xref: /linux/kernel/bpf/queue_stack_maps.c (revision f2835adf8afb2cea248dd10d6eb0444c34b3b51b)
1  // SPDX-License-Identifier: GPL-2.0
2  /*
3   * queue_stack_maps.c: BPF queue and stack maps
4   *
5   * Copyright (c) 2018 Politecnico di Torino
6   */
7  #include <linux/bpf.h>
8  #include <linux/list.h>
9  #include <linux/slab.h>
10  #include <linux/capability.h>
11  #include "percpu_freelist.h"
12  
13  #define QUEUE_STACK_CREATE_FLAG_MASK \
14  	(BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
15  
16  struct bpf_queue_stack {
17  	struct bpf_map map;
18  	raw_spinlock_t lock;
19  	u32 head, tail;
20  	u32 size; /* max_entries + 1 */
21  
22  	char elements[0] __aligned(8);
23  };
24  
25  static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map)
26  {
27  	return container_of(map, struct bpf_queue_stack, map);
28  }
29  
30  static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs)
31  {
32  	return qs->head == qs->tail;
33  }
34  
35  static bool queue_stack_map_is_full(struct bpf_queue_stack *qs)
36  {
37  	u32 head = qs->head + 1;
38  
39  	if (unlikely(head >= qs->size))
40  		head = 0;
41  
42  	return head == qs->tail;
43  }
44  
45  /* Called from syscall */
46  static int queue_stack_map_alloc_check(union bpf_attr *attr)
47  {
48  	if (!capable(CAP_SYS_ADMIN))
49  		return -EPERM;
50  
51  	/* check sanity of attributes */
52  	if (attr->max_entries == 0 || attr->key_size != 0 ||
53  	    attr->value_size == 0 ||
54  	    attr->map_flags & ~QUEUE_STACK_CREATE_FLAG_MASK ||
55  	    !bpf_map_flags_access_ok(attr->map_flags))
56  		return -EINVAL;
57  
58  	if (attr->value_size > KMALLOC_MAX_SIZE)
59  		/* if value_size is bigger, the user space won't be able to
60  		 * access the elements.
61  		 */
62  		return -E2BIG;
63  
64  	return 0;
65  }
66  
67  static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr)
68  {
69  	int ret, numa_node = bpf_map_attr_numa_node(attr);
70  	struct bpf_map_memory mem = {0};
71  	struct bpf_queue_stack *qs;
72  	u64 size, queue_size, cost;
73  
74  	size = (u64) attr->max_entries + 1;
75  	cost = queue_size = sizeof(*qs) + size * attr->value_size;
76  
77  	ret = bpf_map_charge_init(&mem, cost);
78  	if (ret < 0)
79  		return ERR_PTR(ret);
80  
81  	qs = bpf_map_area_alloc(queue_size, numa_node);
82  	if (!qs) {
83  		bpf_map_charge_finish(&mem);
84  		return ERR_PTR(-ENOMEM);
85  	}
86  
87  	memset(qs, 0, sizeof(*qs));
88  
89  	bpf_map_init_from_attr(&qs->map, attr);
90  
91  	bpf_map_charge_move(&qs->map.memory, &mem);
92  	qs->size = size;
93  
94  	raw_spin_lock_init(&qs->lock);
95  
96  	return &qs->map;
97  }
98  
99  /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
100  static void queue_stack_map_free(struct bpf_map *map)
101  {
102  	struct bpf_queue_stack *qs = bpf_queue_stack(map);
103  
104  	/* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
105  	 * so the programs (can be more than one that used this map) were
106  	 * disconnected from events. Wait for outstanding critical sections in
107  	 * these programs to complete
108  	 */
109  	synchronize_rcu();
110  
111  	bpf_map_area_free(qs);
112  }
113  
114  static int __queue_map_get(struct bpf_map *map, void *value, bool delete)
115  {
116  	struct bpf_queue_stack *qs = bpf_queue_stack(map);
117  	unsigned long flags;
118  	int err = 0;
119  	void *ptr;
120  
121  	raw_spin_lock_irqsave(&qs->lock, flags);
122  
123  	if (queue_stack_map_is_empty(qs)) {
124  		memset(value, 0, qs->map.value_size);
125  		err = -ENOENT;
126  		goto out;
127  	}
128  
129  	ptr = &qs->elements[qs->tail * qs->map.value_size];
130  	memcpy(value, ptr, qs->map.value_size);
131  
132  	if (delete) {
133  		if (unlikely(++qs->tail >= qs->size))
134  			qs->tail = 0;
135  	}
136  
137  out:
138  	raw_spin_unlock_irqrestore(&qs->lock, flags);
139  	return err;
140  }
141  
142  
143  static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
144  {
145  	struct bpf_queue_stack *qs = bpf_queue_stack(map);
146  	unsigned long flags;
147  	int err = 0;
148  	void *ptr;
149  	u32 index;
150  
151  	raw_spin_lock_irqsave(&qs->lock, flags);
152  
153  	if (queue_stack_map_is_empty(qs)) {
