]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - include/net/xdp.h
1 /* SPDX-License-Identifier: GPL-2.0-only */
4 * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
6 #ifndef __LINUX_NET_XDP_H__
7 #define __LINUX_NET_XDP_H__
9 #include <linux/skbuff.h> /* skb_shared_info */
12 * DOC: XDP RX-queue information
14 * The XDP RX-queue info (xdp_rxq_info) is associated with the driver
15 * level RX-ring queues. It is information that is specific to how
16 * the driver have configured a given RX-ring queue.
18 * Each xdp_buff frame received in the driver carry a (pointer)
19 * reference to this xdp_rxq_info structure. This provides the XDP
20 * data-path read-access to RX-info for both kernel and bpf-side
23 * For now, direct access is only safe while running in NAPI/softirq
24 * context. Contents is read-mostly and must not be updated during
25 * driver NAPI/softirq poll.
27 * The driver usage API is a register and unregister API.
29 * The struct is not directly tied to the XDP prog. A new XDP prog
30 * can be attached as long as it doesn't change the underlying
31 * RX-ring. If the RX-ring does change significantly, the NIC driver
32 * naturally need to stop the RX-ring before purging and reallocating
33 * memory. In that process the driver MUST call unregistor (which
34 * also apply for driver shutdown and unload). The register API is
35 * also mandatory during RX-ring setup.
39 MEM_TYPE_PAGE_SHARED
= 0, /* Split-page refcnt based model */
40 MEM_TYPE_PAGE_ORDER0
, /* Orig XDP full page model */
42 MEM_TYPE_XSK_BUFF_POOL
,
46 /* XDP flags for ndo_xdp_xmit */
47 #define XDP_XMIT_FLUSH (1U << 0) /* doorbell signal consumer */
48 #define XDP_XMIT_FLAGS_MASK XDP_XMIT_FLUSH
51 u32 type
; /* enum xdp_mem_type, but known size type */
58 struct net_device
*dev
;
61 struct xdp_mem_info mem
;
63 } ____cacheline_aligned
; /* perf critical, avoid false-sharing */
66 struct net_device
*dev
;
73 void *data_hard_start
;
74 struct xdp_rxq_info
*rxq
;
75 struct xdp_txq_info
*txq
;
76 u32 frame_sz
; /* frame size to deduce data_hard_end/reserved tailroom*/
79 /* Reserve memory area at end-of data area.
81 * This macro reserves tailroom in the XDP buffer by limiting the
82 * XDP/BPF data access to data_hard_end. Notice same area (and size)
83 * is used for XDP_PASS, when constructing the SKB via build_skb().
85 #define xdp_data_hard_end(xdp) \
86 ((xdp)->data_hard_start + (xdp)->frame_sz - \
87 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
89 static inline struct skb_shared_info
*
90 xdp_get_shared_info_from_buff(struct xdp_buff
*xdp
)
92 return (struct skb_shared_info
*)xdp_data_hard_end(xdp
);
101 /* Lifetime of xdp_rxq_info is limited to NAPI/enqueue time,
102 * while mem info is valid on remote CPU.
104 struct xdp_mem_info mem
;
105 struct net_device
*dev_rx
; /* used by cpumap */
108 #define XDP_BULK_QUEUE_SIZE 16
109 struct xdp_frame_bulk
{
112 void *q
[XDP_BULK_QUEUE_SIZE
];
115 static __always_inline
void xdp_frame_bulk_init(struct xdp_frame_bulk
*bq
)
117 /* bq->count will be zero'ed when bq->xa gets updated */
121 static inline struct skb_shared_info
*
122 xdp_get_shared_info_from_frame(struct xdp_frame
*frame
)
124 void *data_hard_start
= frame
->data
- frame
->headroom
- sizeof(*frame
);
126 return (struct skb_shared_info
*)(data_hard_start
+ frame
->frame_sz
-
127 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)));
130 struct xdp_cpumap_stats
{
131 unsigned int redirect
;
136 /* Clear kernel pointers in xdp_frame */
137 static inline void xdp_scrub_frame(struct xdp_frame
*frame
)
140 frame
->dev_rx
= NULL
;
143 /* Avoids inlining WARN macro in fast-path */
144 void xdp_warn(const char *msg
, const char *func
, const int line
);
