mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [BUG] usb: core: wBytesPerInterval checked with unsanitized bMaxBurst
@ 2026-11-15 16:37 Cheng Yongkang
  2026-11-16 18:51 ` Alan Stern
  0 siblings, 1 reply; 3+ messages in thread
From: Cheng Yongkang @ 2026-11-15 16:37 UTC (permalink / raw)
  To: Greg Kroah-Hartman; +Cc: Alan Stern, linux-usb, linux-kernel

Hi Greg, Alan,

I hit a corner case in usb_parse_ss_endpoint_companion(). It needs a
SuperSpeed device with an invalid bMaxBurst or Mult, so it may not be
worth fixing.

usb_parse_ss_endpoint_companion() clamps bMaxBurst to 15 and the isoc
Mult to 3 in ep->ss_ep_comp, but then computes max_tx for the
wBytesPerInterval check from the original descriptor. The stored
wBytesPerInterval can then be larger than the clamped endpoint can
carry.

With wMaxPacketSize=1024, bMaxBurst=16, Mult=1 and
wBytesPerInterval=65535, bMaxBurst becomes 15 but wBytesPerInterval
is set to 17408 instead of 16384. usb_submit_urb() uses the clamped
values and rejects isoc frames above 16384, while
usb_endpoint_max_periodic_payload() returns 17408, which xHCI also
uses as the Max ESIT Payload.

On v7.3 with a virtual SuperSpeed device over USB/IP, the kernel
logged "setting to 17408", and 16385- and 17000-byte isoc frames
failed with -EMSGSIZE. I only tested the bMaxBurst case; the Mult
case follows from the same code. The code is unchanged in
next-20261113.

Using ep->ss_ep_comp in the max_tx calculation fixes it:

  -		max_tx = (desc->bMaxBurst + 1) *
  -			(USB_SS_MULT(desc->bmAttributes)) *
  +		max_tx = (ep->ss_ep_comp.bMaxBurst + 1) *
  +			(USB_SS_MULT(ep->ss_ep_comp.bmAttributes)) *
   			usb_endpoint_maxp(&ep->desc);
   	else if (usb_endpoint_xfer_int(&ep->desc))
   		max_tx = usb_endpoint_maxp(&ep->desc) *
  -			(desc->bMaxBurst + 1);
  +			(ep->ss_ep_comp.bMaxBurst + 1);

I can send a patch if you want one.

Thanks,
Cheng Yongkang

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [BUG] usb: core: wBytesPerInterval checked with unsanitized bMaxBurst
  2026-11-15 16:37 [BUG] usb: core: wBytesPerInterval checked with unsanitized bMaxBurst Cheng Yongkang
@ 2026-11-16 18:51 ` Alan Stern
  2026-11-17  3:43   ` [PATCH] usb: core: use the clamped bMaxBurst and Mult for the wBytesPerInterval check Cheng Yongkang
  0 siblings, 1 reply; 3+ messages in thread
From: Alan Stern @ 2026-11-16 18:51 UTC (permalink / raw)
  To: Cheng Yongkang; +Cc: Greg Kroah-Hartman, linux-usb, linux-kernel

On Sun, Nov 15, 2026 at 08:37:01AM -0800, Cheng Yongkang wrote:
> Hi Greg, Alan,
> 
> I hit a corner case in usb_parse_ss_endpoint_companion(). It needs a
> SuperSpeed device with an invalid bMaxBurst or Mult, so it may not be
> worth fixing.
> 
> usb_parse_ss_endpoint_companion() clamps bMaxBurst to 15 and the isoc
> Mult to 3 in ep->ss_ep_comp, but then computes max_tx for the
> wBytesPerInterval check from the original descriptor. The stored
> wBytesPerInterval can then be larger than the clamped endpoint can
> carry.
> 
> With wMaxPacketSize=1024, bMaxBurst=16, Mult=1 and
> wBytesPerInterval=65535, bMaxBurst becomes 15 but wBytesPerInterval
> is set to 17408 instead of 16384. usb_submit_urb() uses the clamped
> values and rejects isoc frames above 16384, while
> usb_endpoint_max_periodic_payload() returns 17408, which xHCI also
> uses as the Max ESIT Payload.
> 
> On v7.3 with a virtual SuperSpeed device over USB/IP, the kernel
> logged "setting to 17408", and 16385- and 17000-byte isoc frames
> failed with -EMSGSIZE. I only tested the bMaxBurst case; the Mult
> case follows from the same code. The code is unchanged in
> next-20261113.
> 
> Using ep->ss_ep_comp in the max_tx calculation fixes it:
> 
>   -		max_tx = (desc->bMaxBurst + 1) *
>   -			(USB_SS_MULT(desc->bmAttributes)) *
>   +		max_tx = (ep->ss_ep_comp.bMaxBurst + 1) *
>   +			(USB_SS_MULT(ep->ss_ep_comp.bmAttributes)) *
>    			usb_endpoint_maxp(&ep->desc);
>    	else if (usb_endpoint_xfer_int(&ep->desc))
>    		max_tx = usb_endpoint_maxp(&ep->desc) *
>   -			(desc->bMaxBurst + 1);
>   +			(ep->ss_ep_comp.bMaxBurst + 1);
> 
> I can send a patch if you want one.

