MIT6.S081-Lab10 mmap 2021Fall 的长尾词是:MIT6.S081-Lab10中2021秋季的mmap实验具体难点和实现细节有哪些?

2026-04-11 03:522阅读0评论SEO资源
  • 内容介绍
  • 文章标签
  • 相关推荐

本文共计3889个文字,预计阅读时间需要16分钟。

MIT6.S081-Lab10 mmap 2021Fall 的长尾词是:MIT6.S081-Lab10中2021秋季的mmap实验具体难点和实现细节有哪些?

开始日期:22.07.15

操作系统:Ubuntu 20.0.4

实验:mmap

实验内容:

本实验涉及使用mmap系统调用来实现文件映射。我们将学习如何将文件内容映射到内存中,进行读写操作,并探讨其应用场景。

实验步骤:

1. 使用mmap()函数将文件映射到内存。

2.在映射的内存区域进行读写操作。

3.使用munmap()函数解除映射。

实验代码及结果:

c

#include #include #include #include

int main() { int fd=open(file.txt, O_RDWR); if (fd==-1) { perror(open); return 1; }

size_t filesize=lseek(fd, 0, SEEK_END); char *map=mmap(NULL, filesize, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); if (map==MAP_FAILED) { perror(mmap); close(fd); return 1; }

// 在映射的内存区域进行读写操作...

munmap(map, filesize); close(fd);

return 0;}

实验总结:

本实验成功实现了使用mmap进行文件映射,并在映射的内存区域进行了读写操作。通过mmap,我们可以方便地在文件和内存之间进行数据交换,提高程序的性能。

开始日期:22.07.15

操作系统:Ubuntu20.0.4

Link:Lab: mmap

目录
  • Lab mmap
    • 写在前面
      • 踩坑
      • 参考材料
    • 实验内容
      • mmap
        • access code
      • result
    • 总结

Lab mmap 写在前面

此部分不涉及实现细节,可以放心阅读

实现简易的mmapmunmap,需要对virtual address和physical address转换过程中使用的函数有一定理解,同时这里也采用了和cow、lazy alloction一样的lazy alloc方式。

实现之后,即可直接从程序中直接访问file。

踩坑
  • 英文阅读问题

    • as if 等价于 like,意思为“就像”,我错误理解为如果munmap被调用之后才调用exit,导致我一直没思路去修改exit

      Modify exit to unmap the process's mapped regions as if munmap had been called.

      修改exit来解除程序中已映射空间的映射,就像调用munmap一样

    • mmaptest.c中,有一句测试概括,这是个否定句,所以read/write按照英文习惯应该翻译为读写,但我一开始按照中文习惯错误翻译为读或写,导致编程时调试了一段时间才过这个测试

      check that mmap doesn't allow read/write mapping of a file opened read-only.

      检查:mmap不允许读写只读文件的映射

  • p->sz

    p->sz不但是程序的virtual space的大小也是physical space的大小

  • 编程过程中,可能出现关于strcut file报错和PROT_READ等定义的报错,在报错文件的开头处添加strcut file#include "fcntl.h"即可

参考材料
  • book-riscv-rev2
  • riscv-privileged
实验内容

警告:此部分涉及实现细节

mmap

实现要求:实现mmapmunmap,分配物理内存时要采用lazy alloc的方式,解除映射时要写回文件

实现思路:按照hint一步步来即可,但有不少地方需要自己发挥

MIT6.S081-Lab10 mmap 2021Fall 的长尾词是:MIT6.S081-Lab10中2021秋季的mmap实验具体难点和实现细节有哪些?

  • mmap的vma应该选择放在哪里?我采用的是先用程序的p->sz当作映射的虚拟地址,当然p->sz有可能已经被使用了(映射了两个文件),那我们就做检查,往后延长即可,注意这样做成功之后,肯定要将原p->sz增加length
  • munmap中要对p->sz、vma结构数据的lengthaddrf等数据进行处理,详细处理见注释,从而舍弃掉将要被解除的区域。同时在munmap中要注意写回文件这个操作,写回文件时要判断是不是第一次写回文件从而判断要不要写回时进行偏移(使用oldsz判断)。除此之外,当解除实际上没有映射到物理空间的vma时,我们不执行uvmunmap(),也不会执行写回文件,因为这个vma对应的pte中的有效位(PTE_V)没有在lazy alloc中被设置。
  • lazy alloc中,我写了个辅助函数mmap_lazyalloc,每次只分配一个page,为了能够在文件的中间部分就被读,当然,分配之前要先进行的检查。最后需要注意readi()被调用时,读文件时的偏移应该是多少,用va减去vma地址即可求出偏移。
  • exit中,类似于munmap,但可以减少一些vma结构数据处理,因为最后结构都会被释放掉。注意在这里写回文件时,使用的是vma的length。而不像munmap使用的是传入的length
  • fork中,我没有采用父子共享vmas的方式(没那么酷 = =),直接copy过去了,copy时vmas不用copy,因为父程序的vmas可能还没lazy alloc。这里需要注意一点,就是p->sz已经被改变了,我们要算出最开始没有vmas的p->sz,用当前的p->sz减掉所有vmas的length即可。
access code

