Re: [PATCH 1/8] xfs: track metadata health status

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Thu, Apr 11, 2019 at 08:29:04AM -0400, Brian Foster wrote:
> On Wed, Apr 10, 2019 at 06:45:32PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> > 
> > Add the necessary in-core metadata fields to keep track of which parts
> > of the filesystem have been observed and which parts were observed to be
> > unhealthy, and print a warning at unmount time if we have unfixed
> > problems.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> >  fs/xfs/Makefile            |    1 
> >  fs/xfs/libxfs/xfs_health.h |  175 ++++++++++++++++++++++++++++++++++++++++
> >  fs/xfs/xfs_health.c        |  192 ++++++++++++++++++++++++++++++++++++++++++++
> >  fs/xfs/xfs_icache.c        |    8 ++
> >  fs/xfs/xfs_inode.h         |    8 ++
> >  fs/xfs/xfs_mount.c         |    1 
> >  fs/xfs/xfs_mount.h         |   23 +++++
> >  fs/xfs/xfs_trace.h         |   73 +++++++++++++++++
> >  8 files changed, 481 insertions(+)
> >  create mode 100644 fs/xfs/libxfs/xfs_health.h
> >  create mode 100644 fs/xfs/xfs_health.c
> > 
> > 
> ...
> > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
> > index e70e7db29026..885decab4735 100644
> > --- a/fs/xfs/xfs_icache.c
> > +++ b/fs/xfs/xfs_icache.c
> > @@ -73,6 +73,8 @@ xfs_inode_alloc(
> >  	INIT_WORK(&ip->i_iodone_work, xfs_end_io);
> >  	INIT_LIST_HEAD(&ip->i_iodone_list);
> >  	spin_lock_init(&ip->i_iodone_lock);
> > +	ip->i_sick = 0;
> > +	ip->i_checked = 0;
> >  
> >  	return ip;
> >  }
> > @@ -133,6 +135,8 @@ xfs_inode_free(
> >  	spin_lock(&ip->i_flags_lock);
> >  	ip->i_flags = XFS_IRECLAIM;
> >  	ip->i_ino = 0;
> > +	ip->i_sick = 0;
> > +	ip->i_checked = 0;
> >  	spin_unlock(&ip->i_flags_lock);
> >  
> 
> FWIW, I'm not totally clear on what the i_checked mask is for yet.

Bleh, I forgot to update the introductory comment. :(

/*
 * <introductory stuff that's in xfs_health.h now>
 *
 * Each health tracking group uses a pair of fields for reporting.  The
 * "checked" field tell us if a given piece of metadata has ever been examined,
 * and the "sick" field tells us if that piece was found to need repairs.
 * Therefore we can conclude that for a given mask:
 *
 *  - checked && sick  => metadata needs repair
 *  - checked && !sick => metadata is ok
 *  - !checked         => has not been examined since mount
 */

In any case, I worked out the need for this new checked field when I was
writing the manual pages describing how all this worked:

https://djwong.org/docs/man/ioctl_xfs_fsop_geometry.2.html
https://djwong.org/docs/man/ioctl_xfs_ag_geometry.2.html
https://djwong.org/docs/man/ioctl_xfs_fsbulkstat.2.html

(See the part "The fields sick and checked indicate...")

@checked is a mask of all the metadata types that scrub has looked at,
whether or not the metadata was any good.  @sick is the mask of all the
metadata that scrub thought was bad, so we now can report to userspace
if something's good, bad, or unchecked.

> That aside, is it necessary to reset these fields in the free/reclaim
> paths?  I wonder if it's sufficient to zero them on alloc and the
> cache hit path just below..?

I think it's not strictly needed, but once we've broken the association
between a (struct xfs_inode *) buffer and a particular inode number, we
ought to zero out the health data just in case that buffer resurfaces
during the rcu grace period.

