What's new in Squashfs 3.4
Aug 27, 2008
- 1. Kernel code improvements:
- 1.1 Internal Squashfs kernel metadata and fragment cache implementations have been merged and optimised. Spinlocks are now used, locks are held for smaller periods and wakeups have been minimised. Small race condition fixed where if two or more processes tried to read the same cache block simultaneously they would both read and decompress it. 10-20% speed improvement has been seen on tests.
- 1.2 NFS export code rewritten following VFS changes in linux-2.6.24
- 1.3 New patches for linux-2.6.25, linux-2.6.26, and linux-2.6.27. Fixed patch for linux-2.6.24.
- 1.4 Fixed small buffer_head leak in squashfs_read_data when handling badly corrupted filesystems.
- 1.5 Fixed bug in get_dir_index_using_offset.
- 2. Unsquashfs improvements:
- 2.1 Unsquashfs has been parallelised. Filesystem reading, writing and decompression is now multi-threaded. Up to 40% speed improvement seen on tests.
- 2.2 Unsquashfs now has a progress bar. Use -no-progress to disable it.
- 2.3 Fixed small bug where unistd.h wasn't being included on some distributions, leading to lseek being used rather than lseek64 - which meant on these distributions Unsquashfs couldn't unsquash filesystems larger than 4GB.
- 3. Mksquashfs improvements:
- 3.1 Removed some small remaining parallelisation bottlenecks. Depending on source filesystem, up to 10% speed improvement. 3.2 Progress bar improved, and moved to separate thread.
- 3.3 Sparse file handling bug in Mksquashfs 3.3 fixed.
- 3.4 Two rare appending restore bugs fixed (when ^C hit twice).