Line data Source code
1 : #include "fd_replay_tile.h"
2 : #include "fd_sched.h"
3 : #include "fd_execrp.h"
4 : #include "fd_vote_tracker.h"
5 : #include "generated/fd_replay_tile_seccomp.h"
6 :
7 : #include "../genesis/fd_genesi_tile.h"
8 : #include "../poh/fd_poh.h"
9 : #include "../poh/fd_poh_tile.h"
10 : #include "../tower/fd_tower_tile.h"
11 : #include "../resolv/fd_resolv_tile.h"
12 : #include "../restore/utils/fd_ssload.h"
13 :
14 : #include "../../disco/tiles.h"
15 : #include "../../disco/fd_txn_m.h"
16 : #include "../../disco/store/fd_store.h"
17 : #include "../../disco/shred/fd_fec_set.h"
18 : #include "../../disco/pack/fd_pack.h"
19 : #include "../../discof/fd_accdb_topo.h"
20 : #include "../../discof/reasm/fd_reasm.h"
21 : #include "../../disco/keyguard/fd_keyload.h"
22 : #include "../../disco/keyguard/fd_keyswitch.h"
23 : #include "../../disco/genesis/fd_genesis_cluster.h"
24 : #include "../../discof/genesis/genesis_hash.h"
25 : #include "../../util/pod/fd_pod.h"
26 : #include "../../flamenco/accdb/fd_accdb_admin_v1.h"
27 : #include "../../flamenco/accdb/fd_accdb_admin_v2.h"
28 : #include "../../flamenco/accdb/fd_accdb_impl_v1.h"
29 : #include "../../flamenco/accdb/fd_accdb_sync.h"
30 : #include "../../flamenco/accdb/fd_vinyl_req_pool.h"
31 : #include "../../flamenco/rewards/fd_rewards.h"
32 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
33 : #include "../../flamenco/progcache/fd_progcache_admin.h"
34 : #include "../../disco/metrics/fd_metrics.h"
35 :
36 : #include "../../flamenco/fd_flamenco_base.h"
37 : #include "../../flamenco/runtime/fd_runtime.h"
38 : #include "../../flamenco/stakes/fd_stakes.h"
39 : #include "../../flamenco/runtime/fd_runtime_stack.h"
40 : #include "../../flamenco/runtime/fd_genesis_parse.h"
41 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
42 : #include "../../flamenco/runtime/program/fd_precompiles.h"
43 : #include "../../flamenco/runtime/program/vote/fd_vote_state_versioned.h"
44 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
45 :
46 : #include <stdio.h>
47 :
48 : /* Replay concepts:
49 :
50 : - Blocks are aggregations of entries aka. microblocks which are
51 : groupings of txns and are constructed by the block producer (see
52 : fd_pack).
53 :
54 : - Entries are grouped into entry batches by the block producer (see
55 : fd_pack / fd_shredder).
56 :
57 : - Entry batches are divided into chunks known as shreds by the block
58 : producer (see fd_shredder).
59 :
60 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
61 : the block producer (see fd_shredder).
62 :
63 : - Shreds are transmitted to the rest of the cluster via the Turbine
64 : protocol (see fd_shredder / fd_shred).
65 :
66 : - Once enough shreds within a FEC set are received to recover the
67 : entirety of the shred data encoded by that FEC set, the receiver
68 : can "complete" the FEC set (see fd_fec_resolver).
69 :
70 : - If shreds in the FEC set are missing such that it can't complete,
71 : the receiver can use the Repair protocol to request missing shreds
72 : in FEC set (see fd_repair).
73 :
74 : - The current Repair protocol does not support requesting coding
75 : shreds. As a result, some FEC sets might be actually complete
76 : (contain all data shreds). Repair currently hacks around this by
77 : forcing completion but the long-term solution is to add support for
78 : fec_repairing coding shreds via Repair.
79 :
80 : - FEC sets are delivered in partial-order to the Replay tile by the
81 : Repair tile. Currently Replay only supports replaying entry batches
82 : so FEC sets need to reassembled into an entry batch before they can
83 : be replayed. The new Dispatcher will change this by taking a FEC
84 : set as input instead. */
85 :
86 0 : #define IN_KIND_SNAP ( 0)
87 0 : #define IN_KIND_GENESIS ( 1)
88 0 : #define IN_KIND_IPECHO ( 2)
89 0 : #define IN_KIND_TOWER ( 3)
90 0 : #define IN_KIND_RESOLV ( 4)
91 0 : #define IN_KIND_POH ( 5)
92 0 : #define IN_KIND_EXECRP ( 6)
93 0 : #define IN_KIND_REPAIR ( 7)
94 0 : #define IN_KIND_TXSEND ( 8)
95 0 : #define IN_KIND_RPC ( 9)
96 0 : #define IN_KIND_GOSSIP_OUT (10)
97 :
98 : #define DEBUG_LOGGING 0
99 :
100 : /* The first bank that the replay tile produces either for genesis
101 : or the snapshot boot will always be at bank index 0. */
102 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
103 :
104 : struct fd_replay_in_link {
105 : fd_wksp_t * mem;
106 : ulong chunk0;
107 : ulong wmark;
108 : ulong mtu;
109 : };
110 :
111 : typedef struct fd_replay_in_link fd_replay_in_link_t;
112 :
113 : struct fd_replay_out_link {
114 : ulong idx;
115 : fd_wksp_t * mem;
116 : ulong chunk0;
117 : ulong wmark;
118 : ulong chunk;
119 : };
120 :
121 : typedef struct fd_replay_out_link fd_replay_out_link_t;
122 :
123 : /* fd_block_id_map is a simple map of block-ids to bank indices. The
124 : map sits on top of an array of fd_block_id_ele_t. This serves as a
125 : translation layer between block ids to bank indices. The data
126 : array is indexed by bank index and the latest observed merkle root
127 : for the bank index is stored in the array. Once the block id has
128 : been observed, the entry is keyed by the latest merkle root (aka the
129 : block id). */
130 :
131 : struct fd_block_id_ele {
132 : fd_hash_t latest_mr;
133 : uint latest_fec_idx;
134 : int block_id_seen;
135 : ulong slot;
136 : ulong next_;
137 : };
138 : typedef struct fd_block_id_ele fd_block_id_ele_t;
139 :
140 : #define MAP_NAME fd_block_id_map
141 : #define MAP_ELE_T fd_block_id_ele_t
142 : #define MAP_KEY_T fd_hash_t
143 0 : #define MAP_KEY latest_mr
144 0 : #define MAP_NEXT next_
145 0 : #define MAP_KEY_EQ(k0,k1) (!memcmp((k0),(k1), sizeof(fd_hash_t)))
146 0 : #define MAP_KEY_HASH(key,seed) (fd_hash((seed),(key),sizeof(fd_hash_t)))
147 : #include "../../util/tmpl/fd_map_chain.c"
148 :
149 : static inline ulong
150 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
151 0 : return (ulong)(ele - ele_arr);
152 0 : }
153 :
154 : struct fd_replay_tile {
155 : fd_wksp_t * wksp;
156 :
157 : uint rng_seed;
158 : fd_rng_t rng[ 1 ];
159 :
160 : fd_accdb_admin_t accdb_admin[1];
161 : fd_accdb_user_t accdb[1];
162 : fd_progcache_join_t progcache[1];
163 :
164 : fd_txncache_t * txncache;
165 : fd_store_t * store;
166 : fd_banks_t banks[1];
167 : ulong frontier_indices[ FD_BANKS_MAX_BANKS ];
168 : ulong frontier_cnt;
169 :
170 : /* This flag is 1 If we have seen a vote signature that our node has
171 : sent out get rooted at least one time. The value is 0 otherwise.
172 : We can't become leader and pack blocks until this flag has been
173 : set. This parallels the Agave 'has_new_vote_been_rooted'. */
174 : int identity_vote_rooted;
175 : int wait_for_vote_to_start_leader;
176 :
177 : /* wfs_enabled is 1 if the validator is booted in
178 : wait_for_supermajority mode. In this mode replay (and, by extension,
179 : downstream consumers) is not allowed to make progress until 80% of
180 : the cluster has published their ContactInfo in Gossip with a
181 : shred version matching expected_shred_version. When this happens,
182 : wfs_complete will be set to 1. */
183 : int wfs_enabled;
184 : int wfs_complete;
185 :
186 : fd_hash_t expected_bank_hash;
187 :
188 : ulong reasm_seed;
189 : fd_reasm_t * reasm;
190 : fd_reasm_fec_t * reasm_evicted; /* evicted FEC by reasm_insert must be stored in returnable_frag, and then drained in after_credit */
191 :
192 : fd_sched_t * sched;
193 :
194 : ulong vote_tracker_seed;
195 : fd_vote_tracker_t * vote_tracker;
196 :
197 : int has_genesis_hash;
198 : char genesis_path[ PATH_MAX ];
199 : fd_hash_t genesis_hash[1];
200 : fd_genesis_t genesis[1];
201 : ulong cluster_type;
202 :
203 : int has_genesis_timestamp;
204 : ulong genesis_timestamp;
205 : int has_expected_genesis_timestamp;
206 : ulong expected_genesis_timestamp;
207 :
208 : #define FD_REPLAY_HARD_FORKS_MAX (64UL)
209 : ulong hard_forks_cnt;
210 : ulong hard_forks[ FD_REPLAY_HARD_FORKS_MAX ];
211 : ulong hard_forks_cnts[ FD_REPLAY_HARD_FORKS_MAX ];
212 :
213 : ushort expected_shred_version;
214 : ushort ipecho_shred_version;
215 :
216 : /* A note on publishing ...
217 :
218 : The watermarks are used to publish our fork-aware structures. For
219 : example, store, banks, and txncache need to be published to release
220 : resources occupied by rooted or dead blocks. In general,
221 : publishing has the effect of pruning forks in those structures,
222 : indicating that it is ok to release the memory being occupied by
223 : the blocks on said forks. Tower is responsible for informing us of
224 : the latest block on the consensus rooted fork. As soon as we can,
225 : we should move the published root as close as possible to the
226 : latest consensus root, publishing/pruning everything on the fork
227 : tree along the way. That is, all the blocks that directly descend
228 : from the current published root (inclusive) to the new published
229 : root (exclusive) on the rooted fork, as well as all the minority
230 : forks that branch from said blocks.
231 :
232 : Ideally, we'd move the published root to the consensus root
233 : immediately upon receiving a new consensus root. However, that's
234 : not always safe to do. One thing we need to be careful about is
235 : making sure that there are no more users/consumers of
236 : soon-to-be-pruned blocks, lest a use-after-free occurs. This can
237 : be done by using a reference counter for each block. Any
238 : concurrent activity, such as transaction execution in the exec
239 : tiles, should retain a refcnt on the block for as
240 : long as it needs access to the shared fork-aware structures related
241 : to that block. Eventually, refcnt on a given block will drop down
242 : to 0 as the block either finishes replaying or gets marked as dead,
243 : and any other tile that has retained a refcnt on the block releases
244 : it. At that point, it becomes a candidate for pruning. The key to
245 : safe publishing then becomes figuring out how far we could advance
246 : the published root, such that every minority fork branching off of
247 : blocks in between the current published root (inclusive) and the
248 : new published root (exclusive) is safe to be pruned. This is a
249 : straightforward tree traversal, where if a block B on the rooted
250 : fork has refcnt 0, and all minority forks branching off of B also
251 : have refcnt 0, then B is safe to be pruned. We advance the
252 : published root to the farthest consecutively prunable block on the
253 : rooted fork. Note that reasm presents the replay tile with a clean
254 : view of the world where every block is chained off of a parent
255 : block. So there are no orpahned/dangling tree nodes to worry
256 : about. The world is a nice single tree as far as replay is
257 : concerned.
258 :
259 : In the following fork tree, every node is a block and the number in
260 : parentheses is the refcnt on the block. The chain marked with
261 : double slashes is the rooted fork. Suppose the published root is
262 : at block P, and consensus root is at block T. We can't publish
263 : past block P because Q has refcnt 1.
264 :
265 :
266 : P(0)
267 : / \\
268 : Q(1) A(0)
269 : / || \
270 : X(0) B(0) C(0)
271 : / || \
272 : Y(0) M(0) R(0)
273 : / || / \
274 : D(2) T(0) J(0) L(0)
275 : ||
276 : ..
277 : ..
278 : ..
279 : ||
280 : blocks we might be actively replaying
281 :
282 :
283 : When refcnt on Q drops to 0, we would be able to advance the
284 : published root to block M, because blocks P, A, and B, as well as
285 : all subtrees branching off of them, have refcnt 0, and therefore
286 : can be pruned. Block M itself cannot be pruned yet because its
287 : child block D has refcnt 2. After publishing/pruning, the fork
288 : tree would be:
289 :
290 :
291 : M(0)
292 : / ||
293 : D(2) T(0)
294 : ||
295 : ..
296 : ..
297 : ..
298 : ||
299 : blocks we might be actively replaying
300 :
301 :
302 : As a result, the shared fork-aware structures can free resources
303 : for blocks P, A, B, and all subtrees branching off of them.
304 :
305 : For the reference counting part, the replay tile is the sole entity
306 : that can update the refcnt. This ensures that all refcnt increment
307 : and decrement attempts are serialized at the replay tile, and that
308 : there are no racy resurrection of a soon-to-be-pruned block. If a
309 : refcnt increment request arrives after a block has been pruned,
310 : replay simply rejects the request.
311 :
312 : A note on the implementation of the above ...
313 :
314 : Upon receiving a new consensus root, we descend down the rooted
315 : fork from the current published root to the new consensus root. On
316 : each node/block of the rooted fork, we do a summation of the refcnt
317 : on the block and all the minority fork blocks branching from the
318 : block. If the summation is 0, the block is safe for pruning. We
319 : advance the published root to the far end of the consecutive run of
320 : 0 refcnt sums originating from the current published root. On our
321 : descent down the minority forks, we also mark any block that hasn't
322 : finished replaying as dead, so we don't waste time executing them.
323 : No more transactions shall be dispatched for execution from dead
324 : blocks.
325 :
326 : Blocks start out with a refcnt of 0. Other tiles may send a
327 : request to the replay tile for a reference on a block. The
328 : transaction dispatcher is another source of refcnt updates. On
329 : every dispatch of a transaction for block B, we increment the
330 : refcnt for B. And on every transaction finalization, we decrement
331 : the refcnt for B. This means that whenever the refcnt on a block
332 : is 0, there is no more reference on that block from the execution
333 : pipeline. While it might be tempting to simply increment the
334 : refcnt once when we start replaying a block, and decrement the
335 : refcnt once when we finish a block, this more fine-grained refcnt
336 : update strategy allows for aborting and potentially immediate
337 : pruning of blocks under interleaved block replay. Upon receiving a
338 : new consensus root, we can simply look at the refcnt on minority
339 : fork blocks, and a refcnt of 0 would imply that the block is safe
340 : for pruning, even if we haven't finished replaying it. Without the
341 : fine-grained refcnt, we would need to first stop dispatching from
342 : the aborted block, and then wait for a full drain of the execution
343 : pipeline to know for sure that there are no more in-flight
344 : transactions executing on the aborted block. Note that this will
345 : allow the refcnt on any block to transiently drop down to 0. We
346 : will not mistakenly prune an actively replaying block, aka a leaf
347 : node, that is chaining off of the rooted fork, because the
348 : consensus root is always an ancestor of the actively replaying tip.
349 : */
350 : fd_hash_t consensus_root; /* The most recent block to have reached max lockout in the tower. */
351 : ulong consensus_root_slot; /* slot number of the above. */
352 : ulong consensus_root_bank_idx; /* bank index of the above. */
353 : ulong published_root_slot; /* slot number of the published root. */
354 : ulong published_root_bank_idx; /* bank index of the published root. */
355 :
356 : /* Randomly generated block id for the initial genesis/snapshot slot.
357 : To be replaced with block id in the snapshot manifest when SIMD-333
358 : is activated. */
359 :
360 : fd_hash_t initial_block_id;
361 :
362 : /* We need to maintain a tile-local mapping of block-ids to bank index
363 : and vice versa. This translation layer is needed for conversion
364 : since tower operates on block-ids and downstream consumers of FEC
365 : sets operate on bank indices. This mapping must happen both ways:
366 : 1. tower sends us block ids and we must map them to bank indices.
367 : 2. when a block is completed, we must map the bank index to a block
368 : id to send a slot complete message to tower. */
369 : ulong block_id_len;
370 : fd_block_id_ele_t * block_id_arr;
371 : ulong block_id_map_seed;
372 : fd_block_id_map_t * block_id_map;
373 :
374 : /* Capture-related configs */
375 : fd_capture_ctx_t * capture_ctx;
376 : FILE * capture_file;
377 : fd_capture_link_buf_t cap_repl_out[1];
378 :
379 : /* Protobuf dumping context for debugging runtime execution and
380 : collecting seed corpora. */
381 : fd_dump_proto_ctx_t * dump_proto_ctx;
382 :
383 : /* Whether the runtime has been booted either from snapshot loading
384 : or from genesis. */
385 : int is_booted;
386 :
387 : /* Buffer to store vote towers that need to be published to the Tower
388 : tile. */
389 :
390 : fd_multi_epoch_leaders_t * mleaders;
391 :
392 : int larger_max_cost_per_block;
393 :
394 : /* When we transition to becoming leader, we can only unbecome the
395 : leader if we have received a block id from the FEC reassembler, and
396 : a message from PoH that the leader slot has ended. After both of
397 : these conditions are met, then we are free to unbecome the leader.
398 : */
399 : uint is_leader : 1;
400 : uint supports_leader : 1;
401 : int recv_poh;
402 : ulong next_leader_slot;
403 : long next_leader_tickcount;
404 : ulong highwater_leader_slot;
405 : ulong reset_slot;
406 : fd_bank_t reset_bank[1];
407 : fd_hash_t reset_block_id;
408 : long reset_timestamp_nanos;
409 : double slot_duration_nanos;
410 : double slot_duration_ticks;
411 : fd_bank_t leader_bank[1];
412 :
413 : fd_pubkey_t identity_pubkey[1];
414 : ulong identity_idx;
415 :
416 : fd_keyswitch_t * keyswitch;
417 : int halt_leader;
418 :
419 : ulong resolv_tile_cnt;
420 :
421 : int in_kind[ 128 ];
422 : fd_replay_in_link_t in[ 128 ];
423 :
424 : fd_replay_out_link_t exec_out[ 1 ];
425 :
426 : fd_replay_out_link_t replay_out[1];
427 :
428 : fd_replay_out_link_t epoch_out[1];
429 :
430 : /* The rpc tile needs to occasionally own a reference to a live bank.
