[PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

[Casey Chen]

Below two paths could overlap each other if we power off a drive quickly
after power it on. There are multiple races in nvme_setup_io_queues()
because of shutdown_lock missing and improper use of NVMEQ_ENABLED bit.

nvme_reset_work()                                nvme_remove()
  nvme_setup_io_queues()                           nvme_dev_disable()
  ...                                              ...
A1  clear NVMEQ_ENABLED bit for admin queue          lock
    retry:                                       B1  nvme_suspend_io_queues()
A2    pci_free_irq() admin queue                 B2  nvme_suspend_queue() admin queue
A3    pci_free_irq_vectors()                         nvme_pci_disable()
A4    nvme_setup_irqs();                         B3    pci_free_irq_vectors()
      ...                                            unlock
A5    queue_request_irq() for admin queue
      set NVMEQ_ENABLED bit
      ...
      nvme_create_io_queues()
A6      result = queue_request_irq();
        set NVMEQ_ENABLED bit
      ...
      fail to allocate enough IO queues:
A7      nvme_suspend_io_queues()
        goto retry

1). If B3 runs in between A1 and A2, it will crash if irqaction haven't
  been freed by A2. B2 is supposed to free admin queue IRQ but it simply
  can't fulfill the job as A1 has cleared NVMEQ_ENABLED bit.

Fix: combine A1 A2 so IRQ get freed as soon as NVMEQ_ENABLED bit get cleared.

2). After solved #1, A2 could race with B3 if A2 is freeing IRQ while B3
  is checking irqaction. A3 also could race with B2 if B2 is freeing
  IRQ while A3 is checking irqaction.

Fix: A2 and A3 take lock for mutual exclusion.

3) A3 could race with B3 since they could run free_msi_irqs() in parallel.

Fix: A3 takes lock for mutual exclusion.

4). A4 could fail to allocate all needed IRQ vectors if A3 and A4 are
  interrupted by B3.

Fix: A4 takes lock for mutual exclusion.

5). If A5/A6 happened after B2/B1, B3 will crash since irqaction is not NULL.
  They are just allocated by A5/A6.

Fix: Lock queue_request_irq() and setting of NVMEQ_ENABLED bit.

6). A7 could get chance to pci_free_irq() for certain IO queue while B3
  is checking irqaction.

Fix: A7 takes lock.

7). nvme_dev->online_queues need to be protected by shutdown_lock. Since it
  is not atomic, both paths could modify it using its own copy.

Co-developed-by: Yuanyuan Zhong <yzhong at purestorage.com>
Signed-off-by: Casey Chen <cachen at purestorage.com>
---
 drivers/nvme/host/pci.c | 37 +++++++++++++++++++++++++++++++------
 1 file changed, 31 insertions(+), 6 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 3aa7245a505f..5abc4c3be454 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1590,8 +1590,13 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 		goto release_cq;
 
 	nvmeq->cq_vector = vector;
-	nvme_init_queue(nvmeq, qid);
 
+	/*
+	 * Return early if locking fails as device is being disabled.
+	 */
+	if (!mutex_trylock(&dev->shutdown_lock))
+		return -ENODEV;
+	nvme_init_queue(nvmeq, qid);
 	if (!polled) {
 		result = queue_request_irq(nvmeq);
 		if (result < 0)
@@ -1599,10 +1604,12 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 	}
 
 	set_bit(NVMEQ_ENABLED, &nvmeq->flags);
+	mutex_unlock(&dev->shutdown_lock);
 	return result;
 
 release_sq:
 	dev->online_queues--;
+	mutex_unlock(&dev->shutdown_lock);
 	adapter_delete_sq(dev, qid);
 release_cq:
 	adapter_delete_cq(dev, qid);
@@ -2176,7 +2183,16 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	if (nr_io_queues == 0)
 		return 0;
 
-	clear_bit(NVMEQ_ENABLED, &adminq->flags);
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock);
+	if (test_and_clear_bit(NVMEQ_ENABLED, &adminq->flags))
+		pci_free_irq(pdev, 0, adminq);
 
 	if (dev->cmb_use_sqes) {
 		result = nvme_cmb_qdepth(dev, nr_io_queues,
@@ -2192,14 +2208,17 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 		result = nvme_remap_bar(dev, size);
 		if (!result)
 			break;
-		if (!--nr_io_queues)
+		if (!--nr_io_queues) {
+			mutex_unlock(&dev->shutdown_lock);
 			return -ENOMEM;
+		}
 	} while (1);
 	adminq->q_db = dev->dbs;
 
  retry:
 	/* Deregister the admin queue's interrupt */
-	pci_free_irq(pdev, 0, adminq);
+	if (test_and_clear_bit(NVMEQ_ENABLED, &adminq->flags))
+		pci_free_irq(pdev, 0, adminq);
 
 	/*
 	 * If we enable msix early due to not intx, disable it again before
@@ -2208,8 +2227,10 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	pci_free_irq_vectors(pdev);
 
 	result = nvme_setup_irqs(dev, nr_io_queues);
-	if (result <= 0)
+	if (result <= 0) {
+		mutex_unlock(&dev->shutdown_lock);
 		return -EIO;
+	}
 
 	dev->num_vecs = result;
 	result = max(result - 1, 1);
@@ -2222,9 +2243,12 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	 * number of interrupts.
 	 */
 	result = queue_request_irq(adminq);
-	if (result)
+	if (result) {
+		mutex_unlock(&dev->shutdown_lock);
 		return result;
+	}
 	set_bit(NVMEQ_ENABLED, &adminq->flags);
+	mutex_unlock(&dev->shutdown_lock);
 
 	result = nvme_create_io_queues(dev);
 	if (result || dev->online_queues < 2)
@@ -2233,6 +2257,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	if (dev->online_queues - 1 < dev->max_qid) {
 		nr_io_queues = dev->online_queues - 1;
 		nvme_disable_io_queues(dev);
+		mutex_lock(&dev->shutdown_lock);
 		nvme_suspend_io_queues(dev);
 		goto retry;
 	}
-- 
2.17.1