summaryrefslogtreecommitdiff
path: root/fs/xfs/xfs_qm.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs/xfs_qm.c')
-rw-r--r--fs/xfs/xfs_qm.c2416
1 files changed, 2416 insertions, 0 deletions
diff --git a/fs/xfs/xfs_qm.c b/fs/xfs/xfs_qm.c
new file mode 100644
index 000000000000..9a0aa76facdf
--- /dev/null
+++ b/fs/xfs/xfs_qm.c
@@ -0,0 +1,2416 @@
+/*
+ * 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_trans.h"
+#include "xfs_sb.h"
+#include "xfs_ag.h"
+#include "xfs_alloc.h"
+#include "xfs_quota.h"
+#include "xfs_mount.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_ialloc_btree.h"
+#include "xfs_dinode.h"
+#include "xfs_inode.h"
+#include "xfs_ialloc.h"
+#include "xfs_itable.h"
+#include "xfs_rtalloc.h"
+#include "xfs_error.h"
+#include "xfs_bmap.h"
+#include "xfs_attr.h"
+#include "xfs_buf_item.h"
+#include "xfs_trans_space.h"
+#include "xfs_utils.h"
+#include "xfs_qm.h"
+#include "xfs_trace.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.
+ */
+struct mutex xfs_Gqm_lock;
+struct xfs_qm *xfs_Gqm;
+uint ndquot;
+
+kmem_zone_t *qm_dqzone;
+kmem_zone_t *qm_dqtrxzone;
+
+STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
+STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
+
+STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
+STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
+STATIC int xfs_qm_shake(struct shrinker *, struct shrink_control *);
+
+static struct shrinker xfs_qm_shaker = {
+ .shrink = xfs_qm_shake,
+ .seeks = DEFAULT_SEEKS,
+};
+
+/*
+ * 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;
+ size_t hsize;
+ uint i;
+
+ /*
+ * Initialize the dquot hash tables.
+ */
+ udqhash = kmem_zalloc_greedy(&hsize,
+ XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
+ XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
+ if (!udqhash)
+ goto out;
+
+ gdqhash = kmem_zalloc_large(hsize);
+ if (!gdqhash)
+ goto out_free_udqhash;
+
+ hsize /= sizeof(xfs_dqhash_t);
+ 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
+ */
+ INIT_LIST_HEAD(&xqm->qm_dqfrlist);
+ xqm->qm_dqfrlist_cnt = 0;
+ mutex_init(&xqm->qm_dqfrlist_lock);
+
+ /*
+ * 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;
+
+ register_shrinker(&xfs_qm_shaker);
+
+ /*
+ * 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;
+ return xqm;
+
+ out_free_udqhash:
+ kmem_free_large(udqhash);
+ out:
+ return NULL;
+}
+
+/*
+ * Destroy the global quota manager when its reference count goes to zero.
+ */
+STATIC void
+xfs_qm_destroy(
+ struct xfs_qm *xqm)
+{
+ struct xfs_dquot *dqp, *n;
+ int hsize, i;
+
+ ASSERT(xqm != NULL);
+ ASSERT(xqm->qm_nrefs == 0);
+ unregister_shrinker(&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_large(xqm->qm_usr_dqhtable);
+ kmem_free_large(xqm->qm_grp_dqhtable);
+ xqm->qm_usr_dqhtable = NULL;
+ xqm->qm_grp_dqhtable = NULL;
+ xqm->qm_dqhashmask = 0;
+
+ /* frlist cleanup */
+ mutex_lock(&xqm->qm_dqfrlist_lock);
+ list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
+ xfs_dqlock(dqp);
+ list_del_init(&dqp->q_freelist);
+ xfs_Gqm->qm_dqfrlist_cnt--;
+ xfs_dqunlock(dqp);
+ xfs_qm_dqdestroy(dqp);
+ }
+ mutex_unlock(&xqm->qm_dqfrlist_lock);
+ mutex_destroy(&xqm->qm_dqfrlist_lock);
+ kmem_free(xqm);
+}
+
+/*
+ * 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.
