2 utcp.c -- Userspace TCP
3 Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include "utcp_priv.h"
47 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
48 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
50 #define UTCP_CLOCK CLOCK_MONOTONIC
54 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
55 r->tv_sec = a->tv_sec - b->tv_sec;
56 r->tv_nsec = a->tv_nsec - b->tv_nsec;
59 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
63 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
64 int64_t diff = (a->tv_sec - b->tv_sec) * 1000000000 + a->tv_sec - b->tv_sec;
68 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
69 if(a->tv_sec == b->tv_sec) {
70 return a->tv_nsec < b->tv_nsec;
72 return a->tv_sec < b->tv_sec;
76 static void timespec_clear(struct timespec *a) {
80 static bool timespec_isset(const struct timespec *a) {
84 static long CLOCK_GRANULARITY; // usec
86 static inline size_t min(size_t a, size_t b) {
90 static inline size_t max(size_t a, size_t b) {
97 #ifndef UTCP_DEBUG_DATALEN
98 #define UTCP_DEBUG_DATALEN 20
101 static void debug(struct utcp_connection *c, const char *format, ...) {
106 clock_gettime(CLOCK_REALTIME, &tv);
107 len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
109 va_start(ap, format);
110 len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
113 if(len > 0 && (size_t)len < sizeof(buf)) {
114 fwrite(buf, len, 1, stderr);
118 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
121 if(len < sizeof(hdr)) {
122 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
126 memcpy(&hdr, pkt, sizeof(hdr));
130 if(len > sizeof(hdr)) {
131 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
137 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
138 char str[datalen * 2 + 1];
141 for(uint32_t i = 0; i < datalen; i++) {
142 *p++ = "0123456789ABCDEF"[data[i] >> 4];
143 *p++ = "0123456789ABCDEF"[data[i] & 15];
148 debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s data %s\n",
149 dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
150 hdr.ctl & SYN ? "SYN" : "",
151 hdr.ctl & RST ? "RST" : "",
152 hdr.ctl & FIN ? "FIN" : "",
153 hdr.ctl & ACK ? "ACK" : "",
158 static void debug_cwnd(struct utcp_connection *c) {
159 debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
162 #define debug(...) do {} while(0)
163 #define print_packet(...) do {} while(0)
164 #define debug_cwnd(...) do {} while(0)
167 static void set_state(struct utcp_connection *c, enum state state) {
170 if(state == ESTABLISHED) {
171 timespec_clear(&c->conn_timeout);
174 debug(c, "state %s\n", strstate[state]);
177 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
178 if(seq != c->snd.last) {
193 static bool is_reliable(struct utcp_connection *c) {
194 return c->flags & UTCP_RELIABLE;
197 static int32_t seqdiff(uint32_t a, uint32_t b) {
202 static bool buffer_wraps(struct buffer *buf) {
203 return buf->size - buf->offset < buf->used;
206 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
207 char *newdata = realloc(buf->data, newsize);
215 if(buffer_wraps(buf)) {
216 // Shift the right part of the buffer until it hits the end of the new buffer.
220 // [345.........|........012]
221 uint32_t tailsize = buf->size - buf->offset;
222 uint32_t newoffset = newsize - tailsize;
223 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
224 buf->offset = newoffset;
231 // Store data into the buffer
232 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
233 debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
235 // Ensure we don't store more than maxsize bytes in total
236 size_t required = offset + len;
238 if(required > buf->maxsize) {
239 if(offset >= buf->maxsize) {
243 len = buf->maxsize - offset;
244 required = buf->maxsize;
247 // Check if we need to resize the buffer
248 if(required > buf->size) {
249 size_t newsize = buf->size;
257 } while(newsize < required);
259 if(newsize > buf->maxsize) {
260 newsize = buf->maxsize;
263 if(!buffer_resize(buf, newsize)) {
268 uint32_t realoffset = buf->offset + offset;
270 if(buf->size - buf->offset < offset) {
271 // The offset wrapped
272 realoffset -= buf->size;
275 if(buf->size - realoffset < len) {
276 // The new chunk of data must be wrapped
277 memcpy(buf->data + realoffset, data, buf->size - realoffset);
278 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
280 memcpy(buf->data + realoffset, data, len);
283 if(required > buf->used) {
284 buf->used = required;
290 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
291 return buffer_put_at(buf, buf->used, data, len);
294 // Copy data from the buffer without removing it.
