sock: Fix sock queue mapping to include device
Subsystems (9)
cover sock: Fix sock queue mapping to include device
The transmit queue selected for a packet is saved in the associated sock for the packet and is subsequently used to avoid recalculating the queue on subsequent sends. The problem is that the corresponding device is not also recorded so that when the queue mapping is referenced it may correspond to a different device than the sending one, resulting in an incorrect queue being used for transmit. A similar problem exists in recording the receive queue in the sock without the corresponding receive device.
This patch set fixes the issue by recording both the device (via ifindex) and the queue in the sock mapping. The pair is set and retrieved atomically. The caller getting the mapping pair checks that both the recorded queue and in the device are valid in the context (for instance, in transmit the returned ifindex is checked against that of the transmitting device to ensure they refer to same device before apply the recorded queue).
This patch set contains:
- Definition of dev_and_queue structure to hold the ifindex and queue number
- Generic functions to get, set, and clear dev_and_queue structure
- Change sk_tx_queue_{get,set,clear} to sk_tx_dev_and_queue_{get,set,clear}
- Modify callers of above to use new interface
- Change sk_rx_queue_{get,set,clear} to sk_rx_dev_and_queue_{get,set,clear}
- Modify callers of above to use new interface
Tom Herbert (3):
sock: Definition and general functions for dev_and_queue structure
sock: Use dev_and_queue structure for TX queue mapping in sock
sock: Use dev_and_queue structure for RX queue mapping in sock
.../mellanox/mlx5/core/en_accel/ktls_rx.c | 10 +- drivers/net/hyperv/netvsc_drv.c | 9 +- include/net/busy_poll.h | 2 +- include/net/request_sock.h | 2 +- include/net/sock.h | 126 +++++++++++++----- net/core/dev.c | 15 ++- net/core/filter.c | 7 +- net/core/sock.c | 10 +- net/ipv4/tcp_input.c | 2 +- 9 files changed, 124 insertions(+), 59 deletions(-)
Signature · 1 line
2.25.1
RFC PATCH 1/3 sock: Definition and general functions for dev_and_queue structure
Add struct dev_and_queue which holds and ifindex and queue pair. Add generic functions to set, get, and clear the pair in a structure.
--- include/net/sock.h | 56 ++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 54 insertions(+), 2 deletions(-)
include/net/sock.h +54 -2
@@ -106,6 +106,16 @@ typedef struct {#endif } socket_lock_t; +struct dev_and_queue { +union { +struct { +int ifindex; +u16 queue; + }; +u64 val64; + }; +}; +struct sock ;struct proto ;struct net ;
@@ -1788,6 +1798,50 @@ static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,return __sk_receive_skb (sk, skb, nested,1 , true); } +#define NO_QUEUE_MAPPING USHRT_MAX + +static inline void __dev_and_queue_get (const struct dev_and_queue *idandq, +int *ifindex,int *queue) +{ +struct dev_and_queue dandq; + + dandq.val64 = idandq->val64; + +if (dandq.ifindex >=0 && dandq.queue !=NO_QUEUE_MAPPING ) { + *ifindex = dandq.ifindex; + *queue = dandq.queue; +return ; + } + + *ifindex =-1 ; + *queue =-1 ; +} + +static inline void __dev_and_queue_set (struct dev_and_queue *odandq, +struct net_device *dev,int queue) +{ +struct dev_and_queue dandq; + +/* queue_mapping accept only upto a 16-bit value */ +if (WARN_ON_ONCE ((unsigned short )queue >=USHRT_MAX )) +return ; + + dandq.ifindex = dev->ifindex; + dandq.queue = queue; + + odandq->val64 = dandq.val64; +} + +static inline void __dev_and_queue_clear (struct dev_and_queue *odandq) +{ +struct dev_and_queue dandq; + + dandq.ifindex =-1 ; + dandq.queue =NO_QUEUE_MAPPING ; + + odandq->val64 = dandq.val64; +} +static inline void sk_tx_queue_set (struct sock *sk,int tx_queue) {/* sk_tx_queue_mapping accept only upto a 16-bit value */
@@ -1796,8 +1850,6 @@ static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) sk->sk_tx_queue_mapping = tx_queue; } -#define NO_QUEUE_MAPPING USHRT_MAX -static inline void sk_tx_queue_clear (struct sock *sk) { sk->sk_tx_queue_mapping =NO_QUEUE_MAPPING ;
Signature · 1 line
2.25.1
RFC PATCH 2/3 sock: Use dev_and_queue structure for TX queue mapping in sock
In reply to Tom HerbertReplace sk_tx_queue_mapping with sk_tx_dev_and_queue_mapping and change associated function to set, get, and clear mapping.
