blob: 8f65c402806d3b807f2da276dc7093bfb9752a78 [file] [log] [blame]
/* buf.c - Buffer management */
/*
* Copyright (c) 2015-2019 Intel Corporation
*
* SPDX-License-Identifier: Apache-2.0
*/
#define LOG_MODULE_NAME net_buf
#define LOG_LEVEL CONFIG_NET_BUF_LOG_LEVEL
#include <zephyr/logging/log.h>
LOG_MODULE_REGISTER(LOG_MODULE_NAME);
#include <stdio.h>
#include <errno.h>
#include <stddef.h>
#include <string.h>
#include <zephyr/sys/byteorder.h>
#include <zephyr/net/buf.h>
#if defined(CONFIG_NET_BUF_LOG)
#define NET_BUF_DBG(fmt, ...) LOG_DBG("(%p) " fmt, k_current_get(), \
##__VA_ARGS__)
#define NET_BUF_ERR(fmt, ...) LOG_ERR(fmt, ##__VA_ARGS__)
#define NET_BUF_WARN(fmt, ...) LOG_WRN(fmt, ##__VA_ARGS__)
#define NET_BUF_INFO(fmt, ...) LOG_INF(fmt, ##__VA_ARGS__)
#else
#define NET_BUF_DBG(fmt, ...)
#define NET_BUF_ERR(fmt, ...)
#define NET_BUF_WARN(fmt, ...)
#define NET_BUF_INFO(fmt, ...)
#endif /* CONFIG_NET_BUF_LOG */
#define NET_BUF_ASSERT(cond, ...) __ASSERT(cond, "" __VA_ARGS__)
#if CONFIG_NET_BUF_WARN_ALLOC_INTERVAL > 0
#define WARN_ALLOC_INTERVAL K_SECONDS(CONFIG_NET_BUF_WARN_ALLOC_INTERVAL)
#else
#define WARN_ALLOC_INTERVAL K_FOREVER
#endif
/* Linker-defined symbol bound to the static pool structs */
STRUCT_SECTION_START_EXTERN(net_buf_pool);
struct net_buf_pool *net_buf_pool_get(int id)
{
struct net_buf_pool *pool;
STRUCT_SECTION_GET(net_buf_pool, id, &pool);
return pool;
}
static int pool_id(struct net_buf_pool *pool)
{
return pool - TYPE_SECTION_START(net_buf_pool);
}
int net_buf_id(struct net_buf *buf)
{
struct net_buf_pool *pool = net_buf_pool_get(buf->pool_id);
size_t struct_size = ROUND_UP(sizeof(struct net_buf) + pool->user_data_size,
__alignof__(struct net_buf));
ptrdiff_t offset = (uint8_t *)buf - (uint8_t *)pool->__bufs;
return offset / struct_size;
}
static inline struct net_buf *pool_get_uninit(struct net_buf_pool *pool,
uint16_t uninit_count)
{
size_t struct_size = ROUND_UP(sizeof(struct net_buf) + pool->user_data_size,
__alignof__(struct net_buf));
size_t byte_offset = (pool->buf_count - uninit_count) * struct_size;
struct net_buf *buf;
buf = (struct net_buf *)(((uint8_t *)pool->__bufs) + byte_offset);
buf->pool_id = pool_id(pool);
buf->user_data_size = pool->user_data_size;
return buf;
}
void net_buf_reset(struct net_buf *buf)
{
__ASSERT_NO_MSG(buf->flags == 0U);
__ASSERT_NO_MSG(buf->frags == NULL);
net_buf_simple_reset(&buf->b);
}
static uint8_t *generic_data_ref(struct net_buf *buf, uint8_t *data)
{
uint8_t *ref_count;
ref_count = data - sizeof(void *);
(*ref_count)++;
return data;
}
static uint8_t *mem_pool_data_alloc(struct net_buf *buf, size_t *size,
k_timeout_t timeout)
{
struct net_buf_pool *buf_pool = net_buf_pool_get(buf->pool_id);
struct k_heap *pool = buf_pool->alloc->alloc_data;
uint8_t *ref_count;
/* Reserve extra space for a ref-count (uint8_t) */
void *b = k_heap_alloc(pool, sizeof(void *) + *size, timeout);
if (b == NULL) {
return NULL;
}
ref_count = (uint8_t *)b;
*ref_count = 1U;
/* Return pointer to the byte following the ref count */