+ */
+ mutex_lock(&xfs_Gqm_lock);
+
+ if (!xfs_Gqm) {
+ xfs_Gqm = xfs_Gqm_init();
+ if (!xfs_Gqm) {
+ mutex_unlock(&xfs_Gqm_lock);
+ return ENOMEM;
+ }
+ }
+
+ /*
+ * 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_Gqm->qm_nrefs++;
+ mutex_unlock(&xfs_Gqm_lock);
+
+ 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, *n;
+
+ ASSERT(xfs_Gqm);
+ ASSERT(xfs_Gqm->qm_nrefs > 0);
+
+ /*
+ * Go thru the freelist and destroy all inactive dquots.
+ */
+ mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
+
+ list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
+ xfs_dqlock(dqp);
+ if (dqp->dq_flags & XFS_DQ_INACTIVE) {
+ ASSERT(dqp->q_mount == NULL);
+ ASSERT(! XFS_DQ_IS_DIRTY(dqp));
+ ASSERT(list_empty(&dqp->q_hashlist));
+ ASSERT(list_empty(&dqp->q_mplist));
+ list_del_init(&dqp->q_freelist);
+ xfs_Gqm->qm_dqfrlist_cnt--;
+ xfs_dqunlock(dqp);
+ xfs_qm_dqdestroy(dqp);
+ } else {
+ xfs_dqunlock(dqp);
+ }
+ }
+ mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
+
+ /*
+ * Destroy the entire XQM. If somebody mounts with quotaon, this'll
+ * be restarted.
+ */
+ mutex_lock(&xfs_Gqm_lock);
+ if (--xfs_Gqm->qm_nrefs == 0) {
+ xfs_qm_destroy(xfs_Gqm);
+ xfs_Gqm = NULL;
+ }
+ mutex_unlock(&xfs_Gqm_lock);
+}
+
+/*
+ * Just destroy the quotainfo structure.
+ */
+void
+xfs_qm_unmount(
+ struct xfs_mount *mp)
+{
+ if (mp->m_quotainfo) {
+ xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
+ 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.
+ *
+ * If we fail here, the mount will continue with quota turned off. We don't
+ * need to inidicate success or failure at all.
+ */
+void
+xfs_qm_mount_quotas(
+ xfs_mount_t *mp)
+{
+ int error = 0;
+ uint sbf;
+
+ /*
+ * If quotas on realtime volumes is not supported, we disable
+ * quotas immediately.
+ */
+ if (mp->m_sb.sb_rextents) {
+ xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
+ 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.
+ */
+ error = xfs_qm_init_quotainfo(mp);
+ if (error) {
+ /*
+ * 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)) {
+ error = xfs_qm_quotacheck(mp);
+ if (error) {
+ /* Quotacheck failed and disabled quotas. */
+ return;
+ }
+ }
+ /*
+ * If one type of quotas is off, then it will lose its
+ * quotachecked status, since we won't be doing accounting for
+ * that type anymore.
+ */
+ if (!XFS_IS_UQUOTA_ON(mp))
+ mp->m_qflags &= ~XFS_UQUOTA_CHKD;
+ if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
+ mp->m_qflags &= ~XFS_OQUOTA_CHKD;
+
+ write_changes:
+ /*
+ * We actually don't have to acquire the m_sb_lock at all.
+ * This can only be called from mount, and that's single threaded. XXX
+ */
+ spin_lock(&mp->m_sb_lock);
+ sbf = mp->m_sb.sb_qflags;
+ mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
+ spin_unlock(&mp->m_sb_lock);
+
+ 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_alert(mp, "%s: Superblock update failed!",
+ __func__);
+ }
+ }
+
+ if (error) {
+ xfs_warn(mp, "Failed to initialize disk quotas.");
+ return;
+ }
+}
+
+/*
+ * Called from the vfsops layer.
+ */
+void
+xfs_qm_unmount_quotas(
+ xfs_mount_t *mp)
+{
+ /*
+ * 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);
+
+ /*
+ * Release the quota inodes.
+ */
+ if (mp->m_quotainfo) {
+ if (mp->m_quotainfo->qi_uquotaip) {
+ IRELE(mp->m_quotainfo->qi_uquotaip);
+ mp->m_quotainfo->qi_uquotaip = NULL;
+ }
+ if (mp->m_quotainfo->qi_gquotaip) {
+ IRELE(mp->m_quotainfo->qi_gquotaip);
+ mp->m_quotainfo->qi_gquotaip = NULL;
+ }
+ }
+}
+
+/*
+ * 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(
+ struct xfs_mount *mp,
+ int sync_mode)
+{
+ struct xfs_quotainfo *q = mp->m_quotainfo;
+ int recl;
+ struct xfs_dquot *dqp;
+ int error;
+
+ if (!q)
+ return 0;
+again:
+ mutex_lock(&q->qi_dqlist_lock);
+ list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
+ xfs_dqlock(dqp);
+ if (! XFS_DQ_IS_DIRTY(dqp)) {
+ xfs_dqunlock(dqp);
+ continue;
+ }
+
+ /* XXX a sentinel would be better */
+ recl = q->qi_dqreclaims;
+ if (!xfs_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.
