lib: fix swapped values, bad setsockopt, and intermittent test failure

[choppsv1]

We had reversed the setting of max read and write message count values for mgmtd connections (both server and client). This effectively allowed clients to send more than they received and the server to process more than it sent (i.e., backwards of what we probably want to keep a loaded system working properly)

Also for the mgmtd loopback connection (i.e., sending CLI based YANG config or reciving oper-state) we were actually setting the recv and send kernel buf (setsockopt() which is given in bytes) to these number of message values. In this case the kernel was ignoring us and using a default value instead which was actually 128K. :)

Finally, fix an intermittent failure in mgmt_notify test that we (again) hit during CI testing of these changes.

[greptile-apps]

Greptile Summary

This PR fixes two distinct bugs in the mgmt message connection layer (lib/mgmt_msg.c):

1. Swapped buffer limits — mgmt_msg_init was assigning max_write_buf to ms->max_read_buf and vice versa, meaning every caller had its read/write message-processing limits inverted. This is now correctly aligned.

2. Wrong setsockopt argument type — msg_client_connect_short_circuit was passing mstate.max_write_buf / mstate.max_read_buf (which count messages, not bytes) to setsockopt_so_sendbuf / setsockopt_so_recvbuf, which expect a byte value. The kernel would have clamped these tiny values to its minimum. The fix replaces them with the dedicated MSG_CONN_SEND_BUF_SIZE / MSG_CONN_RECV_BUF_SIZE constants (now 128 KB each, up from 64 KB), matching what the regular client path (mgmt_msg_connect) and the server acceptance path (msg_conn_accept_init) already used.

Both fixes are clearly correct and the changes are minimal and surgical. No new logic is introduced.

Confidence Score: 5/5

• This PR is safe to merge — all three changes are unambiguous bug fixes with no side-effects beyond correcting the broken behaviour.
• The diff is small, isolated to one file, and each change directly corrects a clearly documented bug. The swapped field assignment is trivially verified against the parameter names. The setsockopt fix brings the short-circuit path in line with the already-correct regular client and server-accept paths. The buffer-size increase (64 KB → 128 KB) is conservative and the previous values were effectively meaningless to the kernel anyway.
• No files require special attention.

Important Files Changed

Filename	Overview
lib/mgmt_msg.c	Three targeted bug fixes: (1) corrects swapped max_read_buf/max_write_buf assignments in mgmt_msg_init, (2) doubles kernel socket buffer constants from 64KB to 128KB, and (3) fixes the short-circuit connection setsockopt calls to use byte-size constants instead of message-count values.

Sequence Diagram

sequenceDiagram
    participant C as msg_client
    participant SC as short_circuit path
    participant S as msg_server

    Note over C,S: Short-circuit connection setup (fixed)
    C->>SC: msg_client_connect_short_circuit()
    SC->>SC: socketpair(AF_UNIX, SOCK_STREAM, 0, sockets)
    SC->>SC: set_nonblocking(sockets[0])
    SC->>SC: setsockopt_so_sendbuf(sockets[0], MSG_CONN_SEND_BUF_SIZE ✅)
    Note right of SC: Previously: mstate.max_write_buf (msg count, not bytes ❌)
    SC->>SC: setsockopt_so_recvbuf(sockets[0], MSG_CONN_RECV_BUF_SIZE ✅)
    Note right of SC: Previously: mstate.max_read_buf (msg count, not bytes ❌)
    SC->>S: server->create(sockets[1], &su) → msg_conn_accept_init()
    S->>S: setsockopt_so_sendbuf(fd, MSG_CONN_SEND_BUF_SIZE) [already correct]
    S->>S: setsockopt_so_recvbuf(fd, MSG_CONN_RECV_BUF_SIZE) [already correct]

    Note over C,S: mgmt_msg_init assignment (fixed)
    C->>C: mgmt_msg_init(ms, max_read_buf, max_write_buf, ...)
    C->>C: ms->max_read_buf = max_read_buf ✅
    Note right of C: Previously assigned max_write_buf here ❌
    C->>C: ms->max_write_buf = max_write_buf ✅
    Note right of C: Previously assigned max_read_buf here ❌

Loading

_{Last reviewed commit: "lib: fix swapped val..."}