projects/linux·blob v4.8-rc1 · refs

291 lines · 7.4 KiB

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/vmalloc.h>
11#include <linux/stacktrace.h>
12#include <linux/perf_event.h>
13#include "percpu_freelist.h"
14
15struct stack_map_bucket {
16 struct pcpu_freelist_node fnode;
17 u32 hash;
18 u32 nr;
19 u64 ip[];
20};
21
22struct bpf_stack_map {
23 struct bpf_map map;
24 void *elems;
25 struct pcpu_freelist freelist;
26 u32 n_buckets;
27 struct stack_map_bucket *buckets[];
28};
29
30static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
31{
32 u32 elem_size = sizeof(struct stack_map_bucket) + smap->map.value_size;
33 int err;
34
35 smap->elems = vzalloc(elem_size * smap->map.max_entries);
36 if (!smap->elems)
37 return -ENOMEM;
38
39 err = pcpu_freelist_init(&smap->freelist);
40 if (err)
41 goto free_elems;
42
43 pcpu_freelist_populate(&smap->freelist, smap->elems, elem_size,
44 smap->map.max_entries);
45 return 0;
46
47free_elems:
48 vfree(smap->elems);
49 return err;
50}
51
52/* Called from syscall */
53static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
54{
55 u32 value_size = attr->value_size;
56 struct bpf_stack_map *smap;
57 u64 cost, n_buckets;
58 int err;
59
60 if (!capable(CAP_SYS_ADMIN))
61 return ERR_PTR(-EPERM);
62
63 if (attr->map_flags)
64 return ERR_PTR(-EINVAL);
65
66 /* check sanity of attributes */
67 if (attr->max_entries == 0 || attr->key_size != 4 ||
68 value_size < 8 || value_size % 8 ||
69 value_size / 8 > sysctl_perf_event_max_stack)
70 return ERR_PTR(-EINVAL);
71
72 /* hash table size must be power of 2 */
73 n_buckets = roundup_pow_of_two(attr->max_entries);
74
75 cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
76 if (cost >= U32_MAX - PAGE_SIZE)
77 return ERR_PTR(-E2BIG);
78
79 smap = kzalloc(cost, GFP_USER | __GFP_NOWARN);
80 if (!smap) {
81 smap = vzalloc(cost);
82 if (!smap)
83 return ERR_PTR(-ENOMEM);
84 }
85
86 err = -E2BIG;
87 cost += n_buckets * (value_size + sizeof(struct stack_map_bucket));
88 if (cost >= U32_MAX - PAGE_SIZE)
89 goto free_smap;
90
91 smap->map.map_type = attr->map_type;
92 smap->map.key_size = attr->key_size;
93 smap->map.value_size = value_size;
94 smap->map.max_entries = attr->max_entries;
95 smap->n_buckets = n_buckets;
96 smap->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
97
98 err = bpf_map_precharge_memlock(smap->map.pages);
99 if (err)
100 goto free_smap;
101
102 err = get_callchain_buffers(sysctl_perf_event_max_stack);
103 if (err)
104 goto free_smap;
105
106 err = prealloc_elems_and_freelist(smap);
107 if (err)
108 goto put_buffers;
109
110 return &smap->map;
111
112put_buffers:
113 put_callchain_buffers();
114free_smap:
115 kvfree(smap);
116 return ERR_PTR(err);
117}
118
119u64 bpf_get_stackid(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5)
120{
121 struct pt_regs *regs = (struct pt_regs *) (long) r1;
122 struct bpf_map *map = (struct bpf_map *) (long) r2;
123 struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
124 struct perf_callchain_entry *trace;
125 struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
126 u32 max_depth = map->value_size / 8;
127 /* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
128 u32 init_nr = sysctl_perf_event_max_stack - max_depth;
129 u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
130 u32 hash, id, trace_nr, trace_len;
131 bool user = flags & BPF_F_USER_STACK;
132 bool kernel = !user;
133 u64 *ips;
134
135 if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
136 BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
137 return -EINVAL;
138
139 trace = get_perf_callchain(regs, init_nr, kernel, user,
140 sysctl_perf_event_max_stack, false, false);
141
142 if (unlikely(!trace))
143 /* couldn't fetch the stack trace */
144 return -EFAULT;
145
146 /* get_perf_callchain() guarantees that trace->nr >= init_nr
147 * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
148 */
149 trace_nr = trace->nr - init_nr;