--- drivers/net/hyperv/netvsc_drv.c | 9 +++++--- include/net/request_sock.h | 2 +- include/net/sock.h | 38 +++++++++++++++++---------------- net/core/dev.c | 8 ++++--- net/core/sock.c | 8 +++---- 5 files changed, 36 insertions(+), 29 deletions(-)
drivers/net/hyperv/netvsc_drv.c +6 -3
@@ -307,7 +307,7 @@ static inline int netvsc_get_tx_queue(struct net_device *ndev,/* If queue index changed record the new value */ if (q_idx != old_idx && sk &&sk_fullsock (sk) &&rcu_access_pointer (sk->sk_dst_cache)) -sk_tx_queue_set (sk, q_idx); +sk_tx_dev_and_queue_set (sk, ndev, q_idx);return q_idx; }
@@ -325,9 +325,12 @@ static inline int netvsc_get_tx_queue(struct net_device *ndev,*/ static u16 netvsc_pick_tx (struct net_device *ndev,struct sk_buff *skb) { -int q_idx =sk_tx_queue_get (skb->sk); +int ifindex, q_idx; -if (q_idx <0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) { +sk_tx_dev_and_queue_get (skb->sk, &ifindex, &q_idx); + +if (ifindex != ndev->ifindex || q_idx <0 || skb->ooo_okay || + q_idx >= ndev->real_num_tx_queues) {/* If forwarding a packet, we use the recorded queue when * available for better cache locality. */
include/net/request_sock.h +1 -1
@@ -95,7 +95,7 @@ reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener, req->rsk_ops = ops;req_to_sk (req)->sk_prot = sk_listener->sk_prot;sk_node_init (&req_to_sk (req)->sk_node); -sk_tx_queue_clear (req_to_sk (req)); +sk_tx_dev_and_queue_clear (req_to_sk (req)); req->saved_syn =NULL ; req->num_timeout =0 ; req->num_retrans =0 ;
include/net/sock.h +20 -18
@@ -149,7 +149,7 @@ typedef __u64 __bitwise __addrpair;* @skc_cookie: socket's cookie value * @skc_node: main hash linkage for various protocol lookup tables * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol -* @skc_tx_queue_mapping: tx queue number for this connection +* @skc_tx_dev_and_queue_mapping: tx ifindex/queue for this connection * @skc_rx_queue_mapping: rx queue number for this connection * @skc_flags: place holder for sk_flags * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
@@ -234,7 +234,7 @@ struct sock_common {struct hlist_node skc_node;struct hlist_nulls_node skc_nulls_node; }; -unsigned short skc_tx_queue_mapping; +struct dev_and_queue skc_tx_dev_and_queue_mapping;#ifdef CONFIG_XPS unsigned short skc_rx_queue_mapping;#endif
@@ -362,7 +362,7 @@ struct sock {#define sk_node __sk_common.skc_node#define sk_nulls_node __sk_common.skc_nulls_node#define sk_refcnt __sk_common.skc_refcnt -#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping +#define sk_tx_dev_and_queue_mapping __sk_common.skc_tx_dev_and_queue_mapping#ifdef CONFIG_XPS #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping#endif
@@ -1842,25 +1842,27 @@ static inline void __dev_and_queue_clear(struct dev_and_queue *odandq) odandq->val64 = dandq.val64; } -static inline void sk_tx_queue_set (struct sock *sk,int tx_queue) +static inline void sk_tx_dev_and_queue_set (struct sock *sk, +struct net_device *dev,int tx_queue) { -/* sk_tx_queue_mapping accept only upto a 16-bit value */ -if (WARN_ON_ONCE ((unsigned short )tx_queue >=USHRT_MAX )) -return ; - sk->sk_tx_queue_mapping = tx_queue; +__dev_and_queue_set (&sk->sk_tx_dev_and_queue_mapping, dev, tx_queue); } -static inline void sk_tx_queue_clear (struct sock *sk) +static inline void sk_tx_dev_and_queue_clear (struct sock *sk) { - sk->sk_tx_queue_mapping =NO_QUEUE_MAPPING ; +__dev_and_queue_clear (&sk->sk_tx_dev_and_queue_mapping); } -static inline int sk_tx_queue_get (const struct sock *sk) +static inline void sk_tx_dev_and_queue_get (const struct sock *sk,int *ifindex, +int *tx_queue) { -if (sk && sk->sk_tx_queue_mapping !=NO_QUEUE_MAPPING ) -return sk->sk_tx_queue_mapping; - -return -1 ; +if (sk) { +__dev_and_queue_get (&sk->sk_tx_dev_and_queue_mapping, + ifindex, tx_queue); + }else { + *ifindex =-1 ; + *tx_queue =-1 ; + } }static inline void sk_rx_queue_set (struct sock *sk,const struct sk_buff *skb)
