From d3ed6dee73c560fad0a8e152c8e233b3fb3a2e44 Mon Sep 17 00:00:00 2001 From: Paolo Abeni Date: Sat, 1 Feb 2025 19:02:51 +0100 Subject: net: harmonize tstats and dstats After the blamed commits below, some UDP tunnel use dstats for accounting. On the xmit path, all the UDP-base tunnels ends up using iptunnel_xmit_stats() for stats accounting, and the latter assumes the relevant (tunnel) network device uses tstats. The end result is some 'funny' stat report for the mentioned UDP tunnel, e.g. when no packet is actually dropped and a bunch of packets are transmitted: gnv2: mtu 1450 qdisc noqueue \ state UNKNOWN mode DEFAULT group default qlen 1000 link/ether ee:7d:09:87:90:ea brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 14916 23 0 15 0 0 TX: bytes packets errors dropped carrier collsns 0 1566 0 0 0 0 Address the issue ensuring the same binary layout for the overlapping fields of dstats and tstats. While this solution is a bit hackish, is smaller and with no performance pitfall compared to other alternatives i.e. supporting both dstat and tstat in iptunnel_xmit_stats() or reverting the blamed commit. With time we should possibly move all the IP-based tunnel (and virtual devices) to dstats. Fixes: c77200c07491 ("bareudp: Handle stats using NETDEV_PCPU_STAT_DSTATS.") Fixes: 6fa6de302246 ("geneve: Handle stats using NETDEV_PCPU_STAT_DSTATS.") Fixes: be226352e8dc ("vxlan: Handle stats using NETDEV_PCPU_STAT_DSTATS.") Signed-off-by: Paolo Abeni Reviewed-by: Guillaume Nault Link: https://patch.msgid.link/2e1c444cf0f63ae472baff29862c4c869be17031.1738432804.git.pabeni@redhat.com Signed-off-by: Jakub Kicinski --- include/linux/netdevice.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include/linux/netdevice.h') diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index 2a59034a5fa2..03bb584c62cf 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -2904,9 +2904,9 @@ struct pcpu_sw_netstats { struct pcpu_dstats { u64_stats_t rx_packets; u64_stats_t rx_bytes; - u64_stats_t rx_drops; u64_stats_t tx_packets; u64_stats_t tx_bytes; + u64_stats_t rx_drops; u64_stats_t tx_drops; struct u64_stats_sync syncp; } __aligned(8 * sizeof(u64)); -- cgit v1.2.3 From 482ad2a4ace2740ca0ff1cbc8f3c7f862f3ab507 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Wed, 5 Feb 2025 15:51:09 +0000 Subject: net: add dev_net_rcu() helper dev->nd_net can change, readers should either use rcu_read_lock() or RTNL. We currently use a generic helper, dev_net() with no debugging support. We probably have many hidden bugs. Add dev_net_rcu() helper for callers using rcu_read_lock() protection. Signed-off-by: Eric Dumazet Reviewed-by: Kuniyuki Iwashima Link: https://patch.msgid.link/20250205155120.1676781-2-edumazet@google.com Signed-off-by: Jakub Kicinski --- include/linux/netdevice.h | 6 ++++++ include/net/net_namespace.h | 2 +- 2 files changed, 7 insertions(+), 1 deletion(-) (limited to 'include/linux/netdevice.h') diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index 03bb584c62cf..c0a86afb85da 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -2663,6 +2663,12 @@ struct net *dev_net(const struct net_device *dev) return read_pnet(&dev->nd_net); } +static inline +struct net *dev_net_rcu(const struct net_device *dev) +{ + return read_pnet_rcu(&dev->nd_net); +} + static inline void dev_net_set(struct net_device *dev, struct net *net) { diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h index 0f5eb9db0c62..7ba1402ca779 100644 --- a/include/net/net_namespace.h +++ b/include/net/net_namespace.h @@ -398,7 +398,7 @@ static inline struct net *read_pnet(const possible_net_t *pnet) #endif } -static inline struct net *read_pnet_rcu(possible_net_t *pnet) +static inline struct net *read_pnet_rcu(const possible_net_t *pnet) { #ifdef CONFIG_NET_NS return rcu_dereference(pnet->net); -- cgit v1.2.3 From 011b0335903832facca86cd8ed05d7d8d94c9c76 Mon Sep 17 00:00:00 2001 From: Paolo Abeni Date: Thu, 6 Feb 2025 22:28:48 +0100 Subject: Revert "net: skb: introduce and use a single page frag cache" This reverts commit dbae2b062824 ("net: skb: introduce and use a single page frag cache"). The intended goal of such change was to counter a performance regression introduced by commit 3226b158e67c ("net: avoid 32 x truesize under-estimation for tiny skbs"). Unfortunately, the blamed commit introduces another regression for the virtio_net driver. Such a driver calls napi_alloc_skb() with a tiny size, so that the whole head frag could fit a 512-byte block. The single