LinuxTV Patchwork vdr 1.7.23: patch for handling symlinks in recordings directory as earlier

login
register
mail settings
Submitter Klaus Schmidinger
Date Feb. 13, 2012, 9:43 a.m.
Message ID <4F38DB57.1000508@tvdr.de>
Download mbox | patch
Permalink /patch/12927/
State New
Headers show

Comments

Klaus Schmidinger - Feb. 13, 2012, 9:43 a.m.
> On 01/27/2012 01:04 PM, Oliver Endriss wrote:
>> On Thursday 26 January 2012 11:07:18 Klaus Schmidinger wrote:
>>> On 25.01.2012 14:11, Oliver Endriss wrote:
>>>> On Wednesday 25 January 2012 10:29:16 Klaus Schmidinger wrote:
>>>>> On 17.01.2012 14:26, sundararaj reel wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I am attaching a patch for vdr 1.7.23 for the problem described here:
>>>>>> http://www.vdr-portal.de/board1-news/board2-vdr-news/p1047199-announce-vdr-developer-version-1-7-23/#post1047199
>>>>>>
>>>>>> There appears to be a problem in listing recordings due to a bug fix
>>>>>> in vdr 1.7.23. "Fixed handling symbolic links in
>>>>>> cRecordings::ScanVideoDir()"
>>>>>>
>>>>>> The attached patch just disables the translation of symbolic links to
>>>>>> "real" paths. So that all recordings appear to be under the same
>>>>>> (recordings) directory tree, as it was earlier.
>>>>>>
>>>>>> Please reply with your results.
>>>>>
>>>>> Can somebody who has actually this use case please confirm
>>>>> whether this patch fixes the problem?
>>>>
>>>> Confirmed.
>>>>
>>>> Without this patch, symbolic links are not displayed
>>>> correctly on my machine.
>>>>
>>>> Oliver
>>>
>>> Thanks.
>>>
>>> I believe the second call to stat() is now superfluous.
>>> Can you please confirm that the following patch still works
>>> as expected?
>>>
>>> --- recording.c 2012/01/25 09:32:39     2.45
>>> +++ recording.c 2012/01/26 10:02:29
>>> @@ -1120,11 +1120,6 @@
>>>                        continue;
>>>                        }sundararaj reel

How did this "sundararaj reel" get in here?

>>>                     Link = 1;
>>> -                 buffer = ReadLink(buffer);
>>> -                 if (!*buffer)
>>> -                    continue;
>>> -                 if (stat(buffer,&st) != 0)
>>> -                    continue;
>>>                     }
>>>                  if (S_ISDIR(st.st_mode)) {
>>>                     if (endswith(buffer, deleted ? DELEXT : RECEXT)) {
>>
>> Yes, it does not make any difference here.
>
> Well, with this patch, symbolic links are not displayed at all on my VDR machine, whereas with sundararaj reel's fix they are displayed correctly.

So what you're saying it that this...

----------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------

...doesn't?

I find that hard to believe, because the only difference here is that
the second version removes the stat() call, which is superfluous if
'buffer' is no longer modified.

Klaus
Lucian Muresan - Feb. 13, 2012, 10:44 a.m.
On 13.02.2012 10:43, Klaus Schmidinger wrote:
>> On 01/27/2012 01:04 PM, Oliver Endriss wrote:
>>> On Thursday 26 January 2012 11:07:18 Klaus Schmidinger wrote:
>>>> On 25.01.2012 14:11, Oliver Endriss wrote:
>>>>> On Wednesday 25 January 2012 10:29:16 Klaus Schmidinger wrote:
>>>>>> On 17.01.2012 14:26, sundararaj reel wrote:

[...]

