lib/cache/lvmetad.c

   1 #include "lib.h"
   2 #include "toolcontext.h"
   3 #include "metadata.h"
   4 #include "device.h"
   5 #include "lvmetad.h"
   6 #include "lvmcache.h"
   7 #include "lvmetad-client.h"
   8 #include "format-text.h" // TODO for disk_locn, used as a DA representation
   9 #include "filter.h"
  10
  11 static int _using_lvmetad = 0;
  12 static daemon_handle _lvmetad;
  13
  14 void lvmetad_init(void)
  15 {
  16         const char *socket = getenv("LVM_LVMETAD_SOCKET");
  17         if (_using_lvmetad) { /* configured by the toolcontext */
  18                 _lvmetad = lvmetad_open(socket ?: DEFAULT_RUN_DIR "/lvmetad.socket");
  19                 if (_lvmetad.socket_fd < 0) {
  20                         log_warn("Failed to connect to lvmetad. Falling back to scanning.");
  21                         _using_lvmetad = 0;
  22                 }
  23         }
  24 }
  25
  26 /*
  27  * Helper; evaluate the reply from lvmetad, check for errors, print diagnostics
  28  * and return a summary success/failure exit code. Frees up the reply resources
  29  * as well.
  30  */
  31 static int _lvmetad_handle_reply(daemon_reply reply, const char *action, const char *object) {
  32         if (reply.error || strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
  33                 log_error("Request to %s %s in lvmetad has failed. Reason: %s",
  34                           action, object, reply.error ? strerror(reply.error) :
  35                           daemon_reply_str(reply, "reason", "Unknown."));
  36                 daemon_reply_destroy(reply);
  37                 return 0;
  38         }
  39
  40         daemon_reply_destroy(reply);
  41         return 1;
  42 }
  43
  44 static int _read_mda(struct lvmcache_info *info,
  45                      struct format_type *fmt,
  46                      const struct dm_config_node *cn)
  47 {
  48         struct metadata_area_ops *ops;
  49
  50         dm_list_iterate_items(ops, &fmt->mda_ops)
  51                 if (ops->mda_import_text && ops->mda_import_text(info, cn))
  52                         return 1;
  53
  54         return 0;
  55 }
  56
  57 static struct lvmcache_info *_pv_populate_lvmcache(
  58         struct cmd_context *cmd, struct dm_config_node *cn, dev_t fallback)
  59 {
  60         struct device *device;
  61         struct id pvid, vgid;
  62         char mda_id[32];
  63         char da_id[32];
  64         int i = 0;
  65         struct dm_config_node *mda = NULL;
  66         struct dm_config_node *da = NULL;
  67         uint64_t offset, size;
  68         struct lvmcache_info *info;
  69         const char *pvid_txt = dm_config_find_str(cn->child, "id", NULL),
  70                    *vgid_txt = dm_config_find_str(cn->child, "vgid", NULL),
  71                    *vgname = dm_config_find_str(cn->child, "vgname", NULL),
  72                    *fmt_name = dm_config_find_str(cn->child, "format", NULL);
  73         dev_t devt = dm_config_find_int(cn->child, "device", 0);
  74         uint64_t devsize = dm_config_find_int(cn->child, "dev_size", 0),
  75                  label_sector = dm_config_find_int(cn->child, "label_sector", 0);
  76
  77         struct format_type *fmt = fmt_name ? get_format_by_name(cmd, fmt_name) : NULL;
  78
  79         if (!fmt) {
  80                 log_warn("No format for PV %s. It is probably missing.", pvid_txt);
  81                 return NULL;
  82         }
  83
  84         device = dev_cache_get_by_devt(devt, cmd->filter);
  85         if (!device && fallback)
  86                 device = dev_cache_get_by_devt(fallback, cmd->filter);
  87
  88         if (!device) {
  89                 log_warn("No device for PV %s.", pvid_txt);
  90                 return NULL;
  91         }
  92
  93         if (!pvid_txt || !id_read_format(&pvid, pvid_txt)) {
  94                 log_warn("Missing or ill-formatted PVID for PV: %s.", pvid_txt);
  95                 return NULL;
  96         }
  97
  98         if (vgid_txt)
  99                 id_read_format(&vgid, vgid_txt);
 100         else
 101                 strcpy((char*)&vgid, fmt->orphan_vg_name);
 102
 103         if (!vgname)
