Commit | Line | Data |
---|---|---|
6ac99e8f MKL |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Copyright (c) 2019 Facebook */ | |
3 | #include <linux/rculist.h> | |
4 | #include <linux/list.h> | |
5 | #include <linux/hash.h> | |
6 | #include <linux/types.h> | |
7 | #include <linux/spinlock.h> | |
8 | #include <linux/bpf.h> | |
8e4597c6 | 9 | #include <linux/btf.h> |
5ce6e77c | 10 | #include <linux/btf_ids.h> |
450af8d0 | 11 | #include <linux/bpf_local_storage.h> |
6ac99e8f MKL |
12 | #include <net/bpf_sk_storage.h> |
13 | #include <net/sock.h> | |
1ed4d924 | 14 | #include <uapi/linux/sock_diag.h> |
6ac99e8f | 15 | #include <uapi/linux/btf.h> |
0fe4b381 | 16 | #include <linux/rcupdate_trace.h> |
6ac99e8f | 17 | |
4cc9ce4e KS |
18 | DEFINE_BPF_STORAGE_CACHE(sk_cache); |
19 | ||
1f00d375 | 20 | static struct bpf_local_storage_data * |
e794bfdd | 21 | bpf_sk_storage_lookup(struct sock *sk, struct bpf_map *map, bool cacheit_lockit) |
6ac99e8f | 22 | { |
1f00d375 KS |
23 | struct bpf_local_storage *sk_storage; |
24 | struct bpf_local_storage_map *smap; | |
6ac99e8f | 25 | |
0fe4b381 KS |
26 | sk_storage = |
27 | rcu_dereference_check(sk->sk_bpf_storage, bpf_rcu_lock_held()); | |
6ac99e8f MKL |
28 | if (!sk_storage) |
29 | return NULL; | |
30 | ||
1f00d375 KS |
31 | smap = (struct bpf_local_storage_map *)map; |
32 | return bpf_local_storage_lookup(sk_storage, smap, cacheit_lockit); | |
6ac99e8f MKL |
33 | } |
34 | ||
e794bfdd | 35 | static int bpf_sk_storage_del(struct sock *sk, struct bpf_map *map) |
6ac99e8f | 36 | { |
1f00d375 | 37 | struct bpf_local_storage_data *sdata; |
6ac99e8f | 38 | |
e794bfdd | 39 | sdata = bpf_sk_storage_lookup(sk, map, false); |
6ac99e8f MKL |
40 | if (!sdata) |
41 | return -ENOENT; | |
42 | ||
1f00d375 | 43 | bpf_selem_unlink(SELEM(sdata)); |
6ac99e8f MKL |
44 | |
45 | return 0; | |
46 | } | |
47 | ||
8f51dfc7 | 48 | /* Called by __sk_destruct() & bpf_sk_storage_clone() */ |
6ac99e8f MKL |
49 | void bpf_sk_storage_free(struct sock *sk) |
50 | { | |
1f00d375 KS |
51 | struct bpf_local_storage_elem *selem; |
52 | struct bpf_local_storage *sk_storage; | |
6ac99e8f MKL |
53 | bool free_sk_storage = false; |
54 | struct hlist_node *n; | |
55 | ||
56 | rcu_read_lock(); | |
57 | sk_storage = rcu_dereference(sk->sk_bpf_storage); | |
58 | if (!sk_storage) { | |
59 | rcu_read_unlock(); | |
60 | return; | |
61 | } | |
62 | ||
63 | /* Netiher the bpf_prog nor the bpf-map's syscall | |
64 | * could be modifying the sk_storage->list now. | |
65 | * Thus, no elem can be added-to or deleted-from the | |
66 | * sk_storage->list by the bpf_prog or by the bpf-map's syscall. | |
67 | * | |
1f00d375 | 68 | * It is racing with bpf_local_storage_map_free() alone |
6ac99e8f MKL |
69 | * when unlinking elem from the sk_storage->list and |
70 | * the map's bucket->list. | |
71 | */ | |
72 | raw_spin_lock_bh(&sk_storage->lock); | |
73 | hlist_for_each_entry_safe(selem, n, &sk_storage->list, snode) { | |
74 | /* Always unlink from map before unlinking from | |
75 | * sk_storage. | |
76 | */ | |
1f00d375 KS |
77 | bpf_selem_unlink_map(selem); |
78 | free_sk_storage = bpf_selem_unlink_storage_nolock(sk_storage, | |
79 | selem, true); | |
6ac99e8f MKL |
80 | } |
81 | raw_spin_unlock_bh(&sk_storage->lock); | |
82 | rcu_read_unlock(); | |
83 | ||
84 | if (free_sk_storage) | |
85 | kfree_rcu(sk_storage, rcu); | |
86 | } | |
87 | ||
e794bfdd | 88 | static void bpf_sk_storage_map_free(struct bpf_map *map) |
f836a56e KS |
89 | { |
90 | struct bpf_local_storage_map *smap; | |
91 | ||
92 | smap = (struct bpf_local_storage_map *)map; | |
93 | bpf_local_storage_cache_idx_free(&sk_cache, smap->cache_idx); | |
bc235cdb | 94 | bpf_local_storage_map_free(smap, NULL); |
6ac99e8f MKL |
95 | } |
96 | ||
e794bfdd | 97 | static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr) |
f836a56e KS |
98 | { |
99 | struct bpf_local_storage_map *smap; | |
100 | ||
101 | smap = bpf_local_storage_map_alloc(attr); | |
102 | if (IS_ERR(smap)) | |
103 | return ERR_CAST(smap); | |
104 | ||
105 | smap->cache_idx = bpf_local_storage_cache_idx_get(&sk_cache); | |
6ac99e8f MKL |
106 | return &smap->map; |
107 | } | |
108 | ||
109 | static int notsupp_get_next_key(struct bpf_map *map, void *key, | |
