This change actually defines the (initial) metadata layout that should be used by AF_XDP userspace (xsk_tx_metadata). The first field is flags which requests appropriate offloads, followed by the offload-specific fields. The supported per-device offloads are exported via netlink (new xsk-flags). The offloads themselves are still implemented in a bit of a framework-y fashion that's left from my initial kfunc attempt. I'm introducing new xsk_tx_metadata_ops which drivers are supposed to implement. The drivers are also supposed to call xsk_tx_metadata_request/xsk_tx_metadata_complete in the right places. Since xsk_tx_metadata_{request,_complete} are static inline, we don't incur any extra overhead doing indirect calls. The benefit of this scheme is as follows: - keeps all metadata layout parsing away from driver code - makes it easy to grep and see which drivers implement what - don't need any extra flags to maintain to keep track of what offloads are implemented; if the callback is implemented - the offload is supported (used by netlink reporting code) Two offloads are defined right now: 1. XDP_TXMD_FLAGS_CHECKSUM: skb-style csum_start+csum_offset 2. XDP_TXMD_FLAGS_TIMESTAMP: writes TX timestamp back into metadata area upon completion (tx_timestamp field) XDP_TXMD_FLAGS_TIMESTAMP is also implemented for XDP_COPY mode: it writes SW timestamp from the skb destructor (note I'm reusing hwtstamps to pass metadata pointer). The struct is forward-compatible and can be extended in the future by appending more fields.
Unified · Split
| @@ -45,7 +45,6 @@ definitions: · 44 unmodified lines | |||
| 45 | - | 45 | - |
| 46 | type: flags | 46 | type: flags |
| 47 | name: xdp-rx-metadata | 47 | name: xdp-rx-metadata |
| 48 | render-max: true | ||
| 49 | entries: | 48 | entries: |
| 50 | - | 49 | - |
| 51 | name: timestamp | 50 | name: timestamp |
| @@ -55,6 +54,18 @@ definitions: · 3 unmodified lines | |||
| 55 | name: hash | 54 | name: hash |
| 56 | doc: | 55 | doc: |
| 57 | Device is capable of exposing receive packet hash via bpf_xdp_metadata_rx_hash(). | 56 | Device is capable of exposing receive packet hash via bpf_xdp_metadata_rx_hash(). |
| 57 | - | ||
| 58 | type: flags | ||
| 59 | name: xsk-flags | ||
| 60 | entries: | ||
| 61 | - | ||
| 62 | name: tx-timestamp | ||
| 63 | doc: | ||
| 64 | HW timestamping egress packets is supported by the driver. | ||
| 65 | - | ||
| 66 | name: tx-checksum | ||
| 67 | doc: | ||
| 68 | L3 checksum HW offload is supported by the driver. | ||
| 58 | 69 | ||
| 59 | attribute-sets: | 70 | attribute-sets: |
| 60 | - | 71 | - |
| @@ -86,6 +97,11 @@ attribute-sets: · 25 unmodified lines | |||
| 86 | See Documentation/networking/xdp-rx-metadata.rst for more details. | 97 | See Documentation/networking/xdp-rx-metadata.rst for more details. |
| 87 | type: u64 | 98 | type: u64 |
| 88 | enum: xdp-rx-metadata | 99 | enum: xdp-rx-metadata |
| 100 | - | ||
| 101 | name: xsk-features | ||
| 102 | doc: Bitmask of enabled AF_XDP features. | ||
| 103 | type: u64 | ||
| 104 | enum: xsk-flags | ||
| 89 | 105 | ||
| 90 | operations: | 106 | operations: |
| 91 | list: | 107 | list: |
| @@ -103,6 +119,7 @@ operations: · 11 unmodified lines | |||
| 103 | - xdp-features | 119 | - xdp-features |
| 104 | - xdp-zc-max-segs | 120 | - xdp-zc-max-segs |
| 105 | - xdp-rx-metadata-features | 121 | - xdp-rx-metadata-features |
| 122 | - xsk-features | ||
| 106 | dump: | 123 | dump: |
| 107 | reply: *dev-all | 124 | reply: *dev-all |
| 108 | - | 125 | - |
| @@ -1865,6 +1865,7 @@ enum netdev_stat_type { · 1864 unmodified lines | |||
| 1865 | * @netdev_ops: Includes several pointers to callbacks, | 1865 | * @netdev_ops: Includes several pointers to callbacks, |
| 1866 | * if one wants to override the ndo_*() functions | 1866 | * if one wants to override the ndo_*() functions |
| 1867 | * @xdp_metadata_ops: Includes pointers to XDP metadata callbacks. | 1867 | * @xdp_metadata_ops: Includes pointers to XDP metadata callbacks. |
| 1868 | * @xsk_tx_metadata_ops: Includes pointers to AF_XDP TX metadata callbacks. | ||
| 1868 | * @ethtool_ops: Management operations | 1869 | * @ethtool_ops: Management operations |
| 1869 | * @l3mdev_ops: Layer 3 master device operations | 1870 | * @l3mdev_ops: Layer 3 master device operations |
