xref: /openbmc/linux/kernel/bpf/stackmap.c (revision d82a6c5ef9dc0aab296936e1aa4ad28fd5162a55)
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->map.value_size = value_size;
110 	smap->n_buckets = n_buckets;
111 
112 	err = get_callchain_buffers(sysctl_perf_event_max_stack);
113 	if (err)
114 		goto free_smap;
115 
116 	err = prealloc_elems_and_freelist(smap);
117 	if (err)
118 		goto put_buffers;
119 
120 	return &smap->map;
121 
122 put_buffers:
123 	put_callchain_buffers();
124 free_smap:
125 	bpf_map_area_free(smap);
126 	return ERR_PTR(err);
127 }
128 
129 static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
130 					  u64 *ips, u32 trace_nr, bool user)
131 {
132 	int i;
133 	struct mmap_unlock_irq_work *work = NULL;
134 	bool irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
135 	struct vm_area_struct *vma, *prev_vma = NULL;
136 	const char *prev_build_id;
137 
138 	/* If the irq_work is in use, fall back to report ips. Same
139 	 * fallback is used for kernel stack (!user) on a stackmap with
140 	 * build_id.
141 	 */
142 	if (!user || !current || !current->mm || irq_work_busy ||
143 	    !mmap_read_trylock(current->mm)) {
144 		/* cannot access current->mm, fall back to ips */
145 		for (i = 0; i < trace_nr; i++) {
146 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
147 			id_offs[i].ip = ips[i];
148 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
149 		}
150 		return;
151 	}
152 
153 	for (i = 0; i < trace_nr; i++) {
154 		if (range_in_vma(prev_vma, ips[i], ips[i])) {
155 			vma = prev_vma;
156 			memcpy(id_offs[i].build_id, prev_build_id,
157 			       BUILD_ID_SIZE_MAX);
158 			goto build_id_valid;
159 		}
160 		vma = find_vma(current->mm, ips[i]);
161 		if (!vma || build_id_parse(vma, id_offs[i].build_id, NULL)) {
162 			/* per entry fall back to ips */
163 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
164 			id_offs[i].ip = ips[i];
165 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
166 			continue;
167 		}
168 build_id_valid:
169 		id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ips[i]
170 			- vma->vm_start;
171 		id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
172 		prev_vma = vma;
173 		prev_build_id = id_offs[i].build_id;
174 	}
175 	bpf_mmap_unlock_mm(work, current->mm);
176 }
177 
178 static struct perf_callchain_entry *
179 get_callchain_entry_for_task(struct task_struct *task, u32 init_nr)
180 {
181 #ifdef CONFIG_STACKTRACE
182 	struct perf_callchain_entry *entry;
183 	int rctx;
184 
185 	entry = get_callchain_entry(&rctx);
186 
187 	if (!entry)
188 		return NULL;
189 
190 	entry->nr = init_nr +
191 		stack_trace_save_tsk(task, (unsigned long *)(entry->ip + init_nr),
192 				     sysctl_perf_event_max_stack - init_nr, 0);
193 
194 	/* stack_trace_save_tsk() works on unsigned long array, while
195 	 * perf_callchain_entry uses u64 array. For 32-bit systems, it is
196 	 * necessary to fix this mismatch.
