Unable to transmit fast via UDP broadcast on a wireless network - c++

I have the written the following code for transmitting UDP packets via broadcasting on a wireless network. The application that I have trying to develop requires the packets to be transmitted very fast, but unfortunately I cannot do so and need to add a sleep time. I find that below 500us sleep time, I am unable to send all the packets successfully.
Why does the sleep time have to be so high?
Is it possible to reduce this time by further optimization of this code?
If I do not process the received packets buffer, is it okay? Or does this create problems?
Note that I am running this code on a wireless radio which runs using OpenWrt.
Thanks in advance.
Code:
#include <sys/types.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <string.h>
#include <netdb.h>
#include <stdio.h>
#include <fcntl.h>
#include <string.h>
#include <sys/time.h>
#include <arpa/inet.h> /* for sockaddr_in */
#define BROADCAST_IP "192.168.255.255"
#define BROADCAST_PORT 45454
int b_sock=-1;
void init_socket()
{
unsigned short b_port = BROADCAST_PORT;
struct sockaddr_in b_addr;
int broadcastPermission;
char* rx_ip = BROADCAST_IP;
if ((b_sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP)) < 0)
perror("socket() failed");
/* Set socket to allow broadcast */
broadcastPermission = 1;
if (setsockopt(b_sock, SOL_SOCKET, SO_BROADCAST, (void *) &broadcastPermission, sizeof(broadcastPermission)) < 0)
perror("setsockopt() failed");
int opts;
opts = fcntl(b_sock,F_GETFL);
if(opts < 0)
perror("fcntl get failed");
opts = (opts | O_NONBLOCK);
if(fcntl(b_sock,F_SETFL,opts) < 0)
perror("fcntl set failed");
memset(&b_addr, 0, sizeof(b_addr)); /* Zero out structure */
b_addr.sin_family = AF_INET; /* Internet address family */
b_addr.sin_addr.s_addr = inet_addr(rx_ip);/* Broadcast IP address */
b_addr.sin_port = htons(b_port); /* Broadcast port */
if (bind(b_sock, (struct sockaddr *) &b_addr, sizeof(b_addr)) < 0)
perror("rx bind() failed");
}
void send_thread_body(long int buf, struct sockaddr_in tx_addr)
{
if(sendto(b_sock, &buf, sizeof(long int), 0, (struct sockaddr *)&tx_addr, sizeof(tx_addr)) < 0)
printf("tx sent diff num bytes than expected: %d\n",buf);
}
int main(int argc, char *argv[])
{
init_socket();
{
timeval start, end;
double diff = 0;
long int num = 0;
char *tx_ip = BROADCAST_IP;
unsigned short tx_port = BROADCAST_PORT;
struct sockaddr_in tx_addr;
memset(&tx_addr, 0, sizeof(tx_addr)); /* Zero out structure */
tx_addr.sin_family = AF_INET; /* Internet address family */
tx_addr.sin_addr.s_addr = inet_addr(tx_ip);/* Broadcast IP address */
tx_addr.sin_port = htons(tx_port); /* Broadcast port */
double next = 0;
double st = 0;
while (num<50000)
{
while (st <= next)
{
gettimeofday(&start,NULL);
st = start.tv_sec*1000 + ((double)start.tv_usec)/1000.0;
}
send_thread_body(num,tx_addr);
gettimeofday(&end, NULL);
diff += ((double)(((end.tv_sec - start.tv_sec)*1000000 + (end.tv_usec - start.tv_usec))))/1000000.0;
num++;
next = end.tv_sec*1000 + ((double)end.tv_usec)/1000.0 + 0.7;
}
printf("Avg time diff: %f\n",diff/50000.0);
}
close(b_sock);
return 0;
}

You are probably overflowing the socket buffer because you set the socket to O_NONBLOCK. Normally (when blocking is enabled), if the socket buffer is full, sendto blocks until there is sufficient buffer space to hold the message for sending.
From http://pubs.opengroup.org/onlinepubs/009695399/functions/sendto.html:
If space is not available at the
sending socket to hold the message to
be transmitted and the socket file
descriptor does not have O_NONBLOCK
set, sendto() shall block until space
is available. If space is not
available at the sending socket to
hold the message to be transmitted and
the socket file descriptor does have
O_NONBLOCK set, sendto() shall fail.
