Hangs when downloading large files

[Greg Ferrar <ferrar@xxxxxxxxxxxxxx>]

Hi,

I've been using libssh successfully for some time as part of my program 
(Sawmill), to fetch files by SFTP. But recently I've gotten two separate 
reports of users who say Sawmill hangs while fetching files. I've 
confirmed this. When SFTP'ing large files (over about 40MB), libssh 
eventually hangs, consistently, at the same point.

It logs in fine, make the connection, downloads part of the data, and 
then hangs.

The traceback at the point of hang is:

#0 0x000000339020cff1 in recv () from /lib64/libpthread.so.0
#1 0x00000000004f2695 in packet_read (session=0x29de150) at 
../src/libssh/packet.c:62
#2 0x00000000004eb1a9 in channel_read (channel=0x29def50, 
buffer=0x2b28d30, bytes=4, is_stderr=<value optimized out>)
at ../src/libssh/channels.c:646
#3 0x00000000004f3727 in sftp_packet_read (sftp=0x29df040) at 
../src/libssh/sftp.c:99
#4 0x00000000004f3d8b in sftp_read_and_dispatch (session=0x9) at 
../src/libssh/sftp.c:158
#5 0x00000000004f461a in sftp_read (handle=0x29ba130, data=0x29c3703, 
len=102316) at ../src/libssh/sftp.c:854
#6 0x000000000070bbda in SFTPSession::ReadFromOpenFile (this=0x29dd6a0, 
buffer=0x7fffc35343d0 "\034\005?,?", bufferLen=16)
at ../src/sftp_session.cpp:305

Here are a couple of interesting points:

 1. If the file is smaller, like 10MB, it completes without any 
problems. Only files more than about 40MB hang.

 2. It happens with /some/ SSH servers, but not others. Specifically, 
if hangs with CentOS 5, x64 (OpenSSH_4.3p2, OpenSSL 0.9.8b 04 May 2006). 
But it completes, downloading the whole 40MB file, on MacOS 10.5, 
PowerPC G5 (OpenSSH_5.2p1, OpenSSL 0.9.7l 28 Sep 2006).

 3. Using the command line "sftp" program from CentOS 5, I am able to 
download the file from either server.

I dug into the code myself a bit, but I don't really understand what's 
happening. It seems as though libssh is expecting more data at the point 
of hang (which seems reasonable, since the file is only partly 
downloaded), but the server has stopped sending.

I'm on libssh 0.4.1 now, but it's still hanging. Is this a known 
problem? Any guesses about what might be going wrong? Need more information?

-
Greg Ferrar