xfs: merge fsync and O_SYNC handling
PATCH 1/1 xfs: merge fsync and O_SYNC handling
The guarantees for O_SYNC are exactly the same as the ones we need to make for an fsync call (and given that Linux O_SYNC is O_DSYNC the equivalent is fdadatasync, but we treat both the same in XFS), except with a range data writeout. Jan Kara has started unifying these two path for filesystems using the generic helpers, and I've started to look at XFS.
The actual transaction commited by xfs_fsync and xfs_write_sync_logforce has a different transaction number, but actually is exactly the same. We'll only use the fsync transaction going forward. One major difference is that xfs_write_sync_logforce never issues a cache flush unless we commit a transaction causing that as a side-effect, which is an obvious bug in the O_SYNC handling. Second all the locking and i_update_size vs i_update_core changes from 978b7237123d007b9fa983af6e0e2fa8f97f9934 never made it to xfs_write_sync_logforce, so we add them back.
To make xfs_fsync easily usable from the O_SYNC path, the filemap_fdatawait call is moved up to xfs_file_fsync, so that we don't wait on the whole file after we already waited for our portion in xfs_write.
We'll also use a plain call to filemap_write_and_wait_range instead of the previous sync_page_rang which did it in two steps including an half-hearted inode write out that doesn't help us.
Once we're done with this also remove the now useless i_update_size tracking.
fs/xfs/linux-2.6/xfs_file.c +14 -5
@@ -172,12 +172,21 @@ xfs_file_release(*/ STATIC intxfs_file_fsync ( -struct file *filp, -struct dentry *dentry, -int datasync) +struct file *file, +struct dentry *dentry, +int datasync) { -xfs_iflags_clear (XFS_I (dentry->d_inode),XFS_ITRUNCATED ); -return -xfs_fsync (XFS_I (dentry->d_inode)); +struct inode *inode = dentry->d_inode; +struct xfs_inode *ip =XFS_I (inode); +int error; + +/* capture size updates in I/O completion before writing the inode. */ + error =filemap_fdatawait (inode->i_mapping); +if (error) +return error; + +xfs_iflags_clear (ip,XFS_ITRUNCATED ); +return -xfs_fsync (ip); }STATIC int
fs/xfs/linux-2.6/xfs_lrw.c +5 -2
@@ -812,18 +812,21 @@ write_retry:/* Handle various SYNC-type writes */ if ((file->f_flags &O_SYNC ) ||IS_SYNC (inode)) { +loff_t end = pos + ret -1 ;int error2;xfs_iunlock (xip, iolock);if (need_i_mutex)mutex_unlock (&inode->i_mutex); - error2 =sync_page_range (inode, mapping, pos, ret); + + error2 =filemap_write_and_wait_range (mapping, pos, end);if (!error) error = error2;if (need_i_mutex)mutex_lock (&inode->i_mutex);xfs_ilock (xip, iolock); - error2 =xfs_write_sync_logforce (mp, xip); + + error2 =xfs_fsync (xip);if (!error) error = error2; }
fs/xfs/xfs_rw.c +0 -84
90 lines not on this page · raw message
fs/xfs/xfs_vnodeops.c +3 -8
@@ -611,7 +611,7 @@ xfs_fsync( xfs_inode_t *ip) {xfs_trans_t *tp; -int error; +int error =0 ;int log_flushed =0 , changed =1 ;xfs_itrace_entry (ip);
@@ -619,14 +619,9 @@ xfs_fsync(if (XFS_FORCED_SHUTDOWN (ip->i_mount))return XFS_ERROR (EIO ); -/* capture size updates in I/O completion before writing the inode. */ - error =xfs_wait_on_pages (ip,0 ,-1 ); -if (error) -return XFS_ERROR (error); -/* * We always need to make sure that the required inode state is safe on -* disk. The vnode might be clean but we still might need to force the +* disk. The inode might be clean but we still might need to force the * log because of committed transactions that haven't hit the disk yet. * Likewise, there could be unflushed non-transactional changes to the * inode core that have to go to disk and this requires us to issue
