This patchset is being developed here: <https://github.com/cyphar/linux/tree/openat2/master> Patch changelog: v14: * The magic-link changes (and O_EMPTYPATH) have been dropped from this series -- they will be developed and sent separately. The main reason is that we need to restrict things other than open(2) (examples include truncate(2) as well as mount(MS_BIND)). This will require a fair amount of extra work, and there's no point stalling openat2(2) for that work to be completed. * Minor rework of 'struct open_how': * To avoid future headaches, make it a non-const argument. * Expand ->flags and ->resolve to 64-bit fields to allow for more flag extensions without needing to add separate fields too early. This requires adding a bit of explicit padding (32 bits) to avoid userspace putting garbage in the alignment padding -- this can be repurposed for future extensions. * upgrade_mask is dropped (and will be a separate field when we add it again in the future) to avoid userspace foot-guns. * Expand -EINVAL checks in build_open_flags(). Rather than silently ignoring silly flag combinations (such as O_TMPFILE|O_PATH or O_PATH|<most flags>), give an -EINVAL. All of the silent ignore semantics were added to open(2) because we couldn't return -EINVAL -- but we can now! * open(2) and openat(2) clean up their flags before passing them to build_open_flags(), so all mixed flags will continue to work. There is one exception which is (O_PATH|O_TMPFILE) -- this is no longer permitted (as far as I can tell this appears to be a bug, and there are no userspace users that I've hit after running this code for a few days). If it turns out that userspace does depend on (O_PATH|O_TMPFILE) working, we can only disallow it for openat2(2). * Don't zero out nd->root in complete_walk() for RCU-walk if we're doing a scoped-lookup (this prevents a needless REF-walk retry). * Attempt all tests on kernels that don't have openat2(2), rather than just skipping everything. v13: <https://lore.kernel.org/lkml/20190930183316.10190-1-cyphar@xxxxxxxxxx/> v12: <https://lore.kernel.org/lkml/20190904201933.10736-1-cyphar@xxxxxxxxxx/> v11: <https://lore.kernel.org/lkml/20190820033406.29796-1-cyphar@xxxxxxxxxx/> <https://lore.kernel.org/lkml/20190728010207.9781-1-cyphar@xxxxxxxxxx/> v10: <https://lore.kernel.org/lkml/20190719164225.27083-1-cyphar@xxxxxxxxxx/> v09: <https://lore.kernel.org/lkml/20190706145737.5299-1-cyphar@xxxxxxxxxx/> v08: <https://lore.kernel.org/lkml/20190520133305.11925-1-cyphar@xxxxxxxxxx/> v07: <https://lore.kernel.org/lkml/20190507164317.13562-1-cyphar@xxxxxxxxxx/> v06: <https://lore.kernel.org/lkml/20190506165439.9155-1-cyphar@xxxxxxxxxx/> v05: <https://lore.kernel.org/lkml/20190320143717.2523-1-cyphar@xxxxxxxxxx/> v04: <https://lore.kernel.org/lkml/20181112142654.341-1-cyphar@xxxxxxxxxx/> v03: <https://lore.kernel.org/lkml/20181009070230.12884-1-cyphar@xxxxxxxxxx/> v02: <https://lore.kernel.org/lkml/20181009065300.11053-1-cyphar@xxxxxxxxxx/> v01: <https://lore.kernel.org/lkml/20180929103453.12025-1-cyphar@xxxxxxxxxx/> For a very long time, extending openat(2) with new features has been incredibly frustrating. This stems from the fact that openat(2) is possibly the most famous counter-example to the mantra "don't silently accept garbage from userspace" -- it doesn't check whether unknown flags are present[1]. This means that (generally) the addition of new flags to openat(2) has been fraught with backwards-compatibility issues (O_TMPFILE has to be defined as __O_TMPFILE|O_DIRECTORY|[O_RDWR or O_WRONLY] to ensure old kernels gave errors, since it's insecure to silently ignore the flag[2]). All new security-related flags therefore have a tough road to being added to openat(2). Furthermore, the need for some sort of control over VFS's path resolution (to avoid malicious paths resulting in inadvertent breakouts) has been a very long-standing desire of many userspace applications. This patchset is a revival of Al Viro's old AT_NO_JUMPS[3] patchset (which was a variant of David Drysdale's O_BENEATH patchset[4] which was a spin-off of the Capsicum project[5]) with a few additions and changes made based on the previous discussion within [6] as well as others I felt were useful. In line with the conclusions of the original discussion of AT_NO_JUMPS, the flag has been split up into separate flags. However, instead of being an openat(2) flag it is provided through a new syscall openat2(2) which provides several other improvements to the openat(2) interface (see the patch description for more details). The following new LOOKUP_* flags are added: * LOOKUP_NO_XDEV blocks all mountpoint crossings (upwards, downwards, or through absolute links). Absolute pathnames alone in openat(2) do not trigger this. Magic-link traversal which implies a vfsmount jump is also blocked (though magic-link jumps on the same vfsmount are permitted). * LOOKUP_NO_MAGICLINKS blocks resolution through /proc/$pid/fd-style links. This is done by blocking the usage of nd_jump_link() during resolution in a filesystem. The term "magic-links" is used to match with the only reference to these links in Documentation/, but I'm happy to change the name. It should be noted that this is different to the scope of ~LOOKUP_FOLLOW in that it applies to all path components. However, you can do openat2(NO_FOLLOW|NO_MAGICLINKS) on a magic-link and it will *not* fail (assuming that no parent component was a magic-link), and you will have an fd for the magic-link. In order to correctly detect magic-links, the introduction of a new LOOKUP_MAGICLINK_JUMPED state flag was required. * LOOKUP_BENEATH disallows escapes to outside the starting dirfd's tree, using techniques such as ".." or absolute links. Absolute paths in openat(2) are also disallowed. Conceptually this flag is to ensure you "stay below" a certain point in the filesystem tree -- but this requires some additional to protect against various races that would allow escape using "..". Currently LOOKUP_BENEATH implies LOOKUP_NO_MAGICLINKS, because it can trivially beam you around the filesystem (breaking the protection). In future, there might be similar safety checks done as in LOOKUP_IN_ROOT, but that requires more discussion. In addition, two new flags are added that expand on the above ideas: * LOOKUP_NO_SYMLINKS does what it says on the tin. No symlink resolution is allowed at all, including magic-links. Just as with LOOKUP_NO_MAGICLINKS this can still be used with NOFOLLOW to open an fd for the symlink as long as no parent path had a symlink component. * LOOKUP_IN_ROOT is an extension of LOOKUP_BENEATH that, rather than blocking attempts to move past the root, forces all such movements to be scoped to the starting point. This provides chroot(2)-like protection but without the cost of a chroot(2) for each filesystem operation, as well as being safe against race attacks that chroot(2) is not. If a race is detected (as with LOOKUP_BENEATH) then an error is generated, and similar to LOOKUP_BENEATH it is not permitted to cross magic-links with LOOKUP_IN_ROOT. The primary need for this is from container runtimes, which currently need to do symlink scoping in userspace[7] when opening paths in a potentially malicious container. There is a long list of CVEs that could have bene mitigated by having RESOLVE_THIS_ROOT (such as CVE-2017-1002101, CVE-2017-1002102, CVE-2018-15664, and CVE-2019-5736, just to name a few). In order to make all of the above more usable, I'm working on libpathrs[8] which is a C-friendly library for safe path resolution. It features a userspace-emulated backend if the kernel doesn't support openat2(2). Hopefully we can get userspace to switch to using it, and thus get openat2(2) support for free once it's ready. [1]: https://lwn.net/Articles/588444/ [2]: https://lore.kernel.org/lkml/CA+55aFyyxJL1LyXZeBsf2ypriraj5ut1XkNDsunRBqgVjZU_6Q@xxxxxxxxxxxxxx [3]: https://lore.kernel.org/lkml/20170429220414.GT29622@xxxxxxxxxxxxxxxxxx [4]: https://lore.kernel.org/lkml/1415094884-18349-1-git-send-email-drysdale@xxxxxxxxxx [5]: https://lore.kernel.org/lkml/1404124096-21445-1-git-send-email-drysdale@xxxxxxxxxx [6]: https://lwn.net/Articles/723057/ [7]: https://github.com/cyphar/filepath-securejoin [8]: https://github.com/openSUSE/libpathrs The current draft of the openat2(2) man-page is included below. --8<--------------------------------------------------------------------------- OPENAT2(2) Linux Programmer's Manual OPENAT2(2) NAME openat2 - open and possibly create a file (extended) SYNOPSIS #include <sys/types.h> #include <sys/stat.h> #include <fcntl.h> int openat2(int dirfd, const char *pathname, struct open_how *how, size_t size); Note: There is no glibc wrapper for this system call; see NOTES. DESCRIPTION The openat2() system call opens the file specified by pathname. If the specified file does not exist, it may optionally (if O_CREAT is specified in how.flags) be created by openat2(). As with openat(2), if pathname is relative, then it is interpreted relative to the directory referred to by the file descriptor dirfd (or the current working directory of the calling process, if dirfd is the special value AT_FDCWD.) If pathname is absolute, then dirfd is ignored (unless how.resolve contains RESOLVE_IN_ROOT, in which case pathname is resolved relative to dirfd.) The openat2() system call is an extension of openat(2) and provides a superset of its functionality. Rather than taking a single flag argument, an extensible structure (how) is passed instead to allow for future extensions. size must be set to sizeof(struct open_how), to facilitate future extensions (see the "Extensibility" section of the NOTES for more detail on how extensions are handled.) The open_how structure The following structure indicates how pathname should be opened, and acts as a superset of the flag and mode arguments to openat(2). struct open_how { __aligned_u64 flags; /* O_* flags. */ __u16 mode; /* Mode for O_{CREAT,TMPFILE}. */ __u16 __padding[3]; /* Must be zeroed. */ __aligned_u64 resolve; /* RESOLVE_* flags. */ }; Any future extensions to openat2() will be implemented as new fields appended to the above structure (or through reuse of pre-existing padding space), with the zero value of the new fields acting as though the extension were not present. The meaning of each field is as follows: flags The file creation and status flags to use for this operation. All of the O_* flags defined for openat(2) are valid openat2() flag values. Unlike openat(2), it is an error to provide openat2() unknown or conflicting flags in flags. mode File mode for the new file, with identical semantics to the mode argument to openat(2). However, unlike openat(2), it is an error to provide openat2() with a mode which contains bits other than 0777. It is an error to provide openat2() a non-zero mode if flags does not contain O_CREAT or O_TMPFILE. resolve Change how the components of pathname will be resolved (see path_resolution(7) for background information.) The primary use case for these flags is to allow trusted programs to restrict how untrusted paths (or paths inside untrusted directories) are resolved. The full list of resolve flags is given below. RESOLVE_NO_XDEV Disallow traversal of mount points during path resolution (including all bind mounts). Users of this flag are encouraged to make its use configurable (unless it is used for a specific security purpose), as bind mounts are very widely used by end-users. Setting this flag indiscrimnately for all uses of openat2() may result in spurious errors on previously-functional systems. RESOLVE_NO_SYMLINKS Disallow resolution of symbolic links during path resolution. This option implies RESOLVE_NO_MAGICLINKS. If the trailing component is a symbolic link, and flags contains both O_PATH and O_NOFOLLOW, then an O_PATH file descriptor referencing the symbolic link will be returned. Users of this flag are encouraged to make its use configurable (unless it is