首先是mmapmunmap,这两个syscall的设置,参考lab syscall即可,在此不赘述

然后是定义

/* proc.h */ struct vma_t { // int fd; struct file *f; int length; int prot; int flags; int offset; int oldsz; uint64 addr; }; // Per-process state struct proc { struct spinlock lock; ... char name[16]; // Process name (debugging) struct vma_t vmas[16]; // VMAs helps the kernel to decide how to handle page faults };

sys_mmap

/* */ uint64 sys_mmap(void) { struct file *f; int length , prot, flags, offset; uint64 addr; // get args, 0:addr, 1:length, 2:prot, 3:flags, 5:offset, 4:fd=>*f if(argaddr(0, &addr) < 0 || argint(1, &length) < 0 || argint(2, &prot) < 0 || argint(3, &flags) < 0 || argint(5, &offset) < 0 || argfd(4, 0, &f) < 0 ) return -1; // mmap doesn't allow read/write mapping of a file opened read-only with MAP_SHARED // "or", "/" also meaning "and" in english when use "doesn't", "not" and so on if(f->readable && !f->writable && (prot & PROT_READ) && (prot & PROT_WRITE) && (flags & MAP_SHARED)) return -1; struct proc *p = myproc(); int oldsz = p->sz; // get va without anything by p->sz if(addr == 0){ uint64 va = p->sz; int is_same = 0; while(1){ // avoid: some VAs may get same addr, becase we use 'lazy alloc' int i ; for(i = 0; i < 16; i++){ if(p->vmas[i].addr <= va && va < (p->vmas[i].addr + p->vmas[i].length)) is_same = 1; } if(!is_same){ addr = va; p->sz = va + length; break; } // reset is_same is_same = 0; if(va >= MAXVA) return -1; va += PGSIZE; } } for(int i = 0; i < 16; i++){ // case: maybe remove and rebuild a vmas[i] when slots full // not case in test // build a vams[i] if(p->vmas[i].addr == 0){ p->vmas[i].f = f; p->vmas[i].length = length; p->vmas[i].prot = prot; p->vmas[i].flags = flags; p->vmas[i].offset = offset; p->vmas[i].addr = addr; p->vmas[i].oldsz = oldsz; // mmap should increase the file's reference count filedup(f); return addr; } } // failed, ret -1 return -1; }

sys_munmap

uint64 sys_munmap(void) { uint64 addr; int length; // get args, 0:addr, 1:length if(argaddr(0, &addr) < 0 || argint(1, &length) < 0) return -1; struct proc *p = myproc(); int has_a_vma = 0; int i; for(i = 0 ; i < 16; i++){ if(p->vmas[i].addr <= addr && addr < (p->vmas[i].addr + p->vmas[i].length)){ has_a_vma = 1; break; } } // not find a vma, so ret -1 if(has_a_vma == 0){ return -1; } pte_t *pte = walk(p->pagetable, addr, 0); int offset = p->vmas[i].f->off; // addr is beginning of vma, so use file->off if(p->vmas[i].oldsz != addr) // addr is't beginning of vma, so use p->vmas[i].offset offset = p->vmas[i].offset; // If an unmapped page has been modified and the file is mapped MAP_SHARED, // write the page back to the file. if((*pte & PTE_V) && p->vmas[i].flags & MAP_SHARED){ begin_op(); ilock(p->vmas[i].f->ip); writei(p->vmas[i].f->ip, 1, addr, offset, length); iunlock(p->vmas[i].f->ip); end_op(); } // If munmap removes all pages of a previous mmap, // it should decrement the reference count of the corresponding struct file. // we keep end of old addr by 'p->vmas[i].addr += length' and 'p->vmas[i].length -= length' // we check by 'addr < (p->vmas[i].addr + p->vmas[i].length' in sys_mmap() // so we can't mmap [length munmap] and we will mmap after [p->vmas[i].length] // figure: // ' p->vmas[i].addr // [process data][ p->vmas[i].length ] // [ p->sz ] // ==> // ' p->vmas[i].addr // [process data][ length munmap ][p->vmas[i].length] // [ p->sz ] if(length < p->vmas[i].length){ p->sz -= length; p->vmas[i].addr += length; p->vmas[i].length -= length; } else if(length == p->vmas[i].length){ p->sz -= length; p->vmas[i].f->ref--; p->vmas[i].f = 0; p->vmas[i].addr = 0; p->vmas[i].prot = 0; p->vmas[i].flags = 0; p->vmas[i].length = 0; p->vmas[i].oldsz = 0; } else return -1; if((*pte & PTE_V) == 0) // don't write and uvmunmap(), only change data of vma return 0; // uvmunmap() after writing // find the VMA for the address range and unmap the specified pages // note: free physical memory => 'do_free = 1' uvmunmap(p->pagetable, addr, length/PGSIZE, 1); // success, ret 0 return 0; }

