2008-12-28 23:28:30

by Evgeniy Dushistov

Subject: [PATCH] minix: fix add link - wrong position calculation

This patch fixes add link method.
Position in directory was calculated in wrong way.

Signed-off-by: Evgeniy Dushistov <[email protected]>
---
fs/minix/dir.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/fs/minix/dir.c b/fs/minix/dir.c
index f704338..4b603b7 100644
--- a/fs/minix/dir.c
+++ b/fs/minix/dir.c
@@ -280,7 +280,7 @@ int minix_add_link(struct dentry *dentry, struct inode *inode)
return -EINVAL;

got_it:
- pos = (page->index >> PAGE_CACHE_SHIFT) + p - (char*)page_address(page);
+ pos = page_offset(page) + p - (char*)page_address(page);
err = __minix_write_begin(NULL, page->mapping, pos, sbi->s_dirsize,
AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
if (err)
--
1.6.0.6

--
/Evgeniy

2008-12-30 23:33:19

by Andrew Morton

Subject: Re: [PATCH] minix: fix add link - wrong position calculation

On Mon, 29 Dec 2008 02:25:50 +0300
Evgeniy Dushistov <[email protected]> wrote:

> This patch fixes add link method.
> Position in directory was calculated in wrong way.
>
> Signed-off-by: Evgeniy Dushistov <[email protected]>
> ---
> fs/minix/dir.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/fs/minix/dir.c b/fs/minix/dir.c
> index f704338..4b603b7 100644
> --- a/fs/minix/dir.c
> +++ b/fs/minix/dir.c
> @@ -280,7 +280,7 @@ int minix_add_link(struct dentry *dentry, struct inode *inode)
> return -EINVAL;
>
> got_it:
> - pos = (page->index >> PAGE_CACHE_SHIFT) + p - (char*)page_address(page);
> + pos = page_offset(page) + p - (char*)page_address(page);
> err = __minix_write_begin(NULL, page->mapping, pos, sbi->s_dirsize,
> AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
> if (err)

Well that's cute. Seems that we broke it over a year ago:

commit 4a66af9eaa9531372cfcb9e20103ed147c729ff9
Author: Nick Piggin <[email protected]>
AuthorDate: Tue Oct 16 01:25:21 2007 -0700
Commit: Linus Torvalds <[email protected]>
CommitDate: Tue Oct 16 09:42:57 2007 -0700

minixfs: convert to new aops

Is minixfs getting THAT little use, or is this bustage more subtle
than it appears?

2008-12-31 01:26:51

by Nick Piggin

Subject: Re: [PATCH] minix: fix add link - wrong position calculation

On Tue, Dec 30, 2008 at 03:32:30PM -0800, Andrew Morton wrote:
> On Mon, 29 Dec 2008 02:25:50 +0300
> Evgeniy Dushistov <[email protected]> wrote:
>
> > This patch fixes add link method.
> > Position in directory was calculated in wrong way.
> >
> > Signed-off-by: Evgeniy Dushistov <[email protected]>
> > ---
> > fs/minix/dir.c | 2 +-
> > 1 files changed, 1 insertions(+), 1 deletions(-)
> >
> > diff --git a/fs/minix/dir.c b/fs/minix/dir.c
> > index f704338..4b603b7 100644
> > --- a/fs/minix/dir.c
> > +++ b/fs/minix/dir.c
> > @@ -280,7 +280,7 @@ int minix_add_link(struct dentry *dentry, struct inode *inode)
> > return -EINVAL;
> >
> > got_it:
> > - pos = (page->index >> PAGE_CACHE_SHIFT) + p - (char*)page_address(page);
> > + pos = page_offset(page) + p - (char*)page_address(page);
> > err = __minix_write_begin(NULL, page->mapping, pos, sbi->s_dirsize,
> > AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
> > if (err)
>
> Well that's cute. Seems that we broke it over a year ago:
>
> commit 4a66af9eaa9531372cfcb9e20103ed147c729ff9
> Author: Nick Piggin <[email protected]>
> AuthorDate: Tue Oct 16 01:25:21 2007 -0700
> Commit: Linus Torvalds <[email protected]>
> CommitDate: Tue Oct 16 09:42:57 2007 -0700
>
> minixfs: convert to new aops
>
> Is minixfs getting THAT little use, or is this bustage more subtle
> than it appears?

Not getting used, I'd suspect (although I definitely have used it -- I
converted it to large block size for testing fsblock, probably I didn't
test softlinks).

But anyway I think it is still useful as a primer for ext2? Probably there
are quite a lot of other filesystems that are less useful than minix ;)

2008-12-31 01:49:35

by Al Viro

Subject: Re: [PATCH] minix: fix add link - wrong position calculation

On Wed, Dec 31, 2008 at 02:26:41AM +0100, Nick Piggin wrote:
> > commit 4a66af9eaa9531372cfcb9e20103ed147c729ff9
> > Author: Nick Piggin <[email protected]>
> > AuthorDate: Tue Oct 16 01:25:21 2007 -0700
> > Commit: Linus Torvalds <[email protected]>
> > CommitDate: Tue Oct 16 09:42:57 2007 -0700
> >
> > minixfs: convert to new aops
> >
> > Is minixfs getting THAT little use, or is this bustage more subtle
> > than it appears?
>
> Not getting used, I'd suspect (although I definitely have used it -- I
> converted it to large block size for testing fsblock, probably I didn't
> test softlinks).

Much simpler: there's one index for which the expressions are equivalent ;-)
Count how many entries does it take on minixfs to get to 4Kb. Now think
how many had used that sucker on directories with >= 128 entries...