Re: [PATCH 4/4] UBI: Implement bitrot checking

[Richard Weinberger]

Am 12.04.2015 um 16:12 schrieb Boris Brezillon:
> Hi Richard,
> 
> Sorry for the late reply.
> 
> On Sun, 29 Mar 2015 14:13:17 +0200
> Richard Weinberger <richard@xxxxxx> wrote:
> 
>> This patch implements bitrot checking for UBI.
>> ubi_wl_trigger_bitrot_check() triggers a re-read of every
>> PEB. If a bitflip is detected PEBs in use will get scrubbed
>> and free ones erased.
> 
> As you'll see, I didn't have much to say about the 'UBI bitrot
> detection' mechanism, so this review is a collection of
> nitpicks :-).
> 
>>
>> Signed-off-by: Richard Weinberger <richard@xxxxxx>
>> ---
>>  drivers/mtd/ubi/build.c |  39 +++++++++++++
>>  drivers/mtd/ubi/ubi.h   |   4 ++
>>  drivers/mtd/ubi/wl.c    | 146 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 189 insertions(+)
>>
>> diff --git a/drivers/mtd/ubi/build.c b/drivers/mtd/ubi/build.c
>> index 9690cf9..f58330b 100644
>> --- a/drivers/mtd/ubi/build.c
>> +++ b/drivers/mtd/ubi/build.c
>> @@ -118,6 +118,9 @@ static struct class_attribute ubi_version =
>>  
>>  static ssize_t dev_attribute_show(struct device *dev,
>>  				  struct device_attribute *attr, char *buf);
>> +static ssize_t trigger_bitrot_check(struct device *dev,
>> +				    struct device_attribute *mattr,
>> +				    const char *data, size_t count);
>>  
>>  /* UBI device attributes (correspond to files in '/<sysfs>/class/ubi/ubiX') */
>>  static struct device_attribute dev_eraseblock_size =
>> @@ -142,6 +145,8 @@ static struct device_attribute dev_bgt_enabled =
>>  	__ATTR(bgt_enabled, S_IRUGO, dev_attribute_show, NULL);
>>  static struct device_attribute dev_mtd_num =
>>  	__ATTR(mtd_num, S_IRUGO, dev_attribute_show, NULL);
>> +static struct device_attribute dev_trigger_bitrot_check =
>> +	__ATTR(trigger_bitrot_check, S_IWUSR, NULL, trigger_bitrot_check);
> 
> How about making this attribute a RW one, so that users could check
> if there's a bitrot check in progress.

As the check will be initiated only by userspace and writing to the trigger
while a check is running will return anyway a EBUSY I don't really see
a point why userspace would check for it.

>>  
>>  /**
>>   * ubi_volume_notify - send a volume change notification.
>> @@ -334,6 +339,36 @@ int ubi_major2num(int major)
>>  	return ubi_num;
>>  }
>>  
>> +/* "Store" method for file '/<sysfs>/class/ubi/ubiX/trigger_bitrot_check' */
>> +static ssize_t trigger_bitrot_check(struct device *dev,
>> +				    struct device_attribute *mattr,
>> +				    const char *data, size_t count)
>> +{
>> +	struct ubi_device *ubi;
>> +	int ret;
>> +
> 
> Maybe that's on purpose, but you do not check the value passed in data
> (in your documention you suggest to do an
> echo 1 > /sys/class/ubi/ubiX/trigger_bitrot_check).

Yeah, the example using "1", but why should I limit it to it?
The idea was that any write will trigger a check.

>> +	ubi = container_of(dev, struct ubi_device, dev);
>> +	ubi = ubi_get_device(ubi->ubi_num);
>> +	if (!ubi) {
>> +		ret = -ENODEV;
>> +		goto out;
>> +	}
>> +
>> +	if (atomic_inc_return(&ubi->bit_rot_work) != 1) {
>> +		ret = -EBUSY;
>> +		goto out_dec;
>> +	}
>> +
>> +	ubi_wl_trigger_bitrot_check(ubi);
>> +	ret = count;
>> +
>> +out_dec:
>> +	atomic_dec(&ubi->bit_rot_work);
>> +out:
>> +	ubi_put_device(ubi);
>> +	return ret;
>> +}
>> +
>>  /* "Show" method for files in '/<sysfs>/class/ubi/ubiX/' */
>>  static ssize_t dev_attribute_show(struct device *dev,
>>  				  struct device_attribute *attr, char *buf)
>> @@ -445,6 +480,9 @@ static int ubi_sysfs_init(struct ubi_device *ubi, int *ref)
>>  	if (err)
>>  		return err;
>>  	err = device_create_file(&ubi->dev, &dev_mtd_num);
>> +	if (err)
>> +		return err;
>> +	err = device_create_file(&ubi->dev, &dev_trigger_bitrot_check);
>>  	return err;
> 
> You don't seem to control the return code, so, how about replacing
> those 2 lines by:
> 
> 	return device_create_file(&ubi->dev, &dev_trigger_bitrot_check);

I did it exactly like the existing code does.
But I can replace the lines...