431 : Replay needs to know if the rpc as a consumer is enabled so it can
432 : increment the bank's refcnt before publishing bank_idx. */
433 : int rpc_enabled;
434 :
435 : # if FD_HAS_FLATCC
436 : /* For dumping blocks to protobuf. For backtest only. */
437 : fd_block_dump_ctx_t * block_dump_ctx;
438 : # endif
439 :
440 : /* We need a few pieces of information to compute the right addresses
441 : for bundle crank information that we need to send to pack. */
442 : struct {
443 : int enabled;
444 : fd_pubkey_t vote_account;
445 : fd_bundle_crank_gen_t gen[1];
446 : } bundle;
447 :
448 : struct {
449 : ulong store_query_acquire;
450 : ulong store_query_release;
451 : fd_histf_t store_query_wait[1];
452 : fd_histf_t store_query_work[1];
453 : ulong store_query_cnt;
454 : ulong store_query_missing_cnt;
455 : ulong store_query_mr;
456 : ulong store_query_missing_mr;
457 :
458 : ulong slots_total;
459 : ulong transactions_total;
460 :
461 : ulong reasm_latest_slot;
462 : ulong reasm_latest_fec_idx;
463 :
464 : ulong sched_full;
465 : ulong reasm_empty;
466 : ulong leader_bid_wait;
467 : ulong banks_full;
468 : ulong storage_root_behind;
469 :
470 : fd_histf_t root_slot_dur[1];
471 : fd_histf_t root_account_dur[1];
472 :
473 : long wksp_sample_ts;
474 : } metrics;
475 :
476 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
477 :
478 : ulong runtime_stack_seed;
479 : fd_runtime_stack_t * runtime_stack;
480 : };
481 :
482 : typedef struct fd_replay_tile fd_replay_tile_t;
483 :
484 : FD_FN_CONST static inline ulong
485 0 : scratch_align( void ) {
486 0 : return 128UL;
487 0 : }
488 : FD_FN_PURE static inline ulong
489 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
490 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
491 :
492 0 : ulong l = FD_LAYOUT_INIT;
493 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
494 0 : l = FD_LAYOUT_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS ) );
495 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
496 0 : l = FD_LAYOUT_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
497 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
498 0 : l = FD_LAYOUT_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
499 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
500 0 : l = FD_LAYOUT_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
501 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
502 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
503 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
504 :
505 0 : # if FD_HAS_FLATCC
506 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
507 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
508 0 : }
509 0 : # endif
510 :
511 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
512 :
513 0 : return l;
514 0 : }
515 :
516 : static inline void
517 0 : metrics_write( fd_replay_tile_t * ctx ) {
518 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_ACQUIRE, ctx->metrics.store_query_acquire );
519 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_RELEASE, ctx->metrics.store_query_release );
520 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WAIT, ctx->metrics.store_query_wait );
521 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WORK, ctx->metrics.store_query_work );
522 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_CNT, ctx->metrics.store_query_cnt );
523 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_MISSING_CNT, ctx->metrics.store_query_missing_cnt );
524 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MR, ctx->metrics.store_query_mr );
525 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MISSING_MR, ctx->metrics.store_query_missing_mr );
526 :
527 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
528 0 : ulong leader_slot = ctx->leader_bank->data ? ctx->leader_bank->data->f.slot : 0UL;
529 :
530 0 : if( FD_LIKELY( ctx->leader_bank->data ) ) {
531 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
532 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
533 0 : } else {
534 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
535 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
536 0 : }
537 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
538 :
539 0 : FD_MGAUGE_SET( REPLAY, LIVE_BANKS, fd_banks_pool_used_cnt( ctx->banks ) );
540 :
541 0 : ulong reasm_free = fd_reasm_free( ctx->reasm );
542 0 : FD_MGAUGE_SET( REPLAY, REASM_FREE, reasm_free );
543 :
544 0 : FD_MCNT_SET( REPLAY, SLOTS_TOTAL, ctx->metrics.slots_total );
545 0 : FD_MCNT_SET( REPLAY, TRANSACTIONS_TOTAL, ctx->metrics.transactions_total );
546 :
547 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_SLOT, ctx->metrics.reasm_latest_slot );
548 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_FEC_IDX, ctx->metrics.reasm_latest_fec_idx );
549 :
550 0 : fd_sched_metrics_write( ctx->sched );
551 :
552 0 : FD_MCNT_SET( REPLAY, SCHED_FULL, ctx->metrics.sched_full );
553 0 : FD_MCNT_SET( REPLAY, REASM_EMPTY, ctx->metrics.reasm_empty );
554 0 : FD_MCNT_SET( REPLAY, LEADER_BID_WAIT, ctx->metrics.leader_bid_wait );
555 0 : FD_MCNT_SET( REPLAY, BANKS_FULL, ctx->metrics.banks_full );
556 0 : FD_MCNT_SET( REPLAY, STORAGE_ROOT_BEHIND, ctx->metrics.storage_root_behind );
557 :
558 0 : FD_MCNT_SET( REPLAY, ACCDB_CREATED, ctx->accdb->base.created_cnt );
559 0 : FD_MCNT_SET( REPLAY, ACCDB_REVERTED, ctx->accdb_admin->base.revert_cnt );
560 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED, ctx->accdb_admin->base.root_cnt );
561 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED_BYTES, ctx->accdb_admin->base.root_tot_sz );
562 0 : FD_MCNT_SET( REPLAY, ACCDB_GC_ROOT, ctx->accdb_admin->base.gc_root_cnt );
563 0 : FD_MCNT_SET( REPLAY, ACCDB_RECLAIMED, ctx->accdb_admin->base.reclaim_cnt );
564 0 : FD_MHIST_COPY( REPLAY, ROOT_SLOT_DURATION_SECONDS, ctx->metrics.root_slot_dur );
565 0 : FD_MHIST_COPY( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS, ctx->metrics.root_account_dur );
566 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_DB, (ulong)ctx->accdb_admin->base.dt_vinyl );
567 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_COPY, (ulong)ctx->accdb_admin->base.dt_copy );
568 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_GC, (ulong)ctx->accdb_admin->base.dt_gc );
569 :
570 0 : fd_progcache_admin_metrics_t const * pcm = &fd_progcache_admin_metrics_g;
571 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, pcm->root_cnt );
572 0 : FD_MCNT_SET( REPLAY, PROGCACHE_GC_ROOT, pcm->gc_root_cnt );
573 0 : long now = fd_log_wallclock();
574 0 : # define PROGCACHE_WKSP_SAMPLE_INTERVAL 695234851L /* sampling the wksp is expensive, so do it no more frequent than every ~700ms */
575 0 : if( FD_UNLIKELY( !ctx->metrics.wksp_sample_ts ||
576 0 : now - ctx->metrics.wksp_sample_ts > PROGCACHE_WKSP_SAMPLE_INTERVAL ) ) {
577 0 : fd_progcache_wksp_metrics_update( ctx->progcache );
578 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_FREE_PARTS, pcm->wksp.free_part_cnt );
579 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_FREE_BYTES, pcm->wksp.free_sz );
580 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_SIZE_BYTES, pcm->wksp.total_sz );
581 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_PART_SIZE_MAX_BYTES, pcm->wksp.free_part_max );
582 0 : ctx->metrics.wksp_sample_ts = now;
583 0 : }
584 0 : }
585 :
586 : static void
587 : publish_epoch_info( fd_replay_tile_t * ctx,
588 : fd_stem_context_t * stem,
589 : fd_bank_t * bank,
590 0 : int current_epoch ) {
591 0 : fd_epoch_schedule_t const * schedule = &bank->data->f.epoch_schedule;
592 0 : ulong epoch = fd_slot_to_epoch( schedule, bank->data->f.slot, NULL ) + fd_ulong_if( current_epoch, 1UL, 0UL );
593 :
594 0 : fd_features_t const * features = &bank->data->f.features;
595 :
596 0 : fd_runtime_stack_t * runtime_stack = ctx->runtime_stack;
597 :
598 0 : fd_epoch_info_msg_t * epoch_info_msg = fd_chunk_to_laddr( ctx->epoch_out->mem, ctx->epoch_out->chunk );
599 :
600 0 : epoch_info_msg->staked_vote_cnt = current_epoch ? runtime_stack->epoch_weights.next_stake_weights_cnt : runtime_stack->epoch_weights.stake_weights_cnt;
601 0 : epoch_info_msg->staked_id_cnt = current_epoch ? runtime_stack->epoch_weights.next_id_weights_cnt : runtime_stack->epoch_weights.id_weights_cnt;
602 0 : epoch_info_msg->epoch_schedule = *schedule;
603 0 : epoch_info_msg->features = *features;
604 0 : epoch_info_msg->epoch = epoch;
605 0 : epoch_info_msg->start_slot = fd_epoch_slot0( schedule, epoch );
606 0 : epoch_info_msg->slot_cnt = fd_epoch_slot_cnt( schedule, epoch );
607 0 : epoch_info_msg->excluded_id_stake = current_epoch ? runtime_stack->epoch_weights.next_id_weights_excluded : runtime_stack->epoch_weights.id_weights_excluded;
608 0 : epoch_info_msg->vote_keyed_lsched = 1UL;
609 :
610 0 : fd_vote_stake_weight_t * stake_weights = fd_type_pun( epoch_info_msg + 1 );
611 0 : fd_vote_stake_weight_t * src_stake_weights = current_epoch ? runtime_stack->epoch_weights.next_stake_weights : runtime_stack->epoch_weights.stake_weights;
612 0 : memcpy( stake_weights, src_stake_weights, epoch_info_msg->staked_vote_cnt * sizeof(fd_vote_stake_weight_t) );
613 :
614 0 : fd_stake_weight_t * id_weights = fd_epoch_info_msg_id_weights( epoch_info_msg );
615 0 : fd_stake_weight_t * src_id_weights = current_epoch ? runtime_stack->epoch_weights.next_id_weights : runtime_stack->epoch_weights.id_weights;
616 0 : fd_memcpy( id_weights, src_id_weights, epoch_info_msg->staked_id_cnt * sizeof(fd_stake_weight_t) );
617 :
618 0 : ulong epoch_info_sz = fd_epoch_info_msg_sz( epoch_info_msg->staked_vote_cnt , epoch_info_msg->staked_id_cnt );
619 :
620 0 : ulong epoch_info_sig = 4UL;
621 0 : fd_stem_publish( stem, ctx->epoch_out->idx, epoch_info_sig, ctx->epoch_out->chunk, epoch_info_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
622 0 : ctx->epoch_out->chunk = fd_dcache_compact_next( ctx->epoch_out->chunk, epoch_info_sz, ctx->epoch_out->chunk0, ctx->epoch_out->wmark );
623 :
624 0 : fd_multi_epoch_leaders_epoch_msg_init( ctx->mleaders, epoch_info_msg );
625 0 : fd_multi_epoch_leaders_epoch_msg_fini( ctx->mleaders );
626 0 : }
627 :
628 : /**********************************************************************/
629 : /* Transaction execution state machine helpers */
630 : /**********************************************************************/
631 :
632 : static void
633 : replay_block_start( fd_replay_tile_t * ctx,
634 : fd_stem_context_t * stem,
635 : ulong bank_idx,
636 : ulong parent_bank_idx,
637 0 : ulong slot ) {
638 0 : long before = fd_log_wallclock();
639 :
640 0 : fd_bank_t bank[1];
641 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, bank_idx ) ) ) {
642 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
643 0 : }
644 0 : if( FD_UNLIKELY( bank->data->flags!=FD_BANK_FLAGS_INIT ) ) {
645 0 : FD_LOG_CRIT(( "invariant violation: bank is not in correct state for bank index %lu", bank_idx ));
646 0 : }
647 :
648 0 : bank->data->preparation_begin_nanos = before;
649 :
650 0 : fd_bank_t parent_bank[1];
651 0 : if( FD_UNLIKELY( !fd_banks_bank_query( parent_bank, ctx->banks, parent_bank_idx ) ) ) {
652 0 : FD_LOG_CRIT(( "invariant violation: parent bank is NULL for bank index %lu", parent_bank_idx ));
653 0 : }
654 0 : ulong parent_slot = parent_bank->data->f.slot;
655 :
656 : /* Clone the bank from the parent. We must special case the first
657 : slot that is executed as the snapshot does not provide a parent
658 : block id. */
659 :
660 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( bank, ctx->banks, bank_idx ) ) ) {
661 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
662 0 : }
663 0 : bank->data->f.slot = slot;
664 0 : bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->data->txncache_fork_id );
665 :
666 : /* Create a new funk txn for the block. */
667 :
668 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank_idx } };
669 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
670 0 : fd_accdb_attach_child ( ctx->accdb_admin, &parent_xid, &xid );
671 0 : fd_progcache_attach_child( ctx->progcache, &parent_xid, &xid );
672 :
673 : /* Update required runtime state and handle potential boundary. */
674 :
675 0 : int is_epoch_boundary = 0;
676 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
677 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, bank, 0 );
678 :
679 0 : ulong max_tick_height;
680 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->data->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
681 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->data->f.ticks_per_slot ));
682 0 : }
683 0 : bank->data->f.max_tick_height = max_tick_height;
684 0 : fd_sched_set_poh_params( ctx->sched, bank->data->idx, bank->data->f.tick_height, bank->data->f.max_tick_height, bank->data->f.hashes_per_tick, &parent_bank->data->f.poh );
685 :
686 0 : FD_LOG_DEBUG(( "replay_block_start: bank_idx=%lu slot=%lu parent_bank_idx=%lu", bank_idx, slot, parent_bank_idx ));
687 0 : }
688 :
689 : static void
690 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
691 0 : if( FD_LIKELY( bank->data->cost_tracker_pool_idx!=ULONG_MAX ) ) {
692 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_query( bank );
693 0 : if( FD_UNLIKELY( cost_tracker->block_cost_limit==0UL ) ) {
694 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
695 0 : } else {
696 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
697 0 : slot_info->cost_tracker.vote_cost = cost_tracker->vote_cost;
698 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
699 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
700 0 : slot_info->cost_tracker.vote_cost_limit = cost_tracker->vote_cost_limit;
701 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
702 0 : }
703 0 : } else {
704 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
705 0 : }
706 0 : }
707 :
708 : static ulong
709 0 : get_identity_balance( fd_replay_tile_t * ctx, fd_funk_txn_xid_t xid ) {
710 0 : ulong identity_balance = ULONG_MAX;
711 0 : fd_accdb_ro_t identity_acc[1];
712 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, identity_acc, &xid, ctx->identity_pubkey ) ) ) {
713 0 : identity_balance = identity_acc->meta->lamports;
714 0 : fd_accdb_close_ro( ctx->accdb, identity_acc );
715 0 : }
716 0 : return identity_balance;
717 0 : }
718 :
719 : static void
720 : publish_slot_completed( fd_replay_tile_t * ctx,
721 : fd_stem_context_t * stem,
722 : fd_bank_t * bank,
723 : int is_initial,
724 0 : int is_leader ) {
725 :
726 0 : ulong slot = bank->data->f.slot;
727 :
728 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->data->idx ];
729 :
730 : /* HACKY: hacky way of checking if we should send a null parent block
731 : id */
732 0 : fd_hash_t parent_block_id = {0};
733 0 : if( FD_UNLIKELY( !is_initial ) ) {
734 0 : parent_block_id = ctx->block_id_arr[ bank->data->parent_idx ].latest_mr;
735 0 : }
736 :
737 0 : fd_hash_t const * bank_hash = &bank->data->f.bank_hash;
738 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->data->f.block_hash_queue );
739 0 : FD_TEST( bank_hash );
740 0 : FD_TEST( block_hash );
741 :
742 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->data->txncache_fork_id, 0UL, block_hash->uc );
743 :
744 0 : fd_epoch_schedule_t const * epoch_schedule = &bank->data->f.epoch_schedule;
745 0 : ulong slot_idx;
746 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
747 :
748 0 : ctx->metrics.slots_total++;
749 0 : ctx->metrics.transactions_total = bank->data->f.txn_count;
750 :
751 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
752 0 : slot_info->slot = slot;
753 0 : slot_info->root_slot = ctx->consensus_root_slot;
754 0 : slot_info->storage_slot = ctx->published_root_slot;
755 0 : slot_info->epoch = epoch;
756 0 : slot_info->slot_in_epoch = slot_idx;
757 0 : slot_info->slots_per_epoch = fd_epoch_slot_cnt( epoch_schedule, epoch );
758 0 : slot_info->block_height = bank->data->f.block_height;
759 0 : slot_info->parent_slot = bank->data->f.parent_slot;
760 0 : slot_info->block_id = block_id_ele->latest_mr;
761 0 : slot_info->parent_block_id = parent_block_id;
762 0 : slot_info->bank_hash = *bank_hash;
763 0 : slot_info->block_hash = *block_hash;
764 0 : slot_info->transaction_count = bank->data->f.txn_count;
765 :
766 0 : fd_inflation_t inflation = bank->data->f.inflation;
767 0 : slot_info->inflation.foundation = inflation.foundation;
768 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
769 0 : slot_info->inflation.terminal = inflation.terminal;
770 0 : slot_info->inflation.initial = inflation.initial;
771 0 : slot_info->inflation.taper = inflation.taper;
772 :
773 0 : fd_rent_t rent = bank->data->f.rent;
774 0 : slot_info->rent.burn_percent = rent.burn_percent;
775 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
776 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
777 :
778 0 : slot_info->first_fec_set_received_nanos = bank->data->first_fec_set_received_nanos;
779 0 : slot_info->preparation_begin_nanos = bank->data->preparation_begin_nanos;
780 0 : slot_info->first_transaction_scheduled_nanos = bank->data->first_transaction_scheduled_nanos;
781 0 : slot_info->last_transaction_finished_nanos = bank->data->last_transaction_finished_nanos;
782 0 : slot_info->completion_time_nanos = fd_log_wallclock();
783 0 : if( !slot_info->first_transaction_scheduled_nanos ) { /* edge case: empty slot */
784 0 : slot_info->first_transaction_scheduled_nanos = slot_info->last_transaction_finished_nanos;
785 0 : }
786 :
787 : /* refcnt should be incremented by 1 for each consumer that uses
788 : `bank_idx`. Each consumer should decrement the bank's refcnt once
789 : they are done using the bank. */
790 0 : bank->data->refcnt++; /* tower_tile */
791 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->data->refcnt++; /* rpc tile */
792 0 : slot_info->bank_idx = bank->data->idx;
793 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for tower, rpc", bank->data->idx, slot, bank->data->refcnt ));
794 :
795 0 : fd_bank_t parent_bank[1];
796 0 : if( FD_LIKELY( fd_banks_get_parent( parent_bank, ctx->banks, bank ) ) ) {
797 0 : ulong total_txn_cnt = bank->data->f.txn_count;
798 0 : ulong nonvote_txn_cnt = bank->data->f.nonvote_txn_count;
799 0 : ulong failed_txn_cnt = bank->data->f.failed_txn_count;
800 0 : ulong nonvote_failed_txn_cnt = bank->data->f.nonvote_failed_txn_count;
801 :
802 0 : slot_info->nonvote_success = nonvote_txn_cnt - nonvote_failed_txn_cnt;
803 0 : slot_info->nonvote_failed = nonvote_failed_txn_cnt;
804 0 : slot_info->vote_failed = failed_txn_cnt - nonvote_failed_txn_cnt;
805 0 : slot_info->vote_success = total_txn_cnt - nonvote_txn_cnt - slot_info->vote_failed;
806 0 : } else {
807 0 : slot_info->vote_failed = ULONG_MAX;
808 0 : slot_info->vote_success = ULONG_MAX;
809 0 : slot_info->nonvote_success = ULONG_MAX;
810 0 : slot_info->nonvote_failed = ULONG_MAX;
811 0 : }
812 :
813 0 : slot_info->is_leader = is_leader;
