Goldwyn Rodrigues
2013-Sep-27 17:06 UTC
[Ocfs2-devel] [PATCH 1/7] Add dlm operations placeholders
Try locking with the new DLM semantics with the dlm operations.
If operation fails with EOPNOTSUPP, try the old semantics.
This would ensure that filesystems are mounted for those who
are upgrading the kernel but not the tools.
Users using old tools will be presented with a notice message
to upgrade the tools.
recover_prep() is called when DLM understands a node is down.
recover_slot() is called once all nodes have acknowledged recover_prep and
recovery can begin.
recover_done() is called once the recovery is complete. It returns the
new membership.
---
fs/ocfs2/stack_user.c | 70 +++++++++++++++++++++++++++++++++++++--------------
1 file changed, 51 insertions(+), 19 deletions(-)
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
index 286edf1..b44490b 100644
--- a/fs/ocfs2/stack_user.c
+++ b/fs/ocfs2/stack_user.c
@@ -799,14 +799,63 @@ static int fs_protocol_compare(struct
ocfs2_protocol_version *existing,
return 0;
}
+static void user_recover_prep(void *arg)
+{
+}
+
+static void user_recover_slot(void *arg, struct dlm_slot *slot)
+{
+}
+
+static void user_recover_done(void *arg, struct dlm_slot *slots,
+ int num_slots, int our_slot,
+ uint32_t generation)
+{
+}
+
+const struct dlm_lockspace_ops ocfs2_ls_ops = {
+ .recover_prep = user_recover_prep,
+ .recover_slot = user_recover_slot,
+ .recover_done = user_recover_done,
+};
+
+static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
+{
+ dlm_release_lockspace(conn->cc_lockspace, 2);
+ conn->cc_lockspace = NULL;
+ ocfs2_live_connection_drop(conn->cc_private);
+ conn->cc_private = NULL;
+ return 0;
+}
+
static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
{
dlm_lockspace_t *fsdlm;
struct ocfs2_live_connection *uninitialized_var(control);
- int rc = 0;
+ int rc = 0, ops_rv;
BUG_ON(conn == NULL);
+ rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN,
+ &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
+
+ if (rc)
+ goto out;
+
+ if (!ops_rv) {
+ /* DLM lockspace creation with newer new_lockspace successful */
+ goto out;
+ } else if (ops_rv != -EOPNOTSUPP) {
+ rc = ops_rv;
+ goto out;
+ }
+
+ conn->cc_lockspace = fsdlm;
+
+ printk(KERN_NOTICE "ocfs2: You seem to be using an older version "
+ "of dlm_controld and/or ocfs2-tools. Please consider "
+ "upgrading.\n");
+
rc = ocfs2_live_connection_new(conn, &control);
if (rc)
goto out;
@@ -823,32 +872,15 @@ static int user_cluster_connect(struct
ocfs2_cluster_connection *conn)
conn->cc_version.pv_major, conn->cc_version.pv_minor,
running_proto.pv_major, running_proto.pv_minor);
rc = -EPROTO;
- ocfs2_live_connection_drop(control);
- goto out;
- }
-
- rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN,
- NULL, NULL, NULL, &fsdlm);
- if (rc) {
- ocfs2_live_connection_drop(control);
+ user_cluster_disconnect(conn);
goto out;
}
conn->cc_private = control;
- conn->cc_lockspace = fsdlm;
out:
return rc;
}
-static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
-{
- dlm_release_lockspace(conn->cc_lockspace, 2);
- conn->cc_lockspace = NULL;
- ocfs2_live_connection_drop(conn->cc_private);
- conn->cc_private = NULL;
- return 0;
-}
-
static int user_cluster_this_node(unsigned int *this_node)
{
int rc;
--
1.8.1.4
--
Goldwyn
Joel Becker
2013-Sep-27 18:59 UTC
[Ocfs2-devel] [PATCH 1/7] Add dlm operations placeholders
