Next: , Up: (dir)

Lziprecover Manual

This manual is for Lziprecover (version 1.24, 20 January 2024).


Copyright © 2009-2024 Antonio Diaz Diaz.

This manual is free documentation: you have unlimited permission to copy, distribute, and modify it.


Next: , Previous: Top, Up: Top

1 Introduction

Lziprecover is a data recovery tool and decompressor for files in the lzip compressed data format (.lz). Lziprecover is able to repair slightly damaged files (up to one single-byte error per member), produce a correct file by merging the good parts of two or more damaged copies, reproduce a missing (zeroed) sector using a reference file, extract data from damaged files, decompress files, and test integrity of files.

Lziprecover can remove the damaged members from multimember files, for example multimember tar.lz archives.

Lziprecover provides random access to the data in multimember files; it only decompresses the members containing the desired data.

Lziprecover facilitates the management of metadata stored as trailing data in lzip files.

Lziprecover is not a replacement for regular backups, but a last line of defense for the case where the backups are also damaged.

The lzip file format is designed for data sharing and long-term archiving, taking into account both data integrity and decoder availability:

A nice feature of the lzip format is that a corrupt byte is easier to repair the nearer it is from the beginning of the file. Therefore, with the help of lziprecover, losing an entire archive just because of a corrupt byte near the beginning is a thing of the past.

Compression may be good for long-term archiving. For compressible data, multiple compressed copies may provide redundancy in a more useful form and may have a better chance of surviving intact than one uncompressed copy using the same amount of storage space. This is especially true if the format provides recovery capabilities like those of lziprecover, which is able to find and combine the good parts of several damaged copies.

Lziprecover is able to recover or decompress files produced by any of the compressors in the lzip family: lzip, plzip, minilzip/lzlib, clzip, and pdlzip.

If the cause of file corruption is a damaged medium, the combination GNU ddrescue + lziprecover is the recommended option for recovering data from damaged lzip files. See ddrescue-example, and ddrescue-example2, for examples.

If a file is too damaged for lziprecover to repair it, all the recoverable data in all members of the file can be extracted with the following command (the resulting file may contain errors and some garbage data may be produced at the end of each damaged member):

     lziprecover -cd --ignore-errors file.lz > file

When recovering data, lziprecover takes as arguments the names of the damaged files and writes zero or more recovered files depending on the operation selected and whether the recovery succeeded or not. The damaged files themselves are kept unchanged.

When decompressing or testing file integrity, lziprecover behaves like lzip or lunzip.

LANGUAGE NOTE: Uncompressed = not compressed = plain data; it may never have been compressed. Decompressed is used to refer to data which have undergone the process of decompression.


Next: , Previous: Introduction, Up: Top

2 Invoking lziprecover

The format for running lziprecover is:

     lziprecover [options] [files]

When decompressing or testing, a hyphen '-' used as a file argument means standard input. It can be mixed with other files and is read just once, the first time it appears in the command line. If no file names are specified, lziprecover decompresses from standard input to standard output. Remember to prepend ./ to any file name beginning with a hyphen, or use '--'.

lziprecover supports the following options:

-h
--help
Print an informative help message describing the options and exit.
-V
--version
Print the version number of lziprecover on the standard output and exit. This version number should be included in all bug reports.


-a
--trailing-error
Exit with error status 2 if any remaining input is detected after decompressing the last member. Such remaining input is usually trailing garbage that can be safely ignored. See concat-example.
-A
--alone-to-lz
Convert lzma-alone files to lzip format without recompressing, just adding a lzip header and trailer. The conversion minimizes the dictionary size of the resulting file (and therefore the amount of memory required to decompress it). Only streamed files with default LZMA properties can be converted; non-streamed lzma-alone files lack the "End Of Stream" marker required in lzip files.

The name of the converted lzip file is derived from that of the original lzma-alone file as follows:

filename.lzma becomes filename.lz
filename.tlz becomes filename.tar.lz
anyothername becomes anyothername.lz

-c
--stdout
Write decompressed data to standard output; keep input files unchanged. This option (or -o) is needed when reading from a named pipe (fifo) or from a device. Use it also to recover as much of the decompressed data as possible when decompressing a corrupt file. -c overrides -o. -c has no effect when merging, removing members, repairing, reproducing, splitting, testing or listing.
-d
--decompress
Decompress the files specified. The integrity of the files specified is checked. If a file does not exist, can't be opened, or the destination file already exists and --force has not been specified, lziprecover continues decompressing the rest of the files and exits with error status 1. If a file fails to decompress, or is a terminal, lziprecover exits immediately with error status 2 without decompressing the rest of the files. A terminal is considered an uncompressed file, and therefore invalid.
-D range
--range-decompress=range
Decompress only a range of bytes starting at decompressed byte position begin and up to byte position end - 1. Byte positions start at 0. This option provides random access to the data in multimember files; it only decompresses the members containing the desired data. In order to guarantee the correctness of the data produced, all members containing any part of the desired data are decompressed and their integrity is checked.

Four formats of range are recognized, 'begin', 'begin-end', 'begin,size', and ',size'. If only begin is specified, end is taken as the end of the file. If only size is specified, begin is taken as the beginning of the file. The bytes produced are sent to standard output unless the option --output is used.


-e
--reproduce
Try to recover a missing (zeroed) sector in file using a reference file and the same version of lzip that created file. If successful, a repaired copy is written to the file file_fixed.lz. file is not modified at all. The exit status is 0 if the member containing the zeroed sector could be repaired, 2 otherwise. Note that file_fixed.lz may still contain errors in the members following the one repaired. See Reproducing one sector, for a complete description of the reproduce mode.
--lzip-level=digit|a|m[length]
Try only the given compression level or match length limit when reproducing a zeroed sector. --lzip-level=a tries all the compression levels (0 to 9), while --lzip-level=m tries all the match length limits (5 to 273).
--lzip-name=name
Set the name of the lzip executable used by --reproduce. If --lzip-name is not specified, 'lzip' is used.
--reference-file=file
Set the reference file used by --reproduce. It must contain the uncompressed data corresponding to the missing compressed data of the zeroed sector, plus some context data before and after them.
-f
--force
Force overwrite of output files.
-i
--ignore-errors
Make --decompress, --test, and --range-decompress ignore format and data errors and continue decompressing the remaining members in the file; keep input files unchanged. For example, the commands 'lziprecover -cd -i file.lz > file' or 'lziprecover -D0 -i file.lz > file' decompress all the recoverable data in all members of 'file.lz' without having to split it first. The '-cd -i' method resyncs to the next member header after each error, and is immune to some format errors that make '-D0 -i' fail. The range decompressed may be smaller than the range requested, because of the errors. The exit status is set to 0 unless other errors are found (I/O errors, for example).

