linux-stable/include/linux/bpf_local_storage.h

172 lines
5.5 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0 */
/*
* Copyright (c) 2019 Facebook
* Copyright 2020 Google LLC.
*/
#ifndef _BPF_LOCAL_STORAGE_H
#define _BPF_LOCAL_STORAGE_H
#include <linux/bpf.h>
#include <linux/filter.h>
#include <linux/rculist.h>
#include <linux/list.h>
#include <linux/hash.h>
#include <linux/types.h>
#include <uapi/linux/btf.h>
#define BPF_LOCAL_STORAGE_CACHE_SIZE 16
bpf: Allow bpf_local_storage to be used by sleepable programs Other maps like hashmaps are already available to sleepable programs. Sleepable BPF programs run under trace RCU. Allow task, sk and inode storage to be used from sleepable programs. This allows sleepable and non-sleepable programs to provide shareable annotations on kernel objects. Sleepable programs run in trace RCU where as non-sleepable programs run in a normal RCU critical section i.e. __bpf_prog_enter{_sleepable} and __bpf_prog_exit{_sleepable}) (rcu_read_lock or rcu_read_lock_trace). In order to make the local storage maps accessible to both sleepable and non-sleepable programs, one needs to call both call_rcu_tasks_trace and call_rcu to wait for both trace and classical RCU grace periods to expire before freeing memory. Paul's work on call_rcu_tasks_trace allows us to have per CPU queueing for call_rcu_tasks_trace. This behaviour can be achieved by setting rcupdate.rcu_task_enqueue_lim=<num_cpus> boot parameter. In light of these new performance changes and to keep the local storage code simple, avoid adding a new flag for sleepable maps / local storage to select the RCU synchronization (trace / classical). Also, update the dereferencing of the pointers to use rcu_derference_check (with either the trace or normal RCU locks held) with a common bpf_rcu_lock_held helper method. Signed-off-by: KP Singh <kpsingh@kernel.org> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Martin KaFai Lau <kafai@fb.com> Link: https://lore.kernel.org/bpf/20211224152916.1550677-2-kpsingh@kernel.org
2021-12-24 15:29:15 +00:00
#define bpf_rcu_lock_held() \
(rcu_read_lock_held() || rcu_read_lock_trace_held() || \
rcu_read_lock_bh_held())
struct bpf_local_storage_map_bucket {
struct hlist_head list;
raw_spinlock_t lock;
};
/* Thp map is not the primary owner of a bpf_local_storage_elem.
* Instead, the container object (eg. sk->sk_bpf_storage) is.
*
* The map (bpf_local_storage_map) is for two purposes
* 1. Define the size of the "local storage". It is
* the map's value_size.
*
* 2. Maintain a list to keep track of all elems such
* that they can be cleaned up during the map destruction.
*
* When a bpf local storage is being looked up for a
* particular object, the "bpf_map" pointer is actually used
* as the "key" to search in the list of elem in
* the respective bpf_local_storage owned by the object.
*
* e.g. sk->sk_bpf_storage is the mini-map with the "bpf_map" pointer
* as the searching key.
*/
struct bpf_local_storage_map {
struct bpf_map map;
/* Lookup elem does not require accessing the map.
*
* Updating/Deleting requires a bucket lock to
* link/unlink the elem from the map. Having
* multiple buckets to improve contention.
*/
struct bpf_local_storage_map_bucket *buckets;
u32 bucket_log;
u16 elem_size;
u16 cache_idx;
};
struct bpf_local_storage_data {
/* smap is used as the searching key when looking up
* from the object's bpf_local_storage.
*
* Put it in the same cacheline as the data to minimize
* the number of cachelines accessed during the cache hit case.
*/
struct bpf_local_storage_map __rcu *smap;
u8 data[] __aligned(8);
};
/* Linked to bpf_local_storage and bpf_local_storage_map */
struct bpf_local_storage_elem {
struct hlist_node map_node; /* Linked to bpf_local_storage_map */
struct hlist_node snode; /* Linked to bpf_local_storage */
struct bpf_local_storage __rcu *local_storage;
struct rcu_head rcu;
/* 8 bytes hole */
/* The data is stored in another cacheline to minimize
* the number of cachelines access during a cache hit.
*/
struct bpf_local_storage_data sdata ____cacheline_aligned;
};
struct bpf_local_storage {
struct bpf_local_storage_data __rcu *cache[BPF_LOCAL_STORAGE_CACHE_SIZE];
struct bpf_local_storage_map __rcu *smap;
struct hlist_head list; /* List of bpf_local_storage_elem */
void *owner; /* The object that owns the above "list" of
* bpf_local_storage_elem.
*/
struct rcu_head rcu;
raw_spinlock_t lock; /* Protect adding/removing from the "list" */
};
/* U16_MAX is much more than enough for sk local storage
* considering a tcp_sock is ~2k.
