Getting the size of the client recv buffer

Niara · 10-08-2006, 09:13 AM

Hello, recently I have been working on a very very simple server, it works well but in some cases is unnecesarily slow. I was wondering if there's a standard call to the client to get the size of recv buffer, so I could adapt my send buffer to the max size instead of checking each time the received size. First I was thinking on send a huge string and use the 'client readed bytes' as max client buffer, but I can see that's a bad solution. I'm using winsock2, and I have taked a look at the ws2 manual reference but with no results.

Thank's in advance
Niara

**Salem** · 10-08-2006, 10:56 AM

No and No.

Both sending and receiving can fragment the message, you have no control over that.

If you call send() with a large buffer, you must look at the return result to see how much was actually sent, and if it's short, then call send() again with the remainder of the buffer until it's all sent.

Likewise with recv(), you can supply a large buffer but recv() won't necessarily wait for it to be filled. Look at the return result to work out how much was received and act accordingly.

If you think it will help, a common technique is to send the message length in front of the message, so you know pretty early on how much data to expect.

Niara · 10-08-2006, 01:50 PM

Hello Salem, thank's for the reply.

I was thought that maybe there was a getclientbuffersize(). For now I'm using (and I will have to continue using) a variation of the beej guide 'sendall()' function. That's a simulation of a send process that I made for implement the function:

Code:

#include <conio.h>
#include <stdio.h>
#include <stdlib.h>

int send(char* ,int );

int main()
{
FILE *arx;
char *bff=NULL;
int ta,tb,tl,q;
int total,restants,n;
int inc;
bool rm;

arx=fopen("test.txt","rb");
fseek(arx,0L,SEEK_END);
ta=ftell(arx);
fseek(arx,0L,SEEK_SET);

tl=0;
tb=27;
bff=(char*)malloc(sizeof(char)*tb);

inc=0;
rm=false;
while(tl<ta)
    {
    if(rm)
        {
        tb++;
        bff=(char*)realloc(bff,sizeof(char)*tb);
        rm=false;
        }
    
    for(q=0;q<tb;q++)
        {
        //if(feof(arx)!=0) {break;}
        if(tl>=ta) {break;}
        bff[q]=getc(arx);
        tl++;
        }
    bff[q]='\0';
    
    total=0;
    restants=strlen(bff);
    while(total<strlen(bff))
        {
        n=send(bff+total,restants);
        
        if(!rm && n==strlen(bff))
            {
            inc=1;
            rm=true;
            }

        if(n==-1) {break;}
        total+=n;
        restants-=n;
        }
    }

if(bff) {free(bff);}
send(NULL,0);

fclose(arx);
getch();
return 0;
}

int send(char* data,int t)
{
static char complert[1024]="";
int tb,q,bl=0;
char  bff[10];

if(!data) {printf("%s",complert);return 0;}

(t<10) ? tb=t : tb=10;

for(q=0;q<tb;q++)
    {
    bff[q]=data[q];
    bl++;
    }
bff[q]='\0';
sprintf(complert,"%s%s",complert,bff);
return bl;
}
//linked/compiled with mingw & devcpp as a *.cpp

it works with sokets simply using the sokets 'send' function instead my simulation call

Code:

n=send(thesocket,bff+total,restants,0);

the modification is the buffer memory reallocation when the client's buffer is bigger than the server buffer, so I can send more bytes. The method is not completed because the server buffer is reallocated incrementing by 1, so if I could know the client buffer size I would be able to realloc at the exact size to optimze the sending process. Sure that I'm not be the first on think about that...

Feel free to use/modify/comment/criticyze that code I posted

Niara

**Salem** · 10-08-2006, 02:11 PM

> #include <conio.h>
The only compiler which needed this header file couldn't do network stuff, so why do you need it?
It's an obsolete header.

> bff=(char*)malloc(sizeof(char)*tb);
Don't cast malloc in C
See the FAQ

Oh now I see
> linked/compiled with mingw & devcpp as a *.cpp
Use a C compiler for your C code then.
Well it's all C except for your use of bool, and that's easy to fix.

> (t<10) ? tb=t : tb=10;
Mmm - if / else would have been clearer, as would
tb = t < 10 ? t : 10;

> sprintf(complert,"%s%s",complert,bff);
Very few C functions (say memmove) work properly if the input and output overlap. Trying to sprintf into the same buffer as you're trying to read from as well is just a recipe for disaster.
Especially as all you needed was strcat()

> bff=(char*)realloc(bff,sizeof(char)*tb);
Classic realloc mis-use.
Always use a temp, and check it for NULL

Code:

temp = realloc ( oldptr, size );
if ( temp ) oldptr = temp;
else // panic

> bff[q]='\0';
You set tb to 10, so when you get here, q is 10
Buffer overflow!