When you added sleeps between your sendto calls, you were effectively throttling down the throughput and preventing the socket buffers from overflowing.
Instead of sleep, you should use a blocking socket. If the socket buffers become full, sendto will block, which is effectively the same thing as sleeping, except that it will automatically stop sleeping the instant the socket is able to hold your next datagram.
To achieve better thoughput, try lumping data into datagrams close to the MTU size (while taking care to save enough room for UDP/IP headers). This should give you smaller header overhead compared to sending very short datagrams.

Related

TCP need to discard info on the buffer or make it faster

I am making a 3d application that works with the data of the accelerometer and gyroscope of the mobile. The problem is that the mobile sends data faster than the computer reads. The application increasingly makes the movements of the mobile with more delay as time passes. For example, at the beginning 2~3s is faithful to what the mobile does in reality, but over 10s it is making the movements that I did 6 seconds before.
I understand that it is reading data from the beginning buffer while the front of the most current data grows and never reaches it. I think the problem is how I read the data that comes to me.
Here is an example code that is implemented in the application. What could I do?
#include <unistd.h>
#include <stdio.h>
#include <sys/socket.h>
#include <stdlib.h>
#include <netinet/in.h>
#include <string.h>
#include <fcntl.h>
#include <algorithm>
#define PORT 8080
int main(int argc, char const *argv[])
{
int server_fd, new_socket, valread;
struct sockaddr_in address;
int opt = 1;
int addrlen = sizeof(address);
char buffer[1024] = {0};
const char *ACK = "ACKDAT\n";
std::string data;
socklen_t len;
char *error;
if ((server_fd = socket(AF_INET, SOCK_STREAM, 0)) == 0)
{
perror("socket failed");
exit(EXIT_FAILURE);
}
if (setsockopt(server_fd, SOL_SOCKET, SO_REUSEADDR | SO_REUSEPORT,&opt, sizeof(opt)))
{
perror("setsockopt");
exit(EXIT_FAILURE);
}
address.sin_family = AF_INET;
address.sin_addr.s_addr = INADDR_ANY;
address.sin_port = htons( PORT );
if (bind(server_fd, (struct sockaddr *)&address, sizeof(address))<0)
{
perror("bind failed");
exit(EXIT_FAILURE);
}
if (listen(server_fd, 3) < 0)
{
perror("listen");
exit(EXIT_FAILURE);
}
if ((new_socket = accept(server_fd, (struct sockaddr *)&address, (socklen_t*)&addrlen))<0)
{
perror("accept");
exit(EXIT_FAILURE);
}
fcntl(new_socket, F_SETFL, O_NONBLOCK);
while(true){
valread = read( new_socket , buffer, 1024);
for(int i = 0;i < 1024 ; i++){
if(buffer[i]!=0){
data = data + buffer[i];
}
buffer[i]=0;
}
if(!data.empty()){
//remove /n from data
data.erase(std::remove(data.begin(), data.end(), '\n'), data.end());
std::cout<<"#"<<data<<"#"<<std::endl;
send(new_socket , ACK , strlen(ACK) , 0 );
}
data.clear();
}
return 0;
}
While Sam Varshavchik's suggestion of using a thread is good, there's another option.
You already set your socket to non-blocking with fcntl(new_socket, F_SETFL, O_NONBLOCK);. So, at each loop you should read everything there is to read and send everything there is to send. If you don't tie one-to-one the reading and writing, both sides will be able to catch up.
The main hint that you need to fix this is that you don't use the read return value, valread. Your code should look like:
while(true){ // main loop
...
valread = read( new_socket , buffer, 1024);
while(valread > 0)
{
// deal with the read values.
// deal with receiving more than one packet per iteration
}
// send code done a single time per loop.
There still plenty of architecture you need to have a clean resilient main loop that sends and receives, but I hope that points you in a useful direction.
You can have all reading from the socket done in a separate std::thread, that does nothing but read from the socket and save the data in your program's internal buffer. Your existing program reads from the internal buffer, rather than a socket, whenever it is capable of doing so.
If the internal buffer is full you'll have to figure out what to do about it, probably discard unwanted data. Perhaps stash a copy of the most recent received record, in a safe place, and put it into the buffer when things get unclogged. The dedicated execution thread will likely need to do some minimal processing, to parse the raw data from the socket into some meaningful, logical form. It's unlikely that you can simply throw away random parts of the raw data stream from the socket. It will need to be logically parsed, in some form or fashion that's particular to your application, so this can be done in a meaningful way.
