Optional Packages: libnet, mailtools Sooner or later, most Perl programmers find themselves needing to send electronic mail from a program. It might be for a CGI application, or sending out autoconfirmations, or generating monthly e-mail invoices, or sending reports to the payroll department. Now, the normal way to send e-mail from Perl looks something like this:
open (SENDMAIL, "|/usr/lib/sendmail -t") or die "sendmail"; print SENDMAIL "From: (goes here)\n"; print SENDMAIL "To: (goes here)\n"; print SENDMAIL "Subject: Test mail\n\n"; print SENDMAIL "Message body goes here"; close SENDMAIL or die "Hey, sendmail failed!";
Nice, simple, and straightforward, right?
Well... no. It's not this simple if you're on a non-Unix machine (such as Windows, NT, VMS, OS/390, or the Mac) or, for whatever reason, you don't have sendmail or a reasonable facsimile installed. And even if you do have sendmail, you might not want to pay the cost to invoke it just to send a single mail message. It is, after all, a fairly big program, and the last thing you really want is for your system monitoring program to drop dead because it couldn't fork off sendmail to tell you it was out of process slots.
So, what we're going to do in this article is build a sendmail replacement suitable both for use in place of a real sendmail and as donor code which you will undoubtedly gut for your own use. (I use the mail sending routines in some system monitoring programs, for example.)
Before we go into the code that actually sends the mail, we need to talk about what exactly e-mail is and how it works.
The single most important thing about e-mail that anyone can know is that e-mail is not a direct delivery, point-to-point system like, for example, the web. Instead, e-mail is a store-and-forward system. What this means is that when you send mail to someone it doesn't go directly to their computer. Instead, it goes to some intermediate machine, which itself might forward the mail to another machine before the mail gets to its final destination.
And, in many cases, the machine you think is the final delivery system isn't. It's common for the machine publicly advertised in e-mail addressed to be just a gateway to another mail system. Corporate mail systems are often like this, as the host you think is the final destination is really a gateway through a firewall, a gateway to a completely different mail system (such as GroupWise or Lotus CC:Mail), or both.
The standards for e-mail, like the standards for most everything else on the Internet, are set down in a series of documents called RFCs. (Short for Request for Comments, although the standards set down are quite a bit stronger than the name might imply. See ftp://ftpeng.cisco.com/fred/rfc-index/rfc.html for a comprehensive list.) The standards for SMTP, the Simple Mail Transport Protocol, are set down in RFC821, while the standards for e-mail headers are set down in RFC822.
The most common way to deliver mail from one system to another is via SMTP, which requires a network connection. Usually that connection is TCP/IP, but SMTP will work over a variety of other network protocols. The other main delivery method is UUCP. UUCP is normally a dialup connection, and doesn't require constant connectivity between hosts. With UUCP, the client system occasionally polls a host system for any pending work. (UUCP is used to transfer more than just mail. It's designed to transfer arbitrary files between two systems, and is often used to move Usenet news in addition to mail.)
There are many different e-mail systems kicking around, but we're going to be talking about the most common type, normally called "Internet e-mail." It's a bit of a misnomer, though, since it's certainly not restricted to machines on the Internet.
Since the code we're going to work with sends mail over the network via SMTP, we need to make sure that it follows the standards properly. Not to worry, though-- SMTP is simple.
Finally, let's talk a bit about the actual message. While it might come as something of a surprise, the message is actually the least interesting part of mail delivery.
E-mail sent via SMTP has three parts, which is probably one part more than you were expecting. The contents of e-mail is also governed by standards, in this case RFC822 and its descendants. (If you're going to be building e-mail messages, you should definitely read this RFC. You might also want to read some of the follow-on RFCs that extend 822.)
The least interesting part of a mail message, at least as far as we're concerned, is the message body. While people care the most about this bit, the mail transport software doesn't care in the least about it. It's just data that needs to be sent on its way. The body consists of everything after the first blank line in the message.
All the lines up to the first blank line in the message are considered the message headers. In general, mail transport software doesn't care very much at all about the message headers. With one exception, the mail software isn't supposed to mess with the headers at all. Nor is it supposed to pay any attention to them.
The single exception to the "leave the headers alone" rule is the Received: header. Each mail system that relays a message is supposed to add a Received: header at the top of each message. You can use these headers to track the systems that a message has passed through.
