Release Notes: Speeding up MS_SYNC writes, setting up kjournald to be parented under init properly, ext3 is no longer experimental in config, fixes for i_blocks getting inconsistent after disk full, sped-up fsyncs in non-journaled data modes a little, no longer considering ENOSPC a fatal error when allocating an inode, a fix for an LVM snapshot deadlock, a fix for a "dump corrupts filesystems" core VFS bug, a fix for an over-zealous ext3 complaint about locked buffers, a fix for a very rare buffer leak, an O_SYNC fix, and a fix for a tiny race where a buffer could be written to disk too soon.
Release Notes: There are no changes in this release; it's just a patch for a newer kernel.
Release Notes: Removal of much debug code, minimized core kernel changes, moving /proc/sys/fs/jbd-debug control up into jbd itself, and removal of i_truncate_sem in favour of leaving the orphaned pages on the inactive_dirty LRU rather than swapping them out.
Release Notes: Removal of compatibility ifdefs for old-style quotas, a fix for an oops in ext3_writepage(), and truncate can zero page->mapping after block_write_full_page().
Release Notes: A fix for an oops in directory readahead on empty (i.e., deleted) directories, handling the new S_NOQUOTA flag for 2.4.10+ quota code, and reverting i_truncate_sem vs. i_sem lock ordering to original design to fix an obscure deadlock (but will deadlock 2.4.11's shmem.c; it's safe with 2.4.10-ac11's shmem.c).
Release Notes: A fix for an oops which could occur at unmount time due to non-empty orphan list, merging in a directory scan speedup heuristic, removal of the abort_write() address_space_operation() by ensuring that all prepare_write() callers always call commit_write(), and a number of changes to suit the new 2.4.10 VM and buffer-layer design.
Release Notes: The buffer-already-revoked assertion failure was fixed by looking up an aliased buffercache buffer and clearing the revoke bits in there, as well as in the journalled data buffer. The page truncation code was reorganized so that the address of block_flushpage() is not taken. This is to simplify merging with Andrea's O_DIRECT patch, which turns block_flushpage() into a macro.
Release Notes: An endianness bug which caused failure on PPC and S390 was fixed.
Release Notes: The checkpoint cleanup code was reworked to reduce another O(n^2) performance bottleneck.