Re: SCSI: race condition between scsi_remove_target and scsi_probe_and_add_lun

[Alexey Ivanov]

+ linux-scsi

> On Mar 20, 2014, at 6:42 PM, Andrey Zonov <andrey.zonov@xxxxxxxxx> wrote:
> 
> Hi,
> 
> I've got kernel panic on my box which works as FibreChannel initiator.
> I was able to reproduce this panic by setting dev_loss_tmo=2 and
> enabling/disabling ports every 5 seconds on the switch in 5 minutes.  I
> added some debug points in the kernel code and that's what I've got so far:
> 
> 1. system is inserting new device into __devices list
> 
> DEBUG: scsi_sysfs_device_initialize(): sdev=ffff88046a931000 7:0:5:0
> Pid: 910, comm: kworker/u:2 Tainted: P           O 3.2.48-swt9004 #33
> Call Trace:
> [<ffffffff81245c42>] ? scsi_alloc_sdev+0x1d2/0x240
> [<ffffffff8123d4bd>] ? scsi_device_lookup_by_target+0x8d/0xc0
> [<ffffffff8124623a>] ? scsi_probe_and_add_lun+0x42a/0xb20
> [<ffffffff811acb7d>] ? kobject_set_name_vargs+0x6d/0x80
> [<ffffffff81230b4f>] ? dev_set_name+0x3f/0x50
> [<ffffffff811ac782>] ? kobject_get+0x12/0x20
> [<ffffffffa000b3e4>] ? fc_host_match+0x14/0x70 [scsi_transport_fc]
> [<ffffffff8123726f>] ? attribute_container_add_device+0x4f/0x160
> [<ffffffff811ac782>] ? kobject_get+0x12/0x20
> [<ffffffff812304e4>] ? get_device+0x14/0x20
> [<ffffffff812459c5>] ? scsi_alloc_target+0x295/0x2d0
> [<ffffffff81230bca>] ? device_release+0x1a/0x80
> [<ffffffff81246bce>] ? __scsi_scan_target+0xce/0x5f0
> [<ffffffff8102ec22>] ? dequeue_task_fair+0x52/0x150
> [<ffffffff8139a91d>] ? __schedule+0x25d/0x7d0
> [<ffffffff812477b6>] ? scsi_scan_target+0xc6/0xe0
> [<ffffffffa000e75f>] ? fc_scsi_scan_rport+0xaf/0xc0 [scsi_transport_fc]
> [<ffffffff8104eb06>] ? process_one_work+0x116/0x3a0
> [<ffffffff8104f1ec>] ? worker_thread+0x14c/0x400
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff81052f06>] ? kthread+0x96/0xa0
> [<ffffffff8139f134>] ? kernel_thread_helper+0x4/0x10
> [<ffffffff81052e70>] ? kthread_worker_fn+0x120/0x120
> [<ffffffff8139f130>] ? gs_change+0xb/0xb
> 
> 2. later in scsi_probe_and_add_lun() this device is removing
> 
> DEBUG: __scsi_remove_device(): sdev=ffff88046a931000 7:0:5:0
> Pid: 910, comm: kworker/u:2 Tainted: P           O 3.2.48-swt9004 #33
> Call Trace:
> [<ffffffff81248c66>] ? __scsi_remove_device+0x46/0x110
> [<ffffffff81246268>] ? scsi_probe_and_add_lun+0x458/0xb20
> [<ffffffff81230b4f>] ? dev_set_name+0x3f/0x50
> [<ffffffff811ac782>] ? kobject_get+0x12/0x20
> [<ffffffff812459c5>] ? scsi_alloc_target+0x295/0x2d0
> [<ffffffff81230bca>] ? device_release+0x1a/0x80
> [<ffffffff81246bce>] ? __scsi_scan_target+0xce/0x5f0
> [<ffffffff8102ec22>] ? dequeue_task_fair+0x52/0x150
> [<ffffffff8139a91d>] ? __schedule+0x25d/0x7d0
> [<ffffffff812477b6>] ? scsi_scan_target+0xc6/0xe0
> [<ffffffffa000e75f>] ? fc_scsi_scan_rport+0xaf/0xc0 [scsi_transport_fc]
> [<ffffffff8104eb06>] ? process_one_work+0x116/0x3a0
> [<ffffffff8104f1ec>] ? worker_thread+0x14c/0x400
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff81052f06>] ? kthread+0x96/0xa0
> [<ffffffff8139f134>] ? kernel_thread_helper+0x4/0x10
> [<ffffffff81052e70>] ? kthread_worker_fn+0x120/0x120
> [<ffffffff8139f130>] ? gs_change+0xb/0xb
> 
> 3. another thread is trying to remove this device because of timeout
> 
> DEBUG: __scsi_remove_device(): sdev=ffff88046a931000 7:0:5:0
> Pid: 4, comm: kworker/0:0 Tainted: P           O 3.2.48-swt9004 #33
> Call Trace:
> [<ffffffff81248c66>] ? __scsi_remove_device+0x46/0x110
> [<ffffffff8139b63a>] ? mutex_lock+0x1a/0x40
> [<ffffffff81248d58>] ? scsi_remove_device+0x28/0x40
> [<ffffffff81242a00>] ? scsi_kmap_atomic_sg+0x180/0x180
> [<ffffffff81248ed1>] ? scsi_remove_target+0x141/0x1e0
> [<ffffffff8104eb06>] ? process_one_work+0x116/0x3a0
