xfs_qm.c

linux-2.6.15.6

	 */
	nowait = (boolean_t)(flags & SYNC_BDFLUSH || (flags & SYNC_WAIT) == 0);

  again:
	xfs_qm_mplist_lock(mp);
	/*
	 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
	 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
	 * when we have the mplist lock, we know that dquots will be consistent
	 * as long as we have it locked.
	 */
	if (! XFS_IS_QUOTA_ON(mp)) {
		xfs_qm_mplist_unlock(mp);
		return (0);
	}
	FOREACH_DQUOT_IN_MP(dqp, mp) {
		/*
		 * If this is vfs_sync calling, then skip the dquots that
		 * don't 'seem' to be dirty. ie. don't acquire dqlock.
		 * This is very similar to what xfs_sync does with inodes.
		 */
		if (flags & SYNC_BDFLUSH) {
			if (! XFS_DQ_IS_DIRTY(dqp))
				continue;
		}

		if (nowait) {
			/*
			 * Try to acquire the dquot lock. We are NOT out of
			 * lock order, but we just don't want to wait for this
			 * lock, unless somebody wanted us to.
			 */
			if (! xfs_qm_dqlock_nowait(dqp))
				continue;
		} else {
			xfs_dqlock(dqp);
		}

		/*
		 * Now, find out for sure if this dquot is dirty or not.
		 */
		if (! XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}

		/* XXX a sentinel would be better */
		recl = XFS_QI_MPLRECLAIMS(mp);
		if (! xfs_qm_dqflock_nowait(dqp)) {
			if (nowait) {
				xfs_dqunlock(dqp);
				continue;
			}
			/*
			 * If we can't grab the flush lock then if the caller
			 * really wanted us to give this our best shot,
			 * see if we can give a push to the buffer before we wait
			 * on the flush lock. At this point, we know that
			 * eventhough the dquot is being flushed,
			 * it has (new) dirty data.
			 */
			xfs_qm_dqflock_pushbuf_wait(dqp);
		}
		/*
		 * Let go of the mplist lock. We don't want to hold it
		 * across a disk write
		 */
		flush_flags = (nowait) ? XFS_QMOPT_DELWRI : XFS_QMOPT_SYNC;
		xfs_qm_mplist_unlock(mp);
		xfs_dqtrace_entry(dqp, "XQM_SYNC: DQFLUSH");
		error = xfs_qm_dqflush(dqp, flush_flags);
		xfs_dqunlock(dqp);
		if (error && XFS_FORCED_SHUTDOWN(mp))
			return(0);	/* Need to prevent umount failure */
		else if (error)
			return (error);

		xfs_qm_mplist_lock(mp);
		if (recl != XFS_QI_MPLRECLAIMS(mp)) {
			if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
				break;

			xfs_qm_mplist_unlock(mp);
			goto again;
		}
	}

	xfs_qm_mplist_unlock(mp);
	return (0);
}


/*
 * This initializes all the quota information that's kept in the
 * mount structure
 */
STATIC int
xfs_qm_init_quotainfo(
	xfs_mount_t	*mp)
{
	xfs_quotainfo_t *qinf;
	int		error;
	xfs_dquot_t	*dqp;

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * Tell XQM that we exist as soon as possible.
	 */
	if ((error = xfs_qm_hold_quotafs_ref(mp))) {
		return (error);
	}

	qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);

	/*
	 * See if quotainodes are setup, and if not, allocate them,
	 * and change the superblock accordingly.
	 */
	if ((error = xfs_qm_init_quotainos(mp))) {
		kmem_free(qinf, sizeof(xfs_quotainfo_t));
		mp->m_quotainfo = NULL;
		return (error);
	}

	spinlock_init(&qinf->qi_pinlock, "xfs_qinf_pin");
	xfs_qm_list_init(&qinf->qi_dqlist, "mpdqlist", 0);
	qinf->qi_dqreclaims = 0;

	/* mutex used to serialize quotaoffs */
	mutex_init(&qinf->qi_quotaofflock, MUTEX_DEFAULT, "qoff");

	/* Precalc some constants */
	qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
	ASSERT(qinf->qi_dqchunklen);
	qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
	do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));

	mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);

	/*
	 * We try to get the limits from the superuser's limits fields.
	 * This is quite hacky, but it is standard quota practice.
	 * We look at the USR dquot with id == 0 first, but if user quotas
	 * are not enabled we goto the GRP dquot with id == 0.
	 * We don't really care to keep separate default limits for user
	 * and group quotas, at least not at this point.
	 */
	error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
			     XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER : 
			     (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
				XFS_DQ_PROJ),
			     XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
			     &dqp);
	if (! error) {
		xfs_disk_dquot_t	*ddqp = &dqp->q_core;