Make --list, --dump, --remove, and --strip ignore format errors. The sizes of the members with errors (especially the last) may be wrong.

-k
--keep
Keep (don't delete) input files during decompression.
-l
--list
Print the uncompressed size, compressed size, and percentage saved of the files specified. Trailing data are ignored. The values produced are correct even for multimember files. If more than one file is given, a final line containing the cumulative sizes is printed. With -v, the dictionary size, the number of members in the file, and the amount of trailing data (if any) are also printed. With -vv, the positions and sizes of each member in multimember files are also printed. With -i, format errors are ignored, and with -ivv, gaps between members are shown. The member numbers shown coincide with the file numbers produced by --split.

If any file is damaged, does not exist, can't be opened, or is not regular, the final exit status is > 0. -lq can be used to check quickly (without decompressing) the structural integrity of the files specified. (Use --test to check the data integrity). -alq additionally checks that none of the files specified contain trailing data.

-m
--merge
Try to produce a correct file by merging the good parts of two or more damaged copies. If successful, a repaired copy is written to the file file_fixed.lz. The exit status is 0 if a correct file could be produced, 2 otherwise. See Merging files, for a complete description of the merge mode.
-o file
--output=file
Place the repaired output into file instead of into file_fixed.lz. If splitting, the names of the files produced are in the form 'rec01file', 'rec02file', etc.

If -c has not been also specified, write the (de)compressed output to file, automatically creating any missing parent directories; keep input files unchanged. This option (or -c) is needed when reading from a named pipe (fifo) or from a device. -o - is equivalent to -c. -o has no effect when testing or listing.

-q
--quiet
Quiet operation. Suppress all messages.


-R
--byte-repair
Try to repair a file with small errors (up to one single-byte error per member). If successful, a repaired copy is written to the file file_fixed.lz. file is not modified at all. The exit status is 0 if the file could be repaired, 2 otherwise. See Repairing one byte, for a complete description of the repair mode.
-s
--split
Search for members in file and write each member in its own file. Gaps between members are detected and each gap is saved in its own file. Trailing data (if any) are saved alone in the last file. You can then use 'lziprecover -t' to test the integrity of the resulting files, decompress those which are undamaged, and try to repair or partially decompress those which are damaged. Gaps may contain garbage or may be members with corrupt headers or trailers. If other lziprecover functions fail to work on a multimember file because of damage in headers or trailers, try to split file and then work on each member individually.

The names of the files produced are in the form 'rec01file', 'rec02file', etc, and are designed so that the use of wildcards in subsequent processing, for example, 'lziprecover -cd rec*file > recovered_data', processes the files in the correct order. The number of digits used in the names varies depending on the number of members in file.

-t
--test
Check integrity of the files specified, but don't decompress them. This really performs a trial decompression and throws away the result. Use it together with -v to see information about the files. If a file fails the test, does not exist, can't be opened, or is a terminal, lziprecover continues testing the rest of the files. A final diagnostic is shown at verbosity level 1 or higher if any file fails the test when testing multiple files.
-v
--verbose
Verbose mode.
When decompressing or testing, further -v's (up to 4) increase the verbosity level, showing status, compression ratio, dictionary size, trailer contents (CRC, data size, member size), and up to 6 bytes of trailing data (if any) both in hexadecimal and as a string of printable ASCII characters.
Two or more -v options show the progress of decompression.
In other modes, increasing verbosity levels show final status, progress of operations, and extra information (for example, the failed areas).
--dump=[member_list][:damaged][:empty][:tdata]
Dump the members listed, the damaged members (if any), the empty members (if any), or the trailing data (if any) of one or more regular multimember files to standard output, or to a file if the option --output is used. If more than one file is given, the elements dumped from all the files are concatenated. If a file does not exist, can't be opened, or is not regular, lziprecover continues processing the rest of the files. If the dump fails in one file, lziprecover exits immediately without processing the rest of the files. Only --dump=tdata can write to a terminal. --dump=damaged implies --ignore-errors.

The argument to --dump is a colon-separated list of the following element specifiers; a member list (1,3-6), a reverse member list (r1,3-6), and the strings "damaged", "empty", and "tdata" (which may be shortened to 'd', 'e', and 't' respectively). A member list selects the members (or gaps) listed, whose numbers coincide with those shown by --list. A reverse member list selects the members listed counting from the last member in the file (r1). Negated versions of both kinds of lists exist (^1,3-6:r^1,3-6) which select all the members except those in the list. The strings "damaged", "empty", and "tdata" select the damaged members, the empty members (those with a data size = 0), and the trailing data respectively. If the same member is selected more than once, for example by '1:r1' in a single-member file, it is dumped just once. See the following examples:

--dump argument Elements dumped
1,3-6 members 1, 3, 4, 5, 6
r1-3 last 3 members in file
^13,15 all but 13th and 15th members in file
r^1 all but last member in file
damaged all damaged members in file
empty all empty members in file
tdata trailing data
1-5:r1:tdata members 1 to 5, last member, trailing data
damaged:tdata damaged members, trailing data
3,12:damaged:tdata members 3, 12, damaged members, trailing data

--remove=[member_list][:damaged][:empty][:tdata]
Remove the members listed, the damaged members (if any), the empty members (if any), or the trailing data (if any) from regular multimember files in place. The date of each file modified is preserved if possible. If all members in a file are selected to be removed, the file is left unchanged and the exit status is set to 2. If a file does not exist, can't be opened, is not regular, or is left unchanged, lziprecover continues processing the rest of the files. In case of I/O error, lziprecover exits immediately without processing the rest of the files. See --dump above for a description of the argument.

This option may be dangerous even if only the trailing data are being removed because the file may be corrupt or the trailing data may contain a forbidden combination of characters. See Trailing data. It is safer to send the output of --strip to a temporary file, check it, and then copy it over the original file. But if you prefer --remove because of its more efficient in-place removal, it is advisable to make a backup before attempting the removal. At least check that 'lzip -cd file.lz | wc -c' and the uncompressed size shown by 'lzip -l file.lz' match before attempting the removal of trailing data.

