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 : : #ifdef USE_LZMA
67 : : #include <lzma.h>
68 : : #endif
69 : :
70 : : #include <unistd.h>
71 : : #include <stdlib.h>
72 : : #include <locale.h>
73 : : #include <pthread.h>
74 : : #include <signal.h>
75 : : #include <sys/stat.h>
76 : : #include <sys/time.h>
77 : : #include <sys/vfs.h>
78 : : #include <unistd.h>
79 : : #include <fcntl.h>
80 : : #include <netdb.h>
81 : : #include <math.h>
82 : : #include <float.h>
83 : : #include <fnmatch.h>
84 : :
85 : :
86 : : /* If fts.h is included before config.h, its indirect inclusions may not
87 : : give us the right LFS aliases of these functions, so map them manually. */
88 : : #ifdef BAD_FTS
89 : : #ifdef _FILE_OFFSET_BITS
90 : : #define open open64
91 : : #define fopen fopen64
92 : : #endif
93 : : #else
94 : : #include <sys/types.h>
95 : : #include <fts.h>
96 : : #endif
97 : :
98 : : #include <cstring>
99 : : #include <vector>
100 : : #include <set>
101 : : #include <unordered_set>
102 : : #include <map>
103 : : #include <string>
104 : : #include <iostream>
105 : : #include <iomanip>
106 : : #include <ostream>
107 : : #include <sstream>
108 : : #include <mutex>
109 : : #include <deque>
110 : : #include <condition_variable>
111 : : #include <exception>
112 : : #include <thread>
113 : : // #include <regex> // on rhel7 gcc 4.8, not competent
114 : : #include <regex.h>
115 : : // #include <algorithm>
116 : : using namespace std;
117 : :
118 : : #include <gelf.h>
119 : : #include <libdwelf.h>
120 : :
121 : : #include <microhttpd.h>
122 : :
123 : : #if MHD_VERSION >= 0x00097002
124 : : // libmicrohttpd 0.9.71 broke API
125 : : #define MHD_RESULT enum MHD_Result
126 : : #else
127 : : #define MHD_RESULT int
128 : : #endif
129 : :
130 : : #ifdef ENABLE_IMA_VERIFICATION
131 : : #include <rpm/rpmlib.h>
132 : : #include <rpm/rpmfi.h>
133 : : #include <rpm/header.h>
134 : : #include <glob.h>
135 : : #endif
136 : :
137 : : #include <curl/curl.h>
138 : : #include <archive.h>
139 : : #include <archive_entry.h>
140 : : #include <sqlite3.h>
141 : :
142 : : #ifdef __linux__
143 : : #include <sys/syscall.h>
144 : : #endif
145 : :
146 : : #ifdef __linux__
147 : : #define tid() syscall(SYS_gettid)
148 : : #else
149 : : #define tid() pthread_self()
150 : : #endif
151 : :
152 : : extern "C" {
153 : : #include "printversion.h"
154 : : #include "system.h"
155 : : }
156 : : #include <json-c/json.h>
157 : :
158 : :
159 : : inline bool
160 : 34844 : string_endswith(const string& haystack, const string& needle)
161 : : {
162 [ + + ]: 34844 : return (haystack.size() >= needle.size() &&
163 : 33773 : equal(haystack.end()-needle.size(), haystack.end(),
164 : 34844 : needle.begin()));
165 : : }
166 : :
167 : :
168 : : // Roll this identifier for every sqlite schema incompatibility.
169 : : #define BUILDIDS "buildids10"
170 : :
171 : : #if SQLITE_VERSION_NUMBER >= 3008000
172 : : #define WITHOUT_ROWID "without rowid"
173 : : #else
174 : : #define WITHOUT_ROWID ""
175 : : #endif
176 : :
177 : : static const char DEBUGINFOD_SQLITE_DDL[] =
178 : : "pragma foreign_keys = on;\n"
179 : : "pragma synchronous = 0;\n" // disable fsync()s - this cache is disposable across a machine crash
180 : : "pragma journal_mode = wal;\n" // https://sqlite.org/wal.html
181 : : "pragma wal_checkpoint = truncate;\n" // clean out any preexisting wal file
182 : : "pragma journal_size_limit = 0;\n" // limit steady state file (between grooming, which also =truncate's)
183 : : "pragma auto_vacuum = incremental;\n" // https://sqlite.org/pragma.html
184 : : "pragma busy_timeout = 1000;\n" // https://sqlite.org/pragma.html
185 : : // NB: all these are overridable with -D option
186 : :
187 : : // Normalization table for interning file names
188 : : "create table if not exists " BUILDIDS "_fileparts (\n"
189 : : " id integer primary key not null,\n"
190 : : " name text unique not null\n"
191 : : " );\n"
192 : : "create table if not exists " BUILDIDS "_files (\n"
193 : : " id integer primary key not null,\n"
194 : : " dirname integer not null,\n"
195 : : " basename integer not null,\n"
196 : : " unique (dirname, basename),\n"
197 : : " foreign key (dirname) references " BUILDIDS "_fileparts(id) on delete cascade,\n"
198 : : " foreign key (basename) references " BUILDIDS "_fileparts(id) on delete cascade\n"
199 : : " );\n"
200 : : "create view if not exists " BUILDIDS "_files_v as\n" // a
201 : : " select f.id, n1.name || '/' || n2.name as name\n"
202 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts n1, " BUILDIDS "_fileparts n2\n"
203 : : " where f.dirname = n1.id and f.basename = n2.id;\n"
204 : :
205 : : // Normalization table for interning buildids
206 : : "create table if not exists " BUILDIDS "_buildids (\n"
207 : : " id integer primary key not null,\n"
208 : : " hex text unique not null);\n"
209 : : // Track the completion of scanning of a given file & sourcetype at given time
210 : : "create table if not exists " BUILDIDS "_file_mtime_scanned (\n"
211 : : " mtime integer not null,\n"
212 : : " file integer not null,\n"
213 : : " size integer not null,\n" // in bytes
214 : : " sourcetype text(1) not null\n"
215 : : " check (sourcetype IN ('F', 'R')),\n"
216 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
217 : : " primary key (file, mtime, sourcetype)\n"
218 : : " ) " WITHOUT_ROWID ";\n"
219 : : "create table if not exists " BUILDIDS "_f_de (\n"
220 : : " buildid integer not null,\n"
221 : : " debuginfo_p integer not null,\n"
222 : : " executable_p integer not null,\n"
223 : : " file integer not null,\n"
224 : : " mtime integer not null,\n"
225 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
226 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
227 : : " primary key (buildid, file, mtime)\n"
228 : : " ) " WITHOUT_ROWID ";\n"
229 : : // Index for faster delete by file identifier and metadata searches
230 : : "create index if not exists " BUILDIDS "_f_de_idx on " BUILDIDS "_f_de (file, mtime);\n"
231 : : "create table if not exists " BUILDIDS "_f_s (\n"
232 : : " buildid integer not null,\n"
233 : : " artifactsrc integer not null,\n"
234 : : " file integer not null,\n" // NB: not necessarily entered into _mtime_scanned
235 : : " mtime integer not null,\n"
236 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
237 : : " foreign key (artifactsrc) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
238 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
239 : : " primary key (buildid, artifactsrc, file, mtime)\n"
240 : : " ) " WITHOUT_ROWID ";\n"
241 : : "create table if not exists " BUILDIDS "_r_de (\n"
242 : : " buildid integer not null,\n"
243 : : " debuginfo_p integer not null,\n"
244 : : " executable_p integer not null,\n"
245 : : " file integer not null,\n"
246 : : " mtime integer not null,\n"
247 : : " content integer not null,\n"
248 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
249 : : " foreign key (content) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
250 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
251 : : " primary key (buildid, debuginfo_p, executable_p, file, content, mtime)\n"
252 : : " ) " WITHOUT_ROWID ";\n"
253 : : // Index for faster delete by archive file identifier
254 : : "create index if not exists " BUILDIDS "_r_de_idx on " BUILDIDS "_r_de (file, mtime);\n"
255 : : // Index for metadata searches
256 : : "create index if not exists " BUILDIDS "_r_de_idx2 on " BUILDIDS "_r_de (content);\n"
257 : : "create table if not exists " BUILDIDS "_r_sref (\n" // outgoing dwarf sourcefile references from rpm
258 : : " buildid integer not null,\n"
259 : : " artifactsrc integer not null,\n"
260 : : " foreign key (artifactsrc) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
261 : : " foreign key (buildid) references " BUILDIDS "_buildids(id) on update cascade on delete cascade,\n"
262 : : " primary key (buildid, artifactsrc)\n"
263 : : " ) " WITHOUT_ROWID ";\n"
264 : : "create table if not exists " BUILDIDS "_r_sdef (\n" // rpm contents that may satisfy sref
265 : : " file integer not null,\n"
266 : : " mtime integer not null,\n"
267 : : " content integer not null,\n"
268 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
269 : : " foreign key (content) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
270 : : " primary key (content, file, mtime)\n"
271 : : " ) " WITHOUT_ROWID ";\n"
272 : : "create table if not exists " BUILDIDS "_r_seekable (\n" // seekable rpm contents
273 : : " file integer not null,\n"
274 : : " content integer not null,\n"
275 : : " type text not null,\n"
276 : : " size integer not null,\n"
277 : : " offset integer not null,\n"
278 : : " mtime integer not null,\n"
279 : : " foreign key (file) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
280 : : " foreign key (content) references " BUILDIDS "_files(id) on update cascade on delete cascade,\n"
281 : : " primary key (file, content)\n"
282 : : " ) " WITHOUT_ROWID ";\n"
283 : : // create views to glue together some of the above tables, for webapi D queries
284 : : // NB: _query_d2 and _query_e2 were added to replace _query_d and _query_e
285 : : // without updating BUILDIDS. They can be renamed back the next time BUILDIDS
286 : : // is updated.
287 : : "create view if not exists " BUILDIDS "_query_d2 as \n"
288 : : "select\n"
289 : : " b.hex as buildid, 'F' as sourcetype, n.file as id0, f0.name as source0, n.mtime as mtime, null as id1, null as source1\n"
290 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_f_de n\n"
291 : : " where b.id = n.buildid and f0.id = n.file and n.debuginfo_p = 1\n"
292 : : "union all select\n"
293 : : " b.hex as buildid, 'R' as sourcetype, n.file as id0, f0.name as source0, n.mtime as mtime, n.content as id1, f1.name as source1\n"
294 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_r_de n\n"
295 : : " where b.id = n.buildid and f0.id = n.file and f1.id = n.content and n.debuginfo_p = 1\n"
296 : : ";"
297 : : // ... and for E queries
298 : : "create view if not exists " BUILDIDS "_query_e2 as \n"
299 : : "select\n"
300 : : " b.hex as buildid, 'F' as sourcetype, n.file as id0, f0.name as source0, n.mtime as mtime, null as id1, null as source1\n"
301 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_f_de n\n"
302 : : " where b.id = n.buildid and f0.id = n.file and n.executable_p = 1\n"
303 : : "union all select\n"
304 : : " b.hex as buildid, 'R' as sourcetype, n.file as id0, f0.name as source0, n.mtime as mtime, n.content as id1, f1.name as source1\n"
305 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_r_de n\n"
306 : : " where b.id = n.buildid and f0.id = n.file and f1.id = n.content and n.executable_p = 1\n"
307 : : ";"
308 : : // ... and for S queries
309 : : "create view if not exists " BUILDIDS "_query_s as \n"
310 : : "select\n"
311 : : " 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"
312 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v fs, " BUILDIDS "_f_s n\n"
313 : : " where b.id = n.buildid and f0.id = n.file and fs.id = n.artifactsrc\n"
314 : : "union all select\n"
315 : : " 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"
316 : : " from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f0, " BUILDIDS "_files_v f1, " BUILDIDS "_files_v fsref, "
317 : : " " BUILDIDS "_r_sdef sd, " BUILDIDS "_r_sref sr, " BUILDIDS "_r_de sde\n"
318 : : " where b.id = sr.buildid and f0.id = sd.file and fsref.id = sde.file and f1.id = sd.content\n"
319 : : " and sr.artifactsrc = sd.content and sde.buildid = sr.buildid\n"
320 : : ";"
321 : : // and for startup overview counts
322 : : "drop view if exists " BUILDIDS "_stats;\n"
323 : : "create view if not exists " BUILDIDS "_stats as\n"
324 : : " select 'file d/e' as label,count(*) as quantity from " BUILDIDS "_f_de\n"
325 : : "union all select 'file s',count(*) from " BUILDIDS "_f_s\n"
326 : : "union all select 'archive d/e',count(*) from " BUILDIDS "_r_de\n"
327 : : "union all select 'archive sref',count(*) from " BUILDIDS "_r_sref\n"
328 : : "union all select 'archive sdef',count(*) from " BUILDIDS "_r_sdef\n"
329 : : "union all select 'buildids',count(*) from " BUILDIDS "_buildids\n"
330 : : "union all select 'filenames',count(*) from " BUILDIDS "_files\n"
331 : : "union all select 'fileparts',count(*) from " BUILDIDS "_fileparts\n"
332 : : "union all select 'files scanned (#)',count(*) from " BUILDIDS "_file_mtime_scanned\n"
333 : : "union all select 'files scanned (mb)',coalesce(sum(size)/1024/1024,0) from " BUILDIDS "_file_mtime_scanned\n"
334 : : #if SQLITE_VERSION_NUMBER >= 3016000
335 : : "union all select 'index db size (mb)',page_count*page_size/1024/1024 as size FROM pragma_page_count(), pragma_page_size()\n"
336 : : #endif
337 : : ";\n"
338 : :
339 : : // schema change history & garbage collection
340 : : //
341 : : // XXX: we could have migration queries here to bring prior-schema
342 : : // data over instead of just dropping it. But that could incur
343 : : // doubled storage costs.
344 : : //
345 : : // buildids10: split the _files table into _parts
346 : : "" // <<< we are here
347 : : // buildids9: widen the mtime_scanned table
348 : : "DROP VIEW IF EXISTS buildids9_stats;\n"
349 : : "DROP INDEX IF EXISTS buildids9_r_de_idx;\n"
350 : : "DROP INDEX IF EXISTS buildids9_f_de_idx;\n"
351 : : "DROP VIEW IF EXISTS buildids9_query_s;\n"
352 : : "DROP VIEW IF EXISTS buildids9_query_e;\n"
353 : : "DROP VIEW IF EXISTS buildids9_query_d;\n"
354 : : "DROP TABLE IF EXISTS buildids9_r_sdef;\n"
355 : : "DROP TABLE IF EXISTS buildids9_r_sref;\n"
356 : : "DROP TABLE IF EXISTS buildids9_r_de;\n"
357 : : "DROP TABLE IF EXISTS buildids9_f_s;\n"
358 : : "DROP TABLE IF EXISTS buildids9_f_de;\n"
359 : : "DROP TABLE IF EXISTS buildids9_file_mtime_scanned;\n"
360 : : "DROP TABLE IF EXISTS buildids9_buildids;\n"
361 : : "DROP TABLE IF EXISTS buildids9_files;\n"
362 : : // buildids8: slim the sref table
363 : : "drop table if exists buildids8_f_de;\n"
364 : : "drop table if exists buildids8_f_s;\n"
365 : : "drop table if exists buildids8_r_de;\n"
366 : : "drop table if exists buildids8_r_sref;\n"
367 : : "drop table if exists buildids8_r_sdef;\n"
368 : : "drop table if exists buildids8_file_mtime_scanned;\n"
369 : : "drop table if exists buildids8_files;\n"
370 : : "drop table if exists buildids8_buildids;\n"
371 : : // buildids7: separate _norm table into dense subtype tables
372 : : "drop table if exists buildids7_f_de;\n"
373 : : "drop table if exists buildids7_f_s;\n"
374 : : "drop table if exists buildids7_r_de;\n"
375 : : "drop table if exists buildids7_r_sref;\n"
376 : : "drop table if exists buildids7_r_sdef;\n"
377 : : "drop table if exists buildids7_file_mtime_scanned;\n"
378 : : "drop table if exists buildids7_files;\n"
379 : : "drop table if exists buildids7_buildids;\n"
380 : : // buildids6: drop bolo/rfolo again, represent sources / rpmcontents in main table
381 : : "drop table if exists buildids6_norm;\n"
382 : : "drop table if exists buildids6_files;\n"
383 : : "drop table if exists buildids6_buildids;\n"
384 : : "drop view if exists buildids6;\n"
385 : : // buildids5: redefine srcfile1 column to be '.'-less (for rpms)
386 : : "drop table if exists buildids5_norm;\n"
387 : : "drop table if exists buildids5_files;\n"
388 : : "drop table if exists buildids5_buildids;\n"
389 : : "drop table if exists buildids5_bolo;\n"
390 : : "drop table if exists buildids5_rfolo;\n"
391 : : "drop view if exists buildids5;\n"
392 : : // buildids4: introduce rpmfile RFOLO
393 : : "drop table if exists buildids4_norm;\n"
394 : : "drop table if exists buildids4_files;\n"
395 : : "drop table if exists buildids4_buildids;\n"
396 : : "drop table if exists buildids4_bolo;\n"
397 : : "drop table if exists buildids4_rfolo;\n"
398 : : "drop view if exists buildids4;\n"
399 : : // buildids3*: split out srcfile BOLO
400 : : "drop table if exists buildids3_norm;\n"
401 : : "drop table if exists buildids3_files;\n"
402 : : "drop table if exists buildids3_buildids;\n"
403 : : "drop table if exists buildids3_bolo;\n"
404 : : "drop view if exists buildids3;\n"
405 : : // buildids2: normalized buildid and filenames into interning tables;
406 : : "drop table if exists buildids2_norm;\n"
407 : : "drop table if exists buildids2_files;\n"
408 : : "drop table if exists buildids2_buildids;\n"
409 : : "drop view if exists buildids2;\n"
410 : : // buildids1: made buildid and artifacttype NULLable, to represent cached-negative
411 : : // lookups from sources, e.g. files or rpms that contain no buildid-indexable content
412 : : "drop table if exists buildids1;\n"
413 : : // buildids: original
414 : : "drop table if exists buildids;\n"
415 : : ;
416 : :
417 : : static const char DEBUGINFOD_SQLITE_CLEANUP_DDL[] =
418 : : "pragma wal_checkpoint = truncate;\n" // clean out any preexisting wal file
419 : : ;
420 : :
421 : :
422 : :
423 : :
424 : : /* Name and version of program. */
425 : : ARGP_PROGRAM_VERSION_HOOK_DEF = print_version;
426 : :
427 : : /* Bug report address. */
428 : : ARGP_PROGRAM_BUG_ADDRESS_DEF = PACKAGE_BUGREPORT;
429 : :
430 : : /* Definitions of arguments for argp functions. */
431 : : static const struct argp_option options[] =
432 : : {
433 : : { NULL, 0, NULL, 0, "Scanners:", 1 },
434 : : { "scan-file-dir", 'F', NULL, 0, "Enable ELF/DWARF file scanning.", 0 },
435 : : { "scan-rpm-dir", 'R', NULL, 0, "Enable RPM scanning.", 0 },
436 : : { "scan-deb-dir", 'U', NULL, 0, "Enable DEB scanning.", 0 },
437 : : { "scan-archive", 'Z', "EXT=CMD", 0, "Enable arbitrary archive scanning.", 0 },
438 : : // "source-oci-imageregistry" ...
439 : :
440 : : { NULL, 0, NULL, 0, "Options:", 2 },
441 : : { "logical", 'L', NULL, 0, "Follow symlinks, default=ignore.", 0 },
442 : : { "rescan-time", 't', "SECONDS", 0, "Number of seconds to wait between rescans, 0=disable.", 0 },
443 : : { "groom-time", 'g', "SECONDS", 0, "Number of seconds to wait between database grooming, 0=disable.", 0 },
444 : : { "maxigroom", 'G', NULL, 0, "Run a complete database groom/shrink pass at startup.", 0 },
445 : : { "concurrency", 'c', "NUM", 0, "Limit scanning thread concurrency to NUM, default=#CPUs.", 0 },
446 : : { "connection-pool", 'C', "NUM", OPTION_ARG_OPTIONAL,
447 : : "Use webapi connection pool with NUM threads, default=unlim.", 0 },
448 : : { "include", 'I', "REGEX", 0, "Include files matching REGEX, default=all.", 0 },
449 : : { "exclude", 'X', "REGEX", 0, "Exclude files matching REGEX, default=none.", 0 },
450 : : { "port", 'p', "NUM", 0, "HTTP port to listen on, default 8002.", 0 },
451 : : { "database", 'd', "FILE", 0, "Path to sqlite database.", 0 },
452 : : { "ddl", 'D', "SQL", 0, "Apply extra sqlite ddl/pragma to connection.", 0 },
453 : : { "verbose", 'v', NULL, 0, "Increase verbosity.", 0 },
454 : : { "regex-groom", 'r', NULL, 0,"Uses regexes from -I and -X arguments to groom the database.",0},
455 : : #define ARGP_KEY_FDCACHE_FDS 0x1001
456 : : { "fdcache-fds", ARGP_KEY_FDCACHE_FDS, "NUM", OPTION_HIDDEN, NULL, 0 },
457 : : #define ARGP_KEY_FDCACHE_MBS 0x1002
458 : : { "fdcache-mbs", ARGP_KEY_FDCACHE_MBS, "MB", 0, "Maximum total size of archive file fdcache.", 0 },
459 : : #define ARGP_KEY_FDCACHE_PREFETCH 0x1003
460 : : { "fdcache-prefetch", ARGP_KEY_FDCACHE_PREFETCH, "NUM", 0, "Number of archive files to prefetch into fdcache.", 0 },
461 : : #define ARGP_KEY_FDCACHE_MINTMP 0x1004
462 : : { "fdcache-mintmp", ARGP_KEY_FDCACHE_MINTMP, "NUM", 0, "Minimum free space% on tmpdir.", 0 },
463 : : #define ARGP_KEY_FDCACHE_PREFETCH_MBS 0x1005
464 : : { "fdcache-prefetch-mbs", ARGP_KEY_FDCACHE_PREFETCH_MBS, "MB", OPTION_HIDDEN, NULL, 0},
465 : : #define ARGP_KEY_FDCACHE_PREFETCH_FDS 0x1006
466 : : { "fdcache-prefetch-fds", ARGP_KEY_FDCACHE_PREFETCH_FDS, "NUM", OPTION_HIDDEN, NULL, 0},
467 : : #define ARGP_KEY_FORWARDED_TTL_LIMIT 0x1007
468 : : {"forwarded-ttl-limit", ARGP_KEY_FORWARDED_TTL_LIMIT, "NUM", 0, "Limit of X-Forwarded-For hops, default 8.", 0},
469 : : #define ARGP_KEY_PASSIVE 0x1008
470 : : { "passive", ARGP_KEY_PASSIVE, NULL, 0, "Do not scan or groom, read-only database.", 0 },
471 : : #define ARGP_KEY_DISABLE_SOURCE_SCAN 0x1009
472 : : { "disable-source-scan", ARGP_KEY_DISABLE_SOURCE_SCAN, NULL, 0, "Do not scan dwarf source info.", 0 },
473 : : #define ARGP_SCAN_CHECKPOINT 0x100A
474 : : { "scan-checkpoint", ARGP_SCAN_CHECKPOINT, "NUM", 0, "Number of files scanned before a WAL checkpoint.", 0 },
475 : : #ifdef ENABLE_IMA_VERIFICATION
476 : : #define ARGP_KEY_KOJI_SIGCACHE 0x100B
477 : : { "koji-sigcache", ARGP_KEY_KOJI_SIGCACHE, NULL, 0, "Do a koji specific mapping of rpm paths to get IMA signatures.", 0 },
478 : : #endif
479 : : #define ARGP_KEY_METADATA_MAXTIME 0x100C
480 : : { "metadata-maxtime", ARGP_KEY_METADATA_MAXTIME, "SECONDS", 0,
481 : : "Number of seconds to limit metadata query run time, 0=unlimited.", 0 },
482 : : { NULL, 0, NULL, 0, NULL, 0 },
483 : : };
484 : :
485 : : /* Short description of program. */
486 : : static const char doc[] = "Serve debuginfo-related content across HTTP from files under PATHs.";
487 : :
488 : : /* Strings for arguments in help texts. */
489 : : static const char args_doc[] = "[PATH ...]";
490 : :
491 : : /* Prototype for option handler. */
492 : : static error_t parse_opt (int key, char *arg, struct argp_state *state);
493 : :
494 : : static unsigned default_concurrency();
495 : :
496 : : /* Data structure to communicate with argp functions. */
497 : : static struct argp argp =
498 : : {
499 : : options, parse_opt, args_doc, doc, NULL, NULL, NULL
500 : : };
501 : :
502 : :
503 : : static string db_path;
504 : : static sqlite3 *db; // single connection, serialized across all our threads!
505 : : static sqlite3 *dbq; // webapi query-servicing readonly connection, serialized ditto!
506 : : static unsigned verbose;
507 : : static volatile sig_atomic_t interrupted = 0;
508 : : static volatile sig_atomic_t forced_rescan_count = 0;
509 : : static volatile sig_atomic_t sigusr1 = 0;
510 : : static volatile sig_atomic_t forced_groom_count = 0;
511 : : static volatile sig_atomic_t sigusr2 = 0;
512 : : static unsigned http_port = 8002;
513 : : static unsigned rescan_s = 300;
514 : : static unsigned groom_s = 86400;
515 : : static bool maxigroom = false;
516 : : static unsigned concurrency = default_concurrency();
517 : : static int connection_pool = 0;
518 : : static set<string> source_paths;
519 : : static bool scan_files = false;
520 : : static map<string,string> scan_archives;
521 : : static vector<string> extra_ddl;
522 : : static regex_t file_include_regex;
523 : : static regex_t file_exclude_regex;
524 : : static bool regex_groom = false;
525 : : static bool traverse_logical;
526 : : static long fdcache_mbs;
527 : : static long fdcache_prefetch;
528 : : static long fdcache_mintmp;
529 : : static unsigned forwarded_ttl_limit = 8;
530 : : static bool scan_source_info = true;
531 : : static string tmpdir;
532 : : static bool passive_p = false;
533 : : static long scan_checkpoint = 256;
534 : : #ifdef ENABLE_IMA_VERIFICATION
535 : : static bool requires_koji_sigcache_mapping = false;
536 : : #endif
537 : : static unsigned metadata_maxtime_s = 5;
538 : :
539 : : static void set_metric(const string& key, double value);
540 : : static void inc_metric(const string& key);
541 : : static void add_metric(const string& metric,
542 : : double value);
543 : : static void set_metric(const string& metric,
544 : : const string& lname, const string& lvalue,
545 : : double value);
546 : : static void inc_metric(const string& metric,
547 : : const string& lname, const string& lvalue);
548 : : static void add_metric(const string& metric,
549 : : const string& lname, const string& lvalue,
550 : : double value);
551 : : static void inc_metric(const string& metric,
552 : : const string& lname, const string& lvalue,
553 : : const string& rname, const string& rvalue);
554 : : static void add_metric(const string& metric,
555 : : const string& lname, const string& lvalue,
556 : : const string& rname, const string& rvalue,
557 : : double value);
558 : :
559 : :
560 : : class tmp_inc_metric { // a RAII style wrapper for exception-safe scoped increment & decrement
561 : : string m, n, v;
562 : : public:
563 : 2246 : tmp_inc_metric(const string& mname, const string& lname, const string& lvalue):
564 [ + - + - ]: 2246 : m(mname), n(lname), v(lvalue)
565 : : {
566 [ + - ]: 2246 : add_metric (m, n, v, 1);
567 [ - - - - : 2246 : }
- - ]
568 : 2246 : ~tmp_inc_metric()
569 : : {
570 : 2246 : add_metric (m, n, v, -1);
571 [ - + - + : 2246 : }
- + ]
572 : : };
573 : :
574 : : class tmp_ms_metric { // a RAII style wrapper for exception-safe scoped timing
575 : : string m, n, v;
576 : : struct timespec ts_start;
577 : : public:
578 : 110208 : tmp_ms_metric(const string& mname, const string& lname, const string& lvalue):
579 [ + - + - ]: 110208 : m(mname), n(lname), v(lvalue)
580 : : {
581 : 110203 : clock_gettime (CLOCK_MONOTONIC, & ts_start);
582 [ - - - - ]: 110215 : }
583 : 110221 : ~tmp_ms_metric()
584 : : {
585 : 110221 : struct timespec ts_end;
586 : 110221 : clock_gettime (CLOCK_MONOTONIC, & ts_end);
587 : 110220 : double deltas = (ts_end.tv_sec - ts_start.tv_sec)
588 : 110220 : + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
589 : :
590 [ + - ]: 110220 : add_metric (m + "_milliseconds_sum", n, v, (deltas*1000.0));
591 [ + - ]: 110222 : inc_metric (m + "_milliseconds_count", n, v);
592 [ + + - + : 192672 : }
- + ]
593 : : };
594 : :
595 : :
596 : : /* Handle program arguments. */
597 : : static error_t
598 : 1174 : parse_opt (int key, char *arg,
599 : : struct argp_state *state __attribute__ ((unused)))
600 : : {
601 : 1174 : int rc;
602 [ + + + + : 1174 : switch (key)
+ + + + +
- + + - -
+ + + + +
+ + + + +
- + - + ]
603 : : {
604 : 280 : case 'v': verbose ++; break;
605 : 76 : case 'd':
606 : : /* When using the in-memory database make sure it is shareable,
607 : : so we can open it twice as read/write and read-only. */
608 [ + + ]: 76 : if (strcmp (arg, ":memory:") == 0)
609 : 1188 : db_path = "file::memory:?cache=shared";
610 : : else
611 [ + - ]: 124 : db_path = string(arg);
612 : : break;
613 : 76 : case 'p': http_port = (unsigned) atoi(arg);
614 [ + - ]: 76 : if (http_port == 0 || http_port > 65535)
615 : 0 : argp_failure(state, 1, EINVAL, "port number");
616 : : break;
617 : 48 : case 'F': scan_files = true; break;
618 : 24 : case 'R':
619 [ + - + - : 24 : scan_archives[".rpm"]="cat"; // libarchive groks rpm natively
- + ]
620 : 24 : break;
621 : 18 : case 'U':
622 [ + - + - : 18 : scan_archives[".deb"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
623 [ + - + - : 18 : scan_archives[".ddeb"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
624 [ + - + - : 18 : scan_archives[".ipk"]="(bsdtar -O -x -f - data.tar\\*)<";
- + ]
625 : : // .udeb too?
626 : 18 : break;
627 : 38 : case 'Z':
628 : 38 : {
629 [ - + ]: 38 : char* extension = strchr(arg, '=');
630 [ - + ]: 38 : if (arg[0] == '\0')
631 : 0 : argp_failure(state, 1, EINVAL, "missing EXT");
632 [ + + ]: 38 : else if (extension)
633 [ + - + - : 20 : scan_archives[string(arg, (extension-arg))]=string(extension+1);
- + - + -
- ]
634 : : else
635 [ + - + - : 18 : scan_archives[string(arg)]=string("cat");
- + - + -
- ]
636 : : }
637 : : break;
638 : 8 : case 'L':
639 [ - + ]: 8 : if (passive_p)
640 : 0 : argp_failure(state, 1, EINVAL, "-L option inconsistent with passive mode");
641 : 8 : traverse_logical = true;
642 : 8 : break;
643 : 0 : case 'D':
644 [ # # ]: 0 : if (passive_p)
645 : 0 : argp_failure(state, 1, EINVAL, "-D option inconsistent with passive mode");
646 [ # # ]: 0 : extra_ddl.push_back(string(arg));
647 : 0 : break;
648 : 62 : case 't':
649 [ - + ]: 62 : if (passive_p)
650 : 0 : argp_failure(state, 1, EINVAL, "-t option inconsistent with passive mode");
651 : 62 : rescan_s = (unsigned) atoi(arg);
652 : 62 : break;
653 : 62 : case 'g':
654 [ - + ]: 62 : if (passive_p)
655 : 0 : argp_failure(state, 1, EINVAL, "-g option inconsistent with passive mode");
656 : 62 : groom_s = (unsigned) atoi(arg);
657 : 62 : break;
658 : 0 : case 'G':
659 [ # # ]: 0 : if (passive_p)
660 : 0 : argp_failure(state, 1, EINVAL, "-G option inconsistent with passive mode");
661 : 0 : maxigroom = true;
662 : 0 : break;
663 : 0 : case 'c':
664 [ # # ]: 0 : if (passive_p)
665 : 0 : argp_failure(state, 1, EINVAL, "-c option inconsistent with passive mode");
666 : 0 : concurrency = (unsigned) atoi(arg);
667 [ # # ]: 0 : if (concurrency < 1) concurrency = 1;
668 : : break;
669 : 6 : case 'C':
670 [ + + ]: 6 : if (arg)
671 : : {
672 : 4 : connection_pool = atoi(arg);
673 [ + - ]: 4 : if (connection_pool < 2)
674 : 0 : argp_failure(state, 1, EINVAL, "-C NUM minimum 2");
675 : : }
676 : : break;
677 : 4 : case 'I':
678 : : // NB: no problem with unconditional free here - an earlier failed regcomp would exit program
679 [ - + ]: 4 : if (passive_p)
680 : 0 : argp_failure(state, 1, EINVAL, "-I option inconsistent with passive mode");
681 : 4 : regfree (&file_include_regex);
682 : 4 : rc = regcomp (&file_include_regex, arg, REG_EXTENDED|REG_NOSUB);
683 [ + - ]: 4 : if (rc != 0)
684 : 0 : argp_failure(state, 1, EINVAL, "regular expression");
685 : : break;
686 : 6 : case 'X':
687 [ - + ]: 6 : if (passive_p)
688 : 0 : argp_failure(state, 1, EINVAL, "-X option inconsistent with passive mode");
689 : 6 : regfree (&file_exclude_regex);
690 : 6 : rc = regcomp (&file_exclude_regex, arg, REG_EXTENDED|REG_NOSUB);
691 [ + - ]: 6 : if (rc != 0)
692 : 0 : argp_failure(state, 1, EINVAL, "regular expression");
693 : : break;
694 : 4 : case 'r':
695 [ - + ]: 4 : if (passive_p)
696 : 0 : argp_failure(state, 1, EINVAL, "-r option inconsistent with passive mode");
697 : 4 : regex_groom = true;
698 : 4 : break;
699 : : case ARGP_KEY_FDCACHE_FDS:
700 : : // deprecated
701 : : break;
702 : 4 : case ARGP_KEY_FDCACHE_MBS:
703 : 4 : fdcache_mbs = atol (arg);
704 : 4 : break;
705 : 4 : case ARGP_KEY_FDCACHE_PREFETCH:
706 : 4 : fdcache_prefetch = atol (arg);
707 : 4 : break;
708 : 4 : case ARGP_KEY_FDCACHE_MINTMP:
709 : 4 : fdcache_mintmp = atol (arg);
710 [ + - ]: 4 : if( fdcache_mintmp > 100 || fdcache_mintmp < 0 )
711 : 0 : argp_failure(state, 1, EINVAL, "fdcache mintmp percent");
712 : : break;
713 : 4 : case ARGP_KEY_FORWARDED_TTL_LIMIT:
714 : 4 : forwarded_ttl_limit = (unsigned) atoi(arg);
715 : 4 : break;
716 : 106 : case ARGP_KEY_ARG:
717 [ + - ]: 106 : source_paths.insert(string(arg));
718 : 106 : break;
719 : : case ARGP_KEY_FDCACHE_PREFETCH_FDS:
720 : : // deprecated
721 : : break;
722 : : case ARGP_KEY_FDCACHE_PREFETCH_MBS:
723 : : // deprecated
724 : : break;
725 : 2 : case ARGP_KEY_PASSIVE:
726 : 2 : passive_p = true;
727 [ + - ]: 2 : if (source_paths.size() > 0
728 [ + - ]: 2 : || maxigroom
729 [ + - ]: 2 : || extra_ddl.size() > 0
730 [ + - + - ]: 4 : || traverse_logical)
731 : : // other conflicting options tricky to check
732 : 0 : argp_failure(state, 1, EINVAL, "inconsistent options with passive mode");
733 : : break;
734 : 0 : case ARGP_KEY_DISABLE_SOURCE_SCAN:
735 : 0 : scan_source_info = false;
736 : 0 : break;
737 : 2 : case ARGP_SCAN_CHECKPOINT:
738 : 2 : scan_checkpoint = atol (arg);
739 [ + - ]: 2 : if (scan_checkpoint < 0)
740 : 0 : argp_failure(state, 1, EINVAL, "scan checkpoint");
741 : : break;
742 : 0 : case ARGP_KEY_METADATA_MAXTIME:
743 : 0 : metadata_maxtime_s = (unsigned) atoi(arg);
744 : 0 : break;
745 : : #ifdef ENABLE_IMA_VERIFICATION
746 : : case ARGP_KEY_KOJI_SIGCACHE:
747 : : requires_koji_sigcache_mapping = true;
748 : : break;
749 : : #endif
750 : : // case 'h': argp_state_help (state, stderr, ARGP_HELP_LONG|ARGP_HELP_EXIT_OK);
751 : : default: return ARGP_ERR_UNKNOWN;
752 : : }
753 : :
754 : : return 0;
755 : : }
756 : :
757 : :
758 : : ////////////////////////////////////////////////////////////////////////
759 : :
760 : :
761 : : static void add_mhd_response_header (struct MHD_Response *r,
762 : : const char *h, const char *v);
763 : :
764 : : // represent errors that may get reported to an ostream and/or a libmicrohttpd connection
765 : :
766 : 8 : struct reportable_exception
767 : : {
768 : : int code;
769 : : string message;
770 : :
771 [ - - + - : 106 : reportable_exception(int c, const string& m): code(c), message(m) {}
- - + - +
- ]
772 [ - - - - : 600 : reportable_exception(const string& m): code(503), message(m) {}
- - - - -
- - - + -
- - - - +
- - - - -
- - - - -
- - - - -
- - - - -
- - - + -
- - ]
773 : : reportable_exception(): code(503), message() {}
774 : :
775 : : void report(ostream& o) const; // defined under obatched() class below
776 : :
777 : 630 : MHD_RESULT mhd_send_response(MHD_Connection* c) const {
778 : 1260 : MHD_Response* r = MHD_create_response_from_buffer (message.size(),
779 : 630 : (void*) message.c_str(),
780 : : MHD_RESPMEM_MUST_COPY);
781 : 630 : add_mhd_response_header (r, "Content-Type", "text/plain");
782 : 630 : MHD_RESULT rc = MHD_queue_response (c, code, r);
783 : 630 : MHD_destroy_response (r);
784 : 630 : return rc;
785 : : }
786 : : };
787 : :
788 : :
789 : : struct sqlite_exception: public reportable_exception
790 : : {
791 : 0 : sqlite_exception(int rc, const string& msg):
792 [ # # # # : 0 : reportable_exception(string("sqlite3 error: ") + msg + ": " + string(sqlite3_errstr(rc) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
793 [ # # # # : 0 : inc_metric("error_count","sqlite3",sqlite3_errstr(rc));
# # # # #
# # # # #
# # # # ]
794 [ # # ]: 0 : }
795 : : };
796 : :
797 [ + - - - ]: 4 : struct libc_exception: public reportable_exception
798 : : {
799 : 594 : libc_exception(int rc, const string& msg):
800 [ - + + - : 2376 : reportable_exception(string("libc error: ") + msg + ": " + string(strerror(rc) ?: "?")) {
+ - + - +
- - + - +
- + + - -
- ]
801 [ + - + - : 1188 : inc_metric("error_count","libc",strerror(rc));
+ - + - -
+ + - - -
- - ]
802 [ - - ]: 594 : }
803 : : };
804 : :
805 : :
806 : : struct archive_exception: public reportable_exception
807 : : {
808 : 0 : archive_exception(const string& msg):
809 [ # # # # : 0 : reportable_exception(string("libarchive error: ") + msg) {
# # ]
810 [ # # # # : 0 : inc_metric("error_count","libarchive",msg);
# # # # #
# ]
811 [ # # ]: 0 : }
812 : 0 : archive_exception(struct archive* a, const string& msg):
813 [ # # # # : 0 : reportable_exception(string("libarchive error: ") + msg + ": " + string(archive_error_string(a) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
814 [ # # # # : 0 : inc_metric("error_count","libarchive",msg + ": " + string(archive_error_string(a) ?: "?"));
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# ]
815 [ # # ]: 0 : }
816 : : };
817 : :
818 : :
819 : : struct elfutils_exception: public reportable_exception
820 : : {
821 : 0 : elfutils_exception(int rc, const string& msg):
822 [ # # # # : 0 : reportable_exception(string("elfutils error: ") + msg + ": " + string(elf_errmsg(rc) ?: "?")) {
# # # # #
# # # # #
# # # # #
# ]
823 [ # # # # : 0 : inc_metric("error_count","elfutils",elf_errmsg(rc));
# # # # #
# # # # #
# # # # ]
824 [ # # ]: 0 : }
825 : : };
826 : :
827 : :
828 : : ////////////////////////////////////////////////////////////////////////
829 : :
830 : : template <typename Payload>
831 : : class workq
832 : : {
833 : : unordered_set<Payload> q; // eliminate duplicates
834 : : mutex mtx;
835 : : condition_variable cv;
836 : : bool dead;
837 : : unsigned idlers; // number of threads busy with wait_idle / done_idle
838 : : unsigned fronters; // number of threads busy with wait_front / done_front
839 : :
840 : : public:
841 : 76 : workq() { dead = false; idlers = 0; fronters = 0; }
842 : 76 : ~workq() {}
843 : :
844 : 994 : void push_back(const Payload& p)
845 : : {
846 : 994 : unique_lock<mutex> lock(mtx);
847 [ + - ]: 994 : q.insert (p);
848 [ + - + - : 1988 : set_metric("thread_work_pending","role","scan", q.size());
+ - + - -
+ - + - -
- - ]
849 : 994 : cv.notify_all();
850 : 994 : }
851 : :
852 : : // kill this workqueue, wake up all idlers / scanners
853 : 76 : void nuke() {
854 : 76 : unique_lock<mutex> lock(mtx);
855 : : // optional: q.clear();
856 : 76 : dead = true;
857 : 76 : cv.notify_all();
858 : 76 : }
859 : :
860 : : // clear the workqueue, when scanning is interrupted with USR2
861 : 0 : void clear() {
862 : 0 : unique_lock<mutex> lock(mtx);
863 : 0 : q.clear();
864 [ # # # # : 0 : set_metric("thread_work_pending","role","scan", q.size());
# # # # #
# # # # #
# # ]
865 : : // NB: there may still be some live fronters
866 : 0 : cv.notify_all(); // maybe wake up waiting idlers
867 : 0 : }
868 : :
869 : : // block this scanner thread until there is work to do and no active idler
870 : 1266 : bool wait_front (Payload& p)
871 : : {
872 : 1266 : unique_lock<mutex> lock(mtx);
873 [ + + + + : 4527 : while (!dead && (q.size() == 0 || idlers > 0))
+ + ]
874 [ + - ]: 3261 : cv.wait(lock);
875 [ + + ]: 1266 : if (dead)
876 : : return false;
877 : : else
878 : : {
879 [ + - ]: 994 : p = * q.begin();
880 : 994 : q.erase (q.begin());
881 : 994 : fronters ++; // prevent idlers from starting awhile, even if empty q
882 [ + - + - : 1988 : set_metric("thread_work_pending","role","scan", q.size());
+ - + - -
+ - + - -
- - - - ]
883 : : // NB: don't wake up idlers yet! The consumer is busy
884 : : // processing this element until it calls done_front().