page frag cache uses a 1K fragment for such allocation, and the additional overhead, under small UDP packets flood, makes the page allocator a bottleneck. Thanks to commit bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head"), this revert does not re-introduce the original regression. Actually, in the relevant test on top of this revert, I measure a small but noticeable positive delta, just above noise level. The revert itself required some additional mangling due to the introduction of the SKB_HEAD_ALIGN() helper and local lock infra in the affected code. Suggested-by: Eric Dumazet Fixes: dbae2b062824 ("net: skb: introduce and use a single page frag cache") Signed-off-by: Paolo Abeni Link: https://patch.msgid.link/e649212fde9f0fdee23909ca0d14158d32bb7425.1738877290.git.pabeni@redhat.com Signed-off-by: Jakub Kicinski --- include/linux/netdevice.h | 1 - net/core/dev.c | 17 ++++++++ net/core/skbuff.c | 103 +++------------------------------------------- 3 files changed, 22 insertions(+), 99 deletions(-) (limited to 'include/linux/netdevice.h') diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index c0a86afb85da..365f0e2098d1 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -4115,7 +4115,6 @@ void netif_receive_skb_list(struct list_head *head); gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb); void napi_gro_flush(struct napi_struct *napi, bool flush_old); struct sk_buff *napi_get_frags(struct napi_struct *napi); -void napi_get_frags_check(struct napi_struct *napi); gro_result_t napi_gro_frags(struct napi_struct *napi); static inline void napi_free_frags(struct napi_struct *napi) diff --git a/net/core/dev.c b/net/core/dev.c index b91658e8aedb..55e356a68db6 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -6920,6 +6920,23 @@ netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) list_add_rcu(&napi->dev_list, higher); /* adds after higher */ } +/* Double check that napi_get_frags() allocates skbs with + * skb->head being backed by slab, not a page fragment. + * This is to make sure bug fixed in 3226b158e67c + * ("net: avoid 32 x truesize under-estimation for tiny skbs") + * does not accidentally come back. + */ +static void napi_get_frags_check(struct napi_struct *napi) +{ + struct sk_buff *skb; + + local_bh_disable(); + skb = napi_get_frags(napi); + WARN_ON_ONCE(skb && skb->head_frag); + napi_free_frags(napi); + local_bh_enable(); +} + void netif_napi_add_weight_locked(struct net_device *dev, struct napi_struct *napi, int (*poll)(struct napi_struct *, int), diff --git a/net/core/skbuff.c b/net/core/skbuff.c index a441613a1e6c..6a99c453397f 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -220,67 +220,9 @@ static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) #define NAPI_SKB_CACHE_BULK 16 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) -#if PAGE_SIZE == SZ_4K - -#define NAPI_HAS_SMALL_PAGE_FRAG 1 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc) - -/* specialized page frag allocator using a single order 0 page - * and slicing it into 1K sized fragment. Constrained to systems - * with a very limited amount of 1K fragments fitting a single - * page - to avoid excessive truesize underestimation - */ - -struct page_frag_1k { - void *va; - u16 offset; - bool pfmemalloc; -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp) -{ - struct page *page; - int offset; - - offset = nc->offset - SZ_1K; - if (likely(offset >= 0)) - goto use_frag; - - page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); - if (!page) - return NULL; - - nc->va = page_address(page); - nc->pfmemalloc = page_is_pfmemalloc(page); - offset = PAGE_SIZE - SZ_1K; - page_ref_add(page, offset / SZ_1K); - -use_frag: - nc->offset = offset; - return nc->va + offset; -} -#else - -/* the small page is actually unused in this build; add dummy helpers - * to please the compiler and avoid later preprocessor's conditionals - */ -#define NAPI_HAS_SMALL_PAGE_FRAG 0 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false - -struct page_frag_1k { -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask) -{ - return NULL; -} - -#endif - struct napi_alloc_cache { local_lock_t bh_lock; struct page_frag_cache page; - struct page_frag_1k page_small; unsigned int skb_count; void *skb_cache[NAPI_SKB_CACHE_SIZE]; }; @@ -290,23 +232,6 @@ static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = { .bh_lock = INIT_LOCAL_LOCK(bh_lock), }; -/* Double check that napi_get_frags() allocates skbs with - * skb->head being