path: root/sys/linux/io_uring.txt

# Copyright 2019 syzkaller project authors. All rights reserved.
# Use of this source code is governed by Apache 2 LICENSE that can be found in the LICENSE file.

# See http://kernel.dk/io_uring.pdf

include <uapi/linux/io_uring.h>
# For EPOLL_CTL_ADD, EPOLL_CTL_MOD, EPOLL_CTL_DEL
include <uapi/linux/eventpoll.h>

resource fd_io_uring[fd]
resource ring_ptr[int64]
resource sqes_ptr[int64]
resource ioring_personality_id[int16]

# fs/io_uring.c
define IORING_MAX_ENTRIES	32768
define IORING_MAX_CQ_ENTRIES	(2 * IORING_MAX_ENTRIES)

# First does the setup calling io_uring_setup, than calls mmap to map the ring and
# the sqes. It is hard for the fuzzer to generate correct programs using mmap calls
# with fuzzer-provided mmap length. This wrapper ensures correct length computation.
syz_io_uring_setup(entries int32[1:IORING_MAX_ENTRIES], params ptr[inout, io_uring_params], ring_ptr ptr[out, ring_ptr], sqes_ptr ptr[out, sqes_ptr]) fd_io_uring

io_uring_setup(entries int32[1:IORING_MAX_ENTRIES], params ptr[inout, io_uring_params]) fd_io_uring
io_uring_enter(fd fd_io_uring, to_submit int32[0:IORING_MAX_ENTRIES], min_complete int32[0:IORING_MAX_CQ_ENTRIES], flags flags[io_uring_enter_flags], sigmask ptr[in, sigset_t], size len[sigmask])
io_uring_register$IORING_REGISTER_BUFFERS(fd fd_io_uring, opcode const[IORING_REGISTER_BUFFERS], arg ptr[in, array[iovec_out]], nr_args len[arg])
io_uring_register$IORING_UNREGISTER_BUFFERS(fd fd_io_uring, opcode const[IORING_UNREGISTER_BUFFERS], arg const[0], nr_args const[0])
io_uring_register$IORING_REGISTER_FILES(fd fd_io_uring, opcode const[IORING_REGISTER_FILES], arg ptr[in, array[fd]], nr_args len[arg])
io_uring_register$IORING_UNREGISTER_FILES(fd fd_io_uring, opcode const[IORING_UNREGISTER_FILES], arg const[0], nr_args const[0])
io_uring_register$IORING_REGISTER_EVENTFD(fd fd_io_uring, opcode const[IORING_REGISTER_EVENTFD], arg ptr[in, fd_event], nr_args const[1])
io_uring_register$IORING_UNREGISTER_EVENTFD(fd fd_io_uring, opcode const[IORING_UNREGISTER_EVENTFD], arg const[0], nr_args const[0])
io_uring_register$IORING_REGISTER_FILES_UPDATE(fd fd_io_uring, opcode const[IORING_REGISTER_FILES_UPDATE], arg ptr[in, io_uring_files_update], nr_args len[arg:fds])
io_uring_register$IORING_REGISTER_EVENTFD_ASYNC(fd fd_io_uring, opcode const[IORING_REGISTER_EVENTFD_ASYNC], arg ptr[in, fd_event], nr_args const[1])
io_uring_register$IORING_REGISTER_PROBE(fd fd_io_uring, opcode const[IORING_REGISTER_PROBE], arg ptr[inout, io_uring_probe], nr_args len[arg:ops])
io_uring_register$IORING_REGISTER_PERSONALITY(fd fd_io_uring, opcode const[IORING_REGISTER_PERSONALITY], arg const[0], nr_args const[0]) ioring_personality_id
io_uring_register$IORING_UNREGISTER_PERSONALITY(fd fd_io_uring, opcode const[IORING_UNREGISTER_PERSONALITY], arg const[0], nr_args ioring_personality_id)
# IORING_REGISTER_EVENTFD, IORING_UNREGISTER_EVENTFD >= 5.2
# IORING_REGISTER_FILES_UPDATE >= 5.5
# IORING_REGISTER_EVENTFD_ASYNC, IORING_REGISTER_PROBE, IORING_REGISTER_PERSONALITY, IORING_UNREGISTER_PERSONALITY>= 5.6

