smbd: Slightly simplify downgrade_lease()
[samba.git] / source3 / smbd / oplock.c
index 067681e0f94ae86f698a67f6c37d6ead13bbb318..ded47b2f525745de5898cb327508f0576e10f81d 100644 (file)
 
 #define DBGC_CLASS DBGC_LOCKING
 #include "includes.h"
+#include "lib/util/server_id.h"
 #include "smbd/smbd.h"
 #include "smbd/globals.h"
 #include "messages.h"
-#include "../librpc/gen_ndr/open_files.h"
-
-/****************************************************************************
- Get the number of current exclusive oplocks.
-****************************************************************************/
-
-int32 get_number_of_exclusive_open_oplocks(void)
-{
-  return exclusive_oplocks_open;
-}
+#include "../librpc/gen_ndr/ndr_open_files.h"
 
 /*
  * helper function used by the kernel oplock backends to post the break message
@@ -56,34 +48,34 @@ void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
 
 /****************************************************************************
  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
- disabled (just sets flags) and no byte-range locks in the file. Returns True
- if oplock set.
+ disabled (just sets flags).
 ****************************************************************************/
 
-bool set_file_oplock(files_struct *fsp, int oplock_type)
+NTSTATUS set_file_oplock(files_struct *fsp)
 {
-       if (fsp->oplock_type == LEVEL_II_OPLOCK) {
-               if (koplocks &&
-                   !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
-                       DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
-                                  "don't support them\n"));
-                       return false;
-               }
+       struct smbd_server_connection *sconn = fsp->conn->sconn;
+       struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
+       bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) &&
+                       (koplocks != NULL);
+
+       if (fsp->oplock_type == LEVEL_II_OPLOCK && use_kernel) {
+               DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
+                          "don't support them\n"));
+               return NT_STATUS_NOT_SUPPORTED;
        }
 
        if ((fsp->oplock_type != NO_OPLOCK) &&
-           (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
-           koplocks &&
-           !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
-               return False;
+           use_kernel &&
+           !koplocks->ops->set_oplock(koplocks, fsp, fsp->oplock_type))
+       {
+               return map_nt_error_from_unix(errno);
        }
 
-       fsp->oplock_type = oplock_type;
        fsp->sent_oplock_break = NO_BREAK_SENT;
-       if (oplock_type == LEVEL_II_OPLOCK) {
-               level_II_oplocks_open++;
+       if (fsp->oplock_type == LEVEL_II_OPLOCK) {
+               sconn->oplocks.level_II_open++;
        } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               exclusive_oplocks_open++;
+               sconn->oplocks.exclusive_open++;
        }
 
        DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
@@ -92,39 +84,38 @@ bool set_file_oplock(files_struct *fsp, int oplock_type)
                 fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
                 (int)fsp->open_time.tv_usec ));
 
-       return True;
+       return NT_STATUS_OK;
 }
 
 /****************************************************************************
  Attempt to release an oplock on a file. Decrements oplock count.
 ****************************************************************************/
 
-void release_file_oplock(files_struct *fsp)
+static void release_file_oplock(files_struct *fsp)
 {
+       struct smbd_server_connection *sconn = fsp->conn->sconn;
+       struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
+       bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) &&
+                       (koplocks != NULL);
+
        if ((fsp->oplock_type != NO_OPLOCK) &&
-           (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
-           koplocks) {
+           use_kernel) {
                koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
        }
 
        if (fsp->oplock_type == LEVEL_II_OPLOCK) {
-               level_II_oplocks_open--;
+               sconn->oplocks.level_II_open--;
        } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               exclusive_oplocks_open--;
+               sconn->oplocks.exclusive_open--;
        }
 
-       SMB_ASSERT(exclusive_oplocks_open>=0);
-       SMB_ASSERT(level_II_oplocks_open>=0);
+       SMB_ASSERT(sconn->oplocks.exclusive_open>=0);
+       SMB_ASSERT(sconn->oplocks.level_II_open>=0);
 
-       if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               /* This doesn't matter for close. */
-               fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
-       } else {
-               fsp->oplock_type = NO_OPLOCK;
-       }
+       fsp->oplock_type = NO_OPLOCK;
        fsp->sent_oplock_break = NO_BREAK_SENT;
 
-       flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
+       flush_write_cache(fsp, SAMBA_OPLOCK_RELEASE_FLUSH);
        delete_write_cache(fsp);
 
        TALLOC_FREE(fsp->oplock_timeout);
@@ -136,18 +127,134 @@ void release_file_oplock(files_struct *fsp)
 
 static void downgrade_file_oplock(files_struct *fsp)
 {
+       struct smbd_server_connection *sconn = fsp->conn->sconn;
+       struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
+       bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) &&
+                       (koplocks != NULL);
+
        if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
                DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
                return;
        }
 
-       if (koplocks) {
+       if (use_kernel) {
                koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
        }
        fsp->oplock_type = LEVEL_II_OPLOCK;
-       exclusive_oplocks_open--;
-       level_II_oplocks_open++;
+       sconn->oplocks.exclusive_open--;
+       sconn->oplocks.level_II_open++;
        fsp->sent_oplock_break = NO_BREAK_SENT;
+
+       flush_write_cache(fsp, SAMBA_OPLOCK_RELEASE_FLUSH);
+       delete_write_cache(fsp);
+
+       TALLOC_FREE(fsp->oplock_timeout);
+}
+
+uint32_t get_lease_type(const struct share_mode_data *d,
+                       const struct share_mode_entry *e)
+{
+       if (e->op_type == LEASE_OPLOCK) {
+               return d->leases[e->lease_idx].current_state;
+       }
+       return map_oplock_to_lease_type(e->op_type);
+}
+
+bool update_num_read_oplocks(files_struct *fsp, struct share_mode_lock *lck)
+{
+       struct share_mode_data *d = lck->data;
+       struct byte_range_lock *br_lck;
+       uint32_t num_read_oplocks = 0;
+       uint32_t i;
+
+       if (fsp_lease_type_is_exclusive(fsp)) {
+               const struct share_mode_entry *e = NULL;
+               uint32_t e_lease_type = 0;
+
+               /*
+                * If we're fully exclusive, we don't need a brlock entry
+                */
+               remove_stale_share_mode_entries(d);
+
+               e = find_share_mode_entry(lck, fsp);
+               if (e != NULL) {
+                       e_lease_type = get_lease_type(d, e);
+               }
+
+               if (!lease_type_is_exclusive(e_lease_type)) {
+                       char *timestr = NULL;
+
+                       timestr = timeval_string(talloc_tos(),
+                                                &fsp->open_time,
+                                                true);
+
+                       NDR_PRINT_DEBUG(share_mode_data, d);
+                       DBG_ERR("file [%s] file_id [%s] gen_id [%lu] "
+                               "open_time[%s] lease_type [0x%x] "
+                               "oplock_type [0x%x]\n",
+                               fsp_str_dbg(fsp),
+                               file_id_string_tos(&fsp->file_id),
+                               fsp->fh->gen_id, timestr,
+                               e_lease_type, fsp->oplock_type);
+
+                       smb_panic("Found non-exclusive lease");
+               }
+
+               return true;
+       }
+
+       for (i=0; i<d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+               uint32_t e_lease_type = get_lease_type(d, e);
+
+               if (e_lease_type & SMB2_LEASE_READ) {
+                       num_read_oplocks += 1;
+               }
+       }
+
+       br_lck = brl_get_locks_readonly(fsp);
+       if (br_lck == NULL) {
+               return false;
+       }
+       if (brl_num_read_oplocks(br_lck) == num_read_oplocks) {
+               return true;
+       }
+
+       br_lck = brl_get_locks(talloc_tos(), fsp);
+       if (br_lck == NULL) {
+               return false;
+       }
+       brl_set_num_read_oplocks(br_lck, num_read_oplocks);
+       TALLOC_FREE(br_lck);
+       return true;
+}
+
+/****************************************************************************
+ Remove a file oplock with lock already held. Copes with level II and exclusive.
+****************************************************************************/
+
+bool remove_oplock_under_lock(files_struct *fsp, struct share_mode_lock *lck)
+{
+       bool ret;
+
+       ret = remove_share_oplock(lck, fsp);
+       if (!ret) {
+               DBG_ERR("failed to remove share oplock for "
+                       "file %s, %s, %s\n",
+                       fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
+                       file_id_string_tos(&fsp->file_id));
+       }
+       release_file_oplock(fsp);
+
+       ret = update_num_read_oplocks(fsp, lck);
+       if (!ret) {
+               DBG_ERR("update_num_read_oplocks failed for "
+                       "file %s, %s, %s\n",
+                       fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
+                       file_id_string_tos(&fsp->file_id));
+       }
+
+       return ret;
 }
 
