Обсуждение: Patch for snprintf problem (bug #1000650)

Hello,

  I found the problem in snprintf on linux (maybe another unix) boxes
in info.c (CVS HEAD). The problematic part is something like:

buf = "text";
snprintf(buf,size,"%s append",buf);

buf = "text append" on Windows (MS VC compiler)
buf = " append" on linux (gcc compiler)

I solve it this way (main idea):
snprintf(buf + strlen(buf), " append");

There are more parameters (char *, int, ...) in real usage.
More details is here:
http://pgfoundry.org/tracker/index.php?func=detail&aid=1000650&group_id=1000125&atid=538

The patch is created againist CVS.

Comments are welcome

Luf

> I solve it this way (main idea):
> snprintf(buf + strlen(buf), " append");

Ugh typo mistake:

 snprintf(buf + strlen(buf), sizeof(buf), " append");

My hands are faster than my head :-(
There is buffer overrun erron becouse I have to decrease sizeof(buf) with
strlen(buf).

So I attach second try of this patch againist CVS HEAD.

Comments are welcome

Luf

Ludek Finstrle wrote:
> Hello,
>
>   I found the problem in snprintf on linux (maybe another unix) boxes
> in info.c (CVS HEAD). The problematic part is something like:
>
> buf = "text";
> snprintf(buf,size,"%s append",buf);
>
> buf = "text append" on Windows (MS VC compiler)
> buf = " append" on linux (gcc compiler)
>
> I solve it this way (main idea):
> snprintf(buf + strlen(buf), " append");

Hmm bad news.
If so, it may be better to use the sequence like the following for example.

    char    *query_ptr;
    size_t    bufsize_res;
    int    slen;

    /* Initialize */
    query_ptr = columns_query;
    bufsize_res = sizeof(columns_query);

    if (..)
    {
        if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
        {
            .. error_handling ..
        }
        query_ptr += slen;
        bufsize_res -= slen;
    }

    ...

regards,
Hiroshi Inoue

> > buf = "text";
> > snprintf(buf,size,"%s append",buf);
> >
> > buf = "text append" on Windows (MS VC compiler)
> > buf = " append" on linux (gcc compiler)
> >
> > I solve it this way (main idea):
> > snprintf(buf + strlen(buf), " append");
>
> Hmm bad news.
> If so, it may be better to use the sequence like the following for example.
>
>     char    *query_ptr;
>     size_t    bufsize_res;
>     int    slen;
>
>     /* Initialize */
>     query_ptr = columns_query;
>     bufsize_res = sizeof(columns_query);
>
>     if (..)
>     {
>         if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
>         {
>             .. error_handling ..
>         }
>         query_ptr += slen;
>         bufsize_res -= slen;
>     }
>
>     ...

I'm not sure. There are a lot of strcat, my_strcat, schema_strcat calls
between snprintf. There is only one place where snprintf could be called
after another snprintf.
BTW it's something what snprintf_addlen is doing. Or am I miss something?

I do the patch with minimal changes in mind. In my opinion strcat
may be faster in joining more char *.

Regards,

Luf

> > > buf = "text";
> > > snprintf(buf,size,"%s append",buf);
> > >
> > > buf = "text append" on Windows (MS VC compiler)
> > > buf = " append" on linux (gcc compiler)
> > >
> > > I solve it this way (main idea):
> > > snprintf(buf + strlen(buf), " append");
> >
> > Hmm bad news.
> > If so, it may be better to use the sequence like the following for example.
> >
> >     char    *query_ptr;
> >     size_t    bufsize_res;
> >     int    slen;
> >
> >     /* Initialize */
> >     query_ptr = columns_query;
> >     bufsize_res = sizeof(columns_query);
> >
> >     if (..)
> >     {
> >         if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
> >         {
> >             .. error_handling ..
> >         }
> >         query_ptr += slen;
> >         bufsize_res -= slen;
> >     }
> >
> >     ...
>
> I'm not sure. There are a lot of strcat, my_strcat, schema_strcat calls
> between snprintf. There is only one place where snprintf could be called
> after another snprintf.

I change the snprintf_addlen to snprintf_len as you mentioned.

Third try of patch attached.

Comments are welcome

Luf

> I change the snprintf_addlen to snprintf_len as you mentioned.
>
> Third try of patch attached.

I forgot attach the patch.