return ref_count + sizeof(void *);
}
static void mem_pool_data_unref(struct net_buf *buf, uint8_t *data)
{
struct net_buf_pool *buf_pool = net_buf_pool_get(buf->pool_id);
struct k_heap *pool = buf_pool->alloc->alloc_data;
uint8_t *ref_count;
ref_count = data - sizeof(void *);
if (--(*ref_count)) {
return;
}
/* Need to copy to local variable due to alignment */
k_heap_free(pool, ref_count);
}
const struct net_buf_data_cb net_buf_var_cb = {
.alloc = mem_pool_data_alloc,
.ref = generic_data_ref,
.unref = mem_pool_data_unref,
};
static uint8_t *fixed_data_alloc(struct net_buf *buf, size_t *size,
k_timeout_t timeout)
{
struct net_buf_pool *pool = net_buf_pool_get(buf->pool_id);
const struct net_buf_pool_fixed *fixed = pool->alloc->alloc_data;
*size = pool->alloc->max_alloc_size;
return fixed->data_pool + *size * net_buf_id(buf);
}
static void fixed_data_unref(struct net_buf *buf, uint8_t *data)
{
/* Nothing needed for fixed-size data pools */
}
const struct net_buf_data_cb net_buf_fixed_cb = {
.alloc = fixed_data_alloc,
.unref = fixed_data_unref,
};
#if (K_HEAP_MEM_POOL_SIZE > 0)
static uint8_t *heap_data_alloc(struct net_buf *buf, size_t *size,
k_timeout_t timeout)
{
uint8_t *ref_count;
ref_count = k_malloc(sizeof(void *) + *size);
if (!ref_count) {
return NULL;
}
*ref_count = 1U;
return ref_count + sizeof(void *);
}
static void heap_data_unref(struct net_buf *buf, uint8_t *data)
{
uint8_t *ref_count;
ref_count = data - sizeof(void *);
if (--(*ref_count)) {
return;
}
k_free(ref_count);
}
static const struct net_buf_data_cb net_buf_heap_cb = {
.alloc = heap_data_alloc,
.ref = generic_data_ref,
.unref = heap_data_unref,
};
const struct net_buf_data_alloc net_buf_heap_alloc = {
.cb = &net_buf_heap_cb,
.max_alloc_size = 0,
};
#endif /* K_HEAP_MEM_POOL_SIZE > 0 */
static uint8_t *data_alloc(struct net_buf *buf, size_t *size, k_timeout_t timeout)
{
struct net_buf_pool *pool = net_buf_pool_get(buf->pool_id);
return pool->alloc->cb->alloc(buf, size, timeout);
}
static uint8_t *data_ref(struct net_buf *buf, uint8_t *data)
{
struct net_buf_pool *pool = net_buf_pool_get(buf->pool_id);
return pool->alloc->cb->ref(buf, data);
}
#if defined(CONFIG_NET_BUF_LOG)
struct net_buf *net_buf_alloc_len_debug(struct net_buf_pool *pool, size_t size,
k_timeout_t timeout, const char *func,
int line)
#else
struct net_buf *net_buf_alloc_len(struct net_buf_pool *pool, size_t size,
k_timeout_t timeout)
#endif
{
k_timepoint_t end = sys_timepoint_calc(timeout);
struct net_buf *buf;
k_spinlock_key_t key;
__ASSERT_NO_MSG(pool);
NET_BUF_DBG("%s():%d: pool %p size %zu", func, line, pool, size);
/* We need to prevent race conditions
* when accessing pool->uninit_count.
*/
key = k_spin_lock(&pool->lock);
/* If there are uninitialized buffers we're guaranteed to succeed
* with the allocation one way or another.
*/
if (pool->uninit_count) {
uint16_t uninit_count;
/* If this is not the first access to the pool, we can
* be opportunistic and try to fetch a previously used
* buffer from the LIFO with K_NO_WAIT.
