summaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlm
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ocfs2/dlm')
-rw-r--r--fs/ocfs2/dlm/dlmast.c22
-rw-r--r--fs/ocfs2/dlm/dlmcommon.h21
-rw-r--r--fs/ocfs2/dlm/dlmconvert.c11
-rw-r--r--fs/ocfs2/dlm/dlmdebug.c18
-rw-r--r--fs/ocfs2/dlm/dlmfs.c3
-rw-r--r--fs/ocfs2/dlm/dlmlock.c14
-rw-r--r--fs/ocfs2/dlm/dlmmaster.c227
-rw-r--r--fs/ocfs2/dlm/dlmrecovery.c50
-rw-r--r--fs/ocfs2/dlm/dlmunlock.c11
9 files changed, 300 insertions, 77 deletions
diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c
index 8d17d28ef91c..355593dd8ef8 100644
--- a/fs/ocfs2/dlm/dlmast.c
+++ b/fs/ocfs2/dlm/dlmast.c
@@ -307,8 +307,11 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data)
if (past->type != DLM_AST &&
past->type != DLM_BAST) {
- mlog(ML_ERROR, "Unknown ast type! %d, cookie=%"MLFu64", "
- "name=%.*s\n", past->type, cookie, locklen, name);
+ mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu"
+ "name=%.*s\n", past->type,
+ dlm_get_lock_cookie_node(cookie),
+ dlm_get_lock_cookie_seq(cookie),
+ locklen, name);
ret = DLM_IVLOCKID;
goto leave;
}
@@ -316,9 +319,11 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data)
res = dlm_lookup_lockres(dlm, name, locklen);
if (!res) {
mlog(ML_ERROR, "got %sast for unknown lockres! "
- "cookie=%"MLFu64", name=%.*s, namelen=%u\n",
+ "cookie=%u:%llu, name=%.*s, namelen=%u\n",
past->type == DLM_AST ? "" : "b",
- cookie, locklen, name, locklen);
+ dlm_get_lock_cookie_node(cookie),
+ dlm_get_lock_cookie_seq(cookie),
+ locklen, name, locklen);
ret = DLM_IVLOCKID;
goto leave;
}
@@ -360,9 +365,12 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data)
goto do_ast;
}
- mlog(ML_ERROR, "got %sast for unknown lock! cookie=%"MLFu64", "
- "name=%.*s, namelen=%u\n",
- past->type == DLM_AST ? "" : "b", cookie, locklen, name, locklen);
+ mlog(ML_ERROR, "got %sast for unknown lock! cookie=%u:%llu, "
+ "name=%.*s, namelen=%u\n",
+ past->type == DLM_AST ? "" : "b",
+ dlm_get_lock_cookie_node(cookie),
+ dlm_get_lock_cookie_seq(cookie),
+ locklen, name, locklen);
ret = DLM_NORMAL;
unlock_out:
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 9c772583744a..88cc43df18f1 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -630,6 +630,21 @@ __dlm_lockres_state_to_status(struct dlm_lock_resource *res)
return status;
}
+static inline u8 dlm_get_lock_cookie_node(u64 cookie)
+{
+ u8 ret;
+ cookie >>= 56;
+ ret = (u8)(cookie & 0xffULL);
+ return ret;
+}
+
+static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
+{
+ unsigned long long ret;
+ ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
+ return ret;
+}
+
struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
struct dlm_lockstatus *lksb);
void dlm_lock_get(struct dlm_lock *lock);
@@ -658,6 +673,7 @@ void dlm_complete_thread(struct dlm_ctxt *dlm);
int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
+void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
@@ -762,6 +778,11 @@ int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data);
int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data);
int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data);
int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data);
+int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+ u8 nodenum, u8 *real_master);
+int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
+ struct dlm_lock_resource *res, u8 *real_master);
+
int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res,