io_uring_register$IORING_REGISTER_ENABLE_RINGS(fd fd_io_uring, opcode const[IORING_REGISTER_ENABLE_RINGS], arg const[0], nr_args const[0])
io_uring_register$IORING_REGISTER_RESTRICTIONS(fd fd_io_uring, opcode const[IORING_REGISTER_RESTRICTIONS], arg ptr[in, array[io_uring_restriction_st]], nr_args len[arg])
# IORING_REGISTER_ENABLE_RINGS, IORING_REGISTER_RESTRICTIONS >= 5.10
io_uring_register$IORING_REGISTER_BUFFERS2(fd fd_io_uring, opcode const[IORING_REGISTER_BUFFERS2], arg ptr[in, io_uring_rsrc_register], size bytesize[arg])
io_uring_register$IORING_REGISTER_BUFFERS_UPDATE(fd fd_io_uring, opcode const[IORING_REGISTER_BUFFERS_UPDATE], arg ptr[in, io_uring_rsrc_update2], size bytesize[arg])
io_uring_register$IORING_REGISTER_FILES2(fd fd_io_uring, opcode const[IORING_REGISTER_FILES2], arg ptr[in, io_uring_rsrc_register], size bytesize[arg])
io_uring_register$IORING_REGISTER_FILES_UPDATE2(fd fd_io_uring, opcode const[IORING_REGISTER_FILES_UPDATE2], arg ptr[in, io_uring_rsrc_update2], size bytesize[arg])
# IORING_REGISTER_BUFFERS2, IORING_REGISTER_BUFFERS_UPDATE, IORING_REGISTER_FILES2, IORING_REGISTER_FILES_UPDATE2 >= 5.13
io_uring_register$IORING_REGISTER_IOWQ_AFF(fd fd_io_uring, opcode const[IORING_REGISTER_IOWQ_AFF], arg ptr[in, array[int8]], size bytesize[arg])
io_uring_register$IORING_UNREGISTER_IOWQ_AFF(fd fd_io_uring, opcode const[IORING_UNREGISTER_IOWQ_AFF], arg const[0], nr_args const[0])
# IORING_REGISTER_IOWQ_AFF, IORING_UNREGISTER_IOWQ_AFF >= 5.14
io_uring_register$IORING_REGISTER_IOWQ_MAX_WORKERS(fd fd_io_uring, opcode const[IORING_REGISTER_IOWQ_MAX_WORKERS], arg ptr[in, array[int32, 2]], nr_args const[2])
# IORING_REGISTER_IOWQ_MAX_WORKERS >= 5.15
io_uring_register$IORING_REGISTER_RING_FDS(fd fd_io_uring, opcode const[IORING_REGISTER_RING_FDS], arg ptr[in, array[io_uring_rsrc_register]], nr_args len[arg])
io_uring_register$IORING_UNREGISTER_RING_FDS(fd fd_io_uring, opcode const[IORING_UNREGISTER_RING_FDS], arg ptr[in, array[io_uring_rsrc_register]], nr_args len[arg])
# IORING_REGISTER_RING_FDS, IORING_UNREGISTER_RING_FDS >= 5.18
io_uring_register$IORING_REGISTER_PBUF_RING(fd fd_io_uring, opcode const[IORING_REGISTER_PBUF_RING], arg ptr[in, io_uring_buf_reg], nr_args const[1])
io_uring_register$IORING_UNREGISTER_PBUF_RING(fd fd_io_uring, opcode const[IORING_UNREGISTER_PBUF_RING], arg ptr[in, io_uring_buf_reg], nr_args const[1])
# IORING_REGISTER_PBUF_RING, IORING_UNREGISTER_PBUF_RING >= 5.19
io_uring_register$IORING_REGISTER_SYNC_CANCEL(fd fd_io_uring, opcode const[IORING_REGISTER_SYNC_CANCEL], arg ptr[in, io_uring_sync_cancel_reg], nr_args const[1])
io_uring_register$IORING_REGISTER_FILE_ALLOC_RANGE(fd fd_io_uring, opcode const[IORING_REGISTER_FILE_ALLOC_RANGE], arg ptr[in, io_uring_file_index_range], nr_args const[0])
io_uring_register$IORING_REGISTER_PBUF_STATUS(fd fd_io_uring, opcode const[IORING_REGISTER_PBUF_STATUS], arg ptr[in, io_uring_buf_status], nr_args const[1])
io_uring_register$IORING_REGISTER_NAPI(fd fd_io_uring, opcode const[IORING_REGISTER_NAPI], arg ptr[in, io_uring_napi], nr_args const[1])
io_uring_register$IORING_UNREGISTER_NAPI(fd fd_io_uring, opcode const[IORING_UNREGISTER_NAPI], arg ptr[out, io_uring_napi], nr_args const[1])
io_uring_register$IORING_REGISTER_CLOCK(fd fd_io_uring, opcode const[IORING_REGISTER_CLOCK], arg ptr[in, io_uring_clock_register], nr_args const[0])
io_uring_register$IORING_REGISTER_CLONE_BUFFERS(fd fd_io_uring, opcode const[IORING_REGISTER_CLONE_BUFFERS], arg ptr[in, io_uring_clone_buffers], nr_args const[1])
io_uring_register$IORING_REGISTER_RESIZE_RINGS(fd fd_io_uring, opcode const[IORING_REGISTER_RESIZE_RINGS], arg ptr[in, io_uring_params], nr_args const[1])
io_uring_register$IORING_REGISTER_MEM_REGION(fd fd_io_uring, opcode const[IORING_REGISTER_MEM_REGION], arg ptr[in, io_uring_mem_region_reg], nr_args const[1])

io_uring_register_opcodes = IORING_REGISTER_BUFFERS, IORING_UNREGISTER_BUFFERS, IORING_REGISTER_FILES, IORING_UNREGISTER_FILES, IORING_REGISTER_EVENTFD, IORING_UNREGISTER_EVENTFD, IORING_REGISTER_FILES_UPDATE, IORING_REGISTER_EVENTFD_ASYNC, IORING_REGISTER_PROBE, IORING_REGISTER_PERSONALITY, IORING_UNREGISTER_PERSONALITY, IORING_REGISTER_RESTRICTIONS, IORING_REGISTER_ENABLE_RINGS, IORING_REGISTER_FILES2, IORING_REGISTER_FILES_UPDATE2, IORING_REGISTER_BUFFERS2, IORING_REGISTER_BUFFERS_UPDATE, IORING_REGISTER_IOWQ_AFF, IORING_UNREGISTER_IOWQ_AFF, IORING_REGISTER_IOWQ_MAX_WORKERS, IORING_REGISTER_RING_FDS, IORING_UNREGISTER_RING_FDS, IORING_REGISTER_PBUF_RING, IORING_UNREGISTER_PBUF_RING, IORING_REGISTER_SYNC_CANCEL, IORING_REGISTER_FILE_ALLOC_RANGE, IORING_REGISTER_PBUF_STATUS, IORING_REGISTER_NAPI, IORING_UNREGISTER_NAPI, IORING_REGISTER_CLOCK, IORING_REGISTER_CLONE_BUFFERS, IORING_REGISTER_SEND_MSG_RING, IORING_REGISTER_RESIZE_RINGS, IORING_REGISTER_MEM_REGION

