xref: /openbmc/linux/kernel/bpf/stackmap.c (revision 7070b274)
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 (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 
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))
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 	bool crosstask = task && task != current;
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 	/* get_perf_callchain does not support crosstask user stack walking
418 	 * but returns an empty stack instead of NULL.
419 	 */
420 	if (crosstask && user) {
421 		err = -EOPNOTSUPP;
422 		goto clear;
423 	}
424 
425 	num_elem = size / elem_size;
426 	max_depth = num_elem + skip;
427 	if (sysctl_perf_event_max_stack < max_depth)
428 		max_depth = sysctl_perf_event_max_stack;
429 
430 	if (trace_in)
431 		trace = trace_in;
432 	else if (kernel && task)
433 		trace = get_callchain_entry_for_task(task, max_depth);
434 	else
435 		trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
436 					   crosstask, false);
437 	if (unlikely(!trace))
438 		goto err_fault;
439 
440 	if (trace->nr < skip)
441 		goto err_fault;
442 
443 	trace_nr = trace->nr - skip;
444 	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
445 	copy_len = trace_nr * elem_size;
446 
447 	ips = trace->ip + skip;
448 	if (user && user_build_id)
449 		stack_map_get_build_id_offset(buf, ips, trace_nr, user);
450 	else
451 		memcpy(buf, ips, copy_len);
452 
453 	if (size > copy_len)
454 		memset(buf + copy_len, 0, size - copy_len);
455 	return copy_len;
456 
457 err_fault:
458 	err = -EFAULT;
459 clear:
460 	memset(buf, 0, size);
461 	return err;
462 }
463 
BPF_CALL_4(bpf_get_stack,struct pt_regs *,regs,void *,buf,u32,size,u64,flags)464 BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
465 	   u64, flags)
466 {
467 	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
468 }
469 
470 const struct bpf_func_proto bpf_get_stack_proto = {
471 	.func		= bpf_get_stack,
472 	.gpl_only	= true,
473 	.ret_type	= RET_INTEGER,
474 	.arg1_type	= ARG_PTR_TO_CTX,
475 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
476 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
477 	.arg4_type	= ARG_ANYTHING,
478 };
479 
BPF_CALL_4(bpf_get_task_stack,struct task_struct *,task,void *,buf,u32,size,u64,flags)480 BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
481 	   u32, size, u64, flags)
482 {
483 	struct pt_regs *regs;
484 	long res = -EINVAL;
485 
486 	if (!try_get_task_stack(task))
487 		return -EFAULT;
488 
489 	regs = task_pt_regs(task);
490 	if (regs)
491 		res = __bpf_get_stack(regs, task, NULL, buf, size, flags);
492 	put_task_stack(task);
493 
494 	return res;
495 }
496 
497 const struct bpf_func_proto bpf_get_task_stack_proto = {
498 	.func		= bpf_get_task_stack,
499 	.gpl_only	= false,
500 	.ret_type	= RET_INTEGER,
501 	.arg1_type	= ARG_PTR_TO_BTF_ID,
502 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
503 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
504 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
505 	.arg4_type	= ARG_ANYTHING,
506 };
507 
BPF_CALL_4(bpf_get_stack_pe,struct bpf_perf_event_data_kern *,ctx,void *,buf,u32,size,u64,flags)508 BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
509 	   void *, buf, u32, size, u64, flags)
510 {
511 	struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
512 	struct perf_event *event = ctx->event;
513 	struct perf_callchain_entry *trace;
514 	bool kernel, user;
515 	int err = -EINVAL;
516 	__u64 nr_kernel;
517 
518 	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
519 		return __bpf_get_stack(regs, NULL, NULL, buf, size, flags);
520 
521 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
522 			       BPF_F_USER_BUILD_ID)))
523 		goto clear;
524 
525 	user = flags & BPF_F_USER_STACK;
526 	kernel = !user;
527 
528 	err = -EFAULT;
529 	trace = ctx->data->callchain;
530 	if (unlikely(!trace))
531 		goto clear;
532 
533 	nr_kernel = count_kernel_ip(trace);
534 
535 	if (kernel) {
536 		__u64 nr = trace->nr;
537 
538 		trace->nr = nr_kernel;
539 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
540 
541 		/* restore nr */
542 		trace->nr = nr;
543 	} else { /* user */
544 		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
545 
546 		skip += nr_kernel;
547 		if (skip > BPF_F_SKIP_FIELD_MASK)
548 			goto clear;
549 
550 		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
