kernel/bpf/queue_stack_maps.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * queue_stack_maps.c: BPF queue and stack maps
0004  *
0005  * Copyright (c) 2018 Politecnico di Torino
0006  */
0007 #include <linux/bpf.h>
0008 #include <linux/list.h>
0009 #include <linux/slab.h>
0010 #include <linux/capability.h>
0011 #include <linux/btf_ids.h>
0012 #include "percpu_freelist.h"
0013
0014 #define QUEUE_STACK_CREATE_FLAG_MASK \
0015     (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
0016
0017 struct bpf_queue_stack {
0018     struct bpf_map map;
0019     raw_spinlock_t lock;
0020     u32 head, tail;
0021     u32 size; /* max_entries + 1 */
0022
0023     char elements[] __aligned(8);
0024 };
0025
0026 static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map)
0027 {
0028     return container_of(map, struct bpf_queue_stack, map);
0029 }
0030
0031 static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs)
0032 {
0033     return qs->head == qs->tail;
0034 }
0035
0036 static bool queue_stack_map_is_full(struct bpf_queue_stack *qs)
0037 {
0038     u32 head = qs->head + 1;
0039
0040     if (unlikely(head >= qs->size))
0041         head = 0;
0042
0043     return head == qs->tail;
0044 }
0045
0046 /* Called from syscall */
0047 static int queue_stack_map_alloc_check(union bpf_attr *attr)
0048 {
0049     if (!bpf_capable())
0050         return -EPERM;
0051
0052     /* check sanity of attributes */
0053     if (attr->max_entries == 0 || attr->key_size != 0 ||
0054         attr->value_size == 0 ||
0055         attr->map_flags & ~QUEUE_STACK_CREATE_FLAG_MASK ||
0056         !bpf_map_flags_access_ok(attr->map_flags))
0057         return -EINVAL;
0058
0059     if (attr->value_size > KMALLOC_MAX_SIZE)
0060         /* if value_size is bigger, the user space won't be able to
0061          * access the elements.
0062          */
0063         return -E2BIG;
0064
0065     return 0;
0066 }
0067
0068 static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr)
0069 {
0070     int numa_node = bpf_map_attr_numa_node(attr);
0071     struct bpf_queue_stack *qs;
0072     u64 size, queue_size;
0073
0074     size = (u64) attr->max_entries + 1;
0075     queue_size = sizeof(*qs) + size * attr->value_size;
0076
0077     qs = bpf_map_area_alloc(queue_size, numa_node);
0078     if (!qs)
0079         return ERR_PTR(-ENOMEM);
0080
0081     memset(qs, 0, sizeof(*qs));
0082
0083     bpf_map_init_from_attr(&qs->map, attr);
0084
0085     qs->size = size;
0086
0087     raw_spin_lock_init(&qs->lock);
0088
0089     return &qs->map;
0090 }
0091
0092 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
0093 static void queue_stack_map_free(struct bpf_map *map)
0094 {
0095     struct bpf_queue_stack *qs = bpf_queue_stack(map);
0096
0097     bpf_map_area_free(qs);
0098 }
0099
0100 static int __queue_map_get(struct bpf_map *map, void *value, bool delete)
0101 {
0102     struct bpf_queue_stack *qs = bpf_queue_stack(map);
0103     unsigned long flags;
0104     int err = 0;
0105     void *ptr;
0106
0107     raw_spin_lock_irqsave(&qs->lock, flags);
0108
0109     if (queue_stack_map_is_empty(qs)) {
0110         memset(value, 0, qs->map.value_size);
0111         err = -ENOENT;
0112         goto out;
0113     }
0114
0115     ptr = &qs->elements[qs->tail * qs->map.value_size];
0116     memcpy(value, ptr, qs->map.value_size);
0117
0118     if (delete) {
0119         if (unlikely(++qs->tail >= qs->size))
0120             qs->tail = 0;
0121     }
0122
0123 out:
0124     raw_spin_unlock_irqrestore(&qs->lock, flags);
0125     return err;
0126 }
0127
0128
0129 static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
0130 {
0131     struct bpf_queue_stack *qs = bpf_queue_stack(map);
0132     unsigned long flags;
0133     int err = 0;
0134     void *ptr;
0135     u32 index;
0136
0137     raw_spin_lock_irqsave(&qs->lock, flags);
0138
0139     if (queue_stack_map_is_empty(qs)) {
0140         memset(value, 0, qs->map.value_size);
0141         err = -ENOENT;
0142         goto out;
0143     }
0144
0145     index = qs->head - 1;
0146     if (unlikely(index >= qs->size))
0147         index = qs->size - 1;
0148
0149     ptr = &qs->elements[index * qs->map.value_size];
0150     memcpy(value, ptr, qs->map.value_size);
0151
0152     if (delete)
0153         qs->head = index;
0154
0155 out:
0156     raw_spin_unlock_irqrestore(&qs->lock, flags);
0157     return err;
0158 }
0159
0160 /* Called from syscall or from eBPF program */
0161 static int queue_map_peek_elem(struct bpf_map *map, void *value)
0162 {
0163     return __queue_map_get(map, value, false);
0164 }
0165
0166 /* Called from syscall or from eBPF program */
0167 static int stack_map_peek_elem(struct bpf_map *map, void *value)
0168 {
0169     return __stack_map_get(map, value, false);
0170 }
0171
0172 /* Called from syscall or from eBPF program */
0173 static int queue_map_pop_elem(struct bpf_map *map, void *value)
0174 {
0175     return __queue_map_get(map, value, true);
0176 }
0177
0178 /* Called from syscall or from eBPF program */
0179 static int stack_map_pop_elem(struct bpf_map *map, void *value)
0180 {
0181     return __stack_map_get(map, value, true);
0182 }
0183
0184 /* Called from syscall or from eBPF program */
0185 static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
0186                      u64 flags)
0187 {
0188     struct bpf_queue_stack *qs = bpf_queue_stack(map);
0189     unsigned long irq_flags;
0190     int err = 0;
0191     void *dst;
0192
0193     /* BPF_EXIST is used to force making room for a new element in case the
0194      * map is full
0195      */
0196     bool replace = (flags & BPF_EXIST);
0197
0198     /* Check supported flags for queue and stack maps */
0199     if (flags & BPF_NOEXIST || flags > BPF_EXIST)
0200         return -EINVAL;
0201
0202     raw_spin_lock_irqsave(&qs->lock, irq_flags);
0203
0204     if (queue_stack_map_is_full(qs)) {
0205         if (!replace) {
0206             err = -E2BIG;
0207             goto out;
0208         }
0209         /* advance tail pointer to overwrite oldest element */
0210         if (unlikely(++qs->tail >= qs->size))
0211             qs->tail = 0;
0212     }
0213
0214     dst = &qs->elements[qs->head * qs->map.value_size];
0215     memcpy(dst, value, qs->map.value_size);
0216
0217     if (unlikely(++qs->head >= qs->size))
0218         qs->head = 0;
0219
0220 out:
0221     raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
0222     return err;
0223 }
0224
0225 /* Called from syscall or from eBPF program */
0226 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
0227 {
0228     return NULL;
0229 }
0230
0231 /* Called from syscall or from eBPF program */
0232 static int queue_stack_map_update_elem(struct bpf_map *map, void *key,
0233                        void *value, u64 flags)
0234 {
0235     return -EINVAL;
0236 }
0237
0238 /* Called from syscall or from eBPF program */
0239 static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
0240 {
0241     return -EINVAL;
0242 }
0243
0244 /* Called from syscall */
0245 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
0246                     void *next_key)
0247 {
0248     return -EINVAL;
0249 }
0250
0251 BTF_ID_LIST_SINGLE(queue_map_btf_ids, struct, bpf_queue_stack)
0252 const struct bpf_map_ops queue_map_ops = {
0253     .map_meta_equal = bpf_map_meta_equal,
0254     .map_alloc_check = queue_stack_map_alloc_check,
0255     .map_alloc = queue_stack_map_alloc,
0256     .map_free = queue_stack_map_free,
0257     .map_lookup_elem = queue_stack_map_lookup_elem,
0258     .map_update_elem = queue_stack_map_update_elem,
0259     .map_delete_elem = queue_stack_map_delete_elem,
0260     .map_push_elem = queue_stack_map_push_elem,
0261     .map_pop_elem = queue_map_pop_elem,
0262     .map_peek_elem = queue_map_peek_elem,
0263     .map_get_next_key = queue_stack_map_get_next_key,
0264     .map_btf_id = &queue_map_btf_ids[0],
0265 };
0266
0267 const struct bpf_map_ops stack_map_ops = {
0268     .map_meta_equal = bpf_map_meta_equal,
0269     .map_alloc_check = queue_stack_map_alloc_check,
0270     .map_alloc = queue_stack_map_alloc,
0271     .map_free = queue_stack_map_free,
0272     .map_lookup_elem = queue_stack_map_lookup_elem,
0273     .map_update_elem = queue_stack_map_update_elem,
0274     .map_delete_elem = queue_stack_map_delete_elem,
0275     .map_push_elem = queue_stack_map_push_elem,
0276     .map_pop_elem = stack_map_pop_elem,
0277     .map_peek_elem = stack_map_peek_elem,
0278     .map_get_next_key = queue_stack_map_get_next_key,
0279     .map_btf_id = &queue_map_btf_ids[0],
0280 };