I don't know any Linux distros that use anything but a Linux kernel. Actually, with Windows, you have a choice between the 9x kernel and the NT kernel. It's also not like I can't install any alternatives on Windows. I use Winamp, Mozilla, Sysinternals's process explorer, and several cygwin tools instead of their MS alternatives. So what if they aren't listed by MS, they weren't very hard to find. Microsoft only provides one choice for each, but at least they are easy to use. When I first started using Linux, it wasn't helpful to know that there are 20 text editors I available; I don't know which to use, and many, like vim and emacs require a lengthy tutorial to do even simple things. No obvious choice as to which is the easiest to get started with. Once I got X windows working, the window environs were easier, but I had to edit XF86's config file to get that far.
Re:Blame windows it already looks like Gnome
on
Gnome.org Compromised?
·
· Score: 2, Insightful
First it's "Microsoft bundles too many things with Windows" and now it's "You can't compare Windows to a Linux distro because it only has 3 packages: kernel/GUI/browser"
He does have a point though; when was the last time Microsoft.com was hacked, or down? They run Windows and even IIS for their webserver. They must be doing something right.
Collectively, all the windows opened by Mozilla run under the same process, so if you have even 1 open, the browser remains active.
I keep it active for some of the same reasons to have multiple desktops. I am in the middle of going through some page, or I see something I want to look at later (only once; not enough for a bookmark) then I will keep that page open for days, until I finish.
How about not even needing the mouse or needing to control the cursor at all; as in having keyboard shortcuts that can do everything. Properly written Windows programs (everything from MS that I have used) do not require the use of the mouse cursor for anything, since at least Win3.1.
To localize a traditional Windows app, you create a seperate 'string resource' file that gets compiled into the program. The easy part about this is that there are no extra files to locate; the program already has a single language file embedded in itself. Creating new languages is (almost) as easy as translating the language file and recompiling with another language. A big problem with this is that to release a patch, you need a seperate one for each language. This also makes it a pain for anyone but Microsoft to translate MS programs, since you need to recompile for each language.
Yeah! The BIOS text interface should be enough for anyone. Those extended ASCII characters are really... WOW! I can't believe Megazeux requires a 186 and EGA!
I'm impressed. That's one of the great things about Linux and free/open sofware in general. You can always update, for free, and only the components you want. I don't have any systems that have been running a single install for that long, in Linux because I haven't been using it that long, and Windows because 'upgrade install' usually sucks. But I also haven't reinstalled Windows because it had become damaged somehow since win95.
Threads in NT are not associated with kernel or user mode. Any thread can enter either. If you turn on the 'show kernel times' option in task manager, you can see how much time is spent in each mode. All waiting threads (probably almost all of them) wait on some synchonization primitive (like an event or queue) in kernel mode. Until recently, the Linux kernel has been such that only one kernel thread can be active per cpu at one time, and never preempted. Recent versions (2.5, 2.6) are chaning that. The NT kernel has always been preemptable, reentrant and multithreaded. I'm not sure about Solaris, but I understand it has excellent multithreading.
And I agree with you; this is another empty article that says Linux is better, big suprise seeing it on/.
Actually, the design of the NT kernel and system are great. NT's native api is clean and consistent. As you go up from there though, it gets bad. Win32 is a mess but compatible (enough) and usable. Some things, like IIS and Exchange, however, have plenty of problems and they aren't caused by the basic system.
And that's half the problem; the misconception that it is always Windows's fault when something goes wrong, but it's the user's fault when the same thing happens on Linux. Using Windows perfectly is at least as hard as Linux. I don't have this supposed plague of unfixable Windows problems I always hear about on any of my computers.
What I think they meant about address space is if something depended using some fixed address that would be wrong if the kernel memory split changed, which is bogus. First, if something uses a fixed address it's already incompatible with 64 bit addresses, and designed badly. Second, NT has had the/3GB switch since v4 which makes private memory 3gb instead of 2. The balance set manager runs on demand when free pages are needed, not every second. I don't know where that is from, let alone how it could be abused.
My laptop runs Windows and it hasn't crashed or been restarted in 10+ months (since I upgraded the video driver), and hasn't slowed down. I play games, browse the internet (/w Mozilla), develop programs, keep several things running concurrently... Quality drivers are essential.
Actually, WinNT has always been able to do things like that. It's the win32 layer that gets in the way and insists on drive letters. The C: drive is really a symlink in the object manager to \Device\HarddiskVolume1. \Device serves the same purpose as/dev on unix. Under the hood, NT and unix have more in common. If you have a winnt install handy, run the winobj tool from sysinternals.com.
WinXP and later have 'side-by-side' components; different versions of the same library that a program can request.
