xref: /openbmc/linux/kernel/bpf/stackmap.c (revision 726bd223)
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 "percpu_freelist.h"
13 
14 struct stack_map_bucket {
15 	struct pcpu_freelist_node fnode;
16 	u32 hash;
17 	u32 nr;
18 	u64 ip[];
19 };
20 
21 struct bpf_stack_map {
22 	struct bpf_map map;
23 	void *elems;
24 	struct pcpu_freelist freelist;
25 	u32 n_buckets;
26 	struct stack_map_bucket *buckets[];
27 };
28 
29 static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
30 {
31 	u32 elem_size = sizeof(struct stack_map_bucket) + smap->map.value_size;
32 	int err;
33 
34 	smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries);
35 	if (!smap->elems)
36 		return -ENOMEM;
37 
38 	err = pcpu_freelist_init(&smap->freelist);
39 	if (err)
40 		goto free_elems;
41 
42 	pcpu_freelist_populate(&smap->freelist, smap->elems, elem_size,
43 			       smap->map.max_entries);
44 	return 0;
45 
46 free_elems:
47 	bpf_map_area_free(smap->elems);
48 	return err;
49 }
50 
51 /* Called from syscall */
52 static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
53 {
54 	u32 value_size = attr->value_size;
55 	struct bpf_stack_map *smap;
56 	u64 cost, n_buckets;
57 	int err;
58 
59 	if (!capable(CAP_SYS_ADMIN))
60 		return ERR_PTR(-EPERM);
61 
62 	if (attr->map_flags)
63 		return ERR_PTR(-EINVAL);
64 
65 	/* check sanity of attributes */
66 	if (attr->max_entries == 0 || attr->key_size != 4 ||
67 	    value_size < 8 || value_size % 8 ||
68 	    value_size / 8 > sysctl_perf_event_max_stack)
69 		return ERR_PTR(-EINVAL);
70 
71 	/* hash table size must be power of 2 */
72 	n_buckets = roundup_pow_of_two(attr->max_entries);
73 
74 	cost = n_buckets * sizeof(struct stack_map_bucket *) + sizeof(*smap);
75 	if (cost >= U32_MAX - PAGE_SIZE)
76 		return ERR_PTR(-E2BIG);
77 
78 	smap = bpf_map_area_alloc(cost);
79 	if (!smap)
80 		return ERR_PTR(-ENOMEM);
81 
82 	err = -E2BIG;
83 	cost += n_buckets * (value_size + sizeof(struct stack_map_bucket));
84 	if (cost >= U32_MAX - PAGE_SIZE)
85 		goto free_smap;
86 
87 	smap->map.map_type = attr->map_type;
88 	smap->map.key_size = attr->key_size;
89 	smap->map.value_size = value_size;
90 	smap->map.max_entries = attr->max_entries;
91 	smap->n_buckets = n_buckets;
92 	smap->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
93 
94 	err = bpf_map_precharge_memlock(smap->map.pages);
95 	if (err)
96 		goto free_smap;
97 
98 	err = get_callchain_buffers(sysctl_perf_event_max_stack);
99 	if (err)
100 		goto free_smap;
101 
102 	err = prealloc_elems_and_freelist(smap);
103 	if (err)
104 		goto put_buffers;
105 
106 	return &smap->map;
107 
108 put_buffers:
109 	put_callchain_buffers();
110 free_smap:
111 	bpf_map_area_free(smap);
112 	return ERR_PTR(err);
113 }
114 
115 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
116 	   u64, flags)
117 {
118 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
119 	struct perf_callchain_entry *trace;
120 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
121 	u32 max_depth = map->value_size / 8;
122 	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
123 	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
124 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
125 	u32 hash, id, trace_nr, trace_len;
126 	bool user = flags & BPF_F_USER_STACK;
127 	bool kernel = !user;
128 	u64 *ips;
129 
130 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
131 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
132 		return -EINVAL;
133 
134 	trace = get_perf_callchain(regs, init_nr, kernel, user,
135 				   sysctl_perf_event_max_stack, false, false);
136 
137 	if (unlikely(!trace))
138 		/* couldn't fetch the stack trace */
139 		return -EFAULT;
140 
141 	/* get_perf_callchain() guarantees that trace->nr >= init_nr
142 	 * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
143 	 */
144 	trace_nr = trace->nr - init_nr;
145 
146 	if (trace_nr <= skip)
147 		/* skipping more than usable stack trace */
148 		return -EFAULT;
149 
