diff options
Diffstat (limited to 'storage/innobase/os/os0file.cc')
| -rw-r--r-- | storage/innobase/os/os0file.cc | 1925 |
1 files changed, 1133 insertions, 792 deletions
diff --git a/storage/innobase/os/os0file.cc b/storage/innobase/os/os0file.cc index 49d6b00d271..b4e4f52a0f7 100644 --- a/storage/innobase/os/os0file.cc +++ b/storage/innobase/os/os0file.cc @@ -1,6 +1,6 @@ /*********************************************************************** -Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved. +Copyright (c) 1995, 2012, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 2009, Percona Inc. Portions of this file contain modifications contributed and copyrighted @@ -60,24 +60,29 @@ Created 10/21/1995 Heikki Tuuri #include <libaio.h> #endif +/** Insert buffer segment id */ +static const ulint IO_IBUF_SEGMENT = 0; + +/** Log segment id */ +static const ulint IO_LOG_SEGMENT = 1; + /* This specifies the file permissions InnoDB uses when it creates files in Unix; the value of os_innodb_umask is initialized in ha_innodb.cc to my_umask */ #ifndef __WIN__ /** Umask for creating files */ -UNIV_INTERN ulint os_innodb_umask - = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP; +UNIV_INTERN ulint os_innodb_umask = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP; #else /** Umask for creating files */ -UNIV_INTERN ulint os_innodb_umask = 0; -#endif +UNIV_INTERN ulint os_innodb_umask = 0; +#endif /* __WIN__ */ #ifndef UNIV_HOTBACKUP /* We use these mutexes to protect lseek + file i/o operation, if the OS does not provide an atomic pread or pwrite, or similar */ #define OS_FILE_N_SEEK_MUTEXES 16 -UNIV_INTERN os_mutex_t os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES]; +UNIV_INTERN os_ib_mutex_t os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES]; /* In simulated aio, merge at most this many consecutive i/os */ #define OS_AIO_MERGE_N_CONSECUTIVE 64 @@ -147,10 +152,7 @@ UNIV_INTERN mysql_pfs_key_t innodb_file_temp_key; #endif /* UNIV_PFS_IO */ /** The asynchronous i/o array slot structure */ -typedef struct os_aio_slot_struct os_aio_slot_t; - -/** The asynchronous i/o array slot structure */ -struct os_aio_slot_struct{ +struct os_aio_slot_t{ ibool is_read; /*!< TRUE if a read operation */ ulint pos; /*!< index of the slot in the aio array */ @@ -182,15 +184,12 @@ struct os_aio_slot_struct{ struct iocb control; /* Linux control block for aio */ int n_bytes; /* bytes written/read. */ int ret; /* AIO return code */ -#endif +#endif /* WIN_ASYNC_IO */ }; /** The asynchronous i/o array structure */ -typedef struct os_aio_array_struct os_aio_array_t; - -/** The asynchronous i/o array structure */ -struct os_aio_array_struct{ - os_mutex_t mutex; /*!< the mutex protecting the aio array */ +struct os_aio_array_t{ + os_ib_mutex_t mutex; /*!< the mutex protecting the aio array */ os_event_t not_full; /*!< The event which is set to the signaled state when there is space in @@ -223,7 +222,7 @@ struct os_aio_array_struct{ order. This can be used in WaitForMultipleObjects; used only in Windows */ -#endif +#endif /* __WIN__ */ #if defined(LINUX_NATIVE_AIO) io_context_t* aio_ctx; @@ -235,7 +234,7 @@ struct os_aio_array_struct{ There is one such event for each possible pending IO. The size of the array is equal to n_slots. */ -#endif +#endif /* LINUX_NATIV_AIO */ }; #if defined(LINUX_NATIVE_AIO) @@ -283,7 +282,7 @@ UNIV_INTERN ibool os_has_said_disk_full = FALSE; #if !defined(UNIV_HOTBACKUP) \ && (!defined(HAVE_ATOMIC_BUILTINS) || UNIV_WORD_SIZE < 8) /** The mutex protecting the following counts of pending I/O operations */ -static os_mutex_t os_file_count_mutex; +static os_ib_mutex_t os_file_count_mutex; #endif /* !UNIV_HOTBACKUP && (!HAVE_ATOMIC_BUILTINS || UNIV_WORD_SIZE < 8) */ /** Number of pending os_file_pread() operations */ @@ -336,7 +335,7 @@ ulint os_get_os_version(void) /*===================*/ { - OSVERSIONINFO os_info; + OSVERSIONINFO os_info; os_info.dwOSVersionInfoSize = sizeof(OSVERSIONINFO); @@ -350,15 +349,15 @@ os_get_os_version(void) switch (os_info.dwMajorVersion) { case 3: case 4: - return OS_WINNT; + return(OS_WINNT); case 5: - return (os_info.dwMinorVersion == 0) ? OS_WIN2000 - : OS_WINXP; + return (os_info.dwMinorVersion == 0) + ? OS_WIN2000 : OS_WINXP; case 6: - return (os_info.dwMinorVersion == 0) ? OS_WINVISTA - : OS_WIN7; + return (os_info.dwMinorVersion == 0) + ? OS_WINVISTA : OS_WIN7; default: - return OS_WIN7; + return(OS_WIN7); } } else { ut_error; @@ -377,16 +376,17 @@ static ulint os_file_get_last_error_low( /*=======================*/ - ibool report_all_errors, /*!< in: TRUE if we want an error + bool report_all_errors, /*!< in: TRUE if we want an error message printed of all errors */ - ibool on_error_silent) /*!< in: TRUE then don't print any + bool on_error_silent) /*!< in: TRUE then don't print any diagnostic to the log */ { - ulint err; - #ifdef __WIN__ - err = (ulint) GetLastError(); + ulint err = (ulint) GetLastError(); + if (err == ERROR_SUCCESS) { + return(0); + } if (report_all_errors || (!on_error_silent @@ -469,15 +469,18 @@ os_file_get_last_error_low( return(100 + err); } #else - err = (ulint) errno; + int err = errno; + if (err == 0) { + return(0); + } if (report_all_errors || (err != ENOSPC && err != EEXIST && !on_error_silent)) { ut_print_timestamp(stderr); fprintf(stderr, - " InnoDB: Operating system error number %lu" - " in a file operation.\n", (ulong) err); + " InnoDB: Operating system error number %d" + " in a file operation.\n", err); if (err == ENOENT) { fprintf(stderr, @@ -497,11 +500,11 @@ os_file_get_last_error_low( " the access rights to\n" "InnoDB: the directory.\n"); } else { - if (strerror((int) err) != NULL) { + if (strerror(err) != NULL) { fprintf(stderr, - "InnoDB: Error number %lu" + "InnoDB: Error number %d" " means '%s'.\n", - err, strerror((int) err)); + err, strerror(err)); } @@ -552,10 +555,10 @@ UNIV_INTERN ulint os_file_get_last_error( /*===================*/ - ibool report_all_errors) /*!< in: TRUE if we want an error + bool report_all_errors) /*!< in: TRUE if we want an error message printed of all errors */ { - return(os_file_get_last_error_low(report_all_errors, FALSE)); + return(os_file_get_last_error_low(report_all_errors, false)); } /****************************************************************//** @@ -577,7 +580,7 @@ os_file_handle_error_cond_exit( { ulint err; - err = os_file_get_last_error_low(FALSE, on_error_silent); + err = os_file_get_last_error_low(false, on_error_silent); switch (err) { case OS_FILE_DISK_FULL: @@ -645,7 +648,8 @@ os_file_handle_error_cond_exit( ut_print_timestamp(stderr); fprintf(stderr, " InnoDB: File operation call: " - "'%s'.\n", operation); + "'%s' returned OS error " ULINTPF ".\n", + operation, err); } if (should_exit) { @@ -654,7 +658,9 @@ os_file_handle_error_cond_exit( "operation.\n"); fflush(stderr); - ut_error; + + ut_ad(0); /* Report call stack, etc only in debug code. */ + exit(1); } } @@ -712,19 +718,23 @@ os_file_lock( const char* name) /*!< in: file name */ { struct flock lk; + + ut_ad(!srv_read_only_mode); + lk.l_type = F_WRLCK; lk.l_whence = SEEK_SET; lk.l_start = lk.l_len = 0; + if (fcntl(fd, F_SETLK, &lk) == -1) { - fprintf(stderr, - "InnoDB: Unable to lock %s, error: %d\n", name, errno); + + ib_logf(IB_LOG_LEVEL_ERROR, + "Unable to lock %s, error: %d", name, errno); if (errno == EAGAIN || errno == EACCES) { - fprintf(stderr, - "InnoDB: Check that you do not already have" - " another mysqld process\n" - "InnoDB: using the same InnoDB data" - " or log files.\n"); + ib_logf(IB_LOG_LEVEL_INFO, + "Check that you do not already have " + "another mysqld process using the " + "same InnoDB data or log files."); } return(-1); @@ -742,13 +752,11 @@ void os_io_init_simple(void) /*===================*/ { - ulint i; - #if !defined(HAVE_ATOMIC_BUILTINS) || UNIV_WORD_SIZE < 8 os_file_count_mutex = os_mutex_create(); #endif /* !HAVE_ATOMIC_BUILTINS || UNIV_WORD_SIZE < 8 */ - for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) { + for (ulint i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) { os_file_seek_mutexes[i] = os_mutex_create(); } } @@ -765,6 +773,8 @@ os_file_create_tmpfile(void) FILE* file = NULL; int fd = innobase_mysql_tmpfile(); + ut_ad(!srv_read_only_mode); + if (fd >= 0) { file = fdopen(fd, "w+b"); } @@ -840,7 +850,7 @@ os_file_opendir( } return(dir); -#endif +#endif /* __WIN__ */ } /***********************************************************************//** @@ -874,7 +884,7 @@ os_file_closedir( } return(ret); -#endif +#endif /* __WIN__ */ } /***********************************************************************//** @@ -1054,10 +1064,12 @@ next_file: } /*****************************************************************//** -This function attempts to create a directory named pathname. The new directory -gets default permissions. On Unix the permissions are (0770 & ~umask). If the -directory exists already, nothing is done and the call succeeds, unless the -fail_if_exists arguments is true. +This function attempts to create a directory named pathname. The new +directory gets default permissions. On Unix the permissions are +(0770 & ~umask). If the directory exists already, nothing is done and +the call succeeds, unless the fail_if_exists arguments is true. +If another error occurs, such as a permission error, this does not crash, +but reports the error and returns FALSE. @return TRUE if call succeeds, FALSE on error */ UNIV_INTERN ibool @@ -1075,13 +1087,14 @@ os_file_create_directory( if (!(rcode != 0 || (GetLastError() == ERROR_ALREADY_EXISTS && !fail_if_exists))) { - /* failure */ - os_file_handle_error(pathname, "CreateDirectory"); + + os_file_handle_error_no_exit( + pathname, "CreateDirectory", FALSE); return(FALSE); } - return (TRUE); + return(TRUE); #else int rcode; @@ -1089,13 +1102,13 @@ os_file_create_directory( if (!(rcode == 0 || (errno == EEXIST && !fail_if_exists))) { /* failure */ - os_file_handle_error(pathname, "mkdir"); + os_file_handle_error_no_exit(pathname, "mkdir", FALSE); return(FALSE); } return (TRUE); -#endif +#endif /* __WIN__ */ } /****************************************************************//** @@ -1115,129 +1128,180 @@ os_file_create_simple_func( OS_FILE_READ_WRITE */ ibool* success)/*!< out: TRUE if succeed, FALSE if error */ { -#ifdef __WIN__ os_file_t file; - DWORD create_flag; + ibool retry; + +#ifdef __WIN__ DWORD access; + DWORD create_flag; DWORD attributes = 0; - ibool retry; ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT)); ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT)); -try_again: - ut_a(name); if (create_mode == OS_FILE_OPEN) { + + create_flag = OPEN_EXISTING; + + } else if (srv_read_only_mode) { + create_flag = OPEN_EXISTING; + } else if (create_mode == OS_FILE_CREATE) { + create_flag = CREATE_NEW; + } else if (create_mode == OS_FILE_CREATE_PATH) { - /* create subdirs along the path if needed */ + + ut_a(!srv_read_only_mode); + + /* Create subdirs along the path if needed */ *success = os_file_create_subdirs_if_needed(name); + if (!*success) { - ut_error; + + ib_logf(IB_LOG_LEVEL_ERROR, + "Unable to create subdirectories '%s'", + name); + + return((os_file_t) -1); } + create_flag = CREATE_NEW; create_mode = OS_FILE_CREATE; + } else { - create_flag = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } if (access_type == OS_FILE_READ_ONLY) { access = GENERIC_READ; + } else if (srv_read_only_mode) { + + ib_logf(IB_LOG_LEVEL_INFO, + "read only mode set. Unable to " + "open file '%s' in RW mode, trying RO mode", name); + + access = GENERIC_READ; + } else if (access_type == OS_FILE_READ_WRITE) { access = GENERIC_READ | GENERIC_WRITE; } else { - access = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file access type (%lu) for file '%s'", + access_type, name); + + return((os_file_t) -1); } - file = CreateFile((LPCTSTR) name, - access, - FILE_SHARE_READ | FILE_SHARE_WRITE, - /* file can be read and written also - by other processes */ - NULL, /* default security attributes */ - create_flag, - attributes, - NULL); /*!< no template file */ + do { + /* Use default security attributes and no template file. */ - if (file == INVALID_HANDLE_VALUE) { - *success = FALSE; + file = CreateFile( + (LPCTSTR) name, access, FILE_SHARE_READ, NULL, + create_flag, attributes, NULL); + + if (file == INVALID_HANDLE_VALUE) { - retry = os_file_handle_error(name, - create_mode == OS_FILE_OPEN ? - "open" : "create"); - if (retry) { - goto try_again; + *success = FALSE; + + retry = os_file_handle_error( + name, create_mode == OS_FILE_OPEN ? + "open" : "create"); + + } else { + *success = TRUE; + retry = false; } - } else { - *success = TRUE; - } - return(file); + } while (retry); + #else /* __WIN__ */ - os_file_t file; int create_flag; - ibool retry; ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT)); ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT)); -try_again: - ut_a(name); - if (create_mode == OS_FILE_OPEN) { + if (access_type == OS_FILE_READ_ONLY) { create_flag = O_RDONLY; + } else if (srv_read_only_mode) { + create_flag = O_RDONLY; } else { create_flag = O_RDWR; } + + } else if (srv_read_only_mode) { + + create_flag = O_RDONLY; + } else if (create_mode == OS_FILE_CREATE) { + create_flag = O_RDWR | O_CREAT | O_EXCL; + } else if (create_mode == OS_FILE_CREATE_PATH) { - /* create subdirs along the path if needed */ + + /* Create subdirs along the path if needed */ + *success = os_file_create_subdirs_if_needed(name); + if (!*success) { - return (-1); + + ib_logf(IB_LOG_LEVEL_ERROR, + "Unable to create subdirectories '%s'", + name); + + return((os_file_t) -1); } + create_flag = O_RDWR | O_CREAT | O_EXCL; create_mode = OS_FILE_CREATE; } else { - create_flag = 0; - ut_error; - } - if (create_mode == OS_FILE_CREATE) { - file = open(name, create_flag, S_IRUSR | S_IWUSR - | S_IRGRP | S_IWGRP); - } else { - file = open(name, create_flag); + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } - if (file == -1) { - *success = FALSE; + do { + file = ::open(name, create_flag, os_innodb_umask); - retry = os_file_handle_error(name, - create_mode == OS_FILE_OPEN ? - "open" : "create"); - if (retry) { - goto try_again; + if (file == -1) { + *success = FALSE; + + retry = os_file_handle_error( + name, + create_mode == OS_FILE_OPEN + ? "open" : "create"); + } else { + *success = TRUE; + retry = false; } + + } while (retry); + #ifdef USE_FILE_LOCK - } else if (access_type == OS_FILE_READ_WRITE - && os_file_lock(file, name)) { + if (!srv_read_only_mode + && *success + && access_type == OS_FILE_READ_WRITE + && os_file_lock(file, name)) { + *success = FALSE; close(file); file = -1; -#endif - } else { - *success = TRUE; } +#endif /* USE_FILE_LOCK */ - return(file); #endif /* __WIN__ */ + + return(file); } /****************************************************************//** @@ -1259,12 +1323,13 @@ os_file_create_simple_no_error_handling_func( used by a backup program reading the file */ ibool* success)/*!< out: TRUE if succeed, FALSE if error */ { -#ifdef __WIN__ os_file_t file; - DWORD create_flag; + +#ifdef __WIN__ DWORD access; + DWORD create_flag; DWORD attributes = 0; - DWORD share_mode = FILE_SHARE_READ | FILE_SHARE_WRITE; + DWORD share_mode = FILE_SHARE_READ; ut_a(name); @@ -1273,46 +1338,53 @@ os_file_create_simple_no_error_handling_func( if (create_mode == OS_FILE_OPEN) { create_flag = OPEN_EXISTING; + } else if (srv_read_only_mode) { + create_flag = OPEN_EXISTING; } else if (create_mode == OS_FILE_CREATE) { create_flag = CREATE_NEW; } else { - create_flag = 0; - ut_error; + + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } if (access_type == OS_FILE_READ_ONLY) { access = GENERIC_READ; + } else if (srv_read_only_mode) { + access = GENERIC_READ; } else if (access_type == OS_FILE_READ_WRITE) { access = GENERIC_READ | GENERIC_WRITE; } else if (access_type == OS_FILE_READ_ALLOW_DELETE) { + + ut_a(!srv_read_only_mode); + access = GENERIC_READ; - share_mode = FILE_SHARE_DELETE | FILE_SHARE_READ - | FILE_SHARE_WRITE; /*!< A backup program has to give - mysqld the maximum freedom to - do what it likes with the - file */ + + /*!< 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; } else { - access = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file access type (%lu) for file '%s'", + access_type, name); + + return((os_file_t) -1); } file = CreateFile((LPCTSTR) name, access, share_mode, - NULL, /* default security attributes */ + NULL, // Security attributes create_flag, attributes, - NULL); /*!< no template file */ - - if (file == INVALID_HANDLE_VALUE) { - *success = FALSE; - } else { - *success = TRUE; - } + NULL); // No template file - return(file); + *success = (file != INVALID_HANDLE_VALUE); #else /* __WIN__ */ - os_file_t file; int create_flag; ut_a(name); @@ -1321,40 +1393,59 @@ os_file_create_simple_no_error_handling_func( ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT)); if (create_mode == OS_FILE_OPEN) { + if (access_type == OS_FILE_READ_ONLY) { + create_flag = O_RDONLY; + + } else if (srv_read_only_mode) { + + create_flag = O_RDONLY; + } else { + + ut_a(access_type == OS_FILE_READ_WRITE + || access_type == OS_FILE_READ_ALLOW_DELETE); + create_flag = O_RDWR; } + + } else if (srv_read_only_mode) { + + create_flag = O_RDONLY; + } else if (create_mode == OS_FILE_CREATE) { + create_flag = O_RDWR | O_CREAT | O_EXCL; - } else { - create_flag = 0; - ut_error; - } - if (create_mode == OS_FILE_CREATE) { - file = open(name, create_flag, S_IRUSR | S_IWUSR - | S_IRGRP | S_IWGRP); } else { - file = open(name, create_flag); + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } - if (file == -1) { - *success = FALSE; + file = ::open(name, create_flag, os_innodb_umask); + + *success = file == -1 ? FALSE : TRUE; + #ifdef USE_FILE_LOCK - } else if (access_type == OS_FILE_READ_WRITE - && os_file_lock(file, name)) { + if (!srv_read_only_mode + && *success + && access_type == OS_FILE_READ_WRITE + && os_file_lock(file, name)) { + *success = FALSE; close(file); file = -1; -#endif - } else { - *success = TRUE; + } +#endif /* USE_FILE_LOCK */ - return(file); #endif /* __WIN__ */ + + return(file); } /****************************************************************//** @@ -1364,80 +1455,43 @@ void os_file_set_nocache( /*================*/ int fd /*!< in: file descriptor to alter */ - __attribute__((unused)), - const char* file_name /*!< in: used in the diagnostic message */ - __attribute__((unused)), + __attribute__((unused)), + const char* file_name /*!< in: used in the diagnostic + message */ + __attribute__((unused)), const char* operation_name __attribute__((unused))) - /*!< in: "open" or "create"; used in the - diagnostic message */ + /*!< in: "open" or "create"; used + in the diagnostic message */ { /* some versions of Solaris may not have DIRECTIO_ON */ #if defined(UNIV_SOLARIS) && defined(DIRECTIO_ON) if (directio(fd, DIRECTIO_ON) == -1) { - int errno_save; - errno_save = (int) errno; - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: Failed to set DIRECTIO_ON " - "on file %s: %s: %s, continuing anyway\n", + int errno_save = errno; + + ib_logf(IB_LOG_LEVEL_ERROR, + "Failed to set DIRECTIO_ON on file %s: %s: %s, " + "continuing anyway.", file_name, operation_name, strerror(errno_save)); } #elif defined(O_DIRECT) if (fcntl(fd, F_SETFL, O_DIRECT) == -1) { - int errno_save; - errno_save = (int) errno; - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: Failed to set O_DIRECT " - "on file %s: %s: %s, continuing anyway\n", + int errno_save = errno; + + ib_logf(IB_LOG_LEVEL_ERROR, + "Failed to set O_DIRECT on file %s: %s: %s, " + "continuing anyway", file_name, operation_name, strerror(errno_save)); + if (errno_save == EINVAL) { - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: O_DIRECT is known to result in " - "'Invalid argument' on Linux on tmpfs, " - "see MySQL Bug#26662\n"); + ib_logf(IB_LOG_LEVEL_ERROR, + "O_DIRECT is known to result in 'Invalid " + "argument' on Linux on tmpfs, see MySQL " + "Bug#26662"); } } -#endif +#endif /* defined(UNIV_SOLARIS) && defined(DIRECTIO_ON) */ } - -#ifdef __linux__ -#include <sys/ioctl.h> -#ifndef DFS_IOCTL_ATOMIC_WRITE_SET -#define DFS_IOCTL_ATOMIC_WRITE_SET _IOW(0x95, 2, uint) -#endif -static int os_file_set_atomic_writes(os_file_t file, const char *name) -{ - int atomic_option = 1; - - int ret = ioctl (file, DFS_IOCTL_ATOMIC_WRITE_SET, &atomic_option); - - if (ret) { - fprintf(stderr, - "InnoDB : can't use atomic write on %s, errno %d\n", - name, errno); - return ret; - } - return ret; -} -#else -static int os_file_set_atomic_writes(os_file_t file, const char *name) -{ - fprintf(stderr, - "InnoDB : can't use atomic writes on %s - not implemented on this platform." - "innodb_use_atomic_writes needs to be 0.