read-write locks for buffer cache

948078f5 · Nickolai Zeldovich · 9f969fb4 · 948078f5 · 948078f5 · 948078f5
--- a/bio.c
+++ b/bio.c
@@ -54,7 +54,7 @@ evict(uint key, void *bp)
 {
  struct buf *b = bp;
  acquire(&b->lock);
-  if ((b->flags & (B_BUSY | B_VALID)) == 0)
+  if ((b->flags & (B_BUSYR | B_BUSYW | B_VALID)) == 0)
    return b;
  release(&b->lock);
  return 0;
@@ -65,7 +65,7 @@ evict_valid(uint key, void *bp)
 {
  struct buf *b = bp;
  acquire(&b->lock);
-  if ((b->flags & B_BUSY) == 0)
+  if ((b->flags & (B_BUSYR | B_BUSYW)) == 0)
    return b;
  release(&b->lock);
  return 0;
@@ -75,7 +75,7 @@ evict_valid(uint key, void *bp)
 // If not found, allocate fresh block.
 // In either case, return locked buffer.
 static struct buf*
-bget(uint dev, uint sector)
+bget(uint dev, uint sector, int writer)
 {
  struct buf *b;
@@ -88,8 +88,9 @@ bget(uint dev, uint sector)
    acquire(&b->lock);
    if (b->dev != dev)
      panic("dev mismatch");
-    if (!(b->flags & B_BUSY)) {
+    if (!(b->flags & (B_BUSYW | (writer ? B_BUSYR : 0)))) {
-      b->flags |= B_BUSY;
+      b->flags |= B_BUSYR | (writer ? B_BUSYW : 0);
+      __sync_fetch_and_add(&b->readbusy, 1);
      release(&b->lock);
      rcu_end_read();
      return b;
@@ -108,7 +109,8 @@ bget(uint dev, uint sector)
    victim = ns_enumerate(bufns, evict_valid);
  if (victim == 0)
    panic("bget all busy");
-  victim->flags |= B_BUSY;
+  victim->flags |= B_BUSYR | B_BUSYW;
+  __sync_fetch_and_add(&victim->readbusy, 1);
  ns_remove(bufns, victim->sector, victim);
  release(&victim->lock);
  rcu_delayed(victim, kmfree);
@@ -116,7 +118,8 @@ bget(uint dev, uint sector)
  b = kmalloc(sizeof(*b));
  b->dev = dev;
  b->sector = sector;
-  b->flags = B_BUSY;
+  b->flags = B_BUSYR | B_BUSYW;
+  b->readbusy = 1;
  initlock(&b->lock, "bcache-lock");
  initcondvar(&b->cv, "bcache-cv");
  if (ns_insert(bufns, b->sector, b) < 0) {
@@ -128,11 +131,11 @@ bget(uint dev, uint sector)
 // Return a B_BUSY buf with the contents of the indicated disk sector.
 struct buf*
-bread(uint dev, uint sector)
+bread(uint dev, uint sector, int writer)
 {
  struct buf *b;
-  b = bget(dev, sector);
+  b = bget(dev, sector, writer);
  if(!(b->flags & B_VALID))
    iderw(b);
  return b;
@@ -142,7 +145,7 @@ bread(uint dev, uint sector)
 void
 bwrite(struct buf *b)
 {
-  if((b->flags & B_BUSY) == 0)
+  if((b->flags & B_BUSYW) == 0)
    panic("bwrite");
  b->flags |= B_DIRTY;
  if (writeback)
@@ -154,9 +157,10 @@ void
 brelse(struct buf *b)
 {
  acquire(&b->lock);
-  if((b->flags & B_BUSY) == 0)
+  if((b->flags & (B_BUSYR | B_BUSYW)) == 0)
    panic("brelse");
-  b->flags &= ~B_BUSY;
+  int lastreader = __sync_sub_and_fetch(&b->readbusy, 1);
+  b->flags &= ~(B_BUSYW | ((lastreader==0) ? B_BUSYR : 0));
  release(&b->lock);
  cv_wakeup(&b->cv);
 }

--- a/buf.h
+++ b/buf.h
@@ -2,6 +2,7 @@ struct buf {
  int flags;
  uint dev;
  uint sector;
+  int readbusy;
  struct buf *prev; // LRU cache list
  struct buf *next;
  struct buf *qnext; // disk queue
@@ -9,7 +10,8 @@ struct buf {
  struct spinlock lock;
  uchar data[512];
 };
-#define B_BUSY  0x1  // buffer is locked by some process
+#define B_BUSYR 0x1  // buffer is locked by some process
-#define B_VALID 0x2  // buffer has been read from disk
+#define B_BUSYW 0x2  // buffer is locked by some process
-#define B_DIRTY 0x4  // buffer needs to be written to disk
+#define B_VALID 0x4  // buffer has been read from disk
+#define B_DIRTY 0x8  // buffer needs to be written to disk
--- a/defs.h
+++ b/defs.h
@@ -14,7 +14,7 @@ struct ns;
 // bio.c
 void            binit(void);
-struct buf*     bread(uint, uint);
+struct buf*     bread(uint, uint, int writer);
 void            brelse(struct buf*);
 void            bwrite(struct buf*);

