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 <sd@queasysnail.net>
Link: https://lore.kernel.org/6bf54579233038bc0e76056c5ea459872ce362ab.1739375933.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 365f0e2098d13f40ce6d8865962678b052b39a16..c0a86afb85daa2b50e26a1ca238707a24a1842ad 100644 (file)
--- 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 55e356a68db667982e7e62d09d07feecc14deebe..b91658e8aedb41de57c0b6531183c454ed770b2b 100644 (file)
--- 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 6a99c453397fcf7a5762cda5200ecee3468fd2e3..a441613a1e6c1765f7fc2e40f982b81f8f8fdb96 100644 (file)
--- 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))