Add support for JBD journal callbacks to update MDS last_committed value.

This needs the most recent kernel patch in order to work properly.  If
the kernel patch isn't applied, you will get a message like:

"no journal callback kernel patch, faking it..."

Likewise, ext2 has "fake" support for commit callbacks.

diff --git a/lustre/include/linux/lustre_mds.h b/lustre/include/linux/lustre_mds.h
index 4f9e3e3..1d27fe0 100644 (file)
--- a/lustre/include/linux/lustre_mds.h
+++ b/lustre/include/linux/lustre_mds.h
@@ -155,6 +155,7 @@ struct mds_fs_operations {
         void    (* fs_delete_inode)(struct inode *inode);
         void    (* cl_delete_inode)(struct inode *inode);
         int     (* fs_journal_data)(struct inode *inode, struct file *file);
+        int     (* fs_set_last_rcvd)(struct mds_obd *mds, void *handle);
 };
 
 #define MDS_FSOP_UNLINK         1
@@ -210,6 +211,11 @@ static inline ssize_t mds_fs_readpage(struct mds_obd *mds, struct file *file,
         return mds->mds_fsops->fs_readpage(file, buf, count, offset);
 }
 
+static inline int mds_fs_set_last_rcvd(struct mds_obd *mds, void *handle)
+{
+        return mds->mds_fsops->fs_set_last_rcvd(mds, handle);
+}
+
 static inline ssize_t mds_fs_journal_data(struct mds_obd *mds,
                                           struct inode *inode,
                                           struct file *file)

diff --git a/lustre/include/linux/obd.h b/lustre/include/linux/obd.h
index 14640fd..f48f608 100644 (file)
--- a/lustre/include/linux/obd.h
+++ b/lustre/include/linux/obd.h
@@ -79,6 +79,7 @@ struct mds_obd {
         struct address_space_operations *mds_aops;
         struct mds_fs_operations *mds_fsops;
         struct file *mds_rcvd_filp;
+        __u64 mds_last_committed;
         __u64 mds_last_rcvd;
         __u64 mds_mount_count;
         struct ll_fid mds_rootfid;

diff --git a/lustre/mds/mds_ext2.c b/lustre/mds/mds_ext2.c
index e8b1e91..ac835a2 100644 (file)
--- a/lustre/mds/mds_ext2.c
+++ b/lustre/mds/mds_ext2.c
@@ -83,7 +83,7 @@ static ssize_t mds_ext2_readpage(struct file *file, char *buf, size_t count,
 
 struct mds_fs_operations mds_ext2_fs_ops;
 
-void mds_ext2_delete_inode(struct inode *inode)
+static void mds_ext2_delete_inode(struct inode *inode)
 {
         if (S_ISREG(inode->i_mode))
                 mds_ext2_set_objid(inode, NULL, 0);
@@ -91,7 +91,15 @@ void mds_ext2_delete_inode(struct inode *inode)
         mds_ext2_fs_ops.cl_delete_inode(inode);
 }
 
-int mds_ext2_journal_data(struct inode *inode, struct file *filp)
+static int mds_ext2_set_last_rcvd(struct mds_obd *mds, void *handle)
+{
+        /* Bail for ext2 - can't tell when it is on disk anyways, sync? */
+        mds->mds_last_committed = mds->mds_last_rcvd;
+
+        return 0;
+}
+
+static int mds_ext2_journal_data(struct inode *inode, struct file *filp)
 {
         return 0;
 }
@@ -106,4 +114,5 @@ struct mds_fs_operations mds_ext2_fs_ops = {
         fs_delete_inode:mds_ext2_delete_inode,
         cl_delete_inode:clear_inode,
         fs_journal_data:mds_ext2_journal_data,
+        fs_set_last_rcvd:mds_ext2_set_last_rcvd,
 };

diff --git a/lustre/mds/mds_ext3.c b/lustre/mds/mds_ext3.c
index 5ff07a7..9c6bd38 100644 (file)
--- a/lustre/mds/mds_ext3.c
+++ b/lustre/mds/mds_ext3.c
@@ -140,12 +140,10 @@ static ssize_t mds_ext3_readpage(struct file *file, char *buf, size_t count,
 
 struct mds_fs_operations mds_ext3_fs_ops;
 
-void mds_ext3_delete_inode(struct inode *inode)
+static void mds_ext3_delete_inode(struct inode *inode)
 {
-        void *handle;
-
         if (S_ISREG(inode->i_mode)) {
-                handle = mds_ext3_start(inode, MDS_FSOP_UNLINK);
+                void *handle = mds_ext3_start(inode, MDS_FSOP_UNLINK);
 
                 if (IS_ERR(handle)) {
                         CERROR("unable to start transaction");
@@ -164,7 +162,54 @@ void mds_ext3_delete_inode(struct inode *inode)
                 mds_ext3_fs_ops.cl_delete_inode(inode);
 }
 
