned int cmd, struct compat_timespec __user *up)
2486{
2487        mm_segment_t old_fs = get_fs();
2488        struct timespec kts;
2489        int err;
2490
2491        set_fs(KERNEL_DS);
2492        err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2493        set_fs(old_fs);
2494        if (!err) {
2495                err = put_user(kts.tv_sec, &up->tv_sec);
2496                err |= __put_user(kts.tv_nsec, &up->tv_nsec);
2497        }
2498        return err;
2499}
2500
2501static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2502{
2503        struct ifreq __user *uifr;
2504        int err;
2505
2506        uifr = compat_alloc_user_space(sizeof(struct ifreq));
2507        if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2508                return -EFAULT;
2509
2510        err = dev_ioctl(net, SIOCGIFNAME, uifr);
2511        if (err)
2512                return err;
2513
2514        if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2515                return -EFAULT;
2516
2517        return 0;
2518}
2519
2520static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2521{
2522        struct compat_ifconf ifc32;
2523        struct ifconf ifc;
2524        struct ifconf __user *uifc;
2525        struct compat_ifreq __user *ifr32;
2526        struct ifreq __user *ifr;
2527        unsigned int i, j;
2528        int err;
2529
2530        if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2531                return -EFAULT;
2532
2533        if (ifc32.ifcbuf == 0) {
2534                ifc32.ifc_len = 0;
2535                ifc.ifc_len = 0;
2536                ifc.ifc_req = NULL;
2537                uifc = compat_alloc_user_space(sizeof(struct ifconf));
2538        } else {
2539                size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2540                        sizeof(struct ifreq);
2541                uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2542                ifc.ifc_len = len;
2543                ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2544                ifr32 = compat_ptr(ifc32.ifcbuf);
2545                for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2546                        if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2547                                return -EFAULT;
2548                        ifr++;
2549                        ifr32++;
2550                }
2551        }
2552        if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2553                return -EFAULT;
2554
2555        err = dev_ioctl(net, SIOCGIFCONF, uifc);
2556        if (err)
2557                return err;
2558
2559        if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2560                return -EFAULT;
2561
2562        ifr = ifc.ifc_req;
2563        ifr32 = compat_ptr(ifc32.ifcbuf);
2564        for (i = 0, j = 0;
2565             i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2566             i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2567                if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2568                        return -EFAULT;
2569                ifr32++;
2570                ifr++;
2571        }
2572
2573        if (ifc32.ifcbuf == 0) {
2574                /* Translate from 64-bit structure multiple to
2575                 * a 32-bit one.
2576                 */
2577                i = ifc.ifc_len;
2578                i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2579                ifc32.ifc_len = i;
2580        } else {
2581                ifc32.ifc_len = i;
2582        }
2583        if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2584                return -EFAULT;
2585
2586        return 0;
2587}
2588
2589static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2590{
2591        struct compat_ethtool_rxnfc __user *compat_rxnfc;
2592        bool convert_in = false, convert_out = false;
2593        size_t buf_size = ALIGN(sizeof(struct ifreq), 8);
2594        struct ethtool_rxnfc __user *rxnfc;
2595        struct ifreq __user *ifr;
2596        u32 rule_cnt = 0, actual_rule_cnt;
2597        u32 ethcmd;
2598        u32 data;
2599        int ret;
2600
2601        if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2602                return -EFAULT;
2603
2604        compat_rxnfc = compat_ptr(data);
2605
2606        if (get_user(ethcmd, &compat_rxnfc->cmd))
2607                return -EFAULT;
2608
2609        /* Most ethtool structures are defined without padding.
2610         * Unfortunately struct ethtool_rxnfc is an exception.
2611         */
2612        switch (ethcmd) {
2613        default:
2614                break;
2615        case ETHTOOL_GRXCLSRLALL:
2616                /* Buffer size is variable */
2617                if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2618                        return -EFAULT;
2619                if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2620                        return -ENOMEM;
2621                buf_size += rule_cnt * sizeof(u32);
2622                /* fall through */
2623        case ETHTOOL_GRXRINGS:
2624        case ETHTOOL_GRXCLSRLCNT:
2625        case ETHTOOL_GRXCLSRULE:
2626                convert_out = true;
2627                /* fall through */
2628        case ETHTOOL_SRXCLSRLDEL:
2629        case ETHTOOL_SRXCLSRLINS:
2630                buf_size += sizeof(struct ethtool_rxnfc);
2631                convert_in = true;
2632                break;
2633        }
2634
2635        ifr = compat_alloc_user_space(buf_size);
2636        rxnfc = (void *)ifr + ALIGN(sizeof(struct ifreq), 8);
2637
2638        if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2639                return -EFAULT;
2640
2641        if (put_user(convert_in ? rxnfc : compat_ptr(data),
2642                     &ifr->ifr_ifru.ifru_data))
2643                return -EFAULT;
2644
2645        if (convert_in) {
2646                /* We expect there to be holes between fs.m_u and
2647                 * fs.ring_cookie and at the end of fs, but nowhere else.
