path: root/extra/quagga/quagga-0.99.11-checksum.patch

From: Paul Jakma <paul.jakma@sun.com>
Date: Sun, 16 Nov 2008 18:34:19 +0000 (+0000)
Subject: [lib] Switch Fletcher checksum back to old ospfd version
X-Git-Url: http://code.quagga.net/cgi-bin/gitweb.cgi?p=quagga.git;a=commitdiff_plain;h=5d4b8cf2faba9f5386810a7c70837e5b7fae3572

[lib] Switch Fletcher checksum back to old ospfd version

* lib/checksum.c: (fletcher_checksum) Switch the second phase of the checksum
  back to the old ospfd logic.

  The isisd-derived version:

  a) is very hard to follow
  b) had some kind of subtle bug that caused it be wrong when c0=0 and c1=254
     (potentially fixable by doing the mods before adjusting x and y)

  Additionally:

  - explicitely cast expressions using non-internal variables to int, to ensure
    the result is signed.
  - defensively change the length argument to 'size_t', to ensure the code
    works with that argument being unsigned..

  Thanks to Joakim Tjernlund for the investigative work into this bug.

* tests/test-checksum.c: new file to exercise the checksum code.
---

diff --git a/lib/checksum.c b/lib/checksum.c
index 88ec72a..f6d74d3 100644
--- a/lib/checksum.c
+++ b/lib/checksum.c
@@ -52,34 +52,31 @@ in_cksum(void *parg, int nbytes)
 /* To be consistent, offset is 0-based index, rather than the 1-based 
    index required in the specification ISO 8473, Annex C.1 */
 u_int16_t
-fletcher_checksum(u_char * buffer, int len, u_int16_t offset)
+fletcher_checksum(u_char * buffer, const size_t len, const uint16_t offset)
 {
   u_int8_t *p;
-  int x;
-  int y;
-  u_int32_t mul;
-  u_int32_t c0;
-  u_int32_t c1;
+  int x, y, c0, c1;
   u_int16_t checksum;
   u_int16_t *csum;
-  int i, init_len, partial_len;
-
+  size_t partial_len, i, left = len;
+  
   checksum = 0;
 
+  assert (offset < len);
+
   /*
    * Zero the csum in the packet.
    */
   csum = (u_int16_t *) (buffer + offset);
-  *(csum) = checksum;
+  *(csum) = 0;
 
   p = buffer;
   c0 = 0;
   c1 = 0;
-  init_len = len;
 
-  while (len != 0)
+  while (left != 0)
     {
-      partial_len = MIN(len, MODX);
+      partial_len = MIN(left, MODX);
 
       for (i = 0; i < partial_len; i++)
 	{
@@ -90,27 +87,18 @@ fletcher_checksum(u_char * buffer, int len, u_int16_t offset)
       c0 = c0 % 255;
       c1 = c1 % 255;
 
-      len -= partial_len;
+      left -= partial_len;
     }
-
-  mul = (init_len - offset)*(c0);
-
-  x = mul - c0 - c1;
-  y = c1 - mul - 1;
-
-  if (y > 0)
-    y++;
-  if (x < 0)
-    x--;
-
-  x %= 255;
-  y %= 255;
-
-  if (x == 0)
-    x = 255;
-  if (y == 0)
-    y = 1;
-
+  
+  /* The cast is important, to ensure the mod is taken as a signed value. */
+  x = ((int)(len - offset - 1) * c0 - c1) % 255;
+
+  if (x <= 0)
+    x += 255;
+  y = 510 - c0 - x;
+  if (y > 255)  
+    y -= 255;
+  
   /*
    * Now we write this to the packet.
    * We could skip this step too, since the checksum returned would
diff --git a/lib/checksum.h b/lib/checksum.h
index d3ce930..da1d3cb 100644
--- a/lib/checksum.h
+++ b/lib/checksum.h
@@ -1,2 +1,2 @@
 extern int in_cksum(void *, int);
-extern u_int16_t fletcher_checksum(u_char * buffer, int len, u_int16_t offset);
+extern u_int16_t fletcher_checksum(u_char *, const size_t len, const uint16_t offset);
diff --git a/tests/Makefile.am b/tests/Makefile.am
index c93fa08..4ab507b 100644
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
@@ -6,7 +6,7 @@ AM_LDFLAGS = $(PILDFLAGS)
 
 noinst_PROGRAMS = testsig testbuffer testmemory heavy heavywq heavythread \
 		aspathtest testprivs teststream testbgpcap ecommtest \
-		testbgpmpattr
+		testbgpmpattr testchecksum
 
