Commit | Line | Data |
---|---|---|
5b497af4 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
d0003ec0 | 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
d0003ec0 AS |
3 | */ |
4 | #include <linux/bpf.h> | |
3bd916ee | 5 | #include <linux/btf.h> |
aef2feda | 6 | #include <linux/bpf-cgroup.h> |
fda01efc | 7 | #include <linux/cgroup.h> |
d0003ec0 | 8 | #include <linux/rcupdate.h> |
03e69b50 | 9 | #include <linux/random.h> |
c04167ce | 10 | #include <linux/smp.h> |
2d0e30c3 | 11 | #include <linux/topology.h> |
17ca8cbf | 12 | #include <linux/ktime.h> |
ffeedafb AS |
13 | #include <linux/sched.h> |
14 | #include <linux/uidgid.h> | |
f3694e00 | 15 | #include <linux/filter.h> |
d7a4cb9b | 16 | #include <linux/ctype.h> |
5576b991 | 17 | #include <linux/jiffies.h> |
b4490c5c | 18 | #include <linux/pid_namespace.h> |
47e34cb7 | 19 | #include <linux/poison.h> |
b4490c5c | 20 | #include <linux/proc_ns.h> |
d02c48fa | 21 | #include <linux/sched/task.h> |
ff40e510 | 22 | #include <linux/security.h> |
376040e4 | 23 | #include <linux/btf_ids.h> |
958cf2e2 | 24 | #include <linux/bpf_mem_alloc.h> |
d7a4cb9b AI |
25 | |
26 | #include "../../lib/kstrtox.h" | |
d0003ec0 AS |
27 | |
28 | /* If kernel subsystem is allowing eBPF programs to call this function, | |
29 | * inside its own verifier_ops->get_func_proto() callback it should return | |
30 | * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments | |
31 | * | |
32 | * Different map implementations will rely on rcu in map methods | |
33 | * lookup/update/delete, therefore eBPF programs must run under rcu lock | |
34 | * if program is allowed to access maps, so check rcu_read_lock_held in | |
35 | * all three functions. | |
36 | */ | |
f3694e00 | 37 | BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key) |
d0003ec0 | 38 | { |
694cea39 | 39 | WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); |
f3694e00 | 40 | return (unsigned long) map->ops->map_lookup_elem(map, key); |
d0003ec0 AS |
41 | } |
42 | ||
a2c83fff | 43 | const struct bpf_func_proto bpf_map_lookup_elem_proto = { |
3324b584 DB |
44 | .func = bpf_map_lookup_elem, |
45 | .gpl_only = false, | |
36bbef52 | 46 | .pkt_access = true, |
3324b584 DB |
47 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, |
48 | .arg1_type = ARG_CONST_MAP_PTR, | |
49 | .arg2_type = ARG_PTR_TO_MAP_KEY, | |
d0003ec0 AS |
50 | }; |
51 | ||
f3694e00 DB |
52 | BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key, |
53 | void *, value, u64, flags) | |
d0003ec0 | 54 | { |
694cea39 | 55 | WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); |
f3694e00 | 56 | return map->ops->map_update_elem(map, key, value, flags); |
d0003ec0 AS |
57 | } |
58 | ||
a2c83fff | 59 | const struct bpf_func_proto bpf_map_update_elem_proto = { |
3324b584 DB |
60 | .func = bpf_map_update_elem, |
61 | .gpl_only = false, | |
36bbef52 | 62 | .pkt_access = true, |
3324b584 DB |
63 | .ret_type = RET_INTEGER, |
64 | .arg1_type = ARG_CONST_MAP_PTR, | |
65 | .arg2_type = ARG_PTR_TO_MAP_KEY, | |
66 | .arg3_type = ARG_PTR_TO_MAP_VALUE, | |
67 | .arg4_type = ARG_ANYTHING, | |
d0003ec0 AS |
68 | }; |
69 | ||
f3694e00 | 70 | BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key) |
d0003ec0 | 71 | { |
694cea39 | 72 | WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); |
d0003ec0 AS |
73 | return map->ops->map_delete_elem(map, key); |
74 | } | |
75 | ||
a2c83fff | 76 | const struct bpf_func_proto bpf_map_delete_elem_proto = { |
3324b584 DB |
77 | .func = bpf_map_delete_elem, |
78 | .gpl_only = false, | |
36bbef52 | 79 | .pkt_access = true, |
3324b584 DB |
80 | .ret_type = RET_INTEGER, |
81 | .arg1_type = ARG_CONST_MAP_PTR, | |
82 | .arg2_type = ARG_PTR_TO_MAP_KEY, | |
d0003ec0 | 83 | }; |
03e69b50 | 84 | |
f1a2e44a MV |
85 | BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags) |
86 | { | |
87 | return map->ops->map_push_elem(map, value, flags); | |
88 | } | |
89 | ||
90 | const struct bpf_func_proto bpf_map_push_elem_proto = { | |
91 | .func = bpf_map_push_elem, | |
92 | .gpl_only = false, | |
93 | .pkt_access = true, | |
94 | .ret_type = RET_INTEGER, | |
95 | .arg1_type = ARG_CONST_MAP_PTR, | |
96 | .arg2_type = ARG_PTR_TO_MAP_VALUE, | |
97 | .arg3_type = ARG_ANYTHING, | |
98 | }; | |
99 | ||
100 | BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value) | |
101 | { | |
102 | return map->ops->map_pop_elem(map, value); | |
103 | } | |
104 | ||
105 | const struct bpf_func_proto bpf_map_pop_elem_proto = { | |
106 | .func = bpf_map_pop_elem, | |
107 | .gpl_only = false, | |
f1a2e44a MV |
108 | .ret_type = RET_INTEGER, |
109 | .arg1_type = ARG_CONST_MAP_PTR, | |
16d1e00c | 110 | .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT, |
f1a2e44a MV |
111 | }; |
112 | ||
113 | BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value) | |
114 | { | |
115 | return map->ops->map_peek_elem(map, value); | |
116 | } | |
117 | ||
118 | const struct bpf_func_proto bpf_map_peek_elem_proto = { | |
301a33d5 | 119 | .func = bpf_map_peek_elem, |
f1a2e44a | 120 | .gpl_only = false, |
f1a2e44a MV |
121 | .ret_type = RET_INTEGER, |
122 | .arg1_type = ARG_CONST_MAP_PTR, | |
16d1e00c | 123 | .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT, |
f1a2e44a MV |
124 | }; |
125 | ||
07343110 FZ |
126 | BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu) |
127 | { | |
128 | WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); | |
129 | return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu); | |
130 | } | |
131 | ||
132 | const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = { | |
133 | .func = bpf_map_lookup_percpu_elem, | |
134 | .gpl_only = false, | |
135 | .pkt_access = true, | |
136 | .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
137 | .arg1_type = ARG_CONST_MAP_PTR, | |
138 | .arg2_type = ARG_PTR_TO_MAP_KEY, | |
139 | .arg3_type = ARG_ANYTHING, | |
140 | }; | |
141 | ||
03e69b50 | 142 | const struct bpf_func_proto bpf_get_prandom_u32_proto = { |
3ad00405 | 143 | .func = bpf_user_rnd_u32, |
03e69b50 DB |
144 | .gpl_only = false, |
145 | .ret_type = RET_INTEGER, | |
146 | }; | |
c04167ce | 147 | |
f3694e00 | 148 | BPF_CALL_0(bpf_get_smp_processor_id) |
c04167ce | 149 | { |
80b48c44 | 150 | return smp_processor_id(); |
c04167ce DB |
151 | } |
152 | ||
153 | const struct bpf_func_proto bpf_get_smp_processor_id_proto = { | |
154 | .func = bpf_get_smp_processor_id, | |
155 | .gpl_only = false, | |
156 | .ret_type = RET_INTEGER, | |
157 | }; | |
17ca8cbf | 158 | |
2d0e30c3 DB |
159 | BPF_CALL_0(bpf_get_numa_node_id) |
160 | { | |
161 | return numa_node_id(); | |
162 | } | |
163 | ||
164 | const struct bpf_func_proto bpf_get_numa_node_id_proto = { | |
165 | .func = bpf_get_numa_node_id, | |
166 | .gpl_only = false, | |
167 | .ret_type = RET_INTEGER, | |
168 | }; | |
169 | ||
f3694e00 | 170 | BPF_CALL_0(bpf_ktime_get_ns) |
17ca8cbf DB |
171 | { |
172 | /* NMI safe access to clock monotonic */ | |
173 | return ktime_get_mono_fast_ns(); | |
174 | } | |
175 | ||
176 | const struct bpf_func_proto bpf_ktime_get_ns_proto = { | |
177 | .func = bpf_ktime_get_ns, | |
082b57e3 | 178 | .gpl_only = false, |
17ca8cbf DB |
179 | .ret_type = RET_INTEGER, |
180 | }; | |
ffeedafb | 181 | |
71d19214 MŻ |
182 | BPF_CALL_0(bpf_ktime_get_boot_ns) |
183 | { | |
184 | /* NMI safe access to clock boottime */ | |
185 | return ktime_get_boot_fast_ns(); | |
186 | } | |
187 | ||
188 | const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = { | |
189 | .func = bpf_ktime_get_boot_ns, | |
190 | .gpl_only = false, | |
191 | .ret_type = RET_INTEGER, | |
192 | }; | |
193 | ||
d0551261 DB |
194 | BPF_CALL_0(bpf_ktime_get_coarse_ns) |
195 | { | |
196 | return ktime_get_coarse_ns(); | |
197 | } | |
198 | ||
199 | const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = { | |
200 | .func = bpf_ktime_get_coarse_ns, | |
201 | .gpl_only = false, | |
202 | .ret_type = RET_INTEGER, | |
203 | }; | |
204 | ||
c8996c98 JDB |
205 | BPF_CALL_0(bpf_ktime_get_tai_ns) |
206 | { | |
207 | /* NMI safe access to clock tai */ | |
208 | return ktime_get_tai_fast_ns(); | |
209 | } | |
210 | ||
211 | const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = { | |
212 | .func = bpf_ktime_get_tai_ns, | |
213 | .gpl_only = false, | |
214 | .ret_type = RET_INTEGER, | |
215 | }; | |
216 | ||
f3694e00 | 217 | BPF_CALL_0(bpf_get_current_pid_tgid) |
ffeedafb AS |
218 | { |
219 | struct task_struct *task = current; | |
220 | ||
6088b582 | 221 | if (unlikely(!task)) |
ffeedafb AS |
222 | return -EINVAL; |
223 | ||
224 | return (u64) task->tgid << 32 | task->pid; | |
225 | } | |
226 | ||
227 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto = { | |
228 | .func = bpf_get_current_pid_tgid, | |
229 | .gpl_only = false, | |
230 | .ret_type = RET_INTEGER, | |
231 | }; | |
232 | ||
f3694e00 | 233 | BPF_CALL_0(bpf_get_current_uid_gid) |
ffeedafb AS |
234 | { |
235 | struct task_struct *task = current; | |
236 | kuid_t uid; | |
237 | kgid_t gid; | |
238 | ||
6088b582 | 239 | if (unlikely(!task)) |
ffeedafb AS |
240 | return -EINVAL; |
241 | ||
242 | current_uid_gid(&uid, &gid); | |
243 | return (u64) from_kgid(&init_user_ns, gid) << 32 | | |
6088b582 | 244 | from_kuid(&init_user_ns, uid); |
ffeedafb AS |
245 | } |
246 | ||
247 | const struct bpf_func_proto bpf_get_current_uid_gid_proto = { | |
248 | .func = bpf_get_current_uid_gid, | |
249 | .gpl_only = false, | |
250 | .ret_type = RET_INTEGER, | |
251 | }; | |
252 | ||
f3694e00 | 253 | BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size) |
ffeedafb AS |
254 | { |
255 | struct task_struct *task = current; | |
ffeedafb | 256 | |
074f528e DB |
257 | if (unlikely(!task)) |
258 | goto err_clear; | |
ffeedafb | 259 | |
03b9c7fa | 260 | /* Verifier guarantees that size > 0 */ |
f3f21349 | 261 | strscpy_pad(buf, task->comm, size); |
ffeedafb | 262 | return 0; |
074f528e DB |
263 | err_clear: |
264 | memset(buf, 0, size); | |
265 | return -EINVAL; | |
ffeedafb AS |
266 | } |
267 | ||
268 | const struct bpf_func_proto bpf_get_current_comm_proto = { | |
269 | .func = bpf_get_current_comm, | |
270 | .gpl_only = false, | |
271 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
272 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
273 | .arg2_type = ARG_CONST_SIZE, | |
ffeedafb | 274 | }; |
bf6fa2c8 | 275 | |
d83525ca AS |
276 | #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK) |
277 | ||
278 | static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) | |
279 | { | |
280 | arch_spinlock_t *l = (void *)lock; | |
281 | union { | |
282 | __u32 val; | |
283 | arch_spinlock_t lock; | |
284 | } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED }; | |
285 | ||
286 | compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0"); | |
287 | BUILD_BUG_ON(sizeof(*l) != sizeof(__u32)); | |
288 | BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32)); | |
5861d1e8 | 289 | preempt_disable(); |
d83525ca AS |
290 | arch_spin_lock(l); |
291 | } | |
292 | ||
293 | static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) | |
294 | { | |
295 | arch_spinlock_t *l = (void *)lock; | |
296 | ||
297 | arch_spin_unlock(l); | |
5861d1e8 | 298 | preempt_enable(); |
d83525ca AS |
299 | } |
300 | ||
301 | #else | |
302 | ||
303 | static inline void __bpf_spin_lock(struct bpf_spin_lock *lock) | |
304 | { | |
305 | atomic_t *l = (void *)lock; | |
306 | ||
307 | BUILD_BUG_ON(sizeof(*l) != sizeof(*lock)); | |
308 | do { | |
309 | atomic_cond_read_relaxed(l, !VAL); | |
310 | } while (atomic_xchg(l, 1)); | |
311 | } | |
312 | ||
313 | static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock) | |
314 | { | |
315 | atomic_t *l = (void *)lock; | |
316 | ||
317 | atomic_set_release(l, 0); | |
318 | } | |
319 | ||
320 | #endif | |
321 | ||
322 | static DEFINE_PER_CPU(unsigned long, irqsave_flags); | |
323 | ||
c1b3fed3 | 324 | static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock) |
d83525ca AS |
325 | { |
326 | unsigned long flags; | |
327 | ||
328 | local_irq_save(flags); | |
329 | __bpf_spin_lock(lock); | |
330 | __this_cpu_write(irqsave_flags, flags); | |
c1b3fed3 AS |
331 | } |
332 | ||
