L11: Naming

- Plan
  Last xv6 lecture
  Quiz next week
  Wrap up bcache+icache
  Names 
  Concurrency
  
- Layered picture

UNIX FS API
------
hierachical names
-----
directories
------
inodes
------
blocks

- Explain directories

each process has a current working directory (p->cwd)

a file can have multiple names

directory representation (14 chars, followed by inumber)
  dirloopup
  why iget doesn't take lock?
  what is the invariant for the inode returned for iget?
  what is the implication for ilock?

- Finish homework L10

echo > a
11 ialloc (create 5220, ialloc 4202, bread 4002, bget 3966, iderw 3854)
11 iupdate OMIT (create 5227, iupdate 4227)
29 writei OMIT ("/a" dirent; create 4237, dirlink 4652, writei 4552)

Question: why are the writes in this order?

Question: what happens if concurrent create?

echo > a
(nothing)

Question: how to allocate a block?

echo x > a
28 balloc (writei 4552, bmap 4410, balloc 4104)
574 writei ("x")
11 iupdate (length, writei 4578)
574 writei ("\n")
11 iupdate (length)

Question: what if concurrent app writes to same file?
 (in prev example)
 filewrite 4922, ilock 4302

Question: when can inode and blocks be freed?

rm a
29 writei OMIT ("/a" dirent; sys_unlink 5185)
11 iupdate (link count; sys_unlink 5194)
574 bzero (sys_unlink 5195, iunlockput 4374, iput 4352, itrunc 4361, bfree 4130, bzero 4089)
28 bfree (4145)
11 iupdate (length; itrunc 4480)
11 iupdate (type=free; iput 4363)

Question: why are the writes in this order?

Question: what happens when concurrent unlink?

mkdir d
11 ialloc (d's inode)
11 iupdate (create)
2 iupdate (/'s link count)
28 balloc (d's content, for . and ..)
574 writei (.)
11 iupdate (writei)
574 writei (..)
11 iupdate (writei)
29 writei (/ dirent)

rm d
29 writei (/ dirent)
2 iupdate (/ link count)
11 iupdate (d's link count now 0)
574 bzero (d's dirent block)
28 bfree (d's dirent block)
11 iupdate (length; itrunc)
11 iupdate (type=free; iput)

Question: why must directories be empty for before they are allowed to
be removed?

- Names

pathnames

namei is tricky
  we want concurrency (potentially many disk reads!)
  multiple objects involved (e.g., parent directory and file)
  concurrent changes to name space

- Hard versus symbolic links

symbolic links
  what changes necessary to xv6
  cycles?

why have both hard and symbolic links?

- UNIX API:
  - shared offset between parent and child (file table)
  - TOCTOU problems