From bf2ba432213fade50dd39f2e348085b758c0726e Mon Sep 17 00:00:00 2001 From: Luis Henriques Date: Tue, 6 Jul 2021 14:52:41 +0100 Subject: ceph: reduce contention in ceph_check_delayed_caps() Function ceph_check_delayed_caps() is called from the mdsc->delayed_work workqueue and it can be kept looping for quite some time if caps keep being added back to the mdsc->cap_delay_list. This may result in the watchdog tainting the kernel with the softlockup flag. This patch breaks this loop if the caps have been recently (i.e. during the loop execution). Any new caps added to the list will be handled in the next run. Also, allow schedule_delayed() callers to explicitly set the delay value instead of defaulting to 5s, so we can ensure that it runs soon afterward if it looks like there is more work. Cc: stable@vger.kernel.org URL: https://tracker.ceph.com/issues/46284 Signed-off-by: Luis Henriques Reviewed-by: Jeff Layton Signed-off-by: Ilya Dryomov --- fs/ceph/caps.c | 17 ++++++++++++++++- fs/ceph/mds_client.c | 25 ++++++++++++++++--------- fs/ceph/super.h | 2 +- 3 files changed, 33 insertions(+), 11 deletions(-) (limited to 'fs') diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c index 7bdefd0c789a..2a2900903f8c 100644 --- a/fs/ceph/caps.c +++ b/fs/ceph/caps.c @@ -4150,11 +4150,19 @@ bad: /* * Delayed work handler to process end of delayed cap release LRU list. + * + * If new caps are added to the list while processing it, these won't get + * processed in this run. In this case, the ci->i_hold_caps_max will be + * returned so that the work can be scheduled accordingly. */ -void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) +unsigned long ceph_check_delayed_caps(struct ceph_mds_client *mdsc) { struct inode *inode; struct ceph_inode_info *ci; + struct ceph_mount_options *opt = mdsc->fsc->mount_options; + unsigned long delay_max = opt->caps_wanted_delay_max * HZ; + unsigned long loop_start = jiffies; + unsigned long delay = 0; dout("check_delayed_caps\n"); spin_lock(&mdsc->cap_delay_lock); @@ -4162,6 +4170,11 @@ void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) ci = list_first_entry(&mdsc->cap_delay_list, struct ceph_inode_info, i_cap_delay_list); + if (time_before(loop_start, ci->i_hold_caps_max - delay_max)) { + dout("%s caps added recently. Exiting loop", __func__); + delay = ci->i_hold_caps_max; + break; + } if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && time_before(jiffies, ci->i_hold_caps_max)) break; @@ -4177,6 +4190,8 @@ void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) } } spin_unlock(&mdsc->cap_delay_lock); + + return delay; } /* diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 9db1b39df773..afdc20213876 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -4490,22 +4490,29 @@ void inc_session_sequence(struct ceph_mds_session *s) } /* - * delayed work -- periodically trim expired leases, renew caps with mds + * delayed work -- periodically trim expired leases, renew caps with mds. If + * the @delay parameter is set to 0 or if it's more than 5 secs, the default + * workqueue delay value of 5 secs will be used. */ -static void schedule_delayed(struct ceph_mds_client *mdsc) +static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay) { - int delay = 5; - unsigned hz = round_jiffies_relative(HZ * delay); - schedule_delayed_work(&mdsc->delayed_work, hz); + unsigned long max_delay = HZ * 5; + + /* 5 secs default delay */ + if (!delay || (delay > max_delay)) + delay = max_delay; + schedule_delayed_work(&mdsc->delayed_work, + round_jiffies_relative(delay)); } static void delayed_work(struct work_struct *work) { - int i; struct ceph_mds_client *mdsc = container_of(work, struct ceph_mds_client, delayed_work.work); + unsigned long delay; int renew_interval; int renew_caps; + int i; dout("mdsc delayed_work\n"); @@ -4545,7 +4552,7 @@ static void delayed_work(struct work_struct *work) } mutex_unlock(&mdsc->mutex); - ceph_check_delayed_caps(mdsc); + delay = ceph_check_delayed_caps(mdsc); ceph_queue_cap_reclaim_work(mdsc); @@ -4553,7 +4560,7 @@ static void delayed_work(struct work_struct *work) maybe_recover_session(mdsc); - schedule_delayed(mdsc); + schedule_delayed(mdsc, delay); } int ceph_mdsc_init(struct ceph_fs_client *fsc) @@ -5030,7 +5037,7 @@ void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg) mdsc->mdsmap->m_epoch); mutex_unlock(&mdsc->mutex); - schedule_delayed(mdsc); + schedule_delayed(mdsc, 0); return; bad_unlock: diff --git a/fs/ceph/super.h b/fs/ceph/super.h index 6b6332a5c113..9215a2f4535c 