From 55f4e4a5ec657a017e3bf75299ad71fd1c968dd3 Mon Sep 17 00:00:00 2001 From: The Android Open Source Project Date: Tue, 21 Oct 2008 07:00:00 -0700 Subject: Initial Contribution --- shaper.c | 593 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 593 insertions(+) create mode 100644 shaper.c (limited to 'shaper.c') diff --git a/shaper.c b/shaper.c new file mode 100644 index 0000000..83e5aea --- /dev/null +++ b/shaper.c @@ -0,0 +1,593 @@ +/* Copyright (C) 2007-2008 The Android Open Source Project +** +** This software is licensed under the terms of the GNU General Public +** License version 2, as published by the Free Software Foundation, and +** may be copied, distributed, and modified under those terms. +** +** This program is distributed in the hope that it will be useful, +** but WITHOUT ANY WARRANTY; without even the implied warranty of +** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +** GNU General Public License for more details. +*/ +#include "shaper.h" +#include "vl.h" +#include + +#define SHAPER_CLOCK rt_clock +#define SHAPER_CLOCK_UNIT 1000. + +static int +_packet_is_internal( const uint8_t* data, size_t size ) +{ + const uint8_t* end = data + size; + + /* must have room for Mac + IP header */ + if (data + 40 > end) + return 0; + + if (data[12] != 0x08 || data[13] != 0x00 ) + return 0; + + /* must have valid IP header */ + data += 14; + if ((data[0] >> 4) != 4 || (data[0] & 15) < 5) + return 0; + + /* internal if both source and dest addresses are in 10.x.x.x */ + return ( data[12] == 10 && data[16] == 10); +} + +/* here's how we implement network shaping. we want to limit the network + * rate to a given constant MAX_RATE expressed as bits/second. this means + * that it takes 1/MAX_RATE seconds to send a single bit, and count*8/MAX_RATE + * seconds to send 'count' bytes. + * + * we're going to implement a scheme where, when we send a packet of + * 'count' bytes, no other packet will go through in the same direction for + * at least 'count*8/MAX_RATE' seconds. any successive packet that is "sent" + * in this interval is placed in a queue, associated to a timer + * + * there are different (queue/timer/rate) values for the input and output + * direction of the user vlan. + */ +typedef struct QueuedPacketRec_ { + int64_t expiration; + struct QueuedPacketRec_* next; + size_t size; + void* opaque; + void* data; +} QueuedPacketRec, *QueuedPacket; + + +static QueuedPacket +queued_packet_create( const void* data, + size_t size, + void* opaque, + int do_copy ) +{ + QueuedPacket packet; + size_t packet_size = sizeof(*packet); + + if (do_copy) + packet_size += size; + + packet = qemu_malloc(packet_size); + packet->next = NULL; + packet->expiration = 0; + packet->size = (size_t)size; + packet->opaque = opaque; + + if (do_copy) { + packet->data = (void*)(packet+1); + memcpy( (char*)packet->data, (char*)data, packet->size ); + } else { + packet->data = (void*)data; + } + return packet; +} + +static void +queued_packet_free( QueuedPacket packet ) +{ + if (packet) { + qemu_free( packet ); + } +} + +typedef struct NetShaperRec_ { + QueuedPacket packets; /* list of queued packets, ordered by expiration date */ + int num_packets; + int active; /* is this shaper active ? */ + int64_t block_until; + double max_rate; /* max rate expressed in bytes/second */ + double inv_rate; /* inverse of max rate */ + QEMUTimer* timer; /* QEMU timer */ + + int do_copy; + NetShaperSendFunc send_func; + +} NetShaperRec; + + +void +netshaper_destroy( NetShaper shaper ) +{ + if (shaper) { + shaper->active = 0; + + while (shaper->packets) { + QueuedPacket