Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ef456144 CG |
2 | /* |
3 | * To speed up listener socket lookup, create an array to store all sockets | |
4 | * listening on the same port. This allows a decision to be made after finding | |
538950a1 CG |
5 | * the first socket. An optional BPF program can also be configured for |
6 | * selecting the socket index from the array of available sockets. | |
ef456144 CG |
7 | */ |
8 | ||
9 | #include <net/sock_reuseport.h> | |
538950a1 | 10 | #include <linux/bpf.h> |
736b4602 | 11 | #include <linux/idr.h> |
8217ca65 | 12 | #include <linux/filter.h> |
ef456144 CG |
13 | #include <linux/rcupdate.h> |
14 | ||
15 | #define INIT_SOCKS 128 | |
16 | ||
736b4602 MKL |
17 | DEFINE_SPINLOCK(reuseport_lock); |
18 | ||
736b4602 | 19 | static DEFINE_IDA(reuseport_ida); |
333bb73f KI |
20 | static int reuseport_resurrect(struct sock *sk, struct sock_reuseport *old_reuse, |
21 | struct sock_reuseport *reuse, bool bind_inany); | |
736b4602 | 22 | |
5c040eaf KI |
23 | static int reuseport_sock_index(struct sock *sk, |
24 | const struct sock_reuseport *reuse, | |
25 | bool closed) | |
26 | { | |
27 | int left, right; | |
28 | ||
29 | if (!closed) { | |
30 | left = 0; | |
31 | right = reuse->num_socks; | |
32 | } else { | |
33 | left = reuse->max_socks - reuse->num_closed_socks; | |
34 | right = reuse->max_socks; | |
35 | } | |
36 | ||
37 | for (; left < right; left++) | |
38 | if (reuse->socks[left] == sk) | |
39 | return left; | |
40 | return -1; | |
41 | } | |
42 | ||
43 | static void __reuseport_add_sock(struct sock *sk, | |
44 | struct sock_reuseport *reuse) | |
45 | { | |
46 | reuse->socks[reuse->num_socks] = sk; | |
1cd62c21 | 47 | /* paired with smp_rmb() in reuseport_(select|migrate)_sock() */ |
5c040eaf KI |
48 | smp_wmb(); |
49 | reuse->num_socks++; | |
50 | } | |
51 | ||
52 | static bool __reuseport_detach_sock(struct sock *sk, | |
53 | struct sock_reuseport *reuse) | |
54 | { | |
55 | int i = reuseport_sock_index(sk, reuse, false); | |
56 | ||
57 | if (i == -1) | |
58 | return false; | |
59 | ||
60 | reuse->socks[i] = reuse->socks[reuse->num_socks - 1]; | |
61 | reuse->num_socks--; | |
62 | ||
63 | return true; | |
64 | } | |
65 | ||
333bb73f KI |
66 | static void __reuseport_add_closed_sock(struct sock *sk, |
67 | struct sock_reuseport *reuse) | |
68 | { | |
69 | reuse->socks[reuse->max_socks - reuse->num_closed_socks - 1] = sk; | |
70 | /* paired with READ_ONCE() in inet_csk_bind_conflict() */ | |
71 | WRITE_ONCE(reuse->num_closed_socks, reuse->num_closed_socks + 1); | |
72 | } | |
73 | ||
74 | static bool __reuseport_detach_closed_sock(struct sock *sk, | |
75 | struct sock_reuseport *reuse) | |
76 | { | |
77 | int i = reuseport_sock_index(sk, reuse, true); | |
78 | ||
79 | if (i == -1) | |
80 | return false; | |
81 | ||
82 | reuse->socks[i] = reuse->socks[reuse->max_socks - reuse->num_closed_socks]; | |
83 | /* paired with READ_ONCE() in inet_csk_bind_conflict() */ | |
84 | WRITE_ONCE(reuse->num_closed_socks, reuse->num_closed_socks - 1); | |
85 | ||
86 | return true; | |
87 | } | |
88 | ||
822f9bb1 | 89 | static struct sock_reuseport *__reuseport_alloc(unsigned int max_socks) |
ef456144 | 90 | { |