 104                 vgname = fmt->orphan_vg_name;
 105
 106         info = lvmcache_add(fmt->labeller, (const char *)&pvid, device,
 107                             vgname, (const char *)&vgid, 0);
 108
 109         lvmcache_get_label(info)->sector = label_sector;
 110         lvmcache_set_device_size(info, devsize);
 111         lvmcache_del_das(info);
 112         lvmcache_del_mdas(info);
 113
 114         do {
 115                 sprintf(mda_id, "mda%d", i);
 116                 mda = dm_config_find_node(cn->child, mda_id);
 117                 if (mda)
 118                         _read_mda(info, fmt, mda);
 119                 ++i;
 120         } while (mda);
 121
 122         i = 0;
 123         do {
 124                 sprintf(da_id, "da%d", i);
 125                 da = dm_config_find_node(cn->child, da_id);
 126                 if (da) {
 127                         if (!dm_config_get_uint64(da->child, "offset", &offset)) return_0;
 128                         if (!dm_config_get_uint64(da->child, "size", &size)) return_0;
 129                         lvmcache_add_da(info, offset, size);
 130                 }
 131                 ++i;
 132         } while (da);
 133
 134         return info;
 135 }
 136
 137 struct volume_group *lvmetad_vg_lookup(struct cmd_context *cmd, const char *vgname, const char *vgid)
 138 {
 139         struct volume_group *vg = NULL;
 140         daemon_reply reply;
 141         char uuid[64];
 142         struct format_instance *fid;
 143         struct format_instance_ctx fic;
 144         struct dm_config_node *top;
 145         const char *name;
 146         const char *fmt_name;
 147         struct format_type *fmt;
 148         struct dm_config_node *pvcn;
 149         struct pv_list *pvl;
 150         struct lvmcache_info *info;
 151
 152         if (!_using_lvmetad)
 153                 return NULL;
 154
 155         if (vgid) {
 156                 id_write_format((const struct id*)vgid, uuid, 64);
 157                 reply = daemon_send_simple(_lvmetad, "vg_lookup", "uuid = %s", uuid, NULL);
 158         } else {
 159                 if (!vgname)
 160                         log_error(INTERNAL_ERROR "VG name required (VGID not available)");
 161                 reply = daemon_send_simple(_lvmetad, "vg_lookup", "name = %s", vgname, NULL);
 162         }
 163
 164         if (!strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
 165
 166                 top = dm_config_find_node(reply.cft->root, "metadata");
 167                 name = daemon_reply_str(reply, "name", NULL);
 168
 169                 /* fall back to lvm2 if we don't know better */
 170                 fmt_name = dm_config_find_str(top, "metadata/format", "lvm2");
 171                 if (!(fmt = get_format_by_name(cmd, fmt_name))) {
 172                         log_error(INTERNAL_ERROR
 173                                   "We do not know the format (%s) reported by lvmetad.",
 174                                   fmt_name);
 175                         return NULL;
 176                 }
 177
 178                 fic.type = FMT_INSTANCE_MDAS | FMT_INSTANCE_AUX_MDAS;
 179                 fic.context.vg_ref.vg_name = name;
 180                 fic.context.vg_ref.vg_id = vgid;
 181
 182                 if (!(fid = fmt->ops->create_instance(fmt, &fic)))
 183                         return_NULL;
 184
 185                 pvcn = dm_config_find_node(top, "metadata/physical_volumes")->child;
 186                 while (pvcn) {
 187                         _pv_populate_lvmcache(cmd, pvcn, 0);
 188                         pvcn = pvcn->sib;
 189                 }
 190
 191                 top->key = name;
 192                 vg = import_vg_from_config_tree(reply.cft, fid);
 193
 194                 dm_list_iterate_items(pvl, &vg->pvs) {
 195                         if ((info = lvmcache_info_from_pvid((const char *)&pvl->pv->id, 0))) {
 196                                 pvl->pv->label_sector = lvmcache_get_label(info)->sector;
 197                                 pvl->pv->dev = lvmcache_device(info);
 198                                 lvmcache_fid_add_mdas_pv(info, fid);
