2 * Implements packet I/O in a pthread.
3 * Copyright (C) 2017 Cumulus Networks
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; see the file COPYING; if not, write to the
18 * Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston,
22 /* clang-format off */
24 #include <pthread.h> // for pthread_mutex_unlock, pthread_mutex_lock
26 #include "frr_pthread.h"
27 #include "linklist.h" // for list_delete, list_delete_all_node, lis...
28 #include "log.h" // for zlog_debug, safe_strerror, zlog_err
29 #include "memory.h" // for MTYPE_TMP, XCALLOC, XFREE
30 #include "network.h" // for ERRNO_IO_RETRY
31 #include "stream.h" // for stream_get_endp, stream_getw_from, str...
32 #include "ringbuf.h" // for ringbuf_remain, ringbuf_peek, ringbuf_...
33 #include "thread.h" // for THREAD_OFF, THREAD_ARG, thread, thread...
34 #include "zassert.h" // for assert
36 #include "bgpd/bgp_io.h"
37 #include "bgpd/bgp_debug.h" // for bgp_debug_neighbor_events, bgp_type_str
38 #include "bgpd/bgp_errors.h" // for expanded error reference information
39 #include "bgpd/bgp_fsm.h" // for BGP_EVENT_ADD, bgp_event
40 #include "bgpd/bgp_packet.h" // for bgp_notify_send_with_data, bgp_notify...
41 #include "bgpd/bgpd.h" // for peer, BGP_MARKER_SIZE, bgp_master, bm
44 /* forward declarations */
45 static uint16_t bgp_write(struct peer
*);
46 static uint16_t bgp_read(struct peer
*);
47 static int bgp_process_writes(struct thread
*);
48 static int bgp_process_reads(struct thread
*);
49 static bool validate_header(struct peer
*);
51 /* generic i/o status codes */
52 #define BGP_IO_TRANS_ERR (1 << 0) // EAGAIN or similar occurred
53 #define BGP_IO_FATAL_ERR (1 << 1) // some kind of fatal TCP error
55 /* Thread external API ----------------------------------------------------- */
57 void bgp_writes_on(struct peer
*peer
)
59 struct frr_pthread
*fpt
= bgp_pth_io
;
62 assert(peer
->status
!= Deleted
);
65 assert(peer
->ibuf_work
);
66 assert(!peer
->t_connect_check_r
);
67 assert(!peer
->t_connect_check_w
);
70 thread_add_write(fpt
->master
, bgp_process_writes
, peer
, peer
->fd
,
72 SET_FLAG(peer
->thread_flags
, PEER_THREAD_WRITES_ON
);
75 void bgp_writes_off(struct peer
*peer
)
77 struct frr_pthread
*fpt
= bgp_pth_io
;
80 thread_cancel_async(fpt
->master
, &peer
->t_write
, NULL
);
81 THREAD_OFF(peer
->t_generate_updgrp_packets
);
83 UNSET_FLAG(peer
->thread_flags
, PEER_THREAD_WRITES_ON
);
86 void bgp_reads_on(struct peer
*peer
)
88 struct frr_pthread
*fpt
= bgp_pth_io
;
91 assert(peer
->status
!= Deleted
);
94 assert(peer
->ibuf_work
);
96 assert(!peer
->t_connect_check_r
);
97 assert(!peer
->t_connect_check_w
);
100 thread_add_read(fpt
->master
, bgp_process_reads
, peer
, peer
->fd
,
103 SET_FLAG(peer
->thread_flags
, PEER_THREAD_READS_ON
);
106 void bgp_reads_off(struct peer
*peer
)
108 struct frr_pthread
*fpt
= bgp_pth_io
;
109 assert(fpt
->running
);
111 thread_cancel_async(fpt
->master
, &peer
->t_read
, NULL
);
112 THREAD_OFF(peer
->t_process_packet
);
114 UNSET_FLAG(peer
->thread_flags
, PEER_THREAD_READS_ON
);
117 /* Thread internal functions ----------------------------------------------- */
120 * Called from I/O pthread when a file descriptor has become ready for writing.