110 | void *next_key) | |
111 | { | |
112 | return -ENOTSUPP; | |
113 | } | |
114 | ||
6ac99e8f MKL |
115 | static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key) |
116 | { | |
1f00d375 | 117 | struct bpf_local_storage_data *sdata; |
6ac99e8f MKL |
118 | struct socket *sock; |
119 | int fd, err; | |
120 | ||
121 | fd = *(int *)key; | |
122 | sock = sockfd_lookup(fd, &err); | |
123 | if (sock) { | |
e794bfdd | 124 | sdata = bpf_sk_storage_lookup(sock->sk, map, true); |
6ac99e8f MKL |
125 | sockfd_put(sock); |
126 | return sdata ? sdata->data : NULL; | |
127 | } | |
128 | ||
129 | return ERR_PTR(err); | |
130 | } | |
131 | ||
132 | static int bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key, | |
133 | void *value, u64 map_flags) | |
134 | { | |
1f00d375 | 135 | struct bpf_local_storage_data *sdata; |
6ac99e8f MKL |
136 | struct socket *sock; |
137 | int fd, err; | |
138 | ||
139 | fd = *(int *)key; | |
140 | sock = sockfd_lookup(fd, &err); | |
141 | if (sock) { | |
f836a56e KS |
142 | sdata = bpf_local_storage_update( |
143 | sock->sk, (struct bpf_local_storage_map *)map, value, | |
b00fa38a | 144 | map_flags, GFP_ATOMIC); |
6ac99e8f | 145 | sockfd_put(sock); |
71f150f4 | 146 | return PTR_ERR_OR_ZERO(sdata); |
6ac99e8f MKL |
147 | } |
148 | ||
149 | return err; | |
150 | } | |
151 | ||
152 | static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key) | |
153 | { | |
154 | struct socket *sock; | |
155 | int fd, err; | |
156 | ||
157 | fd = *(int *)key; | |
158 | sock = sockfd_lookup(fd, &err); | |
159 | if (sock) { | |
e794bfdd | 160 | err = bpf_sk_storage_del(sock->sk, map); |
6ac99e8f MKL |
161 | sockfd_put(sock); |
162 | return err; | |
163 | } | |
164 | ||
165 | return err; | |
166 | } | |
167 | ||
1f00d375 | 168 | static struct bpf_local_storage_elem * |
8f51dfc7 | 169 | bpf_sk_storage_clone_elem(struct sock *newsk, |
1f00d375 KS |
170 | struct bpf_local_storage_map *smap, |
171 | struct bpf_local_storage_elem *selem) | |
8f51dfc7 | 172 | { |
1f00d375 | 173 | struct bpf_local_storage_elem *copy_selem; |
8f51dfc7 | 174 | |
b00fa38a | 175 | copy_selem = bpf_selem_alloc(smap, newsk, NULL, true, GFP_ATOMIC); |
8f51dfc7 SF |
176 | if (!copy_selem) |
177 | return NULL; | |
178 | ||
179 | if (map_value_has_spin_lock(&smap->map)) | |
180 | copy_map_value_locked(&smap->map, SDATA(copy_selem)->data, | |
181 | SDATA(selem)->data, true); | |
182 | else | |
183 | copy_map_value(&smap->map, SDATA(copy_selem)->data, | |
184 | SDATA(selem)->data); | |
185 | ||
186 | return copy_selem; | |
187 | } | |
188 | ||
189 | int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk) | |
190 | { | |
1f00d375 KS |
191 | struct bpf_local_storage *new_sk_storage = NULL; |
192 | struct bpf_local_storage *sk_storage; | |
193 | struct bpf_local_storage_elem *selem; | |
8f51dfc7 SF |
194 | int ret = 0; |
195 | ||
196 | RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL); | |
197 | ||
198 | rcu_read_lock(); | |
199 | sk_storage = rcu_dereference(sk->sk_bpf_storage); | |
200 | ||
201 | if (!sk_storage || hlist_empty(&sk_storage->list)) | |
202 | goto out; | |
203 | ||
204 | hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) { | |
1f00d375 KS |
205 | struct bpf_local_storage_elem *copy_selem; |
206 | struct bpf_local_storage_map *smap; | |
8f51dfc7 SF |
207 | struct bpf_map *map; |
208 | ||
209 | smap = rcu_dereference(SDATA(selem)->smap); | |
210 | if (!(smap->map.map_flags & BPF_F_CLONE)) | |
211 | continue; | |
212 | ||
213 | /* Note that for lockless listeners adding new element | |
1f00d375 | 214 | * here can race with cleanup in bpf_local_storage_map_free. |
8f51dfc7 SF |
215 | * Try to grab map refcnt to make sure that it's still |
216 | * alive and prevent concurrent removal. | |
217 | */ | |
1e0bd5a0 | 218 | map = bpf_map_inc_not_zero(&smap->map); |
8f51dfc7 SF |
219 | if (IS_ERR(map)) |
220 | continue; | |
221 | ||
222 | copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem); | |
223 | if (!copy_selem) { | |
224 | ret = -ENOMEM; | |
225 | bpf_map_put(map); | |
226 | goto out; | |
227 | } | |
228 | ||
229 | if (new_sk_storage) { | |
1f00d375 KS |
230 | bpf_selem_link_map(smap, copy_selem); |
231 | bpf_selem_link_storage_nolock(new_sk_storage, copy_selem); | |
