xref: /linux-6.15/kernel/bpf/stackmap.c (revision 8eb943fc)
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[];
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(elem_size * smap->map.max_entries,
52 					 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(&smap->freelist, smap->elems, elem_size,
61 			       smap->map.max_entries);
62 	return 0;
63 
64 free_elems:
65 	bpf_map_area_free(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 (!bpf_capable())
78 		return ERR_PTR(-EPERM);
79 
80 	if (attr->map_flags & ~STACK_CREATE_FLAG_MASK)
81 		return ERR_PTR(-EINVAL);
82 
83 	/* check sanity of attributes */
84 	if (attr->max_entries == 0 || attr->key_size != 4 ||
85 	    value_size < 8 || value_size % 8)
86 		return ERR_PTR(-EINVAL);
87 
88 	BUILD_BUG_ON(sizeof(struct bpf_stack_build_id) % sizeof(u64));
89 	if (attr->map_flags & BPF_F_STACK_BUILD_ID) {
90 		if (value_size % sizeof(struct bpf_stack_build_id) ||
91 		    value_size / sizeof(struct bpf_stack_build_id)
92 		    > sysctl_perf_event_max_stack)
93 			return ERR_PTR(-EINVAL);
94 	} else if (value_size / 8 > sysctl_perf_event_max_stack)
95 		return ERR_PTR(-EINVAL);
96 
97 	/* hash table size must be power of 2 */
98 	n_buckets = roundup_pow_of_two(attr->max_entries);
99 	if (!n_buckets)
100 		return ERR_PTR(-E2BIG);
101 
102 	cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
103 	cost += n_buckets * (value_size + sizeof(struct stack_map_bucket));
104 	smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
105 	if (!smap)
106 		return ERR_PTR(-ENOMEM);
107 
108 	bpf_map_init_from_attr(&smap->map, attr);
109 	smap->n_buckets = n_buckets;
110 
111 	err = get_callchain_buffers(sysctl_perf_event_max_stack);
112 	if (err)
113 		goto free_smap;
114 
115 	err = prealloc_elems_and_freelist(smap);
116 	if (err)
117 		goto put_buffers;
118 
119 	return &smap->map;
120 
121 put_buffers:
122 	put_callchain_buffers();
123 free_smap:
124 	bpf_map_area_free(smap);
125 	return ERR_PTR(err);
126 }
127 
128 static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
129 					  u64 *ips, u32 trace_nr, bool user)
130 {
131 	int i;
132 	struct mmap_unlock_irq_work *work = NULL;
133 	bool irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
134 	struct vm_area_struct *vma, *prev_vma = NULL;
135 	const char *prev_build_id;
136 
137 	/* If the irq_work is in use, fall back to report ips. Same
138 	 * fallback is used for kernel stack (!user) on a stackmap with
139 	 * build_id.
140 	 */
141 	if (!user || !current || !current->mm || irq_work_busy ||
142 	    !mmap_read_trylock(current->mm)) {
143 		/* cannot access current->mm, fall back to ips */
144 		for (i = 0; i < trace_nr; i++) {
145 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
146 			id_offs[i].ip = ips[i];
147 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
148 		}
149 		return;
150 	}
151 
152 	for (i = 0; i < trace_nr; i++) {
153 		if (range_in_vma(prev_vma, ips[i], ips[i])) {
154 			vma = prev_vma;
155 			memcpy(id_offs[i].build_id, prev_build_id,
156 			       BUILD_ID_SIZE_MAX);
157 			goto build_id_valid;
158 		}
159 		vma = find_vma(current->mm, ips[i]);
160 		if (!vma || build_id_parse(vma, id_offs[i].build_id, NULL)) {
161 			/* per entry fall back to ips */
162 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
163 			id_offs[i].ip = ips[i];
164 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
165 			continue;
166 		}
167 build_id_valid:
168 		id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ips[i]
169 			- vma->vm_start;
170 		id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
171 		prev_vma = vma;
172 		prev_build_id = id_offs[i].build_id;
173 	}
174 	bpf_mmap_unlock_mm(work, current->mm);
175 }
176 
177 static struct perf_callchain_entry *
178 get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
179 {
180 #ifdef CONFIG_STACKTRACE
181 	struct perf_callchain_entry *entry;
182 	int rctx;
183 
184 	entry = get_callchain_entry(&rctx);
185 
186 	if (!entry)
187 		return NULL;
188 
189 	entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
190 					 max_depth, 0);
191 
192 	/* stack_trace_save_tsk() works on unsigned long array, while
193 	 * perf_callchain_entry uses u64 array. For 32-bit systems, it is
194 	 * necessary to fix this mismatch.
