stackmap.c

  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
 19struct stack_map_bucket {
 20	struct pcpu_freelist_node fnode;
 21	u32 hash;
 22	u32 nr;
 23	u64 data[];
 24};
 25
 26struct 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
 34static inline bool stack_map_use_build_id(struct bpf_map *map)
 35{
 36	return (map->map_flags & BPF_F_STACK_BUILD_ID);
 37}
 38
 39static 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/**
 46 * stack_map_calculate_max_depth - Calculate maximum allowed stack trace depth
 47 * @size:  Size of the buffer/map value in bytes
 48 * @elem_size:  Size of each stack trace element
 49 * @flags:  BPF stack trace flags (BPF_F_USER_STACK, BPF_F_USER_BUILD_ID, ...)
 50 *
 51 * Return: Maximum number of stack trace entries that can be safely stored
 52 */
 53static u32 stack_map_calculate_max_depth(u32 size, u32 elem_size, u64 flags)
 54{
 55	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 56	u32 max_depth;
 57	u32 curr_sysctl_max_stack = READ_ONCE(sysctl_perf_event_max_stack);
 58
 59	max_depth = size / elem_size;
 60	max_depth += skip;
 61	if (max_depth > curr_sysctl_max_stack)
 62		return curr_sysctl_max_stack;
 63
 64	return max_depth;
 65}
 66
 67static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
 68{
 69	u64 elem_size = sizeof(struct stack_map_bucket) +
 70			(u64)smap->map.value_size;
 71	int err;
 72
 73	smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries,
 74					 smap->map.numa_node);
 75	if (!smap->elems)
 76		return -ENOMEM;
 77
 78	err = pcpu_freelist_init(&smap->freelist);
 79	if (err)
 80		goto free_elems;
 81
 82	pcpu_freelist_populate(&smap->freelist, smap->elems, elem_size,
 83			       smap->map.max_entries);
 84	return 0;
 85
 86free_elems:
 87	bpf_map_area_free(smap->elems);
 88	return err;
 89}
 90
 91/* Called from syscall */
 92static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
 93{
 94	u32 value_size = attr->value_size;
 95	struct bpf_stack_map *smap;
 96	u64 cost, n_buckets;
 97	int err;
 98
 99	if (attr->map_flags & ~STACK_CREATE_FLAG_MASK)
100		return ERR_PTR(-EINVAL);
101
102	/* check sanity of attributes */
103	if (attr->max_entries == 0 || attr->key_size != 4 ||
104	    value_size < 8 || value_size % 8)
105		return ERR_PTR(-EINVAL);
106
107	BUILD_BUG_ON(sizeof(struct bpf_stack_build_id) % sizeof(u64));
108	if (attr->map_flags & BPF_F_STACK_BUILD_ID) {
109		if (value_size % sizeof(struct bpf_stack_build_id) ||
110		    value_size / sizeof(struct bpf_stack_build_id)
111		    > sysctl_perf_event_max_stack)
112			return ERR_PTR(-EINVAL);
113	} else if (value_size / 8 > sysctl_perf_event_max_stack)
114		return ERR_PTR(-EINVAL);
115
116	/* hash table size must be power of 2; roundup_pow_of_two() can overflow
117	 * into UB on 32-bit arches, so check that first
118	 */
119	if (attr->max_entries > 1UL << 31)
120		return ERR_PTR(-E2BIG);
121
122	n_buckets = roundup_pow_of_two(attr->max_entries);
123
124	cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
125	smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
126	if (!smap)
127		return ERR_PTR(-ENOMEM);
128
129	bpf_map_init_from_attr(&smap->map, attr);
130	smap->n_buckets = n_buckets;
131
132	err = get_callchain_buffers(sysctl_perf_event_max_stack);
133	if (err)
134		goto free_smap;
135
136	err = prealloc_elems_and_freelist(smap);
137	if (err)
138		goto put_buffers;
139
140	return &smap->map;
141
142put_buffers:
143	put_callchain_buffers();
144free_smap:
145	bpf_map_area_free(smap);
146	return ERR_PTR(err);
147}
148
149static int fetch_build_id(struct vm_area_struct *vma, unsigned char *build_id, bool may_fault)
150{
151	return may_fault ? build_id_parse(vma, build_id, NULL)
152			 : build_id_parse_nofault(vma, build_id, NULL);
153}
154
155/*
156 * Expects all id_offs[i].ip values to be set to correct initial IPs.
