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/runtime/fd_runtime_stack.h"
39 : #include "../../flamenco/runtime/fd_genesis_parse.h"
40 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
41 : #include "../../flamenco/runtime/program/fd_precompiles.h"
42 : #include "../../flamenco/runtime/program/vote/fd_vote_state_versioned.h"
43 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
44 :
45 : #include <stdio.h>
46 :
47 : /* Replay concepts:
48 :
49 : - Blocks are aggregations of entries aka. microblocks which are
50 : groupings of txns and are constructed by the block producer (see
51 : fd_pack).
52 :
53 : - Entries are grouped into entry batches by the block producer (see
54 : fd_pack / fd_shredder).
55 :
56 : - Entry batches are divided into chunks known as shreds by the block
57 : producer (see fd_shredder).
58 :
59 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
60 : the block producer (see fd_shredder).
61 :
62 : - Shreds are transmitted to the rest of the cluster via the Turbine
63 : protocol (see fd_shredder / fd_shred).
64 :
65 : - Once enough shreds within a FEC set are received to recover the
66 : entirety of the shred data encoded by that FEC set, the receiver
67 : can "complete" the FEC set (see fd_fec_resolver).
68 :
69 : - If shreds in the FEC set are missing such that it can't complete,
70 : the receiver can use the Repair protocol to request missing shreds
71 : in FEC set (see fd_repair).
72 :
73 : - The current Repair protocol does not support requesting coding
74 : shreds. As a result, some FEC sets might be actually complete
75 : (contain all data shreds). Repair currently hacks around this by
76 : forcing completion but the long-term solution is to add support for
77 : fec_repairing coding shreds via Repair.
78 :
79 : - FEC sets are delivered in partial-order to the Replay tile by the
80 : Repair tile. Currently Replay only supports replaying entry batches
81 : so FEC sets need to reassembled into an entry batch before they can
82 : be replayed. The new Dispatcher will change this by taking a FEC
83 : set as input instead. */
84 :
85 0 : #define IN_KIND_SNAP ( 0)
86 0 : #define IN_KIND_GENESIS ( 1)
87 0 : #define IN_KIND_IPECHO ( 2)
88 0 : #define IN_KIND_TOWER ( 3)
89 0 : #define IN_KIND_RESOLV ( 4)
90 0 : #define IN_KIND_POH ( 5)
91 0 : #define IN_KIND_EXECRP ( 6)
92 0 : #define IN_KIND_REPAIR ( 7)
93 0 : #define IN_KIND_TXSEND ( 8)
94 0 : #define IN_KIND_RPC ( 9)
95 0 : #define IN_KIND_GOSSIP_OUT (10)
96 :
97 : #define DEBUG_LOGGING 0
98 :
99 : /* The first bank that the replay tile produces either for genesis
100 : or the snapshot boot will always be at bank index 0. */
101 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
102 :
103 : struct fd_replay_in_link {
104 : fd_wksp_t * mem;
105 : ulong chunk0;
106 : ulong wmark;
107 : ulong mtu;
108 : };
109 :
110 : typedef struct fd_replay_in_link fd_replay_in_link_t;
111 :
112 : struct fd_replay_out_link {
113 : ulong idx;
114 : fd_wksp_t * mem;
115 : ulong chunk0;
116 : ulong wmark;
117 : ulong chunk;
118 : };
119 :
120 : typedef struct fd_replay_out_link fd_replay_out_link_t;
121 :
122 : /* fd_block_id_map is a simple map of block-ids to bank indices. The
123 : map sits on top of an array of fd_block_id_ele_t. This serves as a
124 : translation layer between block ids to bank indices. The data
125 : array is indexed by bank index and the latest observed merkle root
126 : for the bank index is stored in the array. Once the block id has
127 : been observed, the entry is keyed by the latest merkle root (aka the
128 : block id). */
129 :
130 : struct fd_block_id_ele {
131 : fd_hash_t latest_mr;
132 : uint latest_fec_idx;
133 : int block_id_seen;
134 : ulong slot;
135 : ulong next_;
136 : };
137 : typedef struct fd_block_id_ele fd_block_id_ele_t;
138 :
139 : #define MAP_NAME fd_block_id_map
140 : #define MAP_ELE_T fd_block_id_ele_t
141 : #define MAP_KEY_T fd_hash_t
142 0 : #define MAP_KEY latest_mr
143 0 : #define MAP_NEXT next_
144 0 : #define MAP_KEY_EQ(k0,k1) (!memcmp((k0),(k1), sizeof(fd_hash_t)))
145 0 : #define MAP_KEY_HASH(key,seed) (fd_hash((seed),(key),sizeof(fd_hash_t)))
146 : #include "../../util/tmpl/fd_map_chain.c"
147 :
148 : static inline ulong
149 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
150 0 : return (ulong)(ele - ele_arr);
151 0 : }
152 :
153 : struct fd_replay_tile {
154 : fd_wksp_t * wksp;
155 :
156 : fd_accdb_admin_t accdb_admin[1];
157 : fd_accdb_user_t accdb[1];
158 : fd_progcache_admin_t progcache_admin[1];
159 :
160 : fd_txncache_t * txncache;
161 : fd_store_t * store;
162 : fd_banks_t banks[1];
163 :
164 : /* This flag is 1 If we have seen a vote signature that our node has
165 : sent out get rooted at least one time. The value is 0 otherwise.
166 : We can't become leader and pack blocks until this flag has been
167 : set. This parallels the Agave 'has_new_vote_been_rooted'. */
168 : int identity_vote_rooted;
169 : int wait_for_vote_to_start_leader;
170 :
171 : /* wfs_enabled is 1 if the validator is booted in
172 : wait_for_supermajority mode. In this mode replay (and, by extension,
173 : downstream consumers) is not allowed to make progress until 80% of
174 : the cluster has published their ContactInfo in Gossip with a
175 : shred version matching expected_shred_version. When this happens,
176 : wfs_complete will be set to 1. */
177 : int wfs_enabled;
178 : int wfs_complete;
179 :
180 : fd_hash_t expected_bank_hash;
181 :
182 : ulong reasm_seed;
183 : fd_reasm_t * reasm;
184 : fd_reasm_fec_t * reasm_evicted; /* evicted FEC by reasm_insert must be stored in returnable_frag, and then drained in after_credit */
185 :
186 : fd_sched_t * sched;
187 :
188 : ulong vote_tracker_seed;
189 : fd_vote_tracker_t * vote_tracker;
190 :
191 : int has_genesis_hash;
192 : char genesis_path[ PATH_MAX ];
193 : fd_hash_t genesis_hash[1];
194 : fd_genesis_t genesis[1];
195 : ulong cluster_type;
196 :
197 : int has_genesis_timestamp;
198 : ulong genesis_timestamp;
199 : int has_expected_genesis_timestamp;
200 : ulong expected_genesis_timestamp;
201 :
202 : #define FD_REPLAY_HARD_FORKS_MAX (64UL)
203 : ulong hard_forks_cnt;
204 : ulong hard_forks[ FD_REPLAY_HARD_FORKS_MAX ];
205 : ulong hard_forks_cnts[ FD_REPLAY_HARD_FORKS_MAX ];
206 :
207 : ushort expected_shred_version;
208 : ushort ipecho_shred_version;
209 :
210 : /* A note on publishing ...
211 :
212 : The watermarks are used to publish our fork-aware structures. For
213 : example, store, banks, and txncache need to be published to release
214 : resources occupied by rooted or dead blocks. In general,
215 : publishing has the effect of pruning forks in those structures,
216 : indicating that it is ok to release the memory being occupied by
217 : the blocks on said forks. Tower is responsible for informing us of
218 : the latest block on the consensus rooted fork. As soon as we can,
219 : we should move the published root as close as possible to the
220 : latest consensus root, publishing/pruning everything on the fork
221 : tree along the way. That is, all the blocks that directly descend
222 : from the current published root (inclusive) to the new published
223 : root (exclusive) on the rooted fork, as well as all the minority
224 : forks that branch from said blocks.
225 :
226 : Ideally, we'd move the published root to the consensus root
227 : immediately upon receiving a new consensus root. However, that's
228 : not always safe to do. One thing we need to be careful about is
229 : making sure that there are no more users/consumers of
230 : soon-to-be-pruned blocks, lest a use-after-free occurs. This can
231 : be done by using a reference counter for each block. Any
232 : concurrent activity, such as transaction execution in the exec
233 : tiles, should retain a refcnt on the block for as
234 : long as it needs access to the shared fork-aware structures related
235 : to that block. Eventually, refcnt on a given block will drop down
236 : to 0 as the block either finishes replaying or gets marked as dead,
237 : and any other tile that has retained a refcnt on the block releases
238 : it. At that point, it becomes a candidate for pruning. The key to
239 : safe publishing then becomes figuring out how far we could advance
240 : the published root, such that every minority fork branching off of
241 : blocks in between the current published root (inclusive) and the
242 : new published root (exclusive) is safe to be pruned. This is a
243 : straightforward tree traversal, where if a block B on the rooted
244 : fork has refcnt 0, and all minority forks branching off of B also
245 : have refcnt 0, then B is safe to be pruned. We advance the
246 : published root to the farthest consecutively prunable block on the
247 : rooted fork. Note that reasm presents the replay tile with a clean
248 : view of the world where every block is chained off of a parent
249 : block. So there are no orpahned/dangling tree nodes to worry
250 : about. The world is a nice single tree as far as replay is
251 : concerned.
252 :
253 : In the following fork tree, every node is a block and the number in
254 : parentheses is the refcnt on the block. The chain marked with
255 : double slashes is the rooted fork. Suppose the published root is
256 : at block P, and consensus root is at block T. We can't publish
257 : past block P because Q has refcnt 1.
258 :
259 :
260 : P(0)
261 : / \\
262 : Q(1) A(0)
263 : / || \
264 : X(0) B(0) C(0)
265 : / || \
266 : Y(0) M(0) R(0)
267 : / || / \
268 : D(2) T(0) J(0) L(0)
269 : ||
270 : ..
271 : ..
272 : ..
273 : ||
274 : blocks we might be actively replaying
275 :
276 :
277 : When refcnt on Q drops to 0, we would be able to advance the
278 : published root to block M, because blocks P, A, and B, as well as
279 : all subtrees branching off of them, have refcnt 0, and therefore
280 : can be pruned. Block M itself cannot be pruned yet because its
281 : child block D has refcnt 2. After publishing/pruning, the fork
282 : tree would be:
283 :
284 :
285 : M(0)
286 : / ||
287 : D(2) T(0)
288 : ||
289 : ..
290 : ..
291 : ..
292 : ||
293 : blocks we might be actively replaying
294 :
295 :
296 : As a result, the shared fork-aware structures can free resources
297 : for blocks P, A, B, and all subtrees branching off of them.
298 :
299 : For the reference counting part, the replay tile is the sole entity
300 : that can update the refcnt. This ensures that all refcnt increment
301 : and decrement attempts are serialized at the replay tile, and that
302 : there are no racy resurrection of a soon-to-be-pruned block. If a
303 : refcnt increment request arrives after a block has been pruned,
304 : replay simply rejects the request.
305 :
306 : A note on the implementation of the above ...
307 :
308 : Upon receiving a new consensus root, we descend down the rooted
309 : fork from the current published root to the new consensus root. On
310 : each node/block of the rooted fork, we do a summation of the refcnt
311 : on the block and all the minority fork blocks branching from the
312 : block. If the summation is 0, the block is safe for pruning. We
313 : advance the published root to the far end of the consecutive run of
314 : 0 refcnt sums originating from the current published root. On our
315 : descent down the minority forks, we also mark any block that hasn't
316 : finished replaying as dead, so we don't waste time executing them.
317 : No more transactions shall be dispatched for execution from dead
318 : blocks.
319 :
320 : Blocks start out with a refcnt of 0. Other tiles may send a
321 : request to the replay tile for a reference on a block. The
322 : transaction dispatcher is another source of refcnt updates. On
323 : every dispatch of a transaction for block B, we increment the
324 : refcnt for B. And on every transaction finalization, we decrement
325 : the refcnt for B. This means that whenever the refcnt on a block
326 : is 0, there is no more reference on that block from the execution
327 : pipeline. While it might be tempting to simply increment the
328 : refcnt once when we start replaying a block, and decrement the
329 : refcnt once when we finish a block, this more fine-grained refcnt
330 : update strategy allows for aborting and potentially immediate
331 : pruning of blocks under interleaved block replay. Upon receiving a
332 : new consensus root, we can simply look at the refcnt on minority
333 : fork blocks, and a refcnt of 0 would imply that the block is safe
334 : for pruning, even if we haven't finished replaying it. Without the
335 : fine-grained refcnt, we would need to first stop dispatching from
336 : the aborted block, and then wait for a full drain of the execution
337 : pipeline to know for sure that there are no more in-flight
338 : transactions executing on the aborted block. Note that this will
339 : allow the refcnt on any block to transiently drop down to 0. We
340 : will not mistakenly prune an actively replaying block, aka a leaf
341 : node, that is chaining off of the rooted fork, because the
342 : consensus root is always an ancestor of the actively replaying tip.
343 : */
344 : fd_hash_t consensus_root; /* The most recent block to have reached max lockout in the tower. */
345 : ulong consensus_root_slot; /* slot number of the above. */
346 : ulong consensus_root_bank_idx; /* bank index of the above. */
347 : ulong published_root_slot; /* slot number of the published root. */
348 : ulong published_root_bank_idx; /* bank index of the published root. */
349 :
350 : /* Randomly generated block id for the initial genesis/snapshot slot.
351 : To be replaced with block id in the snapshot manifest when SIMD-333
352 : is activated. */
353 :
354 : fd_hash_t initial_block_id;
355 :
356 : /* We need to maintain a tile-local mapping of block-ids to bank index
357 : and vice versa. This translation layer is needed for conversion
358 : since tower operates on block-ids and downstream consumers of FEC
359 : sets operate on bank indices. This mapping must happen both ways:
360 : 1. tower sends us block ids and we must map them to bank indices.
361 : 2. when a block is completed, we must map the bank index to a block
362 : id to send a slot complete message to tower. */
363 : ulong block_id_len;
364 : fd_block_id_ele_t * block_id_arr;
365 : ulong block_id_map_seed;
366 : fd_block_id_map_t * block_id_map;
367 :
368 : /* Capture-related configs */
369 : fd_capture_ctx_t * capture_ctx;
370 : FILE * capture_file;
371 : fd_capture_link_buf_t cap_repl_out[1];
372 :
373 : /* Protobuf dumping context for debugging runtime execution and
374 : collecting seed corpora. */
375 : fd_dump_proto_ctx_t * dump_proto_ctx;
376 :
377 : /* Whether the runtime has been booted either from snapshot loading
378 : or from genesis. */
379 : int is_booted;
380 :
381 : /* Buffer to store vote towers that need to be published to the Tower
382 : tile. */
383 :
384 : fd_multi_epoch_leaders_t * mleaders;
385 :
386 : int larger_max_cost_per_block;
387 :
388 : /* When we transition to becoming leader, we can only unbecome the
389 : leader if we have received a block id from the FEC reassembler, and
390 : a message from PoH that the leader slot has ended. After both of
391 : these conditions are met, then we are free to unbecome the leader.
392 : */
393 : uint is_leader : 1;
394 : uint supports_leader : 1;
395 : int recv_poh;
396 : ulong next_leader_slot;
397 : long next_leader_tickcount;
398 : ulong highwater_leader_slot;
399 : ulong reset_slot;
400 : fd_bank_t reset_bank[1];
401 : fd_hash_t reset_block_id;
402 : long reset_timestamp_nanos;
403 : double slot_duration_nanos;
404 : double slot_duration_ticks;
405 : fd_bank_t leader_bank[1];
406 :
407 : fd_pubkey_t identity_pubkey[1];
408 : ulong identity_idx;
409 :
410 : fd_keyswitch_t * keyswitch;
411 : int halt_leader;
412 :
413 : ulong resolv_tile_cnt;
414 :
415 : int in_kind[ 128 ];
416 : fd_replay_in_link_t in[ 128 ];
417 :
418 : fd_replay_out_link_t exec_out[ 1 ];
419 :
420 : fd_replay_out_link_t replay_out[1];
421 :
422 : fd_replay_out_link_t epoch_out[1];
423 :
424 : /* The rpc tile needs to occasionally own a reference to a live bank.
