2 HAMMER2 DESIGN DOCUMENT
8 * These features have been speced in the media structures.
10 * Implementation work has begun.
12 * A working filesystem with some features implemented is expected by July 2012.
14 * A fully functional filesystem with most (but not all) features is expected
17 * All elements of the filesystem have been designed except for the freemap
18 (which isn't needed for initial work). 8MB per 2GB of filesystem
19 storage has been reserved for the freemap. The design of the freemap
20 is expected to be completely speced by mid-year.
22 * This is my only project this year. I'm not going to be doing any major
23 kernel bug hunting this year.
27 * Multiple roots (allowing snapshots to be mounted). This is implemented
28 via the super-root concept. When mounting a HAMMER2 filesystem you specify
29 a device path and a directory name in the super-root.
31 * HAMMER1 had PFS's. HAMMER2 does not. Instead, in HAMMER2 any directory
32 in the tree can be configured as a PFS, causing all elements recursively
33 underneath that directory to become a part of that PFS.
35 * Writable snapshots. Any subdirectory tree can be snapshotted. Snapshots
36 show up in the super-root. It is possible to snapshot a subdirectory
37 and then later snapshot a parent of that subdirectory... really there are
40 * Directory sub-hierarchy based quotas and space and inode usage tracking.
41 Any directory sub-tree, whether at a mount point or not, tracks aggregate
42 inode use and data space use. This is stored in the directory inode all
45 * Incremental queueless mirroring / mirroring-streams. Because HAMMER2 is
46 block-oriented and copy-on-write each blockref tracks both direct
47 modifications to the referenced data via (modify_tid) and indirect
48 modifications to the referenced data or any sub-tree via (mirror_tid).
49 This makes it possible to do an incremental scan of meta-data that covers
50 only changes made since the mirror_tid recorded in a prior-run.
52 This feature is also intended to be used to locate recently allocated
53 blocks and thus be able to fixup the freemap after a crash.
55 HAMMER2 mirroring works a bit differently than HAMMER1 mirroring in
56 that HAMMER2 does not keep track of 'deleted' records. Instead any
57 recursion by the mirroring code which finds that (modify_tid) has
58 been updated must also send the direct block table or indirect block
59 table state it winds up recursing through so the target can check
60 similar key ranges and locate elements to be deleted. This can be
61 avoided if the mirroring stream is mostly caught up in that very recent
62 deletions will be cached in memory and can be queried, allowing shorter
63 record deletions to be passed in the stream instead.
65 * Will support multiple compression algorithms configured on subdirectory
66 tree basis and on a file basis. Up to 64K block compression will be used.
67 Only compression ratios near powers of 2 that are at least 2:1 (e.g. 2:1,
68 4:1, 8:1, etc) will work in this scheme because physical block allocations
69 in HAMMER2 are always power-of-2.
71 Compression algorithm #0 will mean no compression and no zero-checking.
72 Compression algorithm #1 will mean zero-checking but no other compression.
73 Real compression will be supported starting with algorithm 2.
75 * Zero detection on write (writing all-zeros), which requires the data
76 buffer to be scanned, will be supported as compression algorithm #1.
77 This allows the writing of 0's to create holes and will be the default
78 compression algorithm for HAMMER2.
80 * Copies support for redundancy. The media blockref structure would
81 have become too bloated but I found a clean way to do copies using the
82 blockset structure (which is a set of 8 fully associative blockref's).
84 The design is such that the filesystem should be able to function at
85 full speed even if disks are pulled or inserted, as long as at least one
86 good copy is present. A background task will be needed to resynchronize
87 missing copies (or remove excessive copies in the case where the copies
88 value is reduced on a live filesystem).
90 * Intended to be clusterable, with a multi-master protocol under design
91 but not expected to be fully operational until mid-2013. The media
92 format for HAMMER1 was less condusive to logical clustering than I had
93 hoped so I was never able to get that aspect of my personal goals
94 working with HAMMER1. HAMMER2 effectively solves the issues that cropped
95 up with HAMMER1 (mainly that HAMMER1's B-Tree did not reflect the logical
96 file/directory hierarchy, making cache coherency very difficult).
98 * Hardlinks will be supported. All other standard features will be supported
99 too of course. Hardlinks in this sort of filesystem require significant
102 * The media blockref structure is now large enough to support up to a 192-bit
103 check value, which would typically be a cryptographic hash of some sort.
104 Multiple check value algorithms will be supported with the default being
105 a simple 32-bit iSCSI CRC.
107 * Fully verified deduplication will be supported and automatic (and
108 necessary in many respects).
