Date: Mon, 17 Sep 2012 15:33:57 -0600 From: Warner Losh <imp@bsdimp.com> To: Brooks Davis <brooks@one-eyed-alien.net> Cc: arch@freebsd.org Subject: Re: Importing NetBSD's mtree (and install) Message-ID: <F952C1AF-BE7A-4422-9997-11B3D41795F7@bsdimp.com> In-Reply-To: <20120917201919.GA43626@lor.one-eyed-alien.net>
index | next in thread | previous in thread | raw e-mail
On Sep 17, 2012, at 2:19 PM, Brooks Davis wrote: > As part of an effort to improve our ability to create full system images > without root access I would like to import NetBSD's version of mtree. > By doing so we would gain the -C option which produces mtree files > compatible with libarchive and makefs (one line per file with full > path) and the -N option which allows a stand alone set of passwd and > group files. > > When mated with the -U and -M options to NetBSD's install we will have > most[0] of the pieces require to allow installworld to run as a user and > then build images containing proper permissions. The rest of this post > will focus on my plans for mtree since it is the logical next step. > > NetBSD's mtree is missing a few features present in our mtree. Most of > them looks simple to implement and I plan to do so before any import. > The only exception is -i which implements indenting of old-style mtree > files to match the format of the files in /etc/mtree/. It will either > need another name or to be dropped. I honestly don't see the point in > it so I'm not sure if it's worth keeping if people will need to alter > their scripts regardless, but I'm willing to be convinced otherwise. > > Importing mtree requires importing or implementing some enhancements to > libc. First, the strsvis() function is required which is most > easily handled by importing NetBSD's vis/unvis implementations with the > addition of the VIS_GLOB set of characters. Compatibly wrappers will > be required for existing vis(3) functions and for unvis() due to ABI > changes, but they will be minimal. > > Second, pwcache_userdb(), uid_from_user(), pwcache_groupdb(), and > gid_from_group() provide useful enhancements to the uid_from_user() > and group_from_gid(). They appear to be entirely compatible with our > current implementation so simply importing the enhanced version seems > like the best course. > > Finally, FreeBSD and Mac OS have the functions fflagstostr() and > strtofflags() in libc. NetBSD has flags_to_string() and > string_to_flags() in libutil. The latter could be a very thin wrapper > around fflagstostr() and the former is exactly strtofflags(). I think > the best course is probably to provide compatible wrappers for mtree's > internal use, but I could be convinced to make them more globally > available. > > Does anything about this plan seem seriously objectionable? Looks good to me. I started doing this a while ago an ran into lots of problems, many of which you've outlined here. They weren't hard problems, just numerous enough for me to lose interest in the project before I completed it. Glad to see somebody has pushed through. > I've written some of this up along with a list of missing features in > the wiki. I'll keep progress up to date there: > > http://wiki.freebsd.org/NetBSDMtree > > -- Brooks > > [0] We also lack a tool to build disk images including partition tables, > but Marcel is looking into this. makefs will build the disk image w/o the MBR/GPT tables. For most flash devices, this is sufficient. For more complex situations, like where the boot loader groks FAT but not UFS, that needs some help... Once upon a time, you could use dd + fdisk to create an MBR image, but nothing apart from the kernel understands using it. gpart is so darn easy to use, it is a shame that you have to make a privileged trip to the kernel to use it. Warnerhelp
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?F952C1AF-BE7A-4422-9997-11B3D41795F7>