814 0 : slot_info->transaction_fee = bank->data->f.execution_fees;
815 0 : slot_info->transaction_fee -= (slot_info->transaction_fee>>1); /* burn */
816 0 : slot_info->priority_fee = bank->data->f.priority_fees;
817 0 : slot_info->tips = bank->data->f.tips;
818 0 : slot_info->shred_cnt = bank->data->f.shred_cnt;
819 :
820 0 : FD_BASE58_ENCODE_32_BYTES( ctx->block_id_arr[ bank->data->idx ].latest_mr.uc, block_id_cstr );
821 0 : FD_BASE58_ENCODE_32_BYTES( bank->data->f.bank_hash.uc, bank_hash_cstr );
822 0 : FD_LOG_DEBUG(( "publish_slot_completed: bank_idx=%lu slot=%lu bank_hash=%s block_id=%s", bank->data->idx, slot, bank_hash_cstr, block_id_cstr ));
823 :
824 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() ) );
825 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 );
826 0 : }
827 :
828 : static void
829 : publish_slot_dead( fd_replay_tile_t * ctx,
830 : fd_stem_context_t * stem,
831 : ulong slot,
832 0 : fd_hash_t const * block_id ) {
833 0 : fd_replay_slot_dead_t * slot_dead = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
834 0 : slot_dead->slot = slot;
835 0 : slot_dead->block_id = *block_id;
836 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_SLOT_DEAD, ctx->replay_out->chunk, sizeof(fd_replay_slot_dead_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
837 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_slot_dead_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
838 0 : }
839 :
840 : static void
841 : publish_txn_executed( fd_replay_tile_t * ctx,
842 : fd_stem_context_t * stem,
843 0 : ulong txn_idx ) {
844 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
845 0 : fd_replay_txn_executed_t * txn_executed = fd_type_pun( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ) );
846 0 : *txn_executed->txn = *fd_sched_get_txn( ctx->sched, txn_idx );
847 0 : txn_executed->txn_err = txn_info->txn_err;
848 0 : txn_executed->is_committable = !!(txn_info->flags&FD_SCHED_TXN_IS_COMMITTABLE);
849 0 : txn_executed->is_fees_only = !!(txn_info->flags&FD_SCHED_TXN_IS_FEES_ONLY);
850 0 : txn_executed->tick_parsed = txn_info->tick_parsed;
851 0 : txn_executed->tick_sigverify_disp = txn_info->tick_sigverify_disp;
852 0 : txn_executed->tick_sigverify_done = txn_info->tick_sigverify_done;
853 0 : txn_executed->tick_exec_disp = txn_info->tick_exec_disp;
854 0 : txn_executed->tick_exec_done = txn_info->tick_exec_done;
855 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_TXN_EXECUTED, ctx->replay_out->chunk, sizeof(*txn_executed), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
856 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(*txn_executed), ctx->replay_out->chunk0, ctx->replay_out->wmark );
857 0 : }
858 :
859 : static void
860 : replay_block_finalize( fd_replay_tile_t * ctx,
861 : fd_stem_context_t * stem,
862 0 : fd_bank_t * bank ) {
863 0 : bank->data->last_transaction_finished_nanos = fd_log_wallclock();
864 :
865 0 : FD_TEST( !(bank->data->flags&FD_BANK_FLAGS_FROZEN) );
866 :
867 : /* Set poh hash in bank. */
868 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->data->idx );
869 0 : bank->data->f.poh = *poh;
870 :
871 : /* Set shred count in bank. */
872 0 : bank->data->f.shred_cnt = fd_sched_get_shred_cnt( ctx->sched, bank->data->idx );
873 :
874 : /* Do hashing and other end-of-block processing. */
875 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
876 :
877 : /* Copy out cost tracker fields before freezing */
878 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
879 0 : cost_tracker_snap( bank, slot_info );
880 :
881 : /* fetch identity / vote balance updates infrequently */
882 0 : ulong slot = bank->data->f.slot;
883 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank->data->idx } };
884 0 : slot_info->identity_balance = FD_UNLIKELY( slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
885 :
886 : /* Mark the bank as frozen. */
887 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
888 :
889 : /**********************************************************************/
890 : /* Bank hash comparison, and halt if there's a mismatch after replay */
891 : /**********************************************************************/
892 :
893 0 : fd_hash_t const * bank_hash = &bank->data->f.bank_hash;
894 0 : FD_TEST( bank_hash );
895 :
896 : /* Must be last so we can measure completion time correctly, even
897 : though we could technically do this before the hash cmp and vote
898 : tower stuff. */
899 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */ );
900 :
901 0 : # if FD_HAS_FLATCC
902 : /* If enabled, dump the block to a file and reset the dumping
903 : context state */
904 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
905 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, ctx->accdb, ctx->dump_proto_ctx, ctx->runtime_stack );
906 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
907 0 : }
908 0 : # endif
909 0 : }
910 :
911 : /**********************************************************************/
912 : /* Leader bank management */
913 : /**********************************************************************/
914 :
915 : static fd_bank_t *
916 : prepare_leader_bank( fd_replay_tile_t * ctx,
917 : fd_stem_context_t * stem,
918 : ulong slot,
919 : long now,
920 0 : fd_hash_t const * parent_block_id ) {
921 0 : long before = fd_log_wallclock();
922 :
923 : /* Make sure that we are not already leader. */
924 0 : FD_TEST( ctx->leader_bank->data==NULL );
925 :
926 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 );
927 0 : if( FD_UNLIKELY( !parent_ele ) ) {
928 0 : FD_BASE58_ENCODE_32_BYTES( parent_block_id->key, parent_block_id_b58 );
929 0 : FD_LOG_CRIT(( "invariant violation: parent bank index not found for merkle root %s", parent_block_id_b58 ));
930 0 : }
931 0 : ulong parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent_ele );
932 :
933 0 : fd_bank_t parent_bank[1];
934 0 : if( FD_UNLIKELY( !fd_banks_bank_query( parent_bank, ctx->banks, parent_bank_idx ) ) ) {
935 0 : FD_LOG_CRIT(( "invariant violation: parent bank not found for bank index %lu", parent_bank_idx ));
936 0 : }
937 0 : ulong parent_slot = parent_bank->data->f.slot;
938 :
939 0 : if( FD_UNLIKELY( !fd_banks_new_bank( ctx->leader_bank, ctx->banks, parent_bank_idx, now ) ) ) {
940 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
941 0 : }
942 :
943 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( ctx->leader_bank, ctx->banks, ctx->leader_bank->data->idx ) ) ) {
944 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
945 0 : }
946 :
947 0 : ctx->leader_bank->data->preparation_begin_nanos = before;
948 :
949 0 : ctx->leader_bank->data->f.slot = slot;
950 0 : ctx->leader_bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->data->txncache_fork_id );
951 : /* prepare the funk transaction for the leader bank */
952 0 : fd_funk_txn_xid_t xid = { .ul = { slot, ctx->leader_bank->data->idx } };
953 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
954 0 : fd_accdb_attach_child ( ctx->accdb_admin, &parent_xid, &xid );
955 0 : fd_progcache_attach_child( ctx->progcache, &parent_xid, &xid );
956 :
957 0 : int is_epoch_boundary = 0;
958 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
959 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, ctx->leader_bank, 0 );
960 :
961 0 : ulong max_tick_height;
962 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->data->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
963 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->data->f.ticks_per_slot ));
964 0 : }
965 0 : ctx->leader_bank->data->f.max_tick_height = max_tick_height;
966 :
967 : /* Now that a bank has been created for the leader slot, increment the
968 : reference count until we are done with the leader slot. */
969 0 : ctx->leader_bank->data->refcnt++;
970 :
971 0 : return ctx->leader_bank;
972 0 : }
973 :
974 : static inline void
975 0 : maybe_switch_identity( fd_replay_tile_t * ctx ) {
976 :
977 0 : if( FD_LIKELY( fd_keyswitch_state_query( ctx->keyswitch )!=FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) return;
978 :
979 : /* Switch identity */
980 :
981 0 : FD_LOG_DEBUG(( "keyswitch: switching identity" ));
982 :
983 0 : memcpy( ctx->identity_pubkey, ctx->keyswitch->bytes, 32UL );
984 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
985 :
986 : /* The next leader slot will be incorrect now that the identity has
987 : switched. The next leader slot normally gets updated based on the
988 : reset slot returned by tower. */
989 0 : ulong min_leader_slot = fd_ulong_max( ctx->reset_slot+1UL, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot+1UL ) );
990 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
991 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
992 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
993 0 : } else {
994 0 : ctx->next_leader_tickcount = LONG_MAX;
995 0 : }
996 :
997 0 : ctx->identity_vote_rooted = 0;
998 0 : ctx->identity_idx++;
999 0 : fd_vote_tracker_reset( ctx->vote_tracker );
1000 0 : }
1001 :
1002 : static void
1003 : fini_leader_bank( fd_replay_tile_t * ctx,
1004 0 : fd_stem_context_t * stem ) {
1005 :
1006 0 : FD_TEST( ctx->leader_bank->data!=NULL );
1007 0 : FD_TEST( ctx->is_leader );
1008 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen );
1009 0 : FD_TEST( ctx->recv_poh );
1010 :
1011 0 : ctx->leader_bank->data->last_transaction_finished_nanos = fd_log_wallclock();
1012 :
1013 0 : ulong curr_slot = ctx->leader_bank->data->f.slot;
1014 :
1015 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->data->idx, ctx->leader_bank->data->parent_idx, curr_slot );
1016 :
1017 0 : fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx );
1018 :
1019 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1020 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
1021 0 : fd_funk_txn_xid_t xid = { .ul = { curr_slot, ctx->leader_bank->data->idx } };
1022 0 : slot_info->identity_balance = FD_UNLIKELY( curr_slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
1023 :
1024 0 : fd_banks_mark_bank_frozen( ctx->banks, ctx->leader_bank );
1025 :
1026 0 : fd_hash_t const * bank_hash = &ctx->leader_bank->data->f.bank_hash;
1027 0 : FD_TEST( bank_hash );
1028 :
1029 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
1030 :
1031 : /* The reference on the bank is finally no longer needed. */
1032 0 : ctx->leader_bank->data->refcnt--;
1033 :
1034 : /* We are no longer leader so we can clear the bank index we use for
1035 : being the leader. */
1036 0 : ctx->leader_bank->data = NULL;
1037 0 : ctx->recv_poh = 0;
1038 0 : ctx->is_leader = 0;
1039 :
1040 0 : maybe_switch_identity( ctx );
1041 0 : }
1042 :
1043 : static void
1044 : publish_root_advanced( fd_replay_tile_t * ctx,
1045 0 : fd_stem_context_t * stem ) {
1046 :
1047 0 : fd_bank_t bank[1];
1048 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, ctx->consensus_root_bank_idx ) ) ) {
1049 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank is NULL at bank index %lu", ctx->consensus_root_bank_idx ));
1050 0 : }
1051 :
1052 0 : if( FD_UNLIKELY( bank->data->f.epoch>fd_slot_to_epoch( &bank->data->f.epoch_schedule, bank->data->f.parent_slot, NULL ) )) {
1053 0 : publish_epoch_info( ctx, stem, bank, 1 );
1054 0 : }
1055 :
1056 0 : if( ctx->rpc_enabled ) {
1057 0 : bank->data->refcnt++;
1058 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->data->idx, bank->data->f.slot, bank->data->refcnt ));
1059 0 : }
1060 :
1061 : /* Increment the reference count on the consensus root bank to account
1062 : for the number of resolv tiles that are waiting on it. */
1063 0 : bank->data->refcnt += ctx->resolv_tile_cnt;
1064 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for resolv", bank->data->idx, bank->data->f.slot, bank->data->refcnt ));
1065 :
1066 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1067 0 : msg->bank_idx = bank->data->idx;
1068 :
1069 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() ) );
1070 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 );
1071 0 : }
1072 :
1073 : /* init_funk performs pre-flight checks for the account database and
1074 : program cache. Ensures that the account database was set up
1075 : correctly by bootstrap components (e.g. genesis or snapshot loader).
1076 : Mirrors the account database's fork tree down to the program cache. */
1077 :
1078 : static void
1079 : init_funk( fd_replay_tile_t * ctx,
1080 0 : ulong bank_slot ) {
1081 : /* Ensure that the loaded bank root corresponds to the account
1082 : database's root. */
1083 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1084 0 : if( FD_UNLIKELY( !funk->shmem ) ) {
1085 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to database shared memory objects" ));
1086 0 : }
1087 0 : fd_funk_txn_xid_t const * accdb_pub = fd_funk_last_publish( funk );
1088 0 : if( FD_UNLIKELY( accdb_pub->ul[0]!=bank_slot ) ) {
1089 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb is at slot %lu, but chain state is at slot %lu\n"
1090 0 : "This is a bug in startup components.",
1091 0 : accdb_pub->ul[0], bank_slot ));
1092 0 : }
1093 0 : if( FD_UNLIKELY( fd_funk_last_publish_is_frozen( funk ) ) ) {
1094 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb fork graph is not clean.\n"
1095 0 : "The account database should only contain state for the root slot at this point,\n"
1096 0 : "but there are incomplete database transactions leftover.\n"
1097 0 : "This is a bug in startup components." ));
1098 0 : }
1099 :
1100 : /* The program cache tracks the account database's fork graph at all
1101 : times. Perform initial synchronization: pivot from funk 'root' (a
1102 : sentinel XID) to 'last publish' (the bootstrap root slot). */
1103 0 : if( FD_UNLIKELY( !ctx->progcache->shmem ) ) {
1104 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to program cache" ));
1105 0 : }
1106 0 : fd_progcache_clear( ctx->progcache );
1107 :
1108 0 : fd_funk_txn_xid_t last_publish = fd_accdb_root_get( ctx->accdb_admin );
1109 0 : fd_funk_txn_xid_t root = { .ul = { ULONG_MAX, ULONG_MAX } };
1110 0 : fd_progcache_attach_child( ctx->progcache, &root, &last_publish );
1111 0 : fd_progcache_advance_root( ctx->progcache, &last_publish );
1112 0 : }
1113 :
1114 : static void
1115 0 : init_after_snapshot( fd_replay_tile_t * ctx ) {
1116 : /* Now that the snapshot has been loaded in, we have to refresh the
1117 : stake delegations since the manifest does not contain the full set
1118 : of data required for the stake delegations. See
1119 : fd_stake_delegations.h for why this is required. */
1120 :
1121 0 : fd_bank_t bank[1];
1122 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) ) ) {
1123 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1124 0 : }
1125 :
1126 0 : fd_funk_txn_xid_t xid = { .ul = { bank->data->f.slot, bank->data->idx } };
1127 0 : init_funk( ctx, bank->data->f.slot );
1128 :
1129 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
1130 0 : fd_stake_delegations_refresh( root_delegations, ctx->accdb, &xid );
1131 :
1132 0 : bank->data->f.warmup_cooldown_rate_epoch = fd_slot_to_epoch( &bank->data->f.epoch_schedule,
1133 0 : bank->data->f.features.reduce_stake_warmup_cooldown,
1134 0 : NULL );
1135 :
1136 0 : fd_stakes_init_totals( bank, root_delegations, ctx->accdb, &xid );
1137 :
1138 0 : fd_top_votes_t * top_votes_t_2 = fd_bank_top_votes_t_2_modify( bank );
1139 0 : fd_top_votes_refresh( top_votes_t_2, ctx->accdb, &xid );
1140 :
1141 : /* After both snapshots have been loaded in, we can determine if we should
1142 : start distributing rewards. */
1143 :
1144 0 : fd_rewards_recalculate_partitioned_rewards( ctx->banks, bank, ctx->accdb, &xid, ctx->runtime_stack, ctx->capture_ctx );
1145 :
1146 0 : ulong snapshot_slot = bank->data->f.slot;
1147 0 : if( FD_UNLIKELY( !snapshot_slot ) ) {
1148 : /* Genesis-specific setup. */
1149 : /* FIXME: This branch does not set up a new block exec ctx
1150 : properly. Needs to do whatever prepare_new_block_execution
1151 : does, but just hacking that in breaks stuff. */
1152 0 : fd_runtime_update_leaders( bank, ctx->runtime_stack );
1153 :
1154 0 : ulong hashcnt_per_slot = bank->data->f.hashes_per_tick * bank->data->f.ticks_per_slot;
1155 0 : fd_hash_t * poh = &bank->data->f.poh;
1156 0 : while( hashcnt_per_slot-- ) {
1157 0 : fd_sha256_hash( poh->hash, 32UL, poh->hash );
1158 0 : }
1159 :
1160 0 : int is_epoch_boundary = 0;
1161 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1162 0 : FD_TEST( !is_epoch_boundary );
1163 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
1164 :
1165 0 : snapshot_slot = 0UL;
1166 0 : }
1167 0 : }
1168 :
1169 : static inline int
1170 : maybe_become_leader( fd_replay_tile_t * ctx,
1171 0 : fd_stem_context_t * stem ) {
1172 0 : FD_TEST( ctx->is_booted );
1173 0 : if( FD_LIKELY( ctx->next_leader_slot==ULONG_MAX || ctx->is_leader || (!ctx->identity_vote_rooted && ctx->wait_for_vote_to_start_leader) || ctx->replay_out->idx==ULONG_MAX || !ctx->wfs_complete ) ) return 0;
1174 0 : if( FD_UNLIKELY( ctx->halt_leader ) ) return 0;
1175 0 : if( !ctx->supports_leader ) return 0;
1176 :
1177 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1178 0 : long now = fd_tickcount();
1179 0 : if( FD_LIKELY( now<ctx->next_leader_tickcount ) ) return 0;
1180 :
1181 : /* If a prior leader is still in the process of publishing their slot,
1182 : delay ours to let them finish ... unless they are so delayed that
1183 : we risk getting skipped by the leader following us. 1.2 seconds
1184 : is a reasonable default here, although any value between 0 and 1.6
1185 : seconds could be considered reasonable. This is arbitrary and
1186 : chosen due to intuition. */
1187 0 : if( FD_UNLIKELY( now<ctx->next_leader_tickcount+(long)(3.0*ctx->slot_duration_ticks) ) ) {
1188 0 : FD_TEST( ctx->reset_bank->data );
1189 :
1190 : /* TODO: Make the max_active_descendant calculation more efficient
1191 : by caching it in the bank structure and updating it as banks are
1192 : created and completed. */
1193 0 : ulong max_active_descendant = 0UL;
1194 0 : ulong child_idx = ctx->reset_bank->data->child_idx;
1195 0 : while( child_idx!=ULONG_MAX ) {
1196 0 : fd_bank_t child_bank[1];
1197 0 : fd_banks_bank_query( child_bank, ctx->banks, child_idx );
1198 0 : max_active_descendant = fd_ulong_max( max_active_descendant, child_bank->data->f.slot );
1199 0 : child_idx = child_bank->data->sibling_idx;
1200 0 : }
1201 :
1202 : /* If the max_active_descendant is >= next_leader_slot, we waited
1203 : too long and a leader after us started publishing to try and skip
1204 : us. Just start our leader slot immediately, we might win ... */
1205 0 : if( FD_LIKELY( max_active_descendant>=ctx->reset_slot && max_active_descendant<ctx->next_leader_slot ) ) {
1206 : /* If one of the leaders between the reset slot and our leader
1207 : slot is in the process of publishing (they have a descendant
1208 : bank that is in progress of being replayed), then keep waiting.
