On Wed, Jan 13, 2021 at 04:04:30PM +0100, Christoph Hellwig wrote:
> On Mon, Jan 11, 2021 at 03:23:40PM -0800, Darrick J. Wong wrote:
> > + * 100ths of a second) with the exception of blockgc_timer, which is measured
> > + * in seconds.
> > */
> > xfs_param_t xfs_params = {
> > /* MIN DFLT MAX */
> > @@ -28,8 +28,7 @@ xfs_param_t xfs_params = {
> > .rotorstep = { 1, 1, 255 },
> > .inherit_nodfrg = { 0, 1, 1 },
> > .fstrm_timer = { 1, 30*100, 3600*100},
> > - .eofb_timer = { 1, 300, 3600*24},
> > - .cowb_timer = { 1, 1800, 3600*24},
> > + .blockgc_timer = { 1, 300, 3600*24},
>
> Renaming this is going to break existing scripts. We could either kill off the
> COW timer as it is relatively recent, or we could keep both and use the minimum.
> But removing both and picking an entirely new name seems a little dangerous.
"blockgc_timer" is the internal variable. The xfs_sysctl.c changes at
the bottom of the patch erase speculative_cow_prealloc_lifetime, but the
older knob remains unchanged. See xfs_sysctl.c:
{
.procname = "speculative_prealloc_lifetime",
.data = &xfs_params.blockgc_timer.val,
And from the test system:
# ls /proc/sys/fs/xfs/
total 0
-rw-r--r-- 1 root root 0 Jan 13 15:46 error_level
-rw-r--r-- 1 root root 0 Jan 13 15:46 filestream_centisecs
-rw-r--r-- 1 root root 0 Jan 13 15:46 inherit_noatime
-rw-r--r-- 1 root root 0 Jan 13 15:46 inherit_nodefrag
-rw-r--r-- 1 root root 0 Jan 13 15:46 inherit_nodump
-rw-r--r-- 1 root root 0 Jan 13 15:46 inherit_nosymlinks
-rw-r--r-- 1 root root 0 Jan 13 15:46 inherit_sync
-rw-r--r-- 1 root root 0 Jan 13 15:46 irix_sgid_inherit
-rw-r--r-- 1 root root 0 Jan 13 15:46 irix_symlink_mode
-rw-r--r-- 1 root root 0 Jan 13 15:46 panic_mask
-rw-r--r-- 1 root root 0 Jan 13 15:46 rotorstep
-rw-r--r-- 1 root root 0 Jan 13 15:46 speculative_prealloc_lifetime
-rw-r--r-- 1 root root 0 Jan 13 15:46 stats_clear
-rw-r--r-- 1 root root 0 Jan 13 15:46 xfssyncd_centisecs
--D
>
> Otherwise this looks sane to me.
