Spring 2017 :: CSE 506 — Operating Systems

• Home
• Schedule
• Labs
• Lab Tools
• References

Lab 3 Introduction

In this lab you will implement the basic kernel facilities required to get a protected user-mode environment (i.e., "process") running. You will enhance the JOS kernel to set up the data structures to keep track of user environments, create a single user environment, load a program image into it, and start it running. You will also make the JOS kernel capable of handling any system calls the user environment makes and handling any other exceptions it causes.

Note: In this lab, the terms environment and process are interchangeable - they have roughly the same meaning. We introduce the term "environment" instead of the traditional term "process" in order to stress the point that JOS environments do not provide the same semantics as UNIX processes, even though they are roughly comparable.

Getting Started

Follow these instructions to fetch the new source and merge it with your old code. Remember to use lab3 as the name of the new Git branch.

Lab 3 contains a number of new source files, which you should browse:

In addition, a number of the source files we handed out for lab2 are modified in lab3. To see the differences, you can type:

$ git diff lab2 | more

You may also want to take another look at the lab tools guide, as it includes information on debugging user code that becomes relevant in this lab.

Inline Assembly

In this lab you may find GCC's inline assembly language feature useful, although it is also possible to complete the lab without using it. At the very least, you will need to be able to understand the fragments of inline assembly language (asm statements) that already exist in the source code we gave you. You can find several sources of information on GCC inline assembly language on the class reference materials page.

Part A: User Environments and Exception Handling

The new include file inc/env.h contains basic definitions for user environments in JOS; read it now. The kernel uses the Env data structure to keep track of each user environment. In this lab you will initially create just one environment, but you will need to design the JOS kernel to support multiple environments; lab 4 will take advantage of this feature by allowing a user environment to fork other environments.

As you can see in kern/env.c, the kernel maintains three main global variables pertaining to environments:

struct Env *envs = NULL;                   /* All environments */
struct Env *curenv = NULL;                 /* the current env */
static struct Env *env_free_list;          /* Free environment list */

Once JOS gets up and running, the envs pointer points to an array of Env structures representing all the environments in the system. In our design, the JOS kernel will support a maximum of NENV simultaneously active environments, although there will typically be far fewer running environments at any given time. (NENV is a constant defined in inc/env.h.) Once it is allocated, the envs array will contain a single instance of the Env data structure for each of the NENV possible environments.

The JOS kernel keeps all of the inactive Env structures on the env_free_list. This design allows easy allocation and deallocation of environments, as they merely have to be added to or removed from the free list.

The kernel uses the curenv variable to keep track of the currently executing environment at any given time. During boot up, before the first environment is run, curenv is initially set to NULL.

Environment State

The Env structure is defined in inc/env.h as follows (although more fields will be added in future labs):

struct Env {
        struct Trapframe env_tf;        // Saved registers
        struct Env *env_link;           // Next free Env
        envid_t env_id;                 // Unique environment identifier
        envid_t env_parent_id;          // env_id of this env's parent
        enum EnvType env_type;          // Indicates special system environments
        unsigned env_status;            // Status of the environment
        uint32_t env_runs;              // Number of times environment has run

        // Address space
        pml4e_t *env_pml4e;             // Kernel virtual address of page map level-4
};

Here's what the Env fields are for:

env_tf:

This structure, defined in inc/trap.h, holds the saved register values for the environment while that environment is not running: i.e., when the kernel or a different environment is running. The kernel saves these when switching from user to kernel mode, so that the environment can later be resumed where it left off.

env_link:

This is a link to the next Env on the env_free_list. env_free_list points to the first free environment on the list.

env_id:

The kernel stores here a value that uniquely identifies the environment currently using this Env structure (i.e., using this particular slot in the envs array). After a user environment terminates, the kernel may re-allocate the same Env structure to a different environment - but the new environment will have a different env_id from the old one even though the new environment is re-using the same slot in the envs array.

env_parent_id:

The kernel stores here the env_id of the environment that created this environment. In this way the environments can form a "family tree," which will be useful for making security decisions about which environments are allowed to do what to whom.

env_type:

This is used to distinguish special environments. For most environments, it will be ENV_TYPE_USER. The idle environment is ENV_TYPE_IDLE and we'll introduce a few more special types for special system service environments in later labs.

env_status:

This variable holds one of the following values:

ENV_FREE:

Indicates that the Env structure is inactive, and therefore on the env_free_list.

