smbd: Add share_mode_forall_entries()
[samba.git] / source3 / locking / locking.c
index d7ba65395d03d03a640b06ae946273b98f0374c2..e8820ee03567effa13fa66f5ff993ad5ab793ff7 100644 (file)
@@ -37,6 +37,7 @@
 
 #include "includes.h"
 #include "system/filesys.h"
+#include "lib/util/server_id.h"
 #include "locking/proto.h"
 #include "smbd/globals.h"
 #include "dbwrap/dbwrap.h"
@@ -46,6 +47,8 @@
 #include "messages.h"
 #include "util_tdb.h"
 #include "../librpc/gen_ndr/ndr_open_files.h"
+#include "librpc/gen_ndr/ndr_file_id.h"
+#include "locking/leases_db.h"
 
 #undef DBGC_CLASS
 #define DBGC_CLASS DBGC_LOCKING
@@ -63,10 +66,6 @@ const char *lock_type_name(enum brl_type lock_type)
                        return "READ";
                case WRITE_LOCK:
                        return "WRITE";
-               case PENDING_READ_LOCK:
-                       return "PENDING_READ";
-               case PENDING_WRITE_LOCK:
-                       return "PENDING_WRITE";
                default:
                        return "other";
        }
@@ -101,8 +100,9 @@ void init_strict_lock_struct(files_struct *fsp,
         plock->lock_flav = lp_posix_cifsu_locktype(fsp);
 }
 
-bool strict_lock_default(files_struct *fsp, struct lock_struct *plock)
+bool strict_lock_check_default(files_struct *fsp, struct lock_struct *plock)
 {
+       struct byte_range_lock *br_lck;
        int strict_locking = lp_strict_locking(fsp->conn->params);
        bool ret = False;
 
@@ -115,56 +115,52 @@ bool strict_lock_default(files_struct *fsp, struct lock_struct *plock)
        }
 
        if (strict_locking == Auto) {
-               if  (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && (plock->lock_type == READ_LOCK || plock->lock_type == WRITE_LOCK)) {
-                       DEBUG(10,("is_locked: optimisation - exclusive oplock on file %s\n", fsp_str_dbg(fsp)));
-                       ret = True;
-               } else if ((fsp->oplock_type == LEVEL_II_OPLOCK) &&
-                          (plock->lock_type == READ_LOCK)) {
-                       DEBUG(10,("is_locked: optimisation - level II oplock on file %s\n", fsp_str_dbg(fsp)));
-                       ret = True;
-               } else {
-                       struct byte_range_lock *br_lck;
-
-                       br_lck = brl_get_locks_readonly(fsp);
-                       if (!br_lck) {
-                               return True;
-                       }
-                       ret = brl_locktest(br_lck,
-                                       plock->context.smblctx,
-                                       plock->context.pid,
-                                       plock->start,
-                                       plock->size,
-                                       plock->lock_type,
-                                       plock->lock_flav);
+               uint32_t lease_type = fsp_lease_type(fsp);
+
+               if ((lease_type & SMB2_LEASE_READ) &&
+                    (plock->lock_type == READ_LOCK))
+               {
+                       DBG_DEBUG("optimisation - read lease on file %s\n",
+                                 fsp_str_dbg(fsp));
+                       return true;
                }
-       } else {
-               struct byte_range_lock *br_lck;
 
-               br_lck = brl_get_locks_readonly(fsp);
-               if (!br_lck) {
-                       return True;
+               if ((lease_type & SMB2_LEASE_WRITE) &&
+                    (plock->lock_type == WRITE_LOCK))
+               {
+                       DBG_DEBUG("optimisation - write lease on file %s\n",
+                                 fsp_str_dbg(fsp));
+                       return true;
                }
-               ret = brl_locktest(br_lck,
-                               plock->context.smblctx,
-                               plock->context.pid,
-                               plock->start,
-                               plock->size,
-                               plock->lock_type,
-                               plock->lock_flav);
-       }
-
-       DEBUG(10,("strict_lock_default: flavour = %s brl start=%.0f "
-                       "len=%.0f %s for fnum %llu file %s\n",
-                       lock_flav_name(plock->lock_flav),
-                       (double)plock->start, (double)plock->size,
-                       ret ? "unlocked" : "locked",
-                       (unsigned long long)plock->fnum, fsp_str_dbg(fsp)));
+       }
 
-       return ret;
-}
+       br_lck = brl_get_locks_readonly(fsp);
+       if (!br_lck) {
+               return true;
+       }
+       ret = brl_locktest(br_lck, plock);
 
-void strict_unlock_default(files_struct *fsp, struct lock_struct *plock)
-{
+       if (!ret) {
+               /*
+                * We got a lock conflict. Retry with rw locks to enable
+                * autocleanup. This is the slow path anyway.
+                */
+               br_lck = brl_get_locks(talloc_tos(), fsp);
+               if (br_lck == NULL) {
+                       return true;
+               }
+               ret = brl_locktest(br_lck, plock);
+               TALLOC_FREE(br_lck);
+       }
+
+       DEBUG(10, ("strict_lock_default: flavour = %s brl start=%ju "
+                  "len=%ju %s for fnum %ju file %s\n",
+                  lock_flav_name(plock->lock_flav),
+                  (uintmax_t)plock->start, (uintmax_t)plock->size,
+                  ret ? "unlocked" : "locked",
+                  (uintmax_t)plock->fnum, fsp_str_dbg(fsp)));
+
+       return ret;
 }
 
 /****************************************************************************
@@ -234,78 +230,132 @@ static void decrement_current_lock_count(files_struct *fsp,
  Utility function called by locking requests.
 ****************************************************************************/
 
