In this lab you will add large files and mmap to the xv6 file system.
Large files
In this assignment you'll increase the maximum size of an xv6
file. Currently xv6 files are limited to 268 blocks, or 268*BSIZE
bytes (BSIZE is 1024 in xv6). This limit comes from the fact that an
xv6 inode contains 12 "direct" block numbers and one "singly-indirect"
block number, which refers to a block that holds up to 256 more block
numbers, for a total of 12+256=268. You'll change the xv6 file system
code to support a "doubly-indirect" block in each inode, containing
256 addresses of singly-indirect blocks, each of which can contain up
to 256 addresses of data blocks. The result will be that a file will
be able to consist of up to 256*256+256+11 blocks (11 instead of 12,
because we will sacrifice one of the direct block numbers for the
double-indirect block).
Preliminaries
Modify your Makefile's CPUS definition so that it reads:
CPUS := 1
XXX doesn't seem to speedup things
Add
QEMUEXTRA = -snapshot
right before
QEMUOPTS
The above two steps speed up qemu tremendously when xv6
creates large files.
mkfs initializes the file system to have fewer
than 1000 free data blocks, too few to show off the changes
you'll make. Modify param.h to
set FSSIZE to:
#define FSSIZE 20000 // size of file system in blocks
Download big.c into your xv6 directory,
add it to the UPROGS list, start up xv6, and run big.
It creates as big a file as xv6 will let
it, and reports the resulting size. It should say 140 sectors.
What to Look At
The format of an on-disk inode is defined by struct dinode
in fs.h. You're particularly interested in NDIRECT,
NINDIRECT, MAXFILE, and the addrs[] element
of struct dinode. Look Figure 7.3 in the xv6 text for a
diagram of the standard xv6 inode.
The code that finds a file's data on disk is in bmap()
in fs.c. Have a look at it and make sure you understand
what it's doing. bmap() is called both when reading and
writing a file. When writing, bmap() allocates new
blocks as needed to hold file content, as well as allocating
an indirect block if needed to hold block addresses.
bmap() deals with two kinds of block numbers. The bn
argument is a "logical block" -- a block number relative to the start
of the file. The block numbers in ip->addrs[], and the
argument to bread(), are disk block numbers.
You can view bmap() as mapping a file's logical
block numbers into disk block numbers.
Your Job
Modify bmap() so that it implements a doubly-indirect
block, in addition to direct blocks and a singly-indirect block.
You'll have to have only 11 direct blocks, rather than 12,
to make room for your new doubly-indirect block; you're
not allowed to change the size of an on-disk inode.
The first 11 elements of ip->addrs[] should be
direct blocks; the 12th should be a singly-indirect block
(just like the current one); the 13th should be your new
doubly-indirect block.
You don't have to modify xv6 to handle deletion of files with
doubly-indirect blocks.
If all goes well, big will now report that it
can write sectors. It will take big minutes
to finish.
XXX this runs for a while!
Hints
Make sure you understand bmap(). Write out a diagram of the
relationships between ip->addrs[], the indirect block, the
doubly-indirect block and the singly-indirect blocks it points to, and
data blocks. Make sure you understand why adding a doubly-indirect
block increases the maximum file size by 256*256 blocks (really -1),
since you have to decrease the number of direct blocks by one).
Think about how you'll index the doubly-indirect block, and
the indirect blocks it points to, with the logical block
number.
If you change the definition of NDIRECT, you'll
probably have to change the size of addrs[]
in struct inode in file.h. Make sure that
struct inode and struct dinode have the
same number of elements in their addrs[] arrays.
If you change the definition of NDIRECT, make sure to create a
new fs.img, since mkfs uses NDIRECT too to build the
initial file systems. If you delete fs.img, make on Unix (not
xv6) will build a new one for you.
If your file system gets into a bad state, perhaps by crashing,
delete fs.img (do this from Unix, not xv6). make will build a
new clean file system image for you.
Don't forget to brelse() each block that you
bread().
