1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $ 11 * 12 * Authors: Ross Biro 13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 14 * Florian La Roche, <flla@stud.uni-sb.de> 15 * Alan Cox, <A.Cox@swansea.ac.uk> 16 * 17 * Fixes: 18 * Alan Cox : Numerous verify_area() problems 19 * Alan Cox : Connecting on a connecting socket 20 * now returns an error for tcp. 21 * Alan Cox : sock->protocol is set correctly. 22 * and is not sometimes left as 0. 23 * Alan Cox : connect handles icmp errors on a 24 * connect properly. Unfortunately there 25 * is a restart syscall nasty there. I 26 * can't match BSD without hacking the C 27 * library. Ideas urgently sought! 28 * Alan Cox : Disallow bind() to addresses that are 29 * not ours - especially broadcast ones!! 30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 32 * instead they leave that for the DESTROY timer. 33 * Alan Cox : Clean up error flag in accept 34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 35 * was buggy. Put a remove_sock() in the handler 36 * for memory when we hit 0. Also altered the timer 37 * code. The ACK stuff can wait and needs major 38 * TCP layer surgery. 39 * Alan Cox : Fixed TCP ack bug, removed remove sock 40 * and fixed timer/inet_bh race. 41 * Alan Cox : Added zapped flag for TCP 42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 47 * Rick Sladkey : Relaxed UDP rules for matching packets. 48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 49 * Pauline Middelink : identd support 50 * Alan Cox : Fixed connect() taking signals I think. 51 * Alan Cox : SO_LINGER supported 52 * Alan Cox : Error reporting fixes 53 * Anonymous : inet_create tidied up (sk->reuse setting) 54 * Alan Cox : inet sockets don't set sk->type! 55 * Alan Cox : Split socket option code 56 * Alan Cox : Callbacks 57 * Alan Cox : Nagle flag for Charles & Johannes stuff 58 * Alex : Removed restriction on inet fioctl 59 * Alan Cox : Splitting INET from NET core 60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 62 * Alan Cox : Split IP from generic code 63 * Alan Cox : New kfree_skbmem() 64 * Alan Cox : Make SO_DEBUG superuser only. 65 * Alan Cox : Allow anyone to clear SO_DEBUG 66 * (compatibility fix) 67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 68 * Alan Cox : Allocator for a socket is settable. 69 * Alan Cox : SO_ERROR includes soft errors. 70 * Alan Cox : Allow NULL arguments on some SO_ opts 71 * Alan Cox : Generic socket allocation to make hooks 72 * easier (suggested by Craig Metz). 73 * Michael Pall : SO_ERROR returns positive errno again 74 * Steve Whitehouse: Added default destructor to free 75 * protocol private data. 76 * Steve Whitehouse: Added various other default routines 77 * common to several socket families. 78 * Chris Evans : Call suser() check last on F_SETOWN 79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 81 * Andi Kleen : Fix write_space callback 82 * Chris Evans : Security fixes - signedness again 83 * Arnaldo C. Melo : cleanups, use skb_queue_purge 84 * 85 * To Fix: 86 * 87 * 88 * This program is free software; you can redistribute it and/or 89 * modify it under the terms of the GNU General Public License 90 * as published by the Free Software Foundation; either version 91 * 2 of the License, or (at your option) any later version. 92 */ 93 94 #include <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/types.h> 97 #include <linux/socket.h> 98 #include <linux/in.h> 99 #include <linux/kernel.h> 100 #include <linux/module.h> 101 #include <linux/proc_fs.h> 102 #include <linux/seq_file.h> 103 #include <linux/sched.h> 104 #include <linux/timer.h> 105 #include <linux/string.h> 106 #include <linux/sockios.h> 107 #include <linux/net.h> 108 #include <linux/mm.h> 109 #include <linux/slab.h> 110 #include <linux/interrupt.h> 111 #include <linux/poll.h> 112 #include <linux/tcp.h> 113 #include <linux/init.h> 114 #include <linux/highmem.h> 115 116 #include <asm/uaccess.h> 117 #include <asm/system.h> 118 119 #include <linux/netdevice.h> 120 #include <net/protocol.h> 121 #include <linux/skbuff.h> 122 #include <net/net_namespace.h> 123 #include <net/request_sock.h> 124 #include <net/sock.h> 125 #include <net/xfrm.h> 126 #include <linux/ipsec.h> 127 128 #include <linux/filter.h> 129 130 #ifdef CONFIG_INET 131 #include <net/tcp.h> 132 #endif 133 134 /* 135 * Each address family might have different locking rules, so we have 136 * one slock key per address family: 137 */ 138 static struct lock_class_key af_family_keys[AF_MAX]; 139 static struct lock_class_key af_family_slock_keys[AF_MAX]; 140 141 #ifdef CONFIG_DEBUG_LOCK_ALLOC 142 /* 143 * Make lock validator output more readable. (we pre-construct these 144 * strings build-time, so that runtime initialization of socket 145 * locks is fast): 146 */ 147 static const char *af_family_key_strings[AF_MAX+1] = { 148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 155 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX" 160 }; 161 static const char *af_family_slock_key_strings[AF_MAX+1] = { 162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" , 170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 171 "slock-27" , "slock-28" , "slock-AF_CAN" , 172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 173 "slock-AF_RXRPC" , "slock-AF_MAX" 174 }; 175 static const char *af_family_clock_key_strings[AF_MAX+1] = { 176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" , 184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 185 "clock-27" , "clock-28" , "clock-29" , 186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 187 "clock-AF_RXRPC" , "clock-AF_MAX" 188 }; 189 #endif 190 191 /* 192 * sk_callback_lock locking rules are per-address-family, 193 * so split the lock classes by using a per-AF key: 194 */ 195 static struct lock_class_key af_callback_keys[AF_MAX]; 196 197 /* Take into consideration the size of the struct sk_buff overhead in the 198 * determination of these values, since that is non-constant across 199 * platforms. This makes socket queueing behavior and performance 200 * not depend upon such differences. 201 */ 202 #define _SK_MEM_PACKETS 256 203 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256) 204 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 205 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 206 207 /* Run time adjustable parameters. */ 208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 212 213 /* Maximal space eaten by iovec or ancilliary data plus some space */ 214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 215 216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 217 { 218 struct timeval tv; 219 220 if (optlen < sizeof(tv)) 221 return -EINVAL; 222 if (copy_from_user(&tv, optval, sizeof(tv))) 223 return -EFAULT; 224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 225 return -EDOM; 226 227 if (tv.tv_sec < 0) { 228 static int warned __read_mostly; 229 230 *timeo_p = 0; 231 if (warned < 10 && net_ratelimit()) 232 warned++; 233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) " 234 "tries to set negative timeout\n", 235 current->comm, task_pid_nr(current)); 236 return 0; 237 } 238 *timeo_p = MAX_SCHEDULE_TIMEOUT; 239 if (tv.tv_sec == 0 && tv.tv_usec == 0) 240 return 0; 241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 243 return 0; 244 } 245 246 static void sock_warn_obsolete_bsdism(const char *name) 247 { 248 static int warned; 249 static char warncomm[TASK_COMM_LEN]; 250 if (strcmp(warncomm, current->comm) && warned < 5) { 251 strcpy(warncomm, current->comm); 252 printk(KERN_WARNING "process `%s' is using obsolete " 253 "%s SO_BSDCOMPAT\n", warncomm, name); 254 warned++; 255 } 256 } 257 258 static void sock_disable_timestamp(struct sock *sk) 259 { 260 if (sock_flag(sk, SOCK_TIMESTAMP)) { 261 sock_reset_flag(sk, SOCK_TIMESTAMP); 262 net_disable_timestamp(); 263 } 264 } 265 266 267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 268 { 269 int err = 0; 270 int skb_len; 271 272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 273 number of warnings when compiling with -W --ANK 274 */ 275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 276 (unsigned)sk->sk_rcvbuf) { 277 err = -ENOMEM; 278 goto out; 279 } 280 281 err = sk_filter(sk, skb); 282 if (err) 283 goto out; 284 285 if (!sk_rmem_schedule(sk, skb->truesize)) { 286 err = -ENOBUFS; 287 goto out; 288 } 289 290 skb->dev = NULL; 291 skb_set_owner_r(skb, sk); 292 293 /* Cache the SKB length before we tack it onto the receive 294 * queue. Once it is added it no longer belongs to us and 295 * may be freed by other threads of control pulling packets 296 * from the queue. 