[linux-block.git] / net / core / sock_reuseport.c

// SPDX-License-Identifier: GPL-2.0
/*
 * To speed up listener socket lookup, create an array to store all sockets
 * listening on the same port.  This allows a decision to be made after finding
 * the first socket.  An optional BPF program can also be configured for
 * selecting the socket index from the array of available sockets.
 */

#include <net/sock_reuseport.h>
#include <linux/bpf.h>
#include <linux/rcupdate.h>

#define INIT_SOCKS 128

static DEFINE_SPINLOCK(reuseport_lock);

static struct sock_reuseport *__reuseport_alloc(unsigned int max_socks)
{
	unsigned int size = sizeof(struct sock_reuseport) +
		      sizeof(struct sock *) * max_socks;
	struct sock_reuseport *reuse = kzalloc(size, GFP_ATOMIC);

	if (!reuse)
		return NULL;

	reuse->max_socks = max_socks;

	RCU_INIT_POINTER(reuse->prog, NULL);
	return reuse;
}

int reuseport_alloc(struct sock *sk)
{
	struct sock_reuseport *reuse;

	/* bh lock used since this function call may precede hlist lock in
	 * soft irq of receive path or setsockopt from process context
	 */
	spin_lock_bh(&reuseport_lock);

	/* Allocation attempts can occur concurrently via the setsockopt path
	 * and the bind/hash path.  Nothing to do when we lose the race.
	 */
	if (rcu_dereference_protected(sk->sk_reuseport_cb,
				      lockdep_is_held(&reuseport_lock)))
		goto out;

	reuse = __reuseport_alloc(INIT_SOCKS);
	if (!reuse) {
		spin_unlock_bh(&reuseport_lock);
		return -ENOMEM;
	}

	reuse->socks[0] = sk;
	reuse->num_socks = 1;
	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);

out:
	spin_unlock_bh(&reuseport_lock);

	return 0;
}
EXPORT_SYMBOL(reuseport_alloc);

static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse)
{
	struct sock_reuseport *more_reuse;
	u32 more_socks_size, i;

	more_socks_size = reuse->max_socks * 2U;
	if (more_socks_size > U16_MAX)
		return NULL;

	more_reuse = __reuseport_alloc(more_socks_size);
	if (!more_reuse)
		return NULL;

	more_reuse->max_socks = more_socks_size;
	more_reuse->num_socks = reuse->num_socks;
	more_reuse->prog = reuse->prog;

	memcpy(more_reuse->socks, reuse->socks,
	       reuse->num_socks * sizeof(struct sock *));

	for (i = 0; i < reuse->num_socks; ++i)
		rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
				   more_reuse);

	/* Note: we use kfree_rcu here instead of reuseport_free_rcu so
	 * that reuse and more_reuse can temporarily share a reference
	 * to prog.
	 */
	kfree_rcu(reuse, rcu);
	return more_reuse;
}

/**
 *  reuseport_add_sock - Add a socket to the reuseport group of another.
 *  @sk:  New socket to add to the group.
 *  @sk2: Socket belonging to the existing reuseport group.
 *  May return ENOMEM and not add socket to group under memory pressure.
 */
int reuseport_add_sock(struct sock *sk, struct sock *sk2)
{
	struct sock_reuseport *reuse;

	if (!rcu_access_pointer(sk2->sk_reuseport_cb)) {
		int err = reuseport_alloc(sk2);

		if (err)
			return err;
	}

	spin_lock_bh(&reuseport_lock);
	reuse = rcu_dereference_protected(sk2->sk_reuseport_cb,
					  lockdep_is_held(&reuseport_lock)),
	WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
					    lockdep_is_held(&reuseport_lock)),
		  "socket already in reuseport group");

	if (reuse->num_socks == reuse->max_socks) {
		reuse = reuseport_grow(reuse);
		if (!reuse) {
			spin_unlock_bh(&reuseport_lock);
			return -ENOMEM;
		}
	}

	reuse->socks[reuse->num_socks] = sk;
	/* paired with smp_rmb() in reuseport_select_sock() */
	smp_wmb();
	reuse->num_socks++;
	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);

	spin_unlock_bh(&reuseport_lock);

	return 0;
}

static void reuseport_free_rcu(struct rcu_head *head)
{
	struct sock_reuseport *reuse;

	reuse = container_of(head, struct sock_reuseport, rcu);
	if (reuse->prog)
		bpf_prog_destroy(reuse->prog);
	kfree(reuse);
}

void reuseport_detach_sock(struct sock *sk)
{
	struct sock_reuseport *reuse;
	int i;