+ */
+ xfs_qm_dqflock_pushbuf_wait(dqp);
+ }
+ /*
+ * Let go of the mplist lock. We don't want to hold it
+ * across a disk write.
+ */
+ mutex_unlock(&q->qi_dqlist_lock);
+ error = xfs_qm_dqflush(dqp, sync_mode);
+ xfs_dqunlock(dqp);
+ if (error)
+ return error;
+
+ mutex_lock(&q->qi_dqlist_lock);
+ if (recl != q->qi_dqreclaims) {
+ mutex_unlock(&q->qi_dqlist_lock);
+ /* XXX restart limit */
+ goto again;
+ }
+ }
+
+ mutex_unlock(&q->qi_dqlist_lock);
+ /* return ! busy */
+ return 0;
+}
+/*
+ * Release the group dquot pointers the user dquots may be
+ * carrying around as a hint. mplist is locked on entry and exit.
+ */
+STATIC void
+xfs_qm_detach_gdquots(
+ struct xfs_mount *mp)
+{
+ struct xfs_quotainfo *q = mp->m_quotainfo;
+ struct xfs_dquot *dqp, *gdqp;
+ int nrecl;
+
+ again:
+ ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
+ list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
+ xfs_dqlock(dqp);
+ if ((gdqp = dqp->q_gdquot)) {
+ xfs_dqlock(gdqp);
+ dqp->q_gdquot = NULL;
+ }
+ xfs_dqunlock(dqp);
+
+ if (gdqp) {
+ /*
+ * Can't hold the mplist lock across a dqput.
+ * XXXmust convert to marker based iterations here.
+ */
+ nrecl = q->qi_dqreclaims;
+ mutex_unlock(&q->qi_dqlist_lock);
+ xfs_qm_dqput(gdqp);
+
+ mutex_lock(&q->qi_dqlist_lock);
+ if (nrecl != q->qi_dqreclaims)
+ goto again;
+ }
+ }
+}
+
+/*
+ * Go through all the incore dquots of this file system and take them
+ * off the mplist and hashlist, if the dquot type matches the dqtype
+ * parameter. This is used when turning off quota accounting for
+ * users and/or groups, as well as when the filesystem is unmounting.
+ */
+STATIC int
+xfs_qm_dqpurge_int(
+ struct xfs_mount *mp,
+ uint flags)
+{
+ struct xfs_quotainfo *q = mp->m_quotainfo;
+ struct xfs_dquot *dqp, *n;
+ uint dqtype;
+ int nrecl;
+ int nmisses;
+
+ if (!q)
+ return 0;
+
+ dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
+ dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
+ dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
+
+ mutex_lock(&q->qi_dqlist_lock);
+
+ /*
+ * In the first pass through all incore dquots of this filesystem,
+ * we release the group dquot pointers the user dquots may be
+ * carrying around as a hint. We need to do this irrespective of
+ * what's being turned off.
+ */
+ xfs_qm_detach_gdquots(mp);
+
+ again:
+ nmisses = 0;
+ ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
+ /*
+ * Try to get rid of all of the unwanted dquots. The idea is to
+ * get them off mplist and hashlist, but leave them on freelist.
+ */
+ list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
+ /*
+ * It's OK to look at the type without taking dqlock here.
+ * We're holding the mplist lock here, and that's needed for
+ * a dqreclaim.
+ */
+ if ((dqp->dq_flags & dqtype) == 0)
+ continue;
+
+ if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
+ nrecl = q->qi_dqreclaims;
+ mutex_unlock(&q->qi_dqlist_lock);
+ mutex_lock(&dqp->q_hash->qh_lock);
+ mutex_lock(&q->qi_dqlist_lock);
+
+ /*
+ * XXXTheoretically, we can get into a very long
+ * ping pong game here.
+ * No one can be adding dquots to the mplist at
+ * this point, but somebody might be taking things off.