157 * They will be subsequently:
158 *   - either adjusted in place to a file offset, if build ID fetching
159 *     succeeds; in this case id_offs[i].build_id is set to correct build ID,
160 *     and id_offs[i].status is set to BPF_STACK_BUILD_ID_VALID;
161 *   - or IP will be kept intact, if build ID fetching failed; in this case
162 *     id_offs[i].build_id is zeroed out and id_offs[i].status is set to
163 *     BPF_STACK_BUILD_ID_IP.
164 */
165static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
166					  u32 trace_nr, bool user, bool may_fault)
167{
168	int i;
169	struct mmap_unlock_irq_work *work = NULL;
170	bool irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
171	struct vm_area_struct *vma, *prev_vma = NULL;
172	const char *prev_build_id;
173
174	/* If the irq_work is in use, fall back to report ips. Same
175	 * fallback is used for kernel stack (!user) on a stackmap with
176	 * build_id.
177	 */
178	if (!user || !current || !current->mm || irq_work_busy ||
179	    !mmap_read_trylock(current->mm)) {
180		/* cannot access current->mm, fall back to ips */
181		for (i = 0; i < trace_nr; i++) {
182			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
183			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
184		}
185		return;
186	}
187
188	for (i = 0; i < trace_nr; i++) {
189		u64 ip = READ_ONCE(id_offs[i].ip);
190
191		if (range_in_vma(prev_vma, ip, ip)) {
192			vma = prev_vma;
193			memcpy(id_offs[i].build_id, prev_build_id, BUILD_ID_SIZE_MAX);
194			goto build_id_valid;
195		}
196		vma = find_vma(current->mm, ip);
197		if (!vma || fetch_build_id(vma, id_offs[i].build_id, may_fault)) {
198			/* per entry fall back to ips */
199			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
200			memset(id_offs[i].build_id, 0, BUILD_ID_SIZE_MAX);
201			continue;
202		}
203build_id_valid:
204		id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ip - vma->vm_start;
205		id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
206		prev_vma = vma;
207		prev_build_id = id_offs[i].build_id;
208	}
209	bpf_mmap_unlock_mm(work, current->mm);
210}
211
212static struct perf_callchain_entry *
213get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
214{
215#ifdef CONFIG_STACKTRACE
216	struct perf_callchain_entry *entry;
217	int rctx;
218
219	entry = get_callchain_entry(&rctx);
220
221	if (!entry)
222		return NULL;
223
224	entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
225					 max_depth, 0);
226
227	/* stack_trace_save_tsk() works on unsigned long array, while
228	 * perf_callchain_entry uses u64 array. For 32-bit systems, it is
229	 * necessary to fix this mismatch.