@@ -638,7 +633,7 @@ xfs_fsync(*/ xfs_ilock (ip,XFS_ILOCK_SHARED ); -if (!(ip->i_update_size || ip->i_update_core)) { +if (!ip->i_update_core) {/* * Timestamps/size haven't changed since last inode flush or * inode transaction commit. That means either nothing got
fs/xfs/xfs_rw.h +0 -1
@@ -68,7 +68,6 @@ xfs_get_extsz_hint(* Prototypes for functions in xfs_rw.c. */ extern int xfs_write_clear_setuid (struct xfs_inode *ip); -extern int xfs_write_sync_logforce (struct xfs_mount *mp,struct xfs_inode *ip);extern int xfs_bwrite (struct xfs_mount *mp,struct xfs_buf *bp);extern int xfs_bioerror (struct xfs_buf *bp);extern int xfs_bioerror_relse (struct xfs_buf *bp);
fs/xfs/xfs_trans.h +1 -1
@@ -68,7 +68,7 @@ typedef struct xfs_trans_header {#define XFS_TRANS_GROWFS 14#define XFS_TRANS_STRAT_WRITE 15#define XFS_TRANS_DIOSTRAT 16 -#define XFS_TRANS_WRITE_SYNC 17 +/* 17 was XFS_TRANS_WRITE_SYNC */ #define XFS_TRANS_WRITEID 18#define XFS_TRANS_ADDAFORK 19#define XFS_TRANS_ATTRINVAL 20
fs/xfs/linux-2.6/xfs_aops.c +0 -1
@@ -216,7 +216,6 @@ xfs_setfilesize(if (ip->i_d.di_size < isize) { ip->i_d.di_size = isize; ip->i_update_core =1 ; - ip->i_update_size =1 ;xfs_mark_inode_dirty_sync (ip); }
fs/xfs/xfs_iget.c +0 -1
@@ -82,7 +82,6 @@ xfs_inode_alloc(memset (&ip->i_df,0 ,sizeof (xfs_ifork_t)); ip->i_flags =0 ; ip->i_update_core =0 ; - ip->i_update_size =0 ; ip->i_delayed_blks =0 ;memset (&ip->i_d,0 ,sizeof (xfs_icdinode_t)); ip->i_size =0 ;
fs/xfs/xfs_inode.h +0 -1
@@ -261,7 +261,6 @@ typedef struct xfs_inode {/* Miscellaneous state. */ unsigned short i_flags;/* see defined flags below */ unsigned char i_update_core;/* timestamps/size is dirty */ -unsigned char i_update_size;/* di_size field is dirty */ unsigned int i_delayed_blks;/* count of delay alloc blks */ xfs_icdinode_t i_d;/* most of ondisk inode */
fs/xfs/xfs_inode_item.c +0 -8
@@ -263,14 +263,6 @@ xfs_inode_item_format( }/* -* We don't have to worry about re-ordering here because -* the update_size field is protected by the inode lock -* and we have that held in exclusive mode. -*/ -if (ip->i_update_size) - ip->i_update_size =0 ; - -/* * Make sure to get the latest atime from the Linux inode. */ xfs_synchronize_atime (ip);
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Quoted 24 lines · Christoph Hellwig
The guarantees for
O_SYNCare exactly the same as the ones we need to make for an fsync call (and given that LinuxO_SYNCisO_DSYNCthe equivalent is fdadatasync, but we treat both the same in XFS), except with a range data writeout. Jan Kara has started unifying these two path for filesystems using the generic helpers, and I've started to look at XFS.The actual transaction commited by
xfs_fsyncandxfs_write_sync_logforcehas a different transaction number, but actually is exactly the same. We'll only use the fsync transaction going forward. One major difference is thatxfs_write_sync_logforcenever issues a cache flush unless we commit a transaction causing that as a side-effect, which is an obvious bug in theO_SYNChandling. Second all the locking andi_update_sizevsi_update_corechanges from978b7237123d007b9fa983af6e0e2fa8f97f9934never made it toxfs_write_sync_logforce, so we add them back.To make
xfs_fsynceasily usable from theO_SYNCpath, thefilemap_fdatawaitcall is moved up toxfs_file_fsync, so that we don't wait on the whole file after we already waited for our portion inxfs_write.We'll also use a plain call to
filemap_write_and_wait_rangeinstead of the previoussync_page_rangwhich did it in two steps including an half-hearted inode write out that doesn't help us.
Quoted 5 lines
Once we're done with this also remove the now useless
i_update_sizetracking.Signed-off-by Christoph Hellwig <[email protected]>
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs