1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET An implementation of the SOCKET network access protocol. 4 * 5 * Version: @(#)socket.c 1.1.93 18/02/95 6 * 7 * Authors: Orest Zborowski, <obz@Kodak.COM> 8 * Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * 11 * Fixes: 12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in 13 * shutdown() 14 * Alan Cox : verify_area() fixes 15 * Alan Cox : Removed DDI 16 * Jonathan Kamens : SOCK_DGRAM reconnect bug 17 * Alan Cox : Moved a load of checks to the very 18 * top level. 19 * Alan Cox : Move address structures to/from user 20 * mode above the protocol layers. 21 * Rob Janssen : Allow 0 length sends. 22 * Alan Cox : Asynchronous I/O support (cribbed from the 23 * tty drivers). 24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style) 25 * Jeff Uphoff : Made max number of sockets command-line 26 * configurable. 27 * Matti Aarnio : Made the number of sockets dynamic, 28 * to be allocated when needed, and mr. 29 * Uphoff's max is used as max to be 30 * allowed to allocate. 31 * Linus : Argh. removed all the socket allocation 32 * altogether: it's in the inode now. 33 * Alan Cox : Made sock_alloc()/sock_release() public 34 * for NetROM and future kernel nfsd type 35 * stuff. 36 * Alan Cox : sendmsg/recvmsg basics. 37 * Tom Dyas : Export net symbols. 38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n". 39 * Alan Cox : Added thread locking to sys_* calls 40 * for sockets. May have errors at the 41 * moment. 42 * Kevin Buhr : Fixed the dumb errors in the above. 43 * Andi Kleen : Some small cleanups, optimizations, 44 * and fixed a copy_from_user() bug. 45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0) 46 * Tigran Aivazian : Made listen(2) backlog sanity checks 47 * protocol-independent 48 * 49 * This module is effectively the top level interface to the BSD socket 50 * paradigm. 51 * 52 * Based upon Swansea University Computer Society NET3.039 53 */ 54 55 #include <linux/bpf-cgroup.h> 56 #include <linux/ethtool.h> 57 #include <linux/mm.h> 58 #include <linux/socket.h> 59 #include <linux/file.h> 60 #include <linux/splice.h> 61 #include <linux/net.h> 62 #include <linux/interrupt.h> 63 #include <linux/thread_info.h> 64 #include <linux/rcupdate.h> 65 #include <linux/netdevice.h> 66 #include <linux/proc_fs.h> 67 #include <linux/seq_file.h> 68 #include <linux/mutex.h> 69 #include <linux/if_bridge.h> 70 #include <linux/if_vlan.h> 71 #include <linux/ptp_classify.h> 72 #include <linux/init.h> 73 #include <linux/poll.h> 74 #include <linux/cache.h> 75 #include <linux/module.h> 76 #include <linux/highmem.h> 77 #include <linux/mount.h> 78 #include <linux/pseudo_fs.h> 79 #include <linux/security.h> 80 #include <linux/syscalls.h> 81 #include <linux/compat.h> 82 #include <linux/kmod.h> 83 #include <linux/audit.h> 84 #include <linux/wireless.h> 85 #include <linux/nsproxy.h> 86 #include <linux/magic.h> 87 #include <linux/slab.h> 88 #include <linux/xattr.h> 89 #include <linux/nospec.h> 90 #include <linux/indirect_call_wrapper.h> 91 #include <linux/io_uring/net.h> 92 93 #include <linux/uaccess.h> 94 #include <asm/unistd.h> 95 96 #include <net/compat.h> 97 #include <net/wext.h> 98 #include <net/cls_cgroup.h> 99 100 #include <net/sock.h> 101 #include <linux/netfilter.h> 102 103 #include <linux/if_tun.h> 104 #include <linux/ipv6_route.h> 105 #include <linux/route.h> 106 #include <linux/termios.h> 107 #include <linux/sockios.h> 108 #include <net/busy_poll.h> 109 #include <linux/errqueue.h> 110 #include <linux/ptp_clock_kernel.h> 111 #include <trace/events/sock.h> 112 113 #ifdef CONFIG_NET_RX_BUSY_POLL 114 unsigned int sysctl_net_busy_read __read_mostly; 115 unsigned int sysctl_net_busy_poll __read_mostly; 116 #endif 117 118 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to); 119 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from); 120 static int sock_mmap(struct file *file, struct vm_area_struct *vma); 121 122 static int sock_close(struct inode *inode, struct file *file); 123 static __poll_t sock_poll(struct file *file, 124 struct poll_table_struct *wait); 125 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 126 #ifdef CONFIG_COMPAT 127 static long compat_sock_ioctl(struct file *file, 128 unsigned int cmd, unsigned long arg); 129 #endif 130 static int sock_fasync(int fd, struct file *filp, int on); 131 static ssize_t sock_splice_read(struct file *file, loff_t *ppos, 132 struct pipe_inode_info *pipe, size_t len, 133 unsigned int flags); 134 static void sock_splice_eof(struct file *file); 135 136 #ifdef CONFIG_PROC_FS 137 static void sock_show_fdinfo(struct seq_file *m, struct file *f) 138 { 139 struct socket *sock = f->private_data; 140 const struct proto_ops *ops = READ_ONCE(sock->ops); 141 142 if (ops->show_fdinfo) 143 ops->show_fdinfo(m, sock); 144 } 145 #else 146 #define sock_show_fdinfo NULL 147 #endif 148 149 /* 150 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear 151 * in the operation structures but are done directly via the socketcall() multiplexor. 152 */ 153 154 static const struct file_operations socket_file_ops = { 155 .owner = THIS_MODULE, 156 .read_iter = sock_read_iter, 157 .write_iter = sock_write_iter, 158 .poll = sock_poll, 159 .unlocked_ioctl = sock_ioctl, 160 #ifdef CONFIG_COMPAT 161 .compat_ioctl = compat_sock_ioctl, 162 #endif 163 .uring_cmd = io_uring_cmd_sock, 164 .mmap = sock_mmap, 165 .release = sock_close, 166 .fasync = sock_fasync, 167 .splice_write = splice_to_socket, 168 .splice_read = sock_splice_read, 169 .splice_eof = sock_splice_eof, 170 .show_fdinfo = sock_show_fdinfo, 171 }; 172 173 static const char * const pf_family_names[] = { 174 [PF_UNSPEC] = "PF_UNSPEC", 175 [PF_UNIX] = "PF_UNIX/PF_LOCAL", 176 [PF_INET] = "PF_INET", 177 [PF_AX25] = "PF_AX25", 178 [PF_IPX] = "PF_IPX", 179 [PF_APPLETALK] = "PF_APPLETALK", 180 [PF_NETROM] = "PF_NETROM", 181 [PF_BRIDGE] = "PF_BRIDGE", 182 [PF_ATMPVC] = "PF_ATMPVC", 183 [PF_X25] = "PF_X25", 184 [PF_INET6] = "PF_INET6", 185 [PF_ROSE] = "PF_ROSE", 186 [PF_DECnet] = "PF_DECnet", 187 [PF_NETBEUI] = "PF_NETBEUI", 188 [PF_SECURITY] = "PF_SECURITY", 189 [PF_KEY] = "PF_KEY", 190 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE", 191 [PF_PACKET] = "PF_PACKET", 192 [PF_ASH] = "PF_ASH", 193 [PF_ECONET] = "PF_ECONET", 194 [PF_ATMSVC] = "PF_ATMSVC", 195 [PF_RDS] = "PF_RDS", 196 [PF_SNA] = "PF_SNA", 197 [PF_IRDA] = "PF_IRDA", 198 [PF_PPPOX] = "PF_PPPOX", 199 [PF_WANPIPE] = "PF_WANPIPE", 200 [PF_LLC] = "PF_LLC", 201 [PF_IB] = "PF_IB", 202 [PF_MPLS] = "PF_MPLS", 203 [PF_CAN] = "PF_CAN", 204 [PF_TIPC] = "PF_TIPC", 205 [PF_BLUETOOTH] = "PF_BLUETOOTH", 206 [PF_IUCV] = "PF_IUCV", 207 [PF_RXRPC] = "PF_RXRPC", 208 [PF_ISDN] = "PF_ISDN", 209 [PF_PHONET] = "PF_PHONET", 210 [PF_IEEE802154] = "PF_IEEE802154", 211 [PF_CAIF] = "PF_CAIF", 212 [PF_ALG] = "PF_ALG", 213 [PF_NFC] = "PF_NFC", 214 [PF_VSOCK] = "PF_VSOCK", 215 [PF_KCM] = "PF_KCM", 216 [PF_QIPCRTR] = "PF_QIPCRTR", 217 [PF_SMC] = "PF_SMC", 218 [PF_XDP] = "PF_XDP", 219 [PF_MCTP] = "PF_MCTP", 220 }; 221 222 /* 223 * The protocol list. Each protocol is registered in here. 224 */ 225 226 static DEFINE_SPINLOCK(net_family_lock); 227 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly; 228 229 /* 230 * Support routines. 231 * Move socket addresses back and forth across the kernel/user 232 * divide and look after the messy bits. 233 */ 234 235 /** 236 * move_addr_to_kernel - copy a socket address into kernel space 237 * @uaddr: Address in user space 238 * @kaddr: Address in kernel space 239 * @ulen: Length in user space 240 * 241 * The address is copied into kernel space. If the provided address is 242 * too long an error code of -EINVAL is returned. If the copy gives 243 * invalid addresses -EFAULT is returned. On a success 0 is returned. 244 */ 245 246 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr) 247 { 248 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage)) 249 return -EINVAL; 250 if (ulen == 0) 251 return 0; 252 if (copy_from_user(kaddr, uaddr, ulen)) 253 return -EFAULT; 254 return audit_sockaddr(ulen, kaddr); 255 } 256 257 /** 258 * move_addr_to_user - copy an address to user space 259 * @kaddr: kernel space address 260 * @klen: length of address in kernel 261 * @uaddr: user space address 262 * @ulen: pointer to user length field 263 * 264 * The value pointed to by ulen on entry is the buffer length available. 265 * This is overwritten with the buffer space used. -EINVAL is returned 266 * if an overlong buffer is specified or a negative buffer size. -EFAULT 267 * is returned if either the buffer or the length field are not 268 * accessible. 269 * After copying the data up to the limit the user specifies, the true 270 * length of the data is written over the length limit the user 271 * specified. Zero is returned for a success. 272 */ 273 274 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen, 275 void __user *uaddr, int __user *ulen) 276 { 277 int err; 278 int len; 279 280 BUG_ON(klen > sizeof(struct sockaddr_storage)); 281 err = get_user(len, ulen); 282 if (err) 283 return err; 284 if (len > klen) 285 len = klen; 286 if (len < 0) 287 return -EINVAL; 288 if (len) { 289 if (audit_sockaddr(klen, kaddr)) 290 return -ENOMEM; 291 if (copy_to_user(uaddr, kaddr, len)) 292 return -EFAULT; 293 } 294 /* 295 * "fromlen shall refer to the value before truncation.." 296 * 1003.1g 297 */ 298 return __put_user(klen, ulen); 299 } 300 301 static struct kmem_cache *sock_inode_cachep __ro_after_init; 302 303 static struct inode *sock_alloc_inode(struct super_block *sb) 304 { 305 struct socket_alloc *ei; 306 307 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL); 308 if (!ei) 309 return NULL; 310 init_waitqueue_head(&ei->socket.wq.wait); 311 ei->socket.wq.fasync_list = NULL; 312 ei->socket.wq.flags = 0; 313 314 ei->socket.state = SS_UNCONNECTED; 315 ei->socket.flags = 0; 316 ei->socket.ops = NULL; 317 ei->socket.sk = NULL; 318 ei->socket.file = NULL; 319 320 return &ei->vfs_inode; 321 } 322 323 static void sock_free_inode(struct inode *inode) 324 { 325 struct socket_alloc *ei; 326 327 ei = container_of(inode, struct socket_alloc, vfs_inode); 328 kmem_cache_free(sock_inode_cachep, ei); 329 } 330 331 static void init_once(void *foo) 332 { 333 struct socket_alloc *ei = (struct socket_alloc *)foo; 334 335 inode_init_once(&ei->vfs_inode); 336 } 337 338 static void init_inodecache(void) 339 { 340 sock_inode_cachep = kmem_cache_create("sock_inode_cache", 341 sizeof(struct socket_alloc), 342 0, 343 (SLAB_HWCACHE_ALIGN | 344 SLAB_RECLAIM_ACCOUNT | 345 SLAB_ACCOUNT), 346 init_once); 347 BUG_ON(sock_inode_cachep == NULL); 348 } 349 350 static const struct super_operations sockfs_ops = { 351 .alloc_inode = sock_alloc_inode, 352 .free_inode = sock_free_inode, 353 .statfs = simple_statfs, 354 }; 355 356 /* 357 * sockfs_dname() is called from d_path(). 358 */ 359 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen) 360 { 361 return dynamic_dname(buffer, buflen, "socket:[%lu]", 362 d_inode(dentry)->i_ino); 363 } 364 365 static const struct dentry_operations sockfs_dentry_operations = { 366 .d_dname = sockfs_dname, 367 }; 368 369 static int sockfs_xattr_get(const struct xattr_handler *handler, 370 struct dentry *dentry, struct inode *inode, 371 const char *suffix, void *value, size_t size) 372 { 373 if (value) { 374 if (dentry->d_name.len + 1 > size) 375 return -ERANGE; 376 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1); 377 } 378 return dentry->d_name.len + 1; 379 } 380 381 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname" 382 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX) 383 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1) 384 385 static const struct xattr_handler sockfs_xattr_handler = { 386 .name = XATTR_NAME_SOCKPROTONAME, 387 .get = sockfs_xattr_get, 388 }; 389 390 static int sockfs_security_xattr_set(const struct xattr_handler *handler, 391 struct mnt_idmap *idmap, 392 struct dentry *dentry, struct inode *inode, 393 const char *suffix, const void *value, 394 size_t size, int flags) 395 { 396 /* Handled by LSM. */ 397 return -EAGAIN; 398 } 399 400 static const struct xattr_handler sockfs_security_xattr_handler = { 401 .prefix = XATTR_SECURITY_PREFIX, 402 .set = sockfs_security_xattr_set, 403 }; 404 405 static const struct xattr_handler * const sockfs_xattr_handlers[] = { 406 &sockfs_xattr_handler, 407 &sockfs_security_xattr_handler, 408 NULL 409 }; 410 411 static int sockfs_init_fs_context(struct fs_context *fc) 412 { 413 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC); 414 if (!ctx) 415 return -ENOMEM; 416 ctx->ops = &sockfs_ops; 417 ctx->dops = &sockfs_dentry_operations; 418 ctx->xattr = sockfs_xattr_handlers; 419 return 0; 420 } 421 422 static struct vfsmount *sock_mnt __read_mostly; 423 424 static struct file_system_type sock_fs_type = { 425 .name = "sockfs", 426 .init_fs_context = sockfs_init_fs_context, 427 .kill_sb = kill_anon_super, 428 }; 429 430 /* 431 * Obtains the first available file descriptor and sets it up for use. 432 * 433 * These functions create file structures and maps them to fd space 434 * of the current process. On success it returns file descriptor 435 * and file struct implicitly stored in sock->file. 436 * Note that another thread may close file descriptor before we return 437 * from this function. We use the fact that now we do not refer 438 * to socket after mapping. If one day we will need it, this 439 * function will increment ref. count on file by 1. 