/ZJU_OS_lab1

Operating System(2018-2019Fall&Winter) LAB1

Primary LanguageC

Operating System(2018-2019Fall&Winter)LAB1

Project Name:   Sychronous Mutual Exclusion and Linux Kernel Module
Student Name:   Hu.Zhaodong
Email       :   zhaodonghu94@zju.edu.cn
phone       :   15700080428
Date        :   2018.9-2018.12

TARGET

  • Learn the system call of linux and use the pthread library to programing.
  • Fully understand the importance of Atomic operation to shared memory.
  • Further understand and master the concept of process and thread and their sychronous and mutual exclusion.
  • Learn to program use multi-thread and handle the sychronous and mutual exclusion problem.
  • Learn the mechanism of module in Linux and master how to write a linux module.
  • Futher understanding of process in linux by traversing processes in linux system.

CONTENT1

There are two roads and two lanes for each road and they intersect with each other. Assume the car can only move forward and can not turn or move backword. If when 4 car from 4 direction arrive the cross simultaneously, just like Fig(a). Then they keep move forward and have to stop with every dude entangled with each other. It's a deadlock problem. From the concept of resources allocation of OS, if both 4 guys want to pass through the cross, their need for resources is just like this:

  • The car move to north need a,b.
  • The car move to west need b,c.
  • The car move to south need c,a.
  • The car move to east need d,a.
    Fig1

We need to handle the sychronous problem and prevent the hungry and deadlock of the car when they pass the across. In our system, there are cars from different direction pass through the across. The cars in same direction line up in turn to through the intersection. According to the traffic rule, the car from right side get the privilige of go first. (e.g. In Fig(1), assume there is car1, car2 and car3, the order of will be car1->car2->car3).
The general rule list below:

  • Cars from same direction pass the cross in order and drive right when they arrive the cross.
  • When cars from different direction arrive the cross simultineosly, they car in the right side go first unless this guy get a signal of immediate access.
  • No deadlock.
  • No starving.
  • Not any car thread can control other cars, says, no single point scheduling.
  • No AND semaphore, says the car can not request two road resources simultaneously (e.g. The south car can not judge if a or b are free simultaneously).

DEVICE

CPU     :       intel core i7-4790k (4C8T) 4.6GHz
RAM     :       16G
OS      :       ubuntu-16.04.5-desktop-i386(Kernel 4.6.0)
gcc     :       4.8.5
thread  :       posix  

SOLUTION

Car Behavior

We solve the problem by assign a thread for each car and solve the problem with the perspective of the car thread.

  1. when a car thread generate, the car will get two attribute, direction and number. And the number actually is the car's ID and the number of any two car will not be equal.
  2. The car will enter a wait queue for its direction. To prevent other car operate the wait queue when this car is operating. So we use a mutex lock for the wait queue.
  3. The car will judge if the number of the car in the head of the wait queue is the same of its number. If not, the car will stuck in the wait queue until the queue head number is just its own number. If the car is in the queue head, the car tell the output device (e.g. our monitor) it has arrived the cross and can't wait to start to pass the cross. And at the same time the car behind this car will not have the the opportunity to become the queue head and tell the output device it arrives before the former car quit from the queue and give the car behind it a broadcast signal.
  4. All the car stuck in the queue will receive the broadcast and re-judge if they become the head of the queue. Only the new head car will continue and the others will back to the wait status till another car get out of the queue and give a new broadcast. This setting make sure there can only be one car enter the judge and pass status. That keep the thread safe.
  5. Now the head car tell the monitor it arrived just the same as the car in step 3.
  6. The car will evaluate the status in the cross. According the passing rules, the right-side car has the privilege to ge first. So the current car will judge if there be some car in its right side. So the car thread will check the size of the queue in its right direction. If the size equal zero that means at this time the right side has no car and it could get into the cross directly if the cross is empty. If the size is not equal to zero, the current car must wait till the right side car pass the cross and send a first-signal to it. And the car will use a flag to tell the god if it's in the waiting status.
  7. After get the first signal from the right side. The car then check if the cross is empty. If not empty, wait the dude in the cross to send a signal. If empty, get into it, tell the others some dude is in the cross and pass through.
  8. The current car tell the monitor it is leaving. And the car give the left-side car a first-sinal to tell the car now it could start to enter the cross once the cross is empty.
  9. Then the car get out of the cross and tell others there is no dude in the cross and give the car which is waiting to enter the cross a sinal that "I'm out of the cross!".
  10. Then the car quit the queue and the car behind it can check if they are the new head now.

In general, the car behavior can be conclued to the diagram below:

start->
enter a queue->
wait till become the head->
check the right-side(queue empty?)->
check the cross(cross empty?)->
enter the cross->
leave the cross->
give signal to the left side car->
quit from the queue->
get out of the cross->
end

DeadLock Check

We don't let the car to check if the deadlock happens, we let a character who is just like a policeman to check the deadlock status and give its order.

  1. The checkman will judge if the head car in the 4 diretion is in the waiting status (In the step6 the car tell the god whether it is waiting status.
  2. If not all the directions are in the waiting stauts, then ignore the remain operation and keep check. If all the directions are in the waiting status, then give a signal to the NORTH direction let the head car of the NORTH direction to go first.

