On Thu, Jan 12, 2012 at 01:06:27PM +0100, Pavel Březina wrote:
Dne 11.1.2012 13:47, Jan Zelený napsal(a):
>>https://fedorahosted.org/sssd/ticket/1110
>>
>>[PATCH 1/4]
>>I've removed FIXME and move the DEBUG message to a better place.
>
>I would have Acked that, but I agree with Jakub that this should be done as a
>part of fix for #1125
Sure. Squashed to the review patch.
>>[PATCH 2/4]
>>I've wrapped the data provider part with tevent request so I can reuse
>>the code in periodical updates.
>
>Nack,
>sdap_sudo.c:144 - please use sdap_handler_done() for better clarity
>I'm not quuite sure why are you swicthing to EAGAIN in
>sdap_sudo_load_sudoers_next_base(). I'd understand if you used EOK as well,
>but this way I doubt it has any advantage (doesn't even add lucidity IMO).
In sdap_sudo_reply(), please call return after the first
sdap_handler_done()
In general, try to mark functions that are private to the module as
"static".
The structure sdap_sudo_refresh_state does not have to be present in the
sdap_sudo.h header, it is internal to the sdap_sudo_refresh request. It
is better to define it next to the _send() function.
>>[PATCH 3/4]
>>Adds two new functions in sudo sysdb api that can set/retrieve the value
>>of 'refreshed' attribute (equivalent to 'has_enumerated'
attribute in
>>domain enumeration). If it is true, the first schedule will be delayed
>>for a few seconds so we don't slow down the start up process.
>
>Nack,
>
>I think that re-using sysdb_set_enumerated() and sysdb_has_enumarated() and
>writing wrappers around them is far better than copying the code, which is
>almost identical.
>
>>[PATCH 4/4]
>>Periodical update itself. It adds two new options:
>>ldap_sudo_refresh_enabled (bool) - equivalent to 'enumerate'
>>ldap_sudo_refresh_timeout (number) - equivalent to
>>'ldap_enumeration_refresh_timeout'
>
>Nack,
>Add new options to config description in src/config/, also please review if all
>of your original SUDO-related options are there as well
>
>I think that delaying the refresh for ldap_sudo_refresh_timeout should be used
>instead of hardcoded 10 seconds. Or do you think that default value for the
>option is too high for this?
I've used the value which is hard coded in sdap_id_setup_tasks() so
I assume that original author had some reason for it. Any opinions?1
I think that's OK.
>
>Nitpick: in the definition of sdap_sudo_refresh_timer(), you miss the
"static"
>keyword
>
>You are forgetting to free sudo_ctx in sdap_sudo_refresh_timer, which could be
>potentially deadly on long-running systems, since the refresh_ctx isn't freed
>anywhere as weel
>
>Please use the standard _send/_retry/_done/_recv function naming in
>sdap_sudo_timer.c
I chose naming convention that is introduced in ldap_id_enum.c
>In sdap_sudo_refresh_send, you create request with struct
>sdap_sudo_refresh_state as its state, but in the timeout handling procedure,
>you retrieve callback data of that request as struct sdap_sudo_refresh_ctx.
>Please correct that and keep in mind that since this is entirely different
>struct assigned to the request, freeing the request won't free the sudo_ctx
>created in sdap_sudo_refresh_timer().
sdap_sudo_refresh_timer():
struct sdap_sudo_refresh_ctx *refresh_ctx;
req = sdap_sudo_refresh_send(sudo_ctx, sudo_ctx);
tevent_req_set_callback(req, sdap_sudo_refresh_reschedule,
*refresh_ctx*);
sdap_sudo_refresh_send(TALLOC_CTX *mem_ctx,
struct sdap_sudo_ctx *sudo_ctx):
struct sdap_sudo_refresh_state *state;
req = tevent_req_create(mem_ctx, &state, struct
sdap_sudo_refresh_state);
Now in callback:
sdap_sudo_refresh_reschedule(struct tevent_req *req):
refresh_ctx = tevent_req_callback_data(req, struct
sdap_sudo_refresh_ctx);
*retrieves correct structure*
ret = sdap_sudo_refresh_recv(req, &sudo_ctx, &dp_error, &error);
*retrieves sdap_sudo_refresh_state and returns sudo_ctx*
talloc_zfree(req); *sdap_sudo_refresh_state is freed*
talloc_zfree(sudo_ctx); *sudo_ctx is freed*
refresh_ctx should be allocated for the lifetime of data provider.
The structure sdap_sudo_refresh_ctx should be internal to
sdap_sudo_timer.c and the module should expose a one-time setup function
that would allocate the context for future use.