On Wed, Apr 27, 2011 at 14:13, Andrea Righi wrote:
Introduce a new fadvise flag to drop page cache pages of a single
filesystem.
At the moment it is possible to drop page cache pages via
/proc/sys/vm/drop_pagecache or via posix_fadvise(POSIX_FADV_DONTNEED).
The first method drops the whole page cache while the second can be used
to drop page cache pages of a single file descriptor. However, there's
not a simple way to drop all the pages of a filesystem (we could scan
all the file descriptors and use posix_fadvise(POSIX_FADV_DONTNEED), but
this solution obviously doesn't scale well).
what if you open the mount point and use POSIX_FADV_DONTNEED on that
dir handle ? if you required write access for that level, it'd also
implicitly take care of the permission issue. but maybe this is just
trying to fit existing code in the wrong way.
-mike
On Wed, Apr 27, 2011 at 02:25:17PM -0400, Mike Frysinger wrote:
On Wed, Apr 27, 2011 at 14:13, Andrea Righi wrote:
quoted
Introduce a new fadvise flag to drop page cache pages of a single
filesystem.
At the moment it is possible to drop page cache pages via
/proc/sys/vm/drop_pagecache or via posix_fadvise(POSIX_FADV_DONTNEED).
The first method drops the whole page cache while the second can be used
to drop page cache pages of a single file descriptor. However, there's
not a simple way to drop all the pages of a filesystem (we could scan
all the file descriptors and use posix_fadvise(POSIX_FADV_DONTNEED), but
this solution obviously doesn't scale well).
what if you open the mount point and use POSIX_FADV_DONTNEED on that
dir handle ? if you required write access for that level, it'd also
implicitly take care of the permission issue. but maybe this is just
trying to fit existing code in the wrong way.
-mike
I still prefer the capability check. I think it's much more simple from
the userspace point of view to be able to specify any file or directory
instead of being forced to retrieve the mountpoint.
However, an advantage with the approach you're proposing is that a
non-privileged user can drop the page cache of a filesystem if it has
write permission in the root of that filesystem.
mmmh.. I don't see big problems also with the interface you propose, if
you all think it's better I can implement this in the next version.
Thanks,
-Andrea