io_uring/rw: always setup io_async_rw for read/write requests

read/write requests try to put everything on the stack, and then alloc
and copy if a retry is needed. This necessitates a bunch of nasty code
that deals with intermediate state.

Get rid of this, and have the prep side setup everything that is needed
upfront, which greatly simplifies the opcode handlers.

This includes adding an alloc cache for io_async_rw, to make it cheap
to handle.

In terms of cost, this should be basically free and transparent. For
the worst case of {READ,WRITE}_FIXED which didn't need it before,
performance is unaffected in the normal peak workload that is being
used to test that. Still runs at 122M IOPS.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

include/linux/io_uring_types.h

@@ -300,6 +300,7 @@ struct io_ring_ctx {
 		struct io_hash_table	cancel_table_locked;
 		struct io_alloc_cache	apoll_cache;
 		struct io_alloc_cache	netmsg_cache;
+		struct io_alloc_cache	rw_cache;
 
 		/*
 		 * Any cancelable uring_cmd is added to this list in

io_uring/io_uring.c

@@ -311,6 +311,8 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 			    sizeof(struct async_poll));
 	io_alloc_cache_init(&ctx->netmsg_cache, IO_ALLOC_CACHE_MAX,
 			    sizeof(struct io_async_msghdr));
+	io_alloc_cache_init(&ctx->rw_cache, IO_ALLOC_CACHE_MAX,
+			    sizeof(struct io_async_rw));
 	io_futex_cache_init(ctx);
 	init_completion(&ctx->ref_comp);
 	xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
@@ -2823,6 +2825,7 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	io_eventfd_unregister(ctx);
 	io_alloc_cache_free(&ctx->apoll_cache, io_apoll_cache_free);
 	io_alloc_cache_free(&ctx->netmsg_cache, io_netmsg_cache_free);
+	io_alloc_cache_free(&ctx->rw_cache, io_rw_cache_free);
 	io_futex_cache_free(ctx);
 	io_destroy_buffers(ctx);
 	mutex_unlock(&ctx->uring_lock);

io_uring/opdef.c

@@ -67,7 +67,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
 		.vectored		= 1,
-		.prep			= io_prep_rwv,
+		.prep			= io_prep_readv,
 		.issue			= io_read,
 	},
 	[IORING_OP_WRITEV] = {
@@ -81,7 +81,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
 		.vectored		= 1,
-		.prep			= io_prep_rwv,
+		.prep			= io_prep_writev,
 		.issue			= io_write,
 	},
 	[IORING_OP_FSYNC] = {
@@ -99,7 +99,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.ioprio			= 1,
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
-		.prep			= io_prep_rw_fixed,
+		.prep			= io_prep_read_fixed,
 		.issue			= io_read,
 	},
 	[IORING_OP_WRITE_FIXED] = {
@@ -112,7 +112,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.ioprio			= 1,
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
-		.prep			= io_prep_rw_fixed,
+		.prep			= io_prep_write_fixed,
 		.issue			= io_write,
 	},
 	[IORING_OP_POLL_ADD] = {
@@ -239,7 +239,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.ioprio			= 1,
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
-		.prep			= io_prep_rw,
+		.prep			= io_prep_read,
 		.issue			= io_read,
 	},
 	[IORING_OP_WRITE] = {
@@ -252,7 +252,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.ioprio			= 1,
 		.iopoll			= 1,
 		.iopoll_queue		= 1,
-		.prep			= io_prep_rw,
+		.prep			= io_prep_write,
 		.issue			= io_write,
 	},
 	[IORING_OP_FADVISE] = {
@@ -490,14 +490,12 @@ const struct io_cold_def io_cold_defs[] = {
 	[IORING_OP_READV] = {
 		.async_size		= sizeof(struct io_async_rw),
 		.name			= "READV",
-		.prep_async		= io_readv_prep_async,
 		.cleanup		= io_readv_writev_cleanup,
 		.fail			= io_rw_fail,
 	},
 	[IORING_OP_WRITEV] = {
 		.async_size		= sizeof(struct io_async_rw),
 		.name			= "WRITEV",
-		.prep_async		= io_writev_prep_async,
 		.cleanup		= io_readv_writev_cleanup,
 		.fail			= io_rw_fail,
 	},
@@ -699,6 +697,7 @@ const struct io_cold_def io_cold_defs[] = {
 #endif
 	},
 	[IORING_OP_READ_MULTISHOT] = {
+		.async_size		= sizeof(struct io_async_rw),
 		.name			= "READ_MULTISHOT",
 	},
 	[IORING_OP_WAITID] = {

io_uring/rw.h

@@ -9,21 +9,26 @@ struct io_rw_state {
 };
 
 struct io_async_rw {
+	union {
+		size_t			bytes_done;
+		struct io_cache_entry	cache;
+	};
 	struct io_rw_state		s;
-	const struct iovec		*free_iovec;
-	size_t				bytes_done;
+	struct iovec			*free_iovec;
 	struct wait_page_queue		wpq;
 };
 
-int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe);
-int io_prep_rwv(struct io_kiocb *req, const struct io_uring_sqe *sqe);
-int io_prep_rw_fixed(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_read_fixed(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_write_fixed(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_readv(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_writev(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_read(struct io_kiocb *req, const struct io_uring_sqe *sqe);
+int io_prep_write(struct io_kiocb *req, const struct io_uring_sqe *sqe);
 int io_read(struct io_kiocb *req, unsigned int issue_flags);
-int io_readv_prep_async(struct io_kiocb *req);
 int io_write(struct io_kiocb *req, unsigned int issue_flags);
-int io_writev_prep_async(struct io_kiocb *req);
 void io_readv_writev_cleanup(struct io_kiocb *req);
 void io_rw_fail(struct io_kiocb *req);
 void io_req_rw_complete(struct io_kiocb *req, struct io_tw_state *ts);
 int io_read_mshot_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe);
 int io_read_mshot(struct io_kiocb *req, unsigned int issue_flags);
+void io_rw_cache_free(struct io_cache_entry *entry);