From 15a39f6a151b4e37e8c482c76bdb4240d9b559a3 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Thu, 27 Mar 2008 18:46:54 -0700
Subject: [PATCH] kclient: drop dn lease before unlink/rmdir

---
 src/TODO                |  8 ++++----
 src/kernel/dir.c        | 13 +++++++++----
 src/kernel/inode.c      | 12 ++++++++----
 src/kernel/mds_client.c | 29 ++++++++++++++++++++++++++++-
 src/kernel/mds_client.h |  3 +++
 src/kernel/super.h      |  4 +++-
 6 files changed, 55 insertions(+), 14 deletions(-)

diff --git a/src/TODO b/src/TODO
index bb279f89f1c6a..5c6c3c3a4ca70 100644
--- a/src/TODO
+++ b/src/TODO
@@ -4,8 +4,6 @@ code cleanup
   - probably kill base case in encoder.h, replace with int types, with appropriate swabbing?
 - addr=?
 
-- switch to nsec in ceph_timespec
-
 client leases
 - lease length heuristics
   - mds lock last_change stamp?
@@ -18,6 +16,8 @@ client leases
 - lease renewals (to avoid blocking client ops?)
 
 userspace client
+- pay attention to dn lease mds
+- pay attention to inode lease mds
 - clean up ll_ interface, now that we have leases!
 - obey file_max
 - revoke own caps when they time out,
@@ -28,8 +28,8 @@ userspace client
 - reference count lease validations on path lookup?
 
 kernel client
-- strip i_wr_size etc out of ceph_inode_info
-  - take mds provided versions when appropriate
+- pay attention to dn lease mds
+- pay attention to inode lease mds
 - carry wrbuffer/rdcache caps until data is flushed
   - this should make the utimes bit kick in
 - make sure link/unlink results reflected by inode/dentry cache  (let fill_trace do it?  invalidate?  do actual update?)
diff --git a/src/kernel/dir.c b/src/kernel/dir.c
index 1f441932cb7d2..5af1695dbd84e 100644
--- a/src/kernel/dir.c
+++ b/src/kernel/dir.c
@@ -97,7 +97,7 @@ static unsigned fpos_off(loff_t p)
 
 static int prepopulate_dir(struct dentry *parent,
 			   struct ceph_mds_reply_info *rinfo,
-			   int from_time)
+			   int from_mds, unsigned long from_time)
 {
 	struct qstr dname;
 	struct dentry *dn;
@@ -122,7 +122,8 @@ static int prepopulate_dir(struct dentry *parent,
 			}
 			ceph_init_dentry(dn);
 		}
-		ceph_update_dentry_lease(dn, rinfo->dir_dlease[i], from_time);
+		ceph_update_dentry_lease(dn, rinfo->dir_dlease[i], 
+					 from_mds, from_time);
 
 		/* inode */
 		if (dn->d_inode == NULL) {
@@ -204,6 +205,7 @@ nextfrag:
 
 		/* pre-populate dentry cache */
 		prepopulate_dir(filp->f_dentry, &req->r_reply_info, 
+				le32_to_cpu(req->r_reply->hdr.src.name.num),
 				req->r_from_time);
 	}
 
@@ -504,6 +506,7 @@ static int ceph_dir_unlink(struct inode *dir, struct dentry *dentry)
 	kfree(path);
 	if (IS_ERR(req))
 		return PTR_ERR(req);
+	ceph_mdsc_lease_release(mdsc, dentry);
 	err = ceph_mdsc_do_request(mdsc, req);
 	ceph_mdsc_put_request(req);
 
@@ -550,8 +553,10 @@ static int ceph_dir_rename(struct inode *old_dir, struct dentry *old_dentry,
 
 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
 {
-	dout(20, "d_revalidate ttl %lu now %lu\n", dentry->d_time, jiffies);
-	if (time_after(jiffies, dentry->d_time)) {
+	int mds = (long)dentry->d_fsdata;
+	dout(20, "d_revalidate ttl %lu mds %d now %lu\n", dentry->d_time, 
+	     mds, jiffies);
+	if (mds >= 0 && time_after(jiffies, dentry->d_time)) {
 		dout(20, "d_revalidate - dentry %p expired\n", dentry);
 		d_drop(dentry);
 		return 0;
diff --git a/src/kernel/inode.c b/src/kernel/inode.c
index 4e678b5a4a74c..7de74b6688231 100644
--- a/src/kernel/inode.c
+++ b/src/kernel/inode.c
@@ -173,6 +173,7 @@ void ceph_update_inode_lease(struct inode *inode,
 
 void ceph_update_dentry_lease(struct dentry *dentry, 
 			      struct ceph_mds_reply_lease *lease,
+			      int from_mds,
 			      unsigned long from_time) 
 {
 	__u64 ttl = le32_to_cpu(lease->duration_ms) * HZ;
@@ -184,14 +185,17 @@ void ceph_update_dentry_lease(struct dentry *dentry,
 	     dentry, le16_to_cpu(lease->mask), le32_to_cpu(lease->duration_ms),
 	     ttl);
 	if (lease->mask) {
-		if (ttl > dentry->d_time)
+		if (ttl > dentry->d_time) {
 			dentry->d_time = ttl;
+			dentry->d_fsdata = (void *)(long)from_mds;
+		}
 	} else {
-		dentry->d_time = 0;  /* invalidate */
+		dentry->d_fsdata = (void *)(long)-1;  /* invalidate */
 	}
 }
 
