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