xfs_qm.c

linux-2.6.15.6

/*
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_bit.h"
#include "xfs_log.h"
#include "xfs_inum.h"
#include "xfs_clnt.h"
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_dir.h"
#include "xfs_dir2.h"
#include "xfs_alloc.h"
#include "xfs_dmapi.h"
#include "xfs_quota.h"
#include "xfs_mount.h"
#include "xfs_bmap_btree.h"
#include "xfs_alloc_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_dir_sf.h"
#include "xfs_dir2_sf.h"
#include "xfs_attr_sf.h"
#include "xfs_dinode.h"
#include "xfs_inode.h"
#include "xfs_btree.h"
#include "xfs_ialloc.h"
#include "xfs_itable.h"
#include "xfs_rtalloc.h"
#include "xfs_error.h"
#include "xfs_bmap.h"
#include "xfs_rw.h"
#include "xfs_acl.h"
#include "xfs_cap.h"
#include "xfs_mac.h"
#include "xfs_attr.h"
#include "xfs_buf_item.h"
#include "xfs_trans_space.h"
#include "xfs_utils.h"
#include "xfs_qm.h"

/*
 * The global quota manager. There is only one of these for the entire
 * system, _not_ one per file system. XQM keeps track of the overall
 * quota functionality, including maintaining the freelist and hash
 * tables of dquots.
 */
mutex_t		xfs_Gqm_lock;
struct xfs_qm	*xfs_Gqm;
uint		ndquot;

kmem_zone_t	*qm_dqzone;
kmem_zone_t	*qm_dqtrxzone;
STATIC kmem_shaker_t	xfs_qm_shaker;

STATIC void	xfs_qm_list_init(xfs_dqlist_t *, char *, int);
STATIC void	xfs_qm_list_destroy(xfs_dqlist_t *);

STATIC void	xfs_qm_freelist_init(xfs_frlist_t *);
STATIC void	xfs_qm_freelist_destroy(xfs_frlist_t *);
STATIC int	xfs_qm_mplist_nowait(xfs_mount_t *);
STATIC int	xfs_qm_dqhashlock_nowait(xfs_dquot_t *);

STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
STATIC int	xfs_qm_init_quotainfo(xfs_mount_t *);
STATIC int	xfs_qm_shake(int, gfp_t);

#ifdef DEBUG
extern mutex_t	qcheck_lock;
#endif

#ifdef QUOTADEBUG
#define XQM_LIST_PRINT(l, NXT, title) \
{ \
	xfs_dquot_t	*dqp; int i = 0; \
	cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
	for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
		cmn_err(CE_DEBUG, "   %d.  \"%d (%s)\"   " \
				  "bcnt = %d, icnt = %d, refs = %d", \
			++i, (int) be32_to_cpu(dqp->q_core.d_id), \
			DQFLAGTO_TYPESTR(dqp),	     \
			(int) be64_to_cpu(dqp->q_core.d_bcount), \
			(int) be64_to_cpu(dqp->q_core.d_icount), \
			(int) dqp->q_nrefs);  } \
}
#else
#define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
#endif

/*
 * Initialize the XQM structure.
 * Note that there is not one quota manager per file system.
 */
STATIC struct xfs_qm *
xfs_Gqm_init(void)
{
	xfs_dqhash_t	*udqhash, *gdqhash;
	xfs_qm_t	*xqm;
	uint		i, hsize, flags = KM_SLEEP | KM_MAYFAIL;

	/*
	 * Initialize the dquot hash tables.
	 */
	hsize = XFS_QM_HASHSIZE_HIGH;
	while (!(udqhash = kmem_zalloc(hsize * sizeof(xfs_dqhash_t), flags))) {
		if ((hsize >>= 1) <= XFS_QM_HASHSIZE_LOW)
			flags = KM_SLEEP;
	}
	gdqhash = kmem_zalloc(hsize * sizeof(xfs_dqhash_t), KM_SLEEP);
	ndquot = hsize << 8;

	xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
	xqm->qm_dqhashmask = hsize - 1;
	xqm->qm_usr_dqhtable = udqhash;
	xqm->qm_grp_dqhtable = gdqhash;
	ASSERT(xqm->qm_usr_dqhtable != NULL);
	ASSERT(xqm->qm_grp_dqhtable != NULL);

	for (i = 0; i < hsize; i++) {
		xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
		xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
	}

