PATCH: Check S-record with 0 size

PATCH: Check S-record with 0 size

[H.J. Lu]

Hi,

I checked in this patch to fix S-record with 0 size reported at

http://lists.gnu.org/archive/html/bug-binutils/2014-08/msg00110.html

H.J.
---
diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index bba079f..6484c91 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,3 +1,7 @@
+2014-08-28  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* srec.c (srec_scan): Return error for 0 size.
+
 2014-08-27  H.J. Lu  <hongjiu.lu@intel.com>
 
 	PR ld/17306
diff --git a/bfd/srec.c b/bfd/srec.c
index f11e74a..d979bf5 100644
--- a/bfd/srec.c
+++ b/bfd/srec.c
@@ -486,7 +486,8 @@ srec_scan (bfd *abfd)
 		bufsize = bytes * 2;
 	      }
 
-	    if (bfd_bread (buf, (bfd_size_type) bytes * 2, abfd) != bytes * 2)
+	    if (bytes == 0
+		|| bfd_bread (buf, (bfd_size_type) bytes * 2, abfd) != bytes * 2)
 	      goto error_return;
 
 	    /* Ignore the checksum byte.  */

Re: PATCH: Check S-record with 0 size

[Alan Modra]

On Thu, Aug 28, 2014 at 08:25:34AM -0700, H.J. Lu wrote:
> Hi,
> 
> I checked in this patch to fix S-record with 0 size reported at
> 
> http://lists.gnu.org/archive/html/bug-binutils/2014-08/msg00110.html

We should really be testing for other invalid byte counts.

    	* srec.c (srec_scan): Revert last change.  Report an error for
    	S-records with less than the miniumum byte count.

diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index 6484c91..3e005c9 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,3 +1,8 @@
+2014-08-29  Alan Modra  <amodra@gmail.com>
+
+	* srec.c (srec_scan): Revert last change.  Report an error for
+	S-records with less than the miniumum byte count.
+
 2014-08-28  H.J. Lu  <hongjiu.lu@intel.com>
 
 	* srec.c (srec_scan): Return error for 0 size.
diff --git a/bfd/srec.c b/bfd/srec.c
index d979bf5..42143c7 100644
--- a/bfd/srec.c
+++ b/bfd/srec.c
@@ -453,7 +453,7 @@ srec_scan (bfd *abfd)
 	  {
 	    file_ptr pos;
 	    char hdr[3];
-	    unsigned int bytes;
+	    unsigned int bytes, min_bytes;
 	    bfd_vma address;
 	    bfd_byte *data;
 	    unsigned char check_sum;
@@ -476,6 +476,19 @@ srec_scan (bfd *abfd)
 	      }
 
 	    check_sum = bytes = HEX (hdr + 1);
+	    min_bytes = 3;
+	    if (hdr[0] == '2' || hdr[0] == '8')
+	      min_bytes = 4;
+	    else if (hdr[0] == '3' || hdr[0] == '7')
+	      min_bytes = 5;
+	    if (bytes < min_bytes)
+	      {
+		(*_bfd_error_handler) (_("%B:%d: byte count %d too small\n"),
+				       abfd, lineno, bytes);
+		bfd_set_error (bfd_error_bad_value);
+		goto error_return;
+	      }
+
 	    if (bytes * 2 > bufsize)
 	      {
 		if (buf != NULL)
@@ -486,8 +499,7 @@ srec_scan (bfd *abfd)
 		bufsize = bytes * 2;
 	      }
 
-	    if (bytes == 0
-		|| bfd_bread (buf, (bfd_size_type) bytes * 2, abfd) != bytes * 2)
+	    if (bfd_bread (buf, (bfd_size_type) bytes * 2, abfd) != bytes * 2)
 	      goto error_return;
 
 	    /* Ignore the checksum byte.  */

-- 
Alan Modra
Australia Development Lab, IBM

Re: PATCH: Check S-record with 0 size

[H.J. Lu]

On Thu, Aug 28, 2014 at 6:15 PM, Alan Modra <amodra@gmail.com> wrote:
> On Thu, Aug 28, 2014 at 08:25:34AM -0700, H.J. Lu wrote:
>> Hi,
>>
>> I checked in this patch to fix S-record with 0 size reported at
>>
>> http://lists.gnu.org/archive/html/bug-binutils/2014-08/msg00110.html
>
> We should really be testing for other invalid byte counts.
>
>         * srec.c (srec_scan): Revert last change.  Report an error for
>         S-records with less than the miniumum byte count.
>

I don't think it is desirable.  For the given testcase, now we got

[hjl@gnu-6 binutils]$ ./strings /tmp/ohcrap.txt
BFD: /tmp/ohcrap.txt:1: byte count 0 too small

S700
[hjl@gnu-6 binutils]$

We should generate

[hjl@gnu-6 binutils]$ ./strings /tmp/ohcrap.txt
S700
[hjl@gnu-6 binutils]$

-- 
H.J.

Re: PATCH: Check S-record with 0 size

[Alan Modra]

On Fri, Aug 29, 2014 at 08:06:28AM -0700, H.J. Lu wrote:
> On Thu, Aug 28, 2014 at 6:15 PM, Alan Modra <amodra@gmail.com> wrote:
> > On Thu, Aug 28, 2014 at 08:25:34AM -0700, H.J. Lu wrote:
> >> Hi,
> >>
> >> I checked in this patch to fix S-record with 0 size reported at
> >>
> >> http://lists.gnu.org/archive/html/bug-binutils/2014-08/msg00110.html
> >
> > We should really be testing for other invalid byte counts.
> >
> >         * srec.c (srec_scan): Revert last change.  Report an error for
> >         S-records with less than the miniumum byte count.
> >
> 
> I don't think it is desirable.  For the given testcase, now we got
> 
> [hjl@gnu-6 binutils]$ ./strings /tmp/ohcrap.txt
> BFD: /tmp/ohcrap.txt:1: byte count 0 too small
> 
> S700

The thing is that srec_scan ought to provide this information in the
bulk of the file.  For example, if someone has been editing an
S-record file and gets something wrong, then it is good to know which
line.

On the other hand, srec_object_p shouldn't really be returning any
errors other than bfd_error_wrong_format, or fatal errors like
bfd_error_no_memory.  The difficulty is that srec_object_p does more
than just decide whether the bfd is an srec file;  It parses the
entire file.  In effect, any error results in "this isn't an srec
file".  So if we don't report S-record errors in srec_scan, we won't
have an opportunity to report them later.

I thought about fixing this by extracting "case 'S'" from srec_scan
into a separate function, untangling all the error reporting, then
using the new function in srec_object_p to verify the first line was a
valid S-record.  ie. in place of

  if (b[0] != 'S' || !ISHEX (b[1]) || !ISHEX (b[2]) || !ISHEX (b[3]))
    {
      bfd_set_error (bfd_error_wrong_format);
      return NULL;
    }

That seemed like too much work for little benefit.

-- 
Alan Modra
Australia Development Lab, IBM