Skip to contentSkip to conversation

FAILED: patch "[PATCH] xhci: fix lost bounce buffers on TDs spanning several ring" failed to apply to 5.10-stable tree

PATCH New [email protected] 2 participants 5 messages Sep 8 – Sep 10

Subsystems (3)
Stable branch USB USB xHCI driver
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 5.10-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y
git checkout FETCH_HEAD
git cherry-pick -x ff44dfb03a293bf30e31f98772a1dd316a6071d1
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2026090813-valuables-reuse-daab@gregkh' --subject-prefix 'PATCH 5.10.y' 'HEAD^..'

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From ff44dfb03a293bf30e31f98772a1dd316a6071d1 Mon Sep 17 00:00:00 2001
From: Arthur Gautier <[email protected]>
Date: Mon, 31 Aug 2026 12:04:48 +0300
Subject: [PATCH] xhci: fix lost bounce buffers on TDs spanning several ring
segments

When a TD reaches a link TRB with data that is not aligned to the endpoint's wMaxPacketSize, xhci_align_td() stages the unalignable tail through the bounce buffer of the ring segment holding that link TRB. xhci_unmap_td_bounce_buffer() later unmaps it and, for IN transfers, copies the data back into the URB's buffer.

The enqueue path records the segment that was bounced in td->bounce_seg, under the assumption that a TD never spans more than two ring segments. That assumption does not hold: a TD large enough to span three or more segments crosses several link TRBs and can be bounced at each of them. Only the last one survives in td->bounce_seg, so every earlier bounce buffer is neither copied back nor DMA unmapped.

The URB still completes with actual_length equal to the requested length and no error, so the transfer looks successful while a wMaxPacketSize sized hole in the destination buffer silently keeps its previous contents. It also leaks a DMA mapping per dropped bounce.

Any sufficiently large and fragmented bulk transfer can hit this. It was found with a USB mass storage device behind xHCI backing a dm-verity target with 512 byte hash blocks, where the stale data is detected rather than silently consumed. The device enumerates as SuperSpeed, so wMaxPacketSize is 1024, while dm-bufio issues one 512 byte bio per hash block. verity_prefetch_io() makes the block layer merge hundreds of them into a single request of up to 512 scatterlist entries of 512 bytes each. At 256 TRBs per ring segment such a TD spans three segments, and every segment boundary falls on an odd multiple of 512, i.e. unaligned to wMaxPacketSize. dm-bufio then caches a hash block holding stale data and dm-verity declares the metadata block corrupted:

device-mapper: verity: 8:2: metadata block 10850 is corrupted

A reproducer running this under qemu is available at
https://github.com/baloo/xhci-verity

The bounce state (bounce_buf, bounce_dma, bounce_len, bounce_offs) already lives on the ring segment, so there is nothing extra to track. Keep recording the last bounced segment in td->bounce_seg and, on completion, walk the segments from td->start_seg up to it, unmapping every segment that still has a pending bounce.

Stopping at td->bounce_seg rather than td->end_seg matters: a bounce implies the TD continues past that segment's link TRB, so bounce_seg is always strictly before end_seg, and a later TD may already have started in end_seg and been bounced there. Walking that far would copy a foreign bounce buffer into this URB and unmap it twice. It also keeps the walk correct if a TD ever wraps the whole ring so that end_seg == start_seg.

[mn: Add ring->num_segs check to prevent unlikely infinite for loop.]

