ip_frag: refactor IPv4 fragmentation into a proper library
authorAnatoly Burakov <anatoly.burakov@intel.com>
Wed, 28 May 2014 17:32:36 +0000 (18:32 +0100)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Mon, 16 Jun 2014 16:55:04 +0000 (18:55 +0200)
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
[Thomas: add in doxygen]
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
doc/doxy-api-index.md
doc/doxy-api.conf
examples/ipv4_frag/main.c
lib/librte_ip_frag/Makefile
lib/librte_ip_frag/rte_ip_frag.h
lib/librte_ip_frag/rte_ipv4_fragmentation.c [new file with mode: 0644]
mk/rte.app.mk

index 83303a1..1b8c7f8 100644 (file)
@@ -78,6 +78,7 @@ There are many libraries, so their headers may be grouped by topics:
   [SCTP]               (@ref rte_sctp.h),
   [TCP]                (@ref rte_tcp.h),
   [UDP]                (@ref rte_udp.h),
+  [frag/reass]         (@ref rte_ip_frag.h),
   [LPM route]          (@ref rte_lpm.h),
   [ACL]                (@ref rte_acl.h)
 
index e5a8520..0af7d37 100644 (file)
@@ -35,6 +35,7 @@ INPUT                   = doc/doxy-api-index.md \
                           lib/librte_distributor \
                           lib/librte_ether \
                           lib/librte_hash \
+                          lib/librte_ip_frag \
                           lib/librte_kni \
                           lib/librte_kvargs \
                           lib/librte_lpm \
index ff60015..88dc1f6 100644 (file)
 #include "rte_ip_frag.h"
 #include "main.h"
 
+/*
+ * Default byte size for the IPv4 Maximum Transfer Unit (MTU).
+ * This value includes the size of IPv4 header.
+ */
+#define        IPV4_MTU_DEFAULT        ETHER_MTU
+
+/*
+ * Default payload in bytes for the IPv4 packet.
+ */
+#define        IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
+
 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 
 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
index 4c353da..95e917d 100644 (file)
 
 include $(RTE_SDK)/mk/rte.vars.mk
 
+# library name
+LIB = librte_ip_frag.a
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS) -I$(SRCDIR)
+
+#source files
+SRCS-$(CONFIG_RTE_LIBRTE_IP_FRAG) += rte_ipv4_fragmentation.c
+
 # install this header file
 SYMLINK-$(CONFIG_RTE_LIBRTE_IP_FRAG)-include += rte_ip_frag.h
 SYMLINK-$(CONFIG_RTE_LIBRTE_IP_FRAG)-include += ipv4_frag_tbl.h
index a96a439..7f7e00a 100644 (file)
@@ -31,9 +31,8 @@
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef __INCLUDE_RTE_IPV4_FRAG_H__
-#define __INCLUDE_RTE_IPV4_FRAG_H__
-#include <rte_ip.h>
+#ifndef _RTE_IP_FRAG_H__
+#define _RTE_IP_FRAG_H__
 
