kernel/linux.git/drivers/usb/gadget/function, branch v6.12.93

usb: gadget: f_fs: serialize DMABUF cancel against request completion

2026-06-09T10:25:58+00:00

commit 2796646f6d892c1eb6818c7ca41fdfa12568e8d1 upstream. ffs_epfile_dmabuf_io_complete() calls usb_ep_free_request() on the completed request but leaves priv->req, the back-pointer that ffs_dmabuf_transfer() set on submission, pointing at the freed memory. A later FUNCTIONFS_DMABUF_DETACH ioctl or ffs_epfile_release() on the close path still sees priv->req non-NULL under ffs->eps_lock: if (priv->ep && priv->req) usb_ep_dequeue(priv->ep, priv->req); so usb_ep_dequeue() is called on a freed usb_request. On dummy_hcd the dequeue path only walks a live queue and pointer-compares, so the freed pointer reads without faulting and KASAN requires an explicit check at the FunctionFS call site to surface the use-after-free. On SG-capable in-tree UDCs the dequeue path dereferences the supplied request immediately: * chipidea's ep_dequeue() does container_of(req, struct ci_hw_req, req) and reads hwreq->req.status before acquiring its own lock. * cdnsp's cdnsp_gadget_ep_dequeue() reads request->status first. The narrower option of clearing priv->req via cmpxchg() in the completion does not close the race: the completion runs without eps_lock, so a cancel path holding eps_lock can still observe priv->req non-NULL, race a concurrent completion that clears and frees, and pass the freed pointer to usb_ep_dequeue(). A slightly longer fix that moves the free into the cleanup work is needed. Same class of lifetime race as the recent usbip-vudc timer fix [1]. Take eps_lock in the sole place that mutates priv->req from the callback direction by moving usb_ep_free_request() out of the completion into ffs_dmabuf_cleanup(), the existing work handler scheduled by ffs_dmabuf_signal_done() on ffs->io_completion_wq. Clear priv->req there under eps_lock before freeing, and only clear if priv->req still names our request (a subsequent ffs_dmabuf_transfer() on the same attachment may have queued a new one). This keeps the existing dummy_hcd sync-dequeue invariant: the completion callback is still invoked by the UDC without eps_lock held (dummy_hcd drops its own lock before calling the callback), and the callback now takes no f_fs lock at all. Serialization against the cancel path happens in cleanup, which runs from the workqueue with no f_fs lock held on entry. The priv ref count protects the containing ffs_dmabuf_priv: ffs_dmabuf_transfer() takes a ref via ffs_dmabuf_get(), cleanup drops it via ffs_dmabuf_put(), so priv stays live for the cleanup even after the cancel path's list_del + ffs_dmabuf_put. The ffs_dmabuf_transfer() error path no longer frees usb_req inline: fence->req and fence->ep are set before usb_ep_queue(), so ffs_dmabuf_cleanup() (scheduled by the error-path ffs_dmabuf_signal_done()) owns the free regardless of whether the queue succeeded. Reproduced under KASAN on both detach and close paths against dummy_hcd with an observability hook (kasan_check_byte(priv->req) immediately before usb_ep_dequeue) at the two FunctionFS cancel sites to surface the stale-pointer access; the hook is not part of this patch. The KASAN allocator / free stacks in the captured splats identify the same request: alloc in dummy_alloc_request, free in dummy_timer, fault reached from ffs_epfile_release (close) and from the FUNCTIONFS_DMABUF_DETACH ioctl (detach). With the patch applied, both paths are silent under the same hook. The bug is reached from the FunctionFS device node, which in real deployments is owned by the privileged gadget daemon (adbd, UMS, composite gadget services, etc.); it is not reachable from unprivileged userspace or from a USB host on the cable. FunctionFS mounts default to GLOBAL_ROOT_UID, but the filesystem supports uid=, gid=, and fmode= delegation to a non-root gadget daemon, so on real deployments the attacker may be a less-privileged service rather than root. Fixes: 7b07a2a7ca02 ("usb: gadget: functionfs: Add DMABUF import interface") Link: https://lore.kernel.org/all/20260417163552.807548-1-michael.bommarito@gmail.com/ [1] Cc: stable Assisted-by: Claude:claude-opus-4-7 Signed-off-by: Michael Bommarito Link: https://patch.msgid.link/20260419161227.1587668-1-michael.bommarito@gmail.com Signed-off-by: Greg Kroah-Hartman Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_fs: copy only received bytes on short ep0 read