822f9bb1 | 91 | unsigned int size = sizeof(struct sock_reuseport) + |
ef456144 CG |
92 | sizeof(struct sock *) * max_socks; |
93 | struct sock_reuseport *reuse = kzalloc(size, GFP_ATOMIC); | |
94 | ||
95 | if (!reuse) | |
96 | return NULL; | |
97 | ||
98 | reuse->max_socks = max_socks; | |
99 | ||
538950a1 | 100 | RCU_INIT_POINTER(reuse->prog, NULL); |
ef456144 CG |
101 | return reuse; |
102 | } | |
103 | ||
2dbb9b9e | 104 | int reuseport_alloc(struct sock *sk, bool bind_inany) |
ef456144 CG |
105 | { |
106 | struct sock_reuseport *reuse; | |
035ff358 | 107 | int id, ret = 0; |
ef456144 CG |
108 | |
109 | /* bh lock used since this function call may precede hlist lock in | |
110 | * soft irq of receive path or setsockopt from process context | |
111 | */ | |
112 | spin_lock_bh(&reuseport_lock); | |
1b5f962e CG |
113 | |
114 | /* Allocation attempts can occur concurrently via the setsockopt path | |
115 | * and the bind/hash path. Nothing to do when we lose the race. | |
116 | */ | |
2dbb9b9e MKL |
117 | reuse = rcu_dereference_protected(sk->sk_reuseport_cb, |
118 | lockdep_is_held(&reuseport_lock)); | |
119 | if (reuse) { | |
333bb73f KI |
120 | if (reuse->num_closed_socks) { |
121 | /* sk was shutdown()ed before */ | |
122 | ret = reuseport_resurrect(sk, reuse, NULL, bind_inany); | |
123 | goto out; | |
124 | } | |
125 | ||
2dbb9b9e MKL |
126 | /* Only set reuse->bind_inany if the bind_inany is true. |
127 | * Otherwise, it will overwrite the reuse->bind_inany | |
128 | * which was set by the bind/hash path. | |
129 | */ | |
130 | if (bind_inany) | |
131 | reuse->bind_inany = bind_inany; | |
1b5f962e | 132 | goto out; |
2dbb9b9e | 133 | } |
1b5f962e | 134 | |
ef456144 CG |
135 | reuse = __reuseport_alloc(INIT_SOCKS); |
136 | if (!reuse) { | |
035ff358 JS |
137 | ret = -ENOMEM; |
138 | goto out; | |
ef456144 CG |
139 | } |
140 | ||
035ff358 JS |
141 | id = ida_alloc(&reuseport_ida, GFP_ATOMIC); |
142 | if (id < 0) { | |
143 | kfree(reuse); | |
144 | ret = id; | |
145 | goto out; | |
146 | } | |
147 | ||
148 | reuse->reuseport_id = id; | |
5c040eaf | 149 | reuse->bind_inany = bind_inany; |
ef456144 CG |
150 | reuse->socks[0] = sk; |
151 | reuse->num_socks = 1; | |
152 | rcu_assign_pointer(sk->sk_reuseport_cb, reuse); | |
153 | ||
1b5f962e | 154 | out: |
ef456144 CG |
155 | spin_unlock_bh(&reuseport_lock); |
156 | ||
035ff358 | 157 | return ret; |
ef456144 CG |
158 | } |
159 | EXPORT_SYMBOL(reuseport_alloc); | |
160 | ||
161 | static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse) | |
162 | { | |
163 | struct sock_reuseport *more_reuse; | |
164 | u32 more_socks_size, i; | |
165 | ||
166 | more_socks_size = reuse->max_socks * 2U; | |
333bb73f KI |
167 | if (more_socks_size > U16_MAX) { |
168 | if (reuse->num_closed_socks) { | |
169 | /* Make room by removing a closed sk. | |
170 | * The child has already been migrated. | |
171 | * Only reqsk left at this point. | |
172 | */ | |
173 | struct sock *sk; | |
174 | ||
175 | sk = reuse->socks[reuse->max_socks - reuse->num_closed_socks]; | |
176 | RCU_INIT_POINTER(sk->sk_reuseport_cb, NULL); | |
177 | __reuseport_detach_closed_sock(sk, reuse); | |
178 | ||
179 | return reuse; | |
180 | } | |
181 | ||
ef456144 | 182 | return NULL; |