| 1870 | * @ndisc_ops: Includes callbacks for different IPv6 neighbour | 1871 | * @ndisc_ops: Includes callbacks for different IPv6 neighbour |
| @@ -2128,6 +2129,7 @@ struct net_device { · 257 unmodified lines | |||
| 2128 | unsigned long long priv_flags; | 2129 | unsigned long long priv_flags; |
| 2129 | const struct net_device_ops *netdev_ops; | 2130 | const struct net_device_ops *netdev_ops; |
| 2130 | const struct xdp_metadata_ops *xdp_metadata_ops; | 2131 | const struct xdp_metadata_ops *xdp_metadata_ops; |
| 2132 | const struct xsk_tx_metadata_ops *xsk_tx_metadata_ops; | ||
| 2131 | int ifindex; | 2133 | int ifindex; |
| 2132 | unsigned short gflags; | 2134 | unsigned short gflags; |
| 2133 | unsigned short hard_header_len; | 2135 | unsigned short hard_header_len; |
| @@ -566,6 +566,15 @@ struct ubuf_info_msgzc { · 565 unmodified lines | |||
| 566 | int mm_account_pinned_pages(struct mmpin *mmp, size_t size); | 566 | int mm_account_pinned_pages(struct mmpin *mmp, size_t size); |
| 567 | void mm_unaccount_pinned_pages(struct mmpin *mmp); | 567 | void mm_unaccount_pinned_pages(struct mmpin *mmp); |
| 568 | 568 | ||
| 569 | /* Preserve some data across TX submission and completion. | ||
| 570 | * | ||
| 571 | * Note, this state is stored in the driver. Extending the layout | ||
| 572 | * might need some special care. | ||
| 573 | */ | ||
| 574 | struct xsk_tx_metadata_compl { | ||
| 575 | __u64 *tx_timestamp; | ||
| 576 | }; | ||
| 577 | |||
| 569 | /* This data is invariant across clones and lives at | 578 | /* This data is invariant across clones and lives at |
| 570 | * the end of the header data, ie. at skb->end. | 579 | * the end of the header data, ie. at skb->end. |
| 571 | */ | 580 | */ |
| @@ -578,7 +587,10 @@ struct skb_shared_info { · 6 unmodified lines | |||
| 578 | /* Warning: this field is not always filled in (UFO)! */ | 587 | /* Warning: this field is not always filled in (UFO)! */ |
| 579 | unsigned short gso_segs; | 588 | unsigned short gso_segs; |
| 580 | struct sk_buff *frag_list; | 589 | struct sk_buff *frag_list; |
| 581 | struct skb_shared_hwtstamps hwtstamps; | 590 | union { |
| 591 | struct skb_shared_hwtstamps hwtstamps; | ||
| 592 | struct xsk_tx_metadata_compl xsk_meta; | ||
| 593 | }; | ||
| 582 | unsigned int gso_type; | 594 | unsigned int gso_type; |
| 583 | u32 tskey; | 595 | u32 tskey; |
| 584 | 596 | ||
| @@ -93,12 +93,105 @@ struct xdp_sock { · 92 unmodified lines | |||
| 93 | struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */ | 93 | struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */ |
| 94 | }; | 94 | }; |
| 95 | 95 | ||
| 96 | /* | ||
| 97 | * AF_XDP TX metadata hooks for network devices. | ||
| 98 | * The following hooks can be defined; unless noted otherwise, they are | ||
| 99 | * optional and can be filled with a null pointer. | ||
| 100 | * | ||
| 101 | * void (*tmo_request_timestamp)(void *priv) | ||
| 102 | * Called when AF_XDP frame requested egress timestamp. | ||
| 103 | * | ||
| 104 | * u64 (*tmo_fill_timestamp)(void *priv) | ||
| 105 | * Called when AF_XDP frame, that had requested egress timestamp, | ||
| 106 | * received a completion. The hook needs to return the actual HW timestamp. | ||
| 107 | * | ||
| 108 | * void (*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv) | ||
| 109 | * Called when AF_XDP frame requested HW checksum offload. csum_start | ||
| 110 | * indicates position where checksumming should start. | ||
| 111 | * csum_offset indicates position where checksum should be stored. | ||
| 112 | * | ||
| 113 | */ | ||
| 114 | struct xsk_tx_metadata_ops { | ||
| 115 | void (*tmo_request_timestamp)(void *priv); | ||
| 116 | u64 (*tmo_fill_timestamp)(void *priv); | ||
| 117 | void (*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv); | ||
| 118 | }; | ||
| 119 | |||
| 96 | #ifdef CONFIG_XDP_SOCKETS | 120 | #ifdef CONFIG_XDP_SOCKETS |
| 97 | 121 | ||
| 98 | int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp); | 122 | int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp); |
| 99 | int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp); | 123 | int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp); |
| 100 | void __xsk_map_flush(void); | 124 | void __xsk_map_flush(void); |
| 101 | 125 | ||
| 126 | /** | ||
| 127 | * xsk_tx_metadata_to_compl - Save enough relevant metadata information | ||