100644 --- a/fs/ceph/super.h +++ b/fs/ceph/super.h @@ -1167,7 +1167,7 @@ extern void ceph_flush_snaps(struct ceph_inode_info *ci, extern bool __ceph_should_report_size(struct ceph_inode_info *ci); extern void ceph_check_caps(struct ceph_inode_info *ci, int flags, struct ceph_mds_session *session); -extern void ceph_check_delayed_caps(struct ceph_mds_client *mdsc); +extern unsigned long ceph_check_delayed_caps(struct ceph_mds_client *mdsc); extern void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc); extern int ceph_drop_caps_for_unlink(struct inode *inode); extern int ceph_encode_inode_release(void **p, struct inode *inode, -- cgit v1.2.3 From 8434ffe71c874b9c4e184b88d25de98c2bf5fe3f Mon Sep 17 00:00:00 2001 From: Jeff Layton Date: Tue, 3 Aug 2021 12:47:34 -0400 Subject: ceph: take snap_empty_lock atomically with snaprealm refcount change There is a race in ceph_put_snap_realm. The change to the nref and the spinlock acquisition are not done atomically, so you could decrement nref, and before you take the spinlock, the nref is incremented again. At that point, you end up putting it on the empty list when it shouldn't be there. Eventually __cleanup_empty_realms runs and frees it when it's still in-use. Fix this by protecting the 1->0 transition with atomic_dec_and_lock, and just drop the spinlock if we can get the rwsem. Because these objects can also undergo a 0->1 refcount transition, we must protect that change as well with the spinlock. Increment locklessly unless the value is at 0, in which case we take the spinlock, increment and then take it off the empty list if it did the 0->1 transition. With these changes, I'm removing the dout() messages from these functions, as well as in __put_snap_realm. They've always been racy, and it's better to not print values that may be misleading. Cc: stable@vger.kernel.org URL: https://tracker.ceph.com/issues/46419 Reported-by: Mark Nelson Signed-off-by: Jeff Layton Reviewed-by: Luis Henriques Signed-off-by: Ilya Dryomov --- fs/ceph/snap.c | 34 +++++++++++++++++----------------- 1 file changed, 17 insertions(+), 17 deletions(-) (limited to 'fs') diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c index 4ac0606dcbd4..4c6bd1042c94 100644 --- a/fs/ceph/snap.c +++ b/fs/ceph/snap.c @@ -67,19 +67,19 @@ void ceph_get_snap_realm(struct ceph_mds_client *mdsc, { lockdep_assert_held(&mdsc->snap_rwsem); - dout("get_realm %p %d -> %d\n", realm, - atomic_read(&realm->nref), atomic_read(&realm->nref)+1); /* - * since we _only_ increment realm refs or empty the empty - * list with snap_rwsem held, adjusting the empty list here is - * safe. we do need to protect against concurrent empty list - * additions, however. + * The 0->1 and 1->0 transitions must take the snap_empty_lock + * atomically with the refcount change. Go ahead and bump the + * nref here, unless it's 0, in which case we take the spinlock + * and then do the increment and remove it from the list. */ - if (atomic_inc_return(&realm->nref) == 1) { - spin_lock(&mdsc->snap_empty_lock); + if (atomic_inc_not_zero(&realm->nref)) + return; + + spin_lock(&mdsc->snap_empty_lock); + if (atomic_inc_return(&realm->nref) == 1) list_del_init(&realm->empty_item); - spin_unlock(&mdsc->snap_empty_lock); - } + spin_unlock(&mdsc->snap_empty_lock); } static void __insert_snap_realm(struct rb_root *root, @@ -208,28 +208,28 @@ static void __put_snap_realm(struct ceph_mds_client *mdsc, { lockdep_assert_held_write(&mdsc->snap_rwsem); - dout("__put_snap_realm %llx %p %d -> %d\n", realm->ino, realm, - atomic_read(&realm->nref), atomic_read(&realm->nref)-1); + /* + * We do not require the snap_empty_lock here, as any caller that + * increments the value must hold the snap_rwsem. + */ if (atomic_dec_and_test(&realm->nref)) __destroy_snap_realm(mdsc, realm); } /* - * caller needn't hold any locks + * See comments in ceph_get_snap_realm. Caller needn't hold any locks. */ void ceph_put_snap_realm(struct ceph_mds_client *mdsc, struct ceph_snap_realm *realm) { - dout("put_snap_realm %llx %p %d -> %d\n", realm->ino, realm, - atomic_read(&realm->nref), atomic_read(&realm->nref)-1); - if (!atomic_dec_and_test(&realm->nref)) + if (!atomic_dec_and_lock(&realm->nref, &mdsc->snap_empty_lock)) return; if (down_write_trylock(&mdsc->snap_rwsem)) { + spin_unlock(&mdsc->snap_empty_lock); __destroy_snap_realm(mdsc, realm); up_write(&mdsc->snap_rwsem); } else { - spin_lock(&mdsc->snap_empty_lock); list_add(&realm->empty_item, &mdsc->snap_empty); spin_unlock(&mdsc->snap_empty_lock); } -- cgit v1.2.3