+ */
+ if (nrecl != q->qi_dqreclaims) {
+ mutex_unlock(&dqp->q_hash->qh_lock);
+ goto again;
+ }
+ }
+
+ /*
+ * Take the dquot off the mplist and hashlist. It may remain on
+ * freelist in INACTIVE state.
+ */
+ nmisses += xfs_qm_dqpurge(dqp);
+ }
+ mutex_unlock(&q->qi_dqlist_lock);
+ return nmisses;
+}
+
+int
+xfs_qm_dqpurge_all(
+ xfs_mount_t *mp,
+ uint flags)
+{
+ int ndquots;
+
+ /*
+ * Purge the dquot cache.
+ * None of the dquots should really be busy at this point.
+ */
+ if (mp->m_quotainfo) {
+ while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
+ delay(ndquots * 10);
+ }
+ }
+ return 0;
+}
+
+STATIC int
+xfs_qm_dqattach_one(
+ xfs_inode_t *ip,
+ xfs_dqid_t id,
+ uint type,
+ uint doalloc,
+ xfs_dquot_t *udqhint, /* hint */
+ xfs_dquot_t **IO_idqpp)
+{
+ xfs_dquot_t *dqp;
+ int error;
+
+ ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+ error = 0;
+
+ /*
+ * See if we already have it in the inode itself. IO_idqpp is
+ * &i_udquot or &i_gdquot. This made the code look weird, but
+ * made the logic a lot simpler.
+ */
+ dqp = *IO_idqpp;
+ if (dqp) {
+ trace_xfs_dqattach_found(dqp);
+ return 0;
+ }
+
+ /*
+ * udqhint is the i_udquot field in inode, and is non-NULL only
+ * when the type arg is group/project. Its purpose is to save a
+ * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
+ * the user dquot.
+ */
+ if (udqhint) {
+ ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
+ xfs_dqlock(udqhint);
+
+ /*
+ * No need to take dqlock to look at the id.
+ *
+ * The ID can't change until it gets reclaimed, and it won't
+ * be reclaimed as long as we have a ref from inode and we
+ * hold the ilock.
+ */
+ dqp = udqhint->q_gdquot;
+ if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
+ xfs_dqlock(dqp);
+ XFS_DQHOLD(dqp);
+ ASSERT(*IO_idqpp == NULL);
+ *IO_idqpp = dqp;
+
+ xfs_dqunlock(dqp);
+ xfs_dqunlock(udqhint);
+ return 0;
+ }
+
+ /*
+ * We can't hold a dquot lock when we call the dqget code.
+ * We'll deadlock in no time, because of (not conforming to)
+ * lock ordering - the inodelock comes before any dquot lock,
+ * and we may drop and reacquire the ilock in xfs_qm_dqget().
+ */
+ xfs_dqunlock(udqhint);
+ }
+
+ /*
+ * Find the dquot from somewhere. This bumps the
+ * reference count of dquot and returns it locked.
+ * This can return ENOENT if dquot didn't exist on
+ * disk and we didn't ask it to allocate;
+ * ESRCH if quotas got turned off suddenly.
+ */
+ error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
+ if (error)
+ return error;
+
+ trace_xfs_dqattach_get(dqp);
+
+ /*
+ * dqget may have dropped and re-acquired the ilock, but it guarantees
+ * that the dquot returned is the one that should go in the inode.
+ */
+ *IO_idqpp = dqp;
+ xfs_dqunlock(dqp);
+ return 0;
+}
+
+
+/*
+ * Given a udquot and gdquot, attach a ptr to the group dquot in the
+ * udquot as a hint for future lookups. The idea sounds simple, but the
+ * execution isn't, because the udquot might have a group dquot attached
+ * already and getting rid of that gets us into lock ordering constraints.
+ * The process is complicated more by the fact that the dquots may or may not
+ * be locked on entry.
+ */
+STATIC void
+xfs_qm_dqattach_grouphint(
+ xfs_dquot_t *udq,
+ xfs_dquot_t *gdq)
+{
+ xfs_dquot_t *tmp;
+
+ xfs_dqlock(udq);
+
+ if ((tmp = udq->q_gdquot)) {
+ if (tmp == gdq) {
+ xfs_dqunlock(udq);
+ return;
+ }
+
+ udq->q_gdquot = NULL;
+ /*
+ * We can't keep any dqlocks when calling dqrele,
+ * because the freelist lock comes before dqlocks.