You will need to implement your application's buffer in a thread-safe manner, so that it can be simultaneously accessed by multiple execution threads. If you're not familiar with multi-threaded programming in C++, you will find more information in your C++ textbook, on this topic.

iOS UDP socket "sendto" can't send immediately one by one

I found that the behavior of iOS socket "sendto" interface was different from android or linux kernel. When we call "sendto",we start wireshark capture at the same time . But we can't find the capture data immediately. We make a test that we continuously sending 332 bytes by "sendto" per 40ms interval.And we found that all data packets was send to the network in a short central time instead of a 40MS(or above) intervals. We doubt that the ios kernal made some change for udp.Also it may calls audio delay about 900-1000ms.
I have test in iphone 5s & iphone 6,iOS 10 or 11 by demo use "sendto" api
#include <sys/socket.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <sys/types.h>
int sSock = socket(AF_INET,SOCK_DGRAM,0);
struct sockaddr_in localAddr;
localAddr.sin_family = AF_INET;
localAddr.sin_port = htons(1001);
localAddr.sin_addr.s_addr = htonl(INADDR_ANY);
if(bind(sSock, (struct sockaddr*)&localAddr, sizeof(localAddr)) < 0)
{
NSLog(#"bind error");
return;
}
struct sockaddr_in sendAddr;
sendAddr.sin_family = AF_INET;
sendAddr.sin_port = htons(21000);
inet_pton(AF_INET, "114.190.105.220", &sendAddr.sin_addr);
char sendBuf[332] = {0};
//unsigned char service_type = 0xe0 | IPTOS_LOWDELAY | IPTOS_RELIABILITY;
int service_type = 0x10;
int priority = 6;
//if(setsockopt(sSock, SOL_SOCKET, SO_PRIORITY, &priority, sizeof(priority))<0)
if(setsockopt(sSock,IPPROTO_IP,IP_TOS,(void*)(&service_type),sizeof(service_type)) < 0)
{
NSLog(#"setsockopt failed1,error[%s]",strerror(errno));
}
while(true)
{
sendto(sSock, sendBuf, sizeof(sendBuf), 0, (struct sockaddr*)&sendAddr, sizeof(sendAddr));
time_t timeNow;
time(&timeNow);
struct tm tmNow;
struct timeval tvNow;
localtime_r(&timeNow,&tmNow);
gettimeofday(&tvNow, NULL);
char sTime[256] = {0};
snprintf(sTime, sizeof(sTime)-1, "%04d%02d%02d %02d%02d%02d.%d",tmNow.tm_year+1900,tmNow.tm_mon,tmNow.tm_mday,tmNow.tm_hour,tmNow.tm_min,tmNow.tm_sec,tvNow.tv_usec/1000);
NSLog(#"[%s]interval 40ms send",sTime);
usleep(40*1000);
}
all data packets was send to the network in a short central time instead of a 40MS(or above) intervals.
all packets send at once,wireshark capture like this:
No. Time Protocol
819 41.439392 H264
820 41.439617 H264
821 41.439819 H264

What is the proper process for ICMP echo request/reply on unreachable destinations?

Goal:
I need to be able to ping a network switch to determine whether or not it is available. This is meant to tell the user that either the network cabling is unplugged, the network switch is unavailable, or some other problem lies within the network communication pathway. I realize this is not a comprehensive diagnosis tool, but something is better than nothing.
Design:
I planned on using ICMP with raw sockets to send five (5) ping messages to a particular address in IPv4 dot-notation. I will setup an ICMP filter on the socket and will not be creating my own IP header. Transmission of the ICMP will be through the sendto method and reception through the recvfrom method. This will occur on a single thread (though another thread can be used to break transmission and reception apart). Reception of a message will further be filtered by matching the ID of the received message to the ID that was transmitted. The ID stored will be the running process ID of the application. If an ICMP_ECHOREPLY message is received and the ID of the message and the stored ID match, then a counter is incremented until five (4) has been reached (the counter is zero-based). I will attempt to send a ping, wait for its reply, and repeat this process five (5) times.