lazy alloc是当作page fault(13)来处理的

/* trap.c */ int mmap_lazyalloc(pagetable_t pagetable, uint64 va) { struct proc *p = myproc(); struct file *f; int prot; int has_a_vam = 0; int perm = 0; char *mem; // find va between vma.addr and vma.addr+vma.lenght int i; for(i = 0; i < 16; i++){ if(p->vmas[i].addr <= va && va < (p->vmas[i].addr + p->vmas[i].length)){ has_a_vam = 1; f = p->vmas[i].f; prot = p->vmas[i].prot; break; } } // not find vma, ret -1 if(has_a_vam == 0){ return -1; } // PTE_U controls whether instructions in user mode are allowed to access the page; // if PTE_U is notset, the PTE can be used only in supervisor mode. perm |= PTE_U; // MAYBE sets PTE_R, PTE_W, PTE_X if(prot & PROT_READ){ perm |= PTE_R; } if(prot & PROT_WRITE){ perm |= PTE_W; } if(prot & PROT_EXEC){ perm |= PTE_X; } // big bug: not alloc mem(4096) to all virtual addresses if((mem = kalloc()) == 0){ return -1; } // In mmaptest/makefile() // create a file to be mapped, containing // 1.5 pages of 'A' and half a page of zeros. // so we must set 0 of length after getting mem memset(mem, 0, PGSIZE); // note: mem is new address of phycial memory if(mappages(pagetable, va, PGSIZE, (uint64)mem, perm) == -1){ kfree(mem); return -1; } // we not set PTE_D, becasue we always directly wirite back to file in munmap() // length is the number of bytes to map; it might not be the same as the file's length. // read data from file, then put data to va ilock(f->ip); if(readi(f->ip, 1, va, va - p->vmas[i].addr, PGSIZE) < 0){ // readi offset by 'va - p->vmas[i].addr' iunlock(f->ip); return -1; } iunlock(f->ip); p->vmas[i].offset += PGSIZE; // success, ret 0 return 0; } void usertrap(void) { int which_dev = 0; if((r_sstatus() & SSTATUS_SPP) != 0) panic("usertrap: not from user mode"); // send interrupts and exceptions to kerneltrap(), // since we're now in the kernel. w_stvec((uint64)kernelvec); struct proc *p = myproc(); // save user program counter. p->trapframe->epc = r_sepc(); if(r_scause() == 8){ // system call if(p->killed) exit(-1); // sepc points to the ecall instruction, // but we want to return to the next instruction. p->trapframe->epc += 4; // an interrupt will change sstatus &c registers, // so don't enable until done with those registers. intr_on(); syscall(); // Fill in the page table lazily, in response to page faults. } else if(r_scause() == 13){ uint64 fault_va = r_stval(); int is_alloc = mmap_lazyalloc(p->pagetable, fault_va); if(fault_va > p->sz || is_alloc == -1){ p->killed = 1; } } else if((which_dev = devintr()) != 0){ // ok } else { printf("usertrap(): unexpected scause %p pid=%d\n", r_scause(), p->pid); printf(" sepc=%p stval=%p\n", r_sepc(), r_stval()); p->killed = 1; } if(p->killed) exit(-1); // give up the CPU if this is a timer interrupt. if(which_dev == 2) yield(); usertrapret(); }