-struct byte_range_lock *do_lock(struct messaging_context *msg_ctx,
-                       files_struct *fsp,
-                       uint64_t smblctx,
-                       uint64_t count,
-                       uint64_t offset,
-                       enum brl_type lock_type,
-                       enum brl_flavour lock_flav,
-                       bool blocking_lock,
-                       NTSTATUS *perr,
-                       uint64_t *psmblctx,
-                       struct blocking_lock_record *blr)
+struct do_lock_state {
+       struct files_struct *fsp;
+       TALLOC_CTX *req_mem_ctx;
+       const struct GUID *req_guid;
+       uint64_t smblctx;
+       uint64_t count;
+       uint64_t offset;
+       enum brl_type lock_type;
+       enum brl_flavour lock_flav;
+
+       struct server_id blocker_pid;
+       uint64_t blocker_smblctx;
+       NTSTATUS status;
+};
+
+static void do_lock_fn(
+       struct db_record *rec,
+       bool *modified_dependent,
+       void *private_data)
 {
+       struct do_lock_state *state = private_data;
        struct byte_range_lock *br_lck = NULL;
 
+       br_lck = brl_get_locks_for_locking(talloc_tos(),
+                                          state->fsp,
+                                          state->req_mem_ctx,
+                                          state->req_guid);
+       if (br_lck == NULL) {
+               state->status = NT_STATUS_NO_MEMORY;
+               return;
+       }
+
+       state->status = brl_lock(
+               br_lck,
+               state->smblctx,
+               messaging_server_id(state->fsp->conn->sconn->msg_ctx),
+               state->offset,
+               state->count,
+               state->lock_type,
+               state->lock_flav,
+               &state->blocker_pid,
+               &state->blocker_smblctx);
+
+       TALLOC_FREE(br_lck);
+}
+
+NTSTATUS do_lock(files_struct *fsp,
+                TALLOC_CTX *req_mem_ctx,
+                const struct GUID *req_guid,
+                uint64_t smblctx,
+                uint64_t count,
+                uint64_t offset,
+                enum brl_type lock_type,
+                enum brl_flavour lock_flav,
+                struct server_id *pblocker_pid,
+                uint64_t *psmblctx)
+{
+       struct do_lock_state state = {
+               .fsp = fsp,
+               .req_mem_ctx = req_mem_ctx,
+               .req_guid = req_guid,
+               .smblctx = smblctx,
+               .count = count,
+               .offset = offset,
+               .lock_type = lock_type,
+               .lock_flav = lock_flav,
+       };
+       NTSTATUS status;
+
        /* silently return ok on print files as we don't do locking there */
        if (fsp->print_file) {
-               *perr = NT_STATUS_OK;
-               return NULL;
+               return NT_STATUS_OK;
        }
 
        if (!fsp->can_lock) {
-               *perr = fsp->is_directory ? NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
-               return NULL;
+               if (fsp->is_directory) {
+                       return NT_STATUS_INVALID_DEVICE_REQUEST;
+               }
+               return NT_STATUS_INVALID_HANDLE;
        }
 
        if (!lp_locking(fsp->conn->params)) {
-               *perr = NT_STATUS_OK;
-               return NULL;
+               return NT_STATUS_OK;
        }
 
        /* NOTE! 0 byte long ranges ARE allowed and should be stored  */
 
-       DEBUG(10,("do_lock: lock flavour %s lock type %s start=%.0f len=%.0f "
-               "blocking_lock=%s requested for %s file %s\n",
-               lock_flav_name(lock_flav), lock_type_name(lock_type),
-               (double)offset, (double)count, blocking_lock ? "true" :
-               "false", fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
+       DBG_DEBUG("lock flavour %s lock type %s start=%"PRIu64" len=%"PRIu64" "
+                 "requested for %s file %s\n",
+                 lock_flav_name(lock_flav),
+                 lock_type_name(lock_type),
+                 offset,
+                 count,
+                 fsp_fnum_dbg(fsp),
+                 fsp_str_dbg(fsp));
 
-       br_lck = brl_get_locks(talloc_tos(), fsp);
-       if (!br_lck) {
-               *perr = NT_STATUS_NO_MEMORY;
-               return NULL;
+       status = share_mode_do_locked(fsp->file_id, do_lock_fn, &state);
+       if (!NT_STATUS_IS_OK(status)) {
+               DBG_DEBUG("share_mode_do_locked returned %s\n",
+                         nt_errstr(status));
+               return status;
        }
 
-       *perr = brl_lock(msg_ctx,
-                       br_lck,
-                       smblctx,
-                       messaging_server_id(fsp->conn->sconn->msg_ctx),
-                       offset,
-                       count,
-                       lock_type,
-                       lock_flav,
-                       blocking_lock,
-                       psmblctx,
-                       blr);
+       if (psmblctx != NULL) {
+               *psmblctx = state.blocker_smblctx;
+       }
+       if (pblocker_pid != NULL) {
+               *pblocker_pid = state.blocker_pid;
+       }
 
-       DEBUG(10, ("do_lock: returning status=%s\n", nt_errstr(*perr)));
+       DBG_DEBUG("returning status=%s\n", nt_errstr(state.status));
 
        increment_current_lock_count(fsp, lock_flav);
-       return br_lck;
+
+       return state.status;
 }
 
 /****************************************************************************
  Utility function called by unlocking requests.
 ****************************************************************************/
 
-NTSTATUS do_unlock(struct messaging_context *msg_ctx,
-                       files_struct *fsp,
-                       uint64_t smblctx,
-                       uint64_t count,
-                       uint64_t offset,
-                       enum brl_flavour lock_flav)
+NTSTATUS do_unlock(files_struct *fsp,
+                  uint64_t smblctx,
+                  uint64_t count,
+                  uint64_t offset,
+                  enum brl_flavour lock_flav)
 {
        bool ok = False;
        struct byte_range_lock *br_lck = NULL;
@@ -318,17 +368,19 @@ NTSTATUS do_unlock(struct messaging_context *msg_ctx,
                return NT_STATUS_OK;
        }
 
