xfs: remove support for disabling quota accounting on a mounted file system
authorChristoph Hellwig <hch@lst.de>
Fri, 6 Aug 2021 18:05:36 +0000 (11:05 -0700)
committerDarrick J. Wong <djwong@kernel.org>
Fri, 6 Aug 2021 18:05:36 +0000 (11:05 -0700)
Disabling quota accounting is hairy, racy code with all kinds of pitfalls.
And it has a very strange mind set, as quota accounting (unlike
enforcement) really is a propery of the on-disk format.  There is no good
use case for supporting this.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
fs/xfs/libxfs/xfs_trans_resv.c
fs/xfs/libxfs/xfs_trans_resv.h
fs/xfs/xfs_dquot_item.c
fs/xfs/xfs_dquot_item.h
fs/xfs/xfs_qm.c
fs/xfs/xfs_qm.h
fs/xfs/xfs_qm_syscalls.c
fs/xfs/xfs_trans_dquot.c

index d1a0848cb52eca6698cd2a224f3f78179be2e9ad..ce12c8142bd18e16e8a302f63b856f2c8cdff160 100644 (file)
@@ -798,29 +798,6 @@ xfs_calc_qm_dqalloc_reservation(
                        XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1);
 }
 
-/*
- * Turning off quotas.
- *    the quota off logitems: sizeof(struct xfs_qoff_logitem) * 2
- *    the superblock for the quota flags: sector size
- */
-STATIC uint
-xfs_calc_qm_quotaoff_reservation(
-       struct xfs_mount        *mp)
-{
-       return sizeof(struct xfs_qoff_logitem) * 2 +
-               xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
-}
-
-/*
- * End of turning off quotas.
- *    the quota off logitems: sizeof(struct xfs_qoff_logitem) * 2
- */
-STATIC uint
-xfs_calc_qm_quotaoff_end_reservation(void)
-{
-       return sizeof(struct xfs_qoff_logitem) * 2;
-}
-
 /*
  * Syncing the incore super block changes to disk.
  *     the super block to reflect the changes: sector size
@@ -923,13 +900,6 @@ xfs_trans_resv_calc(
        resp->tr_qm_setqlim.tr_logres = xfs_calc_qm_setqlim_reservation();
        resp->tr_qm_setqlim.tr_logcount = XFS_DEFAULT_LOG_COUNT;
 
-       resp->tr_qm_quotaoff.tr_logres = xfs_calc_qm_quotaoff_reservation(mp);
-       resp->tr_qm_quotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT;
-
-       resp->tr_qm_equotaoff.tr_logres =
-               xfs_calc_qm_quotaoff_end_reservation();
-       resp->tr_qm_equotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT;
-
        resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp);
        resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT;
 
