fat: msdos_dir_read(..) doesn't reset conversion output string length

[slemstick]

msdos_dir_read(..) uses a conversion function, convert_handler->utf16_to_utf8, to convert LFN directory entry names in utf16 format to utf8.

However, the conversion handler sets the string length of the output utf8 string as well. That variable: string_size in msdos_dir_read(..) is never re-initialised in the search algorithm. When the volume becomes sufficiently fragmented, de-allocated LFN directory entry checksums will cause the filename search algorithm to fail, effectively breaking the current attempt to concatenate directory entry filename chunks, but the output string size is now much shorter (10 characters, where it should be sizeof(tmp_dirent.d_name)). Consequently, msdos_dir_read(..) will continue to parse directory entries with a much smaller output string size.

The end result is that attempts to read file names from a directory will output truncated file names (for example, readdir() will "work" as normal but the output filenames are too short). Any attempt to open these truncated file names will, of course, fail.

[Gedare Bloom]

Please remove the spurious whitespace changes, add "Closes #2987." Into the commit message, and format a friendly short commit message that specifies the rtems subsystem first, e.g. "libfs/dosfs: ..."
Please see Developer/Git?

[Sebastian Huber]

Without a real name I cannot apply this patch.

[Sebastian Huber <sebastian.huber@…>]

In 34dda604/rtems:

dosfs: Fix msdos_dir_read()

Set a proper name buffer length for each converter invocation.

Update #2987.

[Sebastian Huber <sebastian.huber@…>]

In e1c3dc09/rtems:

dosfs: Fix msdos_dir_read()

Set a proper name buffer length for each converter invocation.

Close #2987.