| Message ID | 20180620082937.719067-1-arnd@arndb.de |
|---|---|
| State | Superseded |
| Delegated to: | Richard Weinberger |
| Headers | show |
| Series | ubifs: tnc: use monotonic znode timestamp | expand |
Arnd, Am Mittwoch, 20. Juni 2018, 10:29:11 CEST schrieb Arnd Bergmann: > The tnc uses get_seconds() based timestamps to check the age of a znode, > which has two problems: on 32-bit architectures this may overflow in > 2038 or 2106, and it gives incorrect information when the system time > is updated using settimeofday(). > > Using montonic timestamps with ktime_get_seconds() solves both thes > problems. > > Signed-off-by: Arnd Bergmann <arnd@arndb.de> > --- > fs/ubifs/shrinker.c | 2 +- > fs/ubifs/tnc.c | 4 ++-- > fs/ubifs/tnc_misc.c | 2 +- > fs/ubifs/ubifs.h | 2 +- > 4 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/fs/ubifs/shrinker.c b/fs/ubifs/shrinker.c > index 9a9fb94a41c6..9d10cbdec2cc 100644 > --- a/fs/ubifs/shrinker.c > +++ b/fs/ubifs/shrinker.c > @@ -71,7 +71,7 @@ static int shrink_tnc(struct ubifs_info *c, int nr, int age, int *contention) > { > int total_freed = 0; > struct ubifs_znode *znode, *zprev; > - int time = get_seconds(); > + time64_t time = ktime_get_seconds(); ubifs does abs(time - znode->time) >= age) { Is this still legit with time64_t? Thanks, //richard
Am Montag, 2. Juli 2018, 16:24:15 CEST schrieb Richard Weinberger: > Arnd, > > Am Mittwoch, 20. Juni 2018, 10:29:11 CEST schrieb Arnd Bergmann: > > The tnc uses get_seconds() based timestamps to check the age of a znode, > > which has two problems: on 32-bit architectures this may overflow in > > 2038 or 2106, and it gives incorrect information when the system time > > is updated using settimeofday(). > > > > Using montonic timestamps with ktime_get_seconds() solves both thes > > problems. > > > > Signed-off-by: Arnd Bergmann <arnd@arndb.de> > > --- > > fs/ubifs/shrinker.c | 2 +- > > fs/ubifs/tnc.c | 4 ++-- > > fs/ubifs/tnc_misc.c | 2 +- > > fs/ubifs/ubifs.h | 2 +- > > 4 files changed, 5 insertions(+), 5 deletions(-) > > > > diff --git a/fs/ubifs/shrinker.c b/fs/ubifs/shrinker.c > > index 9a9fb94a41c6..9d10cbdec2cc 100644 > > --- a/fs/ubifs/shrinker.c > > +++ b/fs/ubifs/shrinker.c > > @@ -71,7 +71,7 @@ static int shrink_tnc(struct ubifs_info *c, int nr, int age, int *contention) > > { > > int total_freed = 0; > > struct ubifs_znode *znode, *zprev; > > - int time = get_seconds(); > > + time64_t time = ktime_get_seconds(); > > ubifs does > abs(time - znode->time) >= age) { > > Is this still legit with time64_t? Answering my own question, yes. abs() seems to be able to deal with 64bit numbers and time64_t is just a number. Thanks, //richard
diff --git a/fs/ubifs/shrinker.c b/fs/ubifs/shrinker.c index 9a9fb94a41c6..9d10cbdec2cc 100644 --- a/fs/ubifs/shrinker.c +++ b/fs/ubifs/shrinker.c @@ -71,7 +71,7 @@ static int shrink_tnc(struct ubifs_info *c, int nr, int age, int *contention) { int total_freed = 0; struct ubifs_znode *znode, *zprev; - int time = get_seconds(); + time64_t time = ktime_get_seconds(); ubifs_assert(mutex_is_locked(&c->umount_mutex)); ubifs_assert(mutex_is_locked(&c->tnc_mutex)); diff --git a/fs/ubifs/tnc.c b/fs/ubifs/tnc.c index 4a21e7f75e7a..6c497fb48b98 100644 --- a/fs/ubifs/tnc.c +++ b/fs/ubifs/tnc.c @@ -1179,7 +1179,7 @@ int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key, { int err, exact; struct ubifs_znode *znode; - unsigned long time = get_seconds(); + time64_t time = ktime_get_seconds(); dbg_tnck(key, "search key "); ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY); @@ -1315,7 +1315,7 @@ static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key, { int err, exact; struct ubifs_znode *znode; - unsigned long time = get_seconds(); + time64_t time = ktime_get_seconds(); dbg_tnck(key, "search and dirty key "); diff --git a/fs/ubifs/tnc_misc.c b/fs/ubifs/tnc_misc.c index 93f5b7859e6f..49f647bc88f0 100644 --- a/fs/ubifs/tnc_misc.c +++ b/fs/ubifs/tnc_misc.c @@ -435,7 +435,7 @@ struct ubifs_znode *ubifs_load_znode(struct ubifs_info *c, zbr->znode = znode; znode->parent = parent; - znode->time = get_seconds(); + znode->time = ktime_get_seconds(); znode->iip = iip; return znode; diff --git a/fs/ubifs/ubifs.h b/fs/ubifs/ubifs.h index 04bf84d71e7b..54fb6b1cd042 100644 --- a/fs/ubifs/ubifs.h +++ b/fs/ubifs/ubifs.h @@ -758,7 +758,7 @@ struct ubifs_znode { struct ubifs_znode *parent; struct ubifs_znode *cnext; unsigned long flags; - unsigned long time; + time64_t time; int level; int child_cnt; int iip;
The tnc uses get_seconds() based timestamps to check the age of a znode, which has two problems: on 32-bit architectures this may overflow in 2038 or 2106, and it gives incorrect information when the system time is updated using settimeofday(). Using montonic timestamps with ktime_get_seconds() solves both thes problems. Signed-off-by: Arnd Bergmann <arnd@arndb.de> --- fs/ubifs/shrinker.c | 2 +- fs/ubifs/tnc.c | 4 ++-- fs/ubifs/tnc_misc.c | 2 +- fs/ubifs/ubifs.h | 2 +- 4 files changed, 5 insertions(+), 5 deletions(-)