230	 */
231	if (__BITS_PER_LONG != 64) {
232		unsigned long *from = (unsigned long *) entry->ip;
233		u64 *to = entry->ip;
234		int i;
235
236		/* copy data from the end to avoid using extra buffer */
237		for (i = entry->nr - 1; i >= 0; i--)
238			to[i] = (u64)(from[i]);
239	}
240
241	put_callchain_entry(rctx);
242
243	return entry;
244#else /* CONFIG_STACKTRACE */
245	return NULL;
246#endif
247}
248
249static long __bpf_get_stackid(struct bpf_map *map,
250			      struct perf_callchain_entry *trace, u64 flags)
251{
252	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
253	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
254	u32 hash, id, trace_nr, trace_len, i, max_depth;
255	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
256	bool user = flags & BPF_F_USER_STACK;
257	u64 *ips;
258	bool hash_matches;
259
260	if (trace->nr <= skip)
261		/* skipping more than usable stack trace */
262		return -EFAULT;
263
264	max_depth = stack_map_calculate_max_depth(map->value_size, stack_map_data_size(map), flags);
265	trace_nr = min_t(u32, trace->nr - skip, max_depth - skip);
266	trace_len = trace_nr * sizeof(u64);
267	ips = trace->ip + skip;
268	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
269	id = hash & (smap->n_buckets - 1);
270	bucket = READ_ONCE(smap->buckets[id]);
271
272	hash_matches = bucket && bucket->hash == hash;
273	/* fast cmp */
274	if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
275		return id;
276
277	if (stack_map_use_build_id(map)) {
278		struct bpf_stack_build_id *id_offs;
279
280		/* for build_id+offset, pop a bucket before slow cmp */
281		new_bucket = (struct stack_map_bucket *)
282			pcpu_freelist_pop(&smap->freelist);
283		if (unlikely(!new_bucket))
284			return -ENOMEM;
285		new_bucket->nr = trace_nr;
286		id_offs = (struct bpf_stack_build_id *)new_bucket->data;
287		for (i = 0; i < trace_nr; i++)
288			id_offs[i].ip = ips[i];
289		stack_map_get_build_id_offset(id_offs, trace_nr, user, false /* !may_fault */);
290		trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
291		if (hash_matches && bucket->nr == trace_nr &&
292		    memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
293			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
294			return id;
295		}
296		if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
297			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
298			return -EEXIST;
299		}
300	} else {
301		if (hash_matches && bucket->nr == trace_nr &&
302		    memcmp(bucket->data, ips, trace_len) == 0)
303			return id;
304		if (bucket && !(flags & BPF_F_REUSE_STACKID))
305			return -EEXIST;
306
307		new_bucket = (struct stack_map_bucket *)
308			pcpu_freelist_pop(&smap->freelist);
309		if (unlikely(!new_bucket))
310			return -ENOMEM;
311		memcpy(new_bucket->data, ips, trace_len);
312	}
313
314	new_bucket->hash = hash;
315	new_bucket->nr = trace_nr;
316
317	old_bucket = xchg(&smap->buckets[id], new_bucket);
318	if (old_bucket)
319		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
320	return id;
321}
322
323BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
324	   u64, flags)
325{
326	u32 elem_size = stack_map_data_size(map);
327	bool user = flags & BPF_F_USER_STACK;
328	struct perf_callchain_entry *trace;
329	bool kernel = !user;
330	u32 max_depth;
331
332	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
333			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
334		return -EINVAL;
335
336	max_depth = stack_map_calculate_max_depth(map->value_size, elem_size, flags);
337	trace = get_perf_callchain(regs, kernel, user, max_depth,
338				   false, false, 0);
339
340	if (unlikely(!trace))
341		/* couldn't fetch the stack trace */
342		return -EFAULT;
343
344	return __bpf_get_stackid(map, trace, flags);
345}
346
347const struct bpf_func_proto bpf_get_stackid_proto = {
348	.func		= bpf_get_stackid,
349	.gpl_only	= true,
350	.ret_type	= RET_INTEGER,
351	.arg1_type	= ARG_PTR_TO_CTX,
352	.arg2_type	= ARG_CONST_MAP_PTR,
353	.arg3_type	= ARG_ANYTHING,
354};
355
356static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
357{
358	__u64 nr_kernel = 0;
359
360	while (nr_kernel < trace->nr) {
