Messages in this thread | | | From | Arnd Bergmann <> | Subject | Re: [PATCH 01/15] Introduce noop_llseek() | Date | Sat, 21 Nov 2009 17:56:27 +0100 |
| |
On Friday 20 November 2009, Jan Blunck wrote: > On Fri, Nov 20, Jamie Lokier wrote: > > > Jan Blunck wrote: > > > The noop_llseek() is a llseek() operation that filesystems can use that > > > don't want to support seeking (leave the file->f_pos untouched) but still > > > want to let the syscall itself to succeed. > > > > This is weird behaviour: if you want to allow llseek() to succeed but > > don't really support seeking, why does the device even care about the > > value of file->f_pos? > > The device itself does not care about it but it is userspace that is expecting > the seek to succeed. There is a comment in osst that at least there seems to > be a borken version of tar that wants to seek on the device even it that does > not have any effect.
Looking at the question from the other side -- if the device and the user don't care about file->f_pos, what's wrong with calling generic_file_llseek instead of noop_llseek?
Arnd <><
| |