@@ -1989,7 +1991,7 @@ static inline void dst_negative_advice(struct sock *sk)if (ndst != dst) {rcu_assign_pointer (sk->sk_dst_cache, ndst); -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); sk->sk_dst_pending_confirm =0 ; } }
@@ -2000,7 +2002,7 @@ __sk_dst_set(struct sock *sk, struct dst_entry *dst) {struct dst_entry *old_dst; -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); sk->sk_dst_pending_confirm =0 ; old_dst =rcu_dereference_protected (sk->sk_dst_cache,lockdep_sock_is_held (sk));
@@ -2013,7 +2015,7 @@ sk_dst_set(struct sock *sk, struct dst_entry *dst) {struct dst_entry *old_dst; -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); sk->sk_dst_pending_confirm =0 ; old_dst =xchg ((__force struct dst_entry **)&sk->sk_dst_cache, dst);dst_release (old_dst);
net/core/dev.c +5 -3
@@ -3982,11 +3982,13 @@ u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, struct net_device *sb_dev) {struct sock *sk = skb->sk; -int queue_index =sk_tx_queue_get (sk); +int queue_index, ifindex; + +sk_tx_dev_and_queue_get (sk, &ifindex, &queue_index); sb_dev = sb_dev ? : dev; -if (queue_index <0 || skb->ooo_okay || +if (ifindex != dev->ifindex || queue_index <0 || skb->ooo_okay || queue_index >= dev->real_num_tx_queues) {int new_index =get_xps_queue (dev, sb_dev, skb);
@@ -3996,7 +3998,7 @@ u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,if (queue_index != new_index && sk &&sk_fullsock (sk) &&rcu_access_pointer (sk->sk_dst_cache)) -sk_tx_queue_set (sk, new_index); +sk_tx_dev_and_queue_set (sk, dev, new_index); queue_index = new_index; }
net/core/sock.c +4 -4
@@ -542,7 +542,7 @@ struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)struct dst_entry *dst =__sk_dst_get (sk);if (dst && dst->obsolete && dst->ops->check (dst, cookie) ==NULL ) { -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); sk->sk_dst_pending_confirm =0 ;RCU_INIT_POINTER (sk->sk_dst_cache,NULL );dst_release (dst);
@@ -1680,7 +1680,7 @@ static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,if (!try_module_get (prot->owner)) goto out_free_sec; -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); }return sk;
@@ -1749,7 +1749,7 @@ struct sock *sk_alloc(struct net *net, int family, gfp_t priority,cgroup_sk_alloc (&sk->sk_cgrp_data);sock_update_classid (&sk->sk_cgrp_data);sock_update_netprioidx (&sk->sk_cgrp_data); -sk_tx_queue_clear (sk); +sk_tx_dev_and_queue_clear (sk); }return sk;
@@ -1973,7 +1973,7 @@ struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)*/ sk_refcnt_debug_inc (newsk);sk_set_socket (newsk,NULL ); -sk_tx_queue_clear (newsk); +sk_tx_dev_and_queue_clear (newsk);RCU_INIT_POINTER (newsk->sk_wq,NULL );if (newsk->sk_prot->sockets_allocated)
Signature · 1 line
2.25.1
RFC PATCH 3/3 sock: Use dev_and_queue structure for RX queue mapping in sock
In reply to Tom HerbertReplace sk_rx_queue_mapping with sk_rx_dev_and_queue_mapping and change associated function to set, get, and clear mapping.