--strip=[member_list][:damaged][:empty][:tdata]
Copy one or more regular multimember files to standard output (or to a file if the option --output is used), stripping the members listed, the damaged members (if any), the empty members (if any), or the trailing data (if any) from each file. If all members in a file are selected to be stripped, the trailing data (if any) are also stripped even if 'tdata' is not specified. If more than one file is given, the files are concatenated. In this case the trailing data are also stripped from all but the last file even if 'tdata' is not specified. If a file does not exist, can't be opened, or is not regular, lziprecover continues processing the rest of the files. If a file fails to copy, lziprecover exits immediately without processing the rest of the files. See --dump above for a description of the argument.
--empty-error
Exit with error status 2 if any empty member is found in the input files.
--marking-error
Exit with error status 2 if the first LZMA byte is non-zero in any member of the input files. This may be caused by data corruption or by deliberate insertion of tracking information in the file. Use 'lziprecover --clear-marking' to clear any such non-zero bytes.
--loose-trailing
When decompressing, testing, or listing, allow trailing data whose first bytes are so similar to the magic bytes of a lzip header that they can be confused with a corrupt header. Use this option if a file triggers a "corrupt header" error and the cause is not indeed a corrupt header.
--clear-marking
Set to zero the first LZMA byte of each member in the files specified. At verbosity level 1 (-v), print the number of members cleared. The date of each file modified is preserved if possible. This option exists because the first byte of the LZMA stream is ignored by the range decoder, and can therefore be (mis)used to store any value which can then be used as a watermark to track the path of the compressed payload.

Lziprecover also supports the following debug options (for experts):

-E range[,sector_size]
--debug-reproduce=range[,sector_size]
Load the compressed file into memory, set all bytes in the positions specified by range to 0, and try to reproduce a correct compressed file. See --reproduce. See range-format, for a description of range. If a sector_size is specified, set each sector to 0 in sequence and try to reproduce the file, printing to standard output final statistics of the number of sectors reproduced successfully. Exit with nonzero status only in case of fatal error.
-M
--md5sum
Print to standard output the MD5 digests of the input files one per line in the same format produced by the md5sum tool. Lziprecover uses MD5 digests to check the result of some operations. This option can be used to test the correctness of lziprecover's implementation of the MD5 algorithm.
-S[value]
--nrep-stats[=value]
Compare the frequency of sequences of N repeated bytes of a given value in the compressed LZMA streams of the input files with the frequency expected for random data (1 / 2^(8N)). If value is not specified, print the frequency of repeated sequences of all possible byte values. Print cumulative data for all the files, followed by the name of the first file with the longest sequence.


-U 1|Bsize
--unzcrash=1|Bsize
With argument '1', test 1-bit errors in the LZMA stream of the compressed input file like the command 'unzcrash -b1 -p7 -s-20 'lzip -t' file' but in memory, and therefore much faster (30 to 50 times faster). See Unzcrash. This option tests all the members independently in a multimember file, skipping headers and trailers. If a decompression succeeds, the decompressed output is compared with the decompressed output of the original file using MD5 digests. file must not contain errors and must decompress correctly for the comparisons to work.

With argument 'B', test zeroed sectors (blocks of bytes) in the LZMA stream of the compressed input file like the command 'unzcrash --block=size -d1 -p7 -s-(size+20) 'lzip -t' file' but in memory, and therefore much faster. Testing and comparisons work just like with the argument '1' explained above.

By default --unzcrash only prints the interesting cases; CRC mismatches, size mismatches, unsupported marker codes, unexpected EOFs, apparently successful decompressions, and decoder errors detected 50_000 or more bytes beyond the byte (or the start of the block) being tested. At verbosity level 1 (-v) it also prints decoder errors detected 10_000 or more bytes beyond the byte being tested. At verbosity level 2 (-vv) it prints all cases for 1-bit errors or the decoder errors detected beyond the end of the block for zeroed blocks.

-W position,value
--debug-decompress=position,value
Load the compressed file into memory, set the byte at position to value, and decompress the modified compressed data to standard output. If the damaged member can be decompressed to the end (just fails with a CRC mismatch), the members following it are also decompressed.
-X[position,value]
--show-packets[=position,value]
Load the compressed file into memory, optionally set the byte at position to value, decompress the modified compressed data (discarding the output), and print to standard output descriptions of the LZMA packets being decoded.
-Y range
--debug-delay=range
Load the compressed file into memory and then repeatedly decompress it, increasing 256 times each byte of the subset of the compressed data positions specified by range, so as to test all possible one-byte errors. For each decompression error find the error detection delay and print to standard output the maximum delay. The error detection delay is the difference between the position of the error and the position where the decoder realized that the data contains an error. See range-format, for a description of range.
-Z position,value
--debug-byte-repair=position,value
Load the compressed file into memory, set the byte at position to value, and then try to repair the byte error. See --byte-repair.

Numbers given as arguments to options may be expressed in decimal, hexadecimal, or octal (using the same syntax as integer constants in C++), and may be followed by a multiplier and an optional 'B' for "byte".

Table of SI and binary prefixes (unit multipliers):

Prefix Value | Prefix Value
k kilobyte (10^3 = 1000) | Ki kibibyte (2^10 = 1024)
M megabyte (10^6) | Mi mebibyte (2^20)
G gigabyte (10^9) | Gi gibibyte (2^30)
T terabyte (10^12) | Ti tebibyte (2^40)
P petabyte (10^15) | Pi pebibyte (2^50)
E exabyte (10^18) | Ei exbibyte (2^60)
Z zettabyte (10^21) | Zi zebibyte (2^70)
Y yottabyte (10^24) | Yi yobibyte (2^80)
R ronnabyte (10^27) | Ri robibyte (2^90)
Q quettabyte (10^30) | Qi quebibyte (2^100)


Exit status: 0 for a normal exit, 1 for environmental problems (file not found, invalid command-line options, I/O errors, etc), 2 to indicate a corrupt or invalid input file, 3 for an internal consistency error (e.g., bug) which caused lziprecover to panic.