-       DEBUG(10,("do_unlock: unlock start=%.0f len=%.0f requested for %s file %s\n",
-                 (double)offset, (double)count, fsp_fnum_dbg(fsp),
-                 fsp_str_dbg(fsp)));
+       DBG_DEBUG("unlock start=%"PRIu64" len=%"PRIu64" requested for %s file "
+                 "%s\n",
+                 offset,
+                 count,
+                 fsp_fnum_dbg(fsp),
+                 fsp_str_dbg(fsp));
 
        br_lck = brl_get_locks(talloc_tos(), fsp);
        if (!br_lck) {
                return NT_STATUS_NO_MEMORY;
        }
 
-       ok = brl_unlock(msg_ctx,
-                       br_lck,
+       ok = brl_unlock(br_lck,
                        smblctx,
                        messaging_server_id(fsp->conn->sconn->msg_ctx),
                        offset,
@@ -346,63 +398,11 @@ NTSTATUS do_unlock(struct messaging_context *msg_ctx,
        return NT_STATUS_OK;
 }
 
-/****************************************************************************
- Cancel any pending blocked locks.
-****************************************************************************/
-
-NTSTATUS do_lock_cancel(files_struct *fsp,
-                       uint64 smblctx,
-                       uint64_t count,
-                       uint64_t offset,
-                       enum brl_flavour lock_flav,
-                       struct blocking_lock_record *blr)
-{
-       bool ok = False;
-       struct byte_range_lock *br_lck = NULL;
-
-       if (!fsp->can_lock) {
-               return fsp->is_directory ?
-                       NT_STATUS_INVALID_DEVICE_REQUEST : NT_STATUS_INVALID_HANDLE;
-       }
-
-       if (!lp_locking(fsp->conn->params)) {
-               return NT_STATUS_DOS(ERRDOS, ERRcancelviolation);
-       }
-
-       DEBUG(10,("do_lock_cancel: cancel start=%.0f len=%.0f requested for %s file %s\n",
-                 (double)offset, (double)count, fsp_fnum_dbg(fsp),
-                 fsp_str_dbg(fsp)));
-
-       br_lck = brl_get_locks(talloc_tos(), fsp);
-       if (!br_lck) {
-               return NT_STATUS_NO_MEMORY;
-       }
-
-       ok = brl_lock_cancel(br_lck,
-                       smblctx,
-                       messaging_server_id(fsp->conn->sconn->msg_ctx),
-                       offset,
-                       count,
-                       lock_flav,
-                       blr);
-
-       TALLOC_FREE(br_lck);
-
-       if (!ok) {
-               DEBUG(10,("do_lock_cancel: returning ERRcancelviolation.\n" ));
-               return NT_STATUS_DOS(ERRDOS, ERRcancelviolation);
-       }
-
-       decrement_current_lock_count(fsp, lock_flav);
-       return NT_STATUS_OK;
-}
-
 /****************************************************************************
  Remove any locks on this fd. Called from file_close().
 ****************************************************************************/
 
-void locking_close_file(struct messaging_context *msg_ctx,
-                       files_struct *fsp,
+void locking_close_file(files_struct *fsp,
                        enum file_close_type close_type)
 {
        struct byte_range_lock *br_lck;
@@ -422,8 +422,13 @@ void locking_close_file(struct messaging_context *msg_ctx,
        br_lck = brl_get_locks(talloc_tos(),fsp);
 
        if (br_lck) {
-               cancel_pending_lock_requests_by_fid(fsp, br_lck, close_type);
-               brl_close_fnum(msg_ctx, br_lck);
+               /*
+                * Unlocks must trigger dbwrap_watch watchers,
+                * normally in smbd_do_unlocking. Here it's done
+                * implictly, we're closing the file and thus remove a
+                * share mode. This will wake the waiters.
+                */
+               brl_close_fnum(br_lck);
                TALLOC_FREE(br_lck);
        }
 }
@@ -432,19 +437,24 @@ void locking_close_file(struct messaging_context *msg_ctx,
  Print out a share mode.
 ********************************************************************/
 
-char *share_mode_str(TALLOC_CTX *ctx, int num, const struct share_mode_entry *e)
+char *share_mode_str(TALLOC_CTX *ctx, int num,
+                    const struct file_id *id,
+                    const struct share_mode_entry *e)
 {
+       struct server_id_buf tmp;
+       struct file_id_buf ftmp;
+
        return talloc_asprintf(ctx, "share_mode_entry[%d]: "
                 "pid = %s, share_access = 0x%x, private_options = 0x%x, "
                 "access_mask = 0x%x, mid = 0x%llx, type= 0x%x, gen_id = %llu, "
                 "uid = %u, flags = %u, file_id %s, name_hash = 0x%x",
                 num,
-                procid_str_static(&e->pid),
+                server_id_str_buf(e->pid, &tmp),
                 e->share_access, e->private_options,
                 e->access_mask, (unsigned long long)e->op_mid,
                 e->op_type, (unsigned long long)e->share_file_id,
                 (unsigned int)e->uid, (unsigned int)e->flags,
-                file_id_string_tos(&e->id),
+                file_id_str_buf(*id, &ftmp),
                 (unsigned int)e->name_hash);
 }
 
@@ -459,6 +469,32 @@ struct share_mode_lock *get_existing_share_mode_lock(TALLOC_CTX *mem_ctx,
        return get_share_mode_lock(mem_ctx, id, NULL, NULL, NULL);
 }
 