exit

void exit(int status) { struct proc *p = myproc(); if(p == initproc) panic("init exiting"); // Close all open files. for(int fd = 0; fd < NOFILE; fd++){ if(p->ofile[fd]){ struct file *f = p->ofile[fd]; fileclose(f); p->ofile[fd] = 0; } } // 'as if' == 'like' for(int i = 0; i < 16; i++){ if(p->vmas[i].addr){ int offset = p->vmas[i].f->off; // addr is beginning of vma, so use file->off if(p->vmas[i].oldsz != p->vmas[i].addr) // addr is't beginning of vma, so use p->vmas[i].offset offset = p->vmas[i].offset; if(p->vmas[i].flags & MAP_SHARED){ begin_op(); ilock(p->vmas[i].f->ip); writei(p->vmas[i].f->ip, 1, p->vmas[i].addr, offset, p->vmas[i].length); iunlock(p->vmas[i].f->ip); end_op(); } p->sz -= p->vmas[i].length; p->vmas[i].f->ref--; pte_t *pte = walk(p->pagetable, p->vmas[i].addr, 0); if((*pte & PTE_V) == 0) // don't write and uvmunmap(), only change data of vma continue; uvmunmap(p->pagetable, p->vmas[i].addr, p->vmas[i].length/PGSIZE, 1); } } begin_op(); iput(p->cwd); end_op(); p->cwd = 0; acquire(&wait_lock); // Give any children to init. reparent(p); // Parent might be sleeping in wait(). wakeup(p->parent); acquire(&p->lock); p->xstate = status; p->state = ZOMBIE; release(&wait_lock); // Jump into the scheduler, never to return. sched(); panic("zombie exit"); }

fork

int fork(void) { int i, pid; struct proc *np; struct proc *p = myproc(); // Allocate process. if((np = allocproc()) == 0){ return -1; } // virtual map vs. real map // Copy user memory from parent to child. int length = 0; for(int i = 0; i < 16; i++){ if(p->vmas[i].length){ length += p->vmas[i].length; } } // use the first p->sz by (p->sz-length) if(uvmcopy(p->pagetable, np->pagetable, p->sz-length) < 0){ freeproc(np); release(&np->lock); return -1; } for(int i = 0; i < 16; i++){ if(p->vmas[i].addr){ np->vmas[i].f = p->vmas[i].f; np->vmas[i].length = p->vmas[i].length; np->vmas[i].prot = p->vmas[i].prot; np->vmas[i].flags = p->vmas[i].flags; np->vmas[i].offset = 0; // ret offset, becasue we maybe read before fork() np->vmas[i].addr = p->vmas[i].addr; np->vmas[i].oldsz = p->vmas[i].oldsz; filedup(p->vmas[i].f); } } np->sz = p->sz; // copy saved user registers. *(np->trapframe) = *(p->trapframe); // Cause fork to return 0 in the child. np->trapframe->a0 = 0; // increment reference counts on open file descriptors. for(i = 0; i < NOFILE; i++) if(p->ofile[i]) np->ofile[i] = filedup(p->ofile[i]); np->cwd = idup(p->cwd); safestrcpy(np->name, p->name, sizeof(p->name)); pid = np->pid; release(&np->lock); acquire(&wait_lock); np->parent = p; release(&wait_lock); acquire(&np->lock); np->state = RUNNABLE; release(&np->lock); return pid; } result

make[1]: Leaving directory '/home/duile/xv6-labs-2021' == Test running mmaptest == $ make qemu-gdb (3.9s) == Test mmaptest: mmap f == mmaptest: mmap f: OK == Test mmaptest: mmap private == mmaptest: mmap private: OK == Test mmaptest: mmap read-only == mmaptest: mmap read-only: OK == Test mmaptest: mmap read/write == mmaptest: mmap read/write: OK == Test mmaptest: mmap dirty == mmaptest: mmap dirty: OK == Test mmaptest: not-mapped unmap == mmaptest: not-mapped unmap: OK == Test mmaptest: two files == mmaptest: two files: OK == Test mmaptest: fork_test == mmaptest: fork_test: OK == Test usertests == $ make qemu-gdb usertests: OK (141.4s) == Test time == time: OK Score: 140/140 总结

  • 完成日期:22.07.19
  • 耗时30h,1h看材料,2h在review,剩下的大部分时间都在debug
  • vma位置的确定花了很长时间,最后才使用p->sz,因为一开始没清楚vma位置的要求是什么,同时对process的本身结构不清楚。
  • munmap一开始写的时候,根本没有考虑到vma结构数据的改变和删除。导致vma的数量一直在增加。如果测试一多,可能就超过16个了
  • lazy alloc当中内存的分配我一开始是直接分配length大小(可能大于PGSIZE),没有考虑kalloc()的大小是PGSIZE,也没有考虑可能从中间读。但在fork_test中才报错,竟然无法执行kalloc(),很长时间无法复现错误,最后差不多是无意中发现的。
  • 因为写的mit s6.0812021 fall,所以10个lab已经结束了,第二个课程labs完成!其中6成是靠自己完成的,很明显,后面3个lab的才收获很大,前面7个lab的收获一般,因为只有一半是独立完成的。以后刷课一定要力求独立完成。调试的gdb没咋学会,printf倒是使用得很开心,写printf时要注明清楚打印意图,能提高debug效率。以后刷课尽快用上gdb进行更细致的debug,打断点功能是printf无法涉及的。
  • 最近在听《穿越时空的少女(Toki o kakeru shôjo)》的:
    • スケッチ(ロング?バージョン)
    • 変わらないもの(ストリングス・バージョン)

本文共计3889个文字,预计阅读时间需要16分钟。

MIT6.S081-Lab10 mmap 2021Fall 的长尾词是:MIT6.S081-Lab10中2021秋季的mmap实验具体难点和实现细节有哪些?