297 */ 298 skb_len = skb->len; 299 300 skb_queue_tail(&sk->sk_receive_queue, skb); 301 302 if (!sock_flag(sk, SOCK_DEAD)) 303 sk->sk_data_ready(sk, skb_len); 304 out: 305 return err; 306 } 307 EXPORT_SYMBOL(sock_queue_rcv_skb); 308 309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 310 { 311 int rc = NET_RX_SUCCESS; 312 313 if (sk_filter(sk, skb)) 314 goto discard_and_relse; 315 316 skb->dev = NULL; 317 318 if (nested) 319 bh_lock_sock_nested(sk); 320 else 321 bh_lock_sock(sk); 322 if (!sock_owned_by_user(sk)) { 323 /* 324 * trylock + unlock semantics: 325 */ 326 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 327 328 rc = sk->sk_backlog_rcv(sk, skb); 329 330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 331 } else 332 sk_add_backlog(sk, skb); 333 bh_unlock_sock(sk); 334 out: 335 sock_put(sk); 336 return rc; 337 discard_and_relse: 338 kfree_skb(skb); 339 goto out; 340 } 341 EXPORT_SYMBOL(sk_receive_skb); 342 343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 344 { 345 struct dst_entry *dst = sk->sk_dst_cache; 346 347 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 348 sk->sk_dst_cache = NULL; 349 dst_release(dst); 350 return NULL; 351 } 352 353 return dst; 354 } 355 EXPORT_SYMBOL(__sk_dst_check); 356 357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 358 { 359 struct dst_entry *dst = sk_dst_get(sk); 360 361 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 362 sk_dst_reset(sk); 363 dst_release(dst); 364 return NULL; 365 } 366 367 return dst; 368 } 369 EXPORT_SYMBOL(sk_dst_check); 370 371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen) 372 { 373 int ret = -ENOPROTOOPT; 374 #ifdef CONFIG_NETDEVICES 375 struct net *net = sock_net(sk); 376 char devname[IFNAMSIZ]; 377 int index; 378 379 /* Sorry... */ 380 ret = -EPERM; 381 if (!capable(CAP_NET_RAW)) 382 goto out; 383 384 ret = -EINVAL; 385 if (optlen < 0) 386 goto out; 387 388 /* Bind this socket to a particular device like "eth0", 389 * as specified in the passed interface name. If the 390 * name is "" or the option length is zero the socket 391 * is not bound. 392 */ 393 if (optlen > IFNAMSIZ - 1) 394 optlen = IFNAMSIZ - 1; 395 memset(devname, 0, sizeof(devname)); 396 397 ret = -EFAULT; 398 if (copy_from_user(devname, optval, optlen)) 399 goto out; 400 401 if (devname[0] == '\0') { 402 index = 0; 403 } else { 404 struct net_device *dev = dev_get_by_name(net, devname); 405 406 ret = -ENODEV; 407 if (!dev) 408 goto out; 409 410 index = dev->ifindex; 411 dev_put(dev); 412 } 413 414 lock_sock(sk); 415 sk->sk_bound_dev_if = index; 416 sk_dst_reset(sk); 417 release_sock(sk); 418 419 ret = 0; 420 421 out: 422 #endif 423 424 return ret; 425 } 426 427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 428 { 429 if (valbool) 430 sock_set_flag(sk, bit); 431 else 432 sock_reset_flag(sk, bit); 433 } 434 435 /* 436 * This is meant for all protocols to use and covers goings on 437 * at the socket level. Everything here is generic. 438 */ 439 440 int sock_setsockopt(struct socket *sock, int level, int optname, 441 char __user *optval, int optlen) 442 { 443 struct sock *sk=sock->sk; 444 int val; 445 int valbool; 446 struct linger ling; 447 int ret = 0; 448 449 /* 450 * Options without arguments 451 */ 452 453 #ifdef SO_DONTLINGER /* Compatibility item... */ 454 if (optname == SO_DONTLINGER) { 455 lock_sock(sk); 456 sock_reset_flag(sk, SOCK_LINGER); 457 release_sock(sk); 458 return 0; 459 } 460 #endif 461 462 if (optname == SO_BINDTODEVICE) 463 return sock_bindtodevice(sk, optval, optlen); 464 465 if (optlen < sizeof(int)) 466 return -EINVAL; 467 468 if (get_user(val, (int __user *)optval)) 469 return -EFAULT; 470 471 valbool = val?1:0; 472 473 lock_sock(sk); 474 475 switch(optname) { 476 case SO_DEBUG: 477 if (val && !capable(CAP_NET_ADMIN)) { 478 ret = -EACCES; 479 } else 480 sock_valbool_flag(sk, SOCK_DBG, valbool); 481 break; 482 case SO_REUSEADDR: 483 sk->sk_reuse = valbool; 484 break; 485 case SO_TYPE: 486 case SO_ERROR: 487 ret = -ENOPROTOOPT; 488 break; 489 case SO_DONTROUTE: 490 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 491 break; 492 case SO_BROADCAST: 493 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 494 break; 495 case SO_SNDBUF: 496 /* Don't error on this BSD doesn't and if you think 497 about it this is right. Otherwise apps have to 498 play 'guess the biggest size' games. RCVBUF/SNDBUF 499 are treated in BSD as hints */ 500 501 if (val > sysctl_wmem_max) 502 val = sysctl_wmem_max; 503 set_sndbuf: 504 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 505 if ((val * 2) < SOCK_MIN_SNDBUF) 506 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 507 else 508 sk->sk_sndbuf = val * 2; 509 510 /* 511 * Wake up sending tasks if we 512 * upped the value. 513 */ 514 sk->sk_write_space(sk); 515 break; 516 517 case SO_SNDBUFFORCE: 518 if (!capable(CAP_NET_ADMIN)) { 519 ret = -EPERM; 520 break; 521 } 522 goto set_sndbuf; 523 524 case SO_RCVBUF: 525 /* Don't error on this BSD doesn't and if you think 526 about it this is right. Otherwise apps have to 527 play 'guess the biggest size' games. RCVBUF/SNDBUF 528 are treated in BSD as hints */ 529 530 if (val > sysctl_rmem_max) 531 val = sysctl_rmem_max; 532 set_rcvbuf: 533 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 534 /* 535 * We double it on the way in to account for 536 * "struct sk_buff" etc. overhead. Applications 537 * assume that the SO_RCVBUF setting they make will 538 * allow that much actual data to be received on that 539 * socket. 540 * 541 * Applications are unaware that "struct sk_buff" and 542 * other overheads allocate from the receive buffer 543 * during socket buffer allocation. 544 * 545 * And after considering the possible alternatives, 546 * returning the value we actually used in getsockopt 547 * is the most desirable behavior. 