)]}'
{
  "log": [
    {
      "commit": "a22285a6a32390195235171b89d157ed1a1fe932",
      "tree": "3fabc88a029e1af4f2fdcc708e7b62ef3cf3703a",
      "parents": [
        "f0486c68e4bd9a06a5904d3eeb3a0d73a83befb8"
      ],
      "author": {
        "name": "Yan, Zheng",
        "email": "zheng.yan@oracle.com",
        "time": "Sun May 16 10:48:46 2010 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Tue May 25 10:34:50 2010 -0400"
      },
      "message": "Btrfs: Integrate metadata reservation with start_transaction\n\nBesides simplify the code, this change makes sure all metadata\nreservation for normal metadata operations are released after\ncommitting transaction.\n\nChanges since V1:\n\nAdd code that check if unlink and rmdir will free space.\n\nAdd ENOSPC handling for clone ioctl.\n\nSigned-off-by: Yan Zheng \u003czheng.yan@oracle.com\u003e\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    },
    {
      "commit": "5d4f98a28c7d334091c1b7744f48a1acdd2a4ae0",
      "tree": "c611d7d824cbcdb777dd2d8e33e2ed1c5df8a9c6",
      "parents": [
        "5c939df56c3ea018b58e5aa76181284c2053d699"
      ],
      "author": {
        "name": "Yan Zheng",
        "email": "zheng.yan@oracle.com",
        "time": "Wed Jun 10 10:45:14 2009 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Wed Jun 10 11:29:46 2009 -0400"
      },
      "message": "Btrfs: Mixed back reference  (FORWARD ROLLING FORMAT CHANGE)\n\nThis commit introduces a new kind of back reference for btrfs metadata.\nOnce a filesystem has been mounted with this commit, IT WILL NO LONGER\nBE MOUNTABLE BY OLDER KERNELS.\n\nWhen a tree block in subvolume tree is cow\u0027d, the reference counts of all\nextents it points to are increased by one.  At transaction commit time,\nthe old root of the subvolume is recorded in a \"dead root\" data structure,\nand the btree it points to is later walked, dropping reference counts\nand freeing any blocks where the reference count goes to 0.\n\nThe increments done during cow and decrements done after commit cancel out,\nand the walk is a very expensive way to go about freeing the blocks that\nare no longer referenced by the new btree root.  This commit reduces the\ntransaction overhead by avoiding the need for dead root records.\n\nWhen a non-shared tree block is cow\u0027d, we free the old block at once, and the\nnew block inherits old block\u0027s references. When a tree block with reference\ncount \u003e 1 is cow\u0027d, we increase the reference counts of all extents\nthe new block points to by one, and decrease the old block\u0027s reference count by\none.\n\nThis dead tree avoidance code removes the need to modify the reference\ncounts of lower level extents when a non-shared tree block is cow\u0027d.\nBut we still need to update back ref for all pointers in the block.\nThis is because the location of the block is recorded in the back ref\nitem.\n\nWe can solve this by introducing a new type of back ref. The new\nback ref provides information about pointer\u0027s key, level and in which\ntree the pointer lives. This information allow us to find the pointer\nby searching the tree. The shortcoming of the new back ref is that it\nonly works for pointers in tree blocks referenced by their owner trees.\n\nThis is mostly a problem for snapshots, where resolving one of these\nfuzzy back references would be O(number_of_snapshots) and quite slow.\nThe solution used here is to use the fuzzy back references in the common\ncase where a given tree block is only referenced by one root,\nand use the full back references when multiple roots have a reference\non a given block.\n\nThis commit adds per subvolume red-black tree to keep trace of cached\ninodes. The red-black tree helps the balancing code to find cached\ninodes whose inode numbers within a given range.\n\nThis commit improves the balancing code by introducing several data\nstructures to keep the state of balancing. The most important one\nis the back ref cache. It caches how the upper level tree blocks are\nreferenced. This greatly reduce the overhead of checking back ref.\n\nThe improved balancing code scales significantly better with a large\nnumber of snapshots.\n\nThis is a very large commit and was written in a number of\npieces.  But, they depend heavily on the disk format change and were\nsquashed together to make sure git bisect didn\u0027t end up in a\nbad state wrt space balancing or the format change.\n\nSigned-off-by: Yan Zheng \u003czheng.yan@oracle.com\u003e\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    },
    {
      "commit": "1a81af4d1d9c60d4313309f937a1fc5567205a87",
      "tree": "535b65b3948d34f0948613afc6ebdfe693683f33",
      "parents": [
        "af4176b49c5ee15a9c9b10720c92456b28e7aac7"
      ],
      "author": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Wed Mar 25 09:55:11 2009 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Wed Mar 25 09:55:11 2009 -0400"
      },