2648                 */
2649                BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_u) +
2650                             sizeof(compat_rxnfc->fs.m_u) !=
2651                             offsetof(struct ethtool_rxnfc, fs.m_u) +
2652                             sizeof(rxnfc->fs.m_u));
2653                BUILD_BUG_ON(
2654                        offsetof(struct compat_ethtool_rxnfc, fs.location) -
2655                        offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2656                        offsetof(struct ethtool_rxnfc, fs.location) -
2657                        offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2658
2659                if (copy_in_user(rxnfc, compat_rxnfc,
2660                                 (void *)(&rxnfc->fs.m_u + 1) -
2661                                 (void *)rxnfc) ||
2662                    copy_in_user(&rxnfc->fs.ring_cookie,
2663                                 &compat_rxnfc->fs.ring_cookie,
2664                                 (void *)(&rxnfc->fs.location + 1) -
2665                                 (void *)&rxnfc->fs.ring_cookie) ||
2666                    copy_in_user(&rxnfc->rule_cnt, &compat_rxnfc->rule_cnt,
2667                                 sizeof(rxnfc->rule_cnt)))
2668                        return -EFAULT;
2669        }
2670
2671        ret = dev_ioctl(net, SIOCETHTOOL, ifr);
2672        if (ret)
2673                return ret;
2674
2675        if (convert_out) {
2676                if (copy_in_user(compat_rxnfc, rxnfc,
2677                                 (const void *)(&rxnfc->fs.m_u + 1) -
2678                                 (const void *)rxnfc) ||
2679                    copy_in_user(&compat_rxnfc->fs.ring_cookie,
2680                                 &rxnfc->fs.ring_cookie,
2681                                 (const void *)(&rxnfc->fs.location + 1) -
2682                                 (const void *)&rxnfc->fs.ring_cookie) ||
2683                    copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2684                                 sizeof(rxnfc->rule_cnt)))
2685                        return -EFAULT;
2686
2687                if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2688                        /* As an optimisation, we only copy the actual
2689                         * number of rules that the underlying
2690                         * function returned.  Since Mallory might
2691                         * change the rule count in user memory, we
2692                         * check that it is less than the rule count
2693                         * originally given (as the user buffer size),
2694                         * which has been range-checked.
2695                         */
2696                        if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2697                                return -EFAULT;
2698                        if (actual_rule_cnt < rule_cnt)
2699                                rule_cnt = actual_rule_cnt;
2700                        if (copy_in_user(&compat_rxnfc->rule_locs[0],
2701                                         &rxnfc->rule_locs[0],
2702                                         rule_cnt * sizeof(u32)))
2703                                return -EFAULT;
2704                }
2705        }
2706
2707        return 0;
2708}
2709
2710static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2711{
2712        void __user *uptr;
2713        compat_uptr_t uptr32;
2714        struct ifreq __user *uifr;
2715
2716        uifr = compat_alloc_user_space(sizeof(*uifr));
2717        if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2718                return -EFAULT;
2719
2720        if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2721                return -EFAULT;
2722
2723        uptr = compat_ptr(uptr32);
2724
2725        if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2726                return -EFAULT;
2727
2728        return dev_ioctl(net, SIOCWANDEV, uifr);
2729}
2730
2731static int bond_ioctl(struct net *net, unsigned int cmd,
2732                         struct compat_ifreq __user *ifr32)
2733{
2734        struct ifreq kifr;
2735        struct ifreq __user *uifr;
2736        mm_segment_t old_fs;
2737        int err;
2738        u32 data;
2739        void __user *datap;
2740
2741        switch (cmd) {
2742        case SIOCBONDENSLAVE:
2743        case SIOCBONDRELEASE:
2744        case SIOCBONDSETHWADDR:
2745        case SIOCBONDCHANGEACTIVE:
2746                if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2747                        return -EFAULT;
2748
2749                old_fs = get_fs();
2750                set_fs(KERNEL_DS);
2751                err = dev_ioctl(net, cmd,
