Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later
2 : /*
3 : * INET An implementation of the TCP/IP protocol suite for the LINUX
4 : * operating system. INET is implemented using the BSD Socket
5 : * interface as the means of communication with the user level.
6 : *
7 : * Generic socket support routines. Memory allocators, socket lock/release
8 : * handler for protocols to use and generic option handler.
9 : *
10 : * Authors: Ross Biro
11 : * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 : * Florian La Roche, <flla@stud.uni-sb.de>
13 : * Alan Cox, <A.Cox@swansea.ac.uk>
14 : *
15 : * Fixes:
16 : * Alan Cox : Numerous verify_area() problems
17 : * Alan Cox : Connecting on a connecting socket
18 : * now returns an error for tcp.
19 : * Alan Cox : sock->protocol is set correctly.
20 : * and is not sometimes left as 0.
21 : * Alan Cox : connect handles icmp errors on a
22 : * connect properly. Unfortunately there
23 : * is a restart syscall nasty there. I
24 : * can't match BSD without hacking the C
25 : * library. Ideas urgently sought!
26 : * Alan Cox : Disallow bind() to addresses that are
27 : * not ours - especially broadcast ones!!
28 : * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 : * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 : * instead they leave that for the DESTROY timer.
31 : * Alan Cox : Clean up error flag in accept
32 : * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 : * was buggy. Put a remove_sock() in the handler
34 : * for memory when we hit 0. Also altered the timer
35 : * code. The ACK stuff can wait and needs major
36 : * TCP layer surgery.
37 : * Alan Cox : Fixed TCP ack bug, removed remove sock
38 : * and fixed timer/inet_bh race.
39 : * Alan Cox : Added zapped flag for TCP
40 : * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 : * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 : * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 : * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 : * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 : * Rick Sladkey : Relaxed UDP rules for matching packets.
46 : * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 : * Pauline Middelink : identd support
48 : * Alan Cox : Fixed connect() taking signals I think.
49 : * Alan Cox : SO_LINGER supported
50 : * Alan Cox : Error reporting fixes
51 : * Anonymous : inet_create tidied up (sk->reuse setting)
52 : * Alan Cox : inet sockets don't set sk->type!
53 : * Alan Cox : Split socket option code
54 : * Alan Cox : Callbacks
55 : * Alan Cox : Nagle flag for Charles & Johannes stuff
56 : * Alex : Removed restriction on inet fioctl
57 : * Alan Cox : Splitting INET from NET core
58 : * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 : * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 : * Alan Cox : Split IP from generic code
61 : * Alan Cox : New kfree_skbmem()
62 : * Alan Cox : Make SO_DEBUG superuser only.
63 : * Alan Cox : Allow anyone to clear SO_DEBUG
64 : * (compatibility fix)
65 : * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 : * Alan Cox : Allocator for a socket is settable.
67 : * Alan Cox : SO_ERROR includes soft errors.
68 : * Alan Cox : Allow NULL arguments on some SO_ opts
69 : * Alan Cox : Generic socket allocation to make hooks
70 : * easier (suggested by Craig Metz).
71 : * Michael Pall : SO_ERROR returns positive errno again
72 : * Steve Whitehouse: Added default destructor to free
73 : * protocol private data.
74 : * Steve Whitehouse: Added various other default routines
75 : * common to several socket families.
76 : * Chris Evans : Call suser() check last on F_SETOWN
77 : * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 : * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 : * Andi Kleen : Fix write_space callback
80 : * Chris Evans : Security fixes - signedness again
81 : * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 : *
83 : * To Fix:
84 : */
85 :
86 : #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87 :
88 : #include <asm/unaligned.h>
89 : #include <linux/capability.h>
90 : #include <linux/errno.h>
91 : #include <linux/errqueue.h>
92 : #include <linux/types.h>
93 : #include <linux/socket.h>
94 : #include <linux/in.h>
95 : #include <linux/kernel.h>
96 : #include <linux/module.h>
97 : #include <linux/proc_fs.h>
98 : #include <linux/seq_file.h>
99 : #include <linux/sched.h>
100 : #include <linux/sched/mm.h>
101 : #include <linux/timer.h>
102 : #include <linux/string.h>
103 : #include <linux/sockios.h>
104 : #include <linux/net.h>
105 : #include <linux/mm.h>
106 : #include <linux/slab.h>
107 : #include <linux/interrupt.h>
108 : #include <linux/poll.h>
109 : #include <linux/tcp.h>
110 : #include <linux/init.h>
111 : #include <linux/highmem.h>
112 : #include <linux/user_namespace.h>
113 : #include <linux/static_key.h>
114 : #include <linux/memcontrol.h>
115 : #include <linux/prefetch.h>
116 : #include <linux/compat.h>
117 :
118 : #include <linux/uaccess.h>
119 :
120 : #include <linux/netdevice.h>
121 : #include <net/protocol.h>
122 : #include <linux/skbuff.h>
123 : #include <net/net_namespace.h>
124 : #include <net/request_sock.h>
125 : #include <net/sock.h>
126 : #include <linux/net_tstamp.h>
127 : #include <net/xfrm.h>
128 : #include <linux/ipsec.h>
129 : #include <net/cls_cgroup.h>
130 : #include <net/netprio_cgroup.h>
131 : #include <linux/sock_diag.h>
132 :
133 : #include <linux/filter.h>
134 : #include <net/sock_reuseport.h>
135 : #include <net/bpf_sk_storage.h>
136 :
137 : #include <trace/events/sock.h>
138 :
139 : #include <net/tcp.h>
140 : #include <net/busy_poll.h>
141 :
142 : static DEFINE_MUTEX(proto_list_mutex);
143 : static LIST_HEAD(proto_list);
144 :
145 : static void sock_inuse_add(struct net *net, int val);
146 :
147 : /**
148 : * sk_ns_capable - General socket capability test
149 : * @sk: Socket to use a capability on or through
150 : * @user_ns: The user namespace of the capability to use
151 : * @cap: The capability to use
152 : *
153 : * Test to see if the opener of the socket had when the socket was
154 : * created and the current process has the capability @cap in the user
155 : * namespace @user_ns.
156 : */
157 0 : bool sk_ns_capable(const struct sock *sk,
158 : struct user_namespace *user_ns, int cap)
159 : {
160 0 : return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 0 : ns_capable(user_ns, cap);
162 : }
163 : EXPORT_SYMBOL(sk_ns_capable);
164 :
165 : /**
166 : * sk_capable - Socket global capability test
167 : * @sk: Socket to use a capability on or through
168 : * @cap: The global capability to use
169 : *
170 : * Test to see if the opener of the socket had when the socket was
171 : * created and the current process has the capability @cap in all user
172 : * namespaces.
173 : */
174 0 : bool sk_capable(const struct sock *sk, int cap)
175 : {
176 0 : return sk_ns_capable(sk, &init_user_ns, cap);
177 : }
178 : EXPORT_SYMBOL(sk_capable);
179 :
180 : /**
181 : * sk_net_capable - Network namespace socket capability test
182 : * @sk: Socket to use a capability on or through
183 : * @cap: The capability to use
184 : *
185 : * Test to see if the opener of the socket had when the socket was created
186 : * and the current process has the capability @cap over the network namespace
187 : * the socket is a member of.
188 : */
189 0 : bool sk_net_capable(const struct sock *sk, int cap)
190 : {
191 0 : return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
192 : }
193 : EXPORT_SYMBOL(sk_net_capable);
194 :
195 : /*
196 : * Each address family might have different locking rules, so we have
197 : * one slock key per address family and separate keys for internal and
198 : * userspace sockets.
199 : */
200 : static struct lock_class_key af_family_keys[AF_MAX];
201 : static struct lock_class_key af_family_kern_keys[AF_MAX];
202 : static struct lock_class_key af_family_slock_keys[AF_MAX];
203 : static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
204 :
205 : /*
206 : * Make lock validator output more readable. (we pre-construct these
207 : * strings build-time, so that runtime initialization of socket
208 : * locks is fast):
209 : */
210 :
211 : #define _sock_locks(x) \
212 : x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 : x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 : x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 : x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 : x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 : x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 : x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 : x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 : x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 : x "27" , x "28" , x "AF_CAN" , \
222 : x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 : x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 : x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 : x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 : x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
227 : x "AF_MAX"
228 :
229 : static const char *const af_family_key_strings[AF_MAX+1] = {
230 : _sock_locks("sk_lock-")
231 : };
232 : static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 : _sock_locks("slock-")
234 : };
235 : static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 : _sock_locks("clock-")
237 : };
238 :
239 : static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 : _sock_locks("k-sk_lock-")
241 : };
242 : static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 : _sock_locks("k-slock-")
244 : };
245 : static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 : _sock_locks("k-clock-")
247 : };
248 : static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 : _sock_locks("rlock-")
250 : };
251 : static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 : _sock_locks("wlock-")
253 : };
254 : static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 : _sock_locks("elock-")
256 : };
257 :
258 : /*
259 : * sk_callback_lock and sk queues locking rules are per-address-family,
260 : * so split the lock classes by using a per-AF key:
261 : */
262 : static struct lock_class_key af_callback_keys[AF_MAX];
263 : static struct lock_class_key af_rlock_keys[AF_MAX];
264 : static struct lock_class_key af_wlock_keys[AF_MAX];
265 : static struct lock_class_key af_elock_keys[AF_MAX];
266 : static struct lock_class_key af_kern_callback_keys[AF_MAX];
267 :
268 : /* Run time adjustable parameters. */
269 : __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 : EXPORT_SYMBOL(sysctl_wmem_max);
271 : __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 : EXPORT_SYMBOL(sysctl_rmem_max);
273 : __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 : __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275 :
276 : /* Maximal space eaten by iovec or ancillary data plus some space */
277 : int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 : EXPORT_SYMBOL(sysctl_optmem_max);
279 :
280 : int sysctl_tstamp_allow_data __read_mostly = 1;
281 :
282 : DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283 : EXPORT_SYMBOL_GPL(memalloc_socks_key);
284 :
285 : /**
286 : * sk_set_memalloc - sets %SOCK_MEMALLOC
287 : * @sk: socket to set it on
288 : *
289 : * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 : * It's the responsibility of the admin to adjust min_free_kbytes
291 : * to meet the requirements
292 : */
293 0 : void sk_set_memalloc(struct sock *sk)
294 : {
295 0 : sock_set_flag(sk, SOCK_MEMALLOC);
296 0 : sk->sk_allocation |= __GFP_MEMALLOC;
297 0 : static_branch_inc(&memalloc_socks_key);
298 0 : }
299 : EXPORT_SYMBOL_GPL(sk_set_memalloc);
300 :
301 0 : void sk_clear_memalloc(struct sock *sk)
302 : {
303 0 : sock_reset_flag(sk, SOCK_MEMALLOC);
304 0 : sk->sk_allocation &= ~__GFP_MEMALLOC;
305 0 : static_branch_dec(&memalloc_socks_key);
306 :
307 : /*
308 : * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 : * progress of swapping. SOCK_MEMALLOC may be cleared while
310 : * it has rmem allocations due to the last swapfile being deactivated
311 : * but there is a risk that the socket is unusable due to exceeding
312 : * the rmem limits. Reclaim the reserves and obey rmem limits again.
313 : */
314 0 : sk_mem_reclaim(sk);
315 0 : }
316 : EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317 :
318 0 : int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
319 : {
320 0 : int ret;
321 0 : unsigned int noreclaim_flag;
322 :
323 : /* these should have been dropped before queueing */
324 0 : BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325 :
326 0 : noreclaim_flag = memalloc_noreclaim_save();
327 0 : ret = sk->sk_backlog_rcv(sk, skb);
328 0 : memalloc_noreclaim_restore(noreclaim_flag);
329 :
330 0 : return ret;
331 : }
332 : EXPORT_SYMBOL(__sk_backlog_rcv);
333 :
334 0 : static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
335 : {
336 0 : struct __kernel_sock_timeval tv;
337 :
338 0 : if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 : tv.tv_sec = 0;
340 : tv.tv_usec = 0;
341 : } else {
342 0 : tv.tv_sec = timeo / HZ;
343 0 : tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 : }
345 :
346 0 : if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 0 : struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 0 : *(struct old_timeval32 *)optval = tv32;
349 0 : return sizeof(tv32);
350 : }
351 :
352 0 : if (old_timeval) {
353 0 : struct __kernel_old_timeval old_tv;
354 0 : old_tv.tv_sec = tv.tv_sec;
355 0 : old_tv.tv_usec = tv.tv_usec;
356 0 : *(struct __kernel_old_timeval *)optval = old_tv;
357 0 : return sizeof(old_tv);
358 : }
359 :
360 0 : *(struct __kernel_sock_timeval *)optval = tv;
361 0 : return sizeof(tv);
362 : }
363 :
364 19 : static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
365 : bool old_timeval)
366 : {
367 19 : struct __kernel_sock_timeval tv;
368 :
369 38 : if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370 0 : struct old_timeval32 tv32;
371 :
372 0 : if (optlen < sizeof(tv32))
373 0 : return -EINVAL;
374 :
375 0 : if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
376 : return -EFAULT;
377 0 : tv.tv_sec = tv32.tv_sec;
378 0 : tv.tv_usec = tv32.tv_usec;
379 19 : } else if (old_timeval) {
380 19 : struct __kernel_old_timeval old_tv;
381 :
382 19 : if (optlen < sizeof(old_tv))
383 0 : return -EINVAL;
384 19 : if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
385 : return -EFAULT;
386 19 : tv.tv_sec = old_tv.tv_sec;
387 19 : tv.tv_usec = old_tv.tv_usec;
388 : } else {
389 0 : if (optlen < sizeof(tv))
390 : return -EINVAL;
391 0 : if (copy_from_sockptr(&tv, optval, sizeof(tv)))
392 : return -EFAULT;
393 : }
394 19 : if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 : return -EDOM;
396 :
397 19 : if (tv.tv_sec < 0) {
398 0 : static int warned __read_mostly;
399 :
400 0 : *timeo_p = 0;
401 0 : if (warned < 10 && net_ratelimit()) {
402 0 : warned++;
403 0 : pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 : __func__, current->comm, task_pid_nr(current));
405 : }
406 0 : return 0;
407 : }
408 19 : *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 19 : if (tv.tv_sec == 0 && tv.tv_usec == 0)
410 : return 0;
411 19 : if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412 19 : *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
413 : return 0;
414 : }
415 :
416 7 : static bool sock_needs_netstamp(const struct sock *sk)
417 : {
418 7 : switch (sk->sk_family) {
419 : case AF_UNSPEC:
420 : case AF_UNIX:
421 : return false;
422 : default:
423 4 : return true;
424 : }
425 : }
426 :
427 738 : static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
428 : {
429 738 : if (sk->sk_flags & flags) {
430 0 : sk->sk_flags &= ~flags;
431 0 : if (sock_needs_netstamp(sk) &&
432 0 : !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
433 0 : net_disable_timestamp();
434 : }
435 738 : }
436 :
437 :
438 0 : int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
439 : {
440 0 : unsigned long flags;
441 0 : struct sk_buff_head *list = &sk->sk_receive_queue;
442 :
443 0 : if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
444 0 : atomic_inc(&sk->sk_drops);
445 0 : trace_sock_rcvqueue_full(sk, skb);
446 0 : return -ENOMEM;
447 : }
448 :
449 0 : if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
450 0 : atomic_inc(&sk->sk_drops);
451 0 : return -ENOBUFS;
452 : }
453 :
454 0 : skb->dev = NULL;
455 0 : skb_set_owner_r(skb, sk);
456 :
457 : /* we escape from rcu protected region, make sure we dont leak
458 : * a norefcounted dst
459 : */
460 0 : skb_dst_force(skb);
461 :
462 0 : spin_lock_irqsave(&list->lock, flags);
463 0 : sock_skb_set_dropcount(sk, skb);
464 0 : __skb_queue_tail(list, skb);
465 0 : spin_unlock_irqrestore(&list->lock, flags);
466 :
467 0 : if (!sock_flag(sk, SOCK_DEAD))
468 0 : sk->sk_data_ready(sk);
469 : return 0;
470 : }
471 : EXPORT_SYMBOL(__sock_queue_rcv_skb);
472 :
473 0 : int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
474 : {
475 0 : int err;
476 :
477 0 : err = sk_filter(sk, skb);
478 0 : if (err)
479 : return err;
480 :
481 0 : return __sock_queue_rcv_skb(sk, skb);
482 : }
483 : EXPORT_SYMBOL(sock_queue_rcv_skb);
484 :
485 0 : int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
486 : const int nested, unsigned int trim_cap, bool refcounted)
487 : {
488 0 : int rc = NET_RX_SUCCESS;
489 :
490 0 : if (sk_filter_trim_cap(sk, skb, trim_cap))
491 0 : goto discard_and_relse;
492 :
493 0 : skb->dev = NULL;
494 :
495 0 : if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
496 0 : atomic_inc(&sk->sk_drops);
497 0 : goto discard_and_relse;
498 : }
499 0 : if (nested)
500 0 : bh_lock_sock_nested(sk);
501 : else
502 0 : bh_lock_sock(sk);
503 0 : if (!sock_owned_by_user(sk)) {
504 : /*
505 : * trylock + unlock semantics:
506 : */
507 0 : mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
508 :
509 0 : rc = sk_backlog_rcv(sk, skb);
510 :
511 0 : mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
512 0 : } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
513 0 : bh_unlock_sock(sk);
514 0 : atomic_inc(&sk->sk_drops);
515 0 : goto discard_and_relse;
516 : }
517 :
518 0 : bh_unlock_sock(sk);
519 0 : out:
520 0 : if (refcounted)
521 0 : sock_put(sk);
522 0 : return rc;
523 0 : discard_and_relse:
524 0 : kfree_skb(skb);
525 0 : goto out;
526 : }
527 : EXPORT_SYMBOL(__sk_receive_skb);
528 :
529 : INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
530 : u32));
531 : INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
532 : u32));
533 430 : struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
534 : {
535 430 : struct dst_entry *dst = __sk_dst_get(sk);
536 :
537 856 : if (dst && dst->obsolete &&
538 426 : INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
539 : dst, cookie) == NULL) {
540 0 : sk_tx_queue_clear(sk);
541 0 : sk->sk_dst_pending_confirm = 0;
542 0 : RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
543 0 : dst_release(dst);
544 0 : return NULL;
545 : }
546 :
547 : return dst;
548 : }
549 : EXPORT_SYMBOL(__sk_dst_check);
550 :
551 13 : struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
552 : {
553 13 : struct dst_entry *dst = sk_dst_get(sk);
554 :
555 26 : if (dst && dst->obsolete &&
556 13 : INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
557 : dst, cookie) == NULL) {
558 0 : sk_dst_reset(sk);
559 0 : dst_release(dst);
560 0 : return NULL;
561 : }
562 :
563 : return dst;
564 : }
565 : EXPORT_SYMBOL(sk_dst_check);
566 :
567 0 : static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
568 : {
569 0 : int ret = -ENOPROTOOPT;
570 : #ifdef CONFIG_NETDEVICES
571 0 : struct net *net = sock_net(sk);
572 :
573 : /* Sorry... */
574 0 : ret = -EPERM;
575 0 : if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
576 0 : goto out;
577 :
578 0 : ret = -EINVAL;
579 0 : if (ifindex < 0)
580 0 : goto out;
581 :
582 0 : sk->sk_bound_dev_if = ifindex;
583 0 : if (sk->sk_prot->rehash)
584 0 : sk->sk_prot->rehash(sk);
585 0 : sk_dst_reset(sk);
586 :
587 0 : ret = 0;
588 :
589 0 : out:
590 : #endif
591 :
592 0 : return ret;
593 : }
594 :
595 0 : int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
596 : {
597 0 : int ret;
598 :
599 0 : if (lock_sk)
600 0 : lock_sock(sk);
601 0 : ret = sock_bindtoindex_locked(sk, ifindex);
602 0 : if (lock_sk)
603 0 : release_sock(sk);
604 :
605 0 : return ret;
606 : }
607 : EXPORT_SYMBOL(sock_bindtoindex);
608 :
609 0 : static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
610 : {
611 0 : int ret = -ENOPROTOOPT;
612 : #ifdef CONFIG_NETDEVICES
613 0 : struct net *net = sock_net(sk);
614 0 : char devname[IFNAMSIZ];
615 0 : int index;
616 :
617 0 : ret = -EINVAL;
618 0 : if (optlen < 0)
619 0 : goto out;
620 :
621 : /* Bind this socket to a particular device like "eth0",
622 : * as specified in the passed interface name. If the
623 : * name is "" or the option length is zero the socket
624 : * is not bound.