+static bool rename_lease_fn(struct share_mode_lock *lck,
+                           struct share_mode_entry *e,
+                           void *private_data)
+{
+       struct share_mode_data *d = lck->data;
+       NTSTATUS status;
+
+       status = leases_db_rename(&e->client_guid,
+                                 &e->lease_key,
+                                 &d->id,
+                                 d->servicepath,
+                                 d->base_name,
+                                 d->stream_name);
+
+       if (!NT_STATUS_IS_OK(status)) {
+               /* Any error recovery possible here ? */
+               DBG_WARNING("Failed to rename lease key for "
+                           "renamed file %s:%s. %s\n",
+                           d->base_name,
+                           d->stream_name,
+                           nt_errstr(status));
+       }
+
+       return false;
+}
+
 /*******************************************************************
  Sets the service name and filename for rename.
  At this point we emit "file renamed" messages to all
@@ -468,21 +504,22 @@ struct share_mode_lock *get_existing_share_mode_lock(TALLOC_CTX *mem_ctx,
 
 bool rename_share_filename(struct messaging_context *msg_ctx,
                        struct share_mode_lock *lck,
+                       struct file_id id,
                        const char *servicepath,
                        uint32_t orig_name_hash,
                        uint32_t new_name_hash,
                        const struct smb_filename *smb_fname_dst)
 {
        struct share_mode_data *d = lck->data;
-       size_t sp_len;
-       size_t bn_len;
-       size_t sn_len;
-       size_t msg_len;
-       char *frm = NULL;
-       int i;
-       bool strip_two_chars = false;
-       bool has_stream = smb_fname_dst->stream_name != NULL;
+       struct file_rename_message msg = {
+               .id = id,
+               .servicepath = servicepath,
+               .base_name = smb_fname_dst->base_name,
+               .stream_name = smb_fname_dst->stream_name,
+       };
+       uint32_t i;
        struct server_id self_pid = messaging_server_id(msg_ctx);
+       bool ok;
 
        DEBUG(10, ("rename_share_filename: servicepath %s newname %s\n",
                   servicepath, smb_fname_dst->base_name));
@@ -491,55 +528,28 @@ bool rename_share_filename(struct messaging_context *msg_ctx,
         * rename_internal_fsp() and rename_internals() add './' to
         * head of newname if newname does not contain a '/'.
         */
-       if (smb_fname_dst->base_name[0] &&
-           smb_fname_dst->base_name[1] &&
-           smb_fname_dst->base_name[0] == '.' &&
-           smb_fname_dst->base_name[1] == '/') {
-               strip_two_chars = true;
-       }
-
-       d->servicepath = talloc_strdup(d, servicepath);
-       d->base_name = talloc_strdup(d, smb_fname_dst->base_name +
-                                      (strip_two_chars ? 2 : 0));
-       d->stream_name = talloc_strdup(d, smb_fname_dst->stream_name);
-       if (d->base_name == NULL ||
-           (has_stream && d->stream_name == NULL) ||
-           d->servicepath == NULL) {
-               DEBUG(0, ("rename_share_filename: talloc failed\n"));
-               return False;
-       }
-       d->modified = True;
-
-       sp_len = strlen(d->servicepath);
-       bn_len = strlen(d->base_name);
-       sn_len = has_stream ? strlen(d->stream_name) : 0;
 
-       msg_len = MSG_FILE_RENAMED_MIN_SIZE + sp_len + 1 + bn_len + 1 +
-           sn_len + 1;
-
-       /* Set up the name changed message. */
-       frm = talloc_array(d, char, msg_len);
-       if (!frm) {
-               return False;
+       if (strncmp(msg.base_name, "./", 2) == 0) {
+               msg.base_name += 2;
        }
 
-       push_file_id_24(frm, &d->id);
-
-       DEBUG(10,("rename_share_filename: msg_len = %u\n", (unsigned int)msg_len ));
-
-       strlcpy(&frm[24],
-               d->servicepath ? d->servicepath : "",
-               sp_len+1);
-       strlcpy(&frm[24 + sp_len + 1],
-               d->base_name ? d->base_name : "",
-               bn_len+1);
-       strlcpy(&frm[24 + sp_len + 1 + bn_len + 1],
-               d->stream_name ? d->stream_name : "",
-               sn_len+1);
+       d->servicepath = talloc_strdup(d, msg.servicepath);
+       d->base_name = talloc_strdup(d, msg.base_name);
+       d->stream_name = talloc_strdup(d, msg.stream_name);
+       if ((d->servicepath == NULL) ||
+           (d->base_name == NULL) ||
+           ((msg.stream_name != NULL) && (d->stream_name == NULL))) {
+               DBG_WARNING("talloc failed\n");
+               return false;
+       }
+       d->modified = True;
 
        /* Send the messages. */
        for (i=0; i<d->num_share_modes; i++) {
                struct share_mode_entry *se = &d->share_modes[i];
+               DATA_BLOB blob;
+               enum ndr_err_code ndr_err;
+
                if (!is_valid_share_mode_entry(se)) {
                        continue;
                }
@@ -561,16 +571,36 @@ bool rename_share_filename(struct messaging_context *msg_ctx,
                        continue;
                }
 
-               DEBUG(10,("rename_share_filename: sending rename message to "
-                         "pid %s file_id %s sharepath %s base_name %s "
-                         "stream_name %s\n",
-                         procid_str_static(&se->pid),
-                         file_id_string_tos(&d->id),
-                         d->servicepath, d->base_name,
-                       has_stream ? d->stream_name : ""));
+               msg.share_file_id = se->share_file_id;
+
+               ndr_err = ndr_push_struct_blob(
+                       &blob,
+                       talloc_tos(),
+                       &msg,
+                       (ndr_push_flags_fn_t)ndr_push_file_rename_message);
+               if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
+                       DBG_DEBUG("ndr_push_file_rename_message failed: %s\n",
+                                 ndr_errstr(ndr_err));
+                       return false;
+               }
+               if (DEBUGLEVEL >= 10) {
+                       struct server_id_buf tmp;
+                       DBG_DEBUG("sending rename message to %s\n",
+                                 server_id_str_buf(se->pid, &tmp));
+                       NDR_PRINT_DEBUG(file_rename_message, &msg);
+               }
+
+               messaging_send(msg_ctx, se->pid, MSG_SMB_FILE_RENAME, &blob);
+
+               TALLOC_FREE(blob.data);
+       }
 
