Last month's column (available online at http://www.linux-mag.com/2003-03/power_01.html) presented a number of ways to transfer files by hand with ssh and scp and the power of the shell. This month's column looks at ways to transfer and synchronize sets of files automatically with ftp, sftp, and wget. Along the way, you'll also see tips on compressed ssh and detached processes. Let's dig in!
Last month’s column (available online at http://www.linux-mag.com/2003-03/power_01.html) presented a number of ways to transfer files by hand with ssh and scp and the power of the shell. This month’s column looks at ways to transfer and synchronize sets of files automatically with ftp, sftp, and wget. Along the way, you’ll also see tips on compressed ssh and detached processes. Let’s dig in!
Transfer Tips Part I, Continued
First, a note about last month’s column, where we used gzip and bzip2 to compress data transferred through an ssh connection. Separate compression and decompression steps are often unnecessary because many ssh servers have compression built-in.
For example, if the ssh server on host foo supports compression, the command scp -C afile foo: compresses afile, transfers the compressed (intermediate) file to foo, and then uncompresses it into your home directory on foo. The -C option enables compression.
In the same way, the command ssh -C foo compresses all traffic traversing the interactive connection to foo. In fact, compression even works with X forwarding (tunneling X Window System connections through ssh), boosting the performance of tunneled X applications.
ssh compression is most useful over slow network connections like dialup modems. However, if you’re transferring a lot of small files over a slow link, network latency can defeat the time savings from scp -C. That’s because each file has to be negotiated and sent separately — and that “overhead” time can build up. Instead, pipeline the transfer — package all of the files into a single tar archive, and compress/decompress the tar file on-the-fly (with tar -z or ssh -C). And be careful using ssh compression over fast links. On a fast link, the CPU time needed to compress and decompress the data could make the transfer slower!
If you want to experiment with ssh, the time command can help. For example, try sending a big text file from your machine to some host foo with the command time scp -C /usr/dict/words foo:/tmp. Then do it again without the -C option.
For interactive logins, try emitting a lot of text, like head -5000 /usr/dict/words | fmt and compare the scrolling speed in compressed and uncompressed sessions.
By the way, if the ssh server you’re connecting to doesn’t support compression, you’ll get a warning that says, “Remote host refused compression.”
Web browsers and newer utilities like wget and cURL have mostly superseded the old ftp utility. But ftp does transfers interactively, which lets you control its actions. Unlike many FTP protocol-enabled programs, ftp can also send files to an FTP server and even has a little-known configuration file that stores macros and enables automatic logins.
Unfortunately, the FTP protocol is insecure: your password and your data are transmitted in the clear. Traditional FTP is best used for transfers on your internal network or from anonymous-login sites on the Internet. While ftp is insecure, a similar command, sftp, performs FTP-like transfers over an encrypted SSH transport. It also supports the ssh -C option for compressed transfers. However, the remote host has to accept sftp connections; sftp can’t encrypt connections to hosts that have only standard FTP servers.
Your system may have a newer client named ncftp. In fact, some sites have replaced ftp with ncftp. We’re covering old ftp here, though, because it’s widely available (even on non-Unix-like systems).
Listing One shows an example ftp session, shortened and edited to fit the page (many of the status messages have been removed). We log in anonymously to the host ftp.mycorp. biz, get a prompt (ftp>), go to the remote directory /pub/ foo/bar and the local directory /tmp/foobar, retrieve all files whose names end with .rm, then retrieve the file README and rename it readme-xyz. Finally, we upload a file to the server’s /incoming directory.
Listing One: An interactive ftp session
$ ftp ftp.mycorp.biz
Name (ftp.mycorp.biz:jpeek): anonymous
331 Use your e-mail address as password.
ftp> cd /pub/foo/bar
-rw-r–r– … 4378 Sep 8 2000 README
-rw-r–r– … 18312 Sep 7 2000 xyz1.rm
-rw-r–r– … 88115 Sep 7 2000 xyz2.rm
ftp> lcd /tmp/foobar
Local directory now /tmp/foobar
Interactive mode off.
ftp> mget *.rm
226 BINARY Transfer complete
ftp> get README readme-xyz
local: readme-xyz remote: README
ftp> cd /incoming
ftp> put somefile
The ftp client reads a setup file named .netrc in your home directory, which must be readable only by you (use chmod 600 .netrc to set the correct permissions). For each remote host you access, you can make a .netrc entry with the remote username, password (saved in clear text, so use it only for anonymous logins!), and a set of macros (named shortcuts for a series of ftp commands). The special macro init has commands that run each time you connect to the host.
Let’s see a sample .netrc entry that runs the first few commands used in the previous example. (Although we’ve broken the first line for printing, all of it should be entered on one line.) You must place an empty line after each macro.
machine ftp.mycorp.biz login anonymous
You don’t need to run ftp interactively, though. ftp reads commands from the standard input, and writes its prompts and results to standard output. That lets you generate commands on-the-fly — from a shell script, for instance — to drive ftp non-interactively.
The next example shows a fragment of a Bourne shell script that opens a connection to the host $ftpto (with help from a .netrc file). It writes four commands to ftp‘s standard input, using the shell’s “here-document” operator <<pattern. (A here-document feeds text from the script file, line by line, into ftp, until the shell finds the matching pattern on a line by itself. The standard output and standard error are merged (using the shell’s operator 2>&1) and written to the file named in $logfile:
ftp $ftpto <<END_FTP_CMDS >$logfile 2>&1
The first command sets the ascii mode to do automatic end-of-line corrections for text files sent between different types of hosts (from a Linux box to an MS Windows system, for instance). When ftp isn’t reading commands from a prompt, it uses its non-verbose mode; the second command enables verbose mode. The third command sends the file $xfrfile. The fourth command, dir, gets a listing like ls -l, which should show that $xfrfile has been written to the remote host. A quit command isn’t needed; when ftp‘s standard input ends, it quits.