 199                         } /* else probably missing */
 200                 }
 201
 202                 lvmcache_update_vg(vg, 0);
 203         }
 204
 205         daemon_reply_destroy(reply);
 206         return vg;
 207 }
 208
 209 struct _fixup_baton {
 210         int i;
 211         int find;
 212         int ignore;
 213 };
 214
 215 static int _fixup_ignored(struct metadata_area *mda, void *baton) {
 216         struct _fixup_baton *b = baton;
 217         if (b->i == b->find)
 218                 mda_set_ignored(mda, b->ignore);
 219         b->i ++;
 220         return 1;
 221 }
 222
 223 int lvmetad_vg_update(struct volume_group *vg)
 224 {
 225         char *buf = NULL;
 226         daemon_reply reply;
 227         struct dm_hash_node *n;
 228         struct metadata_area *mda;
 229         char mda_id[128], *num;
 230         struct pv_list *pvl;
 231         struct lvmcache_info *info;
 232         struct _fixup_baton baton;
 233
 234         if (!vg)
 235                 return 0;
 236
 237         if (!_using_lvmetad)
 238                 return 1; /* fake it */
 239
 240         /* TODO. This is not entirely correct, since export_vg_to_buffer
 241          * adds trailing nodes to the buffer. We may need to use
 242          * export_vg_to_config_tree and format the buffer ourselves. It
 243          * does, however, work for now, since the garbage is well
 244          * formatted and has no conflicting keys with the rest of the
 245          * request.  */
 246         if (!export_vg_to_buffer(vg, &buf)) {
 247                 log_error("Could not format VG metadata.");
 248                 return 0;
 249         }
 250
 251         reply = daemon_send_simple(_lvmetad, "vg_update", "vgname = %s", vg->name,
 252                                              "metadata = %b", strchr(buf, '{'),
 253                                    NULL);
 254
 255         if (!_lvmetad_handle_reply(reply, "update VG", vg->name))
 256                 return 0;
 257
 258         n = (vg->fid && vg->fid->metadata_areas_index) ?
 259                 dm_hash_get_first(vg->fid->metadata_areas_index) : NULL;
 260         while (n) {
 261                 mda = dm_hash_get_data(vg->fid->metadata_areas_index, n);
 262                 strcpy(mda_id, dm_hash_get_key(vg->fid->metadata_areas_index, n));
 263                 if ((num = strchr(mda_id, '_'))) {
 264                         *num = 0;
 265                         ++num;
 266                         if ((info = lvmcache_info_from_pvid(mda_id, 0))) {
 267                                 memset(&baton, 0, sizeof(baton));
 268                                 baton.find = atoi(num);
 269                                 baton.ignore = mda_is_ignored(mda);
 270                                 lvmcache_foreach_mda(info, _fixup_ignored, &baton);
 271                         }
 272                 }
 273                 n = dm_hash_get_next(vg->fid->metadata_areas_index, n);
 274         }
 275
 276         dm_list_iterate_items(pvl, &vg->pvs) {
 277                 /* NB. the PV fmt pointer is sometimes wrong during vgconvert */
 278                 if (pvl->pv->dev && !lvmetad_pv_found(pvl->pv->id, pvl->pv->dev,
 279                                                       vg->fid ? vg->fid->fmt : pvl->pv->fmt,
 280                                                       pvl->pv->label_sector, NULL))
 281                         return 0;
 282         }
 283
 284         return 1;
 285 }
 286
 287 int lvmetad_vg_remove(struct volume_group *vg)
 288 {
 289         char uuid[64];
 290         daemon_reply reply;
 291
 292         if (!_using_lvmetad)
 293                 return 1; /* just fake it */
 294
 295         id_write_format(&vg->id, uuid, 64);
 296         reply = daemon_send_simple(_lvmetad, "vg_remove", "uuid = %s", uuid, NULL);
 297
 298         return _lvmetad_handle_reply(reply, "remove VG", vg->name);
 299 }
 300
 301 int lvmetad_pv_lookup(struct cmd_context *cmd, struct id pvid)
 302 {
 303         char uuid[64];
 304         daemon_reply reply;
 305         int result = 1;
 306         struct dm_config_node *cn;
 307
 308         if (!_using_lvmetad)
 309                 return_0;
