| Message ID | 20190906152455.22757-1-mic@digikod.net (mailing list archive) |
|---|---|
| Headers | show |
| Series | Add support for O_MAYEXEC | expand |
On Friday, September 6, 2019 11:24:50 AM EDT Mickaël Salaün wrote: > The goal of this patch series is to control script interpretation. A > new O_MAYEXEC flag used by sys_open() is added to enable userspace > script interpreter to delegate to the kernel (and thus the system > security policy) the permission to interpret/execute scripts or other > files containing what can be seen as commands. The problem is that this is only a gentleman's handshake. If I don't tell the kernel that what I'm opening is tantamount to executing it, then the security feature is never invoked. It is simple to strip the flags off of any system call without needing privileges. For example: #define _GNU_SOURCE #include <link.h> #include <fcntl.h> #include <string.h> unsigned int la_version(unsigned int version) { return version; } unsigned int la_objopen(struct link_map *map, Lmid_t lmid, uintptr_t *cookie) { return LA_FLG_BINDTO | LA_FLG_BINDFROM; } typedef int (*openat_t) (int dirfd, const char *pathname, int flags, mode_t mode); static openat_t real_openat = 0L; int my_openat(int dirfd, const char *pathname, int flags, mode_t mode) { flags &= ~O_CLOEXEC; return real_openat(dirfd, pathname, flags, mode); } uintptr_t la_symbind64(Elf64_Sym *sym, unsigned int ndx, uintptr_t *refcook, uintptr_t *defcook, unsigned int *flags, const char *symname) { if (real_openat == 0L && strcmp(symname, "openat") == 0) { real_openat = (openat_t) sym->st_value; return (uintptr_t) my_openat; } return sym->st_value; } gcc -c -g -Wno-unused-parameter -W -Wall -Wundef -O2 -Wp,-D_GLIBCXX_ASSERTIONS -fexceptions -fPIC test.c gcc -o strip-flags.so.0 -shared -Wl,-soname,strip-flags.so.0 -ldl test.o Now, let's make a test program: #include <stdio.h> #include <dirent.h> #include <fcntl.h> #include <unistd.h> int main(void) { int dir_fd, fd; DIR *d = opendir("/etc"); dir_fd = dirfd(d); fd = openat(dir_fd, "passwd", O_RDONLY|O_CLOEXEC); close (fd); closedir(d); return 0; } gcc -g -W -Wall -Wundef test.c -o test OK, let's see what happens. $ strace ./test 2>&1 | grep passwd openat(3, "passwd", O_RDONLY|O_CLOEXEC) = 4 Now with LD_AUDIT $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test 2>&1 | grep passwd openat(3, "passwd", O_RDONLY) = 4 No O_CLOEXEC flag. -Steve
* Steve Grubb: > Now with LD_AUDIT > $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test 2>&1 | grep passwd > openat(3, "passwd", O_RDONLY) = 4 > > No O_CLOEXEC flag. I think you need to explain in detail why you consider this a problem. With LD_PRELOAD and LD_AUDIT, you can already do anything, including scanning other loaded objects for a system call instruction and jumping to that (in case a security module in the kernel performs a PC check to confer additional privileges). Thanks, Florian
On Friday, September 6, 2019 2:57:00 PM EDT Florian Weimer wrote: > * Steve Grubb: > > Now with LD_AUDIT > > $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test > > 2>&1 | grep passwd openat(3, "passwd", O_RDONLY) = 4 > > > > No O_CLOEXEC flag. > > I think you need to explain in detail why you consider this a problem. Because you can strip the O_MAYEXEC flag from being passed into the kernel. Once you do that, you defeat the security mechanism because it never gets invoked. The issue is that the only thing that knows _why_ something is being opened is user space. With this mechanism, you can attempt to pass this reason to the kernel so that it may see if policy permits this. But you can just remove the flag. -Steve > With LD_PRELOAD and LD_AUDIT, you can already do anything, including > scanning other loaded objects for a system call instruction and jumping > to that (in case a security module in the kernel performs a PC check to > confer additional privileges). > > Thanks, > Florian
> On Sep 6, 2019, at 12:07 PM, Steve Grubb <sgrubb@redhat.com> wrote: > >> On Friday, September 6, 2019 2:57:00 PM EDT Florian Weimer wrote: >> * Steve Grubb: >>> Now with LD_AUDIT >>> $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test >>> 2>&1 | grep passwd openat(3, "passwd", O_RDONLY) = 4 >>> >>> No O_CLOEXEC flag. >> >> I think you need to explain in detail why you consider this a problem. > > Because you can strip the O_MAYEXEC flag from being passed into the kernel. > Once you do that, you defeat the security mechanism because it never gets > invoked. The issue is that the only thing that knows _why_ something is being > opened is user space. With this mechanism, you can attempt to pass this > reason to the kernel so that it may see if policy permits this. But you can > just remove the flag. I’m with Florian here. Once you are executing code in a process, you could just emulate some other unapproved code. This series is not intended to provide the kind of absolute protection you’re imagining. What the kernel *could* do is prevent mmapping a non-FMODE_EXEC file with PROT_EXEC, which would indeed have a real effect (in an iOS-like world, for example) but would break many, many things.
