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
25 #include <sys/uio.h> // for writev
27 #include "frr_pthread.h"
28 #include "linklist.h" // for list_delete, list_delete_all_node, lis...
29 #include "log.h" // for zlog_debug, safe_strerror, zlog_err
30 #include "memory.h" // for MTYPE_TMP, XCALLOC, XFREE
31 #include "network.h" // for ERRNO_IO_RETRY
32 #include "stream.h" // for stream_get_endp, stream_getw_from, str...
33 #include "ringbuf.h" // for ringbuf_remain, ringbuf_peek, ringbuf_...
34 #include "thread.h" // for THREAD_OFF, THREAD_ARG, thread...
35 #include "zassert.h" // for assert
37 #include "bgpd/bgp_io.h"
38 #include "bgpd/bgp_debug.h" // for bgp_debug_neighbor_events, bgp_type_str
39 #include "bgpd/bgp_errors.h" // for expanded error reference information
40 #include "bgpd/bgp_fsm.h" // for BGP_EVENT_ADD, bgp_event
41 #include "bgpd/bgp_packet.h" // for bgp_notify_send_with_data, bgp_notify...
42 #include "bgpd/bgp_trace.h" // for frrtraces
43 #include "bgpd/bgpd.h" // for peer, BGP_MARKER_SIZE, bgp_master, bm
46 /* forward declarations */
47 static uint16_t bgp_write(struct peer
*);
48 static uint16_t bgp_read(struct peer
*);
49 static int bgp_process_writes(struct thread
*);
50 static int bgp_process_reads(struct thread
*);
51 static bool validate_header(struct peer
*);
53 /* generic i/o status codes */
54 #define BGP_IO_TRANS_ERR (1 << 0) // EAGAIN or similar occurred
55 #define BGP_IO_FATAL_ERR (1 << 1) // some kind of fatal TCP error
57 /* Thread external API ----------------------------------------------------- */
59 void bgp_writes_on(struct peer
*peer
)
61 struct frr_pthread
*fpt
= bgp_pth_io
;
64 assert(peer
->status
!= Deleted
);
67 assert(peer
->ibuf_work
);
68 assert(!peer
->t_connect_check_r
);
69 assert(!peer
->t_connect_check_w
);
72 thread_add_write(fpt
->master
, bgp_process_writes
, peer
, peer
->fd
,
74 SET_FLAG(peer
->thread_flags
, PEER_THREAD_WRITES_ON
);
77 void bgp_writes_off(struct peer
*peer
)
79 struct frr_pthread
*fpt
= bgp_pth_io
;
82 thread_cancel_async(fpt
->master
, &peer
->t_write
, NULL
);
83 THREAD_OFF(peer
->t_generate_updgrp_packets
);
85 UNSET_FLAG(peer
->thread_flags
, PEER_THREAD_WRITES_ON
);
88 void bgp_reads_on(struct peer
*peer
)
90 struct frr_pthread
*fpt
= bgp_pth_io
;
93 assert(peer
->status
!= Deleted
);
96 assert(peer
->ibuf_work
);
98 assert(!peer
->t_connect_check_r
);
99 assert(!peer
->t_connect_check_w
);
102 thread_add_read(fpt
->master
, bgp_process_reads
, peer
, peer
->fd
,
105 SET_FLAG(peer
->thread_flags
, PEER_THREAD_READS_ON
);
108 void bgp_reads_off(struct peer
*peer
)
110 struct frr_pthread
*fpt
= bgp_pth_io
;
111 assert(fpt
->running
);
113 thread_cancel_async(fpt
->master
, &peer
->t_read
, NULL
);
114 THREAD_OFF(peer
->t_process_packet
);
116 UNSET_FLAG(peer
->thread_flags
, PEER_THREAD_READS_ON
);
119 /* Thread internal functions ----------------------------------------------- */
122 * Called from I/O pthread when a file descriptor has become ready for writing.