-int mds_ext3_journal_data(struct inode *inode, struct file *filp)
+struct mds_cb_data {
+        struct journal_callback cb_jcb;
+        struct mds_obd *cb_mds;
+        __u64 cb_last_rcvd;
+};
+
+static void mds_ext3_callback_func(void *cb_data)
+{
+        struct mds_cb_data *mcb = cb_data;
+
+        CDEBUG(D_EXT2, "got callback for last_rcvd: %Ld\n", mcb->cb_last_rcvd);
+        if (mcb->cb_last_rcvd > mcb->cb_mds->mds_last_committed)
+                mcb->cb_mds->mds_last_committed = mcb->cb_last_rcvd;
+
+        OBD_FREE(mcb, sizeof(*mcb));
+}
+
+static int mds_ext3_set_last_rcvd(struct mds_obd *mds, void *handle)
+{
+        struct mds_cb_data *mcb;
+
+        OBD_ALLOC(mcb, sizeof(*mcb));
+        if (!mcb)
+                RETURN(-ENOMEM);
+
+        mcb->cb_mds = mds;
+        mcb->cb_last_rcvd = mds->mds_last_rcvd;
+
+#ifdef HAVE_JOURNAL_CALLBACK
+        CDEBUG(D_EXT2, "set callback for last_rcvd: %Ld\n",
+               (unsigned long long)mcb->cb_last_rcvd);
+        journal_callback_set(handle, mds_ext3_callback_func, mcb);
+#else
+        {
+        static long next = 0;
+
+        if (time_after(jiffies, next)) {
+                CERROR("no journal callback kernel patch, faking it...\n");
+                next = jiffies + 300 * HZ;
+        }
+        }
+        mds_ext3_callback_func(mcb);
+#endif
+
+        return 0;
+}
+
+static int mds_ext3_journal_data(struct inode *inode, struct file *filp)
 {
         EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
 
@@ -181,4 +226,5 @@ struct mds_fs_operations mds_ext3_fs_ops = {
         fs_delete_inode:mds_ext3_delete_inode,
         cl_delete_inode:clear_inode,
         fs_journal_data:mds_ext3_journal_data,
+        fs_set_last_rcvd:mds_ext3_set_last_rcvd,
 };

diff --git a/lustre/mds/mds_reint.c b/lustre/mds/mds_reint.c
index a6bf9db..09cfdc9 100644 (file)
--- a/lustre/mds/mds_reint.c
+++ b/lustre/mds/mds_reint.c
@@ -36,7 +36,8 @@
 #include <linux/lustre_mds.h>
 #include <linux/obd_class.h>
 
-int mds_update_last_rcvd(struct mds_obd *mds, struct ptlrpc_request *req)
+int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
+                         struct ptlrpc_request *req)
 {
         /* get from req->rq_connection-> or req->rq_client */
         struct mds_client_info mci_data, *mci = &mci_data;
@@ -53,6 +54,8 @@ int mds_update_last_rcvd(struct mds_obd *mds, struct ptlrpc_request *req)
         mci->mci_mcd->mcd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
         mci->mci_mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
         mci->mci_mcd->mcd_xid = cpu_to_le32(req->rq_connection->c_xid_in);
+
+        mds_fs_set_last_rcvd(mds, handle);
         rc = lustre_fwrite(mds->mds_rcvd_filp, (char *)mci->mci_mcd,
                            sizeof(*mci->mci_mcd), &off);
         CDEBUG(D_INODE, "wrote trans #%Ld for client '%s' at %Ld: rc = %d\n",
@@ -91,7 +94,7 @@ static int mds_reint_setattr(struct mds_update_record *rec,
         rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
 
         if (!rc)
-                rc = mds_update_last_rcvd(mds, req);
+                rc = mds_update_last_rcvd(mds, handle, req);
 
         EXIT;
 
@@ -216,7 +219,7 @@ static int mds_reint_create(struct mds_update_record *rec,
         }
 
         if (!rc)
-                rc = mds_update_last_rcvd(mds, req);
+                rc = mds_update_last_rcvd(mds, handle, req);
 
 out_create_commit:
         /* FIXME: keep rc intact */
@@ -296,7 +299,7 @@ static int mds_reint_unlink(struct mds_update_record *rec,
         }
 
         if (!rc)
-                rc = mds_update_last_rcvd(mds, req);
+                rc = mds_update_last_rcvd(mds, handle, req);
         /* FIXME: keep rc intact */
         rc = mds_fs_commit(mds, dir, handle);
 
@@ -355,7 +358,7 @@ static int mds_reint_link(struct mds_update_record *rec,
         rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
 
         if (!rc)
-                rc = mds_update_last_rcvd(mds, req);
+                rc = mds_update_last_rcvd(mds, handle, req);
 
         /* FIXME: keep rc intact */
         rc = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
@@ -416,7 +419,7 @@ static int mds_reint_rename(struct mds_update_record *rec,
         rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
 
         if (!rc)
-                rc = mds_update_last_rcvd(mds, req);
+                rc = mds_update_last_rcvd(mds, handle, req);
 
         /* FIXME: keep rc intact */
         rc = mds_fs_commit(mds, de_tgtdir->d_inode, handle);