kernfs: kernfs_notify() must be useable from non-sleepable contexts

d911d98 ("kernfs: make kernfs_notify() trigger inotify events
too") added fsnotify triggering to kernfs_notify() which requires a
sleepable context.  There are already existing users of
kernfs_notify() which invoke it from an atomic context and in general
it's silly to require a sleepable context for triggering a
notification.

The following is an invalid context bug triggerd by md invoking
sysfs_notify() from IO completion path.

 BUG: sleeping function called from invalid context at kernel/locking/mutex.c:586
 in_atomic(): 1, irqs_disabled(): 1, pid: 0, name: swapper/1
 2 locks held by swapper/1/0:
  #0:  (&(&vblk->vq_lock)->rlock){-.-...}, at: [<ffffffffa0039042>] virtblk_done+0x42/0xe0 [virtio_blk]
  #1:  (&(&bitmap->counts.lock)->rlock){-.....}, at: [<ffffffff81633718>] bitmap_endwrite+0x68/0x240
 irq event stamp: 33518
 hardirqs last  enabled at (33515): [<ffffffff8102544f>] default_idle+0x1f/0x230
 hardirqs last disabled at (33516): [<ffffffff818122ed>] common_interrupt+0x6d/0x72
 softirqs last  enabled at (33518): [<ffffffff810a1272>] _local_bh_enable+0x22/0x50
 softirqs last disabled at (33517): [<ffffffff810a29e0>] irq_enter+0x60/0x80
 CPU: 1 PID: 0 Comm: swapper/1 Not tainted 3.16.0-0.rc2.git2.1.fc21.x86_64 #1
 Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
  0000000000000000 f90db13964f4ee05 ffff88007d403b80 ffffffff81807b4c
  0000000000000000 ffff88007d403ba8 ffffffff810d4f14 0000000000000000
  0000000000441800 ffff880078fa1780 ffff88007d403c38 ffffffff8180caf2
 Call Trace:
  <IRQ>  [<ffffffff81807b4c>] dump_stack+0x4d/0x66
  [<ffffffff810d4f14>] __might_sleep+0x184/0x240
  [<ffffffff8180caf2>] mutex_lock_nested+0x42/0x440
  [<ffffffff812d76a0>] kernfs_notify+0x90/0x150
  [<ffffffff8163377c>] bitmap_endwrite+0xcc/0x240
  [<ffffffffa00de863>] close_write+0x93/0xb0 [raid1]
  [<ffffffffa00df029>] r1_bio_write_done+0x29/0x50 [raid1]
  [<ffffffffa00e0474>] raid1_end_write_request+0xe4/0x260 [raid1]
  [<ffffffff813acb8b>] bio_endio+0x6b/0xa0
  [<ffffffff813b46c4>] blk_update_request+0x94/0x420
  [<ffffffff813bf0ea>] blk_mq_end_io+0x1a/0x70
  [<ffffffffa00392c2>] virtblk_request_done+0x32/0x80 [virtio_blk]
  [<ffffffff813c0648>] __blk_mq_complete_request+0x88/0x120
  [<ffffffff813c070a>] blk_mq_complete_request+0x2a/0x30
  [<ffffffffa0039066>] virtblk_done+0x66/0xe0 [virtio_blk]
  [<ffffffffa002535a>] vring_interrupt+0x3a/0xa0 [virtio_ring]
  [<ffffffff81116177>] handle_irq_event_percpu+0x77/0x340
  [<ffffffff8111647d>] handle_irq_event+0x3d/0x60
  [<ffffffff81119436>] handle_edge_irq+0x66/0x130
  [<ffffffff8101c3e4>] handle_irq+0x84/0x150
  [<ffffffff818146ad>] do_IRQ+0x4d/0xe0
  [<ffffffff818122f2>] common_interrupt+0x72/0x72
  <EOI>  [<ffffffff8105f706>] ? native_safe_halt+0x6/0x10
  [<ffffffff81025454>] default_idle+0x24/0x230
  [<ffffffff81025f9f>] arch_cpu_idle+0xf/0x20
  [<ffffffff810f5adc>] cpu_startup_entry+0x37c/0x7b0
  [<ffffffff8104df1b>] start_secondary+0x25b/0x300