--- .../mellanox/mlx5/core/en_accel/ktls_rx.c | 10 +++-- include/net/busy_poll.h | 2 +- include/net/sock.h | 38 +++++++++---------- net/core/dev.c | 7 +++- net/core/filter.c | 7 ++-- net/core/sock.c | 2 +- net/ipv4/tcp_input.c | 2 +- 7 files changed, 37 insertions(+), 31 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_rx.c +6 -4
@@ -547,11 +547,13 @@ void mlx5e_ktls_handle_ctx_completion(struct mlx5e_icosq_wqe_info *wi)queue_work (rule->priv->tls->rx_wq, &rule->work); } -static int mlx5e_ktls_sk_get_rxq (struct sock *sk) +static int mlx5e_ktls_sk_get_rxq (struct sock *sk,struct net_device *dev) { -int rxq =sk_rx_queue_get (sk); +int ifindex, rxq; -if (unlikely (rxq ==-1 )) +sk_rx_dev_and_queue_get (sk, &ifindex, &rxq); + +if (unlikely (ifindex != dev->ifindex || rxq ==-1 )) rxq =0 ;return rxq;
@@ -583,7 +585,7 @@ int mlx5e_ktls_add_rx(struct net_device *netdev, struct sock *sk, priv_rx->crypto_info = *(struct tls12_crypto_info_aes_gcm_128 *)crypto_info; - rxq =mlx5e_ktls_sk_get_rxq (sk); + rxq =mlx5e_ktls_sk_get_rxq (sk, netdev); priv_rx->rxq = rxq; priv_rx->sk = sk;
include/net/busy_poll.h +1 -1
@@ -128,7 +128,7 @@ static inline void sk_mark_napi_id(struct sock *sk, const struct sk_buff *skb)#ifdef CONFIG_NET_RX_BUSY_POLL WRITE_ONCE (sk->sk_napi_id, skb->napi_id);#endif -sk_rx_queue_set (sk, skb); +sk_rx_dev_and_queue_set (sk, skb); }/* variant used for unconnected sockets */
include/net/sock.h +19 -19
@@ -150,7 +150,7 @@ typedef __u64 __bitwise __addrpair;* @skc_node: main hash linkage for various protocol lookup tables * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol * @skc_tx_dev_and_queue_mapping: tx ifindex/queue for this connection -* @skc_rx_queue_mapping: rx queue number for this connection +* @skc_rx_dev_and_queue_mapping: rx ifindex/queue for this connection * @skc_flags: place holder for sk_flags * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
@@ -236,7 +236,7 @@ struct sock_common { };struct dev_and_queue skc_tx_dev_and_queue_mapping;#ifdef CONFIG_XPS -unsigned short skc_rx_queue_mapping; +struct dev_and_queue skc_rx_dev_and_queue_mapping;#endif union {int skc_incoming_cpu;
@@ -364,7 +364,7 @@ struct sock {#define sk_refcnt __sk_common.skc_refcnt#define sk_tx_dev_and_queue_mapping __sk_common.skc_tx_dev_and_queue_mapping#ifdef CONFIG_XPS -#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping +#define sk_rx_dev_and_queue_mapping __sk_common.skc_rx_dev_and_queue_mapping#endif #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
