libertas if_usb: Fix crash on 64-bit machines
PATCH 1/1 libertas if_usb: Fix crash on 64-bit machines
On a 64-bit kernel, skb->tail is an offset, not a pointer. The libertas usb driver passes it to usb_fill_bulk_urb() anyway, causing interesting crashes. Fix that by using skb->data instead.
This highlights a problem with usb_fill_bulk_urb(). It doesn't notice when dma_map_single() fails and return the error to its caller as it should. In fact it _can't_ currently return the error, since it returns void.
So this problem was showing up only at unmap time, after we'd already suffered memory corruption by doing DMA to a bogus address.
---
drivers/net/wireless/libertas/if_usb.c +1 -1
@@ -508,7 +508,7 @@ static int __if_usb_submit_rx_urb(struct if_usb_card *cardp,/* Fill the receive configuration URB and initialise the Rx call back */ usb_fill_bulk_urb (cardp->rx_urb, cardp->udev,usb_rcvbulkpipe (cardp->udev, cardp->ep_in), - (void *) (skb->tail + (size_t )IPFIELD_ALIGN_OFFSET ), + skb->data +IPFIELD_ALIGN_OFFSET ,MRVDRV_ETH_RX_PACKET_BUFFER_SIZE , callbackfn, cardp);
Signature · 2 lines
David Woodhouse Open Source Technology Centre [email protected] Intel Corporation
From: David Woodhouse <[email protected]>
Date: Fri, 30 Oct 2009 17:45:14 +0000
Quoted 9 lines
On a 64-bit kernel,
skb->tailis an offset, not a pointer. The libertas usb driver passes it tousb_fill_bulk_urb()anyway, causing interesting crashes. Fix that by usingskb->datainstead.This highlights a problem with
usb_fill_bulk_urb(). It doesn't notice whendma_map_single()fails and return the error to its caller as it should. In fact it _can't_ currently return the error, since it returns void.
Quoted 5 lines
So this problem was showing up only at unmap time, after we'd already suffered memory corruption by doing DMA to a bogus address.
Signed-off-by David Woodhouse <[email protected]>
Quoted 8 lines · David Woodhouse
On a 64-bit kernel,
skb->tailis an offset, not a pointer. The libertas usb driver passes it tousb_fill_bulk_urb()anyway, causing interesting crashes. Fix that by usingskb->datainstead.This highlights a problem with
usb_fill_bulk_urb(). It doesn't notice whendma_map_single()fails and return the error to its caller as it should. In fact it _can't_ currently return the error, since it returns void.
This should be fixed. If changing the code to return the error would be too invasive (It is used in 30+ drivers), perhaps the routine should be modified to log a warning when dma mapping fails. I will submit an RFC to do that.
Larry
Quoted 10 lines · On Friday 30 October 2009 19:23:15 Larry Finger wrote:
Quoted 8 lines · David Woodhouse
On a 64-bit kernel,
skb->tailis an offset, not a pointer. The libertas usb driver passes it tousb_fill_bulk_urb()anyway, causing interesting crashes. Fix that by usingskb->datainstead.This highlights a problem with
usb_fill_bulk_urb(). It doesn't notice whendma_map_single()fails and return the error to its caller as it should. In fact it _can't_ currently return the error, since it returns void.
Quoted 4 lines
This should be fixed. If changing the code to return the error would be too invasive (It is used in 30+ drivers), perhaps the routine should be modified to log a warning when dma mapping fails. I will submit an RFC to do that.
err, hold on a sec:
- include/linux/usb.h -
static inline void usb_fill_bulk_urb(struct urb *urb,
struct usb_device *dev, unsigned int pipe, void *transfer_buffer, int buffer_length, usb_complete_t complete_fn, void *context)
{
urb->dev = dev;
urb->pipe = pipe;
urb->transfer_buffer = transfer_buffer;
urb->transfer_buffer_length = buffer_length;
urb->complete = complete_fn;
urb->context = context;
}that's just a fill-in macro.
AFAICT usb_submit_urb does the tricky dma mapping.
Regards,
Chr
Quoted 3 lines · Christian Lamparter
that's just a fill-in macro.
AFAICTusb_submit_urbdoes the tricky dma mapping.
Ah, that makes sense. In that case, all we need to do is make map_urb_for_dma() do the right thing.
Signature · 2 lines
David Woodhouse Open Source Technology Centre [email protected] Intel Corporation
Quoted 6 lines · David Woodhouse
Quoted 2 lines · Christian Lamparter
that's just a fill-in macro.
AFAICTusb_submit_urbdoes the tricky dma mapping.Ah, that makes sense. In that case, all we need to do is make
map_urb_for_dma()do the right thing.
