Line data Source code
1 : //! Postgres protocol messages serialization-deserialization. See
2 : //! <https://www.postgresql.org/docs/devel/protocol-message-formats.html>
3 : //! on message formats.
4 : #![deny(clippy::undocumented_unsafe_blocks)]
5 :
6 : pub mod framed;
7 :
8 : use byteorder::{BigEndian, ReadBytesExt};
9 : use bytes::{Buf, BufMut, Bytes, BytesMut};
10 : use itertools::Itertools;
11 : use serde::{Deserialize, Serialize};
12 : use std::{borrow::Cow, fmt, io, str};
13 :
14 : // re-export for use in utils pageserver_feedback.rs
15 : pub use postgres_protocol::PG_EPOCH;
16 :
17 : pub type Oid = u32;
18 : pub type SystemId = u64;
19 :
20 : pub const INT8_OID: Oid = 20;
21 : pub const INT4_OID: Oid = 23;
22 : pub const TEXT_OID: Oid = 25;
23 :
24 : #[derive(Debug)]
25 : pub enum FeMessage {
26 : // Simple query.
27 : Query(Bytes),
28 : // Extended query protocol.
29 : Parse(FeParseMessage),
30 : Describe(FeDescribeMessage),
31 : Bind(FeBindMessage),
32 : Execute(FeExecuteMessage),
33 : Close(FeCloseMessage),
34 : Sync,
35 : Terminate,
36 : CopyData(Bytes),
37 : CopyDone,
38 : CopyFail,
39 : PasswordMessage(Bytes),
40 : }
41 :
42 : #[derive(Clone, Copy, PartialEq, PartialOrd)]
43 : pub struct ProtocolVersion(u32);
44 :
45 : impl ProtocolVersion {
46 0 : pub const fn new(major: u16, minor: u16) -> Self {
47 0 : Self(((major as u32) << 16) | minor as u32)
48 0 : }
49 0 : pub const fn minor(self) -> u16 {
50 0 : self.0 as u16
51 0 : }
52 24 : pub const fn major(self) -> u16 {
53 24 : (self.0 >> 16) as u16
54 24 : }
55 : }
56 :
57 : impl fmt::Debug for ProtocolVersion {
58 0 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
59 0 : f.debug_list()
60 0 : .entry(&self.major())
61 0 : .entry(&self.minor())
62 0 : .finish()
63 0 : }
64 : }
65 :
66 : #[derive(Debug)]
67 : pub enum FeStartupPacket {
68 : CancelRequest(CancelKeyData),
69 : SslRequest {
70 : direct: bool,
71 : },
72 : GssEncRequest,
73 : StartupMessage {
74 : version: ProtocolVersion,
75 : params: StartupMessageParams,
76 : },
77 : }
78 :
79 : #[derive(Debug, Clone, Default)]
80 : pub struct StartupMessageParamsBuilder {
81 : params: BytesMut,
82 : }
83 :
84 : impl StartupMessageParamsBuilder {
85 : /// Set parameter's value by its name.
86 : /// name and value must not contain a \0 byte
87 25 : pub fn insert(&mut self, name: &str, value: &str) {
88 25 : self.params.put(name.as_bytes());
89 25 : self.params.put(&b"\0"[..]);
90 25 : self.params.put(value.as_bytes());
91 25 : self.params.put(&b"\0"[..]);
92 25 : }
93 :
94 17 : pub fn freeze(self) -> StartupMessageParams {
95 17 : StartupMessageParams {
96 17 : params: self.params.freeze(),
97 17 : }
98 17 : }
99 : }
100 :
101 : #[derive(Debug, Clone, Default)]
102 : pub struct StartupMessageParams {
103 : pub params: Bytes,
104 : }
105 :
106 : impl StartupMessageParams {
107 : /// Get parameter's value by its name.
108 42 : pub fn get(&self, name: &str) -> Option<&str> {
109 58 : self.iter().find_map(|(k, v)| (k == name).then_some(v))
110 42 : }
111 :
112 : /// Split command-line options according to PostgreSQL's logic,
113 : /// taking into account all escape sequences but leaving them as-is.
114 : /// [`None`] means that there's no `options` in [`Self`].
115 24 : pub fn options_raw(&self) -> Option<impl Iterator<Item = &str>> {
116 24 : self.get("options").map(Self::parse_options_raw)
117 24 : }
118 :
119 : /// Split command-line options according to PostgreSQL's logic,
120 : /// applying all escape sequences (using owned strings as needed).
121 : /// [`None`] means that there's no `options` in [`Self`].
122 5 : pub fn options_escaped(&self) -> Option<impl Iterator<Item = Cow<'_, str>>> {
123 5 : self.get("options").map(Self::parse_options_escaped)
124 5 : }
125 :
126 : /// Split command-line options according to PostgreSQL's logic,
127 : /// taking into account all escape sequences but leaving them as-is.
128 30 : pub fn parse_options_raw(input: &str) -> impl Iterator<Item = &str> {
129 30 : // See `postgres: pg_split_opts`.
130 30 : let mut last_was_escape = false;
131 30 : input
132 594 : .split(move |c: char| {
133 : // We split by non-escaped whitespace symbols.
134 594 : let should_split = c.is_ascii_whitespace() && !last_was_escape;
135 594 : last_was_escape = c == '\\' && !last_was_escape;
136 594 : should_split
137 594 : })
138 77 : .filter(|s| !s.is_empty())
139 30 : }
140 :
141 : /// Split command-line options according to PostgreSQL's logic,
142 : /// applying all escape sequences (using owned strings as needed).
143 4 : pub fn parse_options_escaped(input: &str) -> impl Iterator<Item = Cow<'_, str>> {
144 4 : // See `postgres: pg_split_opts`.
