The FatFs module is assuming following conditions on portability.
The dependency diagram shown below is a typical configuration of the embedded system with FatFs module.
(a) If a working disk module with FatFs API is provided, no additional function is needed. (b) To attach existing disk drivers with different API, glue functions are needed to translate the APIs between FatFs and the drivers.
You need to provide only low level disk I/O functions that required by FatFs module and nothing else. If a working disk module for the target is already existing, you need to write only glue functions to attach it to the FatFs module. If not, you need to port any other disk module or write it from scratch. Most of defined functions are not that always required. For example, disk write function is not required in read-only configuration. Following table shows which function is required depends on configuration options.
|Always||Disk I/O functions.|
Samples available in ffsample.zip.
There are many implementations on the web.
|_FS_READONLY == 0|
|_USE_MKFS == 1|
|disk_ioctl (GET_SECTOR_SIZE)||_MAX_SS != _MIN_SS|
|disk_ioctl (CTRL_ERASE_SECTOR)||_USE_ERASE == 1|
|_USE_LFN >= 1||Unicode support functions.|
Available in option/unicode.c.
|_FS_REENTRANT == 1||O/S dependent functions.|
Samples available in option/syscall.c.
|_USE_LFN == 3|
|text (Full, R/W)||10675||7171||6617||13355||10940||11722||13262||8113||9048||6032||7952|
|text (Min, R/W)||6727||4631||4331||8569||7262||7720||9088||5287||5800||3948||5183|
|text (Full, R/O)||4731||3147||2889||6235||5170||5497||6482||3833||3972||2862||3719|
|text (Min, R/O)||3559||2485||2295||4575||4064||4240||5019||2993||3104||2214||2889|
|bss||V*4 + 2||V*4 + 2||V*4 + 2||V*2 + 2||V*4 + 2||V*2 + 2||V*2 + 2||V*4 + 2||V*4 + 2||V*4 + 2||V*4 + 2|
(_FS_TINY == 0)
(_FS_TINY == 1)
These are the memory usage on some target systems with following condition. The memory sizes are in unit of byte, V denotes number of volumes and F denotes number of open files. All samples are optimezed in code size.
FatFs R0.10a options: _FS_READONLY 0 (R/W) or 1 (R/O) _FS_MINIMIZE 0 (Full function) or 3 (Minimized function) _USE_STRFUNC 0 (Disable string functions) _USE_MKFS 0 (Disable f_mkfs function) _USE_FORWARD 0 (Disable f_forward function) _USE_FASTSEEK 0 (Disable fast seek feature) _CODE_PAGE 932 (Japanese Shift-JIS) _USE_LFN 0 (Disable LFN feature) _MAX_SS 512 (Fixed sector size) _FS_RPATH 0 (Disable relative path feature) _FS_LABEL 0 (Disable volume label functions) _VOLUMES V (Number of logical drives to be used) _MULTI_PARTITION 0 (Single partition per drive) _FS_REENTRANT 0 (Disable thread safe) _FS_LOCK 0 (Disable file lock control)
Follwing table shows which API function is removed by configuration options for the module size reduction.
FatFs module supports LFN (long file name). The two different file names, SFN (short file name) and LFN, of a file is transparent on the API except for f_readdir() function. The LFN feature is disabled by default. To enable it, set _USE_LFN to 1, 2 or 3, and add option/unicode.c to the project. The LFN feature requiers a certain working buffer in addition. The buffer size can be configured by _MAX_LFN according to the available memory. The length of an LFN will reach up to 255 characters, so that the _MAX_LFN should be set to 255 for full featured LFN operation. If the size of working buffer is insufficient for the input file name, the file function fails with FR_INVALID_NAME. When enable the LFN feature with re-entrant configuration, _USE_LFN must be set to 2 or 3. In this case, the file function allocates the working buffer on the stack or heap. The working buffer occupies (_MAX_LFN + 1) * 2 bytes.
|Code page||Program size|
When the LFN feature is enabled, the module size will be increased depends on the selected code page. Right table shows how many bytes increased when LFN feature is enabled with some code pages. Especially, in the CJK region, tens of thousands of characters are being used. Unfortunately, it requires a huge OEM-Unicode bidirectional conversion table and the module size will be drastically increased that shown in the table. As the result, the FatFs with LFN feature with those code pages will not able to be implemented to most 8-bit microcontrollers.
