xfs_trans.c

linux 内核源代码

		if (delta < 0) {
			tp->t_blk_res_used += (uint)-delta;
			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
		}
		tp->t_fdblocks_delta += delta;
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
		break;
	case XFS_TRANS_SB_RES_FDBLOCKS:
		/*
		 * The allocation has already been applied to the
		 * in-core superblock's counter.  This should only
		 * be applied to the on-disk superblock.
		 */
		ASSERT(delta < 0);
		tp->t_res_fdblocks_delta += delta;
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
		break;
	case XFS_TRANS_SB_FREXTENTS:
		/*
		 * Track the number of blocks allocated in the
		 * transaction.  Make sure it does not exceed the
		 * number reserved.
		 */
		if (delta < 0) {
			tp->t_rtx_res_used += (uint)-delta;
			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
		}
		tp->t_frextents_delta += delta;
		break;
	case XFS_TRANS_SB_RES_FREXTENTS:
		/*
		 * The allocation has already been applied to the
		 * in-core superblock's counter.  This should only
		 * be applied to the on-disk superblock.
		 */
		ASSERT(delta < 0);
		tp->t_res_frextents_delta += delta;
		break;
	case XFS_TRANS_SB_DBLOCKS:
		ASSERT(delta > 0);
		tp->t_dblocks_delta += delta;
		break;
	case XFS_TRANS_SB_AGCOUNT:
		ASSERT(delta > 0);
		tp->t_agcount_delta += delta;
		break;
	case XFS_TRANS_SB_IMAXPCT:
		tp->t_imaxpct_delta += delta;
		break;
	case XFS_TRANS_SB_REXTSIZE:
		tp->t_rextsize_delta += delta;
		break;
	case XFS_TRANS_SB_RBMBLOCKS:
		tp->t_rbmblocks_delta += delta;
		break;
	case XFS_TRANS_SB_RBLOCKS:
		tp->t_rblocks_delta += delta;
		break;
	case XFS_TRANS_SB_REXTENTS:
		tp->t_rextents_delta += delta;
		break;
	case XFS_TRANS_SB_REXTSLOG:
		tp->t_rextslog_delta += delta;
		break;
	default:
		ASSERT(0);
		return;
	}

	tp->t_flags |= flags;
}

/*
 * xfs_trans_apply_sb_deltas() is called from the commit code
 * to bring the superblock buffer into the current transaction
 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 *
 * For now we just look at each field allowed to change and change
 * it if necessary.
 */
STATIC void
xfs_trans_apply_sb_deltas(
	xfs_trans_t	*tp)
{
	xfs_dsb_t	*sbp;
	xfs_buf_t	*bp;
	int		whole = 0;

	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
	sbp = XFS_BUF_TO_SBP(bp);

	/*
	 * Check that superblock mods match the mods made to AGF counters.
	 */
	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
		tp->t_ag_btree_delta));

	/*
	 * Only update the superblock counters if we are logging them
	 */
	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
		if (tp->t_icount_delta)
			be64_add(&sbp->sb_icount, tp->t_icount_delta);
		if (tp->t_ifree_delta)
			be64_add(&sbp->sb_ifree, tp->t_ifree_delta);
		if (tp->t_fdblocks_delta)
			be64_add(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
		if (tp->t_res_fdblocks_delta)
			be64_add(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
	}

	if (tp->t_frextents_delta)
		be64_add(&sbp->sb_frextents, tp->t_frextents_delta);
	if (tp->t_res_frextents_delta)
		be64_add(&sbp->sb_frextents, tp->t_res_frextents_delta);

	if (tp->t_dblocks_delta) {
		be64_add(&sbp->sb_dblocks, tp->t_dblocks_delta);
		whole = 1;
	}
	if (tp->t_agcount_delta) {
		be32_add(&sbp->sb_agcount, tp->t_agcount_delta);
		whole = 1;
	}
	if (tp->t_imaxpct_delta) {
		sbp->sb_imax_pct += tp->t_imaxpct_delta;
		whole = 1;
	}
	if (tp->t_rextsize_delta) {
		be32_add(&sbp->sb_rextsize, tp->t_rextsize_delta);
		whole = 1;
	}
	if (tp->t_rbmblocks_delta) {
		be32_add(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
		whole = 1;
	}
	if (tp->t_rblocks_delta) {
		be64_add(&sbp->sb_rblocks, tp->t_rblocks_delta);
		whole = 1;
	}
	if (tp->t_rextents_delta) {
		be64_add(&sbp->sb_rextents, tp->t_rextents_delta);
		whole = 1;
	}
	if (tp->t_rextslog_delta) {
		sbp->sb_rextslog += tp->t_rextslog_delta;
		whole = 1;
	}