 /****************************************************************************
@@ -161,22 +268,18 @@ bool remove_oplock(files_struct *fsp)
        bool ret;
        struct share_mode_lock *lck;
 
+       DBG_DEBUG("remove_oplock called for %s\n", fsp_str_dbg(fsp));
+
        /* Remove the oplock flag from the sharemode. */
-       lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
-                                 NULL);
+       lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
        if (lck == NULL) {
-               DEBUG(0,("remove_oplock: failed to lock share entry for "
-                        "file %s\n", fsp_str_dbg(fsp)));
-               return False;
+               DBG_ERR("failed to lock share entry for "
+                        "file %s\n", fsp_str_dbg(fsp));
+               return false;
        }
-       ret = remove_share_oplock(lck, fsp);
-       if (!ret) {
-               DEBUG(0,("remove_oplock: failed to remove share oplock for "
-                        "file %s fnum %d, %s\n",
-                        fsp_str_dbg(fsp), fsp->fnum,
-                        file_id_string_tos(&fsp->file_id)));
-       }
-       release_file_oplock(fsp);
+
+       ret = remove_oplock_under_lock(fsp, lck);
+
        TALLOC_FREE(lck);
        return ret;
 }
@@ -189,8 +292,10 @@ bool downgrade_oplock(files_struct *fsp)
        bool ret;
        struct share_mode_lock *lck;
 
-       lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
-                                 NULL);
+       DEBUG(10, ("downgrade_oplock called for %s\n",
+                  fsp_str_dbg(fsp)));
+
+       lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
        if (lck == NULL) {
                DEBUG(0,("downgrade_oplock: failed to lock share entry for "
                         "file %s\n", fsp_str_dbg(fsp)));
@@ -199,39 +304,422 @@ bool downgrade_oplock(files_struct *fsp)
        ret = downgrade_share_oplock(lck, fsp);
        if (!ret) {
                DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
-                        "for file %s fnum %d, file_id %s\n",
-                        fsp_str_dbg(fsp), fsp->fnum,
+                        "for file %s, %s, file_id %s\n",
+                        fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
                         file_id_string_tos(&fsp->file_id)));
        }
-
        downgrade_file_oplock(fsp);
+
+       ret = update_num_read_oplocks(fsp, lck);
+       if (!ret) {
+               DEBUG(0, ("%s: update_num_read_oplocks failed for "
+                        "file %s, %s, %s\n",
+                         __func__, fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
+                        file_id_string_tos(&fsp->file_id)));
+       }
+
        TALLOC_FREE(lck);
        return ret;
 }
 
-/*
- * Some kernel oplock implementations handle the notification themselves.
- */
-bool should_notify_deferred_opens()
+static void lease_timeout_handler(struct tevent_context *ctx,
+                                 struct tevent_timer *te,
+                                 struct timeval now,
+                                 void *private_data)
+{
+       struct fsp_lease *lease =
+               talloc_get_type_abort(private_data,
+               struct fsp_lease);
+       struct files_struct *fsp;
+       struct share_mode_lock *lck;
+       uint16_t old_epoch = lease->lease.lease_epoch;
+
+       /*
+        * This function runs without any specific impersonation
+        * and must not call any SMB_VFS operations!
+        */
+
+       fsp = file_find_one_fsp_from_lease_key(lease->sconn,
+                                              &lease->lease.lease_key);
+       if (fsp == NULL) {
+               /* race? */
+               TALLOC_FREE(lease->timeout);
+               return;
+       }
+
+       /*
+        * Paranoia check: There can only be one fsp_lease per lease
+        * key
+        */
+       SMB_ASSERT(fsp->lease == lease);
+
+       lck = get_existing_share_mode_lock(
+                       talloc_tos(), fsp->file_id);
+       if (lck == NULL) {
+               /* race? */
+               TALLOC_FREE(lease->timeout);
+               return;
+       }
+
+       fsp_lease_update(lck, fsp_client_guid(fsp), lease);
+
+       if (lease->lease.lease_epoch != old_epoch) {
+               /*
+                * If the epoch changed we need to wait for
+                * the next timeout to happen.
+                */
+               DEBUG(10, ("lease break timeout race (epoch) for file %s - ignoring\n",
+                          fsp_str_dbg(fsp)));
+               TALLOC_FREE(lck);
+               return;
+       }
+
+       if (!(lease->lease.lease_flags & SMB2_LEASE_FLAG_BREAK_IN_PROGRESS)) {
+               /*
+                * If the epoch changed we need to wait for
+                * the next timeout to happen.
+                */
+               DEBUG(10, ("lease break timeout race (flags) for file %s - ignoring\n",
+                          fsp_str_dbg(fsp)));
+               TALLOC_FREE(lck);
+               return;
+       }
+
+       DEBUG(1, ("lease break timed out for file %s -- replying anyway\n",
+                 fsp_str_dbg(fsp)));
+       (void)downgrade_lease(lease->sconn->client->connections,
+                       1,
+                       &fsp->file_id,
+                       &lease->lease.lease_key,
+                       SMB2_LEASE_NONE);
+
+       TALLOC_FREE(lck);
+}
+
+bool fsp_lease_update(struct share_mode_lock *lck,
+                     const struct GUID *client_guid,
+                     struct fsp_lease *lease)
 {
-       return !(koplocks &&
-               (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
+       struct share_mode_data *d = lck->data;
+       int idx;
+       struct share_mode_lease *l = NULL;
+
+       idx = find_share_mode_lease(d, client_guid, &lease->lease.lease_key);
+       if (idx == -1) {
+               DEBUG(1, ("%s: Could not find lease entry\n", __func__));
+               TALLOC_FREE(lease->timeout);
+               lease->lease.lease_state = SMB2_LEASE_NONE;
+               lease->lease.lease_epoch += 1;
+               lease->lease.lease_flags = 0;
+               return false;
+       }
+
+       l = &d->leases[idx];
+
+       DEBUG(10,("%s: refresh lease state\n", __func__));
+
+       /* Ensure we're in sync with current lease state. */
+       if (lease->lease.lease_epoch != l->epoch) {
+               DEBUG(10,("%s: cancel outdated timeout\n", __func__));
+               TALLOC_FREE(lease->timeout);
+       }
+       lease->lease.lease_epoch = l->epoch;
+       lease->lease.lease_state = l->current_state;
+
+       if (l->breaking) {
+               lease->lease.lease_flags |= SMB2_LEASE_FLAG_BREAK_IN_PROGRESS;
+
+               if (lease->timeout == NULL) {
+                       struct timeval t = timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0);
+
+                       DEBUG(10,("%s: setup timeout handler\n", __func__));
+
+                       /*
+                        * lease_timeout_handler() only accesses locking.tdb
+                        * so we don't use any impersonation and use
+                        * the raw tevent context.
+                        */
+                       lease->timeout = tevent_add_timer(lease->sconn->raw_ev_ctx,
+                                                         lease, t,
+                                                         lease_timeout_handler,
+                                                         lease);
+                       if (lease->timeout == NULL) {
+                               DEBUG(0, ("%s: Could not add lease timeout handler\n",
+                                         __func__));
+                       }
+               }
+       } else {
+               lease->lease.lease_flags &= ~SMB2_LEASE_FLAG_BREAK_IN_PROGRESS;
+               TALLOC_FREE(lease->timeout);
+       }
+
+       return true;
 }
 
