fix(websocket): Fix the bug related to multi-threaded access to trans… (IDFGH-17155)

[udoudou]

The close operation on the client->transport resource is performed under the protection of client->lock. esp_websocket_client_send_with_exact_opcode checks the state without holding the lock when determining whether the transport can be accessed. This can lead to some unexpected behaviors:

1. If tx_lock is not enabled, calling esp_transport_ws_send_raw will be abnormally blocked if the transport is already closed.

2. If tx_lock is enabled, esp_transport_close and esp_transport_ws_send_raw may be called concurrently. Since close involves releasing some resources, this may lead to unpredictable behavior.

Description

To fix this issue, I adjusted the implementation of resource mutual exclusion protection.
1、tx_lock has been removed
The original purpose of introducing tx_lock was to allow for parallel processing of receiving and sending, improving transmission efficiency. However, this doesn't necessarily require introducing a new tx_lock. This is because the receiving direction is accessed only by the esp_websocket_client_task. The sending direction may be accessed concurrently by several user tasks and the esp_websocket_client_task. Therefore, the esp_websocket_client_task does not need lock protection when receiving. Lock protection is only applied when the esp_websocket_client_task needs to send or when a user task sends, to check if the transport is available.

2、Change esp_websocket_client_abort_connection to asynchronous close transport.
Modify esp_websocket_client_abort_connection to only modify the state, preventing user tasks from directly executing close transport and conflicting with esp_websocket_client_task's receive transport.

The APIs involved in resource access contention mainly include esp_websocket_client_send_bin, esp_websocket_client_send_bin_partial, esp_websocket_client_send_text, esp_websocket_client_send_text_partial, esp_websocket_client_send_cont_msg, esp_websocket_client_send_fin, esp_websocket_client_send_with_opcode, esp_websocket_client_close, and esp_websocket_client_close_with_code. Ultimately, all of these are involved in esp_websocket_client_send_with_exact_opcode. The main issue is contention involving transport sending, errormsg_buffer, and state when the state is WEBSOCKET_STATE_CONNECTED. After the modification, esp_websocket_client_send_with_exact_opcode accesses and modifies these resources under lock protection. In esp_websocket_client_task, when the state is WEBSOCKET_STATE_CONNECTED, any transport sending, changes to a connectionless state, or access to errormsg_buffer are all performed under lock protection. The close of the transport is uniformly performed by esp_websocket_client_task in a non-WEBSOCKET_STATE_CONNECTED state, ensuring that user task sending will not occur concurrently with the close of the transport, and esp_websocket_client_task does not require lock protection when receiving.

Related

Testing

---

Checklist

Before submitting a Pull Request, please ensure the following:

• 🚨 This PR does not introduce breaking changes.
• All CI checks (GH Actions) pass.
• Documentation is updated as needed.
• Tests are updated or added as necessary.
• Code is well-commented, especially in complex areas.
• Git history is clean — commits are squashed to the minimum necessary.

---

Note

Refactors websocket client concurrency and connection lifecycle handling.

• Removes tx_lock Kconfig/options and all associated separate TX locking; uses a single client->lock to guard all transport sends/state/error buffer access
• Introduces WEBSOCKET_STATE_WAIT_ABORT_CONNECT and changes esp_websocket_client_abort_connection() to be asynchronous (sets state, dispatches disconnect); actual esp_transport_close() now performed by the client task
• Ensures PING/PONG/CLOSE and send paths acquire client->lock and re-check state before accessing transport; adds locked aborts on read/write failures
• Adjusts main task state machine: splits abort handling, reconnect/wait logic, and server-initiated close flow; removes previous WAIT_TIMEOUT handling tied to immediate close
• Cleans up Kconfig (drops tx lock options) and updates SPDX year

^{Written by Cursor Bugbot for commit 3bbe73a. This will update automatically on new commits. Configure here.}

[CLAassistant]

All committers have signed the CLA.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}