ENV_RUNNABLE:

Indicates that the Env structure represents a currently active environment, and the environment is waiting to run on the processor.

ENV_RUNNING:

Indicates that the Env structure represents the currently running environment.

ENV_NOT_RUNNABLE:

Indicates that the Env structure represents a currently active environment, but it is not currently ready to run: for example, because it is waiting for an interprocess communication (IPC) from another environment.

ENV_DYING:

Indicates that the Env structure represents a zombie environment. A zombie environment will be freed the next time it traps to the kernel. We will not use this flag until Lab 4.

env_pml4e:

This variable holds the kernel virtual address of this environment's top-level (4th level) page directory.

Like a Unix process, a JOS environment couples the concepts of "thread" and "address space". The thread is defined primarily by the saved registers (the env_tf field), and the address space is defined by the PML4 page directory pointer, page directory and page tables pointed to by env_pml4e and env_cr3. To run an environment, the kernel must set up the CPU with both the saved registers and the appropriate address space.

Note that in Unix-like systems, individual environments have their own kernel stacks. In JOS, however, only one environment can be active in the kernel at once, so JOS needs only a single kernel stack.

Allocating the Environments Array

In lab 2, you allocated memory in x86_vm_init() for the pages[] array, which is a table the kernel uses to keep track of which pages are free and which are not. You will now need to modify x64_vm_init() further to allocate a similar array of Env structures, called envs.

Creating and Running Environments

You will now write the code in kern/env.c necessary to run a user environment. Because we do not yet have a filesystem, we will set up the kernel to load a static binary image that is embedded within the kernel itself. JOS embeds this binary in the kernel as a ELF executable image.

The Lab 3 GNUmakefile generates a number of binary images in the obj/user/ directory. If you look at kern/Makefrag, you will notice some magic that "links" these binaries directly into the kernel executable as if they were .o files. The -b binary option on the linker command line causes these files to be linked in as "raw" uninterpreted binary files rather than as regular .o files produced by the compiler. (As far as the linker is concerned, these files do not have to be ELF images at all - they could be anything, such as text files or pictures!) If you look at obj/kern/kernel.sym after building the kernel, you will notice that the linker has "magically" produced a number of funny symbols with obscure names like _binary_obj_user_hello_start, _binary_obj_user_hello_end, and _binary_obj_user_hello_size. The linker generates these symbol names by mangling the file names of the binary files; the symbols provide the regular kernel code with a way to reference the embedded binary files.

In i386_init() in kern/init.c you'll see code to run one of these binary images in an environment. However, the critical functions to set up user environments are not complete; you will need to fill them in.

r = -E_NO_MEM;
panic("env_alloc: %e", r);

Below is a call graph of the code up to the point where the user code is invoked. Make sure you understand the purpose of each step.

• start (kern/entry.S)
• i386_init
  • cons_init
  • x86_vm_init
  • env_init
  • trap_init (still incomplete at this point)
  • env_create
  • env_run
    • env_pop_tf

Once you are done you should compile your kernel and run it under QEMU. If all goes well, your system should enter user space and execute the hello binary until it makes a system call with the int instruction. At that point there will be trouble, since JOS has not set up the hardware to allow any kind of transition from user space into the kernel. When the CPU discovers that it is not set up to handle this system call interrupt, it will generate a general protection exception, find that it can't handle that, generate a double fault exception, find that it can't handle that either, and finally give up, reset, and reboot in what's known as a "triple fault". Usually, you would then see the CPU reset and the system reboot. While this is important for legacy applications (see this blog post for an explanation of why), it's a pain for kernel development, so with the 506 patched QEMU you'll instead see a register dump and a "Triple fault." message.

