src/shared/bus-wait-for-jobs.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include "alloc-util.h"
   4 #include "bus-wait-for-jobs.h"
   5 #include "set.h"
   6 #include "bus-util.h"
   7 #include "bus-internal.h"
   8 #include "unit-def.h"
   9 #include "escape.h"
  10 #include "strv.h"
  11
  12 typedef struct BusWaitForJobs {
  13         sd_bus *bus;
  14
  15         /* The set of jobs to wait for, as bus object paths */
  16         Set *jobs;
  17
  18         /* The unit name and job result of the last Job message */
  19         char *name;
  20         char *result;
  21
  22         sd_bus_slot *slot_job_removed;
  23         sd_bus_slot *slot_disconnected;
  24 } BusWaitForJobs;
  25
  26 static int match_disconnected(sd_bus_message *m, void *userdata, sd_bus_error *error) {
  27         assert(m);
  28
  29         log_error("Warning! D-Bus connection terminated.");
  30         sd_bus_close(sd_bus_message_get_bus(m));
  31
  32         return 0;
  33 }
  34
  35 static int match_job_removed(sd_bus_message *m, void *userdata, sd_bus_error *error) {
  36         const char *path, *unit, *result;
  37         BusWaitForJobs *d = userdata;
  38         uint32_t id;
  39         char *found;
  40         int r;
  41
  42         assert(m);
  43         assert(d);
  44
  45         r = sd_bus_message_read(m, "uoss", &id, &path, &unit, &result);
  46         if (r < 0) {
  47                 bus_log_parse_error(r);
  48                 return 0;
  49         }
  50
  51         found = set_remove(d->jobs, (char*) path);
  52         if (!found)
  53                 return 0;
  54
  55         free(found);
  56
  57         (void) free_and_strdup(&d->result, empty_to_null(result));
  58
  59         (void) free_and_strdup(&d->name, empty_to_null(unit));
  60
  61         return 0;
  62 }
  63
  64 BusWaitForJobs* bus_wait_for_jobs_free(BusWaitForJobs *d) {
  65         if (!d)
  66                 return NULL;
  67
  68         set_free(d->jobs);
  69
  70         sd_bus_slot_unref(d->slot_disconnected);
  71         sd_bus_slot_unref(d->slot_job_removed);
  72
  73         sd_bus_unref(d->bus);
  74
  75         free(d->name);
  76         free(d->result);
  77
  78         return mfree(d);
  79 }
  80
  81 int bus_wait_for_jobs_new(sd_bus *bus, BusWaitForJobs **ret) {
  82         _cleanup_(bus_wait_for_jobs_freep) BusWaitForJobs *d = NULL;
  83         int r;
  84
  85         assert(bus);
  86         assert(ret);
  87
  88         d = new(BusWaitForJobs, 1);
  89         if (!d)
  90                 return -ENOMEM;
  91
  92         *d = (BusWaitForJobs) {
  93                 .bus = sd_bus_ref(bus),
  94         };
  95
  96         /* When we are a bus client we match by sender. Direct
  97          * connections OTOH have no initialized sender field, and
  98          * hence we ignore the sender then */
  99         r = sd_bus_match_signal_async(
 100                         bus,
 101                         &d->slot_job_removed,
 102                         bus->bus_client ? "org.freedesktop.systemd1" : NULL,
 103                         "/org/freedesktop/systemd1",
 104                         "org.freedesktop.systemd1.Manager",
 105                         "JobRemoved",
 106                         match_job_removed, NULL, d);
 107         if (r < 0)
 108                 return r;
 109
 110         r = sd_bus_match_signal_async(
 111                         bus,
 112                         &d->slot_disconnected,
 113                         "org.freedesktop.DBus.Local",
 114                         NULL,
 115                         "org.freedesktop.DBus.Local",
 116                         "Disconnected",
 117                         match_disconnected, NULL, d);
 118         if (r < 0)
 119                 return r;
 120
 121         *ret = TAKE_PTR(d);
 122
 123         return 0;
 124 }
 125
 126 static int bus_process_wait(sd_bus *bus) {
 127         int r;
 128
 129         for (;;) {
 130                 r = sd_bus_process(bus, NULL);
 131                 if (r < 0)
 132                         return r;
 133                 if (r > 0)
 134                         return 0;
 135
 136                 r = sd_bus_wait(bus, UINT64_MAX);
 137                 if (r < 0)
 138                         return r;
 139         }
 140 }
 141
 142 static int bus_job_get_service_result(BusWaitForJobs *d, char **result) {
 143         _cleanup_free_ char *dbus_path = NULL;
