cleanupd: Persistent Handle cleanup

This adds the logic to trigger scavending Persistent Handles from a few places:

- when we got the cluster exclusive ph-cleanup glock after starting
- when the current glock holder node died
- after an unclean process exit

This traverses the node local persistent smbXsrv_open_global_persistent.tdb
which is expected to be leightweight and fast.

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Anoop C S <anoopcs@samba.org>

diff --git a/source3/smbd/smbd_cleanupd.c b/source3/smbd/smbd_cleanupd.c
index 0a50b18331185f3405c30ef2fb06dd3aba4d37c5..7d33ac8b1ca2b67f255f70641ca0c9e3b77fa534 100644 (file)
--- a/source3/smbd/smbd_cleanupd.c
+++ b/source3/smbd/smbd_cleanupd.c
@@ -26,9 +26,26 @@
 #include "serverid.h"
 #include "locking/proto.h"
 #include "cleanupdb.h"
+#include "g_lock.h"
+#include "lib/util/util_tdb.h"
+#include "smbd/globals.h"
+#include "librpc/gen_ndr/ndr_open_files.h"
+#include "scavenger.h"
+#include "source3/smbd/smbXsrv_open.h"
+
+struct cleanup_rec {
+       struct cleanup_rec *prev, *next;
+       uint64_t open_persistent_id;
+       struct file_id id;
+       uint32_t name_hash;
+};
 
 struct smbd_cleanupd_state {
        pid_t parent_pid;
+       struct messaging_context *msg;
+       struct g_lock_ctx *glock_ctx;
+       bool got_glock;
+       struct cleanup_rec *cleanup_list;
 };
 
 static void smbd_cleanupd_shutdown(struct messaging_context *msg,
@@ -39,6 +56,7 @@ static void smbd_cleanupd_process_exited(struct messaging_context *msg,
                                         void *private_data, uint32_t msg_type,
                                         struct server_id server_id,
                                         DATA_BLOB *data);
+static void smbd_cleanupd_got_glock(struct tevent_req *subreq);
 
 struct tevent_req *smbd_cleanupd_send(TALLOC_CTX *mem_ctx,
                                      struct tevent_context *ev,
@@ -46,6 +64,7 @@ struct tevent_req *smbd_cleanupd_send(TALLOC_CTX *mem_ctx,
                                      pid_t parent_pid)
 {
        struct tevent_req *req;
+       struct tevent_req *subreq = NULL;
        struct smbd_cleanupd_state *state;
        NTSTATUS status;
 
@@ -54,6 +73,13 @@ struct tevent_req *smbd_cleanupd_send(TALLOC_CTX *mem_ctx,
                return NULL;
        }
        state->parent_pid = parent_pid;
+       state->msg = msg;
+
+       state->glock_ctx = g_lock_ctx_init(state, msg);
+       if (state->glock_ctx == NULL) {
+               tevent_req_nterror(req, NT_STATUS_INTERNAL_ERROR);
+               return tevent_req_post(req, ev);
+       }
 
        status = messaging_register(msg, req, MSG_SHUTDOWN,
                                    smbd_cleanupd_shutdown);
@@ -67,9 +93,111 @@ struct tevent_req *smbd_cleanupd_send(TALLOC_CTX *mem_ctx,
                return tevent_req_post(req, ev);
        }
 
+       if (!lp_persistent_handles()) {
+               return req;
+       }
+
+       subreq = g_lock_lock_send(state,
+                                 ev,
+                                 state->glock_ctx,
+                                 string_term_tdb_data("cleanupd"),
+                                 G_LOCK_WRITE,
+                                 NULL, NULL);
+       if (tevent_req_nomem(subreq, req)) {
+               return tevent_req_post(req, ev);
+       }
+       tevent_req_set_callback(subreq,
+                               smbd_cleanupd_got_glock,
+                               req);
+
        return req;
 }
 
