//
// Cforall Version 1.0.0 Copyright (C) 2020 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// iocall.cfa --
//
// Author           : Thierry Delisle
// Created On       : Wed Jul  1 14:51:00 2020
// Last Modified By :
// Last Modified On :
// Update Count     :
//

#include "bits/defs.hfa"

//=============================================================================================
// I/O uring backend
//=============================================================================================

#if defined(HAVE_LINUX_IO_URING_H)
	#include <stdint.h>
	#include <linux/io_uring.h>

	#include "kernel_private.hfa"

	extern [* struct io_uring_sqe, uint32_t] __submit_alloc( struct __io_data & ring, uint64_t data );
	extern void __submit( struct __io_data & ring, uint32_t idx );

	static inline void ?{}(struct io_uring_sqe & this, uint8_t opcode, int fd) {
		this.opcode = opcode;
		#if !defined(IOSQE_ASYNC)
			this.flags = 0;
		#else
			this.flags = IOSQE_ASYNC;
		#endif
		this.ioprio = 0;
		this.fd = fd;
		this.off = 0;
		this.addr = 0;
		this.len = 0;
		this.rw_flags = 0;
		this.__pad2[0] = this.__pad2[1] = this.__pad2[2] = 0;
	}

	static inline void ?{}(struct io_uring_sqe & this, uint8_t opcode, int fd, void * addr, uint32_t len, uint64_t off ) {
		(this){ opcode, fd };
		this.off = off;
		this.addr = (uint64_t)(uintptr_t)addr;
		this.len = len;
	}

	#define __submit_prelude \
		__io_user_data_t data = { 0, active_thread() }; \
		struct __io_data & ring = *data.thrd->curr_cluster->io; \
		struct io_uring_sqe * sqe; \
		uint32_t idx; \
		[sqe, idx] = __submit_alloc( ring, (uint64_t)(uintptr_t)&data );

	#define __submit_wait \
		/*__cfaabi_bits_print_safe( STDERR_FILENO, "Preparing user data %p for %p\n", &data, data.thrd );*/ \
		verify( sqe->user_data == (uint64_t)(uintptr_t)&data ); \
		__submit( ring, idx ); \
		park( __cfaabi_dbg_ctx ); \
		return data.result;
#endif

//=============================================================================================
// I/O Forwards
//=============================================================================================

// Some forward declarations
#include <unistd.h>

extern "C" {
	#include <sys/types.h>
	#include <sys/socket.h>
	#include <sys/syscall.h>

#if defined(HAVE_PREADV2)
	struct iovec;
	extern ssize_t preadv2 (int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags);
#endif
#if defined(HAVE_PWRITEV2)
	struct iovec;
	extern ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags);
#endif

	extern int fsync(int fd);
	extern int sync_file_range(int fd, int64_t offset, int64_t nbytes, unsigned int flags);

	struct msghdr;
	struct sockaddr;
	extern ssize_t sendmsg(int sockfd, const struct msghdr *msg, int flags);
	extern ssize_t recvmsg(int sockfd, struct msghdr *msg, int flags);
	extern ssize_t send(int sockfd, const void *buf, size_t len, int flags);
	extern ssize_t recv(int sockfd, void *buf, size_t len, int flags);
	extern int accept4(int sockfd, struct sockaddr *addr, socklen_t *addrlen, int flags);
	extern int connect(int sockfd, const struct sockaddr *addr, socklen_t addrlen);

	extern int fallocate(int fd, int mode, uint64_t offset, uint64_t len);
	extern int posix_fadvise(int fd, uint64_t offset, uint64_t len, int advice);
	extern int madvise(void *addr, size_t length, int advice);

	extern int openat(int dirfd, const char *pathname, int flags, mode_t mode);
	extern int close(int fd);

	extern ssize_t read (int fd, void *buf, size_t count);
}

//=============================================================================================
// I/O Interface
//=============================================================================================

//-----------------------------------------------------------------------------
// Asynchronous operations
#if defined(HAVE_PREADV2)
	ssize_t cfa_preadv2(int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags) {
		#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_READV)
			return preadv2(fd, iov, iovcnt, offset, flags);
		#else
			__submit_prelude

			(*sqe){ IORING_OP_READV, fd, iov, iovcnt, offset };

			__submit_wait
		#endif
	}
#endif

#if defined(HAVE_PWRITEV2)
	ssize_t cfa_pwritev2(int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags) {
		#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_WRITEV)
			return pwritev2(fd, iov, iovcnt, offset, flags);
		#else
			__submit_prelude

			(*sqe){ IORING_OP_WRITEV, fd, iov, iovcnt, offset };

			__submit_wait
		#endif
	}
#endif

int cfa_fsync(int fd) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_FSYNC)
		return fsync(fd);
	#else
		__submit_prelude