145 7 : Self::parse_options_raw(input).map(|s| {
146 7 : let mut preserve_next_escape = false;
147 17 : let escape = |c| {
148 : // We should remove '\\' unless it's preceded by '\\'.
149 17 : let should_remove = c == '\\' && !preserve_next_escape;
150 17 : preserve_next_escape = should_remove;
151 17 : should_remove
152 17 : };
153 :
154 7 : match s.contains('\\') {
155 3 : true => Cow::Owned(s.replace(escape, "")),
156 4 : false => Cow::Borrowed(s),
157 : }
158 7 : })
159 4 : }
160 :
161 : /// Iterate through key-value pairs in an arbitrary order.
162 42 : pub fn iter(&self) -> impl Iterator<Item = (&str, &str)> {
163 42 : let params =
164 42 : std::str::from_utf8(&self.params).expect("should be validated as utf8 already");
165 42 : params.split_terminator('\0').tuples()
166 42 : }
167 :
168 : // This function is mostly useful in tests.
169 : #[doc(hidden)]
170 17 : pub fn new<'a, const N: usize>(pairs: [(&'a str, &'a str); N]) -> Self {
171 17 : let mut b = StartupMessageParamsBuilder::default();
172 42 : for (k, v) in pairs {
173 25 : b.insert(k, v)
174 : }
175 17 : b.freeze()
176 17 : }
177 : }
178 :
179 0 : #[derive(Debug, Hash, PartialEq, Eq, Clone, Copy, Serialize, Deserialize)]
180 : pub struct CancelKeyData {
181 : pub backend_pid: i32,
182 : pub cancel_key: i32,
183 : }
184 :
185 1 : pub fn id_to_cancel_key(id: u64) -> CancelKeyData {
186 1 : CancelKeyData {
187 1 : backend_pid: (id >> 32) as i32,
188 1 : cancel_key: (id & 0xffffffff) as i32,
189 1 : }
190 1 : }
191 :
192 : impl fmt::Display for CancelKeyData {
193 1 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
194 1 : let hi = (self.backend_pid as u64) << 32;
195 1 : let lo = (self.cancel_key as u64) & 0xffffffff;
196 1 : let id = hi | lo;
197 1 :
198 1 : // This format is more compact and might work better for logs.
199 1 : f.debug_tuple("CancelKeyData")
200 1 : .field(&format_args!("{:x}", id))
201 1 : .finish()
202 1 : }
203 : }
204 :
205 : use rand::distributions::{Distribution, Standard};
206 : impl Distribution<CancelKeyData> for Standard {
207 0 : fn sample<R: rand::Rng + ?Sized>(&self, rng: &mut R) -> CancelKeyData {
208 0 : CancelKeyData {
209 0 : backend_pid: rng.gen(),
210 0 : cancel_key: rng.gen(),
211 0 : }
212 0 : }
213 : }
214 :
215 : // We only support the simple case of Parse on unnamed prepared statement and
216 : // no params
217 : #[derive(Debug)]
218 : pub struct FeParseMessage {
219 : pub query_string: Bytes,
220 : }
221 :
222 : #[derive(Debug)]
223 : pub struct FeDescribeMessage {
224 : pub kind: u8, // 'S' to describe a prepared statement; or 'P' to describe a portal.
225 : // we only support unnamed prepared stmt or portal
226 : }
227 :
228 : // we only support unnamed prepared stmt and portal
229 : #[derive(Debug)]
230 : pub struct FeBindMessage;
231 :
232 : // we only support unnamed prepared stmt or portal
233 : #[derive(Debug)]
234 : pub struct FeExecuteMessage {
235 : /// max # of rows
236 : pub maxrows: i32,
237 : }
238 :
239 : // we only support unnamed prepared stmt and portal
240 : #[derive(Debug)]
241 : pub struct FeCloseMessage;
242 :
243 : /// An error occurred while parsing or serializing raw stream into Postgres
244 : /// messages.
245 : #[derive(thiserror::Error, Debug)]
246 : pub enum ProtocolError {
247 : /// Invalid packet was received from the client (e.g. unexpected message
248 : /// type or broken len).
249 : #[error("Protocol error: {0}")]
250 : Protocol(String),
251 : /// Failed to parse or, (unlikely), serialize a protocol message.
252 : #[error("Message parse error: {0}")]
253 : BadMessage(String),
254 : }
255 :
256 : impl ProtocolError {
257 : /// Proxy stream.rs uses only io::Error; provide it.
258 0 : pub fn into_io_error(self) -> io::Error {
259 0 : io::Error::new(io::ErrorKind::Other, self.to_string())
260 0 : }
261 : }
262 :
263 : impl FeMessage {
264 : /// Read and parse one message from the `buf` input buffer. If there is at
265 : /// least one valid message, returns it, advancing `buf`; redundant copies
266 : /// are avoided, as thanks to `bytes` crate ptrs in parsed message point
267 : /// directly into the `buf` (processed data is garbage collected after
268 : /// parsed message is dropped).
269 : ///
270 : /// Returns None if `buf` doesn't contain enough data for a single message.
271 : /// For efficiency, tries to reserve large enough space in `buf` for the
272 : /// next message in this case to save the repeated calls.
273 : ///
274 : /// Returns Error if message is malformed, the only possible ErrorKind is
275 : /// InvalidInput.
276 : //
277 : // Inspired by rust-postgres Message::parse.
278 57 : pub fn parse(buf: &mut BytesMut) -> Result<Option<FeMessage>, ProtocolError> {
279 57 : // Every message contains message type byte and 4 bytes len; can't do
280 57 : // much without them.
