Line data Source code
1 : #include "fd_replay_tile.h"
2 : #include "fd_sched.h"
3 : #include "fd_exec.h"
4 : #include "fd_vote_tracker.h"
5 : #include "../../flamenco/accdb/fd_accdb_sync.h"
6 : #include "generated/fd_replay_tile_seccomp.h"
7 :
8 : #include "../genesis/fd_genesi_tile.h"
9 : #include "../poh/fd_poh.h"
10 : #include "../poh/fd_poh_tile.h"
11 : #include "../tower/fd_tower_tile.h"
12 : #include "../resolv/fd_resolv_tile.h"
13 : #include "../restore/utils/fd_ssload.h"
14 :
15 : #include "../../disco/tiles.h"
16 : #include "../../disco/fd_txn_m.h"
17 : #include "../../disco/store/fd_store.h"
18 : #include "../../disco/pack/fd_pack.h"
19 : #include "../../discof/reasm/fd_reasm.h"
20 : #include "../../disco/keyguard/fd_keyload.h"
21 : #include "../../disco/genesis/fd_genesis_cluster.h"
22 : #include "../../util/pod/fd_pod.h"
23 : #include "../../flamenco/accdb/fd_accdb_admin.h"
24 : #include "../../flamenco/accdb/fd_accdb_user.h"
25 : #include "../../flamenco/rewards/fd_rewards.h"
26 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
27 : #include "../../flamenco/progcache/fd_progcache_admin.h"
28 : #include "../../disco/metrics/fd_metrics.h"
29 :
30 : #include "../../flamenco/runtime/fd_runtime.h"
31 : #include "../../flamenco/runtime/fd_runtime_stack.h"
32 : #include "../../flamenco/fd_flamenco_base.h"
33 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
34 :
35 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
36 :
37 : #include <errno.h>
38 : #include <stdio.h>
39 :
40 : /* Replay concepts:
41 :
42 : - Blocks are aggregations of entries aka. microblocks which are
43 : groupings of txns and are constructed by the block producer (see
44 : fd_pack).
45 :
46 : - Entries are grouped into entry batches by the block producer (see
47 : fd_pack / fd_shredder).
48 :
49 : - Entry batches are divided into chunks known as shreds by the block
50 : producer (see fd_shredder).
51 :
52 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
53 : the block producer (see fd_shredder).
54 :
55 : - Shreds are transmitted to the rest of the cluster via the Turbine
56 : protocol (see fd_shredder / fd_shred).
57 :
58 : - Once enough shreds within a FEC set are received to recover the
59 : entirety of the shred data encoded by that FEC set, the receiver
60 : can "complete" the FEC set (see fd_fec_resolver).
61 :
62 : - If shreds in the FEC set are missing such that it can't complete,
63 : the receiver can use the Repair protocol to request missing shreds
64 : in FEC set (see fd_repair).
65 :
66 : - The current Repair protocol does not support requesting coding
67 : shreds. As a result, some FEC sets might be actually complete
68 : (contain all data shreds). Repair currently hacks around this by
69 : forcing completion but the long-term solution is to add support for
70 : fec_repairing coding shreds via Repair.
71 :
72 : - FEC sets are delivered in partial-order to the Replay tile by the
73 : Repair tile. Currently Replay only supports replaying entry batches
74 : so FEC sets need to reassembled into an entry batch before they can
75 : be replayed. The new Dispatcher will change this by taking a FEC
76 : set as input instead. */
77 :
78 0 : #define IN_KIND_SNAP ( 0)
79 0 : #define IN_KIND_GENESIS ( 1)
80 0 : #define IN_KIND_IPECHO ( 2)
81 0 : #define IN_KIND_TOWER ( 3)
82 0 : #define IN_KIND_RESOLV ( 4)
83 0 : #define IN_KIND_POH ( 5)
84 0 : #define IN_KIND_EXEC ( 6)
85 0 : #define IN_KIND_SHRED ( 7)
86 0 : #define IN_KIND_VTXN ( 8)
87 0 : #define IN_KIND_GUI ( 9)
88 0 : #define IN_KIND_RPC (10)
89 :
90 : #define DEBUG_LOGGING 0
91 :
92 : /* The first bank that that the replay tile produces either for genesis
93 : or the snapshot boot will always be at bank index 0. */
94 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
95 :
96 : struct fd_replay_in_link {
97 : fd_wksp_t * mem;
98 : ulong chunk0;
99 : ulong wmark;
100 : ulong mtu;
101 : };
102 :
103 : typedef struct fd_replay_in_link fd_replay_in_link_t;
104 :
105 : struct fd_replay_out_link {
106 : ulong idx;
107 : fd_wksp_t * mem;
108 : ulong chunk0;
109 : ulong wmark;
110 : ulong chunk;
111 : };
112 :
113 : typedef struct fd_replay_out_link fd_replay_out_link_t;
114 :
115 : /* fd_block_id_map is a simple map of block-ids to bank indices. The
116 : map sits on top of an array of fd_block_id_ele_t. This serves as a
117 : translation layer between block ids to bank indices. */
118 :
119 : struct fd_block_id_ele {
120 : fd_hash_t block_id;
121 : ulong slot; /* = FD_SLOT_NULL if not initialized */
122 : ulong next_;
123 : };
124 : typedef struct fd_block_id_ele fd_block_id_ele_t;
125 :
126 : #define MAP_NAME fd_block_id_map
127 : #define MAP_ELE_T fd_block_id_ele_t
128 : #define MAP_KEY_T fd_hash_t
129 0 : #define MAP_KEY block_id
130 0 : #define MAP_NEXT next_
131 0 : #define MAP_KEY_EQ(k0,k1) (!memcmp((k0),(k1), sizeof(fd_hash_t)))
132 0 : #define MAP_KEY_HASH(key,seed) (fd_hash((seed),(key),sizeof(fd_hash_t)))
133 : #include "../../util/tmpl/fd_map_chain.c"
134 :
135 : static inline ulong
136 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
137 0 : return (ulong)(ele - ele_arr);
138 0 : }
139 :
140 : struct fd_replay_tile {
141 : fd_wksp_t * wksp;
142 :
143 : fd_accdb_admin_t accdb_admin[1];
144 : fd_accdb_user_t accdb[1];
145 : fd_progcache_admin_t progcache_admin[1];
146 :
147 : fd_txncache_t * txncache;
148 : fd_store_t * store;
149 : fd_banks_t * banks;
150 :
151 : /* This flag is 1 If we have seen a vote signature that our node has
152 : sent out get rooted at least one time. The value is 0 otherwise.
153 : We can't become leader and pack blocks until this flag has been
154 : set. This parallels the Agave 'has_new_vote_been_rooted'.
155 :
156 : TODO: Add a flag to the toml to make this optional. */
157 : int has_identity_vote_rooted;
158 :
159 : fd_reasm_t * reasm;
160 :
161 : /* Replay state machine. */
162 : fd_sched_t * sched;
163 : ulong exec_cnt;
164 : fd_replay_out_link_t exec_out[ 1 ]; /* Sending work down to exec tiles */
165 :
166 : fd_vote_tracker_t * vote_tracker;
167 :
168 : int has_genesis_hash;
169 : uchar genesis_hash[ 32UL ];
170 : ulong cluster_type;
171 :
172 : #define FD_REPLAY_HARD_FORKS_MAX (64UL)
173 : ulong hard_forks_cnt;
174 : ulong hard_forks[ FD_REPLAY_HARD_FORKS_MAX ];
175 :
176 : ushort expected_shred_version;
177 : ushort ipecho_shred_version;
178 :
179 : /* A note on publishing ...
180 :
181 : The watermarks are used to publish our fork-aware structures. For
182 : example, store, banks, and txncache need to be published to release
183 : resources occupied by rooted or dead blocks. In general,
184 : publishing has the effect of pruning forks in those structures,
185 : indicating that it is ok to release the memory being occupied by
186 : the blocks on said forks. Tower is responsible for informing us of
187 : the latest block on the consensus rooted fork. As soon as we can,
188 : we should move the published root as close as possible to the
189 : latest consensus root, publishing/pruning everything on the fork
190 : tree along the way. That is, all the blocks that directly descend
191 : from the current published root (inclusive) to the new published
192 : root (exclusive) on the rooted fork, as well as all the minority
193 : forks that branch from said blocks.
194 :
195 : Ideally, we'd move the published root to the consensus root
196 : immediately upon receiving a new consensus root. However, that's
197 : not always safe to do. One thing we need to be careful about is
198 : making sure that there are no more users/consumers of
199 : soon-to-be-pruned blocks, lest a use-after-free occurs. This can
200 : be done by using a reference counter for each block. Any
201 : concurrent activity, such as transaction execution in the exec
202 : tiles, should retain a refcnt on the block for as
203 : long as it needs access to the shared fork-aware structures related
204 : to that block. Eventually, refcnt on a given block will drop down
205 : to 0 as the block either finishes replaying or gets marked as dead,
206 : and any other tile that has retained a refcnt on the block releases
207 : it. At that point, it becomes a candidate for pruning. The key to
208 : safe publishing then becomes figuring out how far we could advance
209 : the published root, such that every minority fork branching off of
210 : blocks in between the current published root (inclusive) and the
211 : new published root (exclusive) is safe to be pruned. This is a
212 : straightforward tree traversal, where if a block B on the rooted
213 : fork has refcnt 0, and all minority forks branching off of B also
214 : have refcnt 0, then B is safe to be pruned. We advance the
215 : published root to the farthest consecutively prunable block on the
216 : rooted fork. Note that reasm presents the replay tile with a clean
217 : view of the world where every block is chained off of a parent
218 : block. So there are no orpahned/dangling tree nodes to worry
219 : about. The world is a nice single tree as far as replay is
220 : concerned.
221 :
222 : In the following fork tree, every node is a block and the number in
223 : parentheses is the refcnt on the block. The chain marked with
224 : double slashes is the rooted fork. Suppose the published root is
225 : at block P, and consensus root is at block T. We can't publish
226 : past block P because Q has refcnt 1.
227 :
228 :
229 : P(0)
230 : / \\
231 : Q(1) A(0)
232 : / || \
233 : X(0) B(0) C(0)
234 : / || \
235 : Y(0) M(0) R(0)
236 : / || / \
237 : D(2) T(0) J(0) L(0)
238 : ||
239 : ..
240 : ..
241 : ..
242 : ||
243 : blocks we might be actively replaying
244 :
245 :
246 : When refcnt on Q drops to 0, we would be able to advance the
247 : published root to block M, because blocks P, A, and B, as well as
248 : all subtrees branching off of them, have refcnt 0, and therefore
249 : can be pruned. Block M itself cannot be pruned yet because its
250 : child block D has refcnt 2. After publishing/pruning, the fork
251 : tree would be:
252 :
253 :
254 : M(0)
255 : / ||
256 : D(2) T(0)
257 : ||
258 : ..
259 : ..
260 : ..
261 : ||
262 : blocks we might be actively replaying
263 :
264 :
265 : As a result, the shared fork-aware structures can free resources
266 : for blocks P, A, B, and all subtrees branching off of them.
267 :
268 : For the reference counting part, the replay tile is the sole entity
269 : that can update the refcnt. This ensures that all refcnt increment
270 : and decrement attempts are serialized at the replay tile, and that
271 : there are no racy resurrection of a soon-to-be-pruned block. If a
272 : refcnt increment request arrives after a block has been pruned,
273 : replay simply rejects the request.
274 :
275 : A note on the implementation of the above ...
276 :
277 : Upon receiving a new consensus root, we descend down the rooted
278 : fork from the current published root to the new consensus root. On
279 : each node/block of the rooted fork, we do a summation of the refcnt
280 : on the block and all the minority fork blocks branching from the
281 : block. If the summation is 0, the block is safe for pruning. We
282 : advance the published root to the far end of the consecutive run of
283 : 0 refcnt sums originating from the current published root. On our
284 : descent down the minority forks, we also mark any block that hasn't
285 : finished replaying as dead, so we don't waste time executing them.
286 : No more transactions shall be dispatched for execution from dead
287 : blocks.
288 :
289 : Blocks start out with a refcnt of 0. Other tiles may send a
290 : request to the replay tile for a reference on a block. The
291 : transaction dispatcher is another source of refcnt updates. On
292 : every dispatch of a transaction for block B, we increment the
293 : refcnt for B. And on every transaction finalization, we decrement
294 : the refcnt for B. This means that whenever the refcnt on a block
295 : is 0, there is no more reference on that block from the execution
296 : pipeline. While it might be tempting to simply increment the
297 : refcnt once when we start replaying a block, and decrement the
298 : refcnt once when we finish a block, this more fine-grained refcnt
299 : update strategy allows for aborting and potentially immediate
300 : pruning of blocks under interleaved block replay. Upon receiving a
301 : new consensus root, we can simply look at the refcnt on minority
302 : fork blocks, and a refcnt of 0 would imply that the block is safe
303 : for pruning, even if we haven't finished replaying it. Without the
304 : fine-grained refcnt, we would need to first stop dispatching from
305 : the aborted block, and then wait for a full drain of the execution
306 : pipeline to know for sure that there are no more in-flight
307 : transactions executing on the aborted block. Note that this will
308 : allow the refcnt on any block to transiently drop down to 0. We
309 : will not mistakenly prune an actively replaying block, aka a leaf
310 : node, that is chaining off of the rooted fork, because the
311 : consensus root is always an ancestor of the actively replaying tip.
312 : */
313 : fd_hash_t consensus_root; /* The most recent block to have reached max lockout in the tower. */
314 : ulong consensus_root_slot; /* slot number of the above. */
315 : ulong consensus_root_bank_idx; /* bank index of the above. */
316 : ulong published_root_slot; /* slot number of the published root. */
317 : ulong published_root_bank_idx; /* bank index of the published root. */
318 :
319 : /* We need to maintain a tile-local mapping of block-ids to bank index
320 : and vice versa. This translation layer is needed for conversion
321 : since tower operates on block-ids and downstream consumers of FEC
322 : sets operate on bank indices. This mapping must happen both ways:
323 : 1. tower sends us block ids and we must map them to bank indices.
324 : 2. when a block is completed, we must map the bank index to a block
325 : id to send a slot complete message to tower. */
326 : ulong block_id_len;
327 : fd_block_id_ele_t * block_id_arr;
328 : fd_block_id_map_t * block_id_map;
329 :
330 : /* Capture-related configs */
331 : fd_capture_ctx_t * capture_ctx;
332 : FILE * capture_file;
333 :
334 : /* Whether the runtime has been booted either from snapshot loading
335 : or from genesis. */
336 : int is_booted;
337 :
338 : /* Buffer to store vote towers that need to be published to the Tower
339 : tile. */
340 :
341 : fd_multi_epoch_leaders_t * mleaders;
342 :
343 : int larger_max_cost_per_block;
344 :
345 : fd_pubkey_t identity_pubkey[1]; /* TODO: Keyswitch */
346 :
347 : /* When we transition to becoming leader, we can only unbecome the
348 : leader if we have received a block id from the FEC reassembler, and
349 : a message from PoH that the leader slot has ended. After both of
350 : these conditions are met, then we are free to unbecome the leader.
