from Solidot
大多数人为他们的硬盘分区起名时，多数只是按照字母排序的默认命名方式，顶多是再根据游戏、软件或视频等类别去分门划类。但是这些命名方式太单调乏味了。如果你维护着数量庞大的服务器，你将采用何种命名方案？美国中央情报局（CIA）是以州的名字为它的服务器命名，如阿拉斯加州；每个子网络下的客户端系统则用相关州的城市命名，如朱诺（阿拉斯加州首府）。还有一家公司以非洲国名命名，因此会出现下列有趣但毫无恶意的场景：安哥拉电脑死亡了，刚果取而代之。
cisco某客户最近被黑客盗打，一个月话费暴增十几万
他们的cme就直接裸露在外网，没有acl，没有防火墙，没有cor（话说cor用处也不大。。。）
323终端是p2p的，没有验证，黑客扫描到cme后直接用323终端发起呼叫就可以了，全部是国际长途
大家实施的时候要小心了，十几万自己可扛不住D
转cisco内部邮件如下：
主题: Regarding Toll Fraud on CME
Hello team
Recently TAC received some complaints from end users that huge amount of call charge due to call fraud.
However, these customer didn’t pay enough attention to the CME security and they even place a nearly naked CME router on the public internet (no ACL, no COR, no CDR at all). We cannot help much as they without CDR we cannot identify where the call was initiated.
Cisco do have provided a lot of documents assisting customers set up toll fraud prevention on CCO. Just type “toll fraud” and without login, you will see tons of ways to configure UC products including unity, MPE, CCM, CME, etc security to defeat the possible toll fraud.
Not sure if you have been aware of the following documents on CCO ( No login required). FYI if you have customers with CME but don’t have the CME secured please pass the document to the customers to prevent possible money loss.
http://www.cisco.com/en/US/products/sw/voicesw/ps4625/products_tech_note09186a00809dc487.shtml
www.cisco.com/en/US/prod/collateral/voicesw/ps6788/vcallcon/ps4625/uc_expresstoll.pdf
Overview
Internal vs. External Threats
Toll Restriction Tools
Direct-inward-dial
After-hours Toll Restrictions
Class of Restriction
H.323 / SIP Trunks toll fraud restrictions
Feature Restriction Tools
Transfer Pattern
Transfer-Pattern Blocked
Transfer max-length
Call Forward max-length
No Forward Local Call
Disable Auto-Registration on CME System
Cisco Unity Express Restriction Tools
Secure Cisco Unity Express: AA PSTN access
Cisco Unity Express Restriction Tables
Call Logging
Enhanced CDR
Furthermore, we can advise the customer the following with caution: customer may deny the extra payment to telco.
根据《中华人民共和国电信条例》第三十四条第二、三款“电信用户出现异常的巨额电信费用时，电信业务经营者一经发现，应当尽可能迅速告知电信用户，并采取相应的措施。前款所称巨额电信费用，是指突然出现超过电信用户此前三个月平均电信费用5倍以上的费用”的规定.
Below is a case from google that customer successfully denied the extra payment.
http://www.np.fjaic.gov.cn/yp/ReadNews.asp?NewsID=1210&BigClassName=%E6%94%BF%E5%BA%9C%E4%BF%A1%E6%81%AF%E5%85%AC%E5%BC%80&SmallClassName=%E6%B6%88%E8%B4%B9%E8%AD%A6%E7%A4%BA&SpecialID=13
漏洞的发现者showrun.lee貌似是一个中国人，他/她在blogspot上的博客有很多以中文书写。
去年下半年通告给思科了，思科的安全公告里有记载，已经放出了攻击代码。
其实VTP协议的漏洞在较早的很多文档里就有提及，但实际上需要不少的特定条件才能攻击这个漏洞，交换机必须工作在Server/Client模式，相应的接口要是Trunk端口，要知道VTP域的密码等，但是网络安全往往是不怕一万就怕万一，侥幸心理要不得，指不定在哪个环节就天时地利人和的被人利用了。
这个攻击代码实施的结果会造成交换机重启。
建议VTP还是配置为Transparent模式比较好，虽然在大规模部署VLAN的情况下有所不便，但习惯了之后既可带来思路的清晰，也能避免不必要的安全隐患。
The books in the Solaris System Administration series are designed to teach practicing and aspiring system administrators how to manage Solaris in an IT environment. Solaris Security Essentials covers the main security features in the Solaris operating system, including roles and privileges, cryptographic services, network security, auditing, and Solaris Trusted Extensions.
