io_uring: account user memory freed when exit has been queued

We currently account the memory after the exit work has been run, but
that leaves a gap where a process has closed its ring and until the
memory has been accounted as freed. If the memlocked ulimit is
borderline, then that can introduce spurious setup errors returning
-ENOMEM because the free work hasn't been run yet.

Account this as freed when we close the ring, as not to expose a tiny
gap where setting up a new ring can fail.

Fixes: 85faa7b8346e ("io_uring: punt final io_ring_ctx wait-and-free to workqueue")
Cc: stable@vger.kernel.org # v5.7
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/fs/io_uring.c b/fs/io_uring.c
index fc07baf4392a8dfdd8aeb6572eb2b314dbac5b9a..ca8abde48b6c702690f477e62aa624b67bf52be2 100644 (file)
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7351,9 +7351,6 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)
        io_mem_free(ctx->sq_sqes);
 
        percpu_ref_exit(&ctx->refs);
-       if (ctx->account_mem)
-               io_unaccount_mem(ctx->user,
-                               ring_pages(ctx->sq_entries, ctx->cq_entries));
        free_uid(ctx->user);
        put_cred(ctx->creds);
        kfree(ctx->cancel_hash);
@@ -7438,6 +7435,16 @@ static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
        if (ctx->rings)
                io_cqring_overflow_flush(ctx, true);
        idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx);
+
+       /*
+        * Do this upfront, so we won't have a grace period where the ring
+        * is closed but resources aren't reaped yet. This can cause
+        * spurious failure in setting up a new ring.
+        */
+       if (ctx->account_mem)
+               io_unaccount_mem(ctx->user,
+                               ring_pages(ctx->sq_entries, ctx->cq_entries));
+
        INIT_WORK(&ctx->exit_work, io_ring_exit_work);
        queue_work(system_wq, &ctx->exit_work);
 }