Filezilla Server keeps crashing, even after updating to 9.45

[somebody08]

Please try the next test version: https://filezilla-project.org/codesquid ... _test2.exe

it crashed one more time this morning before I could apply the new DLL

Faulting application name: FileZilla Host.exe, version: 0.9.45.0, time stamp: 0x53d76fae
Faulting module name: ssleay32.dll, version: 1.0.2.0, time stamp: 0x53d76f66
Exception code: 0xc0000005
Fault offset: 0x00007196
Faulting process id: 0x990
Faulting application start time: 0x01cfab876aa3ca23


I've inserted the new ssleay33.dll

[botg]

I'll wait for the next crash then

[somebody08]

crashed this morning, using the new .dll you sent me

Faulting application name: FileZilla Host.exe, version: 0.9.45.0, time stamp: 0x53d76fae
Faulting module name: ssleay32.dll, version: 1.0.2.0, time stamp: 0x53d89540
Exception code: 0xc0000005
Fault offset: 0x00007e08
Faulting process id: 0xd10
Faulting application start time: 0x01cfac163969a693

[botg]

Perfect

Below's disassembly has been created with ssleay32.dll loaded at offset 0x64a20000.

Crash offset in the module is 0x00007e08, making the final address 0x64A27E08

The crash happens in int ssl3_get_client_hello(SSL *s) in OpenSSL's ssl/s3_svr.c

Code: Select all

c=ssl3_choose_cipher(s,s->session->ciphers,
				     SSL_get_ciphers(s));
64A27DF2  mov         ecx,dword ptr [s]  
64A27DF5  push        ecx  
64A27DF6  call        SSL_get_ciphers (64A47A00h)  
64A27DFB  add         esp,4  
64A27DFE  push        eax  
64A27DFF  mov         edx,dword ptr [s]  
64A27E02  mov         eax,dword ptr [edx+0C0h]  
64A27E08  mov         ecx,dword ptr [eax+0BCh]  
64A27E0E  push        ecx  
64A27E0F  mov         edx,dword ptr [s]  
64A27E12  push        edx  
64A27E13  call        ssl3_choose_cipher (64A300A0h)  
64A27E18  add         esp,0Ch  
64A27E1B  mov         dword ptr [c],eax  

This tells me that s->session points to something invalid.

[botg]

I've uploaded a new version, please test https://filezilla-project.org/codesquid ... leay32.dll

New code looks like this:

Code: Select all

c = NULL;
if (s->session)
	{
	c=ssl3_choose_cipher(s,s->session->ciphers,
				SSL_get_ciphers(s));
	}

This either fixes the problem (good), crashes some other place (not so good) or doesn't do anything (terrible)



One thing I wonder: You said it crashed again in the morning. Does it always happen in the morning? What happens mornings?

[somebody08]

here are the historical dates and times of the crashing. they are random, when ever the server has clients connected.

[botg]

Agreed, that looks too random enough for there to be a regular trigger.

Another question: Are you using speed limits in FileZilla Server?

[somebody08]

Agreed, that looks too random enough for there to be a regular trigger.

Another question: Are you using speed limits in FileZilla Server?

No, speed limits are off.

Here are my deviations from the default settings:
Using 6 threads (Intel(R) Xeon(R) CPU E5-2420 0 @ 1.90GHz 6 core, 12 logical)
using custom passive port range 8022 to 9000

the default buffer size were limiting transfer rates to 15Mbps or slower so i increased them
internal transfer buffer size: 524288
socket buffer size: 524288
Enabled FTP over SLL/TLS. Allow explicit FTP over TLS, disallow plain encrypted FTP

[somebody08]

crashed 10pm last night

Faulting application name: FileZilla Host.exe, version: 0.9.45.0, time stamp: 0x53d76fae
Faulting module name: libeay32.dll, version: 1.0.2.0, time stamp: 0x53d76f63
Exception code: 0xc0000005
Fault offset: 0x0000477f
Faulting process id: 0xae8
Faulting application start time: 0x01cfacec1194f0b7

[botg]

Hmpf. Let me see.

[botg]

Happens in n_do_ssl_write, ss_pkt.c

Why is this code even called? It's part of SSLv2. FileZilla Server explicitly disables SSLv2 support.

[somebody08]

is it being called by the client or the server?

[botg]

It all happens in the server, though I don't understand why it happens.

[somebody08]

Is there anything you need me to try?

[botg]

Perhaps. At this point I'm grasping at straws, in particular since the issue is utterly irreproducible on any of my machines. I would like to rule out a potential hardware issue with your machine before analyzing this any further.

Have you ever heard of the Great Internet Mersenne Prime Search? It searches for Mersenne Primes and to do so, the GIMPS tool (aka prime95) requires utmost stability of the system it runs on and contains a torture test specifically designed to detect even the slightest problem with CPU or memory.

Please download prime95 from http://www.mersenne.org/download/ and run its torture test for a couple of hours. Should there be a hardware defect causing problem, this test will likely find it.