xref: /openbmc/linux/kernel/bpf/stackmap.c (revision 2359ccdd)
1 /* Copyright (c) 2016 Facebook
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of version 2 of the GNU General Public
5  * License as published by the Free Software Foundation.
6  */
7 #include <linux/bpf.h>
8 #include <linux/jhash.h>
9 #include <linux/filter.h>
10 #include <linux/stacktrace.h>
11 #include <linux/perf_event.h>
12 #include <linux/elf.h>
13 #include <linux/pagemap.h>
14 #include "percpu_freelist.h"
15 
16 #define STACK_CREATE_FLAG_MASK					\
17 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY |	\
18 	 BPF_F_STACK_BUILD_ID)
19 
20 struct stack_map_bucket {
21 	struct pcpu_freelist_node fnode;
22 	u32 hash;
23 	u32 nr;
24 	u64 data[];
25 };
26 
27 struct bpf_stack_map {
28 	struct bpf_map map;
29 	void *elems;
30 	struct pcpu_freelist freelist;
31 	u32 n_buckets;
32 	struct stack_map_bucket *buckets[];
33 };
34 
35 static inline bool stack_map_use_build_id(struct bpf_map *map)
36 {
37 	return (map->map_flags & BPF_F_STACK_BUILD_ID);
38 }
39 
40 static inline int stack_map_data_size(struct bpf_map *map)
41 {
42 	return stack_map_use_build_id(map) ?
43 		sizeof(struct bpf_stack_build_id) : sizeof(u64);
44 }
45 
46 static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
47 {
48 	u32 elem_size = sizeof(struct stack_map_bucket) + 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 (!capable(CAP_SYS_ADMIN))
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 
100 	cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
101 	if (cost >= U32_MAX - PAGE_SIZE)
102 		return ERR_PTR(-E2BIG);
103 
104 	smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
105 	if (!smap)
106 		return ERR_PTR(-ENOMEM);
107 
108 	err = -E2BIG;
109 	cost += n_buckets * (value_size + sizeof(struct stack_map_bucket));
110 	if (cost >= U32_MAX - PAGE_SIZE)
111 		goto free_smap;
112 
113 	bpf_map_init_from_attr(&smap->map, attr);
114 	smap->map.value_size = value_size;
115 	smap->n_buckets = n_buckets;
116 	smap->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
117 
118 	err = bpf_map_precharge_memlock(smap->map.pages);
119 	if (err)
120 		goto free_smap;
121 
122 	err = get_callchain_buffers(sysctl_perf_event_max_stack);
123 	if (err)
124 		goto free_smap;
125 
126 	err = prealloc_elems_and_freelist(smap);
127 	if (err)
128 		goto put_buffers;
129 
130 	return &smap->map;
131 
132 put_buffers:
133 	put_callchain_buffers();
134 free_smap:
135 	bpf_map_area_free(smap);
136 	return ERR_PTR(err);
137 }
138 
139 #define BPF_BUILD_ID 3
140 /*
141  * Parse build id from the note segment. This logic can be shared between
142  * 32-bit and 64-bit system, because Elf32_Nhdr and Elf64_Nhdr are
143  * identical.
