Module rsyscall.linux.memfd

Classes

class MFD (value, names=None, *, module=None, qualname=None, type=None, start=1)

An enumeration.

Expand source code Browse git

class MFD(enum.IntFlag):
    NONE = 0
    CLOEXEC = lib.MFD_CLOEXEC
    ALLOW_SEALING = lib.MFD_ALLOW_SEALING
    HUGETLB = lib.MFD_HUGETLB
    HUGE_2MB = lib.MFD_HUGE_2MB
    HUGE_1GB = lib.MFD_HUGE_1GB

Ancestors

• enum.IntFlag
• builtins.int
• enum.Flag
• enum.Enum

Class variables

var NONE

var CLOEXEC

var ALLOW_SEALING

var HUGETLB

var HUGE_2MB

var HUGE_1GB

class MemfdTask (sysif: SyscallInterface, near_process: Process, fd_table: FDTable, address_space: AddressSpace, pidns: PidNamespace)

A wrapper around SyscallInterface which tracks the namespaces of the underlying process

Note that this is a base class for the more fully featured Task.

We store namespace objects to represent the namespaces that we believe that underlying processes is in. Since we have complete control over the process, we can make sure this belief is accurate, by updating our stored namespaces when the process changes namespace. That isn't done here; it's done in handle.Task.

Currently, we store only one PidNamespace. But each process actually has two pid namespaces:

• the process's own pid namespace, which determines the pids returned from getpid, clone, and other syscalls.
• the pid namespace that new children will be in.

The two pid namespaces only differ if we call unshare(CLONE.NEWPID). Currently we don't do that because unshare(CLONE.NEWPID) makes monitoring children more complex, since they can be deleted without leaving a zombie at any time if the pid namespace shuts down. But if we did call unshare(CLONE.NEWPID), we'd need to handle this right.

In the analogy to near and far pointers, this is like a segment register, if a segment register was write-only. Then we'd need to maintain the knowledge of what the segment register was set to, outside the segment register itself. That's what we do here.

There actually were systems where segment registers were, if not quite write-only, at least expensive to set and expensive to read. For example, x86_64 - the FS and GS segment registers can only be set via syscall. If you wanted to use segmentation on such systems, you'd probably have a structure much like this one.

Expand source code Browse git

class MemfdTask(FileDescriptorTask[T_fd]):
    async def memfd_create(self, name: WrittenPointer[t.Union[str, os.PathLike]], flags: MFD=MFD.NONE) -> T_fd:
        with name.borrow(self) as name_n:
            fd = await _memfd_create(self.sysif, name_n, flags|MFD.CLOEXEC)
            return self.make_fd_handle(fd)

Ancestors

• FileDescriptorTask
• Task
• typing.Generic

Subclasses

• Task

Class variables

var sysif : SyscallInterface

var near_process : Process

var fd_table : FDTable

var address_space : AddressSpace

var pidns : PidNamespace

Methods

async def memfd_create(self, name: WrittenPointer[t.Union[str, os.PathLike]], flags: MFD = MFD.NONE) ‑> T_fd

Expand source code Browse git

async def memfd_create(self, name: WrittenPointer[t.Union[str, os.PathLike]], flags: MFD=MFD.NONE) -> T_fd:
    with name.borrow(self) as name_n:
        fd = await _memfd_create(self.sysif, name_n, flags|MFD.CLOEXEC)
        return self.make_fd_handle(fd)

Inherited members

• FileDescriptorTask:
  • inherit_fd
  • make_fd_handle
  • unshare_files