>> Well, with this patch, symbolic links are not displayed at all on my
>> VDR machine, whereas with sundararaj reel's fix they are displayed
>> correctly.
>
> So what you're saying it that this...
>
> ----------------------------------------------------------------------------------------------
>
> --- a/recording.c
> +++ b/recording.c
> @@ -1120,9 +1120,13 @@ void cRecordings::ScanVideoDir(const char
> *DirName, bool Foreground, int LinkLev
> continue;
> }
> Link = 1;
> +#if 0
> + // do not resolve the symbolic links in paths to real path
> + // thereby keeping all the recordings under one directory
> buffer = ReadLink(buffer);
> if (!*buffer)
> continue;
> +#endif
> if (stat(buffer, &st) != 0)
> continue;
> }
> ----------------------------------------------------------------------------------------------
>
>
> ...works, while this...
>
> ----------------------------------------------------------------------------------------------
>
> --- recording.c 2012/01/25 09:32:39 2.45
> +++ recording.c 2012/01/26 10:02:29
> @@ -1120,11 +1120,6 @@
> continue;
> }
> Link = 1;
> - buffer = ReadLink(buffer);
> - if (!*buffer)
> - continue;
> - if (stat(buffer, &st) != 0)
> - continue;
> }
> if (S_ISDIR(st.st_mode)) {
> if (endswith(buffer, deleted ? DELEXT : RECEXT)) {
> ----------------------------------------------------------------------------------------------
>
>
> ...doesn't?
>
> I find that hard to believe, because the only difference here is that
> the second version removes the stat() call, which is superfluous if
> 'buffer' is no longer modified.

Haven't really looked at the code, until now, and I also do not exactly 
know what the call to stat does and also didn't try to understand the 
whole picture now, but your patch does not simply remove just the call 
to stat, but also a *continue* statement from the *while* loop, this 
could have strong implications, so just please consider analyzing the 
issue with respect to that.

Regards,
Lucian
Klaus Schmidinger - Feb. 13, 2012, 11 a.m.
On 13.02.2012 11:44, Lucian Muresan wrote:
> On 13.02.2012 10:43, Klaus Schmidinger wrote:
>>> On 01/27/2012 01:04 PM, Oliver Endriss wrote:
>>>> On Thursday 26 January 2012 11:07:18 Klaus Schmidinger wrote:
>>>>> On 25.01.2012 14:11, Oliver Endriss wrote:
>>>>>> On Wednesday 25 January 2012 10:29:16 Klaus Schmidinger wrote:
>>>>>>> On 17.01.2012 14:26, sundararaj reel wrote:
>
> [...]
>
>>> Well, with this patch, symbolic links are not displayed at all on my
>>> VDR machine, whereas with sundararaj reel's fix they are displayed
>>> correctly.
>>
>> So what you're saying it that this...
>>
>> ----------------------------------------------------------------------------------------------
>>
>> --- a/recording.c
>> +++ b/recording.c
>> @@ -1120,9 +1120,13 @@ void cRecordings::ScanVideoDir(const char
>> *DirName, bool Foreground, int LinkLev
>> continue;
>> }
>> Link = 1;
>> +#if 0
>> + // do not resolve the symbolic links in paths to real path
>> + // thereby keeping all the recordings under one directory
>> buffer = ReadLink(buffer);
>> if (!*buffer)
>> continue;
>> +#endif
>> if (stat(buffer, &st) != 0)
>> continue;
>> }
>> ----------------------------------------------------------------------------------------------
>>
>>
>> ...works, while this...
>>
>> ----------------------------------------------------------------------------------------------
>>
>> --- recording.c 2012/01/25 09:32:39 2.45
>> +++ recording.c 2012/01/26 10:02:29
>> @@ -1120,11 +1120,6 @@
>> continue;
>> }
>> Link = 1;
>> - buffer = ReadLink(buffer);
>> - if (!*buffer)
>> - continue;
>> - if (stat(buffer, &st) != 0)
>> - continue;
>> }
>> if (S_ISDIR(st.st_mode)) {
>> if (endswith(buffer, deleted ? DELEXT : RECEXT)) {
>> ----------------------------------------------------------------------------------------------
>>
>>
>> ...doesn't?
>>
>> I find that hard to believe, because the only difference here is that
>> the second version removes the stat() call, which is superfluous if
>> 'buffer' is no longer modified.
>
> Haven't really looked at the code, until now, and I also do not exactly know what the call to stat does and also didn't try to understand the whole picture now, but your patch does not simply remove just the call to stat, but also a *continue* statement from the *while* loop, this could have strong
> implications, so just please consider analyzing the issue with respect to that.

