xref: /linux-6.15/kernel/bpf/stackmap.c (revision 4f4c4fc0)
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(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 (attr->map_flags & ~STACK_CREATE_FLAG_MASK)
78 		return ERR_PTR(-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(-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(-EINVAL);
91 	} else if (value_size / 8 > sysctl_perf_event_max_stack)
92 		return ERR_PTR(-EINVAL);
93 
94 	/* hash table size must be power of 2; roundup_pow_of_two() can overflow
95 	 * into UB on 32-bit arches, so check that first
96 	 */
97 	if (attr->max_entries > 1UL << 31)
98 		return ERR_PTR(-E2BIG);
99 
100 	n_buckets = roundup_pow_of_two(attr->max_entries);
101 
102 	cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
103 	smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
104 	if (!smap)
105 		return ERR_PTR(-ENOMEM);
106 
107 	bpf_map_init_from_attr(&smap->map, attr);
108 	smap->n_buckets = n_buckets;
109 
110 	err = get_callchain_buffers(sysctl_perf_event_max_stack);
111 	if (err)
112 		goto free_smap;
113 
114 	err = prealloc_elems_and_freelist(smap);
115 	if (err)
116 		goto put_buffers;
117 
118 	return &smap->map;
119 
120 put_buffers:
121 	put_callchain_buffers();
122 free_smap:
123 	bpf_map_area_free(smap);
124 	return ERR_PTR(err);
125 }
126 
127 /*
128  * Expects all id_offs[i].ip values to be set to correct initial IPs.
129  * They will be subsequently:
130  *   - either adjusted in place to a file offset, if build ID fetching
131  *     succeeds; in this case id_offs[i].build_id is set to correct build ID,
132  *     and id_offs[i].status is set to BPF_STACK_BUILD_ID_VALID;
133  *   - or IP will be kept intact, if build ID fetching failed; in this case
134  *     id_offs[i].build_id is zeroed out and id_offs[i].status is set to
135  *     BPF_STACK_BUILD_ID_IP.
136  */
137 static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
138 					  u32 trace_nr, bool user)
139 {
140 	int i;
141 	struct mmap_unlock_irq_work *work = NULL;
142 	bool irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
143 	struct vm_area_struct *vma, *prev_vma = NULL;
144 	const char *prev_build_id;
145 
146 	/* If the irq_work is in use, fall back to report ips. Same
147 	 * fallback is used for kernel stack (!user) on a stackmap with
148 	 * build_id.
149 	 */
150 	if (!user || !current || !current->mm || irq_work_busy ||
151 	    !mmap_read_trylock(current->mm)) {
152 		/* cannot access current->mm, fall back to ips */
153 		for (i = 0; i < trace_nr; i++) {
154 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
155 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
156 		}
157 		return;
158 	}
159 
160 	for (i = 0; i < trace_nr; i++) {
161 		u64 ip = READ_ONCE(id_offs[i].ip);
162 
163 		if (range_in_vma(prev_vma, ip, ip)) {
164 			vma = prev_vma;
165 			memcpy(id_offs[i].build_id, prev_build_id, BUILD_ID_SIZE_MAX);
166 			goto build_id_valid;
167 		}
168 		vma = find_vma(current->mm, ip);
169 		if (!vma || build_id_parse_nofault(vma, id_offs[i].build_id, NULL)) {
170 			/* per entry fall back to ips */
171 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
172 			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
173 			continue;
174 		}
175 build_id_valid:
176 		id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ip - vma->vm_start;
177 		id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
178 		prev_vma = vma;
179 		prev_build_id = id_offs[i].build_id;
180 	}
181 	bpf_mmap_unlock_mm(work, current->mm);
182 }
183 
184 static struct perf_callchain_entry *
185 get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
186 {
187 #ifdef CONFIG_STACKTRACE
188 	struct perf_callchain_entry *entry;
189 	int rctx;
190 
191 	entry = get_callchain_entry(&rctx);
192 
193 	if (!entry)
194 		return NULL;
195 
196 	entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
197 					 max_depth, 0);
198 
199 	/* stack_trace_save_tsk() works on unsigned long array, while
200 	 * perf_callchain_entry uses u64 array. For 32-bit systems, it is
201 	 * necessary to fix this mismatch.