197 	 */
198 	if (__BITS_PER_LONG != 64) {
199 		unsigned long *from = (unsigned long *) entry->ip;
200 		u64 *to = entry->ip;
201 		int i;
202 
203 		/* copy data from the end to avoid using extra buffer */
204 		for (i = entry->nr - 1; i >= (int)init_nr; i--)
205 			to[i] = (u64)(from[i]);
206 	}
207 
208 	put_callchain_entry(rctx);
209 
210 	return entry;
211 #else /* CONFIG_STACKTRACE */
212 	return NULL;
213 #endif
214 }
215 
216 static long __bpf_get_stackid(struct bpf_map *map,
217 			      struct perf_callchain_entry *trace, u64 flags)
218 {
219 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
220 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
221 	u32 max_depth = map->value_size / stack_map_data_size(map);
222 	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
223 	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
224 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
225 	u32 hash, id, trace_nr, trace_len;
226 	bool user = flags & BPF_F_USER_STACK;
227 	u64 *ips;
228 	bool hash_matches;
229 
230 	/* get_perf_callchain() guarantees that trace->nr >= init_nr
231 	 * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
232 	 */
233 	trace_nr = trace->nr - init_nr;
234 
235 	if (trace_nr <= skip)
236 		/* skipping more than usable stack trace */
237 		return -EFAULT;
238 
239 	trace_nr -= skip;
240 	trace_len = trace_nr * sizeof(u64);
241 	ips = trace->ip + skip + init_nr;
242 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
243 	id = hash & (smap->n_buckets - 1);
244 	bucket = READ_ONCE(smap->buckets[id]);
245 
246 	hash_matches = bucket && bucket->hash == hash;
247 	/* fast cmp */
248 	if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
249 		return id;
250 
251 	if (stack_map_use_build_id(map)) {
252 		/* for build_id+offset, pop a bucket before slow cmp */
253 		new_bucket = (struct stack_map_bucket *)
254 			pcpu_freelist_pop(&smap->freelist);
255 		if (unlikely(!new_bucket))
256 			return -ENOMEM;
257 		new_bucket->nr = trace_nr;
258 		stack_map_get_build_id_offset(
259 			(struct bpf_stack_build_id *)new_bucket->data,
260 			ips, trace_nr, user);
261 		trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
262 		if (hash_matches && bucket->nr == trace_nr &&
263 		    memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
264 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
265 			return id;
266 		}
267 		if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
268 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
269 			return -EEXIST;
270 		}
271 	} else {
272 		if (hash_matches && bucket->nr == trace_nr &&
273 		    memcmp(bucket->data, ips, trace_len) == 0)
274 			return id;
275 		if (bucket && !(flags & BPF_F_REUSE_STACKID))
276 			return -EEXIST;
277 
278 		new_bucket = (struct stack_map_bucket *)
279 			pcpu_freelist_pop(&smap->freelist);
280 		if (unlikely(!new_bucket))
281 			return -ENOMEM;
282 		memcpy(new_bucket->data, ips, trace_len);
283 	}
284 
285 	new_bucket->hash = hash;
286 	new_bucket->nr = trace_nr;
287 
288 	old_bucket = xchg(&smap->buckets[id], new_bucket);
289 	if (old_bucket)
290 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
291 	return id;
292 }
293 
294 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
295 	   u64, flags)
296 {
297 	u32 max_depth = map->value_size / stack_map_data_size(map);
298 	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
299 	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
300 	bool user = flags & BPF_F_USER_STACK;
301 	struct perf_callchain_entry *trace;
302 	bool kernel = !user;
303 
304 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
305 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
306 		return -EINVAL;
307 
308 	trace = get_perf_callchain(regs, init_nr, kernel, user,
309 				   sysctl_perf_event_max_stack, false, false);
310 
311 	if (unlikely(!trace))
312 		/* couldn't fetch the stack trace */
313 		return -EFAULT;
314 
315 	return __bpf_get_stackid(map, trace, flags);
316 }
317 
318 const struct bpf_func_proto bpf_get_stackid_proto = {
319 	.func		= bpf_get_stackid,
320 	.gpl_only	= true,
321 	.ret_type	= RET_INTEGER,
322 	.arg1_type	= ARG_PTR_TO_CTX,
323 	.arg2_type	= ARG_CONST_MAP_PTR,
324 	.arg3_type	= ARG_ANYTHING,
325 };
326 
327 static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
328 {
329 	__u64 nr_kernel = 0;
330 
331 	while (nr_kernel < trace->nr) {
332 		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
333 			break;
334 		nr_kernel++;
335 	}
336 	return nr_kernel;
337 }
338 
339 BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