295 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
296 // Ensure we don't copy more than is actually stored in the buffer
297 if(offset >= buf->used) {
301 if(buf->used - offset < len) {
302 len = buf->used - offset;
305 uint32_t realoffset = buf->offset + offset;
307 if(buf->size - buf->offset < offset) {
308 // The offset wrapped
309 realoffset -= buf->size;
312 if(buf->size - realoffset < len) {
313 // The data is wrapped
314 memcpy(data, buf->data + realoffset, buf->size - realoffset);
315 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
317 memcpy(data, buf->data + realoffset, len);
323 // Copy data from the buffer without removing it.
324 static ssize_t buffer_call(struct buffer *buf, utcp_recv_t cb, void *arg, size_t offset, size_t len) {
325 // Ensure we don't copy more than is actually stored in the buffer
326 if(offset >= buf->used) {
330 if(buf->used - offset < len) {
331 len = buf->used - offset;
334 uint32_t realoffset = buf->offset + offset;
336 if(buf->size - buf->offset < offset) {
337 // The offset wrapped
338 realoffset -= buf->size;
341 if(buf->size - realoffset < len) {
342 // The data is wrapped
343 ssize_t rx1 = cb(arg, buf->data + realoffset, buf->size - realoffset);
345 if(rx1 < buf->size - realoffset) {
349 ssize_t rx2 = cb(arg, buf->data, len - (buf->size - realoffset));
357 return cb(arg, buf->data + realoffset, len);
361 // Discard data from the buffer.
362 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
363 if(buf->used < len) {
367 if(buf->size - buf->offset < len) {
368 buf->offset -= buf->size;
377 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
378 if(maxsize < minsize) {
382 buf->maxsize = maxsize;
384 return buf->size >= minsize || buffer_resize(buf, minsize);
387 static void buffer_exit(struct buffer *buf) {
389 memset(buf, 0, sizeof(*buf));
392 static uint32_t buffer_free(const struct buffer *buf) {
393 return buf->maxsize - buf->used;
396 // Connections are stored in a sorted list.
397 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
399 static int compare(const void *va, const void *vb) {
402 const struct utcp_connection *a = *(struct utcp_connection **)va;
403 const struct utcp_connection *b = *(struct utcp_connection **)vb;
406 assert(a->src && b->src);
408 int c = (int)a->src - (int)b->src;
414 c = (int)a->dst - (int)b->dst;
418 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
419 if(!utcp->nconnections) {
423 struct utcp_connection key = {
427 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
428 return match ? *match : NULL;
431 static void free_connection(struct utcp_connection *c) {
432 struct utcp *utcp = c->utcp;
433 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
437 int i = cp - utcp->connections;
438 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
439 utcp->nconnections--;
441 buffer_exit(&c->rcvbuf);
442 buffer_exit(&c->sndbuf);
446 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
447 // Check whether this combination of src and dst is free
450 if(find_connection(utcp, src, dst)) {
454 } else { // If src == 0, generate a random port number with the high bit set
455 if(utcp->nconnections >= 32767) {
460 src = rand() | 0x8000;
462 while(find_connection(utcp, src, dst)) {
467 // Allocate memory for the new connection
469 if(utcp->nconnections >= utcp->nallocated) {
470 if(!utcp->nallocated) {
471 utcp->nallocated = 4;
473 utcp->nallocated *= 2;
476 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
482 utcp->connections = new_array;
485 struct utcp_connection *c = calloc(1, sizeof(*c));
491 if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
496 if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
497 buffer_exit(&c->sndbuf);
502 // Fill in the details
511 c->snd.una = c->snd.iss;
512 c->snd.nxt = c->snd.iss + 1;
513 c->snd.last = c->snd.nxt;
514 c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
515 c->snd.ssthresh = ~0;
519 // Add it to the sorted list of connections
521 utcp->connections[utcp->nconnections++] = c;
522 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
527 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
535 // Update RTT variables. See RFC 6298.