202 	 */
203 	if (__BITS_PER_LONG != 64) {
204 		unsigned long *from = (unsigned long *) entry->ip;
205 		u64 *to = entry->ip;
206 		int i;
207 
208 		/* copy data from the end to avoid using extra buffer */
209 		for (i = entry->nr - 1; i >= 0; i--)
210 			to[i] = (u64)(from[i]);
211 	}
212 
213 	put_callchain_entry(rctx);
214 
215 	return entry;
216 #else /* CONFIG_STACKTRACE */
217 	return NULL;
218 #endif
219 }
220 
221 static long __bpf_get_stackid(struct bpf_map *map,
222 			      struct perf_callchain_entry *trace, u64 flags)
223 {
224 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
225 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
226 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
227 	u32 hash, id, trace_nr, trace_len, i;
228 	bool user = flags & BPF_F_USER_STACK;
229 	u64 *ips;
230 	bool hash_matches;
231 
232 	if (trace->nr <= skip)
233 		/* skipping more than usable stack trace */
234 		return -EFAULT;
235 
236 	trace_nr = trace->nr - skip;
237 	trace_len = trace_nr * sizeof(u64);
238 	ips = trace->ip + skip;
239 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
240 	id = hash & (smap->n_buckets - 1);
241 	bucket = READ_ONCE(smap->buckets[id]);
242 
243 	hash_matches = bucket && bucket->hash == hash;
244 	/* fast cmp */
245 	if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
246 		return id;
247 
248 	if (stack_map_use_build_id(map)) {
249 		struct bpf_stack_build_id *id_offs;
250 
251 		/* for build_id+offset, pop a bucket before slow cmp */
252 		new_bucket = (struct stack_map_bucket *)
253 			pcpu_freelist_pop(&smap->freelist);
254 		if (unlikely(!new_bucket))
255 			return -ENOMEM;
256 		new_bucket->nr = trace_nr;
257 		id_offs = (struct bpf_stack_build_id *)new_bucket->data;
258 		for (i = 0; i < trace_nr; i++)
259 			id_offs[i].ip = ips[i];
260 		stack_map_get_build_id_offset(id_offs, 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 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
299 	bool user = flags & BPF_F_USER_STACK;
300 	struct perf_callchain_entry *trace;
301 	bool kernel = !user;
302 
303 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
304 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
305 		return -EINVAL;
306 
307 	max_depth += skip;
308 	if (max_depth > sysctl_perf_event_max_stack)
309 		max_depth = sysctl_perf_event_max_stack;
310 
311 	trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
312 				   false, false);
313 
314 	if (unlikely(!trace))
315 		/* couldn't fetch the stack trace */
316 		return -EFAULT;
317 
318 	return __bpf_get_stackid(map, trace, flags);
319 }
320 
321 const struct bpf_func_proto bpf_get_stackid_proto = {
322 	.func		= bpf_get_stackid,
323 	.gpl_only	= true,
324 	.ret_type	= RET_INTEGER,
325 	.arg1_type	= ARG_PTR_TO_CTX,
326 	.arg2_type	= ARG_CONST_MAP_PTR,
327 	.arg3_type	= ARG_ANYTHING,
328 };
329 
330 static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
331 {
332 	__u64 nr_kernel = 0;
333 
334 	while (nr_kernel < trace->nr) {
335 		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
336 			break;
337 		nr_kernel++;
338 	}
339 	return nr_kernel;
340 }
341 
342 BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
343 	   struct bpf_map *, map, u64, flags)
344 {
345 	struct perf_event *event = ctx->event;
346 	struct perf_callchain_entry *trace;
347 	bool kernel, user;
348 	__u64 nr_kernel;
349 	int ret;
350 
351 	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
352 	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
353 		return bpf_get_stackid((unsigned long)(ctx->regs),
354 				       (unsigned long) map, flags, 0, 0);
355 
356 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
357 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
358 		return -EINVAL;
359 
360 	user = flags & BPF_F_USER_STACK;
361 	kernel = !user;
362 
363 	trace = ctx->data->callchain;
364 	if (unlikely(!trace))
365 		return -EFAULT;
366 
367 	nr_kernel = count_kernel_ip(trace);
368 
369 	if (kernel) {
370 		__u64 nr = trace->nr;
371 
372 		trace->nr = nr_kernel;
373 		ret = __bpf_get_stackid(map, trace, flags);
374 
375 		/* restore nr */
376 		trace->nr = nr;
377 	} else { /* user */
378 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
379 
380 		skip += nr_kernel;
381 		if (skip > BPF_F_SKIP_FIELD_MASK)
382 			return -EFAULT;
383 
384 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
385 		ret = __bpf_get_stackid(map, trace, flags);
386 	}
387 	return ret;
388 }
389 
390 const struct bpf_func_proto bpf_get_stackid_proto_pe = {
391 	.func		= bpf_get_stackid_pe,
392 	.gpl_only	= false,
393 	.ret_type	= RET_INTEGER,
394 	.arg1_type	= ARG_PTR_TO_CTX,
395 	.arg2_type	= ARG_CONST_MAP_PTR,
396 	.arg3_type	= ARG_ANYTHING,
397 };
398 
399 static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
400 			    struct perf_callchain_entry *trace_in,
401 			    void *buf, u32 size, u64 flags)
402 {
403 	u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
404 	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
405 	bool crosstask = task && task != current;
406 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
407 	bool user = flags & BPF_F_USER_STACK;
408 	struct perf_callchain_entry *trace;
409 	bool kernel = !user;
410 	int err = -EINVAL;
411 	u64 *ips;
412 
413 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
414 			       BPF_F_USER_BUILD_ID)))
415 		goto clear;
416 	if (kernel && user_build_id)
417 		goto clear;
418 
419 	elem_size = (user && user_build_id) ? sizeof(struct bpf_stack_build_id)
420 					    : sizeof(u64);
421 	if (unlikely(size % elem_size))
422 		goto clear;
423 
424 	/* cannot get valid user stack for task without user_mode regs */
425 	if (task && user && !user_mode(regs))
426 		goto err_fault;
427 
428 	/* get_perf_callchain does not support crosstask user stack walking
429 	 * but returns an empty stack instead of NULL.
