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

fs/io_uring.c

···
 	struct list_head		file_list;
 	struct fixed_file_data		*file_data;
 	struct llist_node		llist;
+	bool				done;
 };
 
 struct fixed_file_data {
···
 struct io_open {
 	struct file			*file;
 	int				dfd;
+	bool				ignore_nonblock;
 	struct filename			*filename;
 	struct open_how			how;
 	unsigned long			nofile;
···
 	}
 end_req:
 	req_set_fail_links(req);
-	io_req_complete(req, ret);
 	return false;
 }
 #endif
···
 		return ret;
 	}
 	req->open.nofile = rlimit(RLIMIT_NOFILE);
+	req->open.ignore_nonblock = false;
 	req->flags |= REQ_F_NEED_CLEANUP;
 	return 0;
 }
···
 	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);
···
 	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);
···
 		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);
···
 		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);
···
 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;
 
-	if (percpu_ref_is_dying(&ctx->file_data->refs))
+	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(&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)
···
 	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;
 }
 
···
 
 	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;
···
 	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);

fs/proc/self.c

···
 	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 */

mm/filemap.c

···
 
 page_not_up_to_date:
 		/* Get exclusive access to the page ... */
-		if (iocb->ki_flags & IOCB_WAITQ)
+		if (iocb->ki_flags & IOCB_WAITQ) {
+			if (written) {
+				put_page(page);
+				goto out;
+			}
 			error = lock_page_async(page, iocb->ki_waitq);
-		else
+		} else {
 			error = lock_page_killable(page);
+		}
 		if (unlikely(error))
 			goto readpage_error;
 
···
 		}
 
 		if (!PageUptodate(page)) {
-			if (iocb->ki_flags & IOCB_WAITQ)
+			if (iocb->ki_flags & IOCB_WAITQ) {
+				if (written) {
+					put_page(page);
+					goto out;
+				}
 				error = lock_page_async(page, iocb->ki_waitq);
-			else
+			} else {
 				error = lock_page_killable(page);
+			}
 
 			if (unlikely(error))
 				goto readpage_error;