src/journal/journald-server.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2011 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <sys/signalfd.h>
  23 #include <sys/ioctl.h>
  24 #include <linux/sockios.h>
  25 #include <sys/statvfs.h>
  26 #include <sys/mman.h>
  27 #include <sys/timerfd.h>
  28
  29 #include <libudev.h>
  30 #include <systemd/sd-journal.h>
  31 #include <systemd/sd-messages.h>
  32 #include <systemd/sd-daemon.h>
  33
  34 #include "fileio.h"
  35 #include "mkdir.h"
  36 #include "hashmap.h"
  37 #include "journal-file.h"
  38 #include "socket-util.h"
  39 #include "cgroup-util.h"
  40 #include "list.h"
  41 #include "virt.h"
  42 #include "missing.h"
  43 #include "conf-parser.h"
  44 #include "journal-internal.h"
  45 #include "journal-vacuum.h"
  46 #include "journal-authenticate.h"
  47 #include "journald-server.h"
  48 #include "journald-rate-limit.h"
  49 #include "journald-kmsg.h"
  50 #include "journald-syslog.h"
  51 #include "journald-stream.h"
  52 #include "journald-console.h"
  53 #include "journald-native.h"
  54
  55 #ifdef HAVE_ACL
  56 #include <sys/acl.h>
  57 #include <acl/libacl.h>
  58 #include "acl-util.h"
  59 #endif
  60
  61 #ifdef HAVE_SELINUX
  62 #include <selinux/selinux.h>
  63 #endif
  64
  65 #define USER_JOURNALS_MAX 1024
  66
  67 #define DEFAULT_SYNC_INTERVAL_USEC (5*USEC_PER_MINUTE)
  68 #define DEFAULT_RATE_LIMIT_INTERVAL (30*USEC_PER_SEC)
  69 #define DEFAULT_RATE_LIMIT_BURST 1000
  70
  71 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
  72
  73 static const char* const storage_table[] = {
  74         [STORAGE_AUTO] = "auto",
  75         [STORAGE_VOLATILE] = "volatile",
  76         [STORAGE_PERSISTENT] = "persistent",
  77         [STORAGE_NONE] = "none"
  78 };
  79
  80 DEFINE_STRING_TABLE_LOOKUP(storage, Storage);
  81 DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting");
  82
  83 static const char* const split_mode_table[] = {
  84         [SPLIT_NONE] = "none",
  85         [SPLIT_UID] = "uid",
  86         [SPLIT_LOGIN] = "login"
  87 };
  88
  89 DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode);
  90 DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting");
  91
  92 static uint64_t available_space(Server *s, bool verbose) {
  93         char ids[33];
  94         _cleanup_free_ char *p = NULL;
  95         sd_id128_t machine;
  96         struct statvfs ss;
  97         uint64_t sum = 0, ss_avail = 0, avail = 0;
  98         int r;
  99         _cleanup_closedir_ DIR *d = NULL;
 100         usec_t ts;
 101         const char *f;
 102         JournalMetrics *m;
 103
 104         ts = now(CLOCK_MONOTONIC);
 105
 106         if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts
 107             && !verbose)
 108                 return s->cached_available_space;
 109
 110         r = sd_id128_get_machine(&machine);
 111         if (r < 0)
 112                 return 0;
 113
 114         if (s->system_journal) {
 115                 f = "/var/log/journal/";
 116                 m = &s->system_metrics;
 117         } else {
 118                 f = "/run/log/journal/";
 119                 m = &s->runtime_metrics;
 120         }
 121
 122         assert(m);
 123
 124         p = strappend(f, sd_id128_to_string(machine, ids));
 125         if (!p)
 126                 return 0;
 127
 128         d = opendir(p);
 129         if (!d)
 130                 return 0;
 131
 132         if (fstatvfs(dirfd(d), &ss) < 0)
 133                 return 0;
 134
 135         for (;;) {
 136                 struct stat st;
 137                 struct dirent *de;
 138                 union dirent_storage buf;
 139
 140                 r = readdir_r(d, &buf.de, &de);
 141                 if (r != 0)
 142                         break;
 143
 144                 if (!de)
 145                         break;
 146
 147                 if (!endswith(de->d_name, ".journal") &&
 148                     !endswith(de->d_name, ".journal~"))
 149                         continue;
 150
 151                 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
 152                         continue;
 153
 154                 if (!S_ISREG(st.st_mode))
 155                         continue;
 156
 157                 sum += (uint64_t) st.st_blocks * 512UL;
 158         }
 159
 160         ss_avail = ss.f_bsize * ss.f_bavail;
 161         avail = ss_avail > m->keep_free ? ss_avail - m->keep_free : 0;
 162
 163         s->cached_available_space = MIN(m->max_use, avail) > sum ? MIN(m->max_use, avail) - sum : 0;
 164         s->cached_available_space_timestamp = ts;
 165
 166         if (verbose) {
 167                 char    fb1[FORMAT_BYTES_MAX], fb2[FORMAT_BYTES_MAX], fb3[FORMAT_BYTES_MAX],
 168                         fb4[FORMAT_BYTES_MAX], fb5[FORMAT_BYTES_MAX];
 169
 170                 server_driver_message(s, SD_MESSAGE_JOURNAL_USAGE,
 171                                       "%s journal is using %s (max %s, leaving %s of free %s, current limit %s).",
 172                                       s->system_journal ? "Permanent" : "Runtime",
 173                                       format_bytes(fb1, sizeof(fb1), sum),
 174                                       format_bytes(fb2, sizeof(fb2), m->max_use),
 175                                       format_bytes(fb3, sizeof(fb3), m->keep_free),
 176                                       format_bytes(fb4, sizeof(fb4), ss_avail),
 177                                       format_bytes(fb5, sizeof(fb5), MIN(m->max_use, avail)));
 178         }
 179
 180         return s->cached_available_space;
 181 }
 182
 183 static void server_read_file_gid(Server *s) {
 184         const char *g = "systemd-journal";
 185         int r;
 186
 187         assert(s);
 188
 189         if (s->file_gid_valid)
 190                 return;
 191
 192         r = get_group_creds(&g, &s->file_gid);
 193         if (r < 0)
 194                 log_warning("Failed to resolve '%s' group: %s", g, strerror(-r));
 195
 196         /* if we couldn't read the gid, then it will be 0, but that's
 197          * fine and we shouldn't try to resolve the group again, so
 198          * let's just pretend it worked right-away. */
 199         s->file_gid_valid = true;
 200 }
 201
 202 void server_fix_perms(Server *s, JournalFile *f, uid_t uid) {
 203         int r;
 204 #ifdef HAVE_ACL
 205         acl_t acl;
 206         acl_entry_t entry;
 207         acl_permset_t permset;
 208 #endif
 209
 210         assert(f);
 211
 212         server_read_file_gid(s);
 213
 214         r = fchmod_and_fchown(f->fd, 0640, 0, s->file_gid);
 215         if (r < 0)
 216                 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
 217
 218 #ifdef HAVE_ACL
 219         if (uid <= 0)
 220                 return;
 221
 222         acl = acl_get_fd(f->fd);
 223         if (!acl) {
 224                 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
 225                 return;
 226         }
 227
 228         r = acl_find_uid(acl, uid, &entry);
 229         if (r <= 0) {
 230
 231                 if (acl_create_entry(&acl, &entry) < 0 ||
 232                     acl_set_tag_type(entry, ACL_USER) < 0 ||
 233                     acl_set_qualifier(entry, &uid) < 0) {
 234                         log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
 235                         goto finish;
 236                 }
 237         }
 238
 239         /* We do not recalculate the mask unconditionally here,
 240          * so that the fchmod() mask above stays intact. */
 241         if (acl_get_permset(entry, &permset) < 0 ||
 242             acl_add_perm(permset, ACL_READ) < 0 ||
 243             calc_acl_mask_if_needed(&acl) < 0) {
 244                 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
 245                 goto finish;