| 128 | * to perform tx completion in the future. | ||
| 129 | * @meta: pointer to AF_XDP metadata area | ||
| 130 | * @compl: pointer to output struct xsk_tx_metadata_to_compl | ||
| 131 | * | ||
| 132 | * This function should be called by the networking device when | ||
| 133 | * it prepares AF_XDP egress packet. The value of @compl should be stored | ||
| 134 | * and passed to xsk_tx_metadata_complete upon TX completion. | ||
| 135 | */ | ||
| 136 | static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta, | ||
| 137 | struct xsk_tx_metadata_compl *compl) | ||
| 138 | { | ||
| 139 | if (!meta) | ||
| 140 | return; | ||
| 141 | |||
| 142 | if (meta->flags & XDP_TXMD_FLAGS_TIMESTAMP) | ||
| 143 | compl->tx_timestamp = &meta->completion.tx_timestamp; | ||
| 144 | else | ||
| 145 | compl->tx_timestamp = NULL; | ||
| 146 | } | ||
| 147 | |||
| 148 | /** | ||
| 149 | * xsk_tx_metadata_request - Evaluate AF_XDP TX metadata at submission | ||
| 150 | * and call appropriate xsk_tx_metadata_ops operation. | ||
| 151 | * @meta: pointer to AF_XDP metadata area | ||
| 152 | * @ops: pointer to struct xsk_tx_metadata_ops | ||
| 153 | * @priv: pointer to driver-private aread | ||
| 154 | * | ||
| 155 | * This function should be called by the networking device when | ||
| 156 | * it prepares AF_XDP egress packet. | ||
| 157 | */ | ||
| 158 | static inline void xsk_tx_metadata_request(const struct xsk_tx_metadata *meta, | ||
| 159 | const struct xsk_tx_metadata_ops *ops, | ||
| 160 | void *priv) | ||
| 161 | { | ||
| 162 | if (!meta) | ||
| 163 | return; | ||
| 164 | |||
| 165 | if (ops->tmo_request_timestamp) | ||
| 166 | if (meta->flags & XDP_TXMD_FLAGS_TIMESTAMP) | ||
| 167 | ops->tmo_request_timestamp(priv); | ||
| 168 | |||
| 169 | if (ops->tmo_request_checksum) | ||
| 170 | if (meta->flags & XDP_TXMD_FLAGS_CHECKSUM) | ||
| 171 | ops->tmo_request_checksum(meta->request.csum_start, | ||
| 172 | meta->request.csum_offset, priv); | ||
| 173 | } | ||
| 174 | |||
| 175 | /** | ||
| 176 | * xsk_tx_metadata_complete - Evaluate AF_XDP TX metadata at completion | ||
| 177 | * and call appropriate xsk_tx_metadata_ops operation. | ||
| 178 | * @compl: pointer to completion metadata produced from xsk_tx_metadata_to_compl | ||
| 179 | * @ops: pointer to struct xsk_tx_metadata_ops | ||
| 180 | * @priv: pointer to driver-private aread | ||
| 181 | * | ||
| 182 | * This function should be called by the networking device upon | ||
| 183 | * AF_XDP egress completion. | ||
| 184 | */ | ||
| 185 | static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl, | ||
| 186 | const struct xsk_tx_metadata_ops *ops, | ||
| 187 | void *priv) | ||
| 188 | { | ||
| 189 | if (!compl) | ||
| 190 | return; | ||
| 191 | |||
| 192 | *compl->tx_timestamp = ops->tmo_fill_timestamp(priv); | ||
| 193 | } | ||
| 194 | |||
| 102 | #else | 195 | #else |
| 103 | 196 | ||
| 104 | static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) | 197 | static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) |
| @@ -115,6 +208,23 @@ static inline void __xsk_map_flush(void) · 10 unmodified lines | |||
| 115 | { | 208 | { |
| 116 | } | 209 | } |
| 117 | 210 | ||
| 211 | static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta, | ||
| 212 | struct xsk_tx_metadata_compl *compl) | ||
| 213 | { | ||
| 214 | } | ||
| 215 | |||
| 216 | static inline void xsk_tx_metadata_request(struct xsk_tx_metadata *meta, | ||
| 217 | const struct xsk_tx_metadata_ops *ops, | ||
| 218 | void *priv) | ||
| 219 | { | ||
| 220 | } | ||
| 221 | |||
| 222 | static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl, | ||
| 223 | const struct xsk_tx_metadata_ops *ops, | ||
| 224 | void *priv) | ||
| 225 | { | ||
| 226 | } | ||
| 227 | |||
| 118 | #endif /* CONFIG_XDP_SOCKETS */ | 228 | #endif /* CONFIG_XDP_SOCKETS */ |
| 119 | 229 | ||
| 120 | #if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET) | 230 | #if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET) |
| @@ -165,6 +165,14 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr) · 164 unmodified lines | |||
| 165 | return xp_raw_get_data(pool, addr); | 165 | return xp_raw_get_data(pool, addr); |
| 166 | } | 166 | } |
| 167 | 167 | ||
| 168 | static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr) | ||
| 169 | { | ||