625 : */
626 0 : if (optlen > IFNAMSIZ - 1)
627 : optlen = IFNAMSIZ - 1;
628 0 : memset(devname, 0, sizeof(devname));
629 :
630 0 : ret = -EFAULT;
631 0 : if (copy_from_sockptr(devname, optval, optlen))
632 0 : goto out;
633 :
634 0 : index = 0;
635 0 : if (devname[0] != '\0') {
636 0 : struct net_device *dev;
637 :
638 0 : rcu_read_lock();
639 0 : dev = dev_get_by_name_rcu(net, devname);
640 0 : if (dev)
641 0 : index = dev->ifindex;
642 0 : rcu_read_unlock();
643 0 : ret = -ENODEV;
644 0 : if (!dev)
645 0 : goto out;
646 : }
647 :
648 0 : return sock_bindtoindex(sk, index, true);
649 : out:
650 : #endif
651 :
652 : return ret;
653 : }
654 :
655 0 : static int sock_getbindtodevice(struct sock *sk, char __user *optval,
656 : int __user *optlen, int len)
657 : {
658 0 : int ret = -ENOPROTOOPT;
659 : #ifdef CONFIG_NETDEVICES
660 0 : struct net *net = sock_net(sk);
661 0 : char devname[IFNAMSIZ];
662 :
663 0 : if (sk->sk_bound_dev_if == 0) {
664 0 : len = 0;
665 0 : goto zero;
666 : }
667 :
668 0 : ret = -EINVAL;
669 0 : if (len < IFNAMSIZ)
670 0 : goto out;
671 :
672 0 : ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
673 0 : if (ret)
674 0 : goto out;
675 :
676 0 : len = strlen(devname) + 1;
677 :
678 0 : ret = -EFAULT;
679 0 : if (copy_to_user(optval, devname, len))
680 0 : goto out;
681 :
682 0 : zero:
683 0 : ret = -EFAULT;
684 0 : if (put_user(len, optlen))
685 0 : goto out;
686 :
687 : ret = 0;
688 :
689 0 : out:
690 : #endif
691 :
692 0 : return ret;
693 : }
694 :
695 0 : bool sk_mc_loop(struct sock *sk)
696 : {
697 0 : if (dev_recursion_level())
698 : return false;
699 0 : if (!sk)
700 : return true;
701 0 : switch (sk->sk_family) {
702 : case AF_INET:
703 0 : return inet_sk(sk)->mc_loop;
704 : #if IS_ENABLED(CONFIG_IPV6)
705 : case AF_INET6:
706 : return inet6_sk(sk)->mc_loop;
707 : #endif
708 : }
709 0 : WARN_ON_ONCE(1);
710 0 : return true;
711 : }
712 : EXPORT_SYMBOL(sk_mc_loop);
713 :
714 0 : void sock_set_reuseaddr(struct sock *sk)
715 : {
716 0 : lock_sock(sk);
717 0 : sk->sk_reuse = SK_CAN_REUSE;
718 0 : release_sock(sk);
719 0 : }
720 : EXPORT_SYMBOL(sock_set_reuseaddr);
721 :
722 0 : void sock_set_reuseport(struct sock *sk)
723 : {
724 0 : lock_sock(sk);
725 0 : sk->sk_reuseport = true;
726 0 : release_sock(sk);
727 0 : }
728 : EXPORT_SYMBOL(sock_set_reuseport);
729 :
730 0 : void sock_no_linger(struct sock *sk)
731 : {
732 0 : lock_sock(sk);
733 0 : sk->sk_lingertime = 0;
734 0 : sock_set_flag(sk, SOCK_LINGER);
735 0 : release_sock(sk);
736 0 : }
737 : EXPORT_SYMBOL(sock_no_linger);
738 :
739 0 : void sock_set_priority(struct sock *sk, u32 priority)
740 : {
741 0 : lock_sock(sk);
742 0 : sk->sk_priority = priority;
743 0 : release_sock(sk);
744 0 : }
745 : EXPORT_SYMBOL(sock_set_priority);
746 :
747 0 : void sock_set_sndtimeo(struct sock *sk, s64 secs)
748 : {
749 0 : lock_sock(sk);
750 0 : if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
751 0 : sk->sk_sndtimeo = secs * HZ;
752 : else
753 0 : sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
754 0 : release_sock(sk);
755 0 : }
756 : EXPORT_SYMBOL(sock_set_sndtimeo);
757 :
758 3 : static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
759 : {
760 3 : if (val) {
761 3 : sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
762 3 : sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
763 3 : sock_set_flag(sk, SOCK_RCVTSTAMP);
764 3 : sock_enable_timestamp(sk, SOCK_TIMESTAMP);
765 : } else {
766 0 : sock_reset_flag(sk, SOCK_RCVTSTAMP);
767 0 : sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
768 : }
769 3 : }
770 :
771 0 : void sock_enable_timestamps(struct sock *sk)
772 : {
773 0 : lock_sock(sk);
774 0 : __sock_set_timestamps(sk, true, false, true);
775 0 : release_sock(sk);
776 0 : }
777 : EXPORT_SYMBOL(sock_enable_timestamps);
778 :
779 0 : void sock_set_keepalive(struct sock *sk)
780 : {
781 0 : lock_sock(sk);
782 0 : if (sk->sk_prot->keepalive)
783 0 : sk->sk_prot->keepalive(sk, true);
784 0 : sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
785 0 : release_sock(sk);
786 0 : }
787 : EXPORT_SYMBOL(sock_set_keepalive);
788 :
789 83 : static void __sock_set_rcvbuf(struct sock *sk, int val)
790 : {
791 : /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
792 : * as a negative value.
793 : */
794 83 : val = min_t(int, val, INT_MAX / 2);
795 83 : sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
796 :
797 : /* We double it on the way in to account for "struct sk_buff" etc.
798 : * overhead. Applications assume that the SO_RCVBUF setting they make
799 : * will allow that much actual data to be received on that socket.
800 : *
801 : * Applications are unaware that "struct sk_buff" and other overheads
802 : * allocate from the receive buffer during socket buffer allocation.
803 : *
804 : * And after considering the possible alternatives, returning the value
805 : * we actually used in getsockopt is the most desirable behavior.
806 : */
807 83 : WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
808 83 : }
809 :
810 0 : void sock_set_rcvbuf(struct sock *sk, int val)
811 : {
812 0 : lock_sock(sk);
813 0 : __sock_set_rcvbuf(sk, val);
814 0 : release_sock(sk);
815 0 : }
816 : EXPORT_SYMBOL(sock_set_rcvbuf);
817 :
818 0 : void sock_set_mark(struct sock *sk, u32 val)
819 : {
820 0 : lock_sock(sk);
821 0 : sk->sk_mark = val;
822 0 : release_sock(sk);
823 0 : }
824 : EXPORT_SYMBOL(sock_set_mark);
825 :
826 : /*
827 : * This is meant for all protocols to use and covers goings on
828 : * at the socket level. Everything here is generic.
829 : */
830 :
831 351 : int sock_setsockopt(struct socket *sock, int level, int optname,
832 : sockptr_t optval, unsigned int optlen)
833 : {
834 351 : struct sock_txtime sk_txtime;
835 351 : struct sock *sk = sock->sk;
836 351 : int val;
837 351 : int valbool;
838 351 : struct linger ling;
839 351 : int ret = 0;
840 :
841 : /*
842 : * Options without arguments
843 : */
844 :
845 351 : if (optname == SO_BINDTODEVICE)
846 0 : return sock_setbindtodevice(sk, optval, optlen);
847 :
848 351 : if (optlen < sizeof(int))
849 : return -EINVAL;
850 :
851 351 : if (copy_from_sockptr(&val, optval, sizeof(val)))
852 : return -EFAULT;
853 :
854 351 : valbool = val ? 1 : 0;
855 :
856 351 : lock_sock(sk);
857 :
858 351 : switch (optname) {
859 0 : case SO_DEBUG:
860 0 : if (val && !capable(CAP_NET_ADMIN))
861 : ret = -EACCES;
862 : else
863 0 : sock_valbool_flag(sk, SOCK_DBG, valbool);
864 : break;
865 20 : case SO_REUSEADDR:
866 20 : sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
867 20 : break;
868 0 : case SO_REUSEPORT:
869 0 : sk->sk_reuseport = valbool;
870 0 : break;
871 0 : case SO_TYPE:
872 : case SO_PROTOCOL:
873 : case SO_DOMAIN:
874 : case SO_ERROR:
875 0 : ret = -ENOPROTOOPT;
876 0 : break;
877 0 : case SO_DONTROUTE:
878 0 : sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
879 0 : sk_dst_reset(sk);
880 : break;
881 0 : case SO_BROADCAST:
882 0 : sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
883 0 : break;
884 177 : case SO_SNDBUF:
885 : /* Don't error on this BSD doesn't and if you think
886 : * about it this is right. Otherwise apps have to
887 : * play 'guess the biggest size' games. RCVBUF/SNDBUF
888 : * are treated in BSD as hints
889 : */
890 177 : val = min_t(u32, val, sysctl_wmem_max);
891 178 : set_sndbuf:
892 : /* Ensure val * 2 fits into an int, to prevent max_t()
893 : * from treating it as a negative value.
894 : */
895 178 : val = min_t(int, val, INT_MAX / 2);
896 178 : sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
897 178 : WRITE_ONCE(sk->sk_sndbuf,
898 : max_t(int, val * 2, SOCK_MIN_SNDBUF));
899 : /* Wake up sending tasks if we upped the value. */
900 178 : sk->sk_write_space(sk);
901 178 : break;
902 :
903 1 : case SO_SNDBUFFORCE:
904 1 : if (!capable(CAP_NET_ADMIN)) {
905 : ret = -EPERM;
906 : break;
907 : }
908 :
909 : /* No negative values (to prevent underflow, as val will be
910 : * multiplied by 2).
911 : */
912 1 : if (val < 0)
913 0 : val = 0;
914 1 : goto set_sndbuf;
915 :
916 75 : case SO_RCVBUF:
917 : /* Don't error on this BSD doesn't and if you think
918 : * about it this is right. Otherwise apps have to
919 : * play 'guess the biggest size' games. RCVBUF/SNDBUF
920 : * are treated in BSD as hints
921 : */
922 75 : __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
923 : break;
924 :
925 8 : case SO_RCVBUFFORCE:
926 8 : if (!capable(CAP_NET_ADMIN)) {
927 : ret = -EPERM;
928 : break;
929 : }
930 :
931 : /* No negative values (to prevent underflow, as val will be
932 : * multiplied by 2).
