Branch data Line data Source code
1 : : /* Debuginfo-over-http server.
2 : : Copyright (C) 2019-2024 Red Hat, Inc.
3 : : Copyright (C) 2021, 2022 Mark J. Wielaard <mark@klomp.org>
4 : : This file is part of elfutils.
5 : :
6 : : This file is free software; you can redistribute it and/or modify
7 : : it under the terms of the GNU General Public License as published by
8 : : the Free Software Foundation; either version 3 of the License, or
9 : : (at your option) any later version.
10 : :
11 : : elfutils is distributed in the hope that it will be useful, but
12 : : WITHOUT ANY WARRANTY; without even the implied warranty of
13 : : MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 : : GNU General Public License for more details.
15 : :
16 : : You should have received a copy of the GNU General Public License
17 : : along with this program. If not, see <http://www.gnu.org/licenses/>. */
18 : :
19 : :
20 : : /* cargo-cult from libdwfl linux-kernel-modules.c */
21 : : /* In case we have a bad fts we include this before config.h because it
22 : : can't handle _FILE_OFFSET_BITS.
23 : : Everything we need here is fine if its declarations just come first.
24 : : Also, include sys/types.h before fts. On some systems fts.h is not self
25 : : contained. */
26 : : #ifdef BAD_FTS
27 : : #include <sys/types.h>
28 : : #include <fts.h>
29 : : #endif
30 : :
31 : : #ifdef HAVE_CONFIG_H
32 : : #include "config.h"
33 : : #endif
34 : :
35 : : // #define _GNU_SOURCE
36 : : #ifdef HAVE_SCHED_H
37 : : extern "C" {
38 : : #include <sched.h>
39 : : }
40 : : #endif
41 : : #ifdef HAVE_SYS_RESOURCE_H
42 : : extern "C" {
43 : : #include <sys/resource.h>
44 : : }
45 : : #endif
46 : :
47 : : #ifdef HAVE_EXECINFO_H
48 : : extern "C" {
49 : : #include <execinfo.h>
50 : : }
51 : : #endif
52 : : #ifdef HAVE_MALLOC_H
53 : : extern "C" {
54 : : #include <malloc.h>
55 : : }
56 : : #endif
57 : :
58 : : #include "debuginfod.h"
59 : : #include <dwarf.h>
60 : :
61 : : #include <argp.h>
62 : : #ifdef __GNUC__
63 : : #undef __attribute__ /* glibc bug - rhbz 1763325 */
64 : : #endif
65 : :
66 : : #include <unistd.h>
67 : : #include <stdlib.h>
68 : : #include <locale.h>
69 : : #include <pthread.h>
70 : : #include <signal.h>
71 : : #include <sys/stat.h>
72 : : #include <sys/time.h>
73 : : #include <sys/vfs.h>
74 : : #include <unistd.h>
75 : : #include <fcntl.h>
76 : : #include <netdb.h>
77 : : #include <math.h>
78 : : #include <float.h>
79 : :
80 : :
81 : : /* If fts.h is included before config.h, its indirect inclusions may not
82 : : give us the right LFS aliases of these functions, so map them manually. */
83 : : #ifdef BAD_FTS
84 : : #ifdef _FILE_OFFSET_BITS
85 : : #define open open64
86 : : #define fopen fopen64
87 : : #endif
88 : : #else
89 : : #include <sys/types.h>
90 : : #include <fts.h>
91 : : #endif
92 : :
93 : : #include <cstring>
94 : : #include <vector>
95 : : #include <set>
96 : : #include <unordered_set>
97 : : #include <map>
98 : : #include <string>
99 : : #include <iostream>
100 : : #include <iomanip>
101 : : #include <ostream>
102 : : #include <sstream>
103 : : #include <mutex>
104 : : #include <deque>
105 : : #include <condition_variable>
106 : : #include <exception>
107 : : #include <thread>
108 : : // #include <regex> // on rhel7 gcc 4.8, not competent
109 : : #include <regex.h>
110 : : // #include <algorithm>
111 : : using namespace std;
112 : :
113 : : #include <gelf.h>
114 : : #include <libdwelf.h>
115 : :
116 : : #include <microhttpd.h>
117 : :
118 : : #if MHD_VERSION >= 0x00097002
119 : : // libmicrohttpd 0.9.71 broke API
120 : : #define MHD_RESULT enum MHD_Result
121 : : #else
122 : : #define MHD_RESULT int
123 : : #endif
124 : :
125 : : #include <curl/curl.h>
126 : : #include <archive.h>
127 : : #include <archive_entry.h>
128 : : #include <sqlite3.h>
129 : :
130 : : #ifdef __linux__
131 : : #include <sys/syscall.h>
132 : : #endif
133 : :
134 : : #ifdef __linux__
135 : : #define tid() syscall(SYS_gettid)
136 : : #else
137 : : #define tid() pthread_self()
138 : : #endif
139 : :
140 : : extern "C" {
141 : : #include "printversion.h"
142 : : #include "system.h"
143 : : }
144 : :
145 : :
146 : : inline bool
147 : 2024 : string_endswith(const string& haystack, const string& needle)
148 : : {
149 [ + - ]: 2024 : return (haystack.size() >= needle.size() &&
150 : 2024 : equal(haystack.end()-needle.size(), haystack.end(),
151 : 2024 : needle.begin()));
152 : : }
153 : :
154 : :
155 : : // Roll this identifier for every sqlite schema incompatibility.
156 : : #define BUILDIDS "buildids10"
157 : :
158 : : #if SQLITE_VERSION_NUMBER >= 3008000
159 : : #define WITHOUT_ROWID "without rowid"
160 : : #else
161 : : #define WITHOUT_ROWID ""
162 : : #endif
163 : :
164 : : static const char DEBUGINFOD_SQLITE_DDL[] =
165 : : "pragma foreign_keys = on;\n"
166 : : "pragma synchronous = 0;\n" // disable fsync()s - this cache is disposable across a machine crash
167 : : "pragma journal_mode = wal;\n" // https://sqlite.org/wal.html
168 : : "pragma wal_checkpoint = truncate;\n" // clean out any preexisting wal file
169 : : "pragma journal_size_limit = 0;\n" // limit steady state file (between grooming, which also =truncate's)
170 : : "pragma auto_vacuum = incremental;\n" // https://sqlite.org/pragma.html
171 : : "pragma busy_timeout = 1000;\n" // https://sqlite.org/pragma.html
172 : : // NB: all these are overridable with -D option
173 : :
174 : : // Normalization table for interning file names
175 : : "create table if not exists " BUILDIDS "_fileparts (\n"
176 : : " id integer primary key not null,\n"
177 : : " name text unique not null\n"
178 : : " );\n"
179 : : "create table if not exists " BUILDIDS "_files (\n"
180 : : " id integer primary key not null,\n"
181 : : " dirname integer not null,\n"
182 : : " basename integer not null,\n"
183 : : " unique (dirname, basename),\n"
184 : : " foreign key (dirname) references " BUILDIDS "_fileparts(id) on delete cascade,\n"
185 : : " foreign key (basename) references " BUILDIDS "_fileparts(id) on delete cascade\n"
186 : : " );\n"
187 : : "create view if not exists " BUILDIDS "_files_v as\n" // a
188 : : " select f.id, n1.name || '/' || n2.name as name\n"
189 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts n1, " BUILDIDS "_fileparts n2\n"
190 : : " where f.dirname = n1.id and f.basename = n2.id;\n"
191 : :
192 : : // Normalization table for interning buildids
193 : : "create table if not exists " BUILDIDS "_buildids (\n"
194 : : " id integer primary key not null,\n"
195 : : " hex text unique not null);\n"
196 : : // Track the completion of scanning of a given file & sourcetype at given time
197 : : "create table if not exists " BUILDIDS "_file_mtime_scanned (\n"
198 : : " mtime integer not null,\n"
199 : : " file integer not null,\n"
200 : : " size integer not null,\n" // in bytes
201 : : " sourcetype text(1) not null\n"
202 : : " check (sourcetype IN ('F', 'R')),\n"
203 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
204 : : " primary key (file, mtime, sourcetype)\n"
205 : : " ) " WITHOUT_ROWID ";\n"
206 : : "create table if not exists " BUILDIDS "_f_de (\n"
207 : : " buildid integer not null,\n"
208 : : " debuginfo_p integer not null,\n"
209 : : " executable_p integer not null,\n"
210 : : " file integer not null,\n"
211 : : " mtime integer not null,\n"
212 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
213 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
214 : : " primary key (buildid, file, mtime)\n"
215 : : " ) " WITHOUT_ROWID ";\n"
216 : : // Index for faster delete by file identifier
217 : : "create index if not exists " BUILDIDS "_f_de_idx on " BUILDIDS "_f_de (file, mtime);\n"
218 : : "create table if not exists " BUILDIDS "_f_s (\n"
219 : : " buildid integer not null,\n"
220 : : " artifactsrc integer not null,\n"
221 : : " file integer not null,\n" // NB: not necessarily entered into _mtime_scanned
222 : : " mtime integer not null,\n"
223 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
224 : : " foreign key (artifactsrc) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
225 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
226 : : " primary key (buildid, artifactsrc, file, mtime)\n"
227 : : " ) " WITHOUT_ROWID ";\n"
228 : : "create table if not exists " BUILDIDS "_r_de (\n"
229 : : " buildid integer not null,\n"
230 : : " debuginfo_p integer not null,\n"
231 : : " executable_p integer not null,\n"
232 : : " file integer not null,\n"
233 : : " mtime integer not null,\n"
234 : : " content integer not null,\n"
235 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
236 : : " foreign key (content) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
237 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
238 : : " primary key (buildid, debuginfo_p, executable_p, file, content, mtime)\n"
239 : : " ) " WITHOUT_ROWID ";\n"
240 : : // Index for faster delete by archive file identifier
241 : : "create index if not exists " BUILDIDS "_r_de_idx on " BUILDIDS "_r_de (file, mtime);\n"
242 : : "create table if not exists " BUILDIDS "_r_sref (\n" // outgoing dwarf sourcefile references from rpm
243 : : " buildid integer not null,\n"
244 : : " artifactsrc integer not null,\n"
245 : : " foreign key (artifactsrc) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
246 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
247 : : " primary key (buildid, artifactsrc)\n"
248 : : " ) " WITHOUT_ROWID ";\n"
249 : : "create table if not exists " BUILDIDS "_r_sdef (\n" // rpm contents that may satisfy sref
250 : : " file integer not null,\n"
251 : : " mtime integer not null,\n"
252 : : " content integer not null,\n"
253 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
254 : : " foreign key (content) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
255 : : " primary key (content, file, mtime)\n"
256 : : " ) " WITHOUT_ROWID ";\n"
257 : : // create views to glue together some of the above tables, for webapi D queries
258 : : "create view if not exists " BUILDIDS "_query_d as \n"
259 : : "select\n"
260 : : " b.hex as buildid, n.mtime, 'F' as sourcetype, f0.name as source0, n.mtime as mtime, null as source1\n"
261 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_f_de n\n"
262 : : " where b.id = n.buildid and f0.id = n.file and n.debuginfo_p = 1\n"
263 : : "union all select\n"
264 : : " b.hex as buildid, n.mtime, 'R' as sourcetype, f0.name as source0, n.mtime as mtime, f1.name as source1\n"
265 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_r_de n\n"
266 : : " where b.id = n.buildid and f0.id = n.file and f1.id = n.content and n.debuginfo_p = 1\n"
267 : : ";"
268 : : // ... and for E queries
269 : : "create view if not exists " BUILDIDS "_query_e as \n"
270 : : "select\n"
271 : : " b.hex as buildid, n.mtime, 'F' as sourcetype, f0.name as source0, n.mtime as mtime, null as source1\n"
272 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_f_de n\n"
273 : : " where b.id = n.buildid and f0.id = n.file and n.executable_p = 1\n"
274 : : "union all select\n"
275 : : " b.hex as buildid, n.mtime, 'R' as sourcetype, f0.name as source0, n.mtime as mtime, f1.name as source1\n"
276 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_r_de n\n"
277 : : " where b.id = n.buildid and f0.id = n.file and f1.id = n.content and n.executable_p = 1\n"
278 : : ";"
279 : : // ... and for S queries
280 : : "create view if not exists " BUILDIDS "_query_s as \n"
281 : : "select\n"
282 : : " b.hex as buildid, fs.name as artifactsrc, 'F' as sourcetype, f0.name as source0, n.mtime as mtime, null as source1, null as source0ref\n"
283 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v fs, " BUILDIDS "_f_s n\n"
284 : : " where b.id = n.buildid and f0.id = n.file and fs.id = n.artifactsrc\n"
285 : : "union all select\n"
286 : : " b.hex as buildid, f1.name as artifactsrc, 'R' as sourcetype, f0.name as source0, sd.mtime as mtime, f1.name as source1, fsref.name as source0ref\n"
287 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_files_v fsref, "
288 : : " " BUILDIDS "_r_sdef sd, " BUILDIDS "_r_sref sr, " BUILDIDS "_r_de sde\n"
289 : : " where b.id = sr.buildid and f0.id = sd.file and fsref.id = sde.file and f1.id = sd.content\n"
290 : : " and sr.artifactsrc = sd.content and sde.buildid = sr.buildid\n"
291 : : ";"
292 : : // and for startup overview counts
293 : : "drop view if exists " BUILDIDS "_stats;\n"
294 : : "create view if not exists " BUILDIDS "_stats as\n"
295 : : " select 'file d/e' as label,count(*) as quantity from " BUILDIDS "_f_de\n"
296 : : "union all select 'file s',count(*) from " BUILDIDS "_f_s\n"
297 : : "union all select 'archive d/e',count(*) from " BUILDIDS "_r_de\n"
298 : : "union all select 'archive sref',count(*) from " BUILDIDS "_r_sref\n"
299 : : "union all select 'archive sdef',count(*) from " BUILDIDS "_r_sdef\n"
300 : : "union all select 'buildids',count(*) from " BUILDIDS "_buildids\n"
301 : : "union all select 'filenames',count(*) from " BUILDIDS "_files\n"
302 : : "union all select 'fileparts',count(*) from " BUILDIDS "_fileparts\n"
303 : : "union all select 'files scanned (#)',count(*) from " BUILDIDS "_file_mtime_scanned\n"
304 : : "union all select 'files scanned (mb)',coalesce(sum(size)/1024/1024,0) from " BUILDIDS "_file_mtime_scanned\n"
305 : : #if SQLITE_VERSION_NUMBER >= 3016000
306 : : "union all select 'index db size (mb)',page_count*page_size/1024/1024 as size FROM pragma_page_count(), pragma_page_size()\n"
307 : : #endif
308 : : ";\n"
309 : :
310 : : // schema change history & garbage collection
311 : : //
312 : : // XXX: we could have migration queries here to bring prior-schema
313 : : // data over instead of just dropping it. But that could incur
314 : : // doubled storage costs.
315 : : //
316 : : // buildids10: split the _files table into _parts
317 : : "" // <<< we are here
318 : : // buildids9: widen the mtime_scanned table
319 : : "DROP VIEW IF EXISTS buildids9_stats;\n"
320 : : "DROP INDEX IF EXISTS buildids9_r_de_idx;\n"
321 : : "DROP INDEX IF EXISTS buildids9_f_de_idx;\n"
322 : : "DROP VIEW IF EXISTS buildids9_query_s;\n"
323 : : "DROP VIEW IF EXISTS buildids9_query_e;\n"
324 : : "DROP VIEW IF EXISTS buildids9_query_d;\n"
325 : : "DROP TABLE IF EXISTS buildids9_r_sdef;\n"
326 : : "DROP TABLE IF EXISTS buildids9_r_sref;\n"
327 : : "DROP TABLE IF EXISTS buildids9_r_de;\n"
328 : : "DROP TABLE IF EXISTS buildids9_f_s;\n"
329 : : "DROP TABLE IF EXISTS buildids9_f_de;\n"
330 : : "DROP TABLE IF EXISTS buildids9_file_mtime_scanned;\n"
331 : : "DROP TABLE IF EXISTS buildids9_buildids;\n"
332 : : "DROP TABLE IF EXISTS buildids9_files;\n"
333 : : // buildids8: slim the sref table
334 : : "drop table if exists buildids8_f_de;\n"
335 : : "drop table if exists buildids8_f_s;\n"
336 : : "drop table if exists buildids8_r_de;\n"
337 : : "drop table if exists buildids8_r_sref;\n"
338 : : "drop table if exists buildids8_r_sdef;\n"
339 : : "drop table if exists buildids8_file_mtime_scanned;\n"
340 : : "drop table if exists buildids8_files;\n"
341 : : "drop table if exists buildids8_buildids;\n"
342 : : // buildids7: separate _norm table into dense subtype tables
343 : : "drop table if exists buildids7_f_de;\n"
344 : : "drop table if exists buildids7_f_s;\n"
345 : : "drop table if exists buildids7_r_de;\n"
346 : : "drop table if exists buildids7_r_sref;\n"
347 : : "drop table if exists buildids7_r_sdef;\n"
348 : : "drop table if exists buildids7_file_mtime_scanned;\n"
349 : : "drop table if exists buildids7_files;\n"
350 : : "drop table if exists buildids7_buildids;\n"
351 : : // buildids6: drop bolo/rfolo again, represent sources / rpmcontents in main table
352 : : "drop table if exists buildids6_norm;\n"
353 : : "drop table if exists buildids6_files;\n"
354 : : "drop table if exists buildids6_buildids;\n"
355 : : "drop view if exists buildids6;\n"
356 : : // buildids5: redefine srcfile1 column to be '.'-less (for rpms)
357 : : "drop table if exists buildids5_norm;\n"
358 : : "drop table if exists buildids5_files;\n"
359 : : "drop table if exists buildids5_buildids;\n"
360 : : "drop table if exists buildids5_bolo;\n"
361 : : "drop table if exists buildids5_rfolo;\n"
362 : : "drop view if exists buildids5;\n"
363 : : // buildids4: introduce rpmfile RFOLO
364 : : "drop table if exists buildids4_norm;\n"
365 : : "drop table if exists buildids4_files;\n"
366 : : "drop table if exists buildids4_buildids;\n"
367 : : "drop table if exists buildids4_bolo;\n"
368 : : "drop table if exists buildids4_rfolo;\n"
369 : : "drop view if exists buildids4;\n"
370 : : // buildids3*: split out srcfile BOLO
371 : : "drop table if exists buildids3_norm;\n"
372 : : "drop table if exists buildids3_files;\n"
373 : : "drop table if exists buildids3_buildids;\n"
374 : : "drop table if exists buildids3_bolo;\n"
375 : : "drop view if exists buildids3;\n"
376 : : // buildids2: normalized buildid and filenames into interning tables;
377 : : "drop table if exists buildids2_norm;\n"
378 : : "drop table if exists buildids2_files;\n"
379 : : "drop table if exists buildids2_buildids;\n"
380 : : "drop view if exists buildids2;\n"
381 : : // buildids1: made buildid and artifacttype NULLable, to represent cached-negative
382 : : // lookups from sources, e.g. files or rpms that contain no buildid-indexable content
383 : : "drop table if exists buildids1;\n"
384 : : // buildids: original
385 : : "drop table if exists buildids;\n"
386 : : ;
387 : :
388 : : static const char DEBUGINFOD_SQLITE_CLEANUP_DDL[] =
389 : : "pragma wal_checkpoint = truncate;\n" // clean out any preexisting wal file
390 : : ;
391 : :
392 : :
393 : :
394 : :
395 : : /* Name and version of program. */
396 : : ARGP_PROGRAM_VERSION_HOOK_DEF = print_version;
397 : :
398 : : /* Bug report address. */
399 : : ARGP_PROGRAM_BUG_ADDRESS_DEF = PACKAGE_BUGREPORT;
400 : :
401 : : /* Definitions of arguments for argp functions. */
402 : : static const struct argp_option options[] =
403 : : {
404 : : { NULL, 0, NULL, 0, "Scanners:", 1 },
405 : : { "scan-file-dir", 'F', NULL, 0, "Enable ELF/DWARF file scanning.", 0 },
406 : : { "scan-rpm-dir", 'R', NULL, 0, "Enable RPM scanning.", 0 },
407 : : { "scan-deb-dir", 'U', NULL, 0, "Enable DEB scanning.", 0 },
408 : : { "scan-archive", 'Z', "EXT=CMD", 0, "Enable arbitrary archive scanning.", 0 },
409 : : // "source-oci-imageregistry" ...
410 : :
411 : : { NULL, 0, NULL, 0, "Options:", 2 },
412 : : { "logical", 'L', NULL, 0, "Follow symlinks, default=ignore.", 0 },
413 : : { "rescan-time", 't', "SECONDS", 0, "Number of seconds to wait between rescans, 0=disable.", 0 },
414 : : { "groom-time", 'g', "SECONDS", 0, "Number of seconds to wait between database grooming, 0=disable.", 0 },
415 : : { "maxigroom", 'G', NULL, 0, "Run a complete database groom/shrink pass at startup.", 0 },
416 : : { "concurrency", 'c', "NUM", 0, "Limit scanning thread concurrency to NUM, default=#CPUs.", 0 },
417 : : { "connection-pool", 'C', "NUM", OPTION_ARG_OPTIONAL,
418 : : "Use webapi connection pool with NUM threads, default=unlim.", 0 },
419 : : { "include", 'I', "REGEX", 0, "Include files matching REGEX, default=all.", 0 },
420 : : { "exclude", 'X', "REGEX", 0, "Exclude files matching REGEX, default=none.", 0 },
421 : : { "port", 'p', "NUM", 0, "HTTP port to listen on, default 8002.", 0 },
422 : : { "database", 'd', "FILE", 0, "Path to sqlite database.", 0 },
423 : : { "ddl", 'D', "SQL", 0, "Apply extra sqlite ddl/pragma to connection.", 0 },
424 : : { "verbose", 'v', NULL, 0, "Increase verbosity.", 0 },
425 : : { "regex-groom", 'r', NULL, 0,"Uses regexes from -I and -X arguments to groom the database.",0},
426 : : #define ARGP_KEY_FDCACHE_FDS 0x1001
427 : : { "fdcache-fds", ARGP_KEY_FDCACHE_FDS, "NUM", OPTION_HIDDEN, NULL, 0 },
428 : : #define ARGP_KEY_FDCACHE_MBS 0x1002
429 : : { "fdcache-mbs", ARGP_KEY_FDCACHE_MBS, "MB", 0, "Maximum total size of archive file fdcache.", 0 },
430 : : #define ARGP_KEY_FDCACHE_PREFETCH 0x1003
431 : : { "fdcache-prefetch", ARGP_KEY_FDCACHE_PREFETCH, "NUM", 0, "Number of archive files to prefetch into fdcache.", 0 },
432 : : #define ARGP_KEY_FDCACHE_MINTMP 0x1004
433 : : { "fdcache-mintmp", ARGP_KEY_FDCACHE_MINTMP, "NUM", 0, "Minimum free space% on tmpdir.", 0 },
434 : : #define ARGP_KEY_FDCACHE_PREFETCH_MBS 0x1005
435 : : { "fdcache-prefetch-mbs", ARGP_KEY_FDCACHE_PREFETCH_MBS, "MB", OPTION_HIDDEN, NULL, 0},
436 : : #define ARGP_KEY_FDCACHE_PREFETCH_FDS 0x1006
437 : : { "fdcache-prefetch-fds", ARGP_KEY_FDCACHE_PREFETCH_FDS, "NUM", OPTION_HIDDEN, NULL, 0},
438 : : #define ARGP_KEY_FORWARDED_TTL_LIMIT 0x1007
439 : : {"forwarded-ttl-limit", ARGP_KEY_FORWARDED_TTL_LIMIT, "NUM", 0, "Limit of X-Forwarded-For hops, default 8.", 0},
440 : : #define ARGP_KEY_PASSIVE 0x1008
441 : : { "passive", ARGP_KEY_PASSIVE, NULL, 0, "Do not scan or groom, read-only database.", 0 },
442 : : #define ARGP_KEY_DISABLE_SOURCE_SCAN 0x1009
443 : : { "disable-source-scan", ARGP_KEY_DISABLE_SOURCE_SCAN, NULL, 0, "Do not scan dwarf source info.", 0 },
444 : : #define ARGP_SCAN_CHECKPOINT 0x100A
445 : : { "scan-checkpoint", ARGP_SCAN_CHECKPOINT, "NUM", 0, "Number of files scanned before a WAL checkpoint.", 0 },
446 : : { NULL, 0, NULL, 0, NULL, 0 },
447 : : };
448 : :
449 : : /* Short description of program. */
450 : : static const char doc[] = "Serve debuginfo-related content across HTTP from files under PATHs.";
451 : :
452 : : /* Strings for arguments in help texts. */
453 : : static const char args_doc[] = "[PATH ...]";
454 : :
455 : : /* Prototype for option handler. */
456 : : static error_t parse_opt (int key, char *arg, struct argp_state *state);
457 : :
458 : : static unsigned default_concurrency();
459 : :
460 : : /* Data structure to communicate with argp functions. */
461 : : static struct argp argp =
462 : : {
463 : : options, parse_opt, args_doc, doc, NULL, NULL, NULL
464 : : };
465 : :
466 : :
467 : : static string db_path;
468 : : static sqlite3 *db; // single connection, serialized across all our threads!
469 : : static sqlite3 *dbq; // webapi query-servicing readonly connection, serialized ditto!
470 : : static unsigned verbose;
471 : : static volatile sig_atomic_t interrupted = 0;
472 : : static volatile sig_atomic_t forced_rescan_count = 0;
473 : : static volatile sig_atomic_t sigusr1 = 0;
474 : : static volatile sig_atomic_t forced_groom_count = 0;
475 : : static volatile sig_atomic_t sigusr2 = 0;
476 : : static unsigned http_port = 8002;
477 : : static unsigned rescan_s = 300;
478 : : static unsigned groom_s = 86400;
479 : : static bool maxigroom = false;
480 : : static unsigned concurrency = default_concurrency();
481 : : static int connection_pool = 0;
482 : : static set<string> source_paths;
483 : : static bool scan_files = false;
484 : : static map<string,string> scan_archives;
485 : : static vector<string> extra_ddl;
486 : : static regex_t file_include_regex;
487 : : static regex_t file_exclude_regex;
488 : : static bool regex_groom = false;
489 : : static bool traverse_logical;
490 : : static long fdcache_mbs;
491 : : static long fdcache_prefetch;
492 : : static long fdcache_mintmp;
493 : : static unsigned forwarded_ttl_limit = 8;
494 : : static bool scan_source_info = true;
495 : : static string tmpdir;
496 : : static bool passive_p = false;
497 : : static long scan_checkpoint = 256;
498 : :
499 : : static void set_metric(const string& key, double value);
500 : : static void inc_metric(const string& key);
501 : : static void add_metric(const string& metric,
502 : : double value);
503 : : static void set_metric(const string& metric,
504 : : const string& lname, const string& lvalue,
505 : : double value);
506 : : static void inc_metric(const string& metric,
507 : : const string& lname, const string& lvalue);
508 : : static void add_metric(const string& metric,
509 : : const string& lname, const string& lvalue,
510 : : double value);
511 : : static void inc_metric(const string& metric,
512 : : const string& lname, const string& lvalue,
513 : : const string& rname, const string& rvalue);
514 : : static void add_metric(const string& metric,
515 : : const string& lname, const string& lvalue,
516 : : const string& rname, const string& rvalue,
517 : : double value);
518 : :
519 : :
520 : : class tmp_inc_metric { // a RAII style wrapper for exception-safe scoped increment & decrement
521 : : string m, n, v;
522 : : public:
523 : 2608 : tmp_inc_metric(const string& mname, const string& lname, const string& lvalue):
524 [ + - + - ]: 2608 : m(mname), n(lname), v(lvalue)
525 : : {
526 [ + - ]: 2608 : add_metric (m, n, v, 1);
527 [ - - - - : 2608 : }
- - ]
528 : 2608 : ~tmp_inc_metric()
529 : : {
530 : 2608 : add_metric (m, n, v, -1);
531 [ - + - + : 2608 : }
- + ]
532 : : };
533 : :
534 : : class tmp_ms_metric { // a RAII style wrapper for exception-safe scoped timing
535 : : string m, n, v;
536 : : struct timespec ts_start;
537 : : public:
538 : 322294 : tmp_ms_metric(const string& mname, const string& lname, const string& lvalue):
539 [ + - + - ]: 322294 : m(mname), n(lname), v(lvalue)
540 : : {
541 : 322268 : clock_gettime (CLOCK_MONOTONIC, & ts_start);
542 [ - - - - ]: 322400 : }
543 : 322458 : ~tmp_ms_metric()
544 : : {
545 : 322458 : struct timespec ts_end;
546 : 322458 : clock_gettime (CLOCK_MONOTONIC, & ts_end);
547 : 322452 : double deltas = (ts_end.tv_sec - ts_start.tv_sec)
548 : 322452 : + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
549 : :
550 [ + - ]: 322452 : add_metric (m + "_milliseconds_sum", n, v, (deltas*1000.0));
551 [ + + ]: 322463 : inc_metric (m + "_milliseconds_count", n, v);
552 [ + + - + : 597967 : }
- + ]
553 : : };
554 : :
555 : :
556 : : /* Handle program arguments. */
557 : : static error_t
558 : 1148 : parse_opt (int key, char *arg,
559 : : struct argp_state *state __attribute__ ((unused)))
560 : : {
561 : 1148 : int rc;
562 [ + + + + : 1148 : switch (key)
+ + + + +
- + + - -
+ + + + +
+ + + + +
- + + ]
563 : : {
564 : 286 : case 'v': verbose ++; break;
565 : 74 : case 'd':
566 : : /* When using the in-memory database make sure it is shareable,
567 : : so we can open it twice as read/write and read-only. */
568 [ + + ]: 74 : if (strcmp (arg, ":memory:") == 0)
569 : 1162 : db_path = "file::memory:?cache=shared";
570 : : else
571 [ + - ]: 120 : db_path = string(arg);
572 : : break;
573 : 74 : case 'p': http_port = (unsigned) atoi(arg);
574 [ + - ]: 74 : if (http_port == 0 || http_port > 65535)
575 : 0 : argp_failure(state, 1, EINVAL, "port number");
576 : : break;
577 : 50 : case 'F': scan_files = true; break;
578 : 22 : case 'R':
579 [ + - + - : 22 : scan_archives[".rpm"]="cat"; // libarchive groks rpm natively
- + ]
580 : 22 : break;
581 : 14 : case 'U':
582 [ + - + - : 14 : scan_archives[".deb"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
583 [ + - + - : 14 : scan_archives[".ddeb"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
584 [ + - + - : 14 : scan_archives[".ipk"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
585 : : // .udeb too?