281 57 : if buf.len() < 5 {
282 30 : let to_read = 5 - buf.len();
283 30 : buf.reserve(to_read);
284 30 : return Ok(None);
285 27 : }
286 27 :
287 27 : // We shouldn't advance `buf` as probably full message is not there yet,
288 27 : // so can't directly use Bytes::get_u32 etc.
289 27 : let tag = buf[0];
290 27 : let len = (&buf[1..5]).read_u32::<BigEndian>().unwrap();
291 27 : if len < 4 {
292 0 : return Err(ProtocolError::Protocol(format!(
293 0 : "invalid message length {}",
294 0 : len
295 0 : )));
296 27 : }
297 27 :
298 27 : // length field includes itself, but not message type.
299 27 : let total_len = len as usize + 1;
300 27 : if buf.len() < total_len {
301 : // Don't have full message yet.
302 0 : let to_read = total_len - buf.len();
303 0 : buf.reserve(to_read);
304 0 : return Ok(None);
305 27 : }
306 27 :
307 27 : // got the message, advance buffer
308 27 : let mut msg = buf.split_to(total_len).freeze();
309 27 : msg.advance(5); // consume message type and len
310 27 :
311 27 : match tag {
312 2 : b'Q' => Ok(Some(FeMessage::Query(msg))),
313 0 : b'P' => Ok(Some(FeParseMessage::parse(msg)?)),
314 0 : b'D' => Ok(Some(FeDescribeMessage::parse(msg)?)),
315 0 : b'E' => Ok(Some(FeExecuteMessage::parse(msg)?)),
316 0 : b'B' => Ok(Some(FeBindMessage::parse(msg)?)),
317 0 : b'C' => Ok(Some(FeCloseMessage::parse(msg)?)),
318 0 : b'S' => Ok(Some(FeMessage::Sync)),
319 0 : b'X' => Ok(Some(FeMessage::Terminate)),
320 0 : b'd' => Ok(Some(FeMessage::CopyData(msg))),
321 0 : b'c' => Ok(Some(FeMessage::CopyDone)),
322 0 : b'f' => Ok(Some(FeMessage::CopyFail)),
323 25 : b'p' => Ok(Some(FeMessage::PasswordMessage(msg))),
324 0 : tag => Err(ProtocolError::Protocol(format!(
325 0 : "unknown message tag: {tag},'{msg:?}'"
326 0 : ))),
327 : }
328 57 : }
329 : }
330 :
331 : impl FeStartupPacket {
332 : /// Read and parse startup message from the `buf` input buffer. It is
333 : /// different from [`FeMessage::parse`] because startup messages don't have
334 : /// message type byte; otherwise, its comments apply.
335 91 : pub fn parse(buf: &mut BytesMut) -> Result<Option<FeStartupPacket>, ProtocolError> {
336 : /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L118>
337 : const MAX_STARTUP_PACKET_LENGTH: usize = 10000;
338 : const RESERVED_INVALID_MAJOR_VERSION: u16 = 1234;
339 : /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L132>
340 : const CANCEL_REQUEST_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5678);
341 : /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L166>
342 : const NEGOTIATE_SSL_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5679);
343 : /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L167>
344 : const NEGOTIATE_GSS_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5680);
345 :
346 : // <https://github.com/postgres/postgres/blob/04bcf9e19a4261fe9c7df37c777592c2e10c32a7/src/backend/tcop/backend_startup.c#L378-L382>
347 : // First byte indicates standard SSL handshake message
348 : // (It can't be a Postgres startup length because in network byte order
349 : // that would be a startup packet hundreds of megabytes long)
350 91 : if buf.first() == Some(&0x16) {
351 0 : return Ok(Some(FeStartupPacket::SslRequest { direct: true }));
352 91 : }
353 91 :
354 91 : // need at least 4 bytes with packet len
355 91 : if buf.len() < 4 {
356 45 : let to_read = 4 - buf.len();
357 45 : buf.reserve(to_read);
358 45 : return Ok(None);
359 46 : }
360 46 :
361 46 : // We shouldn't advance `buf` as probably full message is not there yet,
362 46 : // so can't directly use Bytes::get_u32 etc.
363 46 : let len = (&buf[0..4]).read_u32::<BigEndian>().unwrap() as usize;
364 46 : // The proposed replacement is `!(8..=MAX_STARTUP_PACKET_LENGTH).contains(&len)`
365 46 : // which is less readable
366 46 : #[allow(clippy::manual_range_contains)]
367 46 : if len < 8 || len > MAX_STARTUP_PACKET_LENGTH {
368 1 : return Err(ProtocolError::Protocol(format!(
369 1 : "invalid startup packet message length {}",
370 1 : len
371 1 : )));
372 45 : }
373 45 :
374 45 : if buf.len() < len {
375 : // Don't have full message yet.
376 0 : let to_read = len - buf.len();
377 0 : buf.reserve(to_read);
378 0 : return Ok(None);
379 45 : }
380 45 :
381 45 : // got the message, advance buffer
382 45 : let mut msg = buf.split_to(len).freeze();
383 45 : msg.advance(4); // consume len
384 45 :
385 45 : let request_code = ProtocolVersion(msg.get_u32());
386 : // StartupMessage, CancelRequest, SSLRequest etc are differentiated by request code.