-/****************************************************************************
- Set up an oplock break message.
-****************************************************************************/
+struct downgrade_lease_additional_state {
+       struct tevent_immediate *im;
+       struct smbXsrv_connection *xconn;
+       uint32_t break_flags;
+       struct smb2_lease_key lease_key;
+       uint32_t break_from;
+       uint32_t break_to;
+       uint16_t new_epoch;
+};
+
+static void downgrade_lease_additional_trigger(struct tevent_context *ev,
+                                              struct tevent_immediate *im,
+                                              void *private_data)
+{
+       struct downgrade_lease_additional_state *state =
+               talloc_get_type_abort(private_data,
+               struct downgrade_lease_additional_state);
+       struct smbXsrv_connection *xconn = state->xconn;
+       NTSTATUS status;
+
+       status = smbd_smb2_send_lease_break(xconn,
+                                           state->new_epoch,
+                                           state->break_flags,
+                                           &state->lease_key,
+                                           state->break_from,
+                                           state->break_to);
+       TALLOC_FREE(state);
+       if (!NT_STATUS_IS_OK(status)) {
+               smbd_server_connection_terminate(xconn,
+                                                nt_errstr(status));
+               return;
+       }
+}
+
+struct downgrade_lease_fsps_state {
+       struct file_id id;
+       struct share_mode_lock *lck;
+       const struct smb2_lease_key *key;
+};
 
-static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
-                                  files_struct *fsp, int cmd)
+static struct files_struct *downgrade_lease_fsps(struct files_struct *fsp,
+                                                void *private_data)
 {
-       char *result = talloc_array(mem_ctx, char, smb_size + 8*2 + 0);
+       struct downgrade_lease_fsps_state *state =
+               (struct downgrade_lease_fsps_state *)private_data;
 
-       if (result == NULL) {
-               DEBUG(0, ("talloc failed\n"));
+       if (fsp->oplock_type != LEASE_OPLOCK) {
+               return NULL;
+       }
+       if (!smb2_lease_key_equal(&fsp->lease->lease.lease_key, state->key)) {
+               return NULL;
+       }
+       if (!file_id_equal(&fsp->file_id, &state->id)) {
                return NULL;
        }
 
+       fsp_lease_update(state->lck, fsp_client_guid(fsp), fsp->lease);
+
+       return NULL;
+}
+
+static NTSTATUS downgrade_share_lease(struct smbd_server_connection *sconn,
+                                     struct share_mode_lock *lck,
+                                     const struct smb2_lease_key *key,
+                                     uint32_t new_lease_state,
+                                     struct share_mode_lease **_l)
+{
+       struct share_mode_data *d = lck->data;
+       struct share_mode_lease *l;
+       int idx;
+
+       *_l = NULL;
+
+       idx = find_share_mode_lease(
+               d, &sconn->client->connections->smb2.client.guid, key);
+       if (idx == -1) {
+               DEBUG(10, ("lease not found\n"));
+               return NT_STATUS_INVALID_PARAMETER;
+       }
+
+       l = &d->leases[idx];
+
+       if (!l->breaking) {
+               DBG_WARNING("Attempt to break from %"PRIu32" to %"PRIu32" - "
+                           "but we're not in breaking state\n",
+                           l->current_state, new_lease_state);
+               return NT_STATUS_UNSUCCESSFUL;
+       }
+
+       /*
+        * Can't upgrade anything: l->breaking_to_requested (and l->current_state)
+        * must be a strict bitwise superset of new_lease_state
+        */
+       if ((new_lease_state & l->breaking_to_requested) != new_lease_state) {
+               DBG_WARNING("Attempt to upgrade from %"PRIu32" to %"PRIu32" "
+                           "- expected %"PRIu32"\n",
+                           l->current_state, new_lease_state,
+                           l->breaking_to_requested);
+               return NT_STATUS_REQUEST_NOT_ACCEPTED;
+       }
+
+       if (l->current_state != new_lease_state) {
+               l->current_state = new_lease_state;
+               d->modified = true;
+       }
+
+       if ((new_lease_state & ~l->breaking_to_required) != 0) {
+               DBG_INFO("lease state %"PRIu32" not fully broken from "
+                        "%"PRIu32" to %"PRIu32"\n",
+                        new_lease_state,
+                        l->current_state,
+                        l->breaking_to_required);
+               l->breaking_to_requested = l->breaking_to_required;
+               if (l->current_state & (~SMB2_LEASE_READ)) {
+                       /*
+                        * Here we break in steps, as windows does
+                        * see the breaking3 and v2_breaking3 tests.
+                        */
+                       l->breaking_to_requested |= SMB2_LEASE_READ;
+               }
+               d->modified = true;
+               *_l = l;
+               return NT_STATUS_OPLOCK_BREAK_IN_PROGRESS;
+       }
+
+       DBG_DEBUG("breaking from %"PRIu32" to %"PRIu32" - "
+                 "expected %"PRIu32"\n",
+                 l->current_state,
+                 new_lease_state,
+                 l->breaking_to_requested);
+
+       l->breaking_to_requested = 0;
+       l->breaking_to_required = 0;
+       l->breaking = false;
+
+       d->modified = true;
+
+       return NT_STATUS_OK;
+}
+
+NTSTATUS downgrade_lease(struct smbXsrv_connection *xconn,
+                        uint32_t num_file_ids,
+                        const struct file_id *ids,
+                        const struct smb2_lease_key *key,
+                        uint32_t lease_state)
+{
+       struct smbd_server_connection *sconn = xconn->client->sconn;
+       struct share_mode_lock *lck;
+       struct share_mode_lease *l = NULL;
+       const struct file_id id = ids[0];
+       uint32_t i;
+       NTSTATUS status;
+
+       DEBUG(10, ("%s: Downgrading %s to %x\n", __func__,
+                  file_id_string_tos(&id), (unsigned)lease_state));
+
+       lck = get_existing_share_mode_lock(talloc_tos(), id);
+       if (lck == NULL) {
+               return NT_STATUS_OBJECT_NAME_NOT_FOUND;
+       }
+       status = downgrade_share_lease(sconn, lck, key, lease_state, &l);
+
+       DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
+                  file_id_string_tos(&id), (unsigned)lease_state, nt_errstr(status)));
+
+       if (NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_BREAK_IN_PROGRESS)) {
+               struct downgrade_lease_additional_state *state;
+
+               state = talloc_zero(xconn,
+                                   struct downgrade_lease_additional_state);
+               if (state == NULL) {
+                       TALLOC_FREE(lck);
+                       return NT_STATUS_NO_MEMORY;
+               }
+
+               state->im = tevent_create_immediate(state);
+               if (state->im == NULL) {
+                       TALLOC_FREE(state);
+                       TALLOC_FREE(lck);
+                       return NT_STATUS_NO_MEMORY;
+               }
+
+               state->xconn = xconn;
+               state->lease_key = l->lease_key;
+               state->break_from = l->current_state;
+               state->break_to = l->breaking_to_requested;
+               if (l->lease_version > 1) {
+                       state->new_epoch = l->epoch;
+               }
+
+               if (l->current_state & (~SMB2_LEASE_READ)) {
+                       state->break_flags =
+                               SMB2_NOTIFY_BREAK_LEASE_FLAG_ACK_REQUIRED;
+               } else {
+                       /*
+                        * This is an async break without
+                        * SMB2_NOTIFY_BREAK_LEASE_FLAG_ACK_REQUIRED
+                        *
+                        * we need to store NONE state in the
+                        * database.
+                        */
+                       l->current_state = 0;
+                       l->breaking_to_requested = 0;
+                       l->breaking_to_required = 0;
+                       l->breaking = false;
+
+                       lck->data->modified = true;
+               }
+
+               tevent_schedule_immediate(state->im,
+                                         xconn->client->raw_ev_ctx,
+                                         downgrade_lease_additional_trigger,
+                                         state);
+       }
+
+       {
+               struct downgrade_lease_fsps_state state = {
+                       .id = id, .lck = lck, .key = key,
+               };
+
+               files_forall(sconn, downgrade_lease_fsps, &state);
+       }
+
+       TALLOC_FREE(lck);
+       DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
+                  file_id_string_tos(&id), (unsigned)lease_state, nt_errstr(status)));
+
+       /*
+        * Dynamic share case. Ensure other opens are copies.
+        * This will only be breaking to NONE.
+        */
+
+       for (i = 1; i < num_file_ids; i++) {
+               lck = get_existing_share_mode_lock(talloc_tos(), ids[i]);
+               if (lck == NULL) {
+                       return NT_STATUS_OBJECT_NAME_NOT_FOUND;
+               }
+
+               {
+                       struct downgrade_lease_fsps_state state = {
+                               .id = ids[i], .lck = lck, .key = key,
+                       };
+
+                       files_forall(sconn, downgrade_lease_fsps, &state);
+               }
+
+               DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
+                       file_id_string_tos(&ids[i]), (unsigned)lease_state, nt_errstr(status)));
+
+               TALLOC_FREE(lck);
+       }
+
+       return status;
+}
+
+/****************************************************************************
+ Set up an oplock break message.
+****************************************************************************/
+
+#define SMB1_BREAK_MESSAGE_LENGTH (smb_size + 8*2)
+
+static void new_break_message_smb1(files_struct *fsp, int cmd,
+                                  char result[SMB1_BREAK_MESSAGE_LENGTH])
+{
        memset(result,'\0',smb_size);
        srv_set_message(result,8,0,true);
        SCVAL(result,smb_com,SMBlockingX);
@@ -243,7 +731,6 @@ static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
        SSVAL(result,smb_vwv2,fsp->fnum);
        SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
        SCVAL(result,smb_vwv3+1,cmd);
-       return result;
 }
 