425 : Replay needs to know if the rpc as a consumer is enabled so it can
426 : increment the bank's refcnt before publishing bank_idx. */
427 : int rpc_enabled;
428 :
429 : # if FD_HAS_FLATCC
430 : /* For dumping blocks to protobuf. For backtest only. */
431 : fd_block_dump_ctx_t * block_dump_ctx;
432 : # endif
433 :
434 : /* We need a few pieces of information to compute the right addresses
435 : for bundle crank information that we need to send to pack. */
436 : struct {
437 : int enabled;
438 : fd_pubkey_t vote_account;
439 : fd_bundle_crank_gen_t gen[1];
440 : } bundle;
441 :
442 : struct {
443 : ulong store_query_acquire;
444 : ulong store_query_release;
445 : fd_histf_t store_query_wait[1];
446 : fd_histf_t store_query_work[1];
447 : ulong store_query_cnt;
448 : ulong store_query_missing_cnt;
449 : ulong store_query_mr;
450 : ulong store_query_missing_mr;
451 :
452 : ulong slots_total;
453 : ulong transactions_total;
454 :
455 : ulong reasm_latest_slot;
456 : ulong reasm_latest_fec_idx;
457 :
458 : ulong sched_full;
459 : ulong reasm_empty;
460 : ulong leader_bid_wait;
461 : ulong banks_full;
462 : ulong storage_root_behind;
463 :
464 : fd_histf_t root_slot_dur[1];
465 : fd_histf_t root_account_dur[1];
466 : } metrics;
467 :
468 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
469 :
470 : ulong runtime_stack_seed;
471 : fd_runtime_stack_t runtime_stack;
472 : };
473 :
474 : typedef struct fd_replay_tile fd_replay_tile_t;
475 :
476 : FD_FN_CONST static inline ulong
477 0 : scratch_align( void ) {
478 0 : return 128UL;
479 0 : }
480 : FD_FN_PURE static inline ulong
481 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
482 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
483 :
484 0 : ulong l = FD_LAYOUT_INIT;
485 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
486 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
487 0 : l = FD_LAYOUT_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
488 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
489 0 : l = FD_LAYOUT_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
490 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
491 0 : l = FD_LAYOUT_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
492 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
493 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
494 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
495 :
496 0 : # if FD_HAS_FLATCC
497 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
498 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
499 0 : }
500 0 : # endif
501 :
502 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
503 :
504 0 : return l;
505 0 : }
506 :
507 : static inline void
508 0 : metrics_write( fd_replay_tile_t * ctx ) {
509 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_ACQUIRE, ctx->metrics.store_query_acquire );
510 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_RELEASE, ctx->metrics.store_query_release );
511 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WAIT, ctx->metrics.store_query_wait );
512 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WORK, ctx->metrics.store_query_work );
513 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_CNT, ctx->metrics.store_query_cnt );
514 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_MISSING_CNT, ctx->metrics.store_query_missing_cnt );
515 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MR, ctx->metrics.store_query_mr );
516 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MISSING_MR, ctx->metrics.store_query_missing_mr );
517 :
518 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
519 0 : ulong leader_slot = ctx->leader_bank->data ? fd_bank_slot_get( ctx->leader_bank ) : 0UL;
520 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
521 :
522 0 : if( FD_LIKELY( ctx->leader_bank->data ) ) {
523 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
524 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
525 0 : } else {
526 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
527 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
528 0 : }
529 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
530 :
531 0 : fd_bank_data_t * bank_pool = fd_banks_get_bank_pool( ctx->banks->data );
532 0 : ulong live_banks = fd_banks_pool_max( bank_pool ) - fd_banks_pool_free( bank_pool );
533 0 : FD_MGAUGE_SET( REPLAY, LIVE_BANKS, live_banks );
534 :
535 0 : ulong reasm_free = fd_reasm_free( ctx->reasm );
536 0 : FD_MGAUGE_SET( REPLAY, REASM_FREE, reasm_free );
537 :
538 0 : FD_MCNT_SET( REPLAY, SLOTS_TOTAL, ctx->metrics.slots_total );
539 0 : FD_MCNT_SET( REPLAY, TRANSACTIONS_TOTAL, ctx->metrics.transactions_total );
540 :
541 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_SLOT, ctx->metrics.reasm_latest_slot );
542 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_FEC_IDX, ctx->metrics.reasm_latest_fec_idx );
543 :
544 0 : FD_MCNT_SET( REPLAY, SCHED_FULL, ctx->metrics.sched_full );
545 0 : FD_MCNT_SET( REPLAY, REASM_EMPTY, ctx->metrics.reasm_empty );
546 0 : FD_MCNT_SET( REPLAY, LEADER_BID_WAIT, ctx->metrics.leader_bid_wait );
547 0 : FD_MCNT_SET( REPLAY, BANKS_FULL, ctx->metrics.banks_full );
548 0 : FD_MCNT_SET( REPLAY, STORAGE_ROOT_BEHIND, ctx->metrics.storage_root_behind );
549 :
550 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, ctx->progcache_admin->metrics.root_cnt );
551 0 : FD_MCNT_SET( REPLAY, PROGCACHE_GC_ROOT, ctx->progcache_admin->metrics.gc_root_cnt );
552 :
553 0 : FD_MCNT_SET( REPLAY, ACCDB_CREATED, ctx->accdb->base.created_cnt );
554 0 : FD_MCNT_SET( REPLAY, ACCDB_REVERTED, ctx->accdb_admin->base.revert_cnt );
555 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED, ctx->accdb_admin->base.root_cnt );
556 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED_BYTES, ctx->accdb_admin->base.root_tot_sz );
557 0 : FD_MCNT_SET( REPLAY, ACCDB_GC_ROOT, ctx->accdb_admin->base.gc_root_cnt );
558 0 : FD_MCNT_SET( REPLAY, ACCDB_RECLAIMED, ctx->accdb_admin->base.reclaim_cnt );
559 0 : FD_MHIST_COPY( REPLAY, ROOT_SLOT_DURATION_SECONDS, ctx->metrics.root_slot_dur );
560 0 : FD_MHIST_COPY( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS, ctx->metrics.root_account_dur );
561 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_DB, (ulong)ctx->accdb_admin->base.dt_vinyl );
562 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_COPY, (ulong)ctx->accdb_admin->base.dt_copy );
563 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_GC, (ulong)ctx->accdb_admin->base.dt_gc );
564 0 : }
565 :
566 : static inline ulong
567 : generate_epoch_info_msg( ulong slot,
568 : ulong epoch,
569 : fd_epoch_schedule_t const * epoch_schedule,
570 : fd_top_votes_t const * top_votes,
571 : fd_vote_stakes_t * vote_stakes,
572 : ushort vote_stakes_fork_idx,
573 : fd_features_t const * features,
574 : fd_epoch_info_msg_t * epoch_info_msg,
575 0 : int current_epoch ) {
576 0 : fd_vote_stake_weight_t * stake_weights = epoch_info_msg->weights;
577 :
578 0 : epoch_info_msg->epoch = epoch;
579 0 : epoch_info_msg->start_slot = fd_epoch_slot0( epoch_schedule, epoch );
580 0 : epoch_info_msg->slot_cnt = fd_epoch_slot_cnt( epoch_schedule, epoch );
581 0 : epoch_info_msg->excluded_stake = 0UL;
582 0 : epoch_info_msg->vote_keyed_lsched = 1UL;
583 :
584 : /* FIXME: SIMD-0180 - hack to (de)activate in testnet vs mainnet.
585 : This code can be removed once the feature is active. */
586 0 : if( (1==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_TESTNET) ||
587 0 : (0==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_MAINNET) ) {
588 0 : epoch_info_msg->vote_keyed_lsched = 0UL;
589 0 : }
590 :
591 0 : ulong idx = 0UL;
592 :
593 0 : if( FD_FEATURE_ACTIVE( slot, features, validator_admission_ticket ) ) {
594 :
595 0 : uchar __attribute__((aligned(FD_TOP_VOTES_ITER_ALIGN))) iter_mem[ FD_TOP_VOTES_ITER_FOOTPRINT ];
596 0 : for( fd_top_votes_iter_t * iter = fd_top_votes_iter_init( top_votes, iter_mem );
597 0 : !fd_top_votes_iter_done( top_votes, iter );
598 0 : fd_top_votes_iter_next( top_votes, iter ) ) {
599 0 : fd_pubkey_t pubkey;
600 0 : fd_top_votes_iter_ele( top_votes, iter, &pubkey, NULL, NULL, NULL, NULL );
601 :
602 0 : ulong stake;
603 0 : fd_pubkey_t node_account;
604 0 : int found;
605 0 : if( current_epoch ) {
606 0 : found = fd_vote_stakes_query_t_1( vote_stakes, vote_stakes_fork_idx, &pubkey, &stake, &node_account );
607 0 : } else {
608 0 : found = fd_vote_stakes_query_t_2( vote_stakes, vote_stakes_fork_idx, &pubkey, &stake, &node_account );
609 0 : }
610 0 : if( FD_UNLIKELY( !found ) ) continue;
611 :
612 0 : stake_weights[ idx ].stake = stake;
613 0 : memcpy( stake_weights[ idx ].id_key.uc, &node_account, sizeof(fd_pubkey_t) );
614 0 : memcpy( stake_weights[ idx ].vote_key.uc, &pubkey, sizeof(fd_pubkey_t) );
615 0 : idx++;
616 0 : }
617 0 : } else {
618 0 : uchar __attribute__((aligned(FD_VOTE_STAKES_ITER_ALIGN))) iter_mem[ FD_VOTE_STAKES_ITER_FOOTPRINT ];
619 0 : for( fd_vote_stakes_iter_t * iter = fd_vote_stakes_fork_iter_init( vote_stakes, vote_stakes_fork_idx, iter_mem );
620 0 : !fd_vote_stakes_fork_iter_done( vote_stakes, vote_stakes_fork_idx, iter );
621 0 : fd_vote_stakes_fork_iter_next( vote_stakes, vote_stakes_fork_idx, iter ) ) {
622 :
623 0 : fd_pubkey_t pubkey;
624 0 : ulong stake_t_1;
625 0 : ulong stake_t_2;
626 0 : fd_pubkey_t node_account_t_1;
627 0 : fd_pubkey_t node_account_t_2;
628 0 : fd_vote_stakes_fork_iter_ele( vote_stakes, vote_stakes_fork_idx, iter, &pubkey, &stake_t_1, &stake_t_2, &node_account_t_1, &node_account_t_2 );
629 :
630 0 : ulong stake = current_epoch ? stake_t_1 : stake_t_2;
631 0 : fd_pubkey_t node_account = current_epoch ? node_account_t_1 : node_account_t_2;
632 0 : if( FD_UNLIKELY( !stake ) ) continue;
633 :
634 0 : stake_weights[ idx ].stake = stake;
635 0 : memcpy( stake_weights[ idx ].id_key.uc, &node_account, sizeof(fd_pubkey_t) );
636 0 : memcpy( stake_weights[ idx ].vote_key.uc, &pubkey, sizeof(fd_pubkey_t) );
637 0 : idx++;
638 0 : }
639 0 : }
640 :
641 0 : epoch_info_msg->staked_cnt = idx;
642 0 : sort_vote_weights_by_stake_vote_inplace( stake_weights, idx );
643 :
644 0 : epoch_info_msg->epoch_schedule = *epoch_schedule;
645 0 : epoch_info_msg->features = *features;
646 :
647 0 : return fd_epoch_info_msg_sz( epoch_info_msg->staked_cnt );
648 0 : }
649 :
650 : static void
651 : publish_epoch_info( fd_replay_tile_t * ctx,
652 : fd_stem_context_t * stem,
653 : fd_bank_t * bank,
654 0 : int current_epoch ) {
655 0 : fd_epoch_schedule_t const * schedule = fd_bank_epoch_schedule_query( bank );
656 0 : ulong epoch = fd_slot_to_epoch( schedule, fd_bank_slot_get( bank ), NULL );
657 :
658 0 : fd_features_t const * features = fd_bank_features_query( bank );
659 :
660 0 : fd_epoch_info_msg_t * epoch_info_msg = fd_chunk_to_laddr( ctx->epoch_out->mem, ctx->epoch_out->chunk );
661 :
662 0 : fd_vote_stakes_t * vote_stakes = fd_bank_vote_stakes_locking_modify( bank );
663 0 : fd_top_votes_t const * top_votes = fd_bank_top_votes_query( bank );
664 0 : ulong epoch_info_sz = generate_epoch_info_msg( fd_bank_slot_get( bank ), epoch+fd_ulong_if( current_epoch, 1UL, 0UL), schedule, top_votes, vote_stakes, bank->data->vote_stakes_fork_id, features, epoch_info_msg, current_epoch );
665 0 : fd_bank_vote_stakes_end_locking_modify( bank );
666 :
667 0 : ulong epoch_info_sig = 4UL;
668 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() ) );
669 0 : ctx->epoch_out->chunk = fd_dcache_compact_next( ctx->epoch_out->chunk, epoch_info_sz, ctx->epoch_out->chunk0, ctx->epoch_out->wmark );
670 :
671 0 : fd_multi_epoch_leaders_epoch_msg_init( ctx->mleaders, epoch_info_msg );
672 0 : fd_multi_epoch_leaders_epoch_msg_fini( ctx->mleaders );
673 :
674 0 : }
675 :
676 : /**********************************************************************/
677 : /* Transaction execution state machine helpers */
678 : /**********************************************************************/
679 :
680 : static void
681 : replay_block_start( fd_replay_tile_t * ctx,
682 : fd_stem_context_t * stem,
683 : ulong bank_idx,
684 : ulong parent_bank_idx,
685 0 : ulong slot ) {
686 0 : long before = fd_log_wallclock();
687 :
688 0 : fd_bank_t bank[1];
689 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, bank_idx ) ) ) {
690 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
691 0 : }
692 0 : if( FD_UNLIKELY( bank->data->flags!=FD_BANK_FLAGS_INIT ) ) {
693 0 : FD_LOG_CRIT(( "invariant violation: bank is not in correct state for bank index %lu", bank_idx ));
694 0 : }
695 :
696 0 : bank->data->preparation_begin_nanos = before;
697 :
698 0 : fd_bank_t parent_bank[1];
699 0 : if( FD_UNLIKELY( !fd_banks_bank_query( parent_bank, ctx->banks, parent_bank_idx ) ) ) {
700 0 : FD_LOG_CRIT(( "invariant violation: parent bank is NULL for bank index %lu", parent_bank_idx ));
701 0 : }
702 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
703 :
704 : /* Clone the bank from the parent. We must special case the first
705 : slot that is executed as the snapshot does not provide a parent
706 : block id. */
707 :
708 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( bank, ctx->banks, bank_idx ) ) ) {
709 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
710 0 : }
711 0 : fd_bank_slot_set( bank, slot );
712 0 : fd_bank_parent_slot_set( bank, parent_slot );
713 0 : bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->data->txncache_fork_id );
714 :
715 : /* Create a new funk txn for the block. */
716 :
717 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank_idx } };
718 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
719 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
720 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
721 :
722 : /* Update required runtime state and handle potential boundary. */
723 :
724 0 : fd_bank_shred_cnt_set( bank, 0UL );
725 0 : fd_bank_execution_fees_set( bank, 0UL );
726 0 : fd_bank_priority_fees_set( bank, 0UL );
727 0 : fd_bank_tips_set( bank, 0UL );
728 0 : fd_bank_identity_vote_idx_set( bank, ULONG_MAX );
729 :
730 0 : fd_bank_block_height_set( bank, fd_bank_block_height_get( bank ) + 1UL );
731 :
732 0 : int is_epoch_boundary = 0;
733 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
734 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, bank, 1 );
735 :
736 0 : ulong max_tick_height;
737 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( fd_bank_ticks_per_slot_get( parent_bank ), slot, &max_tick_height ) ) ) {
738 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, fd_bank_ticks_per_slot_get( parent_bank ) ));
739 0 : }
740 0 : fd_bank_max_tick_height_set( bank, max_tick_height );
741 0 : fd_bank_tick_height_set( bank, fd_bank_max_tick_height_get( parent_bank ) ); /* The parent's max tick height is our starting tick height. */
742 0 : fd_sched_set_poh_params( ctx->sched, bank->data->idx, fd_bank_tick_height_get( bank ), fd_bank_max_tick_height_get( bank ), fd_bank_hashes_per_tick_get( bank ), fd_bank_poh_query( parent_bank ) );
743 :
744 0 : FD_LOG_DEBUG(( "replay_block_start: bank_idx=%lu slot=%lu parent_bank_idx=%lu", bank_idx, slot, parent_bank_idx ));
745 0 : }
746 :
747 : static void
748 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
749 0 : if( bank->data->cost_tracker_pool_idx!=fd_bank_cost_tracker_pool_idx_null( fd_bank_get_cost_tracker_pool( bank->data ) ) ) {
750 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
751 0 : if( FD_UNLIKELY( cost_tracker->block_cost_limit==0UL ) ) {
752 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
753 0 : } else {
754 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
755 0 : slot_info->cost_tracker.vote_cost = cost_tracker->vote_cost;
756 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
757 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
758 0 : slot_info->cost_tracker.vote_cost_limit = cost_tracker->vote_cost_limit;
759 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
760 0 : }
761 0 : fd_bank_cost_tracker_end_locking_query( bank );
762 0 : } else {
763 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
764 0 : }
765 0 : }
766 :
767 : static ulong
768 0 : get_identity_balance( fd_replay_tile_t * ctx, fd_funk_txn_xid_t xid ) {
769 0 : ulong identity_balance = ULONG_MAX;
770 0 : fd_accdb_ro_t identity_acc[1];
771 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, identity_acc, &xid, ctx->identity_pubkey ) ) ) {
772 0 : identity_balance = identity_acc->meta->lamports;
773 0 : fd_accdb_close_ro( ctx->accdb, identity_acc );
774 0 : }
775 0 : return identity_balance;
776 0 : }
777 :
778 : static void
779 : publish_slot_completed( fd_replay_tile_t * ctx,
780 : fd_stem_context_t * stem,
781 : fd_bank_t * bank,
782 : int is_initial,
783 0 : int is_leader ) {
784 :
785 0 : ulong slot = fd_bank_slot_get( bank );
786 :
