utils: Enable mount_setattr when compiling with older glibc

[ao2]

On some systems where bubblewrap is compiled/run it might happen that the kernel is new enough to have the mount_setattr syscall implemented, but the userspace side still needs __NR_mount_setattr explicitly defined to access it, because glibc may lag behind.

Define a fallback definition of __NR_mount_setattr at least for some more common architectures to enable mount_setattr, on such systems.

For instance, this enables the possible usage of mount_setattr with bubblewrap binaries compiled in the Steam Runtime SDK.

[ao2]

cc @smcv

[smcv]

LGTM, have you confirmed (using strace) that a bwrap compiled in a sufficiently old container still calls the mount_setattr syscall?

(For example you could compile in https://gitlab.steamos.cloud/steamrt/scout/sdk and run the resulting binary on a modern system)

[ao2]

LGTM, have you confirmed (using strace) that a bwrap compiled in a sufficiently old container still calls the mount_setattr syscall?

Good idea, I'll do the strace test with and without this MR and report back.

(For example you could compile in https://gitlab.steamos.cloud/steamrt/scout/sdk and run the resulting binary on a modern system)

BTW it looks like we need the changes from #786 to be able to build in the scout SDK.

[ao2]

@smcv I built on scout two binaries, one without the changes in MR, and one with the changes in MR applied, and things seem to work as expected:

$ strace -f ./scout_bwrap_NO_bubblewrap_PR_785 --bind / / true 2>&1 | grep mount_setattr
$ strace -f ./scout_bwrap_bubblewrap_PR_785 --bind / / true 2>&1 | grep mount_setattr
[pid 107883] mount_setattr(7, "", AT_EMPTY_PATH|AT_RECURSIVE, {attr_set=MOUNT_ATTR_NOSUID|MOUNT_ATTR_NODEV, attr_clr=0, propagation=0 /* MS_??? */, userns_fd=0}, 32) = 0

[smcv]

things seem to work as expected

Great, let's apply this then.