The original code was

----------------------------------------------------------------------------------------------
            if (stat(buffer, &st) == 0) {
               int Link = 0;
               if (S_ISLNK(st.st_mode)) {
                  if (LinkLevel > MAX_LINK_LEVEL) {
                     isyslog("max link level exceeded - not scanning %s", *buffer);
                     continue;
                     }
                  Link = 1;
                  buffer = ReadLink(buffer);
                  if (!*buffer)
                     continue;
                  if (stat(buffer, &st) != 0)
                     continue;
                  }
               ...
               }
----------------------------------------------------------------------------------------------

After Sundararaj's patch it looked like this (just leaving out the lines
that his '#if 0' disabled):

----------------------------------------------------------------------------------------------
            if (stat(buffer, &st) == 0) {
               int Link = 0;
               if (S_ISLNK(st.st_mode)) {
                  if (LinkLevel > MAX_LINK_LEVEL) {
                     isyslog("max link level exceeded - not scanning %s", *buffer);
                     continue;
                     }
                  if (stat(buffer, &st) != 0)
                     continue;
                  }
               ...
               }
----------------------------------------------------------------------------------------------

Reducing this to the stat() calls results in

----------------------------------------------------------------------------------------------
            if (stat(buffer, &st) == 0) {
               ...
                  if (stat(buffer, &st) != 0)
                     continue;
                  }
               ...
               }
----------------------------------------------------------------------------------------------

As you can see, 'buffer' is no longer modified between the two calls, so they
will both return the same value. The code sequence is only entered if the first
stat() call returns 0, so the second call will also return 0, and thus the
'continue' statement will never be executed.