2026-06-09T10:25:58+00:00

commit 4e036c10e7f4df5d951c69cc3697bc8e209c6d02 upstream. ffs_ep0_read() allocates its control-OUT data buffer with kmalloc() (not kzalloc) at the Length value from the Setup packet, then copies that full len to userspace regardless of how many bytes were actually received: data = kmalloc(len, GFP_KERNEL); ... ret = __ffs_ep0_queue_wait(ffs, data, len); if ((ret > 0) && (copy_to_user(buf, data, len))) ret = -EFAULT; __ffs_ep0_queue_wait() returns req->actual, which on a short control OUT transfer is strictly less than len. The copy_to_user() call still copies len bytes, so on a short OUT the last (len - ret) bytes of the kmalloc() buffer -- uninitialised slab residue -- are delivered to the FunctionFS daemon. Short ep0 OUT completions are specified USB control-transfer behavior and are produced by in-tree UDCs: * dwc2 continues on req->actual < req->length for ep0 DATA OUT (short-not-ok is the only ep0-OUT stall path). * aspeed_udc ends ep0 OUT on rx_len < ep->ep.maxpacket. * renesas_usbf logs "ep0 short packet" and completes the request. * dwc3 stalls on short IN but not on short OUT. A short ep0 OUT is therefore not evidence of a broken UDC; it is a normal condition f_fs has to cope with. The sibling gadgetfs implementation in drivers/usb/gadget/legacy/inode.c already does this correctly via min(len, dev->req->actual) before copy_to_user(). This patch brings f_fs.c to the same safe pattern rather than trimming at a defensive layer. The bug is reached from the FunctionFS device node, which in real deployments is owned by the privileged gadget daemon (adbd, UMS, composite gadget services, etc.); it is not reachable from unprivileged userspace. Linux host stacks normally reject short-wLength control OUTs before they reach the gadget, so reproducing this required a build that bypasses that host-side check. With the bypass in place, a 1-byte payload on a 64-byte Setup produces 63 bytes of non-canary slab residue in the daemon's read buffer. Fix by copying only ret (actually received) bytes to userspace. Fixes: ddf8abd25994 ("USB: f_fs: the FunctionFS driver") Cc: stable Assisted-by: Claude:claude-opus-4-7 Signed-off-by: Michael Bommarito Link: https://patch.msgid.link/20260419160359.1577270-1-michael.bommarito@gmail.com Signed-off-by: Greg Kroah-Hartman Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_hid: fix device reference leak in hidg_alloc()

2026-06-09T10:25:58+00:00

commit 4f88d65def6f3c90121601b4f62a4c967f3063a6 upstream. hidg_alloc() initializes hidg->dev with device_initialize() before calling dev_set_name(). If dev_set_name() fails, the function currently jumps to err_unlock and returns without calling put_device(). This leaves the device reference unbalanced and prevents hidg_release() from being called. Calling put_device() here is also safe, since hidg_release() only frees resources owned by hidg. The issue was identified by a static analysis tool I developed and confirmed by manual review. Route the dev_set_name() failure path through err_put_device so the device reference is dropped properly. Fixes: 89ff3dfac604 ("usb: gadget: f_hid: fix f_hidg lifetime vs cdev") Cc: stable Reviewed-by: Johan Hovold Signed-off-by: Guangshuo Li Reviewed-by: Johan Hovold johan@kernel.org Link: https://patch.msgid.link/20260413142119.2977716-1-lgs201920130244@gmail.com Signed-off-by: Greg Kroah-Hartman

