Add registered buffer support for vectored io_uring operations. That
allows to pass an iovec, all entries of which must belong to and
point into the same registered buffer specified by sqe->buf_index.
The series covers zerocopy sendmsg and reads / writes. Reads and
writes are implemented as new opcodes, while zerocopy sendmsg
reuses IORING_RECVSEND_FIXED_BUF for the user API.
Results are aligned to what one would expect from registered buffers:
t/io_uring + nullblk, single segment 16K:
34 -> 46 GiB/s
examples/send-zerocopy.c default send size (64KB):
82558 -> 123855 MB/s
The series is placed on top of 6.15 + zcrx + epoll.
v4: Address cast and include warnings
v3: Handle 32 bit where bvec is larger than iovec
v2: Nowarn alloc
Cap bvec caching
Check length overflow
Reject 0 len segments
Check op direction
Other minor changes
Pavel Begunkov (9):
io_uring: introduce struct iou_vec
io_uring: add infra for importing vectored reg buffers
io_uring/rw: implement vectored registered rw
io_uring/rw: defer reg buf vec import
io_uring/net: combine msghdr copy
io_uring/net: pull vec alloc out of msghdr import
io_uring/net: convert to struct iou_vec
io_uring/net: implement vectored reg bufs for zctx
io_uring: cap cached iovec/bvec size
include/linux/io_uring_types.h | 11 ++
include/uapi/linux/io_uring.h | 2 +
io_uring/alloc_cache.h | 9 --
io_uring/net.c | 180 +++++++++++++++++++++------------
io_uring/net.h | 6 +-
io_uring/opdef.c | 39 +++++++
io_uring/rsrc.c | 137 +++++++++++++++++++++++++
io_uring/rsrc.h | 23 +++++
io_uring/rw.c | 99 ++++++++++++++++--
io_uring/rw.h | 6 +-
10 files changed, 420 insertions(+), 92 deletions(-)
--
2.48.1