885 : 994 : return true;
886 : : }
887 : 1266 : }
888 : :
889 : : // notify waitq that scanner thread is done with that last item
890 : 994 : void done_front ()
891 : : {
892 : 994 : unique_lock<mutex> lock(mtx);
893 : 994 : fronters --;
894 [ + + + + ]: 994 : if (q.size() == 0 && fronters == 0)
895 : 93 : cv.notify_all(); // maybe wake up waiting idlers
896 : 994 : }
897 : :
898 : : // block this idler thread until there is no work to do
899 : 582 : void wait_idle ()
900 : : {
901 : 582 : unique_lock<mutex> lock(mtx);
902 : 582 : cv.notify_all(); // maybe wake up waiting scanners
903 [ + + + + : 622 : while (!dead && ((q.size() != 0) || fronters > 0))
+ + ]
904 [ + - ]: 40 : cv.wait(lock);
905 [ + - ]: 582 : idlers ++;
906 : 582 : }
907 : :
908 : 508 : void done_idle ()
909 : : {
910 : 508 : unique_lock<mutex> lock(mtx);
911 : 508 : idlers --;
912 : 508 : cv.notify_all(); // maybe wake up waiting scanners, but probably not (shutting down)
913 : 508 : }
914 : : };
915 : :
916 : : typedef struct stat stat_t;
917 : : typedef pair<string,stat_t> scan_payload;
918 : : inline bool operator< (const scan_payload& a, const scan_payload& b)
919 : : {
920 : : return a.first < b.first; // don't bother compare the stat fields
921 : : }
922 : :
923 : : namespace std { // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56480
924 : : template<> struct hash<::scan_payload>
925 : : {
926 : 3853 : std::size_t operator() (const ::scan_payload& p) const noexcept
927 : : {
928 [ + + + + ]: 3853 : return hash<string>()(p.first);
929 : : }
930 : : };
931 : : template<> struct equal_to<::scan_payload>
932 : : {
933 : 321 : std::size_t operator() (const ::scan_payload& a, const ::scan_payload& b) const noexcept
934 : : {
935 [ - + - - ]: 321 : return a.first == b.first;
936 : : }
937 : : };
938 : : }
939 : :
940 : : static workq<scan_payload> scanq; // just a single one
941 : : // producer & idler: thread_main_fts_source_paths()
942 : : // consumer: thread_main_scanner()
943 : : // idler: thread_main_groom()
944 : :
945 : :
946 : : ////////////////////////////////////////////////////////////////////////
947 : :
948 : : // Unique set is a thread-safe structure that lends 'ownership' of a value
949 : : // to a thread. Other threads requesting the same thing are made to wait.
950 : : // It's like a semaphore-on-demand.
951 : : template <typename T>
952 : : class unique_set
953 : : {
954 : : private:
955 : : set<T> values;
956 : : mutex mtx;
957 : : condition_variable cv;
958 : : public:
959 : 56 : unique_set() {}
960 : 56 : ~unique_set() {}
961 : :
962 : 1512 : void acquire(const T& value)
963 : : {
964 : 1512 : unique_lock<mutex> lock(mtx);
965 [ + + ]: 1923 : while (values.find(value) != values.end())
966 [ + - ]: 411 : cv.wait(lock);
967 [ + - ]: 1512 : values.insert(value);
968 : 1512 : }
969 : :
970 : 1512 : void release(const T& value)
971 : : {
972 : 1512 : unique_lock<mutex> lock(mtx);
973 : : // assert (values.find(value) != values.end());
974 : 1512 : values.erase(value);
975 : 1512 : cv.notify_all();
976 : 1512 : }
977 : : };
978 : :
979 : :
980 : : // This is the object that's instantiate to uniquely hold a value in a
981 : : // RAII-pattern way.
982 : : template <typename T>
983 : : class unique_set_reserver
984 : : {
985 : : private:
986 : : unique_set<T>& please_hold;
987 : : T mine;
988 : : public:
989 : 1512 : unique_set_reserver(unique_set<T>& t, const T& value):
990 [ + - - - ]: 1512 : please_hold(t), mine(value) { please_hold.acquire(mine); }
991 [ + - ]: 1512 : ~unique_set_reserver() { please_hold.release(mine); }
992 : : };
993 : :
994 : :
995 : : ////////////////////////////////////////////////////////////////////////
996 : :
997 : : // periodic_barrier is a concurrency control object that lets N threads
998 : : // periodically (based on counter value) agree to wait at a barrier,
999 : : // let one of them carry out some work, then be set free
1000 : :
1001 : : class periodic_barrier
1002 : : {
1003 : : private:
1004 : : unsigned period; // number of count() reports to trigger barrier activation
1005 : : unsigned threads; // number of threads participating
1006 : : mutex mtx; // protects all the following fields
1007 : : unsigned counter; // count of count() reports in the current generation
1008 : : unsigned generation; // barrier activation generation
1009 : : unsigned waiting; // number of threads waiting for barrier
1010 : : bool dead; // bring out your
1011 : : condition_variable cv;
1012 : : public:
1013 : 68 : periodic_barrier(unsigned t, unsigned p):
1014 : 68 : period(p), threads(t), counter(0), generation(0), waiting(0), dead(false) { }
1015 : : virtual ~periodic_barrier() {}
1016 : :
1017 : : virtual void periodic_barrier_work() noexcept = 0;
1018 : 68 : void nuke() {
1019 : 68 : unique_lock<mutex> lock(mtx);
1020 : 68 : dead = true;
1021 : 68 : cv.notify_all();
1022 : 68 : }
1023 : :
1024 : 1266 : void count()
1025 : : {
1026 : 1266 : unique_lock<mutex> lock(mtx);
1027 : 1266 : unsigned prev_generation = this->generation;
1028 [ + + ]: 1266 : if (counter < period-1) // normal case: counter just freely running
1029 : : {
1030 : 1134 : counter ++;
1031 : 1134 : return;
1032 : : }
1033 [ + + ]: 132 : else if (counter == period-1) // we're the doer
1034 : : {
1035 : 36 : counter = period; // entering barrier holding phase
1036 : 36 : cv.notify_all();
1037 [ + + + + ]: 165 : while (waiting < threads-1 && !dead)
1038 [ + - ]: 93 : cv.wait(lock);
1039 : : // all other threads are now stuck in the barrier
1040 : 36 : this->periodic_barrier_work(); // NB: we're holding the mutex the whole time
1041 : : // reset for next barrier, releasing other waiters
1042 : 36 : counter = 0;
1043 : 36 : generation ++;
1044 : 36 : cv.notify_all();
1045 : 36 : return;
1046 : : }
1047 [ + - ]: 96 : else if (counter == period) // we're a waiter, in holding phase
1048 : : {
1049 : 96 : waiting ++;
1050 : 96 : cv.notify_all();
1051 [ + + + + : 355 : while (counter == period && generation == prev_generation && !dead)
+ + ]
1052 [ + - ]: 163 : cv.wait(lock);
1053 : 96 : waiting --;
1054 : 96 : return;
1055 : : }
1056 : 1266 : }
1057 : : };
1058 : :
1059 : :
1060 : :
1061 : : ////////////////////////////////////////////////////////////////////////
1062 : :
1063 : :
1064 : : // Print a standard timestamp.
1065 : : static ostream&
1066 : 19397 : timestamp (ostream &o)
1067 : : {
1068 : 19397 : char datebuf[80];
1069 : 19397 : char *now2 = NULL;
1070 : 19397 : time_t now_t = time(NULL);
1071 : 19397 : struct tm now;
1072 : 19397 : struct tm *nowp = gmtime_r (&now_t, &now);
1073 [ + - ]: 19398 : if (nowp)
1074 : : {
1075 : 19398 : (void) strftime (datebuf, sizeof (datebuf), "%c", nowp);
1076 : 19398 : now2 = datebuf;
1077 : : }
1078 : :
1079 : 19398 : return o << "[" << (now2 ? now2 : "") << "] "
1080 [ - + ]: 19398 : << "(" << getpid () << "/" << tid() << "): ";
1081 : : }
1082 : :
1083 : :
1084 : : // A little class that impersonates an ostream to the extent that it can
1085 : : // take << streaming operations. It batches up the bits into an internal
1086 : : // stringstream until it is destroyed; then flushes to the original ostream.
1087 : : // It adds a timestamp
1088 : : class obatched
1089 : : {
1090 : : private:
1091 : : ostream& o;
1092 : : stringstream stro;
1093 : : static mutex lock;
1094 : : public:
1095 : 19398 : obatched(ostream& oo, bool timestamp_p = true): o(oo)
1096 : : {
1097 [ + - ]: 19398 : if (timestamp_p)
1098 [ + - ]: 19398 : timestamp(stro);
1099 : 19398 : }
1100 : 19398 : ~obatched()
1101 : : {
1102 : 19398 : unique_lock<mutex> do_not_cross_the_streams(obatched::lock);
1103 [ + - ]: 19398 : o << stro.str();
1104 : 19398 : o.flush();
1105 : 19398 : }
1106 : : operator ostream& () { return stro; }
1107 [ - - + - : 15954 : template <typename T> ostream& operator << (const T& t) { stro << t; return stro; }
+ - + - +
- + - + -
- - - - -
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
- - + - +
- + - + -
+ - + - +
- + - - -
+ - + - +
- + - + -
+ - + - -
- + - - -
+ - - - +
- + - + -
- - - - -
- - - - -
- - + - -
- + - + -
- - + - +
- - - + -
- - + - -
- - - + -
+ - + - +
- + - + -
- - - - ]
1108 : : };
1109 : : mutex obatched::lock; // just the one, since cout/cerr iostreams are not thread-safe
1110 : :
1111 : :
1112 : 692 : void reportable_exception::report(ostream& o) const {
1113 [ + - + - ]: 692 : obatched(o) << message << endl;
1114 : 692 : }
1115 : :
1116 : :
1117 : : ////////////////////////////////////////////////////////////////////////
1118 : :
1119 : :
1120 : : // RAII style sqlite prepared-statement holder that matches { } block lifetime
1121 : :
1122 : : struct sqlite_ps
1123 : : {
1124 : : private:
1125 : : sqlite3* db;
1126 : : const string nickname;
1127 : : const string sql;
1128 : : sqlite3_stmt *pp;
1129 : :
1130 : : sqlite_ps(const sqlite_ps&); // make uncopyable
1131 : : sqlite_ps& operator=(const sqlite_ps &); // make unassignable
1132 : :
1133 : : public:
1134 [ + - - - ]: 7742 : sqlite_ps (sqlite3* d, const string& n, const string& s): db(d), nickname(n), sql(s) {
1135 : : // tmp_ms_metric tick("sqlite3","prep",nickname);
1136 [ + + ]: 7742 : if (verbose > 4)
1137 [ + - + - : 174 : obatched(clog) << nickname << " prep " << sql << endl;
+ - + - +
- - - ]
1138 [ + - ]: 7742 : int rc = sqlite3_prepare_v2 (db, sql.c_str(), -1 /* to \0 */, & this->pp, NULL);
1139 [ - + ]: 7742 : if (rc != SQLITE_OK)
1140 [ # # # # ]: 0 : throw sqlite_exception(rc, "prepare " + sql);
1141 : 7742 : }
1142 : :
1143 : 60456 : sqlite_ps& reset()
1144 : : {
1145 [ + - + - : 120910 : tmp_ms_metric tick("sqlite3","reset",nickname);
- + - - ]
1146 [ + - ]: 60454 : sqlite3_reset(this->pp);
1147 : 60457 : return *this;
1148 : 60457 : }
1149 : :
1150 : 63790 : sqlite_ps& bind(int parameter, const string& str)
1151 : : {
1152 [ + + ]: 63790 : if (verbose > 4)
1153 [ + - + - : 196 : obatched(clog) << nickname << " bind " << parameter << "=" << str << endl;
+ - + - +
- + - ]
1154 : 63790 : int rc = sqlite3_bind_text (this->pp, parameter, str.c_str(), -1, SQLITE_TRANSIENT);
1155 [ - + ]: 63789 : if (rc != SQLITE_OK)
1156 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 bind");
1157 : 63789 : return *this;
1158 : : }
1159 : :
1160 : 22527 : sqlite_ps& bind(int parameter, int64_t value)
1161 : : {
1162 [ + + ]: 22527 : if (verbose > 4)
1163 [ + - + - : 64 : obatched(clog) << nickname << " bind " << parameter << "=" << value << endl;
+ - + - +
- + - ]
1164 : 22527 : int rc = sqlite3_bind_int64 (this->pp, parameter, value);
1165 [ - + ]: 22530 : if (rc != SQLITE_OK)
1166 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 bind");
1167 : 22530 : return *this;
1168 : : }
1169 : :
1170 : : sqlite_ps& bind(int parameter)
1171 : : {
1172 : : if (verbose > 4)
1173 : : obatched(clog) << nickname << " bind " << parameter << "=" << "NULL" << endl;
1174 : : int rc = sqlite3_bind_null (this->pp, parameter);
1175 : : if (rc != SQLITE_OK)
1176 : : throw sqlite_exception(rc, "sqlite3 bind");
1177 : : return *this;
1178 : : }
1179 : :
1180 : :
1181 : 35579 : void step_ok_done() {
1182 [ + - + - : 71158 : tmp_ms_metric tick("sqlite3","step_done",nickname);
- + - - ]
1183 [ + - ]: 35579 : int rc = sqlite3_step (this->pp);
1184 [ + + ]: 35582 : if (verbose > 4)
1185 [ + - + - : 128 : obatched(clog) << nickname << " step-ok-done(" << sqlite3_errstr(rc) << ") " << sql << endl;
+ - + - +
- + - + -
+ - ]
1186 [ + + - + ]: 35582 : if (rc != SQLITE_OK && rc != SQLITE_DONE && rc != SQLITE_ROW)
1187 [ # # # # ]: 0 : throw sqlite_exception(rc, "sqlite3 step");
1188 [ + - ]: 35582 : (void) sqlite3_reset (this->pp);
1189 : 35582 : }
1190 : :
1191 : :
1192 : 14182 : int step() {
1193 [ + - + - : 28364 : tmp_ms_metric tick("sqlite3","step",nickname);
- + - - ]
1194 [ + - ]: 14182 : int rc = sqlite3_step (this->pp);
1195 [ + + ]: 14182 : if (verbose > 4)
1196 [ + - + - : 62 : obatched(clog) << nickname << " step(" << sqlite3_errstr(rc) << ") " << sql << endl;
+ - + - +
- + - + -
+ - ]
1197 : 14182 : return rc;
1198 : 14182 : }
1199 : :
1200 [ + + + + ]: 15335 : ~sqlite_ps () { sqlite3_finalize (this->pp); }
1201 [ + - + - : 3620 : operator sqlite3_stmt* () { return this->pp; }
+ - + - +
- + - + -
+ - + - +
- ]
1202 : : };
1203 : :
1204 : :
1205 : : ////////////////////////////////////////////////////////////////////////
1206 : :
1207 : :
1208 : : struct sqlite_checkpoint_pb: public periodic_barrier
1209 : : {
1210 : : // NB: don't use sqlite_ps since it can throw exceptions during ctor etc.
1211 : 68 : sqlite_checkpoint_pb(unsigned t, unsigned p):
1212 : 136 : periodic_barrier(t, p) { }
1213 : :
1214 : 36 : void periodic_barrier_work() noexcept
1215 : : {
1216 : 36 : (void) sqlite3_exec (db, "pragma wal_checkpoint(truncate);", NULL, NULL, NULL);
1217 : 36 : }
1218 : : };
1219 : :
1220 : : static periodic_barrier* scan_barrier = 0; // initialized in main()
1221 : :
1222 : :
1223 : : ////////////////////////////////////////////////////////////////////////
1224 : :
1225 : : // RAII style templated autocloser
1226 : :
1227 : : template <class Payload, class Ignore>
1228 : : struct defer_dtor
1229 : : {
1230 : : public:
1231 : : typedef Ignore (*dtor_fn) (Payload);
1232 : :
1233 : : private:
1234 : : Payload p;
1235 : : dtor_fn fn;
1236 : :
1237 : : public:
1238 : 5012 : defer_dtor(Payload _p, dtor_fn _fn): p(_p), fn(_fn) {}
1239 : 746 : ~defer_dtor() { (void) (*fn)(p); }
1240 : :
1241 : : private:
1242 : : defer_dtor(const defer_dtor<Payload,Ignore>&); // make uncopyable
1243 : : defer_dtor& operator=(const defer_dtor<Payload,Ignore> &); // make unassignable
1244 : : };
1245 : :
1246 : :
1247 : :
1248 : : ////////////////////////////////////////////////////////////////////////
1249 : :
1250 : :
1251 : : static string
1252 : 4504 : header_censor(const string& str)
1253 : : {
1254 : 4504 : string y;
1255 [ + + ]: 41044 : for (auto&& x : str)
1256 : : {
1257 [ + + ]: 36540 : if (isalnum(x) || x == '/' || x == '.' || x == ',' || x == '_' || x == ':')
1258 [ + - ]: 73074 : y += x;
1259 : : }
1260 : 4504 : return y;
1261 : 0 : }
1262 : :
1263 : :
1264 : : static string
1265 : 2252 : conninfo (struct MHD_Connection * conn)
1266 : : {
1267 : 2252 : char hostname[256]; // RFC1035
1268 : 2252 : char servname[256];
1269 : 2252 : int sts = -1;
1270 : :
1271 [ - + ]: 2252 : if (conn == 0)
1272 : 0 : return "internal";
1273 : :
1274 : : /* Look up client address data. */
1275 : 2252 : const union MHD_ConnectionInfo *u = MHD_get_connection_info (conn,
1276 : : MHD_CONNECTION_INFO_CLIENT_ADDRESS);
1277 [ + - ]: 2252 : struct sockaddr *so = u ? u->client_addr : 0;
1278 : :
1279 [ + - - + ]: 2252 : if (so && so->sa_family == AF_INET) {
1280 : 0 : sts = getnameinfo (so, sizeof (struct sockaddr_in),
1281 : : hostname, sizeof (hostname),
1282 : : servname, sizeof (servname),
1283 : : NI_NUMERICHOST | NI_NUMERICSERV);
1284 [ + - ]: 2252 : } else if (so && so->sa_family == AF_INET6) {
1285 : 2252 : struct sockaddr_in6* addr6 = (struct sockaddr_in6*) so;
1286 [ + - + - : 2252 : if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
- + ]
1287 : 2252 : struct sockaddr_in addr4;
1288 : 2252 : memset (&addr4, 0, sizeof(addr4));
1289 : 2252 : addr4.sin_family = AF_INET;
1290 : 2252 : addr4.sin_port = addr6->sin6_port;
1291 : 2252 : memcpy (&addr4.sin_addr.s_addr, addr6->sin6_addr.s6_addr+12, sizeof(addr4.sin_addr.s_addr));
1292 : 2252 : sts = getnameinfo ((struct sockaddr*) &addr4, sizeof (addr4),
1293 : : hostname, sizeof (hostname),
1294 : : servname, sizeof (servname),
1295 : : NI_NUMERICHOST | NI_NUMERICSERV);
1296 : : } else {
1297 : 0 : sts = getnameinfo (so, sizeof (struct sockaddr_in6),
1298 : : hostname, sizeof (hostname),
1299 : : servname, sizeof (servname),
1300 : : NI_NUMERICHOST | NI_NUMERICSERV);
1301 : : }
1302 : : }
1303 : :
1304 [ - + ]: 2252 : if (sts != 0) {
1305 : 0 : hostname[0] = servname[0] = '\0';
1306 : : }
1307 : :
1308 : : // extract headers relevant to administration
1309 [ - + ]: 2252 : const char* user_agent = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "User-Agent") ?: "";
1310 [ + + ]: 2252 : const char* x_forwarded_for = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "X-Forwarded-For") ?: "";
1311 : : // NB: these are untrustworthy, beware if machine-processing log files
1312 : :
1313 [ + - + - : 6756 : return string(hostname) + string(":") + string(servname) +
+ - + - +
- + - - +
- + - + -
+ - + - +
- - - - -
- ]
1314 [ + - + - : 9450 : string(" UA:") + header_censor(string(user_agent)) +
+ - + - +
- - + + +
- + + + -
+ - - -
- ]
1315 [ + - + - : 6768 : string(" XFF:") + header_censor(string(x_forwarded_for));
+ - + + +
+ - - ]
1316 : : }
1317 : :
1318 : :
1319 : :
1320 : : ////////////////////////////////////////////////////////////////////////
1321 : :
1322 : : /* Wrapper for MHD_add_response_header that logs an error if we
1323 : : couldn't add the specified header. */
1324 : : static void
1325 : 7708 : add_mhd_response_header (struct MHD_Response *r,
1326 : : const char *h, const char *v)
1327 : : {
1328 [ - + ]: 7708 : if (MHD_add_response_header (r, h, v) == MHD_NO)
1329 [ # # # # : 0 : obatched(clog) << "Error: couldn't add '" << h << "' header" << endl;
# # ]
1330 : 7708 : }
1331 : :
1332 : : static void
1333 : 1070 : add_mhd_last_modified (struct MHD_Response *resp, time_t mtime)
1334 : : {
1335 : 1070 : struct tm now;
1336 : 1070 : struct tm *nowp = gmtime_r (&mtime, &now);
1337 [ + - ]: 1070 : if (nowp != NULL)
1338 : : {
1339 : 1070 : char datebuf[80];
1340 : 1070 : size_t rc = strftime (datebuf, sizeof (datebuf), "%a, %d %b %Y %T GMT",
1341 : : nowp);
1342 [ + - ]: 1070 : if (rc > 0 && rc < sizeof (datebuf))
1343 : 1070 : add_mhd_response_header (resp, "Last-Modified", datebuf);
1344 : : }
1345 : :
1346 : 1070 : add_mhd_response_header (resp, "Cache-Control", "public");
1347 : 1070 : }
1348 : :
1349 : : // quote all questionable characters of str for safe passage through a sh -c expansion.
1350 : : static string
1351 : 64 : shell_escape(const string& str)
1352 : : {
1353 : 64 : string y;
1354 [ + + ]: 8318 : for (auto&& x : str)
1355 : : {
1356 [ + + + + ]: 8254 : if (! isalnum(x) && x != '/')
1357 [ + - ]: 650 : y += "\\";
1358 [ + - ]: 16508 : y += x;
1359 : : }
1360 : 64 : return y;
1361 : 0 : }
1362 : :
1363 : :
1364 : : // PR25548: Perform POSIX / RFC3986 style path canonicalization on the input string.
1365 : : //
1366 : : // Namely:
1367 : : // // -> /
1368 : : // /foo/../ -> /
1369 : : // /./ -> /
1370 : : //
1371 : : // This mapping is done on dwarf-side source path names, which may
1372 : : // include these constructs, so we can deal with debuginfod clients
1373 : : // that accidentally canonicalize the paths.
1374 : : //
1375 : : // realpath(3) is close but not quite right, because it also resolves
1376 : : // symbolic links. Symlinks at the debuginfod server have nothing to
1377 : : // do with the build-time symlinks, thus they must not be considered.
1378 : : //
1379 : : // see also curl Curl_dedotdotify() aka RFC3986, which we mostly follow here
1380 : : // see also libc __realpath()
1381 : : // see also llvm llvm::sys::path::remove_dots()
1382 : : static string
1383 : 3702 : canon_pathname (const string& input)
1384 : : {
1385 : 3702 : string i = input; // 5.2.4 (1)
1386 : 3702 : string o;
1387 : :
1388 : 34740 : while (i.size() != 0)
1389 : : {
1390 : : // 5.2.4 (2) A
1391 [ + - - + : 62076 : if (i.substr(0,3) == "../")
- + ]
1392 [ # # # # ]: 0 : i = i.substr(3);
1393 [ + - - + : 62076 : else if(i.substr(0,2) == "./")
- + ]
1394 [ # # # # ]: 0 : i = i.substr(2);
1395 : :
1396 : : // 5.2.4 (2) B
1397 [ + - - + : 62076 : else if (i.substr(0,3) == "/./")
+ + ]
1398 [ + - + + ]: 618 : i = i.substr(2);
1399 [ - + ]: 30660 : else if (i == "/.")
1400 [ # # ]: 0 : i = ""; // no need to handle "/." complete-path-segment case; we're dealing with file names
1401 : :
1402 : : // 5.2.4 (2) C
1403 [ + - - + : 61320 : else if (i.substr(0,4) == "/../") {
+ + ]
1404 [ + - + + ]: 522 : i = i.substr(3);
1405 : 522 : string::size_type sl = o.rfind("/");
1406 [ + - ]: 522 : if (sl != string::npos)
1407 [ + - + - ]: 1044 : o = o.substr(0, sl);
1408 : : else
1409 [ # # ]: 0 : o = "";
1410 [ - + ]: 30138 : } else if (i == "/..")
1411 [ # # ]: 0 : i = ""; // no need to handle "/.." complete-path-segment case; we're dealing with file names
1412 : :
1413 : : // 5.2.4 (2) D
1414 : : // no need to handle these cases; we're dealing with file names
1415 [ - + ]: 30138 : else if (i == ".")
1416 [ # # ]: 0 : i = "";
1417 [ - + ]: 30138 : else if (i == "..")
1418 [ # # ]: 0 : i = "";
1419 : :
1420 : : // POSIX special: map // to /
1421 [ + - - + : 60276 : else if (i.substr(0,2) == "//")
+ + ]
1422 [ + - + + ]: 144 : i = i.substr(1);
1423 : :
1424 : : // 5.2.4 (2) E
1425 : : else {
1426 [ - + ]: 30010 : string::size_type next_slash = i.find("/", (i[0]=='/' ? 1 : 0)); // skip first slash
1427 [ + - + + ]: 60020 : o += i.substr(0, next_slash);
1428 [ + + ]: 30010 : if (next_slash == string::npos)
1429 [ + - + + ]: 38442 : i = "";
1430 : : else
1431 [ + - + + : 49604 : i = i.substr(next_slash);
- - ]
1432 : : }
1433 : : }
1434 : :
1435 [ + - ]: 3702 : return o;
1436 : 3702 : }
1437 : :
1438 : :
1439 : : // Estimate available free space for a given filesystem via statfs(2).
1440 : : // Return true if the free fraction is known to be smaller than the
1441 : : // given minimum percentage. Also update a related metric.
1442 : 2338 : bool statfs_free_enough_p(const string& path, const string& label, long minfree = 0)
1443 : : {
1444 : 2338 : struct statfs sfs;
1445 : 2338 : int rc = statfs(path.c_str(), &sfs);
1446 [ + + ]: 2338 : if (rc == 0)
1447 : : {
1448 : 2288 : double s = (double) sfs.f_bavail / (double) sfs.f_blocks;
1449 [ + - + - : 4576 : set_metric("filesys_free_ratio","purpose",label, s);
- + - - ]
1450 : 2288 : return ((s * 100.0) < minfree);
1451 : : }
1452 : : return false;
1453 : : }
1454 : :
1455 : :
1456 : :
1457 : : // A map-like class that owns a cache of file descriptors (indexed by
1458 : : // file / content names).
1459 : : //
1460 : : // If only it could use fd's instead of file names ... but we can't
1461 : : // dup(2) to create independent descriptors for the same unlinked
1462 : : // files, so would have to use some goofy linux /proc/self/fd/%d
1463 : : // hack such as the following
1464 : :
1465 : : #if 0
1466 : : int superdup(int fd)
1467 : : {
1468 : : #ifdef __linux__
1469 : : char *fdpath = NULL;
1470 : : int rc = asprintf(& fdpath, "/proc/self/fd/%d", fd);
1471 : : int newfd;
1472 : : if (rc >= 0)
1473 : : newfd = open(fdpath, O_RDONLY);
1474 : : else
1475 : : newfd = -1;
1476 : : free (fdpath);
1477 : : return newfd;
1478 : : #else
1479 : : return -1;
1480 : : #endif
1481 : : }
1482 : : #endif
1483 : :
1484 : : class libarchive_fdcache
1485 : : {
1486 : : private:
1487 : : mutex fdcache_lock;
1488 : :
1489 : : typedef pair<string,string> key; // archive, entry
1490 [ + - ]: 276 : struct fdcache_entry
1491 : : {
1492 : : string fd; // file name (probably in $TMPDIR), not an actual open fd (EMFILE)
1493 : : double fd_size_mb; // slightly rounded up megabytes
1494 : : time_t freshness; // when was this entry created or requested last
1495 : : unsigned request_count; // how many requests were made; or 0=prefetch only
1496 : : double latency; // how many seconds it took to extract the file
1497 : : };
1498 : :
1499 : : map<key,fdcache_entry> entries; // optimized for lookup
1500 : : time_t last_cleaning;
1501 : : long max_mbs;
1502 : :
1503 : : public:
1504 : 352 : void set_metrics()
1505 : : {
1506 : 352 : double fdcache_mb = 0.0;
1507 : 352 : double prefetch_mb = 0.0;
1508 : 352 : unsigned fdcache_count = 0;
1509 : 352 : unsigned prefetch_count = 0;
1510 [ + + ]: 3452 : for (auto &i : entries) {
1511 [ + + ]: 3100 : if (i.second.request_count) {
1512 : 3010 : fdcache_mb += i.second.fd_size_mb;
1513 : 3010 : fdcache_count ++;
1514 : : } else {
1515 : 90 : prefetch_mb += i.second.fd_size_mb;
1516 : 90 : prefetch_count ++;
1517 : : }
1518 : : }
1519 [ + - ]: 352 : set_metric("fdcache_bytes", fdcache_mb*1024.0*1024.0);
1520 [ + - ]: 352 : set_metric("fdcache_count", fdcache_count);
1521 [ + - ]: 352 : set_metric("fdcache_prefetch_bytes", prefetch_mb*1024.0*1024.0);
1522 [ + - ]: 352 : set_metric("fdcache_prefetch_count", prefetch_count);
1523 : 352 : }
1524 : :
1525 : 278 : void intern(const string& a, const string& b, string fd, off_t sz,
1526 : : bool requested_p, double lat)
1527 : : {
1528 : 278 : {
1529 : 278 : unique_lock<mutex> lock(fdcache_lock);
1530 : 278 : time_t now = time(NULL);
1531 : : // there is a chance it's already in here, just wasn't found last time
1532 : : // if so, there's nothing to do but count our luck
1533 [ + - ]: 278 : auto i = entries.find(make_pair(a,b));
1534 [ + + ]: 278 : if (i != entries.end())
1535 : : {
1536 [ + - + - : 4 : inc_metric("fdcache_op_count","op","redundant_intern");
+ - + - -
+ + - - -
- - ]
1537 [ + - ]: 2 : if (requested_p) i->second.request_count ++; // repeat prefetch doesn't count
1538 : 2 : i->second.freshness = now;
1539 : : // We need to nuke the temp file, since interning passes
1540 : : // responsibility over the path to this structure. It is
1541 : : // possible that the caller still has an fd open, but that's
1542 : : // OK.
1543 : 2 : unlink (fd.c_str());
1544 : 2 : return;
1545 : : }
1546 : 276 : double mb = (sz+65535)/1048576.0; // round up to 64K block
1547 : 276 : fdcache_entry n = { .fd=fd, .fd_size_mb=mb,
1548 : 276 : .freshness=now, .request_count = requested_p?1U:0U,
1549 [ + - + + ]: 276 : .latency=lat};
1550 [ + - + - : 276 : entries.insert(make_pair(make_pair(a,b),n));
+ - ]
1551 : :
1552 [ + + ]: 276 : if (requested_p)
1553 [ + - + - : 464 : inc_metric("fdcache_op_count","op","enqueue");
+ - + - -
+ - + - -
- - ]
1554 : : else
1555 [ + - + - : 132 : inc_metric("fdcache_op_count","op","prefetch_enqueue");
+ - + - -
+ + - - -
- - ]
1556 : :
1557 [ + + ]: 276 : if (verbose > 3)
1558 [ + - + - : 486 : obatched(clog) << "fdcache interned a=" << a << " b=" << b
- - ]
1559 [ + - + - : 162 : << " fd=" << fd << " mb=" << mb << " front=" << requested_p
+ - + - +
- + - + -
+ - ]
1560 [ + - + - : 162 : << " latency=" << lat << endl;
+ - ]
1561 : :
1562 [ + - ]: 276 : set_metrics();
1563 : 278 : }
1564 : :
1565 : : // NB: we age the cache at lookup time too
1566 [ + - - + : 276 : if (statfs_free_enough_p(tmpdir, "tmpdir", fdcache_mintmp))
- + ]
1567 : : {
1568 [ # # # # : 0 : inc_metric("fdcache_op_count","op","emerg-flush");
# # # # #
# # # #
# ]
1569 [ # # ]: 0 : obatched(clog) << "fdcache emergency flush for filling tmpdir" << endl;
1570 : 0 : this->limit(0); // emergency flush
1571 : : }
1572 : : else // age cache normally
1573 : 276 : this->limit(max_mbs);
1574 : : }
1575 : :
1576 : 1010 : int lookup(const string& a, const string& b)
1577 : : {
1578 : 1010 : int fd = -1;
1579 : 1010 : {
1580 : 1010 : unique_lock<mutex> lock(fdcache_lock);
1581 [ + - ]: 1010 : auto i = entries.find(make_pair(a,b));
1582 [ + + ]: 1010 : if (i != entries.end())
1583 : : {
1584 [ + + ]: 772 : if (i->second.request_count == 0) // was a prefetch!