 310
 311         id_write_format(&pvid, uuid, 64);
 312
 313         reply = daemon_send_simple(_lvmetad, "pv_lookup", "uuid = %s", uuid, NULL);
 314
 315         if (reply.error || strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
 316                 _lvmetad_handle_reply(reply, "lookup PVs", "");
 317                 return_0;
 318         }
 319
 320         cn = dm_config_find_node(reply.cft->root, "physical_volume");
 321         if (!_pv_populate_lvmcache(cmd, cn, 0))
 322                 result = 0;
 323
 324         daemon_reply_destroy(reply);
 325         return result;
 326 }
 327
 328 int lvmetad_pv_lookup_by_devt(struct cmd_context *cmd, dev_t device)
 329 {
 330         int result = 1;
 331         daemon_reply reply;
 332         struct dm_config_node *cn;
 333
 334         if (!_using_lvmetad)
 335                 return_0;
 336
 337         reply = daemon_send_simple(_lvmetad, "pv_lookup", "device = %d", device, NULL);
 338
 339         if (reply.error || strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
 340                 _lvmetad_handle_reply(reply, "lookup PVs", "");
 341                 return_0;
 342         }
 343
 344         cn = dm_config_find_node(reply.cft->root, "physical_volume");
 345         if (!_pv_populate_lvmcache(cmd, cn, device))
 346                 result = 0;
 347
 348         daemon_reply_destroy(reply);
 349         return result;
 350 }
 351
 352 int lvmetad_pv_list_to_lvmcache(struct cmd_context *cmd)
 353 {
 354         daemon_reply reply;
 355         struct dm_config_node *cn;
 356
 357         if (!_using_lvmetad)
 358                 return_0;
 359
 360         reply = daemon_send_simple(_lvmetad, "pv_list", NULL);
 361
 362         if (reply.error || strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
 363                 _lvmetad_handle_reply(reply, "list PVs", "");
 364                 return_0;
 365         }
 366
 367         cn = dm_config_find_node(reply.cft->root, "physical_volumes")->child;
 368         while (cn) {
 369                 _pv_populate_lvmcache(cmd, cn, 0);
 370                 cn = cn->sib;
 371         }
 372
 373         daemon_reply_destroy(reply);
 374         return 1;
 375 }
 376
 377 int lvmetad_vg_list_to_lvmcache(struct cmd_context *cmd)
 378 {
 379         struct volume_group *tmp;
 380         struct id vgid;
 381         const char *vgid_txt;
 382         daemon_reply reply;
 383         struct dm_config_node *cn;
 384
 385         if (!_using_lvmetad)
 386                 return_0;
 387
 388         reply = daemon_send_simple(_lvmetad, "vg_list", NULL);
 389         if (reply.error || strcmp(daemon_reply_str(reply, "response", ""), "OK")) {
 390                 _lvmetad_handle_reply(reply, "list VGs", "");
 391                 return_0;
 392         }
 393
 394         cn = dm_config_find_node(reply.cft->root, "volume_groups")->child;
 395         while (cn) {
 396                 vgid_txt = cn->key;
 397                 id_read_format(&vgid, vgid_txt);
 398                 cn = cn->sib;
 399
 400                 /* the call to lvmetad_vg_lookup will poke the VG into lvmcache */
 401                 tmp = lvmetad_vg_lookup(cmd, NULL, (const char*)&vgid);
 402                 release_vg(tmp);
 403         }
 404
 405         daemon_reply_destroy(reply);
 406         return 1;
 407 }
 408
 409 struct _print_mda_baton {
 410         int i;
 411         char *buffer;
 412 };
 413
 414 static int _print_mda(struct metadata_area *mda, void *baton)
 415 {
 416         int result = 0;
 417         struct _print_mda_baton *b = baton;
 418         char *buf, *mda_txt;
 419
 420         if (!mda->ops->mda_export_text) /* do nothing */
 421                 return 1;
 422
 423         buf = b->buffer;
 424         mda_txt = mda->ops->mda_export_text(mda);
 425         if (!dm_asprintf(&b->buffer, "%s mda%i { %s }", b->buffer ?: "", b->i, mda_txt))
 426                 goto_out;
 427         b->i ++;
 428         result = 1;
 429 out:
 430         dm_free(mda_txt);
 431         dm_free(buf);
 432         return result;
 433 }
 434
 435 static int _print_da(struct disk_locn *da, void *baton)
 436 {
