MDEV-21774 Innodb, Windows : restore file sharing logic in Innodb

recv_sys_t opened redo log files along with log_sys_t. That's why I
removed file sharing logic from InnoDB
in 9ef2d29ff4
But it was actually used to ensure that only one MariaDB instance
will touch the same InnoDB files.

os0file.cc: revert some changes done previously

mapped_file_t::map(): now has arguments read_only, nvme

file_io::open(): now has argument read_only

class file_os_io: make final

log_file_t::open(): now has argument read_only

storage/innobase/include/log0log.h

@@ -468,7 +468,8 @@ public:
   mapped_file_t &operator=(mapped_file_t &&)= delete;
   ~mapped_file_t() noexcept;
 
-  dberr_t map(const char *path, int flags= 0) noexcept;
+  dberr_t map(const char *path, bool read_only= false,
+              bool nvme= false) noexcept;
   dberr_t unmap() noexcept;
   byte *data() noexcept { return m_area.data(); }
 
@@ -482,7 +483,7 @@ class file_io
 public:
   file_io(bool durable_writes= false) : m_durable_writes(durable_writes) {}
   virtual ~file_io() noexcept {};
-  virtual dberr_t open(const char *path) noexcept= 0;
+  virtual dberr_t open(const char *path, bool read_only) noexcept= 0;
   virtual dberr_t rename(const char *old_path,
                          const char *new_path) noexcept= 0;
   virtual dberr_t close() noexcept= 0;
@@ -498,7 +499,7 @@ protected:
   bool m_durable_writes;
 };
 
-class file_os_io : public file_io
+class file_os_io final: public file_io
 {
 public:
   file_os_io()= default;
@@ -508,7 +509,7 @@ public:
   file_os_io &operator=(file_os_io &&rhs);
   ~file_os_io() noexcept;
 
-  dberr_t open(const char *path) noexcept final;
+  dberr_t open(const char *path, bool read_only) noexcept final;
   bool is_opened() const noexcept { return m_fd != OS_FILE_CLOSED; }
   dberr_t rename(const char *old_path, const char *new_path) noexcept final;
   dberr_t close() noexcept final;
@@ -527,7 +528,7 @@ class log_file_t
 public:
   log_file_t(std::string path= "") noexcept : m_path{std::move(path)} {}
 
-  dberr_t open() noexcept;
+  dberr_t open(bool read_only) noexcept;
   bool is_opened() const noexcept;
 
   const std::string &get_path() const noexcept { return m_path; }

storage/innobase/log/log0log.cc

@@ -573,32 +573,27 @@ mapped_file_t::~mapped_file_t() noexcept
     unmap();
 }
 
-dberr_t mapped_file_t::map(const char *path, int flags) noexcept
+dberr_t mapped_file_t::map(const char *path, bool read_only,
+                           bool nvme) noexcept
 {
-  auto fd=
+  auto fd= mysql_file_open(innodb_log_file_key, path,
-      mysql_file_open(innodb_log_file_key, path,
+                           read_only ? O_RDONLY : O_RDWR, MYF(MY_WME));
-                      srv_read_only_mode ? O_RDONLY : O_RDWR, MYF(MY_WME));
-
   if (fd == -1)
     return DB_ERROR;
 
-  MY_STAT stat;
+  const auto file_size= os_file_get_size(path).m_total_size;
-  if (mysql_file_fstat(fd, &stat, MYF(0)))
-  {
-    mysql_file_close(fd, MYF(MY_WME));
-    return DB_ERROR;
-  }
 
-  void *ptr= my_mmap(0, static_cast<size_t>(stat.st_size),
+  const int nvme_flag= nvme ? MAP_SYNC : 0;
-                     srv_read_only_mode ? PROT_READ : PROT_READ | PROT_WRITE,
+  void *ptr= my_mmap(0, static_cast<size_t>(file_size),
-                     MAP_SHARED_VALIDATE | flags, fd, 0);
+                     read_only ? PROT_READ : PROT_READ | PROT_WRITE,
+                     MAP_SHARED_VALIDATE | nvme_flag, fd, 0);
   mysql_file_close(fd, MYF(MY_WME));
 
   if (ptr == MAP_FAILED)
     return DB_ERROR;
 
   m_area= {static_cast<byte *>(ptr),
-           static_cast<span<byte>::index_type>(stat.st_size)};
+           static_cast<span<byte>::index_type>(file_size)};
   return DB_SUCCESS;
 }
 