787 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->data->idx ];
788 :
789 : /* HACKY: hacky way of checking if we should send a null parent block
790 : id */
791 0 : fd_hash_t parent_block_id = {0};
792 0 : if( FD_UNLIKELY( !is_initial ) ) {
793 0 : parent_block_id = ctx->block_id_arr[ bank->data->parent_idx ].latest_mr;
794 0 : }
795 :
796 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
797 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
798 0 : FD_TEST( bank_hash );
799 0 : FD_TEST( block_hash );
800 :
801 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->data->txncache_fork_id, 0UL, block_hash->uc );
802 :
803 0 : fd_epoch_schedule_t const * epoch_schedule = fd_bank_epoch_schedule_query( bank );
804 0 : ulong slot_idx;
805 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
806 :
807 0 : ctx->metrics.slots_total++;
808 0 : ctx->metrics.transactions_total = fd_bank_txn_count_get( bank );
809 :
810 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
811 0 : slot_info->slot = slot;
812 0 : slot_info->root_slot = ctx->consensus_root_slot;
813 0 : slot_info->storage_slot = ctx->published_root_slot;
814 0 : slot_info->epoch = epoch;
815 0 : slot_info->slot_in_epoch = slot_idx;
816 0 : slot_info->slots_per_epoch = fd_epoch_slot_cnt( epoch_schedule, epoch );
817 0 : slot_info->block_height = fd_bank_block_height_get( bank );
818 0 : slot_info->parent_slot = fd_bank_parent_slot_get( bank );
819 0 : slot_info->block_id = block_id_ele->latest_mr;
820 0 : slot_info->parent_block_id = parent_block_id;
821 0 : slot_info->bank_hash = *bank_hash;
822 0 : slot_info->block_hash = *block_hash;
823 0 : slot_info->transaction_count = fd_bank_txn_count_get( bank );
824 :
825 0 : fd_inflation_t inflation = fd_bank_inflation_get( bank );
826 0 : slot_info->inflation.foundation = inflation.foundation;
827 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
828 0 : slot_info->inflation.terminal = inflation.terminal;
829 0 : slot_info->inflation.initial = inflation.initial;
830 0 : slot_info->inflation.taper = inflation.taper;
831 :
832 0 : fd_rent_t rent = fd_bank_rent_get( bank );
833 0 : slot_info->rent.burn_percent = rent.burn_percent;
834 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
835 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
836 :
837 0 : slot_info->first_fec_set_received_nanos = bank->data->first_fec_set_received_nanos;
838 0 : slot_info->preparation_begin_nanos = bank->data->preparation_begin_nanos;
839 0 : slot_info->first_transaction_scheduled_nanos = bank->data->first_transaction_scheduled_nanos;
840 0 : slot_info->last_transaction_finished_nanos = bank->data->last_transaction_finished_nanos;
841 0 : slot_info->completion_time_nanos = fd_log_wallclock();
842 :
843 : /* refcnt should be incremented by 1 for each consumer that uses
844 : `bank_idx`. Each consumer should decrement the bank's refcnt once
845 : they are done using the bank. */
846 0 : bank->data->refcnt++; /* tower_tile */
847 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->data->refcnt++; /* rpc tile */
848 0 : slot_info->bank_idx = bank->data->idx;
849 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for tower, rpc", bank->data->idx, slot, bank->data->refcnt ));
850 :
851 0 : fd_bank_t parent_bank[1];
852 0 : if( FD_LIKELY( fd_banks_get_parent( parent_bank, ctx->banks, bank ) ) ) {
853 0 : ulong total_txn_cnt = fd_bank_txn_count_get( bank ) - fd_bank_txn_count_get( parent_bank );
854 0 : ulong nonvote_txn_cnt = fd_bank_nonvote_txn_count_get( bank ) - fd_bank_nonvote_txn_count_get( parent_bank );
855 0 : ulong failed_txn_cnt = fd_bank_failed_txn_count_get( bank ) - fd_bank_failed_txn_count_get( parent_bank );
856 0 : ulong nonvote_failed_txn_cnt = fd_bank_nonvote_failed_txn_count_get( bank ) - fd_bank_nonvote_failed_txn_count_get( parent_bank );
857 :
858 0 : slot_info->nonvote_success = nonvote_txn_cnt - nonvote_failed_txn_cnt;
859 0 : slot_info->nonvote_failed = nonvote_failed_txn_cnt;
860 0 : slot_info->vote_failed = failed_txn_cnt - nonvote_failed_txn_cnt;
861 0 : slot_info->vote_success = total_txn_cnt - nonvote_txn_cnt - slot_info->vote_failed;
862 0 : } else {
863 0 : slot_info->vote_failed = ULONG_MAX;
864 0 : slot_info->vote_success = ULONG_MAX;
865 0 : slot_info->nonvote_success = ULONG_MAX;
866 0 : slot_info->nonvote_failed = ULONG_MAX;
867 0 : }
868 :
869 0 : slot_info->is_leader = is_leader;
870 0 : slot_info->transaction_fee = fd_bank_execution_fees_get( bank );
871 0 : slot_info->transaction_fee -= (slot_info->transaction_fee>>1); /* burn */
872 0 : slot_info->priority_fee = fd_bank_priority_fees_get( bank );
873 0 : slot_info->tips = fd_bank_tips_get( bank );
874 0 : slot_info->shred_cnt = fd_bank_shred_cnt_get( bank );
875 :
876 0 : FD_BASE58_ENCODE_32_BYTES( ctx->block_id_arr[ bank->data->idx ].latest_mr.uc, block_id_cstr );
877 0 : FD_BASE58_ENCODE_32_BYTES( fd_bank_bank_hash_query( bank )->uc, bank_hash_cstr );
878 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 ));
879 :
880 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() ) );
881 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 );
882 0 : }
883 :
884 : static void
885 : publish_slot_dead( fd_replay_tile_t * ctx,
886 : fd_stem_context_t * stem,
887 : ulong slot,
888 0 : fd_hash_t const * block_id ) {
889 0 : fd_replay_slot_dead_t * slot_dead = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
890 0 : slot_dead->slot = slot;
891 0 : slot_dead->block_id = *block_id;
892 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() ) );
893 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 );
894 0 : }
895 :
896 : static void
897 : publish_txn_executed( fd_replay_tile_t * ctx,
898 : fd_stem_context_t * stem,
899 0 : ulong txn_idx ) {
900 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
901 0 : fd_replay_txn_executed_t * txn_executed = fd_type_pun( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ) );
902 0 : *txn_executed->txn = *fd_sched_get_txn( ctx->sched, txn_idx );
903 0 : txn_executed->txn_err = txn_info->txn_err;
904 0 : txn_executed->is_committable = !!(txn_info->flags&FD_SCHED_TXN_IS_COMMITTABLE);
905 0 : txn_executed->is_fees_only = !!(txn_info->flags&FD_SCHED_TXN_IS_FEES_ONLY);
906 0 : txn_executed->tick_parsed = txn_info->tick_parsed;
907 0 : txn_executed->tick_sigverify_disp = txn_info->tick_sigverify_disp;
908 0 : txn_executed->tick_sigverify_done = txn_info->tick_sigverify_done;
909 0 : txn_executed->tick_exec_disp = txn_info->tick_exec_disp;
910 0 : txn_executed->tick_exec_done = txn_info->tick_exec_done;
911 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() ) );
912 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(*txn_executed), ctx->replay_out->chunk0, ctx->replay_out->wmark );
913 0 : }
914 :
915 : static void
916 : replay_block_finalize( fd_replay_tile_t * ctx,
917 : fd_stem_context_t * stem,
918 0 : fd_bank_t * bank ) {
919 0 : bank->data->last_transaction_finished_nanos = fd_log_wallclock();
920 :
921 0 : FD_TEST( !(bank->data->flags&FD_BANK_FLAGS_FROZEN) );
922 :
923 : /* Set poh hash in bank. */
924 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->data->idx );
925 0 : fd_bank_poh_set( bank, *poh );
926 :
927 : /* Set shred count in bank. */
928 0 : fd_bank_shred_cnt_set( bank, fd_sched_get_shred_cnt( ctx->sched, bank->data->idx ) );
929 :
930 : /* Do hashing and other end-of-block processing. */
931 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
932 :
933 : /* Copy out cost tracker fields before freezing */
934 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
935 0 : cost_tracker_snap( bank, slot_info );
936 :
937 : /* fetch identity / vote balance updates infrequently */
938 0 : ulong slot = fd_bank_slot_get( bank );
939 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank->data->idx } };
940 0 : slot_info->identity_balance = FD_UNLIKELY( slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
941 :
942 : /* Mark the bank as frozen. */
943 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
944 :
945 : /**********************************************************************/
946 : /* Bank hash comparison, and halt if there's a mismatch after replay */
947 : /**********************************************************************/
948 :
949 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
950 0 : FD_TEST( bank_hash );
951 :
952 : /* Must be last so we can measure completion time correctly, even
953 : though we could technically do this before the hash cmp and vote
954 : tower stuff. */
955 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */ );
956 :
957 0 : # if FD_HAS_FLATCC
958 : /* If enabled, dump the block to a file and reset the dumping
959 : context state */
960 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
961 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, ctx->accdb, ctx->dump_proto_ctx );
962 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
963 0 : }
964 0 : # endif
965 0 : }
966 :
967 : /**********************************************************************/
968 : /* Leader bank management */
969 : /**********************************************************************/
970 :
971 : static fd_bank_t *
972 : prepare_leader_bank( fd_replay_tile_t * ctx,
973 : ulong slot,
974 : long now,
975 : fd_hash_t const * parent_block_id,
976 0 : fd_stem_context_t * stem ) {
977 0 : long before = fd_log_wallclock();
978 :
979 : /* Make sure that we are not already leader. */
980 0 : FD_TEST( ctx->leader_bank->data==NULL );
981 :
982 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 );
983 0 : if( FD_UNLIKELY( !parent_ele ) ) {
984 0 : FD_BASE58_ENCODE_32_BYTES( parent_block_id->key, parent_block_id_b58 );
985 0 : FD_LOG_CRIT(( "invariant violation: parent bank index not found for merkle root %s", parent_block_id_b58 ));
986 0 : }
987 0 : ulong parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent_ele );
988 :
989 0 : fd_bank_t parent_bank[1];
990 0 : if( FD_UNLIKELY( !fd_banks_bank_query( parent_bank, ctx->banks, parent_bank_idx ) ) ) {
991 0 : FD_LOG_CRIT(( "invariant violation: parent bank not found for bank index %lu", parent_bank_idx ));
992 0 : }
993 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
994 :
995 0 : if( FD_UNLIKELY( !fd_banks_new_bank( ctx->leader_bank, ctx->banks, parent_bank_idx, now ) ) ) {
996 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
997 0 : }
998 :
999 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( ctx->leader_bank, ctx->banks, ctx->leader_bank->data->idx ) ) ) {
1000 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
1001 0 : }
1002 :
1003 0 : ctx->leader_bank->data->preparation_begin_nanos = before;
1004 :
1005 0 : fd_bank_slot_set( ctx->leader_bank, slot );
1006 0 : fd_bank_parent_slot_set( ctx->leader_bank, parent_slot );
1007 0 : ctx->leader_bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->data->txncache_fork_id );
1008 : /* prepare the funk transaction for the leader bank */
1009 0 : fd_funk_txn_xid_t xid = { .ul = { slot, ctx->leader_bank->data->idx } };
1010 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
1011 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
1012 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
1013 :
1014 0 : fd_bank_execution_fees_set( ctx->leader_bank, 0UL );
1015 0 : fd_bank_priority_fees_set( ctx->leader_bank, 0UL );
1016 0 : fd_bank_shred_cnt_set( ctx->leader_bank, 0UL );
1017 0 : fd_bank_tips_set( ctx->leader_bank, 0UL );
1018 0 : fd_bank_identity_vote_idx_set( ctx->leader_bank, ULONG_MAX );
1019 :
1020 : /* Update block height. */
1021 0 : fd_bank_block_height_set( ctx->leader_bank, fd_bank_block_height_get( ctx->leader_bank ) + 1UL );
1022 :
1023 0 : int is_epoch_boundary = 0;
1024 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1025 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, ctx->leader_bank, 1 );
1026 :
1027 0 : ulong max_tick_height;
1028 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( fd_bank_ticks_per_slot_get( parent_bank ), slot, &max_tick_height ) ) ) {
1029 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, fd_bank_ticks_per_slot_get( parent_bank ) ));
1030 0 : }
1031 0 : fd_bank_max_tick_height_set( ctx->leader_bank, max_tick_height );
1032 0 : fd_bank_tick_height_set( ctx->leader_bank, fd_bank_max_tick_height_get( parent_bank ) ); /* The parent's max tick height is our starting tick height. */
1033 :
1034 : /* Now that a bank has been created for the leader slot, increment the
1035 : reference count until we are done with the leader slot. */
1036 0 : ctx->leader_bank->data->refcnt++;
1037 :
1038 0 : return ctx->leader_bank;
1039 0 : }
1040 :
1041 : static inline void
1042 0 : maybe_switch_identity( fd_replay_tile_t * ctx ) {
1043 :
1044 0 : if( FD_LIKELY( fd_keyswitch_state_query( ctx->keyswitch )!=FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) return;
1045 :
1046 : /* Switch identity */
1047 :
1048 0 : FD_LOG_DEBUG(( "keyswitch: switching identity" ));
1049 :
1050 0 : memcpy( ctx->identity_pubkey, ctx->keyswitch->bytes, 32UL );
1051 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
1052 :
1053 : /* The next leader slot will be incorrect now that the identity has
1054 : switched. The next leader slot normally gets updated based on the
1055 : reset slot returned by tower. */
1056 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 ) );
1057 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
1058 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1059 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1060 0 : } else {
1061 0 : ctx->next_leader_tickcount = LONG_MAX;
1062 0 : }
1063 :
1064 0 : ctx->identity_vote_rooted = 0;
1065 0 : ctx->identity_idx++;
1066 0 : fd_vote_tracker_reset( ctx->vote_tracker );
1067 0 : }
1068 :
1069 : static void
1070 : fini_leader_bank( fd_replay_tile_t * ctx,
1071 0 : fd_stem_context_t * stem ) {
1072 :
1073 0 : FD_TEST( ctx->leader_bank->data!=NULL );
1074 0 : FD_TEST( ctx->is_leader );
1075 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen );
1076 0 : FD_TEST( ctx->recv_poh );
1077 :
1078 0 : ctx->leader_bank->data->last_transaction_finished_nanos = fd_log_wallclock();
1079 :
1080 0 : ulong curr_slot = fd_bank_slot_get( ctx->leader_bank );
1081 :
1082 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->data->idx, ctx->leader_bank->data->parent_idx, curr_slot );
1083 :
1084 0 : fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx );
1085 :
1086 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1087 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
1088 0 : fd_funk_txn_xid_t xid = { .ul = { curr_slot, ctx->leader_bank->data->idx } };
1089 0 : slot_info->identity_balance = FD_UNLIKELY( curr_slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
1090 :
1091 0 : fd_banks_mark_bank_frozen( ctx->banks, ctx->leader_bank );
1092 :
1093 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( ctx->leader_bank );
1094 0 : FD_TEST( bank_hash );
1095 :
1096 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
1097 :
1098 : /* The reference on the bank is finally no longer needed. */
1099 0 : ctx->leader_bank->data->refcnt--;
1100 :
1101 : /* We are no longer leader so we can clear the bank index we use for
1102 : being the leader. */
1103 0 : ctx->leader_bank->data = NULL;
1104 0 : ctx->recv_poh = 0;
1105 0 : ctx->is_leader = 0;
1106 :
1107 0 : maybe_switch_identity( ctx );
1108 0 : }
1109 :
1110 : static void
1111 : publish_root_advanced( fd_replay_tile_t * ctx,
1112 0 : fd_stem_context_t * stem ) {
1113 :
1114 0 : fd_bank_t bank[1];
1115 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, ctx->consensus_root_bank_idx ) ) ) {
1116 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank is NULL at bank index %lu", ctx->consensus_root_bank_idx ));
1117 0 : }
1118 :
1119 0 : if( ctx->rpc_enabled ) {
1120 0 : bank->data->refcnt++;
1121 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->data->idx, fd_bank_slot_get( bank ), bank->data->refcnt ));
1122 0 : }
1123 :
1124 : /* Increment the reference count on the consensus root bank to account
1125 : for the number of resolv tiles that are waiting on it. */
1126 0 : bank->data->refcnt += ctx->resolv_tile_cnt;
1127 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for resolv", bank->data->idx, fd_bank_slot_get( bank ), bank->data->refcnt ));
1128 :
1129 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1130 0 : msg->bank_idx = bank->data->idx;
1131 :
1132 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() ) );
1133 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 );
1134 0 : }
1135 :
1136 : /* init_funk performs pre-flight checks for the account database and
1137 : program cache. Ensures that the account database was set up
1138 : correctly by bootstrap components (e.g. genesis or snapshot loader).
