A MacFUSE-Based Process File System for Mac OS X
referer: http://osxbook.com/book/bonus/chapter11/procfs/
Processes as Files
The process file system ("procfs" for brevity, or simply "/proc", because that's where it is usually mounted) has become a common entity on Unix-like systems. For example, Solaris, Linux, and the various modern BSDs all have procfs. In general, procfs uses the file metaphor to provide both a view of currently running processes and perhaps an interface to control them. However, some procfs implementations export so many types of information that they are nothing short of the proverbial kitchen sink. Implementations can also differ in the format they use to export such information. For example, Solaris procfs favors the binary format philosophy, assuming that developers will write interesting user-space tools that use that "raw" information. In contrast, Linux favors formatted text to export information through procfs, allowing end users and administrators to simply view that information using familiar file utilities.
Tom J. Killian created the first implementation of a process file system. It was for Eighth Edition UNIX. See T. J. Killian, "Processes as Files," USENIX Summer Conference Proceedings, Salt Lake City, UT, USA (June 1984).
/proc
on Mac OS X (not)
Mac OS X does not provide a process file system. It does provide alternative interfaces such as sysctl(3)
and the now obsolete kvm(3)
. The sysctl(3)
interface provides read and write access to a management information base (MIB) whose contents are various categories of kernel information, such as information related to processes, file systems, virtual memory, networking, and debugging in general. When the /dev/kmem
device is available, the kvm(3)
interface provides access to raw kernel memory. Besides, the I/O Kit programming interfaces and tools such as ioreg
and IORegistryExplorer.app
allow user-space inspection of a variety of kernel information on Mac OS X.
The sysctl()
system call was introduced in 4.4BSD as a safe, reliable, and portable (across kernel versions) way to perform user-kernel data exchange.
In 2003, shortly after I was introduced to Mac OS X, I began a quick-and-dirty implementation of procfs for Mac OS X. Meant as nothing more than a "random experiment", my approach was to take procfs
and pseudofs
source from FreeBSD and morph them to work on Mac OS X. Shortly after I had the thing mounting, I ran into a kernel panic. (Writing in-kernel file systems is very conducive to this behavior.) Before I could begin two-machine debugging, one of the two Macintosh computers I had access to had catastrophic hardware failure. Apple gave me extremely poor repair experience (taking weeks to fix one problem and returning the machine with something else broken; repeat this several times over). So much so that by the time the machine finally worked, I had no interest left in debugging procfs for Mac OS X.
A MacFUSE-Based procfs
Fast forward to late 2006. I was wrapping up MacFUSE, a Mac OS X implementation of the FUSE (File System in User Space) mechanism. The first Mac OS X-specific example file system I wrote for MacFUSE was procfs—this time as a user-space file system, of course. I've been meaning to release it as an open source MacFUSE example, but didn't find the time to package it up until now.
The MacFUSE version of procfs makes heavy use of the Mach programming interfaces. Moreover, the implementation uses a set of macros that depend on the C++ version of the pcre (Perl Compatible Regular Expressions) library. The macros are meant to make it easier to extend the file system.
Detailed information on understanding and using the Mach interfaces used by procfs can be found in the book Mac OS X Internals. In particular, refer to Chapters 6 (The xnu Kernel), 7 (Processes), 8 (Virtual Memory), and 9 (Interprocess Communication).
Compiling and Using procfs
Since procfs depends on the pcre libraries, you will first need to download, compile, and install pcre. Note that if you install pcre in a location other than /usr/local/
, you will need to modify the Makefile
in procfs source. The Makefile
also assumes that you have the MacFUSE libraries installed under /usr/local/
.
$ tar -jxvf pcre-<version>.tar.bz2 ... $ cd pcre-<version> $ CFLAGS="-O -g -arch i386 -arch ppc -isysroot /Developer/SDKs/MacOSX10.4u.sdk" \ CXXFLAGS="-O -g -arch i386 -arch ppc -isysroot /Developer/SDKs/MacOSX10.4u.sdk" \ LDFLAGS="-arch i386 -arch ppc" \ ./configure --prefix=/usr/local --disable-dependency-tracking $ make $ sudo make install ...
Assuming you have the MacFUSE Core package installed and have the MacFUSE source tree checked out under /work/macfuse/
, you can compile procfs as follows.
$ cd /work/macfuse/filesystems/procfs $ make ...
Having successfully compiled procfs, you can mount it, say, on the /proc
directory, as follows.
$ sudo mkdir /proc $ sudo chown root:wheel /proc $ sudo ./procfs /proc $ mount ... procfs on /proc (read-only, synchronous)
Now you can explore the process file system on Mac OS X.
$ ls -F /proc 0/ 151/ 212/ 35/ 47/ 1/ 165/ 215/ 36/ 51/ 1166/ 187/ 218/ 38/ 5338/ 1187/ 194/ 221/ 39/ 54/ 1196/ 195/ 23/ 3954/ 59/ 1216/ 196/ 27/ 3976/ 62/ 131/ 200/ 28807/ 40/ 64/ 140/ 201/ 28902/ 42/ 67/ 145/ 202/ 32/ 43/ 867/ 146/ 204/ 33/ 45/ 95/ 15094/ 206/ 34/ 46/ hardware/
There is a directory under /proc
for each process on the system, with the numeric process ID (pid) being the directory name. pid 0 corresponds to the kernel, pid 1 is launchd
, and so on.