\n", - name); -#ifdef _WIN32 - SetLastError(ERROR_INVALID_FUNCTION); -#else - errno = EINVAL; -#endif - return -1; -} -#endif - /****************************************************************//** NOTE! Use the corresponding macro os_file_create(), not directly this function! @@ -1461,145 +1515,155 @@ os_file_create_func( ulint type, /*!< in: OS_DATA_FILE or OS_LOG_FILE */ ibool* success)/*!< out: TRUE if succeed, FALSE if error */ { + os_file_t file; + ibool retry; ibool on_error_no_exit; ibool on_error_silent; #ifdef __WIN__ - os_file_t file; - DWORD share_mode = FILE_SHARE_READ; + DBUG_EXECUTE_IF( + "ib_create_table_fail_disk_full", + *success = FALSE; + SetLastError(ERROR_DISK_FULL); + return((os_file_t) -1); + ); +#else /* __WIN__ */ + DBUG_EXECUTE_IF( + "ib_create_table_fail_disk_full", + *success = FALSE; + errno = ENOSPC; + return((os_file_t) -1); + ); +#endif /* __WIN__ */ + +#ifdef __WIN__ DWORD create_flag; - DWORD attributes; - ibool retry; + DWORD share_mode = FILE_SHARE_READ; on_error_no_exit = create_mode & OS_FILE_ON_ERROR_NO_EXIT ? TRUE : FALSE; + on_error_silent = create_mode & OS_FILE_ON_ERROR_SILENT ? TRUE : FALSE; create_mode &= ~OS_FILE_ON_ERROR_NO_EXIT; create_mode &= ~OS_FILE_ON_ERROR_SILENT; + if (create_mode == OS_FILE_OPEN_RAW) { - DBUG_EXECUTE_IF( - "ib_create_table_fail_disk_full", - *success = FALSE; - SetLastError(ERROR_DISK_FULL); - return((os_file_t) -1); - ); -try_again: - ut_a(name); + ut_a(!srv_read_only_mode); - if (create_mode == OS_FILE_OPEN_RAW) { create_flag = OPEN_EXISTING; - share_mode = FILE_SHARE_WRITE; + + /* 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; + } else if (create_mode == OS_FILE_OPEN || create_mode == OS_FILE_OPEN_RETRY) { + create_flag = OPEN_EXISTING; + + } else if (srv_read_only_mode) { + + create_flag = OPEN_EXISTING; + } else if (create_mode == OS_FILE_CREATE) { + create_flag = CREATE_NEW; + } else if (create_mode == OS_FILE_OVERWRITE) { + create_flag = CREATE_ALWAYS; + } else { - create_flag = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } + DWORD attributes = 0; + +#ifdef UNIV_HOTBACKUP + attributes |= FILE_FLAG_NO_BUFFERING; +#else if (purpose == OS_FILE_AIO) { + +#ifdef WIN_ASYNC_IO /* If specified, use asynchronous (overlapped) io and no buffering of writes in the OS */ - attributes = 0; -#ifdef WIN_ASYNC_IO + if (srv_use_native_aio) { - attributes = attributes | FILE_FLAG_OVERLAPPED; + attributes |= FILE_FLAG_OVERLAPPED; } -#endif -#ifdef UNIV_NON_BUFFERED_IO -# ifndef UNIV_HOTBACKUP - if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) { - /* Do not use unbuffered i/o to log files because - value 2 denotes that we do not flush the log at every - commit, but only once per second */ - } else if (srv_win_file_flush_method - == SRV_WIN_IO_UNBUFFERED) { - attributes = attributes | FILE_FLAG_NO_BUFFERING; - } -# else /* !UNIV_HOTBACKUP */ - attributes = attributes | FILE_FLAG_NO_BUFFERING; -# endif /* !UNIV_HOTBACKUP */ -#endif /* UNIV_NON_BUFFERED_IO */ +#endif /* WIN_ASYNC_IO */ + } else if (purpose == OS_FILE_NORMAL) { - attributes = 0; -#ifdef UNIV_NON_BUFFERED_IO -# ifndef UNIV_HOTBACKUP - if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) { - /* Do not use unbuffered i/o to log files because - value 2 denotes that we do not flush the log at every - commit, but only once per second */ - } else if (srv_win_file_flush_method - == SRV_WIN_IO_UNBUFFERED) { - attributes = attributes | FILE_FLAG_NO_BUFFERING; - } -# else /* !UNIV_HOTBACKUP */ - attributes = attributes | FILE_FLAG_NO_BUFFERING; -# endif /* !UNIV_HOTBACKUP */ -#endif /* UNIV_NON_BUFFERED_IO */ + /* Use default setting. */ } else { - attributes = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown purpose flag (%lu) while opening file '%s'", + purpose, name); + + return((os_file_t)(-1)); } - file = CreateFile((LPCTSTR) name, - GENERIC_READ | GENERIC_WRITE, /* read and write - access */ - share_mode, /* File can be read also by other - processes; we must give the read - permission because of ibbackup. We do - not give the write permission to - others because if one would succeed to - start 2 instances of mysqld on the - SAME files, that could cause severe - database corruption! When opening - raw disk partitions, Microsoft manuals - say that we must give also the write - permission. */ - NULL, /* default security attributes */ - create_flag, - attributes, - NULL); /*!< no template file */ +#ifdef UNIV_NON_BUFFERED_IO + // TODO: Create a bug, this looks wrong. The flush log + // parameter is dynamic. + if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) { - if (file == INVALID_HANDLE_VALUE) { - const char* operation; + /* Do not use unbuffered i/o for the log files because + value 2 denotes that we do not flush the log at every + commit, but only once per second */ - operation = create_mode == OS_FILE_CREATE ? "create" : "open"; + } else if (srv_win_file_flush_method == SRV_WIN_IO_UNBUFFERED) { - *success = FALSE; + attributes |= FILE_FLAG_NO_BUFFERING; + } +#endif /* UNIV_NON_BUFFERED_IO */ - if (on_error_no_exit) { - retry = os_file_handle_error_no_exit( - name, operation, on_error_silent); - } else { - retry = os_file_handle_error(name, operation); - } +#endif /* UNIV_HOTBACKUP */ + DWORD access = GENERIC_READ; - if (retry) { - goto try_again; - } - } else { - *success = TRUE; + if (!srv_read_only_mode) { + access |= GENERIC_WRITE; } - if (srv_use_atomic_writes && type == OS_DATA_FILE && - os_file_set_atomic_writes(file, name)) { - CloseHandle(file); + do { + /* Use default security attributes and no template file. */ + file = CreateFile( + (LPCTSTR) name, access, share_mode, NULL, + create_flag, attributes, NULL); + + if (file == INVALID_HANDLE_VALUE) { + const char* operation; + + operation = (create_mode == OS_FILE_CREATE + && !srv_read_only_mode) + ? "create" : "open"; + *success = FALSE; - file = INVALID_HANDLE_VALUE; - } - return(file); + if (on_error_no_exit) { + retry = os_file_handle_error_no_exit( + name, operation, on_error_silent); + } else { + retry = os_file_handle_error(name, operation); + } + } else { + *success = TRUE; + retry = FALSE; + } + + } while (retry); + #else /* __WIN__ */ - os_file_t file; int create_flag; - ibool retry; const char* mode_str = NULL; on_error_no_exit = create_mode & OS_FILE_ON_ERROR_NO_EXIT @@ -1610,28 +1674,36 @@ try_again: create_mode &= ~OS_FILE_ON_ERROR_NO_EXIT; create_mode &= ~OS_FILE_ON_ERROR_SILENT; - DBUG_EXECUTE_IF( - "ib_create_table_fail_disk_full", - *success = FALSE; - errno = ENOSPC; - return((os_file_t) -1); - ); -try_again: - ut_a(name); - - if (create_mode == OS_FILE_OPEN || create_mode == OS_FILE_OPEN_RAW + if (create_mode == OS_FILE_OPEN + || create_mode == OS_FILE_OPEN_RAW || create_mode == OS_FILE_OPEN_RETRY) { + + mode_str = "OPEN"; + + create_flag = srv_read_only_mode ? O_RDONLY : O_RDWR; + + } else if (srv_read_only_mode) { + mode_str = "OPEN"; - create_flag = O_RDWR; + + create_flag = O_RDONLY; + } else if (create_mode == OS_FILE_CREATE) { + mode_str = "CREATE"; create_flag = O_RDWR | O_CREAT | O_EXCL; + } else if (create_mode == OS_FILE_OVERWRITE) { + mode_str = "OVERWRITE"; create_flag = O_RDWR | O_CREAT | O_TRUNC; + } else { - create_flag = 0; - ut_error; + ib_logf(IB_LOG_LEVEL_ERROR, + "Unknown file create mode (%lu) for file '%s'", + create_mode, name); + + return((os_file_t) -1); } ut_a(type == OS_LOG_FILE || type == OS_DATA_FILE); @@ -1641,69 +1713,75 @@ try_again: /* We let O_SYNC only affect log files; note that we map O_DSYNC to O_SYNC because the datasync options seemed to corrupt files in 2001 in both Linux and Solaris */ - if (type == OS_LOG_FILE - && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) { -# if 0 - fprintf(stderr, "Using O_SYNC for file %s\n", name); -# endif + if (!srv_read_only_mode + && type == OS_LOG_FILE + && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) { - create_flag = create_flag | O_SYNC; + create_flag |= O_SYNC; } #endif /* O_SYNC */ - file = open(name, create_flag, os_innodb_umask); - - if (file == -1) { - const char* operation; + do { + file = ::open(name, create_flag, os_innodb_umask); - operation = create_mode == OS_FILE_CREATE ? "create" : "open"; + if (file == -1) { + const char* operation; - *success = FALSE; + operation = (create_mode == OS_FILE_CREATE + && !srv_read_only_mode) + ? "create" : "open"; - if (on_error_no_exit) { - retry = os_file_handle_error_no_exit( - name, operation, on_error_silent); - } else { - retry = os_file_handle_error(name, operation); - } + *success = FALSE; - if (retry) { - goto try_again; + if (on_error_no_exit) { + retry = os_file_handle_error_no_exit( + name, operation, on_error_silent); + } else { + retry = os_file_handle_error(name, operation); + } } else { - return(file /* -1 */); + *success = TRUE; + retry = false; } - } - /* else */ - *success = TRUE; + } while (retry); /* We disable OS caching (O_DIRECT) only on data files */ - if (type != OS_LOG_FILE - && srv_unix_file_flush_method == SRV_UNIX_O_DIRECT) { + + if (!srv_read_only_mode + && *success + && type != OS_LOG_FILE + && (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT + || srv_unix_file_flush_method == SRV_UNIX_O_DIRECT_NO_FSYNC)) { os_file_set_nocache(file, name, mode_str); } #ifdef USE_FILE_LOCK - if (create_mode != OS_FILE_OPEN_RAW && os_file_lock(file, name)) { + if (!srv_read_only_mode + && *success + && create_mode != OS_FILE_OPEN_RAW + && os_file_lock(file, name)) { if (create_mode == OS_FILE_OPEN_RETRY) { - int i; - ut_print_timestamp(stderr); - fputs(" InnoDB: Retrying to lock" - " the first data file\n", - stderr); - for (i = 0; i < 100; i++) { + + ut_a(!srv_read_only_mode); + + ib_logf(IB_LOG_LEVEL_INFO, + "Retrying to lock the first data file"); + + for (int i = 0; i < 100; i++) { os_thread_sleep(1000000); + if (!os_file_lock(file, name)) { *success = TRUE; return(file); } } - ut_print_timestamp(stderr); - fputs(" InnoDB: Unable to open the first data file\n", - stderr); + + ib_logf(IB_LOG_LEVEL_INFO, + "Unable to open the first data file"); } *success = FALSE; @@ -1711,29 +1789,24 @@ try_again: file = -1; } #endif /* USE_FILE_LOCK */ - if (srv_use_atomic_writes && type == OS_DATA_FILE - && os_file_set_atomic_writes(file, name)) { - close(file); - *success = FALSE; - file = -1; - } - return(file); #endif /* __WIN__ */ + + return(file); } /***********************************************************************//** Deletes a file if it exists. The file has to be closed before calling this. @return TRUE if success */ UNIV_INTERN -ibool +bool os_file_delete_if_exists( /*=====================*/ const char* name) /*!