| 170 | if (!pool->tx_metadata_len) | ||
| 171 | return NULL; | ||
| 172 | |||
| 173 | return xp_raw_get_data(pool, addr) - pool->tx_metadata_len; | ||
| 174 | } | ||
| 175 | |||
| 168 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) | 176 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) |
| 169 | { | 177 | { |
| 170 | struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp); | 178 | struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp); |
| @@ -324,6 +332,11 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr) · 153 unmodified lines | |||
| 324 | return NULL; | 332 | return NULL; |
| 325 | } | 333 | } |
| 326 | 334 | ||
| 335 | static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr) | ||
| 336 | { | ||
| 337 | return NULL; | ||
| 338 | } | ||
| 339 | |||
| 327 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) | 340 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) |
| 328 | { | 341 | { |
| 329 | } | 342 | } |
| @@ -33,6 +33,7 @@ struct xdp_buff_xsk { · 32 unmodified lines | |||
| 33 | }; | 33 | }; |
| 34 | 34 | ||
| 35 | #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb)) | 35 | #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb)) |
| 36 | #define XSK_TX_COMPL_FITS(t) BUILD_BUG_ON(sizeof(struct xsk_tx_metadata_compl) > sizeof(t)) | ||
| 36 | 37 | ||
| 37 | struct xsk_dma_map { | 38 | struct xsk_dma_map { |
| 38 | dma_addr_t *dma_pages; | 39 | dma_addr_t *dma_pages; |
| @@ -234,4 +235,9 @@ static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb) · 195 unmodified lines | |||
| 234 | return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT); | 235 | return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT); |
| 235 | } | 236 | } |
| 236 | 237 | ||
| 238 | static inline bool xp_tx_metadata_enabled(const struct xsk_buff_pool *pool) | ||
| 239 | { | ||
| 240 | return pool->tx_metadata_len > 0; | ||
| 241 | } | ||
| 242 | |||
| 237 | #endif /* XSK_BUFF_POOL_H_ */ | 243 | #endif /* XSK_BUFF_POOL_H_ */ |
| @@ -106,6 +106,41 @@ struct xdp_options { · 105 unmodified lines | |||
| 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ | 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ |
| 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) | 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) |
| 108 | 108 | ||
| 109 | /* Request transmit timestamp. Upon completion, put it into tx_timestamp | ||
| 110 | * field of struct xsk_tx_metadata. | ||
| 111 | */ | ||
| 112 | #define XDP_TXMD_FLAGS_TIMESTAMP (1 << 0) | ||
| 113 | |||
| 114 | /* Request transmit checksum offload. Checksum start position and offset | ||
| 115 | * are communicated via csum_start and csum_offset fields of struct | ||
| 116 | * xsk_tx_metadata. | ||
| 117 | */ | ||
| 118 | #define XDP_TXMD_FLAGS_CHECKSUM (1 << 1) | ||
| 119 | |||
| 120 | /* AF_XDP offloads request. 'request' union member is consumed by the driver | ||
| 121 | * when the packet is being transmitted. 'completion' union member is | ||
| 122 | * filled by the driver when the transmit completion arrives. | ||
| 123 | */ | ||
| 124 | struct xsk_tx_metadata { | ||
| 125 | __u64 flags; | ||
| 126 | |||
| 127 | union { | ||
| 128 | struct { | ||
| 129 | /* XDP_TXMD_FLAGS_CHECKSUM */ | ||
| 130 | |||
| 131 | /* Offset from desc->addr where checksumming should start. */ | ||
| 132 | __u16 csum_start; | ||
| 133 | /* Offset from csum_start where checksum should be stored. */ | ||
| 134 | __u16 csum_offset; | ||
| 135 | } request; | ||
| 136 | |||
| 137 | struct { | ||
| 138 | /* XDP_TXMD_FLAGS_TIMESTAMP */ | ||
| 139 | __u64 tx_timestamp; | ||
| 140 | } completion; | ||
| 141 | }; | ||
| 142 | }; | ||
| 143 | |||
| 109 | /* Rx/Tx descriptor */ | 144 | /* Rx/Tx descriptor */ |
| 110 | struct xdp_desc { | 145 | struct xdp_desc { |
| 111 | __u64 addr; | 146 | __u64 addr; |
| @@ -122,4 +157,7 @@ struct xdp_desc { · 10 unmodified lines | |||
| 122 | */ | 157 | */ |
| 123 | #define XDP_PKT_CONTD (1 << 0) | 158 | #define XDP_PKT_CONTD (1 << 0) |
| 124 | 159 | ||
| 160 | /* TX packet carries valid metadata. */ | ||
| 161 | #define XDP_TX_METADATA (1 << 1) | ||
| 162 | |||
| 125 | #endif /* _LINUX_IF_XDP_H */ | 163 | #endif /* _LINUX_IF_XDP_H */ |
| @@ -53,12 +53,28 @@ enum netdev_xdp_rx_metadata { · 52 unmodified lines | |||
| 53 | NETDEV_XDP_RX_METADATA_MASK = 3, | 53 | NETDEV_XDP_RX_METADATA_MASK = 3, |
