2023-10-01 21:09:13

by Gustavo A. R. Silva

[permalink] [raw]
Subject: [PATCH][next] PCI/P2PDMA: Fix undefined behavior bug in struct pci_p2pdma_pagemap

`struct dev_pagemap` is a flexible structure, which means that it
contains a flexible-array member at the bottom. This could potentially
lead to an overwrite of the objects following `pgmap` in `struct
pci_p2pdma_pagemap`, when `nr_range > 1`.

Fix this by placing the declaration of object `pgmap` at the end of
`struct pci_p2pdma_pagemap`.

-Wflex-array-member-not-at-end is coming in GCC-14, and we are getting
ready to enable it globally.

Fixes: 0afea3814358 ("PCI/P2PDMA: Add provider's pci_dev to pci_p2pdma_pagemap struct")
Fixes: a6e6fe6549f6 ("PCI/P2PDMA: Introduce private pagemap structure")
Cc: [email protected]
Signed-off-by: Gustavo A. R. Silva <[email protected]>
---
drivers/pci/p2pdma.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c
index fa7370f9561a..ab34d3d36a64 100644
--- a/drivers/pci/p2pdma.c
+++ b/drivers/pci/p2pdma.c
@@ -28,9 +28,9 @@ struct pci_p2pdma {
};

struct pci_p2pdma_pagemap {
- struct dev_pagemap pgmap;
struct pci_dev *provider;
u64 bus_offset;
+ struct dev_pagemap pgmap;
};

static struct pci_p2pdma_pagemap *to_p2p_pgmap(struct dev_pagemap *pgmap)
--
2.34.1


2023-10-02 18:58:13

by Gustavo A. R. Silva

[permalink] [raw]
Subject: Re: [PATCH][next] PCI/P2PDMA: Fix undefined behavior bug in struct pci_p2pdma_pagemap



On 10/2/23 18:07, Logan Gunthorpe wrote:
>
>
> On 2023-10-01 15:08, Gustavo A. R. Silva wrote:
>> `struct dev_pagemap` is a flexible structure, which means that it
>> contains a flexible-array member at the bottom. This could potentially
>> lead to an overwrite of the objects following `pgmap` in `struct
>> pci_p2pdma_pagemap`, when `nr_range > 1`.
>>
>> Fix this by placing the declaration of object `pgmap` at the end of
>> `struct pci_p2pdma_pagemap`.
>>
>> -Wflex-array-member-not-at-end is coming in GCC-14, and we are getting
>> ready to enable it globally.
>>
>> Fixes: 0afea3814358 ("PCI/P2PDMA: Add provider's pci_dev to pci_p2pdma_pagemap struct")
>> Fixes: a6e6fe6549f6 ("PCI/P2PDMA: Introduce private pagemap structure")
>> Cc: [email protected]
>> Signed-off-by: Gustavo A. R. Silva <[email protected]>
>
>
> Makes sense to me, thanks.
>
> Although, I'm not sure the fixes tags are appropriate. The
> flexible-array member was introduced in 5.10 (b7b3c01b19) and both the
> "fixed" commits predate that change by a number of releases.

You're right. I'll remove those tags.

>
> Also, it's probably worth noting in the commit message that the p2pdma
> code hardcodes nr_ranges to 1 (in pci_p2pdma_add_resource); so there is
> no way to actually hit any bug with the current code.

Yep. I mention that in this part "This could potentially lead to an
overwrite of the objects following `pgmap` in `struct pci_p2pdma_pagemap`,
when `nr_range > 1`."

>
> I totally agree that the patch should be applied to prevent possible
> bugs being introduced in the future:
>
> Reviewed-by: Logan Gunthorpe <[email protected]>

Thanks for the RB and the feedback. :)
--
Gustavo

2023-10-02 19:20:22

by Gustavo A. R. Silva

[permalink] [raw]
Subject: Re: [PATCH][next] PCI/P2PDMA: Fix undefined behavior bug in struct pci_p2pdma_pagemap


>
> Yes, but the commit message is not clear that nr_range can never be >1
> in the code as it currently is.

