https://bz.apache.org/bugzilla/show_bug.cgi?id=70167

Joe Orton <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|[email protected]       |[email protected]
          Component|All                         |APR
            Version|2.5-HEAD                    |1.7.6
            Product|Apache httpd-2              |APR

--- Comment #10 from Joe Orton <[email protected]> ---
I threw Claude at it. It's a fun side effect of the mod_ssl coalescing filter:

1. since whenever mod_ssl added the coalescing filter, the first part of the
FILE bucket will be split and merged into the same TLS record with the HTTP
headers (that's the point of that filter)

2. that leaves an odd-sized FILE bucket with an offset (e->start is non-zero),
representing the rest of the response body

3. every read from that FILE bucket would try to morph into an MMAP bucket,
which allocates 56 bytes from r->pool, and then FAILS because the offset is not
page-aligned

Yann had a go at fixing (3) indirectly by fixing the APR mmap code to handle
non-page-aligned offsets cleanly, but this was reverted from 1.7.x, see history
at https://github.com/apache/apr/commits/1.7.x/mmap/unix/mmap.c

We can instead attack (3) from two directions: not allocating 56 bytes in the
failure case, and by fudging the read offsets to page-align so that a FILE
bucket *can* successfully morph into MMAP for this case

-> https://github.com/apache/apr/pull/76

-- 
You are receiving this mail because:
You are the assignee for the bug.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to