933 : */
934 8 : __sock_set_rcvbuf(sk, max(val, 0));
935 : break;
936 :
937 4 : case SO_KEEPALIVE:
938 4 : if (sk->sk_prot->keepalive)
939 4 : sk->sk_prot->keepalive(sk, valbool);
940 4 : sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
941 4 : break;
942 :
943 0 : case SO_OOBINLINE:
944 0 : sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
945 0 : break;
946 :
947 0 : case SO_NO_CHECK:
948 0 : sk->sk_no_check_tx = valbool;
949 0 : break;
950 :
951 0 : case SO_PRIORITY:
952 0 : if ((val >= 0 && val <= 6) ||
953 0 : ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
954 0 : sk->sk_priority = val;
955 : else
956 : ret = -EPERM;
957 : break;
958 :
959 0 : case SO_LINGER:
960 0 : if (optlen < sizeof(ling)) {
961 : ret = -EINVAL; /* 1003.1g */
962 : break;
963 : }
964 0 : if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
965 : ret = -EFAULT;
966 : break;
967 : }
968 0 : if (!ling.l_onoff)
969 0 : sock_reset_flag(sk, SOCK_LINGER);
970 : else {
971 : #if (BITS_PER_LONG == 32)
972 : if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
973 : sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
974 : else
975 : #endif
976 0 : sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
977 0 : sock_set_flag(sk, SOCK_LINGER);
978 : }
979 : break;
980 :
981 : case SO_BSDCOMPAT:
982 : break;
983 :
984 33 : case SO_PASSCRED:
985 33 : if (valbool)
986 33 : set_bit(SOCK_PASSCRED, &sock->flags);
987 : else
988 0 : clear_bit(SOCK_PASSCRED, &sock->flags);
989 : break;
990 :
991 3 : case SO_TIMESTAMP_OLD:
992 3 : __sock_set_timestamps(sk, valbool, false, false);
993 3 : break;
994 0 : case SO_TIMESTAMP_NEW:
995 0 : __sock_set_timestamps(sk, valbool, true, false);
996 0 : break;
997 0 : case SO_TIMESTAMPNS_OLD:
998 0 : __sock_set_timestamps(sk, valbool, false, true);
999 0 : break;
1000 0 : case SO_TIMESTAMPNS_NEW:
1001 0 : __sock_set_timestamps(sk, valbool, true, true);
1002 0 : break;
1003 0 : case SO_TIMESTAMPING_NEW:
1004 : case SO_TIMESTAMPING_OLD:
1005 0 : if (val & ~SOF_TIMESTAMPING_MASK) {
1006 : ret = -EINVAL;
1007 : break;
1008 : }
1009 :
1010 0 : if (val & SOF_TIMESTAMPING_OPT_ID &&
1011 0 : !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1012 0 : if (sk->sk_protocol == IPPROTO_TCP &&
1013 : sk->sk_type == SOCK_STREAM) {
1014 0 : if ((1 << sk->sk_state) &
1015 : (TCPF_CLOSE | TCPF_LISTEN)) {
1016 : ret = -EINVAL;
1017 : break;
1018 : }
1019 0 : sk->sk_tskey = tcp_sk(sk)->snd_una;
1020 : } else {
1021 0 : sk->sk_tskey = 0;
1022 : }
1023 : }
1024 :
1025 0 : if (val & SOF_TIMESTAMPING_OPT_STATS &&
1026 : !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1027 : ret = -EINVAL;
1028 : break;
1029 : }
1030 :
1031 0 : sk->sk_tsflags = val;
1032 0 : sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
1033 :
1034 0 : if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1035 0 : sock_enable_timestamp(sk,
1036 : SOCK_TIMESTAMPING_RX_SOFTWARE);
1037 : else
1038 0 : sock_disable_timestamp(sk,
1039 : (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1040 : break;
1041 :
1042 0 : case SO_RCVLOWAT:
1043 0 : if (val < 0)
1044 0 : val = INT_MAX;
1045 0 : if (sock->ops->set_rcvlowat)
1046 0 : ret = sock->ops->set_rcvlowat(sk, val);
1047 : else
1048 0 : WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1049 : break;
1050 :
1051 0 : case SO_RCVTIMEO_OLD:
1052 : case SO_RCVTIMEO_NEW:
1053 0 : ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1054 : optlen, optname == SO_RCVTIMEO_OLD);
1055 0 : break;
1056 :
1057 19 : case SO_SNDTIMEO_OLD:
1058 : case SO_SNDTIMEO_NEW:
1059 19 : ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1060 : optlen, optname == SO_SNDTIMEO_OLD);
1061 19 : break;
1062 :
1063 7 : case SO_ATTACH_FILTER: {
1064 7 : struct sock_fprog fprog;
1065 :
1066 7 : ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1067 7 : if (!ret)
1068 7 : ret = sk_attach_filter(&fprog, sk);
1069 7 : break;
1070 : }
1071 0 : case SO_ATTACH_BPF:
1072 0 : ret = -EINVAL;
1073 0 : if (optlen == sizeof(u32)) {
1074 0 : u32 ufd;
1075 :
1076 0 : ret = -EFAULT;
1077 0 : if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1078 : break;
1079 :
1080 0 : ret = sk_attach_bpf(ufd, sk);
1081 : }
1082 : break;
1083 :
1084 0 : case SO_ATTACH_REUSEPORT_CBPF: {
1085 0 : struct sock_fprog fprog;
1086 :
1087 0 : ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1088 0 : if (!ret)
1089 0 : ret = sk_reuseport_attach_filter(&fprog, sk);
1090 0 : break;
1091 : }
1092 0 : case SO_ATTACH_REUSEPORT_EBPF:
1093 0 : ret = -EINVAL;
1094 0 : if (optlen == sizeof(u32)) {
1095 0 : u32 ufd;
1096 :
1097 0 : ret = -EFAULT;
1098 0 : if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1099 : break;
1100 :
1101 0 : ret = sk_reuseport_attach_bpf(ufd, sk);
1102 : }
1103 : break;
1104 :
1105 0 : case SO_DETACH_REUSEPORT_BPF:
1106 0 : ret = reuseport_detach_prog(sk);
1107 0 : break;
1108 :
1109 0 : case SO_DETACH_FILTER:
1110 0 : ret = sk_detach_filter(sk);
1111 0 : break;
1112 :
1113 : case SO_LOCK_FILTER:
1114 0 : if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1115 : ret = -EPERM;
1116 : else
1117 0 : sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1118 : break;
1119 :
1120 4 : case SO_PASSSEC:
1121 4 : if (valbool)
1122 4 : set_bit(SOCK_PASSSEC, &sock->flags);
1123 : else
1124 0 : clear_bit(SOCK_PASSSEC, &sock->flags);
1125 : break;
1126 : case SO_MARK:
1127 0 : if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1128 : ret = -EPERM;
1129 0 : } else if (val != sk->sk_mark) {
1130 0 : sk->sk_mark = val;
1131 0 : sk_dst_reset(sk);
1132 : }
1133 : break;
1134 :
1135 0 : case SO_RXQ_OVFL:
1136 0 : sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1137 0 : break;
1138 :
1139 0 : case SO_WIFI_STATUS:
1140 0 : sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1141 0 : break;
1142 :
1143 0 : case SO_PEEK_OFF:
1144 0 : if (sock->ops->set_peek_off)
1145 0 : ret = sock->ops->set_peek_off(sk, val);
1146 : else
1147 : ret = -EOPNOTSUPP;
1148 : break;
1149 :
1150 0 : case SO_NOFCS:
1151 0 : sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1152 0 : break;
1153 :
1154 0 : case SO_SELECT_ERR_QUEUE:
1155 0 : sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1156 0 : break;
1157 :
1158 : #ifdef CONFIG_NET_RX_BUSY_POLL
1159 0 : case SO_BUSY_POLL:
1160 : /* allow unprivileged users to decrease the value */
1161 0 : if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1162 : ret = -EPERM;
1163 : else {
1164 0 : if (val < 0)
1165 : ret = -EINVAL;
1166 : else
1167 0 : sk->sk_ll_usec = val;
1168 : }
1169 : break;
1170 0 : case SO_PREFER_BUSY_POLL:
1171 0 : if (valbool && !capable(CAP_NET_ADMIN))
1172 : ret = -EPERM;
1173 : else
1174 0 : WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1175 : break;
1176 : case SO_BUSY_POLL_BUDGET:
1177 0 : if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1178 : ret = -EPERM;
1179 : } else {
1180 0 : if (val < 0 || val > U16_MAX)
1181 : ret = -EINVAL;
1182 : else
1183 0 : WRITE_ONCE(sk->sk_busy_poll_budget, val);
1184 : }
1185 : break;
1186 : #endif
1187 :
1188 0 : case SO_MAX_PACING_RATE:
1189 : {
1190 0 : unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1191 :
1192 0 : if (sizeof(ulval) != sizeof(val) &&
1193 0 : optlen >= sizeof(ulval) &&
1194 0 : copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1195 : ret = -EFAULT;
1196 : break;
1197 : }
1198 0 : if (ulval != ~0UL)
1199 0 : cmpxchg(&sk->sk_pacing_status,
1200 : SK_PACING_NONE,
1201 : SK_PACING_NEEDED);
1202 0 : sk->sk_max_pacing_rate = ulval;
1203 0 : sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1204 0 : break;
1205 : }
1206 : case SO_INCOMING_CPU:
1207 0 : WRITE_ONCE(sk->sk_incoming_cpu, val);
1208 0 : break;
1209 :
1210 0 : case SO_CNX_ADVICE:
1211 0 : if (val == 1)
1212 0 : dst_negative_advice(sk);
1213 : break;
1214 :
1215 0 : case SO_ZEROCOPY:
1216 0 : if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1217 0 : if (!((sk->sk_type == SOCK_STREAM &&
1218 : sk->sk_protocol == IPPROTO_TCP) ||
1219 : (sk->sk_type == SOCK_DGRAM &&
1220 : sk->sk_protocol == IPPROTO_UDP)))
1221 : ret = -ENOTSUPP;
1222 0 : } else if (sk->sk_family != PF_RDS) {
1223 : ret = -ENOTSUPP;
1224 : }
1225 : if (!ret) {
1226 0 : if (val < 0 || val > 1)
1227 : ret = -EINVAL;
1228 : else
1229 0 : sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1230 : }
1231 : break;
1232 :
1233 0 : case SO_TXTIME:
1234 0 : if (optlen != sizeof(struct sock_txtime)) {
1235 : ret = -EINVAL;
1236 : break;
1237 0 : } else if (copy_from_sockptr(&sk_txtime, optval,
1238 : sizeof(struct sock_txtime))) {
1239 : ret = -EFAULT;
1240 : break;
1241 0 : } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1242 : ret = -EINVAL;
1243 : break;
1244 : }
1245 : /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1246 : * scheduler has enough safe guards.
1247 : */
1248 0 : if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1249 0 : !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1250 : ret = -EPERM;
1251 : break;
1252 : }
1253 0 : sock_valbool_flag(sk, SOCK_TXTIME, true);
1254 0 : sk->sk_clockid = sk_txtime.clockid;
1255 0 : sk->sk_txtime_deadline_mode =
1256 0 : !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1257 0 : sk->sk_txtime_report_errors =
1258 0 : !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1259 0 : break;
1260 :
1261 0 : case SO_BINDTOIFINDEX:
1262 0 : ret = sock_bindtoindex_locked(sk, val);
1263 0 : break;
1264 :
1265 0 : default:
1266 0 : ret = -ENOPROTOOPT;
1267 0 : break;
1268 : }
1269 351 : release_sock(sk);
1270 351 : return ret;
1271 : }
1272 : EXPORT_SYMBOL(sock_setsockopt);
1273 :
1274 :
1275 157 : static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1276 : struct ucred *ucred)
1277 : {
1278 157 : ucred->pid = pid_vnr(pid);
1279 157 : ucred->uid = ucred->gid = -1;
1280 157 : if (cred) {
1281 157 : struct user_namespace *current_ns = current_user_ns();
1282 :
1283 157 : ucred->uid = from_kuid_munged(current_ns, cred->euid);
1284 157 : ucred->gid = from_kgid_munged(current_ns, cred->egid);
1285 : }
1286 157 : }
1287 :
1288 55 : static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1289 : {
1290 55 : struct user_namespace *user_ns = current_user_ns();
1291 55 : int i;
1292 :
1293 55 : for (i = 0; i < src->ngroups; i++)
1294 0 : if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1295 : return -EFAULT;
1296 :
1297 : return 0;
1298 : }
1299 :
1300 623 : int sock_getsockopt(struct socket *sock, int level, int optname,
1301 : char __user *optval, int __user *optlen)
1302 : {
1303 623 : struct sock *sk = sock->sk;
1304 :
1305 623 : union {
1306 : int val;
1307 : u64 val64;
1308 : unsigned long ulval;
1309 : struct linger ling;
1310 : struct old_timeval32 tm32;
1311 : struct __kernel_old_timeval tm;
1312 : struct __kernel_sock_timeval stm;
1313 : struct sock_txtime txtime;
1314 : } v;
1315 :
1316 623 : int lv = sizeof(int);
1317 623 : int len;
1318 :
1319 623 : if (get_user(len, optlen))
1320 : return -EFAULT;
1321 623 : if (len < 0)
1322 : return -EINVAL;
1323 :
1324 623 : memset(&v, 0, sizeof(v));
1325 :
1326 623 : switch (optname) {
1327 : case SO_DEBUG:
1328 0 : v.val = sock_flag(sk, SOCK_DBG);
1329 0 : break;
1330 :
1331 : case SO_DONTROUTE:
1332 0 : v.val = sock_flag(sk, SOCK_LOCALROUTE);
1333 0 : break;
1334 :
1335 : case SO_BROADCAST:
1336 0 : v.val = sock_flag(sk, SOCK_BROADCAST);
1337 0 : break;
1338 :
1339 171 : case SO_SNDBUF:
1340 171 : v.val = sk->sk_sndbuf;
1341 171 : break;
1342 :
1343 69 : case SO_RCVBUF:
1344 69 : v.val = sk->sk_rcvbuf;
1345 69 : break;
1346 :
1347 0 : case SO_REUSEADDR:
1348 0 : v.val = sk->sk_reuse;
1349 0 : break;
1350 :
1351 0 : case SO_REUSEPORT:
1352 0 : v.val = sk->sk_reuseport;
1353 0 : break;
1354 :
1355 : case SO_KEEPALIVE:
1356 0 : v.val = sock_flag(sk, SOCK_KEEPOPEN);
1357 0 : break;
1358 :
1359 33 : case SO_TYPE:
1360 33 : v.val = sk->sk_type;
1361 33 : break;
1362 :
1363 1 : case SO_PROTOCOL:
1364 1 : v.val = sk->sk_protocol;
1365 1 : break;
1366 :
1367 0 : case SO_DOMAIN:
1368 0 : v.val = sk->sk_family;
1369 0 : break;
1370 :
1371 0 : case SO_ERROR:
1372 0 : v.val = -sock_error(sk);
1373 0 : if (v.val == 0)
1374 0 : v.val = xchg(&sk->sk_err_soft, 0);
1375 : break;
1376 :
1377 : case SO_OOBINLINE:
1378 0 : v.val = sock_flag(sk, SOCK_URGINLINE);
1379 0 : break;
1380 :
1381 0 : case SO_NO_CHECK:
1382 0 : v.val = sk->sk_no_check_tx;
1383 0 : break;
1384 :
1385 0 : case SO_PRIORITY:
1386 0 : v.val = sk->sk_priority;
1387 0 : break;
1388 :
1389 0 : case SO_LINGER:
1390 0 : lv = sizeof(v.ling);
1391 0 : v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1392 0 : v.ling.l_linger = sk->sk_lingertime / HZ;
1393 0 : break;
1394 :
1395 : case SO_BSDCOMPAT:
1396 : break;
1397 :
1398 : case SO_TIMESTAMP_OLD:
1399 0 : v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1400 0 : !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1401 0 : !sock_flag(sk, SOCK_RCVTSTAMPNS);
1402 0 : break;
1403 :
1404 : case SO_TIMESTAMPNS_OLD:
1405 0 : v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1406 0 : break;
1407 :
1408 : case SO_TIMESTAMP_NEW:
1409 0 : v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1410 0 : break;
1411 :
1412 : case SO_TIMESTAMPNS_NEW:
1413 0 : v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1414 0 : break;
1415 :
1416 0 : case SO_TIMESTAMPING_OLD:
1417 0 : v.val = sk->sk_tsflags;
1418 0 : break;
1419 :
1420 0 : case SO_RCVTIMEO_OLD:
1421 : case SO_RCVTIMEO_NEW:
1422 0 : lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1423 0 : break;
1424 :
1425 0 : case SO_SNDTIMEO_OLD:
1426 : case SO_SNDTIMEO_NEW:
1427 0 : lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1428 0 : break;
1429 :
1430 0 : case SO_RCVLOWAT:
1431 0 : v.val = sk->sk_rcvlowat;
1432 0 : break;
1433 :
1434 0 : case SO_SNDLOWAT:
1435 0 : v.val = 1;
1436 0 : break;
1437 :
1438 0 : case SO_PASSCRED:
1439 0 : v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1440 0 : break;
1441 :
1442 157 : case SO_PEERCRED:
1443 : {
1444 157 : struct ucred peercred;
1445 157 : if (len > sizeof(peercred))
1446 0 : len = sizeof(peercred);
1447 157 : cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1448 314 : if (copy_to_user(optval, &peercred, len))
1449 0 : return -EFAULT;
1450 157 : goto lenout;
1451 : }
1452 :
1453 55 : case SO_PEERGROUPS:
1454 : {
1455 55 : int ret, n;
1456 :
1457 55 : if (!sk->sk_peer_cred)
1458 : return -ENODATA;
1459 :
1460 55 : n = sk->sk_peer_cred->group_info->ngroups;
1461 55 : if (len < n * sizeof(gid_t)) {
1462 0 : len = n * sizeof(gid_t);
1463 0 : return put_user(len, optlen) ? -EFAULT : -ERANGE;
1464 : }
1465 55 : len = n * sizeof(gid_t);
1466 :
1467 55 : ret = groups_to_user((gid_t __user *)optval,
1468 : sk->sk_peer_cred->group_info);
1469 55 : if (ret)
1470 : return ret;
1471 55 : goto lenout;
1472 : }
1473 :
1474 0 : case SO_PEERNAME:
1475 : {
1476 0 : char address[128];
1477 :
1478 0 : lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1479 0 : if (lv < 0)
1480 0 : return -ENOTCONN;
1481 0 : if (lv < len)
1482 : return -EINVAL;
1483 0 : if (copy_to_user(optval, address, len))
1484 : return -EFAULT;
1485 0 : goto lenout;
1486 : }
1487 :
1488 : /* Dubious BSD thing... Probably nobody even uses it, but
1489 : * the UNIX standard wants it for whatever reason... -DaveM
1490 : */
1491 57 : case SO_ACCEPTCONN:
1492 57 : v.val = sk->sk_state == TCP_LISTEN;
1493 57 : break;
1494 :
1495 0 : case SO_PASSSEC:
1496 0 : v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1497 0 : break;
1498 :
1499 : case SO_PEERSEC:
1500 623 : return security_socket_getpeersec_stream(sock, optval, optlen, len);
1501 :
1502 0 : case SO_MARK:
1503 0 : v.val = sk->sk_mark;
1504 0 : break;
1505 :
1506 : case SO_RXQ_OVFL:
1507 0 : v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1508 0 : break;
1509 :
1510 : case SO_WIFI_STATUS:
1511 0 : v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1512 0 : break;
1513 :
1514 0 : case SO_PEEK_OFF:
1515 0 : if (!sock->ops->set_peek_off)
1516 : return -EOPNOTSUPP;
1517 :
1518 0 : v.val = sk->sk_peek_off;
1519 0 : break;
1520 : case SO_NOFCS:
1521 0 : v.val = sock_flag(sk, SOCK_NOFCS);
1522 0 : break;
1523 :
1524 0 : case SO_BINDTODEVICE:
1525 0 : return sock_getbindtodevice(sk, optval, optlen, len);
1526 :
1527 0 : case SO_GET_FILTER:
1528 0 : len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1529 0 : if (len < 0)
1530 : return len;
1531 :
1532 0 : goto lenout;
1533 :
1534 : case SO_LOCK_FILTER:
1535 0 : v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1536 0 : break;
1537 :
1538 : case SO_BPF_EXTENSIONS:
1539 0 : v.val = bpf_tell_extensions();
1540 0 : break;
1541 :
1542 : case SO_SELECT_ERR_QUEUE:
1543 0 : v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1544 0 : break;
1545 :
1546 : #ifdef CONFIG_NET_RX_BUSY_POLL
1547 0 : case SO_BUSY_POLL:
1548 0 : v.val = sk->sk_ll_usec;
1549 0 : break;
1550 : case SO_PREFER_BUSY_POLL:
1551 0 : v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1552 0 : break;
1553 : #endif
1554 :
1555 0 : case SO_MAX_PACING_RATE:
1556 0 : if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1557 0 : lv = sizeof(v.ulval);
1558 0 : v.ulval = sk->sk_max_pacing_rate;
1559 : } else {
1560 : /* 32bit version */
1561 0 : v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1562 : }
1563 : break;
1564 :
1565 : case SO_INCOMING_CPU:
1566 0 : v.val = READ_ONCE(sk->sk_incoming_cpu);
1567 0 : break;
1568 :
1569 0 : case SO_MEMINFO:
1570 : {
1571 0 : u32 meminfo[SK_MEMINFO_VARS];
1572 :
1573 0 : sk_get_meminfo(sk, meminfo);
1574 :
1575 0 : len = min_t(unsigned int, len, sizeof(meminfo));
1576 0 : if (copy_to_user(optval, &meminfo, len))
1577 0 : return -EFAULT;
1578 :
1579 0 : goto lenout;
1580 : }
1581 :
1582 : #ifdef CONFIG_NET_RX_BUSY_POLL
1583 : case SO_INCOMING_NAPI_ID:
1584 0 : v.val = READ_ONCE(sk->sk_napi_id);
1585 :
1586 : /* aggregate non-NAPI IDs down to 0 */
1587 0 : if (v.val < MIN_NAPI_ID)
1588 0 : v.val = 0;
1589 :
1590 : break;
1591 : #endif
1592 :
1593 0 : case SO_COOKIE:
1594 0 : lv = sizeof(u64);
1595 0 : if (len < lv)
1596 : return -EINVAL;
1597 0 : v.val64 = sock_gen_cookie(sk);
1598 0 : break;
1599 :
1600 : case SO_ZEROCOPY:
1601 0 : v.val = sock_flag(sk, SOCK_ZEROCOPY);
1602 0 : break;
1603 :
1604 0 : case SO_TXTIME:
1605 0 : lv = sizeof(v.txtime);
1606 0 : v.txtime.clockid = sk->sk_clockid;
1607 0 : v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1608 0 : SOF_TXTIME_DEADLINE_MODE : 0;
1609 0 : v.txtime.flags |= sk->sk_txtime_report_errors ?
1610 0 : SOF_TXTIME_REPORT_ERRORS : 0;
1611 0 : break;
1612 :
1613 0 : case SO_BINDTOIFINDEX:
1614 0 : v.val = sk->sk_bound_dev_if;
1615 0 : break;
1616 :
1617 : default:
1618 : /* We implement the SO_SNDLOWAT etc to not be settable
1619 : * (1003.1g 7).
1620 : */
1621 : return -ENOPROTOOPT;
1622 : }
1623 :
1624 331 : if (len > lv)
1625 : len = lv;
1626 662 : if (copy_to_user(optval, &v, len))
1627 : return -EFAULT;
1628 331 : lenout:
1629 543 : if (put_user(len, optlen))
1630 0 : return -EFAULT;
1631 : return 0;
1632 : }
1633 :
1634 : /*
1635 : * Initialize an sk_lock.
1636 : *
1637 : * (We also register the sk_lock with the lock validator.)
1638 : */
1639 837 : static inline void sock_lock_init(struct sock *sk)
1640 : {
1641 837 : if (sk->sk_kern_sock)
1642 13 : sock_lock_init_class_and_name(
1643 : sk,
1644 : af_family_kern_slock_key_strings[sk->sk_family],
1645 : af_family_kern_slock_keys + sk->sk_family,
1646 : af_family_kern_key_strings[sk->sk_family],
1647 : af_family_kern_keys + sk->sk_family);
1648 : else
1649 824 : sock_lock_init_class_and_name(
1650 : sk,
1651 : af_family_slock_key_strings[sk->sk_family],
1652 : af_family_slock_keys + sk->sk_family,
1653 : af_family_key_strings[sk->sk_family],
1654 : af_family_keys + sk->sk_family);
1655 837 : }
1656 :
1657 : /*
1658 : * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1659 : * even temporarly, because of RCU lookups. sk_node should also be left as is.
1660 : * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1661 : */
1662 4 : static void sock_copy(struct sock *nsk, const struct sock *osk)
1663 : {
1664 4 : const struct proto *prot = READ_ONCE(osk->sk_prot);
1665 : #ifdef CONFIG_SECURITY_NETWORK
1666 : void *sptr = nsk->sk_security;
1667 : #endif
1668 :
1669 : /* If we move sk_tx_queue_mapping out of the private section,
1670 : * we must check if sk_tx_queue_clear() is called after
1671 : * sock_copy() in sk_clone_lock().
1672 : */
1673 4 : BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1674 : offsetof(struct sock, sk_dontcopy_begin) ||
1675 : offsetof(struct sock, sk_tx_queue_mapping) >=
1676 : offsetof(struct sock, sk_dontcopy_end));
1677 :
1678 4 : memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1679 :
1680 4 : memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1681 4 : prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1682 :
1683 : #ifdef CONFIG_SECURITY_NETWORK
1684 : nsk->sk_security = sptr;
1685 : security_sk_clone(osk, nsk);
1686 : #endif
1687 4 : }
1688 :
1689 837 : static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1690 : int family)
1691 : {
1692 837 : struct sock *sk;
1693 837 : struct kmem_cache *slab;
1694 :
1695 837 : slab = prot->slab;
1696 837 : if (slab != NULL) {
1697 781 : sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1698 781 : if (!sk)
1699 : return sk;
1700 781 : if (want_init_on_alloc(priority))
1701 777 : sk_prot_clear_nulls(sk, prot->obj_size);
1702 : } else
1703 56 : sk = kmalloc(prot->obj_size, priority);
1704 :
1705 837 : if (sk != NULL) {
1706 837 : if (security_sk_alloc(sk, family, priority))
1707 : goto out_free;
1708 :
1709 837 : if (!try_module_get(prot->owner))
1710 : goto out_free_sec;
1711 : }
1712 :
1713 : return sk;
1714 :
1715 : out_free_sec:
1716 : security_sk_free(sk);
1717 : out_free:
1718 : if (slab != NULL)
1719 : kmem_cache_free(slab, sk);
1720 : else
1721 : kfree(sk);
1722 : return NULL;
1723 : }
1724 :
1725 738 : static void sk_prot_free(struct proto *prot, struct sock *sk)
1726 : {
1727 738 : struct kmem_cache *slab;
1728 738 : struct module *owner;
1729 :
1730 738 : owner = prot->owner;
1731 738 : slab = prot->slab;
1732 :
1733 738 : cgroup_sk_free(&sk->sk_cgrp_data);
1734 738 : mem_cgroup_sk_free(sk);
1735 738 : security_sk_free(sk);
1736 738 : if (slab != NULL)
1737 695 : kmem_cache_free(slab, sk);
1738 : else
1739 43 : kfree(sk);
1740 737 : module_put(owner);
1741 737 : }
1742 :
1743 : /**
1744 : * sk_alloc - All socket objects are allocated here
1745 : * @net: the applicable net namespace
1746 : * @family: protocol family
1747 : * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1748 : * @prot: struct proto associated with this new sock instance
1749 : * @kern: is this to be a kernel socket?
1750 : */
1751 833 : struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1752 : struct proto *prot, int kern)
1753 : {
1754 833 : struct sock *sk;
1755 :
1756 833 : sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1757 833 : if (sk) {
1758 833 : sk->sk_family = family;
1759 : /*
1760 : * See comment in struct sock definition to understand
1761 : * why we need sk_prot_creator -acme
1762 : */
1763 833 : sk->sk_prot = sk->sk_prot_creator = prot;
1764 833 : sk->sk_kern_sock = kern;
1765 833 : sock_lock_init(sk);
1766 833 : sk->sk_net_refcnt = kern ? 0 : 1;
1767 833 : if (likely(sk->sk_net_refcnt)) {
1768 820 : get_net(net);
1769 1653 : sock_inuse_add(net, 1);
1770 : }
1771 :
1772 833 : sock_net_set(sk, net);
1773 833 : refcount_set(&sk->sk_wmem_alloc, 1);
1774 :
1775 833 : mem_cgroup_sk_alloc(sk);
1776 833 : cgroup_sk_alloc(&sk->sk_cgrp_data);
1777 833 : sock_update_classid(&sk->sk_cgrp_data);
1778 833 : sock_update_netprioidx(&sk->sk_cgrp_data);
1779 833 : sk_tx_queue_clear(sk);
1780 : }
1781 :
1782 833 : return sk;
1783 : }
1784 : EXPORT_SYMBOL(sk_alloc);
1785 :
1786 : /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1787 : * grace period. This is the case for UDP sockets and TCP listeners.
1788 : */
1789 737 : static void __sk_destruct(struct rcu_head *head)
1790 : {
1791 737 : struct sock *sk = container_of(head, struct sock, sk_rcu);
1792 737 : struct sk_filter *filter;
1793 :
1794 737 : if (sk->sk_destruct)
1795 737 : sk->sk_destruct(sk);
1796 :
1797 738 : filter = rcu_dereference_check(sk->sk_filter,
1798 : refcount_read(&sk->sk_wmem_alloc) == 0);
1799 738 : if (filter) {
1800 0 : sk_filter_uncharge(sk, filter);
1801 0 : RCU_INIT_POINTER(sk->sk_filter, NULL);
1802 : }
1803 :
1804 738 : sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1805 :
1806 : #ifdef CONFIG_BPF_SYSCALL
1807 : bpf_sk_storage_free(sk);
1808 : #endif
1809 :
1810 738 : if (atomic_read(&sk->sk_omem_alloc))
1811 : pr_debug("%s: optmem leakage (%d bytes) detected\n",
1812 : __func__, atomic_read(&sk->sk_omem_alloc));
1813 :
1814 738 : if (sk->sk_frag.page) {
1815 0 : put_page(sk->sk_frag.page);
1816 0 : sk->sk_frag.page = NULL;
1817 : }
1818 :
1819 738 : if (sk->sk_peer_cred)
1820 164 : put_cred(sk->sk_peer_cred);
1821 738 : put_pid(sk->sk_peer_pid);
1822 738 : if (likely(sk->sk_net_refcnt))
1823 738 : put_net(sock_net(sk));
1824 738 : sk_prot_free(sk->sk_prot_creator, sk);
1825 737 : }
1826 :
1827 737 : void sk_destruct(struct sock *sk)
1828 : {
1829 737 : bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1830 :
1831 737 : if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1832 0 : reuseport_detach_sock(sk);
1833 0 : use_call_rcu = true;
1834 : }
1835 :
1836 737 : if (use_call_rcu)
1837 33 : call_rcu(&sk->sk_rcu, __sk_destruct);
1838 : else
1839 704 : __sk_destruct(&sk->sk_rcu);
1840 737 : }
1841 :
1842 738 : static void __sk_free(struct sock *sk)
1843 : {
1844 738 : if (likely(sk->sk_net_refcnt))
1845 738 : sock_inuse_add(sock_net(sk), -1);
1846 :
1847 738 : if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1848 0 : sock_diag_broadcast_destroy(sk);
1849 : else
1850 737 : sk_destruct(sk);
1851 737 : }
1852 :
1853 1102 : void sk_free(struct sock *sk)
1854 : {
1855 : /*
1856 : * We subtract one from sk_wmem_alloc and can know if
1857 : * some packets are still in some tx queue.