1585 : : {
1586 [ + - + - ]: 12 : inc_metric("fdcache_prefetch_saved_milliseconds_count");
1587 [ + - + - ]: 24 : add_metric("fdcache_prefetch_saved_milliseconds_sum", i->second.latency*1000.);
1588 : : }
1589 : 772 : i->second.request_count ++;
1590 : 772 : i->second.freshness = time(NULL);
1591 : : // brag about our success
1592 [ + - + - : 1544 : inc_metric("fdcache_op_count","op","prefetch_access"); // backward compat
+ - + - -
+ - + - -
- - ]
1593 [ + - + - ]: 772 : inc_metric("fdcache_saved_milliseconds_count");
1594 [ + - + - ]: 772 : add_metric("fdcache_saved_milliseconds_sum", i->second.latency*1000.);
1595 [ + - ]: 1010 : fd = open(i->second.fd.c_str(), O_RDONLY);
1596 : : }
1597 : 0 : }
1598 : :
1599 [ + + ]: 1010 : if (fd >= 0)
1600 [ + - + - : 1544 : inc_metric("fdcache_op_count","op","lookup_hit");
+ - - + -
+ - - -
- ]
1601 : : else
1602 [ + - + - : 476 : inc_metric("fdcache_op_count","op","lookup_miss");
+ - - + -
+ - - -
- ]
1603 : :
1604 : : // NB: no need to age the cache after just a lookup
1605 : :
1606 : 1010 : return fd;
1607 : : }
1608 : :
1609 : 198 : int probe(const string& a, const string& b) // just a cache residency check - don't modify state, don't open
1610 : : {
1611 : 198 : unique_lock<mutex> lock(fdcache_lock);
1612 [ + - ]: 198 : auto i = entries.find(make_pair(a,b));
1613 [ + + ]: 198 : if (i != entries.end()) {
1614 [ + - + - : 48 : inc_metric("fdcache_op_count","op","probe_hit");
+ - + - -
+ - + - -
- - ]
1615 : 24 : return true;
1616 : : } else {
1617 [ + - + - : 348 : inc_metric("fdcache_op_count","op","probe_miss");
+ - + - -
+ - + - -
- - ]
1618 : 174 : return false;
1619 : : }
1620 : 198 : }
1621 : :
1622 : 0 : void clear(const string& a, const string& b)
1623 : : {
1624 : 0 : unique_lock<mutex> lock(fdcache_lock);
1625 [ # # ]: 0 : auto i = entries.find(make_pair(a,b));
1626 [ # # ]: 0 : if (i != entries.end()) {
1627 [ # # # # : 0 : inc_metric("fdcache_op_count","op",
# # # # #
# # # # #
# # # # ]
1628 [ # # ]: 0 : i->second.request_count > 0 ? "clear" : "prefetch_clear");
1629 : 0 : unlink (i->second.fd.c_str());
1630 : 0 : entries.erase(i);
1631 [ # # ]: 0 : set_metrics();
1632 : 0 : return;
1633 : : }
1634 : 0 : }
1635 : :
1636 : 428 : void limit(long maxmbs, bool metrics_p = true)
1637 : : {
1638 : 428 : time_t now = time(NULL);
1639 : :
1640 : : // avoid overly frequent limit operations
1641 [ + + + + ]: 428 : if (maxmbs > 0 && (now - this->last_cleaning) < 10) // probably not worth parametrizing
1642 : 276 : return;
1643 : 152 : this->last_cleaning = now;
1644 : :
1645 [ + + + - ]: 152 : if (verbose > 3 && (this->max_mbs != maxmbs))
1646 [ + - + - ]: 176 : obatched(clog) << "fdcache limited to maxmbs=" << maxmbs << endl;
1647 : :
1648 : 152 : unique_lock<mutex> lock(fdcache_lock);
1649 : :
1650 : 152 : this->max_mbs = maxmbs;
1651 : 152 : double total_mb = 0.0;
1652 : :
1653 : 152 : map<double, pair<string,string>> sorted_entries;
1654 [ + + ]: 428 : for (auto &i: entries)
1655 : : {
1656 : 276 : total_mb += i.second.fd_size_mb;
1657 : :
1658 : : // need a scalar quantity that combines these inputs in a sensible way:
1659 : : //
1660 : : // 1) freshness of this entry (last time it was accessed)
1661 : : // 2) size of this entry
1662 : : // 3) number of times it has been accessed (or if just prefetched with 0 accesses)
1663 : : // 4) latency it required to extract
1664 : : //
1665 : : // The lower the "score", the earlier garbage collection will
1666 : : // nuke it, so to prioritize entries for preservation, the
1667 : : // score should be higher, and vice versa.
1668 : 276 : time_t factor_1_freshness = (now - i.second.freshness); // seconds
1669 : 276 : double factor_2_size = i.second.fd_size_mb; // megabytes
1670 : 276 : unsigned factor_3_accesscount = i.second.request_count; // units
1671 : 276 : double factor_4_latency = i.second.latency; // seconds
1672 : :
1673 : : #if 0
1674 : : double score = - factor_1_freshness; // simple LRU
1675 : : #endif
1676 : :
1677 [ + + ]: 276 : double score = 0.
1678 : 276 : - log1p(factor_1_freshness) // penalize old file
1679 : 276 : - log1p(factor_2_size) // penalize large file
1680 : 276 : + factor_4_latency * factor_3_accesscount; // reward slow + repeatedly read files
1681 : :
1682 [ + + ]: 276 : if (verbose > 4)
1683 [ + - ]: 8 : obatched(clog) << "fdcache scored score=" << score
1684 [ + - + - ]: 8 : << " a=" << i.first.first << " b=" << i.first.second
1685 [ + - + - : 12 : << " f1=" << factor_1_freshness << " f2=" << factor_2_size
+ - + - +
- + - +
- ]
1686 [ + - + - : 4 : << " f3=" << factor_3_accesscount << " f4=" << factor_4_latency
+ - + - +
- ]
1687 : 4 : << endl;
1688 : :
1689 [ + - + - ]: 552 : sorted_entries.insert(make_pair(score, i.first));
1690 : : }
1691 : :
1692 : 152 : unsigned cleaned = 0;
1693 : 152 : unsigned entries_original = entries.size();
1694 : 152 : double cleaned_score_min = DBL_MAX;
1695 : 152 : double cleaned_score_max = DBL_MIN;
1696 : :
1697 : : // drop as many entries[] as needed to bring total mb down to the threshold
1698 [ + + ]: 428 : for (auto &i: sorted_entries) // in increasing score order!
1699 : : {
1700 [ - + ]: 276 : if (this->max_mbs > 0 // if this is not a "clear entire table"
1701 [ # # ]: 0 : && total_mb < this->max_mbs) // we've cleared enough to meet threshold
1702 : : break; // stop clearing
1703 : :
1704 [ - + ]: 276 : auto j = entries.find(i.second);
1705 [ - + ]: 276 : if (j == entries.end())
1706 : 0 : continue; // should not happen
1707 : :
1708 [ + + ]: 276 : if (cleaned == 0)
1709 : 34 : cleaned_score_min = i.first;
1710 : 276 : cleaned++;
1711 : 276 : cleaned_score_max = i.first;
1712 : :
1713 [ + + ]: 276 : if (verbose > 3)
1714 [ + - + - ]: 486 : obatched(clog) << "fdcache evicted score=" << i.first
1715 [ + - + - ]: 324 : << " a=" << i.second.first << " b=" << i.second.second
1716 [ + - + - : 486 : << " fd=" << j->second.fd << " mb=" << j->second.fd_size_mb
+ - + - +
- + - ]
1717 [ + - + - : 162 : << " rq=" << j->second.request_count << " lat=" << j->second.latency
+ - + - ]
1718 [ + - + - : 162 : << " fr=" << (now - j->second.freshness)
+ - ]
1719 : 162 : << endl;
1720 [ - + ]: 276 : if (metrics_p)
1721 [ # # # # : 0 : inc_metric("fdcache_op_count","op","evict");
# # # # #
# # # # #
# # ]
1722 : :
1723 : 276 : total_mb -= j->second.fd_size_mb;
1724 : 276 : unlink (j->second.fd.c_str());
1725 : 276 : entries.erase(j);
1726 : : }
1727 : :
1728 [ + + ]: 152 : if (metrics_p)
1729 [ + - + - : 152 : inc_metric("fdcache_op_count","op","evict_cycle");
+ - + - -
+ - + - -
- - ]
1730 : :
1731 [ + - + + ]: 152 : if (verbose > 1 && cleaned > 0)
1732 : : {
1733 [ + - + - : 102 : obatched(clog) << "fdcache evicted num=" << cleaned << " of=" << entries_original
+ - + - ]
1734 [ + - + - : 34 : << " min=" << cleaned_score_min << " max=" << cleaned_score_max
+ - + - +
- ]
1735 : 34 : << endl;
1736 : : }
1737 : :
1738 [ + + + - ]: 152 : if (metrics_p) set_metrics();
1739 : 152 : }
1740 : :
1741 : :
1742 : 76 : ~libarchive_fdcache()
1743 : : {
1744 : : // unlink any fdcache entries in $TMPDIR
1745 : : // don't update metrics; those globals may be already destroyed
1746 : 76 : limit(0, false);
1747 : 76 : }
1748 : : };
1749 : : static libarchive_fdcache fdcache;
1750 : :
1751 : : /* Search ELF_FD for an ELF/DWARF section with name SECTION.
1752 : : If found copy the section to a temporary file and return
1753 : : its file descriptor, otherwise return -1.
1754 : :
1755 : : The temporary file's mtime will be set to PARENT_MTIME.
1756 : : B_SOURCE should be a description of the parent file suitable
1757 : : for printing to the log. */
1758 : :
1759 : : static int
1760 : 14 : extract_section (int elf_fd, int64_t parent_mtime,
1761 : : const string& b_source, const string& section,
1762 : : const timespec& extract_begin)
1763 : : {
1764 : : /* Search the fdcache. */
1765 : 14 : struct stat fs;
1766 : 14 : int fd = fdcache.lookup (b_source, section);
1767 [ - + ]: 14 : if (fd >= 0)
1768 : : {
1769 [ # # ]: 0 : if (fstat (fd, &fs) != 0)
1770 : : {
1771 [ # # ]: 0 : if (verbose)
1772 [ # # ]: 0 : obatched (clog) << "cannot fstate fdcache "
1773 [ # # # # : 0 : << b_source << " " << section << endl;
# # ]
1774 : 0 : close (fd);
1775 : 0 : return -1;
1776 : : }
1777 [ # # ]: 0 : if ((int64_t) fs.st_mtime != parent_mtime)
1778 : : {
1779 [ # # ]: 0 : if (verbose)
1780 [ # # ]: 0 : obatched(clog) << "mtime mismatch for "
1781 [ # # # # : 0 : << b_source << " " << section << endl;
# # ]
1782 : 0 : close (fd);
1783 : 0 : return -1;
1784 : : }
1785 : : /* Success. */
1786 : : return fd;
1787 : : }
1788 : :
1789 : 14 : Elf *elf = elf_begin (elf_fd, ELF_C_READ_MMAP_PRIVATE, NULL);
1790 [ - + ]: 14 : if (elf == NULL)
1791 : : return -1;
1792 : :
1793 : : /* Try to find the section and copy the contents into a separate file. */
1794 : 14 : try
1795 : : {
1796 : 14 : size_t shstrndx;
1797 [ + - ]: 14 : int rc = elf_getshdrstrndx (elf, &shstrndx);
1798 [ - + ]: 14 : if (rc < 0)
1799 [ # # # # ]: 0 : throw elfutils_exception (rc, "getshdrstrndx");
1800 : :
1801 : : Elf_Scn *scn = NULL;
1802 : 482 : while (true)
1803 : : {
1804 [ + - ]: 248 : scn = elf_nextscn (elf, scn);
1805 [ + - ]: 248 : if (scn == NULL)
1806 : : break;
1807 : 248 : GElf_Shdr shdr_storage;
1808 [ + - ]: 248 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_storage);
1809 [ + - ]: 248 : if (shdr == NULL)
1810 : : break;
1811 : :
1812 [ + - ]: 248 : const char *scn_name = elf_strptr (elf, shstrndx, shdr->sh_name);
1813 [ + - ]: 248 : if (scn_name == NULL)
1814 : : break;
1815 [ + + ]: 248 : if (scn_name == section)
1816 : : {
1817 : 14 : Elf_Data *data = NULL;
1818 : :
1819 : : /* We found the desired section. */
1820 [ + - ]: 14 : data = elf_rawdata (scn, NULL);
1821 [ - + ]: 14 : if (data == NULL)
1822 [ # # # # : 0 : throw elfutils_exception (elf_errno (), "elfraw_data");
# # ]
1823 [ + + ]: 14 : if (data->d_buf == NULL)
1824 : : {
1825 [ + - + - ]: 12 : obatched(clog) << "section " << section
1826 [ + - + - ]: 6 : << " is empty" << endl;
1827 : 6 : break;
1828 : : }
1829 : :
1830 : : /* Create temporary file containing the section. */
1831 : 8 : char *tmppath = NULL;
1832 : 8 : rc = asprintf (&tmppath, "%s/debuginfod-section.XXXXXX", tmpdir.c_str());
1833 [ - + ]: 8 : if (rc < 0)
1834 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
1835 : 8 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
1836 [ + - ]: 8 : fd = mkstemp (tmppath);
1837 [ - + ]: 8 : if (fd < 0)
1838 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
1839 : :
1840 [ + - ]: 8 : ssize_t res = write_retry (fd, data->d_buf, data->d_size);
1841 [ + - - + ]: 8 : if (res < 0 || (size_t) res != data->d_size) {
1842 [ # # ]: 0 : close (fd);
1843 : 0 : unlink (tmppath);
1844 [ # # # # ]: 0 : throw libc_exception (errno, "cannot write to temporary file");
1845 : : }
1846 : :
1847 : : /* Set mtime to be the same as the parent file's mtime. */
1848 : 8 : struct timespec tvs[2];
1849 [ - + ]: 8 : if (fstat (elf_fd, &fs) != 0) {
1850 [ # # ]: 0 : close (fd);
1851 : 0 : unlink (tmppath);
1852 [ # # # # ]: 0 : throw libc_exception (errno, "cannot fstat file");
1853 : : }
1854 : :
1855 : 8 : tvs[0].tv_sec = 0;
1856 : 8 : tvs[0].tv_nsec = UTIME_OMIT;
1857 : 8 : tvs[1] = fs.st_mtim;
1858 : 8 : (void) futimens (fd, tvs);
1859 : :
1860 : 8 : struct timespec extract_end;
1861 : 8 : clock_gettime (CLOCK_MONOTONIC, &extract_end);
1862 : 8 : double extract_time = (extract_end.tv_sec - extract_begin.tv_sec)
1863 : 8 : + (extract_end.tv_nsec - extract_begin.tv_nsec)/1.e9;
1864 : :
1865 : : /* Add to fdcache. */
1866 [ + - + - ]: 8 : fdcache.intern (b_source, section, tmppath, data->d_size, true, extract_time);
1867 : 8 : break;
1868 : 14 : }
1869 : 234 : }
1870 : : }
1871 [ - - ]: 0 : catch (const reportable_exception &e)
1872 : : {
1873 [ - - ]: 0 : e.report (clog);
1874 [ - - ]: 0 : close (fd);
1875 : 0 : fd = -1;
1876 : 0 : }
1877 : :
1878 : 14 : elf_end (elf);
1879 : : return fd;
1880 : : }
1881 : :
1882 : : static struct MHD_Response*
1883 : 74 : handle_buildid_f_match (bool internal_req_t,
1884 : : int64_t b_mtime,
1885 : : const string& b_source0,
1886 : : const string& section,
1887 : : int *result_fd)
1888 : : {
1889 : 74 : (void) internal_req_t; // ignored
1890 : :
1891 : 74 : struct timespec extract_begin;
1892 : 74 : clock_gettime (CLOCK_MONOTONIC, &extract_begin);
1893 : :
1894 : 74 : int fd = open(b_source0.c_str(), O_RDONLY);
1895 [ - + ]: 74 : if (fd < 0)
1896 [ # # # # : 0 : throw libc_exception (errno, string("open ") + b_source0);
# # # # ]
1897 : :
1898 : : // NB: use manual close(2) in error case instead of defer_dtor, because
1899 : : // in the normal case, we want to hand the fd over to libmicrohttpd for
1900 : : // file transfer.
1901 : :
1902 : 74 : struct stat s;
1903 : 74 : int rc = fstat(fd, &s);
1904 [ - + ]: 74 : if (rc < 0)
1905 : : {
1906 : 0 : close(fd);
1907 [ # # # # : 0 : throw libc_exception (errno, string("fstat ") + b_source0);
# # # # ]
1908 : : }
1909 : :
1910 [ - + ]: 74 : if ((int64_t) s.st_mtime != b_mtime)
1911 : : {
1912 [ # # ]: 0 : if (verbose)
1913 [ # # # # ]: 0 : obatched(clog) << "mtime mismatch for " << b_source0 << endl;
1914 : 0 : close(fd);
1915 : 0 : return 0;
1916 : : }
1917 : :
1918 [ + + ]: 74 : if (!section.empty ())
1919 : : {
1920 : 6 : int scn_fd = extract_section (fd, s.st_mtime, b_source0, section, extract_begin);
1921 : 6 : close (fd);
1922 : :
1923 [ + + ]: 6 : if (scn_fd >= 0)
1924 : 4 : fd = scn_fd;
1925 : : else
1926 : : {
1927 [ + - ]: 2 : if (verbose)
1928 [ + - ]: 6 : obatched (clog) << "cannot find section " << section
1929 [ + - + - : 2 : << " for " << b_source0 << endl;
+ - ]
1930 : 2 : return 0;
1931 : : }
1932 : :
1933 : 4 : rc = fstat(fd, &s);
1934 [ - + ]: 4 : if (rc < 0)
1935 : : {
1936 : 0 : close (fd);
1937 [ # # # # : 0 : throw libc_exception (errno, string ("fstat ") + b_source0
# # # # #
# # # #
# ]
1938 [ # # # # : 0 : + string (" ") + section);
# # # # #
# ]
1939 : : }
1940 : : }
1941 : :
1942 : 72 : struct MHD_Response* r = MHD_create_response_from_fd ((uint64_t) s.st_size, fd);
1943 [ + - + - : 144 : inc_metric ("http_responses_total","result","file");
+ - - + -
+ - - -
- ]
1944 [ - + ]: 72 : if (r == 0)
1945 : : {
1946 [ # # ]: 0 : if (verbose)
1947 [ # # ]: 0 : obatched(clog) << "cannot create fd-response for " << b_source0
1948 [ # # # # : 0 : << " section=" << section << endl;
# # ]
1949 : 0 : close(fd);
1950 : : }
1951 : : else
1952 : : {
1953 : 72 : add_mhd_response_header (r, "Content-Type", "application/octet-stream");
1954 [ + - ]: 72 : add_mhd_response_header (r, "X-DEBUGINFOD-SIZE",
1955 : 72 : to_string(s.st_size).c_str());
1956 : 72 : add_mhd_response_header (r, "X-DEBUGINFOD-FILE", b_source0.c_str());
1957 : 72 : add_mhd_last_modified (r, s.st_mtime);
1958 [ + - ]: 72 : if (verbose > 1)
1959 [ + - + - : 144 : obatched(clog) << "serving file " << b_source0 << " section=" << section << endl;
+ - + - ]
1960 : : /* libmicrohttpd will close it. */
1961 [ - + ]: 72 : if (result_fd)
1962 : 72 : *result_fd = fd;
1963 : : }
1964 : :
1965 : : return r;
1966 : : }
1967 : :
1968 : :
1969 : : #ifdef USE_LZMA
1970 : : struct lzma_exception: public reportable_exception
1971 : : {
1972 : 0 : lzma_exception(int rc, const string& msg):
1973 : : // liblzma doesn't have a lzma_ret -> string conversion function, so just
1974 : : // report the value.
1975 [ # # # # : 0 : reportable_exception(string ("lzma error: ") + msg + ": error " + to_string(rc)) {
# # # # #
# # # # #
# # # # ]
1976 [ # # # # : 0 : inc_metric("error_count","lzma",to_string(rc));
# # # # #
# # # #
# ]
1977 [ # # ]: 0 : }
1978 : : };
1979 : :
1980 : : // Neither RPM nor deb files support seeking to a specific file in the package.
1981 : : // Instead, to extract a specific file, we normally need to read the archive
1982 : : // sequentially until we find the file. This is very slow for files at the end
1983 : : // of a large package with lots of files, like kernel debuginfo.
1984 : : //
1985 : : // However, if the compression format used in the archive supports seeking, we
1986 : : // can accelerate this. As of July 2024, xz is the only widely-used format that
1987 : : // supports seeking, and usually only in multi-threaded mode. Luckily, the
1988 : : // kernel-debuginfo package in Fedora and its downstreams, and the
1989 : : // linux-image-*-dbg package in Debian and its downstreams, all happen to use
1990 : : // this.
1991 : : //
1992 : : // The xz format [1] ends with an index of independently compressed blocks in
1993 : : // the stream. In RPM and deb files, the xz stream is the last thing in the
1994 : : // file, so we assume that the xz Stream Footer is at the end of the package
1995 : : // file and do everything relative to that. For each file in the archive, we
1996 : : // remember the size and offset of the file data in the uncompressed xz stream,
1997 : : // then we use the index to seek to that offset when we need that file.
1998 : : //
1999 : : // 1: https://xz.tukaani.org/format/xz-file-format.txt
2000 : :
2001 : : // Return whether an archive supports seeking.
2002 : : static bool
2003 : 524 : is_seekable_archive (const string& rps, struct archive* a)
2004 : : {
2005 : : // Only xz supports seeking.
2006 [ + + ]: 524 : if (archive_filter_code (a, 0) != ARCHIVE_FILTER_XZ)
2007 : : return false;
2008 : :
2009 : 336 : int fd = open (rps.c_str(), O_RDONLY);
2010 [ - + ]: 336 : if (fd < 0)
2011 : : return false;
2012 : 336 : defer_dtor<int,int> fd_closer (fd, close);
2013 : :
2014 : : // Seek to the xz Stream Footer. We assume that it's the last thing in the
2015 : : // file, which is true for RPM and deb files.
2016 : 336 : off_t footer_pos = -LZMA_STREAM_HEADER_SIZE;
2017 [ - + ]: 336 : if (lseek (fd, footer_pos, SEEK_END) == -1)
2018 : : return false;
2019 : :
2020 : : // Decode the Stream Footer.
2021 : : uint8_t footer[LZMA_STREAM_HEADER_SIZE];
2022 : : size_t footer_read = 0;
2023 [ + + ]: 672 : while (footer_read < sizeof (footer))
2024 : : {
2025 [ - + ]: 336 : ssize_t bytes_read = read (fd, footer + footer_read,
2026 : : sizeof (footer) - footer_read);
2027 [ - + ]: 336 : if (bytes_read < 0)
2028 : : {
2029 [ # # ]: 0 : if (errno == EINTR)
2030 : 0 : continue;
2031 : : return false;
2032 : : }
2033 [ - + ]: 336 : if (bytes_read == 0)
2034 : : return false;
2035 : 336 : footer_read += bytes_read;
2036 : : }
2037 : :
2038 : 336 : lzma_stream_flags stream_flags;
2039 : 336 : lzma_ret ret = lzma_stream_footer_decode (&stream_flags, footer);
2040 [ - + ]: 336 : if (ret != LZMA_OK)
2041 : : return false;
2042 : :
2043 : : // Seek to the xz Index.
2044 [ - + ]: 336 : if (lseek (fd, footer_pos - stream_flags.backward_size, SEEK_END) == -1)
2045 : : return false;
2046 : :
2047 : : // Decode the Number of Records in the Index. liblzma doesn't have an API for
2048 : : // this if you don't want to decode the whole Index, so we have to do it
2049 : : // ourselves.
2050 : : //
2051 : : // We need 1 byte for the Index Indicator plus 1-9 bytes for the
2052 : : // variable-length integer Number of Records.
2053 : : uint8_t index[10];
2054 : : size_t index_read = 0;
2055 [ + + ]: 672 : while (index_read == 0) {
2056 [ + - ]: 336 : ssize_t bytes_read = read (fd, index, sizeof (index));
2057 [ - + ]: 336 : if (bytes_read < 0)
2058 : : {
2059 [ # # ]: 0 : if (errno == EINTR)
2060 : 0 : continue;
2061 : : return false;
2062 : : }
2063 [ - + ]: 336 : if (bytes_read == 0)
2064 : : return false;
2065 : 336 : index_read += bytes_read;
2066 : : }
2067 : : // The Index Indicator must be 0.
2068 [ - + ]: 336 : if (index[0] != 0)
2069 : : return false;
2070 : :
2071 : 336 : lzma_vli num_records;
2072 : 336 : size_t pos = 0;
2073 : 336 : size_t in_pos = 1;
2074 : 336 : while (true)
2075 : : {
2076 [ - + ]: 336 : if (in_pos >= index_read)
2077 : : {
2078 [ # # ]: 0 : ssize_t bytes_read = read (fd, index, sizeof (index));
2079 [ # # ]: 0 : if (bytes_read < 0)
2080 : : {
2081 [ # # ]: 0 : if (errno == EINTR)
2082 : 0 : continue;
2083 : : return false;
2084 : : }
2085 [ # # ]: 0 : if (bytes_read == 0)
2086 : : return false;
2087 : 0 : index_read = bytes_read;
2088 : 0 : in_pos = 0;
2089 : : }
2090 : 336 : ret = lzma_vli_decode (&num_records, &pos, index, &in_pos, index_read);
2091 [ - + ]: 336 : if (ret == LZMA_STREAM_END)
2092 : : break;
2093 [ # # ]: 0 : else if (ret != LZMA_OK)
2094 : : return false;
2095 : : }
2096 : :
2097 [ + + ]: 336 : if (verbose > 3)
2098 [ + - + - : 572 : obatched(clog) << rps << " has " << num_records << " xz Blocks" << endl;
+ - + - +
- ]
2099 : :
2100 : : // The file is only seekable if it has more than one Block.
2101 : 336 : return num_records > 1;
2102 : 524 : }
2103 : :
2104 : : // Read the Index at the end of an xz file.
2105 : : static lzma_index*
2106 : 94 : read_xz_index (int fd)
2107 : : {
2108 : 94 : off_t footer_pos = -LZMA_STREAM_HEADER_SIZE;
2109 [ - + ]: 94 : if (lseek (fd, footer_pos, SEEK_END) == -1)
2110 [ # # # # ]: 0 : throw libc_exception (errno, "lseek");
2111 : :
2112 : : uint8_t footer[LZMA_STREAM_HEADER_SIZE];
2113 : : size_t footer_read = 0;
2114 [ + + ]: 188 : while (footer_read < sizeof (footer))
2115 : : {
2116 [ - + ]: 94 : ssize_t bytes_read = read (fd, footer + footer_read,
2117 : : sizeof (footer) - footer_read);
2118 [ - + ]: 94 : if (bytes_read < 0)
2119 : : {
2120 [ # # ]: 0 : if (errno == EINTR)
2121 : 0 : continue;
2122 [ # # # # ]: 0 : throw libc_exception (errno, "read");
2123 : : }
2124 [ - + ]: 94 : if (bytes_read == 0)
2125 [ # # # # ]: 0 : throw reportable_exception ("truncated file");
2126 : 94 : footer_read += bytes_read;
2127 : : }
2128 : :
2129 : 94 : lzma_stream_flags stream_flags;
2130 : 94 : lzma_ret ret = lzma_stream_footer_decode (&stream_flags, footer);
2131 [ - + ]: 94 : if (ret != LZMA_OK)
2132 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_stream_footer_decode");
2133 : :
2134 [ - + ]: 94 : if (lseek (fd, footer_pos - stream_flags.backward_size, SEEK_END) == -1)
2135 [ # # # # ]: 0 : throw libc_exception (errno, "lseek");
2136 : :
2137 : 94 : lzma_stream strm = LZMA_STREAM_INIT;
2138 : 94 : lzma_index* index = NULL;
2139 : 94 : ret = lzma_index_decoder (&strm, &index, UINT64_MAX);
2140 [ - + ]: 94 : if (ret != LZMA_OK)
2141 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_index_decoder");
2142 : 94 : defer_dtor<lzma_stream*,void> strm_ender (&strm, lzma_end);
2143 : :
2144 : 94 : uint8_t in_buf[4096];
2145 : 94 : while (true)
2146 : : {
2147 [ + - ]: 94 : if (strm.avail_in == 0)
2148 : : {
2149 [ + - ]: 94 : ssize_t bytes_read = read (fd, in_buf, sizeof (in_buf));
2150 [ - + ]: 94 : if (bytes_read < 0)
2151 : : {
2152 [ # # ]: 0 : if (errno == EINTR)
2153 : 0 : continue;
2154 [ # # # # ]: 0 : throw libc_exception (errno, "read");
2155 : : }
2156 [ - + ]: 94 : if (bytes_read == 0)
2157 [ # # # # ]: 0 : throw reportable_exception ("truncated file");
2158 : 94 : strm.avail_in = bytes_read;
2159 : 94 : strm.next_in = in_buf;
2160 : : }
2161 : :
2162 : 94 : ret = lzma_code (&strm, LZMA_RUN);
2163 [ - + ]: 94 : if (ret == LZMA_STREAM_END)
2164 : : break;
2165 [ # # ]: 0 : else if (ret != LZMA_OK)
2166 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_code index");
2167 : : }
2168 : :
2169 : 94 : ret = lzma_index_stream_flags (index, &stream_flags);
2170 [ - + ]: 94 : if (ret != LZMA_OK)
2171 : : {
2172 : 0 : lzma_index_end (index, NULL);
2173 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_index_stream_flags");
2174 : : }
2175 : 94 : return index;
2176 : 94 : }
2177 : :
2178 : : static void
2179 : 94 : my_lzma_index_end (lzma_index* index)
2180 : : {
2181 : 94 : lzma_index_end (index, NULL);
2182 : 94 : }
2183 : :
2184 : : static void
2185 : 102 : free_lzma_block_filter_options (lzma_block* block)
2186 : : {
2187 [ + + ]: 510 : for (int i = 0; i < LZMA_FILTERS_MAX; i++)
2188 : : {
2189 : 408 : free (block->filters[i].options);
2190 : 408 : block->filters[i].options = NULL;
2191 : : }
2192 : 102 : }
2193 : :
2194 : : static void
2195 : 94 : free_lzma_block_filters (lzma_block* block)
2196 : : {
2197 [ + - ]: 94 : if (block->filters != NULL)
2198 : : {
2199 : 94 : free_lzma_block_filter_options (block);
2200 : 94 : free (block->filters);
2201 : : }
2202 : 94 : }
2203 : :
2204 : : static void
2205 : 94 : extract_xz_blocks_into_fd (const string& srcpath,
2206 : : int src,
2207 : : int dst,
2208 : : lzma_index_iter* iter,
2209 : : uint64_t offset,
2210 : : uint64_t size)
2211 : : {
2212 : : // Seek to the Block. Seeking from the end using the compressed size from the
2213 : : // footer means we don't need to know where the xz stream starts in the
2214 : : // archive.
2215 [ - + ]: 94 : if (lseek (src,
2216 : 94 : (off_t) iter->block.compressed_stream_offset
2217 : 94 : - (off_t) iter->stream.compressed_size,
2218 : : SEEK_END) == -1)
2219 [ # # # # ]: 0 : throw libc_exception (errno, "lseek");
2220 : :
2221 : 94 : offset -= iter->block.uncompressed_file_offset;
2222 : :
2223 : 94 : lzma_block block{};
2224 : 94 : block.filters = (lzma_filter*) calloc (LZMA_FILTERS_MAX + 1,
2225 : : sizeof (lzma_filter));
2226 [ - + ]: 94 : if (block.filters == NULL)
2227 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate lzma_block filters");
2228 : 94 : defer_dtor<lzma_block*,void> filters_freer (&block, free_lzma_block_filters);
2229 : :
2230 : 94 : uint8_t in_buf[4096];
2231 : 94 : uint8_t out_buf[4096];
2232 : 94 : size_t header_read = 0;
2233 : 94 : bool need_log_extracting = verbose > 3;
2234 : : while (true)
2235 : : {
2236 : : // The first byte of the Block is the encoded Block Header Size. Read the
2237 : : // first byte and whatever extra fits in the buffer.
2238 [ + + ]: 196 : while (header_read == 0)
2239 : : {
2240 [ + - ]: 94 : ssize_t bytes_read = read (src, in_buf, sizeof (in_buf));
2241 [ - + ]: 94 : if (bytes_read < 0)
2242 : : {
2243 [ # # ]: 0 : if (errno == EINTR)
2244 : 0 : continue;
2245 [ # # # # ]: 0 : throw libc_exception (errno, "read");
2246 : : }
2247 [ - + ]: 94 : if (bytes_read == 0)
2248 [ # # # # ]: 0 : throw reportable_exception ("truncated file");
2249 : 94 : header_read += bytes_read;
2250 : : }
2251 : :
2252 : 102 : block.header_size = lzma_block_header_size_decode (in_buf[0]);
2253 : :
2254 : : // If we didn't buffer the whole Block Header earlier, get the rest.
2255 : 102 : eu_static_assert (sizeof (in_buf)
2256 : : >= lzma_block_header_size_decode (UINT8_MAX));
2257 [ - + ]: 102 : while (header_read < block.header_size)
2258 : : {
2259 [ # # ]: 0 : ssize_t bytes_read = read (src, in_buf + header_read,
2260 : : sizeof (in_buf) - header_read);
2261 [ # # ]: 0 : if (bytes_read < 0)
2262 : : {
2263 [ # # ]: 0 : if (errno == EINTR)
2264 : 0 : continue;
2265 [ # # # # ]: 0 : throw libc_exception (errno, "read");
2266 : : }
2267 [ # # ]: 0 : if (bytes_read == 0)
2268 [ # # # # ]: 0 : throw reportable_exception ("truncated file");
2269 : 0 : header_read += bytes_read;
2270 : : }
2271 : :
2272 : : // Decode the Block Header.
2273 : 102 : block.check = iter->stream.flags->check;
2274 : 102 : lzma_ret ret = lzma_block_header_decode (&block, NULL, in_buf);
2275 [ - + ]: 102 : if (ret != LZMA_OK)
2276 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_block_header_decode");
2277 : 102 : ret = lzma_block_compressed_size (&block, iter->block.unpadded_size);
2278 [ - + ]: 102 : if (ret != LZMA_OK)
2279 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_block_compressed_size");
2280 : :
2281 : : // Start decoding the Block data.
2282 : 102 : lzma_stream strm = LZMA_STREAM_INIT;
2283 : 102 : ret = lzma_block_decoder (&strm, &block);
2284 [ - + ]: 102 : if (ret != LZMA_OK)
2285 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_block_decoder");
2286 : 102 : defer_dtor<lzma_stream*,void> strm_ender (&strm, lzma_end);
2287 : :
2288 : : // We might still have some input buffered from when we read the header.
2289 : 102 : strm.avail_in = header_read - block.header_size;
2290 : 102 : strm.next_in = in_buf + block.header_size;
2291 : 102 : strm.avail_out = sizeof (out_buf);
2292 : 102 : strm.next_out = out_buf;
2293 : 11518 : while (true)
2294 : : {
2295 [ + + ]: 11518 : if (strm.avail_in == 0)
2296 : : {
2297 [ + - ]: 14 : ssize_t bytes_read = read (src, in_buf, sizeof (in_buf));
2298 [ - + ]: 14 : if (bytes_read < 0)
2299 : : {
2300 [ # # ]: 0 : if (errno == EINTR)
2301 : 0 : continue;
2302 [ # # # # ]: 0 : throw libc_exception (errno, "read");
2303 : : }
2304 [ - + ]: 14 : if (bytes_read == 0)
2305 [ # # # # ]: 0 : throw reportable_exception ("truncated file");
2306 : 14 : strm.avail_in = bytes_read;
2307 : 14 : strm.next_in = in_buf;
2308 : : }
2309 : :
2310 : 11518 : ret = lzma_code (&strm, LZMA_RUN);
2311 [ - + ]: 11518 : if (ret != LZMA_OK && ret != LZMA_STREAM_END)
2312 [ # # # # ]: 0 : throw lzma_exception (ret, "lzma_code block");
2313 : :
2314 : : // Throw away anything we decode until we reach the offset, then
2315 : : // start writing to the destination.
2316 [ + + ]: 11518 : if (strm.total_out > offset)
2317 : : {
2318 : 2898 : size_t bytes_to_write = strm.next_out - out_buf;
2319 : 2898 : uint8_t* buf_to_write = out_buf;
2320 : :
2321 : : // Ignore anything in the buffer before the offset.
2322 [ + + ]: 2898 : if (bytes_to_write > strm.total_out - offset)
2323 : : {
2324 : 88 : buf_to_write += bytes_to_write - (strm.total_out - offset);
2325 : 88 : bytes_to_write = strm.total_out - offset;
2326 : : }
2327 : :
2328 : : // Ignore anything after the size.
2329 [ + + ]: 2898 : if (strm.total_out - offset >= size)
2330 : 94 : bytes_to_write -= strm.total_out - offset - size;
2331 : :
2332 [ + + ]: 2898 : if (need_log_extracting)
2333 : : {
2334 [ + - + - ]: 24 : obatched(clog) << "extracting from xz archive " << srcpath
2335 [ + - + - : 12 : << " size=" << size << endl;
+ - ]
2336 : 12 : need_log_extracting = false;
2337 : : }
2338 : :
2339 [ + + ]: 5796 : while (bytes_to_write > 0)
2340 : : {
2341 [ + - ]: 2898 : ssize_t written = write (dst, buf_to_write, bytes_to_write);
2342 [ - + ]: 2898 : if (written < 0)
2343 : : {
2344 [ # # ]: 0 : if (errno == EAGAIN)
2345 : 0 : continue;
2346 [ # # # # ]: 0 : throw libc_exception (errno, "write");
2347 : : }
2348 : 2898 : bytes_to_write -= written;
2349 : 2898 : buf_to_write += written;
2350 : : }
2351 : :
2352 : : // If we reached the size, we're done.
2353 [ + + ]: 2898 : if (strm.total_out - offset >= size)
2354 : 94 : return;
2355 : : }
2356 : :
2357 : 11424 : strm.avail_out = sizeof (out_buf);
2358 : 11424 : strm.next_out = out_buf;
2359 : :
2360 [ + + ]: 11424 : if (ret == LZMA_STREAM_END)
2361 : : break;
2362 : : }
2363 : :
2364 : : // This Block didn't have enough data. Go to the next one.