536 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
538 debug(c, "invalid rtt\n");
542 struct utcp *utcp = c->utcp;
546 utcp->rttvar = rtt / 2;
548 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
549 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
552 utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
554 if(utcp->rto > MAX_RTO) {
558 debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
561 static void start_retransmit_timer(struct utcp_connection *c) {
562 clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
564 uint32_t rto = c->utcp->rto;
566 while(rto > USEC_PER_SEC) {
567 c->rtrx_timeout.tv_sec++;
571 c->rtrx_timeout.tv_nsec += c->utcp->rto * 1000;
573 if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
574 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
575 c->rtrx_timeout.tv_sec++;
578 debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
581 static void stop_retransmit_timer(struct utcp_connection *c) {
582 timespec_clear(&c->rtrx_timeout);
583 debug(c, "rtrx_timeout cleared\n");
586 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
587 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
593 assert((flags & ~0x1f) == 0);
604 pkt.hdr.src = c->src;
605 pkt.hdr.dst = c->dst;
606 pkt.hdr.seq = c->snd.iss;
608 pkt.hdr.wnd = c->rcvbuf.maxsize;
610 pkt.hdr.aux = 0x0101;
614 pkt.init[3] = flags & 0x7;
616 set_state(c, SYN_SENT);
618 print_packet(c, "send", &pkt, sizeof(pkt));
619 utcp->send(utcp, &pkt, sizeof(pkt));
621 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
622 c->conn_timeout.tv_sec += utcp->timeout;
624 start_retransmit_timer(c);
629 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
630 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
633 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
634 if(c->reapable || c->state != SYN_RECEIVED) {
635 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
639 debug(c, "accepted %p %p\n", c, recv, priv);
642 set_state(c, ESTABLISHED);
645 static void ack(struct utcp_connection *c, bool sendatleastone) {
646 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
647 int32_t cwndleft = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
653 } else if(cwndleft < left) {
656 if(!sendatleastone || cwndleft > c->utcp->mss) {
657 left -= left % c->utcp->mss;
661 debug(c, "cwndleft %d left %d\n", cwndleft, left);
663 if(!left && !sendatleastone) {
670 } *pkt = c->utcp->pkt;
672 pkt->hdr.src = c->src;
673 pkt->hdr.dst = c->dst;
674 pkt->hdr.ack = c->rcv.nxt;
675 pkt->hdr.wnd = c->rcvbuf.maxsize;
680 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
681 pkt->hdr.seq = c->snd.nxt;
683 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
685 c->snd.nxt += seglen;
688 if(seglen && fin_wanted(c, c->snd.nxt)) {
693 if(!c->rtt_start.tv_sec) {
694 // Start RTT measurement
695 clock_gettime(UTCP_CLOCK, &c->rtt_start);
696 c->rtt_seq = pkt->hdr.seq + seglen;
697 debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
700 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
701 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
705 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
707 debug(c, "send() called on closed connection\n");
715 debug(c, "send() called on unconnected connection\n");
730 debug(c, "send() called on closed connection\n");
735 // Exit early if we have nothing to send.
746 // Check if we need to be able to buffer all data
748 if(c->flags & UTCP_NO_PARTIAL) {
749 if(len > buffer_free(&c->sndbuf)) {
750 if(len > c->sndbuf.maxsize) {
760 // Add data to send buffer.
762 if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
763 len = buffer_put(&c->sndbuf, data, len);
779 // Don't send anything yet if the connection has not fully established yet
781 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
787 if(!is_reliable(c)) {
788 c->snd.una = c->snd.nxt = c->snd.last;
789 buffer_discard(&c->sndbuf, c->sndbuf.used);
793 if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
794 start_retransmit_timer(c);
797 if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
798 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
799 c->conn_timeout.tv_sec += c->utcp->timeout;
805 static void swap_ports(struct hdr *hdr) {
806 uint16_t tmp = hdr->src;
811 static void fast_retransmit(struct utcp_connection *c) {
812 if(c->state == CLOSED || c->snd.last == c->snd.una) {
813 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
817 struct utcp *utcp = c->utcp;
824 pkt = malloc(c->utcp->mtu);
830 pkt->hdr.src = c->src;
831 pkt->hdr.dst = c->dst;
832 pkt->hdr.wnd = c->rcvbuf.maxsize;
841 // Send unacked data again.
842 pkt->hdr.seq = c->snd.una;
843 pkt->hdr.ack = c->rcv.nxt;
845 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
847 if(fin_wanted(c, c->snd.una + len)) {
852 buffer_copy(&c->sndbuf, pkt->data, 0, len);
853 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
854 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
864 static void retransmit(struct utcp_connection *c) {
865 if(c->state == CLOSED || c->snd.last == c->snd.una) {
866 debug(c, "retransmit() called but nothing to retransmit!\n");
867 stop_retransmit_timer(c);
871 struct utcp *utcp = c->utcp;
876 } *pkt = c->utcp->pkt;
878 pkt->hdr.src = c->src;
879 pkt->hdr.dst = c->dst;
880 pkt->hdr.wnd = c->rcvbuf.maxsize;
885 // Send our SYN again
886 pkt->hdr.seq = c->snd.iss;
889 pkt->hdr.aux = 0x0101;
893 pkt->data[3] = c->flags & 0x7;
894 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
895 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
900 pkt->hdr.seq = c->snd.nxt;
901 pkt->hdr.ack = c->rcv.nxt;
902 pkt->hdr.ctl = SYN | ACK;
903 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
904 utcp->send(utcp, pkt, sizeof(pkt->hdr));
912 // Send unacked data again.