 246         }
 247
 248         if (acl_set_fd(f->fd, acl) < 0)
 249                 log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
 250
 251 finish:
 252         acl_free(acl);
 253 #endif
 254 }
 255
 256 static JournalFile* find_journal(Server *s, uid_t uid) {
 257         _cleanup_free_ char *p = NULL;
 258         int r;
 259         JournalFile *f;
 260         sd_id128_t machine;
 261
 262         assert(s);
 263
 264         /* We split up user logs only on /var, not on /run. If the
 265          * runtime file is open, we write to it exclusively, in order
 266          * to guarantee proper order as soon as we flush /run to
 267          * /var and close the runtime file. */
 268
 269         if (s->runtime_journal)
 270                 return s->runtime_journal;
 271
 272         if (uid <= 0)
 273                 return s->system_journal;
 274
 275         r = sd_id128_get_machine(&machine);
 276         if (r < 0)
 277                 return s->system_journal;
 278
 279         f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
 280         if (f)
 281                 return f;
 282
 283         if (asprintf(&p, "/var/log/journal/" SD_ID128_FORMAT_STR "/user-%lu.journal",
 284                      SD_ID128_FORMAT_VAL(machine), (unsigned long) uid) < 0)
 285                 return s->system_journal;
 286
 287         while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
 288                 /* Too many open? Then let's close one */
 289                 f = hashmap_steal_first(s->user_journals);
 290                 assert(f);
 291                 journal_file_close(f);
 292         }
 293
 294         r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &f);
 295         if (r < 0)
 296                 return s->system_journal;
 297
 298         server_fix_perms(s, f, uid);
 299
 300         r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
 301         if (r < 0) {
 302                 journal_file_close(f);
 303                 return s->system_journal;
 304         }
 305
 306         return f;
 307 }
 308
 309 void server_rotate(Server *s) {
 310         JournalFile *f;
 311         void *k;
 312         Iterator i;
 313         int r;
 314
 315         log_debug("Rotating...");
 316
 317         if (s->runtime_journal) {
 318                 r = journal_file_rotate(&s->runtime_journal, s->compress, false);
 319                 if (r < 0)
 320                         if (s->runtime_journal)
 321                                 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
 322                         else
 323                                 log_error("Failed to create new runtime journal: %s", strerror(-r));
 324                 else
 325                         server_fix_perms(s, s->runtime_journal, 0);
 326         }
 327
 328         if (s->system_journal) {
 329                 r = journal_file_rotate(&s->system_journal, s->compress, s->seal);
 330                 if (r < 0)
 331                         if (s->system_journal)
 332                                 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
 333                         else
 334                                 log_error("Failed to create new system journal: %s", strerror(-r));
 335
 336                 else
 337                         server_fix_perms(s, s->system_journal, 0);
 338         }
 339
 340         HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
 341                 r = journal_file_rotate(&f, s->compress, s->seal);
 342                 if (r < 0)
 343                         if (f)
 344                                 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
 345                         else
 346                                 log_error("Failed to create user journal: %s", strerror(-r));
 347                 else {
 348                         hashmap_replace(s->user_journals, k, f);
 349                         server_fix_perms(s, f, PTR_TO_UINT32(k));
 350                 }
 351         }
 352 }
 353
 354 void server_sync(Server *s) {
 355         JournalFile *f;
 356         void *k;
 357         Iterator i;
 358         int r;
 359
 360         static const struct itimerspec sync_timer_disable = {};
 361
 362         if (s->system_journal) {
 363                 r = journal_file_set_offline(s->system_journal);
 364                 if (r < 0)
 365                         log_error("Failed to sync system journal: %s", strerror(-r));
 366         }
 367
 368         HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
 369                 r = journal_file_set_offline(f);
 370                 if (r < 0)
 371                         log_error("Failed to sync user journal: %s", strerror(-r));
 372         }
 373
 374         r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_disable, NULL);
 375         if (r < 0)
 376                 log_error("Failed to disable max timer: %m");
 377
 378         s->sync_scheduled = false;
 379 }
 380
 381 void server_vacuum(Server *s) {
 382         char ids[33];
 383         sd_id128_t machine;
 384         int r;
 385
 386         log_debug("Vacuuming...");
 387
 388         s->oldest_file_usec = 0;
 389
 390         r = sd_id128_get_machine(&machine);
 391         if (r < 0) {
 392                 log_error("Failed to get machine ID: %s", strerror(-r));
 393                 return;
 394         }
 395
 396         sd_id128_to_string(machine, ids);
 397
 398         if (s->system_journal) {
 399                 char *p = strappenda("/var/log/journal/", ids);
 400
 401                 r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
 402                 if (r < 0 && r != -ENOENT)
 403                         log_error("Failed to vacuum %s: %s", p, strerror(-r));
 404         }
 405
 406         if (s->runtime_journal) {
 407                 char *p = strappenda("/run/log/journal/", ids);
 408
 409                 r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
 410                 if (r < 0 && r != -ENOENT)
 411                         log_error("Failed to vacuum %s: %s", p, strerror(-r));
 412         }
 413
 414         s->cached_available_space_timestamp = 0;
 415 }
 416
 417 bool shall_try_append_again(JournalFile *f, int r) {
 418
 419         /* -E2BIG            Hit configured limit
 420            -EFBIG            Hit fs limit
 421            -EDQUOT           Quota limit hit
 422            -ENOSPC           Disk full
 423            -EHOSTDOWN        Other machine
 424            -EBUSY            Unclean shutdown
 425            -EPROTONOSUPPORT  Unsupported feature
 426            -EBADMSG          Corrupted
 427            -ENODATA          Truncated
 428            -ESHUTDOWN        Already archived */
 429
 430         if (r == -E2BIG || r == -EFBIG || r == -EDQUOT || r == -ENOSPC)
 431                 log_debug("%s: Allocation limit reached, rotating.", f->path);
 432         else if (r == -EHOSTDOWN)
 433                 log_info("%s: Journal file from other machine, rotating.", f->path);
 434         else if (r == -EBUSY)
 435                 log_info("%s: Unclean shutdown, rotating.", f->path);
 436         else if (r == -EPROTONOSUPPORT)
 437                 log_info("%s: Unsupported feature, rotating.", f->path);
 438         else if (r == -EBADMSG || r == -ENODATA || r == ESHUTDOWN)
 439                 log_warning("%s: Journal file corrupted, rotating.", f->path);
 440         else
 441                 return false;
 442
 443         return true;
 444 }
 445
 446 static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned n) {
 447         JournalFile *f;
 448         bool vacuumed = false;
 449         int r;
 450
 451         assert(s);
 452         assert(iovec);
 453         assert(n > 0);
 454
 455         f = find_journal(s, uid);
 456         if (!f)
 457                 return;