333 | notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock) | |
334 | { | |
335 | __bpf_spin_lock_irqsave(lock); | |
d83525ca AS |
336 | return 0; |
337 | } | |
338 | ||
339 | const struct bpf_func_proto bpf_spin_lock_proto = { | |
340 | .func = bpf_spin_lock, | |
341 | .gpl_only = false, | |
342 | .ret_type = RET_VOID, | |
343 | .arg1_type = ARG_PTR_TO_SPIN_LOCK, | |
4e814da0 | 344 | .arg1_btf_id = BPF_PTR_POISON, |
d83525ca AS |
345 | }; |
346 | ||
c1b3fed3 | 347 | static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock) |
d83525ca AS |
348 | { |
349 | unsigned long flags; | |
350 | ||
351 | flags = __this_cpu_read(irqsave_flags); | |
352 | __bpf_spin_unlock(lock); | |
353 | local_irq_restore(flags); | |
c1b3fed3 AS |
354 | } |
355 | ||
356 | notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock) | |
357 | { | |
358 | __bpf_spin_unlock_irqrestore(lock); | |
d83525ca AS |
359 | return 0; |
360 | } | |
361 | ||
362 | const struct bpf_func_proto bpf_spin_unlock_proto = { | |
363 | .func = bpf_spin_unlock, | |
364 | .gpl_only = false, | |
365 | .ret_type = RET_VOID, | |
366 | .arg1_type = ARG_PTR_TO_SPIN_LOCK, | |
4e814da0 | 367 | .arg1_btf_id = BPF_PTR_POISON, |
d83525ca AS |
368 | }; |
369 | ||
96049f3a AS |
370 | void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, |
371 | bool lock_src) | |
372 | { | |
373 | struct bpf_spin_lock *lock; | |
374 | ||
375 | if (lock_src) | |
db559117 | 376 | lock = src + map->record->spin_lock_off; |
96049f3a | 377 | else |
db559117 | 378 | lock = dst + map->record->spin_lock_off; |
96049f3a | 379 | preempt_disable(); |
c1b3fed3 | 380 | __bpf_spin_lock_irqsave(lock); |
96049f3a | 381 | copy_map_value(map, dst, src); |
c1b3fed3 | 382 | __bpf_spin_unlock_irqrestore(lock); |
96049f3a AS |
383 | preempt_enable(); |
384 | } | |
385 | ||
5576b991 MKL |
386 | BPF_CALL_0(bpf_jiffies64) |
387 | { | |
388 | return get_jiffies_64(); | |
389 | } | |
390 | ||
391 | const struct bpf_func_proto bpf_jiffies64_proto = { | |
392 | .func = bpf_jiffies64, | |
393 | .gpl_only = false, | |
394 | .ret_type = RET_INTEGER, | |
395 | }; | |
396 | ||
bf6fa2c8 YS |
397 | #ifdef CONFIG_CGROUPS |
398 | BPF_CALL_0(bpf_get_current_cgroup_id) | |
399 | { | |
2d3a1e36 YS |
400 | struct cgroup *cgrp; |
401 | u64 cgrp_id; | |
bf6fa2c8 | 402 | |
2d3a1e36 YS |
403 | rcu_read_lock(); |
404 | cgrp = task_dfl_cgroup(current); | |
405 | cgrp_id = cgroup_id(cgrp); | |
406 | rcu_read_unlock(); | |
407 | ||
408 | return cgrp_id; | |
bf6fa2c8 YS |
409 | } |
410 | ||
411 | const struct bpf_func_proto bpf_get_current_cgroup_id_proto = { | |
412 | .func = bpf_get_current_cgroup_id, | |
413 | .gpl_only = false, | |
414 | .ret_type = RET_INTEGER, | |
415 | }; | |
cd339431 | 416 | |
0f09abd1 DB |
417 | BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level) |
418 | { | |
2d3a1e36 | 419 | struct cgroup *cgrp; |
0f09abd1 | 420 | struct cgroup *ancestor; |
2d3a1e36 | 421 | u64 cgrp_id; |
0f09abd1 | 422 | |
2d3a1e36 YS |
423 | rcu_read_lock(); |
424 | cgrp = task_dfl_cgroup(current); | |
0f09abd1 | 425 | ancestor = cgroup_ancestor(cgrp, ancestor_level); |
2d3a1e36 YS |
426 | cgrp_id = ancestor ? cgroup_id(ancestor) : 0; |
427 | rcu_read_unlock(); | |
428 | ||
429 | return cgrp_id; | |
0f09abd1 DB |
430 | } |
431 | ||
432 | const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = { | |
433 | .func = bpf_get_current_ancestor_cgroup_id, | |
434 | .gpl_only = false, | |
435 | .ret_type = RET_INTEGER, | |
436 | .arg1_type = ARG_ANYTHING, | |
437 | }; | |
8a67f2de | 438 | #endif /* CONFIG_CGROUPS */ |
0f09abd1 | 439 | |
d7a4cb9b AI |
440 | #define BPF_STRTOX_BASE_MASK 0x1F |
441 | ||
442 | static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags, | |
443 | unsigned long long *res, bool *is_negative) | |
444 | { | |
445 | unsigned int base = flags & BPF_STRTOX_BASE_MASK; | |
446 | const char *cur_buf = buf; | |
447 | size_t cur_len = buf_len; | |
448 | unsigned int consumed; | |
449 | size_t val_len; | |
450 | char str[64]; | |
451 | ||
452 | if (!buf || !buf_len || !res || !is_negative) | |
453 | return -EINVAL; | |
454 | ||
455 | if (base != 0 && base != 8 && base != 10 && base != 16) | |
456 | return -EINVAL; | |
457 | ||
458 | if (flags & ~BPF_STRTOX_BASE_MASK) | |
459 | return -EINVAL; | |
460 | ||
461 | while (cur_buf < buf + buf_len && isspace(*cur_buf)) | |
462 | ++cur_buf; | |
463 | ||
464 | *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-'); | |
465 | if (*is_negative) | |
466 | ++cur_buf; | |
467 | ||
468 | consumed = cur_buf - buf; | |
469 | cur_len -= consumed; | |
470 | if (!cur_len) | |
471 | return -EINVAL; | |
472 | ||
473 | cur_len = min(cur_len, sizeof(str) - 1); | |
474 | memcpy(str, cur_buf, cur_len); | |
475 | str[cur_len] = '\0'; | |
476 | cur_buf = str; | |
477 | ||
478 | cur_buf = _parse_integer_fixup_radix(cur_buf, &base); | |
479 | val_len = _parse_integer(cur_buf, base, res); | |
480 | ||
481 | if (val_len & KSTRTOX_OVERFLOW) | |
482 | return -ERANGE; | |
483 | ||
484 | if (val_len == 0) | |
485 | return -EINVAL; | |
486 | ||
487 | cur_buf += val_len; | |
488 | consumed += cur_buf - str; | |
489 | ||
490 | return consumed; | |
491 | } | |
492 | ||
493 | static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags, | |
494 | long long *res) | |
495 | { | |
496 | unsigned long long _res; | |
497 | bool is_negative; | |
498 | int err; | |
499 | ||
500 | err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); | |
501 | if (err < 0) | |
502 | return err; | |
503 | if (is_negative) { | |
504 | if ((long long)-_res > 0) | |
505 | return -ERANGE; | |
506 | *res = -_res; | |
507 | } else { | |
508 | if ((long long)_res < 0) | |
509 | return -ERANGE; | |
510 | *res = _res; | |
511 | } | |
512 | return err; | |
513 | } | |
514 | ||
515 | BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags, | |
516 | long *, res) | |
517 | { | |
518 | long long _res; | |
519 | int err; | |
520 | ||
521 | err = __bpf_strtoll(buf, buf_len, flags, &_res); | |
522 | if (err < 0) | |
523 | return err; | |
524 | if (_res != (long)_res) | |
525 | return -ERANGE; | |
526 | *res = _res; | |
527 | return err; | |
528 | } | |
529 | ||
530 | const struct bpf_func_proto bpf_strtol_proto = { | |
531 | .func = bpf_strtol, | |
532 | .gpl_only = false, | |
533 | .ret_type = RET_INTEGER, | |
216e3cd2 | 534 | .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, |
d7a4cb9b AI |
535 | .arg2_type = ARG_CONST_SIZE, |
536 | .arg3_type = ARG_ANYTHING, | |
537 | .arg4_type = ARG_PTR_TO_LONG, | |
538 | }; | |
539 | ||
540 | BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags, | |
541 | unsigned long *, res) | |
542 | { | |
543 | unsigned long long _res; | |
544 | bool is_negative; | |
545 | int err; | |
546 | ||
547 | err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative); | |
548 | if (err < 0) | |
549 | return err; | |
550 | if (is_negative) | |
551 | return -EINVAL; | |
552 | if (_res != (unsigned long)_res) | |
553 | return -ERANGE; | |
554 | *res = _res; | |
555 | return err; | |
556 | } | |
557 | ||
558 | const struct bpf_func_proto bpf_strtoul_proto = { | |
559 | .func = bpf_strtoul, | |
560 | .gpl_only = false, | |
561 | .ret_type = RET_INTEGER, | |
216e3cd2 | 562 | .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, |
d7a4cb9b AI |
563 | .arg2_type = ARG_CONST_SIZE, |
564 | .arg3_type = ARG_ANYTHING, | |
565 | .arg4_type = ARG_PTR_TO_LONG, | |
566 | }; | |
b4490c5c | 567 | |
c5fb1993 HT |
568 | BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2) |
569 | { | |
570 | return strncmp(s1, s2, s1_sz); | |
571 | } | |
572 | ||
dc368e1c | 573 | static const struct bpf_func_proto bpf_strncmp_proto = { |
c5fb1993 HT |
574 | .func = bpf_strncmp, |
575 | .gpl_only = false, | |
576 | .ret_type = RET_INTEGER, | |
c9267aa8 | 577 | .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, |
c5fb1993 HT |
578 | .arg2_type = ARG_CONST_SIZE, |
579 | .arg3_type = ARG_PTR_TO_CONST_STR, | |
580 | }; | |
581 | ||
b4490c5c CN |
582 | BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino, |
583 | struct bpf_pidns_info *, nsdata, u32, size) | |
584 | { | |
585 | struct task_struct *task = current; | |
586 | struct pid_namespace *pidns; | |
587 | int err = -EINVAL; | |
588 | ||
589 | if (unlikely(size != sizeof(struct bpf_pidns_info))) | |
590 | goto clear; | |
591 | ||
592 | if (unlikely((u64)(dev_t)dev != dev)) | |
593 | goto clear; | |
594 | ||
595 | if (unlikely(!task)) | |
596 | goto clear; | |
597 | ||
598 | pidns = task_active_pid_ns(task); | |
599 | if (unlikely(!pidns)) { | |
600 | err = -ENOENT; | |
601 | goto clear; | |
602 | } | |
603 | ||
604 | if (!ns_match(&pidns->ns, (dev_t)dev, ino)) | |
605 | goto clear; | |
606 | ||
607 | nsdata->pid = task_pid_nr_ns(task, pidns); | |
608 | nsdata->tgid = task_tgid_nr_ns(task, pidns); | |
609 | return 0; | |
610 | clear: | |
611 | memset((void *)nsdata, 0, (size_t) size); | |
612 | return err; | |
613 | } | |
614 | ||
615 | const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = { | |
616 | .func = bpf_get_ns_current_pid_tgid, | |
617 | .gpl_only = false, | |
618 | .ret_type = RET_INTEGER, | |
619 | .arg1_type = ARG_ANYTHING, | |
620 | .arg2_type = ARG_ANYTHING, | |
621 | .arg3_type = ARG_PTR_TO_UNINIT_MEM, | |
622 | .arg4_type = ARG_CONST_SIZE, | |
623 | }; | |
6890896b SF |
624 | |
625 | static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = { | |
626 | .func = bpf_get_raw_cpu_id, | |
627 | .gpl_only = false, | |
628 | .ret_type = RET_INTEGER, | |
629 | }; | |
630 | ||
631 | BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map, | |
632 | u64, flags, void *, data, u64, size) | |
633 | { | |
634 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) | |
635 | return -EINVAL; | |
636 | ||
637 | return bpf_event_output(map, flags, data, size, NULL, 0, NULL); | |
638 | } | |
639 | ||
640 | const struct bpf_func_proto bpf_event_output_data_proto = { | |
641 | .func = bpf_event_output_data, | |
642 | .gpl_only = true, | |
643 | .ret_type = RET_INTEGER, | |
644 | .arg1_type = ARG_PTR_TO_CTX, | |
645 | .arg2_type = ARG_CONST_MAP_PTR, | |
646 | .arg3_type = ARG_ANYTHING, | |
216e3cd2 | 647 | .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, |
6890896b SF |
648 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
649 | }; | |
650 | ||
07be4c4a AS |
651 | BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size, |
652 | const void __user *, user_ptr) | |
653 | { | |
654 | int ret = copy_from_user(dst, user_ptr, size); | |
655 | ||
656 | if (unlikely(ret)) { | |
657 | memset(dst, 0, size); | |
658 | ret = -EFAULT; | |
659 | } | |
660 | ||
661 | return ret; | |
662 | } | |
663 | ||
664 | const struct bpf_func_proto bpf_copy_from_user_proto = { | |
665 | .func = bpf_copy_from_user, | |
666 | .gpl_only = false, | |
01685c5b | 667 | .might_sleep = true, |
07be4c4a AS |
668 | .ret_type = RET_INTEGER, |
669 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
670 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
671 | .arg3_type = ARG_ANYTHING, | |
672 | }; | |
673 | ||
376040e4 KY |
674 | BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size, |
675 | const void __user *, user_ptr, struct task_struct *, tsk, u64, flags) | |
676 | { | |
677 | int ret; | |
678 | ||
679 | /* flags is not used yet */ | |
680 | if (unlikely(flags)) | |
681 | return -EINVAL; | |
682 | ||
683 | if (unlikely(!size)) | |
684 | return 0; | |
685 | ||
686 | ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0); | |
687 | if (ret == size) | |
688 | return 0; | |
689 | ||
690 | memset(dst, 0, size); | |
691 | /* Return -EFAULT for partial read */ | |
692 | return ret < 0 ? ret : -EFAULT; | |
693 | } | |
694 | ||
695 | const struct bpf_func_proto bpf_copy_from_user_task_proto = { | |
696 | .func = bpf_copy_from_user_task, | |
0407a65f | 697 | .gpl_only = true, |
01685c5b | 698 | .might_sleep = true, |
376040e4 KY |
699 | .ret_type = RET_INTEGER, |
700 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
701 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
702 | .arg3_type = ARG_ANYTHING, | |
703 | .arg4_type = ARG_PTR_TO_BTF_ID, | |
704 | .arg4_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK], | |
