1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* Copyright (c) 2016 Facebook |
3 | */ |
4 | #include <linux/bpf.h> |
5 | #include <linux/jhash.h> |
6 | #include <linux/filter.h> |
7 | #include <linux/kernel.h> |
8 | #include <linux/stacktrace.h> |
9 | #include <linux/perf_event.h> |
10 | #include <linux/btf_ids.h> |
11 | #include <linux/buildid.h> |
12 | #include "percpu_freelist.h" |
13 | #include "mmap_unlock_work.h" |
14 | |
15 | #define STACK_CREATE_FLAG_MASK \ |
16 | (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY | \ |
17 | BPF_F_STACK_BUILD_ID) |
18 | |
19 | struct stack_map_bucket { |
20 | struct pcpu_freelist_node fnode; |
21 | u32 hash; |
22 | u32 nr; |
23 | u64 data[]; |
24 | }; |
25 | |
26 | struct bpf_stack_map { |
27 | struct bpf_map map; |
28 | void *elems; |
29 | struct pcpu_freelist freelist; |
30 | u32 n_buckets; |
31 | struct stack_map_bucket *buckets[] __counted_by(n_buckets); |
32 | }; |
33 | |
34 | static inline bool stack_map_use_build_id(struct bpf_map *map) |
35 | { |
36 | return (map->map_flags & BPF_F_STACK_BUILD_ID); |
37 | } |
38 | |
39 | static inline int stack_map_data_size(struct bpf_map *map) |
40 | { |
41 | return stack_map_use_build_id(map) ? |
42 | sizeof(struct bpf_stack_build_id) : sizeof(u64); |
43 | } |
44 | |
45 | static int prealloc_elems_and_freelist(struct bpf_stack_map *smap) |
46 | { |
47 | u64 elem_size = sizeof(struct stack_map_bucket) + |
48 | (u64)smap->map.value_size; |
49 | int err; |
50 | |
51 | smap->elems = bpf_map_area_alloc(size: elem_size * smap->map.max_entries, |
52 | numa_node: smap->map.numa_node); |
53 | if (!smap->elems) |
54 | return -ENOMEM; |
55 | |
56 | err = pcpu_freelist_init(&smap->freelist); |
57 | if (err) |
58 | goto free_elems; |
59 | |
60 | pcpu_freelist_populate(s: &smap->freelist, buf: smap->elems, elem_size, |
61 | nr_elems: smap->map.max_entries); |
62 | return 0; |
63 | |
64 | free_elems: |
65 | bpf_map_area_free(base: smap->elems); |
66 | return err; |
67 | } |
68 | |
69 | /* Called from syscall */ |
70 | static struct bpf_map *stack_map_alloc(union bpf_attr *attr) |
71 | { |
72 | u32 value_size = attr->value_size; |
73 | struct bpf_stack_map *smap; |
74 | u64 cost, n_buckets; |
75 | int err; |
76 | |
77 | if (attr->map_flags & ~STACK_CREATE_FLAG_MASK) |
78 | return ERR_PTR(error: -EINVAL); |
79 | |
80 | /* check sanity of attributes */ |
81 | if (attr->max_entries == 0 || attr->key_size != 4 || |
82 | value_size < 8 || value_size % 8) |
83 | return ERR_PTR(error: -EINVAL); |
84 | |
85 | BUILD_BUG_ON(sizeof(struct bpf_stack_build_id) % sizeof(u64)); |
86 | if (attr->map_flags & BPF_F_STACK_BUILD_ID) { |
87 | if (value_size % sizeof(struct bpf_stack_build_id) || |
88 | value_size / sizeof(struct bpf_stack_build_id) |
89 | > sysctl_perf_event_max_stack) |
90 | return ERR_PTR(error: -EINVAL); |
91 | } else if (value_size / 8 > sysctl_perf_event_max_stack) |
92 | return ERR_PTR(error: -EINVAL); |
93 | |
94 | /* hash table size must be power of 2 */ |
95 | n_buckets = roundup_pow_of_two(attr->max_entries); |
96 | if (!n_buckets) |
97 | return ERR_PTR(error: -E2BIG); |
98 | |
99 | cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap); |
