]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet/in_cksum.c
xnu-2782.30.5.tar.gz
[apple/xnu.git] / bsd / netinet / in_cksum.c
index 850273c86516f63b09bfdb28a1aefafa594d788d..4e6d453cf434cd64cc2353f53d6a80cfb19a3129 100644 (file)
@@ -1,16 +1,19 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * Copyright (c) 1988, 1992, 1993
  */
 
 #include <sys/param.h>
+#include <machine/endian.h>
 #include <sys/mbuf.h>
-#include <sys/kdebug.h>
-
-#define DBG_FNC_IN_CKSUM       NETDBG_CODE(DBG_NETIP, (3 << 8))
+#include <kern/debug.h>
+#include <net/dlil.h>
+#include <netinet/in.h>
+#define        _IP_VHL
+#include <netinet/ip.h>
+#include <netinet/ip_var.h>
 
 /*
  * Checksum routine for Internet Protocol family headers (Portable Version).
  * This routine is very heavily used in the network
  * code and should be modified for each CPU to be as fast as possible.
  */
-
-union s_util {
-        char    c[2];
-        u_short s;
-};
+#define REDUCE16 {                                                       \
+       q_util.q = sum;                                                   \
+       l_util.l = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
+       sum = l_util.s[0] + l_util.s[1];                                  \
+       ADDCARRY(sum);                                                    \
+}
 
 union l_util {
-        u_int16_t s[2];
-        u_int32_t l;   
+        uint16_t s[2];
+        uint32_t l;
 };
 
 union q_util {
-        u_int16_t s[4];
-        u_int32_t l[2];
-        u_int64_t q;
-};    
-
-#define ADDCARRY(x)  (x > 65535 ? x -= 65535 : x)
-
-#define REDUCE32                                                          \
-    {                                                                     \
-        q_util.q = sum;                                                   \
-        sum = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3];      \
-    }
-#define REDUCE16                                                          \
-    {                                                                     \
-        q_util.q = sum;                                                   \
-        l_util.l = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
-        sum = l_util.s[0] + l_util.s[1];                                  \
-        ADDCARRY(sum);                                                    \
-    }
-
-#define REDUCE {l_util.l = sum; sum = l_util.s[0] + l_util.s[1]; ADDCARRY(sum);}
-
-                
-#if defined(ppc)
-
-__inline unsigned short
-in_addword(u_short a, u_short b)
-{
-        union l_util l_util;   
-       u_int32_t sum = a + b;
-       REDUCE;
-       return (sum);
-}
+        uint16_t s[4];
+        uint32_t l[2];
+        uint64_t q;
+};
 
-__inline unsigned short
-in_pseudo(u_int a, u_int b, u_int c)
-{
-        u_int64_t sum;
-        union q_util q_util;
-        union l_util l_util;   
+#define        PREDICT_FALSE(_exp)     __builtin_expect((_exp), 0)
 
-        sum = (u_int64_t) a + b + c;
-        REDUCE16;
-        return (sum);
+static uint16_t in_cksumdata(const void *buf, int len);
 
-}
+/*
+ * Portable version of 16-bit 1's complement sum function that works
+ * on a contiguous buffer.  This is used mainly for instances where
+ * the caller is certain about the buffer requirements, e.g. for IP
+ * header checksum calculation, though it is capable of being used
+ * on any arbitrary data span.  The platform-specific cpu_in_cksum()
+ * routine might be better-optmized, so use that instead for large
+ * data span.
+ *
+ * The logic is borrowed from <bsd/netinet/cpu_in_cksum.c>
+ */
 