We'll address this problem shortly, but for now we can use the debugger to check that we're entering user mode. Use make qemu-gdb (or make qemu-nox-gdb) and set a GDB breakpoint at env_pop_tf, which should be the last function you hit before actually entering user mode. Single step through this function using si; the processor should enter user mode after the iret instruction. You should then see the first instruction in the user environment's executable, which is the cmpl instruction at the label start in lib/entry.S. Now use b *0x... to set a breakpoint at the int $0x30 in sys_cputs() in hello (see obj/user/hello.asm for the user-space address). This int is the system call to display a character to the console. If you cannot execute as far as the int, then something is wrong with your address space setup or program loading code; go back and fix it before continuing.

Handling Interrupts and Exceptions

At this point, the first int $0x30 system call instruction in user space is a dead end: once the processor gets into user mode, there is no way to get back out. You will now need to implement basic exception and system call handling, so that it is possible for the kernel to recover control of the processor from user-mode code. The first thing you should do is thoroughly familiarize yourself with the x86 interrupt and exception mechanism.

In this lab we generally follow Intel's terminology for interrupts, exceptions, and the like. However, terms such as exception, trap, interrupt, fault and abort have no standard meaning across architectures or operating systems, and are often used without regard to the subtle distinctions between them on a particular architecture such as the x86. When you see these terms outside of this lab, the meanings might be slightly different.

Basics of Protected Control Transfer

Exceptions and interrupts are both "protected control transfers," which cause the processor to switch from user to kernel mode (CPL=0) without giving the user-mode code any opportunity to interfere with the functioning of the kernel or other environments. In Intel's terminology, an interrupt is a protected control transfer that is caused by an asynchronous event usually external to the processor, such as notification of external device I/O activity. An exception, in contrast, is a protected control transfer caused synchronously by the currently running code, for example due to a divide by zero or an invalid memory access.

In order to ensure that these protected control transfers are actually protected, the processor's interrupt/exception mechanism is designed so that the code currently running when the interrupt or exception occurs does not get to choose arbitrarily where the kernel is entered or how. Instead, the processor ensures that the kernel can be entered only under carefully controlled conditions. In x86, two mechanisms provide this protection:

1. The Interrupt Descriptor Table. The processor ensures that interrupts and exceptions can only cause the kernel to be entered at a few specific, well-defined entry-points determined by the kernel itself, and not by the code running when the interrupt or exception is taken.

   The x86 allows up to 256 different interrupt or exception entry points into the kernel, each with a different interrupt vector. A vector is a number between 0 and 255. An interrupt's vector is determined by the source of the interrupt: different devices, error conditions, and application requests to the kernel generate interrupts with different vectors. The CPU uses the vector as an index into the processor's interrupt descriptor table (IDT), which the kernel sets up in kernel-private memory, much like the GDT. From the appropriate entry in this table the processor loads:

   • the value to load into the instruction pointer (RIP) register, pointing to the kernel code designated to handle that type of exception.
   • the value to load into the code segment (CS) register, which includes in bits 0-1 the privilege level at which the exception handler is to run. (In JOS, all exceptions are handled in kernel mode, privilege level 0.)

2. The Task State Segment. The processor needs a place to save the old processor state before the interrupt or exception occurred, such as the original values of RIP and CS before the processor invoked the exception handler, so that the exception handler can later restore that old state and resume the interrupted code from where it left off. But this save area for the old processor state must in turn be protected from unprivileged user-mode code; otherwise buggy or malicious user code could compromise the kernel.

   For this reason, when an x86 processor takes an interrupt or trap that causes a privilege level change from user to kernel mode, it also switches to a stack in the kernel's memory. A structure called the task state segment (TSS) specifies the segment selector and address where this stack lives. The processor pushes (on this new stack) SS, RSP, EFLAGS, CS, RIP, and an optional error code. Then it loads the CS and RIP from the interrupt descriptor, and sets the RSP and SS to refer to the new stack.