440 * 441 * In any case returned fd MAY BE not valid! 442 * This race condition is unavoidable 443 * with shared fd spaces, we cannot solve it inside kernel, 444 * but we take care of internal coherence yet. 445 */ 446 447 /** 448 * sock_alloc_file - Bind a &socket to a &file 449 * @sock: socket 450 * @flags: file status flags 451 * @dname: protocol name 452 * 453 * Returns the &file bound with @sock, implicitly storing it 454 * in sock->file. If dname is %NULL, sets to "". 455 * 456 * On failure @sock is released, and an ERR pointer is returned. 457 * 458 * This function uses GFP_KERNEL internally. 459 */ 460 461 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname) 462 { 463 struct file *file; 464 465 if (!dname) 466 dname = sock->sk ? sock->sk->sk_prot_creator->name : ""; 467 468 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname, 469 O_RDWR | (flags & O_NONBLOCK), 470 &socket_file_ops); 471 if (IS_ERR(file)) { 472 sock_release(sock); 473 return file; 474 } 475 476 file->f_mode |= FMODE_NOWAIT; 477 sock->file = file; 478 file->private_data = sock; 479 stream_open(SOCK_INODE(sock), file); 480 return file; 481 } 482 EXPORT_SYMBOL(sock_alloc_file); 483 484 static int sock_map_fd(struct socket *sock, int flags) 485 { 486 struct file *newfile; 487 int fd = get_unused_fd_flags(flags); 488 if (unlikely(fd < 0)) { 489 sock_release(sock); 490 return fd; 491 } 492 493 newfile = sock_alloc_file(sock, flags, NULL); 494 if (!IS_ERR(newfile)) { 495 fd_install(fd, newfile); 496 return fd; 497 } 498 499 put_unused_fd(fd); 500 return PTR_ERR(newfile); 501 } 502 503 /** 504 * sock_from_file - Return the &socket bounded to @file. 505 * @file: file 506 * 507 * On failure returns %NULL. 508 */ 509 510 struct socket *sock_from_file(struct file *file) 511 { 512 if (file->f_op == &socket_file_ops) 513 return file->private_data; /* set in sock_alloc_file */ 514 515 return NULL; 516 } 517 EXPORT_SYMBOL(sock_from_file); 518 519 /** 520 * sockfd_lookup - Go from a file number to its socket slot 521 * @fd: file handle 522 * @err: pointer to an error code return 523 * 524 * The file handle passed in is locked and the socket it is bound 525 * to is returned. If an error occurs the err pointer is overwritten 526 * with a negative errno code and NULL is returned. The function checks 527 * for both invalid handles and passing a handle which is not a socket. 528 * 529 * On a success the socket object pointer is returned. 530 */ 531 532 struct socket *sockfd_lookup(int fd, int *err) 533 { 534 struct file *file; 535 struct socket *sock; 536 537 file = fget(fd); 538 if (!file) { 539 *err = -EBADF; 540 return NULL; 541 } 542 543 sock = sock_from_file(file); 544 if (!sock) { 545 *err = -ENOTSOCK; 546 fput(file); 547 } 548 return sock; 549 } 550 EXPORT_SYMBOL(sockfd_lookup); 551 552 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed) 553 { 554 struct fd f = fdget(fd); 555 struct socket *sock; 556 557 *err = -EBADF; 558 if (fd_file(f)) { 559 sock = sock_from_file(fd_file(f)); 560 if (likely(sock)) { 561 *fput_needed = f.word & FDPUT_FPUT; 562 return sock; 563 } 564 *err = -ENOTSOCK; 565 fdput(f); 566 } 567 return NULL; 568 } 569 570 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer, 571 size_t size) 572 { 573 ssize_t len; 574 ssize_t used = 0; 575 576 len = security_inode_listsecurity(d_inode(dentry), buffer, size); 577 if (len < 0) 578 return len; 579 used += len; 580 if (buffer) { 581 if (size < used) 582 return -ERANGE; 583 buffer += len; 584 } 585 586 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1); 587 used += len; 588 if (buffer) { 589 if (size < used) 590 return -ERANGE; 591 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len); 592 buffer += len; 593 } 594 595 return used; 596 } 597 598 static int sockfs_setattr(struct mnt_idmap *idmap, 599 struct dentry *dentry, struct iattr *iattr) 600 { 601 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr); 602 603 if (!err && (iattr->ia_valid & ATTR_UID)) { 604 struct socket *sock = SOCKET_I(d_inode(dentry)); 605 606 if (sock->sk) 607 sock->sk->sk_uid = iattr->ia_uid; 608 else 609 err = -ENOENT; 610 } 611 612 return err; 613 } 614 615 static const struct inode_operations sockfs_inode_ops = { 616 .listxattr = sockfs_listxattr, 617 .setattr = sockfs_setattr, 618 }; 619 620 /** 621 * sock_alloc - allocate a socket 622 * 623 * Allocate a new inode and socket object. The two are bound together 624 * and initialised. The socket is then returned. If we are out of inodes 625 * NULL is returned. This functions uses GFP_KERNEL internally. 626 */ 627 628 struct socket *sock_alloc(void) 629 { 630 struct inode *inode; 631 struct socket *sock; 632 633 inode = new_inode_pseudo(sock_mnt->mnt_sb); 634 if (!inode) 635 return NULL; 636 637 sock = SOCKET_I(inode); 638 639 inode->i_ino = get_next_ino(); 640 inode->i_mode = S_IFSOCK | S_IRWXUGO; 641 inode->i_uid = current_fsuid(); 642 inode->i_gid = current_fsgid(); 643 inode->i_op = &sockfs_inode_ops; 644 645 return sock; 646 } 647 EXPORT_SYMBOL(sock_alloc); 648 649 static void __sock_release(struct socket *sock, struct inode *inode) 650 { 651 const struct proto_ops *ops = READ_ONCE(sock->ops); 652 653 if (ops) { 654 struct module *owner = ops->owner; 655 656 if (inode) 657 inode_lock(inode); 658 ops->release(sock); 659 sock->sk = NULL; 660 if (inode) 661 inode_unlock(inode); 662 sock->ops = NULL; 663 module_put(owner); 664 } 665 666 if (sock->wq.fasync_list) 667 pr_err("%s: fasync list not empty!\n", __func__); 668 669 if (!sock->file) { 670 iput(SOCK_INODE(sock)); 671 return; 672 } 673 sock->file = NULL; 674 } 675 676 /** 677 * sock_release - close a socket 678 * @sock: socket to close 679 * 680 * The socket is released from the protocol stack if it has a release 681 * callback, and the inode is then released if the socket is bound to 682 * an inode not a file. 683 */ 684 void sock_release(struct socket *sock) 685 { 686 __sock_release(sock, NULL); 687 } 688 EXPORT_SYMBOL(sock_release); 689 690 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags) 691 { 692 u8 flags = *tx_flags; 693 694 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) { 695 flags |= SKBTX_HW_TSTAMP; 696 697 /* PTP hardware clocks can provide a free running cycle counter 698 * as a time base for virtual clocks. Tell driver to use the 699 * free running cycle counter for timestamp if socket is bound 700 * to virtual clock. 701 */ 702 if (tsflags & SOF_TIMESTAMPING_BIND_PHC) 703 flags |= SKBTX_HW_TSTAMP_USE_CYCLES; 704 } 705 706 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE) 707 flags |= SKBTX_SW_TSTAMP; 708 709 if (tsflags & SOF_TIMESTAMPING_TX_SCHED) 710 flags |= SKBTX_SCHED_TSTAMP; 711 712 *tx_flags = flags; 713 } 714 EXPORT_SYMBOL(__sock_tx_timestamp); 715 716 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *, 717 size_t)); 718 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *, 719 size_t)); 720 721 static noinline void call_trace_sock_send_length(struct sock *sk, int ret, 722 int flags) 723 { 724 trace_sock_send_length(sk, ret, 0); 725 } 726 727 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg) 728 { 729 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg, 730 inet_sendmsg, sock, msg, 731 msg_data_left(msg)); 732 BUG_ON(ret == -EIOCBQUEUED); 733 734 if (trace_sock_send_length_enabled()) 735 call_trace_sock_send_length(sock->sk, ret, 0); 736 return ret; 737 } 738 739 static int __sock_sendmsg(struct socket *sock, struct msghdr *msg) 740 { 741 int err = security_socket_sendmsg(sock, msg, 742 msg_data_left(msg)); 743 744 return err ?: sock_sendmsg_nosec(sock, msg); 745 } 746 747 /** 748 * sock_sendmsg - send a message through @sock 749 * @sock: socket 750 * @msg: message to send 751 * 752 * Sends @msg through @sock, passing through LSM. 753 * Returns the number of bytes sent, or an error code. 754 */ 755 int sock_sendmsg(struct socket *sock, struct msghdr *msg) 756 { 757 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name; 758 struct sockaddr_storage address; 759 int save_len = msg->msg_namelen; 760 int ret; 761 762 if (msg->msg_name) { 763 memcpy(&address, msg->msg_name, msg->msg_namelen); 764 msg->msg_name = &address; 765 } 766 767 ret = __sock_sendmsg(sock, msg); 768 msg->msg_name = save_addr; 769 msg->msg_namelen = save_len; 770 771 return ret; 772 } 773 EXPORT_SYMBOL(sock_sendmsg); 774 775 /** 776 * kernel_sendmsg - send a message through @sock (kernel-space) 777 * @sock: socket 778 * @msg: message header 779 * @vec: kernel vec 780 * @num: vec array length 781 * @size: total message data size 782 * 783 * Builds the message data with @vec and sends it through @sock. 784 * Returns the number of bytes sent, or an error code. 785 */ 786 787 int kernel_sendmsg(struct socket *sock, struct msghdr *msg, 788 struct kvec *vec, size_t num, size_t size) 789 { 790 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size); 791 return sock_sendmsg(sock, msg); 792 } 793 EXPORT_SYMBOL(kernel_sendmsg); 794 795 /** 796 * kernel_sendmsg_locked - send a message through @sock (kernel-space) 797 * @sk: sock 798 * @msg: message header 799 * @vec: output s/g array 800 * @num: output s/g array length 801 * @size: total message data size 802 * 803 * Builds the message data with @vec and sends it through @sock. 804 * Returns the number of bytes sent, or an error code. 805 * Caller must hold @sk. 806 */ 807 808 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg, 809 struct kvec *vec, size_t num, size_t size) 810 { 811 struct socket *sock = sk->sk_socket; 812 const struct proto_ops *ops = READ_ONCE(sock->ops); 813 814 if (!ops->sendmsg_locked) 815 return sock_no_sendmsg_locked(sk, msg, size); 816 817 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size); 818 819 return ops->sendmsg_locked(sk, msg, msg_data_left(msg)); 820 } 821 EXPORT_SYMBOL(kernel_sendmsg_locked); 822 823 static bool skb_is_err_queue(const struct sk_buff *skb) 824 { 825 /* pkt_type of skbs enqueued on the error queue are set to 826 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do 827 * in recvmsg, since skbs received on a local socket will never 828 * have a pkt_type of PACKET_OUTGOING. 829 */ 830 return skb->pkt_type == PACKET_OUTGOING; 831 } 832 833 /* On transmit, software and hardware timestamps are returned independently. 834 * As the two skb clones share the hardware timestamp, which may be updated 835 * before the software timestamp is received, a hardware TX timestamp may be 836 * returned only if there is no software TX timestamp. Ignore false software 837 * timestamps, which may be made in the __sock_recv_timestamp() call when the 838 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a 839 * hardware timestamp. 