-int
-in_cksum(m, len)
-       register struct mbuf *m;
-       register int len;
+#if ULONG_MAX == 0xffffffffUL
+/* 32-bit version */
+static uint16_t
+in_cksumdata(const void *buf, int mlen)
 {
-       register u_short *w;
-       register int sum = 0;
-       register int mlen = 0;
-       int starting_on_odd  = 0;
-
-
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_START, len,0,0,0,0);
-
-       for (;m && len; m = m->m_next) {
-               if (m->m_len == 0)
-                       continue;
-               mlen = m->m_len;
-               w = mtod(m, u_short *);
-
-               if (len < mlen)
-                       mlen = len;
-
-               sum = xsum_assym(w, mlen, sum, starting_on_odd);
-               len -= mlen;
-               if (mlen & 0x1)
-               {
-                   if (starting_on_odd)
-                       starting_on_odd = 0;
-                   else
-                       starting_on_odd = 1;
+       uint32_t sum, partial;
+       unsigned int final_acc;
+       uint8_t *data = (void *)buf;
+       boolean_t needs_swap, started_on_odd;
+
+       VERIFY(mlen >= 0);
+
+       needs_swap = FALSE;
+       started_on_odd = FALSE;
+
+       sum = 0;
+       partial = 0;
+
+       if ((uintptr_t)data & 1) {
+               /* Align on word boundary */
+               started_on_odd = !started_on_odd;
+#if BYTE_ORDER == LITTLE_ENDIAN
+               partial = *data << 8;
+#else
+               partial = *data;
+#endif
+               ++data;
+               --mlen;
+       }
+       needs_swap = started_on_odd;
+       while (mlen >= 32) {
+               __builtin_prefetch(data + 32);
+               partial += *(uint16_t *)(void *)data;
+               partial += *(uint16_t *)(void *)(data + 2);
+               partial += *(uint16_t *)(void *)(data + 4);
+               partial += *(uint16_t *)(void *)(data + 6);
+               partial += *(uint16_t *)(void *)(data + 8);
+               partial += *(uint16_t *)(void *)(data + 10);
+               partial += *(uint16_t *)(void *)(data + 12);
+               partial += *(uint16_t *)(void *)(data + 14);
+               partial += *(uint16_t *)(void *)(data + 16);
+               partial += *(uint16_t *)(void *)(data + 18);
+               partial += *(uint16_t *)(void *)(data + 20);
+               partial += *(uint16_t *)(void *)(data + 22);
+               partial += *(uint16_t *)(void *)(data + 24);
+               partial += *(uint16_t *)(void *)(data + 26);
+               partial += *(uint16_t *)(void *)(data + 28);
+               partial += *(uint16_t *)(void *)(data + 30);
+               data += 32;
+               mlen -= 32;
+               if (PREDICT_FALSE(partial & 0xc0000000)) {
+                       if (needs_swap)
+                               partial = (partial << 8) +
+                                   (partial >> 24);
+                       sum += (partial >> 16);
+                       sum += (partial & 0xffff);
+                       partial = 0;
                }
        }
+       if (mlen & 16) {
+               partial += *(uint16_t *)(void *)data;
+               partial += *(uint16_t *)(void *)(data + 2);
+               partial += *(uint16_t *)(void *)(data + 4);
+               partial += *(uint16_t *)(void *)(data + 6);
+               partial += *(uint16_t *)(void *)(data + 8);
+               partial += *(uint16_t *)(void *)(data + 10);
+               partial += *(uint16_t *)(void *)(data + 12);
+               partial += *(uint16_t *)(void *)(data + 14);
+               data += 16;
+               mlen -= 16;
+       }
+       /*
+        * mlen is not updated below as the remaining tests
+        * are using bit masks, which are not affected.
+        */
+       if (mlen & 8) {
+               partial += *(uint16_t *)(void *)data;
+               partial += *(uint16_t *)(void *)(data + 2);
+               partial += *(uint16_t *)(void *)(data + 4);
+               partial += *(uint16_t *)(void *)(data + 6);
+               data += 8;
+       }
+       if (mlen & 4) {
+               partial += *(uint16_t *)(void *)data;
+               partial += *(uint16_t *)(void *)(data + 2);
+               data += 4;
+       }
+       if (mlen & 2) {
+               partial += *(uint16_t *)(void *)data;
+               data += 2;
+       }
+       if (mlen & 1) {
+#if BYTE_ORDER == LITTLE_ENDIAN
+               partial += *data;
+#else
+               partial += *data << 8;
+#endif
+               started_on_odd = !started_on_odd;
+       }
 
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_END, 0,0,0,0,0);
-       return (~sum & 0xffff);
+       if (needs_swap)
+               partial = (partial << 8) + (partial >> 24);
+       sum += (partial >> 16) + (partial & 0xffff);
+       sum = (sum >> 16) + (sum & 0xffff);
+
+       final_acc = ((sum >> 16) & 0xffff) + (sum & 0xffff);
+       final_acc = (final_acc >> 16) + (final_acc & 0xffff);
+
+       return (final_acc);
 }
 