 /****************************************************************************
@@ -269,12 +756,11 @@ static files_struct *initial_break_processing(
 {
        files_struct *fsp = NULL;
 
-       if( DEBUGLVL( 3 ) ) {
-               dbgtext( "initial_break_processing: called for %s/%u\n",
-                        file_id_string_tos(&id), (int)file_id);
-               dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
-                       exclusive_oplocks_open, level_II_oplocks_open );
-       }
+       DEBUG(3, ("initial_break_processing: called for %s/%u\n"
+                 "Current oplocks_open (exclusive = %d, levelII = %d)\n",
+                 file_id_string_tos(&id), (int)file_id,
+                 sconn->oplocks.exclusive_open,
+                 sconn->oplocks.level_II_open));
 
        /*
         * We need to search the file open table for the
@@ -286,11 +772,9 @@ static files_struct *initial_break_processing(
 
        if(fsp == NULL) {
                /* The file could have been closed in the meantime - return success. */
-               if( DEBUGLVL( 3 ) ) {
-                       dbgtext( "initial_break_processing: cannot find open file with " );
-                       dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
-                       dbgtext( "allowing break to succeed.\n" );
-               }
+               DEBUG(3, ("initial_break_processing: cannot find open file "
+                         "with file_id %s gen_id = %lu, allowing break to "
+                         "succeed.\n", file_id_string_tos(&id), file_id));
                return NULL;
        }
 
@@ -305,32 +789,35 @@ static files_struct *initial_break_processing(
         */
 
        if(fsp->oplock_type == NO_OPLOCK) {
-               if( DEBUGLVL( 3 ) ) {
-                       dbgtext( "initial_break_processing: file %s ",
-                                fsp_str_dbg(fsp));
-                       dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
-                                file_id_string_tos(&id), fsp->fh->gen_id );
-                       dbgtext( "Allowing break to succeed regardless.\n" );
-               }
+               DEBUG(3, ("initial_break_processing: file %s (file_id = %s "
+                         "gen_id = %lu) has no oplock. Allowing break to "
+                         "succeed regardless.\n", fsp_str_dbg(fsp),
+                         file_id_string_tos(&id), fsp->fh->gen_id));
                return NULL;
        }
 
        return fsp;
 }
 
-static void oplock_timeout_handler(struct event_context *ctx,
-                                  struct timed_event *te,
+static void oplock_timeout_handler(struct tevent_context *ctx,
+                                  struct tevent_timer *te,
                                   struct timeval now,
                                   void *private_data)
 {
        files_struct *fsp = (files_struct *)private_data;
 
+       /*
+        * Note this function doesn't run under any specific impersonation and
+        * is not expected to call any SMB_VFS operation!
+        */
+
+       SMB_ASSERT(fsp->sent_oplock_break != NO_BREAK_SENT);
+
        /* Remove the timed event handler. */
        TALLOC_FREE(fsp->oplock_timeout);
        DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
                  fsp_str_dbg(fsp)));
        remove_oplock(fsp);
-       reply_to_oplock_break_requests(fsp);
 }
 
 /*******************************************************************
@@ -339,22 +826,20 @@ static void oplock_timeout_handler(struct event_context *ctx,
 
 static void add_oplock_timeout_handler(files_struct *fsp)
 {
-       /*
-        * If kernel oplocks already notifies smbds when an oplock break times
-        * out, just return.
-        */
-       if (koplocks &&
-           (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
-               return;
-       }
-
        if (fsp->oplock_timeout != NULL) {
                DEBUG(0, ("Logic problem -- have an oplock event hanging "
                          "around\n"));
        }
 
+       /*
+        * For now we keep the logic and use the
+        * raw event context. We're called from
+        * the messaging system from a raw event context.
+        * Also oplock_timeout_handler doesn't invoke
+        * SMB_VFS calls.
+        */
        fsp->oplock_timeout =
-               tevent_add_timer(fsp->conn->sconn->ev_ctx, fsp,
+               tevent_add_timer(fsp->conn->sconn->raw_ev_ctx, fsp,
                                 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
                                 oplock_timeout_handler, fsp);
 