340 	   struct bpf_map *, map, u64, flags)
341 {
342 	struct perf_event *event = ctx->event;
343 	struct perf_callchain_entry *trace;
344 	bool kernel, user;
345 	__u64 nr_kernel;
346 	int ret;
347 
348 	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
349 	if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
350 		return bpf_get_stackid((unsigned long)(ctx->regs),
351 				       (unsigned long) map, flags, 0, 0);
352 
353 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
354 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
355 		return -EINVAL;
356 
357 	user = flags & BPF_F_USER_STACK;
358 	kernel = !user;
359 
360 	trace = ctx->data->callchain;
361 	if (unlikely(!trace))
362 		return -EFAULT;
363 
364 	nr_kernel = count_kernel_ip(trace);
365 
366 	if (kernel) {
367 		__u64 nr = trace->nr;
368 
369 		trace->nr = nr_kernel;
370 		ret = __bpf_get_stackid(map, trace, flags);
371 
372 		/* restore nr */
373 		trace->nr = nr;
374 	} else { /* user */
375 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
376 
377 		skip += nr_kernel;
378 		if (skip > BPF_F_SKIP_FIELD_MASK)
379 			return -EFAULT;
380 
381 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
382 		ret = __bpf_get_stackid(map, trace, flags);
383 	}
384 	return ret;
385 }
386 
387 const struct bpf_func_proto bpf_get_stackid_proto_pe = {
388 	.func		= bpf_get_stackid_pe,
389 	.gpl_only	= false,
390 	.ret_type	= RET_INTEGER,
391 	.arg1_type	= ARG_PTR_TO_CTX,
392 	.arg2_type	= ARG_CONST_MAP_PTR,
393 	.arg3_type	= ARG_ANYTHING,
394 };
395 
396 static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
397 			    struct perf_callchain_entry *trace_in,
398 			    void *buf, u32 size, u64 flags)
399 {
400 	u32 init_nr, trace_nr, copy_len, elem_size, num_elem;
401 	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
402 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
403 	bool user = flags & BPF_F_USER_STACK;
404 	struct perf_callchain_entry *trace;
405 	bool kernel = !user;
406 	int err = -EINVAL;
407 	u64 *ips;
408 
409 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
410 			       BPF_F_USER_BUILD_ID)))
411 		goto clear;
412 	if (kernel && user_build_id)
413 		goto clear;
414 
415 	elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id)
416 					    : sizeof(u64);
417 	if (unlikely(size % elem_size))
418 		goto clear;
419 
420 	/* cannot get valid user stack for task without user_mode regs */
421 	if (task && user && !user_mode(regs))
422 		goto err_fault;
423 
424 	num_elem = size / elem_size;
425 	if (sysctl_perf_event_max_stack < num_elem)
426 		init_nr = 0;
427 	else
428 		init_nr = sysctl_perf_event_max_stack - num_elem;
429 
430 	if (trace_in)
431 		trace = trace_in;
432 	else if (kernel && task)
433 		trace = get_callchain_entry_for_task(task, init_nr);
434 	else
435 		trace = get_perf_callchain(regs, init_nr, kernel, user,
436 					   sysctl_perf_event_max_stack,
437 					   false, false);
438 	if (unlikely(!trace))
439 		goto err_fault;
440 
441 	trace_nr = trace->nr - init_nr;
442 	if (trace_nr < skip)
443 		goto err_fault;
444 
445 	trace_nr -= skip;
446 	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
447 	copy_len = trace_nr * elem_size;
448 	ips = trace->ip + skip + init_nr;
449 	if (user && user_build_id)
450 		stack_map_get_build_id_offset(buf, ips, trace_nr, user);
451 	else
452 		memcpy(buf, ips, copy_len);
453 
454 	if (size > copy_len)
455 		memset(buf + copy_len, 0, size - copy_len);
456 	return copy_len;
457 
458 err_fault:
459 	err = -EFAULT;
460 clear:
461 	memset(buf, 0, size);
462 	return err;
463 }
464 
465 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
466 	   u64, flags)
467 {
468 	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
469 }
470 
471 const struct bpf_func_proto bpf_get_stack_proto = {
472 	.func		= bpf_get_stack,
473 	.gpl_only	= true,
474 	.ret_type	= RET_INTEGER,
475 	.arg1_type	= ARG_PTR_TO_CTX,
476 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
477 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
478 	.arg4_type	= ARG_ANYTHING,
479 };
480 
481 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
482 	   u32, size, u64, flags)
483 {
484 	struct pt_regs *regs;
485 	long res = -EINVAL;
486 
487 	if (!try_get_task_stack(task))
488 		return -EFAULT;
489 
490 	regs = task_pt_regs(task);
491 	if (regs)
492 		res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
493 	put_task_stack(task);
494 
495 	return res;
496 }
497 
498 const struct bpf_func_proto bpf_get_task_stack_proto = {
499 	.func		= bpf_get_task_stack,
500 	.gpl_only	= false,
501 	.ret_type	= RET_INTEGER,