开始日期:22.07.15

操作系统:Ubuntu 20.0.4

实验:mmap

实验内容:

本实验涉及使用mmap系统调用来实现文件映射。我们将学习如何将文件内容映射到内存中,进行读写操作,并探讨其应用场景。

实验步骤:

1. 使用mmap()函数将文件映射到内存。

2.在映射的内存区域进行读写操作。

3.使用munmap()函数解除映射。

实验代码及结果:

c

#include #include #include #include

int main() { int fd=open(file.txt, O_RDWR); if (fd==-1) { perror(open); return 1; }

size_t filesize=lseek(fd, 0, SEEK_END); char *map=mmap(NULL, filesize, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); if (map==MAP_FAILED) { perror(mmap); close(fd); return 1; }

// 在映射的内存区域进行读写操作...

munmap(map, filesize); close(fd);

return 0;}

实验总结:

本实验成功实现了使用mmap进行文件映射,并在映射的内存区域进行了读写操作。通过mmap,我们可以方便地在文件和内存之间进行数据交换,提高程序的性能。

开始日期:22.07.15

操作系统:Ubuntu20.0.4

Link:Lab: mmap

目录
  • Lab mmap
    • 写在前面
      • 踩坑
      • 参考材料
    • 实验内容
      • mmap
        • access code
      • result
    • 总结

Lab mmap 写在前面

此部分不涉及实现细节,可以放心阅读

实现简易的mmapmunmap,需要对virtual address和physical address转换过程中使用的函数有一定理解,同时这里也采用了和cow、lazy alloction一样的lazy alloc方式。

实现之后,即可直接从程序中直接访问file。

踩坑
  • 英文阅读问题

    • as if 等价于 like,意思为“就像”,我错误理解为如果munmap被调用之后才调用exit,导致我一直没思路去修改exit

      Modify exit to unmap the process's mapped regions as if munmap had been called.

      修改exit来解除程序中已映射空间的映射,就像调用munmap一样

    • mmaptest.c中,有一句测试概括,这是个否定句,所以read/write按照英文习惯应该翻译为读写,但我一开始按照中文习惯错误翻译为读或写,导致编程时调试了一段时间才过这个测试

      check that mmap doesn't allow read/write mapping of a file opened read-only.

      检查:mmap不允许读写只读文件的映射

  • p->sz

    p->sz不但是程序的virtual space的大小也是physical space的大小

  • 编程过程中,可能出现关于strcut file报错和PROT_READ等定义的报错,在报错文件的开头处添加strcut file#include "fcntl.h"即可

参考材料
  • book-riscv-rev2
  • riscv-privileged
实验内容

警告:此部分涉及实现细节

mmap

实现要求:实现mmapmunmap,分配物理内存时要采用lazy alloc的方式,解除映射时要写回文件

实现思路:按照hint一步步来即可,但有不少地方需要自己发挥

MIT6.S081-Lab10 mmap 2021Fall 的长尾词是:MIT6.S081-Lab10中2021秋季的mmap实验具体难点和实现细节有哪些?

  • mmap的vma应该选择放在哪里?我采用的是先用程序的p->sz当作映射的虚拟地址,当然p->sz有可能已经被使用了(映射了两个文件),那我们就做检查,往后延长即可,注意这样做成功之后,肯定要将原p->sz增加length
  • munmap中要对p->sz、vma结构数据的lengthaddrf等数据进行处理,详细处理见注释,从而舍弃掉将要被解除的区域。同时在munmap中要注意写回文件这个操作,写回文件时要判断是不是第一次写回文件从而判断要不要写回时进行偏移(使用oldsz判断)。除此之外,当解除实际上没有映射到物理空间的vma时,我们不执行uvmunmap(),也不会执行写回文件,因为这个vma对应的pte中的有效位(PTE_V)没有在lazy alloc中被设置。
  • lazy alloc中,我写了个辅助函数mmap_lazyalloc,每次只分配一个page,为了能够在文件的中间部分就被读,当然,分配之前要先进行的检查。最后需要注意readi()被调用时,读文件时的偏移应该是多少,用va减去vma地址即可求出偏移。
  • exit中,类似于munmap,但可以减少一些vma结构数据处理,因为最后结构都会被释放掉。注意在这里写回文件时,使用的是vma的length。而不像munmap使用的是传入的length
  • fork中,我没有采用父子共享vmas的方式(没那么酷 = =),直接copy过去了,copy时vmas不用copy,因为父程序的vmas可能还没lazy alloc。这里需要注意一点,就是p->sz已经被改变了,我们要算出最开始没有vmas的p->sz,用当前的p->sz减掉所有vmas的length即可。
access code

