MDEV-23855: Shrink fil_space_t

Merge n_pending_ios, n_pending_ops to std::atomic<uint32_t> n_pending.
Change some more fil_space_t members to uint32_t to reduce
the memory footprint.

fil_space_t::add(), fil_ibd_create(): Attach the already opened
handle to the tablespace, and enforce the fil_system.n_open limit.

dict_boot(): Initialize fil_system.max_assigned_id.

srv_boot(): Call srv_thread_pool_init() before anything else,
so that files should be opened in the correct mode on Windows.

fil_ibd_create(): Create the file in OS_FILE_AIO mode, just like
fil_node_open_file_low() does it.

dict_table_t::is_accessible(): Replaces fil_table_accessible().

Reviewed by: Vladislav Vaintroub

extra/mariabackup/fil_cur.cc

@@ -93,7 +93,6 @@ xb_fil_node_close_file(
 	mutex_enter(&fil_system.mutex);
 
 	ut_ad(node);
-	ut_a(node->n_pending_flushes == 0);
 	ut_a(!node->being_extended);
 
 	if (!node->is_open()) {
@@ -406,7 +405,7 @@ xb_fil_cur_read(
 	retry_count = 10;
 	ret = XB_FIL_CUR_SUCCESS;
 
-	fil_space_t *space = fil_space_t::get_for_io(cursor->space_id);
+	fil_space_t *space = fil_space_t::get(cursor->space_id);
 
 	if (!space) {
 		return XB_FIL_CUR_ERROR;
@@ -455,7 +454,7 @@ xb_fil_cur_read(
 
 	posix_fadvise(cursor->file, offset, to_read, POSIX_FADV_DONTNEED);
 func_exit:
-	space->release_for_io();
+	space->release();
 	return(ret);
 }
 

extra/mariabackup/xtrabackup.cc

@@ -3094,10 +3094,10 @@ xb_load_single_table_tablespace(
 
 		ut_a(space != NULL);
 
-		space->add(file->filepath(), OS_FILE_CLOSED, 0, false, false);
-		/* by opening the tablespace we forcing node and space objects
-		in the cache to be populated with fields from space header */
-		space->get_size();
+		space->add(file->filepath(), file->detach(), 0, false, false);
+		mutex_enter(&fil_system.mutex);
+		space->read_page0();
+		mutex_exit(&fil_system.mutex);
 
 		if (srv_operation == SRV_OPERATION_RESTORE_DELTA
 		    || xb_close_files) {
@@ -3402,7 +3402,7 @@ xb_load_tablespaces()
 /** Destroy the tablespace memory cache. */
 static void xb_data_files_close()
 {
-  fil_close_all_files();
+  fil_space_t::close_all();
   buf_dblwr.close();
 }
 

storage/innobase/btr/btr0bulk.cc

@@ -60,12 +60,10 @@ PageBulk::init()
 		alloc_mtr.start();
 		m_index->set_modified(alloc_mtr);
 
-		ulint	n_reserved;
-		bool	success;
-		success = fsp_reserve_free_extents(&n_reserved,
-						   m_index->table->space,
-						   1, FSP_NORMAL, &alloc_mtr);
-		if (!success) {
+		uint32_t n_reserved;
+		if (!fsp_reserve_free_extents(&n_reserved,
+					      m_index->table->space,
+					      1, FSP_NORMAL, &alloc_mtr)) {
 			alloc_mtr.commit();
 			m_mtr.commit();
 			return(DB_OUT_OF_FILE_SPACE);

storage/innobase/btr/btr0cur.cc

@@ -3318,20 +3318,16 @@ static void btr_cur_prefetch_siblings(const buf_block_t *block,
   uint32_t prev= mach_read_from_4(my_assume_aligned<4>(page + FIL_PAGE_PREV));
   uint32_t next= mach_read_from_4(my_assume_aligned<4>(page + FIL_PAGE_NEXT));
 
