Line data Source code
1 : #include "../../shared/fd_config.h"
2 : #include "../../shared/fd_bootinfo.h"
3 : #include "../../shared/fd_action.h"
4 : #include "../../../disco/metrics/fd_metrics.h"
5 : #include "../../../disco/fd_clock_tile.h"
6 : #include "../../../disco/net/fd_net_tile.h"
7 : #include "../../../util/clock/fd_clock.h"
8 : #include "../../../util/net/fd_pcap.h"
9 :
10 : #include <errno.h>
11 : #include <fcntl.h>
12 : #include <signal.h>
13 : #include <stdio.h>
14 : #include <unistd.h>
15 :
16 0 : #define FD_DUMP_FLAGS_NO_DATA 1
17 0 : #define FD_DUMP_FLAGS_NET_RX 2
18 :
19 : struct dump_ctx {
20 : fd_io_buffered_ostream_t ostream;
21 : ulong pending_sz;
22 :
23 : struct link {
24 : fd_topo_link_t const * topo;
25 : union {
26 : void const * dcache_base;
27 : fd_net_rx_bounds_t net_rx_bounds[1];
28 : };
29 : uint link_hash;
30 : int flags; /* FD_DUMP_FLAGS_* */
31 : } links[ FD_TOPO_MAX_LINKS ];
32 : ulong link_cnt;
33 :
34 : ulong * metrics_base;
35 : long next_stat_log;
36 : ulong frags;
37 : ulong bytes;
38 : ulong last_overrun_frags;
39 :
40 : long warmup_until;
41 :
42 : fd_clock_tile_t clock[ 1 ];
43 : };
44 : typedef struct dump_ctx dump_ctx_t;
45 :
46 : void
47 : dump_cmd_args( int * argc,
48 : char * * * argv,
49 0 : args_t * args ) {
50 0 : char const * out_file = fd_env_strip_cmdline_cstr( argc, argv, "--out-file", NULL, "dump.pcap" );
51 0 : char const * link = fd_env_strip_cmdline_cstr( argc, argv, "--link", NULL, "" );
52 0 : args->dump.once = fd_env_strip_cmdline_int( argc, argv, "--once", NULL, 1 );
53 0 : fd_cstr_fini( fd_cstr_append_cstr_safe( fd_cstr_init( args->dump.pcap_path ), out_file, sizeof(args->dump.pcap_path)-1UL ) );
54 0 : fd_cstr_fini( fd_cstr_append_cstr_safe( fd_cstr_init( args->dump.link_name ), link, sizeof(args->dump.link_name)-1UL ) );
55 0 : }
56 :
57 : static FD_FN_CONST ulong
58 : frag_sz( int flags,
59 0 : ulong line_sz ) {
60 0 : return fd_ulong_if( flags & FD_DUMP_FLAGS_NO_DATA, 0UL, line_sz );
61 0 : }
62 : static FD_FN_PURE void const *
63 : frag_ptr( int flags,
64 : struct link const * link,
65 : ulong chunk,
66 0 : ulong ctl ) {
67 0 : if( FD_UNLIKELY( flags & FD_DUMP_FLAGS_NET_RX ) ) return fd_net_rx_translate_frag( link->net_rx_bounds, chunk, ctl, 0UL );
68 0 : else return fd_chunk_to_laddr_const ( link->dcache_base, chunk );
69 0 : }
70 :
71 : static void
72 : dump_link_once( dump_ctx_t * ctx,
73 0 : struct link const * link ) {
74 0 : fd_frag_meta_t const * mcache = link->topo->mcache;
75 0 : ulong seq0 = fd_mcache_seq0( mcache );
76 0 : ulong seq_init = fd_mcache_seq_query( fd_mcache_seq_laddr_const( mcache ) );
77 0 : ulong depth = fd_mcache_depth( mcache );
78 :
79 0 : ulong frags = 0UL;
80 0 : ulong bytes = 0UL;
81 :
82 0 : int flags = link->flags;
83 :
84 : /* For links that are published reliably, we can trust the value of
85 : seq_init. For unreliable links, we'll just publish one whole
