/*
* Get and lock the buffer for the caller if it is not already
* locked within the given transaction. If it is already locked
* within the transaction, just increment its lock recursion count
* and return a pointer to it.
*
* Use the fast path function xfs_trans_buf_item_match() or the buffer
* cache routine incore_match() to find the buffer
* if it is already owned by this transaction.
*
* If we don't already own the buffer, use get_buf() to get it.
* If it doesn't yet have an associated xfs_buf_log_item structure,
* then allocate one and add the item to this transaction.
*
* If the transaction pointer is NULL, make this just a normal
* get_buf() call.
*/
xfs_buf_t *
xfs_trans_get_buf(xfs_trans_t *tp,
xfs_buftarg_t *target_dev,
xfs_daddr_t blkno,
int len,
uint flags)
{
xfs_buf_t *bp;
xfs_buf_log_item_t *bip;
if (flags == 0)
flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
/*
* Default to a normal get_buf() call if the tp is NULL.
*/
if (tp == NULL) {
bp = xfs_buf_get_flags(target_dev, blkno, len,
flags | BUF_BUSY);
return(bp);
}
/*
* If we find the buffer in the cache with this transaction
* pointer in its b_fsprivate2 field, then we know we already
* have it locked. In this case we just increment the lock
* recursion count and return the buffer to the caller.
*/
if (tp->t_items.lic_next == NULL) {
bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
} else {
bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
}
if (bp != NULL) {
ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
xfs_buftrace("TRANS GET RECUR SHUT", bp);
XFS_BUF_SUPER_STALE(bp);
}
/*
* If the buffer is stale then it was binval'ed
* since last read. This doesn't matter since the
* caller isn't allowed to use the data anyway.
*/
else if (XFS_BUF_ISSTALE(bp)) {
xfs_buftrace("TRANS GET RECUR STALE", bp);
ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
}
ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
ASSERT(bip != NULL);
ASSERT(atomic_read(&bip->bli_refcount) > 0);
bip->bli_recur++;
xfs_buftrace("TRANS GET RECUR", bp);
xfs_buf_item_trace("GET RECUR", bip);
return (bp);
}
/*
* Called when we want to stop a buffer from getting written or read.
* We attach the EIO error, muck with its flags, and call biodone
* so that the proper iodone callbacks get called.
*/
int
xfs_bioerror(
xfs_buf_t *bp)
{
#ifdef XFSERRORDEBUG
ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
#endif
/*
* No need to wait until the buffer is unpinned.
* We aren't flushing it.
*/
xfs_buftrace("XFS IOERROR", bp);
XFS_BUF_ERROR(bp, EIO);
/*
* We're calling biodone, so delete B_DONE flag. Either way
* we have to call the iodone callback, and calling biodone
* probably is the best way since it takes care of
* GRIO as well.
*/
XFS_BUF_UNREAD(bp);
XFS_BUF_UNDELAYWRITE(bp);
XFS_BUF_UNDONE(bp);
XFS_BUF_STALE(bp);
XFS_BUF_CLR_BDSTRAT_FUNC(bp);
xfs_biodone(bp);
return (EIO);
}
/*
* This is called to unpin the buffer associated with the buf log
* item which was previously pinned with a call to xfs_buf_item_pin().
* Just call bunpin() on the buffer to do this.
*
* Also drop the reference to the buf item for the current transaction.
* If the XFS_BLI_STALE flag is set and we are the last reference,
* then free up the buf log item and unlock the buffer.
*/
void
xfs_buf_item_unpin(
xfs_buf_log_item_t *bip,
int stale)
{
xfs_mount_t *mp;
xfs_buf_t *bp;
int freed;
SPLDECL(s);
bp = bip->bli_buf;
ASSERT(bp != NULL);
ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
ASSERT(atomic_read(&bip->bli_refcount) > 0);
xfs_buf_item_trace("UNPIN", bip);
xfs_buftrace("XFS_UNPIN", bp);
freed = atomic_dec_and_test(&bip->bli_refcount);
mp = bip->bli_item.li_mountp;
xfs_bunpin(bp);
if (freed && stale) {
ASSERT(bip->bli_flags & XFS_BLI_STALE);
ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
ASSERT(XFS_BUF_ISSTALE(bp));
ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
xfs_buf_item_trace("UNPIN STALE", bip);
xfs_buftrace("XFS_UNPIN STALE", bp);
/*
* If we get called here because of an IO error, we may
* or may not have the item on the AIL. xfs_trans_delete_ail()
* will take care of that situation.
* xfs_trans_delete_ail() drops the AIL lock.
