On Tue, Sep 03, 2019 at 09:50:09PM -0700, Deepa Dinamani wrote: > If we don't care to warn about the timestamps that are clamped in > memory, maybe we could just warn when they are being written out. > Would something like this be more acceptable? I would also remove the > warning in ext4.h. I think we don't have to check if the inode is 128 > bytes here (Please correct me if I am wrong). If this looks ok, I can > post this. That's better, but it's going to be misleading in many cases. The inode's extra size field is 16 or larger, there will be enough space for the timestamps, so talking about "timestamps on this inode beyond 2038" when ext4 is unable to expand it from say, 24 to 32, won't be true. Certain certain features won't be available, yes --- such as project-id-based quotas, since there won't be room to store the project ID. However, it's not going to impact the ability to store timestamps beyond 2038. The i_extra_isize field is not just about timestamps! Again, the likelihood that there will be file systems that have this problem in 2038 is... extremely low in my judgement. Storage media just doesn't last that long; and distributions such as Red Hat and SuSE very strongly encourage people to reformat file systems and do *not* support upgrades from ext3 to ext4 by using tune2fs. If you do this, their help desk will laugh at you and refuse to help you. Companies like Google will do this kind of upgrades[1], sure. But that's because backing up and reformatting vast numbers of file systems are not practical at scale. (And even Google doesn't maintain the file system image when the servers are old enough to be TCO negative and it's time to replace them.) In contrast, most companies / users don't do this sort of thing at all. It's not an issue for Cell Phones, for example, or most consumer devices, which are lucky if the last more than 3 years before they get desupported and stop getting security updates, and then the lithium ion batttery dies and the device end up in a landfill. Those that might live 20 years (although good luck with that for something like, say, a smart thermostat) aren't going to have a console and no one will be paying attention to the kernel messages anyway. So is it really worth it? For whom are these messages meant? [1] https://www.youtube.com/watch?v=Wp5Ehw7ByuU Cheers, - Ted
