Everything

php

basename() memory access violation

php.internals

Alexander Valyalkin

22 years ago

Here is patch for basename() function, which prevents possible
memory access violation:

=========cut=========
--- string.c    Thu May 13 20:44:32 2004
+++ string_basename.c   Mon Jun 14 20:43:33 2004
@@ -1079,9 +1079,9 @@


/* strip trailing slashes */
-   while (*c == '/'
+    while (c >= s && *c == '/'
#ifdef PHP_WIN32
-          || (*c == '\\' && !IsDBCSLeadByte(*(c-1)))
+           || (c > s && *c == '\\' && !IsDBCSLeadByte(*(c-1)))
#endif
)
c--;
@@ -1092,10 +1092,10 @@
}

#ifdef PHP_WIN32
-   if ((c = strrchr(s, '/')) || ((c = strrchr(s, '\\')) &&
!IsDBCSLeadByte(*(c-1)))) {
+    if ((c = strrchr(s, '/')) || ((c = strrchr(s, '\\')) && c > s &&
!IsDBCSLeadByte(*(c-1)))) {
if (*c == '/') {
char *c2 = strrchr(s, '\\');
-           if (c2 && !IsDBCSLeadByte(*(c2-1)) && c2 > c) {
+            if (c2 && c2 > s && !IsDBCSLeadByte(*(c2-1)) && c2 > c) {
c = c2;
}
}
=========cut==========

--
Using Opera's revolutionary e-mail client: http://www.opera.com/m2/

Ilia A.

22 years ago

Do you have an example script that can be used to reproduce the supposed
memory access violation?

Ilia

On June 14, 2004 01:51 pm, Alexander Valyalkin wrote:

Christian Schneider

22 years ago

Ilia Alshanetsky wrote:

> Do you have an example script that can be used to reproduce the supposed
> memory access violation?

The function php_basename actually DOES access the char at s-1, i.e. one
byte before the string it is passed. If you hand it a buffer gotten from
malloc you get a valgrind hit. If I use emalloc this doesn't happen, I
guess emalloc has some header stuff before the allocation. If for
whatever reason s[-1] contains '/' then things would get even worse.

Conclusion: The internal function php_basename IS accessing one byte
before the buffer passed to it but the PHP environment seems to hide
this behaviour.
Is this a bug which should proactively been fixed? Or do we rely on the
environment to guarantee that we can access s[-1] without problems? I
don't know and it's up to you PHP gods to decide ;-)

- Chris

Stefan Esser

22 years ago

Hi,

> The function php_basename actually DOES access the char at s-1, i.e. one
> byte before the string it is passed. If you hand it a buffer gotten from
> malloc you get a valgrind hit. If I use emalloc this doesn't happen, I
> guess emalloc has some header stuff before the allocation. If for
> whatever reason s[-1] contains '/' then things would get even worse.

First of all: yes it should be fixed, but it is nothing dangerous.
emalloc stores the size of the buffer before the pointer you actually
get. On little endian systems this means a '/' can only be directly
before the buffer if this buffer is > 752MB big. For big endian systems
this this is of course easier ;)

Anyway this size is used only by the memory cache within the ZendEngine
and by the memory_limit check. So by overwriting with a zero you could
make the ZendEngine believe more memory is allocated that really is. <-
No problem. The other thing could be that the block is inserted into the
wrong cache position. Because we can only write 0 bytes this is no
problem, because we would always insert the free memory block in a block
that is assumed to be smaller (so no overflow possible).

Stefan

Andi Gutmans

22 years ago

I don't think it's a good idea to access s[-1] even if the logic works.
This should be fixed IMO.

At 10:30 PM 6/14/2004 +0200, Christian Schneider wrote: