[Bug 1646822] [NEW] Handling $ORIGIN strings within DT_NEEDED sections
Oleksandr Pikozh
o.pikozh at gmail.com
Fri Dec 2 12:58:12 UTC 2016
Public bug reported:
What documentation says:
"$ORIGIN sequences within a DT_NEEDED entry or path passed as a parameter to dlopen() are treated as errors."
(From "System V Application Binary Interface - DRAFT - 10 June 2013" / "Chapter 5 - Program Loading and Dynamic Linking" / "Dynamic Linking" / "Shared Object Dependencies" / "Substitution Sequences" <http://www.sco.com/developers/gabi/latest/ch5.dynamic.html#substitution>.)
What in reality happens:
- Having $ORIGIN string within DT_NEEDED section with shared library that DOES NOT uses versioning, causes '$ORIGIN' to be interpreted as path-to-directory-where-binary-is-located (despite, per documentation, it should/must be treated as error).
- Having $ORIGIN string within DT_NEEDED section with shared library that USES versioning, causes assertion failure (see below) or segmentation fault (not on my system) (however, it doesn't look like intentional/graceful message related to explicitly documented case of ORIGIN within DT_NEEDED — it more looks like that ld.so was put into unexpected state).
On my system, having $ORIGIN within DT_NEEDED together with versioning
causes the following output: "Inconsistency detected by ld.so: dl-
version.c: 224: _dl_check_map_versions: Assertion `needed != NULL'
failed!". Some other people say they observe segmentation fault instead.
I used the attached script test-origin-in-needed.sh to test behavior.
I'd better report the bug to libc maintainers (upstream). Because it
seems to be upstream-related specification violation. But the libc bug
reporting policy says: "Distributions may include their own
modifications to glibc in the binaries and sources you get with the
operating system. If the glibc you are using comes from a complete
operating system distribution, you should report bugs to that
distribution project first." So, I report here first.
1) lsb_release -rd
Description: Ubuntu 16.04.1 LTS
Release: 16.04
2) pt-cache policy libc6
libc6:
Installed: 2.23-0ubuntu4
Candidate: 2.23-0ubuntu4
Version table:
*** 2.23-0ubuntu4 500
500 http://ua.archive.ubuntu.com/ubuntu xenial-updates/main amd64 Packages
100 /var/lib/dpkg/status
2.23-0ubuntu3 500
500 http://ua.archive.ubuntu.com/ubuntu xenial/main amd64 Packages
3) I expected handling of $ORIGIN strings within DT_NEEDED sections to
be conforming "System V Application Binary Interface". Or, at least, to
see non-conforming parts explicitly documented somewhere in GNU
documentation.
4) See above. Having $ORIGIN string within DT_NEEDED section without
versioning causes $ORIGIN to work as if it's within (for example) RPATH
section — violates documentation. Having $ORIGIN string within DT_NEEDED
section with versioning causes assertion failure (formally doesn't
violate documentation, but really it looks more like unexpected state
than graceful condition handling). I used attached script to test.
** Affects: glibc (Ubuntu)
Importance: Undecided
Status: New
** Attachment added: "test-origin-in-needed.sh"
https://bugs.launchpad.net/bugs/1646822/+attachment/4786412/+files/test-origin-in-needed.sh
--
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to glibc in Ubuntu.
https://bugs.launchpad.net/bugs/1646822
Title:
Handling $ORIGIN strings within DT_NEEDED sections
Status in glibc package in Ubuntu:
New
Bug description:
What documentation says:
"$ORIGIN sequences within a DT_NEEDED entry or path passed as a parameter to dlopen() are treated as errors."
(From "System V Application Binary Interface - DRAFT - 10 June 2013" / "Chapter 5 - Program Loading and Dynamic Linking" / "Dynamic Linking" / "Shared Object Dependencies" / "Substitution Sequences" <http://www.sco.com/developers/gabi/latest/ch5.dynamic.html#substitution>.)
What in reality happens:
- Having $ORIGIN string within DT_NEEDED section with shared library that DOES NOT uses versioning, causes '$ORIGIN' to be interpreted as path-to-directory-where-binary-is-located (despite, per documentation, it should/must be treated as error).
- Having $ORIGIN string within DT_NEEDED section with shared library that USES versioning, causes assertion failure (see below) or segmentation fault (not on my system) (however, it doesn't look like intentional/graceful message related to explicitly documented case of ORIGIN within DT_NEEDED — it more looks like that ld.so was put into unexpected state).
On my system, having $ORIGIN within DT_NEEDED together with versioning
causes the following output: "Inconsistency detected by ld.so: dl-
version.c: 224: _dl_check_map_versions: Assertion `needed != NULL'
failed!". Some other people say they observe segmentation fault
instead. I used the attached script test-origin-in-needed.sh to test
behavior.
I'd better report the bug to libc maintainers (upstream). Because it
seems to be upstream-related specification violation. But the libc bug
reporting policy says: "Distributions may include their own
modifications to glibc in the binaries and sources you get with the
operating system. If the glibc you are using comes from a complete
operating system distribution, you should report bugs to that
distribution project first." So, I report here first.
1) lsb_release -rd
Description: Ubuntu 16.04.1 LTS
Release: 16.04
2) pt-cache policy libc6
libc6:
Installed: 2.23-0ubuntu4
Candidate: 2.23-0ubuntu4
Version table:
*** 2.23-0ubuntu4 500
500 http://ua.archive.ubuntu.com/ubuntu xenial-updates/main amd64 Packages
100 /var/lib/dpkg/status
2.23-0ubuntu3 500
500 http://ua.archive.ubuntu.com/ubuntu xenial/main amd64 Packages
3) I expected handling of $ORIGIN strings within DT_NEEDED sections to
be conforming "System V Application Binary Interface". Or, at least,
to see non-conforming parts explicitly documented somewhere in GNU
documentation.
4) See above. Having $ORIGIN string within DT_NEEDED section without
versioning causes $ORIGIN to work as if it's within (for example)
RPATH section — violates documentation. Having $ORIGIN string within
DT_NEEDED section with versioning causes assertion failure (formally
doesn't violate documentation, but really it looks more like
unexpected state than graceful condition handling). I used attached
script to test.
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/1646822/+subscriptions
More information about the foundations-bugs
mailing list