index 7241ab28cf84fe32ee7f854264f472f03591a5eb..fc4e9b369a3ae6bac41417d1aaeb895366aff6e2 100644 (file)
@@ -46,8 +46,6 @@ struct xfs_trans_resv {
        struct xfs_trans_res    tr_growrtfree;  /* grow realtime freeing */
        struct xfs_trans_res    tr_qm_setqlim;  /* adjust quota limits */
        struct xfs_trans_res    tr_qm_dqalloc;  /* allocate quota on disk */
-       struct xfs_trans_res    tr_qm_quotaoff; /* turn quota off */
-       struct xfs_trans_res    tr_qm_equotaoff;/* end of turn quota off */
        struct xfs_trans_res    tr_sb;          /* modify superblock */
        struct xfs_trans_res    tr_fsyncts;     /* update timestamps on fsync */
 };
index 8ed47b739b6ccccf18b58977adaa433eff5e5b02..6a1aae799cf16df4e28eb86b0bee173a32267d6a 100644 (file)
@@ -218,137 +218,3 @@ xfs_qm_dquot_logitem_init(
                                        &xfs_dquot_item_ops);
        lp->qli_dquot = dqp;
 }
-
-/*------------------  QUOTAOFF LOG ITEMS  -------------------*/
-
-static inline struct xfs_qoff_logitem *QOFF_ITEM(struct xfs_log_item *lip)
-{
-       return container_of(lip, struct xfs_qoff_logitem, qql_item);
-}
-
-
-/*
- * This returns the number of iovecs needed to log the given quotaoff item.
- * We only need 1 iovec for an quotaoff item.  It just logs the
- * quotaoff_log_format structure.
- */
-STATIC void
-xfs_qm_qoff_logitem_size(
-       struct xfs_log_item     *lip,
-       int                     *nvecs,
-       int                     *nbytes)
-{
-       *nvecs += 1;
-       *nbytes += sizeof(struct xfs_qoff_logitem);
-}
-
-STATIC void
-xfs_qm_qoff_logitem_format(
-       struct xfs_log_item     *lip,
-       struct xfs_log_vec      *lv)
-{
-       struct xfs_qoff_logitem *qflip = QOFF_ITEM(lip);
-       struct xfs_log_iovec    *vecp = NULL;
-       struct xfs_qoff_logformat *qlf;
-
-       qlf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_QUOTAOFF);
-       qlf->qf_type = XFS_LI_QUOTAOFF;
-       qlf->qf_size = 1;
-       qlf->qf_flags = qflip->qql_flags;
-       xlog_finish_iovec(lv, vecp, sizeof(struct xfs_qoff_logitem));
-}
-
-/*
- * There isn't much you can do to push a quotaoff item.  It is simply
- * stuck waiting for the log to be flushed to disk.
- */
-STATIC uint
-xfs_qm_qoff_logitem_push(
-       struct xfs_log_item     *lip,
-       struct list_head        *buffer_list)
-{
-       return XFS_ITEM_LOCKED;
-}
-
-STATIC xfs_lsn_t
-xfs_qm_qoffend_logitem_committed(
-       struct xfs_log_item     *lip,
-       xfs_lsn_t               lsn)
-{
-       struct xfs_qoff_logitem *qfe = QOFF_ITEM(lip);
-       struct xfs_qoff_logitem *qfs = qfe->qql_start_lip;
-
-       xfs_qm_qoff_logitem_relse(qfs);
-
-       kmem_free(lip->li_lv_shadow);
-       kmem_free(qfe);
-       return (xfs_lsn_t)-1;
-}
-
-STATIC void
-xfs_qm_qoff_logitem_release(
-       struct xfs_log_item     *lip)
-{
-       struct xfs_qoff_logitem *qoff = QOFF_ITEM(lip);
-
-       if (test_bit(XFS_LI_ABORTED, &lip->li_flags)) {
-               if (qoff->qql_start_lip)
-                       xfs_qm_qoff_logitem_relse(qoff->qql_start_lip);
-               xfs_qm_qoff_logitem_relse(qoff);
-       }
-}
-
-static const struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
-       .iop_size       = xfs_qm_qoff_logitem_size,
-       .iop_format     = xfs_qm_qoff_logitem_format,
-       .iop_committed  = xfs_qm_qoffend_logitem_committed,
-       .iop_push       = xfs_qm_qoff_logitem_push,
-       .iop_release    = xfs_qm_qoff_logitem_release,
-};
-
-static const struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
-       .iop_size       = xfs_qm_qoff_logitem_size,
-       .iop_format     = xfs_qm_qoff_logitem_format,
-       .iop_push       = xfs_qm_qoff_logitem_push,
-       .iop_release    = xfs_qm_qoff_logitem_release,
-};
-
-/*
- * Delete the quotaoff intent from the AIL and free it. On success,
- * this should only be called for the start item. It can be used for
- * either on shutdown or abort.
- */
-void
-xfs_qm_qoff_logitem_relse(
-       struct xfs_qoff_logitem *qoff)
-{
-       struct xfs_log_item     *lip = &qoff->qql_item;
-
-       ASSERT(test_bit(XFS_LI_IN_AIL, &lip->li_flags) ||
-              test_bit(XFS_LI_ABORTED, &lip->li_flags) ||
-              XFS_FORCED_SHUTDOWN(lip->li_mountp));
-       xfs_trans_ail_delete(lip, 0);
-       kmem_free(lip->li_lv_shadow);
-       kmem_free(qoff);
-}
-
-/*
- * Allocate and initialize an quotaoff item of the correct quota type(s).
- */
-struct xfs_qoff_logitem *
-xfs_qm_qoff_logitem_init(
-       struct xfs_mount        *mp,
-       struct xfs_qoff_logitem *start,
-       uint                    flags)
-{
-       struct xfs_qoff_logitem *qf;
-
-       qf = kmem_zalloc(sizeof(struct xfs_qoff_logitem), 0);
-
-       xfs_log_item_init(mp, &qf->qql_item, XFS_LI_QUOTAOFF, start ?
-                       &xfs_qm_qoffend_logitem_ops : &xfs_qm_qoff_logitem_ops);
-       qf->qql_item.li_mountp = mp;
-       qf->qql_start_lip = start;
-       qf->qql_flags = flags;
-       return qf;
-}
index 2b86a43d7ce2ec943bb0b505db3834f33ed7ca19..794710c2447493e7e8f9e57f0fba49ef75a3ee96 100644 (file)
@@ -9,7 +9,6 @@
 struct xfs_dquot;
 struct xfs_trans;
 struct xfs_mount;
-struct xfs_qoff_logitem;
 
