Zircon fundamentals

Zircon is the core that powers Fuchsia. It is composed of a kernel and a small set of userspace services, drivers, and libraries necessary for core system functions such as booting.

Although Zircon applies many of the concepts popularized by microkernels, it does not strive to be minimal. Instead, the microkernel-like architecture of Zircon enables Fuchsia to reduce the amount of trusted code running in the system to a few core functions:

• Memory management
• Scheduling
• Inter-process communication

System calls

User space code interacts with the objects in kernel space using system calls. Zircon has system calls to perform low-level operations such as:

• Memory management
• Task and process management
• Inter-process communication (IPC) and synchronization
• Exception handling
• Hardware support services (clocks, entropy, device I/O)

User space processes access system calls through libzircon.so — a virtual Dynamic Shared Object (vDSO) . The Zircon vDSO is a shared library in ELF format that the kernel maps into the address space of each new process. This library is considered "virtual" because it is exposed directly by the kernel image rather than being loaded from a file.

Most system calls operate directly with one or more handles — process-local references to objects living in kernel space represented as a 32-bit integer (zx_handle_t). Each handle declares the privileges, or rights, the holder has to perform actions on the handle itself or the referenced object.

Jobs, processes and threads

Zircon exposes three main kernel objects for running code:

• Thread: Thread of execution within a given address space.
• Process: Set of executable instructions run in a private, isolated address space.
• Job: Group of related processes and jobs. All jobs form a single rooted tree.

Processes form the basis for system capabilities. Each process is granted a set of capabilities through the various handles it holds.

Fuchsia software may or may not run within the confines of a single process. Jobs allow "applications" that are composed of more than one process to be controlled as a single entity.

Inter-process communication

Since processes are isolated by default, the kernel needs to provide a way for them to securely communicate with each other. Zircon includes the following kernel object types for inter-process communication (IPC):

• Event: Signaling interface between two processes.
• Socket: Streaming data transport, similar to a pipe.
• Stream: Streaming data transport that is seekable, like a file.
• Channel: Message-based transport capable of passing both data and a set of handles.
• FIFO: Control plane for shared memory access, optimized for small data payloads.

Among these objects, channels are uniquely suited to assist in launching new processes because they are capable of transferring handles (and therefore, capabilities) across to another process.

Channels have exactly two endpoint handles, each owned by a separate process. Only the owners may read or write messages, but ownership of an endpoint may be transferred from one process to another. When handles are written into a channel, they are removed from the sending process. When a message with handles is read from a channel, the handles are added to the receiving process.

Zircon channels are the basis for service-level IPC protocols described by the Fuchsia Interface Definition Language (FIDL) . FIDL protocols are the primary method of IPC used by Fuchsia programs. You will explore creating and consuming FIDL protocols in more detail later on.

Exercise: Jobs and processes

Let's explore some of these fundamental concepts on a running system. In this exercise, you'll see how jobs and processes interact to form a tree.

Start the emulator

If you do not already have an instance running, start FEMU with networking support:

ffx emu start workstation_eng.x64 --headless

Dump the process list

Connect to a device shell prompt and use the ps command to dump the list of running jobs and processes.

ffx target ssh ps

Below is a trimmed example of what the output looks like:

TASK                     PSS PRIVATE  SHARED   STATE NAME
j: 1027               507.8M  507.4M                 root
  p: 1061             564.4k    564k     36k         bin/bootsvc
  p: 1150            4264.4k   4264k     36k         bin/component_manager
  j: 1479             228.4k    228k
    p: 1583           228.4k    228k     36k         pwrbtn-monitor.cm
  j: 1484             532.4k    532k
    p: 1599           532.4k    532k     36k         svchost.cm
  j: 1544             402.4k    304k
    p: 1633           402.4k    304k    232k         netsvc.cm
  j: 1681             296.4k    296k
    p: 1733           296.4k    296k     36k         console-launcher.cm
  j: 1799            7232.4k   7232k
    p: 1825          7232.4k   7232k     36k         archivist.cm
  j: 1927             660.4k    660k
    p: 1955           660.4k    660k     36k         base-resolver.cm
  j: 2072            1016.4k   1016k
    p: 2088          1016.4k   1016k     36k         driver_manager.cm
  j: 2239             348.4k    348k
    p: 2252           348.4k    348k     36k         device-name-provider.cm
  j: 2364             275.3M  275.3M
    p: 2380          1012.4k   1012k     36k         fshost.cm
    p: 6544           252.1M  252.1M     36k         /pkg/bin/blobfs
    p: 10205         9744.4k   9744k     36k         /pkg/bin/minfs
    p: 10475           12.8M   12.8M     36k         pkgfs

Let's focus on two columns in the output for now:

• TASK: This tells you whether each entry is a job (j) or process (p) followed by their unique id.
• NAME: This provides a little more detail about what piece of the system is running there.

Let's break down some interesting things here based on what we've discussed so far:

1. Every process is connected to a parent job. Some jobs have multiple processes.
2. All jobs trace back to the root job as the ultimate parent, forming a tree.
3. During startup, the system launches a few processes directly into the root job. Most other processes are launched under their own parent jobs.
4. After the initial startup work, many of the entries have a .cm extension. These refer to components, and you will learn more about them later on.
5. Some of these components are core services like filesystems (fshost.cm) and drivers (driver_manager.cm) that live in user space separate from the kernel.

Next, we'll explore how the Zircon enables the fundamentals of Fuchsia's security model.