2365 [ - + ]: 8 : if (lzma_index_iter_next (iter, LZMA_INDEX_ITER_BLOCK))
2366 [ # # # # ]: 0 : throw reportable_exception ("no more blocks");
2367 [ + - ]: 8 : if (strm.total_out > offset)
2368 : 8 : size -= strm.total_out - offset;
2369 : 8 : offset = 0;
2370 : : // If we had any buffered input left, move it to the beginning of the
2371 : : // buffer to decode the next Block Header.
2372 [ + - ]: 8 : if (strm.avail_in > 0)
2373 : : {
2374 : 8 : memmove (in_buf, strm.next_in, strm.avail_in);
2375 : 8 : header_read = strm.avail_in;
2376 : : }
2377 : : else
2378 : : header_read = 0;
2379 : 8 : free_lzma_block_filter_options (&block);
2380 : 102 : }
2381 : 94 : }
2382 : :
2383 : : static int
2384 : 94 : extract_from_seekable_archive (const string& srcpath,
2385 : : char* tmppath,
2386 : : uint64_t offset,
2387 : : uint64_t size)
2388 : : {
2389 [ + - + - : 188 : inc_metric ("seekable_archive_extraction_attempts","type","xz");
+ - - + -
+ - - -
- ]
2390 : 94 : try
2391 : : {
2392 [ + - ]: 94 : int src = open (srcpath.c_str(), O_RDONLY);
2393 [ - + ]: 94 : if (src < 0)
2394 [ # # # # : 0 : throw libc_exception (errno, string("open ") + srcpath);
# # # # ]
2395 : 94 : defer_dtor<int,int> src_closer (src, close);
2396 : :
2397 [ + - ]: 94 : lzma_index* index = read_xz_index (src);
2398 : 94 : defer_dtor<lzma_index*,void> index_ender (index, my_lzma_index_end);
2399 : :
2400 : : // Find the Block containing the offset.
2401 : 94 : lzma_index_iter iter;
2402 : 94 : lzma_index_iter_init (&iter, index);
2403 [ - + ]: 94 : if (lzma_index_iter_locate (&iter, offset))
2404 [ # # # # ]: 0 : throw reportable_exception ("offset not found");
2405 : :
2406 [ + + ]: 94 : if (verbose > 3)
2407 [ + - + - ]: 36 : obatched(clog) << "seeking in xz archive " << srcpath
2408 [ + - + - : 12 : << " offset=" << offset << " block_offset="
+ - ]
2409 [ + - + - ]: 12 : << iter.block.uncompressed_file_offset << endl;
2410 : :
2411 [ + - ]: 94 : int dst = mkstemp (tmppath);
2412 [ - + ]: 94 : if (dst < 0)
2413 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
2414 : :
2415 : 94 : try
2416 : : {
2417 [ + - ]: 94 : extract_xz_blocks_into_fd (srcpath, src, dst, &iter, offset, size);
2418 : : }
2419 : 0 : catch (...)
2420 : : {
2421 : 0 : unlink (tmppath);
2422 [ - - ]: 0 : close (dst);
2423 : 0 : throw;
2424 : 0 : }
2425 : :
2426 [ + - + - : 188 : inc_metric ("seekable_archive_extraction_successes","type","xz");
+ - + - -
+ - + - -
- - ]
2427 : 94 : return dst;
2428 : 94 : }
2429 [ - - ]: 0 : catch (const reportable_exception &e)
2430 : : {
2431 [ - - - - : 0 : inc_metric ("seekable_archive_extraction_failures","type","xz");
- - - - -
- - - - -
- - ]
2432 [ - - ]: 0 : if (verbose)
2433 [ - - - - ]: 0 : obatched(clog) << "failed to extract from seekable xz archive "
2434 [ - - - - : 0 : << srcpath << ": " << e.message << endl;
- - ]
2435 : 0 : return -1;
2436 : 0 : }
2437 : : }
2438 : : #else
2439 : : static bool
2440 : : is_seekable_archive (const string& rps, struct archive* a)
2441 : : {
2442 : : return false;
2443 : : }
2444 : : static int
2445 : : extract_from_seekable_archive (const string& srcpath,
2446 : : char* tmppath,
2447 : : uint64_t offset,
2448 : : uint64_t size)
2449 : : {
2450 : : return -1;
2451 : : }
2452 : : #endif
2453 : :
2454 : :
2455 : : // For security/portability reasons, many distro-package archives have
2456 : : // a "./" in front of path names; others have nothing, others have
2457 : : // "/". Canonicalize them all to a single leading "/", with the
2458 : : // assumption that this matches the dwarf-derived file names too.
2459 : 1296 : string canonicalized_archive_entry_pathname(struct archive_entry *e)
2460 : : {
2461 : 1296 : string fn = archive_entry_pathname(e);
2462 [ - + ]: 1296 : if (fn.size() == 0)
2463 : 0 : return fn;
2464 [ - + ]: 1296 : if (fn[0] == '/')
2465 : 0 : return fn;
2466 [ + + ]: 1296 : if (fn[0] == '.')
2467 [ + - ]: 1082 : return fn.substr(1);
2468 : : else
2469 [ + - + - : 428 : return string("/")+fn;
- - ]
2470 : 1296 : }
2471 : :
2472 : :
2473 : : // NB: takes ownership of, and may reassign, fd.
2474 : : static struct MHD_Response*
2475 : 998 : create_buildid_r_response (int64_t b_mtime0,
2476 : : const string& b_source0,
2477 : : const string& b_source1,
2478 : : const string& section,
2479 : : const string& ima_sig,
2480 : : const char* tmppath,
2481 : : int& fd,
2482 : : off_t size,
2483 : : time_t mtime,
2484 : : const string& metric,
2485 : : const struct timespec& extract_begin)
2486 : : {
2487 [ + + ]: 998 : if (tmppath != NULL)
2488 : : {
2489 : 226 : struct timespec extract_end;
2490 : 226 : clock_gettime (CLOCK_MONOTONIC, &extract_end);
2491 : 226 : double extract_time = (extract_end.tv_sec - extract_begin.tv_sec)
2492 : 226 : + (extract_end.tv_nsec - extract_begin.tv_nsec)/1.e9;
2493 [ + - ]: 452 : fdcache.intern(b_source0, b_source1, tmppath, size, true, extract_time);
2494 : : }
2495 : :
2496 [ + + ]: 998 : if (!section.empty ())
2497 : : {
2498 [ + - + - ]: 8 : int scn_fd = extract_section (fd, b_mtime0,
2499 [ + - - + ]: 16 : b_source0 + ":" + b_source1,
2500 : : section, extract_begin);
2501 : 8 : close (fd);
2502 [ + + ]: 8 : if (scn_fd >= 0)
2503 : 4 : fd = scn_fd;
2504 : : else
2505 : : {
2506 [ + - ]: 4 : if (verbose)
2507 [ + - ]: 12 : obatched (clog) << "cannot find section " << section
2508 : : << " for archive " << b_source0
2509 [ + - + - : 4 : << " file " << b_source1 << endl;
+ - + - +
- ]
2510 : 4 : return 0;
2511 : : }
2512 : :
2513 : 4 : struct stat fs;
2514 [ - + ]: 4 : if (fstat (fd, &fs) < 0)
2515 : : {
2516 : 0 : close (fd);
2517 [ # # # # ]: 0 : throw libc_exception (errno,
2518 [ # # # # : 0 : string ("fstat ") + b_source0 + string (" ") + section);
# # # # #
# # # # #
# # # # #
# ]
2519 : : }
2520 : 4 : size = fs.st_size;
2521 : : }
2522 : :
2523 : 994 : struct MHD_Response* r = MHD_create_response_from_fd (size, fd);
2524 [ - + ]: 994 : if (r == 0)
2525 : : {
2526 [ # # ]: 0 : if (verbose)
2527 [ # # # # ]: 0 : obatched(clog) << "cannot create fd-response for " << b_source0 << endl;
2528 : 0 : close(fd);
2529 : : }
2530 : : else
2531 : : {
2532 [ + - + - : 1988 : inc_metric ("http_responses_total","result",metric);
- + - - ]
2533 : 994 : add_mhd_response_header (r, "Content-Type", "application/octet-stream");
2534 [ + - ]: 994 : add_mhd_response_header (r, "X-DEBUGINFOD-SIZE", to_string(size).c_str());
2535 : 994 : add_mhd_response_header (r, "X-DEBUGINFOD-ARCHIVE", b_source0.c_str());
2536 : 994 : add_mhd_response_header (r, "X-DEBUGINFOD-FILE", b_source1.c_str());
2537 [ - + ]: 994 : if(!ima_sig.empty()) add_mhd_response_header(r, "X-DEBUGINFOD-IMASIGNATURE", ima_sig.c_str());
2538 : 994 : add_mhd_last_modified (r, mtime);
2539 [ - + ]: 994 : if (verbose > 1)
2540 [ + - ]: 2982 : obatched(clog) << "serving " << metric << " " << b_source0
2541 : : << " file " << b_source1
2542 : : << " section=" << section
2543 [ + - + - : 994 : << " IMA signature=" << ima_sig << endl;
+ - + - +
- + - + -
+ - + - ]
2544 : : /* libmicrohttpd will close fd. */
2545 : : }
2546 : : return r;
2547 : : }
2548 : :
2549 : : static struct MHD_Response*
2550 : 1054 : handle_buildid_r_match (bool internal_req_p,
2551 : : int64_t b_mtime,
2552 : : const string& b_source0,
2553 : : const string& b_source1,
2554 : : int64_t b_id0,
2555 : : int64_t b_id1,
2556 : : const string& section,
2557 : : int *result_fd)
2558 : : {
2559 : 1054 : struct timespec extract_begin;
2560 : 1054 : clock_gettime (CLOCK_MONOTONIC, &extract_begin);
2561 : :
2562 : 1054 : struct stat fs;
2563 : 1054 : int rc = stat (b_source0.c_str(), &fs);
2564 [ + + ]: 1054 : if (rc != 0)
2565 [ + - + - : 116 : throw libc_exception (errno, string("stat ") + b_source0);
+ - - + ]
2566 : :
2567 [ - + ]: 996 : if ((int64_t) fs.st_mtime != b_mtime)
2568 : : {
2569 [ # # ]: 0 : if (verbose)
2570 [ # # # # ]: 0 : obatched(clog) << "mtime mismatch for " << b_source0 << endl;
2571 : 0 : return 0;
2572 : : }
2573 : :
2574 : : // Extract the IMA per-file signature (if it exists)
2575 : 996 : string ima_sig = "";
2576 : : #ifdef ENABLE_IMA_VERIFICATION
2577 : : do
2578 : : {
2579 : : FD_t rpm_fd;
2580 : : if(!(rpm_fd = Fopen(b_source0.c_str(), "r.ufdio"))) // read, uncompressed, rpm/rpmio.h
2581 : : {
2582 : : if (verbose) obatched(clog) << "There was an error while opening " << b_source0 << endl;
2583 : : break; // Exit IMA extraction
2584 : : }
2585 : :
2586 : : Header rpm_hdr;
2587 : : if(RPMRC_FAIL == rpmReadPackageFile(NULL, rpm_fd, b_source0.c_str(), &rpm_hdr))
2588 : : {
2589 : : if (verbose) obatched(clog) << "There was an error while reading the header of " << b_source0 << endl;
2590 : : Fclose(rpm_fd);
2591 : : break; // Exit IMA extraction
2592 : : }
2593 : :
2594 : : // Fill sig_tag_data with an alloc'd copy of the array of IMA signatures (if they exist)
2595 : : struct rpmtd_s sig_tag_data;
2596 : : rpmtdReset(&sig_tag_data);
2597 : : do{ /* A do-while so we can break out of the koji sigcache checking on failure */
2598 : : if(requires_koji_sigcache_mapping)
2599 : : {
2600 : : /* NB: Koji builds result in a directory structure like the following
2601 : : - PACKAGE/VERSION/RELEASE
2602 : : - ARCH1
2603 : : - foo.rpm // The rpm known by debuginfod
2604 : : - ...
2605 : : - ARCHN
2606 : : - data
2607 : : - signed // Periodically purged (and not scanned by debuginfod)
2608 : : - sigcache
2609 : : - ARCH1
2610 : : - foo.rpm.sig // An empty rpm header
2611 : : - ...
2612 : : - ARCHN
2613 : : - PACKAGE_KEYID1
2614 : : - ARCH1
2615 : : - foo.rpm.sig // The header of the signed rpm. This is the file we need to extract the IMA signatures
2616 : : - ...
2617 : : - ARCHN
2618 : : - ...
2619 : : - PACKAGE_KEYIDn
2620 : :
2621 : : We therefore need to do a mapping:
2622 : :
2623 : : P/V/R/A/N-V-R.A.rpm ->
2624 : : P/V/R/data/sigcache/KEYID/A/N-V-R.A.rpm.sig
2625 : :
2626 : : There are 2 key insights here
2627 : :
2628 : : 1. We need to go 2 directories down from sigcache to get to the
2629 : : rpm header. So to distinguish ARCH1/foo.rpm.sig and
2630 : : PACKAGE_KEYID1/ARCH1/foo.rpm.sig we can look 2 directories down
2631 : :
2632 : : 2. It's safe to assume that the user will have all of the
2633 : : required verification certs. So we can pick from any of the
2634 : : PACKAGE_KEYID* directories. For simplicity we choose first we
2635 : : match against
2636 : :
2637 : : See: https://pagure.io/koji/issue/3670
2638 : : */
2639 : :
2640 : : // Do the mapping from b_source0 to the koji path for the signed rpm header
2641 : : string signed_rpm_path = b_source0;
2642 : : size_t insert_pos = string::npos;
2643 : : for(int i = 0; i < 2; i++) insert_pos = signed_rpm_path.rfind("/", insert_pos) - 1;
2644 : : string globbed_path = signed_rpm_path.insert(insert_pos + 1, "/data/sigcache/*").append(".sig"); // The globbed path we're seeking
2645 : : glob_t pglob;
2646 : : int grc;
2647 : : if(0 != (grc = glob(globbed_path.c_str(), GLOB_NOSORT, NULL, &pglob)))
2648 : : {
2649 : : // Break out, but only report real errors
2650 : : if (verbose && grc != GLOB_NOMATCH) obatched(clog) << "There was an error (" << strerror(errno) << ") globbing " << globbed_path << endl;
2651 : : break; // Exit koji sigcache check
2652 : : }
2653 : : signed_rpm_path = pglob.gl_pathv[0]; // See insight 2 above
2654 : : globfree(&pglob);
2655 : :
2656 : : if (verbose > 2) obatched(clog) << "attempting IMA signature extraction from koji header " << signed_rpm_path << endl;
2657 : :
2658 : : FD_t sig_rpm_fd;
2659 : : if(NULL == (sig_rpm_fd = Fopen(signed_rpm_path.c_str(), "r")))
2660 : : {
2661 : : if (verbose) obatched(clog) << "There was an error while opening " << signed_rpm_path << endl;
2662 : : break; // Exit koji sigcache check
2663 : : }
2664 : :
2665 : : Header sig_hdr = headerRead(sig_rpm_fd, HEADER_MAGIC_YES /* Validate magic too */ );
2666 : : if (!sig_hdr || 1 != headerGet(sig_hdr, RPMSIGTAG_FILESIGNATURES, &sig_tag_data, HEADERGET_ALLOC))
2667 : : {
2668 : : if (verbose) obatched(clog) << "Unable to extract RPMSIGTAG_FILESIGNATURES from " << signed_rpm_path << endl;
2669 : : }
2670 : : headerFree(sig_hdr); // We can free here since sig_tag_data has an alloc'd copy of the data
2671 : : Fclose(sig_rpm_fd);
2672 : : }
2673 : : }while(false);
2674 : :
2675 : : if(0 == sig_tag_data.count)
2676 : : {
2677 : : // In the general case (or a fallback from the koji sigcache mapping not finding signatures)
2678 : : // we can just (try) extract the signatures from the rpm header
2679 : : if (1 != headerGet(rpm_hdr, RPMTAG_FILESIGNATURES, &sig_tag_data, HEADERGET_ALLOC))
2680 : : {
2681 : : if (verbose) obatched(clog) << "Unable to extract RPMTAG_FILESIGNATURES from " << b_source0 << endl;
2682 : : }
2683 : : }
2684 : : // Search the array for the signature coresponding to b_source1
2685 : : int idx = -1;
2686 : : char *sig = NULL;
2687 : : rpmfi hdr_fi = rpmfiNew(NULL, rpm_hdr, RPMTAG_BASENAMES, RPMFI_FLAGS_QUERY);
2688 : : do
2689 : : {
2690 : : sig = (char*)rpmtdNextString(&sig_tag_data);
2691 : : idx = rpmfiNext(hdr_fi);
2692 : : }
2693 : : while (idx != -1 && 0 != strcmp(b_source1.c_str(), rpmfiFN(hdr_fi)));
2694 : : rpmfiFree(hdr_fi);
2695 : :
2696 : : if(sig && 0 != strlen(sig) && idx != -1)
2697 : : {
2698 : : if (verbose > 2) obatched(clog) << "Found IMA signature for " << b_source1 << ":\n" << sig << endl;
2699 : : ima_sig = sig;
2700 : : inc_metric("http_responses_total","extra","ima-sigs-extracted");
2701 : : }
2702 : : else
2703 : : {
2704 : : if (verbose > 2) obatched(clog) << "Could not find IMA signature for " << b_source1 << endl;
2705 : : }
2706 : :
2707 : : rpmtdFreeData (&sig_tag_data);
2708 : : headerFree(rpm_hdr);
2709 : : Fclose(rpm_fd);
2710 : : } while(false);
2711 : : #endif
2712 : :
2713 : : // check for a match in the fdcache first
2714 [ + - ]: 996 : int fd = fdcache.lookup(b_source0, b_source1);
2715 [ + + ]: 996 : while (fd >= 0) // got one!; NB: this is really an if() with a possible branch out to the end
2716 : : {
2717 : 772 : rc = fstat(fd, &fs);
2718 [ - + ]: 772 : if (rc < 0) // disappeared?
2719 : : {
2720 [ # # ]: 0 : if (verbose)
2721 [ # # # # : 0 : obatched(clog) << "cannot fstat fdcache " << b_source0 << endl;
# # ]
2722 [ # # ]: 0 : close(fd);
2723 [ # # ]: 0 : fdcache.clear(b_source0, b_source1);
2724 : : break; // branch out of if "loop", to try new libarchive fetch attempt
2725 : : }
2726 : :
2727 [ + - + - : 772 : struct MHD_Response* r = create_buildid_r_response (b_mtime, b_source0,
- + - - ]
2728 : : b_source1, section,
2729 : : ima_sig, NULL, fd,
2730 : : fs.st_size,
2731 : : fs.st_mtime,
2732 : : "archive fdcache",
2733 : : extract_begin);
2734 [ + + ]: 772 : if (r == 0)
2735 : : break; // branch out of if "loop", to try new libarchive fetch attempt
2736 [ + - ]: 770 : if (result_fd)
2737 : 770 : *result_fd = fd;
2738 : : return r;
2739 : : // NB: see, we never go around the 'loop' more than once
2740 : : }
2741 : :
2742 : : // no match ... look for a seekable entry
2743 : 226 : bool populate_seekable = ! passive_p;
2744 : 226 : unique_ptr<sqlite_ps> pp (new sqlite_ps (internal_req_p ? db : dbq,
2745 : : "rpm-seekable-query",
2746 : : "select type, size, offset, mtime from " BUILDIDS "_r_seekable "
2747 [ + - + - : 452 : "where file = ? and content = ?"));
+ - + + +
- + - + -
- - ]
2748 [ + - + - : 226 : rc = pp->reset().bind(1, b_id0).bind(2, b_id1).step();
+ - + - ]
2749 [ + + ]: 226 : if (rc != SQLITE_DONE)
2750 : : {
2751 [ - + ]: 94 : if (rc != SQLITE_ROW)
2752 [ # # # # ]: 0 : throw sqlite_exception(rc, "step");
2753 : : // if we found a match in _r_seekable but we fail to extract it, don't
2754 : : // bother populating it again
2755 : 94 : populate_seekable = false;
2756 [ + - ]: 94 : const char* seekable_type = (const char*) sqlite3_column_text (*pp, 0);
2757 [ + - - + ]: 94 : if (seekable_type != NULL && strcmp (seekable_type, "xz") == 0)
2758 : : {
2759 [ + - ]: 94 : int64_t seekable_size = sqlite3_column_int64 (*pp, 1);
2760 [ + - ]: 94 : int64_t seekable_offset = sqlite3_column_int64 (*pp, 2);
2761 [ + - ]: 94 : int64_t seekable_mtime = sqlite3_column_int64 (*pp, 3);
2762 : :
2763 : 94 : char* tmppath = NULL;
2764 [ - + ]: 94 : if (asprintf (&tmppath, "%s/debuginfod-fdcache.XXXXXX", tmpdir.c_str()) < 0)
2765 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
2766 : 94 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
2767 : :
2768 [ + - ]: 94 : fd = extract_from_seekable_archive (b_source0, tmppath,
2769 : : seekable_offset, seekable_size);
2770 [ + - ]: 94 : if (fd >= 0)
2771 : : {
2772 : : // Set the mtime so the fdcache file mtimes propagate to future webapi
2773 : : // clients.
2774 : 94 : struct timespec tvs[2];
2775 : 94 : tvs[0].tv_sec = 0;
2776 : 94 : tvs[0].tv_nsec = UTIME_OMIT;
2777 : 94 : tvs[1].tv_sec = seekable_mtime;
2778 : 94 : tvs[1].tv_nsec = 0;
2779 : 94 : (void) futimens (fd, tvs); /* best effort */
2780 [ + - + - : 94 : struct MHD_Response* r = create_buildid_r_response (b_mtime,
+ - ]
2781 : : b_source0,
2782 : : b_source1,
2783 : : section,
2784 : : ima_sig,
2785 : : tmppath, fd,
2786 : : seekable_size,
2787 : : seekable_mtime,
2788 : : "seekable xz archive",
2789 : : extract_begin);
2790 [ + - ]: 94 : if (r != 0 && result_fd)
2791 : 94 : *result_fd = fd;
2792 : 94 : return r;
2793 : : }
2794 : 94 : }
2795 : : }
2796 : 132 : pp.reset();
2797 : :
2798 : : // still no match ... grumble, must process the archive
2799 [ + - ]: 132 : string archive_decoder = "/dev/null";
2800 [ + - - - ]: 132 : string archive_extension = "";
2801 [ + + ]: 302 : for (auto&& arch : scan_archives)
2802 [ + + ]: 170 : if (string_endswith(b_source0, arch.first))
2803 : : {
2804 [ + - ]: 132 : archive_extension = arch.first;
2805 [ + - ]: 302 : archive_decoder = arch.second;
2806 : : }
2807 : 132 : FILE* fp;
2808 : :
2809 : 132 : defer_dtor<FILE*,int>::dtor_fn dfn;
2810 [ + + ]: 132 : if (archive_decoder != "cat")
2811 : : {
2812 [ + - + - : 48 : string popen_cmd = archive_decoder + " " + shell_escape(b_source0);
+ - - + -
- - - ]
2813 [ + - ]: 24 : fp = popen (popen_cmd.c_str(), "r"); // "e" O_CLOEXEC?
2814 : 24 : dfn = pclose;
2815 [ - + ]: 24 : if (fp == NULL)
2816 [ # # # # : 0 : throw libc_exception (errno, string("popen ") + popen_cmd);
# # # # ]
2817 : 24 : }
2818 : : else
2819 : : {
2820 [ + - ]: 108 : fp = fopen (b_source0.c_str(), "r");
2821 : 108 : dfn = fclose;
2822 [ - + ]: 108 : if (fp == NULL)
2823 [ # # # # : 0 : throw libc_exception (errno, string("fopen ") + b_source0);
# # # # ]
2824 : : }
2825 : 132 : defer_dtor<FILE*,int> fp_closer (fp, dfn);
2826 : :
2827 : 132 : struct archive *a;
2828 [ + - ]: 132 : a = archive_read_new();
2829 [ - + ]: 132 : if (a == NULL)
2830 [ # # # # ]: 0 : throw archive_exception("cannot create archive reader");
2831 : 132 : defer_dtor<struct archive*,int> archive_closer (a, archive_read_free);
2832 : :
2833 [ + - ]: 132 : rc = archive_read_support_format_all(a);
2834 [ - + ]: 132 : if (rc != ARCHIVE_OK)
2835 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all format");
2836 [ + - ]: 132 : rc = archive_read_support_filter_all(a);
2837 [ - + ]: 132 : if (rc != ARCHIVE_OK)
2838 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all filters");
2839 : :
2840 [ + - ]: 132 : rc = archive_read_open_FILE (a, fp);
2841 [ - + ]: 132 : if (rc != ARCHIVE_OK)
2842 : : {
2843 [ # # # # : 0 : obatched(clog) << "cannot open archive from pipe " << b_source0 << endl;
# # ]
2844 [ # # # # ]: 0 : throw archive_exception(a, "cannot open archive from pipe");
2845 : : }
2846 : :
2847 : : // If the archive was scanned in a version without _r_seekable, then we may
2848 : : // need to populate _r_seekable now. This can be removed the next time
2849 : : // BUILDIDS is updated.
2850 [ + + ]: 132 : if (populate_seekable)
2851 : : {
2852 [ + - ]: 130 : populate_seekable = is_seekable_archive (b_source0, a);
2853 [ + - ]: 130 : if (populate_seekable)
2854 : : {
2855 : : // NB: the names are already interned
2856 [ # # ]: 0 : pp.reset(new sqlite_ps (db, "rpm-seekable-insert2",
2857 : : "insert or ignore into " BUILDIDS "_r_seekable (file, content, type, size, offset, mtime) "
2858 : : "values (?, "
2859 : : "(select id from " BUILDIDS "_files "
2860 : : "where dirname = (select id from " BUILDIDS "_fileparts where name = ?) "
2861 : : "and basename = (select id from " BUILDIDS "_fileparts where name = ?) "
2862 [ # # # # : 0 : "), 'xz', ?, ?, ?)"));
# # # # #
# # # ]
2863 : : }
2864 : : }
2865 : :
2866 : : // archive traversal is in five stages:
2867 : : // 1) before we find a matching entry, insert it into _r_seekable if needed or
2868 : : // skip it otherwise
2869 : : // 2) extract the matching entry (set r = result). Also insert it into
2870 : : // _r_seekable if needed
2871 : : // 3) extract some number of prefetched entries (just into fdcache). Also
2872 : : // insert them into _r_seekable if needed
2873 : : // 4) if needed, insert all of the remaining entries into _r_seekable
2874 : : // 5) abort any further processing
2875 : 132 : struct MHD_Response* r = 0; // will set in stage 2
2876 [ + + ]: 132 : unsigned prefetch_count =
2877 : : internal_req_p ? 0 : fdcache_prefetch; // will decrement in stage 3
2878 : :
2879 [ + + - + ]: 1366 : while(r == 0 || prefetch_count > 0 || populate_seekable) // stage 1-4
2880 : : {
2881 [ + - ]: 1346 : if (interrupted)
2882 : : break;
2883 : :
2884 : 1346 : struct archive_entry *e;
2885 [ + - ]: 1346 : rc = archive_read_next_header (a, &e);
2886 [ + + ]: 1346 : if (rc != ARCHIVE_OK)
2887 : : break;
2888 : :
2889 [ + - + + ]: 1236 : if (! S_ISREG(archive_entry_mode (e))) // skip non-files completely
2890 : 974 : continue;
2891 : :
2892 [ + - ]: 262 : string fn = canonicalized_archive_entry_pathname (e);
2893 : :
2894 [ - + ]: 262 : if (populate_seekable)
2895 : : {
2896 : 0 : string dn, bn;
2897 : 0 : size_t slash = fn.rfind('/');
2898 [ # # ]: 0 : if (slash == std::string::npos) {
2899 [ # # ]: 0 : dn = "";
2900 [ # # ]: 0 : bn = fn;
2901 : : } else {
2902 [ # # # # ]: 0 : dn = fn.substr(0, slash);
2903 [ # # # # ]: 0 : bn = fn.substr(slash + 1);
2904 : : }
2905 : :
2906 [ # # ]: 0 : int64_t seekable_size = archive_entry_size (e);
2907 [ # # ]: 0 : int64_t seekable_offset = archive_filter_bytes (a, 0);
2908 [ # # ]: 0 : time_t seekable_mtime = archive_entry_mtime (e);
2909 : :
2910 [ # # ]: 0 : pp->reset();
2911 [ # # ]: 0 : pp->bind(1, b_id0);
2912 [ # # ]: 0 : pp->bind(2, dn);
2913 [ # # ]: 0 : pp->bind(3, bn);
2914 [ # # ]: 0 : pp->bind(4, seekable_size);
2915 [ # # ]: 0 : pp->bind(5, seekable_offset);
2916 [ # # ]: 0 : pp->bind(6, seekable_mtime);
2917 [ # # ]: 0 : rc = pp->step();
2918 [ # # ]: 0 : if (rc != SQLITE_DONE)
2919 [ # # # # ]: 0 : obatched(clog) << "recording seekable file=" << fn
2920 [ # # # # : 0 : << " sqlite3 error: " << (sqlite3_errstr(rc) ?: "?") << endl;
# # # # #
# ]
2921 [ # # ]: 0 : else if (verbose > 2)
2922 [ # # # # : 0 : obatched(clog) << "recorded seekable file=" << fn
# # ]
2923 [ # # # # ]: 0 : << " size=" << seekable_size
2924 [ # # # # ]: 0 : << " offset=" << seekable_offset
2925 [ # # # # : 0 : << " mtime=" << seekable_mtime << endl;
# # ]
2926 [ # # ]: 0 : if (r != 0 && prefetch_count == 0) // stage 4
2927 [ # # ]: 0 : continue;
2928 [ # # # # ]: 0 : }
2929 : :
2930 [ + + + + ]: 262 : if ((r == 0) && (fn != b_source1)) // stage 1
2931 : 66 : continue;
2932 : :
2933 [ + - + + ]: 196 : if (fdcache.probe (b_source0, fn) && // skip if already interned
2934 [ + + ]: 22 : fn != b_source1) // but only if we'd just be prefetching, PR29474
2935 : 20 : continue;
2936 : :
2937 : : // extract this file to a temporary file
2938 : 176 : char* tmppath = NULL;
2939 : 176 : rc = asprintf (&tmppath, "%s/debuginfod-fdcache.XXXXXX", tmpdir.c_str());
2940 [ - + ]: 176 : if (rc < 0)
2941 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
2942 : 176 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
2943 [ + - ]: 176 : fd = mkstemp (tmppath);
2944 [ - + ]: 176 : if (fd < 0)
2945 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
2946 : : // NB: don't unlink (tmppath), as fdcache will take charge of it.
2947 : :
2948 : : // NB: this can take many uninterruptible seconds for a huge file
2949 [ + - ]: 176 : rc = archive_read_data_into_fd (a, fd);
2950 [ - + ]: 176 : if (rc != ARCHIVE_OK) // e.g. ENOSPC!
2951 : : {
2952 [ # # ]: 0 : close (fd);
2953 : 0 : unlink (tmppath);
2954 [ # # # # ]: 0 : throw archive_exception(a, "cannot extract file");
2955 : : }
2956 : :
2957 : : // Set the mtime so the fdcache file mtimes, even prefetched ones,
2958 : : // propagate to future webapi clients.
2959 : 176 : struct timespec tvs[2];
2960 : 176 : tvs[0].tv_sec = 0;
2961 : 176 : tvs[0].tv_nsec = UTIME_OMIT;
2962 [ + - ]: 176 : tvs[1].tv_sec = archive_entry_mtime(e);
2963 [ + - ]: 176 : tvs[1].tv_nsec = archive_entry_mtime_nsec(e);
2964 : 176 : (void) futimens (fd, tvs); /* best effort */
2965 : :
2966 [ + + ]: 176 : if (r != 0) // stage 3
2967 : : {
2968 : 44 : struct timespec extract_end;
2969 : 44 : clock_gettime (CLOCK_MONOTONIC, &extract_end);
2970 : 44 : double extract_time = (extract_end.tv_sec - extract_begin.tv_sec)
2971 : 44 : + (extract_end.tv_nsec - extract_begin.tv_nsec)/1.e9;
2972 : : // NB: now we know we have a complete reusable file; make fdcache
2973 : : // responsible for unlinking it later.
2974 [ + - + - : 44 : fdcache.intern(b_source0, fn,
+ - ]
2975 : : tmppath, archive_entry_size(e),
2976 : : false, extract_time); // prefetched ones go to the prefetch cache
2977 : 44 : prefetch_count --;
2978 [ + - ]: 44 : close (fd); // we're not saving this fd to make a mhd-response from!
2979 : 44 : continue;
2980 : 44 : }
2981 : :
2982 [ + - + - : 132 : r = create_buildid_r_response (b_mtime, b_source0, b_source1, section,
+ - + + ]
2983 : : ima_sig, tmppath, fd,
2984 : : archive_entry_size(e),
2985 : : archive_entry_mtime(e),
2986 [ + - ]: 132 : archive_extension + " archive",
2987 : : extract_begin);
2988 [ + + ]: 132 : if (r == 0)
2989 : : break; // assume no chance of better luck around another iteration; no other copies of same file
2990 [ + - ]: 130 : if (result_fd)
2991 : 130 : *result_fd = fd;
2992 [ + - + - : 1608 : }
+ + ]
2993 : :
2994 : : // XXX: rpm/file not found: delete this R entry?
2995 : 132 : return r;
2996 [ - + + + ]: 1354 : }
2997 : :
2998 : : void
2999 : 646 : add_client_federation_headers(debuginfod_client *client, MHD_Connection* conn){
3000 : : // Transcribe incoming User-Agent:
3001 [ - + ]: 646 : string ua = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "User-Agent") ?: "";
3002 [ + - + - : 650 : string ua_complete = string("User-Agent: ") + ua;
+ - ]
3003 [ + - ]: 646 : debuginfod_add_http_header (client, ua_complete.c_str());
3004 : :
3005 : : // Compute larger XFF:, for avoiding info loss during
3006 : : // federation, and for future cyclicity detection.
3007 [ + - + + : 1270 : string xff = MHD_lookup_connection_value (conn, MHD_HEADER_KIND, "X-Forwarded-For") ?: "";
+ - + - ]
3008 [ + + ]: 646 : if (xff != "")
3009 [ + - - + ]: 52 : xff += string(", "); // comma separated list
3010 : :
3011 : 646 : unsigned int xff_count = 0;
3012 [ + + ]: 974 : for (auto&& i : xff){
3013 [ + + ]: 328 : if (i == ',') xff_count++;
3014 : : }
3015 : :
3016 : : // if X-Forwarded-For: exceeds N hops,
3017 : : // do not delegate a local lookup miss to upstream debuginfods.
3018 [ + + ]: 646 : if (xff_count >= forwarded_ttl_limit)
3019 [ + - + - ]: 8 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found, --forwared-ttl-limit reached \
3020 : 8 : and will not query the upstream servers");
3021 : :
3022 : : // Compute the client's numeric IP address only - so can't merge with conninfo()
3023 [ + - ]: 642 : const union MHD_ConnectionInfo *u = MHD_get_connection_info (conn,
3024 : : MHD_CONNECTION_INFO_CLIENT_ADDRESS);
3025 [ + - ]: 642 : struct sockaddr *so = u ? u->client_addr : 0;
3026 : 642 : char hostname[256] = ""; // RFC1035
3027 [ + - - + ]: 642 : if (so && so->sa_family == AF_INET) {
3028 [ # # ]: 0 : (void) getnameinfo (so, sizeof (struct sockaddr_in), hostname, sizeof (hostname), NULL, 0,
3029 : : NI_NUMERICHOST);
3030 [ + - ]: 642 : } else if (so && so->sa_family == AF_INET6) {
3031 : 642 : struct sockaddr_in6* addr6 = (struct sockaddr_in6*) so;
3032 [ + - + - : 642 : if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
- + ]
3033 : 642 : struct sockaddr_in addr4;
3034 [ + - ]: 642 : memset (&addr4, 0, sizeof(addr4));
3035 : 642 : addr4.sin_family = AF_INET;
3036 : 642 : addr4.sin_port = addr6->sin6_port;
3037 [ + - ]: 642 : memcpy (&addr4.sin_addr.s_addr, addr6->sin6_addr.s6_addr+12, sizeof(addr4.sin_addr.s_addr));
3038 [ + - ]: 642 : (void) getnameinfo ((struct sockaddr*) &addr4, sizeof (addr4),
3039 : : hostname, sizeof (hostname), NULL, 0,
3040 : : NI_NUMERICHOST);
3041 : : } else {
3042 [ # # ]: 0 : (void) getnameinfo (so, sizeof (struct sockaddr_in6), hostname, sizeof (hostname), NULL, 0,
3043 : : NI_NUMERICHOST);
3044 : : }
3045 : : }
3046 : :
3047 [ + - + - : 1288 : string xff_complete = string("X-Forwarded-For: ")+xff+string(hostname);
+ - + - -
+ - + - -
- + ]
3048 [ + - ]: 642 : debuginfod_add_http_header (client, xff_complete.c_str());
3049 [ + + + - : 1360 : }
+ + ]
3050 : :
3051 : : static struct MHD_Response*
3052 : 1128 : handle_buildid_match (bool internal_req_p,
3053 : : int64_t b_mtime,
3054 : : const string& b_stype,
3055 : : const string& b_source0,
3056 : : const string& b_source1,
3057 : : int64_t b_id0,
3058 : : int64_t b_id1,
3059 : : const string& section,
3060 : : int *result_fd)
3061 : : {
3062 : 1128 : try
3063 : : {
3064 [ + + ]: 1128 : if (b_stype == "F")
3065 [ + - ]: 74 : return handle_buildid_f_match(internal_req_p, b_mtime, b_source0,
3066 : : section, result_fd);
3067 [ + - ]: 1054 : else if (b_stype == "R")
3068 [ + + ]: 1054 : return handle_buildid_r_match(internal_req_p, b_mtime, b_source0,
3069 : : b_source1, b_id0, b_id1, section,
3070 : : result_fd);
3071 : : }
3072 [ - + ]: 58 : catch (const reportable_exception &e)
3073 : : {
3074 [ + - ]: 58 : e.report(clog);
3075 : : // Report but swallow libc etc. errors here; let the caller
3076 : : // iterate to other matches of the content.