Niara · 10-08-2006, 03:22 PM

Hello, more thanks for the reply.

Step by step. I have used the conio.h only in the test while implementing the function (basically to call to 'getch()'), not in the real network code. The C mixed with some C++, it's a bad learning practice, I like C, but C++ is more flexible (I'm not expert in neither C nor C++, but since c++ compiler can compile c code -maybe something not, I don't know- I think that there can be any problem). i.e. I don't know how to work with files in cpp, then I do it in c; I like use vectors of structures but I don't know how to do it in c, then I do it in cpp;etc...
The 'sprintf' over the same buffer, yes a very bad habit, I should use the 'strcat' instead. And the buffer overflow, I saw it while coding but never changed it.
But the important thing on the code I posted is the sending process, verify the received bytes and resend the bytes left (my 'send' function is only a simulation, I never used out of that code).

And I will add my own critic: I have seen now that a sending process like that only work with txt files, testing it with images does not work because of the '\0' that I add at the end of the strings, so I have modifyed a bit

Code:

bool sendTheFile(SOCKET s,char *ruta)
{
FILE *arx;
char *bff=NULL;
int ta,tb,tl,q,ba;
int total,restants,n;
bool rm,eok;

if(!(arx=fopen(ruta,"rb"))) {return false;}
fseek(arx,0L,SEEK_END);
ta=ftell(arx);
fseek(arx,0L,SEEK_SET);

tl=0;
tb=100;
bff=(char*)malloc(sizeof(char)*tb);
if(!bff) {return false;}
rm=false;
eok=true;
while(tl<ta)
    {
    if(rm)
        {
        tb+=50;
        bff=(char*)realloc(bff,sizeof(char)*tb);
        if(!bff) {eok=false;break;}
        rm=false;
        }
    
    ba=0;
    for(q=0;q<tb;q++)
        {
        if(feof(arx)!=0) {break;}
        bff[q]=getc(arx);
        tl++;
        ba++;
        }
    
    total=0;
    restants=ba;
    while(total<ba)
        {
        n=send(s,bff+total,restants,0);
        if(!rm && n==ba)
            {
            rm=true;
            }
        if(n==-1) {break;}
        total+=n;
        restants-=n;
        }
    }
    
if(bff) {free(bff);}
fclose(arx);
return eok;
}

Is the same as before, but I deleted the '\0' at the end of the buffer to send. Maybe is not 100% optimized, but it works (for the moment).

Niara

vart · 10-26-2006, 02:25 PM

ftell returns long - you put it into int. Not a good practice.

And you don't check the resulting value for possible errors.

Advice - use fread instead of the loop to fill the byte array from the file.

Niara · 10-27-2006, 11:46 AM

Hey vart, thanks for your observation.

I have already modifyed the code checking for the error return '-1' on the 'ftell()' function. I didn't realized that I was storing the file size on an 'int'

(I have commented something similar: bad learning habits produce errors in such simple codes), I will have to correct that part.

Niara

10-08-2006, 09:13 AM	#1
Niara Registered User Join Date: Mar 2005 Location: Juneda Posts: 229	Getting the size of the client recv buffer Hello, recently I have been working on a very very simple server, it works well but in some cases is unnecesarily slow. I was wondering if there's a standard call to the client to get the size of recv buffer, so I could adapt my send buffer to the max size instead of checking each time the received size. First I was thinking on send a huge string and use the 'client readed bytes' as max client buffer, but I can see that's a bad solution. I'm using winsock2, and I have taked a look at the ws2 manual reference but with no results. Thank's in advance Niara
Niara is online now

10-08-2006, 10:56 AM	#2
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,688	No and No. Both sending and receiving can fragment the message, you have no control over that. If you call send() with a large buffer, you must look at the return result to see how much was actually sent, and if it's short, then call send() again with the remainder of the buffer until it's all sent. Likewise with recv(), you can supply a large buffer but recv() won't necessarily wait for it to be filled. Look at the return result to work out how much was received and act accordingly. If you think it will help, a common technique is to send the message length in front of the message, so you know pretty early on how much data to expect. __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

