Files
kernel-agent/patches/driver/bes2600/tx-sdio-dma-oob-danctnix/0001-bes2600-bounce-SDIO-TX-buffers-to-avoid-DMA-OOB-read.patch
T
claude-noether b04c8cd501 patches/driver/bes2600/*-danctnix + arch/arm64/scs-...: rebased on danctnix baseline (#29 redo)
PR #33's per-series mirrors were generated against the bes2600-dkms
cleanups branch (rooted at fe73571) without rebasing onto the
v7.0-danctnix1 kernel baseline. Result: per-commit diffs carried
stale baseline context (e.g. from_timer rather than the new
timer_container_of API), so the cumulative no longer applied cleanly
to ohm's actual base. pkgrel=6 build #1 failed with 'Hunk #3 FAILED'
in Patch D's sta.c.

Fix: in marfrit/bes2600-dkms, create danctnix-sync branch
(fe73571 + drop-in replace bes2600/ with v7.0-danctnix1's
drivers/staging/bes2600/), rebase cleanups onto it as
cleanups-rebased-on-danctnix, manually resolve the resulting conflicts
keeping each commit's intent + the new baseline context, rebase
Patch H accordingly. Format-patch and re-route to the same series-dir
names as PR #33.

Conflict resolution notes:
- 'remove userspace /dev/bes2600 character device interface' commit:
  the chardev wrapper was removed but two utility funcs that danctnix's
  bes2600_btuart.c depends on (bes2600_chrdev_is_bus_error,
  bes2600_chrdev_switch_subsys_glb) were re-added with EXPORT_SYMBOL_GPL.
  bes2600_switch_bt re-added as static (file-local, called only from
  bes2600_chrdev_switch_subsys_glb).
- Patch D (atomicize ba_lock): re-resolved bes2600_ba_timer's
  timer_container_of() vs from_timer() to keep the new API.
- SCS Makefile @@ hunk counts corrected from -9,6 +9,10 to -9,6 +9,11
  (the original was actually wrong; build-via-fuzz was masking it).

Cumulative b2sum: ka-promote ohm now emits
  eb179c03f35a4dbaec2e40036f0033ef04985bb6b14ab22419d68e5caaa5874f...
  (279 554 bytes, 32 patches resolved).

pkgrel=6 built from this manifest + installed on ohm 2026-05-19 ~23:39.
Functional verification: bes2600 + bes2600_btuart both load, Pattern A
0 over fresh boot, wlan0 associates to newton. srcversion
1A919EED0E6DC2478559B17 differs from pkgrel=5's BEB625FA... — the
reconstruction is functionally equivalent (5 GHz working, no
firmware/driver race conditions) but NOT byte-equivalent (the chardev
utility re-add chose different formatting than the original danctnix
code). Byte-equivalence is not a goal; per-series traceability and
working hardware are.

Closes (proper this time): #29.
Refs: #28, #30, #33 (the half-working attempt), #31, #32.
2026-05-19 23:44:29 +02:00

