Merge branch 'add-multi-buff-support-for-xdp-running-in-generic-mode'

Lorenzo Bianconi says:

====================
add multi-buff support for xdp running in generic mode

Introduce multi-buffer support for xdp running in generic mode not always
linearizing the skb in netif_receive_generic_xdp routine.
Introduce generic percpu page_pools allocator.
====================

Link: https://lore.kernel.org/r/cover.1707729884.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

8 files changed, 231 insertions, 107 deletions

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index b472f2c972d8..bc80fc1d576e 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1926,7 +1926,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 		rcu_read_lock();
 		xdp_prog = rcu_dereference(tun->xdp_prog);
 		if (xdp_prog) {
-			ret = do_xdp_generic(xdp_prog, skb);
+			ret = do_xdp_generic(xdp_prog, &skb);
 			if (ret != XDP_PASS) {
 				rcu_read_unlock();
 				local_bh_enable();
@@ -2516,7 +2516,7 @@ build:
 	skb_record_rx_queue(skb, tfile->queue_index);
 
 	if (skb_xdp) {
-		ret = do_xdp_generic(xdp_prog, skb);
+		ret = do_xdp_generic(xdp_prog, &skb);
 		if (ret != XDP_PASS) {
 			ret = 0;
 			goto out;
diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index de1f13837782..500b9dfccd08 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -729,80 +729,10 @@ static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
 	    skb_shinfo(skb)->nr_frags ||
 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
-		u32 size, len, max_head_size, off, truesize, page_offset;
-		struct sk_buff *nskb;
-		struct page *page;
-		int i, head_off;
-		void *va;
-
-		/* We need a private copy of the skb and data buffers since
-		 * the ebpf program can modify it. We segment the original skb
-		 * into order-0 pages without linearize it.
-		 *
-		 * Make sure we have enough space for linear and paged area
-		 */
-		max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE -
-						  VETH_XDP_HEADROOM);
-		if (skb->len > PAGE_SIZE * MAX_SKB_FRAGS + max_head_size)
-			goto drop;
-
-		size = min_t(u32, skb->len, max_head_size);
-		truesize = SKB_HEAD_ALIGN(size) + VETH_XDP_HEADROOM;
-
-		/* Allocate skb head */
-		va = page_pool_dev_alloc_va(rq->page_pool, &truesize);
-		if (!va)
-			goto drop;
-
-		nskb = napi_build_skb(va, truesize);
-		if (!nskb) {
-			page_pool_free_va(rq->page_pool, va, true);
+		if (skb_pp_cow_data(rq->page_pool, pskb, XDP_PACKET_HEADROOM))
 			goto drop;
-		}
-
-		skb_reserve(nskb, VETH_XDP_HEADROOM);
-		skb_copy_header(nskb, skb);
-		skb_mark_for_recycle(nskb);
-
-		if (skb_copy_bits(skb, 0, nskb->data, size)) {
-			consume_skb(nskb);
-			goto drop;
-		}
-		skb_put(nskb, size);
 
-		head_off = skb_headroom(nskb) - skb_headroom(skb);
-		skb_headers_offset_update(nskb, head_off);
-
-		/* Allocate paged area of new skb */
-		off = size;
-		len = skb->len - off;
-
-		for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
-			size = min_t(u32, len, PAGE_SIZE);
-			truesize = size;
-
-			page = page_pool_dev_alloc(rq->page_pool, &page_offset,
-						   &truesize);
-			if (!page) {
-				consume_skb(nskb);
-				goto drop;
-			}
-
-			skb_add_rx_frag(nskb, i, page, page_offset, size,
-					truesize);
-			if (skb_copy_bits(skb, off,
-					  page_address(page) + page_offset,
-					  size)) {
-				consume_skb(nskb);
-				goto drop;
-			}
-
-			len -= size;
-			off += size;
-		}
-
-		consume_skb(skb);
-		skb = nskb;
+		skb = *pskb;
 	}
 
 	/* SKB "head" area always have tailroom for skb_shared_info */
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 07cefa32eafa..a3f9c95da51e 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -3958,7 +3958,7 @@ static inline void dev_consume_skb_any(struct sk_buff *skb)
 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
 			     struct bpf_prog *xdp_prog);
 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog);
-int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb);
+int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb);
 int netif_rx(struct sk_buff *skb);
 int __netif_rx(struct sk_buff *skb);
 
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 2dde34c29203..696e7680656f 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -3446,6 +3446,10 @@ static inline void skb_frag_ref(struct sk_buff *skb, int f)
 	__skb_frag_ref(&skb_shinfo(skb)->frags[f]);
 }
 