You should allocate indirect blocks and doubly-indirect
blocks only as needed, like the original bmap().
Memory-mapped files
In this assignment you will implement the core of the systems
calls mmap and munmap; see the man pages for an
explanation what they do (run man 2 mmap in your terminal).
The test program mmaptest tells you what should work.
Here are some hints about how you might go about this assignment:
Start with adding the two systems calls to the kernel, as you
done for other systems calls (e.g., sigalarm), but
don't implement them yet; just return an
error. run mmaptest to observe the error.
Keep track for each process what mmap has mapped.
You will need to allocate a struct vma to record the
address, length, permissions, etc. for each virtual memory area
(VMA) that maps a file. Since the xv6 kernel doesn't have a
memory allocator in the kernel, you can use the same approach has
for struct file: have a global array of struct
vmas and have for each process a fixed-sized array of VMAs
(like the file descriptor array).
Implement mmap: allocate a VMA, add it to the process's
table of VMAs, fill in the VMA, and find a hole in the process's
address space where you will map the file. You can assume that no
file will be bigger than 1GB. The VMA will contain a pointer to
a struct file for the file being mapped; you will need to
increase the file's reference count so that the structure doesn't
disappear when the file is closed (hint:
see filedup). You don't have worry about overlapping
VMAs. Run mmaptest: the first mmap should
succeed, but the first access to the mmaped- memory will fail,
because you haven't updated the page fault handler.
Modify the page-fault handler from the lazy-allocation and COW
labs to call a VMA function that handles page faults in VMAs.
This function allocates a page, reads a 4KB from the mmap-ed
file into the page, and maps the page into the address space of
the process. To read the page, you can use readi,
which allows you to specify an offset from where to read in the
file (but you will have to lock/unlock the inode passed
to readi). Don't forget to set the permissions correctly
on the page. Run mmaptest; you should get to the
first munmap.
Implement munmap: find the struct vma for
the address and unmap the specified pages (hint:
use uvmunmap). If munmap removes all pages
from a VMA, you will have to free the VMA (don't forget to
decrement the reference count of the VMA's struct
file); otherwise, you may have to shrink the VMA. You can
assume that munmap will not split a VMA into two VMAs;
that is, we don't unmap a few pages in the middle of a VMA. If
an unmapped page has been modified and the file is
mapped MAP_SHARED, you will have to write the page back
to the file. RISC-V has a dirty bit (D) in a PTE to
record whether a page has ever been written too; add the
declaration to kernel/riscv.h and use it. Modify exit
to call munmap for the process's open VMAs.
Run mmaptest; you should mmaptest, but
probably not forktest.
Modify fork to copy VMAs from parent to child. Don't
forget to increment reference count for a VMA's struct
file. In the page fault handler of the child, it is OK to
allocate a new page instead of sharing the page with the
parent. The latter would be cooler, but it would require more
implementation work. Run mmaptest; make sure you pass
both mmaptest and forktest.
Run usertests to make sure you didn't break anything.
Optional challenges:
If two processes have the same file mmap-ed (as
in forktest), share their physical pages. You will need
reference counts on physical pages.
The solution above allocates a new physical page for each page
read from the mmap-ed file, even though the data is also in kernel
memory in the buffer cache. Modify your implementation to mmap
that memory, instead of allocating a new page. This requires that
file blocks be the same size as pages (set BSIZE to
4096). You will need to pin mmap-ed blocks into the buffer cache.
You will need worry about reference counts.
Remove redundancy between your implementation for lazy
allocation and your implementation of mmapp-ed files. (Hint:
create an VMA for the lazy allocation area.)
Modify exec to use a VMA for different sections of
the binary so that you get on-demand-paged executables. This will
make starting programs faster, because exec will not have
to read any data from the file system.
Implement on-demand paging: don't keep a process in memory,
but let the kernel move some parts of processes to disk when
physical memory is low. Then, page in the paged-out memory when
the process references it.