@@ -365,86 +850,49 @@ static void add_oplock_timeout_handler(files_struct *fsp)
 
 static void send_break_message_smb1(files_struct *fsp, int level)
 {
-       char *break_msg = new_break_message_smb1(talloc_tos(),
-                                       fsp,
-                                       level);
-       if (break_msg == NULL) {
-               exit_server("Could not talloc break_msg\n");
-       }
+       struct smbXsrv_connection *xconn = NULL;
+       char break_msg[SMB1_BREAK_MESSAGE_LENGTH];
+
+       /*
+        * For SMB1 we only have one connection
+        */
+       xconn = fsp->conn->sconn->client->connections;
+
+       new_break_message_smb1(fsp, level, break_msg);
 
        show_msg(break_msg);
-       if (!srv_send_smb(fsp->conn->sconn,
+       if (!srv_send_smb(xconn,
                        break_msg, false, 0,
                        IS_CONN_ENCRYPTED(fsp->conn),
                        NULL)) {
                exit_server_cleanly("send_break_message_smb1: "
                        "srv_send_smb failed.");
        }
-
-       TALLOC_FREE(break_msg);
-}
-
-void break_level2_to_none_async(files_struct *fsp)
-{
-       struct smbd_server_connection *sconn = fsp->conn->sconn;
-
-       if (fsp->oplock_type == NO_OPLOCK) {
-               /* We already got a "break to none" message and we've handled
-                * it.  just ignore. */
-               DEBUG(3, ("process_oplock_async_level2_break_message: already "
-                         "broken to none, ignoring.\n"));
-               return;
-       }
-
-       if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
-               /* Don't tell the client, just downgrade. */
-               DEBUG(3, ("process_oplock_async_level2_break_message: "
-                         "downgrading fake level 2 oplock.\n"));
-               remove_oplock(fsp);
-               return;
-       }
-
-       /* Ensure we're really at level2 state. */
-       SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
-
-       DEBUG(10,("process_oplock_async_level2_break_message: sending break "
-                 "to none message for fid %d, file %s\n", fsp->fnum,
-                 fsp_str_dbg(fsp)));
-
-       /* Now send a break to none message to our client. */
-       if (sconn->using_smb2) {
-               send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
-       } else {
-               send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
-       }
-
-       /* Async level2 request, don't send a reply, just remove the oplock. */
-       remove_oplock(fsp);
 }
 
 /*******************************************************************
- This handles the case of a write triggering a break to none
- message on a level2 oplock.
- When we get this message we may be in any of three states :
- NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
- the client for LEVEL2.
+ This handles the generic oplock break message from another smbd.
 *******************************************************************/
 
-void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
-                                                     void *private_data,
-                                                     uint32_t msg_type,
-                                                     struct server_id src,
-                                                     DATA_BLOB *data)
+static void process_oplock_break_message(struct messaging_context *msg_ctx,
+                                        void *private_data,
+                                        uint32_t msg_type,
+                                        struct server_id src,
+                                        DATA_BLOB *data)
 {
+       struct file_id id;
        struct share_mode_entry msg;
        files_struct *fsp;
+       bool use_kernel;
        struct smbd_server_connection *sconn =
-               talloc_get_type(private_data,
+               talloc_get_type_abort(private_data,
                struct smbd_server_connection);
-
-       if (sconn == NULL) {
-               return;
-       }
+       struct server_id self = messaging_server_id(sconn->msg_ctx);
+       struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
+       uint16_t break_from;
+       uint16_t break_to;
+       bool break_needed = true;
+       struct server_id_buf tmp;
 
        if (data->data == NULL) {
                DEBUG(0, ("Got NULL buffer\n"));
@@ -457,129 +905,169 @@ void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx
        }
 
        /* De-linearize incoming message. */
-       message_to_share_mode_entry(&msg, (char *)data->data);
+       message_to_share_mode_entry(&id, &msg, (char *)data->data);
+       break_to = msg.op_type;
 
-       DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
-                  "%s/%llu\n", server_id_str(talloc_tos(), &src),
-                  file_id_string_tos(&msg.id),
+       DEBUG(10, ("Got oplock break to %u message from pid %s: %s/%llu\n",
+                  (unsigned)break_to, server_id_str_buf(src, &tmp),
+                  file_id_string_tos(&id),
                   (unsigned long long)msg.share_file_id));
 
-       fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
+       fsp = initial_break_processing(sconn, id, msg.share_file_id);
 
        if (fsp == NULL) {
                /* We hit a race here. Break messages are sent, and before we
-                * get to process this message, we have closed the file. 
-                * No need to reply as this is an async message. */
-               DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
+                * get to process this message, we have closed the file. */
+               DEBUG(3, ("Did not find fsp\n"));
                return;
        }
 
-       break_level2_to_none_async(fsp);
-}
-
-/*******************************************************************
- This handles the generic oplock break message from another smbd.
-*******************************************************************/
-
-static void process_oplock_break_message(struct messaging_context *msg_ctx,
-                                        void *private_data,
-                                        uint32_t msg_type,
-                                        struct server_id src,
-                                        DATA_BLOB *data)
-{
-       struct share_mode_entry msg;
-       files_struct *fsp;
-       bool break_to_level2 = False;
-       struct smbd_server_connection *sconn =
-               talloc_get_type(private_data,
-               struct smbd_server_connection);
+       break_from = fsp_lease_type(fsp);
 
-       if (sconn == NULL) {
-               return;
+       if (fsp->oplock_type != LEASE_OPLOCK) {
+               if (fsp->sent_oplock_break != NO_BREAK_SENT) {
+                       /*
+                        * Nothing to do anymore
+                        */
+                       DEBUG(10, ("fsp->sent_oplock_break = %d\n",
+                                  fsp->sent_oplock_break));
+                       return;
+               }
        }
 
-       if (data->data == NULL) {
-               DEBUG(0, ("Got NULL buffer\n"));
-               return;
+       if (!(global_client_caps & CAP_LEVEL_II_OPLOCKS)) {
+               DEBUG(10, ("client_caps without level2 oplocks\n"));
+               break_to &= ~SMB2_LEASE_READ;
        }
 
-       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
-               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
-               return;
+       use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) &&
+                       (koplocks != NULL);
+       if (use_kernel) {
+               DEBUG(10, ("Kernel oplocks don't allow level2\n"));
+               break_to &= ~SMB2_LEASE_READ;
        }
 
-       /* De-linearize incoming message. */
-       message_to_share_mode_entry(&msg, (char *)data->data);
+       if (!lp_level2_oplocks(SNUM(fsp->conn))) {
+               DEBUG(10, ("no level2 oplocks by config\n"));
+               break_to &= ~SMB2_LEASE_READ;
+       }
 
-       DEBUG(10, ("Got oplock break message from pid %s: %s/%llu\n",
-                  server_id_str(talloc_tos(), &src),
-                  file_id_string_tos(&msg.id),
-                  (unsigned long long)msg.share_file_id));
+       if (fsp->oplock_type == LEASE_OPLOCK) {
+               struct share_mode_lock *lck;
+               int idx;
+
+               lck = get_existing_share_mode_lock(
+                       talloc_tos(), fsp->file_id);
+               if (lck == NULL) {
+                       /*
+                        * We hit a race here. Break messages are sent, and
+                        * before we get to process this message, we have closed
+                        * the file.
+                        */
+                       DEBUG(3, ("Did not find share_mode\n"));
+                       return;
+               }
 