On Fri, Sep 27, 2013 at 12:06:47PM -0500, Goldwyn Rodrigues wrote:> Try locking with the new DLM semantics with the dlm operations. > If operation fails with EOPNOTSUPP, try the old semantics. > This would ensure that filesystems are mounted for those who > are upgrading the kernel but not the tools. > Users using old tools will be presented with a notice message > to upgrade the tools. > > recover_prep() is called when DLM understands a node is down. > recover_slot() is called once all nodes have acknowledged recover_prep and > recovery can begin. > recover_done() is called once the recovery is complete. It returns the > new membership.Can recover_prep() pause? What does the cluster do when not all nodes acknowledge the prep? When they die during recover_slot()? Which node recovers the slot?> --- > fs/ocfs2/stack_user.c | 70 +++++++++++++++++++++++++++++++++++++-------------- > 1 file changed, 51 insertions(+), 19 deletions(-) > > diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c > index 286edf1..b44490b 100644 > --- a/fs/ocfs2/stack_user.c > +++ b/fs/ocfs2/stack_user.c > @@ -799,14 +799,63 @@ static int fs_protocol_compare(struct ocfs2_protocol_version *existing, > return 0; > } > > +static void user_recover_prep(void *arg) > +{ > +} > + > +static void user_recover_slot(void *arg, struct dlm_slot *slot) > +{ > +} > + > +static void user_recover_done(void *arg, struct dlm_slot *slots, > + int num_slots, int our_slot, > + uint32_t generation) > +{ > +} > + > +const struct dlm_lockspace_ops ocfs2_ls_ops = { > + .recover_prep = user_recover_prep, > + .recover_slot = user_recover_slot, > + .recover_done = user_recover_done, > +}; > + > +static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn) > +{ > + dlm_release_lockspace(conn->cc_lockspace, 2); > + conn->cc_lockspace = NULL; > + ocfs2_live_connection_drop(conn->cc_private); > + conn->cc_private = NULL; > + return 0; > +} > + > static int user_cluster_connect(struct ocfs2_cluster_connection *conn) > { > dlm_lockspace_t *fsdlm; > struct ocfs2_live_connection *uninitialized_var(control); > - int rc = 0; > + int rc = 0, ops_rv; > > BUG_ON(conn == NULL); > > + rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN, > + &ocfs2_ls_ops, conn, &ops_rv, &fsdlm); > + > + if (rc) > + goto out; > + > + if (!ops_rv) { > + /* DLM lockspace creation with newer new_lockspace successful */ > + goto out; > + } else if (ops_rv != -EOPNOTSUPP) { > + rc = ops_rv; > + goto out; > + } > + > + conn->cc_lockspace = fsdlm;This code returns from !ops_rv before setting cc_lockspace, yet tries to reference it out in user_cluster_disconnect(). That looks like a crashing bug. Also, it looks like you call the new form of dlm_new_lockspace() before the callbacks are even implemented. If I just apply this patch, the code does not work. Please reorder the patch so that ocfs2 works at each patch. Joel> + > + printk(KERN_NOTICE "ocfs2: You seem to be using an older version " > + "of dlm_controld and/or ocfs2-tools. Please consider " > + "upgrading.\n"); > + > rc = ocfs2_live_connection_new(conn, &control); > if (rc) > goto out; > @@ -823,32 +872,15 @@ static int user_cluster_connect(struct ocfs2_cluster_connection *conn) > conn->cc_version.pv_major, conn->cc_version.pv_minor, > running_proto.pv_major, running_proto.pv_minor); > rc = -EPROTO; > - ocfs2_live_connection_drop(control); > - goto out; > - } > - > - rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN, > - NULL, NULL, NULL, &fsdlm); > - if (rc) { > - ocfs2_live_connection_drop(control); > + user_cluster_disconnect(conn); > goto out; > } > > conn->cc_private = control; > - conn->cc_lockspace = fsdlm; > out: > return rc; > } > > -static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn) > -{ > - dlm_release_lockspace(conn->cc_lockspace, 2); > - conn->cc_lockspace = NULL; > - ocfs2_live_connection_drop(conn->cc_private); > - conn->cc_private = NULL; > - return 0; > -} > - > static int user_cluster_this_node(unsigned int *this_node) > { > int rc; > -- > 1.8.1.4 > > > -- > Goldwyn > > _______________________________________________ > Ocfs2-devel mailing list > Ocfs2-devel at oss.oracle.com > https://oss.oracle.com/mailman/listinfo/ocfs2-devel-- "But all my words come back to me In shades of mediocrity. Like emptiness in harmony I need someone to comfort me." http://www.jlbec.org/ jlbec at evilplan.org