After I sent my message, I realized that usb_submit_urb() and friends do no checking and that the test had to be further down the line.
Larry
Quoted 7 lines · On Friday 30 October 2009 19:51:21 David Woodhouse wrote:
Quoted 3 lines · Christian Lamparter
that's just a fill-in macro.
AFAICTusb_submit_urbdoes the tricky dma mapping.Ah, that makes sense. In that case, all we need to do is make
map_urb_for_dma()do the right thing.
well, but while we're on the subject of libertas.
I took a quick look around and wrote down some hopefully _helpful_ comments. That said, I don't have any libertas hw, so I have no idea if the attached code will actually do what its supposed to do... I'll leave it up to the professionals to test & write a real fix(es) with a proper commit message.
Notes:
- most + IPFIELD_ALIGN_OFFSET can be replaced by a single skb_reserve, right after allocation.
- skb_tail_pointer(skb) should be used to get
the right rx_buf pointer.
- setting URB_ZERO_PACKET is pointless for urbs
which are submitted to an IN endpoint.
---
drivers/net/wireless/libertas/if_usb.c +7 -11
@@ -506,17 +506,16 @@ static int __if_usb_submit_rx_urb(struct if_usb_card *cardp, goto rx_ret; } +skb_reserve (skb,IPFIELD_ALIGN_OFFSET ); cardp->rx_skb = skb;/* Fill the receive configuration URB and initialise the Rx call back */ usb_fill_bulk_urb (cardp->rx_urb, cardp->udev,usb_rcvbulkpipe (cardp->udev, cardp->ep_in), - (void *) (skb->tail + (size_t )IPFIELD_ALIGN_OFFSET ), +skb_tail_pointer (skb),MRVDRV_ETH_RX_PACKET_BUFFER_SIZE , callbackfn, cardp); - cardp->rx_urb->transfer_flags |=URB_ZERO_PACKET ; -lbs_deb_usb2 (&cardp->udev->dev,"Pointer for rx_urb %p\n" , cardp->rx_urb);if ((ret =usb_submit_urb (cardp->rx_urb,GFP_ATOMIC ))) {lbs_deb_usbd (&cardp->udev->dev,"Submit Rx URB failed: %d\n" , ret);
@@ -557,7 +556,7 @@ static void if_usb_receive_fwload(struct urb *urb) }if (cardp->fwdnldover) { -__le32 *tmp = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *tmp = (__le32 *)skb->data;if (tmp[0 ] ==cpu_to_le32 (CMD_TYPE_INDICATION ) && tmp[1 ] ==cpu_to_le32 (MACREG_INT_CODE_FIRMWARE_READY )) {
@@ -572,8 +571,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; }if (cardp->bootcmdresp <=0 ) { -memcpy (&bootcmdresp, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (bootcmdresp)); +memcpy (&bootcmdresp, skb->data,sizeof (bootcmdresp));if (le16_to_cpu (cardp->udev->descriptor.bcdDevice) <0x3106 ) {kfree_skb (skb);
@@ -619,8 +617,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; } -memcpy (syncfwheader, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (struct fwsyncheader )); +memcpy (syncfwheader, skb->data,sizeof (struct fwsyncheader ));if (!syncfwheader->cmd) {lbs_deb_usb2 (&cardp->udev->dev,"FW received Blk with correct CRC\n" );
@@ -665,7 +662,6 @@ static inline void process_cmdtypedata(int recvlength, struct sk_buff *skb,return ; } -skb_reserve (skb,IPFIELD_ALIGN_OFFSET );skb_put (skb, recvlength);skb_pull (skb,MESSAGE_HEADER_LEN );
@@ -719,7 +715,7 @@ static void if_usb_receive(struct urb *urb)int recvlength = urb->actual_length;uint8_t *recvbuff =NULL ;uint32_t recvtype =0 ; -__le32 *pkt = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *pkt = (__le32 *)skb->data;uint32_t event;lbs_deb_enter (LBS_DEB_USB );
@@ -732,7 +728,7 @@ static void if_usb_receive(struct urb *urb) goto setup_for_next; } - recvbuff = skb->data +IPFIELD_ALIGN_OFFSET ; + recvbuff = skb->data; recvtype =le32_to_cpu (pkt[0 ]); lbs_deb_usbd(&cardp->udev->dev, "Recv length = 0x%x, Recv type = 0x%X\n",
Is anyone evaluating these suggestions? Should I be expecting properly formatted patch emails?
