xref: /linux/kernel/bpf/bpf_task_storage.c (revision d0d106a2bd21499901299160744e5fe9f4c83ddb)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2020 Facebook
4  * Copyright 2020 Google LLC.
5  */
6 
7 #include <linux/pid.h>
8 #include <linux/sched.h>
9 #include <linux/rculist.h>
10 #include <linux/list.h>
11 #include <linux/hash.h>
12 #include <linux/types.h>
13 #include <linux/spinlock.h>
14 #include <linux/bpf.h>
15 #include <linux/bpf_local_storage.h>
16 #include <linux/filter.h>
17 #include <uapi/linux/btf.h>
18 #include <linux/btf_ids.h>
19 #include <linux/rcupdate_trace.h>
20 
21 DEFINE_BPF_STORAGE_CACHE(task_cache);
22 
23 static DEFINE_PER_CPU(int, bpf_task_storage_busy);
24 
bpf_task_storage_lock(void)25 static void bpf_task_storage_lock(void)
26 {
27 	cant_migrate();
28 	this_cpu_inc(bpf_task_storage_busy);
29 }
30 
bpf_task_storage_unlock(void)31 static void bpf_task_storage_unlock(void)
32 {
33 	this_cpu_dec(bpf_task_storage_busy);
34 }
35 
bpf_task_storage_trylock(void)36 static bool bpf_task_storage_trylock(void)
37 {
38 	cant_migrate();
39 	if (unlikely(this_cpu_inc_return(bpf_task_storage_busy) != 1)) {
40 		this_cpu_dec(bpf_task_storage_busy);
41 		return false;
42 	}
43 	return true;
44 }
45 
task_storage_ptr(void * owner)46 static struct bpf_local_storage __rcu **task_storage_ptr(void *owner)
47 {
48 	struct task_struct *task = owner;
49 
50 	return &task->bpf_storage;
51 }
52 
53 static struct bpf_local_storage_data *
task_storage_lookup(struct task_struct * task,struct bpf_map * map,bool cacheit_lockit)54 task_storage_lookup(struct task_struct *task, struct bpf_map *map,
55 		    bool cacheit_lockit)
56 {
57 	struct bpf_local_storage *task_storage;
58 	struct bpf_local_storage_map *smap;
59 
60 	task_storage =
61 		rcu_dereference_check(task->bpf_storage, bpf_rcu_lock_held());
62 	if (!task_storage)
63 		return NULL;
64 
65 	smap = (struct bpf_local_storage_map *)map;
66 	return bpf_local_storage_lookup(task_storage, smap, cacheit_lockit);
67 }
68 
bpf_task_storage_free(struct task_struct * task)69 void bpf_task_storage_free(struct task_struct *task)
70 {
71 	struct bpf_local_storage *local_storage;
72 
73 	migrate_disable();
74 	rcu_read_lock();
75 
76 	local_storage = rcu_dereference(task->bpf_storage);
77 	if (!local_storage)
78 		goto out;
79 
80 	bpf_task_storage_lock();
81 	bpf_local_storage_destroy(local_storage);
82 	bpf_task_storage_unlock();
83 out:
84 	rcu_read_unlock();
85 	migrate_enable();
86 }
87 
bpf_pid_task_storage_lookup_elem(struct bpf_map * map,void * key)88 static void *bpf_pid_task_storage_lookup_elem(struct bpf_map *map, void *key)
89 {
90 	struct bpf_local_storage_data *sdata;
91 	struct task_struct *task;
92 	unsigned int f_flags;
93 	struct pid *pid;
94 	int fd, err;
95 
96 	fd = *(int *)key;
97 	pid = pidfd_get_pid(fd, &f_flags);
98 	if (IS_ERR(pid))
99 		return ERR_CAST(pid);
100 
101 	/* We should be in an RCU read side critical section, it should be safe
102 	 * to call pid_task.
103 	 */
104 	WARN_ON_ONCE(!rcu_read_lock_held());
105 	task = pid_task(pid, PIDTYPE_PID);
106 	if (!task) {
107 		err = -ENOENT;
108 		goto out;
109 	}
110 
111 	bpf_task_storage_lock();
112 	sdata = task_storage_lookup(task, map, true);
113 	bpf_task_storage_unlock();
114 	put_pid(pid);
115 	return sdata ? sdata->data : NULL;
116 out:
117 	put_pid(pid);
118 	return ERR_PTR(err);
119 }
120 
bpf_pid_task_storage_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)121 static long bpf_pid_task_storage_update_elem(struct bpf_map *map, void *key,
122 					     void *value, u64 map_flags)
123 {
124 	struct bpf_local_storage_data *sdata;
125 	struct task_struct *task;
126 	unsigned int f_flags;
127 	struct pid *pid;
128 	int fd, err;
129 
130 	if ((map_flags & BPF_F_LOCK) && btf_record_has_field(map->record, BPF_UPTR))
131 		return -EOPNOTSUPP;
132 
133 	fd = *(int *)key;
134 	pid = pidfd_get_pid(fd, &f_flags);
135 	if (IS_ERR(pid))
136 		return PTR_ERR(pid);
137 
138 	/* We should be in an RCU read side critical section, it should be safe
139 	 * to call pid_task.