124 static int bgp_process_writes(struct thread
*thread
)
126 static struct peer
*peer
;
127 peer
= THREAD_ARG(thread
);
135 struct frr_pthread
*fpt
= bgp_pth_io
;
137 frr_with_mutex(&peer
->io_mtx
) {
138 status
= bgp_write(peer
);
139 reschedule
= (stream_fifo_head(peer
->obuf
) != NULL
);
143 if (CHECK_FLAG(status
, BGP_IO_TRANS_ERR
)) {
147 if (CHECK_FLAG(status
, BGP_IO_FATAL_ERR
)) {
152 /* If suppress fib pending is enabled, route is advertised to peers when
153 * the status is received from the FIB. The delay is added
154 * to update group packet generate which will allow more routes to be
155 * sent in the update message
158 thread_add_write(fpt
->master
, bgp_process_writes
, peer
,
159 peer
->fd
, &peer
->t_write
);
161 BGP_UPDATE_GROUP_TIMER_ON(&peer
->t_generate_updgrp_packets
,
162 bgp_generate_updgrp_packets
);
169 * Called from I/O pthread when a file descriptor has become ready for reading,
172 * We read as much data as possible, process as many packets as we can and
173 * place them on peer->ibuf for secondary processing by the main thread.
175 static int bgp_process_reads(struct thread
*thread
)
177 /* clang-format off */
178 static struct peer
*peer
; // peer to read from
179 uint16_t status
; // bgp_read status code
180 bool more
= true; // whether we got more data
181 bool fatal
= false; // whether fatal error occurred
182 bool added_pkt
= false; // whether we pushed onto ->ibuf
183 /* clang-format on */
185 peer
= THREAD_ARG(thread
);
187 if (peer
->fd
< 0 || bm
->terminating
)
190 struct frr_pthread
*fpt
= bgp_pth_io
;
192 frr_with_mutex(&peer
->io_mtx
) {
193 status
= bgp_read(peer
);
196 /* error checking phase */
197 if (CHECK_FLAG(status
, BGP_IO_TRANS_ERR
)) {
198 /* no problem; just don't process packets */
202 if (CHECK_FLAG(status
, BGP_IO_FATAL_ERR
)) {
203 /* problem; tear down session */
209 /* static buffer for transferring packets */
210 /* shorter alias to peer's input buffer */
211 struct ringbuf
*ibw
= peer
->ibuf_work
;
212 /* packet size as given by header */
213 uint16_t pktsize
= 0;
215 /* check that we have enough data for a header */
216 if (ringbuf_remain(ibw
) < BGP_HEADER_SIZE
)
219 /* check that header is valid */
220 if (!validate_header(peer
)) {
225 /* header is valid; retrieve packet size */
226 ringbuf_peek(ibw
, BGP_MARKER_SIZE
, &pktsize
, sizeof(pktsize
));
228 pktsize
= ntohs(pktsize
);
230 /* if this fails we are seriously screwed */
231 assert(pktsize
<= BGP_MAX_PACKET_SIZE
);
234 * If we have that much data, chuck it into its own
235 * stream and append to input queue for processing.
237 if (ringbuf_remain(ibw
) >= pktsize
) {
238 struct stream
*pkt
= stream_new(pktsize
);
239 assert(STREAM_WRITEABLE(pkt
) == pktsize
);
240 assert(ringbuf_get(ibw
, pkt
->data
, pktsize
) == pktsize
);
241 stream_set_endp(pkt
, pktsize
);
243 frrtrace(2, frr_bgp
, packet_read
, peer
, pkt
);
244 frr_with_mutex(&peer
->io_mtx
) {
245 stream_fifo_push(peer
->ibuf
, pkt
);
253 /* handle invalid header */
255 /* wipe buffer just in case someone screwed up */
256 ringbuf_wipe(peer
->ibuf_work
);
258 assert(ringbuf_space(peer
->ibuf_work
) >= BGP_MAX_PACKET_SIZE
);
260 thread_add_read(fpt
->master
, bgp_process_reads
, peer
, peer
->fd
,
263 thread_add_timer_msec(bm
->master
, bgp_process_packet
,
264 peer
, 0, &peer
->t_process_packet
);
271 * Flush peer output buffer.