*/
if (pool->uninit_count < pool->buf_count) {
buf = k_lifo_get(&pool->free, K_NO_WAIT);
if (buf) {
k_spin_unlock(&pool->lock, key);
goto success;
}
}
uninit_count = pool->uninit_count--;
k_spin_unlock(&pool->lock, key);
buf = pool_get_uninit(pool, uninit_count);
goto success;
}
k_spin_unlock(&pool->lock, key);
#if defined(CONFIG_NET_BUF_LOG) && (CONFIG_NET_BUF_LOG_LEVEL >= LOG_LEVEL_WRN)
if (K_TIMEOUT_EQ(timeout, K_FOREVER)) {
uint32_t ref = k_uptime_get_32();
buf = k_lifo_get(&pool->free, K_NO_WAIT);
while (!buf) {
#if defined(CONFIG_NET_BUF_POOL_USAGE)
NET_BUF_WARN("%s():%d: Pool %s low on buffers.",
func, line, pool->name);
#else
NET_BUF_WARN("%s():%d: Pool %p low on buffers.",
func, line, pool);
#endif
buf = k_lifo_get(&pool->free, WARN_ALLOC_INTERVAL);
#if defined(CONFIG_NET_BUF_POOL_USAGE)
NET_BUF_WARN("%s():%d: Pool %s blocked for %u secs",
func, line, pool->name,
(k_uptime_get_32() - ref) / MSEC_PER_SEC);
#else
NET_BUF_WARN("%s():%d: Pool %p blocked for %u secs",
func, line, pool,
(k_uptime_get_32() - ref) / MSEC_PER_SEC);
#endif
}
} else {
buf = k_lifo_get(&pool->free, timeout);
}
#else
buf = k_lifo_get(&pool->free, timeout);
#endif
if (!buf) {
NET_BUF_ERR("%s():%d: Failed to get free buffer", func, line);
return NULL;
}
success:
NET_BUF_DBG("allocated buf %p", buf);
if (size) {
#if __ASSERT_ON
size_t req_size = size;
#endif
timeout = sys_timepoint_timeout(end);
buf->__buf = data_alloc(buf, &size, timeout);
if (!buf->__buf) {
NET_BUF_ERR("%s():%d: Failed to allocate data",
func, line);
net_buf_destroy(buf);
return NULL;
}
#if __ASSERT_ON
NET_BUF_ASSERT(req_size <= size);
#endif
} else {
buf->__buf = NULL;
}
buf->ref = 1U;
buf->flags = 0U;
buf->frags = NULL;
buf->size = size;
net_buf_reset(buf);
#if defined(CONFIG_NET_BUF_POOL_USAGE)
atomic_dec(&pool->avail_count);
__ASSERT_NO_MSG(atomic_get(&pool->avail_count) >= 0);
#endif
return buf;
}
#if defined(CONFIG_NET_BUF_LOG)
struct net_buf *net_buf_alloc_fixed_debug(struct net_buf_pool *pool,
k_timeout_t timeout, const char *func,
int line)
{
return net_buf_alloc_len_debug(pool, pool->alloc->max_alloc_size, timeout, func,
line);
}
#else
struct net_buf *net_buf_alloc_fixed(struct net_buf_pool *pool,
k_timeout_t timeout)
{
return net_buf_alloc_len(pool, pool->alloc->max_alloc_size, timeout);
}
#endif
#if defined(CONFIG_NET_BUF_LOG)
struct net_buf *net_buf_alloc_with_data_debug(struct net_buf_pool *pool,
void *data, size_t size,
k_timeout_t timeout,
const char *func, int line)
#else
struct net_buf *net_buf_alloc_with_data(struct net_buf_pool *pool,
void *data, size_t size,
k_timeout_t timeout)
#endif
{
struct net_buf *buf;
#if defined(CONFIG_NET_BUF_LOG)
buf = net_buf_alloc_len_debug(pool, 0, timeout, func, line);
#else
buf = net_buf_alloc_len(pool, 0, timeout);
#endif
if (!buf) {
return NULL;
}
net_buf_simple_init_with_data(&buf->b, data, size);
buf->flags = NET_BUF_EXTERNAL_DATA;
return buf;
}
#if defined(CONFIG_NET_BUF_LOG)
struct net_buf *net_buf_get_debug(struct k_fifo *fifo, k_timeout_t timeout,
const char *func, int line)
#else
struct net_buf *net_buf_get(struct k_fifo *fifo, k_timeout_t timeout)
#endif
{
struct net_buf *buf;
NET_BUF_DBG("%s():%d: fifo %p", func, line, fifo);
buf = k_fifo_get(fifo, timeout);
if (!buf) {
return NULL;
}
NET_BUF_DBG("%s():%d: buf %p fifo %p", func, line, buf, fifo);
return buf;
}
static struct k_spinlock net_buf_slist_lock;