86 : depth. */
87 :
88 : /* We know at this point [seq0, seq_init) were published, but they may
89 : be long overwritten, and there may be more published than that. */
90 :
91 0 : ulong min_seq_seen = seq_init; /* What we actually dumped is [min_seq_seen, seq_init) */
92 0 : for( ulong seq=fd_seq_dec( seq_init, 1UL ); fd_seq_ge( seq, seq0 ); seq=fd_seq_dec( seq, 1UL ) ) {
93 : /* It's not necessary for this to be atomic, since this is a
94 : post-mortem tool. */
95 0 : fd_frag_meta_t const * line = mcache+fd_mcache_line_idx( seq, depth );
96 0 : ulong read_seq = fd_frag_meta_seq_query( line );
97 0 : if( FD_UNLIKELY( read_seq!=seq ) ) break;
98 :
99 0 : min_seq_seen = seq;
100 :
101 0 : ulong sz = frag_sz ( flags, line->sz );
102 0 : void const * buffer = frag_ptr( flags, link, line->chunk, line->ctl );
103 :
104 0 : fd_pcap_ostream_pkt( &ctx->ostream, (long)seq, line, sizeof(fd_frag_meta_t), buffer, sz, link->link_hash );
105 0 : bytes += sz;
106 0 : frags++;
107 0 : }
108 :
109 : /* Now check everything after seq_init. This could potentially loop
110 : forever if the producer is still going, so we cap it at one depth. */
111 0 : for( ulong off=0UL; off<depth; off++ ) {
112 0 : ulong seq = fd_seq_inc( seq_init, off );
113 :
114 0 : fd_frag_meta_t const * line = mcache+fd_mcache_line_idx( seq, depth );
115 0 : ulong read_seq = fd_frag_meta_seq_query( line );
116 :
117 : /* Skip anything we processed in the first loop, also skipping any
118 : with the err control bit set. When an mcache is initialized, all
119 : the frags have err set to true, so this will skip those in
120 : particular as well. */
121 0 : if( FD_UNLIKELY( (fd_seq_le( min_seq_seen, read_seq ) & fd_seq_lt( read_seq, seq_init )) | (line->ctl & (1<<2)) ) ) continue;
122 :
123 0 : ulong sz = frag_sz ( flags, line->sz );
124 0 : void const * buffer = frag_ptr( flags, link, line->chunk, line->ctl );
125 :
126 0 : fd_pcap_ostream_pkt( &ctx->ostream, (long)seq, line, sizeof(fd_frag_meta_t), buffer, sz, link->link_hash );
127 0 : bytes += sz;
128 0 : frags++;
129 0 : }
130 0 : FD_LOG_NOTICE(( "dumped %lu frags, %lu bytes in total from %s:%lu. Link hash: 0x%x",
131 0 : frags, bytes, link->topo->name, link->topo->kind_id, link->link_hash ));
132 0 : }
133 :
134 : static int running = 1;
135 :
136 : static void
137 0 : exit_signal( int sig FD_PARAM_UNUSED ) {
138 0 : running = 0;
139 0 : }
140 :
141 : static int
142 0 : should_shutdown( dump_ctx_t * ctx FD_PARAM_UNUSED ) {
143 0 : return !running;
144 0 : }
145 :
146 : static void
147 : during_frag( dump_ctx_t * ctx,
148 : ulong in_idx,
149 : ulong seq,
150 : ulong sig FD_PARAM_UNUSED ,
151 : ulong chunk,
152 : ulong sz,
153 0 : ulong ctl ) {
154 : /* We have a new candidate fragment to copy into the dump file.
155 : Because we attach read-only and do not backpressure any producers,
156 : this fragment could be overrun at any point during this function.