586 : 14 : break;
587 : 38 : case 'Z':
588 : 38 : {
589 [ - + ]: 38 : char* extension = strchr(arg, '=');
590 [ - + ]: 38 : if (arg[0] == '\0')
591 : 0 : argp_failure(state, 1, EINVAL, "missing EXT");
592 [ + + ]: 38 : else if (extension)
593 [ + - + - : 20 : scan_archives[string(arg, (extension-arg))]=string(extension+1);
- + - + -
- ]
594 : : else
595 [ + - + - : 18 : scan_archives[string(arg)]=string("cat");
- + - + -
- ]
596 : : }
597 : : break;
598 : 8 : case 'L':
599 [ - + ]: 8 : if (passive_p)
600 : 0 : argp_failure(state, 1, EINVAL, "-L option inconsistent with passive mode");
601 : 8 : traverse_logical = true;
602 : 8 : break;
603 : 0 : case 'D':
604 [ # # ]: 0 : if (passive_p)
605 : 0 : argp_failure(state, 1, EINVAL, "-D option inconsistent with passive mode");
606 [ # # ]: 0 : extra_ddl.push_back(string(arg));
607 : 0 : break;
608 : 58 : case 't':
609 [ - + ]: 58 : if (passive_p)
610 : 0 : argp_failure(state, 1, EINVAL, "-t option inconsistent with passive mode");
611 : 58 : rescan_s = (unsigned) atoi(arg);
612 : 58 : break;
613 : 58 : case 'g':
614 [ - + ]: 58 : if (passive_p)
615 : 0 : argp_failure(state, 1, EINVAL, "-g option inconsistent with passive mode");
616 : 58 : groom_s = (unsigned) atoi(arg);
617 : 58 : break;
618 : 0 : case 'G':
619 [ # # ]: 0 : if (passive_p)
620 : 0 : argp_failure(state, 1, EINVAL, "-G option inconsistent with passive mode");
621 : 0 : maxigroom = true;
622 : 0 : break;
623 : 0 : case 'c':
624 [ # # ]: 0 : if (passive_p)
625 : 0 : argp_failure(state, 1, EINVAL, "-c option inconsistent with passive mode");
626 : 0 : concurrency = (unsigned) atoi(arg);
627 [ # # ]: 0 : if (concurrency < 1) concurrency = 1;
628 : : break;
629 : 6 : case 'C':
630 [ + + ]: 6 : if (arg)
631 : : {
632 : 4 : connection_pool = atoi(arg);
633 [ + - ]: 4 : if (connection_pool < 2)
634 : 0 : argp_failure(state, 1, EINVAL, "-C NUM minimum 2");
635 : : }
636 : : break;
637 : 4 : case 'I':
638 : : // NB: no problem with unconditional free here - an earlier failed regcomp would exit program
639 [ - + ]: 4 : if (passive_p)
640 : 0 : argp_failure(state, 1, EINVAL, "-I option inconsistent with passive mode");
641 : 4 : regfree (&file_include_regex);
642 : 4 : rc = regcomp (&file_include_regex, arg, REG_EXTENDED|REG_NOSUB);
643 [ + - ]: 4 : if (rc != 0)
644 : 0 : argp_failure(state, 1, EINVAL, "regular expression");
645 : : break;
646 : 6 : case 'X':
647 [ - + ]: 6 : if (passive_p)
648 : 0 : argp_failure(state, 1, EINVAL, "-X option inconsistent with passive mode");
649 : 6 : regfree (&file_exclude_regex);
650 : 6 : rc = regcomp (&file_exclude_regex, arg, REG_EXTENDED|REG_NOSUB);
651 [ + - ]: 6 : if (rc != 0)
652 : 0 : argp_failure(state, 1, EINVAL, "regular expression");
653 : : break;
654 : 4 : case 'r':
655 [ - + ]: 4 : if (passive_p)
656 : 0 : argp_failure(state, 1, EINVAL, "-r option inconsistent with passive mode");
657 : 4 : regex_groom = true;
658 : 4 : break;
659 : : case ARGP_KEY_FDCACHE_FDS:
660 : : // deprecated
661 : : break;
662 : 4 : case ARGP_KEY_FDCACHE_MBS:
663 : 4 : fdcache_mbs = atol (arg);
664 : 4 : break;
665 : 2 : case ARGP_KEY_FDCACHE_PREFETCH:
666 : 2 : fdcache_prefetch = atol (arg);
667 : 2 : break;
668 : 2 : case ARGP_KEY_FDCACHE_MINTMP:
669 : 2 : fdcache_mintmp = atol (arg);
670 [ + - ]: 2 : if( fdcache_mintmp > 100 || fdcache_mintmp < 0 )
671 : 0 : argp_failure(state, 1, EINVAL, "fdcache mintmp percent");
672 : : break;
673 : 4 : case ARGP_KEY_FORWARDED_TTL_LIMIT:
674 : 4 : forwarded_ttl_limit = (unsigned) atoi(arg);
675 : 4 : break;
676 : 102 : case ARGP_KEY_ARG:
677 [ + - ]: 102 : source_paths.insert(string(arg));
678 : 102 : break;
679 : : case ARGP_KEY_FDCACHE_PREFETCH_FDS:
680 : : // deprecated
681 : : break;
682 : : case ARGP_KEY_FDCACHE_PREFETCH_MBS:
683 : : // deprecated
684 : : break;
685 : 2 : case ARGP_KEY_PASSIVE:
686 : 2 : passive_p = true;
687 [ + - ]: 2 : if (source_paths.size() > 0
688 [ + - ]: 2 : || maxigroom
689 [ + - ]: 2 : || extra_ddl.size() > 0
690 [ + - + - ]: 4 : || traverse_logical)
691 : : // other conflicting options tricky to check
692 : 0 : argp_failure(state, 1, EINVAL, "inconsistent options with passive mode");
693 : : break;
694 : 0 : case ARGP_KEY_DISABLE_SOURCE_SCAN:
695 : 0 : scan_source_info = false;
696 : 0 : break;
697 : 2 : case ARGP_SCAN_CHECKPOINT:
698 : 2 : scan_checkpoint = atol (arg);
699 [ + - ]: 2 : if (scan_checkpoint < 0)
700 : 0 : argp_failure(state, 1, EINVAL, "scan checkpoint");
701 : : break;
702 : : // case 'h': argp_state_help (state, stderr, ARGP_HELP_LONG|ARGP_HELP_EXIT_OK);
703 : : default: return ARGP_ERR_UNKNOWN;
704 : : }
705 : :
706 : : return 0;
707 : : }
708 : :
709 : :
710 : : ////////////////////////////////////////////////////////////////////////
711 : :
712 : :
713 : : static void add_mhd_response_header (struct MHD_Response *r,
714 : : const char *h, const char *v);
715 : :
716 : : // represent errors that may get reported to an ostream and/or a libmicrohttpd connection
717 : :
718 : 8 : struct reportable_exception
719 : : {
720 : : int code;
721 : : string message;
722 : :
723 [ - - + - : 86 : reportable_exception(int c, const string& m): code(c), message(m) {}
- - + - +
- ]
724 [ - - - - : 602 : reportable_exception(const string& m): code(503), message(m) {}
- - - - +
- - - - -
+ - - - -
- - - + -
- - ]
725 : : reportable_exception(): code(503), message() {}
726 : :
727 : : void report(ostream& o) const; // defined under obatched() class below
728 : :
729 : 632 : MHD_RESULT mhd_send_response(MHD_Connection* c) const {
730 : 1264 : MHD_Response* r = MHD_create_response_from_buffer (message.size(),
731 : 632 : (void*) message.c_str(),
732 : : MHD_RESPMEM_MUST_COPY);
733 : 632 : add_mhd_response_header (r, "Content-Type", "text/plain");
734 : 632 : MHD_RESULT rc = MHD_queue_response (c, code, r);
735 : 632 : MHD_destroy_response (r);
736 : 632 : return rc;
737 : : }
738 : : };
739 : :
740 : :
741 : : struct sqlite_exception: public reportable_exception
742 : : {
743 : 0 : sqlite_exception(int rc, const string& msg):
744 [ # # # # : 0 : reportable_exception(string("sqlite3 error: ") + msg + ": " + string(sqlite3_errstr(rc) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
745 [ # # # # : 0 : inc_metric("error_count","sqlite3",sqlite3_errstr(rc));
# # # # #
# # # # #
# # # # ]
746 [ # # ]: 0 : }
747 : : };
748 : :
749 [ + - - - ]: 4 : struct libc_exception: public reportable_exception
750 : : {
751 : 594 : libc_exception(int rc, const string& msg):
752 [ - + + - : 2376 : reportable_exception(string("libc error: ") + msg + ": " + string(strerror(rc) ?: "?")) {
+ - + - +
- - + - +
- + + - -
- ]
753 [ + - + - : 1188 : inc_metric("error_count","libc",strerror(rc));
+ - + - -
+ + - - -
- - ]
754 [ - - ]: 594 : }
755 : : };
756 : :
757 : :
758 : : struct archive_exception: public reportable_exception
759 : : {
760 : 0 : archive_exception(const string& msg):
761 [ # # # # : 0 : reportable_exception(string("libarchive error: ") + msg) {
# # ]
762 [ # # # # : 0 : inc_metric("error_count","libarchive",msg);
# # # # #
# ]
763 [ # # ]: 0 : }
764 : 0 : archive_exception(struct archive* a, const string& msg):
765 [ # # # # : 0 : reportable_exception(string("libarchive error: ") + msg + ": " + string(archive_error_string(a) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
766 [ # # # # : 0 : inc_metric("error_count","libarchive",msg + ": " + string(archive_error_string(a) ?: "?"));
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# ]
767 [ # # ]: 0 : }
768 : : };
769 : :
770 : :
771 : : struct elfutils_exception: public reportable_exception
772 : : {
773 : 0 : elfutils_exception(int rc, const string& msg):
774 [ # # # # : 0 : reportable_exception(string("elfutils error: ") + msg + ": " + string(elf_errmsg(rc) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
775 [ # # # # : 0 : inc_metric("error_count","elfutils",elf_errmsg(rc));
# # # # #
# # # # #
# # # # ]
776 [ # # ]: 0 : }
777 : : };
778 : :
779 : :
780 : : ////////////////////////////////////////////////////////////////////////
781 : :
782 : : template <typename Payload>
783 : : class workq
784 : : {
785 : : unordered_set<Payload> q; // eliminate duplicates
786 : : mutex mtx;
787 : : condition_variable cv;
788 : : bool dead;
789 : : unsigned idlers; // number of threads busy with wait_idle / done_idle
790 : : unsigned fronters; // number of threads busy with wait_front / done_front
791 : :
792 : : public:
793 : 74 : workq() { dead = false; idlers = 0; fronters = 0; }
794 : 74 : ~workq() {}
795 : :
796 : 1186 : void push_back(const Payload& p)
797 : : {
798 : 1186 : unique_lock<mutex> lock(mtx);
799 [ + - ]: 1186 : q.insert (p);
800 [ + - + - : 2372 : set_metric("thread_work_pending","role","scan", q.size());
+ - + - -
+ - + - -
- - ]
801 : 1186 : cv.notify_all();
802 : 1186 : }
803 : :
804 : : // kill this workqueue, wake up all idlers / scanners
805 : 74 : void nuke() {
806 : 74 : unique_lock<mutex> lock(mtx);
807 : : // optional: q.clear();
808 : 74 : dead = true;
809 : 74 : cv.notify_all();
810 : 74 : }
811 : :
812 : : // clear the workqueue, when scanning is interrupted with USR2
813 : 0 : void clear() {
814 : 0 : unique_lock<mutex> lock(mtx);
815 : 0 : q.clear();
816 [ # # # # : 0 : set_metric("thread_work_pending","role","scan", q.size());
# # # # #
# # # # #
# # ]
817 : : // NB: there may still be some live fronters
818 : 0 : cv.notify_all(); // maybe wake up waiting idlers
819 : 0 : }
820 : :
821 : : // block this scanner thread until there is work to do and no active idler
822 : 1450 : bool wait_front (Payload& p)
823 : : {
824 : 1450 : unique_lock<mutex> lock(mtx);
825 [ + + + + : 5290 : while (!dead && (q.size() == 0 || idlers > 0))
+ + ]
826 [ + - ]: 3840 : cv.wait(lock);
827 [ + + ]: 1450 : if (dead)
828 : : return false;
829 : : else
830 : : {
831 [ + - ]: 1186 : p = * q.begin();
832 : 1186 : q.erase (q.begin());
833 : 1186 : fronters ++; // prevent idlers from starting awhile, even if empty q
834 [ + - + - : 2372 : set_metric("thread_work_pending","role","scan", q.size());
+ - + - -
+ - + - -
- - - - ]
835 : : // NB: don't wake up idlers yet! The consumer is busy
836 : : // processing this element until it calls done_front().
837 : 1186 : return true;
838 : : }
839 : 1450 : }
840 : :
841 : : // notify waitq that scanner thread is done with that last item
842 : 1186 : void done_front ()
843 : : {
844 : 1186 : unique_lock<mutex> lock(mtx);
845 : 1186 : fronters --;
846 [ + + + + ]: 1186 : if (q.size() == 0 && fronters == 0)
847 : 94 : cv.notify_all(); // maybe wake up waiting idlers
848 : 1186 : }
849 : :
850 : : // block this idler thread until there is no work to do
851 : 599 : void wait_idle ()
852 : : {
853 : 599 : unique_lock<mutex> lock(mtx);
854 : 600 : cv.notify_all(); // maybe wake up waiting scanners
855 [ + + + + : 653 : while (!dead && ((q.size() != 0) || fronters > 0))
+ + ]
856 [ + - ]: 53 : cv.wait(lock);
857 [ + - ]: 600 : idlers ++;
858 : 600 : }
859 : :
860 : 528 : void done_idle ()
861 : : {
862 : 528 : unique_lock<mutex> lock(mtx);
863 : 528 : idlers --;
864 : 528 : cv.notify_all(); // maybe wake up waiting scanners, but probably not (shutting down)
865 : 528 : }
866 : : };
867 : :
868 : : typedef struct stat stat_t;
869 : : typedef pair<string,stat_t> scan_payload;
870 : : inline bool operator< (const scan_payload& a, const scan_payload& b)
871 : : {
872 : : return a.first < b.first; // don't bother compare the stat fields
873 : : }
874 : :
875 : : namespace std { // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56480
876 : : template<> struct hash<::scan_payload>
877 : : {
878 : 4876 : std::size_t operator() (const ::scan_payload& p) const noexcept
879 : : {
880 [ + + + + ]: 4876 : return hash<string>()(p.first);
881 : : }
882 : : };
883 : : template<> struct equal_to<::scan_payload>
884 : : {
885 : 505 : std::size_t operator() (const ::scan_payload& a, const ::scan_payload& b) const noexcept
886 : : {
887 [ - + - - ]: 505 : return a.first == b.first;
888 : : }
889 : : };
890 : : }
891 : :
892 : : static workq<scan_payload> scanq; // just a single one
893 : : // producer & idler: thread_main_fts_source_paths()
894 : : // consumer: thread_main_scanner()
895 : : // idler: thread_main_groom()
896 : :
897 : :
898 : : ////////////////////////////////////////////////////////////////////////
899 : :
900 : : // Unique set is a thread-safe structure that lends 'ownership' of a value
901 : : // to a thread. Other threads requesting the same thing are made to wait.
902 : : // It's like a semaphore-on-demand.
903 : : template <typename T>
904 : : class unique_set
905 : : {
906 : : private:
907 : : set<T> values;
908 : : mutex mtx;
909 : : condition_variable cv;
910 : : public:
911 : 57 : unique_set() {}
912 : 57 : ~unique_set() {}
913 : :
914 : 1895 : void acquire(const T& value)
915 : : {
916 : 1895 : unique_lock<mutex> lock(mtx);
917 [ + + ]: 2428 : while (values.find(value) != values.end())
918 [ + - ]: 533 : cv.wait(lock);
919 [ + - ]: 1895 : values.insert(value);
920 : 1895 : }
921 : :
922 : 1895 : void release(const T& value)
923 : : {
924 : 1895 : unique_lock<mutex> lock(mtx);
925 : : // assert (values.find(value) != values.end());
926 : 1895 : values.erase(value);
927 : 1895 : cv.notify_all();
928 : 1895 : }
929 : : };
930 : :
931 : :
932 : : // This is the object that's instantiate to uniquely hold a value in a
933 : : // RAII-pattern way.
934 : : template <typename T>
935 : : class unique_set_reserver
936 : : {
937 : : private:
938 : : unique_set<T>& please_hold;
939 : : T mine;
940 : : public:
941 : 1895 : unique_set_reserver(unique_set<T>& t, const T& value):
942 [ + - - - ]: 1895 : please_hold(t), mine(value) { please_hold.acquire(mine); }
943 [ + - ]: 1895 : ~unique_set_reserver() { please_hold.release(mine); }
944 : : };
945 : :
946 : :
947 : : ////////////////////////////////////////////////////////////////////////
948 : :
949 : : // periodic_barrier is a concurrency control object that lets N threads
950 : : // periodically (based on counter value) agree to wait at a barrier,
951 : : // let one of them carry out some work, then be set free
952 : :
953 : : class periodic_barrier
954 : : {
955 : : private:
956 : : unsigned period; // number of count() reports to trigger barrier activation
957 : : unsigned threads; // number of threads participating
958 : : mutex mtx; // protects all the following fields
959 : : unsigned counter; // count of count() reports in the current generation
960 : : unsigned generation; // barrier activation generation
961 : : unsigned waiting; // number of threads waiting for barrier
962 : : bool dead; // bring out your
963 : : condition_variable cv;
964 : : public:
965 : 66 : periodic_barrier(unsigned t, unsigned p):
966 : 66 : period(p), threads(t), counter(0), generation(0), waiting(0), dead(false) { }
967 : : virtual ~periodic_barrier() {}
968 : :
969 : : virtual void periodic_barrier_work() noexcept = 0;
970 : 66 : void nuke() {
971 : 66 : unique_lock<mutex> lock(mtx);
972 : 66 : dead = true;
973 : 66 : cv.notify_all();
974 : 66 : }
975 : :
976 : 1450 : void count()
977 : : {
978 : 1450 : unique_lock<mutex> lock(mtx);
979 : 1450 : unsigned prev_generation = this->generation;
980 [ + + ]: 1450 : if (counter < period-1) // normal case: counter just freely running
981 : : {
982 : 1334 : counter ++;
983 : 1334 : return;
984 : : }
985 [ + + ]: 116 : else if (counter == period-1) // we're the doer
986 : : {
987 : 30 : counter = period; // entering barrier holding phase
988 : 30 : cv.notify_all();
989 [ + + + + ]: 142 : while (waiting < threads-1 && !dead)
990 [ + - ]: 82 : cv.wait(lock);
991 : : // all other threads are now stuck in the barrier
992 : 30 : this->periodic_barrier_work(); // NB: we're holding the mutex the whole time
993 : : // reset for next barrier, releasing other waiters
994 : 30 : counter = 0;
995 : 30 : generation ++;
996 : 30 : cv.notify_all();
997 : 30 : return;
998 : : }
999 [ + - ]: 86 : else if (counter == period) // we're a waiter, in holding phase
1000 : : {
1001 : 86 : waiting ++;
1002 : 86 : cv.notify_all();
1003 [ + + + + : 313 : while (counter == period && generation == prev_generation && !dead)
+ + ]
1004 [ + - ]: 141 : cv.wait(lock);
1005 : 86 : waiting --;
1006 : 86 : return;
1007 : : }
1008 : 1450 : }
1009 : : };
1010 : :
1011 : :
1012 : :
1013 : : ////////////////////////////////////////////////////////////////////////
1014 : :
1015 : :
1016 : : // Print a standard timestamp.
1017 : : static ostream&
1018 : 41519 : timestamp (ostream &o)
1019 : : {
1020 : 41519 : char datebuf[80];
1021 : 41519 : char *now2 = NULL;
1022 : 41519 : time_t now_t = time(NULL);
1023 : 41522 : struct tm now;
1024 : 41522 : struct tm *nowp = gmtime_r (&now_t, &now);
1025 [ + - ]: 41523 : if (nowp)
1026 : : {
1027 : 41523 : (void) strftime (datebuf, sizeof (datebuf), "%c", nowp);
1028 : 41523 : now2 = datebuf;
1029 : : }
1030 : :
1031 : 41523 : return o << "[" << (now2 ? now2 : "") << "] "
1032 [ - + ]: 41523 : << "(" << getpid () << "/" << tid() << "): ";
1033 : : }
1034 : :
1035 : :
1036 : : // A little class that impersonates an ostream to the extent that it can
1037 : : // take << streaming operations. It batches up the bits into an internal
1038 : : // stringstream until it is destroyed; then flushes to the original ostream.
1039 : : // It adds a timestamp
1040 : : class obatched
1041 : : {
1042 : : private:
1043 : : ostream& o;
1044 : : stringstream stro;
1045 : : static mutex lock;
1046 : : public:
1047 : 41524 : obatched(ostream& oo, bool timestamp_p = true): o(oo)
1048 : : {
1049 [ + + ]: 41525 : if (timestamp_p)
1050 [ + - ]: 41521 : timestamp(stro);
1051 : 41527 : }
1052 : 41518 : ~obatched()
1053 : : {
1054 : 41518 : unique_lock<mutex> do_not_cross_the_streams(obatched::lock);
1055 [ + - ]: 41525 : o << stro.str();
1056 : 41525 : o.flush();
1057 : 41525 : }
1058 : : operator ostream& () { return stro; }
1059 [ - - + - : 29815 : template <typename T> ostream& operator << (const T& t) { stro << t; return stro; }
+ - + - +
- + - + -
- - - - -
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
- - + - +
- + - + -
+ - + - +
- + - - -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - -
- - - - -
+ - - - +
- - - - -
- - + - -
- + - - -
+ - - - -
- + - + -
+ - + - +
- + - - -
- - ]
1060 : : };
1061 : : mutex obatched::lock; // just the one, since cout/cerr iostreams are not thread-safe
1062 : :
1063 : :
1064 : 694 : void reportable_exception::report(ostream& o) const {
1065 [ + - + - ]: 694 : obatched(o) << message << endl;
1066 : 694 : }
1067 : :
1068 : :
1069 : : ////////////////////////////////////////////////////////////////////////
1070 : :
1071 : :
1072 : : // RAII style sqlite prepared-statement holder that matches { } block lifetime
1073 : :
1074 : : struct sqlite_ps
1075 : : {
1076 : : private:
1077 : : sqlite3* db;
1078 : : const string nickname;
1079 : : const string sql;
1080 : : sqlite3_stmt *pp;
1081 : :
1082 : : sqlite_ps(const sqlite_ps&); // make uncopyable
1083 : : sqlite_ps& operator=(const sqlite_ps &); // make unassignable
1084 : :
1085 : : public:
1086 [ + - - - ]: 7063 : sqlite_ps (sqlite3* d, const string& n, const string& s): db(d), nickname(n), sql(s) {
1087 : : // tmp_ms_metric tick("sqlite3","prep",nickname);
1088 [ + + ]: 7063 : if (verbose > 4)
1089 [ + - + - : 348 : obatched(clog) << nickname << " prep " << sql << endl;
+ - + - +
- - - ]
1090 [ + - ]: 7063 : int rc = sqlite3_prepare_v2 (db, sql.c_str(), -1 /* to \0 */, & this->pp, NULL);
1091 [ - + ]: 7063 : if (rc != SQLITE_OK)
1092 [ # # # # ]: 0 : throw sqlite_exception(rc, "prepare " + sql);
1093 : 7063 : }
1094 : :
1095 : 177315 : sqlite_ps& reset()
1096 : : {
1097 [ + - + - : 354622 : tmp_ms_metric tick("sqlite3","reset",nickname);
- + - - ]
1098 [ + - ]: 177307 : sqlite3_reset(this->pp);
1099 : 177318 : return *this;
1100 : 177318 : }
1101 : :
1102 : 203529 : sqlite_ps& bind(int parameter, const string& str)
1103 : : {
1104 [ + + ]: 203529 : if (verbose > 4)
1105 [ + - + - : 1418 : obatched(clog) << nickname << " bind " << parameter << "=" << str << endl;
+ - + - +
- + - ]
1106 : 203529 : int rc = sqlite3_bind_text (this->pp, parameter, str.c_str(), -1, SQLITE_TRANSIENT);
1107 [ - + ]: 203521 : if (rc != SQLITE_OK)
1108 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 bind");
1109 : 203521 : return *this;
1110 : : }
1111 : :
1112 : 52023 : sqlite_ps& bind(int parameter, int64_t value)
1113 : : {
1114 [ + + ]: 52023 : if (verbose > 4)
1115 [ + - + - : 636 : obatched(clog) << nickname << " bind " << parameter << "=" << value << endl;
+ - + - +
- + - ]
1116 : 52023 : int rc = sqlite3_bind_int64 (this->pp, parameter, value);
1117 [ - + ]: 52026 : if (rc != SQLITE_OK)
1118 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 bind");
1119 : 52026 : return *this;
1120 : : }
1121 : :
1122 : : sqlite_ps& bind(int parameter)
1123 : : {
1124 : : if (verbose > 4)
1125 : : obatched(clog) << nickname << " bind " << parameter << "=" << "NULL" << endl;
1126 : : int rc = sqlite3_bind_null (this->pp, parameter);
1127 : : if (rc != SQLITE_OK)
1128 : : throw sqlite_exception(rc, "sqlite3 bind");
1129 : : return *this;
1130 : : }
1131 : :
1132 : :
1133 : 109571 : void step_ok_done() {
1134 [ + - + - : 219122 : tmp_ms_metric tick("sqlite3","step_done",nickname);
- + - - ]
1135 [ + - ]: 109551 : int rc = sqlite3_step (this->pp);
1136 [ + + ]: 109578 : if (verbose > 4)
1137 [ + - + - : 832 : obatched(clog) << nickname << " step-ok-done(" << sqlite3_errstr(rc) << ") " << sql << endl;
+ - + - +
- + - + -
+ - ]
1138 [ + + - + ]: 109578 : if (rc != SQLITE_OK && rc != SQLITE_DONE && rc != SQLITE_ROW)
1139 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 step");
1140 [ + - ]: 109578 : (void) sqlite3_reset (this->pp);
1141 : 109578 : }
1142 : :
1143 : :
1144 : 35566 : int step() {
1145 [ + - + - : 71130 : tmp_ms_metric tick("sqlite3","step",nickname);
- + - - ]
1146 [ + - ]: 35564 : int rc = sqlite3_step (this->pp);
1147 [ + + ]: 35567 : if (verbose > 4)
1148 [ + - + - : 378 : obatched(clog) << nickname << " step(" << sqlite3_errstr(rc) << ") " << sql << endl;
+ - + - +
- + - + -
+ - ]
1149 : 35567 : return rc;
1150 : 35567 : }
1151 : :
1152 [ + + + + ]: 14049 : ~sqlite_ps () { sqlite3_finalize (this->pp); }
1153 [ + - + - : 4107 : operator sqlite3_stmt* () { return this->pp; }
+ - ]
1154 : : };
1155 : :
1156 : :
1157 : : ////////////////////////////////////////////////////////////////////////
1158 : :
1159 : :
1160 : : struct sqlite_checkpoint_pb: public periodic_barrier
1161 : : {
1162 : : // NB: don't use sqlite_ps since it can throw exceptions during ctor etc.
1163 : 66 : sqlite_checkpoint_pb(unsigned t, unsigned p):
1164 : 132 : periodic_barrier(t, p) { }
1165 : :
1166 : 30 : void periodic_barrier_work() noexcept
1167 : : {
1168 : 30 : (void) sqlite3_exec (db, "pragma wal_checkpoint(truncate);", NULL, NULL, NULL);
1169 : 30 : }
1170 : : };
1171 : :
1172 : : static periodic_barrier* scan_barrier = 0; // initialized in main()
1173 : :
1174 : :
1175 : : ////////////////////////////////////////////////////////////////////////
1176 : :
1177 : : // RAII style templated autocloser
1178 : :
1179 : : template <class Payload, class Ignore>
1180 : : struct defer_dtor
1181 : : {
1182 : : public:
1183 : : typedef Ignore (*dtor_fn) (Payload);
1184 : :
1185 : : private:
1186 : : Payload p;
1187 : : dtor_fn fn;
1188 : :
1189 : : public:
1190 : 3322 : defer_dtor(Payload _p, dtor_fn _fn): p(_p), fn(_fn) {}
1191 : 714 : ~defer_dtor() { (void) (*fn)(p); }
1192 : :
1193 : : private:
1194 : : defer_dtor(const defer_dtor<Payload,Ignore>&); // make uncopyable
1195 : : defer_dtor& operator=(const defer_dtor<Payload,Ignore> &); // make unassignable
1196 : : };
1197 : :
1198 : :
1199 : :
1200 : : ////////////////////////////////////////////////////////////////////////
1201 : :
1202 : :
1203 : : static string
1204 : 5228 : header_censor(const string& str)
1205 : : {
1206 : 5228 : string y;
1207 [ + + ]: 55324 : for (auto&& x : str)
1208 : : {
1209 [ + + ]: 50096 : if (isalnum(x) || x == '/' || x == '.' || x == ',' || x == '_' || x == ':')
1210 [ + - ]: 100186 : y += x;
1211 : : }
1212 : 5228 : return y;
1213 : 0 : }
1214 : :
1215 : :
1216 : : static string
1217 : 2614 : conninfo (struct MHD_Connection * conn)
1218 : : {
1219 : 2614 : char hostname[256]; // RFC1035
1220 : 2614 : char servname[256];
1221 : 2614 : int sts = -1;
1222 : :
1223 [ - + ]: 2614 : if (conn == 0)
1224 : 0 : return "internal";
1225 : :
1226 : : /* Look up client address data. */
1227 : 2614 : const union MHD_ConnectionInfo *u = MHD_get_connection_info (conn,
1228 : : MHD_CONNECTION_INFO_CLIENT_ADDRESS);
1229 [ + - ]: 2614 : struct sockaddr *so = u ? u->client_addr : 0;
1230 : :
1231 [ + - - + ]: 2614 : if (so && so->sa_family == AF_INET) {
1232 : 0 : sts = getnameinfo (so, sizeof (struct sockaddr_in),
1233 : : hostname, sizeof (hostname),
1234 : : servname, sizeof (servname),
1235 : : NI_NUMERICHOST | NI_NUMERICSERV);
1236 [ + - ]: 2614 : } else if (so && so->sa_family == AF_INET6) {
1237 : 2614 : struct sockaddr_in6* addr6 = (struct sockaddr_in6*) so;
1238 [ + - + - : 2614 : if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
- + ]
1239 : 2614 : struct sockaddr_in addr4;
1240 : 2614 : memset (&addr4, 0, sizeof(addr4));
1241 : 2614 : addr4.sin_family = AF_INET;
1242 : 2614 : addr4.sin_port = addr6->sin6_port;
1243 : 2614 : memcpy (&addr4.sin_addr.s_addr, addr6->sin6_addr.s6_addr+12, sizeof(addr4.sin_addr.s_addr));
1244 : 2614 : sts = getnameinfo ((struct sockaddr*) &addr4, sizeof (addr4),
1245 : : hostname, sizeof (hostname),
1246 : : servname, sizeof (servname),
1247 : : NI_NUMERICHOST | NI_NUMERICSERV);
1248 : : } else {
1249 : 0 : sts = getnameinfo (so, sizeof (struct sockaddr_in6),
1250 : : hostname, sizeof (hostname),
1251 : : servname, sizeof (servname),
1252 : : NI_NUMERICHOST | NI_NUMERICSERV);
1253 : : }
1254 : : }
1255 : :
1256 [ - + ]: 2614 : if (sts != 0) {
1257 : 0 : hostname[0] = servname[0] = '\0';
1258 : : }
1259 : :
1260 : : // extract headers relevant to administration
1261 [ - + ]: 2614 : const char* user_agent = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "User-Agent") ?: "";
1262 [ + + ]: 2614 : const char* x_forwarded_for = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "X-Forwarded-For") ?: "";
1263 : : // NB: these are untrustworthy, beware if machine-processing log files
1264 : :
1265 [ + - + - : 7842 : return string(hostname) + string(":") + string(servname) +
+ - + - +
- + - - +
- + - + -
+ - + - +
- - - - -
- ]
1266 [ + - + - : 11269 : string(" UA:") + header_censor(string(user_agent)) +
+ - + - +
- - + + +
- + + + -
+ - - -
- ]
1267 [ + - + - : 7854 : string(" XFF:") + header_censor(string(x_forwarded_for));
+ - + + +
+ - - ]
1268 : : }
1269 : :
1270 : :
1271 : :
1272 : : ////////////////////////////////////////////////////////////////////////
1273 : :
1274 : : /* Wrapper for MHD_add_response_header that logs an error if we
1275 : : couldn't add the specified header. */
1276 : : static void
1277 : 8626 : add_mhd_response_header (struct MHD_Response *r,
1278 : : const char *h, const char *v)
1279 : : {
1280 [ - + ]: 8626 : if (MHD_add_response_header (r, h, v) == MHD_NO)
1281 [ # # # # : 0 : obatched(clog) << "Error: couldn't add '" << h << "' header" << endl;
# # ]
1282 : 8626 : }
1283 : :
1284 : : static void
1285 : 1311 : add_mhd_last_modified (struct MHD_Response *resp, time_t mtime)
1286 : : {
1287 : 1311 : struct tm now;
1288 : 1311 : struct tm *nowp = gmtime_r (&mtime, &now);
1289 [ + - ]: 1311 : if (nowp != NULL)
1290 : : {
1291 : 1311 : char datebuf[80];
1292 : 1311 : size_t rc = strftime (datebuf, sizeof (datebuf), "%a, %d %b %Y %T GMT",
1293 : : nowp);
1294 [ + - ]: 1311 : if (rc > 0 && rc < sizeof (datebuf))
1295 : 1311 : add_mhd_response_header (resp, "Last-Modified", datebuf);
1296 : : }
1297 : :
1298 : 1311 : add_mhd_response_header (resp, "Cache-Control", "public");
1299 : 1311 : }
1300 : :
1301 : : // quote all questionable characters of str for safe passage through a sh -c expansion.