-               messaging_send_buf(msg_ctx, se->pid, MSG_SMB_FILE_RENAME,
-                                  (uint8 *)frm, msg_len);
+       ok = share_mode_forall_leases(lck, rename_lease_fn, NULL);
+       if (!ok) {
+               /*
+                * Ignore error here. Not sure what to do..
+                */
+               DBG_WARNING("share_mode_forall_leases failed\n");
        }
 
        return True;
@@ -600,14 +630,7 @@ void get_file_infos(struct file_id id,
        }
 
        if (write_time) {
-               struct timespec wt;
-
-               wt = lck->data->changed_write_time;
-               if (null_timespec(wt)) {
-                       wt = lck->data->old_write_time;
-               }
-
-               *write_time = wt;
+               *write_time = get_share_mode_write_time(lck);
        }
 
        TALLOC_FREE(lck);
@@ -624,6 +647,7 @@ bool is_valid_share_mode_entry(const struct share_mode_entry *e)
        num_props += ((e->op_type == NO_OPLOCK) ? 1 : 0);
        num_props += (EXCLUSIVE_OPLOCK_TYPE(e->op_type) ? 1 : 0);
        num_props += (LEVEL_II_OPLOCK_TYPE(e->op_type) ? 1 : 0);
+       num_props += (e->op_type == LEASE_OPLOCK);
 
        if ((num_props > 1) && serverid_exists(&e->pid)) {
                smb_panic("Invalid share mode entry");
@@ -631,6 +655,88 @@ bool is_valid_share_mode_entry(const struct share_mode_entry *e)
        return (num_props != 0);
 }
 
+NTSTATUS remove_lease_if_stale(const struct share_mode_data *d,
+                              const struct GUID *client_guid,
+                              const struct smb2_lease_key *lease_key)
+{
+       uint32_t i;
+       NTSTATUS status;
+
+       for (i=0; i<d->num_share_modes; i++) {
+               const struct share_mode_entry *e = &d->share_modes[i];
+               bool same;
+
+               if (e->stale) {
+                       continue;
+               }
+               if (e->op_type != LEASE_OPLOCK) {
+                       continue;
+               }
+
+               same = smb2_lease_equal(
+                       &e->client_guid,
+                       &e->lease_key,
+                       client_guid,
+                       lease_key);
+               if (same) {
+                       return NT_STATUS_RESOURCE_IN_USE;
+               }
+       }
+
+       status = leases_db_del(client_guid, lease_key, &d->id);
+       if (!NT_STATUS_IS_OK(status)) {
+               DBG_DEBUG("leases_db_del failed: %s\n",
+                         nt_errstr(status));
+       }
+       return status;
+}
+
+/*
+ * See if we need to remove a lease being referred to by a
+ * share mode that is being marked stale or deleted.
+ */
+
+static void remove_share_mode_lease(struct share_mode_data *d,
+                                   struct share_mode_entry *e)
+{
+       uint16_t op_type;
+
+       op_type = e->op_type;
+       e->op_type = NO_OPLOCK;
+
+       d->modified = true;
+
+       if (op_type != LEASE_OPLOCK) {
+               return;
+       }
+
+       remove_lease_if_stale(d, &e->client_guid, &e->lease_key);
+}
+
+bool share_entry_stale_pid(struct share_mode_entry *e)
+{
+       struct server_id_buf buf;
+       bool exists;
+
+       if (e->stale) {
+               return true;
+       }
+
+       exists = serverid_exists(&e->pid);
+       if (exists) {
+               DBG_DEBUG("PID %s still exists\n",
+                         server_id_str_buf(e->pid, &buf));
+               return false;
+       }
+
+       DBG_DEBUG("PID %s does not exist anymore\n",
+                 server_id_str_buf(e->pid, &buf));
+
+       e->stale = true;
+
+       return true;
+}
+
 /*
  * In case d->share_modes[i] conflicts with something or otherwise is
  * being used, we need to make sure the corresponding process still
@@ -639,88 +745,76 @@ bool is_valid_share_mode_entry(const struct share_mode_entry *e)
 bool share_mode_stale_pid(struct share_mode_data *d, uint32_t idx)
 {
        struct share_mode_entry *e;
+       bool stale;
 
        if (idx > d->num_share_modes) {
-               DEBUG(1, ("Asking for index %u, only %u around\n",
-                         idx, (unsigned)d->num_share_modes));
+               DBG_WARNING("Asking for index %"PRIu32", "
+                           "only %"PRIu32" around\n",
+                           idx,
+                           d->num_share_modes);
                return false;
        }
        e = &d->share_modes[idx];
-       if (e->stale) {
-               /*
-                * Checked before
-                */
-               return true;
-       }
-       if (serverid_exists(&e->pid)) {
-               DEBUG(10, ("PID %s (index %u out of %u) still exists\n",
-                          procid_str_static(&e->pid), idx,
-                          (unsigned)d->num_share_modes));
+
+       stale = share_entry_stale_pid(e);
+       if (!stale) {
                return false;
        }
-       DEBUG(10, ("PID %s (index %u out of %u) does not exist anymore\n",
-                  procid_str_static(&e->pid), idx,
-                  (unsigned)d->num_share_modes));
-
-       e->stale = true;
+       d->modified = true;
 
        if (d->num_delete_tokens != 0) {
-               uint32_t i, num_stale;
-
-               /*
-                * We cannot have any delete tokens
-                * if there are no valid share modes.
-                */
-
-               num_stale = 0;
+               uint32_t i;
 
                for (i=0; i<d->num_share_modes; i++) {
-                       if (d->share_modes[i].stale) {
-                               num_stale += 1;
+                       bool valid = !d->share_modes[i].stale;
+                       if (valid) {
+                               break;
                        }
                }
 
-               if (num_stale == d->num_share_modes) {
+               if (i == d->num_share_modes) {
                        /*
-                        * No non-stale share mode found
+                        * No valid (non-stale) share mode found, all
+                        * who might have set the delete token are
+                        * gone.
                         */
                        TALLOC_FREE(d->delete_tokens);
                        d->num_delete_tokens = 0;
                }
        }
 