-u_short
-in_cksum_skip(m, len, skip)
-        register struct mbuf *m;
-        register int len;
-        register int skip;
+#else
+/* 64-bit version */
+static uint16_t
+in_cksumdata(const void *buf, int mlen)
 {
-       register u_short *w;
-       register int sum = 0;
-       register int mlen = 0;
-       int starting_on_odd  = 0;
-
-       len -= skip;
-        for (; skip && m; m = m->m_next) {
-                if (m->m_len > skip) {
-                        mlen = m->m_len - skip;
-                       w = (u_short *)(m->m_data+skip);
-                        goto skip_start;
-                } else {    
-                        skip -= m->m_len;
-                }
-        }
-       for (;m && len; m = m->m_next) {
-               if (m->m_len == 0)
-                       continue;
-               mlen = m->m_len;
-               w = mtod(m, u_short *);
-
-skip_start:
-               if (len < mlen)
-                       mlen = len;
-               sum = xsum_assym(w, mlen, sum, starting_on_odd);
-               len -= mlen;
-               if (mlen & 0x1)
-               {
-                   if (starting_on_odd)
-                       starting_on_odd = 0;
-                   else
-                       starting_on_odd = 1;
+       uint64_t sum, partial;
+       unsigned int final_acc;
+       uint8_t *data = (void *)buf;
+       boolean_t needs_swap, started_on_odd;
+
+       VERIFY(mlen >= 0);
+
+       needs_swap = FALSE;
+       started_on_odd = FALSE;
+
+       sum = 0;
+       partial = 0;
+
+       if ((uintptr_t)data & 1) {
+               /* Align on word boundary */
+               started_on_odd = !started_on_odd;
+#if BYTE_ORDER == LITTLE_ENDIAN
+               partial = *data << 8;
+#else
+               partial = *data;
+#endif
+               ++data;
+               --mlen;
+       }
+       needs_swap = started_on_odd;
+       if ((uintptr_t)data & 2) {
+               if (mlen < 2)
+                       goto trailing_bytes;
+               partial += *(uint16_t *)(void *)data;
+               data += 2;
+               mlen -= 2;
+       }
+       while (mlen >= 64) {
+               __builtin_prefetch(data + 32);
+               __builtin_prefetch(data + 64);
+               partial += *(uint32_t *)(void *)data;
+               partial += *(uint32_t *)(void *)(data + 4);
+               partial += *(uint32_t *)(void *)(data + 8);
+               partial += *(uint32_t *)(void *)(data + 12);
+               partial += *(uint32_t *)(void *)(data + 16);
+               partial += *(uint32_t *)(void *)(data + 20);
+               partial += *(uint32_t *)(void *)(data + 24);
+               partial += *(uint32_t *)(void *)(data + 28);
+               partial += *(uint32_t *)(void *)(data + 32);
+               partial += *(uint32_t *)(void *)(data + 36);
+               partial += *(uint32_t *)(void *)(data + 40);
+               partial += *(uint32_t *)(void *)(data + 44);
+               partial += *(uint32_t *)(void *)(data + 48);
+               partial += *(uint32_t *)(void *)(data + 52);
+               partial += *(uint32_t *)(void *)(data + 56);
+               partial += *(uint32_t *)(void *)(data + 60);
+               data += 64;
+               mlen -= 64;
+               if (PREDICT_FALSE(partial & (3ULL << 62))) {
+                       if (needs_swap)
+                               partial = (partial << 8) +
+                                   (partial >> 56);
+                       sum += (partial >> 32);
+                       sum += (partial & 0xffffffff);
+                       partial = 0;
                }
        }
+       /*
+        * mlen is not updated below as the remaining tests
+        * are using bit masks, which are not affected.
+        */
+       if (mlen & 32) {
+               partial += *(uint32_t *)(void *)data;
+               partial += *(uint32_t *)(void *)(data + 4);
+               partial += *(uint32_t *)(void *)(data + 8);
+               partial += *(uint32_t *)(void *)(data + 12);
+               partial += *(uint32_t *)(void *)(data + 16);
+               partial += *(uint32_t *)(void *)(data + 20);
+               partial += *(uint32_t *)(void *)(data + 24);
+               partial += *(uint32_t *)(void *)(data + 28);
+               data += 32;
+       }
+       if (mlen & 16) {
+               partial += *(uint32_t *)(void *)data;
+               partial += *(uint32_t *)(void *)(data + 4);
+               partial += *(uint32_t *)(void *)(data + 8);
+               partial += *(uint32_t *)(void *)(data + 12);
+               data += 16;
+       }
+       if (mlen & 8) {
+               partial += *(uint32_t *)(void *)data;
+               partial += *(uint32_t *)(void *)(data + 4);
+               data += 8;
+       }
+       if (mlen & 4) {
+               partial += *(uint32_t *)(void *)data;
+               data += 4;
+       }
+       if (mlen & 2) {
+               partial += *(uint16_t *)(void *)data;
+               data += 2;
+       }
+trailing_bytes:
+       if (mlen & 1) {
+#if BYTE_ORDER == LITTLE_ENDIAN
+               partial += *data;
+#else
+               partial += *data << 8;
+#endif
+               started_on_odd = !started_on_odd;
+       }
 