1302 : : static string
1303 : 56 : shell_escape(const string& str)
1304 : : {
1305 : 56 : string y;
1306 [ + + ]: 7308 : for (auto&& x : str)
1307 : : {
1308 [ + + + + ]: 7252 : if (! isalnum(x) && x != '/')
1309 [ + - ]: 574 : y += "\\";
1310 [ + - ]: 14504 : y += x;
1311 : : }
1312 : 56 : return y;
1313 : 0 : }
1314 : :
1315 : :
1316 : : // PR25548: Perform POSIX / RFC3986 style path canonicalization on the input string.
1317 : : //
1318 : : // Namely:
1319 : : // // -> /
1320 : : // /foo/../ -> /
1321 : : // /./ -> /
1322 : : //
1323 : : // This mapping is done on dwarf-side source path names, which may
1324 : : // include these constructs, so we can deal with debuginfod clients
1325 : : // that accidentally canonicalize the paths.
1326 : : //
1327 : : // realpath(3) is close but not quite right, because it also resolves
1328 : : // symbolic links. Symlinks at the debuginfod server have nothing to
1329 : : // do with the build-time symlinks, thus they must not be considered.
1330 : : //
1331 : : // see also curl Curl_dedotdotify() aka RFC3986, which we mostly follow here
1332 : : // see also libc __realpath()
1333 : : // see also llvm llvm::sys::path::remove_dots()
1334 : : static string
1335 : 15047 : canon_pathname (const string& input)
1336 : : {
1337 : 15047 : string i = input; // 5.2.4 (1)
1338 : 15047 : string o;
1339 : :
1340 : 143354 : while (i.size() != 0)
1341 : : {
1342 : : // 5.2.4 (2) A
1343 [ + - - + : 256614 : if (i.substr(0,3) == "../")
- + ]
1344 [ # # # # ]: 0 : i = i.substr(3);
1345 [ + - - + : 256613 : else if(i.substr(0,2) == "./")
- + ]
1346 [ # # # # ]: 0 : i = i.substr(2);
1347 : :
1348 : : // 5.2.4 (2) B
1349 [ + - - + : 256613 : else if (i.substr(0,3) == "/./")
+ + ]
1350 [ + - + + ]: 2826 : i = i.substr(2);
1351 [ - + ]: 126659 : else if (i == "/.")
1352 [ # # ]: 0 : i = ""; // no need to handle "/." complete-path-segment case; we're dealing with file names
1353 : :
1354 : : // 5.2.4 (2) C
1355 [ + - - + : 253317 : else if (i.substr(0,4) == "/../") {
+ + ]
1356 [ + - + + ]: 2016 : i = i.substr(3);
1357 : 2016 : string::size_type sl = o.rfind("/");
1358 [ + - ]: 2016 : if (sl != string::npos)
1359 [ + - + - ]: 4032 : o = o.substr(0, sl);
1360 : : else
1361 [ # # ]: 0 : o = "";
1362 [ - + ]: 124642 : } else if (i == "/..")
1363 [ # # ]: 0 : i = ""; // no need to handle "/.." complete-path-segment case; we're dealing with file names
1364 : :
1365 : : // 5.2.4 (2) D
1366 : : // no need to handle these cases; we're dealing with file names
1367 [ - + ]: 124642 : else if (i == ".")
1368 [ # # ]: 0 : i = "";
1369 [ - + ]: 124642 : else if (i == "..")
1370 [ # # ]: 0 : i = "";
1371 : :
1372 : : // POSIX special: map // to /
1373 [ + - - + : 249284 : else if (i.substr(0,2) == "//")
+ + ]
1374 [ + - + + ]: 152 : i = i.substr(1);
1375 : :
1376 : : // 5.2.4 (2) E
1377 : : else {
1378 [ - + ]: 124506 : string::size_type next_slash = i.find("/", (i[0]=='/' ? 1 : 0)); // skip first slash
1379 [ + - + + ]: 249013 : o += i.substr(0, next_slash);
1380 [ + + ]: 124506 : if (next_slash == string::npos)
1381 [ + - + + ]: 158401 : i = "";
1382 : : else
1383 [ + - + + : 206807 : i = i.substr(next_slash);
- - ]
1384 : : }
1385 : : }
1386 : :
1387 [ + - ]: 15047 : return o;
1388 : 15047 : }
1389 : :
1390 : :
1391 : : // Estimate available free space for a given filesystem via statfs(2).
1392 : : // Return true if the free fraction is known to be smaller than the
1393 : : // given minimum percentage. Also update a related metric.
1394 : 2652 : bool statfs_free_enough_p(const string& path, const string& label, long minfree = 0)
1395 : : {
1396 : 2652 : struct statfs sfs;
1397 : 2652 : int rc = statfs(path.c_str(), &sfs);
1398 [ + + ]: 2652 : if (rc == 0)
1399 : : {
1400 : 2602 : double s = (double) sfs.f_bavail / (double) sfs.f_blocks;
1401 [ + - + - : 5204 : set_metric("filesys_free_ratio","purpose",label, s);
- + - - ]
1402 : 2602 : return ((s * 100.0) < minfree);
1403 : : }
1404 : : return false;
1405 : : }
1406 : :
1407 : :
1408 : :
1409 : : // A map-like class that owns a cache of file descriptors (indexed by
1410 : : // file / content names).
1411 : : //
1412 : : // If only it could use fd's instead of file names ... but we can't
1413 : : // dup(2) to create independent descriptors for the same unlinked
1414 : : // files, so would have to use some goofy linux /proc/self/fd/%d
1415 : : // hack such as the following
1416 : :
1417 : : #if 0
1418 : : int superdup(int fd)
1419 : : {
1420 : : #ifdef __linux__
1421 : : char *fdpath = NULL;
1422 : : int rc = asprintf(& fdpath, "/proc/self/fd/%d", fd);
1423 : : int newfd;
1424 : : if (rc >= 0)
1425 : : newfd = open(fdpath, O_RDONLY);
1426 : : else
1427 : : newfd = -1;
1428 : : free (fdpath);
1429 : : return newfd;
1430 : : #else
1431 : : return -1;
1432 : : #endif
1433 : : }
1434 : : #endif
1435 : :
1436 : : class libarchive_fdcache
1437 : : {
1438 : : private:
1439 : : mutex fdcache_lock;
1440 : :
1441 : : typedef pair<string,string> key; // archive, entry
1442 [ + - ]: 208 : struct fdcache_entry
1443 : : {
1444 : : string fd; // file name (probably in $TMPDIR), not an actual open fd (EMFILE)
1445 : : double fd_size_mb; // slightly rounded up megabytes
1446 : : time_t freshness; // when was this entry created or requested last
1447 : : unsigned request_count; // how many requests were made; or 0=prefetch only
1448 : : double latency; // how many seconds it took to extract the file
1449 : : };
1450 : :
1451 : : map<key,fdcache_entry> entries; // optimized for lookup
1452 : : time_t last_cleaning;
1453 : : long max_mbs;
1454 : :
1455 : : public:
1456 : 282 : void set_metrics()
1457 : : {
1458 : 282 : double fdcache_mb = 0.0;
1459 : 282 : double prefetch_mb = 0.0;
1460 : 282 : unsigned fdcache_count = 0;
1461 : 282 : unsigned prefetch_count = 0;
1462 [ + + ]: 1730 : for (auto &i : entries) {
1463 [ + + ]: 1448 : if (i.second.request_count) {
1464 : 1354 : fdcache_mb += i.second.fd_size_mb;
1465 : 1354 : fdcache_count ++;
1466 : : } else {
1467 : 94 : prefetch_mb += i.second.fd_size_mb;
1468 : 94 : prefetch_count ++;
1469 : : }
1470 : : }
1471 [ + - ]: 282 : set_metric("fdcache_bytes", fdcache_mb*1024.0*1024.0);
1472 [ + - ]: 282 : set_metric("fdcache_count", fdcache_count);
1473 [ + - ]: 282 : set_metric("fdcache_prefetch_bytes", prefetch_mb*1024.0*1024.0);
1474 [ + - ]: 282 : set_metric("fdcache_prefetch_count", prefetch_count);
1475 : 282 : }
1476 : :
1477 : 208 : void intern(const string& a, const string& b, string fd, off_t sz,
1478 : : bool requested_p, double lat)
1479 : : {
1480 : 208 : {
1481 : 208 : unique_lock<mutex> lock(fdcache_lock);
1482 : 208 : time_t now = time(NULL);
1483 : : // there is a chance it's already in here, just wasn't found last time
1484 : : // if so, there's nothing to do but count our luck
1485 [ + - ]: 208 : auto i = entries.find(make_pair(a,b));
1486 [ - + ]: 208 : if (i != entries.end())
1487 : : {
1488 [ # # # # : 0 : inc_metric("fdcache_op_count","op","redundant_intern");
# # # # #
# # # # #
# # ]
1489 [ # # ]: 0 : if (requested_p) i->second.request_count ++; // repeat prefetch doesn't count
1490 : 0 : i->second.freshness = now;
1491 : : // We need to nuke the temp file, since interning passes
1492 : : // responsibility over the path to this structure. It is
1493 : : // possible that the caller still has an fd open, but that's
1494 : : // OK.
1495 : 0 : unlink (fd.c_str());
1496 : 0 : return;
1497 : : }
1498 : 208 : double mb = (sz+65535)/1048576.0; // round up to 64K block
1499 : 208 : fdcache_entry n = { .fd=fd, .fd_size_mb=mb,
1500 : 208 : .freshness=now, .request_count = requested_p?1U:0U,
1501 [ + - + + ]: 208 : .latency=lat};
1502 [ + - + - : 208 : entries.insert(make_pair(make_pair(a,b),n));
+ - ]
1503 : :
1504 [ + + ]: 208 : if (requested_p)
1505 [ + - + - : 320 : inc_metric("fdcache_op_count","op","enqueue");
+ - + - -
+ - + - -
- - ]
1506 : : else
1507 [ + - + - : 144 : inc_metric("fdcache_op_count","op","prefetch_enqueue");
+ - + - -
+ + - - -
- - ]
1508 : :
1509 [ + + ]: 208 : if (verbose > 3)
1510 [ + - + - : 534 : obatched(clog) << "fdcache interned a=" << a << " b=" << b
- - ]
1511 [ + - + - : 178 : << " fd=" << fd << " mb=" << mb << " front=" << requested_p
+ - + - +
- + - + -
+ - ]
1512 [ + - + - : 178 : << " latency=" << lat << endl;
+ - ]
1513 : :
1514 [ + - ]: 208 : set_metrics();
1515 : 208 : }
1516 : :
1517 : : // NB: we age the cache at lookup time too
1518 [ + - - + : 208 : if (statfs_free_enough_p(tmpdir, "tmpdir", fdcache_mintmp))
- + ]
1519 : : {
1520 [ # # # # : 0 : inc_metric("fdcache_op_count","op","emerg-flush");
# # # # #
# # # #
# ]
1521 [ # # ]: 0 : obatched(clog) << "fdcache emergency flush for filling tmpdir" << endl;
1522 : 0 : this->limit(0); // emergency flush
1523 : : }
1524 : : else // age cache normally
1525 : 208 : this->limit(max_mbs);
1526 : : }
1527 : :
1528 : 740 : int lookup(const string& a, const string& b)
1529 : : {
1530 : 740 : int fd = -1;
1531 : 740 : {
1532 : 740 : unique_lock<mutex> lock(fdcache_lock);
1533 [ + - ]: 740 : auto i = entries.find(make_pair(a,b));
1534 [ + + ]: 740 : if (i != entries.end())
1535 : : {
1536 [ + + ]: 576 : if (i->second.request_count == 0) // was a prefetch!
1537 : : {
1538 [ + - + - ]: 16 : inc_metric("fdcache_prefetch_saved_milliseconds_count");
1539 [ + - + - ]: 32 : add_metric("fdcache_prefetch_saved_milliseconds_sum", i->second.latency*1000.);
1540 : : }
1541 : 576 : i->second.request_count ++;
1542 : 576 : i->second.freshness = time(NULL);
1543 : : // brag about our success
1544 [ + - + - : 1152 : inc_metric("fdcache_op_count","op","prefetch_access"); // backward compat
+ - + - -
+ - + - -
- - ]
1545 [ + - + - ]: 576 : inc_metric("fdcache_saved_milliseconds_count");
1546 [ + - + - ]: 576 : add_metric("fdcache_saved_milliseconds_sum", i->second.latency*1000.);
1547 [ + - ]: 740 : fd = open(i->second.fd.c_str(), O_RDONLY);
1548 : : }
1549 : 0 : }
1550 : :
1551 [ + + ]: 740 : if (fd >= 0)
1552 [ + - + - : 1152 : inc_metric("fdcache_op_count","op","lookup_hit");
+ - - + -
+ - - -
- ]
1553 : : else
1554 [ + - + - : 328 : inc_metric("fdcache_op_count","op","lookup_miss");
+ - - + -
+ - - -
- ]
1555 : :
1556 : : // NB: no need to age the cache after just a lookup
1557 : :
1558 : 740 : return fd;
1559 : : }
1560 : :
1561 : 226 : int probe(const string& a, const string& b) // just a cache residency check - don't modify state, don't open
1562 : : {
1563 : 226 : unique_lock<mutex> lock(fdcache_lock);
1564 [ + - ]: 226 : auto i = entries.find(make_pair(a,b));
1565 [ + + ]: 226 : if (i != entries.end()) {
1566 [ + - + - : 52 : inc_metric("fdcache_op_count","op","probe_hit");
+ - + - -
+ - + - -
- - ]
1567 : 26 : return true;
1568 : : } else {
1569 [ + - + - : 400 : inc_metric("fdcache_op_count","op","probe_miss");
+ - + - -
+ - + - -
- - ]
1570 : 200 : return false;
1571 : : }
1572 : 226 : }
1573 : :
1574 : 0 : void clear(const string& a, const string& b)
1575 : : {
1576 : 0 : unique_lock<mutex> lock(fdcache_lock);
1577 [ # # ]: 0 : auto i = entries.find(make_pair(a,b));
1578 [ # # ]: 0 : if (i != entries.end()) {
1579 [ # # # # : 0 : inc_metric("fdcache_op_count","op",
# # # # #
# # # # #
# # # # ]
1580 [ # # ]: 0 : i->second.request_count > 0 ? "clear" : "prefetch_clear");
1581 : 0 : unlink (i->second.fd.c_str());
1582 : 0 : entries.erase(i);
1583 [ # # ]: 0 : set_metrics();
1584 : 0 : return;
1585 : : }
1586 : 0 : }
1587 : :
1588 : 356 : void limit(long maxmbs, bool metrics_p = true)
1589 : : {
1590 : 356 : time_t now = time(NULL);
1591 : :
1592 : : // avoid overly frequent limit operations
1593 [ + + + + ]: 356 : if (maxmbs > 0 && (now - this->last_cleaning) < 10) // probably not worth parametrizing
1594 : 208 : return;
1595 : 148 : this->last_cleaning = now;
1596 : :
1597 [ + + + - ]: 148 : if (verbose > 3 && (this->max_mbs != maxmbs))
1598 [ + - + - ]: 192 : obatched(clog) << "fdcache limited to maxmbs=" << maxmbs << endl;
1599 : :
1600 : 148 : unique_lock<mutex> lock(fdcache_lock);
1601 : :
1602 : 148 : this->max_mbs = maxmbs;
1603 : 148 : double total_mb = 0.0;
1604 : :
1605 : 148 : map<double, pair<string,string>> sorted_entries;
1606 [ + + ]: 356 : for (auto &i: entries)
1607 : : {
1608 : 208 : total_mb += i.second.fd_size_mb;
1609 : :
1610 : : // need a scalar quantity that combines these inputs in a sensible way:
1611 : : //
1612 : : // 1) freshness of this entry (last time it was accessed)
1613 : : // 2) size of this entry
1614 : : // 3) number of times it has been accessed (or if just prefetched with 0 accesses)
1615 : : // 4) latency it required to extract
1616 : : //
1617 : : // The lower the "score", the earlier garbage collection will
1618 : : // nuke it, so to prioritize entries for preservation, the
1619 : : // score should be higher, and vice versa.
1620 : 208 : time_t factor_1_freshness = (now - i.second.freshness); // seconds
1621 : 208 : double factor_2_size = i.second.fd_size_mb; // megabytes
1622 : 208 : unsigned factor_3_accesscount = i.second.request_count; // units
1623 : 208 : double factor_4_latency = i.second.latency; // seconds
1624 : :
1625 : : #if 0
1626 : : double score = - factor_1_freshness; // simple LRU
1627 : : #endif
1628 : :
1629 [ + + ]: 208 : double score = 0.
1630 : 208 : - log1p(factor_1_freshness) // penalize old file
1631 : 208 : - log1p(factor_2_size) // penalize large file
1632 : 208 : + factor_4_latency * factor_3_accesscount; // reward slow + repeatedly read files
1633 : :
1634 [ + + ]: 208 : if (verbose > 4)
1635 [ + - ]: 64 : obatched(clog) << "fdcache scored score=" << score
1636 [ + - + - ]: 64 : << " a=" << i.first.first << " b=" << i.first.second
1637 [ + - + - : 96 : << " f1=" << factor_1_freshness << " f2=" << factor_2_size
+ - + - +
- + - +
- ]
1638 [ + - + - : 32 : << " f3=" << factor_3_accesscount << " f4=" << factor_4_latency
+ - + - +
- ]
1639 : 32 : << endl;
1640 : :
1641 [ + - + - ]: 416 : sorted_entries.insert(make_pair(score, i.first));
1642 : : }
1643 : :
1644 : 148 : unsigned cleaned = 0;
1645 : 148 : unsigned entries_original = entries.size();
1646 : 148 : double cleaned_score_min = DBL_MAX;
1647 : 148 : double cleaned_score_max = DBL_MIN;
1648 : :
1649 : : // drop as many entries[] as needed to bring total mb down to the threshold
1650 [ + + ]: 356 : for (auto &i: sorted_entries) // in increasing score order!
1651 : : {
1652 [ - + ]: 208 : if (this->max_mbs > 0 // if this is not a "clear entire table"
1653 [ # # ]: 0 : && total_mb < this->max_mbs) // we've cleared enough to meet threshold
1654 : : break; // stop clearing
1655 : :
1656 [ - + ]: 208 : auto j = entries.find(i.second);
1657 [ - + ]: 208 : if (j == entries.end())
1658 : 0 : continue; // should not happen
1659 : :
1660 [ + + ]: 208 : if (cleaned == 0)
1661 : 32 : cleaned_score_min = i.first;
1662 : 208 : cleaned++;
1663 : 208 : cleaned_score_max = i.first;
1664 : :
1665 [ + + ]: 208 : if (verbose > 3)
1666 [ + - + - ]: 534 : obatched(clog) << "fdcache evicted score=" << i.first
1667 [ + - + - ]: 356 : << " a=" << i.second.first << " b=" << i.second.second
1668 [ + - + - : 534 : << " fd=" << j->second.fd << " mb=" << j->second.fd_size_mb
+ - + - +
- + - ]
1669 [ + - + - : 178 : << " rq=" << j->second.request_count << " lat=" << j->second.latency
+ - + - ]
1670 [ + - + - : 178 : << " fr=" << (now - j->second.freshness)
+ - ]
1671 : 178 : << endl;
1672 [ - + ]: 208 : if (metrics_p)
1673 [ # # # # : 0 : inc_metric("fdcache_op_count","op","evict");
# # # # #
# # # # #
# # ]
1674 : :
1675 : 208 : total_mb -= j->second.fd_size_mb;
1676 : 208 : unlink (j->second.fd.c_str());
1677 : 208 : entries.erase(j);
1678 : : }
1679 : :
1680 [ + + ]: 148 : if (metrics_p)
1681 [ + - + - : 148 : inc_metric("fdcache_op_count","op","evict_cycle");
+ - + - -
+ - + - -
- - ]
1682 : :
1683 [ + - + + ]: 148 : if (verbose > 1 && cleaned > 0)
1684 : : {
1685 [ + - + - : 96 : obatched(clog) << "fdcache evicted num=" << cleaned << " of=" << entries_original
+ - + - ]
1686 [ + - + - : 32 : << " min=" << cleaned_score_min << " max=" << cleaned_score_max
+ - + - +
- ]
1687 : 32 : << endl;
1688 : : }
1689 : :
1690 [ + + + - ]: 148 : if (metrics_p) set_metrics();
1691 : 148 : }
1692 : :
1693 : :
1694 : 74 : ~libarchive_fdcache()
1695 : : {
1696 : : // unlink any fdcache entries in $TMPDIR
1697 : : // don't update metrics; those globals may be already destroyed
1698 : 74 : limit(0, false);
1699 : 74 : }
1700 : : };
1701 : : static libarchive_fdcache fdcache;
1702 : :
1703 : : /* Search ELF_FD for an ELF/DWARF section with name SECTION.
1704 : : If found copy the section to a temporary file and return
1705 : : its file descriptor, otherwise return -1.
1706 : :
1707 : : The temporary file's mtime will be set to PARENT_MTIME.
1708 : : B_SOURCE should be a description of the parent file suitable
1709 : : for printing to the log. */
1710 : :
1711 : : static int
1712 : 12 : extract_section (int elf_fd, int64_t parent_mtime,
1713 : : const string& b_source, const string& section,
1714 : : const timespec& extract_begin)
1715 : : {
1716 : : /* Search the fdcache. */
1717 : 12 : struct stat fs;
1718 : 12 : int fd = fdcache.lookup (b_source, section);
1719 [ - + ]: 12 : if (fd >= 0)
1720 : : {
1721 [ # # ]: 0 : if (fstat (fd, &fs) != 0)
1722 : : {
1723 [ # # ]: 0 : if (verbose)
1724 [ # # ]: 0 : obatched (clog) << "cannot fstate fdcache "
1725 [ # # # # : 0 : << b_source << " " << section << endl;
# # ]
1726 : 0 : close (fd);
1727 : 0 : return -1;
1728 : : }
1729 [ # # ]: 0 : if ((int64_t) fs.st_mtime != parent_mtime)
1730 : : {
1731 [ # # ]: 0 : if (verbose)
1732 [ # # ]: 0 : obatched(clog) << "mtime mismatch for "
1733 [ # # # # : 0 : << b_source << " " << section << endl;
# # ]
1734 : 0 : close (fd);
1735 : 0 : return -1;
1736 : : }
1737 : : /* Success. */
1738 : : return fd;
1739 : : }
1740 : :
1741 : 12 : Elf *elf = elf_begin (elf_fd, ELF_C_READ_MMAP_PRIVATE, NULL);
1742 [ - + ]: 12 : if (elf == NULL)
1743 : : return -1;
1744 : :
1745 : : /* Try to find the section and copy the contents into a separate file. */
1746 : 12 : try
1747 : : {
1748 : 12 : size_t shstrndx;
1749 [ + - ]: 12 : int rc = elf_getshdrstrndx (elf, &shstrndx);
1750 [ - + ]: 12 : if (rc < 0)
1751 [ # # # # ]: 0 : throw elfutils_exception (rc, "getshdrstrndx");
1752 : :
1753 : : Elf_Scn *scn = NULL;
1754 : 432 : while (true)
1755 : : {
1756 [ + - ]: 222 : scn = elf_nextscn (elf, scn);
1757 [ + - ]: 222 : if (scn == NULL)
1758 : : break;
1759 : 222 : GElf_Shdr shdr_storage;
1760 [ + - ]: 222 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_storage);
1761 [ + - ]: 222 : if (shdr == NULL)
1762 : : break;
1763 : :
1764 [ + - ]: 222 : const char *scn_name = elf_strptr (elf, shstrndx, shdr->sh_name);
1765 [ + - ]: 222 : if (scn_name == NULL)
1766 : : break;
1767 [ + + ]: 222 : if (scn_name == section)
1768 : : {
1769 : 12 : Elf_Data *data = NULL;
1770 : :
1771 : : /* We found the desired section. */
1772 [ + - ]: 12 : data = elf_rawdata (scn, NULL);
1773 [ - + ]: 12 : if (data == NULL)
1774 [ # # # # : 0 : throw elfutils_exception (elf_errno (), "elfraw_data");
# # ]
1775 [ + + ]: 12 : if (data->d_buf == NULL)
1776 : : {
1777 [ + - + - ]: 8 : obatched(clog) << "section " << section
1778 [ + - + - ]: 4 : << " is empty" << endl;
1779 : 4 : break;
1780 : : }
1781 : :
1782 : : /* Create temporary file containing the section. */
1783 : 8 : char *tmppath = NULL;
1784 : 8 : rc = asprintf (&tmppath, "%s/debuginfod-section.XXXXXX", tmpdir.c_str());
1785 [ - + ]: 8 : if (rc < 0)
1786 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
1787 : 8 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
1788 [ + - ]: 8 : fd = mkstemp (tmppath);
1789 [ - + ]: 8 : if (fd < 0)
1790 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
1791 : :
1792 [ + - ]: 8 : ssize_t res = write_retry (fd, data->d_buf, data->d_size);
1793 [ + - - + ]: 8 : if (res < 0 || (size_t) res != data->d_size) {
1794 [ # # ]: 0 : close (fd);
1795 : 0 : unlink (tmppath);
1796 [ # # # # ]: 0 : throw libc_exception (errno, "cannot write to temporary file");
1797 : : }
1798 : :
1799 : : /* Set mtime to be the same as the parent file's mtime. */
1800 : 8 : struct timespec tvs[2];
1801 [ - + ]: 8 : if (fstat (elf_fd, &fs) != 0) {
1802 [ # # ]: 0 : close (fd);
1803 : 0 : unlink (tmppath);
1804 [ # # # # ]: 0 : throw libc_exception (errno, "cannot fstat file");
1805 : : }
1806 : :
1807 : 8 : tvs[0].tv_sec = 0;
1808 : 8 : tvs[0].tv_nsec = UTIME_OMIT;
1809 : 8 : tvs[1] = fs.st_mtim;
1810 : 8 : (void) futimens (fd, tvs);
1811 : :
1812 : 8 : struct timespec extract_end;
1813 : 8 : clock_gettime (CLOCK_MONOTONIC, &extract_end);
1814 : 8 : double extract_time = (extract_end.tv_sec - extract_begin.tv_sec)
1815 : 8 : + (extract_end.tv_nsec - extract_begin.tv_nsec)/1.e9;
1816 : :
1817 : : /* Add to fdcache. */
1818 [ + - + - ]: 8 : fdcache.intern (b_source, section, tmppath, data->d_size, true, extract_time);
1819 : 8 : break;
1820 : 12 : }
1821 : 210 : }
1822 : : }
1823 [ - - ]: 0 : catch (const reportable_exception &e)
1824 : : {
1825 [ - - ]: 0 : e.report (clog);
1826 [ - - ]: 0 : close (fd);
1827 : 0 : fd = -1;
1828 : 0 : }
1829 : :
1830 : 12 : elf_end (elf);
1831 : : return fd;
1832 : : }
1833 : :
1834 : : static struct MHD_Response*
1835 : 583 : handle_buildid_f_match (bool internal_req_t,
1836 : : int64_t b_mtime,
1837 : : const string& b_source0,
1838 : : const string& section,
1839 : : int *result_fd)
1840 : : {
1841 : 583 : (void) internal_req_t; // ignored
1842 : :
1843 : 583 : struct timespec extract_begin;
1844 : 583 : clock_gettime (CLOCK_MONOTONIC, &extract_begin);
1845 : :
1846 : 583 : int fd = open(b_source0.c_str(), O_RDONLY);
1847 [ - + ]: 583 : if (fd < 0)
1848 [ # # # # : 0 : throw libc_exception (errno, string("open ") + b_source0);
# # # # ]
1849 : :
1850 : : // NB: use manual close(2) in error case instead of defer_dtor, because
1851 : : // in the normal case, we want to hand the fd over to libmicrohttpd for
1852 : : // file transfer.