1209 : We probably wouldn't get a leader slot out before they
1210 : finished.
1211 :
1212 : Unless... we are past the deadline to start our slot by more
1213 : than 1.2 seconds, in which case we should probably start it to
1214 : avoid getting skipped by the leader behind us. */
1215 0 : return 0;
1216 0 : }
1217 0 : }
1218 :
1219 0 : long now_nanos = fd_log_wallclock();
1220 :
1221 0 : ctx->is_leader = 1;
1222 0 : ctx->recv_poh = 0;
1223 :
1224 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1225 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1226 :
1227 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1228 :
1229 : /* Acquires bank, sets up initial state, and refcnts it. */
1230 0 : fd_bank_t * bank = prepare_leader_bank( ctx, stem, ctx->next_leader_slot, now_nanos, &ctx->reset_block_id );
1231 0 : fd_funk_txn_xid_t xid = { .ul = { ctx->next_leader_slot, ctx->leader_bank->data->idx } };
1232 :
1233 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1234 0 : fd_pubkey_t tip_receiver_owner = {0};
1235 :
1236 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1237 0 : fd_acct_addr_t tip_payment_config[1];
1238 0 : fd_acct_addr_t tip_receiver[1];
1239 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, bank->data->f.epoch, tip_payment_config, tip_receiver );
1240 :
1241 0 : fd_accdb_ro_t tip_config_acc[1];
1242 0 : if( FD_UNLIKELY( !fd_accdb_open_ro( ctx->accdb, tip_config_acc, &xid, tip_payment_config ) ) ) {
1243 : /* FIXME This should not crash the validator */
1244 0 : FD_BASE58_ENCODE_32_BYTES( tip_payment_config->b, tip_config_acc_b58 );
1245 0 : FD_LOG_CRIT(( "tip payment config account %s does not exist", tip_config_acc_b58 ));
1246 0 : }
1247 0 : ulong tip_cfg_sz = fd_accdb_ref_data_sz( tip_config_acc );
1248 0 : if( FD_UNLIKELY( tip_cfg_sz < sizeof(fd_bundle_crank_tip_payment_config_t) ) ) {
1249 : /* FIXME This should not crash the validator */
1250 0 : FD_LOG_HEXDUMP_CRIT(( "invalid tip payment config account data", fd_accdb_ref_data_const( tip_config_acc ), tip_cfg_sz ));
1251 0 : }
1252 0 : memcpy( config, fd_accdb_ref_data_const( tip_config_acc ), sizeof(fd_bundle_crank_tip_payment_config_t) );
1253 0 : fd_accdb_close_ro( ctx->accdb, tip_config_acc );
1254 :
1255 : /* It is possible that the tip receiver account does not exist yet
1256 : if it is the first time in an epoch. */
1257 0 : fd_accdb_ro_t tip_receiver_acc[1];
1258 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, tip_receiver_acc, &xid, tip_receiver ) ) ) {
1259 0 : tip_receiver_owner = FD_LOAD( fd_pubkey_t, fd_accdb_ref_owner( tip_receiver_acc ) );
1260 0 : fd_accdb_close_ro( ctx->accdb, tip_receiver_acc );
1261 0 : }
1262 0 : }
1263 :
1264 :
1265 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1266 0 : msg->slot = ctx->next_leader_slot;
1267 0 : msg->slot_start_ns = now_nanos;
1268 0 : msg->slot_end_ns = now_nanos+(long)ctx->slot_duration_nanos;
1269 0 : msg->bank = NULL;
1270 0 : msg->bank_idx = bank->data->idx;
1271 0 : msg->ticks_per_slot = bank->data->f.ticks_per_slot;
1272 0 : msg->hashcnt_per_tick = bank->data->f.hashes_per_tick;
1273 0 : msg->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)msg->ticks_per_slot);
1274 0 : msg->bundle->config[0] = config[0];
1275 0 : memcpy( msg->bundle->last_blockhash, bank->data->f.poh.hash, sizeof(fd_hash_t) );
1276 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner.uc, sizeof(fd_pubkey_t) );
1277 :
1278 0 : if( FD_UNLIKELY( msg->hashcnt_per_tick==1UL ) ) {
1279 : /* Low power producer, maximum of one microblock per tick in the slot */
1280 0 : msg->max_microblocks_in_slot = msg->ticks_per_slot;
1281 0 : } else {
1282 : /* See the long comment in after_credit for this limit */
1283 0 : msg->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, msg->ticks_per_slot*(msg->hashcnt_per_tick-1UL) );
1284 0 : }
1285 :
1286 0 : msg->total_skipped_ticks = msg->ticks_per_slot*(ctx->next_leader_slot-ctx->reset_slot);
1287 0 : msg->epoch = fd_slot_to_epoch( &bank->data->f.epoch_schedule, ctx->next_leader_slot, NULL );
1288 :
1289 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_query( bank );
1290 :
1291 0 : msg->limits.slot_max_cost = ctx->larger_max_cost_per_block ? LARGER_MAX_COST_PER_BLOCK : cost_tracker->block_cost_limit;
1292 0 : msg->limits.slot_max_vote_cost = cost_tracker->vote_cost_limit;
1293 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1294 :
1295 0 : if( FD_UNLIKELY( msg->ticks_per_slot+msg->total_skipped_ticks>USHORT_MAX ) ) {
1296 : /* There can be at most USHORT_MAX skipped ticks, because the
1297 : parent_offset field in the shred data is only 2 bytes wide. */
1298 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 ));
1299 0 : }
1300 :
1301 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() ) );
1302 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 );
1303 :
1304 0 : ctx->next_leader_slot = ULONG_MAX;
1305 0 : ctx->next_leader_tickcount = LONG_MAX;
1306 :
1307 0 : return 1;
1308 0 : }
1309 :
1310 : static void
1311 : process_poh_message( fd_replay_tile_t * ctx,
1312 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
1313 :
1314 0 : FD_TEST( ctx->is_booted );
1315 0 : FD_TEST( ctx->is_leader );
1316 0 : FD_TEST( ctx->leader_bank->data!=NULL );
1317 :
1318 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
1319 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
1320 :
1321 : /* Update the poh hash in the bank. We will want to maintain a refcnt
1322 : on the bank until we have recieved the block id for the block after
1323 : it has been shredded. */
1324 :
1325 0 : memcpy( &ctx->leader_bank->data->f.poh, slot_ended->blockhash, sizeof(fd_hash_t) );
1326 :
1327 0 : ctx->recv_poh = 1;
1328 0 : }
1329 :
1330 : static void
1331 : publish_reset( fd_replay_tile_t * ctx,
1332 : fd_stem_context_t * stem,
1333 0 : fd_bank_t * bank ) {
1334 0 : if( FD_UNLIKELY( ctx->replay_out->idx==ULONG_MAX ) ) return;
1335 :
1336 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->data->f.block_hash_queue );
1337 0 : FD_TEST( block_hash );
1338 :
1339 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1340 :
1341 0 : reset->bank_idx = bank->data->idx;
1342 0 : reset->timestamp = fd_log_wallclock();
1343 0 : reset->completed_slot = bank->data->f.slot;
1344 0 : reset->hashcnt_per_tick = bank->data->f.hashes_per_tick;
1345 0 : reset->ticks_per_slot = bank->data->f.ticks_per_slot;
1346 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
1347 0 : fd_memcpy( reset->completed_block_id, ctx->reset_block_id.uc, sizeof(fd_hash_t) );
1348 0 : fd_memcpy( reset->completed_blockhash, block_hash->uc, sizeof(fd_hash_t) );
1349 :
1350 0 : ulong ticks_per_slot = bank->data->f.ticks_per_slot;
1351 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
1352 : /* Low power producer, maximum of one microblock per tick in the slot */
1353 0 : reset->max_microblocks_in_slot = ticks_per_slot;
1354 0 : } else {
1355 : /* See the long comment in after_credit for this limit */
1356 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
1357 0 : }
1358 0 : reset->next_leader_slot = ctx->next_leader_slot;
1359 :
1360 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() ) );
1361 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 );
1362 0 : }
1363 :
1364 : static void
1365 : store_xinsert( fd_store_t * store,
1366 0 : fd_hash_t const * merkle_root ) {
1367 0 : fd_store_pool_t pool = {
1368 0 : .pool = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_mem_gaddr ),
1369 0 : .ele = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_ele_gaddr ),
1370 0 : .ele_max = store->fec_max
1371 0 : };
1372 0 : int err; fd_store_fec_t * fec = fd_store_pool_acquire( &pool, NULL, 1 /* blocking */, &err );
1373 0 : if( FD_UNLIKELY( err!=FD_POOL_SUCCESS ) ) FD_LOG_CRIT(( "store pool: %s", fd_store_pool_strerror( err ) ));
1374 0 : fec->key.merkle_root = *merkle_root;
1375 0 : fec->key.part_idx = 0;
1376 0 : fec->cmr = (fd_hash_t){ 0 };
1377 0 : fec->next = fd_store_pool_idx_null();
1378 0 : fec->data_sz = 0UL;
1379 :
1380 0 : FD_STORE_XLOCK_BEGIN( store ) {
1381 0 : fd_store_map_ele_insert( fd_wksp_laddr_fast( fd_store_wksp( store ), store->map_gaddr ), fec, pool.ele );
1382 0 : } FD_STORE_XLOCK_END;
1383 0 : }
1384 :
1385 : static void
1386 : boot_genesis( fd_replay_tile_t * ctx,
1387 : fd_stem_context_t * stem,
1388 0 : fd_genesis_meta_t const * meta ) {
1389 : /* If we are bootstrapping, we can't wait to wait for our identity
1390 : vote to be rooted as this creates a circular dependency. */
1391 0 : ctx->identity_vote_rooted = 1;
1392 :
1393 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
1394 0 : FD_TEST( meta->bootstrap && meta->has_lthash );
1395 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
1396 :
1397 0 : fd_bank_t bank[1];
1398 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) );
1399 0 : fd_funk_txn_xid_t xid = { .ul = { 0UL, FD_REPLAY_BOOT_BANK_IDX } };
1400 :
1401 : /* Do genesis-related processing in a non-rooted transaction */
1402 0 : fd_funk_txn_xid_t root_xid = { .ul = { LONG_MAX, LONG_MAX } };
1403 0 : fd_funk_txn_xid_t target_xid = { .ul = { 0UL, 0UL } };
1404 0 : fd_accdb_attach_child( ctx->accdb_admin, &root_xid, &target_xid );
1405 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, &xid, NULL, &meta->genesis_hash, &meta->lthash, ctx->genesis, genesis_blob, ctx->runtime_stack );
1406 0 : fd_accdb_advance_root( ctx->accdb_admin, &target_xid );
1407 :
1408 0 : static const fd_txncache_fork_id_t txncache_root = { .val = USHORT_MAX };
1409 0 : bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, txncache_root );
1410 :
1411 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->data->f.block_hash_queue );
1412 0 : fd_txncache_finalize_fork( ctx->txncache, bank->data->txncache_fork_id, 0UL, block_hash->uc );
1413 :
1414 0 : publish_epoch_info( ctx, stem, bank, 0 );
1415 0 : publish_epoch_info( ctx, stem, bank, 1 );
1416 :
1417 : /* We call this after fd_runtime_read_genesis, which sets up the
1418 : slot_bank needed in blockstore_init. */
1419 0 : init_after_snapshot( ctx );
1420 :
1421 0 : ctx->published_root_slot = 0UL;
1422 0 : fd_sched_block_add_done( ctx->sched, bank->data->idx, ULONG_MAX, 0UL );
1423 :
1424 0 : bank->data->f.block_height = 1UL;
1425 :
1426 0 : ctx->consensus_root = ctx->initial_block_id;
1427 0 : ctx->consensus_root_slot = 0UL;
1428 0 : ctx->consensus_root_bank_idx = 0UL;
1429 0 : ctx->published_root_slot = 0UL;
1430 0 : ctx->published_root_bank_idx = 0UL;
1431 :
1432 0 : ctx->reset_slot = 0UL;
1433 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
1434 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1435 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1436 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1437 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1438 0 : } else {
1439 0 : ctx->next_leader_tickcount = LONG_MAX;
1440 0 : }
1441 :
1442 0 : ctx->is_booted = 1;
1443 0 : maybe_become_leader( ctx, stem );
1444 :
1445 0 : fd_hash_t initial_block_id = ctx->initial_block_id;
1446 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &initial_block_id, NULL, 0 /* genesis slot */, 0, 0, 0, 0, 1, 0, ctx->store, &ctx->reasm_evicted ); /* FIXME manifest block_id */
1447 0 : fec->bank_idx = bank->data->idx;
1448 0 : fec->bank_seq = bank->data->bank_seq;
1449 0 : store_xinsert( ctx->store, &initial_block_id );
1450 :
1451 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1452 0 : block_id_ele->latest_mr = initial_block_id;
1453 0 : block_id_ele->slot = 0UL;
1454 :
1455 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1456 :
1457 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1458 0 : cost_tracker_snap( bank, slot_info );
1459 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1460 :
1461 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1462 0 : publish_root_advanced( ctx, stem );
1463 0 : publish_reset( ctx, stem, bank );
1464 0 : }
1465 :
1466 : static inline void
1467 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
1468 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->has_genesis_hash ) ) {
1469 0 : return;
1470 0 : }
1471 :
1472 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, hash_cstr );
1473 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
1474 : /* Map pyth-related clusters to unkwown. */
1475 0 : switch( cluster ) {
1476 0 : case FD_CLUSTER_PYTHNET:
1477 0 : case FD_CLUSTER_PYTHTEST:
1478 0 : cluster = FD_CLUSTER_UNKNOWN;
1479 0 : }
1480 :
1481 0 : if( FD_UNLIKELY( cluster!=ctx->cluster_type ) ) {
1482 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis hash of `%s` which means the cluster is %s "
1483 0 : "but the snapshot you loaded is for a different cluster %s. If you are trying to join the "
1484 0 : "%s cluster, you can delete the genesis.bin file and restart the node to download the correct "
1485 0 : "genesis file automatically.",
1486 0 : ctx->genesis_path,
1487 0 : hash_cstr,
1488 0 : fd_genesis_cluster_name( cluster ),
1489 0 : fd_genesis_cluster_name( ctx->cluster_type ),
1490 0 : fd_genesis_cluster_name( cluster ) ));
1491 0 : }
1492 0 : }
1493 :
1494 : static void
1495 : on_snapshot_message( fd_replay_tile_t * ctx,
1496 : fd_stem_context_t * stem,
1497 : ulong in_idx,
1498 : ulong chunk,
1499 0 : ulong sig ) {
1500 0 : ulong msg = fd_ssmsg_sig_message( sig );
1501 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
1502 : /* An end of message notification indicates the snapshot is loaded.