	if (whole)
		/*
		 * Log the whole thing, the fields are noncontiguous.
		 */
		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
	else
		/*
		 * Since all the modifiable fields are contiguous, we
		 * can get away with this.
		 */
		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
				  offsetof(xfs_dsb_t, sb_frextents) +
				  sizeof(sbp->sb_frextents) - 1);

	tp->t_mountp->m_super->s_dirt = 1;
}

/*
 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
 * and apply superblock counter changes to the in-core superblock.  The
 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
 * applied to the in-core superblock.  The idea is that that has already been
 * done.
 *
 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
 * However, we have to ensure that we only modify each superblock field only
 * once because the application of the delta values may not be atomic. That can
 * lead to ENOSPC races occurring if we have two separate modifcations of the
 * free space counter to put back the entire reservation and then take away
 * what we used.
 *
 * If we are not logging superblock counters, then the inode allocated/free and
 * used block counts are not updated in the on disk superblock. In this case,
 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
 * still need to update the incore superblock with the changes.
 */
STATIC void
xfs_trans_unreserve_and_mod_sb(
	xfs_trans_t	*tp)
{
	xfs_mod_sb_t	msb[14];	/* If you add cases, add entries */
	xfs_mod_sb_t	*msbp;
	xfs_mount_t	*mp = tp->t_mountp;
	/* REFERENCED */
	int		error;
	int		rsvd;
	int64_t		blkdelta = 0;
	int64_t		rtxdelta = 0;

	msbp = msb;
	rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;

	/* calculate free blocks delta */
	if (tp->t_blk_res > 0)
		blkdelta = tp->t_blk_res;

	if ((tp->t_fdblocks_delta != 0) &&
	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
	        blkdelta += tp->t_fdblocks_delta;

	if (blkdelta != 0) {
		msbp->msb_field = XFS_SBS_FDBLOCKS;
		msbp->msb_delta = blkdelta;
		msbp++;
	}

	/* calculate free realtime extents delta */
	if (tp->t_rtx_res > 0)
		rtxdelta = tp->t_rtx_res;

	if ((tp->t_frextents_delta != 0) &&
	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
		rtxdelta += tp->t_frextents_delta;

	if (rtxdelta != 0) {
		msbp->msb_field = XFS_SBS_FREXTENTS;
		msbp->msb_delta = rtxdelta;
		msbp++;
	}

	/* apply remaining deltas */

	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
		if (tp->t_icount_delta != 0) {
			msbp->msb_field = XFS_SBS_ICOUNT;
			msbp->msb_delta = tp->t_icount_delta;
			msbp++;
		}
		if (tp->t_ifree_delta != 0) {
			msbp->msb_field = XFS_SBS_IFREE;
			msbp->msb_delta = tp->t_ifree_delta;
			msbp++;
		}
	}

	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
		if (tp->t_dblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_DBLOCKS;
			msbp->msb_delta = tp->t_dblocks_delta;
			msbp++;
		}
		if (tp->t_agcount_delta != 0) {
			msbp->msb_field = XFS_SBS_AGCOUNT;
			msbp->msb_delta = tp->t_agcount_delta;
			msbp++;
		}
		if (tp->t_imaxpct_delta != 0) {
			msbp->msb_field = XFS_SBS_IMAX_PCT;
			msbp->msb_delta = tp->t_imaxpct_delta;
			msbp++;
		}
		if (tp->t_rextsize_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTSIZE;
			msbp->msb_delta = tp->t_rextsize_delta;
			msbp++;
		}
		if (tp->t_rbmblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_RBMBLOCKS;
			msbp->msb_delta = tp->t_rbmblocks_delta;
			msbp++;
		}
		if (tp->t_rblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_RBLOCKS;
			msbp->msb_delta = tp->t_rblocks_delta;
			msbp++;
		}
		if (tp->t_rextents_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTENTS;
			msbp->msb_delta = tp->t_rextents_delta;
			msbp++;
		}
		if (tp->t_rextslog_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTSLOG;
			msbp->msb_delta = tp->t_rextslog_delta;
			msbp++;
		}
	}

	/*
	 * If we need to change anything, do it.
	 */
	if (msbp > msb) {
		error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
			(uint)(msbp - msb), rsvd);
		ASSERT(error == 0);
	}
}


