My changes have a race with re-used bufs and the code doesn't seem to get shorter
Keep the changes that fixed ip->off race
This reverts commit 3a5fa7ed90.
Conflicts:
defs.h
file.c
file.h
Remove I_BUSY, B_BUSY, and intrans defs and usages
One spinlock per buf to avoid ugly loop in bget
fix race in filewrite (don't update f->off after releasing lock)
Suppose an inode has been used and freed.
It is left marked I_VALID (the bug).
Now ialloc comes along and reuses the
inode. It writes the new inode type to disk
and returns iget(dev, inum) to get the
cache entry. Iget sees that the inode is valid
and doesn't bother refreshing from disk.
Now when the caller iupdates, it will write
out a zero type and the file or directory has
disappeared.
to take inode* instead of minor number.
Unlock console inode during console_read
and console_write. Otherwise background
processes cannot write to console while the
shell is reading it waiting for input.
Various changes made while offline.
+ bwrite sector argument is redundant; use b->sector.
+ reformatting of files for nicer PDF page breaks
+ distinguish between locked, unlocked inodes in type signatures
+ change FD_FILE to FD_INODE
+ move userinit (nee proc0init) to proc.c
+ move ROOTDEV to param.h
+ always parenthesize sizeof argument
- Got rid of dummy proc[0]. Now proc[0] is init.
- Added initcode.S to exec /init, so that /init is
just a regular binary.
- Moved exec out of sysfile to exec.c
- Moved code dealing with fs guts (like struct inode)
from sysfile.c to fs.c. Code dealing with system call
arguments stays in sysfile.c
- Refactored directory routines in fs.c; should be simpler.
- Changed iget to return *unlocked* inode structure.
This solves the lookup-then-use race in namei
without introducing deadlocks.
It also enabled getting rid of the dummy proc[0].
Robert Morris
Frans Kaashoek
Frans Kaashoek
Nickolai Zeldovich
Austin Clements
Russ Cox
rsc
rtm
kolya