The Problem:
After having implemented my design, whenever I ping a particular valid network address (say 192.168.11.15) with an active network participant, I receive ICMP_ECHOREPLY messages for each of the five (5) pings. However, whenever I ping a valid network address (say 192.168.30.30) with inactive network participants (meaning no device is connected to the particular address), I get one (1) ICMP_DEST_UNREACH, and four (4) ICMP_ECHOREPLY messages. The ID in the reply messages match the ID stored within the software. Whenever I perform a 'ping 192.168.30.30' from the commandline, I get 'From 192.168.40.50 icmp_seq=xx Destination Host Unreachable'. Am I not supposed to be receiving ICMP_DEST_UNREACH messages instead of ICMP_ECHOREPLY messages?
The Code:
Ping.h:
#include <netinet/in.h>
#include <linux/ip.h>
#include <linux/ipmc.h>
#include <arpa/inet.h>
#include <cstdio>
#include <cstdlib>
#include <stdint.h>
#include <time.h>
#include <errno.h>
#include <string>
#include <cstring>
#include <netdb.h>
class Ping
{
public:
Ping(std::string host) : _host(host) {}
~Ping() {}
void start()
{
int sock = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP);
if(sock < 0)
{
printf("Failed to create socket!\n");
close(sock);
exit(1);
}
setuid(getuid());
sockaddr_in pingaddr;
memset(&pingaddr, 0, sizeof(sockaddr_in));
pingaddr.sin_family = AF_INET;
hostent *h = gethostbyname(_host.c_str());
if(not h)
{
printf("Failed to get host by name!\n");
close(sock);
exit(1);
}
memcpy(&pingaddr.sin_addr, h->h_addr, sizeof(pingaddr.sin_addr));
// Set the ID of the sender (will go into the ID of the echo msg)
int pid = getpid();
// Only want to receive the following messages
icmp_filter filter;
filter.data = ~((1<<ICMP_SOURCE_QUENCH) |
(1<<ICMP_DEST_UNREACH) |
(1<<ICMP_TIME_EXCEEDED) |
(1<<ICMP_REDIRECT) |
(1<<ICMP_ECHOREPLY));
if(setsockopt(sock, SOL_RAW, ICMP_FILTER, (char *)&filter, sizeof(filter)) < 0)
{
perror("setsockopt(ICMP_FILTER)");
exit(3);
}
// Number of valid echo receptions
int nrec = 0;
// Send the packet
for(int i = 0; i < 5; ++i)
{
char packet[sizeof(icmphdr)];
memset(packet, 0, sizeof(packet));
icmphdr *pkt = (icmphdr *)packet;
pkt->type = ICMP_ECHO;
pkt->code = 0;
pkt->checksum = 0;
pkt->un.echo.id = htons(pid & 0xFFFF);
pkt->un.echo.sequence = i;
pkt->checksum = checksum((uint16_t *)pkt, sizeof(packet));
int bytes = sendto(sock, packet, sizeof(packet), 0, (sockaddr *)&pingaddr, sizeof(sockaddr_in));
if(bytes < 0)
{
printf("Failed to send to receiver\n");
close(sock);
exit(1);
}
else if(bytes != sizeof(packet))
{
printf("Failed to write the whole packet --- bytes: %d, sizeof(packet): %d\n", bytes, sizeof(packet));
close(sock);
exit(1);
}
while(1)
{
char inbuf[192];
memset(inbuf, 0, sizeof(inbuf));
int addrlen = sizeof(sockaddr_in);
bytes = recvfrom(sock, inbuf, sizeof(inbuf), 0, (sockaddr *)&pingaddr, (socklen_t *)&addrlen);
if(bytes < 0)
{
printf("Error on recvfrom\n");
exit(1);
}
else
{
if(bytes < sizeof(iphdr) + sizeof(icmphdr))
{
printf("Incorrect read bytes!\n");
continue;
}
iphdr *iph = (iphdr *)inbuf;