 struct xfs_dq_logitem {
        struct xfs_log_item     qli_item;       /* common portion */
@@ -17,22 +16,6 @@ struct xfs_dq_logitem {
        xfs_lsn_t               qli_flush_lsn;  /* lsn at last flush */
 };
 
-struct xfs_qoff_logitem {
-       struct xfs_log_item     qql_item;       /* common portion */
-       struct xfs_qoff_logitem *qql_start_lip; /* qoff-start logitem, if any */
-       unsigned int            qql_flags;
-};
-
-
 void xfs_qm_dquot_logitem_init(struct xfs_dquot *dqp);
-struct xfs_qoff_logitem        *xfs_qm_qoff_logitem_init(struct xfs_mount *mp,
-               struct xfs_qoff_logitem *start,
-               uint flags);
-void xfs_qm_qoff_logitem_relse(struct xfs_qoff_logitem *);
-struct xfs_qoff_logitem        *xfs_trans_get_qoff_item(struct xfs_trans *tp,
-               struct xfs_qoff_logitem *startqoff,
-               uint flags);
-void xfs_trans_log_quotaoff_item(struct xfs_trans *tp,
-               struct xfs_qoff_logitem *qlp);
 
 #endif /* __XFS_DQUOT_ITEM_H__ */
index fe341f3fd4190155c467a20f346f770fffc71312..580b9dba21122bb90ff175110e154c102994f417 100644 (file)
@@ -185,7 +185,7 @@ out_unlock:
 /*
  * Purge the dquot cache.
  */
-void
+static void
 xfs_qm_dqpurge_all(
        struct xfs_mount        *mp,
        uint                    flags)
index ebbb484c49dc7e5bc1726eb6a219a607998f0610..442a0f97a9d4390bb69a632bd585b950ec2b744e 100644 (file)
@@ -140,9 +140,6 @@ struct xfs_dquot_acct {
 
 extern void            xfs_qm_destroy_quotainfo(struct xfs_mount *);
 
-/* dquot stuff */
-extern void            xfs_qm_dqpurge_all(struct xfs_mount *, uint);
-
 /* quota ops */
 extern int             xfs_qm_scall_trunc_qfiles(struct xfs_mount *, uint);
 extern int             xfs_qm_scall_getquota(struct xfs_mount *mp,
index 13a56e1ea15ce1eb8ad6023442ef4eac829e840c..d16deb75dc83d7e9baf37d43d24f946fb1b14821 100644 (file)
 #include "xfs_qm.h"
 #include "xfs_icache.h"
 
