xfs_qm.c

linux-2.6.15.6

	for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
	     ((dqp != (xfs_dquot_t *) &xfs_Gqm->qm_dqfreelist) &&
	      nreclaimed < howmany); ) {
		xfs_dqlock(dqp);

		/*
		 * We are racing with dqlookup here. Naturally we don't
		 * want to reclaim a dquot that lookup wants.
		 */
		if (dqp->dq_flags & XFS_DQ_WANT) {
			xfs_dqunlock(dqp);
			xfs_qm_freelist_unlock(xfs_Gqm);
			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
				return (nreclaimed);
			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
			goto tryagain;
		}

		/*
		 * If the dquot is inactive, we are assured that it is
		 * not on the mplist or the hashlist, and that makes our
		 * life easier.
		 */
		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
			ASSERT(dqp->q_mount == NULL);
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
			ASSERT(dqp->HL_PREVP == NULL);
			ASSERT(dqp->MPL_PREVP == NULL);
			XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
			nextdqp = dqp->dq_flnext;
			goto off_freelist;
		}

		ASSERT(dqp->MPL_PREVP);
		/*
		 * Try to grab the flush lock. If this dquot is in the process of
		 * getting flushed to disk, we don't want to reclaim it.
		 */
		if (! xfs_qm_dqflock_nowait(dqp)) {
			xfs_dqunlock(dqp);
			dqp = dqp->dq_flnext;
			continue;
		}

		/*
		 * We have the flush lock so we know that this is not in the
		 * process of being flushed. So, if this is dirty, flush it
		 * DELWRI so that we don't get a freelist infested with
		 * dirty dquots.
		 */
		if (XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqtrace_entry(dqp, "DQSHAKE: DQDIRTY");
			/*
			 * We flush it delayed write, so don't bother
			 * releasing the mplock.
			 */
			(void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
			xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
			dqp = dqp->dq_flnext;
			continue;
		}
		/*
		 * We're trying to get the hashlock out of order. This races
		 * with dqlookup; so, we giveup and goto the next dquot if
		 * we couldn't get the hashlock. This way, we won't starve
		 * a dqlookup process that holds the hashlock that is
		 * waiting for the freelist lock.
		 */
		if (! xfs_qm_dqhashlock_nowait(dqp)) {
			xfs_dqfunlock(dqp);
			xfs_dqunlock(dqp);
			dqp = dqp->dq_flnext;
			continue;
		}
		/*
		 * This races with dquot allocation code as well as dqflush_all
		 * and reclaim code. So, if we failed to grab the mplist lock,
		 * giveup everything and start over.
		 */
		hash = dqp->q_hash;
		ASSERT(hash);
		if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
			/* XXX put a sentinel so that we can come back here */
			xfs_dqfunlock(dqp);
			xfs_dqunlock(dqp);
			XFS_DQ_HASH_UNLOCK(hash);
			xfs_qm_freelist_unlock(xfs_Gqm);
			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
				return (nreclaimed);
			goto tryagain;
		}
		xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
#ifdef QUOTADEBUG
		cmn_err(CE_DEBUG, "Shake 0x%p, ID 0x%x\n",
			dqp, be32_to_cpu(dqp->q_core.d_id));
#endif
		ASSERT(dqp->q_nrefs == 0);
		nextdqp = dqp->dq_flnext;
		XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
		XQM_HASHLIST_REMOVE(hash, dqp);
		xfs_dqfunlock(dqp);
		xfs_qm_mplist_unlock(dqp->q_mount);
		XFS_DQ_HASH_UNLOCK(hash);

 off_freelist:
		XQM_FREELIST_REMOVE(dqp);
		xfs_dqunlock(dqp);
		nreclaimed++;
		XQM_STATS_INC(xqmstats.xs_qm_dqshake_reclaims);
		xfs_qm_dqdestroy(dqp);
		dqp = nextdqp;
	}
	xfs_qm_freelist_unlock(xfs_Gqm);
	return (nreclaimed);
}


/*
 * The kmem_shake interface is invoked when memory is running low.
 */
/* ARGSUSED */
STATIC int
xfs_qm_shake(int nr_to_scan, gfp_t gfp_mask)
{
	int	ndqused, nfree, n;

	if (!kmem_shake_allow(gfp_mask))
		return (0);
	if (!xfs_Gqm)
		return (0);

	nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
	/* incore dquots in all f/s's */
	ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;

	ASSERT(ndqused >= 0);

	if (nfree <= ndqused && nfree < ndquot)
		return (0);

	ndqused *= xfs_Gqm->qm_dqfree_ratio;	/* target # of free dquots */
	n = nfree - ndqused - ndquot;		/* # over target */

	return xfs_qm_shake_freelist(MAX(nfree, n));
}


/*
 * Just pop the least recently used dquot off the freelist and
 * recycle it. The returned dquot is locked.
 */
STATIC xfs_dquot_t *
xfs_qm_dqreclaim_one(void)
{
	xfs_dquot_t	*dqpout;
	xfs_dquot_t	*dqp;
	int		restarts;
	int		nflushes;

	restarts = 0;
	dqpout = NULL;
	nflushes = 0;