Please send a patch to the xHCI maintainer.

Alan Stern

^ permalink raw reply	[flat|nested] 3+ messages in thread

* [PATCH] usb: core: use the clamped bMaxBurst and Mult for the wBytesPerInterval check
  2026-11-16 18:51 ` Alan Stern
@ 2026-11-17  3:43   ` Cheng Yongkang
  0 siblings, 0 replies; 3+ messages in thread
From: Cheng Yongkang @ 2026-11-17  3:43 UTC (permalink / raw)
  To: Mathias Nyman; +Cc: Greg Kroah-Hartman, Alan Stern, linux-usb, linux-kernel

usb_parse_ss_endpoint_companion() clamps bMaxBurst to 15 and the isoc
Mult to 3 in ep->ss_ep_comp, but then computes max_tx for the
wBytesPerInterval check from the original descriptor. A device with an
invalid bMaxBurst or Mult can then keep a wBytesPerInterval that the
clamped endpoint cannot carry.

With wMaxPacketSize 1024, bMaxBurst 16 and Mult 1, bMaxBurst becomes 15
but wBytesPerInterval is limited to 17408 instead of 16384.
usb_submit_urb() uses the clamped values and rejects isoc frames above
16384, while usb_endpoint_max_periodic_payload() returns 17408, which
xHCI uses as the Max ESIT Payload. The same endpoint context gets a Max
Burst of 15, so the two fields disagree.

The rest of the code treats ep->ss_ep_comp as the sanitized copy of the
descriptor. usb_parse_endpoint() checks and fixes ep->desc the same
way, and everything after parsing, including usb_submit_urb(),
usb_endpoint_max_periodic_payload() and xHCI, reads only these copies.
max_tx itself already takes wMaxPacketSize from ep->desc. Before commit
842f16905dfc ("USB: remove the usb_host_ss_ep_comp structure"), the
clamps modified the descriptor in place, so max_tx was computed from
the clamped values. That commit moved the clamps to the copy but left
max_tx reading the original descriptor.

Compute max_tx from ep->ss_ep_comp.

Fixes: 842f16905dfc ("USB: remove the usb_host_ss_ep_comp structure")
Signed-off-by: Cheng Yongkang <teel4res@gmail.com>
---
Report: https://lore.kernel.org/linux-usb/20261115163701.226491-1-teel4res@gmail.com/

 drivers/usb/core/config.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/usb/core/config.c b/drivers/usb/core/config.c
index 0e79b7f8a73d..2f031ea6a437 100644
--- a/drivers/usb/core/config.c
+++ b/drivers/usb/core/config.c
@@ -170,12 +170,12 @@ static void usb_parse_ss_endpoint_companion(struct device *ddev, int cfgno,
 	}
 
 	if (usb_endpoint_xfer_isoc(&ep->desc))
-		max_tx = (desc->bMaxBurst + 1) *
-			(USB_SS_MULT(desc->bmAttributes)) *
+		max_tx = (ep->ss_ep_comp.bMaxBurst + 1) *
+			(USB_SS_MULT(ep->ss_ep_comp.bmAttributes)) *
 			usb_endpoint_maxp(&ep->desc);
 	else if (usb_endpoint_xfer_int(&ep->desc))
 		max_tx = usb_endpoint_maxp(&ep->desc) *
-			(desc->bMaxBurst + 1);
+			(ep->ss_ep_comp.bMaxBurst + 1);
 	else
 		max_tx = 999999;
 	/*

base-commit: 3149e8b3237f58d4dd440fc64c45c93404c4b064
-- 
2.34.1


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-17  3:43 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-15 16:37 [BUG] usb: core: wBytesPerInterval checked with unsanitized bMaxBurst Cheng Yongkang
2026-11-16 18:51 ` Alan Stern
2026-11-17  3:43   ` [PATCH] usb: core: use the clamped bMaxBurst and Mult for the wBytesPerInterval check Cheng Yongkang

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®