Re: [PATCH] Btrfs-progs: check out if the swap device

[Tsutomu Itoh <t-itoh@jp.fujitsu.com>]

On 2013/02/14 6:58, Zach Brown wrote:
>>> why it failed.  But it might not be able to because /proc/swaps is
>>> fundamentally unreliable.
>>
>> Then, how should we do?    I have no idea...
>
> Hmm.  I think I'd do something like:
>
> - First always open with O_EXCL.  If it succeeds then there's no reason
>    to check /proc/swaps at all.  (Maybe it doesn't need to try
>    check_mounted() there either?  Not sure if it's protecting against
>    accidentally mounting mounted shared storage or not.)

check_mounted() is necessary for multiple devices, I think.

>
> - Try stat()ing the /proc/swaps paths and the command line path.  If they
>    point to the same inode then print a helpful message that the open
>    might have failed because the file is an active swap file.
>
> - Use realpath() to resolve the relative path into an absolute path.
>    Copy it and escape control chars ("\n\t\\") with their \0xxx octal
>    equivalents.  If the mangled absolute path matches the path in
>    /proc/swaps (without opening), print the helpful message.

I think realpath() is unnecessary if it checks it by using only
stat() information. (if do not compare path)

Am I misunderstanding anything?

Thanks,
Tsutomu

>
> - At no point is failure of any of the /proc/swaps parsing fatal.  It'd
>    carry on ignoring errors until it doesnt have work to do.  It'd only
>    ever print the nice message when it finds a match.
>
> That seems reasonable to me.  It costs nothing in the vast majority of
> invocations when nothing goes wrong, it doesn't *cause* problems, and
> it'd print helpful messages on boring normal systems when someone really
> does accidentally try and mkfs a swapfile.
>
> In very rare cases /proc/swaps won't be of any help.  The user would
> only see the open failure.  That's fine, it's just not worth worrying
> about.
>
> - z