On 11/16/18 10:06 AM, Daniel Polak wrote: >>> The main benefit of read-only is not having to do disk checks but the time >>> for >>> root is negligible. >> well, it's not just time fsck'ing, those checks can fail, and then if >> you don't have OOB you have to go visit the machine ..
True, but the chance on root are next to 0 with a default layout. If you get an outage during installation then ro root didn't help. That also only matters for the most recently written files. An Fsck option upon failure of marking a second partition or perhaps /altroot_fsck as where to boot a kernel from may be an interesting idea, if possible? To ensure ssh is always available. I dislike OOB chips in general but don't have great experience. OTOH, I have been using ro root for so long, perhaps you have better experience? The alternate boot functionality is something I may get involved in as we could make our own OOB chips but I even dislike the Intel support of file/mem access?? > Agree entirely and that is why on our firewalls I always use a RO root > filesystem and whatever needs to be RW is in MFS. > It does require minor modification of among others /etc/rc though. > > Root of the problem is the relative fragility of the current file system > but changing that is of course a major undertaking. > > Daniel I disagree. Maybe with softdep but I have never had a major problem. Sure you may need -y and lose the most recent files but then you could always mount sync? I have had much worse (unexplainable corruption) on Linux with less uptime. ext4 was annoying as well because testdisk couldn't recover deleted files.