Next: , Previous: Invoking lziprecover, Up: Top

3 Protecting data from accidental loss

It is a fact of life that sometimes data becomes corrupt. Software has errors. Hardware may misbehave or fail. RAM may be struck by a cosmic ray. This is why a safe enough integrity checking is needed in compressed formats, and the reason why a data recovery tool is sometimes needed.

There are 3 main types of data corruption that may cause data loss: single-byte errors, multibyte errors (generally affecting a whole sector in a block device), and total device failure.

Lziprecover protects natively against single-byte errors as long as file integrity is checked frequently enough that a second single-byte error does not develop in the same member before the first one is repaired. See Repairing one byte.

Lziprecover also protects against multibyte errors if at least one backup copy of the file is made (see Merging files), or if the error is a zeroed sector and the uncompressed data corresponding to the zeroed sector are available (see Reproducing one sector). If you can choose between merging and reproducing, try merging first because it is usually faster, easier to use, and has a high probability of success.

Lziprecover can't help in case of device failure. The only remedy for total device failure is storing backup copies in separate media.

The extraordinary safety of the lzip format allows lziprecover to exploit the redundance that occurrs naturally when making compressed backups. Lziprecover can recover data that would not be recoverable from files compressed in other formats. Let's see two examples of how much better is lzip compared with gzip and bzip2 with respect to data safety:


Next: , Up: Data safety

3.1 Recovering a file using a damaged backup

Let's suppose that you made a compressed backup of your valuable scientific data and stored two copies on separate media. Years later you notice that both copies are corrupt.

If you compressed the data with gzip and both copies suffer any damage in the data stream, even if it is just one altered bit, the original data can only be recovered by an expert, if at all.

If you used bzip2, and if the file is large enough to contain more than one compressed data block (usually larger than 900 kB uncompressed), and if no block is damaged in both files, then the data can be manually recovered by splitting the files with bzip2recover, checking every block, and then copying the right blocks in the right order into another file.

But if you used lzip, the data can be automatically recovered with 'lziprecover --merge' as long as the damaged areas don't overlap.

Note that each error in a bzip2 file makes a whole block unusable, but each error in a lzip file only affects the damaged bytes, making it possible to recover a file with thousands of errors.


Previous: Merging with a backup, Up: Data safety

3.2 Recovering new messages using an old backup

Let's suppose that you make periodic backups of your email messages stored in one or more mailboxes. (A mailbox is a file containing a possibly large number of email messages). New messages are appended to the end of each mailbox, therefore the initial part of two consecutive backups is identical unless some messages have been changed or deleted in the meantime. The new messages added to each backup are usually a small part of the whole mailbox.

+============================================+
| Older backup containing some messages      |
+============================================+
+============================================+========================+
| Newer backup containing the messages above | plus some new messages |
+============================================+========================+

One day you discover that your mailbox has disappeared because you deleted it inadvertently or because of a bug in your email reader. Not only that. You need to recover a recent message, but the last backup you made of the mailbox (the newer backup above) has lost the data corresponding to a whole sector because of an I/O error in the part containing the old messages.

If you compressed the mailbox with gzip, usually none of the new messages can be recovered even if they are intact because all the data beyond the missing sector can't be decoded.

If you used bzip2, and if the newer backup is large enough that the new messages are in a different compressed data block than the one damaged (usually larger than 900 kB uncompressed), then you can recover the new messages manually with bzip2recover. If the backups are identical except for the new messages appended, you may even recover the whole newer backup by combining the good blocks from both backups.

But if you used lzip, the whole newer backup can be automatically recovered with 'lziprecover --reproduce' as long as the missing bytes can be recovered from the older backup, even if other messages in the common part have been changed or deleted. Mailboxes seem to be especially easy to reproduce. The probability of reproducing a mailbox (see performance-of-reproduce) is almost as high as that of merging two identical backups (see performance-of-merge).


Next: , Previous: Data safety, Up: Top

4 Repairing one byte

Lziprecover can repair perfectly most files with small errors (up to one single-byte error per member), without the need of any extra redundance at all. If the reparation is successful, the repaired file is identical bit for bit to the original. This makes lzip files resistant to bit flip, one of the most common forms of data corruption.

The file is repaired in memory. Therefore, enough virtual memory (RAM + swap) to contain the largest damaged member is required.

The error may be located anywhere in the file except in the first 5 bytes of each member header or in the 'Member size' field of the trailer (last 8 bytes of each member). If the error is in the header it can be easily repaired with a text editor like GNU Moe (see File format). If the error is in the member size, it is enough to ignore the message about 'bad member size' when decompressing.

Bit flip happens when one bit in the file is changed from 0 to 1 or vice versa. It may be caused by bad RAM or even by natural radiation. I have seen a case of bit flip in a file stored on an USB flash drive.

One byte may seem small, but most file corruptions not produced by transmission errors or I/O errors just affect one byte, or even one bit, of the file. Also, unlike magnetic media, where errors usually affect a whole sector, solid-state storage devices tend to produce single-byte errors, making of lzip the perfect format for data stored on such devices.

Repairing a file can take some time. Small files or files with the error located near the beginning can be repaired in a few seconds. But repairing a large file compressed with a large dictionary size and with the error located far from the beginning, may take hours.

On the other hand, errors located near the beginning of the file cause much more loss of data than errors located near the end. So lziprecover repairs more efficiently the worst errors.


Next: , Previous: Repairing one byte, Up: Top

5 Merging files

If you have several copies of a file but all of them are too damaged to repair them individually (see Repairing one byte), lziprecover can try to produce a correct file by merging the good parts of the damaged copies.

The merge may succeed even if some copies of the file have all the headers and trailers damaged, as long as there is at least one copy of every header and trailer intact, even if they are in different copies of the file.

The merge fails if the damaged areas overlap (at least one byte is damaged in all copies), or are adjacent and the boundary can't be determined, or if the copies have too many damaged areas.

All the copies to be merged must have the same size. If any of them is larger or smaller than it should, either because it has been truncated or because it got some garbage data appended at the end, it can be brought to the correct size with the following command before merging it with the other copies:

     ddrescue -s<correct_size> -x<correct_size> file.lz correct_size_file.lz

To give you an idea of its possibilities, when merging two copies, each of them with one damaged area affecting 1 percent of the copy, the probability of obtaining a correct file is about 98 percent. With three such copies the probability rises to 99.97 percent. For large files (a few MB) with small errors (one sector damaged per copy), the probability approaches 100 percent even with only two copies. (Supposing that the errors are randomly located inside each copy).

