On 8/15/21, Guenter Roeck <linux@xxxxxxxxxxxx> wrote: > On 8/14/21 5:19 PM, Stanislav German-Evtushenko wrote: >> On Saturday, August 14, 2021, Guenter Roeck <linux@xxxxxxxxxxxx >> <mailto:linux@xxxxxxxxxxxx>> wrote: >> > On Sun, Aug 15, 2021 at 04:13:45AM +0900, Stanislav German-Evtushenko >> wrote: >> >> If userspace opens the watchdog device self-feeding stops. Sometimes >> >> opening the device happens by accident, e.g. by mistakenly running >> grep >> >> recursively in a wrong directory which leads to the server being >> reset. >> >> >> >> Watchdog device does not handle read operation therefore the issue can >> be >> >> prevented by requiring the device to be opened for writing: >> >> >> >> - Prevent opening the device without FMODE_WRITE >> >> >> >> Signed-off-by: Stanislav German-Evtushenko <ginermail@xxxxxxxxx >> <mailto:ginermail@xxxxxxxxx>> >> > >> > NACK. That would be a major undocumented ABI change. Opening >> /dev/watchdog >> > for reading can be and is used today to test a watchdog. >> > >> > Guenter >> > >> >> I see. This is unfortunate. >> >> In this case I'll try to find the right place in the documentation and >> make it more explicit unless it is already there and I've overlooked. >> > No. That isn't the point. This is used and must not be changed. > Again, people (including me( are _using_ this to test watchdogs. > We are not going to disable that because some people are not > careful when executing commands as root. > > Guenter > I mean I would describe the existing behavior as this is not obvious. As for testing, opening for writing is not harder than for reading but I understand that it would be a breaking change now.