/*
 * xfs_trans_commit
 *
 * Commit the given transaction to the log a/synchronously.
 *
 * XFS disk error handling mechanism is not based on a typical
 * transaction abort mechanism. Logically after the filesystem
 * gets marked 'SHUTDOWN', we can't let any new transactions
 * be durable - ie. committed to disk - because some metadata might
 * be inconsistent. In such cases, this returns an error, and the
 * caller may assume that all locked objects joined to the transaction
 * have already been unlocked as if the commit had succeeded.
 * Do not reference the transaction structure after this call.
 */
 /*ARGSUSED*/
int
_xfs_trans_commit(
	xfs_trans_t	*tp,
	uint		flags,
	int		*log_flushed)
{
	xfs_log_iovec_t		*log_vector;
	int			nvec;
	xfs_mount_t		*mp;
	xfs_lsn_t		commit_lsn;
	/* REFERENCED */
	int			error;
	int			log_flags;
	int			sync;
#define	XFS_TRANS_LOGVEC_COUNT	16
	xfs_log_iovec_t		log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
	void			*commit_iclog;
	int			shutdown;

	commit_lsn = -1;

	/*
	 * Determine whether this commit is releasing a permanent
	 * log reservation or not.
	 */
	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
		log_flags = XFS_LOG_REL_PERM_RESERV;
	} else {
		log_flags = 0;
	}
	mp = tp->t_mountp;

	/*
	 * If there is nothing to be logged by the transaction,
	 * then unlock all of the items associated with the
	 * transaction and free the transaction structure.
	 * Also make sure to return any reserved blocks to
	 * the free pool.
	 */
shut_us_down:
	shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
	if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
		xfs_trans_unreserve_and_mod_sb(tp);
		/*
		 * It is indeed possible for the transaction to be
		 * not dirty but the dqinfo portion to be. All that
		 * means is that we have some (non-persistent) quota
		 * reservations that need to be unreserved.
		 */
		XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
		if (tp->t_ticket) {
			commit_lsn = xfs_log_done(mp, tp->t_ticket,
							NULL, log_flags);
			if (commit_lsn == -1 && !shutdown)
				shutdown = XFS_ERROR(EIO);
		}
		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
		xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
		xfs_trans_free_busy(tp);
		xfs_trans_free(tp);
		XFS_STATS_INC(xs_trans_empty);
		return (shutdown);
	}
	ASSERT(tp->t_ticket != NULL);

	/*
	 * If we need to update the superblock, then do it now.
	 */
	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
		xfs_trans_apply_sb_deltas(tp);
	}
	XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);

	/*
	 * Ask each log item how many log_vector entries it will
	 * need so we can figure out how many to allocate.
	 * Try to avoid the kmem_alloc() call in the common case
	 * by using a vector from the stack when it fits.
	 */
	nvec = xfs_trans_count_vecs(tp);
	if (nvec == 0) {
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
		goto shut_us_down;
	} else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
		log_vector = log_vector_fast;
	} else {
		log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
						   sizeof(xfs_log_iovec_t),
						   KM_SLEEP);
	}

	/*
	 * Fill in the log_vector and pin the logged items, and
	 * then write the transaction to the log.
	 */
	xfs_trans_fill_vecs(tp, log_vector);

	error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));

	/*
	 * The transaction is committed incore here, and can go out to disk
	 * at any time after this call.  However, all the items associated
	 * with the transaction are still locked and pinned in memory.
	 */
	commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);

	tp->t_commit_lsn = commit_lsn;
	if (nvec > XFS_TRANS_LOGVEC_COUNT) {
		kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
	}

	/*
	 * If we got a log write error. Unpin the logitems that we
	 * had pinned, clean up, free trans structure, and return error.
	 */
	if (error || commit_lsn == -1) {
		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
		xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
		return XFS_ERROR(EIO);
	}

	/*
	 * Once the transaction has committed, unused
	 * reservations need to be released and changes to
	 * the superblock need to be reflected in the in-core
	 * version.  Do that now.
	 */
	xfs_trans_unreserve_and_mod_sb(tp);

	sync = tp->t_flags & XFS_TRANS_SYNC;

	/*
	 * Tell the LM to call the transaction completion routine
	 * when the log write with LSN commit_lsn completes (e.g.
	 * when the transaction commit really hits the on-disk log).
	 * After this call we cannot reference tp, because the call
	 * can happen at any time and the call will free the transaction
	 * structure pointed to by tp.  The only case where we call
	 * the completion routine (xfs_trans_committed) directly is
	 * if the log is turned off on a debug kernel or we're
	 * running in simulation mode (the log is explicitly turned
	 * off).
	 */
	tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
	tp->t_logcb.cb_arg = tp;