Note that the LFN feature on the FAT file system is a patent of Microsoft Corporation. This is not the case on FAT32 but most FAT32 drivers come with the LFN feature. FatFs can swich the LFN feature off by configuration option. When enable LFN feature on the commercial products, a license from Microsoft may be required depends on the final destination.
By default, FatFs uses ANSI/OEM code set on the API under LFN configuration. FatFs can also switch the character encoding to Unicode on the API by _LFN_UNICODE option. This means that the FatFs supports the True-LFN feature. For more information, refer to the description in the file name.
The file operations to the different volume is always re-entrant and can work simultaneously. The file operations to the same volume is not re-entrant but it can also be configured to thread-safe by _FS_REENTRANT option. In this case, also the OS dependent synchronization object control functions, ff_cre_syncobj(), ff_del_syncobj(), ff_req_grant() and ff_rel_grant() must be added to the project. There are some examples in the option/syscall.c.
When a file function is called while the volume is in use by any other task, the file function is suspended until that task leaves the file function. If wait time exceeded a period defined by _TIMEOUT, the file function will abort with FR_TIMEOUT. The timeout feature might not be supported by some RTOS.
There is an exception for f_mount(), f_mkfs(), f_fdisk() function. These functions are not re-entrant to the same volume or corresponding physical drive. When use these functions, all other tasks must unmount the volume and avoid to access the volume.
Note that this section describes on the re-entrancy of the FatFs module itself but also the low level disk I/O layer will need to be re-entrant.
FatFs module does not support the read/write collision control of duplicated open to a file. The duplicated open is permitted only when each of open method to a file is read mode. The duplicated open with one or more write mode to a file is always prohibited, and also open file must not be renamed and deleted. A violation of these rules can cause data colluption.
The file lock control can also be available by _FS_LOCK option. The value defines the number of open objects to manage simultaneously. In this case, if any open, rename or remove that violating the file shareing rule that described above is attempted, the file function will fail with FR_LOCKED. If number of open objects, files and sub-directories, gets larger than _FS_LOCK, the f_open(), f_optndir() function will fail with FR_TOO_MANY_OPEN_FILES.
For good read/write throughput on the small embedded systems with limited size of memory, application programmer should consider what process is done in the FatFs module. The file data on the volume is transferred in following sequence by f_read() function.
Figure 1. Sector misaligned read (short)
Figure 2. Sector misaligned read (long)
Figure 3. Fully sector aligned read
The file I/O buffer is a sector buffer to read/write a partial data on the sector. The sector buffer is either file private sector buffer on each file object or shared sector buffer in the file system object. The buffer configuration option _FS_TINY determins which sector buffer is used for the file data transfer. When tiny buffer (1) is selected, data memory consumption is reduced 512 bytes each file object. In this case, FatFs module uses only a sector buffer in the file system object for file data transfer and FAT/directory access. The disadvantage of the tiny buffer configuration is: the FAT data cached in the sector buffer will be lost by file data transfer and it must be reloaded at every cluster boundary. However it will be suitable for most application from view point of the decent performance and low memory comsumption.
Figure 1 shows that a partial sector, sector misaligned part of the file, is transferred via the file I/O buffer. At long data transfer shown in Figure 2, middle of transfer data that covers one or more sector is transferred to the application buffer directly. Figure 3 shows that the case of entier transfer data is aligned to the sector boundary. In this case, file I/O buffer is not used. On the direct transfer, the maximum extent of sectors are read with disk_read() function at a time but the multiple sector transfer is divided at cluster boundary even if it is contiguous.