+int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
+		    unsigned int headroom);
+int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
+			 struct bpf_prog *prog);
 bool napi_pp_put_page(struct page *page, bool napi_safe);
 
 static inline void
diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
index 76481c465375..3828396ae60c 100644
--- a/include/net/page_pool/types.h
+++ b/include/net/page_pool/types.h
@@ -128,6 +128,7 @@ struct page_pool_stats {
 struct page_pool {
 	struct page_pool_params_fast p;
 
+	int cpuid;
 	bool has_init_callback;
 
 	long frag_users;
@@ -203,6 +204,8 @@ struct page *page_pool_alloc_pages(struct page_pool *pool, gfp_t gfp);
 struct page *page_pool_alloc_frag(struct page_pool *pool, unsigned int *offset,
 				  unsigned int size, gfp_t gfp);
 struct page_pool *page_pool_create(const struct page_pool_params *params);
+struct page_pool *page_pool_create_percpu(const struct page_pool_params *params,
+					  int cpuid);
 
 struct xdp_mem_info;
 
diff --git a/net/core/dev.c b/net/core/dev.c
index 7cf15d2bf78d..2d02ca8a3da5 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -153,6 +153,8 @@
 #include <linux/prandom.h>
 #include <linux/once_lite.h>
 #include <net/netdev_rx_queue.h>
+#include <net/page_pool/types.h>
+#include <net/page_pool/helpers.h>
 
 #include "dev.h"
 #include "net-sysfs.h"
@@ -450,6 +452,12 @@ static RAW_NOTIFIER_HEAD(netdev_chain);
 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 EXPORT_PER_CPU_SYMBOL(softnet_data);
 
+/* Page_pool has a lockless array/stack to alloc/recycle pages.
+ * PP consumers must pay attention to run APIs in the appropriate context
+ * (e.g. NAPI context).
+ */
+static DEFINE_PER_CPU_ALIGNED(struct page_pool *, system_page_pool);
+
 #ifdef CONFIG_LOCKDEP
 /*
  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
@@ -4866,6 +4874,12 @@ u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
 			 skb_headlen(skb) + mac_len, true);
+	if (skb_is_nonlinear(skb)) {
+		skb_shinfo(skb)->xdp_frags_size = skb->data_len;
+		xdp_buff_set_frags_flag(xdp);
+	} else {
+		xdp_buff_clear_frags_flag(xdp);
+	}
 
 	orig_data_end = xdp->data_end;
 	orig_data = xdp->data;
@@ -4895,6 +4909,14 @@ u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
 		skb->len += off; /* positive on grow, negative on shrink */
 	}
 
+	/* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
+	 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
+	 */
+	if (xdp_buff_has_frags(xdp))
+		skb->data_len = skb_shinfo(skb)->xdp_frags_size;
+	else
+		skb->data_len = 0;
+
 	/* check if XDP changed eth hdr such SKB needs update */
 	eth = (struct ethhdr *)xdp->data;
 	if ((orig_eth_type != eth->h_proto) ||
@@ -4928,11 +4950,35 @@ u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
 	return act;
 }
 
-static u32 netif_receive_generic_xdp(struct sk_buff *skb,
+static int
+netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog)
+{
+	struct sk_buff *skb = *pskb;
+	int err, hroom, troom;
+
+	if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog))
+		return 0;
+
+	/* In case we have to go down the path and also linearize,
+	 * then lets do the pskb_expand_head() work just once here.
+	 */
+	hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
+	troom = skb->tail + skb->data_len - skb->end;
+	err = pskb_expand_head(skb,
+			       hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
+			       troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
+	if (err)
+		return err;
+
+	return skb_linearize(skb);
+}
+
+static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
 				     struct xdp_buff *xdp,
 				     struct bpf_prog *xdp_prog)
 {
-	u32 act = XDP_DROP;
+	struct sk_buff *skb = *pskb;
+	u32 mac_len, act = XDP_DROP;
 
 	/* Reinjected packets coming from act_mirred or similar should
 	 * not get XDP generic processing.
@@ -4940,41 +4986,36 @@ static u32 netif_receive_generic_xdp(struct sk_buff *skb,
 	if (skb_is_redirected(skb))
 		return XDP_PASS;
 
-	/* XDP packets must be linear and must have sufficient headroom
-	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
-	 * native XDP provides, thus we need to do it here as well.
+	/* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
+	 * bytes. This is the guarantee that also native XDP provides,
+	 * thus we need to do it here as well.
 	 */
+	mac_len = skb->data - skb_mac_header(skb);
+	__skb_push(skb, mac_len);
+
 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
-		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
-		int troom = skb->tail + skb->data_len - skb->end;
-
-		/* In case we have to go down the path and also linearize,
-		 * then lets do the pskb_expand_head() work just once here.
-		 */
-		if (pskb_expand_head(skb,
-				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
-				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
-			goto do_drop;
-		if (skb_linearize(skb))
+		if (netif_skb_check_for_xdp(pskb, xdp_prog))
 			goto do_drop;
 	}
 