154  		memset(value, 0, qs->map.value_size);
155  		err = -ENOENT;
156  		goto out;
157  	}
158  
159  	index = qs->head - 1;
160  	if (unlikely(index >= qs->size))
161  		index = qs->size - 1;
162  
163  	ptr = &qs->elements[index * qs->map.value_size];
164  	memcpy(value, ptr, qs->map.value_size);
165  
166  	if (delete)
167  		qs->head = index;
168  
169  out:
170  	raw_spin_unlock_irqrestore(&qs->lock, flags);
171  	return err;
172  }
173  
174  /* Called from syscall or from eBPF program */
175  static int queue_map_peek_elem(struct bpf_map *map, void *value)
176  {
177  	return __queue_map_get(map, value, false);
178  }
179  
180  /* Called from syscall or from eBPF program */
181  static int stack_map_peek_elem(struct bpf_map *map, void *value)
182  {
183  	return __stack_map_get(map, value, false);
184  }
185  
186  /* Called from syscall or from eBPF program */
187  static int queue_map_pop_elem(struct bpf_map *map, void *value)
188  {
189  	return __queue_map_get(map, value, true);
190  }
191  
192  /* Called from syscall or from eBPF program */
193  static int stack_map_pop_elem(struct bpf_map *map, void *value)
194  {
195  	return __stack_map_get(map, value, true);
196  }
197  
198  /* Called from syscall or from eBPF program */
199  static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
200  				     u64 flags)
201  {
202  	struct bpf_queue_stack *qs = bpf_queue_stack(map);
203  	unsigned long irq_flags;
204  	int err = 0;
205  	void *dst;
206  
207  	/* BPF_EXIST is used to force making room for a new element in case the
208  	 * map is full
209  	 */
210  	bool replace = (flags & BPF_EXIST);
211  
212  	/* Check supported flags for queue and stack maps */
213  	if (flags & BPF_NOEXIST || flags > BPF_EXIST)
214  		return -EINVAL;
215  
216  	raw_spin_lock_irqsave(&qs->lock, irq_flags);
217  
218  	if (queue_stack_map_is_full(qs)) {
219  		if (!replace) {
220  			err = -E2BIG;
221  			goto out;
222  		}
223  		/* advance tail pointer to overwrite oldest element */
224  		if (unlikely(++qs->tail >= qs->size))
225  			qs->tail = 0;
226  	}
227  
228  	dst = &qs->elements[qs->head * qs->map.value_size];
229  	memcpy(dst, value, qs->map.value_size);
230  
231  	if (unlikely(++qs->head >= qs->size))
232  		qs->head = 0;
233  
234  out:
235  	raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
236  	return err;
237  }
238  
239  /* Called from syscall or from eBPF program */
240  static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
241  {
242  	return NULL;
243  }
244  
245  /* Called from syscall or from eBPF program */
246  static int queue_stack_map_update_elem(struct bpf_map *map, void *key,
247  				       void *value, u64 flags)
248  {
249  	return -EINVAL;
250  }
251  
252  /* Called from syscall or from eBPF program */
253  static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
254  {
255  	return -EINVAL;
256  }
257  
258  /* Called from syscall */
259  static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
260  					void *next_key)
261  {
262  	return -EINVAL;
263  }
264  
265  const struct bpf_map_ops queue_map_ops = {
266  	.map_alloc_check = queue_stack_map_alloc_check,
267  	.map_alloc = queue_stack_map_alloc,
268  	.map_free = queue_stack_map_free,
269  	.map_lookup_elem = queue_stack_map_lookup_elem,
270  	.map_update_elem = queue_stack_map_update_elem,
271  	.map_delete_elem = queue_stack_map_delete_elem,
272  	.map_push_elem = queue_stack_map_push_elem,
273  	.map_pop_elem = queue_map_pop_elem,
274  	.map_peek_elem = queue_map_peek_elem,
275  	.map_get_next_key = queue_stack_map_get_next_key,
276  };
277  
278  const struct bpf_map_ops stack_map_ops = {
279  	.map_alloc_check = queue_stack_map_alloc_check,
280  	.map_alloc = queue_stack_map_alloc,
281  	.map_free = queue_stack_map_free,
282  	.map_lookup_elem = queue_stack_map_lookup_elem,
283  	.map_update_elem = queue_stack_map_update_elem,
284  	.map_delete_elem = queue_stack_map_delete_elem,
285  	.map_push_elem = queue_stack_map_push_elem,
286  	.map_pop_elem = stack_map_pop_elem,
287  	.map_peek_elem = stack_map_peek_elem,
288  	.map_get_next_key = queue_stack_map_get_next_key,
289  };
290