1503 : Replay is able to start executing from this point onwards. */
1504 : /* TODO: replay should finish booting. Could make replay a
1505 : state machine and set the state here accordingly. */
1506 0 : ctx->is_booted = 1;
1507 :
1508 0 : fd_bank_t bank[1];
1509 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) ) ) {
1510 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1511 0 : }
1512 :
1513 0 : ulong snapshot_slot = bank->data->f.slot;
1514 :
1515 0 : fd_hash_t bank_hash = bank->data->f.bank_hash;
1516 0 : if( FD_UNLIKELY( ctx->wfs_enabled && memcmp( ctx->expected_bank_hash.uc, bank_hash.uc, sizeof(fd_hash_t) ) ) ) {
1517 0 : FD_BASE58_ENCODE_32_BYTES( ctx->expected_bank_hash.uc, expected_bank_hash_cstr );
1518 0 : FD_BASE58_ENCODE_32_BYTES( bank_hash.uc, actual_bank_hash_cstr );
1519 0 : FD_LOG_ERR(( "[consensus.wait_for_supermajority_with_bank_hash] expected_bank_hash=%s does not match snapshot slot"
1520 0 : "=%lu bank_hash=%s. If you are loading a snapshot from the network, check that the slot matches the "
1521 0 : "cluster restart slot. ", expected_bank_hash_cstr, snapshot_slot, actual_bank_hash_cstr ));
1522 0 : }
1523 0 : if( FD_UNLIKELY( ctx->wfs_enabled ) ) {
1524 0 : FD_LOG_NOTICE(( "waiting for supermajority at snapshot slot %lu", snapshot_slot ));
1525 0 : }
1526 :
1527 : /* FIXME: This is a hack because the block id of the snapshot slot
1528 : is not provided in the snapshot. A possible solution is to get
1529 : the block id of the snapshot slot from repair. */
1530 0 : fd_hash_t manifest_block_id = ctx->initial_block_id;
1531 :
1532 0 : fd_funk_txn_xid_t xid = { .ul = { snapshot_slot, FD_REPLAY_BOOT_BANK_IDX } };
1533 0 : fd_features_restore( bank, ctx->accdb, &xid );
1534 :
1535 0 : FD_TEST( fd_sysvar_cache_restore( bank, ctx->accdb, &xid ) );
1536 :
1537 0 : ctx->consensus_root = manifest_block_id;
1538 0 : ctx->consensus_root_slot = snapshot_slot;
1539 0 : ctx->consensus_root_bank_idx = 0UL;
1540 0 : ctx->published_root_slot = ctx->consensus_root_slot;
1541 0 : ctx->published_root_bank_idx = 0UL;
1542 :
1543 0 : ctx->reset_slot = snapshot_slot;
1544 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
1545 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1546 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1547 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1548 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1549 0 : } else {
1550 0 : ctx->next_leader_tickcount = LONG_MAX;
1551 0 : }
1552 :
1553 0 : fd_sched_block_add_done( ctx->sched, bank->data->idx, ULONG_MAX, snapshot_slot );
1554 0 : FD_TEST( bank->data->idx==0UL );
1555 :
1556 0 : fd_runtime_update_leaders( bank, ctx->runtime_stack );
1557 :
1558 : /* Typically, when we cross an epoch boundary during normal
1559 : operation, we publish the stake weights for the new epoch. But
1560 : since we are starting from a snapshot, we need to publish two
1561 : epochs worth of stake weights: the previous epoch (which is
1562 : needed for voting on the current epoch), and the current epoch
1563 : (which is needed for voting on the next epoch). */
1564 0 : publish_epoch_info( ctx, stem, bank, 0 );
1565 0 : publish_epoch_info( ctx, stem, bank, 1 );
1566 :
1567 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1568 0 : block_id_ele->latest_mr = manifest_block_id;
1569 0 : block_id_ele->slot = snapshot_slot;
1570 0 : block_id_ele->block_id_seen = 1;
1571 0 : block_id_ele->latest_fec_idx = 0U;
1572 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1573 :
1574 : /* We call this after fd_runtime_read_genesis, which sets up the
1575 : slot_bank needed in blockstore_init. */
1576 0 : init_after_snapshot( ctx );
1577 :
1578 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1579 0 : cost_tracker_snap( bank, slot_info );
1580 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1581 :
1582 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1583 0 : publish_root_advanced( ctx, stem );
1584 :
1585 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &manifest_block_id, NULL, snapshot_slot, 0, 0, 0, 0, 1, 0, ctx->store, &ctx->reasm_evicted ); /* FIXME manifest block_id */
1586 0 : fec->bank_idx = bank->data->idx;
1587 0 : fec->bank_seq = bank->data->bank_seq;
1588 0 : store_xinsert( ctx->store, &manifest_block_id );
1589 :
1590 0 : ctx->cluster_type = bank->data->f.cluster_type;
1591 :
1592 0 : maybe_verify_cluster_type( ctx );
1593 :
1594 0 : return;
1595 0 : }
1596 :
1597 0 : switch( msg ) {
1598 0 : case FD_SSMSG_MANIFEST_FULL:
1599 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
1600 : /* We may either receive a full snapshot manifest or an
1601 : incremental snapshot manifest. Note that this external message
1602 : id is only used temporarily because replay cannot yet receive
1603 : the firedancer-internal snapshot manifest message. */
1604 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1605 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 ));
1606 :
1607 0 : fd_bank_t bank[1];
1608 0 : fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
1609 0 : ctx->banks,
1610 0 : fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
1611 0 : ctx->runtime_stack,
1612 0 : msg==FD_SSMSG_MANIFEST_INCREMENTAL );
1613 :
1614 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1615 0 : ctx->hard_forks_cnt = manifest->hard_forks_len;
1616 0 : for( ulong i=0UL; i<manifest->hard_forks_len; i++ ) {
1617 0 : ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
1618 0 : ctx->hard_forks_cnts[ i ] = manifest->hard_forks_cnts[ i ];
1619 0 : }
1620 0 : ctx->has_expected_genesis_timestamp = 1;
1621 0 : ctx->expected_genesis_timestamp = manifest->creation_time_seconds;
1622 0 : break;
1623 0 : }
1624 0 : default: {
1625 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
1626 0 : return;
1627 0 : }
1628 0 : }
1629 :
1630 0 : return;
1631 0 : }
1632 :
1633 : static void
1634 : dispatch_task( fd_replay_tile_t * ctx,
1635 : fd_stem_context_t * stem,
1636 0 : fd_sched_task_t * task ) {
1637 :
1638 0 : switch( task->task_type ) {
1639 0 : case FD_SCHED_TT_TXN_EXEC: {
1640 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
1641 :
1642 0 : fd_bank_t bank[1];
1643 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->txn_exec->bank_idx ) );
1644 :
1645 0 : # if FD_HAS_FLATCC
1646 : /* Add the transaction to the block dumper if necessary. This
1647 : logic doesn't need to be fork-aware since it's only meant to
1648 : be used in backtest. */
1649 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
1650 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
1651 0 : }
1652 0 : # endif
1653 :
1654 0 : bank->data->refcnt++;
1655 :
1656 0 : if( FD_UNLIKELY( !bank->data->first_transaction_scheduled_nanos ) ) bank->data->first_transaction_scheduled_nanos = fd_log_wallclock();
1657 :
1658 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1659 0 : fd_execrp_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1660 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1661 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
1662 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
1663 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) {
1664 0 : exec_msg->capture_txn_idx = ctx->capture_ctx->current_txn_idx++;
1665 0 : }
1666 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_EXEC<<32) | task->txn_exec->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1667 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1668 0 : break;
1669 0 : }
1670 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1671 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
1672 :
1673 0 : fd_bank_t bank[1];
1674 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->txn_sigverify->bank_idx ) );
1675 0 : bank->data->refcnt++;
1676 :
1677 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1678 0 : fd_execrp_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1679 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1680 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
1681 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
1682 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1683 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1684 0 : break;
1685 0 : };
1686 0 : case FD_SCHED_TT_POH_HASH: {
1687 0 : fd_bank_t bank[ 1 ];
1688 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->poh_hash->bank_idx ) );
1689 0 : bank->data->refcnt++;
1690 :
1691 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1692 0 : fd_execrp_poh_hash_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1693 0 : exec_msg->bank_idx = task->poh_hash->bank_idx;
1694 0 : exec_msg->mblk_idx = task->poh_hash->mblk_idx;
1695 0 : exec_msg->hashcnt = task->poh_hash->hashcnt;
1696 0 : memcpy( exec_msg->hash, task->poh_hash->hash, sizeof(fd_hash_t) );
1697 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_POH_HASH<<32) | task->poh_hash->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1698 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1699 0 : break;
1700 0 : };
1701 0 : default: {
1702 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1703 0 : }
1704 0 : }
1705 0 : }
1706 :
1707 : static void
1708 : mark_bank_dead( fd_replay_tile_t * ctx,
1709 : fd_stem_context_t * stem,
1710 0 : ulong bank_idx ) {
1711 0 : fd_bank_t bank[1];
1712 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
1713 0 : fd_banks_mark_bank_dead( ctx->banks, bank_idx );
1714 :
1715 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank_idx ];
1716 0 : if( block_id_ele->block_id_seen ) publish_slot_dead( ctx, stem, block_id_ele->slot, &block_id_ele->latest_mr );
1717 :
1718 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
1719 0 : if( FD_UNLIKELY( !fec ) ) return;
1720 0 : fec->bank_dead = 1;
1721 :
1722 0 : }
1723 :
1724 : /* Returns 1 if charge_busy. */
1725 : static int
1726 : replay( fd_replay_tile_t * ctx,
1727 0 : fd_stem_context_t * stem ) {
1728 :
1729 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
1730 :
1731 0 : int charge_busy = 0;
1732 0 : fd_sched_task_t task[ 1 ];
1733 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
1734 0 : return charge_busy; /* Nothing to execute or do. */
1735 0 : }
1736 :
1737 0 : charge_busy = 1;
1738 :
1739 0 : switch( task->task_type ) {
1740 0 : case FD_SCHED_TT_BLOCK_START: {
1741 0 : replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
1742 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX, NULL );
1743 0 : break;
1744 0 : }
1745 0 : case FD_SCHED_TT_BLOCK_END: {
1746 0 : fd_bank_t bank[1];
1747 0 : fd_banks_bank_query( bank, ctx->banks, task->block_end->bank_idx );
1748 0 : if( FD_LIKELY( !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) replay_block_finalize( ctx, stem, bank );
1749 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX, NULL );
1750 0 : break;
1751 0 : }
1752 0 : case FD_SCHED_TT_TXN_EXEC:
1753 0 : case FD_SCHED_TT_TXN_SIGVERIFY:
1754 0 : case FD_SCHED_TT_POH_HASH: {
1755 : /* Common case: we have a transaction we need to execute. */
1756 0 : dispatch_task( ctx, stem, task );
1757 0 : break;
1758 0 : }
1759 0 : case FD_SCHED_TT_MARK_DEAD: {
1760 0 : fd_bank_t bank[ 1 ];
1761 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->mark_dead->bank_idx ) );
1762 0 : mark_bank_dead( ctx, stem, task->mark_dead->bank_idx );
1763 0 : break;
1764 0 : }
1765 0 : default: {
1766 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1767 0 : }
1768 0 : }
1769 :
1770 0 : return charge_busy;
1771 0 : }
1772 :
1773 : static int
1774 : can_process_fec( fd_replay_tile_t * ctx,
1775 0 : int * evict_banks_out ) {
1776 0 : fd_reasm_fec_t * fec;
1777 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched )==0UL ) ) {
1778 0 : ctx->metrics.sched_full++;
1779 0 : return 0;
1780 0 : }
1781 :
1782 0 : if( FD_UNLIKELY( (fec = fd_reasm_peek( ctx->reasm ))==NULL ) ) {
1783 0 : ctx->metrics.reasm_empty++;
1784 0 : return 0;
1785 0 : }
1786 :
1787 0 : ctx->metrics.reasm_latest_slot = fec->slot;
1788 0 : ctx->metrics.reasm_latest_fec_idx = fec->fec_set_idx;
1789 :
1790 0 : if( FD_UNLIKELY( ctx->is_leader && fec->fec_set_idx==0U && fd_reasm_parent( ctx->reasm, fec )->bank_idx==ctx->leader_bank->data->idx ) ) {
1791 : /* This guards against a rare race where we receive the FEC set for
1792 : the slot right after our leader rotation before we freeze the
1793 : bank for the last slot in our leader rotation. Leader slot
1794 : freezing happens only after if we've received the final PoH hash
1795 : from the poh tile as well as the final FEC set for the leader
1796 : slot. So the race happens when FEC sets are delivered and
1797 : processed sooner than the PoH hash, aka when the
1798 : poh=>shred=>replay path for the block id beats the poh=>replay
1799 : path for the poh hash. To mitigate this race, we must block on
1800 : ingesting the FEC set for the ensuing slot before the leader
1801 : bank freezes, because that would violate ordering invariants in
1802 : banks and sched. */
1803 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen );
1804 0 : FD_TEST( !ctx->recv_poh );
1805 0 : ctx->metrics.leader_bid_wait++;
1806 0 : return 0;
1807 0 : }
1808 :
1809 : /* If fec_set_idx is 0, we need a new bank for a new slot. Banks must
1810 : not be full in this case. */
1811 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) && fec->fec_set_idx==0 ) ) {
1812 0 : ctx->metrics.banks_full++;
1813 : /* We only want to evict banks if sched is drained and banks is no
1814 : longer making progress. Otherwise, sched might not release
1815 : refcnts on the frontier/leaf banks immediately, and the eviction
1816 : will have to wait for sched to drain anyways. */
1817 0 : if( FD_UNLIKELY( fd_sched_is_drained( ctx->sched ) ) ) *evict_banks_out = 1;
1818 0 : return 0;
1819 0 : }
1820 :
1821 : /* Otherwise, banks may not be full, so we can always create a new
1822 : bank if needed. Or, if banks are full, the current fec set's
1823 : ancestor (idx 0) already created a bank for this slot.*/
1824 0 : return 1;
1825 0 : }
1826 :
1827 : /* Returns 0 on successful FEC ingestion, 1 if the block got marked
1828 : dead. */
1829 : static int
1830 : insert_fec_set( fd_replay_tile_t * ctx,
1831 : fd_stem_context_t * stem,
1832 0 : fd_reasm_fec_t * reasm_fec ) {
1833 :
1834 0 : long now = fd_log_wallclock();
1835 :
1836 0 : reasm_fec->parent_bank_idx = fd_reasm_parent( ctx->reasm, reasm_fec )->bank_idx;
1837 :
1838 0 : fd_bank_t parent_bank[1];
1839 0 : FD_TEST( fd_banks_bank_query( parent_bank, ctx->banks, reasm_fec->parent_bank_idx ) );
1840 0 : reasm_fec->parent_bank_seq = parent_bank->data->bank_seq;
1841 :
1842 0 : if( FD_UNLIKELY( reasm_fec->fec_set_idx==0U ) ) {
1843 : /* If the first FEC set for a slot is observed, provision a new bank
1844 : if you are not the leader. Remove any stale block id map entry
1845 : and update the block id entry. */
1846 0 : fd_bank_t bank_[1];
1847 0 : fd_bank_t * bank = NULL;
1848 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) {
1849 0 : bank = ctx->leader_bank;
1850 0 : } else {
1851 0 : bank = fd_banks_new_bank( bank_, ctx->banks, reasm_fec->parent_bank_idx, now );
1852 0 : }
1853 :
1854 0 : reasm_fec->bank_idx = bank->data->idx;
1855 0 : reasm_fec->bank_seq = bank->data->bank_seq;
1856 :
1857 : /* At this point remove any stale entry in the block id map if it
1858 : exists and set the block id as not having been seen yet. This is
1859 : safe because we know that the old entry for this bank index has
1860 : already been pruned away. */
1861 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1862 0 : if( FD_LIKELY( fd_block_id_map_ele_query( ctx->block_id_map, &block_id_ele->latest_mr, NULL, ctx->block_id_arr )==block_id_ele ) ) {
1863 0 : FD_TEST( fd_block_id_map_ele_remove( ctx->block_id_map, &block_id_ele->latest_mr, NULL, ctx->block_id_arr ) );
1864 0 : }
1865 0 : block_id_ele->block_id_seen = 0;
1866 0 : block_id_ele->slot = reasm_fec->slot;
1867 0 : block_id_ele->latest_fec_idx = 0U;
1868 0 : block_id_ele->latest_mr = reasm_fec->key;
1869 0 : } else {
1870 : /* We are continuing to execute through a slot that we already have
1871 : a bank index for. */
1872 0 : reasm_fec->bank_idx = reasm_fec->parent_bank_idx;
1873 0 : reasm_fec->bank_seq = reasm_fec->parent_bank_seq;
1874 :
1875 0 : FD_TEST( reasm_fec->bank_idx!=ULONG_MAX );
1876 :
1877 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1878 0 : if( FD_UNLIKELY( block_id_ele->latest_fec_idx>=reasm_fec->fec_set_idx ) ) {
1879 0 : FD_LOG_WARNING(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because it is at least as old as the latest FEC set (slot=%lu, fec_set_idx=%u)", reasm_fec->slot, reasm_fec->fec_set_idx, block_id_ele->slot, block_id_ele->latest_fec_idx ));
1880 0 : return 0;
1881 0 : }
1882 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1883 0 : block_id_ele->latest_mr = reasm_fec->key;
1884 0 : }
1885 :
1886 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) {
1887 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1888 0 : block_id_ele->block_id_seen = 1;
1889 0 : block_id_ele->latest_mr = reasm_fec->key;
1890 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1891 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1892 0 : }
1893 :
1894 : /* If we are the leader, we don't need to process the FEC set. */
1895 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) return 0;
1896 :
1897 : /* Forks form a partial ordering over FEC sets. The Repair tile
1898 : delivers FEC sets in-order per fork, but FEC set ordering across
1899 : forks is arbitrary */
1900 0 : fd_sched_fec_t sched_fec[ 1 ];
1901 :
1902 : # if DEBUG_LOGGING
1903 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
1904 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->cmr.key, cmr_b58 );
1905 : 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, key_b58, cmr_b58 ));