首先是mmapmunmap,这两个syscall的设置,参考lab syscall即可,在此不赘述

然后是定义

/* proc.h */ struct vma_t { // int fd; struct file *f; int length; int prot; int flags; int offset; int oldsz; uint64 addr; }; // Per-process state struct proc { struct spinlock lock; ... char name[16]; // Process name (debugging) struct vma_t vmas[16]; // VMAs helps the kernel to decide how to handle page faults };

sys_mmap

/* */ uint64 sys_mmap(void) { struct file *f; int length , prot, flags, offset; uint64 addr; // get args, 0:addr, 1:length, 2:prot, 3:flags, 5:offset, 4:fd=>*f if(argaddr(0, &addr) < 0 || argint(1, &length) < 0 || argint(2, &prot) < 0 || argint(3, &flags) < 0 || argint(5, &offset) < 0 || argfd(4, 0, &f) < 0 ) return -1; // mmap doesn't allow read/write mapping of a file opened read-only with MAP_SHARED // "or", "/" also meaning "and" in english when use "doesn't", "not" and so on if(f->readable && !f->writable && (prot & PROT_READ) && (prot & PROT_WRITE) && (flags & MAP_SHARED)) return -1; struct proc *p = myproc(); int oldsz = p->sz; // get va without anything by p->sz if(addr == 0){ uint64 va = p->sz; int is_same = 0; while(1){ // avoid: some VAs may get same addr, becase we use 'lazy alloc' int i ; for(i = 0; i < 16; i++){ if(p->vmas[i].addr <= va && va < (p->vmas[i].addr + p->vmas[i].length)) is_same = 1; } if(!is_same){ addr = va; p->sz = va + length; break; } // reset is_same is_same = 0; if(va >= MAXVA) return -1; va += PGSIZE; } } for(int i = 0; i < 16; i++){ // case: maybe remove and rebuild a vmas[i] when slots full // not case in test // build a vams[i] if(p->vmas[i].addr == 0){ p->vmas[i].f = f; p->vmas[i].length = length; p->vmas[i].prot = prot; p->vmas[i].flags = flags; p->vmas[i].offset = offset; p->vmas[i].addr = addr; p->vmas[i].oldsz = oldsz; // mmap should increase the file's reference count filedup(f); return addr; } } // failed, ret -1 return -1; }

sys_munmap

uint64 sys_munmap(void) { uint64 addr; int length; // get args, 0:addr, 1:length if(argaddr(0, &addr) < 0 || argint(1, &length) < 0) return -1; struct proc *p = myproc(); int has_a_vma = 0; int i; for(i = 0 ; i < 16; i++){ if(p->vmas[i].addr <= addr && addr < (p->vmas[i].addr + p->vmas[i].length)){ has_a_vma = 1; break; } } // not find a vma, so ret -1 if(has_a_vma == 0){ return -1; } pte_t *pte = walk(p->pagetable, addr, 0); int offset = p->vmas[i].f->off; // addr is beginning of vma, so use file->off if(p->vmas[i].oldsz != addr) // addr is't beginning of vma, so use p->vmas[i].offset offset = p->vmas[i].offset; // If an unmapped page has been modified and the file is mapped MAP_SHARED, // write the page back to the file. if((*pte & PTE_V) && p->vmas[i].flags & MAP_SHARED){ begin_op(); ilock(p->vmas[i].f->ip); writei(p->vmas[i].f->ip, 1, addr, offset, length); iunlock(p->vmas[i].f->ip); end_op(); } // If munmap removes all pages of a previous mmap, // it should decrement the reference count of the corresponding struct file. // we keep end of old addr by 'p->vmas[i].addr += length' and 'p->vmas[i].length -= length' // we check by 'addr < (p->vmas[i].addr + p->vmas[i].length' in sys_mmap() // so we can't mmap [length munmap] and we will mmap after [p->vmas[i].length] // figure: // ' p->vmas[i].addr // [process data][ p->vmas[i].length ] // [ p->sz ] // ==> // ' p->vmas[i].addr // [process data][ length munmap ][p->vmas[i].length] // [ p->sz ] if(length < p->vmas[i].length){ p->sz -= length; p->vmas[i].addr += length; p->vmas[i].length -= length; } else if(length == p->vmas[i].length){ p->sz -= length; p->vmas[i].f->ref--; p->vmas[i].f = 0; p->vmas[i].addr = 0; p->vmas[i].prot = 0; p->vmas[i].flags = 0; p->vmas[i].length = 0; p->vmas[i].oldsz = 0; } else return -1; if((*pte & PTE_V) == 0) // don't write and uvmunmap(), only change data of vma return 0; // uvmunmap() after writing // find the VMA for the address range and unmap the specified pages // note: free physical memory => 'do_free = 1' uvmunmap(p->pagetable, addr, length/PGSIZE, 1); // success, ret 0 return 0; }

