summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAlexander Aring <aahringo@redhat.com>2024-04-02 21:18:02 +0200
committerDavid Teigland <teigland@redhat.com>2024-04-09 18:44:49 +0200
commit3ae67760567438ff857e79bd799154b1f7da3b2a (patch)
tree9fbd4209b0121740070377aee2bab8781550ebb4
parentdlm: move rsb root_list to ls_recover() stack (diff)
downloadlinux-3ae67760567438ff857e79bd799154b1f7da3b2a.tar.xz
linux-3ae67760567438ff857e79bd799154b1f7da3b2a.zip
dlm: add new struct to save position in dlm_copy_master_names
Add a new struct to save the current position in the rsb masters_list while sending the rsb names to other nodes. The rsb names are sent in multiple chunks, and for each new chunk, the new "dlm_dir_dump" struct saves the last position in the masters_list. The new struct is also used to save more information to sanity check the recovery process. Signed-off-by: Alexander Aring <aahringo@redhat.com> Signed-off-by: David Teigland <teigland@redhat.com>
-rw-r--r--fs/dlm/dir.c115
-rw-r--r--fs/dlm/dlm_internal.h4
-rw-r--r--fs/dlm/lockspace.c2
-rw-r--r--fs/dlm/recoverd.c5
4 files changed, 116 insertions, 10 deletions
diff --git a/fs/dlm/dir.c b/fs/dlm/dir.c
index 3da00c46cbb3..0dc8a1d9e411 100644
--- a/fs/dlm/dir.c
+++ b/fs/dlm/dir.c
@@ -224,6 +224,80 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
return NULL;
}
+struct dlm_dir_dump {
+ /* init values to match if whole
+ * dump fits to one seq. Sanity check only.
+ */
+ uint64_t seq_init;
+ uint64_t nodeid_init;
+ /* compare local pointer with last lookup,
+ * just a sanity check.
+ */
+ struct list_head *last;
+
+ unsigned int sent_res; /* for log info */
+ unsigned int sent_msg; /* for log info */
+
+ struct list_head list;
+};
+
+static void drop_dir_ctx(struct dlm_ls *ls, int nodeid)
+{
+ struct dlm_dir_dump *dd, *safe;
+
+ write_lock(&ls->ls_dir_dump_lock);
+ list_for_each_entry_safe(dd, safe, &ls->ls_dir_dump_list, list) {
+ if (dd->nodeid_init == nodeid) {
+ log_error(ls, "drop dump seq %llu",
+ (unsigned long long)dd->seq_init);
+ list_del(&dd->list);
+ kfree(dd);
+ }
+ }
+ write_unlock(&ls->ls_dir_dump_lock);
+}
+
+static struct dlm_dir_dump *lookup_dir_dump(struct dlm_ls *ls, int nodeid)
+{
+ struct dlm_dir_dump *iter, *dd = NULL;
+
+ read_lock(&ls->ls_dir_dump_lock);
+ list_for_each_entry(iter, &ls->ls_dir_dump_list, list) {
+ if (iter->nodeid_init == nodeid) {
+ dd = iter;
+ break;
+ }
+ }
+ read_unlock(&ls->ls_dir_dump_lock);
+
+ return dd;
+}
+
+static struct dlm_dir_dump *init_dir_dump(struct dlm_ls *ls, int nodeid)
+{
+ struct dlm_dir_dump *dd;
+
+ dd = lookup_dir_dump(ls, nodeid);
+ if (dd) {
+ log_error(ls, "found ongoing dir dump for node %d, will drop it",
+ nodeid);
+ drop_dir_ctx(ls, nodeid);
+ }
+
+ dd = kzalloc(sizeof(*dd), GFP_ATOMIC);
+ if (!dd)
+ return NULL;
+
+ dd->seq_init = ls->ls_recover_seq;
+ dd->nodeid_init = nodeid;
+
+ write_lock(&ls->ls_dir_dump_lock);
+ list_add(&dd->list, &ls->ls_dir_dump_list);
+ write_unlock(&ls->ls_dir_dump_lock);
+
+ return dd;
+}
+
/* Find the rsb where we left off (or start again), then send rsb names
for rsb's we're master of and whose directory node matches the requesting
node. inbuf is the rsb name last sent, inlen is the name's length */
@@ -234,11 +308,20 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
struct list_head *list;
struct dlm_rsb *r;
int offset = 0, dir_nodeid;
+ struct dlm_dir_dump *dd;
__be16 be_namelen;
read_lock(&ls->ls_masters_lock);
if (inlen > 1) {
+ dd = lookup_dir_dump(ls, nodeid);
+ if (!dd) {
+ log_error(ls, "failed to lookup dir dump context nodeid: %d",
+ nodeid);
+ goto out;
+ }
+
+ /* next chunk in dump */
r = find_rsb_root(ls, inbuf, inlen);
if (!r) {
log_error(ls, "copy_master_names from %d start %d %.*s",