502 	.arg1_type	= ARG_PTR_TO_BTF_ID,
503 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
504 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
505 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
506 	.arg4_type	= ARG_ANYTHING,
507 };
508 
509 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
510 	   void *, buf, u32, size, u64, flags)
511 {
512 	struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
513 	struct perf_event *event = ctx->event;
514 	struct perf_callchain_entry *trace;
515 	bool kernel, user;
516 	int err = -EINVAL;
517 	__u64 nr_kernel;
518 
519 	if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY))
520 		return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
521 
522 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
523 			       BPF_F_USER_BUILD_ID)))
524 		goto clear;
525 
526 	user = flags & BPF_F_USER_STACK;
527 	kernel = !user;
528 
529 	err = -EFAULT;
530 	trace = ctx->data->callchain;
531 	if (unlikely(!trace))
532 		goto clear;
533 
534 	nr_kernel = count_kernel_ip(trace);
535 
536 	if (kernel) {
537 		__u64 nr = trace->nr;
538 
539 		trace->nr = nr_kernel;
540 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
541 
542 		/* restore nr */
543 		trace->nr = nr;
544 	} else { /* user */
545 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
546 
547 		skip += nr_kernel;
548 		if (skip > BPF_F_SKIP_FIELD_MASK)
549 			goto clear;
550 
551 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
552 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
553 	}
554 	return err;
555 
556 clear:
557 	memset(buf, 0, size);
558 	return err;
559 
560 }
561 
562 const struct bpf_func_proto bpf_get_stack_proto_pe = {
563 	.func		= bpf_get_stack_pe,
564 	.gpl_only	= true,
565 	.ret_type	= RET_INTEGER,
566 	.arg1_type	= ARG_PTR_TO_CTX,
567 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
568 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
569 	.arg4_type	= ARG_ANYTHING,
570 };
571 
572 /* Called from eBPF program */
573 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
574 {
575 	return ERR_PTR(-EOPNOTSUPP);
576 }
577 
578 /* Called from syscall */
579 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
580 {
581 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
582 	struct stack_map_bucket *bucket, *old_bucket;
583 	u32 id = *(u32 *)key, trace_len;
584 
585 	if (unlikely(id >= smap->n_buckets))
586 		return -ENOENT;
587 
588 	bucket = xchg(&smap->buckets[id], NULL);
589 	if (!bucket)
590 		return -ENOENT;
591 
592 	trace_len = bucket->nr * stack_map_data_size(map);
593 	memcpy(value, bucket->data, trace_len);
594 	memset(value + trace_len, 0, map->value_size - trace_len);
595 
596 	old_bucket = xchg(&smap->buckets[id], bucket);
597 	if (old_bucket)
598 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
599 	return 0;
600 }
601 
602 static int stack_map_get_next_key(struct bpf_map *map, void *key,
603 				  void *next_key)
604 {
605 	struct bpf_stack_map *smap = container_of(map,
606 						  struct bpf_stack_map, map);
607 	u32 id;
608 
609 	WARN_ON_ONCE(!rcu_read_lock_held());
610 
611 	if (!key) {
612 		id = 0;
613 	} else {
614 		id = *(u32 *)key;
615 		if (id >= smap->n_buckets || !smap->buckets[id])
616 			id = 0;
617 		else
618 			id++;
619 	}
620 
621 	while (id < smap->n_buckets && !smap->buckets[id])
622 		id++;
623 
624 	if (id >= smap->n_buckets)
625 		return -ENOENT;
626 
627 	*(u32 *)next_key = id;
628 	return 0;
629 }
630 
631 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
632 				 u64 map_flags)
633 {
634 	return -EINVAL;
635 }
636 
637 /* Called from syscall or from eBPF program */
638 static int stack_map_delete_elem(struct bpf_map *map, void *key)
639 {
640 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
641 	struct stack_map_bucket *old_bucket;
642 	u32 id = *(u32 *)key;
643 
644 	if (unlikely(id >= smap->n_buckets))
645 		return -E2BIG;
646 
647 	old_bucket = xchg(&smap->buckets[id], NULL);
648 	if (old_bucket) {
649 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
650 		return 0;
651 	} else {
652 		return -ENOENT;
653 	}
654 }
655 
656 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
657 static void stack_map_free(struct bpf_map *map)
658 {
659 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
660 
661 	bpf_map_area_free(smap->elems);
662 	pcpu_freelist_destroy(&smap->freelist);
663 	bpf_map_area_free(smap);
664 	put_callchain_buffers();
665 }
666 
667 static int stack_trace_map_btf_id;
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_btf_name = "bpf_stack_map",
678 	.map_btf_id = &stack_trace_map_btf_id,
679 };
680