summaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlm
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ocfs2/dlm')
-rw-r--r--fs/ocfs2/dlm/dlmcommon.h21
-rw-r--r--fs/ocfs2/dlm/dlmconvert.c2
-rw-r--r--fs/ocfs2/dlm/dlmdomain.c103
-rw-r--r--fs/ocfs2/dlm/dlmmaster.c24
-rw-r--r--fs/ocfs2/dlm/dlmrecovery.c57
-rw-r--r--fs/ocfs2/dlm/dlmthread.c6
6 files changed, 148 insertions, 65 deletions
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 9843ee17ea2..dc8ea666efd 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -176,6 +176,7 @@ struct dlm_mig_lockres_priv
{
struct dlm_lock_resource *lockres;
u8 real_master;
+ u8 extra_ref;
};
struct dlm_assert_master_priv
@@ -602,17 +603,19 @@ enum dlm_query_join_response_code {
JOIN_PROTOCOL_MISMATCH,
};
+struct dlm_query_join_packet {
+ u8 code; /* Response code. dlm_minor and fs_minor
+ are only valid if this is JOIN_OK */
+ u8 dlm_minor; /* The minor version of the protocol the
+ dlm is speaking. */
+ u8 fs_minor; /* The minor version of the protocol the
+ filesystem is speaking. */
+ u8 reserved;
+};
+
union dlm_query_join_response {
u32 intval;
- struct {
- u8 code; /* Response code. dlm_minor and fs_minor
- are only valid if this is JOIN_OK */
- u8 dlm_minor; /* The minor version of the protocol the
- dlm is speaking. */
- u8 fs_minor; /* The minor version of the protocol the
- filesystem is speaking. */
- u8 reserved;
- } packet;
+ struct dlm_query_join_packet packet;
};
struct dlm_lock_request
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
index ecb4d997221..75997b4deaf 100644
--- a/fs/ocfs2/dlm/dlmconvert.c
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -487,7 +487,7 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
"cookie=%u:%llu\n",
dlm_get_lock_cookie_node(be64_to_cpu(cnv->cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cnv->cookie)));
- __dlm_print_one_lock_resource(res);
+ dlm_print_one_lock_resource(res);
goto leave;
}
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 638d2ebb892..0879d86113e 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -713,14 +713,46 @@ static int dlm_query_join_proto_check(char *proto_type, int node,
return rc;
}
+/*
+ * struct dlm_query_join_packet is made up of four one-byte fields. They
+ * are effectively in big-endian order already. However, little-endian
+ * machines swap them before putting the packet on the wire (because
+ * query_join's response is a status, and that status is treated as a u32
+ * on the wire). Thus, a big-endian and little-endian machines will treat
+ * this structure differently.
+ *
+ * The solution is to have little-endian machines swap the structure when
+ * converting from the structure to the u32 representation. This will
+ * result in the structure having the correct format on the wire no matter
+ * the host endian format.
+ */
+static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
+ u32 *wire)
+{
+ union dlm_query_join_response response;
+
+ response.packet = *packet;
+ *wire = cpu_to_be32(response.intval);
+}
+
+static void dlm_query_join_wire_to_packet(u32 wire,
+ struct dlm_query_join_packet *packet)
+{
+ union dlm_query_join_response response;
+
+ response.intval = cpu_to_be32(wire);
+ *packet = response.packet;
+}
+
static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
void **ret_data)
{
struct dlm_query_join_request *query;
- union dlm_query_join_response response = {
- .packet.code = JOIN_DISALLOW,
+ struct dlm_query_join_packet packet = {
+ .code = JOIN_DISALLOW,
};
struct dlm_ctxt *dlm = NULL;
+ u32 response;
u8 nodenum;
query = (struct dlm_query_join_request *) msg->buf;
@@ -737,11 +769,11 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
mlog(0, "node %u is not in our live map yet\n",
query->node_idx);
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
goto respond;
}
- response.packet.code = JOIN_OK_NO_MAP;
+ packet.code = JOIN_OK_NO_MAP;
spin_lock(&dlm_domain_lock);
dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
@@ -760,7 +792,7 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
mlog(0, "disallow join as node %u does not "
"have node %u in its nodemap\n",
query->node_idx, nodenum);
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
goto unlock_respond;
}
}
@@ -780,23 +812,23 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
/*If this is a brand new context and we
* haven't started our join process yet, then
* the other node won the race. */
- response.packet.code = JOIN_OK_NO_MAP;
+ packet.code = JOIN_OK_NO_MAP;
} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
/* Disallow parallel joins. */
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
