net/9p: set error to EREMOTEIO if trans->write returns zero
[cascardo/linux.git] / net / core / skbuff.c
index 17c6bb5..0583e84 100644 (file)
@@ -55,6 +55,7 @@
 #include <linux/cache.h>
 #include <linux/rtnetlink.h>
 #include <linux/init.h>
+#include <linux/scatterlist.h>
 
 #include <net/protocol.h>
 #include <net/dst.h>
@@ -303,7 +304,7 @@ void kfree_skbmem(struct sk_buff *skb)
                if (atomic_dec_and_test(fclone_ref))
                        kmem_cache_free(skbuff_fclone_cache, other);
                break;
-       };
+       }
 }
 
 /**
@@ -414,9 +415,11 @@ struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
        C(csum);
        C(local_df);
        n->cloned = 1;
+       n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
        n->nohdr = 0;
        C(pkt_type);
        C(ip_summed);
+       skb_copy_queue_mapping(n, skb);
        C(priority);
 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
        C(ipvs_property);
@@ -425,6 +428,10 @@ struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
        n->destructor = NULL;
        C(mark);
        __nf_copy(n, skb);
+#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
+    defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
+       C(nf_trace);
+#endif
 #ifdef CONFIG_NET_SCHED
        C(tc_index);
 #ifdef CONFIG_NET_CLS_ACT
@@ -433,8 +440,8 @@ struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
        n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
        C(iif);
 #endif
-       skb_copy_secmark(n, skb);
 #endif
+       skb_copy_secmark(n, skb);
        C(truesize);
        atomic_set(&n->users, 1);
        C(head);
@@ -458,6 +465,7 @@ static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 #endif
        new->sk         = NULL;
        new->dev        = old->dev;
+       skb_copy_queue_mapping(new, old);
        new->priority   = old->priority;
        new->protocol   = old->protocol;
        new->dst        = dst_clone(old->dst);
@@ -481,6 +489,10 @@ static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
        new->destructor = NULL;
        new->mark       = old->mark;
        __nf_copy(new, old);
+#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
+    defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
+       new->nf_trace   = old->nf_trace;
+#endif
 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
        new->ipvs_property = old->ipvs_property;
 #endif
@@ -643,11 +655,10 @@ int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
 
        /* Copy only real data... and, alas, header. This should be
         * optimized for the cases when header is void. */
-       memcpy(data + nhead, skb->head,
 #ifdef NET_SKBUFF_DATA_USES_OFFSET
-               skb->tail);
+       memcpy(data + nhead, skb->head, skb->tail);
 #else
-               skb->tail - skb->head);
+       memcpy(data + nhead, skb->head, skb->tail - skb->head);
 #endif
        memcpy(data + size, skb_end_pointer(skb),
               sizeof(struct skb_shared_info));
@@ -666,15 +677,17 @@ int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
        skb->data    += off;
 #ifdef NET_SKBUFF_DATA_USES_OFFSET
        skb->end      = size;
+       off           = nhead;
 #else
        skb->end      = skb->head + size;
+#endif
        /* {transport,network,mac}_header and tail are relative to skb->head */
        skb->tail             += off;
        skb->transport_header += off;
        skb->network_header   += off;
        skb->mac_header       += off;
-#endif
        skb->cloned   = 0;
+       skb->hdr_len  = 0;
        skb->nohdr    = 0;
        atomic_set(&skb_shinfo(skb)->dataref, 1);
        return 0;
@@ -734,7 +747,9 @@ struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
         */
        struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
                                      gfp_mask);
+       int oldheadroom = skb_headroom(skb);
        int head_copy_len, head_copy_off;
+       int off = 0;
 
        if (!n)
                return NULL;
@@ -744,7 +759,7 @@ struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
        /* Set the tail pointer and length */
        skb_put(n, skb->len);
 
-       head_copy_len = skb_headroom(skb);
+       head_copy_len = oldheadroom;
        head_copy_off = 0;
        if (newheadroom <= head_copy_len)
                head_copy_len = newheadroom;
@@ -758,6 +773,13 @@ struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
 
        copy_skb_header(n, skb);
 
+#ifdef NET_SKBUFF_DATA_USES_OFFSET
+       off                  = newheadroom - oldheadroom;
+#endif
+       n->transport_header += off;
+       n->network_header   += off;
+       n->mac_header       += off;
+
        return n;
 }
 
