drivers/usb/storage: NULL pointer dereference [null-pointer-deref] (CWE 476) problem
Greg Kroah-Hartman
maintainerPATCH 1/1 drivers/usb/storage: NULL pointer dereference [null-pointer-deref] (CWE 476) problem
Quoted 128 lines · Joe Moriarty
Quoted 118 lines · Greg KH
Quoted 117 lines · Joe Moriarty
Quoted 102 lines · Greg KH
Quoted 5 lines · Joe Moriarty
Quoted 4 lines · Greg KH
Quoted 2 lines · Joe Moriarty
The Parfait (version 2.1.0) static code analysis tool found the following NULL pointer dereference problem.
Quoted 7 lines
Quoted 2 lines
What is the "CWE 476" thing in the subject line for?
[JDM]
(CWE 476) stands for Common Weakness Enumeration.
https://secur1ty.com/cwe/CWE-476/It is the type of security flaw related to a NULL pointer dereference
Ok, why put that in the subject line and not just the body of the changelog if you really want to call something out like this?
Quoted 26 lines
Quoted 26 lines
Quoted 19 lines
dev_to_shost()ininclude/scsi/scsi_host.hhas the ability to return NULL if the scsi host device does not have theScsi_host->parentfield set. With the possibilty of a NULL pointer being set for theScsi_Host->parentfield, calls tohost_to_us()have to make sure the return pointer is not null. Changes were made to check for a return value of NULL on calls tohost_to_us().Signed-off-by Joe Moriarty <[email protected]>Reviewed-by Steven Sistare <[email protected]>Acked-by Hakon Bugge <[email protected]>--- drivers/usb/storage/scsiglue.c | 53 ++++++++++++++++++++++++++++++++++++------ 1 file changed, 46 insertions(+), 7 deletions(-)drivers/usb/storage/scsiglue.c +0
-0drivers/usb/storage/scsiglue.c static int slave_alloc (struct scsi_device *sdev) · line 69
Quoted 4 lines
drivers/usb/storage/scsiglue.c +2
-0@@ -66,6 +66,9 @@ static int slave_alloc (struct scsi_device *sdev) {struct us_data *us =host_to_us (sdev->host); +if (!us) +pr_warn ("Error in %s: us = NULL\n" , __func__);It is a driver, you should never use pr_* calls, but rather dev_* calls.
Quoted 6 lines
Quoted 4 lines
Also, you don't exit, are you sure the code keeps working after this happens?
And what is a user supposed to do with this warning message?
[JDM]
??? You need a better email client, inline comments are the norm.
Quoted 2 lines
- The messages are targeted for a developer and not the end user. I can change it to
dev_calls.But an end user sees "
KERN_WARNING" messages, right? If this is a debugging-only thing, then make it as such, and properly recover from it as it could be hit during normal operation.Quoted 43 lines
Quoted 4 lines
Quoted 4 lines
drivers/usb/storage/scsiglue.c +1
-0+/* * Set the INQUIRY transfer length to 36. We don't use any of * the extra data and many devices choke if asked for more or Quoted 8 lines
drivers/usb/storage/scsiglue.c static int slave_configure(struct scsi_device *sdev) · line 109
Quoted 6 lines
drivers/usb/storage/scsiglue.c +3
-0@@ -102,6 +105,11 @@ static int slave_configure(struct scsi_device *sdev) {struct us_data *us =host_to_us (sdev->host); +if (!us) { +pr_warn ("Error in %s: us = NULL\n" , __func__); +return 0 ;Why are you returning a success?
[JDM]
- Yes, I need to return-ENXIOfor theslave_allocroutine.Quoted 12 lines
drivers/usb/storage/scsiglue.c static int target_alloc(struct scsi_target *starget) · line 340
Quoted 9 lines
drivers/usb/storage/scsiglue.c +2
-0+ } +/* * Many devices have trouble transferring more than 32KB at a time, * while others have trouble with more than 64K. At this time we @@ -331,6 +339,11 @@ static int target_alloc(struct scsi_target *starget) {struct us_data *us =host_to_us (dev_to_shost (starget->dev.parent));Can
host_to_us()handle NULL?Quoted 2 lines
Nope, just looked at it, this will never cause the return value to be NULL, your checker needs some more work :(
[JDM]
This what the static code checker is catching.host_to_uswill return NULL if the following occurs.'
dev_to_shost()' ininclude/scsi/scsi_host.hline 757.This is done everytime a slave device is created at the following code '
target_alloc()' indrivers/usb/storage/scsiglue.clinue 340.This means that any call to
host_to_usin the scsiglue module will have the possibility of setting the return value of NULL. In fact, I would need to split out the following embedded call in 'target_alloc()' to avoid a possible NULL pointer dereference.Quoted 9 lines
Quoted 1 line
struct
us_data*us = host_to_us(dev_to_shost(starget->dev.parent));Exactly, your change did nothing, and us is probably not NULL here.
Quoted 2 lines
The question becomes, Can a slave device ever have it's parent field set to NULL (ie:
dev->parent).I don't think it can, do you see how?
Ok, I believe we went off the original problem this patch solves. Since you and I both agree that a slave device can never have it's parent field set to NULL (ie:
dev->parent) then this patch boils down to the following one change.Quoted 9 lines
Review
Quoted 7 lines
include/scsi/scsi_host.h
static inline structScsi_Host*dev_to_shost(struct device *dev){ while (!scsi_is_host_device(dev)) { if (!dev->parent)(quoted diff) +1
-1- return NULL; + BUG();No, never crash the kernel.
Quoted 1 line
There is
Bug()throughout the kernel. So that is invalid.
Nope, we are trying to get rid of them. If you see new ones being added, please point it out.
Quoted 3 lines
Kernel Developers will put in
Bug()statements anywhere the code is not suppose to go. It is easier to debug and then trying to backtrace from a side effect issue.
Not at all, it causes the machine to crash and make it harder to debug. Or it causes a reboot. Even WARN_ON() should not be used for debugging as we are seeing tools hit those and reboot and complain (see the syzkaller threads about this.)
If you want a debugging message, make it a debugging message using dev_dbg(). That's all that is needed.
Quoted 4 lines
Quoted 2 lines
Why not ask the scsi developers about this? They put that line there for a good reason, right?
I submitted the original patch here because the original patch was in the
drivers/usb/storagedirectory of which you are the maintainer.
But this is a scsi core api call, I know nothing about that :)
Quoted 5 lines
Quoted 2 lines
And if NULL is valid to return, then your patch still does not fix the issue at all, which was my main point.
Yes it does. A NULL return from
dev_to_shostwill kernel panic the machine in multiple places in the SCSI kernel modules. Swapping out the NULL with aBUG()will do that.
Neither is good to have, they should all be fixed.
thanks,
greg k-h
---
To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html