705 | .arg5_type = ARG_ANYTHING | |
706 | }; | |
707 | ||
eaa6bcb7 HL |
708 | BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu) |
709 | { | |
710 | if (cpu >= nr_cpu_ids) | |
711 | return (unsigned long)NULL; | |
712 | ||
713 | return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu); | |
714 | } | |
715 | ||
716 | const struct bpf_func_proto bpf_per_cpu_ptr_proto = { | |
717 | .func = bpf_per_cpu_ptr, | |
718 | .gpl_only = false, | |
34d3a78c | 719 | .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY, |
eaa6bcb7 HL |
720 | .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID, |
721 | .arg2_type = ARG_ANYTHING, | |
722 | }; | |
723 | ||
63d9b80d HL |
724 | BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr) |
725 | { | |
726 | return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr); | |
727 | } | |
728 | ||
729 | const struct bpf_func_proto bpf_this_cpu_ptr_proto = { | |
730 | .func = bpf_this_cpu_ptr, | |
731 | .gpl_only = false, | |
34d3a78c | 732 | .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY, |
63d9b80d HL |
733 | .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID, |
734 | }; | |
735 | ||
d9c9e4db FR |
736 | static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype, |
737 | size_t bufsz) | |
738 | { | |
739 | void __user *user_ptr = (__force void __user *)unsafe_ptr; | |
740 | ||
741 | buf[0] = 0; | |
742 | ||
743 | switch (fmt_ptype) { | |
744 | case 's': | |
745 | #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE | |
746 | if ((unsigned long)unsafe_ptr < TASK_SIZE) | |
747 | return strncpy_from_user_nofault(buf, user_ptr, bufsz); | |
748 | fallthrough; | |
749 | #endif | |
750 | case 'k': | |
751 | return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz); | |
752 | case 'u': | |
753 | return strncpy_from_user_nofault(buf, user_ptr, bufsz); | |
754 | } | |
755 | ||
756 | return -EINVAL; | |
757 | } | |
758 | ||
8afcc19f FR |
759 | /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary |
760 | * arguments representation. | |
d9c9e4db | 761 | */ |
e2bb9e01 | 762 | #define MAX_BPRINTF_BIN_ARGS 512 |
d9c9e4db | 763 | |
e2d5b2bb | 764 | /* Support executing three nested bprintf helper calls on a given CPU */ |
0af02eb2 | 765 | #define MAX_BPRINTF_NEST_LEVEL 3 |
e2d5b2bb | 766 | struct bpf_bprintf_buffers { |
e2bb9e01 JO |
767 | char bin_args[MAX_BPRINTF_BIN_ARGS]; |
768 | char buf[MAX_BPRINTF_BUF]; | |
d9c9e4db | 769 | }; |
e2bb9e01 JO |
770 | |
771 | static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs); | |
e2d5b2bb | 772 | static DEFINE_PER_CPU(int, bpf_bprintf_nest_level); |
d9c9e4db | 773 | |
e2bb9e01 | 774 | static int try_get_buffers(struct bpf_bprintf_buffers **bufs) |
d9c9e4db | 775 | { |
e2d5b2bb | 776 | int nest_level; |
d9c9e4db | 777 | |
d9c9e4db | 778 | preempt_disable(); |
e2d5b2bb | 779 | nest_level = this_cpu_inc_return(bpf_bprintf_nest_level); |
0af02eb2 | 780 | if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) { |
e2d5b2bb | 781 | this_cpu_dec(bpf_bprintf_nest_level); |
d9c9e4db FR |
782 | preempt_enable(); |
783 | return -EBUSY; | |
784 | } | |
e2bb9e01 | 785 | *bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]); |
d9c9e4db FR |
786 | |
787 | return 0; | |
788 | } | |
789 | ||
f19a4050 | 790 | void bpf_bprintf_cleanup(struct bpf_bprintf_data *data) |
d9c9e4db | 791 | { |
e2bb9e01 | 792 | if (!data->bin_args && !data->buf) |
f19a4050 JO |
793 | return; |
794 | if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0)) | |
795 | return; | |
796 | this_cpu_dec(bpf_bprintf_nest_level); | |
797 | preempt_enable(); | |
d9c9e4db FR |
798 | } |
799 | ||
800 | /* | |
48cac3f4 | 801 | * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers |
d9c9e4db FR |
802 | * |
803 | * Returns a negative value if fmt is an invalid format string or 0 otherwise. | |
804 | * | |
805 | * This can be used in two ways: | |
78aa1cc9 | 806 | * - Format string verification only: when data->get_bin_args is false |
d9c9e4db | 807 | * - Arguments preparation: in addition to the above verification, it writes in |
78aa1cc9 JO |
808 | * data->bin_args a binary representation of arguments usable by bstr_printf |
809 | * where pointers from BPF have been sanitized. | |
d9c9e4db FR |
810 | * |
811 | * In argument preparation mode, if 0 is returned, safe temporary buffers are | |
48cac3f4 | 812 | * allocated and bpf_bprintf_cleanup should be called to free them after use. |
d9c9e4db | 813 | */ |
48cac3f4 | 814 | int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args, |
78aa1cc9 | 815 | u32 num_args, struct bpf_bprintf_data *data) |
48cac3f4 | 816 | { |
e2bb9e01 | 817 | bool get_buffers = (data->get_bin_args && num_args) || data->get_buf; |
48cac3f4 | 818 | char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end; |
e2bb9e01 | 819 | struct bpf_bprintf_buffers *buffers = NULL; |
48cac3f4 FR |
820 | size_t sizeof_cur_arg, sizeof_cur_ip; |
821 | int err, i, num_spec = 0; | |
d9c9e4db | 822 | u64 cur_arg; |
48cac3f4 | 823 | char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX"; |
d9c9e4db FR |
824 | |
825 | fmt_end = strnchr(fmt, fmt_size, 0); | |
826 | if (!fmt_end) | |
827 | return -EINVAL; | |
828 | fmt_size = fmt_end - fmt; | |
829 | ||
e2bb9e01 JO |
830 | if (get_buffers && try_get_buffers(&buffers)) |
831 | return -EBUSY; | |
48cac3f4 | 832 | |
e2bb9e01 JO |
833 | if (data->get_bin_args) { |
834 | if (num_args) | |
835 | tmp_buf = buffers->bin_args; | |
836 | tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS; | |
78aa1cc9 | 837 | data->bin_args = (u32 *)tmp_buf; |
48cac3f4 FR |
838 | } |
839 | ||
e2bb9e01 JO |
840 | if (data->get_buf) |
841 | data->buf = buffers->buf; | |
842 | ||
d9c9e4db FR |
843 | for (i = 0; i < fmt_size; i++) { |
844 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) { | |
845 | err = -EINVAL; | |
48cac3f4 | 846 | goto out; |
d9c9e4db FR |
847 | } |
848 | ||
849 | if (fmt[i] != '%') | |
850 | continue; | |
851 | ||
852 | if (fmt[i + 1] == '%') { | |
853 | i++; | |
854 | continue; | |
855 | } | |
856 | ||
857 | if (num_spec >= num_args) { | |
858 | err = -EINVAL; | |
48cac3f4 | 859 | goto out; |
d9c9e4db FR |
860 | } |
861 | ||
862 | /* The string is zero-terminated so if fmt[i] != 0, we can | |
863 | * always access fmt[i + 1], in the worst case it will be a 0 | |
864 | */ | |
865 | i++; | |
866 | ||
867 | /* skip optional "[0 +-][num]" width formatting field */ | |
868 | while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' || | |
869 | fmt[i] == ' ') | |
870 | i++; | |
871 | if (fmt[i] >= '1' && fmt[i] <= '9') { | |
872 | i++; | |
873 | while (fmt[i] >= '0' && fmt[i] <= '9') | |
874 | i++; | |
875 | } | |
876 | ||
877 | if (fmt[i] == 'p') { | |
48cac3f4 | 878 | sizeof_cur_arg = sizeof(long); |
d9c9e4db FR |
879 | |
880 | if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') && | |
881 | fmt[i + 2] == 's') { | |
882 | fmt_ptype = fmt[i + 1]; | |
883 | i += 2; | |
884 | goto fmt_str; | |
885 | } | |
886 | ||
887 | if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) || | |
888 | ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' || | |
48cac3f4 FR |
889 | fmt[i + 1] == 'x' || fmt[i + 1] == 's' || |
890 | fmt[i + 1] == 'S') { | |
d9c9e4db | 891 | /* just kernel pointers */ |
48cac3f4 | 892 | if (tmp_buf) |
d9c9e4db | 893 | cur_arg = raw_args[num_spec]; |
48cac3f4 FR |
894 | i++; |
895 | goto nocopy_fmt; | |
896 | } | |
897 | ||
898 | if (fmt[i + 1] == 'B') { | |
899 | if (tmp_buf) { | |
900 | err = snprintf(tmp_buf, | |
901 | (tmp_buf_end - tmp_buf), | |
902 | "%pB", | |
903 | (void *)(long)raw_args[num_spec]); | |
904 | tmp_buf += (err + 1); | |
905 | } | |
906 | ||
907 | i++; | |
908 | num_spec++; | |
909 | continue; | |
d9c9e4db FR |
910 | } |
911 | ||
912 | /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */ | |
913 | if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') || | |
914 | (fmt[i + 2] != '4' && fmt[i + 2] != '6')) { | |
915 | err = -EINVAL; | |
48cac3f4 | 916 | goto out; |
d9c9e4db FR |
917 | } |
918 | ||
919 | i += 2; | |
48cac3f4 FR |
920 | if (!tmp_buf) |
921 | goto nocopy_fmt; | |
d9c9e4db | 922 | |
48cac3f4 FR |
923 | sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16; |
924 | if (tmp_buf_end - tmp_buf < sizeof_cur_ip) { | |
d9c9e4db | 925 | err = -ENOSPC; |
48cac3f4 | 926 | goto out; |
d9c9e4db FR |
927 | } |
928 | ||
929 | unsafe_ptr = (char *)(long)raw_args[num_spec]; | |
48cac3f4 FR |
930 | err = copy_from_kernel_nofault(cur_ip, unsafe_ptr, |
931 | sizeof_cur_ip); | |
d9c9e4db | 932 | if (err < 0) |
48cac3f4 | 933 | memset(cur_ip, 0, sizeof_cur_ip); |
d9c9e4db | 934 | |
48cac3f4 FR |
935 | /* hack: bstr_printf expects IP addresses to be |
936 | * pre-formatted as strings, ironically, the easiest way | |
937 | * to do that is to call snprintf. | |
938 | */ | |
939 | ip_spec[2] = fmt[i - 1]; | |
940 | ip_spec[3] = fmt[i]; | |
941 | err = snprintf(tmp_buf, tmp_buf_end - tmp_buf, | |
942 | ip_spec, &cur_ip); | |
943 | ||
944 | tmp_buf += err + 1; | |
945 | num_spec++; | |
946 | ||
947 | continue; | |
d9c9e4db | 948 | } else if (fmt[i] == 's') { |
d9c9e4db FR |
949 | fmt_ptype = fmt[i]; |
950 | fmt_str: | |
951 | if (fmt[i + 1] != 0 && | |
952 | !isspace(fmt[i + 1]) && | |
953 | !ispunct(fmt[i + 1])) { | |
954 | err = -EINVAL; | |
d9c9e4db FR |
955 | goto out; |
956 | } | |
957 | ||
48cac3f4 FR |
958 | if (!tmp_buf) |
959 | goto nocopy_fmt; | |
960 | ||
961 | if (tmp_buf_end == tmp_buf) { | |
d9c9e4db | 962 | err = -ENOSPC; |
48cac3f4 | 963 | goto out; |
d9c9e4db FR |
964 | } |
965 | ||
966 | unsafe_ptr = (char *)(long)raw_args[num_spec]; | |
967 | err = bpf_trace_copy_string(tmp_buf, unsafe_ptr, | |
48cac3f4 FR |
968 | fmt_ptype, |
969 | tmp_buf_end - tmp_buf); | |
d9c9e4db FR |
970 | if (err < 0) { |
971 | tmp_buf[0] = '\0'; | |
972 | err = 1; | |
973 | } | |
974 | ||
d9c9e4db | 975 | tmp_buf += err; |
48cac3f4 | 976 | num_spec++; |
d9c9e4db | 977 | |
3478cfcf KI |
978 | continue; |
979 | } else if (fmt[i] == 'c') { | |
980 | if (!tmp_buf) | |
981 | goto nocopy_fmt; | |
982 | ||
983 | if (tmp_buf_end == tmp_buf) { | |
984 | err = -ENOSPC; | |
985 | goto out; | |
986 | } | |
987 | ||
988 | *tmp_buf = raw_args[num_spec]; | |
989 | tmp_buf++; | |
990 | num_spec++; | |
991 | ||
48cac3f4 | 992 | continue; |
d9c9e4db FR |
993 | } |
994 | ||
48cac3f4 | 995 | sizeof_cur_arg = sizeof(int); |
d9c9e4db FR |
996 | |
997 | if (fmt[i] == 'l') { | |
48cac3f4 | 998 | sizeof_cur_arg = sizeof(long); |
d9c9e4db FR |
999 | i++; |
1000 | } | |
1001 | if (fmt[i] == 'l') { | |
48cac3f4 | 1002 | sizeof_cur_arg = sizeof(long long); |
d9c9e4db FR |
1003 | i++; |
1004 | } | |
1005 | ||
1006 | if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' && | |
1007 | fmt[i] != 'x' && fmt[i] != 'X') { | |
1008 | err = -EINVAL; | |
48cac3f4 | 1009 | goto out; |
d9c9e4db FR |
1010 | } |
1011 | ||
48cac3f4 | 1012 | if (tmp_buf) |
d9c9e4db | 1013 | cur_arg = raw_args[num_spec]; |
48cac3f4 FR |
1014 | nocopy_fmt: |
1015 | if (tmp_buf) { | |
1016 | tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32)); | |
1017 | if (tmp_buf_end - tmp_buf < sizeof_cur_arg) { | |
1018 | err = -ENOSPC; | |
1019 | goto out; | |
1020 | } | |
1021 | ||
1022 | if (sizeof_cur_arg == 8) { | |
1023 | *(u32 *)tmp_buf = *(u32 *)&cur_arg; | |
1024 | *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1); | |
1025 | } else { | |
1026 | *(u32 *)tmp_buf = (u32)(long)cur_arg; | |
1027 | } | |
1028 | tmp_buf += sizeof_cur_arg; | |
d9c9e4db FR |
1029 | } |
1030 | num_spec++; | |
1031 | } | |
1032 | ||
1033 | err = 0; | |
d9c9e4db | 1034 | out: |
48cac3f4 | 1035 | if (err) |
f19a4050 | 1036 | bpf_bprintf_cleanup(data); |
d9c9e4db FR |
1037 | return err; |
1038 | } | |
1039 | ||
7b15523a | 1040 | BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt, |
78aa1cc9 | 1041 | const void *, args, u32, data_len) |
7b15523a | 1042 | { |
78aa1cc9 JO |
1043 | struct bpf_bprintf_data data = { |
1044 | .get_bin_args = true, | |
1045 | }; | |
7b15523a FR |
1046 | int err, num_args; |
1047 | ||
335ff499 | 1048 | if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 || |
