There is no reliable way to submit and wait in a single syscall, as
io_submit_sqes() may under-consume sqes (in case of an early error).
Then it will wait for not-yet-submitted requests, deadlocking the user
in most cases.
In such cases adjust min_complete, so it won't wait for more than
what have been submitted in the current call to io_uring_enter(). It
may be less than totally in-flight including previous submissions,
but this shouldn't do harm and up to a user.
Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
v2: cap min_complete if submitted partially (Jens Axboe)
v3: update commit message (Jens Axboe)
fs/io_uring.c | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 81219a631a6d..5dfc805ec31c 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -3763,11 +3763,8 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
unsigned int sqe_flags;
req = io_get_req(ctx, statep);
- if (unlikely(!req)) {
- if (!submitted)
- submitted = -EAGAIN;
+ if (unlikely(!req))
break;
- }
if (!io_get_sqring(ctx, req)) {
__io_free_req(req);
break;
@@ -5272,6 +5269,14 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
submitted = io_submit_sqes(ctx, to_submit, f.file, fd,
&cur_mm, false);
mutex_unlock(&ctx->uring_lock);
+
+ if (submitted != to_submit) {
+ if (!submitted) {
+ submitted = -EAGAIN;
+ goto done;
+ }
+ min_complete = min(min_complete, (u32)submitted);
+ }
}
if (flags & IORING_ENTER_GETEVENTS) {
unsigned nr_events = 0;
@@ -5284,7 +5289,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
}
}
-
+done:
percpu_ref_put(&ctx->refs);
out_fput:
fdput(f);
--
2.24.0