-STATIC int
-xfs_qm_log_quotaoff(
-       struct xfs_mount        *mp,
-       struct xfs_qoff_logitem **qoffstartp,
-       uint                    flags)
-{
-       struct xfs_trans        *tp;
-       int                     error;
-       struct xfs_qoff_logitem *qoffi;
-
-       error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_quotaoff, 0, 0, 0, &tp);
-       if (error)
-               goto out;
-
-       qoffi = xfs_trans_get_qoff_item(tp, NULL, flags & XFS_ALL_QUOTA_ACCT);
-       xfs_trans_log_quotaoff_item(tp, qoffi);
-
-       spin_lock(&mp->m_sb_lock);
-       mp->m_sb.sb_qflags = (mp->m_qflags & ~(flags)) & XFS_MOUNT_QUOTA_ALL;
-       spin_unlock(&mp->m_sb_lock);
-
-       xfs_log_sb(tp);
-
-       /*
-        * We have to make sure that the transaction is secure on disk before we
-        * return and actually stop quota accounting. So, make it synchronous.
-        * We don't care about quotoff's performance.
-        */
-       xfs_trans_set_sync(tp);
-       error = xfs_trans_commit(tp);
-       if (error)
-               goto out;
-
-       *qoffstartp = qoffi;
-out:
-       return error;
-}
-
-STATIC int
-xfs_qm_log_quotaoff_end(
-       struct xfs_mount        *mp,
-       struct xfs_qoff_logitem **startqoff,
-       uint                    flags)
-{
-       struct xfs_trans        *tp;
-       int                     error;
-       struct xfs_qoff_logitem *qoffi;
-
-       error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_equotaoff, 0, 0, 0, &tp);
-       if (error)
-               return error;
-
-       qoffi = xfs_trans_get_qoff_item(tp, *startqoff,
-                                       flags & XFS_ALL_QUOTA_ACCT);
-       xfs_trans_log_quotaoff_item(tp, qoffi);
-       *startqoff = NULL;
-
-       /*
-        * We have to make sure that the transaction is secure on disk before we
-        * return and actually stop quota accounting. So, make it synchronous.
-        * We don't care about quotoff's performance.
-        */
-       xfs_trans_set_sync(tp);
-       return xfs_trans_commit(tp);
-}
-
-/*
- * Turn off quota accounting and/or enforcement for all udquots and/or
- * gdquots. Called only at unmount time.
- *
- * This assumes that there are no dquots of this file system cached
- * incore, and modifies the ondisk dquot directly. Therefore, for example,
- * it is an error to call this twice, without purging the cache.
- */
 int
 xfs_qm_scall_quotaoff(
        xfs_mount_t             *mp,
        uint                    flags)
 {
-       struct xfs_quotainfo    *q = mp->m_quotainfo;
-       uint                    dqtype;
-       int                     error;
-       uint                    inactivate_flags;
-       struct xfs_qoff_logitem *qoffstart = NULL;
-
        /*
         * No file system can have quotas enabled on disk but not in core.
         * Note that quota utilities (like quotaoff) _expect_
@@ -111,160 +31,23 @@ xfs_qm_scall_quotaoff(
         */
        if ((mp->m_qflags & flags) == 0)
                return -EEXIST;
-       error = 0;
-
-       flags &= (XFS_ALL_QUOTA_ACCT | XFS_ALL_QUOTA_ENFD);
 
        /*
-        * We don't want to deal with two quotaoffs messing up each other,
-        * so we're going to serialize it. quotaoff isn't exactly a performance
-        * critical thing.
-        * If quotaoff, then we must be dealing with the root filesystem.
+        * We do not support actually turning off quota accounting any more.
+        * Just log a warning and ignore the accounting related flags.
         */
-       ASSERT(q);
-       mutex_lock(&q->qi_quotaofflock);
+       if (flags & XFS_ALL_QUOTA_ACCT)
+               xfs_info(mp, "disabling of quota accounting not supported.");
 
