best way to send files over socket

What is the best way to send files over a socket? How is it actually done?

You use some kind of protocol, so that you can prepend information about the file. This can be as elaborate as you want, but at minimum you probably need the length of the file. That can simply be a 4 byte integer. So the recipient reads four bytes into an int, then uses that to determine how much more to read before the file is complete. You could also do something like 4 bytes for the total length, then a null terminated string containing the file name (include that in the length), then the file.

Files on the WWW use the http protocol, which is a text header ending with "\r\n\r\n" (two MS-DOS style newlines).

I guess i could use something like this to find the size of the file;

Code:

---

fseek(file, 0, SEEK_END);
int size = ftell(file);

---

I could also use that to keep track of how much data has been sent or read too i assume.
Why only 4 bytes MK27? Couldnt i use 1MB? Or is that too much?

4 bytes, as an unsigned integer, can encode file sizes upto 4GB.
The 4 bytes is the amount of data that follows, not the data itself.

Like you would send
00 00 00 06 (4 byte length, = 6)

Then
"hello\0"
are the 6 bytes of data.

Quote:

---

Originally Posted by Annonymous

I guess i could use something like this to find the size of the file;

Code:

---

fseek(file, 0, SEEK_END);
int size = ftell(file);

---

---

Yeah, that's a portable way. Using stat() is probably faster, but stat() is *nix specific.

Quote:

---

I could also use that to keep track of how much data has been sent or read too i assume.

---

No, write/send and read/recv all return the number of bytes done, so you can keep count that way. The issue is that you cannot count on reading the file all at once, because the underlying tcp/ip layer uses packets and those packets may be delayed. Ie, if you just try and read as much as you can, there is a decent chance you'll only actually get part of the file (issues like this are hard to notice on local loopback, but flare up when you start testing with separate computers, esp. if they are on a large network such the internet).

So you need to know how big the file is in advance, then read it in a loop until until done.

Quote:

---

Why only 4 bytes MK27? Couldnt i use 1MB? Or is that too much?

---

The idea with the 4 byte "header" is it is just an int, altho since ints aren't necessarily 4 bytes, you have to use a more specific type:

Code:

---

#include <sys/types.h>
int32_t size = (int32_t)ftell(file);
// 32 bits == 4 bytes

---

Now:

Code:

---

write(sockfd, &size, 4);
write(sockfd, filebuffer, (size_t)size));
// or you could read and write the file in chunks adding up to the correct size

---

Make sense? And the recipient:

Code:

---

int32_t len;
read(sockfd, &len, 4);

---

Now "len" contains the size of the file to be read in.

I suppose a u_int32_t (unsigned) is actually better, as Salem implies, since negative numbers make no sense here and an unsigned type has twice the maximum value.

is there a certain way to declare those unsigned ints?

Some are in <sys/types.h> but I think the most portable place by the standard is <inttypes.h>. That should have int32_t and uint32_t.