-       fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
+               idx = find_share_mode_lease(
+                       lck->data,
+                       fsp_client_guid(fsp),
+                       &fsp->lease->lease.lease_key);
+               if (idx != -1) {
+                       struct share_mode_lease *l;
+                       l = &lck->data->leases[idx];
+
+                       break_from = l->current_state;
+                       break_to &= l->current_state;
+
+                       if (l->breaking) {
+                               break_to &= l->breaking_to_required;
+                               if (l->breaking_to_required != break_to) {
+                                       /*
+                                        * Note we don't increment the epoch
+                                        * here, which might be a bug in
+                                        * Windows too...
+                                        */
+                                       l->breaking_to_required = break_to;
+                                       lck->data->modified = true;
+                               }
+                               break_needed = false;
+                       } else if (l->current_state == break_to) {
+                               break_needed = false;
+                       } else if (l->current_state == SMB2_LEASE_READ) {
+                               l->current_state = SMB2_LEASE_NONE;
+                               /* Need to increment the epoch */
+                               l->epoch += 1;
+                               lck->data->modified = true;
+                       } else {
+                               l->breaking = true;
+                               l->breaking_to_required = break_to;
+                               l->breaking_to_requested = break_to;
+                               /* Need to increment the epoch */
+                               l->epoch += 1;
+                               lck->data->modified = true;
+                       }
 
-       if (fsp == NULL) {
-               /* We hit a race here. Break messages are sent, and before we
-                * get to process this message, we have closed the file. Reply
-                * with 'ok, oplock broken' */
-               DEBUG(3, ("Did not find fsp\n"));
+                       /* Ensure we're in sync with current lease state. */
+                       fsp_lease_update(lck, fsp_client_guid(fsp), fsp->lease);
+               }
 
-               /* We just send the same message back. */
-               messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
-                                  (uint8 *)data->data,
-                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
-               return;
+               TALLOC_FREE(lck);
        }
 
-       if (fsp->sent_oplock_break != NO_BREAK_SENT) {
-               /* Remember we have to inform the requesting PID when the
-                * client replies */
-               msg.pid = src;
-               ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
-                            &fsp->pending_break_messages,
-                            &fsp->num_pending_break_messages);
+       if (!break_needed) {
+               DEBUG(10,("%s: skip break\n", __func__));
                return;
        }
 
-       if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
-           !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               DEBUG(3, ("Already downgraded oplock on %s: %s\n",
+       if ((break_from == SMB2_LEASE_NONE) && !break_needed) {
+               DEBUG(3, ("Already downgraded oplock to none on %s: %s\n",
                          file_id_string_tos(&fsp->file_id),
                          fsp_str_dbg(fsp)));
-               /* We just send the same message back. */
-               messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
-                                  (uint8 *)data->data,
-                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
                return;
        }
 
-       if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
-           !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
-           !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
-           lp_level2_oplocks(SNUM(fsp->conn))) {
-               break_to_level2 = True;
+       DEBUG(10, ("break_from=%u, break_to=%u\n",
+                  (unsigned)break_from, (unsigned)break_to));
+
+       if ((break_from == break_to) && !break_needed) {
+               DEBUG(3, ("Already downgraded oplock to %u on %s: %s\n",
+                         (unsigned)break_to,
+                         file_id_string_tos(&fsp->file_id),
+                         fsp_str_dbg(fsp)));
+               return;
        }
 
        /* Need to wait before sending a break
           message if we sent ourselves this message. */
-       if (procid_is_me(&src)) {
+       if (serverid_equal(&self, &src)) {
                wait_before_sending_break();
        }
 
        if (sconn->using_smb2) {
-               send_break_message_smb2(fsp, break_to_level2 ?
-                       OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
+               send_break_message_smb2(fsp, break_from, break_to);
        } else {
-               send_break_message_smb1(fsp, break_to_level2 ?
-                       OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
+               send_break_message_smb1(fsp, (break_to & SMB2_LEASE_READ) ?
+                                       OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
        }
 
-       fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
+       if ((break_from == SMB2_LEASE_READ) &&
+           (break_to == SMB2_LEASE_NONE)) {
+               /*
+                * This is an async break without a reply and thus no timeout
+                *
+                * leases are handled above.
+                */
+               if (fsp->oplock_type != LEASE_OPLOCK) {
+                       remove_oplock(fsp);
+               }
+               return;
+       }
+       if (fsp->oplock_type == LEASE_OPLOCK) {
+               return;
+       }
 
-       msg.pid = src;
-       ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
-                    &fsp->pending_break_messages,
-                    &fsp->num_pending_break_messages);
+       fsp->sent_oplock_break = (break_to & SMB2_LEASE_READ) ?
+               LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
 
        add_oplock_timeout_handler(fsp);
 }
@@ -598,12 +1086,9 @@ static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
        unsigned long file_id;
        files_struct *fsp;
        struct smbd_server_connection *sconn =
-               talloc_get_type(private_data,
+               talloc_get_type_abort(private_data,
                struct smbd_server_connection);
-
-       if (sconn == NULL) {
-               return;
-       }
+       struct server_id_buf tmp;
 
        if (data->data == NULL) {
                DEBUG(0, ("Got NULL buffer\n"));
@@ -620,7 +1105,7 @@ static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
        file_id = (unsigned long)IVAL(data->data, 24);
 
        DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
-                  server_id_str(talloc_tos(), &src), file_id_string_tos(&id),
+                  server_id_str_buf(src, &tmp), file_id_string_tos(&id),
                   (unsigned int)file_id));
 
        fsp = initial_break_processing(sconn, id, file_id);
@@ -639,7 +1124,7 @@ static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
        }
 
        if (sconn->using_smb2) {
-               send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
+               send_break_message_smb2(fsp, 0, OPLOCKLEVEL_NONE);
        } else {
                send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
        }
@@ -649,150 +1134,200 @@ static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
        add_oplock_timeout_handler(fsp);
 }
 
-void reply_to_oplock_break_requests(files_struct *fsp)
+static bool file_has_read_oplocks(struct files_struct *fsp)
 {
-       int i;
+       struct byte_range_lock *brl;
+       uint32_t num_read_oplocks = 0;
 
-       /*
-        * If kernel oplocks already notifies smbds when oplocks are
-        * broken/removed, just return.
-        */
-       if (koplocks &&
-           (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
-               return;
+       brl = brl_get_locks_readonly(fsp);
+       if (brl == NULL) {
+               return false;
        }
 
-       for (i=0; i<fsp->num_pending_break_messages; i++) {
-               struct share_mode_entry *e = &fsp->pending_break_messages[i];
-               char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
+       num_read_oplocks = brl_num_read_oplocks(brl);
 
-               share_mode_entry_to_message(msg, e);
+       DBG_DEBUG("num_read_oplocks = %"PRIu32"\n", num_read_oplocks);
 
-               messaging_send_buf(fsp->conn->sconn->msg_ctx, e->pid,
-                                  MSG_SMB_BREAK_RESPONSE,
-                                  (uint8 *)msg,
-                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
-       }
-
-       SAFE_FREE(fsp->pending_break_messages);
-       fsp->num_pending_break_messages = 0;
-       if (fsp->oplock_timeout != NULL) {
-               /* Remove the timed event handler. */
-               TALLOC_FREE(fsp->oplock_timeout);
-               fsp->oplock_timeout = NULL;
-       }
-       return;
+       return (num_read_oplocks != 0);
 }
 