913 pkt->hdr.seq = c->snd.una;
914 pkt->hdr.ack = c->rcv.nxt;
916 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
918 if(fin_wanted(c, c->snd.una + len)) {
923 // RFC 5681 slow start after timeout
924 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
925 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
926 c->snd.cwnd = utcp->mss;
929 buffer_copy(&c->sndbuf, pkt->data, 0, len);
930 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
931 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
933 c->snd.nxt = c->snd.una + len;
940 // We shouldn't need to retransmit anything in this state.
944 stop_retransmit_timer(c);
948 start_retransmit_timer(c);
951 if(utcp->rto > MAX_RTO) {
955 c->rtt_start.tv_sec = 0; // invalidate RTT timer
956 c->dupack = 0; // cancel any ongoing fast recovery
962 /* Update receive buffer and SACK entries after consuming data.
966 * |.....0000..1111111111.....22222......3333|
969 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
970 * to remove data from the receive buffer. The idea is to substract "len"
971 * from the offset of all the SACK entries, and then remove/cut down entries
972 * that are shifted to before the start of the receive buffer.
974 * There are three cases:
975 * - the SACK entry is after ^, in that case just change the offset.
976 * - the SACK entry starts before and ends after ^, so we have to
977 * change both its offset and size.
978 * - the SACK entry is completely before ^, in that case delete it.
980 static void sack_consume(struct utcp_connection *c, size_t len) {
981 debug(c, "sack_consume %lu\n", (unsigned long)len);
983 if(len > c->rcvbuf.used) {
984 debug(c, "all SACK entries consumed\n");
989 buffer_discard(&c->rcvbuf, len);
991 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
992 if(len < c->sacks[i].offset) {
993 c->sacks[i].offset -= len;
995 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
996 c->sacks[i].len -= len - c->sacks[i].offset;
997 c->sacks[i].offset = 0;
1000 if(i < NSACKS - 1) {
1001 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1002 c->sacks[NSACKS - 1].len = 0;
1004 c->sacks[i].len = 0;
1010 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1011 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1015 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1016 debug(c, "out of order packet, offset %u\n", offset);
1017 // Packet loss or reordering occured. Store the data in the buffer.
1018 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1020 if(rxd < 0 || (size_t)rxd < len) {
1024 // Make note of where we put it.
1025 for(int i = 0; i < NSACKS; i++) {
1026 if(!c->sacks[i].len) { // nothing to merge, add new entry
1027 debug(c, "new SACK entry %d\n", i);
1028 c->sacks[i].offset = offset;
1029 c->sacks[i].len = rxd;
1031 } else if(offset < c->sacks[i].offset) {
1032 if(offset + rxd < c->sacks[i].offset) { // insert before
1033 if(!c->sacks[NSACKS - 1].len) { // only if room left
1034 debug(c, "insert SACK entry at %d\n", i);
1035 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1036 c->sacks[i].offset = offset;
1037 c->sacks[i].len = rxd;
1039 debug(c, "SACK entries full, dropping packet\n");
1044 debug(c, "merge with start of SACK entry at %d\n", i);
1045 c->sacks[i].offset = offset;
1048 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1049 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1050 debug(c, "merge with end of SACK entry at %d\n", i);
1051 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1052 // TODO: handle potential merge with next entry
1059 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1060 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1064 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1066 ssize_t rxd = c->recv(c, data, len);
1068 if(rxd != (ssize_t)len) {
1069 // TODO: handle the application not accepting all data.
1074 // Check if we can process out-of-order data now.
1075 if(c->sacks[0].len && len >= c->sacks[0].offset) {
1076 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1078 if(len < c->sacks[0].offset + c->sacks[0].len) {
1079 size_t offset = len;
1080 len = c->sacks[0].offset + c->sacks[0].len;
1081 size_t remainder = len - offset;
1082 ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
1084 if(rxd != (ssize_t)remainder) {
1085 // TODO: handle the application not accepting all data.