144  */
145 static inline int stack_map_parse_build_id(void *page_addr,
146 					   unsigned char *build_id,
147 					   void *note_start,
148 					   Elf32_Word note_size)
149 {
150 	Elf32_Word note_offs = 0, new_offs;
151 
152 	/* check for overflow */
153 	if (note_start < page_addr || note_start + note_size < note_start)
154 		return -EINVAL;
155 
156 	/* only supports note that fits in the first page */
157 	if (note_start + note_size > page_addr + PAGE_SIZE)
158 		return -EINVAL;
159 
160 	while (note_offs + sizeof(Elf32_Nhdr) < note_size) {
161 		Elf32_Nhdr *nhdr = (Elf32_Nhdr *)(note_start + note_offs);
162 
163 		if (nhdr->n_type == BPF_BUILD_ID &&
164 		    nhdr->n_namesz == sizeof("GNU") &&
165 		    nhdr->n_descsz == BPF_BUILD_ID_SIZE) {
166 			memcpy(build_id,
167 			       note_start + note_offs +
168 			       ALIGN(sizeof("GNU"), 4) + sizeof(Elf32_Nhdr),
169 			       BPF_BUILD_ID_SIZE);
170 			return 0;
171 		}
172 		new_offs = note_offs + sizeof(Elf32_Nhdr) +
173 			ALIGN(nhdr->n_namesz, 4) + ALIGN(nhdr->n_descsz, 4);
174 		if (new_offs <= note_offs)  /* overflow */
175 			break;
176 		note_offs = new_offs;
177 	}
178 	return -EINVAL;
179 }
180 
181 /* Parse build ID from 32-bit ELF */
182 static int stack_map_get_build_id_32(void *page_addr,
183 				     unsigned char *build_id)
184 {
185 	Elf32_Ehdr *ehdr = (Elf32_Ehdr *)page_addr;
186 	Elf32_Phdr *phdr;
187 	int i;
188 
189 	/* only supports phdr that fits in one page */
190 	if (ehdr->e_phnum >
191 	    (PAGE_SIZE - sizeof(Elf32_Ehdr)) / sizeof(Elf32_Phdr))
192 		return -EINVAL;
193 
194 	phdr = (Elf32_Phdr *)(page_addr + sizeof(Elf32_Ehdr));
195 
196 	for (i = 0; i < ehdr->e_phnum; ++i)
197 		if (phdr[i].p_type == PT_NOTE)
198 			return stack_map_parse_build_id(page_addr, build_id,
199 					page_addr + phdr[i].p_offset,
200 					phdr[i].p_filesz);
201 	return -EINVAL;
202 }
203 
204 /* Parse build ID from 64-bit ELF */
205 static int stack_map_get_build_id_64(void *page_addr,
206 				     unsigned char *build_id)
207 {
208 	Elf64_Ehdr *ehdr = (Elf64_Ehdr *)page_addr;
209 	Elf64_Phdr *phdr;
210 	int i;
211 
212 	/* only supports phdr that fits in one page */
213 	if (ehdr->e_phnum >
214 	    (PAGE_SIZE - sizeof(Elf64_Ehdr)) / sizeof(Elf64_Phdr))
215 		return -EINVAL;
216 
217 	phdr = (Elf64_Phdr *)(page_addr + sizeof(Elf64_Ehdr));
218 
219 	for (i = 0; i < ehdr->e_phnum; ++i)
220 		if (phdr[i].p_type == PT_NOTE)
221 			return stack_map_parse_build_id(page_addr, build_id,
222 					page_addr + phdr[i].p_offset,
223 					phdr[i].p_filesz);
224 	return -EINVAL;
225 }
226 
227 /* Parse build ID of ELF file mapped to vma */
228 static int stack_map_get_build_id(struct vm_area_struct *vma,
229 				  unsigned char *build_id)
230 {
231 	Elf32_Ehdr *ehdr;
232 	struct page *page;
233 	void *page_addr;
234 	int ret;
235 
236 	/* only works for page backed storage  */
237 	if (!vma->vm_file)
238 		return -EINVAL;
239 
240 	page = find_get_page(vma->vm_file->f_mapping, 0);
241 	if (!page)
242 		return -EFAULT;	/* page not mapped */
243 
244 	ret = -EINVAL;
245 	page_addr = page_address(page);
246 	ehdr = (Elf32_Ehdr *)page_addr;
247 
248 	/* compare magic x7f "ELF" */
249 	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0)
250 		goto out;
251 
252 	/* only support executable file and shared object file */
253 	if (ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN)
254 		goto out;
255 
256 	if (ehdr->e_ident[EI_CLASS] == ELFCLASS32)
257 		ret = stack_map_get_build_id_32(page_addr, build_id);
258 	else if (ehdr->e_ident[EI_CLASS] == ELFCLASS64)
259 		ret = stack_map_get_build_id_64(page_addr, build_id);
260 out:
261 	put_page(page);
262 	return ret;
263 }
264 
265 static void stack_map_get_build_id_offset(struct bpf_map *map,
266 					  struct stack_map_bucket *bucket,
267 					  u64 *ips, u32 trace_nr, bool user)
268 {
269 	int i;
270 	struct vm_area_struct *vma;
271 	struct bpf_stack_build_id *id_offs;
272 
273 	bucket->nr = trace_nr;
274 	id_offs = (struct bpf_stack_build_id *)bucket->data;
275 
276 	/*
277 	 * We cannot do up_read() in nmi context, so build_id lookup is
278 	 * only supported for non-nmi events. If at some point, it is
279 	 * possible to run find_vma() without taking the semaphore, we
280 	 * would like to allow build_id lookup in nmi context.