      "message": "Btrfs: make sure btrfs_update_delayed_ref doesn\u0027t increase ref_mod\n\nbtrfs_update_delayed_ref is optimized to add and remove different\nreferences in one pass through the delayed ref tree.  It is a zero\nsum on the total number of refs on a given extent.\n\nBut, the code was recording an extra ref in the head node.  This\nnever made it down to the disk but was used when deciding if it was\nsafe to free the extent while dropping snapshots.\n\nThe fix used here is to make sure the ref_mod count is unchanged\non the head ref when btrfs_update_delayed_ref is called.\n\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    },
    {
      "commit": "c3e69d58e86c3917ae4e9e31b4acf490a7cafe60",
      "tree": "bd4f1e62446a208bdae26f0c36d67e3afbc1cd1d",
      "parents": [
        "1887be66dcc3140a81d1299958a41fc0eedfa64f"
      ],
      "author": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Fri Mar 13 10:17:05 2009 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Tue Mar 24 16:14:26 2009 -0400"
      },
      "message": "Btrfs: process the delayed reference queue in clusters\n\nThe delayed reference queue maintains pending operations that need to\nbe done to the extent allocation tree.  These are processed by\nfinding records in the tree that are not currently being processed one at\na time.\n\nThis is slow because it uses lots of time searching through the rbtree\nand because it creates lock contention on the extent allocation tree\nwhen lots of different procs are running delayed refs at the same time.\n\nThis commit changes things to grab a cluster of refs for processing,\nusing a cursor into the rbtree as the starting point of the next search.\nThis way we walk smoothly through the rbtree.\n\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    },
    {
      "commit": "1887be66dcc3140a81d1299958a41fc0eedfa64f",
      "tree": "3f0666948a83c48c77e0c37dca1c71a6aff9eb1d",
      "parents": [
        "44871b1b24b593996db43495cf4484cc580bdc10"
      ],
      "author": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Fri Mar 13 10:11:24 2009 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Tue Mar 24 16:14:26 2009 -0400"
      },
      "message": "Btrfs: try to cleanup delayed refs while freeing extents\n\nWhen extents are freed, it is likely that we\u0027ve removed the last\ndelayed reference update for the extent.  This checks the delayed\nref tree when things are freed, and if no ref updates area left it\nimmediately processes the delayed ref.\n\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    },
    {
      "commit": "56bec294dea971335d4466b30f2d959f28f6e36d",
      "tree": "fc0b5bbf4bb6ab35582a4c7f58f5ac88f71c38bf",
      "parents": [
        "9fa8cfe706f9c20067c042a064999d5825a35330"
      ],
      "author": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Fri Mar 13 10:10:06 2009 -0400"
      },
      "committer": {
        "name": "Chris Mason",
        "email": "chris.mason@oracle.com",
        "time": "Tue Mar 24 16:14:25 2009 -0400"
      },
      "message": "Btrfs: do extent allocation and reference count updates in the background\n\nThe extent allocation tree maintains a reference count and full\nback reference information for every extent allocated in the\nfilesystem.  For subvolume and snapshot trees, every time\na block goes through COW, the new copy of the block adds a reference\non every block it points to.\n\nIf a btree node points to 150 leaves, then the COW code needs to go\nand add backrefs on 150 different extents, which might be spread all\nover the extent allocation tree.\n\nThese updates currently happen during btrfs_cow_block, and most COWs\nhappen during btrfs_search_slot.  btrfs_search_slot has locks held\non both the parent and the node we are COWing, and so we really want\nto avoid IO during the COW if we can.\n\nThis commit adds an rbtree of pending reference count updates and extent\nallocations.  The tree is ordered by byte number of the extent and byte number\nof the parent for the back reference.  The tree allows us to:\n\n1) Modify back references in something close to disk order, reducing seeks\n2) Significantly reduce the number of modifications made as block pointers\nare balanced around\n3) Do all of the extent insertion and back reference modifications outside\nof the performance critical btrfs_search_slot code.\n\n#3 has the added benefit of greatly reducing the btrfs stack footprint.\nThe extent allocation tree modifications are done without the deep\n(and somewhat recursive) call chains used in the past.\n\nThese delayed back reference updates must be done before the transaction\ncommits, and so the rbtree is tied to the transaction.  Throttling is\nimplemented to help keep the queue of backrefs at a reasonable size.\n\nSince there was a similar mechanism in place for the extent tree\nextents, that is removed and replaced by the delayed reference tree.\n\nYan Zheng \u003cyan.zheng@oracle.com\u003e helped review and fixup this code.\n\nSigned-off-by: Chris Mason \u003cchris.mason@oracle.com\u003e\n"
    }
  ]
}