# The mmap'ed area for SQ and CQ rings are really the same -- the difference is
# accounted for with the usage of offsets.
mmap$IORING_OFF_SQ_RING(addr vma, len len[addr], prot flags[mmap_prot], flags flags[mmap_flags], fd fd_io_uring, offset const[IORING_OFF_SQ_RING]) ring_ptr
mmap$IORING_OFF_CQ_RING(addr vma, len len[addr], prot flags[mmap_prot], flags flags[mmap_flags], fd fd_io_uring, offset const[IORING_OFF_CQ_RING]) ring_ptr
mmap$IORING_OFF_SQES(addr vma, len len[addr], prot flags[mmap_prot], flags flags[mmap_flags], fd fd_io_uring, offset const[IORING_OFF_SQES]) sqes_ptr

# If no flags are specified(0), the io_uring instance is setup for interrupt driven IO.
io_uring_setup_flags = 0, IORING_SETUP_IOPOLL, IORING_SETUP_SQPOLL, IORING_SETUP_SQ_AFF, IORING_SETUP_CQSIZE, IORING_SETUP_CLAMP, IORING_SETUP_ATTACH_WQ, IORING_FEAT_SINGLE_MMAP, IORING_FEAT_NODROP, IORING_FEAT_SUBMIT_STABLE, IORING_FEAT_RW_CUR_POS, IORING_FEAT_FAST_POLL, IORING_FEAT_POLL_32BITS, IORING_SETUP_R_DISABLED, IORING_FEAT_SQPOLL_NONFIXED, IORING_FEAT_NATIVE_WORKERS, IORING_FEAT_RSRC_TAGS, IORING_FEAT_CQE_SKIP, IORING_FEAT_LINKED_FILE, IORING_FEAT_REG_REG_RING, IORING_FEAT_RECVSEND_BUNDLE, IORING_FEAT_MIN_TIMEOUT, IORING_SETUP_SUBMIT_ALL, IORING_SETUP_COOP_TASKRUN, IORING_SETUP_TASKRUN_FLAG, IORING_SETUP_SQE128, IORING_SETUP_CQE32, IORING_SETUP_SINGLE_ISSUER, IORING_SETUP_DEFER_TASKRUN, IORING_SETUP_NO_MMAP, IORING_SETUP_REGISTERED_FD_ONLY, IORING_SETUP_NO_SQARRAY, IORING_SETUP_HYBRID_IOPOLL
# watch out the being tested kernel version
# IORING_FEAT_SINGLE_MMAP >= 5.4
# IORING_FEAT_NODROP, IORING_FEAT_SUBMIT_STABLE >= 5.5
# IORING_FEAT_RW_CUR_POS >= 5.6
# IORING_FEAT_FAST_POLL >= 5.7
# IORING_FEAT_POLL_32BITS >= 5.9
# IORING_SETUP_R_DISABLED >= 5.10 (this shoule be used with IORING_REGISTER_ENABLE_RINGS)
# IORING_FEAT_SQPOLL_NONFIXED >= 5.11
# IORING_FEAT_NATIVE_WORKERS >= 5.12
# IORING_FEAT_RSRC_TAGS >= 5.13
# IORING_FEAT_CQE_SKIP >= 5.17
# IORING_SETUP_SUBMIT_ALL >= 5.18
# IORING_SETUP_COOP_TASKRUN, IORING_SETUP_TASKRUN_FLAG, IORING_SETUP_SQE128, IORING_SETUP_CQE32 >= 5.19
# IORING_SETUP_SINGLE_ISSUER >= 6.0
# IORING_SETUP_DEFER_TASKRUN >= 6.1

io_uring_enter_flags = IORING_ENTER_GETEVENTS, IORING_ENTER_SQ_WAKEUP, IORING_ENTER_SQ_WAIT, IORING_ENTER_EXT_ARG, IORING_ENTER_REGISTERED_RING, IORING_ENTER_ABS_TIMER, IORING_ENTER_EXT_ARG_REG
# IORING_ENTER_EXT_ARG >= 5.11
_ = __NR_mmap2

# Once an io_uring is set up by calling io_uring_setup, the offsets to the member fields
# to be used on the mmap'ed area are set in structs io_sqring_offsets and io_cqring_offsets.
# Except io_sqring_offsets.array, the offsets are static while all depend on how struct io_rings
# is organized in code. The offsets can be marked as resources in syzkaller descriptions but
# this makes it difficult to generate correct programs by the fuzzer. Thus, the offsets are
# hard-coded here (and in the executor).
define SQ_HEAD_OFFSET	0
define SQ_TAIL_OFFSET	64
define SQ_RING_MASK_OFFSET	256
define SQ_RING_ENTRIES_OFFSET	264
define SQ_FLAGS_OFFSET	276
define SQ_DROPPED_OFFSET	272
define CQ_HEAD_OFFSET	128
define CQ_TAIL_OFFSET	192
define CQ_RING_MASK_OFFSET	260
define CQ_RING_ENTRIES_OFFSET	268
define CQ_RING_OVERFLOW_OFFSET	284
define CQ_FLAGS_OFFSET	280

# Notice all offsets are pointing to uint32 values. This is assumed for the
io_uring_offsets = SQ_HEAD_OFFSET, SQ_TAIL_OFFSET, SQ_RING_MASK_OFFSET, SQ_RING_ENTRIES_OFFSET, SQ_DROPPED_OFFSET, CQ_HEAD_OFFSET, CQ_TAIL_OFFSET, CQ_RING_MASK_OFFSET, CQ_RING_ENTRIES_OFFSET, CQ_RING_OVERFLOW_OFFSET, io_uring_flags_offsets

# Also, all values are int32, thus, set nbytes to 4.
syz_memcpy_off$IO_URING_METADATA_GENERIC(ring_ptr ring_ptr, off flags[io_uring_offsets], src ptr[in, int32], src_off const[0], nbytes const[4])