		/*
		 * The warnings and timers set the grace period given to
		 * a user or group before he or she can not perform any
		 * more writing. If it is zero, a default is used.
		 */
		qinf->qi_btimelimit = ddqp->d_btimer ?
			be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
		qinf->qi_itimelimit = ddqp->d_itimer ?
			be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
		qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
			be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
		qinf->qi_bwarnlimit = ddqp->d_bwarns ?
			be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
		qinf->qi_iwarnlimit = ddqp->d_iwarns ?
			be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
		qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
			be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
		qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
		qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
		qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
		qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
		qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
		qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 
		/*
		 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
		 * we don't want this dquot cached. We haven't done a
		 * quotacheck yet, and quotacheck doesn't like incore dquots.
		 */
		xfs_qm_dqdestroy(dqp);
	} else {
		qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
		qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
		qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
		qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
		qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
		qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
	}

	return (0);
}


/*
 * Gets called when unmounting a filesystem or when all quotas get
 * turned off.
 * This purges the quota inodes, destroys locks and frees itself.
 */
void
xfs_qm_destroy_quotainfo(
	xfs_mount_t	*mp)
{
	xfs_quotainfo_t *qi;

	qi = mp->m_quotainfo;
	ASSERT(qi != NULL);
	ASSERT(xfs_Gqm != NULL);

	/*
	 * Release the reference that XQM kept, so that we know
	 * when the XQM structure should be freed. We cannot assume
	 * that xfs_Gqm is non-null after this point.
	 */
	xfs_qm_rele_quotafs_ref(mp);

	spinlock_destroy(&qi->qi_pinlock);
	xfs_qm_list_destroy(&qi->qi_dqlist);

	if (qi->qi_uquotaip) {
		XFS_PURGE_INODE(qi->qi_uquotaip);
		qi->qi_uquotaip = NULL; /* paranoia */
	}
	if (qi->qi_gquotaip) {
		XFS_PURGE_INODE(qi->qi_gquotaip);
		qi->qi_gquotaip = NULL;
	}
	mutex_destroy(&qi->qi_quotaofflock);
	kmem_free(qi, sizeof(xfs_quotainfo_t));
	mp->m_quotainfo = NULL;
}



/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */

/* ARGSUSED */
STATIC void
xfs_qm_list_init(
	xfs_dqlist_t	*list,
	char		*str,
	int		n)
{
	mutex_init(&list->qh_lock, MUTEX_DEFAULT, str);
	list->qh_next = NULL;
	list->qh_version = 0;
	list->qh_nelems = 0;
}

STATIC void
xfs_qm_list_destroy(
	xfs_dqlist_t	*list)
{
	mutex_destroy(&(list->qh_lock));
}


/*
 * Stripped down version of dqattach. This doesn't attach, or even look at the
 * dquots attached to the inode. The rationale is that there won't be any
 * attached at the time this is called from quotacheck.
 */
STATIC int
xfs_qm_dqget_noattach(
	xfs_inode_t	*ip,
	xfs_dquot_t	**O_udqpp,
	xfs_dquot_t	**O_gdqpp)
{
	int		error;
	xfs_mount_t	*mp;
	xfs_dquot_t	*udqp, *gdqp;

	ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
	mp = ip->i_mount;
	udqp = NULL;
	gdqp = NULL;

	if (XFS_IS_UQUOTA_ON(mp)) {
		ASSERT(ip->i_udquot == NULL);
		/*
		 * We want the dquot allocated if it doesn't exist.
		 */
		if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
					 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
					 &udqp))) {
			/*
			 * Shouldn't be able to turn off quotas here.
			 */
			ASSERT(error != ESRCH);
			ASSERT(error != ENOENT);
			return (error);
		}
		ASSERT(udqp);
	}

	if (XFS_IS_OQUOTA_ON(mp)) {
		ASSERT(ip->i_gdquot == NULL);
		if (udqp)
			xfs_dqunlock(udqp);
		error = XFS_IS_GQUOTA_ON(mp) ?
				xfs_qm_dqget(mp, ip,
					     ip->i_d.di_gid, XFS_DQ_GROUP,
					     XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
					     &gdqp) :
				xfs_qm_dqget(mp, ip,
					     ip->i_d.di_projid, XFS_DQ_PROJ,
					     XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
					     &gdqp);
		if (error) {
			if (udqp)
				xfs_qm_dqrele(udqp);
			ASSERT(error != ESRCH);
			ASSERT(error != ENOENT);
			return (error);
		}
		ASSERT(gdqp);

		/* Reacquire the locks in the right order */
		if (udqp) {
			if (! xfs_qm_dqlock_nowait(udqp)) {
				xfs_dqunlock(gdqp);
				xfs_dqlock(udqp);
				xfs_dqlock(gdqp);
			}
		}
	}

	*O_udqpp = udqp;
	*O_gdqpp = gdqp;

#ifdef QUOTADEBUG
	if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
	if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
#endif
	return (0);
}

