On Tue, 14 Sep 2010 22:22:28 +0200, Arnd Bergmann said:
This changes *all* instances of struct file_operations in
the kernel to have a .llseek operation and then changes
the default to no_llseek, which returns -ESPIPE, which
is what we had decided some time ago in a discussion
with Christoph Hellwig.
I don't suppose there's any clean way to throw a build error or a
printk_on_once() or something if we encounter an unconverted 'struct
file_operations', is there? I have this creeping fear that this patch will go
upstream during the merge window - as will 12 new staging/ drivers from authors
who didn't get the memo yet.
Other than the "missed converting a new usage" issue, it looks OK to me.
On Wednesday 15 September 2010, Valdis.Kletnieks@vt.edu wrote:
Show Details
On Tue, 14 Sep 2010 22:22:28 +0200, Arnd Bergmann said:
quoted
This changes all instances of struct file_operations in
the kernel to have a .llseek operation and then changes
the default to no_llseek, which returns -ESPIPE, which
is what we had decided some time ago in a discussion
with Christoph Hellwig.
I don't suppose there's any clean way to throw a build error or a
printk_on_once() or something if we encounter an unconverted 'struct
file_operations', is there? I have this creeping fear that this patch will go
upstream during the merge window - as will 12 new staging/ drivers from authors
who didn't get the memo yet.
Other than the "missed converting a new usage" issue, it looks OK to me.
One advantage of the scripted approach is that I can run the script
repeatedly, and send a patch after -rc1 fixing up all the new file
operations.
Moreover, any new drivers are by definition not regressions and staging
drivers are expected to have issues. IIRC the most common llseek operation
is 'noop_llseek', which usually means it should have been 'no_llseek' to
start with, but since this has been in a released kernel, some random
user application might expect a seek on that file to succeed without
error.
Arnd