# The flags available are: IORING_SQ_NEED_WAKEUP (1) for sq, IORING_CQ_EVENTFD_DISABLED (1) for cq. Use int32[0:1] to represent possible values.
io_uring_flags_offsets = SQ_FLAGS_OFFSET, CQ_FLAGS_OFFSET
syz_memcpy_off$IO_URING_METADATA_FLAGS(ring_ptr ring_ptr, flag_off flags[io_uring_flags_offsets], src ptr[in, int32[0:1]], src_off const[0], nbytes const[4])

io_uring_probe {
	last_op	const[0, int8]
	ops_len	const[0, int8]
	resv	const[0, int16]
	resv2	array[const[0, int32], 3]
	ops	array[io_uring_probe_op, 0:IORING_OP_LAST]
}

io_uring_probe_op {
	op	const[0, int8]
	resv	const[0, int8]
	flags	const[0, int16]
	resv2	const[0, int32]
}

io_uring_files_update {
	offset	int32
	resv	const[0, int32]
	fds	ptr64[in, array[fd]]
}

#
# type template for io_uring_restriction
#

type io_uring_restriction[OPCODE, OPARG] {
	op	const[OPCODE, int16]
	oparg	OPARG
	resv	const[0, int8]
	resv2	array[const[0, int32], 3]
}

io_uring_restriction_st [
	ioring_restriction_register_op		io_uring_restriction[IORING_RESTRICTION_REGISTER_OP, flags[io_uring_register_opcodes, int8]]
	ioring_restriction_sqe_op		io_uring_restriction[IORING_RESTRICTION_SQE_OP, flags[io_uring_register_opcodes, int8]]
	ioring_restriction_sqe_flags_allowed	io_uring_restriction[IORING_RESTRICTION_SQE_FLAGS_ALLOWED, flags[io_uring_register_opcodes, int8]]
	ioring_restriction_sqe_flags_required	io_uring_restriction[IORING_RESTRICTION_SQE_FLAGS_REQUIRED, flags[io_uring_register_opcodes, int8]]
]

io_uring_rsrc_flags = IORING_RSRC_REGISTER_SPARSE

io_uring_rsrc_register {
	nr	len[data, int32]
	flags	flags[io_uring_rsrc_flags, int32]
	resv2	const[0, int64]
	data	ptr64[in, array[iovec_out]]
	tags	ptr64[in, array[int64]]
}

io_uring_rsrc_update2 {
	offset	int32
	resv	const[0, int32]
	data	ptr64[in, array[iovec_out]]
	tags	ptr64[in, array[int64]]
	nr	len[data, int32]
	resv2	const[0, int32]
}

io_uring_buf {
	addr	ptr64[in, array[int8]]
	len	len[addr, int32]
	bid	io_uring_bid[int16]
	resv	const[0, int16]
}

io_uring_buf_array {
	data	array[io_uring_buf]
} [align[4096]]

io_uring_buf_reg {
	ring_addr	ptr64[in, io_uring_buf_array]
	ring_entries	len[ring_addr:data, int32]
	bgid		io_uring_bgid[int16]	(in)
# IOU_PBUF_RING_MMAP not supported yet
	flags		const[0, int16]
	resv		array[const[0, int64], 3]
}

io_uring_params {
	sq_entries	int32	(out)
	cq_entries	int32[0:IORING_MAX_CQ_ENTRIES]	(inout)
	flags		flags[io_uring_setup_flags, int32]	(in)
	sq_thread_cpu	int32[0:3]	(in)
	sq_thread_idle	int32[0:1000]	(in)
	features	int32	(out)
	wq_fd		fd_io_uring[opt]	(in)
	resv		array[const[0, int32], 3]
# We don't really use them (they are hard-coded). Thus, just pass some memory region of their size.
# TODO: Now that per-field directions is supported, can we avoid using hard-coded values for offsets?
	sq_off		array[int32, 10]	(out)
	cq_off		array[int32, 10]	(out)
}

# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# Descriptions for sq_ring and cq_ring manipulation # # # # # # # # # # # # # #
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #

# Retrieve the cqe at the head of the cq_ring and advance the head. The only meaningful
# resource contained within a cqe is by the completion of openat or openat2 calls,
# which produce fd. If that is the case, returns the return value of those. Otherwise,
# for other operations, returns an invalid fd (-1).
syz_io_uring_complete(ring_ptr ring_ptr) fd