Fixes f9c589e142d0 ("xhci: TD-fragment, align the unsplittable case with a bounce buffer")
Cc [email protected]
Suggested-by Michal Pecio <[email protected]>
Signed-off-by Arthur Gautier <[email protected]>
Signed-off-by Mathias Nyman <[email protected]>
Link https://patch.msgid.link/[email protected] open here
Signed-off-by Greg Kroah-Hartman <[email protected]>
drivers/usb/host/xhci-ring.c +24 -8
index 9847c5bfc41b..ec278a9f9540 100644
@@ -824,21 +824,18 @@ static void xhci_giveback_urb_in_irq(struct xhci_hcd *xhci,
 	usb_hcd_giveback_urb(hcd, urb, status);
 }
 
-static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
-		struct xhci_ring *ring, struct xhci_td *td)
+static void xhci_unmap_one_bounce_buffer(struct xhci_hcd *xhci,
+		struct xhci_ring *ring, struct xhci_td *td,
+		struct xhci_segment *seg)
 {
 	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
-	struct xhci_segment *seg = td->bounce_seg;
 	struct urb *urb = td->urb;
 	size_t len;
 
-	if (!ring || !seg || !urb)
-		return;
-
 	if (usb_urb_dir_out(urb)) {
 		dma_unmap_single(dev, seg->bounce_dma, ring->bounce_buf_len,
 				 DMA_TO_DEVICE);
-		return;
+		goto done;
 	}
 
 	dma_unmap_single(dev, seg->bounce_dma, ring->bounce_buf_len,

@@ -854,10 +851,29 @@ static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
 		memcpy(urb->transfer_buffer + seg->bounce_offs, seg->bounce_buf,
 		       seg->bounce_len);
 	}
+done:
 	seg->bounce_len = 0;
 	seg->bounce_offs = 0;
 }
 
+static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
+		struct xhci_ring *ring, struct xhci_td *td)
+{
+	struct xhci_segment *seg;
+	int i = 0;
+
+	if (!td->bounce_seg || !ring || !td->urb)
+		return;
+
+	/* td->bounce_seg is the last one bounced, unmap them all */
+	for (seg = td->start_seg; i++ < ring->num_segs; seg = seg->next) {
+		if (seg->bounce_len)
+			xhci_unmap_one_bounce_buffer(xhci, ring, td, seg);
+		if (seg == td->bounce_seg)
+			break;
+	}
+}
+
 static void xhci_td_cleanup(struct xhci_hcd *xhci, struct xhci_td *td,
 			    struct xhci_ring *ep_ring, int status)
 {

@@ -3685,7 +3701,7 @@ int xhci_queue_bulk_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
 						  &trb_buff_len,
 						  ring->enq_seg)) {
 					send_addr = ring->enq_seg->bounce_dma;
-					/* assuming TD won't span 2 segs */
+					/* TD bounced at least, and last on this seg */
 					td->bounce_seg = ring->enq_seg;
 				}
 			}
maintainer
#1 lore raw

PATCH 1/4 usb: xhci: Remove unused parameters of next_trb()

From: Michal Pecio <[email protected]>

[ Upstream commit 6b2eb0621ffb3ee9f3e497b483e8088436bdb07e ]

The function has two parameters which it doesn't use and hasn't ever used. One caller even puts NULL there, knowing it will work anyway.

Signed-off-by Michal Pecio <[email protected]>
Signed-off-by Mathias Nyman <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
Signed-off-by Greg Kroah-Hartman <[email protected]>

Stable-dep-of: ff44dfb03a29 ("xhci: fix lost bounce buffers on TDs spanning several ring segments")

Signed-off-by Sasha Levin <[email protected]>
---
 drivers/usb/host/xhci-ring.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)
drivers/usb/host/xhci-ring.c +7 -9
index a337e1fda33db..38e118047ff4c 100644
@@ -136,10 +136,8 @@ static void trb_to_noop(union xhci_trb *trb, u32 noop_type)
  * TRB is in a new segment.  This does not skip over link TRBs, and it does not
  * effect the ring dequeue or enqueue pointers.
  */
-static void next_trb(struct xhci_hcd *xhci,
-		struct xhci_ring *ring,
-		struct xhci_segment **seg,
-		union xhci_trb **trb)
+static void next_trb(struct xhci_segment **seg,
+			union xhci_trb **trb)
 {
 	if (trb_is_link(*trb)) {
 		*seg = (*seg)->next;