RFC822 does mandate a minimum set of headers. To be fully compliant, all mail should have at least a From:, Date:, and either a To: or Bcc: header. There are others, of course, listed in the various RFCs. The standards mandate that only headers listed in the RFCs can be in a header--you can't make up your own. RFC822 grants a single exception--headers that begin with X- are allowed.
When one system sends mail to another, it doesn't just open up a connection and start pumping data across. There's a conversation that takes place where the sender tells the receiver who it is, who the mail is for, and who the mail is from. This information is called the envelope because it acts like an envelope for a letter--it encloses the letter but isn't actually a part of it. The information in the envelope is the only information used by the receiving system to determine what it should do with the mail and who should receive it. The recipients listed in the mail headers don't have anything to do with who the mail is ultimately delivered to.
An SMTP transaction is a reasonably simple thing. Your program connects to the remote server, identifies itself, tells the remote machine who the mail is from, who it's to, and then sends the message. Simple, huh? All the details are specified in RFC821, and if you're going to be writing an SMTP client you should definitely read it before writing any code.
There are two gotchas involved in SMTP conversations that might slip past a quick reading of the RFC. First, all lines end with a CRLF (carriage-return and linefeed) pair, not a bare linefeed. \n is not appropriate as a line ending. Use \cM\cJ instead. Second, RFC821 says that the SMTP channel is 7-bit and the high bits of all bytes should be cleared. This restriction is lifted in a later RFC, but it requires a mildly different, and somewhat more complex, conversation between client and server.
Conversations between client and server are fairly straightforward. The client sends a command with an optional parameter and then a CRLF. The server responds with a three-digit status code and a one-line message describing the status. The first digit provides the general message class--2xx for OKs, 3xx for start mail transmission, 4xx for temporary errors, and 5xx for permanent errors. You can check the first digit for the general status of the request and proceed accordingly. It's best to abort the transaction and try again later on 4xx errors, but abort the transaction and give up on 5xx errors -- except in special cases. It's silly to toss a whole mail message if one recipient in a list of a dozen fails, for example. SMTP places some limits on the transmission of commands and messages; they're listed in Table 1. These numbers are what RFC821 requires from a conformant SMTP server and client. Servers can (but don't have to) accept larger values.
According to the standard, SMTP servers are supposed to be listening on TCP port 25, so the client should open a connection there. When the connection is successfully made, the server will send out a single-line message with a status code.
After connecting, the next step is to identify ourselves. This is done with the HELO command (no, that's not a typo). The full command is:
While many servers will still accept a HELO with no (or a bogus) name, many no longer do because of spam. (A common spammer trick was to leave off the name or mis-identify themselves to try to slip past security measures.)
Next, your program needs to say who's sending the mail. This is done with the MAIL command, which looks like this:
MAIL FROM: <email@example.com> CRLF
Note that the contents of <> are real--you need to store the sending e-mail address there. This is the address that all bounced mail will go to. If the mail being sent cannot be bounced (because, for example, it's a bounce notice itself and thus bouncing it would set up a nasty loop) the address can be an empty <>.
After telling the remote system who the mail is from, you need to tell it who the mail is going to. This is done with the RCPT command. Your program sends out one or more RCPTs, which look like this:
RCPT TO: <firstname.lastname@example.org>CRLF
Once again, the addresses must be enclosed in angle brackets. You may send up to 100 RCPTs per mail message, and the server will send back a status on each one. It's not uncommon to have a message with multiple recipients bounce one or two.
Once you've identified the sender and recipients, it's time to send out the mail itself. The DATA command marks the start of the message itself, which continues until you send a line with just a period. If you need to send a line with just a period on it for some reason, the protocol allows for sending a doubled period, which the server converts to a single period. When you do send out the line with a single period to end the mail message, the server will respond with a status message that tells you what it thinks of your mail.
Finally, you need to close down the connection by sending the QUIT command. While it's not uncommon to see programs send a QUIT and drop the connection, you should wait for the final status to make sure that things went properly.
So, we now know how to send mail in general. The next step is sending mail in specific. How do we send mail in Perl? Well, we're going to show you.
There are three common ways to send mail directly from Perl: Net::SMTP, Mail::Mailer, and talking directly to the mail host. We'll go over each of these methods in turn. We'll present an example of sending mail with each of the three different methods.