大意是：
从2009.2.1开始，CCIE RS考生在传统的实验考试之外，还需要回答4-5个开放式的问题，这些问题都在实验大纲的范围之内。考试并没有因此增加难度，准备充分的考生应该能够回答这些问题。考试的时间仍然是8个小时。考生需要同时通过实验和回答问题两个部分才能通过考试获得认证。其他的类型的CCIE考试会在明年实施同样的变化。
从2009.2.17开始，CCIE笔试会有两个变化。首先是考生必须回答完一道问题后再往后继续，不允许稍后往回退；其次，成绩报告有一个更新，总的考试分数和考试通过分数以一个范围报告，范围在300-1000内。
Effective February 1, 2009, Cisco will introduce a new type of question format to CCIE Routing and Switching lab exams. In addition to the live configuration scenarios, candidates will be asked a series of four or five open-ended questions, drawn from a pool of questions based on the material covered on the lab blueprint. No new topics are being added. The exams are not been increased in difficulty and the well-prepared candidate should have no trouble answering the questions. The length of the exam will remain eight hours. Candidates will need to achieve a passing score on both the open-ended questions and the lab portion in order to pass the lab and become certified. Other CCIE tracks will change over the next year, with exact dates announced in advance.
Effective February 17th, 2009, candidates will also see two other changes in CCIE written exams. First, candidates will now be required to answer each question before moving on to the next question; candidates will no longer be allowed to skip a question and come back to it at a later time. Second, there will be an update to the score report. The overall exam score and the exam passing score will now be reported as a scaled score, on a scale from 300-1000. This change will not affect the difficulty of the current set of exams and will assure CCIE written exams will be consistent with Cisco’s other career certification exams.
今天被一个SB房地产的广告恶心坏了。
一大幅楼盘照片的右上角放了一张张爱玲的旧上海滩那种艺术照，竖着写一排字：民国风格，离张爱玲很近。
丫肯定以为这句是整个创意的点睛之笔。
整一个灰砖灰瓦就是民国风格，整一个小户型迎合小资市场就是张爱玲，全南京都是SB1912，全上海都是SB新天地，我不得不说，全中国的房地产暴发户都是脑残。
Donews的文字审查真TM烦，这篇文字贴不了了，只能贴个URL在这儿。
http://cuddletech.com/blog/pivot/entry.php?id=965
In Part 1 we discussed core analysis in general and some basic mdb commands for high level investigation. When you dig deeper things can get confusing and complex because everything is referenced by address. This is where the Solaris Crash Analysis Tool comes in.
Solaris CAT has been around for a long time, but only as of version 5.0 released on June 18th of this year has it been available for Solaris X86/X64. You can find the Solaris CAT 5.0 Release Notes here.
To get started, download CAT 5.0, uncompress and install the package:
# bunzip2 SUNWscat5.0-GA-i386.pkg.bz2# pkgadd -G -d ./SUNWscat5.0-GA-i386.pkg The following packages are available: 1 SUNWscat Solaris Crash Analysis Tool (5.0 GA SV4622M) (i386) 5.0 Select package(s) you wish to process (or 'all' to processall packages). (default: all) [?,??,q]: 1 Processing package instancefrom Solaris Crash Analysis Tool (5.0 GA SV4622M)(i386) 5.0...
The package will, by default, install into /opt/SUNWscat. There are two binaries we’re really interested in, found in the bin/ directory: scat and blast. The scat tool is the CLI interface to Solaris CAT and provides a shell which is a human friendly re-implementation of mdb (no "::" prefixing commands, etc.) The blast tool is a really nice Java GUI interface to the CLI which adds a lot of "just click here" functionality and is excellent for testing and playing around. I highly recommend you point your browser at /opt/SUNWscat/docs/index.html, which includes some minimal but extremely useful HTML documentation.
Authors note: I’m resisting a "scat" joke with amazing strength. Seriously… resisting…. so…. hard….