--D

> Otherwise looks fine:
> 
> Reviewed-by: Brian Foster <bfoster@redhat.com>
> 
> >  	__xfs_inode_free(ip);
> > @@ -449,6 +453,8 @@ xfs_iget_cache_hit(
> >  		ip->i_flags |= XFS_INEW;
> >  		xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
> >  		inode->i_state = I_NEW;
> > +		ip->i_sick = 0;
> > +		ip->i_checked = 0;
> >  
> >  		ASSERT(!rwsem_is_locked(&inode->i_rwsem));
> >  		init_rwsem(&inode->i_rwsem);
> > @@ -1177,6 +1183,8 @@ xfs_reclaim_inode(
> >  	spin_lock(&ip->i_flags_lock);
> >  	ip->i_flags = XFS_IRECLAIM;
> >  	ip->i_ino = 0;
> > +	ip->i_sick = 0;
> > +	ip->i_checked = 0;
> >  	spin_unlock(&ip->i_flags_lock);
> >  
> >  	xfs_iunlock(ip, XFS_ILOCK_EXCL);
> > diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
> > index 88239c2dd824..494e47ef42cb 100644
> > --- a/fs/xfs/xfs_inode.h
> > +++ b/fs/xfs/xfs_inode.h
> > @@ -45,6 +45,14 @@ typedef struct xfs_inode {
> >  	mrlock_t		i_lock;		/* inode lock */
> >  	mrlock_t		i_mmaplock;	/* inode mmap IO lock */
> >  	atomic_t		i_pincount;	/* inode pin count */
> > +
> > +	/*
> > +	 * Bitsets of inode metadata that have been checked and/or are sick.
> > +	 * Callers must hold i_flags_lock before accessing this field.
> > +	 */
> > +	uint16_t		i_checked;
> > +	uint16_t		i_sick;
> > +
> >  	spinlock_t		i_flags_lock;	/* inode i_flags lock */
> >  	/* Miscellaneous state. */
> >  	unsigned long		i_flags;	/* see defined flags below */
> > diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
> > index fd63b0b1307c..6581381c12be 100644
> > --- a/fs/xfs/xfs_mount.c
> > +++ b/fs/xfs/xfs_mount.c
> > @@ -231,6 +231,7 @@ xfs_initialize_perag(
> >  		error = xfs_iunlink_init(pag);
> >  		if (error)
> >  			goto out_hash_destroy;
> > +		spin_lock_init(&pag->pag_state_lock);
> >  	}
> >  
> >  	index = xfs_set_inode_alloc(mp, agcount);
> > diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
> > index 110f927cf943..cf7facc36a5f 100644
> > --- a/fs/xfs/xfs_mount.h
> > +++ b/fs/xfs/xfs_mount.h
> > @@ -60,6 +60,20 @@ struct xfs_error_cfg {
> >  typedef struct xfs_mount {
> >  	struct super_block	*m_super;
> >  	xfs_tid_t		m_tid;		/* next unused tid for fs */
> > +
> > +	/*
> > +	 * Bitsets of per-fs metadata that have been checked and/or are sick.
> > +	 * Callers must hold m_sb_lock to access these two fields.
> > +	 */
> > +	uint8_t			m_fs_checked;
> > +	uint8_t			m_fs_sick;
> > +	/*
> > +	 * Bitsets of rt metadata that have been checked and/or are sick.
> > +	 * Callers must hold m_sb_lock to access this field.
> > +	 */
> > +	uint8_t			m_rt_checked;
> > +	uint8_t			m_rt_sick;
> > +
> >  	struct xfs_ail		*m_ail;		/* fs active log item list */
> >  
> >  	struct xfs_sb		m_sb;		/* copy of fs superblock */
> > @@ -369,6 +383,15 @@ typedef struct xfs_perag {
> >  	xfs_agino_t	pagl_pagino;
> >  	xfs_agino_t	pagl_leftrec;
> >  	xfs_agino_t	pagl_rightrec;
> > +
> > +	/*
> > +	 * Bitsets of per-ag metadata that have been checked and/or are sick.
> > +	 * Callers should hold pag_state_lock before accessing this field.
> > +	 */
> > +	uint16_t	pag_checked;
> > +	uint16_t	pag_sick;
> > +	spinlock_t	pag_state_lock;
> > +
> >  	spinlock_t	pagb_lock;	/* lock for pagb_tree */