Why Not ftp?
Running ftp interactively lets you see and control the transactions. Running ftp from a script is handy for short jobs, but it isn’t robust. If something goes wrong, the script typically can’t recover. It also doesn’t scale well: writing a script with lots of put or get commands is tedious and not very efficient. The handy Expect utility (covered previously in the January 2001 issue, available online at http://www.linux-mag.com/2001-01/guru_01.html) can do a robust job of managing interactive processes such as ftp. ftp‘s verbose status messages (like 226 BINARY Transfer complete) make error-checking fairly easy for Expect.
Of course, there are other ways to transfer files, including scp and a program in a network-savvy scripting language like Perl. There’s also wget, which is designed to do non-interactive FTP transfers.
What wget Does
The GNU wget utility gets files from the Web with the HTTP and HTTPS protocols. It also supports FTP.
For instance, the command wget http://www.linux-mag.com/ gets a copy of the Linux Magazine home page, which is a file named index.html. By default, wget won’t get other files that make a web page display properly, such as image files and stylesheets. The -p option makes wget analyze an HTML page and retrieve those extra files.
Although you can use wget interactively to get one file at a time, it’s even handier when it retrieves files on its own. For instance, you can log into your office server from home, start a long transfer, then log out and let the transfer complete, as shown in the sidebar “Running a Detached Process.”
Running a Detached Process
A process can run detached from its parent shell. This means the shell won’t wait for the spawned process to finish or kill it when you log out.
Let’s see how to run wget detached so you can log out and let it run. You could do this when you’re logged in from home to your office — over a dialup modem, for instance — and you want wget to download a huge file that will be waiting on your office server in the morning:
mysys$ ssh server.mycorp.biz
server$ (wget -o wlog ftp://foo.com/hugef &)
Connection closed by remote host.
If you use ps -l (lowercase “L”) before logging out, you should see that wget‘s PPID — its parent process ID — is 1 (the Linux init process), not the shell. Some ssh servers still won’t let you log out after starting a detached process, though. In that case, try submitting an at job to start wget soon after you log out. (The February “Power Tools” column, available online at http://www.linux-mag.com/2003-02/power_01.html, describes at.)
And wget does much more. For instance, it can retrieve all files from an FTP area or a web site, recursively. On an FTP site, that job isn’t too tough. For a web site, though, HTML pages have to be parsed and interpreted. The links, which could point to other pages on any other server, may need to be followed. Even more complicated, local copies of HTML files may need global changes so their links point to the local versions of linked-to files. That poses some questions: if image files are in a completely separate directory, should wget store them in the current directory or create a new directory? Should wget follow links from a web page to a parent or sibling directory or to another host, and should wget recursively retrieve from those places too? As it turns out, that’s all configurable from the command line, as well as from two setup files: the system-wide wgetrc and the .wgetrc in your home directory. This helps to explain why wget has so many options! Those options and all the details, are in wget‘s info pages. Type info wget. In this column, we’ll see just a few examples.
Mirroring with wget and cron
wget can mirror files from another host — finding and retrieving only the remote files that differ from the local ones. For instance, wget can make a complete copy of the /pub directory (which typically contains the data file tree) from the FTP server ftp.mycorp.biz and write it into your local directory /mirrors/ftp.mycorp.biz/pub.
The next example shows the wget command to do just that, running nightly from cron. (The command must be entered on a single crontab line, not split as we’ve shown it here.)
System administrators often run jobs from root‘s crontab file, but running wget unattended as root can cause trouble. Unexpectedly-huge files may overfill the local filesystem, crackers can make symbolic links in the local filesystem that cause wget to overwrite system files in other directories, and more. To avoid those problems, we’ve used the command su ftp -c ‘wget ….’ so the root cron job runs wget as the user ftp.
One feature wget uses during mirroring is time-stamping. wget will transfer a remote file if it doesn’t exist locally, or if it’s newer than the local version, or if the remote and local files have different sizes. This is a good example of a task that’s hard to manage with a Web browser or ftp.
Another advantage of a non-interactive command like wget is that you can manage it just like other Linux utilities: you can invoke wget from a script or from the command line. You could start the next example over a dialup connection, then leave home to go shopping while the transfer completes.
cURL and rsync and Beyond
If you like wget, you may like cURL even more because it handles URLs for protocols other than HTTP, HTTPS, and FTP. Also, unlike wget, cURL can send files to a server. It’s available from http://curl.haxx.se.
wget‘s mirroring is one-way: making your local files the same as remote files. It doesn’t work in the other direction, though. However, you can do that with rsync, a non-interactive utility that we’ll cover next month. rsync has a smart difference-finding algorithm, uses pipelining to make transfers efficient, and more. But, for now, happy transfers!
Power Tip: Linux Magazine
The Linux Magazine website is packed with information. The online tables of contents give a good overview, but they can’t possibly list every topic or program mentioned in every article. That’s where a Google search comes in. By using Google’s site: operator to restrict the search to linux-mag.com, you can avoid inferior ;-) sources of information. For instance, to find information on the Expect utility, type the following search into the searchbox at http://www.google.com/.
expect Libes site:linux-mag.com
Adding the name of the author of Expect is a guess (which works!) to keep Google from matching every column that contains the generic word “expect.”
Jerry Peek is a freelance writer and instructor who has used Unix and Linux for over 20 years. He’s happy to hear from readers at firstname.lastname@example.org.