78aa1cc9 | 1049 | (data_len && !args)) |
7b15523a FR |
1050 | return -EINVAL; |
1051 | num_args = data_len / 8; | |
1052 | ||
1053 | /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we | |
1054 | * can safely give an unbounded size. | |
1055 | */ | |
78aa1cc9 | 1056 | err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data); |
7b15523a FR |
1057 | if (err < 0) |
1058 | return err; | |
1059 | ||
78aa1cc9 | 1060 | err = bstr_printf(str, str_size, fmt, data.bin_args); |
48cac3f4 | 1061 | |
f19a4050 | 1062 | bpf_bprintf_cleanup(&data); |
7b15523a FR |
1063 | |
1064 | return err + 1; | |
1065 | } | |
1066 | ||
1067 | const struct bpf_func_proto bpf_snprintf_proto = { | |
1068 | .func = bpf_snprintf, | |
1069 | .gpl_only = true, | |
1070 | .ret_type = RET_INTEGER, | |
1071 | .arg1_type = ARG_PTR_TO_MEM_OR_NULL, | |
1072 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
1073 | .arg3_type = ARG_PTR_TO_CONST_STR, | |
216e3cd2 | 1074 | .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, |
7b15523a FR |
1075 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
1076 | }; | |
1077 | ||
b00628b1 AS |
1078 | /* BPF map elements can contain 'struct bpf_timer'. |
1079 | * Such map owns all of its BPF timers. | |
1080 | * 'struct bpf_timer' is allocated as part of map element allocation | |
1081 | * and it's zero initialized. | |
1082 | * That space is used to keep 'struct bpf_timer_kern'. | |
1083 | * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and | |
1084 | * remembers 'struct bpf_map *' pointer it's part of. | |
1085 | * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn. | |
1086 | * bpf_timer_start() arms the timer. | |
1087 | * If user space reference to a map goes to zero at this point | |
1088 | * ops->map_release_uref callback is responsible for cancelling the timers, | |
1089 | * freeing their memory, and decrementing prog's refcnts. | |
1090 | * bpf_timer_cancel() cancels the timer and decrements prog's refcnt. | |
1091 | * Inner maps can contain bpf timers as well. ops->map_release_uref is | |
1092 | * freeing the timers when inner map is replaced or deleted by user space. | |
1093 | */ | |
1094 | struct bpf_hrtimer { | |
1095 | struct hrtimer timer; | |
1096 | struct bpf_map *map; | |
1097 | struct bpf_prog *prog; | |
1098 | void __rcu *callback_fn; | |
1099 | void *value; | |
1100 | }; | |
1101 | ||
1102 | /* the actual struct hidden inside uapi struct bpf_timer */ | |
1103 | struct bpf_timer_kern { | |
1104 | struct bpf_hrtimer *timer; | |
1105 | /* bpf_spin_lock is used here instead of spinlock_t to make | |
c561d110 | 1106 | * sure that it always fits into space reserved by struct bpf_timer |
b00628b1 AS |
1107 | * regardless of LOCKDEP and spinlock debug flags. |
1108 | */ | |
1109 | struct bpf_spin_lock lock; | |
1110 | } __attribute__((aligned(8))); | |
1111 | ||
1112 | static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running); | |
1113 | ||
1114 | static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer) | |
1115 | { | |
1116 | struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer); | |
1117 | struct bpf_map *map = t->map; | |
1118 | void *value = t->value; | |
102acbac | 1119 | bpf_callback_t callback_fn; |
b00628b1 AS |
1120 | void *key; |
1121 | u32 idx; | |
b00628b1 | 1122 | |
3bd916ee | 1123 | BTF_TYPE_EMIT(struct bpf_timer); |
b00628b1 AS |
1124 | callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held()); |
1125 | if (!callback_fn) | |
1126 | goto out; | |
1127 | ||
1128 | /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and | |
1129 | * cannot be preempted by another bpf_timer_cb() on the same cpu. | |
1130 | * Remember the timer this callback is servicing to prevent | |
1131 | * deadlock if callback_fn() calls bpf_timer_cancel() or | |
1132 | * bpf_map_delete_elem() on the same timer. | |
1133 | */ | |
1134 | this_cpu_write(hrtimer_running, t); | |
1135 | if (map->map_type == BPF_MAP_TYPE_ARRAY) { | |
1136 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
1137 | ||
1138 | /* compute the key */ | |
1139 | idx = ((char *)value - array->value) / array->elem_size; | |
1140 | key = &idx; | |
1141 | } else { /* hash or lru */ | |
1142 | key = value - round_up(map->key_size, 8); | |
1143 | } | |
1144 | ||
102acbac | 1145 | callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0); |
bfc6bb74 | 1146 | /* The verifier checked that return value is zero. */ |
b00628b1 AS |
1147 | |
1148 | this_cpu_write(hrtimer_running, NULL); | |
1149 | out: | |
1150 | return HRTIMER_NORESTART; | |
1151 | } | |
1152 | ||
1153 | BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map, | |
1154 | u64, flags) | |
1155 | { | |
1156 | clockid_t clockid = flags & (MAX_CLOCKS - 1); | |
1157 | struct bpf_hrtimer *t; | |
1158 | int ret = 0; | |
1159 | ||
1160 | BUILD_BUG_ON(MAX_CLOCKS != 16); | |
1161 | BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer)); | |
1162 | BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer)); | |
1163 | ||
1164 | if (in_nmi()) | |
1165 | return -EOPNOTSUPP; | |
1166 | ||
1167 | if (flags >= MAX_CLOCKS || | |
1168 | /* similar to timerfd except _ALARM variants are not supported */ | |
1169 | (clockid != CLOCK_MONOTONIC && | |
1170 | clockid != CLOCK_REALTIME && | |
1171 | clockid != CLOCK_BOOTTIME)) | |
1172 | return -EINVAL; | |
1173 | __bpf_spin_lock_irqsave(&timer->lock); | |
1174 | t = timer->timer; | |
1175 | if (t) { | |
1176 | ret = -EBUSY; | |
1177 | goto out; | |
1178 | } | |
1179 | if (!atomic64_read(&map->usercnt)) { | |
1180 | /* maps with timers must be either held by user space | |
1181 | * or pinned in bpffs. | |
1182 | */ | |
1183 | ret = -EPERM; | |
1184 | goto out; | |
1185 | } | |
1186 | /* allocate hrtimer via map_kmalloc to use memcg accounting */ | |
1187 | t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node); | |
1188 | if (!t) { | |
1189 | ret = -ENOMEM; | |
1190 | goto out; | |
1191 | } | |
db559117 | 1192 | t->value = (void *)timer - map->record->timer_off; |
b00628b1 AS |
1193 | t->map = map; |
1194 | t->prog = NULL; | |
1195 | rcu_assign_pointer(t->callback_fn, NULL); | |
1196 | hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT); | |
1197 | t->timer.function = bpf_timer_cb; | |
1198 | timer->timer = t; | |
1199 | out: | |
1200 | __bpf_spin_unlock_irqrestore(&timer->lock); | |
1201 | return ret; | |
1202 | } | |
1203 | ||
1204 | static const struct bpf_func_proto bpf_timer_init_proto = { | |
1205 | .func = bpf_timer_init, | |
1206 | .gpl_only = true, | |
1207 | .ret_type = RET_INTEGER, | |
1208 | .arg1_type = ARG_PTR_TO_TIMER, | |
1209 | .arg2_type = ARG_CONST_MAP_PTR, | |
1210 | .arg3_type = ARG_ANYTHING, | |
1211 | }; | |
1212 | ||
1213 | BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn, | |
1214 | struct bpf_prog_aux *, aux) | |
1215 | { | |
1216 | struct bpf_prog *prev, *prog = aux->prog; | |
1217 | struct bpf_hrtimer *t; | |
1218 | int ret = 0; | |
1219 | ||
1220 | if (in_nmi()) | |
1221 | return -EOPNOTSUPP; | |
1222 | __bpf_spin_lock_irqsave(&timer->lock); | |
1223 | t = timer->timer; | |
1224 | if (!t) { | |
1225 | ret = -EINVAL; | |
1226 | goto out; | |
1227 | } | |
1228 | if (!atomic64_read(&t->map->usercnt)) { | |
1229 | /* maps with timers must be either held by user space | |
1230 | * or pinned in bpffs. Otherwise timer might still be | |
1231 | * running even when bpf prog is detached and user space | |
1232 | * is gone, since map_release_uref won't ever be called. | |
1233 | */ | |
1234 | ret = -EPERM; | |
1235 | goto out; | |
1236 | } | |
1237 | prev = t->prog; | |
1238 | if (prev != prog) { | |
1239 | /* Bump prog refcnt once. Every bpf_timer_set_callback() | |
1240 | * can pick different callback_fn-s within the same prog. | |
1241 | */ | |
1242 | prog = bpf_prog_inc_not_zero(prog); | |
1243 | if (IS_ERR(prog)) { | |
1244 | ret = PTR_ERR(prog); | |
1245 | goto out; | |
1246 | } | |
1247 | if (prev) | |
1248 | /* Drop prev prog refcnt when swapping with new prog */ | |
1249 | bpf_prog_put(prev); | |
1250 | t->prog = prog; | |
1251 | } | |
1252 | rcu_assign_pointer(t->callback_fn, callback_fn); | |
1253 | out: | |
1254 | __bpf_spin_unlock_irqrestore(&timer->lock); | |
1255 | return ret; | |
1256 | } | |
1257 | ||
1258 | static const struct bpf_func_proto bpf_timer_set_callback_proto = { | |
1259 | .func = bpf_timer_set_callback, | |
1260 | .gpl_only = true, | |
1261 | .ret_type = RET_INTEGER, | |
1262 | .arg1_type = ARG_PTR_TO_TIMER, | |
1263 | .arg2_type = ARG_PTR_TO_FUNC, | |
1264 | }; | |
1265 | ||
1266 | BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags) | |
1267 | { | |
1268 | struct bpf_hrtimer *t; | |
1269 | int ret = 0; | |
f71f8530 | 1270 | enum hrtimer_mode mode; |
b00628b1 AS |
1271 | |
1272 | if (in_nmi()) | |
1273 | return -EOPNOTSUPP; | |
f71f8530 | 1274 | if (flags > BPF_F_TIMER_ABS) |
b00628b1 AS |
1275 | return -EINVAL; |
1276 | __bpf_spin_lock_irqsave(&timer->lock); | |
1277 | t = timer->timer; | |
1278 | if (!t || !t->prog) { | |
1279 | ret = -EINVAL; | |
1280 | goto out; | |
1281 | } | |
f71f8530 TK |
1282 | |
1283 | if (flags & BPF_F_TIMER_ABS) | |
1284 | mode = HRTIMER_MODE_ABS_SOFT; | |
1285 | else | |
1286 | mode = HRTIMER_MODE_REL_SOFT; | |
1287 | ||
1288 | hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode); | |
b00628b1 AS |
1289 | out: |
1290 | __bpf_spin_unlock_irqrestore(&timer->lock); | |
1291 | return ret; | |
1292 | } | |
1293 | ||
1294 | static const struct bpf_func_proto bpf_timer_start_proto = { | |
1295 | .func = bpf_timer_start, | |
1296 | .gpl_only = true, | |
1297 | .ret_type = RET_INTEGER, | |
1298 | .arg1_type = ARG_PTR_TO_TIMER, | |
1299 | .arg2_type = ARG_ANYTHING, | |
1300 | .arg3_type = ARG_ANYTHING, | |
1301 | }; | |
1302 | ||
1303 | static void drop_prog_refcnt(struct bpf_hrtimer *t) | |
1304 | { | |
1305 | struct bpf_prog *prog = t->prog; | |
1306 | ||
1307 | if (prog) { | |
1308 | bpf_prog_put(prog); | |
1309 | t->prog = NULL; | |
1310 | rcu_assign_pointer(t->callback_fn, NULL); | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer) | |
1315 | { | |
1316 | struct bpf_hrtimer *t; | |
1317 | int ret = 0; | |
1318 | ||
1319 | if (in_nmi()) | |
1320 | return -EOPNOTSUPP; | |
1321 | __bpf_spin_lock_irqsave(&timer->lock); | |
1322 | t = timer->timer; | |
1323 | if (!t) { | |
1324 | ret = -EINVAL; | |
1325 | goto out; | |
1326 | } | |
1327 | if (this_cpu_read(hrtimer_running) == t) { | |
1328 | /* If bpf callback_fn is trying to bpf_timer_cancel() | |
1329 | * its own timer the hrtimer_cancel() will deadlock | |
1330 | * since it waits for callback_fn to finish | |
1331 | */ | |
1332 | ret = -EDEADLK; | |
1333 | goto out; | |
1334 | } | |
1335 | drop_prog_refcnt(t); | |
1336 | out: | |
1337 | __bpf_spin_unlock_irqrestore(&timer->lock); | |
1338 | /* Cancel the timer and wait for associated callback to finish | |
1339 | * if it was running. | |
1340 | */ | |
1341 | ret = ret ?: hrtimer_cancel(&t->timer); | |
1342 | return ret; | |
1343 | } | |
1344 | ||
1345 | static const struct bpf_func_proto bpf_timer_cancel_proto = { | |
1346 | .func = bpf_timer_cancel, | |
1347 | .gpl_only = true, | |
1348 | .ret_type = RET_INTEGER, | |
1349 | .arg1_type = ARG_PTR_TO_TIMER, | |
1350 | }; | |
1351 | ||
1352 | /* This function is called by map_delete/update_elem for individual element and | |
1353 | * by ops->map_release_uref when the user space reference to a map reaches zero. | |
1354 | */ | |
1355 | void bpf_timer_cancel_and_free(void *val) | |
1356 | { | |
1357 | struct bpf_timer_kern *timer = val; | |
1358 | struct bpf_hrtimer *t; | |
1359 | ||
1360 | /* Performance optimization: read timer->timer without lock first. */ | |
1361 | if (!READ_ONCE(timer->timer)) | |
1362 | return; | |
1363 | ||
1364 | __bpf_spin_lock_irqsave(&timer->lock); | |
1365 | /* re-read it under lock */ | |
1366 | t = timer->timer; | |
1367 | if (!t) | |
1368 | goto out; | |
1369 | drop_prog_refcnt(t); | |
1370 | /* The subsequent bpf_timer_start/cancel() helpers won't be able to use | |
1371 | * this timer, since it won't be initialized. | |
1372 | */ | |
1373 | timer->timer = NULL; | |
1374 | out: | |