+static int cleanup_ph_fn(struct db_record *dbrec,
+                        struct smbXsrv_open_global0 *global,
+                        TDB_DATA *rc_open_global_key,
+                        void *private_data)
+{
+       struct smbd_cleanupd_state *state = talloc_get_type_abort(
+               private_data, struct smbd_cleanupd_state);
+       struct cleanup_rec *rec = NULL;
+
+       if (global == NULL) {
+               return 0;
+       }
+
+       rec = talloc_zero(state, struct cleanup_rec);
+       if (rec == NULL) {
+               DBG_ERR("talloc_zero failed\n");
+               return -1;
+       }
+       *rec = (struct cleanup_rec) {
+               .open_persistent_id = global->open_persistent_id,
+               .id = global->file_id,
+               .name_hash = global->name_hash,
+       };
+
+       DLIST_ADD(state->cleanup_list, rec);
+       return 0;
+}
+
+static NTSTATUS cleanup_ph(struct smbd_cleanupd_state *state)
+{
+       struct cleanup_rec *rec = NULL;
+       NTSTATUS status;
+
+       if (!lp_persistent_handles()) {
+               return NT_STATUS_OK;
+       }
+
+       DBG_INFO("Cleaning up persistent handles\n");
+
+       status = smbXsrv_open_global_traverse_per_rec_persistent_read(
+               cleanup_ph_fn, state);
+       if (!NT_STATUS_IS_OK(status)) {
+               DBG_ERR("PH cleanup failed\n");
+               return status;
+       }
+
+       rec = state->cleanup_list;
+       while (rec != NULL) {
+               struct cleanup_rec *next = rec->next;
+
+               scavenger_schedule_disconnected(state->msg,
+                                               rec->open_persistent_id,
+                                               &rec->id,
+                                               rec->name_hash);
+               DLIST_REMOVE(state->cleanup_list, rec);
+               TALLOC_FREE(rec);
+               rec = next;
+       }
+
+
+       return NT_STATUS_OK;
+}
+
+static void smbd_cleanupd_got_glock(struct tevent_req *subreq)
+{
+       struct tevent_req *req = tevent_req_callback_data(
+               subreq, struct tevent_req);
+       struct smbd_cleanupd_state *state = tevent_req_data(
+               req, struct smbd_cleanupd_state);
+       NTSTATUS status;
+
+       status = g_lock_lock_recv(subreq);
+       TALLOC_FREE(subreq);
+       if (tevent_req_nterror(req, status)) {
+               return;
+       }
+       state->got_glock = true;
+
+       status = cleanup_ph(state);
+       if (!NT_STATUS_IS_OK(status)) {
+               DBG_ERR("Persistent handle cleanup failed\n");
+               return;
+       }
+}
+
 static void smbd_cleanupd_shutdown(struct messaging_context *msg,
                                   void *private_data, uint32_t msg_type,
                                   struct server_id server_id,
@@ -77,6 +205,20 @@ static void smbd_cleanupd_shutdown(struct messaging_context *msg,
 {
        struct tevent_req *req = talloc_get_type_abort(
                private_data, struct tevent_req);
+       struct smbd_cleanupd_state *state = tevent_req_data(
+               req, struct smbd_cleanupd_state);
+       NTSTATUS status;
+
+       if (!state->got_glock) {
+               tevent_req_done(req);
+               return;
+       }
+
+       status = g_lock_unlock(state->glock_ctx,
+                              string_term_tdb_data("cleanupd"));
+       if (tevent_req_nterror(req, status)) {
+               return;
+       }
        tevent_req_done(req);
 }
 
@@ -126,6 +268,8 @@ static void smbd_cleanupd_process_exited(struct messaging_context *msg,
        struct cleanupdb_traverse_state cleanup_state;
        TALLOC_CTX *frame = talloc_stackframe();
        struct cleanup_child *child = NULL;
+       bool unclean = false;
+       NTSTATUS status;
 
        cleanup_state = (struct cleanupdb_traverse_state) {
                .mem_ctx = frame
@@ -170,9 +314,20 @@ static void smbd_cleanupd_process_exited(struct messaging_context *msg,
                                  strerror(ret));
                }
 
+               if (child->unclean) {
+                       unclean = true;
+               }
+
                DBG_DEBUG("cleaned up pid %d\n", (int)child->pid);
        }
 
+       if (unclean) {
+               status = cleanup_ph(state);
+               if (!NT_STATUS_IS_OK(status)) {
+                       DBG_ERR("Persistent handle cleanup failed\n");
+               }
+       }
+
        TALLOC_FREE(frame);
 }