387 45 : let message = match request_code {
388 : CANCEL_REQUEST_CODE => {
389 0 : if msg.remaining() != 8 {
390 0 : return Err(ProtocolError::BadMessage(
391 0 : "CancelRequest message is malformed, backend PID / secret key missing"
392 0 : .to_owned(),
393 0 : ));
394 0 : }
395 0 : FeStartupPacket::CancelRequest(CancelKeyData {
396 0 : backend_pid: msg.get_i32(),
397 0 : cancel_key: msg.get_i32(),
398 0 : })
399 : }
400 : NEGOTIATE_SSL_CODE => {
401 : // Requested upgrade to SSL (aka TLS)
402 21 : FeStartupPacket::SslRequest { direct: false }
403 : }
404 : NEGOTIATE_GSS_CODE => {
405 : // Requested upgrade to GSSAPI
406 0 : FeStartupPacket::GssEncRequest
407 : }
408 24 : version if version.major() == RESERVED_INVALID_MAJOR_VERSION => {
409 0 : return Err(ProtocolError::Protocol(format!(
410 0 : "Unrecognized request code {}",
411 0 : version.minor()
412 0 : )));
413 : }
414 : // TODO bail if protocol major_version is not 3?
415 24 : version => {
416 : // StartupMessage
417 :
418 24 : let s = str::from_utf8(&msg).map_err(|_e| {
419 0 : ProtocolError::BadMessage("StartupMessage params: invalid utf-8".to_owned())
420 24 : })?;
421 24 : let s = s.strip_suffix('\0').ok_or_else(|| {
422 0 : ProtocolError::Protocol(
423 0 : "StartupMessage params: missing null terminator".to_string(),
424 0 : )
425 24 : })?;
426 :
427 24 : FeStartupPacket::StartupMessage {
428 24 : version,
429 24 : params: StartupMessageParams {
430 24 : params: msg.slice_ref(s.as_bytes()),
431 24 : },
432 24 : }
433 : }
434 : };
435 45 : Ok(Some(message))
436 91 : }
437 : }
438 :
439 : impl FeParseMessage {
440 0 : fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
441 : // FIXME: the rust-postgres driver uses a named prepared statement
442 : // for copy_out(). We're not prepared to handle that correctly. For
443 : // now, just ignore the statement name, assuming that the client never
444 : // uses more than one prepared statement at a time.
445 :
446 0 : let _pstmt_name = read_cstr(&mut buf)?;
447 0 : let query_string = read_cstr(&mut buf)?;
448 0 : if buf.remaining() < 2 {
449 0 : return Err(ProtocolError::BadMessage(
450 0 : "Parse message is malformed, nparams missing".to_string(),
451 0 : ));
452 0 : }
453 0 : let nparams = buf.get_i16();
454 0 :
455 0 : if nparams != 0 {
456 0 : return Err(ProtocolError::BadMessage(
457 0 : "query params not implemented".to_string(),
458 0 : ));
459 0 : }
460 0 :
461 0 : Ok(FeMessage::Parse(FeParseMessage { query_string }))
462 0 : }
463 : }
464 :
465 : impl FeDescribeMessage {
466 0 : fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
467 0 : let kind = buf.get_u8();
468 0 : let _pstmt_name = read_cstr(&mut buf)?;
469 :
470 : // FIXME: see FeParseMessage::parse
471 0 : if kind != b'S' {
472 0 : return Err(ProtocolError::BadMessage(
473 0 : "only prepared statemement Describe is implemented".to_string(),
474 0 : ));
475 0 : }
476 0 :
477 0 : Ok(FeMessage::Describe(FeDescribeMessage { kind }))
478 0 : }
479 : }
480 :
481 : impl FeExecuteMessage {
482 0 : fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
483 0 : let portal_name = read_cstr(&mut buf)?;
484 0 : if buf.remaining() < 4 {
485 0 : return Err(ProtocolError::BadMessage(
486 0 : "FeExecuteMessage message is malformed, maxrows missing".to_string(),
487 0 : ));
488 0 : }
489 0 : let maxrows = buf.get_i32();
490 0 :
491 0 : if !portal_name.is_empty() {
492 0 : return Err(ProtocolError::BadMessage(
493 0 : "named portals not implemented".to_string(),
494 0 : ));
495 0 : }
496 0 : if maxrows != 0 {
497 0 : return Err(ProtocolError::BadMessage(
498 0 : "row limit in Execute message not implemented".to_string(),
499 0 : ));
500 0 : }
501 0 :
502 0 : Ok(FeMessage::Execute(FeExecuteMessage { maxrows }))
503 0 : }
504 : }
505 :
506 : impl FeBindMessage {
507 0 : fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
508 0 : let portal_name = read_cstr(&mut buf)?;
509 0 : let _pstmt_name = read_cstr(&mut buf)?;
510 :
511 : // FIXME: see FeParseMessage::parse
512 0 : if !portal_name.is_empty() {
513 0 : return Err(ProtocolError::BadMessage(
514 0 : "named portals not implemented".to_string(),
515 0 : ));
516 0 : }
517 0 :
518 0 : Ok(FeMessage::Bind(FeBindMessage))
519 0 : }
520 : }
521 :
522 : impl FeCloseMessage {
523 0 : fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
524 0 : let _kind = buf.get_u8();
525 0 : let _pstmt_or_portal_name = read_cstr(&mut buf)?;
526 :
527 : // FIXME: we do nothing with Close
528 0 : Ok(FeMessage::Close(FeCloseMessage))
529 0 : }
530 : }
531 :
532 : // Backend
533 :
534 : #[derive(Debug)]
535 : pub enum BeMessage<'a> {
536 : AuthenticationOk,
537 : AuthenticationMD5Password([u8; 4]),
538 : AuthenticationSasl(BeAuthenticationSaslMessage<'a>),
539 : AuthenticationCleartextPassword,
540 : BackendKeyData(CancelKeyData),
541 : BindComplete,
542 : CommandComplete(&'a [u8]),
543 : CopyData(&'a [u8]),
544 : CopyDone,
545 : CopyFail,
546 : CopyInResponse,
547 : CopyOutResponse,
548 : CopyBothResponse,
549 : CloseComplete,
550 : // None means column is NULL
551 : DataRow(&'a [Option<&'a [u8]>]),
552 : // None errcode means internal_error will be sent.