333bb73f | 183 | } |
ef456144 CG |
184 | |
185 | more_reuse = __reuseport_alloc(more_socks_size); | |
186 | if (!more_reuse) | |
187 | return NULL; | |
188 | ||
ef456144 | 189 | more_reuse->num_socks = reuse->num_socks; |
5c040eaf | 190 | more_reuse->num_closed_socks = reuse->num_closed_socks; |
538950a1 | 191 | more_reuse->prog = reuse->prog; |
736b4602 | 192 | more_reuse->reuseport_id = reuse->reuseport_id; |
2dbb9b9e | 193 | more_reuse->bind_inany = reuse->bind_inany; |
f2b2c55e | 194 | more_reuse->has_conns = reuse->has_conns; |
ef456144 CG |
195 | |
196 | memcpy(more_reuse->socks, reuse->socks, | |
197 | reuse->num_socks * sizeof(struct sock *)); | |
5c040eaf KI |
198 | memcpy(more_reuse->socks + |
199 | (more_reuse->max_socks - more_reuse->num_closed_socks), | |
200 | reuse->socks + (reuse->max_socks - reuse->num_closed_socks), | |
201 | reuse->num_closed_socks * sizeof(struct sock *)); | |
40a1227e | 202 | more_reuse->synq_overflow_ts = READ_ONCE(reuse->synq_overflow_ts); |
ef456144 | 203 | |
5c040eaf | 204 | for (i = 0; i < reuse->max_socks; ++i) |
ef456144 CG |
205 | rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb, |
206 | more_reuse); | |
207 | ||
538950a1 CG |
208 | /* Note: we use kfree_rcu here instead of reuseport_free_rcu so |
209 | * that reuse and more_reuse can temporarily share a reference | |
210 | * to prog. | |
211 | */ | |
ef456144 CG |
212 | kfree_rcu(reuse, rcu); |
213 | return more_reuse; | |
214 | } | |
215 | ||
4db428a7 ED |
216 | static void reuseport_free_rcu(struct rcu_head *head) |
217 | { | |
218 | struct sock_reuseport *reuse; | |
219 | ||
220 | reuse = container_of(head, struct sock_reuseport, rcu); | |
8217ca65 | 221 | sk_reuseport_prog_free(rcu_dereference_protected(reuse->prog, 1)); |
035ff358 | 222 | ida_free(&reuseport_ida, reuse->reuseport_id); |
4db428a7 ED |
223 | kfree(reuse); |
224 | } | |
225 | ||
ef456144 CG |
226 | /** |
227 | * reuseport_add_sock - Add a socket to the reuseport group of another. | |
228 | * @sk: New socket to add to the group. | |
229 | * @sk2: Socket belonging to the existing reuseport group. | |
37f3c421 BVA |
230 | * @bind_inany: Whether or not the group is bound to a local INANY address. |
231 | * | |
ef456144 CG |
232 | * May return ENOMEM and not add socket to group under memory pressure. |
233 | */ | |
2dbb9b9e | 234 | int reuseport_add_sock(struct sock *sk, struct sock *sk2, bool bind_inany) |
ef456144 | 235 | { |
4db428a7 | 236 | struct sock_reuseport *old_reuse, *reuse; |
ef456144 | 237 | |
b4ace4f1 | 238 | if (!rcu_access_pointer(sk2->sk_reuseport_cb)) { |
2dbb9b9e | 239 | int err = reuseport_alloc(sk2, bind_inany); |
b4ace4f1 CG |
240 | |
241 | if (err) | |
242 | return err; | |
243 | } | |
244 | ||
ef456144 CG |
245 | spin_lock_bh(&reuseport_lock); |
246 | reuse = rcu_dereference_protected(sk2->sk_reuseport_cb, | |
4db428a7 ED |
247 | lockdep_is_held(&reuseport_lock)); |
248 | old_reuse = rcu_dereference_protected(sk->sk_reuseport_cb, | |
333bb73f KI |
249 | lockdep_is_held(&reuseport_lock)); |
250 | if (old_reuse && old_reuse->num_closed_socks) { | |
251 | /* sk was shutdown()ed before */ | |
252 | int err = reuseport_resurrect(sk, old_reuse, reuse, reuse->bind_inany); | |
253 | ||