1853 : :
1854 : 583 : struct stat s;
1855 : 583 : int rc = fstat(fd, &s);
1856 [ - + ]: 583 : if (rc < 0)
1857 : : {
1858 : 0 : close(fd);
1859 [ # # # # : 0 : throw libc_exception (errno, string("fstat ") + b_source0);
# # # # ]
1860 : : }
1861 : :
1862 [ - + ]: 583 : if ((int64_t) s.st_mtime != b_mtime)
1863 : : {
1864 [ # # ]: 0 : if (verbose)
1865 [ # # # # ]: 0 : obatched(clog) << "mtime mismatch for " << b_source0 << endl;
1866 : 0 : close(fd);
1867 : 0 : return 0;
1868 : : }
1869 : :
1870 [ + + ]: 583 : if (!section.empty ())
1871 : : {
1872 : 6 : int scn_fd = extract_section (fd, s.st_mtime, b_source0, section, extract_begin);
1873 : 6 : close (fd);
1874 : :
1875 [ + + ]: 6 : if (scn_fd >= 0)
1876 : 4 : fd = scn_fd;
1877 : : else
1878 : : {
1879 [ + - ]: 2 : if (verbose)
1880 [ + - ]: 6 : obatched (clog) << "cannot find section " << section
1881 [ + - + - : 2 : << " for " << b_source0 << endl;
+ - ]
1882 : 2 : return 0;
1883 : : }
1884 : :
1885 : 4 : rc = fstat(fd, &s);
1886 [ - + ]: 4 : if (rc < 0)
1887 : : {
1888 : 0 : close (fd);
1889 [ # # # # : 0 : throw libc_exception (errno, string ("fstat ") + b_source0
# # # # #
# # # #
# ]
1890 [ # # # # : 0 : + string (" ") + section);
# # # # #
# ]
1891 : : }
1892 : : }
1893 : :
1894 : 581 : struct MHD_Response* r = MHD_create_response_from_fd ((uint64_t) s.st_size, fd);
1895 [ + - + - : 1162 : inc_metric ("http_responses_total","result","file");
+ - - + -
+ - - -
- ]
1896 [ - + ]: 581 : if (r == 0)
1897 : : {
1898 [ # # ]: 0 : if (verbose)
1899 [ # # ]: 0 : obatched(clog) << "cannot create fd-response for " << b_source0
1900 [ # # # # : 0 : << " section=" << section << endl;
# # ]
1901 : 0 : close(fd);
1902 : : }
1903 : : else
1904 : : {
1905 : 581 : add_mhd_response_header (r, "Content-Type", "application/octet-stream");
1906 [ + - ]: 581 : add_mhd_response_header (r, "X-DEBUGINFOD-SIZE",
1907 : 581 : to_string(s.st_size).c_str());
1908 : 581 : add_mhd_response_header (r, "X-DEBUGINFOD-FILE", b_source0.c_str());
1909 : 581 : add_mhd_last_modified (r, s.st_mtime);
1910 [ + - ]: 581 : if (verbose > 1)
1911 [ + - + - : 1162 : obatched(clog) << "serving file " << b_source0 << " section=" << section << endl;
+ - + - ]
1912 : : /* libmicrohttpd will close it. */
1913 [ - + ]: 581 : if (result_fd)
1914 : 581 : *result_fd = fd;
1915 : : }
1916 : :
1917 : : return r;
1918 : : }
1919 : :
1920 : : // For security/portability reasons, many distro-package archives have
1921 : : // a "./" in front of path names; others have nothing, others have
1922 : : // "/". Canonicalize them all to a single leading "/", with the
1923 : : // assumption that this matches the dwarf-derived file names too.
1924 : 974 : string canonicalized_archive_entry_pathname(struct archive_entry *e)
1925 : : {
1926 : 974 : string fn = archive_entry_pathname(e);
1927 [ - + ]: 974 : if (fn.size() == 0)
1928 : 0 : return fn;
1929 [ - + ]: 974 : if (fn[0] == '/')
1930 : 0 : return fn;
1931 [ + + ]: 974 : if (fn[0] == '.')
1932 [ + - ]: 776 : return fn.substr(1);
1933 : : else
1934 [ + - + - : 396 : return string("/")+fn;
- - ]
1935 : 974 : }
1936 : :
1937 : :
1938 : :
1939 : : static struct MHD_Response*
1940 : 786 : handle_buildid_r_match (bool internal_req_p,
1941 : : int64_t b_mtime,
1942 : : const string& b_source0,
1943 : : const string& b_source1,
1944 : : const string& section,
1945 : : int *result_fd)
1946 : : {
1947 : 786 : struct timespec extract_begin;
1948 : 786 : clock_gettime (CLOCK_MONOTONIC, &extract_begin);
1949 : :
1950 : 786 : struct stat fs;
1951 : 786 : int rc = stat (b_source0.c_str(), &fs);
1952 [ + + ]: 786 : if (rc != 0)
1953 [ + - + - : 116 : throw libc_exception (errno, string("stat ") + b_source0);
+ - - + ]
1954 : :
1955 [ - + ]: 728 : if ((int64_t) fs.st_mtime != b_mtime)
1956 : : {
1957 [ # # ]: 0 : if (verbose)
1958 [ # # # # ]: 0 : obatched(clog) << "mtime mismatch for " << b_source0 << endl;
1959 : 0 : return 0;
1960 : : }
1961 : :
1962 : : // check for a match in the fdcache first
1963 : 728 : int fd = fdcache.lookup(b_source0, b_source1);
1964 [ + + ]: 728 : while (fd >= 0) // got one!; NB: this is really an if() with a possible branch out to the end
1965 : : {
1966 : 576 : rc = fstat(fd, &fs);
1967 [ - + ]: 576 : if (rc < 0) // disappeared?
1968 : : {
1969 [ # # ]: 0 : if (verbose)
1970 [ # # # # ]: 0 : obatched(clog) << "cannot fstat fdcache " << b_source0 << endl;
1971 : 0 : close(fd);
1972 : 0 : fdcache.clear(b_source0, b_source1);
1973 : : break; // branch out of if "loop", to try new libarchive fetch attempt
1974 : : }
1975 : :
1976 [ + + ]: 576 : if (!section.empty ())
1977 : : {
1978 [ + - + - ]: 2 : int scn_fd = extract_section (fd, fs.st_mtime,
1979 [ + - - + ]: 4 : b_source0 + ":" + b_source1,
1980 : : section, extract_begin);
1981 : 2 : close (fd);
1982 [ - + ]: 2 : if (scn_fd >= 0)
1983 : 0 : fd = scn_fd;
1984 : : else
1985 : : {
1986 [ + - ]: 2 : if (verbose)
1987 [ + - ]: 6 : obatched (clog) << "cannot find section " << section
1988 : : << " for archive " << b_source0
1989 [ + - + - : 2 : << " file " << b_source1 << endl;
+ - + - +
- ]
1990 : 2 : return 0;
1991 : : }
1992 : :
1993 : 0 : rc = fstat(fd, &fs);
1994 [ # # ]: 0 : if (rc < 0)
1995 : : {
1996 : 0 : close (fd);
1997 [ # # # # ]: 0 : throw libc_exception (errno,
1998 [ # # # # : 0 : string ("fstat archive ") + b_source0 + string (" file ") + b_source1
# # # # #
# # # # #
# # # # #
# # # #
# ]
1999 [ # # # # : 0 : + string (" section ") + section);
# # # # #
# ]
2000 : : }
2001 : : }
2002 : :
2003 : 574 : struct MHD_Response* r = MHD_create_response_from_fd (fs.st_size, fd);
2004 [ - + ]: 574 : if (r == 0)
2005 : : {
2006 [ # # ]: 0 : if (verbose)
2007 [ # # # # ]: 0 : obatched(clog) << "cannot create fd-response for " << b_source0 << endl;
2008 : 0 : close(fd);
2009 : : break; // branch out of if "loop", to try new libarchive fetch attempt
2010 : : }
2011 : :
2012 [ + - + - : 1148 : inc_metric ("http_responses_total","result","archive fdcache");
+ - - + -
+ - - -
- ]
2013 : :
2014 : 574 : add_mhd_response_header (r, "Content-Type", "application/octet-stream");
2015 [ + - ]: 574 : add_mhd_response_header (r, "X-DEBUGINFOD-SIZE",
2016 : 574 : to_string(fs.st_size).c_str());
2017 : 574 : add_mhd_response_header (r, "X-DEBUGINFOD-ARCHIVE", b_source0.c_str());
2018 : 574 : add_mhd_response_header (r, "X-DEBUGINFOD-FILE", b_source1.c_str());
2019 : 574 : add_mhd_last_modified (r, fs.st_mtime);
2020 [ + - ]: 574 : if (verbose > 1)
2021 [ + - ]: 1722 : obatched(clog) << "serving fdcache archive " << b_source0
2022 : : << " file " << b_source1
2023 [ + - + - : 574 : << " section=" << section << endl;
+ - + - +
- ]
2024 : : /* libmicrohttpd will close it. */
2025 [ + - ]: 574 : if (result_fd)
2026 : 574 : *result_fd = fd;
2027 : : return r;
2028 : : // NB: see, we never go around the 'loop' more than once
2029 : : }
2030 : :
2031 : : // no match ... grumble, must process the archive
2032 : 152 : string archive_decoder = "/dev/null";
2033 [ + - - - ]: 152 : string archive_extension = "";
2034 [ + + ]: 342 : for (auto&& arch : scan_archives)
2035 [ + + ]: 190 : if (string_endswith(b_source0, arch.first))
2036 : : {
2037 [ + - ]: 152 : archive_extension = arch.first;
2038 [ + - ]: 342 : archive_decoder = arch.second;
2039 : : }
2040 : 152 : FILE* fp;
2041 : :
2042 : 152 : defer_dtor<FILE*,int>::dtor_fn dfn;
2043 [ + + ]: 152 : if (archive_decoder != "cat")
2044 : : {
2045 [ + - + - : 48 : string popen_cmd = archive_decoder + " " + shell_escape(b_source0);
+ - - + -
- - - ]
2046 [ + - ]: 24 : fp = popen (popen_cmd.c_str(), "r"); // "e" O_CLOEXEC?
2047 : 24 : dfn = pclose;
2048 [ - + ]: 24 : if (fp == NULL)
2049 [ # # # # : 0 : throw libc_exception (errno, string("popen ") + popen_cmd);
# # # # ]
2050 : 24 : }
2051 : : else
2052 : : {
2053 [ + - ]: 128 : fp = fopen (b_source0.c_str(), "r");
2054 : 128 : dfn = fclose;
2055 [ - + ]: 128 : if (fp == NULL)
2056 [ # # # # : 0 : throw libc_exception (errno, string("fopen ") + b_source0);
# # # # ]
2057 : : }
2058 : 152 : defer_dtor<FILE*,int> fp_closer (fp, dfn);
2059 : :
2060 : 152 : struct archive *a;
2061 [ + - ]: 152 : a = archive_read_new();
2062 [ - + ]: 152 : if (a == NULL)
2063 [ # # # # ]: 0 : throw archive_exception("cannot create archive reader");
2064 : 152 : defer_dtor<struct archive*,int> archive_closer (a, archive_read_free);
2065 : :
2066 [ + - ]: 152 : rc = archive_read_support_format_all(a);
2067 [ - + ]: 152 : if (rc != ARCHIVE_OK)
2068 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all format");
2069 [ + - ]: 152 : rc = archive_read_support_filter_all(a);
2070 [ - + ]: 152 : if (rc != ARCHIVE_OK)
2071 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all filters");
2072 : :
2073 [ + - ]: 152 : rc = archive_read_open_FILE (a, fp);
2074 [ - + ]: 152 : if (rc != ARCHIVE_OK)
2075 : : {
2076 [ # # # # : 0 : obatched(clog) << "cannot open archive from pipe " << b_source0 << endl;
# # ]
2077 [ # # # # ]: 0 : throw archive_exception(a, "cannot open archive from pipe");
2078 : : }
2079 : :
2080 : : // archive traversal is in three stages, no, four stages:
2081 : : // 1) skip entries whose names do not match the requested one
2082 : : // 2) extract the matching entry name (set r = result)
2083 : : // 3) extract some number of prefetched entries (just into fdcache)
2084 : : // 4) abort any further processing
2085 : 152 : struct MHD_Response* r = 0; // will set in stage 2
2086 [ + + ]: 152 : unsigned prefetch_count =
2087 : : internal_req_p ? 0 : fdcache_prefetch; // will decrement in stage 3
2088 : :
2089 [ + + ]: 1536 : while(r == 0 || prefetch_count > 0) // stage 1, 2, or 3
2090 : : {
2091 [ + - ]: 1516 : if (interrupted)
2092 : : break;
2093 : :
2094 : 1516 : struct archive_entry *e;
2095 [ + - ]: 1516 : rc = archive_read_next_header (a, &e);
2096 [ + + ]: 1516 : if (rc != ARCHIVE_OK)
2097 : : break;
2098 : :
2099 [ + - + + ]: 1384 : if (! S_ISREG(archive_entry_mode (e))) // skip non-files completely
2100 : 1084 : continue;
2101 : :
2102 [ + - ]: 300 : string fn = canonicalized_archive_entry_pathname (e);
2103 [ + + + + ]: 300 : if ((r == 0) && (fn != b_source1)) // stage 1
2104 : 76 : continue;
2105 : :
2106 [ + - + + ]: 224 : if (fdcache.probe (b_source0, fn) && // skip if already interned
2107 [ - + ]: 24 : fn != b_source1) // but only if we'd just be prefetching, PR29474
2108 : 24 : continue;
2109 : :
2110 : : // extract this file to a temporary file
2111 : 200 : char* tmppath = NULL;
2112 : 200 : rc = asprintf (&tmppath, "%s/debuginfod-fdcache.XXXXXX", tmpdir.c_str());
2113 [ - + ]: 200 : if (rc < 0)
2114 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
2115 : 200 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
2116 [ + - ]: 200 : fd = mkstemp (tmppath);
2117 [ - + ]: 200 : if (fd < 0)
2118 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
2119 : : // NB: don't unlink (tmppath), as fdcache will take charge of it.
2120 : :
2121 : : // NB: this can take many uninterruptible seconds for a huge file
2122 [ + - ]: 200 : rc = archive_read_data_into_fd (a, fd);
2123 [ - + ]: 200 : if (rc != ARCHIVE_OK) // e.g. ENOSPC!
2124 : : {
2125 [ # # ]: 0 : close (fd);
2126 : 0 : unlink (tmppath);
2127 [ # # # # ]: 0 : throw archive_exception(a, "cannot extract file");
2128 : : }
2129 : :
2130 : : // Set the mtime so the fdcache file mtimes, even prefetched ones,
2131 : : // propagate to future webapi clients.
2132 : 200 : struct timespec tvs[2];
2133 : 200 : tvs[0].tv_sec = 0;
2134 : 200 : tvs[0].tv_nsec = UTIME_OMIT;
2135 [ + - ]: 200 : tvs[1].tv_sec = archive_entry_mtime(e);
2136 [ + - ]: 200 : tvs[1].tv_nsec = archive_entry_mtime_nsec(e);
2137 : 200 : (void) futimens (fd, tvs); /* best effort */
2138 : :
2139 : 200 : struct timespec extract_end;
2140 : 200 : clock_gettime (CLOCK_MONOTONIC, &extract_end);
2141 : 200 : double extract_time = (extract_end.tv_sec - extract_begin.tv_sec)
2142 : 200 : + (extract_end.tv_nsec - extract_begin.tv_nsec)/1.e9;
2143 : :
2144 [ + + ]: 200 : if (r != 0) // stage 3
2145 : : {
2146 : : // NB: now we know we have a complete reusable file; make fdcache
2147 : : // responsible for unlinking it later.
2148 [ + - + - : 48 : fdcache.intern(b_source0, fn,
+ - ]
2149 : : tmppath, archive_entry_size(e),
2150 : : false, extract_time); // prefetched ones go to the prefetch cache
2151 : 48 : prefetch_count --;
2152 [ + - ]: 48 : close (fd); // we're not saving this fd to make a mhd-response from!
2153 : 48 : continue;
2154 : : }
2155 : :
2156 : : // NB: now we know we have a complete reusable file; make fdcache
2157 : : // responsible for unlinking it later.
2158 [ + - + - : 152 : fdcache.intern(b_source0, b_source1,
+ - ]
2159 : : tmppath, archive_entry_size(e),
2160 : : true, extract_time); // requested ones go to the front of the line
2161 : :
2162 [ + + ]: 152 : if (!section.empty ())
2163 : : {
2164 [ + - + - ]: 4 : int scn_fd = extract_section (fd, b_mtime,
2165 [ + - + - : 8 : b_source0 + ":" + b_source1,
- + ]
2166 : : section, extract_begin);
2167 [ + - ]: 4 : close (fd);
2168 [ + - ]: 4 : if (scn_fd >= 0)
2169 : 4 : fd = scn_fd;
2170 : : else
2171 : : {
2172 [ # # ]: 0 : if (verbose)
2173 [ # # # # ]: 0 : obatched (clog) << "cannot find section " << section
2174 : : << " for archive " << b_source0
2175 [ # # # # : 0 : << " file " << b_source1 << endl;
# # # # #
# ]
2176 : 0 : return 0;
2177 : : }
2178 : :
2179 : 4 : rc = fstat(fd, &fs);
2180 [ - + ]: 4 : if (rc < 0)
2181 : : {
2182 [ # # ]: 0 : close (fd);
2183 [ # # # # ]: 0 : throw libc_exception (errno,
2184 [ # # # # : 0 : string ("fstat ") + b_source0 + string (" ") + section);
# # # # #
# # # # #
# # # # #
# ]
2185 : : }
2186 [ + - ]: 4 : r = MHD_create_response_from_fd (fs.st_size, fd);
2187 : : }
2188 : : else
2189 [ + - + - ]: 148 : r = MHD_create_response_from_fd (archive_entry_size(e), fd);
2190 : :
2191 [ + - + - : 304 : inc_metric ("http_responses_total","result",archive_extension + " archive");
+ - + - -
+ + + - -
- - ]
2192 [ - + ]: 152 : if (r == 0)
2193 : : {
2194 [ # # ]: 0 : if (verbose)
2195 [ # # # # : 0 : obatched(clog) << "cannot create fd-response for " << b_source0 << endl;
# # ]
2196 [ # # ]: 0 : close(fd);
2197 : 0 : break; // assume no chance of better luck around another iteration; no other copies of same file
2198 : : }
2199 : : else
2200 : : {
2201 [ + - ]: 152 : add_mhd_response_header (r, "Content-Type",
2202 : : "application/octet-stream");
2203 [ + - ]: 152 : add_mhd_response_header (r, "X-DEBUGINFOD-SIZE",
2204 [ + - + - ]: 152 : to_string(archive_entry_size(e)).c_str());
2205 [ + - ]: 152 : add_mhd_response_header (r, "X-DEBUGINFOD-ARCHIVE", b_source0.c_str());
2206 [ + - ]: 152 : add_mhd_response_header (r, "X-DEBUGINFOD-FILE", b_source1.c_str());
2207 [ + - + - ]: 152 : add_mhd_last_modified (r, archive_entry_mtime(e));
2208 [ + - ]: 152 : if (verbose > 1)
2209 [ + - + - ]: 456 : obatched(clog) << "serving archive " << b_source0
2210 : : << " file " << b_source1
2211 [ + - + - : 152 : << " section=" << section << endl;
+ - + - +
- ]
2212 : : /* libmicrohttpd will close it. */
2213 [ + - ]: 152 : if (result_fd)
2214 : 152 : *result_fd = fd;
2215 : 152 : continue;
2216 : : }
2217 [ - - - - : 1816 : }
+ + ]
2218 : :
2219 : : // XXX: rpm/file not found: delete this R entry?
2220 : : return r;
2221 [ - + + + ]: 880 : }
2222 : :
2223 : :
2224 : : static struct MHD_Response*
2225 : 1369 : handle_buildid_match (bool internal_req_p,
2226 : : int64_t b_mtime,
2227 : : const string& b_stype,
2228 : : const string& b_source0,
2229 : : const string& b_source1,
2230 : : const string& section,
2231 : : int *result_fd)
2232 : : {
2233 : 1369 : try
2234 : : {
2235 [ + + ]: 1369 : if (b_stype == "F")
2236 [ + - ]: 583 : return handle_buildid_f_match(internal_req_p, b_mtime, b_source0,
2237 : : section, result_fd);
2238 [ + - ]: 786 : else if (b_stype == "R")
2239 [ + + ]: 786 : return handle_buildid_r_match(internal_req_p, b_mtime, b_source0,
2240 : : b_source1, section, result_fd);
2241 : : }
2242 [ - + ]: 58 : catch (const reportable_exception &e)
2243 : : {
2244 [ + - ]: 58 : e.report(clog);
2245 : : // Report but swallow libc etc. errors here; let the caller
2246 : : // iterate to other matches of the content.
2247 : 58 : }
2248 : :
2249 : : return 0;
2250 : : }
2251 : :
2252 : :
2253 : : static int
2254 : 3236 : debuginfod_find_progress (debuginfod_client *, long a, long b)
2255 : : {
2256 [ - + ]: 3236 : if (verbose > 4)
2257 [ # # # # : 0 : obatched(clog) << "federated debuginfod progress=" << a << "/" << b << endl;
# # # # ]
2258 : :
2259 : 3236 : return interrupted;
2260 : : }
2261 : :
2262 : :
2263 : : // a little lru pool of debuginfod_client*s for reuse between query threads
2264 : :
2265 : : mutex dc_pool_lock;
2266 : : deque<debuginfod_client*> dc_pool;
2267 : :
2268 : 620 : debuginfod_client* debuginfod_pool_begin()
2269 : : {
2270 : 620 : unique_lock<mutex> lock(dc_pool_lock);
2271 [ + + ]: 620 : if (dc_pool.size() > 0)
2272 : : {
2273 [ + - + - : 1188 : inc_metric("dc_pool_op_count","op","begin-reuse");
+ - + - -
+ - + - -
- - ]
2274 : 594 : debuginfod_client *c = dc_pool.front();
2275 : 594 : dc_pool.pop_front();
2276 : 594 : return c;
2277 : : }
2278 [ + - + - : 52 : inc_metric("dc_pool_op_count","op","begin-new");
+ - + - -
+ - + - -
- - ]
2279 [ + - ]: 26 : return debuginfod_begin();
2280 : 620 : }
2281 : :
2282 : :
2283 : 146 : void debuginfod_pool_groom()
2284 : : {
2285 : 146 : unique_lock<mutex> lock(dc_pool_lock);
2286 [ + + ]: 172 : while (dc_pool.size() > 0)
2287 : : {
2288 [ + - + - : 52 : inc_metric("dc_pool_op_count","op","end");
+ - + - -
+ - + - -
- - ]
2289 [ + - ]: 26 : debuginfod_end(dc_pool.front());
2290 : 26 : dc_pool.pop_front();
2291 : : }
2292 : 146 : }
2293 : :
2294 : :
2295 : 620 : void debuginfod_pool_end(debuginfod_client* c)
2296 : : {
2297 : 620 : unique_lock<mutex> lock(dc_pool_lock);
2298 [ + - + - : 1240 : inc_metric("dc_pool_op_count","op","end-save");
+ - + - -
+ - + - -
- - ]
2299 [ + - ]: 620 : dc_pool.push_front(c); // accelerate reuse, vs. push_back
2300 : 620 : }
2301 : :
2302 : :
2303 : : static struct MHD_Response*
2304 : 1935 : handle_buildid (MHD_Connection* conn,
2305 : : const string& buildid /* unsafe */,
2306 : : string& artifacttype /* unsafe, cleanse on exception/return */,
2307 : : const string& suffix /* unsafe */,
2308 : : int *result_fd)
2309 : : {
2310 : : // validate artifacttype
2311 : 1935 : string atype_code;
2312 [ + + + - ]: 1935 : if (artifacttype == "debuginfo") atype_code = "D";
2313 [ + + + - ]: 1197 : else if (artifacttype == "executable") atype_code = "E";
2314 [ + + + - ]: 577 : else if (artifacttype == "source") atype_code = "S";
2315 [ + + + - ]: 12 : else if (artifacttype == "section") atype_code = "I";
2316 : : else {
2317 [ + - ]: 4 : artifacttype = "invalid"; // PR28242 ensure http_resposes metrics don't propagate unclean user data
2318 [ + - + - ]: 12 : throw reportable_exception("invalid artifacttype");
2319 : : }
2320 : :
2321 [ + + ]: 1931 : if (conn != 0)
2322 [ + - + - : 4406 : inc_metric("http_requests_total", "type", artifacttype);
+ - - + -
- - + ]
2323 : :
2324 : 1931 : string section;
2325 [ + + ]: 1931 : if (atype_code == "I")
2326 : : {
2327 [ - + ]: 8 : if (suffix.size () < 2)
2328 [ # # # # ]: 0 : throw reportable_exception ("invalid section suffix");
2329 : :
2330 : : // Remove leading '/'
2331 [ + - - + ]: 8 : section = suffix.substr(1);
2332 : : }
2333 : :
2334 [ + + - + ]: 2496 : if (atype_code == "S" && suffix == "")
2335 [ # # # # ]: 0 : throw reportable_exception("invalid source suffix");
2336 : :
2337 : : // validate buildid
2338 [ + + ]: 1931 : if ((buildid.size() < 2) || // not empty
2339 [ + + + - : 3858 : (buildid.size() % 2) || // even number
+ - ]
2340 : 1927 : (buildid.find_first_not_of("0123456789abcdef") != string::npos)) // pure tasty lowercase hex
2341 [ + - - + ]: 8 : throw reportable_exception("invalid buildid");
2342 : :
2343 [ + - ]: 1927 : if (verbose > 1)
2344 [ + - + - ]: 5781 : obatched(clog) << "searching for buildid=" << buildid << " artifacttype=" << artifacttype
2345 [ + - + - : 1927 : << " suffix=" << suffix << endl;
+ - + - +
- ]
2346 : :
2347 : : // If invoked from the scanner threads, use the scanners' read-write
2348 : : // connection. Otherwise use the web query threads' read-only connection.
2349 [ + + ]: 1927 : sqlite3 *thisdb = (conn == 0) ? db : dbq;
2350 : :
2351 : 1927 : sqlite_ps *pp = 0;
2352 : :
2353 [ + + ]: 1927 : if (atype_code == "D")
2354 : : {
2355 [ - + ]: 738 : pp = new sqlite_ps (thisdb, "mhd-query-d",
2356 : : "select mtime, sourcetype, source0, source1 from " BUILDIDS "_query_d where buildid = ? "
2357 [ + - + - : 1476 : "order by mtime desc");
+ - + - +
- - - ]
2358 [ + - ]: 738 : pp->reset();
2359 [ + - ]: 738 : pp->bind(1, buildid);
2360 : : }
2361 [ + + ]: 1189 : else if (atype_code == "E")
2362 : : {
2363 [ - + ]: 616 : pp = new sqlite_ps (thisdb, "mhd-query-e",
2364 : : "select mtime, sourcetype, source0, source1 from " BUILDIDS "_query_e where buildid = ? "
2365 [ + - + - : 1232 : "order by mtime desc");
+ - + - +
- - - ]
2366 [ + - ]: 616 : pp->reset();
2367 [ + - ]: 616 : pp->bind(1, buildid);
2368 : : }
2369 [ + + ]: 573 : else if (atype_code == "S")
2370 : : {
2371 : : // PR25548
2372 : : // Incoming source queries may come in with either dwarf-level OR canonicalized paths.
2373 : : // We let the query pass with either one.
2374 : :
2375 [ - + ]: 565 : pp = new sqlite_ps (thisdb, "mhd-query-s",
2376 : : "select mtime, sourcetype, source0, source1 from " BUILDIDS "_query_s where buildid = ? and artifactsrc in (?,?) "
2377 [ + - + - : 1130 : "order by sharedprefix(source0,source0ref) desc, mtime desc");
+ - + - +
- - - ]
2378 [ + - ]: 565 : pp->reset();
2379 [ + - ]: 565 : pp->bind(1, buildid);
2380 : : // NB: we don't store the non-canonicalized path names any more, but old databases
2381 : : // might have them (and no canon ones), so we keep searching for both.
2382 [ + - ]: 565 : pp->bind(2, suffix);
2383 [ + - + - : 1750 : pp->bind(3, canon_pathname(suffix));
- + ]
2384 : : }
2385 [ + - ]: 8 : else if (atype_code == "I")
2386 : : {
2387 [ - + ]: 8 : pp = new sqlite_ps (thisdb, "mhd-query-i",
2388 : : "select mtime, sourcetype, source0, source1, 1 as debug_p from " BUILDIDS "_query_d where buildid = ? "
2389 : : "union all "
2390 : : "select mtime, sourcetype, source0, source1, 0 as debug_p from " BUILDIDS "_query_e where buildid = ? "
2391 [ + - + - : 16 : "order by debug_p desc, mtime desc");
+ - + - +
- - - ]
2392 [ + - ]: 8 : pp->reset();
2393 [ + - ]: 8 : pp->bind(1, buildid);
2394 [ + - ]: 8 : pp->bind(2, buildid);
2395 : : }
2396 : 1927 : unique_ptr<sqlite_ps> ps_closer(pp); // release pp if exception or return
2397 : :
2398 : 1927 : bool do_upstream_section_query = true;
2399 : :
2400 : : // consume all the rows
2401 : 1989 : while (1)
2402 : : {
2403 [ + - ]: 1989 : int rc = pp->step();
2404 [ + + ]: 1989 : if (rc == SQLITE_DONE) break;
2405 [ - + ]: 1369 : if (rc != SQLITE_ROW)
2406 [ # # # # ]: 0 : throw sqlite_exception(rc, "step");
2407 : :
2408 [ + - ]: 1369 : int64_t b_mtime = sqlite3_column_int64 (*pp, 0);
2409 [ + - - + : 1369 : string b_stype = string((const char*) sqlite3_column_text (*pp, 1) ?: ""); /* by DDL may not be NULL */
+ - ]
2410 [ + - - + : 1369 : string b_source0 = string((const char*) sqlite3_column_text (*pp, 2) ?: ""); /* may be NULL */
+ - - - ]
2411 [ + - + + : 1952 : string b_source1 = string((const char*) sqlite3_column_text (*pp, 3) ?: ""); /* may be NULL */
+ - - - ]
2412 : :
2413 [ + - ]: 1369 : if (verbose > 1)
2414 [ + - + - : 4107 : obatched(clog) << "found mtime=" << b_mtime << " stype=" << b_stype
- - ]
2415 [ + - + - : 1369 : << " source0=" << b_source0 << " source1=" << b_source1 << endl;
+ - + - +
- + - +
- ]
2416 : :
2417 : : // Try accessing the located match.
2418 : : // XXX: in case of multiple matches, attempt them in parallel?
2419 [ + - ]: 1369 : auto r = handle_buildid_match (conn ? false : true,
2420 : : b_mtime, b_stype, b_source0, b_source1,
2421 : : section, result_fd);
2422 [ + + ]: 1369 : if (r)
2423 [ + + ]: 1307 : return r;
2424 : :
2425 : : // If a debuginfo file matching BUILDID was found but didn't contain
2426 : : // the desired section, then the section should not exist. Don't
2427 : : // bother querying upstream servers.
2428 [ + + + - : 62 : if (!section.empty () && (sqlite3_column_int (*pp, 4) == 1))
- + ]
2429 : : {
2430 : 4 : struct stat st;
2431 : :
2432 : : // For "F" sourcetype, check if the debuginfo exists. For "R"
2433 : : // sourcetype, check if the debuginfo was interned into the fdcache.
2434 [ - + ]: 6 : if ((b_stype == "F" && (stat (b_source0.c_str (), &st) == 0))
2435 [ + + + - : 6 : || (b_stype == "R" && fdcache.probe (b_source0, b_source1)))
+ - - + ]
2436 : : do_upstream_section_query = false;
2437 : : }
2438 [ + - - + : 2738 : }
+ - - + ]
2439 [ + - ]: 620 : pp->reset();
2440 : :
2441 [ - + ]: 620 : if (!do_upstream_section_query)
2442 [ # # # # ]: 0 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found");
2443 : :
2444 : : // We couldn't find it in the database. Last ditch effort
2445 : : // is to defer to other debuginfo servers.
2446 : :
2447 : 620 : int fd = -1;
2448 [ + - ]: 620 : debuginfod_client *client = debuginfod_pool_begin ();
2449 [ - + ]: 620 : if (client == NULL)
2450 [ # # # # ]: 0 : throw libc_exception(errno, "debuginfod client pool alloc");
2451 : 620 : defer_dtor<debuginfod_client*,void> client_closer (client, debuginfod_pool_end);
2452 : :
2453 [ + - ]: 620 : debuginfod_set_progressfn (client, & debuginfod_find_progress);
2454 : :
2455 [ + - ]: 620 : if (conn)
2456 : : {
2457 : : // Transcribe incoming User-Agent:
2458 [ + - - + : 620 : string ua = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "User-Agent") ?: "";
+ - ]
2459 [ + - + - : 624 : string ua_complete = string("User-Agent: ") + ua;
+ - ]
2460 [ + - ]: 620 : debuginfod_add_http_header (client, ua_complete.c_str());
2461 : :
2462 : : // Compute larger XFF:, for avoiding info loss during
2463 : : // federation, and for future cyclicity detection.
2464 [ + - + + : 1226 : string xff = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "X-Forwarded-For") ?: "";
+ - + - ]
2465 [ + + ]: 620 : if (xff != "")
2466 [ + - - + ]: 36 : xff += string(", "); // comma separated list
2467 : :
2468 : 620 : unsigned int xff_count = 0;
2469 [ + + ]: 860 : for (auto&& i : xff){
2470 [ + + ]: 240 : if (i == ',') xff_count++;
2471 : : }
2472 : :
2473 : : // if X-Forwarded-For: exceeds N hops,
2474 : : // do not delegate a local lookup miss to upstream debuginfods.
2475 [ + + ]: 620 : if (xff_count >= forwarded_ttl_limit)
2476 [ + - + - ]: 8 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found, --forwared-ttl-limit reached \
2477 : 8 : and will not query the upstream servers");
2478 : :
2479 : : // Compute the client's numeric IP address only - so can't merge with conninfo()
2480 [ + - ]: 616 : const union MHD_ConnectionInfo *u = MHD_get_connection_info (conn,
2481 : : MHD_CONNECTION_INFO_CLIENT_ADDRESS);
2482 [ + - ]: 616 : struct sockaddr *so = u ? u->client_addr : 0;
2483 : 616 : char hostname[256] = ""; // RFC1035
2484 [ + - - + ]: 616 : if (so && so->sa_family == AF_INET) {
2485 [ # # ]: 0 : (void) getnameinfo (so, sizeof (struct sockaddr_in), hostname, sizeof (hostname), NULL, 0,
2486 : : NI_NUMERICHOST);
2487 [ + - ]: 616 : } else if (so && so->sa_family == AF_INET6) {
2488 : 616 : struct sockaddr_in6* addr6 = (struct sockaddr_in6*) so;
2489 [ + - + - : 616 : if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
- + ]
2490 : 616 : struct sockaddr_in addr4;
2491 [ + - ]: 616 : memset (&addr4, 0, sizeof(addr4));
2492 : 616 : addr4.sin_family = AF_INET;
2493 : 616 : addr4.sin_port = addr6->sin6_port;
2494 [ + - ]: 616 : memcpy (&addr4.sin_addr.s_addr, addr6->sin6_addr.s6_addr+12, sizeof(addr4.sin_addr.s_addr));
2495 [ + - ]: 616 : (void) getnameinfo ((struct sockaddr*) &addr4, sizeof (addr4),
2496 : : hostname, sizeof (hostname), NULL, 0,
2497 : : NI_NUMERICHOST);
2498 : : } else {
2499 [ # # ]: 0 : (void) getnameinfo (so, sizeof (struct sockaddr_in6), hostname, sizeof (hostname), NULL, 0,
2500 : : NI_NUMERICHOST);
2501 : : }
2502 : : }
2503 : :
2504 [ + - + - : 1236 : string xff_complete = string("X-Forwarded-For: ")+xff+string(hostname);
+ - + - -
+ - + - -
- + ]
2505 [ + - ]: 616 : debuginfod_add_http_header (client, xff_complete.c_str());
2506 [ + + + - : 1300 : }
+ + ]
2507 : :
2508 [ + + ]: 616 : if (artifacttype == "debuginfo")
2509 [ + - ]: 68 : fd = debuginfod_find_debuginfo (client,
2510 [ + - ]: 68 : (const unsigned char*) buildid.c_str(),
2511 : : 0, NULL);
2512 [ + + ]: 548 : else if (artifacttype == "executable")
2513 [ + - ]: 546 : fd = debuginfod_find_executable (client,
2514 [ + - ]: 546 : (const unsigned char*) buildid.c_str(),
2515 : : 0, NULL);
2516 [ + - ]: 2 : else if (artifacttype == "source")
2517 [ + - ]: 2 : fd = debuginfod_find_source (client,
2518 [ + - ]: 2 : (const unsigned char*) buildid.c_str(),
2519 : : 0, suffix.c_str(), NULL);
2520 [ # # ]: 0 : else if (artifacttype == "section")
2521 [ # # ]: 0 : fd = debuginfod_find_section (client,
2522 [ # # ]: 0 : (const unsigned char*) buildid.c_str(),
2523 : : 0, section.c_str(), NULL);
2524 : :
2525 [ + + ]: 616 : if (fd >= 0)
2526 : : {
2527 [ + - ]: 4 : if (conn != 0)
2528 [ + - + - : 624 : inc_metric ("http_responses_total","result","upstream");
+ - + - -
+ - + - -
- - ]
2529 : 4 : struct stat s;
2530 : 4 : int rc = fstat (fd, &s);
2531 [ + - ]: 4 : if (rc == 0)
2532 : : {
2533 [ + - ]: 4 : auto r = MHD_create_response_from_fd ((uint64_t) s.st_size, fd);
2534 [ + - ]: 4 : if (r)
2535 : : {
2536 [ + - ]: 4 : add_mhd_response_header (r, "Content-Type",
2537 : : "application/octet-stream");
2538 : : // Copy the incoming headers
2539 [ + - ]: 4 : const char * hdrs = debuginfod_get_headers(client);
2540 [ + - ]: 4 : string header_dup;
2541 [ + - ]: 4 : if (hdrs)
2542 [ + - - + ]: 4 : header_dup = string(hdrs);
2543 : : // Parse the "header: value\n" lines into (h,v) tuples and pass on
2544 : 20 : while(1)
2545 : : {
2546 : 12 : size_t newline = header_dup.find('\n');
2547 [ + + ]: 12 : if (newline == string::npos) break;
2548 : 8 : size_t colon = header_dup.find(':');
2549 [ + - ]: 8 : if (colon == string::npos) break;
2550 [ + - ]: 8 : string header = header_dup.substr(0,colon);
2551 [ + - ]: 8 : string value = header_dup.substr(colon+1,newline-colon-1);
2552 : : // strip leading spaces from value
2553 : 8 : size_t nonspace = value.find_first_not_of(" ");
2554 [ + - ]: 8 : if (nonspace != string::npos)
2555 [ + - + + ]: 12 : value = value.substr(nonspace);
2556 [ + - ]: 8 : add_mhd_response_header(r, header.c_str(), value.c_str());
2557 [ + - + + : 12 : header_dup = header_dup.substr(newline+1);
+ + - - ]
2558 [ + - ]: 16 : }
2559 : :
2560 [ + - ]: 4 : add_mhd_last_modified (r, s.st_mtime);
2561 [ + - ]: 4 : if (verbose > 1)
2562 [ + - + - : 8 : obatched(clog) << "serving file from upstream debuginfod/cache" << endl;
- - ]
2563 [ + - ]: 4 : if (result_fd)
2564 : 4 : *result_fd = fd;
2565 [ + - ]: 4 : return r; // NB: don't close fd; libmicrohttpd will
2566 : 4 : }
2567 : : }
2568 [ # # ]: 0 : close (fd);
2569 : : }
2570 : : else
2571 [ + + ]: 612 : switch(fd)
2572 : : {
2573 : : case -ENOSYS:
2574 : : break;
2575 : : case -ENOENT:
2576 : : break;
2577 : 532 : default: // some more tricky error
2578 [ + - + - ]: 1064 : throw libc_exception(-fd, "upstream debuginfod query failed");
2579 : : }
2580 : :
2581 [ + - - + ]: 160 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found");
2582 [ - + - + ]: 2547 : }
2583 : :
2584 : :
2585 : : ////////////////////////////////////////////////////////////////////////
2586 : :
2587 : : static map<string,double> metrics; // arbitrary data for /metrics query
2588 : : // NB: store int64_t since all our metrics are integers; prometheus accepts double
2589 : : static mutex metrics_lock;
2590 : : // NB: these objects get released during the process exit via global dtors
2591 : : // do not call them from within other global dtors
2592 : :
2593 : : // utility function for assembling prometheus-compatible
2594 : : // name="escaped-value" strings
2595 : : // https://prometheus.io/docs/instrumenting/exposition_formats/
2596 : : static string
2597 : 724531 : metric_label(const string& name, const string& value)
2598 : : {
2599 : 724531 : string x = name + "=\"";
2600 [ + + ]: 12616420 : for (auto&& c : value)
2601 [ - - - + ]: 11892216 : switch(c)
2602 : : {
2603 [ # # ]: 0 : case '\\': x += "\\\\"; break;
2604 [ # # ]: 0 : case '\"': x += "\\\""; break;
2605 [ # # ]: 0 : case '\n': x += "\\n"; break;
2606 [ + - ]: 23784258 : default: x += c; break;
2607 : : }
2608 [ + - ]: 724204 : x += "\"";
2609 : 724286 : return x;
2610 : 0 : }
2611 : :
2612 : :
2613 : : // add prometheus-format metric name + label tuple (if any) + value
2614 : :
2615 : : static void
2616 : 1276 : set_metric(const string& metric, double value)
2617 : : {
2618 : 1276 : unique_lock<mutex> lock(metrics_lock);
2619 [ + - ]: 1276 : metrics[metric] = value;
2620 : 1276 : }
2621 : : static void
2622 : 592 : inc_metric(const string& metric)
2623 : : {
2624 : 592 : unique_lock<mutex> lock(metrics_lock);
2625 [ + - ]: 592 : metrics[metric] ++;
2626 : 592 : }
2627 : : static void
2628 : 6407 : set_metric(const string& metric,
2629 : : const string& lname, const string& lvalue,
2630 : : double value)
2631 : : {
2632 [ + - + - : 12814 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + - + +
+ - - ]
2633 [ + - ]: 6407 : unique_lock<mutex> lock(metrics_lock);
2634 [ + - ]: 6408 : metrics[key] = value;
2635 [ + - ]: 12816 : }
2636 : :
2637 : : static void
2638 : 349480 : inc_metric(const string& metric,
2639 : : const string& lname, const string& lvalue)
2640 : : {
2641 [ + - + - : 760823 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + + + +
+ - - ]
2642 [ + - ]: 349477 : unique_lock<mutex> lock(metrics_lock);
2643 [ + - ]: 349483 : metrics[key] ++;
2644 [ + - ]: 698946 : }
2645 : : static void
2646 : 337323 : add_metric(const string& metric,
2647 : : const string& lname, const string& lvalue,
2648 : : double value)
2649 : : {
2650 [ + - + - : 736527 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + + + +
+ - - ]
2651 [ + - ]: 337334 : unique_lock<mutex> lock(metrics_lock);
2652 [ + - ]: 337356 : metrics[key] += value;
2653 [ + - ]: 674708 : }
2654 : : static void
2655 : 592 : add_metric(const string& metric,
2656 : : double value)
2657 : : {
2658 : 592 : unique_lock<mutex> lock(metrics_lock);
2659 [ + - ]: 592 : metrics[metric] += value;
2660 : 592 : }
2661 : :
2662 : :
2663 : : // and more for higher arity labels if needed
2664 : :
2665 : : static void
2666 : 7842 : inc_metric(const string& metric,
2667 : : const string& lname, const string& lvalue,
2668 : : const string& rname, const string& rvalue)
2669 : : {
2670 [ + - - + : 15684 : string key = (metric + "{"
- - ]
2671 [ + - + - : 31368 : + metric_label(lname, lvalue) + ","
- + - + +
+ - - ]
2672 [ + - - + : 23526 : + metric_label(rname, rvalue) + "}");
- + ]
2673 [ + - ]: 7842 : unique_lock<mutex> lock(metrics_lock);
2674 [ + - ]: 7842 : metrics[key] ++;
2675 [ + - ]: 15684 : }
2676 : : static void
2677 : 7842 : add_metric(const string& metric,
2678 : : const string& lname, const string& lvalue,
2679 : : const string& rname, const string& rvalue,
2680 : : double value)
2681 : : {
2682 [ + - - + : 15684 : string key = (metric + "{"
- - ]
2683 [ + - + - : 31368 : + metric_label(lname, lvalue) + ","
- + - + +
+ - - ]
2684 [ + - - + : 23526 : + metric_label(rname, rvalue) + "}");
- + ]
2685 [ + - ]: 7842 : unique_lock<mutex> lock(metrics_lock);
2686 [ + - ]: 7842 : metrics[key] += value;
2687 [ + - ]: 15684 : }
2688 : :
2689 : : static struct MHD_Response*
2690 : 713 : handle_metrics (off_t* size)
2691 : : {
2692 : 713 : stringstream o;
2693 : 713 : {
2694 [ + - ]: 713 : unique_lock<mutex> lock(metrics_lock);
2695 [ + + ]: 67949 : for (auto&& i : metrics)
2696 [ + - ]: 67236 : o << i.first
2697 : : << " "
2698 [ + - + - ]: 67236 : << std::setprecision(std::numeric_limits<double>::digits10 + 1)
2699 [ + - + - ]: 67236 : << i.second
2700 : 67236 : << endl;
2701 : 0 : }
2702 [ + - ]: 713 : const string& os = o.str();
2703 [ + - ]: 713 : MHD_Response* r = MHD_create_response_from_buffer (os.size(),
2704 [ + - ]: 713 : (void*) os.c_str(),
2705 : : MHD_RESPMEM_MUST_COPY);
2706 [ + - ]: 713 : if (r != NULL)
2707 : : {
2708 [ + - ]: 713 : *size = os.size();
2709 [ + - ]: 713 : add_mhd_response_header (r, "Content-Type", "text/plain");
2710 : : }
2711 [ + - ]: 1426 : return r;
2712 : 713 : }
2713 : :
2714 : : static struct MHD_Response*
2715 : 0 : handle_root (off_t* size)
2716 : : {
2717 [ # # # # : 0 : static string version = "debuginfod (" + string (PACKAGE_NAME) + ") "
# # # # #
# # # ]
2718 [ # # # # : 0 : + string (PACKAGE_VERSION);
# # # # #
# ]
2719 : 0 : MHD_Response* r = MHD_create_response_from_buffer (version.size (),
2720 : 0 : (void *) version.c_str (),
2721 : : MHD_RESPMEM_PERSISTENT);
2722 [ # # ]: 0 : if (r != NULL)
2723 : : {
2724 : 0 : *size = version.size ();
2725 : 0 : add_mhd_response_header (r, "Content-Type", "text/plain");
2726 : : }
2727 : 0 : return r;
2728 : : }
2729 : :
2730 : :
2731 : : ////////////////////////////////////////////////////////////////////////
2732 : :
2733 : :
2734 : : /* libmicrohttpd callback */
2735 : : static MHD_RESULT
2736 : 5228 : handler_cb (void * /*cls*/,
2737 : : struct MHD_Connection *connection,
2738 : : const char *url,
2739 : : const char *method,
2740 : : const char * /*version*/,
2741 : : const char * /*upload_data*/,
2742 : : size_t * /*upload_data_size*/,
2743 : : void ** ptr)
2744 : : {
2745 : 5228 : struct MHD_Response *r = NULL;
2746 : 5228 : string url_copy = url;
2747 : :
2748 : : /* libmicrohttpd always makes (at least) two callbacks: once just
2749 : : past the headers, and one after the request body is finished
2750 : : being received. If we process things early (first callback) and
2751 : : queue a response, libmicrohttpd would suppress http keep-alive
2752 : : (via connection->read_closed = true). */
2753 : 5228 : static int aptr; /* just some random object to use as a flag */
2754 [ + + ]: 5228 : if (&aptr != *ptr)
2755 : : {
2756 : : /* do never respond on first call */
2757 : 2614 : *ptr = &aptr;
2758 : 2614 : return MHD_YES;
2759 : : }
2760 : 2614 : *ptr = NULL; /* reset when done */
2761 : :
2762 [ + - ]: 2614 : const char *maxsize_string = MHD_lookup_connection_value(connection, MHD_HEADER_KIND, "X-DEBUGINFOD-MAXSIZE");
2763 : 2614 : long maxsize = 0;
2764 [ + + + - ]: 2614 : if (maxsize_string != NULL && maxsize_string[0] != '\0')
2765 : 2 : maxsize = atol(maxsize_string);
2766 : : else
2767 : : maxsize = 0;
2768 : :
2769 : : #if MHD_VERSION >= 0x00097002
2770 : 2614 : enum MHD_Result rc;
2771 : : #else
2772 : : int rc = MHD_NO; // mhd
2773 : : #endif
2774 : 2614 : int http_code = 500;
2775 : 2614 : off_t http_size = -1;
2776 : 2614 : struct timespec ts_start, ts_end;
2777 : 2614 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
2778 : 2614 : double afteryou = 0.0;
2779 [ + - ]: 2614 : string artifacttype, suffix;
2780 : :
2781 : 2614 : try
2782 : : {
2783 [ + - - + : 5859 : if (string(method) != "GET")
- + ]
2784 [ # # # # ]: 0 : throw reportable_exception(400, "we support GET only");
2785 : :
2786 : : /* Start decoding the URL. */
2787 : 2613 : size_t slash1 = url_copy.find('/', 1);
2788 [ + - ]: 2614 : string url1 = url_copy.substr(0, slash1); // ok even if slash1 not found
2789 : :
2790 [ + + + - ]: 4509 : if (slash1 != string::npos && url1 == "/buildid")
2791 : : {
2792 : : // PR27863: block this thread awhile if another thread is already busy
2793 : : // fetching the exact same thing. This is better for Everyone.
2794 : : // The latecomer says "... after you!" and waits.
2795 [ + - + - : 4414 : add_metric ("thread_busy", "role", "http-buildid-after-you", 1);
+ - + - -
+ + - - -
- - ]
2796 : : #ifdef HAVE_PTHREAD_SETNAME_NP
2797 : 1895 : (void) pthread_setname_np (pthread_self(), "mhd-buildid-after-you");
2798 : : #endif
2799 : 1895 : struct timespec tsay_start, tsay_end;
2800 : 1895 : clock_gettime (CLOCK_MONOTONIC, &tsay_start);
2801 [ + + + - ]: 1895 : static unique_set<string> busy_urls;
2802 [ + - ]: 1895 : unique_set_reserver<string> after_you(busy_urls, url_copy);
2803 : 1895 : clock_gettime (CLOCK_MONOTONIC, &tsay_end);
2804 : 1895 : afteryou = (tsay_end.tv_sec - tsay_start.tv_sec) + (tsay_end.tv_nsec - tsay_start.tv_nsec)/1.e9;
2805 [ + - + - : 3790 : add_metric ("thread_busy", "role", "http-buildid-after-you", -1);
+ - + - -
+ + - - -
- - ]
2806 : :
2807 [ + - + - : 3790 : tmp_inc_metric m ("thread_busy", "role", "http-buildid");
+ - + - -
+ - + - -
- - ]
2808 : : #ifdef HAVE_PTHREAD_SETNAME_NP
2809 : 1895 : (void) pthread_setname_np (pthread_self(), "mhd-buildid");
2810 : : #endif
2811 : 1895 : size_t slash2 = url_copy.find('/', slash1+1);
2812 [ - + ]: 1895 : if (slash2 == string::npos)
2813 [ # # # # ]: 0 : throw reportable_exception("/buildid/ webapi error, need buildid");
2814 : :
2815 [ + - ]: 1895 : string buildid = url_copy.substr(slash1+1, slash2-slash1-1);
2816 : :
2817 : 1895 : size_t slash3 = url_copy.find('/', slash2+1);
2818 : :
2819 [ + + ]: 1895 : if (slash3 == string::npos)
2820 : : {
2821 [ + - - + ]: 1322 : artifacttype = url_copy.substr(slash2+1);
2822 [ + - ]: 1322 : suffix = "";
2823 : : }
2824 : : else
2825 : : {
2826 [ + - - + ]: 573 : artifacttype = url_copy.substr(slash2+1, slash3-slash2-1);
2827 [ + - - + : 1197 : suffix = url_copy.substr(slash3); // include the slash in the suffix
+ + ]
2828 : : }
2829 : :
2830 : : // get the resulting fd so we can report its size
2831 : 1895 : int fd;
2832 [ + + ]: 1895 : r = handle_buildid(connection, buildid, artifacttype, suffix, &fd);
2833 [ + - ]: 1271 : if (r)
2834 : : {
2835 : 1271 : struct stat fs;
2836 [ + - ]: 1271 : if (fstat(fd, &fs) == 0)
2837 : 1271 : http_size = fs.st_size;
2838 : : // libmicrohttpd will close (fd);
2839 : : }
2840 : 2519 : }
2841 [ + + ]: 719 : else if (url1 == "/metrics")
2842 : : {
2843 [ + - + - : 1426 : tmp_inc_metric m ("thread_busy", "role", "http-metrics");
+ - + - -
+ - + - -
- - ]
2844 [ + - ]: 713 : artifacttype = "metrics";
2845 [ + - + - : 1426 : inc_metric("http_requests_total", "type", artifacttype);
+ - - + -
- ]
2846 [ + - ]: 713 : r = handle_metrics(& http_size);
2847 : 713 : }
2848 [ - + ]: 6 : else if (url1 == "/")
2849 : : {
2850 [ # # ]: 0 : artifacttype = "/";
2851 [ - - - - : 632 : inc_metric("http_requests_total", "type", artifacttype);
- - - - -
- - + ]
2852 [ # # ]: 0 : r = handle_root(& http_size);
2853 : : }
2854 : : else
2855 [ + - + - : 18 : throw reportable_exception("webapi error, unrecognized '" + url1 + "'");
+ - - + ]
2856 : :
2857 [ - + ]: 1984 : if (r == 0)
2858 [ # # # # ]: 0 : throw reportable_exception("internal error, missing response");
2859 : :
2860 [ + + + - ]: 1984 : if (maxsize > 0 && http_size > maxsize)
2861 : : {
2862 [ + - ]: 2 : MHD_destroy_response(r);
2863 [ + - + - : 6 : throw reportable_exception(406, "File too large, max size=" + std::to_string(maxsize));
+ - - + ]
2864 : : }
2865 : :
2866 [ + - ]: 1982 : rc = MHD_queue_response (connection, MHD_HTTP_OK, r);
2867 : 1982 : http_code = MHD_HTTP_OK;
2868 [ + - ]: 1982 : MHD_destroy_response (r);
2869 : 2614 : }
2870 [ - + ]: 632 : catch (const reportable_exception& e)
2871 : : {
2872 [ + - + - : 1264 : inc_metric("http_responses_total","result","error");
+ - + - -
+ - + - -
- - ]
2873 [ + - ]: 632 : e.report(clog);
2874 : 632 : http_code = e.code;
2875 [ + - ]: 632 : http_size = e.message.size();
2876 [ + - ]: 632 : rc = e.mhd_send_response (connection);
2877 : 632 : }
2878 : :
2879 : 2614 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
2880 : 2614 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
2881 : : // afteryou: delay waiting for other client's identical query to complete
2882 : : // deltas: total latency, including afteryou waiting
2883 [ + - + - : 5228 : obatched(clog) << conninfo(connection)
- - ]
2884 : : << ' ' << method << ' ' << url
2885 [ + - + - : 2614 : << ' ' << http_code << ' ' << http_size
+ - + - +
- + - +
- ]
2886 [ + - + - : 2614 : << ' ' << (int)(afteryou*1000) << '+' << (int)((deltas-afteryou)*1000) << "ms"
+ - + - +
- + - +
- ]
2887 [ + - ]: 2614 : << endl;
2888 : :
2889 : : // related prometheus metrics
2890 : 2614 : string http_code_str = to_string(http_code);
2891 [ + - + - : 5228 : add_metric("http_responses_transfer_bytes_sum",
+ - + - -
+ - + - -
- - ]
2892 : : "code", http_code_str, "type", artifacttype, http_size);
2893 [ + - + - : 5228 : inc_metric("http_responses_transfer_bytes_count",
+ - + - -
+ - + - -
- - ]
2894 : : "code", http_code_str, "type", artifacttype);
2895 : :
2896 [ + - + - : 5228 : add_metric("http_responses_duration_milliseconds_sum",
+ - + - -
+ - + - -
- - ]
2897 : : "code", http_code_str, "type", artifacttype, deltas*1000); // prometheus prefers _seconds and floating point
2898 [ + - + - : 5228 : inc_metric("http_responses_duration_milliseconds_count",
+ - + - -
+ - + - -
- - ]
2899 : : "code", http_code_str, "type", artifacttype);
2900 : :
2901 [ + - + - : 5228 : add_metric("http_responses_after_you_milliseconds_sum",
+ - + - -
+ - + - -
- - ]
2902 : : "code", http_code_str, "type", artifacttype, afteryou*1000);
2903 [ + - + - : 5228 : inc_metric("http_responses_after_you_milliseconds_count",
+ - + - -
+ - + - -
- - - - ]
2904 : : "code", http_code_str, "type", artifacttype);
2905 : :
2906 [ - + ]: 2614 : return rc;
2907 [ + + - + : 12197 : }
+ + ]
2908 : :
2909 : :
2910 : : ////////////////////////////////////////////////////////////////////////
2911 : : // borrowed originally from src/nm.c get_local_names()
2912 : :
2913 : : static void
2914 : 244 : dwarf_extract_source_paths (Elf *elf, set<string>& debug_sourcefiles)
2915 : : noexcept // no exceptions - so we can simplify the altdbg resource release at end
2916 : : {
2917 : 244 : Dwarf* dbg = dwarf_begin_elf (elf, DWARF_C_READ, NULL);
2918 [ - + ]: 244 : if (dbg == NULL)
2919 : 0 : return;
2920 : :
2921 : 244 : Dwarf* altdbg = NULL;
2922 : 244 : int altdbg_fd = -1;
2923 : :
2924 : : // DWZ handling: if we have an unsatisfied debug-alt-link, add an
2925 : : // empty string into the outgoing sourcefiles set, so the caller
2926 : : // should know that our data is incomplete.
2927 : 244 : const char *alt_name_p;
2928 : 244 : const void *alt_build_id; // elfutils-owned memory
2929 : 244 : ssize_t sz = dwelf_dwarf_gnu_debugaltlink (dbg, &alt_name_p, &alt_build_id);
2930 [ + + ]: 244 : if (sz > 0) // got one!
2931 : : {
2932 : 40 : string buildid;
2933 : 40 : unsigned char* build_id_bytes = (unsigned char*) alt_build_id;
2934 [ + + ]: 840 : for (ssize_t idx=0; idx<sz; idx++)
2935 : : {
2936 : 800 : buildid += "0123456789abcdef"[build_id_bytes[idx] >> 4];
2937 : 800 : buildid += "0123456789abcdef"[build_id_bytes[idx] & 0xf];
2938 : : }
2939 : :
2940 [ + - ]: 40 : if (verbose > 3)
2941 : 40 : obatched(clog) << "Need altdebug buildid=" << buildid << endl;
2942 : :
2943 : : // but is it unsatisfied the normal elfutils ways?
2944 : 40 : Dwarf* alt = dwarf_getalt (dbg);
2945 [ + - ]: 40 : if (alt == NULL)
2946 : : {
2947 : : // Yup, unsatisfied the normal way. Maybe we can satisfy it
2948 : : // from our own debuginfod database.
2949 : 40 : int alt_fd;
2950 : 40 : struct MHD_Response *r = 0;
2951 : 40 : try
2952 : : {
2953 [ + - ]: 40 : string artifacttype = "debuginfo";
2954 [ + - + - : 40 : r = handle_buildid (0, buildid, artifacttype, "", &alt_fd);
- + - + -
- ]
2955 : 0 : }
2956 [ - - ]: 0 : catch (const reportable_exception& e)
2957 : : {
2958 : : // swallow exceptions
2959 : 0 : }
2960 : :
2961 : : // NB: this is not actually recursive! This invokes the web-query
2962 : : // path, which cannot get back into the scan code paths.
2963 [ + - ]: 40 : if (r)
2964 : : {
2965 : : // Found it!
2966 : 40 : altdbg_fd = dup(alt_fd); // ok if this fails, downstream failures ok
2967 : 40 : alt = altdbg = dwarf_begin (altdbg_fd, DWARF_C_READ);
2968 : : // NB: must close this dwarf and this fd at the bottom of the function!
2969 : 40 : MHD_destroy_response (r); // will close alt_fd
2970 [ + - ]: 40 : if (alt)
2971 : 40 : dwarf_setalt (dbg, alt);
2972 : : }
2973 : : }
2974 : : else
2975 : : {
2976 : : // NB: dwarf_setalt(alt) inappropriate - already done!
2977 : : // NB: altdbg will stay 0 so nothing tries to redundantly dealloc.