> [<ffffffff8104f1ec>] ? worker_thread+0x14c/0x400
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff81052f06>] ? kthread+0x96/0xa0
> [<ffffffff8139f134>] ? kernel_thread_helper+0x4/0x10
> [<ffffffff81052e70>] ? kthread_worker_fn+0x120/0x120
> [<ffffffff8139f130>] ? gs_change+0xb/0xb
> 
> and it's got dead sdev object.  I don't understand how this can happen
> because __scsi_remove_target() iterating over __devices and getting sdev
> reference under host_lock and that should be enough.
> 
> DEBUG: kref_put(): kref=ffff88046a9312e0 val=-1
> ------------[ cut here ]------------
> WARNING: at lib/kref.c:61 kref_put+0x88/0xc0()
> Hardware name: X9DRi-LN4+/X9DR3-LN4+
> Modules linked in: qla2xxx(O) igb ehci_hcd scsi_transport_fc
> Pid: 4, comm: kworker/0:0 Tainted: P           O 3.2.48-swt9004 #33
> Call Trace:
> [<ffffffff81037bfb>] ? warn_slowpath_common+0x7b/0xc0
> [<ffffffff811ac6c0>] ? kobject_del+0x30/0x30
> [<ffffffff811adb08>] ? kref_put+0x88/0xc0
> [<ffffffff81248cac>] ? __scsi_remove_device+0x8c/0x110
> [<ffffffff8139b63a>] ? mutex_lock+0x1a/0x40
> [<ffffffff81248d58>] ? scsi_remove_device+0x28/0x40
> [<ffffffff81242a00>] ? scsi_kmap_atomic_sg+0x180/0x180
> [<ffffffff81248ed1>] ? scsi_remove_target+0x141/0x1e0
> [<ffffffff8104eb06>] ? process_one_work+0x116/0x3a0
> [<ffffffff8104f1ec>] ? worker_thread+0x14c/0x400
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff8104f0a0>] ? rescuer_thread+0x310/0x310
> [<ffffffff81052f06>] ? kthread+0x96/0xa0
> [<ffffffff8139f134>] ? kernel_thread_helper+0x4/0x10
> [<ffffffff81052e70>] ? kthread_worker_fn+0x120/0x120
> [<ffffffff8139f130>] ? gs_change+0xb/0xb
> 
> Here is the patch which helped me:
> 
> diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
> index 9117d0b..676e5ff 100644
> --- a/drivers/scsi/scsi_sysfs.c
> +++ b/drivers/scsi/scsi_sysfs.c
> @@ -1094,6 +1094,7 @@ static void __scsi_remove_target(struct
> scsi_target *starget)
>        unsigned long flags;
>        struct scsi_device *sdev;
> 
> +       mutex_lock(&shost->scan_mutex);
>        spin_lock_irqsave(shost->host_lock, flags);
>  restart:
>        list_for_each_entry(sdev, &shost->__devices, siblings) {
> @@ -1102,12 +1103,13 @@ static void __scsi_remove_target(struct
> scsi_target *starget)
>                    scsi_device_get(sdev))
>                        continue;
>                spin_unlock_irqrestore(shost->host_lock, flags);
> -               scsi_remove_device(sdev);
> +               __scsi_remove_device(sdev);
>                scsi_device_put(sdev);
>                spin_lock_irqsave(shost->host_lock, flags);
>                goto restart;
>        }
>        spin_unlock_irqrestore(shost->host_lock, flags);
> +       mutex_unlock(&shost->scan_mutex);
> }
> 
> /**
> 
> I'm not sure about the fix is correct, but I was not able to reproduce
> the panic.
> 
> 
> P.S. Here is another patch which help to detect reference count underflow
> 
> diff --git a/include/linux/kref.h b/include/linux/kref.h
> index 484604d..05dd2b3 100644
> --- a/include/linux/kref.h
> +++ b/include/linux/kref.h
> @@ -68,9 +68,13 @@ static inline void kref_get(struct kref *kref)
> static inline int kref_sub(struct kref *kref, unsigned int count,
>             void (*release)(struct kref *kref))
> {
> +       long refs;
> +
>        WARN_ON(release == NULL);
> 
> -       if (atomic_sub_and_test((int) count, &kref->refcount)) {
> +       refs = atomic_sub_return((int) count, &kref->refcount);
> +       WARN_ON(refs < 0);
> +       if (refs == 0) {
>                release(kref);
>                return 1;
>        }
> 
> -- 
> Andrey Zonov
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/