	spin_lock_bh(&reuseport_lock);
	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
					  lockdep_is_held(&reuseport_lock));
	rcu_assign_pointer(sk->sk_reuseport_cb, NULL);

	for (i = 0; i < reuse->num_socks; i++) {
		if (reuse->socks[i] == sk) {
			reuse->socks[i] = reuse->socks[reuse->num_socks - 1];
			reuse->num_socks--;
			if (reuse->num_socks == 0)
				call_rcu(&reuse->rcu, reuseport_free_rcu);
			break;
		}
	}
	spin_unlock_bh(&reuseport_lock);
}
EXPORT_SYMBOL(reuseport_detach_sock);

static struct sock *run_bpf(struct sock_reuseport *reuse, u16 socks,
			    struct bpf_prog *prog, struct sk_buff *skb,
			    int hdr_len)
{
	struct sk_buff *nskb = NULL;
	u32 index;

	if (skb_shared(skb)) {
		nskb = skb_clone(skb, GFP_ATOMIC);
		if (!nskb)
			return NULL;
		skb = nskb;
	}

	/* temporarily advance data past protocol header */
	if (!pskb_pull(skb, hdr_len)) {
		kfree_skb(nskb);
		return NULL;
	}
	index = bpf_prog_run_save_cb(prog, skb);
	__skb_push(skb, hdr_len);

	consume_skb(nskb);

	if (index >= socks)
		return NULL;

	return reuse->socks[index];
}

/**
 *  reuseport_select_sock - Select a socket from an SO_REUSEPORT group.
 *  @sk: First socket in the group.
 *  @hash: When no BPF filter is available, use this hash to select.
 *  @skb: skb to run through BPF filter.
 *  @hdr_len: BPF filter expects skb data pointer at payload data.  If
 *    the skb does not yet point at the payload, this parameter represents
 *    how far the pointer needs to advance to reach the payload.
 *  Returns a socket that should receive the packet (or NULL on error).
 */
struct sock *reuseport_select_sock(struct sock *sk,
				   u32 hash,
				   struct sk_buff *skb,
				   int hdr_len)
{
	struct sock_reuseport *reuse;
	struct bpf_prog *prog;
	struct sock *sk2 = NULL;
	u16 socks;

	rcu_read_lock();
	reuse = rcu_dereference(sk->sk_reuseport_cb);

	/* if memory allocation failed or add call is not yet complete */
	if (!reuse)
		goto out;

	prog = rcu_dereference(reuse->prog);
	socks = READ_ONCE(reuse->num_socks);
	if (likely(socks)) {
		/* paired with smp_wmb() in reuseport_add_sock() */
		smp_rmb();

		if (prog && skb)
			sk2 = run_bpf(reuse, socks, prog, skb, hdr_len);
		else
			sk2 = reuse->socks[reciprocal_scale(hash, socks)];
	}

out:
	rcu_read_unlock();
	return sk2;
}
EXPORT_SYMBOL(reuseport_select_sock);

struct bpf_prog *
reuseport_attach_prog(struct sock *sk, struct bpf_prog *prog)
{
	struct sock_reuseport *reuse;
	struct bpf_prog *old_prog;

	spin_lock_bh(&reuseport_lock);
	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
					  lockdep_is_held(&reuseport_lock));
	old_prog = rcu_dereference_protected(reuse->prog,
					     lockdep_is_held(&reuseport_lock));
	rcu_assign_pointer(reuse->prog, prog);
	spin_unlock_bh(&reuseport_lock);

