How can I resume a large scp file transfer when using port forwarding?

Question

I have a machine a couple of hops away, and I need to set up port forwarding in order to transfer files.

Edit: To be clear, the multiple hops are needed in order to access the remote machine. From my machine, I set up a VPN, where I can access 10.255.x.x - that is the only machine I can connect to via the VPN. Once logged into .x.x, I can then connect to other machines - .y.y being one of those.

From my machine:

ssh -L 4567:localhost:4567 [email protected]

Then from that machine:

ssh -L 4567:localhost:22 [email protected]

I can then

scp -P 4567 me@localhost:/path/to/large/file.gz .

I left this running overnite, only to find that the transfer died at some point.

I've seen a couple of suggestions to use rsync over ssh to resume the transfer, but I'm unclear how to set that up. Is this possible?

Solution 1

With some versions of scp (the version on the source computer seems to be the key), just re-executing the scp command will suffice to resume the transfer. But be careful! If your version doesn't support partial transfers, the partial file will simply be overwritten.

The following rsync switches are handy for resuming a broken transfer if scp doesn't support it:

     --append                append data onto shorter files
     --append-verify         like --append, but with old data in file checksum
 -e, --rsh=COMMAND           specify the remote shell to use
     --progress              show progress during transfer

The command

rsync --append-verify --progress --rsh="ssh -p 4567" me@localhost:/path/to/large/file.gz .

should have the desired effect. Note that the -p switch must be lowercase for ssh.

Solution 2

Use sftp instead of scp

In your case:

sftp -a -P 4567 me@localhost:/path/to/large/file.gz .

From the sftp man page:

-a      Attempt to continue interrupted downloads rather than overwriting existing partial or complete copies of files.  If the remote file contents differ from the partial local copy then the resultant file is likely to be corrupt.

Solution 3

rsync uses ssh by default, you may have to specify the exact ssh command using rsync's -e switch. It also has a --partial which should keep the incomplete file around so it can resume the transfer.

Comments

[Dennis]

What are all these hops for? Wouldn't scp [email protected]:/path/to/large/file.gz . achieve exactly the same? Which version of scp (ssh) is installed on client and server?

[chris]

@Dennis: scp does not support resuming a transfer - what you suggest will restart the download.

[Dennis]

Some versions do (my laptop's scp does, my VPS's doesn't). The reason I'm asking is because the hops seems unnecessary to me, and rsyncing without hops will be a lot easier.

[Alex Allen]

This answer might help you as well: unix.stackexchange.com/a/43097/14084

[chris]

The hops are needed because of the way the network on the destination server is configured. The problem with using rsync over ssh is that I need to use the local port 4567, the file does not exist on the 10.255.x.x server.

[Dennis]

My mistake. I was assuming that 10.255.y.y was the client computer.

[Dennis]

I've updated my answer.

[chris]

@ingenous: I added an explanation for the tunnels.

[dr.dimitru]

and what to do if there is no way to use rsync? In my case I have only ftp available on remote host.

[Dennis]

@Enzo Because, as I said in the first paragraph, not all versions of scp support this. (That may be less of a problem 5 years later.) Also, because I assumed the OP cared more about resuming the download than using a particular tool for it.

[EnzoR]

TheOP is talking explicitly about SCP. You can also switch to HTTP(S) with curl or to FTP(S) and even SFTP with similar results. But the OP is still talking about scp. When the OP will be edited to widen the problem, then your answer will make some sense. At the moment it's offtopic.

[EnzoR]

Moreover, I'd have quoted exactly which scp implemetation supports the resume. I am not aware of any such scp impleemtation. You left your answer in the haze, instead.

[Dennis]

@Enzo How am I supposed to know which scp implementations support resume? Yours either does or it doesn't. If it does, you can just re-run the command; if it doesn't, you can use rsync instead. Both cases are covered in my answer. If you think it isn't helpful, feel free to downvote it.

[EnzoR]

No scp implementation, afaik, supports resume as it's a plain "echo" over an SSH channel. SFTP/FTPS/HTTP(S) can support it and most of the opensource implementations support it. You said that "With some versions of scp (the version on the source computer seems to be the key)", so you say you know. While you now say actually don't. The question is about SCP, the answer needs to be about scp. And the only one answers is "you cannot".

[Dennis]

@Enzo The question is about SCP, the answer needs to be about scp. The question is about an incomplete scp transfer, not about the scp tool. It literally says I've seen a couple of suggestions to use rsync over ssh to resume the transfer, but I'm unclear how to set that up. Is this possible? at the bottom.

[Dallas Clarke]

You can use reput or reget if you log into the shell.