int hlen = (iph->ihl << 2);
bytes -= hlen;
pkt = (icmphdr *)(inbuf + hlen);
int id = ntohs(pkt->un.echo.id);
if(pkt->type == ICMP_ECHOREPLY)
{
printf(" ICMP_ECHOREPLY\n");
if(id == pid)
{
nrec++;
if(i < 5) break;
}
}
else if(pkt->type == ICMP_DEST_UNREACH)
{
printf(" ICMP_DEST_UNREACH\n");
// Extract the original data out of the received message
int offset = sizeof(iphdr) + sizeof(icmphdr) + sizeof(iphdr);
if(((bytes + hlen) - offset) == sizeof(icmphdr))
{
icmphdr *p = reinterpret_cast<icmphdr *>(inbuf + offset);
id = ntohs(p->un.echo.id);
if(origid == pid)
{
printf(" IDs match!\n");
break;
}
}
}
}
}
}
printf("nrec: %d\n", nrec);
}
private:
int32_t checksum(uint16_t *buf, int32_t len)
{
int32_t nleft = len;
int32_t sum = 0;
uint16_t *w = buf;
uint16_t answer = 0;
while(nleft > 1)
{
sum += *w++;
nleft -= 2;
}
if(nleft == 1)
{
*(uint16_t *)(&answer) = *(uint8_t *)w;
sum += answer;
}
sum = (sum >> 16) + (sum & 0xFFFF);
sum += (sum >> 16);
answer = ~sum;
return answer;
}
std::string _host;
};
main.cpp:
#include "Ping.h"
int main()
{
// Ping ping("192.168.11.15");
Ping ping("192.168.30.30");
ping.start();
while(1) sleep(10);
}
In order to compile, just type 'g++ main.cpp -o ping' into the command-line of a Linux box, and it should compile (that is, if all of the source code is installed).
Conclusion:
Can anyone tell me why I am receiving one (1) ICMP_DEST_UNREACH and four (4) ICMP_ECHOREPLY messages from a device that isn't on that particular network address?
NOTE: You can change the network IP address from the main.cpp file. Just change the IP to a device that actually exists on your network or a device that doesn't exist on your network.
I'm also not interested in criticisms about coding style. I know it isn't pretty, has 'C' style casting mixed with C++ casts, has poor memory management, etc, but this is only prototype code. It isn't meant to be pretty.
Ok i found the error. Look at this two lines.
int bytes = sendto(sock, packet, sizeof(packet), 0, (sockaddr *)&pingaddr, sizeof(sockaddr_in));
bytes = recvfrom(sock, inbuf, sizeof(inbuf), 0, (sockaddr *)&pingaddr, (socklen_t *)&addrlen);
both functions uses pingaddr pointer as parameter, but this should avoided because in the sendto() function is used to point the destination IP of the icmp packet but in the recvfrom() is used to get back the IP of the host that's replying.
Let's say pingaddr is set with an IP not reachable. After your first ICMP_REQUEST the first gateway will reply to you with a ICMP_DEST_UNREACH and... here comes the error... when recvfrom is called, pingaddr structure will be overwritten with the IP of the gateway.
SO... from the second ping you'll be pointing to the gateway IP that, obviously, exists and will reply with a ICMP_ECHOREPLY.
SOLUTION:
avoid pass the same sockaddr_in structure pointer to both sendto() and recvfrom().

throughput control via Sleep

I have this code, the point is that I want to receive no more than 600Kbits/s using a UDP socket, I implemented an algorithm using duration during which we receive and the sleep command....