157 : So we copy the relevant data into local memory (in the buffered
158 : ostream peek space) and only commit it to the file later in
159 : after_frag. after_frag is only called if the fragment was not
160 : overrun while we were processing it. */
161 :
162 0 : ctx->pending_sz = 0UL;
163 :
164 0 : struct link const * llink = ctx->links+in_idx;
165 0 : int flags = llink->flags;
166 :
167 0 : sz = frag_sz( flags, sz );
168 :
169 :
170 0 : long pcap_sz = fd_pcap_pkt_sz( sizeof(fd_frag_meta_t), sz );
171 0 : if( FD_UNLIKELY( pcap_sz < 0L ) ) return;
172 :
173 0 : if( (ulong)pcap_sz > fd_io_buffered_ostream_peek_sz( &ctx->ostream ) ) {
174 0 : fd_io_buffered_ostream_flush( &ctx->ostream );
175 0 : if( FD_UNLIKELY( (ulong)pcap_sz > fd_io_buffered_ostream_peek_sz( &ctx->ostream ) ) ) {
176 0 : FD_LOG_ERR(( "packet size %ld too large for pcap, increase write buffer size (%lu)",
177 0 : pcap_sz, fd_io_buffered_ostream_peek_sz( &ctx->ostream ) ));
178 0 : }
179 0 : }
180 :
181 :
182 0 : fd_topo_link_t const * link = ctx->links[ in_idx ].topo;
183 0 : fd_frag_meta_t const * mline = link->mcache + fd_mcache_line_idx( seq, fd_mcache_depth( link->mcache ) );
184 0 : void const * frag_data = frag_ptr( flags, llink, chunk, ctl );
185 :
186 0 : fd_pcap_pkt( fd_io_buffered_ostream_peek( &ctx->ostream ),
187 0 : 0L,
188 0 : mline,
189 0 : sizeof(fd_frag_meta_t),
190 0 : frag_data,
191 0 : sz,
192 0 : ctx->links[ in_idx ].link_hash );
193 0 : ctx->pending_sz = (ulong)pcap_sz;
194 0 : }
195 :
196 : static void
197 : after_frag( dump_ctx_t * ctx,
198 : ulong in_idx FD_PARAM_UNUSED,
199 : ulong seq FD_PARAM_UNUSED,
200 : ulong sig FD_PARAM_UNUSED,
201 : ulong sz,
202 : ulong tsorig FD_PARAM_UNUSED,
203 : ulong tspub,
204 0 : fd_stem_context_t * stem FD_PARAM_UNUSED ) {
205 0 : if( FD_UNLIKELY( !ctx->pending_sz ) ) return;
206 :
207 0 : sz = frag_sz( ctx->links[ in_idx ].flags, sz );
208 0 : long pcap_sz = fd_pcap_pkt_sz( sizeof(fd_frag_meta_t), sz );
209 0 : FD_TEST( (ulong)pcap_sz == ctx->pending_sz );
210 :
211 0 : long tickcount = fd_tickcount();
212 0 : long now = fd_clock_epoch_y( ctx->clock->shmem->epoch, tickcount );
213 0 : if( FD_LIKELY( now>ctx->warmup_until ) ) {
214 0 : if( FD_LIKELY( tspub!=0UL ) ) {
215 0 : long tspub_decomp = fd_frag_meta_ts_decomp( tspub, tickcount );
216 0 : long tspub_wallclock = fd_clock_epoch_y( ctx->clock->shmem->epoch, tspub_decomp );
217 0 : uint * pcap_tss = fd_io_buffered_ostream_peek( &ctx->ostream );
218 0 : pcap_tss[ 0 ] = (uint)(((ulong)tspub_wallclock) / (ulong)1e9);
219 0 : pcap_tss[ 1 ] = (uint)(((ulong)tspub_wallclock) % (ulong)1e9); /* Actually nsec */
220 0 : }
221 0 : fd_io_buffered_ostream_seek( &ctx->ostream, ctx->pending_sz );
222 0 : ctx->frags += 1UL;
223 0 : ctx->bytes += sz;
224 0 : }