1091 if(c->rcvbuf.used) {
1092 sack_consume(c, len);
1099 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
1100 if(!is_reliable(c)) {
1101 c->recv(c, data, len);
1102 c->rcv.nxt = seq + len;
1106 uint32_t offset = seqdiff(seq, c->rcv.nxt);
1108 if(offset + len > c->rcvbuf.maxsize) {
1113 handle_out_of_order(c, offset, data, len);
1115 handle_in_order(c, data, len);
1120 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1121 const uint8_t *ptr = data;
1137 // Drop packets smaller than the header
1141 if(len < sizeof(hdr)) {
1142 print_packet(NULL, "recv", data, len);
1147 // Make a copy from the potentially unaligned data to a struct hdr
1149 memcpy(&hdr, ptr, sizeof(hdr));
1151 // Try to match the packet to an existing connection
1153 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1154 print_packet(c, "recv", data, len);
1156 // Process the header
1161 // Drop packets with an unknown CTL flag
1163 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
1164 print_packet(NULL, "recv", data, len);
1169 // Check for auxiliary headers
1171 const uint8_t *init = NULL;
1173 uint16_t aux = hdr.aux;
1176 size_t auxlen = 4 * (aux >> 8) & 0xf;
1177 uint8_t auxtype = aux & 0xff;
1186 if(!(hdr.ctl & SYN) || auxlen != 4) {
1202 if(!(aux & 0x800)) {
1211 memcpy(&aux, ptr, 2);
1216 bool has_data = len || (hdr.ctl & (SYN | FIN));
1218 // Is it for a new connection?
1221 // Ignore RST packets
1227 // Is it a SYN packet and are we LISTENing?
1229 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1230 // If we don't want to accept it, send a RST back
1231 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1236 // Try to allocate memory, otherwise send a RST back
1237 c = allocate_connection(utcp, hdr.dst, hdr.src);
1244 // Parse auxilliary information
1251 c->flags = init[3] & 0x7;
1253 c->flags = UTCP_TCP;
1257 // Return SYN+ACK, go to SYN_RECEIVED state
1258 c->snd.wnd = hdr.wnd;
1259 c->rcv.irs = hdr.seq;
1260 c->rcv.nxt = c->rcv.irs + 1;
1261 set_state(c, SYN_RECEIVED);
1268 pkt.hdr.src = c->src;
1269 pkt.hdr.dst = c->dst;
1270 pkt.hdr.ack = c->rcv.irs + 1;
1271 pkt.hdr.seq = c->snd.iss;
1272 pkt.hdr.wnd = c->rcvbuf.maxsize;
1273 pkt.hdr.ctl = SYN | ACK;
1276 pkt.hdr.aux = 0x0101;
1280 pkt.data[3] = c->flags & 0x7;
1281 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1282 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1285 print_packet(c, "send", &pkt, sizeof(hdr));
1286 utcp->send(utcp, &pkt, sizeof(hdr));
1289 // No, we don't want your packets, send a RST back
1297 debug(c, "state %s\n", strstate[c->state]);
1299 // In case this is for a CLOSED connection, ignore the packet.
1300 // TODO: make it so incoming packets can never match a CLOSED connection.
1302 if(c->state == CLOSED) {
1303 debug(c, "got packet for closed connection\n");
1307 // It is for an existing connection.
1309 // 1. Drop invalid packets.
1311 // 1a. Drop packets that should not happen in our current state.
1332 // 1b. Discard data that is not in our receive window.
1334 if(is_reliable(c)) {
1337 if(c->state == SYN_SENT) {
1339 } else if(len == 0) {
1340 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1342 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1344 // cut already accepted front overlapping
1345 if(rcv_offset < 0) {
1346 acceptable = len > (size_t) - rcv_offset;
1351 hdr.seq -= rcv_offset;
1354 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1359 debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1361 // Ignore unacceptable RST packets.
1366 // Otherwise, continue processing.
1371 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1374 debug(c, "packet out of order, offset %u bytes", rcv_offset);
1377 if(rcv_offset >= 0) {
1378 c->rcv.nxt = hdr.seq + len;
1384 c->snd.wnd = hdr.wnd; // TODO: move below
1386 // 1c. Drop packets with an invalid ACK.
1387 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1388 // (= snd.una + c->sndbuf.used).
1390 if(!is_reliable(c)) {
1391 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1392 hdr.ack = c->snd.una;
1396 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1397 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1399 // Ignore unacceptable RST packets.