3077 : 58 : }
3078 : :
3079 : : return 0;
3080 : : }
3081 : :
3082 : :
3083 : : static int
3084 : 4 : debuginfod_find_progress (debuginfod_client *, long a, long b)
3085 : : {
3086 [ - + ]: 4 : if (verbose > 4)
3087 [ # # # # : 0 : obatched(clog) << "federated debuginfod progress=" << a << "/" << b << endl;
# # # # ]
3088 : :
3089 : 4 : return interrupted;
3090 : : }
3091 : :
3092 : :
3093 : : // a little lru pool of debuginfod_client*s for reuse between query threads
3094 : :
3095 : : mutex dc_pool_lock;
3096 : : deque<debuginfod_client*> dc_pool;
3097 : :
3098 : 666 : debuginfod_client* debuginfod_pool_begin()
3099 : : {
3100 : 666 : unique_lock<mutex> lock(dc_pool_lock);
3101 [ + + ]: 666 : if (dc_pool.size() > 0)
3102 : : {
3103 [ + - + - : 1268 : inc_metric("dc_pool_op_count","op","begin-reuse");
+ - + - -
+ - + - -
- - ]
3104 : 634 : debuginfod_client *c = dc_pool.front();
3105 : 634 : dc_pool.pop_front();
3106 : 634 : return c;
3107 : : }
3108 [ + - + - : 64 : inc_metric("dc_pool_op_count","op","begin-new");
+ - + - -
+ - + - -
- - ]
3109 [ + - ]: 32 : return debuginfod_begin();
3110 : 666 : }
3111 : :
3112 : :
3113 : 150 : void debuginfod_pool_groom()
3114 : : {
3115 : 150 : unique_lock<mutex> lock(dc_pool_lock);
3116 [ + + ]: 182 : while (dc_pool.size() > 0)
3117 : : {
3118 [ + - + - : 64 : inc_metric("dc_pool_op_count","op","end");
+ - + - -
+ - + - -
- - ]
3119 [ + - ]: 32 : debuginfod_end(dc_pool.front());
3120 : 32 : dc_pool.pop_front();
3121 : : }
3122 : 150 : }
3123 : :
3124 : :
3125 : 666 : void debuginfod_pool_end(debuginfod_client* c)
3126 : : {
3127 : 666 : unique_lock<mutex> lock(dc_pool_lock);
3128 [ + - + - : 1332 : inc_metric("dc_pool_op_count","op","end-save");
+ - + - -
+ - + - -
- - ]
3129 [ + - ]: 666 : dc_pool.push_front(c); // accelerate reuse, vs. push_back
3130 : 666 : }
3131 : :
3132 : :
3133 : : static struct MHD_Response*
3134 : 1712 : handle_buildid (MHD_Connection* conn,
3135 : : const string& buildid /* unsafe */,
3136 : : string& artifacttype /* unsafe, cleanse on exception/return */,
3137 : : const string& suffix /* unsafe */,
3138 : : int *result_fd)
3139 : : {
3140 : : // validate artifacttype
3141 : 1712 : string atype_code;
3142 [ + + + - ]: 1712 : if (artifacttype == "debuginfo") atype_code = "D";
3143 [ + + + - ]: 746 : else if (artifacttype == "executable") atype_code = "E";
3144 [ + + + - ]: 58 : else if (artifacttype == "source") atype_code = "S";
3145 [ + + + - ]: 12 : else if (artifacttype == "section") atype_code = "I";
3146 : : else {
3147 [ + - ]: 4 : artifacttype = "invalid"; // PR28242 ensure http_resposes metrics don't propagate unclean user data
3148 [ + - + - ]: 12 : throw reportable_exception("invalid artifacttype");
3149 : : }
3150 : :
3151 [ + + ]: 1708 : if (conn != 0)
3152 [ + - + - : 3658 : inc_metric("http_requests_total", "type", artifacttype);
+ - - + -
- - + ]
3153 : :
3154 : 1708 : string section;
3155 [ + + ]: 1708 : if (atype_code == "I")
3156 : : {
3157 [ - + ]: 8 : if (suffix.size () < 2)
3158 [ # # # # ]: 0 : throw reportable_exception ("invalid section suffix");
3159 : :
3160 : : // Remove leading '/'
3161 [ + - - + ]: 8 : section = suffix.substr(1);
3162 : : }
3163 : :
3164 [ + + - + ]: 1754 : if (atype_code == "S" && suffix == "")
3165 [ # # # # ]: 0 : throw reportable_exception("invalid source suffix");
3166 : :
3167 : : // validate buildid
3168 [ + + ]: 1708 : if ((buildid.size() < 2) || // not empty
3169 [ + + + - : 3414 : (buildid.size() % 2) || // even number
+ - ]
3170 : 1706 : (buildid.find_first_not_of("0123456789abcdef") != string::npos)) // pure tasty lowercase hex
3171 [ + - - + ]: 4 : throw reportable_exception("invalid buildid");
3172 : :
3173 [ + - ]: 1706 : if (verbose > 1)
3174 [ + - + - ]: 5118 : obatched(clog) << "searching for buildid=" << buildid << " artifacttype=" << artifacttype
3175 [ + - + - : 1706 : << " suffix=" << suffix << endl;
+ - + - +
- ]
3176 : :
3177 : : // If invoked from the scanner threads, use the scanners' read-write
3178 : : // connection. Otherwise use the web query threads' read-only connection.
3179 [ + + ]: 1706 : sqlite3 *thisdb = (conn == 0) ? db : dbq;
3180 : :
3181 : 1706 : sqlite_ps *pp = 0;
3182 : :
3183 [ + + ]: 1706 : if (atype_code == "D")
3184 : : {
3185 [ - + ]: 966 : pp = new sqlite_ps (thisdb, "mhd-query-d",
3186 : : "select mtime, sourcetype, source0, source1, id0, id1 from " BUILDIDS "_query_d2 where buildid = ? "
3187 [ + - + - : 1932 : "order by mtime desc");
+ - + - +
- - - ]
3188 [ + - ]: 966 : pp->reset();
3189 [ + - ]: 966 : pp->bind(1, buildid);
3190 : : }
3191 [ + + ]: 740 : else if (atype_code == "E")
3192 : : {
3193 [ - + ]: 686 : pp = new sqlite_ps (thisdb, "mhd-query-e",
3194 : : "select mtime, sourcetype, source0, source1, id0, id1 from " BUILDIDS "_query_e2 where buildid = ? "
3195 [ + - + - : 1372 : "order by mtime desc");
+ - + - +
- - - ]
3196 [ + - ]: 686 : pp->reset();
3197 [ + - ]: 686 : pp->bind(1, buildid);
3198 : : }
3199 [ + + ]: 54 : else if (atype_code == "S")
3200 : : {
3201 : : // PR25548
3202 : : // Incoming source queries may come in with either dwarf-level OR canonicalized paths.
3203 : : // We let the query pass with either one.
3204 : :
3205 [ - + ]: 46 : pp = new sqlite_ps (thisdb, "mhd-query-s",
3206 : : "select mtime, sourcetype, source0, source1 from " BUILDIDS "_query_s where buildid = ? and artifactsrc in (?,?) "
3207 [ + - + - : 92 : "order by sharedprefix(source0,source0ref) desc, mtime desc");
+ - + - +
- - - ]
3208 [ + - ]: 46 : pp->reset();
3209 [ + - ]: 46 : pp->bind(1, buildid);
3210 : : // NB: we don't store the non-canonicalized path names any more, but old databases
3211 : : // might have them (and no canon ones), so we keep searching for both.
3212 [ + - ]: 46 : pp->bind(2, suffix);
3213 [ + - + - : 730 : pp->bind(3, canon_pathname(suffix));
- + ]
3214 : : }
3215 [ + - ]: 8 : else if (atype_code == "I")
3216 : : {
3217 [ - + ]: 8 : pp = new sqlite_ps (thisdb, "mhd-query-i",
3218 : : "select mtime, sourcetype, source0, source1, 1 as debug_p from " BUILDIDS "_query_d2 where buildid = ? "
3219 : : "union all "
3220 : : "select mtime, sourcetype, source0, source1, 0 as debug_p from " BUILDIDS "_query_e2 where buildid = ? "
3221 [ + - + - : 16 : "order by debug_p desc, mtime desc");
+ - + - +
- - - ]
3222 [ + - ]: 8 : pp->reset();
3223 [ + - ]: 8 : pp->bind(1, buildid);
3224 [ + - ]: 8 : pp->bind(2, buildid);
3225 : : }
3226 : 1706 : unique_ptr<sqlite_ps> ps_closer(pp); // release pp if exception or return
3227 : :
3228 : 1706 : bool do_upstream_section_query = true;
3229 : :
3230 : : // consume all the rows
3231 : 1768 : while (1)
3232 : : {
3233 [ + - ]: 1768 : int rc = pp->step();
3234 [ + + ]: 1768 : if (rc == SQLITE_DONE) break;
3235 [ - + ]: 1128 : if (rc != SQLITE_ROW)
3236 [ # # # # ]: 0 : throw sqlite_exception(rc, "step");
3237 : :
3238 [ + - ]: 1128 : int64_t b_mtime = sqlite3_column_int64 (*pp, 0);
3239 [ + - - + : 1128 : string b_stype = string((const char*) sqlite3_column_text (*pp, 1) ?: ""); /* by DDL may not be NULL */
+ - ]
3240 [ + - - + : 1128 : string b_source0 = string((const char*) sqlite3_column_text (*pp, 2) ?: ""); /* may be NULL */
+ - - - ]
3241 [ + - + + : 1202 : string b_source1 = string((const char*) sqlite3_column_text (*pp, 3) ?: ""); /* may be NULL */
+ - - - ]
3242 : 1128 : int64_t b_id0 = 0, b_id1 = 0;
3243 [ + + + + ]: 1340 : if (atype_code == "D" || atype_code == "E")
3244 : : {
3245 [ + - ]: 1068 : b_id0 = sqlite3_column_int64 (*pp, 4);
3246 [ + - ]: 1068 : b_id1 = sqlite3_column_int64 (*pp, 5);
3247 : : }
3248 : :
3249 [ + - ]: 1128 : if (verbose > 1)
3250 [ + - + - : 3384 : obatched(clog) << "found mtime=" << b_mtime << " stype=" << b_stype
- - ]
3251 [ + - + - : 1128 : << " source0=" << b_source0 << " source1=" << b_source1 << endl;
+ - + - +
- + - +
- ]
3252 : :
3253 : : // Try accessing the located match.
3254 : : // XXX: in case of multiple matches, attempt them in parallel?
3255 [ + - ]: 1128 : auto r = handle_buildid_match (conn ? false : true,
3256 : : b_mtime, b_stype, b_source0, b_source1,
3257 : : b_id0, b_id1, section, result_fd);
3258 [ + + ]: 1128 : if (r)
3259 [ + + ]: 1066 : return r;
3260 : :
3261 : : // If a debuginfo file matching BUILDID was found but didn't contain
3262 : : // the desired section, then the section should not exist. Don't
3263 : : // bother querying upstream servers.
3264 [ + + + - : 62 : if (!section.empty () && (sqlite3_column_int (*pp, 4) == 1))
- + ]
3265 : : {
3266 : 4 : struct stat st;
3267 : :
3268 : : // For "F" sourcetype, check if the debuginfo exists. For "R"
3269 : : // sourcetype, check if the debuginfo was interned into the fdcache.
3270 [ - + ]: 6 : if ((b_stype == "F" && (stat (b_source0.c_str (), &st) == 0))
3271 [ + + + - : 6 : || (b_stype == "R" && fdcache.probe (b_source0, b_source1)))
+ - - + ]
3272 : : do_upstream_section_query = false;
3273 : : }
3274 [ + - - + : 2256 : }
+ - - + ]
3275 [ + - ]: 640 : pp->reset();
3276 : :
3277 [ - + ]: 640 : if (!do_upstream_section_query)
3278 [ # # # # ]: 0 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found");
3279 : :
3280 : : // We couldn't find it in the database. Last ditch effort
3281 : : // is to defer to other debuginfo servers.
3282 : :
3283 : 640 : int fd = -1;
3284 [ + - ]: 640 : debuginfod_client *client = debuginfod_pool_begin ();
3285 [ - + ]: 640 : if (client == NULL)
3286 [ # # # # ]: 0 : throw libc_exception(errno, "debuginfod client pool alloc");
3287 : 640 : defer_dtor<debuginfod_client*,void> client_closer (client, debuginfod_pool_end);
3288 : :
3289 [ + - ]: 640 : debuginfod_set_progressfn (client, & debuginfod_find_progress);
3290 : :
3291 [ + + ]: 640 : if (conn)
3292 [ + + ]: 620 : add_client_federation_headers(client, conn);
3293 : :
3294 [ + + ]: 636 : if (artifacttype == "debuginfo")
3295 [ + - ]: 88 : fd = debuginfod_find_debuginfo (client,
3296 [ + - ]: 88 : (const unsigned char*) buildid.c_str(),
3297 : : 0, NULL);
3298 [ + + ]: 548 : else if (artifacttype == "executable")
3299 [ + - ]: 546 : fd = debuginfod_find_executable (client,
3300 [ + - ]: 546 : (const unsigned char*) buildid.c_str(),
3301 : : 0, NULL);
3302 [ + - ]: 2 : else if (artifacttype == "source")
3303 [ + - ]: 2 : fd = debuginfod_find_source (client,
3304 [ + - ]: 2 : (const unsigned char*) buildid.c_str(),
3305 : : 0, suffix.c_str(), NULL);
3306 [ # # ]: 0 : else if (artifacttype == "section")
3307 [ # # ]: 0 : fd = debuginfod_find_section (client,
3308 [ # # ]: 0 : (const unsigned char*) buildid.c_str(),
3309 : : 0, section.c_str(), NULL);
3310 : :
3311 [ + + ]: 636 : if (fd >= 0)
3312 : : {
3313 [ + - ]: 4 : if (conn != 0)
3314 [ + - + - : 644 : inc_metric ("http_responses_total","result","upstream");
+ - + - -
+ - + - -
- - ]
3315 : 4 : struct stat s;
3316 : 4 : int rc = fstat (fd, &s);
3317 [ + - ]: 4 : if (rc == 0)
3318 : : {
3319 [ + - ]: 4 : auto r = MHD_create_response_from_fd ((uint64_t) s.st_size, fd);
3320 [ + - ]: 4 : if (r)
3321 : : {
3322 [ + - ]: 4 : add_mhd_response_header (r, "Content-Type",
3323 : : "application/octet-stream");
3324 : : // Copy the incoming headers
3325 [ + - ]: 4 : const char * hdrs = debuginfod_get_headers(client);
3326 [ + - ]: 4 : string header_dup;
3327 [ + - ]: 4 : if (hdrs)
3328 [ + - - + ]: 4 : header_dup = string(hdrs);
3329 : : // Parse the "header: value\n" lines into (h,v) tuples and pass on
3330 : 20 : while(1)
3331 : : {
3332 : 12 : size_t newline = header_dup.find('\n');
3333 [ + + ]: 12 : if (newline == string::npos) break;
3334 : 8 : size_t colon = header_dup.find(':');
3335 [ + - ]: 8 : if (colon == string::npos) break;
3336 [ + - ]: 8 : string header = header_dup.substr(0,colon);
3337 [ + - ]: 8 : string value = header_dup.substr(colon+1,newline-colon-1);
3338 : : // strip leading spaces from value
3339 : 8 : size_t nonspace = value.find_first_not_of(" ");
3340 [ + - ]: 8 : if (nonspace != string::npos)
3341 [ + - + + ]: 12 : value = value.substr(nonspace);
3342 [ + - ]: 8 : add_mhd_response_header(r, header.c_str(), value.c_str());
3343 [ + - + + : 12 : header_dup = header_dup.substr(newline+1);
+ + - - ]
3344 [ + - ]: 16 : }
3345 : :
3346 [ + - ]: 4 : add_mhd_last_modified (r, s.st_mtime);
3347 [ + - ]: 4 : if (verbose > 1)
3348 [ + - + - : 8 : obatched(clog) << "serving file from upstream debuginfod/cache" << endl;
- - ]
3349 [ + - ]: 4 : if (result_fd)
3350 : 4 : *result_fd = fd;
3351 [ + - ]: 4 : return r; // NB: don't close fd; libmicrohttpd will
3352 : 4 : }
3353 : : }
3354 [ # # ]: 0 : close (fd);
3355 : : }
3356 : : else
3357 [ + + ]: 632 : switch(fd)
3358 : : {
3359 : : case -ENOSYS:
3360 : : break;
3361 : : case -ENOENT:
3362 : : break;
3363 : 532 : default: // some more tricky error
3364 [ + - + - ]: 1064 : throw libc_exception(-fd, "upstream debuginfod query failed");
3365 : : }
3366 : :
3367 [ + - - + ]: 200 : throw reportable_exception(MHD_HTTP_NOT_FOUND, "not found");
3368 [ - + - + ]: 2344 : }
3369 : :
3370 : :
3371 : : ////////////////////////////////////////////////////////////////////////
3372 : :
3373 : : static map<string,double> metrics; // arbitrary data for /metrics query
3374 : : // NB: store int64_t since all our metrics are integers; prometheus accepts double
3375 : : static mutex metrics_lock;
3376 : : // NB: these objects get released during the process exit via global dtors
3377 : : // do not call them from within other global dtors
3378 : :
3379 : : // utility function for assembling prometheus-compatible
3380 : : // name="escaped-value" strings
3381 : : // https://prometheus.io/docs/instrumenting/exposition_formats/
3382 : : static string
3383 : 281974 : metric_label(const string& name, const string& value)
3384 : : {
3385 : 281974 : string x = name + "=\"";
3386 [ + + ]: 4572062 : for (auto&& c : value)
3387 [ - - - + ]: 4290154 : switch(c)
3388 : : {
3389 [ # # ]: 0 : case '\\': x += "\\\\"; break;
3390 [ # # ]: 0 : case '\"': x += "\\\""; break;
3391 [ # # ]: 0 : case '\n': x += "\\n"; break;
3392 [ + - ]: 8580261 : default: x += c; break;
3393 : : }
3394 [ + - ]: 281908 : x += "\"";
3395 : 281930 : return x;
3396 : 0 : }
3397 : :
3398 : :
3399 : : // add prometheus-format metric name + label tuple (if any) + value
3400 : :
3401 : : static void
3402 : 1560 : set_metric(const string& metric, double value)
3403 : : {
3404 : 1560 : unique_lock<mutex> lock(metrics_lock);
3405 [ + - ]: 1560 : metrics[metric] = value;
3406 : 1560 : }
3407 : : static void
3408 : 784 : inc_metric(const string& metric)
3409 : : {
3410 : 784 : unique_lock<mutex> lock(metrics_lock);
3411 [ + - ]: 784 : metrics[metric] ++;
3412 : 784 : }
3413 : : static void
3414 : 5742 : set_metric(const string& metric,
3415 : : const string& lname, const string& lvalue,
3416 : : double value)
3417 : : {
3418 [ + - + - : 11484 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + - + +
+ - - ]
3419 [ + - ]: 5742 : unique_lock<mutex> lock(metrics_lock);
3420 [ + - ]: 5742 : metrics[key] = value;
3421 [ + - ]: 11484 : }
3422 : :
3423 : : static void
3424 : 126032 : inc_metric(const string& metric,
3425 : : const string& lname, const string& lvalue)
3426 : : {
3427 [ + - + - : 271462 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + + + +
+ - - ]
3428 [ + - ]: 126034 : unique_lock<mutex> lock(metrics_lock);
3429 [ + - ]: 126034 : metrics[key] ++;
3430 [ + - ]: 252067 : }
3431 : : static void
3432 : 123199 : add_metric(const string& metric,
3433 : : const string& lname, const string& lvalue,
3434 : : double value)
3435 : : {
3436 [ + - + - : 265796 : string key = (metric + "{" + metric_label(lname, lvalue) + "}");
- + + + +
+ - - ]
3437 [ + - ]: 123202 : unique_lock<mutex> lock(metrics_lock);
3438 [ + - ]: 123204 : metrics[key] += value;
3439 [ + - ]: 246408 : }
3440 : : static void
3441 : 784 : add_metric(const string& metric,
3442 : : double value)
3443 : : {
3444 : 784 : unique_lock<mutex> lock(metrics_lock);
3445 [ + - ]: 784 : metrics[metric] += value;
3446 : 784 : }
3447 : :
3448 : :
3449 : : // and more for higher arity labels if needed
3450 : :
3451 : : static void
3452 : 6756 : inc_metric(const string& metric,
3453 : : const string& lname, const string& lvalue,
3454 : : const string& rname, const string& rvalue)
3455 : : {
3456 [ + - - + : 13512 : string key = (metric + "{"
- - ]
3457 [ + - + - : 27024 : + metric_label(lname, lvalue) + ","
- + - + +
+ - - ]
3458 [ + - - + : 20268 : + metric_label(rname, rvalue) + "}");
- + ]
3459 [ + - ]: 6756 : unique_lock<mutex> lock(metrics_lock);
3460 [ + - ]: 6756 : metrics[key] ++;
3461 [ + - ]: 13512 : }
3462 : : static void
3463 : 6756 : add_metric(const string& metric,
3464 : : const string& lname, const string& lvalue,
3465 : : const string& rname, const string& rvalue,
3466 : : double value)
3467 : : {
3468 [ + - - + : 13512 : string key = (metric + "{"
- - ]
3469 [ + - + - : 27024 : + metric_label(lname, lvalue) + ","
- + - + +
+ - - ]
3470 [ + - - + : 20268 : + metric_label(rname, rvalue) + "}");
- + ]
3471 [ + - ]: 6756 : unique_lock<mutex> lock(metrics_lock);
3472 [ + - ]: 6756 : metrics[key] += value;
3473 [ + - ]: 13512 : }
3474 : :
3475 : : static struct MHD_Response*
3476 : 708 : handle_metrics (off_t* size)
3477 : : {
3478 : 708 : stringstream o;
3479 : 708 : {
3480 [ + - ]: 708 : unique_lock<mutex> lock(metrics_lock);
3481 [ + + ]: 76206 : for (auto&& i : metrics)
3482 [ + - ]: 75498 : o << i.first
3483 : : << " "
3484 [ + - + - ]: 75498 : << std::setprecision(std::numeric_limits<double>::digits10 + 1)
3485 [ + - + - ]: 75498 : << i.second
3486 : 75498 : << endl;
3487 : 0 : }
3488 [ + - ]: 708 : const string& os = o.str();
3489 [ + - ]: 708 : MHD_Response* r = MHD_create_response_from_buffer (os.size(),
3490 [ + - ]: 708 : (void*) os.c_str(),
3491 : : MHD_RESPMEM_MUST_COPY);
3492 [ + - ]: 708 : if (r != NULL)
3493 : : {
3494 [ + - ]: 708 : *size = os.size();
3495 [ + - ]: 708 : add_mhd_response_header (r, "Content-Type", "text/plain");
3496 : : }
3497 [ + - ]: 1416 : return r;
3498 : 708 : }
3499 : :
3500 : :
3501 : : static struct MHD_Response*
3502 : 26 : handle_metadata (MHD_Connection* conn,
3503 : : string key, string value, off_t* size)
3504 : : {
3505 : 26 : MHD_Response* r;
3506 : 26 : sqlite3 *thisdb = dbq;
3507 : :
3508 : : // Query locally for matching e, d files
3509 : 26 : string op;
3510 [ + + ]: 26 : if (key == "glob")
3511 [ + - ]: 22 : op = "glob";
3512 [ + - ]: 4 : else if (key == "file")
3513 [ + - ]: 4 : op = "=";
3514 : : else
3515 [ # # # # ]: 0 : throw reportable_exception("/metadata webapi error, unsupported key");
3516 : :
3517 : : // Since PR30378, the file names are segmented into two tables. We
3518 : : // could do a glob/= search over the _files_v view that combines
3519 : : // them, but that means that the entire _files_v thing has to be
3520 : : // materialized & scanned to do the query. Slow! Instead, we can
3521 : : // segment the incoming file/glob pattern into dirname / basename
3522 : : // parts, and apply them to the corresponding table. This is done
3523 : : // by splitting the value at the last "/". If absent, the same
3524 : : // convention as is used in register_file_name().
3525 : :
3526 : 26 : string dirname, bname; // basename is a "poisoned" identifier on some distros
3527 : 26 : size_t slash = value.rfind('/');
3528 [ - + ]: 26 : if (slash == std::string::npos) {
3529 [ # # ]: 0 : dirname = "";
3530 [ # # ]: 0 : bname = value;
3531 : : } else {
3532 [ + - - + ]: 26 : dirname = value.substr(0, slash);
3533 [ + - - + ]: 26 : bname = value.substr(slash+1);
3534 : : }
3535 : :
3536 : : // NB: further optimization is possible: replacing the 'glob' op
3537 : : // with simple equality, if the corresponding value segment lacks
3538 : : // metacharacters. sqlite may or may not be smart enough to do so,
3539 : : // so we help out.
3540 [ + - - - ]: 26 : string metacharacters = "[]*?";
3541 [ + + + + : 48 : string dop = (op == "glob" && dirname.find_first_of(metacharacters) == string::npos) ? "=" : op;
+ - + - -
- ]
3542 [ + + - + : 48 : string bop = (op == "glob" && bname.find_first_of(metacharacters) == string::npos) ? "=" : op;
- - + - -
- ]
3543 : :
3544 : 26 : string sql = string(
3545 : : // explicit query r_de and f_de once here, rather than the query_d and query_e
3546 : : // separately, because they scan the same tables, so we'd double the work
3547 : : "select d1.executable_p, d1.debuginfo_p, 0 as source_p, "
3548 : : " b1.hex, f1d.name || '/' || f1b.name as file, a1.name as archive "
3549 : : "from " BUILDIDS "_r_de d1, " BUILDIDS "_files f1, " BUILDIDS "_fileparts f1b, " BUILDIDS "_fileparts f1d, "
3550 : : BUILDIDS "_buildids b1, " BUILDIDS "_files_v a1 "
3551 : : "where f1.id = d1.content and a1.id = d1.file and d1.buildid = b1.id "
3552 [ + - + - : 78 : " and f1d.name " + dop + " ? and f1b.name " + bop + " ? and f1.dirname = f1d.id and f1.basename = f1b.id "
- + - + -
+ ]
3553 : : "union all \n"
3554 : : "select d2.executable_p, d2.debuginfo_p, 0, "
3555 : : " b2.hex, f2d.name || '/' || f2b.name, NULL "
3556 : : "from " BUILDIDS "_f_de d2, " BUILDIDS "_files f2, " BUILDIDS "_fileparts f2b, " BUILDIDS "_fileparts f2d, "
3557 : : BUILDIDS "_buildids b2 "
3558 : : "where f2.id = d2.file and d2.buildid = b2.id "
3559 [ + - + - : 78 : " and f2d.name " + dop + " ? and f2b.name " + bop + " ? "
- + - + -
+ - - ]
3560 [ - + ]: 26 : " and f2.dirname = f2d.id and f2.basename = f2b.id");
3561 : :
3562 : : // NB: we could query source file names too, thusly:
3563 : : //
3564 : : // select * from " BUILDIDS "_buildids b, " BUILDIDS "_files_v f1, " BUILDIDS "_r_sref sr
3565 : : // where b.id = sr.buildid and f1.id = sr.artifactsrc and f1.name " + op + "?"
3566 : : // UNION ALL something with BUILDIDS "_f_s"
3567 : : //
3568 : : // But the first part of this query cannot run fast without the same index temp-created
3569 : : // during "maxigroom":
3570 : : // create index " BUILDIDS "_r_sref_arc on " BUILDIDS "_r_sref(artifactsrc);
3571 : : // and unfortunately this index is HUGE. It's similar to the size of the _r_sref
3572 : : // table, which is already the largest part of a debuginfod index. Adding that index
3573 : : // would nearly double the .sqlite db size.
3574 : :
3575 [ + - + - : 26 : sqlite_ps *pp = new sqlite_ps (thisdb, "mhd-query-meta-glob", sql);
+ - + - ]
3576 [ + - ]: 26 : pp->reset();
3577 [ + - ]: 26 : pp->bind(1, dirname);
3578 [ + - ]: 26 : pp->bind(2, bname);
3579 [ + - ]: 26 : pp->bind(3, dirname);
3580 [ + - ]: 26 : pp->bind(4, bname);
3581 [ + - ]: 26 : unique_ptr<sqlite_ps> ps_closer(pp); // release pp if exception or return
3582 : :
3583 [ + - ]: 26 : json_object *metadata = json_object_new_object();
3584 [ - + - - : 26 : if (!metadata) throw libc_exception(ENOMEM, "json allocation");
- - ]
3585 : 26 : defer_dtor<json_object*,int> metadata_d(metadata, json_object_put);
3586 [ + - ]: 26 : json_object *metadata_arr = json_object_new_array();
3587 [ - + - - : 26 : if (!metadata_arr) throw libc_exception(ENOMEM, "json allocation");
- - ]
3588 [ + - ]: 26 : json_object_object_add(metadata, "results", metadata_arr);
3589 : : // consume all the rows
3590 : 26 : struct timespec ts_start;
3591 : 26 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
3592 : :
3593 : 26 : int rc;
3594 : 26 : bool metadata_complete = true;
3595 [ + - + + ]: 68 : while (SQLITE_DONE != (rc = pp->step()))
3596 : : {
3597 : : // break out of loop if we have searched too long
3598 : 16 : struct timespec ts_end;
3599 : 16 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
3600 : 16 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
3601 [ + - - + ]: 16 : if (metadata_maxtime_s > 0 && deltas > metadata_maxtime_s)
3602 : : {
3603 : 0 : metadata_complete = false;
3604 : 0 : break;
3605 : : }
3606 : :
3607 [ - + - - : 16 : if (rc != SQLITE_ROW) throw sqlite_exception(rc, "step");
- - ]
3608 : :
3609 [ + - ]: 16 : int m_executable_p = sqlite3_column_int (*pp, 0);
3610 [ + - ]: 16 : int m_debuginfo_p = sqlite3_column_int (*pp, 1);
3611 [ + - ]: 16 : int m_source_p = sqlite3_column_int (*pp, 2);
3612 [ + - - + : 16 : string m_buildid = (const char*) sqlite3_column_text (*pp, 3) ?: ""; // should always be non-null
+ - ]
3613 [ + - - + : 16 : string m_file = (const char*) sqlite3_column_text (*pp, 4) ?: "";
+ - - - ]
3614 [ + - - + : 16 : string m_archive = (const char*) sqlite3_column_text (*pp, 5) ?: "";
+ - - - ]
3615 : :
3616 : : // Confirm that m_file matches in the fnmatch(FNM_PATHNAME)
3617 : : // sense, since sqlite's GLOB operator is a looser filter.
3618 [ + - + - : 16 : if (key == "glob" && fnmatch(value.c_str(), m_file.c_str(), FNM_PATHNAME) != 0)
- + ]
3619 [ # # ]: 0 : continue;
3620 : :
3621 : 48 : auto add_metadata = [metadata_arr, m_buildid, m_file, m_archive](const string& type) {
3622 : 16 : json_object* entry = json_object_new_object();
3623 [ - + - - : 16 : if (NULL == entry) throw libc_exception (ENOMEM, "cannot allocate json");
- - ]
3624 : 16 : defer_dtor<json_object*,int> entry_d(entry, json_object_put);
3625 : :
3626 : 144 : auto add_entry_metadata = [entry](const char* k, string v) {
3627 : 64 : json_object* s;
3628 [ + - ]: 64 : if(v != "") {
3629 : 64 : s = json_object_new_string(v.c_str());
3630 [ - + - - : 64 : if (NULL == s) throw libc_exception (ENOMEM, "cannot allocate json");
- - ]
3631 : 64 : json_object_object_add(entry, k, s);
3632 : : }
3633 : 64 : };
3634 : :
3635 [ + - + - ]: 16 : add_entry_metadata("type", type.c_str());
3636 [ + - + - ]: 16 : add_entry_metadata("buildid", m_buildid);
3637 [ + - + - ]: 16 : add_entry_metadata("file", m_file);
3638 [ + - + - : 32 : if (m_archive != "") add_entry_metadata("archive", m_archive);
+ - ]
3639 [ - + ]: 16 : if (verbose > 3)
3640 [ # # ]: 0 : obatched(clog) << "metadata found local "
3641 : : << json_object_to_json_string_ext(entry,
3642 [ # # # # : 0 : JSON_C_TO_STRING_PRETTY)
# # ]
3643 : 0 : << endl;
3644 : :
3645 : : // Increase ref count to switch its ownership
3646 [ + - + - ]: 16 : json_object_array_add(metadata_arr, json_object_get(entry));
3647 [ + - + - : 32 : };
+ - ]
3648 : :
3649 [ + - + - : 32 : if (m_executable_p) add_metadata("executable");
+ - ]
3650 [ - + - - : 16 : if (m_debuginfo_p) add_metadata("debuginfo");
- - ]
3651 [ - + - - : 16 : if (m_source_p) add_metadata("source");
- - ]
3652 [ - - - - : 64 : }
+ - + - +
- ]
3653 [ + - ]: 26 : pp->reset();
3654 : :
3655 [ + - ]: 26 : unsigned num_local_results = json_object_array_length(metadata_arr);
3656 : :
3657 : : // Query upstream as well
3658 [ + - ]: 26 : debuginfod_client *client = debuginfod_pool_begin();
3659 [ + - ]: 26 : if (client != NULL)
3660 : : {
3661 [ + - ]: 26 : add_client_federation_headers(client, conn);
3662 : :
3663 : 26 : int upstream_metadata_fd;
3664 : 26 : char *upstream_metadata_file = NULL;
3665 [ + - ]: 26 : upstream_metadata_fd = debuginfod_find_metadata(client, key.c_str(), (char*)value.c_str(),
3666 : : &upstream_metadata_file);
3667 [ + + ]: 26 : if (upstream_metadata_fd >= 0) {
3668 : : /* json-c >= 0.13 has json_object_from_fd(). */
3669 [ + - ]: 16 : json_object *upstream_metadata_json = json_object_from_file(upstream_metadata_file);
3670 : 16 : free (upstream_metadata_file);
3671 : 16 : json_object *upstream_metadata_json_arr;
3672 : 16 : json_object *upstream_complete;
3673 [ - + ]: 16 : if (NULL != upstream_metadata_json &&
3674 [ + - + - : 32 : json_object_object_get_ex(upstream_metadata_json, "results", &upstream_metadata_json_arr) &&
- + ]
3675 [ + - ]: 16 : json_object_object_get_ex(upstream_metadata_json, "complete", &upstream_complete))
3676 : : {
3677 [ + - ]: 16 : metadata_complete &= json_object_get_boolean(upstream_complete);
3678 [ + - + + ]: 20 : for (int i = 0, n = json_object_array_length(upstream_metadata_json_arr); i < n; i++)
3679 : : {
3680 [ + - ]: 4 : json_object *entry = json_object_array_get_idx(upstream_metadata_json_arr, i);
3681 [ - + ]: 4 : if (verbose > 3)
3682 [ # # ]: 0 : obatched(clog) << "metadata found remote "
3683 : : << json_object_to_json_string_ext(entry,
3684 [ # # # # : 0 : JSON_C_TO_STRING_PRETTY)
# # ]
3685 : 0 : << endl;
3686 : :
3687 [ + - ]: 4 : json_object_get(entry); // increment reference count
3688 [ + - ]: 4 : json_object_array_add(metadata_arr, entry);
3689 : : }
3690 [ + - ]: 16 : json_object_put(upstream_metadata_json);
3691 : : }
3692 [ + - ]: 16 : close(upstream_metadata_fd);
3693 : : }
3694 [ + - ]: 26 : debuginfod_pool_end (client);
3695 : : }
3696 : :
3697 [ + - ]: 26 : unsigned num_total_results = json_object_array_length(metadata_arr);
3698 : :
3699 [ + - ]: 26 : if (verbose > 2)
3700 [ + - + - ]: 78 : obatched(clog) << "metadata found local=" << num_local_results
3701 [ + - + - ]: 26 : << " remote=" << (num_total_results-num_local_results)
3702 [ + - + - : 26 : << " total=" << num_total_results
+ - ]
3703 : 26 : << endl;
3704 : :
3705 [ + - + - ]: 26 : json_object_object_add(metadata, "complete", json_object_new_boolean(metadata_complete));
3706 [ + - ]: 26 : const char* metadata_str = json_object_to_json_string(metadata);
3707 [ - + ]: 26 : if (!metadata_str)
3708 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate json");
3709 [ + - ]: 26 : r = MHD_create_response_from_buffer (strlen(metadata_str),
3710 : : (void*) metadata_str,
3711 : : MHD_RESPMEM_MUST_COPY);
3712 : 26 : *size = strlen(metadata_str);
3713 [ + - ]: 26 : if (r)
3714 [ + - ]: 26 : add_mhd_response_header(r, "Content-Type", "application/json");
3715 : 26 : return r;
3716 [ + - - + : 52 : }
- + - + -
+ - + -
+ ]
3717 : :
3718 : :
3719 : : static struct MHD_Response*
3720 : 0 : handle_root (off_t* size)
3721 : : {
3722 [ # # # # : 0 : static string version = "debuginfod (" + string (PACKAGE_NAME) + ") "
# # # # #
# # # ]
3723 [ # # # # : 0 : + string (PACKAGE_VERSION);
# # # # #
# ]
3724 : 0 : MHD_Response* r = MHD_create_response_from_buffer (version.size (),
3725 : 0 : (void *) version.c_str (),
3726 : : MHD_RESPMEM_PERSISTENT);
3727 [ # # ]: 0 : if (r != NULL)
3728 : : {
3729 : 0 : *size = version.size ();
3730 : 0 : add_mhd_response_header (r, "Content-Type", "text/plain");
3731 : : }
3732 : 0 : return r;
3733 : : }
3734 : :
3735 : :
3736 : : ////////////////////////////////////////////////////////////////////////
3737 : :
3738 : :
3739 : : /* libmicrohttpd callback */
3740 : : static MHD_RESULT
3741 : 4504 : handler_cb (void * /*cls*/,
3742 : : struct MHD_Connection *connection,
3743 : : const char *url,
3744 : : const char *method,
3745 : : const char * /*version*/,
3746 : : const char * /*upload_data*/,
3747 : : size_t * /*upload_data_size*/,
3748 : : void ** ptr)
3749 : : {
3750 : 4504 : struct MHD_Response *r = NULL;
3751 : 4504 : string url_copy = url;
3752 : :
3753 : : /* libmicrohttpd always makes (at least) two callbacks: once just
3754 : : past the headers, and one after the request body is finished
3755 : : being received. If we process things early (first callback) and
3756 : : queue a response, libmicrohttpd would suppress http keep-alive
3757 : : (via connection->read_closed = true). */
3758 : 4503 : static int aptr; /* just some random object to use as a flag */
3759 [ + + ]: 4503 : if (&aptr != *ptr)
3760 : : {
3761 : : /* do never respond on first call */
3762 : 2251 : *ptr = &aptr;
3763 : 2251 : return MHD_YES;
3764 : : }
3765 : 2252 : *ptr = NULL; /* reset when done */
3766 : :
3767 [ + - ]: 2252 : const char *maxsize_string = MHD_lookup_connection_value(connection, MHD_HEADER_KIND, "X-DEBUGINFOD-MAXSIZE");
3768 : 2251 : long maxsize = 0;
3769 [ + + + - ]: 2251 : if (maxsize_string != NULL && maxsize_string[0] != '\0')
3770 : 2 : maxsize = atol(maxsize_string);
3771 : : else
3772 : : maxsize = 0;
3773 : :
3774 : : #if MHD_VERSION >= 0x00097002
3775 : 2251 : enum MHD_Result rc;
3776 : : #else
3777 : : int rc = MHD_NO; // mhd
3778 : : #endif
3779 : 2251 : int http_code = 500;
3780 : 2251 : off_t http_size = -1;
3781 : 2251 : struct timespec ts_start, ts_end;
3782 : 2251 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
3783 : 2252 : double afteryou = 0.0;
3784 [ + - ]: 2252 : string artifacttype, suffix;
3785 : 2252 : string urlargs; // for logging
3786 : :
3787 : 2252 : try
3788 : : {
3789 [ + - - + : 5133 : if (string(method) != "GET")
- + ]
3790 [ # # # # ]: 0 : throw reportable_exception(400, "we support GET only");
3791 : :
3792 : : /* Start decoding the URL. */
3793 : 2251 : size_t slash1 = url_copy.find('/', 1);
3794 [ + - ]: 2252 : string url1 = url_copy.substr(0, slash1); // ok even if slash1 not found
3795 : :
3796 [ + + + - ]: 3764 : if (slash1 != string::npos && url1 == "/buildid")
3797 : : {
3798 : : // PR27863: block this thread awhile if another thread is already busy
3799 : : // fetching the exact same thing. This is better for Everyone.