-	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
+	__skb_pull(*pskb, mac_len);
+
+	act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
 	switch (act) {
 	case XDP_REDIRECT:
 	case XDP_TX:
 	case XDP_PASS:
 		break;
 	default:
-		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
+		bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
 		fallthrough;
 	case XDP_ABORTED:
-		trace_xdp_exception(skb->dev, xdp_prog, act);
+		trace_xdp_exception((*pskb)->dev, xdp_prog, act);
 		fallthrough;
 	case XDP_DROP:
 	do_drop:
-		kfree_skb(skb);
+		kfree_skb(*pskb);
 		break;
 	}
 
@@ -5012,24 +5053,24 @@ void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
 
 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
 
-int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
+int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb)
 {
 	if (xdp_prog) {
 		struct xdp_buff xdp;
 		u32 act;
 		int err;
 
-		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
+		act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
 		if (act != XDP_PASS) {
 			switch (act) {
 			case XDP_REDIRECT:
-				err = xdp_do_generic_redirect(skb->dev, skb,
+				err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
 							      &xdp, xdp_prog);
 				if (err)
 					goto out_redir;
 				break;
 			case XDP_TX:
-				generic_xdp_tx(skb, xdp_prog);
+				generic_xdp_tx(*pskb, xdp_prog);
 				break;
 			}
 			return XDP_DROP;
@@ -5037,7 +5078,7 @@ int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
 	}
 	return XDP_PASS;
 out_redir:
-	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
+	kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
 	return XDP_DROP;
 }
 EXPORT_SYMBOL_GPL(do_xdp_generic);
@@ -5360,7 +5401,8 @@ another_round:
 		int ret2;
 
 		migrate_disable();
-		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
+		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
+				      &skb);
 		migrate_enable();
 
 		if (ret2 != XDP_PASS) {
@@ -11724,6 +11766,27 @@ static void __init net_dev_struct_check(void)
  *
  */
 
+/* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
+#define SYSTEM_PERCPU_PAGE_POOL_SIZE	((1 << 20) / PAGE_SIZE)
+
+static int net_page_pool_create(int cpuid)
+{
+#if IS_ENABLED(CONFIG_PAGE_POOL)
+	struct page_pool_params page_pool_params = {
+		.pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
+		.nid = NUMA_NO_NODE,
+	};
+	struct page_pool *pp_ptr;
+
+	pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
+	if (IS_ERR(pp_ptr))
+		return -ENOMEM;
+
+	per_cpu(system_page_pool, cpuid) = pp_ptr;
+#endif
+	return 0;
+}
+
 /*
  *       This is called single threaded during boot, so no need
  *       to take the rtnl semaphore.
@@ -11776,6 +11839,9 @@ static int __init net_dev_init(void)
 		init_gro_hash(&sd->backlog);
 		sd->backlog.poll = process_backlog;
 		sd->backlog.weight = weight_p;
+
+		if (net_page_pool_create(i))
+			goto out;
 	}
 
 	dev_boot_phase = 0;
@@ -11803,6 +11869,19 @@ static int __init net_dev_init(void)
 	WARN_ON(rc < 0);
 	rc = 0;
 out:
+	if (rc < 0) {
+		for_each_possible_cpu(i) {
+			struct page_pool *pp_ptr;
+
+			pp_ptr = per_cpu(system_page_pool, i);
+			if (!pp_ptr)
+				continue;
+
+			page_pool_destroy(pp_ptr);
+			per_cpu(system_page_pool, i) = NULL;
+		}
+	}
+
 	return rc;
 }
 
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index 4933762e5a6b..89c835fcf094 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -171,13 +171,16 @@ static void page_pool_producer_unlock(struct page_pool *pool,
 }
 
 static int page_pool_init(struct page_pool *pool,
-			  const struct page_pool_params *params)
+			  const struct page_pool_params *params,
+			  int cpuid)
 {
 	unsigned int ring_qsize = 1024; /* Default */
 
 	memcpy(&pool->p, &params->fast, sizeof(pool->p));
 	memcpy(&pool->slow, &params->slow, sizeof(pool->slow));
 
+	pool->cpuid = cpuid;
+
 	/* Validate only known flags were used */
 	if (pool->p.flags & ~(PP_FLAG_ALL))
 		return -EINVAL;
@@ -253,10 +256,12 @@ static void page_pool_uninit(struct page_pool *pool)
 }
 
 /**
- * page_pool_create() - create a page pool.
+ * page_pool_create_percpu() - create a page pool for a given cpu.
  * @params: parameters, see struct page_pool_params
+ * @cpuid: cpu identifier
  */
-struct page_pool *page_pool_create(const struct page_pool_params *params)
+struct page_pool *
+page_pool_create_percpu(const struct page_pool_params *params, int cpuid)
 {
 	struct page_pool *pool;
 	int err;
@@ -265,7 +270,7 @@ struct page_pool *page_pool_create(const struct page_pool_params *params)
 	if (!pool)
 		return ERR_PTR(-ENOMEM);
 