+       remove_share_mode_lease(d, e);
+
        d->modified = true;
        return true;
 }
 
-/*******************************************************************
- Fill a share mode entry.
-********************************************************************/
-
-static void fill_share_mode_entry(struct share_mode_entry *e,
-                                 files_struct *fsp,
-                                 uid_t uid, uint64_t mid, uint16 op_type)
+void remove_stale_share_mode_entries(struct share_mode_data *d)
 {
-       ZERO_STRUCTP(e);
-       e->pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
-       e->share_access = fsp->share_access;
-       e->private_options = fsp->fh->private_options;
-       e->access_mask = fsp->access_mask;
-       e->op_mid = mid;
-       e->op_type = op_type;
-       e->time.tv_sec = fsp->open_time.tv_sec;
-       e->time.tv_usec = fsp->open_time.tv_usec;
-       e->id = fsp->file_id;
-       e->share_file_id = fsp->fh->gen_id;
-       e->uid = (uint32)uid;
-       e->flags = fsp->posix_open ? SHARE_MODE_FLAG_POSIX_OPEN : 0;
-       e->name_hash = fsp->name_hash;
+       uint32_t i;
+
+       i = 0;
+       while (i < d->num_share_modes) {
+               if (d->share_modes[i].stale) {
+                       struct share_mode_entry *m = d->share_modes;
+                       m[i] = m[d->num_share_modes-1];
+                       d->num_share_modes -= 1;
+                       continue;
+               }
+               i += 1;
+       }
 }
 
-static bool add_share_mode_entry(struct share_mode_data *d,
-                                const struct share_mode_entry *entry)
+bool set_share_mode(struct share_mode_lock *lck,
+                   struct files_struct *fsp,
+                   uid_t uid,
+                   uint64_t mid,
+                   uint16_t op_type,
+                   uint32_t share_access,
+                   uint32_t access_mask)
 {
-       struct share_mode_entry *tmp;
+       struct share_mode_data *d = lck->data;
+       struct share_mode_entry *tmp, *e;
 
        tmp = talloc_realloc(d, d->share_modes, struct share_mode_entry,
                             d->num_share_modes+1);
@@ -728,28 +822,41 @@ static bool add_share_mode_entry(struct share_mode_data *d,
                return false;
        }
        d->share_modes = tmp;
-       d->share_modes[d->num_share_modes] = *entry;
+       e = &d->share_modes[d->num_share_modes];
        d->num_share_modes += 1;
        d->modified = true;
-       return true;
-}
 
-bool set_share_mode(struct share_mode_lock *lck, files_struct *fsp,
-                   uid_t uid, uint64_t mid, uint16 op_type)
-{
-       struct share_mode_entry entry;
-       fill_share_mode_entry(&entry, fsp, uid, mid, op_type);
-       return add_share_mode_entry(lck->data, &entry);
+       ZERO_STRUCTP(e);
+       e->pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       e->share_access = share_access;
+       e->private_options = fsp->fh->private_options;
+       e->access_mask = access_mask;
+       e->op_mid = mid;
+       e->op_type = op_type;
+
+       if (op_type == LEASE_OPLOCK) {
+               const struct GUID *client_guid = fsp_client_guid(fsp);
+               e->client_guid = *client_guid;
+               e->lease_key = fsp->lease->lease.lease_key;
+       }
+
+       e->time.tv_sec = fsp->open_time.tv_sec;
+       e->time.tv_usec = fsp->open_time.tv_usec;
+       e->share_file_id = fsp->fh->gen_id;
+       e->uid = (uint32_t)uid;
+       e->flags = (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) ?
+               SHARE_MODE_FLAG_POSIX_OPEN : 0;
+       e->name_hash = fsp->name_hash;
+
+       return true;
 }
 
 static struct share_mode_entry *find_share_mode_entry(
        struct share_mode_lock *lck, files_struct *fsp)
 {
        struct share_mode_data *d = lck->data;
-       struct server_id pid;
-       int i;
-
-       pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
+       uint32_t i;
 
        for (i=0; i<d->num_share_modes; i++) {
                struct share_mode_entry *e = &d->share_modes[i];
@@ -760,9 +867,6 @@ static struct share_mode_entry *find_share_mode_entry(
                if (!serverid_equal(&pid, &e->pid)) {
                        continue;
                }
-               if (!file_id_equal(&fsp->file_id, &e->id)) {
-                       continue;
-               }
                if (fsp->fh->gen_id != e->share_file_id) {
                        continue;
                }
@@ -772,8 +876,7 @@ static struct share_mode_entry *find_share_mode_entry(
 }
 
 /*******************************************************************
- Del the share mode of a file for this process. Return the number of
- entries left.
+ Del the share mode of a file for this process.
 ********************************************************************/
 
 bool del_share_mode(struct share_mode_lock *lck, files_struct *fsp)
@@ -784,6 +887,7 @@ bool del_share_mode(struct share_mode_lock *lck, files_struct *fsp)
        if (e == NULL) {
                return False;
        }
+       remove_share_mode_lease(lck->data, e);
        *e = lck->data->share_modes[lck->data->num_share_modes-1];
        lck->data->num_share_modes -= 1;
        lck->data->modified = True;
@@ -831,6 +935,7 @@ bool mark_share_mode_disconnected(struct share_mode_lock *lck,
 
 bool remove_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
 {
+       struct share_mode_data *d = lck->data;
        struct share_mode_entry *e;
 
        e = find_share_mode_entry(lck, fsp);
@@ -838,21 +943,9 @@ bool remove_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
                return False;
        }
 
