2022-01-07 09:12:19

by Jiasheng Jiang

Subject: [PATCH v2] ide: Check for null pointer after calling devm_ioremap

In linux-stable-5.15.13, this file has been removed and combined
to `drivers/ata/pata_platform.c` without this bug.
But in the older LTS kernels, like 5.10.90, this bug still exists.
As the possible failure of the devres_alloc(), the devm_ioremap() and
devm_ioport_map() may return NULL pointer.
And then, the 'base' and 'alt_base' are used in plat_ide_setup_ports().
Therefore, it should be better to add the check in order to avoid the
dereference of the NULL pointer.
Actually, it introduced the bug from commit 8cb1f567f4c0
("ide: Platform IDE driver") and we can know from the commit message
that it tended to be similar to the `drivers/ata/pata_platform.c`.
But actually, even the first time pata_platform was built,
commit a20c9e820864 ("[PATCH] ata: Generic platform_device libata driver"),
there was no the bug, as there was a check after the ioremap().
So possibly the bug was caused by ide itself.

Fixes: 8cb1f567f4c0 ("ide: Platform IDE driver")
Cc: [email protected]#5.10.90
Signed-off-by: Jiasheng Jiang <[email protected]>
---
Changelog

v1 -> v2

* Change 1. Correct the fixes tag and commit message.
---
drivers/ide/ide_platform.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/drivers/ide/ide_platform.c b/drivers/ide/ide_platform.c
index 91639fd6c276..5500c5afb3ca 100644
--- a/drivers/ide/ide_platform.c
+++ b/drivers/ide/ide_platform.c
@@ -85,6 +85,10 @@ static int plat_ide_probe(struct platform_device *pdev)
alt_base = devm_ioport_map(&pdev->dev,
res_alt->start, resource_size(res_alt));
}
+ if (!base || !!alt_base) {
+ ret = -ENOMEM;
+ goto out;
+ }

memset(&hw, 0, sizeof(hw));
plat_ide_setup_ports(&hw, base, alt_base, pdata, res_irq->start);
--
2.25.1

2022-01-07 09:29:12

by David Laight

Subject: RE: [PATCH v2] ide: Check for null pointer after calling devm_ioremap

From: Jiasheng Jiang
> Sent: 07 January 2022 09:12
>
> In linux-stable-5.15.13, this file has been removed and combined
> to `drivers/ata/pata_platform.c` without this bug.
> But in the older LTS kernels, like 5.10.90, this bug still exists.
> As the possible failure of the devres_alloc(), the devm_ioremap() and
> devm_ioport_map() may return NULL pointer.
> And then, the 'base' and 'alt_base' are used in plat_ide_setup_ports().
> Therefore, it should be better to add the check in order to avoid the
> dereference of the NULL pointer.
> Actually, it introduced the bug from commit 8cb1f567f4c0
> ("ide: Platform IDE driver") and we can know from the commit message
> that it tended to be similar to the `drivers/ata/pata_platform.c`.
> But actually, even the first time pata_platform was built,
> commit a20c9e820864 ("[PATCH] ata: Generic platform_device libata driver"),
> there was no the bug, as there was a check after the ioremap().
> So possibly the bug was caused by ide itself.
>
> Fixes: 8cb1f567f4c0 ("ide: Platform IDE driver")
> Cc: [email protected]#5.10.90
> Signed-off-by: Jiasheng Jiang <[email protected]>
> ---
> Changelog
>
> v1 -> v2
>
> * Change 1. Correct the fixes tag and commit message.
> ---
> drivers/ide/ide_platform.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/drivers/ide/ide_platform.c b/drivers/ide/ide_platform.c
> index 91639fd6c276..5500c5afb3ca 100644
> --- a/drivers/ide/ide_platform.c
> +++ b/drivers/ide/ide_platform.c
> @@ -85,6 +85,10 @@ static int plat_ide_probe(struct platform_device *pdev)
> alt_base = devm_ioport_map(&pdev->dev,
> res_alt->start, resource_size(res_alt));
> }
> + if (!base || !!alt_base) {
> + ret = -ENOMEM;
> + goto out;
> + }

That !!alt_base doesn't look right.
Without looking at the rest of the code maybe:
if (!base && !alt_base)
may be correct.

It also rather makes me wonder about the actual failure return value.
If devm_ioport_map() returns a 'port number' for inb()/outb() then
zero is technically a valid value!

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)