2 * bulkfree pass needs to do a vchain flush from the root to avoid
3 accidently freeing live in-process chains.
5 * Need backend synchronization / serialization when the frontend detaches
6 a XOP. modify_tid tests won't be enough, the backend may wind up executing
7 the XOP out of order after the detach.
9 * xop_start - only start synchronized elements
11 * See if we can remove hammer2_inode_repoint()
13 * FIXME - logical buffer associated with write-in-progress on backend
14 disappears once the cluster validates, even if more backend nodes
17 * FIXME - backend ops need per-node transactions using spmp to protect
20 * FIXME - modifying backend ops are not currently validating the cluster.
21 That probably needs to be done by the frontend in hammer2_xop_start()
23 * modify_tid handling probably broken w/ the XOP code for the moment.
25 * embedded transactions in XOPs - interlock early completion
27 * remove current incarnation of EAGAIN
29 * mtx locks should not track td_locks count?. They can be acquired by one
30 thread and released by another. Need API function for exclusive locks.
32 * Convert xops and hammer2_update_spans() from cluster back into chain calls
34 * syncthr leaves inode locks for entire sync, which is wrong.
36 * recovery scan vs unmount. At the moment an unmount does its flushes,
37 and if successful the freemap will be fully up-to-date, but the mount
38 code doesn't know that and the last flush batch will probably match
39 the PFS root mirror_tid. If it was a large cpdup the (unnecessary)
40 recovery pass at mount time can be extensive. Add a CLEAN flag to the
41 volume header to optimize out the unnecessary recovery pass.
43 * More complex transaction sequencing and flush merging. Right now it is
44 all serialized against flushes.
46 * adding new pfs - freeze and force remaster
48 * removing a pfs - freeze and force remaster
50 * bulkfree - sync between passes and enforce serialization of operation
52 * bulkfree - signal check, allow interrupt
54 * bulkfree - sub-passes when kernel memory block isn't large enough
56 * bulkfree - limit kernel memory allocation for bmap space
58 * bulkfree - must include any detached vnodes in scan so open unlinked files
59 are not ripped out from under the system.
61 * bulkfree - must include all volume headers in scan so they can be used
62 for recovery or automatic snapshot retrieval.
64 * bulkfree - snapshot duplicate sub-tree cache and tests needed to reduce
67 * Currently the check code (bref.methods / crc, sha, etc) is being checked
68 every single blasted time a chain is locked, even if the underlying buffer
69 was previously checked for that chain. This needs an optimization to
70 (significantly) improve performance.
72 * flush synchronization boundary crossing check and current flush chain
75 * snapshot creation must allocate and separately pass a new pmp for the pfs
76 degenerate 'cluster' representing the snapshot. This theoretically will
77 also allow a snapshot to be generated inside a cluster of more than one
80 * snapshot copy currently also copies uuids and can confuse cluster code
82 * hidden dir or other dirs/files/modifications made to PFS before
83 additional cluster entries added.
85 * transaction on cluster - multiple trans structures, subtrans
87 * inode always contains target cluster/chain, not hardlink
89 * chain refs in cluster, cluster refs
91 * check inode shared lock ... can end up in endless loop if following
92 hardlink because ip->chain is not updated in the exclusive lock cycle
93 when following hardlink.
95 cpdup /build/boomdata/jails/bleeding-edge/usr/share/man/man4 /mnt/x3
98 * The block freeing code. At the very least a bulk scan is needed
99 to implement freeing blocks.
101 * Crash stability. Right now the allocation table on-media is not
102 properly synchronized with the flush. This needs to be adjusted
103 such that H2 can do an incremental scan on mount to fixup
104 allocations on mount as part of its crash recovery mechanism.
106 * We actually have to start checking and acting upon the CRCs being
109 * Remaining known hardlink issues need to be addressed.
111 * Core 'copies' mechanism needs to be implemented to support multiple
112 copies on the same media.
114 * Core clustering mechanism needs to be implemented to support
115 mirroring and basic multi-master operation from a single host
116 (multi-host requires additional network protocols and won't
119 * make sure we aren't using a shared lock during RB_SCAN's?