@@ -1865,34 +1865,34 @@ static inline void sk_tx_dev_and_queue_get(const struct sock *sk, int *ifindex, } } -static inline void sk_rx_queue_set (struct sock *sk,const struct sk_buff *skb) +static inline void sk_rx_dev_and_queue_set (struct sock *sk, +const struct sk_buff *skb) {#ifdef CONFIG_XPS -if (skb_rx_queue_recorded (skb)) { -u16 rx_queue =skb_get_rx_queue (skb); - -if (WARN_ON_ONCE (rx_queue ==NO_QUEUE_MAPPING )) -return ; - - sk->sk_rx_queue_mapping = rx_queue; - } +if (skb->dev &&skb_rx_queue_recorded (skb)) +__dev_and_queue_set (&sk->sk_rx_dev_and_queue_mapping, skb->dev, +skb_get_rx_queue (skb));#endif } -static inline void sk_rx_queue_clear (struct sock *sk) +static inline void sk_rx_dev_and_queue_clear (struct sock *sk) {#ifdef CONFIG_XPS - sk->sk_rx_queue_mapping =NO_QUEUE_MAPPING ; +__dev_and_queue_clear (&sk->sk_rx_dev_and_queue_mapping);#endif }#ifdef CONFIG_XPS -static inline int sk_rx_queue_get (const struct sock *sk) +static inline void sk_rx_dev_and_queue_get (const struct sock *sk, +int *ifindex,int *rx_queue) { -if (sk && sk->sk_rx_queue_mapping !=NO_QUEUE_MAPPING ) -return sk->sk_rx_queue_mapping; - -return -1 ; +if (sk) { +__dev_and_queue_get (&sk->sk_rx_dev_and_queue_mapping, + ifindex, rx_queue); + }else { + *ifindex =-1 ; + *rx_queue =-1 ; + } }#endif
net/core/dev.c +5 -2
@@ -3939,9 +3939,12 @@ static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, dev_maps =rcu_dereference (sb_dev->xps_rxqs_map);if (dev_maps) { -int tci =sk_rx_queue_get (sk); +int tci, ifindex; -if (tci >=0 && tci < dev->num_rx_queues) +sk_rx_dev_and_queue_get (sk, &ifindex, &tci); + +if (dev->ifindex == ifindex && + tci >=0 && tci < dev->num_rx_queues) queue_index =__get_xps_queue_idx (dev, skb, dev_maps, tci); }
net/core/filter.c +4 -3
@@ -8044,11 +8044,12 @@ u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,case offsetof(struct bpf_sock , rx_queue_mapping):#ifdef CONFIG_XPS *insn++ =BPF_LDX_MEM ( - BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), + BPF_FIELD_SIZEOF(struct sock, + sk_rx_dev_and_queue_mapping.queue), si->dst_reg, si->src_reg, - bpf_target_off(struct sock, sk_rx_queue_mapping, + bpf_target_off(struct sock, sk_rx_dev_and_queue_mapping.queue, sizeof_field(struct sock, - sk_rx_queue_mapping), + sk_rx_dev_and_queue_mapping.queue), target_size)); *insn++ =BPF_JMP_IMM (BPF_JNE , si->dst_reg,NO_QUEUE_MAPPING ,1 );
net/core/sock.c +1 -1
@@ -3000,7 +3000,7 @@ void sock_init_data(struct socket *sock, struct sock *sk)WRITE_ONCE (sk->sk_pacing_shift,10 ); sk->sk_incoming_cpu =-1 ; -sk_rx_queue_clear (sk); +sk_rx_dev_and_queue_clear (sk);/* * Before updating sk_refcnt, we must commit prior changes to memory * (Documentation/RCU/rculist_nulls.txt for details)
net/ipv4/tcp_input.c +1 -1
@@ -6746,7 +6746,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,tcp_rsk (req)->snt_isn = isn;tcp_rsk (req)->txhash =net_tx_rndhash ();tcp_openreq_init_rwin (req, sk, dst); -sk_rx_queue_set (req_to_sk (req), skb); +sk_rx_dev_and_queue_set (req_to_sk (req), skb);if (!want_cookie) {tcp_reqsk_record_syn (sk, req, skb); fastopen_sk =tcp_try_fastopen (sk, skb, req, &foc, dst);
Signature · 1 line
2.25.1