Folks, at some point somebody decided to use ->i_size for
all pipes (for amount of bytes in flight, AFAICS). Which might be
a nice thing, except that for FIFOs it leads to interesting
problems.
Think what happens if you get FIFO inode dirtied (for any
reason - chown(), atime, whatever) while you have pending data
in it. And then closed by everyone.
As far as I can see _none_ of the filesystems is prepared
to deal with that. Result: entries of inode table with type S_FIFO
and size != 0. And yes, it actually happens - /dev/xconsole is
one of the usual victims.
Potential solutions:
a) don't use i_size for pipes.
b) teach all ->write_inode() instances to write zero size for FIFOs,
regardless of ->i_size.
IMO (a) is easier and less prone to fsckups, so I'd go for
that variant. Linus, your opinion?
Al
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]