Building File Systems with.ppt

《Building File Systems with.ppt》由会员分享，可在线阅读，更多相关《Building File Systems with.ppt（40页珍藏版）》请在麦多课文档分享上搜索。

1、Building File Systems with,Xavid Pretzer SIPB IAP 2009,What is FUSE?,Stands for “File system in USErspace”,Whats a File System?,A file system maps file paths (e.g., /etc/hostname) to file contents and metadata Metadata includes modification times, permissions, etc. File systems are mounted over a pa

2、rticular directory,What is Userspace?,Your operating system has (at least) two modes: kernel (trusted) and user Kernelspace code has real ultimate power and can only be modified by root Base system software like filesystems are traditionally kernel modules and not changeable by normal users,FUSE,Mak

3、es it easy to write new filesystems without knowing how the kernel works without breaking unrelated things more quickly/easily than traditional file systems built as a kernel module Makes it safe for sysadmins to let users they dont trust use custom file systems,Other Key Features,Cross-platform: Li

4、nux/BSD/OS X Wide language support: natively in C, with bindings in C+, Java, C#, Haskell, TCL, Python, Perl, Shell Script, SWIG, OCaml, Pliant, Ruby, Lua, Erlang, PHP (My examples use Python) Low-level interface for more efficient file systems,What do people do with FUSE?,Hardware-based: ext2, iso,

5、 ZFS Network-based: NFS, smb, SSH Nontradtional: Gmail, MySQL Loopback: compression, conversion, encryption, virus scanning, versioning Synthetic: search results, application interaction, dynamic conf files,Using FUSE Filesystems,To mount: ./hello.py /somedir To unmount: fusermount -u /somedir,How F

6、USE Works,Application makes a file-related syscall Kernel figures out that the file is in a mounted FUSE filesystem The FUSE kernel module forwards the request to your userspace FUSE app Your app tells FUSE how to reply,Writing FUSE Filesystems,Writing a FUSE Filesystem,Write an ordinary application

7、 that defines certain functions/methods that FUSE will call to handle operations 35 possible operations Many operations have useful defaults Useful filesystems can define only 4 Full-featured ones will need to define most,Defining FUSE Operations,In C, you define functions and put pointers to them o

8、n a struct In python-fuse, operations are methods on a subclass of fuse.Fuse You can set your Fuse subclasss file_class attribute to a class that implements the file operations, or implement them on your Fuse subclass,FUSE Operations,Directory Operations File Operations Metadata Operations Some othe

9、r stuff,Directory Operations,readdir(path): yield directory entries for each file in the directory mkdir(path, mode): create a directory rmdir(path): delete an empty directory,Basic File Operations,mknod(path, mode, dev): create a file (or device) unlink(path): delete a file rename(old, new): move a

10、nd/or rename a file,Reading and Writing Files,open(path, flags): open a file read(path, length, offset, fh) write(path, buf, offset, fh) truncate(path, len, fh): cut off at length flush(path, fh): one handle is closed release(path, fh): file handle is completely closed (no errors),Metadata Operation

11、s,getattr(path): read metadata chmod(path, mode): alter permissions chown(path, uid, gid): alter ownership,Meta Operations,fsinit(self): initialize filesystem state after being mounted start threads, for example,Other Operations,statfs(path) fsdestroy() create(path, flags, mode) utimens(path, times)

12、 readlink(path) symlink(target, name) link(target, name) fsync(path, fdatasync, fh) ,Metadata Format,self.st_size: size in bytes st_mode: type and permissions self.st_uid: owner id self.st_gid: group id self.st_atime: access time (often fudged) self.st_mtime: modification time self.st_ctime: metadat

13、a change time self.st_ino: doesnt matter too much self.st_dev: 0 for normal files/directories self.st_nlink: 2 for dirs, 1 for files (generally),FUSE Context,GetContext() within a Fuse object returns a dict with: uid: accessing users user ID gid: accessing users group ID pid: accessing processs ID U

14、seful for nonstandard permission models and other user-specific behavior,Errors in FUSE,Dont have access to the users terminal (if any), and can only send predefined codes from the errno module Return -the error code to indicate failure Can log arbitrary messages to a log file for debugging,Useful E

15、rrors,errno.ENOSYS: Function not implemented errno.EROFS: Read-only file system errno.EPERM: Operation not permitted errno.EACCES: Permission denied errno.ENOENT: No such file or directory errno.EIO: I/O error errno.EEXIST: File exists errno.ENOTDIR: Not a directory errno.EISDIR: Is a directory errn

