From: Jeff Layton <jlayton@kernel.org>
Date: Wed, 1 Apr 2020 17:07:52 -0400
Subject: ceph: convert mdsc->cap_dirty to a per-session list
Git-commit: 1cf03a68e791b1673bc4daaa88a0820f34f538f8
Patch-mainline: v5.8-rc1
References: bsc#1167104
This is a per-sb list now, but that makes it difficult to tell when
the cap is the last dirty one associated with the session. Switch
this to be a per-session list, but continue using the
mdsc->cap_dirty_lock to protect the lists.
This list is only ever walked in ceph_flush_dirty_caps, so change that
to walk the sessions array and then flush the caps for inodes on each
session's list.
If the auth cap ever changes while the inode has dirty caps, then
move the inode to the appropriate session for the new auth_cap. Also,
ensure that we never remove an auth cap while the inode is still on the
s_cap_dirty list.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Acked-by: Luis Henriques <lhenriques@suse.com>
---
fs/ceph/caps.c | 78 ++++++++++++++++++++++++++++++++++++++++++---------
fs/ceph/mds_client.c | 2 -
fs/ceph/mds_client.h | 5 +--
fs/ceph/super.h | 16 +++++++++-
4 files changed, 84 insertions(+), 17 deletions(-)
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -603,6 +603,27 @@ static void __check_cap_issue(struct cep
}
}
+/**
+ * change_auth_cap_ses - move inode to appropriate lists when auth caps change
+ * @ci: inode to be moved
+ * @session: new auth caps session
+ */
+static void change_auth_cap_ses(struct ceph_inode_info *ci,
+ struct ceph_mds_session *session)
+{
+ lockdep_assert_held(&ci->i_ceph_lock);
+
+ if (list_empty(&ci->i_dirty_item) && list_empty(&ci->i_flushing_item))
+ return;
+
+ spin_lock(&session->s_mdsc->cap_dirty_lock);
+ if (!list_empty(&ci->i_dirty_item))
+ list_move(&ci->i_dirty_item, &session->s_cap_dirty);
+ if (!list_empty(&ci->i_flushing_item))
+ list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
+ spin_unlock(&session->s_mdsc->cap_dirty_lock);
+}
+
/*
* Add a capability under the given MDS session.
*
@@ -729,6 +750,9 @@ void ceph_add_cap(struct inode *inode,
if (flags & CEPH_CAP_FLAG_AUTH) {
if (!ci->i_auth_cap ||
ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
+ if (ci->i_auth_cap &&
+ ci->i_auth_cap->session != cap->session)
+ change_auth_cap_ses(ci, cap->session);
ci->i_auth_cap = cap;
cap->mds_wanted = wanted;
}
@@ -1058,8 +1082,10 @@ void __ceph_remove_cap(struct ceph_cap *
/* remove from inode's cap rbtree, and clear auth cap */
rb_erase(&cap->ci_node, &ci->i_caps);
- if (ci->i_auth_cap == cap)
+ if (ci->i_auth_cap == cap) {
+ WARN_ON_ONCE(!list_empty(&ci->i_dirty_item));
ci->i_auth_cap = NULL;
+ }
/* remove from session list */
spin_lock(&session->s_cap_lock);
@@ -1632,6 +1658,8 @@ int __ceph_mark_dirty_caps(struct ceph_i
ceph_cap_string(was | mask));
ci->i_dirty_caps |= mask;
if (was == 0) {
+ struct ceph_mds_session *session = ci->i_auth_cap->session;
+
WARN_ON_ONCE(ci->i_prealloc_cap_flush);
swap(ci->i_prealloc_cap_flush, *pcf);
@@ -1644,7 +1672,7 @@ int __ceph_mark_dirty_caps(struct ceph_i
&ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
BUG_ON(!list_empty(&ci->i_dirty_item));
spin_lock(&mdsc->cap_dirty_lock);
- list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
+ list_add(&ci->i_dirty_item, &session->s_cap_dirty);
spin_unlock(&mdsc->cap_dirty_lock);
if (ci->i_flushing_caps == 0) {
ihold(inode);
@@ -3596,15 +3624,9 @@ retry:
tcap->issue_seq = t_seq - 1;
tcap->issued |= issued;
tcap->implemented |= issued;
- if (cap == ci->i_auth_cap)
+ if (cap == ci->i_auth_cap) {
ci->i_auth_cap = tcap;
-
- if (!list_empty(&ci->i_cap_flush_list) &&
- ci->i_auth_cap == tcap) {
- spin_lock(&mdsc->cap_dirty_lock);
- list_move_tail(&ci->i_flushing_item,
- &tcap->session->s_cap_flushing);
- spin_unlock(&mdsc->cap_dirty_lock);
+ change_auth_cap_ses(ci, tcap->session);
}
}