>>  }
> 
> [...]
> 
>> diff --git a/drivers/mtd/ubi/wl.c b/drivers/mtd/ubi/wl.c
>> index 9b11db9..784bb52 100644
>> --- a/drivers/mtd/ubi/wl.c
>> +++ b/drivers/mtd/ubi/wl.c
>> @@ -1447,6 +1447,150 @@ static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
>>  }
>>  
>>  /**
>> + * bitrot_check_worker - physical eraseblock bitrot check worker function.
>> + * @ubi: UBI device description object
>> + * @wl_wrk: the work object
>> + * @shutdown: non-zero if the worker has to free memory and exit
>> + *
>> + * This function reads a physical eraseblock and schedules scrubbing if
>> + * bit flips are detected.
>> + */
>> +static int bitrot_check_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
>> +			       int shutdown)
>> +{
>> +	struct ubi_wl_entry *e = wl_wrk->e;
>> +	int err;
>> +
>> +	kfree(wl_wrk);
>> +	if (shutdown) {
>> +		dbg_wl("cancel bitrot check of PEB %d", e->pnum);
>> +		wl_entry_destroy(ubi, e);
>> +		return 0;
>> +	}
>> +
>> +	mutex_lock(&ubi->buf_mutex);
>> +	err = ubi_io_read(ubi, ubi->peb_buf, e->pnum, 0, ubi->peb_size);
>> +	mutex_unlock(&ubi->buf_mutex);
>> +	if (err == UBI_IO_BITFLIPS) {
>> +		dbg_wl("found bitflips in PEB %d", e->pnum);
>> +		spin_lock(&ubi->wl_lock);
>> +
>> +		if (in_pq(ubi, e)) {
>> +			prot_queue_del(ubi, e->pnum);
>> +			wl_tree_add(e, &ubi->scrub);
>> +			spin_unlock(&ubi->wl_lock);
>> +
>> +			err = ensure_wear_leveling(ubi, 1);
>> +		}
>> +		else if (in_wl_tree(e, &ubi->used)) {
>> +			rb_erase(&e->u.rb, &ubi->used);
>> +			wl_tree_add(e, &ubi->scrub);
>> +			spin_unlock(&ubi->wl_lock);
>> +
>> +			err = ensure_wear_leveling(ubi, 1);
>> +		}
>> +		else if (in_wl_tree(e, &ubi->free)) {
>> +			rb_erase(&e->u.rb, &ubi->free);
>> +			spin_unlock(&ubi->wl_lock);
>> +
>> +			wl_wrk = prepare_erase_work(e, -1, -1, 1);
>> +			if (IS_ERR(wl_wrk)) {
>> +				err = PTR_ERR(wl_wrk);
>> +				goto out;
>> +			}
>> +
>> +			__schedule_ubi_work(ubi, wl_wrk);
>> +			err = 0;
>> +		}
>> +		/*
>> +		 * e is target of a move operation, all we can do is kicking
>> +		 * wear leveling such that we can catch it later or wear
>> +		 * leveling itself scrubbs the PEB.
>> +		 */
>> +		else if (ubi->move_to == e || ubi->move_from == e) {
>> +			spin_unlock(&ubi->wl_lock);
>> +
>> +			err = ensure_wear_leveling(ubi, 1);
>> +		}
>> +		/*
>> +		 * e is member of a fastmap pool. We are not allowed to
>> +		 * remove it from that pool as the on-flash fastmap data
>> +		 * structure refers to it. Let's schedule a new fastmap write
>> +		 * such that the said PEB can get released.
>> +		 */
>> +		else {
>> +			ubi_schedule_fm_work(ubi);
>> +			spin_unlock(&ubi->wl_lock);
>> +
>> +			err = 0;
>> +		}
> 
> Nitpick, but checkpatch complains about 'else' or 'else if' statements
> that are not on the '}' line.

I like it as is because I can nicely place the comment above the else {.
And checkpatch is not our lawmaker.

>> +	}
>> +	else {
>> +		/*
>> +		 * Ignore read errors as we return only work related errors.
>> +		 * Read errors will be logged by ubi_io_read().
>> +		 */
>> +		err = 0;
>> +	}
> 
> Nitpicking again, but you can avoid another level of indentation by
> doing the following:
> 
> 	if (err != UBI_IO_BITFLIPS) {
> 		err = 0;
> 		goto out;
> 	}
> 
> 	dbg_wl("found bitflips in PEB %d", e->pnum);
> 	spin_lock(&ubi->wl_lock);
> 	/* ... */
> 
>> +
>> +out:
>> +	atomic_dec(&ubi->bit_rot_work);
>> +	ubi_assert(atomic_read(&ubi->bit_rot_work) >= 0);
> 
> How about replacing those two lines by:
> 
> 	ubi_assert(atomic_dec_return(&ubi->bit_rot_work) >= 0);

True, I always forget how many nice atomic_* helper we have. :)

Thanks,
//richard
--
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/