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
index f66e2d818ccd..8285228d9e37 100644
--- a/fs/ocfs2/dlm/dlmconvert.c
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -284,8 +284,10 @@ enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
if (lock->ml.convert_type != LKM_IVMODE) {
__dlm_print_one_lock_resource(res);
mlog(ML_ERROR, "converting a remote lock that is already "
- "converting! (cookie=%"MLFu64", conv=%d)\n",
- lock->ml.cookie, lock->ml.convert_type);
+ "converting! (cookie=%u:%llu, conv=%d)\n",
+ dlm_get_lock_cookie_node(lock->ml.cookie),
+ dlm_get_lock_cookie_seq(lock->ml.cookie),
+ lock->ml.convert_type);
status = DLM_DENIED;
goto bail;
}
@@ -513,8 +515,9 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data)
leave:
if (!lock)
mlog(ML_ERROR, "did not find lock to convert on grant queue! "
- "cookie=%"MLFu64"\n",
- cnv->cookie);
+ "cookie=%u:%llu\n",
+ dlm_get_lock_cookie_node(cnv->cookie),
+ dlm_get_lock_cookie_seq(cnv->cookie));
else
dlm_lock_put(lock);
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index 54f61b76ab51..c7eae5d3324e 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -72,8 +72,10 @@ void __dlm_print_one_lock_resource(struct dlm_lock_resource *res)
lock = list_entry(iter2, struct dlm_lock, list);
spin_lock(&lock->spinlock);
mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, "
- "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
- lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie,
+ "cookie=%u:%llu, ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
+ lock->ml.type, lock->ml.convert_type, lock->ml.node,
+ dlm_get_lock_cookie_node(lock->ml.cookie),
+ dlm_get_lock_cookie_seq(lock->ml.cookie),
list_empty(&lock->ast_list) ? 'y' : 'n',
lock->ast_pending ? 'y' : 'n',
list_empty(&lock->bast_list) ? 'y' : 'n',
@@ -85,8 +87,10 @@ void __dlm_print_one_lock_resource(struct dlm_lock_resource *res)
lock = list_entry(iter2, struct dlm_lock, list);
spin_lock(&lock->spinlock);
mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, "
- "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
- lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie,
+ "cookie=%u:%llu, ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
+ lock->ml.type, lock->ml.convert_type, lock->ml.node,
+ dlm_get_lock_cookie_node(lock->ml.cookie),
+ dlm_get_lock_cookie_seq(lock->ml.cookie),
list_empty(&lock->ast_list) ? 'y' : 'n',
lock->ast_pending ? 'y' : 'n',
list_empty(&lock->bast_list) ? 'y' : 'n',
@@ -98,8 +102,10 @@ void __dlm_print_one_lock_resource(struct dlm_lock_resource *res)
lock = list_entry(iter2, struct dlm_lock, list);
spin_lock(&lock->spinlock);
mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, "
- "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
- lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie,
+ "cookie=%u:%llu, ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n",
+ lock->ml.type, lock->ml.convert_type, lock->ml.node,
+ dlm_get_lock_cookie_node(lock->ml.cookie),
+ dlm_get_lock_cookie_seq(lock->ml.cookie),
list_empty(&lock->ast_list) ? 'y' : 'n',
lock->ast_pending ? 'y' : 'n',
list_empty(&lock->bast_list) ? 'y' : 'n',
diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c
index dd2d24dc25e0..7e88e24b3471 100644
--- a/fs/ocfs2/dlm/dlmfs.c
+++ b/fs/ocfs2/dlm/dlmfs.c
@@ -596,7 +596,8 @@ static int __init init_dlmfs_fs(void)
dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
sizeof(struct dlmfs_inode_private),
- 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