	return old_prog;
}
EXPORT_SYMBOL(reuseport_attach_prog);
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
ef456144 CG	2	/*
	3	* To speed up listener socket lookup, create an array to store all sockets
	4	* listening on the same port. This allows a decision to be made after finding
538950a1 CG	5	* the first socket. An optional BPF program can also be configured for
538950a1 CG	6	* selecting the socket index from the array of available sockets.
ef456144 CG	7	*/
	8
	9	#include <net/sock_reuseport.h>
538950a1	10	#include <linux/bpf.h>
ef456144 CG	11	#include <linux/rcupdate.h>
	12
	13	#define INIT_SOCKS 128
	14
	15	static DEFINE_SPINLOCK(reuseport_lock);
	16
822f9bb1	17	static struct sock_reuseport *__reuseport_alloc(unsigned int max_socks)
ef456144	18	{
822f9bb1	19	unsigned int size = sizeof(struct sock_reuseport) +
ef456144 CG	20	sizeof(struct sock ) max_socks;
	21	struct sock_reuseport *reuse = kzalloc(size, GFP_ATOMIC);
	22
	23	if (!reuse)
	24	return NULL;
	25
	26	reuse->max_socks = max_socks;
	27
538950a1	28	RCU_INIT_POINTER(reuse->prog, NULL);
ef456144 CG	29	return reuse;
	30	}
	31
	32	int reuseport_alloc(struct sock *sk)
	33	{
	34	struct sock_reuseport *reuse;
	35
	36	/* bh lock used since this function call may precede hlist lock in
	37	* soft irq of receive path or setsockopt from process context
	38	*/
	39	spin_lock_bh(&reuseport_lock);
1b5f962e CG	40
	41	/* Allocation attempts can occur concurrently via the setsockopt path
	42	* and the bind/hash path. Nothing to do when we lose the race.
	43	*/
	44	if (rcu_dereference_protected(sk->sk_reuseport_cb,
	45	lockdep_is_held(&reuseport_lock)))
	46	goto out;
	47
ef456144 CG	48	reuse = __reuseport_alloc(INIT_SOCKS);
	49	if (!reuse) {
	50	spin_unlock_bh(&reuseport_lock);
	51	return -ENOMEM;
	52	}
	53
	54	reuse->socks[0] = sk;
	55	reuse->num_socks = 1;
	56	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
	57
1b5f962e	58	out:
ef456144 CG	59	spin_unlock_bh(&reuseport_lock);
	60
	61	return 0;
	62	}
	63	EXPORT_SYMBOL(reuseport_alloc);
	64
	65	static struct sock_reuseport reuseport_grow(struct sock_reuseport reuse)
	66	{
	67	struct sock_reuseport *more_reuse;
	68	u32 more_socks_size, i;
	69
	70	more_socks_size = reuse->max_socks * 2U;
	71	if (more_socks_size > U16_MAX)
	72	return NULL;
	73
	74	more_reuse = __reuseport_alloc(more_socks_size);
	75	if (!more_reuse)
	76	return NULL;
	77
	78	more_reuse->max_socks = more_socks_size;
	79	more_reuse->num_socks = reuse->num_socks;
538950a1	80	more_reuse->prog = reuse->prog;
ef456144 CG	81
	82	memcpy(more_reuse->socks, reuse->socks,
	83	reuse->num_socks * sizeof(struct sock *));
	84
	85	for (i = 0; i < reuse->num_socks; ++i)
	86	rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
	87	more_reuse);
	88
538950a1 CG	89	/* Note: we use kfree_rcu here instead of reuseport_free_rcu so
	90	* that reuse and more_reuse can temporarily share a reference
	91	* to prog.
	92	*/
ef456144 CG	93	kfree_rcu(reuse, rcu);
	94	return more_reuse;
	95	}
	96
	97	/**
	98	* reuseport_add_sock - Add a socket to the reuseport group of another.
	99	* @sk: New socket to add to the group.
	100	* @sk2: Socket belonging to the existing reuseport group.
	101	* May return ENOMEM and not add socket to group under memory pressure.
	102	*/
b4ace4f1	103	int reuseport_add_sock(struct sock sk, struct sock sk2)
ef456144 CG	104	{
	105	struct sock_reuseport *reuse;
	106
b4ace4f1 CG	107	if (!rcu_access_pointer(sk2->sk_reuseport_cb)) {
	108	int err = reuseport_alloc(sk2);
	109
	110	if (err)
	111	return err;
	112	}
	113
ef456144 CG	114	spin_lock_bh(&reuseport_lock);
	115	reuse = rcu_dereference_protected(sk2->sk_reuseport_cb,
	116	lockdep_is_held(&reuseport_lock)),
	117	WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
	118	lockdep_is_held(&reuseport_lock)),
	119	"socket already in reuseport group");
	120
	121	if (reuse->num_socks == reuse->max_socks) {
	122	reuse = reuseport_grow(reuse);
	123	if (!reuse) {
	124	spin_unlock_bh(&reuseport_lock);
	125	return -ENOMEM;
	126	}
	127	}
	128
	129	reuse->socks[reuse->num_socks] = sk;
