Transcript File Systems I

Operating Systems
CMPSCI 377
Lecture 15: File Systems I
Emery Berger
University of Massachusetts, Amherst
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
Course Outline

Processes & Threads





CPU Scheduling
Synchronization & Deadlock
Memory Management
File Systems & I/O
Distributed Systems
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
2
Files: OS Abstraction

Files: another OS-provided abstraction over
hardware resources
OS Abstraction
Hardware Resource
Processes
Threads
Address space
CPU
Files
Disk
Memory
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
3
File System Abstraction

Applications operate on files in a file system




Device-independent interface
open(), close(), link(), read(),
write(), rename()
Device-level interface

Manage disk in terms of sectors, tracks, & blocks

seek(), readblock(), writeblock()
OS converts these to operations on raw
hardware (disk)
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
4
User Expectations on Data

Persistence


Speed


Ability to store lots of data
Sharing/protection


Quick access to data
Size


Data lives across jobs, power outages, crashes
Users can share or restrict access to data when
appropriate
Ease of use

User can easily find, examine & modify data
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
5
Hardware/OS Support for Data

hardware





operating
system
Persistence: disks: non-volatile memory
Speed: random access devices
Size: disk capacity growing fast
Persistence: redundancy, fault-tolerance
Sharing/protection: UNIX privileges
Ease of use:
 Names associated with data (files)
 Hierarchical directories
 Transparent mapping of devices
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
6
Files

Named collection of related information recorded
on secondary storage




Can contain programs (source, binary) or data
Can be structured or unstructured



Logical unit of storage on a device
e.g., helloworld.cpp, resume.doc
IBM mainframe OS: series of records (structured)
UNIX: file = stream of bytes (unstructured)
Files have attributes:

name, type, location, size, protection, creation time...
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
7
User Interface to File System

Data operations:




create(), delete()
open(), close()
read(), write(), seek()
Naming operations:


get & set attributes (ownership, protection)
hard link, soft links, rename
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
8
OS File-Related Data Structures

Open file table – shared by all processes with open file





open count
file attributes (ownership, protection, etc.)
location(s) of file on disk
pointers to location(s) of file in memory
Per-process file table – one for each file




Pointer to entry in open file table
Current position in file (offset)
Mode in which process accesses file (r, w, r/w...)
Pointers to file buffer
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
9
File Operations: Creating a File

Create(name)

Allocates disk space


Creates file descriptor for file



Checks disk quotas, permissions, etc.
(name, location on disk, attributes)
Adds file descriptor to directory that contains file
May mark file with type attribute (esp. Mac)



Advantages: error detection, launch appropriate app
Disadvantages: not supported everywhere, complicates file
system and OS, less flexible
UNIX: no types, Windows: file extensions indicate type
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
10
File Operations: Deleting a File

Delete(name)



Find directory containing file
Free disk blocks used by file
Remove file descriptor from directory
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
11
File Operations:
Opening Files

fd = open(name, mode)


Check if file is already open by another process
If not:



Check protection of file against requested mode




Find file
Copy file descriptor into system-wide open file table
If not OK, abort
Increment open count
Create entry in process’s file table pointing to entry in
system-wide table
Initialize current file pointer to start of file
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
12
File Operations: Closing Files

close (fd)



Remove entry for file in process’s file table
Decrement open count in system-wide file table
If open count == 0:

Remove entry from system-wide file table
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
13
OS File Operations: Reading File

Random access: read(fd, from, size, buf)

OS reads “size” bytes from file position “from” into
“buf ”
for (i = from; i < from + size; i++)
buf[i – from] = file[i];

Sequential access: read(fd, size, buf)

OS reads “size” bytes from current file position “from”
into “buf ”; increments file position
for (i = 0; i < size; i++)
buf[i] = file[fp + i];
fp += size;
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
14
OS File Operations

write


seek


Similar to read, but copies from buffer to file
Just updates fp
memory mapping file



map portion of virtual address space to file
read/writes to that memory = OS reads/writes
from corresponding location in file
avoids need for explicit reads/writes
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
15
File Access Methods

Common file access patterns from programmer’s
perspective

Sequential: data processed in order



Keyed: address block based on key table


Most programs use this method
Example: compiler reading source file
Example: database search, hash table, dictionary
Common file access patterns from OS perspective


Sequential: pointer to next byte; update on read/write
Random: address any block directly given offset in file
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
16
Naming & Directories

Need method of retrieving files from disk!



OS uses numbers (inodes) for files
People prefer names...
OS provides directory to map names to file
descriptors
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
17
Flat File Systems

One level directory

One namespace for entire disk, every name
unique



Directory contains (name, index) pairs
Used by Apple, CP/M, DOS 1.0, first MacOS
Two level directories

Separate directories for each user
Not a problem for 140K disks, but...
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
18
Hierarchical File Systems

Tree-structured name space


Used by all modern operating systems
Directory becomes special file on disk




Marked by special flag bit
User programs may read directories, but only
system may manipulate directories
Each directory contains (name, inode) pairs,
names need not be unique in file system
Distinguished root directory (UNIX) or drive
names (Windows)
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
19
Referential Naming

Hard links (UNIX: ln command)


Allows multiple links to single file
Example: “ln A B”:



OS maintains reference counts


Deletes file only after last link to it is deleted
Problem: users could create circular links with
directories


initially: A -> file # 100
after: A, B -> file # 100
Reference counting can’t reclaim cycles
Solution: can’t hard link directories
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
20
Referential Naming

Soft (symbolic) links (UNIX: ln -s)


Makes symbolic pointer from one file to another
“ln -s A B”:





Removing B does not affect A
Removing A: dangling pointer
Problem: Circular links can cause infinite loops


Initially, A -> file #100
After, A -> file #100, B -> A
E.g., list all files in directory & its subdirectories
Solution: (lame) Limit number of links traversed
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
21
Directory Operations






Search for file: locate entry
Create file: adds directory listing
Delete file: removes directory listing
List directory: list all files (ls)
Rename file
Traverse file system
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
22
Protection



OS must allow users to control access to files
Grant or deny access to file operations depending
on protection information
Access lists and groups (Windows)



Access list for each file with user name and access type
Lists can become large & tedious to maintain
Access control bits (UNIX)



Three categories of user (owner, group, world)
Three types of access privileges (read, write, execute)
One bit per operation (111101000 = rwxr-x----)
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
23
Summary

File systems provide





Naming
Protection
Persistence
Fast access
Next time: file system implementation
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
24