Skip to contentSkip to conversation

USB static analysis patches

PATCH Superseded Liam Merwick 2 participants 7 messages Jan 30, 2019 – Feb 5, 2019

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover USB static analysis patches

These two patches are a result of running the Parfait static analysis tool on QEMU. The first patch is an optimisation to move an assignment after a test and the second patch handles a number of potential NULL pointer dereferences if usb_ep_get() returns NULL.

Liam Merwick (2):
usb: rearrange usb_ep_get()
usb: deal with potential Null pointer returned by usb_ep_get()

 hw/usb/core.c | 29 ++++++++++++++++++++++++-----
 1 file changed, 24 insertions(+), 5 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/2 usb: rearrange usb_ep_get()

There is no need to calculate the 'eps' variable in usb_ep_get() if 'ep' is the control endpoint. Instead the calculation should be done after validating the input and the resulting pointer also validated before returning an entry indexed on the endpoint 'ep'.

---
 hw/usb/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
hw/usb/core.c +2 -1
index 241ae66b1505..1aa0051b2b2d 100644
@@ -720,12 +720,13 @@ struct USBEndpoint *usb_ep_get(USBDevice *dev, int pid, int ep)
     if (dev == NULL) {
         return NULL;
     }
-    eps = (pid == USB_TOKEN_IN) ? dev->ep_in : dev->ep_out;
     if (ep == 0) {
         return &dev->ep_ctl;
     }
     assert(pid == USB_TOKEN_IN || pid == USB_TOKEN_OUT);
     assert(ep > 0 && ep <= USB_MAX_ENDPOINTS);
+    eps = (pid == USB_TOKEN_IN) ? dev->ep_in : dev->ep_out;
+    assert(eps != NULL);
     return eps + ep - 1;
 }
 
Signature · 1 line
1.8.3.1
#2 lore raw

Re: [Qemu-devel] [PATCH 1/2] usb: rearrange usb_ep_get()

Quoted 19 lines · Liam Merwick

There is no need to calculate the 'eps' variable in usb_ep_get() if 'ep' is the control endpoint. Instead the calculation should be done after validating the input and the resulting pointer also validated before returning an entry indexed on the endpoint 'ep'.

---
 hw/usb/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
hw/usb/core.c +0 -0
index 241ae66b1505..1aa0051b2b2d 100644
@@ -720,12 +720,13 @@ struct USBEndpoint *usb_ep_get(USBDevice *dev, int pid, int ep)
     if (dev == NULL) {

hw/usb/core.c struct USBEndpoint *usb_ep_get(USBDevice *dev, int pid, int ep) · line 729

Quoted 10 lines
hw/usb/core.c +2 -1
         return NULL;
     }
-    eps = (pid == USB_TOKEN_IN) ? dev->ep_in : dev->ep_out;
     if (ep == 0) {
         return &dev->ep_ctl;
     }
     assert(pid == USB_TOKEN_IN || pid == USB_TOKEN_OUT);
     assert(ep > 0 && ep <= USB_MAX_ENDPOINTS);
+    eps = (pid == USB_TOKEN_IN) ? dev->ep_in : dev->ep_out;
+    assert(eps != NULL);

That assert is rather pointless. It's impossible for eps to be NULL at this point.

cheers,
Gerd

#3 lore raw

PATCH 2/2 usb: deal with potential Null pointer returned by usb_ep_get()

In reply to Liam Merwick

From: Liam Merwick <[email protected]>

usb_ep_get() can return a Null pointer in the (albeit unlikely) case that a NULL USBDevice is passed in via the 'dev' parameter. Before dereferencing the return value from usb_ep_get(), check its validity and use a default (invalid) value where needed.

Reported by the Parfait static code analysis tool

---
 hw/usb/core.c | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
hw/usb/core.c +22 -4
index 1aa0051b2b2d..9d46a4b41e99 100644
@@ -733,19 +733,23 @@ struct USBEndpoint *usb_ep_get(USBDevice *dev, int pid, int ep)
 uint8_t usb_ep_get_type(USBDevice *dev, int pid, int ep)
 {
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
-    return uep->type;
+    return uep ? uep->type : USB_ENDPOINT_XFER_INVALID;
 }
 
 void usb_ep_set_type(USBDevice *dev, int pid, int ep, uint8_t type)
 {
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
-    uep->type = type;
+    if (uep) {
+        uep->type = type;
+    }
 }
 
 void usb_ep_set_ifnum(USBDevice *dev, int pid, int ep, uint8_t ifnum)
 {
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
-    uep->ifnum = ifnum;
+    if (uep) {
+        uep->ifnum = ifnum;
+    }
 }
 
 void usb_ep_set_max_packet_size(USBDevice *dev, int pid, int ep,
@@ -754,6 +758,10 @@ void usb_ep_set_max_packet_size(USBDevice *dev, int pid, int ep,
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
     int size, microframes;
 
+    if (!uep) {
+        return;
+    }
+
     size = raw & 0x7ff;
     switch ((raw >> 11) & 3) {
     case 1:
@@ -774,6 +782,10 @@ void usb_ep_set_max_streams(USBDevice *dev, int pid, int ep, uint8_t raw)
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
     int MaxStreams;
 