195 	 */
196 	if (__BITS_PER_LONG != 64) {
197 		unsigned long *from = (unsigned long *) entry->ip;
198 		u64 *to = entry->ip;
199 		int i;
200 
201 		/* copy data from the end to avoid using extra buffer */
202 		for (i = entry->nr - 1; i >= 0; i--)
203 			to[i] = (u64)(from[i]);
204 	}
205 
206 	put_callchain_entry(rctx);
207 
208 	return entry;
209 #else /* CONFIG_STACKTRACE */
210 	return NULL;
211 #endif
212 }
213 
214 static long __bpf_get_stackid(struct bpf_map *map,
215 			      struct perf_callchain_entry *trace, u64 flags)
216 {
217 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
218 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
219 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
220 	u32 hash, id, trace_nr, trace_len;
221 	bool user = flags & BPF_F_USER_STACK;
222 	u64 *ips;
223 	bool hash_matches;
224 
225 	if (trace->nr <= skip)
226 		/* skipping more than usable stack trace */
227 		return -EFAULT;
228 
229 	trace_nr = trace->nr - skip;
230 	trace_len = trace_nr * sizeof(u64);
231 	ips = trace->ip + skip;
232 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
233 	id = hash & (smap->n_buckets - 1);
234 	bucket = READ_ONCE(smap->buckets[id]);
235 
236 	hash_matches = bucket && bucket->hash == hash;
237 	/* fast cmp */
238 	if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
239 		return id;
240 
241 	if (stack_map_use_build_id(map)) {
242 		/* for build_id+offset, pop a bucket before slow cmp */
243 		new_bucket = (struct stack_map_bucket *)
244 			pcpu_freelist_pop(&smap->freelist);
245 		if (unlikely(!new_bucket))
246 			return -ENOMEM;
247 		new_bucket->nr = trace_nr;
248 		stack_map_get_build_id_offset(
249 			(struct bpf_stack_build_id *)new_bucket->data,
250 			ips, trace_nr, user);
251 		trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
252 		if (hash_matches && bucket->nr == trace_nr &&
253 		    memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
254 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
255 			return id;
256 		}
257 		if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
258 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
259 			return -EEXIST;
260 		}
261 	} else {
262 		if (hash_matches && bucket->nr == trace_nr &&
263 		    memcmp(bucket->data, ips, trace_len) == 0)
264 			return id;
265 		if (bucket && !(flags & BPF_F_REUSE_STACKID))
266 			return -EEXIST;
267 
268 		new_bucket = (struct stack_map_bucket *)
269 			pcpu_freelist_pop(&smap->freelist);
270 		if (unlikely(!new_bucket))
271 			return -ENOMEM;
272 		memcpy(new_bucket->data, ips, trace_len);
273 	}
274 
275 	new_bucket->hash = hash;
276 	new_bucket->nr = trace_nr;
277 
278 	old_bucket = xchg(&smap->buckets[id], new_bucket);
279 	if (old_bucket)
280 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
281 	return id;
282 }
283 
284 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
285 	   u64, flags)
286 {
287 	u32 max_depth = map->value_size / stack_map_data_size(map);
288 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
289 	bool user = flags & BPF_F_USER_STACK;
290 	struct perf_callchain_entry *trace;
291 	bool kernel = !user;
292 
293 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
294 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
295 		return -EINVAL;
296 
297 	max_depth += skip;
298 	if (max_depth > sysctl_perf_event_max_stack)
299 		max_depth = sysctl_perf_event_max_stack;
300 
301 	trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
302 				   false, false);
303 
304 	if (unlikely(!trace))
305 		/* couldn't fetch the stack trace */
306 		return -EFAULT;
307 
308 	return __bpf_get_stackid(map, trace, flags);
309 }
310 
311 const struct bpf_func_proto bpf_get_stackid_proto = {
312 	.func		= bpf_get_stackid,
313 	.gpl_only	= true,
314 	.ret_type	= RET_INTEGER,
315 	.arg1_type	= ARG_PTR_TO_CTX,
316 	.arg2_type	= ARG_CONST_MAP_PTR,
317 	.arg3_type	= ARG_ANYTHING,
318 };
319 
320 static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
321 {
322 	__u64 nr_kernel = 0;
323 
324 	while (nr_kernel < trace->nr) {
325 		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
326 			break;
327 		nr_kernel++;
328 	}
329 	return nr_kernel;
330 }
331 
332 BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
333 	   struct bpf_map *, map, u64, flags)
334 {
335 	struct perf_event *event = ctx->event;
336 	struct perf_callchain_entry *trace;
337 	bool kernel, user;
338 	__u64 nr_kernel;
339 	int ret;
340 
341 	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
342 	if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