140 	 */
141 	WARN_ON_ONCE(!rcu_read_lock_held());
142 	task = pid_task(pid, PIDTYPE_PID);
143 	if (!task) {
144 		err = -ENOENT;
145 		goto out;
146 	}
147 
148 	bpf_task_storage_lock();
149 	sdata = bpf_local_storage_update(
150 		task, (struct bpf_local_storage_map *)map, value, map_flags,
151 		true, GFP_ATOMIC);
152 	bpf_task_storage_unlock();
153 
154 	err = PTR_ERR_OR_ZERO(sdata);
155 out:
156 	put_pid(pid);
157 	return err;
158 }
159 
task_storage_delete(struct task_struct * task,struct bpf_map * map,bool nobusy)160 static int task_storage_delete(struct task_struct *task, struct bpf_map *map,
161 			       bool nobusy)
162 {
163 	struct bpf_local_storage_data *sdata;
164 
165 	sdata = task_storage_lookup(task, map, false);
166 	if (!sdata)
167 		return -ENOENT;
168 
169 	if (!nobusy)
170 		return -EBUSY;
171 
172 	bpf_selem_unlink(SELEM(sdata), false);
173 
174 	return 0;
175 }
176 
bpf_pid_task_storage_delete_elem(struct bpf_map * map,void * key)177 static long bpf_pid_task_storage_delete_elem(struct bpf_map *map, void *key)
178 {
179 	struct task_struct *task;
180 	unsigned int f_flags;
181 	struct pid *pid;
182 	int fd, err;
183 
184 	fd = *(int *)key;
185 	pid = pidfd_get_pid(fd, &f_flags);
186 	if (IS_ERR(pid))
187 		return PTR_ERR(pid);
188 
189 	/* We should be in an RCU read side critical section, it should be safe
190 	 * to call pid_task.
191 	 */
192 	WARN_ON_ONCE(!rcu_read_lock_held());
193 	task = pid_task(pid, PIDTYPE_PID);
194 	if (!task) {
195 		err = -ENOENT;
196 		goto out;
197 	}
198 
199 	bpf_task_storage_lock();
200 	err = task_storage_delete(task, map, true);
201 	bpf_task_storage_unlock();
202 out:
203 	put_pid(pid);
204 	return err;
205 }
206 
207 /* Called by bpf_task_storage_get*() helpers */
__bpf_task_storage_get(struct bpf_map * map,struct task_struct * task,void * value,u64 flags,gfp_t gfp_flags,bool nobusy)208 static void *__bpf_task_storage_get(struct bpf_map *map,
209 				    struct task_struct *task, void *value,
210 				    u64 flags, gfp_t gfp_flags, bool nobusy)
211 {
212 	struct bpf_local_storage_data *sdata;
213 
214 	sdata = task_storage_lookup(task, map, nobusy);
215 	if (sdata)
216 		return sdata->data;
217 
218 	/* only allocate new storage, when the task is refcounted */
219 	if (refcount_read(&task->usage) &&
220 	    (flags & BPF_LOCAL_STORAGE_GET_F_CREATE) && nobusy) {
221 		sdata = bpf_local_storage_update(
222 			task, (struct bpf_local_storage_map *)map, value,
223 			BPF_NOEXIST, false, gfp_flags);
224 		return IS_ERR(sdata) ? NULL : sdata->data;
225 	}
226 
227 	return NULL;
228 }
229 
230 /* *gfp_flags* is a hidden argument provided by the verifier */
BPF_CALL_5(bpf_task_storage_get_recur,struct bpf_map *,map,struct task_struct *,task,void *,value,u64,flags,gfp_t,gfp_flags)231 BPF_CALL_5(bpf_task_storage_get_recur, struct bpf_map *, map, struct task_struct *,
232 	   task, void *, value, u64, flags, gfp_t, gfp_flags)
233 {
234 	bool nobusy;
235 	void *data;
236 
237 	WARN_ON_ONCE(!bpf_rcu_lock_held());
238 	if (flags & ~BPF_LOCAL_STORAGE_GET_F_CREATE || !task)
239 		return (unsigned long)NULL;
240 
241 	nobusy = bpf_task_storage_trylock();
242 	data = __bpf_task_storage_get(map, task, value, flags,
243 				      gfp_flags, nobusy);
244 	if (nobusy)
245 		bpf_task_storage_unlock();
246 	return (unsigned long)data;
247 }
248 
249 /* *gfp_flags* is a hidden argument provided by the verifier */
BPF_CALL_5(bpf_task_storage_get,struct bpf_map *,map,struct task_struct *,task,void *,value,u64,flags,gfp_t,gfp_flags)250 BPF_CALL_5(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
251 	   task, void *, value, u64, flags, gfp_t, gfp_flags)
252 {
253 	void *data;
254 
255 	WARN_ON_ONCE(!bpf_rcu_lock_held());
256 	if (flags & ~BPF_LOCAL_STORAGE_GET_F_CREATE || !task)
257 		return (unsigned long)NULL;
258 
259 	bpf_task_storage_lock();
260 	data = __bpf_task_storage_get(map, task, value, flags,
261 				      gfp_flags, true);
262 	bpf_task_storage_unlock();
263 	return (unsigned long)data;
264 }
265 
BPF_CALL_2(bpf_task_storage_delete_recur,struct bpf_map *,map,struct task_struct *,task)266 BPF_CALL_2(bpf_task_storage_delete_recur, struct bpf_map *, map, struct task_struct *,
267 	   task)
268 {
269 	bool nobusy;
270 	int ret;
271 
272 	WARN_ON_ONCE(!bpf_rcu_lock_held());
273 	if (!task)
274 		return -EINVAL;
275 
276 	nobusy = bpf_task_storage_trylock();
277 	/* This helper must only be called from places where the lifetime of the task
278 	 * is guaranteed. Either by being refcounted or by being protected
279 	 * by an RCU read-side critical section.