@@ -359,9 +357,9 @@ static int xhci_abort_cmd_ring(struct xhci_hcd *xhci, unsigned long flags)
 	 * avoiding corrupting the command ring pointer in case the command ring
 	 * is stopped by the time the upper dword is written.
 	 */
-	next_trb(xhci, NULL, &new_seg, &new_deq);
+	next_trb(&new_seg, &new_deq);
 	if (trb_is_link(new_deq))
-		next_trb(xhci, NULL, &new_seg, &new_deq);
+		next_trb(&new_seg, &new_deq);
 
 	crcr = xhci_trb_virt_to_dma(new_seg, new_deq);
 	xhci_write_64(xhci, crcr | CMD_RING_ABORT, &xhci->op_regs->cmd_ring);

@@ -654,7 +652,7 @@ void xhci_find_new_dequeue_state(struct xhci_hcd *xhci,
 		    link_trb_toggles_cycle(new_deq))
 			state->new_cycle_state ^= 0x1;
 
-		next_trb(xhci, ep_ring, &new_seg, &new_deq);
+		next_trb(&new_seg, &new_deq);
 
 		/* Search wrapped around, bail out */
 		if (new_deq == ep->ring->dequeue) {

@@ -703,7 +701,7 @@ static void td_to_noop(struct xhci_hcd *xhci, struct xhci_ring *ep_ring,
 		if (trb == td->last_trb)
 			break;
 
-		next_trb(xhci, ep_ring, &seg, &trb);
+		next_trb(&seg, &trb);
 	}
 }
 

@@ -2121,7 +2119,7 @@ static int sum_trb_lengths(struct xhci_hcd *xhci, struct xhci_ring *ring,
 	union xhci_trb *trb = ring->dequeue;
 	struct xhci_segment *seg = ring->deq_seg;
 
-	for (sum = 0; trb != stop_trb; next_trb(xhci, ring, &seg, &trb)) {
+	for (sum = 0; trb != stop_trb; next_trb(&seg, &trb)) {
 		if (!trb_is_noop(trb) && !trb_is_link(trb))
 			sum += TRB_LEN(le32_to_cpu(trb->generic.field[2]));
 	}
Signature · 1 line
2.53.0
maintainer
#2 lore raw

PATCH 2/4 usb: xhci: Fix sum_trb_lengths()

From: Michal Pecio <[email protected]>

[ Upstream commit ae71f9b88e5a15fed17a432e21c30ee9463ed1a4 ]

This function is supposed to sum the lengths of all transfer TRBs in a TD up to a point, but it starts summing at the current dequeue since it only ever gets called on the first pending TD.

This won't work when there are cancelled TDs at the beginning of the ring. The function tries to exclude No-Ops from the count, but not all cancelled TDs are No-Op'ed - not those the HW stopped on.

The absolutely obvious fix is to start counting at the TD's first TRB. And remove the now-useless 'ring' parameter, and 'xhci' too.

Signed-off-by Michal Pecio <[email protected]>
Signed-off-by Mathias Nyman <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
Signed-off-by Greg Kroah-Hartman <[email protected]>

Stable-dep-of: ff44dfb03a29 ("xhci: fix lost bounce buffers on TDs spanning several ring segments")

Signed-off-by Sasha Levin <[email protected]>
---
 drivers/usb/host/xhci-ring.c | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
drivers/usb/host/xhci-ring.c +7 -8
index 38e118047ff4c..3e7664a2476db 100644
@@ -2111,13 +2111,12 @@ static int finish_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 	return xhci_td_cleanup(xhci, td, ep_ring, td->status);
 }
 