John
Quoted 109 lines · Christian Lamparter
Quoted 7 lines · On Friday 30 October 2009 19:51:21 David Woodhouse wrote:
Quoted 3 lines · Christian Lamparter
that's just a fill-in macro.
AFAICTusb_submit_urbdoes the tricky dma mapping.Ah, that makes sense. In that case, all we need to do is make
map_urb_for_dma()do the right thing.well, but while we're on the subject of libertas.
I took a quick look around and wrote down some hopefully _helpful_ comments. That said, I don't have any libertas hw, so I have no idea if the attached code will actually do what its supposed to do... I'll leave it up to the professionals to test & write a real fix(es) with a proper commit message.
Notes:
- most +IPFIELD_ALIGN_OFFSETcan be replaced by a singleskb_reserve, right after allocation.-
skb_tail_pointer(skb)should be used to get
the rightrx_bufpointer.- setting
URB_ZERO_PACKETis pointless for urbs
which are submitted to an IN endpoint.
---drivers/net/wireless/libertas/if_usb.c +7
-11@@ -506,17 +506,16 @@ static int __if_usb_submit_rx_urb(struct if_usb_card *cardp, goto rx_ret; } +skb_reserve (skb,IPFIELD_ALIGN_OFFSET ); cardp->rx_skb = skb;/* Fill the receive configuration URB and initialise the Rx call back */ usb_fill_bulk_urb (cardp->rx_urb, cardp->udev,usb_rcvbulkpipe (cardp->udev, cardp->ep_in), - (void *) (skb->tail + (size_t )IPFIELD_ALIGN_OFFSET ), +skb_tail_pointer (skb),MRVDRV_ETH_RX_PACKET_BUFFER_SIZE , callbackfn, cardp); - cardp->rx_urb->transfer_flags |=URB_ZERO_PACKET ; -lbs_deb_usb2 (&cardp->udev->dev,"Pointer for rx_urb %p\n" , cardp->rx_urb);if ((ret =usb_submit_urb (cardp->rx_urb,GFP_ATOMIC ))) {lbs_deb_usbd (&cardp->udev->dev,"Submit Rx URB failed: %d\n" , ret);@@ -557,7 +556,7 @@ static void if_usb_receive_fwload(struct urb *urb) }if (cardp->fwdnldover) { -__le32 *tmp = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *tmp = (__le32 *)skb->data;if (tmp[0 ] ==cpu_to_le32 (CMD_TYPE_INDICATION ) && tmp[1 ] ==cpu_to_le32 (MACREG_INT_CODE_FIRMWARE_READY )) {@@ -572,8 +571,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; }if (cardp->bootcmdresp <=0 ) { -memcpy (&bootcmdresp, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (bootcmdresp)); +memcpy (&bootcmdresp, skb->data,sizeof (bootcmdresp));if (le16_to_cpu (cardp->udev->descriptor.bcdDevice) <0x3106 ) {kfree_skb (skb);@@ -619,8 +617,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; } -memcpy (syncfwheader, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (struct fwsyncheader )); +memcpy (syncfwheader, skb->data,sizeof (struct fwsyncheader ));if (!syncfwheader->cmd) {lbs_deb_usb2 (&cardp->udev->dev,"FW received Blk with correct CRC\n" );@@ -665,7 +662,6 @@ static inline void process_cmdtypedata(int recvlength, struct sk_buff *skb,return ; } -skb_reserve (skb,IPFIELD_ALIGN_OFFSET );skb_put (skb, recvlength);skb_pull (skb,MESSAGE_HEADER_LEN );@@ -719,7 +715,7 @@ static void if_usb_receive(struct urb *urb)int recvlength = urb->actual_length;uint8_t *recvbuff =NULL ;uint32_t recvtype =0 ; -__le32 *pkt = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *pkt = (__le32 *)skb->data;uint32_t event;lbs_deb_enter (LBS_DEB_USB );@@ -732,7 +728,7 @@ static void if_usb_receive(struct urb *urb) goto setup_for_next; } - recvbuff = skb->data +IPFIELD_ALIGN_OFFSET ; + recvbuff = skb->data; recvtype =le32_to_cpu (pkt[0 ]); lbs_deb_usbd(&cardp->udev->dev, "Recv length = 0x%x, Recv type = 0x%X\n",
Signature · 2 lines
John W. Linville Someday the world will need a hero, and you [email protected] might be all we have. Be ready.
Quoted 2 lines · John W. Linville
Is anyone evaluating these suggestions? Should I be expecting properly formatted patch emails?
The suggestions make sense, but not for stable. Let's fix the crasher bug first, and worry about the rest later.