1375 | __bpf_spin_unlock_irqrestore(&timer->lock); | |
1376 | if (!t) | |
1377 | return; | |
1378 | /* Cancel the timer and wait for callback to complete if it was running. | |
1379 | * If hrtimer_cancel() can be safely called it's safe to call kfree(t) | |
1380 | * right after for both preallocated and non-preallocated maps. | |
1381 | * The timer->timer = NULL was already done and no code path can | |
1382 | * see address 't' anymore. | |
1383 | * | |
1384 | * Check that bpf_map_delete/update_elem() wasn't called from timer | |
1385 | * callback_fn. In such case don't call hrtimer_cancel() (since it will | |
1386 | * deadlock) and don't call hrtimer_try_to_cancel() (since it will just | |
1387 | * return -1). Though callback_fn is still running on this cpu it's | |
1388 | * safe to do kfree(t) because bpf_timer_cb() read everything it needed | |
1389 | * from 't'. The bpf subprog callback_fn won't be able to access 't', | |
1390 | * since timer->timer = NULL was already done. The timer will be | |
1391 | * effectively cancelled because bpf_timer_cb() will return | |
1392 | * HRTIMER_NORESTART. | |
1393 | */ | |
1394 | if (this_cpu_read(hrtimer_running) != t) | |
1395 | hrtimer_cancel(&t->timer); | |
1396 | kfree(t); | |
1397 | } | |
1398 | ||
c0a5a21c KKD |
1399 | BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr) |
1400 | { | |
1401 | unsigned long *kptr = map_value; | |
1402 | ||
1403 | return xchg(kptr, (unsigned long)ptr); | |
1404 | } | |
1405 | ||
1406 | /* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg() | |
47e34cb7 DM |
1407 | * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to |
1408 | * denote type that verifier will determine. | |
c0a5a21c | 1409 | */ |
dc368e1c | 1410 | static const struct bpf_func_proto bpf_kptr_xchg_proto = { |
c0a5a21c KKD |
1411 | .func = bpf_kptr_xchg, |
1412 | .gpl_only = false, | |
1413 | .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, | |
1414 | .ret_btf_id = BPF_PTR_POISON, | |
1415 | .arg1_type = ARG_PTR_TO_KPTR, | |
1416 | .arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE, | |
1417 | .arg2_btf_id = BPF_PTR_POISON, | |
1418 | }; | |
1419 | ||
263ae152 JK |
1420 | /* Since the upper 8 bits of dynptr->size is reserved, the |
1421 | * maximum supported size is 2^24 - 1. | |
1422 | */ | |
1423 | #define DYNPTR_MAX_SIZE ((1UL << 24) - 1) | |
1424 | #define DYNPTR_TYPE_SHIFT 28 | |
13bbbfbe JK |
1425 | #define DYNPTR_SIZE_MASK 0xFFFFFF |
1426 | #define DYNPTR_RDONLY_BIT BIT(31) | |
1427 | ||
540ccf96 | 1428 | static bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr) |
13bbbfbe JK |
1429 | { |
1430 | return ptr->size & DYNPTR_RDONLY_BIT; | |
1431 | } | |
263ae152 | 1432 | |
b5964b96 JK |
1433 | void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr) |
1434 | { | |
1435 | ptr->size |= DYNPTR_RDONLY_BIT; | |
1436 | } | |
1437 | ||
263ae152 JK |
1438 | static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type) |
1439 | { | |
1440 | ptr->size |= type << DYNPTR_TYPE_SHIFT; | |
1441 | } | |
1442 | ||
b5964b96 JK |
1443 | static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr) |
1444 | { | |
1445 | return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT; | |
1446 | } | |
1447 | ||
26662d73 | 1448 | u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr) |
13bbbfbe JK |
1449 | { |
1450 | return ptr->size & DYNPTR_SIZE_MASK; | |
1451 | } | |
1452 | ||
987d0242 JK |
1453 | static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size) |
1454 | { | |
1455 | u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK; | |
1456 | ||
1457 | ptr->size = new_size | metadata; | |
1458 | } | |
1459 | ||
bc34dee6 | 1460 | int bpf_dynptr_check_size(u32 size) |
263ae152 JK |
1461 | { |
1462 | return size > DYNPTR_MAX_SIZE ? -E2BIG : 0; | |
1463 | } | |
1464 | ||
bc34dee6 JK |
1465 | void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data, |
1466 | enum bpf_dynptr_type type, u32 offset, u32 size) | |
263ae152 JK |
1467 | { |
1468 | ptr->data = data; | |
1469 | ptr->offset = offset; | |
1470 | ptr->size = size; | |
1471 | bpf_dynptr_set_type(ptr, type); | |
1472 | } | |
1473 | ||
bc34dee6 | 1474 | void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr) |
263ae152 JK |
1475 | { |
1476 | memset(ptr, 0, sizeof(*ptr)); | |
1477 | } | |
1478 | ||
27060531 | 1479 | static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len) |
13bbbfbe | 1480 | { |
26662d73 | 1481 | u32 size = __bpf_dynptr_size(ptr); |
13bbbfbe JK |
1482 | |
1483 | if (len > size || offset > size - len) | |
1484 | return -E2BIG; | |
1485 | ||
1486 | return 0; | |
1487 | } | |
1488 | ||
263ae152 JK |
1489 | BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr) |
1490 | { | |
1491 | int err; | |
1492 | ||
00f14641 RS |
1493 | BTF_TYPE_EMIT(struct bpf_dynptr); |
1494 | ||
263ae152 JK |
1495 | err = bpf_dynptr_check_size(size); |
1496 | if (err) | |
1497 | goto error; | |
1498 | ||
1499 | /* flags is currently unsupported */ | |
1500 | if (flags) { | |
1501 | err = -EINVAL; | |
1502 | goto error; | |
1503 | } | |
1504 | ||
1505 | bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size); | |
1506 | ||
1507 | return 0; | |
1508 | ||
1509 | error: | |
1510 | bpf_dynptr_set_null(ptr); | |
1511 | return err; | |
1512 | } | |
1513 | ||
dc368e1c | 1514 | static const struct bpf_func_proto bpf_dynptr_from_mem_proto = { |
263ae152 JK |
1515 | .func = bpf_dynptr_from_mem, |
1516 | .gpl_only = false, | |
1517 | .ret_type = RET_INTEGER, | |
1518 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
1519 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
1520 | .arg3_type = ARG_ANYTHING, | |
1521 | .arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT, | |
1522 | }; | |
1523 | ||
27060531 | 1524 | BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src, |
f8d3da4e | 1525 | u32, offset, u64, flags) |
13bbbfbe | 1526 | { |
b5964b96 | 1527 | enum bpf_dynptr_type type; |
13bbbfbe JK |
1528 | int err; |
1529 | ||
f8d3da4e | 1530 | if (!src->data || flags) |
13bbbfbe JK |
1531 | return -EINVAL; |
1532 | ||
1533 | err = bpf_dynptr_check_off_len(src, offset, len); | |
1534 | if (err) | |
1535 | return err; | |
1536 | ||
b5964b96 | 1537 | type = bpf_dynptr_get_type(src); |
13bbbfbe | 1538 | |
b5964b96 JK |
1539 | switch (type) { |
1540 | case BPF_DYNPTR_TYPE_LOCAL: | |
1541 | case BPF_DYNPTR_TYPE_RINGBUF: | |
1542 | /* Source and destination may possibly overlap, hence use memmove to | |
1543 | * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr | |
1544 | * pointing to overlapping PTR_TO_MAP_VALUE regions. | |
1545 | */ | |
1546 | memmove(dst, src->data + src->offset + offset, len); | |
1547 | return 0; | |
1548 | case BPF_DYNPTR_TYPE_SKB: | |
1549 | return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len); | |
05421aec JK |
1550 | case BPF_DYNPTR_TYPE_XDP: |
1551 | return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len); | |
b5964b96 JK |
1552 | default: |
1553 | WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type); | |
1554 | return -EFAULT; | |
1555 | } | |
13bbbfbe JK |
1556 | } |
1557 | ||
dc368e1c | 1558 | static const struct bpf_func_proto bpf_dynptr_read_proto = { |
13bbbfbe JK |
1559 | .func = bpf_dynptr_read, |
1560 | .gpl_only = false, | |
1561 | .ret_type = RET_INTEGER, | |
1562 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
1563 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
27060531 | 1564 | .arg3_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY, |
13bbbfbe | 1565 | .arg4_type = ARG_ANYTHING, |
f8d3da4e | 1566 | .arg5_type = ARG_ANYTHING, |
13bbbfbe JK |
1567 | }; |
1568 | ||
27060531 | 1569 | BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src, |
f8d3da4e | 1570 | u32, len, u64, flags) |
13bbbfbe | 1571 | { |
b5964b96 | 1572 | enum bpf_dynptr_type type; |
13bbbfbe JK |
1573 | int err; |
1574 | ||
540ccf96 | 1575 | if (!dst->data || __bpf_dynptr_is_rdonly(dst)) |
13bbbfbe JK |
1576 | return -EINVAL; |
1577 | ||
1578 | err = bpf_dynptr_check_off_len(dst, offset, len); | |
1579 | if (err) | |
1580 | return err; | |
1581 | ||
b5964b96 | 1582 | type = bpf_dynptr_get_type(dst); |
13bbbfbe | 1583 | |
b5964b96 JK |
1584 | switch (type) { |
1585 | case BPF_DYNPTR_TYPE_LOCAL: | |
1586 | case BPF_DYNPTR_TYPE_RINGBUF: | |
1587 | if (flags) | |
1588 | return -EINVAL; | |
1589 | /* Source and destination may possibly overlap, hence use memmove to | |
1590 | * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr | |
1591 | * pointing to overlapping PTR_TO_MAP_VALUE regions. | |
1592 | */ | |
1593 | memmove(dst->data + dst->offset + offset, src, len); | |
1594 | return 0; | |
1595 | case BPF_DYNPTR_TYPE_SKB: | |
1596 | return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len, | |
1597 | flags); | |
05421aec JK |
1598 | case BPF_DYNPTR_TYPE_XDP: |
1599 | if (flags) | |
1600 | return -EINVAL; | |
1601 | return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len); | |
b5964b96 JK |
1602 | default: |
1603 | WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type); | |
1604 | return -EFAULT; | |
1605 | } | |
13bbbfbe JK |
1606 | } |
1607 | ||
dc368e1c | 1608 | static const struct bpf_func_proto bpf_dynptr_write_proto = { |
13bbbfbe JK |
1609 | .func = bpf_dynptr_write, |
1610 | .gpl_only = false, | |
1611 | .ret_type = RET_INTEGER, | |
27060531 | 1612 | .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY, |
13bbbfbe JK |
1613 | .arg2_type = ARG_ANYTHING, |
1614 | .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, | |
1615 | .arg4_type = ARG_CONST_SIZE_OR_ZERO, | |
f8d3da4e | 1616 | .arg5_type = ARG_ANYTHING, |
13bbbfbe JK |
1617 | }; |
1618 | ||
27060531 | 1619 | BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len) |
34d4ef57 | 1620 | { |
b5964b96 | 1621 | enum bpf_dynptr_type type; |
34d4ef57 JK |
1622 | int err; |
1623 | ||
1624 | if (!ptr->data) | |
1625 | return 0; | |
1626 | ||
1627 | err = bpf_dynptr_check_off_len(ptr, offset, len); | |
1628 | if (err) | |
1629 | return 0; | |
1630 | ||
540ccf96 | 1631 | if (__bpf_dynptr_is_rdonly(ptr)) |
34d4ef57 JK |
1632 | return 0; |
1633 | ||
b5964b96 JK |
1634 | type = bpf_dynptr_get_type(ptr); |
1635 | ||
1636 | switch (type) { | |
1637 | case BPF_DYNPTR_TYPE_LOCAL: | |
1638 | case BPF_DYNPTR_TYPE_RINGBUF: | |
1639 | return (unsigned long)(ptr->data + ptr->offset + offset); | |
1640 | case BPF_DYNPTR_TYPE_SKB: | |
05421aec JK |
1641 | case BPF_DYNPTR_TYPE_XDP: |
1642 | /* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */ | |
b5964b96 JK |
1643 | return 0; |
1644 | default: | |
1645 | WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type); | |
1646 | return 0; | |
1647 | } | |
34d4ef57 JK |
1648 | } |
1649 | ||
dc368e1c | 1650 | static const struct bpf_func_proto bpf_dynptr_data_proto = { |
34d4ef57 JK |
1651 | .func = bpf_dynptr_data, |
1652 | .gpl_only = false, | |
1653 | .ret_type = RET_PTR_TO_DYNPTR_MEM_OR_NULL, | |
27060531 | 1654 | .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY, |
34d4ef57 JK |
1655 | .arg2_type = ARG_ANYTHING, |
1656 | .arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO, | |
1657 | }; | |
1658 | ||
f470378c | 1659 | const struct bpf_func_proto bpf_get_current_task_proto __weak; |
a396eda5 | 1660 | const struct bpf_func_proto bpf_get_current_task_btf_proto __weak; |
f470378c JF |
1661 | const struct bpf_func_proto bpf_probe_read_user_proto __weak; |
1662 | const struct bpf_func_proto bpf_probe_read_user_str_proto __weak; | |
1663 | const struct bpf_func_proto bpf_probe_read_kernel_proto __weak; | |
1664 | const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak; | |
dd6e10fb | 1665 | const struct bpf_func_proto bpf_task_pt_regs_proto __weak; |
f470378c | 1666 | |
6890896b SF |
1667 | const struct bpf_func_proto * |
1668 | bpf_base_func_proto(enum bpf_func_id func_id) | |
1669 | { | |
1670 | switch (func_id) { | |
1671 | case BPF_FUNC_map_lookup_elem: | |
1672 | return &bpf_map_lookup_elem_proto; | |
1673 | case BPF_FUNC_map_update_elem: | |
1674 | return &bpf_map_update_elem_proto; | |
1675 | case BPF_FUNC_map_delete_elem: | |
1676 | return &bpf_map_delete_elem_proto; | |
1677 | case BPF_FUNC_map_push_elem: | |