*/
if (bip->bli_flags & XFS_BLI_STALE_INODE) {
xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
XFS_BUF_SET_FSPRIVATE(bp, NULL);
XFS_BUF_CLR_IODONE_FUNC(bp);
} else {
AIL_LOCK(mp,s);
xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip, s);
xfs_buf_item_relse(bp);
ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
}
xfs_buf_relse(bp);
}
/*
* Wrapper around bdstrat so that we can stop data from going to disk in case
* we are shutting down the filesystem. Typically user data goes thru this
* path; one of the exceptions is the superblock.
*/
void
xfsbdstrat(
struct xfs_mount *mp,
struct xfs_buf *bp)
{
ASSERT(mp);
if (!XFS_FORCED_SHUTDOWN(mp)) {
xfs_buf_iorequest(bp);
return;
}
xfs_buftrace("XFSBDSTRAT IOERROR", bp);
xfs_bioerror_relse(bp);
}
/*
* This is called to pin the buffer associated with the buf log
* item in memory so it cannot be written out. Simply call bpin()
* on the buffer to do this.
*/
void
xfs_buf_item_pin(
xfs_buf_log_item_t *bip)
{
xfs_buf_t *bp;
bp = bip->bli_buf;
ASSERT(XFS_BUF_ISBUSY(bp));
ASSERT(atomic_read(&bip->bli_refcount) > 0);
ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
(bip->bli_flags & XFS_BLI_STALE));
xfs_buf_item_trace("PIN", bip);
xfs_buftrace("XFS_PIN", bp);
xfs_bpin(bp);
}
/*
* Wrapper around bdstrat so that we can stop data
* from going to disk in case we are shutting down the filesystem.
* Typically user data goes thru this path; one of the exceptions
* is the superblock.
*/
int
xfsbdstrat(
struct xfs_mount *mp,
struct xfs_buf *bp)
{
ASSERT(mp);
if (!XFS_FORCED_SHUTDOWN(mp)) {
/* Grio redirection would go here
* if (XFS_BUF_IS_GRIO(bp)) {
*/
pagebuf_iorequest(bp);
return 0;
}
xfs_buftrace("XFSBDSTRAT IOERROR", bp);
return (xfs_bioerror_relse(bp));
}
/*
* All xfs metadata buffers except log state machine buffers
* get this attached as their b_bdstrat callback function.
* This is so that we can catch a buffer
* after prematurely unpinning it to forcibly shutdown the filesystem.
*/
int
xfs_bdstrat_cb(struct xfs_buf *bp)
{
if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
/*
* Metadata write that didn't get logged but
* written delayed anyway. These aren't associated
* with a transaction, and can be ignored.
*/
if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
(XFS_BUF_ISREAD(bp)) == 0)
return (xfs_bioerror_relse(bp));
else
return (xfs_bioerror(bp));
}
xfs_buf_iorequest(bp);
return 0;
}
/*
* Same as xfs_bioerror, except that we are releasing the buffer
* here ourselves, and avoiding the biodone call.
* This is meant for userdata errors; metadata bufs come with
* iodone functions attached, so that we can track down errors.
*/
int
xfs_bioerror_relse(
xfs_buf_t *bp)
{
int64_t fl;
ASSERT(XFS_BUF_IODONE_FUNC(bp) != xfs_buf_iodone_callbacks);
ASSERT(XFS_BUF_IODONE_FUNC(bp) != xlog_iodone);
xfs_buftrace("XFS IOERRELSE", bp);
fl = XFS_BUF_BFLAGS(bp);
/*
* No need to wait until the buffer is unpinned.
* We aren't flushing it.
*
* chunkhold expects B_DONE to be set, whether
* we actually finish the I/O or not. We don't want to
* change that interface.
*/
XFS_BUF_UNREAD(bp);
XFS_BUF_UNDELAYWRITE(bp);
XFS_BUF_DONE(bp);
XFS_BUF_STALE(bp);
XFS_BUF_CLR_IODONE_FUNC(bp);
XFS_BUF_CLR_BDSTRAT_FUNC(bp);
if (!(fl & XFS_B_ASYNC)) {
/*
* Mark b_error and B_ERROR _both_.
* Lot's of chunkcache code assumes that.
* There's no reason to mark error for
* ASYNC buffers.
*/
XFS_BUF_ERROR(bp, EIO);
XFS_BUF_V_IODONESEMA(bp);
} else {
xfs_buf_relse(bp);
}
return (EIO);
}
/*
* This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
* failed to get the inode flush lock but did get the inode locked SHARED.
* Here we're trying to see if the inode buffer is incore, and if so whether it's
* marked delayed write. If that's the case, we'll initiate a bawrite on that
* buffer to expedite the process.
*
* We aren't holding the AIL lock (or the flush lock) when this gets called,
* so it is inherently race-y.