-  if (prev != FIL_NULL)
-  {
-    ut_a(index->table->space->acquire_for_io());
+  if (prev == FIL_NULL);
+  else if (index->table->space->acquire())
     buf_read_page_background(index->table->space,
                              page_id_t(block->page.id().space(), prev),
                              block->zip_size(), false);
-  }
-  if (next != FIL_NULL)
-  {
-    ut_a(index->table->space->acquire_for_io());
+  if (next == FIL_NULL);
+  else if (index->table->space->acquire())
     buf_read_page_background(index->table->space,
                              page_id_t(block->page.id().space(), next),
                              block->zip_size(), false);
-  }
 }
 
 /*************************************************************//**
@@ -3679,7 +3675,7 @@ btr_cur_pessimistic_insert(
 	dberr_t		err;
 	bool		inherit = false;
 	bool		success;
-	ulint		n_reserved	= 0;
+	uint32_t	n_reserved	= 0;
 
 	ut_ad(dtuple_check_typed(entry));
 	ut_ad(thr || !(~flags & (BTR_NO_LOCKING_FLAG | BTR_NO_UNDO_LOG_FLAG)));
@@ -3711,7 +3707,7 @@ btr_cur_pessimistic_insert(
 		of the index tree, so that the insert will not fail because
 		of lack of space */
 
-		ulint	n_extents = cursor->tree_height / 16 + 3;
+		uint32_t n_extents = uint32_t(cursor->tree_height / 16 + 3);
 
 		success = fsp_reserve_free_extents(&n_reserved,
 						   index->table->space,
@@ -4878,8 +4874,8 @@ btr_cur_pessimistic_update(
 	dberr_t		err;
 	dberr_t		optim_err;
 	roll_ptr_t	roll_ptr;
-	ibool		was_first;
-	ulint		n_reserved	= 0;
+	bool		was_first;
+	uint32_t	n_reserved	= 0;
 
 	*offsets = NULL;
 	*big_rec = NULL;
@@ -5041,7 +5037,7 @@ btr_cur_pessimistic_update(
 		of the index tree, so that the update will not fail because
 		of lack of space */
 
-		ulint	n_extents = cursor->tree_height / 16 + 3;
+		uint32_t n_extents = uint32_t(cursor->tree_height / 16 + 3);
 
 		if (!fsp_reserve_free_extents(
 		            &n_reserved, index->table->space, n_extents,
@@ -5643,7 +5639,7 @@ btr_cur_pessimistic_delete(
 	page_zip_des_t*	page_zip;
 	dict_index_t*	index;
 	rec_t*		rec;
-	ulint		n_reserved	= 0;
+	uint32_t	n_reserved	= 0;
 	bool		success;
 	ibool		ret		= FALSE;
 	mem_heap_t*	heap;
@@ -5672,7 +5668,7 @@ btr_cur_pessimistic_delete(
 		of the index tree, so that the node pointer updates will
 		not fail because of lack of space */
 
-		ulint	n_extents = cursor->tree_height / 32 + 1;
+		uint32_t n_extents = uint32_t(cursor->tree_height / 32 + 1);
 
 		success = fsp_reserve_free_extents(&n_reserved,
 						   index->table->space,
@@ -7315,7 +7311,7 @@ btr_store_big_rec_extern_fields(
 		for (ulint blob_npages = 0;; ++blob_npages) {
 			buf_block_t*	block;
 			const ulint	commit_freq = 4;
-			ulint		r_extents;
+			uint32_t	r_extents;
 
 			ut_ad(page_align(field_ref) == page_align(rec));
 
