1
Q
What is the File System for?
A
File System provides:
- Large and cheap storage space
- Non-volatility
- Sharing information between processes
2
Q
What is a File System?
A
It is a collection of files + directory structures
3
Q
What are files?
A
A container for storing information.
File abstracts away the different kinds of information into a single struct - devices, pipes, text files, data files.
4
Q
Where are files stored?
A
Generally, they are stored on disk.
5
Q
what are the different kinds of representations a file has?
A
Three kinds of files:
- byte sequence
- record sequence
- tree of records
6
Q
What kind of types a file has?
A
User files:
- ASCII
- Binary
System files
- Directories
- Special files
7
Q
How many kinds of access-to-file we have?
A
Sequential access:
- Read all bytes/records from the beginning
- cannot jump around
Random access
- bytes/records read in any order
- All files of modren OSs are random access
- read/write functions can
- recieve an offset parameter
- seek(offset, start_from)
- start_from = {0,1,2}
- 0 - beginning
- 1 - current
- 2 - end
- start_from = {0,1,2}
8
Q
What are the file attributes?
A
General info - user ID, Group ID, dates, times
Location & size - pointer to a device and location on it
Flags that store information for the system
Another special flags - Protection, password..
9
Q
What are the file’s operations?
A
Create ; Delete
Open ; Close
Write ; Read ; Seek
Get attributes
Set attributes
10
Q
What are the directory’s operations?
A
Create entry ; Delete entry (?)
Search for a file
Create/Delete a directory file
List a directory
Rename a file
Link a file to a directory
Traverse a file system
11
Q
What are the dot and dotdot directory entries?
What’s the relation between a process and a path?
A
. - the current directory.
.. - the parent directory
Each process has its own working directory, which is shared by its threads.
12
Q
Using which tool files(inodes) are shared?
A
Links
13
Q
What are soft links and hard links?
A
Soft
- Containing a path name. access is slower
- If source is removed the link becomes broken
Hard
- information about a shared file is duplicated in sharing directories.
- fast, points to file
- Link count must be maintained.
- if source is removed link still accessible
14
Q
What is the difference between shared and exclusive locks?
A
Exclusive - protects updates to file resources and can be owned by only one transaction at a time.
Acquiring an exclusive lock demands waiting if another process is currently holding an exclusive lock or a shared lock of the same resource.
shared - can be owned by several processes at a time.
Acquiring a shared lock demands waiting if another process is currently holding an exclusive lock.
A new request for a shared lock must wait if there’s a request for an exclusive lock on a resource that already has a shared lock.
15
Q
How can i block or unblock a file?
A
Using the command flock(file descriptor, operation)
- operation:
- LOCK_SH - place a shared lock.
- LOCK_EX - place an exclusive lock.
- LOCK_UN - remove an existing lock held by this process.
16
Q
What happens to the lock when a file is closed or a process terminates?
A
File lock is removed.
17
Q
Can we lock parts of the file?
A
Yes.
Any part of the file may be blocked.
18
Q
What are the concerns of the user/kernel regarding files?
A
User:
- File names
- operations allowed
- location(directory structures)
System:
- Storage of files and directories
- Disk space management
- implementation efficiency and reliability.
19
Q
Disk allocation:
Advantages/Disadvantages of:
- Contiguous allocation?
- Linked list of disk blocks
- Linked list using in-memory Files Allocation Table
A
Contiguous:
- Simple; access is fast
- external fragmentation
- How much size to allocate at creation time?
Linked list of disk blocks:
- No fragmentation and easy allocation
- slow random access
- n disk accesses to get to the n’th block
- because the pointer to the next block is held within the block.
- n disk accesses to get to the n’th block
- weird block size
Linked list using in-memory Files Allocation Table
- none of the above disadvantages.
- Uses a table to store the pointers of all blocks in the linked list that represent files - last file block has a special EOF symbol.
- we don’t need allocations to be continguous - segmentation free
- Can access the nth block of some file using the table and not using the table blocks.
- A very large table in memory
20
Q
What is an inode and who holds it?
A
It is a place, on disk, where data about the file is stored.
Each file struct holds inode. (might be two files pointing to the same inode..)
There is a struct called dinode which holds the:
- File Attributes
- Time last accessed
- Time last modified
- Size
- File type, protection bits
- Nlinks
- Number of directory entries pointing to this i-node.
- Address of disk block 0
- Address of disk block 1
- ..
- Address of block of pointer
- Containing additional disk addresses
21
Q
What is the classic Unix disk structure?
A
- Boot Sector
- Super Block
- Number of inodes
- Number of Blocks
- Number of free blocks
- Pointer to free blocks list
- Pointer to free i-node list
- …
- i-nodes
- Data blocks
22
Q
Unix inodes - how bytes counting work?