*/
STATIC void
xfs_inode_item_pushbuf(
xfs_inode_log_item_t *iip)
{
xfs_inode_t *ip;
xfs_mount_t *mp;
xfs_buf_t *bp;
uint dopush;
ip = iip->ili_inode;
ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
/*
* The ili_pushbuf_flag keeps others from
* trying to duplicate our effort.
*/
ASSERT(iip->ili_pushbuf_flag != 0);
ASSERT(iip->ili_push_owner == current_pid());
/*
* If a flush is not in progress anymore, chances are that the
* inode was taken off the AIL. So, just get out.
*/
if (completion_done(&ip->i_flush) ||
((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0)) {
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
mp = ip->i_mount;
bp = xfs_incore(mp->m_ddev_targp, iip->ili_format.ilf_blkno,
iip->ili_format.ilf_len, XFS_INCORE_TRYLOCK);
if (bp != NULL) {
if (XFS_BUF_ISDELAYWRITE(bp)) {
/*
* We were racing with iflush because we don't hold
* the AIL lock or the flush lock. However, at this point,
* we have the buffer, and we know that it's dirty.
* So, it's possible that iflush raced with us, and
* this item is already taken off the AIL.
* If not, we can flush it async.
*/
dopush = ((iip->ili_item.li_flags & XFS_LI_IN_AIL) &&
!completion_done(&ip->i_flush));
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_buftrace("INODE ITEM PUSH", bp);
if (XFS_BUF_ISPINNED(bp)) {
xfs_log_force(mp, (xfs_lsn_t)0,
XFS_LOG_FORCE);
}
if (dopush) {
int error;
error = xfs_bawrite(mp, bp);
if (error)
xfs_fs_cmn_err(CE_WARN, mp,
"xfs_inode_item_pushbuf: pushbuf error %d on iip %p, bp %p",
error, iip, bp);
} else {
xfs_buf_relse(bp);
}
} else {
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_buf_relse(bp);
}
return;
}
/*
* We have to be careful about resetting pushbuf flag too early (above).
* Even though in theory we can do it as soon as we have the buflock,
* we don't want others to be doing work needlessly. They'll come to
* this function thinking that pushing the buffer is their
* responsibility only to find that the buffer is still locked by
* another doing the same thing
*/
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
/*
* This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
* failed to get the inode flush lock but did get the inode locked SHARED.
* Here we're trying to see if the inode buffer is incore, and if so whether it's
* marked delayed write. If that's the case, we'll initiate a bawrite on that
* buffer to expedite the process.
*
* We aren't holding the AIL_LOCK (or the flush lock) when this gets called,
* so it is inherently race-y.
*/
STATIC void
xfs_inode_item_pushbuf(
xfs_inode_log_item_t *iip)
{
xfs_inode_t *ip;
xfs_mount_t *mp;
xfs_buf_t *bp;
uint dopush;
ip = iip->ili_inode;
ASSERT(ismrlocked(&(ip->i_lock), MR_ACCESS));
/*
* The ili_pushbuf_flag keeps others from
* trying to duplicate our effort.
*/
ASSERT(iip->ili_pushbuf_flag != 0);
ASSERT(iip->ili_push_owner == get_thread_id());
/*
* If flushlock isn't locked anymore, chances are that the
* inode flush completed and the inode was taken off the AIL.
* So, just get out.
*/
if ((valusema(&(ip->i_flock)) > 0) ||
((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0)) {
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
mp = ip->i_mount;
bp = xfs_incore(mp->m_ddev_targ, iip->ili_format.ilf_blkno,
iip->ili_format.ilf_len, XFS_INCORE_TRYLOCK);
if (bp != NULL) {
if (XFS_BUF_ISDELAYWRITE(bp)) {
/*
* We were racing with iflush because we don't hold
* the AIL_LOCK or the flush lock. However, at this point,
* we have the buffer, and we know that it's dirty.
* So, it's possible that iflush raced with us, and
* this item is already taken off the AIL.
* If not, we can flush it async.
*/
dopush = ((iip->ili_item.li_flags & XFS_LI_IN_AIL) &&
(valusema(&(ip->i_flock)) <= 0));
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_buftrace("INODE ITEM PUSH", bp);
if (XFS_BUF_ISPINNED(bp)) {
xfs_log_force(mp, (xfs_lsn_t)0,
XFS_LOG_FORCE);
}
if (dopush) {
xfs_bawrite(mp, bp);
} else {
xfs_buf_relse(bp);
}
} else {
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_buf_relse(bp);
}
return;
}
/*
* We have to be careful about resetting pushbuf flag too early (above).
* Eventhough in theory we can do it as soon as we have the buflock,
* we don't want others to be doing work needlessly. They'll come to
* this function thinking that pushing the buffer is there responsibility
* only to find that the buffer is still locked by another doing the
* same thing.XXX
*/
iip->ili_pushbuf_flag = 0;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