void net_buf_slist_put(sys_slist_t *list, struct net_buf *buf)
{
k_spinlock_key_t key;
__ASSERT_NO_MSG(list);
__ASSERT_NO_MSG(buf);
key = k_spin_lock(&net_buf_slist_lock);
sys_slist_append(list, &buf->node);
k_spin_unlock(&net_buf_slist_lock, key);
}
struct net_buf *net_buf_slist_get(sys_slist_t *list)
{
struct net_buf *buf;
k_spinlock_key_t key;
__ASSERT_NO_MSG(list);
key = k_spin_lock(&net_buf_slist_lock);
buf = (void *)sys_slist_get(list);
k_spin_unlock(&net_buf_slist_lock, key);
return buf;
}
void net_buf_put(struct k_fifo *fifo, struct net_buf *buf)
{
__ASSERT_NO_MSG(fifo);
__ASSERT_NO_MSG(buf);
k_fifo_put(fifo, buf);
}
#if defined(CONFIG_NET_BUF_LOG)
void net_buf_unref_debug(struct net_buf *buf, const char *func, int line)
#else
void net_buf_unref(struct net_buf *buf)
#endif
{
__ASSERT_NO_MSG(buf);
while (buf) {
struct net_buf *frags = buf->frags;
struct net_buf_pool *pool;
#if defined(CONFIG_NET_BUF_LOG)
if (!buf->ref) {
NET_BUF_ERR("%s():%d: buf %p double free", func, line,
buf);
return;
}
#endif
NET_BUF_DBG("buf %p ref %u pool_id %u frags %p", buf, buf->ref,
buf->pool_id, buf->frags);
if (--buf->ref > 0) {
return;
}
buf->data = NULL;
buf->frags = NULL;
pool = net_buf_pool_get(buf->pool_id);
#if defined(CONFIG_NET_BUF_POOL_USAGE)
atomic_inc(&pool->avail_count);
__ASSERT_NO_MSG(atomic_get(&pool->avail_count) <= pool->buf_count);
#endif
if (pool->destroy) {
pool->destroy(buf);
} else {
net_buf_destroy(buf);
}
buf = frags;
}
}
struct net_buf *net_buf_ref(struct net_buf *buf)
{
__ASSERT_NO_MSG(buf);
NET_BUF_DBG("buf %p (old) ref %u pool_id %u",
buf, buf->ref, buf->pool_id);
buf->ref++;
return buf;
}
struct net_buf *net_buf_clone(struct net_buf *buf, k_timeout_t timeout)
{
k_timepoint_t end = sys_timepoint_calc(timeout);
struct net_buf_pool *pool;
struct net_buf *clone;
__ASSERT_NO_MSG(buf);
pool = net_buf_pool_get(buf->pool_id);
clone = net_buf_alloc_len(pool, 0, timeout);
if (!clone) {
return NULL;
}
/* If the pool supports data referencing use that. Otherwise
* we need to allocate new data and make a copy.
*/
if (pool->alloc->cb->ref && !(buf->flags & NET_BUF_EXTERNAL_DATA)) {
clone->__buf = buf->__buf ? data_ref(buf, buf->__buf) : NULL;
clone->data = buf->data;
clone->len = buf->len;
clone->size = buf->size;
} else {
size_t size = buf->size;
timeout = sys_timepoint_timeout(end);
clone->__buf = data_alloc(clone, &size, timeout);
if (!clone->__buf || size < buf->size) {
net_buf_destroy(clone);
return NULL;
}
clone->size = size;
clone->data = clone->__buf + net_buf_headroom(buf);
net_buf_add_mem(clone, buf->data, buf->len);
}
/* user_data_size should be the same for buffers from the same pool */
__ASSERT(buf->user_data_size == clone->user_data_size, "Unexpected user data size");
memcpy(clone->user_data, buf->user_data, clone->user_data_size);
return clone;
}
int net_buf_user_data_copy(struct net_buf *dst, const struct net_buf *src)
{
__ASSERT_NO_MSG(dst);
__ASSERT_NO_MSG(src);
if (dst == src) {
return 0;
}
if (dst->user_data_size < src->user_data_size) {
return -EINVAL;
}
memcpy(dst->user_data, src->user_data, src->user_data_size);
return 0;
}
struct net_buf *net_buf_frag_last(struct net_buf *buf)
{
__ASSERT_NO_MSG(buf);
while (buf->frags) {
buf = buf->frags;
}
return buf;
}
void net_buf_frag_insert(struct net_buf *parent, struct net_buf *frag)
{
__ASSERT_NO_MSG(parent);
__ASSERT_NO_MSG(frag);
if (parent->frags) {
