/*
* Delayed write.
*
* The buffer is marked dirty, but is not queued for I/O.
* This routine should be used when the buffer is expected
* to be modified again soon, typically a small write that
* partially fills a buffer.
*
* NB: magnetic tapes cannot be delayed; they must be
* written in the order that the writes are requested.
*
* Described in Leffler, et al. (pp. 208-213).
*/
void
bdwrite(struct buf *bp)
{
int s;
/*
* If the block hasn't been seen before:
* (1) Mark it as having been seen,
* (2) Charge for the write.
* (3) Make sure it's on its vnode's correct block list,
* (4) If a buffer is rewritten, move it to end of dirty list
*/
if (!ISSET(bp->b_flags, B_DELWRI)) {
SET(bp->b_flags, B_DELWRI);
s = splbio();
reassignbuf(bp);
splx(s);
curproc->p_stats->p_ru.ru_oublock++; /* XXX */
}
/* If this is a tape block, write the block now. */
if (major(bp->b_dev) < nblkdev &&
bdevsw[major(bp->b_dev)].d_type == D_TAPE) {
bawrite(bp);
return;
}
/* Otherwise, the "write" is done, so mark and release the buffer. */
CLR(bp->b_flags, B_NEEDCOMMIT);
SET(bp->b_flags, B_DONE);
brelse(bp);
}
/*
* Synchronous write.
* Release buffer on completion.
*/
int
bwrite(register struct buf *bp)
{
int rv;
if(bp->b_flags & B_INVAL) {
brelse(bp);
return (0);
} else {
int wasdelayed;
if(!(bp->b_flags & B_BUSY))
panic("bwrite: not busy");
wasdelayed = bp->b_flags & B_DELWRI;
bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_ASYNC|B_DELWRI);
if(wasdelayed)
reassignbuf(bp, bp->b_vp);
bp->b_flags |= B_DIRTY;
bp->b_vp->v_numoutput++;
VOP_STRATEGY(bp);
rv = biowait(bp);
brelse(bp);
return (rv);
}
}
/*
* Must be called at splbio()
*/
void
buf_undirty(struct buf *bp)
{
splassert(IPL_BIO);
if (ISSET(bp->b_flags, B_DELWRI)) {
CLR(bp->b_flags, B_DELWRI);
reassignbuf(bp);
}
}
/*
* Must be called at splbio()
*/
void
buf_undirty(struct buf *bp)
{
splassert(IPL_BIO);
#ifdef DIAGNOSTIC
if (!ISSET(bp->b_flags, B_BUSY))
panic("Trying to undirty buffer on freelist!");
#endif
if (ISSET(bp->b_flags, B_DELWRI)) {
CLR(bp->b_flags, B_DELWRI);
reassignbuf(bp);
}
}
/*
* Delayed write.
*
* The buffer is marked dirty, but is not queued for I/O.
* This routine should be used when the buffer is expected
* to be modified again soon, typically a small write that
* partially fills a buffer.
*
* NB: magnetic tapes cannot be delayed; they must be
* written in the order that the writes are requested.
*/
void
bdwrite(register struct buf *bp)
{
if(!(bp->b_flags & B_BUSY))
panic("bdwrite: not busy");
if(bp->b_flags & B_INVAL) {
brelse(bp);
}
if(bp->b_flags & B_TAPE) {
bwrite(bp);
return;
}
bp->b_flags &= ~(B_READ|B_DONE);
bp->b_flags |= B_DIRTY|B_DELWRI;
reassignbuf(bp, bp->b_vp);
brelse(bp);
return;
}
/*
* Asynchronous write.
* Start I/O on a buffer, but do not wait for it to complete.
* The buffer is released when the I/O completes.
*/
void
bawrite(register struct buf *bp)
{
if(!(bp->b_flags & B_BUSY))
panic("bawrite: not busy");
if(bp->b_flags & B_INVAL)
brelse(bp);
else {
int wasdelayed;
wasdelayed = bp->b_flags & B_DELWRI;
bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
if(wasdelayed)
reassignbuf(bp, bp->b_vp);
bp->b_flags |= B_DIRTY | B_ASYNC;
bp->b_vp->v_numoutput++;
VOP_STRATEGY(bp);
}
}
/*
* Release a buffer on to the free lists.
* Described in Bach (p. 46).
*/
void
brelse(struct buf *bp)
{
struct bqueues *bufq;
int s;
/* Block disk interrupts. */
s = splbio();
/*
* Determine which queue the buffer should be on, then put it there.