   Although the TSS is large and can potentially serve a variety of purposes, JOS only uses it to define the kernel stack that the processor should switch to when it transfers from user to kernel mode. Since "kernel mode" in JOS is privilege level 0 on the x86, the processor uses the RSP0 and SS0 fields of the TSS to define the kernel stack when entering kernel mode. JOS doesn't use any other TSS fields.

Types of Exceptions and Interrupts

All of the synchronous exceptions that the x86 processor can generate internally use interrupt vectors between 0 and 31, and therefore map to IDT entries 0-31. For example, a page fault always causes an exception through vector 14. Interrupt vectors greater than 31 are only used by software interrupts, which can be generated by the int instruction, or asynchronous hardware interrupts, caused by external devices when they need attention.

In this section we will extend JOS to handle the internally generated x86 exceptions in vectors 0-31. In the next section we will make JOS handle software interrupt vector 48 (0x30), which JOS (fairly arbitrarily) uses as its system call interrupt vector. In Lab 4 we will extend JOS to handle externally generated hardware interrupts such as the clock interrupt.

An Example

Let's put these pieces together and trace through an example. Let's say the processor is executing code in a user environment and encounters a divide instruction that attempts to divide by zero.

1. The processor switches to the stack defined by the SS0 and RSP0 fields of the TSS, which in JOS will hold the values GD_KD and KSTACKTOP, respectively.
2. The processor pushes the exception parameters on the kernel stack, starting at address KSTACKTOP:

                        +--------------------+ KSTACKTOP     
                        | 0x00000 | old SS   |     " - 8
                        |      old RSP       |     " - 16
                        |     old EFLAGS     |     " - 24
                        | 0x00000 | old CS   |     " - 32
                        |      old RIP       |     " - 40 <---- RSP 
                        +--------------------+             
   	

3. Because we're handling a divide error, which is interrupt vector 0 on the x86, the processor reads IDT entry 0 and sets CS:RIP to point to the handler function defined there.
4. The handler function takes control and handles the exception, for example by terminating the user environment.

For certain types of x86 exceptions, in addition to the "standard" five words above, the processor pushes onto the stack another word containing an error code. The page fault exception, number 14, is an important example. See the 80386 manual to determine for which exception numbers the processor pushes an error code, and what the error code means in that case. When the processor pushes an error code, the stack would look as follows at the beginning of the exception handler when coming in from user mode:

                     +--------------------+ KSTACKTOP             
                     | 0x00000 | old SS   |     " - 8
                     |      old RSP       |     " - 16
                     |     old EFLAGS     |     " - 24
                     | 0x00000 | old CS   |     " - 32
                     |      old RIP       |     " - 40
                     |     error code     |     " - 48 <---- RSP
                     +--------------------+             
	

Nested Exceptions and Interrupts

The processor can take exceptions and interrupts both from kernel and user mode. It is only when entering the kernel from user mode, however, that the x86 processor automatically switches stacks before pushing its old register state onto the stack and invoking the appropriate exception handler through the IDT. If the processor is already in kernel mode when the interrupt or exception occurs (the low 2 bits of the CS register are already zero), then the kernel just pushes more values on the same kernel stack. In this way, the kernel can gracefully handle nested exceptions caused by code within the kernel itself. This capability is an important tool in implementing protection, as we will see later in the section on system calls.

If the processor is already in kernel mode and takes a nested exception, since it does not need to switch stacks, it does not save the old SS or RSP registers. For exception types that do not push an error code, the kernel stack therefore looks like the following on entry to the exception handler:

                     +--------------------+ <---- old RSP
                     |     old EFLAGS     |     " - 8
                     | 0x00000 | old CS   |     " - 16
                     |      old RIP       |     " - 24
                     +--------------------+             

For exception types that push an error code, the processor pushes the error code immediately after the old RIP, as before.