225 0 : ctx->pending_sz = 0UL;
226 0 : }
227 :
228 : static void
229 0 : log_metrics( dump_ctx_t * ctx ) {
230 0 : FD_LOG_NOTICE(( "dumped %lu frags, %lu bytes", ctx->frags, ctx->bytes ));
231 0 : ctx->frags = 0UL;
232 0 : ctx->bytes = 0UL;
233 :
234 0 : ulong overrun_frags = 0UL;
235 0 : for( ulong i=0UL; i<ctx->link_cnt; i++) {
236 0 : volatile ulong const * link_metrics = fd_metrics_link_in( ctx->metrics_base, i );
237 0 : overrun_frags += link_metrics[ FD_METRICS_COUNTER_LINK_FRAG_POLLING_OVERRUN_OFF ];
238 0 : overrun_frags += link_metrics[ FD_METRICS_COUNTER_LINK_FRAG_READING_OVERRUN_OFF ];
239 0 : }
240 0 : if( FD_UNLIKELY( overrun_frags != ctx->last_overrun_frags ) ) {
241 : /* Note: We expect overruns at startup because we have no way to
242 : know the current seq to start polling from, so we start at 0
243 : which has often been overrun. */
244 0 : if( FD_LIKELY( ctx->last_overrun_frags != 0 ) ) {
245 0 : long overrun_diff = fd_seq_diff( overrun_frags, ctx->last_overrun_frags );
246 0 : FD_LOG_WARNING(( "overrun detected, %ld frags dropped", overrun_diff ));
247 0 : }
248 0 : ctx->last_overrun_frags = overrun_frags;
249 0 : }
250 0 : }
251 :
252 : static void
253 0 : during_housekeeping( dump_ctx_t * ctx ) {
254 0 : long now = fd_clock_tile_now( ctx->clock );
255 0 : if( FD_UNLIKELY( now >= fd_clock_tile_recal_next( ctx->clock ) ) ) {
256 0 : fd_clock_tile_recal( ctx->clock );
257 0 : }
258 0 : if( FD_UNLIKELY( now > ctx->next_stat_log ) ) {
259 0 : ctx->next_stat_log = now + (long)1e9;
260 0 : log_metrics( ctx );
261 0 : }
262 0 : }
263 :
264 : #define STEM_BURST (0UL)
265 : #define STEM_CALLBACK_CONTEXT_TYPE dump_ctx_t
266 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(dump_ctx_t)
267 0 : #define STEM_CALLBACK_DURING_FRAG during_frag
268 0 : #define STEM_CALLBACK_AFTER_FRAG after_frag
269 0 : #define STEM_CALLBACK_SHOULD_SHUTDOWN should_shutdown
270 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
271 : #include "../../../disco/stem/fd_stem.c"
272 :
273 : void
274 : dump_cmd_fn( args_t * args,
275 0 : fd_config_t * config ) {
276 0 : char * tokens[ 16 ];
277 0 : ulong token_count = fd_cstr_tokenize( tokens, 16UL, args->dump.link_name, ',' );
278 :
279 0 : dump_ctx_t ctx = { 0 };
280 :
281 0 : uchar write_buf[ 131072 ];
282 0 : int fd = open( args->dump.pcap_path, O_WRONLY | O_CREAT | O_TRUNC,
283 0 : S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH );
284 0 : if( FD_UNLIKELY( fd < 0 ) )
285 0 : FD_LOG_ERR(( "open() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
286 0 : if( FD_UNLIKELY( &ctx.ostream != fd_io_buffered_ostream_init( &ctx.ostream, fd, write_buf, sizeof(write_buf) ) ) )
287 0 : FD_LOG_ERR(( "fd_io_buffered_ostream_init() failed" ));
288 :
289 0 : FD_TEST( 0==fd_pcap_ostream_hdr( &ctx.ostream, FD_PCAP_LINK_LAYER_USER0 ) );