lazy alloc是当作page fault(13)来处理的

/* trap.c */ int mmap_lazyalloc(pagetable_t pagetable, uint64 va) { struct proc *p = myproc(); struct file *f; int prot; int has_a_vam = 0; int perm = 0; char *mem; // find va between vma.addr and vma.addr+vma.lenght int i; for(i = 0; i < 16; i++){ if(p->vmas[i].addr <= va && va < (p->vmas[i].addr + p->vmas[i].length)){ has_a_vam = 1; f = p->vmas[i].f; prot = p->vmas[i].prot; break; } } // not find vma, ret -1 if(has_a_vam == 0){ return -1; } // PTE_U controls whether instructions in user mode are allowed to access the page; // if PTE_U is notset, the PTE can be used only in supervisor mode. perm |= PTE_U; // MAYBE sets PTE_R, PTE_W, PTE_X if(prot & PROT_READ){ perm |= PTE_R; } if(prot & PROT_WRITE){ perm |= PTE_W; } if(prot & PROT_EXEC){ perm |= PTE_X; } // big bug: not alloc mem(4096) to all virtual addresses if((mem = kalloc()) == 0){ return -1; } // In mmaptest/makefile() // create a file to be mapped, containing // 1.5 pages of 'A' and half a page of zeros. // so we must set 0 of length after getting mem memset(mem, 0, PGSIZE); // note: mem is new address of phycial memory if(mappages(pagetable, va, PGSIZE, (uint64)mem, perm) == -1){ kfree(mem); return -1; } // we not set PTE_D, becasue we always directly wirite back to file in munmap() // length is the number of bytes to map; it might not be the same as the file's length. // read data from file, then put data to va ilock(f->ip); if(readi(f->ip, 1, va, va - p->vmas[i].addr, PGSIZE) < 0){ // readi offset by 'va - p->vmas[i].addr' iunlock(f->ip); return -1; } iunlock(f->ip); p->vmas[i].offset += PGSIZE; // success, ret 0 return 0; } void usertrap(void) { int which_dev = 0; if((r_sstatus() & SSTATUS_SPP) != 0) panic("usertrap: not from user mode"); // send interrupts and exceptions to kerneltrap(), // since we're now in the kernel. w_stvec((uint64)kernelvec); struct proc *p = myproc(); // save user program counter. p->trapframe->epc = r_sepc(); if(r_scause() == 8){ // system call if(p->killed) exit(-1); // sepc points to the ecall instruction, // but we want to return to the next instruction. p->trapframe->epc += 4; // an interrupt will change sstatus &c registers, // so don't enable until done with those registers. intr_on(); syscall(); // Fill in the page table lazily, in response to page faults. } else if(r_scause() == 13){ uint64 fault_va = r_stval(); int is_alloc = mmap_lazyalloc(p->pagetable, fault_va); if(fault_va > p->sz || is_alloc == -1){ p->killed = 1; } } else if((which_dev = devintr()) != 0){ // ok } else { printf("usertrap(): unexpected scause %p pid=%d\n", r_scause(), p->pid); printf(" sepc=%p stval=%p\n", r_sepc(), r_stval()); p->killed = 1; } if(p->killed) exit(-1); // give up the CPU if this is a timer interrupt. if(which_dev == 2) yield(); usertrapret(); }

exit

void exit(int status) { struct proc *p = myproc(); if(p == initproc) panic("init exiting"); // Close all open files. for(int fd = 0; fd < NOFILE; fd++){ if(p->ofile[fd]){ struct file *f = p->ofile[fd]; fileclose(f); p->ofile[fd] = 0; } } // 'as if' == 'like' for(int i = 0; i < 16; i++){ if(p->vmas[i].addr){ int offset = p->vmas[i].f->off; // addr is beginning of vma, so use file->off if(p->vmas[i].oldsz != p->vmas[i].addr) // addr is't beginning of vma, so use p->vmas[i].offset offset = p->vmas[i].offset; if(p->vmas[i].flags & MAP_SHARED){ begin_op(); ilock(p->vmas[i].f->ip); writei(p->vmas[i].f->ip, 1, p->vmas[i].addr, offset, p->vmas[i].length); iunlock(p->vmas[i].f->ip); end_op(); } p->sz -= p->vmas[i].length; p->vmas[i].f->ref--; pte_t *pte = walk(p->pagetable, p->vmas[i].addr, 0); if((*pte & PTE_V) == 0) // don't write and uvmunmap(), only change data of vma continue; uvmunmap(p->pagetable, p->vmas[i].addr, p->vmas[i].length/PGSIZE, 1); } } begin_op(); iput(p->cwd); end_op(); p->cwd = 0; acquire(&wait_lock); // Give any children to init. reparent(p); // Parent might be sleeping in wait(). wakeup(p->parent); acquire(&p->lock); p->xstate = status; p->state = ZOMBIE; release(&wait_lock); // Jump into the scheduler, never to return. sched(); panic("zombie exit"); }