We’ll focus on the CLI here. Invocation is a little unusual; add /opt/SUNWscat/bin to your path and then change to the directory containing your dumps (usual /var/crash/hostname/), for the .0 dumps use "scat 0", for the .1 dumps use "scat 1", and so on. You’ll fine the "online help" within the CLI exceptional, lets look:
# export PATH=$PATH:/opt/SUNWscat/bin# cd /var/crash/ev2-r01-s10/# ls -ltotal 14205330-rw-r--r-- 1 root root 2 Aug 25 07:49 bounds-rw-r--r-- 1 root root 1444762 Aug 25 07:43 unix.0-rw-r--r-- 1 root root 7268106240 Aug 25 07:49 vmcore.0# scat 0 Solaris[TM] CAT 5.0 for Solaris 11 64-bit x86 SV4622M, Jul 3 2008 Copyright © 2008 Sun Microsystems, Inc. All rights reserved. Use is subject to license terms. Feedback regarding the tool should be sent to SolarisCAT_Feedback@Sun.COM Visit the Solaris CAT blog at http://blogs.sun.com/SolarisCAT opening unix.0 vmcore.0 ...dumphdr...symtab...core...doneloading core data: modules...symbols...ctftype: unknown type struct panic_trap_infoCTF...done core file: /var/crash/xxxxxxxx/vmcore.0user: Super-User (root:0)release: 5.11 (64-bit)version: snv_67machine: i86pcnode name: xxxxxxxxxxxxxxxxxxsystem type: i86pchostid: xxxxxxxxdump_conflags: 0x10000 (DUMP_KERNEL) on /dev/dsk/c0t0d0s1(24.0G)time of crash: Mon Aug 25 07:41:00 GMT 2008 (core is 13 days old)age of system: 91 days 22 hours 49 minutes 50.97 secondspanic CPU: 1 (8 CPUs, 31.9G memory)panic string: page_free pp=ffffff0007243bd8, pfn=11228e, lckcnt=0, cowcnt=0 slckcnt = 0 sanity checks: settings...vmem...WARNING: FSS thread 0xffffff097d1e3400 on CPU2 using 99%CPUWARNING: FSS thread 0xffffff09fddbab40 on CPU3 using 99%CPUsysent...clock...misc...NOTE: system has 54 non-global zonesdoneSolarisCAT(vmcore.0/11X)>
When CAT is unleashed on a dump several "sanity checks" are run which can point out glaring known issues. There is an HTML document in the docs/ directory which outlines all the various sanity checks. These checks alone make CAT a must-have tool! Sanity check output will come in two varieties, "WARNING" which indicates something out of whack that may have been the cause or contributor to the crash, and "NOTE" which is unlikely the cause but of interest. We can see in the example above two warnings telling me that 2 threads were consuming 99% of a CPU… thats handy! It also notes that I’m running 54 zones.
The available commands a broken down into categories which you can see using the "help" command. The first group are for "Initial Investigation:" and include: analyze, coreinfo, msgbuf, panic, stack, stat, and toolinfo. Lets look at the "analyze" commands output:
SolarisCAT(vmcore.0/11X)> analyze core file: /var/crash/xxxxxx/vmcore.0user: Super-User (root:0)release: 5.11 (64-bit)version: snv_67machine: i86pcnode name: xxxxxxsystem type: i86pchostid: xxxxxdump_conflags: 0x10000 (DUMP_KERNEL) on /dev/dsk/c0t0d0s1(24.0G)time of crash: Mon Aug 25 07:41:00 GMT 2008 (core is 13 days old)age of system: 91 days 22 hours 49 minutes 50.97 secondspanic CPU: 1 (8 CPUs, 31.9G memory)panic string: page_free pp=ffffff0007243bd8, pfn=11228e, lckcnt=0, cowcnt=0 slckcnt = 0 ==== panic thread: 0xfffffffef4ce5dc0 ==== CPU: 1 ======== panic user (LWP_SYS) thread: 0xfffffffef4ce5dc0 PID: 10156 on CPU: 1 ====cmd: /opt/local/sbin/httpd -k startt_procp: 0xffffffff06595e50 p_as: 0xffffffff093490e0 size: 47374336 RSS: 3125248 hat: 0xffffffff092a9480 cpuset: 1 zone: address translation failed for zone_name addr: 8 bytes @ 0x3 t_stk: 0xffffff00486bcf10 sp: 0xffffff00486bc880 t_stkbase: 0xffffff00486b8000t_pri: 