840 */ 841 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp) 842 { 843 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb); 844 } 845 846 static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index) 847 { 848 bool cycles = READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_BIND_PHC; 849 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 850 struct net_device *orig_dev; 851 ktime_t hwtstamp; 852 853 rcu_read_lock(); 854 orig_dev = dev_get_by_napi_id(skb_napi_id(skb)); 855 if (orig_dev) { 856 *if_index = orig_dev->ifindex; 857 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles); 858 } else { 859 hwtstamp = shhwtstamps->hwtstamp; 860 } 861 rcu_read_unlock(); 862 863 return hwtstamp; 864 } 865 866 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb, 867 int if_index) 868 { 869 struct scm_ts_pktinfo ts_pktinfo; 870 struct net_device *orig_dev; 871 872 if (!skb_mac_header_was_set(skb)) 873 return; 874 875 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo)); 876 877 if (!if_index) { 878 rcu_read_lock(); 879 orig_dev = dev_get_by_napi_id(skb_napi_id(skb)); 880 if (orig_dev) 881 if_index = orig_dev->ifindex; 882 rcu_read_unlock(); 883 } 884 ts_pktinfo.if_index = if_index; 885 886 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb); 887 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO, 888 sizeof(ts_pktinfo), &ts_pktinfo); 889 } 890 891 /* 892 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP) 893 */ 894 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 895 struct sk_buff *skb) 896 { 897 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP); 898 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW); 899 struct scm_timestamping_internal tss; 900 int empty = 1, false_tstamp = 0; 901 struct skb_shared_hwtstamps *shhwtstamps = 902 skb_hwtstamps(skb); 903 int if_index; 904 ktime_t hwtstamp; 905 u32 tsflags; 906 907 /* Race occurred between timestamp enabling and packet 908 receiving. Fill in the current time for now. */ 909 if (need_software_tstamp && skb->tstamp == 0) { 910 __net_timestamp(skb); 911 false_tstamp = 1; 912 } 913 914 if (need_software_tstamp) { 915 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) { 916 if (new_tstamp) { 917 struct __kernel_sock_timeval tv; 918 919 skb_get_new_timestamp(skb, &tv); 920 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW, 921 sizeof(tv), &tv); 922 } else { 923 struct __kernel_old_timeval tv; 924 925 skb_get_timestamp(skb, &tv); 926 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD, 927 sizeof(tv), &tv); 928 } 929 } else { 930 if (new_tstamp) { 931 struct __kernel_timespec ts; 932 933 skb_get_new_timestampns(skb, &ts); 934 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW, 935 sizeof(ts), &ts); 936 } else { 937 struct __kernel_old_timespec ts; 938 939 skb_get_timestampns(skb, &ts); 940 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD, 941 sizeof(ts), &ts); 942 } 943 } 944 } 945 946 memset(&tss, 0, sizeof(tss)); 947 tsflags = READ_ONCE(sk->sk_tsflags); 948 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE && 949 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE || 950 skb_is_err_queue(skb) || 951 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) && 952 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0)) 953 empty = 0; 954 if (shhwtstamps && 955 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE && 956 (tsflags & SOF_TIMESTAMPING_RX_HARDWARE || 957 skb_is_err_queue(skb) || 958 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) && 959 !skb_is_swtx_tstamp(skb, false_tstamp)) { 960 if_index = 0; 961 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV) 962 hwtstamp = get_timestamp(sk, skb, &if_index); 963 else 964 hwtstamp = shhwtstamps->hwtstamp; 965 966 if (tsflags & SOF_TIMESTAMPING_BIND_PHC) 967 hwtstamp = ptp_convert_timestamp(&hwtstamp, 968 READ_ONCE(sk->sk_bind_phc)); 969 970 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) { 971 empty = 0; 972 973 if ((tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) && 974 !skb_is_err_queue(skb)) 975 put_ts_pktinfo(msg, skb, if_index); 976 } 977 } 978 if (!empty) { 979 if (sock_flag(sk, SOCK_TSTAMP_NEW)) 980 put_cmsg_scm_timestamping64(msg, &tss); 981 else 982 put_cmsg_scm_timestamping(msg, &tss); 983 984 if (skb_is_err_queue(skb) && skb->len && 985 SKB_EXT_ERR(skb)->opt_stats) 986 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS, 987 skb->len, skb->data); 988 } 989 } 990 EXPORT_SYMBOL_GPL(__sock_recv_timestamp); 991 992 #ifdef CONFIG_WIRELESS 993 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 994 struct sk_buff *skb) 995 { 996 int ack; 997 998 if (!sock_flag(sk, SOCK_WIFI_STATUS)) 999 return; 1000 if (!skb->wifi_acked_valid) 1001 return; 1002 1003 ack = skb->wifi_acked; 1004 1005 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack); 1006 } 1007 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status); 1008 #endif 1009 1010 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk, 1011 struct sk_buff *skb) 1012 { 1013 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount) 1014 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL, 1015 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount); 1016 } 1017 1018 static void sock_recv_mark(struct msghdr *msg, struct sock *sk, 1019 struct sk_buff *skb) 1020 { 1021 if (sock_flag(sk, SOCK_RCVMARK) && skb) { 1022 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */ 1023 __u32 mark = skb->mark; 1024 1025 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark); 1026 } 1027 } 1028 1029 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 1030 struct sk_buff *skb) 1031 { 1032 sock_recv_timestamp(msg, sk, skb); 1033 sock_recv_drops(msg, sk, skb); 1034 sock_recv_mark(msg, sk, skb); 1035 } 1036 EXPORT_SYMBOL_GPL(__sock_recv_cmsgs); 1037 1038 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *, 1039 size_t, int)); 1040 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *, 1041 size_t, int)); 1042 1043 static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags) 1044 { 1045 trace_sock_recv_length(sk, ret, flags); 1046 } 1047 1048 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg, 1049 int flags) 1050 { 1051 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg, 1052 inet6_recvmsg, 1053 inet_recvmsg, sock, msg, 1054 msg_data_left(msg), flags); 1055 if (trace_sock_recv_length_enabled()) 1056 call_trace_sock_recv_length(sock->sk, ret, flags); 1057 return ret; 1058 } 1059 1060 /** 1061 * sock_recvmsg - receive a message from @sock 1062 * @sock: socket 1063 * @msg: message to receive 1064 * @flags: message flags 1065 * 1066 * Receives @msg from @sock, passing through LSM. Returns the total number 1067 * of bytes received, or an error. 1068 */ 1069 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags) 1070 { 1071 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags); 1072 1073 return err ?: sock_recvmsg_nosec(sock, msg, flags); 1074 } 1075 EXPORT_SYMBOL(sock_recvmsg); 1076 1077 /** 1078 * kernel_recvmsg - Receive a message from a socket (kernel space) 1079 * @sock: The socket to receive the message from 1080 * @msg: Received message 1081 * @vec: Input s/g array for message data 1082 * @num: Size of input s/g array 1083 * @size: Number of bytes to read 1084 * @flags: Message flags (MSG_DONTWAIT, etc...) 1085 * 1086 * On return the msg structure contains the scatter/gather array passed in the 1087 * vec argument. The array is modified so that it consists of the unfilled 1088 * portion of the original array. 1089 * 1090 * The returned value is the total number of bytes received, or an error. 1091 */ 1092 1093 int kernel_recvmsg(struct socket *sock, struct msghdr *msg, 1094 struct kvec *vec, size_t num, size_t size, int flags) 1095 { 1096 msg->msg_control_is_user = false; 1097 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size); 1098 return sock_recvmsg(sock, msg, flags); 1099 } 1100 EXPORT_SYMBOL(kernel_recvmsg); 1101 1102 static ssize_t sock_splice_read(struct file *file, loff_t *ppos, 1103 struct pipe_inode_info *pipe, size_t len, 1104 unsigned int flags) 1105 { 1106 struct socket *sock = file->private_data; 1107 const struct proto_ops *ops; 1108 1109 ops = READ_ONCE(sock->ops); 1110 if (unlikely(!ops->splice_read)) 1111 return copy_splice_read(file, ppos, pipe, len, flags); 1112 1113 return ops->splice_read(sock, ppos, pipe, len, flags); 1114 } 1115 1116 static void sock_splice_eof(struct file *file) 1117 { 1118 struct socket *sock = file->private_data; 1119 const struct proto_ops *ops; 1120 1121 ops = READ_ONCE(sock->ops); 1122 if (ops->splice_eof) 1123 ops->splice_eof(sock); 1124 } 1125 1126 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to) 1127 { 1128 struct file *file = iocb->ki_filp; 1129 struct socket *sock = file->private_data; 1130 struct msghdr msg = {.msg_iter = *to, 1131 .msg_iocb = iocb}; 1132 ssize_t res; 1133 1134 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT)) 1135 msg.msg_flags = MSG_DONTWAIT; 1136 1137 if (iocb->ki_pos != 0) 1138 return -ESPIPE; 1139 1140 if (!iov_iter_count(to)) /* Match SYS5 behaviour */ 1141 return 0; 1142 1143 res = sock_recvmsg(sock, &msg, msg.msg_flags); 1144 *to = msg.msg_iter; 1145 return res; 1146 } 1147 1148 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from) 1149 { 1150 struct file *file = iocb->ki_filp; 1151 struct socket *sock = file->private_data; 1152 struct msghdr msg = {.msg_iter = *from, 1153 .msg_iocb = iocb}; 1154 ssize_t res; 1155 1156 if (iocb->ki_pos != 0) 1157 return -ESPIPE; 1158 1159 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT)) 1160 msg.msg_flags = MSG_DONTWAIT; 1161 1162 if (sock->type == SOCK_SEQPACKET) 1163 msg.msg_flags |= MSG_EOR; 1164 1165 res = __sock_sendmsg(sock, &msg); 1166 *from = msg.msg_iter; 1167 return res; 1168 } 1169 1170 /* 1171 * Atomic setting of ioctl hooks to avoid race 1172 * with module unload. 1173 */ 1174 1175 static DEFINE_MUTEX(br_ioctl_mutex); 1176 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br, 1177 unsigned int cmd, struct ifreq *ifr, 1178 void __user *uarg); 1179 1180 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br, 1181 unsigned int cmd, struct ifreq *ifr, 1182 void __user *uarg)) 1183 { 1184 mutex_lock(&br_ioctl_mutex); 1185 br_ioctl_hook = hook; 1186 mutex_unlock(&br_ioctl_mutex); 1187 } 1188 EXPORT_SYMBOL(brioctl_set); 1189 1190 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd, 1191 struct ifreq *ifr, void __user *uarg) 1192 { 1193 int err = -ENOPKG; 1194 1195 if (!br_ioctl_hook) 1196 request_module("bridge"); 1197 1198 mutex_lock(&br_ioctl_mutex); 1199 if (br_ioctl_hook) 1200 err = br_ioctl_hook(net, br, cmd, ifr, uarg); 1201 mutex_unlock(&br_ioctl_mutex); 1202 1203 return err; 1204 } 1205 1206 static DEFINE_MUTEX(vlan_ioctl_mutex); 1207 static int (*vlan_ioctl_hook) (struct net *, void __user *arg); 1208 1209 void vlan_ioctl_set(int (*hook) (struct net *, void __user *)) 1210 { 1211 mutex_lock(&vlan_ioctl_mutex); 1212 vlan_ioctl_hook = hook; 1213 mutex_unlock(&vlan_ioctl_mutex); 1214 } 1215 EXPORT_SYMBOL(vlan_ioctl_set); 1216 1217 static long sock_do_ioctl(struct net *net, struct socket *sock, 1218 unsigned int cmd, unsigned long arg) 1219 { 1220 const struct proto_ops *ops = READ_ONCE(sock->ops); 1221 struct ifreq ifr; 1222 bool need_copyout; 1223 int err; 1224 void __user *argp = (void __user *)arg; 1225 void __user *data; 1226 1227 err = ops->ioctl(sock, cmd, arg); 1228 1229 /* 1230 * If this ioctl is unknown try to hand it down 1231 * to the NIC driver. 1232 */ 1233 if (err != -ENOIOCTLCMD) 1234 return err; 1235 1236 if (!is_socket_ioctl_cmd(cmd)) 1237 return -ENOTTY; 1238 1239 if (get_user_ifreq(&ifr, &data, argp)) 1240 return -EFAULT; 1241 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout); 1242 if (!err && need_copyout) 1243 if (put_user_ifreq(&ifr, argp)) 1244 return -EFAULT; 1245 1246 return err; 1247 } 1248 1249 /* 1250 * With an ioctl, arg may well be a user mode pointer, but we don't know 1251 * what to do with it - that's up to the protocol still. 1252 */ 1253 1254 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg) 1255 { 1256 const struct proto_ops *ops; 1257 struct socket *sock; 1258 struct sock *sk; 1259 void __user *argp = (void __user *)arg; 1260 int pid, err; 1261 struct net *net; 1262 1263 sock = file->private_data; 1264 ops = READ_ONCE(sock->ops); 1265 sk = sock->sk; 1266 net = sock_net(sk); 1267 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) { 1268 struct ifreq ifr; 1269 void __user *data; 1270 bool need_copyout; 1271 if (get_user_ifreq(&ifr, &data, argp)) 1272 return -EFAULT; 1273 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout); 1274 if (!err && need_copyout) 1275 if (put_user_ifreq(&ifr, argp)) 1276 return -EFAULT; 1277 } else 1278 #ifdef CONFIG_WEXT_CORE 1279 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) { 1280 err = wext_handle_ioctl(net, cmd, argp); 1281 } else 1282 #endif 1283 switch (cmd) { 1284 case FIOSETOWN: 1285 case SIOCSPGRP: 1286 err = -EFAULT; 1287 if (get_user(pid, (int __user *)argp)) 1288 break; 1289 err = f_setown(sock->file, pid, 1); 1290 break; 1291 case FIOGETOWN: 1292 case SIOCGPGRP: 1293 err = put_user(f_getown(sock->file), 1294 (int __user *)argp); 1295 break; 1296 case SIOCGIFBR: 1297 case SIOCSIFBR: 1298 case SIOCBRADDBR: 1299 case SIOCBRDELBR: 1300 err = br_ioctl_call(net, NULL, cmd, NULL, argp); 1301 break; 1302 case SIOCGIFVLAN: 1303 case SIOCSIFVLAN: 1304 err = -ENOPKG; 1305 if (!vlan_ioctl_hook) 1306 request_module("8021q"); 1307 1308 mutex_lock(&vlan_ioctl_mutex); 1309 if (vlan_ioctl_hook) 1310 err = vlan_ioctl_hook(net, argp); 1311 mutex_unlock(&vlan_ioctl_mutex); 1312 break; 1313 case SIOCGSKNS: 1314 err = -EPERM; 1315 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1316 break; 1317 1318 err = open_related_ns(&net->ns, get_net_ns); 1319 break; 1320 case SIOCGSTAMP_OLD: 1321 case SIOCGSTAMPNS_OLD: 1322 if (!ops->gettstamp) { 1323 err = -ENOIOCTLCMD; 1324 break; 1325 } 1326 err = ops->gettstamp(sock, argp, 1327 cmd == SIOCGSTAMP_OLD, 1328 !IS_ENABLED(CONFIG_64BIT)); 1329 break; 1330 case SIOCGSTAMP_NEW: 1331 case SIOCGSTAMPNS_NEW: 1332 if (!ops->gettstamp) { 1333 err = -ENOIOCTLCMD; 1334 break; 1335 } 1336 err = ops->gettstamp(sock, argp, 1337 cmd == SIOCGSTAMP_NEW, 1338 false); 1339 break; 1340 1341 case SIOCGIFCONF: 1342 err = dev_ifconf(net, argp); 1343 break; 1344 1345 default: 1346 err = sock_do_ioctl(net, sock, cmd, arg); 1347 break; 1348 } 1349 return err; 1350 } 1351 1352 /** 1353 * sock_create_lite - creates a socket 1354 * @family: protocol family (AF_INET, ...) 1355 * @type: communication type (SOCK_STREAM, ...) 1356 * @protocol: protocol (0, ...) 1357 * @res: new socket 1358 * 1359 * Creates a new socket and assigns it to @res, passing through LSM. 1360 * The new socket initialization is not complete, see kernel_accept(). 1361 * Returns 0 or an error. On failure @res is set to %NULL. 1362 * This function internally uses GFP_KERNEL. 1363 */ 1364 1365 int sock_create_lite(int family, int type, int protocol, struct socket **res) 1366 { 1367 int err; 1368 struct socket *sock = NULL; 1369 1370 err = security_socket_create(family, type, protocol, 1); 1371 if (err) 1372 goto out; 1373 1374 sock = sock_alloc(); 1375 if (!sock) { 1376 err = -ENOMEM; 1377 goto out; 1378 } 1379 1380 sock->type = type; 1381 err = security_socket_post_create(sock, family, type, protocol, 1); 1382 if (err) 1383 goto out_release; 1384 1385 out: 1386 *res = sock; 1387 return err; 1388 out_release: 1389 sock_release(sock); 1390 sock = NULL; 1391 goto out; 1392 } 1393 EXPORT_SYMBOL(sock_create_lite); 1394 1395 /* No kernel lock held - perfect */ 1396 static __poll_t sock_poll(struct file *file, poll_table *wait) 1397 { 1398 struct socket *sock = file->private_data; 1399 const struct proto_ops *ops = READ_ONCE(sock->ops); 1400 __poll_t events = poll_requested_events(wait), flag = 0; 1401 1402 if (!ops->poll) 1403 return 0; 1404 1405 if (sk_can_busy_loop(sock->sk)) { 1406 /* poll once if requested by the syscall */ 1407 if (events & POLL_BUSY_LOOP) 1408 sk_busy_loop(sock->sk, 1); 1409 1410 /* if this socket can poll_ll, tell the system call */ 1411 flag = POLL_BUSY_LOOP; 1412 } 1413 1414 return ops->poll(file, sock, wait) | flag; 1415 } 1416 1417 static int sock_mmap(struct file *file, struct vm_area_struct *vma) 1418 { 1419 struct socket *sock = file->private_data; 1420 1421 return READ_ONCE(sock->ops)->mmap(file, sock, vma); 1422 } 1423 1424 static int sock_close(struct inode *inode, struct file *filp) 1425 { 1426 __sock_release(SOCKET_I(inode), inode); 1427 return 0; 1428 } 1429 1430 /* 1431 * Update the socket async list 1432 * 1433 * Fasync_list locking strategy. 1434 * 1435 * 1. fasync_list is modified only under process context socket lock 1436 * i.e. under semaphore. 1437 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock) 1438 * or under socket lock 1439 */ 1440 1441 static int sock_fasync(int fd, struct file *filp, int on) 1442 { 1443 struct socket *sock = filp->private_data; 1444 struct sock *sk = sock->sk; 1445 struct socket_wq *wq = &sock->wq; 1446 1447 if (sk == NULL) 1448 return -EINVAL; 1449 1450 lock_sock(sk); 1451 fasync_helper(fd, filp, on, &wq->fasync_list); 1452 1453 if (!wq->fasync_list) 1454 sock_reset_flag(sk, SOCK_FASYNC); 1455 else 1456 sock_set_flag(sk, SOCK_FASYNC); 1457 1458 release_sock(sk); 1459 return 0; 1460 } 1461 1462 /* This function may be called only under rcu_lock */ 1463 1464 int sock_wake_async(struct socket_wq *wq, int how, int band) 1465 { 1466 if (!wq || !wq->fasync_list) 1467 return -1; 1468 1469 switch (how) { 1470 case SOCK_WAKE_WAITD: 1471 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags)) 1472 break; 1473 goto call_kill; 1474 case SOCK_WAKE_SPACE: 1475 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags)) 1476 break; 1477 fallthrough; 1478 case SOCK_WAKE_IO: 1479 call_kill: 1480 kill_fasync(&wq->fasync_list, SIGIO, band); 1481 break; 1482 case SOCK_WAKE_URG: 1483 kill_fasync(&wq->fasync_list, SIGURG, band); 1484 } 1485 1486 return 0; 1487 } 1488 EXPORT_SYMBOL(sock_wake_async); 1489 1490 /** 1491 * __sock_create - creates a socket 1492 * @net: net namespace 1493 * @family: protocol family (AF_INET, ...) 1494 * @type: communication type (SOCK_STREAM, ...) 1495 * @protocol: protocol (0, ...) 1496 * @res: new socket 1497 * @kern: boolean for kernel space sockets 1498 * 1499 * Creates a new socket and assigns it to @res, passing through LSM. 1500 * Returns 0 or an error. On failure @res is set to %NULL. @kern must 1501 * be set to true if the socket resides in kernel space. 1502 * This function internally uses GFP_KERNEL. 1503 */ 1504 1505 int __sock_create(struct net *net, int family, int type, int protocol, 1506 struct socket **res, int kern) 1507 { 1508 int err; 1509 struct socket *sock; 1510 const struct net_proto_family *pf; 1511 1512 /* 1513 * Check protocol is in range 1514 */ 1515 if (family < 0 || family >= NPROTO) 1516 return -EAFNOSUPPORT; 1517 if (type < 0 || type >= SOCK_MAX) 1518 return -EINVAL; 1519 1520 /* Compatibility. 1521 1522 This uglymoron is moved from INET layer to here to avoid 1523 deadlock in module load. 1524 */ 1525 if (family == PF_INET && type == SOCK_PACKET) { 1526 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n", 1527 current->comm); 1528 family = PF_PACKET; 1529 } 1530 1531 err = security_socket_create(family, type, protocol, kern); 1532 if (err) 1533 return err; 1534 1535 /* 1536 * Allocate the socket and allow the family to set things up. if 1537 * the protocol is 0, the family is instructed to select an appropriate 1538 * default. 1539 */ 1540 sock = sock_alloc(); 1541 if (!sock) { 1542 net_warn_ratelimited("socket: no more sockets\n"); 1543 return -ENFILE; /* Not exactly a match, but its the 1544 closest posix thing */ 1545 } 1546 1547 sock->type = type; 1548 1549 #ifdef CONFIG_MODULES 1550 /* Attempt to load a protocol module if the find failed. 1551 * 1552 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user 1553 * requested real, full-featured networking support upon configuration. 1554 * Otherwise module support will break! 1555 */ 1556 if (rcu_access_pointer(net_families[family]) == NULL) 1557 request_module("net-pf-%d", family); 1558 #endif 1559 1560 rcu_read_lock(); 1561 pf = rcu_dereference(net_families[family]); 1562 err = -EAFNOSUPPORT; 1563 if (!pf) 1564 goto out_release; 1565 1566 /* 1567 * We will call the ->create function, that possibly is in a loadable 1568 * module, so we have to bump that loadable module refcnt first. 1569 */ 1570 if (!try_module_get(pf->owner)) 1571 goto out_release; 1572 1573 /* Now protected by module ref count */ 1574 rcu_read_unlock(); 1575 1576 err = pf->create(net, sock, protocol, kern); 1577 if (err < 0) { 1578 /* ->create should release the allocated sock->sk object on error 1579 * and make sure sock->sk is set to NULL to avoid use-after-free 1580 */ 1581 DEBUG_NET_WARN_ONCE(sock->sk, 1582 "%ps must clear sock->sk on failure, family: %d, type: %d, protocol: %d\n", 1583 pf->create, family, type, protocol); 1584 goto out_module_put; 1585 } 1586 1587 /* 1588 * Now to bump the refcnt of the [loadable] module that owns this 1589 * socket at sock_release time we decrement its refcnt. 1590 */ 1591 if (!try_module_get(sock->ops->owner)) 1592 goto out_module_busy; 1593 1594 /* 1595 * Now that we're done with the ->create function, the [loadable] 1596 * module can have its refcnt decremented 1597 */ 1598 module_put(pf->owner); 1599 err = security_socket_post_create(sock, family, type, protocol, kern); 1600 if (err) 1601 goto out_sock_release; 1602 *res = sock; 1603 1604 return 0; 1605 1606 out_module_busy: 1607 err = -EAFNOSUPPORT; 1608 out_module_put: 1609 sock->ops = NULL; 1610 module_put(pf->owner); 1611 out_sock_release: 1612 sock_release(sock); 1613 return err; 1614 1615 out_release: 1616 rcu_read_unlock(); 1617 goto out_sock_release; 1618 } 1619 EXPORT_SYMBOL(__sock_create); 1620 1621 /** 1622 * sock_create - creates a socket 1623 * @family: protocol family (AF_INET, ...) 1624 * @type: communication type (SOCK_STREAM, ...) 1625 * @protocol: protocol (0, ...) 1626 * @res: new socket 1627 * 1628 * A wrapper around __sock_create(). 1629 * Returns 0 or an error. This function internally uses GFP_KERNEL. 1630 */ 1631 1632 int sock_create(int family, int type, int protocol, struct socket **res) 1633 { 1634 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0); 1635 } 1636 EXPORT_SYMBOL(sock_create); 1637 1638 /** 1639 * sock_create_kern - creates a socket (kernel space) 1640 * @net: net namespace 1641 * @family: protocol family (AF_INET, ...) 1642 * @type: communication type (SOCK_STREAM, ...) 1643 * @protocol: protocol (0, ...) 1644 * @res: new socket 1645 * 1646 * A wrapper around __sock_create(). 1647 * Returns 0 or an error. This function internally uses GFP_KERNEL. 1648 */ 1649 1650 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res) 1651 { 1652 return __sock_create(net, family, type, protocol, res, 1); 1653 } 1654 EXPORT_SYMBOL(sock_create_kern); 1655 1656 static struct socket *__sys_socket_create(int family, int type, int protocol) 1657 { 1658 struct socket *sock; 1659 int retval; 1660 1661 /* Check the SOCK_* constants for consistency. */ 1662 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC); 1663 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK); 1664 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK); 1665 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK); 1666 1667 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) 1668 return ERR_PTR(-EINVAL); 1669 type &= SOCK_TYPE_MASK; 1670 1671 retval = sock_create(family, type, protocol, &sock); 1672 if (retval < 0) 1673 return ERR_PTR(retval); 1674 1675 return sock; 1676 } 1677 1678 struct file *__sys_socket_file(int family, int type, int protocol) 1679 { 1680 struct socket *sock; 1681 int flags; 1682 1683 sock = __sys_socket_create(family, type, protocol); 1684 if (IS_ERR(sock)) 1685 return ERR_CAST(sock); 1686 1687 flags = type & ~SOCK_TYPE_MASK; 1688 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK)) 1689 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK; 1690 1691 return sock_alloc_file(sock, flags, NULL); 1692 } 1693 1694 /* A hook for bpf progs to attach to and update socket protocol. 1695 * 1696 * A static noinline declaration here could cause the compiler to 1697 * optimize away the function. A global noinline declaration will 1698 * keep the definition, but may optimize away the callsite. 1699 * Therefore, __weak is needed to ensure that the call is still 1700 * emitted, by telling the compiler that we don't know what the 1701 * function might eventually be. 1702 */ 1703 1704 __bpf_hook_start(); 1705 1706 __weak noinline int update_socket_protocol(int family, int type, int protocol) 1707 { 1708 return protocol; 1709 } 1710 1711 __bpf_hook_end(); 1712 1713 int __sys_socket(int family, int type, int protocol) 1714 { 1715 struct socket *sock; 1716 int flags; 1717 1718 sock = __sys_socket_create(family, type, 1719 update_socket_protocol(family, type, protocol)); 1720 if (IS_ERR(sock)) 1721 return PTR_ERR(sock); 1722 1723 flags = type & ~SOCK_TYPE_MASK; 1724 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK)) 1725 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK; 1726 1727 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK)); 1728 } 1729 1730 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol) 1731 { 1732 return __sys_socket(family, type, protocol); 1733 } 1734 1735 /* 1736 * Create a pair of connected sockets. 1737 */ 1738 1739 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec) 1740 { 1741 struct socket *sock1, *sock2; 1742 int fd1, fd2, err; 1743 struct file *newfile1, *newfile2; 1744 int flags; 1745 1746 flags = type & ~SOCK_TYPE_MASK; 1747 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) 1748 return -EINVAL; 1749 type &= SOCK_TYPE_MASK; 1750 1751 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK)) 1752 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK; 1753 1754 /* 1755 * reserve descriptors and make sure we won't fail 1756 * to return them to userland. 1757 */ 1758 fd1 = get_unused_fd_flags(flags); 1759 if (unlikely(fd1 < 0)) 1760 return fd1; 1761 1762 fd2 = get_unused_fd_flags(flags); 1763 if (unlikely(fd2 < 0)) { 1764 put_unused_fd(fd1); 1765 return fd2; 1766 } 1767 1768 err = put_user(fd1, &usockvec[0]); 1769 if (err) 1770 goto out; 1771 1772 err = put_user(fd2, &usockvec[1]); 1773 if (err) 1774 goto out; 1775 1776 /* 1777 * Obtain the first socket and check if the underlying protocol 1778 * supports the socketpair call. 1779 */ 1780 1781 err = sock_create(family, type, protocol, &sock1); 1782 if (unlikely(err < 0)) 1783 goto out; 1784 1785 err = sock_create(family, type, protocol, &sock2); 1786 if (unlikely(err < 0)) { 1787 sock_release(sock1); 1788 goto out; 1789 } 1790 1791 err = security_socket_socketpair(sock1, sock2); 1792 if (unlikely(err)) { 1793 sock_release(sock2); 1794 sock_release(sock1); 1795 goto out; 1796 } 1797 1798 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2); 1799 if (unlikely(err < 0)) { 1800 sock_release(sock2); 1801 sock_release(sock1); 1802 goto out; 1803 } 1804 1805 newfile1 = sock_alloc_file(sock1, flags, NULL); 1806 if (IS_ERR(newfile1)) { 1807 err = PTR_ERR(newfile1); 1808 sock_release(sock2); 1809 goto out; 1810 } 1811 1812 newfile2 = sock_alloc_file(sock2, flags, NULL); 1813 if (IS_ERR(newfile2)) { 1814 err = PTR_ERR(newfile2); 1815 fput(newfile1); 1816 goto out; 1817 } 1818 1819 audit_fd_pair(fd1, fd2); 1820 1821 fd_install(fd1, newfile1); 1822 fd_install(fd2, newfile2); 1823 return 0; 1824 1825 out: 1826 put_unused_fd(fd2); 1827 put_unused_fd(fd1); 1828 return err; 1829 } 1830 1831 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol, 1832 int __user *, usockvec) 1833 { 1834 return __sys_socketpair(family, type, protocol, usockvec); 1835 } 1836 1837 int __sys_bind_socket(struct socket *sock, struct sockaddr_storage *address, 1838 int addrlen) 1839 { 1840 int err; 1841 1842 err = security_socket_bind(sock, (struct sockaddr *)address, 1843 addrlen); 1844 if (!err) 1845 err = READ_ONCE(sock->ops)->bind(sock, 1846 (struct sockaddr *)address, 1847 addrlen); 1848 return err; 1849 } 1850 1851 /* 1852 * Bind a name to a socket. Nothing much to do here since it's 1853 * the protocol's responsibility to handle the local address. 1854 * 1855 * We move the socket address to kernel space before we call 1856 * the protocol layer (having also checked the address is ok). 1857 */ 1858 1859 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen) 1860 { 1861 struct socket *sock; 1862 struct sockaddr_storage address; 1863 int err, fput_needed; 1864 1865 sock = sockfd_lookup_light(fd, &err, &fput_needed); 1866 if (sock) { 1867 err = move_addr_to_kernel(umyaddr, addrlen, &address); 1868 if (!err) 1869 err = __sys_bind_socket(sock, &address, addrlen); 1870 fput_light(sock->file, fput_needed); 1871 } 1872 return err; 1873 } 1874 1875 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen) 1876 { 1877 return __sys_bind(fd, umyaddr, addrlen); 1878 } 1879 1880 /* 1881 * Perform a listen. Basically, we allow the protocol to do anything 1882 * necessary for a listen, and if that works, we mark the socket as 1883 * ready for listening. 1884 */ 1885 int __sys_listen_socket(struct socket *sock, int backlog) 1886 { 1887 int somaxconn, err; 1888 1889 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn); 1890 if ((unsigned int)backlog > somaxconn) 1891 backlog = somaxconn; 1892 1893 err = security_socket_listen(sock, backlog); 1894 if (!err) 1895 err = READ_ONCE(sock->ops)->listen(sock, backlog); 1896 return err; 1897 } 1898 1899 int __sys_listen(int fd, int backlog) 1900 { 1901 struct socket *sock; 1902 int err, fput_needed; 1903 1904 sock = sockfd_lookup_light(fd, &err, &fput_needed); 1905 if (sock) { 1906 err = __sys_listen_socket(sock, backlog); 1907 fput_light(sock->file, fput_needed); 1908 } 1909 return err; 1910 } 1911 1912 SYSCALL_DEFINE2(listen, int, fd, int, backlog) 1913 { 1914 return __sys_listen(fd, backlog); 1915 } 1916 1917 struct file *do_accept(struct file *file, struct proto_accept_arg *arg, 1918 struct sockaddr __user *upeer_sockaddr, 1919 int __user *upeer_addrlen, int flags) 1920 { 1921 struct socket *sock, *newsock; 1922 struct file *newfile; 1923 int err, len; 1924 struct sockaddr_storage address; 1925 const struct proto_ops *ops; 1926 1927 sock = sock_from_file(file); 1928 if (!sock) 1929 return ERR_PTR(-ENOTSOCK); 1930 1931 newsock = sock_alloc(); 1932 if (!newsock) 1933 return ERR_PTR(-ENFILE); 1934 ops = READ_ONCE(sock->ops); 1935 1936 newsock->type = sock->type; 1937 newsock->ops = ops; 1938 1939 /* 1940 * We don't need try_module_get here, as the listening socket (sock) 1941 * has the protocol module (sock->ops->owner) held. 1942 */ 1943 __module_get(ops->owner); 1944 1945 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name); 1946 if (IS_ERR(newfile)) 1947 return newfile; 1948 1949 err = security_socket_accept(sock, newsock); 1950 if (err) 1951 goto out_fd; 1952 1953 arg->flags |= sock->file->f_flags; 1954 err = ops->accept(sock, newsock, arg); 1955 if (err < 0) 1956 goto out_fd; 1957 1958 if (upeer_sockaddr) { 1959 len = ops->getname(newsock, (struct sockaddr *)&address, 2); 1960 if (len < 0) { 1961 err = -ECONNABORTED; 1962 goto out_fd; 1963 } 1964 err = move_addr_to_user(&address, 1965 len, upeer_sockaddr, upeer_addrlen); 1966 if (err < 0) 1967 goto out_fd; 1968 } 1969 1970 /* File flags are not inherited via accept() unlike another OSes. */ 1971 return newfile; 1972 out_fd: 1973 fput(newfile); 1974 return ERR_PTR(err); 1975 } 1976 1977 static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr, 1978 int __user *upeer_addrlen, int flags) 1979 { 1980 struct proto_accept_arg arg = { }; 1981 struct file *newfile; 1982 int newfd; 1983 1984 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) 1985 return -EINVAL; 1986 1987 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK)) 1988 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK; 1989 1990 newfd = get_unused_fd_flags(flags); 1991 if (unlikely(newfd < 0)) 1992 return newfd; 1993 1994 newfile = do_accept(file, &arg, upeer_sockaddr, upeer_addrlen, 1995 flags); 1996 if (IS_ERR(newfile)) { 1997 put_unused_fd(newfd); 1998 return PTR_ERR(newfile); 1999 } 2000 fd_install(newfd, newfile); 2001 return newfd; 2002 } 2003 2004 /* 2005 * For accept, we attempt to create a new socket, set up the link 2006 * with the client, wake up the client, then return the new 2007 * connected fd. We collect the address of the connector in kernel 2008 * space and move it to user at the very end. This is unclean because 2009 * we open the socket then return an error. 2010 * 2011 * 1003.1g adds the ability to recvmsg() to query connection pending 2012 * status to recvmsg. We need to add that support in a way thats 2013 * clean when we restructure accept also. 2014 */ 2015 2016 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr, 2017 int __user *upeer_addrlen, int flags) 2018 { 2019 int ret = -EBADF; 2020 struct fd f; 2021 2022 f = fdget(fd); 2023 if (fd_file(f)) { 2024 ret = __sys_accept4_file(fd_file(f), upeer_sockaddr, 2025 upeer_addrlen, flags); 2026 fdput(f); 2027 } 2028 2029 return ret; 2030 } 2031 2032 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr, 2033 int __user *, upeer_addrlen, int, flags) 2034 { 2035 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags); 2036 } 2037 2038 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr, 2039 int __user *, upeer_addrlen) 2040 { 2041 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0); 2042 } 2043 2044 /* 2045 * Attempt to connect to a socket with the server address. The address 2046 * is in user space so we verify it is OK and move it to kernel space. 2047 * 2048 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to 2049 * break bindings 2050 * 2051 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and 2052 * other SEQPACKET protocols that take time to connect() as it doesn't 2053 * include the -EINPROGRESS status for such sockets. 2054 */ 2055 2056 int __sys_connect_file(struct file *file, struct sockaddr_storage *address, 2057 int addrlen, int file_flags) 2058 { 2059 struct socket *sock; 2060 int err; 2061 2062 sock = sock_from_file(file); 2063 if (!sock) { 2064 err = -ENOTSOCK; 2065 goto out; 2066 } 2067 2068 err = 2069 security_socket_connect(sock, (struct sockaddr *)address, addrlen); 2070 if (err) 2071 goto out; 2072 2073 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address, 2074 addrlen, sock->file->f_flags | file_flags); 2075 out: 2076 return err; 2077 } 2078 2079 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen) 2080 { 2081 int ret = -EBADF; 2082 struct fd f; 2083 2084 f = fdget(fd); 2085 if (fd_file(f)) { 2086 struct sockaddr_storage address; 2087 2088 ret = move_addr_to_kernel(uservaddr, addrlen, &address); 2089 if (!ret) 2090 ret = __sys_connect_file(fd_file(f), &address, addrlen, 0); 2091 fdput(f); 2092 } 2093 2094 return ret; 2095 } 2096 2097 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr, 2098 int, addrlen) 2099 { 2100 return __sys_connect(fd, uservaddr, addrlen); 2101 } 2102 2103 /* 2104 * Get the local address ('name') of a socket object. Move the obtained 2105 * name to user space. 2106 */ 2107 2108 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr, 2109 int __user *usockaddr_len) 2110 { 2111 struct socket *sock; 2112 struct sockaddr_storage address; 2113 int err, fput_needed; 2114 2115 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2116 if (!sock) 2117 goto out; 2118 2119 err = security_socket_getsockname(sock); 2120 if (err) 2121 goto out_put; 2122 2123 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0); 2124 if (err < 0) 2125 goto out_put; 2126 /* "err" is actually length in this case */ 2127 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len); 2128 2129 out_put: 2130 fput_light(sock->file, fput_needed); 2131 out: 2132 return err; 2133 } 2134 2135 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr, 2136 int __user *, usockaddr_len) 2137 { 2138 return __sys_getsockname(fd, usockaddr, usockaddr_len); 2139 } 2140 2141 /* 2142 * Get the remote address ('name') of a socket object. Move the obtained 2143 * name to user space. 2144 */ 2145 2146 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr, 2147 int __user *usockaddr_len) 2148 { 2149 struct socket *sock; 2150 struct sockaddr_storage address; 2151 int err, fput_needed; 2152 2153 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2154 if (sock != NULL) { 2155 const struct proto_ops *ops = READ_ONCE(sock->ops); 2156 2157 err = security_socket_getpeername(sock); 2158 if (err) { 2159 fput_light(sock->file, fput_needed); 2160 return err; 2161 } 2162 2163 err = ops->getname(sock, (struct sockaddr *)&address, 1); 2164 if (err >= 0) 2165 /* "err" is actually length in this case */ 2166 err = move_addr_to_user(&address, err, usockaddr, 2167 usockaddr_len); 2168 fput_light(sock->file, fput_needed); 2169 } 2170 return err; 2171 } 2172 2173 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr, 2174 int __user *, usockaddr_len) 2175 { 2176 return __sys_getpeername(fd, usockaddr, usockaddr_len); 2177 } 2178 2179 /* 2180 * Send a datagram to a given address. We move the address into kernel 2181 * space and check the user space data area is readable before invoking 2182 * the protocol. 2183 */ 2184 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags, 2185 struct sockaddr __user *addr, int addr_len) 2186 { 2187 struct socket *sock; 2188 struct sockaddr_storage address; 2189 int err; 2190 struct msghdr msg; 2191 int fput_needed; 2192 2193 err = import_ubuf(ITER_SOURCE, buff, len, &msg.msg_iter); 2194 if (unlikely(err)) 2195 return err; 2196 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2197 if (!sock) 2198 goto out; 2199 2200 msg.msg_name = NULL; 2201 msg.msg_control = NULL; 2202 msg.msg_controllen = 0; 2203 msg.msg_namelen = 0; 2204 msg.msg_ubuf = NULL; 2205 if (addr) { 2206 err = move_addr_to_kernel(addr, addr_len, &address); 2207 if (err < 0) 2208 goto out_put; 2209 msg.msg_name = (struct sockaddr *)&address; 2210 msg.msg_namelen = addr_len; 2211 } 2212 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS; 2213 if (sock->file->f_flags & O_NONBLOCK) 2214 flags |= MSG_DONTWAIT; 2215 msg.msg_flags = flags; 2216 err = __sock_sendmsg(sock, &msg); 2217 2218 out_put: 2219 fput_light(sock->file, fput_needed); 2220 out: 2221 return err; 2222 } 2223 2224 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len, 2225 unsigned int, flags, struct sockaddr __user *, addr, 2226 int, addr_len) 2227 { 2228 return __sys_sendto(fd, buff, len, flags, addr, addr_len); 2229 } 2230 2231 /* 2232 * Send a datagram down a socket. 2233 */ 2234 2235 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len, 2236 unsigned int, flags) 2237 { 2238 return __sys_sendto(fd, buff, len, flags, NULL, 0); 2239 } 2240 2241 /* 2242 * Receive a frame from the socket and optionally record the address of the 2243 * sender. We verify the buffers are writable and if needed move the 2244 * sender address from kernel to user space. 2245 */ 2246 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags, 2247 struct sockaddr __user *addr, int __user *addr_len) 2248 { 2249 struct sockaddr_storage address; 2250 struct msghdr msg = { 2251 /* Save some cycles and don't copy the address if not needed */ 2252 .msg_name = addr ? (struct sockaddr *)&address : NULL, 2253 }; 2254 struct socket *sock; 2255 int err, err2; 2256 int fput_needed; 2257 2258 err = import_ubuf(ITER_DEST, ubuf, size, &msg.msg_iter); 2259 if (unlikely(err)) 2260 return err; 2261 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2262 if (!sock) 2263 goto out; 2264 2265 if (sock->file->f_flags & O_NONBLOCK) 2266 flags |= MSG_DONTWAIT; 2267 err = sock_recvmsg(sock, &msg, flags); 2268 2269 if (err >= 0 && addr != NULL) { 2270 err2 = move_addr_to_user(&address, 2271 msg.msg_namelen, addr, addr_len); 2272 if (err2 < 0) 2273 err = err2; 2274 } 2275 2276 fput_light(sock->file, fput_needed); 2277 out: 2278 return err; 2279 } 2280 2281 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size, 2282 unsigned int, flags, struct sockaddr __user *, addr, 2283 int __user *, addr_len) 2284 { 2285 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len); 2286 } 2287 2288 /* 2289 * Receive a datagram from a socket. 