150
151 if (trace_nr <= skip)
152 /* skipping more than usable stack trace */
153 return -EFAULT;
154
155 trace_nr -= skip;
156 trace_len = trace_nr * sizeof(u64);
157 ips = trace->ip + skip + init_nr;
158 hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
159 id = hash & (smap->n_buckets - 1);
160 bucket = READ_ONCE(smap->buckets[id]);
161
162 if (bucket && bucket->hash == hash) {
163 if (flags & BPF_F_FAST_STACK_CMP)
164 return id;
165 if (bucket->nr == trace_nr &&
166 memcmp(bucket->ip, ips, trace_len) == 0)
167 return id;
168 }
169
170 /* this call stack is not in the map, try to add it */
171 if (bucket && !(flags & BPF_F_REUSE_STACKID))
172 return -EEXIST;
173
174 new_bucket = (struct stack_map_bucket *)
175 pcpu_freelist_pop(&smap->freelist);
176 if (unlikely(!new_bucket))
177 return -ENOMEM;
178
179 memcpy(new_bucket->ip, ips, trace_len);
180 new_bucket->hash = hash;
181 new_bucket->nr = trace_nr;
182
183 old_bucket = xchg(&smap->buckets[id], new_bucket);
184 if (old_bucket)
185 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
186 return id;
187}
188
189const struct bpf_func_proto bpf_get_stackid_proto = {
190 .func = bpf_get_stackid,
191 .gpl_only = true,
192 .ret_type = RET_INTEGER,
193 .arg1_type = ARG_PTR_TO_CTX,
194 .arg2_type = ARG_CONST_MAP_PTR,
195 .arg3_type = ARG_ANYTHING,
196};
197
198/* Called from eBPF program */
199static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
200{
201 return NULL;
202}
203
204/* Called from syscall */
205int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
206{
207 struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
208 struct stack_map_bucket *bucket, *old_bucket;
209 u32 id = *(u32 *)key, trace_len;
210
211 if (unlikely(id >= smap->n_buckets))
212 return -ENOENT;
213
214 bucket = xchg(&smap->buckets[id], NULL);
215 if (!bucket)
216 return -ENOENT;
217
218 trace_len = bucket->nr * sizeof(u64);
219 memcpy(value, bucket->ip, trace_len);
220 memset(value + trace_len, 0, map->value_size - trace_len);
221
222 old_bucket = xchg(&smap->buckets[id], bucket);
223 if (old_bucket)
224 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
225 return 0;
226}
227
228static int stack_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
229{
230 return -EINVAL;
231}
232
233static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
234 u64 map_flags)
235{
236 return -EINVAL;
237}
238
239/* Called from syscall or from eBPF program */
240static int stack_map_delete_elem(struct bpf_map *map, void *key)
241{
242 struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
243 struct stack_map_bucket *old_bucket;
244 u32 id = *(u32 *)key;
245
246 if (unlikely(id >= smap->n_buckets))
247 return -E2BIG;
248
249 old_bucket = xchg(&smap->buckets[id], NULL);
250 if (old_bucket) {
251 pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
252 return 0;
253 } else {
254 return -ENOENT;
255 }
256}
257
258/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
259static void stack_map_free(struct bpf_map *map)
260{
261 struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
262
263 /* wait for bpf programs to complete before freeing stack map */
264 synchronize_rcu();
265
266 vfree(smap->elems);
267 pcpu_freelist_destroy(&smap->freelist);
268 kvfree(smap);
269 put_callchain_buffers();
270}
271
272static const struct bpf_map_ops stack_map_ops = {
273 .map_alloc = stack_map_alloc,
274 .map_free = stack_map_free,
275 .map_get_next_key = stack_map_get_next_key,
276 .map_lookup_elem = stack_map_lookup_elem,
277 .map_update_elem = stack_map_update_elem,
278 .map_delete_elem = stack_map_delete_elem,
279};
280
281static struct bpf_map_type_list stack_map_type __read_mostly = {
282 .ops = &stack_map_ops,
283 .type = BPF_MAP_TYPE_STACK_TRACE,
284};
285
286static int __init register_stack_map(void)
287{
288 bpf_register_map_type(&stack_map_type);
289 return 0;
290}
291late_initcall(register_stack_map);