How does /dev work on OpenBSD?

A10netsurf

I am attempting to configure OpenBSD to be more flash storage friendly. I have been reading about turning various partitions into MFSes. Some guides also mention making /dev a MFS. I do not see a /dev mount point entry in fstab, however. Presumably the device files/nodes do exist on the disk by default if MFS is not used. This raises the question: when OpenBSD or an application writes to such a device file (whether representing a virtual or hardware device) to communicate with it, is the data actually written to the disk? If so, would that not be a significant performance bottleneck and cause of wear on the disk? The special virtual device /dev/pf in particular comes to mind as a device file that potentially could be written to a lot (despite not being an actual disk). Also, if /dev is simply located on the root partition, which hypothetically is located on sd0, does that mean that if /dev/sd2 is written to all that data is also written to sd0? How do(es) /dev or device files/nodes on OpenBSD actually work by default? If writes to device files do not actually cause writes to the disk that /dev is located on, why make a /dev MFS? Or is there a misunderstanding on the part of the authors of these guides? If someone here is familiar with the internals of FreeBSD and/or NetBSD I would be interested to know it works there as well. It seems that on Linux a tmpfs is used by default for /dev and there is also devtmpfs as an option.

nia

Traditional Unix systems use a so-called static /dev. This includes NetBSD and OpenBSD (as far as I know). FreeBSD uses a dynamic devfs file system that creates nodes as the kernel prompts it.

Under the static model, /dev is populated by a script with all the potential nodes the system might need. These are "special files". They don't take up space on the underlying file system since they simply contain a pointer identifier that tells the kernel how to re-route the open(), read(), and write() operations.

You can create such a special file anywhere in the file system, not just /dev (unless the "nodev" mount option is used). This is how chroot sandboxes work. As long as the special file contains the right information, the kernel will know what to do with it.

A10netsurf

Does this re-routing mean that the IO never touches the underlying disk of /(dev) on NetBSD and OpenBSD but is immediately re-routed to an area in kernel memory space?

rvp

A10netsurf Does this re-routing mean that the IO never touches the underlying disk of /(dev) on NetBSD and OpenBSD

It works like for regular files. If the disk devices are read or written, then their access or mod. times will get updated. So, a fsck or dump/restore etc will have to change the timestamps on disk. Mounting a disk device and doing read/writes on it, however, won't change the timestamps every time you do a read or write on the FS.

You don't have to worry about disk devices, which're read maybe read/written once for FS consistency checks. No, what you should care about are the TTY devices in /dev which'll have to be updated every time you type something.

NetBSD has a nodevmtime flag which you can pass to mount for this. But, OpenBSD only has noatime. Use that after shifting /dev onto its own partition. (Using that flag on the root partition can interefere with programs that look at the access time--like mailers.)

A10netsurf

Thanks for your responses. I was able to create a working /dev MFS through combining information from some guides and trial and error some days ago, which should eliminate any metadata writes to the eMMC. Of course, I also better understand now how /dev works on *BSDs.