[PATCH 0/1] Fix OOB checking problem in nand_base.c

[PATCH 0/1] Fix OOB checking problem in nand_base.c

[Stanley.Miao]

on TI OMAPL138 EVM board, the NAND Flash have 22 bytes free in OOB area.
When yaffs2 writes the 28-byte tags into NAND OOB area, the NAND subsystem
doesn't give any WARNING or ERROR.

Move the checking of validity of oob data from nand_do_write_oob into
nand_write_oob to fix this problem.

[PATCH 1/1] mtd: nand: move the checking the validity of oob into nand_write_oob

[Stanley.Miao]

nand_write_oob will invoke nand_do_write_oob or nand_do_write_ops depending
on if ops->datbuf is NULL. nand_do_write_oob checked the validity of oob
but nand_do_write_ops didn't. Now move the check into nand_write_oob to
ensure the validity of oobbuf.

Signed-off-by: Stanley.Miao <stanley.miao@windriver.com>
---
 drivers/mtd/nand/nand_base.c |   69 ++++++++++++++++++++++-------------------
 1 files changed, 37 insertions(+), 32 deletions(-)

diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
index 8f2958f..29e2a06 100644
--- a/drivers/mtd/nand/nand_base.c
+++ b/drivers/mtd/nand/nand_base.c
@@ -2113,40 +2113,9 @@ static int nand_write(struct mtd_info *mtd, loff_t to, size_t len,
 static int nand_do_write_oob(struct mtd_info *mtd, loff_t to,
 			     struct mtd_oob_ops *ops)
 {
-	int chipnr, page, status, len;
+	int chipnr, page, status;
 	struct nand_chip *chip = mtd->priv;
 
-	DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, len = %i\n",
-			 __func__, (unsigned int)to, (int)ops->ooblen);
-
-	if (ops->mode == MTD_OOB_AUTO)
-		len = chip->ecc.layout->oobavail;
-	else
-		len = mtd->oobsize;
-
-	/* Do not allow write past end of page */
-	if ((ops->ooboffs + ops->ooblen) > len) {
-		DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt to write "
-				"past end of page\n", __func__);
-		return -EINVAL;
-	}
-
-	if (unlikely(ops->ooboffs >= len)) {
-		DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt to start "
-				"write outside oob\n", __func__);
-		return -EINVAL;
-	}
-
-	/* Do not allow reads past end of device */
-	if (unlikely(to >= mtd->size ||
-		     ops->ooboffs + ops->ooblen >
-			((mtd->size >> chip->page_shift) -
-			 (to >> chip->page_shift)) * len)) {
-		DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt write beyond "
-				"end of device\n", __func__);
-		return -EINVAL;
-	}
-
 	chipnr = (int)(to >> chip->chip_shift);
 	chip->select_chip(mtd, chipnr);
 
@@ -2203,6 +2172,42 @@ static int nand_write_oob(struct mtd_info *mtd, loff_t to,
 		return -EINVAL;
 	}
 
+	if (ops->oobbuf) {
+		int len;
+		DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, ooblen = %i\n",
+				__func__, (unsigned int)to, (int)ops->ooblen);
+
+		if (ops->mode == MTD_OOB_AUTO)
+			len = chip->ecc.layout->oobavail;
+		else
+			len = mtd->oobsize;
+
+		/* Do not allow write past end of page */
+		if ((ops->ooboffs + ops->ooblen) > len) {
+			DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt to write "
+					"past end of page\n", __func__);
+			return -EINVAL;
+		}
+
+		if (unlikely(ops->ooboffs >= len)) {
+			DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt to start "
+					"write outside oob\n", __func__);
+			return -EINVAL;
+		}
+
+		/* Do not allow reads past end of device */
+		if (unlikely(to >= mtd->size ||
+					ops->ooboffs + ops->ooblen >
+					((mtd->size >> chip->page_shift) -
+					 (to >> chip->page_shift)) * len)) {
+			DEBUG(MTD_DEBUG_LEVEL0, "%s: Attempt write beyond "
+					"end of device\n", __func__);
+			return -EINVAL;
+		}
+
+
+	}
+
 	nand_get_device(chip, mtd, FL_WRITING);
 
 	switch(ops->mode) {
-- 
1.5.4.3

Re: [PATCH 1/1] mtd: nand: move the checking the validity of oob into nand_write_oob