 458
 459         if (journal_file_rotate_suggested(f, s->max_file_usec)) {
 460                 log_debug("%s: Journal header limits reached or header out-of-date, rotating.", f->path);
 461                 server_rotate(s);
 462                 server_vacuum(s);
 463                 vacuumed = true;
 464
 465                 f = find_journal(s, uid);
 466                 if (!f)
 467                         return;
 468         }
 469
 470         r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
 471         if (r >= 0) {
 472                 server_schedule_sync(s);
 473                 return;
 474         }
 475
 476         if (vacuumed || !shall_try_append_again(f, r)) {
 477                 size_t size = 0;
 478                 unsigned i;
 479                 for (i = 0; i < n; i++)
 480                         size += iovec[i].iov_len;
 481
 482                 log_error("Failed to write entry (%d items, %zu bytes), ignoring: %s", n, size, strerror(-r));
 483                 return;
 484         }
 485
 486         server_rotate(s);
 487         server_vacuum(s);
 488
 489         f = find_journal(s, uid);
 490         if (!f)
 491                 return;
 492
 493         log_debug("Retrying write.");
 494         r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
 495         if (r < 0) {
 496                 size_t size = 0;
 497                 unsigned i;
 498                 for (i = 0; i < n; i++)
 499                         size += iovec[i].iov_len;
 500
 501                 log_error("Failed to write entry (%d items, %zu bytes) despite vacuuming, ignoring: %s", n, size, strerror(-r));
 502         }
 503 }
 504
 505 static void dispatch_message_real(
 506                 Server *s,
 507                 struct iovec *iovec, unsigned n, unsigned m,
 508                 struct ucred *ucred,
 509                 struct timeval *tv,
 510                 const char *label, size_t label_len,
 511                 const char *unit_id,
 512                 pid_t object_pid) {
 513
 514         char    pid[sizeof("_PID=") + DECIMAL_STR_MAX(pid_t)],
 515                 uid[sizeof("_UID=") + DECIMAL_STR_MAX(uid_t)],
 516                 gid[sizeof("_GID=") + DECIMAL_STR_MAX(gid_t)],
 517                 owner_uid[sizeof("_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)],
 518                 source_time[sizeof("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t)],
 519                 boot_id[sizeof("_BOOT_ID=") + 32] = "_BOOT_ID=",
 520                 machine_id[sizeof("_MACHINE_ID=") + 32] = "_MACHINE_ID=",
 521                 o_uid[sizeof("OBJECT_UID=") + DECIMAL_STR_MAX(uid_t)],
 522                 o_gid[sizeof("OBJECT_GID=") + DECIMAL_STR_MAX(gid_t)],
 523                 o_owner_uid[sizeof("OBJECT_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)];
 524         uid_t object_uid;
 525         gid_t object_gid;
 526
 527         char *x;
 528         sd_id128_t id;
 529         int r;
 530         char *t, *c;
 531         uid_t realuid = 0, owner = 0, journal_uid;
 532         bool owner_valid = false;
 533 #ifdef HAVE_AUDIT
 534         char    audit_session[sizeof("_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
 535                 audit_loginuid[sizeof("_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)],
 536                 o_audit_session[sizeof("OBJECT_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
 537                 o_audit_loginuid[sizeof("OBJECT_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)];
 538
 539         uint32_t audit;
 540         uid_t loginuid;
 541 #endif
 542
 543         assert(s);
 544         assert(iovec);
 545         assert(n > 0);
 546         assert(n + N_IOVEC_META_FIELDS + (object_pid ? N_IOVEC_OBJECT_FIELDS : 0) <= m);
 547
 548         if (ucred) {
 549                 realuid = ucred->uid;
 550
 551                 sprintf(pid, "_PID=%lu", (unsigned long) ucred->pid);
 552                 IOVEC_SET_STRING(iovec[n++], pid);
 553
 554                 sprintf(uid, "_UID=%lu", (unsigned long) ucred->uid);
 555                 IOVEC_SET_STRING(iovec[n++], uid);
 556
 557                 sprintf(gid, "_GID=%lu", (unsigned long) ucred->gid);
 558                 IOVEC_SET_STRING(iovec[n++], gid);
 559
 560                 r = get_process_comm(ucred->pid, &t);
 561                 if (r >= 0) {
 562                         x = strappenda("_COMM=", t);
 563                         free(t);
 564                         IOVEC_SET_STRING(iovec[n++], x);
 565                 }
 566
 567                 r = get_process_exe(ucred->pid, &t);
 568                 if (r >= 0) {
 569                         x = strappenda("_EXE=", t);
 570                         free(t);
 571                         IOVEC_SET_STRING(iovec[n++], x);
 572                 }
 573
 574                 r = get_process_cmdline(ucred->pid, 0, false, &t);
 575                 if (r >= 0) {
 576                         x = strappenda("_CMDLINE=", t);
 577                         free(t);
 578                         IOVEC_SET_STRING(iovec[n++], x);
 579                 }
 580
 581                 r = get_process_capeff(ucred->pid, &t);
 582                 if (r >= 0) {
 583                         x = strappenda("_CAP_EFFECTIVE=", t);
 584                         free(t);
 585                         IOVEC_SET_STRING(iovec[n++], x);
 586                 }
 587
 588 #ifdef HAVE_AUDIT
 589                 r = audit_session_from_pid(ucred->pid, &audit);
 590                 if (r >= 0) {
 591                         sprintf(audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit);
 592                         IOVEC_SET_STRING(iovec[n++], audit_session);
 593                 }
 594
 595                 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
 596                 if (r >= 0) {
 597                         sprintf(audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
 598                         IOVEC_SET_STRING(iovec[n++], audit_loginuid);
 599                 }
 600 #endif
 601
 602                 r = cg_pid_get_path_shifted(ucred->pid, NULL, &c);
 603                 if (r >= 0) {
 604                         char *session = NULL;
 605
 606                         x = strappenda("_SYSTEMD_CGROUP=", c);
 607                         IOVEC_SET_STRING(iovec[n++], x);
 608
 609                         r = cg_path_get_session(c, &t);
 610                         if (r >= 0) {
 611                                 session = strappenda("_SYSTEMD_SESSION=", t);
 612                                 free(t);
 613                                 IOVEC_SET_STRING(iovec[n++], session);
 614                         }
 615
 616                         if (cg_path_get_owner_uid(c, &owner) >= 0) {
 617                                 owner_valid = true;
 618
 619                                 sprintf(owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
 620                                 IOVEC_SET_STRING(iovec[n++], owner_uid);
 621                         }
 622
 623                         if (cg_path_get_unit(c, &t) >= 0) {
 624                                 x = strappenda("_SYSTEMD_UNIT=", t);
 625                                 free(t);
 626                                 IOVEC_SET_STRING(iovec[n++], x);
 627                         } else if (unit_id && !session) {
 628                                 x = strappenda("_SYSTEMD_UNIT=", unit_id);
 629                                 IOVEC_SET_STRING(iovec[n++], x);
 630                         }
 631
 632                         if (cg_path_get_user_unit(c, &t) >= 0) {
 633                                 x = strappenda("_SYSTEMD_USER_UNIT=", t);
 634                                 free(t);
 635                                 IOVEC_SET_STRING(iovec[n++], x);