124 lines
4.4 KiB
Diff

From 2f9b4c719faf9563895c064439a7da25f35c8fc7 Mon Sep 17 00:00:00 2001
From: Markus Fritsche <fritsche.markus@gmail.com>
Date: Thu, 23 Apr 2026 11:58:31 +0200
Subject: [PATCH 07/29] bes2600: bounce SDIO TX buffers to avoid DMA OOB read
The SDIO TX path rounds the DMA transfer length up to the host's
current block size and hands that length to dma_map_sg() via
sg_set_buf(&sg[scatters], tx_buffer->buf, align) in sdio_tx_work().
tx_buffer->buf typically aliases into an skb linear head whose
allocated size matches tx_buffer->len, not the block-aligned
align. The DMA engine (swiotlb / dw_mci IDMAC) therefore reads up
to one block past the end of the skb. On a PineTab2 with KFENCE
enabled this fires as:
BUG: KFENCE: out-of-bounds read in __pi_memcpy_generic
Out-of-bounds read at ... (704B right of kfence-#...):
__pi_memcpy_generic
swiotlb_tbl_map_single
swiotlb_map
dma_direct_map_sg
__dma_map_sg_attrs
dma_map_sg_attrs
dw_mci_pre_dma_transfer
__dw_mci_start_request
...
bes_sdio_memcpy_to_io_helper+0x18c/0x288 [bes2600]
sdio_tx_work+0x2b4/0x4a0 [bes2600]
allocated by ... pskb_expand_head / validate_xmit_skb / tcp_*
In addition to being undefined behavior, the padding bytes (which
come from whatever memory follows the skb) are transmitted to the
peer, leaking kernel memory on the air.
Allocate a driver-owned DMA-page bounce buffer sized to
MAX_SDIO_TRANSFER_LEN and use it as the scatter-gather backing for
sdio_tx_work. Each TX buffer is copied into its bounce slot and the
tail (align - tx_buffer->len bytes) is zeroed. This mirrors the
existing bounce pattern already used by bes2600_sdio_memcpy_toio()
via single_gathered_buffer; a separate allocation is used for the
TX path because single_gathered_buffer is only serialised via
sdio_claim_host and sdio_tx_work accumulates scatter entries before
claiming the bus.
Signed-off-by: Markus Fritsche <fritsche.markus@gmail.com>
---
bes2600/bes2600_sdio.c | 39 ++++++++++++++++++++++++++++++++++++++-
1 file changed, 38 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/bes2600/bes2600_sdio.c b/drivers/staging/bes2600/bes2600_sdio.c
index f172d53..b9d836f 100644
--- a/drivers/staging/bes2600/bes2600_sdio.c
+++ b/drivers/staging/bes2600/bes2600_sdio.c
@@ -95,6 +95,7 @@ struct sbus_priv {
struct work_struct tx_work;
struct scatterlist tx_sg[BES_SDIO_TX_MULTIPLE_NUM + 1];
struct scatterlist tx_sg_nosignal[BES_SDIO_TX_MULTIPLE_NUM_NOSIGNAL + 1];
+ u8 *tx_bounce;
u32 tx_data_cnt;
u32 tx_xfer_cnt;
u32 tx_proc_cnt;
@@ -1136,7 +1137,26 @@ static void sdio_tx_work(struct work_struct *work)
}
}
- sg_set_buf(&sg[scatters], tx_buffer->buf, align);
+ /*
+ * The transfer length is rounded up to the SDIO block
+ * size, but tx_buffer->buf is only tx_buffer->len bytes
+ * long (it usually aliases into an skb linear head).
+ * Copy into a driver-owned bounce buffer and zero-pad
+ * to the aligned size; otherwise DMA reads past the
+ * skb and leaks adjacent kernel memory on the wire --
+ * observed as KFENCE OOB reads from
+ * bes_sdio_memcpy_to_io_helper via dma_map_sg.
+ */
+ if (WARN_ON_ONCE(total_len + align > MAX_SDIO_TRANSFER_LEN))
+ goto flush_previous;
+ memcpy(self->tx_bounce + total_len,
+ tx_buffer->buf, tx_buffer->len);
+ if (align > tx_buffer->len)
+ memset(self->tx_bounce + total_len +
+ tx_buffer->len, 0,
+ align - tx_buffer->len);
+ sg_set_buf(&sg[scatters],
+ self->tx_bounce + total_len, align);
total_len += align;
++scatters;
/*del_node:*/
@@ -1857,6 +1877,17 @@ static int bes2600_sdio_probe(struct sdio_func *func,
if (!self->single_gathered_buffer)
return -ENOMEM;
#endif
+#ifdef BES_SDIO_TX_MULTIPLE_ENABLE
+ self->tx_bounce = (u8 *)__get_dma_pages(GFP_KERNEL,
+ get_order(MAX_SDIO_TRANSFER_LEN));
+ if (!self->tx_bounce) {
+#ifndef SDIO_HOST_ADMA_SUPPORT
+ free_pages((unsigned long)self->single_gathered_buffer,
+ get_order(MAX_SDIO_TRANSFER_LEN));
+#endif
+ return -ENOMEM;
+ }
+#endif
#ifdef BES_SDIO_RXTX_TOGGLE
self->fw_started = false;
#endif
@@ -1984,6 +2015,12 @@ static void bes2600_sdio_remove(struct sdio_func *func)
if (self->single_gathered_buffer) {
free_pages((unsigned long)self->single_gathered_buffer, get_order(MAX_SDIO_TRANSFER_LEN));
}
+#endif
+#ifdef BES_SDIO_TX_MULTIPLE_ENABLE
+ if (self->tx_bounce) {
+ free_pages((unsigned long)self->tx_bounce,
+ get_order(MAX_SDIO_TRANSFER_LEN));
+ }
#endif
kfree(self);
}
--
2.54.0