1906 : # endif
1907 :
1908 0 : sched_fec->shred_cnt = reasm_fec->data_cnt;
1909 0 : sched_fec->is_last_in_batch = !!reasm_fec->data_complete;
1910 0 : sched_fec->is_last_in_block = !!reasm_fec->slot_complete;
1911 0 : sched_fec->bank_idx = reasm_fec->bank_idx;
1912 0 : sched_fec->parent_bank_idx = reasm_fec->parent_bank_idx;
1913 0 : sched_fec->slot = reasm_fec->slot;
1914 0 : sched_fec->parent_slot = reasm_fec->slot - reasm_fec->parent_off;
1915 0 : sched_fec->is_first_in_block = reasm_fec->fec_set_idx==0U;
1916 0 : fd_funk_txn_xid_t const root = fd_accdb_root_get( ctx->accdb_admin );
1917 0 : fd_funk_txn_xid_copy( sched_fec->alut_ctx->xid, &root );
1918 0 : sched_fec->alut_ctx->accdb[0] = ctx->accdb[0];
1919 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
1920 :
1921 0 : fd_bank_t bank[1];
1922 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, sched_fec->bank_idx ) );
1923 0 : if( sched_fec->is_first_in_block ) {
1924 0 : bank->data->refcnt++;
1925 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for sched", bank->data->idx, sched_fec->slot, bank->data->refcnt ));
1926 0 : }
1927 :
1928 : /* Read FEC set from the store. This should happen before we try to
1929 : ingest the FEC set. This allows us to filter out frags that were
1930 : in-flight when we published away minority forks that the frags land
1931 : on. These frags would have no bank to execute against, because
1932 : their corresponding banks, or parent banks, have also been pruned
1933 : during publishing. A query against store will rightfully tell us
1934 : that the underlying data is not found, implying that this is for a
1935 : minority fork that we can safeljy ignore. */
1936 :
1937 0 : ulong wait = (ulong)fd_log_wallclock();
1938 0 : ulong work = wait;
1939 0 : FD_STORE_SLOCK_BEGIN( ctx->store ) {
1940 0 : ctx->metrics.store_query_acquire++;
1941 0 : work = (ulong)fd_log_wallclock();
1942 0 : fd_histf_sample( ctx->metrics.store_query_wait, work - wait );
1943 :
1944 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->store, &reasm_fec->key );
1945 0 : ctx->metrics.store_query_cnt++;
1946 0 : if( FD_UNLIKELY( !store_fec ) ) {
1947 :
1948 : /* The only case in which a FEC is not found in the store after
1949 : repair has notified is if the FEC was on a minority fork that
1950 : has already been published away. In this case we abandon the
1951 : entire slice because it is no longer relevant. */
1952 :
1953 0 : ctx->metrics.store_query_missing_cnt++;
1954 0 : ctx->metrics.store_query_missing_mr = reasm_fec->key.ul[0];
1955 0 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
1956 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, key_b58 ));
1957 0 : return 0;
1958 0 : }
1959 0 : sched_fec->fec = store_fec;
1960 0 : if( FD_UNLIKELY( !fd_sched_fec_ingest( ctx->sched, sched_fec ) ) ) { /* FIXME this critical section is unnecessarily complex. should refactor to just be held for the memcpy and block_offs. */
1961 0 : mark_bank_dead( ctx, stem, sched_fec->bank_idx );
1962 0 : return 1;
1963 0 : }
1964 0 : } FD_STORE_SLOCK_END;
1965 :
1966 0 : ctx->metrics.store_query_release++;
1967 0 : fd_histf_sample( ctx->metrics.store_query_work, (ulong)fd_log_wallclock() - work );
1968 0 : return 0;
1969 0 : }
1970 :
1971 : static void
1972 : process_fec_set( fd_replay_tile_t * ctx,
1973 : fd_stem_context_t * stem,
1974 0 : fd_reasm_fec_t * reasm_fec ) {
1975 :
1976 0 : fd_reasm_fec_t * parent = fd_reasm_parent( ctx->reasm, reasm_fec );
1977 0 : if( FD_UNLIKELY( !parent ) ) {
1978 0 : FD_LOG_WARNING(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because it is unconnected in reasm", reasm_fec->slot, reasm_fec->fec_set_idx ));
1979 0 : return;
1980 0 : }
1981 :
1982 0 : if( FD_UNLIKELY( parent->bank_dead ) ) {
1983 : /* Inherit the dead flag from the parent. If a dead slot is
1984 : completed, we publish the slot as dead. Don't insert FECs for
1985 : dead slots. */
1986 0 : reasm_fec->bank_dead = 1;
1987 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) publish_slot_dead( ctx, stem, reasm_fec->slot, &reasm_fec->key );
1988 0 : FD_LOG_DEBUG(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because parent bank is marked dead", reasm_fec->slot, reasm_fec->fec_set_idx ));
1989 0 : return;
1990 0 : }
1991 :
1992 : /* Standard case, the parent FEC has a valid corresponding bank. */
1993 0 : fd_bank_t parent_fec_bank[1];
1994 0 : if( FD_LIKELY( fd_banks_bank_query( parent_fec_bank, ctx->banks, parent->bank_idx ) &&
1995 0 : parent_fec_bank->data->bank_seq==parent->bank_seq ) ) {
1996 0 : insert_fec_set( ctx, stem, reasm_fec );
1997 0 : return;
1998 0 : }
1999 :
2000 : /* In the case the FEC doesn't directly connect, iterate up the reasm
2001 : tree to find the closest valid slot complete that corresponds to a
2002 : valid bank. */
2003 :
2004 : /* First keep track of all of the slot completes up to and including
2005 : the fec we want to insert off of. */
2006 0 : fd_reasm_fec_t * path[ FD_BANKS_MAX_BANKS ];
2007 0 : ulong path_cnt = 0UL;
2008 0 : path[ path_cnt++ ] = reasm_fec;
2009 :
2010 0 : for( fd_reasm_fec_t * curr = reasm_fec;; ) {
2011 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2012 0 : FD_TEST( curr );
2013 0 : if( FD_LIKELY( !curr->slot_complete ) ) continue;
2014 :
2015 0 : fd_bank_t curr_bank[1];
2016 0 : if( FD_LIKELY( fd_banks_bank_query( curr_bank, ctx->banks, curr->bank_idx ) && curr_bank->data->bank_seq==curr->bank_seq ) ) break;
2017 :
2018 0 : FD_TEST( path_cnt<FD_BANKS_MAX_BANKS );
2019 0 : path[ path_cnt++ ] = curr;
2020 0 : }
2021 :
2022 0 : for( ulong i=path_cnt; i>0UL; i-- ) {
2023 0 : fd_reasm_fec_t * leaf = path[ i-1 ];
2024 :
2025 : /* If there's not capacity in the sched or banks, return early and
2026 : drop the FEC. We have inserted as much as we can for now. */
2027 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched ) < (leaf->fec_set_idx/FD_FEC_SHRED_CNT + 1) ) ) return;
2028 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) ) ) return;
2029 :
2030 : /* Gather all FECs for this slot; */
2031 0 : fd_reasm_fec_t * slot_fecs[ FD_FEC_BLK_MAX ];
2032 0 : fd_reasm_fec_t * curr = leaf;
2033 0 : for(;;) {
2034 0 : slot_fecs[ curr->fec_set_idx/FD_FEC_SHRED_CNT ] = curr;
2035 0 : if( curr->fec_set_idx==0U ) break;
2036 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2037 0 : FD_TEST( curr );
2038 0 : }
2039 0 : FD_LOG_NOTICE(( "backfilling FEC sets for slot %lu from fec_set_idx %u to fec_set_idx %u", leaf->slot, leaf->fec_set_idx, curr->fec_set_idx ));
2040 :
2041 0 : for( ulong j=0UL; j<=leaf->fec_set_idx/FD_FEC_SHRED_CNT; j++ ) {
2042 0 : if( FD_UNLIKELY( insert_fec_set( ctx, stem, slot_fecs[ j ] ) ) ) return;
2043 0 : }
2044 0 : }
2045 0 : }
2046 :
2047 : /* accdb_advance_root moves account records from the unrooted to the
2048 : rooted database. */
2049 :
2050 : static inline ulong
2051 0 : accdb_root_op_total( fd_replay_tile_t const * ctx ) {
2052 0 : return ctx->accdb_admin->base.root_cnt +
2053 0 : ctx->accdb_admin->base.reclaim_cnt;
2054 0 : }
2055 :
2056 : static void
2057 : accdb_advance_root( fd_replay_tile_t * ctx,
2058 : ulong slot,
2059 0 : ulong bank_idx ) {
2060 0 : fd_funk_txn_xid_t xid = { .ul[0] = slot, .ul[1] = bank_idx };
2061 0 : FD_LOG_DEBUG(( "advancing root to slot=%lu", slot ));
2062 :
2063 0 : long rooted_accounts = -(long)accdb_root_op_total( ctx );
2064 0 : long t0 = fd_tickcount();
2065 0 : fd_accdb_advance_root( ctx->accdb_admin, &xid );
2066 0 : rooted_accounts += (long)accdb_root_op_total( ctx );
2067 0 : long t1 = fd_tickcount();
2068 0 : long root_accounts_dt = t1 - t0;
2069 0 : fd_histf_sample( ctx->metrics.root_slot_dur, (ulong)root_accounts_dt );
2070 0 : fd_histf_sample( ctx->metrics.root_account_dur, (ulong)root_accounts_dt / (ulong)fd_long_max( rooted_accounts, 1L ) );
2071 :
2072 0 : fd_progcache_advance_root( ctx->progcache, &xid );
2073 0 : long t2 = fd_tickcount();
2074 0 : FD_MCNT_INC( REPLAY, PROGCACHE_TIME_SECONDS, (ulong)( t2-t1 ) );
2075 0 : }
2076 :
2077 : static int
2078 0 : advance_published_root( fd_replay_tile_t * ctx ) {
2079 :
2080 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 );
2081 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2082 0 : FD_BASE58_ENCODE_32_BYTES( ctx->consensus_root.key, consensus_root_b58 );
2083 0 : FD_LOG_CRIT(( "invariant violation: block id ele not found for consensus root %s", consensus_root_b58 ));
2084 0 : }
2085 0 : ulong target_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2086 :
2087 : /* If the identity vote has been seen on a bank that should be rooted,
2088 : then we are now ready to produce blocks. */
2089 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2090 0 : fd_bank_t root_bank[1];
2091 0 : if( FD_UNLIKELY( !fd_banks_bank_query( root_bank, ctx->banks, target_bank_idx ) ) ) FD_LOG_CRIT(( "invariant violation: root bank not found for bank index %lu", target_bank_idx ));
2092 0 : if( root_bank->data->f.identity_vote_idx==ctx->identity_idx ) ctx->identity_vote_rooted = 1;
2093 0 : }
2094 :
2095 0 : ulong advanceable_root_idx = ULONG_MAX;
2096 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank_idx, &advanceable_root_idx ) ) ) {
2097 0 : ctx->metrics.storage_root_behind++;
2098 0 : return 0;
2099 0 : }
2100 :
2101 0 : fd_bank_t bank[1];
2102 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, advanceable_root_idx ) );
2103 :
2104 0 : if( FD_UNLIKELY( advanceable_root_idx >= ctx->block_id_len ) ) {
2105 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele out of bounds [0, %lu) index %lu", ctx->block_id_len, advanceable_root_idx ));
2106 0 : }
2107 0 : fd_block_id_ele_t * advanceable_root_ele = &ctx->block_id_arr[ advanceable_root_idx ];
2108 :
2109 0 : ulong advanceable_root_slot = bank->data->f.slot;
2110 0 : accdb_advance_root( ctx, advanceable_root_slot, bank->data->idx );
2111 :
2112 0 : fd_txncache_advance_root( ctx->txncache, bank->data->txncache_fork_id );
2113 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
2114 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
2115 :
2116 : /* Reasm also prunes from the store during its publish. */
2117 :
2118 0 : fd_reasm_publish( ctx->reasm, &advanceable_root_ele->latest_mr, ctx->store );
2119 :
2120 0 : ctx->published_root_slot = advanceable_root_slot;
2121 0 : ctx->published_root_bank_idx = advanceable_root_idx;
2122 :
2123 0 : return 1;
2124 0 : }
2125 :
2126 : static void
2127 : after_credit( fd_replay_tile_t * ctx,
2128 : fd_stem_context_t * stem,
2129 : int * opt_poll_in,
2130 0 : int * charge_busy ) {
2131 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->wfs_complete ) ) return;
2132 :
2133 0 : if( FD_UNLIKELY( maybe_become_leader( ctx, stem ) ) ) {
2134 0 : *charge_busy = 1;
2135 0 : *opt_poll_in = 0;
2136 0 : return;
2137 0 : }
2138 :
2139 : /* If we are leader, we can only unbecome the leader iff we have
2140 : received the poh hash from the poh tile and block id from reasm.
2141 : We have to do an additional check against the slot of the leader
2142 : bank because we lazily remove entries from the block id arr. */
2143 0 : if( FD_UNLIKELY( ctx->is_leader &&
2144 0 : ctx->recv_poh &&
2145 0 : ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen &&
2146 0 : ctx->block_id_arr[ ctx->leader_bank->data->idx ].slot==ctx->leader_bank->data->f.slot ) ) {
2147 :
2148 0 : fini_leader_bank( ctx, stem );
2149 0 : *charge_busy = 1;
2150 0 : *opt_poll_in = 0;
2151 0 : return;
2152 0 : }
2153 :
2154 0 : ulong bank_idx;
2155 0 : while( (bank_idx=fd_sched_pruned_block_next( ctx->sched ))!=ULONG_MAX ) {
2156 0 : fd_bank_t bank[1];
2157 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
2158 0 : bank->data->refcnt--;
2159 0 : FD_LOG_DEBUG(( "bank (idx=%lu) refcnt decremented to %lu for sched", bank->data->idx, bank->data->refcnt ));
2160 0 : }
2161 :
2162 : /* If the published_root is not caught up to the consensus root, then
2163 : we should try to advance the published root. */
2164 0 : if( FD_UNLIKELY( ctx->consensus_root_bank_idx!=ctx->published_root_bank_idx && advance_published_root( ctx ) ) ) {
2165 0 : *charge_busy = 1;
2166 0 : *opt_poll_in = 0;
2167 0 : return;
2168 0 : }
2169 :
2170 0 : fd_banks_prune_cancel_info_t cancel_info[ 1 ];
2171 0 : int pruned = fd_banks_prune_one_dead_bank( ctx->banks, cancel_info );
2172 0 : if( FD_UNLIKELY( pruned==2 ) ) {
2173 0 : fd_txncache_cancel_fork( ctx->txncache, cancel_info->txncache_fork_id );
2174 0 : fd_funk_txn_xid_t xid = { .ul = { cancel_info->slot, cancel_info->bank_idx } };
2175 0 : fd_accdb_cancel ( ctx->accdb_admin, &xid );
2176 0 : fd_progcache_cancel( ctx->progcache, &xid );
2177 0 : *charge_busy = 1;
2178 0 : *opt_poll_in = 0;
2179 0 : return;
2180 0 : }
2181 :
2182 : /* if reasm evicted is set, publish starting from reasm_evicted down
2183 : to the leaf node to repair so repair can re-request for it */
2184 :
2185 0 : if( FD_UNLIKELY( ctx->reasm_evicted ) ) {
2186 0 : fd_replay_fec_evicted_t evicted = (fd_replay_fec_evicted_t){ .mr = ctx->reasm_evicted->key, .slot = ctx->reasm_evicted->slot, .fec_set_idx = ctx->reasm_evicted->fec_set_idx, .bank_idx = ctx->reasm_evicted->bank_idx };
2187 0 : fd_memcpy( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ), &evicted, sizeof(fd_replay_fec_evicted_t) );
2188 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_REASM_EVICTED, ctx->replay_out->chunk, sizeof(fd_replay_fec_evicted_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2189 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_fec_evicted_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2190 :
2191 : /* eviction policy only evicts chains of nodes until there is a
2192 : fork, so guaranteed that the evict path is always the left-child */
2193 0 : fd_reasm_pool_release( ctx->reasm, ctx->reasm_evicted );
2194 0 : ctx->reasm_evicted = fd_reasm_child( ctx->reasm, ctx->reasm_evicted ); /* indexes into pool, safe to use */
2195 :
2196 0 : *charge_busy = 1;
2197 0 : *opt_poll_in = 0;
2198 0 : return;
2199 0 : }
2200 :
2201 : /* Mark a frontier eviction victim bank as dead. As refcnts on said
2202 : banks are drained, they will be pruned away. */
2203 0 : if( FD_UNLIKELY( ctx->frontier_cnt ) ) {
2204 0 : *charge_busy = 1;
2205 0 : *opt_poll_in = 0;
2206 0 : bank_idx = ctx->frontier_indices[ --ctx->frontier_cnt ];
2207 0 : fd_bank_t bank[1];
2208 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
2209 0 : if( FD_UNLIKELY( ctx->is_leader && bank_idx==ctx->leader_bank->data->idx ) ) return;
2210 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2211 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2212 :
2213 : /* evict it from reasm */
2214 :
2215 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->data->idx ];
2216 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
2217 0 : FD_TEST( fec );
2218 0 : fd_reasm_fec_t * evicted_head = fd_reasm_remove( ctx->reasm, fec, ctx->store );
2219 0 : if( FD_UNLIKELY( ctx->reasm_evicted ) ) {
2220 : /* already have a chain we are evicting. Prepend the new chain to the existing chain */
2221 0 : fec->child = fd_reasm_pool_idx( ctx->reasm, ctx->reasm_evicted );
2222 0 : }
2223 0 : ctx->reasm_evicted = evicted_head;
2224 0 : return;
2225 0 : }
2226 :
2227 : /* If the reassembler has a fec that is ready, we should process it
2228 : and pass it to the scheduler. */
2229 0 : int evict_banks = 0;
2230 0 : if( FD_LIKELY( can_process_fec( ctx, &evict_banks ) ) ) {
2231 0 : fd_reasm_fec_t * fec = fd_reasm_pop( ctx->reasm );
2232 0 : process_fec_set( ctx, stem, fec );
2233 0 : *charge_busy = 1;
2234 0 : *opt_poll_in = 0;
2235 0 : return;
2236 0 : }
2237 :
2238 0 : if( FD_UNLIKELY( evict_banks ) ) {
2239 0 : FD_LOG_WARNING(( "banks are full and partially executed frontier banks are being evicted" ));
2240 0 : fd_banks_get_frontier( ctx->banks, ctx->frontier_indices, &ctx->frontier_cnt );
2241 0 : *charge_busy = 1;
2242 0 : *opt_poll_in = 0;
2243 0 : return;
2244 0 : }
2245 :
2246 0 : *charge_busy = replay( ctx, stem );
2247 0 : *opt_poll_in = !*charge_busy;
2248 0 : }
2249 :
2250 : static int
2251 : before_frag( fd_replay_tile_t * ctx,
2252 : ulong in_idx,
2253 : ulong seq FD_PARAM_UNUSED,
2254 0 : ulong sig ) {
2255 :
2256 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_GOSSIP_OUT && sig!=FD_GOSSIP_UPDATE_TAG_WFS_DONE ) ) return 1;
2257 0 : return 0;
2258 0 : }
2259 :
2260 : static void
2261 : process_exec_task_done( fd_replay_tile_t * ctx,
2262 : fd_stem_context_t * stem,
2263 : fd_execrp_task_done_msg_t * msg,
2264 0 : ulong sig ) {
2265 :
2266 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
2267 :
2268 0 : fd_bank_t bank[1];
2269 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, msg->bank_idx ) );
2270 0 : FD_TEST( bank->data );
2271 0 : bank->data->refcnt--;
2272 :
2273 0 : switch( sig>>32 ) {
2274 0 : case FD_EXECRP_TT_TXN_EXEC: {
2275 0 : ulong txn_idx = msg->txn_exec->txn_idx;
2276 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2277 : /* Query the txn signature against our recently generated vote
2278 : txn signatures. If the query is successful, then we have
2279 : seen our own vote transaction land and this should be marked
2280 : in the bank. We go through this exercise until we've seen
2281 : our vote rooted. */
2282 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, txn_idx );
2283 :
2284 0 : fd_pubkey_t * identity_pubkey_out = NULL;
2285 0 : if( fd_vote_tracker_query_sig( ctx->vote_tracker, fd_type_pun_const( txn_p->payload+TXN( txn_p )->signature_off ), &identity_pubkey_out ) && fd_pubkey_eq( identity_pubkey_out, ctx->identity_pubkey ) ) {
2286 0 : bank->data->f.identity_vote_idx = ctx->identity_idx;
2287 0 : }
2288 0 : }
2289 0 : if( FD_UNLIKELY( !msg->txn_exec->is_committable && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2290 : /* Every transaction in a valid block has to execute.