100 | smap = bpf_map_area_alloc(size: cost, numa_node: bpf_map_attr_numa_node(attr)); |
101 | if (!smap) |
102 | return ERR_PTR(error: -ENOMEM); |
103 | |
104 | bpf_map_init_from_attr(map: &smap->map, attr); |
105 | smap->n_buckets = n_buckets; |
106 | |
107 | err = get_callchain_buffers(max_stack: sysctl_perf_event_max_stack); |
108 | if (err) |
109 | goto free_smap; |
110 | |
111 | err = prealloc_elems_and_freelist(smap); |
112 | if (err) |
113 | goto put_buffers; |
114 | |
115 | return &smap->map; |
116 | |
117 | put_buffers: |
118 | put_callchain_buffers(); |
119 | free_smap: |
120 | bpf_map_area_free(base: smap); |
121 | return ERR_PTR(error: err); |
122 | } |
123 | |
124 | static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs, |
125 | u64 *ips, u32 trace_nr, bool user) |
126 | { |
127 | int i; |
128 | struct mmap_unlock_irq_work *work = NULL; |
129 | bool irq_work_busy = bpf_mmap_unlock_get_irq_work(work_ptr: &work); |
130 | struct vm_area_struct *vma, *prev_vma = NULL; |
131 | const char *prev_build_id; |
132 | |
133 | /* If the irq_work is in use, fall back to report ips. Same |
134 | * fallback is used for kernel stack (!user) on a stackmap with |
135 | * build_id. |
136 | */ |
137 | if (!user || !current || !current->mm || irq_work_busy || |
138 | !mmap_read_trylock(current->mm)) { |
139 | /* cannot access current->mm, fall back to ips */ |
140 | for (i = 0; i < trace_nr; i++) { |
141 | id_offs[i].status = BPF_STACK_BUILD_ID_IP; |
142 | id_offs[i].ip = ips[i]; |
143 | memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX); |
144 | } |
145 | return; |
146 | } |
147 | |
148 | for (i = 0; i < trace_nr; i++) { |
149 | if (range_in_vma(vma: prev_vma, start: ips[i], end: ips[i])) { |
150 | vma = prev_vma; |
151 | memcpy(id_offs[i].build_id, prev_build_id, |
152 | BUILD_ID_SIZE_MAX); |
153 | goto build_id_valid; |
154 | } |
155 | vma = find_vma(current->mm, addr: ips[i]); |
156 | if (!vma || build_id_parse(vma, build_id: id_offs[i].build_id, NULL)) { |
157 | /* per entry fall back to ips */ |
158 | id_offs[i].status = BPF_STACK_BUILD_ID_IP; |
159 | id_offs[i].ip = ips[i]; |
160 | memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX); |
161 | continue; |
162 | } |
163 | build_id_valid: |
164 | id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ips[i] |
165 | - vma->vm_start; |
166 | id_offs[i].status = BPF_STACK_BUILD_ID_VALID; |
167 | prev_vma = vma; |
168 | prev_build_id = id_offs[i].build_id; |
169 | } |
170 | bpf_mmap_unlock_mm(work, current->mm); |
171 | } |
172 | |
173 | static struct perf_callchain_entry * |
174 | get_callchain_entry_for_task(struct task_struct *task, u32 max_depth) |
175 | { |
176 | #ifdef CONFIG_STACKTRACE |
177 | struct perf_callchain_entry *entry; |
178 | int rctx; |
179 | |
180 | entry = get_callchain_entry(rctx: &rctx); |
181 | |
182 | if (!entry) |
183 | return NULL; |
184 | |
185 | entry->nr = stack_trace_save_tsk(task, store: (unsigned long *)entry->ip, |
186 | size: max_depth, skipnr: 0); |
187 | |
188 | /* stack_trace_save_tsk() works on unsigned long array, while |
189 | * perf_callchain_entry uses u64 array. For 32-bit systems, it is |
190 | * necessary to fix this mismatch. |
