[Olsr-users] problem when disconnection two network cables

[Henning Rogge]

Hi,

the analysis of the bug sounds reasonable. Please post the patch here
and/or the olsr-dev list! :)

Henning

On Fri, Jun 12, 2015 at 4:03 PM, Ferry Huberts <(spam-protected)> wrote:
> You can post the patch here, or on the dev list
>
> Henning, WDYT?
>
> On 12/06/15 15:55, Koen Swinters wrote:
>>
>> Hi all,
>>
>> We are using olsrd-0.6.8 in our system.
>>
>> I have a problem with the following setup:
>>
>> Node 8100 ------------- node 3
>>
>>       |             X                 |
>>
>> Node 8120 ------------- node 425
>>
>> Each node has a wired connection to the 3 other nodes (each node has 3
>> Ethernet interfaces). Olsrd is running on each node.
>>
>> When I disconnect the cable between node 3 and node 8120 and after this
>> the cable between node 8120 and node 425 => node 425 cannot reach node
>> 8120 anymore (there is still a path via node 3 and node 8100) => the
>> routing table on node 425 is not updated.
>>
>> Remark: when there is no physical Ethernet connection our systems
>> disables the eth interface (ethx)
>>
>> I`ve investigated this problem and found the following problem in olsrd:
>>
>> ·disconnect the cable between node 3 and node 8120 => Link down between
>> 3 and 8120 => disable interface
>>
>> oLine 371 in interfaces.c: "olsr_remove_interface(struct olsr_if* iface)"
>>
>> ·Line 379 in "interfaces.c":
>> "olsr_delete_link_entry_by_ip(&ifp->ip_addr)"
>>
>> §Line 411 in "link_set.c":  "olsr_delete_link_entry(link);"
>>
>> ·Line 375 in
>> "olsr_delete_neighbor_table(&link->neighbor->neighbor_main_addr)";
>>
>> ·This function will delete the neighbor from the neighbor table
>>
>> ·(Line 177 in neighbor_table.c)
>>
>> ·This function sets "changes_neighborhood = true;" (Line 219 in
>> neighbor_table.c)
>>
>> oIn "olsr_scheduler(void)" (line 486 in scheduler.c)
>>
>> ·function "olsr_process_changes()" (line 180 in olsr.c)
>>
>> ·Because "changes_neighborhood == TRUE and olsr_cnf->lq_level > 1"
>> "olsr_calculate_lq_mpr() "is called
>>
>> ·In "olsr_calculate_lq_mpr()" (line 51 in lq_mpr.c) a new mpr is
>> calculated (8100 will now be mpr for 8120) and the function
>> "signal_link_changes(true);" is called => link_changes  == true
>>
>> ·In the next step  ANSN  is increased
>>
>> oAt the tc_interval  a tc message is generated with this ANSN in => the
>> neighbors see a change in the ANSN number(when the packet are received)
>> an will update their tables
>>
>> ·Link Down between node 8120 and 425
>>
>> oLine 371 in interfaces.c: "olsr_remove_interface(struct olsr_if* iface)"
>>
>> ·Line 379 in "interfaces.c":
>> "olsr_delete_link_entry_by_ip(&ifp->ip_addr)"
>>
>> §Line 411 in "link_set.c":  "olsr_delete_link_entry(link);"
>>
>> ·Line 375 in
>> "olsr_delete_neighbor_table(&link->neighbor->neighbor_main_addr)";
>>
>> ·This function will delete the neighbor from the neighbor table
>>
>> ·(Line 177 in neighbor_table.c)
>>
>> ·This function sets "changes_neighborhood = true;" (Line 219 in
>> neighbor_table.c)
>>
>> oIn "olsr_scheduler(void)" (line 486 in scheduler.c)
>>
>> ·function "olsr_process_changes()" (line 180 in olsr.c)
>>
>> ·Because "changes_neighborhood == TRUE and olsr_cnf->lq_level > 1"
>> "olsr_calculate_lq_mpr() "is called
>>
>> ·In "olsr_calculate_lq_mpr()" (line 51 in lq_mpr.c) no new mpr is
>> calculated (8100 will stay mpr for 8120) and the function
>> "signal_link_changes(true);" is not called => link_changes  == false
>>
>> ·In the next step  ANSN  is not increased
>>
>> oAt the tc_interval  a tc message is generated with old ANSN in => the
>> neighbors see no change in the ANSN number(when the packet are received)
>> an will no update their tables
>>
>> Solution:
>>
>> ·When a neighbor is deleted from the neighbor table (line 177 in
>> neighbor_table: olsr_delete_neighbor_table(const union olsr_ip_addr
>> *neighbor_addr)) also call the function signal_link_changes(true);
>> because there is a change in the neighbor table.
>>
>> I’ve tested this and it is working correct (routing tables are updated
>> as expected).
>>
>> Where can I post this fix, or is there a better solution?
>>
>> Regards,
>>
>> koen
>>
>>
>>
>
> --
> Ferry Huberts