On Wed, Nov 03, 2010 at 10:38:33AM -0600, Eric Blake wrote:
On 11/03/2010 07:23 AM, Daniel P. Berrange wrote:
I was wondering about reusing saferead/write but they have a different semantic on blocking,
Yep, saferead/write cannot be used on any FD with O_NONBLOCK set, because they'll just spin in a 100% CPU loop whenever EAGAIN occurs, until they get nbytes worth of data.
Is it worth teaching saferead/write to use fcntl() to determine if an fd is O_NONBLOCK? And if so, should it outright reject an O_NONBLOCK fd (to help us diagnose bugs) or be documented as allowing an EAGAIN failure on non-blocking fds?
The API design doesn't lend itself to this, because it would result in a horribly inefficient way of reading data fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); fcntl(fd) read(fd, buf, 1024); ... Daniel -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://deltacloud.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|