mlog(0, "node %u trying to join, but recovery "
"is ongoing.\n", bit);
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
} else if (test_bit(bit, dlm->recovery_map)) {
mlog(0, "node %u trying to join, but it "
"still needs recovery.\n", bit);
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
} else if (test_bit(bit, dlm->domain_map)) {
mlog(0, "node %u trying to join, but it "
"is still in the domain! needs recovery?\n",
bit);
- response.packet.code = JOIN_DISALLOW;
+ packet.code = JOIN_DISALLOW;
} else {
/* Alright we're fully a part of this domain
* so we keep some state as to who's joining
@@ -807,19 +839,15 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
if (dlm_query_join_proto_check("DLM", bit,
&dlm->dlm_locking_proto,
&query->dlm_proto)) {
- response.packet.code =
- JOIN_PROTOCOL_MISMATCH;
+ packet.code = JOIN_PROTOCOL_MISMATCH;
} else if (dlm_query_join_proto_check("fs", bit,
&dlm->fs_locking_proto,
&query->fs_proto)) {
- response.packet.code =
- JOIN_PROTOCOL_MISMATCH;
+ packet.code = JOIN_PROTOCOL_MISMATCH;
} else {
- response.packet.dlm_minor =
- query->dlm_proto.pv_minor;
- response.packet.fs_minor =
- query->fs_proto.pv_minor;
- response.packet.code = JOIN_OK;
+ packet.dlm_minor = query->dlm_proto.pv_minor;
+ packet.fs_minor = query->fs_proto.pv_minor;
+ packet.code = JOIN_OK;
__dlm_set_joining_node(dlm, query->node_idx);
}
}
@@ -830,9 +858,10 @@ unlock_respond:
spin_unlock(&dlm_domain_lock);
respond:
- mlog(0, "We respond with %u\n", response.packet.code);
+ mlog(0, "We respond with %u\n", packet.code);
- return response.intval;
+ dlm_query_join_packet_to_wire(&packet, &response);
+ return response;
}
static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
@@ -937,7 +966,7 @@ static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
sizeof(unsigned long))) {
mlog(ML_ERROR,
"map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
- map_size, BITS_TO_LONGS(O2NM_MAX_NODES));
+ map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
return -EINVAL;
}
@@ -968,7 +997,8 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
{
int status;
struct dlm_query_join_request join_msg;
- union dlm_query_join_response join_resp;
+ struct dlm_query_join_packet packet;
+ u32 join_resp;
mlog(0, "querying node %d\n", node);
@@ -984,11 +1014,12 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
sizeof(join_msg), node,
- &join_resp.intval);
+ &join_resp);
if (status < 0 && status != -ENOPROTOOPT) {
mlog_errno(status);
goto bail;
}
+ dlm_query_join_wire_to_packet(join_resp, &packet);
/* -ENOPROTOOPT from the net code means the other side isn't
listening for our message type -- that's fine, it means
@@ -997,10 +1028,10 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
if (status == -ENOPROTOOPT) {
status = 0;
*response = JOIN_OK_NO_MAP;
- } else if (join_resp.packet.code == JOIN_DISALLOW ||
- join_resp.packet.code == JOIN_OK_NO_MAP) {
- *response = join_resp.packet.code;
- } else if (join_resp.packet.code == JOIN_PROTOCOL_MISMATCH) {
+ } else if (packet.code == JOIN_DISALLOW ||
+ packet.code == JOIN_OK_NO_MAP) {
+ *response = packet.code;
+ } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
mlog(ML_NOTICE,
"This node requested DLM locking protocol %u.%u and "
"filesystem locking protocol %u.%u. At least one of "
@@ -1012,14 +1043,12 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
dlm->fs_locking_proto.pv_minor,
node);
status = -EPROTO;
- *response = join_resp.packet.code;
- } else if (join_resp.packet.code == JOIN_OK) {
- *response = join_resp.packet.code;
+ *response = packet.code;
+ } else if (packet.code == JOIN_OK) {
+ *response = packet.code;
/* Use the same locking protocol as the remote node */
- dlm->dlm_locking_proto.pv_minor =
- join_resp.packet.dlm_minor;
- dlm->fs_locking_proto.pv_minor =
- join_resp.packet.fs_minor;
+ dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
+ dlm->fs_locking_proto.pv_minor = packet.fs_minor;
mlog(0,
"Node %d responds JOIN_OK with DLM locking protocol "
"%u.%u and fs locking protocol %u.%u\n",
@@ -1031,11 +1060,11 @@ static int dlm_request_join(struct dlm_ctxt *dlm,
} else {
status = -EINVAL;
mlog(ML_ERROR, "invalid response %d from node %u\n",
- join_resp.packet.code, node);
+ packet.code, node);
}
mlog(0, "status %d, node %d response is %d\n", status, node,
- *response);
+ *response);
bail:
return status;
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index a54d33d95ad..ea6b8957786 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -1663,7 +1663,12 @@ way_up_top:
dlm_put_mle(tmpmle);
}
send_response:
-
+ /*
+ * __dlm_lookup_lockres() grabbed a reference to this lockres.
