Re: bsqldb SEGFAULT abend

On Mar 31,  8:19pm, [email protected] ("James K. Lowden") wrote:
-- Subject: Re: [freetds] bsqldb SEGFAULT abend

| Ronildo wrote:
| > 
| > Program received signal SIGSEGV, Segmentation fault.
| > [Switching to Thread -1208239392 (LWP 20221)]
| > 0x00b5143d in realloc () from /lib/libc.so.6
| > (gdb) bt
| > #0  0x00b5143d in realloc () from /lib/libc.so.6
| > #1  0x00132379 in tds_get_data (tds=0x8097ad0, curcol=0x80cad10) at
| > token.c:2043
| > #2  0x0013265a in tds_process_row (tds=0x8097ad0) at token.c:2168
| > #3  0x00134989 in tds_process_tokens (tds=0x8097ad0,
| > #result_type=0xbf940188,
| >     done_flags=0x0, flag=5384) at token.c:623
| > #4  0x00120a47 in dbnextrow (dbproc=0x8097360) at dblib.c:1994
| > #5  0x0804aae5 in main (argc=Cannot access memory at address 0x4
| > ) at bsqldb.c:514
| 
| Well, that's bad.  It looks to me like memory corruption, because all
| that's ever supposed to be stored in blob->textpointer is a pointer
| returned by malloc/realloc.  
| 
| One possibilty is that determine_adjusted_size() somehow returns a
| negative value, leading realloc() to ask for something huge.  (There are
| quite a few places in the code where int is used where size_t would be
| preferable.  This might be one.)  
| 
| One thing you could do is add 
| 
| 	assert(new_blob_size > 0);
| 
| as soon as it's assigned.  Also see MALLOC_CHECK_ in your realloc() man
| page.  
| 
| I consider any segment fault a serious error, and I very much wnat to get
| to the bottom of it.  It will be hard to reproduce, though, even with the
| details you've provided.  The more you're able to do on your end, the
| better.  
| 
| (Then again, it seems Frediano commits a fix everytime I make a speech
| like that....)
| 

Run it through valgrind...

christos