< in: file path as a null-terminated string */ { #ifdef __WIN__ - BOOL ret; + bool ret; ulint count = 0; loop: /* In Windows, deleting an .ibd file may fail if ibbackup is copying @@ -1742,31 +1815,30 @@ loop: ret = DeleteFile((LPCTSTR) name); if (ret) { - return(TRUE); + return(true); } - if (GetLastError() == ERROR_FILE_NOT_FOUND) { + DWORD lasterr = GetLastError(); + if (lasterr == ERROR_FILE_NOT_FOUND + || lasterr == ERROR_PATH_NOT_FOUND) { /* the file does not exist, this not an error */ - return(TRUE); + return(true); } count++; if (count > 100 && 0 == (count % 10)) { - fprintf(stderr, - "InnoDB: Warning: cannot delete file %s\n" - "InnoDB: Are you running ibbackup" - " to back up the file?\n", name); + os_file_get_last_error(true); /* print error information */ - os_file_get_last_error(TRUE); /* print error information */ + ib_logf(IB_LOG_LEVEL_WARN, "Delete of file %s failed.", name); } os_thread_sleep(1000000); /* sleep for a second */ if (count > 2000) { - return(FALSE); + return(false); } goto loop; @@ -1778,18 +1850,18 @@ loop: if (ret != 0 && errno != ENOENT) { os_file_handle_error_no_exit(name, "delete", FALSE); - return(FALSE); + return(false); } - return(TRUE); -#endif + return(true); +#endif /* __WIN__ */ } /***********************************************************************//** Deletes a file. The file has to be closed before calling this. @return TRUE if success */ UNIV_INTERN -ibool +bool os_file_delete( /*===========*/ const char* name) /*!< in: file path as a null-terminated @@ -1805,32 +1877,32 @@ loop: ret = DeleteFile((LPCTSTR) name); if (ret) { - return(TRUE); + return(true); } if (GetLastError() == ERROR_FILE_NOT_FOUND) { /* If the file does not exist, we classify this as a 'mild' error and return */ - return(FALSE); + return(false); } count++; if (count > 100 && 0 == (count % 10)) { + os_file_get_last_error(true); /* print error information */ + fprintf(stderr, "InnoDB: Warning: cannot delete file %s\n" "InnoDB: Are you running ibbackup" " to back up the file?\n", name); - - os_file_get_last_error(TRUE); /* print error information */ } os_thread_sleep(1000000); /* sleep for a second */ if (count > 2000) { - return(FALSE); + return(false); } goto loop; @@ -1842,10 +1914,10 @@ loop: if (ret != 0) { os_file_handle_error_no_exit(name, "delete", FALSE); - return(FALSE); + return(false); } - return(TRUE); + return(true); #endif } @@ -1862,6 +1934,19 @@ os_file_rename_func( string */ const char* newpath)/*!< in: new file path */ { +#ifdef UNIV_DEBUG + os_file_type_t type; + ibool exists; + + /* New path must not exist. */ + ut_ad(os_file_status(newpath, &exists, &type)); + ut_ad(!exists); + + /* Old path must exist. */ + ut_ad(os_file_status(oldpath, &exists, &type)); + ut_ad(exists); +#endif /* UNIV_DEBUG */ + #ifdef __WIN__ BOOL ret; @@ -1886,7 +1971,7 @@ os_file_rename_func( } return(TRUE); -#endif +#endif /* __WIN__ */ } /***********************************************************************//** @@ -1926,7 +2011,7 @@ os_file_close_func( } return(TRUE); -#endif +#endif /* __WIN__ */ } #ifdef UNIV_HOTBACKUP @@ -1962,7 +2047,7 @@ os_file_close_no_error_handling( } return(TRUE); -#endif +#endif /* __WIN__ */ } #endif /* UNIV_HOTBACKUP */ @@ -1991,7 +2076,7 @@ os_file_get_size( return(offset); #else return((os_offset_t) lseek(file, 0, SEEK_END)); -#endif +#endif /* __WIN__ */ } /***********************************************************************//** @@ -2014,28 +2099,6 @@ os_file_set_size( current_size = 0; -#ifdef HAVE_POSIX_FALLOCATE - if (srv_use_posix_fallocate) { - if (posix_fallocate(file, current_size, size) == -1) { - fprintf(stderr, - "InnoDB: Error: preallocating data for" - " file %s failed at\n" - "InnoDB: offset 0 size %lld. Operating system" - " error number %d.\n" - "InnoDB: Check that the disk is not full" - " or a disk quota exceeded.\n" - "InnoDB: Some operating system error numbers" - " are described at\n" - "InnoDB: " - REFMAN "operating-system-error-codes.html\n", - name, (longlong)size, errno); - - return (FALSE); - } - return (TRUE); - } -#endif - /* Write up to 1 megabyte at a time. */ buf_size = ut_min(64, (ulint) (size / UNIV_PAGE_SIZE)) * UNIV_PAGE_SIZE; @@ -2246,10 +2309,7 @@ os_file_flush_func( return(TRUE); } - ut_print_timestamp(stderr); - - fprintf(stderr, - " InnoDB: Error: the OS said file flush did not succeed\n"); + ib_logf(IB_LOG_LEVEL_ERROR, "The OS said file flush did not succeed"); os_file_handle_error(NULL, "flush"); @@ -2286,9 +2346,9 @@ os_file_pread( offs = (off_t) offset; if (sizeof(off_t) <= 4) { - if (UNIV_UNLIKELY(offset != (os_offset_t) offs)) { - fprintf(stderr, - "InnoDB: Error: file read at offset > 4 GB\n"); + if (offset != (os_offset_t) offs) { + ib_logf(IB_LOG_LEVEL_ERROR, + "File read at offset > 4 GB"); } } @@ -2389,16 +2449,16 @@ os_file_pwrite( off_t offs; ut_ad(n); + ut_ad(!srv_read_only_mode); /* If off_t is > 4 bytes in size, then we assume we can pass a 64-bit address */ offs = (off_t) offset; if (sizeof(off_t) <= 4) { - if (UNIV_UNLIKELY(offset != (os_offset_t) offs)) { - fprintf(stderr, - "InnoDB: Error: file write" - " at offset > 4 GB\n"); + if (offset != (os_offset_t) offs) { + ib_logf(IB_LOG_LEVEL_ERROR, + "File write at offset > 4 GB."); } } @@ -2473,7 +2533,7 @@ func_exit: return(ret); } -#endif +#endif /* !UNIV_HOTBACKUP */ } #endif @@ -2574,11 +2634,9 @@ try_again: return(TRUE); } - fprintf(stderr, - "InnoDB: Error: tried to read "ULINTPF" bytes at offset " - UINT64PF"\n" - "InnoDB: Was only able to read %ld.\n", - n, offset, (lint) ret); + ib_logf(IB_LOG_LEVEL_ERROR, + "Tried to read "ULINTPF" bytes at offset " UINT64PF". " + "Was only able to read %ld.", n, offset, (lint) ret); #endif /* __WIN__ */ #ifdef __WIN__ error_handling: @@ -2596,7 +2654,7 @@ error_handling: (ulong) GetLastError() #else (ulong) errno -#endif +#endif /* __WIN__ */ ); fflush(stderr); @@ -2754,6 +2812,8 @@ os_file_write_func( os_offset_t offset, /*!< in: file offset where to write */ ulint n) /*!< in: number of bytes to write */ { + ut_ad(!srv_read_only_mode); + #ifdef __WIN__ BOOL ret; DWORD len; @@ -2913,8 +2973,8 @@ retry: (ulint) errno); if (strerror(errno) != NULL) { fprintf(stderr, - "InnoDB: Error number %lu means '%s'.\n", - (ulint) errno, strerror(errno)); + "InnoDB: Error number %d means '%s'.\n", + errno, strerror(errno)); } fprintf(stderr, @@ -2937,15 +2997,15 @@ UNIV_INTERN ibool os_file_status( /*===========*/ - const char* path, /*!< in: pathname of the file */ + const char* path, /*!< in: pathname of the file */ ibool* exists, /*!< out: TRUE if file exists */ os_file_type_t* type) /*!< out: type of the file (if it exists) */ { #ifdef __WIN__ int ret; - struct _stat statinfo; + struct _stat64 statinfo; - ret = _stat(path, &statinfo); + ret = _stat64(path, &statinfo); if (ret && (errno == ENOENT || errno == ENOTDIR)) { /* file does not exist */ *exists = FALSE; @@ -3004,47 +3064,73 @@ os_file_status( /*******************************************************************//** This function returns information about the specified file -@return TRUE if stat information found */ +@return DB_SUCCESS if all OK */ UNIV_INTERN -ibool +dberr_t os_file_get_status( /*===============*/ const char* path, /*!< in: pathname of the file */ - os_file_stat_t* stat_info) /*!< information of a file in a + os_file_stat_t* stat_info, /*!< information of a file in a directory */ + bool check_rw_perm) /*!< in: for testing whether the + file can be opened in RW mode */ { -#ifdef __WIN__ int ret; - struct _stat statinfo; - ret = _stat(path, &statinfo); +#ifdef __WIN__ + struct _stat64 statinfo; + + ret = _stat64(path, &statinfo); + if (ret && (errno == ENOENT || errno == ENOTDIR)) { /* file does not exist */ - return(FALSE); + return(DB_NOT_FOUND); + } else if (ret) { /* file exists, but stat call failed */ os_file_handle_error_no_exit(path, "stat", FALSE); - return(FALSE); - } - if (_S_IFDIR & statinfo.st_mode) { + return(DB_FAIL); + + } else if (_S_IFDIR & statinfo.st_mode) { stat_info->type = OS_FILE_TYPE_DIR; } else if (_S_IFREG & statinfo.st_mode) { + + DWORD access = GENERIC_READ; + + if (!srv_read_only_mode) { + access |= GENERIC_WRITE; + } + stat_info->type = OS_FILE_TYPE_FILE; + + /* Check if we can open it in read-only mode. */ + + if (check_rw_perm) { + HANDLE fh; + + fh = CreateFile( + (LPCTSTR) path, // File to open + access, + 0, // No sharing + NULL, // Default security + OPEN_EXISTING, // Existing file only + FILE_ATTRIBUTE_NORMAL, // Normal file + NULL); // No attr. template + + if (fh == INVALID_HANDLE_VALUE) { + stat_info->rw_perm = false; + } else { + stat_info->rw_perm = true; + CloseHandle(fh); + } + } } else { stat_info->type = OS_FILE_TYPE_UNKNOWN; } - - stat_info->ctime = statinfo.st_ctime; - stat_info->atime = statinfo.st_atime; - stat_info->mtime = statinfo.st_mtime; - stat_info->size = statinfo.st_size; - - return(TRUE); #else - int ret; struct stat statinfo; ret = stat(path, &statinfo); @@ -3052,32 +3138,49 @@ os_file_get_status( if (ret && (errno == ENOENT || errno == ENOTDIR)) { /* file does not exist */ - return(FALSE); + return(DB_NOT_FOUND); + } else if (ret) { /* file exists, but stat call failed */ os_file_handle_error_no_exit(path, "stat", FALSE); - return(FALSE); - } + return(DB_FAIL); - if (S_ISDIR(statinfo.st_mode)) { + } else if (S_ISDIR(statinfo.st_mode)) { stat_info->type = OS_FILE_TYPE_DIR; } else if (S_ISLNK(statinfo.st_mode)) { stat_info->type = OS_FILE_TYPE_LINK; } else if (S_ISREG(statinfo.st_mode)) { stat_info->type = OS_FILE_TYPE_FILE; + + if (check_rw_perm) { + int fh; + int access; + + access = !srv_read_only_mode ? O_RDWR : O_RDONLY; + + fh = ::open(path, access, os_innodb_umask); + + if (fh == -1) { + stat_info->rw_perm = false; + } else { + stat_info->rw_perm = true; + close(fh); + } + } } else { stat_info->type = OS_FILE_TYPE_UNKNOWN; } +#endif /* _WIN_ */ + stat_info->ctime = statinfo.st_ctime; stat_info->atime = statinfo.st_atime; stat_info->mtime = statinfo.st_mtime; - stat_info->size = statinfo.st_size; + stat_info->size = statinfo.st_size; - return(TRUE); -#endif + return(DB_SUCCESS); } /* path name separator character */ @@ -3088,6 +3191,153 @@ os_file_get_status( #endif /****************************************************************//** +This function returns a new path name after replacing the basename +in an old path with a new basename. The old_path is a full path +name including the extension. The tablename is in the normal +form "databasename/tablename". The new base name is found after +the forward slash. Both input strings are null terminated. + +This function allocates memory to be returned. It is the callers +responsibility to free the return value after it is no longer needed. + +@return own: new full pathname */ +UNIV_INTERN +char* +os_file_make_new_pathname( +/*======================*/ + const char* old_path, /*!< in: pathname */ + const char* tablename) /*!< in: contains new base name */ +{ + ulint dir_len; + char* last_slash; + char* base_name; + char* new_path; + ulint new_path_len; + + /* Split the tablename into its database and table name components. + They are separated by a '/'. */ + last_slash = strrchr((char*) tablename, '/'); + base_name = last_slash ? last_slash + 1 : (char*) tablename; + + /* Find the offset of the last slash. We will strip off the + old basename.ibd which starts after that slash. */ + last_slash = strrchr((char*) old_path, OS_FILE_PATH_SEPARATOR); + dir_len = last_slash ? last_slash - old_path : strlen(old_path); + + /* allocate a new path and move the old directory path to it. */ + new_path_len = dir_len + strlen(base_name) + sizeof "/.ibd"; + new_path = static_cast<char*>(mem_alloc(new_path_len)); + memcpy(new_path, old_path, dir_len); + + ut_snprintf(new_path + dir_len, + new_path_len - dir_len, + "%c%s.ibd", + OS_FILE_PATH_SEPARATOR, + base_name); + + return(new_path); +} + +/****************************************************************//** +This function returns a remote path name by combining a data directory +path provided in a DATA DIRECTORY clause with the tablename which is +in the form 'database/tablename'. It strips the file basename (which +is the tablename) found after the last directory in the path provided. +The full filepath created will include the database name as a directory +under the path provided. The filename is the tablename with the '.ibd' +extension. All input and output strings are null-terminated. + +This function allocates memory to be returned. It is the callers +responsibility to free the return value after it is no longer needed. + +@return own: A full pathname; data_dir_path/databasename/tablename.ibd */ +UNIV_INTERN +char* +os_file_make_remote_pathname( +/*=========================*/ + const char* data_dir_path, /*!< in: pathname */ + const char* tablename, /*!< in: tablename */ + const char* extention) /*!< in: file extention; ibd,cfg */ +{ + ulint data_dir_len; + char* last_slash; + char* new_path; + ulint new_path_len; + + ut_ad(extention && strlen(extention) == 3); + + /* Find the offset of the last slash. We will strip off the + old basename or tablename which starts after that slash. */ + last_slash = strrchr((char*) data_dir_path, OS_FILE_PATH_SEPARATOR); + data_dir_len = last_slash ? last_slash - data_dir_path : strlen(data_dir_path); + + /* allocate a new path and move the old directory path to it. */ + new_path_len = data_dir_len + strlen(tablename) + + sizeof "/." + strlen(extention); + new_path = static_cast<char*>(mem_alloc(new_path_len)); + memcpy(new_path, data_dir_path, data_dir_len); + ut_snprintf(new_path + data_dir_len, + new_path_len - data_dir_len, + "%c%s.%s", + OS_FILE_PATH_SEPARATOR, + tablename, + extention); + + srv_normalize_path_for_win(new_path); + + return(new_path); +} + +/****************************************************************//** +This function reduces a null-terminated full remote path name into +the path that is sent by MySQL for DATA DIRECTORY clause. It replaces +the 'databasename/tablename.ibd' found at the end of the path with just +'tablename'. + +Since the result is always smaller than the path sent in, no new memory +is allocated. The caller should allocate memory for the path sent in. +This function manipulates that path in place. + +If the path format is not as expected, just return. The result is used +to inform a SHOW CREATE TABLE command. */ +UNIV_INTERN +void +os_file_make_data_dir_path( +/*========================*/ + char* data_dir_path) /*!< in/out: full path/data_dir_path */ +{ + char* ptr; + char* tablename; + ulint tablename_len; + + /* Replace the period before the extension with a null byte. */ + ptr = strrchr((char*) data_dir_path, '.'); + if (!ptr) { + return; + } + ptr[0] = '\0'; + + /* The tablename starts after the last slash. */ + ptr = strrchr((char*) data_dir_path, OS_FILE_PATH_SEPARATOR); + if (!ptr) { + return; + } + ptr[0] = '\0'; + tablename = ptr + 1; + + /* The databasename starts after the next to last slash. */ + ptr = strrchr((char*) data_dir_path, OS_FILE_PATH_SEPARATOR); + if (!ptr) { + return; + } + tablename_len = ut_strlen(tablename); + + ut_memmove(++ptr, tablename, tablename_len); + + ptr[tablename_len] = '\0'; +} + +/****************************************************************//** The function os_file_dirname returns a directory component of a null-terminated pathname string. In the usual case, dirname returns the string up to, but not including, the final '/', and basename @@ -3151,11 +3401,18 @@ os_file_create_subdirs_if_needed( /*=============================*/ const char* path) /*!< in: path name */ { - char* subdir; - ibool success, subdir_exists; - os_file_type_t type; + if (srv_read_only_mode) { + + ib_logf(IB_LOG_LEVEL_ERROR, + "read only mode set. Can't create subdirectories '%s'", + path); + + return(FALSE); + + } + + char* subdir = os_file_dirname(path); - subdir = os_file_dirname(path); if (strlen(subdir) == 1 && (*subdir == OS_FILE_PATH_SEPARATOR || *subdir == '.')) { /* subdir is root or cwd, nothing to do */ @@ -3165,15 +3422,21 @@ os_file_create_subdirs_if_needed( } /* Test if subdir exists */ - success = os_file_status(subdir, &subdir_exists, &type); + os_file_type_t type; + ibool subdir_exists; + ibool success = os_file_status(subdir, &subdir_exists, &type); + if (success && !subdir_exists) { + /* subdir does not exist, create it */ success = os_file_create_subdirs_if_needed(subdir); + if (!success) { mem_free(subdir); return(FALSE); } + success = os_file_create_directory(subdir, FALSE); } @@ -3195,7 +3458,7 @@ os_aio_array_get_nth_slot( { ut_a(index < array->n_slots); - return((array->slots) + index); + return(&array->slots[index]); } #if defined(LINUX_NATIVE_AIO) @@ -3297,43 +3560,74 @@ os_aio_native_aio_supported(void) /*=============================*/ { int fd; - byte* buf; - byte* ptr; - struct io_event io_event; io_context_t io_ctx; - struct iocb iocb; - struct iocb* p_iocb; - int err; + char name[1000]; if (!os_aio_linux_create_io_ctx(1, &io_ctx)) { /* The platform does not support native aio. */ return(FALSE); - } + } else if (!srv_read_only_mode) { + /* Now check if tmpdir supports native aio ops. */ + fd = innobase_mysql_tmpfile(); - /* Now check if tmpdir supports native aio ops. */ - fd = innobase_mysql_tmpfile(); + if (fd < 0) { + ib_logf(IB_LOG_LEVEL_WARN, + "Unable to create temp file to check " + "native AIO support."); - if (fd < 0) { - ut_print_timestamp(stderr); - fprintf(stderr, " InnoDB: Error: unable to create " - "temp file to check native AIO support.\n"); + return(FALSE); + } + } else { - return(FALSE); + srv_normalize_path_for_win(srv_log_group_home_dir); + + ulint dirnamelen = strlen(srv_log_group_home_dir); + ut_a(dirnamelen < (sizeof name) - 10 - sizeof "ib_logfile"); + memcpy(name, srv_log_group_home_dir, dirnamelen); + + /* Add a path separator if needed. */ + if (dirnamelen && name[dirnamelen - 1] != SRV_PATH_SEPARATOR) { + name[dirnamelen++] = SRV_PATH_SEPARATOR; + } + + strcpy(name + dirnamelen, "ib_logfile0"); + + fd = ::open(name, O_RDONLY); + + if (fd == -1) { + + ib_logf(IB_LOG_LEVEL_WARN, + "Unable to open \"%s\" to check " + "native AIO read support.", name); + + return(FALSE); + } } + struct io_event io_event; + memset(&io_event, 0x0, sizeof(io_event)); - buf = static_cast<byte*>(ut_malloc(UNIV_PAGE_SIZE * 2)); - ptr = static_cast<byte*>(ut_align(buf, UNIV_PAGE_SIZE)); + byte* buf = static_cast<byte*>(ut_malloc(UNIV_PAGE_SIZE * 2)); + byte* ptr = static_cast<byte*>(ut_align(buf, UNIV_PAGE_SIZE)); + + struct iocb iocb; /* Suppress valgrind warning. */ memset(buf, 0x00, UNIV_PAGE_SIZE * 2); - memset(&iocb, 0x0, sizeof(iocb)); - p_iocb = &iocb; - io_prep_pwrite(p_iocb, fd, ptr, UNIV_PAGE_SIZE, 0); - err = io_submit(io_ctx, 1, &p_iocb); + struct iocb* p_iocb = &iocb; + + if (!srv_read_only_mode) { + io_prep_pwrite(p_iocb, fd, ptr, UNIV_PAGE_SIZE, 0); + } else { + ut_a(UNIV_PAGE_SIZE >= 512); + io_prep_pread(p_iocb, fd, ptr, 512, 0); + } + + int err = io_submit(io_ctx, 1, &p_iocb); + if (err >= 1) { /* Now collect the submitted IO request. */ err = io_getevents(io_ctx, 1, 1, &io_event, NULL); @@ -3348,22 +3642,18 @@ os_aio_native_aio_supported(void) case -EINVAL: case -ENOSYS: - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: Error: Linux Native AIO is not" - " supported on tmpdir.