Remark

  • Something about pthread:

    pthread_cond_wait is a interesting function, we pass the two parameters to it, the pthread_cond_t and the pthread_mutex_t. When the function is invoked, we let the thread to wait until the cond is satisfied, and meanwhile we find the mutex passed in, and unlock the mutex! So when this thread is waiting for the conditon, other thread could do something in the critcal section. (It's just like if we don't pull, we hand over my pit so others could use.) And when the condition is satisfied, we regain the possesion of the mutex(that is lock the dirMut again) The process just like that:

    invoke->unlock the mutex->wait->conditon satisfied->lock again->continue
    

    And we usually usr the pthread_cond_t with the while loop and pthread_mutex_t. Check the example below:

    //thread A
    pthread_mutex_lock(mutex);
    while(!ConditionStatisfied())pthread_cond_wait(&conditon,&mutex);
    pthread_mutex_unlock(mutex);
    
    //thread B
    pthread_mutex_lock(mutex);
    signal(&condition);
    pthread_mutex_unlock(mutex);

    thread A will wait the conditon to be satified and release the control of the critical section at the same time. And thread B could regain the control of critical section and then signal the thread A the conditon is satisfied. Then thread B hand out the critical section. After the thread A is awake the thread A will try to regain the control of the critical section so it will get the mutex after the thread B release the mutex. And before it continue, thread A should also check if the condition in the while loop is still statisfied. That prevent the situation that the thread A is signaled but the condition in the while loop has lost it power. The internal in the thread A could be describ like this:

    unlock(mutex);
    wait();
    lock(mutex);

    And the pthread_cond_wait() merges them as a atomic operation.

  • A subtle method:

    Some subtle things may happen when the car in the same direction is wait in the queue. When after a car thread created and assigned values, the car will enter a queue contains parteners have same direction. The car in the thread will enter a while loop to check if they are the head of the queue. If not, they will be in the waiting status and waiting the out car to give the signal. In my design, when a car is stuck in the condition outQueue[direction] using the pthread_cond_wait(), this car thread will release the mutex and let the car thread still not enter the queue to enter the queue and wait the same signal (Yes, the signal give by the car get out of the queue). No there is more than one thread is waiting. But the pthread_cond_signal() can only wake one thread. If the car thread in the queue which is waken by the out car thread is not in the queue head yet. The waken car thread will re-check the conditon in the while loop and found it still not statisfied (still not in the head), the car thread will back to wait. And the former car out the queue and out the cross but no new car in the same direction wake up from the sleep. That's a disaster! So we use the pthread_cond_broadcast() to wake all the car thread wait in this direction and only the head car can successfully re-check the conditon in the while loop and continue the trip. The check-failed car will back to the waiting status.

    car1->car2->car3->car4->...->cark->...
     ^     ^     ^     ^          ^
    wait  wait  wait  wait       wait
     |     |     |     |          |
      Check the while(inTheHead())
     |     |     |     |          |
     T     F     F     F          F
     |     |     |     |          |
     Go   wait  wait  wait       wait
    

    All the car in the queue waiting for the signal will wake but only the head car will continue and the rest will back to sleep.

  • A problem in the deadlock check:

    In this design, the car dosen't check if there is a deadlock. All car need to do is to tell the god(the checkman) it is in the waiting conditon due the right-side cars. So the deadlock check thread have to check the conditon again and agian. Besides, the check thread has to check 4 flags for the 4 direction. we could make sure the flags to check will not be change once the check thread is working in its critical section to gain the whole 4 mutex of the flags. But we can't guarantee after the flags stand, the car thread will stop move. The must be situation like this: A car is in the waiting status, and the check thread notice that. The car may continue when the condition statified.But simultenously, the car thread check the 3 other directions and find they are all in the waiting status and give the "NORTH FIRST" signal. Actually there is no deadlock actually. A more persuasive solution is waiting to be figure out.

Code

The code could get from the carCross.c file in the zip. Or just visit my github repository. And details are commited in my source file.
Use the instruction below to compile and run:

//compile
gcc -pthread -o carCross carCross.c
//run (can use any argv[1] with the size below 100 to test)
./carCross wsnewsnewsnewsneewsesewn

CONTENT2

Write a linux kernel module, which has the function to traverse all the process. This module can output the name, pid, process state, parent name. Besides, it can also statistic the nubmer of the different type processes, include TASK_RUNNING, TASK_ZOMBIE, TASK_INTERRUPTIBLE, TASK_UNINTERRRUPTIBLE, TASK_STOP etc. And write a user-mode program to output the result in the monitor. Give each line of the code some commit.

DEVICE

CPU     :       intel core i7-4790k (4C8T) 4.6GHz
RAM     :       16G
OS      :       ubuntu-16.04.5-desktop-i386(Kernel 4.6.0)
gcc     :       4.8.5
thread  :       posix  

SOLUTION

We use a pointer to point the init_task, then we traverse all the task_struct to find out the PID, DESCRIPTOR, FATHER DESCRIPTOR, FATHER PID, STATE and output them in the monitor. In the last we output the statistic information of different state process. For some reason the printk() will only output the information to /var/log/kern.log. We could use the command dmesg to show the output information. But it either show two much information or the quantity of the imformation is too much so the dmesg can't show them all. So we write another user programm to open the /var/log/kern.log and output the imformation we want.

CODE

kernel module

Visit process_module.c in my Github or just visit the file process_module.c in the same directory.
usage:

//build
make
//run the module
insmod process_module.ko
//remove the module
rmmod process_module.ko
//remove the make file
make clean

Makefile

Visit Makefile in my github or just visit the file Makefile in the same directory.

TARGET=process_module
KDIR=/usr/src/linux
PWD=$(shell pwd)
obj-m += $(TARGET).o
default:
	make -C $(KDIR) M=$(PWD) modules
clean:
	make -C $(KDIR) M=$(PWD) clean

usage:

make
make clean

user interface

Visit user_Interface.c in my github or just visit the file user_Interface.c in the same directory.
After we make the module ,we could just compile the user-Interface.c by gcc and then run the executable file. And this program will insert the module, find the information in /var/log/kern.log and out put. After printing the program will let the system to remove the module.
usage:

//do it after the make
gcc -o user user_Interface.c
./user