1678 | return &bpf_map_push_elem_proto; | |
1679 | case BPF_FUNC_map_pop_elem: | |
1680 | return &bpf_map_pop_elem_proto; | |
1681 | case BPF_FUNC_map_peek_elem: | |
1682 | return &bpf_map_peek_elem_proto; | |
07343110 FZ |
1683 | case BPF_FUNC_map_lookup_percpu_elem: |
1684 | return &bpf_map_lookup_percpu_elem_proto; | |
6890896b SF |
1685 | case BPF_FUNC_get_prandom_u32: |
1686 | return &bpf_get_prandom_u32_proto; | |
1687 | case BPF_FUNC_get_smp_processor_id: | |
1688 | return &bpf_get_raw_smp_processor_id_proto; | |
1689 | case BPF_FUNC_get_numa_node_id: | |
1690 | return &bpf_get_numa_node_id_proto; | |
1691 | case BPF_FUNC_tail_call: | |
1692 | return &bpf_tail_call_proto; | |
1693 | case BPF_FUNC_ktime_get_ns: | |
1694 | return &bpf_ktime_get_ns_proto; | |
71d19214 MŻ |
1695 | case BPF_FUNC_ktime_get_boot_ns: |
1696 | return &bpf_ktime_get_boot_ns_proto; | |
c8996c98 JDB |
1697 | case BPF_FUNC_ktime_get_tai_ns: |
1698 | return &bpf_ktime_get_tai_ns_proto; | |
457f4436 AN |
1699 | case BPF_FUNC_ringbuf_output: |
1700 | return &bpf_ringbuf_output_proto; | |
1701 | case BPF_FUNC_ringbuf_reserve: | |
1702 | return &bpf_ringbuf_reserve_proto; | |
1703 | case BPF_FUNC_ringbuf_submit: | |
1704 | return &bpf_ringbuf_submit_proto; | |
1705 | case BPF_FUNC_ringbuf_discard: | |
1706 | return &bpf_ringbuf_discard_proto; | |
1707 | case BPF_FUNC_ringbuf_query: | |
1708 | return &bpf_ringbuf_query_proto; | |
c5fb1993 HT |
1709 | case BPF_FUNC_strncmp: |
1710 | return &bpf_strncmp_proto; | |
8a67f2de SF |
1711 | case BPF_FUNC_strtol: |
1712 | return &bpf_strtol_proto; | |
1713 | case BPF_FUNC_strtoul: | |
1714 | return &bpf_strtoul_proto; | |
6890896b SF |
1715 | default: |
1716 | break; | |
1717 | } | |
1718 | ||
2c78ee89 | 1719 | if (!bpf_capable()) |
6890896b SF |
1720 | return NULL; |
1721 | ||
1722 | switch (func_id) { | |
1723 | case BPF_FUNC_spin_lock: | |
1724 | return &bpf_spin_lock_proto; | |
1725 | case BPF_FUNC_spin_unlock: | |
1726 | return &bpf_spin_unlock_proto; | |
6890896b SF |
1727 | case BPF_FUNC_jiffies64: |
1728 | return &bpf_jiffies64_proto; | |
b7906b70 | 1729 | case BPF_FUNC_per_cpu_ptr: |
eaa6bcb7 | 1730 | return &bpf_per_cpu_ptr_proto; |
b7906b70 | 1731 | case BPF_FUNC_this_cpu_ptr: |
63d9b80d | 1732 | return &bpf_this_cpu_ptr_proto; |
b00628b1 AS |
1733 | case BPF_FUNC_timer_init: |
1734 | return &bpf_timer_init_proto; | |
1735 | case BPF_FUNC_timer_set_callback: | |
1736 | return &bpf_timer_set_callback_proto; | |
1737 | case BPF_FUNC_timer_start: | |
1738 | return &bpf_timer_start_proto; | |
1739 | case BPF_FUNC_timer_cancel: | |
1740 | return &bpf_timer_cancel_proto; | |
c0a5a21c KKD |
1741 | case BPF_FUNC_kptr_xchg: |
1742 | return &bpf_kptr_xchg_proto; | |
5679ff2f KKD |
1743 | case BPF_FUNC_for_each_map_elem: |
1744 | return &bpf_for_each_map_elem_proto; | |
1745 | case BPF_FUNC_loop: | |
1746 | return &bpf_loop_proto; | |
20571567 DV |
1747 | case BPF_FUNC_user_ringbuf_drain: |
1748 | return &bpf_user_ringbuf_drain_proto; | |
8addbfc7 KKD |
1749 | case BPF_FUNC_ringbuf_reserve_dynptr: |
1750 | return &bpf_ringbuf_reserve_dynptr_proto; | |
1751 | case BPF_FUNC_ringbuf_submit_dynptr: | |
1752 | return &bpf_ringbuf_submit_dynptr_proto; | |
1753 | case BPF_FUNC_ringbuf_discard_dynptr: | |
1754 | return &bpf_ringbuf_discard_dynptr_proto; | |
1755 | case BPF_FUNC_dynptr_from_mem: | |
1756 | return &bpf_dynptr_from_mem_proto; | |
1757 | case BPF_FUNC_dynptr_read: | |
1758 | return &bpf_dynptr_read_proto; | |
1759 | case BPF_FUNC_dynptr_write: | |
1760 | return &bpf_dynptr_write_proto; | |
1761 | case BPF_FUNC_dynptr_data: | |
1762 | return &bpf_dynptr_data_proto; | |
c4bcfb38 YS |
1763 | #ifdef CONFIG_CGROUPS |
1764 | case BPF_FUNC_cgrp_storage_get: | |
1765 | return &bpf_cgrp_storage_get_proto; | |
1766 | case BPF_FUNC_cgrp_storage_delete: | |
1767 | return &bpf_cgrp_storage_delete_proto; | |
c501bf55 TH |
1768 | case BPF_FUNC_get_current_cgroup_id: |
1769 | return &bpf_get_current_cgroup_id_proto; | |
1770 | case BPF_FUNC_get_current_ancestor_cgroup_id: | |
1771 | return &bpf_get_current_ancestor_cgroup_id_proto; | |
c4bcfb38 | 1772 | #endif |
f470378c JF |
1773 | default: |
1774 | break; | |
1775 | } | |
1776 | ||
1777 | if (!perfmon_capable()) | |
1778 | return NULL; | |
1779 | ||
1780 | switch (func_id) { | |
61ca36c8 TK |
1781 | case BPF_FUNC_trace_printk: |
1782 | return bpf_get_trace_printk_proto(); | |
f470378c JF |
1783 | case BPF_FUNC_get_current_task: |
1784 | return &bpf_get_current_task_proto; | |
a396eda5 DX |
1785 | case BPF_FUNC_get_current_task_btf: |
1786 | return &bpf_get_current_task_btf_proto; | |
f470378c JF |
1787 | case BPF_FUNC_probe_read_user: |
1788 | return &bpf_probe_read_user_proto; | |
1789 | case BPF_FUNC_probe_read_kernel: | |
71330842 | 1790 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1791 | NULL : &bpf_probe_read_kernel_proto; |
f470378c JF |
1792 | case BPF_FUNC_probe_read_user_str: |
1793 | return &bpf_probe_read_user_str_proto; | |
1794 | case BPF_FUNC_probe_read_kernel_str: | |
71330842 | 1795 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1796 | NULL : &bpf_probe_read_kernel_str_proto; |
61ca36c8 TK |
1797 | case BPF_FUNC_snprintf_btf: |
1798 | return &bpf_snprintf_btf_proto; | |
7b15523a FR |
1799 | case BPF_FUNC_snprintf: |
1800 | return &bpf_snprintf_proto; | |
dd6e10fb DX |
1801 | case BPF_FUNC_task_pt_regs: |
1802 | return &bpf_task_pt_regs_proto; | |
10aceb62 DM |
1803 | case BPF_FUNC_trace_vprintk: |
1804 | return bpf_get_trace_vprintk_proto(); | |
6890896b SF |
1805 | default: |
1806 | return NULL; | |
1807 | } | |
1808 | } | |
13379059 | 1809 | |
1512217c DM |
1810 | void __bpf_obj_drop_impl(void *p, const struct btf_record *rec); |
1811 | ||
f0c5941f KKD |
1812 | void bpf_list_head_free(const struct btf_field *field, void *list_head, |
1813 | struct bpf_spin_lock *spin_lock) | |
1814 | { | |
1815 | struct list_head *head = list_head, *orig_head = list_head; | |
1816 | ||
1817 | BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head)); | |
1818 | BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head)); | |
1819 | ||
1820 | /* Do the actual list draining outside the lock to not hold the lock for | |
1821 | * too long, and also prevent deadlocks if tracing programs end up | |
1822 | * executing on entry/exit of functions called inside the critical | |
1823 | * section, and end up doing map ops that call bpf_list_head_free for | |
1824 | * the same map value again. | |
1825 | */ | |
1826 | __bpf_spin_lock_irqsave(spin_lock); | |
1827 | if (!head->next || list_empty(head)) | |
1828 | goto unlock; | |
1829 | head = head->next; | |
1830 | unlock: | |
1831 | INIT_LIST_HEAD(orig_head); | |
1832 | __bpf_spin_unlock_irqrestore(spin_lock); | |
1833 | ||
1834 | while (head != orig_head) { | |
1835 | void *obj = head; | |
1836 | ||
30465003 | 1837 | obj -= field->graph_root.node_offset; |
f0c5941f | 1838 | head = head->next; |
958cf2e2 KKD |
1839 | /* The contained type can also have resources, including a |
1840 | * bpf_list_head which needs to be freed. | |
1841 | */ | |
958cf2e2 | 1842 | migrate_disable(); |
1512217c | 1843 | __bpf_obj_drop_impl(obj, field->graph_root.value_rec); |
958cf2e2 | 1844 | migrate_enable(); |
f0c5941f KKD |
1845 | } |
1846 | } | |
1847 | ||
9c395c1b DM |
1848 | /* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are |
1849 | * 'rb_node *', so field name of rb_node within containing struct is not | |
1850 | * needed. | |
1851 | * | |
1852 | * Since bpf_rb_tree's node type has a corresponding struct btf_field with | |
1853 | * graph_root.node_offset, it's not necessary to know field name | |
1854 | * or type of node struct | |
1855 | */ | |
1856 | #define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \ | |
1857 | for (pos = rb_first_postorder(root); \ | |
1858 | pos && ({ n = rb_next_postorder(pos); 1; }); \ | |
1859 | pos = n) | |
1860 | ||
1861 | void bpf_rb_root_free(const struct btf_field *field, void *rb_root, | |
1862 | struct bpf_spin_lock *spin_lock) | |
1863 | { | |
1864 | struct rb_root_cached orig_root, *root = rb_root; | |
1865 | struct rb_node *pos, *n; | |
1866 | void *obj; | |
1867 | ||
1868 | BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root)); | |
1869 | BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root)); | |
1870 | ||
1871 | __bpf_spin_lock_irqsave(spin_lock); | |
1872 | orig_root = *root; | |
1873 | *root = RB_ROOT_CACHED; | |
1874 | __bpf_spin_unlock_irqrestore(spin_lock); | |
1875 | ||
1876 | bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) { | |
1877 | obj = pos; | |
1878 | obj -= field->graph_root.node_offset; | |
1879 | ||
9c395c1b DM |
1880 | |
1881 | migrate_disable(); | |
1512217c | 1882 | __bpf_obj_drop_impl(obj, field->graph_root.value_rec); |
9c395c1b DM |
1883 | migrate_enable(); |
1884 | } | |
1885 | } | |
1886 | ||
958cf2e2 KKD |
1887 | __diag_push(); |
1888 | __diag_ignore_all("-Wmissing-prototypes", | |
1889 | "Global functions as their definitions will be in vmlinux BTF"); | |
1890 | ||
400031e0 | 1891 | __bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign) |
958cf2e2 KKD |
1892 | { |
1893 | struct btf_struct_meta *meta = meta__ign; | |
1894 | u64 size = local_type_id__k; | |
1895 | void *p; | |
1896 | ||
958cf2e2 KKD |
1897 | p = bpf_mem_alloc(&bpf_global_ma, size); |
1898 | if (!p) | |
1899 | return NULL; | |
1900 | if (meta) | |
cd2a8079 | 1901 | bpf_obj_init(meta->record, p); |
958cf2e2 KKD |
1902 | return p; |
1903 | } | |
1904 | ||
1512217c | 1905 | /* Must be called under migrate_disable(), as required by bpf_mem_free */ |
c8e18754 DM |
1906 | void __bpf_obj_drop_impl(void *p, const struct btf_record *rec) |
1907 | { | |
1512217c DM |
1908 | if (rec && rec->refcount_off >= 0 && |
1909 | !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) { | |
1910 | /* Object is refcounted and refcount_dec didn't result in 0 | |
1911 | * refcount. Return without freeing the object | |
1912 | */ | |
1913 | return; | |
1914 | } | |
1915 | ||
c8e18754 DM |
1916 | if (rec) |
1917 | bpf_obj_free_fields(rec, p); | |
7e26cd12 DM |
1918 | |
1919 | if (rec && rec->refcount_off >= 0) | |
1920 | bpf_mem_free_rcu(&bpf_global_ma, p); | |
1921 | else | |
1922 | bpf_mem_free(&bpf_global_ma, p); | |
c8e18754 DM |
1923 | } |
1924 | ||
400031e0 | 1925 | __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign) |
ac9f0605 KKD |
1926 | { |
1927 | struct btf_struct_meta *meta = meta__ign; | |
1928 | void *p = p__alloc; | |
1929 | ||
c8e18754 | 1930 | __bpf_obj_drop_impl(p, meta ? meta->record : NULL); |
ac9f0605 KKD |
1931 | } |
1932 | ||
7c50b1cb DM |
1933 | __bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign) |
1934 | { | |
1935 | struct btf_struct_meta *meta = meta__ign; | |
1936 | struct bpf_refcount *ref; | |
1937 | ||
1938 | /* Could just cast directly to refcount_t *, but need some code using | |
1939 | * bpf_refcount type so that it is emitted in vmlinux BTF | |
1940 | */ | |
4ab07209 | 1941 | ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off); |
7793fc3b DM |
1942 | if (!refcount_inc_not_zero((refcount_t *)ref)) |
1943 | return NULL; | |
7c50b1cb | 1944 | |
7793fc3b DM |
1945 | /* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null |
1946 | * in verifier.c | |
1947 | */ | |
7c50b1cb DM |
1948 | return (void *)p__refcounted_kptr; |
1949 | } | |
1950 | ||
0a1f7bfe DM |
1951 | static int __bpf_list_add(struct bpf_list_node_kern *node, |
1952 | struct bpf_list_head *head, | |
d2dcc67d | 1953 | bool tail, struct btf_record *rec, u64 off) |
8cab76ec | 1954 | { |
0a1f7bfe | 1955 | struct list_head *n = &node->list_head, *h = (void *)head; |
8cab76ec | 1956 | |
3e81740a DM |
1957 | /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't |
1958 | * called on its fields, so init here | |
1959 | */ | |
8cab76ec KKD |
1960 | if (unlikely(!h->next)) |
1961 | INIT_LIST_HEAD(h); | |
c3c510ce DM |
1962 | |
1963 | /* node->owner != NULL implies !list_empty(n), no need to separately | |
1964 | * check the latter | |
1965 | */ | |
1966 | if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) { | |