1858 : * If not null, sock_wfree() will call __sk_free(sk) later
1859 : */
1860 1102 : if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1861 662 : __sk_free(sk);
1862 1101 : }
1863 : EXPORT_SYMBOL(sk_free);
1864 :
1865 837 : static void sk_init_common(struct sock *sk)
1866 : {
1867 837 : skb_queue_head_init(&sk->sk_receive_queue);
1868 837 : skb_queue_head_init(&sk->sk_write_queue);
1869 837 : skb_queue_head_init(&sk->sk_error_queue);
1870 :
1871 837 : rwlock_init(&sk->sk_callback_lock);
1872 837 : lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1873 : af_rlock_keys + sk->sk_family,
1874 : af_family_rlock_key_strings[sk->sk_family]);
1875 837 : lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1876 : af_wlock_keys + sk->sk_family,
1877 : af_family_wlock_key_strings[sk->sk_family]);
1878 837 : lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1879 : af_elock_keys + sk->sk_family,
1880 : af_family_elock_key_strings[sk->sk_family]);
1881 837 : lockdep_set_class_and_name(&sk->sk_callback_lock,
1882 : af_callback_keys + sk->sk_family,
1883 : af_family_clock_key_strings[sk->sk_family]);
1884 837 : }
1885 :
1886 : /**
1887 : * sk_clone_lock - clone a socket, and lock its clone
1888 : * @sk: the socket to clone
1889 : * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1890 : *
1891 : * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1892 : */
1893 4 : struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1894 : {
1895 4 : struct proto *prot = READ_ONCE(sk->sk_prot);
1896 4 : struct sk_filter *filter;
1897 4 : bool is_charged = true;
1898 4 : struct sock *newsk;
1899 :
1900 4 : newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1901 4 : if (!newsk)
1902 0 : goto out;
1903 :
1904 4 : sock_copy(newsk, sk);
1905 :
1906 4 : newsk->sk_prot_creator = prot;
1907 :
1908 : /* SANITY */
1909 4 : if (likely(newsk->sk_net_refcnt))
1910 4 : get_net(sock_net(newsk));
1911 4 : sk_node_init(&newsk->sk_node);
1912 4 : sock_lock_init(newsk);
1913 4 : bh_lock_sock(newsk);
1914 4 : newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1915 4 : newsk->sk_backlog.len = 0;
1916 :
1917 4 : atomic_set(&newsk->sk_rmem_alloc, 0);
1918 :
1919 : /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
1920 4 : refcount_set(&newsk->sk_wmem_alloc, 1);
1921 :
1922 4 : atomic_set(&newsk->sk_omem_alloc, 0);
1923 4 : sk_init_common(newsk);
1924 :
1925 4 : newsk->sk_dst_cache = NULL;
1926 4 : newsk->sk_dst_pending_confirm = 0;
1927 4 : newsk->sk_wmem_queued = 0;
1928 4 : newsk->sk_forward_alloc = 0;
1929 4 : atomic_set(&newsk->sk_drops, 0);
1930 4 : newsk->sk_send_head = NULL;
1931 4 : newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1932 4 : atomic_set(&newsk->sk_zckey, 0);
1933 :
1934 4 : sock_reset_flag(newsk, SOCK_DONE);
1935 :
1936 : /* sk->sk_memcg will be populated at accept() time */
1937 4 : newsk->sk_memcg = NULL;
1938 :
1939 4 : cgroup_sk_clone(&newsk->sk_cgrp_data);
1940 :
1941 4 : rcu_read_lock();
1942 4 : filter = rcu_dereference(sk->sk_filter);
1943 4 : if (filter != NULL)
1944 : /* though it's an empty new sock, the charging may fail
1945 : * if sysctl_optmem_max was changed between creation of
1946 : * original socket and cloning
1947 : */
1948 0 : is_charged = sk_filter_charge(newsk, filter);
1949 4 : RCU_INIT_POINTER(newsk->sk_filter, filter);
1950 4 : rcu_read_unlock();
1951 :
1952 4 : if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1953 : /* We need to make sure that we don't uncharge the new
1954 : * socket if we couldn't charge it in the first place
1955 : * as otherwise we uncharge the parent's filter.
1956 : */
1957 0 : if (!is_charged)
1958 0 : RCU_INIT_POINTER(newsk->sk_filter, NULL);
1959 0 : sk_free_unlock_clone(newsk);
1960 0 : newsk = NULL;
1961 0 : goto out;
1962 : }
1963 4 : RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1964 :
1965 4 : if (bpf_sk_storage_clone(sk, newsk)) {
1966 : sk_free_unlock_clone(newsk);
1967 : newsk = NULL;
1968 : goto out;
1969 : }
1970 :
1971 : /* Clear sk_user_data if parent had the pointer tagged
1972 : * as not suitable for copying when cloning.
1973 : */
1974 4 : if (sk_user_data_is_nocopy(newsk))
1975 0 : newsk->sk_user_data = NULL;
1976 :
1977 4 : newsk->sk_err = 0;
1978 4 : newsk->sk_err_soft = 0;
1979 4 : newsk->sk_priority = 0;
1980 4 : newsk->sk_incoming_cpu = raw_smp_processor_id();
1981 4 : if (likely(newsk->sk_net_refcnt))
1982 4 : sock_inuse_add(sock_net(newsk), 1);
1983 :
1984 : /* Before updating sk_refcnt, we must commit prior changes to memory
1985 : * (Documentation/RCU/rculist_nulls.rst for details)
1986 : */
1987 4 : smp_wmb();
1988 4 : refcount_set(&newsk->sk_refcnt, 2);
1989 :
1990 : /* Increment the counter in the same struct proto as the master
1991 : * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1992 : * is the same as sk->sk_prot->socks, as this field was copied
1993 : * with memcpy).
1994 : *
1995 : * This _changes_ the previous behaviour, where
1996 : * tcp_create_openreq_child always was incrementing the
1997 : * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1998 : * to be taken into account in all callers. -acme
1999 : */
2000 4 : sk_refcnt_debug_inc(newsk);
2001 4 : sk_set_socket(newsk, NULL);
2002 4 : sk_tx_queue_clear(newsk);
2003 4 : RCU_INIT_POINTER(newsk->sk_wq, NULL);
2004 :
2005 4 : if (newsk->sk_prot->sockets_allocated)
2006 4 : sk_sockets_allocated_inc(newsk);
2007 :
2008 4 : if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2009 0 : net_enable_timestamp();
2010 4 : out:
2011 4 : return newsk;
2012 : }
2013 : EXPORT_SYMBOL_GPL(sk_clone_lock);
2014 :
2015 0 : void sk_free_unlock_clone(struct sock *sk)
2016 : {
2017 : /* It is still raw copy of parent, so invalidate
2018 : * destructor and make plain sk_free() */
2019 0 : sk->sk_destruct = NULL;
2020 0 : bh_unlock_sock(sk);
2021 0 : sk_free(sk);
2022 0 : }
2023 : EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2024 :
2025 8 : void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2026 : {
2027 8 : u32 max_segs = 1;
2028 :
2029 8 : sk_dst_set(sk, dst);
2030 8 : sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2031 8 : if (sk->sk_route_caps & NETIF_F_GSO)
2032 8 : sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2033 8 : sk->sk_route_caps &= ~sk->sk_route_nocaps;
2034 8 : if (sk_can_gso(sk)) {
2035 8 : if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2036 0 : sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2037 : } else {
2038 8 : sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2039 8 : sk->sk_gso_max_size = dst->dev->gso_max_size;
2040 8 : max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2041 : }
2042 : }
2043 8 : sk->sk_gso_max_segs = max_segs;
2044 8 : }
2045 : EXPORT_SYMBOL_GPL(sk_setup_caps);
2046 :
2047 : /*
2048 : * Simple resource managers for sockets.
2049 : */
2050 :
2051 :
2052 : /*
2053 : * Write buffer destructor automatically called from kfree_skb.
2054 : */
2055 3005 : void sock_wfree(struct sk_buff *skb)
2056 : {
2057 3005 : struct sock *sk = skb->sk;
2058 3005 : unsigned int len = skb->truesize;
2059 :
2060 3005 : if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2061 : /*
2062 : * Keep a reference on sk_wmem_alloc, this will be released
2063 : * after sk_write_space() call
2064 : */
2065 3006 : WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2066 3006 : sk->sk_write_space(sk);
2067 3006 : len = 1;
2068 : }
2069 : /*
2070 : * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2071 : * could not do because of in-flight packets
2072 : */
2073 3005 : if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2074 76 : __sk_free(sk);
2075 3006 : }
2076 : EXPORT_SYMBOL(sock_wfree);
2077 :
2078 : /* This variant of sock_wfree() is used by TCP,
2079 : * since it sets SOCK_USE_WRITE_QUEUE.
2080 : */
2081 62 : void __sock_wfree(struct sk_buff *skb)
2082 : {
2083 62 : struct sock *sk = skb->sk;
2084 :
2085 62 : if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2086 0 : __sk_free(sk);
2087 62 : }
2088 :
2089 3010 : void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2090 : {
2091 3010 : skb_orphan(skb);
2092 3010 : skb->sk = sk;
2093 : #ifdef CONFIG_INET
2094 3010 : if (unlikely(!sk_fullsock(sk))) {
2095 4 : skb->destructor = sock_edemux;
2096 4 : sock_hold(sk);
2097 4 : return;
2098 : }
2099 : #endif
2100 3006 : skb->destructor = sock_wfree;
2101 3006 : skb_set_hash_from_sk(skb, sk);
2102 : /*
2103 : * We used to take a refcount on sk, but following operation
2104 : * is enough to guarantee sk_free() wont free this sock until
2105 : * all in-flight packets are completed
2106 : */
2107 3006 : refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2108 : }
2109 : EXPORT_SYMBOL(skb_set_owner_w);
2110 :
2111 0 : static bool can_skb_orphan_partial(const struct sk_buff *skb)
2112 : {
2113 : #ifdef CONFIG_TLS_DEVICE
2114 : /* Drivers depend on in-order delivery for crypto offload,
2115 : * partial orphan breaks out-of-order-OK logic.
2116 : */
2117 : if (skb->decrypted)
2118 : return false;
2119 : #endif
2120 0 : return (skb->destructor == sock_wfree ||
2121 : (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2122 : }
2123 :
2124 : /* This helper is used by netem, as it can hold packets in its
2125 : * delay queue. We want to allow the owner socket to send more
2126 : * packets, as if they were already TX completed by a typical driver.
2127 : * But we also want to keep skb->sk set because some packet schedulers
2128 : * rely on it (sch_fq for example).
2129 : */
2130 0 : void skb_orphan_partial(struct sk_buff *skb)
2131 : {
2132 0 : if (skb_is_tcp_pure_ack(skb))
2133 : return;
2134 :
2135 0 : if (can_skb_orphan_partial(skb)) {
2136 0 : struct sock *sk = skb->sk;
2137 :
2138 0 : if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2139 0 : WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2140 0 : skb->destructor = sock_efree;
2141 : }
2142 : } else {
2143 0 : skb_orphan(skb);
2144 : }
2145 : }
2146 : EXPORT_SYMBOL(skb_orphan_partial);
2147 :
2148 : /*
2149 : * Read buffer destructor automatically called from kfree_skb.
2150 : */
2151 752 : void sock_rfree(struct sk_buff *skb)
2152 : {
2153 752 : struct sock *sk = skb->sk;
2154 752 : unsigned int len = skb->truesize;
2155 :
2156 752 : atomic_sub(len, &sk->sk_rmem_alloc);
2157 752 : sk_mem_uncharge(sk, len);
2158 752 : }
2159 : EXPORT_SYMBOL(sock_rfree);
2160 :
2161 : /*
2162 : * Buffer destructor for skbs that are not used directly in read or write
2163 : * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2164 : */
2165 0 : void sock_efree(struct sk_buff *skb)
2166 : {
2167 0 : sock_put(skb->sk);
2168 0 : }
2169 : EXPORT_SYMBOL(sock_efree);
2170 :
2171 : /* Buffer destructor for prefetch/receive path where reference count may
2172 : * not be held, e.g. for listen sockets.
2173 : */
2174 : #ifdef CONFIG_INET
2175 0 : void sock_pfree(struct sk_buff *skb)
2176 : {
2177 0 : if (sk_is_refcounted(skb->sk))
2178 0 : sock_gen_put(skb->sk);
2179 0 : }
2180 : EXPORT_SYMBOL(sock_pfree);
2181 : #endif /* CONFIG_INET */
2182 :
2183 47 : kuid_t sock_i_uid(struct sock *sk)
2184 : {
2185 47 : kuid_t uid;
2186 :
2187 47 : read_lock_bh(&sk->sk_callback_lock);
2188 47 : uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2189 47 : read_unlock_bh(&sk->sk_callback_lock);
2190 47 : return uid;
2191 : }
2192 : EXPORT_SYMBOL(sock_i_uid);
2193 :
2194 0 : unsigned long sock_i_ino(struct sock *sk)
2195 : {
2196 0 : unsigned long ino;
2197 :
2198 0 : read_lock_bh(&sk->sk_callback_lock);
2199 0 : ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2200 0 : read_unlock_bh(&sk->sk_callback_lock);
2201 0 : return ino;
2202 : }
2203 : EXPORT_SYMBOL(sock_i_ino);
2204 :
2205 : /*
2206 : * Allocate a skb from the socket's send buffer.
2207 : */
2208 279 : struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2209 : gfp_t priority)
2210 : {
2211 279 : if (force ||
2212 279 : refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2213 279 : struct sk_buff *skb = alloc_skb(size, priority);
2214 :
2215 279 : if (skb) {
2216 279 : skb_set_owner_w(skb, sk);
2217 279 : return skb;
2218 : }
2219 : }
2220 : return NULL;
2221 : }
2222 : EXPORT_SYMBOL(sock_wmalloc);
2223 :
2224 0 : static void sock_ofree(struct sk_buff *skb)
2225 : {
2226 0 : struct sock *sk = skb->sk;
2227 :
2228 0 : atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2229 0 : }
2230 :
2231 0 : struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2232 : gfp_t priority)
2233 : {
2234 0 : struct sk_buff *skb;
2235 :
2236 : /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2237 0 : if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2238 : sysctl_optmem_max)
2239 : return NULL;
2240 :
2241 0 : skb = alloc_skb(size, priority);
2242 0 : if (!skb)
2243 : return NULL;
2244 :
2245 0 : atomic_add(skb->truesize, &sk->sk_omem_alloc);
2246 0 : skb->sk = sk;
2247 0 : skb->destructor = sock_ofree;
2248 0 : return skb;
2249 : }
2250 :
2251 : /*
2252 : * Allocate a memory block from the socket's option memory buffer.
2253 : */
2254 0 : void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2255 : {
2256 0 : if ((unsigned int)size <= sysctl_optmem_max &&
2257 0 : atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2258 0 : void *mem;
2259 : /* First do the add, to avoid the race if kmalloc
2260 : * might sleep.
2261 : */
2262 0 : atomic_add(size, &sk->sk_omem_alloc);
2263 0 : mem = kmalloc(size, priority);
2264 0 : if (mem)
2265 0 : return mem;
2266 0 : atomic_sub(size, &sk->sk_omem_alloc);
2267 : }
2268 : return NULL;
2269 : }
2270 : EXPORT_SYMBOL(sock_kmalloc);
2271 :
2272 : /* Free an option memory block. Note, we actually want the inline
2273 : * here as this allows gcc to detect the nullify and fold away the
2274 : * condition entirely.
2275 : */
2276 0 : static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2277 : const bool nullify)
2278 : {
2279 0 : if (WARN_ON_ONCE(!mem))
2280 : return;
2281 0 : if (nullify)
2282 0 : kfree_sensitive(mem);
2283 : else
2284 0 : kfree(mem);
2285 0 : atomic_sub(size, &sk->sk_omem_alloc);
2286 : }
2287 :
2288 0 : void sock_kfree_s(struct sock *sk, void *mem, int size)
2289 : {
2290 0 : __sock_kfree_s(sk, mem, size, false);
2291 0 : }
2292 : EXPORT_SYMBOL(sock_kfree_s);
2293 :
2294 0 : void sock_kzfree_s(struct sock *sk, void *mem, int size)
2295 : {
2296 0 : __sock_kfree_s(sk, mem, size, true);
2297 0 : }
2298 : EXPORT_SYMBOL(sock_kzfree_s);
2299 :
2300 : /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2301 : I think, these locks should be removed for datagram sockets.
