From 1ddfbbb194e3aa668b33469eb547132a7a7f940a Mon Sep 17 00:00:00 2001 From: Frans Kaashoek Date: Mon, 29 Aug 2011 17:18:40 -0400 Subject: [PATCH] Revert "Introduce and use sleeplocks instead of BUSY flags" 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 3a5fa7ed9020eaf8ab843a16d26db7393b2ec072. Conflicts: defs.h file.c file.h --- bio.c | 45 ++++++++++++++++++++++----------------------- buf.h | 7 +++---- defs.h | 5 ----- file.c | 2 +- file.h | 7 +++---- fs.c | 32 ++++++++++++++++---------------- ide.c | 2 +- log.c | 14 +++++++++----- pipe.c | 2 +- spinlock.c | 42 ++++-------------------------------------- spinlock.h | 7 +------ sysfile.c | 1 - 12 files changed, 61 insertions(+), 105 deletions(-) diff --git a/bio.c b/bio.c index f11328d..6a3968b 100644 --- a/bio.c +++ b/bio.c @@ -13,7 +13,9 @@ // * Only one process at a time can use a buffer, // so do not keep them longer than necessary. // -// The implementation uses two state flags internally: +// The implementation uses three state flags internally: +// * B_BUSY: the block has been returned from bread +// and has not been passed back to brelse. // * B_VALID: the buffer data has been initialized // with the associated disk block contents. // * B_DIRTY: the buffer data has been modified @@ -49,8 +51,6 @@ binit(void) b->next = bcache.head.next; b->prev = &bcache.head; b->dev = -1; - initlock(&b->lock, "buf"); - initsleeplock(&b->sleeplock); bcache.head.next->prev = b; bcache.head.next = b; } @@ -58,43 +58,42 @@ binit(void) // Look through buffer cache for sector on device dev. // If not found, allocate fresh block. -// In either case, return sleep-locked buffer. +// In either case, return locked buffer. static struct buf* bget(uint dev, uint sector) { struct buf *b; acquire(&bcache.lock); + + loop: // Try for cached block. for(b = bcache.head.next; b != &bcache.head; b = b->next){ - acquire(&b->lock); if(b->dev == dev && b->sector == sector){ - release(&bcache.lock); - acquire_sleeplock(&b->sleeplock, &b->lock); - release(&b->lock); - return b; + if(!(b->flags & B_BUSY)){ + b->flags |= B_BUSY; + release(&bcache.lock); + return b; + } + sleep(b, &bcache.lock); + goto loop; } - release(&b->lock); } // Allocate fresh block. for(b = bcache.head.prev; b != &bcache.head; b = b->prev){ - acquire(&b->lock); - if (!acquired_sleeplock(&b->sleeplock)) { - release(&bcache.lock); + if((b->flags & B_BUSY) == 0){ b->dev = dev; b->sector = sector; - b->flags = 0; - acquire_sleeplock(&b->sleeplock, &b->lock); - release(&b->lock); + b->flags = B_BUSY; + release(&bcache.lock); return b; } - release(&b->lock); } panic("bget: no buffers"); } -// Return a locked buf with the contents of the indicated disk sector. +// Return a B_BUSY buf with the contents of the indicated disk sector. struct buf* bread(uint dev, uint sector) { @@ -110,7 +109,7 @@ bread(uint dev, uint sector) void bwrite(struct buf *b) { - if(!acquired_sleeplock(&b->sleeplock)) + if((b->flags & B_BUSY) == 0) panic("bwrite"); b->flags |= B_DIRTY; iderw(b); @@ -120,11 +119,11 @@ bwrite(struct buf *b) void brelse(struct buf *b) { - if(!acquired_sleeplock(&b->sleeplock)) + if((b->flags & B_BUSY) == 0) panic("brelse"); acquire(&bcache.lock); - acquire(&b->lock); + b->next->prev = b->prev; b->prev->next = b->next; b->next = bcache.head.next; @@ -132,8 +131,8 @@ brelse(struct buf *b) bcache.head.next->prev = b; bcache.head.next = b; - release_sleeplock(&b->sleeplock); - release(&b->lock); + b->flags &= ~B_BUSY; + wakeup(b); release(&bcache.lock); } diff --git a/buf.h b/buf.h index cfeb8c6..9c586f2 100644 --- a/buf.h +++ b/buf.h @@ -2,13 +2,12 @@ struct buf { int flags; uint dev; uint sector; - struct spinlock lock; - struct sleeplock sleeplock; struct buf *prev; // LRU cache list struct buf *next; struct buf *qnext; // disk queue uchar data[512]; }; -#define B_VALID 0x1 // buffer has been read from disk -#define B_DIRTY 0x2 // buffer needs to be written to disk +#define B_BUSY 0x1 // buffer is locked by some process +#define