| 1 |
Von: Roman Zippel [[email protected]] |
| 2 |
Gesendet: Mittwoch, 30. August 2000 03:10 |
| 3 |
An: Alexander Viro |
| 4 |
Cc: Matthew Wilcox; Roman Zippel; linux-fsdevel; linux-kernel; |
| 5 |
[email protected] |
| 6 |
Betreff: Re: hfs support for blocksize != 512 |
| 7 |
|
| 8 |
|
| 9 |
Hi, |
| 10 |
|
| 11 |
> > hfs. For example reading from a file might require a read from a btree |
| 12 |
> > file (extent file), with what another file write can be busy with (e.g. |
| 13 |
> > reordering the btree nodes). |
| 14 |
> |
| 15 |
> And? |
| 16 |
|
| 17 |
The point is: the thing I like about Linux is its simple interfaces, it's |
| 18 |
the basic idea of unix - keep it simple. That is true for most parts - the |
| 19 |
basic idea is simple and the real complexity is hidden behind it. But |
| 20 |
that's currently not true for vfs interface, a fs maintainer has to fight |
| 21 |
right now with fscking complex vfs interface and with a possible fscking |
| 22 |
complex fs implementation. E2fs or affs have a pretty simple structure and |
| 23 |
I believe you that it's not that hard to fix, maybe there is also a simple |
| 24 |
solution for hfs. But I'd like you to forget about that and think about |
| 25 |
the big picture (how Linus nicely states it). What we should aim at with |
| 26 |
the vfs interface is simplicity, I want to use a fscking simple semaphore |
| 27 |
to protect something like anywhere else, I don't want to juggle with lots |
| 28 |
blocks wich have to be updated atomically. Maybe you get once right, but |
| 29 |
it will follow you as a nightmare, you add one feature (e.g. quota), you |
| 30 |
add another feature (like btrees), you so still damned fscking sure to get |
| 31 |
and keeping it right? |
| 32 |
So and? What I'd really like to see from you is to be a bit more |
| 33 |
supportive for other peoples problems, I really don't expect you to solve |
| 34 |
these problems, but if someone approaches a different solution, you're |
| 35 |
pretty quick to refuse it. |
| 36 |
So lets get back to the vfs interface, fs currently have to do pretty much |
| 37 |
all there changes atomically, they have to grab all the buffers they need |
| 38 |
and do all changes at once. How can you be sure that this is possible for |
| 39 |
every possible fs? How do you make sure you don't create other problems |
| 40 |
like livelocks? We currently have problem that things like kswapd require |
| 41 |
an asynchronous interface, but fs prefer to synchronize it. Currently you |
| 42 |
pushing all the burden of an asynchronous interface into the fs, which |
| 43 |
want to rather avoid that. Why don't you think for a moment in the other |
| 44 |
direction? Currently I'm playing with the idea of a kernel thread for |
| 45 |
asynchronous io (maybe one per fs), that thread takes the io requests e.g. |
| 46 |
from kswapd and the io thread can safely sleep on it, while kswapd can |
| 47 |
continue its job, but I don't know yet, where to put, whether in the fs |
| 48 |
specific part or whether it can be made generic enough to be put into the |
| 49 |
generic part. Can we please think for a moment in that direction? At some |
| 50 |
point you have to synchronize the io anyway (at latest when it hits the |
| 51 |
device), but I would pretty much prefer if a fs would get some help at |
| 52 |
some earlier point. |
| 53 |
(Anyway, I need some sleep now as well... :) ) |
| 54 |
|
| 55 |
bye, Roman |
| 56 |
|
| 57 |
|
| 58 |
** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/ |