 636                         } else if (unit_id && session) {
 637                                 x = strappenda("_SYSTEMD_USER_UNIT=", unit_id);
 638                                 IOVEC_SET_STRING(iovec[n++], x);
 639                         }
 640
 641                         free(c);
 642                 }
 643
 644 #ifdef HAVE_SELINUX
 645                 if (label) {
 646                         x = alloca(sizeof("_SELINUX_CONTEXT=") + label_len);
 647
 648                         *((char*) mempcpy(stpcpy(x, "_SELINUX_CONTEXT="), label, label_len)) = 0;
 649                         IOVEC_SET_STRING(iovec[n++], x);
 650                 } else {
 651                         security_context_t con;
 652
 653                         if (getpidcon(ucred->pid, &con) >= 0) {
 654                                 x = strappenda("_SELINUX_CONTEXT=", con);
 655
 656                                 freecon(con);
 657                                 IOVEC_SET_STRING(iovec[n++], x);
 658                         }
 659                 }
 660 #endif
 661         }
 662         assert(n <= m);
 663
 664         if (object_pid) {
 665                 r = get_process_uid(object_pid, &object_uid);
 666                 if (r >= 0) {
 667                         sprintf(o_uid, "OBJECT_UID=%lu", (unsigned long) object_uid);
 668                         IOVEC_SET_STRING(iovec[n++], o_uid);
 669                 }
 670
 671                 r = get_process_gid(object_pid, &object_gid);
 672                 if (r >= 0) {
 673                         sprintf(o_gid, "OBJECT_GID=%lu", (unsigned long) object_gid);
 674                         IOVEC_SET_STRING(iovec[n++], o_gid);
 675                 }
 676
 677                 r = get_process_comm(object_pid, &t);
 678                 if (r >= 0) {
 679                         x = strappenda("OBJECT_COMM=", t);
 680                         free(t);
 681                         IOVEC_SET_STRING(iovec[n++], x);
 682                 }
 683
 684                 r = get_process_exe(object_pid, &t);
 685                 if (r >= 0) {
 686                         x = strappenda("OBJECT_EXE=", t);
 687                         free(t);
 688                         IOVEC_SET_STRING(iovec[n++], x);
 689                 }
 690
 691                 r = get_process_cmdline(object_pid, 0, false, &t);
 692                 if (r >= 0) {
 693                         x = strappenda("OBJECT_CMDLINE=", t);
 694                         free(t);
 695                         IOVEC_SET_STRING(iovec[n++], x);
 696                 }
 697
 698 #ifdef HAVE_AUDIT
 699                 r = audit_session_from_pid(object_pid, &audit);
 700                 if (r >= 0) {
 701                         sprintf(o_audit_session, "OBJECT_AUDIT_SESSION=%lu", (unsigned long) audit);
 702                         IOVEC_SET_STRING(iovec[n++], o_audit_session);
 703                 }
 704
 705                 r = audit_loginuid_from_pid(object_pid, &loginuid);
 706                 if (r >= 0) {
 707                         sprintf(o_audit_loginuid, "OBJECT_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
 708                         IOVEC_SET_STRING(iovec[n++], o_audit_loginuid);
 709                 }
 710 #endif
 711
 712                 r = cg_pid_get_path_shifted(object_pid, NULL, &c);
 713                 if (r >= 0) {
 714                         x = strappenda("OBJECT_SYSTEMD_CGROUP=", c);
 715                         IOVEC_SET_STRING(iovec[n++], x);
 716
 717                         r = cg_path_get_session(c, &t);
 718                         if (r >= 0) {
 719                                 x = strappenda("OBJECT_SYSTEMD_SESSION=", t);
 720                                 free(t);
 721                                 IOVEC_SET_STRING(iovec[n++], x);
 722                         }
 723
 724                         if (cg_path_get_owner_uid(c, &owner) >= 0) {
 725                                 sprintf(o_owner_uid, "OBJECT_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
 726                                 IOVEC_SET_STRING(iovec[n++], o_owner_uid);
 727                         }
 728
 729                         if (cg_path_get_unit(c, &t) >= 0) {
 730                                 x = strappenda("OBJECT_SYSTEMD_UNIT=", t);
 731                                 free(t);
 732                                 IOVEC_SET_STRING(iovec[n++], x);
 733                         }
 734
 735                         if (cg_path_get_user_unit(c, &t) >= 0) {
 736                                 x = strappenda("OBJECT_SYSTEMD_USER_UNIT=", t);
 737                                 free(t);
 738                                 IOVEC_SET_STRING(iovec[n++], x);
 739                         }
 740
 741                         free(c);
 742                 }
 743         }
 744         assert(n <= m);
 745
 746         if (tv) {
 747                 sprintf(source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu", (unsigned long long) timeval_load(tv));
 748                 IOVEC_SET_STRING(iovec[n++], source_time);
 749         }
 750
 751         /* Note that strictly speaking storing the boot id here is
 752          * redundant since the entry includes this in-line
 753          * anyway. However, we need this indexed, too. */
 754         r = sd_id128_get_boot(&id);
 755         if (r >= 0) {
 756                 sd_id128_to_string(id, boot_id + strlen("_BOOT_ID="));
 757                 IOVEC_SET_STRING(iovec[n++], boot_id);
 758         }
 759
 760         r = sd_id128_get_machine(&id);
 761         if (r >= 0) {
 762                 sd_id128_to_string(id, machine_id + strlen("_MACHINE_ID="));
 763                 IOVEC_SET_STRING(iovec[n++], machine_id);
 764         }
 765
 766         t = gethostname_malloc();
 767         if (t) {
 768                 x = strappenda("_HOSTNAME=", t);
 769                 free(t);
 770                 IOVEC_SET_STRING(iovec[n++], x);
 771         }
 772
 773         assert(n <= m);
 774
 775         if (s->split_mode == SPLIT_UID && realuid > 0)
 776                 /* Split up strictly by any UID */
 777                 journal_uid = realuid;
 778         else if (s->split_mode == SPLIT_LOGIN && realuid > 0 && owner_valid && owner > 0)
 779                 /* Split up by login UIDs, this avoids creation of
 780                  * individual journals for system UIDs.  We do this
 781                  * only if the realuid is not root, in order not to
 782                  * accidentally leak privileged information to the
 783                  * user that is logged by a privileged process that is
 784                  * part of an unprivileged session.*/
 785                 journal_uid = owner;
 786         else
 787                 journal_uid = 0;
 788
 789         write_to_journal(s, journal_uid, iovec, n);
 790 }
 791
 792 void server_driver_message(Server *s, sd_id128_t message_id, const char *format, ...) {
 793         char mid[11 + 32 + 1];
 794         char buffer[16 + LINE_MAX + 1];
 795         struct iovec iovec[N_IOVEC_META_FIELDS + 4];
 796         int n = 0;
 797         va_list ap;
 798         struct ucred ucred = {};
 799
 800         assert(s);
 801         assert(format);
 802
 803         IOVEC_SET_STRING(iovec[n++], "PRIORITY=6");
 804         IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver");
 805
 806         memcpy(buffer, "MESSAGE=", 8);
 807         va_start(ap, format);
 808         vsnprintf(buffer + 8, sizeof(buffer) - 8, format, ap);
 809         va_end(ap);
 810         char_array_0(buffer);
 811         IOVEC_SET_STRING(iovec[n++], buffer);
 812
 813         if (!sd_id128_equal(message_id, SD_ID128_NULL)) {
 814                 snprintf(mid, sizeof(mid), MESSAGE_ID(message_id));
 815                 char_array_0(mid);
 816                 IOVEC_SET_STRING(iovec[n++], mid);
