client: add ExecQueryRelayLocalInfile for proxy-friendly LOAD DATA LOCAL INFILE

[nzin-alayacare]

Summary

Add (*Conn).ExecQueryRelayLocalInfile and the companion helper cleanupLocalInfileAfterRelayError to client/conn.go.

Closes #1161

Motivation

The existing Execute / exec path (and the LOCAL INFILE handling in client/auth.go) reads the file from the local filesystem. In a proxy scenario the file lives on the end-user's machine, not on the proxy. The proxy must relay the 0xfb request and the subsequent file-data packets between the upstream MySQL server and the downstream client.

There is no API today to intercept the 0xfb LOCAL INFILE response and hand off the transfer to a callback. This PR fills that gap.

New API

// ExecQueryRelayLocalInfile sends COM_QUERY and handles the LOCAL INFILE protocol
// when the server responds with a 0xfb packet. relayFile receives the full 0xfb
// request payload; the implementation must forward the request to the application
// client, read file-data packets from that client, and write each to this
// connection via WritePacket (4-byte header + payload) until an empty packet
// signals EOF. The final OK or ERR from the upstream server is read and returned.
//
// If relayFile returns an error the connection is resynchronized before the error
// is returned so the connection can be reused.
func (c *Conn) ExecQueryRelayLocalInfile(
    query string,
    relayFile func(localInfileRequestPayload []byte) error,
) (*mysql.Result, error)

cleanupLocalInfileAfterRelayError is unexported; it sends the empty EOF packet and reads the server's OK/ERR to bring the connection back to a usable state when relayFile fails mid-transfer.

Notes

• Pure additive change — no existing API is modified.
• Requires CLIENT_LOCAL_FILES to be set on both the server side (see server: add CLIENT_LOCAL_FILES to default server capabilities #1163) and the client side (via SetCapability).
• Tested in production at alayacare/castledm.

Made with Cursor

[gemini-code-assist]

Code Review

This pull request introduces cleanupLocalInfileAfterRelayError and ExecQueryRelayLocalInfile to client/conn.go to support relaying LOCAL INFILE requests and handling connection resynchronization on errors. The review feedback correctly identifies several potential runtime panics where the code accesses the first element of a packet slice (resp, data, and data2) without first verifying that the slice is not empty. It is recommended to add length checks before accessing these slices and to consistently trace errors using errors.Trace.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[lance6716]

Oh for previous PR codex found some problems:

[P1] server/resp.go (line 31): when CLIENT_SESSION_TRACK is negotiated but SERVER_SESSION_STATE_CHANGED is not set, writeOK still appends a zero-length session-state block via AppendOKSessionTrackSuffix. In the MySQL OK-packet format, that field is only present when the status bit is set; otherwise the packet should stop after the length-encoded info string. This makes auth OK / normal OK packets malformed for strict clients.

[P2] server/server_conf.go (line 137): the new SetCapability / UnsetCapability API lets callers flip any server capability bit, but some of those bits are not independently configurable. A concrete example is CLIENT_SSL: enabling it on a server without tlsConfig will advertise TLS in the greeting and then fail the SSLRequest path during handshake. The setter should either validate a safe whitelist or tie each flag to the underlying server config.

Can you create a new PR to fix it?

[nzin-alayacare]

[P1] server/resp.go (line 31): when CLIENT_SESSION_TRACK is negotiated but SERVER_SESSION_STATE_CHANGED is not set, writeOK still appends a zero-length session-state block via AppendOKSessionTrackSuffix. In the MySQL OK-packet format, that field is only present when the status bit is set; otherwise the packet should stop after the length-encoded info string. This makes auth OK / normal OK packets malformed for strict clients.

[P2] server/server_conf.go (line 137): the new SetCapability / UnsetCapability API lets callers flip any server capability bit, but some of those bits are not independently configurable. A concrete example is CLIENT_SSL: enabling it on a server without tlsConfig will advertise TLS in the greeting and then fail the SSLRequest path during handshake. The setter should either validate a safe whitelist or tie each flag to the underlying server config.

sure -> #1167

[lance6716]

rest lgtm