191 | */ |
192 | if (__BITS_PER_LONG != 64) { |
193 | unsigned long *from = (unsigned long *) entry->ip; |
194 | u64 *to = entry->ip; |
195 | int i; |
196 | |
197 | /* copy data from the end to avoid using extra buffer */ |
198 | for (i = entry->nr - 1; i >= 0; i--) |
199 | to[i] = (u64)(from[i]); |
200 | } |
201 | |
202 | put_callchain_entry(rctx); |
203 | |
204 | return entry; |
205 | #else /* CONFIG_STACKTRACE */ |
206 | return NULL; |
207 | #endif |
208 | } |
209 | |
210 | static long __bpf_get_stackid(struct bpf_map *map, |
211 | struct perf_callchain_entry *trace, u64 flags) |
212 | { |
213 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map); |
214 | struct stack_map_bucket *bucket, *new_bucket, *old_bucket; |
215 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK; |
216 | u32 hash, id, trace_nr, trace_len; |
217 | bool user = flags & BPF_F_USER_STACK; |
218 | u64 *ips; |
219 | bool hash_matches; |
220 | |
221 | if (trace->nr <= skip) |
222 | /* skipping more than usable stack trace */ |
223 | return -EFAULT; |
224 | |
225 | trace_nr = trace->nr - skip; |
226 | trace_len = trace_nr * sizeof(u64); |
227 | ips = trace->ip + skip; |
228 | hash = jhash2(k: (u32 *)ips, length: trace_len / sizeof(u32), initval: 0); |
229 | id = hash & (smap->n_buckets - 1); |
230 | bucket = READ_ONCE(smap->buckets[id]); |
231 | |
232 | hash_matches = bucket && bucket->hash == hash; |
233 | /* fast cmp */ |
234 | if (hash_matches && flags & BPF_F_FAST_STACK_CMP) |
235 | return id; |
236 | |
237 | if (stack_map_use_build_id(map)) { |
238 | /* for build_id+offset, pop a bucket before slow cmp */ |
239 | new_bucket = (struct stack_map_bucket *) |
240 | pcpu_freelist_pop(&smap->freelist); |
241 | if (unlikely(!new_bucket)) |
242 | return -ENOMEM; |
243 | new_bucket->nr = trace_nr; |
244 | stack_map_get_build_id_offset( |
245 | id_offs: (struct bpf_stack_build_id *)new_bucket->data, |
246 | ips, trace_nr, user); |
247 | trace_len = trace_nr * sizeof(struct bpf_stack_build_id); |
248 | if (hash_matches && bucket->nr == trace_nr && |
249 | memcmp(p: bucket->data, q: new_bucket->data, size: trace_len) == 0) { |
250 | pcpu_freelist_push(&smap->freelist, &new_bucket->fnode); |
251 | return id; |
252 | } |
253 | if (bucket && !(flags & BPF_F_REUSE_STACKID)) { |
254 | pcpu_freelist_push(&smap->freelist, &new_bucket->fnode); |
255 | return -EEXIST; |
256 | } |
257 | } else { |
258 | if (hash_matches && bucket->nr == trace_nr && |
259 | memcmp(p: bucket->data, q: ips, size: trace_len) == 0) |
260 | return id; |
261 | if (bucket && !(flags & BPF_F_REUSE_STACKID)) |
262 | return -EEXIST; |
263 | |
264 | new_bucket = (struct stack_map_bucket *) |
265 | pcpu_freelist_pop(&smap->freelist); |
266 | if (unlikely(!new_bucket)) |
267 | return -ENOMEM; |
268 | memcpy(new_bucket->data, ips, trace_len); |
269 | } |
270 | |
271 | new_bucket->hash = hash; |
272 | new_bucket->nr = trace_nr; |
273 | |
274 | old_bucket = xchg(&smap->buckets[id], new_bucket); |
275 | if (old_bucket) |
276 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode); |
277 | return id; |
278 | } |
279 | |
280 | BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map, |
281 | u64, flags) |