145 #define XDP_WARN(msg) xdp_warn(msg, __func__, __LINE__)
147 struct xdp_frame
*xdp_convert_zc_to_xdp_frame(struct xdp_buff
*xdp
);
150 void xdp_convert_frame_to_buff(struct xdp_frame
*frame
, struct xdp_buff
*xdp
)
152 xdp
->data_hard_start
= frame
->data
- frame
->headroom
- sizeof(*frame
);
153 xdp
->data
= frame
->data
;
154 xdp
->data_end
= frame
->data
+ frame
->len
;
155 xdp
->data_meta
= frame
->data
- frame
->metasize
;
156 xdp
->frame_sz
= frame
->frame_sz
;
160 int xdp_update_frame_from_buff(struct xdp_buff
*xdp
,
161 struct xdp_frame
*xdp_frame
)
163 int metasize
, headroom
;
165 /* Assure headroom is available for storing info */
166 headroom
= xdp
->data
- xdp
->data_hard_start
;
167 metasize
= xdp
->data
- xdp
->data_meta
;
168 metasize
= metasize
> 0 ? metasize
: 0;
169 if (unlikely((headroom
- metasize
) < sizeof(*xdp_frame
)))
172 /* Catch if driver didn't reserve tailroom for skb_shared_info */
173 if (unlikely(xdp
->data_end
> xdp_data_hard_end(xdp
))) {
174 XDP_WARN("Driver BUG: missing reserved tailroom");
178 xdp_frame
->data
= xdp
->data
;
179 xdp_frame
->len
= xdp
->data_end
- xdp
->data
;
180 xdp_frame
->headroom
= headroom
- sizeof(*xdp_frame
);
181 xdp_frame
->metasize
= metasize
;
182 xdp_frame
->frame_sz
= xdp
->frame_sz
;
187 /* Convert xdp_buff to xdp_frame */
189 struct xdp_frame
*xdp_convert_buff_to_frame(struct xdp_buff
*xdp
)
191 struct xdp_frame
*xdp_frame
;
193 if (xdp
->rxq
->mem
.type
== MEM_TYPE_XSK_BUFF_POOL
)
194 return xdp_convert_zc_to_xdp_frame(xdp
);
196 /* Store info in top of packet */
197 xdp_frame
= xdp
->data_hard_start
;
198 if (unlikely(xdp_update_frame_from_buff(xdp
, xdp_frame
) < 0))
201 /* rxq only valid until napi_schedule ends, convert to xdp_mem_info */
202 xdp_frame
->mem
= xdp
->rxq
->mem
;
207 void xdp_return_frame(struct xdp_frame
*xdpf
);
208 void xdp_return_frame_rx_napi(struct xdp_frame
*xdpf
);
209 void xdp_return_buff(struct xdp_buff
*xdp
);
210 void xdp_flush_frame_bulk(struct xdp_frame_bulk
*bq
);
211 void xdp_return_frame_bulk(struct xdp_frame
*xdpf
,
212 struct xdp_frame_bulk
*bq
);
214 /* When sending xdp_frame into the network stack, then there is no
215 * return point callback, which is needed to release e.g. DMA-mapping
216 * resources with page_pool. Thus, have explicit function to release
219 void __xdp_release_frame(void *data
, struct xdp_mem_info
*mem
);
220 static inline void xdp_release_frame(struct xdp_frame
*xdpf
)
222 struct xdp_mem_info
*mem
= &xdpf
->mem
;
224 /* Curr only page_pool needs this */
225 if (mem
->type
== MEM_TYPE_PAGE_POOL
)
226 __xdp_release_frame(xdpf
->data
, mem
);
229 int xdp_rxq_info_reg(struct xdp_rxq_info
*xdp_rxq
,
230 struct net_device
*dev
, u32 queue_index
, unsigned int napi_id
);
231 void xdp_rxq_info_unreg(struct xdp_rxq_info
*xdp_rxq
);
232 void xdp_rxq_info_unused(struct xdp_rxq_info
*xdp_rxq
);
233 bool xdp_rxq_info_is_reg(struct xdp_rxq_info
*xdp_rxq
);
234 int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info
*xdp_rxq
,
235 enum xdp_mem_type type
, void *allocator
);
236 void xdp_rxq_info_unreg_mem_model(struct xdp_rxq_info
*xdp_rxq
);
238 /* Drivers not supporting XDP metadata can use this helper, which
239 * rejects any room expansion for metadata as a result.
241 static __always_inline
void
242 xdp_set_data_meta_invalid(struct xdp_buff
*xdp
)
244 xdp
->data_meta
= xdp
->data
+ 1;
247 static __always_inline
bool
248 xdp_data_meta_unsupported(const struct xdp_buff
*xdp
)
250 return unlikely(xdp
->data_meta
> xdp
->data
);
253 struct xdp_attachment_info
{
254 struct bpf_prog
*prog
;
259 void xdp_attachment_setup(struct xdp_attachment_info
*info
,
260 struct netdev_bpf
*bpf
);
262 #define DEV_MAP_BULK_SIZE XDP_BULK_QUEUE_SIZE
264 #endif /* __LINUX_NET_XDP_H__ */