Done:
https://lore.kernel.org/linux-hardening/ZRsUL%2FhATNruwtla@work/

Thanks!
--
Gustavo

2023-10-02 22:10:58

by Logan Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH][next] PCI/P2PDMA: Fix undefined behavior bug in struct pci_p2pdma_pagemap



On 2023-10-01 15:08, Gustavo A. R. Silva wrote:
> `struct dev_pagemap` is a flexible structure, which means that it
> contains a flexible-array member at the bottom. This could potentially
> lead to an overwrite of the objects following `pgmap` in `struct
> pci_p2pdma_pagemap`, when `nr_range > 1`.
>
> Fix this by placing the declaration of object `pgmap` at the end of
> `struct pci_p2pdma_pagemap`.
>
> -Wflex-array-member-not-at-end is coming in GCC-14, and we are getting
> ready to enable it globally.
>
> Fixes: 0afea3814358 ("PCI/P2PDMA: Add provider's pci_dev to pci_p2pdma_pagemap struct")
> Fixes: a6e6fe6549f6 ("PCI/P2PDMA: Introduce private pagemap structure")
> Cc: [email protected]
> Signed-off-by: Gustavo A. R. Silva <[email protected]>


Makes sense to me, thanks.

Although, I'm not sure the fixes tags are appropriate. The
flexible-array member was introduced in 5.10 (b7b3c01b19) and both the
"fixed" commits predate that change by a number of releases.

Also, it's probably worth noting in the commit message that the p2pdma
code hardcodes nr_ranges to 1 (in pci_p2pdma_add_resource); so there is
no way to actually hit any bug with the current code.

I totally agree that the patch should be applied to prevent possible
bugs being introduced in the future:

Reviewed-by: Logan Gunthorpe <[email protected]>

Logan

2023-10-02 22:15:06

by Logan Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH][next] PCI/P2PDMA: Fix undefined behavior bug in struct pci_p2pdma_pagemap



On 2023-10-02 12:40, Gustavo A. R. Silva wrote:
>
>
> On 10/2/23 18:07, Logan Gunthorpe wrote:
>>
>>
>> On 2023-10-01 15:08, Gustavo A. R. Silva wrote:
>>> `struct dev_pagemap` is a flexible structure, which means that it
>>> contains a flexible-array member at the bottom. This could potentially
>>> lead to an overwrite of the objects following `pgmap` in `struct
>>> pci_p2pdma_pagemap`, when `nr_range > 1`.
>>>
>>> Fix this by placing the declaration of object `pgmap` at the end of
>>> `struct pci_p2pdma_pagemap`.
>>>
>>> -Wflex-array-member-not-at-end is coming in GCC-14, and we are getting
>>> ready to enable it globally.
>>>
>>> Fixes: 0afea3814358 ("PCI/P2PDMA: Add provider's pci_dev to pci_p2pdma_pagemap struct")
>>> Fixes: a6e6fe6549f6 ("PCI/P2PDMA: Introduce private pagemap structure")
>>> Cc: [email protected]
>>> Signed-off-by: Gustavo A. R. Silva <[email protected]>
>>
>>
>> Makes sense to me, thanks.
>>
>> Although, I'm not sure the fixes tags are appropriate. The
>> flexible-array member was introduced in 5.10 (b7b3c01b19) and both the
>> "fixed" commits predate that change by a number of releases.
>
> You're right. I'll remove those tags.
>
>>
>> Also, it's probably worth noting in the commit message that the p2pdma
>> code hardcodes nr_ranges to 1 (in pci_p2pdma_add_resource); so there is
>> no way to actually hit any bug with the current code.
>
> Yep. I mention that in this part "This could potentially lead to an
> overwrite of the objects following `pgmap` in `struct pci_p2pdma_pagemap`,
> when `nr_range > 1`."

Yes, but the commit message is not clear that nr_range can never be >1
in the code as it currently is.

Logan