351 : */
352 : int is_leader;
353 : int recv_poh;
354 : int recv_block_id;
355 : ulong next_leader_slot;
356 : long next_leader_tickcount;
357 : ulong highwater_leader_slot;
358 : ulong reset_slot;
359 : fd_bank_t * reset_bank;
360 : fd_hash_t reset_block_id;
361 : long reset_timestamp_nanos;
362 : double slot_duration_nanos;
363 : double slot_duration_ticks;
364 : fd_bank_t * leader_bank; /* ==NULL if not currently the leader */
365 :
366 : ulong resolv_tile_cnt;
367 :
368 : int in_kind[ 64 ];
369 : fd_replay_in_link_t in[ 64 ];
370 :
371 : fd_replay_out_link_t replay_out[1];
372 :
373 : fd_replay_out_link_t stake_out[1];
374 :
375 : /* The gui tile needs to reliably own a reference to the most recent
376 : completed active bank. Replay needs to know if the gui as a
377 : consumer is enabled so it can increment the bank's refcnt before
378 : publishing the bank_idx to the gui. */
379 : int gui_enabled;
380 : int rpc_enabled;
381 :
382 : # if FD_HAS_FLATCC
383 : /* For dumping blocks to protobuf. For backtest only. */
384 : fd_block_dump_ctx_t * block_dump_ctx;
385 : # endif
386 :
387 : /* We need a few pieces of information to compute the right addresses
388 : for bundle crank information that we need to send to pack. */
389 : struct {
390 : int enabled;
391 : fd_pubkey_t vote_account;
392 : fd_bundle_crank_gen_t gen[1];
393 : } bundle;
394 :
395 : struct {
396 : fd_histf_t store_read_wait[ 1 ];
397 : fd_histf_t store_read_work[ 1 ];
398 : fd_histf_t store_publish_wait[ 1 ];
399 : fd_histf_t store_publish_work[ 1 ];
400 : fd_histf_t store_link_wait[ 1 ];
401 : fd_histf_t store_link_work[ 1 ];
402 :
403 : ulong slots_total;
404 : ulong transactions_total;
405 : } metrics;
406 :
407 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
408 :
409 : fd_runtime_stack_t runtime_stack;
410 : };
411 :
412 : typedef struct fd_replay_tile fd_replay_tile_t;
413 :
414 : FD_FN_CONST static inline ulong
415 0 : scratch_align( void ) {
416 0 : return 128UL;
417 0 : }
418 : FD_FN_PURE static inline ulong
419 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
420 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
421 :
422 0 : ulong l = FD_LAYOUT_INIT;
423 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
424 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
425 0 : l = FD_LAYOUT_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
426 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
427 0 : l = FD_LAYOUT_APPEND( l, fd_reasm_align(), fd_reasm_footprint( 1 << 20 ) );
428 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.max_live_slots ) );
429 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
430 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
431 :
432 0 : # if FD_HAS_FLATCC
433 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
434 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
435 0 : }
436 0 : # endif
437 :
438 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
439 :
440 0 : return l;
441 0 : }
442 :
443 : static inline void
444 0 : metrics_write( fd_replay_tile_t * ctx ) {
445 0 : FD_MHIST_COPY( REPLAY, STORE_LINK_WAIT, ctx->metrics.store_link_wait );
446 0 : FD_MHIST_COPY( REPLAY, STORE_LINK_WORK, ctx->metrics.store_link_work );
447 0 : FD_MHIST_COPY( REPLAY, STORE_READ_WAIT, ctx->metrics.store_read_wait );
448 0 : FD_MHIST_COPY( REPLAY, STORE_READ_WORK, ctx->metrics.store_read_work );
449 0 : FD_MHIST_COPY( REPLAY, STORE_PUBLISH_WAIT, ctx->metrics.store_publish_wait );
450 0 : FD_MHIST_COPY( REPLAY, STORE_PUBLISH_WORK, ctx->metrics.store_publish_work );
451 :
452 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
453 0 : ulong leader_slot = ctx->leader_bank ? fd_bank_slot_get( ctx->leader_bank ) : 0UL;
454 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
455 :
456 0 : if( FD_LIKELY( ctx->leader_bank ) ) {
457 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
458 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
459 0 : } else {
460 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
461 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
462 0 : }
463 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
464 :
465 0 : fd_bank_t * bank_pool = fd_banks_get_bank_pool( ctx->banks );
466 0 : ulong live_banks = fd_banks_pool_max( bank_pool ) - fd_banks_pool_free( bank_pool );
467 0 : FD_MGAUGE_SET( REPLAY, LIVE_BANKS, live_banks );
468 :
469 0 : FD_MCNT_SET( REPLAY, SLOTS_TOTAL, ctx->metrics.slots_total );
470 0 : FD_MCNT_SET( REPLAY, TRANSACTIONS_TOTAL, ctx->metrics.transactions_total );
471 :
472 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, ctx->progcache_admin->metrics.root_cnt );
473 0 : FD_MCNT_SET( REPLAY, PROGCACHE_GC_ROOT, ctx->progcache_admin->metrics.gc_root_cnt );
474 0 : }
475 :
476 : static inline ulong
477 : generate_stake_weight_msg( ulong epoch,
478 : fd_epoch_schedule_t const * epoch_schedule,
479 : fd_vote_states_t const * epoch_stakes,
480 0 : ulong * stake_weight_msg_out ) {
481 0 : fd_stake_weight_msg_t * stake_weight_msg = (fd_stake_weight_msg_t *)fd_type_pun( stake_weight_msg_out );
482 0 : fd_vote_stake_weight_t * stake_weights = stake_weight_msg->weights;
483 :
484 0 : stake_weight_msg->epoch = epoch;
485 0 : stake_weight_msg->start_slot = fd_epoch_slot0( epoch_schedule, epoch );
486 0 : stake_weight_msg->slot_cnt = epoch_schedule->slots_per_epoch;
487 0 : stake_weight_msg->excluded_stake = 0UL;
488 0 : stake_weight_msg->vote_keyed_lsched = 1UL;
489 :
490 : /* FIXME: SIMD-0180 - hack to (de)activate in testnet vs mainnet.
491 : This code can be removed once the feature is active. */
492 0 : if( (1==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_TESTNET) ||
493 0 : (0==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_MAINNET) ) {
494 0 : stake_weight_msg->vote_keyed_lsched = 0UL;
495 0 : }
496 :
497 : /* epoch_stakes from manifest are already filtered (stake>0), but not sorted */
498 0 : fd_vote_states_iter_t iter_[1];
499 0 : ulong idx = 0UL;
500 0 : for( fd_vote_states_iter_t * iter = fd_vote_states_iter_init( iter_, epoch_stakes ); !fd_vote_states_iter_done( iter ); fd_vote_states_iter_next( iter ) ) {
501 0 : fd_vote_state_ele_t * vote_state = fd_vote_states_iter_ele( iter );
502 0 : if( FD_UNLIKELY( !vote_state->stake ) ) continue;
503 :
504 0 : stake_weights[ idx ].stake = vote_state->stake;
505 0 : memcpy( stake_weights[ idx ].id_key.uc, &vote_state->node_account, sizeof(fd_pubkey_t) );
506 0 : memcpy( stake_weights[ idx ].vote_key.uc, &vote_state->vote_account, sizeof(fd_pubkey_t) );
507 0 : idx++;
508 0 : }
509 0 : stake_weight_msg->staked_cnt = idx;
510 0 : sort_vote_weights_by_stake_vote_inplace( stake_weights, idx );
511 :
512 0 : return fd_stake_weight_msg_sz( idx );
513 0 : }
514 :
515 : static void
516 : publish_stake_weights( fd_replay_tile_t * ctx,
517 : fd_stem_context_t * stem,
518 : fd_bank_t * bank,
519 0 : int current_epoch ) {
520 0 : fd_epoch_schedule_t const * schedule = fd_bank_epoch_schedule_query( bank );
521 0 : ulong epoch = fd_slot_to_epoch( schedule, fd_bank_slot_get( bank ), NULL );
522 :
523 0 : fd_vote_states_t const * vote_states_prev;
524 0 : if( FD_LIKELY( current_epoch ) ) vote_states_prev = fd_bank_vote_states_prev_locking_query( bank );
525 0 : else vote_states_prev = fd_bank_vote_states_prev_prev_locking_query( bank );
526 :
527 0 : ulong * stake_weights_msg = fd_chunk_to_laddr( ctx->stake_out->mem, ctx->stake_out->chunk );
528 0 : ulong stake_weights_sz = generate_stake_weight_msg( epoch+fd_ulong_if( current_epoch, 1UL, 0UL), schedule, vote_states_prev, stake_weights_msg );
529 0 : ulong stake_weights_sig = 4UL;
530 0 : fd_stem_publish( stem, ctx->stake_out->idx, stake_weights_sig, ctx->stake_out->chunk, stake_weights_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
531 0 : ctx->stake_out->chunk = fd_dcache_compact_next( ctx->stake_out->chunk, stake_weights_sz, ctx->stake_out->chunk0, ctx->stake_out->wmark );
532 :
533 0 : if( FD_LIKELY( current_epoch ) ) fd_bank_vote_states_prev_end_locking_query( bank );
534 0 : else fd_bank_vote_states_prev_prev_end_locking_query( bank );
535 :
536 0 : fd_multi_epoch_leaders_stake_msg_init( ctx->mleaders, fd_type_pun_const( stake_weights_msg ) );
537 0 : fd_multi_epoch_leaders_stake_msg_fini( ctx->mleaders );
538 0 : }
539 :
540 : /**********************************************************************/
541 : /* Transaction execution state machine helpers */
542 : /**********************************************************************/
543 :
544 : static fd_bank_t *
545 : replay_block_start( fd_replay_tile_t * ctx,
546 : fd_stem_context_t * stem,
547 : ulong bank_idx,
548 : ulong parent_bank_idx,
549 0 : ulong slot ) {
550 0 : long before = fd_log_wallclock();
551 :
552 : /* Switch to a new block that we don't have a bank for. */
553 :
554 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
555 0 : if( FD_UNLIKELY( !bank ) ) {
556 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
557 0 : }
558 0 : if( FD_UNLIKELY( bank->flags!=FD_BANK_FLAGS_INIT ) ) {
559 0 : FD_LOG_CRIT(( "invariant violation: bank is not in correct state for bank index %lu", bank_idx ));
560 0 : }
561 :
562 0 : bank->preparation_begin_nanos = before;
563 :
564 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
565 0 : if( FD_UNLIKELY( !parent_bank ) ) {
566 0 : FD_LOG_CRIT(( "invariant violation: parent bank is NULL for bank index %lu", parent_bank_idx ));
567 0 : }
568 0 : if( FD_UNLIKELY( !(parent_bank->flags&FD_BANK_FLAGS_FROZEN) ) ) {
569 0 : FD_LOG_CRIT(( "invariant violation: parent bank is not frozen for bank index %lu", parent_bank_idx ));
570 0 : }
571 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
572 :
573 : /* Clone the bank from the parent. We must special case the first
574 : slot that is executed as the snapshot does not provide a parent
575 : block id. */
576 :
577 0 : bank = fd_banks_clone_from_parent( ctx->banks, bank_idx, parent_bank_idx );
578 0 : if( FD_UNLIKELY( !bank ) ) {
579 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
580 0 : }
581 0 : fd_bank_slot_set( bank, slot );
582 0 : fd_bank_parent_slot_set( bank, parent_slot );
583 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
584 :
585 : /* Create a new funk txn for the block. */
586 :
587 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank_idx } };
588 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
589 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
590 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
591 :
592 : /* Update any required runtime state and handle any potential epoch
593 : boundary change. */
594 :
595 0 : if( ctx->capture_ctx ) {
596 0 : fd_solcap_writer_set_slot( ctx->capture_ctx->capture, slot );
597 0 : }
598 :
599 0 : fd_bank_shred_cnt_set( bank, 0UL );
600 0 : fd_bank_execution_fees_set( bank, 0UL );
601 0 : fd_bank_priority_fees_set( bank, 0UL );
602 0 : fd_bank_tips_set( bank, 0UL );
603 :
604 0 : fd_bank_has_identity_vote_set( bank, 0 );
605 :
606 : /* Set the tick height. */
607 0 : fd_bank_tick_height_set( bank, fd_bank_max_tick_height_get( bank ) );
608 :
609 : /* Update block height. */
610 0 : fd_bank_block_height_set( bank, fd_bank_block_height_get( bank ) + 1UL );
611 :
612 0 : ulong * max_tick_height = fd_bank_max_tick_height_modify( bank );
613 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
614 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS != fd_runtime_compute_max_tick_height( ticks_per_slot, slot, max_tick_height ) ) ) {
615 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, ticks_per_slot ));
616 0 : }
617 :
618 0 : int is_epoch_boundary = 0;
619 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
620 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_stake_weights( ctx, stem, bank, 1 );
621 :
622 0 : return bank;
623 0 : }
624 :
625 : static void
626 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
627 0 : if( bank->cost_tracker_pool_idx!=fd_bank_cost_tracker_pool_idx_null( fd_bank_get_cost_tracker_pool( bank ) ) ) {
628 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
629 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
630 0 : slot_info->cost_tracker.vote_cost = cost_tracker->vote_cost;
631 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
632 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
633 0 : slot_info->cost_tracker.vote_cost_limit = cost_tracker->vote_cost_limit;
634 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
635 0 : fd_bank_cost_tracker_end_locking_query( bank );
636 0 : } else {
637 0 : memset( &slot_info->cost_tracker, 0, sizeof(slot_info->cost_tracker) );
638 0 : }
639 0 : }
640 :
641 : static ulong
642 0 : get_identity_balance( fd_replay_tile_t * ctx, fd_funk_txn_xid_t xid ) {
643 0 : ulong identity_balance = ULONG_MAX;
644 0 : fd_txn_account_t identity_acc[1];
645 0 : int err = fd_txn_account_init_from_funk_readonly( identity_acc,
646 0 : ctx->identity_pubkey,
647 0 : ctx->accdb->funk,
648 0 : &xid );
649 0 : if( FD_LIKELY( !err && identity_acc->meta ) ) identity_balance = identity_acc->meta->lamports;
650 :
651 0 : return identity_balance;
652 0 : }
653 :
654 : static void
655 : publish_slot_completed( fd_replay_tile_t * ctx,
656 : fd_stem_context_t * stem,
657 : fd_bank_t * bank,
658 : int is_initial,
659 0 : int is_leader ) {
660 :
661 0 : ulong slot = fd_bank_slot_get( bank );
662 :
663 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->idx ];