553 : ErrorResponse(&'a str, Option<&'a [u8; 5]>),
554 : /// Single byte - used in response to SSLRequest/GSSENCRequest.
555 : EncryptionResponse(bool),
556 : NoData,
557 : ParameterDescription,
558 : ParameterStatus {
559 : name: &'a [u8],
560 : value: &'a [u8],
561 : },
562 : ParseComplete,
563 : ReadyForQuery,
564 : RowDescription(&'a [RowDescriptor<'a>]),
565 : XLogData(XLogDataBody<'a>),
566 : NoticeResponse(&'a str),
567 : NegotiateProtocolVersion {
568 : version: ProtocolVersion,
569 : options: &'a [&'a str],
570 : },
571 : KeepAlive(WalSndKeepAlive),
572 : /// Batch of interpreted, shard filtered WAL records,
573 : /// ready for the pageserver to ingest
574 : InterpretedWalRecords(InterpretedWalRecordsBody<'a>),
575 :
576 : Raw(u8, &'a [u8]),
577 : }
578 :
579 : /// Common shorthands.
580 : impl<'a> BeMessage<'a> {
581 : /// A [`BeMessage::ParameterStatus`] holding the client encoding, i.e. UTF-8.
582 : /// This is a sensible default, given that:
583 : /// * rust strings only support this encoding out of the box.
584 : /// * tokio-postgres, postgres-jdbc (and probably more) mandate it.
585 : ///
586 : /// TODO: do we need to report `server_encoding` as well?
587 : pub const CLIENT_ENCODING: Self = Self::ParameterStatus {
588 : name: b"client_encoding",
589 : value: b"UTF8",
590 : };
591 :
592 : pub const INTEGER_DATETIMES: Self = Self::ParameterStatus {
593 : name: b"integer_datetimes",
594 : value: b"on",
595 : };
596 :
597 : /// Build a [`BeMessage::ParameterStatus`] holding the server version.
598 2 : pub fn server_version(version: &'a str) -> Self {
599 2 : Self::ParameterStatus {
600 2 : name: b"server_version",
601 2 : value: version.as_bytes(),
602 2 : }
603 2 : }
604 : }
605 :
606 : #[derive(Debug)]
607 : pub enum BeAuthenticationSaslMessage<'a> {
608 : Methods(&'a [&'a str]),
609 : Continue(&'a [u8]),
610 : Final(&'a [u8]),
611 : }
612 :
613 : #[derive(Debug)]
614 : pub enum BeParameterStatusMessage<'a> {
615 : Encoding(&'a str),
616 : ServerVersion(&'a str),
617 : }
618 :
619 : // One row description in RowDescription packet.
620 : #[derive(Debug)]
621 : pub struct RowDescriptor<'a> {
622 : pub name: &'a [u8],
623 : pub tableoid: Oid,
624 : pub attnum: i16,
625 : pub typoid: Oid,
626 : pub typlen: i16,
627 : pub typmod: i32,
628 : pub formatcode: i16,
629 : }
630 :
631 : impl Default for RowDescriptor<'_> {
632 0 : fn default() -> RowDescriptor<'static> {
633 0 : RowDescriptor {
634 0 : name: b"",
635 0 : tableoid: 0,
636 0 : attnum: 0,
637 0 : typoid: 0,
638 0 : typlen: 0,
639 0 : typmod: 0,
640 0 : formatcode: 0,
641 0 : }
642 0 : }
643 : }
644 :
645 : impl RowDescriptor<'_> {
646 : /// Convenience function to create a RowDescriptor message for an int8 column
647 0 : pub const fn int8_col(name: &[u8]) -> RowDescriptor {
648 0 : RowDescriptor {
649 0 : name,
650 0 : tableoid: 0,
651 0 : attnum: 0,
652 0 : typoid: INT8_OID,
653 0 : typlen: 8,
654 0 : typmod: 0,
655 0 : formatcode: 0,
656 0 : }
657 0 : }
658 :
659 2 : pub const fn text_col(name: &[u8]) -> RowDescriptor {
660 2 : RowDescriptor {
661 2 : name,
662 2 : tableoid: 0,
663 2 : attnum: 0,
664 2 : typoid: TEXT_OID,
665 2 : typlen: -1,
666 2 : typmod: 0,
667 2 : formatcode: 0,
668 2 : }
669 2 : }
670 : }
671 :
672 : #[derive(Debug)]
673 : pub struct XLogDataBody<'a> {
674 : pub wal_start: u64,
675 : pub wal_end: u64, // current end of WAL on the server
676 : pub timestamp: i64,
677 : pub data: &'a [u8],
678 : }
679 :
680 : #[derive(Debug)]
681 : pub struct WalSndKeepAlive {
682 : pub wal_end: u64, // current end of WAL on the server
683 : pub timestamp: i64,
684 : pub request_reply: bool,
685 : }
686 :
687 : /// Batch of interpreted WAL records used in the interpreted
688 : /// safekeeper to pageserver protocol.
689 : ///
690 : /// Note that the pageserver uses the RawInterpretedWalRecordsBody
691 : /// counterpart of this from the neondatabase/rust-postgres repo.
692 : /// If you're changing this struct, you likely need to change its
693 : /// twin as well.