 testsig_SOURCES = test-sig.c
 testbuffer_SOURCES = test-buffer.c
@@ -20,6 +20,7 @@ aspathtest_SOURCES = aspath_test.c
 testbgpcap_SOURCES = bgp_capability_test.c
 ecommtest_SOURCES = ecommunity_test.c
 testbgpmpattr_SOURCES =  bgp_mp_attr_test.c
+testchecksum_SOURCES = test-checksum.c
 
 testsig_LDADD = ../lib/libzebra.la @LIBCAP@
 testbuffer_LDADD = ../lib/libzebra.la @LIBCAP@
@@ -33,3 +34,4 @@ aspathtest_LDADD = ../lib/libzebra.la @LIBCAP@ -lm ../bgpd/libbgp.a
 testbgpcap_LDADD = ../lib/libzebra.la @LIBCAP@ -lm ../bgpd/libbgp.a
 ecommtest_LDADD = ../lib/libzebra.la @LIBCAP@ -lm ../bgpd/libbgp.a
 testbgpmpattr_LDADD = ../lib/libzebra.la @LIBCAP@ -lm ../bgpd/libbgp.a
+testchecksum_LDADD = ../lib/libzebra.la @LIBCAP@ 
diff --git a/tests/test-checksum.c b/tests/test-checksum.c
new file mode 100644
index 0000000..d218840
--- /dev/null
+++ b/tests/test-checksum.c
@@ -0,0 +1,499 @@
+#include <zebra.h>
+#include <stdlib.h>
+#include <time.h>
+
+#include "checksum.h"
+
+struct thread_master *master;
+
+struct acc_vals {
+  int c0;
+  int c1;
+};
+
+struct csum_vals {
+  struct acc_vals a;
+  int x; 
+  int y;
+};
+
+static struct csum_vals ospfd_vals, isisd_vals;
+
+typedef size_t testsz_t;
+typedef uint16_t testoff_t;
+
+/* Fletcher Checksum -- Refer to RFC1008. */
+#define MODX                 4102
+
+/* Accumulator phase of checksum */
+static 
+struct acc_vals
+accumulate (u_char *buffer, testsz_t len, testoff_t off)
+{
+  u_int8_t *p;
+  u_int16_t *csum;
+  int i, init_len, partial_len;
+  struct acc_vals ret;
+  
+  csum = (u_int16_t *) (buffer + off);
+  *(csum) = 0;
+  
+  p = buffer;
+  ret.c0 = 0;
+  ret.c1 = 0;
+  init_len = len;
+  
+  while (len != 0)
+    {
+      partial_len = MIN(len, MODX);
+
+      for (i = 0; i < partial_len; i++)
+	{
+	  ret.c0 = ret.c0 + *(p++);
+	  ret.c1 += ret.c0;
+	}
+
+      ret.c0 = ret.c0 % 255;
+      ret.c1 = ret.c1 % 255;
+
+      len -= partial_len;
+    }
+  return ret;
+}
+
+/* The final reduction phase.
+ * This one should be the original ospfd version 
+ */
+static u_int16_t 
+reduce_ospfd (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+
+  x = ((len - off - 1) * c0 - c1) % 255;
+  
+  if (x <= 0)
+    x += 255;
+  y = 510 - c0 - x;
+  if (y > 255)
+    y -= 255;
+
+   /* take care endian issue. */
+   return htons ((x << 8) + y); 
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+/* slightly different concatenation */
+static u_int16_t 
+reduce_ospfd1 (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+
+  x = ((len - off - 1) * c0 - c1) % 255;
+  if (x <= 0)
+    x += 255;
+  y = 510 - c0 - x;
+  if (y > 255)
+    y -= 255;
+
+   /* take care endian issue. */
+   return htons ((x << 8) | (y & 0xff)); 
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+/* original isisd version */
+static u_int16_t 
+reduce_isisd (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+  u_int32_t mul;
+  
+  mul = (len - off)*(c0);
+  x = mul - c0 - c1;
+  y = c1 - mul - 1;
+
+  if (y > 0)
+    y++;
+  if (x < 0)
+    x--;
+
+  x %= 255;
+  y %= 255;
+
+  if (x == 0)
+    x = 255;
+  if (y == 0)
+    y = 1;
+
+  return htons ((x << 8) | (y & 0xff));
+
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+/* Is the -1 in y wrong perhaps? */
+static u_int16_t 
+reduce_isisd_yfix (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+  u_int32_t mul;
+  
+  mul = (len - off)*(c0);
+  x = mul - c0 - c1;