Klaus
Lucian Muresan - Feb. 13, 2012, 11:27 a.m.
On 13.02.2012 12:00, Klaus Schmidinger wrote:
> On 13.02.2012 11:44, Lucian Muresan wrote:
>> On 13.02.2012 10:43, Klaus Schmidinger wrote:
>>>> On 01/27/2012 01:04 PM, Oliver Endriss wrote:
>>>>> On Thursday 26 January 2012 11:07:18 Klaus Schmidinger wrote:
>>>>>> On 25.01.2012 14:11, Oliver Endriss wrote:
>>>>>>> On Wednesday 25 January 2012 10:29:16 Klaus Schmidinger wrote:
>>>>>>>> On 17.01.2012 14:26, sundararaj reel wrote:
>>
>> [...]
>>
>>>> Well, with this patch, symbolic links are not displayed at all on my
>>>> VDR machine, whereas with sundararaj reel's fix they are displayed
>>>> correctly.
>>>
>>> So what you're saying it that this...
>>>
>>> ----------------------------------------------------------------------------------------------
>>>
>>>
>>> --- a/recording.c
>>> +++ b/recording.c
>>> @@ -1120,9 +1120,13 @@ void cRecordings::ScanVideoDir(const char
>>> *DirName, bool Foreground, int LinkLev
>>> continue;
>>> }
>>> Link = 1;
>>> +#if 0
>>> + // do not resolve the symbolic links in paths to real path
>>> + // thereby keeping all the recordings under one directory
>>> buffer = ReadLink(buffer);
>>> if (!*buffer)
>>> continue;
>>> +#endif
>>> if (stat(buffer, &st) != 0)
>>> continue;
>>> }
>>> ----------------------------------------------------------------------------------------------
>>>
>>>
>>>
>>> ...works, while this...
>>>
>>> ----------------------------------------------------------------------------------------------
>>>
>>>
>>> --- recording.c 2012/01/25 09:32:39 2.45
>>> +++ recording.c 2012/01/26 10:02:29
>>> @@ -1120,11 +1120,6 @@
>>> continue;
>>> }
>>> Link = 1;
>>> - buffer = ReadLink(buffer);
>>> - if (!*buffer)
>>> - continue;
>>> - if (stat(buffer, &st) != 0)
>>> - continue;
>>> }
>>> if (S_ISDIR(st.st_mode)) {
>>> if (endswith(buffer, deleted ? DELEXT : RECEXT)) {
>>> ----------------------------------------------------------------------------------------------
>>>
>>>
>>>
>>> ...doesn't?
>>>
>>> I find that hard to believe, because the only difference here is that
>>> the second version removes the stat() call, which is superfluous if
>>> 'buffer' is no longer modified.
>>
>> Haven't really looked at the code, until now, and I also do not
>> exactly know what the call to stat does and also didn't try to
>> understand the whole picture now, but your patch does not simply
>> remove just the call to stat, but also a *continue* statement from the
>> *while* loop, this could have strong
>> implications, so just please consider analyzing the issue with respect
>> to that.
>
> The original code was
>
> ----------------------------------------------------------------------------------------------
>
> if (stat(buffer, &st) == 0) {
> int Link = 0;
> if (S_ISLNK(st.st_mode)) {
> if (LinkLevel > MAX_LINK_LEVEL) {
> isyslog("max link level exceeded - not scanning %s", *buffer);
> continue;
> }
> Link = 1;
> buffer = ReadLink(buffer);
> if (!*buffer)
> continue;
> if (stat(buffer, &st) != 0)
> continue;
> }
> ...
> }
> ----------------------------------------------------------------------------------------------
>
>
> After Sundararaj's patch it looked like this (just leaving out the lines
> that his '#if 0' disabled):
>
> ----------------------------------------------------------------------------------------------
>
> if (stat(buffer, &st) == 0) {
> int Link = 0;
> if (S_ISLNK(st.st_mode)) {
> if (LinkLevel > MAX_LINK_LEVEL) {
> isyslog("max link level exceeded - not scanning %s", *buffer);
> continue;
> }
> if (stat(buffer, &st) != 0)
> continue;
> }
> ...
> }
> ----------------------------------------------------------------------------------------------
>
>
> Reducing this to the stat() calls results in
>
> ----------------------------------------------------------------------------------------------
>
> if (stat(buffer, &st) == 0) {
> ...
> if (stat(buffer, &st) != 0)
> continue;
> }
> ...
> }
> ----------------------------------------------------------------------------------------------
>
>
> As you can see, 'buffer' is no longer modified between the two calls, so
> they
> will both return the same value. The code sequence is only entered if
> the first
> stat() call returns 0, so the second call will also return 0, and thus the
> 'continue' statement will never be executed.

Now I looked a bit closer, and noticed that the "first" call to stat is 
actually not to *stat*, but to *lstat* in vanilla vdr-1.7.23, so if you 
remove the call to the second one, could it be that you're not really 
cutting redundancy, maybe it actually makes a difference?