8f51dfc7 | 232 | } else { |
b00fa38a | 233 | ret = bpf_local_storage_alloc(newsk, smap, copy_selem, GFP_ATOMIC); |
8f51dfc7 SF |
234 | if (ret) { |
235 | kfree(copy_selem); | |
236 | atomic_sub(smap->elem_size, | |
237 | &newsk->sk_omem_alloc); | |
238 | bpf_map_put(map); | |
239 | goto out; | |
240 | } | |
241 | ||
1f00d375 KS |
242 | new_sk_storage = |
243 | rcu_dereference(copy_selem->local_storage); | |
8f51dfc7 SF |
244 | } |
245 | bpf_map_put(map); | |
246 | } | |
247 | ||
248 | out: | |
249 | rcu_read_unlock(); | |
250 | ||
251 | /* In case of an error, don't free anything explicitly here, the | |
252 | * caller is responsible to call bpf_sk_storage_free. | |
253 | */ | |
254 | ||
255 | return ret; | |
256 | } | |
257 | ||
b00fa38a JK |
258 | /* *gfp_flags* is a hidden argument provided by the verifier */ |
259 | BPF_CALL_5(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk, | |
260 | void *, value, u64, flags, gfp_t, gfp_flags) | |
6ac99e8f | 261 | { |
1f00d375 | 262 | struct bpf_local_storage_data *sdata; |
6ac99e8f | 263 | |
0fe4b381 | 264 | WARN_ON_ONCE(!bpf_rcu_lock_held()); |
592a3498 | 265 | if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE) |
6ac99e8f MKL |
266 | return (unsigned long)NULL; |
267 | ||
e794bfdd | 268 | sdata = bpf_sk_storage_lookup(sk, map, true); |
6ac99e8f MKL |
269 | if (sdata) |
270 | return (unsigned long)sdata->data; | |
271 | ||
272 | if (flags == BPF_SK_STORAGE_GET_F_CREATE && | |
273 | /* Cannot add new elem to a going away sk. | |
274 | * Otherwise, the new elem may become a leak | |
275 | * (and also other memory issues during map | |
276 | * destruction). | |
277 | */ | |
278 | refcount_inc_not_zero(&sk->sk_refcnt)) { | |
f836a56e KS |
279 | sdata = bpf_local_storage_update( |
280 | sk, (struct bpf_local_storage_map *)map, value, | |
b00fa38a | 281 | BPF_NOEXIST, gfp_flags); |
6ac99e8f MKL |
282 | /* sk must be a fullsock (guaranteed by verifier), |
283 | * so sock_gen_put() is unnecessary. | |
284 | */ | |
285 | sock_put(sk); | |
286 | return IS_ERR(sdata) ? | |
287 | (unsigned long)NULL : (unsigned long)sdata->data; | |
288 | } | |
289 | ||
290 | return (unsigned long)NULL; | |
291 | } | |
292 | ||
293 | BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk) | |
294 | { | |
0fe4b381 | 295 | WARN_ON_ONCE(!bpf_rcu_lock_held()); |
592a3498 MKL |
296 | if (!sk || !sk_fullsock(sk)) |
297 | return -EINVAL; | |
298 | ||
6ac99e8f MKL |
299 | if (refcount_inc_not_zero(&sk->sk_refcnt)) { |
300 | int err; | |
301 | ||
e794bfdd | 302 | err = bpf_sk_storage_del(sk, map); |
6ac99e8f MKL |
303 | sock_put(sk); |
304 | return err; | |
305 | } | |
306 | ||
307 | return -ENOENT; | |
308 | } | |
309 | ||
e794bfdd MKL |
310 | static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap, |
311 | void *owner, u32 size) | |
f836a56e | 312 | { |
9e838b02 MKL |
313 | struct sock *sk = (struct sock *)owner; |
314 | ||
315 | /* same check as in sock_kmalloc() */ | |
316 | if (size <= sysctl_optmem_max && | |
317 | atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { | |
318 | atomic_add(size, &sk->sk_omem_alloc); | |
319 | return 0; | |
320 | } | |
321 | ||
322 | return -ENOMEM; | |
f836a56e KS |
323 | } |
324 | ||
e794bfdd MKL |
325 | static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap, |
326 | void *owner, u32 size) | |
f836a56e KS |
327 | { |
328 | struct sock *sk = owner; | |
329 | ||
330 | atomic_sub(size, &sk->sk_omem_alloc); | |
331 | } | |
332 | ||
333 | static struct bpf_local_storage __rcu ** | |
e794bfdd | 334 | bpf_sk_storage_ptr(void *owner) |
f836a56e KS |
335 | { |
336 | struct sock *sk = owner; | |
337 | ||
338 | return &sk->sk_bpf_storage; | |
339 | } | |
340 | ||
2872e9ac | 341 | static int sk_storage_map_btf_id; |
6ac99e8f | 342 | const struct bpf_map_ops sk_storage_map_ops = { |
f4d05259 | 343 | .map_meta_equal = bpf_map_meta_equal, |
1f00d375 | 344 | .map_alloc_check = bpf_local_storage_map_alloc_check, |
e794bfdd MKL |
345 | .map_alloc = bpf_sk_storage_map_alloc, |
346 | .map_free = bpf_sk_storage_map_free, | |
6ac99e8f MKL |
347 | .map_get_next_key = notsupp_get_next_key, |
348 | .map_lookup_elem = bpf_fd_sk_storage_lookup_elem, | |
349 | .map_update_elem = bpf_fd_sk_storage_update_elem, | |