		(*sqe){ IORING_OP_FSYNC, fd };

		__submit_wait
	#endif
}

int cfa_sync_file_range(int fd, int64_t offset, int64_t nbytes, unsigned int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_SYNC_FILE_RANGE)
		return sync_file_range(fd, offset, nbytes, flags);
	#else
		__submit_prelude

		(*sqe){ IORING_OP_SYNC_FILE_RANGE, fd };
		sqe->off = offset;
		sqe->len = nbytes;
		sqe->sync_range_flags = flags;

		__submit_wait
	#endif
}


ssize_t cfa_sendmsg(int sockfd, const struct msghdr *msg, int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_SENDMSG)
		return sendmsg(sockfd, msg, flags);
	#else
		__submit_prelude

		(*sqe){ IORING_OP_SENDMSG, sockfd, msg, 1, 0 };
		sqe->msg_flags = flags;

		__submit_wait
	#endif
}

ssize_t cfa_recvmsg(int sockfd, struct msghdr *msg, int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_RECVMSG)
		return recvmsg(sockfd, msg, flags);
	#else
		__submit_prelude

		(*sqe){ IORING_OP_RECVMSG, sockfd, msg, 1, 0 };
		sqe->msg_flags = flags;

		__submit_wait
	#endif
}

ssize_t cfa_send(int sockfd, const void *buf, size_t len, int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_SEND)
		return send( sockfd, buf, len, flags );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_SEND, sockfd };
		sqe->addr = (uint64_t)buf;
		sqe->len = len;
		sqe->msg_flags = flags;

		__submit_wait
	#endif
}

ssize_t cfa_recv(int sockfd, void *buf, size_t len, int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_RECV)
		return recv( sockfd, buf, len, flags );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_RECV, sockfd };
		sqe->addr = (uint64_t)buf;
		sqe->len = len;
		sqe->msg_flags = flags;

		__submit_wait
	#endif
}

int cfa_accept4(int sockfd, struct sockaddr *addr, socklen_t *addrlen, int flags) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_ACCEPT)
		return accept4( sockfd, addr, addrlen, flags );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_ACCEPT, sockfd };
		sqe->addr = addr;
		sqe->addr2 = addrlen;
		sqe->accept_flags = flags;

		__submit_wait
	#endif
}

int cfa_connect(int sockfd, const struct sockaddr *addr, socklen_t addrlen) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_CONNECT)
		return connect( sockfd, addr, addrlen );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_CONNECT, sockfd };
		sqe->addr = (uint64_t)addr;
		sqe->off = addrlen;

		__submit_wait
	#endif
}

int cfa_fallocate(int fd, int mode, uint64_t offset, uint64_t len) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_FALLOCATE)
		return fallocate( fd, mode, offset, len );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_FALLOCATE, fd };
		sqe->off = offset;
		sqe->len = length;
		sqe->mode = mode;

		__submit_wait
	#endif
}

int cfa_fadvise(int fd, uint64_t offset, uint64_t len, int advice) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_FADVISE)
		return posix_fadvise( fd, offset, len, advice );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_FADVISE, fd };
		sqe->off = (uint64_t)offset;
		sqe->len = length;
		sqe->fadvise_advice = advice;

		__submit_wait
	#endif
}

int cfa_madvise(void *addr, size_t length, int advice) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_MADVISE)
		return madvise( addr, length, advice );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_MADVISE, 0 };
		sqe->addr = (uint64_t)addr;
		sqe->len = length;
		sqe->fadvise_advice = advice;

		__submit_wait
	#endif
}

int cfa_openat(int dirfd, const char *pathname, int flags, mode_t mode) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_OPENAT)
		return openat( dirfd, pathname, flags, mode );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_OPENAT, dirfd };
		sqe->addr = (uint64_t)pathname;
		sqe->open_flags = flags;
		sqe->mode = mode;

		__submit_wait
	#endif
}

int cfa_close(int fd) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_CLOSE)
		return close( fd );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_CLOSE, fd };

		__submit_wait
	#endif
}


ssize_t cfa_read(int fd, void *buf, size_t count) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_READ)
		return read( fd, buf, count );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_READ, fd, buf, count, 0 };

		__submit_wait
	#endif
}

ssize_t cfa_write(int fd, void *buf, size_t count) {
	#if !defined(HAVE_LINUX_IO_URING_H) || !defined(IORING_OP_WRITE)
		return read( fd, buf, count );
	#else
		__submit_prelude

		(*sqe){ IORING_OP_WRITE, fd, buf, count, 0 };

		__submit_wait
	#endif
}

//-----------------------------------------------------------------------------
// Check if a function is asynchronous