	/*
	 * Freelist of all dquots of all file systems
	 */
	xfs_qm_freelist_init(&(xqm->qm_dqfreelist));

	/*
	 * dquot zone. we register our own low-memory callback.
	 */
	if (!qm_dqzone) {
		xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
						"xfs_dquots");
		qm_dqzone = xqm->qm_dqzone;
	} else
		xqm->qm_dqzone = qm_dqzone;

	xfs_qm_shaker = kmem_shake_register(xfs_qm_shake);

	/*
	 * The t_dqinfo portion of transactions.
	 */
	if (!qm_dqtrxzone) {
		xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
						   "xfs_dqtrx");
		qm_dqtrxzone = xqm->qm_dqtrxzone;
	} else
		xqm->qm_dqtrxzone = qm_dqtrxzone;

	atomic_set(&xqm->qm_totaldquots, 0);
	xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
	xqm->qm_nrefs = 0;
#ifdef DEBUG
	mutex_init(&qcheck_lock, MUTEX_DEFAULT, "qchk");
#endif
	return xqm;
}

/*
 * Destroy the global quota manager when its reference count goes to zero.
 */
STATIC void
xfs_qm_destroy(
	struct xfs_qm	*xqm)
{
	int		hsize, i;

	ASSERT(xqm != NULL);
	ASSERT(xqm->qm_nrefs == 0);
	kmem_shake_deregister(xfs_qm_shaker);
	hsize = xqm->qm_dqhashmask + 1;
	for (i = 0; i < hsize; i++) {
		xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
		xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
	}
	kmem_free(xqm->qm_usr_dqhtable, hsize * sizeof(xfs_dqhash_t));
	kmem_free(xqm->qm_grp_dqhtable, hsize * sizeof(xfs_dqhash_t));
	xqm->qm_usr_dqhtable = NULL;
	xqm->qm_grp_dqhtable = NULL;
	xqm->qm_dqhashmask = 0;
	xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
#ifdef DEBUG
	mutex_destroy(&qcheck_lock);
#endif
	kmem_free(xqm, sizeof(xfs_qm_t));
}

/*
 * Called at mount time to let XQM know that another file system is
 * starting quotas. This isn't crucial information as the individual mount
 * structures are pretty independent, but it helps the XQM keep a
 * global view of what's going on.
 */
/* ARGSUSED */
STATIC int
xfs_qm_hold_quotafs_ref(
	struct xfs_mount *mp)
{
	/*
	 * Need to lock the xfs_Gqm structure for things like this. For example,
	 * the structure could disappear between the entry to this routine and
	 * a HOLD operation if not locked.
	 */
	XFS_QM_LOCK(xfs_Gqm);

	if (xfs_Gqm == NULL)
		xfs_Gqm = xfs_Gqm_init();
	/*
	 * We can keep a list of all filesystems with quotas mounted for
	 * debugging and statistical purposes, but ...
	 * Just take a reference and get out.
	 */
	XFS_QM_HOLD(xfs_Gqm);
	XFS_QM_UNLOCK(xfs_Gqm);

	return 0;
}


/*
 * Release the reference that a filesystem took at mount time,
 * so that we know when we need to destroy the entire quota manager.
 */
/* ARGSUSED */
STATIC void
xfs_qm_rele_quotafs_ref(
	struct xfs_mount *mp)
{
	xfs_dquot_t	*dqp, *nextdqp;

	ASSERT(xfs_Gqm);
	ASSERT(xfs_Gqm->qm_nrefs > 0);