B_VALID 0x2 // buffer has been read from disk +#define B_DIRTY 0x4 // buffer needs to be written to disk diff --git a/defs.h b/defs.h index 0a55759..19d559b 100644 --- a/defs.h +++ b/defs.h @@ -5,7 +5,6 @@ struct inode; struct pipe; struct proc; struct spinlock; -struct sleeplock; struct stat; struct superblock; @@ -130,10 +129,6 @@ void initlock(struct spinlock*, char*); void release(struct spinlock*); void pushcli(void); void popcli(void); -void initsleeplock(struct sleeplock*); -void acquire_sleeplock(struct sleeplock*, struct spinlock*); -void release_sleeplock(struct sleeplock*); -int acquired_sleeplock(struct sleeplock*); // string.c int memcmp(const void*, const void*, uint); diff --git a/file.c b/file.c index df90e07..f28365e 100644 --- a/file.c +++ b/file.c @@ -2,8 +2,8 @@ #include "defs.h" #include "param.h" #include "fs.h" -#include "spinlock.h" #include "file.h" +#include "spinlock.h" struct devsw devsw[NDEV]; struct { diff --git a/file.h b/file.h index 2b4a2cb..d123696 100644 --- a/file.h +++ b/file.h @@ -15,9 +15,7 @@ struct inode { uint dev; // Device number uint inum; // Inode number int ref; // Reference count - int flags; // I_VALID - struct spinlock lock; - struct sleeplock sleeplock; + int flags; // I_BUSY, I_VALID short type; // copy of disk inode short major; @@ -27,7 +25,8 @@ struct inode { uint addrs[NDIRECT+1]; }; -#define I_VALID 0x1 +#define I_BUSY 0x1 +#define I_VALID 0x2 // device implementations diff --git a/fs.c b/fs.c index 6bfe963..b755c78 100644 --- a/fs.c +++ b/fs.c @@ -113,8 +113,11 @@ bfree(int dev, uint b) // It is an error to use an inode without holding a reference to it. // // Processes are only allowed to read and write inode -// metadata and contents when holding the inode's sleeplock. -// Callers are responsible for locking +// metadata and contents when holding the inode's lock, +// represented by the I_BUSY flag in the in-memory copy. +// Because inode locks are held during disk accesses, +// they are implemented using a flag rather than with +// spin locks. Callers are responsible for locking // inodes before passing them to routines in this file; leaving // this responsibility with the caller makes it possible for them // to create arbitrarily-sized atomic operations. @@ -213,7 +216,6 @@ iget(uint dev, uint inum) ip->inum = inum; ip->ref = 1; ip->flags = 0; - initsleeplock(&ip->sleeplock); release(&icache.lock); return ip; @@ -230,7 +232,7 @@ idup(struct inode *ip) return ip; } -// Acquire the sleeplock for a given inode. +// Lock the given inode. void ilock(struct inode *ip) { @@ -241,7 +243,9 @@ ilock(struct inode *ip) panic("ilock"); acquire(&icache.lock); - acquire_sleeplock(&ip->sleeplock, &icache.lock); + while(ip->flags & I_BUSY) + sleep(ip, &icache.lock); + ip->flags |= I_BUSY; release(&icache.lock); if(!(ip->flags & I_VALID)){ @@ -264,11 +268,12 @@ ilock(struct inode *ip) void iunlock(struct inode *ip) { - if(ip == 0 || !acquired_sleeplock(&ip->sleeplock) || ip->ref < 1) + if(ip == 0 || !(ip->flags & I_BUSY) || ip->ref < 1) panic("iunlock"); acquire(&icache.lock); - release_sleeplock(&ip->sleeplock); + ip->flags &= ~I_BUSY; + wakeup(ip); release(&icache.lock); } @@ -279,15 +284,14 @@ iput(struct inode *ip) acquire(&icache.lock); if(ip->ref == 1 && (ip->flags & I_VALID) && ip->nlink == 0){ // inode is no longer used: truncate and free inode. - if(acquired_sleeplock(&ip->sleeplock)) + if(ip->flags & I_BUSY) panic("iput busy"); - acquire_sleeplock(&ip->sleeplock, &icache.lock); + ip->flags |= I_BUSY; release(&icache.lock); itrunc(ip); ip->type = 0; iupdate(ip); acquire(&icache.lock); - release_sleeplock(&ip->sleeplock); ip->flags = 0; wakeup(ip); } @@ -429,14 +433,10 @@ writei(struct inode *ip, char *src, uint off, uint n) return devsw[ip->major].write(ip, src, n); } - if(off > ip->size || off + n < off) { - panic("writei1"); + if(off > ip->size || off + n < off) return -1; - } - if(off + n > MAXFILE*BSIZE) { - panic("writei2"); + if(off + n > MAXFILE*BSIZE) return -1; - } for(tot=0; totdev, bmap(ip, off/BSIZE)); diff --git a/ide.c b/ide.c index e41437e..4165831 100644 --- a/ide.c +++ b/ide.c @@ -127,7 +127,7 @@ iderw(struct buf *b) { struct buf **pp; - if(!acquired_sleeplock(&b->sleeplock)) + if(!