343 		return bpf_get_stackid((unsigned long)(ctx->regs),
344 				       (unsigned long) map, flags, 0, 0);
345 
346 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
347 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
348 		return -EINVAL;
349 
350 	user = flags & BPF_F_USER_STACK;
351 	kernel = !user;
352 
353 	trace = ctx->data->callchain;
354 	if (unlikely(!trace))
355 		return -EFAULT;
356 
357 	nr_kernel = count_kernel_ip(trace);
358 
359 	if (kernel) {
360 		__u64 nr = trace->nr;
361 
362 		trace->nr = nr_kernel;
363 		ret = __bpf_get_stackid(map, trace, flags);
364 
365 		/* restore nr */
366 		trace->nr = nr;
367 	} else { /* user */
368 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
369 
370 		skip += nr_kernel;
371 		if (skip > BPF_F_SKIP_FIELD_MASK)
372 			return -EFAULT;
373 
374 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
375 		ret = __bpf_get_stackid(map, trace, flags);
376 	}
377 	return ret;
378 }
379 
380 const struct bpf_func_proto bpf_get_stackid_proto_pe = {
381 	.func		= bpf_get_stackid_pe,
382 	.gpl_only	= false,
383 	.ret_type	= RET_INTEGER,
384 	.arg1_type	= ARG_PTR_TO_CTX,
385 	.arg2_type	= ARG_CONST_MAP_PTR,
386 	.arg3_type	= ARG_ANYTHING,
387 };
388 
389 static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
390 			    struct perf_callchain_entry *trace_in,
391 			    void *buf, u32 size, u64 flags)
392 {
393 	u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
394 	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
395 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
396 	bool user = flags & BPF_F_USER_STACK;
397 	struct perf_callchain_entry *trace;
398 	bool kernel = !user;
399 	int err = -EINVAL;
400 	u64 *ips;
401 
402 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
403 			       BPF_F_USER_BUILD_ID)))
404 		goto clear;
405 	if (kernel && user_build_id)
406 		goto clear;
407 
408 	elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id)
409 					    : sizeof(u64);
410 	if (unlikely(size % elem_size))
411 		goto clear;
412 
413 	/* cannot get valid user stack for task without user_mode regs */
414 	if (task && user && !user_mode(regs))
415 		goto err_fault;
416 
417 	num_elem = size / elem_size;
418 	max_depth = num_elem + skip;
419 	if (sysctl_perf_event_max_stack < max_depth)
420 		max_depth = sysctl_perf_event_max_stack;
421 
422 	if (trace_in)
423 		trace = trace_in;
424 	else if (kernel && task)
425 		trace = get_callchain_entry_for_task(task, max_depth);
426 	else
427 		trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
428 					   false, false);
429 	if (unlikely(!trace))
430 		goto err_fault;
431 
432 	if (trace->nr < skip)
433 		goto err_fault;
434 
435 	trace_nr = trace->nr - skip;
436 	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
437 	copy_len = trace_nr * elem_size;
438 
439 	ips = trace->ip + skip;
440 	if (user && user_build_id)
441 		stack_map_get_build_id_offset(buf, ips, trace_nr, user);
442 	else
443 		memcpy(buf, ips, copy_len);
444 
445 	if (size > copy_len)
446 		memset(buf + copy_len, 0, size - copy_len);
447 	return copy_len;
448 
449 err_fault:
450 	err = -EFAULT;
451 clear:
452 	memset(buf, 0, size);
453 	return err;
454 }
455 
456 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
457 	   u64, flags)
458 {
459 	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
460 }
461 
462 const struct bpf_func_proto bpf_get_stack_proto = {
463 	.func		= bpf_get_stack,
464 	.gpl_only	= true,
465 	.ret_type	= RET_INTEGER,
466 	.arg1_type	= ARG_PTR_TO_CTX,
467 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
468 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
469 	.arg4_type	= ARG_ANYTHING,
470 };
471 
472 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
473 	   u32, size, u64, flags)
474 {
475 	struct pt_regs *regs;
476 	long res = -EINVAL;
477 
478 	if (!try_get_task_stack(task))
479 		return -EFAULT;
480 
481 	regs = task_pt_regs(task);
482 	if (regs)
483 		res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
484 	put_task_stack(task);
485 
486 	return res;
487 }
488 
489 const struct bpf_func_proto bpf_get_task_stack_proto = {
490 	.func		= bpf_get_task_stack,
491 	.gpl_only	= false,
492 	.ret_type	= RET_INTEGER,
493 	.arg1_type	= ARG_PTR_TO_BTF_ID,
494 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
495 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
496 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
497 	.arg4_type	= ARG_ANYTHING,
498 };
499 
500 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
501 	   void *, buf, u32, size, u64, flags)
502 {
