| Message ID | ZSpHPCaQ5DDA9Ysl@Sun |
|---|---|
| State | New |
| Headers | show |
| Series | [v2,-next] usb/usbip: fix wrong data added to platform device | expand |
On Sat, Oct 14, 2023 at 03:46:04PM +0800, Hongren Zheng wrote: > .data of platform_device_info will be copied into .platform_data of > struct device via platform_device_add_data. platform_device_add_data() > However, vhcis[i] contains a spinlock, is dynamically allocated and > used by other code, so it is not meant to be copied. The workaround > was to use void *vhci as an agent, but it was removed in the commit > suggested below. > > This patch adds back the workaround and changes the way of using > platform_data accordingly. Good learning to me, thank you for the fix! Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
On Sat, Oct 14, 2023 at 03:46:04PM +0800, Hongren Zheng wrote: > .data of platform_device_info will be copied into .platform_data of > struct device via platform_device_add_data. That is crazy. > However, vhcis[i] contains a spinlock, is dynamically allocated and > used by other code, so it is not meant to be copied. The workaround > was to use void *vhci as an agent, but it was removed in the commit > suggested below. Again, crazy, I'm amazed it works at all. > This patch adds back the workaround and changes the way of using > platform_data accordingly. > > Reported-by: syzbot+e0dbc33630a092ccf033@syzkaller.appspotmail.com > Closes: https://lore.kernel.org/r/00000000000029242706077f3145@google.com/ > Reported-by: syzbot+6867a9777f4b8dc4e256@syzkaller.appspotmail.com > Closes: https://lore.kernel.org/r/0000000000007634c1060793197c@google.com/ > Fixes: b8aaf639b403 ("usbip: Use platform_device_register_full()") > Tested-by: syzbot+6867a9777f4b8dc4e256@syzkaller.appspotmail.com > Link: https://lore.kernel.org/r/0000000000007ac87d0607979b6b@google.com/ > Signed-off-by: Hongren Zheng <i@zenithal.me> > --- > drivers/usb/usbip/vhci_hcd.c | 15 ++++++++------- > 1 file changed, 8 insertions(+), 7 deletions(-) I'll take this now, as it fixes a reported issue, but really, the root problem should be fixed instead. This is not how to treat platform data at all. thanks, greg k-h
diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c index f845b91848b9..82650c11e451 100644 --- a/drivers/usb/usbip/vhci_hcd.c +++ b/drivers/usb/usbip/vhci_hcd.c @@ -1139,7 +1139,7 @@ static int hcd_name_to_id(const char *name) static int vhci_setup(struct usb_hcd *hcd) { - struct vhci *vhci = dev_get_platdata(hcd->self.controller); + struct vhci *vhci = *((void **)dev_get_platdata(hcd->self.controller)); if (usb_hcd_is_primary_hcd(hcd)) { vhci->vhci_hcd_hs = hcd_to_vhci_hcd(hcd); @@ -1257,7 +1257,7 @@ static int vhci_get_frame_number(struct usb_hcd *hcd) /* FIXME: suspend/resume */ static int vhci_bus_suspend(struct usb_hcd *hcd) { - struct vhci *vhci = dev_get_platdata(hcd->self.controller); + struct vhci *vhci = *((void **)dev_get_platdata(hcd->self.controller)); unsigned long flags; dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__); @@ -1271,7 +1271,7 @@ static int vhci_bus_suspend(struct usb_hcd *hcd) static int vhci_bus_resume(struct usb_hcd *hcd) { - struct vhci *vhci = dev_get_platdata(hcd->self.controller); + struct vhci *vhci = *((void **)dev_get_platdata(hcd->self.controller)); int rc = 0; unsigned long flags; @@ -1338,7 +1338,7 @@ static const struct hc_driver vhci_hc_driver = { static int vhci_hcd_probe(struct platform_device *pdev) { - struct vhci *vhci = dev_get_platdata(&pdev->dev); + struct vhci *vhci = *((void **)dev_get_platdata(&pdev->dev)); struct usb_hcd *hcd_hs; struct usb_hcd *hcd_ss; int ret; @@ -1396,7 +1396,7 @@ static int vhci_hcd_probe(struct platform_device *pdev) static void vhci_hcd_remove(struct platform_device *pdev) { - struct vhci *vhci = dev_get_platdata(&pdev->dev); + struct vhci *vhci = *((void **)dev_get_platdata(&pdev->dev)); /* * Disconnects the root hub, @@ -1431,7 +1431,7 @@ static int vhci_hcd_suspend(struct platform_device *pdev, pm_message_t state) if (!hcd) return 0; - vhci = dev_get_platdata(hcd->self.controller); + vhci = *((void **)dev_get_platdata(hcd->self.controller)); spin_lock_irqsave(&vhci->lock, flags); @@ -1522,10 +1522,11 @@ static int __init vhci_hcd_init(void) goto err_driver_register; for (i = 0; i < vhci_num_controllers; i++) { + void *vhci = &vhcis[i]; struct platform_device_info pdevinfo = { .name = driver_name, .id = i, - .data = &vhcis[i], + .data = &vhci, .size_data = sizeof(void *), };
.data of platform_device_info will be copied into .platform_data of struct device via platform_device_add_data. However, vhcis[i] contains a spinlock, is dynamically allocated and used by other code, so it is not meant to be copied. The workaround was to use void *vhci as an agent, but it was removed in the commit suggested below. This patch adds back the workaround and changes the way of using platform_data accordingly. Reported-by: syzbot+e0dbc33630a092ccf033@syzkaller.appspotmail.com Closes: https://lore.kernel.org/r/00000000000029242706077f3145@google.com/ Reported-by: syzbot+6867a9777f4b8dc4e256@syzkaller.appspotmail.com Closes: https://lore.kernel.org/r/0000000000007634c1060793197c@google.com/ Fixes: b8aaf639b403 ("usbip: Use platform_device_register_full()") Tested-by: syzbot+6867a9777f4b8dc4e256@syzkaller.appspotmail.com Link: https://lore.kernel.org/r/0000000000007ac87d0607979b6b@google.com/ Signed-off-by: Hongren Zheng <i@zenithal.me> --- drivers/usb/usbip/vhci_hcd.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) v2: code style change suggested by Andy Shevchenko