#if defined (WIN32)
#include <winsock2.h>
typedef int socklen_t;
#elif defined (linux)
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <unistd.h>
#define INVALID_SOCKET -1
#define SOCKET_ERROR -1
#define closesocket(s) close(s)
typedef int SOCKET;
typedef struct sockaddr_in SOCKADDR_IN;
typedef struct sockaddr SOCKADDR;
#endif
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#define RCVBUFSIZE 4
#define PORT 4444
#define ECHOMAX 255
int main(void)
{
#if defined (WIN32)
WSADATA WSAData;
int erreur = WSAStartup(MAKEWORD(2,2), &WSAData);
#else
int erreur = 0;
#endif
int recvMsgSize;
char echoBuffer[RCVBUFSIZE];
//unsigned int echoStringLen;
//int bytesRcvd, totalBytesRcvd; //bytes received in a single recv()
SOCKET sock;
SOCKADDR_IN sin;
SOCKADDR_IN SenderAddr;
int SenderAddrSize = sizeof(SenderAddr);
if(!erreur)
{
sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
sin.sin_addr.s_addr = inet_addr("127.0.0.1");
sin.sin_family = AF_INET;
sin.sin_port = htons(4444);
memset(&sin.sin_zero, '\0', sizeof(sin.sin_zero));
bind(sock, (SOCKADDR*)&sin, sizeof(sin));
//totalBytesRcvd = 0;
printf("Received: ");
int speed_limit= 614400; //600Kbits/s
int one_second=1000;
int elapsed; //elapsed time
int transmit=0; // how much i receive during 'elapsed'
int expected_receive; //what I'm excpected to receive
int transmit_delta; //diference
int time_sleep; //how much to sleep
clock_t start_time= clock();
for(;;)
{
if((recvMsgSize=recvfrom(sock, echoBuffer,1024, 0, (SOCKADDR *)&SenderAddr, &SenderAddrSize)) > 0)
{
transmit+=recvMsgSize;
clock_t tempo= clock();
elapsed=(tempo-start_time)/CLOCKS_PER_SEC;
// Check moment speed every five second, you can choose any value
if(elapsed>5)
{
start_time= tempo;
expected_receive=speed_limit*elapsed/8;
transmit_delta=expected_receive-transmit;
if(transmit_delta>0)
{
time_sleep=8*transmit_delta*one_second/speed_limit;
Sleep(time_sleep);
}
transmit=0;
}
echoBuffer[recvMsgSize]='\0';
printf(echoBuffer);
}
}
printf("\n");
getchar() ;
closesocket(sock);
#if defined (WIN32)
WSACleanup();
#endif
}
return EXIT_SUCCESS;
}
the problem is that it receive the message but it blocks the receiving process from time to time...which I guess will cause loss of data especially when I'm using UDP...
any alternative solution is welcolme...
thanks in advance..
Better not to sleep - just read and discard data while you are over the threshold, then resume processing it once you are ready to do so (after interval expires).
If the data is arriving over UDP, there is no way to reduce the throughput at the receiving end. You'd have to tell the sending computer to send less data. The receiver's only choice is the receive the UDP packets, or let them get dropped, and do something with them, or not.
If you use TCP, then the TCP protocol includes a built-in way to do what you want (called "source quench") ... the TCP protocol layer of the receiver will send a message to the sender telling the sender to slow down if the receiver is having problems keeping up. (and you can influence whether "the receiver is having problems keeping up" by reading the data slower than usual, if you want).
If you use non-blocking sockets with select() you can just ignore the data until you're ready to read it. Not reading data will cause the TCP connection to "slow down". However, this way you do not have exact control about the amount of data received. It's just an approximation. All techniques I know of that limit bandwidth in a more accurate way are implemented on the sender side.

Send and Receive a file in socket programming in Linux with C/C++ (GCC/G++)

I would like to implement a client-server architecture running on Linux using sockets and C/C++ language that is capable of sending and receiving files. Is there any library that makes this task easy? Could anyone please provide an example?
The most portable solution is just to read the file in chunks, and then write the data out to the socket, in a loop (and likewise, the other way around when receiving the file). You allocate a buffer, read into that buffer, and write from that buffer into your socket (you could also use send and recv, which are socket-specific ways of writing and reading data). The outline would look something like this:
while (1) {
// Read data into buffer. We may not have enough to fill up buffer, so we
// store how many bytes were actually read in bytes_read.
int bytes_read = read(input_file, buffer, sizeof(buffer));
if (bytes_read == 0) // We're done reading from the file
break;
if (bytes_read < 0) {
// handle errors
}
// You need a loop for the write, because not all of the data may be written
// in one call; write will return how many bytes were written. p keeps
// track of where in the buffer we are, while we decrement bytes_read
// to keep track of how many bytes are left to write.
void *p = buffer;
while (bytes_read > 0) {
int bytes_written = write(output_socket, p, bytes_read);
if (bytes_written <= 0) {
// handle errors
}
bytes_read -= bytes_written;
p += bytes_written;
}
}
Make sure to read the documentation for read and write carefully, especially when handling errors. Some of the error codes mean that you should just try again, for instance just looping again with a continue statement, while others mean something is broken and you need to stop.