1407 // 2. Handle RST packets
1412 if(!(hdr.ctl & ACK)) {
1416 // The peer has refused our connection.
1417 set_state(c, CLOSED);
1418 errno = ECONNREFUSED;
1421 c->recv(c, NULL, 0);
1424 if(c->poll && !c->reapable) {
1435 // We haven't told the application about this connection yet. Silently delete.
1447 // The peer has aborted our connection.
1448 set_state(c, CLOSED);
1452 c->recv(c, NULL, 0);
1455 if(c->poll && !c->reapable) {
1468 // As far as the application is concerned, the connection has already been closed.
1469 // If it has called utcp_close() already, we can immediately free this connection.
1475 // Otherwise, immediately move to the CLOSED state.
1476 set_state(c, CLOSED);
1489 if(!(hdr.ctl & ACK)) {
1494 // 3. Advance snd.una
1496 advanced = seqdiff(hdr.ack, c->snd.una);
1500 if(c->rtt_start.tv_sec) {
1501 if(c->rtt_seq == hdr.ack) {
1502 struct timespec now;
1503 clock_gettime(UTCP_CLOCK, &now);
1504 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1505 update_rtt(c, diff);
1506 c->rtt_start.tv_sec = 0;
1507 } else if(c->rtt_seq < hdr.ack) {
1508 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1509 c->rtt_start.tv_sec = 0;
1513 int32_t data_acked = advanced;
1521 // TODO: handle FIN as well.
1526 assert(data_acked >= 0);
1529 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1530 assert(data_acked <= bufused);
1534 buffer_discard(&c->sndbuf, data_acked);
1538 // Also advance snd.nxt if possible
1539 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1540 c->snd.nxt = hdr.ack;
1543 c->snd.una = hdr.ack;
1546 if(c->dupack >= 3) {
1547 debug(c, "fast recovery ended\n");
1548 c->snd.cwnd = c->snd.ssthresh;
1554 // Increase the congestion window according to RFC 5681
1555 if(c->snd.cwnd < c->snd.ssthresh) {
1556 c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1558 c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1561 if(c->snd.cwnd > c->sndbuf.maxsize) {
1562 c->snd.cwnd = c->sndbuf.maxsize;
1567 // Check if we have sent a FIN that is now ACKed.
1570 if(c->snd.una == c->snd.last) {
1571 set_state(c, FIN_WAIT_2);
1577 if(c->snd.una == c->snd.last) {
1578 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1579 c->conn_timeout.tv_sec += utcp->timeout;
1580 set_state(c, TIME_WAIT);
1589 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1591 debug(c, "duplicate ACK %d\n", c->dupack);
1593 if(c->dupack == 3) {
1594 // RFC 5681 fast recovery
1595 debug(c, "fast recovery started\n", c->dupack);
1596 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1597 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1598 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1600 if(c->snd.cwnd > c->sndbuf.maxsize) {
1601 c->snd.cwnd = c->sndbuf.maxsize;
1607 } else if(c->dupack > 3) {
1608 c->snd.cwnd += utcp->mss;
1610 if(c->snd.cwnd > c->sndbuf.maxsize) {
1611 c->snd.cwnd = c->sndbuf.maxsize;
1617 // We got an ACK which indicates the other side did get one of our packets.
1618 // Reset the retransmission timer to avoid going to slow start,
1619 // but don't touch the connection timeout.
1620 start_retransmit_timer(c);
1627 if(c->snd.una == c->snd.last) {
1628 stop_retransmit_timer(c);
1629 timespec_clear(&c->conn_timeout);
1630 } else if(is_reliable(c)) {
1631 start_retransmit_timer(c);
1632 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1633 c->conn_timeout.tv_sec += utcp->timeout;
1638 // 5. Process SYN stuff
1644 // This is a SYNACK. It should always have ACKed the SYN.
1649 c->rcv.irs = hdr.seq;
1650 c->rcv.nxt = hdr.seq;
1654 set_state(c, FIN_WAIT_1);
1656 set_state(c, ESTABLISHED);
1659 // TODO: notify application of this somehow.
1663 // This is a retransmit of a SYN, send back the SYNACK.
1673 // Ehm, no. We should never receive a second SYN.
1683 // SYN counts as one sequence number
1687 // 6. Process new data
1689 if(c->state == SYN_RECEIVED) {
1690 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1695 // Are we still LISTENing?