430 	 */
431 	if (crosstask && user) {
432 		err = -EOPNOTSUPP;
433 		goto clear;
434 	}
435 
436 	num_elem = size / elem_size;
437 	max_depth = num_elem + skip;
438 	if (sysctl_perf_event_max_stack < max_depth)
439 		max_depth = sysctl_perf_event_max_stack;
440 
441 	if (trace_in)
442 		trace = trace_in;
443 	else if (kernel && task)
444 		trace = get_callchain_entry_for_task(task, max_depth);
445 	else
446 		trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
447 					   crosstask, false);
448 	if (unlikely(!trace))
449 		goto err_fault;
450 
451 	if (trace->nr < skip)
452 		goto err_fault;
453 
454 	trace_nr = trace->nr - skip;
455 	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
456 	copy_len = trace_nr * elem_size;
457 
458 	ips = trace->ip + skip;
459 	if (user && user_build_id) {
460 		struct bpf_stack_build_id *id_offs = buf;
461 		u32 i;
462 
463 		for (i = 0; i < trace_nr; i++)
464 			id_offs[i].ip = ips[i];
465 		stack_map_get_build_id_offset(buf, trace_nr, user);
466 	} else {
467 		memcpy(buf, ips, copy_len);
468 	}
469 
470 	if (size > copy_len)
471 		memset(buf + copy_len, 0, size - copy_len);
472 	return copy_len;
473 
474 err_fault:
475 	err = -EFAULT;
476 clear:
477 	memset(buf, 0, size);
478 	return err;
479 }
480 
481 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
482 	   u64, flags)
483 {
484 	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
485 }
486 
487 const struct bpf_func_proto bpf_get_stack_proto = {
488 	.func		= bpf_get_stack,
489 	.gpl_only	= true,
490 	.ret_type	= RET_INTEGER,
491 	.arg1_type	= ARG_PTR_TO_CTX,
492 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
493 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
494 	.arg4_type	= ARG_ANYTHING,
495 };
496 
497 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
498 	   u32, size, u64, flags)
499 {
500 	struct pt_regs *regs;
501 	long res = -EINVAL;
502 
503 	if (!try_get_task_stack(task))
504 		return -EFAULT;
505 
506 	regs = task_pt_regs(task);
507 	if (regs)
508 		res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
509 	put_task_stack(task);
510 
511 	return res;
512 }
513 
514 const struct bpf_func_proto bpf_get_task_stack_proto = {
515 	.func		= bpf_get_task_stack,
516 	.gpl_only	= false,
517 	.ret_type	= RET_INTEGER,
518 	.arg1_type	= ARG_PTR_TO_BTF_ID,
519 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
520 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
521 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
522 	.arg4_type	= ARG_ANYTHING,
523 };
524 
525 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
526 	   void *, buf, u32, size, u64, flags)
527 {
528 	struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
529 	struct perf_event *event = ctx->event;
530 	struct perf_callchain_entry *trace;
531 	bool kernel, user;
532 	int err = -EINVAL;
533 	__u64 nr_kernel;
534 
535 	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
536 		return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
537 
538 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
539 			       BPF_F_USER_BUILD_ID)))
540 		goto clear;
541 
542 	user = flags & BPF_F_USER_STACK;
543 	kernel = !user;
544 
545 	err = -EFAULT;
546 	trace = ctx->data->callchain;
547 	if (unlikely(!trace))
548 		goto clear;
549 
550 	nr_kernel = count_kernel_ip(trace);
551 
552 	if (kernel) {
553 		__u64 nr = trace->nr;
554 
555 		trace->nr = nr_kernel;
556 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
557 
558 		/* restore nr */
559 		trace->nr = nr;
560 	} else { /* user */
561 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
562 
563 		skip += nr_kernel;
564 		if (skip > BPF_F_SKIP_FIELD_MASK)
565 			goto clear;
566 
567 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
568 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
569 	}
570 	return err;
571 
572 clear:
573 	memset(buf, 0, size);