122 static int bgp_process_writes(struct thread
*thread
)
124 static struct peer
*peer
;
125 peer
= THREAD_ARG(thread
);
133 struct frr_pthread
*fpt
= bgp_pth_io
;
135 pthread_mutex_lock(&peer
->io_mtx
);
137 status
= bgp_write(peer
);
138 reschedule
= (stream_fifo_head(peer
->obuf
) != NULL
);
140 pthread_mutex_unlock(&peer
->io_mtx
);
143 if (CHECK_FLAG(status
, BGP_IO_TRANS_ERR
)) {
147 if (CHECK_FLAG(status
, BGP_IO_FATAL_ERR
)) {
153 thread_add_write(fpt
->master
, bgp_process_writes
, peer
,
154 peer
->fd
, &peer
->t_write
);
156 BGP_TIMER_ON(peer
->t_generate_updgrp_packets
,
157 bgp_generate_updgrp_packets
, 0);
164 * Called from I/O pthread when a file descriptor has become ready for reading,
167 * We read as much data as possible, process as many packets as we can and
168 * place them on peer->ibuf for secondary processing by the main thread.
170 static int bgp_process_reads(struct thread
*thread
)
172 /* clang-format off */
173 static struct peer
*peer
; // peer to read from
174 uint16_t status
; // bgp_read status code
175 bool more
= true; // whether we got more data
176 bool fatal
= false; // whether fatal error occurred
177 bool added_pkt
= false; // whether we pushed onto ->ibuf
178 /* clang-format on */
180 peer
= THREAD_ARG(thread
);
182 if (peer
->fd
< 0 || bm
->terminating
)
185 struct frr_pthread
*fpt
= bgp_pth_io
;
187 pthread_mutex_lock(&peer
->io_mtx
);
189 status
= bgp_read(peer
);
191 pthread_mutex_unlock(&peer
->io_mtx
);
193 /* error checking phase */
194 if (CHECK_FLAG(status
, BGP_IO_TRANS_ERR
)) {
195 /* no problem; just don't process packets */
199 if (CHECK_FLAG(status
, BGP_IO_FATAL_ERR
)) {
200 /* problem; tear down session */
206 /* static buffer for transferring packets */
207 static unsigned char pktbuf
[BGP_MAX_PACKET_SIZE
];
208 /* shorter alias to peer's input buffer */
209 struct ringbuf
*ibw
= peer
->ibuf_work
;
210 /* packet size as given by header */
211 uint16_t pktsize
= 0;
213 /* check that we have enough data for a header */
214 if (ringbuf_remain(ibw
) < BGP_HEADER_SIZE
)
217 /* check that header is valid */
218 if (!validate_header(peer
)) {
223 /* header is valid; retrieve packet size */
224 ringbuf_peek(ibw
, BGP_MARKER_SIZE
, &pktsize
, sizeof(pktsize
));
226 pktsize
= ntohs(pktsize
);
228 /* if this fails we are seriously screwed */
229 assert(pktsize
<= BGP_MAX_PACKET_SIZE
);
232 * If we have that much data, chuck it into its own
233 * stream and append to input queue for processing.
235 if (ringbuf_remain(ibw
) >= pktsize
) {
236 struct stream
*pkt
= stream_new(pktsize
);
237 assert(ringbuf_get(ibw
, pktbuf
, pktsize
) == pktsize
);
238 stream_put(pkt
, pktbuf
, pktsize
);
240 pthread_mutex_lock(&peer
->io_mtx
);
242 stream_fifo_push(peer
->ibuf
, pkt
);
244 pthread_mutex_unlock(&peer
->io_mtx
);
251 assert(ringbuf_space(peer
->ibuf_work
) >= BGP_MAX_PACKET_SIZE
);
253 /* handle invalid header */
255 /* wipe buffer just in case someone screwed up */
256 ringbuf_wipe(peer
->ibuf_work
);
258 thread_add_read(fpt
->master
, bgp_process_reads
, peer
, peer
->fd
,
261 thread_add_timer_msec(bm
->master
, bgp_process_packet
,
262 peer
, 0, &peer
->t_process_packet
);
269 * Flush peer output buffer.
271 * This function pops packets off of peer->obuf and writes them to peer->fd.
272 * The amount of packets written is equal to the minimum of peer->wpkt_quanta
273 * and the number of packets on the output buffer, unless an error occurs.
275 * If write() returns an error, the appropriate FSM event is generated.
277 * The return value is equal to the number of packets written
278 * (which may be zero).
