2 * hidden dir or other dirs/files/modifications made to PFS before
3 additional cluster entries added.
5 * transaction on cluster - multiple trans structures, subtrans
7 * inode always contains target cluster/chain, not hardlink
9 * cluster_modify_ip -> data returned mod to all chains
10 * and hammer2_cluster_data() -> same thing
12 * chain refs in cluster, cluster refs
14 * check inode shared lock ... can end up in endless loop if following
15 hardlink because ip->chain is not updated in the exclusive lock cycle
16 when following hardlink.
18 cpdup /build/boomdata/jails/bleeding-edge/usr/share/man/man4 /mnt/x3
21 * The block freeing code. At the very least a bulk scan is needed
22 to implement freeing blocks.
24 * Crash stability. Right now the allocation table on-media is not
25 properly synchronized with the flush. This needs to be adjusted
26 such that H2 can do an incremental scan on mount to fixup
27 allocations on mount as part of its crash recovery mechanism.
29 * We actually have to start checking and acting upon the CRCs being
32 * Remaining known hardlink issues need to be addressed.
34 * Core 'copies' mechanism needs to be implemented to support multiple
35 copies on the same media.
37 * Core clustering mechanism needs to be implemented to support
38 mirroring and basic multi-master operation from a single host
39 (multi-host requires additional network protocols and won't
42 * make sure we aren't using a shared lock during RB_SCAN's?
44 * overwrite in write_file case w/compression - if device block size changes
45 the block has to be deleted and reallocated. See hammer2_assign_physical()
48 * freemap / clustering. Set block size on 2MB boundary so the cluster code
49 can be used for reading.
51 * need API layer for shared buffers (unfortunately).
53 * add magic number to inode header, add parent inode number too, to
54 help with brute-force recovery.
56 * modifications past our flush point do not adjust vchain.
57 need to make vchain dynamic so we can (see flush_scan2).??
59 * MINIOSIZE/RADIX set to 1KB for now to avoid buffer cache deadlocks
60 on multiple locked inodes. Fix so we can use LBUFSIZE! Or,
61 alternatively, allow a smaller I/O size based on the sector size
64 * When making a snapshot, do not allow the snapshot to be mounted until
65 the in-memory chain has been freed in order to break the shared core.
67 * Snapshotting a sub-directory does not snapshot any
68 parent-directory-spanning hardlinks.
70 * Snapshot / flush-synchronization point. remodified data that crosses
71 the synchronization boundary is not currently reallocated. see
72 hammer2_chain_modify(), explicit check (requires logical buffer cache
75 * on fresh mount with multiple hardlinks present separate lookups will
76 result in separate vnodes pointing to separate inodes pointing to a
77 common chain (the hardlink target).
79 When the hardlink target consolidates upward only one vp/ip will be
80 adjusted. We need code to fixup the other chains (probably put in
81 inode_lock_*()) which will be pointing to an older deleted hardlink
84 * Filesystem must ensure that modify_tid is not too large relative to
85 the iterator in the volume header, on load, or flush sequencing will
86 not work properly. We should be able to just override it, but we
87 should complain if it happens.
89 * Kernel-side needs to clean up transaction queues and make appropriate
92 * Userland side needs to do the same for any initiated transactions.
94 * Nesting problems in the flusher.
96 * Inefficient vfsync due to thousands of file buffers, one per-vnode.
97 (need to aggregate using a device buffer?)
99 * Use bp->b_dep to interlock the buffer with the chain structure so the
100 strategy code can calculate the crc and assert that the chain is marked
101 modified (not yet flushed).
103 * Deleted inode not reachable via tree for volume flush but still reachable
104 via fsync/inactive/reclaim. Its tree can be destroyed at that point.
106 * The direct write code needs to invalidate any underlying physical buffers.
107 Direct write needs to be implemented.
109 * Make sure a resized block (hammer2_chain_resize()) calculates a new
110 hash code in the parent bref
112 * The freemap allocator needs to getblk/clrbuf/bdwrite any partial
113 block allocations (less than 64KB) that allocate out of a new 64K
114 block, to avoid causing a read-before-write I/O.
116 * Check flush race upward recursion setting SUBMODIFIED vs downward
117 recursion checking SUBMODIFIED then locking (must clear before the
118 recursion and might need additional synchronization)
120 * There is definitely a flush race in the hardlink implementation between
121 the forwarding entries and the actual (hidden) hardlink inode.
123 This will require us to associate a small hard-link-adjust structure
124 with the chain whenever we create or delete hardlinks, on top of
125 adjusting the hardlink inode itself. Any actual flush to the media
126 has to synchronize the correct nlinks value based on whether related
127 created or deleted hardlinks were also flushed.
129 * When a directory entry is created and also if an indirect block is
130 created and entries moved into it, the directory seek position can
131 potentially become incorrect during a scan.
133 * When a directory entry is deleted a directory seek position depending
134 on that key can cause readdir to skip entries.
136 * TWO PHASE COMMIT - store two data offsets in the chain, and
137 hammer2_chain_delete() needs to leave the chain intact if MODIFIED2 is
138 set on its buffer until the flusher gets to it?
143 * If a file is unlinked buts its descriptors is left open and used, we
144 should allow data blocks on-media to be reused since there is no
145 topology left to point at them.