# Submit sqe into the sq_ring
syz_io_uring_submit(ring_ptr ring_ptr, sqes_ptr sqes_ptr, sqe ptr[in, io_uring_sqe_u])

io_uring_sqe_u [
	IORING_OP_NOP			io_uring_sqe$nop
	IORING_OP_READV			io_uring_sqe_readv
	IORING_OP_WRITEV		io_uring_sqe$writev
	IORING_OP_FSYNC			io_uring_sqe$fsync
	IORING_OP_READ_FIXED		io_uring_sqe$read_fixed
	IORING_OP_WRITE_FIXED		io_uring_sqe$write_fixed
	IORING_OP_POLL_ADD		io_uring_sqe$poll_add
	IORING_OP_POLL_REMOVE		io_uring_sqe$poll_remove
	IORING_OP_SYNC_FILE_RANGE	io_uring_sqe$sync_file_range
	IORING_OP_SENDMSG		io_uring_sqe$sendmsg
	IORING_OP_RECVMSG		io_uring_sqe$recvmsg
	IORING_OP_TIMEOUT		io_uring_sqe$timeout
	IORING_OP_TIMEOUT_REMOVE	io_uring_sqe$timeout_remove
	IORING_OP_ACCEPT		io_uring_sqe$accept
	IORING_OP_ASYNC_CANCEL		io_uring_sqe$async_cancel
	IORING_OP_LINK_TIMEOUT		io_uring_sqe$link_timeout
	IORING_OP_CONNECT		io_uring_sqe$connect
	IORING_OP_FALLOCATE		io_uring_sqe$fallocate
	IORING_OP_OPENAT		io_uring_sqe$openat
	IORING_OP_CLOSE			io_uring_sqe$close
	IORING_OP_FILES_UPDATE		io_uring_sqe$files_update
	IORING_OP_STATX			io_uring_sqe$statx
	IORING_OP_READ			io_uring_sqe_read
	IORING_OP_WRITE			io_uring_sqe$write
	IORING_OP_FADVISE		io_uring_sqe$fadvise
	IORING_OP_MADVISE		io_uring_sqe$madvise
	IORING_OP_SEND			io_uring_sqe$send
	IORING_OP_RECV			io_uring_sqe_recv
	IORING_OP_OPENAT2		io_uring_sqe$openat2
	IORING_OP_EPOLL_CTL		io_uring_sqe_epoll_ctl
	IORING_OP_SPLICE		io_uring_sqe$splice
	IORING_OP_PROVIDE_BUFFERS	io_uring_sqe$provide_buffers
	IORING_OP_REMOVE_BUFFERS	io_uring_sqe$remove_buffers
	IORING_OP_TEE			io_uring_sqe$tee
	IORING_OP_SHUTDOWN		io_uring_sqe$shutdown
	IORING_OP_RENAMEAT		io_uring_sqe$renameat
	IORING_OP_UNLINKAT		io_uring_sqe$unlinkat
	IORING_OP_MKDIRAT		io_uring_sqe$mkdirat
	IORING_OP_SYMLINKAT		io_uring_sqe$symlinkat
	IORING_OP_LINKAT		io_uring_sqe$linkat
	IORING_OP_MSG_RING		io_uring_sqe$msg_ring
# 	TODO undocumented 6.0
#	IORING_OP_FSETXATTR		io_uring_sqe$fsetxatt
#	IORING_OP_SETXATTR		io_uring_sqe$setxatt
#	IORING_OP_FGETXATTR		io_uring_sqe$fgetxatt
#	IORING_OP_GETXATTR		io_uring_sqe$etxatt
#	IORING_OP_SOCKET		io_uring_sqe$socket
#	IORING_OP_URING_CMD		io_uring_sqe$uring_cmd
#	IORING_OP_SEND_ZC		io_uring_sqe$send_zc
]

# io_uring_enter_opcodes = IORING_OP_NOP, IORING_OP_READV, IORING_OP_WRITEV, IORING_OP_FSYNC, IORING_OP_READ_FIXED, IORING_OP_WRITE_FIXED, IORING_OP_POLL_ADD, IORING_OP_POLL_REMOVE, IORING_OP_SYNC_FILE_RANGE, IORING_OP_SENDMSG, IORING_OP_RECVMSG, IORING_OP_TIMEOUT, IORING_OP_TIMEOUT_REMOVE, IORING_OP_ACCEPT, IORING_OP_ASYNC_CANCEL, IORING_OP_LINK_TIMEOUT, IORING_OP_CONNECT, IORING_OP_FALLOCATE, IORING_OP_OPENAT, IORING_OP_CLOSE, IORING_OP_FILES_UPDATE, IORING_OP_STATX, IORING_OP_READ, IORING_OP_WRITE, IORING_OP_FADVISE, IORING_OP_MADVISE, IORING_OP_SEND, IORING_OP_RECV, IORING_OP_OPENAT2, IORING_OP_EPOLL_CTL, IORING_OP_SPLICE, IORING_OP_PROVIDE_BUFFERS, IORING_OP_REMOVE_BUFFERS, IORING_OP_TEE, IORING_OP_SHUTDOWN, IORING_OP_RENAMEAT, IORING_OP_UNLINKAT, IORING_OP_MKDIRAT, IORING_OP_SYMLINKAT, IORING_OP_LINKAT, IORING_OP_MSG_RING, IORING_OP_FSETXATTR, IORING_OP_SETXATTR, IORING_OP_FGETXATTR, IORING_OP_GETXATTR, IORING_OP_SOCKET, IORING_OP_URING_CMD, IORING_OP_SEND_ZC

# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# io_uring submission queue entry (io_uring_sqe) descriptions # # # # # # # # #
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #

#
# sqe type template
#

type io_uring_sqe[OP, IOPRIO, FD, OFF, ADDR, LEN, MISC_FLAGS, USER_DATA, MISC] {
	opcode		const[OP, int8]
	flags		flags[iosqe_flags, int8]
	ioprio		IOPRIO
	fd		FD
	off		OFF
	addr		ADDR
	len		LEN
	misc_flags	MISC_FLAGS
	user_data	flags[USER_DATA, int64]
# This is a union of different possibilites with a padding at the end.
	misc		MISC
} [size[SIZEOF_IO_URING_SQE]]

define SIZEOF_IO_URING_SQE	64

# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #

#
# Instantiation of sqes for each op
#

type io_uring_sqe$nop io_uring_sqe[IORING_OP_NOP, const[0, int16], const[0, int32], const[0, int64], const[0, int64], const[0, int32], const[0, int32], zero_flag, array[const[0, int64], 3]]

io_uring_sqe_readv [
	pass_iovec		io_uring_sqe[IORING_OP_READV, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], ptr[in, array[iovec_out]], len[addr, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, personality_only_misc]
	use_registered_buffer	io_uring_sqe[IORING_OP_READV, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], const[0, int64], const[0, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, buf_index_personality_misc]
]