1139 : Mirrors the account database's fork tree down to the program cache. */
1140 :
1141 : static void
1142 : init_funk( fd_replay_tile_t * ctx,
1143 0 : ulong bank_slot ) {
1144 : /* Ensure that the loaded bank root corresponds to the account
1145 : database's root. */
1146 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1147 0 : if( FD_UNLIKELY( !funk->shmem ) ) {
1148 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to database shared memory objects" ));
1149 0 : }
1150 0 : fd_funk_txn_xid_t const * accdb_pub = fd_funk_last_publish( funk );
1151 0 : if( FD_UNLIKELY( accdb_pub->ul[0]!=bank_slot ) ) {
1152 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb is at slot %lu, but chain state is at slot %lu\n"
1153 0 : "This is a bug in startup components.",
1154 0 : accdb_pub->ul[0], bank_slot ));
1155 0 : }
1156 0 : if( FD_UNLIKELY( fd_funk_last_publish_is_frozen( funk ) ) ) {
1157 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb fork graph is not clean.\n"
1158 0 : "The account database should only contain state for the root slot at this point,\n"
1159 0 : "but there are incomplete database transactions leftover.\n"
1160 0 : "This is a bug in startup components." ));
1161 0 : }
1162 :
1163 : /* The program cache tracks the account database's fork graph at all
1164 : times. Perform initial synchronization: pivot from funk 'root' (a
1165 : sentinel XID) to 'last publish' (the bootstrap root slot). */
1166 0 : if( FD_UNLIKELY( !ctx->progcache_admin->funk->shmem ) ) {
1167 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to program cache" ));
1168 0 : }
1169 0 : fd_progcache_clear( ctx->progcache_admin );
1170 :
1171 0 : fd_funk_txn_xid_t last_publish = fd_accdb_root_get( ctx->accdb_admin );
1172 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, fd_funk_root( ctx->progcache_admin->funk ), &last_publish );
1173 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, &last_publish );
1174 0 : }
1175 :
1176 : static void
1177 0 : init_after_snapshot( fd_replay_tile_t * ctx ) {
1178 : /* Now that the snapshot has been loaded in, we have to refresh the
1179 : stake delegations since the manifest does not contain the full set
1180 : of data required for the stake delegations. See
1181 : fd_stake_delegations.h for why this is required. */
1182 :
1183 0 : fd_bank_t bank[1];
1184 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) ) ) {
1185 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1186 0 : }
1187 :
1188 0 : fd_funk_txn_xid_t xid = { .ul = { fd_bank_slot_get( bank ), bank->data->idx } };
1189 0 : init_funk( ctx, fd_bank_slot_get( bank ) );
1190 :
1191 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
1192 :
1193 0 : fd_stake_delegations_refresh( root_delegations, ctx->accdb, &xid );
1194 :
1195 0 : fd_top_votes_t * top_votes = fd_bank_top_votes_modify( bank );
1196 :
1197 0 : fd_vote_stakes_t * vote_stakes = fd_bank_vote_stakes_locking_modify( bank );
1198 0 : ushort fork_idx = bank->data->vote_stakes_fork_id;
1199 :
1200 0 : uchar __attribute__((aligned(FD_VOTE_STAKES_ITER_ALIGN))) iter_mem[ FD_VOTE_STAKES_ITER_FOOTPRINT ];
1201 0 : for( fd_vote_stakes_iter_t * iter = fd_vote_stakes_fork_iter_init( vote_stakes, fork_idx, iter_mem );
1202 0 : !fd_vote_stakes_fork_iter_done( vote_stakes, fork_idx, iter );
1203 0 : fd_vote_stakes_fork_iter_next( vote_stakes, fork_idx, iter ) ) {
1204 0 : fd_pubkey_t pubkey;
1205 0 : fd_pubkey_t node_account_t_2;
1206 0 : ulong stake_t_2;
1207 0 : fd_vote_stakes_fork_iter_ele( vote_stakes, fork_idx, iter, &pubkey, NULL, &stake_t_2, NULL, &node_account_t_2 );
1208 :
1209 0 : int is_valid = 1;
1210 0 : fd_accdb_ro_t acc[1];
1211 0 : if( FD_UNLIKELY( !fd_accdb_open_ro( ctx->accdb, acc, &xid, &pubkey ) ) ) {
1212 0 : is_valid = 0;
1213 0 : } else if( FD_UNLIKELY( !fd_vsv_is_correct_size_and_initialized( acc->meta ) ) ) {
1214 0 : fd_accdb_close_ro( ctx->accdb, acc );
1215 0 : is_valid = 0;
1216 0 : }
1217 :
1218 0 : if( FD_LIKELY( is_valid ) ) {
1219 0 : fd_vote_block_timestamp_t last_vote = fd_vsv_get_vote_block_timestamp( fd_account_data( acc->meta ), acc->meta->dlen );
1220 0 : fd_top_votes_insert( top_votes, &pubkey, &node_account_t_2, stake_t_2, last_vote.slot, last_vote.timestamp );
1221 0 : fd_accdb_close_ro( ctx->accdb, acc );
1222 0 : } else {
1223 0 : fd_top_votes_insert( top_votes, &pubkey, &node_account_t_2, stake_t_2, 0UL, 0L );
1224 0 : fd_top_votes_invalidate( top_votes, &pubkey );
1225 0 : }
1226 0 : }
1227 :
1228 0 : fd_bank_vote_stakes_end_locking_modify( bank );
1229 :
1230 : /* After both snapshots have been loaded in, we can determine if we should
1231 : start distributing rewards. */
1232 :
1233 0 : fd_rewards_recalculate_partitioned_rewards( ctx->banks, bank, ctx->accdb, &xid, &ctx->runtime_stack, ctx->capture_ctx );
1234 :
1235 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1236 0 : if( FD_UNLIKELY( !snapshot_slot ) ) {
1237 : /* Genesis-specific setup. */
1238 : /* FIXME: This branch does not set up a new block exec ctx
1239 : properly. Needs to do whatever prepare_new_block_execution
1240 : does, but just hacking that in breaks stuff. */
1241 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1242 :
1243 0 : ulong hashcnt_per_slot = fd_bank_hashes_per_tick_get( bank ) * fd_bank_ticks_per_slot_get( bank );
1244 0 : fd_hash_t * poh = fd_bank_poh_modify( bank );
1245 0 : while( hashcnt_per_slot-- ) {
1246 0 : fd_sha256_hash( poh->hash, 32UL, poh->hash );
1247 0 : }
1248 :
1249 0 : int is_epoch_boundary = 0;
1250 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1251 0 : FD_TEST( !is_epoch_boundary );
1252 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
1253 :
1254 0 : snapshot_slot = 0UL;
1255 0 : }
1256 0 : }
1257 :
1258 : static inline int
1259 : maybe_become_leader( fd_replay_tile_t * ctx,
1260 0 : fd_stem_context_t * stem ) {
1261 0 : FD_TEST( ctx->is_booted );
1262 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;
1263 0 : if( FD_UNLIKELY( ctx->halt_leader ) ) return 0;
1264 0 : if( !ctx->supports_leader ) return 0;
1265 :
1266 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1267 0 : long now = fd_tickcount();
1268 0 : if( FD_LIKELY( now<ctx->next_leader_tickcount ) ) return 0;
1269 :
1270 : /* If a prior leader is still in the process of publishing their slot,
1271 : delay ours to let them finish ... unless they are so delayed that
1272 : we risk getting skipped by the leader following us. 1.2 seconds
1273 : is a reasonable default here, although any value between 0 and 1.6
1274 : seconds could be considered reasonable. This is arbitrary and
1275 : chosen due to intuition. */
1276 0 : if( FD_UNLIKELY( now<ctx->next_leader_tickcount+(long)(3.0*ctx->slot_duration_ticks) ) ) {
1277 0 : FD_TEST( ctx->reset_bank->data );
1278 :
1279 : /* TODO: Make the max_active_descendant calculation more efficient
1280 : by caching it in the bank structure and updating it as banks are
1281 : created and completed. */
1282 0 : ulong max_active_descendant = 0UL;
1283 0 : ulong child_idx = ctx->reset_bank->data->child_idx;
1284 0 : while( child_idx!=ULONG_MAX ) {
1285 0 : fd_bank_t child_bank[1];
1286 0 : fd_banks_bank_query( child_bank, ctx->banks, child_idx );
1287 0 : max_active_descendant = fd_ulong_max( max_active_descendant, fd_bank_slot_get( child_bank ) );
1288 0 : child_idx = child_bank->data->sibling_idx;
1289 0 : }
1290 :
1291 : /* If the max_active_descendant is >= next_leader_slot, we waited
1292 : too long and a leader after us started publishing to try and skip
1293 : us. Just start our leader slot immediately, we might win ... */
1294 0 : if( FD_LIKELY( max_active_descendant>=ctx->reset_slot && max_active_descendant<ctx->next_leader_slot ) ) {
1295 : /* If one of the leaders between the reset slot and our leader
1296 : slot is in the process of publishing (they have a descendant
1297 : bank that is in progress of being replayed), then keep waiting.
1298 : We probably wouldn't get a leader slot out before they
1299 : finished.
1300 :
1301 : Unless... we are past the deadline to start our slot by more
1302 : than 1.2 seconds, in which case we should probably start it to
1303 : avoid getting skipped by the leader behind us. */
1304 0 : return 0;
1305 0 : }
1306 0 : }
1307 :
1308 0 : long now_nanos = fd_log_wallclock();
1309 :
1310 0 : ctx->is_leader = 1;
1311 0 : ctx->recv_poh = 0;
1312 :
1313 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1314 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1315 :
1316 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1317 :
1318 : /* Acquires bank, sets up initial state, and refcnts it. */
1319 0 : fd_bank_t * bank = prepare_leader_bank( ctx, ctx->next_leader_slot, now_nanos, &ctx->reset_block_id, stem );
1320 0 : fd_funk_txn_xid_t xid = { .ul = { ctx->next_leader_slot, ctx->leader_bank->data->idx } };
1321 :
1322 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1323 0 : fd_pubkey_t tip_receiver_owner = {0};
1324 :
1325 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1326 0 : fd_acct_addr_t tip_payment_config[1];
1327 0 : fd_acct_addr_t tip_receiver[1];
1328 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, fd_bank_epoch_get( bank ), tip_payment_config, tip_receiver );
1329 :
1330 0 : fd_accdb_ro_t tip_config_acc[1];
1331 0 : if( FD_UNLIKELY( !fd_accdb_open_ro( ctx->accdb, tip_config_acc, &xid, tip_payment_config ) ) ) {
1332 : /* FIXME This should not crash the validator */
1333 0 : FD_BASE58_ENCODE_32_BYTES( tip_payment_config->b, tip_config_acc_b58 );
1334 0 : FD_LOG_CRIT(( "tip payment config account %s does not exist", tip_config_acc_b58 ));
1335 0 : }
1336 0 : ulong tip_cfg_sz = fd_accdb_ref_data_sz( tip_config_acc );
1337 0 : if( FD_UNLIKELY( tip_cfg_sz < sizeof(fd_bundle_crank_tip_payment_config_t) ) ) {
1338 : /* FIXME This should not crash the validator */
1339 0 : FD_LOG_HEXDUMP_CRIT(( "invalid tip payment config account data", fd_accdb_ref_data_const( tip_config_acc ), tip_cfg_sz ));
1340 0 : }
1341 0 : memcpy( config, fd_accdb_ref_data_const( tip_config_acc ), sizeof(fd_bundle_crank_tip_payment_config_t) );
1342 0 : fd_accdb_close_ro( ctx->accdb, tip_config_acc );
1343 :
1344 : /* It is possible that the tip receiver account does not exist yet
1345 : if it is the first time in an epoch. */
1346 0 : fd_accdb_ro_t tip_receiver_acc[1];
1347 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, tip_receiver_acc, &xid, tip_receiver ) ) ) {
1348 0 : tip_receiver_owner = FD_LOAD( fd_pubkey_t, fd_accdb_ref_owner( tip_receiver_acc ) );
1349 0 : fd_accdb_close_ro( ctx->accdb, tip_receiver_acc );
1350 0 : }
1351 0 : }
1352 :
1353 :
1354 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1355 0 : msg->slot = ctx->next_leader_slot;
1356 0 : msg->slot_start_ns = now_nanos;
1357 0 : msg->slot_end_ns = now_nanos+(long)ctx->slot_duration_nanos;
1358 0 : msg->bank = NULL;
1359 0 : msg->bank_idx = bank->data->idx;
1360 0 : msg->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1361 0 : msg->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1362 0 : msg->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)msg->ticks_per_slot);
1363 0 : msg->bundle->config[0] = config[0];
1364 0 : memcpy( msg->bundle->last_blockhash, fd_bank_poh_query( bank )->hash, sizeof(fd_hash_t) );
1365 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner.uc, sizeof(fd_pubkey_t) );
1366 :
1367 0 : if( FD_UNLIKELY( msg->hashcnt_per_tick==1UL ) ) {
1368 : /* Low power producer, maximum of one microblock per tick in the slot */
1369 0 : msg->max_microblocks_in_slot = msg->ticks_per_slot;
1370 0 : } else {
1371 : /* See the long comment in after_credit for this limit */
1372 0 : msg->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, msg->ticks_per_slot*(msg->hashcnt_per_tick-1UL) );
1373 0 : }
1374 :
1375 0 : msg->total_skipped_ticks = msg->ticks_per_slot*(ctx->next_leader_slot-ctx->reset_slot);
1376 0 : msg->epoch = fd_slot_to_epoch( fd_bank_epoch_schedule_query( bank ), ctx->next_leader_slot, NULL );
1377 :
1378 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
1379 :
1380 0 : msg->limits.slot_max_cost = ctx->larger_max_cost_per_block ? LARGER_MAX_COST_PER_BLOCK : cost_tracker->block_cost_limit;
1381 0 : msg->limits.slot_max_vote_cost = cost_tracker->vote_cost_limit;
1382 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1383 :
1384 0 : fd_bank_cost_tracker_end_locking_query( bank );
1385 :
1386 0 : if( FD_UNLIKELY( msg->ticks_per_slot+msg->total_skipped_ticks>USHORT_MAX ) ) {
1387 : /* There can be at most USHORT_MAX skipped ticks, because the
1388 : parent_offset field in the shred data is only 2 bytes wide. */
1389 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 ));
1390 0 : }
1391 :
1392 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() ) );
1393 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 );
1394 :
1395 0 : ctx->next_leader_slot = ULONG_MAX;
1396 0 : ctx->next_leader_tickcount = LONG_MAX;
1397 :
1398 0 : return 1;
1399 0 : }
1400 :
1401 : static void
1402 : process_poh_message( fd_replay_tile_t * ctx,
1403 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
1404 :
1405 0 : FD_TEST( ctx->is_booted );
1406 0 : FD_TEST( ctx->is_leader );
1407 0 : FD_TEST( ctx->leader_bank->data!=NULL );
1408 :
1409 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
1410 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
1411 :
1412 : /* Update the poh hash in the bank. We will want to maintain a refcnt
1413 : on the bank until we have recieved the block id for the block after
1414 : it has been shredded. */
1415 :
1416 0 : memcpy( fd_bank_poh_modify( ctx->leader_bank ), slot_ended->blockhash, sizeof(fd_hash_t) );
1417 :
1418 0 : ctx->recv_poh = 1;
1419 0 : }
1420 :
1421 : static void
1422 : publish_reset( fd_replay_tile_t * ctx,
1423 : fd_stem_context_t * stem,
1424 0 : fd_bank_t * bank ) {
1425 0 : if( FD_UNLIKELY( ctx->replay_out->idx==ULONG_MAX ) ) return;
1426 :
1427 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1428 0 : FD_TEST( block_hash );
1429 :
1430 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1431 :
1432 0 : reset->bank_idx = bank->data->idx;
1433 0 : reset->timestamp = fd_log_wallclock();
1434 0 : reset->completed_slot = fd_bank_slot_get( bank );
1435 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1436 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1437 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
1438 0 : fd_memcpy( reset->completed_block_id, ctx->reset_block_id.uc, sizeof(fd_hash_t) );
1439 0 : fd_memcpy( reset->completed_blockhash, block_hash->uc, sizeof(fd_hash_t) );
1440 :
1441 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1442 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
1443 : /* Low power producer, maximum of one microblock per tick in the slot */
1444 0 : reset->max_microblocks_in_slot = ticks_per_slot;
1445 0 : } else {
1446 : /* See the long comment in after_credit for this limit */
1447 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
1448 0 : }
1449 0 : reset->next_leader_slot = ctx->next_leader_slot;
1450 :
1451 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() ) );
1452 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 );
1453 0 : }
1454 :
1455 : static void
1456 : store_xinsert( fd_store_t * store,
1457 0 : fd_hash_t const * merkle_root ) {
1458 0 : fd_store_pool_t pool = {
1459 0 : .pool = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_mem_gaddr ),
1460 0 : .ele = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_ele_gaddr ),
1461 0 : .ele_max = store->fec_max
1462 0 : };
1463 0 : int err; fd_store_fec_t * fec = fd_store_pool_acquire( &pool, NULL, 1 /* blocking */, &err );
1464 0 : if( FD_UNLIKELY( err!=FD_POOL_SUCCESS ) ) FD_LOG_CRIT(( "store pool: %s", fd_store_pool_strerror( err ) ));
1465 0 : fec->key.merkle_root = *merkle_root;
1466 0 : fec->key.part_idx = 0;
1467 0 : fec->cmr = (fd_hash_t){ 0 };
1468 0 : fec->next = fd_store_pool_idx_null();
1469 0 : fec->data_sz = 0UL;
1470 :
1471 0 : FD_STORE_XLOCK_BEGIN( store ) {
1472 0 : fd_store_map_ele_insert( fd_wksp_laddr_fast( fd_store_wksp( store ), store->map_gaddr ), fec, pool.ele );
1473 0 : } FD_STORE_XLOCK_END;
1474 0 : }
1475 :
1476 : static void
1477 : boot_genesis( fd_replay_tile_t * ctx,
1478 : fd_stem_context_t * stem,
1479 0 : fd_genesis_meta_t const * meta ) {
1480 : /* If we are bootstrapping, we can't wait to wait for our identity
1481 : vote to be rooted as this creates a circular dependency. */
1482 0 : ctx->identity_vote_rooted = 1;
1483 :
1484 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
1485 0 : FD_TEST( meta->bootstrap && meta->has_lthash );
1486 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
1487 :
1488 0 : fd_bank_t bank[1];
1489 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) );
1490 0 : fd_funk_txn_xid_t xid = { .ul = { 0UL, FD_REPLAY_BOOT_BANK_IDX } };
1491 :
1492 : /* Do genesis-related processing in a non-rooted transaction */
1493 0 : fd_funk_txn_xid_t root_xid = { .ul = { LONG_MAX, LONG_MAX } };
1494 0 : fd_funk_txn_xid_t target_xid = { .ul = { 0UL, 0UL } };
1495 0 : fd_accdb_attach_child( ctx->accdb_admin, &root_xid, &target_xid );
1496 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, &xid, NULL, &meta->genesis_hash, &meta->lthash, ctx->genesis, genesis_blob, &ctx->runtime_stack );
1497 0 : fd_accdb_advance_root( ctx->accdb_admin, &target_xid );
1498 :
1499 0 : static const fd_txncache_fork_id_t txncache_root = { .val = USHORT_MAX };
1500 0 : bank->data->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, txncache_root );
1501 :
1502 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1503 0 : fd_txncache_finalize_fork( ctx->txncache, bank->data->txncache_fork_id, 0UL, block_hash->uc );
1504 :
1505 0 : publish_epoch_info( ctx, stem, bank, 0 );
1506 0 : publish_epoch_info( ctx, stem, bank, 1 );
1507 :
1508 : /* We call this after fd_runtime_read_genesis, which sets up the
1509 : slot_bank needed in blockstore_init. */
1510 0 : init_after_snapshot( ctx );
1511 :
1512 0 : ctx->published_root_slot = 0UL;
1513 0 : fd_sched_block_add_done( ctx->sched, bank->data->idx, ULONG_MAX, 0UL );
1514 :
1515 0 : fd_bank_block_height_set( bank, 1UL );
1516 :
1517 0 : ctx->consensus_root = ctx->initial_block_id;
1518 0 : ctx->consensus_root_slot = 0UL;
1519 0 : ctx->consensus_root_bank_idx = 0UL;
1520 0 : ctx->published_root_slot = 0UL;
1521 0 : ctx->published_root_bank_idx = 0UL;
1522 :
1523 0 : ctx->reset_slot = 0UL;
1524 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
1525 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1526 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1527 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1528 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1529 0 : } else {
1530 0 : ctx->next_leader_tickcount = LONG_MAX;
1531 0 : }
1532 :
1533 0 : ctx->is_booted = 1;
1534 0 : maybe_become_leader( ctx, stem );
1535 :
1536 0 : fd_hash_t initial_block_id = ctx->initial_block_id;
1537 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 */
1538 0 : fec->bank_idx = bank->data->idx;
1539 0 : fec->bank_seq = bank->data->bank_seq;
1540 0 : store_xinsert( ctx->store, &initial_block_id );
1541 :
1542 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1543 0 : block_id_ele->latest_mr = initial_block_id;
1544 0 : block_id_ele->slot = 0UL;
1545 :
1546 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1547 :
1548 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1549 0 : cost_tracker_snap( bank, slot_info );
1550 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1551 :
1552 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1553 0 : publish_root_advanced( ctx, stem );
1554 0 : publish_reset( ctx, stem, bank );
1555 0 : }
1556 :
1557 : static inline void
1558 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
1559 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->has_genesis_hash ) ) {
1560 0 : return;
1561 0 : }
1562 :
1563 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, hash_cstr );
1564 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
1565 : /* Map pyth-related clusters to unkwown. */
1566 0 : switch( cluster ) {
1567 0 : case FD_CLUSTER_PYTHNET:
1568 0 : case FD_CLUSTER_PYTHTEST:
1569 0 : cluster = FD_CLUSTER_UNKNOWN;
1570 0 : }
1571 :
1572 0 : if( FD_UNLIKELY( cluster!=ctx->cluster_type ) ) {
1573 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis hash of `%s` which means the cluster is %s "
1574 0 : "but the snapshot you loaded is for a different cluster %s. If you are trying to join the "
1575 0 : "%s cluster, you can delete the genesis.bin file and restart the node to download the correct "
1576 0 : "genesis file automatically.",
1577 0 : ctx->genesis_path,
1578 0 : hash_cstr,
1579 0 : fd_genesis_cluster_name( cluster ),
1580 0 : fd_genesis_cluster_name( ctx->cluster_type ),
1581 0 : fd_genesis_cluster_name( cluster ) ));
1582 0 : }
1583 0 : }
1584 :
1585 : static void
1586 : on_snapshot_message( fd_replay_tile_t * ctx,
1587 : fd_stem_context_t * stem,
1588 : ulong in_idx,
1589 : ulong chunk,
1590 0 : ulong sig ) {
1591 0 : ulong msg = fd_ssmsg_sig_message( sig );
1592 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
1593 : /* An end of message notification indicates the snapshot is loaded.
