[BUG] 'make install' may overwrite existing directories
From: Alejandro Colomar <hidden>
Date: 2022-10-20 10:58:16
Attachments
- OpenPGP_signature [application/pgp-signature] 833 bytes
From: Alejandro Colomar <hidden>
Date: 2022-10-20 10:58:16
Hi, The project Makefile defines the install target as: $ sed -n '3236,3245p' Makefile install: all $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(bindir_SQ)' $(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(gitexec_instdir_SQ)' $(INSTALL) $(INSTALL_STRIP) $(PROGRAMS) '$(DESTDIR_SQ)$(gitexec_instdir_SQ)' $(INSTALL) $(SCRIPTS) '$(DESTDIR_SQ)$(gitexec_instdir_SQ)' $(INSTALL) -m 644 $(SCRIPT_LIB) '$(DESTDIR_SQ)$(gitexec_instdir_SQ)' $(INSTALL) $(INSTALL_STRIP) $(install_bindir_xprograms) '$(DESTDIR_SQ)$(bindir_SQ)' $(INSTALL) $(BINDIR_PROGRAMS_NO_X) '$(DESTDIR_SQ)$(bindir_SQ)' Notice that the directories are being installed even if they already exist. This may be problematic in case the existing directories have a different mode, or have SELinux security contexts, or any other characteristic that install(1) would overwrite. A similar bug report in an NGINX project can be found here: <https://github.com/nginx/unit/issues/769> The two solutions available are: - Make the install target be a dummy target that depends on the files being installed, and write proper make(1) rules for those targets. This would be preferred. You can check the Linux man-pages Makefile for an implementation of the 'install' target that does this. - test(1) if the directory exists before actually creating it. This is a simple workaround. Cheers, Alex -- <http://www.alejandro-colomar.es/>