+ 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
+ SLAB_MEM_SPREAD),
dlmfs_init_once, NULL);
if (!dlmfs_inode_cache)
return -ENOMEM;
diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c
index 671d4ff222cc..6fea28318d6d 100644
--- a/fs/ocfs2/dlm/dlmlock.c
+++ b/fs/ocfs2/dlm/dlmlock.c
@@ -141,13 +141,23 @@ static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
res->lockname.len)) {
kick_thread = 1;
call_ast = 1;
+ } else {
+ mlog(0, "%s: returning DLM_NORMAL to "
+ "node %u for reco lock\n", dlm->name,
+ lock->ml.node);
}
} else {
/* for NOQUEUE request, unless we get the
* lock right away, return DLM_NOTQUEUED */
- if (flags & LKM_NOQUEUE)
+ if (flags & LKM_NOQUEUE) {
status = DLM_NOTQUEUED;
- else {
+ if (dlm_is_recovery_lock(res->lockname.name,
+ res->lockname.len)) {
+ mlog(0, "%s: returning NOTQUEUED to "
+ "node %u for reco lock\n", dlm->name,
+ lock->ml.node);
+ }
+ } else {
dlm_lock_get(lock);
list_add_tail(&lock->list, &res->blocked);
kick_thread = 1;
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 847dd3cc4cf5..940be4c13b1f 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -239,6 +239,8 @@ static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res,
u8 target);
+static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
+ struct dlm_lock_resource *res);
int dlm_is_host_down(int errno)
@@ -677,6 +679,7 @@ struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
struct dlm_node_iter iter;
unsigned int namelen;
int tries = 0;
+ int bit, wait_on_recovery = 0;
BUG_ON(!lockid);
@@ -762,6 +765,18 @@ lookup:
dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
set_bit(dlm->node_num, mle->maybe_map);
list_add(&mle->list, &dlm->master_list);
+
+ /* still holding the dlm spinlock, check the recovery map
+ * to see if there are any nodes that still need to be
+ * considered. these will not appear in the mle nodemap
+ * but they might own this lockres. wait on them. */
+ bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
+ if (bit < O2NM_MAX_NODES) {
+ mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+ "recover before lock mastery can begin\n",
+ dlm->name, namelen, (char *)lockid, bit);
+ wait_on_recovery = 1;
+ }
}
/* at this point there is either a DLM_MLE_BLOCK or a
@@ -779,6 +794,39 @@ lookup:
spin_unlock(&dlm->master_lock);
spin_unlock(&dlm->spinlock);
+ while (wait_on_recovery) {
+ /* any cluster changes that occurred after dropping the
+ * dlm spinlock would be detectable be a change on the mle,
+ * so we only need to clear out the recovery map once. */
+ if (dlm_is_recovery_lock(lockid, namelen)) {
+ mlog(ML_NOTICE, "%s: recovery map is not empty, but "
+ "must master $RECOVERY lock now\n", dlm->name);
+ if (!dlm_pre_master_reco_lockres(dlm, res))
+ wait_on_recovery = 0;
+ else {
+ mlog(0, "%s: waiting 500ms for heartbeat state "
+ "change\n", dlm->name);
+ msleep(500);
+ }
+ continue;
+ }
+
+ dlm_kick_recovery_thread(dlm);
+ msleep(100);
+ dlm_wait_for_recovery(dlm);
+
+ spin_lock(&dlm->spinlock);
+ bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
+ if (bit < O2NM_MAX_NODES) {
+ mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+ "recover before lock mastery can begin\n",
+ dlm->name, namelen, (char *)lockid, bit);
+ wait_on_recovery = 1;
+ } else
+ wait_on_recovery = 0;
+ spin_unlock(&dlm->spinlock);
+ }
+
/* must wait for lock to be mastered elsewhere */
if (blocked)
goto wait;
@@ -792,7 +840,15 @@ redo_request:
mlog_errno(ret);
if (mle->master != O2NM_MAX_NODES) {
/* found a master ! */
- break;