\n" - "InnoDB: You can either move tmpdir to a" - " file system that supports native AIO\n" - "InnoDB: or you can set" - " innodb_use_native_aio to FALSE to avoid" - " this message.\n"); + ib_logf(IB_LOG_LEVEL_ERROR, + "Linux Native AIO not supported. You can either " + "move %s to a file system that supports native " + "AIO or you can set innodb_use_native_aio to " + "FALSE to avoid this message.", + srv_read_only_mode ? name : "tmpdir"); /* fall through. */ default: - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: Error: Linux Native AIO check" - " on tmpdir returned error[%d]\n", -err); + ib_logf(IB_LOG_LEVEL_ERROR, + "Linux Native AIO check on %s returned error[%d]", + srv_read_only_mode ? name : "tmpdir", -err); } return(FALSE); @@ -3385,34 +3675,33 @@ os_aio_array_create( ulint n_segments) /*!< in: number of segments in the aio array */ { os_aio_array_t* array; - ulint i; - os_aio_slot_t* slot; #ifdef WIN_ASYNC_IO OVERLAPPED* over; #elif defined(LINUX_NATIVE_AIO) struct io_event* io_event = NULL; -#endif +#endif /* WIN_ASYNC_IO */ ut_a(n > 0); ut_a(n_segments > 0); - array = static_cast<os_aio_array_t*>(ut_malloc(sizeof(os_aio_array_t))); + array = static_cast<os_aio_array_t*>(ut_malloc(sizeof(*array))); + memset(array, 0x0, sizeof(*array)); - array->mutex = os_mutex_create(); - array->not_full = os_event_create(NULL); - array->is_empty = os_event_create(NULL); + array->mutex = os_mutex_create(); + array->not_full = os_event_create(); + array->is_empty = os_event_create(); os_event_set(array->is_empty); - array->n_slots = n; - array->n_segments = n_segments; - array->n_reserved = 0; - array->cur_seg = 0; + array->n_slots = n; + array->n_segments = n_segments; array->slots = static_cast<os_aio_slot_t*>( - ut_malloc(n * sizeof(os_aio_slot_t))); + ut_malloc(n * sizeof(*array->slots))); + + memset(array->slots, 0x0, sizeof(n * sizeof(*array->slots))); #ifdef __WIN__ array->handles = static_cast<HANDLE*>(ut_malloc(n * sizeof(HANDLE))); -#endif +#endif /* __WIN__ */ #if defined(LINUX_NATIVE_AIO) array->aio_ctx = NULL; @@ -3430,7 +3719,7 @@ os_aio_array_create( array->aio_ctx = static_cast<io_context**>( ut_malloc(n_segments * sizeof(*array->aio_ctx))); - for (i = 0; i < n_segments; ++i) { + for (ulint i = 0; i < n_segments; ++i) { if (!os_aio_linux_create_io_ctx(n/n_segments, &array->aio_ctx[i])) { /* If something bad happened during aio setup @@ -3463,7 +3752,9 @@ os_aio_array_create( skip_native_aio: #endif /* LINUX_NATIVE_AIO */ - for (i = 0; i < n; i++) { + for (ulint i = 0; i < n; i++) { + os_aio_slot_t* slot; + slot = os_aio_array_get_nth_slot(array, i); slot->pos = i; @@ -3471,18 +3762,17 @@ skip_native_aio: #ifdef WIN_ASYNC_IO slot->handle = CreateEvent(NULL,TRUE, FALSE, NULL); - over = &(slot->control); + over = &slot->control; over->hEvent = slot->handle; - *((array->handles) + i) = over->hEvent; + array->handles[i] = over->hEvent; #elif defined(LINUX_NATIVE_AIO) - memset(&slot->control, 0x0, sizeof(slot->control)); slot->n_bytes = 0; slot->ret = 0; -#endif +#endif /* WIN_ASYNC_IO */ } return(array); @@ -3494,7 +3784,7 @@ static void os_aio_array_free( /*==============*/ - os_aio_array_t* array) /*!< in, own: array to free */ + os_aio_array_t*& array) /*!< in, own: array to free */ { #ifdef WIN_ASYNC_IO ulint i; @@ -3521,6 +3811,8 @@ os_aio_array_free( ut_free(array->slots); ut_free(array); + + array = 0; } /*********************************************************************** @@ -3541,93 +3833,100 @@ os_aio_init( ulint n_slots_sync) /*<! in: number of slots in the sync aio array */ { - ulint i; - ulint n_segments = 2 + n_read_segs + n_write_segs; - - ut_ad(n_segments >= 4); - os_io_init_simple(); #if defined(LINUX_NATIVE_AIO) /* Check if native aio is supported on this system and tmpfs */ - if (srv_use_native_aio - && !os_aio_native_aio_supported()) { + if (srv_use_native_aio && !os_aio_native_aio_supported()) { + + ib_logf(IB_LOG_LEVEL_WARN, "Linux Native AIO disabled."); - ut_print_timestamp(stderr); - fprintf(stderr, - " InnoDB: Warning: Linux Native AIO" - " disabled.\n"); srv_use_native_aio = FALSE; } #endif /* LINUX_NATIVE_AIO */ - for (i = 0; i < n_segments; i++) { - srv_set_io_thread_op_info(i, "not started yet"); - } - + srv_reset_io_thread_op_info(); - /* fprintf(stderr, "Array n per seg %lu\n", n_per_seg); */ + os_aio_read_array = os_aio_array_create( + n_read_segs * n_per_seg, n_read_segs); - os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1); - if (os_aio_ibuf_array == NULL) { - goto err_exit; + if (os_aio_read_array == NULL) { + return(FALSE); } - srv_io_thread_function[0] = "insert buffer thread"; + ulint start = (srv_read_only_mode) ? 0 : 2; + ulint n_segs = n_read_segs + start; - os_aio_log_array = os_aio_array_create(n_per_seg, 1); - if (os_aio_log_array == NULL) { - goto err_exit; + /* 0 is the ibuf segment and 1 is the insert buffer segment. */ + for (ulint i = start; i < n_segs; ++i) { + ut_a(i < SRV_MAX_N_IO_THREADS); + srv_io_thread_function[i] = "read thread"; } - srv_io_thread_function[1] = "log thread"; + ulint n_segments = n_read_segs; - os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg, - n_read_segs); - if (os_aio_read_array == NULL) { - goto err_exit; - } + if (!srv_read_only_mode) { - for (i = 2; i < 2 + n_read_segs; i++) { - ut_a(i < SRV_MAX_N_IO_THREADS); - srv_io_thread_function[i] = "read thread"; - } + os_aio_log_array = os_aio_array_create(n_per_seg, 1); - os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg, - n_write_segs); - if (os_aio_write_array == NULL) { - goto err_exit; - } + if (os_aio_log_array == NULL) { + return(FALSE); + } - for (i = 2 + n_read_segs; i < n_segments; i++) { - ut_a(i < SRV_MAX_N_IO_THREADS); - srv_io_thread_function[i] = "write thread"; + ++n_segments; + + srv_io_thread_function[1] = "log thread"; + + os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1); + + if (os_aio_ibuf_array == NULL) { + return(FALSE); + } + + ++n_segments; + + srv_io_thread_function[0] = "insert buffer thread"; + + os_aio_write_array = os_aio_array_create( + n_write_segs * n_per_seg, n_write_segs); + + if (os_aio_write_array == NULL) { + return(FALSE); + } + + n_segments += n_write_segs; + + for (ulint i = start + n_read_segs; i < n_segments; ++i) { + ut_a(i < SRV_MAX_N_IO_THREADS); + srv_io_thread_function[i] = "write thread"; + } + + ut_ad(n_segments >= 4); + } else { + ut_ad(n_segments > 0); } os_aio_sync_array = os_aio_array_create(n_slots_sync, 1); + if (os_aio_sync_array == NULL) { - goto err_exit; + return(FALSE); } - os_aio_n_segments = n_segments; os_aio_validate(); - os_aio_segment_wait_events = static_cast<os_event_struct_t**>( - ut_malloc(n_segments * sizeof(void*))); + os_aio_segment_wait_events = static_cast<os_event_t*>( + ut_malloc(n_segments * sizeof *os_aio_segment_wait_events)); - for (i = 0; i < n_segments; i++) { - os_aio_segment_wait_events[i] = os_event_create(NULL); + for (ulint i = 0; i < n_segments; ++i) { + os_aio_segment_wait_events[i] = os_event_create(); } - os_last_printout = time(NULL); + os_last_printout = ut_time(); return(TRUE); -err_exit: - return(FALSE); - } /*********************************************************************** @@ -3637,20 +3936,25 @@ void os_aio_free(void) /*=============*/ { - ulint i; + if (os_aio_ibuf_array != 0) { + os_aio_array_free(os_aio_ibuf_array); + } + + if (os_aio_log_array != 0) { + os_aio_array_free(os_aio_log_array); + } + + if (os_aio_write_array != 0) { + os_aio_array_free(os_aio_write_array); + } + + if (os_aio_sync_array != 0) { + os_aio_array_free(os_aio_sync_array); + } - os_aio_array_free(os_aio_ibuf_array); - os_aio_ibuf_array = NULL; - os_aio_array_free(os_aio_log_array); - os_aio_log_array = NULL; os_aio_array_free(os_aio_read_array); - os_aio_read_array = NULL; - os_aio_array_free(os_aio_write_array); - os_aio_write_array = NULL; - os_aio_array_free(os_aio_sync_array); - os_aio_sync_array = NULL; - for (i = 0; i < os_aio_n_segments; i++) { + for (ulint i = 0; i < os_aio_n_segments; i++) { os_event_free(os_aio_segment_wait_events[i]); } @@ -3686,14 +3990,20 @@ void os_aio_wake_all_threads_at_shutdown(void) /*=====================================*/ { - ulint i; - #ifdef WIN_ASYNC_IO /* This code wakes up all ai/o threads in Windows native aio */ os_aio_array_wake_win_aio_at_shutdown(os_aio_read_array); - os_aio_array_wake_win_aio_at_shutdown(os_aio_write_array); - os_aio_array_wake_win_aio_at_shutdown(os_aio_ibuf_array); - os_aio_array_wake_win_aio_at_shutdown(os_aio_log_array); + if (os_aio_write_array != 0) { + os_aio_array_wake_win_aio_at_shutdown(os_aio_write_array); + } + + if (os_aio_ibuf_array != 0) { + os_aio_array_wake_win_aio_at_shutdown(os_aio_ibuf_array); + } + + if (os_aio_log_array != 0) { + os_aio_array_wake_win_aio_at_shutdown(os_aio_log_array); + } #elif defined(LINUX_NATIVE_AIO) @@ -3705,12 +4015,14 @@ os_aio_wake_all_threads_at_shutdown(void) if (srv_use_native_aio) { return; } + /* Fall through to simulated AIO handler wakeup if we are not using native AIO. */ -#endif +#endif /* !WIN_ASYNC_AIO */ + /* This loop wakes up all simulated ai/o threads */ - for (i = 0; i < os_aio_n_segments; i++) { + for (ulint i = 0; i < os_aio_n_segments; i++) { os_event_set(os_aio_segment_wait_events[i]); } @@ -3724,6 +4036,7 @@ void os_aio_wait_until_no_pending_writes(void) /*=====================================*/ { + ut_ad(!srv_read_only_mode); os_event_wait(os_aio_write_array->is_empty); } @@ -3742,10 +4055,14 @@ os_aio_get_segment_no_from_slot( ulint seg_len; if (array == os_aio_ibuf_array) { - segment = 0; + ut_ad(!srv_read_only_mode); + + segment = IO_IBUF_SEGMENT; } else if (array == os_aio_log_array) { - segment = 1; + ut_ad(!srv_read_only_mode); + + segment = IO_LOG_SEGMENT; } else if (array == os_aio_read_array) { seg_len = os_aio_read_array->n_slots @@ -3753,7 +4070,9 @@ os_aio_get_segment_no_from_slot( segment = 2 + slot->pos / seg_len; } else { + ut_ad(!srv_read_only_mode); ut_a(array == os_aio_write_array); + seg_len = os_aio_write_array->n_slots / os_aio_write_array->n_segments; @@ -3774,15 +4093,19 @@ os_aio_get_array_and_local_segment( os_aio_array_t** array, /*!< out: aio wait array */ ulint global_segment)/*!