| 54 | }; | 54 | }; |
| 55 | 55 | ||
| 56 | /** | ||
| 57 | * enum netdev_xsk_flags | ||
| 58 | * @NETDEV_XSK_FLAGS_TX_TIMESTAMP: HW timestamping egress packets is supported | ||
| 59 | * by the driver. | ||
| 60 | * @NETDEV_XSK_FLAGS_TX_CHECKSUM: L3 checksum HW offload is supported by the | ||
| 61 | * driver. | ||
| 62 | */ | ||
| 63 | enum netdev_xsk_flags { | ||
| 64 | NETDEV_XSK_FLAGS_TX_TIMESTAMP = 1, | ||
| 65 | NETDEV_XSK_FLAGS_TX_CHECKSUM = 2, | ||
| 66 | |||
| 67 | /* private: */ | ||
| 68 | NETDEV_XSK_FLAGS_MASK = 3, | ||
| 69 | }; | ||
| 70 | |||
| 56 | enum { | 71 | enum { |
| 57 | NETDEV_A_DEV_IFINDEX = 1, | 72 | NETDEV_A_DEV_IFINDEX = 1, |
| 58 | NETDEV_A_DEV_PAD, | 73 | NETDEV_A_DEV_PAD, |
| 59 | NETDEV_A_DEV_XDP_FEATURES, | 74 | NETDEV_A_DEV_XDP_FEATURES, |
| 60 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, | 75 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, |
| 61 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, | 76 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, |
| 77 | NETDEV_A_DEV_XSK_FEATURES, | ||
| 62 | 78 | ||
| 63 | __NETDEV_A_DEV_MAX, | 79 | __NETDEV_A_DEV_MAX, |
| 64 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) | 80 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) |
| @@ -6,6 +6,7 @@ · 5 unmodified lines | |||
| 6 | #include <net/net_namespace.h> | 6 | #include <net/net_namespace.h> |
| 7 | #include <net/sock.h> | 7 | #include <net/sock.h> |
| 8 | #include <net/xdp.h> | 8 | #include <net/xdp.h> |
| 9 | #include <net/xdp_sock.h> | ||
| 9 | 10 | ||
| 10 | #include "netdev-genl-gen.h" | 11 | #include "netdev-genl-gen.h" |
| 11 | 12 | ||
| @@ -13,6 +14,7 @@ static int · 1 unmodified line | |||
| 13 | netdev_nl_dev_fill(struct net_device *netdev, struct sk_buff *rsp, | 14 | netdev_nl_dev_fill(struct net_device *netdev, struct sk_buff *rsp, |
| 14 | const struct genl_info *info) | 15 | const struct genl_info *info) |
| 15 | { | 16 | { |
| 17 | u64 xsk_features = 0; | ||
| 16 | u64 xdp_rx_meta = 0; | 18 | u64 xdp_rx_meta = 0; |
| 17 | void *hdr; | 19 | void *hdr; |
| 18 | 20 | ||
| @@ -26,11 +28,20 @@ netdev_nl_dev_fill(struct net_device *netdev, struct sk_buff *rsp, · 7 unmodified lines | |||
| 26 | XDP_METADATA_KFUNC_xxx | 28 | XDP_METADATA_KFUNC_xxx |
| 27 | #undef XDP_METADATA_KFUNC | 29 | #undef XDP_METADATA_KFUNC |
| 28 | 30 | ||
| 31 | if (netdev->xsk_tx_metadata_ops) { | ||
| 32 | if (netdev->xsk_tx_metadata_ops->tmo_fill_timestamp) | ||
| 33 | xsk_features |= NETDEV_XSK_FLAGS_TX_TIMESTAMP; | ||
| 34 | if (netdev->xsk_tx_metadata_ops->tmo_request_checksum) | ||
| 35 | xsk_features |= NETDEV_XSK_FLAGS_TX_CHECKSUM; | ||
| 36 | } | ||
| 37 | |||
| 29 | if (nla_put_u32(rsp, NETDEV_A_DEV_IFINDEX, netdev->ifindex) || | 38 | if (nla_put_u32(rsp, NETDEV_A_DEV_IFINDEX, netdev->ifindex) || |
| 30 | nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_FEATURES, | 39 | nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_FEATURES, |
| 31 | netdev->xdp_features, NETDEV_A_DEV_PAD) || | 40 | netdev->xdp_features, NETDEV_A_DEV_PAD) || |
| 32 | nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, | 41 | nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, |
| 33 | xdp_rx_meta, NETDEV_A_DEV_PAD)) { | 42 | xdp_rx_meta, NETDEV_A_DEV_PAD) || |
| 43 | nla_put_u64_64bit(rsp, NETDEV_A_DEV_XSK_FEATURES, | ||
| 44 | xsk_features, NETDEV_A_DEV_PAD)) { | ||
| 34 | genlmsg_cancel(rsp, hdr); | 45 | genlmsg_cancel(rsp, hdr); |
| 35 | return -EINVAL; | 46 | return -EINVAL; |
| 36 | } | 47 | } |
| @@ -571,6 +571,13 @@ static u32 xsk_get_num_desc(struct sk_buff *skb) · 570 unmodified lines | |||
| 571 | 571 | ||
| 572 | static void xsk_destruct_skb(struct sk_buff *skb) | 572 | static void xsk_destruct_skb(struct sk_buff *skb) |
| 573 | { | 573 | { |
| 574 | struct xsk_tx_metadata_compl *compl = &skb_shinfo(skb)->xsk_meta; | ||
| 575 | |||
| 576 | if (compl->tx_timestamp) { | ||
| 577 | /* sw completion timestamp, not a real one */ | ||
| 578 | *compl->tx_timestamp = ktime_get_tai_fast_ns(); | ||
| 579 | } | ||
| 580 | |||
| 574 | xsk_cq_submit_locked(xdp_sk(skb->sk), xsk_get_num_desc(skb)); | 581 | xsk_cq_submit_locked(xdp_sk(skb->sk), xsk_get_num_desc(skb)); |
| 575 | sock_wfree(skb); | 582 | sock_wfree(skb); |
| 576 | } | 583 | } |
