nvme: do not let the user delete a ctrl before a complete initialization
authorMaurizio Lombardi <mlombard@redhat.com>
Thu, 11 May 2023 11:07:41 +0000 (13:07 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 9 Jun 2023 08:34:15 +0000 (10:34 +0200)
[ Upstream commit 2eb94dd56a4a4e3fe286def3e2ba207804a37345 ]

If a userspace application performes a "delete_controller" command
early during the ctrl initialization, the delete operation
may race against the init code and the kernel will crash.

nvme nvme5: Connect command failed: host path error
nvme nvme5: failed to connect queue: 0 ret=880
PF: supervisor write access in kernel mode
PF: error_code(0x0002) - not-present page
 blk_mq_quiesce_queue+0x18/0x90
 nvme_tcp_delete_ctrl+0x24/0x40 [nvme_tcp]
 nvme_do_delete_ctrl+0x7f/0x8b [nvme_core]
 nvme_sysfs_delete.cold+0x8/0xd [nvme_core]
 kernfs_fop_write_iter+0x124/0x1b0
 new_sync_write+0xff/0x190
 vfs_write+0x1ef/0x280

Fix the crash by checking the NVME_CTRL_STARTED_ONCE bit;
if it's not set it means that the nvme controller is still
in the process of getting initialized and the kernel
will return an -EBUSY error to userspace.
Set the NVME_CTRL_STARTED_ONCE later in the nvme_start_ctrl()
function, after the controller start operation is completed.

Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/nvme/host/core.c

index f502e03..2e22c78 100644 (file)
@@ -3540,6 +3540,9 @@ static ssize_t nvme_sysfs_delete(struct device *dev,
 {
        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
 
+       if (!test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags))
+               return -EBUSY;
+
        if (device_remove_file_self(dev, attr))
                nvme_delete_ctrl_sync(ctrl);
        return count;
@@ -4980,7 +4983,7 @@ void nvme_start_ctrl(struct nvme_ctrl *ctrl)
         * that were missed. We identify persistent discovery controllers by
         * checking that they started once before, hence are reconnecting back.
         */
-       if (test_and_set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
+       if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
            nvme_discovery_ctrl(ctrl))
                nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
 
@@ -4991,6 +4994,7 @@ void nvme_start_ctrl(struct nvme_ctrl *ctrl)
        }
 
        nvme_change_uevent(ctrl, "NVME_EVENT=connected");
+       set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
 }
 EXPORT_SYMBOL_GPL(nvme_start_ctrl);