Skip to contentSkip to conversation

Failure to enqueue packets for cryptodev-scheduler_multicore can lead to program crashes

DPDK/cryptodev Bug 1537 [email protected] 1 participant 1 message Sep 9, 2024

lore
Share & tools
lore
#0 lore raw

https://bugs.dpdk.org/show_bug.cgi?id=1537

            Bug ID: 1537
           Summary: Failure to enqueue packets for
                    cryptodev-scheduler_multicore can lead to program
                    crashes
           Product: DPDK
           Version: 22.11
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: critical
          Priority: Normal
         Component: cryptodev
          Assignee: [email protected]
          Reporter: [email protected]

Target Milestone: ---

Created attachment 288
--> https://bugs.dpdk.org/attachment.cgi?id=288&action=edit
Incorrect Variable Usage

Failure Scenario:
The cryptodev scheduler is running in Multi-core mode, with OpenSSL PMD
attached as workers.
When a batch of packets to be enqueued contains invalid encrypted ESP
(Encapsulating Security Payload) packets, the program crashes.

---------------------------------
Specific Issues:
1. Incorrect Variable Usage:
- The variable pending_deq_ops was incorrectly used instead of
pending_enq_ops. This leads to failed condition checks and prevents the scheduler_retrieve_session function from executing. Consequently, the content of the op->sym->session variable becomes abnormal, causing the program to crash when the OpenSSL PMD accesses this variable.
2. Lack of Exception Handling:
- There is no handling for exceptional cases when enqueuing packets. When
invalid packets are enqueued, the rte_cryptodev_enqueue_burst function always returns a value of processed_ops that is less than pending_enq_ops, leading the program into an infinite retry loop.

--------------------------------
Resolution Suggestions:
1. Correct Variable Usage:
- In the faulty code, replace pending_deq_ops with pending_enq_ops.

2. Documentation Update:
- Add documentation stating that packets should always be added to the
processed operations queue in the worker PMD.

- Alternatively, implement a maximum retry count check to prevent the
program from entering an infinite retry loop.

Signature · 2 lines
You are receiving this mail because:
You are the assignee for the bug.