On Tue, Sep 5, 2017 at 2:07 PM, Eryu Guan [off-list ref] wrote:

On Wed, Aug 30, 2017 at 05:51:44PM +0300, Amir Goldstein wrote:

quoted

Using command line options --start-sector and --end-sector, only
operations acting on the specified target device range will be
replayed.

Single vebbose mode (-v) prints out only replayed operations.
Double verbose mode (-vv) prints out also skipped operations.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 src/log-writes/log-writes.c | 33 +++++++++++++++++++++++++++++++--
 src/log-writes/log-writes.h |  2 ++
 src/log-writes/replay-log.c | 31 +++++++++++++++++++++++++++++++
 3 files changed, 64 insertions(+), 2 deletions(-)

diff --git a/src/log-writes/log-writes.c b/src/log-writes/log-writes.c
index ba66a5c..d832c2a 100644
--- a/src/log-writes/log-writes.c
+++ b/src/log-writes/log-writes.c

@@ -119,6 +119,24 @@ int log_discard(struct log *log, struct log_write_entry *entry)

 /*
  * @log: the log we are replaying.
+ * @entry: entry to be replayed.
+ *
+ * @return: 0 if we should replay the entry, > 0 if we should skip it.
+ *
+ * Should we skip the entry in our log or replay onto the replay device.
+ */
+int log_should_skip(struct log *log, struct log_write_entry *entry)
+{
+     if (!entry->nr_sectors)
+             return 0;
+     if (entry->sector + entry->nr_sectors < log->start_sector ||
+         entry->sector > log->end_sector)

Seems values from entry can't be used directly, need le64_to_cpu first I
think.

quoted

+             return 1;
+     return 0;
+}
+
+/*
+ * @log: the log we are replaying.
  * @entry: where we put the entry.
  * @read_data: read the entry data as well, entry must be log->sectorsize sized
  * if this is set.

@@ -137,6 +155,7 @@ int log_replay_next_entry(struct log *log, struct log_write_entry *entry,
      char *buf;
      ssize_t ret;
      off_t offset;
+     u64 skip = 0;

int skip? and log_should_skip returns int too.

Right. Thanks

FYI, this is also a debugging option.
I used to to replay operations on a given range that was different between
good and bad buffers to narrow down the suspects.