Zircon is an object-based kernel. User mode code almost exclusively interacts with OS resources via object handles. A handle can be thought of as an active session with a specific OS subsystem scoped to a particular resource.
Zircon actively manages the following resources:
- processor time
- memory and address spaces
- device-io memory
- signaling and waiting
- inter-process communication
Kernel objects for applications
Memory and address space
Kernel objects for drivers
Kernel object lifetime
Kernel objects are reference-counted. Most kernel objects are created during a 'create' syscall and are held alive by the first handle, given as the output of the create syscall. The caller gets the numeric id of the handle and the handle itself is placed in the handle table of the process.
A handle is held alive as long it exists in the handle table. Handles are removed from the handle table by:
Closing them via
zx_handle_closewhich decrements the reference count of the corresponding kernel object. Usually, when the last handle is closed the kernel object reference count will reach 0 which causes the kernel object to be destroyed.
When a channel endpoint holding an unread message containing a handle is destroyed, all of the pending messages will also be destroyed, closing any handles contained in the messages while doing so.
When the process that owns the handle table is destroyed. The kernel effectively iterates over the entire handle table closing each handle in turn.
The reference count increases when new handles (referring to the same object)
are created via
zx_handle_duplicate, but also when a direct pointer
reference (by some kernel code) is acquired; therefore a kernel object lifetime
might be longer than the lifetime of the code that created it. A separate count
of active handles referencing an object is also maintained, allowing the kernel
to trigger specific behaviors when the handle count of an object reaches zero,
even if the kernel is keeping the object alive behind the scenes because of a
direct pointer reference.
There are three important cases in which kernel objects are kept alive when there are no outstanding handles to them:
The object is referenced by a handle in a message which has not been consumed. This can happen via the channel APIs. While such handle is in the channel the kernel keeps the object alive, and with a non-zero active handle count.
The outcome of the last case is that a single thread can keep its process and the entire lineage of jobs up to the root job alive.
Peered object and the peer-closed state
Currently, the kernel defines the following object types as "peered" objects.
|Name||Peer-Closed Signal Name|
All peered objects are created in pairs, which are internally linked to each
other in a peer relationship. When the active handle count of a peered object
reaches 0, if that object still has a link to its peer, the peer object will be
placed in the
PEER_CLOSED state, causing the link to be destroyed, the
ZX_*_PEER_CLOSED signal to become asserted on the peer, and for
syscalls involving the object's peer (for example,
zx_channel_write) to return
When the final handle to an object is closed via a call to
zx_handle_close_many, it is guaranteed that the object's peer (if any)
will be placed into the
PEER_CLOSED state, asserting its associated signal in
the process, before the
zx_handle_close syscall returns from the kernel.
Note that objects are placed into
PEER_CLOSED when their peer's active handle
count has hit zero, even of the peer object continues to live because of a
direct pointer reference held by the kernel.
Kernel Object security
Kernel objects do not have an intrinsic notion of security and do not do authorization checks; security rights are held by each handle. A single process can have two different handles to the same object with different rights.