-       return (~sum & 0xffff);
+       if (needs_swap)
+               partial = (partial << 8) + (partial >> 56);
+       sum += (partial >> 32) + (partial & 0xffffffff);
+       sum = (sum >> 32) + (sum & 0xffffffff);
+
+       final_acc = (sum >> 48) + ((sum >> 32) & 0xffff) +
+           ((sum >> 16) & 0xffff) + (sum & 0xffff);
+       final_acc = (final_acc >> 16) + (final_acc & 0xffff);
+       final_acc = (final_acc >> 16) + (final_acc & 0xffff);
+
+       return (final_acc);
 }
-#else
+#endif /* ULONG_MAX != 0xffffffffUL */
 
-u_short 
-in_addword(u_short a, u_short b)
-{       
-        union l_util l_util;   
-        u_int32_t sum = a + b;
-        REDUCE(sum);
-        return (sum);
-}       
+/*
+ * Perform 16-bit 1's complement sum on a contiguous span.
+ */
+uint16_t
+b_sum16(const void *buf, int len)
+{
+       return (in_cksumdata(buf, len));
+}
+
+uint16_t inet_cksum_simple(struct mbuf *, int);
+/*
+ * For the exported _in_cksum symbol in BSDKernel symbol set.
+ */
+uint16_t
+inet_cksum_simple(struct mbuf *m, int len)
+{
+       return (inet_cksum(m, 0, 0, len));
+}
+
+uint16_t
+in_addword(uint16_t a, uint16_t b)
+{
+       uint64_t sum = a + b;
+
+       ADDCARRY(sum);
+       return (sum);
+}
 
-u_short
-in_pseudo(u_int a, u_int b, u_int c)
+uint16_t
+in_pseudo(uint32_t a, uint32_t b, uint32_t c)
 {
-        u_int64_t sum;  
+        uint64_t sum;
         union q_util q_util;
-        union l_util l_util;   
+        union l_util l_util;
 
-        sum = (u_int64_t) a + b + c;
+        sum = (uint64_t)a + b + c;
         REDUCE16;
         return (sum);
 }
 
+uint16_t
+in_pseudo64(uint64_t a, uint64_t b, uint64_t c)
+{
+       uint64_t sum;
+       union q_util q_util;
+       union l_util l_util;
 
-int
-in_cksum(m, len)
-       register struct mbuf *m;
-       register int len;
+       sum = a + b + c;
+       REDUCE16;
+       return (sum);
+}
+
+/*
+ * May be used on IP header with options.
+ */
+uint16_t
+in_cksum_hdr_opt(const struct ip *ip)
 {
-       register u_short *w;
-       register int sum = 0;
-       register int mlen = 0;
-       int byte_swapped = 0;
-       union s_util s_util;
-       union l_util l_util;   
-
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_START, len,0,0,0,0);
-
-       for (;m && len; m = m->m_next) {
-               if (m->m_len == 0)
-                       continue;
-               w = mtod(m, u_short *);
-               if (mlen == -1) {
-                       /*
-                        * The first byte of this mbuf is the continuation
-                        * of a word spanning between this mbuf and the
-                        * last mbuf.
-                        *
-                        * s_util.c[0] is already saved when scanning previous
-                        * mbuf.
-                        */
-                       s_util.c[1] = *(char *)w;
-                       sum += s_util.s;
-                       w = (u_short *)((char *)w + 1);
-                       mlen = m->m_len - 1;
-                       len--;
-               } else
-                       mlen = m->m_len;
-               if (len < mlen)
-                       mlen = len;
-               len -= mlen;
-               /*
-                * Force to even boundary.
-                */
-               if ((1 & (int) w) && (mlen > 0)) {
-                       REDUCE;
-                       sum <<= 8;
-                       s_util.c[0] = *(u_char *)w;
-                       w = (u_short *)((char *)w + 1);
-                       mlen--;
-                       byte_swapped = 1;
-               }
-               /*
-                * Unroll the loop to make overhead from
-                * branches &c small.
-                */
-               while ((mlen -= 32) >= 0) {
-                       sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
-                       sum += w[4]; sum += w[5]; sum += w[6]; sum += w[7];
-                       sum += w[8]; sum += w[9]; sum += w[10]; sum += w[11];
-                       sum += w[12]; sum += w[13]; sum += w[14]; sum += w[15];
-                       w += 16;
-               }
-               mlen += 32;
-               while ((mlen -= 8) >= 0) {
-                       sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
-                       w += 4;
-               }
-               mlen += 8;
-               if (mlen == 0 && byte_swapped == 0)
-                       continue;
-               REDUCE;
-               while ((mlen -= 2) >= 0) {
-                       sum += *w++;
-               }
-               if (byte_swapped) {
-                       REDUCE;
-                       sum <<= 8;
-                       byte_swapped = 0;
-                       if (mlen == -1) {
-                               s_util.c[1] = *(char *)w;
-                               sum += s_util.s;
-                               mlen = 0;
-                       } else
-                               mlen = -1;
-               } else if (mlen == -1)
-                       s_util.c[0] = *(char *)w;
-       }
-       if (len)
-               printf("cksum: out of data\n");
-       if (mlen == -1) {
-               /* The last mbuf has odd # of bytes. Follow the
-                  standard (the odd byte may be shifted left by 8 bits
-                  or not as determined by endian-ness of the machine) */
-               s_util.c[1] = 0;
-               sum += s_util.s;
+       return (~b_sum16(ip, (IP_VHL_HL(ip->ip_vhl) << 2)) & 0xffff);
+}
+
+/*
+ * A wrapper around the simple in_cksum_hdr() and the more complicated
+ * inet_cksum(); the former is chosen if the IP header is simple,
+ * contiguous and 32-bit aligned.  Also does some stats accounting.
+ */
+uint16_t
+ip_cksum_hdr_dir(struct mbuf *m, uint32_t hlen, int out)
+{
+       struct ip *ip = mtod(m, struct ip *);
+
+       if (out) {
+               ipstat.ips_snd_swcsum++;
+               ipstat.ips_snd_swcsum_bytes += hlen;
+       } else {
+               ipstat.ips_rcv_swcsum++;
+               ipstat.ips_rcv_swcsum_bytes += hlen;
        }
-       REDUCE;
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_END, 0,0,0,0,0);
-       return (~sum & 0xffff);
+
+       if (hlen == sizeof (*ip) &&
+           m->m_len >= sizeof (*ip) && IP_HDR_ALIGNED_P(ip))
+               return (in_cksum_hdr(ip));
+
+       return (inet_cksum(m, 0, 0, hlen));
 }
 
-int
-in_cksum_skip(m, len, skip)
-       register struct mbuf *m;
-       register u_short len;
-       register u_short skip;
+/*
+ * m MUST contain at least an IP header, if nxt is specified;
+ * nxt is the upper layer protocol number;
+ * off is an offset where TCP/UDP/ICMP header starts;
+ * len is a total length of a transport segment (e.g. TCP header + TCP payload)
+ */
+uint16_t
+inet_cksum(struct mbuf *m, uint32_t nxt, uint32_t off, uint32_t len)
 {
-       register u_short *w;
-       register int sum = 0;
-       register int mlen = 0;
-       int byte_swapped = 0;
-       union s_util s_util;
-       union l_util l_util;   
-
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_START, len,0,0,0,0);
-
-       len -= skip;
-        for (; skip && m; m = m->m_next) {
-                if (m->m_len > skip) {
-                        mlen = m->m_len - skip;
-                       w = (u_short *)(m->m_data+skip);
-                        goto skip_start;
-                } else {    
-                        skip -= m->m_len;
-                }
-        }
-       for (;m && len; m = m->m_next) {
-               if (m->m_len == 0)
-                       continue;
-               w = mtod(m, u_short *);
-
-               if (mlen == -1) {
-                       /*
-                        * The first byte of this mbuf is the continuation
-                        * of a word spanning between this mbuf and the
-                        * last mbuf.
-                        *
-                        * s_util.c[0] is already saved when scanning previous
-                        * mbuf.
-                        */
-                       s_util.c[1] = *(char *)w;
-                       sum += s_util.s;
-                       w = (u_short *)((char *)w + 1);
-                       mlen = m->m_len - 1;
-                       len--;
-               } else {
-                 mlen = m->m_len;
-               }
-skip_start:
-               if (len < mlen)
-                   mlen = len;
+       uint32_t sum;
+
+       sum = m_sum16(m, off, len);
+
+       /* include pseudo header checksum? */
+       if (nxt != 0) {
+               struct ip *ip;
+               unsigned char buf[sizeof ((*ip))] __attribute__((aligned(8)));
+               uint32_t mlen;
 