350 | .map_delete_elem = bpf_fd_sk_storage_delete_elem, | |
1f00d375 KS |
351 | .map_check_btf = bpf_local_storage_map_check_btf, |
352 | .map_btf_name = "bpf_local_storage_map", | |
2872e9ac | 353 | .map_btf_id = &sk_storage_map_btf_id, |
e794bfdd MKL |
354 | .map_local_storage_charge = bpf_sk_storage_charge, |
355 | .map_local_storage_uncharge = bpf_sk_storage_uncharge, | |
356 | .map_owner_storage_ptr = bpf_sk_storage_ptr, | |
6ac99e8f MKL |
357 | }; |
358 | ||
359 | const struct bpf_func_proto bpf_sk_storage_get_proto = { | |
360 | .func = bpf_sk_storage_get, | |
361 | .gpl_only = false, | |
362 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
363 | .arg1_type = ARG_CONST_MAP_PTR, | |
592a3498 | 364 | .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, |
6ac99e8f MKL |
365 | .arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL, |
366 | .arg4_type = ARG_ANYTHING, | |
367 | }; | |
368 | ||
f7c6cb1d SF |
369 | const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = { |
370 | .func = bpf_sk_storage_get, | |
371 | .gpl_only = false, | |
372 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
373 | .arg1_type = ARG_CONST_MAP_PTR, | |
374 | .arg2_type = ARG_PTR_TO_CTX, /* context is 'struct sock' */ | |
375 | .arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL, | |
376 | .arg4_type = ARG_ANYTHING, | |
377 | }; | |
378 | ||
6ac99e8f MKL |
379 | const struct bpf_func_proto bpf_sk_storage_delete_proto = { |
380 | .func = bpf_sk_storage_delete, | |
381 | .gpl_only = false, | |
382 | .ret_type = RET_INTEGER, | |
383 | .arg1_type = ARG_CONST_MAP_PTR, | |
592a3498 | 384 | .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, |
30897832 KS |
385 | }; |
386 | ||
8e4597c6 MKL |
387 | static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog) |
388 | { | |
389 | const struct btf *btf_vmlinux; | |
390 | const struct btf_type *t; | |
391 | const char *tname; | |
392 | u32 btf_id; | |
393 | ||
394 | if (prog->aux->dst_prog) | |
395 | return false; | |
396 | ||
397 | /* Ensure the tracing program is not tracing | |
398 | * any bpf_sk_storage*() function and also | |
399 | * use the bpf_sk_storage_(get|delete) helper. | |
400 | */ | |
401 | switch (prog->expected_attach_type) { | |
a50a85e4 | 402 | case BPF_TRACE_ITER: |
8e4597c6 MKL |
403 | case BPF_TRACE_RAW_TP: |
404 | /* bpf_sk_storage has no trace point */ | |
405 | return true; | |
406 | case BPF_TRACE_FENTRY: | |
407 | case BPF_TRACE_FEXIT: | |
408 | btf_vmlinux = bpf_get_btf_vmlinux(); | |
409 | btf_id = prog->aux->attach_btf_id; | |
410 | t = btf_type_by_id(btf_vmlinux, btf_id); | |
411 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
412 | return !!strncmp(tname, "bpf_sk_storage", | |
413 | strlen("bpf_sk_storage")); | |
414 | default: | |
415 | return false; | |
416 | } | |
417 | ||
418 | return false; | |
419 | } | |
420 | ||
b00fa38a JK |
421 | /* *gfp_flags* is a hidden argument provided by the verifier */ |
422 | BPF_CALL_5(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk, | |
423 | void *, value, u64, flags, gfp_t, gfp_flags) | |
8e4597c6 | 424 | { |
0fe4b381 | 425 | WARN_ON_ONCE(!bpf_rcu_lock_held()); |
afa79d08 | 426 | if (in_hardirq() || in_nmi()) |
8e4597c6 MKL |
427 | return (unsigned long)NULL; |
428 | ||
b00fa38a JK |
429 | return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags, |
430 | gfp_flags); | |
8e4597c6 MKL |
431 | } |
432 | ||
433 | BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map, | |
434 | struct sock *, sk) | |
435 | { | |
0fe4b381 | 436 | WARN_ON_ONCE(!bpf_rcu_lock_held()); |
afa79d08 | 437 | if (in_hardirq() || in_nmi()) |
8e4597c6 MKL |
438 | return -EPERM; |
439 | ||
440 | return ____bpf_sk_storage_delete(map, sk); | |
441 | } | |
442 | ||
443 | const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = { | |
444 | .func = bpf_sk_storage_get_tracing, | |
445 | .gpl_only = false, | |
446 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
447 | .arg1_type = ARG_CONST_MAP_PTR, | |
448 | .arg2_type = ARG_PTR_TO_BTF_ID, | |
449 | .arg2_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
450 | .arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL, | |
451 | .arg4_type = ARG_ANYTHING, | |
452 | .allowed = bpf_sk_storage_tracing_allowed, | |
453 | }; | |
454 | ||
