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* cluster_modify_ip -> data returned mod to all chains 25* and hammer2_cluster_data() -> same thing 26 27* chain refs in cluster, cluster refs 28 29* check inode shared lock ... can end up in endless loop if following 30 hardlink because ip->chain is not updated in the exclusive lock cycle 31 when following hardlink. 32 33cpdup /build/boomdata/jails/bleeding-edge/usr/share/man/man4 /mnt/x3 34 35 36 * The block freeing code. At the very least a bulk scan is needed 37 to implement freeing blocks. 38 39 * Crash stability. Right now the allocation table on-media is not 40 properly synchronized with the flush. This needs to be adjusted 41 such that H2 can do an incremental scan on mount to fixup 42 allocations on mount as part of its crash recovery mechanism. 43 44 * We actually have to start checking and acting upon the CRCs being 45 generated. 46 47 * Remaining known hardlink issues need to be addressed. 48 49 * Core 'copies' mechanism needs to be implemented to support multiple 50 copies on the same media. 51 52 * Core clustering mechanism needs to be implemented to support 53 mirroring and basic multi-master operation from a single host 54 (multi-host requires additional network protocols and won't 55 be as easy). 56 57* make sure we aren't using a shared lock during RB_SCAN's? 58 59* overwrite in write_file case w/compression - if device block size changes 60 the block has to be deleted and reallocated. See hammer2_assign_physical() 61 in vnops. 62 63* freemap / clustering. Set block size on 2MB boundary so the cluster code 64 can be used for reading. 65 66* need API layer for shared buffers (unfortunately). 67 68* add magic number to inode header, add parent inode number too, to 69 help with brute-force recovery. 70 71* modifications past our flush point do not adjust vchain. 72 need to make vchain dynamic so we can (see flush_scan2).?? 73 74* MINIOSIZE/RADIX set to 1KB for now to avoid buffer cache deadlocks 75 on multiple locked inodes. Fix so we can use LBUFSIZE! Or, 76 alternatively, allow a smaller I/O size based on the sector size 77 (not optimal though). 78 79* When making a snapshot, do not allow the snapshot to be mounted until 80 the in-memory chain has been freed in order to break the shared core. 81 82* Snapshotting a sub-directory does not snapshot any 83 parent-directory-spanning hardlinks. 84 85* Snapshot / flush-synchronization point. remodified data that crosses 86 the synchronization boundary is not currently reallocated. see 87 hammer2_chain_modify(), explicit check (requires logical buffer cache 88 buffer handling). 89 90* on fresh mount with multiple hardlinks present separate lookups will 91 result in separate vnodes pointing to separate inodes pointing to a 92 common chain (the hardlink target). 93 94 When the hardlink target consolidates upward only one vp/ip will be 95 adjusted. We need code to fixup the other chains (probably put in 96 inode_lock_*()) which will be pointing to an older deleted hardlink 97 target. 98 99* Filesystem must ensure that modify_tid is not too large relative to 100 the iterator in the volume header, on load, or flush sequencing will 101 not work properly. We should be able to just override it, but we 102 should complain if it happens. 103 104* Kernel-side needs to clean up transaction queues and make appropriate 105 callbacks. 106 107* Userland side needs to do the same for any initiated transactions. 108 109* Nesting problems in the flusher. 110 111* Inefficient vfsync due to thousands of file buffers, one per-vnode. 112 (need to aggregate using a device buffer?) 113 114* Use bp->b_dep to interlock the buffer with the chain structure so the 115 strategy code can calculate the crc and assert that the chain is marked 116 modified (not yet flushed). 117 118* Deleted inode not reachable via tree for volume flush but still reachable 119 via fsync/inactive/reclaim. Its tree can be destroyed at that point. 120 121* The direct write code needs to invalidate any underlying physical buffers. 122 Direct write needs to be implemented. 123 124* Make sure a resized block (hammer2_chain_resize()) calculates a new 125 hash code in the parent bref 126 127* The freemap allocator needs to getblk/clrbuf/bdwrite any partial 128 block allocations (less than 64KB) that allocate out of a new 64K 129 block, to avoid causing a read-before-write I/O. 130 131* Check flush race upward recursion setting SUBMODIFIED vs downward 132 recursion checking SUBMODIFIED then locking (must clear before the 133 recursion and might need additional synchronization) 134 135* There is definitely a flush race in the hardlink implementation between 136 the forwarding entries and the actual (hidden) hardlink inode. 137 138 This will require us to associate a small hard-link-adjust structure 139 with the chain whenever we create or delete hardlinks, on top of 140 adjusting the hardlink inode itself. Any actual flush to the media 141 has to synchronize the correct nlinks value based on whether related 142 created or deleted hardlinks were also flushed. 143 144* When a directory entry is created and also if an indirect block is 145 created and entries moved into it, the directory seek position can 146 potentially become incorrect during a scan. 147 148* When a directory entry is deleted a directory seek position depending 149 on that key can cause readdir to skip entries. 150 151* TWO PHASE COMMIT - store two data offsets in the chain, and 152 hammer2_chain_delete() needs to leave the chain intact if MODIFIED2 is 153 set on its buffer until the flusher gets to it? 154 155 156 OPTIMIZATIONS 157 158* If a file is unlinked buts its descriptors is left open and used, we 159 should allow data blocks on-media to be reused since there is no 160 topology left to point at them. 161