-       if (EXCLUSIVE_OPLOCK_TYPE(e->op_type)) {
-               /*
-                * Going from exclusive or batch,
-                * we always go through FAKE_LEVEL_II
-                * first.
-                */
-               if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-                       smb_panic("remove_share_oplock: logic error");
-               }
-               e->op_type = FAKE_LEVEL_II_OPLOCK;
-       } else {
-               e->op_type = NO_OPLOCK;
-       }
-       lck->data->modified = True;
-       return True;
+       remove_share_mode_lease(d, e);
+       d->modified = True;
+       return true;
 }
 
 /*******************************************************************
@@ -869,6 +962,7 @@ bool downgrade_share_oplock(struct share_mode_lock *lck, files_struct *fsp)
        }
 
        e->op_type = LEVEL_II_OPLOCK;
+       lck->data->flags |= SHARE_MODE_HAS_READ_LEASE;
        lck->data->modified = True;
        return True;
 }
@@ -906,6 +1000,27 @@ static bool add_delete_on_close_token(struct share_mode_data *d,
        return true;
 }
 
+void reset_delete_on_close_lck(files_struct *fsp,
+                              struct share_mode_lock *lck)
+{
+       struct share_mode_data *d = lck->data;
+       uint32_t i;
+
+       for (i=0; i<d->num_delete_tokens; i++) {
+               struct delete_token *dt = &d->delete_tokens[i];
+
+               if (dt->name_hash == fsp->name_hash) {
+                       d->modified = true;
+
+                       /* Delete this entry. */
+                       TALLOC_FREE(dt->delete_nt_token);
+                       TALLOC_FREE(dt->delete_token);
+                       *dt = d->delete_tokens[d->num_delete_tokens-1];
+                       d->num_delete_tokens -= 1;
+               }
+       }
+}
+
 /****************************************************************************
  Sets the delete on close flag over all share modes on this file.
  Modify the share mode entry for all files open
@@ -919,54 +1034,63 @@ static bool add_delete_on_close_token(struct share_mode_data *d,
 
 void set_delete_on_close_lck(files_struct *fsp,
                        struct share_mode_lock *lck,
-                       bool delete_on_close,
                        const struct security_token *nt_tok,
                        const struct security_unix_token *tok)
 {
+       struct messaging_context *msg_ctx = fsp->conn->sconn->msg_ctx;
        struct share_mode_data *d = lck->data;
-       int i;
+       uint32_t i;
        bool ret;
+       DATA_BLOB fid_blob = {};
+       enum ndr_err_code ndr_err;
 
-       if (delete_on_close) {
-               SMB_ASSERT(nt_tok != NULL);
-               SMB_ASSERT(tok != NULL);
-       } else {
-               SMB_ASSERT(nt_tok == NULL);
-               SMB_ASSERT(tok == NULL);
-       }
+       SMB_ASSERT(nt_tok != NULL);
+       SMB_ASSERT(tok != NULL);
 
        for (i=0; i<d->num_delete_tokens; i++) {
                struct delete_token *dt = &d->delete_tokens[i];
                if (dt->name_hash == fsp->name_hash) {
                        d->modified = true;
-                       if (delete_on_close == false) {
-                               /* Delete this entry. */
-                               TALLOC_FREE(dt->delete_nt_token);
-                               TALLOC_FREE(dt->delete_token);
-                               *dt = d->delete_tokens[
-                                       d->num_delete_tokens-1];
-                               d->num_delete_tokens -= 1;
-                       } else {
-                               /* Replace this token with the
-                                  given tok. */
-                               TALLOC_FREE(dt->delete_nt_token);
-                               dt->delete_nt_token = dup_nt_token(dt, nt_tok);
-                               SMB_ASSERT(dt->delete_nt_token != NULL);
-                               TALLOC_FREE(dt->delete_token);
-                               dt->delete_token = copy_unix_token(dt, tok);
-                               SMB_ASSERT(dt->delete_token != NULL);
-                       }
+
+                       /* Replace this token with the given tok. */
+                       TALLOC_FREE(dt->delete_nt_token);
+                       dt->delete_nt_token = dup_nt_token(dt, nt_tok);
+                       SMB_ASSERT(dt->delete_nt_token != NULL);
+                       TALLOC_FREE(dt->delete_token);
+                       dt->delete_token = copy_unix_token(dt, tok);
+                       SMB_ASSERT(dt->delete_token != NULL);
+
                        return;
                }
        }
 
-       if (!delete_on_close) {
-               /* Nothing to delete - not found. */
-               return;
-       }
-
        ret = add_delete_on_close_token(lck->data, fsp->name_hash, nt_tok, tok);
        SMB_ASSERT(ret);
+
+       ndr_err = ndr_push_struct_blob(&fid_blob, talloc_tos(), &fsp->file_id,
+                                      (ndr_push_flags_fn_t)ndr_push_file_id);
+       if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
+               DEBUG(10, ("ndr_push_file_id failed: %s\n",
+                          ndr_errstr(ndr_err)));
+       }
+
+       for (i=0; i<d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+               NTSTATUS status;
+
+               status = messaging_send(
+                       msg_ctx, e->pid, MSG_SMB_NOTIFY_CANCEL_DELETED,
+                       &fid_blob);
+
+               if (!NT_STATUS_IS_OK(status)) {
+                       struct server_id_buf tmp;
+                       DEBUG(10, ("%s: messaging_send to %s returned %s\n",
+                                  __func__, server_id_str_buf(e->pid, &tmp),
+                                  nt_errstr(status)));
+               }
+       }
+
+       TALLOC_FREE(fid_blob.data);
 }
 
 bool set_delete_on_close(files_struct *fsp, bool delete_on_close,
@@ -986,13 +1110,9 @@ bool set_delete_on_close(files_struct *fsp, bool delete_on_close,
        }
 
        if (delete_on_close) {
-               set_delete_on_close_lck(fsp, lck, true,
-                       nt_tok,
-                       tok);
+               set_delete_on_close_lck(fsp, lck, nt_tok, tok);
        } else {
-               set_delete_on_close_lck(fsp, lck, false,
-                       NULL,
-                       NULL);
+               reset_delete_on_close_lck(fsp, lck);
        }
 
        if (fsp->is_directory) {
@@ -1013,14 +1133,13 @@ static struct delete_token *find_delete_on_close_token(
 {
        uint32_t i;
 