110 * Non-verified de-duplication will be supported as a configurable option on
111 a file or subdirectory tree. Non-verified deduplication would use the
112 largest available check code (192 bits) and not bother to verify data
113 matches during the dedup pass, which is necessary on extremely large
114 filesystems with a great deal of deduplicable data (as otherwise a large
115 chunk of the media would have to be read to implement the dedup).
117 This feature is intended only for those files where occassional corruption
118 is ok, such as in a large data store of farmed web content.
122 HAMMER2 generally implements a copy-on-write block design for the filesystem,
123 which is very different from HAMMER1's B-Tree design. Because the design
124 is copy-on-write it can be trivially snapshotted simply by referencing an
125 existing block, and because the media structures logically match a standard
126 filesystem directory/file hierarchy snapshots and other similar operations
127 can be trivially performed on an entire subdirectory tree at any level in
130 The copy-on-write nature of the filesystem implies that any modification
131 whatsoever will have to eventually synchronize new disk blocks all the way
132 to the super-root of the filesystem and the volume header itself. This forms
133 the basis for crash recovery. All disk writes are to new blocks except for
134 the volume header, thus allowing all writes to run concurrently except for
135 the volume header update at the end.
137 Clearly this method requires intermediate modifications to the chain to be
138 cached so multiple modifications can be aggregated prior to being
139 synchronized. One advantage, however, is that the cache can be flushed at
140 any time WITHOUT having to allocate yet another new block when further
141 modifications are made as long as the volume header has not yet been flushed.
142 This means that buffer cache overhead is very well bounded and can handle
143 filesystem operations of any complexity even on boxes with very small amounts
146 I intend to implement a shortcut to make fsync()'s run fast, and that is to
147 allow deep updates to blockrefs to shortcut to auxillary space in the
148 volume header to satisfy the fsync requirement. The related blockref is
149 then recorded when the filesystem is mounted after a crash and the update
150 chain is reconstituted when a matching blockref is encountered again during
151 normal operation of the filesystem.
153 Basically this means that no real work needs to be done at mount-time
156 Directories are hashed, and another major design element is that directory
157 entries ARE INODES. They are one and the same. In addition to directory
158 entries being inodes the data for very small files (512 bytes or smaller)
159 can be directly embedded in the inode (overloaded onto the same space that
160 the direct blockref array uses). This should result in very high
163 Inode numbers are not spatially referenced, which complicates NFS servers
164 but doesn't complicate anything else. The inode number is stored in the
165 inode itself, an absolutely necessary feature in order to support the
166 hugely flexible snapshots that we want to have in HAMMER2.
170 Hardlinks are a particularly sticky problem for HAMMER2 due to the lack of
171 a spatial reference to the inode number. We do not want to have to have
172 an index of inode numbers for any basic HAMMER2 feature if we can help it.
174 Hardlinks are handled by placing the inode for a multiply-hardlinked file
175 in the closest common parent directory. If "a/x" and "a/y" are hardlinked
176 the inode for the hardlinked file will be placed in directory "a", e.g.
177 "a/3239944", but it will be invisible and will be in an out-of-band namespace.
178 The directory entries "a/x" and "a/y" will be given the same inode number
179 but in fact just be placemarks that cause HAMMER2 to recurse upwards through
180 the directory tree to find the invisible inode number.
182 Because directories are hashed and a different namespace (hash key range)
183 is used for hardlinked inodes, standard directory scans are able to trivially
184 skip this invisible namespace and inode-specific lookups can restrict their
185 lookup to within this space.
187 The nature of snapshotting makes handling link-count 2->1 and 1->2 cases
188 trivial. Basically the inode media structure is copied as needed to break-up
189 or re-form the standard directory entry/inode. There are no backpointers in
190 HAMMER2 and no reference counts on the blocks (see FREEMAP NOTES below), so
191 it is an utterly trivial operation.
195 In order to implement fast snapshots (and writable snapshots for that
196 matter), HAMMER2 does NOT ref-count allocations. The freemap which
197 is still under design just won't do that. All the freemap does is
198 keep track of 100% free blocks.
200 This not only trivializes all the snapshot features it also trivializes
201 hardlink handling and solves the problem of keeping the freemap sychronized
202 in the event of a crash. Now all we have to do after a crash is make
203 sure blocks allocated before the freemap was flushed are properly
204 marked as allocated in the allocmap. This is a trivial exercise using the
205 same algorithm the mirror streaming code uses (which is very similar to
206 HAMMER1)... an incremental meta-data scan that covers only the blocks that
207 might have been allocated between the last allocation map sync and now.
209 Thus the freemap does not have to be synchronized during a fsync().
211 The complexity is in figuring out what can be freed... that is, when one
212 can mark blocks in the freemap as being free. HAMMER2 implements this as
213 a background task which essentially must scan available meta-data to
214 determine which blocks are not being referenced.
