Use SEEK_HOLE for hole detection

Based on patch by Pavel Raiskup.

Use SEEK_HOLE/SEEK_DATA feature of lseek on systems that support
it.  This can make archiving of sparse files much faster.

Implement the --hole-detection option to allow users to select
hole-detection method.

* src/common.h (hole_detection_method): New enum.
(hole_detection): New global.
* src/sparse.c  (sparse_scan_file_wholesparse): New function as a
method for detecting sparse files without any data.
(sparse_scan_file_raw): Rename from sparse_scan_file; with edits.
(sparse_scan_file_seek): New function.
(sparse_scan_file): Reimplement function.
* src/tar.c: New option --hole-detection

* tests/checkseekhole.c: New file.
* tests/.gitignore: Mention two test binaries.
* tests/Makefile.am: Add new tests.
* tests/testsuite.at (AT_SEEKHOLE_PREREQ): New macro.
Include sparse06.at.
* tests/sparse06.at: New test case.
* tests/sparse02.at: Force raw hole-detection method.
* tests/sparsemv.at: Likewise.
* tests/sparsemvp.at: Likewise.

* doc/tar.1: Document --hole-detection option.
* doc/tar.texi: Document hole-detection algorithms and
command-line options.
* NEWS: Document hole-detection.
This commit is contained in:
Sergey Poznyakoff
2015-12-06 00:14:55 +02:00
parent 589ba77faf
commit b684326e69
14 changed files with 425 additions and 78 deletions
+20 -1
View File
@@ -1,4 +1,4 @@
GNU tar NEWS - User visible changes. 2015-11-02
GNU tar NEWS - User visible changes. 2015-12-06
Please send GNU tar bug reports to <bug-tar@gnu.org>
@@ -48,6 +48,25 @@ read from null-delimited file lists is treated as a file name.
This restores the documented behavior, which was broken in version
1.27.
* Sparse file detection
Tar now uses SEEK_DATA/SEEK_HOLE on systems that support it. This
allows for considerable speed-up in sparse-file detection.
New option --hole-detection is provided, that allows the user to
select the algorithm used for hole detection. Available arguments
are:
--hole-detection=seek
Use lseek(2) SEEK_DATA and SEEK_HOLE "whence" parameters.
--hole-detection=raw
Scan entire file before storing it to determine where holes
are located.
The default is to use "seek" whenever possible, and fall back to
"raw" otherwise.
version 1.28, 2014-07-28