@@ -7588,7 +7584,7 @@ static void btr_check_blob_fil_page_type(const buf_block_t& block, bool read)
   if (UNIV_LIKELY(type == FIL_PAGE_TYPE_BLOB))
     return;
   /* FIXME: take the tablespace as a parameter */
-  if (fil_space_t *space= fil_space_acquire_silent(block.page.id().space()))
+  if (fil_space_t *space= fil_space_t::get(block.page.id().space()))
   {
     /* Old versions of InnoDB did not initialize FIL_PAGE_TYPE on BLOB
     pages.  Do not print anything about the type mismatch when reading

storage/innobase/buf/buf0buf.cc

@@ -415,7 +415,7 @@ static bool buf_tmp_page_decrypt(byte* tmp_frame, byte* src_frame)
 static bool buf_page_decrypt_after_read(buf_page_t *bpage,
                                         const fil_node_t &node)
 {
-	ut_ad(node.space->pending_io());
+	ut_ad(node.space->referenced());
 	ut_ad(node.space->id == bpage->id().space());
 	const auto flags = node.space->flags;
 
@@ -475,7 +475,7 @@ static bool buf_page_decrypt_after_read(buf_page_t *bpage,
 		slot->release();
 		ut_ad(!write_size
 		      || fil_page_type_validate(node.space, dst_frame));
-		ut_ad(node.space->pending_io());
+		ut_ad(node.space->referenced());
 		return write_size != 0;
 	}
 
@@ -516,7 +516,7 @@ static bool buf_page_decrypt_after_read(buf_page_t *bpage,
 		goto decompress;
 	}
 
-	ut_ad(node.space->pending_io());
+	ut_ad(node.space->referenced());
 	return true;
 }
 #endif /* !UNIV_INNOCHECKSUM */
@@ -2768,7 +2768,7 @@ buf_zip_decompress(
 	ulint		size = page_zip_get_size(&block->page.zip);
 	/* The tablespace will not be found if this function is called
 	during IMPORT. */
-	fil_space_t* space= fil_space_t::get_for_io(block->page.id().space());
+	fil_space_t* space= fil_space_t::get(block->page.id().space());
 	const unsigned key_version = mach_read_from_4(
 		frame + FIL_PAGE_FILE_FLUSH_LSN_OR_KEY_VERSION);
 	fil_space_crypt_t* crypt_data = space ? space->crypt_data : NULL;
@@ -2805,7 +2805,7 @@ buf_zip_decompress(
 		if (page_zip_decompress(&block->page.zip,
 					block->frame, TRUE)) {
 			if (space) {
-				space->release_for_io();
+				space->release();
 			}
 			return(TRUE);
 		}
@@ -2824,7 +2824,7 @@ buf_zip_decompress(
 		/* Copy to uncompressed storage. */
 		memcpy(block->frame, frame, block->zip_size());
 		if (space) {
-			space->release_for_io();
+			space->release();
 		}
 
 		return(TRUE);
@@ -2848,7 +2848,7 @@ buf_zip_decompress(
 			dict_set_corrupted_by_space(space);
 		}
 
-		space->release_for_io();
+		space->release();
 	}
 
 	return(FALSE);
@@ -3160,10 +3160,10 @@ buf_page_get_low(
 			asserting. */
 			if (page_id.space() == TRX_SYS_SPACE) {
 			} else if (page_id.space() == SRV_TMP_SPACE_ID) {
-			} else if (fil_space_t* space= fil_space_t::get_for_io(
+			} else if (fil_space_t* space= fil_space_t::get(
 					   page_id.space())) {
 				bool set = dict_set_corrupted_by_space(space);
-				space->release_for_io();
+				space->release();
 				if (set) {
 					return NULL;
 				}
@@ -3374,8 +3374,7 @@ buf_page_get_low(
 	if (mode != BUF_GET_IF_IN_POOL
 	    && mode != BUF_GET_IF_IN_POOL_OR_WATCH) {
 	} else if (!ibuf_debug) {
-	} else if (fil_space_t* space
-		   = fil_space_t::get_for_io(page_id.space())) {
+	} else if (fil_space_t* space = fil_space_t::get(page_id.space())) {
 		/* Try to evict the block from the buffer pool, to use the
 		insert buffer (change buffer) as much as possible. */
 