254 | spin_unlock_bh(&reuseport_lock); | |
255 | return err; | |
256 | } | |
257 | ||
4db428a7 ED |
258 | if (old_reuse && old_reuse->num_socks != 1) { |
259 | spin_unlock_bh(&reuseport_lock); | |
260 | return -EBUSY; | |
261 | } | |
ef456144 | 262 | |
5c040eaf | 263 | if (reuse->num_socks + reuse->num_closed_socks == reuse->max_socks) { |
ef456144 CG |
264 | reuse = reuseport_grow(reuse); |
265 | if (!reuse) { | |
266 | spin_unlock_bh(&reuseport_lock); | |
267 | return -ENOMEM; | |
268 | } | |
269 | } | |
270 | ||
5c040eaf | 271 | __reuseport_add_sock(sk, reuse); |
ef456144 CG |
272 | rcu_assign_pointer(sk->sk_reuseport_cb, reuse); |
273 | ||
274 | spin_unlock_bh(&reuseport_lock); | |
275 | ||
4db428a7 ED |
276 | if (old_reuse) |
277 | call_rcu(&old_reuse->rcu, reuseport_free_rcu); | |
ef456144 CG |
278 | return 0; |
279 | } | |
76c6d988 | 280 | EXPORT_SYMBOL(reuseport_add_sock); |
ef456144 | 281 | |
333bb73f KI |
282 | static int reuseport_resurrect(struct sock *sk, struct sock_reuseport *old_reuse, |
283 | struct sock_reuseport *reuse, bool bind_inany) | |
284 | { | |
285 | if (old_reuse == reuse) { | |
286 | /* If sk was in the same reuseport group, just pop sk out of | |
287 | * the closed section and push sk into the listening section. | |
288 | */ | |
289 | __reuseport_detach_closed_sock(sk, old_reuse); | |
290 | __reuseport_add_sock(sk, old_reuse); | |
291 | return 0; | |
292 | } | |
293 | ||
294 | if (!reuse) { | |
295 | /* In bind()/listen() path, we cannot carry over the eBPF prog | |
296 | * for the shutdown()ed socket. In setsockopt() path, we should | |
297 | * not change the eBPF prog of listening sockets by attaching a | |
298 | * prog to the shutdown()ed socket. Thus, we will allocate a new | |
299 | * reuseport group and detach sk from the old group. | |
300 | */ | |
301 | int id; | |
302 | ||
303 | reuse = __reuseport_alloc(INIT_SOCKS); | |
304 | if (!reuse) | |
305 | return -ENOMEM; | |
306 | ||
307 | id = ida_alloc(&reuseport_ida, GFP_ATOMIC); | |
308 | if (id < 0) { | |
309 | kfree(reuse); | |
310 | return id; | |
311 | } | |
312 | ||
313 | reuse->reuseport_id = id; | |
314 | reuse->bind_inany = bind_inany; | |
315 | } else { | |
316 | /* Move sk from the old group to the new one if | |
317 | * - all the other listeners in the old group were close()d or | |
318 | * shutdown()ed, and then sk2 has listen()ed on the same port | |
319 | * OR | |
320 | * - sk listen()ed without bind() (or with autobind), was | |
321 | * shutdown()ed, and then listen()s on another port which | |
322 | * sk2 listen()s on. | |
323 | */ | |
324 | if (reuse->num_socks + reuse->num_closed_socks == reuse->max_socks) { | |
325 | reuse = reuseport_grow(reuse); | |
326 | if (!reuse) | |
327 | return -ENOMEM; | |
328 | } | |
329 | } | |
330 | ||
331 | __reuseport_detach_closed_sock(sk, old_reuse); | |
332 | __reuseport_add_sock(sk, reuse); | |
333 | rcu_assign_pointer(sk->sk_reuseport_cb, reuse); | |
334 | ||
335 | if (old_reuse->num_socks + old_reuse->num_closed_socks == 0) | |
336 | call_rcu(&old_reuse->rcu, reuseport_free_rcu); | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
ef456144 CG |
341 | void reuseport_detach_sock(struct sock *sk) |
342 | { | |