$ cd /proc/867 $ ls -l total 0 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 . 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 .. 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 carbon 0 -r--r--r-- 1 root wheel 0 May 7 23:44 cmdline 0 -r--r--r-- 1 root wheel 0 May 7 23:44 jobc 0 -r--r--r-- 1 root wheel 0 May 7 23:44 paddr 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 pcred 0 -r--r--r-- 1 root wheel 0 May 7 23:44 pgid 0 -r--r--r-- 1 root wheel 0 May 7 23:44 ppid 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 task 0 -r--r--r-- 1 root wheel 0 May 7 23:44 tdev 0 -r--r--r-- 1 root wheel 0 May 7 23:44 tpgid 0 dr-xr-xr-x 1 root wheel 0 May 7 23:44 ucred 0 -r--r--r-- 1 root wheel 0 May 7 23:44 wchan
For a given pid, the /proc/pid/
directory has some files containing certain BSD-level information. The carbon/
subdirectory contains additional information for processes that are also Carbon processes. The pcred/
and ucred/
subdirectories contains files exporting process and user credentials, respectively.
$ pwd /proc/867 $ ls -F carbon name psn $ cat carbon/name carbon/psn Safari 0:12451841 $ cat cmdline /Applications/Safari.app/Contents/MacOS/Safari -psn_0_12451841 $ ls -F ucred groups uid $ cat ucred/groups ucred/uid 501(singh) 81(appserveradm) 79(appserverusr) 80(admin) 501(singh)
The task/
subdirectory contains a variety of Mach task information: scheduling data, virtual memory regions, Mach ports, threads, register state, and so on.
$ pwd /proc/867 $ cd task $ ls -F absolutetime_info/ ports/ tokens/ basic_info/ role vmmap events_info/ thread_times_info/ mach_name threads/ $ cat role BACKGROUND_APPLICATION $ cat vmmap 00000000-00001000 4K ---/--- COPY - DEFAULT uwir=0 sub=0 00001000-000dc000 876K r-x/rwx COPY - DEFAULT uwir=0 sub=0 ... 15a54000-15ada000 536K r--/rwx COPY - DEFAULT uwir=0 sub=0 15ade000-15ede000 4096K rw-/rwx COPY - DEFAULT uwir=0 sub=0 $ ls -F basic_info policy suspend_count user_time resident_size system_time virtual_size $ ls -F events_info cow_faults messages_received syscalls_mach csw messages_sent syscalls_unix faults pageins $ ls -F ports 1003/ 1f03/ 2f03/ 4003/ 507/ 6423/ 7b03/ 8e03/ a887/ d093/ ... 1e03/ 2e03/ 3f03/ 5003/ 62b7/ 79f7/ 8d03/ a7c7/ d03/ $ ls -F ports/1003 msgcount qlimit seqno sorights task_rights $ cat ports/1003/task_rights RECEIVE $ ls -F threads 7b03/ 7c03/ 7d03/ 7e03/ 7f03/ 8003/ 8103/ 8203/ $ ls -F threads/7b03 basic_info/ states/ $ ls -F threads/7b03/basic_info cpu_usage policy sleep_time system_time flags run_state suspend_count user_time $ ls -F threads/7b03/states debug/ exception/ float/ thread/ $ ls -F threads/7b03/states/thread cs eax ebx edi eflags es esp gs ds ebp ecx edx eip esi fs ss $ cat threads/7b03/states/thread/esp bfffef6c
The hardware/
subdirectory contains subdirectories for several pieces of hardware on the machine, allowing certain hardware data to be read from virtual files.
In a subsequent version of procfs, the /hardware/
subdirectory was moved to appear under the /system/
subdirectory, which in turn now contains hardware/
and firmware
subdirectories.
$ ls -F /proc/hardware cpus/ lightsensor/ motionsensor/ mouse/ tpm/ $ ls -F /proc/hardware/cpus 0/ 1/ $ ls -F /proc/hardware/cpus/0 data $ cat /proc/hardware/cpus/0/data slot 0 (master), running type 7, subtype 4 6751211 ticks (user 91036 system 141938 idle 6517257 nice 980) cpu uptime 18h 45m 12s
In particular, if the machine has a sudden motion sensor or an ambient light sensor, you can retrieve "live" readings from these devices through procfs files.
$ cat /proc/hardware/motionsensor/data -7 -1 0 $ cat /proc/hardware/motionsensor/data -37 -1 0 $ cat /proc/hardware/lightsensor/data 839 895 $ cat /proc/hardware/lightsensor/data 49 893
Compatibility
Although there's no universal standard for either the hierarchy of file system objects in procfs or the contents of these objects, some implementations are rather popular because of the prevalence of their underlying operating systems. For example, FreeBSD contains linprocfs, a process file system that emulates a subset of Linux procfs. linprocfs was necessary for Linux binary emulation to work completely. This procfs implementation doesn't attempt to conform to any other implementation. For somebody with enough determination and time, it could be an interesting project to create a Linux-compatible (or Solaris-compatible, or whatever) version of MacFUSE-based procfs for Mac OS X.
