I initially wanted to see if I could make a faster (and easier to use) tool for flashing drives than dd.
In order to acheive this, we use the sendfile64 syscall, which allows for copying data from one file descriptor to another completely in kernel space.
This is opposed to calling read and write repeatedly, which copies data in and out of user space.
As of writing this, when copying a file to a block device, both cp and dd employ the method of repeatedly calling read and write.
You can also use this as a general replacement for cp in cases where cp is otherwise inefficent (run your own benchmarks!)
In my personal (non-rigorous) benchmarks, it seems to match dd in terms of time spent (when an optimal block size is found and used). Even so, I consider this project a success, as I'm able to get good speeds without messing with dd options (and finding an optimal blocksize), and theoretically not wasting resources on context switching. I want to collect more data, so please open issues with your results :)
zcp <source> <destination>Initial inspiration from: https://jvns.ca/blog/2016/01/23/sendfile-a-new-to-me-system-call/ https://blog.plenz.com/2014-04/so-you-want-to-write-to-a-file-real-fast.html
ioctl for getting block device size in Rust: http://syhpoon.ca/posts/how-to-get-block-device-size-on-linux-with-rust/