-static void process_oplock_break_response(struct messaging_context *msg_ctx,
-                                         void *private_data,
-                                         uint32_t msg_type,
-                                         struct server_id src,
-                                         DATA_BLOB *data)
+struct break_to_none_state {
+       struct smbd_server_connection *sconn;
+       struct file_id id;
+       struct smb2_lease_key lease_key;
+       struct GUID client_guid;
+};
+static void do_break_to_none(struct tevent_context *ctx,
+                            struct tevent_immediate *im,
+                            void *private_data);
+
+/****************************************************************************
+ This function is called on any file modification or lock request. If a file
+ is level 2 oplocked then it must tell all other level 2 holders to break to
+ none.
+****************************************************************************/
+
+static void contend_level2_oplocks_begin_default(files_struct *fsp,
+                                             enum level2_contention_type type)
 {
-       struct share_mode_entry msg;
-       struct smbd_server_connection *sconn =
-               talloc_get_type(private_data,
-               struct smbd_server_connection);
+       struct smbd_server_connection *sconn = fsp->conn->sconn;
+       struct tevent_immediate *im;
+       struct break_to_none_state *state;
+       bool has_read_oplocks;
 
-       if (sconn == NULL) {
-               return;
-       }
+       /*
+        * If this file is level II oplocked then we need
+        * to grab the shared memory lock and inform all
+        * other files with a level II lock that they need
+        * to flush their read caches. We keep the lock over
+        * the shared memory area whilst doing this.
+        */
 
-       if (data->data == NULL) {
-               DEBUG(0, ("Got NULL buffer\n"));
+       if (fsp_lease_type_is_exclusive(fsp)) {
+               /*
+                * There can't be any level2 oplocks, we're alone.
+                */
                return;
        }
 
-       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
-               DEBUG(0, ("Got invalid msg len %u\n",
-                         (unsigned int)data->length));
+       has_read_oplocks = file_has_read_oplocks(fsp);
+       if (!has_read_oplocks) {
+               DEBUG(10, ("No read oplocks around\n"));
                return;
        }
 
-       /* De-linearize incoming message. */
-       message_to_share_mode_entry(&msg, (char *)data->data);
-
-       DEBUG(10, ("Got oplock break response from pid %s: %s/%llu mid %llu\n",
-                  server_id_str(talloc_tos(), &src),
-                  file_id_string_tos(&msg.id),
-                  (unsigned long long)msg.share_file_id,
-                  (unsigned long long)msg.op_mid));
-
-       schedule_deferred_open_message_smb(sconn, msg.op_mid);
-}
-
-static void process_open_retry_message(struct messaging_context *msg_ctx,
-                                      void *private_data,
-                                      uint32_t msg_type,
-                                      struct server_id src,
-                                      DATA_BLOB *data)
-{
-       struct share_mode_entry msg;
-       struct smbd_server_connection *sconn;
+       /*
+        * When we get here we might have a brlock entry locked. Also
+        * locking the share mode entry would violate the locking
+        * order. Breaking level2 oplocks to none is asynchronous
+        * anyway, so we postpone this into an immediate event.
+        */
 
-       if (data->data == NULL) {
-               DEBUG(0, ("Got NULL buffer\n"));
+       state = talloc_zero(sconn, struct break_to_none_state);
+       if (state == NULL) {
+               DEBUG(1, ("talloc failed\n"));
                return;
        }
+       state->sconn = sconn;
+       state->id = fsp->file_id;
+
+       if (fsp->oplock_type == LEASE_OPLOCK) {
+               state->client_guid = *fsp_client_guid(fsp);
+               state->lease_key = fsp->lease->lease.lease_key;
+               DEBUG(10, ("Breaking through lease key %"PRIu64"/%"PRIu64"\n",
+                          state->lease_key.data[0],
+                          state->lease_key.data[1]));
+       }
 
-       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
-               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
+       im = tevent_create_immediate(state);
+       if (im == NULL) {
+               DEBUG(1, ("tevent_create_immediate failed\n"));
+               TALLOC_FREE(state);
                return;
        }
 
-       /* De-linearize incoming message. */
-       message_to_share_mode_entry(&msg, (char *)data->data);
+       /*
+        * do_break_to_none() only operates on the
+        * locking.tdb and sends network packets to
+        * the client. That doesn't require any
+        * impersonation, so we just use the
+        * raw tevent context here.
+        */
+       tevent_schedule_immediate(im, sconn->raw_ev_ctx, do_break_to_none, state);
+}
 
-       DEBUG(10, ("Got open retry msg from pid %s: %s mid %llu\n",
-                  server_id_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
-                  (unsigned long long)msg.op_mid));
+static void send_break_to_none(struct messaging_context *msg_ctx,
+                              const struct file_id *id,
+                              const struct share_mode_entry *e)
+{
+       char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 
-       sconn = msg_ctx_to_sconn(msg_ctx);
-       if (sconn != NULL) {
-               schedule_deferred_open_message_smb(sconn, msg.op_mid);
-       }
-}
+       share_mode_entry_to_message(msg, id, e);
+       /* Overload entry->op_type */
+       SSVAL(msg, OP_BREAK_MSG_OP_TYPE_OFFSET, NO_OPLOCK);
 
-/****************************************************************************
- This function is called on any file modification or lock request. If a file
- is level 2 oplocked then it must tell all other level 2 holders to break to
- none.
-****************************************************************************/
+       messaging_send_buf(msg_ctx, e->pid, MSG_SMB_BREAK_REQUEST,
+                          (uint8_t *)msg, sizeof(msg));
+}
 
-static void contend_level2_oplocks_begin_default(files_struct *fsp,
-                                             enum level2_contention_type type)
+static void do_break_to_none(struct tevent_context *ctx,
+                            struct tevent_immediate *im,
+                            void *private_data)
 {
-       int i;
+       struct break_to_none_state *state = talloc_get_type_abort(
+               private_data, struct break_to_none_state);
+       uint32_t i;
        struct share_mode_lock *lck;
+       struct share_mode_data *d;
 
        /*
-        * If this file is level II oplocked then we need
-        * to grab the shared memory lock and inform all
-        * other files with a level II lock that they need
-        * to flush their read caches. We keep the lock over
-        * the shared memory area whilst doing this.
+        * Note this function doesn't run under any specific impersonation and
+        * is not expected to call any SMB_VFS operation!
         */
 
-       if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
-               return;
-
-       lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
-                                 NULL);
+       lck = get_existing_share_mode_lock(talloc_tos(), state->id);
        if (lck == NULL) {
-               DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
-                        "share mode entry for file %s.\n", fsp_str_dbg(fsp)));
-               return;
+               DEBUG(1, ("%s: failed to lock share mode entry for file %s.\n",
+                         __func__, file_id_string_tos(&state->id)));
+               goto done;
        }
+       d = lck->data;
 
-       DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
-                 lck->num_share_modes ));
+       /*
+        * Walk leases and oplocks separately: We have to send one break per
+        * lease. If we have multiple share_mode_entry having a common lease,
+        * we would break the lease twice if we don't walk the leases list
+        * separately.
+        */
 
-       for(i = 0; i < lck->num_share_modes; i++) {
-               struct share_mode_entry *share_entry = &lck->share_modes[i];
-               char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
+       for (i=0; i<d->num_leases; i++) {
+               struct share_mode_lease *l = &d->leases[i];
+               struct share_mode_entry *e = NULL;
+               uint32_t j;
 