282 | { |
283 | u32 max_depth = map->value_size / stack_map_data_size(map); |
284 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK; |
285 | bool user = flags & BPF_F_USER_STACK; |
286 | struct perf_callchain_entry *trace; |
287 | bool kernel = !user; |
288 | |
289 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | |
290 | BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID))) |
291 | return -EINVAL; |
292 | |
293 | max_depth += skip; |
294 | if (max_depth > sysctl_perf_event_max_stack) |
295 | max_depth = sysctl_perf_event_max_stack; |
296 | |
297 | trace = get_perf_callchain(regs, init_nr: 0, kernel, user, max_stack: max_depth, |
298 | crosstask: false, add_mark: false); |
299 | |
300 | if (unlikely(!trace)) |
301 | /* couldn't fetch the stack trace */ |
302 | return -EFAULT; |
303 | |
304 | return __bpf_get_stackid(map, trace, flags); |
305 | } |
306 | |
307 | const struct bpf_func_proto bpf_get_stackid_proto = { |
308 | .func = bpf_get_stackid, |
309 | .gpl_only = true, |
310 | .ret_type = RET_INTEGER, |
311 | .arg1_type = ARG_PTR_TO_CTX, |
312 | .arg2_type = ARG_CONST_MAP_PTR, |
313 | .arg3_type = ARG_ANYTHING, |
314 | }; |
315 | |
316 | static __u64 count_kernel_ip(struct perf_callchain_entry *trace) |
317 | { |
318 | __u64 nr_kernel = 0; |
319 | |
320 | while (nr_kernel < trace->nr) { |
321 | if (trace->ip[nr_kernel] == PERF_CONTEXT_USER) |
322 | break; |
323 | nr_kernel++; |
324 | } |
325 | return nr_kernel; |
326 | } |
327 | |
328 | BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx, |
329 | struct bpf_map *, map, u64, flags) |
330 | { |
331 | struct perf_event *event = ctx->event; |
332 | struct perf_callchain_entry *trace; |
333 | bool kernel, user; |
334 | __u64 nr_kernel; |
335 | int ret; |
336 | |
337 | /* perf_sample_data doesn't have callchain, use bpf_get_stackid */ |
338 | if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) |
339 | return bpf_get_stackid(regs: (unsigned long)(ctx->regs), |
340 | map: (unsigned long) map, flags, ur_1: 0, ur_2: 0); |
341 | |
342 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | |
343 | BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID))) |
344 | return -EINVAL; |
345 | |
346 | user = flags & BPF_F_USER_STACK; |
347 | kernel = !user; |
348 | |
349 | trace = ctx->data->callchain; |
350 | if (unlikely(!trace)) |
351 | return -EFAULT; |
352 | |
353 | nr_kernel = count_kernel_ip(trace); |
354 | |
355 | if (kernel) { |
356 | __u64 nr = trace->nr; |
357 | |
358 | trace->nr = nr_kernel; |
359 | ret = __bpf_get_stackid(map, trace, flags); |
360 | |
361 | /* restore nr */ |
362 | trace->nr = nr; |
363 | } else { /* user */ |
364 | u64 skip = flags & BPF_F_SKIP_FIELD_MASK; |
365 | |
366 | skip += nr_kernel; |
367 | if (skip > BPF_F_SKIP_FIELD_MASK) |
368 | return -EFAULT; |
369 | |
370 | flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip; |
371 | ret = __bpf_get_stackid(map, trace, flags); |
372 | } |
373 | return ret; |
374 | } |
375 | |
376 | const struct bpf_func_proto bpf_get_stackid_proto_pe = { |
377 | .func = bpf_get_stackid_pe, |
378 | .gpl_only = false, |
379 | .ret_type = RET_INTEGER, |
380 | .arg1_type = ARG_PTR_TO_CTX, |