455 | const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = { | |
456 | .func = bpf_sk_storage_delete_tracing, | |
457 | .gpl_only = false, | |
458 | .ret_type = RET_INTEGER, | |
459 | .arg1_type = ARG_CONST_MAP_PTR, | |
460 | .arg2_type = ARG_PTR_TO_BTF_ID, | |
461 | .arg2_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
462 | .allowed = bpf_sk_storage_tracing_allowed, | |
463 | }; | |
464 | ||
1ed4d924 MKL |
465 | struct bpf_sk_storage_diag { |
466 | u32 nr_maps; | |
467 | struct bpf_map *maps[]; | |
468 | }; | |
469 | ||
470 | /* The reply will be like: | |
471 | * INET_DIAG_BPF_SK_STORAGES (nla_nest) | |
472 | * SK_DIAG_BPF_STORAGE (nla_nest) | |
473 | * SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32) | |
474 | * SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit) | |
475 | * SK_DIAG_BPF_STORAGE (nla_nest) | |
476 | * SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32) | |
477 | * SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit) | |
478 | * .... | |
479 | */ | |
480 | static int nla_value_size(u32 value_size) | |
481 | { | |
482 | /* SK_DIAG_BPF_STORAGE (nla_nest) | |
483 | * SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32) | |
484 | * SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit) | |
485 | */ | |
486 | return nla_total_size(0) + nla_total_size(sizeof(u32)) + | |
487 | nla_total_size_64bit(value_size); | |
488 | } | |
489 | ||
490 | void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag) | |
491 | { | |
492 | u32 i; | |
493 | ||
494 | if (!diag) | |
495 | return; | |
496 | ||
497 | for (i = 0; i < diag->nr_maps; i++) | |
498 | bpf_map_put(diag->maps[i]); | |
499 | ||
500 | kfree(diag); | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free); | |
503 | ||
504 | static bool diag_check_dup(const struct bpf_sk_storage_diag *diag, | |
505 | const struct bpf_map *map) | |
506 | { | |
507 | u32 i; | |
508 | ||
509 | for (i = 0; i < diag->nr_maps; i++) { | |
510 | if (diag->maps[i] == map) | |
511 | return true; | |
512 | } | |
513 | ||
514 | return false; | |
515 | } | |
516 | ||
517 | struct bpf_sk_storage_diag * | |
518 | bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs) | |
519 | { | |
520 | struct bpf_sk_storage_diag *diag; | |
521 | struct nlattr *nla; | |
522 | u32 nr_maps = 0; | |
523 | int rem, err; | |
524 | ||
1f00d375 | 525 | /* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as |
1ed4d924 MKL |
526 | * the map_alloc_check() side also does. |
527 | */ | |
2c78ee89 | 528 | if (!bpf_capable()) |
1ed4d924 MKL |
529 | return ERR_PTR(-EPERM); |
530 | ||
531 | nla_for_each_nested(nla, nla_stgs, rem) { | |
532 | if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD) | |
533 | nr_maps++; | |
534 | } | |
535 | ||
fe0bdaec | 536 | diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL); |
1ed4d924 MKL |
537 | if (!diag) |
538 | return ERR_PTR(-ENOMEM); | |
539 | ||
540 | nla_for_each_nested(nla, nla_stgs, rem) { | |
541 | struct bpf_map *map; | |
542 | int map_fd; | |
543 | ||
544 | if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD) | |
545 | continue; | |
546 | ||
547 | map_fd = nla_get_u32(nla); | |
548 | map = bpf_map_get(map_fd); | |
549 | if (IS_ERR(map)) { | |
550 | err = PTR_ERR(map); | |
551 | goto err_free; | |
552 | } | |
553 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) { | |
554 | bpf_map_put(map); | |
555 | err = -EINVAL; | |
556 | goto err_free; | |
557 | } | |
558 | if (diag_check_dup(diag, map)) { | |
559 | bpf_map_put(map); | |
560 | err = -EEXIST; | |
561 | goto err_free; | |
562 | } | |
563 | diag->maps[diag->nr_maps++] = map; | |
564 | } | |
565 | ||
566 | return diag; | |
567 | ||
568 | err_free: | |
569 | bpf_sk_storage_diag_free(diag); | |
570 | return ERR_PTR(err); | |
571 | } | |
572 | EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc); | |
573 | ||
1f00d375 | 574 | static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb) |
1ed4d924 MKL |
575 | { |
576 | struct nlattr *nla_stg, *nla_value; | |
1f00d375 | 577 | struct bpf_local_storage_map *smap; |
1ed4d924 MKL |
578 | |
579 | /* It cannot exceed max nlattr's payload */ | |
1f00d375 | 580 | BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE); |
1ed4d924 MKL |
581 | |