-       /*
-        * If we're just turning off quota enforcement, change mp and go.
-        */
-       if ((flags & XFS_ALL_QUOTA_ACCT) == 0) {
-               mp->m_qflags &= ~(flags);
-
-               spin_lock(&mp->m_sb_lock);
-               mp->m_sb.sb_qflags = mp->m_qflags;
-               spin_unlock(&mp->m_sb_lock);
-               mutex_unlock(&q->qi_quotaofflock);
-
-               /* XXX what to do if error ? Revert back to old vals incore ? */
-               return xfs_sync_sb(mp, false);
-       }
-
-       dqtype = 0;
-       inactivate_flags = 0;
-       /*
-        * If accounting is off, we must turn enforcement off, clear the
-        * quota 'CHKD' certificate to make it known that we have to
-        * do a quotacheck the next time this quota is turned on.
-        */
-       if (flags & XFS_UQUOTA_ACCT) {
-               dqtype |= XFS_QMOPT_UQUOTA;
-               flags |= (XFS_UQUOTA_CHKD | XFS_UQUOTA_ENFD);
-               inactivate_flags |= XFS_UQUOTA_ACTIVE;
-       }
-       if (flags & XFS_GQUOTA_ACCT) {
-               dqtype |= XFS_QMOPT_GQUOTA;
-               flags |= (XFS_GQUOTA_CHKD | XFS_GQUOTA_ENFD);
-               inactivate_flags |= XFS_GQUOTA_ACTIVE;
-       }
-       if (flags & XFS_PQUOTA_ACCT) {
-               dqtype |= XFS_QMOPT_PQUOTA;
-               flags |= (XFS_PQUOTA_CHKD | XFS_PQUOTA_ENFD);
-               inactivate_flags |= XFS_PQUOTA_ACTIVE;
-       }
-
-       /*
-        * Nothing to do?  Don't complain. This happens when we're just
-        * turning off quota enforcement.
-        */
-       if ((mp->m_qflags & flags) == 0)
-               goto out_unlock;
-
-       /*
-        * Write the LI_QUOTAOFF log record, and do SB changes atomically,
-        * and synchronously. If we fail to write, we should abort the
-        * operation as it cannot be recovered safely if we crash.
-        */
-       error = xfs_qm_log_quotaoff(mp, &qoffstart, flags);
-       if (error)
-               goto out_unlock;
-
-       /*
-        * Next we clear the XFS_MOUNT_*DQ_ACTIVE bit(s) in the mount struct
-        * to take care of the race between dqget and quotaoff. We don't take
-        * any special locks to reset these bits. All processes need to check
-        * these bits *after* taking inode lock(s) to see if the particular
-        * quota type is in the process of being turned off. If *ACTIVE, it is
-        * guaranteed that all dquot structures and all quotainode ptrs will all
-        * stay valid as long as that inode is kept locked.
-        *
-        * There is no turning back after this.
-        */
-       mp->m_qflags &= ~inactivate_flags;
-
-       /*
-        * Give back all the dquot reference(s) held by inodes.
-        * Here we go thru every single incore inode in this file system, and
-        * do a dqrele on the i_udquot/i_gdquot that it may have.
-        * Essentially, as long as somebody has an inode locked, this guarantees
-        * that quotas will not be turned off. This is handy because in a
-        * transaction once we lock the inode(s) and check for quotaon, we can
-        * depend on the quota inodes (and other things) being valid as long as
-        * we keep the lock(s).
-        */
-       error = xfs_dqrele_all_inodes(mp, flags);
-       ASSERT(!error);
-
-       /*
-        * Next we make the changes in the quota flag in the mount struct.
-        * This isn't protected by a particular lock directly, because we
-        * don't want to take a mrlock every time we depend on quotas being on.
-        */
-       mp->m_qflags &= ~flags;
-
-       /*
-        * Go through all the dquots of this file system and purge them,
-        * according to what was turned off.
-        */
-       xfs_qm_dqpurge_all(mp, dqtype);
-
-       /*
-        * Transactions that had started before ACTIVE state bit was cleared
-        * could have logged many dquots, so they'd have higher LSNs than
-        * the first QUOTAOFF log record does. If we happen to crash when
-        * the tail of the log has gone past the QUOTAOFF record, but
-        * before the last dquot modification, those dquots __will__
-        * recover, and that's not good.
-        *
-        * So, we have QUOTAOFF start and end logitems; the start
-        * logitem won't get overwritten until the end logitem appears...
-        */
-       error = xfs_qm_log_quotaoff_end(mp, &qoffstart, flags);
-       if (error) {
-               /* We're screwed now. Shutdown is the only option. */
-               xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
-               goto out_unlock;
-       }
-
-       /*
-        * If all quotas are completely turned off, close shop.
-        */
-       if (mp->m_qflags == 0) {
-               mutex_unlock(&q->qi_quotaofflock);
-               xfs_qm_destroy_quotainfo(mp);
-               return 0;
-       }
-
-       /*
-        * Release our quotainode references if we don't need them anymore.
-        */
-       if ((dqtype & XFS_QMOPT_UQUOTA) && q->qi_uquotaip) {
-               xfs_irele(q->qi_uquotaip);
-               q->qi_uquotaip = NULL;
-       }
-       if ((dqtype & XFS_QMOPT_GQUOTA) && q->qi_gquotaip) {
-               xfs_irele(q->qi_gquotaip);
-               q->qi_gquotaip = NULL;
-       }
-       if ((dqtype & XFS_QMOPT_PQUOTA) && q->qi_pquotaip) {
-               xfs_irele(q->qi_pquotaip);
-               q->qi_pquotaip = NULL;
-       }
+       mutex_lock(&mp->m_quotainfo->qi_quotaofflock);
+       mp->m_qflags &= ~(flags & XFS_ALL_QUOTA_ENFD);
+       spin_lock(&mp->m_sb_lock);
+       mp->m_sb.sb_qflags = mp->m_qflags;
+       spin_unlock(&mp->m_sb_lock);
+       mutex_unlock(&mp->m_quotainfo->qi_quotaofflock);
 