381 | .arg2_type = ARG_CONST_MAP_PTR, |
382 | .arg3_type = ARG_ANYTHING, |
383 | }; |
384 | |
385 | static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task, |
386 | struct perf_callchain_entry *trace_in, |
387 | void *buf, u32 size, u64 flags) |
388 | { |
389 | u32 trace_nr, copy_len, elem_size, num_elem, max_depth; |
390 | bool user_build_id = flags & BPF_F_USER_BUILD_ID; |
391 | u32 skip = flags & BPF_F_SKIP_FIELD_MASK; |
392 | bool user = flags & BPF_F_USER_STACK; |
393 | struct perf_callchain_entry *trace; |
394 | bool kernel = !user; |
395 | int err = -EINVAL; |
396 | u64 *ips; |
397 | |
398 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | |
399 | BPF_F_USER_BUILD_ID))) |
400 | goto clear; |
401 | if (kernel && user_build_id) |
402 | goto clear; |
403 | |
404 | elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id) |
405 | : sizeof(u64); |
406 | if (unlikely(size % elem_size)) |
407 | goto clear; |
408 | |
409 | /* cannot get valid user stack for task without user_mode regs */ |
410 | if (task && user && !user_mode(regs)) |
411 | goto err_fault; |
412 | |
413 | num_elem = size / elem_size; |
414 | max_depth = num_elem + skip; |
415 | if (sysctl_perf_event_max_stack < max_depth) |
416 | max_depth = sysctl_perf_event_max_stack; |
417 | |
418 | if (trace_in) |
419 | trace = trace_in; |
420 | else if (kernel && task) |
421 | trace = get_callchain_entry_for_task(task, max_depth); |
422 | else |
423 | trace = get_perf_callchain(regs, init_nr: 0, kernel, user, max_stack: max_depth, |
424 | crosstask: false, add_mark: false); |
425 | if (unlikely(!trace)) |
426 | goto err_fault; |
427 | |
428 | if (trace->nr < skip) |
429 | goto err_fault; |
430 | |
431 | trace_nr = trace->nr - skip; |
432 | trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem; |
433 | copy_len = trace_nr * elem_size; |
434 | |
435 | ips = trace->ip + skip; |
436 | if (user && user_build_id) |
437 | stack_map_get_build_id_offset(id_offs: buf, ips, trace_nr, user); |
438 | else |
439 | memcpy(buf, ips, copy_len); |
440 | |
441 | if (size > copy_len) |
442 | memset(buf + copy_len, 0, size - copy_len); |
443 | return copy_len; |
444 | |
445 | err_fault: |
446 | err = -EFAULT; |
447 | clear: |
448 | memset(buf, 0, size); |
449 | return err; |
450 | } |
451 | |
452 | BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size, |
453 | u64, flags) |
454 | { |
455 | return __bpf_get_stack(regs, NULL, NULL, buf, size, flags); |
456 | } |
457 | |
458 | const struct bpf_func_proto bpf_get_stack_proto = { |
459 | .func = bpf_get_stack, |
460 | .gpl_only = true, |
461 | .ret_type = RET_INTEGER, |
462 | .arg1_type = ARG_PTR_TO_CTX, |
463 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, |
464 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, |
465 | .arg4_type = ARG_ANYTHING, |
466 | }; |
467 | |
468 | BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf, |
469 | u32, size, u64, flags) |
470 | { |
471 | struct pt_regs *regs; |
472 | long res = -EINVAL; |
473 | |
474 | if (!try_get_task_stack(tsk: task)) |
475 | return -EFAULT; |
476 | |
477 | regs = task_pt_regs(task); |
478 | if (regs) |
479 | res = __bpf_get_stack(regs, task, NULL, buf, size, flags); |