273 * This function pops packets off of peer->obuf and writes them to peer->fd.
274 * The amount of packets written is equal to the minimum of peer->wpkt_quanta
275 * and the number of packets on the output buffer, unless an error occurs.
277 * If write() returns an error, the appropriate FSM event is generated.
279 * The return value is equal to the number of packets written
280 * (which may be zero).
282 static uint16_t bgp_write(struct peer
*peer
)
286 int update_last_write
= 0;
290 uint32_t wpkt_quanta_old
;
296 unsigned int total_written
;
298 wpkt_quanta_old
= atomic_load_explicit(&peer
->bgp
->wpkt_quanta
,
299 memory_order_relaxed
);
300 struct stream
*ostreams
[wpkt_quanta_old
];
301 struct stream
**streams
= ostreams
;
302 struct iovec iov
[wpkt_quanta_old
];
304 s
= stream_fifo_head(peer
->obuf
);
310 while (count
< wpkt_quanta_old
&& iovsz
< array_size(iov
) && s
) {
312 iov
[iovsz
].iov_base
= stream_pnt(s
);
313 iov
[iovsz
].iov_len
= STREAM_READABLE(s
);
314 writenum
+= STREAM_READABLE(s
);
324 num
= writev(peer
->fd
, iov
, iovsz
);
327 if (!ERRNO_IO_RETRY(errno
)) {
328 BGP_EVENT_ADD(peer
, TCP_fatal_error
);
329 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
331 SET_FLAG(status
, BGP_IO_TRANS_ERR
);
335 } else if (num
!= writenum
) {
336 unsigned int msg_written
= 0;
337 unsigned int ic
= iovsz
;
339 for (unsigned int i
= 0; i
< ic
; i
++) {
340 size_t ss
= iov
[i
].iov_len
;
342 if (ss
> (unsigned int) num
)
351 total_written
+= msg_written
;
353 assert(total_written
< count
);
355 memmove(&iov
, &iov
[msg_written
],
356 sizeof(iov
[0]) * iovsz
);
357 streams
= &streams
[msg_written
];
358 stream_forward_getp(streams
[0], num
);
359 iov
[0].iov_base
= stream_pnt(streams
[0]);
360 iov
[0].iov_len
= STREAM_READABLE(streams
[0]);
364 assert(writenum
> 0);
366 total_written
= strmsz
;
369 } while (num
!= writenum
);
371 /* Handle statistics */
372 for (unsigned int i
= 0; i
< total_written
; i
++) {
373 s
= stream_fifo_pop(peer
->obuf
);
375 assert(s
== ostreams
[i
]);
377 /* Retrieve BGP packet type. */
378 stream_set_getp(s
, BGP_MARKER_SIZE
+ 2);
379 type
= stream_getc(s
);
383 atomic_fetch_add_explicit(&peer
->open_out
, 1,
384 memory_order_relaxed
);
387 atomic_fetch_add_explicit(&peer
->update_out
, 1,
388 memory_order_relaxed
);
392 atomic_fetch_add_explicit(&peer
->notify_out
, 1,
393 memory_order_relaxed
);
394 /* Double start timer. */
397 /* Overflow check. */
398 if (peer
->v_start
>= (60 * 2))
399 peer
->v_start
= (60 * 2);
402 * Handle Graceful Restart case where the state changes
403 * to Connect instead of Idle.
