Re: [RFC,net-next, 2/5] vrf: track associations between VRF devices and tables

On Fri, 12 Jun 2020 18:49:34 +0200 Andrea Mayer andrea.mayer@uniroma2.it wrote:

• /* shared_tables:

• * count how many distinct tables does not comply with the
  

• * strict mode requirement.
  

• * shared_table value must be 0 in order to switch to strict mode.
  

• *
  

• * example of evolution of shared_table:
  

• *                                                        | time
  

• * add  vrf0 --> table 100        shared_tables = 0       | t0
  

• * add  vrf1 --> table 101        shared_tables = 0       | t1
  

• * add  vrf2 --> table 100        shared_tables = 1       | t2
  

• * add  vrf3 --> table 100        shared_tables = 1       | t3
  

• * add  vrf4 --> table 101        shared_tables = 2       v t4
  

• *
  

• * shared_tables is a "step function" (or "staircase function")
  

• * and is increased by one when the second vrf is associated to a table
  

• *
  

• * at t2, vrf0 and vrf2 are bound to table 100: shared_table = 1.
  

• *
  

• * at t3, another dev (vrf3) is bound to the same table 100 but the
  

• * shared_table counters is still 1.
  

• * This means that no matter how many new vrfs will register on the
  

• * table 100, the shared_table will not increase (considering only
  

• * table 100).
  

• *
  

• * at t4, vrf4 is bound to table 101, and shared_table = 2.
  

• *
  

• * Looking at the value of shared_tables we can immediately know if
  

• * the strict_mode can or cannot be enforced. Indeed, strict_mode
  

• * can be enforced iff shared_table = 0.
  

• *
  

• * Conversely, shared_table is decreased when a vrf is de-associated
  

• * from a table with exactly two associated vrfs.
  

• */
  

• int shared_tables;

Should this be unsigned? Should it be a fixed size?