+ */
+ xfs_dqunlock(udq);
+ /*
+ * we took a hard reference once upon a time in dqget,
+ * so give it back when the udquot no longer points at it
+ * dqput() does the unlocking of the dquot.
+ */
+ xfs_qm_dqrele(tmp);
+
+ xfs_dqlock(udq);
+ xfs_dqlock(gdq);
+
+ } else {
+ ASSERT(XFS_DQ_IS_LOCKED(udq));
+ xfs_dqlock(gdq);
+ }
+
+ ASSERT(XFS_DQ_IS_LOCKED(udq));
+ ASSERT(XFS_DQ_IS_LOCKED(gdq));
+ /*
+ * Somebody could have attached a gdquot here,
+ * when we dropped the uqlock. If so, just do nothing.
+ */
+ if (udq->q_gdquot == NULL) {
+ XFS_DQHOLD(gdq);
+ udq->q_gdquot = gdq;
+ }
+
+ xfs_dqunlock(gdq);
+ xfs_dqunlock(udq);
+}
+
+
+/*
+ * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
+ * into account.
+ * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
+ * Inode may get unlocked and relocked in here, and the caller must deal with
+ * the consequences.
+ */
+int
+xfs_qm_dqattach_locked(
+ xfs_inode_t *ip,
+ uint flags)
+{
+ xfs_mount_t *mp = ip->i_mount;
+ uint nquotas = 0;
+ int error = 0;
+
+ if (!XFS_IS_QUOTA_RUNNING(mp) ||
+ !XFS_IS_QUOTA_ON(mp) ||
+ !XFS_NOT_DQATTACHED(mp, ip) ||
+ ip->i_ino == mp->m_sb.sb_uquotino ||
+ ip->i_ino == mp->m_sb.sb_gquotino)
+ return 0;
+
+ ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+
+ if (XFS_IS_UQUOTA_ON(mp)) {
+ error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
+ flags & XFS_QMOPT_DQALLOC,
+ NULL, &ip->i_udquot);
+ if (error)
+ goto done;
+ nquotas++;
+ }
+
+ ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+ if (XFS_IS_OQUOTA_ON(mp)) {
+ error = XFS_IS_GQUOTA_ON(mp) ?
+ xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
+ flags & XFS_QMOPT_DQALLOC,
+ ip->i_udquot, &ip->i_gdquot) :
+ xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
+ flags & XFS_QMOPT_DQALLOC,
+ ip->i_udquot, &ip->i_gdquot);
+ /*
+ * Don't worry about the udquot that we may have
+ * attached above. It'll get detached, if not already.
+ */
+ if (error)
+ goto done;
+ nquotas++;
+ }
+
+ /*
+ * Attach this group quota to the user quota as a hint.
+ * This WON'T, in general, result in a thrash.
+ */
+ if (nquotas == 2) {
+ ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+ ASSERT(ip->i_udquot);
+ ASSERT(ip->i_gdquot);
+
+ /*
+ * We may or may not have the i_udquot locked at this point,
+ * but this check is OK since we don't depend on the i_gdquot to
+ * be accurate 100% all the time. It is just a hint, and this
+ * will succeed in general.
+ */
+ if (ip->i_udquot->q_gdquot == ip->i_gdquot)
+ goto done;
+ /*
+ * Attach i_gdquot to the gdquot hint inside the i_udquot.
+ */
+ xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
+ }
+
+ done:
+#ifdef DEBUG
+ if (!error) {
+ if (XFS_IS_UQUOTA_ON(mp))
+ ASSERT(ip->i_udquot);
+ if (XFS_IS_OQUOTA_ON(mp))
+ ASSERT(ip->i_gdquot);
+ }
+ ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
+#endif
+ return error;
+}
+
+int
+xfs_qm_dqattach(
+ struct xfs_inode *ip,
+ uint flags)
+{
+ int error;
+
+ xfs_ilock(ip, XFS_ILOCK_EXCL);
+ error = xfs_qm_dqattach_locked(ip, flags);
+ xfs_iunlock(ip, XFS_ILOCK_EXCL);
+
+ return error;
+}
+
+/*
+ * Release dquots (and their references) if any.
+ * The inode should be locked EXCL except when this's called by
+ * xfs_ireclaim.
