On Tue, Sep 5, 2017 at 2:03 PM, Eryu Guan [off-list ref] wrote:

On Wed, Aug 30, 2017 at 05:51:42PM +0300, Amir Goldstein wrote:

quoted

Imported Josef Bacik's code from:
https://github.com/josefbacik/log-writes.git

Specialized program for replaying a write log that was recorded by
device mapper log-writes target.  The tools is used to perform
crash consistency tests, allowing to run an arbitrary check tool
(fsck) at specified checkpoints in the write log.

[Amir:]
- Add project Makefile and SOURCE files
- Document the replay-log auxiliary program

Cc: Josef Bacik <redacted>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---

...

quoted

+static int zero_range(struct log *log, u64 start, u64 len)
+{
+     u64 bufsize = len;
+     ssize_t ret;
+     char *buf = NULL;
+
+     if (log->max_zero_size < len) {
+             if (log_writes_verbose)
+                     printf("discard len %llu larger than max %llu\n",
+                            (unsigned long long)len,
+                            (unsigned long long)log->max_zero_size);
+             return 0;
+     }
+
+     while (!buf) {
+             buf = malloc(sizeof(char) * len);

                                            ^^^^ shouldn't this be bufsize?

Yeh, look like is should be...
FYI, zero_range() is  used to emulate DISCARD that
was recorded on a device that supports DISCARD but then
replayed on a device that does not support DISCARD
The only time I tested this scenario is when I replayed lof to /dev/null.

quoted

+/*
+ * @log: the log we are manipulating.
+ * @entry_num: the entry we want.
+ *
+ * Seek to the given entry in the log, starting at 0 and ending at
+ * log->nr_entries - 1.
+ */
+int log_seek_entry(struct log *log, u64 entry_num)
+{
+     u64 i = 0;
+
+     if (entry_num >= log->nr_entries) {
+             fprintf(stderr, "Invalid entry number\n");
+             return -1;
+     }
+
+     if (lseek(log->logfd, log->sectorsize, SEEK_SET) == (off_t)-1) {
+             fprintf(stderr, "Error seeking in file: %d\n", errno);
+             return -1;
+     }

Hmm, we reset the log position to the first log entry by seeking to
log->sectorsize, shouldn't log->cur_entry be reset to 0 too? Though it
doesn't make any difference for now, because log_seek_entry() is only
called at init time, log->cur_entry is 0 anyway. But still, I think it
should be fixed.

True.

BTW, better to add some comments about the seek, it's not so obvious
it's seeking off the log super block on first read :)

...

quoted

+
+/*
+ * Basic info about the log for userspace.
+ */
+struct log_write_super {
+     __le64 magic;
+     __le64 version;
+     __le64 nr_entries;
+     __le32 sectorsize;
+};
+
+/*
+ * sector - the sector we wrote.
+ * nr_sectors - the number of sectors we wrote.
+ * flags - flags for this log entry.
+ * data_len - the size of the data in this log entry, this is for private log
+ * entry stuff, the MARK data provided by userspace for example.
+ */
+struct log_write_entry {
+     __le64 sector;
+     __le64 nr_sectors;
+     __le64 flags;
+     __le64 data_len;

This has to match the in-kernel log_write_entry structure, but the
data_len field is not used in this userspace program, better to add
comments to explain that.

OK. also should_stop() should strncmp() with data_len instead of strcmp
so there is a use for data_len...

quoted

+};
+
+#define LOG_IGNORE_DISCARD (1 << 0)
+#define LOG_DISCARD_NOT_SUPP (1 << 1)
+
+struct log {
+     int logfd;
+     int replayfd;
+     unsigned long flags;
+     u64 sectorsize;
+     u64 nr_entries;
+     u64 cur_entry;
+     u64 max_zero_size;
+     off_t cur_pos;

cur_pos is not used, can be removed?

I think it is best if I used it in patch
("replay-log: add validations for corrupt log entries")
every time I added lseek(log->logfd, 0, SEEK_CUR)
for printing offset in debug logs.