216 Part of the ongoing design work is finding ways to reduce the scope of this
217 meta-data scan so the entire filesystem's meta-data does not need to be
218 scanned (though in tests with HAMMER1, even full meta-data scans have
219 turned out to be fairly low cost). In other words, its an area that we
220 can continue to improve on as the filesystem matures. Not only that, but
221 we can completely change the freemap algorithms without creating
222 incompatibilities (at worse simply having to require that a R+W mount do
223 a full meta-data scan when upgrading or downgrading the freemap algorithm).
227 Clustering, as always, is the most difficult bit but we have some advantages
228 with HAMMER2 that we did not have with HAMMER1. First, HAMMER2's media
229 structures generally follow the kernel's filesystem hiearchy. Second,
230 HAMMER2's writable snapshots make it possible to implement several forms
231 of multi-master clustering.
233 The general mechanics for most of the multi-master clustering implementations
236 (a) Use the copies mechanism to specify all elements of the cluster,
237 both local and remote (networked).
239 (b) The core synchronization state operates just as it does for copies,
240 simply requiring a fully-flushed ack from the remote in order to
241 mark the blocks as having been fully synchronized.
243 The mirror_tid may be used to locate these blocks, allowing the
244 synchronization state to be updated on the fly at a much later
245 time without requiring the state to be maintained in-memory.
246 (also for crash recovery resynchronization purposes).
248 (c) Data/meta-data can be retrieved from those copies which are marked
249 as being synchronized, with priority given to the local storage
250 relative to any given physical machine.
252 This means that e.g. even in a master-slave orientation the slave
253 may be able to satisfy a request from a program when the slave
254 happens to be the local storage.
256 (d) Transaction id synchronization between all elements of the cluster,
257 typically through masking (assigning a cluster number using the low
258 3 bits of the transaction id).
260 (e) General access (synchronized or otherwise) may require cache
261 coherency mechanisms to run over the network.
263 Implementing cache coherency is a major complexity issue.
265 (f) General access (synchronized or otherwise) may require quorum
266 agreement, using the synchronization flags in the blockrefs
267 to determine whether agreement has been reached.
269 Implementing quorum voting is a major complexity issue.
271 There are lots of ways to implement multi-master environments using the
272 above core features but the implementation is going to be fairly complex
273 even with HAMMER2's feature set.
275 Keep in mind that modifications propagate all the way to the super-root
276 and volume header, so in any clustered arrangement the use of (modify_tid)
277 and (mirror_tid) is critical in determining the synchronization state of
278 portion(s) of the filesystem.
280 Specifically, since any modification propagates to the root the (mirror_tid)
281 in higher level directories is going to be in a constant state of flux. This
282 state of flux DOES NOT invalidate the cache state for these higher levels
283 of directories. Instead, the (modify_tid) is used on a node-by-node basis
284 to determine cache state at any given level, and (mirror_tid) is used to
285 determine whether any recursively underlying state is desynchronized.
287 * Simple semi-synchronized multi-master environment.
289 In this environment all nodes are considered masters and modifications
290 can be made on any of them, and then propagate to the others
291 asynchronously via HAMMER2 mirror streams. One difference here is
292 that kernel can activate these userland-managed streams automatically
293 when the copies configuration is used to specify the cluster.
295 The only type of conflict which isn't readily resolvable by comparing
296 the (modify_tid) is when file data is updated. In this case user
297 intervention might be required but, theoretically, it should be
298 possible to automate most merges using a multi-way patch and, if not,
299 choosing one and creating backup copies if the others to allow the
300 user or sysop to resolve the conflict later.
302 * Simple fully synchronized fail-over environment.
304 In this environment there is one designated master and the remaining
305 nodes are slaves. If the master fails all remaining nodes agree on a
306 new master, possibly with the requirement that a quorum be achieved
307 (if you don't want to allow the cluster to split).
309 If network splits are allowed the each sub-cluster operates in this
310 mode but recombining the clusters reverts to the first algorithm.
311 If not allowed whomever no longer has a quorum will be forced to stall.
313 In this environment the current designated master is responsible for
314 managing locks for modifying operations. The designated master will
315 proactively tell the other nodes to mark the blocks related to the
316 modifying operation as no longer being synchronized while any local
317 data at the node that acquired the lock (master or slave) remains
318 marked as being synchronized.
320 The node that succesfully gets the lock then issues the modifying
321 operation to both its local copy and to the master, marking the
322 master as being desynchronized until the master acknowledges receipt.