 144
 145         assert(d);
 146         assert(d->name);
 147         assert(result);
 148
 149         if (!endswith(d->name, ".service"))
 150                 return -EINVAL;
 151
 152         dbus_path = unit_dbus_path_from_name(d->name);
 153         if (!dbus_path)
 154                 return -ENOMEM;
 155
 156         return sd_bus_get_property_string(d->bus,
 157                                           "org.freedesktop.systemd1",
 158                                           dbus_path,
 159                                           "org.freedesktop.systemd1.Service",
 160                                           "Result",
 161                                           NULL,
 162                                           result);
 163 }
 164
 165 static void log_job_error_with_service_result(const char* service, const char *result, const char* const* extra_args) {
 166         _cleanup_free_ char *service_shell_quoted = NULL;
 167         const char *systemctl = "systemctl", *journalctl = "journalctl";
 168
 169         static const struct {
 170                 const char *result, *explanation;
 171         } explanations[] = {
 172                 { "resources",   "of unavailable resources or another system error" },
 173                 { "protocol",    "the service did not take the steps required by its unit configuration" },
 174                 { "timeout",     "a timeout was exceeded" },
 175                 { "exit-code",   "the control process exited with error code" },
 176                 { "signal",      "a fatal signal was delivered to the control process" },
 177                 { "core-dump",   "a fatal signal was delivered causing the control process to dump core" },
 178                 { "watchdog",    "the service failed to send watchdog ping" },
 179                 { "start-limit", "start of the service was attempted too often" }
 180         };
 181
 182         assert(service);
 183
 184         service_shell_quoted = shell_maybe_quote(service, 0);
 185
 186         if (!strv_isempty((char**) extra_args)) {
 187                 _cleanup_free_ char *t = NULL;
 188
 189                 t = strv_join((char**) extra_args, " ");
 190                 systemctl = strjoina("systemctl ", t ? : "<args>");
 191                 journalctl = strjoina("journalctl ", t ? : "<args>");
 192         }
 193
 194         if (!isempty(result)) {
 195                 size_t i;
 196
 197                 for (i = 0; i < ELEMENTSOF(explanations); ++i)
 198                         if (streq(result, explanations[i].result))
 199                                 break;
 200
 201                 if (i < ELEMENTSOF(explanations)) {
 202                         log_error("Job for %s failed because %s.\n"
 203                                   "See \"%s status %s\" and \"%s -xeu %s\" for details.\n",
 204                                   service,
 205                                   explanations[i].explanation,
 206                                   systemctl,
 207                                   service_shell_quoted ?: "<service>",
 208                                   journalctl,
 209                                   service_shell_quoted ?: "<service>");
 210                         goto finish;
 211                 }
 212         }
 213
 214         log_error("Job for %s failed.\n"
 215                   "See \"%s status %s\" and \"%s -xeu %s\" for details.\n",
 216                   service,
 217                   systemctl,
 218                   service_shell_quoted ?: "<service>",
 219                   journalctl,
 220                   service_shell_quoted ?: "<service>");
 221
 222 finish:
 223         /* For some results maybe additional explanation is required */
 224         if (streq_ptr(result, "start-limit"))
 225                 log_info("To force a start use \"%1$s reset-failed %2$s\"\n"
 226                          "followed by \"%1$s start %2$s\" again.",
 227                          systemctl,
 228                          service_shell_quoted ?: "<service>");
 229 }
 230
 231 static int check_wait_response(BusWaitForJobs *d, bool quiet, const char* const* extra_args) {
 232         assert(d);
 233         assert(d->name);
 234         assert(d->result);
 235
 236         if (!quiet) {
 237                 if (streq(d->result, "canceled"))
 238                         log_error("Job for %s canceled.", strna(d->name));