+ * The reference is released by dlm_assert_master_worker() under
+ * the call to dlm_dispatch_assert_master(). If
+ * dlm_assert_master_worker() isn't called, we drop it here.
+ */
if (dispatch_assert) {
if (response != DLM_MASTER_RESP_YES)
mlog(ML_ERROR, "invalid response %d\n", response);
@@ -1678,7 +1683,11 @@ send_response:
if (ret < 0) {
mlog(ML_ERROR, "failed to dispatch assert master work\n");
response = DLM_MASTER_RESP_ERROR;
+ dlm_lockres_put(res);
}
+ } else {
+ if (res)
+ dlm_lockres_put(res);
}
dlm_put(dlm);
@@ -1695,9 +1704,9 @@ send_response:
* can periodically run all locks owned by this node
* and re-assert across the cluster...
*/
-int dlm_do_assert_master(struct dlm_ctxt *dlm,
- struct dlm_lock_resource *res,
- void *nodemap, u32 flags)
+static int dlm_do_assert_master(struct dlm_ctxt *dlm,
+ struct dlm_lock_resource *res,
+ void *nodemap, u32 flags)
{
struct dlm_assert_master assert;
int to, tmpret;
@@ -2348,7 +2357,7 @@ int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
"but it is already dropped!\n", dlm->name,
res->lockname.len, res->lockname.name, node);
- __dlm_print_one_lock_resource(res);
+ dlm_print_one_lock_resource(res);
}
ret = 0;
goto done;
@@ -2408,7 +2417,7 @@ static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data)
mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
"but it is already dropped!\n", dlm->name,
res->lockname.len, res->lockname.name, node);
- __dlm_print_one_lock_resource(res);
+ dlm_print_one_lock_resource(res);
}
dlm_lockres_put(res);
@@ -2933,6 +2942,9 @@ static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
dlm_lockres_clear_refmap_bit(lock->ml.node, res);
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* In a normal unlock, we would have added a
+ * DLM_UNLOCK_FREE_LOCK action. Force it. */
+ dlm_lock_put(lock);
}
}
queue++;
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 91f747b8a53..bcb9260c373 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -519,9 +519,9 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm)
return 0;
master_here:
- mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
- task_pid_nr(dlm->dlm_reco_thread_task),
- dlm->name, dlm->reco.dead_node, dlm->node_num);
+ mlog(ML_NOTICE, "(%d) Node %u is the Recovery Master for the Dead Node "
+ "%u for Domain %s\n", task_pid_nr(dlm->dlm_reco_thread_task),
+ dlm->node_num, dlm->reco.dead_node, dlm->name);
status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
if (status < 0) {
@@ -1191,7 +1191,7 @@ static int dlm_add_lock_to_array(struct dlm_lock *lock,
(ml->type == LKM_EXMODE ||
memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
mlog(ML_ERROR, "mismatched lvbs!\n");
- __dlm_print_one_lock_resource(lock->lockres);
+ dlm_print_one_lock_resource(lock->lockres);
BUG();
}
memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
@@ -1327,6 +1327,7 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
(struct dlm_migratable_lockres *)msg->buf;
int ret = 0;
u8 real_master;
+ u8 extra_refs = 0;
char *buf = NULL;
struct dlm_work_item *item = NULL;
struct dlm_lock_resource *res = NULL;
@@ -1404,16 +1405,28 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
__dlm_insert_lockres(dlm, res);
spin_unlock(&dlm->spinlock);
+ /* Add an extra ref for this lock-less lockres lest the
+ * dlm_thread purges it before we get the chance to add
+ * locks to it */
+ dlm_lockres_get(res);
+
+ /* There are three refs that need to be put.
+ * 1. Taken above.
+ * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
+ * 3. dlm_lookup_lockres()
+ * The first one is handled at the end of this function. The
+ * other two are handled in the worker thread after locks have
+ * been attached. Yes, we don't wait for purge time to match
+ * kref_init. The lockres will still have atleast one ref
+ * added because it is in the hash __dlm_insert_lockres() */
+ extra_refs++;
+
/* now that the new lockres is inserted,
* make it usable by other processes */
spin_lock(&res->spinlock);
res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
spin_unlock(&res->spinlock);
wake_up(&res->wq);
-
- /* add an extra ref for just-allocated lockres
- * otherwise the lockres will be purged immediately */
- dlm_lockres_get(res);
}
/* at this point we have allocated everything we need,
@@ -1443,12 +1456,17 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
item->u.ml.lockres = res; /* already have a ref */
item->u.ml.real_master = real_master;
+ item->u.ml.extra_ref = extra_refs;
spin_lock(&dlm->work_lock);
list_add_tail(&item->list, &dlm->work_list);
spin_unlock(&dlm->work_lock);
queue_work(dlm->dlm_worker, &dlm->dispatched_work);
leave:
+ /* One extra ref taken needs to be put here */
+ if (extra_refs)
+ dlm_lockres_put(res);
+
dlm_put(dlm);
if (ret < 0) {
if (buf)
@@ -1464,17 +1482,19 @@ leave:
static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
{
- struct dlm_ctxt *dlm = data;
+ struct dlm_ctxt *dlm;
struct dlm_migratable_lockres *mres;
int ret = 0;
struct dlm_lock_resource *res;
u8 real_master;
+ u8 extra_ref;
dlm = item->dlm;
mres = (struct dlm_migratable_lockres *)data;
res = item->u.ml.lockres;
real_master = item->u.ml.real_master;
+ extra_ref = item->u.ml.extra_ref;
if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
/* this case is super-rare. only occurs if
@@ -1517,6 +1537,12 @@ again:
}
leave:
+ /* See comment in dlm_mig_lockres_handler() */
+ if (res) {
+ if (extra_ref)
+ dlm_lockres_put(res);
+ dlm_lockres_put(res);
+ }
kfree(data);
mlog_exit(ret);
}
@@ -1644,7 +1670,8 @@ int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
/* retry!? */
BUG();
}
- }
+ } else /* put.. incase we are not the master */
+ dlm_lockres_put(res);
spin_unlock(&res->spinlock);
}
spin_unlock(&dlm->spinlock);
@@ -1921,6 +1948,7 @@ void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
"Recovering res %s:%.*s, is already on recovery list!\n",
dlm->name, res->lockname.len, res->lockname.name);
list_del_init(&res->recovering);
+ dlm_lockres_put(res);
}
/* We need to hold a reference while on the recovery list */
dlm_lockres_get(res);
@@ -2130,11 +2158,16 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
assert_spin_locked(&dlm->spinlock);
assert_spin_locked(&res->spinlock);
+ /* We do two dlm_lock_put(). One for removing from list and the other is
+ * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
+
/* TODO: check pending_asts, pending_basts here */
list_for_each_entry_safe(lock, next, &res->granted, list) {
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
@@ -2142,6 +2175,8 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
@@ -2149,6 +2184,8 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
index cebd089f895..4060bb328bc 100644
--- a/fs/ocfs2/dlm/dlmthread.c
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -176,12 +176,14 @@ static int dlm_purge_lockres(struct dlm_ctxt *dlm,
res->lockname.name, master);
if (!master) {
+ /* drop spinlock... retake below */
+ spin_unlock(&dlm->spinlock);
+
spin_lock(&res->spinlock);
/* This ensures that clear refmap is sent after the set */
__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
spin_unlock(&res->spinlock);
- /* drop spinlock to do messaging, retake below */
- spin_unlock(&dlm->spinlock);
+
/* clear our bit from the master's refmap, ignore errors */
ret = dlm_drop_lockres_ref(dlm, res);
if (ret < 0) {
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2024-12-29 20:04:19 +0000