324 In this environment any node can access data from local storage if
325 the designated master copy is marked synchronized AND its (modify_tid)
326 matches the slave copy's (modify_tid).
328 However, if a slave disconnects from the master then reconnects the
329 slave will have lost the master's desynchronization stream and must
330 mark its root blockref for the master copy HAMMER2_BREF_DESYNCHLD as
331 well as clear the SYNC1/SYNC2 bits. Setting DESYNCCHLD forces on-demand
332 recursive reverification that the master and slave are (or are not) in
333 sync in order to reestablish on the slave the synchronization state of
336 That might be a bit confusing but the whole point here is to allow
337 read accesses to the filesystem to be satisfied by any node in a
338 multi-master cluster, not just by the current designated master.
340 * Fully cache coherent and synchronized multi-master environment.
342 In this environment a quorum is required to perform any modifying
343 action. All nodes are masters (there is no 'designated' master)
344 and all nodes connect to all other nodes in a cross-bar.
346 The quorum is specified by copies setup in the root volume configuration.
347 A quorum of nodes in the cluster must agree on the copies configuration.
348 If they do not the cluster cannot proceed to mount. Any other nodes
349 not in the quorum which are in the cluster which disagree with the
350 configuration will inherit the copies configuration from the quorum.
352 Any modifying action will initiate a lock request locally to all nodes
353 in the cluster. The modifying action is allowed to proceed the instant
354 a quorum of nodes respond in the affirmative (even if some have not
355 yet responded or are down). The modifying action is considered complete
356 once the two-phase commit protocol succeeds. The modifying action
357 typically creates and commits a temporary snapshot on at least a quorum
358 of masters as phase-1 and then ties the snapshot back into the main
361 These locks are cache-coherency locks and may be passively maintained
362 in order to aggregate multiple operations under the same lock and thus
363 under the same transaction from the point of view of the rest of the
366 A lock request which interferes with a passively maintained lock will
367 force the two-phase commit protocol to complete and then transfer
368 ownership to the requesting entity, thus avoiding having to deal with
369 deadlock protocols at this point in the state machine.
371 Since any node can initiate concurrent lock requests to many other nodes
372 it is possible to deadlock. When two nodes initiate conflicting lock
373 requests to the cluster the one achieving the quorum basically wins and
374 the other is forced to retry (go back one paragraph). In this situation
375 no deadlock will occur.
377 If three are more nodes initiate conflicting lock requests to the
378 cluster a deadlock can occur whereby none of the nodes achieve a quorum.
379 In this case every node will know which of the other nodes was granted
380 the lock(s). Deadlock resolution then proceeds simultaniously on the
381 three nodes (since they have the same information), whereby the lock
382 holders on the losing end of the algorithm transfer their locks to one
383 of the other nodes. The lock state and knowledge of the lock state is
384 updated in real time on all nodes until a quorum is achieved.
386 * Fully cache coherent and synchronized multi-master environment with
387 passive read locking.
389 This is a more complex form of clustering than the previous form.
390 Take the previous form and add the ability to passively hold SHARED
391 locks in addition to the EXCLUSIVE locks the previous form is able
394 The advantage of being able to passively hold a shared lock on a sub-tree
395 (locks can be held on single nodes or entire sub-trees) is that it is
396 then possible for all nodes to validate a node (modify_tid) or entire
397 sub-tree (mirror_tid) with a very short network transaction and then
398 satisfy a large number of requests from local storage.
400 * Fully cache coherent and synchronized multi-master environment with
401 passive read locking and slave-only nodes.
403 This is the MOST complex form of clustering we intend to support.
404 In a multi-master environment requiring a quorum of masters to operate
405 we implement all of the above plus ALSO allow additional nodes to be
406 added to the cluster as slave-only nodes.
408 The difference between a slave-only node and setting up a manual
409 mirror-stream from the cluster to a read-only snapshot on another
410 HAMMER2 filesystem is that the slave-only node will be fully
411 cache coherent with either the cluster proper (if connected to a quorum
412 of masters), or to one or more other nodes in the cluster (if not
413 connected to a quorum of masters), EVEN if the slave itself is not
414 completely caught up.
416 So if the slave-only cluster node is connected to the rest of the cluster
417 over a slow connection you basically get a combination of local disk
418 speeds for any data that is locally in sync and network-limited speeds
419 for any data that is not locally in sync.
421 slave-only cluster nodes run a standard mirror-stream in the background
422 to pull in the data as quickly as possible.
424 This is in constrast to a manual mirror-stream to a read-only
425 snapshot (basically a simple slave), which has no ability to bypass
426 the local storage to handle out-of-date requests (in fact has no ability
427 to detect that the local storage is out-of-date anyway).
projects / dragonfly.git / blob