694 : #[derive(Debug)]
695 : pub struct InterpretedWalRecordsBody<'a> {
696 : /// End of raw WAL in [`Self::data`]
697 : pub streaming_lsn: u64,
698 : /// Current end of WAL on the server
699 : pub commit_lsn: u64,
700 : pub data: &'a [u8],
701 : }
702 :
703 : pub static HELLO_WORLD_ROW: BeMessage = BeMessage::DataRow(&[Some(b"hello world")]);
704 :
705 : // single text column
706 : pub static SINGLE_COL_ROWDESC: BeMessage = BeMessage::RowDescription(&[RowDescriptor {
707 : name: b"data",
708 : tableoid: 0,
709 : attnum: 0,
710 : typoid: TEXT_OID,
711 : typlen: -1,
712 : typmod: 0,
713 : formatcode: 0,
714 : }]);
715 :
716 : /// Call f() to write body of the message and prepend it with 4-byte len as
717 : /// prescribed by the protocol.
718 75 : fn write_body<R>(buf: &mut BytesMut, f: impl FnOnce(&mut BytesMut) -> R) -> R {
719 75 : let base = buf.len();
720 75 : buf.extend_from_slice(&[0; 4]);
721 75 :
722 75 : let res = f(buf);
723 75 :
724 75 : let size = i32::try_from(buf.len() - base).expect("message too big to transmit");
725 75 : (&mut buf[base..]).put_slice(&size.to_be_bytes());
726 75 :
727 75 : res
728 75 : }
729 :
730 : /// Safe write of s into buf as cstring (String in the protocol).
731 56 : fn write_cstr(s: impl AsRef<[u8]>, buf: &mut BytesMut) -> Result<(), ProtocolError> {
732 56 : let bytes = s.as_ref();
733 56 : if bytes.contains(&0) {
734 0 : return Err(ProtocolError::BadMessage(
735 0 : "string contains embedded null".to_owned(),
736 0 : ));
737 56 : }
738 56 : buf.put_slice(bytes);
739 56 : buf.put_u8(0);
740 56 : Ok(())
741 56 : }
742 :
743 : /// Read cstring from buf, advancing it.
744 11 : pub fn read_cstr(buf: &mut Bytes) -> Result<Bytes, ProtocolError> {
745 11 : let pos = buf
746 11 : .iter()
747 156 : .position(|x| *x == 0)
748 11 : .ok_or_else(|| ProtocolError::BadMessage("missing cstring terminator".to_owned()))?;
749 11 : let result = buf.split_to(pos);
750 11 : buf.advance(1); // drop the null terminator
751 11 : Ok(result)
752 11 : }
753 :
754 : pub const SQLSTATE_INTERNAL_ERROR: &[u8; 5] = b"XX000";
755 : pub const SQLSTATE_ADMIN_SHUTDOWN: &[u8; 5] = b"57P01";
756 : pub const SQLSTATE_SUCCESSFUL_COMPLETION: &[u8; 5] = b"00000";
757 :
758 : impl BeMessage<'_> {
759 : /// Serialize `message` to the given `buf`.
760 : /// Apart from smart memory managemet, BytesMut is good here as msg len
761 : /// precedes its body and it is handy to write it down first and then fill
762 : /// the length. With Write we would have to either calc it manually or have
763 : /// one more buffer.
764 96 : pub fn write(buf: &mut BytesMut, message: &BeMessage) -> Result<(), ProtocolError> {
765 96 : match message {
766 0 : BeMessage::Raw(code, data) => {
767 0 : buf.put_u8(*code);
768 0 : write_body(buf, |b| b.put_slice(data))
769 : }
770 12 : BeMessage::AuthenticationOk => {
771 12 : buf.put_u8(b'R');
772 12 : write_body(buf, |buf| {
773 12 : buf.put_i32(0); // Specifies that the authentication was successful.
774 12 : });
775 12 : }
776 :
777 2 : BeMessage::AuthenticationCleartextPassword => {
778 2 : buf.put_u8(b'R');
779 2 : write_body(buf, |buf| {
780 2 : buf.put_i32(3); // Specifies that clear text password is required.
781 2 : });
782 2 : }
783 :
784 0 : BeMessage::AuthenticationMD5Password(salt) => {
785 0 : buf.put_u8(b'R');
786 0 : write_body(buf, |buf| {
787 0 : buf.put_i32(5); // Specifies that an MD5-encrypted password is required.
788 0 : buf.put_slice(&salt[..]);
789 0 : });
790 0 : }
791 :
792 30 : BeMessage::AuthenticationSasl(msg) => {
793 30 : buf.put_u8(b'R');
794 30 : write_body(buf, |buf| {
795 : use BeAuthenticationSaslMessage::*;
796 30 : match msg {
797 13 : Methods(methods) => {
798 13 : buf.put_i32(10); // Specifies that SASL auth method is used.
799 25 : for method in methods.iter() {
800 25 : write_cstr(method, buf)?;
801 : }
802 13 : buf.put_u8(0); // zero terminator for the list
803 : }
804 11 : Continue(extra) => {
805 11 : buf.put_i32(11); // Continue SASL auth.
806 11 : buf.put_slice(extra);
807 11 : }
808 6 : Final(extra) => {
809 6 : buf.put_i32(12); // Send final SASL message.