10-08-2006, 02:11 PM	#4
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,688	> #include <conio.h> The only compiler which needed this header file couldn't do network stuff, so why do you need it? It's an obsolete header. > bff=(char)malloc(sizeof(char)tb); Don't cast malloc in C See the FAQ Oh now I see > linked/compiled with mingw & devcpp as a .cpp Use a C compiler for your C code then. Well it's all C except for your use of bool, and that's easy to fix. > (t<10) ? tb=t : tb=10; Mmm - if / else would have been clearer, as would tb = t < 10 ? t : 10; > sprintf(complert,"%s%s",complert,bff); Very few C functions (say memmove) work properly if the input and output overlap. Trying to sprintf into the same buffer as you're trying to read from as well is just a recipe for disaster. Especially as all you needed was strcat() > bff=(char)realloc(bff,sizeof(char)*tb); Classic realloc mis-use. Always use a temp, and check it for NULL Code: temp = realloc ( oldptr, size ); if ( temp ) oldptr = temp; else // panic > bff[q]='\0'; You set tb to 10, so when you get here, q is 10 Buffer overflow! __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

10-08-2006, 03:22 PM	#5
Niara Registered User Join Date: Mar 2005 Location: Juneda Posts: 229	Hello, more thanks for the reply. Step by step. I have used the conio.h only in the test while implementing the function (basically to call to 'getch()'), not in the real network code. The C mixed with some C++, it's a bad learning practice, I like C, but C++ is more flexible (I'm not expert in neither C nor C++, but since c++ compiler can compile c code -maybe something not, I don't know- I think that there can be any problem). i.e. I don't know how to work with files in cpp, then I do it in c; I like use vectors of structures but I don't know how to do it in c, then I do it in cpp;etc... The 'sprintf' over the same buffer, yes a very bad habit, I should use the 'strcat' instead. And the buffer overflow, I saw it while coding but never changed it. But the important thing on the code I posted is the sending process, verify the received bytes and resend the bytes left (my 'send' function is only a simulation, I never used out of that code). And I will add my own critic: I have seen now that a sending process like that only work with txt files, testing it with images does not work because of the '\0' that I add at the end of the strings, so I have modifyed a bit Code: bool sendTheFile(SOCKET s,char ruta) { FILE arx; char bff=NULL; int ta,tb,tl,q,ba; int total,restants,n; bool rm,eok; if(!(arx=fopen(ruta,"rb"))) {return false;} fseek(arx,0L,SEEK_END); ta=ftell(arx); fseek(arx,0L,SEEK_SET); tl=0; tb=100; bff=(char)malloc(sizeof(char)tb); if(!bff) {return false;} rm=false; eok=true; while(tl<ta) { if(rm) { tb+=50; bff=(char)realloc(bff,sizeof(char)*tb); if(!bff) {eok=false;break;} rm=false; } ba=0; for(q=0;q<tb;q++) { if(feof(arx)!=0) {break;} bff[q]=getc(arx); tl++; ba++; } total=0; restants=ba; while(total<ba) { n=send(s,bff+total,restants,0); if(!rm && n==ba) { rm=true; } if(n==-1) {break;} total+=n; restants-=n; } } if(bff) {free(bff);} fclose(arx); return eok; } Is the same as before, but I deleted the '\0' at the end of the buffer to send. Maybe is not 100% optimized, but it works (for the moment). Niara
Niara is online now

10-26-2006, 02:25 PM	#6
vart CSharpener Join Date: Oct 2006 Posts: 5,325	ftell returns long - you put it into int. Not a good practice. And you don't check the resulting value for possible errors. Advice - use fread instead of the loop to fill the byte array from the file.
vart is offline

10-27-2006, 11:46 AM	#7
Niara Registered User Join Date: Mar 2005 Location: Juneda Posts: 229	Hey vart, thanks for your observation. I have already modifyed the code checking for the error return '-1' on the 'ftell()' function. I didn't realized that I was storing the file size on an 'int' (I have commented something similar: bad learning habits produce errors in such simple codes), I will have to correct that part. Niara
Niara is online now

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
char Handling, probably typical newbie stuff	Neolyth	C Programming	16	06-21-2009 04:05 AM
Adventures in labyrinth generation.	guesst	Game Programming	8	10-12-2008 01:30 PM
Client works on a LAN but don't send all the data	Niara	Networking/Device Communication	9	01-04-2007 04:44 PM
Error with a vector	Tropicalia	C++ Programming	20	09-28-2006 07:45 PM
AdjustWindowRect/Setting client size	JaWiB	Windows Programming	4	06-23-2006 11:43 PM