On Thu, Feb 7, 2019 at 11:22 AM Brian Ledbetter <brian.ledbet...@gmail.com> wrote:
> > > On Thursday, February 7, 2019 at 3:48:32 AM UTC-5, Nadav Har'El wrote: > >> To support the case where the /etc filesystem is read-only, we can do the >> following: In the read-only filesystem, put in /etc/resolv.conf a symbolic >> link to /tmp/resolv.conf, assuming we always mount a writable /tmp. Then, >> when dhcp.cc will write to "/etc/resolv.conf" (overwritten, e.g., open() >> with O_TRUNC), it will actually write to that file on /tmp and succeed. >> > > Ha - That's quite a bit easier than the path I was thinking of, trying to > write some sort of mmap'ed VFS hack to point /etc/resolv.conf to a blob of > kernel memory. This is definitely a better approach. > > Are we guaranteed that every OSv instance will have a writable /tmp > directory? Is that a part of the core manifest? > This is a question for Waldek, who's the expert (and author) of the readonly-filesystem stuff. But I seem to remember it is - that we always mount a writeable, in-memory, file system, because many applications assume they can write stuff there. If we don't do that yet, maybe we should? You can test this by building an image with "fs=rofs". For example, scripts/build fs=rofs image=rogue -- You received this message because you are subscribed to the Google Groups "OSv Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to osv-dev+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