@@ -1118,7 +1140,7 @@ fault:
  *     traversing fragment lists and such.
  */
 
-int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
+int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
 {
        int i, copy;
        int start = skb_headlen(skb);
@@ -1129,7 +1151,7 @@ int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
        if ((copy = start - offset) > 0) {
                if (copy > len)
                        copy = len;
-               memcpy(skb->data + offset, from, copy);
+               skb_copy_to_linear_data_offset(skb, offset, from, copy);
                if ((len -= copy) == 0)
                        return 0;
                offset += copy;
@@ -1356,7 +1378,7 @@ void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
        long csstart;
 
        if (skb->ip_summed == CHECKSUM_PARTIAL)
-               csstart = skb_transport_offset(skb);
+               csstart = skb->csum_start - skb_headroom(skb);
        else
                csstart = skb_headlen(skb);
 
@@ -1696,6 +1718,11 @@ next_skb:
                st->stepped_offset += frag->size;
        }
 
+       if (st->frag_data) {
+               kunmap_skb_frag(st->frag_data);
+               st->frag_data = NULL;
+       }
+
        if (st->cur_skb->next) {
                st->cur_skb = st->cur_skb->next;
                st->frag_idx = 0;
@@ -1915,6 +1942,7 @@ struct sk_buff *skb_segment(struct sk_buff *skb, int features)
                tail = nskb;
 
                nskb->dev = skb->dev;
+               skb_copy_queue_mapping(nskb, skb);
                nskb->priority = skb->priority;
                nskb->protocol = skb->protocol;
                nskb->dst = dst_clone(skb->dst);
@@ -2002,6 +2030,190 @@ void __init skb_init(void)
                                                NULL, NULL);
 }
 