280 	 */
281 	ret = task_storage_delete(task, map, nobusy);
282 	if (nobusy)
283 		bpf_task_storage_unlock();
284 	return ret;
285 }
286 
BPF_CALL_2(bpf_task_storage_delete,struct bpf_map *,map,struct task_struct *,task)287 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
288 	   task)
289 {
290 	int ret;
291 
292 	WARN_ON_ONCE(!bpf_rcu_lock_held());
293 	if (!task)
294 		return -EINVAL;
295 
296 	bpf_task_storage_lock();
297 	/* This helper must only be called from places where the lifetime of the task
298 	 * is guaranteed. Either by being refcounted or by being protected
299 	 * by an RCU read-side critical section.
300 	 */
301 	ret = task_storage_delete(task, map, true);
302 	bpf_task_storage_unlock();
303 	return ret;
304 }
305 
notsupp_get_next_key(struct bpf_map * map,void * key,void * next_key)306 static int notsupp_get_next_key(struct bpf_map *map, void *key, void *next_key)
307 {
308 	return -ENOTSUPP;
309 }
310 
task_storage_map_alloc(union bpf_attr * attr)311 static struct bpf_map *task_storage_map_alloc(union bpf_attr *attr)
312 {
313 	return bpf_local_storage_map_alloc(attr, &task_cache, true);
314 }
315 
task_storage_map_free(struct bpf_map * map)316 static void task_storage_map_free(struct bpf_map *map)
317 {
318 	bpf_local_storage_map_free(map, &task_cache, &bpf_task_storage_busy);
319 }
320 
321 BTF_ID_LIST_GLOBAL_SINGLE(bpf_local_storage_map_btf_id, struct, bpf_local_storage_map)
322 const struct bpf_map_ops task_storage_map_ops = {
323 	.map_meta_equal = bpf_map_meta_equal,
324 	.map_alloc_check = bpf_local_storage_map_alloc_check,
325 	.map_alloc = task_storage_map_alloc,
326 	.map_free = task_storage_map_free,
327 	.map_get_next_key = notsupp_get_next_key,
328 	.map_lookup_elem = bpf_pid_task_storage_lookup_elem,
329 	.map_update_elem = bpf_pid_task_storage_update_elem,
330 	.map_delete_elem = bpf_pid_task_storage_delete_elem,
331 	.map_check_btf = bpf_local_storage_map_check_btf,
332 	.map_mem_usage = bpf_local_storage_map_mem_usage,
333 	.map_btf_id = &bpf_local_storage_map_btf_id[0],
334 	.map_owner_storage_ptr = task_storage_ptr,
335 };
336 
337 const struct bpf_func_proto bpf_task_storage_get_recur_proto = {
338 	.func = bpf_task_storage_get_recur,
339 	.gpl_only = false,
340 	.ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
341 	.arg1_type = ARG_CONST_MAP_PTR,
342 	.arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL,
343 	.arg2_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
344 	.arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL,
345 	.arg4_type = ARG_ANYTHING,
346 };
347 
348 const struct bpf_func_proto bpf_task_storage_get_proto = {
349 	.func = bpf_task_storage_get,
350 	.gpl_only = false,
351 	.ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
352 	.arg1_type = ARG_CONST_MAP_PTR,
353 	.arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL,
354 	.arg2_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
355 	.arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL,
356 	.arg4_type = ARG_ANYTHING,
357 };
358 
359 const struct bpf_func_proto bpf_task_storage_delete_recur_proto = {
360 	.func = bpf_task_storage_delete_recur,
361 	.gpl_only = false,
362 	.ret_type = RET_INTEGER,
363 	.arg1_type = ARG_CONST_MAP_PTR,
364 	.arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL,
365 	.arg2_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
366 };
367 
368 const struct bpf_func_proto bpf_task_storage_delete_proto = {
369 	.func = bpf_task_storage_delete,
370 	.gpl_only = false,
371 	.ret_type = RET_INTEGER,
372 	.arg1_type = ARG_CONST_MAP_PTR,
373 	.arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL,
374 	.arg2_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
375 };
376