480 | put_task_stack(tsk: task); |
481 | |
482 | return res; |
483 | } |
484 | |
485 | const struct bpf_func_proto bpf_get_task_stack_proto = { |
486 | .func = bpf_get_task_stack, |
487 | .gpl_only = false, |
488 | .ret_type = RET_INTEGER, |
489 | .arg1_type = ARG_PTR_TO_BTF_ID, |
490 | .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK], |
491 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, |
492 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, |
493 | .arg4_type = ARG_ANYTHING, |
494 | }; |
495 | |
496 | BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx, |
497 | void *, buf, u32, size, u64, flags) |
498 | { |
499 | struct pt_regs *regs = (struct pt_regs *)(ctx->regs); |
500 | struct perf_event *event = ctx->event; |
501 | struct perf_callchain_entry *trace; |
502 | bool kernel, user; |
503 | int err = -EINVAL; |
504 | __u64 nr_kernel; |
505 | |
506 | if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) |
507 | return __bpf_get_stack(regs, NULL, NULL, buf, size, flags); |
508 | |
509 | if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | |
510 | BPF_F_USER_BUILD_ID))) |
511 | goto clear; |
512 | |
513 | user = flags & BPF_F_USER_STACK; |
514 | kernel = !user; |
515 | |
516 | err = -EFAULT; |
517 | trace = ctx->data->callchain; |
518 | if (unlikely(!trace)) |
519 | goto clear; |
520 | |
521 | nr_kernel = count_kernel_ip(trace); |
522 | |
523 | if (kernel) { |
524 | __u64 nr = trace->nr; |
525 | |
526 | trace->nr = nr_kernel; |
527 | err = __bpf_get_stack(regs, NULL, trace_in: trace, buf, size, flags); |
528 | |
529 | /* restore nr */ |
530 | trace->nr = nr; |
531 | } else { /* user */ |
532 | u64 skip = flags & BPF_F_SKIP_FIELD_MASK; |
533 | |
534 | skip += nr_kernel; |
535 | if (skip > BPF_F_SKIP_FIELD_MASK) |
536 | goto clear; |
537 | |
538 | flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip; |
539 | err = __bpf_get_stack(regs, NULL, trace_in: trace, buf, size, flags); |
540 | } |
541 | return err; |
542 | |
543 | clear: |
544 | memset(buf, 0, size); |
545 | return err; |
546 | |
547 | } |
548 | |
549 | const struct bpf_func_proto bpf_get_stack_proto_pe = { |
550 | .func = bpf_get_stack_pe, |
551 | .gpl_only = true, |
552 | .ret_type = RET_INTEGER, |
553 | .arg1_type = ARG_PTR_TO_CTX, |
554 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, |
555 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, |
556 | .arg4_type = ARG_ANYTHING, |
557 | }; |
558 | |
559 | /* Called from eBPF program */ |
560 | static void *stack_map_lookup_elem(struct bpf_map *map, void *key) |
561 | { |
562 | return ERR_PTR(error: -EOPNOTSUPP); |
563 | } |
564 | |
565 | /* Called from syscall */ |
566 | int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
567 | { |
568 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map); |
569 | struct stack_map_bucket *bucket, *old_bucket; |
570 | u32 id = *(u32 *)key, trace_len; |
571 | |
572 | if (unlikely(id >= smap->n_buckets)) |
573 | return -ENOENT; |
574 | |
575 | bucket = xchg(&smap->buckets[id], NULL); |
576 | if (!bucket) |
577 | return -ENOENT; |
578 | |
579 | trace_len = bucket->nr * stack_map_data_size(map); |
580 | memcpy(value, bucket->data, trace_len); |
