File handle writes on broken connection results in out-of-memory

[jozef]

Hello,

one-liner that will allocate gigabytes of memory:

perl -MNet::SFTP::Foreign -lE 'use Net::SFTP::Foreign::Constants qw(:flags); my $sftp = Net::SFTP::Foreign->new("localhost"); my $fh = $sftp->open("/tmp/test-pls-delete", SSH2_FXF_WRITE|SSH2_FXF_CREAT|SSH2_FXF_TRUNC) or die $sftp->error; print $fh "x"x1000; kill('HUP',$sftp->{pid}); for (1..100_000) { print $fh "x"x1000 or die "print failed" };'

or pretty formatted:

#!/usr/bin/env perl

use Net::SFTP::Foreign;
use Net::SFTP::Foreign::Constants qw(:flags);

my $sftp = Net::SFTP::Foreign->new("localhost");
my $fh =
    $sftp->open( "/tmp/test-pls-delete",
    SSH2_FXF_WRITE | SSH2_FXF_CREAT | SSH2_FXF_TRUNC )
    or die $sftp->error;
print $fh "x" x 1000;

# loosing ssh process results in memory over allocation somewhere in write()
kill( 'HUP', $sftp->{pid} );

for ( 1 .. 100_000 ) {
    print $fh "x" x 1000 or die "print failed";
}

kill( 'HUP', $sftp->{pid} ); line triggers lost-connection conditions, without it script would create ~100MB file in /tmp.

Quick work-around is to set autodie=1.

Would be good if write()/print() functions will return fail in case of broken connection and at the same time not overfill message buffers.

Best regards
Jozef

[salva]

Hi, thank you for the report!

I have just released version 1.92_03 with a fix for this issue: https://metacpan.org/release/SALVA/Net-SFTP-Foreign-1.92_03

[salva]

Hi!

Have you been able to test the new version?

[jozef]

Hello, thank you for the quick fix! I've tested locally right away → works good. I'll be able to test on our staging servers following week with real traffic.

[jozef]

Works good, thank you!