+ if (mle->master <= nodenum)
+ break;
+ /* if our master request has not reached the master
+ * yet, keep going until it does. this is how the
+ * master will know that asserts are needed back to
+ * the lower nodes. */
+ mlog(0, "%s:%.*s: requests only up to %u but master "
+ "is %u, keep going\n", dlm->name, namelen,
+ lockid, nodenum, mle->master);
}
}
@@ -860,7 +916,19 @@ recheck:
/* check if another node has already become the owner */
spin_lock(&res->spinlock);
if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
+ mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name,
+ res->lockname.len, res->lockname.name, res->owner);
spin_unlock(&res->spinlock);
+ /* this will cause the master to re-assert across
+ * the whole cluster, freeing up mles */
+ ret = dlm_do_master_request(mle, res->owner);
+ if (ret < 0) {
+ /* give recovery a chance to run */
+ mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
+ msleep(500);
+ goto recheck;
+ }
+ ret = 0;
goto leave;
}
spin_unlock(&res->spinlock);
@@ -1244,13 +1312,14 @@ int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data)
{
u8 response = DLM_MASTER_RESP_MAYBE;
struct dlm_ctxt *dlm = data;
- struct dlm_lock_resource *res;
+ struct dlm_lock_resource *res = NULL;
struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
char *name;
unsigned int namelen;
int found, ret;
int set_maybe;
+ int dispatch_assert = 0;
if (!dlm_grab(dlm))
return DLM_MASTER_RESP_NO;
@@ -1287,7 +1356,6 @@ way_up_top:
}
if (res->owner == dlm->node_num) {
- u32 flags = DLM_ASSERT_MASTER_MLE_CLEANUP;
spin_unlock(&res->spinlock);
// mlog(0, "this node is the master\n");
response = DLM_MASTER_RESP_YES;
@@ -1300,16 +1368,7 @@ way_up_top:
* caused all nodes up to this one to
* create mles. this node now needs to
* go back and clean those up. */
- mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
- dlm->node_num, res->lockname.len, res->lockname.name);
- ret = dlm_dispatch_assert_master(dlm, res, 1,
- request->node_idx,
- flags);
- if (ret < 0) {
- mlog(ML_ERROR, "failed to dispatch assert "
- "master work\n");
- response = DLM_MASTER_RESP_ERROR;
- }
+ dispatch_assert = 1;
goto send_response;
} else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
spin_unlock(&res->spinlock);
@@ -1357,9 +1416,13 @@ way_up_top:
}
} else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
set_maybe = 0;
- if (tmpmle->master == dlm->node_num)
+ if (tmpmle->master == dlm->node_num) {
response = DLM_MASTER_RESP_YES;
- else
+ /* this node will be the owner.
+ * go back and clean the mles on any
+ * other nodes */
+ dispatch_assert = 1;
+ } else
response = DLM_MASTER_RESP_NO;
} else {
// mlog(0, "this node is attempting to "
@@ -1398,8 +1461,8 @@ way_up_top:
mle = (struct dlm_master_list_entry *)
kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
if (!mle) {
- // bad bad bad... this sucks.
response = DLM_MASTER_RESP_ERROR;
+ mlog_errno(-ENOMEM);
goto send_response;
}
spin_lock(&dlm->spinlock);
@@ -1418,25 +1481,19 @@ way_up_top:
// mlog(0, "mle was found\n");
set_maybe = 1;
spin_lock(&tmpmle->spinlock);
+ if (tmpmle->master == dlm->node_num) {
+ mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n");
+ BUG();
+ }
if (tmpmle->type == DLM_MLE_BLOCK)
response = DLM_MASTER_RESP_NO;
else if (tmpmle->type == DLM_MLE_MIGRATION) {
mlog(0, "migration mle was found (%u->%u)\n",
tmpmle->master, tmpmle->new_master);
- if (tmpmle->master == dlm->node_num) {
- mlog(ML_ERROR, "no lockres, but migration mle "
- "says that this node is master!\n");