+/**
+ *     skb_to_sgvec - Fill a scatter-gather list from a socket buffer
+ *     @skb: Socket buffer containing the buffers to be mapped
+ *     @sg: The scatter-gather list to map into
+ *     @offset: The offset into the buffer's contents to start mapping
+ *     @len: Length of buffer space to be mapped
+ *
+ *     Fill the specified scatter-gather list with mappings/pointers into a
+ *     region of the buffer space attached to a socket buffer.
+ */
+int
+skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
+{
+       int start = skb_headlen(skb);
+       int i, copy = start - offset;
+       int elt = 0;
+
+       if (copy > 0) {
+               if (copy > len)
+                       copy = len;
+               sg[elt].page = virt_to_page(skb->data + offset);
+               sg[elt].offset = (unsigned long)(skb->data + offset) % PAGE_SIZE;
+               sg[elt].length = copy;
+               elt++;
+               if ((len -= copy) == 0)
+                       return elt;
+               offset += copy;
+       }
+
+       for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+               int end;
+
+               BUG_TRAP(start <= offset + len);
+
+               end = start + skb_shinfo(skb)->frags[i].size;
+               if ((copy = end - offset) > 0) {
+                       skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+                       if (copy > len)
+                               copy = len;
+                       sg[elt].page = frag->page;
+                       sg[elt].offset = frag->page_offset+offset-start;
+                       sg[elt].length = copy;
+                       elt++;
+                       if (!(len -= copy))
+                               return elt;
+                       offset += copy;
+               }
+               start = end;
+       }
+
+       if (skb_shinfo(skb)->frag_list) {
+               struct sk_buff *list = skb_shinfo(skb)->frag_list;
+
+               for (; list; list = list->next) {
+                       int end;
+
+                       BUG_TRAP(start <= offset + len);
+
+                       end = start + list->len;
+                       if ((copy = end - offset) > 0) {
+                               if (copy > len)
+                                       copy = len;
+                               elt += skb_to_sgvec(list, sg+elt, offset - start, copy);
+                               if ((len -= copy) == 0)
+                                       return elt;
+                               offset += copy;
+                       }
+                       start = end;
+               }
+       }
+       BUG_ON(len);
+       return elt;
+}
+
+/**
+ *     skb_cow_data - Check that a socket buffer's data buffers are writable
+ *     @skb: The socket buffer to check.
+ *     @tailbits: Amount of trailing space to be added
+ *     @trailer: Returned pointer to the skb where the @tailbits space begins
+ *
+ *     Make sure that the data buffers attached to a socket buffer are
+ *     writable. If they are not, private copies are made of the data buffers
+ *     and the socket buffer is set to use these instead.
+ *
+ *     If @tailbits is given, make sure that there is space to write @tailbits
+ *     bytes of data beyond current end of socket buffer.  @trailer will be
+ *     set to point to the skb in which this space begins.
+ *
+ *     The number of scatterlist elements required to completely map the
+ *     COW'd and extended socket buffer will be returned.
+ */
+int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
+{
+       int copyflag;
+       int elt;
+       struct sk_buff *skb1, **skb_p;
+
+       /* If skb is cloned or its head is paged, reallocate
+        * head pulling out all the pages (pages are considered not writable
+        * at the moment even if they are anonymous).
+        */
+       if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
+           __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
+               return -ENOMEM;
+
+       /* Easy case. Most of packets will go this way. */
+       if (!skb_shinfo(skb)->frag_list) {
+               /* A little of trouble, not enough of space for trailer.
+                * This should not happen, when stack is tuned to generate
+                * good frames. OK, on miss we reallocate and reserve even more
+                * space, 128 bytes is fair. */
+
+               if (skb_tailroom(skb) < tailbits &&
+                   pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
+                       return -ENOMEM;
+
+               /* Voila! */
+               *trailer = skb;
+               return 1;
+       }
+
+       /* Misery. We are in troubles, going to mincer fragments... */
+
+       elt = 1;
+       skb_p = &skb_shinfo(skb)->frag_list;
+       copyflag = 0;
+
+       while ((skb1 = *skb_p) != NULL) {
+               int ntail = 0;
+
+               /* The fragment is partially pulled by someone,
+                * this can happen on input. Copy it and everything
+                * after it. */
+
+               if (skb_shared(skb1))
+                       copyflag = 1;
+
+               /* If the skb is the last, worry about trailer. */
+
+               if (skb1->next == NULL && tailbits) {
+                       if (skb_shinfo(skb1)->nr_frags ||
+                           skb_shinfo(skb1)->frag_list ||
+                           skb_tailroom(skb1) < tailbits)
+                               ntail = tailbits + 128;
+               }
+
+               if (copyflag ||
+                   skb_cloned(skb1) ||
+                   ntail ||
+                   skb_shinfo(skb1)->nr_frags ||
+                   skb_shinfo(skb1)->frag_list) {
+                       struct sk_buff *skb2;
+
+                       /* Fuck, we are miserable poor guys... */
+                       if (ntail == 0)
+                               skb2 = skb_copy(skb1, GFP_ATOMIC);
+                       else
+                               skb2 = skb_copy_expand(skb1,
+                                                      skb_headroom(skb1),
+                                                      ntail,
+                                                      GFP_ATOMIC);
+                       if (unlikely(skb2 == NULL))
+                               return -ENOMEM;
+
+                       if (skb1->sk)
+                               skb_set_owner_w(skb2, skb1->sk);
+
+                       /* Looking around. Are we still alive?
+                        * OK, link new skb, drop old one */
+
+                       skb2->next = skb1->next;
+                       *skb_p = skb2;
+                       kfree_skb(skb1);
+                       skb1 = skb2;
+               }
+               elt++;
+               *trailer = skb1;
+               skb_p = &skb1->next;
+       }
+
+       return elt;
+}
+
 EXPORT_SYMBOL(___pskb_trim);
 EXPORT_SYMBOL(__kfree_skb);
 EXPORT_SYMBOL(kfree_skb);
@@ -2012,7 +2224,6 @@ EXPORT_SYMBOL(pskb_copy);
 EXPORT_SYMBOL(pskb_expand_head);
 EXPORT_SYMBOL(skb_checksum);
 EXPORT_SYMBOL(skb_clone);
-EXPORT_SYMBOL(skb_clone_fraglist);
 EXPORT_SYMBOL(skb_copy);
 EXPORT_SYMBOL(skb_copy_and_csum_bits);
 EXPORT_SYMBOL(skb_copy_and_csum_dev);
@@ -2036,3 +2247,6 @@ EXPORT_SYMBOL(skb_seq_read);
 EXPORT_SYMBOL(skb_abort_seq_read);
 EXPORT_SYMBOL(skb_find_text);
 EXPORT_SYMBOL(skb_append_datato_frags);
+
+EXPORT_SYMBOL_GPL(skb_to_sgvec);
+EXPORT_SYMBOL_GPL(skb_cow_data);