 437         struct _print_mda_baton *b;
 438         char *buf;
 439
 440         if (!da)
 441                 return 1;
 442
 443         b = baton;
 444         buf = b->buffer;
 445         if (!dm_asprintf(&b->buffer, "%s da%i { offset = %" PRIu64
 446                          " size = %" PRIu64 " }",
 447                          b->buffer ?: "", b->i, da->offset, da->size))
 448         {
 449                 dm_free(buf);
 450                 return_0;
 451         }
 452         b->i ++;
 453         dm_free(buf);
 454
 455         return 1;
 456 }
 457
 458 static const char *_print_mdas(struct lvmcache_info *info)
 459 {
 460         struct _print_mda_baton baton = { .i = 0, .buffer = NULL };
 461
 462         if (!lvmcache_foreach_mda(info, &_print_mda, &baton))
 463                 return NULL;
 464         baton.i = 0;
 465         if (!lvmcache_foreach_da(info, &_print_da, &baton))
 466                 return NULL;
 467
 468         return baton.buffer;
 469 }
 470
 471 int lvmetad_pv_found(struct id pvid, struct device *device, const struct format_type *fmt,
 472                      uint64_t label_sector, struct volume_group *vg)
 473 {
 474         char uuid[64];
 475         daemon_reply reply;
 476         struct lvmcache_info *info;
 477         const char *mdas = NULL;
 478         char *pvmeta;
 479         char *buf = NULL;
 480
 481         if (!_using_lvmetad)
 482                 return 1;
 483
 484         id_write_format(&pvid, uuid, 64);
 485
 486         /* FIXME A more direct route would be much preferable. */
 487         if ((info = lvmcache_info_from_pvid((const char *)&pvid, 0)))
 488                 mdas = _print_mdas(info);
 489
 490         if (!dm_asprintf(&pvmeta,
 491                          "{ device = %" PRIu64 "\n"
 492                          "  dev_size = %" PRIu64 "\n"
 493                          "  format = \"%s\"\n"
 494                          "  label_sector = %" PRIu64 "\n"
 495                          "  id = \"%s\"\n"
 496                          "  %s"
 497                          "}", device->dev,
 498                          info ? lvmcache_device_size(info) : 0,
 499                          fmt->name, label_sector, uuid, mdas ?: ""))
 500                 return_0;
 501
 502         if (vg) {
 503                 /*
 504                  * TODO. This is not entirely correct, since export_vg_to_buffer
 505                  * adds trailing garbage to the buffer. We may need to use
 506                  * export_vg_to_config_tree and format the buffer ourselves. It
 507                  * does, however, work for now, since the garbage is well
 508                  * formatted and has no conflicting keys with the rest of the
 509                  * request.
 510                  */
 511                 export_vg_to_buffer(vg, &buf);
 512                 reply = daemon_send_simple(_lvmetad,
 513                                            "pv_found",
 514                                            "pvmeta = %b", pvmeta,
 515                                            "vgname = %s", vg->name,
 516                                            "metadata = %b", strchr(buf, '{'),
 517                                            NULL);
 518         } else {
 519                 /* There are no MDAs on this PV. */
 520                 reply = daemon_send_simple(_lvmetad,
 521                                            "pv_found",
 522                                            "pvmeta = %b", pvmeta,
 523                                            NULL);
 524         }
 525
 526         dm_free(pvmeta);
 527         return _lvmetad_handle_reply(reply, "update PV", uuid);
 528 }
 529
 530 int lvmetad_pv_gone(dev_t device)
 531 {
 532         daemon_reply reply =
 533                 daemon_send_simple(_lvmetad, "pv_gone", "device = %d", device, NULL);
 534
 535         return _lvmetad_handle_reply(reply, "drop PV", "");
 536 }
 537
 538 int lvmetad_active(void)
 539 {
 540         return _using_lvmetad;
 541 }
 542
 543 void lvmetad_set_active(int active)
 544 {
 545         _using_lvmetad = active;
 546 }
 547
 548 /*
 549  * The following code implements pvscan --lvmetad.
