Skip to contentSkip to conversation

nvme-pci: fix race condition between reset and nvme_dev_disable()

PATCH Applied Maurizio Lombardi 2 participants 4 messages Oct 15, 2024 – Oct 17, 2024 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 nvme-pci: fix race condition between reset and nvme_dev_disable()

nvme_dev_disable() modifies the dev->online_queues field, therefore nvme_pci_update_nr_queues() should avoid racing against it, otherwise we could end up passing invalid values to blk_mq_update_nr_hw_queues().

 WARNING: CPU: 39 PID: 61303 at drivers/pci/msi/api.c:347
          pci_irq_get_affinity+0x187/0x210
 Workqueue: nvme-reset-wq nvme_reset_work [nvme]
 RIP: 0010:pci_irq_get_affinity+0x187/0x210
 Call Trace:
  <TASK>
  ? blk_mq_pci_map_queues+0x87/0x3c0
  ? pci_irq_get_affinity+0x187/0x210
  blk_mq_pci_map_queues+0x87/0x3c0
  nvme_pci_map_queues+0x189/0x460 [nvme]
  blk_mq_update_nr_hw_queues+0x2a/0x40
  nvme_reset_work+0x1be/0x2a0 [nvme]

Fix the bug by locking the shutdown_lock mutex before using dev->online_queues. Give up if nvme_dev_disable() is running or if it has been executed already.

Fixes 949928c1c731 ("NVMe: Fix possible queue use after freed")

---

v2: return true in case of success, false in case of failure

 drivers/nvme/host/pci.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)
drivers/nvme/host/pci.c +16 -3
index 7990c3f22ecf..4b9fda0b1d9a 100644
@@ -2506,17 +2506,29 @@ static unsigned int nvme_pci_nr_maps(struct nvme_dev *dev)
 	return 1;
 }
 
-static void nvme_pci_update_nr_queues(struct nvme_dev *dev)
+static bool nvme_pci_update_nr_queues(struct nvme_dev *dev)
 {
 	if (!dev->ctrl.tagset) {
 		nvme_alloc_io_tag_set(&dev->ctrl, &dev->tagset, &nvme_mq_ops,
 				nvme_pci_nr_maps(dev), sizeof(struct nvme_iod));
-		return;
+		return true;
+	}
+
+	/* Give up if we are racing with nvme_dev_disable() */
+	if (!mutex_trylock(&dev->shutdown_lock))
+		return false;
+
+	/* Check if nvme_dev_disable() has been executed already */
+	if (!dev->online_queues) {
+		mutex_unlock(&dev->shutdown_lock);
+		return false;
 	}
 
 	blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1);
 	/* free previously allocated queues that are no longer usable */
 	nvme_free_queues(dev, dev->online_queues);
+	mutex_unlock(&dev->shutdown_lock);
+	return true;
 }
 
 static int nvme_pci_enable(struct nvme_dev *dev)
@@ -2797,7 +2809,8 @@ static void nvme_reset_work(struct work_struct *work)
 		nvme_dbbuf_set(dev);
 		nvme_unquiesce_io_queues(&dev->ctrl);
 		nvme_wait_freeze(&dev->ctrl);
-		nvme_pci_update_nr_queues(dev);
+		if (!nvme_pci_update_nr_queues(dev))
+			goto out;
 		nvme_unfreeze(&dev->ctrl);
 	} else {
 		dev_warn(dev->ctrl.device, "IO queues lost\n");
Signature · 1 line
2.43.5
maintainer
#1 lore raw
Quoted 14 lines · Maurizio Lombardi
drivers/nvme/host/pci.c +8 -2
-static void nvme_pci_update_nr_queues(struct nvme_dev *dev)
+static bool nvme_pci_update_nr_queues(struct nvme_dev *dev)
 {
 	if (!dev->ctrl.tagset) {
 		nvme_alloc_io_tag_set(&dev->ctrl, &dev->tagset, &nvme_mq_ops,
 				nvme_pci_nr_maps(dev), sizeof(struct nvme_iod));
-		return;
+		return true;
+	}
+
+	/* Give up if we are racing with nvme_dev_disable() */
+	if (!mutex_trylock(&dev->shutdown_lock))
+		return false;
+

drivers/nvme/host/pci.c static unsigned int nvme_pci_nr_maps(struct nvme_dev *dev) · line 2530

Quoted 10 lines
drivers/nvme/host/pci.c +5 -0
+	/* Check if nvme_dev_disable() has been executed already */
+	if (!dev->online_queues) {
+		mutex_unlock(&dev->shutdown_lock);
+		return false;
 	}
 
 	blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1);
 	/* free previously allocated queues that are no longer usable */
 	nvme_free_queues(dev, dev->online_queues);
+	mutex_unlock(&dev->shutdown_lock);

I believe mutex_unlock needs to be above blk_mq_update_nr_hw_queues(). That function needs all the queues to be frozen, so any older IO that times out is going to need this lock in order to reclaim it.

maintainer
#2 lore raw
Quoted 21 lines · Keith Busch
Quoted 20 lines · Maurizio Lombardi
drivers/nvme/host/pci.c +12 -2
-static void nvme_pci_update_nr_queues(struct nvme_dev *dev)
+static bool nvme_pci_update_nr_queues(struct nvme_dev *dev)
 {
 	if (!dev->ctrl.tagset) {
 		nvme_alloc_io_tag_set(&dev->ctrl, &dev->tagset, &nvme_mq_ops,
 				nvme_pci_nr_maps(dev), sizeof(struct nvme_iod));
-		return;
+		return true;
+	}
+
+	/* Give up if we are racing with nvme_dev_disable() */
+	if (!mutex_trylock(&dev->shutdown_lock))
+		return false;
+
+	/* Check if nvme_dev_disable() has been executed already */
+	if (!dev->online_queues) {
+		mutex_unlock(&dev->shutdown_lock);
+		return false;
 	}
Quoted 8 lines

drivers/nvme/host/pci.c static unsigned int nvme_pci_nr_maps(struct nvme_dev *dev) · line 2530

Quoted 4 lines
drivers/nvme/host/pci.c +1 -0
 	blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1);
 	/* free previously allocated queues that are no longer usable */
 	nvme_free_queues(dev, dev->online_queues);
+	mutex_unlock(&dev->shutdown_lock);

I believe mutex_unlock needs to be above blk_mq_update_nr_hw_queues(). That function needs all the queues to be frozen, so any older IO that times out is going to need this lock in order to reclaim it.

Oops, unlocking there doesn't solve your problem. I guess your patch should be safe as-is since the nvme driver already waits for the freeze prior to updating the hw queues.

maintainer Applied
#3 lore raw
In reply to Maurizio Lombardi
Quoted 3 lines · Maurizio Lombardi

nvme_dev_disable() modifies the dev->online_queues field, therefore nvme_pci_update_nr_queues() should avoid racing against it, otherwise we could end up passing invalid values to blk_mq_update_nr_hw_queues().

Thanks, applied to nvme-6.12.