121 * overwrite in write_file case w/compression - if device block size changes
122 the block has to be deleted and reallocated. See hammer2_assign_physical()
125 * freemap / clustering. Set block size on 2MB boundary so the cluster code
126 can be used for reading.
128 * need API layer for shared buffers (unfortunately).
130 * add magic number to inode header, add parent inode number too, to
131 help with brute-force recovery.
133 * modifications past our flush point do not adjust vchain.
134 need to make vchain dynamic so we can (see flush_scan2).??
136 * MINIOSIZE/RADIX set to 1KB for now to avoid buffer cache deadlocks
137 on multiple locked inodes. Fix so we can use LBUFSIZE! Or,
138 alternatively, allow a smaller I/O size based on the sector size
139 (not optimal though).
141 * When making a snapshot, do not allow the snapshot to be mounted until
142 the in-memory chain has been freed in order to break the shared core.
144 * Snapshotting a sub-directory does not snapshot any
145 parent-directory-spanning hardlinks.
147 * Snapshot / flush-synchronization point. remodified data that crosses
148 the synchronization boundary is not currently reallocated. see
149 hammer2_chain_modify(), explicit check (requires logical buffer cache
152 * on fresh mount with multiple hardlinks present separate lookups will
153 result in separate vnodes pointing to separate inodes pointing to a
154 common chain (the hardlink target).
156 When the hardlink target consolidates upward only one vp/ip will be
157 adjusted. We need code to fixup the other chains (probably put in
158 inode_lock_*()) which will be pointing to an older deleted hardlink
161 * Filesystem must ensure that modify_tid is not too large relative to
162 the iterator in the volume header, on load, or flush sequencing will
163 not work properly. We should be able to just override it, but we
164 should complain if it happens.
166 * Kernel-side needs to clean up transaction queues and make appropriate
169 * Userland side needs to do the same for any initiated transactions.
171 * Nesting problems in the flusher.
173 * Inefficient vfsync due to thousands of file buffers, one per-vnode.
174 (need to aggregate using a device buffer?)
176 * Use bp->b_dep to interlock the buffer with the chain structure so the
177 strategy code can calculate the crc and assert that the chain is marked
178 modified (not yet flushed).
180 * Deleted inode not reachable via tree for volume flush but still reachable
181 via fsync/inactive/reclaim. Its tree can be destroyed at that point.
183 * The direct write code needs to invalidate any underlying physical buffers.
184 Direct write needs to be implemented.
186 * Make sure a resized block (hammer2_chain_resize()) calculates a new
187 hash code in the parent bref
189 * The freemap allocator needs to getblk/clrbuf/bdwrite any partial
190 block allocations (less than 64KB) that allocate out of a new 64K
191 block, to avoid causing a read-before-write I/O.
193 * Check flush race upward recursion setting SUBMODIFIED vs downward
194 recursion checking SUBMODIFIED then locking (must clear before the
195 recursion and might need additional synchronization)
197 * There is definitely a flush race in the hardlink implementation between
198 the forwarding entries and the actual (hidden) hardlink inode.
200 This will require us to associate a small hard-link-adjust structure
201 with the chain whenever we create or delete hardlinks, on top of
202 adjusting the hardlink inode itself. Any actual flush to the media
203 has to synchronize the correct nlinks value based on whether related
204 created or deleted hardlinks were also flushed.
206 * When a directory entry is created and also if an indirect block is
207 created and entries moved into it, the directory seek position can
208 potentially become incorrect during a scan.
210 * When a directory entry is deleted a directory seek position depending
211 on that key can cause readdir to skip entries.
213 * TWO PHASE COMMIT - store two data offsets in the chain, and
214 hammer2_chain_delete() needs to leave the chain intact if MODIFIED2 is
215 set on its buffer until the flusher gets to it?
220 * If a file is unlinked buts its descriptors is left open and used, we
221 should allow data blocks on-media to be reused since there is no
222 topology left to point at them.