- BUG();
- }
/* real master can respond on its own */
response = DLM_MASTER_RESP_NO;
- } else {
- if (tmpmle->master == dlm->node_num) {
- response = DLM_MASTER_RESP_YES;
- set_maybe = 0;
- } else
- response = DLM_MASTER_RESP_MAYBE;
- }
+ } else
+ response = DLM_MASTER_RESP_MAYBE;
if (set_maybe)
set_bit(request->node_idx, tmpmle->maybe_map);
spin_unlock(&tmpmle->spinlock);
@@ -1449,6 +1506,24 @@ way_up_top:
dlm_put_mle(tmpmle);
}
send_response:
+
+ if (dispatch_assert) {
+ if (response != DLM_MASTER_RESP_YES)
+ mlog(ML_ERROR, "invalid response %d\n", response);
+ if (!res) {
+ mlog(ML_ERROR, "bad lockres while trying to assert!\n");
+ BUG();
+ }
+ mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
+ dlm->node_num, res->lockname.len, res->lockname.name);
+ ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
+ DLM_ASSERT_MASTER_MLE_CLEANUP);
+ if (ret < 0) {
+ mlog(ML_ERROR, "failed to dispatch assert master work\n");
+ response = DLM_MASTER_RESP_ERROR;
+ }
+ }
+
dlm_put(dlm);
return response;
}
@@ -1471,8 +1546,11 @@ static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
int to, tmpret;
struct dlm_node_iter iter;
int ret = 0;
+ int reassert;
BUG_ON(namelen > O2NM_MAX_NAME_LEN);
+again:
+ reassert = 0;
/* note that if this nodemap is empty, it returns 0 */
dlm_node_iter_init(nodemap, &iter);
@@ -1504,9 +1582,17 @@ static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
"got %d.\n", namelen, lockname, to, r);
dlm_dump_lock_resources(dlm);
BUG();
+ } else if (r == EAGAIN) {
+ mlog(0, "%.*s: node %u create mles on other "
+ "nodes and requests a re-assert\n",
+ namelen, lockname, to);
+ reassert = 1;
}
}
+ if (reassert)
+ goto again;
+
return ret;
}
@@ -1528,6 +1614,8 @@ int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data)
char *name;
unsigned int namelen;
u32 flags;
+ int master_request = 0;
+ int ret = 0;
if (!dlm_grab(dlm))
return 0;
@@ -1642,11 +1730,22 @@ ok:
// mlog(0, "woo! got an assert_master from node %u!\n",
// assert->node_idx);
if (mle) {
- int extra_ref;
+ int extra_ref = 0;
+ int nn = -1;
spin_lock(&mle->spinlock);
- extra_ref = !!(mle->type == DLM_MLE_BLOCK
- || mle->type == DLM_MLE_MIGRATION);
+ if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
+ extra_ref = 1;
+ else {
+ /* MASTER mle: if any bits set in the response map
+ * then the calling node needs to re-assert to clear
+ * up nodes that this node contacted */
+ while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
+ nn+1)) < O2NM_MAX_NODES) {
+ if (nn != dlm->node_num && nn != assert->node_idx)
+ master_request = 1;
+ }
+ }
mle->master = assert->node_idx;
atomic_set(&mle->woken, 1);
wake_up(&mle->wq);
@@ -1677,10 +1776,15 @@ ok:
}
done:
+ ret = 0;
if (res)
dlm_lockres_put(res);
dlm_put(dlm);
- return 0;
+ if (master_request) {
+ mlog(0, "need to tell master to reassert\n");
+ ret = EAGAIN; // positive. negative would shoot down the node.
+ }
+ return ret;
kill:
/* kill the caller! */
@@ -1713,6 +1817,10 @@ int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
item->u.am.request_from = request_from;
item->u.am.flags = flags;
+ if (ignore_higher)
+ mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
+ res->lockname.name);
+
spin_lock(&dlm->work_lock);
list_add_tail(&item->list, &dlm->work_list);
spin_unlock(&dlm->work_lock);
@@ -1775,6 +1883,61 @@ static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
mlog(0, "finished with dlm_assert_master_worker\n");
}
+/* SPECIAL CASE for the $RECOVERY lock used by the recovery thread.