Some types of solid-state device (NAND flash, for example) can produce bursts of scattered single-bit errors. Lziprecover is able to merge files with thousands of such scattered errors by grouping the errors into clusters and then merging the files as if each cluster were a single error.

Here is a real case of successful merging. Two copies of the file 'icecat-3.5.3-x86.tar.lz' (compressed size 9 MB) became corrupt while stored on the same NAND flash device. One of the copies had 76 single-bit errors scattered in an area of 1020 bytes, and the other had 3028 such errors in an area of 31729 bytes. Lziprecover produced a correct file, identical to the original, in just 5 seconds:

     lziprecover -vvm a/icecat-3.5.3-x86.tar.lz b/icecat-3.5.3-x86.tar.lz
     Merging member 1 of 1  (2552 errors)
       2552 errors have been grouped in 16 clusters.
       Trying variation 2 of 2, block 2
     Input files merged successfully.

Note that the number of errors reported by lziprecover (2552) is lower than the number of corrupt bytes (3104) because contiguous corrupt bytes are counted as a single multibyte error.


Example 1: Recover a compressed backup from two copies on CD-ROM with error-checked merging of copies. See the ddrescue manual for details about ddrescue.
     ddrescue -d -r1 -b2048 /dev/cdrom cdimage1 mapfile1
     mount -t iso9660 -o loop,ro cdimage1 /mnt/cdimage
     cp /mnt/cdimage/backup.tar.lz rescued1.tar.lz
     umount /mnt/cdimage
       (insert second copy in the CD drive)
     ddrescue -d -r1 -b2048 /dev/cdrom cdimage2 mapfile2
     mount -t iso9660 -o loop,ro cdimage2 /mnt/cdimage
     cp /mnt/cdimage/backup.tar.lz rescued2.tar.lz
     umount /mnt/cdimage
     lziprecover -m -v -o backup.tar.lz rescued1.tar.lz rescued2.tar.lz
       Input files merged successfully.
     lziprecover -tv backup.tar.lz
       backup.tar.lz: ok

Example 2: Recover the first volume of those created with the command 'lzip -b 32MiB -S 650MB big_db' from two copies, 'big_db1_00001.lz' and 'big_db2_00001.lz', with member 07 damaged in the first copy, member 18 damaged in the second copy, and member 12 damaged in both copies. The correct file produced is saved in 'big_db_00001.lz'.
     lziprecover -m -v -o big_db_00001.lz big_db1_00001.lz big_db2_00001.lz
       Input files merged successfully.
     lziprecover -tv big_db_00001.lz
       big_db_00001.lz: ok


Next: , Previous: Merging files, Up: Top

6 Reproducing one sector

Lziprecover can recover a zeroed sector in a lzip file by concatenating the decompressed contents of the file up to the beginning of the zeroed sector and the uncompressed data corresponding to the zeroed sector, and then feeding the concatenated data to the same version of lzip that created the file. For this to work, a reference file is required containing the uncompressed data corresponding to the missing compressed data of the zeroed sector, plus some context data before and after them. It is possible to recover a large file using just a few kB of reference data.

The difficult part is finding a suitable reference file. It must contain the exact data required (possibly mixed with other data). Containing similar data is not enough.

A zeroed sector may be caused by the incomplete recovery of a damaged storage device (with I/O errors) using, for example, ddrescue. The reproduction can't be done if the zeroed sector overlaps with the first 15 bytes of a member, or if the zeroed sector is smaller than 8 bytes.

The file is reproduced in memory. Therefore, enough virtual memory (RAM + swap) to contain the damaged member is required.

To understand how it works, take any lzipped file, say 'foo.lz', decompress it (keeping the original), and try to reproduce an artificially zeroed sector in it by running the following commands:

     lzip -kd foo.lz
     lziprecover -vv --debug-reproduce=65536,512 --reference-file=foo foo.lz

which should produce an output like the following:

     Reproducing:    foo.lz
     Reference file: foo
     Testing sectors of size 512 at file positions 65536 to 66047
       (master mpos = 65536, dpos = 296892)
     foo: Match found at offset 296892
     Reproduction succeeded at pos 65536
     
             1 sectors tested
             1 reproductions returned with zero status
               all comparisons passed

Using 'foo' as reference file guarantees that any zeroed sector in 'foo.lz' can be reproduced because both files contain the same data. In real use, the reference file needs to contain the data corresponding to the zeroed sector, but the rest of the data (if any) may differ between both files. The reference data may be obtained from the partial decompression of the damaged file itself if it contains repeated data. For example if the damaged file is a compressed tarball containing several partially modified versions of the same file.

The offset reported by lziprecover is the position in the reference file of the first byte that could not be decompressed. This is the first byte that will be compressed to reproduce the zeroed sector.

The reproduce mode tries to reproduce the missing compressed data originally present in the zeroed sector. It is based on the perfect reproducibility of lzip files (lzip produces identical compressed output from identical input). Therefore, the same version of lzip that created the file to be reproduced should be used to reproduce the zeroed sector. Near versions may also work because the output of lzip changes infrequently. If reproducing a tar.lz archive created with tarlz, the version of lzip, clzip, or minilzip corresponding to the version of the lzlib library used by tarlz to create the archive should be used.

When recovering a tar.lz archive and using as reference a file from the filesystem, if the zeroed sector encodes (part of) a tar header, the archive can't be reproduced. Therefore, the less overhead (smaller headers) a tar archive has, the more probable is that the zeroed sector does not include a header, and that the archive can be reproduced. The tarlz format has minimum overhead. It uses basic ustar headers, and only adds extended pax headers when they are required.