548 */ 549 if ((val * 2) < SOCK_MIN_RCVBUF) 550 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 551 else 552 sk->sk_rcvbuf = val * 2; 553 break; 554 555 case SO_RCVBUFFORCE: 556 if (!capable(CAP_NET_ADMIN)) { 557 ret = -EPERM; 558 break; 559 } 560 goto set_rcvbuf; 561 562 case SO_KEEPALIVE: 563 #ifdef CONFIG_INET 564 if (sk->sk_protocol == IPPROTO_TCP) 565 tcp_set_keepalive(sk, valbool); 566 #endif 567 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 568 break; 569 570 case SO_OOBINLINE: 571 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 572 break; 573 574 case SO_NO_CHECK: 575 sk->sk_no_check = valbool; 576 break; 577 578 case SO_PRIORITY: 579 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 580 sk->sk_priority = val; 581 else 582 ret = -EPERM; 583 break; 584 585 case SO_LINGER: 586 if (optlen < sizeof(ling)) { 587 ret = -EINVAL; /* 1003.1g */ 588 break; 589 } 590 if (copy_from_user(&ling,optval,sizeof(ling))) { 591 ret = -EFAULT; 592 break; 593 } 594 if (!ling.l_onoff) 595 sock_reset_flag(sk, SOCK_LINGER); 596 else { 597 #if (BITS_PER_LONG == 32) 598 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 599 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 600 else 601 #endif 602 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 603 sock_set_flag(sk, SOCK_LINGER); 604 } 605 break; 606 607 case SO_BSDCOMPAT: 608 sock_warn_obsolete_bsdism("setsockopt"); 609 break; 610 611 case SO_PASSCRED: 612 if (valbool) 613 set_bit(SOCK_PASSCRED, &sock->flags); 614 else 615 clear_bit(SOCK_PASSCRED, &sock->flags); 616 break; 617 618 case SO_TIMESTAMP: 619 case SO_TIMESTAMPNS: 620 if (valbool) { 621 if (optname == SO_TIMESTAMP) 622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 623 else 624 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 625 sock_set_flag(sk, SOCK_RCVTSTAMP); 626 sock_enable_timestamp(sk); 627 } else { 628 sock_reset_flag(sk, SOCK_RCVTSTAMP); 629 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 630 } 631 break; 632 633 case SO_RCVLOWAT: 634 if (val < 0) 635 val = INT_MAX; 636 sk->sk_rcvlowat = val ? : 1; 637 break; 638 639 case SO_RCVTIMEO: 640 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 641 break; 642 643 case SO_SNDTIMEO: 644 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 645 break; 646 647 case SO_ATTACH_FILTER: 648 ret = -EINVAL; 649 if (optlen == sizeof(struct sock_fprog)) { 650 struct sock_fprog fprog; 651 652 ret = -EFAULT; 653 if (copy_from_user(&fprog, optval, sizeof(fprog))) 654 break; 655 656 ret = sk_attach_filter(&fprog, sk); 657 } 658 break; 659 660 case SO_DETACH_FILTER: 661 ret = sk_detach_filter(sk); 662 break; 663 664 case SO_PASSSEC: 665 if (valbool) 666 set_bit(SOCK_PASSSEC, &sock->flags); 667 else 668 clear_bit(SOCK_PASSSEC, &sock->flags); 669 break; 670 case SO_MARK: 671 if (!capable(CAP_NET_ADMIN)) 672 ret = -EPERM; 673 else { 674 sk->sk_mark = val; 675 } 676 break; 677 678 /* We implement the SO_SNDLOWAT etc to 679 not be settable (1003.1g 5.3) */ 680 default: 681 ret = -ENOPROTOOPT; 682 break; 683 } 684 release_sock(sk); 685 return ret; 686 } 687 688 689 int sock_getsockopt(struct socket *sock, int level, int optname, 690 char __user *optval, int __user *optlen) 691 { 692 struct sock *sk = sock->sk; 693 694 union { 695 int val; 696 struct linger ling; 697 struct timeval tm; 698 } v; 699 700 unsigned int lv = sizeof(int); 701 int len; 702 703 if (get_user(len, optlen)) 704 return -EFAULT; 705 if (len < 0) 706 return -EINVAL; 707 708 switch(optname) { 709 case SO_DEBUG: 710 v.val = sock_flag(sk, SOCK_DBG); 711 break; 712 713 case SO_DONTROUTE: 714 v.val = sock_flag(sk, SOCK_LOCALROUTE); 715 break; 716 717 case SO_BROADCAST: 718 v.val = !!sock_flag(sk, SOCK_BROADCAST); 719 break; 720 721 case SO_SNDBUF: 722 v.val = sk->sk_sndbuf; 723 break; 724 725 case SO_RCVBUF: 726 v.val = sk->sk_rcvbuf; 727 break; 728 729 case SO_REUSEADDR: 730 v.val = sk->sk_reuse; 731 break; 732 733 case SO_KEEPALIVE: 734 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 735 break; 736 737 case SO_TYPE: 738 v.val = sk->sk_type; 739 break; 740 741 case SO_ERROR: 742 v.val = -sock_error(sk); 743 if (v.val==0) 744 v.val = xchg(&sk->sk_err_soft, 0); 745 break; 746 747 case SO_OOBINLINE: 748 v.val = !!sock_flag(sk, SOCK_URGINLINE); 749 break; 750 751 case SO_NO_CHECK: 752 v.val = sk->sk_no_check; 753 break; 754 755 case SO_PRIORITY: 756 v.val = sk->sk_priority; 757 break; 758 759 case SO_LINGER: 760 lv = sizeof(v.ling); 761 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 762 v.ling.l_linger = sk->sk_lingertime / HZ; 763 break; 764 765 case SO_BSDCOMPAT: 766 sock_warn_obsolete_bsdism("getsockopt"); 767 break; 768 769 case SO_TIMESTAMP: 770 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 771 !sock_flag(sk, SOCK_RCVTSTAMPNS); 772 break; 773 774 case SO_TIMESTAMPNS: 775 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 776 break; 777 778 case SO_RCVTIMEO: 779 lv=sizeof(struct timeval); 780 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 781 v.tm.tv_sec = 0; 782 v.tm.tv_usec = 0; 783 } else { 784 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 785 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 786 } 787 break; 788 789 case SO_SNDTIMEO: 790 lv=sizeof(struct timeval); 791 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 792 v.tm.tv_sec = 0; 793 v.tm.tv_usec = 0; 794 } else { 795 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 796 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 797 } 798 break; 799 800 case SO_RCVLOWAT: 801 v.val = sk->sk_rcvlowat; 802 break; 803 804 case SO_SNDLOWAT: 805 v.val=1; 806 break; 807 808 case SO_PASSCRED: 809 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 810 break; 811 812 case SO_PEERCRED: 813 if (len > sizeof(sk->sk_peercred)) 814 len = sizeof(sk->sk_peercred); 815 if (copy_to_user(optval, &sk->sk_peercred, len)) 816 return -EFAULT; 817 goto lenout; 818 819 case SO_PEERNAME: 820 { 821 char address[128]; 822 823 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 824 return -ENOTCONN; 825 if (lv < len) 826 return -EINVAL; 827 if (copy_to_user(optval, address, len)) 828 return -EFAULT; 829 goto lenout; 830 } 831 832 /* Dubious BSD thing... Probably nobody even uses it, but 833 * the UNIX standard wants it for whatever reason... -DaveM 834 */ 835 case SO_ACCEPTCONN: 836 v.val = sk->sk_state == TCP_LISTEN; 837 break; 838 839 case SO_PASSSEC: 840 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 841 break; 842 843 case SO_PEERSEC: 844 return security_socket_getpeersec_stream(sock, optval, optlen, len); 845 846 case SO_MARK: 847 v.val = sk->sk_mark; 848 break; 849 850 default: 851 return -ENOPROTOOPT; 852 } 853 854 if (len > lv) 855 len = lv; 856 if (copy_to_user(optval, &v, len)) 857 return -EFAULT; 858 lenout: 859 if (put_user(len, optlen)) 860 return -EFAULT; 861 return 0; 862 } 863 864 /* 865 * Initialize an sk_lock. 866 * 867 * (We also register the sk_lock with the lock validator.) 868 */ 869 static inline void sock_lock_init(struct sock *sk) 870 { 871 sock_lock_init_class_and_name(sk, 872 af_family_slock_key_strings[sk->sk_family], 873 af_family_slock_keys + sk->sk_family, 874 af_family_key_strings[sk->sk_family], 875 af_family_keys + sk->sk_family); 876 } 877 878 static void sock_copy(struct sock *nsk, const struct sock *osk) 879 { 880 #ifdef CONFIG_SECURITY_NETWORK 881 void *sptr = nsk->sk_security; 882 #endif 883 884 memcpy(nsk, osk, osk->sk_prot->obj_size); 885 #ifdef CONFIG_SECURITY_NETWORK 886 nsk->sk_security = sptr; 887 security_sk_clone(osk, nsk); 888 #endif 889 } 890 891 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 892 int family) 893 { 894 struct sock *sk; 895 struct kmem_cache *slab; 896 897 slab = prot->slab; 898 if (slab != NULL) 899 sk = kmem_cache_alloc(slab, priority); 900 else 901 sk = kmalloc(prot->obj_size, priority); 902 903 if (sk != NULL) { 904 if (security_sk_alloc(sk, family, priority)) 905 goto out_free; 906 907 if (!try_module_get(prot->owner)) 908 goto out_free_sec; 909 } 910 911 return sk; 912 913 out_free_sec: 914 security_sk_free(sk); 915 out_free: 916 if (slab != NULL) 917 kmem_cache_free(slab, sk); 918 else 919 kfree(sk); 920 return NULL; 921 } 922 923 static void sk_prot_free(struct proto *prot, struct sock *sk) 924 { 925 struct kmem_cache *slab; 926 struct module *owner; 927 928 owner = prot->owner; 929 slab = prot->slab; 930 931 security_sk_free(sk); 932 if (slab != NULL) 933 kmem_cache_free(slab, sk); 934 else 935 kfree(sk); 936 module_put(owner); 937 } 938 939 /** 940 * sk_alloc - All socket objects are allocated here 941 * @net: the applicable net namespace 942 * @family: protocol family 943 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 944 * @prot: struct proto associated with this new sock instance 945 * @zero_it: if we should zero the newly allocated sock 946 */ 947 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 948 struct proto *prot) 949 { 950 struct sock *sk; 951 952 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 953 if (sk) { 954 sk->sk_family = family; 955 /* 956 * See comment in struct sock definition to understand 957 * why we need sk_prot_creator -acme 958 */ 959 sk->sk_prot = sk->sk_prot_creator = prot; 960 sock_lock_init(sk); 961 sock_net_set(sk, get_net(net)); 962 } 963 964 return sk; 965 } 966 967 void sk_free(struct sock *sk) 968 { 969 struct sk_filter *filter; 970 971 if (sk->sk_destruct) 972 sk->sk_destruct(sk); 973 974 filter = rcu_dereference(sk->sk_filter); 975 if (filter) { 976 sk_filter_uncharge(sk, filter); 977 rcu_assign_pointer(sk->sk_filter, NULL); 978 } 979 980 sock_disable_timestamp(sk); 981 982 if (atomic_read(&sk->sk_omem_alloc)) 983 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 984 __func__, atomic_read(&sk->sk_omem_alloc)); 985 986 put_net(sock_net(sk)); 987 sk_prot_free(sk->sk_prot_creator, sk); 988 } 989 990 /* 991 * Last sock_put should drop referrence to sk->sk_net. It has already 992 * been dropped in sk_change_net. Taking referrence to stopping namespace 993 * is not an option. 994 * Take referrence to a socket to remove it from hash _alive_ and after that 995 * destroy it in the context of init_net. 996 */ 997 void sk_release_kernel(struct sock *sk) 998 { 999 if (sk == NULL || sk->sk_socket == NULL) 1000 return; 1001 1002 sock_hold(sk); 1003 sock_release(sk->sk_socket); 1004 release_net(sock_net(sk)); 1005 sock_net_set(sk, get_net(&init_net)); 1006 sock_put(sk); 1007 } 1008 EXPORT_SYMBOL(sk_release_kernel); 1009 1010 struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 1011 { 1012 struct sock *newsk; 1013 1014 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1015 if (newsk != NULL) { 1016 struct sk_filter *filter; 1017 1018 sock_copy(newsk, sk); 1019 1020 /* SANITY */ 1021 get_net(sock_net(newsk)); 1022 sk_node_init(&newsk->sk_node); 1023 sock_lock_init(newsk); 1024 bh_lock_sock(newsk); 1025 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1026 1027 atomic_set(&newsk->sk_rmem_alloc, 0); 1028 atomic_set(&newsk->sk_wmem_alloc, 0); 1029 atomic_set(&newsk->sk_omem_alloc, 0); 1030 skb_queue_head_init(&newsk->sk_receive_queue); 1031 skb_queue_head_init(&newsk->sk_write_queue); 1032 #ifdef CONFIG_NET_DMA 1033 skb_queue_head_init(&newsk->sk_async_wait_queue); 1034 #endif 1035 1036 rwlock_init(&newsk->sk_dst_lock); 1037 rwlock_init(&newsk->sk_callback_lock); 1038 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1039 af_callback_keys + newsk->sk_family, 1040 af_family_clock_key_strings[newsk->sk_family]); 1041 1042 newsk->sk_dst_cache = NULL; 1043 newsk->sk_wmem_queued = 0; 1044 newsk->sk_forward_alloc = 0; 1045 newsk->sk_send_head = NULL; 1046 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1047 1048 sock_reset_flag(newsk, SOCK_DONE); 1049 skb_queue_head_init(&newsk->sk_error_queue); 1050 1051 filter = newsk->sk_filter; 1052 if (filter != NULL) 1053 sk_filter_charge(newsk, filter); 1054 1055 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1056 /* It is still raw copy of parent, so invalidate 1057 * destructor and make plain sk_free() */ 1058 newsk->sk_destruct = NULL; 1059 sk_free(newsk); 1060 newsk = NULL; 1061 goto out; 1062 } 1063 1064 newsk->sk_err = 0; 1065 newsk->sk_priority = 0; 1066 atomic_set(&newsk->sk_refcnt, 2); 1067 1068 /* 1069 * Increment the counter in the same struct proto as the master 1070 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1071 * is the same as sk->sk_prot->socks, as this field was copied 1072 * with memcpy). 1073 * 1074 * This _changes_ the previous behaviour, where 1075 * tcp_create_openreq_child always was incrementing the 1076 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1077 * to be taken into account in all callers. -acme 1078 */ 1079 sk_refcnt_debug_inc(newsk); 1080 newsk->sk_socket = NULL; 1081 newsk->sk_sleep = NULL; 1082 1083 if (newsk->sk_prot->sockets_allocated) 1084 atomic_inc(newsk->sk_prot->sockets_allocated); 1085 } 1086 out: 1087 return newsk; 1088 } 1089 1090 EXPORT_SYMBOL_GPL(sk_clone); 1091 1092 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1093 { 1094 __sk_dst_set(sk, dst); 1095 sk->sk_route_caps = dst->dev->features; 1096 if (sk->sk_route_caps & NETIF_F_GSO) 1097 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1098 if (sk_can_gso(sk)) { 1099 if (dst->header_len) { 1100 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1101 } else { 1102 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1103 sk->sk_gso_max_size = dst->dev->gso_max_size; 1104 } 1105 } 1106 } 1107 EXPORT_SYMBOL_GPL(sk_setup_caps); 1108 1109 void __init sk_init(void) 1110 { 1111 if (num_physpages <= 4096) { 1112 sysctl_wmem_max = 32767; 1113 sysctl_rmem_max = 32767; 1114 sysctl_wmem_default = 32767; 1115 sysctl_rmem_default = 32767; 1116 } else if (num_physpages >= 131072) { 1117 sysctl_wmem_max = 131071; 1118 sysctl_rmem_max = 131071; 1119 } 1120 } 1121 1122 /* 1123 * Simple resource managers for sockets. 1124 */ 1125 1126 1127 /* 1128 * Write buffer destructor automatically called from kfree_skb. 1129 */ 1130 void sock_wfree(struct sk_buff *skb) 1131 { 1132 struct sock *sk = skb->sk; 1133 1134 /* In case it might be waiting for more memory. */ 1135 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 1136 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 1137 sk->sk_write_space(sk); 1138 sock_put(sk); 1139 } 1140 1141 /* 1142 * Read buffer destructor automatically called from kfree_skb. 1143 */ 1144 void sock_rfree(struct sk_buff *skb) 1145 { 1146 struct sock *sk = skb->sk; 1147 1148 skb_truesize_check(skb); 1149 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1150 sk_mem_uncharge(skb->sk, skb->truesize); 1151 } 1152 1153 1154 int sock_i_uid(struct sock *sk) 1155 { 1156 int uid; 1157 1158 read_lock(&sk->sk_callback_lock); 1159 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1160 read_unlock(&sk->sk_callback_lock); 1161 return uid; 1162 } 1163 1164 unsigned long sock_i_ino(struct sock *sk) 1165 { 1166 unsigned long ino; 1167 1168 read_lock(&sk->sk_callback_lock); 1169 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1170 read_unlock(&sk->sk_callback_lock); 1171 return ino; 1172 } 1173 1174 /* 1175 * Allocate a skb from the socket's send buffer. 1176 */ 1177 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1178 gfp_t priority) 1179 { 1180 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1181 struct sk_buff * skb = alloc_skb(size, priority); 1182 if (skb) { 1183 skb_set_owner_w(skb, sk); 1184 return skb; 1185 } 1186 } 1187 return NULL; 1188 } 1189 1190 /* 1191 * Allocate a skb from the socket's receive buffer. 1192 */ 1193 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1194 gfp_t priority) 1195 { 1196 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1197 struct sk_buff *skb = alloc_skb(size, priority); 1198 if (skb) { 1199 skb_set_owner_r(skb, sk); 1200 return skb; 1201 } 1202 } 1203 return NULL; 1204 } 1205 1206 /* 1207 * Allocate a memory block from the socket's option memory buffer. 