664 :
665 : /* HACKY: hacky way of checking if we should send a null parent block
666 : id */
667 0 : fd_hash_t parent_block_id = {0};
668 0 : if( FD_UNLIKELY( !is_initial ) ) {
669 0 : parent_block_id = ctx->block_id_arr[ bank->parent_idx ].block_id;
670 0 : }
671 :
672 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
673 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
674 0 : FD_TEST( bank_hash );
675 0 : FD_TEST( block_hash );
676 :
677 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
678 :
679 0 : fd_epoch_schedule_t const * epoch_schedule = fd_bank_epoch_schedule_query( bank );
680 0 : ulong slot_idx;
681 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
682 :
683 0 : ctx->metrics.slots_total++;
684 0 : ctx->metrics.transactions_total = fd_bank_txn_count_get( bank );
685 :
686 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
687 0 : slot_info->slot = slot;
688 0 : slot_info->root_slot = ctx->consensus_root_slot;
689 0 : slot_info->storage_slot = ctx->published_root_slot;
690 0 : slot_info->epoch = epoch;
691 0 : slot_info->slot_in_epoch = slot_idx;
692 0 : slot_info->block_height = fd_bank_block_height_get( bank );
693 0 : slot_info->parent_slot = fd_bank_parent_slot_get( bank );
694 0 : slot_info->block_id = block_id_ele->block_id;
695 0 : slot_info->parent_block_id = parent_block_id;
696 0 : slot_info->bank_hash = *bank_hash;
697 0 : slot_info->block_hash = *block_hash;
698 0 : slot_info->transaction_count = fd_bank_txn_count_get( bank );
699 :
700 0 : fd_inflation_t inflation = fd_bank_inflation_get( bank );
701 0 : slot_info->inflation.foundation = inflation.foundation;
702 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
703 0 : slot_info->inflation.terminal = inflation.terminal;
704 0 : slot_info->inflation.initial = inflation.initial;
705 0 : slot_info->inflation.taper = inflation.taper;
706 :
707 0 : fd_rent_t rent = fd_bank_rent_get( bank );
708 0 : slot_info->rent.burn_percent = rent.burn_percent;
709 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
710 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
711 :
712 0 : slot_info->first_fec_set_received_nanos = bank->first_fec_set_received_nanos;
713 0 : slot_info->preparation_begin_nanos = bank->preparation_begin_nanos;
714 0 : slot_info->first_transaction_scheduled_nanos = bank->first_transaction_scheduled_nanos;
715 0 : slot_info->last_transaction_finished_nanos = bank->last_transaction_finished_nanos;
716 0 : slot_info->completion_time_nanos = fd_log_wallclock();
717 :
718 : /* refcnt should be incremented by 1 for each consumer that uses
719 : `bank_idx`. Each consumer should decrement the bank's refcnt once
720 : they are done usin the bank. */
721 0 : bank->refcnt++; /* tower_tile */
722 0 : if( FD_LIKELY( ctx->gui_enabled ) ) bank->refcnt++; /* gui tile */
723 0 : slot_info->bank_idx = bank->idx;
724 :
725 0 : slot_info->parent_bank_idx = ULONG_MAX;
726 0 : fd_bank_t * parent_bank = fd_banks_get_parent( ctx->banks, bank );
727 0 : if( FD_LIKELY( parent_bank && ctx->gui_enabled ) ) {
728 0 : parent_bank->refcnt++;
729 0 : slot_info->parent_bank_idx = parent_bank->idx;
730 0 : }
731 :
732 0 : slot_info->is_leader = is_leader;
733 :
734 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_SLOT_COMPLETED, ctx->replay_out->chunk, sizeof(fd_replay_slot_completed_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
735 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_slot_completed_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
736 0 : }
737 :
738 : static void
739 : replay_block_finalize( fd_replay_tile_t * ctx,
740 : fd_stem_context_t * stem,
741 0 : fd_bank_t * bank ) {
742 :
743 0 : bank->last_transaction_finished_nanos = fd_log_wallclock();
744 :
745 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) fd_solcap_writer_flush( ctx->capture_ctx->capture );
746 :
747 0 : FD_TEST( !(bank->flags&FD_BANK_FLAGS_FROZEN) );
748 :
749 : /* Set poh hash in bank. */
750 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->idx );
751 0 : fd_bank_poh_set( bank, *poh );
752 :
753 : /* Set shred count in bank. */
754 0 : fd_bank_shred_cnt_set( bank, fd_sched_get_shred_cnt( ctx->sched, bank->idx ) );
755 :
756 : /* Do hashing and other end-of-block processing. */
757 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
758 :
759 : /* Copy out cost tracker fields before freezing */
760 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
761 0 : cost_tracker_snap( bank, slot_info );
762 :
763 : /* fetch identity / vote balance updates infrequently */
764 0 : ulong slot = fd_bank_slot_get( bank );
765 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank->idx } };
766 0 : slot_info->identity_balance = FD_UNLIKELY( slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
767 :
768 : /* Mark the bank as frozen. */
769 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
770 :
771 : /**********************************************************************/
772 : /* Bank hash comparison, and halt if there's a mismatch after replay */
773 : /**********************************************************************/
774 :
775 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
776 0 : FD_TEST( bank_hash );
777 :
778 : /* Must be last so we can measure completion time correctly, even
779 : though we could technically do this before the hash cmp and vote
780 : tower stuff. */
781 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */ );
782 :
783 0 : # if FD_HAS_FLATCC
784 : /* If enabled, dump the block to a file and reset the dumping
785 : context state */
786 0 : if( FD_UNLIKELY( ctx->capture_ctx && ctx->capture_ctx->dump_block_to_pb ) ) {
787 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, ctx->accdb->funk, ctx->capture_ctx );
788 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
789 0 : }
790 0 : # endif
791 0 : }
792 :
793 : /**********************************************************************/
794 : /* Leader bank management */
795 : /**********************************************************************/
796 :
797 : static fd_bank_t *
798 : prepare_leader_bank( fd_replay_tile_t * ctx,
799 : ulong slot,
800 : long now,
801 : fd_hash_t const * parent_block_id,
802 0 : fd_stem_context_t * stem ) {
803 0 : long before = fd_log_wallclock();
804 :
805 : /* Make sure that we are not already leader. */
806 0 : FD_TEST( ctx->leader_bank==NULL );
807 :
808 0 : fd_block_id_ele_t * parent_ele = fd_block_id_map_ele_query( ctx->block_id_map, parent_block_id, NULL, ctx->block_id_arr );
809 0 : if( FD_UNLIKELY( !parent_ele ) ) {
810 0 : FD_LOG_CRIT(( "invariant violation: parent bank index not found for merkle root %s", FD_BASE58_ENC_32_ALLOCA( parent_block_id->uc ) ));
811 0 : }
812 0 : ulong parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent_ele );
813 :
814 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
815 0 : if( FD_UNLIKELY( !parent_bank ) ) {
816 0 : FD_LOG_CRIT(( "invariant violation: parent bank not found for bank index %lu", parent_bank_idx ));
817 0 : }
818 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
819 :
820 0 : ctx->leader_bank = fd_banks_new_bank( ctx->banks, parent_bank_idx, now );
821 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
822 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
823 0 : }
824 :
825 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( ctx->banks, ctx->leader_bank->idx, parent_bank_idx ) ) ) {
826 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
827 0 : }
828 :
829 0 : ctx->leader_bank->preparation_begin_nanos = before;
830 :
831 0 : fd_bank_slot_set( ctx->leader_bank, slot );
832 0 : fd_bank_parent_slot_set( ctx->leader_bank, parent_slot );
833 0 : ctx->leader_bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
834 : /* prepare the funk transaction for the leader bank */
835 0 : fd_funk_txn_xid_t xid = { .ul = { slot, ctx->leader_bank->idx } };
836 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
837 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
838 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
839 :
840 0 : fd_bank_execution_fees_set( ctx->leader_bank, 0UL );
841 0 : fd_bank_priority_fees_set( ctx->leader_bank, 0UL );
842 0 : fd_bank_shred_cnt_set( ctx->leader_bank, 0UL );
843 0 : fd_bank_tips_set( ctx->leader_bank, 0UL );
844 :
845 : /* Set the tick height. */
846 0 : fd_bank_tick_height_set( ctx->leader_bank, fd_bank_max_tick_height_get( ctx->leader_bank ) );
847 :
848 : /* Update block height. */
849 0 : fd_bank_block_height_set( ctx->leader_bank, fd_bank_block_height_get( ctx->leader_bank ) + 1UL );
850 :
851 0 : ulong * max_tick_height = fd_bank_max_tick_height_modify( ctx->leader_bank );
852 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( ctx->leader_bank );
853 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS != fd_runtime_compute_max_tick_height( ticks_per_slot, slot, max_tick_height ) ) ) {
854 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, ticks_per_slot ));
855 0 : }
856 :
857 0 : int is_epoch_boundary = 0;
858 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
859 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_stake_weights( ctx, stem, ctx->leader_bank, 1 );
860 :
861 : /* Now that a bank has been created for the leader slot, increment the
862 : reference count until we are done with the leader slot. */
863 0 : ctx->leader_bank->refcnt++;
864 :
865 0 : return ctx->leader_bank;
866 0 : }
867 :
868 : static void
869 : fini_leader_bank( fd_replay_tile_t * ctx,
870 0 : fd_stem_context_t * stem ) {
871 :
872 0 : FD_TEST( ctx->leader_bank!=NULL );
873 0 : FD_TEST( ctx->is_leader );
874 0 : FD_TEST( ctx->recv_block_id );
875 0 : FD_TEST( ctx->recv_poh );
876 :
877 0 : ctx->leader_bank->last_transaction_finished_nanos = fd_log_wallclock();
878 :
879 0 : ulong curr_slot = fd_bank_slot_get( ctx->leader_bank );
880 :
881 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->idx, ctx->leader_bank->parent_idx, curr_slot );
882 :
883 : /* Do hashing and other end-of-block processing */
884 0 : fd_funk_txn_map_t * txn_map = fd_funk_txn_map( ctx->accdb->funk );
885 0 : if( FD_UNLIKELY( !txn_map->map ) ) {
886 0 : FD_LOG_ERR(( "Could not find valid funk transaction map" ));
887 0 : }
888 :
889 0 : fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx );
890 :
891 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
892 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
893 0 : fd_funk_txn_xid_t xid = { .ul = { curr_slot, ctx->leader_bank->idx } };
894 0 : slot_info->identity_balance = FD_UNLIKELY( curr_slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
895 :
896 0 : fd_banks_mark_bank_frozen( ctx->banks, ctx->leader_bank );
897 :
898 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( ctx->leader_bank );
899 0 : FD_TEST( bank_hash );
900 :
901 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
902 :
903 : /* The reference on the bank is finally no longer needed. */
904 0 : ctx->leader_bank->refcnt--;
905 :
906 : /* We are no longer leader so we can clear the bank index we use for
907 : being the leader. */
908 0 : ctx->leader_bank = NULL;
909 0 : ctx->recv_block_id = 0;
910 0 : ctx->recv_poh = 0;
911 0 : ctx->is_leader = 0;
912 0 : }
913 :
914 : static void
915 : publish_root_advanced( fd_replay_tile_t * ctx,
916 0 : fd_stem_context_t * stem ) {
917 :
918 : /* FIXME: for now we want to send the child of the consensus root to
919 : avoid data races with funk root advancing. This is a temporary
920 : hack because currently it is not safe to query against the xid for
921 : the root that is being advanced in funk. This doesn't eliminate
922 : the data race that exists in funk, but reduces how often it occurs.
923 :
924 : Case that causes a data race:
925 : replay: we are advancing the root from slot A->B
926 : resolv: we are resolving ALUTs against slot B */
927 :
928 0 : fd_bank_t * consensus_root_bank = fd_banks_bank_query( ctx->banks, ctx->consensus_root_bank_idx );
929 0 : if( FD_UNLIKELY( !consensus_root_bank ) ) {
930 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank is NULL at bank index %lu", ctx->consensus_root_bank_idx ));
931 0 : }
932 :
933 0 : if( FD_UNLIKELY( consensus_root_bank->child_idx==ULONG_MAX ) ) {
934 0 : return;
935 0 : }
936 :
937 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, consensus_root_bank->child_idx );
938 0 : if( FD_UNLIKELY( !bank ) ) {
939 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank child is NULL at bank index %lu", consensus_root_bank->child_idx ));
940 0 : }
941 :
942 : /* Increment the reference count on the consensus root bank to account
943 : for the number of exec tiles that are waiting on it. */
944 0 : bank->refcnt += ctx->resolv_tile_cnt;
945 :
946 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
947 0 : msg->bank_idx = bank->idx;
948 :
949 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_ROOT_ADVANCED, ctx->replay_out->chunk, sizeof(fd_replay_root_advanced_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
950 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_root_advanced_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
951 0 : }
952 :
953 : /* init_funk performs pre-flight checks for the account database and
954 : program cache. Ensures that the account database was set up
955 : correctly by bootstrap components (e.g. genesis or snapshot loader).
