I have a very minimal config file with no mod_autoindex that reacts
strangely to an HTTP 0.9 GET for /.
On AIX, I get back the raw contents of the directory file (gibberish).
It turns out that default_handler opened the htdocs directory via
apr_file_open()/open() which caused the raw directory to be returned.
On Linux, I get a 500 error and I/O errors in the log (I guess Linux
doesn't like normal I/O on directories :) ).
r->finfo.filetype is APR_DIR. r->handler is httpd/unix-directory.
Why don't we decline if filetype is APR_DIR? (Conceivably it could be
useful to read from some of the other filetypes?)
Here is a silly patch. For now I'm filtering out APR_DIR files.
WW1D? ("1" == "1.3")
Index: server/core.c
===================================================================
RCS file: /home/cvs/httpd-2.0/server/core.c,v
retrieving revision 1.138
diff -u -r1.138 core.c
--- server/core.c 27 Jan 2002 12:52:07 -0000 1.138
+++ server/core.c 29 Jan 2002 00:29:43 -0000
@@ -2900,8 +2900,9 @@
* have a / in the middle, which probably captures the original
* intent, but may cause problems at first - Ben 7th Jan 01
*/
- if (strcmp(r->handler, "default-handler")
- && !ap_strchr_c(r->handler, '/'))
+ if ((strcmp(r->handler, "default-handler")
+ && !ap_strchr_c(r->handler, '/'))
+ || r->finfo.filetype == APR_DIR)
return DECLINED;
d = (core_dir_config *)ap_get_module_config(r->per_dir_config,
Now the client should portably get a 500 error. Better yet, error_log
has
[warn] [client 127.0.0.1] handler "httpd/unix-directory" not found for:
/tmp/xxx/htdocs/
On Linux, it did have
[error] (21)Is a directory: core_output_filter: Error reading from bucket.
[info] (22)Invalid argument: core_output_filter: writing data to the network
--
Jeff Trawick | [EMAIL PROTECTED] | PGP public key at web site:
http://www.geocities.com/SiliconValley/Park/9289/
Born in Roswell... married an alien...