3800 : : // The latecomer says "... after you!" and waits.
3801 [ + - + - : 3646 : add_metric ("thread_busy", "role", "http-buildid-after-you", 1);
+ - + - -
+ + - - -
- - ]
3802 : : #ifdef HAVE_PTHREAD_SETNAME_NP
3803 : 1512 : (void) pthread_setname_np (pthread_self(), "mhd-buildid-after-you");
3804 : : #endif
3805 : 1512 : struct timespec tsay_start, tsay_end;
3806 : 1512 : clock_gettime (CLOCK_MONOTONIC, &tsay_start);
3807 [ + + + - ]: 1512 : static unique_set<string> busy_urls;
3808 [ + - ]: 1512 : unique_set_reserver<string> after_you(busy_urls, url_copy);
3809 : 1512 : clock_gettime (CLOCK_MONOTONIC, &tsay_end);
3810 : 1512 : afteryou = (tsay_end.tv_sec - tsay_start.tv_sec) + (tsay_end.tv_nsec - tsay_start.tv_nsec)/1.e9;
3811 [ + - + - : 3024 : add_metric ("thread_busy", "role", "http-buildid-after-you", -1);
+ - + - -
+ + - - -
- - ]
3812 : :
3813 [ + - + - : 3024 : tmp_inc_metric m ("thread_busy", "role", "http-buildid");
+ - + - -
+ - + - -
- - ]
3814 : : #ifdef HAVE_PTHREAD_SETNAME_NP
3815 : 1512 : (void) pthread_setname_np (pthread_self(), "mhd-buildid");
3816 : : #endif
3817 : 1512 : size_t slash2 = url_copy.find('/', slash1+1);
3818 [ - + ]: 1512 : if (slash2 == string::npos)
3819 [ # # # # ]: 0 : throw reportable_exception("/buildid/ webapi error, need buildid");
3820 : :
3821 [ + - ]: 1512 : string buildid = url_copy.substr(slash1+1, slash2-slash1-1);
3822 : :
3823 : 1512 : size_t slash3 = url_copy.find('/', slash2+1);
3824 : :
3825 [ + + ]: 1512 : if (slash3 == string::npos)
3826 : : {
3827 [ + - - + ]: 1458 : artifacttype = url_copy.substr(slash2+1);
3828 [ + - ]: 1458 : suffix = "";
3829 : : }
3830 : : else
3831 : : {
3832 [ + - - + ]: 54 : artifacttype = url_copy.substr(slash2+1, slash3-slash2-1);
3833 [ + - - + : 676 : suffix = url_copy.substr(slash3); // include the slash in the suffix
+ + ]
3834 : : }
3835 : :
3836 : : // get the resulting fd so we can report its size
3837 : 1512 : int fd;
3838 [ + + ]: 1512 : r = handle_buildid(connection, buildid, artifacttype, suffix, &fd);
3839 [ + - ]: 890 : if (r)
3840 : : {
3841 : 890 : struct stat fs;
3842 [ + - ]: 890 : if (fstat(fd, &fs) == 0)
3843 : 890 : http_size = fs.st_size;
3844 : : // libmicrohttpd will close (fd);
3845 : : }
3846 : 2134 : }
3847 [ + + ]: 740 : else if (url1 == "/metrics")
3848 : : {
3849 [ + - + - : 1416 : tmp_inc_metric m ("thread_busy", "role", "http-metrics");
+ - + - -
+ - + - -
- - ]
3850 [ + - ]: 708 : artifacttype = "metrics";
3851 [ + - + - : 1416 : inc_metric("http_requests_total", "type", artifacttype);
+ - - + -
- ]
3852 [ + - ]: 708 : r = handle_metrics(& http_size);
3853 : 708 : }
3854 [ + + ]: 32 : else if (url1 == "/metadata")
3855 : : {
3856 [ + - + - : 52 : tmp_inc_metric m ("thread_busy", "role", "http-metadata");
+ - + - -
+ - + - -
- - ]
3857 [ + - ]: 26 : const char* key = MHD_lookup_connection_value(connection, MHD_GET_ARGUMENT_KIND, "key");
3858 [ + - ]: 26 : const char* value = MHD_lookup_connection_value(connection, MHD_GET_ARGUMENT_KIND, "value");
3859 [ - + ]: 26 : if (NULL == value || NULL == key)
3860 [ # # # # ]: 0 : throw reportable_exception("/metadata webapi error, need key and value");
3861 : :
3862 [ + - + - : 26 : urlargs = string("?key=") + string(key) + string("&value=") + string(value); // apprx., for logging
+ - + - +
- + - + -
- + - + -
+ - + - +
- + + + -
- - - -
- ]
3863 [ + - ]: 26 : artifacttype = "metadata";
3864 [ + - + - : 52 : inc_metric("http_requests_total", "type", artifacttype);
+ - - + -
- ]
3865 [ + - + - : 26 : r = handle_metadata(connection, key, value, &http_size);
+ - - + +
+ - - ]
3866 : 26 : }
3867 [ - + ]: 6 : else if (url1 == "/")
3868 : : {
3869 [ # # ]: 0 : artifacttype = "/";
3870 [ - - - - : 630 : inc_metric("http_requests_total", "type", artifacttype);
- - - - -
- - + ]
3871 [ # # ]: 0 : r = handle_root(& http_size);
3872 : : }
3873 : : else
3874 [ + - + - : 18 : throw reportable_exception("webapi error, unrecognized '" + url1 + "'");
+ - - + ]
3875 : :
3876 [ - + ]: 1624 : if (r == 0)
3877 [ # # # # ]: 0 : throw reportable_exception("internal error, missing response");
3878 : :
3879 [ + + + - ]: 1624 : if (maxsize > 0 && http_size > maxsize)
3880 : : {
3881 [ + - ]: 2 : MHD_destroy_response(r);
3882 [ + - + - : 6 : throw reportable_exception(406, "File too large, max size=" + std::to_string(maxsize));
+ - - + ]
3883 : : }
3884 : :
3885 [ + - ]: 1622 : rc = MHD_queue_response (connection, MHD_HTTP_OK, r);
3886 : 1622 : http_code = MHD_HTTP_OK;
3887 [ + - ]: 1622 : MHD_destroy_response (r);
3888 : 2252 : }
3889 [ - + ]: 630 : catch (const reportable_exception& e)
3890 : : {
3891 [ + - + - : 1260 : inc_metric("http_responses_total","result","error");
+ - + - -
+ - + - -
- - ]
3892 [ + - ]: 630 : e.report(clog);
3893 : 630 : http_code = e.code;
3894 [ + - ]: 630 : http_size = e.message.size();
3895 [ + - ]: 630 : rc = e.mhd_send_response (connection);
3896 : 630 : }
3897 : :
3898 : 2252 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
3899 : 2252 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
3900 : : // afteryou: delay waiting for other client's identical query to complete
3901 : : // deltas: total latency, including afteryou waiting
3902 [ + - + - : 4504 : obatched(clog) << conninfo(connection)
- - ]
3903 : : << ' ' << method << ' ' << url << urlargs
3904 [ + - + - : 2252 : << ' ' << http_code << ' ' << http_size
+ - + - +
- + - + -
+ - ]
3905 [ + - + - : 2252 : << ' ' << (int)(afteryou*1000) << '+' << (int)((deltas-afteryou)*1000) << "ms"
+ - + - +
- + - +
- ]
3906 [ + - ]: 2252 : << endl;
3907 : :
3908 : : // related prometheus metrics
3909 : 2252 : string http_code_str = to_string(http_code);
3910 [ + - + - : 4504 : add_metric("http_responses_transfer_bytes_sum",
+ - + - -
+ - + - -
- - ]
3911 : : "code", http_code_str, "type", artifacttype, http_size);
3912 [ + - + - : 4504 : inc_metric("http_responses_transfer_bytes_count",
+ - + - -
+ - + - -
- - ]
3913 : : "code", http_code_str, "type", artifacttype);
3914 : :
3915 [ + - + - : 4504 : add_metric("http_responses_duration_milliseconds_sum",
+ - + - -
+ - + - -
- - ]
3916 : : "code", http_code_str, "type", artifacttype, deltas*1000); // prometheus prefers _seconds and floating point
3917 [ + - + - : 4504 : inc_metric("http_responses_duration_milliseconds_count",
+ - + - -
+ - + - -
- - ]
3918 : : "code", http_code_str, "type", artifacttype);
3919 : :
3920 [ + - + - : 4504 : add_metric("http_responses_after_you_milliseconds_sum",
+ - + - -
+ - + - -
- - ]
3921 : : "code", http_code_str, "type", artifacttype, afteryou*1000);
3922 [ + - + - : 4504 : inc_metric("http_responses_after_you_milliseconds_count",
+ - + - -
+ - + - -
- - - - ]
3923 : : "code", http_code_str, "type", artifacttype);
3924 : :
3925 [ - + ]: 2252 : return rc;
3926 [ + + + + : 9849 : }
- + + + ]
3927 : :
3928 : :
3929 : : ////////////////////////////////////////////////////////////////////////
3930 : : // borrowed originally from src/nm.c get_local_names()
3931 : :
3932 : : static void
3933 : 376 : dwarf_extract_source_paths (Elf *elf, set<string>& debug_sourcefiles)
3934 : : noexcept // no exceptions - so we can simplify the altdbg resource release at end
3935 : : {
3936 : 376 : Dwarf* dbg = dwarf_begin_elf (elf, DWARF_C_READ, NULL);
3937 [ - + ]: 376 : if (dbg == NULL)
3938 : 0 : return;
3939 : :
3940 : 376 : Dwarf* altdbg = NULL;
3941 : 376 : int altdbg_fd = -1;
3942 : :
3943 : : // DWZ handling: if we have an unsatisfied debug-alt-link, add an
3944 : : // empty string into the outgoing sourcefiles set, so the caller
3945 : : // should know that our data is incomplete.
3946 : 376 : const char *alt_name_p;
3947 : 376 : const void *alt_build_id; // elfutils-owned memory
3948 : 376 : ssize_t sz = dwelf_dwarf_gnu_debugaltlink (dbg, &alt_name_p, &alt_build_id);
3949 [ + + ]: 376 : if (sz > 0) // got one!
3950 : : {
3951 : 200 : string buildid;
3952 : 200 : unsigned char* build_id_bytes = (unsigned char*) alt_build_id;
3953 [ + + ]: 4200 : for (ssize_t idx=0; idx<sz; idx++)
3954 : : {
3955 : 4000 : buildid += "0123456789abcdef"[build_id_bytes[idx] >> 4];
3956 : 4000 : buildid += "0123456789abcdef"[build_id_bytes[idx] & 0xf];
3957 : : }
3958 : :
3959 [ + + ]: 200 : if (verbose > 3)
3960 : 156 : obatched(clog) << "Need altdebug buildid=" << buildid << endl;
3961 : :
3962 : : // but is it unsatisfied the normal elfutils ways?
3963 : 200 : Dwarf* alt = dwarf_getalt (dbg);
3964 [ + - ]: 200 : if (alt == NULL)
3965 : : {
3966 : : // Yup, unsatisfied the normal way. Maybe we can satisfy it
3967 : : // from our own debuginfod database.
3968 : 200 : int alt_fd;
3969 : 200 : struct MHD_Response *r = 0;
3970 : 200 : try
3971 : : {
3972 [ + - ]: 200 : string artifacttype = "debuginfo";
3973 [ + - + + : 220 : r = handle_buildid (0, buildid, artifacttype, "", &alt_fd);
- + - + -
+ ]
3974 : 20 : }
3975 [ - + ]: 20 : catch (const reportable_exception& e)
3976 : : {
3977 : : // swallow exceptions
3978 : 20 : }
3979 : :
3980 : : // NB: this is not actually recursive! This invokes the web-query
3981 : : // path, which cannot get back into the scan code paths.
3982 [ + - ]: 180 : if (r)
3983 : : {
3984 : : // Found it!
3985 : 180 : altdbg_fd = dup(alt_fd); // ok if this fails, downstream failures ok
3986 : 180 : alt = altdbg = dwarf_begin (altdbg_fd, DWARF_C_READ);
3987 : : // NB: must close this dwarf and this fd at the bottom of the function!
3988 : 180 : MHD_destroy_response (r); // will close alt_fd
3989 [ + - ]: 180 : if (alt)
3990 : 180 : dwarf_setalt (dbg, alt);
3991 : : }
3992 : : }
3993 : : else
3994 : : {
3995 : : // NB: dwarf_setalt(alt) inappropriate - already done!
3996 : : // NB: altdbg will stay 0 so nothing tries to redundantly dealloc.
3997 : : }
3998 : :
3999 [ + + ]: 200 : if (alt)
4000 : : {
4001 [ + + ]: 180 : if (verbose > 3)
4002 : 156 : obatched(clog) << "Resolved altdebug buildid=" << buildid << endl;
4003 : : }
4004 : : else // (alt == NULL) - signal possible presence of poor debuginfo
4005 : : {
4006 [ - + ]: 20 : debug_sourcefiles.insert("");
4007 [ - + ]: 20 : if (verbose > 3)
4008 : 0 : obatched(clog) << "Unresolved altdebug buildid=" << buildid << endl;
4009 : : }
4010 : 200 : }
4011 : :
4012 : 376 : Dwarf_Off offset = 0;
4013 : 376 : Dwarf_Off old_offset;
4014 : 376 : size_t hsize;
4015 : :
4016 [ + + ]: 2714 : while (dwarf_nextcu (dbg, old_offset = offset, &offset, &hsize, NULL, NULL, NULL) == 0)
4017 : : {
4018 : 2338 : Dwarf_Die cudie_mem;
4019 : 2338 : Dwarf_Die *cudie = dwarf_offdie (dbg, old_offset + hsize, &cudie_mem);
4020 : :
4021 [ - + ]: 2338 : if (cudie == NULL)
4022 : 36 : continue;
4023 [ + + ]: 2338 : if (dwarf_tag (cudie) != DW_TAG_compile_unit)
4024 : 36 : continue;
4025 : :
4026 [ - + ]: 2302 : const char *cuname = dwarf_diename(cudie) ?: "unknown";
4027 : :
4028 : 2302 : Dwarf_Files *files;
4029 : 2302 : size_t nfiles;
4030 [ - + ]: 2302 : if (dwarf_getsrcfiles (cudie, &files, &nfiles) != 0)
4031 : 0 : continue;
4032 : :
4033 : : // extract DW_AT_comp_dir to resolve relative file names
4034 : 2302 : const char *comp_dir = "";
4035 : 2302 : const char *const *dirs;
4036 : 2302 : size_t ndirs;
4037 [ - + ]: 2302 : if (dwarf_getsrcdirs (files, &dirs, &ndirs) == 0 &&
4038 [ - + ]: 2302 : dirs[0] != NULL)
4039 : : comp_dir = dirs[0];
4040 : : if (comp_dir == NULL)
4041 : : comp_dir = "";
4042 : :
4043 [ + + ]: 2302 : if (verbose > 3)
4044 : 536 : obatched(clog) << "searching for sources for cu=" << cuname << " comp_dir=" << comp_dir
4045 : 268 : << " #files=" << nfiles << " #dirs=" << ndirs << endl;
4046 : :
4047 [ + - - - ]: 2302 : if (comp_dir[0] == '\0' && cuname[0] != '/')
4048 : : {
4049 [ # # ]: 0 : if (verbose > 3)
4050 : 0 : obatched(clog) << "skipping cu=" << cuname << " due to empty comp_dir" << endl;
4051 : 0 : continue;
4052 : : }
4053 : :
4054 [ + + ]: 34906 : for (size_t f = 1; f < nfiles; f++)
4055 : : {
4056 : 32604 : const char *hat = dwarf_filesrc (files, f, NULL, NULL);
4057 [ - + ]: 32604 : if (hat == NULL)
4058 : 0 : continue;
4059 : :
4060 [ + + ]: 32604 : if (string(hat) == "<built-in>"
4061 [ + - + - : 97812 : || string_endswith(hat, "<built-in>")) // gcc intrinsics, don't bother record
+ - + + ]
4062 : 478 : continue;
4063 : :
4064 [ + + ]: 32126 : string waldo;
4065 [ + + ]: 32126 : if (hat[0] == '/') // absolute
4066 [ - + ]: 19528 : waldo = (string (hat));
4067 [ + - ]: 12598 : else if (comp_dir[0] != '\0') // comp_dir relative
4068 [ - + - + : 22322 : waldo = (string (comp_dir) + string("/") + string (hat));
- + - + +
+ ]
4069 : : else
4070 : : {
4071 [ # # ]: 0 : if (verbose > 3)
4072 : 0 : obatched(clog) << "skipping hat=" << hat << " due to empty comp_dir" << endl;
4073 [ # # ]: 0 : continue;
4074 : : }
4075 : :
4076 : : // NB: this is the 'waldo' that a dbginfo client will have
4077 : : // to supply for us to give them the file The comp_dir
4078 : : // prefixing is a definite complication. Otherwise we'd
4079 : : // have to return a setof comp_dirs (one per CU!) with
4080 : : // corresponding filesrc[] names, instead of one absolute
4081 : : // resoved set. Maybe we'll have to do that anyway. XXX
4082 : :
4083 [ + + ]: 32126 : if (verbose > 4)
4084 [ - + ]: 32 : obatched(clog) << waldo
4085 [ - + ]: 16 : << (debug_sourcefiles.find(waldo)==debug_sourcefiles.end() ? " new" : " dup") << endl;
4086 : :
4087 [ + - ]: 32126 : debug_sourcefiles.insert (waldo);
4088 : 32604 : }
4089 : : }
4090 : :
4091 : 376 : dwarf_end(dbg);
4092 [ + + ]: 376 : if (altdbg)
4093 : 180 : dwarf_end(altdbg);
4094 [ + + ]: 376 : if (altdbg_fd >= 0)
4095 : 180 : close(altdbg_fd);
4096 : : }
4097 : :
4098 : :
4099 : :
4100 : : static void
4101 : 1434 : elf_classify (int fd, bool &executable_p, bool &debuginfo_p, string &buildid, set<string>& debug_sourcefiles)
4102 : : {
4103 : 1434 : Elf *elf = elf_begin (fd, ELF_C_READ_MMAP_PRIVATE, NULL);
4104 [ + - ]: 1434 : if (elf == NULL)
4105 : : return;
4106 : :
4107 : 1434 : try // catch our types of errors and clean up the Elf* object
4108 : : {
4109 [ + - + + ]: 1434 : if (elf_kind (elf) != ELF_K_ELF)
4110 : : {
4111 [ + - ]: 708 : elf_end (elf);
4112 : 708 : return;
4113 : : }
4114 : :
4115 : 726 : GElf_Ehdr ehdr_storage;
4116 [ + - ]: 726 : GElf_Ehdr *ehdr = gelf_getehdr (elf, &ehdr_storage);
4117 [ - + ]: 726 : if (ehdr == NULL)
4118 : : {
4119 [ # # ]: 0 : elf_end (elf);
4120 : : return;
4121 : : }
4122 : 726 : auto elf_type = ehdr->e_type;
4123 : :
4124 : 726 : const void *build_id; // elfutils-owned memory
4125 [ + - ]: 726 : ssize_t sz = dwelf_elf_gnu_build_id (elf, & build_id);
4126 [ - + ]: 726 : if (sz <= 0)
4127 : : {
4128 : : // It's not a diagnostic-worthy error for an elf file to lack build-id.
4129 : : // It might just be very old.
4130 [ # # ]: 0 : elf_end (elf);
4131 : : return;
4132 : : }
4133 : :
4134 : : // build_id is a raw byte array; convert to hexadecimal *lowercase*
4135 : 726 : unsigned char* build_id_bytes = (unsigned char*) build_id;
4136 [ + + ]: 15237 : for (ssize_t idx=0; idx<sz; idx++)
4137 : : {
4138 [ + - ]: 14511 : buildid += "0123456789abcdef"[build_id_bytes[idx] >> 4];
4139 [ + - ]: 29024 : buildid += "0123456789abcdef"[build_id_bytes[idx] & 0xf];
4140 : : }
4141 : :
4142 : : // now decide whether it's an executable - namely, any allocatable section has
4143 : : // PROGBITS;
4144 [ + + ]: 726 : if (elf_type == ET_EXEC || elf_type == ET_DYN)
4145 : : {
4146 : 664 : size_t shnum;
4147 [ + - ]: 664 : int rc = elf_getshdrnum (elf, &shnum);
4148 [ - + ]: 664 : if (rc < 0)
4149 [ # # # # ]: 0 : throw elfutils_exception(rc, "getshdrnum");
4150 : :
4151 : 664 : executable_p = false;
4152 [ + + ]: 13122 : for (size_t sc = 0; sc < shnum; sc++)
4153 : : {
4154 [ + - ]: 12796 : Elf_Scn *scn = elf_getscn (elf, sc);
4155 [ - + ]: 12795 : if (scn == NULL)
4156 : 0 : continue;
4157 : :
4158 : 12795 : GElf_Shdr shdr_mem;
4159 [ + - ]: 12795 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem);
4160 [ - + ]: 12796 : if (shdr == NULL)
4161 : 0 : continue;
4162 : :
4163 : : // allocated (loadable / vm-addr-assigned) section with available content?
4164 [ + + + + ]: 12796 : if ((shdr->sh_type == SHT_PROGBITS) && (shdr->sh_flags & SHF_ALLOC))
4165 : : {
4166 [ - + ]: 338 : if (verbose > 4)
4167 [ # # # # : 0 : obatched(clog) << "executable due to SHF_ALLOC SHT_PROGBITS sc=" << sc << endl;
# # ]
4168 : 338 : executable_p = true;
4169 : 338 : break; // no need to keep looking for others
4170 : : }
4171 : : } // iterate over sections
4172 : : } // executable_p classification
4173 : :
4174 : : // now decide whether it's a debuginfo - namely, if it has any .debug* or .zdebug* sections
4175 : : // logic mostly stolen from fweimer@redhat.com's elfclassify drafts
4176 : 726 : size_t shstrndx;
4177 [ + - ]: 726 : int rc = elf_getshdrstrndx (elf, &shstrndx);
4178 [ - + ]: 726 : if (rc < 0)
4179 [ # # # # ]: 0 : throw elfutils_exception(rc, "getshdrstrndx");
4180 : :
4181 : : Elf_Scn *scn = NULL;
4182 : : bool symtab_p = false;
4183 : : bool bits_alloc_p = false;
4184 : 38502 : while (true)
4185 : : {
4186 [ + - ]: 19614 : scn = elf_nextscn (elf, scn);
4187 [ + + ]: 19527 : if (scn == NULL)
4188 : : break;
4189 : 19177 : GElf_Shdr shdr_storage;
4190 [ + - ]: 19177 : GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_storage);
4191 [ - + ]: 19177 : if (shdr == NULL)
4192 : : break;
4193 [ + - ]: 19177 : const char *section_name = elf_strptr (elf, shstrndx, shdr->sh_name);
4194 [ - + ]: 19264 : if (section_name == NULL)
4195 : : break;
4196 [ + + ]: 19264 : if (startswith (section_name, ".debug_line") ||
4197 [ - + ]: 18888 : startswith (section_name, ".zdebug_line"))
4198 : : {
4199 : 376 : debuginfo_p = true;
4200 [ + - ]: 376 : if (scan_source_info)
4201 : 376 : dwarf_extract_source_paths (elf, debug_sourcefiles);
4202 : : break; // expecting only one .*debug_line, so no need to look for others
4203 : : }
4204 [ + + ]: 18888 : else if (startswith (section_name, ".debug_") ||
4205 [ + # ]: 17688 : startswith (section_name, ".zdebug_"))
4206 : : {
4207 : 1142 : debuginfo_p = true;
4208 : : // NB: don't break; need to parse .debug_line for sources
4209 : : }
4210 [ + + ]: 17746 : else if (shdr->sh_type == SHT_SYMTAB)
4211 : : {
4212 : : symtab_p = true;
4213 : : }
4214 : 17722 : else if (shdr->sh_type != SHT_NOBITS
4215 [ + + ]: 17722 : && shdr->sh_type != SHT_NOTE
4216 [ + + ]: 8301 : && (shdr->sh_flags & SHF_ALLOC) != 0)
4217 : : {
4218 : 18888 : bits_alloc_p = true;
4219 : : }
4220 : 18888 : }
4221 : :
4222 : : // For more expansive elf/split-debuginfo classification, we
4223 : : // want to identify as debuginfo "strip -s"-produced files
4224 : : // without .debug_info* (like libicudata), but we don't want to
4225 : : // identify "strip -g" executables (with .symtab left there).
4226 [ - + ]: 726 : if (symtab_p && !bits_alloc_p)
4227 : 0 : debuginfo_p = true;
4228 : : }
4229 [ # # ]: 0 : catch (const reportable_exception& e)
4230 : : {
4231 [ # # ]: 0 : e.report(clog);
4232 : 0 : }
4233 : 726 : elf_end (elf);
4234 : : }
4235 : :
4236 : :
4237 : : // Intern the given file name in two parts (dirname & basename) and
4238 : : // return the resulting file's id.
4239 : : static int64_t
4240 : 9461 : register_file_name(sqlite_ps& ps_upsert_fileparts,
4241 : : sqlite_ps& ps_upsert_file,
4242 : : sqlite_ps& ps_lookup_file,
4243 : : const string& name)
4244 : : {
4245 : 9461 : std::size_t slash = name.rfind('/');
4246 [ + + ]: 9461 : string dirname, filename;
4247 [ + + ]: 9461 : if (slash == std::string::npos)
4248 : : {
4249 [ + - ]: 90 : dirname = "";
4250 [ + - ]: 90 : filename = name;
4251 : : }
4252 : : else
4253 : : {
4254 [ + - - + ]: 9371 : dirname = name.substr(0, slash);
4255 [ + - - + : 9371 : filename = name.substr(slash+1);
- - ]
4256 : : }
4257 : : // NB: see also handle_metadata()
4258 : :
4259 : : // intern the two substrings
4260 : 9462 : ps_upsert_fileparts
4261 [ + - ]: 9462 : .reset()
4262 [ + - ]: 9462 : .bind(1, dirname)
4263 [ + - ]: 9462 : .step_ok_done();
4264 : 9462 : ps_upsert_fileparts
4265 [ + - ]: 9462 : .reset()
4266 [ + - ]: 9462 : .bind(1, filename)
4267 [ + - ]: 9462 : .step_ok_done();
4268 : :
4269 : : // intern the tuple
4270 : 9462 : ps_upsert_file
4271 [ + - ]: 9462 : .reset()
4272 [ + - ]: 9462 : .bind(1, dirname)
4273 [ + - ]: 9462 : .bind(2, filename)
4274 [ + - ]: 9462 : .step_ok_done();
4275 : :
4276 : : // look up the tuple's id
4277 : 9462 : ps_lookup_file
4278 [ + - ]: 9462 : .reset()
4279 [ + - ]: 9462 : .bind(1, dirname)
4280 [ + - ]: 9462 : .bind(2, filename);
4281 [ + - ]: 9462 : int rc = ps_lookup_file.step();
4282 [ - + - - : 9462 : if (rc != SQLITE_ROW) throw sqlite_exception(rc, "step");
- - ]
4283 : :
4284 [ + - ]: 9462 : int64_t id = sqlite3_column_int64 (ps_lookup_file, 0);
4285 [ + - ]: 9462 : ps_lookup_file.reset();
4286 [ + + ]: 9462 : return id;
4287 [ + + ]: 18044 : }
4288 : :
4289 : :
4290 : :
4291 : : static void
4292 : 831 : scan_source_file (const string& rps, const stat_t& st,
4293 : : sqlite_ps& ps_upsert_buildids,
4294 : : sqlite_ps& ps_upsert_fileparts,
4295 : : sqlite_ps& ps_upsert_file,
4296 : : sqlite_ps& ps_lookup_file,
4297 : : sqlite_ps& ps_upsert_de,
4298 : : sqlite_ps& ps_upsert_s,
4299 : : sqlite_ps& ps_query,
4300 : : sqlite_ps& ps_scan_done,
4301 : : unsigned& fts_cached,
4302 : : unsigned& fts_executable,
4303 : : unsigned& fts_debuginfo,
4304 : : unsigned& fts_sourcefiles)
4305 : : {
4306 : 831 : int64_t fileid = register_file_name(ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, rps);
4307 : :
4308 : : /* See if we know of it already. */
4309 : 832 : int rc = ps_query
4310 : 832 : .reset()
4311 : 832 : .bind(1, fileid)
4312 : 832 : .bind(2, st.st_mtime)
4313 : 832 : .step();
4314 : 832 : ps_query.reset();
4315 [ + + ]: 832 : if (rc == SQLITE_ROW) // i.e., a result, as opposed to DONE (no results)
4316 : : // no need to recheck a file/version we already know
4317 : : // specifically, no need to elf-begin a file we already determined is non-elf
4318 : : // (so is stored with buildid=NULL)
4319 : : {
4320 : 432 : fts_cached++;
4321 : 432 : return;
4322 : : }
4323 : :
4324 : 400 : bool executable_p = false, debuginfo_p = false; // E and/or D
4325 [ + - ]: 400 : string buildid;
4326 [ + - ]: 400 : set<string> sourcefiles;
4327 : :
4328 [ + - ]: 400 : int fd = open (rps.c_str(), O_RDONLY);
4329 : 400 : try
4330 : : {
4331 [ + - ]: 400 : if (fd >= 0)
4332 [ + - ]: 400 : elf_classify (fd, executable_p, debuginfo_p, buildid, sourcefiles);
4333 : : else
4334 [ # # # # : 0 : throw libc_exception(errno, string("open ") + rps);
# # # # ]
4335 [ + - + - : 800 : add_metric ("scanned_bytes_total","source","file",
+ - - + -
+ - - -
- ]
4336 [ + - ]: 400 : st.st_size);
4337 [ + - + - : 800 : inc_metric ("scanned_files_total","source","file");
+ - + - -
+ - + - -
- - ]
4338 : : }
4339 : : // NB: we catch exceptions here too, so that we can
4340 : : // cache the corrupt-elf case (!executable_p &&
4341 : : // !debuginfo_p) just below, just as if we had an
4342 : : // EPERM error from open(2).
4343 [ - - ]: 0 : catch (const reportable_exception& e)
4344 : : {
4345 [ - - ]: 0 : e.report(clog);
4346 : 0 : }
4347 : :
4348 [ + - ]: 400 : if (fd >= 0)
4349 [ + - ]: 400 : close (fd);
4350 : :
4351 [ + + ]: 400 : if (buildid == "")
4352 : : {
4353 : : // no point storing an elf file without buildid
4354 : 338 : executable_p = false;
4355 : 338 : debuginfo_p = false;
4356 : : }
4357 : : else
4358 : : {
4359 : : // register this build-id in the interning table
4360 : 62 : ps_upsert_buildids
4361 [ + - ]: 62 : .reset()
4362 [ + - ]: 62 : .bind(1, buildid)
4363 [ + - ]: 62 : .step_ok_done();
4364 : : }
4365 : :
4366 [ + + ]: 400 : if (executable_p)
4367 : 38 : fts_executable ++;
4368 [ + + ]: 400 : if (debuginfo_p)
4369 : 38 : fts_debuginfo ++;
4370 [ + + + + ]: 400 : if (executable_p || debuginfo_p)
4371 : : {
4372 : 62 : ps_upsert_de
4373 [ + - ]: 62 : .reset()
4374 [ + - ]: 62 : .bind(1, buildid)
4375 [ + + + - ]: 86 : .bind(2, debuginfo_p ? 1 : 0)
4376 [ + + + - ]: 86 : .bind(3, executable_p ? 1 : 0)
4377 [ + - ]: 62 : .bind(4, fileid)
4378 [ + - ]: 62 : .bind(5, st.st_mtime)
4379 [ + - ]: 62 : .step_ok_done();
4380 : : }
4381 [ + + ]: 400 : if (executable_p)
4382 [ + - + - : 76 : inc_metric("found_executable_total","source","files");
+ - + - -
+ - + - -
- - ]
4383 [ + + ]: 400 : if (debuginfo_p)
4384 [ + - + - : 76 : inc_metric("found_debuginfo_total","source","files");
+ - + - -
+ - + - -
- - ]
4385 : :
4386 [ + + + - ]: 438 : if (sourcefiles.size() && buildid != "")
4387 : : {
4388 : 38 : fts_sourcefiles += sourcefiles.size();
4389 : :
4390 [ + + ]: 3220 : for (auto&& dwarfsrc : sourcefiles)
4391 : : {
4392 [ + - ]: 3182 : char *srp = realpath(dwarfsrc.c_str(), NULL);
4393 [ - + ]: 3182 : if (srp == NULL) // also if DWZ unresolved dwarfsrc=""
4394 : 0 : continue; // unresolvable files are not a serious problem
4395 : : // throw libc_exception(errno, "fts/file realpath " + srcpath);
4396 [ + - ]: 3182 : string srps = string(srp);
4397 : 3182 : free (srp);
4398 : :
4399 : 3182 : struct stat sfs;
4400 : 3182 : rc = stat(srps.c_str(), &sfs);
4401 [ - + ]: 3182 : if (rc != 0)
4402 [ # # ]: 0 : continue;
4403 : :
4404 [ + - ]: 3182 : if (verbose > 2)
4405 [ + - + - : 9546 : obatched(clog) << "recorded buildid=" << buildid << " file=" << srps
- - ]
4406 [ + - + - : 3182 : << " mtime=" << sfs.st_mtime
+ - + - ]
4407 [ + - + - : 3182 : << " as source " << dwarfsrc << endl;
+ - ]
4408 : :
4409 : : // PR25548: store canonicalized dwarfsrc path
4410 [ + - ]: 3182 : string dwarfsrc_canon = canon_pathname (dwarfsrc);
4411 [ + + ]: 3182 : if (dwarfsrc_canon != dwarfsrc)
4412 : : {
4413 [ + + ]: 562 : if (verbose > 3)
4414 [ + - + - : 20 : obatched(clog) << "canonicalized src=" << dwarfsrc << " alias=" << dwarfsrc_canon << endl;
+ - + - +
- ]
4415 : : }
4416 : :
4417 [ + - ]: 3182 : int64_t fileid1 = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, dwarfsrc_canon);
4418 [ + - ]: 3182 : int64_t fileid2 = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, srps);
4419 : :
4420 : 3182 : ps_upsert_s
4421 [ + - ]: 3182 : .reset()
4422 [ + - ]: 3182 : .bind(1, buildid)
4423 [ + - ]: 3182 : .bind(2, fileid1)
4424 [ + - ]: 3182 : .bind(3, fileid2)
4425 [ + - ]: 3182 : .bind(4, sfs.st_mtime)
4426 [ + - ]: 3182 : .step_ok_done();
4427 : :
4428 [ + - + - : 6364 : inc_metric("found_sourcerefs_total","source","files");
+ - + - -
+ - + + -
- - - - -
- ]
4429 [ + - ]: 6364 : }
4430 : : }
4431 : :
4432 : 400 : ps_scan_done
4433 [ + - ]: 400 : .reset()
4434 [ + - ]: 400 : .bind(1, fileid)
4435 [ + - ]: 400 : .bind(2, st.st_mtime)
4436 [ + - ]: 400 : .bind(3, st.st_size)
4437 [ + - ]: 400 : .step_ok_done();
4438 : :
4439 [ + - ]: 400 : if (verbose > 2)
4440 [ + - + - ]: 1200 : obatched(clog) << "recorded buildid=" << buildid << " file=" << rps
4441 [ + - + - : 400 : << " mtime=" << st.st_mtime << " atype="
+ - + - ]
4442 : : << (executable_p ? "E" : "")
4443 [ + - + + : 1124 : << (debuginfo_p ? "D" : "") << endl;
+ - + + +
- + - ]
4444 [ + + ]: 462 : }
4445 : :
4446 : :
4447 : :
4448 : :
4449 : :
4450 : : // Analyze given archive file of given age; record buildids / exec/debuginfo-ness of its
4451 : : // constituent files with given upsert statements.
