xfs: avoid ilock games in the quota scrubber
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 9 May 2018 17:02:00 +0000 (10:02 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Wed, 16 May 2018 00:57:05 +0000 (17:57 -0700)
Refactor the quota scrubber to take the quotaofflock and grab the quota
inode in the setup function so that we can treat quota in the same
"scrub in the context of this inode" (i.e. sc->ip) manner as we treat
any other inode.  We do have to drop the quota inode's ILOCK_EXCL to use
dqiterate, but since dquots have their own individual locks the ILOCK
wasn't helping us anyway.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
fs/xfs/scrub/quota.c
fs/xfs/scrub/scrub.c
fs/xfs/scrub/scrub.h

index ba87c3a..d3d0897 100644 (file)
@@ -66,12 +66,24 @@ xfs_scrub_setup_quota(
        struct xfs_inode                *ip)
 {
        uint                            dqtype;
+       int                             error;
+
+       if (!XFS_IS_QUOTA_RUNNING(sc->mp) || !XFS_IS_QUOTA_ON(sc->mp))
+               return -ENOENT;
 
        dqtype = xfs_scrub_quota_to_dqtype(sc);
        if (dqtype == 0)
                return -EINVAL;
+       sc->has_quotaofflock = true;
+       mutex_lock(&sc->mp->m_quotainfo->qi_quotaofflock);
        if (!xfs_this_quota_on(sc->mp, dqtype))
                return -ENOENT;
+       error = xfs_scrub_setup_fs(sc, ip);
+       if (error)
+               return error;
+       sc->ip = xfs_quota_inode(sc->mp, dqtype);
+       xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
+       sc->ilock_flags = XFS_ILOCK_EXCL;
        return 0;
 }
 
@@ -202,7 +214,6 @@ xfs_scrub_quota(
        struct xfs_bmbt_irec            irec = { 0 };
        struct xfs_scrub_quota_info     sqi;
        struct xfs_mount                *mp = sc->mp;
-       struct xfs_inode                *ip;
        struct xfs_quotainfo            *qi = mp->m_quotainfo;
        xfs_fileoff_t                   max_dqid_off;
        xfs_fileoff_t                   off = 0;
@@ -210,25 +221,12 @@ xfs_scrub_quota(
        int                             nimaps;
        int                             error = 0;
 
-       if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
-               return -ENOENT;
-
-       mutex_lock(&qi->qi_quotaofflock);
        dqtype = xfs_scrub_quota_to_dqtype(sc);
-       if (!xfs_this_quota_on(sc->mp, dqtype)) {
-               error = -ENOENT;
-               goto out_unlock_quota;
-       }
-
-       /* Attach to the quota inode and set sc->ip so that reporting works. */
-       ip = xfs_quota_inode(sc->mp, dqtype);
-       sc->ip = ip;
 
        /* Look for problem extents. */
-       xfs_ilock(ip, XFS_ILOCK_EXCL);
-       if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
+       if (sc->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
                xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino);
-               goto out_unlock_inode;
+               goto out;
        }
        max_dqid_off = ((xfs_dqid_t)-1) / qi->qi_dqperchunk;
        while (1) {
@@ -237,11 +235,11 @@ xfs_scrub_quota(
 
                off = irec.br_startoff + irec.br_blockcount;
                nimaps = 1;
-               error = xfs_bmapi_read(ip, off, -1, &irec, &nimaps,
+               error = xfs_bmapi_read(sc->ip, off, -1, &irec, &nimaps,
                                XFS_BMAPI_ENTIRE);
                if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, off,
                                &error))
-                       goto out_unlock_inode;
+                       goto out;
                if (!nimaps)
                        break;
                if (irec.br_startblock == HOLESTARTBLOCK)
@@ -267,26 +265,25 @@ xfs_scrub_quota(
                    irec.br_startoff + irec.br_blockcount > max_dqid_off + 1)
                        xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, off);
        }
-       xfs_iunlock(ip, XFS_ILOCK_EXCL);
        if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
                goto out;
 
-       /* Check all the quota items. */
+       /*
+        * Check all the quota items.  Now that we've checked the quota inode
+        * data fork we have to drop ILOCK_EXCL to use the regular dquot
+        * functions.
+        */
+       xfs_iunlock(sc->ip, sc->ilock_flags);
+       sc->ilock_flags = 0;
        sqi.sc = sc;
        sqi.last_id = 0;
        error = xfs_qm_dqiterate(mp, dqtype, xfs_scrub_quota_item, &sqi);
+       sc->ilock_flags = XFS_ILOCK_EXCL;
+       xfs_ilock(sc->ip, sc->ilock_flags);
        if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK,
                        sqi.last_id * qi->qi_dqperchunk, &error))
                goto out;
 
 out:
-       /* We set sc->ip earlier, so make sure we clear it now. */
-       sc->ip = NULL;
-out_unlock_quota:
-       mutex_unlock(&qi->qi_quotaofflock);
        return error;
-
-out_unlock_inode:
-       xfs_iunlock(ip, XFS_ILOCK_EXCL);
-       goto out;
 }
index 26c7596..67509d8 100644 (file)
@@ -42,6 +42,8 @@
 #include "xfs_refcount_btree.h"
 #include "xfs_rmap.h"
 #include "xfs_rmap_btree.h"
+#include "xfs_quota.h"
+#include "xfs_qm.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
@@ -166,6 +168,8 @@ xfs_scrub_teardown(
                        iput(VFS_I(sc->ip));
                sc->ip = NULL;
        }
+       if (sc->has_quotaofflock)
+               mutex_unlock(&sc->mp->m_quotainfo->qi_quotaofflock);
        if (sc->buf) {
                kmem_free(sc->buf);
                sc->buf = NULL;
index 0d92af8..5d79731 100644 (file)
@@ -73,6 +73,7 @@ struct xfs_scrub_context {
        void                            *buf;
        uint                            ilock_flags;
        bool                            try_harder;
+       bool                            has_quotaofflock;
 
        /* State tracking for single-AG operations. */
        struct xfs_scrub_ag             sa;