1208 */ 1209 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1210 { 1211 if ((unsigned)size <= sysctl_optmem_max && 1212 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1213 void *mem; 1214 /* First do the add, to avoid the race if kmalloc 1215 * might sleep. 1216 */ 1217 atomic_add(size, &sk->sk_omem_alloc); 1218 mem = kmalloc(size, priority); 1219 if (mem) 1220 return mem; 1221 atomic_sub(size, &sk->sk_omem_alloc); 1222 } 1223 return NULL; 1224 } 1225 1226 /* 1227 * Free an option memory block. 1228 */ 1229 void sock_kfree_s(struct sock *sk, void *mem, int size) 1230 { 1231 kfree(mem); 1232 atomic_sub(size, &sk->sk_omem_alloc); 1233 } 1234 1235 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1236 I think, these locks should be removed for datagram sockets. 1237 */ 1238 static long sock_wait_for_wmem(struct sock * sk, long timeo) 1239 { 1240 DEFINE_WAIT(wait); 1241 1242 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1243 for (;;) { 1244 if (!timeo) 1245 break; 1246 if (signal_pending(current)) 1247 break; 1248 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1249 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1250 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1251 break; 1252 if (sk->sk_shutdown & SEND_SHUTDOWN) 1253 break; 1254 if (sk->sk_err) 1255 break; 1256 timeo = schedule_timeout(timeo); 1257 } 1258 finish_wait(sk->sk_sleep, &wait); 1259 return timeo; 1260 } 1261 1262 1263 /* 1264 * Generic send/receive buffer handlers 1265 */ 1266 1267 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1268 unsigned long header_len, 1269 unsigned long data_len, 1270 int noblock, int *errcode) 1271 { 1272 struct sk_buff *skb; 1273 gfp_t gfp_mask; 1274 long timeo; 1275 int err; 1276 1277 gfp_mask = sk->sk_allocation; 1278 if (gfp_mask & __GFP_WAIT) 1279 gfp_mask |= __GFP_REPEAT; 1280 1281 timeo = sock_sndtimeo(sk, noblock); 1282 while (1) { 1283 err = sock_error(sk); 1284 if (err != 0) 1285 goto failure; 1286 1287 err = -EPIPE; 1288 if (sk->sk_shutdown & SEND_SHUTDOWN) 1289 goto failure; 1290 1291 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1292 skb = alloc_skb(header_len, gfp_mask); 1293 if (skb) { 1294 int npages; 1295 int i; 1296 1297 /* No pages, we're done... */ 1298 if (!data_len) 1299 break; 1300 1301 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1302 skb->truesize += data_len; 1303 skb_shinfo(skb)->nr_frags = npages; 1304 for (i = 0; i < npages; i++) { 1305 struct page *page; 1306 skb_frag_t *frag; 1307 1308 page = alloc_pages(sk->sk_allocation, 0); 1309 if (!page) { 1310 err = -ENOBUFS; 1311 skb_shinfo(skb)->nr_frags = i; 1312 kfree_skb(skb); 1313 goto failure; 1314 } 1315 1316 frag = &skb_shinfo(skb)->frags[i]; 1317 frag->page = page; 1318 frag->page_offset = 0; 1319 frag->size = (data_len >= PAGE_SIZE ? 1320 PAGE_SIZE : 1321 data_len); 1322 data_len -= PAGE_SIZE; 1323 } 1324 1325 /* Full success... */ 1326 break; 1327 } 1328 err = -ENOBUFS; 1329 goto failure; 1330 } 1331 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1332 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1333 err = -EAGAIN; 1334 if (!timeo) 1335 goto failure; 1336 if (signal_pending(current)) 1337 goto interrupted; 1338 timeo = sock_wait_for_wmem(sk, timeo); 1339 } 1340 1341 skb_set_owner_w(skb, sk); 1342 return skb; 1343 1344 interrupted: 1345 err = sock_intr_errno(timeo); 1346 failure: 1347 *errcode = err; 1348 return NULL; 1349 } 1350 1351 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1352 int noblock, int *errcode) 1353 { 1354 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1355 } 1356 1357 static void __lock_sock(struct sock *sk) 1358 { 1359 DEFINE_WAIT(wait); 1360 1361 for (;;) { 1362 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1363 TASK_UNINTERRUPTIBLE); 1364 spin_unlock_bh(&sk->sk_lock.slock); 1365 schedule(); 1366 spin_lock_bh(&sk->sk_lock.slock); 1367 if (!sock_owned_by_user(sk)) 1368 break; 1369 } 1370 finish_wait(&sk->sk_lock.wq, &wait); 1371 } 1372 1373 static void __release_sock(struct sock *sk) 1374 { 1375 struct sk_buff *skb = sk->sk_backlog.head; 1376 1377 do { 1378 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1379 bh_unlock_sock(sk); 1380 1381 do { 1382 struct sk_buff *next = skb->next; 1383 1384 skb->next = NULL; 1385 sk->sk_backlog_rcv(sk, skb); 1386 1387 /* 1388 * We are in process context here with softirqs 1389 * disabled, use cond_resched_softirq() to preempt. 1390 * This is safe to do because we've taken the backlog 1391 * queue private: 1392 */ 1393 cond_resched_softirq(); 1394 1395 skb = next; 1396 } while (skb != NULL); 1397 1398 bh_lock_sock(sk); 1399 } while ((skb = sk->sk_backlog.head) != NULL); 1400 } 1401 1402 /** 1403 * sk_wait_data - wait for data to arrive at sk_receive_queue 1404 * @sk: sock to wait on 1405 * @timeo: for how long 1406 * 1407 * Now socket state including sk->sk_err is changed only under lock, 1408 * hence we may omit checks after joining wait queue. 1409 * We check receive queue before schedule() only as optimization; 1410 * it is very likely that release_sock() added new data. 1411 */ 1412 int sk_wait_data(struct sock *sk, long *timeo) 1413 { 1414 int rc; 1415 DEFINE_WAIT(wait); 1416 1417 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1418 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1419 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1420 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1421 finish_wait(sk->sk_sleep, &wait); 1422 return rc; 1423 } 1424 1425 EXPORT_SYMBOL(sk_wait_data); 1426 1427 /** 1428 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1429 * @sk: socket 1430 * @size: memory size to allocate 1431 * @kind: allocation type 1432 * 1433 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1434 * rmem allocation. This function assumes that protocols which have 1435 * memory_pressure use sk_wmem_queued as write buffer accounting. 1436 */ 1437 int __sk_mem_schedule(struct sock *sk, int size, int kind) 1438 { 1439 struct proto *prot = sk->sk_prot; 1440 int amt = sk_mem_pages(size); 1441 int allocated; 1442 1443 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1444 allocated = atomic_add_return(amt, prot->memory_allocated); 1445 1446 /* Under limit. */ 1447 if (allocated <= prot->sysctl_mem[0]) { 1448 if (prot->memory_pressure && *prot->memory_pressure) 1449 *prot->memory_pressure = 0; 1450 return 1; 1451 } 1452 1453 /* Under pressure. */ 1454 if (allocated > prot->sysctl_mem[1]) 1455 if (prot->enter_memory_pressure) 1456 prot->enter_memory_pressure(); 1457 1458 /* Over hard limit. */ 1459 if (allocated > prot->sysctl_mem[2]) 1460 goto suppress_allocation; 1461 1462 /* guarantee minimum buffer size under pressure */ 1463 if (kind == SK_MEM_RECV) { 1464 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1465 return 1; 1466 } else { /* SK_MEM_SEND */ 1467 if (sk->sk_type == SOCK_STREAM) { 1468 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 1469 return 1; 1470 } else if (atomic_read(&sk->sk_wmem_alloc) < 1471 prot->sysctl_wmem[0]) 1472 return 1; 1473 } 1474 1475 if (prot->memory_pressure) { 1476 if (!*prot->memory_pressure || 1477 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) * 1478 sk_mem_pages(sk->sk_wmem_queued + 1479 atomic_read(&sk->sk_rmem_alloc) + 1480 sk->sk_forward_alloc)) 1481 return 1; 1482 } 1483 1484 suppress_allocation: 1485 1486 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 1487 sk_stream_moderate_sndbuf(sk); 1488 1489 /* Fail only if socket is _under_ its sndbuf. 1490 * In this case we cannot block, so that we have to fail. 1491 */ 1492 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 1493 return 1; 1494 } 1495 1496 /* Alas. Undo changes. */ 1497 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 1498 atomic_sub(amt, prot->memory_allocated); 1499 return 0; 1500 } 1501 1502 EXPORT_SYMBOL(__sk_mem_schedule); 1503 1504 /** 1505 * __sk_reclaim - reclaim memory_allocated 1506 * @sk: socket 1507 */ 1508 void __sk_mem_reclaim(struct sock *sk) 1509 { 1510 struct proto *prot = sk->sk_prot; 1511 1512 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT, 1513 prot->memory_allocated); 1514 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 1515 1516 if (prot->memory_pressure && *prot->memory_pressure && 1517 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0])) 1518 *prot->memory_pressure = 0; 1519 } 1520 1521 EXPORT_SYMBOL(__sk_mem_reclaim); 1522 1523 1524 /* 1525 * Set of default routines for initialising struct proto_ops when 1526 * the protocol does not support a particular function. In certain 1527 * cases where it makes no sense for a protocol to have a "do nothing" 1528 * function, some default processing is provided. 1529 */ 1530 1531 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1532 { 1533 return -EOPNOTSUPP; 1534 } 1535 1536 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1537 int len, int flags) 1538 { 1539 return -EOPNOTSUPP; 1540 } 1541 1542 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1543 { 1544 return -EOPNOTSUPP; 1545 } 1546 1547 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1548 { 1549 return -EOPNOTSUPP; 1550 } 1551 1552 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1553 int *len, int peer) 1554 { 1555 return -EOPNOTSUPP; 1556 } 1557 1558 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1559 { 1560 return 0; 1561 } 1562 1563 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1564 { 1565 return -EOPNOTSUPP; 1566 } 1567 1568 int sock_no_listen(struct socket *sock, int backlog) 1569 { 1570 return -EOPNOTSUPP; 1571 } 1572 1573 int sock_no_shutdown(struct socket *sock, int how) 1574 { 1575 return -EOPNOTSUPP; 1576 } 1577 1578 int sock_no_setsockopt(struct socket *sock, int level, int optname, 1579 char __user *optval, int optlen) 1580 { 1581 return -EOPNOTSUPP; 1582 } 1583 1584 int sock_no_getsockopt(struct socket *sock, int level, int optname, 1585 char __user *optval, int __user *optlen) 1586 { 1587 return -EOPNOTSUPP; 1588 } 1589 1590 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1591 size_t len) 1592 { 1593 return -EOPNOTSUPP; 1594 } 1595 1596 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1597 size_t len, int flags) 1598 { 1599 return -EOPNOTSUPP; 1600 } 1601 1602 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1603 { 1604 /* Mirror missing mmap method error code */ 1605 return -ENODEV; 1606 } 1607 1608 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1609 { 1610 ssize_t res; 1611 struct msghdr msg = {.msg_flags = flags}; 1612 struct kvec iov; 1613 char *kaddr = kmap(page); 1614 iov.iov_base = kaddr + offset; 1615 iov.iov_len = size; 1616 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1617 kunmap(page); 1618 return res; 1619 } 1620 1621 /* 1622 * Default Socket Callbacks 1623 */ 1624 1625 static void sock_def_wakeup(struct sock *sk) 1626 { 1627 read_lock(&sk->sk_callback_lock); 1628 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1629 wake_up_interruptible_all(sk->sk_sleep); 1630 read_unlock(&sk->sk_callback_lock); 1631 } 1632 1633 static void sock_def_error_report(struct sock *sk) 1634 { 1635 read_lock(&sk->sk_callback_lock); 1636 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1637 wake_up_interruptible(sk->sk_sleep); 1638 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 1639 read_unlock(&sk->sk_callback_lock); 1640 } 1641 1642 static void sock_def_readable(struct sock *sk, int len) 1643 { 1644 read_lock(&sk->sk_callback_lock); 1645 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1646 wake_up_interruptible_sync(sk->sk_sleep); 1647 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 1648 read_unlock(&sk->sk_callback_lock); 1649 } 1650 1651 static void sock_def_write_space(struct sock *sk) 1652 { 1653 read_lock(&sk->sk_callback_lock); 1654 1655 /* Do not wake up a writer until he can make "significant" 1656 * progress. --DaveM 1657 */ 1658 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1659 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1660 wake_up_interruptible_sync(sk->sk_sleep); 1661 1662 /* Should agree with poll, otherwise some programs break */ 1663 if (sock_writeable(sk)) 1664 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 1665 } 1666 1667 read_unlock(&sk->sk_callback_lock); 1668 } 1669 1670 static void sock_def_destruct(struct sock *sk) 1671 { 1672 kfree(sk->sk_protinfo); 1673 } 1674 1675 void sk_send_sigurg(struct sock *sk) 1676 { 1677 if (sk->sk_socket && sk->sk_socket->file) 1678 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1679 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 1680 } 1681 1682 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1683 unsigned long expires) 1684 { 1685 if (!mod_timer(timer, expires)) 1686 sock_hold(sk); 1687 } 1688 1689 EXPORT_SYMBOL(sk_reset_timer); 1690 1691 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1692 { 1693 if (timer_pending(timer) && del_timer(timer)) 1694 __sock_put(sk); 1695 } 1696 1697 EXPORT_SYMBOL(sk_stop_timer); 1698 1699 void sock_init_data(struct socket *sock, struct sock *sk) 1700 { 1701 skb_queue_head_init(&sk->sk_receive_queue); 1702 skb_queue_head_init(&sk->sk_write_queue); 1703 skb_queue_head_init(&sk->sk_error_queue); 1704 #ifdef CONFIG_NET_DMA 1705 skb_queue_head_init(&sk->sk_async_wait_queue); 1706 #endif 1707 1708 sk->sk_send_head = NULL; 1709 1710 init_timer(&sk->sk_timer); 1711 1712 sk->sk_allocation = GFP_KERNEL; 1713 sk->sk_rcvbuf = sysctl_rmem_default; 1714 sk->sk_sndbuf = sysctl_wmem_default; 1715 sk->sk_state = TCP_CLOSE; 1716 sk->sk_socket = sock; 1717 1718 sock_set_flag(sk, SOCK_ZAPPED); 1719 1720 if (sock) { 1721 sk->sk_type = sock->type; 1722 sk->sk_sleep = &sock->wait; 1723 sock->sk = sk; 1724 } else 1725 sk->sk_sleep = NULL; 1726 1727 rwlock_init(&sk->sk_dst_lock); 1728 rwlock_init(&sk->sk_callback_lock); 1729 lockdep_set_class_and_name(&sk->sk_callback_lock, 1730 af_callback_keys + sk->sk_family, 1731 af_family_clock_key_strings[sk->sk_family]); 1732 1733 sk->sk_state_change = sock_def_wakeup; 1734 sk->sk_data_ready = sock_def_readable; 1735 sk->sk_write_space = sock_def_write_space; 1736 sk->sk_error_report = sock_def_error_report; 1737 sk->sk_destruct = sock_def_destruct; 1738 1739 sk->sk_sndmsg_page = NULL; 1740 sk->sk_sndmsg_off = 0; 1741 1742 sk->sk_peercred.pid = 0; 1743 sk->sk_peercred.uid = -1; 1744 sk->sk_peercred.gid = -1; 1745 sk->sk_write_pending = 0; 1746 sk->sk_rcvlowat = 1; 1747 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1748 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1749 1750 sk->sk_stamp = ktime_set(-1L, 0); 1751 1752 atomic_set(&sk->sk_refcnt, 1); 1753 atomic_set(&sk->sk_drops, 0); 1754 } 1755 1756 void lock_sock_nested(struct sock *sk, int subclass) 1757 { 1758 might_sleep(); 1759 spin_lock_bh(&sk->sk_lock.slock); 1760 if (sk->sk_lock.owned) 1761 __lock_sock(sk); 1762 sk->sk_lock.owned = 1; 1763 spin_unlock(&sk->sk_lock.slock); 1764 /* 1765 * The sk_lock has mutex_lock() semantics here: 1766 */ 1767 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1768 