343 | struct sock_reuseport *reuse; | |
ef456144 CG |
344 | |
345 | spin_lock_bh(&reuseport_lock); | |
346 | reuse = rcu_dereference_protected(sk->sk_reuseport_cb, | |
347 | lockdep_is_held(&reuseport_lock)); | |
5dc4c4b7 | 348 | |
333bb73f KI |
349 | /* reuseport_grow() has detached a closed sk */ |
350 | if (!reuse) | |
351 | goto out; | |
352 | ||
035ff358 JS |
353 | /* Notify the bpf side. The sk may be added to a sockarray |
354 | * map. If so, sockarray logic will remove it from the map. | |
355 | * | |
356 | * Other bpf map types that work with reuseport, like sockmap, | |
357 | * don't need an explicit callback from here. They override sk | |
358 | * unhash/close ops to remove the sk from the map before we | |
359 | * get to this point. | |
5dc4c4b7 | 360 | */ |
035ff358 | 361 | bpf_sk_reuseport_detach(sk); |
5dc4c4b7 | 362 | |
ef456144 | 363 | rcu_assign_pointer(sk->sk_reuseport_cb, NULL); |
333bb73f KI |
364 | |
365 | if (!__reuseport_detach_closed_sock(sk, reuse)) | |
366 | __reuseport_detach_sock(sk, reuse); | |
5c040eaf KI |
367 | |
368 | if (reuse->num_socks + reuse->num_closed_socks == 0) | |
369 | call_rcu(&reuse->rcu, reuseport_free_rcu); | |
ef456144 | 370 | |
333bb73f | 371 | out: |
ef456144 CG |
372 | spin_unlock_bh(&reuseport_lock); |
373 | } | |
374 | EXPORT_SYMBOL(reuseport_detach_sock); | |
375 | ||
333bb73f KI |
376 | void reuseport_stop_listen_sock(struct sock *sk) |
377 | { | |
378 | if (sk->sk_protocol == IPPROTO_TCP) { | |
379 | struct sock_reuseport *reuse; | |
380 | ||
381 | spin_lock_bh(&reuseport_lock); | |
382 | ||
383 | reuse = rcu_dereference_protected(sk->sk_reuseport_cb, | |
384 | lockdep_is_held(&reuseport_lock)); | |
385 | ||
386 | if (sock_net(sk)->ipv4.sysctl_tcp_migrate_req) { | |
387 | /* Migration capable, move sk from the listening section | |
388 | * to the closed section. | |
389 | */ | |
390 | bpf_sk_reuseport_detach(sk); | |
391 | ||
392 | __reuseport_detach_sock(sk, reuse); | |
393 | __reuseport_add_closed_sock(sk, reuse); | |
394 | ||
395 | spin_unlock_bh(&reuseport_lock); | |
396 | return; | |
397 | } | |
398 | ||
399 | spin_unlock_bh(&reuseport_lock); | |
400 | } | |
401 | ||
402 | /* Not capable to do migration, detach immediately */ | |
403 | reuseport_detach_sock(sk); | |
404 | } | |
405 | EXPORT_SYMBOL(reuseport_stop_listen_sock); | |
406 | ||
8217ca65 MKL |
407 | static struct sock *run_bpf_filter(struct sock_reuseport *reuse, u16 socks, |
408 | struct bpf_prog *prog, struct sk_buff *skb, | |
409 | int hdr_len) | |
538950a1 CG |
410 | { |
411 | struct sk_buff *nskb = NULL; | |
412 | u32 index; | |
413 | ||
414 | if (skb_shared(skb)) { | |
415 | nskb = skb_clone(skb, GFP_ATOMIC); | |
416 | if (!nskb) | |
417 | return NULL; | |
418 | skb = nskb; | |
419 | } | |
420 | ||
421 | /* temporarily advance data past protocol header */ | |
422 | if (!pskb_pull(skb, hdr_len)) { | |
00ce3a15 | 423 | kfree_skb(nskb); |
538950a1 CG |
424 | return NULL; |
425 | } | |
426 | index = bpf_prog_run_save_cb(prog, skb); | |
427 | __skb_push(skb, hdr_len); | |
428 | ||
429 | consume_skb(nskb); | |
430 | ||
431 | if (index >= socks) | |
432 | return NULL; | |
433 | ||
434 | return reuse->socks[index]; | |
435 | } | |
436 | ||
1cd62c21 KI |