503 	struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
504 	struct perf_event *event = ctx->event;
505 	struct perf_callchain_entry *trace;
506 	bool kernel, user;
507 	int err = -EINVAL;
508 	__u64 nr_kernel;
509 
510 	if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
511 		return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
512 
513 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
514 			       BPF_F_USER_BUILD_ID)))
515 		goto clear;
516 
517 	user = flags & BPF_F_USER_STACK;
518 	kernel = !user;
519 
520 	err = -EFAULT;
521 	trace = ctx->data->callchain;
522 	if (unlikely(!trace))
523 		goto clear;
524 
525 	nr_kernel = count_kernel_ip(trace);
526 
527 	if (kernel) {
528 		__u64 nr = trace->nr;
529 
530 		trace->nr = nr_kernel;
531 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
532 
533 		/* restore nr */
534 		trace->nr = nr;
535 	} else { /* user */
536 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
537 
538 		skip += nr_kernel;
539 		if (skip > BPF_F_SKIP_FIELD_MASK)
540 			goto clear;
541 
542 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
543 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
544 	}
545 	return err;
546 
547 clear:
548 	memset(buf, 0, size);
549 	return err;
550 
551 }
552 
553 const struct bpf_func_proto bpf_get_stack_proto_pe = {
554 	.func		= bpf_get_stack_pe,
555 	.gpl_only	= true,
556 	.ret_type	= RET_INTEGER,
557 	.arg1_type	= ARG_PTR_TO_CTX,
558 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
559 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
560 	.arg4_type	= ARG_ANYTHING,
561 };
562 
563 /* Called from eBPF program */
564 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
565 {
566 	return ERR_PTR(-EOPNOTSUPP);
567 }
568 
569 /* Called from syscall */
570 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
571 {
572 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
573 	struct stack_map_bucket *bucket, *old_bucket;
574 	u32 id = *(u32 *)key, trace_len;
575 
576 	if (unlikely(id >= smap->n_buckets))
577 		return -ENOENT;
578 
579 	bucket = xchg(&smap->buckets[id], NULL);
580 	if (!bucket)
581 		return -ENOENT;
582 
583 	trace_len = bucket->nr * stack_map_data_size(map);
584 	memcpy(value, bucket->data, trace_len);
585 	memset(value + trace_len, 0, map->value_size - trace_len);
586 
587 	old_bucket = xchg(&smap->buckets[id], bucket);
588 	if (old_bucket)
589 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
590 	return 0;
591 }
592 
593 static int stack_map_get_next_key(struct bpf_map *map, void *key,
594 				  void *next_key)
595 {
596 	struct bpf_stack_map *smap = container_of(map,
597 						  struct bpf_stack_map, map);
598 	u32 id;
599 
600 	WARN_ON_ONCE(!rcu_read_lock_held());
601 
602 	if (!key) {
603 		id = 0;
604 	} else {
605 		id = *(u32 *)key;
606 		if (id >= smap->n_buckets || !smap->buckets[id])
607 			id = 0;
608 		else
609 			id++;
610 	}
611 
612 	while (id < smap->n_buckets && !smap->buckets[id])
613 		id++;
614 
615 	if (id >= smap->n_buckets)
616 		return -ENOENT;
617 
618 	*(u32 *)next_key = id;
619 	return 0;
620 }
621 
622 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
623 				 u64 map_flags)
624 {
625 	return -EINVAL;
626 }
627 
628 /* Called from syscall or from eBPF program */
629 static int stack_map_delete_elem(struct bpf_map *map, void *key)
630 {
631 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
632 	struct stack_map_bucket *old_bucket;
633 	u32 id = *(u32 *)key;
634 
635 	if (unlikely(id >= smap->n_buckets))
636 		return -E2BIG;
637 
638 	old_bucket = xchg(&smap->buckets[id], NULL);
639 	if (old_bucket) {
640 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
641 		return 0;
642 	} else {
643 		return -ENOENT;
644 	}
645 }
646 
647 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
648 static void stack_map_free(struct bpf_map *map)
649 {
650 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
651 
652 	bpf_map_area_free(smap->elems);
653 	pcpu_freelist_destroy(&smap->freelist);
654 	bpf_map_area_free(smap);
655 	put_callchain_buffers();
656 }
657 
658 static int stack_trace_map_btf_id;
659 const struct bpf_map_ops stack_trace_map_ops = {
660 	.map_meta_equal = bpf_map_meta_equal,
661 	.map_alloc = stack_map_alloc,
662 	.map_free = stack_map_free,
663 	.map_get_next_key = stack_map_get_next_key,
664 	.map_lookup_elem = stack_map_lookup_elem,
665 	.map_update_elem = stack_map_update_elem,
666 	.map_delete_elem = stack_map_delete_elem,
667 	.map_check_btf = map_check_no_btf,
668 	.map_btf_name = "bpf_stack_map",
669 	.map_btf_id = &stack_trace_map_btf_id,
670 };
671