type io_uring_sqe$writev io_uring_sqe[IORING_OP_WRITEV, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], ptr[in, array[iovec_in]], len[addr, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, buf_index_personality_misc]
type io_uring_sqe$fsync io_uring_sqe[IORING_OP_FSYNC, const[0, int16], fd_or_fixed_fd_index, const[0, int64], const[0, int64], const[0, int32], flags[io_uring_fsync_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$read_fixed io_uring_sqe[IORING_OP_READ_FIXED, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], int64, int32, flags[rwf_flags, int32], sqe_user_data_not_openat, buf_index_personality_misc]
type io_uring_sqe$write_fixed io_uring_sqe[IORING_OP_WRITE_FIXED, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], int64, int32, flags[rwf_flags, int32], sqe_user_data_not_openat, buf_index_personality_misc]
type io_uring_sqe$poll_add io_uring_sqe[IORING_OP_POLL_ADD, const[0, int16], fd_or_fixed_fd_index, const[0, int64], const[0, int64], const[0, int32], io_uring_sqe_poll_add_misc_flags, sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$poll_remove io_uring_sqe[IORING_OP_POLL_REMOVE, const[0, int16], const[0, int32], const[0, int64], flags[sqe_user_data, int64], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$sync_file_range io_uring_sqe[IORING_OP_SYNC_FILE_RANGE, const[0, int16], fd_or_fixed_fd_index, fileoff[int64], const[0, int64], int32, flags[sync_file_flags, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_SYNC_FILE_RANGE >= 5.2
type io_uring_sqe$sendmsg io_uring_sqe[IORING_OP_SENDMSG, const[0, int16], sock, const[0, int64], ptr[in, send_msghdr], const[0, int32], flags[send_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$recvmsg io_uring_sqe[IORING_OP_RECVMSG, flags[iouring_recv_ioprio, int16], sock, const[0, int64], ptr[inout, recv_msghdr], const[0, int32], flags[recv_flags, int32], sqe_user_data_not_openat, buf_group_personality_misc]
# IORING_OP_SENDMSG, IORING_OP_RECVMSG >= 5.3
type io_uring_sqe$timeout io_uring_sqe[IORING_OP_TIMEOUT, const[0, int16], const[0, int32], io_uring_timeout_completion_event_count, ptr[in, timespec], const[1, int32], flags[io_uring_timeout_flags, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_TIMEOUT >= 5.4
type io_uring_sqe$timeout_remove io_uring_sqe[IORING_OP_TIMEOUT_REMOVE, const[0, int16], const[0, int32], const[0, int64], flags[sqe_user_data, int64], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$accept io_uring_sqe[IORING_OP_ACCEPT, flags[iouring_accept_flags, int16], sock, ptr[inout, len[addr, int32]], ptr[out, sockaddr_storage, opt], const[0, int32], flags[accept_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$async_cancel io_uring_sqe[IORING_OP_ASYNC_CANCEL, const[0, int16], const[0, int32], const[0, int64], flags[sqe_user_data, int64], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$link_timeout io_uring_sqe[IORING_OP_LINK_TIMEOUT, const[0, int16], const[0, int32], const[0, int64], ptr[in, timespec], const[1, int32], flags[io_uring_timeout_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$connect io_uring_sqe[IORING_OP_CONNECT, const[0, int16], sock, len[addr, int32], ptr[in, sockaddr_storage], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_TIMEOUT_REMOVE, IORING_OP_ACCEPT, IORING_OP_ASYNC_CANCEL, IORING_OP_LINK_TIMEOUT, IORING_OP_CONNECT >= 5.5
type io_uring_sqe$fallocate io_uring_sqe[IORING_OP_FALLOCATE, const[0, int16], fd_or_fixed_fd_index, fileoff[int64], const[0, int64], int32, const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$openat io_uring_sqe[IORING_OP_OPENAT, const[0, int16], fd_dir[opt], const[0, int64], ptr64[in, filename], flags[open_mode, int32], flags[open_flags, int32], sqe_user_data_openat, personality_only_misc]
type io_uring_sqe$close io_uring_sqe[IORING_OP_CLOSE, const[0, int16], fd, const[0, int64], const[0, int64], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$files_update io_uring_sqe[IORING_OP_FILES_UPDATE, const[0, int16], const[0, int32], fileoff[int64], ptr[in, array[fd]], len[addr, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$statx io_uring_sqe[IORING_OP_STATX, const[0, int16], fd_dir[opt], ptr[out, statx], ptr64[in, filename], flags[statx_mask, int32], flags[statx_flags, int32], sqe_user_data_not_openat, personality_only_misc]

iouring_accept_flags = IORING_ACCEPT_MULTISHOT, IORING_ACCEPT_DONTWAIT, IORING_ACCEPT_POLL_FIRST

io_uring_sqe_read [
	pass_buffer		io_uring_sqe[IORING_OP_READ, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], buffer[out], bytesize[addr, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, personality_only_misc]
	use_registered_buffer	io_uring_sqe[IORING_OP_READ, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], const[0, int64], const[0, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, buf_index_personality_misc]
]

type io_uring_sqe$write io_uring_sqe[IORING_OP_WRITE, flags[ioprio_priorities, int16], fd_or_fixed_fd_index, fileoff[int64], buffer[in], bytesize[addr, int32], flags[rwf_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$fadvise io_uring_sqe[IORING_OP_FADVISE, const[0, int16], fd_or_fixed_fd_index, fileoff[int64], const[0, int64], int32, flags[fadvise_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$madvise io_uring_sqe[IORING_OP_MADVISE, const[0, int16], const[0, int32], const[0, int64], vma, len[addr, int32], flags[madvise_flags, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$send io_uring_sqe[IORING_OP_SEND, const[0, int16], sock, const[0, int64], buffer[in], len[addr, int32], flags[send_flags, int32], sqe_user_data_not_openat, personality_only_misc]

iouring_recv_ioprio = IORING_RECVSEND_POLL_FIRST, IORING_RECV_MULTISHOT, IORING_RECVSEND_FIXED_BUF, IORING_RECVSEND_BUNDLE