2291 : Otherwise, we should mark the block as dead. */
2292 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2293 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2294 0 : }
2295 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2296 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2297 0 : }
2298 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, txn_idx, exec_tile_idx, NULL );
2299 0 : FD_TEST( res==0 );
2300 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
2301 0 : txn_info->flags |= FD_SCHED_TXN_EXEC_DONE;
2302 0 : if( FD_LIKELY( !(txn_info->flags&FD_SCHED_TXN_SIGVERIFY_DONE)||!txn_info->txn_err ) ) { /* Set execution status if sigverify hasn't happened yet or if sigverify was a success. */
2303 0 : txn_info->txn_err = msg->txn_exec->txn_err;
2304 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_committable, FD_SCHED_TXN_IS_COMMITTABLE, 0UL );
2305 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_fees_only, FD_SCHED_TXN_IS_FEES_ONLY, 0UL );
2306 0 : }
2307 0 : if( FD_UNLIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) { /* UNLIKELY because generally exec happens before sigverify. */
2308 0 : publish_txn_executed( ctx, stem, txn_idx );
2309 0 : }
2310 0 : break;
2311 0 : }
2312 0 : case FD_EXECRP_TT_TXN_SIGVERIFY: {
2313 0 : ulong txn_idx = msg->txn_sigverify->txn_idx;
2314 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
2315 0 : txn_info->flags |= FD_SCHED_TXN_SIGVERIFY_DONE;
2316 0 : if( FD_UNLIKELY( msg->txn_sigverify->err ) ) {
2317 0 : txn_info->txn_err = FD_RUNTIME_TXN_ERR_SIGNATURE_FAILURE;
2318 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_COMMITTABLE;
2319 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_FEES_ONLY;
2320 0 : }
2321 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2322 : /* Every transaction in a valid block has to sigverify.
2323 : Otherwise, we should mark the block as dead. Also freeze the
2324 : bank if possible. */
2325 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2326 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2327 0 : }
2328 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2329 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2330 0 : }
2331 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, txn_idx, exec_tile_idx, NULL );
2332 0 : FD_TEST( res==0 );
2333 0 : if( FD_LIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) {
2334 0 : publish_txn_executed( ctx, stem, txn_idx );
2335 0 : }
2336 0 : break;
2337 0 : }
2338 0 : case FD_EXECRP_TT_POH_HASH: {
2339 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_POH_HASH, ULONG_MAX, exec_tile_idx, msg->poh_hash );
2340 0 : if( FD_UNLIKELY( res<0 && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2341 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2342 0 : }
2343 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2344 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2345 0 : }
2346 0 : break;
2347 0 : }
2348 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
2349 0 : }
2350 :
2351 : /* Reference counter just decreased, and an exec tile just got freed
2352 : up. If there's a need to be more aggressively pruning, we could
2353 : check here if more slots just became publishable and publish. Not
2354 : publishing here shouldn't bloat the fork tree too much though. We
2355 : mark minority forks dead as soon as we can, and execution dispatch
2356 : stops on dead blocks. So shortly afterwards, dead blocks should be
2357 : eligible for pruning as in-flight transactions retire from the
2358 : execution pipeline. */
2359 :
2360 0 : }
2361 :
2362 : static void
2363 : process_tower_slot_done( fd_replay_tile_t * ctx,
2364 : fd_stem_context_t * stem,
2365 : fd_tower_slot_done_t const * msg,
2366 0 : ulong seq ) {
2367 0 : fd_bank_t replay_bank[1];
2368 0 : if( FD_UNLIKELY( !fd_banks_bank_query( replay_bank, ctx->banks, msg->replay_bank_idx ) ) ) FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", msg->replay_bank_idx ));
2369 0 : replay_bank->data->refcnt--;
2370 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for tower", replay_bank->data->idx, msg->replay_slot, replay_bank->data->refcnt ));
2371 :
2372 0 : ctx->reset_block_id = msg->reset_block_id;
2373 0 : ctx->reset_slot = msg->reset_slot;
2374 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
2375 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 ) );
2376 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
2377 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
2378 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
2379 0 : } else {
2380 0 : ctx->next_leader_tickcount = LONG_MAX;
2381 0 : }
2382 :
2383 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 );
2384 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2385 0 : FD_BASE58_ENCODE_32_BYTES( msg->reset_block_id.key, reset_block_id_b58 );
2386 0 : FD_LOG_CRIT(( "invariant violation: block id ele doesn't exist for reset block id: %s, slot: %lu", reset_block_id_b58, msg->reset_slot ));
2387 0 : }
2388 0 : ulong reset_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2389 :
2390 0 : fd_bank_t bank[1];
2391 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, reset_bank_idx ) ) ) {
2392 0 : FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", reset_bank_idx ));
2393 0 : }
2394 :
2395 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) FD_TEST( msg->root_slot<=msg->reset_slot );
2396 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
2397 :
2398 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2399 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2400 :
2401 0 : reset->bank_idx = bank->data->idx;
2402 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2403 0 : reset->completed_slot = ctx->reset_slot;
2404 0 : reset->hashcnt_per_tick = bank->data->f.hashes_per_tick;
2405 0 : reset->ticks_per_slot = bank->data->f.ticks_per_slot;
2406 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
2407 :
2408 0 : fd_memcpy( reset->completed_block_id, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
2409 :
2410 0 : fd_blockhashes_t const * block_hash_queue = &bank->data->f.block_hash_queue;
2411 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2412 0 : FD_TEST( last_hash );
2413 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2414 :
2415 0 : ulong ticks_per_slot = bank->data->f.ticks_per_slot;
2416 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
2417 : /* Low power producer, maximum of one microblock per tick in the slot */
2418 0 : reset->max_microblocks_in_slot = ticks_per_slot;
2419 0 : } else {
2420 : /* See the long comment in after_credit for this limit */
2421 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
2422 0 : }
2423 0 : reset->next_leader_slot = ctx->next_leader_slot;
2424 :
2425 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() ) );
2426 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 );
2427 0 : }
2428 :
2429 0 : FD_LOG_INFO(( "tower_slot_done(reset_slot=%lu, next_leader_slot=%lu, vote_slot=%lu, replay_slot=%lu, root_slot=%lu, seqno=%lu)", msg->reset_slot, ctx->next_leader_slot, msg->vote_slot, msg->replay_slot, msg->root_slot, seq ));
2430 0 : maybe_become_leader( ctx, stem );
2431 :
2432 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) {
2433 :
2434 0 : FD_TEST( msg->root_slot>=ctx->consensus_root_slot );
2435 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 );
2436 0 : FD_TEST( block_id_ele );
2437 0 : ctx->consensus_root_slot = msg->root_slot;
2438 0 : ctx->consensus_root = msg->root_block_id;
2439 0 : ctx->consensus_root_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2440 :
2441 0 : publish_root_advanced( ctx, stem );
2442 :
2443 0 : fd_sched_root_notify( ctx->sched, ctx->consensus_root_bank_idx );
2444 0 : }
2445 :
2446 0 : ulong distance = 0UL;
2447 0 : fd_bank_t * parent = bank;
2448 0 : while( parent ) {
2449 0 : if( FD_UNLIKELY( parent->data->idx==ctx->consensus_root_bank_idx ) ) break;
2450 0 : parent = fd_banks_get_parent( bank, ctx->banks, parent );
2451 0 : distance++;
2452 0 : }
2453 :
2454 0 : FD_MGAUGE_SET( REPLAY, ROOT_DISTANCE, distance );
2455 0 : }
2456 :
2457 : static void
2458 : process_fec_complete( fd_replay_tile_t * ctx,
2459 : fd_stem_context_t * stem,
2460 0 : uchar const * shred_buf ) {
2461 0 : fd_shred_t const * shred = (fd_shred_t const *)fd_type_pun_const( shred_buf );
2462 :
2463 0 : fd_hash_t const * merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ );
2464 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) );
2465 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) );
2466 :
2467 0 : int data_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_DATA_COMPLETE );
2468 0 : int slot_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_SLOT_COMPLETE );
2469 :
2470 0 : if( FD_UNLIKELY( shred->slot - shred->data.parent_off == fd_reasm_slot0( ctx->reasm ) && shred->fec_set_idx == 0) ) {
2471 0 : chained_merkle_root = &fd_reasm_root( ctx->reasm )->key;
2472 0 : }
2473 :
2474 0 : if( FD_UNLIKELY( fd_reasm_query( ctx->reasm, merkle_root ) ) ) return;
2475 0 : fd_reasm_fec_t * fec = 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, ctx->store, &ctx->reasm_evicted );
2476 :
2477 0 : if( FD_UNLIKELY( !fec ) ) {
2478 : /* reasm failed to insert. We don't want to just put this back on
2479 : the returnable_frag queue because it's unclear whether this FEC
2480 : is truly something we want to process. Therefore our best option
2481 : is to punt it and "go around." reasm_insert populates it's last
2482 : pool element with the data of the failed insert, so we make sure
2483 : to publish the failed insert data to repair in after_credit. */
2484 0 : return;
2485 0 : }
2486 :
2487 0 : if( FD_UNLIKELY( ctx->reasm_evicted && ctx->reasm_evicted->bank_idx != ULONG_MAX ) ) {
2488 0 : mark_bank_dead( ctx, stem, ctx->reasm_evicted->bank_idx );
2489 0 : fd_sched_block_abandon( ctx->sched, ctx->reasm_evicted->bank_idx );
2490 0 : }
2491 0 : }
2492 :
2493 : static void
2494 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
2495 0 : fd_bank_t bank[1];
2496 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
2497 0 : bank->data->refcnt--;
2498 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for resolv", bank->data->idx, bank->data->f.slot, bank->data->refcnt ));
2499 0 : }
2500 :
2501 : static void
2502 : process_vote_txn_sent( fd_replay_tile_t * ctx,
2503 0 : fd_txn_m_t * txnm ) {
2504 : /* The send tile has signed and sent a vote. Add this vote to the
2505 : vote tracker. We go through this exercise until the client has
2506 : seen a vote corresponding to the current identity rooted. */
2507 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2508 0 : uchar * payload = (uchar *)txnm + sizeof(fd_txn_m_t);
2509 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
2510 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
2511 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
2512 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
2513 0 : }
2514 : /* The identity of the validator that the signed the vote will
2515 : always be the first signer in the vote transaction. */
2516 0 : fd_pubkey_t * vote_identity = fd_type_pun( payload+txn->acct_addr_off );
2517 0 : fd_vote_tracker_insert( ctx->vote_tracker, vote_identity, fd_type_pun_const( payload+txn->signature_off ) );
2518 0 : }
2519 0 : }
2520 :
2521 : static inline void
2522 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
2523 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
2524 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
2525 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
2526 0 : }
2527 0 : }
2528 :
2529 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_forks_cnt!=ULONG_MAX && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
2530 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
2531 :
2532 0 : ushort actual_shred_version = compute_shred_version( ctx->genesis_hash->uc, ctx->hard_forks, ctx->hard_forks_cnts, ctx->hard_forks_cnt );
2533 :
2534 0 : if( FD_UNLIKELY( expected_shred_version!=actual_shred_version ) ) {
2535 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, genesis_hash_b58 );
2536 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` combined with the hard_forks from the loaded snapshot have produced "
2537 0 : "a shred version of %hu but the entrypoint you connected to on boot reported a shred version of %hu. "
2538 0 : "This likely means that the genesis.bin file you have is for a different cluster than the one you "
2539 0 : "are trying to connect to, you can delete it and restart the node to download the correct genesis "
2540 0 : "file automatically.", ctx->genesis_path, actual_shred_version, expected_shred_version ));
2541 0 : }
2542 0 : }
2543 0 : }
2544 :
2545 : static inline void
2546 0 : maybe_verify_genesis_timestamp( fd_replay_tile_t * ctx ) {
2547 0 : if( FD_LIKELY( !ctx->has_expected_genesis_timestamp || !ctx->has_genesis_timestamp ) ) return;
2548 0 : if( FD_LIKELY( ctx->genesis_timestamp==ctx->expected_genesis_timestamp ) ) return;
2549 :
2550 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis timestamp of %lu but the snapshot you loaded has a genesis "
2551 0 : "timestamp of %lu. This either means that the genesis.bin file you have is for a different cluster than "
2552 0 : "the one you are trying to connect to, or you have loaded a snapshot for the wrong cluster. In either "
2553 0 : "case, you can delete the problematic file and restart the node to download the correct one automatically.",
2554 0 : ctx->genesis_path, ctx->genesis_timestamp, ctx->expected_genesis_timestamp ));
2555 0 : }
2556 :
2557 : static void
2558 : process_tower_optimistic_confirmed( fd_replay_tile_t * ctx,
2559 : fd_stem_context_t * stem,
2560 0 : fd_tower_slot_confirmed_t const * msg ) {
2561 :
2562 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->block_id, NULL, ctx->block_id_arr );
2563 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2564 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_b58 );
2565 0 : FD_LOG_WARNING(( "missing bank for confirmed block_id: %s level %d", block_id_b58, msg->level ));
2566 0 : return;
2567 0 : }
2568 :
2569 0 : ulong bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2570 0 : fd_bank_t bank_[1]; fd_bank_t * bank = fd_banks_bank_query( bank_, ctx->banks, bank_idx );
2571 :
2572 :
2573 0 : if( FD_UNLIKELY( !bank ) ) {
2574 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_cstr );
2575 0 : FD_LOG_WARNING(( "failed to query optimistically confirmed bank for block id %s", block_id_cstr ));
2576 0 : return;
2577 0 : }
2578 :
2579 0 : if( ctx->rpc_enabled ) {
2580 0 : bank->data->refcnt++;
2581 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->data->idx, bank->data->f.slot, bank->data->refcnt ));
2582 0 : }
2583 :
2584 0 : fd_replay_oc_advanced_t * replay_msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2585 0 : replay_msg->bank_idx = bank_idx;
2586 0 : replay_msg->slot = msg->slot;
2587 :
2588 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_OC_ADVANCED, ctx->replay_out->chunk, sizeof(fd_replay_oc_advanced_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2589 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_oc_advanced_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2590 0 : }
2591 :
2592 : static inline int
2593 : returnable_frag( fd_replay_tile_t * ctx,
2594 : ulong in_idx,
2595 : ulong seq,
2596 : ulong sig,
2597 : ulong chunk,
2598 : ulong sz,
2599 : ulong ctl,
2600 : ulong tsorig,
2601 : ulong tspub,
2602 0 : fd_stem_context_t * stem ) {
2603 0 : (void)seq;
2604 0 : (void)ctl;
2605 0 : (void)tsorig;
2606 0 : (void)tspub;
2607 :
2608 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
2609 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 ));
2610 :
2611 0 : switch( ctx->in_kind[in_idx] ) {
2612 0 : case IN_KIND_GENESIS: {
2613 0 : fd_genesis_meta_t const * meta = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2614 0 : ctx->has_genesis_hash = 1;
2615 0 : ctx->has_genesis_timestamp = 1;
2616 0 : ctx->genesis_timestamp = meta->creation_time_seconds;
2617 0 : *ctx->genesis_hash = meta->genesis_hash;
2618 0 : if( FD_LIKELY( meta->bootstrap ) ) {
2619 0 : boot_genesis( ctx, stem, meta );
2620 0 : } else {
2621 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
2622 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
2623 0 : }
2624 0 : ctx->has_genesis_timestamp = 1;
2625 0 : ctx->genesis_timestamp = ctx->genesis->creation_time;
2626 :
2627 0 : maybe_verify_cluster_type( ctx );
2628 0 : maybe_verify_shred_version( ctx );
2629 0 : maybe_verify_genesis_timestamp( ctx );
2630 0 : break;
2631 0 : }
2632 0 : case IN_KIND_IPECHO: {
2633 0 : FD_TEST( sig && sig<=USHORT_MAX );
2634 0 : ctx->ipecho_shred_version = (ushort)sig;
2635 0 : maybe_verify_shred_version( ctx );
2636 0 : break;
2637 0 : }
2638 0 : case IN_KIND_SNAP: {
2639 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
2640 0 : maybe_verify_shred_version( ctx );
2641 0 : maybe_verify_genesis_timestamp( ctx );
2642 0 : break;
2643 0 : }
2644 0 : case IN_KIND_EXECRP: {
2645 0 : process_exec_task_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
2646 0 : break;
2647 0 : }
2648 0 : case IN_KIND_POH: {
2649 0 : process_poh_message( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2650 0 : break;
2651 0 : }
2652 0 : case IN_KIND_RESOLV: {
2653 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2654 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
2655 0 : break;
2656 0 : }
2657 0 : case IN_KIND_TOWER: {
2658 0 : if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_DONE ) ) {
2659 0 : process_tower_slot_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), seq );
2660 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_CONFIRMED ) ) {
2661 0 : fd_tower_slot_confirmed_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2662 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_OPTIMISTIC && !msg->fwd ) process_tower_optimistic_confirmed( ctx, stem, msg );
2663 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_DUPLICATE ) fd_reasm_confirm( ctx->reasm, &msg->block_id );
2664 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_IGNORED ) ) {
2665 0 : fd_tower_slot_ignored_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2666 0 : fd_tower_slot_done_t ignored = {
2667 0 : .replay_slot = msg->slot,
2668 0 : .replay_bank_idx = msg->bank_idx,
2669 0 : .vote_slot = ULONG_MAX,
2670 0 : .reset_slot = ctx->reset_slot, /* Use most recent reset slot */
2671 0 : .reset_block_id = ctx->reset_block_id,
2672 0 : .root_slot = ULONG_MAX
2673 0 : };
2674 0 : process_tower_slot_done( ctx, stem, &ignored, seq );
2675 0 : }
2676 0 : break;
2677 0 : }
2678 0 : case IN_KIND_REPAIR: {
2679 : /* TODO: This message/sz should be defined. */
2680 0 : if( sz!=0 && fd_disco_shred_out_msg_type( sig )==FD_SHRED_OUT_MSG_TYPE_FEC ) {
2681 : /* If receive a FEC complete message. */
2682 0 : process_fec_complete( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2683 0 : }
2684 0 : break;
2685 0 : }
2686 0 : case IN_KIND_TXSEND: {
2687 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2688 0 : break;
2689 0 : }
2690 0 : case IN_KIND_GOSSIP_OUT: {
2691 0 : FD_TEST( sig==FD_GOSSIP_UPDATE_TAG_WFS_DONE );
2692 0 : ctx->wfs_complete = 1;
2693 0 : FD_LOG_NOTICE(( "Done waiting for supermajority. More than 80 percent of cluster stake has joined." ));
2694 0 : break;
2695 0 : }
2696 0 : case IN_KIND_RPC: {
2697 0 : fd_bank_t bank[1];
2698 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, sig ) );
2699 0 : bank->data->refcnt--;
2700 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for %s", bank->data->idx, bank->data->f.slot, bank->data->refcnt, ctx->in_kind[ in_idx ]==IN_KIND_RPC ? "rpc" : "gui" ));