-/* sum trb lengths from ring dequeue up to stop_trb, _excluding_ stop_trb */
-static int sum_trb_lengths(struct xhci_hcd *xhci, struct xhci_ring *ring,
-			   union xhci_trb *stop_trb)
+/* sum trb lengths from the first trb up to stop_trb, _excluding_ stop_trb */
+static u32 sum_trb_lengths(struct xhci_td *td, union xhci_trb *stop_trb)
 {
 	u32 sum;
-	union xhci_trb *trb = ring->dequeue;
-	struct xhci_segment *seg = ring->deq_seg;
+	union xhci_trb *trb = td->first_trb;
+	struct xhci_segment *seg = td->start_seg;
 
 	for (sum = 0; trb != stop_trb; next_trb(&seg, &trb)) {
 		if (!trb_is_noop(trb) && !trb_is_link(trb))

@@ -2308,7 +2307,7 @@ static int process_isoc_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		goto finish_td;
 
 	if (sum_trbs_for_length)
-		frame->actual_length = sum_trb_lengths(xhci, ep->ring, ep_trb) +
+		frame->actual_length = sum_trb_lengths(td, ep_trb) +
 			ep_trb_len - remaining;
 	else
 		frame->actual_length = requested;

@@ -2392,7 +2391,7 @@ static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		goto finish_td;
 	case COMP_STOPPED_LENGTH_INVALID:
 		/* stopped on ep trb with invalid length, exclude it */
-		td->urb->actual_length = sum_trb_lengths(xhci, ep_ring, ep_trb);
+		td->urb->actual_length = sum_trb_lengths(td, ep_trb);
 		goto finish_td;
 	case COMP_USB_TRANSACTION_ERROR:
 		if (xhci->quirks & XHCI_NO_SOFT_RETRY ||

@@ -2414,7 +2413,7 @@ static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		td->urb->actual_length = requested - remaining;
 	else
 		td->urb->actual_length =
-			sum_trb_lengths(xhci, ep_ring, ep_trb) +
+			sum_trb_lengths(td, ep_trb) +
 			ep_trb_len - remaining;
 finish_td:
 	if (remaining > requested) {
Signature · 1 line
2.53.0
maintainer
#3 lore raw

PATCH 3/4 usb: xhci: refactor xhci_td_cleanup() to return void

In reply to Sasha Levin

From: Niklas Neronin <[email protected]>

[ Upstream commit 7acfea2866acc6c888f880e28a249bc5ce069ad0 ]

The function is modified to return 'void' instead of an integer since it invariably returns '0'. Additionally, multiple functions which only return xhci_td_cleanup() are also refactored to return void.

This change eliminates the need for callers to handle a return value that does not convey meaningful information and improve code readability, as it becomes immediately clear that the function does not produce a significant output.

Signed-off-by Niklas Neronin <[email protected]>
Signed-off-by Mathias Nyman <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
Signed-off-by Greg Kroah-Hartman <[email protected]>

[5.10 backport: keep the existing if-based halt cleanup, stream_id arguments, last_trb/last_trb_seg fields and num_trbs_free accounting. Convert only the six existing cleanup/completion functions and their return statements; the newer upstream halt/cancellation paths are absent here. This supplies the void xhci_td_cleanup() context required by ff44dfb03a293bf30e31f98772a1dd316a6071d1 without adding functions.]

Stable-dep-of: ff44dfb03a29 ("xhci: fix lost bounce buffers on TDs spanning several ring segments")

Signed-off-by Sasha Levin <[email protected]>
---
 drivers/usb/host/xhci-ring.c | 53 +++++++++++++++++-------------------
 1 file changed, 25 insertions(+), 28 deletions(-)
drivers/usb/host/xhci-ring.c +25 -28
index 3e7664a2476db..eeb4d5de682d6 100644
@@ -771,8 +771,8 @@ static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
 	seg->bounce_offs = 0;
 }
 
-static int xhci_td_cleanup(struct xhci_hcd *xhci, struct xhci_td *td,
-			   struct xhci_ring *ep_ring, int status)
+static void xhci_td_cleanup(struct xhci_hcd *xhci, struct xhci_td *td,
+			    struct xhci_ring *ep_ring, int status)
 {
 	struct urb *urb = NULL;
 