used for a specific security purpose), as symbolic links are very widely used by end-users. Setting this flag indiscrimnately for all uses of openat2() may result in spurious errors on previously-functional systems. RESOLVE_NO_MAGICLINKS Disallow all magic link resolution during path resolution. If the trailing component is a magic link, and flags contains both O_PATH and O_NOFOLLOW, then an O_PATH file descriptor referencing the magic link will be returned. Magic-links are symbolic link-like objects that are most notably found in proc(5) (examples include /proc/[pid]/exe and /proc/[pid]/fd/*.) Due to the potential danger of unknowingly opening these magic links, it may be preferable for users to disable their resolution entirely (see symboliclink(7) for more details.) RESOLVE_BENEATH Do not permit the path resolution to succeed if any component of the resolution is not a descendant of the directory indicated by dirfd. This results in absolute symbolic links (and absolute values of pathname) to be rejected. Currently, this flag also disables magic link resolution. However, this may change in the future. The caller should explicitly specify RESOLVE_NO_MAGICLINKS to ensure that magic links are not resolved. RESOLVE_IN_ROOT Treat dirfd as the root directory while resolving pathname (as though the user called chroot(2) with dirfd as the argument.) Absolute symbolic links and ".." path components will be scoped to dirfd. If pathname is an absolute path, it is also treated relative to dirfd. However, unlike chroot(2) (which changes the filesystem root permanently for a process), RESOLVE_IN_ROOT allows a program to efficiently restrict path resolution for only certain operations. It also has several hardening features (such detecting escape attempts during .. resolution) which chroot(2) does not. Currently, this flag also disables magic link resolution. However, this may change in the future. The caller should explicitly specify RESOLVE_NO_MAGICLINKS to ensure that magic links are not resolved. It is an error to provide openat2() unknown flags in resolve. RETURN VALUE On success, a new file descriptor is returned. On error, -1 is returned, and errno is set appropriately. ERRORS The set of errors returned by openat2() includes all of the errors returned by openat(2), as well as the following additional errors: EINVAL An unknown flag or invalid value was specified in how. EINVAL mode is non-zero, but flags does not contain O_CREAT or O_TMPFILE. EINVAL size was smaller than any known version of struct open_how. E2BIG An extension was specified in how, which the current kernel does not support (see the "Extensibility" section of the NOTES for more detail on how extensions are handled.) EAGAIN resolve contains either RESOLVE_IN_ROOT or RESOLVE_BENEATH, and the kernel could not ensure that a ".." component didn't escape (due to a race condition or potential attack.) Callers may choose to retry the openat2() call. EXDEV resolve contains either RESOLVE_IN_ROOT or RESOLVE_BENEATH, and an escape from the root during path resolution was detected. EXDEV resolve contains RESOLVE_NO_XDEV, and a path component attempted to cross a mount point. ELOOP resolve contains RESOLVE_NO_SYMLINKS, and one of the path components was a symbolic link (or magic link). ELOOP resolve contains RESOLVE_NO_MAGICLINKS, and one of the path components was a magic link. VERSIONS openat2() was added to Linux in kernel 5.FOO. CONFORMING TO This system call is Linux-specific. The semantics of RESOLVE_BENEATH were modelled after FreeBSD's O_BENEATH. NOTES Glibc does not provide a wrapper for this system call; call it using systemcall(2). Extensibility In order to allow for struct open_how to be extended in future kernel revisions, openat2() requires userspace to specify the size of struct open_how structure they are passing. By providing this information, it is possible for openat2() to provide both forwards- and backwards-compatibility — with size acting as an