> >  	struct rb_root	pagb_tree;	/* ordered tree of busy extents */
> >  	unsigned int	pagb_gen;	/* generation count for pagb_tree */
> > diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
> > index 47fb07d86efd..f079841c7af6 100644
> > --- a/fs/xfs/xfs_trace.h
> > +++ b/fs/xfs/xfs_trace.h
> > @@ -3440,6 +3440,79 @@ DEFINE_AGINODE_EVENT(xfs_iunlink);
> >  DEFINE_AGINODE_EVENT(xfs_iunlink_remove);
> >  DEFINE_AG_EVENT(xfs_iunlink_map_prev_fallback);
> >  
> > +DECLARE_EVENT_CLASS(xfs_fs_corrupt_class,
> > +	TP_PROTO(struct xfs_mount *mp, unsigned int flags),
> > +	TP_ARGS(mp, flags),
> > +	TP_STRUCT__entry(
> > +		__field(dev_t, dev)
> > +		__field(unsigned int, flags)
> > +	),
> > +	TP_fast_assign(
> > +		__entry->dev = mp->m_super->s_dev;
> > +		__entry->flags = flags;
> > +	),
> > +	TP_printk("dev %d:%d flags 0x%x",
> > +		  MAJOR(__entry->dev), MINOR(__entry->dev),
> > +		  __entry->flags)
> > +);
> > +#define DEFINE_FS_CORRUPT_EVENT(name)	\
> > +DEFINE_EVENT(xfs_fs_corrupt_class, name,	\
> > +	TP_PROTO(struct xfs_mount *mp, unsigned int flags), \
> > +	TP_ARGS(mp, flags))
> > +DEFINE_FS_CORRUPT_EVENT(xfs_fs_mark_sick);
> > +DEFINE_FS_CORRUPT_EVENT(xfs_fs_mark_healthy);
> > +DEFINE_FS_CORRUPT_EVENT(xfs_rt_mark_sick);
> > +DEFINE_FS_CORRUPT_EVENT(xfs_rt_mark_healthy);
> > +
> > +DECLARE_EVENT_CLASS(xfs_ag_corrupt_class,
> > +	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, unsigned int flags),
> > +	TP_ARGS(mp, agno, flags),
> > +	TP_STRUCT__entry(
> > +		__field(dev_t, dev)
> > +		__field(xfs_agnumber_t, agno)
> > +		__field(unsigned int, flags)
> > +	),
> > +	TP_fast_assign(
> > +		__entry->dev = mp->m_super->s_dev;
> > +		__entry->agno = agno;
> > +		__entry->flags = flags;
> > +	),
> > +	TP_printk("dev %d:%d agno %u flags 0x%x",
> > +		  MAJOR(__entry->dev), MINOR(__entry->dev),
> > +		  __entry->agno, __entry->flags)
> > +);
> > +#define DEFINE_AG_CORRUPT_EVENT(name)	\
> > +DEFINE_EVENT(xfs_ag_corrupt_class, name,	\
> > +	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, \
> > +		 unsigned int flags), \
> > +	TP_ARGS(mp, agno, flags))
> > +DEFINE_AG_CORRUPT_EVENT(xfs_ag_mark_sick);
> > +DEFINE_AG_CORRUPT_EVENT(xfs_ag_mark_healthy);
> > +
> > +DECLARE_EVENT_CLASS(xfs_inode_corrupt_class,
> > +	TP_PROTO(struct xfs_inode *ip, unsigned int flags),
> > +	TP_ARGS(ip, flags),
> > +	TP_STRUCT__entry(
> > +		__field(dev_t, dev)
> > +		__field(xfs_ino_t, ino)
> > +		__field(unsigned int, flags)
> > +	),
> > +	TP_fast_assign(
> > +		__entry->dev = ip->i_mount->m_super->s_dev;
> > +		__entry->ino = ip->i_ino;
> > +		__entry->flags = flags;
> > +	),
> > +	TP_printk("dev %d:%d ino 0x%llx flags 0x%x",
> > +		  MAJOR(__entry->dev), MINOR(__entry->dev),
> > +		  __entry->ino, __entry->flags)
> > +);
> > +#define DEFINE_INODE_CORRUPT_EVENT(name)	\
> > +DEFINE_EVENT(xfs_inode_corrupt_class, name,	\
> > +	TP_PROTO(struct xfs_inode *ip, unsigned int flags), \
> > +	TP_ARGS(ip, flags))
> > +DEFINE_INODE_CORRUPT_EVENT(xfs_inode_mark_sick);
> > +DEFINE_INODE_CORRUPT_EVENT(xfs_inode_mark_healthy);
> > +
> >  #endif /* _TRACE_XFS_H */
> >  
> >  #undef TRACE_INCLUDE_PATH
> >