-               len -= mlen;
                /*
-                * Force to even boundary.
+                * Sanity check
+                *
+                * Use m_length2() instead of m_length(), as we cannot rely on
+                * the caller setting m_pkthdr.len correctly, if the mbuf is
+                * a M_PKTHDR one.
                 */
-               if ((1 & (int) w) && (mlen > 0)) {
-                       REDUCE;
-                       sum <<= 8;
-                       s_util.c[0] = *(u_char *)w;
-                       w = (u_short *)((char *)w + 1);
-                       mlen--;
-                       byte_swapped = 1;
+               if ((mlen = m_length2(m, NULL)) < sizeof (*ip)) {
+                       panic("%s: mbuf %p too short (%d) for IPv4 header",
+                           __func__, m, mlen);
+                       /* NOTREACHED */
                }
+
                /*
-                * Unroll the loop to make overhead from
-                * branches &c small.
+                * In case the IP header is not contiguous, or not 32-bit
+                * aligned, copy it to a local buffer.  Note here that we
+                * expect the data pointer to point to the IP header.
                 */
-               while ((mlen -= 32) >= 0) {
-                       sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
-                       sum += w[4]; sum += w[5]; sum += w[6]; sum += w[7];
-                       sum += w[8]; sum += w[9]; sum += w[10]; sum += w[11];
-                       sum += w[12]; sum += w[13]; sum += w[14]; sum += w[15];
-                       w += 16;
-               }
-               mlen += 32;
-               while ((mlen -= 8) >= 0) {
-                       sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
-                       w += 4;
-               }
-               mlen += 8;
-               if (mlen == 0 && byte_swapped == 0)
-                       continue;
-               REDUCE;
-               while ((mlen -= 2) >= 0) {
-                       sum += *w++;
+               if ((sizeof (*ip) > m->m_len) ||
+                   !IP_HDR_ALIGNED_P(mtod(m, caddr_t))) {
+                       m_copydata(m, 0, sizeof (*ip), (caddr_t)buf);
+                       ip = (struct ip *)(void *)buf;
+               } else {
+                       ip = (struct ip *)(void *)(m->m_data);
                }
-               if (byte_swapped) {
-                       REDUCE;
-                       sum <<= 8;
-                       byte_swapped = 0;
-                       if (mlen == -1) {
-                               s_util.c[1] = *(char *)w;
-                               sum += s_util.s;
-                               mlen = 0;
-                       } else
-                               mlen = -1;
-               } else if (mlen == -1)
-                       s_util.c[0] = *(char *)w;
-       }
-       if (len)
-               printf("cksum: out of data\n");
-       if (mlen == -1) {
-               /* The last mbuf has odd # of bytes. Follow the
-                  standard (the odd byte may be shifted left by 8 bits
-                  or not as determined by endian-ness of the machine) */
-               s_util.c[1] = 0;
-               sum += s_util.s;
+
+               /* add pseudo header checksum */
+               sum += in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
+                   htonl(len + nxt));
+
+               /* fold in carry bits */
+               ADDCARRY(sum);
        }
-       REDUCE;
-       KERNEL_DEBUG(DBG_FNC_IN_CKSUM | DBG_FUNC_END, 0,0,0,0,0);
+
        return (~sum & 0xffff);
 }
-
-#endif