__ceph_remove_cap(cap, false);
@@ -4025,15 +4047,16 @@ void ceph_check_delayed_caps(struct ceph
/*
* Flush all dirty caps to the mds
*/
-void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
+static void flush_dirty_session_caps(struct ceph_mds_session *s)
{
+ struct ceph_mds_client *mdsc = s->s_mdsc;
struct ceph_inode_info *ci;
struct inode *inode;
dout("flush_dirty_caps\n");
spin_lock(&mdsc->cap_dirty_lock);
- while (!list_empty(&mdsc->cap_dirty)) {
- ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
+ while (!list_empty(&s->s_cap_dirty)) {
+ ci = list_first_entry(&s->s_cap_dirty, struct ceph_inode_info,
i_dirty_item);
inode = &ci->vfs_inode;
ihold(inode);
@@ -4047,6 +4070,35 @@ void ceph_flush_dirty_caps(struct ceph_m
dout("flush_dirty_caps done\n");
}
+static void iterate_sessions(struct ceph_mds_client *mdsc,
+ void (*cb)(struct ceph_mds_session *))
+{
+ int mds;
+
+ mutex_lock(&mdsc->mutex);
+ for (mds = 0; mds < mdsc->max_sessions; ++mds) {
+ struct ceph_mds_session *s;
+
+ if (!mdsc->sessions[mds])
+ continue;
+
+ s = ceph_get_mds_session(mdsc->sessions[mds]);
+ if (!s)
+ continue;
+
+ mutex_unlock(&mdsc->mutex);
+ cb(s);
+ ceph_put_mds_session(s);
+ mutex_lock(&mdsc->mutex);
+ }
+ mutex_unlock(&mdsc->mutex);
+}
+
+void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
+{
+ iterate_sessions(mdsc, flush_dirty_session_caps);
+}
+
void __ceph_get_fmode(struct ceph_inode_info *ci, int fmode)
{
int i;
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -647,6 +647,7 @@ static struct ceph_mds_session *register
s->s_cap_reconnect = 0;
s->s_cap_iterator = NULL;
INIT_LIST_HEAD(&s->s_cap_releases);
+ INIT_LIST_HEAD(&s->s_cap_dirty);
INIT_LIST_HEAD(&s->s_cap_flushing);
mdsc->sessions[mds] = s;
@@ -3893,7 +3894,6 @@ int ceph_mdsc_init(struct ceph_fs_client
spin_lock_init(&mdsc->snap_flush_lock);
mdsc->last_cap_flush_tid = 1;
INIT_LIST_HEAD(&mdsc->cap_flush_list);
- INIT_LIST_HEAD(&mdsc->cap_dirty);
INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
mdsc->num_cap_flushing = 0;
spin_lock_init(&mdsc->cap_dirty_lock);
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -179,8 +179,10 @@ struct ceph_mds_session {
struct list_head s_cap_releases; /* waiting cap_release messages */
struct ceph_cap *s_cap_iterator;
- /* protected by mutex */
+ /* both protected by s_mdsc->cap_dirty_lock */
+ struct list_head s_cap_dirty; /* inodes w/ dirty caps */
struct list_head s_cap_flushing; /* inodes w/ flushing caps */
+
unsigned long s_renew_requested; /* last time we sent a renew req */
u64 s_renew_seq;
@@ -379,7 +381,6 @@ struct ceph_mds_client {
u64 last_cap_flush_tid;
struct list_head cap_flush_list;
- struct list_head cap_dirty; /* inodes with dirty caps */
struct list_head cap_dirty_migrating; /* ...that are migration... */
int num_cap_flushing; /* # caps we are flushing */
spinlock_t cap_dirty_lock; /* protects above items */
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -328,7 +328,21 @@ struct ceph_inode_info {
struct rb_root i_caps; /* cap list */
struct ceph_cap *i_auth_cap; /* authoritative cap, if any */
unsigned i_dirty_caps, i_flushing_caps; /* mask of dirtied fields */
- struct list_head i_dirty_item, i_flushing_item;
+ /*
+ * Link to the the auth cap's session's s_cap_dirty list. s_cap_dirty
+ * is protected by the mdsc->cap_dirty_lock, but each individual item
+ * is also protected by the inode's i_ceph_lock. Walking s_cap_dirty
+ * requires the mdsc->cap_dirty_lock. List presence for an item can
+ * be tested under the i_ceph_lock. Changing anything requires both.
+ */
+ struct list_head i_dirty_item;
+
+ /*
+ * Link to session's s_cap_flushing list. Protected by
+ * mdsc->cap_dirty_lock.
+ */
+ struct list_head i_flushing_item;
+
/* we need to track cap writeback on a per-cap-bit basis, to allow
* overlapping, pipelined cap flushes to the mds. we can probably
* reduce the tid to 8 bits if we're concerned about inode size. */