xref: /openbmc/linux/kernel/bpf/bpf_task_storage.c (revision 0b26ca68)
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/bpf_lsm.h>
19 #include <linux/btf_ids.h>
20 #include <linux/fdtable.h>
21 
22 DEFINE_BPF_STORAGE_CACHE(task_cache);
23 
24 static struct bpf_local_storage __rcu **task_storage_ptr(void *owner)
25 {
26 	struct task_struct *task = owner;
27 	struct bpf_storage_blob *bsb;
28 
29 	bsb = bpf_task(task);
30 	if (!bsb)
31 		return NULL;
32 	return &bsb->storage;
33 }
34 
35 static struct bpf_local_storage_data *
36 task_storage_lookup(struct task_struct *task, struct bpf_map *map,
37 		    bool cacheit_lockit)
38 {
39 	struct bpf_local_storage *task_storage;
40 	struct bpf_local_storage_map *smap;
41 	struct bpf_storage_blob *bsb;
42 
43 	bsb = bpf_task(task);
44 	if (!bsb)
45 		return NULL;
46 
47 	task_storage = rcu_dereference(bsb->storage);
48 	if (!task_storage)
49 		return NULL;
50 
51 	smap = (struct bpf_local_storage_map *)map;
52 	return bpf_local_storage_lookup(task_storage, smap, cacheit_lockit);
53 }
54 
55 void bpf_task_storage_free(struct task_struct *task)
56 {
57 	struct bpf_local_storage_elem *selem;
58 	struct bpf_local_storage *local_storage;
59 	bool free_task_storage = false;
60 	struct bpf_storage_blob *bsb;
61 	struct hlist_node *n;
62 
63 	bsb = bpf_task(task);
64 	if (!bsb)
65 		return;
66 
67 	rcu_read_lock();
68 
69 	local_storage = rcu_dereference(bsb->storage);
70 	if (!local_storage) {
71 		rcu_read_unlock();
72 		return;
73 	}
74 
75 	/* Neither the bpf_prog nor the bpf-map's syscall
76 	 * could be modifying the local_storage->list now.
77 	 * Thus, no elem can be added-to or deleted-from the
78 	 * local_storage->list by the bpf_prog or by the bpf-map's syscall.
79 	 *
80 	 * It is racing with bpf_local_storage_map_free() alone
81 	 * when unlinking elem from the local_storage->list and
82 	 * the map's bucket->list.
83 	 */
84 	raw_spin_lock_bh(&local_storage->lock);
85 	hlist_for_each_entry_safe(selem, n, &local_storage->list, snode) {
86 		/* Always unlink from map before unlinking from
87 		 * local_storage.
88 		 */
89 		bpf_selem_unlink_map(selem);
90 		free_task_storage = bpf_selem_unlink_storage_nolock(
91 			local_storage, selem, false);
92 	}
93 	raw_spin_unlock_bh(&local_storage->lock);
94 	rcu_read_unlock();
95 
96 	/* free_task_storage should always be true as long as
97 	 * local_storage->list was non-empty.
98 	 */
99 	if (free_task_storage)
100 		kfree_rcu(local_storage, rcu);
101 }
102 
103 static void *bpf_pid_task_storage_lookup_elem(struct bpf_map *map, void *key)
104 {
105 	struct bpf_local_storage_data *sdata;
106 	struct task_struct *task;
107 	unsigned int f_flags;
108 	struct pid *pid;
109 	int fd, err;
110 
111 	fd = *(int *)key;
112 	pid = pidfd_get_pid(fd, &f_flags);
113 	if (IS_ERR(pid))
114 		return ERR_CAST(pid);
115 
116 	/* We should be in an RCU read side critical section, it should be safe
117 	 * to call pid_task.
118 	 */
119 	WARN_ON_ONCE(!rcu_read_lock_held());
120 	task = pid_task(pid, PIDTYPE_PID);
121 	if (!task) {
122 		err = -ENOENT;
123 		goto out;
124 	}
125 
126 	sdata = task_storage_lookup(task, map, true);
127 	put_pid(pid);
128 	return sdata ? sdata->data : NULL;
129 out:
130 	put_pid(pid);
131 	return ERR_PTR(err);
132 }
133 
134 static int bpf_pid_task_storage_update_elem(struct bpf_map *map, void *key,
135 					    void *value, u64 map_flags)
136 {
137 	struct bpf_local_storage_data *sdata;
138 	struct task_struct *task;
139 	unsigned int f_flags;
140 	struct pid *pid;
141 	int fd, err;
142 
143 	fd = *(int *)key;
144 	pid = pidfd_get_pid(fd, &f_flags);
145 	if (IS_ERR(pid))
146 		return PTR_ERR(pid);
147 
148 	/* We should be in an RCU read side critical section, it should be safe
149 	 * to call pid_task.
150 	 */
151 	WARN_ON_ONCE(!rcu_read_lock_held());
152 	task = pid_task(pid, PIDTYPE_PID);
153 	if (!task || !task_storage_ptr(task)) {
154 		err = -ENOENT;
155 		goto out;
156 	}
157 
158 	sdata = bpf_local_storage_update(
159 		task, (struct bpf_local_storage_map *)map, value, map_flags);
160 
161 	err = PTR_ERR_OR_ZERO(sdata);
162 out:
163 	put_pid(pid);
164 	return err;
165 }
166 
167 static int task_storage_delete(struct task_struct *task, struct bpf_map *map)
168 {
169 	struct bpf_local_storage_data *sdata;
170 
171 	sdata = task_storage_lookup(task, map, false);
172 	if (!sdata)
173 		return -ENOENT;
174 
175 	bpf_selem_unlink(SELEM(sdata));
176 
177 	return 0;
178 }
179 
180 static int bpf_pid_task_storage_delete_elem(struct bpf_map *map, void *key)
181 {
182 	struct task_struct *task;
183 	unsigned int f_flags;
184 	struct pid *pid;
185 	int fd, err;
186 
187 	fd = *(int *)key;
188 	pid = pidfd_get_pid(fd, &f_flags);
189 	if (IS_ERR(pid))
190 		return PTR_ERR(pid);
191 
192 	/* We should be in an RCU read side critical section, it should be safe
193 	 * to call pid_task.