@@ -813,8 +813,6 @@ static int xhci_td_cleanup(struct xhci_hcd *xhci, struct xhci_td *td,
 			status = 0;
 		xhci_giveback_urb_in_irq(xhci, td, status);
 	}
-
-	return 0;
 }
 
 static int xhci_reset_halted_ep(struct xhci_hcd *xhci, unsigned int slot_id,

@@ -2065,9 +2063,9 @@ int xhci_is_vendor_info_code(struct xhci_hcd *xhci, unsigned int trb_comp_code)
 	return 0;
 }
 
-static int finish_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
-		     struct xhci_ring *ep_ring, struct xhci_td *td,
-		     u32 trb_comp_code)
+static void finish_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
+		      struct xhci_ring *ep_ring, struct xhci_td *td,
+		      u32 trb_comp_code)
 {
 	struct xhci_ep_ctx *ep_ctx;
 

@@ -2080,7 +2078,7 @@ static int finish_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		 * stopped TDs.  A stopped TD may be restarted, so don't update
 		 * the ring dequeue pointer or take this TD off any lists yet.
 		 */
-		return 0;
+		return;
 	}
 	if (trb_comp_code == COMP_STALL_ERROR ||
 		xhci_requires_manual_halt_cleanup(xhci, ep_ctx,

@@ -2108,7 +2106,7 @@ static int finish_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		inc_deq(xhci, ep_ring);
 	}
 
-	return xhci_td_cleanup(xhci, td, ep_ring, td->status);
+	xhci_td_cleanup(xhci, td, ep_ring, td->status);
 }
 
 /* sum trb lengths from the first trb up to stop_trb, _excluding_ stop_trb */

@@ -2128,9 +2126,9 @@ static u32 sum_trb_lengths(struct xhci_td *td, union xhci_trb *stop_trb)
 /*
  * Process control tds, update urb status and actual_length.
  */
-static int process_ctrl_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
-		struct xhci_ring *ep_ring,  struct xhci_td *td,
-			   union xhci_trb *ep_trb, struct xhci_transfer_event *event)
+static void process_ctrl_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
+			    struct xhci_ring *ep_ring,  struct xhci_td *td,
+			    union xhci_trb *ep_trb, struct xhci_transfer_event *event)
 {
 	struct xhci_ep_ctx *ep_ctx;
 	u32 trb_comp_code;

@@ -2210,7 +2208,7 @@ static int process_ctrl_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		td->urb_length_set = true;
 		td->urb->actual_length = requested - remaining;
 		xhci_dbg(xhci, "Waiting for status stage event\n");
-		return 0;
+		return;
 	}
 
 	/* at status stage */

@@ -2218,15 +2216,15 @@ static int process_ctrl_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		td->urb->actual_length = requested;
 
 finish_td:
-	return finish_td(xhci, ep, ep_ring, td, trb_comp_code);
+	finish_td(xhci, ep, ep_ring, td, trb_comp_code);
 }
 
 /*
  * Process isochronous tds, update urb packet status and actual_length.
  */
-static int process_isoc_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
-		struct xhci_ring *ep_ring, struct xhci_td *td,
-		union xhci_trb *ep_trb, struct xhci_transfer_event *event)
+static void process_isoc_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
+			    struct xhci_ring *ep_ring, struct xhci_td *td,
+			    union xhci_trb *ep_trb, struct xhci_transfer_event *event)
 {
 	struct urb_priv *urb_priv;
 	int idx;

@@ -2319,14 +2317,13 @@ static int process_isoc_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 	if (td->error_mid_td && ep_trb != td->last_trb) {
 		xhci_dbg(xhci, "Error mid isoc TD, wait for final completion event\n");
 		td->urb_length_set = true;
-		return 0;
+		return;
 	}
-
-	return finish_td(xhci, ep, ep_ring, td, trb_comp_code);
+	finish_td(xhci, ep, ep_ring, td, trb_comp_code);
 }
 