Signature · 2 lines
David Woodhouse Open Source Technology Centre [email protected] Intel Corporation
Quoted 6 lines · David Woodhouse
Quoted 2 lines · John W. Linville
Is anyone evaluating these suggestions? Should I be expecting properly formatted patch emails?
The suggestions make sense, but not for stable. Let's fix the crasher bug first, and worry about the rest later.
Sounds good to me...
Dan
Quoted 9 lines · Dan Williams
Quoted 6 lines · David Woodhouse
Quoted 2 lines · John W. Linville
Is anyone evaluating these suggestions? Should I be expecting properly formatted patch emails?
The suggestions make sense, but not for stable. Let's fix the crasher bug first, and worry about the rest later.
Sounds good to me...
Yes, of course.
So, Christian, will you be following-up?
Thanks!
John
Signature · 2 lines
John W. Linville Someday the world will need a hero, and you [email protected] might be all we have. Be ready.
PATCH 1/1 libertas if_usb: tiny usb-rx overhaul
This patch reorganizes libertas' if_usb rx routines.
- URB_ZERO_PACKET flag has no use for urbs for
incoming endpoints.
- skb_tail_pointer(skb) should be used to get
the right rx_buf pointer.
- most + IPFIELD_ALIGN_OFFSET can be prevented by moving skb_reserve up right after allocation.
---
any Tested-by: ?
---
drivers/net/wireless/libertas/if_usb.c +7 -11
@@ -506,17 +506,16 @@ static int __if_usb_submit_rx_urb(struct if_usb_card *cardp, goto rx_ret; } +skb_reserve (skb,IPFIELD_ALIGN_OFFSET ); cardp->rx_skb = skb;/* Fill the receive configuration URB and initialise the Rx call back */ usb_fill_bulk_urb (cardp->rx_urb, cardp->udev,usb_rcvbulkpipe (cardp->udev, cardp->ep_in), - skb->data +IPFIELD_ALIGN_OFFSET , +skb_tail_pointer (skb),MRVDRV_ETH_RX_PACKET_BUFFER_SIZE , callbackfn, cardp); - cardp->rx_urb->transfer_flags |=URB_ZERO_PACKET ; -lbs_deb_usb2 (&cardp->udev->dev,"Pointer for rx_urb %p\n" , cardp->rx_urb);if ((ret =usb_submit_urb (cardp->rx_urb,GFP_ATOMIC ))) {lbs_deb_usbd (&cardp->udev->dev,"Submit Rx URB failed: %d\n" , ret);
@@ -557,7 +556,7 @@ static void if_usb_receive_fwload(struct urb *urb) }if (cardp->fwdnldover) { -__le32 *tmp = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *tmp = (__le32 *)skb->data;if (tmp[0 ] ==cpu_to_le32 (CMD_TYPE_INDICATION ) && tmp[1 ] ==cpu_to_le32 (MACREG_INT_CODE_FIRMWARE_READY )) {
@@ -572,8 +571,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; }if (cardp->bootcmdresp <=0 ) { -memcpy (&bootcmdresp, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (bootcmdresp)); +memcpy (&bootcmdresp, skb->data,sizeof (bootcmdresp));if (le16_to_cpu (cardp->udev->descriptor.bcdDevice) <0x3106 ) {kfree_skb (skb);
@@ -619,8 +617,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; } -memcpy (syncfwheader, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (struct fwsyncheader )); +memcpy (syncfwheader, skb->data,sizeof (struct fwsyncheader ));if (!syncfwheader->cmd) {lbs_deb_usb2 (&cardp->udev->dev,"FW received Blk with correct CRC\n" );
@@ -665,7 +662,6 @@ static inline void process_cmdtypedata(int recvlength, struct sk_buff *skb,return ; } -skb_reserve (skb,IPFIELD_ALIGN_OFFSET );skb_put (skb, recvlength);skb_pull (skb,MESSAGE_HEADER_LEN );
@@ -719,7 +715,7 @@ static void if_usb_receive(struct urb *urb)int recvlength = urb->actual_length;uint8_t *recvbuff =NULL ;uint32_t recvtype =0 ; -__le32 *pkt = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *pkt = (__le32 *)skb->data;uint32_t event;lbs_deb_enter (LBS_DEB_USB );
@@ -732,7 +728,7 @@ static void if_usb_receive(struct urb *urb) goto setup_for_next; } - recvbuff = skb->data +IPFIELD_ALIGN_OFFSET ; + recvbuff = skb->data; recvtype =le32_to_cpu (pkt[0 ]); lbs_deb_usbd(&cardp->udev->dev, "Recv length = 0x%x, Recv type = 0x%X\n",
Re: [PATCH] libertas if_usb: tiny usb-rx overhaul
Quoted 5 lines · Christian Lamparter
This patch reorganizes libertas'
if_usbrx routines.-
URB_ZERO_PACKETflag has no use for urbs for
incoming endpoints.
