You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by "Gregory (Grisha) Trubetskoy" <gr...@ispol.com> on 2003/08/19 17:04:38 UTC

[PATCH] mod_rewrite RewriteLock permissions fix

When the RewriteLock creates the global mutex on a system where file-based
locking is the default mechanism, the file is owned by root, and after
setuid/fork children will not have permissions to it. This should fix it.

Grisha

--- modules/mappers/mod_rewrite.c       19 Aug 2003 00:52:43 -0000      1.228
+++ modules/mappers/mod_rewrite.c       19 Aug 2003 14:55:32 -0000
@@ -2372,6 +2372,11 @@
                      "file %s", lockname);
         return rc;
     }
+    else {
+#if !defined(OS2) && !defined(WIN32) && !defined(BEOS) && !defined(NETWARE)
+       chown(lockname, unixd_config.user_id, -1);
+#endif
+    }

 #ifdef MOD_REWRITE_SET_MUTEX_PERMS
     rc = unixd_set_global_mutex_perms(rewrite_mapr_lock_acquire);


Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by Jeff Trawick <tr...@attglobal.net>.
Gregory (Grisha) Trubetskoy wrote:

> Hmm... I think it went into MAIN as opposed to APACHE_2_0_BRANCH. It's not
> in 2.0.47, which still has the RewriteLock problem.

Right, but it is listed in the STATUS file on the APACHE_2_0_BRANCH 
branch as something proposed for merging into APACHE_2_0_BRANCH.

JimJag had some comments there today that need to be properly addressed 
though.



Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by "Gregory (Grisha) Trubetskoy" <gr...@apache.org>.
Hmm... I think it went into MAIN as opposed to APACHE_2_0_BRANCH. It's not
in 2.0.47, which still has the RewriteLock problem.

Grisha

On Wed, 20 Aug 2003, Jeff Trawick wrote:

> I humbly submit that the patch already listed in
> APACHE_2_0_BRANCH/STATUS is what should be available in the next 2.0.x
> release:
>
>      * Unix: Handle permissions settings for flock-based mutexes in
>        unixd_set_global|proc_mutex_perms().  Allow the functions to
>        be called for any type of mutex.  PR 20312
>          modules/mappers/mod_rewrite.c 1.153
>          modules/ssl/mod_ssl.h 1.136
>          modules/ssl/ssl_engine_config.c 1.81
>          modules/ssl/ssl_engine_mutex.c 1.26
>          os/unix/unixd.c 1.58
>          os/unix/unixd.h 1.38
>          +1: trawick, jerenkrantz
>

Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by "Gregory (Grisha) Trubetskoy" <gr...@apache.org>.
On Wed, 20 Aug 2003, Jeff Trawick wrote:

> The issues with your patch are:
>
> . your patch does the chown() for more than just flock()-ed files
>
> . there is a better place to put the non-portable flock() code, both to
> keep it out of mod_rewrite and to help solve the problem for any other
> modules
>

I agree - my patch was just a copycat from something similar done in
mod_ssl - I'm guessing the code i was copying from was probably removed by
the patch you mentioned.

Grisha

Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by Jeff Trawick <tr...@attglobal.net>.
Gregory (Grisha) Trubetskoy wrote:

> Most likely it *is* fixed in 2.1 - I saw this problem on 2.0, but I looked
> at the code of 2.1 and saw that the code looks the same, so assumed (I
> know, assuming is bad...) that it is still broken.

hey, that's a pretty mean trick :)

> But this can be applied to 2.0, where I think it isn't fixed yet.

I humbly submit that the patch already listed in 
APACHE_2_0_BRANCH/STATUS is what should be available in the next 2.0.x 
release:

     * Unix: Handle permissions settings for flock-based mutexes in
       unixd_set_global|proc_mutex_perms().  Allow the functions to
       be called for any type of mutex.  PR 20312
         modules/mappers/mod_rewrite.c 1.153
         modules/ssl/mod_ssl.h 1.136
         modules/ssl/ssl_engine_config.c 1.81
         modules/ssl/ssl_engine_mutex.c 1.26
         os/unix/unixd.c 1.58
         os/unix/unixd.h 1.38
         +1: trawick, jerenkrantz

The parts that fix mod_rewrite specifically are the changes to 
mod_rewrite.c, unixd.c, and unixd.h.  the ssl changes are just a cleanup 
made possible by the infrastructure knowing when a chown() is necessary.