2978 : : }
2979 : :
2980 [ + - ]: 40 : if (alt)
2981 : : {
2982 [ + - ]: 40 : if (verbose > 3)
2983 : 40 : obatched(clog) << "Resolved altdebug buildid=" << buildid << endl;
2984 : : }
2985 : : else // (alt == NULL) - signal possible presence of poor debuginfo
2986 : : {
2987 [ # # ]: 0 : debug_sourcefiles.insert("");
2988 [ # # ]: 0 : if (verbose > 3)
2989 : 0 : obatched(clog) << "Unresolved altdebug buildid=" << buildid << endl;
2990 : : }
2991 : 40 : }
2992 : :
2993 : 244 : Dwarf_Off offset = 0;
2994 : 244 : Dwarf_Off old_offset;
2995 : 244 : size_t hsize;
2996 : :
2997 [ + + ]: 8618 : while (dwarf_nextcu (dbg, old_offset = offset, &offset, &hsize, NULL, NULL, NULL) == 0)
2998 : : {
2999 : 8374 : Dwarf_Die cudie_mem;
3000 : 8374 : Dwarf_Die *cudie = dwarf_offdie (dbg, old_offset + hsize, &cudie_mem);
3001 : :
3002 [ - + ]: 8374 : if (cudie == NULL)
3003 : 20 : continue;
3004 [ + + ]: 8374 : if (dwarf_tag (cudie) != DW_TAG_compile_unit)
3005 : 20 : continue;
3006 : :
3007 [ - + ]: 8354 : const char *cuname = dwarf_diename(cudie) ?: "unknown";
3008 : :
3009 : 8354 : Dwarf_Files *files;
3010 : 8354 : size_t nfiles;
3011 [ - + ]: 8354 : if (dwarf_getsrcfiles (cudie, &files, &nfiles) != 0)
3012 : 0 : continue;
3013 : :
3014 : : // extract DW_AT_comp_dir to resolve relative file names
3015 : 8354 : const char *comp_dir = "";
3016 : 8354 : const char *const *dirs;
3017 : 8354 : size_t ndirs;
3018 [ - + ]: 8354 : if (dwarf_getsrcdirs (files, &dirs, &ndirs) == 0 &&
3019 [ - + ]: 8354 : dirs[0] != NULL)
3020 : : comp_dir = dirs[0];
3021 : : if (comp_dir == NULL)
3022 : : comp_dir = "";
3023 : :
3024 [ + + ]: 8354 : if (verbose > 3)
3025 : 12864 : obatched(clog) << "searching for sources for cu=" << cuname << " comp_dir=" << comp_dir
3026 : 6432 : << " #files=" << nfiles << " #dirs=" << ndirs << endl;
3027 : :
3028 [ + - - - ]: 8354 : if (comp_dir[0] == '\0' && cuname[0] != '/')
3029 : : {
3030 [ # # ]: 0 : if (verbose > 3)
3031 : 0 : obatched(clog) << "skipping cu=" << cuname << " due to empty comp_dir" << endl;
3032 : 0 : continue;
3033 : : }
3034 : :
3035 [ + + ]: 134847 : for (size_t f = 1; f < nfiles; f++)
3036 : : {
3037 : 126493 : const char *hat = dwarf_filesrc (files, f, NULL, NULL);
3038 [ - + ]: 126503 : if (hat == NULL)
3039 : 0 : continue;
3040 : :
3041 [ + + - + ]: 240251 : if (string(hat) == "<built-in>") // gcc intrinsics, don't bother record
3042 : 0 : continue;
3043 : :
3044 [ + + ]: 126498 : string waldo;
3045 [ + + ]: 126498 : if (hat[0] == '/') // absolute
3046 [ - + ]: 81142 : waldo = (string (hat));
3047 [ + - ]: 45356 : else if (comp_dir[0] != '\0') // comp_dir relative
3048 [ - + - + : 77966 : waldo = (string (comp_dir) + string("/") + string (hat));
- + - + +
+ ]
3049 : : else
3050 : : {
3051 [ # # ]: 0 : if (verbose > 3)
3052 : 0 : obatched(clog) << "skipping hat=" << hat << " due to empty comp_dir" << endl;
3053 [ # # ]: 0 : continue;
3054 : : }
3055 : :
3056 : : // NB: this is the 'waldo' that a dbginfo client will have
3057 : : // to supply for us to give them the file The comp_dir
3058 : : // prefixing is a definite complication. Otherwise we'd
3059 : : // have to return a setof comp_dirs (one per CU!) with
3060 : : // corresponding filesrc[] names, instead of one absolute
3061 : : // resoved set. Maybe we'll have to do that anyway. XXX
3062 : :
3063 [ + + ]: 126501 : if (verbose > 4)
3064 [ - + ]: 100 : obatched(clog) << waldo
3065 [ - + ]: 50 : << (debug_sourcefiles.find(waldo)==debug_sourcefiles.end() ? " new" : " dup") << endl;
3066 : :
3067 [ + - ]: 126501 : debug_sourcefiles.insert (waldo);
3068 : 126493 : }
3069 : : }
3070 : :
3071 : 244 : dwarf_end(dbg);
3072 [ + + ]: 244 : if (altdbg)
3073 : 40 : dwarf_end(altdbg);
3074 [ + + ]: 244 : if (altdbg_fd >= 0)
3075 : 40 : close(altdbg_fd);
3076 : : }
3077 : :
3078 : :
3079 : :
3080 : : static void
3081 : 1280 : elf_classify (int fd, bool &executable_p, bool &debuginfo_p, string &buildid, set<string>& debug_sourcefiles)
3082 : : {
3083 : 1280 : Elf *elf = elf_begin (fd, ELF_C_READ_MMAP_PRIVATE, NULL);
3084 [ + - ]: 1280 : if (elf == NULL)
3085 : : return;
3086 : :
3087 : 1280 : try // catch our types of errors and clean up the Elf* object
3088 : : {
3089 [ + - + + ]: 1280 : if (elf_kind (elf) != ELF_K_ELF)
3090 : : {
3091 [ + - ]: 792 : elf_end (elf);
3092 : 832 : return;
3093 : : }
3094 : :
3095 : 488 : GElf_Ehdr ehdr_storage;
3096 [ + - ]: 488 : GElf_Ehdr *ehdr = gelf_getehdr (elf, &ehdr_storage);
3097 [ - + ]: 488 : if (ehdr == NULL)
3098 : : {
3099 [ # # ]: 0 : elf_end (elf);
3100 : : return;
3101 : : }
3102 : 488 : auto elf_type = ehdr->e_type;
3103 : :
3104 : 488 : const void *build_id; // elfutils-owned memory
3105 [ + - ]: 488 : ssize_t sz = dwelf_elf_gnu_build_id (elf, & build_id);
3106 [ + + ]: 487 : if (sz <= 0)
3107 : : {
3108 : : // It's not a diagnostic-worthy error for an elf file to lack build-id.
3109 : : // It might just be very old.
3110 [ + - ]: 40 : elf_end (elf);
3111 : : return;
3112 : : }
3113 : :
3114 : : // build_id is a raw byte array; convert to hexadecimal *lowercase*
3115 : 447 : unsigned char* build_id_bytes = (unsigned char*) build_id;
3116 [ + + ]: 9401 : for (ssize_t idx=0; idx<sz; idx++)
3117 : : {
3118 [ + - ]: 8954 : buildid += "0123456789abcdef"[build_id_bytes[idx] >> 4];
3119 [ + - ]: 17907 : buildid += "0123456789abcdef"[build_id_bytes[idx] & 0xf];
3120 : : }
3121 : :
3122 : : // now decide whether it's an executable - namely, any allocatable section has
3123 : : // PROGBITS;
3124 [ + + ]: 447 : if (elf_type == ET_EXEC || elf_type == ET_DYN)
3125 : : {
3126 : 398 : size_t shnum;
3127 [ + - ]: 398 : int rc = elf_getshdrnum (elf, &shnum);
3128 [ - + ]: 398 : if (rc < 0)
3129 [ # # # # ]: 0 : throw elfutils_exception(rc, "getshdrnum");
3130 : :
3131 : 398 : executable_p = false;
3132 [ + + ]: 7054 : for (size_t sc = 0; sc < shnum; sc++)
3133 : : {
3134 [ + - ]: 6878 : Elf_Scn *scn = elf_getscn (elf, sc);
3135 [ - + ]: 6877 : if (scn == NULL)
3136 : 0 : continue;
3137 : :
3138 : 6877 : GElf_Shdr shdr_mem;
3139 [ + - ]: 6877 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem);
3140 [ - + ]: 6878 : if (shdr == NULL)
3141 : 0 : continue;
3142 : :
3143 : : // allocated (loadable / vm-addr-assigned) section with available content?
3144 [ + + + + ]: 6878 : if ((shdr->sh_type == SHT_PROGBITS) && (shdr->sh_flags & SHF_ALLOC))
3145 : : {
3146 [ + + ]: 222 : if (verbose > 4)
3147 [ + - + - : 32 : obatched(clog) << "executable due to SHF_ALLOC SHT_PROGBITS sc=" << sc << endl;
+ - ]
3148 : 222 : executable_p = true;
3149 : 222 : break; // no need to keep looking for others
3150 : : }
3151 : : } // iterate over sections
3152 : : } // executable_p classification
3153 : :
3154 : : // now decide whether it's a debuginfo - namely, if it has any .debug* or .zdebug* sections
3155 : : // logic mostly stolen from fweimer@redhat.com's elfclassify drafts
3156 : 447 : size_t shstrndx;
3157 [ + - ]: 447 : int rc = elf_getshdrstrndx (elf, &shstrndx);
3158 [ - + ]: 448 : if (rc < 0)
3159 [ # # # # ]: 0 : throw elfutils_exception(rc, "getshdrstrndx");
3160 : :
3161 : : Elf_Scn *scn = NULL;
3162 : : bool symtab_p = false;
3163 : : bool bits_alloc_p = false;
3164 : 23544 : while (true)
3165 : : {
3166 [ + - ]: 11996 : scn = elf_nextscn (elf, scn);
3167 [ + + ]: 11967 : if (scn == NULL)
3168 : : break;
3169 : 11763 : GElf_Shdr shdr_storage;
3170 [ + - ]: 11763 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_storage);
3171 [ - + ]: 11763 : if (shdr == NULL)
3172 : : break;
3173 [ + - ]: 11763 : const char *section_name = elf_strptr (elf, shstrndx, shdr->sh_name);
3174 [ - + ]: 11792 : if (section_name == NULL)
3175 : : break;
3176 [ + + ]: 11792 : if (startswith (section_name, ".debug_line") ||
3177 [ - + ]: 11548 : startswith (section_name, ".zdebug_line"))
3178 : : {
3179 : 244 : debuginfo_p = true;
3180 [ + - ]: 244 : if (scan_source_info)
3181 : 244 : dwarf_extract_source_paths (elf, debug_sourcefiles);
3182 : : break; // expecting only one .*debug_line, so no need to look for others
3183 : : }
3184 [ + + ]: 11548 : else if (startswith (section_name, ".debug_") ||
3185 [ + # ]: 10753 : startswith (section_name, ".zdebug_"))
3186 : : {
3187 : 770 : debuginfo_p = true;
3188 : : // NB: don't break; need to parse .debug_line for sources
3189 : : }
3190 [ + + ]: 10778 : else if (shdr->sh_type == SHT_SYMTAB)
3191 : : {
3192 : : symtab_p = true;
3193 : : }
3194 : 10754 : else if (shdr->sh_type != SHT_NOBITS
3195 [ + + ]: 10754 : && shdr->sh_type != SHT_NOTE
3196 [ + + ]: 5517 : && (shdr->sh_flags & SHF_ALLOC) != 0)
3197 : : {
3198 : 11548 : bits_alloc_p = true;
3199 : : }
3200 : 11548 : }
3201 : :
3202 : : // For more expansive elf/split-debuginfo classification, we
3203 : : // want to identify as debuginfo "strip -s"-produced files
3204 : : // without .debug_info* (like libicudata), but we don't want to
3205 : : // identify "strip -g" executables (with .symtab left there).
3206 [ - + ]: 448 : if (symtab_p && !bits_alloc_p)
3207 : 0 : debuginfo_p = true;
3208 : : }
3209 [ # # ]: 0 : catch (const reportable_exception& e)
3210 : : {
3211 [ # # ]: 0 : e.report(clog);
3212 : 0 : }
3213 : 448 : elf_end (elf);
3214 : : }
3215 : :
3216 : :
3217 : : // Intern the given file name in two parts (dirname & basename) and
3218 : : // return the resulting file's id.
3219 : : static int64_t
3220 : 30850 : register_file_name(sqlite_ps& ps_upsert_fileparts,
3221 : : sqlite_ps& ps_upsert_file,
3222 : : sqlite_ps& ps_lookup_file,
3223 : : const string& name)
3224 : : {
3225 : 30850 : std::size_t slash = name.rfind('/');
3226 [ + + ]: 30850 : string dirname, filename;
3227 [ + + ]: 30850 : if (slash == std::string::npos)
3228 : : {
3229 [ + - ]: 90 : dirname = "";
3230 [ + - ]: 90 : filename = name;
3231 : : }
3232 : : else
3233 : : {
3234 [ + - - + ]: 30760 : dirname = name.substr(0, slash);
3235 [ + - - + : 30758 : filename = name.substr(slash+1);
- - ]
3236 : : }
3237 : :
3238 : : // intern the two substrings
3239 : 30849 : ps_upsert_fileparts
3240 [ + - ]: 30849 : .reset()
3241 [ + - ]: 30847 : .bind(1, dirname)
3242 [ + - ]: 30850 : .step_ok_done();
3243 : 30850 : ps_upsert_fileparts
3244 [ + - ]: 30850 : .reset()
3245 [ + - ]: 30850 : .bind(1, filename)
3246 [ + - ]: 30849 : .step_ok_done();
3247 : :
3248 : : // intern the tuple
3249 : 30850 : ps_upsert_file
3250 [ + - ]: 30850 : .reset()
3251 [ + - ]: 30850 : .bind(1, dirname)
3252 [ + - ]: 30849 : .bind(2, filename)
3253 [ + - ]: 30850 : .step_ok_done();
3254 : :
3255 : : // look up the tuple's id
3256 : 30850 : ps_lookup_file
3257 [ + - ]: 30850 : .reset()
3258 [ + - ]: 30850 : .bind(1, dirname)
3259 [ + - ]: 30850 : .bind(2, filename);
3260 [ + - ]: 30850 : int rc = ps_lookup_file.step();
3261 [ - + - - : 30850 : if (rc != SQLITE_ROW) throw sqlite_exception(rc, "step");
- - ]
3262 : :
3263 [ + - ]: 30850 : int64_t id = sqlite3_column_int64 (ps_lookup_file, 0);
3264 [ + - ]: 30850 : ps_lookup_file.reset();
3265 [ + + ]: 30850 : return id;
3266 [ + + ]: 59528 : }
3267 : :
3268 : :
3269 : :
3270 : : static void
3271 : 974 : scan_source_file (const string& rps, const stat_t& st,
3272 : : sqlite_ps& ps_upsert_buildids,
3273 : : sqlite_ps& ps_upsert_fileparts,
3274 : : sqlite_ps& ps_upsert_file,
3275 : : sqlite_ps& ps_lookup_file,
3276 : : sqlite_ps& ps_upsert_de,
3277 : : sqlite_ps& ps_upsert_s,
3278 : : sqlite_ps& ps_query,
3279 : : sqlite_ps& ps_scan_done,
3280 : : unsigned& fts_cached,
3281 : : unsigned& fts_executable,
3282 : : unsigned& fts_debuginfo,
3283 : : unsigned& fts_sourcefiles)
3284 : : {
3285 : 974 : int64_t fileid = register_file_name(ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, rps);
3286 : :
3287 : : /* See if we know of it already. */
3288 : 974 : int rc = ps_query
3289 : 974 : .reset()
3290 : 974 : .bind(1, fileid)
3291 : 974 : .bind(2, st.st_mtime)
3292 : 974 : .step();
3293 : 974 : ps_query.reset();
3294 [ + + ]: 974 : if (rc == SQLITE_ROW) // i.e., a result, as opposed to DONE (no results)
3295 : : // no need to recheck a file/version we already know
3296 : : // specifically, no need to elf-begin a file we already determined is non-elf
3297 : : // (so is stored with buildid=NULL)
3298 : : {
3299 : 368 : fts_cached++;
3300 : 368 : return;
3301 : : }
3302 : :
3303 : 606 : bool executable_p = false, debuginfo_p = false; // E and/or D
3304 [ + - ]: 606 : string buildid;
3305 [ + - ]: 606 : set<string> sourcefiles;
3306 : :
3307 [ + - ]: 606 : int fd = open (rps.c_str(), O_RDONLY);
3308 : 606 : try
3309 : : {
3310 [ + - ]: 606 : if (fd >= 0)
3311 [ + - ]: 606 : elf_classify (fd, executable_p, debuginfo_p, buildid, sourcefiles);
3312 : : else
3313 [ # # # # : 0 : throw libc_exception(errno, string("open ") + rps);
# # # # ]
3314 [ + - + - : 1212 : add_metric ("scanned_bytes_total","source","file",
+ - - + -
+ - - -
- ]
3315 [ + - ]: 606 : st.st_size);
3316 [ + - + - : 1212 : inc_metric ("scanned_files_total","source","file");
+ - + - -
+ - + - -
- - ]
3317 : : }
3318 : : // NB: we catch exceptions here too, so that we can
3319 : : // cache the corrupt-elf case (!executable_p &&
3320 : : // !debuginfo_p) just below, just as if we had an
3321 : : // EPERM error from open(2).
3322 [ - - ]: 0 : catch (const reportable_exception& e)
3323 : : {
3324 [ - - ]: 0 : e.report(clog);
3325 : 0 : }
3326 : :
3327 [ + - ]: 606 : if (fd >= 0)
3328 [ + - ]: 606 : close (fd);
3329 : :
3330 [ + + ]: 606 : if (buildid == "")
3331 : : {
3332 : : // no point storing an elf file without buildid
3333 : 510 : executable_p = false;
3334 : 510 : debuginfo_p = false;
3335 : : }
3336 : : else
3337 : : {
3338 : : // register this build-id in the interning table
3339 : 96 : ps_upsert_buildids
3340 [ + - ]: 96 : .reset()
3341 [ + - ]: 96 : .bind(1, buildid)
3342 [ + - ]: 96 : .step_ok_done();
3343 : : }
3344 : :
3345 [ + + ]: 606 : if (executable_p)
3346 : 72 : fts_executable ++;
3347 [ + + ]: 606 : if (debuginfo_p)
3348 : 72 : fts_debuginfo ++;
3349 [ + + + + ]: 606 : if (executable_p || debuginfo_p)
3350 : : {
3351 : 96 : ps_upsert_de
3352 [ + - ]: 96 : .reset()
3353 [ + - ]: 96 : .bind(1, buildid)
3354 [ + + + - ]: 120 : .bind(2, debuginfo_p ? 1 : 0)
3355 [ + + + - ]: 120 : .bind(3, executable_p ? 1 : 0)
3356 [ + - ]: 96 : .bind(4, fileid)
3357 [ + - ]: 96 : .bind(5, st.st_mtime)
3358 [ + - ]: 96 : .step_ok_done();
3359 : : }
3360 [ + + ]: 606 : if (executable_p)
3361 [ + - + - : 144 : inc_metric("found_executable_total","source","files");
+ - + - -
+ - + - -
- - ]
3362 [ + + ]: 606 : if (debuginfo_p)
3363 [ + - + - : 144 : inc_metric("found_debuginfo_total","source","files");
+ - + - -
+ - + - -
- - ]
3364 : :
3365 [ + + + - ]: 678 : if (sourcefiles.size() && buildid != "")
3366 : : {
3367 : 72 : fts_sourcefiles += sourcefiles.size();
3368 : :
3369 [ + + ]: 14310 : for (auto&& dwarfsrc : sourcefiles)
3370 : : {
3371 [ + - ]: 14238 : char *srp = realpath(dwarfsrc.c_str(), NULL);
3372 [ + + ]: 14238 : if (srp == NULL) // also if DWZ unresolved dwarfsrc=""
3373 : 220 : continue; // unresolvable files are not a serious problem
3374 : : // throw libc_exception(errno, "fts/file realpath " + srcpath);
3375 [ + - ]: 14018 : string srps = string(srp);
3376 : 14018 : free (srp);
3377 : :
3378 : 14018 : struct stat sfs;
3379 : 14018 : rc = stat(srps.c_str(), &sfs);
3380 [ - + ]: 14018 : if (rc != 0)
3381 [ # # ]: 0 : continue;
3382 : :
3383 [ + - ]: 14018 : if (verbose > 2)
3384 [ + - + - : 42054 : obatched(clog) << "recorded buildid=" << buildid << " file=" << srps
- - ]
3385 [ + - + - : 14018 : << " mtime=" << sfs.st_mtime
+ - + - ]
3386 [ + - + - : 14018 : << " as source " << dwarfsrc << endl;
+ - ]
3387 : :
3388 : : // PR25548: store canonicalized dwarfsrc path
3389 [ + - ]: 14018 : string dwarfsrc_canon = canon_pathname (dwarfsrc);
3390 [ + + ]: 14018 : if (dwarfsrc_canon != dwarfsrc)
3391 : : {
3392 [ + + ]: 2382 : if (verbose > 3)
3393 [ + - + - : 3768 : obatched(clog) << "canonicalized src=" << dwarfsrc << " alias=" << dwarfsrc_canon << endl;
+ - + - +
- ]
3394 : : }
3395 : :
3396 [ + - ]: 14018 : int64_t fileid1 = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, dwarfsrc_canon);
3397 [ + - ]: 14018 : int64_t fileid2 = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, srps);
3398 : :
3399 : 14018 : ps_upsert_s
3400 [ + - ]: 14018 : .reset()
3401 [ + - ]: 14018 : .bind(1, buildid)
3402 [ + - ]: 14017 : .bind(2, fileid1)
3403 [ + - ]: 14018 : .bind(3, fileid2)
3404 [ + - ]: 14018 : .bind(4, sfs.st_mtime)
3405 [ + - ]: 14017 : .step_ok_done();
3406 : :
3407 [ + - + - : 28036 : inc_metric("found_sourcerefs_total","source","files");
+ - + - -
+ - + + -
- - - - -
- ]
3408 [ + - ]: 28256 : }
3409 : : }
3410 : :
3411 : 606 : ps_scan_done
3412 [ + - ]: 606 : .reset()
3413 [ + - ]: 606 : .bind(1, fileid)
3414 [ + - ]: 606 : .bind(2, st.st_mtime)
3415 [ + - ]: 606 : .bind(3, st.st_size)
3416 [ + - ]: 606 : .step_ok_done();
3417 : :
3418 [ + - ]: 606 : if (verbose > 2)
3419 [ + - + - ]: 1818 : obatched(clog) << "recorded buildid=" << buildid << " file=" << rps
3420 [ + - + - : 606 : << " mtime=" << st.st_mtime << " atype="
+ - + - ]
3421 : : << (executable_p ? "E" : "")
3422 [ + - + + : 1674 : << (debuginfo_p ? "D" : "") << endl;
+ - + + +
- + - ]
3423 [ + + ]: 702 : }
3424 : :
3425 : :
3426 : :
3427 : :
3428 : :
3429 : : // Analyze given archive file of given age; record buildids / exec/debuginfo-ness of its
3430 : : // constituent files with given upsert statements.
3431 : : static void
3432 : 362 : archive_classify (const string& rps, string& archive_extension, int64_t archiveid,
3433 : : sqlite_ps& ps_upsert_buildids, sqlite_ps& ps_upsert_fileparts, sqlite_ps& ps_upsert_file,
3434 : : sqlite_ps& ps_lookup_file,
3435 : : sqlite_ps& ps_upsert_de, sqlite_ps& ps_upsert_sref, sqlite_ps& ps_upsert_sdef,
3436 : : time_t mtime,
3437 : : unsigned& fts_executable, unsigned& fts_debuginfo, unsigned& fts_sref, unsigned& fts_sdef,
3438 : : bool& fts_sref_complete_p)
3439 : : {
3440 : 362 : string archive_decoder = "/dev/null";
3441 [ + + ]: 896 : for (auto&& arch : scan_archives)
3442 [ + + ]: 534 : if (string_endswith(rps, arch.first))
3443 : : {
3444 [ + - ]: 362 : archive_extension = arch.first;
3445 [ + - ]: 896 : archive_decoder = arch.second;
3446 : : }
3447 : :
3448 : 362 : FILE* fp;
3449 : 362 : defer_dtor<FILE*,int>::dtor_fn dfn;
3450 [ + + ]: 362 : if (archive_decoder != "cat")
3451 : : {
3452 [ + - + - : 64 : string popen_cmd = archive_decoder + " " + shell_escape(rps);
+ - - + -
- - - ]
3453 [ + - ]: 32 : fp = popen (popen_cmd.c_str(), "r"); // "e" O_CLOEXEC?
3454 : 32 : dfn = pclose;
3455 [ - + ]: 32 : if (fp == NULL)
3456 [ # # # # : 0 : throw libc_exception (errno, string("popen ") + popen_cmd);
# # # # ]
3457 : 32 : }
3458 : : else
3459 : : {
3460 [ + - ]: 330 : fp = fopen (rps.c_str(), "r");
3461 : 330 : dfn = fclose;
3462 [ - + ]: 330 : if (fp == NULL)
3463 [ # # # # : 0 : throw libc_exception (errno, string("fopen ") + rps);
# # # # ]
3464 : : }
3465 : 362 : defer_dtor<FILE*,int> fp_closer (fp, dfn);
3466 : :
3467 : 362 : struct archive *a;
3468 [ + - ]: 362 : a = archive_read_new();
3469 [ - + ]: 362 : if (a == NULL)
3470 [ # # # # ]: 0 : throw archive_exception("cannot create archive reader");
3471 : 362 : defer_dtor<struct archive*,int> archive_closer (a, archive_read_free);
3472 : :
3473 [ + - ]: 362 : int rc = archive_read_support_format_all(a);
3474 [ - + ]: 362 : if (rc != ARCHIVE_OK)
3475 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all formats");
3476 [ + - ]: 362 : rc = archive_read_support_filter_all(a);
3477 [ - + ]: 362 : if (rc != ARCHIVE_OK)
3478 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all filters");
3479 : :
3480 [ + - ]: 362 : rc = archive_read_open_FILE (a, fp);
3481 [ - + ]: 362 : if (rc != ARCHIVE_OK)
3482 : : {
3483 [ # # # # : 0 : obatched(clog) << "cannot open archive from pipe " << rps << endl;
# # ]
3484 [ # # # # ]: 0 : throw archive_exception(a, "cannot open archive from pipe");
3485 : : }
3486 : :
3487 [ + + ]: 362 : if (verbose > 3)
3488 [ + - + - : 692 : obatched(clog) << "libarchive scanning " << rps << " id " << archiveid << endl;
+ - + - +
- ]
3489 : :
3490 : : bool any_exceptions = false;
3491 : 2564 : while(1) // parse archive entries
3492 : : {
3493 [ + - ]: 2564 : if (interrupted)
3494 : : break;
3495 : :
3496 : 2564 : try
3497 : : {
3498 : 2564 : struct archive_entry *e;
3499 [ + - ]: 2564 : rc = archive_read_next_header (a, &e);
3500 [ + + ]: 2564 : if (rc != ARCHIVE_OK)
3501 : : break;
3502 : :
3503 [ + - + + ]: 2202 : if (! S_ISREG(archive_entry_mode (e))) // skip non-files completely
3504 : 1528 : continue;
3505 : :
3506 [ + - ]: 674 : string fn = canonicalized_archive_entry_pathname (e);
3507 : :
3508 [ + + ]: 674 : if (verbose > 3)
3509 [ + - + - : 1268 : obatched(clog) << "libarchive checking " << fn << endl;
+ - - - ]
3510 : :
3511 : : // extract this file to a temporary file
3512 : 674 : char* tmppath = NULL;
3513 : 674 : rc = asprintf (&tmppath, "%s/debuginfod-classify.XXXXXX", tmpdir.c_str());
3514 [ - + ]: 674 : if (rc < 0)
3515 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
3516 : 674 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
3517 [ + - ]: 674 : int fd = mkstemp (tmppath);
3518 [ - + ]: 674 : if (fd < 0)
3519 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
3520 : 674 : unlink (tmppath); // unlink now so OS will release the file as soon as we close the fd
3521 : 674 : defer_dtor<int,int> minifd_closer (fd, close);
3522 : :
3523 [ + - ]: 674 : rc = archive_read_data_into_fd (a, fd);
3524 [ - + ]: 674 : if (rc != ARCHIVE_OK) {
3525 [ # # ]: 0 : close (fd);
3526 [ # # # # ]: 0 : throw archive_exception(a, "cannot extract file");
3527 : : }
3528 : :
3529 : : // finally ... time to run elf_classify on this bad boy and update the database
3530 : 674 : bool executable_p = false, debuginfo_p = false;
3531 [ + - ]: 674 : string buildid;
3532 [ + - ]: 674 : set<string> sourcefiles;
3533 [ + - ]: 674 : elf_classify (fd, executable_p, debuginfo_p, buildid, sourcefiles);
3534 : : // NB: might throw
3535 : :
3536 [ + + ]: 674 : if (buildid != "") // intern buildid
3537 : : {
3538 : 352 : ps_upsert_buildids
3539 [ + - ]: 352 : .reset()
3540 [ + - ]: 352 : .bind(1, buildid)
3541 [ + - ]: 352 : .step_ok_done();
3542 : : }
3543 : :
3544 [ + - ]: 674 : int64_t fileid = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, fn);
3545 : :
3546 [ + + ]: 674 : if (sourcefiles.size() > 0) // sref records needed
3547 : : {
3548 : : // NB: we intern each source file once. Once raw, as it
3549 : : // appears in the DWARF file list coming back from
3550 : : // elf_classify() - because it'll end up in the
3551 : : // _norm.artifactsrc column. We don't also put another
3552 : : // version with a '.' at the front, even though that's
3553 : : // how rpm/cpio packs names, because we hide that from
3554 : : // the database for storage efficiency.
3555 : :
3556 [ + + ]: 616 : for (auto&& s : sourcefiles)
3557 : : {
3558 [ - + ]: 464 : if (s == "")
3559 : : {
3560 : 0 : fts_sref_complete_p = false;
3561 : 0 : continue;
3562 : : }
3563 : :
3564 : : // PR25548: store canonicalized source path
3565 : 464 : const string& dwarfsrc = s;
3566 [ + - ]: 464 : string dwarfsrc_canon = canon_pathname (dwarfsrc);
3567 [ + + ]: 464 : if (dwarfsrc_canon != dwarfsrc)
3568 : : {
3569 [ + - ]: 32 : if (verbose > 3)
3570 [ + - + - : 64 : obatched(clog) << "canonicalized src=" << dwarfsrc << " alias=" << dwarfsrc_canon << endl;
+ - + - +
- - - ]
3571 : : }
3572 : :
3573 [ + - ]: 464 : int64_t srcfileid = register_file_name(ps_upsert_fileparts, ps_upsert_file, ps_lookup_file,
3574 : : dwarfsrc_canon);
3575 : :
3576 : 464 : ps_upsert_sref
3577 [ + - ]: 464 : .reset()
3578 [ + - ]: 464 : .bind(1, buildid)
3579 [ + - ]: 463 : .bind(2, srcfileid)
3580 [ + - ]: 464 : .step_ok_done();
3581 : :
3582 [ + - ]: 464 : fts_sref ++;
3583 : 464 : }
3584 : : }
3585 : :
3586 [ + + ]: 674 : if (executable_p)
3587 : 150 : fts_executable ++;
3588 [ + + ]: 674 : if (debuginfo_p)
3589 : 202 : fts_debuginfo ++;
3590 : :
3591 [ + + + + ]: 674 : if (executable_p || debuginfo_p)
3592 : : {
3593 : 352 : ps_upsert_de
3594 [ + - ]: 352 : .reset()
3595 [ + - ]: 352 : .bind(1, buildid)
3596 [ + + + - ]: 502 : .bind(2, debuginfo_p ? 1 : 0)
3597 [ + + + - ]: 554 : .bind(3, executable_p ? 1 : 0)
3598 [ + - ]: 352 : .bind(4, archiveid)
3599 [ + - ]: 352 : .bind(5, mtime)
3600 [ + - ]: 352 : .bind(6, fileid)
3601 [ + - ]: 352 : .step_ok_done();
3602 : : }
3603 : : else // potential source - sdef record
3604 : : {
3605 : 322 : fts_sdef ++;
3606 : 322 : ps_upsert_sdef
3607 [ + - ]: 322 : .reset()
3608 [ + - ]: 322 : .bind(1, archiveid)
3609 [ + - ]: 322 : .bind(2, mtime)
3610 [ + - ]: 322 : .bind(3, fileid)
3611 [ + - ]: 322 : .step_ok_done();
3612 : : }
3613 : :
3614 [ + - + + : 674 : if ((verbose > 2) && (executable_p || debuginfo_p))
+ + ]
3615 [ + - + - ]: 1056 : obatched(clog) << "recorded buildid=" << buildid << " rpm=" << rps << " file=" << fn
3616 [ + - + - : 351 : << " mtime=" << mtime << " atype="
+ - + - +
- + - ]
3617 : : << (executable_p ? "E" : "")
3618 : : << (debuginfo_p ? "D" : "")
3619 [ + - + + : 703 : << " sourcefiles=" << sourcefiles.size() << endl;
+ - + + +
- + - + -
+ - ]
3620 : :
3621 [ + + + + ]: 1532 : }
3622 [ - - ]: 0 : catch (const reportable_exception& e)
3623 : : {
3624 [ - - ]: 0 : e.report(clog);
3625 : 0 : any_exceptions = true;
3626 : : // NB: but we allow the libarchive iteration to continue, in
3627 : : // case we can still gather some useful information. That
3628 : : // would allow some webapi queries to work, until later when
3629 : : // this archive is rescanned. (Its vitals won't go into the
3630 : : // _file_mtime_scanned table until after a successful scan.)
3631 : 0 : }
3632 : : }
3633 : :
3634 [ - + ]: 362 : if (any_exceptions)
3635 [ # # # # ]: 0 : throw reportable_exception("exceptions encountered during archive scan");
3636 [ + + ]: 377 : }
3637 : :
3638 : :
3639 : :
3640 : : // scan for archive files such as .rpm
3641 : : static void
3642 : 702 : scan_archive_file (const string& rps, const stat_t& st,
3643 : : sqlite_ps& ps_upsert_buildids,
3644 : : sqlite_ps& ps_upsert_fileparts,
3645 : : sqlite_ps& ps_upsert_file,
3646 : : sqlite_ps& ps_lookup_file,
3647 : : sqlite_ps& ps_upsert_de,
3648 : : sqlite_ps& ps_upsert_sref,
3649 : : sqlite_ps& ps_upsert_sdef,
3650 : : sqlite_ps& ps_query,
3651 : : sqlite_ps& ps_scan_done,
3652 : : unsigned& fts_cached,
3653 : : unsigned& fts_executable,
3654 : : unsigned& fts_debuginfo,
3655 : : unsigned& fts_sref,
3656 : : unsigned& fts_sdef)
3657 : : {
3658 : : // intern the archive file name
3659 : 702 : int64_t archiveid = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, rps);
3660 : :
3661 : : /* See if we know of it already. */
3662 : 702 : int rc = ps_query
3663 : 702 : .reset()
3664 : 702 : .bind(1, archiveid)
3665 : 702 : .bind(2, st.st_mtime)
3666 : 702 : .step();
3667 : 702 : ps_query.reset();
3668 [ + + ]: 702 : if (rc == SQLITE_ROW) // i.e., a result, as opposed to DONE (no results)
3669 : : // no need to recheck a file/version we already know
3670 : : // specifically, no need to parse this archive again, since we already have
3671 : : // it as a D or E or S record,
3672 : : // (so is stored with buildid=NULL)
3673 : : {
3674 : 340 : fts_cached ++;
3675 : 340 : return;
3676 : : }
3677 : :
3678 : : // extract the archive contents
3679 : 362 : unsigned my_fts_executable = 0, my_fts_debuginfo = 0, my_fts_sref = 0, my_fts_sdef = 0;
3680 : 362 : bool my_fts_sref_complete_p = true;
3681 : 362 : bool any_exceptions = false;
3682 : 362 : try
3683 : : {
3684 [ + - ]: 362 : string archive_extension;
3685 : 362 : archive_classify (rps, archive_extension, archiveid,
3686 : : ps_upsert_buildids, ps_upsert_fileparts, ps_upsert_file, ps_lookup_file,
3687 : : ps_upsert_de, ps_upsert_sref, ps_upsert_sdef, // dalt
3688 [ + - ]: 362 : st.st_mtime,
3689 : : my_fts_executable, my_fts_debuginfo, my_fts_sref, my_fts_sdef,
3690 : : my_fts_sref_complete_p);
3691 [ + - + - : 723 : add_metric ("scanned_bytes_total","source",archive_extension + " archive",
+ - - + +
+ - - -
- ]
3692 [ + - ]: 361 : st.st_size);
3693 [ + - + - : 724 : inc_metric ("scanned_files_total","source",archive_extension + " archive");
+ - + - -
+ + + - -
- - ]
3694 [ + - + - : 724 : add_metric("found_debuginfo_total","source",archive_extension + " archive",
+ - + - -
+ + + - -
- - ]
3695 : : my_fts_debuginfo);
3696 [ + - + - : 724 : add_metric("found_executable_total","source",archive_extension + " archive",
+ - + - -
+ + + - -
- - ]
3697 : : my_fts_executable);
3698 [ + - + - : 740 : add_metric("found_sourcerefs_total","source",archive_extension + " archive",
+ - + - -
+ + + - +
- - - - -
- ]
3699 : : my_fts_sref);
3700 : 362 : }
3701 [ - - ]: 0 : catch (const reportable_exception& e)
3702 : : {
3703 [ - - ]: 0 : e.report(clog);
3704 : 0 : any_exceptions = true;
3705 : 0 : }
3706 : :
3707 [ + - ]: 362 : if (verbose > 2)
3708 [ + - ]: 1086 : obatched(clog) << "scanned archive=" << rps
3709 [ + - + - ]: 362 : << " mtime=" << st.st_mtime
3710 [ + - ]: 362 : << " executables=" << my_fts_executable
3711 [ + - + - ]: 362 : << " debuginfos=" << my_fts_debuginfo
3712 [ + - + - ]: 362 : << " srefs=" << my_fts_sref
3713 [ + - + - ]: 362 : << " sdefs=" << my_fts_sdef
3714 [ + - + - : 362 : << " exceptions=" << any_exceptions
+ - + - ]
3715 : 362 : << endl;
3716 : :
3717 : 362 : fts_executable += my_fts_executable;
3718 : 362 : fts_debuginfo += my_fts_debuginfo;
3719 : 362 : fts_sref += my_fts_sref;
3720 : 362 : fts_sdef += my_fts_sdef;
3721 : :
3722 [ - + ]: 362 : if (any_exceptions)
3723 [ # # # # ]: 0 : throw reportable_exception("exceptions encountered during archive scan");
3724 : :
3725 [ + - ]: 362 : if (my_fts_sref_complete_p) // leave incomplete?
3726 : 362 : ps_scan_done
3727 : 362 : .reset()
3728 : 362 : .bind(1, archiveid)
3729 : 362 : .bind(2, st.st_mtime)
3730 : 362 : .bind(3, st.st_size)
3731 : 362 : .step_ok_done();
3732 : : }
3733 : :
3734 : :
3735 : :
3736 : : ////////////////////////////////////////////////////////////////////////
3737 : :
3738 : :
3739 : :
3740 : : // The thread that consumes file names off of the scanq. We hold
3741 : : // the persistent sqlite_ps's at this level and delegate file/archive
3742 : : // scanning to other functions.
3743 : : static void
3744 : 264 : scan ()
3745 : : {
3746 : : // all the prepared statements fit to use, the _f_ set:
3747 [ + - + - : 528 : sqlite_ps ps_f_upsert_buildids (db, "file-buildids-intern", "insert or ignore into " BUILDIDS "_buildids VALUES (NULL, ?);");
+ - - - ]
3748 [ + - + - : 528 : sqlite_ps ps_f_upsert_fileparts (db, "file-fileparts-intern", "insert or ignore into " BUILDIDS "_fileparts VALUES (NULL, ?);");
+ - + - -
- ]
3749 [ + - ]: 264 : sqlite_ps ps_f_upsert_file (db, "file-file-intern", "insert or ignore into " BUILDIDS "_files VALUES (NULL, \n"
3750 : : "(select id from " BUILDIDS "_fileparts where name = ?),\n"
3751 [ + - + - : 528 : "(select id from " BUILDIDS "_fileparts where name = ?));");
+ - + - -
- ]
3752 [ + - ]: 264 : sqlite_ps ps_f_lookup_file (db, "file-file-lookup",
3753 : : "select f.id\n"
3754 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts p1, " BUILDIDS "_fileparts p2 \n"
3755 [ + - + - : 528 : " where f.dirname = p1.id and f.basename = p2.id and p1.name = ? and p2.name = ?;\n");
+ - + - -
- ]
3756 [ + - ]: 264 : sqlite_ps ps_f_upsert_de (db, "file-de-upsert",
3757 : : "insert or ignore into " BUILDIDS "_f_de "
3758 : : "(buildid, debuginfo_p, executable_p, file, mtime) "
3759 : : "values ((select id from " BUILDIDS "_buildids where hex = ?),"
3760 [ + - + - : 528 : " ?,?,?,?);");
+ - + - -
- ]
3761 [ + - ]: 264 : sqlite_ps ps_f_upsert_s (db, "file-s-upsert",
3762 : : "insert or ignore into " BUILDIDS "_f_s "
3763 : : "(buildid, artifactsrc, file, mtime) "
3764 : : "values ((select id from " BUILDIDS "_buildids where hex = ?),"
3765 [ + - + - : 528 : " ?,?,?);");
+ - + - -
- ]
3766 [ + - ]: 264 : sqlite_ps ps_f_query (db, "file-negativehit-find",
3767 : : "select 1 from " BUILDIDS "_file_mtime_scanned where sourcetype = 'F' "
3768 [ + - + - : 528 : "and file = ? and mtime = ?;");
+ - + - -
- ]
3769 [ + - ]: 264 : sqlite_ps ps_f_scan_done (db, "file-scanned",
3770 : : "insert or ignore into " BUILDIDS "_file_mtime_scanned (sourcetype, file, mtime, size)"
3771 [ + - + - : 528 : "values ('F', ?,?,?);");
+ - + - -
- ]
3772 : :
3773 : : // and now for the _r_ set
3774 [ + - + - : 528 : sqlite_ps ps_r_upsert_buildids (db, "rpm-buildid-intern", "insert or ignore into " BUILDIDS "_buildids VALUES (NULL, ?);");
+ - + - -
- ]
3775 [ + - + - : 528 : sqlite_ps ps_r_upsert_fileparts (db, "rpm-fileparts-intern", "insert or ignore into " BUILDIDS "_fileparts VALUES (NULL, ?);");
+ - + - -
- ]
3776 [ + - ]: 264 : sqlite_ps ps_r_upsert_file (db, "rpm-file-intern", "insert or ignore into " BUILDIDS "_files VALUES (NULL, \n"
3777 : : "(select id from " BUILDIDS "_fileparts where name = ?),\n"
3778 [ + - + - : 528 : "(select id from " BUILDIDS "_fileparts where name = ?));");
+ - + - -
- ]
3779 [ + - ]: 264 : sqlite_ps ps_r_lookup_file (db, "rpm-file-lookup",
3780 : : "select f.id\n"
3781 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts p1, " BUILDIDS "_fileparts p2 \n"
3782 [ + - + - : 528 : " where f.dirname = p1.id and f.basename = p2.id and p1.name = ? and p2.name = ?;\n");
+ - + - -
- ]
3783 [ + - ]: 264 : sqlite_ps ps_r_upsert_de (db, "rpm-de-insert",
3784 : : "insert or ignore into " BUILDIDS "_r_de (buildid, debuginfo_p, executable_p, file, mtime, content) values ("
3785 [ + - + - : 528 : "(select id from " BUILDIDS "_buildids where hex = ?), ?, ?, ?, ?, ?);");
+ - + - -
- ]
3786 [ + - ]: 264 : sqlite_ps ps_r_upsert_sref (db, "rpm-sref-insert",
3787 : : "insert or ignore into " BUILDIDS "_r_sref (buildid, artifactsrc) values ("
3788 : : "(select id from " BUILDIDS "_buildids where hex = ?), "
3789 [ + - + - : 528 : "?);");
+ - + - -
- ]
3790 [ + - ]: 264 : sqlite_ps ps_r_upsert_sdef (db, "rpm-sdef-insert",
3791 : : "insert or ignore into " BUILDIDS "_r_sdef (file, mtime, content) values ("
3792 [ + - + - : 528 : "?, ?, ?);");
+ - + - -
- ]
3793 [ + - ]: 264 : sqlite_ps ps_r_query (db, "rpm-negativehit-query",
3794 : : "select 1 from " BUILDIDS "_file_mtime_scanned where "
3795 [ + - + - : 528 : "sourcetype = 'R' and file = ? and mtime = ?;");
+ - + - -
- ]
3796 [ + - ]: 264 : sqlite_ps ps_r_scan_done (db, "rpm-scanned",
3797 : : "insert or ignore into " BUILDIDS "_file_mtime_scanned (sourcetype, file, mtime, size)"
3798 [ + - + - : 528 : "values ('R', ?, ?, ?);");
+ - + - -
- ]
3799 : :
3800 : :
3801 : 264 : unsigned fts_cached = 0, fts_executable = 0, fts_debuginfo = 0, fts_sourcefiles = 0;
3802 : 264 : unsigned fts_sref = 0, fts_sdef = 0;
3803 : :
3804 [ + - + - : 528 : add_metric("thread_count", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
3805 [ + - + - : 528 : add_metric("thread_busy", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
3806 [ + + ]: 1714 : while (! interrupted)
3807 : : {
3808 [ + - ]: 1452 : scan_payload p;
3809 : :
3810 [ + - + - : 2902 : add_metric("thread_busy", "role", "scan", -1);
+ - + - -
+ - + - -
- - ]
3811 : : // NB: threads may be blocked within either of these two waiting
3812 : : // states, if the work queue happens to run dry. That's OK.
3813 [ + - + - ]: 1450 : if (scan_barrier) scan_barrier->count();
3814 [ + - ]: 1450 : bool gotone = scanq.wait_front(p);
3815 [ + - + - : 2899 : add_metric("thread_busy", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
3816 : :
3817 [ + + - + ]: 1450 : if (! gotone) continue; // go back to waiting
3818 : :
3819 : 1186 : try
3820 : : {
3821 : 1186 : bool scan_archive = false;
3822 [ + + ]: 2486 : for (auto&& arch : scan_archives)
3823 [ + + ]: 1300 : if (string_endswith(p.first, arch.first))
3824 : 702 : scan_archive = true;
3825 : :
3826 [ + + ]: 1186 : if (scan_archive)
3827 [ + - ]: 702 : scan_archive_file (p.first, p.second,
3828 : : ps_r_upsert_buildids,
3829 : : ps_r_upsert_fileparts,
3830 : : ps_r_upsert_file,
3831 : : ps_r_lookup_file,
3832 : : ps_r_upsert_de,
3833 : : ps_r_upsert_sref,
3834 : : ps_r_upsert_sdef,
3835 : : ps_r_query,
3836 : : ps_r_scan_done,
3837 : : fts_cached,
3838 : : fts_executable,
3839 : : fts_debuginfo,
3840 : : fts_sref,
3841 : : fts_sdef);
3842 : :
3843 [ + + ]: 1186 : if (scan_files) // NB: maybe "else if" ?
3844 [ + - ]: 974 : scan_source_file (p.first, p.second,
3845 : : ps_f_upsert_buildids,
3846 : : ps_f_upsert_fileparts,
3847 : : ps_f_upsert_file,
3848 : : ps_f_lookup_file,
3849 : : ps_f_upsert_de,
3850 : : ps_f_upsert_s,
3851 : : ps_f_query,
3852 : : ps_f_scan_done,
3853 : : fts_cached, fts_executable, fts_debuginfo, fts_sourcefiles);
3854 : : }
3855 [ - - ]: 0 : catch (const reportable_exception& e)
3856 : : {
3857 [ - - ]: 0 : e.report(cerr);
3858 : 0 : }
3859 : :
3860 [ + - ]: 1186 : scanq.done_front(); // let idlers run
3861 : :
3862 : 1186 : if (fts_cached || fts_executable || fts_debuginfo || fts_sourcefiles || fts_sref || fts_sdef)
3863 : : {} // NB: not just if a successful scan - we might have encountered -ENOSPC & failed
3864 [ + - + - ]: 1186 : (void) statfs_free_enough_p(db_path, "database"); // report sqlite filesystem size
3865 [ + - + - ]: 1186 : (void) statfs_free_enough_p(tmpdir, "tmpdir"); // this too, in case of fdcache/tmpfile usage
3866 : :
3867 : : // finished a scanning step -- not a "loop", because we just
3868 : : // consume the traversal loop's work, whenever
3869 [ + - + - : 2372 : inc_metric("thread_work_total","role","scan");
+ - + - -
+ - + + -
- - - - -
- ]
3870 : 1450 : }
3871 : :
3872 [ + - + - : 525 : add_metric("thread_busy", "role", "scan", -1);
+ - + - -
+ - + - -
- - ]
3873 : 263 : }
3874 : :
3875 : :
3876 : : // Use this function as the thread entry point, so it can catch our
3877 : : // fleet of exceptions (incl. the sqlite_ps ctors) and report.
3878 : : static void*
3879 : 264 : thread_main_scanner (void* arg)
3880 : : {
3881 : 264 : (void) arg;
3882 [ + + ]: 792 : while (! interrupted)
3883 : 264 : try
3884 : : {
3885 [ + - ]: 264 : scan();
3886 : : }
3887 [ - - ]: 0 : catch (const reportable_exception& e)
3888 : : {
3889 [ - - ]: 0 : e.report(cerr);
3890 : 0 : }
3891 : 264 : return 0;
3892 : : }
3893 : :
3894 : :
3895 : :
3896 : : // The thread that traverses all the source_paths and enqueues all the
3897 : : // matching files into the file/archive scan queue.
3898 : : static void
3899 : 120 : scan_source_paths()
3900 : : {
3901 : : // NB: fedora 31 glibc/fts(3) crashes inside fts_read() on empty
3902 : : // path list.
3903 [ + + ]: 120 : if (source_paths.empty())
3904 : 2 : return;
3905 : :
3906 : : // Turn the source_paths into an fts(3)-compatible char**. Since
3907 : : // source_paths[] does not change after argv processing, the
3908 : : // c_str()'s are safe to keep around awile.
3909 : 118 : vector<const char *> sps;
3910 [ + + ]: 318 : for (auto&& sp: source_paths)
3911 [ + - ]: 200 : sps.push_back(sp.c_str());
3912 [ + - - - ]: 118 : sps.push_back(NULL);
3913 : :
3914 [ + + + - ]: 222 : FTS *fts = fts_open ((char * const *)sps.data(),
3915 : : (traverse_logical ? FTS_LOGICAL : FTS_PHYSICAL|FTS_XDEV)
3916 : : | FTS_NOCHDIR /* multithreaded */,
3917 : : NULL);
3918 [ - + ]: 118 : if (fts == NULL)
3919 [ # # # # ]: 0 : throw libc_exception(errno, "cannot fts_open");
3920 : 118 : defer_dtor<FTS*,int> fts_cleanup (fts, fts_close);
3921 : :
3922 : 118 : struct timespec ts_start, ts_end;
3923 : 118 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
3924 : 118 : unsigned fts_scanned = 0, fts_regex = 0;
3925 : :
3926 : 118 : FTSENT *f;
3927 [ + - + + ]: 2374 : while ((f = fts_read (fts)) != NULL)
3928 : : {
3929 [ + - ]: 2138 : if (interrupted) break;
3930 : :
3931 [ - + ]: 2138 : if (sigusr2 != forced_groom_count) // stop early if groom triggered
3932 : : {
3933 [ # # ]: 0 : scanq.clear(); // clear previously issued work for scanner threads
3934 : : break;
3935 : : }
3936 : :
3937 : 2138 : fts_scanned ++;
3938 : :
3939 [ + - ]: 2138 : if (verbose > 2)
3940 [ + - + - : 4276 : obatched(clog) << "fts traversing " << f->fts_path << endl;
+ - ]
3941 : :
3942 [ + + + + : 2138 : switch (f->fts_info)
+ ]
3943 : : {
3944 : 1274 : case FTS_F:
3945 : 1274 : {
3946 : : /* Found a file. Convert it to an absolute path, so
3947 : : the buildid database does not have relative path
3948 : : names that are unresolvable from a subsequent run
3949 : : in a different cwd. */
3950 [ + - ]: 1274 : char *rp = realpath(f->fts_path, NULL);
3951 [ - + ]: 1274 : if (rp == NULL)
3952 : 0 : continue; // ignore dangling symlink or such
3953 [ + - ]: 1274 : string rps = string(rp);
3954 : 1274 : free (rp);
3955 : :
3956 [ + - ]: 1274 : bool ri = !regexec (&file_include_regex, rps.c_str(), 0, 0, 0);
3957 [ + - ]: 1274 : bool rx = !regexec (&file_exclude_regex, rps.c_str(), 0, 0, 0);
3958 [ + + ]: 1274 : if (!ri || rx)
3959 : : {
3960 [ + - ]: 88 : if (verbose > 3)
3961 [ + - ]: 176 : obatched(clog) << "fts skipped by regex "
3962 [ + + + - : 96 : << (!ri ? "I" : "") << (rx ? "X" : "") << endl;
+ + + - +
- ]
3963 : 88 : fts_regex ++;
3964 [ + + ]: 88 : if (!ri)
3965 [ + - + - : 8 : inc_metric("traversed_total","type","file-skipped-I");
+ - + - -
+ - + - -
- - ]
3966 [ + + ]: 88 : if (rx)
3967 [ + - + - : 168 : inc_metric("traversed_total","type","file-skipped-X");
+ - + - -
+ - + - -
- - ]
3968 : : }
3969 : : else
3970 : : {
3971 [ + - + - ]: 1186 : scanq.push_back (make_pair(rps, *f->fts_statp));
3972 [ + - + - : 2372 : inc_metric("traversed_total","type","file");
+ - + - -
+ - + - -
- - - - ]
3973 : : }
3974 : 0 : }
3975 : 1274 : break;
3976 : :
3977 : 4 : case FTS_ERR:
3978 : 4 : case FTS_NS:
3979 : : // report on some types of errors because they may reflect fixable misconfiguration
3980 : 4 : {
3981 [ + - + - : 8 : auto x = libc_exception(f->fts_errno, string("fts traversal ") + string(f->fts_path));
+ - + - -
+ - + -
- ]
3982 [ + - ]: 4 : x.report(cerr);
3983 : 0 : }
3984 [ + - + - : 8 : inc_metric("traversed_total","type","error");
+ - + - -
+ - + - -
- - ]
3985 : 4 : break;
3986 : :
3987 : 12 : case FTS_SL: // ignore, but count because debuginfod -L would traverse these
3988 [ + - + - : 24 : inc_metric("traversed_total","type","symlink");
+ - + - -
+ - + - -
- - ]
3989 : 12 : break;
3990 : :
3991 : 424 : case FTS_D: // ignore
3992 [ + - + - : 848 : inc_metric("traversed_total","type","directory");
+ - + - -
+ - + - -
- - ]
3993 : 424 : break;
3994 : :
3995 : 424 : default: // ignore
3996 [ + - + - : 848 : inc_metric("traversed_total","type","other");
+ - + - -
+ - + - -
- - ]
3997 : 424 : break;
3998 : : }
3999 : : }
4000 : 118 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
4001 : 118 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
4002 : :
4003 [ + - + - : 354 : obatched(clog) << "fts traversed source paths in " << deltas << "s, scanned=" << fts_scanned
+ - + - ]
4004 [ + - + - : 118 : << ", regex-skipped=" << fts_regex << endl;
+ - ]
4005 [ + - ]: 236 : }
4006 : :
4007 : :
4008 : : static void*
4009 : 66 : thread_main_fts_source_paths (void* arg)
4010 : : {
4011 : 66 : (void) arg; // ignore; we operate on global data
4012 : :
4013 [ + - + - : 132 : set_metric("thread_tid", "role","traverse", tid());
+ - - + -
+ - - -
- ]
4014 [ + - + - : 132 : add_metric("thread_count", "role", "traverse", 1);
+ - - + -
+ - - -
- ]
4015 : :
4016 : 66 : time_t last_rescan = 0;
4017 : :
4018 [ + - ]: 297 : while (! interrupted)
4019 : : {
4020 : 297 : sleep (1);
4021 : 297 : scanq.wait_idle(); // don't start a new traversal while scanners haven't finished the job
4022 : 297 : scanq.done_idle(); // release the hounds
4023 [ + + ]: 297 : if (interrupted) break;
4024 : :
4025 : 231 : time_t now = time(NULL);
4026 : 231 : bool rescan_now = false;
4027 [ + + ]: 231 : if (last_rescan == 0) // at least one initial rescan is documented even for -t0
4028 : 64 : rescan_now = true;
4029 [ + + + + ]: 231 : if (rescan_s > 0 && (long)now > (long)(last_rescan + rescan_s))
4030 : 231 : rescan_now = true;
4031 [ + + ]: 231 : if (sigusr1 != forced_rescan_count)
4032 : : {
4033 : 60 : forced_rescan_count = sigusr1;
4034 : 60 : rescan_now = true;
4035 : : }
4036 [ + + ]: 231 : if (rescan_now)
4037 : : {
4038 [ + - + - : 240 : set_metric("thread_busy", "role","traverse", 1);
+ - - + -
+ - - -
- ]
4039 : 120 : try
4040 : : {
4041 [ + - ]: 120 : scan_source_paths();
4042 : : }
4043 [ - - ]: 0 : catch (const reportable_exception& e)
4044 : : {
4045 [ - - ]: 0 : e.report(cerr);
4046 : 0 : }
4047 : 120 : last_rescan = time(NULL); // NB: now was before scanning
4048 : : // finished a traversal loop
4049 [ + - + - : 240 : inc_metric("thread_work_total", "role","traverse");
+ - - + -
+ - - -
- ]
4050 [ + - + - : 240 : set_metric("thread_busy", "role","traverse", 0);
+ - - + -
+ - - -
- ]
4051 : : }
4052 : : }
4053 : :
4054 : 66 : return 0;
4055 : : }
4056 : :
4057 : :
4058 : :
4059 : : ////////////////////////////////////////////////////////////////////////
4060 : :
4061 : : static void
4062 : 72 : database_stats_report()
4063 : : {
4064 : 72 : sqlite_ps ps_query (db, "database-overview",
4065 [ + - + - : 144 : "select label,quantity from " BUILDIDS "_stats");
+ - - - ]
4066 : :
4067 [ + - + - ]: 144 : obatched(clog) << "database record counts:" << endl;
4068 : 1656 : while (1)
4069 : : {
4070 [ + - ]: 864 : if (interrupted) break;
4071 [ + - ]: 864 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
4072 : : break;
4073 : :
4074 [ + - ]: 864 : int rc = ps_query.step();
4075 [ + + ]: 864 : if (rc == SQLITE_DONE) break;
4076 [ - + ]: 792 : if (rc != SQLITE_ROW)
4077 [ # # # # ]: 0 : throw sqlite_exception(rc, "step");
4078 : :
4079 [ + - ]: 792 : obatched(clog)
4080 [ + - - + : 792 : << ((const char*) sqlite3_column_text(ps_query, 0) ?: (const char*) "NULL")
+ - ]
4081 : : << " "
4082 [ + - + - : 1584 : << (sqlite3_column_text(ps_query, 1) ?: (const unsigned char*) "NULL")
- + + - ]
4083 : 792 : << endl;
4084 : :
4085 [ + - + - : 1584 : set_metric("groom", "statistic",
- + + - +
- + - + -
- + + + -
- - - ]
4086 [ + - ]: 792 : ((const char*) sqlite3_column_text(ps_query, 0) ?: (const char*) "NULL"),
4087 : : (sqlite3_column_double(ps_query, 1)));
4088 : 792 : }
4089 : 72 : }
4090 : :
4091 : :
4092 : : // Do a round of database grooming that might take many minutes to run.
4093 : 72 : void groom()
4094 : : {
4095 [ + - ]: 144 : obatched(clog) << "grooming database" << endl;
4096 : :
4097 : 72 : struct timespec ts_start, ts_end;
4098 : 72 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
4099 : :
4100 : : // scan for files that have disappeared
4101 : 72 : sqlite_ps files (db, "check old files",
4102 : : "select distinct s.mtime, s.file, f.name from "
4103 : : BUILDIDS "_file_mtime_scanned s, " BUILDIDS "_files_v f "
4104 [ + - + - : 144 : "where f.id = s.file");
+ - - - ]
4105 : : // NB: Because _ftime_mtime_scanned can contain both F and
4106 : : // R records for the same file, this query would return duplicates if the
4107 : : // DISTINCT qualifier were not there.
4108 [ + - ]: 72 : files.reset();
4109 : :
4110 : : // DECISION TIME - we enumerate stale fileids/mtimes
4111 [ + - ]: 72 : deque<pair<int64_t,int64_t> > stale_fileid_mtime;
4112 : :
4113 : 72 : time_t time_start = time(NULL);
4114 : 304 : while(1)
4115 : : {
4116 : : // PR28514: limit grooming iteration to O(rescan time), to avoid
4117 : : // slow filesystem tests over many files locking out rescans for
4118 : : // too long.
4119 [ + + - + ]: 188 : if (rescan_s > 0 && (long)time(NULL) > (long)(time_start + rescan_s))
4120 : : {
4121 [ # # # # : 0 : inc_metric("groomed_total", "decision", "aborted");
# # # # #
# # # # #
# # ]
4122 : 0 : break;
4123 : : }
4124 : :
4125 [ + - ]: 188 : if (interrupted) break;
4126 : :
4127 [ + - ]: 188 : int rc = files.step();
4128 [ + + ]: 188 : if (rc != SQLITE_ROW)
4129 : : break;
4130 : :
4131 [ + - ]: 116 : int64_t mtime = sqlite3_column_int64 (files, 0);
4132 [ + - ]: 116 : int64_t fileid = sqlite3_column_int64 (files, 1);
4133 [ + - - + ]: 116 : const char* filename = ((const char*) sqlite3_column_text (files, 2) ?: "");
4134 : 116 : struct stat s;
4135 : 116 : bool regex_file_drop = 0;
4136 : :
4137 [ + + ]: 116 : if (regex_groom)
4138 : : {
4139 [ + - ]: 16 : bool reg_include = !regexec (&file_include_regex, filename, 0, 0, 0);
4140 [ + - ]: 16 : bool reg_exclude = !regexec (&file_exclude_regex, filename, 0, 0, 0);
4141 : 16 : regex_file_drop = !reg_include || reg_exclude; // match logic of scan_source_paths
4142 : : }
4143 : :
4144 : 116 : rc = stat(filename, &s);
4145 [ + + - + ]: 116 : if ( regex_file_drop || rc < 0 || (mtime != (int64_t) s.st_mtime) )
4146 : : {
4147 [ + - ]: 24 : if (verbose > 2)
4148 [ + - + - : 48 : obatched(clog) << "groom: stale file=" << filename << " mtime=" << mtime << endl;
+ - + - +
- ]
4149 [ + - ]: 24 : stale_fileid_mtime.push_back(make_pair(fileid,mtime));
4150 [ + - + - : 48 : inc_metric("groomed_total", "decision", "stale");
+ - + - -
+ - + - -
- - ]
4151 [ + - + - : 48 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
4152 : : }
4153 : : else
4154 [ + - + - : 184 : inc_metric("groomed_total", "decision", "fresh");
+ - + - -
+ - + - -
- - ]
4155 : :
4156 [ + - ]: 116 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
4157 : : break;
4158 : 116 : }
4159 [ + - ]: 72 : files.reset();
4160 : :
4161 : : // ACTION TIME
4162 : :
4163 : : // Now that we know which file/mtime tuples are stale, actually do
4164 : : // the deletion from the database. Doing this during the SELECT
4165 : : // iteration above results in undefined behaviour in sqlite, as per
4166 : : // https://www.sqlite.org/isolation.html
4167 : :
4168 : : // We could shuffle stale_fileid_mtime[] here. It'd let aborted
4169 : : // sequences of nuke operations resume at random locations, instead
4170 : : // of just starting over. But it doesn't matter much either way,
4171 : : // as long as we make progress.
4172 : :
4173 [ + - + - : 144 : sqlite_ps files_del_f_de (db, "nuke f_de", "delete from " BUILDIDS "_f_de where file = ? and mtime = ?");
+ - + - -
- ]
4174 [ + - + - : 144 : sqlite_ps files_del_r_de (db, "nuke r_de", "delete from " BUILDIDS "_r_de where file = ? and mtime = ?");
+ - + - -
- ]
4175 [ + - ]: 72 : sqlite_ps files_del_scan (db, "nuke f_m_s", "delete from " BUILDIDS "_file_mtime_scanned "
4176 [ + - + - : 144 : "where file = ? and mtime = ?");
+ - + - -
- ]
4177 : :
4178 [ + + ]: 96 : while (! stale_fileid_mtime.empty())
4179 : : {
4180 : 24 : auto stale = stale_fileid_mtime.front();
4181 : 24 : stale_fileid_mtime.pop_front();
4182 [ + - + - : 48 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
4183 : :
4184 : : // PR28514: limit grooming iteration to O(rescan time), to avoid
4185 : : // slow nuke_* queries over many files locking out rescans for too
4186 : : // long. We iterate over the files in random() sequence to avoid
4187 : : // partial checks going over the same set.