This patch fixes it by punting the notification delivery through a
work item.  This ends up adding an extra pointer to kernfs_elem_attr
enlarging kernfs_node by a pointer, which is not ideal but not a very
big deal either.  If this turns out to be an actual issue, we can move
kernfs_elem_attr->size to kernfs_node->iattr later.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Josh Boyer <jwboyer@fedoraproject.org>
Cc: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/kernfs/file.c

@@ -39,6 +39,19 @@ struct kernfs_open_node {
 	struct list_head	files; /* goes through kernfs_open_file.list */
 };
 
+/*
+ * kernfs_notify() may be called from any context and bounces notifications
+ * through a work item.  To minimize space overhead in kernfs_node, the
+ * pending queue is implemented as a singly linked list of kernfs_nodes.
+ * The list is terminated with the self pointer so that whether a
+ * kernfs_node is on the list or not can be determined by testing the next
+ * pointer for NULL.
+ */
+#define KERNFS_NOTIFY_EOL			((void *)&kernfs_notify_list)
+
+static DEFINE_SPINLOCK(kernfs_notify_lock);
+static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL;
+
 static struct kernfs_open_file *kernfs_of(struct file *file)
 {
 	return ((struct seq_file *)file->private_data)->private;
@@ -783,37 +796,38 @@ static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait)
 	return DEFAULT_POLLMASK|POLLERR|POLLPRI;
 }
 
-/**
- * kernfs_notify - notify a kernfs file
- * @kn: file to notify
- *
- * Notify @kn such that poll(2) on @kn wakes up.
- */
-void kernfs_notify(struct kernfs_node *kn)
+static void kernfs_notify_workfn(struct work_struct *work)
 {
-	struct kernfs_root *root = kernfs_root(kn);
+	struct kernfs_node *kn;
 	struct kernfs_open_node *on;
 	struct kernfs_super_info *info;
-	unsigned long flags;
-
-	if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
+repeat:
+	/* pop one off the notify_list */
+	spin_lock_irq(&kernfs_notify_lock);
+	kn = kernfs_notify_list;
+	if (kn == KERNFS_NOTIFY_EOL) {
+		spin_unlock_irq(&kernfs_notify_lock);
 		return;
+	}
+	kernfs_notify_list = kn->attr.notify_next;
+	kn->attr.notify_next = NULL;
+	spin_unlock_irq(&kernfs_notify_lock);
 
 	/* kick poll */
-	spin_lock_irqsave(&kernfs_open_node_lock, flags);
+	spin_lock_irq(&kernfs_open_node_lock);
 
 	on = kn->attr.open;
 	if (on) {
 		atomic_inc(&on->event);
 		wake_up_interruptible(&on->poll);
 	}
 
-	spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
+	spin_unlock_irq(&kernfs_open_node_lock);
 
 	/* kick fsnotify */
 	mutex_lock(&kernfs_mutex);
 
-	list_for_each_entry(info, &root->supers, node) {
+	list_for_each_entry(info, &kernfs_root(kn)->supers, node) {
 		struct inode *inode;
 		struct dentry *dentry;
 
@@ -833,6 +847,33 @@ void kernfs_notify(struct kernfs_node *kn)
 	}
 
 	mutex_unlock(&kernfs_mutex);
+	kernfs_put(kn);
+	goto repeat;
+}
+
+/**
+ * kernfs_notify - notify a kernfs file
+ * @kn: file to notify
+ *
+ * Notify @kn such that poll(2) on @kn wakes up.  Maybe be called from any
+ * context.
+ */
+void kernfs_notify(struct kernfs_node *kn)
+{
+	static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn);
+	unsigned long flags;
+
+	if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
+		return;
+
+	spin_lock_irqsave(&kernfs_notify_lock, flags);
+	if (!kn->attr.notify_next) {
+		kernfs_get(kn);
+		kn->attr.notify_next = kernfs_notify_list;
+		kernfs_notify_list = kn;
+		schedule_work(&kernfs_notify_work);
+	}
+	spin_unlock_irqrestore(&kernfs_notify_lock, flags);
 }
 EXPORT_SYMBOL_GPL(kernfs_notify);
 

include/linux/kernfs.h

@@ -91,6 +91,7 @@ struct kernfs_elem_attr {
 	const struct kernfs_ops	*ops;
 	struct kernfs_open_node	*open;
 	loff_t			size;
+	struct kernfs_node	*notify_next;	/* for kernfs_notify() */
 };
 
 /*