437 | static struct sock *reuseport_select_sock_by_hash(struct sock_reuseport *reuse, |
438 | u32 hash, u16 num_socks) | |
439 | { | |
440 | int i, j; | |
441 | ||
442 | i = j = reciprocal_scale(hash, num_socks); | |
443 | while (reuse->socks[i]->sk_state == TCP_ESTABLISHED) { | |
444 | i++; | |
445 | if (i >= num_socks) | |
446 | i = 0; | |
447 | if (i == j) | |
448 | return NULL; | |
449 | } | |
450 | ||
451 | return reuse->socks[i]; | |
452 | } | |
453 | ||
ef456144 CG |
454 | /** |
455 | * reuseport_select_sock - Select a socket from an SO_REUSEPORT group. | |
456 | * @sk: First socket in the group. | |
538950a1 CG |
457 | * @hash: When no BPF filter is available, use this hash to select. |
458 | * @skb: skb to run through BPF filter. | |
459 | * @hdr_len: BPF filter expects skb data pointer at payload data. If | |
460 | * the skb does not yet point at the payload, this parameter represents | |
461 | * how far the pointer needs to advance to reach the payload. | |
ef456144 CG |
462 | * Returns a socket that should receive the packet (or NULL on error). |
463 | */ | |
538950a1 CG |
464 | struct sock *reuseport_select_sock(struct sock *sk, |
465 | u32 hash, | |
466 | struct sk_buff *skb, | |
467 | int hdr_len) | |
ef456144 CG |
468 | { |
469 | struct sock_reuseport *reuse; | |
538950a1 | 470 | struct bpf_prog *prog; |
ef456144 CG |
471 | struct sock *sk2 = NULL; |
472 | u16 socks; | |
473 | ||
474 | rcu_read_lock(); | |
475 | reuse = rcu_dereference(sk->sk_reuseport_cb); | |
476 | ||
477 | /* if memory allocation failed or add call is not yet complete */ | |
478 | if (!reuse) | |
479 | goto out; | |
480 | ||
538950a1 | 481 | prog = rcu_dereference(reuse->prog); |
ef456144 CG |
482 | socks = READ_ONCE(reuse->num_socks); |
483 | if (likely(socks)) { | |
5c040eaf | 484 | /* paired with smp_wmb() in __reuseport_add_sock() */ |
ef456144 CG |
485 | smp_rmb(); |
486 | ||
8217ca65 MKL |
487 | if (!prog || !skb) |
488 | goto select_by_hash; | |
489 | ||
490 | if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) | |
491 | sk2 = bpf_run_sk_reuseport(reuse, sk, prog, skb, hash); | |
492 | else | |
493 | sk2 = run_bpf_filter(reuse, socks, prog, skb, hdr_len); | |
e94a62f5 | 494 | |
8217ca65 | 495 | select_by_hash: |
e94a62f5 | 496 | /* no bpf or invalid bpf result: fall back to hash usage */ |
1cd62c21 KI |
497 | if (!sk2) |
498 | sk2 = reuseport_select_sock_by_hash(reuse, hash, socks); | |
ef456144 CG |
499 | } |
500 | ||
501 | out: | |
502 | rcu_read_unlock(); | |
503 | return sk2; | |
504 | } | |
505 | EXPORT_SYMBOL(reuseport_select_sock); | |
538950a1 | 506 | |
1cd62c21 KI |
507 | /** |
508 | * reuseport_migrate_sock - Select a socket from an SO_REUSEPORT group. | |
509 | * @sk: close()ed or shutdown()ed socket in the group. | |
510 | * @migrating_sk: ESTABLISHED/SYN_RECV full socket in the accept queue or | |
511 | * NEW_SYN_RECV request socket during 3WHS. | |
512 | * @skb: skb to run through BPF filter. | |
513 | * Returns a socket (with sk_refcnt +1) that should accept the child socket | |
514 | * (or NULL on error). | |
515 | */ | |
516 | struct sock *reuseport_migrate_sock(struct sock *sk, | |
517 | struct sock *migrating_sk, | |
518 | struct sk_buff *skb) | |
519 | { | |