usb: gadget: uvc: hold opts->lock across XU walks in uvc_function_bind

2026-06-09T10:25:58+00:00

commit 68aa70648b625fa684bc0b71bbfd905f4943ca20 upstream. uvc_function_bind() walks &opts->extension_units twice without holding opts->lock: - directly, for the iExtension string-descriptor fixup loop; - indirectly, four times via uvc_copy_descriptors() (once per speed), where the helper iterates uvc->desc.extension_units (which aliases &opts->extension_units) to size and emit XU descriptors. The configfs side (uvcg_extension_make / uvcg_extension_drop, in drivers/usb/gadget/function/uvc_configfs.c) takes opts->lock around its list_add_tail / list_del operations. A privileged userspace process that holds the configfs subtree open and writes the gadget UDC name to bind the function while concurrently rmdir()'ing an extensions subdir can race uvcg_extension_drop() against the bind-time list walks and dereference a freed struct uvcg_extension. Hold opts->lock from the start of the XU string-descriptor fixup through the last uvc_copy_descriptors() call, releasing on the descriptor-error path via a new error_unlock label that drops the lock before falling through to the existing error label. This matches the locking discipline of the configfs callbacks and removes the only remaining unsynchronised reader of the XU list during bind. Reachability: only privileged processes that can mount configfs and write to gadget UDC files can trigger the race, so this is a correctness fix rather than a security boundary. Fixes: 0525210c9840 ("usb: gadget: uvc: Allow definition of XUs in configfs") Cc: stable Signed-off-by: Kai Aizen Link: https://patch.msgid.link/20260430175643.67120-1-kai.aizen.dev@gmail.com Signed-off-by: Greg Kroah-Hartman Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_hid: don't call cdev_init while cdev in use

2026-04-22T11:18:59+00:00

commit 81ebd43cc0d6d106ce7b6ccbf7b5e40ca7f5503d upstream. When calling unbind, then bind again, cdev_init reinitialized the cdev, even though there may still be references to it. That's the case when the /dev/hidg* device is still opened. This obviously unsafe behavior like oopes. This fixes this by using cdev_alloc to put the cdev on the heap. That way, we can simply allocate a new one in hidg_bind. Closes: https://lore.kernel.org/linux-usb/CAN9vWDKZn0Ts5JyV2_xcAmbnBEi0znMLg_USMFrShRryXrgWGQ@mail.gmail.com/T/#m2cb0dba3633b67b2a679c98499508267d1508881 Cc: stable Signed-off-by: Michael Zimmermann Link: https://patch.msgid.link/20260327192209.59945-1-sigmaepsilon92@gmail.com Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_phonet: fix skb frags[] overflow in pn_rx_complete()

2026-04-22T11:18:58+00:00

commit c088d5dd2fffb4de1fb8e7f57751c8b82942180a upstream. A broken/bored/mean USB host can overflow the skb_shared_info->frags[] array on a Linux gadget exposing a Phonet function by sending an unbounded sequence of full-page OUT transfers. pn_rx_complete() finalizes the skb only when req->actual < req->length, where req->length is set to PAGE_SIZE by the gadget. If the host always sends exactly PAGE_SIZE bytes per transfer, fp->rx.skb will never be reset and each completion will add another fragment via skb_add_rx_frag(). Once nr_frags exceeds MAX_SKB_FRAGS (default 17), subsequent frag stores overwrite memory adjacent to the shinfo on the heap. Drop the skb and account a length error when the frag limit is reached, matching the fix applied in t7xx by commit f0813bcd2d9d ("net: wwan: t7xx: fix potential skb->frags overflow in RX path"). Cc: stable Assisted-by: gregkh_clanker_t1000 Link: https://patch.msgid.link/2026040705-fruit-unloved-0701@gregkh Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_ncm: validate minimum block_len in ncm_unwrap_ntb()