backed by slab, not a page fragment. - * This is to make sure bug fixed in 3226b158e67c - * ("net: avoid 32 x truesize under-estimation for tiny skbs") - * does not accidentally come back. - */ -void napi_get_frags_check(struct napi_struct *napi) -{ - struct sk_buff *skb; - - local_bh_disable(); - skb = napi_get_frags(napi); - WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag); - napi_free_frags(napi); - local_bh_enable(); -} - void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) { struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); @@ -813,10 +738,8 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) /* If requested length is either too small or too big, * we use kmalloc() for skb->head allocation. - * When the small frag allocator is available, prefer it over kmalloc - * for small fragments */ - if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) || + if (len <= SKB_WITH_OVERHEAD(1024) || len > SKB_WITH_OVERHEAD(PAGE_SIZE) || (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, @@ -826,32 +749,16 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) goto skb_success; } + len = SKB_HEAD_ALIGN(len); + if (sk_memalloc_socks()) gfp_mask |= __GFP_MEMALLOC; local_lock_nested_bh(&napi_alloc_cache.bh_lock); nc = this_cpu_ptr(&napi_alloc_cache); - if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) { - /* we are artificially inflating the allocation size, but - * that is not as bad as it may look like, as: - * - 'len' less than GRO_MAX_HEAD makes little sense - * - On most systems, larger 'len' values lead to fragment - * size above 512 bytes - * - kmalloc would use the kmalloc-1k slab for such values - * - Builds with smaller GRO_MAX_HEAD will very likely do - * little networking, as that implies no WiFi and no - * tunnels support, and 32 bits arches. - */ - len = SZ_1K; - data = page_frag_alloc_1k(&nc->page_small, gfp_mask); - pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small); - } else { - len = SKB_HEAD_ALIGN(len); - - data = page_frag_alloc(&nc->page, len, gfp_mask); - pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); - } + data = page_frag_alloc(&nc->page, len, gfp_mask); + pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); local_unlock_nested_bh(&napi_alloc_cache.bh_lock); if (unlikely(!data)) -- cgit v1.2.3 From 0892b840318daa6ae739b7cdec5ecdfca4006689 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Thu, 13 Feb 2025 08:49:44 -0800 Subject: Reapply "net: skb: introduce and use a single page frag cache" This reverts commit 011b0335903832facca86cd8ed05d7d8d94c9c76. Sabrina reports that the revert may trigger warnings due to intervening changes, especially the ability to rise MAX_SKB_FRAGS. Let's drop it and revisit once that part is also ironed out. Fixes: 011b03359038 ("Revert "net: skb: introduce and use a single page frag cache"") Reported-by: Sabrina Dubroca Link: https://lore.kernel.org/6bf54579233038bc0e76056c5ea459872ce362ab.1739375933.git.pabeni@redhat.com Signed-off-by: Jakub Kicinski --- include/linux/netdevice.h | 1 + net/core/dev.c | 17 -------- net/core/skbuff.c | 103 +++++++++++++++++++++++++++++++++++++++++++--- 3 files changed, 99 insertions(+), 22 deletions(-) (limited to 'include/linux/netdevice.h') diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index 365f0e2098d1..c0a86afb85da 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -4115,6 +4115,7 @@ void netif_receive_skb_list(struct list_head *head); gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb); void napi_gro_flush(struct napi_struct *napi, bool flush_old); struct sk_buff *napi_get_frags(struct napi_struct *napi); +void napi_get_frags_check(struct napi_struct *napi); gro_result_t napi_gro_frags(struct napi_struct *napi); static inline void napi_free_frags(struct napi_struct *napi) diff --git a/net/core/dev.c b/net/core/dev.c index 55e356a68db6..b91658e8aedb 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -6920,23 +6920,6 @@ netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) list_add_rcu(&napi->dev_list, higher); /* adds after higher */ } -/* Double check that napi_get_frags() allocates skbs with - * skb->head being backed by slab, not a page fragment. - * This is to make sure bug fixed in 3226b158e67c - * ("net: avoid 32 x truesize under-estimation for tiny skbs") - * does not accidentally come back. - */ -static void napi_get_frags_check(struct napi_struct *napi) -{ - struct sk_buff *skb; - - local_bh_disable(); - skb = napi_get_frags(napi); - WARN_ON_ONCE(skb && skb->head_frag); - napi_free_frags(napi); - local_bh_enable(); -} - void netif_napi_add_weight_locked(struct net_device *dev, struct napi_struct *napi, int (*poll)(struct napi_struct *, int), diff --git a/net/core/skbuff.c b/net/core/skbuff.c index 6a99c453397f..a441613a1e6c 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -220,9 +220,67 @@ static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) #define NAPI_SKB_CACHE_BULK 16 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) +#if PAGE_SIZE == SZ_4K + +#define NAPI_HAS_SMALL_PAGE_FRAG 1 +#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc) + +/* specialized page frag allocator using a single order 0 page + * and slicing it into 1K sized fragment. Constrained to systems + * with a very limited amount of 1K fragments fitting a single + * page - to avoid excessive truesize underestimation + */ + +struct page_frag_1k { + void *va; + u16 offset; + bool pfmemalloc; +}; + +static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp) +{ + struct page *page; + int offset; + + offset = nc->offset - SZ_1K; + if (likely(offset >= 0)) + goto use_frag; + + page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); + if (!page) + return NULL; + + nc->va = page_address(page); + nc->pfmemalloc = page_is_pfmemalloc(page); + offset = PAGE_SIZE - SZ_1K; + page_ref_add(page, offset / SZ_1K); + +use_frag: + nc->offset = offset; + return nc->va + offset; +} +#else + +/* the small page is actually unused in this build; add dummy helpers + * to please the compiler and avoid later preprocessor's conditionals + */ +#define NAPI_HAS_SMALL_PAGE_FRAG 0 +#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false + +struct page_frag_1k { +}; + +static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask) +{ + return NULL; +} + +#endif + struct napi_alloc_cache { local_lock_t bh_lock; struct page_frag_cache page; + struct page_frag_1k page_small; unsigned int skb_count; void *skb_cache[NAPI_SKB_CACHE_SIZE]; }; @@ -232,6 +290,23 @@ static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = { .bh_lock = INIT_LOCAL_LOCK(bh_lock), }; +/* Double check that napi_get_frags() allocates skbs with + * skb->head being backed by slab, not a page fragment. + * This is to make sure bug fixed in 3226b158e67c + * ("net: avoid 32 x truesize under-estimation for tiny skbs") + * does not accidentally come back. + */ +void napi_get_frags_check(struct napi_struct *napi) +{ + struct sk_buff *skb; + + local_bh_disable(); + skb = napi_get_frags(napi); + WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag); + napi_free_frags(napi); + local_bh_enable(); +} + void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) { struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); @@ -738,8 +813,10 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) /* If requested length is either too small or too big, * we use kmalloc() for skb->head allocation. + * When the small frag allocator is available, prefer it over kmalloc + * for small fragments */ - if (len <= SKB_WITH_OVERHEAD(1024) || + if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) || len > SKB_WITH_OVERHEAD(PAGE_SIZE) || (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, @@ -749,16 +826,32 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) goto skb_success; } - len = SKB_HEAD_ALIGN(len); - if (sk_memalloc_socks()) gfp_mask |= __GFP_MEMALLOC; local_lock_nested_bh(&napi_alloc_cache.bh_lock); nc = this_cpu_ptr(&napi_alloc_cache); + if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) { + /* we are artificially inflating the allocation size, but + * that is not as bad as it may look like, as: + * - 'len' less than GRO_MAX_HEAD makes little sense + * - On most systems, larger 'len' values lead to fragment + * size above 512 bytes + * - kmalloc would use the kmalloc-1k slab for such values + * - Builds with smaller GRO_MAX_HEAD will very likely do + * little networking, as that implies no WiFi and no + * tunnels support, and 32 bits arches. + */ + len = SZ_1K; - data = page_frag_alloc(&nc->page, len, gfp_mask); - pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); + data = page_frag_alloc_1k(&nc->page_small, gfp_mask); + pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small); + } else { + len = SKB_HEAD_ALIGN(len); + + data = page_frag_alloc(&nc->page, len, gfp_mask); + pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); + } local_unlock_nested_bh(&napi_alloc_cache.bh_lock); if (unlikely(!data)) -- cgit v1.2.3