Lucian
Klaus Schmidinger - Feb. 13, 2012, 11:54 a.m.
On 13.02.2012 12:27, Lucian Muresan wrote:
> On 13.02.2012 12:00, Klaus Schmidinger wrote:
>> On 13.02.2012 11:44, Lucian Muresan wrote:
>>> On 13.02.2012 10:43, Klaus Schmidinger wrote:
>>>>> On 01/27/2012 01:04 PM, Oliver Endriss wrote:
>>>>>> On Thursday 26 January 2012 11:07:18 Klaus Schmidinger wrote:
>>>>>>> On 25.01.2012 14:11, Oliver Endriss wrote:
>>>>>>>> On Wednesday 25 January 2012 10:29:16 Klaus Schmidinger wrote:
>>>>>>>>> On 17.01.2012 14:26, sundararaj reel wrote:
>>>
>>> [...]
>>>
>>>>> Well, with this patch, symbolic links are not displayed at all on my
>>>>> VDR machine, whereas with sundararaj reel's fix they are displayed
>>>>> correctly.
>>>>
>>>> So what you're saying it that this...
>>>>
>>>> ----------------------------------------------------------------------------------------------
>>>>
>>>>
>>>> --- a/recording.c
>>>> +++ b/recording.c
>>>> @@ -1120,9 +1120,13 @@ void cRecordings::ScanVideoDir(const char
>>>> *DirName, bool Foreground, int LinkLev
>>>> continue;
>>>> }
>>>> Link = 1;
>>>> +#if 0
>>>> + // do not resolve the symbolic links in paths to real path
>>>> + // thereby keeping all the recordings under one directory
>>>> buffer = ReadLink(buffer);
>>>> if (!*buffer)
>>>> continue;
>>>> +#endif
>>>> if (stat(buffer, &st) != 0)
>>>> continue;
>>>> }
>>>> ----------------------------------------------------------------------------------------------
>>>>
>>>>
>>>>
>>>> ...works, while this...
>>>>
>>>> ----------------------------------------------------------------------------------------------
>>>>
>>>>
>>>> --- recording.c 2012/01/25 09:32:39 2.45
>>>> +++ recording.c 2012/01/26 10:02:29
>>>> @@ -1120,11 +1120,6 @@
>>>> continue;
>>>> }
>>>> Link = 1;
>>>> - buffer = ReadLink(buffer);
>>>> - if (!*buffer)
>>>> - continue;
>>>> - if (stat(buffer, &st) != 0)
>>>> - continue;
>>>> }
>>>> if (S_ISDIR(st.st_mode)) {
>>>> if (endswith(buffer, deleted ? DELEXT : RECEXT)) {
>>>> ----------------------------------------------------------------------------------------------
>>>>
>>>>
>>>>
>>>> ...doesn't?
>>>>
>>>> I find that hard to believe, because the only difference here is that
>>>> the second version removes the stat() call, which is superfluous if
>>>> 'buffer' is no longer modified.
>>>
>>> Haven't really looked at the code, until now, and I also do not
>>> exactly know what the call to stat does and also didn't try to
>>> understand the whole picture now, but your patch does not simply
>>> remove just the call to stat, but also a *continue* statement from the
>>> *while* loop, this could have strong
>>> implications, so just please consider analyzing the issue with respect
>>> to that.
>>
>> The original code was
>>
>> ----------------------------------------------------------------------------------------------
>>
>> if (stat(buffer, &st) == 0) {
>> int Link = 0;
>> if (S_ISLNK(st.st_mode)) {
>> if (LinkLevel > MAX_LINK_LEVEL) {
>> isyslog("max link level exceeded - not scanning %s", *buffer);
>> continue;
>> }
>> Link = 1;
>> buffer = ReadLink(buffer);
>> if (!*buffer)
>> continue;
>> if (stat(buffer, &st) != 0)
>> continue;
>> }
>> ...
>> }
>> ----------------------------------------------------------------------------------------------
>>
>>
>> After Sundararaj's patch it looked like this (just leaving out the lines
>> that his '#if 0' disabled):
>>
>> ----------------------------------------------------------------------------------------------
>>
>> if (stat(buffer, &st) == 0) {
>> int Link = 0;
>> if (S_ISLNK(st.st_mode)) {
>> if (LinkLevel > MAX_LINK_LEVEL) {
>> isyslog("max link level exceeded - not scanning %s", *buffer);
>> continue;
>> }
>> if (stat(buffer, &st) != 0)
>> continue;
>> }
>> ...
>> }
>> ----------------------------------------------------------------------------------------------
>>
>>
>> Reducing this to the stat() calls results in
>>
>> ----------------------------------------------------------------------------------------------
>>
>> if (stat(buffer, &st) == 0) {
>> ...
>> if (stat(buffer, &st) != 0)
>> continue;
>> }
>> ...
>> }
>> ----------------------------------------------------------------------------------------------
>>
>>
>> As you can see, 'buffer' is no longer modified between the two calls, so
>> they
>> will both return the same value. The code sequence is only entered if
>> the first
>> stat() call returns 0, so the second call will also return 0, and thus the
>> 'continue' statement will never be executed.
>
> Now I looked a bit closer, and noticed that the "first" call to stat is actually not to *stat*, but to *lstat* in vanilla vdr-1.7.23, so if you remove the call to the second one, could it be that you're not really cutting redundancy, maybe it actually makes a difference?