-static int skip_isoc_td(struct xhci_hcd *xhci, struct xhci_td *td,
-			struct xhci_virt_ep *ep, int status)
+static void skip_isoc_td(struct xhci_hcd *xhci, struct xhci_td *td,
+			 struct xhci_virt_ep *ep, int status)
 {
 	struct urb_priv *urb_priv;
 	struct usb_iso_packet_descriptor *frame;

@@ -2348,15 +2345,15 @@ static int skip_isoc_td(struct xhci_hcd *xhci, struct xhci_td *td,
 	ep->ring->num_trbs_free += td->num_trbs - 1;
 	inc_deq(xhci, ep->ring);
 
-	return xhci_td_cleanup(xhci, td, ep->ring, status);
+	xhci_td_cleanup(xhci, td, ep->ring, status);
 }
 
 /*
  * Process bulk and interrupt tds, update urb status and actual_length.
  */
-static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
-		struct xhci_ring *ep_ring, struct xhci_td *td,
-		union xhci_trb *ep_trb, struct xhci_transfer_event *event)
+static void process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
+				 struct xhci_ring *ep_ring, struct xhci_td *td,
+				 union xhci_trb *ep_trb, struct xhci_transfer_event *event)
 {
 	struct xhci_slot_ctx *slot_ctx;
 	u32 trb_comp_code;

@@ -2403,7 +2400,7 @@ static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 
 		xhci_handle_halted_endpoint(xhci, ep, ep_ring->stream_id, td,
 					    EP_SOFT_RESET);
-		return 0;
+		return;
 	default:
 		/* do nothing */
 		break;

@@ -2422,7 +2419,7 @@ static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_virt_ep *ep,
 		td->urb->actual_length = 0;
 	}
 
-	return finish_td(xhci, ep, ep_ring, td, trb_comp_code);
+	finish_td(xhci, ep, ep_ring, td, trb_comp_code);
 }
 
 /*
Signature · 1 line
2.53.0
maintainer
#4 lore raw

PATCH 4/4 xhci: fix lost bounce buffers on TDs spanning several ring segments

In reply to Sasha Levin

From: Arthur Gautier <[email protected]>

[ Upstream commit ff44dfb03a293bf30e31f98772a1dd316a6071d1 ]

When a TD reaches a link TRB with data that is not aligned to the endpoint's wMaxPacketSize, xhci_align_td() stages the unalignable tail through the bounce buffer of the ring segment holding that link TRB. xhci_unmap_td_bounce_buffer() later unmaps it and, for IN transfers, copies the data back into the URB's buffer.

The enqueue path records the segment that was bounced in td->bounce_seg, under the assumption that a TD never spans more than two ring segments. That assumption does not hold: a TD large enough to span three or more segments crosses several link TRBs and can be bounced at each of them. Only the last one survives in td->bounce_seg, so every earlier bounce buffer is neither copied back nor DMA unmapped.

The URB still completes with actual_length equal to the requested length and no error, so the transfer looks successful while a wMaxPacketSize sized hole in the destination buffer silently keeps its previous contents. It also leaks a DMA mapping per dropped bounce.

Any sufficiently large and fragmented bulk transfer can hit this. It was found with a USB mass storage device behind xHCI backing a dm-verity target with 512 byte hash blocks, where the stale data is detected rather than silently consumed. The device enumerates as SuperSpeed, so wMaxPacketSize is 1024, while dm-bufio issues one 512 byte bio per hash block. verity_prefetch_io() makes the block layer merge hundreds of them into a single request of up to 512 scatterlist entries of 512 bytes each. At 256 TRBs per ring segment such a TD spans three segments, and every segment boundary falls on an odd multiple of 512, i.e. unaligned to wMaxPacketSize. dm-bufio then caches a hash block holding stale data and dm-verity declares the metadata block corrupted:

device-mapper: verity: 8:2: metadata block 10850 is corrupted

A reproducer running this under qemu is available at
https://github.com/baloo/xhci-verity

The bounce state (bounce_buf, bounce_dma, bounce_len, bounce_offs) already lives on the ring segment, so there is nothing extra to track. Keep recording the last bounced segment in td->bounce_seg and, on completion, walk the segments from td->start_seg up to it, unmapping every segment that still has a pending bounce.

