io_uring: do grab_env() just before punting

Currently io_steal_work() is disabled, and every linked request should
go through task_work for initialisation. Do io_req_work_grab_env()
just before io-wq punting and for the whole link, so any request
reachable by io_steal_work() is prepared.

This is also interesting for another reason -- it localises
io_req_work_grab_env() into one place just before io-wq punting, helping
to to better manage req->work lifetime and add some neat
cleanup/optimisations later.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/fs/io_uring.c b/fs/io_uring.c
index caf9083..9bc4339 100644 (file)
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1101,7 +1101,7 @@ static void __io_commit_cqring(struct io_ring_ctx *ctx)
        }
 }
 
-static inline void io_req_work_grab_env(struct io_kiocb *req)
+static void io_req_work_grab_env(struct io_kiocb *req)
 {
        const struct io_op_def *def = &io_op_defs[req->opcode];
 
@@ -1150,8 +1150,7 @@ static inline void io_req_work_drop_env(struct io_kiocb *req)
        }
 }
 
-static inline void io_prep_async_work(struct io_kiocb *req,
-                                     struct io_kiocb **link)
+static void io_prep_async_work(struct io_kiocb *req)
 {
        const struct io_op_def *def = &io_op_defs[req->opcode];
 
@@ -1164,15 +1163,22 @@ static inline void io_prep_async_work(struct io_kiocb *req,
        }
 
        io_req_work_grab_env(req);
-       *link = io_prep_linked_timeout(req);
 }
 
-static inline void io_queue_async_work(struct io_kiocb *req)
+static void io_prep_async_link(struct io_kiocb *req)
 {
-       struct io_ring_ctx *ctx = req->ctx;
-       struct io_kiocb *link;
+       struct io_kiocb *cur;
 
-       io_prep_async_work(req, &link);
+       io_prep_async_work(req);
+       if (req->flags & REQ_F_LINK_HEAD)
+               list_for_each_entry(cur, &req->link_list, link_list)
+                       io_prep_async_work(cur);
+}
+
+static void __io_queue_async_work(struct io_kiocb *req)
+{
+       struct io_ring_ctx *ctx = req->ctx;
+       struct io_kiocb *link = io_prep_linked_timeout(req);
 
        trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
                                        &req->work, req->flags);
@@ -1182,6 +1188,13 @@ static inline void io_queue_async_work(struct io_kiocb *req)
                io_queue_linked_timeout(link);
 }
 
+static void io_queue_async_work(struct io_kiocb *req)
+{
+       /* init ->work of the whole link before punting */
+       io_prep_async_link(req);
+       __io_queue_async_work(req);
+}
+
 static void io_kill_timeout(struct io_kiocb *req)
 {
        int ret;
@@ -1215,7 +1228,8 @@ static void __io_queue_deferred(struct io_ring_ctx *ctx)
                if (req_need_defer(req))
                        break;
                list_del_init(&req->list);
-               io_queue_async_work(req);
+               /* punt-init is done before queueing for defer */
+               __io_queue_async_work(req);
        } while (!list_empty(&ctx->defer_list));
 }
 
@@ -1791,7 +1805,7 @@ static void io_put_req(struct io_kiocb *req)
 
 static struct io_wq_work *io_steal_work(struct io_kiocb *req)
 {
-       struct io_kiocb *nxt = NULL;
+       struct io_kiocb *timeout, *nxt = NULL;
 
        /*
         * A ref is owned by io-wq in which context we're. So, if that's the
@@ -1805,18 +1819,10 @@ static struct io_wq_work *io_steal_work(struct io_kiocb *req)
        if (!nxt)
                return NULL;
 
-       if ((nxt->flags & REQ_F_ISREG) && io_op_defs[nxt->opcode].hash_reg_file)
-               io_wq_hash_work(&nxt->work, file_inode(nxt->file));
-
-       io_req_task_queue(nxt);
-       /*
-        * If we're going to return actual work, here should be timeout prep:
-        *
-        * link = io_prep_linked_timeout(nxt);
-        * if (link)
-        *      nxt->flags |= REQ_F_QUEUE_TIMEOUT;
-        */
-       return NULL;
+       timeout = io_prep_linked_timeout(nxt);
+       if (timeout)
+               nxt->flags |= REQ_F_QUEUE_TIMEOUT;
+       return &nxt->work;
 }
 
 /*
@@ -5369,8 +5375,8 @@ static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
                ret = io_req_defer_prep(req, sqe);
                if (ret < 0)
                        return ret;
-               io_req_work_grab_env(req);
        }
+       io_prep_async_link(req);
 
        spin_lock_irq(&ctx->completion_lock);
        if (!req_need_defer(req) && list_empty(&ctx->defer_list)) {
@@ -5984,7 +5990,6 @@ fail_req:
                        ret = io_req_defer_prep(req, sqe);
                        if (unlikely(ret < 0))
                                goto fail_req;
-                       io_req_work_grab_env(req);
                }
 
                /*