Luf

Fri, Jun 09, 2006 at 02:13:23PM +0200, Ludek Finstrle napsal(a):
> > > > buf = "text";
> > > > snprintf(buf,size,"%s append",buf);
> > > >
> > > > buf = "text append" on Windows (MS VC compiler)
> > > > buf = " append" on linux (gcc compiler)
> > > >
> > > > I solve it this way (main idea):
> > > > snprintf(buf + strlen(buf), " append");
> > >
> > > Hmm bad news.
> > > If so, it may be better to use the sequence like the following for example.
> > >
> > >     char    *query_ptr;
> > >     size_t    bufsize_res;
> > >     int    slen;
> > >
> > >     /* Initialize */
> > >     query_ptr = columns_query;
> > >     bufsize_res = sizeof(columns_query);
> > >
> > >     if (..)
> > >     {
> > >         if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
> > >         {
> > >             .. error_handling ..
> > >         }
> > >         query_ptr += slen;
> > >         bufsize_res -= slen;
> > >     }
> > >
> > >     ...
> >
> > I'm not sure. There are a lot of strcat, my_strcat, schema_strcat calls
> > between snprintf. There is only one place where snprintf could be called
> > after another snprintf.
>
> I change the snprintf_addlen to snprintf_len as you mentioned.

I take a look at all snprintf in psqlodbc an I find one more place
where it's problematic.
If we accept snprintf_add I change one call to it.

Additional patch attached, comments are welcome

Luf

> > > > buf = "text";
> > > > snprintf(buf,size,"%s append",buf);
> > > >
> > > > buf = "text append" on Windows (MS VC compiler)
> > > > buf = " append" on linux (gcc compiler)
> > > >
> > > > I solve it this way (main idea):
> > > > snprintf(buf + strlen(buf), " append");
> > >
> > > Hmm bad news.
> > > If so, it may be better to use the sequence like the following for example.
> > >
> > >     char    *query_ptr;
> > >     size_t    bufsize_res;
> > >     int    slen;
> > >
> > >     /* Initialize */
> > >     query_ptr = columns_query;
> > >     bufsize_res = sizeof(columns_query);
> > >
> > >     if (..)
> > >     {
> > >         if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
> > >         {
> > >             .. error_handling ..
> > >         }
> > >         query_ptr += slen;
> > >         bufsize_res -= slen;
> > >     }
> > >
> > >     ...
> >
> > I'm not sure. There are a lot of strcat, my_strcat, schema_strcat calls
> > between snprintf. There is only one place where snprintf could be called
> > after another snprintf.
>
> I change the snprintf_addlen to snprintf_len as you mentioned.
>
> Third try of patch attached.

I make patch againist CVS after yours huge commit. What's your opinion?
Could we fix linux bug? Or do you prefer ifdef or what's your
idea about change these lines:

snprintf(buf,bufsize,"%s adding 1",buf);
strcat(buf," adding 2");
snprintf(buf,bufsize,"%s adding 3",buf);

I think it's hard to maintain code you mentioned above. This calls
are only in catalog information (relatively short query - used not so
often).

Patch attached contains previous psqlodbc_snprintf_3try and
psqlodbc_snprintf_next.

Regards,

Luf

Ludek Finstrle wrote:
>
> I make patch againist CVS after yours huge commit. What's your opinion?
>

Is the second parameter of snprintf_add needed ?
Aren't the parameter values always strlen(the first parameter) ?

Is snprintf_len needed instead of snprintf ?
Though the current code ignores snprintf errors, it's simply
my negligence..

regards,
Hiroshi Inoue

Tue, Jun 13, 2006 at 11:12:58PM +0900, Hiroshi Inoue wrote:
> Ludek Finstrle wrote:
> >
> > I make patch againist CVS after yours huge commit. What's your opinion?
> >
>
> Is the second parameter of snprintf_add needed ?
> Aren't the parameter values always strlen(the first parameter) ?

You're right. I think more about it and "add" means add to the end
so I changed the patch as you pointed.

> Is snprintf_len needed instead of snprintf ?
> Though the current code ignores snprintf errors, it's simply
> my negligence..

I'm voting for keeping safer snprintf_len. But I can change it if
you wish.

New patch attached.

Regards,

Luf

Ludek Finstrle wrote:
> Tue, Jun 13, 2006 at 11:12:58PM +0900, Hiroshi Inoue wrote:
>
>> Ludek Finstrle wrote:
>>
>>> I make patch againist CVS after yours huge commit. What's your opinion?
>>>
>>>
>> Is the second parameter of snprintf_add needed ?
>> Aren't the parameter values always strlen(the first parameter) ?
>>
>
> You're right. I think more about it and "add" means add to the end
> so I changed the patch as you pointed.
>
>
>> Is snprintf_len needed instead of snprintf ?
>> Though the current code ignores snprintf errors, it's simply
>> my negligence..
>>
>
> I'm voting for keeping safer snprintf_len. But I can change it if
> you wish.
>
> New patch attached.
>

OK please commit it.

regards,
Hiroshi Inoue

> OK please commit it.

Patch commited.

Thanks,

Luf