4452 : : static void
4453 : 394 : archive_classify (const string& rps, string& archive_extension, int64_t archiveid,
4454 : : sqlite_ps& ps_upsert_buildids, sqlite_ps& ps_upsert_fileparts, sqlite_ps& ps_upsert_file,
4455 : : sqlite_ps& ps_lookup_file,
4456 : : sqlite_ps& ps_upsert_de, sqlite_ps& ps_upsert_sref, sqlite_ps& ps_upsert_sdef,
4457 : : sqlite_ps& ps_upsert_seekable,
4458 : : time_t mtime,
4459 : : unsigned& fts_executable, unsigned& fts_debuginfo, unsigned& fts_sref, unsigned& fts_sdef,
4460 : : bool& fts_sref_complete_p)
4461 : : {
4462 : 394 : string archive_decoder = "/dev/null";
4463 [ + + ]: 1020 : for (auto&& arch : scan_archives)
4464 [ + + ]: 626 : if (string_endswith(rps, arch.first))
4465 : : {
4466 [ + - ]: 394 : archive_extension = arch.first;
4467 [ + - ]: 1020 : archive_decoder = arch.second;
4468 : : }
4469 : :
4470 : 394 : FILE* fp;
4471 : 394 : defer_dtor<FILE*,int>::dtor_fn dfn;
4472 [ + + ]: 394 : if (archive_decoder != "cat")
4473 : : {
4474 [ + - + - : 80 : string popen_cmd = archive_decoder + " " + shell_escape(rps);
+ - - + -
- - - ]
4475 [ + - ]: 40 : fp = popen (popen_cmd.c_str(), "r"); // "e" O_CLOEXEC?
4476 : 40 : dfn = pclose;
4477 [ - + ]: 40 : if (fp == NULL)
4478 [ # # # # : 0 : throw libc_exception (errno, string("popen ") + popen_cmd);
# # # # ]
4479 : 40 : }
4480 : : else
4481 : : {
4482 [ + - ]: 354 : fp = fopen (rps.c_str(), "r");
4483 : 354 : dfn = fclose;
4484 [ - + ]: 354 : if (fp == NULL)
4485 [ # # # # : 0 : throw libc_exception (errno, string("fopen ") + rps);
# # # # ]
4486 : : }
4487 : 394 : defer_dtor<FILE*,int> fp_closer (fp, dfn);
4488 : :
4489 : 394 : struct archive *a;
4490 [ + - ]: 394 : a = archive_read_new();
4491 [ - + ]: 394 : if (a == NULL)
4492 [ # # # # ]: 0 : throw archive_exception("cannot create archive reader");
4493 : 394 : defer_dtor<struct archive*,int> archive_closer (a, archive_read_free);
4494 : :
4495 [ + - ]: 394 : int rc = archive_read_support_format_all(a);
4496 [ - + ]: 394 : if (rc != ARCHIVE_OK)
4497 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all formats");
4498 [ + - ]: 394 : rc = archive_read_support_filter_all(a);
4499 [ - + ]: 394 : if (rc != ARCHIVE_OK)
4500 [ # # # # ]: 0 : throw archive_exception(a, "cannot select all filters");
4501 : :
4502 [ + - ]: 394 : rc = archive_read_open_FILE (a, fp);
4503 [ - + ]: 394 : if (rc != ARCHIVE_OK)
4504 : : {
4505 [ # # # # : 0 : obatched(clog) << "cannot open archive from pipe " << rps << endl;
# # ]
4506 [ # # # # ]: 0 : throw archive_exception(a, "cannot open archive from pipe");
4507 : : }
4508 : :
4509 [ + + ]: 394 : if (verbose > 3)
4510 [ + - + - : 708 : obatched(clog) << "libarchive scanning " << rps << " id " << archiveid << endl;
+ - + - +
- ]
4511 : :
4512 [ + - ]: 394 : bool seekable = is_seekable_archive (rps, a);
4513 [ + - + + ]: 394 : if (verbose> 2 && seekable)
4514 [ + - + - : 64 : obatched(clog) << rps << " is seekable" << endl;
+ - ]
4515 : :
4516 : : bool any_exceptions = false;
4517 : 3734 : while(1) // parse archive entries
4518 : : {
4519 [ + - ]: 3734 : if (interrupted)
4520 : : break;
4521 : :
4522 : 3734 : try
4523 : : {
4524 : 3734 : struct archive_entry *e;
4525 [ + - ]: 3734 : rc = archive_read_next_header (a, &e);
4526 [ + + ]: 3734 : if (rc != ARCHIVE_OK)
4527 : : break;
4528 : :
4529 [ + - + + ]: 3340 : if (! S_ISREG(archive_entry_mode (e))) // skip non-files completely
4530 : 2306 : continue;
4531 : :
4532 [ + - ]: 1034 : string fn = canonicalized_archive_entry_pathname (e);
4533 : :
4534 [ + + ]: 1034 : if (verbose > 3)
4535 [ + - + - : 1744 : obatched(clog) << "libarchive checking " << fn << endl;
+ - - - ]
4536 : :
4537 [ + - ]: 1034 : int64_t seekable_size = archive_entry_size (e);
4538 [ + - ]: 1034 : int64_t seekable_offset = archive_filter_bytes (a, 0);
4539 [ + - ]: 1034 : time_t seekable_mtime = archive_entry_mtime (e);
4540 : :
4541 : : // extract this file to a temporary file
4542 : 1034 : char* tmppath = NULL;
4543 : 1034 : rc = asprintf (&tmppath, "%s/debuginfod-classify.XXXXXX", tmpdir.c_str());
4544 [ - + ]: 1034 : if (rc < 0)
4545 [ # # # # ]: 0 : throw libc_exception (ENOMEM, "cannot allocate tmppath");
4546 : 1034 : defer_dtor<void*,void> tmmpath_freer (tmppath, free);
4547 [ + - ]: 1034 : int fd = mkstemp (tmppath);
4548 [ - + ]: 1034 : if (fd < 0)
4549 [ # # # # ]: 0 : throw libc_exception (errno, "cannot create temporary file");
4550 : 1034 : unlink (tmppath); // unlink now so OS will release the file as soon as we close the fd
4551 : 1034 : defer_dtor<int,int> minifd_closer (fd, close);
4552 : :
4553 [ + - ]: 1034 : rc = archive_read_data_into_fd (a, fd);
4554 [ - + ]: 1034 : if (rc != ARCHIVE_OK) {
4555 [ # # ]: 0 : close (fd);
4556 [ # # # # ]: 0 : throw archive_exception(a, "cannot extract file");
4557 : : }
4558 : :
4559 : : // finally ... time to run elf_classify on this bad boy and update the database
4560 : 1034 : bool executable_p = false, debuginfo_p = false;
4561 [ + - ]: 1034 : string buildid;
4562 [ + - ]: 1034 : set<string> sourcefiles;
4563 [ + - ]: 1034 : elf_classify (fd, executable_p, debuginfo_p, buildid, sourcefiles);
4564 : : // NB: might throw
4565 : :
4566 [ + + ]: 1034 : if (buildid != "") // intern buildid
4567 : : {
4568 : 664 : ps_upsert_buildids
4569 [ + - ]: 664 : .reset()
4570 [ + - ]: 664 : .bind(1, buildid)
4571 [ + - ]: 664 : .step_ok_done();
4572 : : }
4573 : :
4574 [ + - ]: 1034 : int64_t fileid = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, fn);
4575 : :
4576 [ + + ]: 1034 : if (sourcefiles.size() > 0) // sref records needed
4577 : : {
4578 : : // NB: we intern each source file once. Once raw, as it
4579 : : // appears in the DWARF file list coming back from
4580 : : // elf_classify() - because it'll end up in the
4581 : : // _norm.artifactsrc column. We don't also put another
4582 : : // version with a '.' at the front, even though that's
4583 : : // how rpm/cpio packs names, because we hide that from
4584 : : // the database for storage efficiency.
4585 : :
4586 [ + + ]: 796 : for (auto&& s : sourcefiles)
4587 : : {
4588 [ + + ]: 494 : if (s == "")
4589 : : {
4590 : 20 : fts_sref_complete_p = false;
4591 : 20 : continue;
4592 : : }
4593 : :
4594 : : // PR25548: store canonicalized source path
4595 : 474 : const string& dwarfsrc = s;
4596 [ + - ]: 474 : string dwarfsrc_canon = canon_pathname (dwarfsrc);
4597 [ + + ]: 474 : if (dwarfsrc_canon != dwarfsrc)
4598 : : {
4599 [ + - ]: 28 : if (verbose > 3)
4600 [ + - + - : 56 : obatched(clog) << "canonicalized src=" << dwarfsrc << " alias=" << dwarfsrc_canon << endl;
+ - + - +
- - - ]
4601 : : }
4602 : :
4603 [ + - ]: 474 : int64_t srcfileid = register_file_name(ps_upsert_fileparts, ps_upsert_file, ps_lookup_file,
4604 : : dwarfsrc_canon);
4605 : :
4606 : 474 : ps_upsert_sref
4607 [ + - ]: 474 : .reset()
4608 [ + - ]: 474 : .bind(1, buildid)
4609 [ + - ]: 474 : .bind(2, srcfileid)
4610 [ + - ]: 474 : .step_ok_done();
4611 : :
4612 [ + - ]: 474 : fts_sref ++;
4613 : 474 : }
4614 : : }
4615 : :
4616 [ + + ]: 1034 : if (executable_p)
4617 : 300 : fts_executable ++;
4618 [ + + ]: 1034 : if (debuginfo_p)
4619 : 364 : fts_debuginfo ++;
4620 : :
4621 [ + + + + ]: 1034 : if (executable_p || debuginfo_p)
4622 : : {
4623 : 664 : ps_upsert_de
4624 [ + - ]: 664 : .reset()
4625 [ + - ]: 664 : .bind(1, buildid)
4626 [ + + + - ]: 964 : .bind(2, debuginfo_p ? 1 : 0)
4627 [ + + + - ]: 1028 : .bind(3, executable_p ? 1 : 0)
4628 [ + - ]: 664 : .bind(4, archiveid)
4629 [ + - ]: 664 : .bind(5, mtime)
4630 [ + - ]: 664 : .bind(6, fileid)
4631 [ + - ]: 664 : .step_ok_done();
4632 [ + + ]: 664 : if (seekable)
4633 : 336 : ps_upsert_seekable
4634 [ + - ]: 336 : .reset()
4635 [ + - ]: 336 : .bind(1, archiveid)
4636 [ + - ]: 336 : .bind(2, fileid)
4637 [ + - ]: 336 : .bind(3, seekable_size)
4638 [ + - ]: 336 : .bind(4, seekable_offset)
4639 [ + - ]: 336 : .bind(5, seekable_mtime)
4640 [ + - ]: 336 : .step_ok_done();
4641 : : }
4642 : : else // potential source - sdef record
4643 : : {
4644 : 370 : fts_sdef ++;
4645 : 370 : ps_upsert_sdef
4646 [ + - ]: 370 : .reset()
4647 [ + - ]: 370 : .bind(1, archiveid)
4648 [ + - ]: 370 : .bind(2, mtime)
4649 [ + - ]: 370 : .bind(3, fileid)
4650 [ + - ]: 370 : .step_ok_done();
4651 : : }
4652 : :
4653 [ + - + + : 1034 : if ((verbose > 2) && (executable_p || debuginfo_p))
+ + ]
4654 : : {
4655 [ + - ]: 664 : obatched ob(clog);
4656 [ + - + - ]: 664 : auto& o = ob << "recorded buildid=" << buildid << " rpm=" << rps << " file=" << fn
4657 [ + - + - : 664 : << " mtime=" << mtime << " atype="
+ - + - +
- + - ]
4658 : : << (executable_p ? "E" : "")
4659 : : << (debuginfo_p ? "D" : "")
4660 [ + - + + : 1328 : << " sourcefiles=" << sourcefiles.size();
+ - + + +
- + - +
- ]
4661 [ + + ]: 664 : if (seekable)
4662 [ + - + - ]: 336 : o << " seekable size=" << seekable_size
4663 [ + - + - ]: 336 : << " offset=" << seekable_offset
4664 [ + - + - ]: 336 : << " mtime=" << seekable_mtime;
4665 [ + - ]: 664 : o << endl;
4666 : 664 : }
4667 : :
4668 [ + + + + ]: 2570 : }
4669 [ - - ]: 0 : catch (const reportable_exception& e)
4670 : : {
4671 [ - - ]: 0 : e.report(clog);
4672 : 0 : any_exceptions = true;
4673 : : // NB: but we allow the libarchive iteration to continue, in
4674 : : // case we can still gather some useful information. That
4675 : : // would allow some webapi queries to work, until later when
4676 : : // this archive is rescanned. (Its vitals won't go into the
4677 : : // _file_mtime_scanned table until after a successful scan.)
4678 : 0 : }
4679 : : }
4680 : :
4681 [ - + ]: 394 : if (any_exceptions)
4682 [ # # # # ]: 0 : throw reportable_exception("exceptions encountered during archive scan");
4683 [ + + ]: 418 : }
4684 : :
4685 : :
4686 : :
4687 : : // scan for archive files such as .rpm
4688 : : static void
4689 : 758 : scan_archive_file (const string& rps, const stat_t& st,
4690 : : sqlite_ps& ps_upsert_buildids,
4691 : : sqlite_ps& ps_upsert_fileparts,
4692 : : sqlite_ps& ps_upsert_file,
4693 : : sqlite_ps& ps_lookup_file,
4694 : : sqlite_ps& ps_upsert_de,
4695 : : sqlite_ps& ps_upsert_sref,
4696 : : sqlite_ps& ps_upsert_sdef,
4697 : : sqlite_ps& ps_upsert_seekable,
4698 : : sqlite_ps& ps_query,
4699 : : sqlite_ps& ps_scan_done,
4700 : : unsigned& fts_cached,
4701 : : unsigned& fts_executable,
4702 : : unsigned& fts_debuginfo,
4703 : : unsigned& fts_sref,
4704 : : unsigned& fts_sdef)
4705 : : {
4706 : : // intern the archive file name
4707 : 758 : int64_t archiveid = register_file_name (ps_upsert_fileparts, ps_upsert_file, ps_lookup_file, rps);
4708 : :
4709 : : /* See if we know of it already. */
4710 : 758 : int rc = ps_query
4711 : 758 : .reset()
4712 : 758 : .bind(1, archiveid)
4713 : 758 : .bind(2, st.st_mtime)
4714 : 758 : .step();
4715 : 758 : ps_query.reset();
4716 [ + + ]: 758 : if (rc == SQLITE_ROW) // i.e., a result, as opposed to DONE (no results)
4717 : : // no need to recheck a file/version we already know
4718 : : // specifically, no need to parse this archive again, since we already have
4719 : : // it as a D or E or S record,
4720 : : // (so is stored with buildid=NULL)
4721 : : {
4722 : 364 : fts_cached ++;
4723 : 364 : return;
4724 : : }
4725 : :
4726 : : // extract the archive contents
4727 : 394 : unsigned my_fts_executable = 0, my_fts_debuginfo = 0, my_fts_sref = 0, my_fts_sdef = 0;
4728 : 394 : bool my_fts_sref_complete_p = true;
4729 : 394 : bool any_exceptions = false;
4730 : 394 : try
4731 : : {
4732 [ + - ]: 394 : string archive_extension;
4733 : 394 : archive_classify (rps, archive_extension, archiveid,
4734 : : ps_upsert_buildids, ps_upsert_fileparts, ps_upsert_file, ps_lookup_file,
4735 : : ps_upsert_de, ps_upsert_sref, ps_upsert_sdef, ps_upsert_seekable, // dalt
4736 [ + - ]: 394 : st.st_mtime,
4737 : : my_fts_executable, my_fts_debuginfo, my_fts_sref, my_fts_sdef,
4738 : : my_fts_sref_complete_p);
4739 [ + - + - : 788 : add_metric ("scanned_bytes_total","source",archive_extension + " archive",
+ - - + +
+ - - -
- ]
4740 [ + - ]: 394 : st.st_size);
4741 [ + - + - : 788 : inc_metric ("scanned_files_total","source",archive_extension + " archive");
+ - + - -
+ + + - -
- - ]
4742 [ + - + - : 788 : add_metric("found_debuginfo_total","source",archive_extension + " archive",
+ - + - -
+ + + - -
- - ]
4743 : : my_fts_debuginfo);
4744 [ + - + - : 788 : add_metric("found_executable_total","source",archive_extension + " archive",
+ - + - -
+ + + - -
- - ]
4745 : : my_fts_executable);
4746 [ + - + - : 804 : add_metric("found_sourcerefs_total","source",archive_extension + " archive",
+ - + - -
+ + + - +
- - - - -
- ]
4747 : : my_fts_sref);
4748 : 394 : }
4749 [ - - ]: 0 : catch (const reportable_exception& e)
4750 : : {
4751 [ - - ]: 0 : e.report(clog);
4752 : 0 : any_exceptions = true;
4753 : 0 : }
4754 : :
4755 [ + - ]: 394 : if (verbose > 2)
4756 [ + - ]: 1182 : obatched(clog) << "scanned archive=" << rps
4757 [ + - + - ]: 394 : << " mtime=" << st.st_mtime
4758 [ + - ]: 394 : << " executables=" << my_fts_executable
4759 [ + - + - ]: 394 : << " debuginfos=" << my_fts_debuginfo
4760 [ + - + - ]: 394 : << " srefs=" << my_fts_sref
4761 [ + - + - ]: 394 : << " sdefs=" << my_fts_sdef
4762 [ + - + - : 394 : << " exceptions=" << any_exceptions
+ - + - ]
4763 : 394 : << endl;
4764 : :
4765 : 394 : fts_executable += my_fts_executable;
4766 : 394 : fts_debuginfo += my_fts_debuginfo;
4767 : 394 : fts_sref += my_fts_sref;
4768 : 394 : fts_sdef += my_fts_sdef;
4769 : :
4770 [ - + ]: 394 : if (any_exceptions)
4771 [ # # # # ]: 0 : throw reportable_exception("exceptions encountered during archive scan");
4772 : :
4773 [ + + ]: 394 : if (my_fts_sref_complete_p) // leave incomplete?
4774 : 392 : ps_scan_done
4775 : 392 : .reset()
4776 : 392 : .bind(1, archiveid)
4777 : 392 : .bind(2, st.st_mtime)
4778 : 392 : .bind(3, st.st_size)
4779 : 392 : .step_ok_done();
4780 : : }
4781 : :
4782 : :
4783 : :
4784 : : ////////////////////////////////////////////////////////////////////////
4785 : :
4786 : :
4787 : :
4788 : : // The thread that consumes file names off of the scanq. We hold
4789 : : // the persistent sqlite_ps's at this level and delegate file/archive
4790 : : // scanning to other functions.
4791 : : static void
4792 : 272 : scan ()
4793 : : {
4794 : : // all the prepared statements fit to use, the _f_ set:
4795 [ + - + - : 544 : sqlite_ps ps_f_upsert_buildids (db, "file-buildids-intern", "insert or ignore into " BUILDIDS "_buildids VALUES (NULL, ?);");
+ - - - ]
4796 [ + - + - : 544 : sqlite_ps ps_f_upsert_fileparts (db, "file-fileparts-intern", "insert or ignore into " BUILDIDS "_fileparts VALUES (NULL, ?);");
+ - + - -
- ]
4797 [ + - ]: 272 : sqlite_ps ps_f_upsert_file (db, "file-file-intern", "insert or ignore into " BUILDIDS "_files VALUES (NULL, \n"
4798 : : "(select id from " BUILDIDS "_fileparts where name = ?),\n"
4799 [ + - + - : 544 : "(select id from " BUILDIDS "_fileparts where name = ?));");
+ - + - -
- ]
4800 [ + - ]: 272 : sqlite_ps ps_f_lookup_file (db, "file-file-lookup",
4801 : : "select f.id\n"
4802 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts p1, " BUILDIDS "_fileparts p2 \n"
4803 [ + - + - : 544 : " where f.dirname = p1.id and f.basename = p2.id and p1.name = ? and p2.name = ?;\n");
+ - + - -
- ]
4804 [ + - ]: 272 : sqlite_ps ps_f_upsert_de (db, "file-de-upsert",
4805 : : "insert or ignore into " BUILDIDS "_f_de "
4806 : : "(buildid, debuginfo_p, executable_p, file, mtime) "
4807 : : "values ((select id from " BUILDIDS "_buildids where hex = ?),"
4808 [ + - + - : 544 : " ?,?,?,?);");
+ - + - -
- ]
4809 [ + - ]: 272 : sqlite_ps ps_f_upsert_s (db, "file-s-upsert",
4810 : : "insert or ignore into " BUILDIDS "_f_s "
4811 : : "(buildid, artifactsrc, file, mtime) "
4812 : : "values ((select id from " BUILDIDS "_buildids where hex = ?),"
4813 [ + - + - : 544 : " ?,?,?);");
+ - + - -
- ]
4814 [ + - ]: 272 : sqlite_ps ps_f_query (db, "file-negativehit-find",
4815 : : "select 1 from " BUILDIDS "_file_mtime_scanned where sourcetype = 'F' "
4816 [ + - + - : 544 : "and file = ? and mtime = ?;");
+ - + - -
- ]
4817 [ + - ]: 272 : sqlite_ps ps_f_scan_done (db, "file-scanned",
4818 : : "insert or ignore into " BUILDIDS "_file_mtime_scanned (sourcetype, file, mtime, size)"
4819 [ + - + - : 544 : "values ('F', ?,?,?);");
+ - + - -
- ]
4820 : :
4821 : : // and now for the _r_ set
4822 [ + - + - : 544 : sqlite_ps ps_r_upsert_buildids (db, "rpm-buildid-intern", "insert or ignore into " BUILDIDS "_buildids VALUES (NULL, ?);");
+ - + - -
- ]
4823 [ + - + - : 544 : sqlite_ps ps_r_upsert_fileparts (db, "rpm-fileparts-intern", "insert or ignore into " BUILDIDS "_fileparts VALUES (NULL, ?);");
+ - + - -
- ]
4824 [ + - ]: 272 : sqlite_ps ps_r_upsert_file (db, "rpm-file-intern", "insert or ignore into " BUILDIDS "_files VALUES (NULL, \n"
4825 : : "(select id from " BUILDIDS "_fileparts where name = ?),\n"
4826 [ + - + - : 544 : "(select id from " BUILDIDS "_fileparts where name = ?));");
+ - + - -
- ]
4827 [ + - ]: 272 : sqlite_ps ps_r_lookup_file (db, "rpm-file-lookup",
4828 : : "select f.id\n"
4829 : : " from " BUILDIDS "_files f, " BUILDIDS "_fileparts p1, " BUILDIDS "_fileparts p2 \n"
4830 [ + - + - : 544 : " where f.dirname = p1.id and f.basename = p2.id and p1.name = ? and p2.name = ?;\n");
+ - + - -
- ]
4831 [ + - ]: 272 : sqlite_ps ps_r_upsert_de (db, "rpm-de-insert",
4832 : : "insert or ignore into " BUILDIDS "_r_de (buildid, debuginfo_p, executable_p, file, mtime, content) values ("
4833 [ + - + - : 544 : "(select id from " BUILDIDS "_buildids where hex = ?), ?, ?, ?, ?, ?);");
+ - + - -
- ]
4834 [ + - ]: 272 : sqlite_ps ps_r_upsert_sref (db, "rpm-sref-insert",
4835 : : "insert or ignore into " BUILDIDS "_r_sref (buildid, artifactsrc) values ("
4836 : : "(select id from " BUILDIDS "_buildids where hex = ?), "
4837 [ + - + - : 544 : "?);");
+ - + - -
- ]
4838 [ + - ]: 272 : sqlite_ps ps_r_upsert_sdef (db, "rpm-sdef-insert",
4839 : : "insert or ignore into " BUILDIDS "_r_sdef (file, mtime, content) values ("
4840 [ + - + - : 544 : "?, ?, ?);");
+ - + - -
- ]
4841 [ + - ]: 272 : sqlite_ps ps_r_upsert_seekable (db, "rpm-seekable-insert",
4842 : : "insert or ignore into " BUILDIDS "_r_seekable (file, content, type, size, offset, mtime) "
4843 [ + - + - : 544 : "values (?, ?, 'xz', ?, ?, ?);");
+ - + - -
- ]
4844 [ + - ]: 272 : sqlite_ps ps_r_query (db, "rpm-negativehit-query",
4845 : : "select 1 from " BUILDIDS "_file_mtime_scanned where "
4846 [ + - + - : 544 : "sourcetype = 'R' and file = ? and mtime = ?;");
+ - + - -
- ]
4847 [ + - ]: 272 : sqlite_ps ps_r_scan_done (db, "rpm-scanned",
4848 : : "insert or ignore into " BUILDIDS "_file_mtime_scanned (sourcetype, file, mtime, size)"
4849 [ + - + - : 544 : "values ('R', ?, ?, ?);");
+ - + - -
- ]
4850 : :
4851 : :
4852 : 272 : unsigned fts_cached = 0, fts_executable = 0, fts_debuginfo = 0, fts_sourcefiles = 0;
4853 : 272 : unsigned fts_sref = 0, fts_sdef = 0;
4854 : :
4855 [ + - + - : 544 : add_metric("thread_count", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
4856 [ + - + - : 544 : add_metric("thread_busy", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
4857 [ + + ]: 1538 : while (! interrupted)
4858 : : {
4859 [ + - ]: 1266 : scan_payload p;
4860 : :
4861 [ + - + - : 2532 : add_metric("thread_busy", "role", "scan", -1);
+ - + - -
+ - + - -
- - ]
4862 : : // NB: threads may be blocked within either of these two waiting
4863 : : // states, if the work queue happens to run dry. That's OK.
4864 [ + - + - ]: 1266 : if (scan_barrier) scan_barrier->count();
4865 [ + - ]: 1266 : bool gotone = scanq.wait_front(p);
4866 [ + - + - : 2532 : add_metric("thread_busy", "role", "scan", 1);
+ - + - -
+ - + - -
- - ]
4867 : :
4868 [ + + - + ]: 1266 : if (! gotone) continue; // go back to waiting
4869 : :
4870 : 994 : try
4871 : : {
4872 : 994 : bool scan_archive = false;
4873 [ + + ]: 2437 : for (auto&& arch : scan_archives)
4874 [ + + ]: 1443 : if (string_endswith(p.first, arch.first))
4875 : 758 : scan_archive = true;
4876 : :
4877 [ + + ]: 994 : if (scan_archive)
4878 [ + - ]: 758 : scan_archive_file (p.first, p.second,
4879 : : ps_r_upsert_buildids,
4880 : : ps_r_upsert_fileparts,
4881 : : ps_r_upsert_file,
4882 : : ps_r_lookup_file,
4883 : : ps_r_upsert_de,
4884 : : ps_r_upsert_sref,
4885 : : ps_r_upsert_sdef,
4886 : : ps_r_upsert_seekable,
4887 : : ps_r_query,
4888 : : ps_r_scan_done,
4889 : : fts_cached,
4890 : : fts_executable,
4891 : : fts_debuginfo,
4892 : : fts_sref,
4893 : : fts_sdef);
4894 : :
4895 [ + + ]: 994 : if (scan_files) // NB: maybe "else if" ?
4896 [ + - ]: 832 : scan_source_file (p.first, p.second,
4897 : : ps_f_upsert_buildids,
4898 : : ps_f_upsert_fileparts,
4899 : : ps_f_upsert_file,
4900 : : ps_f_lookup_file,
4901 : : ps_f_upsert_de,
4902 : : ps_f_upsert_s,
4903 : : ps_f_query,
4904 : : ps_f_scan_done,
4905 : : fts_cached, fts_executable, fts_debuginfo, fts_sourcefiles);
4906 : : }
4907 [ - - ]: 0 : catch (const reportable_exception& e)
4908 : : {
4909 [ - - ]: 0 : e.report(cerr);
4910 : 0 : }
4911 : :
4912 [ + - ]: 994 : scanq.done_front(); // let idlers run
4913 : :
4914 : 994 : if (fts_cached || fts_executable || fts_debuginfo || fts_sourcefiles || fts_sref || fts_sdef)
4915 : : {} // NB: not just if a successful scan - we might have encountered -ENOSPC & failed
4916 [ + - + - ]: 994 : (void) statfs_free_enough_p(db_path, "database"); // report sqlite filesystem size
4917 [ + - + - ]: 994 : (void) statfs_free_enough_p(tmpdir, "tmpdir"); // this too, in case of fdcache/tmpfile usage
4918 : :
4919 : : // finished a scanning step -- not a "loop", because we just
4920 : : // consume the traversal loop's work, whenever
4921 [ + - + - : 1988 : inc_metric("thread_work_total","role","scan");
+ - + - -
+ - + + -
- - - - -
- ]
4922 : 1266 : }
4923 : :
4924 [ + - + - : 544 : add_metric("thread_busy", "role", "scan", -1);
+ - + - -
+ - + - -
- - ]
4925 : 272 : }
4926 : :
4927 : :
4928 : : // Use this function as the thread entry point, so it can catch our
4929 : : // fleet of exceptions (incl. the sqlite_ps ctors) and report.
4930 : : static void*
4931 : 272 : thread_main_scanner (void* arg)
4932 : : {
4933 : 272 : (void) arg;
4934 [ + + ]: 816 : while (! interrupted)
4935 : 272 : try
4936 : : {
4937 [ + - ]: 272 : scan();
4938 : : }
4939 [ - - ]: 0 : catch (const reportable_exception& e)
4940 : : {
4941 [ - - ]: 0 : e.report(cerr);
4942 : 0 : }
4943 : 272 : return 0;
4944 : : }
4945 : :
4946 : :
4947 : :
4948 : : // The thread that traverses all the source_paths and enqueues all the
4949 : : // matching files into the file/archive scan queue.
4950 : : static void
4951 : 120 : scan_source_paths()
4952 : : {
4953 : : // NB: fedora 31 glibc/fts(3) crashes inside fts_read() on empty
4954 : : // path list.
4955 [ + + ]: 120 : if (source_paths.empty())
4956 : 2 : return;
4957 : :
4958 : : // Turn the source_paths into an fts(3)-compatible char**. Since
4959 : : // source_paths[] does not change after argv processing, the
4960 : : // c_str()'s are safe to keep around awile.
4961 : 118 : vector<const char *> sps;
4962 [ + + ]: 320 : for (auto&& sp: source_paths)
4963 [ + - ]: 202 : sps.push_back(sp.c_str());
4964 [ + - - - ]: 118 : sps.push_back(NULL);
4965 : :
4966 [ + + + - ]: 222 : FTS *fts = fts_open ((char * const *)sps.data(),
4967 : : (traverse_logical ? FTS_LOGICAL : FTS_PHYSICAL|FTS_XDEV)
4968 : : | FTS_NOCHDIR /* multithreaded */,
4969 : : NULL);
4970 [ - + ]: 118 : if (fts == NULL)
4971 [ # # # # ]: 0 : throw libc_exception(errno, "cannot fts_open");
4972 : 118 : defer_dtor<FTS*,int> fts_cleanup (fts, fts_close);
4973 : :
4974 : 118 : struct timespec ts_start, ts_end;
4975 : 118 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
4976 : 118 : unsigned fts_scanned = 0, fts_regex = 0;
4977 : :
4978 : 118 : FTSENT *f;
4979 [ + - + + ]: 2230 : while ((f = fts_read (fts)) != NULL)
4980 : : {
4981 [ + - ]: 1994 : if (interrupted) break;
4982 : :
4983 [ - + ]: 1994 : if (sigusr2 != forced_groom_count) // stop early if groom triggered
4984 : : {
4985 [ # # ]: 0 : scanq.clear(); // clear previously issued work for scanner threads
4986 : : break;
4987 : : }
4988 : :
4989 : 1994 : fts_scanned ++;
4990 : :
4991 [ + - ]: 1994 : if (verbose > 2)
4992 [ + - + - : 3988 : obatched(clog) << "fts traversing " << f->fts_path << endl;
+ - ]
4993 : :
4994 [ + + + + : 1994 : switch (f->fts_info)
+ ]
4995 : : {
4996 : 1098 : case FTS_F:
4997 : 1098 : {
4998 : : /* Found a file. Convert it to an absolute path, so
4999 : : the buildid database does not have relative path
5000 : : names that are unresolvable from a subsequent run
5001 : : in a different cwd. */
5002 [ + - ]: 1098 : char *rp = realpath(f->fts_path, NULL);
5003 [ - + ]: 1098 : if (rp == NULL)
5004 : 0 : continue; // ignore dangling symlink or such
5005 [ + - ]: 1098 : string rps = string(rp);
5006 : 1098 : free (rp);
5007 : :
5008 [ + - ]: 1098 : bool ri = !regexec (&file_include_regex, rps.c_str(), 0, 0, 0);
5009 [ + - ]: 1098 : bool rx = !regexec (&file_exclude_regex, rps.c_str(), 0, 0, 0);
5010 [ + + ]: 1098 : if (!ri || rx)
5011 : : {
5012 [ + - ]: 104 : if (verbose > 3)
5013 [ + - ]: 208 : obatched(clog) << "fts skipped by regex "
5014 [ + + + - : 128 : << (!ri ? "I" : "") << (rx ? "X" : "") << endl;
+ + + - +
- ]
5015 : 104 : fts_regex ++;
5016 [ + + ]: 104 : if (!ri)
5017 [ + - + - : 24 : inc_metric("traversed_total","type","file-skipped-I");
+ - + - -
+ - + - -
- - ]
5018 [ + + ]: 104 : if (rx)
5019 [ + - + - : 184 : inc_metric("traversed_total","type","file-skipped-X");
+ - + - -
+ - + - -
- - ]
5020 : : }
5021 : : else
5022 : : {
5023 [ + - + - ]: 994 : scanq.push_back (make_pair(rps, *f->fts_statp));
5024 [ + - + - : 1988 : inc_metric("traversed_total","type","file");
+ - + - -
+ - + - -
- - - - ]
5025 : : }
5026 : 0 : }
5027 : 1098 : break;
5028 : :
5029 : 4 : case FTS_ERR:
5030 : 4 : case FTS_NS:
5031 : : // report on some types of errors because they may reflect fixable misconfiguration
5032 : 4 : {
5033 [ + - + - : 8 : auto x = libc_exception(f->fts_errno, string("fts traversal ") + string(f->fts_path));
+ - + - -
+ - + -
- ]
5034 [ + - ]: 4 : x.report(cerr);
5035 : 0 : }
5036 [ + - + - : 8 : inc_metric("traversed_total","type","error");
+ - + - -
+ - + - -
- - ]
5037 : 4 : break;
5038 : :
5039 : 12 : case FTS_SL: // ignore, but count because debuginfod -L would traverse these
5040 [ + - + - : 24 : inc_metric("traversed_total","type","symlink");
+ - + - -
+ - + - -
- - ]
5041 : 12 : break;
5042 : :
5043 : 440 : case FTS_D: // ignore
5044 [ + - + - : 880 : inc_metric("traversed_total","type","directory");
+ - + - -
+ - + - -
- - ]
5045 : 440 : break;
5046 : :
5047 : 440 : default: // ignore
5048 [ + - + - : 880 : inc_metric("traversed_total","type","other");
+ - + - -
+ - + - -
- - ]
5049 : 440 : break;
5050 : : }
5051 : : }
5052 : 118 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
5053 : 118 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
5054 : :
5055 [ + - + - : 354 : obatched(clog) << "fts traversed source paths in " << deltas << "s, scanned=" << fts_scanned
+ - + - ]
5056 [ + - + - : 118 : << ", regex-skipped=" << fts_regex << endl;
+ - ]
5057 [ + - ]: 236 : }
5058 : :
5059 : :
5060 : : static void*
5061 : 68 : thread_main_fts_source_paths (void* arg)
5062 : : {
5063 : 68 : (void) arg; // ignore; we operate on global data
5064 : :
5065 [ + - + - : 136 : set_metric("thread_tid", "role","traverse", tid());
+ - - + -
+ - - -
- ]
5066 [ + - + - : 136 : add_metric("thread_count", "role", "traverse", 1);
+ - - + -
+ - - -
- ]
5067 : :
5068 : 68 : time_t last_rescan = 0;
5069 : :
5070 [ + - ]: 288 : while (! interrupted)
5071 : : {
5072 : 288 : sleep (1);
5073 : 288 : scanq.wait_idle(); // don't start a new traversal while scanners haven't finished the job
5074 : 288 : scanq.done_idle(); // release the hounds
5075 [ + + ]: 288 : if (interrupted) break;
5076 : :
5077 : 220 : time_t now = time(NULL);
5078 : 220 : bool rescan_now = false;
5079 [ + + ]: 220 : if (last_rescan == 0) // at least one initial rescan is documented even for -t0
5080 : 66 : rescan_now = true;
5081 [ + + + + ]: 220 : if (rescan_s > 0 && (long)now > (long)(last_rescan + rescan_s))
5082 : 220 : rescan_now = true;
5083 [ + + ]: 220 : if (sigusr1 != forced_rescan_count)
5084 : : {
5085 : 56 : forced_rescan_count = sigusr1;
5086 : 56 : rescan_now = true;
5087 : : }
5088 [ + + ]: 220 : if (rescan_now)
5089 : : {
5090 [ + - + - : 240 : set_metric("thread_busy", "role","traverse", 1);
+ - - + -
+ - - -
- ]
5091 : 120 : try
5092 : : {
5093 [ + - ]: 120 : scan_source_paths();
5094 : : }
5095 [ - - ]: 0 : catch (const reportable_exception& e)
5096 : : {
5097 [ - - ]: 0 : e.report(cerr);
5098 : 0 : }
5099 : 120 : last_rescan = time(NULL); // NB: now was before scanning
5100 : : // finished a traversal loop
5101 [ + - + - : 240 : inc_metric("thread_work_total", "role","traverse");
+ - - + -
+ - - -
- ]
5102 [ + - + - : 240 : set_metric("thread_busy", "role","traverse", 0);
+ - - + -
+ - - -
- ]
5103 : : }
5104 : : }
5105 : :
5106 : 68 : return 0;
5107 : : }
5108 : :
5109 : :
5110 : :
5111 : : ////////////////////////////////////////////////////////////////////////
5112 : :
5113 : : static void
5114 : 74 : database_stats_report()
5115 : : {
5116 : 74 : sqlite_ps ps_query (db, "database-overview",
5117 [ + - + - : 148 : "select label,quantity from " BUILDIDS "_stats");
+ - - - ]
5118 : :
5119 [ + - + - ]: 148 : obatched(clog) << "database record counts:" << endl;
5120 : 1702 : while (1)
5121 : : {
5122 [ + - ]: 888 : if (interrupted) break;
5123 [ + - ]: 888 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
5124 : : break;
5125 : :
5126 [ + - ]: 888 : int rc = ps_query.step();
5127 [ + + ]: 888 : if (rc == SQLITE_DONE) break;
5128 [ - + ]: 814 : if (rc != SQLITE_ROW)
5129 [ # # # # ]: 0 : throw sqlite_exception(rc, "step");
5130 : :
5131 [ + - ]: 814 : obatched(clog)
5132 [ + - - + : 814 : << ((const char*) sqlite3_column_text(ps_query, 0) ?: (const char*) "NULL")
+ - ]
5133 : : << " "
5134 [ + - + - : 1628 : << (sqlite3_column_text(ps_query, 1) ?: (const unsigned char*) "NULL")
- + + - ]
5135 : 814 : << endl;
5136 : :
5137 [ + - + - : 1628 : set_metric("groom", "statistic",
- + + - +
- + - + -
- + + + -
- - - ]
5138 [ + - ]: 814 : ((const char*) sqlite3_column_text(ps_query, 0) ?: (const char*) "NULL"),
5139 : : (sqlite3_column_double(ps_query, 1)));
5140 : 814 : }
5141 : 74 : }
5142 : :
5143 : :
5144 : : // Do a round of database grooming that might take many minutes to run.