2701 0 : break;
2702 0 : }
2703 0 : default:
2704 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
2705 0 : }
2706 :
2707 0 : return 0;
2708 0 : }
2709 :
2710 : static inline fd_replay_out_link_t
2711 : out1( fd_topo_t const * topo,
2712 : fd_topo_tile_t const * tile,
2713 0 : char const * name ) {
2714 0 : ulong idx = ULONG_MAX;
2715 :
2716 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2717 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2718 0 : if( !strcmp( link->name, name ) ) {
2719 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 ));
2720 0 : idx = i;
2721 0 : }
2722 0 : }
2723 :
2724 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
2725 :
2726 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2727 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2728 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2729 :
2730 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2731 0 : }
2732 :
2733 : static void
2734 : privileged_init( fd_topo_t * topo,
2735 0 : fd_topo_tile_t * tile ) {
2736 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2737 :
2738 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2739 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2740 :
2741 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
2742 :
2743 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
2744 0 : ctx->identity_idx = 0UL;
2745 :
2746 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
2747 0 : tile->replay.bundle.enabled = 0;
2748 0 : }
2749 :
2750 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2751 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2752 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
2753 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2754 0 : }
2755 0 : }
2756 :
2757 0 : FD_TEST( fd_rng_secure( &ctx->rng_seed, sizeof(ctx->rng_seed) ) );
2758 :
2759 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->reasm_seed, sizeof(ulong) ) ) ) {
2760 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2761 0 : }
2762 :
2763 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->vote_tracker_seed, sizeof(ulong) ) ) ) {
2764 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2765 0 : }
2766 :
2767 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->block_id_map_seed, sizeof(ulong) ) ) ) {
2768 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2769 0 : }
2770 :
2771 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->initial_block_id, sizeof(fd_hash_t) ) ) ) {
2772 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2773 0 : }
2774 :
2775 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->runtime_stack_seed, sizeof(ulong) ) ) ) {
2776 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2777 0 : }
2778 0 : }
2779 :
2780 : static void
2781 : unprivileged_init( fd_topo_t * topo,
2782 0 : fd_topo_tile_t * tile ) {
2783 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2784 :
2785 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
2786 :
2787 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2788 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2789 0 : void * runtime_stack_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS ) );
2790 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 );
2791 0 : void * block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
2792 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
2793 0 : void * reasm_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
2794 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
2795 0 : void * vinyl_req_pool_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
2796 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
2797 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
2798 0 : void * dump_proto_ctx_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
2799 0 : # if FD_HAS_FLATCC
2800 0 : void * block_dump_ctx = NULL;
2801 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2802 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
2803 0 : }
2804 0 : # endif
2805 :
2806 0 : ctx->runtime_stack = fd_runtime_stack_join( fd_runtime_stack_new( runtime_stack_mem, FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS, ctx->runtime_stack_seed ) );
2807 0 : FD_TEST( ctx->runtime_stack );
2808 :
2809 0 : ctx->wksp = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].wksp;
2810 :
2811 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
2812 0 : FD_TEST( store_obj_id!=ULONG_MAX );
2813 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
2814 0 : FD_TEST( ctx->store );
2815 :
2816 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
2817 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
2818 0 : ulong banks_locks_obj_id = fd_pod_query_ulong( topo->props, "banks_locks", ULONG_MAX );
2819 0 : FD_TEST( banks_locks_obj_id!=ULONG_MAX );
2820 :
2821 0 : FD_TEST( fd_banks_join( ctx->banks, fd_topo_obj_laddr( topo, banks_obj_id ), fd_topo_obj_laddr( topo, banks_locks_obj_id ) ) );
2822 :
2823 0 : FD_MGAUGE_SET( REPLAY, MAX_LIVE_BANKS, fd_banks_pool_max_cnt( ctx->banks ) );
2824 :
2825 0 : ctx->frontier_cnt = 0UL;
2826 :
2827 0 : fd_bank_t bank[1];
2828 0 : FD_TEST( fd_banks_init_bank( bank, ctx->banks ) );
2829 0 : bank->data->f.slot = 0UL;
2830 0 : FD_TEST( bank->data->idx==FD_REPLAY_BOOT_BANK_IDX );
2831 :
2832 0 : ctx->consensus_root_slot = ULONG_MAX;
2833 0 : ctx->consensus_root = ctx->initial_block_id;
2834 0 : ctx->published_root_slot = ULONG_MAX;
2835 :
2836 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
2837 0 : ctx->ipecho_shred_version = 0;
2838 0 : fd_memcpy( ctx->genesis_path, tile->replay.genesis_path, sizeof(ctx->genesis_path) );
2839 0 : ctx->has_genesis_hash = 0;
2840 0 : ctx->has_genesis_timestamp = 0;
2841 0 : ctx->has_expected_genesis_timestamp = 0;
2842 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
2843 0 : ctx->hard_forks_cnt = ULONG_MAX;
2844 :
2845 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2846 0 : ctx->bundle.enabled = 1;
2847 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
2848 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
2849 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
2850 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2851 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
2852 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
2853 0 : }
2854 0 : } else {
2855 0 : ctx->bundle.enabled = 0;
2856 0 : }
2857 :
2858 0 : fd_features_t * features = &bank->data->f.features;
2859 0 : fd_features_enable_cleaned_up( features );
2860 :
2861 0 : char const * one_off_features[ 16UL ];
2862 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(one_off_features)/sizeof(one_off_features[0]) );
2863 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) one_off_features[ i ] = tile->replay.enable_features[i];
2864 0 : fd_features_enable_one_offs( features, one_off_features, (uint)tile->replay.enable_features_cnt, 0UL );
2865 :
2866 0 : fd_topo_obj_t const * vinyl_data = fd_topo_find_tile_obj( topo, tile, "vinyl_data" );
2867 :
2868 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ) )!=ULONG_MAX );
2869 0 : FD_TEST( fd_progcache_shmem_join( ctx->progcache, fd_topo_obj_laddr( topo, progcache_obj_id ) ) );
2870 :
2871 0 : ulong funk_obj_id; FD_TEST( (funk_obj_id = fd_pod_query_ulong( topo->props, "funk", ULONG_MAX ) )!=ULONG_MAX );
2872 0 : ulong funk_locks_obj_id; FD_TEST( (funk_locks_obj_id = fd_pod_query_ulong( topo->props, "funk_locks", ULONG_MAX ) )!=ULONG_MAX );
2873 0 : ulong max_depth = tile->replay.max_live_slots + tile->replay.write_delay_slots;
2874 0 : if( !vinyl_data ) {
2875 0 : FD_TEST( fd_accdb_admin_v1_init( ctx->accdb_admin,
2876 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2877 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ) ) );
2878 0 : } else {
2879 0 : fd_topo_obj_t const * vinyl_rq = fd_topo_find_tile_obj( topo, tile, "vinyl_rq" );
2880 0 : fd_topo_obj_t const * vinyl_req_pool = fd_topo_find_tile_obj( topo, tile, "vinyl_rpool" );
2881 0 : FD_TEST( fd_accdb_admin_v2_init( ctx->accdb_admin,
2882 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2883 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ),
2884 0 : fd_topo_obj_laddr( topo, vinyl_rq->id ),
2885 0 : topo->workspaces[ vinyl_data->wksp_id ].wksp,
2886 0 : fd_topo_obj_laddr( topo, vinyl_req_pool->id ),
2887 0 : vinyl_rq->id,
2888 0 : max_depth ) );
2889 0 : fd_accdb_admin_v2_delay_set( ctx->accdb_admin, tile->replay.write_delay_slots );
2890 0 : }
2891 0 : fd_accdb_init_from_topo( ctx->accdb, topo, tile, max_depth );
2892 :
2893 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
2894 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
2895 0 : FD_TEST( txncache_shmem );
2896 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
2897 0 : FD_TEST( ctx->txncache );
2898 :
2899 0 : ctx->capture_ctx = NULL;
2900 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
2901 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
2902 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
2903 0 : ctx->capture_ctx->capture_solcap = 1;
2904 0 : }
2905 :
2906 0 : ctx->dump_proto_ctx = NULL;
2907 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2908 0 : ctx->dump_proto_ctx = dump_proto_ctx_mem;
2909 0 : ctx->dump_proto_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
2910 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) {
2911 0 : ctx->dump_proto_ctx->dump_block_to_pb = !!tile->replay.dump_block_to_pb;
2912 0 : }
2913 0 : }
2914 :
2915 0 : # if FD_HAS_FLATCC
2916 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2917 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
2918 0 : } else {
2919 0 : ctx->block_dump_ctx = NULL;
2920 0 : }
2921 0 : # endif
2922 :
2923 0 : ctx->is_booted = 0;
2924 :
2925 0 : ctx->larger_max_cost_per_block = tile->replay.larger_max_cost_per_block;
2926 :
2927 0 : FD_TEST( fd_rng_new( ctx->rng, ctx->rng_seed, 0UL ) );
2928 :
2929 0 : ctx->reasm = fd_reasm_join( fd_reasm_new( reasm_mem, tile->replay.fec_max, ctx->reasm_seed ) );
2930 0 : FD_TEST( ctx->reasm );
2931 :
2932 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, ctx->rng, tile->replay.sched_depth, tile->replay.max_live_slots, fd_topo_tile_name_cnt( topo, "execrp" ) ) );
2933 0 : FD_TEST( ctx->sched );
2934 :
2935 0 : FD_TEST( fd_vinyl_req_pool_new( vinyl_req_pool_mem, 1UL, 1UL ) );
2936 :
2937 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, ctx->vote_tracker_seed ) );
2938 0 : FD_TEST( ctx->vote_tracker );
2939 :
2940 0 : ctx->identity_vote_rooted = 0;
2941 :
2942 0 : ctx->wait_for_vote_to_start_leader = tile->replay.wait_for_vote_to_start_leader;
2943 :
2944 0 : ctx->wfs_enabled = memcmp( tile->replay.wait_for_supermajority_with_bank_hash.uc, ((fd_pubkey_t){ 0 }).uc, sizeof(fd_pubkey_t) );
2945 0 : ctx->expected_bank_hash = tile->replay.wait_for_supermajority_with_bank_hash;
2946 0 : ctx->wfs_complete = !ctx->wfs_enabled;
2947 :
2948 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
2949 0 : FD_TEST( ctx->mleaders );
2950 :
2951 0 : ctx->is_leader = 0;
2952 0 : ctx->supports_leader = fd_topo_find_tile( topo, "pack", 0UL )!=ULONG_MAX;
2953 0 : ctx->reset_slot = 0UL;
2954 0 : fd_memset( ctx->reset_bank, 0, sizeof(fd_bank_t) );
2955 0 : ctx->reset_block_id = ctx->initial_block_id;
2956 0 : ctx->reset_timestamp_nanos = 0UL;
2957 0 : ctx->next_leader_slot = ULONG_MAX;
2958 0 : ctx->next_leader_tickcount = LONG_MAX;
2959 0 : ctx->highwater_leader_slot = ULONG_MAX;
2960 0 : ctx->slot_duration_nanos = 350L*1000L*1000L; /* TODO: Not fixed ... not always 350ms ... */
2961 0 : ctx->slot_duration_ticks = (double)ctx->slot_duration_nanos*fd_tempo_tick_per_ns( NULL );
2962 0 : ctx->leader_bank->data = NULL;
2963 :
2964 0 : ctx->block_id_len = tile->replay.max_live_slots;
2965 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
2966 0 : ctx->block_id_map = fd_block_id_map_join( fd_block_id_map_new( block_id_map_mem, chain_cnt, ctx->block_id_map_seed ) );
2967 0 : FD_TEST( ctx->block_id_map );
2968 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) ctx->block_id_arr[ i ].block_id_seen = 0;
2969 :
2970 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
2971 :
2972 0 : ctx->keyswitch = fd_keyswitch_join( fd_topo_obj_laddr( topo, tile->id_keyswitch_obj_id ) );
2973 0 : FD_TEST( ctx->keyswitch );
2974 0 : ctx->halt_leader = 0;
2975 :
2976 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
2977 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
2978 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
2979 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
2980 :
2981 0 : if( FD_LIKELY( link->dcache ) ) {
2982 0 : ctx->in[ i ].mem = link_wksp->wksp;
2983 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
2984 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
2985 0 : ctx->in[ i ].mtu = link->mtu;
2986 0 : }
2987 :
2988 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
2989 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
2990 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
2991 0 : else if( !strcmp( link->name, "execrp_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXECRP;
2992 0 : else if( !strcmp( link->name, "tower_out" ) ) ctx->in_kind[ i ] = IN_KIND_TOWER;
2993 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
2994 0 : else if( !strcmp( link->name, "resolv_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
2995 0 : else if( !strcmp( link->name, "repair_out" ) ) ctx->in_kind[ i ] = IN_KIND_REPAIR;
2996 0 : else if( !strcmp( link->name, "txsend_out" ) ) ctx->in_kind[ i ] = IN_KIND_TXSEND;
2997 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
2998 0 : else if( !strcmp( link->name, "gossip_out" ) ) ctx->in_kind[ i ] = IN_KIND_GOSSIP_OUT;
2999 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
3000 0 : }
3001 :
3002 0 : *ctx->epoch_out = out1( topo, tile, "replay_epoch" ); FD_TEST( ctx->epoch_out->idx!=ULONG_MAX );
3003 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
3004 0 : *ctx->exec_out = out1( topo, tile, "replay_execrp" ); FD_TEST( ctx->exec_out->idx!=ULONG_MAX );
3005 :
3006 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
3007 :
3008 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
3009 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "cap_repl", 0UL );
3010 0 : FD_TEST( idx!=ULONG_MAX );
3011 0 : fd_topo_link_t * link = &topo->links[ tile->out_link_id[ idx ] ];
3012 :
3013 :
3014 0 : fd_capture_link_buf_t * cap_repl_out = ctx->cap_repl_out;
3015 0 : cap_repl_out->base.vt = &fd_capture_link_buf_vt;
3016 0 : cap_repl_out->idx = idx;
3017 0 : cap_repl_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
3018 0 : cap_repl_out->chunk0 = fd_dcache_compact_chunk0( cap_repl_out->mem, link->dcache );
3019 0 : cap_repl_out->wmark = fd_dcache_compact_wmark( cap_repl_out->mem, link->dcache, link->mtu );
3020 0 : cap_repl_out->chunk = cap_repl_out->chunk0;
3021 0 : cap_repl_out->mcache = link->mcache;
3022 0 : cap_repl_out->depth = fd_mcache_depth( link->mcache );
3023 0 : cap_repl_out->seq = 0UL;
3024 :
3025 0 : ctx->capture_ctx->capctx_type.buf = cap_repl_out;
3026 0 : ctx->capture_ctx->capture_link = &cap_repl_out->base;
3027 0 : ctx->capture_ctx->current_txn_idx = 0UL;
3028 :
3029 :
3030 0 : ulong consumer_tile_idx = fd_topo_find_tile( topo, "solcap", 0UL );
3031 0 : fd_topo_tile_t * consumer_tile = &topo->tiles[ consumer_tile_idx ];
3032 0 : cap_repl_out->fseq = NULL;
3033 0 : for( ulong j = 0UL; j < consumer_tile->in_cnt; j++ ) {
3034 0 : if( FD_UNLIKELY( consumer_tile->in_link_id[ j ] == link->id ) ) {
3035 0 : cap_repl_out->fseq = fd_fseq_join( fd_topo_obj_laddr( topo, consumer_tile->in_link_fseq_obj_id[ j ] ) );
3036 0 : FD_TEST( cap_repl_out->fseq );
3037 0 : break;
3038 0 : }
3039 0 : }
3040 0 : }
3041 :
3042 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
3043 :
3044 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WAIT ),
3045 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WAIT ) ) );
3046 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WORK ),
3047 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WORK ) ) );
3048 :
3049 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_slot_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_SLOT_DURATION_SECONDS ),
3050 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_SLOT_DURATION_SECONDS ) ) );
3051 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_account_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ),
3052 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ) ) );
3053 :
3054 : /* Ensure precompiles are available, crash fast otherwise */
3055 0 : fd_precompiles();
3056 :
3057 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
3058 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
3059 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
3060 0 : }
3061 :
3062 : static ulong
3063 : populate_allowed_seccomp( fd_topo_t const * topo FD_FN_UNUSED,
3064 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3065 : ulong out_cnt,
3066 0 : struct sock_filter * out ) {
3067 :
3068 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() );
3069 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
3070 0 : }
3071 :
3072 : static ulong
3073 : populate_allowed_fds( fd_topo_t const * topo FD_FN_UNUSED,
3074 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3075 : ulong out_fds_cnt,
3076 0 : int * out_fds ) {
3077 :
3078 0 : if( FD_UNLIKELY( out_fds_cnt<2UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
3079 :
3080 0 : ulong out_cnt = 0UL;
3081 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
3082 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
3083 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
3084 0 : return out_cnt;
3085 0 : }
3086 :
3087 : static inline void
3088 0 : during_housekeeping( fd_replay_tile_t * ctx ) {
3089 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_UNHALT_PENDING ) ) {
3090 0 : FD_CRIT( ctx->halt_leader, "state machine corruption" );
3091 0 : FD_LOG_DEBUG(( "keyswitch: unhalting leader" ));
3092 0 : ctx->halt_leader = 0;
3093 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
3094 0 : }
3095 :
3096 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) {
3097 0 : FD_LOG_DEBUG(( "keyswitch: halting leader" ));
3098 0 : ctx->halt_leader = 1;
3099 0 : if( !ctx->is_leader ) maybe_switch_identity( ctx );
3100 0 : }
3101 0 : }
3102 :
3103 : #undef DEBUG_LOGGING
3104 :
3105 : /* counting carefully, after_credit can generate at most 7 frags and
3106 : returnable_frag boot_genesis can also generate at most 7 frags, so 14
3107 : is a conservative bound. */
3108 0 : #define STEM_BURST (14UL)
3109 :
3110 : /* fd_tempo_lazy_default( 16384 ) where 16384 is the minimum out-link
3111 : depth (i.e. cr_max) but excludes replay_epoch, which is so infrequent
3112 : credit availability is a non-issue. */
3113 0 : #define STEM_LAZY ((long)36865)
3114 :
3115 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
3116 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
3117 :
3118 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
3119 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
3120 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
3121 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
3122 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
3123 :
3124 : #include "../../disco/stem/fd_stem.c"
3125 :
3126 : fd_topo_run_tile_t fd_tile_replay = {
3127 : .name = "replay",
3128 : .populate_allowed_seccomp = populate_allowed_seccomp,
3129 : .populate_allowed_fds = populate_allowed_fds,
3130 : .scratch_align = scratch_align,
3131 : .scratch_footprint = scratch_footprint,
3132 : .privileged_init = privileged_init,
3133 : .unprivileged_init = unprivileged_init,
3134 : .run = stem_run,
3135 : };
|