2302 : */
2303 0 : static long sock_wait_for_wmem(struct sock *sk, long timeo)
2304 : {
2305 0 : DEFINE_WAIT(wait);
2306 :
2307 0 : sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2308 0 : for (;;) {
2309 0 : if (!timeo)
2310 : break;
2311 0 : if (signal_pending(current))
2312 : break;
2313 0 : set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2314 0 : prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2315 0 : if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2316 : break;
2317 0 : if (sk->sk_shutdown & SEND_SHUTDOWN)
2318 : break;
2319 0 : if (sk->sk_err)
2320 : break;
2321 0 : timeo = schedule_timeout(timeo);
2322 : }
2323 0 : finish_wait(sk_sleep(sk), &wait);
2324 0 : return timeo;
2325 : }
2326 :
2327 :
2328 : /*
2329 : * Generic send/receive buffer handlers
2330 : */
2331 :
2332 2727 : struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2333 : unsigned long data_len, int noblock,
2334 : int *errcode, int max_page_order)
2335 : {
2336 2727 : struct sk_buff *skb;
2337 2727 : long timeo;
2338 2727 : int err;
2339 :
2340 2727 : timeo = sock_sndtimeo(sk, noblock);
2341 2727 : for (;;) {
2342 2727 : err = sock_error(sk);
2343 2727 : if (err != 0)
2344 0 : goto failure;
2345 :
2346 2727 : err = -EPIPE;
2347 2727 : if (sk->sk_shutdown & SEND_SHUTDOWN)
2348 0 : goto failure;
2349 :
2350 2727 : if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2351 : break;
2352 :
2353 0 : sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2354 0 : set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2355 0 : err = -EAGAIN;
2356 0 : if (!timeo)
2357 0 : goto failure;
2358 0 : if (signal_pending(current))
2359 0 : goto interrupted;
2360 0 : timeo = sock_wait_for_wmem(sk, timeo);
2361 : }
2362 2727 : skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2363 : errcode, sk->sk_allocation);
2364 2727 : if (skb)
2365 2727 : skb_set_owner_w(skb, sk);
2366 : return skb;
2367 :
2368 0 : interrupted:
2369 0 : err = sock_intr_errno(timeo);
2370 0 : failure:
2371 0 : *errcode = err;
2372 0 : return NULL;
2373 : }
2374 : EXPORT_SYMBOL(sock_alloc_send_pskb);
2375 :
2376 13 : struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2377 : int noblock, int *errcode)
2378 : {
2379 13 : return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2380 : }
2381 : EXPORT_SYMBOL(sock_alloc_send_skb);
2382 :
2383 0 : int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2384 : struct sockcm_cookie *sockc)
2385 : {
2386 0 : u32 tsflags;
2387 :
2388 0 : switch (cmsg->cmsg_type) {
2389 : case SO_MARK:
2390 0 : if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2391 : return -EPERM;
2392 0 : if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2393 : return -EINVAL;
2394 0 : sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2395 0 : break;
2396 0 : case SO_TIMESTAMPING_OLD:
2397 0 : if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2398 : return -EINVAL;
2399 :
2400 0 : tsflags = *(u32 *)CMSG_DATA(cmsg);
2401 0 : if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2402 : return -EINVAL;
2403 :
2404 0 : sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2405 0 : sockc->tsflags |= tsflags;
2406 0 : break;
2407 : case SCM_TXTIME:
2408 0 : if (!sock_flag(sk, SOCK_TXTIME))
2409 : return -EINVAL;
2410 0 : if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2411 : return -EINVAL;
2412 0 : sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2413 0 : break;
2414 : /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2415 : case SCM_RIGHTS:
2416 : case SCM_CREDENTIALS:
2417 : break;
2418 : default:
2419 : return -EINVAL;
2420 : }
2421 : return 0;
2422 : }
2423 : EXPORT_SYMBOL(__sock_cmsg_send);
2424 :
2425 0 : int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2426 : struct sockcm_cookie *sockc)
2427 : {
2428 0 : struct cmsghdr *cmsg;
2429 0 : int ret;
2430 :
2431 0 : for_each_cmsghdr(cmsg, msg) {
2432 0 : if (!CMSG_OK(msg, cmsg))
2433 : return -EINVAL;
2434 0 : if (cmsg->cmsg_level != SOL_SOCKET)
2435 0 : continue;
2436 0 : ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2437 0 : if (ret)
2438 0 : return ret;
2439 : }
2440 : return 0;
2441 : }
2442 : EXPORT_SYMBOL(sock_cmsg_send);
2443 :
2444 0 : static void sk_enter_memory_pressure(struct sock *sk)
2445 : {
2446 0 : if (!sk->sk_prot->enter_memory_pressure)
2447 : return;
2448 :
2449 0 : sk->sk_prot->enter_memory_pressure(sk);
2450 : }
2451 :
2452 14 : static void sk_leave_memory_pressure(struct sock *sk)
2453 : {
2454 14 : if (sk->sk_prot->leave_memory_pressure) {
2455 13 : sk->sk_prot->leave_memory_pressure(sk);
2456 : } else {
2457 1 : unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2458 :
2459 1 : if (memory_pressure && READ_ONCE(*memory_pressure))
2460 0 : WRITE_ONCE(*memory_pressure, 0);
2461 : }
2462 14 : }
2463 :
2464 : #define SKB_FRAG_PAGE_ORDER get_order(32768)
2465 : DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2466 :
2467 : /**
2468 : * skb_page_frag_refill - check that a page_frag contains enough room
2469 : * @sz: minimum size of the fragment we want to get
2470 : * @pfrag: pointer to page_frag
2471 : * @gfp: priority for memory allocation
2472 : *
2473 : * Note: While this allocator tries to use high order pages, there is
2474 : * no guarantee that allocations succeed. Therefore, @sz MUST be
2475 : * less or equal than PAGE_SIZE.
2476 : */
2477 2133 : bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2478 : {
2479 2133 : if (pfrag->page) {
2480 2124 : if (page_ref_count(pfrag->page) == 1) {
2481 209 : pfrag->offset = 0;
2482 209 : return true;
2483 : }
2484 1915 : if (pfrag->offset + sz <= pfrag->size)
2485 : return true;
2486 81 : put_page(pfrag->page);
2487 : }
2488 :
2489 90 : pfrag->offset = 0;
2490 90 : if (SKB_FRAG_PAGE_ORDER &&
2491 90 : !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2492 : /* Avoid direct reclaim but allow kswapd to wake */
2493 90 : pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2494 90 : __GFP_COMP | __GFP_NOWARN |
2495 : __GFP_NORETRY,
2496 90 : SKB_FRAG_PAGE_ORDER);
2497 90 : if (likely(pfrag->page)) {
2498 90 : pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2499 90 : return true;
2500 : }
2501 : }
2502 0 : pfrag->page = alloc_page(gfp);
2503 0 : if (likely(pfrag->page)) {
2504 0 : pfrag->size = PAGE_SIZE;
2505 0 : return true;
2506 : }
2507 : return false;
2508 : }
2509 : EXPORT_SYMBOL(skb_page_frag_refill);
2510 :
2511 411 : bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2512 : {
2513 411 : if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2514 : return true;
2515 :
2516 0 : sk_enter_memory_pressure(sk);
2517 0 : sk_stream_moderate_sndbuf(sk);
2518 : return false;
2519 : }
2520 : EXPORT_SYMBOL(sk_page_frag_refill);
2521 :
2522 0 : void __lock_sock(struct sock *sk)
2523 : __releases(&sk->sk_lock.slock)
2524 : __acquires(&sk->sk_lock.slock)
2525 : {
2526 0 : DEFINE_WAIT(wait);
2527 :
2528 0 : for (;;) {
2529 0 : prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2530 : TASK_UNINTERRUPTIBLE);
2531 0 : spin_unlock_bh(&sk->sk_lock.slock);
2532 0 : schedule();
2533 0 : spin_lock_bh(&sk->sk_lock.slock);
2534 0 : if (!sock_owned_by_user(sk))
2535 : break;
2536 : }
2537 0 : finish_wait(&sk->sk_lock.wq, &wait);
2538 0 : }
2539 :
2540 109 : void __release_sock(struct sock *sk)
2541 : __releases(&sk->sk_lock.slock)
2542 : __acquires(&sk->sk_lock.slock)
2543 : {
2544 109 : struct sk_buff *skb, *next;
2545 :
2546 109 : while ((skb = sk->sk_backlog.head) != NULL) {
2547 141 : sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2548 :
2549 141 : spin_unlock_bh(&sk->sk_lock.slock);
2550 :
2551 142 : do {
2552 142 : next = skb->next;
2553 142 : prefetch(next);
2554 142 : WARN_ON_ONCE(skb_dst_is_noref(skb));
2555 142 : skb_mark_not_on_list(skb);
2556 142 : sk_backlog_rcv(sk, skb);
2557 :
2558 142 : cond_resched();
2559 :
2560 142 : skb = next;
2561 142 : } while (skb != NULL);
2562 :
2563 391 : spin_lock_bh(&sk->sk_lock.slock);
2564 : }
2565 :
2566 : /*
2567 : * Doing the zeroing here guarantee we can not loop forever
2568 : * while a wild producer attempts to flood us.
2569 : */
2570 109 : sk->sk_backlog.len = 0;
2571 109 : }
2572 :
2573 0 : void __sk_flush_backlog(struct sock *sk)
2574 : {
2575 0 : spin_lock_bh(&sk->sk_lock.slock);
2576 0 : __release_sock(sk);
2577 0 : spin_unlock_bh(&sk->sk_lock.slock);
2578 0 : }
2579 :
2580 : /**
2581 : * sk_wait_data - wait for data to arrive at sk_receive_queue
2582 : * @sk: sock to wait on
2583 : * @timeo: for how long
2584 : * @skb: last skb seen on sk_receive_queue
2585 : *
2586 : * Now socket state including sk->sk_err is changed only under lock,
2587 : * hence we may omit checks after joining wait queue.
2588 : * We check receive queue before schedule() only as optimization;
2589 : * it is very likely that release_sock() added new data.
2590 : */
2591 0 : int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2592 : {
2593 0 : DEFINE_WAIT_FUNC(wait, woken_wake_function);
2594 0 : int rc;
2595 :
2596 0 : add_wait_queue(sk_sleep(sk), &wait);
2597 0 : sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2598 0 : rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2599 0 : sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2600 0 : remove_wait_queue(sk_sleep(sk), &wait);
2601 0 : return rc;
2602 : }
2603 : EXPORT_SYMBOL(sk_wait_data);
2604 :
2605 : /**
2606 : * __sk_mem_raise_allocated - increase memory_allocated
2607 : * @sk: socket
2608 : * @size: memory size to allocate
2609 : * @amt: pages to allocate
2610 : * @kind: allocation type
2611 : *
2612 : * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2613 : */
2614 14 : int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2615 : {
2616 14 : struct proto *prot = sk->sk_prot;
2617 14 : long allocated = sk_memory_allocated_add(sk, amt);
2618 14 : bool charged = true;
2619 :
2620 14 : if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2621 : !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2622 : goto suppress_allocation;
2623 :
2624 : /* Under limit. */
2625 14 : if (allocated <= sk_prot_mem_limits(sk, 0)) {
2626 14 : sk_leave_memory_pressure(sk);
2627 14 : return 1;
2628 : }
2629 :
2630 : /* Under pressure. */
2631 0 : if (allocated > sk_prot_mem_limits(sk, 1))
2632 0 : sk_enter_memory_pressure(sk);
2633 :
2634 : /* Over hard limit. */
2635 0 : if (allocated > sk_prot_mem_limits(sk, 2))
2636 0 : goto suppress_allocation;
2637 :
2638 : /* guarantee minimum buffer size under pressure */
2639 0 : if (kind == SK_MEM_RECV) {
2640 0 : if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2641 : return 1;
2642 :
2643 : } else { /* SK_MEM_SEND */
2644 0 : int wmem0 = sk_get_wmem0(sk, prot);
2645 :
2646 0 : if (sk->sk_type == SOCK_STREAM) {
2647 0 : if (sk->sk_wmem_queued < wmem0)
2648 : return 1;
2649 0 : } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2650 : return 1;
2651 : }
2652 : }
2653 :
2654 0 : if (sk_has_memory_pressure(sk)) {
2655 0 : u64 alloc;
2656 :
2657 0 : if (!sk_under_memory_pressure(sk))
2658 : return 1;
2659 0 : alloc = sk_sockets_allocated_read_positive(sk);
2660 0 : if (sk_prot_mem_limits(sk, 2) > alloc *
2661 0 : sk_mem_pages(sk->sk_wmem_queued +
2662 0 : atomic_read(&sk->sk_rmem_alloc) +
2663 0 : sk->sk_forward_alloc))
2664 : return 1;
2665 : }
2666 :
2667 0 : suppress_allocation:
2668 :
2669 0 : if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2670 0 : sk_stream_moderate_sndbuf(sk);
2671 :
2672 : /* Fail only if socket is _under_ its sndbuf.
2673 : * In this case we cannot block, so that we have to fail.
2674 : */
2675 0 : if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2676 : return 1;
2677 : }
2678 :
2679 0 : if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2680 0 : trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2681 :
2682 0 : sk_memory_allocated_sub(sk, amt);
2683 :
2684 0 : if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2685 : mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2686 :
2687 0 : return 0;
2688 : }
2689 : EXPORT_SYMBOL(__sk_mem_raise_allocated);
2690 :
2691 : /**
2692 : * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2693 : * @sk: socket
2694 : * @size: memory size to allocate
2695 : * @kind: allocation type
2696 : *
2697 : * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2698 : * rmem allocation. This function assumes that protocols which have
2699 : * memory_pressure use sk_wmem_queued as write buffer accounting.
2700 : */
2701 13 : int __sk_mem_schedule(struct sock *sk, int size, int kind)
2702 : {
2703 13 : int ret, amt = sk_mem_pages(size);
2704 :
2705 13 : sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2706 13 : ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2707 13 : if (!ret)
2708 0 : sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2709 13 : return ret;
2710 : }
2711 : EXPORT_SYMBOL(__sk_mem_schedule);
2712 :
2713 : /**
2714 : * __sk_mem_reduce_allocated - reclaim memory_allocated
2715 : * @sk: socket
2716 : * @amount: number of quanta
2717 : *
2718 : * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2719 : */
2720 17 : void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2721 : {
2722 17 : sk_memory_allocated_sub(sk, amount);
2723 :
2724 17 : if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2725 : mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2726 :
2727 17 : if (sk_under_memory_pressure(sk) &&
2728 0 : (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2729 0 : sk_leave_memory_pressure(sk);
2730 17 : }
2731 : EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2732 :
2733 : /**
2734 : * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2735 : * @sk: socket
2736 : * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2737 : */
2738 17 : void __sk_mem_reclaim(struct sock *sk, int amount)
2739 : {
2740 17 : amount >>= SK_MEM_QUANTUM_SHIFT;
2741 17 : sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2742 17 : __sk_mem_reduce_allocated(sk, amount);
2743 0 : }
2744 : EXPORT_SYMBOL(__sk_mem_reclaim);
2745 :
2746 0 : int sk_set_peek_off(struct sock *sk, int val)
2747 : {
2748 0 : sk->sk_peek_off = val;
2749 0 : return 0;
2750 : }
2751 : EXPORT_SYMBOL_GPL(sk_set_peek_off);
2752 :
2753 : /*
2754 : * Set of default routines for initialising struct proto_ops when
2755 : * the protocol does not support a particular function. In certain
2756 : * cases where it makes no sense for a protocol to have a "do nothing"
2757 : * function, some default processing is provided.
2758 : */
2759 :
2760 0 : int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2761 : {
2762 0 : return -EOPNOTSUPP;
2763 : }
2764 : EXPORT_SYMBOL(sock_no_bind);
2765 :
2766 0 : int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2767 : int len, int flags)
2768 : {
2769 0 : return -EOPNOTSUPP;
2770 : }
2771 : EXPORT_SYMBOL(sock_no_connect);
2772 :
2773 0 : int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2774 : {
2775 0 : return -EOPNOTSUPP;
2776 : }
2777 : EXPORT_SYMBOL(sock_no_socketpair);
2778 :
2779 0 : int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2780 : bool kern)
2781 : {
2782 0 : return -EOPNOTSUPP;
2783 : }
2784 : EXPORT_SYMBOL(sock_no_accept);
2785 :
2786 0 : int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2787 : int peer)
2788 : {
2789 0 : return -EOPNOTSUPP;
2790 : }
2791 : EXPORT_SYMBOL(sock_no_getname);
2792 :
2793 0 : int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2794 : {
2795 0 : return -EOPNOTSUPP;
2796 : }
2797 : EXPORT_SYMBOL(sock_no_ioctl);
2798 :
2799 0 : int sock_no_listen(struct socket *sock, int backlog)
2800 : {
2801 0 : return -EOPNOTSUPP;
2802 : }
2803 : EXPORT_SYMBOL(sock_no_listen);
2804 :
2805 0 : int sock_no_shutdown(struct socket *sock, int how)
2806 : {
2807 0 : return -EOPNOTSUPP;
2808 : }
2809 : EXPORT_SYMBOL(sock_no_shutdown);
2810 :
2811 0 : int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2812 : {
2813 0 : return -EOPNOTSUPP;
2814 : }
2815 : EXPORT_SYMBOL(sock_no_sendmsg);
2816 :
2817 0 : int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2818 : {
2819 0 : return -EOPNOTSUPP;
2820 : }
2821 : EXPORT_SYMBOL(sock_no_sendmsg_locked);
2822 :
2823 0 : int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2824 : int flags)
2825 : {
2826 0 : return -EOPNOTSUPP;
2827 : }
2828 : EXPORT_SYMBOL(sock_no_recvmsg);
2829 :
2830 0 : int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2831 : {
2832 : /* Mirror missing mmap method error code */
2833 0 : return -ENODEV;
2834 : }
2835 : EXPORT_SYMBOL(sock_no_mmap);
2836 :
2837 : /*
2838 : * When a file is received (via SCM_RIGHTS, etc), we must bump the
2839 : * various sock-based usage counts.
