netfs: Fix the request's work item to not require a ref

When the netfs_io_request struct's work item is queued, it must be supplied
with a ref to the work item struct to prevent it being deallocated whilst
on the queue or whilst it is being processed.  This is tricky to manage as
we have to get a ref before we try and queue it and then we may find it's
already queued and is thus already holding a ref - in which case we have to
try and get rid of the ref again.

The problem comes if we're in BH or IRQ context and need to drop the ref:
if netfs_put_request() reduces the count to 0, we have to do the cleanup -
but the cleanup may need to wait.

Fix this by adding a new work item to the request, ->cleanup_work, and
dispatching that when the refcount hits zero.  That can then synchronously
cancel any outstanding work on the main work item before doing the cleanup.

Adding a new work item also deals with another problem upstream where it's
sometimes changing the work func in the put function and requeuing it -
which has occasionally in the past caused the cleanup to happen
incorrectly.

As a bonus, this allows us to get rid of the 'was_async' parameter from a
bunch of functions.  This indicated whether the put function might not be
permitted to sleep.

Fixes: 3d3c95046742 ("netfs: Provide readahead and readpage netfs helpers")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/20250519090707.2848510-4-dhowells@redhat.com
cc: Paulo Alcantara <pc@manguebit.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: Steve French <stfrench@microsoft.com>
cc: linux-cifs@vger.kernel.org
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>

1 files changed, 28 insertions, 33 deletions

diff --git a/fs/netfs/write_collect.c b/fs/netfs/write_collect.c
index 3fca59e6475d..7241d1fd2c14 100644
--- a/fs/netfs/write_collect.c
+++ b/fs/netfs/write_collect.c
@@ -280,7 +280,7 @@ reassess_streams:
 							 struct netfs_io_subrequest, rreq_link);
 			stream->front = front;
 			spin_unlock(&wreq->lock);
-			netfs_put_subrequest(remove, false,
+			netfs_put_subrequest(remove,
 					     notes & SAW_FAILURE ?
 					     netfs_sreq_trace_put_cancel :
 					     netfs_sreq_trace_put_done);
@@ -356,30 +356,21 @@ need_retry:
 /*
  * Perform the collection of subrequests, folios and encryption buffers.
  */
-void netfs_write_collection_worker(struct work_struct *work)
+static bool netfs_write_collection(struct netfs_io_request *wreq)
 {
-	struct netfs_io_request *wreq = container_of(work, struct netfs_io_request, work);
 	struct netfs_inode *ictx = netfs_inode(wreq->inode);
 	size_t transferred;
 	int s;
 
 	_enter("R=%x", wreq->debug_id);
 
-	netfs_see_request(wreq, netfs_rreq_trace_see_work);
-	if (!test_bit(NETFS_RREQ_IN_PROGRESS, &wreq->flags)) {
-		netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
-		return;
-	}
-
 	netfs_collect_write_results(wreq);
 
 	/* We're done when the app thread has finished posting subreqs and all
 	 * the queues in all the streams are empty.
 	 */
-	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags)) {
-		netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
-		return;
-	}
+	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags))
+		return false;
 	smp_rmb(); /* Read ALL_QUEUED before lists. */
 
 	transferred = LONG_MAX;
@@ -387,10 +378,8 @@ void netfs_write_collection_worker(struct work_struct *work)
 		struct netfs_io_stream *stream = &wreq->io_streams[s];
 		if (!stream->active)
 			continue;
-		if (!list_empty(&stream->subrequests)) {
-			netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
-			return;
-		}
+		if (!list_empty(&stream->subrequests))
+			return false;
 		if (stream->transferred < transferred)
 			transferred = stream->transferred;
 	}
@@ -430,6 +419,7 @@ void netfs_write_collection_worker(struct work_struct *work)
 	_debug("finished");
 	trace_netfs_rreq(wreq, netfs_rreq_trace_wake_ip);
 	clear_and_wake_up_bit(NETFS_RREQ_IN_PROGRESS, &wreq->flags);
+	/* As we cleared NETFS_RREQ_IN_PROGRESS, we acquired its ref. */
 
 	if (wreq->iocb) {
 		size_t written = min(wreq->transferred, wreq->len);
@@ -440,27 +430,36 @@ void netfs_write_collection_worker(struct work_struct *work)
 		wreq->iocb = VFS_PTR_POISON;
 	}
 
-	netfs_clear_subrequests(wreq, false);
-	netfs_put_request(wreq, false, netfs_rreq_trace_put_work_complete);
+	netfs_clear_subrequests(wreq);
+	return true;
+}
+
+void netfs_write_collection_worker(struct work_struct *work)
+{
+	struct netfs_io_request *rreq = container_of(work, struct netfs_io_request, work);
+
+	netfs_see_request(rreq, netfs_rreq_trace_see_work);
+	if (test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags)) {
+		if (netfs_write_collection(rreq))
+			/* Drop the ref from the IN_PROGRESS flag. */
+			netfs_put_request(rreq, netfs_rreq_trace_put_work_ip);
+		else
+			netfs_see_request(rreq, netfs_rreq_trace_see_work_complete);
+	}
 }
 
 /*
  * Wake the collection work item.
  */
-void netfs_wake_write_collector(struct netfs_io_request *wreq, bool was_async)
+void netfs_wake_write_collector(struct netfs_io_request *wreq)
 {
-	if (!work_pending(&wreq->work)) {
-		netfs_get_request(wreq, netfs_rreq_trace_get_work);
-		if (!queue_work(system_unbound_wq, &wreq->work))
-			netfs_put_request(wreq, was_async, netfs_rreq_trace_put_work_nq);
-	}
+	queue_work(system_unbound_wq, &wreq->work);
 }
 
 /**
  * netfs_write_subrequest_terminated - Note the termination of a write operation.
  * @_op: The I/O request that has terminated.
  * @transferred_or_error: The amount of data transferred or an error code.
- * @was_async: The termination was asynchronous
  *
  * This tells the library that a contributory write I/O operation has
  * terminated, one way or another, and that it should collect the results.
@@ -470,17 +469,13 @@ void netfs_wake_write_collector(struct netfs_io_request *wreq, bool was_async)
  * negative error code.  The library will look after reissuing I/O operations
  * as appropriate and writing downloaded data to the cache.
  *
- * If @was_async is true, the caller might be running in softirq or interrupt
- * context and we can't sleep.
- *
  * When this is called, ownership of the subrequest is transferred back to the
  * library, along with a ref.
  *
  * Note that %_op is a void* so that the function can be passed to
  * kiocb::term_func without the need for a casting wrapper.
  */
-void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error,
-				       bool was_async)
+void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error)
 {
 	struct netfs_io_subrequest *subreq = _op;
 	struct netfs_io_request *wreq = subreq->rreq;
@@ -543,8 +538,8 @@ void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error,
 	 * transferring a ref to it if we were the ones to do so.
 	 */
 	if (list_is_first(&subreq->rreq_link, &stream->subrequests))
-		netfs_wake_write_collector(wreq, was_async);
+		netfs_wake_write_collector(wreq);
 
-	netfs_put_subrequest(subreq, was_async, netfs_sreq_trace_put_terminated);
+	netfs_put_subrequest(subreq, netfs_sreq_trace_put_terminated);
 }
 EXPORT_SYMBOL(netfs_write_subrequest_terminated);