In <[EMAIL PROTECTED]>, on 03/20/03 
   at 06:28 PM, "Mitch \(WebCob\)" <[EMAIL PROTECTED]> said:

>Sam wrote:

>>> What is the reason to use soft links in the shared folders?
>>>
>>> I would think that hardlinks (with an independance of file
>>> permissions on
>>> the original file) would provide advantages from a group
>>> membership and maintenance point of view...
>>
>>The sharable mailbox may be on a different filesystem.

>I was 99% certain you were goign to say that... and a valid consideration
>-
>so to avoid breaking things we'd probably have to allow the possibiliy of
>both unless people were sure they didn't need the softlink option
>(./configure option?)

>>> COULD the shared folder softlink be changed to hardlinks? Would
>>> it be a BIG change?
>>
>>First of all, there needs to be a good reason for a change.
>>"Just for the heck of it" isn't going to cut the mustard.

>Of course. Should go without saying but often needs to be said anyways
>;-) Seriously though I think there are reasons to consider the change -
>it's a matter of whether or not the impact is worth it.

>1) copied messages could be handled with hardlinks instead of copies,
>resulting in less data transfer and less data storage
>2) shared folders could be accessed without reliance on common group
>membership (which admittedly could be more than a "simple" softlink ->
>hardlink change). Shared folders by hardlink would be faster (by how
>much?) without having to resolve the soft link, but the permissions issue
>was more of a problem for me before and I see it as a happy coincidense
>of changing to a hardlink based system.

>On FreeBSD 4.x (maybe this is different elsewhere, so I note it here for
>later arguments ;-) one can maintain hardlinks to a single file with
>completely different ownership information.

>i.e.
>touch stuff1
>chown user1:user2 stuff1
>chmod 700 stuff1
>ln stuff1 stuff2
>chown user1:user1 stuff2
>chmod 700 stuff2

>two names, one file, completely different and independant owners /
>groups, ONE data file.

>On BSD there are limits to the number of groups a user can belong to,
>which can be increased by rebuilding a custom kernel (16 is the limit I
>think) which unfortunately limits the complexity of managing shared
>mailing lists... you can't have each list be a group, or people run out
>of group membership options. Changing the limit is cautioned against
>without a makeworld and also the warning that some applications MAY have
>been hardcoded to the original value.

>of course with alternative authentication systems (mysql) this can
>probably be bypassed, but I was never confident that one wouldn't end up
>bumping into the OS limit at some point as the daemon tried to access
>files it would rely on the OS environment of permissions one it sets that
>user id...

>With hardlinks, a user can own their own copy of their messages which
>just HAPPEN to be linked to the shared copy.

>am I making sense? what do you think?

>m/

Butting in here, but adding & supporting that extra sort of 'option'
complexity w/r file storage doesn't seem to me to be an 'improvement' to
courier-mta.

Instead, what you want seems better suited to a DB-driven approach.  There
can be overhead in that - LOTS of it, but not always as bad as it might
seem...

If the DB stores just the path to the file containing the message, *and*
to whom it may be released, then it is up to the implementor where the
files are stored and how, and what tradeoffs are tolerable w/r speed,
security, portability, redundancy, etc.

I have been looking at ways to do something similar with integrating
postgresql with courier-mta.  have a look at what powerdns.com are doing
with their PowerMail product, which I regard as more concept-inspirational
than ready-for-production for what I need.

Regards,

Bill Hacker
-- 



-------------------------------------------------------
This SF.net email is sponsored by: Tablet PC.  
Does your code think in ink? You could win a Tablet PC. 
Get a free Tablet PC hat just for playing. What are you waiting for? 
http://ads.sourceforge.net/cgi-bin/redirect.pl?micr5043en
_______________________________________________
courier-users mailing list
[EMAIL PROTECTED]
Unsubscribe: https://lists.sourceforge.net/lists/listinfo/courier-users

Reply via email to