956 : Mirrors the account database's fork tree down to the program cache. */
957 :
958 : static void
959 : init_funk( fd_replay_tile_t * ctx,
960 0 : ulong bank_slot ) {
961 : /* Ensure that the loaded bank root corresponds to the account
962 : database's root. */
963 0 : fd_funk_t * funk = ctx->accdb_admin->funk;
964 0 : if( FD_UNLIKELY( !funk->shmem ) ) {
965 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to database shared memory objects" ));
966 0 : }
967 0 : fd_funk_txn_xid_t const * accdb_pub = fd_funk_last_publish( funk );
968 0 : if( FD_UNLIKELY( accdb_pub->ul[0]!=bank_slot ) ) {
969 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb is at slot %lu, but chain state is at slot %lu\n"
970 0 : "This is a bug in startup components.",
971 0 : accdb_pub->ul[0], bank_slot ));
972 0 : }
973 0 : if( FD_UNLIKELY( fd_funk_last_publish_is_frozen( funk ) ) ) {
974 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb fork graph is not clean.\n"
975 0 : "The account database should only contain state for the root slot at this point,\n"
976 0 : "but there are incomplete database transactions leftover.\n"
977 0 : "This is a bug in startup components." ));
978 0 : }
979 :
980 : /* The program cache tracks the account database's fork graph at all
981 : times. Perform initial synchronization: pivot from funk 'root' (a
982 : sentinel XID) to 'last publish' (the bootstrap root slot). */
983 0 : if( FD_UNLIKELY( !ctx->progcache_admin->funk->shmem ) ) {
984 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to program cache" ));
985 0 : }
986 0 : fd_progcache_clear( ctx->progcache_admin );
987 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, fd_funk_root( ctx->progcache_admin->funk ), fd_funk_last_publish( ctx->accdb->funk ) );
988 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, fd_funk_last_publish( ctx->accdb->funk ) );
989 0 : }
990 :
991 : static void
992 0 : init_after_snapshot( fd_replay_tile_t * ctx ) {
993 : /* Now that the snapshot has been loaded in, we have to refresh the
994 : stake delegations since the manifest does not contain the full set
995 : of data required for the stake delegations. See
996 : fd_stake_delegations.h for why this is required. */
997 :
998 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
999 0 : if( FD_UNLIKELY( !bank ) ) {
1000 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1001 0 : }
1002 :
1003 0 : fd_funk_txn_xid_t xid = { .ul = { fd_bank_slot_get( bank ), bank->idx } };
1004 0 : init_funk( ctx, fd_bank_slot_get( bank ) );
1005 :
1006 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
1007 :
1008 0 : fd_stake_delegations_refresh( root_delegations, ctx->accdb->funk, &xid );
1009 :
1010 : /* After both snapshots have been loaded in, we can determine if we should
1011 : start distributing rewards. */
1012 :
1013 0 : fd_rewards_recalculate_partitioned_rewards( ctx->banks, bank, ctx->accdb->funk, &xid, &ctx->runtime_stack, ctx->capture_ctx );
1014 :
1015 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1016 0 : if( FD_UNLIKELY( !snapshot_slot ) ) {
1017 : /* Genesis-specific setup. */
1018 : /* FIXME: This branch does not set up a new block exec ctx
1019 : properly. Needs to do whatever prepare_new_block_execution
1020 : does, but just hacking that in breaks stuff. */
1021 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1022 :
1023 0 : ulong hashcnt_per_slot = fd_bank_hashes_per_tick_get( bank ) * fd_bank_ticks_per_slot_get( bank );
1024 0 : fd_hash_t * poh = fd_bank_poh_modify( bank );
1025 0 : while( hashcnt_per_slot-- ) {
1026 0 : fd_sha256_hash( poh->hash, 32UL, poh->hash );
1027 0 : }
1028 :
1029 0 : int is_epoch_boundary = 0;
1030 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1031 0 : FD_TEST( !is_epoch_boundary );
1032 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
1033 :
1034 0 : snapshot_slot = 0UL;
1035 0 : }
1036 :
1037 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) fd_solcap_writer_flush( ctx->capture_ctx->capture );
1038 0 : }
1039 :
1040 : static inline int
1041 : maybe_become_leader( fd_replay_tile_t * ctx,
1042 0 : fd_stem_context_t * stem ) {
1043 0 : FD_TEST( ctx->is_booted );
1044 0 : if( FD_LIKELY( ctx->next_leader_slot==ULONG_MAX || ctx->is_leader || !ctx->has_identity_vote_rooted || ctx->replay_out->idx==ULONG_MAX ) ) return 0;
1045 :
1046 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1047 0 : long now = fd_tickcount();
1048 0 : if( FD_LIKELY( now<ctx->next_leader_tickcount ) ) return 0;
1049 :
1050 : /* TODO:
1051 : if( FD_UNLIKELY( ctx->halted_switching_key ) ) return 0; */
1052 :
1053 : /* If a prior leader is still in the process of publishing their slot,
1054 : delay ours to let them finish ... unless they are so delayed that
1055 : we risk getting skipped by the leader following us. 1.2 seconds
1056 : is a reasonable default here, although any value between 0 and 1.6
1057 : seconds could be considered reasonable. This is arbitrary and
1058 : chosen due to intuition. */
1059 0 : if( FD_UNLIKELY( now<ctx->next_leader_tickcount+(long)(3.0*ctx->slot_duration_ticks) ) ) {
1060 0 : FD_TEST( ctx->reset_bank );
1061 :
1062 : /* TODO: Make the max_active_descendant calculation more efficient
1063 : by caching it in the bank structure and updating it as banks are
1064 : created and completed. */
1065 0 : ulong max_active_descendant = 0UL;
1066 0 : ulong child_idx = ctx->reset_bank->child_idx;
1067 0 : while( child_idx!=ULONG_MAX ) {
1068 0 : fd_bank_t const * child_bank = fd_banks_bank_query( ctx->banks, child_idx );
1069 0 : max_active_descendant = fd_ulong_max( max_active_descendant, fd_bank_slot_get( child_bank ) );
1070 0 : child_idx = child_bank->sibling_idx;
1071 0 : }
1072 :
1073 : /* If the max_active_descendant is >= next_leader_slot, we waited
1074 : too long and a leader after us started publishing to try and skip
1075 : us. Just start our leader slot immediately, we might win ... */
1076 0 : if( FD_LIKELY( max_active_descendant>=ctx->reset_slot && max_active_descendant<ctx->next_leader_slot ) ) {
1077 : /* If one of the leaders between the reset slot and our leader
1078 : slot is in the process of publishing (they have a descendant
1079 : bank that is in progress of being replayed), then keep waiting.
1080 : We probably wouldn't get a leader slot out before they
1081 : finished.
1082 :
1083 : Unless... we are past the deadline to start our slot by more
1084 : than 1.2 seconds, in which case we should probably start it to
1085 : avoid getting skipped by the leader behind us. */
1086 0 : return 0;
1087 0 : }
1088 0 : }
1089 :
1090 0 : long now_nanos = fd_log_wallclock();
1091 :
1092 0 : ctx->is_leader = 1;
1093 0 : ctx->recv_poh = 0;
1094 0 : ctx->recv_block_id = 0;
1095 :
1096 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1097 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1098 :
1099 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1100 :
1101 : /* Acquires bank, sets up initial state, and refcnts it. */
1102 0 : fd_bank_t * bank = prepare_leader_bank( ctx, ctx->next_leader_slot, now_nanos, &ctx->reset_block_id, stem );
1103 0 : fd_funk_txn_xid_t xid = { .ul = { ctx->next_leader_slot, ctx->leader_bank->idx } };
1104 :
1105 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1106 0 : fd_acct_addr_t tip_receiver_owner[1] = { 0 };
1107 :
1108 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1109 0 : fd_acct_addr_t tip_payment_config[1];
1110 0 : fd_acct_addr_t tip_receiver[1];
1111 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, fd_bank_epoch_get( bank ), tip_payment_config, tip_receiver );
1112 :
1113 0 : fd_txn_account_t tip_config_acc[1];
1114 0 : int err = fd_txn_account_init_from_funk_readonly( tip_config_acc,
1115 0 : (fd_hash_t *)tip_payment_config->b,
1116 0 : ctx->accdb->funk,
1117 0 : &xid );
1118 0 : if( FD_UNLIKELY( err ) ) {
1119 0 : FD_LOG_CRIT(( "failed to initialize tip payment config account: err=%d", err ));
1120 0 : }
1121 0 : memcpy( config, fd_txn_account_get_data( tip_config_acc ), sizeof(fd_bundle_crank_tip_payment_config_t) );
1122 :
1123 : /* It is possible that the tip receiver account does not exist yet
1124 : if it is the first time in an epoch. */
1125 0 : fd_txn_account_t tip_receiver_acc[1];
1126 0 : err = fd_txn_account_init_from_funk_readonly( tip_receiver_acc,
1127 0 : (fd_hash_t *)tip_receiver->b,
1128 0 : ctx->accdb->funk,
1129 0 : &xid );
1130 0 : if( FD_LIKELY( !err ) ) {
1131 0 : memcpy( tip_receiver_owner, tip_receiver_acc->meta->owner, sizeof(fd_acct_addr_t) );
1132 0 : }
1133 0 : }
1134 :
1135 :
1136 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1137 0 : msg->slot = ctx->next_leader_slot;
1138 0 : msg->slot_start_ns = now_nanos;
1139 0 : msg->slot_end_ns = now_nanos+(long)ctx->slot_duration_nanos;
1140 0 : msg->bank = NULL;
1141 0 : msg->bank_idx = bank->idx;
1142 0 : msg->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1143 0 : msg->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1144 0 : msg->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)msg->ticks_per_slot);
1145 0 : msg->bundle->config[0] = config[0];
1146 0 : memcpy( msg->bundle->last_blockhash, (fd_hash_t *)fd_bank_poh_query( bank )->hash, 32UL );
1147 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner, 32UL );
1148 :
1149 :
1150 0 : if( FD_UNLIKELY( msg->hashcnt_per_tick==1UL ) ) {
1151 : /* Low power producer, maximum of one microblock per tick in the slot */
1152 0 : msg->max_microblocks_in_slot = msg->ticks_per_slot;
1153 0 : } else {
1154 : /* See the long comment in after_credit for this limit */
1155 0 : msg->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, msg->ticks_per_slot*(msg->hashcnt_per_tick-1UL) );
1156 0 : }
1157 :
1158 0 : msg->total_skipped_ticks = msg->ticks_per_slot*(ctx->next_leader_slot-ctx->reset_slot);
1159 0 : msg->epoch = fd_slot_to_epoch( fd_bank_epoch_schedule_query( bank ), ctx->next_leader_slot, NULL );
1160 :
1161 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
1162 :
1163 0 : msg->limits.slot_max_cost = ctx->larger_max_cost_per_block ? LARGER_MAX_COST_PER_BLOCK : cost_tracker->block_cost_limit;
1164 0 : msg->limits.slot_max_vote_cost = cost_tracker->vote_cost_limit;
1165 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1166 :
1167 0 : fd_bank_cost_tracker_end_locking_query( bank );
1168 :
1169 0 : if( FD_UNLIKELY( msg->ticks_per_slot+msg->total_skipped_ticks>USHORT_MAX ) ) {
1170 : /* There can be at most USHORT_MAX skipped ticks, because the
1171 : parent_offset field in the shred data is only 2 bytes wide. */
1172 0 : FD_LOG_ERR(( "too many skipped ticks %lu for slot %lu, chain must halt", msg->ticks_per_slot+msg->total_skipped_ticks, ctx->next_leader_slot ));
1173 0 : }
1174 :
1175 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_BECAME_LEADER, ctx->replay_out->chunk, sizeof(fd_became_leader_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1176 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_became_leader_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
1177 :
1178 0 : ctx->next_leader_slot = ULONG_MAX;
1179 0 : ctx->next_leader_tickcount = LONG_MAX;
1180 :
1181 0 : return 1;
1182 0 : }
1183 :
1184 : static void
1185 : process_poh_message( fd_replay_tile_t * ctx,
1186 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
1187 :
1188 0 : FD_TEST( ctx->is_booted );
1189 0 : FD_TEST( ctx->is_leader );
1190 0 : FD_TEST( ctx->leader_bank!=NULL );
1191 :
1192 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
1193 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
1194 :
1195 : /* Update the poh hash in the bank. We will want to maintain a refcnt
1196 : on the bank until we have recieved the block id for the block after
1197 : it has been shredded. */
1198 :
1199 0 : memcpy( fd_bank_poh_modify( ctx->leader_bank ), slot_ended->blockhash, sizeof(fd_hash_t) );
1200 :
1201 0 : ctx->recv_poh = 1;
1202 0 : }
1203 :
1204 : static void
1205 : publish_reset( fd_replay_tile_t * ctx,
1206 : fd_stem_context_t * stem,
1207 0 : fd_bank_t * bank ) {
1208 0 : if( FD_UNLIKELY( ctx->replay_out->idx==ULONG_MAX ) ) return;
1209 :
1210 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1211 0 : FD_TEST( block_hash );
1212 :
1213 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1214 :
1215 0 : reset->bank_idx = bank->idx;
1216 0 : reset->timestamp = fd_log_wallclock();
1217 0 : reset->completed_slot = fd_bank_slot_get( bank );
1218 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1219 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1220 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
1221 0 : fd_memcpy( reset->completed_blockhash, block_hash->uc, sizeof(fd_hash_t) );
1222 :
1223 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1224 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
1225 : /* Low power producer, maximum of one microblock per tick in the slot */
1226 0 : reset->max_microblocks_in_slot = ticks_per_slot;
1227 0 : } else {
1228 : /* See the long comment in after_credit for this limit */
1229 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
1230 0 : }
1231 0 : reset->next_leader_slot = ctx->next_leader_slot;
1232 :
1233 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++;
1234 :
1235 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_RESET, ctx->replay_out->chunk, sizeof(fd_poh_reset_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1236 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_poh_reset_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
1237 0 : }
1238 :
1239 : static void
1240 : boot_genesis( fd_replay_tile_t * ctx,
1241 : fd_stem_context_t * stem,
1242 : ulong in_idx,
1243 0 : ulong chunk ) {
1244 : /* If we are bootstrapping, we can't wait to wait for our identity
1245 : vote to be rooted as this creates a circular dependency. */
1246 0 : ctx->has_identity_vote_rooted = 1;
1247 :
1248 0 : uchar const * lthash = (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1249 0 : uchar const * genesis_hash = (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk )+sizeof(fd_lthash_value_t);
1250 :
1251 : // TODO: Do not pass the fd_types type between tiles, it have offsets
1252 : // that are unsafe and can't be validated as being in-bounds. Need to
1253 : // pass an actual owned genesis type.