 817         }
 818
 819         ucred.pid = getpid();
 820         ucred.uid = getuid();
 821         ucred.gid = getgid();
 822
 823         dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0, NULL, 0);
 824 }
 825
 826 void server_dispatch_message(
 827                 Server *s,
 828                 struct iovec *iovec, unsigned n, unsigned m,
 829                 struct ucred *ucred,
 830                 struct timeval *tv,
 831                 const char *label, size_t label_len,
 832                 const char *unit_id,
 833                 int priority,
 834                 pid_t object_pid) {
 835
 836         int rl, r;
 837         _cleanup_free_ char *path = NULL;
 838         char *c;
 839
 840         assert(s);
 841         assert(iovec || n == 0);
 842
 843         if (n == 0)
 844                 return;
 845
 846         if (LOG_PRI(priority) > s->max_level_store)
 847                 return;
 848
 849         /* Stop early in case the information will not be stored
 850          * in a journal. */
 851         if (s->storage == STORAGE_NONE)
 852                 return;
 853
 854         if (!ucred)
 855                 goto finish;
 856
 857         r = cg_pid_get_path_shifted(ucred->pid, NULL, &path);
 858         if (r < 0)
 859                 goto finish;
 860
 861         /* example: /user/lennart/3/foobar
 862          *          /system/dbus.service/foobar
 863          *
 864          * So let's cut of everything past the third /, since that is
 865          * where user directories start */
 866
 867         c = strchr(path, '/');
 868         if (c) {
 869                 c = strchr(c+1, '/');
 870                 if (c) {
 871                         c = strchr(c+1, '/');
 872                         if (c)
 873                                 *c = 0;
 874                 }
 875         }
 876
 877         rl = journal_rate_limit_test(s->rate_limit, path,
 878                                      priority & LOG_PRIMASK, available_space(s, false));
 879
 880         if (rl == 0)
 881                 return;
 882
 883         /* Write a suppression message if we suppressed something */
 884         if (rl > 1)
 885                 server_driver_message(s, SD_MESSAGE_JOURNAL_DROPPED,
 886                                       "Suppressed %u messages from %s", rl - 1, path);
 887
 888 finish:
 889         dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len, unit_id, object_pid);
 890 }
 891
 892
 893 static int system_journal_open(Server *s) {
 894         int r;
 895         char *fn;
 896         sd_id128_t machine;
 897         char ids[33];
 898
 899         r = sd_id128_get_machine(&machine);
 900         if (r < 0)
 901                 return r;
 902
 903         sd_id128_to_string(machine, ids);
 904
 905         if (!s->system_journal &&
 906             (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) &&
 907             access("/run/systemd/journal/flushed", F_OK) >= 0) {
 908
 909                 /* If in auto mode: first try to create the machine
 910                  * path, but not the prefix.
 911                  *
 912                  * If in persistent mode: create /var/log/journal and
 913                  * the machine path */
 914
 915                 if (s->storage == STORAGE_PERSISTENT)
 916                         (void) mkdir("/var/log/journal/", 0755);
 917
 918                 fn = strappenda("/var/log/journal/", ids);
 919                 (void) mkdir(fn, 0755);
 920
 921                 fn = strappenda(fn, "/system.journal");
 922                 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &s->system_journal);
 923
 924                 if (r >= 0)
 925                         server_fix_perms(s, s->system_journal, 0);
 926                 else if (r < 0) {
 927                         if (r != -ENOENT && r != -EROFS)
 928                                 log_warning("Failed to open system journal: %s", strerror(-r));
 929
 930                         r = 0;
 931                 }
 932         }
 933
 934         if (!s->runtime_journal &&
 935             (s->storage != STORAGE_NONE)) {
 936
 937                 fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL);
 938                 if (!fn)
 939                         return -ENOMEM;
 940
 941                 if (s->system_journal) {
 942
 943                         /* Try to open the runtime journal, but only
 944                          * if it already exists, so that we can flush
 945                          * it into the system journal */
 946
 947                         r = journal_file_open(fn, O_RDWR, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
 948                         free(fn);
 949
 950                         if (r < 0) {
 951                                 if (r != -ENOENT)
 952                                         log_warning("Failed to open runtime journal: %s", strerror(-r));
 953
 954                                 r = 0;
 955                         }
 956
 957                 } else {
 958
 959                         /* OK, we really need the runtime journal, so create
 960                          * it if necessary. */
 961
 962                         (void) mkdir_parents(fn, 0755);
 963                         r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
 964                         free(fn);
 965
 966                         if (r < 0) {
 967                                 log_error("Failed to open runtime journal: %s", strerror(-r));
 968                                 return r;
 969                         }
 970                 }
 971
 972                 if (s->runtime_journal)
 973                         server_fix_perms(s, s->runtime_journal, 0);
 974         }
 975
 976         available_space(s, true);
 977
 978         return r;
 979 }
 980
 981 int server_flush_to_var(Server *s) {
 982         int r;
 983         sd_id128_t machine;
 984         sd_journal *j = NULL;
 985
 986         assert(s);
 987
 988         if (s->storage != STORAGE_AUTO &&
 989             s->storage != STORAGE_PERSISTENT)
 990                 return 0;
 991
 992         if (!s->runtime_journal)
 993                 return 0;
 994
 995         system_journal_open(s);
 996
 997         if (!s->system_journal)
 998                 return 0;
 999
1000         log_debug("Flushing to /var...");
1001
1002         r = sd_id128_get_machine(&machine);
1003         if (r < 0) {
1004                 log_error("Failed to get machine id: %s", strerror(-r));
1005                 return r;
1006         }
1007
1008         r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
1009         if (r < 0) {
1010                 log_error("Failed to read runtime journal: %s", strerror(-r));
1011                 return r;
1012         }
1013
1014         sd_journal_set_data_threshold(j, 0);
1015
1016         SD_JOURNAL_FOREACH(j) {
1017                 Object *o = NULL;
1018                 JournalFile *f;
1019
1020                 f = j->current_file;
1021                 assert(f && f->current_offset > 0);
1022
1023                 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1024                 if (r < 0) {
1025                         log_error("Can't read entry: %s", strerror(-r));
1026                         goto finish;
1027                 }
1028
1029                 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1030                 if (r >= 0)
1031                         continue;
1032
1033                 if (!shall_try_append_again(s->system_journal, r)) {
1034                         log_error("Can't write entry: %s", strerror(-r));
1035                         goto finish;
1036                 }
1037
1038                 server_rotate(s);
1039                 server_vacuum(s);
1040
1041                 if (!s->system_journal) {