2752                                (struct ifreq __user __force *) &kifr);
2753                set_fs(old_fs);
2754
2755                return err;
2756        case SIOCBONDSLAVEINFOQUERY:
2757        case SIOCBONDINFOQUERY:
2758                uifr = compat_alloc_user_space(sizeof(*uifr));
2759                if (copy_in_user(&uifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2760                        return -EFAULT;
2761
2762                if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2763                        return -EFAULT;
2764
2765                datap = compat_ptr(data);
2766                if (put_user(datap, &uifr->ifr_ifru.ifru_data))
2767                        return -EFAULT;
2768
2769                return dev_ioctl(net, cmd, uifr);
2770        default:
2771                return -EINVAL;
2772        }
2773}
2774
2775static int siocdevprivate_ioctl(struct net *net, unsigned int cmd,
2776                                 struct compat_ifreq __user *u_ifreq32)
2777{
2778        struct ifreq __user *u_ifreq64;
2779        char tmp_buf[IFNAMSIZ];
2780        void __user *data64;
2781        u32 data32;
2782
2783        if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2784                           IFNAMSIZ))
2785                return -EFAULT;
2786        if (__get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2787                return -EFAULT;
2788        data64 = compat_ptr(data32);
2789
2790        u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2791
2792        /* Don't check these user accesses, just let that get trapped
2793         * in the ioctl handler instead.
2794         */
2795        if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2796                         IFNAMSIZ))
2797                return -EFAULT;
2798        if (__put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2799                return -EFAULT;
2800
2801        return dev_ioctl(net, cmd, u_ifreq64);
2802}
2803
2804static int dev_ifsioc(struct net *net, struct socket *sock,
2805                         unsigned int cmd, struct compat_ifreq __user *uifr32)
2806{
2807        struct ifreq __user *uifr;
2808        int err;
2809
2810        uifr = compat_alloc_user_space(sizeof(*uifr));
2811        if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2812                return -EFAULT;
2813
2814        err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2815
2816        if (!err) {
2817                switch (cmd) {
2818                case SIOCGIFFLAGS:
2819                case SIOCGIFMETRIC:
2820                case SIOCGIFMTU:
2821                case SIOCGIFMEM:
2822                case SIOCGIFHWADDR:
2823                case SIOCGIFINDEX:
2824                case SIOCGIFADDR:
2825                case SIOCGIFBRDADDR:
2826                case SIOCGIFDSTADDR:
2827                case SIOCGIFNETMASK:
2828                case SIOCGIFPFLAGS:
2829                case SIOCGIFTXQLEN:
2830                case SIOCGMIIPHY:
2831                case SIOCGMIIREG:
2832                        if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2833                                err = -EFAULT;
2834                        break;
2835                }
2836        }
2837        return err;
2838}
2839
2840static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2841                        struct compat_ifreq __user *uifr32)
2842{
2843        struct ifreq ifr;
2844        struct compat_ifmap __user *uifmap32;
2845        mm_segment_t old_fs;
2846        int err;
2847
2848        uifmap32 = &uifr32->ifr_ifru.ifru_map;
2849        err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2850        err |= __get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2851        err |= __get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2852        err |= __get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2853        err |= __get_user(ifr.ifr_map.irq, &uifmap32->irq);
2854        err |= __get_user(ifr.ifr_map.dma, &uifmap32->dma);
2855        err |= __get_user(ifr.ifr_map.port, &uifmap32->port);
2856        if (err)
2857                return -EFAULT;
2858
2859        old_fs = get_fs();
2860        set_fs(KERNEL_DS);
2861        err = dev_ioctl(net, cmd, (void  __user __force *)&ifr);
2862        set_fs(old_fs);
2863
2864        if (cmd == SIOCGIFMAP && !err) {
2865                err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2866                err |= __put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2867                err |= __put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2868                err |= __put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2869                err |= __put_user(ifr.ifr_map.irq, &uifmap32->irq);