A
I didn’t get it. You try.
23
Q
Tell me in which block byte 355,000 can be found?
A
- First 10 blocks - 10K:*
- 0 - 10240
- Indircet block - 10K + 256K = 266K =*
- 10240 - 272384
- double indirect block -*
- refers to a list of 256K indirect blocks - each of size 256KB.
- 355,000 = 272384 + X*
- X = 82616. That is, it is placed within the 0th indirect block, in the 80th block. (starting at byte 81920), at byte number 696*
24
Q
What is the file descriptor table for? who holds it?
A
- It is for quick access of a proccess to its files.
- Each process has a file descriptors table
- One entry per each open file.
25
* Where should we keep the file offset?*
* what if two processes want to read from the same file but starting from a different offset?*
* What if i want a couple of threads to write to the same file one after the other?*
* we keep the offset within the file struct .*
* Two files might refer to the same inode but have different offset, thus no problem for them to read different parts of the data.*
* For the mission of writing/reading together **Unix** provided the following solution:*
* Parent's file descriptors-table-entries and child's file descriptors-table-entries point to the **same** file struct, thus holding the same offset and flags.*
* Unrelated processes point to **different** files thus have different offsets.*
26
Can a user access the i-node of a file?
**No.**
Each process has its own *file descriptors table* that points to the entries in the **kernel's** *open files description table.*
*The kernel's open files description table entries point to the i-node of the file.*
27
What does the *open* call do?
It adds an entry to both the *kernel's open file descriptions* and the *process'* file description table.
28
*How is a directory implemented?*
*A simple directory(like in MS-DOS):*
* *fixed size entries*
* *disk addresses & attributes in diretory entry*
*Directory entries simply point to i-nodes.*
29
There are two ways of handling long file names:
* *In-line*
* *In a heap*
what does *in-line* means?
*In-line* means it is included in the file. That is, there is no pointer referring to some outsource which holds the content, it is "physically" here.
30
How **Unix** *directories* are implemented?
Implemented using struct *dirent* which holds :
* *name*
* *inode array*
31
*What are the Pros/Cons with having large blocks?*
*_Pros:_*
* *Typically a multiple number of sectors*
* *Because its size requires division*
* *In sequential access, less blocks to read/write - less seek/search.*
*_Cons:_*
* *High internal fragmentatios*
* *In Random access larger transfer time, larger memory buffers.*
32
What are the Pros/Cons with having small blocks?
*_Pros:_*
* Smaller internal fragmentatios
* Faster random access
*_Cons:_*
* Slower sequential access(*more seeks*)
33
What are the disk access time parameters?
* A - Average seek-time
* Avg time for head to get above a cylinder.
* B - Average time to get to track block.
* C - Rotation time
* Time for disk to complete full rotation
* D - Transfer time
* (block size/track size)\*rotation time
## Footnote
*Average time to access block is given by:*
**A + B + D.**
34
*How can we keep track of free blocks?*
***As a linked list of blocks***
* *Adresses of the first n free blocks are stored at the _super block._*
* *First n-1 blocks are free to be assigned.*
* *Last block contains address of a block which contains n more addresses to free blocks..*
* Addresses of many free blocks are retrieved with one disk access.*
* Unix maintains a single block of free-block addresses in memory.*
* Whenever the last free block is reached, the next block of free-blocks is read and used.*
***As a bit map***
35
Didn't get it.. you try
36
***File system consistency***
* *we count the number of references in free lists*
* *we count the number of references in inodes*
*_what results are troubling?_*
*_How is consistency mainted?_*
* *Both counts 0 - block is missing*
* *Add to free list*
* *More than once in free list*
* *delete all references but one*
* *More than once in files - **Trouble***
* *two or more different files write/read from the same place. which to delete? problem.*
* *In both file and free list*
* *Delete from free list.*
***_Hard links may solve the problem_***
* *count - number of references to each i-node which is found by _descending_ down the file system tree.*
* *Compare count with the link-count field in the i-node struct.*
* *If different, correct link-count field.*
37
What are the hazards if:
* *Link-count \>* actual links number
* *Link-count \<* actual links number
_*Link-count \>* actual links number_
* This file will not be removed.
* Even after removing all actual links - there's a fake one which tell the kernel to avoid removal.
* A waste of memory.
_*Link-count \<* actual links number_
* The file will be removed even though there's a file pointing to it. It is a much greater problem
38
How does Unix enhance performance?
**Using *filename* caching**
39
What is the **Buffer Cache?**
Who maintains that?
It is a pool of internal **data buffers** - the buffer cache.
The *kernel* maintains that.
_Important notes_
* Data *written to disk* is cached, for later use.