-       DEBUG(10, ("find_delete_on_close_token: name_hash = 0x%x\n",
-                  (unsigned int)name_hash));
+       DBG_DEBUG("name_hash = 0x%"PRIx32"\n", name_hash);
 
        for (i=0; i<d->num_delete_tokens; i++) {
                struct delete_token *dt = &d->delete_tokens[i];
 
-               DEBUG(10, ("find__delete_on_close_token: dt->name_hash = 0x%x\n",
-                          (unsigned int)dt->name_hash ));
+               DBG_DEBUG("dt->name_hash = 0x%"PRIx32"\n",
+                         dt->name_hash);
                if (dt->name_hash == name_hash) {
                        return dt;
                }
@@ -1057,11 +1176,12 @@ bool is_delete_on_close_set(struct share_mode_lock *lck, uint32_t name_hash)
 bool set_sticky_write_time(struct file_id fileid, struct timespec write_time)
 {
        struct share_mode_lock *lck;
+       struct file_id_buf ftmp;
 
-       DEBUG(5,("set_sticky_write_time: %s id=%s\n",
+       DBG_INFO("%s id=%s\n",
                 timestring(talloc_tos(),
                            convert_timespec_to_time_t(write_time)),
-                file_id_string_tos(&fileid)));
+                file_id_str_buf(fileid, &ftmp));
 
        lck = get_existing_share_mode_lock(talloc_tos(), fileid);
        if (lck == NULL) {
@@ -1099,3 +1219,165 @@ bool set_write_time(struct file_id fileid, struct timespec write_time)
        TALLOC_FREE(lck);
        return True;
 }
+
+struct timespec get_share_mode_write_time(struct share_mode_lock *lck)
+{
+       struct share_mode_data *d = lck->data;
+
+       if (!null_timespec(d->changed_write_time)) {
+               return d->changed_write_time;
+       }
+       return d->old_write_time;
+}
+
+bool file_has_open_streams(files_struct *fsp)
+{
+       struct share_mode_lock *lock = NULL;
+       struct share_mode_data *d = NULL;
+       uint32_t i;
+
+       lock = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
+       if (lock == NULL) {
+               return false;
+       }
+       d = lock->data;
+
+       for (i = 0; i < d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+
+               if (share_mode_stale_pid(d, i)) {
+                       continue;
+               }
+
+               if (e->private_options &
+                   NTCREATEX_OPTIONS_PRIVATE_STREAM_BASEOPEN)
+               {
+                       TALLOC_FREE(lock);
+                       return true;
+               }
+       }
+
+       TALLOC_FREE(lock);
+       return false;
+}
+
+bool share_mode_forall_entries(
+       struct share_mode_lock *lck,
+       bool (*fn)(struct share_mode_entry *e,
+                  bool *modified,
+                  void *private_data),
+       void *private_data)
+{
+       struct share_mode_data *d = lck->data;
+       uint32_t i;
+
+       for (i=0; i<d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+               struct server_id pid = e->pid;
+               uint64_t share_file_id = e->share_file_id;
+               bool ok, stop;
+               bool modified = false;
+
+               ok = is_valid_share_mode_entry(e);
+               if (!ok) {
+                       continue;
+               }
+
+               stop = fn(e, &modified, private_data);
+
+               if (modified || e->stale) {
+                       d->modified = true;
+               }
+
+               if (modified) {
+                       /*
+                        * In a later commit we will sort the share
+                        * mode array keyed by pid and
+                        * share_file_id. Make sure that from within
+                        * this routine those values don't change.
+                        */
+                       SMB_ASSERT(server_id_equal(&pid, &e->pid));
+                       SMB_ASSERT(share_file_id == e->share_file_id);
+               }
+
+               if (stop) {
+                       return true;
+               }
+       }
+
+       return true;
+}
+
+/*
+ * Walk share mode entries, looking at every lease only once
+ */
+
+bool share_mode_forall_leases(
+       struct share_mode_lock *lck,
+       bool (*fn)(struct share_mode_lock *lck,
+                  struct share_mode_entry *e,
+                  void *private_data),
+       void *private_data)
+{
+       struct share_mode_data *d = lck->data;
+       uint32_t *leases = NULL;
+       uint32_t num_leases = 0;
+       uint32_t i;
+
+       leases = talloc_array(talloc_tos(), uint32_t, d->num_share_modes);
+       if (leases == NULL) {
+               return false;
+       }
+
+       for (i=0; i<d->num_share_modes; i++) {
+               struct share_mode_entry *e = &d->share_modes[i];
+               uint32_t j;
+               bool ok, stop;
+
+               ok = is_valid_share_mode_entry(e);
+               if (!ok) {
+                       continue;
+               }
+
+               if (e->op_type != LEASE_OPLOCK) {
+                       continue;
+               }
+
+               /*
+                * See if we have already seen "e"'s lease. This is
+                * O(n^2). If we sort "leases", we can get this down
+                * to O(n).
+                */
+
+               for (j=0; j<num_leases; j++) {
+                       uint32_t idx = leases[j];
+                       struct share_mode_entry *l = &d->share_modes[idx];
+
+                       if (smb2_lease_equal(&e->client_guid,
+                                            &e->lease_key,
+                                            &l->client_guid,
+                                            &l->lease_key)) {
+                               break;
+                       }
+               }
+               if (j < num_leases) {
+                       /*
+                        * Don't look at "e"'s lease, we've already
+                        * seen it.
+                        */
+                       continue;
+               }
+
+               stop = fn(lck, e, private_data);
+               if (stop) {
+                       TALLOC_FREE(leases);
+                       return true;
+               }
+
+               leases[num_leases] = i;
+               num_leases += 1;
+       }
+
+       TALLOC_FREE(leases);
+       return true;
+}