| @@ -655,8 +662,10 @@ static struct sk_buff *xsk_build_skb_zerocopy(struct xdp_sock *xs, · 78 unmodified lines | |||
| 655 | static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, | 662 | static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, |
| 656 | struct xdp_desc *desc) | 663 | struct xdp_desc *desc) |
| 657 | { | 664 | { |
| 665 | struct xsk_tx_metadata *meta = NULL; | ||
| 658 | struct net_device *dev = xs->dev; | 666 | struct net_device *dev = xs->dev; |
| 659 | struct sk_buff *skb = xs->skb; | 667 | struct sk_buff *skb = xs->skb; |
| 668 | bool first_frag = false; | ||
| 660 | int err; | 669 | int err; |
| 661 | 670 | ||
| 662 | if (dev->priv_flags & IFF_TX_SKB_NO_LINEAR) { | 671 | if (dev->priv_flags & IFF_TX_SKB_NO_LINEAR) { |
| @@ -687,6 +696,8 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, · 24 unmodified lines | |||
| 687 | kfree_skb(skb); | 696 | kfree_skb(skb); |
| 688 | goto free_err; | 697 | goto free_err; |
| 689 | } | 698 | } |
| 699 | |||
| 700 | first_frag = true; | ||
| 690 | } else { | 701 | } else { |
| 691 | int nr_frags = skb_shinfo(skb)->nr_frags; | 702 | int nr_frags = skb_shinfo(skb)->nr_frags; |
| 692 | struct page *page; | 703 | struct page *page; |
| @@ -709,12 +720,35 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, · 16 unmodified lines | |||
| 709 | 720 | ||
| 710 | skb_add_rx_frag(skb, nr_frags, page, 0, len, 0); | 721 | skb_add_rx_frag(skb, nr_frags, page, 0, len, 0); |
| 711 | } | 722 | } |
| 723 | |||
| 724 | if (first_frag && desc->options & XDP_TX_METADATA) { | ||
| 725 | if (unlikely(xs->pool->tx_metadata_len == 0)) { | ||
| 726 | err = -EINVAL; | ||
| 727 | goto free_err; | ||
| 728 | } | ||
| 729 | |||
| 730 | meta = buffer - xs->pool->tx_metadata_len; | ||
| 731 | |||
| 732 | if (meta->flags & XDP_TXMD_FLAGS_CHECKSUM) { | ||
| 733 | if (unlikely(meta->request.csum_start + | ||
| 734 | meta->request.csum_offset + | ||
| 735 | sizeof(__sum16) > len)) { | ||
| 736 | err = -EINVAL; | ||
| 737 | goto free_err; | ||
| 738 | } | ||
| 739 | |||
| 740 | skb->csum_start = hr + meta->request.csum_start; | ||
| 741 | skb->csum_offset = meta->request.csum_offset; | ||
| 742 | skb->ip_summed = CHECKSUM_PARTIAL; | ||
| 743 | } | ||
| 744 | } | ||
| 712 | } | 745 | } |
| 713 | 746 | ||
| 714 | skb->dev = dev; | 747 | skb->dev = dev; |
| 715 | skb->priority = READ_ONCE(xs->sk.sk_priority); | 748 | skb->priority = READ_ONCE(xs->sk.sk_priority); |
| 716 | skb->mark = READ_ONCE(xs->sk.sk_mark); | 749 | skb->mark = READ_ONCE(xs->sk.sk_mark); |
| 717 | skb->destructor = xsk_destruct_skb; | 750 | skb->destructor = xsk_destruct_skb; |
| 751 | xsk_tx_metadata_to_compl(meta, &skb_shinfo(skb)->xsk_meta); | ||
| 718 | xsk_set_destructor_arg(skb); | 752 | xsk_set_destructor_arg(skb); |
| 719 | 753 | ||
| 720 | return skb; | 754 | return skb; |
| @@ -137,7 +137,7 @@ static inline bool xskq_cons_read_addr_unchecked(struct xsk_queue *q, u64 *addr) · 136 unmodified lines | |||
| 137 | 137 | ||
| 138 | static inline bool xp_unused_options_set(u32 options) | 138 | static inline bool xp_unused_options_set(u32 options) |
| 139 | { | 139 | { |
| 140 | return options & ~XDP_PKT_CONTD; | 140 | return options & ~(XDP_PKT_CONTD | XDP_TX_METADATA); |
| 141 | } | 141 | } |
| 142 | 142 | ||
| 143 | static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool, | 143 | static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool, |
| @@ -26,11 +26,11 @@ · 25 unmodified lines | |||
| 26 | */ | 26 | */ |
| 27 | #define XDP_USE_NEED_WAKEUP (1 << 3) | 27 | #define XDP_USE_NEED_WAKEUP (1 << 3) |
| 28 | /* By setting this option, userspace application indicates that it can | 28 | /* By setting this option, userspace application indicates that it can |
| 29 | * handle multiple descriptors per packet thus enabling xsk core to split | 29 | * handle multiple descriptors per packet thus enabling AF_XDP to split |
| 30 | * multi-buffer XDP frames into multiple Rx descriptors. Without this set | 30 | * multi-buffer XDP frames into multiple Rx descriptors. Without this set |
| 31 | * such frames will be dropped by xsk. | 31 | * such frames will be dropped. |
| 32 | */ | 32 | */ |
| 33 | #define XDP_USE_SG (1 << 4) | 33 | #define XDP_USE_SG (1 << 4) |
| 34 | 34 | ||
| 35 | /* Flags for xsk_umem_config flags */ | 35 | /* Flags for xsk_umem_config flags */ |