* Algorithms instruct the buffer to _delay-write._
* If data is not found in cache, it is read from disk to cache.
* Algoritms instruct the buffer to _pre-cache._
40
* What are the buffer's states?*
* How buffers are implemented?*
*Buffers are categorized into:*
* *busy- currently being used.*
* *Clean - available for use, sync with block content on disk.*
* *Free - empty and haven't been used yet.*
* *Dirty - needs to be moved to write list.*
* Each has a header, that includes the pair .*
* The Buffer Cache is implemented using **LRU list:***
* *Buffers are on a doubly-linked list in LRU order.*
* *Each hash-queue entry points to a linked list of buffers that have the same hash value.*
* *A block may be in only **one** hash-queue.*
* *A **free block** is on the free-list(maintained by the kernel, remember?) _in addition_ to being on a hash-queue.*
* *When looking for a particular block, the hash-queue for it is searched. When in need of a new block , it is removed from the free list.*
41
What happens if the block is:
1. Found in its *hash queue* and is *free*
2. Found in its *hash queue* and is *busy*
3. Not found in the *hash queue* and there are free buffers
4. Not found in the hash queue and in searching the free list for a free buffer, one or more "*delayed-write"* buffers are found.
5. Not found in the *hash queue* and free list is empty.
1. *Found in its hash queue and is free*
* *_buffer is marked busy_*
* *_buffer is removed from free list_*
2. *Found in its has queue and is busy*
* *_process sleeps until buffer is freed, then **recheak**._*
3. *Not found in the hash queue and there are free buffers*
* *_A free buffer is allocated from the pool_*
4. *Not found in the hash queue and in searching the free list for a free buffer, one or more "delayed-write" buffers are found.*
* *_write delayed-write buffer(s) to disk_*
* *_move them to the most recently used side of the list and find a free buffer_*
5. *Not found in the hash queue and free list is empty.*
* *_Block requesting process, when scheduled, go through hash-queue again._*
42
Process B looks up for block **b** to find that its buffer is locked.
He waits until the buffer of block B is available. When is returned, it needs to search again in the hash table for block *b.*
*_Why?_*
Because while waiting some other process **C** might have gotten the buffer, might have loaded it with another block **c.**
43
Why not pure LRU?
* Some blocks should be written as quickly as possible.
* Insert critical blocks at the head of the queue ,to be replaced soon and written to disk.
* Have a system daemon that calls *sync* every 30 seconds, to help in updating blocks.
* Some blocks are likely to be used again. (partly filled blocks being written)
* Insert to the end to stay longer in the cache.
44
What is special about the ***NTFS***?
***NTFS:***
* A journaling file system - a file system that keeps track of changed not yet commited to the file system's main part by *_recording_* the intentions of such changes.
* Each file is represented by a record in a special file called the *master file table(MFT)*
45
*MFT - Master File Table*
* *what does it hold?*
* *does its records are of the same size for process?*
* *Can it save file's data?*
* *Is there any limit of file size?*
* *Where does the MFT address lie?*
MFT - *Master File Table*
* Each file/directory has one or more **1K records** in the *MFT.*
* A record contains file attributes and a list of block numbers.
* Larger files need more then one MFT record for the list of blocks - records are **extended** by pointing to other records.
* Disk blocks are described by sequence of records, each of which is a series of *runs.*
* *A run* is a continguous sequence of blocks and represented by a pair: (offset, length)
* Case the file is very small - data can be kept *directly.*
* The boot sector contains the MFT address
* NTFS tries to allocate blocks continguosuly
* That is why *run is* a contiguous sequence of blocks.
* No upper bound on file size
46
Is it possible that a short file uses more MFT records than a longer file?
**Yes.**
**Assume files A,B where B is a longer.**
A - most of its blocks are not sequential.
* A few *runs* structs.
* Might have a few MFT records.
B - all of his blocks are sequential.
* Only one *run* struct is needed.
* Only one MFT record is needed.
47
*How is **NTFS**(new technology file system) implemented with directories?*
*With small directories:*
* *Standard info*
* *A directory entry contains the MFT index for the file, length of the file's name, file's name, and various fields and flags.*
*With large directories:*
* *Organized as B+ trees*
* *Supports transparent file compression*
* *Compresses in groups of 16 blocks.*
* *Can select to compress whole volume, specific directories or files.*
48
Is compression good or bad for performance?
Disk throughput - increased/decreased?
* CPU works harder/lighter?
* As function of compression unit size - RAM access slowed down or increased?
**Disk throughput is increased**
* *because size is smaller.*
**CPU works harder**
* Because it needs to compress/decompress
**Access to RAM slowed down**
49
*_DFS - Distributed File Systems_*
* What is it?