3(FSS) pctcpu: 0.380035t_lwp: 0xfffffffefe61ab60 lwp_regs: 0xffffff00486bcf10 mstate: LMS_SYSTEM ms_prev: LMS_SYSTEM ms_state_start: 2 minutes 31.229022230 seconds earlier ms_start: 2 minutes 31.343582414 seconds earlierpsrset: 0 last CPU: 1 idle: 0 ticks (0 seconds)start: Mon Aug 25 07:41:00 2008age: 0 seconds (0 seconds)syscall: #131 memcntl(, 0x0) ()tstate: TS_ONPROC - thread is being run on a processortflg: T_PANIC - thread initiated a system panic T_DFLTSTK - stack is default sizetpflg: TP_MSACCT - collect micro-state accounting informationtsched: TS_LOAD - thread is in memory TS_DONT_SWAP - thread/LWP should not be swapped TS_RUNQMATCHpflag: SMSACCT - process is keeping micro-state accounting SMSFORK - child inherits micro-state accounting pc: unix:vpanic_common+0x13b: addq $0xf0,%rsp unix:vpanic_common+0x13b()unix:panic+0x9c()unix:page_free+0x22e()unix:page_destroy+0x100()genunix:fs_dispose+0x2e()genunix:fop_dispose+0xdc()genunix:pvn_getdirty+0x1f0()zfs:zfs_putpage+0x129()genunix:fop_putpage+0x65()genunix:segvn_sync+0x39f()genunix:as_ctl+0x1f2()genunix:memcntl+0x709()unix:_syscall32_save+0xbf()-- switch to user thread's user stack --
This output provides a vast array of useful details, including:
As noted in Part 1, what most people are really looking for when doing core analysis is to determine which application was responsable, and this output provides that data in great clarity. Lets dig into it a bit more explicitly… based on the above "analyze" output we can see that….
For many administrators this might be as much as you wanted to know, right there. But lets look at a couple more commands.
You’ll recall that during the sanity checks at startup it noted 2 threads consuming full CPU’s. We can feed the thread address to the "thread" command to get details on them:
SolarisCAT(vmcore.0/11X)> thread 0xffffff097d1e3400==== user (LWP_SYS) thread: 0xffffff097d1e3400 PID: 27446 on CPU: 2 ====cmd: nano svn-commit.tmpt_procp: 0xffffffff2e908ab0 p_as: 0xffffffff10402ee0 size: 2772992 RSS: 1642496 hat: 0xffffffff102f6b48 cpuset: 2 zone: address translation failed for zone_name addr: 8 bytes @ 0x2 t_stk: 0xffffff004e47ef10 sp: 0xffffff003d3fcf08 t_stkbase: 0xffffff004e47a000t_pri: 26(FSS) pctcpu: 99.306175t_lwp: 0xffffffff202a78b0 lwp_regs: 0xffffff004e47ef10 mstate: LMS_SYSTEM ms_prev: LMS_USER ms_state_start: 2 minutes 31.228983791 seconds earlier ms_start: 39 days 19 hours 11 minutes 8.989252296 seconds earlierpsrset: 0 last CPU: 2 idle: 9 ticks (0.09 seconds)start: Wed Jul 16 12:30:07 2008age: 3438653 seconds (39 days 19 hours 10 minutes 53 seconds)syscall: #98 sigaction(, 0x0) ()tstate: TS_ONPROC - thread is being run on a processortflg: T_DFLTSTK - stack is default sizetpflg: TP_TWAIT - wait to be freed by lwp_wait TP_MSACCT - collect micro-state accounting informationtsched: TS_LOAD - thread is in memory TS_DONT_SWAP - thread/LWP should not be swapped TS_RUNQMATCHpflag: SMSACCT - process is keeping micro-state accounting SMSFORK - child inherits micro-state accounting pc: unix:panic_idle+0x23: jmp -0x2 (unix:panic_idle+0x23) unix:panic_idle+0x23()0xffffff003d3fcf60()-- error reading next frame @ 0x0 --
So using the "thread" command we can get full granularity on a given thread. In fact, using the "tlist" command you can dump this information for every thread on the system at the time of crash.
Another nifty command is "tunables". This will display the "current value" (at time of the dump) and the default value. If someone’s been experimenting on the production systems this will clue you in.