-	err = page_pool_init(pool, params);
+	err = page_pool_init(pool, params, cpuid);
 	if (err < 0)
 		goto err_free;
 
@@ -282,6 +287,16 @@ err_free:
 	kfree(pool);
 	return ERR_PTR(err);
 }
+EXPORT_SYMBOL(page_pool_create_percpu);
+
+/**
+ * page_pool_create() - create a page pool
+ * @params: parameters, see struct page_pool_params
+ */
+struct page_pool *page_pool_create(const struct page_pool_params *params)
+{
+	return page_pool_create_percpu(params, -1);
+}
 EXPORT_SYMBOL(page_pool_create);
 
 static void page_pool_return_page(struct page_pool *pool, struct page *page);
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index edbbef563d4d..0d9a489e6ae1 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -895,6 +895,98 @@ static bool is_pp_page(struct page *page)
 	return (page->pp_magic & ~0x3UL) == PP_SIGNATURE;
 }
 
+int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
+		    unsigned int headroom)
+{
+#if IS_ENABLED(CONFIG_PAGE_POOL)
+	u32 size, truesize, len, max_head_size, off;
+	struct sk_buff *skb = *pskb, *nskb;
+	int err, i, head_off;
+	void *data;
+
+	/* XDP does not support fraglist so we need to linearize
+	 * the skb.
+	 */
+	if (skb_has_frag_list(skb))
+		return -EOPNOTSUPP;
+
+	max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE - headroom);
+	if (skb->len > max_head_size + MAX_SKB_FRAGS * PAGE_SIZE)
+		return -ENOMEM;
+
+	size = min_t(u32, skb->len, max_head_size);
+	truesize = SKB_HEAD_ALIGN(size) + headroom;
+	data = page_pool_dev_alloc_va(pool, &truesize);
+	if (!data)
+		return -ENOMEM;
+
+	nskb = napi_build_skb(data, truesize);
+	if (!nskb) {
+		page_pool_free_va(pool, data, true);
+		return -ENOMEM;
+	}
+
+	skb_reserve(nskb, headroom);
+	skb_copy_header(nskb, skb);
+	skb_mark_for_recycle(nskb);
+
+	err = skb_copy_bits(skb, 0, nskb->data, size);
+	if (err) {
+		consume_skb(nskb);
+		return err;
+	}
+	skb_put(nskb, size);
+
+	head_off = skb_headroom(nskb) - skb_headroom(skb);
+	skb_headers_offset_update(nskb, head_off);
+
+	off = size;
+	len = skb->len - off;
+	for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
+		struct page *page;
+		u32 page_off;
+
+		size = min_t(u32, len, PAGE_SIZE);
+		truesize = size;
+
+		page = page_pool_dev_alloc(pool, &page_off, &truesize);
+		if (!data) {
+			consume_skb(nskb);
+			return -ENOMEM;
+		}
+
+		skb_add_rx_frag(nskb, i, page, page_off, size, truesize);
+		err = skb_copy_bits(skb, off, page_address(page) + page_off,
+				    size);
+		if (err) {
+			consume_skb(nskb);
+			return err;
+		}
+
+		len -= size;
+		off += size;
+	}
+
+	consume_skb(skb);
+	*pskb = nskb;
+
+	return 0;
+#else
+	return -EOPNOTSUPP;
+#endif
+}
+EXPORT_SYMBOL(skb_pp_cow_data);
+
+int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
+			 struct bpf_prog *prog)
+{
+	if (!prog->aux->xdp_has_frags)
+		return -EINVAL;
+
+	return skb_pp_cow_data(pool, pskb, XDP_PACKET_HEADROOM);
+}
+EXPORT_SYMBOL(skb_cow_data_for_xdp);
+
 #if IS_ENABLED(CONFIG_PAGE_POOL)
 bool napi_pp_put_page(struct page *page, bool napi_safe)
 {
@@ -923,9 +1015,10 @@ bool napi_pp_put_page(struct page *page, bool napi_safe)
 	 */
 	if (napi_safe || in_softirq()) {
 		const struct napi_struct *napi = READ_ONCE(pp->p.napi);
+		unsigned int cpuid = smp_processor_id();
 
-		allow_direct = napi &&
-			READ_ONCE(napi->list_owner) == smp_processor_id();
+		allow_direct = napi && READ_ONCE(napi->list_owner) == cpuid;
+		allow_direct |= (pp->cpuid == cpuid);
 	}
 
 	/* Driver set this to memory recycling info. Reset it on recycle.