packet = shaper->packets; + shaper->packets = packet->next; + packet->next = NULL; + queued_packet_free(packet); + } + + qemu_del_timer(shaper->timer); + qemu_free_timer(shaper->timer); + shaper->timer = NULL; + qemu_free(shaper); + } +} + +/* this function is called when the shaper's timer expires */ +static void +netshaper_expires( NetShaper shaper ) +{ + QueuedPacket packet; + + while ((packet = shaper->packets) != NULL) { + int64_t now = qemu_get_clock( SHAPER_CLOCK ); + + if (packet->expiration > now) + break; + + shaper->packets = packet->next; + shaper->send_func( packet->data, packet->size, packet->opaque ); + queued_packet_free(packet); + shaper->num_packets--; + } + + /* reprogram timer if needed */ + if (shaper->packets) { + shaper->block_until = shaper->packets->expiration; + qemu_mod_timer( shaper->timer, shaper->block_until ); + } else { + shaper->block_until = -1; + } +} + + +NetShaper +netshaper_create( int do_copy, + NetShaperSendFunc send_func ) +{ + NetShaper shaper = qemu_malloc(sizeof(*shaper)); + + init_timers(); + + shaper->active = 0; + shaper->packets = NULL; + shaper->num_packets = 0; + shaper->timer = qemu_new_timer( SHAPER_CLOCK, + (QEMUTimerCB*) netshaper_expires, + shaper ); + shaper->send_func = send_func; + shaper->max_rate = 1e6; + shaper->inv_rate = 0.; + + shaper->block_until = -1; /* magic value, means to not block */ + + return shaper; +} + +void +netshaper_set_rate( NetShaper shaper, + double rate ) +{ + /* send all current packets when changing the rate */ + while (shaper->packets) { + QueuedPacket packet = shaper->packets; + shaper->packets = packet->next; + shaper->send_func(packet->data, packet->size, packet->opaque); + qemu_free(packet); + shaper->num_packets = 0; + } + + shaper->max_rate = rate; + if (rate > 1.) { + shaper->inv_rate = (8.*SHAPER_CLOCK_UNIT)/rate; /* qemu_get_clock returns time in ms */ + shaper->active = 1; /* for the real-time clock */ + } else { + shaper->active = 0; + } + + shaper->block_until = -1; +} + +void +netshaper_send_aux( NetShaper shaper, + void* data, + size_t size, + void* opaque ) +{ + int64_t now; + + if (!shaper->active || _packet_is_internal(data, size)) { + shaper->send_func( data, size, opaque ); + return; + } + + now = qemu_get_clock( SHAPER_CLOCK ); + if (now >= shaper->block_until) { + shaper->send_func( data, size, opaque ); + shaper->block_until = now + size*shaper->inv_rate; + //fprintf(stderr, "NETSHAPER: block for %.2fms\n", (shaper->block_until - now)*1.0 ); + return; + } + + /* create new packet, add it to the queue */ + { + QueuedPacket packet; + + packet = queued_packet_create( data, size, opaque, shaper->do_copy ); + + packet->expiration = shaper->block_until; + + { + QueuedPacket *pnode, node; + + pnode = &shaper->packets; + for (;;) { + node = *pnode; + if (node == NULL || node->expiration > packet->expiration ) + break; + pnode = &node->next; + } + packet->next = *pnode; + *pnode = packet; + + if (packet == shaper->packets) + qemu_mod_timer( shaper->timer, packet->expiration ); + } + shaper->num_packets += 1; + } + shaper->block_until += size*shaper->inv_rate; + //fprintf(stderr, "NETSHAPER: block2 for %.2fms\n", (shaper->block_until - now)*1.0 ); +} + +void +netshaper_send( NetShaper shaper, + void* data, + size_t size ) +{ + netshaper_send_aux(shaper, data, size, NULL); +} + + +int +netshaper_can_send( NetShaper shaper ) +{ + int64_t now; + + if (!shaper->active || shaper->block_until < 0) + return 1; + + if (shaper->packets) + return 0; + + now = qemu_get_clock( SHAPER_CLOCK ); + return (now >= shaper->block_until); +} + + + + + + +/* this type is used to model a session connection/state + * if session->packet is != NULL, then the connection is delayed + */ +typedef struct SessionRec_ { + int64_t expiration; + struct SessionRec_* next; + unsigned src_ip; + unsigned dst_ip; + unsigned short src_port; + unsigned short dst_port; + uint8_t protocol; + QueuedPacket packet; + +} SessionRec, *Session; + +#define _PROTOCOL_TCP 6 +#define _PROTOCOL_UDP 17 + + + +static void +session_free( Session session ) +{ + if (session) { + if (session->packet) { + queued_packet_free(session->packet); + session->packet = NULL; + } + qemu_free( session ); + } +} + + +#if 0 /* useful for debugging */ +static const char* +session_to_string( Session session ) +{ + static char temp[256]; + const char* format = (session->protocol == _PROTOCOL_TCP) ? "TCP" : "UDP"; + sprintf( temp, "%s[%d.%d.%d.%d:%d / %d.%d.%d.%d:%d]", format, + (session->src_ip >> 24) & 255, (session->src_ip >> 16) & 255, + (session->src_ip >> 8) & 255, (session->src_ip) & 255, session->src_port, + (session->dst_ip >> 24) & 255, (session->dst_ip >> 16) & 255, + (session->dst_ip >> 8) & 255, (session->dst_ip) & 255, session->dst_port); + + return temp; +} +#endif + +/* returns TRUE if this corresponds to a SYN packet */ +int +_packet_SYN_flags( const void* _data, size_t size, Session info ) +{ + const uint8_t* data = (const uint8_t*)_data; + const uint8_t* end = data + size; + + /* enough room for a Ethernet MAC packet ? */ + if (data + 14 > end - 4) + return 0; + + /* is it an IP packet ? */ + if (data[12] != 0x8 || data[13] != 0) + return 0; + + data += 14; + end -= 4; + + if (data + 20 > end) + return 0; + + /* IP version must be 4, and the header length in words at least 5 */ + if ((data[0] & 0xF) < 5 || (data[0] >> 4) != 4) + return 0; + + /* time-to-live must be > 0 */ + if (data[8] == 0) + return 0; + + /* must be TCP or UDP packet */ + if (data[9] != _PROTOCOL_TCP && data[9] != _PROTOCOL_UDP) + return 0; + + info->protocol = data[9]; + info->src_ip = (data[12] << 24) | (data[13] << 16) | (data[14] << 8) | data[15]; + info->dst_ip = (data[16] << 24) | (data[17] << 16) | (data[18] << 8) | data[19]; + + data += 4*(data[0] & 15); + if (data + 20 > end) + return 0; + + info->src_port = (unsigned short)((data[0] << 8) | data[1]); + info->dst_port = (unsigned short)((data[2] << 8) | data[3]); + + return (data[13] & 0x1f); +} + + +typedef struct NetDelayRec_ +{ + Session sessions; + int num_sessions; + QEMUTimer* timer; + int active; + int min_ms; + int max_ms; + + NetShaperSendFunc send_func; + +} NetDelayRec; + + +static Session* +netdelay_lookup_session( NetDelay delay, Session info ) +{ + Session* pnode = &delay->sessions; + Session node; + + for (;;) { + node = *pnode; + if (node == NULL) + break; + + if (node->src_ip == info->src_ip && + node->dst_ip == info->dst_ip && + node->src_port == info->src_port && + node->dst_port == info->dst_port && + node->protocol == info->protocol ) + break; + + pnode = &node->next; + } + return pnode; +} + + + +/* called by the delay's timer on expiration */ +static void +netdelay_expires( NetDelay delay ) +{ + Session session; + int64_t now = qemu_get_clock( SHAPER_CLOCK ); + int rearm = 0; + int64_t rearm_time = 0; + + for (session = delay->sessions; session != NULL; session = session->next) + { + QueuedPacket packet = session->packet; + + if (packet == NULL) + continue; + + if (session->expiration <= now) { + /* send the SYN packet now */ + //fprintf(stderr, "NetDelay:RST: sending creation for %s\n", session_to_string(session) ); + delay->send_func( packet->data, packet->size, packet->opaque ); + session->packet = NULL; + queued_packet_free( packet ); + } else { + if (!rearm) { + rearm = 1; + rearm_time = session->expiration; + } + else if ( session->expiration < rearm_time ) + rearm_time = session->expiration; + } + } + + if (rearm) + qemu_mod_timer( delay->timer, rearm_time ); +} + + +NetDelay +netdelay_create( NetShaperSendFunc send_func ) +{ + NetDelay delay = qemu_malloc(sizeof(*delay)); + + init_timers(); + + delay->sessions = NULL; + delay->num_sessions = 0; + delay->timer = qemu_new_timer( SHAPER_CLOCK, + (QEMUTimerCB*) netdelay_expires, + delay ); + delay->active = 0; + delay->min_ms = 0; + delay->max_ms = 0; + + delay->send_func = send_func; + + return delay; +} + + +void +netdelay_set_latency( NetDelay delay, int min_ms, int max_ms ) +{ + /* when changing the latency, accept all sessions */ + while (delay->sessions) { + Session session = delay->sessions; + delay->sessions = session->next; + session->next = NULL; + if (session->packet) { + QueuedPacket packet = session->packet; + delay->send_func( packet->data, packet->size, packet->opaque ); + } + session_free(session); + delay->num_sessions--; + } + + delay->min_ms = min_ms; + delay->max_ms = max_ms; + delay->active = (min_ms <= max_ms) && min_ms > 0; +} + +void +netdelay_send( NetDelay delay, const void* data, size_t size ) +{ + netdelay_send_aux(delay, data, size, NULL); +} + + +void +netdelay_send_aux( NetDelay delay, const void* data, size_t size, void* opaque ) +{ + if (delay->active && !_packet_is_internal(data, size)) { + SessionRec info[1]; + int flags; + + flags = _packet_SYN_flags( data, size, info ); + if ((flags & 0x05) != 0) + { /* FIN or RST: drop connection */ + Session* lookup = netdelay_lookup_session( delay, info ); + Session session = *lookup; + if (session != NULL) { + //fprintf(stderr, "NetDelay:RST: dropping %s\n", session_to_string(info) ); + + *lookup = session->next; + session_free( session ); + delay->num_sessions -= 1; + } + } + else if ((flags & 0x12) == 0x02) + { + /* SYN: create connection */ + Session* lookup = netdelay_lookup_session( delay, info ); + Session session = *lookup; + + if (session != NULL) { + if (session->packet != NULL) { + /* this is a SYN re-transmission, since we didn't + * send the original SYN packet yet, just eat this one + */ + //fprintf(stderr, "NetDelay:RST: swallow SYN re-send for %s\n", session_to_string(info) ); + return; + } + } else { + /* establish a new session slightly in the future */ + int latency = delay->min_ms; + int range = delay->max_ms - delay->min_ms; + + if (delay > 0) + latency += rand() % range; + + //fprintf(stderr, "NetDelay:RST: delay creation for %s\n", session_to_string(info) ); + session = qemu_malloc( sizeof(*session) ); + + session->next = delay->sessions; + delay->sessions = session; + delay->num_sessions += 1; + + session->expiration = qemu_get_clock( SHAPER_CLOCK ) + latency; + + session->src_ip = info->src_ip; + session->dst_ip = info->dst_ip; + session->src_port = info->src_port; + session->dst_port = info->dst_port; + session->protocol = info->protocol; + + session->packet = queued_packet_create( data, size, opaque, 1 ); + + netdelay_expires(delay); + return; + } + } + } + + delay->send_func( (void*)data, size, opaque ); +} + + +void +netdelay_destroy( NetDelay delay ) +{ + if (delay) { + while (delay->sessions) { + Session session = delay->sessions; + delay->sessions = session->next; + session_free(session); + delay->num_sessions -= 1; + } + delay->active = 0; + qemu_free( delay ); + } +} + -- cgit v1.1