On 05/09/2016 06:50 AM, Joe Lawrence wrote:
> On 05/05/2016 04:40 PM, Joe Lawrence wrote:
>> On 05/05/2016 03:58 PM, Bart Van Assche wrote:
>>> On 03/28/2016 02:29 PM, Bart Van Assche wrote:
>>>> Avoid that the sd driver registers a BDI device with a name that
>>>> is still in use. This patch avoids that the following warning gets
>>>> triggered:
>>>>
>>>> [ ... ]
>>>
>>> (replying to my own e-mail)
>>>
>>> If anyone could review this patch that would be very welcome.
>>
>> I *think* I may be hitting this same problem running some tests here at Stratus
>> ... snip...
>
> Good news = With your v3 patch, I didn't see the "sysfs: cannot create
> duplicate filename '/devices/virtual/bdi/65:0'" warning during my
> weekend testing (573 surprise disk HBA removals).
>
> Bad news = I still crashed in add_disk > sysfs_create_link >
> sysfs_do_create_link_sd on a NULL target_kobj->sd ... unfortunately I
> don't have kdump working, so all I have is a serial console output to
> work with for now.
(replying to an e-mail of one month ago)
Hello Joe,
Earlier today I discovered a subtle bug in v3 of this patch. It would
be appreciated if you could give v4 a try. The only substantial
difference between v3 and v4 is that a "if (dev->class != &sdev_class)"
test has been added.
Thanks,
Bart.
[PATCH] Fix a bdi reregistration race, v4
Avoid that the sd driver registers a BDI device with a name that
is still in use. This patch avoids that the following warning gets
triggered:
WARNING: CPU: 7 PID: 203 at fs/sysfs/dir.c:31 sysfs_warn_dup+0x68/0x80()
sysfs: cannot create duplicate filename '/devices/virtual/bdi/8:32'
Workqueue: events_unbound async_run_entry_fn
Call Trace:
[<ffffffff814ff5a4>] dump_stack+0x4c/0x65
[<ffffffff810746ba>] warn_slowpath_common+0x8a/0xc0
[<ffffffff81074736>] warn_slowpath_fmt+0x46/0x50
[<ffffffff81237ca8>] sysfs_warn_dup+0x68/0x80
[<ffffffff81237d8e>] sysfs_create_dir_ns+0x7e/0x90
[<ffffffff81291f58>] kobject_add_internal+0xa8/0x320
[<ffffffff812923a0>] kobject_add+0x60/0xb0
[<ffffffff8138c937>] device_add+0x107/0x5e0
[<ffffffff8138d018>] device_create_groups_vargs+0xd8/0x100
[<ffffffff8138d05c>] device_create_vargs+0x1c/0x20
[<ffffffff8117f233>] bdi_register+0x63/0x2a0
[<ffffffff8117f497>] bdi_register_dev+0x27/0x30
[<ffffffff81281549>] add_disk+0x1a9/0x4e0
[<ffffffffa00c5739>] sd_probe_async+0x119/0x1d0 [sd_mod]
[<ffffffff8109a81a>] async_run_entry_fn+0x4a/0x140
[<ffffffff81091078>] process_one_work+0x1d8/0x7c0
[<ffffffff81091774>] worker_thread+0x114/0x460
[<ffffffff81097878>] kthread+0xf8/0x110
[<ffffffff8150801f>] ret_from_fork+0x3f/0x70
Signed-off-by: Bart Van Assche <bart.vanassche@xxxxxxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Hannes Reinecke <hare@xxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx>
---
drivers/scsi/scsi_sysfs.c | 39 ++++++++++++++++++++++++++++++++++++++-
1 file changed, 38 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
index 0734927..ac10f0c 100644
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -1273,9 +1273,35 @@ int scsi_sysfs_add_sdev(struct scsi_device *sdev)
return error;
}
+/**
+ * scsi_filter_sd - Look up the device structure embedded in a disk structure
+ * @dev: A sdev_gendev device
+ * @data: A struct device pointer
+ *
+ * sdev_gendev devices have two children - the sdev_dev device and for SCSI
+ * disks, the device embedded in a scsi_disk.
+ */
+static int scsi_filter_sd(struct device *dev, void *data)
+{
+ struct device **childp = data;
+
+ if (dev->class != &sdev_class)
+ *childp = dev;
+ return 0;
+}
+
+/* Caller must call put_device() if this function does not return NULL. */
+static struct device *scsi_get_sd(struct device *dev)
+{
+ struct device *child = NULL;
+
+ device_for_each_child(dev, &child, scsi_filter_sd);
+ return get_device(child);
+}
+
void __scsi_remove_device(struct scsi_device *sdev)
{
- struct device *dev = &sdev->sdev_gendev;
+ struct device *dev = &sdev->sdev_gendev, *sdp = NULL;
/*
* This cleanup path is not reentrant and while it is impossible
@@ -1290,6 +1316,7 @@ void __scsi_remove_device(struct scsi_device *sdev)
return;
bsg_unregister_queue(sdev->request_queue);
+ sdp = scsi_get_sd(dev);
device_unregister(&sdev->sdev_dev);
transport_remove_device(dev);
scsi_dh_remove_device(sdev);
@@ -1306,6 +1333,16 @@ void __scsi_remove_device(struct scsi_device *sdev)
blk_cleanup_queue(sdev->request_queue);
cancel_work_sync(&sdev->requeue_work);
+ /*
+ * blk_cleanup_queue() unregisters the BDI device. The name of the
+ * BDI device is derived from the dev_t of the /dev/sd<n> device.
+ * Keep a reference to the /dev/sd<n> device until the BDI device
+ * has been unregistered to avoid that a BDI device with the same
+ * name gets registered before blk_cleanup_queue() has finished.
+ */
+ if (sdp)
+ put_device(sdp);
+
if (sdev->host->hostt->slave_destroy)
sdev->host->hostt->slave_destroy(sdev);
transport_destroy_device(dev);
--
2.8.3
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