For sending the file to a socket, there is a system call, sendfile that does just what you want. It tells the kernel to send a file from one file descriptor to another, and then the kernel can take care of the rest. There is a caveat that the source file descriptor must support mmap (as in, be an actual file, not a socket), and the destination must be a socket (so you can't use it to copy files, or send data directly from one socket to another); it is designed to support the usage you describe, of sending a file to a socket. It doesn't help with receiving the file, however; you would need to do the loop yourself for that. I cannot tell you why there is a sendfile call but no analogous recvfile.
Beware that sendfile is Linux specific; it is not portable to other systems. Other systems frequently have their own version of sendfile, but the exact interface may vary (FreeBSD, Mac OS X, Solaris).
In Linux 2.6.17, the splice system call was introduced, and as of 2.6.23 is used internally to implement sendfile. splice is a more general purpose API than sendfile. For a good description of splice and tee, see the rather good explanation from Linus himself. He points out how using splice is basically just like the loop above, using read and write, except that the buffer is in the kernel, so the data doesn't have to transferred between the kernel and user space, or may not even ever pass through the CPU (known as "zero-copy I/O").
Do aman 2 sendfile. You only need to open the source file on the client and destination file on the server, then call sendfile and the kernel will chop and move the data.
Minimal runnable POSIX read + write example
Usage:
get two computers on a LAN.
For example, this will work if both computers are connected to your home router in most cases, which is how I tested it.
On the server computer:
Find the server local IP with ifconfig, e.g. 192.168.0.10
Run:
./server output.tmp 12345
On the client computer:
printf 'ab\ncd\n' > input.tmp
./client input.tmp 192.168.0.10 12345
Outcome: a file output.tmp is created on the sever computer containing 'ab\ncd\n'!
server.c
/*
Receive a file over a socket.
Saves it to output.tmp by default.
Interface:
./executable [<output_file> [<port>]]
Defaults:
- output_file: output.tmp
- port: 12345
*/
#define _XOPEN_SOURCE 700
#include <stdio.h>
#include <stdlib.h>
#include <arpa/inet.h>
#include <fcntl.h>
#include <netdb.h> /* getprotobyname */
#include <netinet/in.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <unistd.h>
int main(int argc, char **argv) {
char *file_path = "output.tmp";
char buffer[BUFSIZ];
char protoname[] = "tcp";
int client_sockfd;
int enable = 1;
int filefd;
int i;
int server_sockfd;
socklen_t client_len;
ssize_t read_return;
struct protoent *protoent;
struct sockaddr_in client_address, server_address;
unsigned short server_port = 12345u;
if (argc > 1) {
file_path = argv[1];
if (argc > 2) {
server_port = strtol(argv[2], NULL, 10);
}
}
/* Create a socket and listen to it.. */
protoent = getprotobyname(protoname);
if (protoent == NULL) {
perror("getprotobyname");
exit(EXIT_FAILURE);
}
server_sockfd = socket(
AF_INET,
SOCK_STREAM,
protoent->p_proto
);
if (server_sockfd == -1) {
perror("socket");
exit(EXIT_FAILURE);
}
if (setsockopt(server_sockfd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable)) < 0) {
perror("setsockopt(SO_REUSEADDR) failed");
exit(EXIT_FAILURE);
}
server_address.sin_family = AF_INET;
server_address.sin_addr.s_addr = htonl(INADDR_ANY);
server_address.sin_port = htons(server_port);
if (bind(
server_sockfd,
(struct sockaddr*)&server_address,
sizeof(server_address)
) == -1
) {
perror("bind");
exit(EXIT_FAILURE);
}
if (listen(server_sockfd, 5) == -1) {
perror("listen");
exit(EXIT_FAILURE);
}
fprintf(stderr, "listening on port %d\n", server_port);
while (1) {
client_len = sizeof(client_address);
puts("waiting for client");
client_sockfd = accept(
server_sockfd,
(struct sockaddr*)&client_address,
&client_len
);
filefd = open(file_path,
O_WRONLY | O_CREAT | O_TRUNC,
S_IRUSR | S_IWUSR);
if (filefd == -1) {
perror("open");
exit(EXIT_FAILURE);
}
do {
read_return = read(client_sockfd, buffer, BUFSIZ);
if (read_return == -1) {
perror("read");
exit(EXIT_FAILURE);
}
if (write(filefd, buffer, read_return) == -1) {
perror("write");
exit(EXIT_FAILURE);
}
} while (read_return > 0);
close(filefd);
close(client_sockfd);
}
return EXIT_SUCCESS;
}
client.c
/*
Send a file over a socket.