1254 0 : fd_genesis_solana_global_t const * genesis = fd_type_pun( (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk )+sizeof(fd_hash_t)+sizeof(fd_lthash_value_t) );
1255 :
1256 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1257 0 : FD_TEST( bank );
1258 0 : fd_funk_txn_xid_t xid = { .ul = { 0UL, FD_REPLAY_BOOT_BANK_IDX } };
1259 :
1260 : /* Do genesis-related processing in a non-rooted transaction */
1261 0 : fd_funk_txn_xid_t root_xid; fd_funk_txn_xid_set_root( &root_xid );
1262 0 : fd_funk_txn_xid_t target_xid = { .ul = { 0UL, 0UL } };
1263 0 : fd_accdb_attach_child( ctx->accdb_admin, &root_xid, &target_xid );
1264 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, &xid, NULL, fd_type_pun_const( genesis_hash ), fd_type_pun_const( lthash ), genesis, &ctx->runtime_stack );
1265 0 : fd_accdb_advance_root( ctx->accdb_admin, &target_xid );
1266 :
1267 0 : static const fd_txncache_fork_id_t txncache_root = { .val = USHORT_MAX };
1268 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, txncache_root );
1269 :
1270 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1271 0 : fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
1272 :
1273 0 : publish_stake_weights( ctx, stem, bank, 0 );
1274 0 : publish_stake_weights( ctx, stem, bank, 1 );
1275 :
1276 : /* We call this after fd_runtime_read_genesis, which sets up the
1277 : slot_bank needed in blockstore_init. */
1278 0 : init_after_snapshot( ctx );
1279 :
1280 : /* Initialize store for genesis case, similar to snapshot case */
1281 0 : fd_hash_t genesis_block_id = { .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
1282 0 : fd_store_exacq( ctx->store );
1283 0 : if( FD_UNLIKELY( fd_store_root( ctx->store ) ) ) {
1284 0 : FD_LOG_CRIT(( "invariant violation: store root is not 0 for genesis" ));
1285 0 : }
1286 0 : fd_store_insert( ctx->store, 0, &genesis_block_id );
1287 0 : ctx->store->slot0 = 0UL; /* Genesis slot */
1288 0 : fd_store_exrel( ctx->store );
1289 :
1290 0 : ctx->published_root_slot = 0UL;
1291 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, 0UL );
1292 :
1293 0 : fd_bank_block_height_set( bank, 1UL );
1294 :
1295 0 : ctx->consensus_root = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
1296 0 : ctx->consensus_root_slot = 0UL;
1297 0 : ctx->consensus_root_bank_idx = 0UL;
1298 0 : ctx->published_root_slot = 0UL;
1299 0 : ctx->published_root_bank_idx = 0UL;
1300 :
1301 0 : ctx->reset_slot = 0UL;
1302 0 : ctx->reset_bank = bank;
1303 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1304 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1305 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1306 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1307 0 : } else {
1308 0 : ctx->next_leader_tickcount = LONG_MAX;
1309 0 : }
1310 :
1311 0 : ctx->is_booted = 1;
1312 0 : maybe_become_leader( ctx, stem );
1313 :
1314 0 : fd_hash_t initial_block_id = { .ul = { FD_RUNTIME_INITIAL_BLOCK_ID } };
1315 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &initial_block_id, NULL, 0 /* genesis slot */, 0, 0, 0, 0, 1, 0 ); /* FIXME manifest block_id */
1316 0 : fec->bank_idx = 0UL;
1317 :
1318 :
1319 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1320 0 : FD_TEST( block_id_ele );
1321 0 : block_id_ele->block_id = initial_block_id;
1322 0 : block_id_ele->slot = 0UL;
1323 :
1324 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1325 :
1326 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1327 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1328 :
1329 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1330 0 : publish_root_advanced( ctx, stem );
1331 0 : publish_reset( ctx, stem, bank );
1332 0 : }
1333 :
1334 : static inline void
1335 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
1336 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->has_genesis_hash ) ) {
1337 0 : return;
1338 0 : }
1339 :
1340 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash, hash_cstr );
1341 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
1342 : /* Map pyth-related clusters to unkwown. */
1343 0 : switch( cluster ) {
1344 0 : case FD_CLUSTER_PYTHNET:
1345 0 : case FD_CLUSTER_PYTHTEST:
1346 0 : cluster = FD_CLUSTER_UNKNOWN;
1347 0 : }
1348 :
1349 0 : if( cluster!=ctx->cluster_type ) {
1350 0 : FD_LOG_ERR(( "cluster type mismatch: (genesis) %s != (snapshot) %s", fd_genesis_cluster_name( cluster ), fd_genesis_cluster_name( ctx->cluster_type ) ));
1351 0 : }
1352 0 : }
1353 :
1354 : static void
1355 : on_snapshot_message( fd_replay_tile_t * ctx,
1356 : fd_stem_context_t * stem,
1357 : ulong in_idx,
1358 : ulong chunk,
1359 0 : ulong sig ) {
1360 0 : ulong msg = fd_ssmsg_sig_message( sig );
1361 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
1362 : /* An end of message notification indicates the snapshot is loaded.
1363 : Replay is able to start executing from this point onwards. */
1364 : /* TODO: replay should finish booting. Could make replay a
1365 : state machine and set the state here accordingly. */
1366 0 : ctx->is_booted = 1;
1367 :
1368 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1369 0 : if( FD_UNLIKELY( !bank ) ) {
1370 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1371 0 : }
1372 :
1373 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1374 : /* FIXME: This is a hack because the block id of the snapshot slot
1375 : is not provided in the snapshot. A possible solution is to get
1376 : the block id of the snapshot slot from repair. */
1377 0 : fd_hash_t manifest_block_id = { .ul = { FD_RUNTIME_INITIAL_BLOCK_ID } };
1378 :
1379 0 : fd_store_exacq( ctx->store );
1380 0 : FD_TEST( !fd_store_root( ctx->store ) );
1381 0 : fd_store_insert( ctx->store, 0, &manifest_block_id );
1382 0 : ctx->store->slot0 = snapshot_slot; /* FIXME manifest_block_id */
1383 0 : fd_store_exrel( ctx->store );
1384 :
1385 : /* Typically, when we cross an epoch boundary during normal
1386 : operation, we publish the stake weights for the new epoch. But
1387 : since we are starting from a snapshot, we need to publish two
1388 : epochs worth of stake weights: the previous epoch (which is
1389 : needed for voting on the current epoch), and the current epoch
1390 : (which is needed for voting on the next epoch). */
1391 0 : publish_stake_weights( ctx, stem, bank, 0 );
1392 0 : publish_stake_weights( ctx, stem, bank, 1 );
1393 :
1394 0 : ctx->consensus_root = manifest_block_id;
1395 0 : ctx->consensus_root_slot = snapshot_slot;
1396 0 : ctx->consensus_root_bank_idx = 0UL;
1397 0 : ctx->published_root_slot = ctx->consensus_root_slot;
1398 0 : ctx->published_root_bank_idx = 0UL;
1399 :
1400 0 : ctx->reset_slot = snapshot_slot;
1401 0 : ctx->reset_bank = bank;
1402 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1403 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1404 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1405 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1406 0 : } else {
1407 0 : ctx->next_leader_tickcount = LONG_MAX;
1408 0 : }
1409 :
1410 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, snapshot_slot );
1411 0 : FD_TEST( bank->idx==0UL );
1412 :
1413 0 : fd_funk_txn_xid_t xid = { .ul = { snapshot_slot, FD_REPLAY_BOOT_BANK_IDX } };
1414 :
1415 0 : fd_features_restore( bank, ctx->accdb->funk, &xid );
1416 :
1417 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1418 :
1419 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1420 0 : FD_TEST( block_id_ele );
1421 0 : block_id_ele->block_id = manifest_block_id;
1422 0 : block_id_ele->slot = snapshot_slot;
1423 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1424 :
1425 : /* We call this after fd_runtime_read_genesis, which sets up the
1426 : slot_bank needed in blockstore_init. */
1427 0 : init_after_snapshot( ctx );
1428 :
1429 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1430 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1431 :
1432 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1433 0 : publish_root_advanced( ctx, stem );
1434 :
1435 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &manifest_block_id, NULL, snapshot_slot, 0, 0, 0, 0, 1, 0 ); /* FIXME manifest block_id */
1436 0 : fec->bank_idx = 0UL;
1437 :
1438 0 : ctx->cluster_type = fd_bank_cluster_type_get( bank );
1439 :
1440 0 : maybe_verify_cluster_type( ctx );
1441 :
1442 0 : return;
1443 0 : }
1444 :
1445 0 : switch( msg ) {
1446 0 : case FD_SSMSG_MANIFEST_FULL:
1447 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
1448 : /* We may either receive a full snapshot manifest or an
1449 : incremental snapshot manifest. Note that this external message
1450 : id is only used temporarily because replay cannot yet receive
1451 : the firedancer-internal snapshot manifest message. */
1452 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1453 0 : FD_LOG_ERR(( "chunk %lu from in %d corrupt, not in range [%lu,%lu]", chunk, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
1454 :
1455 0 : fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
1456 0 : ctx->banks,
1457 0 : fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
1458 0 : ctx->runtime_stack.stakes.vote_credits );
1459 :
1460 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1461 0 : ctx->hard_forks_cnt = manifest->hard_forks_len;
1462 0 : for( ulong i=0UL; i<manifest->hard_forks_len; i++ ) ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
1463 0 : break;
1464 0 : }
1465 0 : default: {
1466 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
1467 0 : return;
1468 0 : }
1469 0 : }
1470 :
1471 0 : return;
1472 0 : }
1473 :
1474 : static void
1475 : dispatch_task( fd_replay_tile_t * ctx,
1476 : fd_stem_context_t * stem,
1477 0 : fd_sched_task_t * task ) {
1478 :
1479 0 : switch( task->task_type ) {
1480 0 : case FD_SCHED_TT_TXN_EXEC: {
1481 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
1482 :
1483 : /* FIXME: this should be done during txn parsing so that we don't
1484 : have to loop over all accounts a second time. */
1485 : /* Insert or reverify invoked programs for this epoch, if needed. */
1486 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_exec->bank_idx );
1487 :
1488 0 : # if FD_HAS_FLATCC
1489 : /* Add the transaction to the block dumper if necessary. This
1490 : logic doesn't need to be fork-aware since it's only meant to
1491 : be used in backtest. */
1492 0 : if( FD_UNLIKELY( ctx->capture_ctx && ctx->capture_ctx->dump_block_to_pb ) ) {
1493 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
1494 0 : }
1495 0 : # endif
1496 :
1497 0 : bank->refcnt++;
1498 :
1499 0 : if( FD_UNLIKELY( !bank->first_transaction_scheduled_nanos ) ) bank->first_transaction_scheduled_nanos = fd_log_wallclock();
1500 :
1501 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1502 0 : fd_exec_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1503 0 : memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1504 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
1505 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
1506 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_TXN_EXEC<<32) | task->txn_exec->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1507 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1508 0 : break;
1509 0 : }
1510 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1511 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
1512 :
1513 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_sigverify->bank_idx );
1514 0 : bank->refcnt++;
1515 :
1516 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1517 0 : fd_exec_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1518 0 : memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1519 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
1520 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
1521 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1522 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1523 0 : break;
1524 0 : };
1525 0 : default: {
1526 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1527 0 : }
1528 0 : }
1529 0 : }
1530 :
1531 : /* Returns 1 if charge_busy. */
1532 : static int
1533 : replay( fd_replay_tile_t * ctx,
1534 0 : fd_stem_context_t * stem ) {
1535 :
1536 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
1537 :
1538 0 : int charge_busy = 0;
1539 0 : fd_sched_task_t task[ 1 ];
1540 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
1541 0 : return charge_busy; /* Nothing to execute or do. */
1542 0 : }
1543 :
1544 0 : charge_busy = 1;
1545 :
1546 0 : switch( task->task_type ) {
1547 0 : case FD_SCHED_TT_BLOCK_START: {
1548 0 : replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
1549 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX );
1550 0 : break;
1551 0 : }
1552 0 : case FD_SCHED_TT_BLOCK_END: {
1553 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->block_end->bank_idx );
1554 0 : if( FD_LIKELY( !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) replay_block_finalize( ctx, stem, bank );
1555 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX );
1556 0 : break;
1557 0 : }
1558 0 : case FD_SCHED_TT_TXN_EXEC:
1559 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1560 : /* Likely/common case: we have a transaction we actually need to
1561 : execute. */
1562 0 : dispatch_task( ctx, stem, task );
1563 0 : break;
1564 0 : }
1565 0 : default: {
1566 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1567 0 : }
1568 0 : }
1569 :
1570 0 : return charge_busy;
1571 0 : }
1572 :
1573 : static int
1574 0 : can_process_fec( fd_replay_tile_t * ctx ) {
1575 0 : fd_reasm_fec_t * fec;
1576 0 : if( FD_UNLIKELY( !fd_sched_can_ingest( ctx->sched, 1UL ) ) ) return 0;
1577 0 : if( FD_UNLIKELY( (fec = fd_reasm_peek( ctx->reasm ))==NULL ) ) return 0;
1578 :
1579 0 : if( FD_UNLIKELY( ctx->is_leader && fec->fec_set_idx==0U && fd_reasm_parent( ctx->reasm, fec )->bank_idx==ctx->leader_bank->idx ) ) {
1580 : /* There's a race that's exceedingly rare, where we receive the
1581 : FEC set for the slot right after our leader rotation before we
1582 : freeze the bank for the last slot in our leader rotation.
1583 : Leader slot freezing happens only after if we've received the
1584 : final PoH hash from the poh tile as well as the final FEC set
1585 : for the leader slot. So the race happens when FEC sets are
1586 : delivered and processed sooner than the PoH hash, aka when the
1587 : poh=>shred=>replay path for the block id somehow beats the
1588 : poh=>replay path for the poh hash. To mitigate this race,
1589 : we must block on ingesting the FEC set for the ensuing slot
1590 : before the leader bank freezes, because that would violate
1591 : ordering invariants in banks and sched. */
1592 0 : FD_TEST( ctx->recv_block_id );
1593 0 : FD_TEST( !ctx->recv_poh );
1594 0 : return 0;
1595 0 : }
1596 :
1597 : /* If fec_set_idx is 0, we need a new bank for a new slot. Banks must
1598 : not be full in this case. */
1599 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) && fec->fec_set_idx==0 ) ) return 0;
1600 :
1601 : /* Otherwise, banks may not be full, so we can always create a new
1602 : bank if needed. Or, if banks are full, the current fec set's
1603 : ancestor (idx 0) already created a bank for this slot.*/
1604 0 : return 1;
1605 0 : }
1606 :
1607 : static void
1608 : process_fec_set( fd_replay_tile_t * ctx,
1609 0 : fd_reasm_fec_t * reasm_fec ) {
1610 0 : long now = fd_log_wallclock();
1611 :
1612 : /* Linking only requires a shared lock because the fields that are
1613 : modified are only read on publish which uses exclusive lock. */
1614 :
1615 0 : long shacq_start, shacq_end, shrel_end;
1616 :
1617 0 : FD_STORE_SHARED_LOCK( ctx->store, shacq_start, shacq_end, shrel_end ) {
1618 0 : if( FD_UNLIKELY( !fd_store_link( ctx->store, &reasm_fec->key, &reasm_fec->cmr ) ) ) FD_LOG_WARNING(( "failed to link %s %s. slot %lu fec_set_idx %u", FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ), FD_BASE58_ENC_32_ALLOCA( &reasm_fec->cmr ), reasm_fec->slot, reasm_fec->fec_set_idx ));
1619 0 : } FD_STORE_SHARED_LOCK_END;
1620 0 : fd_histf_sample( ctx->metrics.store_link_wait, (ulong)fd_long_max( shacq_end - shacq_start, 0L ) );
1621 0 : fd_histf_sample( ctx->metrics.store_link_work, (ulong)fd_long_max( shrel_end - shacq_end, 0L ) );
1622 :
1623 : /* Update the reasm_fec with the correct bank index and parent bank
1624 : index. If the FEC belongs to a leader, we have already allocated
1625 : a bank index for the FEC and it just needs to be propagated to the
1626 : reasm_fec. */
1627 :
1628 0 : reasm_fec->parent_bank_idx = fd_reasm_parent( ctx->reasm, reasm_fec )->bank_idx;
1629 :
1630 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1631 : /* If we are the leader we just need to copy in the bank index that
1632 : the leader slot is using. */
1633 0 : FD_TEST( ctx->leader_bank!=NULL );
1634 0 : reasm_fec->bank_idx = ctx->leader_bank->idx;
1635 0 : } else if( FD_UNLIKELY( reasm_fec->fec_set_idx==0U ) ) {
1636 : /* If we are seeing a FEC with fec set idx 0, this means that we are
1637 : starting a new slot, and we need a new bank index. */
1638 0 : reasm_fec->bank_idx = fd_banks_new_bank( ctx->banks, reasm_fec->parent_bank_idx, now )->idx;
1639 0 : } else {
1640 : /* We are continuing to execute through a slot that we already have
1641 : a bank index for. */
1642 0 : reasm_fec->bank_idx = reasm_fec->parent_bank_idx;
1643 0 : }
1644 :
1645 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) {
1646 : /* Once the block id for a block is known it must be added to the
1647 : leader block mapping. */
1648 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1649 0 : FD_TEST( block_id_ele );
1650 :
1651 : /* If an entry already exists for this bank index in the block id
1652 : map, we can safely remove it and replace it with the new entry.