SolarisCAT(vmcore.0/11X)> tunables Tunable Name Current Default Value Units Description Value physmem 8386375 * pages Physical memory installed in system. freemem 376628 * pages Available memory. avefree 338943 * pages Average free memory in the last 30 seconds.........
Using the "dispq" command we can look at the dispatch queues (run queue). This answers "what other processes were running on CPU at the time of the crash", again, using the thread address we can dig into them with "thread":
SolarisCAT(vmcore.0/11X)> dispq CPU thread pri PID cmd 0 @ 0xfffffffffbc26bb0 0xffffff003d005c80 -1 (idle) pri 60 -=> 0xffffff004337dc80 60 0 sched 1 @ 0xfffffffec6634000 P 0xfffffffef4ce5dc0 P 3 10156 /opt/local/sbin/httpd -k start 2 @ 0xfffffffec662f000 0xffffff097d1e3400 26 27446 nano svn-commit.tmp 3 @ 0xfffffffec66f4800 0xffffff09fddbab40 25 21329 java -jar xxxxx.jar --ui=console 4 @ 0xfffffffec66ea800 0xffffff003d414c80 -1 (idle) pri 60 -=> 0xffffff0048b12c80 60 0 sched 5 @ 0xfffffffec6770800 0xffffff003d4b0c80 -1 (idle) 6 @ 0xfffffffec6770000 0xffffff003d53bc80 -1 (idle) 7 @ 0xfffffffec6762000 0xffffff003d58fc80 -1 (idle)
part thread pri PID cmd 0 @ 0xfffffffffbc4eef0
There are far too many to go through in a blog entry… but lets look at my personal favorite, "zfs". The "zfs" command can show us the pool(s), their configuration, read/write/checksum/error stats, and even ARC stats!
SolarisCAT(vmcore.0/11X)> zfs -eZFS spa @ 0xfffffffec6c21540 Pool name: zones State: ACTIVE VDEV Address State Aux Description 0xfffffffec0a9e040 FAULTED - root
READ WRITE FREE CLAIM IOCTL OPS 0 0 0 0 0 BYTES 0 0 0 0 0
EREAD 0 EWRITE 0 ECKSUM 0
VDEV Address State Aux Description 0xfffffffec0a9eac0 FAULTED - /dev/dsk/c0t1d0s0
READ WRITE FREE CLAIM IOCTL OPS 74356305 578263155 0 0 0 BYTES 757G 10.4T 0 0 0
EREAD 0 EWRITE 0 ECKSUM 0SolarisCAT(vmcore.0/11X)> zfs arc
ARC (Adaptive Replacement Cache) Stats:
hits 77708247444 misses 1930348 demand_data_hits 74303514929 demand_data_misses 1325511 demand_metadata_hits 620388795 demand_metadata_misses 160708 prefetch_data_hits 1361651307....
I hope this helps you get an idea of how easy it is to really dig deeply into your core dumps using Solaris CAT to hide the oddities of mdb from you. Its a powerful and robust tool, and I’m glad that we have it.
Happy dump divin’! You’ll be amazed how much you’ll learn about your system.
OSNews recently posted the following: Linux Foundation’s Zemlin Declares Solaris All But Dead, which in turn links to the InfoWorld article: Is Sun Solaris on its deathbed?
Mr. Zemlin hammers on the fact that Solaris has only two redeeming qualities, ZFS and DTrace, which he degrades as "minor features". Both these technologies are revolutionary advances in computing and storage… "minor" my ass.
"That’s literally like noticing the view from a third-story building as it burns to the ground." And the Linux community is working on rival technology, Zemlin adds.
That quote is jibberish. And asserting that the Linux community is working on rival technologies? Ummmm. SystemTap is a horrible failure (following image from OSCon) and there is no rival for ZFS.
"The only people I hear talk about DTrace [Solaris's technology for assessing program and OS behaviours] and ZFS [the Zettabyte File System] as competitive features [are] Sun Microsystems sales representatives. It’s not something I believe is impacting the market in any way,"
Its quotes like this one that show that Mr. Zemlin is living in a cave… I mean, he is the Executive Director of the Linux Foundation. Honestly, do you expect him to say anything else regardless of validity? He’s surrounded by Linux users, vendors, and advocates based on his station. Users excited about (Open)Solaris aren’t likely going to be hanging out with this guy; rather, he’s going to hear from Linux users interested in the excitement around ZFS, DTrace, etc, and ask why they can’t be provided to Linux. Of course, its these very users we’d like to stop looking and start trying OpenSolaris itself.