< in: global segment number */ { - ulint segment; + ulint segment; ut_a(global_segment < os_aio_n_segments); - if (global_segment == 0) { + if (srv_read_only_mode) { + *array = os_aio_read_array; + + return(global_segment); + } else if (global_segment == IO_IBUF_SEGMENT) { *array = os_aio_ibuf_array; segment = 0; - } else if (global_segment == 1) { + } else if (global_segment == IO_LOG_SEGMENT) { *array = os_aio_log_array; segment = 0; @@ -3830,7 +4153,7 @@ os_aio_array_reserve_slot( struct iocb* iocb; off_t aio_offset; -#endif +#endif /* WIN_ASYNC_IO */ ulint i; ulint counter; ulint slots_per_seg; @@ -3838,7 +4161,7 @@ os_aio_array_reserve_slot( #ifdef WIN_ASYNC_IO ut_a((len & 0xFFFFFFFFUL) == len); -#endif +#endif /* WIN_ASYNC_IO */ /* No need of a mutex. Only reading constant fields */ slots_per_seg = array->n_slots / array->n_segments; @@ -3871,9 +4194,11 @@ loop: local segment and do a full scan of the array. We are guaranteed to find a slot in full scan. */ for (i = local_seg * slots_per_seg, counter = 0; - counter < array->n_slots; i++, counter++) { + counter < array->n_slots; + i++, counter++) { i %= array->n_slots; + slot = os_aio_array_get_nth_slot(array, i); if (slot->reserved == FALSE) { @@ -3897,7 +4222,7 @@ found: } slot->reserved = TRUE; - slot->reservation_time = time(NULL); + slot->reservation_time = ut_time(); slot->message1 = message1; slot->message2 = message2; slot->file = file; @@ -3909,7 +4234,7 @@ found: slot->io_already_done = FALSE; #ifdef WIN_ASYNC_IO - control = &(slot->control); + control = &slot->control; control->Offset = (DWORD) offset & 0xFFFFFFFF; control->OffsetHigh = (DWORD) (offset >> 32); ResetEvent(slot->handle); @@ -3940,7 +4265,6 @@ found: iocb->data = (void*) slot; slot->n_bytes = 0; slot->ret = 0; - /*fprintf(stderr, "Filled up Linux native iocb.\n");*/ skip_native_aio: #endif /* LINUX_NATIVE_AIO */ @@ -3958,9 +4282,6 @@ os_aio_array_free_slot( os_aio_array_t* array, /*!< in: aio array */ os_aio_slot_t* slot) /*!< in: pointer to slot */ { - ut_ad(array); - ut_ad(slot); - os_mutex_enter(array->mutex); ut_ad(slot->reserved); @@ -4009,36 +4330,42 @@ os_aio_simulated_wake_handler_thread( arrays */ { os_aio_array_t* array; - os_aio_slot_t* slot; ulint segment; - ulint n; - ulint i; ut_ad(!srv_use_native_aio); segment = os_aio_get_array_and_local_segment(&array, global_segment); - n = array->n_slots / array->n_segments; + ulint n = array->n_slots / array->n_segments; + + segment *= n; /* Look through n slots after the segment * n'th slot */ os_mutex_enter(array->mutex); - for (i = 0; i < n; i++) { - slot = os_aio_array_get_nth_slot(array, i + segment * n); + for (ulint i = 0; i < n; ++i) { + const os_aio_slot_t* slot; + + slot = os_aio_array_get_nth_slot(array, segment + i); if (slot->reserved) { + /* Found an i/o request */ - break; + os_mutex_exit(array->mutex); + + os_event_t event; + + event = os_aio_segment_wait_events[global_segment]; + + os_event_set(event); + + return; } } os_mutex_exit(array->mutex); - - if (i < n) { - os_event_set(os_aio_segment_wait_events[global_segment]); - } } /**********************************************************************//** @@ -4048,8 +4375,6 @@ void os_aio_simulated_wake_handler_threads(void) /*=======================================*/ { - ulint i; - if (srv_use_native_aio) { /* We do not use simulated aio: do nothing */ @@ -4058,7 +4383,7 @@ os_aio_simulated_wake_handler_threads(void) os_aio_recommend_sleep_for_read_threads = FALSE; - for (i = 0; i < os_aio_n_segments; i++) { + for (ulint i = 0; i < os_aio_n_segments; i++) { os_aio_simulated_wake_handler_thread(i); } } @@ -4080,7 +4405,6 @@ background threads too eagerly to allow for coalescing during readahead requests. */ #ifdef __WIN__ os_aio_array_t* array; - ulint g; if (srv_use_native_aio) { /* We do not use simulated aio: do nothing */ @@ -4090,12 +4414,12 @@ readahead requests. */ os_aio_recommend_sleep_for_read_threads = TRUE; - for (g = 0; g < os_aio_n_segments; g++) { - os_aio_get_array_and_local_segment(&array, g); + for (ulint i = 0; i < os_aio_n_segments; i++) { + os_aio_get_array_and_local_segment(&array, i); if (array == os_aio_read_array) { - os_event_reset(os_aio_segment_wait_events[g]); + os_event_reset(os_aio_segment_wait_events[i]); } } #endif /* __WIN__ */ @@ -4193,11 +4517,10 @@ os_aio_func( ibool retval; BOOL ret = TRUE; DWORD len = (DWORD) n; - struct fil_node_struct * dummy_mess1; + struct fil_node_t* dummy_mess1; void* dummy_mess2; ulint dummy_type; #endif /* WIN_ASYNC_IO */ - ibool retry; ulint wake_later; ut_ad(file); @@ -4235,6 +4558,7 @@ os_aio_func( return(os_file_read_func(file, buf, offset, n)); } + ut_ad(!srv_read_only_mode); ut_a(type == OS_FILE_WRITE); return(os_file_write_func(name, file, buf, offset, n)); @@ -4243,9 +4567,12 @@ os_aio_func( try_again: switch (mode) { case OS_AIO_NORMAL: - array = (type == OS_FILE_READ) - ? os_aio_read_array - : os_aio_write_array; + if (type == OS_FILE_READ) { + array = os_aio_read_array; + } else { + ut_ad(!srv_read_only_mode); + array = os_aio_write_array; + } break; case OS_AIO_IBUF: ut_ad(type == OS_FILE_READ); @@ -4254,14 +4581,21 @@ try_again: wake_later = FALSE; - array = os_aio_ibuf_array; + if (srv_read_only_mode) { + array = os_aio_read_array; + } else { + array = os_aio_ibuf_array; + } break; case OS_AIO_LOG: - array = os_aio_log_array; + if (srv_read_only_mode) { + array = os_aio_read_array; + } else { + array = os_aio_log_array; + } break; case OS_AIO_SYNC: array = os_aio_sync_array; - #if defined(LINUX_NATIVE_AIO) /* In Linux native AIO we don't use sync IO array. */ ut_a(!srv_use_native_aio); @@ -4286,7 +4620,7 @@ try_again: if (!os_aio_linux_dispatch(array, slot)) { goto err_exit; } -#endif +#endif /* WIN_ASYNC_IO */ } else { if (!wake_later) { os_aio_simulated_wake_handler_thread( @@ -4295,6 +4629,7 @@ try_again: } } } else if (type == OS_FILE_WRITE) { + ut_ad(!srv_read_only_mode); if (srv_use_native_aio) { os_n_file_writes++; #ifdef WIN_ASYNC_IO @@ -4305,7 +4640,7 @@ try_again: if (!os_aio_linux_dispatch(array, slot)) { goto err_exit; } -#endif +#endif /* WIN_ASYNC_IO */ } else { if (!wake_later) { os_aio_simulated_wake_handler_thread( @@ -4329,11 +4664,10 @@ try_again: we must use the same wait mechanism as for async i/o */ - retval = os_aio_windows_handle(ULINT_UNDEFINED, - slot->pos, - &dummy_mess1, - &dummy_mess2, - &dummy_type); + retval = os_aio_windows_handle( + ULINT_UNDEFINED, slot->pos, + &dummy_mess1, &dummy_mess2, + &dummy_type); return(retval); } @@ -4352,10 +4686,8 @@ err_exit: #endif /* LINUX_NATIVE_AIO || WIN_ASYNC_IO */ os_aio_array_free_slot(array, slot); - retry = os_file_handle_error(name, - type == OS_FILE_READ - ? "aio read" : "aio write"); - if (retry) { + if (os_file_handle_error( + name,type == OS_FILE_READ ? "aio read" : "aio write")) { goto try_again; } @@ -4405,8 +4737,8 @@ os_aio_windows_handle( BOOL retry = FALSE; if (segment == ULINT_UNDEFINED) { - array = os_aio_sync_array; segment = 0; + array = os_aio_sync_array; } else { segment = os_aio_get_array_and_local_segment(&array, segment); } @@ -4420,16 +4752,21 @@ os_aio_windows_handle( n = array->n_slots / array->n_segments; if (array == os_aio_sync_array) { + WaitForSingleObject( os_aio_array_get_nth_slot(array, pos)->handle, INFINITE); + i = pos; + } else { - srv_set_io_thread_op_info(orig_seg, "wait Windows aio"); - i = WaitForMultipleObjects((DWORD) n, - array->handles + segment * n, - FALSE, - INFINITE); + if (orig_seg != ULINT_UNDEFINED) { + srv_set_io_thread_op_info(orig_seg, "wait Windows aio"); + } + + i = WaitForMultipleObjects( + (DWORD) n, array->handles + segment * n, + FALSE, INFINITE); } os_mutex_enter(array->mutex); @@ -4449,8 +4786,8 @@ os_aio_windows_handle( ut_a(slot->reserved); if (orig_seg != ULINT_UNDEFINED) { - srv_set_io_thread_op_info(orig_seg, - "get windows aio return value"); + srv_set_io_thread_op_info( + orig_seg, "get windows aio return value"); } ret = GetOverlappedResult(slot->file, &(slot->control), &len, TRUE); @@ -4753,7 +5090,7 @@ found: *type = slot->type; - if ((slot->ret == 0) && (slot->n_bytes == (long) slot->len)) { + if (slot->ret == 0 && slot->n_bytes == (long) slot->len) { ret = TRUE; } else { @@ -4802,8 +5139,6 @@ os_aio_simulated_handle( { os_aio_array_t* array; ulint segment; - os_aio_slot_t* slot; - os_aio_slot_t* slot2; os_aio_slot_t* consecutive_ios[OS_AIO_MERGE_N_CONSECUTIVE]; ulint n_consecutive; ulint total_len; @@ -4816,7 +5151,7 @@ os_aio_simulated_handle( ibool ret; ibool any_reserved; ulint n; - ulint i; + os_aio_slot_t* aio_slot; /* Fix compiler warning */ *consecutive_ios = NULL; @@ -4854,7 +5189,9 @@ restart: os_mutex_enter(array->mutex); - for (i = 0; i < n; i++) { + for (ulint i = 0; i < n; i++) { + os_aio_slot_t* slot; + slot = os_aio_array_get_nth_slot(array, i + segment * n); if (!slot->reserved) { @@ -4868,8 +5205,8 @@ restart: (ulong) i); } + aio_slot = slot; ret = TRUE; - goto slot_io_done; } else { any_reserved = TRUE; @@ -4879,9 +5216,7 @@ restart: /* There is no completed request. If there is no pending request at all, and the system is being shut down, exit. */ - if (UNIV_UNLIKELY - (!any_reserved - && srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS)) { + if (!any_reserved && srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) { os_mutex_exit(array->mutex); *message1 = NULL; *message2 = NULL; @@ -4897,12 +5232,15 @@ restart: biggest_age = 0; lowest_offset = IB_UINT64_MAX; - for (i = 0; i < n; i++) { + for (ulint i = 0; i < n; i++) { + os_aio_slot_t* slot; + slot = os_aio_array_get_nth_slot(array, i + segment * n); if (slot->reserved) { - age = (ulint) difftime(time(NULL), - slot->reservation_time); + + age = (ulint) difftime( + ut_time(), slot->reservation_time); if ((age >= 2 && age > biggest_age) || (age >= 2 && age == biggest_age @@ -4926,9 +5264,11 @@ restart: lowest_offset = IB_UINT64_MAX; - for (i = 0; i < n; i++) { - slot = os_aio_array_get_nth_slot(array, - i + segment * n); + for (ulint i = 0; i < n; i++) { + os_aio_slot_t* slot; + + slot = os_aio_array_get_nth_slot( + array, i + segment * n); if (slot->reserved && slot->offset < lowest_offset) { @@ -4954,25 +5294,28 @@ restart: ut_ad(n_consecutive != 0); ut_ad(consecutive_ios[0] != NULL); - slot = consecutive_ios[0]; + aio_slot = consecutive_ios[0]; /* Check if there are several consecutive blocks to read or write */ consecutive_loop: - for (i = 0; i < n; i++) { - slot2 = os_aio_array_get_nth_slot(array, i + segment * n); + for (ulint i = 0; i < n; i++) { + os_aio_slot_t* slot; + + slot = os_aio_array_get_nth_slot(array, i + segment * n); - if (slot2->reserved && slot2 != slot - && slot2->offset == slot->offset + slot->len - && slot2->type == slot->type - && slot2->file == slot->file) { + if (slot->reserved + && slot != aio_slot + && slot->offset == slot->offset + aio_slot->len + && slot->type == aio_slot->type + && slot->file == aio_slot->file) { /* Found a consecutive i/o request */ - consecutive_ios[n_consecutive] = slot2; + consecutive_ios[n_consecutive] = slot; n_consecutive++; - slot = slot2; + aio_slot = slot; if (n_consecutive < OS_AIO_MERGE_N_CONSECUTIVE) { @@ -4990,15 +5333,15 @@ consecutive_loop: i/o */ total_len = 0; - slot = consecutive_ios[0]; + aio_slot = consecutive_ios[0]; - for (i = 0; i < n_consecutive; i++) { + for (ulint i = 0; i < n_consecutive; i++) { total_len += consecutive_ios[i]->len; } if (n_consecutive == 1) { /* We can use the buffer of the i/o request */ - combined_buf = slot->buf; + combined_buf = aio_slot->buf; combined_buf2 = NULL; } else { combined_buf2 = static_cast<byte*>( @@ -5016,50 +5359,41 @@ consecutive_loop: os_mutex_exit(array->mutex); - if (slot->type == OS_FILE_WRITE && n_consecutive > 1) { + if (aio_slot->type == OS_FILE_WRITE && n_consecutive > 1) { /* Copy the buffers to the combined buffer */ offs = 0; - for (i = 0; i < n_consecutive; i++) { + for (ulint i = 0; i < n_consecutive; i++) { ut_memcpy(combined_buf + offs, consecutive_ios[i]->buf, consecutive_ios[i]->len); + offs += consecutive_ios[i]->len; } } srv_set_io_thread_op_info(global_segment, "doing file i/o"); - if (os_aio_print_debug) { - fprintf(stderr, - "InnoDB: doing i/o of type %lu at offset " UINT64PF - ", length %lu\n", - (ulong) slot->type, slot->offset, (ulong) total_len); - } - /* Do the i/o with ordinary, synchronous i/o functions: */ - if (slot->type == OS_FILE_WRITE) { - ret = os_file_write(slot->name, slot->file, combined_buf, - slot->offset, total_len); + if (aio_slot->type == OS_FILE_WRITE) { + ut_ad(!srv_read_only_mode); + ret = os_file_write( + aio_slot->name, aio_slot->file, combined_buf, + aio_slot->offset, total_len); } else { - ret = os_file_read(slot->file, combined_buf, - slot->offset, total_len); + ret = os_file_read( + aio_slot->file, combined_buf, + aio_slot->offset, total_len); } ut_a(ret); srv_set_io_thread_op_info(global_segment, "file i/o done"); -#if 0 - fprintf(stderr, - "aio: %lu consecutive %lu:th segment, first offs %lu blocks\n", - n_consecutive, global_segment, slot->offset / UNIV_PAGE_SIZE); -#endif - - if (slot->type == OS_FILE_READ && n_consecutive > 1) { + if (aio_slot->type == OS_FILE_READ && n_consecutive > 1) { /* Copy the combined buffer to individual buffers */ offs = 0; - for (i = 0; i < n_consecutive; i++) { + for (ulint i = 0; i < n_consecutive; i++) { ut_memcpy(consecutive_ios[i]->buf, combined_buf + offs, consecutive_ios[i]->len); @@ -5075,7 +5409,7 @@ consecutive_loop: /* Mark the i/os done in slots */ - for (i = 0; i < n_consecutive; i++) { + for (ulint i = 0; i < n_consecutive; i++) { consecutive_ios[i]->io_already_done = TRUE; } @@ -5085,16 +5419,16 @@ consecutive_loop: slot_io_done: - ut_a(slot->reserved); + ut_a(aio_slot->reserved); - *message1 = slot->message1; - *message2 = slot->message2; + *message1 = aio_slot->message1; + *message2 = aio_slot->message2; - *type = slot->type; + *type = aio_slot->type; os_mutex_exit(array->mutex); - os_aio_array_free_slot(array, slot); + os_aio_array_free_slot(array, aio_slot); return(ret); @@ -5113,30 +5447,20 @@ recommended_sleep: os_event_wait(os_aio_segment_wait_events[global_segment]); - if (os_aio_print_debug) { - fprintf(stderr, - "InnoDB: i/o handler thread for i/o" - " segment %lu wakes up\n", - (ulong) global_segment); - } - goto restart; } /**********************************************************************//** Validates the consistency of an aio array. -@return TRUE if ok */ +@return true if ok */ static -ibool +bool os_aio_array_validate( /*==================*/ os_aio_array_t* array) /*!< in: aio wait array */ { - os_aio_slot_t* slot; - ulint n_reserved = 0; ulint i; - - ut_a(array); + ulint n_reserved = 0; os_mutex_enter(array->mutex); @@ -5144,6 +5468,8 @@ os_aio_array_validate( ut_a(array->n_segments > 0); for (i = 0; i < array->n_slots; i++) { + os_aio_slot_t* slot; + slot = os_aio_array_get_nth_slot(array, i); if (slot->reserved) { @@ -5156,7 +5482,7 @@ os_aio_array_validate( os_mutex_exit(array->mutex); - return(TRUE); + return(true); } /**********************************************************************//** @@ -5168,10 +5494,22 @@ os_aio_validate(void) /*=================*/ { os_aio_array_validate(os_aio_read_array); - os_aio_array_validate(os_aio_write_array); - os_aio_array_validate(os_aio_ibuf_array); - os_aio_array_validate(os_aio_log_array); - os_aio_array_validate(os_aio_sync_array); + + if (os_aio_write_array != 0) { + os_aio_array_validate(os_aio_write_array); + } + + if (os_aio_ibuf_array != 0) { + os_aio_array_validate(os_aio_ibuf_array); + } + + if (os_aio_log_array != 0) { + os_aio_array_validate(os_aio_log_array); + } + + if (os_aio_sync_array != 0) { + os_aio_array_validate(os_aio_sync_array); + } return(TRUE); } @@ -5211,65 +5549,36 @@ os_aio_print_segment_info( } /**********************************************************************//** -Prints info of the aio arrays. */ +Prints info about the aio array. */ UNIV_INTERN void -os_aio_print( -/*=========*/ - FILE* file) /*!< in: file where to print */ +os_aio_print_array( +/*==============*/ + FILE* file, /*!< in: file where to print */ + os_aio_array_t* array) /*!< in: aio array to print */ { - os_aio_array_t* array; - os_aio_slot_t* slot; - ulint n_reserved; - ulint n_res_seg[SRV_MAX_N_IO_THREADS]; - time_t current_time; - double time_elapsed; - double avg_bytes_read; - ulint i; - - for (i = 0; i < srv_n_file_io_threads; i++) { - fprintf(file, "I/O thread %lu state: %s (%s)", (ulong) i, - srv_io_thread_op_info[i], - srv_io_thread_function[i]); - -#ifndef __WIN__ - if (os_aio_segment_wait_events[i]->is_set) { - fprintf(file, " ev set"); - } -#endif - - fprintf(file, "\n"); - } - - fputs("Pending normal aio reads:", file); - - array = os_aio_read_array; -loop: - ut_a(array); + ulint n_reserved = 0; + ulint n_res_seg[SRV_MAX_N_IO_THREADS]; os_mutex_enter(array->mutex); ut_a(array->n_slots > 0); ut_a(array->n_segments > 0); - n_reserved = 0; - memset(n_res_seg, 0x0, sizeof(n_res_seg)); - for (i = 0; i < array->n_slots; i++) { - ulint seg_no; + for (ulint i = 0; i < array->n_slots; ++i) { + os_aio_slot_t* slot; + ulint seg_no; slot = os_aio_array_get_nth_slot(array, i); seg_no = (i * array->n_segments) / array->n_slots; + if (slot->reserved) { - n_reserved++; - n_res_seg[seg_no]++; -#if 0 - fprintf(stderr, "Reserved slot, messages %p %p\n", - (void*) slot->message1, - (void*) slot->message2); -#endif + ++n_reserved; + ++n_res_seg[seg_no]; + ut_a(slot->len > 0); } } @@ -5281,38 +5590,61 @@ loop: os_aio_print_segment_info(file, n_res_seg, array); os_mutex_exit(array->mutex); +} - if (array == os_aio_read_array) { - fputs(", aio writes:", file); +/**********************************************************************//** +Prints info of the aio arrays. */ +UNIV_INTERN +void +os_aio_print( +/*=========*/ + FILE* file) /*!< in: file where to print */ +{ + time_t current_time; + double time_elapsed; + double avg_bytes_read; - array = os_aio_write_array; + for (ulint i = 0; i < srv_n_file_io_threads; ++i) { + fprintf(file, "I/O thread %lu state: %s (%s)", + (ulong) i, + srv_io_thread_op_info[i], + srv_io_thread_function[i]); - goto loop; +#ifndef __WIN__ + if (os_aio_segment_wait_events[i]->is_set) { + fprintf(file, " ev set"); + } +#endif /* __WIN__ */ + + fprintf(file, "\n"); } - if (array == os_aio_write_array) { - fputs(",\n ibuf aio reads:", file); - array = os_aio_ibuf_array; + fputs("Pending normal aio reads:", file); - goto loop; + os_aio_print_array(file, os_aio_read_array); + + if (os_aio_write_array != 0) { + fputs(", aio writes:", file); + os_aio_print_array(file, os_aio_write_array); } - if (array == os_aio_ibuf_array) { - fputs(", log i/o's:", file); - array = os_aio_log_array; + if (os_aio_ibuf_array != 0) { + fputs(",\n ibuf aio reads:", file); + os_aio_print_array(file, os_aio_ibuf_array); + } - goto loop; + if (os_aio_log_array != 0) { + fputs(", log i/o's:", file); + os_aio_print_array(file, os_aio_log_array); } - if (array == os_aio_log_array) { + if (os_aio_sync_array != 0) { fputs(", sync i/o's:", file); - array = os_aio_sync_array; - - goto loop; + os_aio_print_array(file, os_aio_sync_array); } putc('\n', file); - current_time = time(NULL); + current_time = ut_time(); time_elapsed = 0.001 + difftime(current_time, os_last_printout); fprintf(file, @@ -5320,7 +5652,8 @@ loop: "%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n", (ulong) fil_n_pending_log_flushes, (ulong) fil_n_pending_tablespace_flushes, - (ulong) os_n_file_reads, (ulong) os_n_file_writes, + (ulong) os_n_file_reads, + (ulong) os_n_file_writes, (ulong) os_n_fsyncs); if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) { @@ -5392,21 +5725,29 @@ os_aio_all_slots_free(void) os_mutex_exit(array->mutex); - array = os_aio_write_array; + if (!srv_read_only_mode) { + ut_a(os_aio_write_array == 0); - os_mutex_enter(array->mutex); + array = os_aio_write_array; - n_res += array->n_reserved; + os_mutex_enter(array->mutex); - os_mutex_exit(array->mutex); + n_res += array->n_reserved; - array = os_aio_ibuf_array; + os_mutex_exit(array->mutex); - os_mutex_enter(array->mutex); + ut_a(os_aio_ibuf_array == 0); - n_res += array->n_reserved; + array = os_aio_ibuf_array; - os_mutex_exit(array->mutex); + os_mutex_enter(array->mutex); + + n_res += array->n_reserved; + + os_mutex_exit(array->mutex); + } + + ut_a(os_aio_log_array == 0); array = os_aio_log_array; |