George Spelvin
2020-Mar-28 16:43 UTC
[Ocfs2-devel] [RFC PATCH v1 29/50] fs/ocfs2/journal: Use prandom_u32() and not /dev/random for timeout
get_random_bytes() is expensive crypto-quality random numbers.
If we're just doing random backoff, prandom_u32() is plenty.
(Not to mention fetching 8 bytes of seed material only to
reduce it modulo 5000 is a huge waste.)
Also, convert timeouts to jiffies at compile time; convert
milliseconds to jiffies before picking a random number in the
range to take advantage of compile-time constant folding.
Signed-off-by: George Spelvin <lkml at sdf.org>
Cc: Mark Fasheh <mark at fasheh.com>
Cc: Joel Becker <jlbec at evilplan.org>
Cc: Joseph Qi <joseph.qi at linux.alibaba.com>
Cc: ocfs2-devel at oss.oracle.com
---
fs/ocfs2/journal.c | 7 ++-----
1 file changed, 2 insertions(+), 5 deletions(-)
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 68ba354cf3610..939a12e57fa8b 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -1884,11 +1884,8 @@ int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
*/
static inline unsigned long ocfs2_orphan_scan_timeout(void)
{
- unsigned long time;
-
- get_random_bytes(&time, sizeof(time));
- time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
- return msecs_to_jiffies(time);
+ return msecs_to_jiffies(ORPHAN_SCAN_SCHEDULE_TIMEOUT) +
+ prandom_u32_max(5 * HZ);
}
/*
--
2.26.0
Joseph Qi
2020-Mar-30 12:09 UTC
[Ocfs2-devel] [RFC PATCH v1 29/50] fs/ocfs2/journal: Use prandom_u32() and not /dev/random for timeout
Sorry for the late reply since I might miss this mail. On 2019/3/21 11:07, George Spelvin wrote:> get_random_bytes() is expensive crypto-quality random numbers. > If we're just doing random backoff, prandom_u32() is plenty. > > (Not to mention fetching 8 bytes of seed material only to > reduce it modulo 5000 is a huge waste.) > > Also, convert timeouts to jiffies at compile time; convert > milliseconds to jiffies before picking a random number in the > range to take advantage of compile-time constant folding. > > Signed-off-by: George Spelvin <lkml at sdf.org> > Cc: Mark Fasheh <mark at fasheh.com> > Cc: Joel Becker <jlbec at evilplan.org> > Cc: Joseph Qi <joseph.qi at linux.alibaba.com> > Cc: ocfs2-devel at oss.oracle.com > --- > fs/ocfs2/journal.c | 7 ++----- > 1 file changed, 2 insertions(+), 5 deletions(-) > > diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c > index 68ba354cf3610..939a12e57fa8b 100644 > --- a/fs/ocfs2/journal.c > +++ b/fs/ocfs2/journal.c > @@ -1884,11 +1884,8 @@ int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) > */ > static inline unsigned long ocfs2_orphan_scan_timeout(void) > { > - unsigned long time; > - > - get_random_bytes(&time, sizeof(time)); > - time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); > - return msecs_to_jiffies(time); > + return msecs_to_jiffies(ORPHAN_SCAN_SCHEDULE_TIMEOUT) + > + prandom_u32_max(5 * HZ);Seems better include the prandom_u32_max() into msecs_to_jiffies()? Thanks, Joseph> } > > /* >