Quoted 7 lines
-
skb_tail_pointer(skb)should be used to get
the rightrx_bufpointer.- most +
IPFIELD_ALIGN_OFFSETcan be prevented by movingskb_reserveup right after allocation.Signed-off-by Christian Lamparter <[email protected]>
With 5.110.22p23 firmware if anyone cares... Thanks!
Quoted 87 lines
---
any Tested-by: ?
---drivers/net/wireless/libertas/if_usb.c +7
-11@@ -506,17 +506,16 @@ static int __if_usb_submit_rx_urb(struct if_usb_card *cardp, goto rx_ret; } +skb_reserve (skb,IPFIELD_ALIGN_OFFSET ); cardp->rx_skb = skb;/* Fill the receive configuration URB and initialise the Rx call back */ usb_fill_bulk_urb (cardp->rx_urb, cardp->udev,usb_rcvbulkpipe (cardp->udev, cardp->ep_in), - skb->data +IPFIELD_ALIGN_OFFSET , +skb_tail_pointer (skb),MRVDRV_ETH_RX_PACKET_BUFFER_SIZE , callbackfn, cardp); - cardp->rx_urb->transfer_flags |=URB_ZERO_PACKET ; -lbs_deb_usb2 (&cardp->udev->dev,"Pointer for rx_urb %p\n" , cardp->rx_urb);if ((ret =usb_submit_urb (cardp->rx_urb,GFP_ATOMIC ))) {lbs_deb_usbd (&cardp->udev->dev,"Submit Rx URB failed: %d\n" , ret);@@ -557,7 +556,7 @@ static void if_usb_receive_fwload(struct urb *urb) }if (cardp->fwdnldover) { -__le32 *tmp = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *tmp = (__le32 *)skb->data;if (tmp[0 ] ==cpu_to_le32 (CMD_TYPE_INDICATION ) && tmp[1 ] ==cpu_to_le32 (MACREG_INT_CODE_FIRMWARE_READY )) {@@ -572,8 +571,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; }if (cardp->bootcmdresp <=0 ) { -memcpy (&bootcmdresp, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (bootcmdresp)); +memcpy (&bootcmdresp, skb->data,sizeof (bootcmdresp));if (le16_to_cpu (cardp->udev->descriptor.bcdDevice) <0x3106 ) {kfree_skb (skb);@@ -619,8 +617,7 @@ static void if_usb_receive_fwload(struct urb *urb)return ; } -memcpy (syncfwheader, skb->data +IPFIELD_ALIGN_OFFSET , -sizeof (struct fwsyncheader )); +memcpy (syncfwheader, skb->data,sizeof (struct fwsyncheader ));if (!syncfwheader->cmd) {lbs_deb_usb2 (&cardp->udev->dev,"FW received Blk with correct CRC\n" );@@ -665,7 +662,6 @@ static inline void process_cmdtypedata(int recvlength, struct sk_buff *skb,return ; } -skb_reserve (skb,IPFIELD_ALIGN_OFFSET );skb_put (skb, recvlength);skb_pull (skb,MESSAGE_HEADER_LEN );@@ -719,7 +715,7 @@ static void if_usb_receive(struct urb *urb)int recvlength = urb->actual_length;uint8_t *recvbuff =NULL ;uint32_t recvtype =0 ; -__le32 *pkt = (__le32 *)(skb->data +IPFIELD_ALIGN_OFFSET ); +__le32 *pkt = (__le32 *)skb->data;uint32_t event;lbs_deb_enter (LBS_DEB_USB );@@ -732,7 +728,7 @@ static void if_usb_receive(struct urb *urb) goto setup_for_next; } - recvbuff = skb->data +IPFIELD_ALIGN_OFFSET ; + recvbuff = skb->data; recvtype =le32_to_cpu (pkt[0 ]); lbs_deb_usbd(&cardp->udev->dev, "Recv length = 0x%x, Recv type = 0x%X\n",_______________________________________________
libertas-dev mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/libertas-dev
Quoted 7 lines · David Woodhouse
Quoted 3 lines · Christian Lamparter
that's just a fill-in macro.
AFAICTusb_submit_urbdoes the tricky dma mapping.Ah, that makes sense. In that case, all we need to do is make
map_urb_for_dma()do the right thing.
map_urb_for_dma() does little more than call dma_map_single(). That's what you'll have to fix.
Alan Stern