[Bug-ddrescue] Bad blocks skipping are too slow

[Alexander Sashnov]

Hello,

I'm trying to recover data from 1Tb seagate freeagent disk 
with following command
(I use new 1.5 Tb disk as the destination):

# ddrescue -d -f -n -v -i25G --skip-size=1M --try-again 
/dev/sda /dev/sdc rescue.log
--------------
About to copy 975204 MBytes from /dev/sda to /dev/sdc
    Starting positions: infile = 25000 MB, outfile = 
25000 MB
    Copy block size: 128 sectors Initial skip size: 1954 
sectors
Sector size: 512 Bytes

Press Ctrl-C to interrupt
Initial status (read from logfile)
rescued: 17812 MB, errsize: 2605 MB, errors: 1834801
...
--------------


Command was ran at 16:00. At 16:23 I've seen at console 
following:

--------------
Current status
rescued: 17812 MB, errsize: 2605 MB, current rate: 0 B/s
   ipos: 25022 MB, errors: 1834792, average rate:       0 
B/s
   opos: 25022 MB, time since last successful read: 
 24.5 m
--------------


My concern is 'too slow progress' here.

Also I could see following in /var/log/kern.log (Ubuntu 
12.04 x64, 3.2.0-37-generic #58-Ubuntu SMP)

Apr 1 16:00:42 umbrella kernel: [13016.728404] 
end_request: I/O error, dev sda, sector 48871829
Apr 1 16:03:02 umbrella kernel: [13156.564408] 
end_request: I/O error, dev sda, sector 48871954
Apr 1 16:05:25 umbrella kernel: [13299.572396] 
end_request: I/O error, dev sda, sector 48871956
Apr 1 16:07:30 umbrella kernel: [13424.540397] 
end_request: I/O error, dev sda, sector 48871958
Apr 1 16:09:47 umbrella kernel: [13562.064398] 
end_request: I/O error, dev sda, sector 48871959
Apr 1 16:11:35 umbrella kernel: [13669.324543] 
end_request: I/O error, dev sda, sector 48872080
Apr 1 16:14:02 umbrella kernel: [13816.340532] 
end_request: I/O error, dev sda, sector 48872082
Apr 1 16:16:05 umbrella kernel: [13939.336397] 
end_request: I/O error, dev sda, sector 48872084
Apr 1 16:18:32 umbrella kernel: [14086.976381] 
end_request: I/O error, dev sda, sector 48872086
Apr 1 16:20:56 umbrella kernel: [14230.372555] 
end_request: I/O error, dev sda, sector 48872087

Why it tries sectors with too small distance???

If I use 128 sectors as cluster size and 1M skip block I 
wouldn't expected to see IO errors nearer than 128 sectors 
between them.


My suggestion is to use ioctl(SGIO,...)  as 'hdparm 
--read-sector' does.
Most block devices are support SCSI protocol directly or 
via libata in Linux kernel.

---
Alexander Sashnov.