 /**
  * @file
  *
  */
 
-/*
- * Default byte size for the IPv4 Maximum Transfer Unit (MTU).
- * This value includes the size of IPv4 header.
- */
-#define        IPV4_MTU_DEFAULT        ETHER_MTU
-
-/*
- * Default payload in bytes for the IPv4 packet.
- */
-#define        IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
-
-/*
- * MAX number of fragments per packet allowed.
- */
-#define        IPV4_MAX_FRAGS_PER_PACKET       0x80
-
-
-/* Debug on/off */
-#ifdef RTE_IPV4_FRAG_DEBUG
-
-#define        RTE_IPV4_FRAG_ASSERT(exp)                                       \
-if (!(exp))    {                                                       \
-       rte_panic("function %s, line%d\tassert \"" #exp "\" failed\n",  \
-               __func__, __LINE__);                                    \
-}
-
-#else /*RTE_IPV4_FRAG_DEBUG*/
-
-#define RTE_IPV4_FRAG_ASSERT(exp)      do { } while(0)
-
-#endif /*RTE_IPV4_FRAG_DEBUG*/
-
-/* Fragment Offset */
-#define        IPV4_HDR_DF_SHIFT                       14
-#define        IPV4_HDR_MF_SHIFT                       13
-#define        IPV4_HDR_FO_SHIFT                       3
-
-#define        IPV4_HDR_DF_MASK                        (1 << IPV4_HDR_DF_SHIFT)
-#define        IPV4_HDR_MF_MASK                        (1 << IPV4_HDR_MF_SHIFT)
-
-#define        IPV4_HDR_FO_MASK                        ((1 << IPV4_HDR_FO_SHIFT) - 1)
-
-static inline void __fill_ipv4hdr_frag(struct ipv4_hdr *dst,
-               const struct ipv4_hdr *src, uint16_t len, uint16_t fofs,
-               uint16_t dofs, uint32_t mf)
-{
-       rte_memcpy(dst, src, sizeof(*dst));
-       fofs = (uint16_t)(fofs + (dofs >> IPV4_HDR_FO_SHIFT));
-       fofs = (uint16_t)(fofs | mf << IPV4_HDR_MF_SHIFT);
-       dst->fragment_offset = rte_cpu_to_be_16(fofs);
-       dst->total_length = rte_cpu_to_be_16(len);
-       dst->hdr_checksum = 0;
-}
-
-static inline void __free_fragments(struct rte_mbuf *mb[], uint32_t num)
-{
-       uint32_t i;
-       for (i = 0; i != num; i++)
-               rte_pktmbuf_free(mb[i]);
-}
-
 /**
  * IPv4 fragmentation.
  *
@@ -113,6 +51,8 @@ static inline void __free_fragments(struct rte_mbuf *mb[], uint32_t num)
  *   The input packet.
  * @param pkts_out
  *   Array storing the output fragments.
+ * @param nb_pkts_out
+ *   Number of fragments.
  * @param mtu_size
  *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
  *   datagrams. This value includes the size of the IPv4 header.
@@ -123,129 +63,13 @@ static inline void __free_fragments(struct rte_mbuf *mb[], uint32_t num)
  * @return
  *   Upon successful completion - number of output fragments placed
  *   in the pkts_out array.
- *   Otherwise - (-1) * <errno>.
+ *   Otherwise - (-1) * errno.
  */
-static inline int32_t rte_ipv4_fragmentation(struct rte_mbuf *pkt_in,
+int32_t rte_ipv4_fragmentation(struct rte_mbuf *pkt_in,
        struct rte_mbuf **pkts_out,
        uint16_t nb_pkts_out,
        uint16_t mtu_size,
        struct rte_mempool *pool_direct,
-       struct rte_mempool *pool_indirect)
-{
-       struct rte_mbuf *in_seg = NULL;
-       struct ipv4_hdr *in_hdr;
-       uint32_t out_pkt_pos, in_seg_data_pos;
-       uint32_t more_in_segs;
-       uint16_t fragment_offset, flag_offset, frag_size;
-
-       frag_size = (uint16_t)(mtu_size - sizeof(struct ipv4_hdr));
-
-       /* Fragment size should be a multiply of 8. */
-       RTE_IPV4_FRAG_ASSERT((frag_size & IPV4_HDR_FO_MASK) == 0);
-
-       /* Fragment size should be a multiply of 8. */
-       RTE_IPV4_FRAG_ASSERT(IPV4_MAX_FRAGS_PER_PACKET * frag_size >=
-           (uint16_t)(pkt_in->pkt.pkt_len - sizeof (struct ipv4_hdr)));
-
-       in_hdr = (struct ipv4_hdr*) pkt_in->pkt.data;
-       flag_offset = rte_cpu_to_be_16(in_hdr->fragment_offset);
-
-       /* If Don't Fragment flag is set */
-       if (unlikely ((flag_offset & IPV4_HDR_DF_MASK) != 0))
-               return (-ENOTSUP);
-
-       /* Check that pkts_out is big enough to hold all fragments */