574 	return err;
575 
576 }
577 
578 const struct bpf_func_proto bpf_get_stack_proto_pe = {
579 	.func		= bpf_get_stack_pe,
580 	.gpl_only	= true,
581 	.ret_type	= RET_INTEGER,
582 	.arg1_type	= ARG_PTR_TO_CTX,
583 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
584 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
585 	.arg4_type	= ARG_ANYTHING,
586 };
587 
588 /* Called from eBPF program */
589 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
590 {
591 	return ERR_PTR(-EOPNOTSUPP);
592 }
593 
594 /* Called from syscall */
595 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
596 {
597 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
598 	struct stack_map_bucket *bucket, *old_bucket;
599 	u32 id = *(u32 *)key, trace_len;
600 
601 	if (unlikely(id >= smap->n_buckets))
602 		return -ENOENT;
603 
604 	bucket = xchg(&smap->buckets[id], NULL);
605 	if (!bucket)
606 		return -ENOENT;
607 
608 	trace_len = bucket->nr * stack_map_data_size(map);
609 	memcpy(value, bucket->data, trace_len);
610 	memset(value + trace_len, 0, map->value_size - trace_len);
611 
612 	old_bucket = xchg(&smap->buckets[id], bucket);
613 	if (old_bucket)
614 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
615 	return 0;
616 }
617 
618 static int stack_map_get_next_key(struct bpf_map *map, void *key,
619 				  void *next_key)
620 {
621 	struct bpf_stack_map *smap = container_of(map,
622 						  struct bpf_stack_map, map);
623 	u32 id;
624 
625 	WARN_ON_ONCE(!rcu_read_lock_held());
626 
627 	if (!key) {
628 		id = 0;
629 	} else {
630 		id = *(u32 *)key;
631 		if (id >= smap->n_buckets || !smap->buckets[id])
632 			id = 0;
633 		else
634 			id++;
635 	}
636 
637 	while (id < smap->n_buckets && !smap->buckets[id])
638 		id++;
639 
640 	if (id >= smap->n_buckets)
641 		return -ENOENT;
642 
643 	*(u32 *)next_key = id;
644 	return 0;
645 }
646 
647 static long stack_map_update_elem(struct bpf_map *map, void *key, void *value,
648 				  u64 map_flags)
649 {
650 	return -EINVAL;
651 }
652 
653 /* Called from syscall or from eBPF program */
654 static long stack_map_delete_elem(struct bpf_map *map, void *key)
655 {
656 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
657 	struct stack_map_bucket *old_bucket;
658 	u32 id = *(u32 *)key;
659 
660 	if (unlikely(id >= smap->n_buckets))
661 		return -E2BIG;
662 
663 	old_bucket = xchg(&smap->buckets[id], NULL);
664 	if (old_bucket) {
665 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
666 		return 0;
667 	} else {
668 		return -ENOENT;
669 	}
670 }
671 
672 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
673 static void stack_map_free(struct bpf_map *map)
674 {
675 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
676 
677 	bpf_map_area_free(smap->elems);
678 	pcpu_freelist_destroy(&smap->freelist);
679 	bpf_map_area_free(smap);
680 	put_callchain_buffers();
681 }
682 
683 static u64 stack_map_mem_usage(const struct bpf_map *map)
684 {
685 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
686 	u64 value_size = map->value_size;
687 	u64 n_buckets = smap->n_buckets;
688 	u64 enties = map->max_entries;
689 	u64 usage = sizeof(*smap);
690 
691 	usage += n_buckets * sizeof(struct stack_map_bucket *);
692 	usage += enties * (sizeof(struct stack_map_bucket) + value_size);
693 	return usage;
694 }
695 
696 BTF_ID_LIST_SINGLE(stack_trace_map_btf_ids, struct, bpf_stack_map)
697 const struct bpf_map_ops stack_trace_map_ops = {
698 	.map_meta_equal = bpf_map_meta_equal,
699 	.map_alloc = stack_map_alloc,
700 	.map_free = stack_map_free,
701 	.map_get_next_key = stack_map_get_next_key,
702 	.map_lookup_elem = stack_map_lookup_elem,
703 	.map_update_elem = stack_map_update_elem,
704 	.map_delete_elem = stack_map_delete_elem,
705 	.map_check_btf = map_check_no_btf,
706 	.map_mem_usage = stack_map_mem_usage,
707 	.map_btf_id = &stack_trace_map_btf_ids[0],
708 };
709