5145 : 74 : void groom()
5146 : : {
5147 [ + - ]: 148 : obatched(clog) << "grooming database" << endl;
5148 : :
5149 : 74 : struct timespec ts_start, ts_end;
5150 : 74 : clock_gettime (CLOCK_MONOTONIC, &ts_start);
5151 : :
5152 : : // scan for files that have disappeared
5153 : 74 : sqlite_ps files (db, "check old files",
5154 : : "select distinct s.mtime, s.file, f.name from "
5155 : : BUILDIDS "_file_mtime_scanned s, " BUILDIDS "_files_v f "
5156 [ + - + - : 148 : "where f.id = s.file");
+ - - - ]
5157 : : // NB: Because _ftime_mtime_scanned can contain both F and
5158 : : // R records for the same file, this query would return duplicates if the
5159 : : // DISTINCT qualifier were not there.
5160 [ + - ]: 74 : files.reset();
5161 : :
5162 : : // DECISION TIME - we enumerate stale fileids/mtimes
5163 [ + - ]: 74 : deque<pair<int64_t,int64_t> > stale_fileid_mtime;
5164 : :
5165 : 74 : time_t time_start = time(NULL);
5166 : 338 : while(1)
5167 : : {
5168 : : // PR28514: limit grooming iteration to O(rescan time), to avoid
5169 : : // slow filesystem tests over many files locking out rescans for
5170 : : // too long.
5171 [ + + - + ]: 206 : if (rescan_s > 0 && (long)time(NULL) > (long)(time_start + rescan_s))
5172 : : {
5173 [ # # # # : 0 : inc_metric("groomed_total", "decision", "aborted");
# # # # #
# # # # #
# # ]
5174 : 0 : break;
5175 : : }
5176 : :
5177 [ + - ]: 206 : if (interrupted) break;
5178 : :
5179 [ + - ]: 206 : int rc = files.step();
5180 [ + + ]: 206 : if (rc != SQLITE_ROW)
5181 : : break;
5182 : :
5183 [ + - ]: 132 : int64_t mtime = sqlite3_column_int64 (files, 0);
5184 [ + - ]: 132 : int64_t fileid = sqlite3_column_int64 (files, 1);
5185 [ + - - + ]: 132 : const char* filename = ((const char*) sqlite3_column_text (files, 2) ?: "");
5186 : 132 : struct stat s;
5187 : 132 : bool regex_file_drop = 0;
5188 : :
5189 [ + + ]: 132 : if (regex_groom)
5190 : : {
5191 [ + - ]: 16 : bool reg_include = !regexec (&file_include_regex, filename, 0, 0, 0);
5192 [ + - ]: 16 : bool reg_exclude = !regexec (&file_exclude_regex, filename, 0, 0, 0);
5193 : 16 : regex_file_drop = !reg_include || reg_exclude; // match logic of scan_source_paths
5194 : : }
5195 : :
5196 : 132 : rc = stat(filename, &s);
5197 [ + + - + ]: 132 : if ( regex_file_drop || rc < 0 || (mtime != (int64_t) s.st_mtime) )
5198 : : {
5199 [ + - ]: 24 : if (verbose > 2)
5200 [ + - + - : 48 : obatched(clog) << "groom: stale file=" << filename << " mtime=" << mtime << endl;
+ - + - +
- ]
5201 [ + - ]: 24 : stale_fileid_mtime.push_back(make_pair(fileid,mtime));
5202 [ + - + - : 48 : inc_metric("groomed_total", "decision", "stale");
+ - + - -
+ - + - -
- - ]
5203 [ + - + - : 48 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
5204 : : }
5205 : : else
5206 [ + - + - : 216 : inc_metric("groomed_total", "decision", "fresh");
+ - + - -
+ - + - -
- - ]
5207 : :
5208 [ + - ]: 132 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
5209 : : break;
5210 : 132 : }
5211 [ + - ]: 74 : files.reset();
5212 : :
5213 : : // ACTION TIME
5214 : :
5215 : : // Now that we know which file/mtime tuples are stale, actually do
5216 : : // the deletion from the database. Doing this during the SELECT
5217 : : // iteration above results in undefined behaviour in sqlite, as per
5218 : : // https://www.sqlite.org/isolation.html
5219 : :
5220 : : // We could shuffle stale_fileid_mtime[] here. It'd let aborted
5221 : : // sequences of nuke operations resume at random locations, instead
5222 : : // of just starting over. But it doesn't matter much either way,
5223 : : // as long as we make progress.
5224 : :
5225 [ + - + - : 148 : sqlite_ps files_del_f_de (db, "nuke f_de", "delete from " BUILDIDS "_f_de where file = ? and mtime = ?");
+ - + - -
- ]
5226 [ + - + - : 148 : sqlite_ps files_del_r_de (db, "nuke r_de", "delete from " BUILDIDS "_r_de where file = ? and mtime = ?");
+ - + - -
- ]
5227 [ + - ]: 74 : sqlite_ps files_del_scan (db, "nuke f_m_s", "delete from " BUILDIDS "_file_mtime_scanned "
5228 [ + - + - : 148 : "where file = ? and mtime = ?");
+ - + - -
- ]
5229 : :
5230 [ + + ]: 98 : while (! stale_fileid_mtime.empty())
5231 : : {
5232 : 24 : auto stale = stale_fileid_mtime.front();
5233 : 24 : stale_fileid_mtime.pop_front();
5234 [ + - + - : 48 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
5235 : :
5236 : : // PR28514: limit grooming iteration to O(rescan time), to avoid
5237 : : // slow nuke_* queries over many files locking out rescans for too
5238 : : // long. We iterate over the files in random() sequence to avoid
5239 : : // partial checks going over the same set.
5240 [ - + - - ]: 24 : if (rescan_s > 0 && (long)time(NULL) > (long)(time_start + rescan_s))
5241 : : {
5242 [ # # # # : 0 : inc_metric("groomed_total", "action", "aborted");
# # # # #
# # # # #
# # ]
5243 : 0 : break;
5244 : : }
5245 : :
5246 [ + - ]: 24 : if (interrupted) break;
5247 : :
5248 : 24 : int64_t fileid = stale.first;
5249 : 24 : int64_t mtime = stale.second;
5250 [ + - + - : 24 : files_del_f_de.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
5251 [ + - + - : 24 : files_del_r_de.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
5252 [ + - + - : 24 : files_del_scan.reset().bind(1,fileid).bind(2,mtime).step_ok_done();
+ - + - ]
5253 [ + - + - : 48 : inc_metric("groomed_total", "action", "cleaned");
+ - + - -
+ - + - -
- - ]
5254 : :
5255 [ + - ]: 24 : if (sigusr1 != forced_rescan_count) // stop early if scan triggered
5256 : : break;
5257 : : }
5258 : 74 : stale_fileid_mtime.clear(); // no need for this any longer
5259 [ + - + - : 148 : set_metric("thread_work_pending","role","groom", stale_fileid_mtime.size());
+ - + - -
+ - + - -
- - ]
5260 : :
5261 : : // delete buildids with no references in _r_de or _f_de tables;
5262 : : // cascades to _r_sref & _f_s records
5263 [ + - ]: 74 : sqlite_ps buildids_del (db, "nuke orphan buildids",
5264 : : "delete from " BUILDIDS "_buildids "
5265 : : "where not exists (select 1 from " BUILDIDS "_f_de d where " BUILDIDS "_buildids.id = d.buildid) "
5266 [ + - + - : 148 : "and not exists (select 1 from " BUILDIDS "_r_de d where " BUILDIDS "_buildids.id = d.buildid)");
+ - + - -
- ]
5267 [ + - + - ]: 74 : buildids_del.reset().step_ok_done();
5268 : :
5269 [ - + ]: 74 : if (interrupted) return;
5270 : :
5271 : : // NB: "vacuum" is too heavy for even daily runs: it rewrites the entire db, so is done as maxigroom -G
5272 [ + - + - : 222 : { sqlite_ps g (db, "incremental vacuum", "pragma incremental_vacuum"); g.reset().step_ok_done(); }
+ - + - +
- + - -
- ]
5273 : : // https://www.sqlite.org/lang_analyze.html#approx
5274 [ + - + - : 222 : { sqlite_ps g (db, "analyze setup", "pragma analysis_limit = 1000;\n"); g.reset().step_ok_done(); }
+ - + - +
- + - -
- ]
5275 [ + - + - : 148 : { sqlite_ps g (db, "analyze", "analyze"); g.reset().step_ok_done(); }
+ - - + +
- + - -
- ]
5276 [ + - + - : 222 : { sqlite_ps g (db, "analyze reload", "analyze sqlite_schema"); g.reset().step_ok_done(); }
+ - + - +
- + - -
- ]
5277 [ + - + - : 148 : { sqlite_ps g (db, "optimize", "pragma optimize"); g.reset().step_ok_done(); }
+ - - + +
- + - -
- ]
5278 [ + - + - : 222 : { sqlite_ps g (db, "wal checkpoint", "pragma wal_checkpoint=truncate"); g.reset().step_ok_done(); }
+ - + - +
- + - -
- ]
5279 : :
5280 [ + - ]: 74 : database_stats_report();
5281 : :
5282 [ + - + - ]: 74 : (void) statfs_free_enough_p(db_path, "database"); // report sqlite filesystem size
5283 : :
5284 [ + - ]: 74 : sqlite3_db_release_memory(db); // shrink the process if possible
5285 [ + - ]: 74 : sqlite3_db_release_memory(dbq); // ... for both connections
5286 [ + - ]: 74 : debuginfod_pool_groom(); // and release any debuginfod_client objects we've been holding onto
5287 : : #if HAVE_MALLOC_TRIM
5288 : 74 : malloc_trim(0); // PR31103: release memory allocated for temporary purposes
5289 : : #endif
5290 : :
5291 : : #if 0 /* PR31265: don't jettison cache unnecessarily */
5292 : : fdcache.limit(0); // release the fdcache contents
5293 : : fdcache.limit(fdcache_mbs); // restore status quo parameters
5294 : : #endif
5295 : :
5296 : 74 : clock_gettime (CLOCK_MONOTONIC, &ts_end);
5297 : 74 : double deltas = (ts_end.tv_sec - ts_start.tv_sec) + (ts_end.tv_nsec - ts_start.tv_nsec)/1.e9;
5298 : :
5299 [ + - + - : 148 : obatched(clog) << "groomed database in " << deltas << "s" << endl;
+ - + - ]
5300 : 74 : }
5301 : :
5302 : :
5303 : : static void*
5304 : 74 : thread_main_groom (void* /*arg*/)
5305 : : {
5306 [ + - + - : 148 : set_metric("thread_tid", "role", "groom", tid());
+ - - + -
+ - - -
- ]
5307 [ + - + - : 148 : add_metric("thread_count", "role", "groom", 1);
+ - - + -
+ - - -
- ]
5308 : :
5309 : 74 : time_t last_groom = 0;
5310 : :
5311 : 514 : while (1)
5312 : : {
5313 : 294 : sleep (1);
5314 : 294 : scanq.wait_idle(); // PR25394: block scanners during grooming!
5315 [ + + ]: 294 : if (interrupted) break;
5316 : :
5317 : 220 : time_t now = time(NULL);
5318 : 220 : bool groom_now = false;
5319 [ + + ]: 220 : if (last_groom == 0) // at least one initial groom is documented even for -g0
5320 : 68 : groom_now = true;
5321 [ + + + + ]: 220 : if (groom_s > 0 && (long)now > (long)(last_groom + groom_s))
5322 : 220 : groom_now = true;
5323 [ + + ]: 220 : if (sigusr2 != forced_groom_count)
5324 : : {
5325 : 6 : forced_groom_count = sigusr2;
5326 : 6 : groom_now = true;
5327 : : }
5328 [ + + ]: 220 : if (groom_now)
5329 : : {
5330 [ + - + - : 148 : set_metric("thread_busy", "role", "groom", 1);
+ - - + -
+ - - -
- ]
5331 : 74 : try
5332 : : {
5333 [ + - ]: 74 : groom ();
5334 : : }
5335 [ - - ]: 0 : catch (const sqlite_exception& e)
5336 : : {
5337 [ - - - - : 0 : obatched(cerr) << e.message << endl;
- - ]
5338 : 0 : }
5339 : 74 : last_groom = time(NULL); // NB: now was before grooming
5340 : : // finished a grooming loop
5341 [ + - + - : 148 : inc_metric("thread_work_total", "role", "groom");
+ - - + -
+ - - -
- ]
5342 [ + - + - : 148 : set_metric("thread_busy", "role", "groom", 0);
+ - - + -
+ - - -
- ]
5343 : : }
5344 : :
5345 : 220 : scanq.done_idle();
5346 : 220 : }
5347 : :
5348 : 74 : return 0;
5349 : : }
5350 : :
5351 : :
5352 : : ////////////////////////////////////////////////////////////////////////
5353 : :
5354 : :
5355 : : static void
5356 : 76 : signal_handler (int /* sig */)
5357 : : {
5358 : 76 : interrupted ++;
5359 : :
5360 [ + + ]: 76 : if (db)
5361 : 74 : sqlite3_interrupt (db);
5362 [ + - ]: 76 : if (dbq)
5363 : 76 : sqlite3_interrupt (dbq);
5364 : :
5365 : : // NB: don't do anything else in here
5366 : 76 : }
5367 : :
5368 : : static void
5369 : 56 : sigusr1_handler (int /* sig */)
5370 : : {
5371 : 56 : sigusr1 ++;
5372 : : // NB: don't do anything else in here
5373 : 56 : }
5374 : :
5375 : : static void
5376 : 6 : sigusr2_handler (int /* sig */)
5377 : : {
5378 : 6 : sigusr2 ++;
5379 : : // NB: don't do anything else in here
5380 : 6 : }
5381 : :
5382 : :
5383 : : static void // error logging callback from libmicrohttpd internals
5384 : 0 : error_cb (void *arg, const char *fmt, va_list ap)
5385 : : {
5386 : 0 : (void) arg;
5387 [ # # # # : 0 : inc_metric("error_count","libmicrohttpd",fmt);
# # # # #
# # # #
# ]
5388 : 0 : char errmsg[512];
5389 : 0 : (void) vsnprintf (errmsg, sizeof(errmsg), fmt, ap); // ok if slightly truncated
5390 [ # # ]: 0 : obatched(cerr) << "libmicrohttpd error: " << errmsg; // MHD_DLOG calls already include \n
5391 : 0 : }
5392 : :
5393 : :
5394 : : // A user-defined sqlite function, to score the sharedness of the
5395 : : // prefix of two strings. This is used to compare candidate debuginfo
5396 : : // / source-rpm names, so that the closest match
5397 : : // (directory-topology-wise closest) is found. This is important in
5398 : : // case the same sref (source file name) is in many -debuginfo or
5399 : : // -debugsource RPMs, such as when multiple versions/releases of the
5400 : : // same package are in the database.
5401 : :
5402 : 264 : static void sqlite3_sharedprefix_fn (sqlite3_context* c, int argc, sqlite3_value** argv)
5403 : : {
5404 [ - + ]: 264 : if (argc != 2)
5405 : 0 : sqlite3_result_error(c, "expect 2 string arguments", -1);
5406 [ + - + + ]: 528 : else if ((sqlite3_value_type(argv[0]) != SQLITE_TEXT) ||
5407 : 264 : (sqlite3_value_type(argv[1]) != SQLITE_TEXT))
5408 : 6 : sqlite3_result_null(c);
5409 : : else
5410 : : {
5411 : 258 : const unsigned char* a = sqlite3_value_text (argv[0]);
5412 : 258 : const unsigned char* b = sqlite3_value_text (argv[1]);
5413 : 258 : int i = 0;
5414 [ + + + - : 31870 : while (*a != '\0' && *b != '\0' && *a++ == *b++)
+ + + + ]
5415 : 31130 : i++;
5416 : 258 : sqlite3_result_int (c, i);
5417 : : }
5418 : 264 : }
5419 : :
5420 : :
5421 : : static unsigned
5422 : 148 : default_concurrency() // guaranteed >= 1
5423 : : {
5424 : : // Prior to PR29975 & PR29976, we'd just use this:
5425 : 148 : unsigned sth = std::thread::hardware_concurrency();
5426 : : // ... but on many-CPU boxes, admins or distros may throttle
5427 : : // resources in such a way that debuginfod would mysteriously fail.
5428 : : // So we reduce the defaults:
5429 : :
5430 : 148 : unsigned aff = 0;
5431 : : #ifdef HAVE_SCHED_GETAFFINITY
5432 : 148 : {
5433 : 148 : int ret;
5434 : 148 : cpu_set_t mask;
5435 : 148 : CPU_ZERO(&mask);
5436 : 148 : ret = sched_getaffinity(0, sizeof(mask), &mask);
5437 [ + - ]: 148 : if (ret == 0)
5438 : 148 : aff = CPU_COUNT(&mask);
5439 : : }
5440 : : #endif
5441 : :
5442 : 148 : unsigned fn = 0;
5443 : : #ifdef HAVE_GETRLIMIT
5444 : 148 : {
5445 : 148 : struct rlimit rlim;
5446 : 148 : int rc = getrlimit(RLIMIT_NOFILE, &rlim);
5447 [ + - ]: 148 : if (rc == 0)
5448 [ + - ]: 296 : fn = max((rlim_t)1, (rlim.rlim_cur - 100) / 4);
5449 : : // at least 2 fds are used by each listener thread etc.
5450 : : // plus a bunch to account for shared libraries and such
5451 : : }
5452 : : #endif
5453 : :
5454 [ - + - + : 148 : unsigned d = min(max(sth, 1U),
- + ]
5455 [ - + ]: 148 : min(max(aff, 1U),
5456 [ - + ]: 148 : max(fn, 1U)));
5457 : 148 : return d;
5458 : : }
5459 : :
5460 : :
5461 : : // 30879: Something to help out in case of an uncaught exception.
5462 : 0 : void my_terminate_handler()
5463 : : {
5464 : : #if defined(__GLIBC__)
5465 : 0 : void *array[40];
5466 : 0 : int size = backtrace (array, 40);
5467 : 0 : backtrace_symbols_fd (array, size, STDERR_FILENO);
5468 : : #endif
5469 : : #if defined(__GLIBCXX__) || defined(__GLIBCPP__)
5470 : 0 : __gnu_cxx::__verbose_terminate_handler();
5471 : : #endif
5472 : 0 : abort();
5473 : : }
5474 : :
5475 : :
5476 : : int
5477 : 76 : main (int argc, char *argv[])
5478 : : {
5479 : 76 : (void) setlocale (LC_ALL, "");
5480 : 76 : (void) bindtextdomain (PACKAGE_TARNAME, LOCALEDIR);
5481 : 76 : (void) textdomain (PACKAGE_TARNAME);
5482 : :
5483 : 76 : std::set_terminate(& my_terminate_handler);
5484 : :
5485 : : /* Tell the library which version we are expecting. */
5486 : 76 : elf_version (EV_CURRENT);
5487 : :
5488 [ + - - + ]: 152 : tmpdir = string(getenv("TMPDIR") ?: "/tmp");
5489 : :
5490 : : /* Set computed default values. */
5491 [ - + + - : 76 : db_path = string(getenv("HOME") ?: "/") + string("/.debuginfod.sqlite"); /* XDG? */
+ - - + -
+ + - -
- ]
5492 : 76 : int rc = regcomp (& file_include_regex, ".*", REG_EXTENDED|REG_NOSUB); // match everything
5493 [ - + ]: 76 : if (rc != 0)
5494 : 0 : error (EXIT_FAILURE, 0, "regcomp failure: %d", rc);
5495 : 76 : rc = regcomp (& file_exclude_regex, "^$", REG_EXTENDED|REG_NOSUB); // match nothing
5496 [ - + ]: 76 : if (rc != 0)
5497 : 0 : error (EXIT_FAILURE, 0, "regcomp failure: %d", rc);
5498 : :
5499 : : // default parameters for fdcache are computed from system stats
5500 : 76 : struct statfs sfs;
5501 : 76 : rc = statfs(tmpdir.c_str(), &sfs);
5502 [ - + ]: 76 : if (rc < 0)
5503 : 0 : fdcache_mbs = 1024; // 1 gigabyte
5504 : : else
5505 : 76 : fdcache_mbs = sfs.f_bavail * sfs.f_bsize / 1024 / 1024 / 4; // 25% of free space
5506 : 76 : fdcache_mintmp = 25; // emergency flush at 25% remaining (75% full)
5507 : 76 : fdcache_prefetch = 64; // guesstimate storage is this much less costly than re-decompression
5508 : :
5509 : : /* Parse and process arguments. */
5510 : 76 : int remaining;
5511 : 76 : (void) argp_parse (&argp, argc, argv, ARGP_IN_ORDER, &remaining, NULL);
5512 [ - + ]: 76 : if (remaining != argc)
5513 : 0 : error (EXIT_FAILURE, 0,
5514 : 0 : "unexpected argument: %s", argv[remaining]);
5515 : :
5516 [ + + + + : 76 : if (scan_archives.size()==0 && !scan_files && source_paths.size()>0)
- + ]
5517 [ # # ]: 0 : obatched(clog) << "warning: without -F -R -U -Z, ignoring PATHs" << endl;
5518 : :
5519 : 76 : fdcache.limit(fdcache_mbs);
5520 : :
5521 : 76 : (void) signal (SIGPIPE, SIG_IGN); // microhttpd can generate it incidentally, ignore
5522 : 76 : (void) signal (SIGINT, signal_handler); // ^C
5523 : 76 : (void) signal (SIGHUP, signal_handler); // EOF
5524 : 76 : (void) signal (SIGTERM, signal_handler); // systemd
5525 : 76 : (void) signal (SIGUSR1, sigusr1_handler); // end-user
5526 : 76 : (void) signal (SIGUSR2, sigusr2_handler); // end-user
5527 : :
5528 : : /* Get database ready. */
5529 [ + + ]: 76 : if (! passive_p)
5530 : : {
5531 : 74 : rc = sqlite3_open_v2 (db_path.c_str(), &db, (SQLITE_OPEN_READWRITE
5532 : : |SQLITE_OPEN_URI
5533 : : |SQLITE_OPEN_PRIVATECACHE
5534 : : |SQLITE_OPEN_CREATE
5535 : : |SQLITE_OPEN_FULLMUTEX), /* thread-safe */
5536 : : NULL);
5537 [ - + ]: 74 : if (rc == SQLITE_CORRUPT)
5538 : : {
5539 : 0 : (void) unlink (db_path.c_str());
5540 : 0 : error (EXIT_FAILURE, 0,
5541 : : "cannot open %s, deleted database: %s", db_path.c_str(), sqlite3_errmsg(db));
5542 : : }
5543 [ - + ]: 74 : else if (rc)
5544 : : {
5545 : 0 : error (EXIT_FAILURE, 0,
5546 : : "cannot open %s, consider deleting database: %s", db_path.c_str(), sqlite3_errmsg(db));
5547 : : }
5548 : : }
5549 : :
5550 : : // open the readonly query variant
5551 : : // NB: PRIVATECACHE allows web queries to operate in parallel with
5552 : : // much other grooming/scanning operation.
5553 : 76 : rc = sqlite3_open_v2 (db_path.c_str(), &dbq, (SQLITE_OPEN_READONLY
5554 : : |SQLITE_OPEN_URI
5555 : : |SQLITE_OPEN_PRIVATECACHE
5556 : : |SQLITE_OPEN_FULLMUTEX), /* thread-safe */
5557 : : NULL);
5558 [ - + ]: 76 : if (rc)
5559 : : {
5560 : 0 : error (EXIT_FAILURE, 0,
5561 : : "cannot open %s, consider deleting database: %s", db_path.c_str(), sqlite3_errmsg(dbq));
5562 : : }
5563 : :
5564 : :
5565 [ + - ]: 152 : obatched(clog) << "opened database " << db_path
5566 [ + + + - : 78 : << (db?" rw":"") << (dbq?" ro":"") << endl;
- + + - +
- ]
5567 [ + - + - ]: 152 : obatched(clog) << "sqlite version " << sqlite3_version << endl;
5568 [ + + + - : 226 : obatched(clog) << "service mode " << (passive_p ? "passive":"active") << endl;
+ - ]
5569 : :
5570 : : // add special string-prefix-similarity function used in rpm sref/sdef resolution
5571 : 76 : rc = sqlite3_create_function(dbq, "sharedprefix", 2, SQLITE_UTF8, NULL,
5572 : : & sqlite3_sharedprefix_fn, NULL, NULL);
5573 [ - + ]: 76 : if (rc != SQLITE_OK)
5574 : 0 : error (EXIT_FAILURE, 0,
5575 : : "cannot create sharedprefix function: %s", sqlite3_errmsg(dbq));
5576 : :
5577 [ + + ]: 76 : if (! passive_p)
5578 : : {
5579 [ + + ]: 74 : if (verbose > 3)
5580 [ + - + - ]: 84 : obatched(clog) << "ddl: " << DEBUGINFOD_SQLITE_DDL << endl;
5581 : 74 : rc = sqlite3_exec (db, DEBUGINFOD_SQLITE_DDL, NULL, NULL, NULL);
5582 [ - + ]: 74 : if (rc != SQLITE_OK)
5583 : : {
5584 : 0 : error (EXIT_FAILURE, 0,
5585 : : "cannot run database schema ddl: %s", sqlite3_errmsg(db));
5586 : : }
5587 : : }
5588 : :
5589 [ + - + - : 152 : obatched(clog) << "libmicrohttpd version " << MHD_get_version() << endl;
+ - ]
5590 : :
5591 : : /* If '-C' wasn't given or was given with no arg, pick a reasonable default
5592 : : for the number of worker threads. */
5593 [ + + ]: 76 : if (connection_pool == 0)
5594 : 72 : connection_pool = default_concurrency();
5595 : :
5596 : : /* Note that MHD_USE_EPOLL and MHD_USE_THREAD_PER_CONNECTION don't
5597 : : work together. */
5598 : 76 : unsigned int use_epoll = 0;
5599 : : #if MHD_VERSION >= 0x00095100
5600 : 76 : use_epoll = MHD_USE_EPOLL;
5601 : : #endif
5602 : :
5603 : 76 : unsigned int mhd_flags = (
5604 : : #if MHD_VERSION >= 0x00095300
5605 : : MHD_USE_INTERNAL_POLLING_THREAD
5606 : : #else
5607 : : MHD_USE_SELECT_INTERNALLY
5608 : : #endif
5609 : : | MHD_USE_DUAL_STACK
5610 : : | use_epoll
5611 : : #if MHD_VERSION >= 0x00095200
5612 : : | MHD_USE_ITC
5613 : : #endif
5614 : : | MHD_USE_DEBUG); /* report errors to stderr */
5615 : :
5616 : : // Start httpd server threads. Use a single dual-homed pool.
5617 : 76 : MHD_Daemon *d46 = MHD_start_daemon (mhd_flags, http_port,
5618 : : NULL, NULL, /* default accept policy */
5619 : : handler_cb, NULL, /* handler callback */
5620 : : MHD_OPTION_EXTERNAL_LOGGER,
5621 : : error_cb, NULL,
5622 : : MHD_OPTION_THREAD_POOL_SIZE,
5623 : : (int)connection_pool,
5624 : : MHD_OPTION_END);
5625 : :
5626 : 76 : MHD_Daemon *d4 = NULL;
5627 [ - + ]: 76 : if (d46 == NULL)
5628 : : {
5629 : : // Cannot use dual_stack, use ipv4 only
5630 : 0 : mhd_flags &= ~(MHD_USE_DUAL_STACK);
5631 [ # # ]: 0 : d4 = MHD_start_daemon (mhd_flags, http_port,
5632 : : NULL, NULL, /* default accept policy */
5633 : : handler_cb, NULL, /* handler callback */
5634 : : MHD_OPTION_EXTERNAL_LOGGER,
5635 : : error_cb, NULL,
5636 : : (connection_pool
5637 : : ? MHD_OPTION_THREAD_POOL_SIZE
5638 : : : MHD_OPTION_END),
5639 : : (connection_pool
5640 : : ? (int)connection_pool
5641 : : : MHD_OPTION_END),
5642 : : MHD_OPTION_END);
5643 [ # # ]: 0 : if (d4 == NULL)
5644 : : {
5645 : 0 : sqlite3 *database = db;
5646 : 0 : sqlite3 *databaseq = dbq;
5647 : 0 : db = dbq = 0; // for signal_handler not to freak
5648 : 0 : sqlite3_close (databaseq);
5649 : 0 : sqlite3_close (database);
5650 : 0 : error (EXIT_FAILURE, 0, "cannot start http server at port %d",
5651 : : http_port);
5652 : : }
5653 : :
5654 : : }
5655 : 76 : obatched(clog) << "started http server on"
5656 : : << (d4 != NULL ? " IPv4 " : " IPv4 IPv6 ")
5657 [ + - + - : 152 : << "port=" << http_port << endl;
+ - + - +
- ]
5658 : :
5659 : : // add maxigroom sql if -G given
5660 [ - + ]: 76 : if (maxigroom)
5661 : : {
5662 [ # # ]: 0 : obatched(clog) << "maxigrooming database, please wait." << endl;
5663 : : // NB: this index alone can nearly double the database size!
5664 : : // NB: this index would be necessary to run source-file metadata searches fast
5665 [ # # ]: 0 : extra_ddl.push_back("create index if not exists " BUILDIDS "_r_sref_arc on " BUILDIDS "_r_sref(artifactsrc);");
5666 [ # # ]: 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);");
5667 [ # # ]: 0 : extra_ddl.push_back("drop index if exists " BUILDIDS "_r_sref_arc;");
5668 : :
5669 : : // NB: we don't maxigroom the _files interning table. It'd require a temp index on all the
5670 : : // tables that have file foreign-keys, which is a lot.
5671 : :
5672 : : // NB: with =delete, may take up 3x disk space total during vacuum process
5673 : : // vs. =off (only 2x but may corrupt database if program dies mid-vacuum)
5674 : : // vs. =wal (>3x observed, but safe)
5675 [ # # ]: 0 : extra_ddl.push_back("pragma journal_mode=delete;");
5676 [ # # ]: 0 : extra_ddl.push_back("vacuum;");
5677 [ # # ]: 0 : extra_ddl.push_back("pragma journal_mode=wal;");
5678 : : }
5679 : :
5680 : : // run extra -D sql if given
5681 [ + + ]: 76 : if (! passive_p)
5682 [ - + ]: 74 : for (auto&& i: extra_ddl)
5683 : : {
5684 [ # # ]: 0 : if (verbose > 1)
5685 [ # # # # ]: 0 : obatched(clog) << "extra ddl:\n" << i << endl;
5686 : 0 : rc = sqlite3_exec (db, i.c_str(), NULL, NULL, NULL);
5687 [ # # # # ]: 0 : if (rc != SQLITE_OK && rc != SQLITE_DONE && rc != SQLITE_ROW)
5688 : 0 : error (0, 0,
5689 : : "warning: cannot run database extra ddl %s: %s", i.c_str(), sqlite3_errmsg(db));
5690 : :
5691 [ # # ]: 0 : if (maxigroom)
5692 [ # # ]: 0 : obatched(clog) << "maxigroomed database" << endl;
5693 : : }
5694 : :
5695 [ + + ]: 76 : if (! passive_p)
5696 [ + - + - ]: 148 : obatched(clog) << "search concurrency " << concurrency << endl;
5697 : 76 : obatched(clog) << "webapi connection pool " << connection_pool
5698 [ + - - + : 76 : << (connection_pool ? "" : " (unlimited)") << endl;
+ - + - ]
5699 [ + + ]: 76 : if (! passive_p) {
5700 [ + - + - ]: 148 : obatched(clog) << "rescan time " << rescan_s << endl;
5701 [ + - + - ]: 148 : obatched(clog) << "scan checkpoint " << scan_checkpoint << endl;
5702 : : }
5703 [ + - + - ]: 152 : obatched(clog) << "fdcache mbs " << fdcache_mbs << endl;
5704 [ + - + - ]: 152 : obatched(clog) << "fdcache prefetch " << fdcache_prefetch << endl;
5705 [ + - + - ]: 152 : obatched(clog) << "fdcache tmpdir " << tmpdir << endl;
5706 [ + - + - ]: 152 : obatched(clog) << "fdcache tmpdir min% " << fdcache_mintmp << endl;
5707 [ + + ]: 76 : if (! passive_p)
5708 [ + - + - ]: 148 : obatched(clog) << "groom time " << groom_s << endl;
5709 [ + - + - ]: 152 : obatched(clog) << "forwarded ttl limit " << forwarded_ttl_limit << endl;
5710 : :
5711 [ + + ]: 76 : if (scan_archives.size()>0)
5712 : : {
5713 : 54 : obatched ob(clog);
5714 [ + - ]: 54 : auto& o = ob << "accepting archive types ";
5715 [ + + ]: 170 : for (auto&& arch : scan_archives)
5716 [ + - + - : 116 : o << arch.first << "(" << arch.second << ") ";
+ - + - ]
5717 [ + - ]: 54 : o << endl;
5718 : 54 : }
5719 : 76 : const char* du = getenv(DEBUGINFOD_URLS_ENV_VAR);
5720 [ + + + + ]: 76 : if (du && du[0] != '\0') // set to non-empty string?
5721 [ + - + - ]: 32 : obatched(clog) << "upstream debuginfod servers: " << du << endl;
5722 : :
5723 [ + + ]: 76 : vector<pthread_t> all_threads;
5724 : :
5725 [ + + ]: 76 : if (! passive_p)
5726 : : {
5727 : 74 : pthread_t pt;
5728 : 74 : rc = pthread_create (& pt, NULL, thread_main_groom, NULL);
5729 [ - + ]: 74 : if (rc)
5730 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to groom database\n");
5731 : : else
5732 : : {
5733 : : #ifdef HAVE_PTHREAD_SETNAME_NP
5734 : 74 : (void) pthread_setname_np (pt, "groom");
5735 : : #endif
5736 [ + - ]: 74 : all_threads.push_back(pt);
5737 : : }
5738 : :
5739 [ + + + + ]: 74 : if (scan_files || scan_archives.size() > 0)
5740 : : {
5741 [ + - ]: 68 : if (scan_checkpoint > 0)
5742 [ + - ]: 68 : scan_barrier = new sqlite_checkpoint_pb(concurrency, (unsigned) scan_checkpoint);
5743 : :
5744 : 68 : rc = pthread_create (& pt, NULL, thread_main_fts_source_paths, NULL);
5745 [ - + ]: 68 : if (rc)
5746 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to traverse source paths\n");
5747 : : #ifdef HAVE_PTHREAD_SETNAME_NP
5748 : 68 : (void) pthread_setname_np (pt, "traverse");
5749 : : #endif
5750 [ + - ]: 68 : all_threads.push_back(pt);
5751 : :
5752 [ + + ]: 340 : for (unsigned i=0; i<concurrency; i++)
5753 : : {
5754 : 272 : rc = pthread_create (& pt, NULL, thread_main_scanner, NULL);
5755 [ - + ]: 272 : if (rc)
5756 : 0 : error (EXIT_FAILURE, rc, "cannot spawn thread to scan source files / archives\n");
5757 : : #ifdef HAVE_PTHREAD_SETNAME_NP
5758 : 272 : (void) pthread_setname_np (pt, "scan");
5759 : : #endif
5760 [ + - ]: 272 : all_threads.push_back(pt);
5761 : : }
5762 : : }
5763 : : }
5764 : :
5765 : : /* Trivial main loop! */
5766 [ + - + - ]: 76 : set_metric("ready", 1);
5767 [ + + ]: 214 : while (! interrupted)
5768 [ + - ]: 138 : pause ();
5769 [ + - ]: 76 : scanq.nuke(); // wake up any remaining scanq-related threads, let them die
5770 [ + + + - ]: 76 : if (scan_barrier) scan_barrier->nuke(); // ... in case they're stuck in a barrier
5771 [ + - + - ]: 76 : set_metric("ready", 0);
5772 : :
5773 [ + - ]: 76 : if (verbose)
5774 [ + - + - : 152 : obatched(clog) << "stopping" << endl;
- - ]
5775 : :
5776 : : /* Join all our threads. */
5777 [ + + ]: 490 : for (auto&& it : all_threads)
5778 [ + - ]: 414 : pthread_join (it, NULL);
5779 : :
5780 : : /* Stop all the web service threads. */
5781 [ + - + - ]: 76 : if (d46) MHD_stop_daemon (d46);
5782 [ - + - - ]: 76 : if (d4) MHD_stop_daemon (d4);
5783 : :
5784 [ + + ]: 76 : if (! passive_p)
5785 : : {
5786 : : /* With all threads known dead, we can clean up the global resources. */
5787 [ + - ]: 74 : rc = sqlite3_exec (db, DEBUGINFOD_SQLITE_CLEANUP_DDL, NULL, NULL, NULL);
5788 [ - + ]: 74 : if (rc != SQLITE_OK)
5789 : : {
5790 [ # # # # ]: 0 : error (0, 0,
5791 : : "warning: cannot run database cleanup ddl: %s", sqlite3_errmsg(db));
5792 : : }
5793 : : }
5794 : :
5795 [ + - ]: 76 : debuginfod_pool_groom ();
5796 [ + + ]: 76 : delete scan_barrier;
5797 : :
5798 : : // NB: no problem with unconditional free here - an earlier failed regcomp would exit program
5799 [ + - ]: 76 : (void) regfree (& file_include_regex);
5800 [ + - ]: 76 : (void) regfree (& file_exclude_regex);
5801 : :
5802 : 76 : sqlite3 *database = db;
5803 : 76 : sqlite3 *databaseq = dbq;
5804 : 76 : db = dbq = 0; // for signal_handler not to freak
5805 [ + - ]: 76 : (void) sqlite3_close (databaseq);
5806 [ + + ]: 76 : if (! passive_p)
5807 [ + - ]: 74 : (void) sqlite3_close (database);
5808 : :
5809 [ + + ]: 76 : return 0;
5810 : 76 : }
|