2290 */ 2291 2292 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size, 2293 unsigned int, flags) 2294 { 2295 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL); 2296 } 2297 2298 static bool sock_use_custom_sol_socket(const struct socket *sock) 2299 { 2300 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags); 2301 } 2302 2303 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 2304 int optname, sockptr_t optval, int optlen) 2305 { 2306 const struct proto_ops *ops; 2307 char *kernel_optval = NULL; 2308 int err; 2309 2310 if (optlen < 0) 2311 return -EINVAL; 2312 2313 err = security_socket_setsockopt(sock, level, optname); 2314 if (err) 2315 goto out_put; 2316 2317 if (!compat) 2318 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname, 2319 optval, &optlen, 2320 &kernel_optval); 2321 if (err < 0) 2322 goto out_put; 2323 if (err > 0) { 2324 err = 0; 2325 goto out_put; 2326 } 2327 2328 if (kernel_optval) 2329 optval = KERNEL_SOCKPTR(kernel_optval); 2330 ops = READ_ONCE(sock->ops); 2331 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock)) 2332 err = sock_setsockopt(sock, level, optname, optval, optlen); 2333 else if (unlikely(!ops->setsockopt)) 2334 err = -EOPNOTSUPP; 2335 else 2336 err = ops->setsockopt(sock, level, optname, optval, 2337 optlen); 2338 kfree(kernel_optval); 2339 out_put: 2340 return err; 2341 } 2342 EXPORT_SYMBOL(do_sock_setsockopt); 2343 2344 /* Set a socket option. Because we don't know the option lengths we have 2345 * to pass the user mode parameter for the protocols to sort out. 2346 */ 2347 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval, 2348 int optlen) 2349 { 2350 sockptr_t optval = USER_SOCKPTR(user_optval); 2351 bool compat = in_compat_syscall(); 2352 int err, fput_needed; 2353 struct socket *sock; 2354 2355 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2356 if (!sock) 2357 return err; 2358 2359 err = do_sock_setsockopt(sock, compat, level, optname, optval, optlen); 2360 2361 fput_light(sock->file, fput_needed); 2362 return err; 2363 } 2364 2365 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname, 2366 char __user *, optval, int, optlen) 2367 { 2368 return __sys_setsockopt(fd, level, optname, optval, optlen); 2369 } 2370 2371 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level, 2372 int optname)); 2373 2374 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 2375 int optname, sockptr_t optval, sockptr_t optlen) 2376 { 2377 int max_optlen __maybe_unused = 0; 2378 const struct proto_ops *ops; 2379 int err; 2380 2381 err = security_socket_getsockopt(sock, level, optname); 2382 if (err) 2383 return err; 2384 2385 if (!compat) 2386 copy_from_sockptr(&max_optlen, optlen, sizeof(int)); 2387 2388 ops = READ_ONCE(sock->ops); 2389 if (level == SOL_SOCKET) { 2390 err = sk_getsockopt(sock->sk, level, optname, optval, optlen); 2391 } else if (unlikely(!ops->getsockopt)) { 2392 err = -EOPNOTSUPP; 2393 } else { 2394 if (WARN_ONCE(optval.is_kernel || optlen.is_kernel, 2395 "Invalid argument type")) 2396 return -EOPNOTSUPP; 2397 2398 err = ops->getsockopt(sock, level, optname, optval.user, 2399 optlen.user); 2400 } 2401 2402 if (!compat) 2403 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname, 2404 optval, optlen, max_optlen, 2405 err); 2406 2407 return err; 2408 } 2409 EXPORT_SYMBOL(do_sock_getsockopt); 2410 2411 /* 2412 * Get a socket option. Because we don't know the option lengths we have 2413 * to pass a user mode parameter for the protocols to sort out. 2414 */ 2415 int __sys_getsockopt(int fd, int level, int optname, char __user *optval, 2416 int __user *optlen) 2417 { 2418 int err, fput_needed; 2419 struct socket *sock; 2420 bool compat; 2421 2422 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2423 if (!sock) 2424 return err; 2425 2426 compat = in_compat_syscall(); 2427 err = do_sock_getsockopt(sock, compat, level, optname, 2428 USER_SOCKPTR(optval), USER_SOCKPTR(optlen)); 2429 2430 fput_light(sock->file, fput_needed); 2431 return err; 2432 } 2433 2434 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname, 2435 char __user *, optval, int __user *, optlen) 2436 { 2437 return __sys_getsockopt(fd, level, optname, optval, optlen); 2438 } 2439 2440 /* 2441 * Shutdown a socket. 2442 */ 2443 2444 int __sys_shutdown_sock(struct socket *sock, int how) 2445 { 2446 int err; 2447 2448 err = security_socket_shutdown(sock, how); 2449 if (!err) 2450 err = READ_ONCE(sock->ops)->shutdown(sock, how); 2451 2452 return err; 2453 } 2454 2455 int __sys_shutdown(int fd, int how) 2456 { 2457 int err, fput_needed; 2458 struct socket *sock; 2459 2460 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2461 if (sock != NULL) { 2462 err = __sys_shutdown_sock(sock, how); 2463 fput_light(sock->file, fput_needed); 2464 } 2465 return err; 2466 } 2467 2468 SYSCALL_DEFINE2(shutdown, int, fd, int, how) 2469 { 2470 return __sys_shutdown(fd, how); 2471 } 2472 2473 /* A couple of helpful macros for getting the address of the 32/64 bit 2474 * fields which are the same type (int / unsigned) on our platforms. 2475 */ 2476 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member) 2477 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen) 2478 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags) 2479 2480 struct used_address { 2481 struct sockaddr_storage name; 2482 unsigned int name_len; 2483 }; 2484 2485 int __copy_msghdr(struct msghdr *kmsg, 2486 struct user_msghdr *msg, 2487 struct sockaddr __user **save_addr) 2488 { 2489 ssize_t err; 2490 2491 kmsg->msg_control_is_user = true; 2492 kmsg->msg_get_inq = 0; 2493 kmsg->msg_control_user = msg->msg_control; 2494 kmsg->msg_controllen = msg->msg_controllen; 2495 kmsg->msg_flags = msg->msg_flags; 2496 2497 kmsg->msg_namelen = msg->msg_namelen; 2498 if (!msg->msg_name) 2499 kmsg->msg_namelen = 0; 2500 2501 if (kmsg->msg_namelen < 0) 2502 return -EINVAL; 2503 2504 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage)) 2505 kmsg->msg_namelen = sizeof(struct sockaddr_storage); 2506 2507 if (save_addr) 2508 *save_addr = msg->msg_name; 2509 2510 if (msg->msg_name && kmsg->msg_namelen) { 2511 if (!save_addr) { 2512 err = move_addr_to_kernel(msg->msg_name, 2513 kmsg->msg_namelen, 2514 kmsg->msg_name); 2515 if (err < 0) 2516 return err; 2517 } 2518 } else { 2519 kmsg->msg_name = NULL; 2520 kmsg->msg_namelen = 0; 2521 } 2522 2523 if (msg->msg_iovlen > UIO_MAXIOV) 2524 return -EMSGSIZE; 2525 2526 kmsg->msg_iocb = NULL; 2527 kmsg->msg_ubuf = NULL; 2528 return 0; 2529 } 2530 2531 static int copy_msghdr_from_user(struct msghdr *kmsg, 2532 struct user_msghdr __user *umsg, 2533 struct sockaddr __user **save_addr, 2534 struct iovec **iov) 2535 { 2536 struct user_msghdr msg; 2537 ssize_t err; 2538 2539 if (copy_from_user(&msg, umsg, sizeof(*umsg))) 2540 return -EFAULT; 2541 2542 err = __copy_msghdr(kmsg, &msg, save_addr); 2543 if (err) 2544 return err; 2545 2546 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE, 2547 msg.msg_iov, msg.msg_iovlen, 2548 UIO_FASTIOV, iov, &kmsg->msg_iter); 2549 return err < 0 ? err : 0; 2550 } 2551 2552 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys, 2553 unsigned int flags, struct used_address *used_address, 2554 unsigned int allowed_msghdr_flags) 2555 { 2556 unsigned char ctl[sizeof(struct cmsghdr) + 20] 2557 __aligned(sizeof(__kernel_size_t)); 2558 /* 20 is size of ipv6_pktinfo */ 2559 unsigned char *ctl_buf = ctl; 2560 int ctl_len; 2561 ssize_t err; 2562 2563 err = -ENOBUFS; 2564 2565 if (msg_sys->msg_controllen > INT_MAX) 2566 goto out; 2567 flags |= (msg_sys->msg_flags & allowed_msghdr_flags); 2568 ctl_len = msg_sys->msg_controllen; 2569 if ((MSG_CMSG_COMPAT & flags) && ctl_len) { 2570 err = 2571 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl, 2572 sizeof(ctl)); 2573 if (err) 2574 goto out; 2575 ctl_buf = msg_sys->msg_control; 2576 ctl_len = msg_sys->msg_controllen; 2577 } else if (ctl_len) { 2578 BUILD_BUG_ON(sizeof(struct cmsghdr) != 2579 CMSG_ALIGN(sizeof(struct cmsghdr))); 2580 if (ctl_len > sizeof(ctl)) { 2581 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL); 2582 if (ctl_buf == NULL) 2583 goto out; 2584 } 2585 err = -EFAULT; 2586 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len)) 2587 goto out_freectl; 2588 msg_sys->msg_control = ctl_buf; 2589 msg_sys->msg_control_is_user = false; 2590 } 2591 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS; 2592 msg_sys->msg_flags = flags; 2593 2594 if (sock->file->f_flags & O_NONBLOCK) 2595 msg_sys->msg_flags |= MSG_DONTWAIT; 2596 /* 2597 * If this is sendmmsg() and current destination address is same as 2598 * previously succeeded address, omit asking LSM's decision. 2599 * used_address->name_len is initialized to UINT_MAX so that the first 2600 * destination address never matches. 2601 */ 2602 if (used_address && msg_sys->msg_name && 2603 used_address->name_len == msg_sys->msg_namelen && 2604 !memcmp(&used_address->name, msg_sys->msg_name, 2605 used_address->name_len)) { 2606 err = sock_sendmsg_nosec(sock, msg_sys); 2607 goto out_freectl; 2608 } 2609 err = __sock_sendmsg(sock, msg_sys); 2610 /* 2611 * If this is sendmmsg() and sending to current destination address was 2612 * successful, remember it. 2613 */ 2614 if (used_address && err >= 0) { 2615 used_address->name_len = msg_sys->msg_namelen; 2616 if (msg_sys->msg_name) 2617 memcpy(&used_address->name, msg_sys->msg_name, 2618 used_address->name_len); 2619 } 2620 2621 out_freectl: 2622 if (ctl_buf != ctl) 2623 sock_kfree_s(sock->sk, ctl_buf, ctl_len); 2624 out: 2625 return err; 2626 } 2627 2628 static int sendmsg_copy_msghdr(struct msghdr *msg, 2629 struct user_msghdr __user *umsg, unsigned flags, 2630 struct iovec **iov) 2631 { 2632 int err; 2633 2634 if (flags & MSG_CMSG_COMPAT) { 2635 struct compat_msghdr __user *msg_compat; 2636 2637 msg_compat = (struct compat_msghdr __user *) umsg; 2638 err = get_compat_msghdr(msg, msg_compat, NULL, iov); 2639 } else { 2640 err = copy_msghdr_from_user(msg, umsg, NULL, iov); 2641 } 2642 if (err < 0) 2643 return err; 2644 2645 return 0; 2646 } 2647 2648 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg, 2649 struct msghdr *msg_sys, unsigned int flags, 2650 struct used_address *used_address, 2651 unsigned int allowed_msghdr_flags) 2652 { 2653 struct sockaddr_storage address; 2654 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack; 2655 ssize_t err; 2656 2657 msg_sys->msg_name = &address; 2658 2659 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov); 2660 if (err < 0) 2661 return err; 2662 2663 err = ____sys_sendmsg(sock, msg_sys, flags, used_address, 2664 allowed_msghdr_flags); 2665 kfree(iov); 2666 return err; 2667 } 2668 2669 /* 2670 * BSD sendmsg interface 2671 */ 2672 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg, 2673 unsigned int flags) 2674 { 2675 return ____sys_sendmsg(sock, msg, flags, NULL, 0); 2676 } 2677 2678 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags, 2679 bool forbid_cmsg_compat) 2680 { 2681 int fput_needed, err; 2682 struct msghdr msg_sys; 2683 struct socket *sock; 2684 2685 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT)) 2686 return -EINVAL; 2687 2688 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2689 if (!sock) 2690 goto out; 2691 2692 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0); 2693 2694 fput_light(sock->file, fput_needed); 2695 out: 2696 return err; 2697 } 2698 2699 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags) 2700 { 2701 return __sys_sendmsg(fd, msg, flags, true); 2702 } 2703 2704 /* 2705 * Linux sendmmsg interface 2706 */ 2707 2708 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen, 2709 unsigned int flags, bool forbid_cmsg_compat) 2710 { 2711 int fput_needed, err, datagrams; 2712 struct socket *sock; 2713 struct mmsghdr __user *entry; 2714 struct compat_mmsghdr __user *compat_entry; 2715 struct msghdr msg_sys; 2716 struct used_address used_address; 2717 unsigned int oflags = flags; 2718 2719 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT)) 2720 return -EINVAL; 2721 2722 if (vlen > UIO_MAXIOV) 2723 vlen = UIO_MAXIOV; 2724 2725 datagrams = 0; 2726 2727 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2728 if (!sock) 2729 return err; 2730 2731 used_address.name_len = UINT_MAX; 2732 entry = mmsg; 2733 compat_entry = (struct compat_mmsghdr __user *)mmsg; 2734 err = 0; 2735 flags |= MSG_BATCH; 2736 2737 while (datagrams < vlen) { 2738 if (datagrams == vlen - 1) 2739 flags = oflags; 2740 2741 if (MSG_CMSG_COMPAT & flags) { 2742 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry, 2743 &msg_sys, flags, &used_address, MSG_EOR); 2744 if (err < 0) 2745 break; 2746 err = __put_user(err, &compat_entry->msg_len); 2747 ++compat_entry; 2748 } else { 2749 err = ___sys_sendmsg(sock, 2750 (struct user_msghdr __user *)entry, 2751 &msg_sys, flags, &used_address, MSG_EOR); 2752 if (err < 0) 2753 break; 2754 err = put_user(err, &entry->msg_len); 2755 ++entry; 2756 } 2757 2758 if (err) 2759 break; 2760 ++datagrams; 2761 if (msg_data_left(&msg_sys)) 2762 break; 2763 cond_resched(); 2764 } 2765 2766 fput_light(sock->file, fput_needed); 2767 2768 /* We only return an error if no datagrams were able to be sent */ 2769 if (datagrams != 0) 2770 return datagrams; 2771 2772 return err; 2773 } 2774 2775 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg, 2776 unsigned int, vlen, unsigned int, flags) 2777 { 2778 return __sys_sendmmsg(fd, mmsg, vlen, flags, true); 2779 } 2780 2781 static int recvmsg_copy_msghdr(struct msghdr *msg, 2782 struct user_msghdr __user *umsg, unsigned flags, 