local_bh_enable(); 1769 } 1770 1771 EXPORT_SYMBOL(lock_sock_nested); 1772 1773 void release_sock(struct sock *sk) 1774 { 1775 /* 1776 * The sk_lock has mutex_unlock() semantics: 1777 */ 1778 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1779 1780 spin_lock_bh(&sk->sk_lock.slock); 1781 if (sk->sk_backlog.tail) 1782 __release_sock(sk); 1783 sk->sk_lock.owned = 0; 1784 if (waitqueue_active(&sk->sk_lock.wq)) 1785 wake_up(&sk->sk_lock.wq); 1786 spin_unlock_bh(&sk->sk_lock.slock); 1787 } 1788 EXPORT_SYMBOL(release_sock); 1789 1790 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1791 { 1792 struct timeval tv; 1793 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1794 sock_enable_timestamp(sk); 1795 tv = ktime_to_timeval(sk->sk_stamp); 1796 if (tv.tv_sec == -1) 1797 return -ENOENT; 1798 if (tv.tv_sec == 0) { 1799 sk->sk_stamp = ktime_get_real(); 1800 tv = ktime_to_timeval(sk->sk_stamp); 1801 } 1802 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 1803 } 1804 EXPORT_SYMBOL(sock_get_timestamp); 1805 1806 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 1807 { 1808 struct timespec ts; 1809 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1810 sock_enable_timestamp(sk); 1811 ts = ktime_to_timespec(sk->sk_stamp); 1812 if (ts.tv_sec == -1) 1813 return -ENOENT; 1814 if (ts.tv_sec == 0) { 1815 sk->sk_stamp = ktime_get_real(); 1816 ts = ktime_to_timespec(sk->sk_stamp); 1817 } 1818 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 1819 } 1820 EXPORT_SYMBOL(sock_get_timestampns); 1821 1822 void sock_enable_timestamp(struct sock *sk) 1823 { 1824 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1825 sock_set_flag(sk, SOCK_TIMESTAMP); 1826 net_enable_timestamp(); 1827 } 1828 } 1829 1830 /* 1831 * Get a socket option on an socket. 1832 * 1833 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1834 * asynchronous errors should be reported by getsockopt. We assume 1835 * this means if you specify SO_ERROR (otherwise whats the point of it). 1836 */ 1837 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1838 char __user *optval, int __user *optlen) 1839 { 1840 struct sock *sk = sock->sk; 1841 1842 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1843 } 1844 1845 EXPORT_SYMBOL(sock_common_getsockopt); 1846 1847 #ifdef CONFIG_COMPAT 1848 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1849 char __user *optval, int __user *optlen) 1850 { 1851 struct sock *sk = sock->sk; 1852 1853 if (sk->sk_prot->compat_getsockopt != NULL) 1854 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1855 optval, optlen); 1856 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1857 } 1858 EXPORT_SYMBOL(compat_sock_common_getsockopt); 1859 #endif 1860 1861 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1862 struct msghdr *msg, size_t size, int flags) 1863 { 1864 struct sock *sk = sock->sk; 1865 int addr_len = 0; 1866 int err; 1867 1868 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1869 flags & ~MSG_DONTWAIT, &addr_len); 1870 if (err >= 0) 1871 msg->msg_namelen = addr_len; 1872 return err; 1873 } 1874 1875 EXPORT_SYMBOL(sock_common_recvmsg); 1876 1877 /* 1878 * Set socket options on an inet socket. 1879 */ 1880 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1881 char __user *optval, int optlen) 1882 { 1883 struct sock *sk = sock->sk; 1884 1885 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1886 } 1887 1888 EXPORT_SYMBOL(sock_common_setsockopt); 1889 1890 #ifdef CONFIG_COMPAT 1891 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1892 char __user *optval, int optlen) 1893 { 1894 struct sock *sk = sock->sk; 1895 1896 if (sk->sk_prot->compat_setsockopt != NULL) 1897 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1898 optval, optlen); 1899 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1900 } 1901 EXPORT_SYMBOL(compat_sock_common_setsockopt); 1902 #endif 1903 1904 void sk_common_release(struct sock *sk) 1905 { 1906 if (sk->sk_prot->destroy) 1907 sk->sk_prot->destroy(sk); 1908 1909 /* 1910 * Observation: when sock_common_release is called, processes have 1911 * no access to socket. But net still has. 1912 * Step one, detach it from networking: 1913 * 1914 * A. Remove from hash tables. 1915 */ 1916 1917 sk->sk_prot->unhash(sk); 1918 1919 /* 1920 * In this point socket cannot receive new packets, but it is possible 1921 * that some packets are in flight because some CPU runs receiver and 1922 * did hash table lookup before we unhashed socket. They will achieve 1923 * receive queue and will be purged by socket destructor. 1924 * 1925 * Also we still have packets pending on receive queue and probably, 1926 * our own packets waiting in device queues. sock_destroy will drain 1927 * receive queue, but transmitted packets will delay socket destruction 1928 * until the last reference will be released. 1929 */ 1930 1931 sock_orphan(sk); 1932 1933 xfrm_sk_free_policy(sk); 1934 1935 sk_refcnt_debug_release(sk); 1936 sock_put(sk); 1937 } 1938 1939 EXPORT_SYMBOL(sk_common_release); 1940 1941 static DEFINE_RWLOCK(proto_list_lock); 1942 static LIST_HEAD(proto_list); 1943 1944 #ifdef CONFIG_PROC_FS 1945 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1946 struct prot_inuse { 1947 int val[PROTO_INUSE_NR]; 1948 }; 1949 1950 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 1951 1952 #ifdef CONFIG_NET_NS 1953 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 1954 { 1955 int cpu = smp_processor_id(); 1956 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val; 1957 } 1958 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 1959 1960 int sock_prot_inuse_get(struct net *net, struct proto *prot) 1961 { 1962 int cpu, idx = prot->inuse_idx; 1963 int res = 0; 1964 1965 for_each_possible_cpu(cpu) 1966 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 1967 1968 return res >= 0 ? res : 0; 1969 } 1970 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 1971 1972 static int sock_inuse_init_net(struct net *net) 1973 { 1974 net->core.inuse = alloc_percpu(struct prot_inuse); 1975 return net->core.inuse ? 0 : -ENOMEM; 1976 } 1977 1978 static void sock_inuse_exit_net(struct net *net) 1979 { 1980 free_percpu(net->core.inuse); 1981 } 1982 1983 static struct pernet_operations net_inuse_ops = { 1984 .init = sock_inuse_init_net, 1985 .exit = sock_inuse_exit_net, 1986 }; 1987 1988 static __init int net_inuse_init(void) 1989 { 1990 if (register_pernet_subsys(&net_inuse_ops)) 1991 panic("Cannot initialize net inuse counters"); 1992 1993 return 0; 1994 } 1995 1996 core_initcall(net_inuse_init); 1997 #else 1998 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 1999 2000 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2001 { 2002 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val; 2003 } 2004 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2005 2006 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2007 { 2008 int cpu, idx = prot->inuse_idx; 2009 int res = 0; 2010 2011 for_each_possible_cpu(cpu) 2012 res += per_cpu(prot_inuse, cpu).val[idx]; 2013 2014 return res >= 0 ? res : 0; 2015 } 2016 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2017 #endif 2018 2019 static void assign_proto_idx(struct proto *prot) 2020 { 2021 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2022 2023 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2024 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n"); 2025 return; 2026 } 2027 2028 set_bit(prot->inuse_idx, proto_inuse_idx); 2029 } 2030 2031 static void release_proto_idx(struct proto *prot) 2032 { 2033 