 550  */
 551
 552 struct _pvscan_lvmetad_baton {
 553         struct volume_group *vg;
 554         struct format_instance *fid;
 555 };
 556
 557 static int _pvscan_lvmetad_single(struct metadata_area *mda, void *baton)
 558 {
 559         struct _pvscan_lvmetad_baton *b = baton;
 560         struct volume_group *this = mda->ops->vg_read(b->fid, "", mda);
 561         if ((this && !b->vg) || this->seqno > b->vg->seqno)
 562                 b->vg = this;
 563         else release_vg(this);
 564         return 1;
 565 }
 566
 567 static dev_t _parse_devt(const char *str) { /* Oh. */
 568         char *where = (char *) str;
 569         int major = strtol(str, &where, 10);
 570         int minor;
 571
 572         if (where == str)
 573                 return -1;
 574
 575         if (*where != ':')
 576                 return -1;
 577
 578         str = ++where;
 579         minor = strtol(str, &where, 10);
 580
 581         if (where == str)
 582                 return -1;
 583
 584         if (*where)
 585                 return -1;
 586
 587         return MKDEV(major, minor);
 588 }
 589
 590 int pvscan_lvmetad(struct cmd_context *cmd, int argc, char **argv)
 591 {
 592         struct device *dev;
 593         struct label *label;
 594         struct lvmcache_info *info;
 595         struct physical_volume pv;
 596         struct _pvscan_lvmetad_baton baton;
 597         /* Create a dummy instance. */
 598         struct format_instance_ctx fic = { .type = 0 };
 599
 600         if (argc != 1) {
 601                 log_error("Exactly one device parameter required.");
 602                 return 0;
 603         }
 604
 605         if (!lvmetad_active()) {
 606                 log_error("Cannot proceed since lvmetad is not active.");
 607                 return 0;
 608         }
 609
 610         dev = dev_cache_get(argv[0], NULL);
 611         if (!dev && _parse_devt(argv[0]) != -1)
 612                 dev = dev_cache_get_by_devt(_parse_devt(argv[0]), NULL);
 613
 614         if (!dev) {
 615                 if (_parse_devt(argv[0]) == -1) {
 616                         log_error("For devices that do not exist, we need a MAJOR:MINOR pair.");
 617                         return 0;
 618                 }
 619
 620                 if (!lvmetad_pv_gone(_parse_devt(argv[0])))
 621                         goto fatal;
 622
 623                 log_info("Device %s not found and was wiped from lvmetad.", argv[0]);
 624                 return 1;
 625         }
 626
 627         if (!label_read(dev, &label, 0)) {
 628                 log_warn("No PV label found on %s.", dev_name(dev));
 629                 if (!lvmetad_pv_gone(dev->dev))
 630                         goto fatal;
 631                 return 1;
 632         }
 633
 634         info = (struct lvmcache_info *) label->info;
 635         memset(&pv, 0, sizeof(pv));
 636
 637         baton.vg = NULL;
 638         baton.fid = lvmcache_fmt(info)->ops->create_instance(lvmcache_fmt(info),
 639                                                              &fic);
 640
 641         lvmcache_foreach_mda(info, _pvscan_lvmetad_single, &baton);
 642
 643         /*
 644          * NB. If this command failed and we are relying on lvmetad to have an
 645          * *exact* image of the system, the lvmetad instance that went out of
 646          * sync needs to be killed.
 647          */
 648         if (!lvmetad_pv_found(*(struct id *)dev->pvid, dev, lvmcache_fmt(info),
 649                               label->sector, baton.vg))
 650                 goto fatal;
 651
 652         release_vg(baton.vg);
 653         return 1;
 654 fatal:
 655         release_vg(baton.vg);
 656         /* FIXME kill lvmetad automatically if we can */
 657         log_error("Update of lvmetad failed. This is a serious problem.\n  "
 658                   "It is strongly recommended that you restart lvmetad immediately.");
 659         return 0;
 660 }
 661