d2dcc67d | 1967 | /* Only called from BPF prog, no need to migrate_disable */ |
cc0d76ca | 1968 | __bpf_obj_drop_impl((void *)n - off, rec); |
d2dcc67d DM |
1969 | return -EINVAL; |
1970 | } | |
1971 | ||
8cab76ec | 1972 | tail ? list_add_tail(n, h) : list_add(n, h); |
c3c510ce | 1973 | WRITE_ONCE(node->owner, head); |
d2dcc67d DM |
1974 | |
1975 | return 0; | |
8cab76ec KKD |
1976 | } |
1977 | ||
d2dcc67d DM |
1978 | __bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head, |
1979 | struct bpf_list_node *node, | |
1980 | void *meta__ign, u64 off) | |
8cab76ec | 1981 | { |
0a1f7bfe | 1982 | struct bpf_list_node_kern *n = (void *)node; |
d2dcc67d DM |
1983 | struct btf_struct_meta *meta = meta__ign; |
1984 | ||
0a1f7bfe | 1985 | return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off); |
8cab76ec KKD |
1986 | } |
1987 | ||
d2dcc67d DM |
1988 | __bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head, |
1989 | struct bpf_list_node *node, | |
1990 | void *meta__ign, u64 off) | |
8cab76ec | 1991 | { |
0a1f7bfe | 1992 | struct bpf_list_node_kern *n = (void *)node; |
d2dcc67d DM |
1993 | struct btf_struct_meta *meta = meta__ign; |
1994 | ||
0a1f7bfe | 1995 | return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off); |
8cab76ec KKD |
1996 | } |
1997 | ||
1998 | static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail) | |
1999 | { | |
2000 | struct list_head *n, *h = (void *)head; | |
c3c510ce | 2001 | struct bpf_list_node_kern *node; |
8cab76ec | 2002 | |
3e81740a DM |
2003 | /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't |
2004 | * called on its fields, so init here | |
2005 | */ | |
8cab76ec KKD |
2006 | if (unlikely(!h->next)) |
2007 | INIT_LIST_HEAD(h); | |
2008 | if (list_empty(h)) | |
2009 | return NULL; | |
c3c510ce | 2010 | |
8cab76ec | 2011 | n = tail ? h->prev : h->next; |
c3c510ce DM |
2012 | node = container_of(n, struct bpf_list_node_kern, list_head); |
2013 | if (WARN_ON_ONCE(READ_ONCE(node->owner) != head)) | |
2014 | return NULL; | |
2015 | ||
8cab76ec | 2016 | list_del_init(n); |
c3c510ce | 2017 | WRITE_ONCE(node->owner, NULL); |
8cab76ec KKD |
2018 | return (struct bpf_list_node *)n; |
2019 | } | |
2020 | ||
400031e0 | 2021 | __bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head) |
8cab76ec KKD |
2022 | { |
2023 | return __bpf_list_del(head, false); | |
2024 | } | |
2025 | ||
400031e0 | 2026 | __bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head) |
8cab76ec KKD |
2027 | { |
2028 | return __bpf_list_del(head, true); | |
2029 | } | |
2030 | ||
bd1279ae DM |
2031 | __bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root, |
2032 | struct bpf_rb_node *node) | |
2033 | { | |
c3c510ce | 2034 | struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node; |
bd1279ae | 2035 | struct rb_root_cached *r = (struct rb_root_cached *)root; |
c3c510ce | 2036 | struct rb_node *n = &node_internal->rb_node; |
bd1279ae | 2037 | |
c3c510ce DM |
2038 | /* node_internal->owner != root implies either RB_EMPTY_NODE(n) or |
2039 | * n is owned by some other tree. No need to check RB_EMPTY_NODE(n) | |
2040 | */ | |
2041 | if (READ_ONCE(node_internal->owner) != root) | |
404ad75a DM |
2042 | return NULL; |
2043 | ||
bd1279ae DM |
2044 | rb_erase_cached(n, r); |
2045 | RB_CLEAR_NODE(n); | |
c3c510ce | 2046 | WRITE_ONCE(node_internal->owner, NULL); |
bd1279ae DM |
2047 | return (struct bpf_rb_node *)n; |
2048 | } | |
2049 | ||
2050 | /* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF | |
2051 | * program | |
2052 | */ | |
0a1f7bfe DM |
2053 | static int __bpf_rbtree_add(struct bpf_rb_root *root, |
2054 | struct bpf_rb_node_kern *node, | |
d2dcc67d | 2055 | void *less, struct btf_record *rec, u64 off) |
bd1279ae DM |
2056 | { |
2057 | struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node; | |
0a1f7bfe | 2058 | struct rb_node *parent = NULL, *n = &node->rb_node; |
bd1279ae | 2059 | bpf_callback_t cb = (bpf_callback_t)less; |
bd1279ae DM |
2060 | bool leftmost = true; |
2061 | ||
c3c510ce DM |
2062 | /* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately |
2063 | * check the latter | |
2064 | */ | |
2065 | if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) { | |
d2dcc67d | 2066 | /* Only called from BPF prog, no need to migrate_disable */ |
cc0d76ca | 2067 | __bpf_obj_drop_impl((void *)n - off, rec); |
d2dcc67d DM |
2068 | return -EINVAL; |
2069 | } | |
2070 | ||
bd1279ae DM |
2071 | while (*link) { |
2072 | parent = *link; | |
2073 | if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) { | |
2074 | link = &parent->rb_left; | |
2075 | } else { | |
2076 | link = &parent->rb_right; | |
2077 | leftmost = false; | |
2078 | } | |
2079 | } | |
2080 | ||
d2dcc67d DM |
2081 | rb_link_node(n, parent, link); |
2082 | rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost); | |
c3c510ce | 2083 | WRITE_ONCE(node->owner, root); |
d2dcc67d | 2084 | return 0; |
bd1279ae DM |
2085 | } |
2086 | ||
d2dcc67d DM |
2087 | __bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node, |
2088 | bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b), | |
2089 | void *meta__ign, u64 off) | |
bd1279ae | 2090 | { |
d2dcc67d | 2091 | struct btf_struct_meta *meta = meta__ign; |
0a1f7bfe | 2092 | struct bpf_rb_node_kern *n = (void *)node; |
d2dcc67d | 2093 | |
0a1f7bfe | 2094 | return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off); |
bd1279ae DM |
2095 | } |
2096 | ||
2097 | __bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root) | |
2098 | { | |
2099 | struct rb_root_cached *r = (struct rb_root_cached *)root; | |
2100 | ||
2101 | return (struct bpf_rb_node *)rb_first_cached(r); | |
2102 | } | |
2103 | ||
90660309 DV |
2104 | /** |
2105 | * bpf_task_acquire - Acquire a reference to a task. A task acquired by this | |
2106 | * kfunc which is not stored in a map as a kptr, must be released by calling | |
2107 | * bpf_task_release(). | |
2108 | * @p: The task on which a reference is being acquired. | |
2109 | */ | |
400031e0 | 2110 | __bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p) |
90660309 | 2111 | { |
d02c48fa DV |
2112 | if (refcount_inc_not_zero(&p->rcu_users)) |
2113 | return p; | |
2114 | return NULL; | |
90660309 DV |
2115 | } |
2116 | ||
90660309 | 2117 | /** |
25c5e92d | 2118 | * bpf_task_release - Release the reference acquired on a task. |
90660309 DV |
2119 | * @p: The task on which a reference is being released. |
2120 | */ | |
400031e0 | 2121 | __bpf_kfunc void bpf_task_release(struct task_struct *p) |
90660309 | 2122 | { |
d02c48fa | 2123 | put_task_struct_rcu_user(p); |
90660309 DV |
2124 | } |
2125 | ||
fda01efc DV |
2126 | #ifdef CONFIG_CGROUPS |
2127 | /** | |
2128 | * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by | |
2129 | * this kfunc which is not stored in a map as a kptr, must be released by | |
2130 | * calling bpf_cgroup_release(). | |
2131 | * @cgrp: The cgroup on which a reference is being acquired. | |
2132 | */ | |
400031e0 | 2133 | __bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp) |
fda01efc | 2134 | { |
1d712839 | 2135 | return cgroup_tryget(cgrp) ? cgrp : NULL; |
fda01efc DV |
2136 | } |
2137 | ||
fda01efc | 2138 | /** |
36aa10ff | 2139 | * bpf_cgroup_release - Release the reference acquired on a cgroup. |
fda01efc DV |
2140 | * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to |
2141 | * not be freed until the current grace period has ended, even if its refcount | |
2142 | * drops to 0. | |
2143 | * @cgrp: The cgroup on which a reference is being released. | |
2144 | */ | |
400031e0 | 2145 | __bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp) |
fda01efc | 2146 | { |
fda01efc DV |
2147 | cgroup_put(cgrp); |
2148 | } | |
5ca78670 DV |
2149 | |
2150 | /** | |
2151 | * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor | |
2152 | * array. A cgroup returned by this kfunc which is not subsequently stored in a | |
2153 | * map, must be released by calling bpf_cgroup_release(). | |
2154 | * @cgrp: The cgroup for which we're performing a lookup. | |
2155 | * @level: The level of ancestor to look up. | |
2156 | */ | |
400031e0 | 2157 | __bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level) |
5ca78670 DV |
2158 | { |
2159 | struct cgroup *ancestor; | |
2160 | ||
2161 | if (level > cgrp->level || level < 0) | |
2162 | return NULL; | |
2163 | ||
20c09d92 | 2164 | /* cgrp's refcnt could be 0 here, but ancestors can still be accessed */ |
5ca78670 | 2165 | ancestor = cgrp->ancestors[level]; |
20c09d92 AS |
2166 | if (!cgroup_tryget(ancestor)) |
2167 | return NULL; | |
5ca78670 DV |
2168 | return ancestor; |
2169 | } | |
332ea1f6 TH |
2170 | |
2171 | /** | |
2172 | * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this | |
2173 | * kfunc which is not subsequently stored in a map, must be released by calling | |
2174 | * bpf_cgroup_release(). | |
30a2d832 | 2175 | * @cgid: cgroup id. |
332ea1f6 TH |
2176 | */ |
2177 | __bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid) | |
2178 | { | |
2179 | struct cgroup *cgrp; | |
2180 | ||
2181 | cgrp = cgroup_get_from_id(cgid); | |
2182 | if (IS_ERR(cgrp)) | |
2183 | return NULL; | |
2184 | return cgrp; | |
2185 | } | |
b5ad4cdc FZ |
2186 | |
2187 | /** | |
2188 | * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test | |
2189 | * task's membership of cgroup ancestry. | |
2190 | * @task: the task to be tested | |
2191 | * @ancestor: possible ancestor of @task's cgroup | |
2192 | * | |
2193 | * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor. | |
2194 | * It follows all the same rules as cgroup_is_descendant, and only applies | |
2195 | * to the default hierarchy. | |
2196 | */ | |
2197 | __bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task, | |
2198 | struct cgroup *ancestor) | |
2199 | { | |
2200 | return task_under_cgroup_hierarchy(task, ancestor); | |
2201 | } | |
fda01efc DV |
2202 | #endif /* CONFIG_CGROUPS */ |
2203 | ||
3f0e6f2b DV |
2204 | /** |
2205 | * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up | |
2206 | * in the root pid namespace idr. If a task is returned, it must either be | |
2207 | * stored in a map, or released with bpf_task_release(). | |
2208 | * @pid: The pid of the task being looked up. | |
2209 | */ | |
400031e0 | 2210 | __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid) |
3f0e6f2b DV |
2211 | { |
2212 | struct task_struct *p; | |
2213 | ||
2214 | rcu_read_lock(); | |
2215 | p = find_task_by_pid_ns(pid, &init_pid_ns); | |
2216 | if (p) | |
d02c48fa | 2217 | p = bpf_task_acquire(p); |
3f0e6f2b DV |
2218 | rcu_read_unlock(); |
2219 | ||
2220 | return p; | |
2221 | } | |
2222 | ||
66e3a13e | 2223 | /** |
7ce60b11 DV |
2224 | * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data. |
2225 | * @ptr: The dynptr whose data slice to retrieve | |
2226 | * @offset: Offset into the dynptr | |
3bda08b6 DR |
2227 | * @buffer__opt: User-provided buffer to copy contents into. May be NULL |
2228 | * @buffer__szk: Size (in bytes) of the buffer if present. This is the | |
2229 | * length of the requested slice. This must be a constant. | |
66e3a13e JK |
2230 | * |
2231 | * For non-skb and non-xdp type dynptrs, there is no difference between | |
2232 | * bpf_dynptr_slice and bpf_dynptr_data. | |
2233 | * | |
3bda08b6 DR |
2234 | * If buffer__opt is NULL, the call will fail if buffer_opt was needed. |
2235 | * | |
66e3a13e JK |
2236 | * If the intention is to write to the data slice, please use |
2237 | * bpf_dynptr_slice_rdwr. | |
2238 | * | |
2239 | * The user must check that the returned pointer is not null before using it. | |
2240 | * | |
2241 | * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice | |
2242 | * does not change the underlying packet data pointers, so a call to | |
2243 | * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in | |
2244 | * the bpf program. | |
2245 | * | |
7ce60b11 | 2246 | * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only |
66e3a13e JK |
2247 | * data slice (can be either direct pointer to the data or a pointer to the user |
2248 | * provided buffer, with its contents containing the data, if unable to obtain | |
2249 | * direct pointer) | |
2250 | */ | |