280 static uint16_t bgp_write(struct peer
*peer
)
285 int update_last_write
= 0;
286 unsigned int count
= 0;
289 uint32_t wpkt_quanta_old
;
291 wpkt_quanta_old
= atomic_load_explicit(&peer
->bgp
->wpkt_quanta
,
292 memory_order_relaxed
);
294 while (count
< wpkt_quanta_old
&& (s
= stream_fifo_head(peer
->obuf
))) {
297 writenum
= stream_get_endp(s
) - stream_get_getp(s
);
298 num
= write(peer
->fd
, stream_pnt(s
), writenum
);
301 if (!ERRNO_IO_RETRY(errno
)) {
302 BGP_EVENT_ADD(peer
, TCP_fatal_error
);
303 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
305 SET_FLAG(status
, BGP_IO_TRANS_ERR
);
309 } else if (num
!= writenum
)
310 stream_forward_getp(s
, num
);
312 } while (num
!= writenum
);
314 /* Retrieve BGP packet type. */
315 stream_set_getp(s
, BGP_MARKER_SIZE
+ 2);
316 type
= stream_getc(s
);
320 atomic_fetch_add_explicit(&peer
->open_out
, 1,
321 memory_order_relaxed
);
324 atomic_fetch_add_explicit(&peer
->update_out
, 1,
325 memory_order_relaxed
);
329 atomic_fetch_add_explicit(&peer
->notify_out
, 1,
330 memory_order_relaxed
);
331 /* Double start timer. */
334 /* Overflow check. */
335 if (peer
->v_start
>= (60 * 2))
336 peer
->v_start
= (60 * 2);
339 * Handle Graceful Restart case where the state changes
340 * to Connect instead of Idle.
342 BGP_EVENT_ADD(peer
, BGP_Stop
);
345 case BGP_MSG_KEEPALIVE
:
346 atomic_fetch_add_explicit(&peer
->keepalive_out
, 1,
347 memory_order_relaxed
);
349 case BGP_MSG_ROUTE_REFRESH_NEW
:
350 case BGP_MSG_ROUTE_REFRESH_OLD
:
351 atomic_fetch_add_explicit(&peer
->refresh_out
, 1,
352 memory_order_relaxed
);
354 case BGP_MSG_CAPABILITY
:
355 atomic_fetch_add_explicit(&peer
->dynamic_cap_out
, 1,
356 memory_order_relaxed
);
362 stream_free(stream_fifo_pop(peer
->obuf
));
363 update_last_write
= 1;
368 * Update last_update if UPDATEs were written.
369 * Note: that these are only updated at end,
370 * not per message (i.e., per loop)
373 atomic_store_explicit(&peer
->last_update
, bgp_clock(),
374 memory_order_relaxed
);
376 /* If we TXed any flavor of packet */
377 if (update_last_write
)
378 atomic_store_explicit(&peer
->last_write
, bgp_clock(),
379 memory_order_relaxed
);
386 * Reads a chunk of data from peer->fd into peer->ibuf_work.
388 * @return status flag (see top-of-file)
390 static uint16_t bgp_read(struct peer
*peer
)
392 size_t readsize
; // how many bytes we want to read
393 ssize_t nbytes
; // how many bytes we actually read
395 static uint8_t ibw
[BGP_MAX_PACKET_SIZE
* BGP_READ_PACKET_MAX
];
397 readsize
= MIN(ringbuf_space(peer
->ibuf_work
), sizeof(ibw
));
398 nbytes
= read(peer
->fd
, ibw
, readsize
);
400 /* EAGAIN or EWOULDBLOCK; come back later */
401 if (nbytes
< 0 && ERRNO_IO_RETRY(errno
)) {
402 SET_FLAG(status
, BGP_IO_TRANS_ERR
);
403 /* Fatal error; tear down session */
404 } else if (nbytes
< 0) {
405 flog_err(EC_BGP_UPDATE_RCV
,
406 "%s [Error] bgp_read_packet error: %s", peer
->host
,
407 safe_strerror(errno
));
409 if (peer
->status
== Established
) {
410 if (CHECK_FLAG(peer
->sflags
, PEER_STATUS_NSF_MODE
)) {
411 peer
->last_reset
= PEER_DOWN_NSF_CLOSE_SESSION
;
412 SET_FLAG(peer
->sflags
, PEER_STATUS_NSF_WAIT
);
414 peer
->last_reset
= PEER_DOWN_CLOSE_SESSION
;
417 BGP_EVENT_ADD(peer
, TCP_fatal_error
);
418 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
419 /* Received EOF / TCP session closed */
420 } else if (nbytes
== 0) {
421 if (bgp_debug_neighbor_events(peer
))
422 zlog_debug("%s [Event] BGP connection closed fd %d",
423 peer
->host
, peer
->fd
);
425 if (peer
->status
== Established
) {
426 if (CHECK_FLAG(peer
->sflags
, PEER_STATUS_NSF_MODE
)) {
427 peer
->last_reset
= PEER_DOWN_NSF_CLOSE_SESSION
;
428 SET_FLAG(peer
->sflags
, PEER_STATUS_NSF_WAIT
);
430 peer
->last_reset
= PEER_DOWN_CLOSE_SESSION
;
433 BGP_EVENT_ADD(peer
, TCP_connection_closed
);
434 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
436 assert(ringbuf_put(peer
->ibuf_work
, ibw
, nbytes
)
444 * Called after we have read a BGP packet header. Validates marker, message
445 * type and packet length. If any of these aren't correct, sends a notify.