582 | nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE); | |
583 | if (!nla_stg) | |
584 | return -EMSGSIZE; | |
585 | ||
586 | smap = rcu_dereference(sdata->smap); | |
587 | if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id)) | |
588 | goto errout; | |
589 | ||
590 | nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE, | |
591 | smap->map.value_size, | |
592 | SK_DIAG_BPF_STORAGE_PAD); | |
593 | if (!nla_value) | |
594 | goto errout; | |
595 | ||
596 | if (map_value_has_spin_lock(&smap->map)) | |
597 | copy_map_value_locked(&smap->map, nla_data(nla_value), | |
598 | sdata->data, true); | |
599 | else | |
600 | copy_map_value(&smap->map, nla_data(nla_value), sdata->data); | |
601 | ||
602 | nla_nest_end(skb, nla_stg); | |
603 | return 0; | |
604 | ||
605 | errout: | |
606 | nla_nest_cancel(skb, nla_stg); | |
607 | return -EMSGSIZE; | |
608 | } | |
609 | ||
610 | static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb, | |
611 | int stg_array_type, | |
612 | unsigned int *res_diag_size) | |
613 | { | |
614 | /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */ | |
615 | unsigned int diag_size = nla_total_size(0); | |
1f00d375 KS |
616 | struct bpf_local_storage *sk_storage; |
617 | struct bpf_local_storage_elem *selem; | |
618 | struct bpf_local_storage_map *smap; | |
1ed4d924 MKL |
619 | struct nlattr *nla_stgs; |
620 | unsigned int saved_len; | |
621 | int err = 0; | |
622 | ||
623 | rcu_read_lock(); | |
624 | ||
625 | sk_storage = rcu_dereference(sk->sk_bpf_storage); | |
626 | if (!sk_storage || hlist_empty(&sk_storage->list)) { | |
627 | rcu_read_unlock(); | |
628 | return 0; | |
629 | } | |
630 | ||
631 | nla_stgs = nla_nest_start(skb, stg_array_type); | |
632 | if (!nla_stgs) | |
633 | /* Continue to learn diag_size */ | |
634 | err = -EMSGSIZE; | |
635 | ||
636 | saved_len = skb->len; | |
637 | hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) { | |
638 | smap = rcu_dereference(SDATA(selem)->smap); | |
639 | diag_size += nla_value_size(smap->map.value_size); | |
640 | ||
641 | if (nla_stgs && diag_get(SDATA(selem), skb)) | |
642 | /* Continue to learn diag_size */ | |
643 | err = -EMSGSIZE; | |
644 | } | |
645 | ||
646 | rcu_read_unlock(); | |
647 | ||
648 | if (nla_stgs) { | |
649 | if (saved_len == skb->len) | |
650 | nla_nest_cancel(skb, nla_stgs); | |
651 | else | |
652 | nla_nest_end(skb, nla_stgs); | |
653 | } | |
654 | ||
655 | if (diag_size == nla_total_size(0)) { | |
656 | *res_diag_size = 0; | |
657 | return 0; | |
658 | } | |
659 | ||
660 | *res_diag_size = diag_size; | |
661 | return err; | |
662 | } | |
663 | ||
664 | int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag, | |
665 | struct sock *sk, struct sk_buff *skb, | |
666 | int stg_array_type, | |
667 | unsigned int *res_diag_size) | |
668 | { | |
669 | /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */ | |
670 | unsigned int diag_size = nla_total_size(0); | |
1f00d375 KS |
671 | struct bpf_local_storage *sk_storage; |
672 | struct bpf_local_storage_data *sdata; | |
1ed4d924 MKL |
673 | struct nlattr *nla_stgs; |
674 | unsigned int saved_len; | |
675 | int err = 0; | |
676 | u32 i; | |
677 | ||
678 | *res_diag_size = 0; | |
679 | ||
680 | /* No map has been specified. Dump all. */ | |
681 | if (!diag->nr_maps) | |
682 | return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type, | |
683 | res_diag_size); | |
684 | ||
685 | rcu_read_lock(); | |
686 | sk_storage = rcu_dereference(sk->sk_bpf_storage); | |
687 | if (!sk_storage || hlist_empty(&sk_storage->list)) { | |
688 | rcu_read_unlock(); | |
689 | return 0; | |
690 | } | |
691 | ||
692 | nla_stgs = nla_nest_start(skb, stg_array_type); | |
693 | if (!nla_stgs) | |
694 | /* Continue to learn diag_size */ | |
695 | err = -EMSGSIZE; | |
696 | ||
697 | saved_len = skb->len; | |
698 | for (i = 0; i < diag->nr_maps; i++) { | |
1f00d375 KS |
699 | sdata = bpf_local_storage_lookup(sk_storage, |
700 | (struct bpf_local_storage_map *)diag->maps[i], | |
1ed4d924 MKL |
701 | false); |
702 | ||
703 | if (!sdata) | |
704 | continue; | |
705 | ||
706 | diag_size += nla_value_size(diag->maps[i]->value_size); | |
707 | ||
708 | if (nla_stgs && diag_get(sdata, skb)) | |
709 | /* Continue to learn diag_size */ | |
710 | err = -EMSGSIZE; | |