	/* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
 startagain:
	xfs_qm_freelist_lock(xfs_Gqm);

	FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
		xfs_dqlock(dqp);

		/*
		 * We are racing with dqlookup here. Naturally we don't
		 * want to reclaim a dquot that lookup wants. We release the
		 * freelist lock and start over, so that lookup will grab
		 * both the dquot and the freelistlock.
		 */
		if (dqp->dq_flags & XFS_DQ_WANT) {
			ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
			xfs_dqtrace_entry(dqp, "DQRECLAIM: DQWANT");
			xfs_dqunlock(dqp);
			xfs_qm_freelist_unlock(xfs_Gqm);
			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
				return (NULL);
			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
			goto startagain;
		}

		/*
		 * If the dquot is inactive, we are assured that it is
		 * not on the mplist or the hashlist, and that makes our
		 * life easier.
		 */
		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
			ASSERT(dqp->q_mount == NULL);
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
			ASSERT(dqp->HL_PREVP == NULL);
			ASSERT(dqp->MPL_PREVP == NULL);
			XQM_FREELIST_REMOVE(dqp);
			xfs_dqunlock(dqp);
			dqpout = dqp;
			XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
			break;
		}

		ASSERT(dqp->q_hash);
		ASSERT(dqp->MPL_PREVP);

		/*
		 * Try to grab the flush lock. If this dquot is in the process of
		 * getting flushed to disk, we don't want to reclaim it.
		 */
		if (! xfs_qm_dqflock_nowait(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}

		/*
		 * We have the flush lock so we know that this is not in the
		 * process of being flushed. So, if this is dirty, flush it
		 * DELWRI so that we don't get a freelist infested with
		 * dirty dquots.
		 */
		if (XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqtrace_entry(dqp, "DQRECLAIM: DQDIRTY");
			/*
			 * We flush it delayed write, so don't bother
			 * releasing the freelist lock.
			 */
			(void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
			xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
			continue;
		}

		if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
			xfs_dqfunlock(dqp);
			xfs_dqunlock(dqp);
			continue;
		}

		if (! xfs_qm_dqhashlock_nowait(dqp))
			goto mplistunlock;

		ASSERT(dqp->q_nrefs == 0);
		xfs_dqtrace_entry(dqp, "DQRECLAIM: UNLINKING");
		XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
		XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
		XQM_FREELIST_REMOVE(dqp);
		dqpout = dqp;
		XFS_DQ_HASH_UNLOCK(dqp->q_hash);
 mplistunlock:
		xfs_qm_mplist_unlock(dqp->q_mount);
		xfs_dqfunlock(dqp);
		xfs_dqunlock(dqp);
		if (dqpout)
			break;
	}

	xfs_qm_freelist_unlock(xfs_Gqm);
	return (dqpout);
}


/*------------------------------------------------------------------*/

/*
 * Return a new incore dquot. Depending on the number of
 * dquots in the system, we either allocate a new one on the kernel heap,
 * or reclaim a free one.
 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
 * to reclaim an existing one from the freelist.
 */
boolean_t
xfs_qm_dqalloc_incore(
	xfs_dquot_t **O_dqpp)
{
	xfs_dquot_t	*dqp;

	/*
	 * Check against high water mark to see if we want to pop
	 * a nincompoop dquot off the freelist.
	 */
	if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
		/*
		 * Try to recycle a dquot from the freelist.
		 */
		if ((dqp = xfs_qm_dqreclaim_one())) {
			XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
			/*
			 * Just zero the core here. The rest will get
			 * reinitialized by caller. XXX we shouldn't even
			 * do this zero ...
			 */
			memset(&dqp->q_core, 0, sizeof(dqp->q_core));
			*O_dqpp = dqp;
			return (B_FALSE);
		}
		XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
	}

	/*
	 * Allocate a brand new dquot on the kernel heap and return it
	 * to the caller to initialize.
	 */
	ASSERT(xfs_Gqm->qm_dqzone != NULL);
	*O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
	atomic_inc(&xfs_Gqm->qm_totaldquots);

	return (B_TRUE);
}


/*
 * Start a transaction and write the incore superblock changes to
 * disk. flags parameter indicates which fields have changed.
 */
int
xfs_qm_write_sb_changes(
	xfs_mount_t	*mp,
	__int64_t	flags)
{
	xfs_trans_t	*tp;
	int		error;