@@ -246,8 +329,25 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
goto out;
}
list = r->res_masters_list.next;
+
+ /* sanity checks */
+ if (dd->last != &r->res_masters_list ||
+ dd->seq_init != ls->ls_recover_seq) {
+ log_error(ls, "failed dir dump sanity check seq_init: %llu seq: %llu",
+ (unsigned long long)dd->seq_init,
+ (unsigned long long)ls->ls_recover_seq);
+ goto out;
+ }
} else {
+ dd = init_dir_dump(ls, nodeid);
+ if (!dd) {
+ log_error(ls, "failed to allocate dir dump context");
+ goto out;
+ }
+
+ /* start dump */
list = ls->ls_masters_list.next;
+ dd->last = list;
}
for (offset = 0; list != &ls->ls_masters_list; list = list->next) {
@@ -269,7 +369,7 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
be_namelen = cpu_to_be16(0);
memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
offset += sizeof(__be16);
- ls->ls_recover_dir_sent_msg++;
+ dd->sent_msg++;
goto out;
}
@@ -278,7 +378,8 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
offset += sizeof(__be16);
memcpy(outbuf + offset, r->res_name, r->res_length);
offset += r->res_length;
- ls->ls_recover_dir_sent_res++;
+ dd->sent_res++;
+ dd->last = list;
}
/*
@@ -288,10 +389,18 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
if ((list == &ls->ls_masters_list) &&
(offset + sizeof(uint16_t) <= outlen)) {
+ /* end dump */
be_namelen = cpu_to_be16(0xFFFF);
memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
offset += sizeof(__be16);
- ls->ls_recover_dir_sent_msg++;
+ dd->sent_msg++;
+ log_rinfo(ls, "dlm_recover_directory nodeid %d sent %u res out %u messages",
+ nodeid, dd->sent_res, dd->sent_msg);
+
+ write_lock(&ls->ls_dir_dump_lock);
+ list_del_init(&dd->list);
+ write_unlock(&ls->ls_dir_dump_lock);
+ kfree(dd);
}
out:
read_unlock(&ls->ls_masters_lock);
diff --git a/fs/dlm/dlm_internal.h b/fs/dlm/dlm_internal.h
index f434325d5bc8..e03a379832d5 100644
--- a/fs/dlm/dlm_internal.h
+++ b/fs/dlm/dlm_internal.h
@@ -660,8 +660,6 @@ struct dlm_ls {
struct mutex ls_requestqueue_mutex;
struct dlm_rcom *ls_recover_buf;
int ls_recover_nodeid; /* for debugging */
- unsigned int ls_recover_dir_sent_res; /* for log info */
- unsigned int ls_recover_dir_sent_msg; /* for log info */
unsigned int ls_recover_locks_in; /* for log info */
uint64_t ls_rcom_seq;
spinlock_t ls_rcom_spin;
@@ -676,6 +674,8 @@ struct dlm_ls {
struct list_head ls_masters_list; /* root resources */
rwlock_t ls_masters_lock; /* protect root_list */
+ struct list_head ls_dir_dump_list; /* root resources */
+ rwlock_t ls_dir_dump_lock; /* protect root_list */
const struct dlm_lockspace_ops *ls_ops;
void *ls_ops_arg;
diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index da756e5c0f6c..af7769f8e38c 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -582,6 +582,8 @@ static int new_lockspace(const char *name, const char *cluster,
init_waitqueue_head(&ls->ls_wait_general);
INIT_LIST_HEAD(&ls->ls_masters_list);
rwlock_init(&ls->ls_masters_lock);
+ INIT_LIST_HEAD(&ls->ls_dir_dump_list);
+ rwlock_init(&ls->ls_dir_dump_lock);
spin_lock(&lslist_lock);
ls->ls_create_count = 1;
diff --git a/fs/dlm/recoverd.c b/fs/dlm/recoverd.c
index f6acc7351625..0b1a62167798 100644
--- a/fs/dlm/recoverd.c
+++ b/fs/dlm/recoverd.c
@@ -173,8 +173,6 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
goto fail_root_list;
}
- ls->ls_recover_dir_sent_res = 0;
- ls->ls_recover_dir_sent_msg = 0;
ls->ls_recover_locks_in = 0;
dlm_set_recover_status(ls, DLM_RS_NODES);
@@ -211,9 +209,6 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
dlm_release_masters_list(ls);
- log_rinfo(ls, "dlm_recover_directory %u out %u messages",
- ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
-
/*
* We may have outstanding operations that are waiting for a reply from
* a failed node. Mark these to be resent after recovery. Unlock and