1653 : This is safe because we know that the old entry for this fork
1654 : index has already been pruned away. */
1655 0 : if( FD_LIKELY( block_id_ele->slot!=FD_SLOT_NULL && fd_block_id_map_ele_query( ctx->block_id_map, &block_id_ele->block_id, NULL, ctx->block_id_arr ) ) ) {
1656 0 : FD_TEST( fd_block_id_map_ele_remove( ctx->block_id_map, &block_id_ele->block_id, NULL, ctx->block_id_arr ) );
1657 0 : }
1658 :
1659 0 : block_id_ele->block_id = reasm_fec->key;
1660 0 : block_id_ele->slot = reasm_fec->slot;
1661 :
1662 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1663 :
1664 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1665 0 : ctx->recv_block_id = 1;
1666 0 : }
1667 0 : }
1668 :
1669 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1670 0 : return;
1671 0 : }
1672 :
1673 : /* Forks form a partial ordering over FEC sets. The Repair tile
1674 : delivers FEC sets in-order per fork, but FEC set ordering across
1675 : forks is arbitrary */
1676 0 : fd_sched_fec_t sched_fec[ 1 ];
1677 :
1678 : # if DEBUG_LOGGING
1679 : FD_LOG_INFO(( "replay processing FEC set for slot %lu fec_set_idx %u, mr %s cmr %s", reasm_fec->slot, reasm_fec->fec_set_idx, FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ), FD_BASE58_ENC_32_ALLOCA( &reasm_fec->cmr ) ));
1680 : # endif
1681 :
1682 : /* Read FEC set from the store. This should happen before we try to
1683 : ingest the FEC set. This allows us to filter out frags that were
1684 : in-flight when we published away minority forks that the frags land
1685 : on. These frags would have no bank to execute against, because
1686 : their corresponding banks, or parent banks, have also been pruned
1687 : during publishing. A query against store will rightfully tell us
1688 : that the underlying data is not found, implying that this is for a
1689 : minority fork that we can safely ignore. */
1690 0 : FD_STORE_SHARED_LOCK( ctx->store, shacq_start, shacq_end, shrel_end ) {
1691 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->store, &reasm_fec->key );
1692 0 : if( FD_UNLIKELY( !store_fec ) ) {
1693 : /* The only case in which a FEC is not found in the store after
1694 : repair has notified is if the FEC was on a minority fork that
1695 : has already been published away. In this case we abandon the
1696 : entire slice because it is no longer relevant. */
1697 0 : FD_LOG_WARNING(( "store fec for slot: %lu is on minority fork already pruned by publish. abandoning slice. root: %lu. pruned merkle: %s", reasm_fec->slot, ctx->consensus_root_slot, FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ) ));
1698 0 : return;
1699 0 : }
1700 0 : FD_TEST( store_fec );
1701 0 : sched_fec->fec = store_fec;
1702 0 : sched_fec->shred_cnt = reasm_fec->data_cnt;
1703 0 : } FD_STORE_SHARED_LOCK_END;
1704 :
1705 0 : fd_histf_sample( ctx->metrics.store_read_wait, (ulong)fd_long_max( shacq_end - shacq_start, 0UL ) );
1706 0 : fd_histf_sample( ctx->metrics.store_read_work, (ulong)fd_long_max( shrel_end - shacq_end, 0UL ) );
1707 :
1708 0 : sched_fec->is_last_in_batch = !!reasm_fec->data_complete;
1709 0 : sched_fec->is_last_in_block = !!reasm_fec->slot_complete;
1710 0 : sched_fec->bank_idx = reasm_fec->bank_idx;
1711 0 : sched_fec->parent_bank_idx = reasm_fec->parent_bank_idx;
1712 0 : sched_fec->slot = reasm_fec->slot;
1713 0 : sched_fec->parent_slot = reasm_fec->slot - reasm_fec->parent_off;
1714 0 : sched_fec->is_first_in_block = reasm_fec->fec_set_idx==0U;
1715 0 : fd_funk_txn_xid_copy( sched_fec->alut_ctx->xid, fd_funk_last_publish( ctx->accdb->funk ) );
1716 0 : sched_fec->alut_ctx->accdb[0] = ctx->accdb[0];
1717 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
1718 :
1719 0 : if( FD_UNLIKELY( !fd_sched_fec_ingest( ctx->sched, sched_fec ) ) ) {
1720 0 : fd_banks_mark_bank_dead( ctx->banks, fd_banks_bank_query( ctx->banks, sched_fec->bank_idx ) );
1721 0 : }
1722 0 : }
1723 :
1724 : static void
1725 : funk_publish( fd_replay_tile_t * ctx,
1726 : ulong slot,
1727 0 : ulong bank_idx ) {
1728 0 : fd_funk_txn_xid_t xid = { .ul[0] = slot, .ul[1] = bank_idx };
1729 0 : FD_LOG_DEBUG(( "publishing slot=%lu", slot ));
1730 :
1731 : /* This is the standard case. Publish all transactions up to and
1732 : including the watermark. This will publish any in-prep ancestors
1733 : of root_txn as well. */
1734 0 : fd_accdb_advance_root( ctx->accdb_admin, &xid );
1735 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, &xid );
1736 0 : }
1737 :
1738 : static int
1739 0 : advance_published_root( fd_replay_tile_t * ctx ) {
1740 :
1741 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &ctx->consensus_root, NULL, ctx->block_id_arr );
1742 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
1743 0 : FD_LOG_CRIT(( "invariant violation: block id ele not found for consensus root %s", FD_BASE58_ENC_32_ALLOCA( &ctx->consensus_root ) ));
1744 0 : }
1745 0 : ulong target_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
1746 :
1747 0 : fd_sched_root_notify( ctx->sched, target_bank_idx );
1748 :
1749 : /* If the identity vote has been seen on a bank that should be rooted,
1750 : then we are now ready to produce blocks. */
1751 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
1752 0 : fd_bank_t * root_bank = fd_banks_bank_query( ctx->banks, target_bank_idx );
1753 0 : if( FD_UNLIKELY( !root_bank ) ) FD_LOG_CRIT(( "invariant violation: root bank not found for bank index %lu", target_bank_idx ));
1754 0 : if( FD_LIKELY( fd_bank_has_identity_vote_get( root_bank ) ) ) ctx->has_identity_vote_rooted = 1;
1755 0 : }
1756 :
1757 0 : ulong advanceable_root_idx = ULONG_MAX;
1758 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank_idx, &advanceable_root_idx ) ) ) return 0;
1759 :
1760 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, advanceable_root_idx );
1761 0 : FD_TEST( bank );
1762 :
1763 0 : fd_block_id_ele_t * advanceable_root_ele = &ctx->block_id_arr[ advanceable_root_idx ];
1764 0 : if( FD_UNLIKELY( !advanceable_root_ele ) ) {
1765 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele not found for bank index %lu", advanceable_root_idx ));
1766 0 : }
1767 :
1768 0 : long exacq_start, exacq_end, exrel_end;
1769 0 : FD_STORE_EXCLUSIVE_LOCK( ctx->store, exacq_start, exacq_end, exrel_end ) {
1770 0 : fd_store_publish( ctx->store, &advanceable_root_ele->block_id );
1771 0 : } FD_STORE_EXCLUSIVE_LOCK_END;
1772 :
1773 0 : fd_histf_sample( ctx->metrics.store_publish_wait, (ulong)fd_long_max( exacq_end-exacq_start, 0UL ) );
1774 0 : fd_histf_sample( ctx->metrics.store_publish_work, (ulong)fd_long_max( exrel_end-exacq_end, 0UL ) );
1775 :
1776 0 : ulong advanceable_root_slot = fd_bank_slot_get( bank );
1777 0 : funk_publish( ctx, advanceable_root_slot, bank->idx );
1778 :
1779 0 : fd_txncache_advance_root( ctx->txncache, bank->txncache_fork_id );
1780 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
1781 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
1782 0 : fd_reasm_publish( ctx->reasm, &advanceable_root_ele->block_id );
1783 :
1784 0 : ctx->published_root_slot = advanceable_root_slot;
1785 0 : ctx->published_root_bank_idx = advanceable_root_idx;
1786 :
1787 0 : return 1;
1788 0 : }
1789 :
1790 : static void
1791 : after_credit( fd_replay_tile_t * ctx,
1792 : fd_stem_context_t * stem,
1793 : int * opt_poll_in,
1794 0 : int * charge_busy ) {
1795 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return;
1796 :
1797 0 : if( FD_UNLIKELY( maybe_become_leader( ctx, stem ) ) ) {
1798 0 : *charge_busy = 1;
1799 0 : *opt_poll_in = 0;
1800 0 : return;
1801 0 : }
1802 :
1803 : /* If we are leader, we can only unbecome the leader iff we have
1804 : received the poh hash from the poh tile and block id from reasm. */
1805 0 : if( FD_UNLIKELY( ctx->is_leader && ctx->recv_block_id && ctx->recv_poh ) ) {
1806 0 : fini_leader_bank( ctx, stem );
1807 0 : *charge_busy = 1;
1808 0 : *opt_poll_in = 0;
1809 0 : return;
1810 0 : }
1811 :
1812 : /* If the published_root is not caught up to the consensus root, then
1813 : we should try to advance the published root. */
1814 0 : if( FD_UNLIKELY( ctx->consensus_root_bank_idx!=ctx->published_root_bank_idx && advance_published_root( ctx ) ) ) {
1815 0 : *charge_busy = 1;
1816 0 : *opt_poll_in = 0;
1817 0 : return;
1818 0 : }
1819 :
1820 : /* If the reassembler has a fec that is ready, we should process it
1821 : and pass it to the scheduler. */
1822 :
1823 : /* FIXME: The reasm logic needs to get reworked to support
1824 : equivocation more robustly. */
1825 0 : if( FD_LIKELY( can_process_fec( ctx ) ) ) {
1826 0 : fd_reasm_fec_t * fec = fd_reasm_peek( ctx->reasm );
1827 :
1828 : /* If fec->eqvoc is set that means that equivocation mid-block was
1829 : detected in fd_reasm_t. We need to replay up to and including
1830 : the equivocating FEC on a new bank. */
1831 :
1832 0 : if( FD_UNLIKELY( fec->eqvoc ) ) {
1833 0 : FD_LOG_WARNING(( "Block equivocation detected at slot %lu", fec->slot ));
1834 :
1835 : /* We need to figure out which and how many FECs we need to
1836 : (re)insert into the scheduler. We work backwards from the
1837 : equivocating FEC, querying for chained merkle roots until we
1838 : reach the first FEC in the slot.
1839 : TODO: replace the magic number with a constant for the max
1840 : number of fecs possible in a slot with fix-32. */
1841 0 : fd_reasm_fec_t * fecs[ 1024 ] = { [0] = fec };
1842 0 : ulong fec_cnt = 1UL;
1843 0 : while( fecs[ fec_cnt-1UL ]->fec_set_idx!=0UL ) {
1844 0 : fec = fd_reasm_query( ctx->reasm, &fecs[ fec_cnt-1UL ]->cmr );
1845 0 : fecs[ fec_cnt++ ] = fec;
1846 0 : }
1847 :
1848 : /* If we don't have enough space in the scheduler to ingest all of
1849 : FECs, we can't proceed yet. */
1850 0 : if( FD_UNLIKELY( !fd_sched_can_ingest( ctx->sched, fec_cnt ) ) ) return;
1851 :
1852 : /* Now that we have validated that sched can ingest all of the
1853 : required FECs, it is finally safe to remove the equivocating
1854 : fec from the reasm deque. */
1855 0 : fd_reasm_out( ctx->reasm );
1856 :
1857 : /* Now we can process all of the FECs. */
1858 0 : for( ulong i=fec_cnt; i>0UL; i-- ) {
1859 0 : process_fec_set( ctx, fecs[i-1UL] );
1860 0 : }
1861 0 : } else {
1862 : /* Standard case. */
1863 0 : fec = fd_reasm_out( ctx->reasm );
1864 0 : process_fec_set( ctx, fec );
1865 0 : }
1866 :
1867 0 : *charge_busy = 1;
1868 0 : *opt_poll_in = 0;
1869 0 : return;
1870 0 : }
1871 :
1872 0 : *charge_busy = replay( ctx, stem );
1873 0 : *opt_poll_in = !*charge_busy;
1874 0 : }
1875 :
1876 : static int
1877 : before_frag( fd_replay_tile_t * ctx,
1878 : ulong in_idx,
1879 : ulong seq FD_PARAM_UNUSED,
1880 0 : ulong sig FD_PARAM_UNUSED ) {
1881 :
1882 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_SHRED ) ) {
1883 : /* If reasm is full, we can not insert any more FEC sets. We must
1884 : not consume any frags from shred_out until reasm can process more
1885 : FEC sets. */
1886 :
1887 0 : if( FD_UNLIKELY( !fd_reasm_free( ctx->reasm ) ) ) {
1888 0 : return -1;
1889 0 : }
1890 0 : }
1891 :
1892 0 : return 0;
1893 0 : }
1894 :
1895 : static void
1896 : process_solcap_account_update( fd_replay_tile_t * ctx,
1897 0 : fd_capture_ctx_account_update_msg_t const * msg ) {
1898 :
1899 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
1900 0 : if( FD_UNLIKELY( !bank ) ) {
1901 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", msg->bank_idx ));
1902 0 : }
1903 :
1904 0 : if( FD_UNLIKELY( !ctx->capture_ctx || !ctx->capture_ctx->capture ) ) return;
1905 0 : if( FD_UNLIKELY( fd_bank_slot_get( bank )<ctx->capture_ctx->solcap_start_slot ) ) return;
1906 :
1907 0 : uchar const * account_data = (uchar const *)fd_type_pun_const( msg )+sizeof(fd_capture_ctx_account_update_msg_t);
1908 0 : fd_solcap_write_account( ctx->capture_ctx->capture, &msg->pubkey, &msg->info, account_data, msg->data_sz );
1909 0 : }
1910 :
1911 : static void
1912 : process_exec_task_done( fd_replay_tile_t * ctx,
1913 : fd_exec_task_done_msg_t * msg,
1914 0 : ulong sig ) {
1915 0 : if( FD_UNLIKELY( sig==0UL ) ) {
1916 : // FIXME remove this branch with new solcap
1917 0 : process_solcap_account_update( ctx, fd_type_pun( msg ) );
1918 0 : return;
1919 0 : }
1920 :
1921 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
1922 :
1923 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
1924 0 : bank->refcnt--;
1925 :
1926 0 : switch( sig>>32 ) {
1927 0 : case FD_EXEC_TT_TXN_EXEC: {
1928 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
1929 : /* Query the txn signature against our recently generated vote
1930 : txn signatures. If the query is successful, then we have
1931 : seen our own vote transaction land and this should be marked
1932 : in the bank. We go through this exercise until we've seen
1933 : our vote rooted. */
1934 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, msg->txn_exec->txn_idx );
1935 0 : if( fd_vote_tracker_query_sig( ctx->vote_tracker, fd_type_pun_const( txn_p->payload+TXN( txn_p )->signature_off ) ) ) {
1936 0 : *fd_bank_has_identity_vote_modify( bank ) += 1;
1937 0 : }
1938 0 : }
1939 0 : if( FD_UNLIKELY( msg->txn_exec->err && !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) {
1940 : /* Every transaction in a valid block has to execute.
