Hi,
On 18/01/18 16:04, Andreas Gruenbacher wrote:
Yes, that should make it a bit faster in that case, which seems like it is worth doing. Is there any measurable effect I wonder?Here's a revised version of Bob's gfs2_log_header_v2 patches from January 4: The first patch cleanus up get_log_header in a preliminary step. The second patch incorporates the following chages: * Fix a NULL pointer access when recovering a node's own journal (sdp->sd_qc_inode is NULL in gfs2_write_log_header in that case).* Move the call to gfs2_log_bmap out of gfs2_log_write. With that,lh_crc can be computed in gfs2_write_log_header and no separate gfs2_set_log_csum function is needed anymore. * Change lh_crc to only cover the rest of the block excluding the v1 log header: there is no point in crc-protecting the v1 header a second time. * Since lh_crc covers the rest of the block and not only a few header fields, it makes sense to use crc32c, which is much faster on architectures that implement it in hardware. The third patch is unchanged from Bob's last version.
Steve.
Thanks, Andreas Andreas Gruenbacher (1): gfs2: Get rid of gfs2_log_header_in Bob Peterson (2): GFS2: Introduce new gfs2_log_header_v2 GFS2: Log the reason for log flushes in every log header fs/gfs2/aops.c | 3 +- fs/gfs2/file.c | 4 +- fs/gfs2/glops.c | 19 ++++++--- fs/gfs2/incore.h | 1 - fs/gfs2/log.c | 83 +++++++++++++++++++++++++++++----------- fs/gfs2/log.h | 12 ++---- fs/gfs2/lops.c | 16 ++++---- fs/gfs2/lops.h | 3 ++ fs/gfs2/ops_fstype.c | 2 +- fs/gfs2/quota.c | 3 +- fs/gfs2/recovery.c | 57 ++++++++++++--------------- fs/gfs2/rgrp.c | 3 +- fs/gfs2/super.c | 13 +++++-- fs/gfs2/trace_gfs2.h | 11 ++++-- fs/gfs2/trans.c | 3 +- include/uapi/linux/gfs2_ondisk.h | 46 +++++++++++++++++++++- 16 files changed, 187 insertions(+), 92 deletions(-)