implicit version number (because new extension fields will always be appended, the size will always increase.) This extensibility design is very similar to other system calls such as perf_setattr(2), perf_event_open(2), and clone(3). If we let usize be the size of the structure according to userspace and ksize be the size of the structure which the kernel supports, then there are only three cases to consider: * If ksize equals usize, then there is no version mismatch and how can be used verbatim. * If ksize is larger than usize, then there are some extensions the kernel supports which the userspace program is unaware of. Because all extensions must have their zero values be a no-op, the kernel treats all of the extension fields not set by userspace to have zero values. This provides backwards- compatibility. * If ksize is smaller than usize, then there are some extensions which the userspace program is aware of but the kernel does not support. Because all extensions must have their zero values be a no-op, the kernel can safely ignore the unsupported extension fields if they are all-zero. If any unsupported extension fields are non-zero, then -1 is returned and errno is set to E2BIG. This provides forwards-compatibility. Therefore, most userspace programs will not need to have any special handling of extensions. However, if a userspace program wishes to determine what extensions the running kernel supports, they may conduct a binary search on size (to find the largest value which doesn't produce an error of E2BIG.) SEE ALSO openat(2), path_resolution(7), symboliclink(7) Linux 2019-10-10 OPENAT2(2) --8<--------------------------------------------------------------------------- Aleksa Sarai (6): namei: O_BENEATH-style resolution restriction flags namei: LOOKUP_IN_ROOT: chroot-like path resolution namei: permit ".." resolution with LOOKUP_{IN_ROOT,BENEATH} open: introduce openat2(2) syscall selftests: add openat2(2) selftests Documentation: path-lookup: mention LOOKUP_MAGICLINK_JUMPED CREDITS | 4 +- Documentation/filesystems/path-lookup.rst | 18 +- arch/alpha/kernel/syscalls/syscall.tbl | 1 + arch/arm/tools/syscall.tbl | 1 + arch/arm64/include/asm/unistd.h | 2 +- arch/arm64/include/asm/unistd32.h | 2 + arch/ia64/kernel/syscalls/syscall.tbl | 1 + arch/m68k/kernel/syscalls/syscall.tbl | 1 + arch/microblaze/kernel/syscalls/syscall.tbl | 1 + arch/mips/kernel/syscalls/syscall_n32.tbl | 1 + arch/mips/kernel/syscalls/syscall_n64.tbl | 1 + arch/mips/kernel/syscalls/syscall_o32.tbl | 1 + arch/parisc/kernel/syscalls/syscall.tbl | 1 + arch/powerpc/kernel/syscalls/syscall.tbl | 1 + arch/s390/kernel/syscalls/syscall.tbl | 1 + arch/sh/kernel/syscalls/syscall.tbl | 1 + arch/sparc/kernel/syscalls/syscall.tbl | 1 + arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + arch/xtensa/kernel/syscalls/syscall.tbl | 1 + fs/namei.c | 167 +++++- fs/open.c | 154 ++++-- include/linux/fcntl.h | 12 +- include/linux/namei.h | 12 + include/linux/syscalls.h | 3 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/fcntl.h | 41 ++ tools/testing/selftests/Makefile | 1 + tools/testing/selftests/openat2/.gitignore | 1 + tools/testing/selftests/openat2/Makefile | 8 + tools/testing/selftests/openat2/helpers.c | 109 ++++ tools/testing/selftests/openat2/helpers.h | 107 ++++ .../testing/selftests/openat2/openat2_test.c | 297 ++++++++++ .../selftests/openat2/rename_attack_test.c | 160 ++++++ .../testing/selftests/openat2/resolve_test.c | 523 ++++++++++++++++++ 35 files changed, 1571 insertions(+), 71 deletions(-) create mode 100644 tools/testing/selftests/openat2/.gitignore create mode 100644 tools/testing/selftests/openat2/Makefile create mode 100644 tools/testing/selftests/openat2/helpers.c create mode 100644 tools/testing/selftests/openat2/helpers.h create mode 100644 tools/testing/selftests/openat2/openat2_test.c create mode 100644 tools/testing/selftests/openat2/rename_attack_test.c create mode 100644 tools/testing/selftests/openat2/resolve_test.c -- 2.23.0