The issues with your patch are:

. your patch does the chown() for more than just flock()-ed files

. there is a better place to put the non-portable flock() code, both to 
keep it out of mod_rewrite and to help solve the problem for any other 
modules



Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by "Gregory (Grisha) Trubetskoy" <gr...@apache.org>.
Most likely it *is* fixed in 2.1 - I saw this problem on 2.0, but I looked
at the code of 2.1 and saw that the code looks the same, so assumed (I
know, assuming is bad...) that it is still broken. I haven't actually
tried verifying that it is a problem with 2.1 (it didn't occur to me to
look at unixd_set_global_mutex_perms() as in 2.0 IIRC it was something
SysV-specific).

But this can be applied to 2.0, where I think it isn't fixed yet.

P.S. My OS is FreeBSD 4.8 the preferred locking is flock() indeed.

Grisha

On Tue, 19 Aug 2003, Jeff Trawick wrote:

> Gregory (Grisha) Trubetskoy wrote:
>
> > When the RewriteLock creates the global mutex on a system where file-based
> > locking is the default mechanism, the file is owned by root, and after
> > setuid/fork children will not have permissions to it. This should fix it.
>
> hmmm...  some joker allegedly fixed this in 2.1-dev recently, yet you're
> apparently using 2.1-dev already based on the looks of your patch :(
>
> >
> > Grisha
> >
> > --- modules/mappers/mod_rewrite.c       19 Aug 2003 00:52:43 -0000      1.228
> > +++ modules/mappers/mod_rewrite.c       19 Aug 2003 14:55:32 -0000
> > @@ -2372,6 +2372,11 @@
> >                       "file %s", lockname);
> >          return rc;
> >      }
> > +    else {
> > +#if !defined(OS2) && !defined(WIN32) && !defined(BEOS) && !defined(NETWARE)
> > +       chown(lockname, unixd_config.user_id, -1);
> > +#endif
>
> the unixd_set_global_mutex_perms() is supposed to do the chown() now
> when necessary...   what is the mutex type for this mutex on your
> system, and can you check why unixd_set_proc_mutex_perms() isn't doing
> the chown() for you?
>
> (btw, the chown() is only needed when flock mutexes are used, as that is
> the implementation where the lock file has to be re-opened in the child
> process)
>
> > +    }
> >
> >  #ifdef MOD_REWRITE_SET_MUTEX_PERMS
> >      rc = unixd_set_global_mutex_perms(rewrite_mapr_lock_acquire);
>
>

Re: [PATCH] mod_rewrite RewriteLock permissions fix

Posted by Jeff Trawick <tr...@attglobal.net>.
Gregory (Grisha) Trubetskoy wrote:

> When the RewriteLock creates the global mutex on a system where file-based
> locking is the default mechanism, the file is owned by root, and after
> setuid/fork children will not have permissions to it. This should fix it.

hmmm...  some joker allegedly fixed this in 2.1-dev recently, yet you're 
apparently using 2.1-dev already based on the looks of your patch :(

> 
> Grisha
> 
> --- modules/mappers/mod_rewrite.c       19 Aug 2003 00:52:43 -0000      1.228
> +++ modules/mappers/mod_rewrite.c       19 Aug 2003 14:55:32 -0000
> @@ -2372,6 +2372,11 @@
>                       "file %s", lockname);
>          return rc;
>      }
> +    else {
> +#if !defined(OS2) && !defined(WIN32) && !defined(BEOS) && !defined(NETWARE)
> +       chown(lockname, unixd_config.user_id, -1);
> +#endif

the unixd_set_global_mutex_perms() is supposed to do the chown() now 
when necessary...   what is the mutex type for this mutex on your 
system, and can you check why unixd_set_proc_mutex_perms() isn't doing 
the chown() for you?

(btw, the chown() is only needed when flock mutexes are used, as that is 
the implementation where the lock file has to be re-opened in the child 
process)

> +    }
> 
>  #ifdef MOD_REWRITE_SET_MUTEX_PERMS
>      rc = unixd_set_global_mutex_perms(rewrite_mapr_lock_acquire);