/*
 * Create an inode and return with a reference already taken, but unlocked
 * This is how we create quota inodes
 */
STATIC int
xfs_qm_qino_alloc(
	xfs_mount_t	*mp,
	xfs_inode_t	**ip,
	__int64_t	sbfields,
	uint		flags)
{
	xfs_trans_t	*tp;
	int		error;
	unsigned long s;
	cred_t		zerocr;
	int		committed;

	tp = xfs_trans_alloc(mp,XFS_TRANS_QM_QINOCREATE);
	if ((error = xfs_trans_reserve(tp,
				      XFS_QM_QINOCREATE_SPACE_RES(mp),
				      XFS_CREATE_LOG_RES(mp), 0,
				      XFS_TRANS_PERM_LOG_RES,
				      XFS_CREATE_LOG_COUNT))) {
		xfs_trans_cancel(tp, 0);
		return (error);
	}
	memset(&zerocr, 0, sizeof(zerocr));

	if ((error = xfs_dir_ialloc(&tp, mp->m_rootip, S_IFREG, 1, 0,
				   &zerocr, 0, 1, ip, &committed))) {
		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
				 XFS_TRANS_ABORT);
		return (error);
	}

	/*
	 * Keep an extra reference to this quota inode. This inode is
	 * locked exclusively and joined to the transaction already.
	 */
	ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip));
	VN_HOLD(XFS_ITOV((*ip)));

	/*
	 * Make the changes in the superblock, and log those too.
	 * sbfields arg may contain fields other than *QUOTINO;
	 * VERSIONNUM for example.
	 */
	s = XFS_SB_LOCK(mp);
	if (flags & XFS_QMOPT_SBVERSION) {
#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
		unsigned oldv = mp->m_sb.sb_versionnum;
#endif
		ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp->m_sb));
		ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
				   XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
		       (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
			XFS_SB_GQUOTINO | XFS_SB_QFLAGS));

		XFS_SB_VERSION_ADDQUOTA(&mp->m_sb);
		mp->m_sb.sb_uquotino = NULLFSINO;
		mp->m_sb.sb_gquotino = NULLFSINO;

		/* qflags will get updated _after_ quotacheck */
		mp->m_sb.sb_qflags = 0;
#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
		cmn_err(CE_NOTE,
			"Old superblock version %x, converting to %x.",
			oldv, mp->m_sb.sb_versionnum);
#endif
	}
	if (flags & XFS_QMOPT_UQUOTA)
		mp->m_sb.sb_uquotino = (*ip)->i_ino;
	else
		mp->m_sb.sb_gquotino = (*ip)->i_ino;
	XFS_SB_UNLOCK(mp, s);
	xfs_mod_sb(tp, sbfields);

	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
				     NULL))) {
		xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
		return (error);
	}
	return (0);
}


STATIC int
xfs_qm_reset_dqcounts(
	xfs_mount_t	*mp,
	xfs_buf_t	*bp,
	xfs_dqid_t	id,
	uint		type)
{
	xfs_disk_dquot_t	*ddq;
	int			j;

	xfs_buftrace("RESET DQUOTS", bp);
	/*
	 * Reset all counters and timers. They'll be
	 * started afresh by xfs_qm_quotacheck.
	 */
#ifdef DEBUG
	j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
	do_div(j, sizeof(xfs_dqblk_t));
	ASSERT(XFS_QM_DQPERBLK(mp) == j);
#endif
	ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
	for (j = 0; j < XFS_QM_DQPERBLK(mp); j++) {
		/*
		 * Do a sanity check, and if needed, repair the dqblk. Don't
		 * output any warnings because it's perfectly possible to
		 * find unitialized dquot blks. See comment in xfs_qm_dqcheck.
		 */
		(void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
				      "xfs_quotacheck");
		ddq->d_bcount = 0;
		ddq->d_icount = 0;
		ddq->d_rtbcount = 0;
		ddq->d_btimer = 0;
		ddq->d_itimer = 0;
		ddq->d_rtbtimer = 0;
		ddq->d_bwarns = 0;
		ddq->d_iwarns = 0;
		ddq->d_rtbwarns = 0;
		ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
	}

	return (0);
}

STATIC int
xfs_qm_dqiter_bufs(
	xfs_mount_t	*mp,
	xfs_dqid_t	firstid,
	xfs_fsblock_t	bno,
	xfs_filblks_t	blkcnt,
	uint		flags)
{
	xfs_buf_t	*bp;
	int		error;
	int		notcommitted;
	int		incr;
	int		type;

	ASSERT(blkcnt > 0);
	notcommitted = 0;
	incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
		XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
	type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
		(flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
	error = 0;

	/*
	 * Blkcnt arg can be a very big number, and might even be
	 * larger than the log itself. So, we have to break it up into
	 * manageable-sized transactions.
	 * Note that we don't start a permanent transaction here; we might
	 * not be able to get a log reservation for the whole thing up front,
	 * and we don't really care to either, because we just discard
	 * everything if we were to crash in the middle of this loop.
	 */
	while (blkcnt--) {
		error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
			      XFS_FSB_TO_DADDR(mp, bno),
			      (int)XFS_QI_DQCHUNKLEN(mp), 0, &bp);
		if (error)
			break;

		(void) xfs_qm_reset_dqcounts(mp, bp, firstid, type);
		xfs_bdwrite(mp, bp);
		/*
		 * goto the next block.
		 */
		bno++;
		firstid += XFS_QM_DQPERBLK(mp);
	}