361		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
362			break;
363		nr_kernel++;
364	}
365	return nr_kernel;
366}
367
368BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
369	   struct bpf_map *, map, u64, flags)
370{
371	struct perf_event *event = ctx->event;
372	struct perf_callchain_entry *trace;
373	bool kernel, user;
374	__u64 nr_kernel;
375	int ret;
376
377	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
378	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
379		return bpf_get_stackid((unsigned long)(ctx->regs),
380				       (unsigned long) map, flags, 0, 0);
381
382	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
383			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
384		return -EINVAL;
385
386	user = flags & BPF_F_USER_STACK;
387	kernel = !user;
388
389	trace = ctx->data->callchain;
390	if (unlikely(!trace))
391		return -EFAULT;
392
393	nr_kernel = count_kernel_ip(trace);
394	__u64 nr = trace->nr; /* save original */
395
396	if (kernel) {
397		trace->nr = nr_kernel;
398		ret = __bpf_get_stackid(map, trace, flags);
399	} else { /* user */
400		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
401
402		skip += nr_kernel;
403		if (skip > BPF_F_SKIP_FIELD_MASK)
404			return -EFAULT;
405
406		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
407		ret = __bpf_get_stackid(map, trace, flags);
408	}
409
410	/* restore nr */
411	trace->nr = nr;
412
413	return ret;
414}
415
416const struct bpf_func_proto bpf_get_stackid_proto_pe = {
417	.func		= bpf_get_stackid_pe,
418	.gpl_only	= false,
419	.ret_type	= RET_INTEGER,
420	.arg1_type	= ARG_PTR_TO_CTX,
421	.arg2_type	= ARG_CONST_MAP_PTR,
422	.arg3_type	= ARG_ANYTHING,
423};
424
425static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
426			    struct perf_callchain_entry *trace_in,
427			    void *buf, u32 size, u64 flags, bool may_fault)
428{
429	u32 trace_nr, copy_len, elem_size, max_depth;
430	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
431	bool crosstask = task && task != current;
432	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
433	bool user = flags & BPF_F_USER_STACK;
434	struct perf_callchain_entry *trace;
435	bool kernel = !user;
436	int err = -EINVAL;
437	u64 *ips;
438
439	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
440			       BPF_F_USER_BUILD_ID)))
441		goto clear;
442	if (kernel && user_build_id)
443		goto clear;
444
445	elem_size = user_build_id ? sizeof(struct bpf_stack_build_id) : sizeof(u64);
446	if (unlikely(size % elem_size))
447		goto clear;
448
449	/* cannot get valid user stack for task without user_mode regs */
450	if (task && user && !user_mode(regs))
451		goto err_fault;
452
453	/* get_perf_callchain does not support crosstask user stack walking
454	 * but returns an empty stack instead of NULL.
455	 */
456	if (crosstask && user) {
457		err = -EOPNOTSUPP;
458		goto clear;
459	}
460
461	max_depth = stack_map_calculate_max_depth(size, elem_size, flags);
462
463	if (may_fault)
464		rcu_read_lock(); /* need RCU for perf's callchain below */
465
466	if (trace_in) {
467		trace = trace_in;
468		trace->nr = min_t(u32, trace->nr, max_depth);
469	} else if (kernel && task) {
470		trace = get_callchain_entry_for_task(task, max_depth);
471	} else {
472		trace = get_perf_callchain(regs, kernel, user, max_depth,
473					   crosstask, false, 0);
474	}
475
476	if (unlikely(!trace) || trace->nr < skip) {
477		if (may_fault)
478			rcu_read_unlock();
479		goto err_fault;
480	}
481
482	trace_nr = trace->nr - skip;
483	copy_len = trace_nr * elem_size;
484
485	ips = trace->ip + skip;
486	if (user_build_id) {
487		struct bpf_stack_build_id *id_offs = buf;
488		u32 i;
489
490		for (i = 0; i < trace_nr; i++)
491			id_offs[i].ip = ips[i];
492	} else {
493		memcpy(buf, ips, copy_len);
494	}
495
496	/* trace/ips should not be dereferenced after this point */
497	if (may_fault)
498		rcu_read_unlock();
499
500	if (user_build_id)
501		stack_map_get_build_id_offset(buf, trace_nr, user, may_fault);
502
503	if (size > copy_len)
504		memset(buf + copy_len, 0, size - copy_len);
505	return copy_len;
506
507err_fault:
508	err = -EFAULT;
