Hi all, Here's v3 of sparse inode chunk suport for XFS. The primary update for this version is an update to how inodes are aligned when sparse inode support is enabled. Inode chunks are currently aligned to cluster size to support single I/O per inode cluster. This means that the minimum block range between two non-adjacent inode chunks is the cluster size. Cluster size is also the granularity of sparse allocation. Therefore, it is possible to allocate a cluster size chunk that cannot be converted to an inode record due to overlap on both sides (ambiguous metadata). The only recourse in this situation is to undo the allocation and likely return ENOSPC. Given the added complexity of that and the already complicated inode allocation path, an approach that avoids this potential condition in the first place is preferred. To address this situation, inode alignment is increased from cluster size to chunk size (by mkfs) when sparse inode chunks are enabled. This guarantees that the minimum block range between two non-adjacent inode chunks is at least big enough for one full chunk. This greatly simplifies sparse inode record management. Allocations occur at cluster size granularity and are shifted into inode records that align to chunk size. In other words, for any particular sparse allocation, a well known record startino is determined by aligning the agbno of the allocation to the chunk size. The increased inode chunk alignment does limit the ability to allocate full inode chunks on a significantly populated fs, but what is lost in that regard is regained by the ability to allocate sparse records in any AG that can satisfy the minimum free space requirement[1]. Other changes in this version include marking the feature as experimental, a block allocation agbno range limit to avoid invalid inode records at AG boundaries, DEBUG mode allocation logic to improve test coverage, etc. This series survives xfstests regression runs on basic v5 configurations as well as some longer term debug mode fsstress testing. Thoughts, reviews, flames appreciated! Brian [1] - This can also be mitigated by future work to consider allocation of inode chunks in batches rather than one at a time. v3: - Rebase to latest for-next (bulkstat rework, data structure shuffling, etc.). - Fix issparse helper logic. - Update inode alignment model w/ spinodes enabled. All inode records are chunk size aligned, sparse allocations cluster size aligned (both enforced on mount). - Reworked sparse inode record merge logic to coincide w/ new alignment model. - Mark feature as experimental (warn on mount). - Include and use block allocation agbno range limit to prevent allocation of invalid inode records. - Add some DEBUG bits to improve sparse alloc. test coverage. v2: http://oss.sgi.com/archives/xfs/2014-11/msg00007.html - Use a manually set feature bit instead of dynamic based on the existence of sparse inode chunks. - Add sb/mp fields for sparse alloc. granularity (use instead of cluster size). - Undo xfs_inobt_insert() loop removal to avoid breakage of larger page size arches. - Rename sparse record overlap helper and do XFS_LOOKUP_LE search. - Use byte of pad space in inobt record for inode count field. - Convert bitmap mgmt to use generic bitmap code. - Rename XFS_INODES_PER_SPCHUNK to XFS_INODES_PER_HOLEMASK_BIT. - Add fs geometry bit for sparse inodes. - Rebase to latest for-next (bulkstat refactor). v1: http://oss.sgi.com/archives/xfs/2014-07/msg00355.html Brian Foster (18): xfs: add sparse inode chunk alignment superblock field xfs: use sparse chunk alignment for min. inode allocation requirement xfs: sparse inode chunks feature helpers and mount requirements xfs: introduce inode record hole mask for sparse inode chunks xfs: create macros/helpers for dealing with sparse inode chunks xfs: pass inode count through ordered icreate log item xfs: handle sparse inode chunks in icreate log recovery xfs: helpers to convert holemask to/from generic bitmap xfs: support min/max agbno args in block allocator xfs: allocate sparse inode chunks on full chunk allocation failure xfs: randomly do sparse inode allocations in DEBUG mode xfs: filter out sparse regions from individual inode allocation xfs: update free inode record logic to support sparse inode records xfs: only free allocated regions of inode chunks xfs: skip unallocated regions of inode chunks in xfs_ifree_cluster() xfs: use actual inode count for sparse records in bulkstat/inumbers xfs: add fs geometry bit for sparse inode chunks xfs: enable sparse inode chunks for v5 superblocks fs/xfs/libxfs/xfs_alloc.c | 42 ++- fs/xfs/libxfs/xfs_alloc.h | 2 + fs/xfs/libxfs/xfs_format.h | 33 +- fs/xfs/libxfs/xfs_fs.h | 1 + fs/xfs/libxfs/xfs_ialloc.c | 651 ++++++++++++++++++++++++++++++++++++--- fs/xfs/libxfs/xfs_ialloc.h | 17 +- fs/xfs/libxfs/xfs_ialloc_btree.c | 4 +- fs/xfs/libxfs/xfs_sb.c | 31 +- fs/xfs/xfs_fsops.c | 4 +- fs/xfs/xfs_inode.c | 28 +- fs/xfs/xfs_itable.c | 14 +- fs/xfs/xfs_log_recover.c | 23 +- fs/xfs/xfs_mount.c | 16 + fs/xfs/xfs_mount.h | 2 + fs/xfs/xfs_trace.h | 47 +++ 15 files changed, 836 insertions(+), 79 deletions(-) -- 1.8.3.1 _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
