[DPDK/other Bug 1883] rte_hash with big cap and key size of 64bytes and more can lead to key store corruption

[DPDK/other Bug 1883] rte_hash with big cap and key size of 64bytes and more can lead to key store corruption

[bugzilla]

http://bugs.dpdk.org/show_bug.cgi?id=1883

            Bug ID: 1883
           Summary: rte_hash with big cap and key size of 64bytes and more
                    can lead to key store corruption
           Product: DPDK
           Version: 26.03
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: normal
          Priority: Normal
         Component: other
          Assignee: dev@dpdk.org
          Reporter: oleksandrn@interfacemasters.com
  Target Milestone: ---

Created attachment 340
  --> http://bugs.dpdk.org/attachment.cgi?id=340&action=edit
simple repro

I was experimenting with table size of 2^26 and key size of 64 bytes, and
noticed, that deletion was failing for some earlier added entries, after some
digging it seems like the main problem is overflow in key_store offset
calculation, seems like there are multiple places, but as example

new_k = RTE_PTR_ADD(keys, slot_id * h->key_entry_size);

slot_id is u32, and h->key_entry_size is u32 as well, in my case at high
occupancy slot_id  can be very high(up to 67mil) which multiplied by 80 will
overflow u32 and will overwrite earlier entries.

lazy fix seem to be making h->key_entry_size u64

Attached simple repro.

-- 
You are receiving this mail because:
You are the assignee for the bug.