6.1 Performance of --reproduce

Reproduce mode is especially useful when recovering a corrupt backup (or a corrupt source tarball) that is part of a series. Usually only a small fraction of the data changes from one backup to the next or from one version of a source tarball to the next. This makes sometimes possible to reproduce a given corrupted version using reference data from a near version. The following two tables show the fraction of reproducible sectors (reproducible sectors divided by total sectors in archive) for some archives, using sector sizes of 512 and 4096 bytes. 'mailbox-aug.tar.lz' is a backup of some of my mailboxes. 'backup-feb.tar.lz' and 'backup-apr.tar.lz' are real backups of my own working directory:

Reference file File Reproducible (512)
backup-feb.tar backup-apr.tar.lz 3273 / 4342 = 75.38%
backup-apr.tar backup-feb.tar.lz 3259 / 4161 = 78.32%
gawk-5.0.0.tar gawk-5.0.1.tar.lz 4369 / 5844 = 74.76%
gawk-5.0.1.tar gawk-5.0.0.tar.lz 4379 / 5603 = 78.15%
gmp-6.1.1.tar gmp-6.1.2.tar.lz 2454 / 3787 = 64.8%
gmp-6.1.2.tar gmp-6.1.1.tar.lz 2461 / 3782 = 65.07%

Reference file File Reproducible (4096)
mailbox-mar.tar mailbox-aug.tar.lz 4036 / 4252 = 94.92%
backup-feb.tar backup-apr.tar.lz 264 / 542 = 48.71%
backup-apr.tar backup-feb.tar.lz 264 / 520 = 50.77%
gawk-5.0.0.tar gawk-5.0.1.tar.lz 327 / 730 = 44.79%
gawk-5.0.1.tar gawk-5.0.0.tar.lz 326 / 700 = 46.57%
gmp-6.1.1.tar gmp-6.1.2.tar.lz 175 / 473 = 37%
gmp-6.1.2.tar gmp-6.1.1.tar.lz 181 / 472 = 38.35%

Note that the "performance of reproduce" is a probability, not a partial recovery. The data are either recovered fully (with the probability X shown in the last column of the tables above) or not recovered at all (with probability 1 - X).

Example 1: Recover a damaged source tarball with a zeroed sector of 512 bytes at file position 1019904, using as reference another source tarball for a different version of the software.

     lziprecover -vv -e --reference-file=gmp-6.1.1.tar gmp-6.1.2.tar.lz
     Reproducing bad area in member 1 of 1
       (begin = 1019904, size = 512, value = 0x00)
       (master mpos = 1019904, dpos = 6292134)
     warning: gmp-6.1.1.tar: Partial match found at offset 6277798, len 8716.
     Reference data may be mixed with other data.
     Trying level -9
       Reproducing position 1015808
     Member reproduced successfully.
     Copy of input file reproduced successfully.

Example 2: Recover a damaged backup with a zeroed sector of 4096 bytes at file position 1019904, using as reference a previous backup. The damaged backup comes from a damaged partition copied with ddrescue.
     ddrescue -b4096 -r10 /dev/sdc1 hdimage mapfile
     mount -o loop,ro hdimage /mnt/hdimage
     cp /mnt/hdimage/backup.tar.lz backup.tar.lz
     umount /mnt/hdimage
     lzip -t backup.tar.lz
       backup.tar.lz: Decoder error at pos 1020530
     lziprecover -vv -e --reference-file=old_backup.tar backup.tar.lz
     Reproducing bad area in member 1 of 1
       (begin = 1019904, size = 4096, value = 0x00)
       (master mpos = 1019903, dpos = 5857954)
     warning: old_backup.tar: Partial match found at offset 5743778, len 9546.
     Reference data may be mixed with other data.
     Trying level -9
       Reproducing position 1015808
     Member reproduced successfully.
     Copy of input file reproduced successfully.

Example 3: Recover a damaged backup with a zeroed sector of 4096 bytes at file position 1019904, using as reference a file from the filesystem. (If the zeroed sector encodes (part of) a tar header, the tarball can't be reproduced).
     # List the contents of the backup tarball to locate the damaged member.
     tarlz -n0 -tvf backup.tar.lz
       [...]
       example.txt
     tarlz: Skipping to next header.
     tarlz: backup.tar.lz: Archive ends unexpectedly.
     # Find in the filesystem the last file listed and use it as reference.
     lziprecover -vv -e --reference-file=/somedir/example.txt backup.tar.lz
     Reproducing bad area in member 1 of 1
       (begin = 1019904, size = 4096, value = 0x00)
       (master mpos = 1019903, dpos = 5857954)
     /somedir/example.txt: Match found at offset 9378
     Trying level -9
       Reproducing position 1015808
     Member reproduced successfully.
     Copy of input file reproduced successfully.

If 'backup.tar.lz' is a multimember file with more than one member damaged and lziprecover shows the message 'One member reproduced. Copy of input file still contains errors.', the procedure shown in the example above can be repeated until all the members have been reproduced.

'tarlz --keep-damaged -n0 -xf backup.tar.lz example.txt' produces a partial copy of the reference file 'example.txt' that may help locate a complete copy in the filesystem or in another backup, even if 'example.txt' has been renamed.


Next: , Previous: Reproducing one sector, Up: Top

7 Options supporting the tar.lz format

Tarlz is a massively parallel (multi-threaded) combined implementation of the tar archiver and the lzip compressor.

Tarlz creates tar archives using a simplified and safer variant of the POSIX pax format compressed in lzip format, keeping the alignment between tar members and lzip members. The resulting multimember tar.lz archive is backward compatible with standard tar tools like GNU tar, which treat it like any other tar.lz archive.

Multimember tar.lz archives have some safety advantages over solidly compressed tar.lz archives. For example, in case of corruption, tarlz can extract all the undamaged members from the tar.lz archive, skipping over the damaged members, just like the standard (uncompressed) tar. Keeping the alignment between tar members and lzip members minimizes the amount of data lost in case of corruption. In this chapter we'll explain the ways in which lziprecover can recover and process multimember tar.lz archives.