509clear:
510	memset(buf, 0, size);
511	return err;
512}
513
514BPF_CALL_4(bpf_get_stack, struct pt_regs *, regs, void *, buf, u32, size,
515	   u64, flags)
516{
517	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags, false /* !may_fault */);
518}
519
520const struct bpf_func_proto bpf_get_stack_proto = {
521	.func		= bpf_get_stack,
522	.gpl_only	= true,
523	.ret_type	= RET_INTEGER,
524	.arg1_type	= ARG_PTR_TO_CTX,
525	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
526	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
527	.arg4_type	= ARG_ANYTHING,
528};
529
530BPF_CALL_4(bpf_get_stack_sleepable, struct pt_regs *, regs, void *, buf, u32, size,
531	   u64, flags)
532{
533	return __bpf_get_stack(regs, NULL, NULL, buf, size, flags, true /* may_fault */);
534}
535
536const struct bpf_func_proto bpf_get_stack_sleepable_proto = {
537	.func		= bpf_get_stack_sleepable,
538	.gpl_only	= true,
539	.ret_type	= RET_INTEGER,
540	.arg1_type	= ARG_PTR_TO_CTX,
541	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
542	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
543	.arg4_type	= ARG_ANYTHING,
544};
545
546static long __bpf_get_task_stack(struct task_struct *task, void *buf, u32 size,
547				 u64 flags, bool may_fault)
548{
549	struct pt_regs *regs;
550	long res = -EINVAL;
551
552	if (!try_get_task_stack(task))
553		return -EFAULT;
554
555	regs = task_pt_regs(task);
556	if (regs)
557		res = __bpf_get_stack(regs, task, NULL, buf, size, flags, may_fault);
558	put_task_stack(task);
559
560	return res;
561}
562
563BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
564	   u32, size, u64, flags)
565{
566	return __bpf_get_task_stack(task, buf, size, flags, false /* !may_fault */);
567}
568
569const struct bpf_func_proto bpf_get_task_stack_proto = {
570	.func		= bpf_get_task_stack,
571	.gpl_only	= false,
572	.ret_type	= RET_INTEGER,
573	.arg1_type	= ARG_PTR_TO_BTF_ID,
574	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
575	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
576	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
577	.arg4_type	= ARG_ANYTHING,
578};
579
580BPF_CALL_4(bpf_get_task_stack_sleepable, struct task_struct *, task, void *, buf,
581	   u32, size, u64, flags)
582{
583	return __bpf_get_task_stack(task, buf, size, flags, true /* !may_fault */);
584}
585
586const struct bpf_func_proto bpf_get_task_stack_sleepable_proto = {
587	.func		= bpf_get_task_stack_sleepable,
588	.gpl_only	= false,
589	.ret_type	= RET_INTEGER,
590	.arg1_type	= ARG_PTR_TO_BTF_ID,
591	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
592	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
593	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
594	.arg4_type	= ARG_ANYTHING,
595};
596
597BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx,
598	   void *, buf, u32, size, u64, flags)
599{
600	struct pt_regs *regs = (struct pt_regs *)(ctx->regs);
601	struct perf_event *event = ctx->event;
602	struct perf_callchain_entry *trace;
603	bool kernel, user;
604	int err = -EINVAL;
605	__u64 nr_kernel;
606
607	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
608		return __bpf_get_stack(regs, NULL, NULL, buf, size, flags, false /* !may_fault */);
609
610	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
611			       BPF_F_USER_BUILD_ID)))
612		goto clear;
613
614	user = flags & BPF_F_USER_STACK;
615	kernel = !user;
616
617	err = -EFAULT;
618	trace = ctx->data->callchain;
619	if (unlikely(!trace))
620		goto clear;
621
622	nr_kernel = count_kernel_ip(trace);
623
624	if (kernel) {
625		__u64 nr = trace->nr;
626
627		trace->nr = nr_kernel;
628		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags, false /* !may_fault */);
629
630		/* restore nr */
631		trace->nr = nr;
632	} else { /* user */
633		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
634
635		skip += nr_kernel;
636		if (skip > BPF_F_SKIP_FIELD_MASK)
637			goto clear;
638
639		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
640		err = __bpf_get_stack(regs, NULL, trace, buf, size, flags, false /* !may_fault */);
641	}
642	return err;
643
644clear:
645	memset(buf, 0, size);
