RTEMS | Port FatFS to RTEMS with Benchmarking (#25)

[Sepehr Ganji (@Sepehr)]

Sepehr Ganji commented: https://gitlab.rtems.org/groups/rtems/-/work_items/25#note_129286


I've wrapped up the tests by fixing the `fsformat` and `fserror` test:

1. **Fixed Parameter Structure Mismatch in Formatting**  
   The test suite used a custom `mkfs_parm` structure with mismatched fields compared to FatFS’s `MKFS_PARM`. This caused parameter corruption when calling `f_mkfs()`. I wrote a conversion function in `fatfs_format_disk()` to translate between the two structures, handling field mapping and unit conversions correctly.

2. **Removed Invalid Parameter Test Expectation**  
   The tests expected FatFS to reject invalid parameters (e.g., `num_fat=7`), but FatFS clamps invalid values instead of failing. This mismatch caused assertion errors. I removed the invalid parameter test and added documentation clarifying FatFS’s actual behavior.

3. **Corrected Hardcoded Block Size in RTEMS Wrapper**  
   The wrapper always set `st_blksize = 512`, ignoring FatFS’s actual cluster size. This caused cluster-related assertions to fail. I fixed this by reading the cluster size from `fs_info->fatfs.csize` and calculating the correct values for both `st_blksize` and `st_blocks`.

4. **Aligned Test Expectations with FatFS Defaults**  
   The tests assumed specific default cluster sizes, but FatFS determines them dynamically based on volume size and constraints. For example, the test expected 8 sectors/cluster but FatFS chose 4. I updated the test expectations to match FatFS’s real behavior, ensuring consistency.

5. **Enabled Large Volume Formatting**  
   The tests tried formatting volumes up to 128GB using `FM_FAT` (FAT12/16), which is invalid because FAT16 only supports up to ~2GB. FatFS returned `FR_MKFS_ABORTED` for larger volumes. I changed the format type to `FM_ANY`, allowing FatFS to automatically select the appropriate FAT type (12/16/32), enabling successful formatting of all volumes.

6. **Fixed Incorrect Error on Invalid Path Traversal**  
   When unlinking a path like `"tmp/file/dir"` where `"file"` was not a directory, the test expected `ENOTDIR` but got `EACCES`. The issue was in `rtems_fatfs_eval_token()`, which attempted to traverse through regular files. I added a check to validate directory nodes before traversal, returning `ENOTDIR` immediately when invalid, achieving POSIX compliance.

7. **Implemented Proper lseek() EOVERFLOW Handling**  
   The `lseek()` test with extremely large offsets expected `EOVERFLOW` but returned `EINVAL` due to integer overflow during arithmetic. I fixed this by adding proactive overflow checks before performing calculations in both `SEEK_CUR` and `SEEK_END`. Now operations that would exceed the maximum representable offset correctly return `EOVERFLOW`.

-- 
View it on GitLab: https://gitlab.rtems.org/groups/rtems/-/work_items/25#note_129286
You're receiving this email because of your account on gitlab.rtems.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/bugs/attachments/20250822/1d1f76d7/attachment.htm>