581 | memset(value + trace_len, 0, map->value_size - trace_len); |
582 | |
583 | old_bucket = xchg(&smap->buckets[id], bucket); |
584 | if (old_bucket) |
585 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode); |
586 | return 0; |
587 | } |
588 | |
589 | static int stack_map_get_next_key(struct bpf_map *map, void *key, |
590 | void *next_key) |
591 | { |
592 | struct bpf_stack_map *smap = container_of(map, |
593 | struct bpf_stack_map, map); |
594 | u32 id; |
595 | |
596 | WARN_ON_ONCE(!rcu_read_lock_held()); |
597 | |
598 | if (!key) { |
599 | id = 0; |
600 | } else { |
601 | id = *(u32 *)key; |
602 | if (id >= smap->n_buckets || !smap->buckets[id]) |
603 | id = 0; |
604 | else |
605 | id++; |
606 | } |
607 | |
608 | while (id < smap->n_buckets && !smap->buckets[id]) |
609 | id++; |
610 | |
611 | if (id >= smap->n_buckets) |
612 | return -ENOENT; |
613 | |
614 | *(u32 *)next_key = id; |
615 | return 0; |
616 | } |
617 | |
618 | static long stack_map_update_elem(struct bpf_map *map, void *key, void *value, |
619 | u64 map_flags) |
620 | { |
621 | return -EINVAL; |
622 | } |
623 | |
624 | /* Called from syscall or from eBPF program */ |
625 | static long stack_map_delete_elem(struct bpf_map *map, void *key) |
626 | { |
627 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map); |
628 | struct stack_map_bucket *old_bucket; |
629 | u32 id = *(u32 *)key; |
630 | |
631 | if (unlikely(id >= smap->n_buckets)) |
632 | return -E2BIG; |
633 | |
634 | old_bucket = xchg(&smap->buckets[id], NULL); |
635 | if (old_bucket) { |
636 | pcpu_freelist_push(&smap->freelist, &old_bucket->fnode); |
637 | return 0; |
638 | } else { |
639 | return -ENOENT; |
640 | } |
641 | } |
642 | |
643 | /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ |
644 | static void stack_map_free(struct bpf_map *map) |
645 | { |
646 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map); |
647 | |
648 | bpf_map_area_free(base: smap->elems); |
649 | pcpu_freelist_destroy(s: &smap->freelist); |
650 | bpf_map_area_free(base: smap); |
651 | put_callchain_buffers(); |
652 | } |
653 | |
654 | static u64 stack_map_mem_usage(const struct bpf_map *map) |
655 | { |
656 | struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map); |
657 | u64 value_size = map->value_size; |
658 | u64 n_buckets = smap->n_buckets; |
659 | u64 enties = map->max_entries; |
660 | u64 usage = sizeof(*smap); |
661 | |
662 | usage += n_buckets * sizeof(struct stack_map_bucket *); |
663 | usage += enties * (sizeof(struct stack_map_bucket) + value_size); |
664 | return usage; |
665 | } |
666 | |
667 | BTF_ID_LIST_SINGLE(stack_trace_map_btf_ids, struct, bpf_stack_map) |
668 | const struct bpf_map_ops stack_trace_map_ops = { |
669 | .map_meta_equal = bpf_map_meta_equal, |
670 | .map_alloc = stack_map_alloc, |
671 | .map_free = stack_map_free, |
672 | .map_get_next_key = stack_map_get_next_key, |
673 | .map_lookup_elem = stack_map_lookup_elem, |
674 | .map_update_elem = stack_map_update_elem, |
675 | .map_delete_elem = stack_map_delete_elem, |
676 | .map_check_btf = map_check_no_btf, |
677 | .map_mem_usage = stack_map_mem_usage, |
678 | .map_btf_id = &stack_trace_map_btf_ids[0], |
679 | }; |
680 | |