281 	 *
282 	 * Same fallback is used for kernel stack (!user) on a stackmap
283 	 * with build_id.
284 	 */
285 	if (!user || !current || !current->mm || in_nmi() ||
286 	    down_read_trylock(&current->mm->mmap_sem) == 0) {
287 		/* cannot access current->mm, fall back to ips */
288 		for (i = 0; i < trace_nr; i++) {
289 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
290 			id_offs[i].ip = ips[i];
291 		}
292 		return;
293 	}
294 
295 	for (i = 0; i < trace_nr; i++) {
296 		vma = find_vma(current->mm, ips[i]);
297 		if (!vma || stack_map_get_build_id(vma, id_offs[i].build_id)) {
298 			/* per entry fall back to ips */
299 			id_offs[i].status = BPF_STACK_BUILD_ID_IP;
300 			id_offs[i].ip = ips[i];
301 			continue;
302 		}
303 		id_offs[i].offset = (vma->vm_pgoff << PAGE_SHIFT) + ips[i]
304 			- vma->vm_start;
305 		id_offs[i].status = BPF_STACK_BUILD_ID_VALID;
306 	}
307 	up_read(&current->mm->mmap_sem);
308 }
309 
310 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
311 	   u64, flags)
312 {
313 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
314 	struct perf_callchain_entry *trace;
315 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
316 	u32 max_depth = map->value_size / stack_map_data_size(map);
317 	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
318 	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
319 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
320 	u32 hash, id, trace_nr, trace_len;
321 	bool user = flags & BPF_F_USER_STACK;
322 	bool kernel = !user;
323 	u64 *ips;
324 	bool hash_matches;
325 
326 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
327 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
328 		return -EINVAL;
329 
330 	trace = get_perf_callchain(regs, init_nr, kernel, user,
331 				   sysctl_perf_event_max_stack, false, false);
332 
333 	if (unlikely(!trace))
334 		/* couldn't fetch the stack trace */
335 		return -EFAULT;
336 
337 	/* get_perf_callchain() guarantees that trace->nr >= init_nr
338 	 * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
339 	 */
340 	trace_nr = trace->nr - init_nr;
341 
342 	if (trace_nr <= skip)
343 		/* skipping more than usable stack trace */
344 		return -EFAULT;
345 
346 	trace_nr -= skip;
347 	trace_len = trace_nr * sizeof(u64);
348 	ips = trace->ip + skip + init_nr;
349 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
350 	id = hash & (smap->n_buckets - 1);
351 	bucket = READ_ONCE(smap->buckets[id]);
352 
353 	hash_matches = bucket && bucket->hash == hash;
354 	/* fast cmp */
355 	if (hash_matches && flags & BPF_F_FAST_STACK_CMP)
356 		return id;
357 
358 	if (stack_map_use_build_id(map)) {
359 		/* for build_id+offset, pop a bucket before slow cmp */
360 		new_bucket = (struct stack_map_bucket *)
361 			pcpu_freelist_pop(&smap->freelist);
362 		if (unlikely(!new_bucket))
363 			return -ENOMEM;
364 		stack_map_get_build_id_offset(map, new_bucket, ips,
365 					      trace_nr, user);
366 		trace_len = trace_nr * sizeof(struct bpf_stack_build_id);
367 		if (hash_matches && bucket->nr == trace_nr &&
368 		    memcmp(bucket->data, new_bucket->data, trace_len) == 0) {
369 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
370 			return id;
371 		}
372 		if (bucket && !(flags & BPF_F_REUSE_STACKID)) {
373 			pcpu_freelist_push(&smap->freelist, &new_bucket->fnode);
374 			return -EEXIST;
375 		}
376 	} else {
377 		if (hash_matches && bucket->nr == trace_nr &&
378 		    memcmp(bucket->data, ips, trace_len) == 0)
379 			return id;
380 		if (bucket && !(flags & BPF_F_REUSE_STACKID))
381 			return -EEXIST;