Interface:
./executable [<input_path> [<sever_hostname> [<port>]]]
Defaults:
- input_path: input.tmp
- server_hostname: 127.0.0.1
- port: 12345
*/
#define _XOPEN_SOURCE 700
#include <stdio.h>
#include <stdlib.h>
#include <arpa/inet.h>
#include <fcntl.h>
#include <netdb.h> /* getprotobyname */
#include <netinet/in.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <unistd.h>
int main(int argc, char **argv) {
char protoname[] = "tcp";
struct protoent *protoent;
char *file_path = "input.tmp";
char *server_hostname = "127.0.0.1";
char *server_reply = NULL;
char *user_input = NULL;
char buffer[BUFSIZ];
in_addr_t in_addr;
in_addr_t server_addr;
int filefd;
int sockfd;
ssize_t i;
ssize_t read_return;
struct hostent *hostent;
struct sockaddr_in sockaddr_in;
unsigned short server_port = 12345;
if (argc > 1) {
file_path = argv[1];
if (argc > 2) {
server_hostname = argv[2];
if (argc > 3) {
server_port = strtol(argv[3], NULL, 10);
}
}
}
filefd = open(file_path, O_RDONLY);
if (filefd == -1) {
perror("open");
exit(EXIT_FAILURE);
}
/* Get socket. */
protoent = getprotobyname(protoname);
if (protoent == NULL) {
perror("getprotobyname");
exit(EXIT_FAILURE);
}
sockfd = socket(AF_INET, SOCK_STREAM, protoent->p_proto);
if (sockfd == -1) {
perror("socket");
exit(EXIT_FAILURE);
}
/* Prepare sockaddr_in. */
hostent = gethostbyname(server_hostname);
if (hostent == NULL) {
fprintf(stderr, "error: gethostbyname(\"%s\")\n", server_hostname);
exit(EXIT_FAILURE);
}
in_addr = inet_addr(inet_ntoa(*(struct in_addr*)*(hostent->h_addr_list)));
if (in_addr == (in_addr_t)-1) {
fprintf(stderr, "error: inet_addr(\"%s\")\n", *(hostent->h_addr_list));
exit(EXIT_FAILURE);
}
sockaddr_in.sin_addr.s_addr = in_addr;
sockaddr_in.sin_family = AF_INET;
sockaddr_in.sin_port = htons(server_port);
/* Do the actual connection. */
if (connect(sockfd, (struct sockaddr*)&sockaddr_in, sizeof(sockaddr_in)) == -1) {
perror("connect");
return EXIT_FAILURE;
}
while (1) {
read_return = read(filefd, buffer, BUFSIZ);
if (read_return == 0)
break;
if (read_return == -1) {
perror("read");
exit(EXIT_FAILURE);
}
/* TODO use write loop: https://stackoverflow.com/questions/24259640/writing-a-full-buffer-using-write-system-call */
if (write(sockfd, buffer, read_return) == -1) {
perror("write");
exit(EXIT_FAILURE);
}
}
free(user_input);
free(server_reply);
close(filefd);
exit(EXIT_SUCCESS);
}
GitHub upstream.
Further comments
Possible improvements:
Currently output.tmp gets overwritten each time a send is done.
This begs for the creation of a simple protocol that allows to pass a filename so that multiple files can be uploaded, e.g.: filename up to the first newline character, max filename 256 chars, and the rest until socket closure are the contents. Of course, that would require sanitation to avoid a path transversal vulnerability.
Alternatively, we could make a server that hashes the files to find filenames, and keeps a map from original paths to hashes on disk (on a database).
Only one client can connect at a time.
This is specially harmful if there are slow clients whose connections last for a long time: the slow connection halts everyone down.
One way to work around that is to fork a process / thread for each accept, start listening again immediately, and use file lock synchronization on the files.
Add timeouts, and close clients if they take too long. Or else it would be easy to do a DoS.
poll or select are some options: How to implement a timeout in read function call?
A simple HTTP wget implementation is shown at: How to make an HTTP get request in C without libcurl?
Tested on Ubuntu 15.10.
This file will serve you as a good sendfile example : http://tldp.org/LDP/LGNET/91/misc/tranter/server.c.txt