1594 : Replay is able to start executing from this point onwards. */
1595 : /* TODO: replay should finish booting. Could make replay a
1596 : state machine and set the state here accordingly. */
1597 0 : ctx->is_booted = 1;
1598 :
1599 0 : fd_bank_t bank[1];
1600 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ) ) ) {
1601 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1602 0 : }
1603 :
1604 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1605 :
1606 0 : fd_hash_t bank_hash = fd_bank_bank_hash_get( bank );
1607 0 : if( FD_UNLIKELY( ctx->wfs_enabled && memcmp( ctx->expected_bank_hash.uc, bank_hash.uc, sizeof(fd_hash_t) ) ) ) {
1608 0 : FD_BASE58_ENCODE_32_BYTES( ctx->expected_bank_hash.uc, expected_bank_hash_cstr );
1609 0 : FD_BASE58_ENCODE_32_BYTES( bank_hash.uc, actual_bank_hash_cstr );
1610 0 : FD_LOG_ERR(( "[consensus.wait_for_supermajority_with_bank_hash] expected_bank_hash=%s does not match snapshot slot"
1611 0 : "=%lu bank_hash=%s. If you are loading a snapshot from the network, check that the slot matches the "
1612 0 : "cluster restart slot. ", expected_bank_hash_cstr, snapshot_slot, actual_bank_hash_cstr ));
1613 0 : }
1614 0 : if( FD_UNLIKELY( ctx->wfs_enabled ) ) {
1615 0 : FD_LOG_NOTICE(( "waiting for supermajority at snapshot slot %lu", snapshot_slot ));
1616 0 : }
1617 :
1618 : /* FIXME: This is a hack because the block id of the snapshot slot
1619 : is not provided in the snapshot. A possible solution is to get
1620 : the block id of the snapshot slot from repair. */
1621 0 : fd_hash_t manifest_block_id = ctx->initial_block_id;
1622 :
1623 0 : fd_funk_txn_xid_t xid = { .ul = { snapshot_slot, FD_REPLAY_BOOT_BANK_IDX } };
1624 0 : fd_features_restore( bank, ctx->accdb, &xid );
1625 :
1626 : /* Typically, when we cross an epoch boundary during normal
1627 : operation, we publish the stake weights for the new epoch. But
1628 : since we are starting from a snapshot, we need to publish two
1629 : epochs worth of stake weights: the previous epoch (which is
1630 : needed for voting on the current epoch), and the current epoch
1631 : (which is needed for voting on the next epoch). */
1632 0 : publish_epoch_info( ctx, stem, bank, 0 );
1633 0 : publish_epoch_info( ctx, stem, bank, 1 );
1634 :
1635 0 : ctx->consensus_root = manifest_block_id;
1636 0 : ctx->consensus_root_slot = snapshot_slot;
1637 0 : ctx->consensus_root_bank_idx = 0UL;
1638 0 : ctx->published_root_slot = ctx->consensus_root_slot;
1639 0 : ctx->published_root_bank_idx = 0UL;
1640 :
1641 0 : ctx->reset_slot = snapshot_slot;
1642 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
1643 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1644 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1645 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1646 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1647 0 : } else {
1648 0 : ctx->next_leader_tickcount = LONG_MAX;
1649 0 : }
1650 :
1651 0 : fd_sched_block_add_done( ctx->sched, bank->data->idx, ULONG_MAX, snapshot_slot );
1652 0 : FD_TEST( bank->data->idx==0UL );
1653 :
1654 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1655 :
1656 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1657 0 : block_id_ele->latest_mr = manifest_block_id;
1658 0 : block_id_ele->slot = snapshot_slot;
1659 0 : block_id_ele->block_id_seen = 1;
1660 0 : block_id_ele->latest_fec_idx = 0U;
1661 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1662 :
1663 : /* We call this after fd_runtime_read_genesis, which sets up the
1664 : slot_bank needed in blockstore_init. */
1665 0 : init_after_snapshot( ctx );
1666 :
1667 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1668 0 : cost_tracker_snap( bank, slot_info );
1669 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1670 :
1671 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1672 0 : publish_root_advanced( ctx, stem );
1673 :
1674 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 */
1675 0 : fec->bank_idx = bank->data->idx;
1676 0 : fec->bank_seq = bank->data->bank_seq;
1677 0 : store_xinsert( ctx->store, &manifest_block_id );
1678 :
1679 0 : ctx->cluster_type = fd_bank_cluster_type_get( bank );
1680 :
1681 0 : maybe_verify_cluster_type( ctx );
1682 :
1683 0 : return;
1684 0 : }
1685 :
1686 0 : switch( msg ) {
1687 0 : case FD_SSMSG_MANIFEST_FULL:
1688 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
1689 : /* We may either receive a full snapshot manifest or an
1690 : incremental snapshot manifest. Note that this external message
1691 : id is only used temporarily because replay cannot yet receive
1692 : the firedancer-internal snapshot manifest message. */
1693 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1694 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 ));
1695 :
1696 0 : fd_bank_t bank[1];
1697 0 : fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
1698 0 : ctx->banks,
1699 0 : fd_banks_bank_query( bank, ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
1700 0 : &ctx->runtime_stack,
1701 0 : msg==FD_SSMSG_MANIFEST_INCREMENTAL );
1702 :
1703 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1704 0 : ctx->hard_forks_cnt = manifest->hard_forks_len;
1705 0 : for( ulong i=0UL; i<manifest->hard_forks_len; i++ ) {
1706 0 : ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
1707 0 : ctx->hard_forks_cnts[ i ] = manifest->hard_forks_cnts[ i ];
1708 0 : }
1709 0 : ctx->has_expected_genesis_timestamp = 1;
1710 0 : ctx->expected_genesis_timestamp = manifest->creation_time_millis;
1711 0 : break;
1712 0 : }
1713 0 : default: {
1714 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
1715 0 : return;
1716 0 : }
1717 0 : }
1718 :
1719 0 : return;
1720 0 : }
1721 :
1722 : static void
1723 : dispatch_task( fd_replay_tile_t * ctx,
1724 : fd_stem_context_t * stem,
1725 0 : fd_sched_task_t * task ) {
1726 :
1727 0 : switch( task->task_type ) {
1728 0 : case FD_SCHED_TT_TXN_EXEC: {
1729 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
1730 :
1731 0 : fd_bank_t bank[1];
1732 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->txn_exec->bank_idx ) );
1733 :
1734 0 : # if FD_HAS_FLATCC
1735 : /* Add the transaction to the block dumper if necessary. This
1736 : logic doesn't need to be fork-aware since it's only meant to
1737 : be used in backtest. */
1738 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
1739 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
1740 0 : }
1741 0 : # endif
1742 :
1743 0 : bank->data->refcnt++;
1744 :
1745 0 : if( FD_UNLIKELY( !bank->data->first_transaction_scheduled_nanos ) ) bank->data->first_transaction_scheduled_nanos = fd_log_wallclock();
1746 :
1747 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1748 0 : fd_execrp_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1749 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1750 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
1751 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
1752 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) {
1753 0 : exec_msg->capture_txn_idx = ctx->capture_ctx->current_txn_idx++;
1754 0 : }
1755 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() ) );
1756 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1757 0 : break;
1758 0 : }
1759 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1760 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
1761 :
1762 0 : fd_bank_t bank[1];
1763 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->txn_sigverify->bank_idx ) );
1764 0 : bank->data->refcnt++;
1765 :
1766 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1767 0 : fd_execrp_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1768 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1769 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
1770 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
1771 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 );
1772 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1773 0 : break;
1774 0 : };
1775 0 : case FD_SCHED_TT_POH_HASH: {
1776 0 : fd_bank_t bank[ 1 ];
1777 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->poh_hash->bank_idx ) );
1778 0 : bank->data->refcnt++;
1779 :
1780 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1781 0 : fd_execrp_poh_hash_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1782 0 : exec_msg->bank_idx = task->poh_hash->bank_idx;
1783 0 : exec_msg->mblk_idx = task->poh_hash->mblk_idx;
1784 0 : exec_msg->hashcnt = task->poh_hash->hashcnt;
1785 0 : memcpy( exec_msg->hash, task->poh_hash->hash, sizeof(fd_hash_t) );
1786 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 );
1787 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1788 0 : break;
1789 0 : };
1790 0 : default: {
1791 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1792 0 : }
1793 0 : }
1794 0 : }
1795 :
1796 : static void
1797 : mark_bank_dead( fd_replay_tile_t * ctx,
1798 : fd_stem_context_t * stem,
1799 0 : ulong bank_idx ) {
1800 0 : fd_bank_t bank[1];
1801 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
1802 0 : fd_banks_mark_bank_dead( ctx->banks, bank_idx );
1803 :
1804 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank_idx ];
1805 0 : if( block_id_ele->block_id_seen ) publish_slot_dead( ctx, stem, block_id_ele->slot, &block_id_ele->latest_mr );
1806 :
1807 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
1808 0 : if( FD_UNLIKELY( !fec ) ) return;
1809 0 : fec->bank_dead = 1;
1810 :
1811 0 : }
1812 :
1813 : /* Returns 1 if charge_busy. */
1814 : static int
1815 : replay( fd_replay_tile_t * ctx,
1816 0 : fd_stem_context_t * stem ) {
1817 :
1818 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
1819 :
1820 0 : int charge_busy = 0;
1821 0 : fd_sched_task_t task[ 1 ];
1822 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
1823 0 : return charge_busy; /* Nothing to execute or do. */
1824 0 : }
1825 :
1826 0 : charge_busy = 1;
1827 :
1828 0 : switch( task->task_type ) {
1829 0 : case FD_SCHED_TT_BLOCK_START: {
1830 0 : replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
1831 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX, NULL );
1832 0 : break;
1833 0 : }
1834 0 : case FD_SCHED_TT_BLOCK_END: {
1835 0 : fd_bank_t bank[1];
1836 0 : fd_banks_bank_query( bank, ctx->banks, task->block_end->bank_idx );
1837 0 : if( FD_LIKELY( !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) replay_block_finalize( ctx, stem, bank );
1838 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX, NULL );
1839 0 : break;
1840 0 : }
1841 0 : case FD_SCHED_TT_TXN_EXEC:
1842 0 : case FD_SCHED_TT_TXN_SIGVERIFY:
1843 0 : case FD_SCHED_TT_POH_HASH: {
1844 : /* Common case: we have a transaction we need to execute. */
1845 0 : dispatch_task( ctx, stem, task );
1846 0 : break;
1847 0 : }
1848 0 : case FD_SCHED_TT_MARK_DEAD: {
1849 0 : fd_bank_t bank[ 1 ];
1850 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, task->mark_dead->bank_idx ) );
1851 0 : mark_bank_dead( ctx, stem, task->mark_dead->bank_idx );
1852 0 : break;
1853 0 : }
1854 0 : default: {
1855 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1856 0 : }
1857 0 : }
1858 :
1859 0 : return charge_busy;
1860 0 : }
1861 :
1862 : static int
1863 : can_process_fec( fd_replay_tile_t * ctx,
1864 0 : int * evict_banks_out ) {
1865 0 : fd_reasm_fec_t * fec;
1866 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched )==0UL ) ) {
1867 0 : ctx->metrics.sched_full++;
1868 0 : return 0;
1869 0 : }
1870 :
1871 0 : if( FD_UNLIKELY( (fec = fd_reasm_peek( ctx->reasm ))==NULL ) ) {
1872 0 : ctx->metrics.reasm_empty++;
1873 0 : return 0;
1874 0 : }
1875 :
1876 0 : ctx->metrics.reasm_latest_slot = fec->slot;
1877 0 : ctx->metrics.reasm_latest_fec_idx = fec->fec_set_idx;
1878 :
1879 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 ) ) {
1880 : /* This guards against a rare race where we receive the FEC set for
1881 : the slot right after our leader rotation before we freeze the
1882 : bank for the last slot in our leader rotation. Leader slot
1883 : freezing happens only after if we've received the final PoH hash
1884 : from the poh tile as well as the final FEC set for the leader
1885 : slot. So the race happens when FEC sets are delivered and
1886 : processed sooner than the PoH hash, aka when the
1887 : poh=>shred=>replay path for the block id beats the poh=>replay
1888 : path for the poh hash. To mitigate this race, we must block on
1889 : ingesting the FEC set for the ensuing slot before the leader
1890 : bank freezes, because that would violate ordering invariants in
1891 : banks and sched. */
1892 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen );
1893 0 : FD_TEST( !ctx->recv_poh );
1894 0 : ctx->metrics.leader_bid_wait++;
1895 0 : return 0;
1896 0 : }
1897 :
1898 : /* If fec_set_idx is 0, we need a new bank for a new slot. Banks must
1899 : not be full in this case. */
1900 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) && fec->fec_set_idx==0 ) ) {
1901 0 : ctx->metrics.banks_full++;
1902 : /* We only want to evict banks if sched is drained and banks is no
1903 : longer making progress. Otherwise, sched might not release
1904 : refcnts on the frontier/leaf banks immediately, and the eviction
1905 : will have to wait for sched to drain anyways. */
1906 0 : if( FD_UNLIKELY( fd_sched_is_drained( ctx->sched ) ) ) *evict_banks_out = 1;
1907 0 : return 0;
1908 0 : }
1909 :
1910 : /* Otherwise, banks may not be full, so we can always create a new
1911 : bank if needed. Or, if banks are full, the current fec set's
1912 : ancestor (idx 0) already created a bank for this slot.*/
1913 0 : return 1;
1914 0 : }
1915 :
1916 : static void
1917 : insert_fec_set( fd_replay_tile_t * ctx,
1918 : fd_stem_context_t * stem,
1919 0 : fd_reasm_fec_t * reasm_fec ) {
1920 :
1921 0 : long now = fd_log_wallclock();
1922 :
1923 0 : reasm_fec->parent_bank_idx = fd_reasm_parent( ctx->reasm, reasm_fec )->bank_idx;
1924 :
1925 0 : fd_bank_t parent_bank[1];
1926 0 : FD_TEST( fd_banks_bank_query( parent_bank, ctx->banks, reasm_fec->parent_bank_idx ) );
1927 0 : reasm_fec->parent_bank_seq = parent_bank->data->bank_seq;
1928 :
1929 0 : if( FD_UNLIKELY( reasm_fec->fec_set_idx==0U ) ) {
1930 : /* If the first FEC set for a slot is observed, provision a new bank
1931 : if you are not the leader. Remove any stale block id map entry
1932 : and update the block id entry. */
1933 0 : fd_bank_t bank_[1];
1934 0 : fd_bank_t * bank = NULL;
1935 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) {
1936 0 : bank = ctx->leader_bank;
1937 0 : } else {
1938 0 : bank = fd_banks_new_bank( bank_, ctx->banks, reasm_fec->parent_bank_idx, now );
1939 0 : }
1940 :
1941 0 : reasm_fec->bank_idx = bank->data->idx;
1942 0 : reasm_fec->bank_seq = bank->data->bank_seq;
1943 :
1944 : /* At this point remove any stale entry in the block id map if it
1945 : exists and set the block id as not having been seen yet. This is
1946 : safe because we know that the old entry for this bank index has
1947 : already been pruned away. */
1948 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1949 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 ) ) {
1950 0 : FD_TEST( fd_block_id_map_ele_remove( ctx->block_id_map, &block_id_ele->latest_mr, NULL, ctx->block_id_arr ) );
1951 0 : }
1952 0 : block_id_ele->block_id_seen = 0;
1953 0 : block_id_ele->slot = reasm_fec->slot;
1954 0 : block_id_ele->latest_fec_idx = 0U;
1955 0 : block_id_ele->latest_mr = reasm_fec->key;
1956 0 : } else {
1957 : /* We are continuing to execute through a slot that we already have
1958 : a bank index for. */
1959 0 : reasm_fec->bank_idx = reasm_fec->parent_bank_idx;
1960 0 : reasm_fec->bank_seq = reasm_fec->parent_bank_seq;
1961 :
1962 0 : FD_TEST( reasm_fec->bank_idx!=ULONG_MAX );
1963 :
1964 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1965 0 : if( FD_UNLIKELY( block_id_ele->latest_fec_idx>=reasm_fec->fec_set_idx ) ) {
1966 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 ));
1967 0 : return;
1968 0 : }
1969 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1970 0 : block_id_ele->latest_mr = reasm_fec->key;
1971 0 : }
1972 :
1973 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) {
1974 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1975 0 : block_id_ele->block_id_seen = 1;
1976 0 : block_id_ele->latest_mr = reasm_fec->key;
1977 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1978 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1979 0 : }
1980 :
1981 : /* If we are the leader, we don't need to process the FEC set. */
1982 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) return;
1983 :
1984 : /* Forks form a partial ordering over FEC sets. The Repair tile