194 	 */
195 	WARN_ON_ONCE(!rcu_read_lock_held());
196 	task = pid_task(pid, PIDTYPE_PID);
197 	if (!task) {
198 		err = -ENOENT;
199 		goto out;
200 	}
201 
202 	err = task_storage_delete(task, map);
203 out:
204 	put_pid(pid);
205 	return err;
206 }
207 
208 BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
209 	   task, void *, value, u64, flags)
210 {
211 	struct bpf_local_storage_data *sdata;
212 
213 	if (flags & ~(BPF_LOCAL_STORAGE_GET_F_CREATE))
214 		return (unsigned long)NULL;
215 
216 	/* explicitly check that the task_storage_ptr is not
217 	 * NULL as task_storage_lookup returns NULL in this case and
218 	 * bpf_local_storage_update expects the owner to have a
219 	 * valid storage pointer.
220 	 */
221 	if (!task || !task_storage_ptr(task))
222 		return (unsigned long)NULL;
223 
224 	sdata = task_storage_lookup(task, map, true);
225 	if (sdata)
226 		return (unsigned long)sdata->data;
227 
228 	/* This helper must only be called from places where the lifetime of the task
229 	 * is guaranteed. Either by being refcounted or by being protected
230 	 * by an RCU read-side critical section.
231 	 */
232 	if (flags & BPF_LOCAL_STORAGE_GET_F_CREATE) {
233 		sdata = bpf_local_storage_update(
234 			task, (struct bpf_local_storage_map *)map, value,
235 			BPF_NOEXIST);
236 		return IS_ERR(sdata) ? (unsigned long)NULL :
237 					     (unsigned long)sdata->data;
238 	}
239 
240 	return (unsigned long)NULL;
241 }
242 
243 BPF_CALL_2(bpf_task_storage_delete, struct bpf_map *, map, struct task_struct *,
244 	   task)
245 {
246 	if (!task)
247 		return -EINVAL;
248 
249 	/* This helper must only be called from places where the lifetime of the task
250 	 * is guaranteed. Either by being refcounted or by being protected
251 	 * by an RCU read-side critical section.
252 	 */
253 	return task_storage_delete(task, map);
254 }
255 
256 static int notsupp_get_next_key(struct bpf_map *map, void *key, void *next_key)
257 {
258 	return -ENOTSUPP;
259 }
260 
261 static struct bpf_map *task_storage_map_alloc(union bpf_attr *attr)
262 {
263 	struct bpf_local_storage_map *smap;
264 
265 	smap = bpf_local_storage_map_alloc(attr);
266 	if (IS_ERR(smap))
267 		return ERR_CAST(smap);
268 
269 	smap->cache_idx = bpf_local_storage_cache_idx_get(&task_cache);
270 	return &smap->map;
271 }
272 
273 static void task_storage_map_free(struct bpf_map *map)
274 {
275 	struct bpf_local_storage_map *smap;
276 
277 	smap = (struct bpf_local_storage_map *)map;
278 	bpf_local_storage_cache_idx_free(&task_cache, smap->cache_idx);
279 	bpf_local_storage_map_free(smap);
280 }
281 
282 static int task_storage_map_btf_id;
283 const struct bpf_map_ops task_storage_map_ops = {
284 	.map_meta_equal = bpf_map_meta_equal,
285 	.map_alloc_check = bpf_local_storage_map_alloc_check,
286 	.map_alloc = task_storage_map_alloc,
287 	.map_free = task_storage_map_free,
288 	.map_get_next_key = notsupp_get_next_key,
289 	.map_lookup_elem = bpf_pid_task_storage_lookup_elem,
290 	.map_update_elem = bpf_pid_task_storage_update_elem,
291 	.map_delete_elem = bpf_pid_task_storage_delete_elem,
292 	.map_check_btf = bpf_local_storage_map_check_btf,
293 	.map_btf_name = "bpf_local_storage_map",
294 	.map_btf_id = &task_storage_map_btf_id,
295 	.map_owner_storage_ptr = task_storage_ptr,
296 };
297 
298 BTF_ID_LIST_SINGLE(bpf_task_storage_btf_ids, struct, task_struct)
299 
300 const struct bpf_func_proto bpf_task_storage_get_proto = {
301 	.func = bpf_task_storage_get,
302 	.gpl_only = false,
303 	.ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
304 	.arg1_type = ARG_CONST_MAP_PTR,
305 	.arg2_type = ARG_PTR_TO_BTF_ID,
306 	.arg2_btf_id = &bpf_task_storage_btf_ids[0],
307 	.arg3_type = ARG_PTR_TO_MAP_VALUE_OR_NULL,
308 	.arg4_type = ARG_ANYTHING,
309 };
310 
311 const struct bpf_func_proto bpf_task_storage_delete_proto = {
312 	.func = bpf_task_storage_delete,
313 	.gpl_only = false,
314 	.ret_type = RET_INTEGER,
315 	.arg1_type = ARG_CONST_MAP_PTR,
316 	.arg2_type = ARG_PTR_TO_BTF_ID,
317 	.arg2_btf_id = &bpf_task_storage_btf_ids[0],
318 };
319