Question regarding direntry read error in fat filesystem when a device is removed

Question regarding direntry read error in fat filesystem when a device is removed

[YS Kim]

Hi,

I have a question about too many error messages from FAT when
unexpected device removal occurs.
I got the following error messages when I unplugged a USB
drive(formatted as FAT) during direntry reading.

sd 0:0:0:0: rejecting I/O to device being removed
FAT: Directory bread(block 1478192) failed
sd 0:0:0:0: rejecting I/O to device being removed
FAT: Directory bread(block 1478193) failed
sd 0:0:0:0: rejecting I/O to device being removed
FAT: Directory bread(block 1478194) failed
sd 0:0:0:0: rejecting I/O to device being removed
FAT: Directory bread(block 1478195) failed
...

>From the code of fat__get_entry in dir.c, I found that it continues to
read the next device block
even if the block reading has been failed. Because of this continuous
read, a lot of error messages
come up if the device itself is removed during reading fairly many
files, and it causes the malfunction
of my embedded system.

I think this code is to make the filesystem available to read the next
block when a storage contains
bad blocks, so that a user can backup remainings in case that some
portion of a storage are dead.

However, if the storage is a NAND flash memory, the flash driver is
supposed to perform bad block
management (because of the NAND charateristics) before the filesystem
encounters the bad block.

Thus, if I modify the code to just return error when fat__get_entry
fails to read blocks, could it be
cause any side effects? Except the purpose of backup/recovery, is
there any other reason for
the continuous block read?

Re: Question regarding direntry read error in fat filesystem when a device is removed

[Bodo Eggert]

YS Kim <historic59@gmail.com> wrote:

> I have a question about too many error messages from FAT when
> unexpected device removal occurs.
> I got the following error messages when I unplugged a USB
> drive(formatted as FAT) during direntry reading.
> 
> sd 0:0:0:0: rejecting I/O to device being removed
> FAT: Directory bread(block 1478192) failed
> sd 0:0:0:0: rejecting I/O to device being removed
> FAT: Directory bread(block 1478193) failed
> sd 0:0:0:0: rejecting I/O to device being removed
> FAT: Directory bread(block 1478194) failed
> sd 0:0:0:0: rejecting I/O to device being removed
> FAT: Directory bread(block 1478195) failed
> ...
> 
> From the code of fat__get_entry in dir.c, I found that it continues to
> read the next device block
> even if the block reading has been failed. Because of this continuous
> read, a lot of error messages
> come up if the device itself is removed during reading fairly many
> files, and it causes the malfunction
> of my embedded system.
[...]
> Thus, if I modify the code to just return error when fat__get_entry
> fails to read blocks, could it be
> cause any side effects? Except the purpose of backup/recovery, is
> there any other reason for
> the continuous block read?

I guess aborting the directory read would work as expected, but what about
suppressing the messages if you're not sure?