On Sat, 19 Mar 2011, Michael Tokarev wrote: > Hello. > > I'm trying to get rid of the ugly hack in busybox, which is > activated by CONFIG_STANDALONE_SHELL in busybox configuration. > > What it does is: when you run its shell, ash, and run a command, > such as dd or ls, and this command is provided by this busybox > binary too, busybox executes itself to run this command, instead > of trying to search it in $PATH as it's usually done. > > So effectively, all busybox commands becomes "shell builtins" > like cd and echo (which is a built-in in most shells nowadays).
yes, this is known and wanted as busybox alternatives are quite powerful, no? (at least more powerful than the smallish klibc ones) > This "standalone shell" mode helps when you're working in some > rescue environment where you want as less filesystem access as > possible, so that just one (probably statically linked) executable > is enough for everything. But such a usage case is very, well, > special. In all other cases, such behavour is unexpected and > confusing at best, and there's no good way to control which > implementation of commands will be used if both busybox and > an external binary provides the same command. > > So I'm trying to get rid of CONFIG_STANDALONE_SHELL in busybox, > and initramfs is the only user of this feature currently. I'm not conviced by the potential gain yet? > The plan is to introduce (sym)links in initramfs pointing to > busybox for all applets it provides, instead of just one > executable which when executes itself. > > The patchset that follows is a bunch of very small changes for > initramfs hooks that does the following: > > [001] don't move klibc's sh.shared to sh, link instead > this is a tiny "optional" change, for consistency, so > to say: when using symlinks it becomes more obvious which > implementation is being used. Moreover, by not moving the > original sh.shared we keep it even in case we'll use > something else in the future. > > [002] don't warn about md-root need busybox: it doesn't anymore > unrelated cleanup patch but it is in the same place I'll > touch later: we believed mdadm needs busybox in initramfs, > but it has been fixed long ago > > [003] don't copy busybox to sh, use proper name and symlink > > [004] rename hooks/busybox to hooks/zz-busybox to reorder it to be last hook > this is in order to ensure that busybox hook will be run last, > in order to "fill the gaps", -- to create links to busybox only > for those commands which don't already exist in initramfs. > > [005] create links in initramfs to busybox with other names if not already > exist > the final thing > > The whole thing is also available in a git repository, > git://git.corpit.ru/initramfs-tools.git in create-links-to-busybox > branch. > > This series is based on maks/mkinitramfs_cp branch of initramfs > git repository on alioth, since it requires commit 11e9453a29cbc1 > "mkinitramfs: copy over on build instead of using symlink tree", > because it uses symlinks heavily. > > This is just one possible approach. Another approach will be > to move whole hooks/busybox into busybox package, just like > hooks/klibc actually, -- I'm not sure what kernel team prefers. did newer busybox gain a switchroot utility? thanks for your work and sorry if I misunderstood the purpose. -- maks -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/[email protected]