447 * Assumes that there are at least BGP_HEADER_SIZE readable bytes in the input
450 static bool validate_header(struct peer
*peer
)
454 struct ringbuf
*pkt
= peer
->ibuf_work
;
456 static uint8_t m_correct
[BGP_MARKER_SIZE
] = {
457 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
458 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
459 uint8_t m_rx
[BGP_MARKER_SIZE
] = {0x00};
461 if (ringbuf_peek(pkt
, 0, m_rx
, BGP_MARKER_SIZE
) != BGP_MARKER_SIZE
)
464 if (memcmp(m_correct
, m_rx
, BGP_MARKER_SIZE
) != 0) {
465 bgp_notify_send(peer
, BGP_NOTIFY_HEADER_ERR
,
466 BGP_NOTIFY_HEADER_NOT_SYNC
);
470 /* Get size and type in network byte order. */
471 ringbuf_peek(pkt
, BGP_MARKER_SIZE
, &size
, sizeof(size
));
472 ringbuf_peek(pkt
, BGP_MARKER_SIZE
+ 2, &type
, sizeof(type
));
476 /* BGP type check. */
477 if (type
!= BGP_MSG_OPEN
&& type
!= BGP_MSG_UPDATE
478 && type
!= BGP_MSG_NOTIFY
&& type
!= BGP_MSG_KEEPALIVE
479 && type
!= BGP_MSG_ROUTE_REFRESH_NEW
480 && type
!= BGP_MSG_ROUTE_REFRESH_OLD
481 && type
!= BGP_MSG_CAPABILITY
) {
482 if (bgp_debug_neighbor_events(peer
))
483 zlog_debug("%s unknown message type 0x%02x", peer
->host
,
486 bgp_notify_send_with_data(peer
, BGP_NOTIFY_HEADER_ERR
,
487 BGP_NOTIFY_HEADER_BAD_MESTYPE
, &type
,
492 /* Minimum packet length check. */
493 if ((size
< BGP_HEADER_SIZE
) || (size
> BGP_MAX_PACKET_SIZE
)
494 || (type
== BGP_MSG_OPEN
&& size
< BGP_MSG_OPEN_MIN_SIZE
)
495 || (type
== BGP_MSG_UPDATE
&& size
< BGP_MSG_UPDATE_MIN_SIZE
)
496 || (type
== BGP_MSG_NOTIFY
&& size
< BGP_MSG_NOTIFY_MIN_SIZE
)
497 || (type
== BGP_MSG_KEEPALIVE
&& size
!= BGP_MSG_KEEPALIVE_MIN_SIZE
)
498 || (type
== BGP_MSG_ROUTE_REFRESH_NEW
499 && size
< BGP_MSG_ROUTE_REFRESH_MIN_SIZE
)
500 || (type
== BGP_MSG_ROUTE_REFRESH_OLD
501 && size
< BGP_MSG_ROUTE_REFRESH_MIN_SIZE
)
502 || (type
== BGP_MSG_CAPABILITY
503 && size
< BGP_MSG_CAPABILITY_MIN_SIZE
)) {
504 if (bgp_debug_neighbor_events(peer
)) {
505 zlog_debug("%s bad message length - %d for %s",
507 type
== 128 ? "ROUTE-REFRESH"
508 : bgp_type_str
[(int)type
]);
511 uint16_t nsize
= htons(size
);
513 bgp_notify_send_with_data(peer
, BGP_NOTIFY_HEADER_ERR
,
514 BGP_NOTIFY_HEADER_BAD_MESLEN
,
515 (unsigned char *)&nsize
, 2);