Re: [moto-users] Apache crashes on page load

Well, the good news is I can replicate the crash :) The bad news is 
we've got one bug masking another. The worse news is if I just payed 
attention to the forums I would have known about this already.

The masking bug is the interesting one. Whatever is going wrong in the 
PosixFileIO extension is just something annoying. The masking bug is 
caused by stack traces. Well, not exactly ... thats kind of a 
simplification. The masking bug is actually caused by stack traces 
generated by throwing exceptions from within sharedmem.c . Here's the 
problem:

Somewhere someone (PosixFileIO) is trying to free some memory that's 
already been freed. The memory manager doesn't like this very much 
because it corrupts the bookkeeping. So sharedmem tries to throw an 
exception (half way through the free ... ick). The exception itself 
tries to generate a stack trace ... which of course tries to allocate 
memory from sharedmem (double ick)... which of course is a little busy 
right now trying to throw an exception. This is when moto puts up the 
white flag and concedes defeat.

There is more than one bad thing about all of this:

1) Sharedmem is too trusting of the programmers who use it. It should 
be able to detect frees of un-allocated memory before it gets into a 
critical section. . The implementation of new bookkeeping must be 
considered in detail. Heap allocation / de-allocation is already one of 
the most expensive operations in moto ... we don't want to slow it down 
further.

2) If the memory manager is going to throw exceptions when bad things 
happen than those exceptions must not attempt to allocate memory from 
shared space  ... what if the exception we want to throw is an 
'OutOfMemoryException' ! We need to consider these sorts of exceptions 
as special ... perhaps using system malloc to store the stack trace.

I'm looking into the shared memory stuff now. The most obvious 'fix' 
for the first issue is simply to check that the previous and subsequent 
nodes in the alist agree that this node is still allocated before 
shared_free enters the critical section. This will work for 99.9999% of 
all double free's ... however since alist nodes are in the same 'space' 
as objects (anodeAdress = address - sizeof(ANode)) there are 
pathological cases where this will fail.

-Dave

On Tuesday, February 4, 2003, at 01:26  AM, Dean Mills wrote:

> I just downloaded and installed moto 0.18 as per the instructions, and
> created a test.moto page using the Midafternoon Commander example on 
> the
> project's website. When I attempt to load the page, Apache crashes. 
> This is
> the contents of the error_log created, it's dying in sharemem.c in the
> deleteFreeNode() function...
>
> ### Pre-Loading Extension 'libmx_cstdlib.so' from
> '/usr/moto/mx/cstdlib/libmx_cstdlib.so'
> ### Pre-Loading Extension 'libmx_codex_util.so' from
> '/usr/moto/mx/codex/util/libmx_codex_util.so'
> ### Pre-Loading Extension 'libmx_codex_io.so' from
> '/usr/moto/mx/codex/io/libmx_codex_io.so'
> ### Pre-Loading Extension 'libmx_codex_http.so' from
> '/usr/moto/mx/codex/http/libmx_codex_http.so'
> ### Pre-Loading Extension 'libmx_codex_db_mysql.so' from
> '/usr/moto/mx/codex/db/mysql/libmx_codex_db_mysql.so'
> ### Pre-Loading Extension 'libmx_moto.so' from
> '/usr/moto/mx/moto/libmx_moto.so'
> [Mon Feb  3 22:05:16 2003] [info] mod_unique_id: using ip addr 
> 127.0.0.1
> ### Pre-Loading Extension 'libmx_cstdlib.so' from
> '/usr/moto/mx/cstdlib/libmx_cstdlib.so'
> ### Pre-Loading Extension 'libmx_codex_util.so' from
> '/usr/moto/mx/codex/util/libmx_codex_util.so'
> ### Pre-Loading Extension 'libmx_codex_io.so' from
> '/usr/moto/mx/codex/io/libmx_codex_io.so'
> ### Pre-Loading Extension 'libmx_codex_http.so' from
> '/usr/moto/mx/codex/http/libmx_codex_http.so'
> ### Pre-Loading Extension 'libmx_codex_db_mysql.so' from
> '/usr/moto/mx/codex/db/mysql/libmx_codex_db_mysql.so'
> ### Pre-Loading Extension 'libmx_moto.so' from
> '/usr/moto/mx/moto/libmx_moto.so'
> [Mon Feb  3 22:05:17 2003] [info] mod_unique_id: using ip addr 
> 127.0.0.1
> [Mon Feb  3 22:05:17 2003] [info] created shared memory segment #851968
> [Mon Feb  3 22:05:18 2003] [notice] Apache/1.3.27 (Unix) 
> mod_moto/0.18.0
> mod_tsunami/1.1 FrontPage/4.0.4.3 mod_ssl/2.8.12 OpenSSL/0.9.6h
> configured -- resuming normal operations
> [Mon Feb  3 22:05:18 2003] [info] Server built: Feb  2 2003 23:53:57
> [Mon Feb  3 22:05:18 2003] [notice] Accept mutex: sysvsem (Default: 
> sysvsem)
> ### Pool doesn't match!
> ### Trying to delete a free node that doesn't exist
> sharedmem.c:443: *** uncaught exception: NullPointerException
>
> ### Pool doesn't match!
> ### Trying to delete a free node that doesn't exist
> sharedmem.c:443: *** uncaught exception: NullPointerException
>
> [Mon Feb  3 22:05:42 2003] [notice] child pid 16509 exit signal 
> Aborted (6)
> [Mon Feb  3 22:05:42 2003] [notice] child pid 16508 exit signal 
> Aborted (6)
>
> As you can see from the [notice] line above, I am running mod_tsunami, 
> FPSE,
> mod_ssl, as well am using mm 1.2.2 and PHP 4.23
> Base system is RH 7.3, Quad Xeon 550's w/ 1GB RAM. Configured moto as
> below...
>
> ./configure --with-apache=/usr/apache --with-mysql=/var/mysql 
> --prefix=/usr/
> moto
> make -j 4 *
> make install
>
> * Failed, seems the threads ran into each other, had to do a `make 
> clean &&
> make` and then all was well.
>
> Sorry in advance if this is a FAQ, I'm just rarin' to get moto going! 
> :)
> TIA!
>
> D.Mills
>
>