2840 : */
2841 54 : void __receive_sock(struct file *file)
2842 : {
2843 54 : struct socket *sock;
2844 :
2845 54 : sock = sock_from_file(file);
2846 54 : if (sock) {
2847 : sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2848 54 : sock_update_classid(&sock->sk->sk_cgrp_data);
2849 : }
2850 54 : }
2851 :
2852 0 : ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2853 : {
2854 0 : ssize_t res;
2855 0 : struct msghdr msg = {.msg_flags = flags};
2856 0 : struct kvec iov;
2857 0 : char *kaddr = kmap(page);
2858 0 : iov.iov_base = kaddr + offset;
2859 0 : iov.iov_len = size;
2860 0 : res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2861 0 : kunmap(page);
2862 0 : return res;
2863 : }
2864 : EXPORT_SYMBOL(sock_no_sendpage);
2865 :
2866 0 : ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2867 : int offset, size_t size, int flags)
2868 : {
2869 0 : ssize_t res;
2870 0 : struct msghdr msg = {.msg_flags = flags};
2871 0 : struct kvec iov;
2872 0 : char *kaddr = kmap(page);
2873 :
2874 0 : iov.iov_base = kaddr + offset;
2875 0 : iov.iov_len = size;
2876 0 : res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2877 0 : kunmap(page);
2878 0 : return res;
2879 : }
2880 : EXPORT_SYMBOL(sock_no_sendpage_locked);
2881 :
2882 : /*
2883 : * Default Socket Callbacks
2884 : */
2885 :
2886 352 : static void sock_def_wakeup(struct sock *sk)
2887 : {
2888 352 : struct socket_wq *wq;
2889 :
2890 352 : rcu_read_lock();
2891 352 : wq = rcu_dereference(sk->sk_wq);
2892 613 : if (skwq_has_sleeper(wq))
2893 81 : wake_up_interruptible_all(&wq->wait);
2894 352 : rcu_read_unlock();
2895 352 : }
2896 :
2897 0 : static void sock_def_error_report(struct sock *sk)
2898 : {
2899 0 : struct socket_wq *wq;
2900 :
2901 0 : rcu_read_lock();
2902 0 : wq = rcu_dereference(sk->sk_wq);
2903 0 : if (skwq_has_sleeper(wq))
2904 0 : wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2905 0 : sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2906 0 : rcu_read_unlock();
2907 0 : }
2908 :
2909 3330 : void sock_def_readable(struct sock *sk)
2910 : {
2911 3330 : struct socket_wq *wq;
2912 :
2913 3330 : rcu_read_lock();
2914 3330 : wq = rcu_dereference(sk->sk_wq);
2915 6656 : if (skwq_has_sleeper(wq))
2916 2868 : wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2917 : EPOLLRDNORM | EPOLLRDBAND);
2918 3330 : sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2919 3330 : rcu_read_unlock();
2920 3330 : }
2921 :
2922 21 : static void sock_def_write_space(struct sock *sk)
2923 : {
2924 21 : struct socket_wq *wq;
2925 :
2926 21 : rcu_read_lock();
2927 :
2928 : /* Do not wake up a writer until he can make "significant"
2929 : * progress. --DaveM
2930 : */
2931 21 : if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2932 21 : wq = rcu_dereference(sk->sk_wq);
2933 42 : if (skwq_has_sleeper(wq))
2934 14 : wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2935 : EPOLLWRNORM | EPOLLWRBAND);
2936 :
2937 : /* Should agree with poll, otherwise some programs break */
2938 21 : if (sock_writeable(sk))
2939 21 : sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2940 : }
2941 :
2942 21 : rcu_read_unlock();
2943 21 : }
2944 :
2945 0 : static void sock_def_destruct(struct sock *sk)
2946 : {
2947 0 : }
2948 :
2949 0 : void sk_send_sigurg(struct sock *sk)
2950 : {
2951 0 : if (sk->sk_socket && sk->sk_socket->file)
2952 0 : if (send_sigurg(&sk->sk_socket->file->f_owner))
2953 0 : sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2954 0 : }
2955 : EXPORT_SYMBOL(sk_send_sigurg);
2956 :
2957 369 : void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2958 : unsigned long expires)
2959 : {
2960 369 : if (!mod_timer(timer, expires))
2961 20 : sock_hold(sk);
2962 369 : }
2963 : EXPORT_SYMBOL(sk_reset_timer);
2964 :
2965 21 : void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2966 : {
2967 21 : if (del_timer(timer))
2968 9 : __sock_put(sk);
2969 21 : }
2970 : EXPORT_SYMBOL(sk_stop_timer);
2971 :
2972 0 : void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
2973 : {
2974 0 : if (del_timer_sync(timer))
2975 0 : __sock_put(sk);
2976 0 : }
2977 : EXPORT_SYMBOL(sk_stop_timer_sync);
2978 :
2979 833 : void sock_init_data(struct socket *sock, struct sock *sk)
2980 : {
2981 833 : sk_init_common(sk);
2982 833 : sk->sk_send_head = NULL;
2983 :
2984 833 : timer_setup(&sk->sk_timer, NULL, 0);
2985 :
2986 833 : sk->sk_allocation = GFP_KERNEL;
2987 833 : sk->sk_rcvbuf = sysctl_rmem_default;
2988 833 : sk->sk_sndbuf = sysctl_wmem_default;
2989 833 : sk->sk_state = TCP_CLOSE;
2990 833 : sk_set_socket(sk, sock);
2991 :
2992 833 : sock_set_flag(sk, SOCK_ZAPPED);
2993 :
2994 833 : if (sock) {
2995 554 : sk->sk_type = sock->type;
2996 554 : RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2997 554 : sock->sk = sk;
2998 554 : sk->sk_uid = SOCK_INODE(sock)->i_uid;
2999 : } else {
3000 279 : RCU_INIT_POINTER(sk->sk_wq, NULL);
3001 279 : sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3002 : }
3003 :
3004 833 : rwlock_init(&sk->sk_callback_lock);
3005 833 : if (sk->sk_kern_sock)
3006 13 : lockdep_set_class_and_name(
3007 : &sk->sk_callback_lock,
3008 : af_kern_callback_keys + sk->sk_family,
3009 : af_family_kern_clock_key_strings[sk->sk_family]);
3010 : else
3011 820 : lockdep_set_class_and_name(
3012 : &sk->sk_callback_lock,
3013 : af_callback_keys + sk->sk_family,
3014 : af_family_clock_key_strings[sk->sk_family]);
3015 :
3016 833 : sk->sk_state_change = sock_def_wakeup;
3017 833 : sk->sk_data_ready = sock_def_readable;
3018 833 : sk->sk_write_space = sock_def_write_space;
3019 833 : sk->sk_error_report = sock_def_error_report;
3020 833 : sk->sk_destruct = sock_def_destruct;
3021 :
3022 833 : sk->sk_frag.page = NULL;
3023 833 : sk->sk_frag.offset = 0;
3024 833 : sk->sk_peek_off = -1;
3025 :
3026 833 : sk->sk_peer_pid = NULL;
3027 833 : sk->sk_peer_cred = NULL;
3028 833 : sk->sk_write_pending = 0;
3029 833 : sk->sk_rcvlowat = 1;
3030 833 : sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3031 833 : sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3032 :
3033 833 : sk->sk_stamp = SK_DEFAULT_STAMP;
3034 : #if BITS_PER_LONG==32
3035 : seqlock_init(&sk->sk_stamp_seq);
3036 : #endif
3037 833 : atomic_set(&sk->sk_zckey, 0);
3038 :
3039 : #ifdef CONFIG_NET_RX_BUSY_POLL
3040 833 : sk->sk_napi_id = 0;
3041 833 : sk->sk_ll_usec = sysctl_net_busy_read;
3042 : #endif
3043 :
3044 833 : sk->sk_max_pacing_rate = ~0UL;
3045 833 : sk->sk_pacing_rate = ~0UL;
3046 833 : WRITE_ONCE(sk->sk_pacing_shift, 10);
3047 833 : sk->sk_incoming_cpu = -1;
3048 :
3049 833 : sk_rx_queue_clear(sk);
3050 : /*
3051 : * Before updating sk_refcnt, we must commit prior changes to memory
3052 : * (Documentation/RCU/rculist_nulls.rst for details)
3053 : */
3054 833 : smp_wmb();
3055 833 : refcount_set(&sk->sk_refcnt, 1);
3056 833 : atomic_set(&sk->sk_drops, 0);
3057 833 : }
3058 : EXPORT_SYMBOL(sock_init_data);
3059 :
3060 1121 : void lock_sock_nested(struct sock *sk, int subclass)
3061 : {
3062 1121 : might_sleep();
3063 1121 : spin_lock_bh(&sk->sk_lock.slock);
3064 1121 : if (sk->sk_lock.owned)
3065 0 : __lock_sock(sk);
3066 1121 : sk->sk_lock.owned = 1;
3067 1121 : spin_unlock(&sk->sk_lock.slock);
3068 : /*
3069 : * The sk_lock has mutex_lock() semantics here:
3070 : */
3071 1121 : mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3072 1121 : local_bh_enable();
3073 1121 : }
3074 : EXPORT_SYMBOL(lock_sock_nested);
3075 :
3076 1121 : void release_sock(struct sock *sk)
3077 : {
3078 1121 : spin_lock_bh(&sk->sk_lock.slock);
3079 1121 : if (sk->sk_backlog.tail)
3080 105 : __release_sock(sk);
3081 :
3082 : /* Warning : release_cb() might need to release sk ownership,
3083 : * ie call sock_release_ownership(sk) before us.
3084 : */
3085 1121 : if (sk->sk_prot->release_cb)
3086 722 : sk->sk_prot->release_cb(sk);
3087 :
3088 1121 : sock_release_ownership(sk);
3089 1121 : if (waitqueue_active(&sk->sk_lock.wq))
3090 0 : wake_up(&sk->sk_lock.wq);
3091 1121 : spin_unlock_bh(&sk->sk_lock.slock);
3092 1121 : }
3093 : EXPORT_SYMBOL(release_sock);
3094 :
3095 : /**
3096 : * lock_sock_fast - fast version of lock_sock
3097 : * @sk: socket
3098 : *
3099 : * This version should be used for very small section, where process wont block
3100 : * return false if fast path is taken:
3101 : *
3102 : * sk_lock.slock locked, owned = 0, BH disabled
3103 : *
3104 : * return true if slow path is taken:
3105 : *
3106 : * sk_lock.slock unlocked, owned = 1, BH enabled
3107 : */
3108 39 : bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3109 : {
3110 39 : might_sleep();
3111 39 : spin_lock_bh(&sk->sk_lock.slock);
3112 :
3113 39 : if (!sk->sk_lock.owned)
3114 : /*
3115 : * Note : We must disable BH
3116 : */
3117 : return false;
3118 :
3119 0 : __lock_sock(sk);
3120 0 : sk->sk_lock.owned = 1;
3121 0 : spin_unlock(&sk->sk_lock.slock);
3122 : /*
3123 : * The sk_lock has mutex_lock() semantics here:
3124 : */
3125 0 : mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3126 0 : __acquire(&sk->sk_lock.slock);
3127 0 : local_bh_enable();
3128 0 : return true;
3129 : }
3130 : EXPORT_SYMBOL(lock_sock_fast);
3131 :
3132 0 : int sock_gettstamp(struct socket *sock, void __user *userstamp,
3133 : bool timeval, bool time32)
3134 : {
3135 0 : struct sock *sk = sock->sk;
3136 0 : struct timespec64 ts;
3137 :
3138 0 : sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3139 0 : ts = ktime_to_timespec64(sock_read_timestamp(sk));
3140 0 : if (ts.tv_sec == -1)
3141 : return -ENOENT;
3142 0 : if (ts.tv_sec == 0) {
3143 0 : ktime_t kt = ktime_get_real();
3144 0 : sock_write_timestamp(sk, kt);
3145 0 : ts = ktime_to_timespec64(kt);
3146 : }
3147 :
3148 0 : if (timeval)
3149 0 : ts.tv_nsec /= 1000;
3150 :
3151 : #ifdef CONFIG_COMPAT_32BIT_TIME
3152 : if (time32)
3153 : return put_old_timespec32(&ts, userstamp);
3154 : #endif
3155 : #ifdef CONFIG_SPARC64
3156 : /* beware of padding in sparc64 timeval */
3157 : if (timeval && !in_compat_syscall()) {
3158 : struct __kernel_old_timeval __user tv = {
3159 : .tv_sec = ts.tv_sec,
3160 : .tv_usec = ts.tv_nsec,
3161 : };
3162 : if (copy_to_user(userstamp, &tv, sizeof(tv)))
3163 : return -EFAULT;
3164 : return 0;
3165 : }
3166 : #endif
3167 0 : return put_timespec64(&ts, userstamp);
3168 : }
3169 : EXPORT_SYMBOL(sock_gettstamp);
3170 :
3171 3 : void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3172 : {
3173 3 : if (!sock_flag(sk, flag)) {
3174 3 : unsigned long previous_flags = sk->sk_flags;
3175 :
3176 3 : sock_set_flag(sk, flag);
3177 : /*
3178 : * we just set one of the two flags which require net
3179 : * time stamping, but time stamping might have been on
3180 : * already because of the other one
3181 : */
3182 3 : if (sock_needs_netstamp(sk) &&
3183 0 : !(previous_flags & SK_FLAGS_TIMESTAMP))
3184 0 : net_enable_timestamp();
3185 : }
3186 3 : }
3187 :
3188 0 : int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3189 : int level, int type)
3190 : {
3191 0 : struct sock_exterr_skb *serr;
3192 0 : struct sk_buff *skb;
3193 0 : int copied, err;
3194 :
3195 0 : err = -EAGAIN;
3196 0 : skb = sock_dequeue_err_skb(sk);
3197 0 : if (skb == NULL)
3198 0 : goto out;
3199 :
3200 0 : copied = skb->len;
3201 0 : if (copied > len) {
3202 0 : msg->msg_flags |= MSG_TRUNC;
3203 0 : copied = len;
3204 : }
3205 0 : err = skb_copy_datagram_msg(skb, 0, msg, copied);
3206 0 : if (err)
3207 0 : goto out_free_skb;
3208 :
3209 0 : sock_recv_timestamp(msg, sk, skb);
3210 :
3211 0 : serr = SKB_EXT_ERR(skb);
3212 0 : put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3213 :
3214 0 : msg->msg_flags |= MSG_ERRQUEUE;
3215 0 : err = copied;
3216 :
3217 0 : out_free_skb:
3218 0 : kfree_skb(skb);
3219 0 : out:
3220 0 : return err;
3221 : }
3222 : EXPORT_SYMBOL(sock_recv_errqueue);
3223 :
3224 : /*
3225 : * Get a socket option on an socket.
3226 : *
3227 : * FIX: POSIX 1003.1g is very ambiguous here. It states that
3228 : * asynchronous errors should be reported by getsockopt. We assume
3229 : * this means if you specify SO_ERROR (otherwise whats the point of it).
3230 : */
3231 8 : int sock_common_getsockopt(struct socket *sock, int level, int optname,
3232 : char __user *optval, int __user *optlen)
3233 : {
3234 8 : struct sock *sk = sock->sk;
3235 :
3236 8 : return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3237 : }
3238 : EXPORT_SYMBOL(sock_common_getsockopt);
3239 :
3240 0 : int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3241 : int flags)
3242 : {
3243 0 : struct sock *sk = sock->sk;
3244 0 : int addr_len = 0;
3245 0 : int err;
3246 :
3247 0 : err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3248 : flags & ~MSG_DONTWAIT, &addr_len);
3249 0 : if (err >= 0)
3250 0 : msg->msg_namelen = addr_len;
3251 0 : return err;
3252 : }
3253 : EXPORT_SYMBOL(sock_common_recvmsg);
3254 :
3255 : /*
3256 : * Set socket options on an inet socket.
