Colin Percival
2005-Mar-31 16:00 UTC
MNT_NOEXEC on root filesystem with diskless PXE boot?
Tom Alsberg wrote:> Perhaps this should go to -STABLE, I just couldn't be sure.It will get more attention on freebsd-stable@, so I'm CCing that list.> We are trying out FreeBSD 5.4-PRERELEASE on diskless clients. I > noticed one problem, being that when setting the LD_LIBRARY_PATH > (or for that matter, LD_PRELOAD, and LD_LIBMAP_DISABLE) environment > variables, nothing will run, as /libexec/ld-elf.so.1 complains: > > Cannot execute objects on / > > According to the sources, this was added in 5.4, and will happen > if / is mounted noexec.Yes, that's quite correct -- although I can't imagine how a bug which caused / to be labelled as "noexec" managed to avoid causing major problems until now. I don't know anything about NFS, but hopefully someone on -stable will be able to work out what's going on from the rest of your email (quoted below). Colin Percival> In this case, / is mounted by the BTX PXE loader over NFS (from a > FreeBSD 5.3 server, right now). "mount" does not show the noexec > flag. However, with the attached little C program I verified that > statfs really returns this flag (0x00000006). > > Now, I see that on FreeBSD 5.3 diskless clients this flag is also > returned on / - just it happened that nobody looked at it until > the change in rtld.c of FreeBSD 5.4: > > if (fs.f_flags & MNT_NOEXEC) { > _rtld_error("Cannot execute objects on %s\n", fs.f_mntonname); > close(fd); > return NULL; > } > > I didn't yet understand (didn't check much) - why does statfs report > the MNT_NOEXEC flag on the / filesystem (and only the / filesystem, > when it's mounted from NFS by the bootloader - not any other > NFS filesystems)? BTW, this happens also with NetApp as the NFS > server - just to rule out any possibility of relation here. > > Ideas appreciated, > -- Tom > > > > ------------------------------------------------------------------------ > > #include <stdio.h> > #include <fcntl.h> > #include <sys/param.h> > #include <sys/mount.h> > > > int main(int argc, char *argv[]) > { > if (argc != 2) { > fprintf(stderr, "invalid number of arguments"); > return -1; > } > > struct statfs stbuf; > > if (statfs(argv[1], &stbuf) != 0) { > perror("fstatfs"); > return -1; > } > > printf("FLAGS: 0x%08X\n", stbuf.f_flags); > if (stbuf.f_flags & MNT_NOEXEC) > printf("MNT_NOEXEC\n"); > > return 0; > }
Sten Spans
2005-Aug-24 15:54 UTC
NFS - MNT_NOEXEC [WAS Re: MNT_NOEXEC on root filesystem with diskless PXE boot?]
On Thu, 31 Mar 2005, Colin Percival wrote:> Tom Alsberg wrote: >> Perhaps this should go to -STABLE, I just couldn't be sure. > > It will get more attention on freebsd-stable@, so I'm CCing that list. > >> We are trying out FreeBSD 5.4-PRERELEASE on diskless clients. I >> noticed one problem, being that when setting the LD_LIBRARY_PATH >> (or for that matter, LD_PRELOAD, and LD_LIBMAP_DISABLE) environment >> variables, nothing will run, as /libexec/ld-elf.so.1 complains: >> >> Cannot execute objects on / >> >> According to the sources, this was added in 5.4, and will happen >> if / is mounted noexec. > > Yes, that's quite correct -- although I can't imagine how a bug which > caused / to be labelled as "noexec" managed to avoid causing major > problems until now. > > I don't know anything about NFS, but hopefully someone on -stable > will be able to work out what's going on from the rest of your > email (quoted below). > > Colin Percival > >> In this case, / is mounted by the BTX PXE loader over NFS (from a >> FreeBSD 5.3 server, right now). "mount" does not show the noexec >> flag. However, with the attached little C program I verified that >> statfs really returns this flag (0x00000006). >> >> Now, I see that on FreeBSD 5.3 diskless clients this flag is also >> returned on / - just it happened that nobody looked at it until >> the change in rtld.c of FreeBSD 5.4: >> >> if (fs.f_flags & MNT_NOEXEC) { >> _rtld_error("Cannot execute objects on %s\n", fs.f_mntonname); >> close(fd); >> return NULL; >> } >> >> I didn't yet understand (didn't check much) - why does statfs report >> the MNT_NOEXEC flag on the / filesystem (and only the / filesystem, >> when it's mounted from NFS by the bootloader - not any other >> NFS filesystems)? BTW, this happens also with NetApp as the NFS >> server - just to rule out any possibility of relation here. >> >> Ideas appreciated, >> -- Tom >> >> >> >> ------------------------------------------------------------------------ >> >> #include <stdio.h> >> #include <fcntl.h> >> #include <sys/param.h> >> #include <sys/mount.h> >> >> >> int main(int argc, char *argv[]) >> { >> if (argc != 2) { >> fprintf(stderr, "invalid number of arguments"); >> return -1; >> } >> >> struct statfs stbuf; >> >> if (statfs(argv[1], &stbuf) != 0) { >> perror("fstatfs"); >> return -1; >> } >> >> printf("FLAGS: 0x%08X\n", stbuf.f_flags); >> if (stbuf.f_flags & MNT_NOEXEC) >> printf("MNT_NOEXEC\n"); >> >> return 0; >> }I've managed to reproduce this with a simple: www1# gcc -o noexec noexec.c www1# umount /buildroot/ www1# mount nfs3:/export/vol1/build/BUILDROOT-WWW /buildroot www1# ./noexec /buildroot/ exec www1# umount /buildroot/ www1# mount -o -r=32768 nfs3:/export/vol1/build/BUILDROOT-WWW /buildroot www1# ./noexec /buildroot/ noexec www1# Pretty funky shit this :). PR: 85268 http://www.freebsd.org/cgi/query-pr.cgi?pr=85268 Anybody care to take a deeper look ? -- Sten Spans "There is a crack in everything, that's how the light gets in." Leonard Cohen - Anthem