1985 : delivers FEC sets in-order per fork, but FEC set ordering across
1986 : forks is arbitrary */
1987 0 : fd_sched_fec_t sched_fec[ 1 ];
1988 :
1989 : # if DEBUG_LOGGING
1990 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
1991 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->cmr.key, cmr_b58 );
1992 : 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 ));
1993 : # endif
1994 :
1995 0 : sched_fec->shred_cnt = reasm_fec->data_cnt;
1996 0 : sched_fec->is_last_in_batch = !!reasm_fec->data_complete;
1997 0 : sched_fec->is_last_in_block = !!reasm_fec->slot_complete;
1998 0 : sched_fec->bank_idx = reasm_fec->bank_idx;
1999 0 : sched_fec->parent_bank_idx = reasm_fec->parent_bank_idx;
2000 0 : sched_fec->slot = reasm_fec->slot;
2001 0 : sched_fec->parent_slot = reasm_fec->slot - reasm_fec->parent_off;
2002 0 : sched_fec->is_first_in_block = reasm_fec->fec_set_idx==0U;
2003 0 : fd_funk_txn_xid_t const root = fd_accdb_root_get( ctx->accdb_admin );
2004 0 : fd_funk_txn_xid_copy( sched_fec->alut_ctx->xid, &root );
2005 0 : sched_fec->alut_ctx->accdb[0] = ctx->accdb[0];
2006 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
2007 :
2008 0 : fd_bank_t bank[1];
2009 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, sched_fec->bank_idx ) );
2010 0 : if( sched_fec->is_first_in_block ) {
2011 0 : bank->data->refcnt++;
2012 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for sched", bank->data->idx, sched_fec->slot, bank->data->refcnt ));
2013 0 : }
2014 :
2015 : /* Read FEC set from the store. This should happen before we try to
2016 : ingest the FEC set. This allows us to filter out frags that were
2017 : in-flight when we published away minority forks that the frags land
2018 : on. These frags would have no bank to execute against, because
2019 : their corresponding banks, or parent banks, have also been pruned
2020 : during publishing. A query against store will rightfully tell us
2021 : that the underlying data is not found, implying that this is for a
2022 : minority fork that we can safeljy ignore. */
2023 :
2024 0 : ulong wait = (ulong)fd_log_wallclock();
2025 0 : ulong work = wait;
2026 0 : FD_STORE_SLOCK_BEGIN( ctx->store ) {
2027 0 : ctx->metrics.store_query_acquire++;
2028 0 : work = (ulong)fd_log_wallclock();
2029 0 : fd_histf_sample( ctx->metrics.store_query_wait, work - wait );
2030 :
2031 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->store, &reasm_fec->key );
2032 0 : ctx->metrics.store_query_cnt++;
2033 0 : if( FD_UNLIKELY( !store_fec ) ) {
2034 :
2035 : /* The only case in which a FEC is not found in the store after
2036 : repair has notified is if the FEC was on a minority fork that
2037 : has already been published away. In this case we abandon the
2038 : entire slice because it is no longer relevant. */
2039 :
2040 0 : ctx->metrics.store_query_missing_cnt++;
2041 0 : ctx->metrics.store_query_missing_mr = reasm_fec->key.ul[0];
2042 0 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
2043 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 ));
2044 0 : return;
2045 0 : }
2046 0 : sched_fec->fec = store_fec;
2047 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. */
2048 0 : mark_bank_dead( ctx, stem, sched_fec->bank_idx );
2049 0 : }
2050 0 : } FD_STORE_SLOCK_END;
2051 :
2052 0 : ctx->metrics.store_query_release++;
2053 0 : fd_histf_sample( ctx->metrics.store_query_work, (ulong)fd_log_wallclock() - work );
2054 0 : }
2055 :
2056 : static void
2057 : process_fec_set( fd_replay_tile_t * ctx,
2058 : fd_stem_context_t * stem,
2059 0 : fd_reasm_fec_t * reasm_fec ) {
2060 :
2061 0 : fd_reasm_fec_t * parent = fd_reasm_parent( ctx->reasm, reasm_fec );
2062 0 : if( FD_UNLIKELY( !parent ) ) {
2063 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 ));
2064 0 : return;
2065 0 : }
2066 :
2067 0 : if( FD_UNLIKELY( parent->bank_dead ) ) {
2068 : /* Inherit the dead flag from the parent. If a dead slot is
2069 : completed, we publish the slot as dead. Don't insert FECs for
2070 : dead slots. */
2071 0 : reasm_fec->bank_dead = 1;
2072 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) publish_slot_dead( ctx, stem, reasm_fec->slot, &reasm_fec->key );
2073 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 ));
2074 0 : return;
2075 0 : }
2076 :
2077 : /* Standard case, the parent FEC has a valid corresponding bank. */
2078 0 : fd_bank_t parent_fec_bank[1];
2079 0 : if( FD_LIKELY( fd_banks_bank_query( parent_fec_bank, ctx->banks, parent->bank_idx ) &&
2080 0 : parent_fec_bank->data->bank_seq==parent->bank_seq ) ) {
2081 0 : insert_fec_set( ctx, stem, reasm_fec );
2082 0 : return;
2083 0 : }
2084 :
2085 : /* In the case the FEC doesn't directly connect, iterate up the reasm
2086 : tree to find the closest valid slot complete that corresponds to a
2087 : valid bank. */
2088 :
2089 : /* First keep track of all of the slot completes up to and including
2090 : the fec we want to insert off of. */
2091 0 : fd_reasm_fec_t * path[ FD_BANKS_MAX_BANKS ];
2092 0 : ulong path_cnt = 0UL;
2093 0 : path[ path_cnt++ ] = reasm_fec;
2094 :
2095 0 : for( fd_reasm_fec_t * curr = reasm_fec;; ) {
2096 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2097 0 : FD_TEST( curr );
2098 0 : if( FD_LIKELY( !curr->slot_complete ) ) continue;
2099 :
2100 0 : fd_bank_t curr_bank[1];
2101 0 : if( FD_LIKELY( fd_banks_bank_query( curr_bank, ctx->banks, curr->bank_idx ) && curr_bank->data->bank_seq==curr->bank_seq ) ) break;
2102 :
2103 0 : FD_TEST( path_cnt<=FD_BANKS_MAX_BANKS );
2104 0 : path[ path_cnt++ ] = curr;
2105 0 : }
2106 :
2107 0 : for( ulong i=path_cnt; i>0UL; i-- ) {
2108 0 : fd_reasm_fec_t * leaf = path[ i-1 ];
2109 :
2110 : /* If there's not capacity in the sched or banks, return early and
2111 : drop the FEC. We have inserted as much as we can for now. */
2112 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched ) < (leaf->fec_set_idx/FD_FEC_SHRED_CNT + 1) ) ) return;
2113 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) ) ) return;
2114 :
2115 : /* Gather all FECs for this slot; */
2116 0 : fd_reasm_fec_t * slot_fecs[ FD_FEC_BLK_MAX ];
2117 0 : fd_reasm_fec_t * curr = leaf;
2118 0 : for(;;) {
2119 0 : slot_fecs[ curr->fec_set_idx/FD_FEC_SHRED_CNT ] = curr;
2120 0 : if( curr->fec_set_idx==0U ) break;
2121 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2122 0 : FD_TEST( curr );
2123 0 : }
2124 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 ));
2125 :
2126 0 : for( ulong j=0UL; j<=leaf->fec_set_idx/FD_FEC_SHRED_CNT; j++ ) {
2127 0 : insert_fec_set( ctx, stem, slot_fecs[ j ] );
2128 0 : }
2129 0 : }
2130 0 : }
2131 :
2132 : /* accdb_advance_root moves account records from the unrooted to the
2133 : rooted database. */
2134 :
2135 : static inline ulong
2136 0 : accdb_root_op_total( fd_replay_tile_t const * ctx ) {
2137 0 : return ctx->accdb_admin->base.root_cnt +
2138 0 : ctx->accdb_admin->base.reclaim_cnt;
2139 0 : }
2140 :
2141 : static void
2142 : accdb_advance_root( fd_replay_tile_t * ctx,
2143 : ulong slot,
2144 0 : ulong bank_idx ) {
2145 0 : fd_funk_txn_xid_t xid = { .ul[0] = slot, .ul[1] = bank_idx };
2146 0 : FD_LOG_DEBUG(( "advancing root to slot=%lu", slot ));
2147 :
2148 0 : long rooted_accounts = -(long)accdb_root_op_total( ctx );
2149 0 : long root_accounts_dt = -fd_tickcount();
2150 0 : fd_accdb_advance_root( ctx->accdb_admin, &xid );
2151 0 : rooted_accounts += (long)accdb_root_op_total( ctx );
2152 0 : root_accounts_dt += fd_tickcount();
2153 0 : fd_histf_sample( ctx->metrics.root_slot_dur, (ulong)root_accounts_dt );
2154 0 : fd_histf_sample( ctx->metrics.root_account_dur, (ulong)root_accounts_dt / (ulong)fd_long_max( rooted_accounts, 1L ) );
2155 :
2156 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, &xid );
2157 0 : }
2158 :
2159 : static int
2160 0 : advance_published_root( fd_replay_tile_t * ctx ) {
2161 :
2162 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 );
2163 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2164 0 : FD_BASE58_ENCODE_32_BYTES( ctx->consensus_root.key, consensus_root_b58 );
2165 0 : FD_LOG_CRIT(( "invariant violation: block id ele not found for consensus root %s", consensus_root_b58 ));
2166 0 : }
2167 0 : ulong target_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2168 :
2169 : /* If the identity vote has been seen on a bank that should be rooted,
2170 : then we are now ready to produce blocks. */
2171 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2172 0 : fd_bank_t root_bank[1];
2173 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 ));
2174 0 : if( fd_bank_identity_vote_idx_get( root_bank )==ctx->identity_idx ) ctx->identity_vote_rooted = 1;
2175 0 : }
2176 :
2177 0 : ulong advanceable_root_idx = ULONG_MAX;
2178 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank_idx, &advanceable_root_idx ) ) ) {
2179 0 : ctx->metrics.storage_root_behind++;
2180 0 : return 0;
2181 0 : }
2182 :
2183 0 : fd_bank_t bank[1];
2184 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, advanceable_root_idx ) );
2185 :
2186 0 : if( FD_UNLIKELY( advanceable_root_idx >= ctx->block_id_len ) ) {
2187 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele out of bounds [0, %lu) index %lu", ctx->block_id_len, advanceable_root_idx ));
2188 0 : }
2189 0 : fd_block_id_ele_t * advanceable_root_ele = &ctx->block_id_arr[ advanceable_root_idx ];
2190 :
2191 0 : ulong advanceable_root_slot = fd_bank_slot_get( bank );
2192 0 : accdb_advance_root( ctx, advanceable_root_slot, bank->data->idx );
2193 :
2194 0 : fd_txncache_advance_root( ctx->txncache, bank->data->txncache_fork_id );
2195 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
2196 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
2197 :
2198 : /* Set metrics pointers. */
2199 :
2200 :
2201 : /* Reasm also prunes from the store during its publish. */
2202 :
2203 0 : fd_reasm_publish( ctx->reasm, &advanceable_root_ele->latest_mr, ctx->store );
2204 :
2205 0 : ctx->published_root_slot = advanceable_root_slot;
2206 0 : ctx->published_root_bank_idx = advanceable_root_idx;
2207 :
2208 0 : return 1;
2209 0 : }
2210 :
2211 : static void
2212 : after_credit( fd_replay_tile_t * ctx,
2213 : fd_stem_context_t * stem,
2214 : int * opt_poll_in,
2215 0 : int * charge_busy ) {
2216 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->wfs_complete ) ) return;
2217 :
2218 0 : if( FD_UNLIKELY( maybe_become_leader( ctx, stem ) ) ) {
2219 0 : *charge_busy = 1;
2220 0 : *opt_poll_in = 0;
2221 0 : return;
2222 0 : }
2223 :
2224 : /* If we are leader, we can only unbecome the leader iff we have
2225 : received the poh hash from the poh tile and block id from reasm.
2226 : We have to do an additional check against the slot of the leader
2227 : bank because we lazily remove entries from the block id arr. */
2228 0 : if( FD_UNLIKELY( ctx->is_leader &&
2229 0 : ctx->recv_poh &&
2230 0 : ctx->block_id_arr[ ctx->leader_bank->data->idx ].block_id_seen &&
2231 0 : ctx->block_id_arr[ ctx->leader_bank->data->idx ].slot==fd_bank_slot_get( ctx->leader_bank ) ) ) {
2232 :
2233 0 : fini_leader_bank( ctx, stem );
2234 0 : *charge_busy = 1;
2235 0 : *opt_poll_in = 0;
2236 0 : return;
2237 0 : }
2238 :
2239 0 : ulong bank_idx;
2240 0 : while( (bank_idx=fd_sched_pruned_block_next( ctx->sched ))!=ULONG_MAX ) {
2241 0 : fd_bank_t bank[1];
2242 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
2243 0 : bank->data->refcnt--;
2244 0 : FD_LOG_DEBUG(( "bank (idx=%lu) refcnt decremented to %lu for sched", bank->data->idx, bank->data->refcnt ));
2245 0 : }
2246 :
2247 : /* If the published_root is not caught up to the consensus root, then
2248 : we should try to advance the published root. */
2249 0 : if( FD_UNLIKELY( ctx->consensus_root_bank_idx!=ctx->published_root_bank_idx && advance_published_root( ctx ) ) ) {
2250 0 : *charge_busy = 1;
2251 0 : *opt_poll_in = 0;
2252 0 : return;
2253 0 : }
2254 :
2255 0 : if( FD_UNLIKELY( fd_banks_prune_dead_banks( ctx->banks ) ) ) {
2256 : // FIXME: anything pruned from banks should also be pruned from txncache and accdb
2257 0 : *charge_busy = 1;
2258 0 : *opt_poll_in = 0;
2259 0 : return;
2260 0 : }
2261 :
2262 : /* if reasm evicted is set, publish starting from reasm_evicted down
2263 : to the leaf node to repair so repair can re-request for it */
2264 :
2265 0 : if( FD_UNLIKELY( ctx->reasm_evicted ) ) {
2266 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 };
2267 0 : fd_memcpy( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ), &evicted, sizeof(fd_replay_fec_evicted_t) );
2268 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() ) );
2269 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 );
2270 :
2271 : /* eviction policy only evicts chains of nodes until there is a
2272 : fork, so guaranteed that the evict path is always the left-child */
2273 0 : fd_reasm_pool_release( ctx->reasm, ctx->reasm_evicted );
2274 0 : ctx->reasm_evicted = fd_reasm_child( ctx->reasm, ctx->reasm_evicted ); /* indexes into pool, safe to use */
2275 :
2276 0 : *charge_busy = 1;
2277 0 : *opt_poll_in = 0;
2278 0 : return;
2279 0 : }
2280 :
2281 : /* If the reassembler has a fec that is ready, we should process it
2282 : and pass it to the scheduler. */
2283 0 : int evict_banks = 0;
2284 0 : if( FD_LIKELY( can_process_fec( ctx, &evict_banks ) ) ) {
2285 0 : fd_reasm_fec_t * fec = fd_reasm_pop( ctx->reasm );
2286 0 : process_fec_set( ctx, stem, fec );
2287 0 : *charge_busy = 1;
2288 0 : *opt_poll_in = 0;
2289 0 : return;
2290 0 : }
2291 :
2292 0 : if( FD_UNLIKELY( evict_banks ) ) {
2293 0 : FD_LOG_WARNING(( "banks are full and partially executed frontier banks are being evicted" ));
2294 0 : ulong frontier_cnt = 0UL;
2295 0 : ulong frontier_indices[ FD_BANKS_MAX_BANKS ];
2296 0 : fd_banks_get_frontier( ctx->banks, frontier_indices, &frontier_cnt );
2297 :
2298 : /* Mark all frontier banks as dead. As refcnts on said banks are
2299 : drained, they will be pruned away. */
2300 0 : for( ulong i=0UL; i<frontier_cnt; i++ ) {
2301 0 : fd_bank_t bank[1];
2302 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, frontier_indices[i] ) );
2303 0 : if( FD_UNLIKELY( ctx->is_leader && frontier_indices[i]==ctx->leader_bank->data->idx ) ) continue;
2304 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2305 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2306 :
2307 : /* evict it from reasm - we can guarantee this is a leaf because
2308 : no new bank is allocated until a slot boundary. If a fec has
2309 : children that are of a different slot, then it would never be
2310 : evicted from banks because that means the bank finished
2311 : executing on that slot. */
2312 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->data->idx ];
2313 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
2314 0 : FD_TEST( fec && fec->child == ULONG_MAX );
2315 0 : ctx->reasm_evicted = fd_reasm_remove( ctx->reasm, fec, ctx->store );
2316 0 : }
2317 0 : }
2318 :
2319 0 : *charge_busy = replay( ctx, stem );
2320 0 : *opt_poll_in = !*charge_busy;
2321 0 : }
2322 :
2323 : static int
2324 : before_frag( fd_replay_tile_t * ctx,
2325 : ulong in_idx,
2326 : ulong seq FD_PARAM_UNUSED,
2327 0 : ulong sig ) {
2328 :
2329 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_GOSSIP_OUT && sig!=FD_GOSSIP_UPDATE_TAG_WFS_DONE ) ) return 1;
2330 0 : return 0;
2331 0 : }
2332 :
2333 : static void
2334 : process_exec_task_done( fd_replay_tile_t * ctx,
2335 : fd_stem_context_t * stem,
2336 : fd_execrp_task_done_msg_t * msg,
2337 0 : ulong sig ) {
2338 :
2339 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
2340 :
2341 0 : fd_bank_t bank[1];
2342 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, msg->bank_idx ) );
2343 0 : FD_TEST( bank->data );
2344 0 : bank->data->refcnt--;
2345 :
2346 0 : switch( sig>>32 ) {
2347 0 : case FD_EXECRP_TT_TXN_EXEC: {
2348 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2349 : /* Query the txn signature against our recently generated vote
2350 : txn signatures. If the query is successful, then we have
2351 : seen our own vote transaction land and this should be marked
2352 : in the bank. We go through this exercise until we've seen
2353 : our vote rooted. */
2354 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, msg->txn_exec->txn_idx );
2355 :
2356 0 : fd_pubkey_t * identity_pubkey_out = NULL;
2357 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 ) ) {
2358 0 : fd_bank_identity_vote_idx_set( bank, ctx->identity_idx );
2359 0 : }
2360 0 : }
2361 0 : if( FD_UNLIKELY( !msg->txn_exec->is_committable && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2362 : /* Every transaction in a valid block has to execute.