-       if (unlikely (frag_size * nb_pkts_out <
-           (uint16_t)(pkt_in->pkt.pkt_len - sizeof (struct ipv4_hdr))))
-               return (-EINVAL);
-
-       in_seg = pkt_in;
-       in_seg_data_pos = sizeof(struct ipv4_hdr);
-       out_pkt_pos = 0;
-       fragment_offset = 0;
-
-       more_in_segs = 1;
-       while (likely(more_in_segs)) {
-               struct rte_mbuf *out_pkt = NULL, *out_seg_prev = NULL;
-               uint32_t more_out_segs;
-               struct ipv4_hdr *out_hdr;
-
-               /* Allocate direct buffer */
-               out_pkt = rte_pktmbuf_alloc(pool_direct);
-               if (unlikely(out_pkt == NULL)) {
-                       __free_fragments(pkts_out, out_pkt_pos);
-                       return (-ENOMEM);
-               }
-
-               /* Reserve space for the IP header that will be built later */
-               out_pkt->pkt.data_len = sizeof(struct ipv4_hdr);
-               out_pkt->pkt.pkt_len = sizeof(struct ipv4_hdr);
-
-               out_seg_prev = out_pkt;
-               more_out_segs = 1;
-               while (likely(more_out_segs && more_in_segs)) {
-                       struct rte_mbuf *out_seg = NULL;
-                       uint32_t len;
-
-                       /* Allocate indirect buffer */
-                       out_seg = rte_pktmbuf_alloc(pool_indirect);
-                       if (unlikely(out_seg == NULL)) {
-                               rte_pktmbuf_free(out_pkt);
-                               __free_fragments(pkts_out, out_pkt_pos);
-                               return (-ENOMEM);
-                       }
-                       out_seg_prev->pkt.next = out_seg;
-                       out_seg_prev = out_seg;
-
-                       /* Prepare indirect buffer */
-                       rte_pktmbuf_attach(out_seg, in_seg);
-                       len = mtu_size - out_pkt->pkt.pkt_len;
-                       if (len > (in_seg->pkt.data_len - in_seg_data_pos)) {
-                               len = in_seg->pkt.data_len - in_seg_data_pos;
-                       }
-                       out_seg->pkt.data = (char*) in_seg->pkt.data + (uint16_t)in_seg_data_pos;
-                       out_seg->pkt.data_len = (uint16_t)len;
-                       out_pkt->pkt.pkt_len = (uint16_t)(len +
-                           out_pkt->pkt.pkt_len);
-                       out_pkt->pkt.nb_segs += 1;
-                       in_seg_data_pos += len;
-
-                       /* Current output packet (i.e. fragment) done ? */
-                       if (unlikely(out_pkt->pkt.pkt_len >= mtu_size)) {
-                               more_out_segs = 0;
-                       }
-
-                       /* Current input segment done ? */
-                       if (unlikely(in_seg_data_pos == in_seg->pkt.data_len)) {
-                               in_seg = in_seg->pkt.next;
-                               in_seg_data_pos = 0;
-
-                               if (unlikely(in_seg == NULL)) {
-                                       more_in_segs = 0;
-                               }
-                       }
-               }
-
-               /* Build the IP header */
-
-               out_hdr = (struct ipv4_hdr*) out_pkt->pkt.data;
-
-               __fill_ipv4hdr_frag(out_hdr, in_hdr,
-                   (uint16_t)out_pkt->pkt.pkt_len,
-                   flag_offset, fragment_offset, more_in_segs);
-
-               fragment_offset = (uint16_t)(fragment_offset +
-                   out_pkt->pkt.pkt_len - sizeof(struct ipv4_hdr));
-
-               out_pkt->ol_flags |= PKT_TX_IP_CKSUM;
-               out_pkt->pkt.vlan_macip.f.l3_len = sizeof(struct ipv4_hdr);
-
-               /* Write the fragment to the output list */
-               pkts_out[out_pkt_pos] = out_pkt;
-               out_pkt_pos ++;
-       }
-
-       return (out_pkt_pos);
-}
+       struct rte_mempool *pool_indirect);
 