646	return err;
647
648}
649
650const struct bpf_func_proto bpf_get_stack_proto_pe = {
651	.func		= bpf_get_stack_pe,
652	.gpl_only	= true,
653	.ret_type	= RET_INTEGER,
654	.arg1_type	= ARG_PTR_TO_CTX,
655	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
656	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
657	.arg4_type	= ARG_ANYTHING,
658};
659
660/* Called from eBPF program */
661static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
662{
663	return ERR_PTR(-EOPNOTSUPP);
664}
665
666/* Called from syscall */
667static int stack_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
668					    void *value, u64 flags)
669{
670	return bpf_stackmap_extract(map, key, value, true);
671}
672
673/* Called from syscall */
674int bpf_stackmap_extract(struct bpf_map *map, void *key, void *value,
675			 bool delete)
676{
677	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
678	struct stack_map_bucket *bucket, *old_bucket;
679	u32 id = *(u32 *)key, trace_len;
680
681	if (unlikely(id >= smap->n_buckets))
682		return -ENOENT;
683
684	bucket = xchg(&smap->buckets[id], NULL);
685	if (!bucket)
686		return -ENOENT;
687
688	trace_len = bucket->nr * stack_map_data_size(map);
689	memcpy(value, bucket->data, trace_len);
690	memset(value + trace_len, 0, map->value_size - trace_len);
691
692	if (delete)
693		old_bucket = bucket;
694	else
695		old_bucket = xchg(&smap->buckets[id], bucket);
696	if (old_bucket)
697		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
698	return 0;
699}
700
701static int stack_map_get_next_key(struct bpf_map *map, void *key,
702				  void *next_key)
703{
704	struct bpf_stack_map *smap = container_of(map,
705						  struct bpf_stack_map, map);
706	u32 id;
707
708	WARN_ON_ONCE(!rcu_read_lock_held());
709
710	if (!key) {
711		id = 0;
712	} else {
713		id = *(u32 *)key;
714		if (id >= smap->n_buckets || !smap->buckets[id])
715			id = 0;
716		else
717			id++;
718	}
719
720	while (id < smap->n_buckets && !smap->buckets[id])
721		id++;
722
723	if (id >= smap->n_buckets)
724		return -ENOENT;
725
726	*(u32 *)next_key = id;
727	return 0;
728}
729
730static long stack_map_update_elem(struct bpf_map *map, void *key, void *value,
731				  u64 map_flags)
732{
733	return -EINVAL;
734}
735
736/* Called from syscall or from eBPF program */
737static long stack_map_delete_elem(struct bpf_map *map, void *key)
738{
739	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
740	struct stack_map_bucket *old_bucket;
741	u32 id = *(u32 *)key;
742
743	if (unlikely(id >= smap->n_buckets))
744		return -E2BIG;
745
746	old_bucket = xchg(&smap->buckets[id], NULL);
747	if (old_bucket) {
748		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
749		return 0;
750	} else {
751		return -ENOENT;
752	}
753}
754
755/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
756static void stack_map_free(struct bpf_map *map)
757{
758	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
759
760	bpf_map_area_free(smap->elems);
761	pcpu_freelist_destroy(&smap->freelist);
762	bpf_map_area_free(smap);
763	put_callchain_buffers();
764}
765
766static u64 stack_map_mem_usage(const struct bpf_map *map)
767{
768	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
769	u64 value_size = map->value_size;
770	u64 n_buckets = smap->n_buckets;
771	u64 enties = map->max_entries;
772	u64 usage = sizeof(*smap);
773
774	usage += n_buckets * sizeof(struct stack_map_bucket *);
775	usage += enties * (sizeof(struct stack_map_bucket) + value_size);
776	return usage;
777}
778
779BTF_ID_LIST_SINGLE(stack_trace_map_btf_ids, struct, bpf_stack_map)
780const struct bpf_map_ops stack_trace_map_ops = {
781	.map_meta_equal = bpf_map_meta_equal,
782	.map_alloc = stack_map_alloc,
783	.map_free = stack_map_free,
784	.map_get_next_key = stack_map_get_next_key,
785	.map_lookup_elem = stack_map_lookup_elem,
786	.map_lookup_and_delete_elem = stack_map_lookup_and_delete_elem,
787	.map_update_elem = stack_map_update_elem,
788	.map_delete_elem = stack_map_delete_elem,
789	.map_check_btf = map_check_no_btf,
790	.map_mem_usage = stack_map_mem_usage,
791	.map_btf_id = &stack_trace_map_btf_ids[0],
792};