On 2019-09-06, Andy Lutomirski <luto@amacapital.net> wrote: > > On Sep 6, 2019, at 12:07 PM, Steve Grubb <sgrubb@redhat.com> wrote: > > > >> On Friday, September 6, 2019 2:57:00 PM EDT Florian Weimer wrote: > >> * Steve Grubb: > >>> Now with LD_AUDIT > >>> $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test > >>> 2>&1 | grep passwd openat(3, "passwd", O_RDONLY) = 4 > >>> > >>> No O_CLOEXEC flag. > >> > >> I think you need to explain in detail why you consider this a problem. > > > > Because you can strip the O_MAYEXEC flag from being passed into the kernel. > > Once you do that, you defeat the security mechanism because it never gets > > invoked. The issue is that the only thing that knows _why_ something is being > > opened is user space. With this mechanism, you can attempt to pass this > > reason to the kernel so that it may see if policy permits this. But you can > > just remove the flag. > > I’m with Florian here. Once you are executing code in a process, you > could just emulate some other unapproved code. This series is not > intended to provide the kind of absolute protection you’re imagining. I also agree, though I think that there is a separate argument to be made that there are two possible problems with O_MAYEXEC (which might not be really big concerns): * It's very footgun-prone if you didn't call O_MAYEXEC yourself and you pass the descriptor elsewhere. You need to check f_flags to see if it contains O_MAYEXEC. Maybe there is an argument to be made that passing O_MAYEXECs around isn't a valid use-case, but in that case there should be some warnings about that. * There's effectively a TOCTOU flaw (even if you are sure O_MAYEXEC is in f_flags) -- if the filesystem becomes re-mounted noexec (or the file has a-x permissions) after you've done the check you won't get hit with an error when you go to use the file descriptor later. To fix both you'd need to do what you mention later: > What the kernel *could* do is prevent mmapping a non-FMODE_EXEC file > with PROT_EXEC, which would indeed have a real effect (in an iOS-like > world, for example) but would break many, many things. And I think this would be useful (with the two possible ways of executing .text split into FMODE_EXEC and FMODE_MAP_EXEC, as mentioned in a sister subthread), but would have to be opt-in for the obvious reason you outlined. However, we could make it the default for openat2(2) -- assuming we can agree on what the semantics of a theoretical FMODE_EXEC should be. And of course we'd need to do FMODE_UPGRADE_EXEC (which would need to also permit fexecve(2) though probably not PROT_EXEC -- I don't think you can mmap() an O_PATH descriptor).
On Fri, 6 Sep 2019, Mickaël Salaün wrote: > Furthermore, the security policy can also be delegated to an LSM, either > a MAC system or an integrity system. For instance, the new kernel > MAY_OPENEXEC flag closes a major IMA measurement/appraisal interpreter > integrity gap by bringing the ability to check the use of scripts [2]. To clarify, scripts are already covered by IMA if they're executed directly, and the gap is when invoking a script as a parameter to the interpreter (and for any sourced files). In that case only the interpreter is measured/appraised, unless there's a rule also covering the script file(s). See: https://en.opensuse.org/SDB:Ima_evm#script-behaviour In theory you could probably also close the gap by modifying the interpreters to check for the execute bit on any file opened for interpretation (as earlier suggested by Steve Grubb), and then you could have IMA measure/appraise all files with +x. I suspect this could get messy in terms of unwanted files being included, and the MAY_OPENEXEC flag has cleaner semantics.