Stopping at td->bounce_seg rather than td->end_seg matters: a bounce implies the TD continues past that segment's link TRB, so bounce_seg is always strictly before end_seg, and a later TD may already have started in end_seg and been bounced there. Walking that far would copy a foreign bounce buffer into this URB and unmap it twice. It also keeps the walk correct if a TD ever wraps the whole ring so that end_seg == start_seg.

[mn: Add ring->num_segs check to prevent unlikely infinite for loop.]

Fixes f9c589e142d0 ("xhci: TD-fragment, align the unsplittable case with a bounce buffer")
Cc [email protected]
Suggested-by Michal Pecio <[email protected]>
Signed-off-by Arthur Gautier <[email protected]>
Signed-off-by Mathias Nyman <[email protected]>
Link https://patch.msgid.link/[email protected] open here
Signed-off-by Greg Kroah-Hartman <[email protected]>
Signed-off-by Sasha Levin <[email protected]>
---
 drivers/usb/host/xhci-ring.c | 32 ++++++++++++++++++++++++--------
 1 file changed, 24 insertions(+), 8 deletions(-)
drivers/usb/host/xhci-ring.c +24 -8
index eeb4d5de682d6..491e6c488dd30 100644
@@ -737,21 +737,18 @@ static void xhci_giveback_urb_in_irq(struct xhci_hcd *xhci,
 	usb_hcd_giveback_urb(hcd, urb, status);
 }
 
-static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
-		struct xhci_ring *ring, struct xhci_td *td)
+static void xhci_unmap_one_bounce_buffer(struct xhci_hcd *xhci,
+		struct xhci_ring *ring, struct xhci_td *td,
+		struct xhci_segment *seg)
 {
 	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
-	struct xhci_segment *seg = td->bounce_seg;
 	struct urb *urb = td->urb;
 	size_t len;
 
-	if (!ring || !seg || !urb)
-		return;
-
 	if (usb_urb_dir_out(urb)) {
 		dma_unmap_single(dev, seg->bounce_dma, ring->bounce_buf_len,
 				 DMA_TO_DEVICE);
-		return;
+		goto done;
 	}
 
 	dma_unmap_single(dev, seg->bounce_dma, ring->bounce_buf_len,

@@ -767,10 +764,29 @@ static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
 		memcpy(urb->transfer_buffer + seg->bounce_offs, seg->bounce_buf,
 		       seg->bounce_len);
 	}
+done:
 	seg->bounce_len = 0;
 	seg->bounce_offs = 0;
 }
 
+static void xhci_unmap_td_bounce_buffer(struct xhci_hcd *xhci,
+		struct xhci_ring *ring, struct xhci_td *td)
+{
+	struct xhci_segment *seg;
+	int i = 0;
+
+	if (!td->bounce_seg || !ring || !td->urb)
+		return;
+
+	/* td->bounce_seg is the last one bounced, unmap them all */
+	for (seg = td->start_seg; i++ < ring->num_segs; seg = seg->next) {
+		if (seg->bounce_len)
+			xhci_unmap_one_bounce_buffer(xhci, ring, td, seg);
+		if (seg == td->bounce_seg)
+			break;
+	}
+}
+
 static void xhci_td_cleanup(struct xhci_hcd *xhci, struct xhci_td *td,
 			    struct xhci_ring *ep_ring, int status)
 {

@@ -3524,7 +3540,7 @@ int xhci_queue_bulk_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
 						  &trb_buff_len,
 						  ring->enq_seg)) {
 					send_addr = ring->enq_seg->bounce_dma;
-					/* assuming TD won't span 2 segs */
+					/* TD bounced at least, and last on this seg */
 					td->bounce_seg = ring->enq_seg;
 				}
 			}
Signature · 1 line
2.53.0