	/*
	 * Go thru the freelist and destroy all inactive dquots.
	 */
	xfs_qm_freelist_lock(xfs_Gqm);

	for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
	     dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
		xfs_dqlock(dqp);
		nextdqp = dqp->dq_flnext;
		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);
			xfs_qm_dqdestroy(dqp);
		} else {
			xfs_dqunlock(dqp);
		}
		dqp = nextdqp;
	}
	xfs_qm_freelist_unlock(xfs_Gqm);

	/*
	 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
	 * be restarted.
	 */
	XFS_QM_LOCK(xfs_Gqm);
	XFS_QM_RELE(xfs_Gqm);
	if (xfs_Gqm->qm_nrefs == 0) {
		xfs_qm_destroy(xfs_Gqm);
		xfs_Gqm = NULL;
	}
	XFS_QM_UNLOCK(xfs_Gqm);
}

/*
 * This is called at mount time from xfs_mountfs to initialize the quotainfo
 * structure and start the global quotamanager (xfs_Gqm) if it hasn't done
 * so already.	Note that the superblock has not been read in yet.
 */
void
xfs_qm_mount_quotainit(
	xfs_mount_t	*mp,
	uint		flags)
{
	/*
	 * User, projects or group quotas has to be on.
	 */
	ASSERT(flags & (XFSMNT_UQUOTA | XFSMNT_PQUOTA | XFSMNT_GQUOTA));

	/*
	 * Initialize the flags in the mount structure. From this point
	 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
	 * Note that we enforce nothing if accounting is off.
	 * ie.	XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
	 * It isn't necessary to take the quotaoff lock to do this; this is
	 * called from mount.
	 */
	if (flags & XFSMNT_UQUOTA) {
		mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
		if (flags & XFSMNT_UQUOTAENF)
			mp->m_qflags |= XFS_UQUOTA_ENFD;
	}
	if (flags & XFSMNT_GQUOTA) {
		mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
		if (flags & XFSMNT_GQUOTAENF)
			mp->m_qflags |= XFS_OQUOTA_ENFD;
	} else if (flags & XFSMNT_PQUOTA) {
		mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
		if (flags & XFSMNT_PQUOTAENF)
			mp->m_qflags |= XFS_OQUOTA_ENFD;
	}
}

/*
 * Just destroy the quotainfo structure.
 */
void
xfs_qm_unmount_quotadestroy(
	xfs_mount_t	*mp)
{
	if (mp->m_quotainfo)
		xfs_qm_destroy_quotainfo(mp);
}


/*
 * This is called from xfs_mountfs to start quotas and initialize all
 * necessary data structures like quotainfo.  This is also responsible for
 * running a quotacheck as necessary.  We are guaranteed that the superblock
 * is consistently read in at this point.
 */
int
xfs_qm_mount_quotas(
	xfs_mount_t	*mp,
	int		mfsi_flags)
{
	unsigned long	s;
	int		error = 0;
	uint		sbf;


	/*
	 * If quotas on realtime volumes is not supported, we disable
	 * quotas immediately.
	 */
	if (mp->m_sb.sb_rextents) {
		cmn_err(CE_NOTE,
			"Cannot turn on quotas for realtime filesystem %s",
			mp->m_fsname);
		mp->m_qflags = 0;
		goto write_changes;
	}

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * Allocate the quotainfo structure inside the mount struct, and
	 * create quotainode(s), and change/rev superblock if necessary.
	 */
	if ((error = xfs_qm_init_quotainfo(mp))) {
		/*
		 * We must turn off quotas.
		 */
		ASSERT(mp->m_quotainfo == NULL);
		mp->m_qflags = 0;
		goto write_changes;
	}
	/*
	 * If any of the quotas are not consistent, do a quotacheck.
	 */
	if (XFS_QM_NEED_QUOTACHECK(mp) &&
		!(mfsi_flags & XFS_MFSI_NO_QUOTACHECK)) {
		if ((error = xfs_qm_quotacheck(mp))) {
			/* Quotacheck has failed and quotas have
			 * been disabled.
			 */
			return XFS_ERROR(error);
		}
	}