(b->flags & B_BUSY)) panic("iderw: buf not busy"); if((b->flags & (B_VALID|B_DIRTY)) == B_VALID) panic("iderw: nothing to do"); diff --git a/log.c b/log.c index d2dcba7..6b5c329 100644 --- a/log.c +++ b/log.c @@ -43,9 +43,9 @@ struct logheader { struct { struct spinlock lock; - struct sleeplock sleeplock; int start; int size; + int intrans; int dev; struct logheader lh; } log; @@ -60,7 +60,6 @@ initlog(void) struct superblock sb; initlock(&log.lock, "log"); - initsleeplock(&log.sleeplock); readsb(ROOTDEV, &sb); log.start = sb.size - sb.nlog; log.size = sb.nlog; @@ -134,7 +133,10 @@ void begin_trans(void) { acquire(&log.lock); - acquire_sleeplock(&log.sleeplock, &log.lock); + while (log.intrans) { + sleep(&log, &log.lock); + } + log.intrans = 1; release(&log.lock); } @@ -147,8 +149,10 @@ commit_trans(void) log.lh.n = 0; write_head(); // Reclaim log } + acquire(&log.lock); - release_sleeplock(&log.sleeplock); + log.intrans = 0; + wakeup(&log); release(&log.lock); } @@ -167,7 +171,7 @@ log_write(struct buf *b) if (log.lh.n >= LOGSIZE || log.lh.n >= log.size - 1) panic("too big a transaction"); - if (!acquired_sleeplock(&log.sleeplock)) + if (!log.intrans) panic("write outside of trans"); // cprintf("log_write: %d %d\n", b->sector, log.lh.n); diff --git a/pipe.c b/pipe.c index bbac4dc..f76ed5c 100644 --- a/pipe.c +++ b/pipe.c @@ -4,8 +4,8 @@ #include "mmu.h" #include "proc.h" #include "fs.h" -#include "spinlock.h" #include "file.h" +#include "spinlock.h" #define PIPESIZE 512 diff --git a/spinlock.c b/spinlock.c index c1f6a96..a16621c 100644 --- a/spinlock.c +++ b/spinlock.c @@ -17,9 +17,10 @@ initlock(struct spinlock *lk, char *name) lk->cpu = 0; } -// Acquire a spin lock. Loops (spins) until the lock is acquired. -// Holding a lock for a long time may cause other CPUs to waste time spinning to acquire it. -// Spinlocks shouldn't be held across sleep(); for those cases, use sleeplocks. +// Acquire the lock. +// Loops (spins) until the lock is acquired. +// Holding a lock for a long time may cause +// other CPUs to waste time spinning to acquire it. void acquire(struct spinlock *lk) { @@ -114,38 +115,3 @@ popcli(void) sti(); } -void -initsleeplock(struct sleeplock *l) -{ - l->locked = 0; -} - -// Grab the sleeplock that is protected by spinl. Sleeplocks allow a process to lock -// a data structure for long times, including across sleeps. Other processes that try -// to acquire a sleeplock will be put to sleep when another process hold the sleeplock. -// To update status of the sleeplock atomically, the caller must hold spinl -void -acquire_sleeplock(struct sleeplock *sleepl, struct spinlock *spinl) -{ - while (sleepl->locked) { - sleep(sleepl, spinl); - } - sleepl->locked = 1; -} - -// Release the sleeplock that is protected by a spin lock -// Caller must hold the spinlock that protects the sleeplock -void -release_sleeplock(struct sleeplock *sleepl) -{ - sleepl->locked = 0; - wakeup(sleepl); -} - -// Is the sleeplock acquired? -// Caller must hold the spinlock that protects the sleeplock -int -acquired_sleeplock(struct sleeplock *sleepl) -{ - return sleepl->locked; -} diff --git a/spinlock.h b/spinlock.h index f6a69b7..fdda016 100644 --- a/spinlock.h +++ b/spinlock.h @@ -1,4 +1,4 @@ -// Mutual exclusion lock for short code fragments +// Mutual exclusion lock. struct spinlock { uint locked; // Is the lock held? @@ -9,8 +9,3 @@ struct spinlock { // that locked the lock. }; -// Lock that maybe held across sleeps -struct sleeplock { - uint locked; // Is the lock held? -}; - diff --git a/sysfile.c b/sysfile.c index 11895df..c9d3594 100644 --- a/sysfile.c +++ b/sysfile.c @@ -5,7 +5,6 @@ #include "mmu.h" #include "proc.h" #include "fs.h" -#include "spinlock.h" #include "file.h" #include "fcntl.h"