2783 struct sockaddr __user **uaddr, 2784 struct iovec **iov) 2785 { 2786 ssize_t err; 2787 2788 if (MSG_CMSG_COMPAT & flags) { 2789 struct compat_msghdr __user *msg_compat; 2790 2791 msg_compat = (struct compat_msghdr __user *) umsg; 2792 err = get_compat_msghdr(msg, msg_compat, uaddr, iov); 2793 } else { 2794 err = copy_msghdr_from_user(msg, umsg, uaddr, iov); 2795 } 2796 if (err < 0) 2797 return err; 2798 2799 return 0; 2800 } 2801 2802 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys, 2803 struct user_msghdr __user *msg, 2804 struct sockaddr __user *uaddr, 2805 unsigned int flags, int nosec) 2806 { 2807 struct compat_msghdr __user *msg_compat = 2808 (struct compat_msghdr __user *) msg; 2809 int __user *uaddr_len = COMPAT_NAMELEN(msg); 2810 struct sockaddr_storage addr; 2811 unsigned long cmsg_ptr; 2812 int len; 2813 ssize_t err; 2814 2815 msg_sys->msg_name = &addr; 2816 cmsg_ptr = (unsigned long)msg_sys->msg_control; 2817 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT); 2818 2819 /* We assume all kernel code knows the size of sockaddr_storage */ 2820 msg_sys->msg_namelen = 0; 2821 2822 if (sock->file->f_flags & O_NONBLOCK) 2823 flags |= MSG_DONTWAIT; 2824 2825 if (unlikely(nosec)) 2826 err = sock_recvmsg_nosec(sock, msg_sys, flags); 2827 else 2828 err = sock_recvmsg(sock, msg_sys, flags); 2829 2830 if (err < 0) 2831 goto out; 2832 len = err; 2833 2834 if (uaddr != NULL) { 2835 err = move_addr_to_user(&addr, 2836 msg_sys->msg_namelen, uaddr, 2837 uaddr_len); 2838 if (err < 0) 2839 goto out; 2840 } 2841 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT), 2842 COMPAT_FLAGS(msg)); 2843 if (err) 2844 goto out; 2845 if (MSG_CMSG_COMPAT & flags) 2846 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr, 2847 &msg_compat->msg_controllen); 2848 else 2849 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr, 2850 &msg->msg_controllen); 2851 if (err) 2852 goto out; 2853 err = len; 2854 out: 2855 return err; 2856 } 2857 2858 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg, 2859 struct msghdr *msg_sys, unsigned int flags, int nosec) 2860 { 2861 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack; 2862 /* user mode address pointers */ 2863 struct sockaddr __user *uaddr; 2864 ssize_t err; 2865 2866 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov); 2867 if (err < 0) 2868 return err; 2869 2870 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec); 2871 kfree(iov); 2872 return err; 2873 } 2874 2875 /* 2876 * BSD recvmsg interface 2877 */ 2878 2879 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg, 2880 struct user_msghdr __user *umsg, 2881 struct sockaddr __user *uaddr, unsigned int flags) 2882 { 2883 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0); 2884 } 2885 2886 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags, 2887 bool forbid_cmsg_compat) 2888 { 2889 int fput_needed, err; 2890 struct msghdr msg_sys; 2891 struct socket *sock; 2892 2893 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT)) 2894 return -EINVAL; 2895 2896 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2897 if (!sock) 2898 goto out; 2899 2900 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0); 2901 2902 fput_light(sock->file, fput_needed); 2903 out: 2904 return err; 2905 } 2906 2907 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg, 2908 unsigned int, flags) 2909 { 2910 return __sys_recvmsg(fd, msg, flags, true); 2911 } 2912 2913 /* 2914 * Linux recvmmsg interface 2915 */ 2916 2917 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg, 2918 unsigned int vlen, unsigned int flags, 2919 struct timespec64 *timeout) 2920 { 2921 int fput_needed, err, datagrams; 2922 struct socket *sock; 2923 struct mmsghdr __user *entry; 2924 struct compat_mmsghdr __user *compat_entry; 2925 struct msghdr msg_sys; 2926 struct timespec64 end_time; 2927 struct timespec64 timeout64; 2928 2929 if (timeout && 2930 poll_select_set_timeout(&end_time, timeout->tv_sec, 2931 timeout->tv_nsec)) 2932 return -EINVAL; 2933 2934 datagrams = 0; 2935 2936 sock = sockfd_lookup_light(fd, &err, &fput_needed); 2937 if (!sock) 2938 return err; 2939 2940 if (likely(!(flags & MSG_ERRQUEUE))) { 2941 err = sock_error(sock->sk); 2942 if (err) { 2943 datagrams = err; 2944 goto out_put; 2945 } 2946 } 2947 2948 entry = mmsg; 2949 compat_entry = (struct compat_mmsghdr __user *)mmsg; 2950 2951 while (datagrams < vlen) { 2952 /* 2953 * No need to ask LSM for more than the first datagram. 2954 */ 2955 if (MSG_CMSG_COMPAT & flags) { 2956 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry, 2957 &msg_sys, flags & ~MSG_WAITFORONE, 2958 datagrams); 2959 if (err < 0) 2960 break; 2961 err = __put_user(err, &compat_entry->msg_len); 2962 ++compat_entry; 2963 } else { 2964 err = ___sys_recvmsg(sock, 2965 (struct user_msghdr __user *)entry, 2966 &msg_sys, flags & ~MSG_WAITFORONE, 2967 datagrams); 2968 if (err < 0) 2969 break; 2970 err = put_user(err, &entry->msg_len); 2971 ++entry; 2972 } 2973 2974 if (err) 2975 break; 2976 ++datagrams; 2977 2978 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */ 2979 if (flags & MSG_WAITFORONE) 2980 flags |= MSG_DONTWAIT; 2981 2982 if (timeout) { 2983 ktime_get_ts64(&timeout64); 2984 *timeout = timespec64_sub(end_time, timeout64); 2985 if (timeout->tv_sec < 0) { 2986 timeout->tv_sec = timeout->tv_nsec = 0; 2987 break; 2988 } 2989 2990 /* Timeout, return less than vlen datagrams */ 2991 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0) 2992 break; 2993 } 2994 2995 /* Out of band data, return right away */ 2996 if (msg_sys.msg_flags & MSG_OOB) 2997 break; 2998 cond_resched(); 2999 } 3000 3001 if (err == 0) 3002 goto out_put; 3003 3004 if (datagrams == 0) { 3005 datagrams = err; 3006 goto out_put; 3007 } 3008 3009 /* 3010 * We may return less entries than requested (vlen) if the 3011 * sock is non block and there aren't enough datagrams... 3012 */ 3013 if (err != -EAGAIN) { 3014 /* 3015 * ... or if recvmsg returns an error after we 3016 * received some datagrams, where we record the 3017 * error to return on the next call or if the 3018 * app asks about it using getsockopt(SO_ERROR). 3019 */ 3020 WRITE_ONCE(sock->sk->sk_err, -err); 3021 } 3022 out_put: 3023 fput_light(sock->file, fput_needed); 3024 3025 return datagrams; 3026 } 3027 3028 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, 3029 unsigned int vlen, unsigned int flags, 3030 struct __kernel_timespec __user *timeout, 3031 struct old_timespec32 __user *timeout32) 3032 { 3033 int datagrams; 3034 struct timespec64 timeout_sys; 3035 3036 if (timeout && get_timespec64(&timeout_sys, timeout)) 3037 return -EFAULT; 3038 3039 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32)) 3040 return -EFAULT; 3041 3042 if (!timeout && !timeout32) 3043 return do_recvmmsg(fd, mmsg, vlen, flags, NULL); 3044 3045 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys); 3046 3047 if (datagrams <= 0) 3048 return datagrams; 3049 3050 if (timeout && put_timespec64(&timeout_sys, timeout)) 3051 datagrams = -EFAULT; 3052 3053 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32)) 3054 datagrams = -EFAULT; 3055 3056 return datagrams; 3057 } 3058 3059 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg, 3060 unsigned int, vlen, unsigned int, flags, 3061 struct __kernel_timespec __user *, timeout) 3062 { 3063 if (flags & MSG_CMSG_COMPAT) 3064 return -EINVAL; 3065 3066 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL); 3067 } 3068 3069 #ifdef CONFIG_COMPAT_32BIT_TIME 3070 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg, 3071 unsigned int, vlen, unsigned int, flags, 3072 struct old_timespec32 __user *, timeout) 3073 { 3074 if (flags & MSG_CMSG_COMPAT) 3075 return -EINVAL; 3076 3077 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout); 3078 } 3079 #endif 3080 3081 #ifdef __ARCH_WANT_SYS_SOCKETCALL 3082 /* Argument list sizes for sys_socketcall */ 3083 #define AL(x) ((x) * sizeof(unsigned long)) 3084 static const unsigned char nargs[21] = { 3085 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3), 3086 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6), 3087 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3), 3088 AL(4), AL(5), AL(4) 3089 }; 3090 3091 #undef AL 3092 3093 /* 3094 * System call vectors. 3095 * 3096 * Argument checking cleaned up. Saved 20% in size. 3097 * This function doesn't need to set the kernel lock because 3098 * it is set by the callees. 3099 */ 3100 3101 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args) 3102 { 3103 unsigned long a[AUDITSC_ARGS]; 3104 unsigned long a0, a1; 3105 int err; 3106 unsigned int len; 3107 3108 if (call < 1 || call > SYS_SENDMMSG) 3109 return -EINVAL; 3110 call = array_index_nospec(call, SYS_SENDMMSG + 1); 3111 3112 len = nargs[call]; 3113 if (len > sizeof(a)) 3114 return -EINVAL; 3115 3116 /* copy_from_user should be SMP safe. */ 3117 if (copy_from_user(a, args, len)) 3118 return -EFAULT; 3119 3120 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a); 3121 if (err) 3122 return err; 3123 3124 a0 = a[0]; 3125 a1 = a[1]; 3126 3127 switch (call) { 3128 case SYS_SOCKET: 3129 err = __sys_socket(a0, a1, a[2]); 3130 break; 3131 case SYS_BIND: 3132 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]); 3133 break; 3134 case SYS_CONNECT: 3135 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]); 3136 break; 3137 case SYS_LISTEN: 3138 err = __sys_listen(a0, a1); 3139 break; 3140 case SYS_ACCEPT: 3141 err = __sys_accept4(a0, (struct sockaddr __user *)a1, 3142 (int __user *)a[2], 0); 3143 break; 3144 case SYS_GETSOCKNAME: 3145 err = 3146 __sys_getsockname(a0, (struct sockaddr __user *)a1, 3147 (int __user *)a[2]); 3148 break; 3149 case SYS_GETPEERNAME: 3150 err = 3151 __sys_getpeername(a0, (struct sockaddr __user *)a1, 3152 (int __user *)a[2]); 3153 break; 3154 case SYS_SOCKETPAIR: 3155 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]); 3156 break; 3157 case SYS_SEND: 3158 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3], 3159 NULL, 0); 3160 break; 3161 case SYS_SENDTO: 3162 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3], 3163 (struct sockaddr __user *)a[4], a[5]); 3164 break; 3165 case SYS_RECV: 3166 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3], 3167 NULL, NULL); 3168 break; 3169 case SYS_RECVFROM: 3170 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3], 3171 (struct sockaddr __user *)a[4], 3172 (int __user *)a[5]); 3173 break; 3174 case SYS_SHUTDOWN: 3175 err = __sys_shutdown(a0, a1); 3176 break; 3177 case SYS_SETSOCKOPT: 3178 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3], 3179 a[4]); 3180 break; 3181 case SYS_GETSOCKOPT: 3182 err = 3183 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3], 3184 (int __user *)a[4]); 3185 break; 3186 case SYS_SENDMSG: 3187 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1, 3188 a[2], true); 3189 break; 3190 case SYS_SENDMMSG: 3191 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2], 3192 a[3], true); 3193 break; 3194 case SYS_RECVMSG: 3195 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1, 3196 a[2], true); 3197 break; 3198 case SYS_RECVMMSG: 3199 if (IS_ENABLED(CONFIG_64BIT)) 3200 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1, 3201 a[2], a[3], 3202 (struct __kernel_timespec __user *)a[4], 3203 NULL); 3204 else 3205 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1, 3206 a[2], a[3], NULL, 3207 (struct old_timespec32 __user *)a[4]); 3208 break; 3209 case SYS_ACCEPT4: 3210 err = __sys_accept4(a0, (struct sockaddr __user *)a1, 3211 (int __user *)a[2], a[3]); 3212 break; 3213 default: 3214 err = -EINVAL; 3215 break; 3216 } 3217 return err; 3218 } 3219 3220 #endif /* __ARCH_WANT_SYS_SOCKETCALL */ 3221 3222 /** 3223 * sock_register - add a socket protocol handler 3224 * @ops: description of protocol 3225 * 3226 * This function is called by a protocol handler that wants to 3227 * advertise its address family, and have it linked into the 3228 * socket interface. The value ops->family corresponds to the 3229 * socket system call protocol family. 3230 */ 3231 int sock_register(const struct net_proto_family *ops) 3232 { 3233 int err; 3234 3235 if (ops->family >= NPROTO) { 3236 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO); 3237 return -ENOBUFS; 3238 } 3239 3240 spin_lock(&net_family_lock); 3241 if (rcu_dereference_protected(net_families[ops->family], 3242 lockdep_is_held(&net_family_lock))) 3243 err = -EEXIST; 3244 else { 3245 rcu_assign_pointer(net_families[ops->family], ops); 3246 err = 0; 3247 } 3248 spin_unlock(&net_family_lock); 3249 3250 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]); 3251 return err; 3252 } 3253 EXPORT_SYMBOL(sock_register); 3254 3255 /** 3256 * sock_unregister - remove a protocol handler 3257 * @family: protocol family to remove 3258 * 3259 * This function is called by a protocol handler that wants to 3260 * remove its address family, and have it unlinked from the 3261 * new socket creation. 