Now you got me ;-)
I had switched back to 'stat()' instead of 'lstat()' and forgotten to reintroduce
the 'lstat()'. Therefore I saw two identical calls to stat().
With the first call being lstat(), the second call to stat() is of course necessary
for the following S_ISDIR() check.

Sorry for the confusion, and thanks for clarifying this.
So now this whole code sequence looks like this:

----------------------------------------------------------------------------------------------
            if (lstat(buffer, &st) == 0) {
               int Link = 0;
               if (S_ISLNK(st.st_mode)) {
                  if (LinkLevel > MAX_LINK_LEVEL) {
                     isyslog("max link level exceeded - not scanning %s", *buffer);
                     continue;
                     }
                  Link = 1;
                  if (stat(buffer, &st) != 0)
                     continue;
                  }
----------------------------------------------------------------------------------------------

Klaus
Lucian Muresan - Feb. 13, 2012, 12:52 p.m.
On 13.02.2012 12:54, Klaus Schmidinger wrote:
> On 13.02.2012 12:27, Lucian Muresan wrote:
>> On 13.02.2012 12:00, Klaus Schmidinger wrote:
[...]
>> Now I looked a bit closer, and noticed that the "first" call to stat
>> is actually not to *stat*, but to *lstat* in vanilla vdr-1.7.23, so if
>> you remove the call to the second one, could it be that you're not
>> really cutting redundancy, maybe it actually makes a difference?
>
> Now you got me ;-)
> I had switched back to 'stat()' instead of 'lstat()' and forgotten to
> reintroduce
> the 'lstat()'. Therefore I saw two identical calls to stat().
> With the first call being lstat(), the second call to stat() is of
> course necessary
> for the following S_ISDIR() check.
>
> Sorry for the confusion, and thanks for clarifying this.
> So now this whole code sequence looks like this:
>
> ----------------------------------------------------------------------------------------------
>
> if (lstat(buffer, &st) == 0) {
> int Link = 0;
> if (S_ISLNK(st.st_mode)) {
> if (LinkLevel > MAX_LINK_LEVEL) {
> isyslog("max link level exceeded - not scanning %s", *buffer);
> continue;
> }
> Link = 1;
> if (stat(buffer, &st) != 0)
> continue;
> }

No problem at all, I'm glad it's now sorted out for the next developer 
version.

Lucian

Patch

--- a/recording.c
+++ b/recording.c
@@ -1120,9 +1120,13 @@  void cRecordings::ScanVideoDir(const char *DirName, bool Foreground, int LinkLev
                      continue;
                      }
                   Link = 1;
+#if 0
+                 // do not resolve the symbolic links in paths to real path
+                 // thereby keeping all the recordings under one directory
                   buffer = ReadLink(buffer);
                   if (!*buffer)
                      continue;
+#endif
                   if (stat(buffer, &st) != 0)
                      continue;
                   }
----------------------------------------------------------------------------------------------

...works, while this...

----------------------------------------------------------------------------------------------
--- recording.c 2012/01/25 09:32:39     2.45
+++ recording.c 2012/01/26 10:02:29
@@ -1120,11 +1120,6 @@ 
                      continue;
                      }
                   Link = 1;
-                 buffer = ReadLink(buffer);
-                 if (!*buffer)
-                    continue;
-                 if (stat(buffer, &st) != 0)
-                    continue;
                   }
                if (S_ISDIR(st.st_mode)) {
                   if (endswith(buffer, deleted ? DELEXT : RECEXT)) {

Privacy Policy