 #endif
diff --git a/lib/librte_ip_frag/rte_ipv4_fragmentation.c b/lib/librte_ip_frag/rte_ipv4_fragmentation.c
new file mode 100644 (file)
index 0000000..2d33a7b
--- /dev/null
@@ -0,0 +1,239 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stddef.h>
+#include <stdint.h>
+#include <errno.h>
+
+#include <rte_byteorder.h>
+#include <rte_memcpy.h>
+#include <rte_mempool.h>
+#include <rte_debug.h>
+#include <rte_mbuf.h>
+#include <rte_ip.h>
+
+#include "rte_ip_frag.h"
+
+/*
+ * MAX number of fragments per packet allowed.
+ */
+#define        IPV4_MAX_FRAGS_PER_PACKET       0x80
+
+/* Debug on/off */
+#ifdef RTE_IPV4_FRAG_DEBUG
+
+#define        RTE_IPV4_FRAG_ASSERT(exp)                                       \
+if (!(exp))    {                                                       \
+       rte_panic("function %s, line%d\tassert \"" #exp "\" failed\n",  \
+               __func__, __LINE__);                                    \
+}
+
+#else /*RTE_IPV4_FRAG_DEBUG*/
+
+#define RTE_IPV4_FRAG_ASSERT(exp)      do { } while(0)
+
+#endif /*RTE_IPV4_FRAG_DEBUG*/
+
+/* Fragment Offset */
+#define        IPV4_HDR_DF_SHIFT                       14
+#define        IPV4_HDR_MF_SHIFT                       13
+#define        IPV4_HDR_FO_SHIFT                       3
+
+#define        IPV4_HDR_DF_MASK                        (1 << IPV4_HDR_DF_SHIFT)
+#define        IPV4_HDR_MF_MASK                        (1 << IPV4_HDR_MF_SHIFT)
+
+#define        IPV4_HDR_FO_MASK                        ((1 << IPV4_HDR_FO_SHIFT) - 1)
+
+static inline void __fill_ipv4hdr_frag(struct ipv4_hdr *dst,
+               const struct ipv4_hdr *src, uint16_t len, uint16_t fofs,
+               uint16_t dofs, uint32_t mf)
+{
+       rte_memcpy(dst, src, sizeof(*dst));
+       fofs = (uint16_t)(fofs + (dofs >> IPV4_HDR_FO_SHIFT));
+       fofs = (uint16_t)(fofs | mf << IPV4_HDR_MF_SHIFT);
+       dst->fragment_offset = rte_cpu_to_be_16(fofs);
+       dst->total_length = rte_cpu_to_be_16(len);
+       dst->hdr_checksum = 0;
+}
+
+static inline void __free_fragments(struct rte_mbuf *mb[], uint32_t num)
+{
+       uint32_t i;
+       for (i = 0; i != num; i++)
+               rte_pktmbuf_free(mb[i]);
+}
+
+/**
+ * IPv4 fragmentation.
+ *
+ * This function implements the fragmentation of IPv4 packets.
+ *
+ * @param pkt_in
+ *   The input packet.
+ * @param pkts_out
+ *   Array storing the output fragments.
+ * @param mtu_size
+ *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
+ *   datagrams. This value includes the size of the IPv4 header.
+ * @param pool_direct
+ *   MBUF pool used for allocating direct buffers for the output fragments.
+ * @param pool_indirect
+ *   MBUF pool used for allocating indirect buffers for the output fragments.
+ * @return
+ *   Upon successful completion - number of output fragments placed
+ *   in the pkts_out array.
+ *   Otherwise - (-1) * <errno>.
+ */
+int32_t
+rte_ipv4_fragmentation(struct rte_mbuf *pkt_in,
+       struct rte_mbuf **pkts_out,
+       uint16_t nb_pkts_out,
+       uint16_t mtu_size,
+       struct rte_mempool *pool_direct,
+       struct rte_mempool *pool_indirect)
+{
+       struct rte_mbuf *in_seg = NULL;
+       struct ipv4_hdr *in_hdr;
+       uint32_t out_pkt_pos, in_seg_data_pos;
+       uint32_t more_in_segs;
+       uint16_t fragment_offset, flag_offset, frag_size;
+
+       frag_size = (uint16_t)(mtu_size - sizeof(struct ipv4_hdr));
+
+       /* Fragment size should be a multiply of 8. */
+       RTE_IPV4_FRAG_ASSERT((frag_size & IPV4_HDR_FO_MASK) == 0);
+
+       /* Fragment size should be a multiply of 8. */
+       RTE_IPV4_FRAG_ASSERT(IPV4_MAX_FRAGS_PER_PACKET * frag_size >=
+           (uint16_t)(pkt_in->pkt.pkt_len - sizeof (struct ipv4_hdr)));
+
+       in_hdr = (struct ipv4_hdr*) pkt_in->pkt.data;
+       flag_offset = rte_cpu_to_be_16(in_hdr->fragment_offset);
+
+       /* If Don't Fragment flag is set */
+       if (unlikely ((flag_offset & IPV4_HDR_DF_MASK) != 0))
+               return (-ENOTSUP);
+
+       /* Check that pkts_out is big enough to hold all fragments */
+       if (unlikely (frag_size * nb_pkts_out <