net_buf_frag_last(frag)->frags = parent->frags;
}
/* Take ownership of the fragment reference */
parent->frags = frag;
}
struct net_buf *net_buf_frag_add(struct net_buf *head, struct net_buf *frag)
{
__ASSERT_NO_MSG(frag);
if (!head) {
return net_buf_ref(frag);
}
net_buf_frag_insert(net_buf_frag_last(head), frag);
return head;
}
#if defined(CONFIG_NET_BUF_LOG)
struct net_buf *net_buf_frag_del_debug(struct net_buf *parent,
struct net_buf *frag,
const char *func, int line)
#else
struct net_buf *net_buf_frag_del(struct net_buf *parent, struct net_buf *frag)
#endif
{
struct net_buf *next_frag;
__ASSERT_NO_MSG(frag);
if (parent) {
__ASSERT_NO_MSG(parent->frags);
__ASSERT_NO_MSG(parent->frags == frag);
parent->frags = frag->frags;
}
next_frag = frag->frags;
frag->frags = NULL;
#if defined(CONFIG_NET_BUF_LOG)
net_buf_unref_debug(frag, func, line);
#else
net_buf_unref(frag);
#endif
return next_frag;
}
size_t net_buf_linearize(void *dst, size_t dst_len, struct net_buf *src,
size_t offset, size_t len)
{
struct net_buf *frag;
size_t to_copy;
size_t copied;
len = MIN(len, dst_len);
frag = src;
/* find the right fragment to start copying from */
while (frag && offset >= frag->len) {
offset -= frag->len;
frag = frag->frags;
}
/* traverse the fragment chain until len bytes are copied */
copied = 0;
while (frag && len > 0) {
to_copy = MIN(len, frag->len - offset);
memcpy((uint8_t *)dst + copied, frag->data + offset, to_copy);
copied += to_copy;
/* to_copy is always <= len */
len -= to_copy;
frag = frag->frags;
/* after the first iteration, this value will be 0 */
offset = 0;
}
return copied;
}
/* This helper routine will append multiple bytes, if there is no place for
* the data in current fragment then create new fragment and add it to
* the buffer. It assumes that the buffer has at least one fragment.
*/
size_t net_buf_append_bytes(struct net_buf *buf, size_t len,
const void *value, k_timeout_t timeout,
net_buf_allocator_cb allocate_cb, void *user_data)
{
struct net_buf *frag = net_buf_frag_last(buf);
size_t added_len = 0;
const uint8_t *value8 = value;
size_t max_size;
do {
uint16_t count = MIN(len, net_buf_tailroom(frag));
net_buf_add_mem(frag, value8, count);
len -= count;
added_len += count;
value8 += count;
if (len == 0) {
return added_len;
}
if (allocate_cb) {
frag = allocate_cb(timeout, user_data);
} else {
struct net_buf_pool *pool;
/* Allocate from the original pool if no callback has
* been provided.
*/
pool = net_buf_pool_get(buf->pool_id);
max_size = pool->alloc->max_alloc_size;
frag = net_buf_alloc_len(pool,
max_size ? MIN(len, max_size) : len,
timeout);
}
if (!frag) {
return added_len;
}
net_buf_frag_add(buf, frag);
} while (1);
/* Unreachable */
return 0;
}
size_t net_buf_data_match(const struct net_buf *buf, size_t offset, const void *data, size_t len)
{
const uint8_t *dptr = data;
const uint8_t *bptr;
size_t compared = 0;
size_t to_compare;
if (!buf || !data) {
return compared;
}
/* find the right fragment to start comparison */
while (buf && offset >= buf->len) {
offset -= buf->len;
buf = buf->frags;
}
while (buf && len > 0) {
bptr = buf->data + offset;
to_compare = MIN(len, buf->len - offset);
for (size_t i = 0; i < to_compare; ++i) {
if (dptr[compared] != bptr[i]) {
return compared;
}
compared++;
}
len -= to_compare;
buf = buf->frags;
offset = 0;
}
return compared;
}