The DLL path for Windows is configurable too. First it searches the list of known system libraries (about 10 used to avoid spoofing) then the program's directory, then the system directories, then environment path. You can use PATH in a similar way to LD_LIBRARY_PATH, although it isn't exactlly the same.
The term non-threaded they use in the article is a misnomer. Compared to Linux, there is no distinction between a kernel thread and a user thread. Any thread can enter kernel mode or user mode. Locking for things like interrupts are abstracted by interrupt request levels (IRQLs) in the NT kernel. Here's an example: Let's say that your network adapter just recieved a packet. The NIC generates an interrupt. If the interrupt isn't masked, the CPU interrputs what it is doing to handle the interrupt. The HAL gets the interrupt and sets the IRQL to the level of the interrupt, 9. All lower interrupts (including the scheduler) are masked. Then control goes to the NIC's driver. A well written driver will do a minimum of processing and schedule a deffered procedure call (DPC). The DPC is queued and executed later when the worker thread (in the system process) gets scheduled. Then the kernel is running at IRQL passive, and can be interrupted by anything (the scheduler, another interrupt), the same as all normal (user) threads. The kernel uses IRQLs to abstract interrupts and tries to spend a minimum of time at an elevated IRQL. The NT kernel is multithreaded, preemptable and reentrant, things the Linux kernel has only recently implemented (2.5).
Runas doesn't require the Administrator password; only the password of the account you want to run as. As for setuid; I can't think of any built-in way to do that. The SUD tool I mentioned does. First you create an encrypted 'cfg' file that stores a user, password and a specific program. Then you give any users you want to run that program read access to the cfg file so they can run the program without knowing the target account's password. Just make a shortcut like "su -C c:\prog\sudothis.cfg". It is all explained in the included readme file.
Here's what you do about badly written programs that require too many priveleges: Create a runas script to run only those programs as admin. Like sudo in unix. Or the tool SUD from http://www.loa.espci.fr/winnt/sud/sud.htm. A proper su command for WinNT. Along the same lines, locking down IE and OE with a seperate restricted user is a good idea too.
1. Yes. I am trying to say that a poor design decision in Outlook or Outlook Express does not mean a basic design flaw in Windows (as the parent imples).
2. Win2k gold is 3 years old now; there little excuse to not download the patch for OE by now... It's not on AutoUpdate?
3. Windows 9x isn't secure. If cost is the issue, I don't know what to say. Mabye you can find some NT 3.51 licences cheap somewhere, since it is EOL. Even NT 3.51 is much more secure than 9x. As for applications that require more privileges than they should (like installing drivers), Windows is plagued with those programs, and I personally blame the developers of the applications. As a workaround, you could just run only those programs as admin with a runas script or something. Or run as admin by default and risky programs (like OE and IE) as a lesser user. (if you trust your users)
4. It should be default. I think Microsoft has some crappy defaults too. Still, it is easy enough to implement yourself (in this case).
5. You are right, those can be just as good. The parent implied that Windows(NT) couldn't be secure because you can't mount readonly or noexec. I wanted to show that you can use ACLs to create the same effect. (Personally, I perfer ACLs because of the extra granularity, esp for special cases.)
In NT4, most of the win32 subsystem, including the graphcis system was moved into win32k.sys. It runs in kernel mode but is definately not part of the kernel (ntoskrnl.exe). It was moved there to cut down on the context switching overhead for out-of-process function calls to the win32 subsystem (in csrss.exe and csrsrv.dll). You can run the kernel perfectly fine without loading win32; like the first phase of setup or the recovery console.
Most of ntdll.dll's entry points (all the Nt* functions) simply load a kernel function ID into AX and use a software interrupt to enter kernel mode. It represents NT's system call interface. It has 285 of those in xpsp1. The first release of NT(3.1) has 178.
How this compares to send me a fscking html-with-vbscript that will be executed while in the preview pane of Outlook Express and downloads another executable that has the power to install itself as a device driver and run in kernel mode?????
This is a problem with OE, not Windows, and it has been fixed for a long time. Programs can only install device drivers when priveleged. Do you run your mail client as root? Normal users can't install drivers. Also, Outlook and other Internet Explorer based programs put downloaded files, like attachments, into the current user's profile under "Temporary Internet Files". You could always deny everyone execute access in that directory to get the same effect as noexec. You can also deny users from writing and creating files by the use of ACLs for an effect like mounting read-only. If anything, the security model of WinNT is more flexible than a standard Linux system.