1941 : Otherwise, we should mark the block as dead. Also freeze the
1942 : bank if possible. */
1943 0 : fd_banks_mark_bank_dead( ctx->banks, bank );
1944 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
1945 0 : }
1946 0 : if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
1947 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
1948 0 : }
1949 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, msg->txn_exec->txn_idx, exec_tile_idx );
1950 0 : break;
1951 0 : }
1952 0 : case FD_EXEC_TT_TXN_SIGVERIFY: {
1953 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) {
1954 : /* Every transaction in a valid block has to sigverify.
1955 : Otherwise, we should mark the block as dead. Also freeze the
1956 : bank if possible. */
1957 0 : fd_banks_mark_bank_dead( ctx->banks, bank );
1958 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
1959 0 : }
1960 0 : if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
1961 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
1962 0 : }
1963 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, msg->txn_sigverify->txn_idx, exec_tile_idx );
1964 0 : break;
1965 0 : }
1966 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
1967 0 : }
1968 :
1969 : /* Reference counter just decreased, and an exec tile just got freed
1970 : up. If there's a need to be more aggressively pruning, we could
1971 : check here if more slots just became publishable and publish. Not
1972 : publishing here shouldn't bloat the fork tree too much though. We
1973 : mark minority forks dead as soon as we can, and execution dispatch
1974 : stops on dead blocks. So shortly afterwards, dead blocks should be
1975 : eligible for pruning as in-flight transactions retire from the
1976 : execution pipeline. */
1977 :
1978 0 : }
1979 :
1980 : static void
1981 : process_tower_slot_done( fd_replay_tile_t * ctx,
1982 : fd_stem_context_t * stem,
1983 0 : fd_tower_slot_done_t const * msg ) {
1984 0 : fd_bank_t * replay_bank = fd_banks_bank_query( ctx->banks, msg->replay_bank_idx );
1985 0 : if( FD_UNLIKELY( !replay_bank ) ) FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", msg->replay_bank_idx ));
1986 0 : replay_bank->refcnt--;
1987 :
1988 0 : ctx->reset_block_id = msg->reset_block_id;
1989 0 : ctx->reset_slot = msg->reset_slot;
1990 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1991 0 : ulong min_leader_slot = fd_ulong_max( msg->reset_slot+1UL, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot+1UL ) );
1992 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
1993 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1994 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1995 0 : } else {
1996 0 : ctx->next_leader_tickcount = LONG_MAX;
1997 0 : }
1998 :
1999 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->reset_block_id, NULL, ctx->block_id_arr );
2000 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2001 0 : FD_LOG_CRIT(( "invariant violation: block id ele doesn't exist for reset block id: %s, slot: %lu", FD_BASE58_ENC_32_ALLOCA( &msg->reset_block_id ), msg->reset_slot ));
2002 0 : }
2003 0 : ulong reset_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2004 :
2005 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, reset_bank_idx );
2006 0 : if( FD_UNLIKELY( !bank ) ) {
2007 0 : FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", reset_bank_idx ));
2008 0 : }
2009 :
2010 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) FD_TEST( msg->root_slot<=msg->reset_slot );
2011 0 : ctx->reset_bank = bank;
2012 :
2013 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2014 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2015 :
2016 0 : reset->bank_idx = bank->idx;
2017 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2018 0 : reset->completed_slot = ctx->reset_slot;
2019 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
2020 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2021 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
2022 :
2023 0 : fd_memcpy( reset->completed_block_id, &block_id_ele->block_id, sizeof(fd_hash_t) );
2024 :
2025 0 : fd_blockhashes_t const * block_hash_queue = fd_bank_block_hash_queue_query( bank );
2026 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2027 0 : FD_TEST( last_hash );
2028 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2029 :
2030 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2031 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
2032 : /* Low power producer, maximum of one microblock per tick in the slot */
2033 0 : reset->max_microblocks_in_slot = ticks_per_slot;
2034 0 : } else {
2035 : /* See the long comment in after_credit for this limit */
2036 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
2037 0 : }
2038 0 : reset->next_leader_slot = ctx->next_leader_slot;
2039 :
2040 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++;
2041 :
2042 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_RESET, ctx->replay_out->chunk, sizeof(fd_poh_reset_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2043 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_poh_reset_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2044 0 : }
2045 :
2046 0 : FD_LOG_INFO(( "tower_slot_done(reset_slot=%lu, next_leader_slot=%lu, vote_slot=%lu)", msg->reset_slot, ctx->next_leader_slot, msg->vote_slot ));
2047 0 : maybe_become_leader( ctx, stem );
2048 :
2049 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) {
2050 :
2051 0 : FD_TEST( msg->root_slot>=ctx->consensus_root_slot );
2052 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->root_block_id, NULL, ctx->block_id_arr );
2053 0 : FD_TEST( block_id_ele );
2054 :
2055 0 : ctx->consensus_root_slot = msg->root_slot;
2056 0 : ctx->consensus_root = msg->root_block_id;
2057 0 : ctx->consensus_root_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2058 :
2059 0 : publish_root_advanced( ctx, stem );
2060 0 : }
2061 :
2062 0 : ulong distance = 0UL;
2063 0 : fd_bank_t * parent = bank;
2064 0 : while( parent ) {
2065 0 : if( FD_UNLIKELY( parent->idx==ctx->consensus_root_bank_idx ) ) break;
2066 0 : parent = fd_banks_get_parent( ctx->banks, parent );
2067 0 : distance++;
2068 0 : }
2069 :
2070 0 : FD_MGAUGE_SET( REPLAY, ROOT_DISTANCE, distance );
2071 0 : }
2072 :
2073 : static void
2074 : process_fec_complete( fd_replay_tile_t * ctx,
2075 0 : uchar const * shred_buf ) {
2076 0 : fd_shred_t const * shred = (fd_shred_t const *)fd_type_pun_const( shred_buf );
2077 :
2078 0 : fd_hash_t const * merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ );
2079 0 : fd_hash_t const * chained_merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) );
2080 0 : int is_leader_fec = *(int const *) fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) + sizeof(fd_hash_t) );
2081 :
2082 0 : int data_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_DATA_COMPLETE );
2083 0 : int slot_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_SLOT_COMPLETE );
2084 :
2085 0 : FD_TEST( !fd_reasm_query( ctx->reasm, merkle_root ) );
2086 0 : if( FD_UNLIKELY( shred->slot - shred->data.parent_off == fd_reasm_slot0( ctx->reasm ) && shred->fec_set_idx == 0) ) {
2087 0 : chained_merkle_root = &fd_reasm_root( ctx->reasm )->key;
2088 0 : }
2089 :
2090 0 : FD_TEST( fd_reasm_free( ctx->reasm ) );
2091 :
2092 0 : FD_TEST( fd_reasm_insert( ctx->reasm, merkle_root, chained_merkle_root, shred->slot, shred->fec_set_idx, shred->data.parent_off, (ushort)(shred->idx - shred->fec_set_idx + 1), data_complete, slot_complete, is_leader_fec ) );
2093 0 : }
2094 :
2095 : static void
2096 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
2097 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2098 0 : FD_TEST( bank );
2099 :
2100 0 : bank->refcnt--;
2101 0 : }
2102 :
2103 : static void
2104 : process_vote_txn_sent( fd_replay_tile_t * ctx,
2105 0 : fd_txn_m_t * txnm ) {
2106 : /* The send tile has signed and sent a vote. Add this vote to the
2107 : vote tracker. We go through this exercise until we've seen our
2108 : vote rooted. */
2109 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
2110 0 : uchar * payload = ((uchar *)txnm) + sizeof(fd_txn_m_t);
2111 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
2112 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
2113 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
2114 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
2115 0 : }
2116 0 : fd_vote_tracker_insert( ctx->vote_tracker, fd_type_pun_const( payload+txn->signature_off ) );
2117 0 : }
2118 0 : }
2119 :
2120 : static inline void
2121 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
2122 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
2123 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
2124 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
2125 0 : }
2126 0 : }
2127 :
2128 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_forks_cnt!=ULONG_MAX && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
2129 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
2130 :
2131 0 : union {
2132 0 : uchar c[ 32 ];
2133 0 : ushort s[ 16 ];
2134 0 : } running_hash;
2135 0 : fd_memcpy( running_hash.c, ctx->genesis_hash, sizeof(fd_hash_t) );
2136 :
2137 0 : ulong processed = 0UL;
2138 0 : ulong min_value = 0UL;
2139 0 : while( processed<ctx->hard_forks_cnt ) {
2140 0 : ulong min_index = ULONG_MAX;
2141 0 : for( ulong i=0UL; i<ctx->hard_forks_cnt; i++ ) {
2142 0 : if( ctx->hard_forks[ i ]>=min_value && (min_index==ULONG_MAX || ctx->hard_forks[ i ]<ctx->hard_forks[ min_index ] ) ) {
2143 0 : min_index = i;
2144 0 : }
2145 0 : }
2146 :
2147 0 : FD_TEST( min_index!=ULONG_MAX );
2148 0 : min_value = ctx->hard_forks[ min_index ];
2149 0 : ulong min_count = 0UL;
2150 0 : for( ulong i=0UL; i<ctx->hard_forks_cnt; i++ ) {
2151 0 : if( ctx->hard_forks[ i ]==min_value ) min_count++;
2152 0 : }
2153 :
2154 0 : uchar data[ 48UL ];
2155 0 : fd_memcpy( data, running_hash.c, sizeof(fd_hash_t) );
2156 0 : fd_memcpy( data+32UL, &min_value, sizeof(ulong) );
2157 0 : fd_memcpy( data+40UL, &min_count, sizeof(ulong) );
2158 :
2159 0 : FD_TEST( fd_sha256_hash( data, 48UL, running_hash.c ) );
2160 0 : processed += min_count;
2161 0 : min_value += 1UL;
2162 0 : }
2163 :
2164 0 : ushort xor = 0;
2165 0 : for( ulong i=0UL; i<16UL; i++ ) xor ^= running_hash.s[ i ];
2166 :
2167 0 : xor = fd_ushort_bswap( xor );
2168 0 : xor = fd_ushort_if( xor<USHORT_MAX, (ushort)(xor + 1), USHORT_MAX );
2169 :
2170 0 : if( FD_UNLIKELY( expected_shred_version!=xor ) ) {
2171 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from genesis hash %s and hard forks", expected_shred_version, xor, FD_BASE58_ENC_32_ALLOCA( &ctx->genesis_hash ) ));
2172 0 : }
2173 0 : }
2174 0 : }
2175 :
2176 : static inline int
2177 : returnable_frag( fd_replay_tile_t * ctx,
2178 : ulong in_idx,
2179 : ulong seq,
2180 : ulong sig,
2181 : ulong chunk,
2182 : ulong sz,
2183 : ulong ctl,
2184 : ulong tsorig,
2185 : ulong tspub,
2186 0 : fd_stem_context_t * stem ) {
2187 0 : (void)seq;
2188 0 : (void)ctl;
2189 0 : (void)tsorig;
2190 0 : (void)tspub;
2191 :
2192 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
2193 0 : FD_LOG_ERR(( "chunk %lu %lu from in %d corrupt, not in range [%lu,%lu]", chunk, sz, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
2194 :
2195 0 : switch( ctx->in_kind[in_idx] ) {
2196 0 : case IN_KIND_GENESIS: {
2197 0 : uchar const * src = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2198 0 : ctx->has_genesis_hash = 1;
2199 0 : if( FD_LIKELY( sig==GENESI_SIG_BOOTSTRAP_COMPLETED ) ) {
2200 0 : boot_genesis( ctx, stem, in_idx, chunk );
2201 0 : fd_memcpy( ctx->genesis_hash, src+sizeof(fd_lthash_value_t), sizeof(fd_hash_t) );
2202 0 : } else {
2203 0 : fd_memcpy( ctx->genesis_hash, src, sizeof(fd_hash_t) );
2204 0 : }
2205 :
2206 0 : maybe_verify_cluster_type( ctx );
2207 0 : maybe_verify_shred_version( ctx );
2208 0 : break;
2209 0 : }
2210 0 : case IN_KIND_IPECHO: {
2211 0 : FD_TEST( sig && sig<=USHORT_MAX );
2212 0 : ctx->ipecho_shred_version = (ushort)sig;
2213 0 : maybe_verify_shred_version( ctx );
2214 0 : break;
2215 0 : }
2216 0 : case IN_KIND_SNAP:
2217 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
2218 0 : maybe_verify_shred_version( ctx );
2219 0 : break;
2220 0 : case IN_KIND_EXEC: {
2221 0 : process_exec_task_done( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
2222 0 : break;
2223 0 : }
2224 0 : case IN_KIND_POH: {
2225 0 : process_poh_message( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2226 0 : break;
2227 0 : }
2228 0 : case IN_KIND_RESOLV: {
2229 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2230 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
2231 0 : break;
2232 0 : }
2233 0 : case IN_KIND_TOWER: {
2234 0 : if ( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_DONE ) ) process_tower_slot_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2235 0 : else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_CONFIRMED ) ) {
2236 0 : fd_tower_slot_confirmed_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2237 :
2238 : /* Implement replay plugin API here */
2239 :
2240 0 : switch( msg->kind ) {
2241 0 : case FD_TOWER_SLOT_CONFIRMED_OPTIMISTIC: break;
2242 0 : case FD_TOWER_SLOT_CONFIRMED_ROOTED: break;
2243 0 : }
2244 0 : };
2245 0 : break;
2246 0 : }
2247 0 : case IN_KIND_SHRED: {
2248 : /* TODO: This message/sz should be defined. */
2249 0 : if( sz==FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) + sizeof(fd_hash_t) + sizeof(int) ) {
2250 : /* If receive a FEC complete message. */
2251 0 : process_fec_complete( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2252 0 : }
2253 0 : break;
2254 0 : }
2255 0 : case IN_KIND_VTXN: {
2256 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2257 0 : break;
2258 0 : }
2259 0 : case IN_KIND_RPC:
2260 0 : case IN_KIND_GUI: {
2261 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sig );
2262 0 : FD_TEST( bank );
2263 0 : bank->refcnt--;
2264 0 : break;
2265 0 : }
2266 0 : default:
2267 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
2268 0 : }
2269 :
2270 0 : return 0;
2271 0 : }
2272 :
2273 : static inline fd_replay_out_link_t
2274 : out1( fd_topo_t const * topo,
2275 : fd_topo_tile_t const * tile,
2276 0 : char const * name ) {
2277 0 : ulong idx = ULONG_MAX;
2278 :
2279 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2280 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2281 0 : if( !strcmp( link->name, name ) ) {
2282 0 : if( FD_UNLIKELY( idx!=ULONG_MAX ) ) FD_LOG_ERR(( "tile %s:%lu had multiple output links named %s but expected one", tile->name, tile->kind_id, name ));
2283 0 : idx = i;
2284 0 : }
2285 0 : }
2286 :
2287 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
2288 :
2289 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2290 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2291 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2292 :
2293 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2294 0 : }
2295 :
2296 : static void
2297 : privileged_init( fd_topo_t * topo,
2298 0 : fd_topo_tile_t * tile ) {
2299 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2300 :
2301 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2302 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2303 :
2304 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
2305 :
2306 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
2307 :
2308 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
2309 0 : tile->replay.bundle.enabled = 0;
2310 0 : }
2311 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2312 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2313 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
2314 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2315 0 : }
2316 0 : }
2317 0 : }
2318 :
2319 : static void
2320 : unprivileged_init( fd_topo_t * topo,
2321 0 : fd_topo_tile_t * tile ) {
2322 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2323 :
2324 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
2325 :
2326 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2327 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2328 0 : void * block_id_arr_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
2329 0 : void * block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
2330 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
2331 0 : void * reasm_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_reasm_align(), fd_reasm_footprint( 1 << 20 ) );
2332 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.max_live_slots ) );
2333 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
2334 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
2335 0 : # if FD_HAS_FLATCC
2336 0 : void * block_dump_ctx = NULL;
2337 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2338 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
2339 0 : }
2340 0 : # endif
2341 :
2342 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
2343 0 : FD_TEST( store_obj_id!=ULONG_MAX );
2344 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
2345 0 : FD_TEST( ctx->store );
2346 :
2347 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
2348 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
2349 0 : ctx->banks = fd_banks_join( fd_topo_obj_laddr( topo, banks_obj_id ) );
2350 0 : FD_TEST( ctx->banks );
2351 :
2352 0 : fd_bank_t * bank_pool = fd_banks_get_bank_pool( ctx->banks );
2353 0 : FD_MGAUGE_SET( REPLAY, MAX_LIVE_BANKS, fd_banks_pool_max( bank_pool ) );
2354 :
2355 0 : fd_bank_t * bank = fd_banks_init_bank( ctx->banks );
2356 0 : fd_bank_slot_set( bank, 0UL );
2357 0 : FD_TEST( bank );
2358 0 : FD_TEST( bank->idx==FD_REPLAY_BOOT_BANK_IDX );
2359 :
2360 0 : ctx->consensus_root_slot = ULONG_MAX;
2361 0 : ctx->consensus_root = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
2362 0 : ctx->published_root_slot = ULONG_MAX;
2363 :
2364 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
2365 0 : ctx->ipecho_shred_version = 0;
2366 0 : ctx->has_genesis_hash = 0;
2367 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
2368 0 : ctx->hard_forks_cnt = ULONG_MAX;
2369 :
2370 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2371 0 : ctx->bundle.enabled = 1;
2372 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
2373 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
2374 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
2375 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2376 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
2377 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
2378 0 : }
2379 0 : } else {
2380 0 : ctx->bundle.enabled = 0;
2381 0 : }
2382 :
2383 0 : fd_features_t * features = fd_bank_features_modify( bank );
2384 0 : fd_features_enable_cleaned_up( features, &FD_RUNTIME_CLUSTER_VERSION );
2385 :
2386 0 : char const * one_off_features[ 16UL ];
2387 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(one_off_features)/sizeof(one_off_features[0]) );
2388 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) one_off_features[ i ] = tile->replay.enable_features[i];
2389 0 : fd_features_enable_one_offs( features, one_off_features, (uint)tile->replay.enable_features_cnt, 0UL );
2390 :
2391 0 : FD_TEST( fd_accdb_admin_join ( ctx->accdb_admin, fd_topo_obj_laddr( topo, tile->replay.funk_obj_id ) ) );
2392 0 : FD_TEST( fd_accdb_user_join ( ctx->accdb, fd_topo_obj_laddr( topo, tile->replay.funk_obj_id ) ) );
2393 0 : FD_TEST( fd_progcache_admin_join( ctx->progcache_admin, fd_topo_obj_laddr( topo, tile->replay.progcache_obj_id ) ) );
2394 :
2395 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
2396 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
2397 0 : FD_TEST( txncache_shmem );
2398 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
2399 0 : FD_TEST( ctx->txncache );
2400 :
2401 0 : ctx->capture_ctx = NULL;
2402 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) || strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2403 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
2404 0 : }
2405 :
2406 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
2407 0 : ctx->capture_ctx->checkpt_freq = ULONG_MAX;
2408 0 : ctx->capture_file = fopen( tile->replay.solcap_capture, "w+" );
2409 0 : if( FD_UNLIKELY( !ctx->capture_file ) ) FD_LOG_ERR(( "fopen(%s) failed (%d-%s)", tile->replay.solcap_capture, errno, fd_io_strerror( errno ) ));
2410 :
2411 0 : ctx->capture_ctx->capture_txns = 0;
2412 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
2413 0 : fd_solcap_writer_init( ctx->capture_ctx->capture, ctx->capture_file );
2414 0 : }
2415 :
2416 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2417 0 : ctx->capture_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
2418 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) ctx->capture_ctx->dump_block_to_pb = tile->replay.dump_block_to_pb;
2419 0 : }
2420 :
2421 0 : # if FD_HAS_FLATCC
2422 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2423 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
2424 0 : } else {
2425 0 : ctx->block_dump_ctx = NULL;
2426 0 : }
2427 0 : # endif
2428 :
2429 0 : ctx->exec_cnt = fd_topo_tile_name_cnt( topo, "exec" );
2430 :
2431 0 : ctx->is_booted = 0;
2432 :
2433 0 : ctx->larger_max_cost_per_block = tile->replay.larger_max_cost_per_block;
2434 :
2435 0 : ctx->reasm = fd_reasm_join( fd_reasm_new( reasm_mem, 1 << 20, 0 ) );
2436 0 : FD_TEST( ctx->reasm );
2437 :
2438 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, tile->replay.max_live_slots, ctx->exec_cnt ), tile->replay.max_live_slots );
2439 0 : FD_TEST( ctx->sched );
2440 :
2441 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, 0UL ) );
2442 0 : FD_TEST( ctx->vote_tracker );
2443 :
2444 0 : ctx->has_identity_vote_rooted = 0;
2445 :
2446 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
2447 0 : FD_TEST( ctx->mleaders );
2448 :
2449 0 : ctx->is_leader = 0;
2450 0 : ctx->reset_slot = 0UL;
2451 0 : ctx->reset_bank = NULL;
2452 0 : ctx->reset_block_id = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
2453 0 : ctx->reset_timestamp_nanos = 0UL;
2454 0 : ctx->next_leader_slot = ULONG_MAX;
2455 0 : ctx->next_leader_tickcount = LONG_MAX;
2456 0 : ctx->highwater_leader_slot = ULONG_MAX;
2457 0 : ctx->slot_duration_nanos = 350L*1000L*1000L; /* TODO: Not fixed ... not always 400ms ... */
2458 0 : ctx->slot_duration_ticks = (double)ctx->slot_duration_nanos*fd_tempo_tick_per_ns( NULL );
2459 0 : ctx->leader_bank = NULL;
2460 :
2461 : /* TODO: We need a real seed here. */
2462 0 : ctx->block_id_len = tile->replay.max_live_slots;
2463 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
2464 0 : ctx->block_id_map = fd_block_id_map_join( fd_block_id_map_new( block_id_map_mem, chain_cnt, 999UL ) );
2465 0 : FD_TEST( ctx->block_id_map );
2466 :
2467 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) {
2468 0 : ctx->block_id_arr[ i ].slot = FD_SLOT_NULL;
2469 0 : }
2470 :
2471 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
2472 :
2473 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
2474 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
2475 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
2476 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
2477 :
2478 0 : if( FD_LIKELY( link->dcache ) ) {
2479 0 : ctx->in[ i ].mem = link_wksp->wksp;
2480 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
2481 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
2482 0 : ctx->in[ i ].mtu = link->mtu;
2483 0 : }
2484 :
2485 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
2486 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
2487 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
2488 0 : else if( !strcmp( link->name, "exec_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXEC;
2489 0 : else if( !strcmp( link->name, "tower_out" ) ) ctx->in_kind[ i ] = IN_KIND_TOWER;
2490 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
2491 0 : else if( !strcmp( link->name, "resolv_repla" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
2492 0 : else if( !strcmp( link->name, "shred_out" ) ) ctx->in_kind[ i ] = IN_KIND_SHRED;
2493 0 : else if( !strcmp( link->name, "send_out" ) ) ctx->in_kind[ i ] = IN_KIND_VTXN;
2494 0 : else if( !strcmp( link->name, "gui_replay" ) ) ctx->in_kind[ i ] = IN_KIND_GUI;
2495 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
2496 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
2497 0 : }
2498 :
2499 0 : *ctx->stake_out = out1( topo, tile, "replay_stake" ); FD_TEST( ctx->stake_out->idx!=ULONG_MAX );
2500 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
2501 :
2502 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "replay_exec", 0UL );
2503 0 : FD_TEST( idx!=ULONG_MAX );
2504 0 : fd_topo_link_t * link = &topo->links[ tile->out_link_id[ idx ] ];
2505 :
2506 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
2507 0 : exec_out->idx = idx;
2508 0 : exec_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
2509 0 : exec_out->chunk0 = fd_dcache_compact_chunk0( exec_out->mem, link->dcache );
2510 0 : exec_out->wmark = fd_dcache_compact_wmark( exec_out->mem, link->dcache, link->mtu );
2511 0 : exec_out->chunk = exec_out->chunk0;
2512 :
2513 0 : ctx->gui_enabled = fd_topo_find_tile( topo, "gui", 0UL )!=ULONG_MAX;
2514 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
2515 :
2516 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
2517 :
2518 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_link_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_LINK_WAIT ),
2519 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_LINK_WAIT ) ) );
2520 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_link_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_LINK_WORK ),
2521 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_LINK_WORK ) ) );
2522 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_read_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_READ_WAIT ),
2523 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_READ_WAIT ) ) );
2524 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_read_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_READ_WORK ),
2525 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_READ_WORK ) ) );
2526 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_publish_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_PUBLISH_WAIT ),
2527 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_PUBLISH_WAIT ) ) );
2528 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_publish_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_PUBLISH_WORK ),
2529 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_PUBLISH_WORK ) ) );
2530 :
2531 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
2532 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
2533 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
2534 0 : }
2535 :
2536 : static ulong
2537 : populate_allowed_seccomp( fd_topo_t const * topo FD_FN_UNUSED,
2538 : fd_topo_tile_t const * tile FD_FN_UNUSED,
2539 : ulong out_cnt,
2540 0 : struct sock_filter * out ) {
2541 :
2542 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() );
2543 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
2544 0 : }
2545 :
2546 : static ulong
2547 : populate_allowed_fds( fd_topo_t const * topo FD_FN_UNUSED,
2548 : fd_topo_tile_t const * tile FD_FN_UNUSED,
2549 : ulong out_fds_cnt,
2550 0 : int * out_fds ) {
2551 :
2552 0 : if( FD_UNLIKELY( out_fds_cnt<2UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
2553 :
2554 0 : ulong out_cnt = 0UL;
2555 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
2556 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
2557 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
2558 0 : return out_cnt;
2559 0 : }
2560 :
2561 : #undef DEBUG_LOGGING
2562 :
2563 : /* counting carefully, after_credit can generate at most 7 frags and
2564 : returnable_frag boot_genesis can also generate at most 7 frags, so 14
2565 : is a conservative bound. */
2566 0 : #define STEM_BURST (14UL)
2567 :
2568 : /* TODO: calculate this properly/fix stem to work with larger numbers of links */
2569 : /* 1000 chosen empirically as anything larger slowed down replay times. Need to calculate
2570 : this properly. */
2571 0 : #define STEM_LAZY ((long)10e3)
2572 :
2573 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
2574 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
2575 :
2576 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
2577 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
2578 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
2579 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
2580 :
2581 : #include "../../disco/stem/fd_stem.c"
2582 :
2583 : fd_topo_run_tile_t fd_tile_replay = {
2584 : .name = "replay",
2585 : .populate_allowed_seccomp = populate_allowed_seccomp,
2586 : .populate_allowed_fds = populate_allowed_fds,
2587 : .scratch_align = scratch_align,
2588 : .scratch_footprint = scratch_footprint,
2589 : .privileged_init = privileged_init,
2590 : .unprivileged_init = unprivileged_init,
2591 : .run = stem_run,
2592 : };
|
