2010-10-27 17:56:41

by Dmitriy Paliy

[permalink] [raw]
Subject: [PATCH v2] Fix obexd crash for empty listing invalid cache

This fixes obexd crash in 3-way calling scenario when listing response is
empty. Valid cache and empty pbap buffer mean that cache was already attempted
to be created within a single session, but no data was available. Hence, it
is not notified and no such file error returned. New cache is not created
within current obex session or unless path is changed. Such removes necessity
of querying and filtering contacts for each incoming call in the other case,
which is extensive for large phone books. On the other hand, if user updates
contacts, cache will not be renewed till obex session is closed or path is
changed. Therefore TODO: note is added that clear of cache should be defined
besides of end of session or change of path.
---
plugins/pbap.c | 14 ++++++++++++++
1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/plugins/pbap.c b/plugins/pbap.c
index 11cb678..2a90d4d 100644
--- a/plugins/pbap.c
+++ b/plugins/pbap.c
@@ -751,6 +751,20 @@ static void *vobject_list_open(const char *name, int oflag, mode_t mode,
/* PullvCardListing always get the contacts from the cache */

if (pbap->cache.valid) {
+ /*
+ * Valid cache and empty buffer mean that cache was already
+ * created within a single session, but no data is available.
+ * New cache will not be created in current obex session or
+ * unless path is changed. If user updates contacts, cache
+ * will not be renewed, and, therefore:
+ * TODO: Define clear cache besides end of session or change
+ * of path.
+ */
+ if (!pbap->buffer) {
+ ret = -ENOENT;
+ goto fail;
+ }
+
cache_ready_notify(pbap);
goto done;
}
--
1.7.0.4



2010-10-28 08:07:33

by Dmitriy Paliy

[permalink] [raw]
Subject: Re: [PATCH v2] Fix obexd crash for empty listing invalid cache

Hi Johan,

On Wed, 2010-10-27 at 21:03 +0200, ext Johan Hedberg wrote:
> Hi Dmitriy,
>
> On Wed, Oct 27, 2010, Dmitriy Paliy wrote:
> > This fixes obexd crash in 3-way calling scenario when listing response is
> > empty. Valid cache and empty pbap buffer mean that cache was already attempted
> > to be created within a single session, but no data was available. Hence, it
> > is not notified and no such file error returned. New cache is not created
> > within current obex session or unless path is changed. Such removes necessity
> > of querying and filtering contacts for each incoming call in the other case,
> > which is extensive for large phone books. On the other hand, if user updates
> > contacts, cache will not be renewed till obex session is closed or path is
> > changed. Therefore TODO: note is added that clear of cache should be defined
> > besides of end of session or change of path.
>
> The commit message seems to have a max width of 78 characters which
> means that it's not viewable with git log on a 80 character terminal
> (git log indents the output by 4 characters). Please keep the max commit
> message width to 74 characters or so. Also, if possible try to split it
> up into multiple paragraphs. Paragraphs longer than 6 lines tend to be a
> bit harder to follow.
>
> > + * TODO: Define clear cache besides end of session or change
> > + * of path.
>
> That doesn't sound like proper english to me and I'm not sure what
> you're trying to say. Should it be "Define a clear distinction between
> end of session and change of path"?

Thanks for comments. What I meant is following. Cache is created at the
moment when there is an incoming call in the middle of another ongoing
call. Clear cache and create a new one when the call is retried, which
is one of possible solutions to fix this bug, looks inefficient to me.
Therefore, already created cache is valid either till session is closed
or path is changed. However, if user updates his contacts, then changes
will not affect cache within these limits. What I was trying to say in
the todo note is that in order to improve logic when cache is created
and cleared, besides two facts written above (end of session and change
of path), another way to renew cache should be defined.

Br,
Dmitriy


2010-10-27 19:03:47

by Johan Hedberg

[permalink] [raw]
Subject: Re: [PATCH v2] Fix obexd crash for empty listing invalid cache

Hi Dmitriy,

On Wed, Oct 27, 2010, Dmitriy Paliy wrote:
> This fixes obexd crash in 3-way calling scenario when listing response is
> empty. Valid cache and empty pbap buffer mean that cache was already attempted
> to be created within a single session, but no data was available. Hence, it
> is not notified and no such file error returned. New cache is not created
> within current obex session or unless path is changed. Such removes necessity
> of querying and filtering contacts for each incoming call in the other case,
> which is extensive for large phone books. On the other hand, if user updates
> contacts, cache will not be renewed till obex session is closed or path is
> changed. Therefore TODO: note is added that clear of cache should be defined
> besides of end of session or change of path.

The commit message seems to have a max width of 78 characters which
means that it's not viewable with git log on a 80 character terminal
(git log indents the output by 4 characters). Please keep the max commit
message width to 74 characters or so. Also, if possible try to split it
up into multiple paragraphs. Paragraphs longer than 6 lines tend to be a
bit harder to follow.

> + * TODO: Define clear cache besides end of session or change
> + * of path.

That doesn't sound like proper english to me and I'm not sure what
you're trying to say. Should it be "Define a clear distinction between
end of session and change of path"?

Johan