+ * We cannot wait for node recovery to complete to begin mastering this
+ * lockres because this lockres is used to kick off recovery! ;-)
+ * So, do a pre-check on all living nodes to see if any of those nodes
+ * think that $RECOVERY is currently mastered by a dead node. If so,
+ * we wait a short time to allow that node to get notified by its own
+ * heartbeat stack, then check again. All $RECOVERY lock resources
+ * mastered by dead nodes are purged when the hearbeat callback is
+ * fired, so we can know for sure that it is safe to continue once
+ * the node returns a live node or no node. */
+static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
+ struct dlm_lock_resource *res)
+{
+ struct dlm_node_iter iter;
+ int nodenum;
+ int ret = 0;
+ u8 master = DLM_LOCK_RES_OWNER_UNKNOWN;
+
+ spin_lock(&dlm->spinlock);
+ dlm_node_iter_init(dlm->domain_map, &iter);
+ spin_unlock(&dlm->spinlock);
+
+ while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+ /* do not send to self */
+ if (nodenum == dlm->node_num)
+ continue;
+ ret = dlm_do_master_requery(dlm, res, nodenum, &master);
+ if (ret < 0) {
+ mlog_errno(ret);
+ if (!dlm_is_host_down(ret))
+ BUG();
+ /* host is down, so answer for that node would be
+ * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */
+ }
+
+ if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
+ /* check to see if this master is in the recovery map */
+ spin_lock(&dlm->spinlock);
+ if (test_bit(master, dlm->recovery_map)) {
+ mlog(ML_NOTICE, "%s: node %u has not seen "
+ "node %u go down yet, and thinks the "
+ "dead node is mastering the recovery "
+ "lock. must wait.\n", dlm->name,
+ nodenum, master);
+ ret = -EAGAIN;
+ }
+ spin_unlock(&dlm->spinlock);
+ mlog(0, "%s: reco lock master is %u\n", dlm->name,
+ master);
+ break;
+ }
+ }
+ return ret;
+}
+
/*
* DLM_MIGRATE_LOCKRES
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 1e232000f3f7..805cbabac051 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -58,7 +58,7 @@ static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node);
static int dlm_recovery_thread(void *data);
void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
-static void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
+void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
static int dlm_do_recovery(struct dlm_ctxt *dlm);
static int dlm_pick_recovery_master(struct dlm_ctxt *dlm);
@@ -78,15 +78,9 @@ static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm,
u8 send_to,
struct dlm_lock_resource *res,
int total_locks);
-static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
- struct dlm_lock_resource *res,
- u8 *real_master);
static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res,
struct dlm_migratable_lockres *mres);
-static int dlm_do_master_requery(struct dlm_ctxt *dlm,
- struct dlm_lock_resource *res,
- u8 nodenum, u8 *real_master);
static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm);
static int dlm_send_all_done_msg(struct dlm_ctxt *dlm,
u8 dead_node, u8 send_to);
@@ -165,7 +159,7 @@ void dlm_dispatch_work(void *data)
* RECOVERY THREAD
*/
-static void dlm_kick_recovery_thread(struct dlm_ctxt *dlm)
+void dlm_kick_recovery_thread(struct dlm_ctxt *dlm)
{
/* wake the recovery thread
* this will wake the reco thread in one of three places
@@ -750,10 +744,12 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
dlm->name, dlm->reco.dead_node, dlm->reco.new_master,
dead_node, reco_master);
mlog(ML_ERROR, "%s: name=%.*s master=%u locks=%u/%u flags=%u "
- "entry[0]={c=%"MLFu64",l=%u,f=%u,t=%d,ct=%d,hb=%d,n=%u}\n",
+ "entry[0]={c=%u:%llu,l=%u,f=%u,t=%d,ct=%d,hb=%d,n=%u}\n",
dlm->name, mres->lockname_len, mres->lockname, mres->master,
mres->num_locks, mres->total_locks, mres->flags,
- mres->ml[0].cookie, mres->ml[0].list, mres->ml[0].flags,
+ dlm_get_lock_cookie_node(mres->ml[0].cookie),