1042                         log_notice("Didn't flush runtime journal since rotation of system journal wasn't successful.");
1043                         r = -EIO;
1044                         goto finish;
1045                 }
1046
1047                 log_debug("Retrying write.");
1048                 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1049                 if (r < 0) {
1050                         log_error("Can't write entry: %s", strerror(-r));
1051                         goto finish;
1052                 }
1053         }
1054
1055 finish:
1056         journal_file_post_change(s->system_journal);
1057
1058         journal_file_close(s->runtime_journal);
1059         s->runtime_journal = NULL;
1060
1061         if (r >= 0)
1062                 rm_rf("/run/log/journal", false, true, false);
1063
1064         sd_journal_close(j);
1065
1066         return r;
1067 }
1068
1069 int process_event(Server *s, struct epoll_event *ev) {
1070         assert(s);
1071         assert(ev);
1072
1073         if (ev->data.fd == s->signal_fd) {
1074                 struct signalfd_siginfo sfsi;
1075                 ssize_t n;
1076
1077                 if (ev->events != EPOLLIN) {
1078                         log_error("Got invalid event from epoll.");
1079                         return -EIO;
1080                 }
1081
1082                 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
1083                 if (n != sizeof(sfsi)) {
1084
1085                         if (n >= 0)
1086                                 return -EIO;
1087
1088                         if (errno == EINTR || errno == EAGAIN)
1089                                 return 1;
1090
1091                         return -errno;
1092                 }
1093
1094                 if (sfsi.ssi_signo == SIGUSR1) {
1095                         touch("/run/systemd/journal/flushed");
1096                         server_flush_to_var(s);
1097                         server_sync(s);
1098                         return 1;
1099                 }
1100
1101                 if (sfsi.ssi_signo == SIGUSR2) {
1102                         server_rotate(s);
1103                         server_vacuum(s);
1104                         return 1;
1105                 }
1106
1107                 log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1108
1109                 return 0;
1110
1111         } else if (ev->data.fd == s->sync_timer_fd) {
1112                 int r;
1113                 uint64_t t;
1114
1115                 log_debug("Got sync request from epoll.");
1116
1117                 r = read(ev->data.fd, (void *)&t, sizeof(t));
1118                 if (r < 0)
1119                         return 0;
1120
1121                 server_sync(s);
1122                 return 1;
1123
1124         } else if (ev->data.fd == s->dev_kmsg_fd) {
1125                 int r;
1126
1127                 if (ev->events != EPOLLIN) {
1128                         log_error("Got invalid event from epoll.");
1129                         return -EIO;
1130                 }
1131
1132                 r = server_read_dev_kmsg(s);
1133                 if (r < 0)
1134                         return r;
1135
1136                 return 1;
1137
1138         } else if (ev->data.fd == s->native_fd ||
1139                    ev->data.fd == s->syslog_fd) {
1140
1141                 if (ev->events != EPOLLIN) {
1142                         log_error("Got invalid event from epoll.");
1143                         return -EIO;
1144                 }
1145
1146                 for (;;) {
1147                         struct msghdr msghdr;
1148                         struct iovec iovec;
1149                         struct ucred *ucred = NULL;
1150                         struct timeval *tv = NULL;
1151                         struct cmsghdr *cmsg;
1152                         char *label = NULL;
1153                         size_t label_len = 0;
1154                         union {
1155                                 struct cmsghdr cmsghdr;
1156
1157                                 /* We use NAME_MAX space for the
1158                                  * SELinux label here. The kernel
1159                                  * currently enforces no limit, but
1160                                  * according to suggestions from the
1161                                  * SELinux people this will change and
1162                                  * it will probably be identical to
1163                                  * NAME_MAX. For now we use that, but
1164                                  * this should be updated one day when
1165                                  * the final limit is known.*/
1166                                 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1167                                             CMSG_SPACE(sizeof(struct timeval)) +
1168                                             CMSG_SPACE(sizeof(int)) + /* fd */
1169                                             CMSG_SPACE(NAME_MAX)]; /* selinux label */
1170                         } control;
1171                         ssize_t n;
1172                         int v;
1173                         int *fds = NULL;
1174                         unsigned n_fds = 0;
1175
1176                         if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1177                                 log_error("SIOCINQ failed: %m");
1178                                 return -errno;
1179                         }
1180
1181                         if (s->buffer_size < (size_t) v) {
1182                                 void *b;
1183                                 size_t l;
1184
1185                                 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1186                                 b = realloc(s->buffer, l+1);
1187
1188                                 if (!b) {
1189                                         log_error("Couldn't increase buffer.");
1190                                         return -ENOMEM;
1191                                 }
1192
1193                                 s->buffer_size = l;
1194                                 s->buffer = b;
1195                         }
1196
1197                         zero(iovec);
1198                         iovec.iov_base = s->buffer;
1199                         iovec.iov_len = s->buffer_size;
1200
1201                         zero(control);
1202                         zero(msghdr);
1203                         msghdr.msg_iov = &iovec;
1204                         msghdr.msg_iovlen = 1;
1205                         msghdr.msg_control = &control;
1206                         msghdr.msg_controllen = sizeof(control);
1207
1208                         n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1209                         if (n < 0) {
1210
1211                                 if (errno == EINTR || errno == EAGAIN)
1212                                         return 1;
1213
1214                                 log_error("recvmsg() failed: %m");
1215                                 return -errno;
1216                         }
1217
1218                         for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1219
1220                                 if (cmsg->cmsg_level == SOL_SOCKET &&
1221                                     cmsg->cmsg_type == SCM_CREDENTIALS &&
1222                                     cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1223                                         ucred = (struct ucred*) CMSG_DATA(cmsg);
1224                                 else if (cmsg->cmsg_level == SOL_SOCKET &&
1225                                          cmsg->cmsg_type == SCM_SECURITY) {
1226                                         label = (char*) CMSG_DATA(cmsg);
1227                                         label_len = cmsg->cmsg_len - CMSG_LEN(0);
