aboutsummaryrefslogtreecommitdiffstats
path: root/fetchmail.man
diff options
context:
space:
mode:
authorEric S. Raymond <esr@thyrsus.com>1996-10-19 19:03:31 +0000
committerEric S. Raymond <esr@thyrsus.com>1996-10-19 19:03:31 +0000
commit07513709eed70fe7fcf7361eaf17640ed74683bf (patch)
treeab22ca5334e422d176327e5162c7390dc3c369aa /fetchmail.man
parent5bd3f8287149511024b95c01fb4212e60ec6fe13 (diff)
downloadfetchmail-07513709eed70fe7fcf7361eaf17640ed74683bf.tar.gz
fetchmail-07513709eed70fe7fcf7361eaf17640ed74683bf.tar.bz2
fetchmail-07513709eed70fe7fcf7361eaf17640ed74683bf.zip
Avoid the socket-exhaustion bug.
svn path=/trunk/; revision=354
Diffstat (limited to 'fetchmail.man')
-rw-r--r--fetchmail.man10
1 files changed, 10 insertions, 0 deletions
diff --git a/fetchmail.man b/fetchmail.man
index 03a72a69..b2165186 100644
--- a/fetchmail.man
+++ b/fetchmail.man
@@ -642,6 +642,16 @@ to the mailserver. This creates a risk that name/password pairs
might be snaffled with a packet sniffer or more sophisticated
monitoring software.
.PP
+Under Linux, if fetchmail is run in daemon mode with the network
+inaccessible, each poll leaves a socket allocated but in CLOSE state
+(this is visible in netstat(1)'s output). For some reason, these
+sockets aren't garbage-collected until \fIfetchmail\fR exits. When
+whatever kernel table is involved fills up, fetchmail can no longer
+run even if the network is up. This appears \fInot\fR to be a socket
+leak in \fIfetchmail\fR, but rather some glitch or misfeature in the system
+network code. To avoid this problem, fetchmail commits seppuku after
+too many unsuccessful socket opens.
+.PP
Send comments, bug reports, gripes, and the like to Eric S. Raymond
<esr@thyrsus.com>.
.SH SEE ALSO