2870                err |= __put_user(ifr.ifr_map.dma, &uifmap32->dma);
2871                err |= __put_user(ifr.ifr_map.port, &uifmap32->port);
2872                if (err)
2873                        err = -EFAULT;
2874        }
2875        return err;
2876}
2877
2878static int compat_siocshwtstamp(struct net *net, struct compat_ifreq __user *uifr32)
2879{
2880        void __user *uptr;
2881        compat_uptr_t uptr32;
2882        struct ifreq __user *uifr;
2883
2884        uifr = compat_alloc_user_space(sizeof(*uifr));
2885        if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2886                return -EFAULT;
2887
2888        if (get_user(uptr32, &uifr32->ifr_data))
2889                return -EFAULT;
2890
2891        uptr = compat_ptr(uptr32);
2892
2893        if (put_user(uptr, &uifr->ifr_data))
2894                return -EFAULT;
2895
2896        return dev_ioctl(net, SIOCSHWTSTAMP, uifr);
2897}
2898
2899struct rtentry32 {
2900        u32             rt_pad1;
2901        struct sockaddr rt_dst;         /* target address               */
2902        struct sockaddr rt_gateway;     /* gateway addr (RTF_GATEWAY)   */
2903        struct sockaddr rt_genmask;     /* target network mask (IP)     */
2904        unsigned short  rt_flags;
2905        short           rt_pad2;
2906        u32             rt_pad3;
2907        unsigned char   rt_tos;
2908        unsigned char   rt_class;
2909        short           rt_pad4;
2910        short           rt_metric;      /* +1 for binary compatibility! */
2911        /* char * */ u32 rt_dev;        /* forcing the device at add    */
2912        u32             rt_mtu;         /* per route MTU/Window         */
2913        u32             rt_window;      /* Window clamping              */
2914        unsigned short  rt_irtt;        /* Initial RTT                  */
2915};
2916
2917struct in6_rtmsg32 {
2918        struct in6_addr         rtmsg_dst;
2919        struct in6_addr         rtmsg_src;
2920        struct in6_addr         rtmsg_gateway;
2921        u32                     rtmsg_type;
2922        u16                     rtmsg_dst_len;
2923        u16                     rtmsg_src_len;
2924        u32                     rtmsg_metric;
2925        u32                     rtmsg_info;
2926        u32                     rtmsg_flags;
2927        s32                     rtmsg_ifindex;
2928};
2929
2930static int routing_ioctl(struct net *net, struct socket *sock,
2931                         unsigned int cmd, void __user *argp)
2932{
2933        int ret;
2934        void *r = NULL;
2935        struct in6_rtmsg r6;
2936        struct rtentry r4;
2937        char devname[16];
2938        u32 rtdev;
2939        mm_segment_t old_fs = get_fs();
2940
2941        if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
2942                struct in6_rtmsg32 __user *ur6 = argp;
2943                ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
2944                        3 * sizeof(struct in6_addr));
2945                ret |= __get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
2946                ret |= __get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
2947                ret |= __get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
2948                ret |= __get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
2949                ret |= __get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
2950                ret |= __get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
2951                ret |= __get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
2952
2953                r = (void *) &r6;
2954        } else { /* ipv4 */
2955                struct rtentry32 __user *ur4 = argp;
2956                ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
2957                                        3 * sizeof(struct sockaddr));
2958                ret |= __get_user(r4.rt_flags, &(ur4->rt_flags));
2959                ret |= __get_user(r4.rt_metric, &(ur4->rt_metric));
2960                ret |= __get_user(r4.rt_mtu, &(ur4->rt_mtu));
2961                ret |= __get_user(r4.rt_window, &(ur4->rt_window));
2962                ret |= __get_user(r4.rt_irtt, &(ur4->rt_irtt));
2963                ret |= __get_user(rtdev, &(ur4->rt_dev));
2964                if (rtdev) {
2965                        ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
2966                        r4.rt_dev = (char __user __force *)devname;
2967                        devname[15] = 0;
2968                } else
2969                        r4.rt_dev = NULL;
2970
2971                r = (void *) &r4;