1228                                 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1229                                            cmsg->cmsg_type == SO_TIMESTAMP &&
1230                                            cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1231                                         tv = (struct timeval*) CMSG_DATA(cmsg);
1232                                 else if (cmsg->cmsg_level == SOL_SOCKET &&
1233                                          cmsg->cmsg_type == SCM_RIGHTS) {
1234                                         fds = (int*) CMSG_DATA(cmsg);
1235                                         n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1236                                 }
1237                         }
1238
1239                         if (ev->data.fd == s->syslog_fd) {
1240                                 char *e;
1241
1242                                 if (n > 0 && n_fds == 0) {
1243                                         e = memchr(s->buffer, '\n', n);
1244                                         if (e)
1245                                                 *e = 0;
1246                                         else
1247                                                 s->buffer[n] = 0;
1248
1249                                         server_process_syslog_message(s, strstrip(s->buffer), ucred, tv, label, label_len);
1250                                 } else if (n_fds > 0)
1251                                         log_warning("Got file descriptors via syslog socket. Ignoring.");
1252
1253                         } else {
1254                                 if (n > 0 && n_fds == 0)
1255                                         server_process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
1256                                 else if (n == 0 && n_fds == 1)
1257                                         server_process_native_file(s, fds[0], ucred, tv, label, label_len);
1258                                 else if (n_fds > 0)
1259                                         log_warning("Got too many file descriptors via native socket. Ignoring.");
1260                         }
1261
1262                         close_many(fds, n_fds);
1263                 }
1264
1265                 return 1;
1266
1267         } else if (ev->data.fd == s->stdout_fd) {
1268
1269                 if (ev->events != EPOLLIN) {
1270                         log_error("Got invalid event from epoll.");
1271                         return -EIO;
1272                 }
1273
1274                 stdout_stream_new(s);
1275                 return 1;
1276
1277         } else {
1278                 StdoutStream *stream;
1279
1280                 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1281                         log_error("Got invalid event from epoll.");
1282                         return -EIO;
1283                 }
1284
1285                 /* If it is none of the well-known fds, it must be an
1286                  * stdout stream fd. Note that this is a bit ugly here
1287                  * (since we rely that none of the well-known fds
1288                  * could be interpreted as pointer), but nonetheless
1289                  * safe, since the well-known fds would never get an
1290                  * fd > 4096, i.e. beyond the first memory page */
1291
1292                 stream = ev->data.ptr;
1293
1294                 if (stdout_stream_process(stream) <= 0)
1295                         stdout_stream_free(stream);
1296
1297                 return 1;
1298         }
1299
1300         log_error("Unknown event.");
1301         return 0;
1302 }
1303
1304 static int open_signalfd(Server *s) {
1305         sigset_t mask;
1306         struct epoll_event ev;
1307
1308         assert(s);
1309
1310         assert_se(sigemptyset(&mask) == 0);
1311         sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1);
1312         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1313
1314         s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1315         if (s->signal_fd < 0) {
1316                 log_error("signalfd(): %m");
1317                 return -errno;
1318         }
1319
1320         zero(ev);
1321         ev.events = EPOLLIN;
1322         ev.data.fd = s->signal_fd;
1323
1324         if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1325                 log_error("epoll_ctl(): %m");
1326                 return -errno;
1327         }
1328
1329         return 0;
1330 }
1331
1332 static int server_parse_proc_cmdline(Server *s) {
1333         _cleanup_free_ char *line = NULL;
1334         char *w, *state;
1335         int r;
1336         size_t l;
1337
1338         if (detect_container(NULL) > 0)
1339                 return 0;
1340
1341         r = read_one_line_file("/proc/cmdline", &line);
1342         if (r < 0) {
1343                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
1344                 return 0;
1345         }
1346
1347         FOREACH_WORD_QUOTED(w, l, line, state) {
1348                 _cleanup_free_ char *word;
1349
1350                 word = strndup(w, l);
1351                 if (!word)
1352                         return -ENOMEM;
1353
1354                 if (startswith(word, "systemd.journald.forward_to_syslog=")) {
1355                         r = parse_boolean(word + 35);
1356                         if (r < 0)
1357                                 log_warning("Failed to parse forward to syslog switch %s. Ignoring.", word + 35);
1358                         else
1359                                 s->forward_to_syslog = r;
1360                 } else if (startswith(word, "systemd.journald.forward_to_kmsg=")) {
1361                         r = parse_boolean(word + 33);
1362                         if (r < 0)
1363                                 log_warning("Failed to parse forward to kmsg switch %s. Ignoring.", word + 33);
1364                         else
1365                                 s->forward_to_kmsg = r;
1366                 } else if (startswith(word, "systemd.journald.forward_to_console=")) {
1367                         r = parse_boolean(word + 36);
1368                         if (r < 0)
1369                                 log_warning("Failed to parse forward to console switch %s. Ignoring.", word + 36);
1370                         else
1371                                 s->forward_to_console = r;
1372                 } else if (startswith(word, "systemd.journald"))
1373                         log_warning("Invalid systemd.journald parameter. Ignoring.");
1374         }
1375
1376         return 0;
1377 }
1378
1379 static int server_parse_config_file(Server *s) {
1380         static const char fn[] = "/etc/systemd/journald.conf";
1381         _cleanup_fclose_ FILE *f = NULL;
1382         int r;
1383
1384         assert(s);
1385
1386         f = fopen(fn, "re");
1387         if (!f) {
1388                 if (errno == ENOENT)
1389                         return 0;
1390
1391                 log_warning("Failed to open configuration file %s: %m", fn);
1392                 return -errno;
1393         }
1394
1395         r = config_parse(NULL, fn, f, "Journal\0", config_item_perf_lookup,
1396                          (void*) journald_gperf_lookup, false, false, s);
1397         if (r < 0)
1398                 log_warning("Failed to parse configuration file: %s", strerror(-r));
1399
1400         return r;
1401 }
1402
1403 static int server_open_sync_timer(Server *s) {
1404         int r;
1405         struct epoll_event ev;
1406
1407         assert(s);
1408
1409         s->sync_timer_fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
1410         if (s->sync_timer_fd < 0)
1411                 return -errno;
1412
1413         zero(ev);
1414         ev.events = EPOLLIN;
