Message ID | 151062258598.8554.8157038002895095232.stgit@dwillia2-desk3.amr.corp.intel.com |
---|---|
State | Not Applicable, archived |
Headers | show |
Series | fs, dax: unify IOMAP_F_DIRTY read vs write handling policy in the dax core | expand |
On Mon, Nov 13, 2017 at 05:27:54PM -0800, Dan Williams wrote: > While reviewing whether MAP_SYNC should strengthen its current guarantee > of syncing writes from the initiating process to also include > third-party readers observing dirty metadata, Dave pointed out that the > check of IOMAP_WRITE is misplaced. > > The policy of what to with IOMAP_F_DIRTY should be separated from the > generic filesystem mechanism of reporting dirty metadata. Move this > policy to the fs-dax core to simplify the per-filesystem iomap handlers, > and further centralize code that implements the MAP_SYNC policy. This > otherwise should not change behavior, it just makes it easier to change > behavior in the future. Yes, this was what I was looking for on my last read-through. :) > Cc: Jan Kara <jack@suse.cz> > Cc: Christoph Hellwig <hch@lst.de> > Cc: Darrick J. Wong <darrick.wong@oracle.com> > Cc: Ross Zwisler <ross.zwisler@linux.intel.com> > Reported-by: Dave Chinner <david@fromorbit.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > This is an additional cleanup I want to include in the 4.15 merge > request for the MAP_SYNC feature. Please review, I'm looking to send the > pull request towards the end of the week. > > > fs/dax.c | 15 +++++++++++++-- > fs/ext4/inode.c | 2 +- > fs/xfs/xfs_iomap.c | 6 +++--- > 3 files changed, 17 insertions(+), 6 deletions(-) > > diff --git a/fs/dax.c b/fs/dax.c > index 78233c716757..27ba300660ff 100644 > --- a/fs/dax.c > +++ b/fs/dax.c > @@ -1079,6 +1079,17 @@ static int dax_fault_return(int error) > return VM_FAULT_SIGBUS; > } > > +/* > + * MAP_SYNC on a dax mapping guarantees dirty metadata is > + * flushed on write-faults (non-cow), but not read-faults. > + */ > +static bool dax_fault_is_synchronous(unsigned long flags, > + struct vm_area_struct *vma, struct iomap *iomap) > +{ > + return (flags & IOMAP_WRITE) && (vma->vm_flags & VM_SYNC) > + && (iomap->flags & IOMAP_F_DIRTY); > +} > + > static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, > const struct iomap_ops *ops) > { > @@ -1170,7 +1181,7 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, > goto finish_iomap; > } > > - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); > + sync = dax_fault_is_synchronous(flags, vma, &iomap); > > switch (iomap.type) { > case IOMAP_MAPPED: > @@ -1390,7 +1401,7 @@ static int dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp, > if (iomap.offset + iomap.length < pos + PMD_SIZE) > goto finish_iomap; > > - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); > + sync = dax_fault_is_synchronous(iomap_flags, vma, &iomap); > > switch (iomap.type) { > case IOMAP_MAPPED: > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c > index 13a198924a0f..ee4d907a4251 100644 > --- a/fs/ext4/inode.c > +++ b/fs/ext4/inode.c > @@ -3479,7 +3479,7 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, > } > > iomap->flags = 0; > - if ((flags & IOMAP_WRITE) && ext4_inode_datasync_dirty(inode)) > + if (ext4_inode_datasync_dirty(inode)) > iomap->flags |= IOMAP_F_DIRTY; > iomap->bdev = inode->i_sb->s_bdev; > iomap->dax_dev = sbi->s_daxdev; > diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c > index b43be199fbdf..888b60189983 100644 > --- a/fs/xfs/xfs_iomap.c > +++ b/fs/xfs/xfs_iomap.c > @@ -1087,9 +1087,9 @@ xfs_file_iomap_begin( > trace_xfs_iomap_found(ip, offset, length, 0, &imap); > } > > - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && > - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) > - iomap->flags |= IOMAP_F_DIRTY; > + if (xfs_ipincount(ip)) > + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) Why not if (xfs_ipincount(ip) && (ip->... & ~XFS_ILOG_TIMESTAMP)) ? Otherwise looks ok (just this patch, will rvb the rest separately), Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> --D > + iomap->flags |= IOMAP_F_DIRTY; > > xfs_bmbt_to_iomap(ip, iomap, &imap); > > > -- > To unsubscribe from this list: send the line "unsubscribe linux-xfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Nov 13, 2017 at 05:27:54PM -0800, Dan Williams wrote: > diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c > index b43be199fbdf..888b60189983 100644 > --- a/fs/xfs/xfs_iomap.c > +++ b/fs/xfs/xfs_iomap.c > @@ -1087,9 +1087,9 @@ xfs_file_iomap_begin( > trace_xfs_iomap_found(ip, offset, length, 0, &imap); > } > > - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && > - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) > - iomap->flags |= IOMAP_F_DIRTY; > + if (xfs_ipincount(ip)) > + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) > + iomap->flags |= IOMAP_F_DIRTY; Please make a helper for this, and use it in xfs_file_fsync() where the same dirty checks are done. e.g. static inline bool xfs_inode_need_fsync(struct xfs_inode *ip, bool datasync) { if (!xfs_ipincount(ip)) return false; if (xfs_inode_clean(ip)) return false; if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) return false; return true; } Cheers, Dave.
> - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && > - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) > - iomap->flags |= IOMAP_F_DIRTY; > + if (xfs_ipincount(ip)) > + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) > + iomap->flags |= IOMAP_F_DIRTY; Please don't split up the conditional and make it harder to read. With that fixed up this looks fine to me: Reviewed-by: Christoph Hellwig <hch@lst.de>
On Tue, Nov 14, 2017 at 03:12:28PM +1100, Dave Chinner wrote: > On Mon, Nov 13, 2017 at 05:27:54PM -0800, Dan Williams wrote: > > diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c > > index b43be199fbdf..888b60189983 100644 > > --- a/fs/xfs/xfs_iomap.c > > +++ b/fs/xfs/xfs_iomap.c > > @@ -1087,9 +1087,9 @@ xfs_file_iomap_begin( > > trace_xfs_iomap_found(ip, offset, length, 0, &imap); > > } > > > > - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && > > - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) > > - iomap->flags |= IOMAP_F_DIRTY; > > + if (xfs_ipincount(ip)) > > + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) > > + iomap->flags |= IOMAP_F_DIRTY; > > Please make a helper for this, and use it in xfs_file_fsync() where > the same dirty checks are done. e.g. Let's keep that sort of cleanup separate. I initially considered it when doing the MAP_SYNC work, but fsync works a little different in that it looks at ili_last_lsn, so your little helper below won't work as-is. You can send it as an incremental patch later and we can see if we can come up with something workable.
On Mon 13-11-17 17:27:54, Dan Williams wrote: > While reviewing whether MAP_SYNC should strengthen its current guarantee > of syncing writes from the initiating process to also include > third-party readers observing dirty metadata, Dave pointed out that the > check of IOMAP_WRITE is misplaced. > > The policy of what to with IOMAP_F_DIRTY should be separated from the > generic filesystem mechanism of reporting dirty metadata. Move this > policy to the fs-dax core to simplify the per-filesystem iomap handlers, > and further centralize code that implements the MAP_SYNC policy. This > otherwise should not change behavior, it just makes it easier to change > behavior in the future. > > Cc: Jan Kara <jack@suse.cz> > Cc: Christoph Hellwig <hch@lst.de> > Cc: Darrick J. Wong <darrick.wong@oracle.com> > Cc: Ross Zwisler <ross.zwisler@linux.intel.com> > Reported-by: Dave Chinner <david@fromorbit.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > This is an additional cleanup I want to include in the 4.15 merge > request for the MAP_SYNC feature. Please review, I'm looking to send the > pull request towards the end of the week. The patch looks fine to me. You can add: Reviewed-by: Jan Kara <jack@suse.cz> Honza > > > fs/dax.c | 15 +++++++++++++-- > fs/ext4/inode.c | 2 +- > fs/xfs/xfs_iomap.c | 6 +++--- > 3 files changed, 17 insertions(+), 6 deletions(-) > > diff --git a/fs/dax.c b/fs/dax.c > index 78233c716757..27ba300660ff 100644 > --- a/fs/dax.c > +++ b/fs/dax.c > @@ -1079,6 +1079,17 @@ static int dax_fault_return(int error) > return VM_FAULT_SIGBUS; > } > > +/* > + * MAP_SYNC on a dax mapping guarantees dirty metadata is > + * flushed on write-faults (non-cow), but not read-faults. > + */ > +static bool dax_fault_is_synchronous(unsigned long flags, > + struct vm_area_struct *vma, struct iomap *iomap) > +{ > + return (flags & IOMAP_WRITE) && (vma->vm_flags & VM_SYNC) > + && (iomap->flags & IOMAP_F_DIRTY); > +} > + > static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, > const struct iomap_ops *ops) > { > @@ -1170,7 +1181,7 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, > goto finish_iomap; > } > > - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); > + sync = dax_fault_is_synchronous(flags, vma, &iomap); > > switch (iomap.type) { > case IOMAP_MAPPED: > @@ -1390,7 +1401,7 @@ static int dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp, > if (iomap.offset + iomap.length < pos + PMD_SIZE) > goto finish_iomap; > > - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); > + sync = dax_fault_is_synchronous(iomap_flags, vma, &iomap); > > switch (iomap.type) { > case IOMAP_MAPPED: > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c > index 13a198924a0f..ee4d907a4251 100644 > --- a/fs/ext4/inode.c > +++ b/fs/ext4/inode.c > @@ -3479,7 +3479,7 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, > } > > iomap->flags = 0; > - if ((flags & IOMAP_WRITE) && ext4_inode_datasync_dirty(inode)) > + if (ext4_inode_datasync_dirty(inode)) > iomap->flags |= IOMAP_F_DIRTY; > iomap->bdev = inode->i_sb->s_bdev; > iomap->dax_dev = sbi->s_daxdev; > diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c > index b43be199fbdf..888b60189983 100644 > --- a/fs/xfs/xfs_iomap.c > +++ b/fs/xfs/xfs_iomap.c > @@ -1087,9 +1087,9 @@ xfs_file_iomap_begin( > trace_xfs_iomap_found(ip, offset, length, 0, &imap); > } > > - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && > - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) > - iomap->flags |= IOMAP_F_DIRTY; > + if (xfs_ipincount(ip)) > + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) > + iomap->flags |= IOMAP_F_DIRTY; > > xfs_bmbt_to_iomap(ip, iomap, &imap); > >
diff --git a/fs/dax.c b/fs/dax.c index 78233c716757..27ba300660ff 100644 --- a/fs/dax.c +++ b/fs/dax.c @@ -1079,6 +1079,17 @@ static int dax_fault_return(int error) return VM_FAULT_SIGBUS; } +/* + * MAP_SYNC on a dax mapping guarantees dirty metadata is + * flushed on write-faults (non-cow), but not read-faults. + */ +static bool dax_fault_is_synchronous(unsigned long flags, + struct vm_area_struct *vma, struct iomap *iomap) +{ + return (flags & IOMAP_WRITE) && (vma->vm_flags & VM_SYNC) + && (iomap->flags & IOMAP_F_DIRTY); +} + static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, const struct iomap_ops *ops) { @@ -1170,7 +1181,7 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, goto finish_iomap; } - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); + sync = dax_fault_is_synchronous(flags, vma, &iomap); switch (iomap.type) { case IOMAP_MAPPED: @@ -1390,7 +1401,7 @@ static int dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp, if (iomap.offset + iomap.length < pos + PMD_SIZE) goto finish_iomap; - sync = (vma->vm_flags & VM_SYNC) && (iomap.flags & IOMAP_F_DIRTY); + sync = dax_fault_is_synchronous(iomap_flags, vma, &iomap); switch (iomap.type) { case IOMAP_MAPPED: diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c index 13a198924a0f..ee4d907a4251 100644 --- a/fs/ext4/inode.c +++ b/fs/ext4/inode.c @@ -3479,7 +3479,7 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, } iomap->flags = 0; - if ((flags & IOMAP_WRITE) && ext4_inode_datasync_dirty(inode)) + if (ext4_inode_datasync_dirty(inode)) iomap->flags |= IOMAP_F_DIRTY; iomap->bdev = inode->i_sb->s_bdev; iomap->dax_dev = sbi->s_daxdev; diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index b43be199fbdf..888b60189983 100644 --- a/fs/xfs/xfs_iomap.c +++ b/fs/xfs/xfs_iomap.c @@ -1087,9 +1087,9 @@ xfs_file_iomap_begin( trace_xfs_iomap_found(ip, offset, length, 0, &imap); } - if ((flags & IOMAP_WRITE) && xfs_ipincount(ip) && - (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) - iomap->flags |= IOMAP_F_DIRTY; + if (xfs_ipincount(ip)) + if (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP) + iomap->flags |= IOMAP_F_DIRTY; xfs_bmbt_to_iomap(ip, iomap, &imap);
While reviewing whether MAP_SYNC should strengthen its current guarantee of syncing writes from the initiating process to also include third-party readers observing dirty metadata, Dave pointed out that the check of IOMAP_WRITE is misplaced. The policy of what to with IOMAP_F_DIRTY should be separated from the generic filesystem mechanism of reporting dirty metadata. Move this policy to the fs-dax core to simplify the per-filesystem iomap handlers, and further centralize code that implements the MAP_SYNC policy. This otherwise should not change behavior, it just makes it easier to change behavior in the future. Cc: Jan Kara <jack@suse.cz> Cc: Christoph Hellwig <hch@lst.de> Cc: Darrick J. Wong <darrick.wong@oracle.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Reported-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- This is an additional cleanup I want to include in the 4.15 merge request for the MAP_SYNC feature. Please review, I'm looking to send the pull request towards the end of the week. fs/dax.c | 15 +++++++++++++-- fs/ext4/inode.c | 2 +- fs/xfs/xfs_iomap.c | 6 +++--- 3 files changed, 17 insertions(+), 6 deletions(-)