2026-04-22T11:18:57+00:00

commit 8f993d30b95dc9557a8a96ceca11abed674c8acb upstream. The block_len read from the host-supplied NTB header is checked against ntb_max but has no lower bound. When block_len is smaller than opts->ndp_size, the bounds check of: ndp_index > (block_len - opts->ndp_size) will underflow producing a huge unsigned value that ndp_index can never exceed, defeating the check entirely. The same underflow occurs in the datagram index checks against block_len - opts->dpe_size. With those checks neutered, a malicious USB host can choose ndp_index and datagram offsets that point past the actual transfer, and the skb_put_data() copies adjacent kernel memory into the network skb. Fix this by rejecting block lengths that cannot hold at least the NTB header plus one NDP. This will make block_len - opts->ndp_size and block_len - opts->dpe_size both well-defined. Commit 8d2b1a1ec9f5 ("CDC-NCM: avoid overflow in sanity checking") fixed a related class of issues on the host side of NCM. Fixes: 2b74b0a04d3e ("USB: gadget: f_ncm: add bounds checks to ncm_unwrap_ntb()") Cc: stable Assisted-by: gregkh_clanker_t1000 Link: https://patch.msgid.link/2026040753-baffle-handheld-624d@gregkh Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_uac1_legacy: validate control request size

2026-04-11T12:24:51+00:00

commit 6e0e34d85cd46ceb37d16054e97a373a32770f6c upstream. f_audio_complete() copies req->length bytes into a 4-byte stack variable: u32 data = 0; memcpy(&data, req->buf, req->length); req->length is derived from the host-controlled USB request path, which can lead to a stack out-of-bounds write. Validate req->actual against the expected payload size for the supported control selectors and decode only the expected amount of data. This avoids copying a host-influenced length into a fixed-size stack object. Signed-off-by: Taegu Ha Cc: stable Link: https://patch.msgid.link/20260401191311.3604898-1-hataegu0826@gmail.com Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_hid: move list and spinlock inits from bind to alloc

2026-04-11T12:24:51+00:00

commit 4e0a88254ad59f6c53a34bf5fa241884ec09e8b2 upstream. There was an issue when you did the following: - setup and bind an hid gadget - open /dev/hidg0 - use the resulting fd in EPOLL_CTL_ADD - unbind the UDC - bind the UDC - use the fd in EPOLL_CTL_DEL When CONFIG_DEBUG_LIST was enabled, a list_del corruption was reported within remove_wait_queue (via ep_remove_wait_queue). After some debugging I found out that the queues, which f_hid registers via poll_wait were the problem. These were initialized using init_waitqueue_head inside hidg_bind. So effectively, the bind function re-initialized the queues while there were still items in them. The solution is to move the initialization from hidg_bind to hidg_alloc to extend their lifetimes to the lifetime of the function instance. Additionally, I found many other possibly problematic init calls in the bind function, which I moved as well. Signed-off-by: Michael Zimmermann Cc: stable Link: https://patch.msgid.link/20260331184844.2388761-1-sigmaepsilon92@gmail.com Signed-off-by: Greg Kroah-Hartman

usb: gadget: f_rndis: Fix net_device lifecycle with device_move

2026-04-11T12:24:51+00:00

commit e367599529dc42578545a7f85fde517b35b3cda7 upstream. The net_device is allocated during function instance creation and registered during the bind phase with the gadget device as its sysfs parent. When the function unbinds, the parent device is destroyed, but the net_device survives, resulting in dangling sysfs symlinks: console:/ # ls -l /sys/class/net/usb0 lrwxrwxrwx ... /sys/class/net/usb0 -> /sys/devices/platform/.../gadget.0/net/usb0 console:/ # ls -l /sys/devices/platform/.../gadget.0/net/usb0 ls: .../gadget.0/net/usb0: No such file or directory Use device_move() to reparent the net_device between the gadget device tree and /sys/devices/virtual across bind and unbind cycles. During the final unbind, calling device_move(NULL) moves the net_device to the virtual device tree before the gadget device is destroyed. On rebinding, device_move() reparents the device back under the new gadget, ensuring proper sysfs topology and power management ordering. To maintain compatibility with legacy composite drivers (e.g., multi.c), the borrowed_net flag is used to indicate whether the network device is shared and pre-registered during the legacy driver's bind phase. Fixes: f466c6353819 ("usb: gadget: f_rndis: convert to new function interface with backward compatibility") Cc: stable@vger.kernel.org Signed-off-by: Kuen-Han Tsai Link: https://patch.msgid.link/20260320-usb-net-lifecycle-v1-7-4886b578161b@google.com Signed-off-by: Greg Kroah-Hartman