"Customers are pretty aware that Unix is a more expensive legacy architecture. They continue to support it because they don’t want to change their legacy apps over to a new platform because of the costs," Zemlin said. "But they know now they eventually need to do it because Unix just doesn’t have the combined might of all the different organizations and individuals that are developing [for] Linux."
Now, lets get some things straight here. OpenSolaris is free, as in both beer and freedom. It runs beautifully on non-Sun hardware, and in fact I’d say at the present, it runs better on Dell/Supermicro/etc. hardware than Sun Fire servers(which has got to be turned around soon). It binary compatible, supports legacy SPARC, cutting edge Niagara SPARC, as well as X86 32 and 64bit, in fact I’d say that Solaris handles intermixing 32/64 bit applications significantly better than Linux.
Linux still has huge problems with quality support offerings. Sun offers a significantly higher quality of support because so much is in house.
End-to-end, Solaris/OpenSolaris is drastically cheaper than Linux. This doesn’t even take into consideration the savings you get by using ZFS rather than buying high cost NAS solutions for your Linux infrastructure.
Given Sun’s own Linux support on its Sparc and x86 servers, Zemlin suggests that it should make ZFS and DTrace available under a Linux-compatible license.
And there it is… Linux advocates bash Sun & Solaris right, left and center, but after it all they just want to rape the OS. Not share, not collaborate… rape. I have never been an advocate of GPLv2 release of the code for this explicit reason. Its like a bad date, she smiles and winks a little, calls you names but takes you home anyway, and in the morning your wallet is gone. Is it possible that I’m wrong? Perhaps, but once you’ve gone down that road you can’t turn back and I’m afraid that Sun will give up the advantages that it has, throwing away thousands of man hours of development effort and end up becoming a SGI, forever doomed to embrace Linux or else.
Now, there was one very important point made in the article, but not by Mr. Zemlin of course:
One company that is moving from Solaris to Linux is Sesame Workshop, famous for TV shows such as Sesame Street. A key reason is that more people are available to support Linux than Solaris, says Noah Broadwater, vice president of information services at Sesame Workshop. "I honestly have one person who is certified on Solaris. I have four people who are certified on Linux," Broadwater said.
First, lets point out that the customer is moving from an existing Solaris SPARC install. This is likely an infrastructure that is 3+ years old and perhaps like the vast bulk of Sun customers still running on Solaris 8; regardless, there is a very big difference between moving from an older SPARC install to X86 Linux versus moving from Solaris/X86 to Linux/X86.
The point about qualified people is a very big problem! I’ve been extremely interested in working toward solving it but had little traction thus far. There are too few of us blogging, too few learning resources, and certification is expensive ($300*2 for examines, making certification $600 assuming no training and that you pass the first time). On top of it all, many of us, new and old alike, aren’t convinced that certifications are actually useful in instructing new admins.
I’d go so far as to suggest that Sun Ed. and the community (SysAdmin OpenSolaris Community Group) should work together on improving the certification criteria and tests, reducing it from 2 to 1 test, and then dropping the cost under $100.
So long as we lack skilled and interested SysAdmin’s in the field we’ll continue to deal with Linux advocates bashing Solaris with their hand out.
We need:
Jeff Doyle 2006年离开Juniper之后，开了一家貌似是咨询性质的公司“Jeff Doyle and Associates"，还在Network World的网站上开了一个博客Jeff Doyle on IP Routing，内容还是他老大擅长的IP Routing，IPv6，MPLS。
好像他有意写一本关于MPLS的书，但担心市场上已有的MPLS书籍太多，还在blog上征求读者的意见，当然老大拥泵众多，冲着《Routing TCP/IP》，大多数人都非常期待Jeff在MPLS上也发挥一把。
最近几篇，Jeff开始写关于JUNOS的文章，貌似是一个系列，这恐怕跟Juniper近期在全球范围内借JUNOS诞生十周年的机会，大张旗鼓的宣传有关，Jeff既然running a company，当然也就是生意人了，人在江湖，老东家有求自然要应的，价格肯定不会亏待滴，呵呵。
很久不读Routing TCP/IP，现在看Jeff的文章，还是那种平实、容易亲近的风格，值得一看。