 239                 else if (streq(d->result, "timeout"))
 240                         log_error("Job for %s timed out.", strna(d->name));
 241                 else if (streq(d->result, "dependency"))
 242                         log_error("A dependency job for %s failed. See 'journalctl -xe' for details.", strna(d->name));
 243                 else if (streq(d->result, "invalid"))
 244                         log_error("%s is not active, cannot reload.", strna(d->name));
 245                 else if (streq(d->result, "assert"))
 246                         log_error("Assertion failed on job for %s.", strna(d->name));
 247                 else if (streq(d->result, "unsupported"))
 248                         log_error("Operation on or unit type of %s not supported on this system.", strna(d->name));
 249                 else if (streq(d->result, "collected"))
 250                         log_error("Queued job for %s was garbage collected.", strna(d->name));
 251                 else if (streq(d->result, "once"))
 252                         log_error("Unit %s was started already once and can't be started again.", strna(d->name));
 253                 else if (!STR_IN_SET(d->result, "done", "skipped")) {
 254
 255                         if (d->name && endswith(d->name, ".service")) {
 256                                 _cleanup_free_ char *result = NULL;
 257                                 int q;
 258
 259                                 q = bus_job_get_service_result(d, &result);
 260                                 if (q < 0)
 261                                         log_debug_errno(q, "Failed to get Result property of unit %s: %m", d->name);
 262
 263                                 log_job_error_with_service_result(d->name, result, extra_args);
 264                         } else
 265                                 log_error("Job failed. See \"journalctl -xe\" for details.");
 266                 }
 267         }
 268
 269         if (STR_IN_SET(d->result, "canceled", "collected"))
 270                 return -ECANCELED;
 271         else if (streq(d->result, "timeout"))
 272                 return -ETIME;
 273         else if (streq(d->result, "dependency"))
 274                 return -EIO;
 275         else if (streq(d->result, "invalid"))
 276                 return -ENOEXEC;
 277         else if (streq(d->result, "assert"))
 278                 return -EPROTO;
 279         else if (streq(d->result, "unsupported"))
 280                 return -EOPNOTSUPP;
 281         else if (streq(d->result, "once"))
 282                 return -ESTALE;
 283         else if (STR_IN_SET(d->result, "done", "skipped"))
 284                 return 0;
 285
 286         return log_debug_errno(SYNTHETIC_ERRNO(EIO),
 287                                "Unexpected job result, assuming server side newer than us: %s", d->result);
 288 }
 289
 290 int bus_wait_for_jobs(BusWaitForJobs *d, bool quiet, const char* const* extra_args) {
 291         int r = 0;
 292
 293         assert(d);
 294
 295         while (!set_isempty(d->jobs)) {
 296                 int q;
 297
 298                 q = bus_process_wait(d->bus);
 299                 if (q < 0)
 300                         return log_error_errno(q, "Failed to wait for response: %m");
 301
 302                 if (d->name && d->result) {
 303                         q = check_wait_response(d, quiet, extra_args);
 304                         /* Return the first error as it is most likely to be
 305                          * meaningful. */
 306                         if (q < 0 && r == 0)
 307                                 r = q;
 308
 309                         log_full_errno_zerook(LOG_DEBUG, q,
 310                                               "Got result %s/%m for job %s", d->result, d->name);
 311                 }
 312
 313                 d->name = mfree(d->name);
 314                 d->result = mfree(d->result);
 315         }
 316
 317         return r;
 318 }
 319
 320 int bus_wait_for_jobs_add(BusWaitForJobs *d, const char *path) {
 321         assert(d);
 322
 323         return set_put_strdup(&d->jobs, path);
 324 }
 325
 326 int bus_wait_for_jobs_one(BusWaitForJobs *d, const char *path, bool quiet) {
 327         int r;
 328
 329         r = bus_wait_for_jobs_add(d, path);
 330         if (r < 0)
 331                 return log_oom();
 332
 333         return bus_wait_for_jobs(d, quiet, NULL);
 334 }