#ifdef QUOTADEBUG
	cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
#endif
	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
	if ((error = xfs_trans_reserve(tp, 0,
				      mp->m_sb.sb_sectsize + 128, 0,
				      0,
				      XFS_DEFAULT_LOG_COUNT))) {
		xfs_trans_cancel(tp, 0);
		return (error);
	}

	xfs_mod_sb(tp, flags);
	(void) xfs_trans_commit(tp, 0, NULL);

	return (0);
}


/* --------------- utility functions for vnodeops ---------------- */


/*
 * Given an inode, a uid and gid (from cred_t) make sure that we have
 * allocated relevant dquot(s) on disk, and that we won't exceed inode
 * quotas by creating this file.
 * This also attaches dquot(s) to the given inode after locking it,
 * and returns the dquots corresponding to the uid and/or gid.
 *
 * in	: inode (unlocked)
 * out	: udquot, gdquot with references taken and unlocked
 */
int
xfs_qm_vop_dqalloc(
	xfs_mount_t	*mp,
	xfs_inode_t	*ip,
	uid_t		uid,
	gid_t		gid,
	prid_t		prid,
	uint		flags,
	xfs_dquot_t	**O_udqpp,
	xfs_dquot_t	**O_gdqpp)
{
	int		error;
	xfs_dquot_t	*uq, *gq;
	uint		lockflags;

	if (!XFS_IS_QUOTA_ON(mp))
		return 0;

	lockflags = XFS_ILOCK_EXCL;
	xfs_ilock(ip, lockflags);

	if ((flags & XFS_QMOPT_INHERIT) &&
	    XFS_INHERIT_GID(ip, XFS_MTOVFS(mp)))
		gid = ip->i_d.di_gid;

	/*
	 * Attach the dquot(s) to this inode, doing a dquot allocation
	 * if necessary. The dquot(s) will not be locked.
	 */
	if (XFS_NOT_DQATTACHED(mp, ip)) {
		if ((error = xfs_qm_dqattach(ip, XFS_QMOPT_DQALLOC |
					    XFS_QMOPT_ILOCKED))) {
			xfs_iunlock(ip, lockflags);
			return (error);
		}
	}

	uq = gq = NULL;
	if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
		if (ip->i_d.di_uid != uid) {
			/*
			 * What we need is the dquot that has this uid, and
			 * if we send the inode to dqget, the uid of the inode
			 * takes priority over what's sent in the uid argument.
			 * We must unlock inode here before calling dqget if
			 * we're not sending the inode, because otherwise
			 * we'll deadlock by doing trans_reserve while
			 * holding ilock.
			 */
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
						 XFS_DQ_USER,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &uq))) {
				ASSERT(error != ENOENT);
				return (error);
			}
			/*
			 * Get the ilock in the right order.
			 */
			xfs_dqunlock(uq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			/*
			 * Take an extra reference, because we'll return
			 * this to caller
			 */
			ASSERT(ip->i_udquot);
			uq = ip->i_udquot;
			xfs_dqlock(uq);
			XFS_DQHOLD(uq);
			xfs_dqunlock(uq);
		}
	}
	if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
		if (ip->i_d.di_gid != gid) {
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
						 XFS_DQ_GROUP,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &gq))) {
				if (uq)
					xfs_qm_dqrele(uq);
				ASSERT(error != ENOENT);
				return (error);
			}
			xfs_dqunlock(gq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			ASSERT(ip->i_gdquot);
			gq = ip->i_gdquot;
			xfs_dqlock(gq);
			XFS_DQHOLD(gq);
			xfs_dqunlock(gq);
		}
	} else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
		if (ip->i_d.di_projid != prid) {
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
						 XFS_DQ_PROJ,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &gq))) {
				if (uq)
					xfs_qm_dqrele(uq);
				ASSERT(error != ENOENT);
				return (error);
			}
			xfs_dqunlock(gq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			ASSERT(ip->i_gdquot);
			gq = ip->i_gdquot;
			xfs_dqlock(gq);
			XFS_DQHOLD(gq);
			xfs_dqunlock(gq);
		}
	}
	if (uq)
		xfs_dqtrace_entry_ino(uq, "DQALLOC", ip);

	xfs_iunlock(ip, lockflags);
	if (O_udqpp)
		*O_udqpp = uq;
	else if (uq)
		xfs_qm_dqrele(uq);
	if (O_gdqpp)
		*O_gdqpp = gq;
	else if (gq)
		xfs_qm_dqrele(gq);
	return (0);
}

/*
 * Actually transfer ownership, and do dquot modifications.
 * These were already reserved.
 */
xfs_dquot_t *
xfs_qm_vop_chown(
	xfs_trans_t	*tp,
	xfs_inode_t	*ip,
	xfs_dquot_t	**IO_olddq,
	xfs_dquot_t	*newdq)
{
	xfs_dquot_t	*prevdq;
	uint		bfield = XFS_IS_REALTIME_INODE(ip) ?
				 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;

	ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
	ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));

	/* old dquot */
	prevdq = *IO_olddq;
	ASSERT(prevdq);
	ASSERT(prevdq != newdq);

	xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
	xfs_trans_mod_dquot(tp,