In languages like Python that can treat code as data, the code is not stored in the binary form the CPU executes, but as a list of instructions for the language's interpreter. So, as far as the CPU is concerned, those pieces of code are really data. If the code was compiled beforehand, then the reference to the code structure will be a pointer to the actual function, stored with the rest of code. If the language has a JIT (just-in-time compiler), it has a risk of being broken, as mentioned earlier with the.NET JIT. It will have to be aware of the NX flag and set it explicitlly.
Slashdot Mirror
I don't know any Linux distros that use anything but a Linux kernel. Actually, with Windows, you have a choice between the 9x kernel and the NT kernel.
It's also not like I can't install any alternatives on Windows. I use Winamp, Mozilla, Sysinternals's process explorer, and several cygwin tools instead of their MS alternatives. So what if they aren't listed by MS, they weren't very hard to find. Microsoft only provides one choice for each, but at least they are easy to use. When I first started using Linux, it wasn't helpful to know that there are 20 text editors I available; I don't know which to use, and many, like vim and emacs require a lengthy tutorial to do even simple things. No obvious choice as to which is the easiest to get started with.
Once I got X windows working, the window environs were easier, but I had to edit XF86's config file to get that far.
First it's "Microsoft bundles too many things with Windows" and now it's "You can't compare Windows to a Linux distro because it only has 3 packages: kernel/GUI/browser"
Pick one.
He does have a point though; when was the last time Microsoft.com was hacked, or down? They run Windows and even IIS for their webserver. They must be doing something right.
Collectively, all the windows opened by Mozilla run under the same process, so if you have even 1 open, the browser remains active.
I keep it active for some of the same reasons to have multiple desktops. I am in the middle of going through some page, or I see something I want to look at later (only once; not enough for a bookmark) then I will keep that page open for days, until I finish.
How about not even needing the mouse or needing to control the cursor at all; as in having keyboard shortcuts that can do everything.
Properly written Windows programs (everything from MS that I have used) do not require the use of the mouse cursor for anything, since at least Win3.1.
To localize a traditional Windows app, you create a seperate 'string resource' file that gets compiled into the program. The easy part about this is that there are no extra files to locate; the program already has a single language file embedded in itself. Creating new languages is (almost) as easy as translating the language file and recompiling with another language. A big problem with this is that to release a patch, you need a seperate one for each language. This also makes it a pain for anyone but Microsoft to translate MS programs, since you need to recompile for each language.
Yeah! The BIOS text interface should be enough for anyone. Those extended ASCII characters are really... WOW! I can't believe Megazeux requires a 186 and EGA!
I'm impressed. That's one of the great things about Linux and free/open sofware in general. You can always update, for free, and only the components you want.
I don't have any systems that have been running a single install for that long, in Linux because I haven't been using it that long, and Windows because 'upgrade install' usually sucks. But I also haven't reinstalled Windows because it had become damaged somehow since win95.
Threads in NT are not associated with kernel or user mode. Any thread can enter either. If you turn on the 'show kernel times' option in task manager, you can see how much time is spent in each mode. All waiting threads (probably almost all of them) wait on some synchonization primitive (like an event or queue) in kernel mode. Until recently, the Linux kernel has been such that only one kernel thread can be active per cpu at one time, and never preempted. Recent versions (2.5, 2.6) are chaning that. The NT kernel has always been preemptable, reentrant and multithreaded. I'm not sure about Solaris, but I understand it has excellent multithreading.
/.
And I agree with you; this is another empty article that says Linux is better, big suprise seeing it on
Actually, the design of the NT kernel and system are great. NT's native api is clean and consistent. As you go up from there though, it gets bad. Win32 is a mess but compatible (enough) and usable. Some things, like IIS and Exchange, however, have plenty of problems and they aren't caused by the basic system.
And that's half the problem; the misconception that it is always Windows's fault when something goes wrong, but it's the user's fault when the same thing happens on Linux. Using Windows perfectly is at least as hard as Linux. I don't have this supposed plague of unfixable Windows problems I always hear about on any of my computers.
What I think they meant about address space is if something depended using some fixed address that would be wrong if the kernel memory split changed, which is bogus. First, if something uses a fixed address it's already incompatible with 64 bit addresses, and designed badly. Second, NT has had the /3GB switch since v4 which makes private memory 3gb instead of 2.
The balance set manager runs on demand when free pages are needed, not every second. I don't know where that is from, let alone how it could be abused.
My laptop runs Windows and it hasn't crashed or been restarted in 10+ months (since I upgraded the video driver), and hasn't slowed down. I play games, browse the internet (/w Mozilla), develop programs, keep several things running concurrently... Quality drivers are essential.
Actually, WinNT has always been able to do things like that. It's the win32 layer that gets in the way and insists on drive letters. The C: drive is really a symlink in the object manager to \Device\HarddiskVolume1. \Device serves the same purpose as /dev on unix.