711 | } | |
712 | rcu_read_unlock(); | |
713 | ||
714 | if (nla_stgs) { | |
715 | if (saved_len == skb->len) | |
716 | nla_nest_cancel(skb, nla_stgs); | |
717 | else | |
718 | nla_nest_end(skb, nla_stgs); | |
719 | } | |
720 | ||
721 | if (diag_size == nla_total_size(0)) { | |
722 | *res_diag_size = 0; | |
723 | return 0; | |
724 | } | |
725 | ||
726 | *res_diag_size = diag_size; | |
727 | return err; | |
728 | } | |
729 | EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put); | |
5ce6e77c YS |
730 | |
731 | struct bpf_iter_seq_sk_storage_map_info { | |
732 | struct bpf_map *map; | |
733 | unsigned int bucket_id; | |
734 | unsigned skip_elems; | |
735 | }; | |
736 | ||
1f00d375 | 737 | static struct bpf_local_storage_elem * |
5ce6e77c | 738 | bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info, |
1f00d375 | 739 | struct bpf_local_storage_elem *prev_selem) |
c69d2ddb | 740 | __acquires(RCU) __releases(RCU) |
5ce6e77c | 741 | { |
1f00d375 KS |
742 | struct bpf_local_storage *sk_storage; |
743 | struct bpf_local_storage_elem *selem; | |
5ce6e77c | 744 | u32 skip_elems = info->skip_elems; |
1f00d375 | 745 | struct bpf_local_storage_map *smap; |
5ce6e77c YS |
746 | u32 bucket_id = info->bucket_id; |
747 | u32 i, count, n_buckets; | |
1f00d375 | 748 | struct bpf_local_storage_map_bucket *b; |
5ce6e77c | 749 | |
1f00d375 | 750 | smap = (struct bpf_local_storage_map *)info->map; |
5ce6e77c YS |
751 | n_buckets = 1U << smap->bucket_log; |
752 | if (bucket_id >= n_buckets) | |
753 | return NULL; | |
754 | ||
755 | /* try to find next selem in the same bucket */ | |
756 | selem = prev_selem; | |
757 | count = 0; | |
758 | while (selem) { | |
c69d2ddb | 759 | selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)), |
1f00d375 | 760 | struct bpf_local_storage_elem, map_node); |
5ce6e77c YS |
761 | if (!selem) { |
762 | /* not found, unlock and go to the next bucket */ | |
763 | b = &smap->buckets[bucket_id++]; | |
c69d2ddb | 764 | rcu_read_unlock(); |
5ce6e77c YS |
765 | skip_elems = 0; |
766 | break; | |
767 | } | |
c69d2ddb | 768 | sk_storage = rcu_dereference(selem->local_storage); |
5ce6e77c YS |
769 | if (sk_storage) { |
770 | info->skip_elems = skip_elems + count; | |
771 | return selem; | |
772 | } | |
773 | count++; | |
774 | } | |
775 | ||
776 | for (i = bucket_id; i < (1U << smap->bucket_log); i++) { | |
777 | b = &smap->buckets[i]; | |
c69d2ddb | 778 | rcu_read_lock(); |
5ce6e77c | 779 | count = 0; |
c69d2ddb YS |
780 | hlist_for_each_entry_rcu(selem, &b->list, map_node) { |
781 | sk_storage = rcu_dereference(selem->local_storage); | |
5ce6e77c YS |
782 | if (sk_storage && count >= skip_elems) { |
783 | info->bucket_id = i; | |
784 | info->skip_elems = count; | |
785 | return selem; | |
786 | } | |
787 | count++; | |
788 | } | |
c69d2ddb | 789 | rcu_read_unlock(); |
5ce6e77c YS |
790 | skip_elems = 0; |
791 | } | |
792 | ||
793 | info->bucket_id = i; | |
794 | info->skip_elems = 0; | |
795 | return NULL; | |
796 | } | |
797 | ||
798 | static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos) | |
799 | { | |
1f00d375 | 800 | struct bpf_local_storage_elem *selem; |
5ce6e77c YS |
801 | |
802 | selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL); | |
803 | if (!selem) | |
804 | return NULL; | |
805 | ||
806 | if (*pos == 0) | |
807 | ++*pos; | |
808 | return selem; | |
809 | } | |
810 | ||
811 | static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v, | |
812 | loff_t *pos) | |
813 | { | |
814 | struct bpf_iter_seq_sk_storage_map_info *info = seq->private; | |
815 | ||
816 | ++*pos; | |
817 | ++info->skip_elems; | |
818 | return bpf_sk_storage_map_seq_find_next(seq->private, v); | |
819 | } | |
820 | ||
821 | struct bpf_iter__bpf_sk_storage_map { | |
822 | __bpf_md_ptr(struct bpf_iter_meta *, meta); | |
823 | __bpf_md_ptr(struct bpf_map *, map); | |
824 | __bpf_md_ptr(struct sock *, sk); | |
825 | __bpf_md_ptr(void *, value); | |
826 | }; | |
827 | ||
828 | DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta, | |
829 | struct bpf_map *map, struct sock *sk, | |
830 | void *value) | |
831 | ||
832 | static int __bpf_sk_storage_map_seq_show(struct seq_file *seq, | |
1f00d375 | 833 | struct bpf_local_storage_elem *selem) |
5ce6e77c YS |
834 | { |
