$ the-wire · showcase
bnxt XDP leak fixes, gdb remote parser overflows
By RepoJournal · Filed · About Linux · Composed from the cited sources · methodology
The bnxt driver shipped two XDP leak fixes that matter to anyone running AF_XDP redirects, while kgdb's remote protocol parser got a family of buffer-overflow fixes.
eth: bnxt: free XDP frags when xdp_do_redirect() fails torvalds/linux
When xdp_do_redirect() failed, bnxt_rx_xdp() recycled the head page but not the frags of a multi-buffer frame, and since the agg ring slots were already refilled those pages leaked permanently, blocking page pool destruction. Any redirect to an AF_XDP socket in copy mode whose fill ring runs dry hits this with -ENOMEM; a socket bound without XDP_USE_SG fails with -ENOSPC.
eth: bnxt: don't leak buffers when XDP agg buffer allocation fails torvalds/linux
If a replacement agg buffer can't be allocated, bnxt_rx_pkt() drops the packet without returning its head buffer, and rx_prod still advances so the device receives a BD pointing at a stale DMA address the driver no longer owns. The change also returns the agg buffers consumed before the failing one.
net/packet: check TX_RING fragments before memory accounting torvalds/linux
tpacket_fill_skb() charged the full payload length to skb->truesize and sk_wmem_alloc before checking MAX_SKB_FRAGS, letting userspace with PACKET_VNET_HDR submit a ring-sized frame far above the device MTU. A near-INT_MAX frame can saturate sk_wmem_alloc before the fragment loop rejects it, and a subsequent refcount_add(0x40000001) can temporarily turn REFCOUNT_SATURATED into 1.
Merge tag 'kgdb-7.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/danielt/linux torvalds/linux
Daniel Thompson's pull request describes "a family of related buffer overflows in the gdb remote protocol parsing" from Fang Xieyan, plus a memory leak fixed when the environment is already full and the removal of a CMD_BUFLEN that held different values in different files. Only relevant to you if you attach gdb over kgdb, but the parsing fixes are worth backporting.
Keep the keyboard backlight level when the lock blanks again before a wake (#15908) basecamp/omarchy
In the long tail: the lock's second blank before a wake saved the 0 that the first blank had set, so unlock restored zero backlight. Meanwhile Omarchy now supports the ASUS Zenbook A16's speakers and microphones on its Snapdragon X2 Elite hardware, and stopped a hidden bar from swallowing clicks over fullscreen windows under Hyprland 0.56.