On 07/09/2019 00:44, Aleksa Sarai wrote: > On 2019-09-06, Andy Lutomirski <luto@amacapital.net> wrote: >>> On Sep 6, 2019, at 12:07 PM, Steve Grubb <sgrubb@redhat.com> wrote: >>> >>>> On Friday, September 6, 2019 2:57:00 PM EDT Florian Weimer wrote: >>>> * Steve Grubb: >>>>> Now with LD_AUDIT >>>>> $ LD_AUDIT=/home/sgrubb/test/openflags/strip-flags.so.0 strace ./test >>>>> 2>&1 | grep passwd openat(3, "passwd", O_RDONLY) = 4 >>>>> >>>>> No O_CLOEXEC flag. >>>> >>>> I think you need to explain in detail why you consider this a problem. Right, LD_PRELOAD and such things are definitely not part of the threat model for O_MAYEXEC, on purpose, because this must be addressed with other security mechanism (e.g. correct file system access-control, IMA policy, SELinux or other LSM security policies). This is a requirement for O_MAYEXEC to be useful. An interpreter is just a flexible program which is generic and doesn't have other purpose other than behaving accordingly to external rules (i.e. scripts). If you don't trust your interpreter, it should not be executable in the first place. O_MAYEXEC enables to restrict the use of (some) interpreters accordingly to a *global* system security policy. >>> >>> Because you can strip the O_MAYEXEC flag from being passed into the kernel. >>> Once you do that, you defeat the security mechanism because it never gets >>> invoked. The issue is that the only thing that knows _why_ something is being >>> opened is user space. With this mechanism, you can attempt to pass this >>> reason to the kernel so that it may see if policy permits this. But you can >>> just remove the flag. >> >> I’m with Florian here. Once you are executing code in a process, you >> could just emulate some other unapproved code. This series is not >> intended to provide the kind of absolute protection you’re imagining. > > I also agree, though I think that there is a separate argument to be > made that there are two possible problems with O_MAYEXEC (which might > not be really big concerns): > > * It's very footgun-prone if you didn't call O_MAYEXEC yourself and > you pass the descriptor elsewhere. You need to check f_flags to see > if it contains O_MAYEXEC. Maybe there is an argument to be made that > passing O_MAYEXECs around isn't a valid use-case, but in that case > there should be some warnings about that. That could be an issue if you don't trust your system, especially if the mount points (and the "noexec" option) can be changed by untrusted users. As I said above, there is a requirement for basic security properties as a meaningful file system access control, and obviously not letting any user change mount points (which can lead to much sever security issues anyway). If a process A pass a FD to an interpreter B, then the interpreter B must trust the process A. Moreover, being able to tell if the FD was open with O_MAYEXEC and relying on it may create a wrong feeling of security. As I said in a previous email, being able to probe for O_MAYEXEC does not make sense because it would not be enough to know the system policy (either this flag is enforced or not, for mount points, based on xattr, time…). The main goal of O_MAYEXEC is to ask the kernel, on a trusted link (hence without LD_PRELOAD-like interfering), for a file which is allowed to be interpreted/executed by this interpreter. To be able to correctly handle the case you pointed out (FD passing), either an existing or a new LSM should handle this behavior according to the origin of the FD and the chain of processes getting it. Some advanced LSM rules could tie interpreters with scripts dedicated to them, and have different behavior for the same scripts but with different interpreters. > > * There's effectively a TOCTOU flaw (even if you are sure O_MAYEXEC is > in f_flags) -- if the filesystem becomes re-mounted noexec (or the > file has a-x permissions) after you've done the check you won't get > hit with an error when you go to use the file descriptor later. Again, the threat model needs to be appropriate to make O_MAYEXEC useful. The security policies of the system need to be seen as a whole, and updated as such. As for most file system access control on Linux, it may be possible to have TOCTOU, but the whole system should be designed to protect against that. For example, changing file access control (e.g. mount point options) without a reboot may lead to inconsistent security properties, which is why such thing are discouraged by some access control systems (e.g. SELinux). > > To fix both you'd need to do what you mention later: > >> What the kernel *could* do is prevent mmapping a non-FMODE_EXEC file >> with PROT_EXEC, which would indeed have a real effect (in an iOS-like >> world, for example) but would break many, many things. > > And I think this would be useful (with the two possible ways of > executing .text split into FMODE_EXEC and FMODE_MAP_EXEC, as mentioned > in a sister subthread), but would have to be opt-in for the obvious > reason you outlined. However, we could make it the default for > openat2(2) -- assuming we can agree on what the semantics of a > theoretical FMODE_EXEC should be. > > And of course we'd need to do FMODE_UPGRADE_EXEC (which would need to > also permit fexecve(2) though probably not PROT_EXEC -- I don't think > you can mmap() an O_PATH descriptor). The mmapping restriction may be interesting but it is a different use case. This series address the interpreter/script problem. Either the script may be mapped executable is the choice of the interpreter. In most cases, no script are mapped as such, exactly because they are interpreted by a process but not by the CPU. -- Mickaël Salaün Les données à caractère personnel recueillies et traitées dans le cadre de cet échange, le sont à seule fin d’exécution d’une relation professionnelle et s’opèrent dans cette seule finalité et pour la durée nécessaire à cette relation. Si vous souhaitez faire usage de vos droits de consultation, de rectification et de suppression de vos données, veuillez contacter contact.rgpd@sgdsn.gouv.fr. Si vous avez reçu ce message par erreur, nous vous remercions d’en informer l’expéditeur et de détruire le message. The personal data collected and processed during this exchange aims solely at completing a business relationship and is limited to the necessary duration of that relationship. If you wish to use your rights of consultation, rectification and deletion of your data, please contact: contact.rgpd@sgdsn.gouv.fr. If you have received this message in error, we thank you for informing the sender and destroying the message.