551 		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags);
552 	}
553 	return err;
554 
555 clear:
556 	memset(buf, 0, size);
557 	return err;
558 
559 }
560 
561 const struct bpf_func_proto bpf_get_stack_proto_pe = {
562 	.func		= bpf_get_stack_pe,
563 	.gpl_only	= true,
564 	.ret_type	= RET_INTEGER,
565 	.arg1_type	= ARG_PTR_TO_CTX,
566 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
567 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
568 	.arg4_type	= ARG_ANYTHING,
569 };
570 
571 /* Called from eBPF program */
stack_map_lookup_elem(struct bpf_map * map,void * key)572 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
573 {
574 	return ERR_PTR(-EOPNOTSUPP);
575 }
576 
577 /* Called from syscall */
bpf_stackmap_copy(struct bpf_map * map,void * key,void * value)578 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
579 {
580 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
581 	struct stack_map_bucket *bucket, *old_bucket;
582 	u32 id = *(u32 *)key, trace_len;
583 
584 	if (unlikely(id >= smap->n_buckets))
585 		return -ENOENT;
586 
587 	bucket = xchg(&smap->buckets[id], NULL);
588 	if (!bucket)
589 		return -ENOENT;
590 
591 	trace_len = bucket->nr * stack_map_data_size(map);
592 	memcpy(value, bucket->data, trace_len);
593 	memset(value + trace_len, 0, map->value_size - trace_len);
594 
595 	old_bucket = xchg(&smap->buckets[id], bucket);
596 	if (old_bucket)
597 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
598 	return 0;
599 }
600 
stack_map_get_next_key(struct bpf_map * map,void * key,void * next_key)601 static int stack_map_get_next_key(struct bpf_map *map, void *key,
602 				  void *next_key)
603 {
604 	struct bpf_stack_map *smap = container_of(map,
605 						  struct bpf_stack_map, map);
606 	u32 id;
607 
608 	WARN_ON_ONCE(!rcu_read_lock_held());
609 
610 	if (!key) {
611 		id = 0;
612 	} else {
613 		id = *(u32 *)key;
614 		if (id >= smap->n_buckets || !smap->buckets[id])
615 			id = 0;
616 		else
617 			id++;
618 	}
619 
620 	while (id < smap->n_buckets && !smap->buckets[id])
621 		id++;
622 
623 	if (id >= smap->n_buckets)
624 		return -ENOENT;
625 
626 	*(u32 *)next_key = id;
627 	return 0;
628 }
629 
stack_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)630 static long stack_map_update_elem(struct bpf_map *map, void *key, void *value,
631 				  u64 map_flags)
632 {
633 	return -EINVAL;
634 }
635 
636 /* Called from syscall or from eBPF program */
stack_map_delete_elem(struct bpf_map * map,void * key)637 static long stack_map_delete_elem(struct bpf_map *map, void *key)
638 {
639 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
640 	struct stack_map_bucket *old_bucket;
641 	u32 id = *(u32 *)key;
642 
643 	if (unlikely(id >= smap->n_buckets))
644 		return -E2BIG;
645 
646 	old_bucket = xchg(&smap->buckets[id], NULL);
647 	if (old_bucket) {
648 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
649 		return 0;
650 	} else {
651 		return -ENOENT;
652 	}
653 }
654 
655 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
stack_map_free(struct bpf_map * map)656 static void stack_map_free(struct bpf_map *map)
657 {
658 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
659 
660 	bpf_map_area_free(smap->elems);
661 	pcpu_freelist_destroy(&smap->freelist);
662 	bpf_map_area_free(smap);
663 	put_callchain_buffers();
664 }
665 
stack_map_mem_usage(const struct bpf_map * map)666 static u64 stack_map_mem_usage(const struct bpf_map *map)
667 {
668 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
669 	u64 value_size = map->value_size;
670 	u64 n_buckets = smap->n_buckets;
671 	u64 enties = map->max_entries;
672 	u64 usage = sizeof(*smap);
673 
674 	usage += n_buckets * sizeof(struct stack_map_bucket *);
675 	usage += enties * (sizeof(struct stack_map_bucket) + value_size);
676 	return usage;
677 }
678 
679 BTF_ID_LIST_SINGLE(stack_trace_map_btf_ids, struct, bpf_stack_map)
680 const struct bpf_map_ops stack_trace_map_ops = {
681 	.map_meta_equal = bpf_map_meta_equal,
682 	.map_alloc = stack_map_alloc,
683 	.map_free = stack_map_free,
684 	.map_get_next_key = stack_map_get_next_key,
685 	.map_lookup_elem = stack_map_lookup_elem,
686 	.map_update_elem = stack_map_update_elem,
687 	.map_delete_elem = stack_map_delete_elem,
688 	.map_check_btf = map_check_no_btf,
689 	.map_mem_usage = stack_map_mem_usage,
690 	.map_btf_id = &stack_trace_map_btf_ids[0],
691 };
692