4 badblock-guess: Quickly recover most of the data from a damaged disk
10 badblock-guess will try to find all readable sectors of the disk in minimal
11 time. It is similiar to:
12 dd if=<src_dev> of=<dst_dev> bs=512 conv=noerror,sync
14 but dd(1) solution can last for many weeks/months on heavily corrupted disk
15 media as each bad sector attempted to be read costs about 5secs (your disk
16 bad-blocks performance may significantly vary). badblock-guess tries to first
17 find and recover most of the data from healthy zones of the disk and later it
18 will start trying to recover smaller bits of information spread between/around
19 badblock zones - the principle of its operation is the fact that badblocks
20 usually occur in chunks. If you don't user-terminate badblock-guess and leave
21 it to finish completely itself, its execution time should be approx. the same
22 as for dd(1). You may also use it to guess the remaining execution time as
23 during its later execution phases it will be showing the remaining number of
24 sectors ("TODO=x") while most of them are probably badblocks in this phase of
27 This program will not recover data of any sector with failing read command.
28 If dd(1) command above finishes for you in a reasonable time, you don't need
29 this program. No special read methods are used. No vendor-specific dependencies
30 exist. No IDE, SCSI or any other specific device is required.
36 GNU GENERAL PUBLIC LICENSE, Version 2, June 1991.
38 See the file COPYING for its details. It is also available at:
39 http://www.gnu.org/licenses/gpl.txt
45 Syntax: badblock-guess <src_dev> [<dst_dev (OVERWRITTEN & DESTROYED!!!)>]
47 Possible device cases (watch out for <src_dev> vs. <dst_dev> differences!):
49 No <dst_dev> is specified: Just the <src_dev> is scanned for errors and
50 the detected badblocks list is output.
52 <src_dev> is a harddrive (/dev/hdc): Other harddrive of equal or higher
53 capacity is recommended
55 <src_dev> is a partition (/dev/hdc1): Partition on any other physical
56 drive in the system with the exactly same partition size is
59 <dst_dev> is a file: File must exist and it must be at least the size
60 as <src_dev>. If it is larger, the trailing data will remain
61 unchanged. You can create blank file of size 123456789 by:
62 dd if=/dev/zero of=/tmp/outimg bs=1 count=1 seek=$[123456789-1]
63 The size of the original <src_dev> can be queried by:
65 Be aware the size "sectors" is incorrectly the size in bytes,
66 depending on your Linux kernel version. dd(1) command above
67 requires the bytes size (use "bs=512" to use sectors value).
69 All the numbers are always expressed as sector (that means 512 bytes)
72 Never terminate badblock-guess by CTRL-C or kill(1) if you want to utilize its
73 results - always use 'f' followed by ENTER ('f' for 'finish'). You may not see
74 'f' while typing it - type it blindly (without quotes - just the one letter!).
75 During finishing the program will produce the badblocks list output to its
76 stdout while it will be clearing the 'bad' or 'not-yet-done' zones of the
77 target disk (if any specified).
79 TODO: Resuming of the operation from the badblocks list to continue the
80 scanning is not yet implemented.
86 Although not required it is strongly recommended to turn of drive readahead
87 during severe disk failure recoveries. You can use
89 /sbin/hdparm -A0a0 /dev/hdX (or /dev/sdX etc.)
91 for your drive to do it. You may need to install some extra package with
92 hdparm(8) for your Linux distribution.
94 Be aware that the disk performance will be critically hit - you may expect read
95 performance about 90KB/s (approx. 7.5GB/24hours if no badblocks read retrying
102 Linux kernels have internal blocksize 1KB but the partitions can be sized by
103 512 sectors and thus the last odd partition sector gets inaccessible.
104 This isn't a problem for a the disk devices as AFAIK all the disks have always
105 even number of sectors.
107 This program is aware of this flaw and it will not report last such sector as
108 BAD - just the appropriate warning is supplied (to stderr). This may be visible
109 for example while recovering NTFS partition with odd cylinder number (=>odd
110 sector count) as NTFS uses the last partition sector for its superblock backup.
111 Fortunately CHKDSK will fix it back, of course.
120 And you should have now the binary file "badblock-guess" compiled out.
121 You may need to install the following packages of your Linux distribution:
122 e2fsprogs, e2fsprogs-devel
123 glib, glib-devel (this is NOT glibc!)
124 other standard C compilation tools and libraries...
126 Compiled out binary is fully statically linked, you can bring it with you on
127 the floppy everywhere (running Linux is still required, of course!).
130 Operation description
131 ---------------------
133 During its run it will update its progress line. All the numbers are always
134 expressed as sector (that means 512 bytes) number/count.
136 @342342/819223,TODO=8192,bad=0,largest=8192,hunks=1
139 A=currently reding sector 342342 ...
140 B=... out of total sector 819223 of the disk (or partition)
141 C=8192 secters weren't yet attempted to be read
142 D=0 sectors were found with proof of read failure
143 E=currently read hunk of 8192 sectors
144 =also there is a maximum size of hunk 8192 sectors
145 F=total remaining hunks to be processed
147 Initially there is just one hunk (0-media_size) to be read. When no errors on
148 the disk are found, this one hunk is finished and no output (bad sectors list)
151 When we find bad a sector, we divide our todo-listed hunks by the schema:
153 > successfuly read 10 ->[1 bad]<todo 1><todo 2><todo 4><todo 9><todo 19>|end
155 We always read the biggest hunk "todo", "todo" hunks of the size are read in
156 backwards order (to approach the found bad blocks from the other side).
158 In this case the program would start with the status line (*):
160 @27/46,TODO=35,bad=1,largest=19,hunks=5
163 A=we start reading the first sector of the largest hunk <todo 19> (10+1+1+2+4+9)
164 B=total number of sectors on the disk (10+1+1+2+4+9+19)
165 C=still have to read all <todo> hunks (1+2+4+9+19)
166 D=we found just one confirmed bad sector <bad 1> (1)
167 E=the largest hunk "todo" on the disk is <todo 19> (19)
168 F=total number of the remaining "todo" hunks <t1><t2><t4><t9><t19> (1+1+1+1+1)
170 If you terminate badblock-guess run (by 'f' key, see above!), you can be sure
171 that you will loose the data of at most <largest> sectors after any bad sector
172 found. But you still can loose <TODO> sectors of data.
174 (*) The status line is printed once per a second and thus in real you can't
175 predict the exact states of the progress status line, of course.
178 Output badblock list format
179 ---------------------------
181 Output badblock list has form
183 first_sector-behind_last_sector <tab> ;<comments>
185 "behind_last_sector" is last_bad_sector+1. Thus single bad sector would be:
188 The following three types of output lines can occur. The first one is just
189 a shortcut for the second+third - such output is chosen when TODO blocks are
190 consequently following BAD blocks.
192 1000-9192 ;BAD=100, TODO(@1100-@9192)=8092