Therefore taking effort to sector aligned read/write accesss eliminates buffered data transfer and the read/write performance will be improved. Besides the effect, cached FAT data will not be flushed by file data transfer at the tiny configuration, so that it can achieve same performance as non-tiny configuration with small memory footprint.
To maximize the write performance of flash memory media, such as SDC, CFC and U Disk, it must be controlled in consideration of its characteristitcs.
The write throughput of the flash memory media becomes the worst at single sector write transaction. The write throughput increases as the number of sectors per a write transaction. This effect more appers at faster interface speed and the performance ratio often becomes grater than ten. This graph is clearly explaining how fast is multiple block write (W:16K, 32 sectors) than single block write (W:100, 1 sector), and also larger card tends to be slow at single block write. The number of write transactions also affects the life time of the flash memory media. Therefore the application program should write the data in large block as possible. The ideal write chunk size and alighment is size of sector, and size of cluster is the best. Of course all layers between the application and the storage device must have consideration on multiple sector write, however most of open-source disk drivers lack it. Do not split a multiple sector write request into single sector write transactions or the write throughput gets poor. Note that FatFs module and its sample disk drivers supprt multiple sector read/write feature.
When remove a file with f_remove() function, the data clusters occupied by the file are marked 'free' on the FAT. But the data sectors containing the file data are not that applied any process, so that the file data left occupies a part of the flash memory array as 'live block'. If the file data is forced erased on removing the file, those data blocks will be turned in to the free block pool. This may skip internal block erase operation to the data block on next write operation. As the result the write performance might be improved. FatFs can manage this feature by setting _USE_ERASE to 1. Note that this is an expectation of internal process of the flash memory storage and not that always effective. Also f_remove() function will take a time when remove a large file. Most applications will not need this feature.
If a write operation to the FAT volume is interrupted due to any accidental failure, such as sudden blackout, incorrect disk removal and unrecoverable disk error, the FAT structure on the volume can be broken. Following images shows the critical section of the FatFs module.
An interruption in the red section can cause a cross link; as a result, the object being changed can be lost. If an interruption in the yellow section is occured, there is one or more possibility listed below.
Each case does not affect the files that not opened in write mode. To minimize risk of data loss, the critical section can be minimized by minimizing the time that file is opened in write mode or using f_sync() function as shown in Figure 5.
These are examples of extended use of FatFs APIs. New item will be added whenever a useful code is found.
FatFs has being developped as a personal project of author, ChaN. It is free from the code anyone else wrote. Following code block shows a copy of the FatFs license document that included in the source files.
/*----------------------------------------------------------------------------/ / FatFs - FAT file system module R0.10b (C)ChaN, 2014 /-----------------------------------------------------------------------------/ / FatFs module is a generic FAT file system module for small embedded systems. / This is a free software that opened for education, research and commercial / developments under license policy of following trems. / / Copyright (C) 2014, ChaN, all right reserved. / / * The FatFs module is a free software and there is NO WARRANTY. / * No restriction on use. You can use, modify and redistribute it for / personal, non-profit or commercial products UNDER YOUR RESPONSIBILITY. / * Redistributions of source code must retain the above copyright notice. / /-----------------------------------------------------------------------------/
Therefore FatFs license is one of the BSD-style licenses but there is a significant feature. Because FatFs is for embedded projects, the conditions of redistributions in binary form, such as embedded code, hex file, binary library or any forms without source code, are not specified in order to extend usability for commercial products. The documentation of the distributions need not include about FatFs and its license document, and it may also. This is equivalent to the BSD 1-Clause License. Of course FatFs is compatible with the projects under GNU GPL. When redistribute the FatFs with any modification or branch it as a folk, the license can also be changed to GNU GPL or BSD-style license.