2363 : Otherwise, we should mark the block as dead. */
2364 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2365 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2366 0 : }
2367 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2368 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2369 0 : }
2370 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, msg->txn_exec->txn_idx, exec_tile_idx, NULL );
2371 0 : FD_TEST( res==0 );
2372 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, msg->txn_exec->txn_idx );
2373 0 : txn_info->flags |= FD_SCHED_TXN_EXEC_DONE;
2374 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. */
2375 0 : txn_info->txn_err = msg->txn_exec->txn_err;
2376 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_committable, FD_SCHED_TXN_IS_COMMITTABLE, 0UL );
2377 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_fees_only, FD_SCHED_TXN_IS_FEES_ONLY, 0UL );
2378 0 : }
2379 0 : if( FD_UNLIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) { /* UNLIKELY because generally exec happens before sigverify. */
2380 0 : publish_txn_executed( ctx, stem, msg->txn_exec->txn_idx );
2381 0 : }
2382 0 : break;
2383 0 : }
2384 0 : case FD_EXECRP_TT_TXN_SIGVERIFY: {
2385 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, msg->txn_sigverify->txn_idx );
2386 0 : txn_info->flags |= FD_SCHED_TXN_SIGVERIFY_DONE;
2387 0 : if( FD_UNLIKELY( msg->txn_sigverify->err ) ) {
2388 0 : txn_info->txn_err = FD_RUNTIME_TXN_ERR_SIGNATURE_FAILURE;
2389 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_COMMITTABLE;
2390 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_FEES_ONLY;
2391 0 : }
2392 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2393 : /* Every transaction in a valid block has to sigverify.
2394 : Otherwise, we should mark the block as dead. Also freeze the
2395 : bank if possible. */
2396 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2397 0 : fd_sched_block_abandon( ctx->sched, bank->data->idx );
2398 0 : }
2399 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2400 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2401 0 : }
2402 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, msg->txn_sigverify->txn_idx, exec_tile_idx, NULL );
2403 0 : FD_TEST( res==0 );
2404 0 : if( FD_LIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) {
2405 0 : publish_txn_executed( ctx, stem, msg->txn_exec->txn_idx );
2406 0 : }
2407 0 : break;
2408 0 : }
2409 0 : case FD_EXECRP_TT_POH_HASH: {
2410 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_POH_HASH, ULONG_MAX, exec_tile_idx, msg->poh_hash );
2411 0 : if( FD_UNLIKELY( res<0 && !(bank->data->flags&FD_BANK_FLAGS_DEAD) ) ) {
2412 0 : mark_bank_dead( ctx, stem, bank->data->idx );
2413 0 : }
2414 0 : if( FD_UNLIKELY( (bank->data->flags&FD_BANK_FLAGS_DEAD) && bank->data->refcnt==0UL ) ) {
2415 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
2416 0 : }
2417 0 : break;
2418 0 : }
2419 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
2420 0 : }
2421 :
2422 : /* Reference counter just decreased, and an exec tile just got freed
2423 : up. If there's a need to be more aggressively pruning, we could
2424 : check here if more slots just became publishable and publish. Not
2425 : publishing here shouldn't bloat the fork tree too much though. We
2426 : mark minority forks dead as soon as we can, and execution dispatch
2427 : stops on dead blocks. So shortly afterwards, dead blocks should be
2428 : eligible for pruning as in-flight transactions retire from the
2429 : execution pipeline. */
2430 :
2431 0 : }
2432 :
2433 : static void
2434 : process_tower_slot_done( fd_replay_tile_t * ctx,
2435 : fd_stem_context_t * stem,
2436 : fd_tower_slot_done_t const * msg,
2437 0 : ulong seq ) {
2438 0 : fd_bank_t replay_bank[1];
2439 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 ));
2440 0 : replay_bank->data->refcnt--;
2441 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 ));
2442 :
2443 0 : ctx->reset_block_id = msg->reset_block_id;
2444 0 : ctx->reset_slot = msg->reset_slot;
2445 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
2446 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 ) );
2447 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
2448 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
2449 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
2450 0 : } else {
2451 0 : ctx->next_leader_tickcount = LONG_MAX;
2452 0 : }
2453 :
2454 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 );
2455 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2456 0 : FD_BASE58_ENCODE_32_BYTES( msg->reset_block_id.key, reset_block_id_b58 );
2457 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 ));
2458 0 : }
2459 0 : ulong reset_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2460 :
2461 0 : fd_bank_t bank[1];
2462 0 : if( FD_UNLIKELY( !fd_banks_bank_query( bank, ctx->banks, reset_bank_idx ) ) ) {
2463 0 : FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", reset_bank_idx ));
2464 0 : }
2465 :
2466 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) FD_TEST( msg->root_slot<=msg->reset_slot );
2467 0 : fd_memcpy( ctx->reset_bank, bank, sizeof(fd_bank_t) );
2468 :
2469 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2470 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2471 :
2472 0 : reset->bank_idx = bank->data->idx;
2473 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2474 0 : reset->completed_slot = ctx->reset_slot;
2475 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
2476 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2477 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
2478 :
2479 0 : fd_memcpy( reset->completed_block_id, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
2480 :
2481 0 : fd_blockhashes_t const * block_hash_queue = fd_bank_block_hash_queue_query( bank );
2482 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2483 0 : FD_TEST( last_hash );
2484 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2485 :
2486 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2487 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
2488 : /* Low power producer, maximum of one microblock per tick in the slot */
2489 0 : reset->max_microblocks_in_slot = ticks_per_slot;
2490 0 : } else {
2491 : /* See the long comment in after_credit for this limit */
2492 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
2493 0 : }
2494 0 : reset->next_leader_slot = ctx->next_leader_slot;
2495 :
2496 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() ) );
2497 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 );
2498 0 : }
2499 :
2500 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 ));
2501 0 : maybe_become_leader( ctx, stem );
2502 :
2503 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) {
2504 :
2505 0 : FD_TEST( msg->root_slot>=ctx->consensus_root_slot );
2506 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 );
2507 0 : FD_TEST( block_id_ele );
2508 0 : ctx->consensus_root_slot = msg->root_slot;
2509 0 : ctx->consensus_root = msg->root_block_id;
2510 0 : ctx->consensus_root_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2511 :
2512 0 : publish_root_advanced( ctx, stem );
2513 :
2514 0 : fd_sched_root_notify( ctx->sched, ctx->consensus_root_bank_idx );
2515 0 : }
2516 :
2517 0 : ulong distance = 0UL;
2518 0 : fd_bank_t * parent = bank;
2519 0 : while( parent ) {
2520 0 : if( FD_UNLIKELY( parent->data->idx==ctx->consensus_root_bank_idx ) ) break;
2521 0 : parent = fd_banks_get_parent( bank, ctx->banks, parent );
2522 0 : distance++;
2523 0 : }
2524 :
2525 0 : FD_MGAUGE_SET( REPLAY, ROOT_DISTANCE, distance );
2526 0 : }
2527 :
2528 : static void
2529 : process_fec_complete( fd_replay_tile_t * ctx,
2530 : fd_stem_context_t * stem,
2531 0 : uchar const * shred_buf ) {
2532 0 : fd_shred_t const * shred = (fd_shred_t const *)fd_type_pun_const( shred_buf );
2533 :
2534 0 : fd_hash_t const * merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ );
2535 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) );
2536 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) );
2537 :
2538 0 : int data_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_DATA_COMPLETE );
2539 0 : int slot_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_SLOT_COMPLETE );
2540 :
2541 0 : if( FD_UNLIKELY( shred->slot - shred->data.parent_off == fd_reasm_slot0( ctx->reasm ) && shred->fec_set_idx == 0) ) {
2542 0 : chained_merkle_root = &fd_reasm_root( ctx->reasm )->key;
2543 0 : }
2544 :
2545 0 : if( FD_UNLIKELY( fd_reasm_query( ctx->reasm, merkle_root ) ) ) return;
2546 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 );
2547 :
2548 0 : if( FD_UNLIKELY( !fec ) ) {
2549 : /* reasm failed to insert. We don't want to just put this back on
2550 : the returnable_frag queue because it's unclear whether this FEC
2551 : is truly something we want to process. Therefore our best option
2552 : is to punt it and "go around." reasm_insert populates it's last
2553 : pool element with the data of the failed insert, so we make sure
2554 : to publish the failed insert data to repair in after_credit. */
2555 0 : return;
2556 0 : }
2557 :
2558 0 : if( FD_UNLIKELY( ctx->reasm_evicted && ctx->reasm_evicted->bank_idx != ULONG_MAX ) ) {
2559 0 : mark_bank_dead( ctx, stem, ctx->reasm_evicted->bank_idx );
2560 0 : fd_sched_block_abandon( ctx->sched, ctx->reasm_evicted->bank_idx );
2561 0 : }
2562 0 : }
2563 :
2564 : static void
2565 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
2566 0 : fd_bank_t bank[1];
2567 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, bank_idx ) );
2568 0 : bank->data->refcnt--;
2569 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for resolv", bank->data->idx, fd_bank_slot_get( bank ), bank->data->refcnt ));
2570 0 : }
2571 :
2572 : static void
2573 : process_vote_txn_sent( fd_replay_tile_t * ctx,
2574 0 : fd_txn_m_t * txnm ) {
2575 : /* The send tile has signed and sent a vote. Add this vote to the
2576 : vote tracker. We go through this exercise until the client has
2577 : seen a vote corresponding to the current identity rooted. */
2578 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2579 0 : uchar * payload = (uchar *)txnm + sizeof(fd_txn_m_t);
2580 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
2581 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
2582 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
2583 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
2584 0 : }
2585 : /* The identity of the validator that the signed the vote will
2586 : always be the first signer in the vote transaction. */
2587 0 : fd_pubkey_t * vote_identity = fd_type_pun( payload+txn->acct_addr_off );
2588 0 : fd_vote_tracker_insert( ctx->vote_tracker, vote_identity, fd_type_pun_const( payload+txn->signature_off ) );
2589 0 : }
2590 0 : }
2591 :
2592 : static inline void
2593 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
2594 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
2595 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
2596 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
2597 0 : }
2598 0 : }
2599 :
2600 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_forks_cnt!=ULONG_MAX && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
2601 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
2602 :
2603 0 : ushort actual_shred_version = compute_shred_version( ctx->genesis_hash->uc, ctx->hard_forks, ctx->hard_forks_cnts, ctx->hard_forks_cnt );
2604 :
2605 0 : if( FD_UNLIKELY( expected_shred_version!=actual_shred_version ) ) {
2606 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, genesis_hash_b58 );
2607 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` combined with the hard_forks from the loaded snapshot have produced "
2608 0 : "a shred version of %hu but the entrypoint you connected to on boot reported a shred version of %hu. "
2609 0 : "This likely means that the genesis.bin file you have is for a different cluster than the one you "
2610 0 : "are trying to connect to, you can delete it and restart the node to download the correct genesis "
2611 0 : "file automatically.", ctx->genesis_path, actual_shred_version, expected_shred_version ));
2612 0 : }
2613 0 : }
2614 0 : }
2615 :
2616 : static inline void
2617 0 : maybe_verify_genesis_timestamp( fd_replay_tile_t * ctx ) {
2618 0 : if( FD_LIKELY( !ctx->has_expected_genesis_timestamp || !ctx->has_genesis_timestamp ) ) return;
2619 0 : if( FD_LIKELY( ctx->genesis_timestamp==ctx->expected_genesis_timestamp ) ) return;
2620 :
2621 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis timestamp of %lu but the snapshot you loaded has a genesis "
2622 0 : "timestamp of %lu. This either means that the genesis.bin file you have is for a different cluster than "
2623 0 : "the one you are trying to connect to, or you have loaded a snapshot for the wrong cluster. In either "
2624 0 : "case, you can delete the problematic file and restart the node to download the correct one automatically.",
2625 0 : ctx->genesis_path, ctx->genesis_timestamp, ctx->expected_genesis_timestamp ));
2626 0 : }
2627 :
2628 : static void
2629 : process_tower_optimistic_confirmed( fd_replay_tile_t * ctx,
2630 : fd_stem_context_t * stem,
2631 0 : fd_tower_slot_confirmed_t const * msg ) {
2632 :
2633 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 );
2634 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2635 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_b58 );
2636 0 : FD_LOG_WARNING(( "missing bank for confirmed block_id: %s level %d", block_id_b58, msg->level ));
2637 0 : return;
2638 0 : }
2639 :
2640 0 : ulong bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2641 0 : fd_bank_t bank_[1]; fd_bank_t * bank = fd_banks_bank_query( bank_, ctx->banks, bank_idx );
2642 :
2643 :
2644 0 : if( FD_UNLIKELY( !bank ) ) {
2645 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_cstr );
2646 0 : FD_LOG_WARNING(( "failed to query optimistically confirmed bank for block id %s", block_id_cstr ));
2647 0 : return;
2648 0 : }
2649 :
2650 0 : if( ctx->rpc_enabled ) {
2651 0 : bank->data->refcnt++;
2652 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->data->idx, fd_bank_slot_get( bank ), bank->data->refcnt ));
2653 0 : }
2654 :
2655 0 : fd_replay_oc_advanced_t * replay_msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2656 0 : replay_msg->bank_idx = bank_idx;
2657 0 : replay_msg->slot = msg->slot;
2658 :
2659 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() ) );
2660 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 );
2661 0 : }
2662 :
2663 : static inline int
2664 : returnable_frag( fd_replay_tile_t * ctx,
2665 : ulong in_idx,
2666 : ulong seq,
2667 : ulong sig,
2668 : ulong chunk,
2669 : ulong sz,
2670 : ulong ctl,
2671 : ulong tsorig,
2672 : ulong tspub,
2673 0 : fd_stem_context_t * stem ) {
2674 0 : (void)seq;
2675 0 : (void)ctl;
2676 0 : (void)tsorig;
2677 0 : (void)tspub;
2678 :
2679 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
2680 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 ));
2681 :
2682 0 : switch( ctx->in_kind[in_idx] ) {
2683 0 : case IN_KIND_GENESIS: {
2684 0 : fd_genesis_meta_t const * meta = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2685 0 : ctx->has_genesis_hash = 1;
2686 0 : ctx->has_genesis_timestamp = 1;
2687 0 : ctx->genesis_timestamp = meta->creation_time_millis;
2688 0 : *ctx->genesis_hash = meta->genesis_hash;
2689 0 : if( FD_LIKELY( meta->bootstrap ) ) {
2690 0 : boot_genesis( ctx, stem, meta );
2691 0 : } else {
2692 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
2693 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
2694 0 : }
2695 0 : ctx->has_genesis_timestamp = 1;
2696 0 : ctx->genesis_timestamp = ctx->genesis->creation_time;
2697 :
2698 0 : maybe_verify_cluster_type( ctx );
2699 0 : maybe_verify_shred_version( ctx );
2700 0 : maybe_verify_genesis_timestamp( ctx );
2701 0 : break;
2702 0 : }
2703 0 : case IN_KIND_IPECHO: {
2704 0 : FD_TEST( sig && sig<=USHORT_MAX );
2705 0 : ctx->ipecho_shred_version = (ushort)sig;
2706 0 : maybe_verify_shred_version( ctx );
2707 0 : break;
2708 0 : }
2709 0 : case IN_KIND_SNAP: {
2710 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
2711 0 : maybe_verify_shred_version( ctx );
2712 0 : maybe_verify_genesis_timestamp( ctx );
2713 0 : break;
2714 0 : }
2715 0 : case IN_KIND_EXECRP: {
2716 0 : process_exec_task_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
2717 0 : break;
2718 0 : }
2719 0 : case IN_KIND_POH: {
2720 0 : process_poh_message( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2721 0 : break;
2722 0 : }
2723 0 : case IN_KIND_RESOLV: {
2724 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2725 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
2726 0 : break;
2727 0 : }
2728 0 : case IN_KIND_TOWER: {
2729 0 : if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_DONE ) ) {
2730 0 : process_tower_slot_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), seq );
2731 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_CONFIRMED ) ) {
2732 0 : fd_tower_slot_confirmed_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2733 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_OPTIMISTIC && !msg->fwd ) process_tower_optimistic_confirmed( ctx, stem, msg );
2734 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_DUPLICATE ) fd_reasm_confirm( ctx->reasm, &msg->block_id );
2735 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_IGNORED ) ) {
2736 0 : fd_tower_slot_ignored_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2737 0 : fd_tower_slot_done_t ignored = {
2738 0 : .replay_slot = msg->slot,
2739 0 : .replay_bank_idx = msg->bank_idx,
2740 0 : .vote_slot = ULONG_MAX,
2741 0 : .reset_slot = ctx->reset_slot, /* Use most recent reset slot */
2742 0 : .reset_block_id = ctx->reset_block_id,
2743 0 : .root_slot = ULONG_MAX
2744 0 : };
2745 0 : process_tower_slot_done( ctx, stem, &ignored, seq );
2746 0 : }
2747 0 : break;
2748 0 : }
2749 0 : case IN_KIND_REPAIR: {
2750 : /* TODO: This message/sz should be defined. */
2751 0 : if( sz!=0 && fd_disco_shred_out_msg_type( sig )==FD_SHRED_OUT_MSG_TYPE_FEC ) {
2752 : /* If receive a FEC complete message. */
2753 0 : process_fec_complete( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2754 0 : }
2755 0 : break;
2756 0 : }
2757 0 : case IN_KIND_TXSEND: {
2758 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2759 0 : break;
2760 0 : }
2761 0 : case IN_KIND_GOSSIP_OUT: {
2762 0 : FD_TEST( sig==FD_GOSSIP_UPDATE_TAG_WFS_DONE );
2763 0 : ctx->wfs_complete = 1;
2764 0 : FD_LOG_NOTICE(( "Done waiting for supermajority. More than 80 percent of cluster stake has joined." ));
2765 0 : break;
2766 0 : }
2767 0 : case IN_KIND_RPC: {
2768 0 : fd_bank_t bank[1];
2769 0 : FD_TEST( fd_banks_bank_query( bank, ctx->banks, sig ) );
2770 0 : bank->data->refcnt--;
2771 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for %s", bank->data->idx, fd_bank_slot_get( bank ), bank->data->refcnt, ctx->in_kind[ in_idx ]==IN_KIND_RPC ? "rpc" : "gui" ));
2772 0 : break;
2773 0 : }
2774 0 : default:
2775 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
2776 0 : }
2777 :
2778 0 : return 0;
2779 0 : }
2780 :
2781 : static inline fd_replay_out_link_t
2782 : out1( fd_topo_t const * topo,
2783 : fd_topo_tile_t const * tile,