+  y = c1 - mul;
+
+  if (y > 0)
+    y++;
+  if (x < 0)
+    x--;
+
+  x %= 255;
+  y %= 255;
+
+  if (x == 0)
+    x = 255;
+  if (y == 0)
+    y = 1;
+
+  return htons ((x << 8) | (y & 0xff));
+
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+/* Move the mods yp */
+static u_int16_t 
+reduce_isisd_mod (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+  u_int32_t mul;
+  
+  mul = (len - off)*(c0);
+  x = mul - c1 - c0;
+  y = c1 - mul - 1;
+
+  x %= 255;
+  y %= 255;
+
+  if (y > 0)
+    y++;
+  if (x < 0)
+    x--;
+
+  if (x == 0)
+    x = 255;
+  if (y == 0)
+    y = 1;
+
+  return htons ((x << 8) | (y & 0xff));
+
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+/* Move the mods up + fix y */
+static u_int16_t 
+reduce_isisd_mody (struct csum_vals *vals, testsz_t len, testoff_t off)
+{
+#define x vals->x
+#define y vals->y
+#define c0 vals->a.c0
+#define c1 vals->a.c1
+  u_int32_t mul;
+  
+  mul = (len - off)*(c0);
+  x = mul - c0 - c1;
+  y = c1 - mul;
+
+  x %= 255;
+  y %= 255;
+
+  if (y > 0)
+    y++;
+  if (x < 0)
+    x--;
+
+  if (x == 0)
+    x = 255;
+  if (y == 0)
+    y = 1;
+
+  return htons ((x << 8) | (y & 0xff));
+
+#undef x
+#undef y
+#undef c0
+#undef c1
+}
+
+struct reductions_t {
+  const char *name;
+  u_int16_t (*f) (struct csum_vals *, testsz_t, testoff_t);
+} reducts[] = {
+  { .name = "ospfd",		.f = reduce_ospfd },
+  { .name = "ospfd-1",		.f = reduce_ospfd1 },
+  { .name = "isisd", 		.f = reduce_isisd },
+  { .name = "isisd-yfix", 	.f = reduce_isisd_yfix },
+  { .name = "isisd-mod", 	.f = reduce_isisd_mod },
+  { .name = "isisd-mody", 	.f = reduce_isisd_mody },
+  { NULL, NULL },
+};
+
+/* The original ospfd checksum */
+static u_int16_t
+ospfd_checksum (u_char *buffer, testsz_t len, testoff_t off)
+{
+  u_char *sp, *ep, *p, *q;
+  int c0 = 0, c1 = 0;
+  int x, y;
+  u_int16_t checksum, *csum;
+
+  csum = (u_int16_t *) (buffer + off);
+  *(csum) = 0;
+  
+  sp = buffer;
+
+  for (ep = sp + len; sp < ep; sp = q)
+    {
+      q = sp + MODX;
+      if (q > ep)
+        q = ep;
+      for (p = sp; p < q; p++)
+        {
+          c0 += *p;
+          c1 += c0;
+        }
+      c0 %= 255;
+      c1 %= 255;
+    }
+  
+  ospfd_vals.a.c0 = c0;
+  ospfd_vals.a.c1 = c1;
+  
+  //printf ("%s: len %u, off %u, c0 %d, c1 %d\n",
+  //        __func__, len, off, c0, c1);
+
+  x = ((int)(len - off - 1) * (int)c0 - (int)c1) % 255;
+  
+  if (x <= 0)
+    x += 255;
+  y = 510 - c0 - x;
+  if (y > 255)
+    y -= 255;
+  
+  ospfd_vals.x = x;
+  ospfd_vals.y = y;
+  
+  buffer[off] = x;
+  buffer[off + 1] = y;
+  
+  /* take care endian issue. */
+  checksum = htons ((x << 8) | (y & 0xff));
+
+  return (checksum);
+}
+
+/* the original, broken isisd checksum */
+static u_int16_t
+iso_csum_create (u_char * buffer, testsz_t len, testoff_t off)
+{
+
+  u_int8_t *p;
+  int x;
+  int y;
+  u_int32_t mul;
+  u_int32_t c0;
+  u_int32_t c1;
+  u_int16_t checksum, *csum;
+  int i, init_len, partial_len;
+
+  checksum = 0;
+  
+  csum = (u_int16_t *) (buffer + off);
+  *(csum) = checksum;
+  
+  p = buffer;
+  c0 = 0;
+  c1 = 0;
+  init_len = len;
+  
+  while (len != 0)
+    {
+      partial_len = MIN(len, MODX);
+
+      for (i = 0; i < partial_len; i++)
+	{
+	  c0 = c0 + *(p++);
+	  c1 += c0;
+	}
+
+      c0 = c0 % 255;
+      c1 = c1 % 255;
+
+      len -= partial_len;
+    }
+
+  isisd_vals.a.c0 = c0;