A MacFUSE-Based Process File System for Mac OS X的更多相关文章
- File System Programming --- (二)
File System Basics The file systems in OS X and iOS handle the persistent storage of data files, app ...
- 让Mac OS X中的PHP支持GD
GD库已经是近乎于是现在主流PHP程序的标配了,所以也必须让Mac OS X中的PHP支持GD.在网上搜索了好多,最终按照这个方式成功实现,如何让Mac OS X支持PHP,请查看<让PHP跑在 ...
- Mac OS X 下安装使用 Docker
它依赖于 LXC(Linux Container),能从网络上获得配置好的 Linux 镜像,非常容易在隔离的系统中运行自己的应用.也因为它的底层核心是个 LXC,所以在 Mac OS X 下需要在 ...
- Mac OS 下安装mysql环境
传送门:Mac下安装与配置MySQL mac 上怎么重置mysql的root的密码? 一.下载mysql 进入官方下载地址:https://www.mysql.com/downloads/ 1.找 ...
- 5105 pa3 Distributed File System based on Quorum Protocol
1 Design document 1.1 System overview We implemented a distributed file system using a quorum based ...
- filebench - File system and storage benchmark - 模拟生成各种各样的应用的负载 - A Model Based File System Workload Generator
兼容posix 接口的文件系统中我们不仅要测试 posix 接口是否兼容.随机读,随机写,顺序读,顺序写等读写模式下的性能.我们还要测试在不同工作负载条件下的文件系统的性能的情况:Filebench ...
- 部署ActiveMQ的Share File System Master-Slave
之前在项目里用MQ是用单节点,因为业务量不大没有主从.这样风险很大,会有单点问题.新项目起来了,需要一个高可用的MQ,故研究了下AMQ的几种master-slave方式: 1.Pure Master- ...
- Supervisor: A Process Control System
Supervisor: 进程控制系统 概述:Supervisor是一个 Client/Server模式的系统,允许用户在类unix操作系统上监视和控制多个进程,或者可以说是多个程序. 它与launch ...
- Sharing The Application Tier File System in Oracle E-Business Suite Release 12.2
The most current version of this document can be obtained in My Oracle Support Knowledge Document 13 ...
随机推荐
- python字典基本操作
字典是python中五中基本数据类型之一,虽然它的赋值稍微麻烦点,但用起来真的是很方便.它用键值对来存放数据,所谓键值对,就是一个键,对应一个值,如果后面对前面的键再次赋值,第一次的值就被覆盖掉.像是 ...
- shell之seq
seq 用于生成从一个数到另一个数之间的所有整数 seq [选项]... 尾数 seq [选项]... 首数 尾数 seq [选项]... 首数 增量 尾数 例如: 1. -s 指定分隔符,默认分隔 ...
- LG2512/BZOJ1045 「HAOI2008」糖果传递 中位数
问题描述 LG2512 BZOJ1045 题解 这是一个链状问题的环状版本. 问题最终变为给定数轴上的\(n\)个点,找出一个到他们的距离之和尽量小的点,而这个点就是这些数中的中位数. 网络流24题的 ...
- goto命令
GOTO会点编程的朋友就会知道这是跳转的意思. 在批处理中允许以“:XXX”来构建一个标号,然后用GOTO XXX跳转到标号:XXX处,然后执行标号后的命令. 例: }=={} goto noparm ...
- python 面试真题
0.Python是什么? Python是一种解释型语言.但是跟C和C的衍生语言不同,Python代码在运行之前不需要编译.其他解释型语言还包括PHP和Ruby. Python是动态类型语言,指的是在声 ...
- Taro/JS/H5/小程序:纯前端解决小程序微信支付统一下单和调起支付
这个文章不会说具体0到1的代码流程,我会着重讲几个问题的解决 准备以下依赖 "md5": "^2.2.1", "xml-js": " ...
- [RN] React Native 使用 realm 数据库
React Native 使用 realm 数据库 realm 是一款专为移动 端开发的高性能数据库,其宣称自己是最快的 react-native 数据库. realm 整体的优点有这么四点: ...
- TCP/UDP通信中server和client是如何知道对方IP地址的
在TCP通信中 client是主动连接的一方,client对server的IP的地址提前已知的.如果是未知则是没办法通信的. server是在accpet返回的时候知道的,因为数据包中包含客户端的IP ...
- NTT&FFT(快速?变换)
NTT&FFT 预先知识:无 我觉得我们可以从NTT/FFT讲起? 两个其实本质相同,都是求 多项式乘积 的算法 FFT \((x,y)\)指复数,我们可以不用管它 首先我们构造单位根\(\o ...
- 收藏一份devmem源码
/* * devmem2.c: Simple program to read/write from/to any location in memory. * * Copyright (C) 2000, ...