-               if (!is_valid_share_mode_entry(share_entry)) {
+               if ((l->current_state & SMB2_LEASE_READ) == 0) {
+                       continue;
+               }
+               if (smb2_lease_equal(&state->client_guid,
+                                    &state->lease_key,
+                                    &l->client_guid,
+                                    &l->lease_key)) {
+                       DEBUG(10, ("Don't break our own lease\n"));
+                       continue;
+               }
+
+               for (j=0; j<d->num_share_modes; j++) {
+                       e = &d->share_modes[j];
+
+                       if (!is_valid_share_mode_entry(e)) {
+                               continue;
+                       }
+                       if (e->lease_idx == i) {
+                               break;
+                       }
+               }
+               if (j == d->num_share_modes) {
+                       DEBUG(0, ("leases[%"PRIu32"] has no share mode\n",
+                                 i));
+                       continue;
+               }
+
+               DEBUG(10, ("Breaking lease# %"PRIu32" with share_entry# "
+                          "%"PRIu32"\n", i, j));
+
+               send_break_to_none(state->sconn->msg_ctx, &state->id, e);
+       }
+
+       for(i = 0; i < d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+
+               if (!is_valid_share_mode_entry(e)) {
+                       continue;
+               }
+               if (e->op_type == LEASE_OPLOCK) {
+                       /*
+                        * Took care of those in the loop above
+                        */
                        continue;
                }
 
@@ -800,98 +1335,61 @@ static void contend_level2_oplocks_begin_default(files_struct *fsp,
                 * As there could have been multiple writes waiting at the
                 * lock_share_entry gate we may not be the first to
                 * enter. Hence the state of the op_types in the share mode
-                * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
+                * entries may be partly NO_OPLOCK and partly LEVEL_II
                 * oplock. It will do no harm to re-send break messages to
                 * those smbd's that are still waiting their turn to remove
                 * their LEVEL_II state, and also no harm to ignore existing
                 * NO_OPLOCK states. JRA.
                 */
 
-               DEBUG(10,("release_level_2_oplocks_on_change: "
-                         "share_entry[%i]->op_type == %d\n",
-                         i, share_entry->op_type ));
+               DEBUG(10, ("%s: share_entry[%i]->op_type == %d\n", __func__,
+                          i, e->op_type ));
 
-               if (share_entry->op_type == NO_OPLOCK) {
+               if (e->op_type == NO_OPLOCK) {
                        continue;
                }
 
                /* Paranoia .... */
-               if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
-                       DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
-                                "share mode entry %d is an exlusive "
-                                "oplock !\n", i ));
+               if (EXCLUSIVE_OPLOCK_TYPE(e->op_type)) {
+                       DEBUG(0,("%s: PANIC. "
+                                "share mode entry %d is an exclusive "
+                                "oplock !\n", __func__, i ));
                        TALLOC_FREE(lck);
                        abort();
                }
 
-               share_mode_entry_to_message(msg, share_entry);
-
-               /*
-                * Deal with a race condition when breaking level2
-                * oplocks. Don't send all the messages and release
-                * the lock, this allows someone else to come in and
-                * get a level2 lock before any of the messages are
-                * processed, and thus miss getting a break message.
-                * Ensure at least one entry (the one we're breaking)
-                * is processed immediately under the lock and becomes
-                * set as NO_OPLOCK to stop any waiter getting a level2.
-                * Bugid #5980.
-                */
-
-               if (procid_is_me(&share_entry->pid)) {
-                       struct files_struct *cur_fsp =
-                               initial_break_processing(fsp->conn->sconn,
-                                       share_entry->id,
-                                       share_entry->share_file_id);
-                       wait_before_sending_break();
-                       if (cur_fsp != NULL) {
-                               break_level2_to_none_async(cur_fsp);
-                       } else {
-                               DEBUG(3, ("release_level_2_oplocks_on_change: "
-                               "Did not find fsp, ignoring\n"));
-                       }
-               } else {
-                       messaging_send_buf(fsp->conn->sconn->msg_ctx,
-                                       share_entry->pid,
-                                       MSG_SMB_ASYNC_LEVEL2_BREAK,
-                                       (uint8 *)msg,
-                                       MSG_SMB_SHARE_MODE_ENTRY_SIZE);
-               }
+               send_break_to_none(state->sconn->msg_ctx, &state->id, e);
        }
 
        /* We let the message receivers handle removing the oplock state
           in the share mode lock db. */
 
        TALLOC_FREE(lck);
+done:
+       TALLOC_FREE(state);
+       return;
 }
 
 void smbd_contend_level2_oplocks_begin(files_struct *fsp,
                                  enum level2_contention_type type)
 {
-       if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
-               koplocks->ops->contend_level2_oplocks_begin(fsp, type);
-               return;
-       }
-
        contend_level2_oplocks_begin_default(fsp, type);
 }
 
 void smbd_contend_level2_oplocks_end(files_struct *fsp,
                                enum level2_contention_type type)
 {
-       /* Only kernel oplocks implement this so far */
-       if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
-               koplocks->ops->contend_level2_oplocks_end(fsp, type);
-       }
+       return;
 }
 
 /****************************************************************************
  Linearize a share mode entry struct to an internal oplock break message.
 ****************************************************************************/
 
-void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
+void share_mode_entry_to_message(char *msg, const struct file_id *id,
+                                const struct share_mode_entry *e)
 {
-       SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
+       SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32_t)e->pid.pid);
        SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
        SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
        SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
@@ -899,7 +1397,11 @@ void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
        SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
        SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
        SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
-       push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
+       /*
+        * "id" used to be part of share_mode_entry, thus the strange
+        * place to put this. Feel free to move somewhere else :-)
+        */
+       push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, id);
        SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
        SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
        SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
@@ -911,7 +1413,9 @@ void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
  De-linearize an internal oplock break message to a share mode entry struct.
 ****************************************************************************/
 
-void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
+void message_to_share_mode_entry(struct file_id *id,
+                                struct share_mode_entry *e,
+                                const char *msg)
 {
        e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
        e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
@@ -921,10 +1425,14 @@ void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
        e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
        e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
        e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
-       pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
+       /*
+        * "id" used to be part of share_mode_entry, thus the strange
+        * place to put this. Feel free to move somewhere else :-)
+        */
+       pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, id);
        e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
-       e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
-       e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
+       e->uid = (uint32_t)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
+       e->flags = (uint16_t)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
        e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
        e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
 }
@@ -939,25 +1447,20 @@ bool init_oplocks(struct smbd_server_connection *sconn)
 
        messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_REQUEST,
                           process_oplock_break_message);
-       messaging_register(sconn->msg_ctx, sconn, MSG_SMB_ASYNC_LEVEL2_BREAK,
-                          process_oplock_async_level2_break_message);
-       messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_RESPONSE,
-                          process_oplock_break_response);
        messaging_register(sconn->msg_ctx, sconn, MSG_SMB_KERNEL_BREAK,
                           process_kernel_oplock_break);
-       messaging_register(sconn->msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
-                          process_open_retry_message);
-
-       if (lp_kernel_oplocks()) {
-#if HAVE_KERNEL_OPLOCKS_IRIX
-               koplocks = irix_init_kernel_oplocks(NULL);
-#elif HAVE_KERNEL_OPLOCKS_LINUX
-               koplocks = linux_init_kernel_oplocks(NULL);
-#elif HAVE_ONEFS
-#error Isilon, please check if the NULL context is okay here. Thanks!
-               koplocks = onefs_init_kernel_oplocks(NULL);
+       return true;
+}
+
+void init_kernel_oplocks(struct smbd_server_connection *sconn)
+{
+       struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
+
+       /* only initialize once */
+       if (koplocks == NULL) {
+#if HAVE_KERNEL_OPLOCKS_LINUX
+               koplocks = linux_init_kernel_oplocks(sconn);
 #endif
+               sconn->oplocks.kernel_ops = koplocks;
        }
-
-       return True;
 }