In my post
Re: ssh returns before remote output complete?
that still hasn't surfaced on the ssh list as far as I have been
able to determine, I made one miniscule error -- the cpio copy-
out command had the "B" flag appended. But I only accounted for
512 bytes for the loss of the last transmitted buffer -- ssh2
tends not to finish sending the final 512 byte buffer over a
link so one needs to compensate by adding 512 bytes of junk at
the end of a transmission, and also, the process must be done
manually so one can ^C it when all done.
This is the fixed version of the now working cpio over ssh2
(fssh-2.0.9) command:
cd /
find proj -depth -print0 | ( cpio -0oa -H crc ; \
dd if=/dev/zero bs=512 count=1 2>/dev/null ) | \
ssh2 -l root ruth '( cd /fscd ; cpio -idum -H crc )'
But if one wants the B flag still in, one must increase the dd
bs correspondingly:
cd /
find proj -depth -print0 | ( cpio -0oaB -H crc ; \
dd if=/dev/zero bs=5120 count=1 2>/dev/null ) | \
ssh2 -l root ruth '( cd /fscd ; cpio -idum -H crc )'
Again, does anyone know the real reason this problem occurs?!
Thank,
________________________________________________________________
Leif Erlingsson, Katrinebergsvagen 70, 146 50 Tullinge, Sweden
TEL +46 8 778-5038, MOB +46 709 14-0631, URL http://www.lege.com
PS1: The above command is copying 1.37 GB of data, so it
isn't finished yet, but once it does, I'll report the
outcome. Should be OK though!
PS2: Background:
Date: Tue, 17 Nov 1998 11:35:42 +0100
From: Roman Maeder <[EMAIL PROTECTED]>
Subject: ssh returns before remote output complete?
I am observing occasional cases where an ssh connection returns
before the remote pipe has finished its output, resulting in
loss of the end of the remote output.
....
Leif:
Myself, I have been able to consistently repeat the problem
given identical circumstances each time. (I have a dedicated
10 Mbps ethernet cable between two compters.)
E.g., if I don't append the 5120 null bytes with the above
cpio command, this is how the destination timestamps look like:
.
.
.
drwxrwxr-x 10 leif adm 3072 Jun 22 15:14 domains
drwxrwxr-x 3 leif adm 1024 Jun 24 20:00 mailx2
drwxrwxrwt 3 root root 3072 Jul 9 17:34 tmp
drwxr-xr-x 6 root root 1024 Jul 10 03:04 ..
drwxrwxr-x 2 leif adm 5120 Jul 10 06:32 scancem
lrwxrwxrwx 1 leif adm 23 Jul 10 06:35 trickokn ->
/usr/local/doc/trickokn
drwxrwxr-x 2 leif adm 1024 Jul 10 06:38 solaris
drwx------ 6 leif adm 3072 Jul 10 07:04 n-plex
While on the source, this is how the last date-stamps look like:
.
.
.
drwxrwxr-x 10 leif adm 3072 Jun 22 15:14 domains/
drwxrwxr-x 3 leif adm 2048 Jun 24 20:00 mailx2/
drwxrwxrwt 3 root root 3072 Jul 9 17:34 tmp/
And the other directories have these datestamps:
drwxrwxr-x 2 leif adm 5120 Sep 11 1998 scancem/
lrwxrwxrwx 1 leif adm 23 Apr 9 1998 trickokn ->
/usr/local/doc/trickokn/
drwxrwxr-x 2 leif adm 1024 Oct 30 1998 solaris/
drwxrwxr-x 6 leif adm 3072 Jun 9 00:21 n-plex/
Executing this command instead, I found that the last 485 bytes
wasn't written:
ssh2 -l root ruth.swip.net \
'( cd /usr/src ; find . -type d -print )' \
2>&1 | tee /tmp/files-as-root
By adding 512 bytes of nulls like below, I got all my real data
through (and plenty of "end-of-string" - characters on top of
that. :-) ):
ssh2 -l root ruth.swip.net \
'( cd /usr/src ; find . -type d -print ; \
dd if=/dev/zero bs=512 count=1 2>/dev/null )' \
2>&1 | tee /tmp/files-as-root