+ */
+void
+xfs_qm_dqdetach(
+ xfs_inode_t *ip)
+{
+ if (!(ip->i_udquot || ip->i_gdquot))
+ return;
+
+ trace_xfs_dquot_dqdetach(ip);
+
+ ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
+ ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
+ if (ip->i_udquot) {
+ xfs_qm_dqrele(ip->i_udquot);
+ ip->i_udquot = NULL;
+ }
+ if (ip->i_gdquot) {
+ xfs_qm_dqrele(ip->i_gdquot);
+ ip->i_gdquot = NULL;
+ }
+}
+
+int
+xfs_qm_sync(
+ struct xfs_mount *mp,
+ int flags)
+{
+ struct xfs_quotainfo *q = mp->m_quotainfo;
+ int recl, restarts;
+ struct xfs_dquot *dqp;
+ int error;
+
+ if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
+ return 0;
+
+ restarts = 0;
+
+ again:
+ mutex_lock(&q->qi_dqlist_lock);
+ /*
+ * 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)) {
+ mutex_unlock(&q->qi_dqlist_lock);
+ return 0;
+ }
+ ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
+ list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
+ /*
+ * 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_TRYLOCK) {
+ if (!XFS_DQ_IS_DIRTY(dqp))
+ continue;
+ 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 = q->qi_dqreclaims;
+ if (!xfs_dqflock_nowait(dqp)) {
+ if (flags & SYNC_TRYLOCK) {
+ 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, so
+ * see if we can give a push to the buffer before we wait
+ * on the flush lock. At this point, we know that
+ * even though 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
+ */
+ mutex_unlock(&q->qi_dqlist_lock);
+ error = xfs_qm_dqflush(dqp, flags);
+ xfs_dqunlock(dqp);
+ if (error && XFS_FORCED_SHUTDOWN(mp))
+ return 0; /* Need to prevent umount failure */
+ else if (error)
+ return error;
+
+ mutex_lock(&q->qi_dqlist_lock);
+ if (recl != q->qi_dqreclaims) {
+ if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
+ break;
+
+ mutex_unlock(&q->qi_dqlist_lock);
+ goto again;
+ }
+ }
+
+ mutex_unlock(&q->qi_dqlist_lock);
+ return 0;
+}
+
+/*
+ * The hash chains and the mplist use the same xfs_dqhash structure as
+ * their list head, but we can take the mplist qh_lock and one of the
+ * hash qh_locks at the same time without any problem as they aren't
+ * related.
+ */
+static struct lock_class_key xfs_quota_mplist_class;
+
+/*
+ * 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);
+ mp->m_quotainfo = NULL;
+ return error;
+ }
+
+ INIT_LIST_HEAD(&qinf->qi_dqlist);
+ mutex_init(&qinf->qi_dqlist_lock);
+ lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
+
+ qinf->qi_dqreclaims = 0;
+
+ /* mutex used to serialize quotaoffs */
+ mutex_init(&qinf->qi_quotaofflock);
+
+ /* 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);
+
+ ASSERT(list_empty(&qi->qi_dqlist));
+ mutex_destroy(&qi->qi_dqlist_lock);
+
+ if (qi->qi_uquotaip) {
+ IRELE(qi->qi_uquotaip);
+ qi->qi_uquotaip = NULL; /* paranoia */
+ }
+ if (qi->qi_gquotaip) {
+ IRELE(qi->qi_gquotaip);
+ qi->qi_gquotaip = NULL;
+ }
+ mutex_destroy(&qi->qi_quotaofflock);
+ kmem_free(qi);
+ 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);
+ INIT_LIST_HEAD(&list->qh_list);
+ list->qh_version = 0;
+ list->qh_nelems = 0;
+}
+
+STATIC void
+xfs_qm_list_destroy(
+ xfs_dqlist_t *list)
+{
+ mutex_destroy(&(list->qh_lock));
+}
+
+/*
+ * 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;
+ 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;
+ }
+
+ error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
+ if (error) {
+ xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
+ XFS_TRANS_ABORT);
+ return error;
+ }
+
+ /*
+ * Make the changes in the superblock, and log those too.
+ * sbfields arg may contain fields other than *QUOTINO;
+ * VERSIONNUM for example.
+ */
+ spin_lock(&mp->m_sb_lock);
+ if (flags & XFS_QMOPT_SBVERSION) {
+ 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;
+