Opened 5 years ago
Last modified 5 years ago
#15095 assigned bug
if_free not called on (some?) attach failures
| Reported by: | bga | Owned by: | bga |
|---|---|---|---|
| Priority: | normal | Milestone: | Unscheduled |
| Component: | Drivers/Network | Version: | R1/Development |
| Keywords: | Cc: | ||
| Blocked By: | Blocking: | ||
| Platform: | All |
Description
If during attach() a device for some reason fails to initialize, there are cases where gDeviceCount will not be decremented (specially when using iflib) as the place where this happens currently is never called in that case. So even if the device is not successfully initialized, the driver is not unloaded and the /dev entry still hangs around.
We probably want to move the handling to a higher abstraction or simply remove it (it is not obvious to me if it is used outside the compatibility layer).
Change History (1)
comment:1 by , 5 years ago
| Component: | Network & Internet → Drivers/Network |
|---|---|
| Summary: | gDeviceCount management in the FreeBSD compatibility layer needs improvement. → if_free not called on (some?) attach failures |
Note:
See TracTickets
for help on using tickets.
The problem appears to be that if_free is not called on attach failures. iflib calls that when deregistering the device. Perhaps when attach fails, we should call if_free ourselves and not rely on the driver to do it?
Or, how do other ethernet drivers handle this? I think most of them call detach() inside attach() to tear down the internal state? Is iflib doing that?