All changes to fakechroot we carry in Debian for glibc >= 2.37

[josch]

I'm creating this merge request to keep track of the patch stack we carry in Debian on top of fakechroot 2.20.1.

These are mostly the same patches that are also applied by Fedora: https://src.fedoraproject.org/rpms/fakechroot

This pull request contains commits from the following pull requests:

• check return value of dladdr #70
• glibc 2.33+ compatibility #85
• Wrap fstatat and fstatat64 (glibc 2.33) #86
• src/lckpwdf.c: create an empty /etc/.pwd.lock #95
• Wrap all functions accessing /etc/passwd, /etc/group and /etc/shadow for glibc >= 2.34 #98
• Glibc2.34 #108

As well as the last useful commit that was pushed to master since the release of 2.20.1: b42d1fb

This is the current patch stack in Debian: https://sources.debian.org/src/fakechroot/unstable/debian/patches/

[sergiomb2]

First commit is already in master , can you rebase with master please?

[sergiomb2]

• Stat fix compilation #73 is also included
• Add support for arch install scripts #75 not included but just to be used in Arch Linux
• The strcpy writes the terminating null byte as well. #79 is also included
• Fix typo in AC_PATH_PROG for ldconfig #80 is not included but it is a trivial fix (IIRC)
• Fix issue #92 #93 not included
• Scripting very basic debian package creation in makedeb.sh #96 not included but just to used in Debian
• rel2abs: Only call getcwd_real for relative paths #100 not included

as note #98 for glibc 2.32 added src/__nss_files_fopen.c and later for glibc 2.34 remove it

[josch]

I added two more commits which are needed to prevent #107 under glibc 2.34.

[sergiomb2]

I added two more commits which are needed to prevent #107 under glibc 2.34.

It failed to build on "../build-aux/test-driver: line 107: ./t/rm.t: Permission denied".
I just hat modify the file permissions (mode) to 755 [1] to fix the build , everything else is fine .

Thank you.

[1]
diff --git a/test/t/rm.t b/test/t/rm.t
new file mode 100755

[josch]

Thank you @sergiomb2 I added the executable flag to the test.

I also added a new commit that fixes ldd.fakechroot on mips64el, ppc64el and s390x -- see #111 for details.

[josch]

I found a rather important problem with my patch stack. The fix is:

diff --git a/src/__lstat64_time64.c b/src/__lstat64_time64.c
index e3e8400..1637b9c 100644
--- a/src/__lstat64_time64.c
+++ b/src/__lstat64_time64.c
@@ -20,7 +20,7 @@
 
 #include <config.h>
 
-#ifdef HAVE___lstat64_time64
+#ifdef HAVE___LSTAT64_TIME64
 
 #define _ATFILE_SOURCE
 #define _POSIX_C_SOURCE 200809L

Because of that "typo" (probably some copy-paste mess-up of mine), touch --no-dereference --reference ... would fail on 32bit arches like armel, armhf and i386 because __lstat64_time64 was not wrapped by fakechroot. I added a check to test/t/touch.t to test running touch in that way to prevent future regressions.

In the process I also observed that during a normal build, test/t/touch.t is never run because test/touch.inc.sh tries to find testtree/usr/bin/touch instead of $testtree/usr/bin/touch.

I pushed all fixes here as well as to #108.

[josch]

I pushed another commit which wraps __lutimes64 which is used by glibc 2.37 and later on armel and armhf.

[josch]

This branch now has another fix by @yorickvP to fix #114

[sergiomb2]

this fix " __readlink_chk, __readlinkat_chk: properly pass buffer length " seems to me that fixes cp.t .

but now socket test fail with "af_unix server socket created bind: File name too long"

Anyway I need to reduce my workload , and fakechroot seems abandoned to me , I think it was replaced with fakeroot , I most probably I will orphan the package in Fedora .

It was a pleasure meet you

[josch]

fakechroot seems abandoned to me

I think it is.

I think it was replaced with fakeroot

No. fakeroot makes the wrapped program appear as if the user running it is actually root (uid 0). fakechroot makes the program believe it is run inside a chroot. You usually use both fakeroot and fakechroot at the same time. They work together.

I most probably I will orphan the package in Fedora .

I think that makes sense. I will continue maintaining it in Debian but it's not much fun at this point...

[sergiomb2]

OK , I'm building it for Fedora 41 without socket-af_unix.t .
we can add to this PR is one typo fix #80