There is one important caveat to the processor's nested exception capability. If the processor takes an exception while already in kernel mode, and cannot push its old state onto the kernel stack for any reason such as lack of stack space, then there is nothing the processor can do to recover, so it simply resets itself. Needless to say, the kernel should be designed so that this can't happen.

Setting Up the IDT

You should now have the basic information you need in order to set up the IDT and handle exceptions in JOS. For now, you will set up the IDT to handle interrupt vectors 0-31 (the processor exceptions) and interrupts 32-47 (the device IRQs).

The header files inc/trap.h and kern/trap.h contain important definitions related to interrupts and exceptions that you will need to become familiar with. The file kern/trap.h contains definitions that are strictly private to the kernel, while inc/trap.h contains definitions that may also be useful to user-level programs and libraries.

Note: Some of the exceptions in the range 0-31 are defined by Intel to be reserved. Since they will never be generated by the processor, it doesn't really matter how you handle them. Do whatever you think is cleanest.

The overall flow of control that you should achieve is depicted below:

      IDT                   trapentry.S         trap.c
   
+----------------+                        
|   &handler1    |--------> handler1:          trap (struct Trapframe *tf)
|                |            // do stuff      {
|                |            call trap          // handle the exception/interrupt
|                |            // undo stuff    }
+----------------+
|   &handler2    |--------> handler2:
|                |            // do stuff
|                |            call trap
|                |            // undo stuff
+----------------+
       .
       .
       .
+----------------+
|   &handlerX    |--------> handlerX:
|                |             // do stuff
|                |             call trap
|                |             // undo stuff
+----------------+

Each exception or interrupt should have its own handler in trapentry.S and trap_init() should initialize the IDT with the addresses of these handlers. Each of the handlers should build a struct Trapframe (see inc/trap.h) on the stack and call trap() (in trap.c) with a pointer to the Trapframe.

trap() handles the exception/interrupt or dispatches to a specific handler function. If and when trap() returns, the code in trapentry.S restores the old CPU state saved in the Trapframe and then uses the iret instruction to return from the exception.

This completes part A.

Part B: Page Faults, Breakpoints Exceptions, and System Calls

Now that your kernel has basic exception handling capabilities, you will refine it to provide important operating system primitives that depend on exception handling.

Handling Page Faults

The page fault exception, interrupt vector 14 (T_PGFLT), is a particularly important one that we will exercise heavily throughout this lab and the next. When the processor takes a page fault, it stores the linear address that caused the fault in a special processor control register, CR2. In trap.c we have provided the beginnings of a special function, page_fault_handler(), to handle page fault exceptions.

You will further refine the kernel's page fault handling below, as you implement system calls.

The Breakpoint Exception

The breakpoint exception, interrupt vector 3 (T_BRKPT), is normally used to allow debuggers to insert breakpoints in a program's code by temporarily replacing the relevant program instruction with the special 1-byte int3 software interrupt instruction. In JOS we will abuse this exception slightly by turning it into a primitive pseudo-system call that any user environment can use to invoke the JOS kernel monitor. This usage is actually somewhat appropriate if we think of the JOS kernel monitor as a primitive debugger. The user-mode implementation of panic() in lib/panic.c, for example, performs an int3 after displaying its panic message.

System calls

User processes ask the kernel to do things for them by invoking system calls. When the user process invokes a system call, the processor enters kernel mode, the processor and the kernel cooperate to save the user process's state, the kernel executes appropriate code in order to carry out the system call, and then resumes the user process. The exact details of how the user process gets the kernel's attention and how it specifies which call it wants to execute vary from system to system.

In the JOS kernel, we will use the int instruction, which causes a processor interrupt. In particular, we will use int $0x30 as the system call interrupt. We have defined the constant T_SYSCALL to 48 (0x30) for you. You will have to set up the interrupt descriptor to allow user processes to cause that interrupt. Note that interrupt 0x30 cannot be generated by hardware, so there is no ambiguity caused by allowing user code to generate it.