405 BGP_EVENT_ADD(peer
, BGP_Stop
);
408 case BGP_MSG_KEEPALIVE
:
409 atomic_fetch_add_explicit(&peer
->keepalive_out
, 1,
410 memory_order_relaxed
);
412 case BGP_MSG_ROUTE_REFRESH_NEW
:
413 case BGP_MSG_ROUTE_REFRESH_OLD
:
414 atomic_fetch_add_explicit(&peer
->refresh_out
, 1,
415 memory_order_relaxed
);
417 case BGP_MSG_CAPABILITY
:
418 atomic_fetch_add_explicit(&peer
->dynamic_cap_out
, 1,
419 memory_order_relaxed
);
425 update_last_write
= 1;
430 * Update last_update if UPDATEs were written.
431 * Note: that these are only updated at end,
432 * not per message (i.e., per loop)
435 atomic_store_explicit(&peer
->last_update
, bgp_clock(),
436 memory_order_relaxed
);
438 /* If we TXed any flavor of packet */
439 if (update_last_write
)
440 atomic_store_explicit(&peer
->last_write
, bgp_clock(),
441 memory_order_relaxed
);
448 * Reads a chunk of data from peer->fd into peer->ibuf_work.
450 * @return status flag (see top-of-file)
452 static uint16_t bgp_read(struct peer
*peer
)
454 size_t readsize
; // how many bytes we want to read
455 ssize_t nbytes
; // how many bytes we actually read
457 static uint8_t ibw
[BGP_MAX_PACKET_SIZE
* BGP_READ_PACKET_MAX
];
459 readsize
= MIN(ringbuf_space(peer
->ibuf_work
), sizeof(ibw
));
460 nbytes
= read(peer
->fd
, ibw
, readsize
);
462 /* EAGAIN or EWOULDBLOCK; come back later */
463 if (nbytes
< 0 && ERRNO_IO_RETRY(errno
)) {
464 SET_FLAG(status
, BGP_IO_TRANS_ERR
);
465 /* Fatal error; tear down session */
466 } else if (nbytes
< 0) {
467 flog_err(EC_BGP_UPDATE_RCV
,
468 "%s [Error] bgp_read_packet error: %s", peer
->host
,
469 safe_strerror(errno
));
471 if (peer
->status
== Established
) {
472 if ((CHECK_FLAG(peer
->flags
, PEER_FLAG_GRACEFUL_RESTART
)
473 || CHECK_FLAG(peer
->flags
,
474 PEER_FLAG_GRACEFUL_RESTART_HELPER
))
475 && CHECK_FLAG(peer
->sflags
, PEER_STATUS_NSF_MODE
)) {
476 peer
->last_reset
= PEER_DOWN_NSF_CLOSE_SESSION
;
477 SET_FLAG(peer
->sflags
, PEER_STATUS_NSF_WAIT
);
479 peer
->last_reset
= PEER_DOWN_CLOSE_SESSION
;
482 BGP_EVENT_ADD(peer
, TCP_fatal_error
);
483 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
484 /* Received EOF / TCP session closed */
485 } else if (nbytes
== 0) {
486 if (bgp_debug_neighbor_events(peer
))
487 zlog_debug("%s [Event] BGP connection closed fd %d",
488 peer
->host
, peer
->fd
);
490 if (peer
->status
== Established
) {
491 if ((CHECK_FLAG(peer
->flags
, PEER_FLAG_GRACEFUL_RESTART
)
492 || CHECK_FLAG(peer
->flags
,
493 PEER_FLAG_GRACEFUL_RESTART_HELPER
))
494 && CHECK_FLAG(peer
->sflags
, PEER_STATUS_NSF_MODE
)) {
495 peer
->last_reset
= PEER_DOWN_NSF_CLOSE_SESSION
;
496 SET_FLAG(peer
->sflags
, PEER_STATUS_NSF_WAIT
);
498 peer
->last_reset
= PEER_DOWN_CLOSE_SESSION
;
501 BGP_EVENT_ADD(peer
, TCP_connection_closed
);
502 SET_FLAG(status
, BGP_IO_FATAL_ERR
);
504 assert(ringbuf_put(peer
->ibuf_work
, ibw
, nbytes
)
512 * Called after we have read a BGP packet header. Validates marker, message
513 * type and packet length. If any of these aren't correct, sends a notify.