290 :
291 0 : fd_topo_t * topo = &config->topo;
292 0 : fd_bootinfo_adopt( config );
293 0 : fd_bootinfo_check_layout( config );
294 0 : fd_topo_join_workspaces( topo, FD_SHMEM_JOIN_MODE_READ_ONLY, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
295 0 : fd_topo_fill( topo );
296 :
297 0 : for( ulong i=0UL; i<topo->link_cnt; i++) {
298 0 : fd_topo_link_t * link = &topo->links[ i ];
299 0 : int found = (token_count==0UL);
300 0 : for( ulong j=0UL; (!found)&(j<token_count); j++ ) found |= !strcmp( tokens[ j ], topo->links[ i ].name );
301 0 : if( found ) {
302 0 : if( FD_UNLIKELY( NULL==link->mcache ) )
303 0 : FD_LOG_ERR(( "link %s:%lu mcache is null", link->name, link->kind_id ));
304 0 : ctx.links[ ctx.link_cnt ].topo = link;
305 0 : ctx.links[ ctx.link_cnt ].dcache_base = link->mtu ? fd_topo_obj_wksp_base( topo, link->dcache_obj_id ) : NULL;
306 0 : ctx.links[ ctx.link_cnt ].link_hash = (uint)((fd_hash( 17UL, link->name, strlen( link->name ) ) << 8) | link->kind_id);
307 :
308 0 : int flags = fd_int_if( !link->mtu, FD_DUMP_FLAGS_NO_DATA, 0 );
309 0 : if( FD_UNLIKELY( strstr( link->name, "net_" ) ) ) {
310 0 : flags |= FD_DUMP_FLAGS_NET_RX;
311 : /* this overwrites dcache_base since it's part of the union, but
312 : that's okay. */
313 0 : uchar * dcache = fd_dcache_join( fd_topo_obj_laddr( topo, link->dcache_obj_id ) );
314 0 : fd_net_rx_bounds_init( ctx.links[ ctx.link_cnt ].net_rx_bounds, dcache );
315 0 : fd_dcache_leave( dcache );
316 0 : }
317 0 : ctx.links[ ctx.link_cnt ].flags = flags;
318 :
319 0 : ctx.link_cnt++;
320 0 : }
321 0 : }
322 :
323 0 : if( args->dump.once ) {
324 0 : for( ulong i=0UL; i<ctx.link_cnt; i++ ) {
325 0 : dump_link_once( &ctx, &ctx.links[ i ] );
326 0 : }
327 0 : } else {
328 0 : struct sigaction sa = {
329 0 : .sa_handler = exit_signal,
330 0 : .sa_flags = 0,
331 0 : };
332 0 : if( FD_UNLIKELY( sigaction( SIGTERM, &sa, NULL ) ) )
333 0 : FD_LOG_ERR(( "sigaction(SIGTERM) failed (%i-%s)", errno, fd_io_strerror( errno ) ));
334 0 : if( FD_UNLIKELY( sigaction( SIGINT, &sa, NULL ) ) )
335 0 : FD_LOG_ERR(( "sigaction(SIGINT) failed (%i-%s)", errno, fd_io_strerror( errno ) ));
336 :
337 0 : fd_frag_meta_t const * mcaches[ FD_TOPO_MAX_LINKS ];
338 0 : for( ulong i=0UL; i<ctx.link_cnt; i++ ) {
339 0 : mcaches[ i ] = ctx.links[ i ].topo->mcache;
340 0 : }
341 :
342 0 : uchar fseq_mem[ FD_TOPO_MAX_LINKS ][ FD_FSEQ_FOOTPRINT ] __attribute__((aligned((FD_FSEQ_ALIGN))));
343 0 : ulong * fseqs[ FD_TOPO_MAX_LINKS ];
344 0 : for( ulong i=0UL; i<ctx.link_cnt; i++ ) {
345 0 : fseqs[ i ] = fd_fseq_join( fd_fseq_new( fseq_mem[ i ], 0UL ) );
346 0 : }
347 :
348 0 : fd_rng_t _rng[1];
349 0 : fd_rng_t * rng = fd_rng_join( fd_rng_new( _rng, (uint)fd_tickcount(), 0UL ) );
350 :