-int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
+int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 
+		    int mds)
 {
 	struct ceph_mds_reply_info *rinfo = &req->r_reply_info;
 	int err = 0;
@@ -264,7 +268,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 			}
 		}
 		ceph_update_dentry_lease(dn, rinfo->trace_dlease[d], 
-					 req->r_from_time);
+					 mds, req->r_from_time);
 
 		/* inode */
 		if (d+1 == rinfo->trace_numi) {
diff --git a/src/kernel/mds_client.c b/src/kernel/mds_client.c
index 83bd90f8e724d..f8b5e9c572c25 100644
--- a/src/kernel/mds_client.c
+++ b/src/kernel/mds_client.c
@@ -892,7 +892,7 @@ void ceph_mdsc_handle_reply(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
 	dout(10, "handle_reply tid %lld result %d\n", tid, result);
 
 	/* insert trace into our cache */
-	err = ceph_fill_trace(mdsc->client->sb, req);
+	err = ceph_fill_trace(mdsc->client->sb, req, mds);
 	if (err)
 		goto done;
 	if (result == 0 && req->r_expects_cap) {
@@ -1446,6 +1446,33 @@ bad:
 }
 
 
+void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct dentry *dn)
+{
+	struct ceph_msg *msg;
+	struct ceph_mds_lease *lease;
+	int mds = (long)dn->d_fsdata;
+	
+	if (mds < 0) {
+		dout(10, "lease_release dentry %p -- no lease\n", dn);
+		return;
+	}
+	dout(10, "lease_release dentry %p to mds%d\n", dn, mds);
+
+	msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, sizeof(*lease) + 
+			   sizeof(__u32) + dn->d_name.len, 0, 0, 0);
+	if (IS_ERR(msg))
+		return;
+	lease = msg->front.iov_base;
+	lease->action = CEPH_MDS_LEASE_RELEASE;
+	lease->mask = CEPH_LOCK_DN;
+	lease->ino = cpu_to_le64(dn->d_parent->d_inode->i_ino); /* ?? */
+	*(__le32*)(lease+1) = cpu_to_le32(dn->d_name.len);
+	memcpy((void *)(lease + 1) + 4, dn->d_name.name, dn->d_name.len);
+
+	send_msg_mds(mdsc, msg, mds);
+}
+
+
 /*
  * delayed work -- renew caps with mds
  */
diff --git a/src/kernel/mds_client.h b/src/kernel/mds_client.h
index c04d03174c316..5091006c4cafc 100644
--- a/src/kernel/mds_client.h
+++ b/src/kernel/mds_client.h
@@ -130,6 +130,9 @@ extern int ceph_mdsc_update_cap_wanted(struct ceph_inode_info *ci, int wanted);
 extern void ceph_mdsc_handle_lease(struct ceph_mds_client *mdsc,
 				   struct ceph_msg *msg);
 
+extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, 
+				    struct dentry *dn);
+
 extern struct ceph_mds_request *
 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op,
 			 ceph_ino_t ino1, const char *path1,
diff --git a/src/kernel/super.h b/src/kernel/super.h
index 8e09082547882..34741ee0de3db 100644
--- a/src/kernel/super.h
+++ b/src/kernel/super.h
@@ -322,6 +322,7 @@ extern void ceph_update_inode_lease(struct inode *inode,
 				    unsigned long from_time);
 extern void ceph_update_dentry_lease(struct dentry *dentry, 
 				     struct ceph_mds_reply_lease *lease,
+				     int from_mds,
 				     unsigned long from_time);
 
 extern struct ceph_inode_cap *ceph_find_cap(struct inode *inode, int want);
@@ -363,7 +364,8 @@ extern const struct file_operations ceph_dir_fops;
 extern struct dentry_operations ceph_dentry_ops;
 
 extern char *ceph_build_dentry_path(struct dentry *dentry, int *len);
-extern int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req);
+extern int ceph_fill_trace(struct super_block *sb, 
+			   struct ceph_mds_request *req, int mds);
 extern int ceph_do_lookup(struct super_block *sb, struct dentry *dentry, int m);
 
 static inline void ceph_init_dentry(struct dentry *dentry) {
-- 
2.39.5