On 06/18/2014 02:07 PM, Pavel Reichl wrote:
Hello,
could we push the first already ACKed patch to master, please?
I have updated the other 3 patches as asked for, but these are not so
important and can wait (or be forgotten).
Thanks,
PR
On Fri, 2014-04-18 at 10:32 +0200, Jakub Hrozek wrote:
> On Thu, Apr 17, 2014 at 07:01:35PM +0200, Pavel Reichl wrote:
>> Hello,
>>
>> while working on
https://fedorahosted.org/sssd/ticket/2102 I noticed a
>> few minor problems regarding netgroups.
>>
>> 1st patch:
>> First check returned value then it is safe to access results.
>>
>> 2nd patch:
>> It is my opinion that function lookup_netgr_step is too long and can be
>> safely (with minimum of changes) divided into 2 separate functions.
>>
>> 3rd patch:
>> Pavel Brezina asked me to solve memory leak in lookup_netgr_step.
>>
>> 4th patch:
>> Fixes two code style issues.
>>
>> Thank for review!
>>
>> Pavel Reichl
>>
>
>> From d8f7815a69b676c65da189d01a999f2acfc7deec Mon Sep 17 00:00:00 2001
>> From: Pavel Reichl <preichl(a)redhat.com>
>> Date: Wed, 26 Feb 2014 16:58:24 +0000
>> Subject: [PATCH 2/5] NSS: sysdb_getnetgr check return value first
>>
>> Output parameter was accessed before return value was checked.
>
> ACK, nice catch!
Thanks.
Additional ack.
>
>> From da8ce47f42c6e551432a23ff813c938af6c5721a Mon Sep 17 00:00:00 2001
>> From: Pavel Reichl <preichl(a)redhat.com>
>> Date: Thu, 17 Apr 2014 16:14:11 +0000
>> Subject: [PATCH 3/5] NSS: sysdb_getnetgr refactor
>
>> + ret = create_dummy_netgr(step_ctx);
>> + if (ret != EOK) {
>> + return ret;
>
> I think there is a change in behaviour. Previously, if creating the
> negative entry failed, we would have just ignored the failure, the new
> code errors out. I think the failure can be ignored, because at worst,
> there would be a slowdown at the next lookup.
Agreed, fixed.
Patch 02:
You should return all errors from the create_dummy_netgr
properly. Here:
- netgr = talloc_zero(step_ctx->nctx, struct getent_ctx);
- if (netgr == NULL) {
- DEBUG(SSSDBG_CRIT_FAILURE, "talloc_zero failed, ignored.\n");
and here
- ret = set_netgroup_entry(step_ctx->nctx, netgr);
- if (ret != EOK) {
- DEBUG(SSSDBG_CRIT_FAILURE, "...");
- }
The caller can than ignore the errors (as it already
does). This is not a big deal, but I think it will be
nicer that way. Currently you only report
talloc_strdup failure to the caller and ignore the other
possible errors, which is a little inconsistent.
Also , I am not sure if create_dummy_netgr is good name
for the helper function. It creates negative cache entry,
so maybe something like create_negative_netgr?
>
>> From 85a1744b803e9d25a82f6ce83a9eaf3d9fdffe16 Mon Sep 17 00:00:00 2001
>> From: Pavel Reichl <preichl(a)redhat.com>
>> Date: Thu, 17 Apr 2014 16:52:43 +0000
>> Subject: [PATCH 4/5] NSS: fix memory leak in sysdb_getnetgr
>
> I'm not opposed to pushing this patch, but in general I think 'leaks'
> like this are not a big issue. The string was allocated on top of a
> per-request context, so when the request finishes, the string would be
> freed together with step_ctx.
Agree, I was asked to fix this by Pavel Brezina, but I don't think he
cares much about this.
Patch 03: Ack.
>
> Such growths are much more dangerous in long-running requests like
> nested group resolution, the frontend is usually quite fast and
> efficient.
>
>> From eab0ebf66cdc977656ef8964c7546f54212a08dd Mon Sep 17 00:00:00 2001
>> From: Pavel Reichl <preichl(a)redhat.com>
>> Date: Thu, 17 Apr 2014 17:02:31 +0000
>> Subject: [PATCH 5/5] NSS: minor code style improvements
>> @@ -644,10 +644,8 @@ static void lookup_netgr_dp_callback(uint16_t err_maj,
uint32_t err_min,
>>
>> /* ok the backend returned, search to see if we have updated results */
>> ret = lookup_netgr_step(step_ctx);
>> - if (ret != EOK) {
>> - if (ret == EAGAIN) {
>> - return;
>> - }
>> + if (ret == EAGAIN) {
>> + return;
>> }
>>
>> /* We have results to return */
>
> This patch is OK, but I wonder if you could also refactor the result
> handling. Currently nss_setent_notify_error() is used also for EOK which
> might be a bit confusing in comparison with tevent where tevent_req_done !=
> tevent_req_error(EOK). I think this would be more readable:
>
> if (ret == EOK) {
> setent_notify_done();
> } else {
> nss_setent_notify_error();
> }
Fixed.
Patch 04: Ack.
Michal