351 0 : uchar __attribute__((aligned(FD_STEM_SCRATCH_ALIGN))) scratch[ stem_scratch_footprint( ctx.link_cnt, 0UL, 0UL ) ];
352 0 : uchar __attribute__((aligned(FD_METRICS_ALIGN))) metrics_mem[ FD_METRICS_FOOTPRINT( ctx.link_cnt ) ];
353 :
354 0 : ctx.metrics_base = fd_metrics_join( fd_metrics_new( metrics_mem, ctx.link_cnt ) );
355 0 : fd_metrics_register( ctx.metrics_base );
356 :
357 : /* The purpose of the warmup period is to ensure that all frags in
358 : the pcap were from after the dump was started. fd_stem currently
359 : has no way to start from the most recently published frags when
360 : joining links, so at startup we get a full ~depth set of callbacks
361 : for old frags. This code assumes that we are caught up and only
362 : processing new frags by the time the warmup timer expires. */
363 0 : fd_clock_tile_init( ctx.clock );
364 0 : ctx.warmup_until = fd_clock_tile_now( ctx.clock ) + (long)1e9;
365 0 : ctx.next_stat_log = ctx.warmup_until + (long)1e9;
366 :
367 0 : fd_stem_sleep_t sleep[ 1 ] = {{ .shmem = NULL }}; /* no sleep object: spin */
368 0 : stem_run1( ctx.link_cnt, /* in_cnt */
369 0 : mcaches, /* in_mcache */
370 0 : fseqs, /* in_fseq */
371 0 : 0UL, /* out_cnt */
372 0 : NULL, /* out_mcache */
373 0 : 0UL, /* cons_cnt */
374 0 : NULL, /* _cons_out */
375 0 : NULL, /* _cons_fseq */
376 0 : NULL, /* cons_slow */
377 0 : 0UL, /* burst */
378 0 : 0UL, /* lazy */
379 0 : rng, /* rng */
380 0 : scratch, /* scratch */
381 0 : &ctx, /* ctx */
382 0 : sleep ); /* sleep */
383 :
384 0 : for( ulong i=0UL; i<ctx.link_cnt; i++ ) {
385 0 : struct link const * link = &ctx.links[ i ];
386 0 : volatile ulong const * link_metrics = fd_metrics_link_in( ctx.metrics_base, i );
387 0 : ulong frags = link_metrics[ FD_METRICS_COUNTER_LINK_FRAG_CONSUMED_OFF ];
388 0 : ulong bytes = link_metrics[ FD_METRICS_COUNTER_LINK_FRAG_CONSUMED_BYTES_OFF ];
389 0 : FD_LOG_NOTICE(( "dumped %lu frags, %lu bytes in total from %s:%lu. Link hash: 0x%x",
390 0 : frags, bytes, link->topo->name, link->topo->kind_id, link->link_hash ));
391 0 : }
392 :
393 0 : fd_metrics_delete( fd_metrics_leave( ctx.metrics_base ) );
394 0 : ctx.metrics_base = NULL;
395 :
396 0 : fd_rng_delete( fd_rng_leave( rng ) );
397 :
398 0 : for( ulong i=0UL; i<ctx.link_cnt; i++ ) {
399 0 : fd_fseq_delete( fd_fseq_leave( fseqs[ i ] ) );
400 0 : }
401 0 : }
402 :
403 0 : fd_topo_leave_workspaces( topo );
404 :
405 0 : fd_io_buffered_ostream_flush( &ctx.ostream );
406 0 : fd_io_buffered_ostream_fini( &ctx.ostream );
407 0 : close( fd );
408 0 : }
409 :
410 : action_t fd_action_dump = {
411 : .name = "dump",
412 : .args = dump_cmd_args,
413 : .fn = dump_cmd_fn,
414 : .perm = NULL,
415 : .description = "Dump tango links to a packet capture file",
416 : .is_diagnostic = 1
417 : };
|
