From: Robert Hodaszi robert.hodaszi@digi.com
[ Upstream commit 387602d8a75574fafb451b7a8215e78dfd67ee63 ]
Don't set WDM_READ flag in wdm_in_callback() for ZLP-s, otherwise when userspace tries to poll for available data, it might - incorrectly - believe there is something available, and when it tries to non-blocking read it, it might get stuck in the read loop.
For example this is what glib does for non-blocking read (briefly):
1. poll() 2. if poll returns with non-zero, starts a read data loop: a. loop on poll() (EINTR disabled) b. if revents was set, reads data I. if read returns with EINTR or EAGAIN, goto 2.a. II. otherwise return with data
So if ZLP sets WDM_READ (#1), we expect data, and try to read it (#2). But as that was a ZLP, and we are doing non-blocking read, wdm_read() returns with EAGAIN (#2.b.I), so loop again, and try to read again (#2.a.).
With glib, we might stuck in this loop forever, as EINTR is disabled (#2.a).
Signed-off-by: Robert Hodaszi robert.hodaszi@digi.com Acked-by: Oliver Neukum oneukum@suse.com Link: https://lore.kernel.org/r/20250403144004.3889125-1-robert.hodaszi@digi.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org Signed-off-by: Sasha Levin sashal@kernel.org ---
## Backport Analysis: YES
This commit should be backported to stable kernel trees for the following reasons:
### 1. **Clear User-Facing Bug Fix** The commit fixes a critical bug that causes userspace applications to get stuck in infinite poll/read loops when dealing with Zero-Length Packets (ZLPs). This is a regression that affects real-world applications using glib and similar libraries with cdc-wdm devices.
### 2. **Minimal and Surgical Changes** The fix is extremely contained and follows the stable tree criteria: - **Small code change**: Only ~20 lines modified, mostly removing problematic code - **Well-defined scope**: Changes are limited to `wdm_in_callback()` function - **Clear logic**: Simple check for `length == 0` and skip setting `WDM_READ` flag - **No architectural changes**: Doesn't modify core data structures or algorithms
### 3. **Low Regression Risk** The changes are inherently safe: - **Removing problematic behavior**: The fix removes the incorrect setting of `WDM_READ` for ZLPs - **Maintains existing functionality**: All other packet handling remains unchanged - **Proper fallback**: ZLPs still trigger service work via `schedule_work(&desc->service_outs_intr)` - **Backwards compatible**: No API changes that could affect other drivers
### 4. **Critical Subsystem Impact** The USB CDC-WDM driver is used by many mobile modems and WWAN devices. Getting stuck in infinite loops renders these devices unusable, which is a severe user impact.
### 5. **Alignment with Historical Backports** This commit follows the same pattern as the successful backports shown in the examples: - Similar to **Commit #1 (YES)**: Fixes flag management issue that causes device to become unusable - Similar to **Commit #2 (YES)**: Addresses race condition in flag handling - Similar to **Commit #3 (YES)**: Prevents userspace applications from hanging
### 6. **Code Analysis Details**
**Key changes that support backporting:**
1. **ZLP Detection**: `if (length == 0)` - Simple, safe check 2. **Flag Management Fix**: Prevents incorrect `WDM_READ` setting for ZLPs 3. **Cleanup**: Removes `reslength` field and related zero-length handling code that was causing the infinite loop 4. **Proper Service Handling**: ZLPs are still processed via work queue (`skip_zlp` goto)
**Risk mitigation:** - The `service_interrupt_work()` function change adds a condition `(desc->length || desc->rerr)` that ensures `WDM_READ` is only set when there's actual data or an error to report - This prevents the exact scenario that caused infinite loops
### 7. **Real-World Impact** From the kernel repository examination, this appears to be addressing a recently discovered issue that affects applications using glib's non- blocking I/O patterns with CDC-WDM devices. The infinite loop scenario described is a critical usability bug.
**Conclusion**: This is a textbook example of a stable-worthy fix - it addresses a clear user-facing bug with minimal, surgical changes that have low regression risk and high benefit for affected users.
drivers/usb/class/cdc-wdm.c | 23 +++++++++-------------- 1 file changed, 9 insertions(+), 14 deletions(-)
diff --git a/drivers/usb/class/cdc-wdm.c b/drivers/usb/class/cdc-wdm.c index 16e7fa4d488d3..ecd6d1f39e498 100644 --- a/drivers/usb/class/cdc-wdm.c +++ b/drivers/usb/class/cdc-wdm.c @@ -92,7 +92,6 @@ struct wdm_device { u16 wMaxCommand; u16 wMaxPacketSize; __le16 inum; - int reslength; int length; int read; int count; @@ -214,6 +213,11 @@ static void wdm_in_callback(struct urb *urb) if (desc->rerr == 0 && status != -EPIPE) desc->rerr = status;
+ if (length == 0) { + dev_dbg(&desc->intf->dev, "received ZLP\n"); + goto skip_zlp; + } + if (length + desc->length > desc->wMaxCommand) { /* The buffer would overflow */ set_bit(WDM_OVERFLOW, &desc->flags); @@ -222,18 +226,18 @@ static void wdm_in_callback(struct urb *urb) if (!test_bit(WDM_OVERFLOW, &desc->flags)) { memmove(desc->ubuf + desc->length, desc->inbuf, length); desc->length += length; - desc->reslength = length; } } skip_error:
if (desc->rerr) { /* - * Since there was an error, userspace may decide to not read - * any data after poll'ing. + * If there was a ZLP or an error, userspace may decide to not + * read any data after poll'ing. * We should respond to further attempts from the device to send * data, so that we can get unstuck. */ +skip_zlp: schedule_work(&desc->service_outs_intr); } else { set_bit(WDM_READ, &desc->flags); @@ -585,15 +589,6 @@ static ssize_t wdm_read goto retry; }
- if (!desc->reslength) { /* zero length read */ - dev_dbg(&desc->intf->dev, "zero length - clearing WDM_READ\n"); - clear_bit(WDM_READ, &desc->flags); - rv = service_outstanding_interrupt(desc); - spin_unlock_irq(&desc->iuspin); - if (rv < 0) - goto err; - goto retry; - } cntr = desc->length; spin_unlock_irq(&desc->iuspin); } @@ -1016,7 +1011,7 @@ static void service_interrupt_work(struct work_struct *work)
spin_lock_irq(&desc->iuspin); service_outstanding_interrupt(desc); - if (!desc->resp_count) { + if (!desc->resp_count && (desc->length || desc->rerr)) { set_bit(WDM_READ, &desc->flags); wake_up(&desc->wait); }