1<!doctype html public "-//W3C//DTD HTML 4.01 Transitional//EN"
2        "http://www.w3.org/TR/html4/loose.dtd">
3
4<html>
5
6<head>
7
8<title>Postfix and NFS</title>
9
10<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
11
12</head>
13
14<body>
15
16<h1><img src="postfix-logo.jpg" width="203" height="98" ALT="">Postfix and NFS</h1>
17
18<hr>
19
20<h2> Postfix support status for NFS </h2>
21
22<p> What is the status of support for Postfix on NFS? The answer
23is that Postfix itself is supported when you use NFS, but there is
24no promise that an NFS-related problem will promptly receive a
25Postfix workaround, or that a workaround will even be possible.
26</p>
27
28<p> That said, Postfix will in many cases work very well on NFS,
29because Postfix implements a number of workarounds (see below).
30Good NFS implementations seldom if ever give problems with Postfix,
31so Wietse recommends that you spend your money wisely.  </p>
32
33<h2> Postfix file locking and NFS </h2>
34
35<p> For the Postfix mail queue, it does not matter how well NFS
36file locking works. The reason is that you cannot share Postfix
37queues among multiple running Postfix instances. You can use NFS
38to switch a Postfix mail queue from one NFS client to another one,
39but only one NFS client can access a Postfix mail queue at any
40particular point in time. </p>
41
42<p> For mailbox file sharing with NFS, your options are to use
43<b>fcntl</b> (kernel locks), <b>dotlock</b> (<i>username</i>.lock
44files), to use both locking methods simultaneously, or to switch
45to maildir format. The maildir format uses one file per message and
46needs no file locking support in Postfix or in other mail software.
47</p>
48
49<p> Many sites that use mailbox format play safe and use both locking
50methods simultaneously. </p>
51
52<blockquote>
53<pre>
54/etc/postfix/main.cf:
55    virtual_mailbox_lock = fcntl, dotlock
56    mailbox_delivery_lock = fcntl, dotlock
57</pre>
58</blockquote>
59
60<h2> Postfix NFS workarounds </h2>
61
62<p> The list below summarizes the workarounds that exist for running
63Postfix on NFS as of the middle of 2003. As a reminder, Postfix
64itself is still supported when it runs on NFS, but there is no
65promise that an NFS-related problem will promptly receive a Postfix
66workaround, or that a workaround will even be possible.  </p>
67
68<ul>
69
70<li> <p> Problem: when renaming a file, the operation may succeed
71but report an error anyway<sup>[1]</sup>. </p>
72
73<p> Workaround: when rename(old, new) reports an error, Postfix
74checks if the new name exists and the old name is gone. If the check
75succeeds, Postfix assumes that the rename() operation completed
76normally.  </p>
77
78<li> <p> Problem: when creating a directory, the operation may succeed
79but report an error anyway<sup>[1]</sup>. </p>
80
81<p> Workaround: when mkdir(new) reports an EEXIST error, Postfix
82checks if the new name resolves to a directory. If the check succeeds,
83Postfix assumes that the mkdir() operation completed normally.  </p>
84
85<li> <p> Problem: when creating a hardlink to a file, the operation
86may succeed but report an error anyway<sup>[1]</sup>. </p>
87
88<p> Workaround: when link(old, new) fails, Postfix compares the
89device and inode number of the old and new files. When the two files
90are identical, Postfix assumes that the link() operation completed
91normally. </p>
92
93<li> <p> Problem: when creating a dotlock (<i>username</i>.lock)
94file, the operation may succeed but report an error anyway<sup>[1]</sup>.
95</p>
96
97<p> Workaround: in this case, the only safe action is to back off
98and try again later.  </p>
99
100<li> <p> Problem: when a file server's "time of day" clock is not
101synchronized with the client's "time of day" clock, email deliveries
102are delayed by a minute or more.  </p>
103
104<p> Workaround: Postfix explicitly sets file time stamps to avoid
105delays with new mail (Postfix uses "last modified" file time stamps
106to decide when a queue file is ready for delivery).  </p>
107
108</ul>
109
110<p> <sup>[1]</sup> How can an operation succeed and report an error
111anyway? </p>
112
113<p> Suppose that an NFS server executes a client request successfully,
114and that the server's reply to the client is lost.  After some time
115the client retransmits the request to the server.  Normally, the
116server remembers that it already completed the request (it keeps a
117list of recently-completed requests and replies), and simply
118retransmits the reply. </p>
119
120<p> However, when the server has rebooted or when it has been very
121busy, the server no longer remembers that it already completed the
122request, and repeats the operation.  This causes no problems with
123file read/write requests (they contain a file offset and can therefore
124be repeated safely), but fails with non-idempotent operations. For
125example, when the server executes a retransmitted rename() request,
126the server reports an ENOENT error because the old name does not
127exist; and when the server executes a retransmitted link(), mkdir()
128or create() request, the server reports an EEXIST error because the
129name already exists. </p>
130
131<p> Thus, successful, non-idempotent, NFS operations will report
132false errors when the server reply is lost, the client retransmits
133the request, and the server does not remember that it already
134completed the request.  </p>
135
136</body>
137</html>
138