150 	trace_nr -= skip;
151 	trace_len = trace_nr * sizeof(u64);
152 	ips = trace->ip + skip + init_nr;
153 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
154 	id = hash & (smap->n_buckets - 1);
155 	bucket = READ_ONCE(smap->buckets[id]);
156 
157 	if (bucket && bucket->hash == hash) {
158 		if (flags & BPF_F_FAST_STACK_CMP)
159 			return id;
160 		if (bucket->nr == trace_nr &&
161 		    memcmp(bucket->ip, ips, trace_len) == 0)
162 			return id;
163 	}
164 
165 	/* this call stack is not in the map, try to add it */
166 	if (bucket && !(flags & BPF_F_REUSE_STACKID))
167 		return -EEXIST;
168 
169 	new_bucket = (struct stack_map_bucket *)
170 		pcpu_freelist_pop(&smap->freelist);
171 	if (unlikely(!new_bucket))
172 		return -ENOMEM;
173 
174 	memcpy(new_bucket->ip, ips, trace_len);
175 	new_bucket->hash = hash;
176 	new_bucket->nr = trace_nr;
177 
178 	old_bucket = xchg(&smap->buckets[id], new_bucket);
179 	if (old_bucket)
180 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
181 	return id;
182 }
183 
184 const struct bpf_func_proto bpf_get_stackid_proto = {
185 	.func		= bpf_get_stackid,
186 	.gpl_only	= true,
187 	.ret_type	= RET_INTEGER,
188 	.arg1_type	= ARG_PTR_TO_CTX,
189 	.arg2_type	= ARG_CONST_MAP_PTR,
190 	.arg3_type	= ARG_ANYTHING,
191 };
192 
193 /* Called from eBPF program */
194 static void *stack_map_lookup_elem(struct bpf_map *map, void *key)
195 {
196 	return NULL;
197 }
198 
199 /* Called from syscall */
200 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
201 {
202 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
203 	struct stack_map_bucket *bucket, *old_bucket;
204 	u32 id = *(u32 *)key, trace_len;
205 
206 	if (unlikely(id >= smap->n_buckets))
207 		return -ENOENT;
208 
209 	bucket = xchg(&smap->buckets[id], NULL);
210 	if (!bucket)
211 		return -ENOENT;
212 
213 	trace_len = bucket->nr * sizeof(u64);
214 	memcpy(value, bucket->ip, trace_len);
215 	memset(value + trace_len, 0, map->value_size - trace_len);
216 
217 	old_bucket = xchg(&smap->buckets[id], bucket);
218 	if (old_bucket)
219 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
220 	return 0;
221 }
222 
223 static int stack_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
224 {
225 	return -EINVAL;
226 }
227 
228 static int stack_map_update_elem(struct bpf_map *map, void *key, void *value,
229 				 u64 map_flags)
230 {
231 	return -EINVAL;
232 }
233 
234 /* Called from syscall or from eBPF program */
235 static int stack_map_delete_elem(struct bpf_map *map, void *key)
236 {
237 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
238 	struct stack_map_bucket *old_bucket;
239 	u32 id = *(u32 *)key;
240 
241 	if (unlikely(id >= smap->n_buckets))
242 		return -E2BIG;
243 
244 	old_bucket = xchg(&smap->buckets[id], NULL);
245 	if (old_bucket) {
246 		pcpu_freelist_push(&smap->freelist, &old_bucket->fnode);
247 		return 0;
248 	} else {
249 		return -ENOENT;
250 	}
251 }
252 
253 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
254 static void stack_map_free(struct bpf_map *map)
255 {
256 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
257 
258 	/* wait for bpf programs to complete before freeing stack map */
259 	synchronize_rcu();
260 
261 	bpf_map_area_free(smap->elems);
262 	pcpu_freelist_destroy(&smap->freelist);
263 	bpf_map_area_free(smap);
264 	put_callchain_buffers();
265 }
266 
267 static const struct bpf_map_ops stack_map_ops = {
268 	.map_alloc = stack_map_alloc,
269 	.map_free = stack_map_free,
270 	.map_get_next_key = stack_map_get_next_key,
271 	.map_lookup_elem = stack_map_lookup_elem,
272 	.map_update_elem = stack_map_update_elem,
273 	.map_delete_elem = stack_map_delete_elem,
274 };
275 
276 static struct bpf_map_type_list stack_map_type __ro_after_init = {
277 	.ops = &stack_map_ops,
278 	.type = BPF_MAP_TYPE_STACK_TRACE,
279 };
280 
281 static int __init register_stack_map(void)
282 {
283 	bpf_register_map_type(&stack_map_type);
284 	return 0;
285 }
286 late_initcall(register_stack_map);
287