7.1 Recovering damaged multimember tar.lz archives

If you have several copies of the damaged archive, try merging them first because merging has a high probability of success. See Merging files. If the command below prints something like 'Input files merged successfully.' you are done and 'archive.tar.lz' now contains the recovered archive:

     lziprecover -m -v -o archive.tar.lz a/archive.tar.lz b/archive.tar.lz

If you only have one copy of the damaged archive with a zeroed block of data caused by an I/O error, you may try to reproduce the archive. See Reproducing one sector. If the command below prints something like 'Copy of input file reproduced successfully.' you are done and 'archive_fixed.tar.lz' now contains the recovered archive:

     lziprecover -vv -e --reference-file=old_archive.tar archive.tar.lz

If you only have one copy of the damaged archive, you may try to repair the archive, but this has a lower probability of success. See Repairing one byte. If the command below prints something like 'Copy of input file repaired successfully.' you are done and 'archive_fixed.tar.lz' now contains the recovered archive:

     lziprecover -v -R archive.tar.lz

If all the above fails, and the archive was created with tarlz, you may save the damaged members for later and then copy the good members to another archive. If the two commands below succeed, 'bad_members.tar.lz' will contain all the damaged members and 'archive_cleaned.tar.lz' will contain a good archive with the damaged members removed:

     lziprecover -v --dump=damaged -o bad_members.tar.lz archive.tar.lz
     lziprecover -v --strip=damaged -o archive_cleaned.tar.lz archive.tar.lz

You can then use 'tarlz --keep-damaged' to recover as much data as possible from each damaged member in 'bad_members.tar.lz':

     mkdir tmp
     cd tmp
     tarlz --keep-damaged -xvf ../bad_members.tar.lz

7.2 Processing multimember tar.lz archives

Lziprecover is able to copy a list of members from a file to another. For example the command 'lziprecover --dump=1-10:r1:tdata archive.tar.lz > subarch.tar.lz' creates a subset archive containing the first ten members, the end-of-file blocks, and the trailing data (if any) of 'archive.tar.lz'. The 'r1' part selects the last member, which in an appendable tar.lz archive contains the end-of-file blocks.


Next: , Previous: Tarlz, Up: Top

8 Names of the files produced by lziprecover

The name of the fixed file produced by --byte-repair and --merge is made by appending the string '_fixed.lz' to the original file name. If the original file name ends with one of the extensions '.tar.lz', '.lz', or '.tlz', the string '_fixed' is inserted before the extension.


Next: , Previous: File names, Up: Top

9 File format

Perfection is reached, not when there is no longer anything to add, but when there is no longer anything to take away.
-- Antoine de Saint-Exupery


In the diagram below, a box like this:
+---+
|   | <-- the vertical bars might be missing
+---+

represents one byte; a box like this:

+==============+
|              |
+==============+

represents a variable number of bytes.


A lzip file consists of one or more independent "members" (compressed data sets). The members simply appear one after another in the file, with no additional information before, between, or after them. Each member can encode in compressed form up to 16 EiB - 1 byte of uncompressed data. The size of a multimember file is unlimited.

Each member has the following structure:

+--+--+--+--+----+----+=============+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ID string | VN | DS | LZMA stream | CRC32 |   Data size   |  Member size  |
+--+--+--+--+----+----+=============+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

All multibyte values are stored in little endian order.

'ID string (the "magic" bytes)'
A four byte string, identifying the lzip format, with the value "LZIP" (0x4C, 0x5A, 0x49, 0x50).
'VN (version number, 1 byte)'
Just in case something needs to be modified in the future. 1 for now.
'DS (coded dictionary size, 1 byte)'
The dictionary size is calculated by taking a power of 2 (the base size) and subtracting from it a fraction between 0/16 and 7/16 of the base size.
Bits 4-0 contain the base 2 logarithm of the base size (12 to 29).
Bits 7-5 contain the numerator of the fraction (0 to 7) to subtract from the base size to obtain the dictionary size.
Example: 0xD3 = 2^19 - 6 * 2^15 = 512 KiB - 6 * 32 KiB = 320 KiB
Valid values for dictionary size range from 4 KiB to 512 MiB.
'LZMA stream'
The LZMA stream, finished by an "End Of Stream" marker. Uses default values for encoder properties. See Stream format for a complete description.
'CRC32 (4 bytes)'
Cyclic Redundancy Check (CRC) of the original uncompressed data.
'Data size (8 bytes)'
Size of the original uncompressed data.
'Member size (8 bytes)'
Total size of the member, including header and trailer. This field acts as a distributed index, improves the checking of stream integrity, and facilitates the safe recovery of undamaged members from multimember files. Lzip limits the member size to 2 PiB to prevent the data size field from overflowing.


Next: , Previous: File format, Up: Top

10 Extra data appended to the file

Sometimes extra data are found appended to a lzip file after the last member. Such trailing data may be:

Trailing data are in no way part of the lzip file format, but tools reading lzip files are expected to behave as correctly and usefully as possible in the presence of trailing data.

Trailing data can be safely ignored in most cases. In some cases, like that of user-added data, they are expected to be ignored. In those cases where a file containing trailing data must be rejected, the option --trailing-error can be used. See --trailing-error.

Lziprecover facilitates the management of metadata stored as trailing data in lzip files. See the following examples:

Example 1: Add a comment or description to a compressed file.

     # First append the comment as trailing data to a lzip file
     echo 'This file contains this and that' >> file.lz
     # This command prints the comment to standard output
     lziprecover --dump=tdata file.lz
     # This command outputs file.lz without the comment
     lziprecover --strip=tdata file.lz > stripped_file.lz
     # This command removes the comment from file.lz
     lziprecover --remove=tdata file.lz

Example 2: Add and check a cryptographically secure hash. (This may be convenient, but a separate copy of the hash must be kept in a safe place to guarantee that both file and hash have not been maliciously replaced).
     sha256sum < file.lz >> file.lz
     lziprecover --strip=tdata file.lz | sha256sum -c \
       <(lziprecover --dump=tdata file.lz)


Next: , Previous: Trailing data, Up: Top

11 A small tutorial with examples

Example 1: Extract all the files from archive 'foo.tar.lz'.

       tar -xf foo.tar.lz
     or
       lziprecover -cd foo.tar.lz | tar -xf -

Example 2: Restore a regular file from its compressed version 'file.lz'. If the operation is successful, 'file.lz' is removed.
     lziprecover -d file.lz

Example 3: Check the integrity of the compressed file 'file.lz' and show status.
     lziprecover -tv file.lz

Example 4: The right way of concatenating the decompressed output of two or more compressed files. See Trailing data.
     Don't do this
       cat file1.lz file2.lz file3.lz | lziprecover -d -
     Do this instead
       lziprecover -cd file1.lz file2.lz file3.lz
     You may also concatenate the compressed files like this
       lziprecover --strip=tdata file1.lz file2.lz file3.lz > file123.lz
     Or keeping the trailing data of the last file like this
       lziprecover --strip=empty file1.lz file2.lz file3.lz > file123.lz