--- a/fs.c
+++ b/fs.c
@@ -32,7 +32,7 @@ readsb(int dev, struct superblock *sb)
 {
  struct buf *bp;
-  bp = bread(dev, 1);
+  bp = bread(dev, 1, 0);
  memmove(sb, bp->data, sizeof(*sb));
  brelse(bp);
 }
@@ -43,7 +43,7 @@ bzero(int dev, int bno)
 {
  struct buf *bp;
-  bp = bread(dev, bno);
+  bp = bread(dev, bno, 1);
  memset(bp->data, 0, BSIZE);
  bwrite(bp);
  brelse(bp);
@@ -62,7 +62,7 @@ balloc(uint dev)
  bp = 0;
  readsb(dev, &sb);
  for(b = 0; b < sb.size; b += BPB){
-    bp = bread(dev, BBLOCK(b, sb.ninodes));
+    bp = bread(dev, BBLOCK(b, sb.ninodes), 1);
    for(bi = 0; bi < BPB; bi++){
      m = 1 << (bi % 8);
      if((bp->data[bi/8] & m) == 0){  // Is block free?
@@ -88,7 +88,7 @@ bfree(int dev, uint b)
  bzero(dev, b);
  readsb(dev, &sb);
-  bp = bread(dev, BBLOCK(b, sb.ninodes));
+  bp = bread(dev, BBLOCK(b, sb.ninodes), 1);
  bi = b % BPB;
  m = 1 << (bi % 8);
  if((bp->data[bi/8] & m) == 0)
@@ -160,7 +160,7 @@ ialloc(uint dev, short type)
  readsb(dev, &sb);
  for(inum = 1; inum < sb.ninodes; inum++){  // loop over inode blocks
-    bp = bread(dev, IBLOCK(inum));
+    bp = bread(dev, IBLOCK(inum), 1);
    dip = (struct dinode*)bp->data + inum%IPB;
    int seemsfree = (dip->type == 0);
    brelse(bp);
@@ -191,7 +191,7 @@ iupdate(struct inode *ip)
  struct buf *bp;
  struct dinode *dip;
-  bp = bread(ip->dev, IBLOCK(ip->inum));
+  bp = bread(ip->dev, IBLOCK(ip->inum), 1);
  dip = (struct dinode*)bp->data + ip->inum%IPB;
  dip->type = ip->type;
  dip->major = ip->major;
@@ -268,7 +268,7 @@ iget(uint dev, uint inum)
    goto retry;
  }
-  struct buf *bp = bread(ip->dev, IBLOCK(ip->inum));
+  struct buf *bp = bread(ip->dev, IBLOCK(ip->inum), 0);
  struct dinode *dip = (struct dinode*)bp->data + ip->inum%IPB;
  ip->type = dip->type;
  ip->major = dip->major;
@@ -394,7 +394,7 @@ bmap(struct inode *ip, uint bn)
    // Load indirect block, allocating if necessary.
    if((addr = ip->addrs[NDIRECT]) == 0)
      ip->addrs[NDIRECT] = addr = balloc(ip->dev);
-    bp = bread(ip->dev, addr);
+    bp = bread(ip->dev, addr, 1);
    a = (uint*)bp->data;
    if((addr = a[bn]) == 0){
      a[bn] = addr = balloc(ip->dev);
@@ -425,7 +425,7 @@ itrunc(struct inode *ip)
  }
  if(ip->addrs[NDIRECT]){
-    bp = bread(ip->dev, ip->addrs[NDIRECT]);
+    bp = bread(ip->dev, ip->addrs[NDIRECT], 0);
    a = (uint*)bp->data;
    for(j = 0; j < NINDIRECT; j++){
      if(a[j])
@@ -471,7 +471,7 @@ readi(struct inode *ip, char *dst, uint off, uint n)
    n = ip->size - off;
  for(tot=0; tot<n; tot+=m, off+=m, dst+=m){
-    bp = bread(ip->dev, bmap(ip, off/BSIZE));
+    bp = bread(ip->dev, bmap(ip, off/BSIZE), 0);
    m = min(n - tot, BSIZE - off%BSIZE);
    memmove(dst, bp->data + off%BSIZE, m);
    brelse(bp);
@@ -499,7 +499,7 @@ writei(struct inode *ip, char *src, uint off, uint n)
    n = MAXFILE*BSIZE - off;
  for(tot=0; tot<n; tot+=m, off+=m, src+=m){
-    bp = bread(ip->dev, bmap(ip, off/BSIZE));
+    bp = bread(ip->dev, bmap(ip, off/BSIZE), 1);
    m = min(n - tot, BSIZE - off%BSIZE);
    memmove(bp->data + off%BSIZE, src, m);
    bwrite(bp);
@@ -536,7 +536,7 @@ dirlookup(struct inode *dp, char *name, uint *poff)
    panic("dirlookup not DIR");
  for(off = 0; off < dp->size; off += BSIZE){
-    bp = bread(dp->dev, bmap(dp, off / BSIZE));
+    bp = bread(dp->dev, bmap(dp, off / BSIZE), 0);
    for(de = (struct dirent*)bp->data;
        de < (struct dirent*)(bp->data + BSIZE);
        de++){

--- a/ide.c
+++ b/ide.c
@@ -130,7 +130,7 @@ iderw(struct buf *b)
 {
  struct buf **pp;
-  if(!(b->flags & B_BUSY))
+  if(!(b->flags & (B_BUSYR | B_BUSYW)))
    panic("iderw: buf not busy");
  if((b->flags & (B_VALID|B_DIRTY)) == B_VALID)
    panic("iderw: nothing to do");