Package: grep Version: 2.27-1 Severity: serious Tags: patch fixed-upstream Control: forwarded -1 https://debbugs.gnu.org/cgi/bugreport.cgi?bug=25283
This is a bug that was reported to debian-testing. Severity is based on the fact that grep is a tool where a wrong return value has the potential to cause much (and hard to detect) harm in scripts. The patch is in the upstream bug. Original report copied from Lars Bro https://lists.debian.org/debian-testing/2017/01/msg00000.html follows: I am using Debian Stretch, obtained with debootstrap (as of Jan 02 2017). Linux kernel of the host is 4.4.6, and the following is run in a chroot. File system is ext4. When I try to run "./configure" of one of the applications I use, it fails, claiming that gmake is not present. When looking into the "configure script" I find this construct: /bin/sh -c 'make --version | grep GNU' >/dev/null 2>&1 And when I test it, I see this: root@mini:/# /bin/sh -c 'make --version | grep GNU' >/dev/null 2>&1 root@mini:/# echo $? 2 root@mini:/# /bin/sh -c 'make --version | grep GNU' >/dev/null 2>/dev/null root@mini:/# echo $? 0 root@mini:/# There is apparently something that goes wrong when "dup(2)" is used as in 2>&1 I did an strace of the two constructs: This is the failing: "/bin/sh -c 'make --version | gre"..., 8192) = 55 open("/dev/null", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3 fcntl(1, F_DUPFD, 10) = 11 close(1) = 0 fcntl(11, F_SETFD, FD_CLOEXEC) = 0 dup2(3, 1) = 1 close(3) = 0 fcntl(2, F_DUPFD, 10) = 12 close(2) = 0 fcntl(12, F_SETFD, FD_CLOEXEC) = 0 dup2(1, 2) [pid 11678] read(0, "GNU Make 4.1\nBuilt for x86_64-pc"..., 32768) = 297 [pid 11678] splice(0, NULL, 1, NULL, 32768, SPLICE_F_MOVE) = -1 EINVAL (Invalid argument) And this one works read(0, /bin/sh -c 'make --version | grep GNU' > /dev/null 2>/dev/null " /bin/sh -c 'make --version | gr"..., 8192) = 64 open("/dev/null", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3 fcntl(1, F_DUPFD, 10) = 11 close(1) = 0 fcntl(11, F_SETFD, FD_CLOEXEC) = 0 dup2(3, 1) = 1 close(3) = 0 open("/dev/null", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3 fcntl(2, F_DUPFD, 10) = 12 close(2) = 0 fcntl(12, F_SETFD, FD_CLOEXEC) = 0 dup2(3, 2) = 2 close(3) [pid 11689] read(0, "GNU Make 4.1\nBuilt for x86_64-pc"..., 32768) = 297 [pid 11689] splice(0, NULL, 1, NULL, 32768, SPLICE_F_MOVE) = 0 This suggests that there is something with the "splice(2)" system call. I tried with grep_2.25-6 wich was in Stretch before. It works as below: [pid 11821] <... read resumed> "GNU Make 4.1\nBuilt for x86_64- pc"..., 32768) = 297 ... [pid 11821] read(0, "", 32768) = 0 [pid 11821] write(1, "GNU Make 4.1\nLicense GPLv3+: GNU"..., 91) = 91 It seems to work because it uses read(0,...) and write(1,...) instead of splice().