520 | struct sock_reuseport *reuse; | |
521 | struct sock *nsk = NULL; | |
522 | u16 socks; | |
523 | u32 hash; | |
524 | ||
525 | rcu_read_lock(); | |
526 | ||
527 | reuse = rcu_dereference(sk->sk_reuseport_cb); | |
528 | if (!reuse) | |
529 | goto out; | |
530 | ||
531 | socks = READ_ONCE(reuse->num_socks); | |
532 | if (unlikely(!socks)) | |
533 | goto out; | |
534 | ||
535 | /* paired with smp_wmb() in __reuseport_add_sock() */ | |
536 | smp_rmb(); | |
537 | ||
538 | hash = migrating_sk->sk_hash; | |
539 | if (sock_net(sk)->ipv4.sysctl_tcp_migrate_req) | |
540 | nsk = reuseport_select_sock_by_hash(reuse, hash, socks); | |
541 | ||
542 | if (nsk && unlikely(!refcount_inc_not_zero(&nsk->sk_refcnt))) | |
543 | nsk = NULL; | |
544 | ||
545 | out: | |
546 | rcu_read_unlock(); | |
547 | return nsk; | |
548 | } | |
549 | EXPORT_SYMBOL(reuseport_migrate_sock); | |
550 | ||
8217ca65 | 551 | int reuseport_attach_prog(struct sock *sk, struct bpf_prog *prog) |
538950a1 CG |
552 | { |
553 | struct sock_reuseport *reuse; | |
554 | struct bpf_prog *old_prog; | |
555 | ||
333bb73f KI |
556 | if (sk_unhashed(sk)) { |
557 | int err; | |
8217ca65 | 558 | |
333bb73f KI |
559 | if (!sk->sk_reuseport) |
560 | return -EINVAL; | |
561 | ||
562 | err = reuseport_alloc(sk, false); | |
8217ca65 MKL |
563 | if (err) |
564 | return err; | |
565 | } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) { | |
566 | /* The socket wasn't bound with SO_REUSEPORT */ | |
567 | return -EINVAL; | |
568 | } | |
569 | ||
538950a1 CG |
570 | spin_lock_bh(&reuseport_lock); |
571 | reuse = rcu_dereference_protected(sk->sk_reuseport_cb, | |
572 | lockdep_is_held(&reuseport_lock)); | |
573 | old_prog = rcu_dereference_protected(reuse->prog, | |
574 | lockdep_is_held(&reuseport_lock)); | |
575 | rcu_assign_pointer(reuse->prog, prog); | |
576 | spin_unlock_bh(&reuseport_lock); | |
577 | ||
8217ca65 MKL |
578 | sk_reuseport_prog_free(old_prog); |
579 | return 0; | |
538950a1 CG |
580 | } |
581 | EXPORT_SYMBOL(reuseport_attach_prog); | |
99f3a064 MKL |
582 | |
583 | int reuseport_detach_prog(struct sock *sk) | |
584 | { | |
585 | struct sock_reuseport *reuse; | |
586 | struct bpf_prog *old_prog; | |
587 | ||
99f3a064 MKL |
588 | old_prog = NULL; |
589 | spin_lock_bh(&reuseport_lock); | |
590 | reuse = rcu_dereference_protected(sk->sk_reuseport_cb, | |
591 | lockdep_is_held(&reuseport_lock)); | |
333bb73f KI |
592 | |
593 | /* reuse must be checked after acquiring the reuseport_lock | |
594 | * because reuseport_grow() can detach a closed sk. | |
595 | */ | |
596 | if (!reuse) { | |
597 | spin_unlock_bh(&reuseport_lock); | |
598 | return sk->sk_reuseport ? -ENOENT : -EINVAL; | |
599 | } | |
600 | ||
601 | if (sk_unhashed(sk) && reuse->num_closed_socks) { | |
602 | spin_unlock_bh(&reuseport_lock); | |
603 | return -ENOENT; | |
604 | } | |
605 | ||
e3f0d761 PM |
606 | old_prog = rcu_replace_pointer(reuse->prog, old_prog, |
607 | lockdep_is_held(&reuseport_lock)); | |
99f3a064 MKL |
608 | spin_unlock_bh(&reuseport_lock); |
609 | ||
610 | if (!old_prog) | |
611 | return -ENOENT; | |
612 | ||
613 | sk_reuseport_prog_free(old_prog); | |
614 | return 0; | |
615 | } | |
616 | EXPORT_SYMBOL(reuseport_detach_prog); |