1697 utcp->accept(c, c->src);
1700 if(c->state != ESTABLISHED) {
1701 set_state(c, CLOSED);
1711 // This should never happen.
1726 // Ehm no, We should never receive more data after a FIN.
1736 handle_incoming_data(c, hdr.seq, ptr, len);
1739 // 7. Process FIN stuff
1741 if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1745 // This should never happen.
1752 set_state(c, CLOSE_WAIT);
1756 set_state(c, CLOSING);
1760 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1761 c->conn_timeout.tv_sec += utcp->timeout;
1762 set_state(c, TIME_WAIT);
1769 // Ehm, no. We should never receive a second FIN.
1779 // FIN counts as one sequence number
1783 // Inform the application that the peer closed its end of the connection.
1786 c->recv(c, NULL, 0);
1790 // Now we send something back if:
1791 // - we received data, so we have to send back an ACK
1792 // -> sendatleastone = true
1793 // - or we got an ack, so we should maybe send a bit more data
1794 // -> sendatleastone = false
1796 if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1811 hdr.ack = hdr.seq + len;
1813 hdr.ctl = RST | ACK;
1816 print_packet(c, "send", &hdr, sizeof(hdr));
1817 utcp->send(utcp, &hdr, sizeof(hdr));
1822 int utcp_shutdown(struct utcp_connection *c, int dir) {
1823 debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1831 debug(c, "shutdown() called on closed connection\n");
1836 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1841 // TCP does not have a provision for stopping incoming packets.
1842 // The best we can do is to just ignore them.
1843 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1847 // The rest of the code deals with shutting down writes.
1848 if(dir == UTCP_SHUT_RD) {
1852 // Only process shutting down writes once.
1870 set_state(c, FIN_WAIT_1);
1878 set_state(c, CLOSING);
1891 if(!timespec_isset(&c->rtrx_timeout)) {
1892 start_retransmit_timer(c);
1898 static bool reset_connection(struct utcp_connection *c) {
1905 debug(c, "abort() called on closed connection\n");
1922 set_state(c, CLOSED);
1930 set_state(c, CLOSED);
1940 hdr.seq = c->snd.nxt;
1945 print_packet(c, "send", &hdr, sizeof(hdr));
1946 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1950 // Closes all the opened connections
1951 void utcp_abort_all_connections(struct utcp *utcp) {
1957 for(int i = 0; i < utcp->nconnections; i++) {
1958 struct utcp_connection *c = utcp->connections[i];
1960 if(c->reapable || c->state == CLOSED) {
1964 utcp_recv_t old_recv = c->recv;
1965 utcp_poll_t old_poll = c->poll;
1967 reset_connection(c);
1971 old_recv(c, NULL, 0);
1974 if(old_poll && !c->reapable) {
1983 int utcp_close(struct utcp_connection *c) {
1984 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1994 int utcp_abort(struct utcp_connection *c) {
1995 if(!reset_connection(c)) {
2004 * One call to this function will loop through all connections,
2005 * checking if something needs to be resent or not.
2006 * The return value is the time to the next timeout in milliseconds,
2007 * or maybe a negative value if the timeout is infinite.
2009 struct timespec utcp_timeout(struct utcp *utcp) {
2010 struct timespec now;
2011 clock_gettime(UTCP_CLOCK, &now);
2012 struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2014 for(int i = 0; i < utcp->nconnections; i++) {
2015 struct utcp_connection *c = utcp->connections[i];
2021 // delete connections that have been utcp_close()d.