| 36 | #define XDP_UMEM_UNALIGNED_CHUNK_FLAG (1 << 0) | 36 | #define XDP_UMEM_UNALIGNED_CHUNK_FLAG (1 << 0) |
| @@ -106,6 +106,41 @@ struct xdp_options { · 69 unmodified lines | |||
| 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ | 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ |
| 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) | 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) |
| 108 | 108 | ||
| 109 | /* Request transmit timestamp. Upon completion, put it into tx_timestamp | ||
| 110 | * field of union xsk_tx_metadata. | ||
| 111 | */ | ||
| 112 | #define XDP_TXMD_FLAGS_TIMESTAMP (1 << 0) | ||
| 113 | |||
| 114 | /* Request transmit checksum offload. Checksum start position and offset | ||
| 115 | * are communicated via csum_start and csum_offset fields of union | ||
| 116 | * xsk_tx_metadata. | ||
| 117 | */ | ||
| 118 | #define XDP_TXMD_FLAGS_CHECKSUM (1 << 1) | ||
| 119 | |||
| 120 | /* AF_XDP offloads request. 'request' union member is consumed by the driver | ||
| 121 | * when the packet is being transmitted. 'completion' union member is | ||
| 122 | * filled by the driver when the transmit completion arrives. | ||
| 123 | */ | ||
| 124 | struct xsk_tx_metadata { | ||
| 125 | __u64 flags; | ||
| 126 | |||
| 127 | union { | ||
| 128 | struct { | ||
| 129 | /* XDP_TXMD_FLAGS_CHECKSUM */ | ||
| 130 | |||
| 131 | /* Offset from desc->addr where checksumming should start. */ | ||
| 132 | __u16 csum_start; | ||
| 133 | /* Offset from csum_start where checksum should be stored. */ | ||
| 134 | __u16 csum_offset; | ||
| 135 | } request; | ||
| 136 | |||
| 137 | struct { | ||
| 138 | /* XDP_TXMD_FLAGS_TIMESTAMP */ | ||
| 139 | __u64 tx_timestamp; | ||
| 140 | } completion; | ||
| 141 | }; | ||
| 142 | }; | ||
| 143 | |||
| 109 | /* Rx/Tx descriptor */ | 144 | /* Rx/Tx descriptor */ |
| 110 | struct xdp_desc { | 145 | struct xdp_desc { |
| 111 | __u64 addr; | 146 | __u64 addr; |
| @@ -113,9 +148,16 @@ struct xdp_desc { · 1 unmodified line | |||
| 113 | __u32 options; | 148 | __u32 options; |
| 114 | }; | 149 | }; |
| 115 | 150 | ||
| 116 | /* Flag indicating packet constitutes of multiple buffers*/ | 151 | /* UMEM descriptor is __u64 */ |
| 152 | |||
| 153 | /* Flag indicating that the packet continues with the buffer pointed out by the | ||
| 154 | * next frame in the ring. The end of the packet is signalled by setting this | ||
| 155 | * bit to zero. For single buffer packets, every descriptor has 'options' set | ||
| 156 | * to 0 and this maintains backward compatibility. | ||
| 157 | */ | ||
| 117 | #define XDP_PKT_CONTD (1 << 0) | 158 | #define XDP_PKT_CONTD (1 << 0) |
| 118 | 159 | ||
| 119 | /* UMEM descriptor is __u64 */ | 160 | /* TX packet carries valid metadata. */ |
| 161 | #define XDP_TX_METADATA (1 << 1) | ||
| 120 | 162 | ||
| 121 | #endif /* _LINUX_IF_XDP_H */ | 163 | #endif /* _LINUX_IF_XDP_H */ |
| @@ -53,12 +53,28 @@ enum netdev_xdp_rx_metadata { · 52 unmodified lines | |||
| 53 | NETDEV_XDP_RX_METADATA_MASK = 3, | 53 | NETDEV_XDP_RX_METADATA_MASK = 3, |
| 54 | }; | 54 | }; |
| 55 | 55 | ||
| 56 | /** | ||
| 57 | * enum netdev_xsk_flags | ||
| 58 | * @NETDEV_XSK_FLAGS_TX_TIMESTAMP: HW timestamping egress packets is supported | ||
| 59 | * by the driver. | ||
| 60 | * @NETDEV_XSK_FLAGS_TX_CHECKSUM: L3 checksum HW offload is supported by the | ||
| 61 | * driver. | ||
| 62 | */ | ||
| 63 | enum netdev_xsk_flags { | ||
| 64 | NETDEV_XSK_FLAGS_TX_TIMESTAMP = 1, | ||
| 65 | NETDEV_XSK_FLAGS_TX_CHECKSUM = 2, | ||
| 66 | |||
| 67 | /* private: */ | ||
| 68 | NETDEV_XSK_FLAGS_MASK = 3, | ||
| 69 | }; | ||
| 70 | |||
| 56 | enum { | 71 | enum { |
| 57 | NETDEV_A_DEV_IFINDEX = 1, | 72 | NETDEV_A_DEV_IFINDEX = 1, |
| 58 | NETDEV_A_DEV_PAD, | 73 | NETDEV_A_DEV_PAD, |
| 59 | NETDEV_A_DEV_XDP_FEATURES, | 74 | NETDEV_A_DEV_XDP_FEATURES, |
| 60 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, | 75 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, |
| 61 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, | 76 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, |
| 77 | NETDEV_A_DEV_XSK_FEATURES, | ||
| 62 | 78 | ||
| 63 | __NETDEV_A_DEV_MAX, | 79 | __NETDEV_A_DEV_MAX, |
| 64 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) | 80 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) |