io_uring_sqe_recv [
	pass_buffer		io_uring_sqe[IORING_OP_RECV, const[0, int16], sock, const[0, int64], buffer[inout], len[addr, int32], flags[recv_flags, int32], sqe_user_data_not_openat, personality_only_misc]
	use_registered_buffer	io_uring_sqe[IORING_OP_RECV, flags[iouring_recv_ioprio, int16], sock, const[0, int64], const[0, int64], const[0, int32], flags[recv_flags, int32], sqe_user_data_not_openat, buf_group_personality_misc]
]

type io_uring_sqe$openat2 io_uring_sqe[IORING_OP_OPENAT2, const[0, int16], fd_dir[opt], ptr[in, open_how], ptr64[in, filename], bytesize[off, int32], const[0, int32], sqe_user_data_openat, personality_only_misc]
type io_uring_sqe$epoll_ctl_t[EPOLL_OP, EPOLL_EVENTS] io_uring_sqe[IORING_OP_EPOLL_CTL, const[0, int16], fd_epoll, EPOLL_EVENTS, fd, const[EPOLL_OP, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]

io_uring_sqe_epoll_ctl [
	add	io_uring_sqe$epoll_ctl_t[EPOLL_CTL_ADD, ptr[in, epoll_event]]
	del	io_uring_sqe$epoll_ctl_t[EPOLL_CTL_DEL, const[0, int64]]
	mod	io_uring_sqe$epoll_ctl_t[EPOLL_CTL_MOD, ptr[in, epoll_event]]
]

# IORING_OP_EPOLL_CTL, IORING_OP_SEND, IORING_OP_FALLOCATE, IORING_OP_MADVISE, IORING_OP_FADVISE, IORING_OP_RECV
# IORING_OP_OPENAT, IORING_OP_OPENAT2, IORING_OP_CLOSE, IORING_OP_STATX, IORING_OP_READ, IORING_OP_WRITE >= 5.6

type io_uring_sqe$splice io_uring_sqe[IORING_OP_SPLICE, const[0, int16], fd_or_fixed_fd_index, fileoff[int64], io_uring_sqe_splice_off_in, int32, flags[splice_flags, int32], sqe_user_data_not_openat, io_uring_sqe_splice_misc]
type io_uring_sqe$provide_buffers io_uring_sqe[IORING_OP_PROVIDE_BUFFERS, const[0, int16], int32, io_uring_bid[int64], buffer[in], int32, const[0, int32], sqe_user_data_not_openat, buf_group_personality_misc]
type io_uring_sqe$remove_buffers io_uring_sqe[IORING_OP_PROVIDE_BUFFERS, const[0, int16], int32, const[0, int64], const[0, int64], const[0, int32], const[0, int32], sqe_user_data_not_openat, buf_group_personality_misc]
# IORING_OP_SPLICE, IORING_OP_PROVIDE_BUFFERS >= 5.7
type io_uring_sqe$tee io_uring_sqe[IORING_OP_TEE, const[0, int16], fd_or_fixed_fd_index, const[0, int64], const[0, int64], int32, flags[splice_flags, int32], sqe_user_data_not_openat, io_uring_sqe_tee_misc]
# IORING_OP_TEE >= 5.8

type io_uring_sqe$shutdown io_uring_sqe[IORING_OP_SHUTDOWN, const[0, int16], sock, const[0, int64], const[0, int64], flags[shutdown_flags, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$renameat io_uring_sqe[IORING_OP_RENAMEAT, const[0, int16], fd_dir, ptr64[in, filename], ptr64[in, filename], fd_dir, const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$unlinkat io_uring_sqe[IORING_OP_UNLINKAT, const[0, int16], fd_dir, const[0, int64], ptr64[in, filename], const[0, int32], flags[unlinkat_flags, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_SHUTDOWN, IORING_OP_RENAMEAT, IORING_OP_UNLINKAT >= 5.11

type io_uring_sqe$mkdirat io_uring_sqe[IORING_OP_MKDIRAT, const[0, int16], fd_dir, const[0, int64], ptr64[in, filename], flags[open_mode, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$symlinkat io_uring_sqe[IORING_OP_SYMLINKAT, const[0, int16], fd_dir, ptr64[in, filename], ptr64[in, filename], const[0, int32], const[0, int32], sqe_user_data_not_openat, personality_only_misc]
type io_uring_sqe$linkat io_uring_sqe[IORING_OP_LINKAT, const[0, int16], fd_dir, ptr64[in, filename], ptr64[in, filename], fd_dir, flags[linkat_flags, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_MKDIRAT, IORING_OP_SYMLINKAT, IORING_OP_LINKAT >= 5.15

iouring_msg_ring_cmd_flags = IORING_MSG_DATA, IORING_MSG_SEND_FD
msg_ring_flags = IORING_MSG_RING_CQE_SKIP, IORING_MSG_RING_FLAGS_PASS

type io_uring_sqe$msg_ring io_uring_sqe[IORING_OP_MSG_RING, const[0, int16], fd_io_uring, flags[iouring_msg_ring_cmd_flags, int64], buffer[in], len[addr, int32], flags[msg_ring_flags, int32], sqe_user_data_not_openat, personality_only_misc]
# IORING_OP_MSG_RING >= 5.18

# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #

#
# Flags, enumerations, and misc fields of sqe ops
#

iosqe_flags = IOSQE_IO_DRAIN, IOSQE_IO_LINK, IOSQE_IO_HARDLINK, IOSQE_ASYNC, IOSQE_BUFFER_SELECT, IOSQE_FIXED_FILE, IOSQE_CQE_SKIP_SUCCESS
# should not use BIT as flags
# IOSQE_IO_DRAIN >= 5.2
# IOSQE_IO_LINK >= 5.3
# IOSQE_IO_HARDLINK >= 5.5
# IOSQE_ASYNC >= 5.6
# IOSQE_BUFFER_SELECT >= 5.7
# IOSQE_FIXED_FILE >= 5.1
# IOSQE_CQE_SKIP_SUCCESS >= 5.17

fd_or_fixed_fd_index [
	fd		fd
# Use the registered files (io_uring_register$IORING_REGISTER_FILES) when IOSQE_FIXED_FILE_BIT is set in sqe.
# To ease collisions, limit the indices.
	fd_index	int32[0:10]
]