-out_unlock:
-       if (error && qoffstart)
-               xfs_qm_qoff_logitem_relse(qoffstart);
-       mutex_unlock(&q->qi_quotaofflock);
-       return error;
+       /* XXX what to do if error ? Revert back to old vals incore ? */
+       return xfs_sync_sb(mp, false);
 }
 
 STATIC int
index 48e09ea30ee53999e20d4640f9f9ad7b6698bce3..b7e4b05a559bdbe039893276201dcc96267dc5fb 100644 (file)
@@ -843,44 +843,6 @@ xfs_trans_reserve_quota_icreate(
                        dblocks, 1, XFS_QMOPT_RES_REGBLKS);
 }
 
-/*
- * This routine is called to allocate a quotaoff log item.
- */
-struct xfs_qoff_logitem *
-xfs_trans_get_qoff_item(
-       struct xfs_trans        *tp,
-       struct xfs_qoff_logitem *startqoff,
-       uint                    flags)
-{
-       struct xfs_qoff_logitem *q;
-
-       ASSERT(tp != NULL);
-
-       q = xfs_qm_qoff_logitem_init(tp->t_mountp, startqoff, flags);
-       ASSERT(q != NULL);
-
-       /*
-        * Get a log_item_desc to point at the new item.
-        */
-       xfs_trans_add_item(tp, &q->qql_item);
-       return q;
-}
-
-
-/*
- * This is called to mark the quotaoff logitem as needing
- * to be logged when the transaction is committed.  The logitem must
- * already be associated with the given transaction.
- */
-void
-xfs_trans_log_quotaoff_item(
-       struct xfs_trans        *tp,
-       struct xfs_qoff_logitem *qlp)
-{
-       tp->t_flags |= XFS_TRANS_DIRTY;
-       set_bit(XFS_LI_DIRTY, &qlp->qql_item.li_flags);
-}
-
 STATIC void
 xfs_trans_alloc_dqinfo(
        xfs_trans_t     *tp)