382 
383 		new_bucket = (struct stack_map_bucket *)
384 			pcpu_freelist_pop(&smap->freelist);
385 		if (unlikely(!new_bucket))
386 			return -ENOMEM;
387 		memcpy(new_bucket->data, ips, trace_len);
388 	}
389 
390 	new_bucket->hash = hash;
391 	new_bucket->nr = trace_nr;
392 
393 	old_bucket = xchg(&smap->buckets[id], new_bucket);
394 	if (old_bucket)
395 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
396 	return id;
397 }
398 
399 const struct bpf_func_proto bpf_get_stackid_proto = {
400 	.func		= bpf_get_stackid,
401 	.gpl_only	= true,
402 	.ret_type	= RET_INTEGER,
403 	.arg1_type	= ARG_PTR_TO_CTX,
404 	.arg2_type	= ARG_CONST_MAP_PTR,
405 	.arg3_type	= ARG_ANYTHING,
406 };
407 
408 /* Called from eBPF program */
409 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
410 {
411 	return NULL;
412 }
413 
414 /* Called from syscall */
415 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
416 {
417 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
418 	struct stack_map_bucket *bucket, *old_bucket;
419 	u32 id = *(u32 *)key, trace_len;
420 
421 	if (unlikely(id >= smap->n_buckets))
422 		return -ENOENT;
423 
424 	bucket = xchg(&smap->buckets[id], NULL);
425 	if (!bucket)
426 		return -ENOENT;
427 
428 	trace_len = bucket->nr * stack_map_data_size(map);
429 	memcpy(value, bucket->data, trace_len);
430 	memset(value + trace_len, 0, map->value_size - trace_len);
431 
432 	old_bucket = xchg(&smap->buckets[id], bucket);
433 	if (old_bucket)
434 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
435 	return 0;
436 }
437 
438 static int stack_map_get_next_key(struct bpf_map *map, void *key,
439 				  void *next_key)
440 {
441 	struct bpf_stack_map *smap = container_of(map,
442 						  struct bpf_stack_map, map);
443 	u32 id;
444 
445 	WARN_ON_ONCE(!rcu_read_lock_held());
446 
447 	if (!key) {
448 		id = 0;
449 	} else {
450 		id = *(u32 *)key;
451 		if (id >= smap->n_buckets || !smap->buckets[id])
452 			id = 0;
453 		else
454 			id++;
455 	}
456 
457 	while (id < smap->n_buckets && !smap->buckets[id])
458 		id++;
459 
460 	if (id >= smap->n_buckets)
461 		return -ENOENT;
462 
463 	*(u32 *)next_key = id;
464 	return 0;
465 }
466 
467 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
468 				 u64 map_flags)
469 {
470 	return -EINVAL;
471 }
472 
473 /* Called from syscall or from eBPF program */
474 static int stack_map_delete_elem(struct bpf_map *map, void *key)
475 {
476 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
477 	struct stack_map_bucket *old_bucket;
478 	u32 id = *(u32 *)key;
479 
480 	if (unlikely(id >= smap->n_buckets))
481 		return -E2BIG;
482 
483 	old_bucket = xchg(&smap->buckets[id], NULL);
484 	if (old_bucket) {
485 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
486 		return 0;
487 	} else {
488 		return -ENOENT;
489 	}
490 }
491 
492 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
493 static void stack_map_free(struct bpf_map *map)
494 {
495 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
496 
497 	/* wait for bpf programs to complete before freeing stack map */
498 	synchronize_rcu();
499 
500 	bpf_map_area_free(smap->elems);
501 	pcpu_freelist_destroy(&smap->freelist);
502 	bpf_map_area_free(smap);
503 	put_callchain_buffers();
504 }
505 
506 const struct bpf_map_ops stack_map_ops = {
507 	.map_alloc = stack_map_alloc,
508 	.map_free = stack_map_free,
509 	.map_get_next_key = stack_map_get_next_key,
510 	.map_lookup_elem = stack_map_lookup_elem,
511 	.map_update_elem = stack_map_update_elem,
512 	.map_delete_elem = stack_map_delete_elem,
513 };
514