*/
#define BPF_LOCAL_STORAGE_MAX_VALUE_SIZE \
min_t(u32, \
(KMALLOC_MAX_SIZE - MAX_BPF_STACK - \
sizeof(struct bpf_local_storage_elem)), \
(U16_MAX - sizeof(struct bpf_local_storage_elem)))
#define SELEM(_SDATA) \
container_of((_SDATA), struct bpf_local_storage_elem, sdata)
#define SDATA(_SELEM) (&(_SELEM)->sdata)
#define BPF_LOCAL_STORAGE_CACHE_SIZE 16
struct bpf_local_storage_cache {
spinlock_t idx_lock;
u64 idx_usage_counts[BPF_LOCAL_STORAGE_CACHE_SIZE];
};
#define DEFINE_BPF_STORAGE_CACHE(name) \
static struct bpf_local_storage_cache name = { \
.idx_lock = __SPIN_LOCK_UNLOCKED(name.idx_lock), \
}
/* Helper functions for bpf_local_storage */
int bpf_local_storage_map_alloc_check(union bpf_attr *attr);
struct bpf_map *
bpf_local_storage_map_alloc(union bpf_attr *attr,
struct bpf_local_storage_cache *cache);
struct bpf_local_storage_data *
bpf_local_storage_lookup(struct bpf_local_storage *local_storage,
struct bpf_local_storage_map *smap,
bool cacheit_lockit);
void bpf_local_storage_destroy(struct bpf_local_storage *local_storage);
void bpf_local_storage_map_free(struct bpf_map *map,
struct bpf_local_storage_cache *cache,
int __percpu *busy_counter);
int bpf_local_storage_map_check_btf(const struct bpf_map *map,
const struct btf *btf,
const struct btf_type *key_type,
const struct btf_type *value_type);
void bpf_selem_link_storage_nolock(struct bpf_local_storage *local_storage,
struct bpf_local_storage_elem *selem);
bpf: Repurpose use_trace_rcu to reuse_now in bpf_local_storage This patch re-purpose the use_trace_rcu to mean if the freed memory can be reused immediately or not. The use_trace_rcu is renamed to reuse_now. Other than the boolean test is reversed, it should be a no-op. The following explains the reason for the rename and how it will be used in a later patch. In a later patch, bpf_mem_cache_alloc/free will be used in the bpf_local_storage. The bpf mem allocator will reuse the freed memory immediately. Some of the free paths in bpf_local_storage does not support memory to be reused immediately. These paths are the "delete" elem cases from the bpf_*_storage_delete() helper and the map_delete_elem() syscall. Note that "delete" elem before the owner's (sk/task/cgrp/inode) lifetime ended is not the common usage for the local storage. The common free path, bpf_local_storage_destroy(), can reuse the memory immediately. This common path means the storage stays with its owner until the owner is destroyed. The above mentioned "delete" elem paths that cannot reuse immediately always has the 'use_trace_rcu == true'. The cases that is safe for immediate reuse always have 'use_trace_rcu == false'. Instead of adding another arg in a later patch, this patch re-purpose this arg to reuse_now and have the test logic reversed. In a later patch, 'reuse_now == true' will free to the bpf_mem_cache_free() where the memory can be reused immediately. 'reuse_now == false' will go through the call_rcu_tasks_trace(). Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org> Link: https://lore.kernel.org/r/20230308065936.1550103-7-martin.lau@linux.dev Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 06:59:25 +00:00
void bpf_selem_unlink(struct bpf_local_storage_elem *selem, bool reuse_now);
void bpf_selem_link_map(struct bpf_local_storage_map *smap,
struct bpf_local_storage_elem *selem);
struct bpf_local_storage_elem *
bpf_selem_alloc(struct bpf_local_storage_map *smap, void *owner, void *value,
bool charge_mem, gfp_t gfp_flags);
void bpf_selem_free(struct bpf_local_storage_elem *selem,
struct bpf_local_storage_map *smap,
bool reuse_now);
int
bpf_local_storage_alloc(void *owner,
struct bpf_local_storage_map *smap,
struct bpf_local_storage_elem *first_selem,
gfp_t gfp_flags);
struct bpf_local_storage_data *
bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
void *value, u64 map_flags, gfp_t gfp_flags);
u64 bpf_local_storage_map_mem_usage(const struct bpf_map *map);
bpf: Allow bpf_local_storage to be used by sleepable programs Other maps like hashmaps are already available to sleepable programs. Sleepable BPF programs run under trace RCU. Allow task, sk and inode storage to be used from sleepable programs. This allows sleepable and non-sleepable programs to provide shareable annotations on kernel objects. Sleepable programs run in trace RCU where as non-sleepable programs run in a normal RCU critical section i.e. __bpf_prog_enter{_sleepable} and __bpf_prog_exit{_sleepable}) (rcu_read_lock or rcu_read_lock_trace). In order to make the local storage maps accessible to both sleepable and non-sleepable programs, one needs to call both call_rcu_tasks_trace and call_rcu to wait for both trace and classical RCU grace periods to expire before freeing memory. Paul's work on call_rcu_tasks_trace allows us to have per CPU queueing for call_rcu_tasks_trace. This behaviour can be achieved by setting rcupdate.rcu_task_enqueue_lim=<num_cpus> boot parameter. In light of these new performance changes and to keep the local storage code simple, avoid adding a new flag for sleepable maps / local storage to select the RCU synchronization (trace / classical). Also, update the dereferencing of the pointers to use rcu_derference_check (with either the trace or normal RCU locks held) with a common bpf_rcu_lock_held helper method. Signed-off-by: KP Singh <kpsingh@kernel.org> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Martin KaFai Lau <kafai@fb.com> Link: https://lore.kernel.org/bpf/20211224152916.1550677-2-kpsingh@kernel.org
2021-12-24 15:29:15 +00:00
#endif /* _BPF_LOCAL_STORAGE_H */