16、o.ENOTEMPTY: Directory not empty,Examples,Example: hello.py,Minimal synthetic file system Holds a single immutable file with a pre-defined message Could easily be adapted to run arbitrary code to generate the file contents Uses 4 operations readdir, open, read, getattr,readdir,fuse.fuse_python_api =

17、 (0, 2)hello_path = /hello hello_str = Hello World!n class HelloFS(Fuse):def readdir(self, path, offset):for r in ., , hello_path1:yield fuse.Direntry(r),open,hello_path = /hello hello_str = Hello World!n class HelloFS(Fuse):# .def open(self, path, flags):if path != hello_path:return -errno.ENOENTac

18、cmode = os.O_RDONLY | os.O_WRONLY | os.O_RDWRif (flags & accmode) != os.O_RDONLY:return -errno.EACCES,read,def read(self, path, size, offset):if path != hello_path:return -errno.ENOENTslen = len(hello_str)if offset slen:size = slen - offsetbuf = hello_stroffset:offset+sizeelse:buf = return buf,Helpe

19、r Stat subclass,class MyStat(fuse.Stat):def _init_(self):self.st_mode = 0self.st_ino = 0self.st_dev = 0self.st_nlink = 0self.st_uid = 0self.st_gid = 0self.st_size = 0self.st_atime = 0self.st_mtime = 0self.st_ctime = 0,getattr,def getattr(self, path):st = MyStat()if path = /:st.st_mode = stat.S_IFDIR

20、 | 0755st.st_nlink = 2elif path = hello_path:st.st_mode = stat.S_IFREG | 0444st.st_nlink = 1st.st_size = len(hello_str)else:return -errno.ENOENTreturn st,Boilerplate Main,def main():usage=“nUserspace hello examplenn“ + Fuse.fusageserver = HelloFS(version=“%prog “+ fuse._version_,usage=usage,dash_s_d

21、o=setsingle)server.parse(errex=1)server.main()if _name_ = _main_:main(),Example: xmp.py,Mirrors a local file hierarchy Simple to implement using functions in the os module Shows how many operations work Usage: ./xmp.py -o root=/mit/sipb/ /tmp/mntdir,_init_ and fsinit,fuse.fuse_python_api = (0, 2)# W

22、e use a custom file class and fsinit feature_assert(stateful_files, has_init)class Xmp(Fuse):def _init_(self, *args, *kw):Fuse._init_(self, *args, *kw)self.root = /self.file_class = self.XmpFiledef fsinit(self):os.chdir(self.root),Main with Options,def main():server = Xmp(version=“%prog “ + fuse._ve

23、rsion_, usage=Fuse.fusage)server.parser.add_option(mountopt=“root“, metavar=“PATH“, default=/,help=“mirror PATH def: %default“)server.parse(values=server, errex=1)if server.fuse_args.mount_expected():os.chdir(server.root)server.main(),Operations on Fuse Subclass,def getattr(self, path):return os.lst

24、at(“.“ + path)def readdir(self, path, offset):for e in os.listdir(“.“ + path):yield fuse.Direntry(e)def truncate(self, path, len):f = open(“.“ + path, “a“)f.truncate(len)f.close()# .,Operations on File class,class XmpFile(object):# Called for opendef _init_(self, path, flags, *mode):self.file = os.f

25、dopen(os.open(“.“ + path, flags, *mode),flag2mode(flags)self.fd = self.file.fileno()def read(self, length, offset):self.file.seek(offset)return self.file.read(length)def write(self, buf, offset):self.file.seek(offset)self.file.write(buf)return len(buf)# .,Examples,For full details, see xmp.py http:/

26、stuff.mit.edu/iap/2009/fuse/examples/ /mit/sipb-iap/www/2009/fuse/examples/ Also look at pyhesiodfs.py there Used on Debathena machines for /mit/ Very simple, yet useful and used widely,Try it Out,ssh iap-fuse.xvm.mit.edu Play with the examples: http:/stuff.mit.edu/iap/2009/fuse/examples/ /mit/sipb-

27、iap/www/2009/fuse/examples/ Ask me questions Write your own! Some fun ideas are at: http:/stuff.mit.edu/iap/2009/fuse/practice.html,fuse_lowlevel.h,C only Uses numeric ino identifiers instead of always passing full paths Less friendly interface (more similar to kernel interface) allows FUSE to add less overhead,