From c8799fc4674fe5bb9b9391f9eac202250b8370e1 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Thu, 7 Jul 2016 15:22:38 +0800 Subject: ceph: optimize cap flush waiting Add a 'wake' flag to ceph_cap_flush struct, which indicates if there is someone waiting for it to finish. When getting flush ack message, we check the 'wake' flag in corresponding ceph_cap_flush struct to decide if we should wake up waiters. One corner case is that the acked cap flush has 'wake' flags is set, but it is not the first one on the flushing list. We do not wake up waiters in this case, set 'wake' flags of preceding ceph_cap_flush struct instead Signed-off-by: Yan, Zheng --- fs/ceph/caps.c | 91 ++++++++++++++++++++++++++++++++++++---------------- fs/ceph/mds_client.c | 8 +++++ fs/ceph/super.h | 1 + 3 files changed, 73 insertions(+), 27 deletions(-) (limited to 'fs/ceph') diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c index 736e1c86bcf3..99115cae1652 100644 --- a/fs/ceph/caps.c +++ b/fs/ceph/caps.c @@ -1473,6 +1473,37 @@ static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc) return 0; } +/* + * Remove cap_flush from the mdsc's or inode's flushing cap list. + * Return true if caller needs to wake up flush waiters. + */ +static bool __finish_cap_flush(struct ceph_mds_client *mdsc, + struct ceph_inode_info *ci, + struct ceph_cap_flush *cf) +{ + struct ceph_cap_flush *prev; + bool wake = cf->wake; + if (mdsc) { + /* are there older pending cap flushes? */ + if (wake && cf->g_list.prev != &mdsc->cap_flush_list) { + prev = list_prev_entry(cf, g_list); + prev->wake = true; + wake = false; + } + list_del(&cf->g_list); + } else if (ci) { + if (wake && cf->i_list.prev != &ci->i_cap_flush_list) { + prev = list_prev_entry(cf, i_list); + prev->wake = true; + wake = false; + } + list_del(&cf->i_list); + } else { + BUG_ON(1); + } + return wake; +} + /* * Add dirty inode to the flushing list. Assigned a seq number so we * can wait for caps to flush without starving. @@ -1480,7 +1511,7 @@ static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc) * Called under i_ceph_lock. */ static int __mark_caps_flushing(struct inode *inode, - struct ceph_mds_session *session, + struct ceph_mds_session *session, bool wake, u64 *flush_tid, u64 *oldest_flush_tid) { struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; @@ -1503,6 +1534,7 @@ static int __mark_caps_flushing(struct inode *inode, swap(cf, ci->i_prealloc_cap_flush); cf->caps = flushing; + cf->wake = wake; spin_lock(&mdsc->cap_dirty_lock); list_del_init(&ci->i_dirty_item); @@ -1808,7 +1840,7 @@ ack: } if (cap == ci->i_auth_cap && ci->i_dirty_caps) { - flushing = __mark_caps_flushing(inode, session, + flushing = __mark_caps_flushing(inode, session, false, &flush_tid, &oldest_flush_tid); } else { @@ -1885,8 +1917,8 @@ retry: if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) goto out; - flushing = __mark_caps_flushing(inode, session, &flush_tid, - &oldest_flush_tid); + flushing = __mark_caps_flushing(inode, session, true, + &flush_tid, &oldest_flush_tid); /* __send_cap drops i_ceph_lock */ delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want, @@ -1902,7 +1934,8 @@ retry: if (!list_empty(&ci->i_cap_flush_list)) { struct ceph_cap_flush *cf = list_last_entry(&ci->i_cap_flush_list, - struct ceph_cap_flush, i_list); + struct ceph_cap_flush, i_list); + cf->wake = true; flush_tid = cf->tid; } flushing = ci->i_flushing_caps; @@ -3022,7 +3055,9 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, unsigned seq = le32_to_cpu(m->seq); int dirty = le32_to_cpu(m->dirty); int cleaned = 0; - int drop = 0; + bool drop = false; + bool wake_ci = 0; + bool wake_mdsc = 0; list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) { if (cf->tid == flush_tid) @@ -3030,7 +3065,8 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, if (cf->caps == 0) /* capsnap */ continue; if (cf->tid <= flush_tid) { - list_del(&cf->i_list); + if (__finish_cap_flush(NULL, ci, cf)) + wake_ci = true; list_add_tail(&cf->i_list, &to_remove); } else { cleaned &= ~cf->caps; @@ -3052,14 +3088,9 