CS 161: Operating Systems

Tuesday/Thursday 1:00-2:30

Pierce 301

Web Work Answers

Below are our solutions to the web work, along with explanations as to why we think they are correct. Note that some of the questions are ambiguous and can be interpreted several ways, some of which yield other, equally correct answers. If you're unsure of your understanding of the material and want to know whether your answer was also correct, or if you feel like our solution is incorrect, feel free to reach out to the course staff. However, the web work is graded only on completion, not on exactly matching the solutions below, so there's no need to argue your case if you're confident you're also right.

4/10

1. Which of the following pieces of existing code do you expect you will need to modify in Assignment 4?

kern/vm/addrespace.c: We certainly hope not
kern/fs/sfs/sfs_dir.c: If you want your directory operations to be recoverable, this is undoubtedly yes.
kern/fs/sfs/sfs_jphys.c: In theory you shouldn't need to modify this; if you find yourself wanting to, talk with us -- it suggests that either you're doing something wrong or we didn't think of some feature jphys should support.
kern/vfs/vnode.c: Ideally you should not need to touch this, since the vnode operations are not specific to a given file system and you are adding journaling specifically to SFS.
kern/vfs/buf.c: This probably will need changes to enforce WAL.
kern/syscall/more_syscalls.c: Yes, you need to interface this to your file table from assignment 2.
userland/sbin/mksfs/mksfs.c: If you change the ondisk format for SFS, then you'll need to change this, but you may not actually need to do that.

2. Which of the following is not necessary to do to use an SFS file system?

disk161 create
/sbin/mksfs
dumpsfs (This is a handy debugging tool, but its use is purely optional.)
mount

3. Prior to adding journaling, SFS maintains consistency via synchronous ordered writes.

False! It doesn't maintain consistency at all.

4. If you unmount an SFS volume and then run sfsck, you should not get any errors.

True.

5. 5. Which of the following seem like record types you might need to support in this assignment?

Change bit 57 in block 1234 from 0 to 1: This seems like a nice idempotent way to describe a bitmap change.
Create a file named foo in directory D: This feels like a high level logical log record, so, "No!"
Change the fields in inode 23 from to an initialized state: This too seems like a nice idempotent way to initialize an inode
Here is an old copy of block B, here is a new copy of block B: Nah -- this feels like a low level physical log record.
Allocate a block: This is too vague
Here is a block of user data that needs to be written to block 9753: You don't have to log user data, so probably not.
Remove the directory entry containing name="dog" inode=825 from slot 3 of this directory.: Ah yes, this too seems like a nice idempotent way of removing a directory entry.

6. How many different record types can you have in an OS/161 journal?

128

7. What do we call the unique identifier we use to refer to a log record?

Log Sequence Number, LSN

8. Which of the following are tools or programs that can be used to test your file system?

sfsck, The DOom Counter, frack, poisondisk are all good test features; mksfs simply makes a file system, something you'll have to do, but isn't really a test feature. And crashme isn't a program we give you.

9. New records in jphys appear at the HEAD or the TAIL?

The HEAD

10. Is it possible to use the physical journal provided in OS/161 to implement multithreaded recovery?

That would be, "No!"

11. Which of the following accurately completes the following statement: A single journal transaction contains ...

Multiple log records from a single high-level file system operations

12. Assuming UNDO/REDO logging, Which of the following log records describe idempotent operations?

Set the first unset bit in the freemap to 1: Not quite -- you don't which bit is the first unset bit and the freemap might be in a different state during recovery than it was when you wrote this log record.
Clear bit 97 in block B: Not quite -- what was the value of bit 97 before this operation?
Create a directory entry in (formerly empty) slot 5 of directory D containing "mynewfile.txt" and inode 123. This seems pretty complete, so yes!
Mark block 8765 newly allocated in the block bitmap. This too seems nicely specified
Remove the directory entry at slot 4 in directory D. That's a little risky; what if there is a new entry in slot 4?