4188 [ - + - - ]: 24 : if (rescan_s > 0 && (long)time(NULL) > (long)(time_start + rescan_s))
4189 : : {
4190 [ # # # # : 0 : inc_metric("groomed_total", "action", "aborted");
# # # # #
# # # # #
# # ]
4191 : 0 : break;
4192 : : }
4193 : :
4194 [ + - ]: 24 : if (interrupted) break;
4195 : :
4196 : 24 : int64_t fileid = stale.first;
4197 : 24 : int64_t mtime = stale.second;
4198 [ + - + - : 24 : files_del_f_de.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
4199 [ + - + - : 24 : files_del_r_de.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
4200 [ + - + - : 24 : files_del_scan.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
4201 [ + - + - : 48 : inc_metric("groomed_total", "action", "cleaned");
+ - + - -
+ - + - -
- - ]
4202 : :
4203 [ + - ]: 24 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
4204 : : break;
4205 : : }
4206 : 72 : stale_fileid_mtime.clear(); // no need for this any longer
4207 [ + - + - : 144 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
4208 : :
4209 : : // delete buildids with no references in _r_de or _f_de tables;
4210 : : // cascades to _r_sref & _f_s records
4211 [ + - ]: 72 : sqlite_ps buildids_del (db, "nuke orphan buildids",
4212 : : "delete from " BUILDIDS "_buildids "
4213 : : "where not exists (select 1 from " BUILDIDS "_f_de d where " BUILDIDS "_buildids.id = d.buildid) "
4214 [ + - + - : 144 : "and not exists (select 1 from " BUILDIDS "_r_de d where " BUILDIDS "_buildids.id = d.buildid)");
+ - + - -
- ]
4215 [ + - + - ]: 72 : buildids_del.reset().step_ok_done();
4216 : :
4217 [ - + ]: 72 : if (interrupted) return;
4218 : :
4219 : : // NB: "vacuum" is too heavy for even daily runs: it rewrites the entire db, so is done as maxigroom -G
4220 [ + - + - : 144 : sqlite_ps g1 (db, "incremental vacuum", "pragma incremental_vacuum");
+ - + - -
- ]
4221 [ + - + - ]: 72 : g1.reset().step_ok_done();
4222 [ + - + - : 144 : sqlite_ps g2 (db, "optimize", "pragma optimize");
+ - - + -
- ]
4223 [ + - + - ]: 72 : g2.reset().step_ok_done();
4224 [ + - + - : 144 : sqlite_ps g3 (db, "wal checkpoint", "pragma wal_checkpoint=truncate");
+ - + - -
- ]
4225 [ + - + - ]: 72 : g3.reset().step_ok_done();
4226 : :
4227 [ + - ]: 72 : database_stats_report();
4228 : :
4229 [ + - + - ]: 72 : (void) statfs_free_enough_p(db_path, "database"); // report sqlite filesystem size
4230 : :
4231 [ + - ]: 72 : sqlite3_db_release_memory(db); // shrink the process if possible
4232 [ + - ]: 72 : sqlite3_db_release_memory(dbq); // ... for both connections
4233 [ + - ]: 72 : debuginfod_pool_groom(); // and release any debuginfod_client objects we've been holding onto
4234 : : #if HAVE_MALLOC_TRIM
4235 : 72 : malloc_trim(0); // PR31103: release memory allocated for temporary purposes
4236 : : #endif
4237 : :
4238 : : #if 0 /* PR31265: don't jettison cache unnecessarily */
4239 : : fdcache.limit(0); // release the fdcache contents
4240 : : fdcache.limit(fdcache_mbs); // restore status quo parameters
4241 : : #endif
4242 : :
4243 : 72 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
4244 : 72 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
4245 : :
4246 [ + - + - : 144 : obatched(clog) << "groomed database in " << deltas << "s" << endl;
+ - + - ]
4247 : 72 : }
4248 : :
4249 : :
4250 : : static void*
4251 : 72 : thread_main_groom (void* /*arg*/)
4252 : : {
4253 [ + - + - : 144 : set_metric("thread_tid", "role", "groom", tid());
+ - - + -
+ - - -
- ]
4254 [ + - + - : 144 : add_metric("thread_count", "role", "groom", 1);
+ - - + -
+ - - -
- ]
4255 : :
4256 : 72 : time_t last_groom = 0;
4257 : :
4258 : 534 : while (1)
4259 : : {
4260 : 303 : sleep (1);
4261 : 303 : scanq.wait_idle(); // PR25394: block scanners during grooming!
4262 [ + + ]: 303 : if (interrupted) break;
4263 : :
4264 : 231 : time_t now = time(NULL);
4265 : 231 : bool groom_now = false;
4266 [ + + ]: 231 : if (last_groom == 0) // at least one initial groom is documented even for -g0
4267 : 66 : groom_now = true;
4268 [ + + + + ]: 231 : if (groom_s > 0 && (long)now > (long)(last_groom + groom_s))
4269 : 231 : groom_now = true;
4270 [ + + ]: 231 : if (sigusr2 != forced_groom_count)
4271 : : {
4272 : 6 : forced_groom_count = sigusr2;
4273 : 6 : groom_now = true;
4274 : : }
4275 [ + + ]: 231 : if (groom_now)
4276 : : {
4277 [ + - + - : 144 : set_metric("thread_busy", "role", "groom", 1);
+ - - + -
+ - - -
- ]
4278 : 72 : try
4279 : : {
4280 [ + - ]: 72 : groom ();
4281 : : }
4282 [ - - ]: 0 : catch (const sqlite_exception& e)
4283 : : {
4284 [ - - - - : 0 : obatched(cerr) << e.message << endl;
- - ]
4285 : 0 : }
4286 : 72 : last_groom = time(NULL); // NB: now was before grooming
4287 : : // finished a grooming loop
4288 [ + - + - : 144 : inc_metric("thread_work_total", "role", "groom");
+ - - + -
+ - - -
- ]
4289 [ + - + - : 144 : set_metric("thread_busy", "role", "groom", 0);
+ - - + -
+ - - -
- ]
4290 : : }
4291 : :
4292 : 231 : scanq.done_idle();
4293 : 231 : }
4294 : :
4295 : 72 : return 0;
4296 : : }
4297 : :
4298 : :
4299 : : ////////////////////////////////////////////////////////////////////////
4300 : :
4301 : :
4302 : : static void
4303 : 74 : signal_handler (int /* sig */)
4304 : : {
4305 : 74 : interrupted ++;
4306 : :
4307 [ + + ]: 74 : if (db)
4308 : 72 : sqlite3_interrupt (db);
4309 [ + - ]: 74 : if (dbq)
4310 : 74 : sqlite3_interrupt (dbq);
4311 : :
4312 : : // NB: don't do anything else in here
4313 : 74 : }
4314 : :
4315 : : static void
4316 : 60 : sigusr1_handler (int /* sig */)
4317 : : {
4318 : 60 : sigusr1 ++;
4319 : : // NB: don't do anything else in here
4320 : 60 : }
4321 : :
4322 : : static void
4323 : 6 : sigusr2_handler (int /* sig */)
4324 : : {
4325 : 6 : sigusr2 ++;
4326 : : // NB: don't do anything else in here
4327 : 6 : }
4328 : :
4329 : :
4330 : : static void // error logging callback from libmicrohttpd internals
4331 : 0 : error_cb (void *arg, const char *fmt, va_list ap)
4332 : : {
4333 : 0 : (void) arg;
4334 [ # # # # : 0 : inc_metric("error_count","libmicrohttpd",fmt);
# # # # #
# # # #
# ]
4335 : 0 : char errmsg[512];
4336 : 0 : (void) vsnprintf (errmsg, sizeof(errmsg), fmt, ap); // ok if slightly truncated
4337 [ # # ]: 0 : obatched(cerr) << "libmicrohttpd error: " << errmsg; // MHD_DLOG calls already include \n
4338 : 0 : }
4339 : :
4340 : :
4341 : : // A user-defined sqlite function, to score the sharedness of the
4342 : : // prefix of two strings. This is used to compare candidate debuginfo
4343 : : // / source-rpm names, so that the closest match
4344 : : // (directory-topology-wise closest) is found. This is important in
4345 : : // case the same sref (source file name) is in many -debuginfo or
4346 : : // -debugsource RPMs, such as when multiple versions/releases of the
4347 : : // same package are in the database.
4348 : :
4349 : 805 : static void sqlite3_sharedprefix_fn (sqlite3_context* c, int argc, sqlite3_value** argv)
4350 : : {
4351 [ - + ]: 805 : if (argc != 2)
4352 : 0 : sqlite3_result_error(c, "expect 2 string arguments", -1);
4353 [ + - + + ]: 1610 : else if ((sqlite3_value_type(argv[0]) != SQLITE_TEXT) ||
4354 : 805 : (sqlite3_value_type(argv[1]) != SQLITE_TEXT))
4355 : 515 : sqlite3_result_null(c);
4356 : : else
4357 : : {
4358 : 290 : const unsigned char* a = sqlite3_value_text (argv[0]);
4359 : 290 : const unsigned char* b = sqlite3_value_text (argv[1]);
4360 : 290 : int i = 0;
4361 [ + + + - : 35878 : while (*a != '\0' && *b != '\0' && *a++ == *b++)
+ + + + ]
4362 : 35048 : i++;
4363 : 290 : sqlite3_result_int (c, i);
4364 : : }
4365 : 805 : }
4366 : :
4367 : :
4368 : : static unsigned
4369 : 144 : default_concurrency() // guaranteed >= 1
4370 : : {
4371 : : // Prior to PR29975 & PR29976, we'd just use this:
4372 : 144 : unsigned sth = std::thread::hardware_concurrency();
4373 : : // ... but on many-CPU boxes, admins or distros may throttle
4374 : : // resources in such a way that debuginfod would mysteriously fail.
4375 : : // So we reduce the defaults:
4376 : :
4377 : 144 : unsigned aff = 0;
4378 : : #ifdef HAVE_SCHED_GETAFFINITY
4379 : 144 : {
4380 : 144 : int ret;
4381 : 144 : cpu_set_t mask;
4382 : 144 : CPU_ZERO(&mask);
4383 : 144 : ret = sched_getaffinity(0, sizeof(mask), &mask);
4384 [ + - ]: 144 : if (ret == 0)
4385 : 144 : aff = CPU_COUNT(&mask);
4386 : : }
4387 : : #endif
4388 : :
4389 : 144 : unsigned fn = 0;
4390 : : #ifdef HAVE_GETRLIMIT
4391 : 144 : {
4392 : 144 : struct rlimit rlim;
4393 : 144 : int rc = getrlimit(RLIMIT_NOFILE, &rlim);
4394 [ + - ]: 144 : if (rc == 0)
4395 [ + - ]: 288 : fn = max((rlim_t)1, (rlim.rlim_cur - 100) / 4);
4396 : : // at least 2 fds are used by each listener thread etc.
4397 : : // plus a bunch to account for shared libraries and such
4398 : : }
4399 : : #endif
4400 : :
4401 [ - + - + : 144 : unsigned d = min(max(sth, 1U),
- + ]
4402 [ - + ]: 144 : min(max(aff, 1U),
4403 [ - + ]: 144 : max(fn, 1U)));
4404 : 144 : return d;
4405 : : }
4406 : :
4407 : :
4408 : : // 30879: Something to help out in case of an uncaught exception.
4409 : 0 : void my_terminate_handler()
4410 : : {
4411 : : #if defined(__GLIBC__)
4412 : 0 : void *array[40];
4413 : 0 : int size = backtrace (array, 40);
4414 : 0 : backtrace_symbols_fd (array, size, STDERR_FILENO);
4415 : : #endif
4416 : : #if defined(__GLIBCXX__) || defined(__GLIBCPP__)
4417 : 0 : __gnu_cxx::__verbose_terminate_handler();
4418 : : #endif
4419 : 0 : abort();
4420 : : }
4421 : :
4422 : :
4423 : : int
4424 : 74 : main (int argc, char *argv[])
4425 : : {
4426 : 74 : (void) setlocale (LC_ALL, "");
4427 : 74 : (void) bindtextdomain (PACKAGE_TARNAME, LOCALEDIR);
4428 : 74 : (void) textdomain (PACKAGE_TARNAME);
4429 : :
4430 : 74 : std::set_terminate(& my_terminate_handler);
4431 : :
4432 : : /* Tell the library which version we are expecting. */
4433 : 74 : elf_version (EV_CURRENT);
4434 : :
4435 [ + - - + ]: 148 : tmpdir = string(getenv("TMPDIR") ?: "/tmp");
4436 : :
4437 : : /* Set computed default values. */
4438 [ - + + - : 74 : db_path = string(getenv("HOME") ?: "/") + string("/.debuginfod.sqlite"); /* XDG? */
+ - - + -
+ + - -
- ]
4439 : 74 : int rc = regcomp (& file_include_regex, ".*", REG_EXTENDED|REG_NOSUB); // match everything
4440 [ - + ]: 74 : if (rc != 0)
4441 : 0 : error (EXIT_FAILURE, 0, "regcomp failure: %d", rc);
4442 : 74 : rc = regcomp (& file_exclude_regex, "^$", REG_EXTENDED|REG_NOSUB); // match nothing
4443 [ - + ]: 74 : if (rc != 0)
4444 : 0 : error (EXIT_FAILURE, 0, "regcomp failure: %d", rc);
4445 : :
4446 : : // default parameters for fdcache are computed from system stats
4447 : 74 : struct statfs sfs;
4448 : 74 : rc = statfs(tmpdir.c_str(), &sfs);
4449 [ - + ]: 74 : if (rc < 0)
4450 : 0 : fdcache_mbs = 1024; // 1 gigabyte
4451 : : else
4452 : 74 : fdcache_mbs = sfs.f_bavail * sfs.f_bsize / 1024 / 1024 / 4; // 25% of free space
4453 : 74 : fdcache_mintmp = 25; // emergency flush at 25% remaining (75% full)
4454 : 74 : fdcache_prefetch = 64; // guesstimate storage is this much less costly than re-decompression
4455 : :
4456 : : /* Parse and process arguments. */
4457 : 74 : int remaining;
4458 : 74 : (void) argp_parse (&argp, argc, argv, ARGP_IN_ORDER, &remaining, NULL);
4459 [ - + ]: 74 : if (remaining != argc)
4460 : 0 : error (EXIT_FAILURE, 0,
4461 : 0 : "unexpected argument: %s", argv[remaining]);
4462 : :
4463 [ + + + + : 74 : if (scan_archives.size()==0 && !scan_files && source_paths.size()>0)
- + ]
4464 [ # # ]: 0 : obatched(clog) << "warning: without -F -R -U -Z, ignoring PATHs" << endl;
4465 : :
4466 : 74 : fdcache.limit(fdcache_mbs);
4467 : :
4468 : 74 : (void) signal (SIGPIPE, SIG_IGN); // microhttpd can generate it incidentally, ignore
4469 : 74 : (void) signal (SIGINT, signal_handler); // ^C
4470 : 74 : (void) signal (SIGHUP, signal_handler); // EOF
4471 : 74 : (void) signal (SIGTERM, signal_handler); // systemd
4472 : 74 : (void) signal (SIGUSR1, sigusr1_handler); // end-user
4473 : 74 : (void) signal (SIGUSR2, sigusr2_handler); // end-user
4474 : :
4475 : : /* Get database ready. */
4476 [ + + ]: 74 : if (! passive_p)
4477 : : {
4478 : 72 : rc = sqlite3_open_v2 (db_path.c_str(), &db, (SQLITE_OPEN_READWRITE
4479 : : |SQLITE_OPEN_URI
4480 : : |SQLITE_OPEN_PRIVATECACHE
4481 : : |SQLITE_OPEN_CREATE
4482 : : |SQLITE_OPEN_FULLMUTEX), /* thread-safe */
4483 : : NULL);
4484 [ - + ]: 72 : if (rc == SQLITE_CORRUPT)
4485 : : {
4486 : 0 : (void) unlink (db_path.c_str());
4487 : 0 : error (EXIT_FAILURE, 0,
4488 : : "cannot open %s, deleted database: %s", db_path.c_str(), sqlite3_errmsg(db));
4489 : : }
4490 [ - + ]: 72 : else if (rc)
4491 : : {
4492 : 0 : error (EXIT_FAILURE, 0,
4493 : : "cannot open %s, consider deleting database: %s", db_path.c_str(), sqlite3_errmsg(db));
4494 : : }
4495 : : }
4496 : :
4497 : : // open the readonly query variant
4498 : : // NB: PRIVATECACHE allows web queries to operate in parallel with
4499 : : // much other grooming/scanning operation.
4500 : 74 : rc = sqlite3_open_v2 (db_path.c_str(), &dbq, (SQLITE_OPEN_READONLY
4501 : : |SQLITE_OPEN_URI
4502 : : |SQLITE_OPEN_PRIVATECACHE
4503 : : |SQLITE_OPEN_FULLMUTEX), /* thread-safe */
4504 : : NULL);
4505 [ - + ]: 74 : if (rc)
4506 : : {
4507 : 0 : error (EXIT_FAILURE, 0,
4508 : : "cannot open %s, consider deleting database: %s", db_path.c_str(), sqlite3_errmsg(dbq));
4509 : : }
4510 : :
4511 : :
4512 [ + - ]: 148 : obatched(clog) << "opened database " << db_path
4513 [ + + + - : 76 : << (db?" rw":"") << (dbq?" ro":"") << endl;
- + + - +
- ]
4514 [ + - + - ]: 148 : obatched(clog) << "sqlite version " << sqlite3_version << endl;
4515 [ + + + - : 220 : obatched(clog) << "service mode " << (passive_p ? "passive":"active") << endl;
+ - ]
4516 : :
4517 : : // add special string-prefix-similarity function used in rpm sref/sdef resolution
4518 : 74 : rc = sqlite3_create_function(dbq, "sharedprefix", 2, SQLITE_UTF8, NULL,
4519 : : & sqlite3_sharedprefix_fn, NULL, NULL);
4520 [ - + ]: 74 : if (rc != SQLITE_OK)
4521 : 0 : error (EXIT_FAILURE, 0,
4522 : : "cannot create sharedprefix function: %s", sqlite3_errmsg(dbq));
4523 : :
4524 [ + + ]: 74 : if (! passive_p)
4525 : : {
4526 [ + + ]: 72 : if (verbose > 3)
4527 [ + - + - ]: 92 : obatched(clog) << "ddl: " << DEBUGINFOD_SQLITE_DDL << endl;
4528 : 72 : rc = sqlite3_exec (db, DEBUGINFOD_SQLITE_DDL, NULL, NULL, NULL);
4529 [ - + ]: 72 : if (rc != SQLITE_OK)
4530 : : {
4531 : 0 : error (EXIT_FAILURE, 0,
4532 : : "cannot run database schema ddl: %s", sqlite3_errmsg(db));
4533 : : }
4534 : : }
4535 : :
4536 [ + - + - : 148 : obatched(clog) << "libmicrohttpd version " << MHD_get_version() << endl;
+ - ]
4537 : :
4538 : : /* If '-C' wasn't given or was given with no arg, pick a reasonable default
4539 : : for the number of worker threads. */
4540 [ + + ]: 74 : if (connection_pool == 0)
4541 : 70 : connection_pool = default_concurrency();
4542 : :
4543 : : /* Note that MHD_USE_EPOLL and MHD_USE_THREAD_PER_CONNECTION don't
4544 : : work together. */
4545 : 74 : unsigned int use_epoll = 0;
4546 : : #if MHD_VERSION >= 0x00095100
4547 : 74 : use_epoll = MHD_USE_EPOLL;
4548 : : #endif
4549 : :
4550 : 74 : unsigned int mhd_flags = (
4551 : : #if MHD_VERSION >= 0x00095300
4552 : : MHD_USE_INTERNAL_POLLING_THREAD
4553 : : #else
4554 : : MHD_USE_SELECT_INTERNALLY
4555 : : #endif
4556 : : | MHD_USE_DUAL_STACK
4557 : : | use_epoll
4558 : : #if MHD_VERSION >= 0x00095200
4559 : : | MHD_USE_ITC
4560 : : #endif
4561 : : | MHD_USE_DEBUG); /* report errors to stderr */
4562 : :
4563 : : // Start httpd server threads. Use a single dual-homed pool.
4564 : 74 : MHD_Daemon *d46 = MHD_start_daemon (mhd_flags, http_port,
4565 : : NULL, NULL, /* default accept policy */
4566 : : handler_cb, NULL, /* handler callback */
4567 : : MHD_OPTION_EXTERNAL_LOGGER,
4568 : : error_cb, NULL,
4569 : : MHD_OPTION_THREAD_POOL_SIZE,
4570 : : (int)connection_pool,
4571 : : MHD_OPTION_END);
4572 : :
4573 : 74 : MHD_Daemon *d4 = NULL;
4574 [ - + ]: 74 : if (d46 == NULL)
4575 : : {
4576 : : // Cannot use dual_stack, use ipv4 only
4577 : 0 : mhd_flags &= ~(MHD_USE_DUAL_STACK);
4578 [ # # ]: 0 : d4 = MHD_start_daemon (mhd_flags, http_port,
4579 : : NULL, NULL, /* default accept policy */
4580 : : handler_cb, NULL, /* handler callback */
4581 : : MHD_OPTION_EXTERNAL_LOGGER,
4582 : : error_cb, NULL,
4583 : : (connection_pool
4584 : : ? MHD_OPTION_THREAD_POOL_SIZE
4585 : : : MHD_OPTION_END),
4586 : : (connection_pool
4587 : : ? (int)connection_pool
4588 : : : MHD_OPTION_END),
4589 : : MHD_OPTION_END);
4590 [ # # ]: 0 : if (d4 == NULL)
4591 : : {
4592 : 0 : sqlite3 *database = db;
4593 : 0 : sqlite3 *databaseq = dbq;
4594 : 0 : db = dbq = 0; // for signal_handler not to freak
4595 : 0 : sqlite3_close (databaseq);
4596 : 0 : sqlite3_close (database);
4597 : 0 : error (EXIT_FAILURE, 0, "cannot start http server at port %d",
4598 : : http_port);
4599 : : }
4600 : :
4601 : : }
4602 : 74 : obatched(clog) << "started http server on"
4603 : : << (d4 != NULL ? " IPv4 " : " IPv4 IPv6 ")
4604 [ + - + - : 148 : << "port=" << http_port << endl;
+ - + - +
- ]
4605 : :
4606 : : // add maxigroom sql if -G given
4607 [ - + ]: 74 : if (maxigroom)
4608 : : {
4609 [ # # ]: 0 : obatched(clog) << "maxigrooming database, please wait." << endl;
4610 [ # # ]: 0 : extra_ddl.push_back("create index if not exists " BUILDIDS "_r_sref_arc on " BUILDIDS "_r_sref(artifactsrc);");
4611 [ # # ]: 0 : extra_ddl.push_back("delete from " BUILDIDS "_r_sdef where not exists (select 1 from " BUILDIDS "_r_sref b where " BUILDIDS "_r_sdef.content = b.artifactsrc);");
4612 [ # # ]: 0 : extra_ddl.push_back("drop index if exists " BUILDIDS "_r_sref_arc;");
4613 : :
4614 : : // NB: we don't maxigroom the _files interning table. It'd require a temp index on all the
4615 : : // tables that have file foreign-keys, which is a lot.
4616 : :
4617 : : // NB: with =delete, may take up 3x disk space total during vacuum process
4618 : : // vs. =off (only 2x but may corrupt database if program dies mid-vacuum)
4619 : : // vs. =wal (>3x observed, but safe)
4620 [ # # ]: 0 : extra_ddl.push_back("pragma journal_mode=delete;");
4621 [ # # ]: 0 : extra_ddl.push_back("vacuum;");
4622 [ # # ]: 0 : extra_ddl.push_back("pragma journal_mode=wal;");
4623 : : }
4624 : :
4625 : : // run extra -D sql if given
4626 [ + + ]: 74 : if (! passive_p)
4627 [ - + ]: 72 : for (auto&& i: extra_ddl)
4628 : : {
4629 [ # # ]: 0 : if (verbose > 1)
4630 [ # # # # ]: 0 : obatched(clog) << "extra ddl:\n" << i << endl;
4631 : 0 : rc = sqlite3_exec (db, i.c_str(), NULL, NULL, NULL);
4632 [ # # # # ]: 0 : if (rc != SQLITE_OK && rc != SQLITE_DONE && rc != SQLITE_ROW)
4633 : 0 : error (0, 0,
4634 : : "warning: cannot run database extra ddl %s: %s", i.c_str(), sqlite3_errmsg(db));
4635 : :
4636 [ # # ]: 0 : if (maxigroom)
4637 [ # # ]: 0 : obatched(clog) << "maxigroomed database" << endl;
4638 : : }
4639 : :
4640 [ + + ]: 74 : if (! passive_p)
4641 [ + - + - ]: 144 : obatched(clog) << "search concurrency " << concurrency << endl;
4642 : 74 : obatched(clog) << "webapi connection pool " << connection_pool
4643 [ + - - + : 74 : << (connection_pool ? "" : " (unlimited)") << endl;
+ - + - ]
4644 [ + + ]: 74 : if (! passive_p) {
4645 [ + - + - ]: 144 : obatched(clog) << "rescan time " << rescan_s << endl;
4646 [ + - + - ]: 144 : obatched(clog) << "scan checkpoint " << scan_checkpoint << endl;
4647 : : }
4648 [ + - + - ]: 148 : obatched(clog) << "fdcache mbs " << fdcache_mbs << endl;
4649 [ + - + - ]: 148 : obatched(clog) << "fdcache prefetch " << fdcache_prefetch << endl;
4650 [ + - + - ]: 148 : obatched(clog) << "fdcache tmpdir " << tmpdir << endl;
4651 [ + - + - ]: 148 : obatched(clog) << "fdcache tmpdir min% " << fdcache_mintmp << endl;
4652 [ + + ]: 74 : if (! passive_p)
4653 [ + - + - ]: 144 : obatched(clog) << "groom time " << groom_s << endl;
4654 [ + - + - ]: 148 : obatched(clog) << "forwarded ttl limit " << forwarded_ttl_limit << endl;
4655 : :
4656 [ + + ]: 74 : if (scan_archives.size()>0)
4657 : : {
4658 : 50 : obatched ob(clog);
4659 [ + - ]: 50 : auto& o = ob << "accepting archive types ";
4660 [ + + ]: 152 : for (auto&& arch : scan_archives)
4661 [ + - + - : 102 : o << arch.first << "(" << arch.second << ") ";
+ - + - ]
4662 [ + - ]: 50 : o << endl;
4663 : 50 : }
4664 : 74 : const char* du = getenv(DEBUGINFOD_URLS_ENV_VAR);
4665 [ + + + + ]: 74 : if (du && du[0] != '\0') // set to non-empty string?
4666 [ + - + - ]: 28 : obatched(clog) << "upstream debuginfod servers: " << du << endl;
4667 : :
4668 [ + + ]: 74 : vector<pthread_t> all_threads;
4669 : :
4670 [ + + ]: 74 : if (! passive_p)
4671 : : {
4672 : 72 : pthread_t pt;
4673 : 72 : rc = pthread_create (& pt, NULL, thread_main_groom, NULL);
4674 [ - + ]: 72 : if (rc)
4675 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to groom database\n");
4676 : : else
4677 : : {
4678 : : #ifdef HAVE_PTHREAD_SETNAME_NP
4679 : 72 : (void) pthread_setname_np (pt, "groom");
4680 : : #endif
4681 [ + - ]: 72 : all_threads.push_back(pt);
4682 : : }
4683 : :
4684 [ + + + + ]: 72 : if (scan_files || scan_archives.size() > 0)
4685 : : {
4686 [ + - ]: 66 : if (scan_checkpoint > 0)
4687 [ + - ]: 66 : scan_barrier = new sqlite_checkpoint_pb(concurrency, (unsigned) scan_checkpoint);
4688 : :
4689 : 66 : rc = pthread_create (& pt, NULL, thread_main_fts_source_paths, NULL);
4690 [ - + ]: 66 : if (rc)
4691 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to traverse source paths\n");
4692 : : #ifdef HAVE_PTHREAD_SETNAME_NP
4693 : 66 : (void) pthread_setname_np (pt, "traverse");
4694 : : #endif
4695 [ + - ]: 66 : all_threads.push_back(pt);
4696 : :
4697 [ + + ]: 330 : for (unsigned i=0; i<concurrency; i++)
4698 : : {
4699 : 264 : rc = pthread_create (& pt, NULL, thread_main_scanner, NULL);
4700 [ - + ]: 264 : if (rc)
4701 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to scan source files / archives\n");
4702 : : #ifdef HAVE_PTHREAD_SETNAME_NP
4703 : 264 : (void) pthread_setname_np (pt, "scan");
4704 : : #endif
4705 [ + - ]: 264 : all_threads.push_back(pt);
4706 : : }
4707 : : }
4708 : : }
4709 : :
4710 : : /* Trivial main loop! */
4711 [ + - + - ]: 74 : set_metric("ready", 1);
4712 [ + + ]: 214 : while (! interrupted)
4713 [ + - ]: 140 : pause ();
4714 [ + - ]: 74 : scanq.nuke(); // wake up any remaining scanq-related threads, let them die
4715 [ + + + - ]: 74 : if (scan_barrier) scan_barrier->nuke(); // ... in case they're stuck in a barrier
4716 [ + - + - ]: 74 : set_metric("ready", 0);
4717 : :
4718 [ + - ]: 74 : if (verbose)
4719 [ + - + - : 148 : obatched(clog) << "stopping" << endl;
- - ]
4720 : :
4721 : : /* Join all our threads. */
4722 [ + + ]: 476 : for (auto&& it : all_threads)
4723 [ + - ]: 402 : pthread_join (it, NULL);
4724 : :
4725 : : /* Stop all the web service threads. */
4726 [ + - + - ]: 74 : if (d46) MHD_stop_daemon (d46);
4727 [ - + - - ]: 74 : if (d4) MHD_stop_daemon (d4);
4728 : :
4729 [ + + ]: 74 : if (! passive_p)
4730 : : {
4731 : : /* With all threads known dead, we can clean up the global resources. */
4732 [ + - ]: 72 : rc = sqlite3_exec (db, DEBUGINFOD_SQLITE_CLEANUP_DDL, NULL, NULL, NULL);
4733 [ - + ]: 72 : if (rc != SQLITE_OK)
4734 : : {
4735 [ # # # # ]: 0 : error (0, 0,
4736 : : "warning: cannot run database cleanup ddl: %s", sqlite3_errmsg(db));
4737 : : }
4738 : : }
4739 : :
4740 [ + - ]: 74 : debuginfod_pool_groom ();
4741 [ + + ]: 74 : delete scan_barrier;
4742 : :
4743 : : // NB: no problem with unconditional free here - an earlier failed regcomp would exit program
4744 [ + - ]: 74 : (void) regfree (& file_include_regex);
4745 [ + - ]: 74 : (void) regfree (& file_exclude_regex);
4746 : :
4747 : 74 : sqlite3 *database = db;
4748 : 74 : sqlite3 *databaseq = dbq;
4749 : 74 : db = dbq = 0; // for signal_handler not to freak
4750 [ + - ]: 74 : (void) sqlite3_close (databaseq);
4751 [ + + ]: 74 : if (! passive_p)
4752 [ + - ]: 72 : (void) sqlite3_close (database);
4753 : :
4754 [ + + ]: 74 : return 0;
4755 : 74 : }
|