# 0 for normal file integrity sync, IORING_FSYNC_DATASYNC to provide data sync only semantics
io_uring_fsync_flags = 0, IORING_FSYNC_DATASYNC

# 0 for relative, IORING_TIMEOUT_ABS for absolute timeout value
io_uring_timeout_flags = 0, IORING_TIMEOUT_ABS, IORING_TIMEOUT_UPDATE, IORING_TIMEOUT_BOOTTIME, IORING_TIMEOUT_REALTIME, IORING_LINK_TIMEOUT_UPDATE, IORING_TIMEOUT_ETIME_SUCCESS, IORING_TIMEOUT_MULTISHOT, IORING_TIMEOUT_CLOCK_MASK, IORING_TIMEOUT_UPDATE_MASK

# The timeout condition is met when either the specific timeout expries, or the
# specified number of events have completed. If not set, defaults to 1. Use a
# limited range to allow utilization of this value to meet timeout condition besides
# the timeout expiration.
type io_uring_timeout_completion_event_count int64[0:10]

# An already issued request can be attempted to be cancelled using ASYNC_CANCEL
# operation. This operation identifies the operations using what's passed as
# with user_data in their sqe. To ease collisions of ASYNC_CANCEL operation with
# already submitted ones, use a limited range of values for user_data field.
# Among all operations that can be achieved by submitting to the io_uring, only
# openat and openat2 returns a useful resource (fd) that we can use for other
# systemcalls. The resulting fds are returned within io_uring_cqe.res. The only way
# to identify cqes for those operations is to keep track of their user data. Thus,
# use a separate set of sqe_user_data range for openat and openat2.
sqe_user_data_not_openat = 0, 1
sqe_user_data_openat = 0x12345, 0x23456
sqe_user_data = 0, 1, 0x12345, 0x23456

# The buffer id (bid) and the buffer group id (bgid) are registered using
# IORING_OP_PROVIDE_BUFFERS. Use the ids in a limited range to ease collisions
# with other operations.
type io_uring_bid[T] T[0:3]
type io_uring_bgid[T] T[0:3]

zero_flag = 0

io_uring_sqe_poll_add_misc_flags {
	misc_flags		flags[pollfd_events, int16]
# 2 bytes of padding to fill what is left from the union of flags
	fill_flags_union	const[0, int16]
}

io_uring_sqe_splice_off_in {
	splice_off_in_unused	const[0, int32]
	splice_off_in		fd
}

# Descriptions for MISC field of io_uring_sqe. The content for most are common
# while there are a few specials.

personality_only_misc {
	buf_index_unused	const[0, int16]
	ioring_personality_id	ioring_personality_id[opt]
	pad_unused		array[const[0, int8], 20]
}

buf_index_personality_misc {
	buf_index		io_uring_bid[int16]
	ioring_personality_id	ioring_personality_id[opt]
	pad_unused		array[const[0, int8], 20]
}

buf_group_personality_misc {
	buf_group		io_uring_bgid[int16]
	ioring_personality_id	ioring_personality_id[opt]
	pad_unused		array[const[0, int8], 20]
}

io_uring_sqe_splice_misc {
	buf_unused		const[0, int16]
	ioring_personality_id	ioring_personality_id[opt]
	splice_fd_in		fd
	pad_unused		array[const[0, int64], 2]
}

io_uring_sqe_tee_misc {
	buf_unused		const[0, int16]
	ioring_personality_id	ioring_personality_id[opt]
	splice_fd_in		fd
	pad_unused		array[const[0, int64], 2]
}

io_uring_sync_cancel_reg {
	addr	int64
	fd	fd
	flags	flags[iouring_async_cancel_flags, int32]
	timeout	kernel_timespec
	opcode	int8
	pad	array[const[0, int8], 7]
	pad2	array[const[0, int64], 3]
}

kernel_timespec {
	tv_sec	int64
	tv_nsec	int64
}

iouring_async_cancel_flags = IORING_ASYNC_CANCEL_ALL, IORING_ASYNC_CANCEL_FD, IORING_ASYNC_CANCEL_ANY, IORING_ASYNC_CANCEL_FD_FIXED, IORING_ASYNC_CANCEL_USERDATA, IORING_ASYNC_CANCEL_OP

io_uring_file_index_range {
	off	int32
	len	int32
	resv	int64
}

io_uring_buf_status {
	buf_group	int32	(in)
	head		int32	(out)
	resv		array[const[0, int32], 8]
}

io_uring_napi {
	busy_poll_to		int32
	prefer_busy_poll	int8
	pad			array[const[0, int8], 3]
	resv			const[0, int64]
}

io_uring_clock_register {
	clockid	flags[clock_type, int32]
	resv	array[const[0, int32], 3]
}

io_uring_clone_buffers {
	src_fd	fd
	flags	const[1, int32]
	pad	array[const[0, int32], 6]
}

io_uring_mem_region_reg {
	region_uptr	ptr64[inout, io_uring_region_desc]
	flags		flags[io_uring_region_flags, int64]
	resv		array[const[0, int64], 2]
}

io_uring_region_flags = IORING_MEM_REGION_REG_WAIT_ARG

io_uring_region_desc {
	user_addr	ptr64[inout, array[int8]]
	size		len[user_addr, int64]
	flags		flags[io_uring_region_desc_flags, int32]
	id		int32
	mmap_offset	int64
	resv		array[const[0, int64], 4]
}

io_uring_region_desc_flags = IORING_MEM_REGION_TYPE_USER