* what is *mount?*
* remote file access can be "*stateful"* or *"stateless".* explain.
* What is it used for?
*_DFS - Distributed File Systems_*
* A collection of interconnected machines that **do not share memory or a clock.**
* We can use *file naming scheme.* (**{hostname:path}**)
* This method is not **transparent**.
* we want the kernel to hide details
* This can be solved using _mount_
* __*mounting* is a process by which the OS makes files and directories on a storage device available for users to access via the computer's file system.
* A server exports part of its file system
* We don't want clients to be exposed to the server's data..
* *stateless protocol -* server remembers nothing about previous requests from a client
* *stateful protocol -* more efficient(information kept in server's kernel)
* _can be different machines and different OSs_
* _Any machine can be both client and server_
* _Clients access directories by *mounting*_
* *_remote mounts are invisible_*
* *_guess it means A request B that request C.._*
* *_Servers specify their exported directories upon boot, listed in /etc/exports_*
* *_File sharing: accessing a file in a directory mounted by other (sharing) clients._*
50
NFS(v3) protocols
* How mounting requests work?
* What *file handles* contain?
* What is the file operations protocol?
NFS(v3) protocols
* *send*: {host-name;target-directory *path}, get*: *file handle*
* *File handle contains:*
* *File system type*
* *Disk ID*
* *i-node number*
* *Protection information*
* *File operations - directory and file access:*
* *lookup* provides a *file handle*
* *reads* and *writes* have all the needed information by using file handles - offsets are absolute.
* No *open* or *close* calls.
* Server **does not** keep open files tables.
* Files cannot be locked
51
*Remote Procedure Calls (RPC)*
* How does it activate code on a remote machine?
* using send/receive protocol
* Proccess on one machine calls a procedure on another machine. What must be maintained?
* - synchronous(!) - *blocking* send
* Problems?
* What is the general scheme?
*Remote Procedure Calls (RPC)*
* How does it activate code on a remote machine?
* using send/receive protocol
* Proccess on one machine calls a procedure on another machine. What must be maintained?
* synchronous(!) - *blocking* send
* Problems?
* Different address spaces
* Parameters and results have to be passed - by they refer to addresses the client/server don't have
* Machines can crash...
* general scheme
* A client *stub* and a server *stub*
* A stub is a piece of code that converts parameters passed between *client* and *sever*
* Server *stub* blocked on recieve.
* Client *stub* blocks for returning message
52
What is ***NFS*?**
*NFS* is a distributed file system protocol allowing a user on a client computer to access files over a _computer network_ much like local storage is accessed.
53
How ***NFS*** is implemented?
***_NFS implemenation:_***
* Client & Server have:
* Virtual File System layer.
* Keeps *v-nodes* for *open-files.*
* *v-node* = {*i-node* | *r-node*}
* *NFS* module
* Used for remote requests
* Client creates *r-node* in its internal tables and stores the *file handle.*
54
_What are the steps in NFS mounting?_
_What are the steps in opening a remote file?_
*_NFS mounting_*
* mount program invoked manually by admin
* input: {remote target | local dir pathname}
* remote target = {remote server name | dir path name}
* *_local dir is where the file will stay in the **client's** VFS_*
* server sends *file handle* representing directory
* kernel construct a *v-node* for remote directory
* Client(NFS) creates r-node for remote directory
_Opening a remote file_
* Client kernel parses a path name to reach desired directory
* Find a pointer to *r-node* in *v-node*
* Asks *NFS* client to open the file
* *NFS* clients look at the last name of the path name and gets back a *file-handle.*
* *NFS* client creates an *r-node* entry for the open file, stores the *file-handle* in it and the VFS creates a *v-node*, pointing to the *r-node*
* Calling process is given a file descriptor in return, pointing to the *v-node.*
* Any subsequent calls this file descriptor will be traced by the VFS to the *r-node* and the suitable *read/write* operations will be performed.
55
_What is an obsolete value?_
_What is read-ahead?_
_obsolete value:_
*A value which is not in use anymore.*
_read-ahead_
*Reading data in advance so that it is immedietly available when requested.*
56
*_NFS: what are the performance issues?_*
*_NFS: performance issues_*
* Data sent in 8KB chunks
* not necesarilly efficient
* *Read-ahead* might not be coherent.
* Client caching is important for efficiecy
* If the policy is **not** *write-through NFS* exposed to problems with coherency and data loss
* Cached block discarded
* Data block after 3 seconds
* Directory block after 30 seconds
* Every 30 seconds all dirty cached blocks are written.
* Check with the server whenever a cached file is opened
* if not in used discard from cache.
Operating Systems
flashcards
Decks in class (12)
# Cards