+ dlm_get_lock_cookie_seq(mres->ml[0].cookie),
+ mres->ml[0].list, mres->ml[0].flags,
mres->ml[0].type, mres->ml[0].convert_type,
mres->ml[0].highest_blocked, mres->ml[0].node);
BUG();
@@ -1316,9 +1312,8 @@ leave:
-static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
- struct dlm_lock_resource *res,
- u8 *real_master)
+int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
+ struct dlm_lock_resource *res, u8 *real_master)
{
struct dlm_node_iter iter;
int nodenum;
@@ -1360,8 +1355,10 @@ static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
ret = dlm_do_master_requery(dlm, res, nodenum, real_master);
if (ret < 0) {
mlog_errno(ret);
- BUG();
- /* TODO: need to figure a way to restart this */
+ if (!dlm_is_host_down(ret))
+ BUG();
+ /* host is down, so answer for that node would be
+ * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */
}
if (*real_master != DLM_LOCK_RES_OWNER_UNKNOWN) {
mlog(0, "lock master is %u\n", *real_master);
@@ -1372,9 +1369,8 @@ static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
}
-static int dlm_do_master_requery(struct dlm_ctxt *dlm,
- struct dlm_lock_resource *res,
- u8 nodenum, u8 *real_master)
+int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+ u8 nodenum, u8 *real_master)
{
int ret = -EINVAL;
struct dlm_master_requery req;
@@ -1519,9 +1515,11 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
/* lock is always created locally first, and
* destroyed locally last. it must be on the list */
if (!lock) {
+ u64 c = ml->cookie;
mlog(ML_ERROR, "could not find local lock "
- "with cookie %"MLFu64"!\n",
- ml->cookie);
+ "with cookie %u:%llu!\n",
+ dlm_get_lock_cookie_node(c),
+ dlm_get_lock_cookie_seq(c));
BUG();
}
BUG_ON(lock->ml.node != ml->node);
@@ -1739,6 +1737,13 @@ static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
} else
continue;
+ if (!list_empty(&res->recovering)) {
+ mlog(0, "%s:%.*s: lockres was "
+ "marked RECOVERING, owner=%u\n",
+ dlm->name, res->lockname.len,
+ res->lockname.name, res->owner);
+ list_del_init(&res->recovering);
+ }
spin_lock(&res->spinlock);
dlm_change_lockres_owner(dlm, res, new_master);
res->state &= ~DLM_LOCK_RES_RECOVERING;
@@ -2258,7 +2263,10 @@ int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data)
mlog(0, "%u not in domain/live_nodes map "
"so setting it in reco map manually\n",
br->dead_node);
- set_bit(br->dead_node, dlm->recovery_map);
+ /* force the recovery cleanup in __dlm_hb_node_down
+ * both of these will be cleared in a moment */
+ set_bit(br->dead_node, dlm->domain_map);
+ set_bit(br->dead_node, dlm->live_nodes_map);
__dlm_hb_node_down(dlm, br->dead_node);
}
spin_unlock(&dlm->spinlock);
diff --git a/fs/ocfs2/dlm/dlmunlock.c b/fs/ocfs2/dlm/dlmunlock.c
index c95f08d2e925..7b1a27542674 100644
--- a/fs/ocfs2/dlm/dlmunlock.c
+++ b/fs/ocfs2/dlm/dlmunlock.c
@@ -244,8 +244,10 @@ leave:
if (actions & DLM_UNLOCK_FREE_LOCK) {
/* this should always be coupled with list removal */
BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK));
- mlog(0, "lock %"MLFu64" should be gone now! refs=%d\n",
- lock->ml.cookie, atomic_read(&lock->lock_refs.refcount)-1);
+ mlog(0, "lock %u:%llu should be gone now! refs=%d\n",
+ dlm_get_lock_cookie_node(lock->ml.cookie),
+ dlm_get_lock_cookie_seq(lock->ml.cookie),
+ atomic_read(&lock->lock_refs.refcount)-1);
dlm_lock_put(lock);
}
if (actions & DLM_UNLOCK_CALL_AST)
@@ -493,8 +495,9 @@ int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data)
not_found:
if (!found)
mlog(ML_ERROR, "failed to find lock to unlock! "
- "cookie=%"MLFu64"\n",
- unlock->cookie);
+ "cookie=%u:%llu\n",
+ dlm_get_lock_cookie_node(unlock->cookie),
+ dlm_get_lock_cookie_seq(unlock->cookie));
else {
/* send the lksb->status back to the other node */
status = lksb->status;
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-21 07:26:42 +0100