*/
/* If it's locked, don't report an error; try again later. */
if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR))
CLR(bp->b_flags, B_ERROR);
/* If it's not cacheable, or an error, mark it invalid. */
if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR)))
SET(bp->b_flags, B_INVAL);
if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) {
/*
* If it's invalid or empty, dissociate it from its vnode
* and put on the head of the appropriate queue.
*/
if (LIST_FIRST(&bp->b_dep) != NULL)
buf_deallocate(bp);
if (ISSET(bp->b_flags, B_DELWRI)) {
CLR(bp->b_flags, B_DELWRI);
}
if (bp->b_vp) {
reassignbuf(bp);
brelvp(bp);
}
if (bp->b_bufsize <= 0) {
/* no data */
bufq = &bufqueues[BQ_EMPTY];
numemptybufs++;
} else {
/* invalid data */
bufq = &bufqueues[BQ_CLEAN];
numfreepages += btoc(bp->b_bufsize);
numcleanpages += btoc(bp->b_bufsize);
}
binsheadfree(bp, bufq);
} else {
/*
* It has valid data. Put it on the end of the appropriate
* queue, so that it'll stick around for as long as possible.
*/
if (ISSET(bp->b_flags, B_LOCKED))
/* locked in core */
bufq = &bufqueues[BQ_LOCKED];
else {
numfreepages += btoc(bp->b_bufsize);
if (!ISSET(bp->b_flags, B_DELWRI)) {
numcleanpages += btoc(bp->b_bufsize);
bufq = &bufqueues[BQ_CLEAN];
} else {
numdirtypages += btoc(bp->b_bufsize);
bufq = &bufqueues[BQ_DIRTY];
}
}
if (ISSET(bp->b_flags, B_AGE))
binsheadfree(bp, bufq);
else
binstailfree(bp, bufq);
}
/* Unlock the buffer. */
CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE | B_DEFERRED));
/* Wake up syncer and cleaner processes waiting for buffers */
if (nobuffers) {
wakeup(&nobuffers);
nobuffers = 0;
}
/* Wake up any processes waiting for any buffer to become free. */
if (needbuffer && (numcleanpages > locleanpages)) {
needbuffer--;
wakeup_one(&needbuffer);
}
splx(s);
/* Wake up any processes waiting for _this_ buffer to become free. */
if (ISSET(bp->b_flags, B_WANTED)) {
CLR(bp->b_flags, B_WANTED);
wakeup(bp);
}
}
/*
* Block write. Described in Bach (p.56)
*/
int
bwrite(struct buf *bp)
{
int rv, async, wasdelayed, s;
struct vnode *vp;
struct mount *mp;
/*
* Remember buffer type, to switch on it later. If the write was
* synchronous, but the file system was mounted with MNT_ASYNC,
* convert it to a delayed write.
* XXX note that this relies on delayed tape writes being converted
* to async, not sync writes (which is safe, but ugly).
*/
async = ISSET(bp->b_flags, B_ASYNC);
if (!async && bp->b_vp && bp->b_vp->v_mount &&
ISSET(bp->b_vp->v_mount->mnt_flag, MNT_ASYNC)) {
bdwrite(bp);
return (0);
}
/*
* Collect statistics on synchronous and asynchronous writes.
* Writes to block devices are charged to their associated
* filesystem (if any).
*/
if ((vp = bp->b_vp) != NULL) {
if (vp->v_type == VBLK)
mp = vp->v_specmountpoint;
else
mp = vp->v_mount;
if (mp != NULL) {
if (async)
mp->mnt_stat.f_asyncwrites++;
else
mp->mnt_stat.f_syncwrites++;
}
}
wasdelayed = ISSET(bp->b_flags, B_DELWRI);
CLR(bp->b_flags, (B_READ | B_DONE | B_ERROR | B_DELWRI));
s = splbio();
/*
* If not synchronous, pay for the I/O operation and make
* sure the buf is on the correct vnode queue. We have
* to do this now, because if we don't, the vnode may not
* be properly notified that its I/O has completed.
*/
if (wasdelayed) {
reassignbuf(bp);
} else
curproc->p_stats->p_ru.ru_oublock++;
/* Initiate disk write. Make sure the appropriate party is charged. */
bp->b_vp->v_numoutput++;
splx(s);
SET(bp->b_flags, B_WRITEINPROG);
VOP_STRATEGY(bp);
if (async)
return (0);
/*
* If I/O was synchronous, wait for it to complete.
*/
rv = biowait(bp);
/* Release the buffer. */
brelse(bp);
return (rv);
}