Example 5: Decompress 'file.lz' partially until 10 KiB of decompressed data are produced.
     lziprecover -D 0,10KiB file.lz

Example 6: Decompress 'file.lz' partially from decompressed byte at offset 10000 to decompressed byte at offset 14999 (5000 bytes are produced).
     lziprecover -D 10000-15000 file.lz

Example 7: Repair a corrupt byte in the file 'file.lz'. (Indented lines are abridged diagnostic messages from lziprecover).
     lziprecover -v -R file.lz
       Copy of input file repaired successfully.
     lziprecover -tv file_fixed.lz
       file_fixed.lz: ok
     mv file_fixed.lz file.lz

Example 8: Split the multimember file 'file.lz' and write each member in its own 'recXXXfile.lz' file. Then use 'lziprecover -t' to test the integrity of the resulting files.
     lziprecover -s file.lz
     lziprecover -tv rec*file.lz


Next: , Previous: Examples, Up: Top

12 Testing the robustness of decompressors

See --unzcrash, for a faster way of testing the robustness of lzip.

The lziprecover package also includes unzcrash, a program written to test robustness to decompression of corrupted data, inspired by unzcrash.c from Julian Seward's bzip2. Type 'make unzcrash' in the lziprecover source directory to build it.

By default, unzcrash reads the file specified and then repeatedly decompresses it, increasing 256 times each byte of the compressed data, so as to test all possible one-byte errors. Note that it may take years or even centuries to test all possible one-byte errors in a large file (tens of MB).

If the option --block is given, unzcrash reads the file specified and then repeatedly decompresses it, setting all bytes in each successive block to the value given, so as to test all possible full sector errors.

If the option --truncate is given, unzcrash reads the file specified and then repeatedly decompresses it, truncating the file to increasing lengths, so as to test all possible truncation points.

None of the three test modes described above should cause any invalid memory accesses. If any of them does, please, report it as a bug to the maintainers of the decompressor being tested.

Unzcrash really executes as a subprocess the shell command specified in the first non-option argument, and then writes the file specified in the second non-option argument to the standard input of the subprocess, modifying the corresponding byte each time. Therefore unzcrash can be used to test any decompressor (not only lzip), or even other decoder programs having a suitable command-line syntax.

If the decompressor returns with zero status, unzcrash compares the output of the decompressor for the original and corrupt files. If the outputs differ, it means that the decompressor returned a false negative; it failed to recognize the corruption and produced garbage output. The only exception is when a multimember file is truncated just after the last byte of a member, producing a shorter but valid compressed file. Except in this latter case, please, report any false negative as a bug.

In order to compare the outputs, unzcrash needs a 'zcmp' program able to understand the format being tested. For example the 'zcmp' provided by zutils. If the 'zcmp' program used does not understand the format being tested, all the comparisons fail because the compressed files are compared without being decompressed first. Use --zcmp=false to disable comparisons.

The format for running unzcrash is:

     unzcrash [options] 'lzip -t' file

The compressed file must not contain errors and the decompressor being tested must decompress it correctly for the comparisons to work.

unzcrash supports the following options:

-h
--help
Print an informative help message describing the options and exit.
-V
--version
Print the version number of unzcrash on the standard output and exit. This version number should be included in all bug reports.
-b range
--bits=range
Test N-bit errors only, instead of testing all the 255 wrong values for each byte. 'N-bit error' means any value differing from the original value in N bit positions, not a value differing from the original value in the bit position N.
The number of N-bit errors per byte (N = 1 to 8) is: 8 28 56 70 56 28 8 1

Examples of range Tests errors of N-bits
1 1
1,2,3 1, 2, 3
2-4 2, 3, 4
1,3-5,8 1, 3, 4, 5, 8
1-3,5-8 1, 2, 3, 5, 6, 7, 8

-B[size][,value]
--block[=size][,value]
Test block errors of given size, simulating a whole sector I/O error. size defaults to 512 bytes. value defaults to 0. By default, only contiguous, non-overlapping blocks are tested, but this may be changed with the option --delta.
-d n
--delta=n
Test one byte, block, or truncation size every n bytes. If --delta is not specified, unzcrash tests all the bytes, non-overlapping blocks, or truncation sizes. Values of n smaller than the block size result in overlapping blocks. (Which is convenient for testing because there are usually too few non-overlapping blocks in a file).
-e position,value
--set-byte=position,value
Set byte at position to value in the internal buffer after reading and testing file but before the first test call to the decompressor. Byte positions start at 0. If value is preceded by '+', it is added to the original value of the byte at position. If value is preceded by 'f' (flip), it is XORed with the original value of the byte at position. This option can be used to run tests with a changed dictionary size, for example.
-n
--no-check
Skip initial test of file and 'zcmp'. May speed up things a lot when testing many (or large) known good files.
-p bytes
--position=bytes
First byte position to test in the file. Defaults to 0. Negative values are relative to the end of the file.
-q
--quiet
Quiet operation. Suppress all messages.
-s bytes
--size=bytes
Number of byte positions to test. If not specified, the rest of the file is tested (from --position to end of file). Negative values are relative to the rest of the file.
-t
--truncate
Test all possible truncation points in the range specified by --position and --size.
-v
--verbose
Verbose mode.
-z
--zcmp=<command>
Set zcmp command name and options. Defaults to 'zcmp'. Use --zcmp=false to disable comparisons. If testing a decompressor different from the one used by default by zcmp, it is needed to force unzcrash and zcmp to use the same decompressor with a command like 'unzcrash --zcmp='zcmp --lz=plzip' 'plzip -t' file'

Exit status: 0 for a normal exit, 1 for environmental problems (file not found, invalid command-line options, I/O errors, etc), 2 to indicate a corrupt or invalid input file, 3 for an internal consistency error (e.g., bug) which caused unzcrash to panic.


Next: , Previous: Unzcrash, Up: Top

13 Reporting bugs

There are probably bugs in lziprecover. There are certainly errors and omissions in this manual. If you report them, they will get fixed. If you don't, no one will ever know about them and they will remain unfixed for all eternity, if not longer.

If you find a bug in lziprecover, please send electronic mail to lzip-bug@nongnu.org. Include the version number, which you can find by running 'lziprecover --version'.


Previous: Problems, Up: Top

Concept index