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, spin_lock(&mdsc->cap_dirty_lock); - if (!list_empty(&to_remove)) { - u64 oldest_flush_tid; - list_for_each_entry(cf, &to_remove, i_list) - list_del(&cf->g_list); - - oldest_flush_tid = __get_oldest_flush_tid(mdsc); - if (oldest_flush_tid == 0 || oldest_flush_tid > flush_tid) - wake_up_all(&mdsc->cap_flushing_wq); + list_for_each_entry(cf, &to_remove, i_list) { + if (__finish_cap_flush(mdsc, NULL, cf)) + wake_mdsc = true; } if (ci->i_flushing_caps == 0) { @@ -3079,7 +3110,7 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, if (ci->i_dirty_caps == 0) { dout(" inode %p now clean\n", inode); BUG_ON(!list_empty(&ci->i_dirty_item)); - drop = 1; + drop = true; if (ci->i_wr_ref == 0 && ci->i_wrbuffer_ref_head == 0) { BUG_ON(!ci->i_head_snapc); @@ -3091,7 +3122,6 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, } } spin_unlock(&mdsc->cap_dirty_lock); - wake_up_all(&ci->i_cap_wq); out: spin_unlock(&ci->i_ceph_lock); @@ -3102,6 +3132,11 @@ out: list_del(&cf->i_list); ceph_free_cap_flush(cf); } + + if (wake_ci) + wake_up_all(&ci->i_cap_wq); + if (wake_mdsc) + wake_up_all(&mdsc->cap_flushing_wq); if (drop) iput(inode); } @@ -3120,7 +3155,9 @@ static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; u64 follows = le64_to_cpu(m->snap_follows); struct ceph_cap_snap *capsnap; - int flushed = 0; + bool flushed = false; + bool wake_ci = false; + bool wake_mdsc = false; dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", inode, ci, session->s_mds, follows); @@ -3134,7 +3171,7 @@ static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, flush_tid, capsnap->cap_flush.tid); break; } - flushed = 1; + flushed = true; break; } else { dout(" skipping cap_snap %p follows %lld\n", @@ -3142,31 +3179,31 @@ static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, } } if (flushed) { - u64 oldest_flush_tid; WARN_ON(capsnap->dirty_pages || capsnap->writing); dout(" removing %p cap_snap %p follows %lld\n", inode, capsnap, follows); list_del(&capsnap->ci_item); - list_del(&capsnap->cap_flush.i_list); + if (__finish_cap_flush(NULL, ci, &capsnap->cap_flush)) + wake_ci = true; spin_lock(&mdsc->cap_dirty_lock); if (list_empty(&ci->i_cap_flush_list)) list_del_init(&ci->i_flushing_item); - list_del(&capsnap->cap_flush.g_list); - - oldest_flush_tid = __get_oldest_flush_tid(mdsc); - if (oldest_flush_tid == 0 || oldest_flush_tid > flush_tid) - wake_up_all(&mdsc->cap_flushing_wq); + if (__finish_cap_flush(mdsc, NULL, &capsnap->cap_flush)) + wake_mdsc = true; spin_unlock(&mdsc->cap_dirty_lock); - wake_up_all(&ci->i_cap_wq); } spin_unlock(&ci->i_ceph_lock); if (flushed) { ceph_put_snap_context(capsnap->context); ceph_put_cap_snap(capsnap); + if (wake_ci) + wake_up_all(&ci->i_cap_wq); + if (wake_mdsc) + wake_up_all(&mdsc->cap_flushing_wq); iput(inode); } } diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index fa9036af5445..cdc6a17f5867 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -1212,6 +1212,8 @@ static void remove_session_caps(struct ceph_mds_session *session) dout("remove_session_caps on %p\n", session); iterate_session_caps(session, remove_session_caps_cb, fsc); + wake_up_all(&fsc->mdsc->cap_flushing_wq); + spin_lock(&session->s_cap_lock); if (session->s_nr_caps > 0) { struct inode *inode; @@ -3536,6 +3538,12 @@ void ceph_mdsc_sync(struct ceph_mds_client *mdsc) ceph_flush_dirty_caps(mdsc); spin_lock(&mdsc->cap_dirty_lock); want_flush = mdsc->last_cap_flush_tid; + if (!list_empty(&mdsc->cap_flush_list)) { + struct ceph_cap_flush *cf = + list_last_entry(&mdsc->cap_flush_list, + struct ceph_cap_flush, g_list); + cf->wake = true; + } spin_unlock(&mdsc->cap_dirty_lock); dout("sync want tid %lld flush_seq %lld\n", diff --git a/fs/ceph/super.h b/fs/ceph/super.h index b097d474f888..3e3fa9163059 100644 --- a/fs/ceph/super.h +++ b/fs/ceph/super.h @@ -150,6 +150,7 @@ struct ceph_cap { struct ceph_cap_flush { u64 tid; int caps; /* 0 means capsnap */ + bool wake; /* wake up flush waiters when finish ? */ struct list_head g_list; // global struct list_head i_list; // per inode }; -- cgit v1.2.3