[PATCH for-5.15 v2] io_uring: fix lacking of protection for compl_nr

[PATCH for-5.15 v2] io_uring: fix lacking of protection for compl_nr

[Hao Xu]

coml_nr in ctx_flush_and_put() is not protected by uring_lock, this
may cause problems when accessing it parallelly:

say coml_nr > 0

  ctx_flush_and put                  other context
   if (compl_nr)                      get mutex
                                      coml_nr > 0
                                      do flush
                                          coml_nr = 0
                                      release mutex
        get mutex
           do flush (*)
        release mutex

in (*) place, we call io_cqring_ev_posted() and users likely get
none events there.

Fixes: 2c32395d8111 ("io_uring: fix __tctx_task_work() ctx race")
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
---
 fs/io_uring.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c755efdac71f..c3bd2b3fc46b 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2005,7 +2005,8 @@ static void ctx_flush_and_put(struct io_ring_ctx *ctx)
 		return;
 	if (ctx->submit_state.compl_nr) {
 		mutex_lock(&ctx->uring_lock);
-		io_submit_flush_completions(ctx);
+		if (ctx->submit_state.compl_nr)
+			io_submit_flush_completions(ctx);
 		mutex_unlock(&ctx->uring_lock);
 	}
 	percpu_ref_put(&ctx->refs);
-- 
2.24.4

Re: [PATCH for-5.15 v2] io_uring: fix lacking of protection for compl_nr

[Jens Axboe]

On 8/20/21 4:19 PM, Hao Xu wrote:
> coml_nr in ctx_flush_and_put() is not protected by uring_lock, this
> may cause problems when accessing it parallelly:
> 
> say coml_nr > 0
> 
>   ctx_flush_and put                  other context
>    if (compl_nr)                      get mutex
>                                       coml_nr > 0
>                                       do flush
>                                           coml_nr = 0
>                                       release mutex
>         get mutex
>            do flush (*)
>         release mutex
> 
> in (*) place, we call io_cqring_ev_posted() and users likely get
> none events there.

Applied, thanks.

-- 
Jens Axboe