1415         ev.data.fd = s->sync_timer_fd;
1416
1417         r = epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->sync_timer_fd, &ev);
1418         if (r < 0) {
1419                 log_error("Failed to add idle timer fd to epoll object: %m");
1420                 return -errno;
1421         }
1422
1423         return 0;
1424 }
1425
1426 int server_schedule_sync(Server *s) {
1427         int r;
1428
1429         assert(s);
1430
1431         if (s->sync_scheduled)
1432                 return 0;
1433
1434         if (s->sync_interval_usec) {
1435                 struct itimerspec sync_timer_enable = {};
1436
1437                 timespec_store(&sync_timer_enable.it_value, s->sync_interval_usec);
1438
1439                 r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_enable, NULL);
1440                 if (r < 0)
1441                         return -errno;
1442         }
1443
1444         s->sync_scheduled = true;
1445
1446         return 0;
1447 }
1448
1449 int server_init(Server *s) {
1450         int n, r, fd;
1451
1452         assert(s);
1453
1454         zero(*s);
1455         s->sync_timer_fd = s->syslog_fd = s->native_fd = s->stdout_fd =
1456                 s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
1457         s->compress = true;
1458         s->seal = true;
1459
1460         s->sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC;
1461         s->sync_scheduled = false;
1462
1463         s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
1464         s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
1465
1466         s->forward_to_syslog = true;
1467
1468         s->max_level_store = LOG_DEBUG;
1469         s->max_level_syslog = LOG_DEBUG;
1470         s->max_level_kmsg = LOG_NOTICE;
1471         s->max_level_console = LOG_INFO;
1472
1473         memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
1474         memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
1475
1476         server_parse_config_file(s);
1477         server_parse_proc_cmdline(s);
1478         if (!!s->rate_limit_interval ^ !!s->rate_limit_burst) {
1479                 log_debug("Setting both rate limit interval and burst from %llu,%u to 0,0",
1480                           (long long unsigned) s->rate_limit_interval,
1481                           s->rate_limit_burst);
1482                 s->rate_limit_interval = s->rate_limit_burst = 0;
1483         }
1484
1485         mkdir_p("/run/systemd/journal", 0755);
1486
1487         s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1488         if (!s->user_journals)
1489                 return log_oom();
1490
1491         s->mmap = mmap_cache_new();
1492         if (!s->mmap)
1493                 return log_oom();
1494
1495         s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1496         if (s->epoll_fd < 0) {
1497                 log_error("Failed to create epoll object: %m");
1498                 return -errno;
1499         }
1500
1501         n = sd_listen_fds(true);
1502         if (n < 0) {
1503                 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1504                 return n;
1505         }
1506
1507         for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1508
1509                 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/journal/socket", 0) > 0) {
1510
1511                         if (s->native_fd >= 0) {
1512                                 log_error("Too many native sockets passed.");
1513                                 return -EINVAL;
1514                         }
1515
1516                         s->native_fd = fd;
1517
1518                 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/journal/stdout", 0) > 0) {
1519
1520                         if (s->stdout_fd >= 0) {
1521                                 log_error("Too many stdout sockets passed.");
1522                                 return -EINVAL;
1523                         }
1524
1525                         s->stdout_fd = fd;
1526
1527                 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1528
1529                         if (s->syslog_fd >= 0) {
1530                                 log_error("Too many /dev/log sockets passed.");
1531                                 return -EINVAL;
1532                         }
1533
1534                         s->syslog_fd = fd;
1535
1536                 } else {
1537                         log_error("Unknown socket passed.");
1538                         return -EINVAL;
1539                 }
1540         }
1541
1542         r = server_open_syslog_socket(s);
1543         if (r < 0)
1544                 return r;
1545
1546         r = server_open_native_socket(s);
1547         if (r < 0)
1548                 return r;
1549
1550         r = server_open_stdout_socket(s);
1551         if (r < 0)
1552                 return r;
1553
1554         r = server_open_dev_kmsg(s);
1555         if (r < 0)
1556                 return r;
1557
1558         r = server_open_kernel_seqnum(s);
1559         if (r < 0)
1560                 return r;
1561
1562         r = server_open_sync_timer(s);
1563         if (r < 0)
1564                 return r;
1565
1566         r = open_signalfd(s);
1567         if (r < 0)
1568                 return r;
1569
1570         s->udev = udev_new();
1571         if (!s->udev)
1572                 return -ENOMEM;
1573
1574         s->rate_limit = journal_rate_limit_new(s->rate_limit_interval,
1575                                                s->rate_limit_burst);
1576         if (!s->rate_limit)
1577                 return -ENOMEM;
1578
1579         r = system_journal_open(s);
1580         if (r < 0)
1581                 return r;
1582
1583         return 0;
1584 }
1585
1586 void server_maybe_append_tags(Server *s) {
1587 #ifdef HAVE_GCRYPT
1588         JournalFile *f;
1589         Iterator i;
1590         usec_t n;
1591
1592         n = now(CLOCK_REALTIME);
1593
1594         if (s->system_journal)
1595                 journal_file_maybe_append_tag(s->system_journal, n);
1596
1597         HASHMAP_FOREACH(f, s->user_journals, i)
1598                 journal_file_maybe_append_tag(f, n);
1599 #endif
1600 }
1601
1602 void server_done(Server *s) {
1603         JournalFile *f;
1604         assert(s);
1605
1606         while (s->stdout_streams)
1607                 stdout_stream_free(s->stdout_streams);
1608
1609         if (s->system_journal)
1610                 journal_file_close(s->system_journal);
1611
1612         if (s->runtime_journal)
1613                 journal_file_close(s->runtime_journal);
1614
1615         while ((f = hashmap_steal_first(s->user_journals)))
1616                 journal_file_close(f);
1617
1618         hashmap_free(s->user_journals);
1619
1620         if (s->epoll_fd >= 0)
1621                 close_nointr_nofail(s->epoll_fd);
1622
1623         if (s->signal_fd >= 0)
1624                 close_nointr_nofail(s->signal_fd);
1625
1626         if (s->syslog_fd >= 0)
1627                 close_nointr_nofail(s->syslog_fd);
1628
1629         if (s->native_fd >= 0)
1630                 close_nointr_nofail(s->native_fd);
1631
1632         if (s->stdout_fd >= 0)
1633                 close_nointr_nofail(s->stdout_fd);
1634
1635         if (s->dev_kmsg_fd >= 0)
1636                 close_nointr_nofail(s->dev_kmsg_fd);
1637
1638         if (s->sync_timer_fd >= 0)
1639                 close_nointr_nofail(s->sync_timer_fd);
1640
1641         if (s->rate_limit)
1642                 journal_rate_limit_free(s->rate_limit);
1643
1644         if (s->kernel_seqnum)
1645                 munmap(s->kernel_seqnum, sizeof(uint64_t));
1646
1647         free(s->buffer);
1648         free(s->tty_path);
1649
1650         if (s->mmap)
1651                 mmap_cache_unref(s->mmap);
1652
1653         if (s->udev)
1654                 udev_unref(s->udev);
1655 }