fork

int fork(void) { int i, pid; struct proc *np; struct proc *p = myproc(); // Allocate process. if((np = allocproc()) == 0){ return -1; } // virtual map vs. real map // Copy user memory from parent to child. int length = 0; for(int i = 0; i < 16; i++){ if(p->vmas[i].length){ length += p->vmas[i].length; } } // use the first p->sz by (p->sz-length) if(uvmcopy(p->pagetable, np->pagetable, p->sz-length) < 0){ freeproc(np); release(&np->lock); return -1; } for(int i = 0; i < 16; i++){ if(p->vmas[i].addr){ np->vmas[i].f = p->vmas[i].f; np->vmas[i].length = p->vmas[i].length; np->vmas[i].prot = p->vmas[i].prot; np->vmas[i].flags = p->vmas[i].flags; np->vmas[i].offset = 0; // ret offset, becasue we maybe read before fork() np->vmas[i].addr = p->vmas[i].addr; np->vmas[i].oldsz = p->vmas[i].oldsz; filedup(p->vmas[i].f); } } np->sz = p->sz; // copy saved user registers. *(np->trapframe) = *(p->trapframe); // Cause fork to return 0 in the child. np->trapframe->a0 = 0; // increment reference counts on open file descriptors. for(i = 0; i < NOFILE; i++) if(p->ofile[i]) np->ofile[i] = filedup(p->ofile[i]); np->cwd = idup(p->cwd); safestrcpy(np->name, p->name, sizeof(p->name)); pid = np->pid; release(&np->lock); acquire(&wait_lock); np->parent = p; release(&wait_lock); acquire(&np->lock); np->state = RUNNABLE; release(&np->lock); return pid; } result

make[1]: Leaving directory '/home/duile/xv6-labs-2021' == Test running mmaptest == $ make qemu-gdb (3.9s) == Test mmaptest: mmap f == mmaptest: mmap f: OK == Test mmaptest: mmap private == mmaptest: mmap private: OK == Test mmaptest: mmap read-only == mmaptest: mmap read-only: OK == Test mmaptest: mmap read/write == mmaptest: mmap read/write: OK == Test mmaptest: mmap dirty == mmaptest: mmap dirty: OK == Test mmaptest: not-mapped unmap == mmaptest: not-mapped unmap: OK == Test mmaptest: two files == mmaptest: two files: OK == Test mmaptest: fork_test == mmaptest: fork_test: OK == Test usertests == $ make qemu-gdb usertests: OK (141.4s) == Test time == time: OK Score: 140/140 总结

  • 完成日期:22.07.19
  • 耗时30h,1h看材料,2h在review,剩下的大部分时间都在debug
  • vma位置的确定花了很长时间,最后才使用p->sz,因为一开始没清楚vma位置的要求是什么,同时对process的本身结构不清楚。
  • munmap一开始写的时候,根本没有考虑到vma结构数据的改变和删除。导致vma的数量一直在增加。如果测试一多,可能就超过16个了
  • lazy alloc当中内存的分配我一开始是直接分配length大小(可能大于PGSIZE),没有考虑kalloc()的大小是PGSIZE,也没有考虑可能从中间读。但在fork_test中才报错,竟然无法执行kalloc(),很长时间无法复现错误,最后差不多是无意中发现的。
  • 因为写的mit s6.0812021 fall,所以10个lab已经结束了,第二个课程labs完成!其中6成是靠自己完成的,很明显,后面3个lab的才收获很大,前面7个lab的收获一般,因为只有一半是独立完成的。以后刷课一定要力求独立完成。调试的gdb没咋学会,printf倒是使用得很开心,写printf时要注明清楚打印意图,能提高debug效率。以后刷课尽快用上gdb进行更细致的debug,打断点功能是printf无法涉及的。
  • 最近在听《穿越时空的少女(Toki o kakeru shôjo)》的:
    • スケッチ(ロング?バージョン)
    • 変わらないもの(ストリングス・バージョン)