835 | struct bpf_iter_seq_sk_storage_map_info *info = seq->private; | |
836 | struct bpf_iter__bpf_sk_storage_map ctx = {}; | |
1f00d375 | 837 | struct bpf_local_storage *sk_storage; |
5ce6e77c YS |
838 | struct bpf_iter_meta meta; |
839 | struct bpf_prog *prog; | |
840 | int ret = 0; | |
841 | ||
842 | meta.seq = seq; | |
843 | prog = bpf_iter_get_info(&meta, selem == NULL); | |
844 | if (prog) { | |
845 | ctx.meta = &meta; | |
846 | ctx.map = info->map; | |
847 | if (selem) { | |
c69d2ddb | 848 | sk_storage = rcu_dereference(selem->local_storage); |
1f00d375 | 849 | ctx.sk = sk_storage->owner; |
5ce6e77c YS |
850 | ctx.value = SDATA(selem)->data; |
851 | } | |
852 | ret = bpf_iter_run_prog(prog, &ctx); | |
853 | } | |
854 | ||
855 | return ret; | |
856 | } | |
857 | ||
858 | static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v) | |
859 | { | |
860 | return __bpf_sk_storage_map_seq_show(seq, v); | |
861 | } | |
862 | ||
863 | static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v) | |
c69d2ddb | 864 | __releases(RCU) |
5ce6e77c | 865 | { |
c69d2ddb | 866 | if (!v) |
5ce6e77c | 867 | (void)__bpf_sk_storage_map_seq_show(seq, v); |
c69d2ddb YS |
868 | else |
869 | rcu_read_unlock(); | |
5ce6e77c YS |
870 | } |
871 | ||
872 | static int bpf_iter_init_sk_storage_map(void *priv_data, | |
873 | struct bpf_iter_aux_info *aux) | |
874 | { | |
875 | struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data; | |
876 | ||
877 | seq_info->map = aux->map; | |
878 | return 0; | |
879 | } | |
880 | ||
5e7b3020 YS |
881 | static int bpf_iter_attach_map(struct bpf_prog *prog, |
882 | union bpf_iter_link_info *linfo, | |
883 | struct bpf_iter_aux_info *aux) | |
5ce6e77c | 884 | { |
5e7b3020 YS |
885 | struct bpf_map *map; |
886 | int err = -EINVAL; | |
887 | ||
888 | if (!linfo->map.map_fd) | |
889 | return -EBADF; | |
890 | ||
891 | map = bpf_map_get_with_uref(linfo->map.map_fd); | |
892 | if (IS_ERR(map)) | |
893 | return PTR_ERR(map); | |
5ce6e77c YS |
894 | |
895 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
5e7b3020 | 896 | goto put_map; |
5ce6e77c | 897 | |
5e7b3020 YS |
898 | if (prog->aux->max_rdonly_access > map->value_size) { |
899 | err = -EACCES; | |
900 | goto put_map; | |
901 | } | |
5ce6e77c | 902 | |
5e7b3020 | 903 | aux->map = map; |
5ce6e77c | 904 | return 0; |
5e7b3020 YS |
905 | |
906 | put_map: | |
907 | bpf_map_put_with_uref(map); | |
908 | return err; | |
909 | } | |
910 | ||
911 | static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux) | |
912 | { | |
913 | bpf_map_put_with_uref(aux->map); | |
5ce6e77c YS |
914 | } |
915 | ||
916 | static const struct seq_operations bpf_sk_storage_map_seq_ops = { | |
917 | .start = bpf_sk_storage_map_seq_start, | |
918 | .next = bpf_sk_storage_map_seq_next, | |
919 | .stop = bpf_sk_storage_map_seq_stop, | |
920 | .show = bpf_sk_storage_map_seq_show, | |
921 | }; | |
922 | ||
923 | static const struct bpf_iter_seq_info iter_seq_info = { | |
924 | .seq_ops = &bpf_sk_storage_map_seq_ops, | |
925 | .init_seq_private = bpf_iter_init_sk_storage_map, | |
926 | .fini_seq_private = NULL, | |
927 | .seq_priv_size = sizeof(struct bpf_iter_seq_sk_storage_map_info), | |
928 | }; | |
929 | ||
930 | static struct bpf_iter_reg bpf_sk_storage_map_reg_info = { | |
931 | .target = "bpf_sk_storage_map", | |
5e7b3020 YS |
932 | .attach_target = bpf_iter_attach_map, |
933 | .detach_target = bpf_iter_detach_map, | |
b76f2226 YS |
934 | .show_fdinfo = bpf_iter_map_show_fdinfo, |
935 | .fill_link_info = bpf_iter_map_fill_link_info, | |
5ce6e77c YS |
936 | .ctx_arg_info_size = 2, |
937 | .ctx_arg_info = { | |
938 | { offsetof(struct bpf_iter__bpf_sk_storage_map, sk), | |
939 | PTR_TO_BTF_ID_OR_NULL }, | |
940 | { offsetof(struct bpf_iter__bpf_sk_storage_map, value), | |
20b2aff4 | 941 | PTR_TO_BUF | PTR_MAYBE_NULL }, |
5ce6e77c YS |
942 | }, |
943 | .seq_info = &iter_seq_info, | |
944 | }; | |
945 | ||
946 | static int __init bpf_sk_storage_map_iter_init(void) | |
947 | { | |
948 | bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id = | |
949 | btf_sock_ids[BTF_SOCK_TYPE_SOCK]; | |
950 | return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info); | |
951 | } | |
952 | late_initcall(bpf_sk_storage_map_iter_init); |