515 * Assumes that there are at least BGP_HEADER_SIZE readable bytes in the input
518 static bool validate_header(struct peer
*peer
)
522 struct ringbuf
*pkt
= peer
->ibuf_work
;
524 static const uint8_t m_correct
[BGP_MARKER_SIZE
] = {
525 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
526 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
527 uint8_t m_rx
[BGP_MARKER_SIZE
] = {0x00};
529 if (ringbuf_peek(pkt
, 0, m_rx
, BGP_MARKER_SIZE
) != BGP_MARKER_SIZE
)
532 if (memcmp(m_correct
, m_rx
, BGP_MARKER_SIZE
) != 0) {
533 bgp_notify_send(peer
, BGP_NOTIFY_HEADER_ERR
,
534 BGP_NOTIFY_HEADER_NOT_SYNC
);
538 /* Get size and type in network byte order. */
539 ringbuf_peek(pkt
, BGP_MARKER_SIZE
, &size
, sizeof(size
));
540 ringbuf_peek(pkt
, BGP_MARKER_SIZE
+ 2, &type
, sizeof(type
));
544 /* BGP type check. */
545 if (type
!= BGP_MSG_OPEN
&& type
!= BGP_MSG_UPDATE
546 && type
!= BGP_MSG_NOTIFY
&& type
!= BGP_MSG_KEEPALIVE
547 && type
!= BGP_MSG_ROUTE_REFRESH_NEW
548 && type
!= BGP_MSG_ROUTE_REFRESH_OLD
549 && type
!= BGP_MSG_CAPABILITY
) {
550 if (bgp_debug_neighbor_events(peer
))
551 zlog_debug("%s unknown message type 0x%02x", peer
->host
,
554 bgp_notify_send_with_data(peer
, BGP_NOTIFY_HEADER_ERR
,
555 BGP_NOTIFY_HEADER_BAD_MESTYPE
, &type
,
560 /* Minimum packet length check. */
561 if ((size
< BGP_HEADER_SIZE
) || (size
> BGP_MAX_PACKET_SIZE
)
562 || (type
== BGP_MSG_OPEN
&& size
< BGP_MSG_OPEN_MIN_SIZE
)
563 || (type
== BGP_MSG_UPDATE
&& size
< BGP_MSG_UPDATE_MIN_SIZE
)
564 || (type
== BGP_MSG_NOTIFY
&& size
< BGP_MSG_NOTIFY_MIN_SIZE
)
565 || (type
== BGP_MSG_KEEPALIVE
&& size
!= BGP_MSG_KEEPALIVE_MIN_SIZE
)
566 || (type
== BGP_MSG_ROUTE_REFRESH_NEW
567 && size
< BGP_MSG_ROUTE_REFRESH_MIN_SIZE
)
568 || (type
== BGP_MSG_ROUTE_REFRESH_OLD
569 && size
< BGP_MSG_ROUTE_REFRESH_MIN_SIZE
)
570 || (type
== BGP_MSG_CAPABILITY
571 && size
< BGP_MSG_CAPABILITY_MIN_SIZE
)) {
572 if (bgp_debug_neighbor_events(peer
)) {
573 zlog_debug("%s bad message length - %d for %s",
575 type
== 128 ? "ROUTE-REFRESH"
576 : bgp_type_str
[(int)type
]);
579 uint16_t nsize
= htons(size
);
581 bgp_notify_send_with_data(peer
, BGP_NOTIFY_HEADER_ERR
,
582 BGP_NOTIFY_HEADER_BAD_MESLEN
,
583 (unsigned char *)&nsize
, 2);