// Macro magic to reduce the size of the following switch case
#define IS_DEFINED_APPLY(f, ...) f(__VA_ARGS__)
#define IS_DEFINED_SECOND(first, second, ...) second
#define IS_DEFINED_TEST(expansion) _CFA_IO_FEATURE_##expansion
#define IS_DEFINED(macro) IS_DEFINED_APPLY( IS_DEFINED_SECOND,IS_DEFINED_TEST(macro) false, true)

bool has_user_level_blocking( fptr_t func ) {
	#if defined(HAVE_LINUX_IO_URING_H)
		#if defined(HAVE_PREADV2)
			if( /*func == (fptr_t)preadv2 || */
				func == (fptr_t)cfa_preadv2 )
				#define _CFA_IO_FEATURE_IORING_OP_READV ,
				return IS_DEFINED(IORING_OP_READV);
		#endif

		#if defined(HAVE_PWRITEV2)
			if( /*func == (fptr_t)pwritev2 || */
				func == (fptr_t)cfa_pwritev2 )
				#define _CFA_IO_FEATURE_IORING_OP_WRITEV ,
				return IS_DEFINED(IORING_OP_WRITEV);
		#endif

		if( /*func == (fptr_t)fsync || */
			func == (fptr_t)cfa_fsync )
			#define _CFA_IO_FEATURE_IORING_OP_FSYNC ,
			return IS_DEFINED(IORING_OP_FSYNC);

		if( /*func == (fptr_t)ync_file_range || */
			func == (fptr_t)cfa_sync_file_range )
			#define _CFA_IO_FEATURE_IORING_OP_SYNC_FILE_RANGE ,
			return IS_DEFINED(IORING_OP_SYNC_FILE_RANGE);

		if( /*func == (fptr_t)sendmsg || */
			func == (fptr_t)cfa_sendmsg )
			#define _CFA_IO_FEATURE_IORING_OP_SENDMSG ,
			return IS_DEFINED(IORING_OP_SENDMSG);

		if( /*func == (fptr_t)recvmsg || */
			func == (fptr_t)cfa_recvmsg )
			#define _CFA_IO_FEATURE_IORING_OP_RECVMSG ,
			return IS_DEFINED(IORING_OP_RECVMSG);

		if( /*func == (fptr_t)send || */
			func == (fptr_t)cfa_send )
			#define _CFA_IO_FEATURE_IORING_OP_SEND ,
			return IS_DEFINED(IORING_OP_SEND);

		if( /*func == (fptr_t)recv || */
			func == (fptr_t)cfa_recv )
			#define _CFA_IO_FEATURE_IORING_OP_RECV ,
			return IS_DEFINED(IORING_OP_RECV);

		if( /*func == (fptr_t)accept4 || */
			func == (fptr_t)cfa_accept4 )
			#define _CFA_IO_FEATURE_IORING_OP_ACCEPT ,
			return IS_DEFINED(IORING_OP_ACCEPT);

		if( /*func == (fptr_t)connect || */
			func == (fptr_t)cfa_connect )
			#define _CFA_IO_FEATURE_IORING_OP_CONNECT ,
			return IS_DEFINED(IORING_OP_CONNECT);

		if( /*func == (fptr_t)fallocate || */
			func == (fptr_t)cfa_fallocate )
			#define _CFA_IO_FEATURE_IORING_OP_FALLOCATE ,
			return IS_DEFINED(IORING_OP_FALLOCATE);

		if( /*func == (fptr_t)posix_fadvise || */
			func == (fptr_t)cfa_fadvise )
			#define _CFA_IO_FEATURE_IORING_OP_FADVISE ,
			return IS_DEFINED(IORING_OP_FADVISE);

		if( /*func == (fptr_t)madvise || */
			func == (fptr_t)cfa_madvise )
			#define _CFA_IO_FEATURE_IORING_OP_MADVISE ,
			return IS_DEFINED(IORING_OP_MADVISE);

		if( /*func == (fptr_t)openat || */
			func == (fptr_t)cfa_openat )
			#define _CFA_IO_FEATURE_IORING_OP_OPENAT ,
			return IS_DEFINED(IORING_OP_OPENAT);

		if( /*func == (fptr_t)close || */
			func == (fptr_t)cfa_close )
			#define _CFA_IO_FEATURE_IORING_OP_CLOSE ,
			return IS_DEFINED(IORING_OP_CLOSE);

		if( /*func == (fptr_t)read || */
			func == (fptr_t)cfa_read )
			#define _CFA_IO_FEATURE_IORING_OP_READ ,
			return IS_DEFINED(IORING_OP_READ);

		if( /*func == (fptr_t)write || */
			func == (fptr_t)cfa_write )
			#define _CFA_IO_FEATURE_IORING_OP_WRITE ,
			return IS_DEFINED(IORING_OP_WRITE);
	#endif

	return false;
}