2784 0 : char const * name ) {
2785 0 : ulong idx = ULONG_MAX;
2786 :
2787 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2788 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2789 0 : if( !strcmp( link->name, name ) ) {
2790 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 ));
2791 0 : idx = i;
2792 0 : }
2793 0 : }
2794 :
2795 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
2796 :
2797 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2798 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2799 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2800 :
2801 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2802 0 : }
2803 :
2804 : static void
2805 : privileged_init( fd_topo_t * topo,
2806 0 : fd_topo_tile_t * tile ) {
2807 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2808 :
2809 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2810 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2811 :
2812 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
2813 :
2814 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
2815 0 : ctx->identity_idx = 0UL;
2816 :
2817 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
2818 0 : tile->replay.bundle.enabled = 0;
2819 0 : }
2820 :
2821 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2822 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2823 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
2824 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2825 0 : }
2826 0 : }
2827 :
2828 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->reasm_seed, sizeof(ulong) ) ) ) {
2829 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2830 0 : }
2831 :
2832 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->vote_tracker_seed, sizeof(ulong) ) ) ) {
2833 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2834 0 : }
2835 :
2836 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->block_id_map_seed, sizeof(ulong) ) ) ) {
2837 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2838 0 : }
2839 :
2840 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->initial_block_id, sizeof(fd_hash_t) ) ) ) {
2841 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2842 0 : }
2843 :
2844 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->runtime_stack_seed, sizeof(ulong) ) ) ) {
2845 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2846 0 : }
2847 0 : }
2848 :
2849 : static void
2850 : unprivileged_init( fd_topo_t * topo,
2851 0 : fd_topo_tile_t * tile ) {
2852 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2853 :
2854 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
2855 :
2856 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2857 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2858 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 );
2859 0 : void * block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
2860 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
2861 0 : void * reasm_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
2862 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
2863 0 : void * vinyl_req_pool_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
2864 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
2865 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
2866 0 : void * dump_proto_ctx_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
2867 0 : # if FD_HAS_FLATCC
2868 0 : void * block_dump_ctx = NULL;
2869 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2870 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
2871 0 : }
2872 0 : # endif
2873 :
2874 0 : FD_TEST( fd_vote_rewards_map_join( fd_vote_rewards_map_new( ctx->runtime_stack.stakes.vote_map_mem, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, ctx->runtime_stack_seed ) ) );
2875 :
2876 0 : ctx->wksp = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].wksp;
2877 :
2878 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
2879 0 : FD_TEST( store_obj_id!=ULONG_MAX );
2880 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
2881 0 : FD_TEST( ctx->store );
2882 :
2883 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
2884 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
2885 0 : ulong banks_locks_obj_id = fd_pod_query_ulong( topo->props, "banks_locks", ULONG_MAX );
2886 0 : FD_TEST( banks_locks_obj_id!=ULONG_MAX );
2887 :
2888 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 ) ) );
2889 :
2890 0 : fd_bank_data_t * bank_pool = fd_banks_get_bank_pool( ctx->banks->data );
2891 0 : FD_MGAUGE_SET( REPLAY, MAX_LIVE_BANKS, fd_banks_pool_max( bank_pool ) );
2892 :
2893 0 : fd_bank_t bank[1];
2894 0 : FD_TEST( fd_banks_init_bank( bank, ctx->banks ) );
2895 0 : fd_bank_slot_set( bank, 0UL );
2896 0 : FD_TEST( bank->data->idx==FD_REPLAY_BOOT_BANK_IDX );
2897 :
2898 0 : ctx->consensus_root_slot = ULONG_MAX;
2899 0 : ctx->consensus_root = ctx->initial_block_id;
2900 0 : ctx->published_root_slot = ULONG_MAX;
2901 :
2902 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
2903 0 : ctx->ipecho_shred_version = 0;
2904 0 : fd_memcpy( ctx->genesis_path, tile->replay.genesis_path, sizeof(ctx->genesis_path) );
2905 0 : ctx->has_genesis_hash = 0;
2906 0 : ctx->has_genesis_timestamp = 0;
2907 0 : ctx->has_expected_genesis_timestamp = 0;
2908 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
2909 0 : ctx->hard_forks_cnt = ULONG_MAX;
2910 :
2911 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2912 0 : ctx->bundle.enabled = 1;
2913 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
2914 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
2915 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
2916 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2917 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
2918 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
2919 0 : }
2920 0 : } else {
2921 0 : ctx->bundle.enabled = 0;
2922 0 : }
2923 :
2924 0 : fd_features_t * features = fd_bank_features_modify( bank );
2925 0 : fd_features_enable_cleaned_up( features );
2926 :
2927 0 : char const * one_off_features[ 16UL ];
2928 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(one_off_features)/sizeof(one_off_features[0]) );
2929 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) one_off_features[ i ] = tile->replay.enable_features[i];
2930 0 : fd_features_enable_one_offs( features, one_off_features, (uint)tile->replay.enable_features_cnt, 0UL );
2931 :
2932 0 : fd_topo_obj_t const * vinyl_data = fd_topo_find_tile_obj( topo, tile, "vinyl_data" );
2933 :
2934 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ) )!=ULONG_MAX );
2935 0 : ulong progcache_locks_obj_id; FD_TEST( (progcache_locks_obj_id = fd_pod_query_ulong( topo->props, "progcache_locks", ULONG_MAX ) )!=ULONG_MAX );
2936 0 : FD_TEST( fd_progcache_admin_join( ctx->progcache_admin,
2937 0 : fd_topo_obj_laddr( topo, progcache_obj_id ),
2938 0 : fd_topo_obj_laddr( topo, progcache_locks_obj_id ) ) );
2939 :
2940 0 : ulong funk_obj_id; FD_TEST( (funk_obj_id = fd_pod_query_ulong( topo->props, "funk", ULONG_MAX ) )!=ULONG_MAX );
2941 0 : ulong funk_locks_obj_id; FD_TEST( (funk_locks_obj_id = fd_pod_query_ulong( topo->props, "funk_locks", ULONG_MAX ) )!=ULONG_MAX );
2942 0 : ulong max_depth = tile->replay.max_live_slots + tile->replay.write_delay_slots;
2943 0 : if( !vinyl_data ) {
2944 0 : FD_TEST( fd_accdb_admin_v1_init( ctx->accdb_admin,
2945 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2946 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ) ) );
2947 0 : } else {
2948 0 : fd_topo_obj_t const * vinyl_rq = fd_topo_find_tile_obj( topo, tile, "vinyl_rq" );
2949 0 : fd_topo_obj_t const * vinyl_req_pool = fd_topo_find_tile_obj( topo, tile, "vinyl_rpool" );
2950 0 : FD_TEST( fd_accdb_admin_v2_init( ctx->accdb_admin,
2951 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2952 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ),
2953 0 : fd_topo_obj_laddr( topo, vinyl_rq->id ),
2954 0 : topo->workspaces[ vinyl_data->wksp_id ].wksp,
2955 0 : fd_topo_obj_laddr( topo, vinyl_req_pool->id ),
2956 0 : vinyl_rq->id,
2957 0 : max_depth ) );
2958 0 : fd_accdb_admin_v2_delay_set( ctx->accdb_admin, tile->replay.write_delay_slots );
2959 0 : }
2960 0 : fd_accdb_init_from_topo( ctx->accdb, topo, tile, max_depth );
2961 :
2962 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
2963 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
2964 0 : FD_TEST( txncache_shmem );
2965 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
2966 0 : FD_TEST( ctx->txncache );
2967 :
2968 0 : ctx->capture_ctx = NULL;
2969 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
2970 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
2971 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
2972 0 : ctx->capture_ctx->capture_solcap = 1;
2973 0 : }
2974 :
2975 0 : ctx->dump_proto_ctx = NULL;
2976 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2977 0 : ctx->dump_proto_ctx = dump_proto_ctx_mem;
2978 0 : ctx->dump_proto_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
2979 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) {
2980 0 : ctx->dump_proto_ctx->dump_block_to_pb = !!tile->replay.dump_block_to_pb;
2981 0 : }
2982 0 : }
2983 :
2984 0 : # if FD_HAS_FLATCC
2985 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2986 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
2987 0 : } else {
2988 0 : ctx->block_dump_ctx = NULL;
2989 0 : }
2990 0 : # endif
2991 :
2992 0 : ctx->is_booted = 0;
2993 :
2994 0 : ctx->larger_max_cost_per_block = tile->replay.larger_max_cost_per_block;
2995 :
2996 0 : ctx->reasm = fd_reasm_join( fd_reasm_new( reasm_mem, tile->replay.fec_max, ctx->reasm_seed ) );
2997 0 : FD_TEST( ctx->reasm );
2998 :
2999 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, tile->replay.sched_depth, tile->replay.max_live_slots, fd_topo_tile_name_cnt( topo, "execrp" ) ) );
3000 0 : FD_TEST( ctx->sched );
3001 :
3002 0 : FD_TEST( fd_vinyl_req_pool_new( vinyl_req_pool_mem, 1UL, 1UL ) );
3003 :
3004 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, ctx->vote_tracker_seed ) );
3005 0 : FD_TEST( ctx->vote_tracker );
3006 :
3007 0 : ctx->identity_vote_rooted = 0;
3008 :
3009 0 : ctx->wait_for_vote_to_start_leader = tile->replay.wait_for_vote_to_start_leader;
3010 :
3011 0 : ctx->wfs_enabled = memcmp( tile->replay.wait_for_supermajority_with_bank_hash.uc, ((fd_pubkey_t){ 0 }).uc, sizeof(fd_pubkey_t) );
3012 0 : ctx->expected_bank_hash = tile->replay.wait_for_supermajority_with_bank_hash;
3013 0 : ctx->wfs_complete = !ctx->wfs_enabled;
3014 :
3015 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
3016 0 : FD_TEST( ctx->mleaders );
3017 :
3018 0 : ctx->is_leader = 0;
3019 0 : ctx->supports_leader = fd_topo_find_tile( topo, "pack", 0UL )!=ULONG_MAX;
3020 0 : ctx->reset_slot = 0UL;
3021 0 : fd_memset( ctx->reset_bank, 0, sizeof(fd_bank_t) );
3022 0 : ctx->reset_block_id = ctx->initial_block_id;
3023 0 : ctx->reset_timestamp_nanos = 0UL;
3024 0 : ctx->next_leader_slot = ULONG_MAX;
3025 0 : ctx->next_leader_tickcount = LONG_MAX;
3026 0 : ctx->highwater_leader_slot = ULONG_MAX;
3027 0 : ctx->slot_duration_nanos = 350L*1000L*1000L; /* TODO: Not fixed ... not always 350ms ... */
3028 0 : ctx->slot_duration_ticks = (double)ctx->slot_duration_nanos*fd_tempo_tick_per_ns( NULL );
3029 0 : ctx->leader_bank->data = NULL;
3030 :
3031 0 : ctx->block_id_len = tile->replay.max_live_slots;
3032 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
3033 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 ) );
3034 0 : FD_TEST( ctx->block_id_map );
3035 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) ctx->block_id_arr[ i ].block_id_seen = 0;
3036 :
3037 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
3038 :
3039 0 : ctx->keyswitch = fd_keyswitch_join( fd_topo_obj_laddr( topo, tile->id_keyswitch_obj_id ) );
3040 0 : FD_TEST( ctx->keyswitch );
3041 0 : ctx->halt_leader = 0;
3042 :
3043 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
3044 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
3045 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
3046 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
3047 :
3048 0 : if( FD_LIKELY( link->dcache ) ) {
3049 0 : ctx->in[ i ].mem = link_wksp->wksp;
3050 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
3051 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
3052 0 : ctx->in[ i ].mtu = link->mtu;
3053 0 : }
3054 :
3055 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
3056 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
3057 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
3058 0 : else if( !strcmp( link->name, "execrp_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXECRP;
3059 0 : else if( !strcmp( link->name, "tower_out" ) ) ctx->in_kind[ i ] = IN_KIND_TOWER;
3060 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
3061 0 : else if( !strcmp( link->name, "resolv_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
3062 0 : else if( !strcmp( link->name, "repair_out" ) ) ctx->in_kind[ i ] = IN_KIND_REPAIR;
3063 0 : else if( !strcmp( link->name, "txsend_out" ) ) ctx->in_kind[ i ] = IN_KIND_TXSEND;
3064 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
3065 0 : else if( !strcmp( link->name, "gossip_out" ) ) ctx->in_kind[ i ] = IN_KIND_GOSSIP_OUT;
3066 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
3067 0 : }
3068 :
3069 0 : *ctx->epoch_out = out1( topo, tile, "replay_epoch" ); FD_TEST( ctx->epoch_out->idx!=ULONG_MAX );
3070 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
3071 0 : *ctx->exec_out = out1( topo, tile, "replay_execrp" ); FD_TEST( ctx->exec_out->idx!=ULONG_MAX );
3072 :
3073 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
3074 :
3075 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
3076 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "cap_repl", 0UL );
3077 0 : FD_TEST( idx!=ULONG_MAX );
3078 0 : fd_topo_link_t * link = &topo->links[ tile->out_link_id[ idx ] ];
3079 :
3080 :
3081 0 : fd_capture_link_buf_t * cap_repl_out = ctx->cap_repl_out;
3082 0 : cap_repl_out->base.vt = &fd_capture_link_buf_vt;
3083 0 : cap_repl_out->idx = idx;
3084 0 : cap_repl_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
3085 0 : cap_repl_out->chunk0 = fd_dcache_compact_chunk0( cap_repl_out->mem, link->dcache );
3086 0 : cap_repl_out->wmark = fd_dcache_compact_wmark( cap_repl_out->mem, link->dcache, link->mtu );
3087 0 : cap_repl_out->chunk = cap_repl_out->chunk0;
3088 0 : cap_repl_out->mcache = link->mcache;
3089 0 : cap_repl_out->depth = fd_mcache_depth( link->mcache );
3090 0 : cap_repl_out->seq = 0UL;
3091 :
3092 0 : ctx->capture_ctx->capctx_type.buf = cap_repl_out;
3093 0 : ctx->capture_ctx->capture_link = &cap_repl_out->base;
3094 0 : ctx->capture_ctx->current_txn_idx = 0UL;
3095 :
3096 :
3097 0 : ulong consumer_tile_idx = fd_topo_find_tile( topo, "solcap", 0UL );
3098 0 : fd_topo_tile_t * consumer_tile = &topo->tiles[ consumer_tile_idx ];
3099 0 : cap_repl_out->fseq = NULL;
3100 0 : for( ulong j = 0UL; j < consumer_tile->in_cnt; j++ ) {
3101 0 : if( FD_UNLIKELY( consumer_tile->in_link_id[ j ] == link->id ) ) {
3102 0 : cap_repl_out->fseq = fd_fseq_join( fd_topo_obj_laddr( topo, consumer_tile->in_link_fseq_obj_id[ j ] ) );
3103 0 : FD_TEST( cap_repl_out->fseq );
3104 0 : break;
3105 0 : }
3106 0 : }
3107 0 : }
3108 :
3109 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
3110 :
3111 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WAIT ),
3112 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WAIT ) ) );
3113 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WORK ),
3114 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WORK ) ) );
3115 :
3116 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_slot_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_SLOT_DURATION_SECONDS ),
3117 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_SLOT_DURATION_SECONDS ) ) );
3118 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_account_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ),
3119 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ) ) );
3120 :
3121 : /* Ensure precompiles are available, crash fast otherwise */
3122 0 : fd_precompiles();
3123 :
3124 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
3125 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
3126 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
3127 0 : }
3128 :
3129 : static ulong
3130 : populate_allowed_seccomp( fd_topo_t const * topo FD_FN_UNUSED,
3131 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3132 : ulong out_cnt,
3133 0 : struct sock_filter * out ) {
3134 :
3135 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() );
3136 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
3137 0 : }
3138 :
3139 : static ulong
3140 : populate_allowed_fds( fd_topo_t const * topo FD_FN_UNUSED,
3141 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3142 : ulong out_fds_cnt,
3143 0 : int * out_fds ) {
3144 :
3145 0 : if( FD_UNLIKELY( out_fds_cnt<2UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
3146 :
3147 0 : ulong out_cnt = 0UL;
3148 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
3149 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
3150 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
3151 0 : return out_cnt;
3152 0 : }
3153 :
3154 : static inline void
3155 0 : during_housekeeping( fd_replay_tile_t * ctx ) {
3156 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_UNHALT_PENDING ) ) {
3157 0 : FD_CRIT( ctx->halt_leader, "state machine corruption" );
3158 0 : FD_LOG_DEBUG(( "keyswitch: unhalting leader" ));
3159 0 : ctx->halt_leader = 0;
3160 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
3161 0 : }
3162 :
3163 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) {
3164 0 : FD_LOG_DEBUG(( "keyswitch: halting leader" ));
3165 0 : ctx->halt_leader = 1;
3166 0 : if( !ctx->is_leader ) maybe_switch_identity( ctx );
3167 0 : }
3168 0 : }
3169 :
3170 : #undef DEBUG_LOGGING
3171 :
3172 : /* counting carefully, after_credit can generate at most 7 frags and
3173 : returnable_frag boot_genesis can also generate at most 7 frags, so 14
3174 : is a conservative bound. */
3175 0 : #define STEM_BURST (14UL)
3176 :
3177 : /* fd_tempo_lazy_default( 16384 ) where 16384 is the minimum out-link
3178 : depth (i.e. cr_max) but excludes replay_epoch, which is so infrequent
3179 : credit availability is a non-issue. */
3180 0 : #define STEM_LAZY ((long)36865)
3181 :
3182 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
3183 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
3184 :
3185 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
3186 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
3187 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
3188 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
3189 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
3190 :
3191 : #include "../../disco/stem/fd_stem.c"
3192 :
3193 : fd_topo_run_tile_t fd_tile_replay = {
3194 : .name = "replay",
3195 : .populate_allowed_seccomp = populate_allowed_seccomp,
3196 : .populate_allowed_fds = populate_allowed_fds,
3197 : .scratch_align = scratch_align,
3198 : .scratch_footprint = scratch_footprint,
3199 : .privileged_init = privileged_init,
3200 : .unprivileged_init = unprivileged_init,
3201 : .run = stem_run,
3202 : };
|