	130	/* paired with smp_rmb() in reuseport_select_sock() */
	131	smp_wmb();
	132	reuse->num_socks++;
	133	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
	134
	135	spin_unlock_bh(&reuseport_lock);
	136
	137	return 0;
	138	}
ef456144	139
538950a1 CG	140	static void reuseport_free_rcu(struct rcu_head *head)
	141	{
	142	struct sock_reuseport *reuse;
	143
	144	reuse = container_of(head, struct sock_reuseport, rcu);
	145	if (reuse->prog)
	146	bpf_prog_destroy(reuse->prog);
	147	kfree(reuse);
	148	}
	149
ef456144 CG	150	void reuseport_detach_sock(struct sock *sk)
	151	{
	152	struct sock_reuseport *reuse;
	153	int i;
	154
	155	spin_lock_bh(&reuseport_lock);
	156	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
	157	lockdep_is_held(&reuseport_lock));
	158	rcu_assign_pointer(sk->sk_reuseport_cb, NULL);
	159
	160	for (i = 0; i < reuse->num_socks; i++) {
	161	if (reuse->socks[i] == sk) {
	162	reuse->socks[i] = reuse->socks[reuse->num_socks - 1];
	163	reuse->num_socks--;
	164	if (reuse->num_socks == 0)
538950a1	165	call_rcu(&reuse->rcu, reuseport_free_rcu);
ef456144 CG	166	break;
	167	}
	168	}
	169	spin_unlock_bh(&reuseport_lock);
	170	}
	171	EXPORT_SYMBOL(reuseport_detach_sock);
	172
538950a1 CG	173	static struct sock run_bpf(struct sock_reuseport reuse, u16 socks,
	174	struct bpf_prog prog, struct sk_buff skb,
	175	int hdr_len)
	176	{
	177	struct sk_buff *nskb = NULL;
	178	u32 index;
	179
	180	if (skb_shared(skb)) {
	181	nskb = skb_clone(skb, GFP_ATOMIC);
	182	if (!nskb)
	183	return NULL;
	184	skb = nskb;
	185	}
	186
	187	/* temporarily advance data past protocol header */
	188	if (!pskb_pull(skb, hdr_len)) {
00ce3a15	189	kfree_skb(nskb);
538950a1 CG	190	return NULL;
	191	}
	192	index = bpf_prog_run_save_cb(prog, skb);
	193	__skb_push(skb, hdr_len);
	194
	195	consume_skb(nskb);
	196
	197	if (index >= socks)
	198	return NULL;
	199
	200	return reuse->socks[index];
	201	}
	202
ef456144 CG	203	/**
	204	* reuseport_select_sock - Select a socket from an SO_REUSEPORT group.
	205	* @sk: First socket in the group.
538950a1 CG	206	* @hash: When no BPF filter is available, use this hash to select.
	207	* @skb: skb to run through BPF filter.
	208	* @hdr_len: BPF filter expects skb data pointer at payload data. If
	209	* the skb does not yet point at the payload, this parameter represents
	210	* how far the pointer needs to advance to reach the payload.
ef456144 CG	211	* Returns a socket that should receive the packet (or NULL on error).
ef456144 CG	212	*/
538950a1 CG	213	struct sock reuseport_select_sock(struct sock sk,
	214	u32 hash,
	215	struct sk_buff *skb,
	216	int hdr_len)
ef456144 CG	217	{
ef456144 CG	218	struct sock_reuseport *reuse;
538950a1	219	struct bpf_prog *prog;
ef456144 CG	220	struct sock *sk2 = NULL;
	221	u16 socks;
	222
	223	rcu_read_lock();
	224	reuse = rcu_dereference(sk->sk_reuseport_cb);
	225
	226	/* if memory allocation failed or add call is not yet complete */
	227	if (!reuse)
	228	goto out;
	229
538950a1	230	prog = rcu_dereference(reuse->prog);
ef456144 CG	231	socks = READ_ONCE(reuse->num_socks);
	232	if (likely(socks)) {
	233	/* paired with smp_wmb() in reuseport_add_sock() */
	234	smp_rmb();
	235
538950a1 CG	236	if (prog && skb)
	237	sk2 = run_bpf(reuse, socks, prog, skb, hdr_len);
	238	else
	239	sk2 = reuse->socks[reciprocal_scale(hash, socks)];
ef456144 CG	240	}
	241
	242	out:
	243	rcu_read_unlock();
	244	return sk2;
	245	}
	246	EXPORT_SYMBOL(reuseport_select_sock);
538950a1 CG	247
	248	struct bpf_prog *
	249	reuseport_attach_prog(struct sock sk, struct bpf_prog prog)
	250	{
	251	struct sock_reuseport *reuse;
	252	struct bpf_prog *old_prog;
	253
	254	spin_lock_bh(&reuseport_lock);
	255	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
	256	lockdep_is_held(&reuseport_lock));
	257	old_prog = rcu_dereference_protected(reuse->prog,
	258	lockdep_is_held(&reuseport_lock));
	259	rcu_assign_pointer(reuse->prog, prog);
	260	spin_unlock_bh(&reuseport_lock);
	261
	262	return old_prog;
	263	}
	264	EXPORT_SYMBOL(reuseport_attach_prog);