810 6 : buf.put_slice(extra);
811 6 : }
812 : }
813 30 : Ok(())
814 30 : })?;
815 : }
816 :
817 0 : BeMessage::BackendKeyData(key_data) => {
818 0 : buf.put_u8(b'K');
819 0 : write_body(buf, |buf| {
820 0 : buf.put_i32(key_data.backend_pid);
821 0 : buf.put_i32(key_data.cancel_key);
822 0 : });
823 0 : }
824 :
825 0 : BeMessage::BindComplete => {
826 0 : buf.put_u8(b'2');
827 0 : write_body(buf, |_| {});
828 0 : }
829 :
830 0 : BeMessage::CloseComplete => {
831 0 : buf.put_u8(b'3');
832 0 : write_body(buf, |_| {});
833 0 : }
834 :
835 2 : BeMessage::CommandComplete(cmd) => {
836 2 : buf.put_u8(b'C');
837 2 : write_body(buf, |buf| write_cstr(cmd, buf))?;
838 : }
839 :
840 0 : BeMessage::CopyData(data) => {
841 0 : buf.put_u8(b'd');
842 0 : write_body(buf, |buf| {
843 0 : buf.put_slice(data);
844 0 : });
845 0 : }
846 :
847 0 : BeMessage::CopyDone => {
848 0 : buf.put_u8(b'c');
849 0 : write_body(buf, |_| {});
850 0 : }
851 :
852 0 : BeMessage::CopyFail => {
853 0 : buf.put_u8(b'f');
854 0 : write_body(buf, |_| {});
855 0 : }
856 :
857 0 : BeMessage::CopyInResponse => {
858 0 : buf.put_u8(b'G');
859 0 : write_body(buf, |buf| {
860 0 : buf.put_u8(1); // copy_is_binary
861 0 : buf.put_i16(0); // numAttributes
862 0 : });
863 0 : }
864 :
865 0 : BeMessage::CopyOutResponse => {
866 0 : buf.put_u8(b'H');
867 0 : write_body(buf, |buf| {
868 0 : buf.put_u8(0); // copy_is_binary
869 0 : buf.put_i16(0); // numAttributes
870 0 : });
871 0 : }
872 :
873 0 : BeMessage::CopyBothResponse => {
874 0 : buf.put_u8(b'W');
875 0 : write_body(buf, |buf| {
876 0 : // doesn't matter, used only for replication
877 0 : buf.put_u8(0); // copy_is_binary
878 0 : buf.put_i16(0); // numAttributes
879 0 : });
880 0 : }
881 :
882 2 : BeMessage::DataRow(vals) => {
883 2 : buf.put_u8(b'D');
884 2 : write_body(buf, |buf| {
885 2 : buf.put_u16(vals.len() as u16); // num of cols
886 2 : for val_opt in vals.iter() {
887 2 : if let Some(val) = val_opt {
888 2 : buf.put_u32(val.len() as u32);
889 2 : buf.put_slice(val);
890 2 : } else {
891 0 : buf.put_i32(-1);
892 0 : }
893 : }
894 2 : });
895 2 : }
896 :
897 : // ErrorResponse is a zero-terminated array of zero-terminated fields.
898 : // First byte of each field represents type of this field. Set just enough fields
899 : // to satisfy rust-postgres client: 'S' -- severity, 'C' -- error, 'M' -- error
900 : // message text.
901 1 : BeMessage::ErrorResponse(error_msg, pg_error_code) => {
902 1 : // 'E' signalizes ErrorResponse messages
903 1 : buf.put_u8(b'E');
904 1 : write_body(buf, |buf| {
905 1 : buf.put_u8(b'S'); // severity
906 1 : buf.put_slice(b"ERROR\0");
907 1 :
908 1 : buf.put_u8(b'C'); // SQLSTATE error code
909 1 : buf.put_slice(&terminate_code(
910 1 : pg_error_code.unwrap_or(SQLSTATE_INTERNAL_ERROR),
911 1 : ));
912 1 :
913 1 : buf.put_u8(b'M'); // the message
914 1 : write_cstr(error_msg, buf)?;
915 :
916 1 : buf.put_u8(0); // terminator
917 1 : Ok(())
918 1 : })?;
919 : }
920 :
921 : // NoticeResponse has the same format as ErrorResponse. From doc: "The frontend should display the
922 : // message but continue listening for ReadyForQuery or ErrorResponse"
923 0 : BeMessage::NoticeResponse(error_msg) => {
924 0 : // For all the errors set Severity to Error and error code to
925 0 : // 'internal error'.
