[RFC PATCH] scsi: ses: Fix out-of-bound write at ses_enclosure_data_process()

[RFC PATCH] scsi: ses: Fix out-of-bound write at ses_enclosure_data_process()

[Tianchen Ding]

Our modified KFENCE reported a memory corruption:

[   52.584914] BUG: KFENCE: memory corruption in ses_enclosure_data_process+0x24b/0x310 [ses]

[   52.584917] Corrupted memory at 0xffff88982de06ff0 [ 0x00 . . . . . . . . . . . . . . . ] (in kfence-#1624698):
[   52.607212]  ses_enclosure_data_process+0x24b/0x310 [ses]
[   52.607215]  ses_intf_add+0x444/0x542 [ses]
[   52.621369]  class_interface_register+0x110/0x120
[   52.621373]  ses_init+0x13/0x1000 [ses]
[   52.621377]  do_one_initcall+0x41/0x1c0
[   52.621380]  do_init_module+0x5c/0x260
[   52.621382]  __do_sys_finit_module+0xb1/0x110
[   52.621386]  do_syscall_64+0x2d/0x40
[   52.621388]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

[   52.621393] kfence-#1624698 [0xffff88982de06fc0-0xffff88982de06fe0, size=33, cache=kmalloc-64] allocated by task 1033:
[   52.670344]  ses_enclosure_data_process+0x2ae/0x310 [ses]
[   52.670347]  ses_intf_add+0x444/0x542 [ses]
[   52.670353]  class_interface_register+0x110/0x120
[   52.688165]  ses_init+0x13/0x1000 [ses]
[   52.688169]  do_one_initcall+0x41/0x1c0
[   52.688172]  do_init_module+0x5c/0x260
[   52.688174]  __do_sys_finit_module+0xb1/0x110
[   52.688177]  do_syscall_64+0x2d/0x40
[   52.688179]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

This is because we check desc_ptr >= buf + page7_len first but then
write '\0' to desc_ptr[len+4], while this address may be out of bound.

Fixes: 21fab1d0595e ("[SCSI] ses: update enclosure data on hot add")
Signed-off-by: Tianchen Ding <dtcccc@linux.alibaba.com>
---
 drivers/scsi/ses.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/scsi/ses.c b/drivers/scsi/ses.c
index 0a1734f34587..06b991e27c84 100644
--- a/drivers/scsi/ses.c
+++ b/drivers/scsi/ses.c
@@ -559,11 +559,11 @@ static void ses_enclosure_data_process(struct enclosure_device *edev,
 			struct enclosure_component *ecomp;
 
 			if (desc_ptr) {
-				if (desc_ptr >= buf + page7_len) {
+				len = (desc_ptr[2] << 8) + desc_ptr[3];
+				desc_ptr += 4;
+				if (desc_ptr + len > buf + page7_len) {
 					desc_ptr = NULL;
 				} else {
-					len = (desc_ptr[2] << 8) + desc_ptr[3];
-					desc_ptr += 4;
 					/* Add trailing zero - pushes into
 					 * reserved space */
 					desc_ptr[len] = '\0';
-- 
2.33.0

Re: [RFC PATCH] scsi: ses: Fix out-of-bound write at ses_enclosure_data_process()

[Tianchen Ding]

Hi.

There seems to be some hardware problems with this machine, and it 
cannot bootup stably now... :-(

We guess the given desc_ptr[2] and [3] may be wrong on this machine, 
causing the calculated len becoming wrong too.

However, from the view of memory side, this is a potential oob write. 
Should we fix it?

On 2022/4/15 11:23, Tianchen Ding wrote:
> Our modified KFENCE reported a memory corruption:
> 
> [   52.584914] BUG: KFENCE: memory corruption in ses_enclosure_data_process+0x24b/0x310 [ses]
> 
> [   52.584917] Corrupted memory at 0xffff88982de06ff0 [ 0x00 . . . . . . . . . . . . . . . ] (in kfence-#1624698):
> [   52.607212]  ses_enclosure_data_process+0x24b/0x310 [ses]
> [   52.607215]  ses_intf_add+0x444/0x542 [ses]
> [   52.621369]  class_interface_register+0x110/0x120
> [   52.621373]  ses_init+0x13/0x1000 [ses]
> [   52.621377]  do_one_initcall+0x41/0x1c0
> [   52.621380]  do_init_module+0x5c/0x260
> [   52.621382]  __do_sys_finit_module+0xb1/0x110
> [   52.621386]  do_syscall_64+0x2d/0x40
> [   52.621388]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> 
> [   52.621393] kfence-#1624698 [0xffff88982de06fc0-0xffff88982de06fe0, size=33, cache=kmalloc-64] allocated by task 1033:
> [   52.670344]  ses_enclosure_data_process+0x2ae/0x310 [ses]
> [   52.670347]  ses_intf_add+0x444/0x542 [ses]
> [   52.670353]  class_interface_register+0x110/0x120
> [   52.688165]  ses_init+0x13/0x1000 [ses]
> [   52.688169]  do_one_initcall+0x41/0x1c0
> [   52.688172]  do_init_module+0x5c/0x260
> [   52.688174]  __do_sys_finit_module+0xb1/0x110
> [   52.688177]  do_syscall_64+0x2d/0x40
> [   52.688179]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> 
> This is because we check desc_ptr >= buf + page7_len first but then
> write '\0' to desc_ptr[len+4], while this address may be out of bound.
> 
> Fixes: 21fab1d0595e ("[SCSI] ses: update enclosure data on hot add")
> Signed-off-by: Tianchen Ding <dtcccc@linux.alibaba.com>
> ---
>   drivers/scsi/ses.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/scsi/ses.c b/drivers/scsi/ses.c
> index 0a1734f34587..06b991e27c84 100644
> --- a/drivers/scsi/ses.c
> +++ b/drivers/scsi/ses.c
> @@ -559,11 +559,11 @@ static void ses_enclosure_data_process(struct enclosure_device *edev,
>   			struct enclosure_component *ecomp;
>   
>   			if (desc_ptr) {
> -				if (desc_ptr >= buf + page7_len) {
> +				len = (desc_ptr[2] << 8) + desc_ptr[3];
> +				desc_ptr += 4;
> +				if (desc_ptr + len > buf + page7_len) {
>   					desc_ptr = NULL;
>   				} else {
> -					len = (desc_ptr[2] << 8) + desc_ptr[3];
> -					desc_ptr += 4;
>   					/* Add trailing zero - pushes into
>   					 * reserved space */
>   					desc_ptr[len] = '\0';