Re: MAXSTRINGLEN applied to printf()?

[Josh Stone <jistone at redhat dot com>]

On 08/09/2012 12:51 PM, halcyonic@gmail.com wrote:
> Sorry...didn't want to spam a lot of code and output.

Thanks - what you have here is not nearly large enough that I'd call it
spam, and it's much better than us trying to guess what's going on.

> So, for example, I get a long output that contains this fragment:
> 
> ... ,"localhost.localdomain"localhost.localdomain_2012-8-8-23-47-30.bz2", ...
> 
> ...generated by the following probe (I've set string lengths to be
> 2048).  The thing to notice is that any string concatenation I do
> should always involve the addition of a pair of quotes, so I should
> never be able to get an output that involves an odd number of quotes
> (as per above).

But if your concatenation goes beyond MAXSTRINGLEN, then one of the pair
of quotes may be silently truncated.

> So either I'm somehow getting intermingled output (I
> don't think I am...the rest of the output looks perfectly fine), or
> somehow a string is getting silently truncated somewhere (fyi, all
> the filenames it's listing are of the form
> localhost.localdomain_2012-8-8-23-47-30.bz2).
> 
> Any theories appreciated.
> 
> Thanks, Nick
> 
> -----
> probe syscall.getdents.return {
>     if ((execname() != "stap") && !is_fd_blacklisted(pid(), $fd)) {
>     printf("{ \"arglist\":")
>     arglist = "[ "
>     if ($return > 0) {
>         total_entries = 0
>         dirent = $dirent
>         total_bytes = 0
>         current_bytes = 0
>         while (total_bytes < $return) {
>           if (dirent == 0) {
>               break
>           }
>           nextarg = clean_string(user_string_warn(@cast(dirent, "struct linux_dirent")->d_name))

What is clean_string(), something to sanitize to printable characters?
You might like user_string_quoted(), or "text_strn(str, 0, 1)" to quote
it after the fact.

>           len = @cast(dirent, "struct linux_dirent")->d_reclen
>           formatlen = strlen(nextarg) + 2
>           dirent += len
>           total_bytes += len
>           total_entries += 1
> 
>           if (total_entries < 256) {
>               if (current_bytes + formatlen > 2048) {
>                 printf("%s", arglist)
>                 arglist = ""
>                 current_bytes = 0
>               }
>               
>               arglist .= "\"".nextarg."\""
>               current_bytes += formatlen 
>               if (total_bytes < $return) {
>                 arglist .= ","
>               }
>           }          

Ok, I see you're trying to avoid MAXSTRINGLEN here.  I think your bug
may be simply when you add "," to arglist without also incrementing
current_bytes, so arglist is longer than you think when you check if a
printf is due.

Also, don't forget that a \0 terminator has to be present within
MAXSTRINGLEN too, so you really only have 2047 bytes to play with.

>         }
>     }
>     printf("%s],", arglist)
> #    arglist = substr(arglist, 0, strlen(arglist)-1)."]"
> #    outstr = "{ "
> #    outstr .= "\"arglist\": "
> #    outstr .= arglist.","
>     outstr .= "\"count\": "
>     outstr .= sprintf("%u", $count).","
>     outstr .= "\"execname\": \""
>     outstr .= clean_string(execname())."\","
>     outstr .= "\"fd\": "
>     outstr .= sprintf("%d", $fd).","
>     outstr .= "\"op\": \""
>     outstr .= clean_string("GETDENTS")."\","
>     outstr .= "\"pid\": "
>     outstr .= sprintf("%d", pid()).","
>     outstr .= "\"ppid\": "
>     outstr .= sprintf("%d", ppid()).","
>     outstr .= "\"return\": "
>     outstr .= sprintf("%d", $return).","
>     outstr .= "\"timestamp\": "
>     outstr .= sprintf("%d", gettimeofday_ms()).","
>     outstr .= "\"total_bytes\": "
>     outstr .= sprintf("%u", total_bytes).","
>     outstr .= "\"total_entries\": "
>     outstr .= sprintf("%u", total_entries).","
>     outstr .= "\"uid\": "
>     outstr .= sprintf("%d", uid())."}\n"
>     printf("%s", outstr)
>   }
> }
> -----

With your MAXTRINGLEN=2048, you're probably not hitting limits here at
the end, but every single one of these ".", ".=", and "sprintf" create
string temporaries.  The code we generate for this will be pretty
inefficient, with lots of string copies.  So I'd recommend building as
much as you can directly into that final output format, e.g.

  printf("\"count\":%u,\"execname\":%s, ...\n",
         $count, clean_string(execname()), ...)


Josh