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2034 clear_bit(prot->inuse_idx, proto_inuse_idx); 2035 } 2036 #else 2037 static inline void assign_proto_idx(struct proto *prot) 2038 { 2039 } 2040 2041 static inline void release_proto_idx(struct proto *prot) 2042 { 2043 } 2044 #endif 2045 2046 int proto_register(struct proto *prot, int alloc_slab) 2047 { 2048 char *request_sock_slab_name = NULL; 2049 char *timewait_sock_slab_name; 2050 2051 if (alloc_slab) { 2052 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2053 SLAB_HWCACHE_ALIGN, NULL); 2054 2055 if (prot->slab == NULL) { 2056 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 2057 prot->name); 2058 goto out; 2059 } 2060 2061 if (prot->rsk_prot != NULL) { 2062 static const char mask[] = "request_sock_%s"; 2063 2064 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 2065 if (request_sock_slab_name == NULL) 2066 goto out_free_sock_slab; 2067 2068 sprintf(request_sock_slab_name, mask, prot->name); 2069 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 2070 prot->rsk_prot->obj_size, 0, 2071 SLAB_HWCACHE_ALIGN, NULL); 2072 2073 if (prot->rsk_prot->slab == NULL) { 2074 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 2075 prot->name); 2076 goto out_free_request_sock_slab_name; 2077 } 2078 } 2079 2080 if (prot->twsk_prot != NULL) { 2081 static const char mask[] = "tw_sock_%s"; 2082 2083 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 2084 2085 if (timewait_sock_slab_name == NULL) 2086 goto out_free_request_sock_slab; 2087 2088 sprintf(timewait_sock_slab_name, mask, prot->name); 2089 prot->twsk_prot->twsk_slab = 2090 kmem_cache_create(timewait_sock_slab_name, 2091 prot->twsk_prot->twsk_obj_size, 2092 0, SLAB_HWCACHE_ALIGN, 2093 NULL); 2094 if (prot->twsk_prot->twsk_slab == NULL) 2095 goto out_free_timewait_sock_slab_name; 2096 } 2097 } 2098 2099 write_lock(&proto_list_lock); 2100 list_add(&prot->node, &proto_list); 2101 assign_proto_idx(prot); 2102 write_unlock(&proto_list_lock); 2103 return 0; 2104 2105 out_free_timewait_sock_slab_name: 2106 kfree(timewait_sock_slab_name); 2107 out_free_request_sock_slab: 2108 if (prot->rsk_prot && prot->rsk_prot->slab) { 2109 kmem_cache_destroy(prot->rsk_prot->slab); 2110 prot->rsk_prot->slab = NULL; 2111 } 2112 out_free_request_sock_slab_name: 2113 kfree(request_sock_slab_name); 2114 out_free_sock_slab: 2115 kmem_cache_destroy(prot->slab); 2116 prot->slab = NULL; 2117 out: 2118 return -ENOBUFS; 2119 } 2120 2121 EXPORT_SYMBOL(proto_register); 2122 2123 void proto_unregister(struct proto *prot) 2124 { 2125 write_lock(&proto_list_lock); 2126 release_proto_idx(prot); 2127 list_del(&prot->node); 2128 write_unlock(&proto_list_lock); 2129 2130 if (prot->slab != NULL) { 2131 kmem_cache_destroy(prot->slab); 2132 prot->slab = NULL; 2133 } 2134 2135 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2136 const char *name = kmem_cache_name(prot->rsk_prot->slab); 2137 2138 kmem_cache_destroy(prot->rsk_prot->slab); 2139 kfree(name); 2140 prot->rsk_prot->slab = NULL; 2141 } 2142 2143 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2144 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 2145 2146 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2147 kfree(name); 2148 prot->twsk_prot->twsk_slab = NULL; 2149 } 2150 } 2151 2152 EXPORT_SYMBOL(proto_unregister); 2153 2154 #ifdef CONFIG_PROC_FS 2155 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2156 __acquires(proto_list_lock) 2157 { 2158 read_lock(&proto_list_lock); 2159 return seq_list_start_head(&proto_list, *pos); 2160 } 2161 2162 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2163 { 2164 return seq_list_next(v, &proto_list, pos); 2165 } 2166 2167 static void proto_seq_stop(struct seq_file *seq, void *v) 2168 __releases(proto_list_lock) 2169 { 2170 read_unlock(&proto_list_lock); 2171 } 2172 2173 static char proto_method_implemented(const void *method) 2174 { 2175 return method == NULL ? 'n' : 'y'; 2176 } 2177 2178 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2179 { 2180 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 2181 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2182 proto->name, 2183 proto->obj_size, 2184 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 2185 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 2186 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 2187 proto->max_header, 2188 proto->slab == NULL ? "no" : "yes", 2189 module_name(proto->owner), 2190 proto_method_implemented(proto->close), 2191 proto_method_implemented(proto->connect), 2192 proto_method_implemented(proto->disconnect), 2193 proto_method_implemented(proto->accept), 2194 proto_method_implemented(proto->ioctl), 2195 proto_method_implemented(proto->init), 2196 proto_method_implemented(proto->destroy), 2197 proto_method_implemented(proto->shutdown), 2198 proto_method_implemented(proto->setsockopt), 2199 proto_method_implemented(proto->getsockopt), 2200 proto_method_implemented(proto->sendmsg), 2201 proto_method_implemented(proto->recvmsg), 2202 proto_method_implemented(proto->sendpage), 2203 proto_method_implemented(proto->bind), 2204 proto_method_implemented(proto->backlog_rcv), 2205 proto_method_implemented(proto->hash), 2206 proto_method_implemented(proto->unhash), 2207 proto_method_implemented(proto->get_port), 2208 proto_method_implemented(proto->enter_memory_pressure)); 2209 } 2210 2211 static int proto_seq_show(struct seq_file *seq, void *v) 2212 { 2213 if (v == &proto_list) 2214 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2215 "protocol", 2216 "size", 2217 "sockets", 2218 "memory", 2219 "press", 2220 "maxhdr", 2221 "slab", 2222 "module", 2223 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2224 else 2225 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2226 return 0; 2227 } 2228 2229 static const struct seq_operations proto_seq_ops = { 2230 .start = proto_seq_start, 2231 .next = proto_seq_next, 2232 .stop = proto_seq_stop, 2233 .show = proto_seq_show, 2234 }; 2235 2236 static int proto_seq_open(struct inode *inode, struct file *file) 2237 { 2238 return seq_open(file, &proto_seq_ops); 2239 } 2240 2241 static const struct file_operations proto_seq_fops = { 2242 .owner = THIS_MODULE, 2243 .open = proto_seq_open, 2244 .read = seq_read, 2245 .llseek = seq_lseek, 2246 .release = seq_release, 2247 }; 2248 2249 static int __init proto_init(void) 2250 { 2251 /* register /proc/net/protocols */ 2252 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 2253 } 2254 2255 subsys_initcall(proto_init); 2256 2257 #endif /* PROC_FS */ 2258 2259 EXPORT_SYMBOL(sk_alloc); 2260 EXPORT_SYMBOL(sk_free); 2261 EXPORT_SYMBOL(sk_send_sigurg); 2262 EXPORT_SYMBOL(sock_alloc_send_skb); 2263 EXPORT_SYMBOL(sock_init_data); 2264 EXPORT_SYMBOL(sock_kfree_s); 2265 EXPORT_SYMBOL(sock_kmalloc); 2266 EXPORT_SYMBOL(sock_no_accept); 2267 EXPORT_SYMBOL(sock_no_bind); 2268 EXPORT_SYMBOL(sock_no_connect); 2269 EXPORT_SYMBOL(sock_no_getname); 2270 EXPORT_SYMBOL(sock_no_getsockopt); 2271 EXPORT_SYMBOL(sock_no_ioctl); 2272 EXPORT_SYMBOL(sock_no_listen); 2273 EXPORT_SYMBOL(sock_no_mmap); 2274 EXPORT_SYMBOL(sock_no_poll); 2275 EXPORT_SYMBOL(sock_no_recvmsg); 2276 EXPORT_SYMBOL(sock_no_sendmsg); 2277 EXPORT_SYMBOL(sock_no_sendpage); 2278 EXPORT_SYMBOL(sock_no_setsockopt); 2279 EXPORT_SYMBOL(sock_no_shutdown); 2280 EXPORT_SYMBOL(sock_no_socketpair); 2281 EXPORT_SYMBOL(sock_rfree); 2282 EXPORT_SYMBOL(sock_setsockopt); 2283 EXPORT_SYMBOL(sock_wfree); 2284 EXPORT_SYMBOL(sock_wmalloc); 2285 EXPORT_SYMBOL(sock_i_uid); 2286 EXPORT_SYMBOL(sock_i_ino); 2287 EXPORT_SYMBOL(sysctl_optmem_max); 2288