Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 31 Mar 2005 01:32:13 -0600
From:      Dan Nelson <dnelson@allantgroup.com>
To:        Matthew Luckie <mjl@luckie.org.nz>
Cc:        freebsd-hackers@freebsd.org
Subject:   Re: kqueue and ordinary files
Message-ID:  <20050331073213.GE46288@dan.emsphone.com>
In-Reply-To: <424BA0B5.2000302@luckie.org.nz>
References:  <424BA0B5.2000302@luckie.org.nz>

next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (Mar 31), Matthew Luckie said:
> Does kqueue signal EOF on an ordinary file when there is nothing left
> to read?
> 
> The code at http://www.wand.net.nz/~mjl12/kqfile.c.txt
> 
> cc -Wall -o kqfile kqfile.c
> ./kqfile kqueue.c
> 
> doesn't ever get EOF notification as far as i can tell.  as in, it
> isn't signaled in kevent.flags, nor does kqueue signal the file is
> ready for reading and then read(2) return 0.
> 
> ident 3 filter 0xffffffff flags 0x0001 fflags 0x0000 data 128
> read 128 bytes
> 
> how should i detect that the file no longer has anything left to read
> with kqueue?  at the moment I use select but would like to use kqueue
> where available.

You can get it indirectly by examining the data field.  You can see
that the call just before the final kqueue returns data=60, so if your
read call returns 60, you're done.  The current behaviour is useful for
things like tail or syslog watchers, so that they get an EVFILT_READ
event when the file grows.  They may be better off registering an
EVFILT_VNODE/NOTE_EXTEND event though, so you could make a case for
returning EV_EOF on EVFILT_READ instead of blocking.

-- 
	Dan Nelson
	dnelson@allantgroup.com



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050331073213.GE46288>