1 2* Currently the check code (bref.methods / crc, sha, etc) is being checked 3 every single blasted time a chain is locked, even if the underlying buffer 4 was previously checked for that chain. This needs an optimization to 5 (significantly) improve performance. 6 7* flush synchronization boundary crossing check and current flush chain 8 interlock needed. 9 10* snapshot creation must allocate and separately pass a new pmp for the pfs 11 degenerate 'cluster' representing the snapshot. This theoretically will 12 also allow a snapshot to be generated inside a cluster of more than one 13 node. 14 15* snapshot copy currently also copies uuids and can confuse cluster code 16 17* hidden dir or other dirs/files/modifications made to PFS before 18 additional cluster entries added. 19 20* transaction on cluster - multiple trans structures, subtrans 21 22* inode always contains target cluster/chain, not hardlink 23 24* chain refs in cluster, cluster refs 25 26* check inode shared lock ... can end up in endless loop if following 27 hardlink because ip->chain is not updated in the exclusive lock cycle 28 when following hardlink. 29 30cpdup /build/boomdata/jails/bleeding-edge/usr/share/man/man4 /mnt/x3 31 32 33 * The block freeing code. At the very least a bulk scan is needed 34 to implement freeing blocks. 35 36 * Crash stability. Right now the allocation table on-media is not 37 properly synchronized with the flush. This needs to be adjusted 38 such that H2 can do an incremental scan on mount to fixup 39 allocations on mount as part of its crash recovery mechanism. 40 41 * We actually have to start checking and acting upon the CRCs being 42 generated. 43 44 * Remaining known hardlink issues need to be addressed. 45 46 * Core 'copies' mechanism needs to be implemented to support multiple 47 copies on the same media. 48 49 * Core clustering mechanism needs to be implemented to support 50 mirroring and basic multi-master operation from a single host 51 (multi-host requires additional network protocols and won't 52 be as easy). 53 54* make sure we aren't using a shared lock during RB_SCAN's? 55 56* overwrite in write_file case w/compression - if device block size changes 57 the block has to be deleted and reallocated. See hammer2_assign_physical() 58 in vnops. 59 60* freemap / clustering. Set block size on 2MB boundary so the cluster code 61 can be used for reading. 62 63* need API layer for shared buffers (unfortunately). 64 65* add magic number to inode header, add parent inode number too, to 66 help with brute-force recovery. 67 68* modifications past our flush point do not adjust vchain. 69 need to make vchain dynamic so we can (see flush_scan2).?? 70 71* MINIOSIZE/RADIX set to 1KB for now to avoid buffer cache deadlocks 72 on multiple locked inodes. Fix so we can use LBUFSIZE! Or, 73 alternatively, allow a smaller I/O size based on the sector size 74 (not optimal though). 75 76* When making a snapshot, do not allow the snapshot to be mounted until 77 the in-memory chain has been freed in order to break the shared core. 78 79* Snapshotting a sub-directory does not snapshot any 80 parent-directory-spanning hardlinks. 81 82* Snapshot / flush-synchronization point. remodified data that crosses 83 the synchronization boundary is not currently reallocated. see 84 hammer2_chain_modify(), explicit check (requires logical buffer cache 85 buffer handling). 86 87* on fresh mount with multiple hardlinks present separate lookups will 88 result in separate vnodes pointing to separate inodes pointing to a 89 common chain (the hardlink target). 90 91 When the hardlink target consolidates upward only one vp/ip will be 92 adjusted. We need code to fixup the other chains (probably put in 93 inode_lock_*()) which will be pointing to an older deleted hardlink 94 target. 95 96* Filesystem must ensure that modify_tid is not too large relative to 97 the iterator in the volume header, on load, or flush sequencing will 98 not work properly. We should be able to just override it, but we 99 should complain if it happens. 100 101* Kernel-side needs to clean up transaction queues and make appropriate 102 callbacks. 103 104* Userland side needs to do the same for any initiated transactions. 105 106* Nesting problems in the flusher. 107 108* Inefficient vfsync due to thousands of file buffers, one per-vnode. 109 (need to aggregate using a device buffer?) 110 111* Use bp->b_dep to interlock the buffer with the chain structure so the 112 strategy code can calculate the crc and assert that the chain is marked 113 modified (not yet flushed). 114 115* Deleted inode not reachable via tree for volume flush but still reachable 116 via fsync/inactive/reclaim. Its tree can be destroyed at that point. 117 118* The direct write code needs to invalidate any underlying physical buffers. 119 Direct write needs to be implemented. 120 121* Make sure a resized block (hammer2_chain_resize()) calculates a new 122 hash code in the parent bref 123 124* The freemap allocator needs to getblk/clrbuf/bdwrite any partial 125 block allocations (less than 64KB) that allocate out of a new 64K 126 block, to avoid causing a read-before-write I/O. 127 128* Check flush race upward recursion setting SUBMODIFIED vs downward 129 recursion checking SUBMODIFIED then locking (must clear before the 130 recursion and might need additional synchronization) 131 132* There is definitely a flush race in the hardlink implementation between 133 the forwarding entries and the actual (hidden) hardlink inode. 134 135 This will require us to associate a small hard-link-adjust structure 136 with the chain whenever we create or delete hardlinks, on top of 137 adjusting the hardlink inode itself. Any actual flush to the media 138 has to synchronize the correct nlinks value based on whether related 139 created or deleted hardlinks were also flushed. 140 141* When a directory entry is created and also if an indirect block is 142 created and entries moved into it, the directory seek position can 143 potentially become incorrect during a scan. 144 145* When a directory entry is deleted a directory seek position depending 146 on that key can cause readdir to skip entries. 147 148* TWO PHASE COMMIT - store two data offsets in the chain, and 149 hammer2_chain_delete() needs to leave the chain intact if MODIFIED2 is 150 set on its buffer until the flusher gets to it? 151 152 153 OPTIMIZATIONS 154 155* If a file is unlinked buts its descriptors is left open and used, we 156 should allow data blocks on-media to be reused since there is no 157 topology left to point at them. 158