@@ -630,14 +625,14 @@ file_os_io::~file_os_io() noexcept
     close();
 }
 
-dberr_t file_os_io::open(const char *path) noexcept
+dberr_t file_os_io::open(const char *path, bool read_only) noexcept
 {
   ut_ad(!is_opened());
 
   bool success;
   auto tmp_fd= os_file_create(
       innodb_log_file_key, path, OS_FILE_OPEN | OS_FILE_ON_ERROR_NO_EXIT,
-      OS_FILE_NORMAL, OS_LOG_FILE, srv_read_only_mode, &success);
+      OS_FILE_NORMAL, OS_LOG_FILE, read_only, &success);
   if (!success)
     return DB_ERROR;
 
@@ -685,7 +680,7 @@ dberr_t file_os_io::flush_data_only() noexcept
 static bool is_pmem(const char *path) noexcept
 {
   mapped_file_t mf;
-  return mf.map(path, MAP_SYNC) == DB_SUCCESS ? true : false;
+  return mf.map(path, true, true) == DB_SUCCESS ? true : false;
 }
 
 class file_pmem_io final : public file_io
@@ -693,9 +688,9 @@ class file_pmem_io final : public file_io
 public:
   file_pmem_io() noexcept : file_io(true) {}
 
-  dberr_t open(const char *path) noexcept final
+  dberr_t open(const char *path, bool read_only) noexcept final
   {
-    return m_file.map(path, MAP_SYNC);
+    return m_file.map(path, read_only, true);
   }
   dberr_t rename(const char *old_path, const char *new_path) noexcept final
   {
@@ -725,7 +720,7 @@ private:
 };
 #endif
 
-dberr_t log_file_t::open() noexcept
+dberr_t log_file_t::open(bool read_only) noexcept
 {
   ut_a(!is_opened());
 
@@ -737,7 +732,7 @@ dberr_t log_file_t::open() noexcept
   auto ptr= std::unique_ptr<file_io>(new file_os_io);
 #endif
 
-  if (dberr_t err= ptr->open(m_path.c_str()))
+  if (dberr_t err= ptr->open(m_path.c_str(), read_only))
     return err;
 
   m_file= std::move(ptr);
@@ -795,7 +790,7 @@ dberr_t log_file_t::flush_data_only() noexcept
 void log_t::file::open_file(std::string path)
 {
   fd= log_file_t(std::move(path));
-  if (const dberr_t err= fd.open())
+  if (const dberr_t err= fd.open(srv_read_only_mode))
     ib::fatal() << "open(" << fd.get_path() << ") returned " << err;
 }
 

storage/innobase/log/log0recv.cc

@@ -598,7 +598,7 @@ void recv_sys_t::open_log_files_if_needed()
   for (auto &&path : get_existing_log_files_paths())
   {
     recv_sys.files.emplace_back(std::move(path));
-    ut_a(recv_sys.files.back().open() == DB_SUCCESS);
+    ut_a(recv_sys.files.back().open(true) == DB_SUCCESS);
   }
 }
 

storage/innobase/os/os0file.cc

@@ -2136,7 +2136,7 @@ os_file_create_simple_func(
 
 		file = CreateFile(
 			(LPCTSTR) name, access,
-			FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
+			FILE_SHARE_READ | FILE_SHARE_DELETE,
 			NULL, create_flag, attributes, NULL);
 
 		if (file == INVALID_HANDLE_VALUE) {
@@ -2404,8 +2404,9 @@ os_file_create_func(
 	);
 
 	DWORD		create_flag;
-	const DWORD	share_mode =
+	DWORD		share_mode = read_only
-		FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
+		? FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE
+		: FILE_SHARE_READ | FILE_SHARE_DELETE;
 
 	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW) {
 		WAIT_ALLOW_WRITES();
@@ -2423,6 +2424,12 @@ os_file_create_func(
 
 		ut_a(!read_only);
 
+		/* On Windows Physical devices require admin privileges and
+		have to have the write-share mode set. See the remarks
+		section for the CreateFile() function documentation in MSDN. */
+
+		share_mode |= FILE_SHARE_WRITE;
+
 		create_flag = OPEN_EXISTING;
 
 	} else if (create_mode == OS_FILE_OPEN
@@ -2605,8 +2612,9 @@ os_file_create_simple_no_error_handling_func(
 	DWORD		access;
 	DWORD		create_flag;
 	DWORD		attributes	= 0;
-	const DWORD	share_mode =
+	DWORD		share_mode = read_only
-		FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
+		? FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE
+		: FILE_SHARE_READ | FILE_SHARE_DELETE;
 
 	ut_a(name);
 
@@ -2652,6 +2660,12 @@ os_file_create_simple_no_error_handling_func(
 
 		access = GENERIC_READ;
 
+		/*!< A backup program has to give mysqld the maximum
+		freedom to do what it likes with the file */
+
+		share_mode |= FILE_SHARE_DELETE | FILE_SHARE_WRITE
+			| FILE_SHARE_READ;
+
 	} else {
 
 		ib::error()