Mail::Mailer, included as part of the MailTools module, sits on top of Net::SMTP, so it requires the libnet bundle to run. Both modules are well worth the time, so install (or get your sysadmin to install) them into your production tree. They're worth it. Anyway, as you can see from the sample program, sending mail with Mail::Mailer's pretty straightforward. The whole thing's only 14 lines, and that's including comments.
The one thing to keep in mind with Mail::Mailer is that it figures out the sender and recipients from the headers of the mail message, so you lose the separation of the envelope and headers. That's not necessarily bad, just something to remember.
Net::SMTP, which is included as part of the libnet bundle, puts a fairly thin wrapper around the SMTP conversation, just barely encapsulating it. There's one function call for each step of the SMTP conversation, with just enough code wrapped around things to perform tasks like error checking.
As you can see from the example, it's a bit more verbose than using Mail::Mailer, though not by much. You've got full control over the SMTP conversation this way, although you don't get fine-grained response codes.
Finally, there's the fully manual way. It's significantly longer than any of our other methods -- too long for this magazine, but you can find my smailer.pl program on the TPJ web site. The reason it's so long is that we include the code that actually sends the mail out. In the other examples that code is encapsulated in other modules.
We're not going to spend too much time going over this code, because you can't see it here. However, we'll touch on a few high points. You'll notice, for example, that we use syswrite to write the data out, while we use <> to read it in, and on two different filehandles to boot. Why?
Well, the big reason is to compensate for a variety of odd socket and stdio implementations on the platforms that Perl runs on. syswrite is guaranteed to work for writing, and <> works okay for reading. They each use a separate filehandle to make sure there aren't any underlying stdio buffer collisions.
You'll also note that the code doesn't really care much which non-2xx response code it gets--it considers them all errors and returns the status to the caller of the subroutine, who can hopefully do something reasonable with the return values.
Other than that, the code is pretty straightforward. The smailer() subroutine is written such that you ought to be able to just cut and paste it into your own code and have it run. (You can, for example, fill in some of the empty variables at the top of the subroutine if you want to avoid the overhead of computing them on every invocation.)
Now that we've covered the three different methods, the next question is--which one is the right one? The answer, of course, is "it depends."
The hand-rolled method is the lightest of the three methods: It's snappy and takes up very little memory, although it does sacrifice some flexibility. For instance, it doesn't handle ESMTP, the extension to SMTP that allows for delivery notifications and 8-bit characters in mail messages.
Net::SMTP is more flexible than the hand-rolled way of doing things, but you pay for that increased flexibility in several ways. It's fully object-oriented, which means that it incurs Perl's not insubstantial method call overhead. It's also layered on top of the IO modules, which are themselves quite large. (This is only a consideration if you're not already using them, of course.) On the other hand, it does talk ESMTP, and exposes SMTP in easy-to-use pieces.
Mail::Mailer is the most flexible of the three methods. In addition to connecting via SMTP, it can also send mail by invoking sendmail, or by firing off mail or an equivalent. It's also got the highest overhead in terms of both speed and memory usage. It's layered on top of Net::SMTP, and so inherits the Net::SMTP overhead as well as generating some of its own.
In general, I'd recommend using Mail::Mailer for general-purpose mailing work. While it's got some overhead, the delays you might incur will be far less than typical network latency. The extra flexibility and robustness are worth it.
If you find the overhead of Mail::Mailer is a bit much, especially if your need to send mail using minimal computational resources, go for the hand-rolled approach. Also, if you're not using any of the IO modules, and you're not sending mail that's got 8-bit characters in it, the hand-rolled method should work well.
Finally, for the very curious, it is possible to write a reasonable emulation of sendmail with Perl and these mail routines, at least enough to send mail with it. I have an implementation that I mess with on and off. Versions occasionally make it to my CPAN directory, so you can always check there to see how it's going.
Dan Sugalski is the VMS Systems Administrator for the Oregon University System's ITS department. He's been involved in the VMS Perl port for a few years, likes threading, mail, doing obscure things in XS, and tilting at windmills. To reach him you can either leave your message and a plate of cookies under the bushes by your front door or send mail to email@example.com . One of the big downsides to presenting code to send e-mail is the possibility of it being used for spamming. While there's no way short of violence to stop antisocial morons, it never hurts to have the guys in expensive suits and thin watches on your side. So, the copyright notice: All the code presented in this article is Copyright 1999 Dan Sugalski (The text, of course, is covered by TPJ's copyright.) It may be used for any purpose you like as long as you don't misrepresent it as your own, or send out unsolicited mass e-mail of any sort.