The application will pass the system call number and the system call arguments in registers. This way, the kernel won't need to grub around in the user environment's stack or instruction stream. The system call number will go in %rax, and the arguments (up to five of them) will go in %rdx, %rcx, %rbx, %rdi, and %rsi, respectively. The kernel passes the return value back in %eax. The assembly code to invoke a system call has been written for you, in syscall() in lib/syscall.c. You should read through it and make sure you understand what is going on.

rax                - syscall number
rdx, rcx, rbx, rdi - arg1, arg2, arg3, arg4
rsi                - return pc
rbp                - return rsp
rsp                - trashed by sysenter

User-mode startup

A user program starts running at the top of lib/entry.S. After some setup, this code calls libmain(), in lib/libmain.c. You should modify libmain() to initialize the global pointer thisenv to point at this environment's struct Env in the envs[] array. (Note that lib/entry.S has already defined envs to point at the UENVS mapping you set up in Part A.) Hint: look in inc/env.h and use sys_getenvid.

libmain() then calls umain, which, in the case of the hello program, is in user/hello.c. Note that after printing "hello, world", it tries to access env->env_id. This is why it faulted earlier. Now that you've initialized env properly, it should not fault. If it still faults, you probably haven't mapped the UENVS area user-readable (back in Part A in pmap.c; this is the first time we've actually used the UENVS area).

Page faults and memory protection

Memory protection is a crucial feature of an operating system, ensuring that bugs in one program cannot corrupt other programs or corrupt the operating system itself.

Operating systems usually rely on hardware support to implement memory protection. The OS keeps the hardware informed about which virtual addresses are valid and which are not. When a program tries to access an invalid address or one for which it has no permissions, the processor stops the program at the instruction causing the fault and then traps into the kernel with information about the attempted operation. If the fault is fixable, the kernel can fix it and let the program continue running. If the fault is not fixable, then the program cannot continue, since it will never get past the instruction causing the fault.

As an example of a fixable fault, consider an automatically extended stack. In many systems the kernel initially allocates a single stack page, and then if a program faults accessing pages further down the stack, the kernel will allocate those pages automatically and let the program continue. By doing this, the kernel only allocates as much stack memory as the program needs, but the program can work under the illusion that it has an arbitrarily large stack.

System calls present an interesting problem for memory protection. Most system call interfaces let user programs pass pointers to the kernel. These pointers point at user buffers to be read or written. The kernel then dereferences these pointers while carrying out the system call. There are two problems with this:

1. A page fault in the kernel is potentially a lot more serious than a page fault in a user program. If the kernel page-faults while manipulating its own data structures, that's a kernel bug, and the fault handler should panic the kernel (and hence the whole system). But when the kernel is dereferencing pointers given to it by the user program, it needs a way to remember that any page faults these dereferences cause are actually on behalf of the user program.
2. The kernel typically has more memory permissions than the user program. The user program might pass a pointer to a system call that points to memory that the kernel can read or write but that the program cannot. The kernel must be careful not to be tricked into dereferencing such a pointer, since that might reveal private information or destroy the integrity of the kernel.

For both of these reasons the kernel must be extremely careful when handling pointers presented by user programs.

You will now solve these two problems with a single mechanism that scrutinizes all pointers passed from userspace into the kernel. When a program passes the kernel a pointer, the kernel will check that the address is in the user part of the address space, and that the page table would allow the memory operation.

Thus, the kernel will never suffer a page fault due to dereferencing a user-supplied pointer. If the kernel does page fault, it should panic and terminate.

[00001000] user_mem_check assertion failure for va 00000001
[00001000] free env 00001000
Destroyed the only environment - nothing more to do!

Note that the same mechanism you just implemented also works for malicious user applications (such as user/evilhello).

[00000000] new env 00001000
[00001000] user_mem_check assertion failure for va 800420000c
[00001000] free env 00001000

This completes the lab. Make sure you pass all the make grade tests, and hand in your work with make handin. If submission fails, double check that you have committed all of your changes, and read any error messages carefully before emailing the course staff for help.