2022 if(c->state == CLOSED) {
2024 debug(c, "reaping\n");
2032 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2037 c->recv(c, NULL, 0);
2040 if(c->poll && !c->reapable) {
2047 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2048 debug(c, "retransmitting after timeout\n");
2053 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2055 uint32_t len = buffer_free(&c->sndbuf);
2060 } else if(c->state == CLOSED) {
2065 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2066 next = c->conn_timeout;
2069 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2070 next = c->rtrx_timeout;
2074 struct timespec diff;
2076 timespec_sub(&next, &now, &diff);
2081 bool utcp_is_active(struct utcp *utcp) {
2086 for(int i = 0; i < utcp->nconnections; i++)
2087 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2094 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2100 struct utcp *utcp = calloc(1, sizeof(*utcp));
2106 if(!CLOCK_GRANULARITY) {
2107 struct timespec res;
2108 clock_getres(UTCP_CLOCK, &res);
2109 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2112 utcp->accept = accept;
2113 utcp->pre_accept = pre_accept;
2116 utcp_set_mtu(utcp, DEFAULT_MTU);
2117 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2118 utcp->rto = START_RTO; // usec
2123 void utcp_exit(struct utcp *utcp) {
2128 for(int i = 0; i < utcp->nconnections; i++) {
2129 struct utcp_connection *c = utcp->connections[i];
2133 c->recv(c, NULL, 0);
2136 if(c->poll && !c->reapable) {
2141 buffer_exit(&c->rcvbuf);
2142 buffer_exit(&c->sndbuf);
2146 free(utcp->connections);
2150 uint16_t utcp_get_mtu(struct utcp *utcp) {
2151 return utcp ? utcp->mtu : 0;
2154 uint16_t utcp_get_mss(struct utcp *utcp) {
2155 return utcp ? utcp->mss : 0;
2158 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2163 if(mtu <= sizeof(struct hdr)) {
2167 if(mtu > utcp->mtu) {
2168 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2178 utcp->mss = mtu - sizeof(struct hdr);
2181 void utcp_reset_timers(struct utcp *utcp) {
2186 struct timespec now, then;
2188 clock_gettime(UTCP_CLOCK, &now);
2192 then.tv_sec += utcp->timeout;
2194 for(int i = 0; i < utcp->nconnections; i++) {
2195 struct utcp_connection *c = utcp->connections[i];
2201 if(timespec_isset(&c->rtrx_timeout)) {
2202 c->rtrx_timeout = now;
2205 if(timespec_isset(&c->conn_timeout)) {
2206 c->conn_timeout = then;
2209 c->rtt_start.tv_sec = 0;
2212 if(utcp->rto > START_RTO) {
2213 utcp->rto = START_RTO;
2217 int utcp_get_user_timeout(struct utcp *u) {
2218 return u ? u->timeout : 0;
2221 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2223 u->timeout = timeout;
2227 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2228 return c ? c->sndbuf.maxsize : 0;
2231 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2241 return buffer_free(&c->sndbuf);
2248 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2253 c->sndbuf.maxsize = size;
2255 if(c->sndbuf.maxsize != size) {
2256 c->sndbuf.maxsize = -1;
2259 c->do_poll = buffer_free(&c->sndbuf);
2262 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2263 return c ? c->rcvbuf.maxsize : 0;
2266 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2267 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2268 return buffer_free(&c->rcvbuf);
2274 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2279 c->rcvbuf.maxsize = size;
2281 if(c->rcvbuf.maxsize != size) {
2282 c->rcvbuf.maxsize = -1;
2286 size_t utcp_get_sendq(struct utcp_connection *c) {
2287 return c->sndbuf.used;
2290 size_t utcp_get_recvq(struct utcp_connection *c) {
2291 return c->rcvbuf.used;
2294 bool utcp_get_nodelay(struct utcp_connection *c) {
2295 return c ? c->nodelay : false;
2298 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2300 c->nodelay = nodelay;
2304 bool utcp_get_keepalive(struct utcp_connection *c) {
2305 return c ? c->keepalive : false;
2308 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2310 c->keepalive = keepalive;
2314 size_t utcp_get_outq(struct utcp_connection *c) {
2315 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2318 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2324 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2327 c->do_poll = buffer_free(&c->sndbuf);
2331 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2333 utcp->accept = accept;
2334 utcp->pre_accept = pre_accept;
2338 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2339 if(!c || c->reapable) {
2343 if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2348 // If we expect data, start the connection timer.
2349 if(!timespec_isset(&c->conn_timeout)) {
2350 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2351 c->conn_timeout.tv_sec += c->utcp->timeout;
2354 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2355 if(c->snd.una == c->snd.last) {
2356 timespec_clear(&c->conn_timeout);
2361 void utcp_offline(struct utcp *utcp, bool offline) {
2362 struct timespec now;
2363 clock_gettime(UTCP_CLOCK, &now);
2365 for(int i = 0; i < utcp->nconnections; i++) {
2366 struct utcp_connection *c = utcp->connections[i];
2372 utcp_expect_data(c, offline);
2375 if(timespec_isset(&c->rtrx_timeout)) {
2376 c->rtrx_timeout = now;
2379 utcp->connections[i]->rtt_start.tv_sec = 0;
2383 if(!offline && utcp->rto > START_RTO) {
2384 utcp->rto = START_RTO;
2388 void utcp_set_clock_granularity(long granularity) {
2389 CLOCK_GRANULARITY = granularity;