2251 | __bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset, | |
3bda08b6 | 2252 | void *buffer__opt, u32 buffer__szk) |
66e3a13e JK |
2253 | { |
2254 | enum bpf_dynptr_type type; | |
2255 | u32 len = buffer__szk; | |
2256 | int err; | |
2257 | ||
2258 | if (!ptr->data) | |
c45eac53 | 2259 | return NULL; |
66e3a13e JK |
2260 | |
2261 | err = bpf_dynptr_check_off_len(ptr, offset, len); | |
2262 | if (err) | |
c45eac53 | 2263 | return NULL; |
66e3a13e JK |
2264 | |
2265 | type = bpf_dynptr_get_type(ptr); | |
2266 | ||
2267 | switch (type) { | |
2268 | case BPF_DYNPTR_TYPE_LOCAL: | |
2269 | case BPF_DYNPTR_TYPE_RINGBUF: | |
2270 | return ptr->data + ptr->offset + offset; | |
2271 | case BPF_DYNPTR_TYPE_SKB: | |
6f5a630d AS |
2272 | if (buffer__opt) |
2273 | return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt); | |
2274 | else | |
2275 | return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len); | |
66e3a13e JK |
2276 | case BPF_DYNPTR_TYPE_XDP: |
2277 | { | |
2278 | void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len); | |
5426700e | 2279 | if (!IS_ERR_OR_NULL(xdp_ptr)) |
66e3a13e JK |
2280 | return xdp_ptr; |
2281 | ||
3bda08b6 DR |
2282 | if (!buffer__opt) |
2283 | return NULL; | |
2284 | bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false); | |
2285 | return buffer__opt; | |
66e3a13e JK |
2286 | } |
2287 | default: | |
2288 | WARN_ONCE(true, "unknown dynptr type %d\n", type); | |
c45eac53 | 2289 | return NULL; |
66e3a13e JK |
2290 | } |
2291 | } | |
2292 | ||
2293 | /** | |
7ce60b11 DV |
2294 | * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data. |
2295 | * @ptr: The dynptr whose data slice to retrieve | |
2296 | * @offset: Offset into the dynptr | |
3bda08b6 DR |
2297 | * @buffer__opt: User-provided buffer to copy contents into. May be NULL |
2298 | * @buffer__szk: Size (in bytes) of the buffer if present. This is the | |
2299 | * length of the requested slice. This must be a constant. | |
66e3a13e JK |
2300 | * |
2301 | * For non-skb and non-xdp type dynptrs, there is no difference between | |
2302 | * bpf_dynptr_slice and bpf_dynptr_data. | |
2303 | * | |
3bda08b6 DR |
2304 | * If buffer__opt is NULL, the call will fail if buffer_opt was needed. |
2305 | * | |
66e3a13e JK |
2306 | * The returned pointer is writable and may point to either directly the dynptr |
2307 | * data at the requested offset or to the buffer if unable to obtain a direct | |
2308 | * data pointer to (example: the requested slice is to the paged area of an skb | |
2309 | * packet). In the case where the returned pointer is to the buffer, the user | |
2310 | * is responsible for persisting writes through calling bpf_dynptr_write(). This | |
2311 | * usually looks something like this pattern: | |
2312 | * | |
2313 | * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer)); | |
2314 | * if (!eth) | |
2315 | * return TC_ACT_SHOT; | |
2316 | * | |
2317 | * // mutate eth header // | |
2318 | * | |
2319 | * if (eth == buffer) | |
2320 | * bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0); | |
2321 | * | |
2322 | * Please note that, as in the example above, the user must check that the | |
2323 | * returned pointer is not null before using it. | |
2324 | * | |
2325 | * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr | |
2326 | * does not change the underlying packet data pointers, so a call to | |
2327 | * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in | |
2328 | * the bpf program. | |
2329 | * | |
7ce60b11 | 2330 | * Return: NULL if the call failed (eg invalid dynptr), pointer to a |
66e3a13e JK |
2331 | * data slice (can be either direct pointer to the data or a pointer to the user |
2332 | * provided buffer, with its contents containing the data, if unable to obtain | |
2333 | * direct pointer) | |
2334 | */ | |
2335 | __bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset, | |
3bda08b6 | 2336 | void *buffer__opt, u32 buffer__szk) |
66e3a13e | 2337 | { |
540ccf96 | 2338 | if (!ptr->data || __bpf_dynptr_is_rdonly(ptr)) |
c45eac53 | 2339 | return NULL; |
66e3a13e JK |
2340 | |
2341 | /* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice. | |
2342 | * | |
2343 | * For skb-type dynptrs, it is safe to write into the returned pointer | |
2344 | * if the bpf program allows skb data writes. There are two possiblities | |
2345 | * that may occur when calling bpf_dynptr_slice_rdwr: | |
2346 | * | |
2347 | * 1) The requested slice is in the head of the skb. In this case, the | |
2348 | * returned pointer is directly to skb data, and if the skb is cloned, the | |
2349 | * verifier will have uncloned it (see bpf_unclone_prologue()) already. | |
2350 | * The pointer can be directly written into. | |
2351 | * | |
2352 | * 2) Some portion of the requested slice is in the paged buffer area. | |
2353 | * In this case, the requested data will be copied out into the buffer | |
2354 | * and the returned pointer will be a pointer to the buffer. The skb | |
2355 | * will not be pulled. To persist the write, the user will need to call | |
2356 | * bpf_dynptr_write(), which will pull the skb and commit the write. | |
2357 | * | |
2358 | * Similarly for xdp programs, if the requested slice is not across xdp | |
2359 | * fragments, then a direct pointer will be returned, otherwise the data | |
2360 | * will be copied out into the buffer and the user will need to call | |
2361 | * bpf_dynptr_write() to commit changes. | |
2362 | */ | |
3bda08b6 | 2363 | return bpf_dynptr_slice(ptr, offset, buffer__opt, buffer__szk); |
66e3a13e JK |
2364 | } |
2365 | ||
987d0242 JK |
2366 | __bpf_kfunc int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end) |
2367 | { | |
2368 | u32 size; | |
2369 | ||
2370 | if (!ptr->data || start > end) | |
2371 | return -EINVAL; | |
2372 | ||
26662d73 | 2373 | size = __bpf_dynptr_size(ptr); |
987d0242 JK |
2374 | |
2375 | if (start > size || end > size) | |
2376 | return -ERANGE; | |
2377 | ||
2378 | ptr->offset += start; | |
2379 | bpf_dynptr_set_size(ptr, end - start); | |
2380 | ||
2381 | return 0; | |
2382 | } | |
2383 | ||
540ccf96 JK |
2384 | __bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr) |
2385 | { | |
2386 | return !ptr->data; | |
2387 | } | |
2388 | ||
2389 | __bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr) | |
2390 | { | |
2391 | if (!ptr->data) | |
2392 | return false; | |
2393 | ||
2394 | return __bpf_dynptr_is_rdonly(ptr); | |
2395 | } | |
2396 | ||
26662d73 JK |
2397 | __bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr) |
2398 | { | |
2399 | if (!ptr->data) | |
2400 | return -EINVAL; | |
2401 | ||
2402 | return __bpf_dynptr_size(ptr); | |
2403 | } | |
2404 | ||
361f129f JK |
2405 | __bpf_kfunc int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr, |
2406 | struct bpf_dynptr_kern *clone__uninit) | |
2407 | { | |
2408 | if (!ptr->data) { | |
2409 | bpf_dynptr_set_null(clone__uninit); | |
2410 | return -EINVAL; | |
2411 | } | |
2412 | ||
2413 | *clone__uninit = *ptr; | |
2414 | ||
2415 | return 0; | |
2416 | } | |
2417 | ||
400031e0 | 2418 | __bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj) |
fd264ca0 YS |
2419 | { |
2420 | return obj; | |
2421 | } | |
2422 | ||
400031e0 | 2423 | __bpf_kfunc void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) |
a35b9af4 YS |
2424 | { |
2425 | return obj__ign; | |
2426 | } | |
2427 | ||
400031e0 | 2428 | __bpf_kfunc void bpf_rcu_read_lock(void) |
9bb00b28 YS |
2429 | { |
2430 | rcu_read_lock(); | |
2431 | } | |
2432 | ||
400031e0 | 2433 | __bpf_kfunc void bpf_rcu_read_unlock(void) |
9bb00b28 YS |
2434 | { |
2435 | rcu_read_unlock(); | |
2436 | } | |
2437 | ||
958cf2e2 KKD |
2438 | __diag_pop(); |
2439 | ||
2440 | BTF_SET8_START(generic_btf_ids) | |
13379059 AS |
2441 | #ifdef CONFIG_KEXEC_CORE |
2442 | BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE) | |
2443 | #endif | |
958cf2e2 | 2444 | BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL) |
ac9f0605 | 2445 | BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE) |
7793fc3b | 2446 | BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL) |
d2dcc67d DM |
2447 | BTF_ID_FLAGS(func, bpf_list_push_front_impl) |
2448 | BTF_ID_FLAGS(func, bpf_list_push_back_impl) | |
8cab76ec KKD |
2449 | BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL) |
2450 | BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL) | |
d02c48fa | 2451 | BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL) |
90660309 | 2452 | BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE) |
404ad75a | 2453 | BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL) |
d2dcc67d | 2454 | BTF_ID_FLAGS(func, bpf_rbtree_add_impl) |
bd1279ae DM |
2455 | BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL) |
2456 | ||
fda01efc | 2457 | #ifdef CONFIG_CGROUPS |
1d712839 | 2458 | BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL) |
fda01efc | 2459 | BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE) |
20c09d92 | 2460 | BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL) |
332ea1f6 | 2461 | BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL) |
b5ad4cdc | 2462 | BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU) |
fda01efc | 2463 | #endif |
3f0e6f2b | 2464 | BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL) |
958cf2e2 | 2465 | BTF_SET8_END(generic_btf_ids) |
13379059 | 2466 | |
958cf2e2 | 2467 | static const struct btf_kfunc_id_set generic_kfunc_set = { |
13379059 | 2468 | .owner = THIS_MODULE, |
958cf2e2 | 2469 | .set = &generic_btf_ids, |
13379059 AS |
2470 | }; |
2471 | ||
cfe14564 | 2472 | |
90660309 DV |
2473 | BTF_ID_LIST(generic_dtor_ids) |
2474 | BTF_ID(struct, task_struct) | |
2475 | BTF_ID(func, bpf_task_release) | |
fda01efc DV |
2476 | #ifdef CONFIG_CGROUPS |
2477 | BTF_ID(struct, cgroup) | |
2478 | BTF_ID(func, bpf_cgroup_release) | |
2479 | #endif | |
90660309 | 2480 | |
cfe14564 | 2481 | BTF_SET8_START(common_btf_ids) |
fd264ca0 | 2482 | BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 2483 | BTF_ID_FLAGS(func, bpf_rdonly_cast) |
9bb00b28 YS |
2484 | BTF_ID_FLAGS(func, bpf_rcu_read_lock) |
2485 | BTF_ID_FLAGS(func, bpf_rcu_read_unlock) | |
66e3a13e JK |
2486 | BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL) |
2487 | BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL) | |
6018e1f4 AN |
2488 | BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW) |
2489 | BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL) | |
2490 | BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY) | |
987d0242 | 2491 | BTF_ID_FLAGS(func, bpf_dynptr_adjust) |
540ccf96 JK |
2492 | BTF_ID_FLAGS(func, bpf_dynptr_is_null) |
2493 | BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly) | |
26662d73 | 2494 | BTF_ID_FLAGS(func, bpf_dynptr_size) |
361f129f | 2495 | BTF_ID_FLAGS(func, bpf_dynptr_clone) |
cfe14564 YS |
2496 | BTF_SET8_END(common_btf_ids) |
2497 | ||
2498 | static const struct btf_kfunc_id_set common_kfunc_set = { | |
2499 | .owner = THIS_MODULE, | |
2500 | .set = &common_btf_ids, | |
2501 | }; | |
2502 | ||
13379059 AS |
2503 | static int __init kfunc_init(void) |
2504 | { | |
2fcc6081 | 2505 | int ret; |
90660309 DV |
2506 | const struct btf_id_dtor_kfunc generic_dtors[] = { |
2507 | { | |
2fcc6081 DV |
2508 | .btf_id = generic_dtor_ids[0], |
2509 | .kfunc_btf_id = generic_dtor_ids[1] | |
90660309 | 2510 | }, |
fda01efc DV |
2511 | #ifdef CONFIG_CGROUPS |
2512 | { | |
2fcc6081 DV |
2513 | .btf_id = generic_dtor_ids[2], |
2514 | .kfunc_btf_id = generic_dtor_ids[3] | |
fda01efc DV |
2515 | }, |
2516 | #endif | |
90660309 | 2517 | }; |
8cab76ec KKD |
2518 | |
2519 | ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set); | |
90660309 DV |
2520 | ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set); |
2521 | ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set); | |
cfe14564 | 2522 | ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors, |
90660309 DV |
2523 | ARRAY_SIZE(generic_dtors), |
2524 | THIS_MODULE); | |
cfe14564 | 2525 | return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set); |
13379059 AS |
2526 | } |
2527 | ||
2528 | late_initcall(kfunc_init); |