Under the hood, NT and unix have more in common. If you have a winnt install handy, run the winobj tool from sysinternals.com.
WinXP and later have 'side-by-side' components; different versions of the same library that a program can request.
The DLL path for Windows is configurable too. First it searches the list of known system libraries (about 10 used to avoid spoofing) then the program's directory, then the system directories, then environment path. You can use PATH in a similar way to LD_LIBRARY_PATH, although it isn't exactlly the same.
The term non-threaded they use in the article is a misnomer. Compared to Linux, there is no distinction between a kernel thread and a user thread. Any thread can enter kernel mode or user mode. Locking for things like interrupts are abstracted by interrupt request levels (IRQLs) in the NT kernel.
Here's an example:
Let's say that your network adapter just recieved a packet. The NIC generates an interrupt. If the interrupt isn't masked, the CPU interrputs what it is doing to handle the interrupt. The HAL gets the interrupt and sets the IRQL to the level of the interrupt, 9. All lower interrupts (including the scheduler) are masked. Then control goes to the NIC's driver. A well written driver will do a minimum of processing and schedule a deffered procedure call (DPC). The DPC is queued and executed later when the worker thread (in the system process) gets scheduled. Then the kernel is running at IRQL passive, and can be interrupted by anything (the scheduler, another interrupt), the same as all normal (user) threads.
The kernel uses IRQLs to abstract interrupts and tries to spend a minimum of time at an elevated IRQL. The NT kernel is multithreaded, preemptable and reentrant, things the Linux kernel has only recently implemented (2.5).
Runas doesn't require the Administrator password; only the password of the account you want to run as.
As for setuid; I can't think of any built-in way to do that. The SUD tool I mentioned does. First you create an encrypted 'cfg' file that stores a user, password and a specific program. Then you give any users you want to run that program read access to the cfg file so they can run the program without knowing the target account's password. Just make a shortcut like "su -C c:\prog\sudothis.cfg". It is all explained in the included readme file.
Here's what you do about badly written programs that require too many priveleges: Create a runas script to run only those programs as admin. Like sudo in unix. Or the tool SUD from http://www.loa.espci.fr/winnt/sud/sud.htm. A proper su command for WinNT.
Along the same lines, locking down IE and OE with a seperate restricted user is a good idea too.
1. Yes. I am trying to say that a poor design decision in Outlook or Outlook Express does not mean a basic design flaw in Windows (as the parent imples).
2. Win2k gold is 3 years old now; there little excuse to not download the patch for OE by now... It's not on AutoUpdate?
3. Windows 9x isn't secure. If cost is the issue, I don't know what to say. Mabye you can find some NT 3.51 licences cheap somewhere, since it is EOL. Even NT 3.51 is much more secure than 9x.
As for applications that require more privileges than they should (like installing drivers), Windows is plagued with those programs, and I personally blame the developers of the applications. As a workaround, you could just run only those programs as admin with a runas script or something. Or run as admin by default and risky programs (like OE and IE) as a lesser user. (if you trust your users)
4. It should be default. I think Microsoft has some crappy defaults too. Still, it is easy enough to implement yourself (in this case).
5. You are right, those can be just as good. The parent implied that Windows(NT) couldn't be secure because you can't mount readonly or noexec. I wanted to show that you can use ACLs to create the same effect. (Personally, I perfer ACLs because of the extra granularity, esp for special cases.)
In NT4, most of the win32 subsystem, including the graphcis system was moved into win32k.sys. It runs in kernel mode but is definately not part of the kernel (ntoskrnl.exe). It was moved there to cut down on the context switching overhead for out-of-process function calls to the win32 subsystem (in csrss.exe and csrsrv.dll). You can run the kernel perfectly fine without loading win32; like the first phase of setup or the recovery console.
NT3.1 gold is build 511.
Most of ntdll.dll's entry points (all the Nt* functions) simply load a kernel function ID into AX and use a software interrupt to enter kernel mode. It represents NT's system call interface.
It has 285 of those in xpsp1.
The first release of NT(3.1) has 178.
You can also deny users from writing and creating files by the use of ACLs for an effect like mounting read-only. If anything, the security model of WinNT is more flexible than a standard Linux system.
In languages like Python that can treat code as data, the code is not stored in the binary form the CPU executes, but as a list of instructions for the language's interpreter. So, as far as the CPU is concerned, those pieces of code are really data. .NET JIT. It will have to be aware of the NX flag and set it explicitlly.
If the code was compiled beforehand, then the reference to the code structure will be a pointer to the actual function, stored with the rest of code.
If the language has a JIT (just-in-time compiler), it has a risk of being broken, as mentioned earlier with the