+    if (!uep) {
+        return;
+    }
+
     MaxStreams = raw & 0x1f;
     if (MaxStreams) {
         uep->max_streams = 1 << MaxStreams;
@@ -785,7 +797,9 @@ void usb_ep_set_max_streams(USBDevice *dev, int pid, int ep, uint8_t raw)
 void usb_ep_set_halted(USBDevice *dev, int pid, int ep, bool halted)
 {
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
-    uep->halted = halted;
+    if (uep) {
+        uep->halted = halted;
+    }
 }
 
 USBPacket *usb_ep_find_packet_by_id(USBDevice *dev, int pid, int ep,
@@ -794,6 +808,10 @@ USBPacket *usb_ep_find_packet_by_id(USBDevice *dev, int pid, int ep,
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep);
     USBPacket *p;
 
+    if (!uep) {
+        return NULL;
+    }
+
     QTAILQ_FOREACH(p, &uep->queue, queue) {
         if (p->id == id) {
             return p;
Signature · 1 line
1.8.3.1
#4 lore raw

Re: [Qemu-devel] [PATCH 2/2] usb: deal with potential Null pointer returned by usb_ep_get()

Quoted 4 lines · Liam Merwick

From: Liam Merwick <[email protected]>

usb_ep_get() can return a Null pointer in the (albeit unlikely) case that a NULL USBDevice is passed in via the 'dev' parameter.

That should never ever happen.

Quoted 1 line

Reported by the Parfait static code analysis tool

Try add "assert(dev != NULL)" to usb_ep_get() instead of sprinkling pointless checks all over the place.

thanks,
Gerd

#5 lore raw

Re: [Qemu-devel] [PATCH 2/2] usb: deal with potential Null pointer returned by usb_ep_get()

Quoted 11 lines · Gerd Hoffmann
Quoted 4 lines · Liam Merwick

From: Liam Merwick <[email protected]>

usb_ep_get() can return a Null pointer in the (albeit unlikely) case that a NULL USBDevice is passed in via the 'dev' parameter.

That should never ever happen.

Quoted 1 line

Reported by the Parfait static code analysis tool

Try add "assert(dev != NULL)" to usb_ep_get() instead of sprinkling pointless checks all over the place.

Adding "assert(dev != NULL)" to usb_ep_get() isn't sufficient for that tool unless the 'if (dev== NULL)' check is removed which seems a backwards step even if that NULL USBDevice case is impossible.

Adding an assert like below in 7 places in hw/usb/core.c that call usb_ep_get() would resolve it but would that  pass coding conventions (checkpatch.pl seems OK with it)?

 uint8_t usb_ep_get_type(USBDevice *dev, int pid, int ep) {
+    assert(dev);
     struct USBEndpoint *uep = usb_ep_get(dev, pid, ep); return uep->type;
 }

(the other option below it seems like too much code churn).

 uint8_t usb_ep_get_type(USBDevice *dev, int pid, int ep) {
-    struct USBEndpoint *uep = usb_ep_get(dev, pid, ep); +    struct USBEndpoint *uep;
+    assert(dev);
+    uep = usb_ep_get(dev, pid, ep);
     return uep->type;
 }

So that's kinda a long way of saying I'll drop this patch unless someone thinks it still adds a benefit and will send a v2 with a modified Patch1 and a patch that adds two asserts to hw/usb/hcd-xhci.c

Regards,

Liam

#6 lore raw

Re: [Qemu-devel] [PATCH 2/2] usb: deal with potential Null pointer returned by usb_ep_get()

Quoted 8 lines · Liam Merwick
Quoted 7 lines · Gerd Hoffmann
Quoted 4 lines · Liam Merwick

From: Liam Merwick <[email protected]>

usb_ep_get() can return a Null pointer in the (albeit unlikely) case that a NULL USBDevice is passed in via the 'dev' parameter.

That should never ever happen.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Reported by the Parfait static code analysis tool

Try add "assert(dev != NULL)" to usb_ep_get() instead of sprinkling pointless checks all over the place.

Adding "assert(dev != NULL)" to usb_ep_get() isn't sufficient for that tool unless the 'if (dev== NULL)' check is removed which seems a backwards step even if that NULL USBDevice case is impossible.

Looked at the code again.

The usb device emulation (hw/usb/dev-*.c) never ever calls usb_ep_get() with dev == NULL. There are some places in usb host adapter emulation (hw/usb/hcd-*) which might do this. uhci for example has this ...

	[ ... ]
        USBDevice *dev = uhci_find_device(s, (td->token >> 8) & 0x7f);
        USBEndpoint *ep = usb_ep_get(dev, pid, (td->token >> 15) & 0xf);

        if (ep == NULL) {
	[ ... ]

... and uhci_find_device can return NULL.

So, I'd suggest to check all usb_ep_get() callers, fix them if needed, then remove the 'if (dev== NULL)' check in usb_ep_get() and add the assert() instead.

cheers,
Gerd