Merge tag 'io_uring-5.10-2020-11-20' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "Mostly regression or stable fodder:

   - Disallow async path resolution of /proc/self

   - Tighten constraints for segmented async buffered reads

   - Fix double completion for a retry error case

   - Fix for fixed file life times (Pavel)"

* tag 'io_uring-5.10-2020-11-20' of git://git.kernel.dk/linux-block:
  io_uring: order refnode recycling
  io_uring: get an active ref_node from files_data
  io_uring: don't double complete failed reissue request
  mm: never attempt async page lock if we've transferred data already
  io_uring: handle -EOPNOTSUPP on path resolution
  proc: don't allow async path resolution of /proc/self components

2 files changed, 49 insertions, 15 deletions

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 4ead291b2976..a8c136a1cf4e 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -205,6 +205,7 @@ struct fixed_file_ref_node {
 	struct list_head		file_list;
 	struct fixed_file_data		*file_data;
 	struct llist_node		llist;
+	bool				done;
 };
 
 struct fixed_file_data {
@@ -478,6 +479,7 @@ struct io_sr_msg {
 struct io_open {
 	struct file			*file;
 	int				dfd;
+	bool				ignore_nonblock;
 	struct filename			*filename;
 	struct open_how			how;
 	unsigned long			nofile;
@@ -2577,7 +2579,6 @@ static bool io_resubmit_prep(struct io_kiocb *req, int error)
 	}
 end_req:
 	req_set_fail_links(req);
-	io_req_complete(req, ret);
 	return false;
 }
 #endif
@@ -3795,6 +3796,7 @@ static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe
 		return ret;
 	}
 	req->open.nofile = rlimit(RLIMIT_NOFILE);
+	req->open.ignore_nonblock = false;
 	req->flags |= REQ_F_NEED_CLEANUP;
 	return 0;
 }
@@ -3838,7 +3840,7 @@ static int io_openat2(struct io_kiocb *req, bool force_nonblock)
 	struct file *file;
 	int ret;
 
-	if (force_nonblock)
+	if (force_nonblock && !req->open.ignore_nonblock)
 		return -EAGAIN;
 
 	ret = build_open_flags(&req->open.how, &op);
@@ -3853,6 +3855,21 @@ static int io_openat2(struct io_kiocb *req, bool force_nonblock)
 	if (IS_ERR(file)) {
 		put_unused_fd(ret);
 		ret = PTR_ERR(file);
+		/*
+		 * A work-around to ensure that /proc/self works that way
+		 * that it should - if we get -EOPNOTSUPP back, then assume
+		 * that proc_self_get_link() failed us because we're in async
+		 * context. We should be safe to retry this from the task
+		 * itself with force_nonblock == false set, as it should not
+		 * block on lookup. Would be nice to know this upfront and
+		 * avoid the async dance, but doesn't seem feasible.
+		 */
+		if (ret == -EOPNOTSUPP && io_wq_current_is_worker()) {
+			req->open.ignore_nonblock = true;
+			refcount_inc(&req->refs);
+			io_req_task_queue(req);
+			return 0;
+		}
 	} else {
 		fsnotify_open(file);
 		fd_install(ret, file);
@@ -6957,9 +6974,7 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
 		return -ENXIO;
 
 	spin_lock(&data->lock);
-	if (!list_empty(&data->ref_list))
-		ref_node = list_first_entry(&data->ref_list,
-				struct fixed_file_ref_node, node);
+	ref_node = data->node;
 	spin_unlock(&data->lock);
 	if (ref_node)
 		percpu_ref_kill(&ref_node->refs);
@@ -7308,10 +7323,6 @@ static void __io_file_put_work(struct fixed_file_ref_node *ref_node)
 		kfree(pfile);
 	}
 
-	spin_lock(&file_data->lock);
-	list_del(&ref_node->node);
-	spin_unlock(&file_data->lock);
-
 	percpu_ref_exit(&ref_node->refs);
 	kfree(ref_node);
 	percpu_ref_put(&file_data->refs);
@@ -7338,17 +7349,32 @@ static void io_file_put_work(struct work_struct *work)
 static void io_file_data_ref_zero(struct percpu_ref *ref)
 {
 	struct fixed_file_ref_node *ref_node;
+	struct fixed_file_data *data;
 	struct io_ring_ctx *ctx;
-	bool first_add;
+	bool first_add = false;
 	int delay = HZ;
 
 	ref_node = container_of(ref, struct fixed_file_ref_node, refs);
-	ctx = ref_node->file_data->ctx;
+	data = ref_node->file_data;
+	ctx = data->ctx;
+
+	spin_lock(&data->lock);
+	ref_node->done = true;
+
+	while (!list_empty(&data->ref_list)) {
+		ref_node = list_first_entry(&data->ref_list,
+					struct fixed_file_ref_node, node);
+		/* recycle ref nodes in order */
+		if (!ref_node->done)
+			break;
+		list_del(&ref_node->node);
+		first_add |= llist_add(&ref_node->llist, &ctx->file_put_llist);
+	}
+	spin_unlock(&data->lock);
 
-	if (percpu_ref_is_dying(&ctx->file_data->refs))
+	if (percpu_ref_is_dying(&data->refs))
 		delay = 0;
 
-	first_add = llist_add(&ref_node->llist, &ctx->file_put_llist);
 	if (!delay)
 		mod_delayed_work(system_wq, &ctx->file_put_work, 0);
 	else if (first_add)
@@ -7372,6 +7398,7 @@ static struct fixed_file_ref_node *alloc_fixed_file_ref_node(
 	INIT_LIST_HEAD(&ref_node->node);
 	INIT_LIST_HEAD(&ref_node->file_list);
 	ref_node->file_data = ctx->file_data;
+	ref_node->done = false;
 	return ref_node;
 }
 
@@ -7467,7 +7494,7 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
 
 	file_data->node = ref_node;
 	spin_lock(&file_data->lock);
-	list_add(&ref_node->node, &file_data->ref_list);
+	list_add_tail(&ref_node->node, &file_data->ref_list);
 	spin_unlock(&file_data->lock);
 	percpu_ref_get(&file_data->refs);
 	return ret;
@@ -7626,7 +7653,7 @@ static int __io_sqe_files_update(struct io_ring_ctx *ctx,
 	if (needs_switch) {
 		percpu_ref_kill(&data->node->refs);
 		spin_lock(&data->lock);
-		list_add(&ref_node->node, &data->ref_list);
+		list_add_tail(&ref_node->node, &data->ref_list);
 		data->node = ref_node;
 		spin_unlock(&data->lock);
 		percpu_ref_get(&ctx->file_data->refs);
diff --git a/fs/proc/self.c b/fs/proc/self.c
index 72cd69bcaf4a..cc71ce3466dc 100644
--- a/fs/proc/self.c
+++ b/fs/proc/self.c
@@ -16,6 +16,13 @@ static const char *proc_self_get_link(struct dentry *dentry,
 	pid_t tgid = task_tgid_nr_ns(current, ns);
 	char *name;
 
+	/*
+	 * Not currently supported. Once we can inherit all of struct pid,
+	 * we can allow this.
+	 */
+	if (current->flags & PF_KTHREAD)
+		return ERR_PTR(-EOPNOTSUPP);
+
 	if (!tgid)
 		return ERR_PTR(-ENOENT);
 	/* max length of unsigned int in decimal + NULL term */