926 0 :
927 0 : // 'N' signalizes NoticeResponse messages
928 0 : buf.put_u8(b'N');
929 0 : write_body(buf, |buf| {
930 0 : buf.put_u8(b'S'); // severity
931 0 : buf.put_slice(b"NOTICE\0");
932 0 :
933 0 : buf.put_u8(b'C'); // SQLSTATE error code
934 0 : buf.put_slice(&terminate_code(SQLSTATE_INTERNAL_ERROR));
935 0 :
936 0 : buf.put_u8(b'M'); // the message
937 0 : write_cstr(error_msg.as_bytes(), buf)?;
938 :
939 0 : buf.put_u8(0); // terminator
940 0 : Ok(())
941 0 : })?;
942 : }
943 :
944 0 : BeMessage::NoData => {
945 0 : buf.put_u8(b'n');
946 0 : write_body(buf, |_| {});
947 0 : }
948 :
949 21 : BeMessage::EncryptionResponse(should_negotiate) => {
950 21 : let response = if *should_negotiate { b'S' } else { b'N' };
951 21 : buf.put_u8(response);
952 : }
953 :
954 13 : BeMessage::ParameterStatus { name, value } => {
955 13 : buf.put_u8(b'S');
956 13 : write_body(buf, |buf| {
957 13 : write_cstr(name, buf)?;
958 13 : write_cstr(value, buf)
959 13 : })?;
960 : }
961 :
962 0 : BeMessage::ParameterDescription => {
963 0 : buf.put_u8(b't');
964 0 : write_body(buf, |buf| {
965 0 : // we don't support params, so always 0
966 0 : buf.put_i16(0);
967 0 : });
968 0 : }
969 :
970 0 : BeMessage::ParseComplete => {
971 0 : buf.put_u8(b'1');
972 0 : write_body(buf, |_| {});
973 0 : }
974 :
975 11 : BeMessage::ReadyForQuery => {
976 11 : buf.put_u8(b'Z');
977 11 : write_body(buf, |buf| {
978 11 : buf.put_u8(b'I');
979 11 : });
980 11 : }
981 :
982 2 : BeMessage::RowDescription(rows) => {
983 2 : buf.put_u8(b'T');
984 2 : write_body(buf, |buf| {
985 2 : buf.put_i16(rows.len() as i16); // # of fields
986 2 : for row in rows.iter() {
987 2 : write_cstr(row.name, buf)?;
988 2 : buf.put_i32(0); /* table oid */
989 2 : buf.put_i16(0); /* attnum */
990 2 : buf.put_u32(row.typoid);
991 2 : buf.put_i16(row.typlen);
992 2 : buf.put_i32(-1); /* typmod */
993 2 : buf.put_i16(0); /* format code */
994 : }
995 2 : Ok(())
996 2 : })?;
997 : }
998 :
999 0 : BeMessage::XLogData(body) => {
1000 0 : buf.put_u8(b'd');
1001 0 : write_body(buf, |buf| {
1002 0 : buf.put_u8(b'w');
1003 0 : buf.put_u64(body.wal_start);
1004 0 : buf.put_u64(body.wal_end);
1005 0 : buf.put_i64(body.timestamp);
1006 0 : buf.put_slice(body.data);
1007 0 : });
1008 0 : }
1009 :
1010 0 : BeMessage::KeepAlive(req) => {
1011 0 : buf.put_u8(b'd');
1012 0 : write_body(buf, |buf| {
1013 0 : buf.put_u8(b'k');
1014 0 : buf.put_u64(req.wal_end);
1015 0 : buf.put_i64(req.timestamp);
1016 0 : buf.put_u8(u8::from(req.request_reply));
1017 0 : });
1018 0 : }
1019 :
1020 0 : BeMessage::NegotiateProtocolVersion { version, options } => {
1021 0 : buf.put_u8(b'v');
1022 0 : write_body(buf, |buf| {
1023 0 : buf.put_u32(version.0);
1024 0 : buf.put_u32(options.len() as u32);
1025 0 : for option in options.iter() {
1026 0 : write_cstr(option, buf)?;
1027 : }
1028 0 : Ok(())
1029 0 : })?
1030 : }
1031 :
1032 0 : BeMessage::InterpretedWalRecords(rec) => {
1033 0 : // We use the COPY_DATA_TAG for our custom message
1034 0 : // since this tag is interpreted as raw bytes.
1035 0 : buf.put_u8(b'd');
1036 0 : write_body(buf, |buf| {
1037 0 : buf.put_u8(b'0'); // matches INTERPRETED_WAL_RECORD_TAG in postgres-protocol
1038 0 : // dependency
1039 0 : buf.put_u64(rec.streaming_lsn);
1040 0 : buf.put_u64(rec.commit_lsn);
1041 0 : buf.put_slice(rec.data);
1042 0 : });
1043 0 : }
1044 : }
1045 96 : Ok(())
1046 96 : }
1047 : }
1048 :
1049 1 : fn terminate_code(code: &[u8; 5]) -> [u8; 6] {
1050 1 : let mut terminated = [0; 6];
1051 5 : for (i, &elem) in code.iter().enumerate() {
1052 5 : terminated[i] = elem;
1053 5 : }
1054 :
1055 1 : terminated
1056 1 : }
1057 :
1058 : #[cfg(test)]
1059 : mod tests {
1060 : use super::*;
1061 :
1062 : #[test]
1063 1 : fn test_startup_message_params_options_escaped() {
1064 4 : fn split_options(params: &StartupMessageParams) -> Vec<Cow<'_, str>> {
1065 4 : params
1066 4 : .options_escaped()
1067 4 : .expect("options are None")
1068 4 : .collect()
1069 4 : }
1070 :
1071 4 : let make_params = |options| StartupMessageParams::new([("options", options)]);
1072 :
1073 1 : let params = StartupMessageParams::new([]);
1074 1 : assert!(params.options_escaped().is_none());
1075 :
1076 1 : let params = make_params("");
1077 1 : assert!(split_options(¶ms).is_empty());
1078 :
1079 1 : let params = make_params("foo");
1080 1 : assert_eq!(split_options(¶ms), ["foo"]);
1081 :
1082 1 : let params = make_params(" foo bar ");
1083 1 : assert_eq!(split_options(¶ms), ["foo", "bar"]);
1084 :
1085 1 : let params = make_params("foo\\ bar \\ \\\\ baz\\ lol");
1086 1 : assert_eq!(split_options(¶ms), ["foo bar", " \\", "baz ", "lol"]);
1087 1 : }
1088 :
1089 : #[test]
1090 1 : fn parse_fe_startup_packet_regression() {
1091 1 : let data = [0, 0, 0, 7, 0, 0, 0, 0];
1092 1 : FeStartupPacket::parse(&mut BytesMut::from_iter(data)).unwrap_err();
1093 1 : }
1094 :
1095 : #[test]
1096 1 : fn cancel_key_data() {
1097 1 : let key = CancelKeyData {
1098 1 : backend_pid: -1817212860,
1099 1 : cancel_key: -1183897012,
1100 1 : };
1101 1 : assert_eq!(format!("{key}"), "CancelKeyData(93af8844b96f2a4c)");
1102 1 : }
1103 : }
|