[Artem Bityutskiy]

On Thu, 2010-02-04 at 15:18 +0800, Stanley.Miao wrote:
> nand_write_oob will invoke nand_do_write_oob or nand_do_write_ops depending
> on if ops->datbuf is NULL. nand_do_write_oob checked the validity of oob
> but nand_do_write_ops didn't. Now move the check into nand_write_oob to
> ensure the validity of oobbuf.
> 
> Signed-off-by: Stanley.Miao <stanley.miao@windriver.com>
> ---
>  drivers/mtd/nand/nand_base.c |   69 ++++++++++++++++++++++-------------------
>  1 files changed, 37 insertions(+), 32 deletions(-)
> 
> diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
> index 8f2958f..29e2a06 100644
> --- a/drivers/mtd/nand/nand_base.c
> +++ b/drivers/mtd/nand/nand_base.c
> @@ -2113,40 +2113,9 @@ static int nand_write(struct mtd_info *mtd, loff_t to, size_t len,
>  static int nand_do_write_oob(struct mtd_info *mtd, loff_t to,
>  			     struct mtd_oob_ops *ops)
>  {

...

> @@ -2203,6 +2172,42 @@ static int nand_write_oob(struct mtd_info *mtd, loff_t to,
>  		return -EINVAL;
>  	}
>  
> +	if (ops->oobbuf) {
> +		int len;
> +		DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, ooblen = %i\n",
> +				__func__, (unsigned int)to, (int)ops->ooblen);

Why this check is not done for !obs->oobbuf cas as well?


-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

Re: [PATCH 1/1] mtd: nand: move the checking the validity of oob into nand_write_oob

[Maxim Levitsky]

On Mon, 2010-02-15 at 16:07 +0200, Artem Bityutskiy wrote: 
> On Thu, 2010-02-04 at 15:18 +0800, Stanley.Miao wrote:
> > nand_write_oob will invoke nand_do_write_oob or nand_do_write_ops depending
> > on if ops->datbuf is NULL. nand_do_write_oob checked the validity of oob
> > but nand_do_write_ops didn't. Now move the check into nand_write_oob to
> > ensure the validity of oobbuf.
> > 
> > Signed-off-by: Stanley.Miao <stanley.miao@windriver.com>
> > ---
> >  drivers/mtd/nand/nand_base.c |   69 ++++++++++++++++++++++-------------------
> >  1 files changed, 37 insertions(+), 32 deletions(-)
> > 
> > diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
> > index 8f2958f..29e2a06 100644
> > --- a/drivers/mtd/nand/nand_base.c
> > +++ b/drivers/mtd/nand/nand_base.c
> > @@ -2113,40 +2113,9 @@ static int nand_write(struct mtd_info *mtd, loff_t to, size_t len,
> >  static int nand_do_write_oob(struct mtd_info *mtd, loff_t to,
> >  			     struct mtd_oob_ops *ops)
> >  {
> 
> ...
> 
> > @@ -2203,6 +2172,42 @@ static int nand_write_oob(struct mtd_info *mtd, loff_t to,
> >  		return -EINVAL;
> >  	}
> >  
> > +	if (ops->oobbuf) {
> > +		int len;
> > +		DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, ooblen = %i\n",
> > +				__func__, (unsigned int)to, (int)ops->ooblen);
> 
> Why this check is not done for !obs->oobbuf cas as well?
> 
> 

I address this and other problems in my patchset, please test if it
works.

Best regards,
Maxim Levitsky