+           (uint16_t)(pkt_in->pkt.pkt_len - sizeof (struct ipv4_hdr))))
+               return (-EINVAL);
+
+       in_seg = pkt_in;
+       in_seg_data_pos = sizeof(struct ipv4_hdr);
+       out_pkt_pos = 0;
+       fragment_offset = 0;
+
+       more_in_segs = 1;
+       while (likely(more_in_segs)) {
+               struct rte_mbuf *out_pkt = NULL, *out_seg_prev = NULL;
+               uint32_t more_out_segs;
+               struct ipv4_hdr *out_hdr;
+
+               /* Allocate direct buffer */
+               out_pkt = rte_pktmbuf_alloc(pool_direct);
+               if (unlikely(out_pkt == NULL)) {
+                       __free_fragments(pkts_out, out_pkt_pos);
+                       return (-ENOMEM);
+               }
+
+               /* Reserve space for the IP header that will be built later */
+               out_pkt->pkt.data_len = sizeof(struct ipv4_hdr);
+               out_pkt->pkt.pkt_len = sizeof(struct ipv4_hdr);
+
+               out_seg_prev = out_pkt;
+               more_out_segs = 1;
+               while (likely(more_out_segs && more_in_segs)) {
+                       struct rte_mbuf *out_seg = NULL;
+                       uint32_t len;
+
+                       /* Allocate indirect buffer */
+                       out_seg = rte_pktmbuf_alloc(pool_indirect);
+                       if (unlikely(out_seg == NULL)) {
+                               rte_pktmbuf_free(out_pkt);
+                               __free_fragments(pkts_out, out_pkt_pos);
+                               return (-ENOMEM);
+                       }
+                       out_seg_prev->pkt.next = out_seg;
+                       out_seg_prev = out_seg;
+
+                       /* Prepare indirect buffer */
+                       rte_pktmbuf_attach(out_seg, in_seg);
+                       len = mtu_size - out_pkt->pkt.pkt_len;
+                       if (len > (in_seg->pkt.data_len - in_seg_data_pos)) {
+                               len = in_seg->pkt.data_len - in_seg_data_pos;
+                       }
+                       out_seg->pkt.data = (char*) in_seg->pkt.data + (uint16_t)in_seg_data_pos;
+                       out_seg->pkt.data_len = (uint16_t)len;
+                       out_pkt->pkt.pkt_len = (uint16_t)(len +
+                           out_pkt->pkt.pkt_len);
+                       out_pkt->pkt.nb_segs += 1;
+                       in_seg_data_pos += len;
+
+                       /* Current output packet (i.e. fragment) done ? */
+                       if (unlikely(out_pkt->pkt.pkt_len >= mtu_size)) {
+                               more_out_segs = 0;
+                       }
+
+                       /* Current input segment done ? */
+                       if (unlikely(in_seg_data_pos == in_seg->pkt.data_len)) {
+                               in_seg = in_seg->pkt.next;
+                               in_seg_data_pos = 0;
+
+                               if (unlikely(in_seg == NULL)) {
+                                       more_in_segs = 0;
+                               }
+                       }
+               }
+
+               /* Build the IP header */
+
+               out_hdr = (struct ipv4_hdr*) out_pkt->pkt.data;
+
+               __fill_ipv4hdr_frag(out_hdr, in_hdr,
+                   (uint16_t)out_pkt->pkt.pkt_len,
+                   flag_offset, fragment_offset, more_in_segs);
+
+               fragment_offset = (uint16_t)(fragment_offset +
+                   out_pkt->pkt.pkt_len - sizeof(struct ipv4_hdr));
+
+               out_pkt->ol_flags |= PKT_TX_IP_CKSUM;
+               out_pkt->pkt.vlan_macip.f.l3_len = sizeof(struct ipv4_hdr);
+
+               /* Write the fragment to the output list */
+               pkts_out[out_pkt_pos] = out_pkt;
+               out_pkt_pos ++;
+       }
+
+       return (out_pkt_pos);
+}
index 914c568..14a26c1 100644 (file)
@@ -117,6 +117,10 @@ ifeq ($(CONFIG_RTE_LIBRTE_MBUF),y)
 LDLIBS += -lrte_mbuf
 endif
 
+ifeq ($(CONFIG_RTE_LIBRTE_IP_FRAG),y)
+LDLIBS += -lrte_ip_frag
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_ETHER),y)
 LDLIBS += -lethdev
 endif