| @@ -58,6 +58,19 @@ const char *netdev_xdp_rx_metadata_str(enum netdev_xdp_rx_metadata value) · 57 unmodified lines | |||
| 58 | return netdev_xdp_rx_metadata_strmap[value]; | 58 | return netdev_xdp_rx_metadata_strmap[value]; |
| 59 | } | 59 | } |
| 60 | 60 | ||
| 61 | static const char * const netdev_xsk_flags_strmap[] = { | ||
| 62 | [0] = "tx-timestamp", | ||
| 63 | [1] = "tx-checksum", | ||
| 64 | }; | ||
| 65 | |||
| 66 | const char *netdev_xsk_flags_str(enum netdev_xsk_flags value) | ||
| 67 | { | ||
| 68 | value = ffs(value) - 1; | ||
| 69 | if (value < 0 || value >= (int)MNL_ARRAY_SIZE(netdev_xsk_flags_strmap)) | ||
| 70 | return NULL; | ||
| 71 | return netdev_xsk_flags_strmap[value]; | ||
| 72 | } | ||
| 73 | |||
| 61 | /* Policies */ | 74 | /* Policies */ |
| 62 | struct ynl_policy_attr netdev_dev_policy[NETDEV_A_DEV_MAX + 1] = { | 75 | struct ynl_policy_attr netdev_dev_policy[NETDEV_A_DEV_MAX + 1] = { |
| 63 | [NETDEV_A_DEV_IFINDEX] = { .name = "ifindex", .type = YNL_PT_U32, }, | 76 | [NETDEV_A_DEV_IFINDEX] = { .name = "ifindex", .type = YNL_PT_U32, }, |
| @@ -65,6 +78,7 @@ struct ynl_policy_attr netdev_dev_policy[NETDEV_A_DEV_MAX + 1] = { · 1 unmodified line | |||
| 65 | [NETDEV_A_DEV_XDP_FEATURES] = { .name = "xdp-features", .type = YNL_PT_U64, }, | 78 | [NETDEV_A_DEV_XDP_FEATURES] = { .name = "xdp-features", .type = YNL_PT_U64, }, |
| 66 | [NETDEV_A_DEV_XDP_ZC_MAX_SEGS] = { .name = "xdp-zc-max-segs", .type = YNL_PT_U32, }, | 79 | [NETDEV_A_DEV_XDP_ZC_MAX_SEGS] = { .name = "xdp-zc-max-segs", .type = YNL_PT_U32, }, |
| 67 | [NETDEV_A_DEV_XDP_RX_METADATA_FEATURES] = { .name = "xdp-rx-metadata-features", .type = YNL_PT_U64, }, | 80 | [NETDEV_A_DEV_XDP_RX_METADATA_FEATURES] = { .name = "xdp-rx-metadata-features", .type = YNL_PT_U64, }, |
| 81 | [NETDEV_A_DEV_XSK_FEATURES] = { .name = "xsk-features", .type = YNL_PT_U64, }, | ||
| 68 | }; | 82 | }; |
| 69 | 83 | ||
| 70 | struct ynl_policy_nest netdev_dev_nest = { | 84 | struct ynl_policy_nest netdev_dev_nest = { |
| @@ -116,6 +130,11 @@ int netdev_dev_get_rsp_parse(const struct nlmsghdr *nlh, void *data) · 45 unmodified lines | |||
| 116 | return MNL_CB_ERROR; | 130 | return MNL_CB_ERROR; |
| 117 | dst->_present.xdp_rx_metadata_features = 1; | 131 | dst->_present.xdp_rx_metadata_features = 1; |
| 118 | dst->xdp_rx_metadata_features = mnl_attr_get_u64(attr); | 132 | dst->xdp_rx_metadata_features = mnl_attr_get_u64(attr); |
| 133 | } else if (type == NETDEV_A_DEV_XSK_FEATURES) { | ||
| 134 | if (ynl_attr_validate(yarg, attr)) | ||
| 135 | return MNL_CB_ERROR; | ||
| 136 | dst->_present.xsk_features = 1; | ||
| 137 | dst->xsk_features = mnl_attr_get_u64(attr); | ||
| 119 | } | 138 | } |
| 120 | } | 139 | } |
| 121 | 140 | ||
| @@ -19,6 +19,7 @@ extern const struct ynl_family ynl_netdev_family; · 18 unmodified lines | |||
| 19 | const char *netdev_op_str(int op); | 19 | const char *netdev_op_str(int op); |
| 20 | const char *netdev_xdp_act_str(enum netdev_xdp_act value); | 20 | const char *netdev_xdp_act_str(enum netdev_xdp_act value); |
| 21 | const char *netdev_xdp_rx_metadata_str(enum netdev_xdp_rx_metadata value); | 21 | const char *netdev_xdp_rx_metadata_str(enum netdev_xdp_rx_metadata value); |
| 22 | const char *netdev_xsk_flags_str(enum netdev_xsk_flags value); | ||
| 22 | 23 | ||
| 23 | /* Common nested types */ | 24 | /* Common nested types */ |
| 24 | /* ============== NETDEV_CMD_DEV_GET ============== */ | 25 | /* ============== NETDEV_CMD_DEV_GET ============== */ |
| @@ -50,12 +51,14 @@ struct netdev_dev_get_rsp { · 25 unmodified lines | |||
| 50 | __u32 xdp_features:1; | 51 | __u32 xdp_features:1; |
| 51 | __u32 xdp_zc_max_segs:1; | 52 | __u32 xdp_zc_max_segs:1; |
| 52 | __u32 xdp_rx_metadata_features:1; | 53 | __u32 xdp_rx_metadata_features:1; |
| 54 | __u32 xsk_features:1; | ||
| 53 | } _present; | 55 | } _present; |
| 54 | 56 | ||
| 55 | __u32 ifindex; | 57 | __u32 ifindex; |
| 56 | __u64 xdp_features; | 58 | __u64 xdp_features; |
| 57 | __u32 xdp_zc_max_segs; | 59 | __u32 xdp_zc_max_segs; |
| 58 | __u64 xdp_rx_metadata_features; | 60 | __u64 xdp_rx_metadata_features; |
| 61 | __u64 xsk_features; | ||
| 59 | }; | 62 | }; |
| 60 | 63 | ||
| 61 | void netdev_dev_get_rsp_free(struct netdev_dev_get_rsp *rsp); | 64 | void netdev_dev_get_rsp_free(struct netdev_dev_get_rsp *rsp); |