2972        }
2973
2974        if (ret) {
2975                ret = -EFAULT;
2976                goto out;
2977        }
2978
2979        set_fs(KERNEL_DS);
2980        ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
2981        set_fs(old_fs);
2982
2983out:
2984        return ret;
2985}
2986
2987/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
2988 * for some operations; this forces use of the newer bridge-utils that
2989 * use compatible ioctls
2990 */
2991static int old_bridge_ioctl(compat_ulong_t __user *argp)
2992{
2993        compat_ulong_t tmp;
2994
2995        if (get_user(tmp, argp))
2996                return -EFAULT;
2997        if (tmp == BRCTL_GET_VERSION)
2998                return BRCTL_VERSION + 1;
2999        return -EINVAL;
3000}
3001
3002static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3003                         unsigned int cmd, unsigned long arg)
3004{
3005        void __user *argp = compat_ptr(arg);
3006        struct sock *sk = sock->sk;
3007        struct net *net = sock_net(sk);
3008
3009        if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3010                return siocdevprivate_ioctl(net, cmd, argp);
3011
3012        switch (cmd) {
3013        case SIOCSIFBR:
3014        case SIOCGIFBR:
3015                return old_bridge_ioctl(argp);
3016        case SIOCGIFNAME:
3017                return dev_ifname32(net, argp);
3018        case SIOCGIFCONF:
3019                return dev_ifconf(net, argp);
3020        case SIOCETHTOOL:
3021                return ethtool_ioctl(net, argp);
3022        case SIOCWANDEV:
3023                return compat_siocwandev(net, argp);
3024        case SIOCGIFMAP:
3025        case SIOCSIFMAP:
3026                return compat_sioc_ifmap(net, cmd, argp);
3027        case SIOCBONDENSLAVE:
3028        case SIOCBONDRELEASE:
3029        case SIOCBONDSETHWADDR:
3030        case SIOCBONDSLAVEINFOQUERY:
3031        case SIOCBONDINFOQUERY:
3032        case SIOCBONDCHANGEACTIVE:
3033                return bond_ioctl(net, cmd, argp);
3034        case SIOCADDRT:
3035        case SIOCDELRT:
3036                return routing_ioctl(net, sock, cmd, argp);
3037        case SIOCGSTAMP:
3038                return do_siocgstamp(net, sock, cmd, argp);
3039        case SIOCGSTAMPNS:
3040                return do_siocgstampns(net, sock, cmd, argp);
3041        case SIOCSHWTSTAMP:
3042                return compat_siocshwtstamp(net, argp);
3043
3044        case FIOSETOWN:
3045        case SIOCSPGRP:
3046        case FIOGETOWN:
3047        case SIOCGPGRP:
3048        case SIOCBRADDBR:
3049        case SIOCBRDELBR:
3050        case SIOCGIFVLAN:
3051        case SIOCSIFVLAN:
3052        case SIOCADDDLCI:
3053        case SIOCDELDLCI:
3054                return sock_ioctl(file, cmd, arg);
3055
3056        case SIOCGIFFLAGS:
3057        case SIOCSIFFLAGS:
3058        case SIOCGIFMETRIC:
3059        case SIOCSIFMETRIC:
3060        case SIOCGIFMTU:
3061        case SIOCSIFMTU:
3062        case SIOCGIFMEM:
3063        case SIOCSIFMEM:
3064        case SIOCGIFHWADDR:
3065        case SIOCSIFHWADDR:
3066        case SIOCADDMULTI:
3067        case SIOCDELMULTI:
3068        case SIOCGIFINDEX:
3069        case SIOCGIFADDR:
3070        case SIOCSIFADDR:
3071        case SIOCSIFHWBROADCAST:
3072        case SIOCDIFADDR:
3073        case SIOCGIFBRDADDR:
3074        case SIOCSIFBRDADDR:
3075        case SIOCGIFDSTADDR:
3076        case SIOCSIFDSTADDR:
3077        case SIOCGIFNETMASK:
3078        case SIOCSIFNETMASK:
3079        case SIOCSIFPFLAGS:
3080        case SIOCGIFPFLAGS:
3081        case SIOCGIFTXQLEN:
3082        case SIOCSIFTXQLEN:
3083        case SIOCBRADDIF:
3084        case SIOCBRDELIF:
3085        case SIOCSIFNAME:
3086        case SIOCGMIIPHY:
3087        case SIOCGMIIREG:
3088        case SIOCSMIIREG:
3089                return dev_ifsioc(net, sock, cmd, argp);
3090
3091        case SIOCSARP:
3092        case SIOCGARP:
3093        case SIOCDARP:
3094        case SIOCATMARK:
3095                return sock_do_ioctl(net, sock, cmd, arg);
3096        }
3097
3098        /* Prevent warning from compat_sys_ioctl, these always
3099         * result in -EINVAL in the native case anyway. */
3100        switch (cmd) {
3101        case SIOCRTMSG:
3102        case SIOCGIFCOUNT:
3103        case SIOCSRARP:
3104        case SIOCGRARP:
3105        case SIOCDRARP:
3106        case SIOCSIFLINK:
3107        case SIOCGIFSLAVE:
3108        case SIOCSIFSLAVE:
3109                return -EINVAL;
3110        }
3111
3112        return -ENOIOCTLCMD;
3113}
3114
3115static long compat_sock_ioctl(struct file *file, unsigned cmd,
3116                              unsigned long arg)
3117{