@@ -3386,7 +3385,7 @@ buf_page_get_low(
 		/* Blocks cannot be relocated or enter or exit the
 		buf_pool while we are holding the buf_pool.mutex. */
 		const bool evicted = buf_LRU_free_page(&fix_block->page, true);
-		space->release_for_io();
+		space->release();
 
 		if (evicted) {
 			hash_lock = buf_pool.page_hash.lock_get(fold);
@@ -4108,7 +4107,7 @@ after decryption normal page checksum does not match.
 static dberr_t buf_page_check_corrupt(buf_page_t *bpage,
                                       const fil_node_t &node)
 {
-	ut_ad(node.space->pending_io());
+	ut_ad(node.space->referenced());
 
 	byte* dst_frame = (bpage->zip.data) ? bpage->zip.data :
 		((buf_block_t*) bpage)->frame;

storage/innobase/buf/buf0dblwr.cc

@@ -363,7 +363,7 @@ void buf_dblwr_t::recover()
       continue;
     }
 
-    fil_space_t *space= fil_space_t::get_for_io(space_id);
+    fil_space_t *space= fil_space_t::get(space_id);
 
     if (!space)
       /* The tablespace that this page once belonged to does not exist */
@@ -379,7 +379,7 @@ void buf_dblwr_t::recover()
                    << " is beyond the end of tablespace " << space->name
                    << " (" << space->size << " pages)";
 next_page:
-      space->release_for_io();
+      space->release();
       continue;
     }
 
@@ -420,7 +420,7 @@ void buf_dblwr_t::recover()
 
     /* Write the good page from the doublewrite buffer to the intended
     position. */
-    space->reacquire_for_io();
+    space->reacquire();
     fio= space->io(IORequestWrite,
                    os_offset_t{page_id.page_no()} * physical_size,
                    physical_size, page);
@@ -506,10 +506,10 @@ static void buf_dblwr_check_page_lsn(const page_t* page, const fil_space_t& s)
 
 static void buf_dblwr_check_page_lsn(const buf_page_t &b, const byte *page)
 {
-  if (fil_space_t *space= fil_space_t::get_for_io(b.id().space()))
+  if (fil_space_t *space= fil_space_t::get(b.id().space()))
   {
     buf_dblwr_check_page_lsn(page, *space);
-    space->release_for_io();
+    space->release();
   }
 }
 
@@ -583,7 +583,7 @@ bool buf_dblwr_t::flush_buffered_writes(const ulint size)
   }
 #endif /* UNIV_DEBUG */
   /* Write out the first block of the doublewrite buffer */
-  ut_a(fil_system.sys_space->acquire_for_io());
+  ut_a(fil_system.sys_space->acquire());
   fil_system.sys_space->io(IORequestWrite,
                            os_offset_t{block1.page_no()} <<
                            srv_page_size_shift,
@@ -593,7 +593,7 @@ bool buf_dblwr_t::flush_buffered_writes(const ulint size)
   if (old_first_free > size)
   {
     /* Write out the second block of the doublewrite buffer. */
-    ut_a(fil_system.sys_space->acquire_for_io());
+    ut_a(fil_system.sys_space->acquire());
     fil_system.sys_space->io(IORequestWrite,
                              os_offset_t{block2.page_no()} <<
                              srv_page_size_shift,
@@ -687,7 +687,7 @@ void buf_dblwr_t::add_to_batch(fil_space_t *space, const IORequest &request,
   ut_ad(request.bpage);
   ut_ad(request.bpage->in_file());
   ut_ad(space->id == request.bpage->id().space());
-  ut_ad(space->pending_io());
+  ut_ad(space->referenced());
   ut_ad(!srv_read_only_mode);
 
   const ulint buf_size= 2 * block_size();