 write_changes:
	/*
	 * We actually don't have to acquire the SB_LOCK at all.
	 * This can only be called from mount, and that's single threaded. XXX
	 */
	s = XFS_SB_LOCK(mp);
	sbf = mp->m_sb.sb_qflags;
	mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
	XFS_SB_UNLOCK(mp, s);

	if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
		if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
			/*
			 * We could only have been turning quotas off.
			 * We aren't in very good shape actually because
			 * the incore structures are convinced that quotas are
			 * off, but the on disk superblock doesn't know that !
			 */
			ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
			xfs_fs_cmn_err(CE_ALERT, mp,
				"XFS mount_quotas: Superblock update failed!");
		}
	}

	if (error) {
		xfs_fs_cmn_err(CE_WARN, mp,
			"Failed to initialize disk quotas.");
	}
	return XFS_ERROR(error);
}

/*
 * Called from the vfsops layer.
 */
int
xfs_qm_unmount_quotas(
	xfs_mount_t	*mp)
{
	xfs_inode_t	*uqp, *gqp;
	int		error = 0;

	/*
	 * Release the dquots that root inode, et al might be holding,
	 * before we flush quotas and blow away the quotainfo structure.
	 */
	ASSERT(mp->m_rootip);
	xfs_qm_dqdetach(mp->m_rootip);
	if (mp->m_rbmip)
		xfs_qm_dqdetach(mp->m_rbmip);
	if (mp->m_rsumip)
		xfs_qm_dqdetach(mp->m_rsumip);

	/*
	 * Flush out the quota inodes.
	 */
	uqp = gqp = NULL;
	if (mp->m_quotainfo) {
		if ((uqp = mp->m_quotainfo->qi_uquotaip) != NULL) {
			xfs_ilock(uqp, XFS_ILOCK_EXCL);
			xfs_iflock(uqp);
			error = xfs_iflush(uqp, XFS_IFLUSH_SYNC);
			xfs_iunlock(uqp, XFS_ILOCK_EXCL);
			if (unlikely(error == EFSCORRUPTED)) {
				XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
						 XFS_ERRLEVEL_LOW, mp);
				goto out;
			}
		}
		if ((gqp = mp->m_quotainfo->qi_gquotaip) != NULL) {
			xfs_ilock(gqp, XFS_ILOCK_EXCL);
			xfs_iflock(gqp);
			error = xfs_iflush(gqp, XFS_IFLUSH_SYNC);
			xfs_iunlock(gqp, XFS_ILOCK_EXCL);
			if (unlikely(error == EFSCORRUPTED)) {
				XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
						 XFS_ERRLEVEL_LOW, mp);
				goto out;
			}
		}
	}
	if (uqp) {
		 XFS_PURGE_INODE(uqp);
		 mp->m_quotainfo->qi_uquotaip = NULL;
	}
	if (gqp) {
		XFS_PURGE_INODE(gqp);
		mp->m_quotainfo->qi_gquotaip = NULL;
	}
out:
	return XFS_ERROR(error);
}

/*
 * Flush all dquots of the given file system to disk. The dquots are
 * _not_ purged from memory here, just their data written to disk.
 */
STATIC int
xfs_qm_dqflush_all(
	xfs_mount_t	*mp,
	int		flags)
{
	int		recl;
	xfs_dquot_t	*dqp;
	int		niters;
	int		error;

	if (mp->m_quotainfo == NULL)
		return (0);
	niters = 0;
again:
	xfs_qm_mplist_lock(mp);
	FOREACH_DQUOT_IN_MP(dqp, mp) {
		xfs_dqlock(dqp);
		if (! XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}
		xfs_dqtrace_entry(dqp, "FLUSHALL: DQDIRTY");
		/* XXX a sentinel would be better */
		recl = XFS_QI_MPLRECLAIMS(mp);
		if (! xfs_qm_dqflock_nowait(dqp)) {
			/*
			 * If we can't grab the flush lock then check
			 * to see if the dquot has been flushed delayed
			 * write.  If so, grab its buffer and send it
			 * out immediately.  We'll be able to acquire
			 * the flush lock when the I/O completes.