3118        struct socket *sock = file->private_data;
3119        int ret = -ENOIOCTLCMD;
3120        struct sock *sk;
3121        struct net *net;
3122
3123        sk = sock->sk;
3124        net = sock_net(sk);
3125
3126        if (sock->ops->compat_ioctl)
3127                ret = sock->ops->compat_ioctl(sock, cmd, arg);
3128
3129        if (ret == -ENOIOCTLCMD &&
3130            (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3131                ret = compat_wext_handle_ioctl(net, cmd, arg);
3132
3133        if (ret == -ENOIOCTLCMD)
3134                ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3135
3136        return ret;
3137}
3138#endif
3139
3140int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3141{
3142        return sock->ops->bind(sock, addr, addrlen);
3143}
3144EXPORT_SYMBOL(kernel_bind);
3145
3146int kernel_listen(struct socket *sock, int backlog)
3147{
3148        return sock->ops->listen(sock, backlog);
3149}
3150EXPORT_SYMBOL(kernel_listen);
3151
3152int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3153{
3154        struct sock *sk = sock->sk;
3155        int err;
3156
3157        err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3158                               newsock);
3159        if (err < 0)
3160                goto done;
3161
3162        err = sock->ops->accept(sock, *newsock, flags);
3163        if (err < 0) {
3164                sock_release(*newsock);
3165                *newsock = NULL;
3166                goto done;
3167        }
3168
3169        (*newsock)->ops = sock->ops;
3170        __module_get((*newsock)->ops->owner);
3171
3172done:
3173        return err;
3174}
3175EXPORT_SYMBOL(kernel_accept);
3176
3177int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3178                   int flags)
3179{
3180        return sock->ops->connect(sock, addr, addrlen, flags);
3181}
3182EXPORT_SYMBOL(kernel_connect);
3183
3184int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3185                         int *addrlen)
3186{
3187        return sock->ops->getname(sock, addr, addrlen, 0);
3188}
3189EXPORT_SYMBOL(kernel_getsockname);
3190
3191int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3192                         int *addrlen)
3193{
3194        return sock->ops->getname(sock, addr, addrlen, 1);
3195}
3196EXPORT_SYMBOL(kernel_getpeername);
3197
3198int kernel_getsockopt(struct socket *sock, int level, int optname,
3199                        char *optval, int *optlen)
3200{
3201        mm_segment_t oldfs = get_fs();
3202        char __user *uoptval;
3203        int __user *uoptlen;
3204        int err;
3205
3206        uoptval = (char __user __force *) optval;
3207        uoptlen = (int __user __force *) optlen;
3208
3209        set_fs(KERNEL_DS);
3210        if (level == SOL_SOCKET)
3211                err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3212        else
3213                err = sock->ops->getsockopt(sock, level, optname, uoptval,
3214                                            uoptlen);
3215        set_fs(oldfs);
3216        return err;
3217}
3218EXPORT_SYMBOL(kernel_getsockopt);
3219
3220int kernel_setsockopt(struct socket *sock, int level, int optname,
3221                        char *optval, unsigned int optlen)
3222{
3223        mm_segment_t oldfs = get_fs();
3224        char __user *uoptval;
3225        int err;
3226
3227        uoptval = (char __user __force *) optval;
3228
3229        set_fs(KERNEL_DS);
3230        if (level == SOL_SOCKET)
3231                err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3232        else
3233                err = sock->ops->setsockopt(sock, level, optname, uoptval,
3234                                            optlen);
3235        set_fs(oldfs);
3236        return err;
3237}
3238EXPORT_SYMBOL(kernel_setsockopt);
3239
3240int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3241                    size_t size, int flags)
3242{
3243        sock_update_classid(sock->sk);
3244
3245        if (sock->ops->sendpage)
3246                return sock->ops->sendpage(sock, page, offset, size, flags);
3247
3248        return sock_no_sendpage(sock, page, offset, size, flags);
3249}
3250EXPORT_SYMBOL(kernel_sendpage);
3251
3252int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3253{
3254        mm_segment_t oldfs = get_fs();
3255        int err;
3256
3257        set_fs(KERNEL_DS);
3258        err = sock->ops->ioctl(sock, cmd, arg);
3259        set_fs(oldfs);
3260
3261        return err;
3262}
3263EXPORT_SYMBOL(kernel_sock_ioctl);
3264
3265int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3266{
3267        return sock->ops->shutdown(sock, how);
3268}
3269EXPORT_SYMBOL(kernel_sock_shutdown);
3270