+  isisd_vals.a.c1 = c1;
+  
+  mul = (init_len - off) * c0;
+
+  x = mul - c1 - c0;
+  y = c1 - mul - 1;
+
+  if (y > 0)
+    y++;
+  if (x < 0)
+    x--;
+
+  x %= 255;
+  y %= 255;
+
+  if (x == 0)
+    x = 255;
+  if (y == 0)
+    y = 1;
+  
+  isisd_vals.x = x;
+  isisd_vals.y = y;
+  
+  checksum = htons((x << 8) | (y & 0xFF));
+  
+  *(csum) = checksum;
+  
+  /* return the checksum for user usage */
+  return checksum;
+}
+
+static int
+verify (u_char * buffer, testsz_t len)
+{
+  u_int8_t *p;
+  u_int32_t c0;
+  u_int32_t c1;
+  u_int16_t checksum;
+  int i, partial_len;
+ 
+  p = buffer;
+  checksum = 0;
+
+  c0 = 0;
+  c1 = 0;
+
+  while (len)
+    {
+      partial_len = MIN(len, 5803);
+
+      for (i = 0; i < partial_len; i++)
+        {
+          c0 = c0 + *(p++);
+          c1 += c0;
+        }
+      c0 = c0 % 255;
+      c1 = c1 % 255;
+
+      len -= partial_len;
+    }
+
+  if (c0 == 0 && c1 == 0)
+    return 0;
+
+  return 1;
+}
+
+int
+main(int argc, char **argv)
+{
+/* 60017 65629 702179 */
+#define MAXDATALEN 60017
+#define BUFSIZE MAXDATALEN + sizeof(u_int16_t)
+  u_char buffer[BUFSIZE];
+  int exercise = 0;
+#define EXERCISESTEP 257
+  
+  srandom (time (NULL));
+  
+  while (1) {
+    u_int16_t ospfd, isisd, lib;
+    
+    exercise += EXERCISESTEP;
+    exercise %= MAXDATALEN;
+    
+    for (int i = 0; i < exercise; i += sizeof (long int)) {
+      long int rand = random ();
+      
+      for (int j = sizeof (long int); j > 0; j--)
+        buffer[i + (sizeof (long int) - j)] = (rand >> (j * 8)) & 0xff;
+    }
+    
+    ospfd = ospfd_checksum (buffer, exercise + sizeof(u_int16_t), exercise);
+    if (verify (buffer, exercise + sizeof(u_int16_t)))
+      printf ("verify: ospfd failed\n");
+    isisd = iso_csum_create (buffer, exercise + sizeof(u_int16_t), exercise);
+    if (verify (buffer, exercise + sizeof(u_int16_t)))
+      printf ("verify: isisd failed\n");
+    lib = fletcher_checksum (buffer, exercise + sizeof(u_int16_t), exercise);
+    if (verify (buffer, exercise + sizeof(u_int16_t)))
+      printf ("verify: lib failed\n");
+    
+    if (ospfd != lib) {
+      printf ("Mismatch in values at size %u\n"
+              "ospfd: 0x%04x\tc0: %d\tc1: %d\tx: %d\ty: %d\n"
+              "isisd: 0x%04x\tc0: %d\tc1: %d\tx: %d\ty: %d\n"
+              "lib: 0x%04x\n",
+              exercise,
+              ospfd, ospfd_vals.a.c0, ospfd_vals.a.c1, ospfd_vals.x, ospfd_vals.y,
+              isisd, isisd_vals.a.c0, isisd_vals.a.c1, isisd_vals.x, isisd_vals.y,
+              lib
+              );
+      
+      /* Investigate reduction phase discrepencies */
+      if (ospfd_vals.a.c0 == isisd_vals.a.c0
+          && ospfd_vals.a.c1 == isisd_vals.a.c1) {
+        printf ("\n");
+        for (int i = 0; reducts[i].name != NULL; i++) {	
+          ospfd = reducts[i].f (&ospfd_vals,
+                                exercise + sizeof (u_int16_t),
+                                exercise);
+          printf ("%20s: x: %02x, y %02x, checksum 0x%04x\n",
+                  reducts[i].name, ospfd_vals.x & 0xff, ospfd_vals.y & 0xff, ospfd);
+        }
+      }
+              
+      printf ("\n  u_char testdata [] = {\n  ");
+      for (int i = 0; i < exercise; i++) {
+        printf ("0x%02x,%s",
+                buffer[i],
+                (i + 1) % 8 ? " " : "\n  ");
+      }
+      printf ("\n}\n");
+      exit (1);
+    }
+  }
+}