Hi, The goal of this patch series is to control script interpretation. A new O_MAYEXEC flag used by sys_open() is added to enable userspace script interpreter to delegate to the kernel (and thus the system security policy) the permission to interpret/execute scripts or other files containing what can be seen as commands. This second series mainly differ from the previous one [1] by moving the basic security policy from Yama to the filesystem subsystem. This policy can be enforced by the system administrator through a sysctl configuration consistent with the mount points. Furthermore, the security policy can also be delegated to an LSM, either a MAC system or an integrity system. For instance, the new kernel MAY_OPENEXEC flag closes a major IMA measurement/appraisal interpreter integrity gap by bringing the ability to check the use of scripts [2]. Other uses are expected, such as for openat2(2) [3], SGX integration [4], and bpffs [5]. Userspace need to adapt to take advantage of this new feature. For example, the PEP 578 [6] (Runtime Audit Hooks) enables Python 3.8 to be extended with policy enforcement points related to code interpretation, which can be used to align with the PowerShell audit features. Additional Python security improvements (e.g. a limited interpreter withou -c, stdin piping of code) are on their way. The initial idea come from CLIP OS and the original implementation has been used for more than 10 years: https://github.com/clipos-archive/clipos4_doc An introduction to O_MAYEXEC was given at the Linux Security Summit Europe 2018 - Linux Kernel Security Contributions by ANSSI: https://www.youtube.com/watch?v=chNjCRtPKQY&t=17m15s The "write xor execute" principle was explained at Kernel Recipes 2018 - CLIP OS: a defense-in-depth OS: https://www.youtube.com/watch?v=PjRE0uBtkHU&t=11m14s This patch series can be applied on top of v5.3-rc7. This can be tested with CONFIG_SYSCTL. I would really appreciate constructive comments on this patch series. # Changes since v1 * move code from Yama to the FS subsystem * set __FMODE_EXEC when using O_MAYEXEC to make this information available through the new fanotify/FAN_OPEN_EXEC event * only match regular files (not directories nor other types), which follows the same semantic as commit 73601ea5b7b1 ("fs/open.c: allow opening only regular files during execve()") * improve tests [1] https://lore.kernel.org/lkml/20181212081712.32347-1-mic@digikod.net/ [2] https://lore.kernel.org/lkml/1544647356.4028.105.camel@linux.ibm.com/ [3] https://lore.kernel.org/lkml/20190904201933.10736-6-cyphar@cyphar.com/ [4] https://lore.kernel.org/lkml/CALCETrVovr8XNZSroey7pHF46O=kj_c5D9K8h=z2T_cNrpvMig@mail.gmail.com/ [5] https://lore.kernel.org/lkml/CALCETrVeZ0eufFXwfhtaG_j+AdvbzEWE0M3wjXMWVEO7pj+xkw@mail.gmail.com/ [6] https://www.python.org/dev/peps/pep-0578/ Regards, Mickaël Salaün (5): fs: Add support for an O_MAYEXEC flag on sys_open() fs: Add a MAY_EXECMOUNT flag to infer the noexec mount propertie fs: Enable to enforce noexec mounts or file exec through O_MAYEXEC selftest/exec: Add tests for O_MAYEXEC enforcing doc: Add documentation for the fs.open_mayexec_enforce sysctl Documentation/admin-guide/sysctl/fs.rst | 43 +++ fs/fcntl.c | 2 +- fs/namei.c | 70 +++++ fs/open.c | 6 + include/linux/fcntl.h | 2 +- include/linux/fs.h | 7 + include/uapi/asm-generic/fcntl.h | 3 + kernel/sysctl.c | 7 + tools/testing/selftests/exec/.gitignore | 1 + tools/testing/selftests/exec/Makefile | 4 +- tools/testing/selftests/exec/omayexec.c | 317 ++++++++++++++++++++ tools/testing/selftests/kselftest_harness.h | 3 + 12 files changed, 462 insertions(+), 3 deletions(-) create mode 100644 tools/testing/selftests/exec/omayexec.c