3262 * 3263 * If protocol handler is a module, then it can use module reference 3264 * counts to protect against new references. If protocol handler is not 3265 * a module then it needs to provide its own protection in 3266 * the ops->create routine. 3267 */ 3268 void sock_unregister(int family) 3269 { 3270 BUG_ON(family < 0 || family >= NPROTO); 3271 3272 spin_lock(&net_family_lock); 3273 RCU_INIT_POINTER(net_families[family], NULL); 3274 spin_unlock(&net_family_lock); 3275 3276 synchronize_rcu(); 3277 3278 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]); 3279 } 3280 EXPORT_SYMBOL(sock_unregister); 3281 3282 bool sock_is_registered(int family) 3283 { 3284 return family < NPROTO && rcu_access_pointer(net_families[family]); 3285 } 3286 3287 static int __init sock_init(void) 3288 { 3289 int err; 3290 /* 3291 * Initialize the network sysctl infrastructure. 3292 */ 3293 err = net_sysctl_init(); 3294 if (err) 3295 goto out; 3296 3297 /* 3298 * Initialize skbuff SLAB cache 3299 */ 3300 skb_init(); 3301 3302 /* 3303 * Initialize the protocols module. 3304 */ 3305 3306 init_inodecache(); 3307 3308 err = register_filesystem(&sock_fs_type); 3309 if (err) 3310 goto out; 3311 sock_mnt = kern_mount(&sock_fs_type); 3312 if (IS_ERR(sock_mnt)) { 3313 err = PTR_ERR(sock_mnt); 3314 goto out_mount; 3315 } 3316 3317 /* The real protocol initialization is performed in later initcalls. 3318 */ 3319 3320 #ifdef CONFIG_NETFILTER 3321 err = netfilter_init(); 3322 if (err) 3323 goto out; 3324 #endif 3325 3326 ptp_classifier_init(); 3327 3328 out: 3329 return err; 3330 3331 out_mount: 3332 unregister_filesystem(&sock_fs_type); 3333 goto out; 3334 } 3335 3336 core_initcall(sock_init); /* early initcall */ 3337 3338 #ifdef CONFIG_PROC_FS 3339 void socket_seq_show(struct seq_file *seq) 3340 { 3341 seq_printf(seq, "sockets: used %d\n", 3342 sock_inuse_get(seq->private)); 3343 } 3344 #endif /* CONFIG_PROC_FS */ 3345 3346 /* Handle the fact that while struct ifreq has the same *layout* on 3347 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data, 3348 * which are handled elsewhere, it still has different *size* due to 3349 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit, 3350 * resulting in struct ifreq being 32 and 40 bytes respectively). 3351 * As a result, if the struct happens to be at the end of a page and 3352 * the next page isn't readable/writable, we get a fault. To prevent 3353 * that, copy back and forth to the full size. 3354 */ 3355 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg) 3356 { 3357 if (in_compat_syscall()) { 3358 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr; 3359 3360 memset(ifr, 0, sizeof(*ifr)); 3361 if (copy_from_user(ifr32, arg, sizeof(*ifr32))) 3362 return -EFAULT; 3363 3364 if (ifrdata) 3365 *ifrdata = compat_ptr(ifr32->ifr_data); 3366 3367 return 0; 3368 } 3369 3370 if (copy_from_user(ifr, arg, sizeof(*ifr))) 3371 return -EFAULT; 3372 3373 if (ifrdata) 3374 *ifrdata = ifr->ifr_data; 3375 3376 return 0; 3377 } 3378 EXPORT_SYMBOL(get_user_ifreq); 3379 3380 int put_user_ifreq(struct ifreq *ifr, void __user *arg) 3381 { 3382 size_t size = sizeof(*ifr); 3383 3384 if (in_compat_syscall()) 3385 size = sizeof(struct compat_ifreq); 3386 3387 if (copy_to_user(arg, ifr, size)) 3388 return -EFAULT; 3389 3390 return 0; 3391 } 3392 EXPORT_SYMBOL(put_user_ifreq); 3393 3394 #ifdef CONFIG_COMPAT 3395 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32) 3396 { 3397 compat_uptr_t uptr32; 3398 struct ifreq ifr; 3399 void __user *saved; 3400 int err; 3401 3402 if (get_user_ifreq(&ifr, NULL, uifr32)) 3403 return -EFAULT; 3404 3405 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu)) 3406 return -EFAULT; 3407 3408 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc; 3409 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32); 3410 3411 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL); 3412 if (!err) { 3413 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved; 3414 if (put_user_ifreq(&ifr, uifr32)) 3415 err = -EFAULT; 3416 } 3417 return err; 3418 } 3419 3420 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */ 3421 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd, 3422 struct compat_ifreq __user *u_ifreq32) 3423 { 3424 struct ifreq ifreq; 3425 void __user *data; 3426 3427 if (!is_socket_ioctl_cmd(cmd)) 3428 return -ENOTTY; 3429 if (get_user_ifreq(&ifreq, &data, u_ifreq32)) 3430 return -EFAULT; 3431 ifreq.ifr_data = data; 3432 3433 return dev_ioctl(net, cmd, &ifreq, data, NULL); 3434 } 3435 3436 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock, 3437 unsigned int cmd, unsigned long arg) 3438 { 3439 void __user *argp = compat_ptr(arg); 3440 struct sock *sk = sock->sk; 3441 struct net *net = sock_net(sk); 3442 const struct proto_ops *ops; 3443 3444 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) 3445 return sock_ioctl(file, cmd, (unsigned long)argp); 3446 3447 switch (cmd) { 3448 case SIOCWANDEV: 3449 return compat_siocwandev(net, argp); 3450 case SIOCGSTAMP_OLD: 3451 case SIOCGSTAMPNS_OLD: 3452 ops = READ_ONCE(sock->ops); 3453 if (!ops->gettstamp) 3454 return -ENOIOCTLCMD; 3455 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD, 3456 !COMPAT_USE_64BIT_TIME); 3457 3458 case SIOCETHTOOL: 3459 case SIOCBONDSLAVEINFOQUERY: 3460 case SIOCBONDINFOQUERY: 3461 case SIOCSHWTSTAMP: 3462 case SIOCGHWTSTAMP: 3463 return compat_ifr_data_ioctl(net, cmd, argp); 3464 3465 case FIOSETOWN: 3466 case SIOCSPGRP: 3467 case FIOGETOWN: 3468 case SIOCGPGRP: 3469 case SIOCBRADDBR: 3470 case SIOCBRDELBR: 3471 case SIOCGIFVLAN: 3472 case SIOCSIFVLAN: 3473 case SIOCGSKNS: 3474 case SIOCGSTAMP_NEW: 3475 case SIOCGSTAMPNS_NEW: 3476 case SIOCGIFCONF: 3477 case SIOCSIFBR: 3478 case SIOCGIFBR: 3479 return sock_ioctl(file, cmd, arg); 3480 3481 case SIOCGIFFLAGS: 3482 case SIOCSIFFLAGS: 3483 case SIOCGIFMAP: 3484 case SIOCSIFMAP: 3485 case SIOCGIFMETRIC: 3486 case SIOCSIFMETRIC: 3487 case SIOCGIFMTU: 3488 case SIOCSIFMTU: 3489 case SIOCGIFMEM: 3490 case SIOCSIFMEM: 3491 case SIOCGIFHWADDR: 3492 case SIOCSIFHWADDR: 3493 case SIOCADDMULTI: 3494 case SIOCDELMULTI: 3495 case SIOCGIFINDEX: 3496 case SIOCGIFADDR: 3497 case SIOCSIFADDR: 3498 case SIOCSIFHWBROADCAST: 3499 case SIOCDIFADDR: 3500 case SIOCGIFBRDADDR: 3501 case SIOCSIFBRDADDR: 3502 case SIOCGIFDSTADDR: 3503 case SIOCSIFDSTADDR: 3504 case SIOCGIFNETMASK: 3505 case SIOCSIFNETMASK: 3506 case SIOCSIFPFLAGS: 3507 case SIOCGIFPFLAGS: 3508 case SIOCGIFTXQLEN: 3509 case SIOCSIFTXQLEN: 3510 case SIOCBRADDIF: 3511 case SIOCBRDELIF: 3512 case SIOCGIFNAME: 3513 case SIOCSIFNAME: 3514 case SIOCGMIIPHY: 3515 case SIOCGMIIREG: 3516 case SIOCSMIIREG: 3517 case SIOCBONDENSLAVE: 3518 case SIOCBONDRELEASE: 3519 case SIOCBONDSETHWADDR: 3520 case SIOCBONDCHANGEACTIVE: 3521 case SIOCSARP: 3522 case SIOCGARP: 3523 case SIOCDARP: 3524 case SIOCOUTQ: 3525 case SIOCOUTQNSD: 3526 case SIOCATMARK: 3527 return sock_do_ioctl(net, sock, cmd, arg); 3528 } 3529 3530 return -ENOIOCTLCMD; 3531 } 3532 3533 static long compat_sock_ioctl(struct file *file, unsigned int cmd, 3534 unsigned long arg) 3535 { 3536 struct socket *sock = file->private_data; 3537 const struct proto_ops *ops = READ_ONCE(sock->ops); 3538 int ret = -ENOIOCTLCMD; 3539 struct sock *sk; 3540 struct net *net; 3541 3542 sk = sock->sk; 3543 net = sock_net(sk); 3544 3545 if (ops->compat_ioctl) 3546 ret = ops->compat_ioctl(sock, cmd, arg); 3547 3548 if (ret == -ENOIOCTLCMD && 3549 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)) 3550 ret = compat_wext_handle_ioctl(net, cmd, arg); 3551 3552 if (ret == -ENOIOCTLCMD) 3553 ret = compat_sock_ioctl_trans(file, sock, cmd, arg); 3554 3555 return ret; 3556 } 3557 #endif 3558 3559 /** 3560 * kernel_bind - bind an address to a socket (kernel space) 3561 * @sock: socket 3562 * @addr: address 3563 * @addrlen: length of address 3564 * 3565 * Returns 0 or an error. 3566 */ 3567 3568 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen) 3569 { 3570 struct sockaddr_storage address; 3571 3572 memcpy(&address, addr, addrlen); 3573 3574 return READ_ONCE(sock->ops)->bind(sock, (struct sockaddr *)&address, 3575 addrlen); 3576 } 3577 EXPORT_SYMBOL(kernel_bind); 3578 3579 /** 3580 * kernel_listen - move socket to listening state (kernel space) 3581 * @sock: socket 3582 * @backlog: pending connections queue size 3583 * 3584 * Returns 0 or an error. 3585 */ 3586 3587 int kernel_listen(struct socket *sock, int backlog) 3588 { 3589 return READ_ONCE(sock->ops)->listen(sock, backlog); 3590 } 3591 EXPORT_SYMBOL(kernel_listen); 3592 3593 /** 3594 * kernel_accept - accept a connection (kernel space) 3595 * @sock: listening socket 3596 * @newsock: new connected socket 3597 * @flags: flags 3598 * 3599 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0. 3600 * If it fails, @newsock is guaranteed to be %NULL. 3601 * Returns 0 or an error. 3602 */ 3603 3604 int kernel_accept(struct socket *sock, struct socket **newsock, int flags) 3605 { 3606 struct sock *sk = sock->sk; 3607 const struct proto_ops *ops = READ_ONCE(sock->ops); 3608 struct proto_accept_arg arg = { 3609 .flags = flags, 3610 .kern = true, 3611 }; 3612 int err; 3613 3614 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, 3615 newsock); 3616 if (err < 0) 3617 goto done; 3618 3619 err = ops->accept(sock, *newsock, &arg); 3620 if (err < 0) { 3621 sock_release(*newsock); 3622 *newsock = NULL; 3623 goto done; 3624 } 3625 3626 (*newsock)->ops = ops; 3627 __module_get(ops->owner); 3628 3629 done: 3630 return err; 3631 } 3632 EXPORT_SYMBOL(kernel_accept); 3633 3634 /** 3635 * kernel_connect - connect a socket (kernel space) 3636 * @sock: socket 3637 * @addr: address 3638 * @addrlen: address length 3639 * @flags: flags (O_NONBLOCK, ...) 3640 * 3641 * For datagram sockets, @addr is the address to which datagrams are sent 3642 * by default, and the only address from which datagrams are received. 3643 * For stream sockets, attempts to connect to @addr. 3644 * Returns 0 or an error code. 3645 */ 3646 3647 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen, 3648 int flags) 3649 { 3650 struct sockaddr_storage address; 3651 3652 memcpy(&address, addr, addrlen); 3653 3654 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address, 3655 addrlen, flags); 3656 } 3657 EXPORT_SYMBOL(kernel_connect); 3658 3659 /** 3660 * kernel_getsockname - get the address which the socket is bound (kernel space) 3661 * @sock: socket 3662 * @addr: address holder 3663 * 3664 * Fills the @addr pointer with the address which the socket is bound. 3665 * Returns the length of the address in bytes or an error code. 3666 */ 3667 3668 int kernel_getsockname(struct socket *sock, struct sockaddr *addr) 3669 { 3670 return READ_ONCE(sock->ops)->getname(sock, addr, 0); 3671 } 3672 EXPORT_SYMBOL(kernel_getsockname); 3673 3674 /** 3675 * kernel_getpeername - get the address which the socket is connected (kernel space) 3676 * @sock: socket 3677 * @addr: address holder 3678 * 3679 * Fills the @addr pointer with the address which the socket is connected. 3680 * Returns the length of the address in bytes or an error code. 3681 */ 3682 3683 int kernel_getpeername(struct socket *sock, struct sockaddr *addr) 3684 { 3685 return READ_ONCE(sock->ops)->getname(sock, addr, 1); 3686 } 3687 EXPORT_SYMBOL(kernel_getpeername); 3688 3689 /** 3690 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space) 3691 * @sock: socket 3692 * @how: connection part 3693 * 3694 * Returns 0 or an error. 3695 */ 3696 3697 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how) 3698 { 3699 return READ_ONCE(sock->ops)->shutdown(sock, how); 3700 } 3701 EXPORT_SYMBOL(kernel_sock_shutdown); 3702 3703 /** 3704 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket 3705 * @sk: socket 3706 * 3707 * This routine returns the IP overhead imposed by a socket i.e. 3708 * the length of the underlying IP header, depending on whether 3709 * this is an IPv4 or IPv6 socket and the length from IP options turned 3710 * on at the socket. Assumes that the caller has a lock on the socket. 3711 */ 3712 3713 u32 kernel_sock_ip_overhead(struct sock *sk) 3714 { 3715 struct inet_sock *inet; 3716 struct ip_options_rcu *opt; 3717 u32 overhead = 0; 3718 #if IS_ENABLED(CONFIG_IPV6) 3719 struct ipv6_pinfo *np; 3720 struct ipv6_txoptions *optv6 = NULL; 3721 #endif /* IS_ENABLED(CONFIG_IPV6) */ 3722 3723 if (!sk) 3724 return overhead; 3725 3726 switch (sk->sk_family) { 3727 case AF_INET: 3728 inet = inet_sk(sk); 3729 overhead += sizeof(struct iphdr); 3730 opt = rcu_dereference_protected(inet->inet_opt, 3731 sock_owned_by_user(sk)); 3732 if (opt) 3733 overhead += opt->opt.optlen; 3734 return overhead; 3735 #if IS_ENABLED(CONFIG_IPV6) 3736 case AF_INET6: 3737 np = inet6_sk(sk); 3738 overhead += sizeof(struct ipv6hdr); 3739 if (np) 3740 optv6 = rcu_dereference_protected(np->opt, 3741 sock_owned_by_user(sk)); 3742 if (optv6) 3743 overhead += (optv6->opt_flen + optv6->opt_nflen); 3744 return overhead; 3745 #endif /* IS_ENABLED(CONFIG_IPV6) */ 3746 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */ 3747 return overhead; 3748 } 3749 } 3750 EXPORT_SYMBOL(kernel_sock_ip_overhead); 3751