3257 : */
3258 11 : int sock_common_setsockopt(struct socket *sock, int level, int optname,
3259 : sockptr_t optval, unsigned int optlen)
3260 : {
3261 11 : struct sock *sk = sock->sk;
3262 :
3263 11 : return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3264 : }
3265 : EXPORT_SYMBOL(sock_common_setsockopt);
3266 :
3267 39 : void sk_common_release(struct sock *sk)
3268 : {
3269 39 : if (sk->sk_prot->destroy)
3270 39 : sk->sk_prot->destroy(sk);
3271 :
3272 : /*
3273 : * Observation: when sk_common_release is called, processes have
3274 : * no access to socket. But net still has.
3275 : * Step one, detach it from networking:
3276 : *
3277 : * A. Remove from hash tables.
3278 : */
3279 :
3280 39 : sk->sk_prot->unhash(sk);
3281 :
3282 : /*
3283 : * In this point socket cannot receive new packets, but it is possible
3284 : * that some packets are in flight because some CPU runs receiver and
3285 : * did hash table lookup before we unhashed socket. They will achieve
3286 : * receive queue and will be purged by socket destructor.
3287 : *
3288 : * Also we still have packets pending on receive queue and probably,
3289 : * our own packets waiting in device queues. sock_destroy will drain
3290 : * receive queue, but transmitted packets will delay socket destruction
3291 : * until the last reference will be released.
3292 : */
3293 :
3294 39 : sock_orphan(sk);
3295 :
3296 39 : xfrm_sk_free_policy(sk);
3297 :
3298 39 : sk_refcnt_debug_release(sk);
3299 :
3300 39 : sock_put(sk);
3301 39 : }
3302 : EXPORT_SYMBOL(sk_common_release);
3303 :
3304 0 : void sk_get_meminfo(const struct sock *sk, u32 *mem)
3305 : {
3306 0 : memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3307 :
3308 0 : mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3309 0 : mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3310 0 : mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3311 0 : mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3312 0 : mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3313 0 : mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3314 0 : mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3315 0 : mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3316 0 : mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3317 0 : }
3318 :
3319 : #ifdef CONFIG_PROC_FS
3320 : #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3321 : struct prot_inuse {
3322 : int val[PROTO_INUSE_NR];
3323 : };
3324 :
3325 : static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3326 :
3327 1564 : void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3328 : {
3329 1564 : __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3330 1564 : }
3331 : EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3332 :
3333 0 : int sock_prot_inuse_get(struct net *net, struct proto *prot)
3334 : {
3335 0 : int cpu, idx = prot->inuse_idx;
3336 0 : int res = 0;
3337 :
3338 0 : for_each_possible_cpu(cpu)
3339 0 : res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3340 :
3341 0 : return res >= 0 ? res : 0;
3342 : }
3343 : EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3344 :
3345 1562 : static void sock_inuse_add(struct net *net, int val)
3346 : {
3347 1562 : this_cpu_add(*net->core.sock_inuse, val);
3348 1562 : }
3349 :
3350 0 : int sock_inuse_get(struct net *net)
3351 : {
3352 0 : int cpu, res = 0;
3353 :
3354 0 : for_each_possible_cpu(cpu)
3355 0 : res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3356 :
3357 0 : return res;
3358 : }
3359 :
3360 : EXPORT_SYMBOL_GPL(sock_inuse_get);
3361 :
3362 1 : static int __net_init sock_inuse_init_net(struct net *net)
3363 : {
3364 1 : net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3365 1 : if (net->core.prot_inuse == NULL)
3366 : return -ENOMEM;
3367 :
3368 1 : net->core.sock_inuse = alloc_percpu(int);
3369 1 : if (net->core.sock_inuse == NULL)
3370 0 : goto out;
3371 :
3372 : return 0;
3373 :
3374 0 : out:
3375 0 : free_percpu(net->core.prot_inuse);
3376 0 : return -ENOMEM;
3377 : }
3378 :
3379 0 : static void __net_exit sock_inuse_exit_net(struct net *net)
3380 : {
3381 0 : free_percpu(net->core.prot_inuse);
3382 0 : free_percpu(net->core.sock_inuse);
3383 0 : }
3384 :
3385 : static struct pernet_operations net_inuse_ops = {
3386 : .init = sock_inuse_init_net,
3387 : .exit = sock_inuse_exit_net,
3388 : };
3389 :
3390 1 : static __init int net_inuse_init(void)
3391 : {
3392 1 : if (register_pernet_subsys(&net_inuse_ops))
3393 0 : panic("Cannot initialize net inuse counters");
3394 :
3395 1 : return 0;
3396 : }
3397 :
3398 : core_initcall(net_inuse_init);
3399 :
3400 8 : static int assign_proto_idx(struct proto *prot)
3401 : {
3402 8 : prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3403 :
3404 8 : if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3405 0 : pr_err("PROTO_INUSE_NR exhausted\n");
3406 0 : return -ENOSPC;
3407 : }
3408 :
3409 8 : set_bit(prot->inuse_idx, proto_inuse_idx);
3410 8 : return 0;
3411 : }
3412 :
3413 0 : static void release_proto_idx(struct proto *prot)
3414 : {
3415 0 : if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3416 0 : clear_bit(prot->inuse_idx, proto_inuse_idx);
3417 0 : }
3418 : #else
3419 : static inline int assign_proto_idx(struct proto *prot)
3420 : {
3421 : return 0;
3422 : }
3423 :
3424 : static inline void release_proto_idx(struct proto *prot)
3425 : {
3426 : }
3427 :
3428 : static void sock_inuse_add(struct net *net, int val)
3429 : {
3430 : }
3431 : #endif
3432 :
3433 0 : static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3434 : {
3435 0 : if (!twsk_prot)
3436 : return;
3437 0 : kfree(twsk_prot->twsk_slab_name);
3438 0 : twsk_prot->twsk_slab_name = NULL;
3439 0 : kmem_cache_destroy(twsk_prot->twsk_slab);
3440 0 : twsk_prot->twsk_slab = NULL;
3441 : }
3442 :
3443 0 : static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3444 : {
3445 0 : if (!rsk_prot)
3446 : return;
3447 0 : kfree(rsk_prot->slab_name);
3448 0 : rsk_prot->slab_name = NULL;
3449 0 : kmem_cache_destroy(rsk_prot->slab);
3450 0 : rsk_prot->slab = NULL;
3451 : }
3452 :
3453 6 : static int req_prot_init(const struct proto *prot)
3454 : {
3455 6 : struct request_sock_ops *rsk_prot = prot->rsk_prot;
3456 :
3457 6 : if (!rsk_prot)
3458 : return 0;
3459 :
3460 2 : rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3461 1 : prot->name);
3462 1 : if (!rsk_prot->slab_name)
3463 : return -ENOMEM;
3464 :
3465 2 : rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3466 : rsk_prot->obj_size, 0,
3467 1 : SLAB_ACCOUNT | prot->slab_flags,
3468 : NULL);
3469 :
3470 1 : if (!rsk_prot->slab) {
3471 0 : pr_crit("%s: Can't create request sock SLAB cache!\n",
3472 : prot->name);
3473 0 : return -ENOMEM;
3474 : }
3475 : return 0;
3476 : }
3477 :
3478 8 : int proto_register(struct proto *prot, int alloc_slab)
3479 : {
3480 8 : int ret = -ENOBUFS;
3481 :
3482 8 : if (alloc_slab) {
3483 12 : prot->slab = kmem_cache_create_usercopy(prot->name,
3484 : prot->obj_size, 0,
3485 : SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3486 6 : prot->slab_flags,
3487 : prot->useroffset, prot->usersize,
3488 : NULL);
3489 :
3490 6 : if (prot->slab == NULL) {
3491 0 : pr_crit("%s: Can't create sock SLAB cache!\n",
3492 : prot->name);
3493 0 : goto out;
3494 : }
3495 :
3496 6 : if (req_prot_init(prot))
3497 0 : goto out_free_request_sock_slab;
3498 :
3499 6 : if (prot->twsk_prot != NULL) {
3500 1 : prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3501 :
3502 1 : if (prot->twsk_prot->twsk_slab_name == NULL)
3503 0 : goto out_free_request_sock_slab;
3504 :
3505 2 : prot->twsk_prot->twsk_slab =
3506 1 : kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3507 : prot->twsk_prot->twsk_obj_size,
3508 : 0,
3509 : SLAB_ACCOUNT |
3510 : prot->slab_flags,
3511 : NULL);
3512 1 : if (prot->twsk_prot->twsk_slab == NULL)
3513 0 : goto out_free_timewait_sock_slab;
3514 : }
3515 : }
3516 :
3517 8 : mutex_lock(&proto_list_mutex);
3518 8 : ret = assign_proto_idx(prot);
3519 8 : if (ret) {
3520 0 : mutex_unlock(&proto_list_mutex);
3521 0 : goto out_free_timewait_sock_slab;
3522 : }
3523 8 : list_add(&prot->node, &proto_list);
3524 8 : mutex_unlock(&proto_list_mutex);
3525 8 : return ret;
3526 :
3527 0 : out_free_timewait_sock_slab:
3528 0 : if (alloc_slab && prot->twsk_prot)
3529 0 : tw_prot_cleanup(prot->twsk_prot);
3530 0 : out_free_request_sock_slab:
3531 0 : if (alloc_slab) {
3532 0 : req_prot_cleanup(prot->rsk_prot);
3533 :
3534 0 : kmem_cache_destroy(prot->slab);
3535 0 : prot->slab = NULL;
3536 : }
3537 0 : out:
3538 : return ret;
3539 : }
3540 : EXPORT_SYMBOL(proto_register);
3541 :
3542 0 : void proto_unregister(struct proto *prot)
3543 : {
3544 0 : mutex_lock(&proto_list_mutex);
3545 0 : release_proto_idx(prot);
3546 0 : list_del(&prot->node);
3547 0 : mutex_unlock(&proto_list_mutex);
3548 :
3549 0 : kmem_cache_destroy(prot->slab);
3550 0 : prot->slab = NULL;
3551 :
3552 0 : req_prot_cleanup(prot->rsk_prot);
3553 0 : tw_prot_cleanup(prot->twsk_prot);
3554 0 : }
3555 : EXPORT_SYMBOL(proto_unregister);
3556 :
3557 0 : int sock_load_diag_module(int family, int protocol)
3558 : {
3559 0 : if (!protocol) {
3560 0 : if (!sock_is_registered(family))
3561 : return -ENOENT;
3562 :
3563 0 : return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3564 : NETLINK_SOCK_DIAG, family);
3565 : }
3566 :
3567 : #ifdef CONFIG_INET
3568 0 : if (family == AF_INET &&
3569 0 : protocol != IPPROTO_RAW &&
3570 0 : protocol < MAX_INET_PROTOS &&
3571 0 : !rcu_access_pointer(inet_protos[protocol]))
3572 0 : return -ENOENT;
3573 : #endif
3574 :
3575 0 : return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3576 : NETLINK_SOCK_DIAG, family, protocol);
3577 : }
3578 : EXPORT_SYMBOL(sock_load_diag_module);
3579 :
3580 : #ifdef CONFIG_PROC_FS
3581 0 : static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3582 : __acquires(proto_list_mutex)
3583 : {
3584 0 : mutex_lock(&proto_list_mutex);
3585 0 : return seq_list_start_head(&proto_list, *pos);
3586 : }
3587 :
3588 0 : static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3589 : {
3590 0 : return seq_list_next(v, &proto_list, pos);
3591 : }
3592 :
3593 0 : static void proto_seq_stop(struct seq_file *seq, void *v)
3594 : __releases(proto_list_mutex)
3595 : {
3596 0 : mutex_unlock(&proto_list_mutex);
3597 0 : }
3598 :
3599 0 : static char proto_method_implemented(const void *method)
3600 : {
3601 0 : return method == NULL ? 'n' : 'y';
3602 : }
3603 0 : static long sock_prot_memory_allocated(struct proto *proto)
3604 : {
3605 0 : return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3606 : }
3607 :
3608 0 : static const char *sock_prot_memory_pressure(struct proto *proto)
3609 : {
3610 0 : return proto->memory_pressure != NULL ?
3611 0 : proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3612 : }
3613 :
3614 0 : static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3615 : {
3616 :
3617 0 : seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3618 : "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3619 0 : proto->name,
3620 : proto->obj_size,
3621 : sock_prot_inuse_get(seq_file_net(seq), proto),
3622 : sock_prot_memory_allocated(proto),
3623 : sock_prot_memory_pressure(proto),
3624 : proto->max_header,
3625 0 : proto->slab == NULL ? "no" : "yes",
3626 : module_name(proto->owner),
3627 0 : proto_method_implemented(proto->close),
3628 0 : proto_method_implemented(proto->connect),
3629 0 : proto_method_implemented(proto->disconnect),
3630 0 : proto_method_implemented(proto->accept),
3631 0 : proto_method_implemented(proto->ioctl),
3632 0 : proto_method_implemented(proto->init),
3633 0 : proto_method_implemented(proto->destroy),
3634 0 : proto_method_implemented(proto->shutdown),
3635 0 : proto_method_implemented(proto->setsockopt),
3636 0 : proto_method_implemented(proto->getsockopt),
3637 0 : proto_method_implemented(proto->sendmsg),
3638 0 : proto_method_implemented(proto->recvmsg),
3639 0 : proto_method_implemented(proto->sendpage),
3640 0 : proto_method_implemented(proto->bind),
3641 0 : proto_method_implemented(proto->backlog_rcv),
3642 0 : proto_method_implemented(proto->hash),
3643 0 : proto_method_implemented(proto->unhash),
3644 0 : proto_method_implemented(proto->get_port),
3645 0 : proto_method_implemented(proto->enter_memory_pressure));
3646 0 : }
3647 :
3648 0 : static int proto_seq_show(struct seq_file *seq, void *v)
3649 : {
3650 0 : if (v == &proto_list)
3651 0 : seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3652 : "protocol",
3653 : "size",
3654 : "sockets",
3655 : "memory",
3656 : "press",
3657 : "maxhdr",
3658 : "slab",
3659 : "module",
3660 : "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3661 : else
3662 0 : proto_seq_printf(seq, list_entry(v, struct proto, node));
3663 0 : return 0;
3664 : }
3665 :
3666 : static const struct seq_operations proto_seq_ops = {
3667 : .start = proto_seq_start,
3668 : .next = proto_seq_next,
3669 : .stop = proto_seq_stop,
3670 : .show = proto_seq_show,
3671 : };
3672 :
3673 1 : static __net_init int proto_init_net(struct net *net)
3674 : {
3675 1 : if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3676 : sizeof(struct seq_net_private)))
3677 0 : return -ENOMEM;
3678 :
3679 : return 0;
3680 : }
3681 :
3682 0 : static __net_exit void proto_exit_net(struct net *net)
3683 : {
3684 0 : remove_proc_entry("protocols", net->proc_net);
3685 0 : }
3686 :
3687 :
3688 : static __net_initdata struct pernet_operations proto_net_ops = {
3689 : .init = proto_init_net,
3690 : .exit = proto_exit_net,
3691 : };
3692 :
3693 1 : static int __init proto_init(void)
3694 : {
3695 1 : return register_pernet_subsys(&proto_net_ops);
3696 : }
3697 :
3698 : subsys_initcall(proto_init);
3699 :
3700 : #endif /* PROC_FS */
3701 :
3702 : #ifdef CONFIG_NET_RX_BUSY_POLL
3703 0 : bool sk_busy_loop_end(void *p, unsigned long start_time)
3704 : {
3705 0 : struct sock *sk = p;
3706 :
3707 0 : return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3708 0 : sk_busy_loop_timeout(sk, start_time);
3709 : }
3710 : EXPORT_SYMBOL(sk_busy_loop_end);
3711 : #endif /* CONFIG_NET_RX_BUSY_POLL */
3712 :
3713 0 : int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3714 : {
3715 0 : if (!sk->sk_prot->bind_add)
3716 : return -EOPNOTSUPP;
3717 0 : return sk->sk_prot->bind_add(sk, addr, addr_len);
3718 : }
3719 : EXPORT_SYMBOL(sock_bind_add);
|
