Re: bsqldb SEGFAULT abend

Hi James,

Thanks for your help. I made the change to allocate width+1 bytes for the
destination buffer, and asked SQL Server to CAST(CMD_LINE AS VARCHAR). Tests
run fine without any errors both under MALLOC_CHECK_ and valgrind.

For the time being those changes will suffice. Hope you guys have the
opportunity to implement SYBTEXT proper handling in the near future.

Best Regards,

Ronildo

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of James K. Lowden
Sent: quinta-feira, 1 de abril de 2010 21:26
To: FreeTDS Development Group
Subject: Re: [freetds] bsqldb SEGFAULT abend

Hi Ronildo, 

You found two bugs in bsqldb.  

Ronildo wrote:
> Please let me share some thoughts from the tests I've been running:
> 
> 1 - If got it right from James' first answer, we expect
> (dblib.c)+(buffering.h) to buffer_transfer_bound_data bytes limited to:
> 
>  data[c].buffer = calloc(1, metadata[c].width);

Change that line to 

   data[c].buffer = calloc(1, 1 + metadata[c].width);  

http://freetds.cvs.sourceforge.net/viewvc/freetds/freetds/src/dblib/dblib.c?
revision=1.320&view=markup&pathrev=BRANCH0_82

Your log shows:

> buffering.h:231:buffer_transfer_bound_data(0x92ad368 4040 -1 0x92ad360
> 0) dblib.c:7083:copy_data_to_host_var(35 [SYBTEXT] len 30 => 47 
> [SYBCHAR] len -1)

We're copying 30 bytes and asking for a null-terminated string.  We're
promising the destination buffer is at least 36 bytes.  

Line 7143, per valgrind, memcpy(3) copies srclen bytes to dest.  That's
actually OK, because bsqldb calloc'ed 30 bytes.  But then this happens:  

	memcpy(dest, src, srclen);
	for (i = srclen; i < destlen - 1; i++)
		dest[i] = ' ';
	dest[i] = '\0';

If i == srclen, dest[i] refers to 1 byte past the end of dest.  With the
above change, we allocate width+1 bytes, so dest will be 36 bytes.  

Bug #2 is even dumber, but I don't have a proper fix yet: 
get_printable_size() doesn't handle all data types.  In particular it
doesn't handle SYBTEXT.  That's because the "size" of SYBTEXT is 2^31 or so,
which should be handled specially.  Instead it returns 0, causing
print_results() to set the width to the length of the *name* of the column.
After which ensues a whole peck of trouble!  

If you changed your SQL to convert the column to varchar, your memory error
will go away.  I'll look to add proper SYBTEXT handling.  No reason it can't
be made to work.  

Regards, 

--jkl




_______________________________________________
FreeTDS mailing list
[email protected]
http://lists.ibiblio.org/mailman/listinfo/freetds