lustre/utils/lustre_lfsck.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2015, Intel Corporation.
  24  */
  25 /*
  26  * lustre/utils/lustre_lfsck.c
  27  *
  28  * Lustre user-space tools for LFSCK.
  29  *
  30  * Author: Fan Yong <yong.fan@whamcloud.com>
  31  */
  32
  33 #include <stdio.h>
  34 #include <unistd.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37 #include <errno.h>
  38 #include <getopt.h>
  39 #include <sys/ioctl.h>
  40 #include <time.h>
  41
  42 #include "obdctl.h"
  43
  44 #include <lustre/lustre_lfsck_user.h>
  45 #include <lnet/lnetctl.h>
  46 #include <lustre_ioctl.h>
  47 /* Needs to be last to avoid clashes */
  48 #include <libcfs/util/ioctl.h>
  49
  50 static struct option long_opt_start[] = {
  51         {"device",              required_argument, 0, 'M'},
  52         {"all",                 no_argument,       0, 'A'},
  53         {"create_ostobj",       optional_argument, 0, 'c'},
  54         {"create-ostobj",       optional_argument, 0, 'c'},
  55         {"create_mdtobj",       optional_argument, 0, 'C'},
  56         {"create-mdtobj",       optional_argument, 0, 'C'},
  57         {"error",               required_argument, 0, 'e'},
  58         {"help",                no_argument,       0, 'h'},
  59         {"dryrun",              optional_argument, 0, 'n'},
  60         {"orphan",              no_argument,       0, 'o'},
  61         {"reset",               no_argument,       0, 'r'},
  62         {"speed",               required_argument, 0, 's'},
  63         {"type",                required_argument, 0, 't'},
  64         {"window_size",         required_argument, 0, 'w'},
  65         {"window-size",         required_argument, 0, 'w'},
  66         {0,                     0,                 0,  0 }
  67 };
  68
  69 static struct option long_opt_stop[] = {
  70         {"device",      required_argument, 0, 'M'},
  71         {"all",         no_argument,       0, 'A'},
  72         {"help",        no_argument,       0, 'h'},
  73         {0,             0,                 0,  0 }
  74 };
  75
  76 static struct option long_opt_query[] = {
  77         {"device",      required_argument, 0, 'M'},
  78         {"type",        required_argument, 0, 't'},
  79         {"help",        no_argument,       0, 'h'},
  80         {"wait",        no_argument,       0, 'w'},
  81         {0,             0,                 0,  0 }
  82 };
  83
  84 struct lfsck_type_name {
  85         char            *ltn_name;
  86         enum lfsck_type  ltn_type;
  87 };
  88
  89 static struct lfsck_type_name lfsck_types_names[] = {
  90         { "scrub",      LFSCK_TYPE_SCRUB },
  91         { "layout",     LFSCK_TYPE_LAYOUT },
  92         { "namespace",  LFSCK_TYPE_NAMESPACE },
  93         { "default",    LFSCK_TYPES_DEF },
  94         { "all",        LFSCK_TYPES_SUPPORTED },
  95         { NULL,         0 }
  96 };
  97
  98 static enum lfsck_type lfsck_name2type(const char *name)
  99 {
 100         int i;
 101
 102         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 103                 if (strcmp(lfsck_types_names[i].ltn_name, name) == 0)
 104                         return lfsck_types_names[i].ltn_type;
 105         }
 106         return -1;
 107 }
 108
 109 static const char *lfsck_type2name(__u16 type)
 110 {
 111         int i;
 112
 113         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 114                 if (type == lfsck_types_names[i].ltn_type)
 115                         return lfsck_types_names[i].ltn_name;
 116         }
 117
 118         return NULL;
 119 }
 120
 121 static void usage_start(void)
 122 {
 123         fprintf(stderr, "start LFSCK\n"
 124                 "usage:\n"
 125                 "lfsck_start <-M | --device {MDT,OST}_device>\n"
 126                 "            [-A | --all] [-c | --create_ostobj [on | off]]\n"
 127                 "            [-C | --create_mdtobj [on | off]]\n"
 128                 "            [-e | --error {continue | abort}] [-h | --help]\n"
 129                 "            [-n | --dryrun [on | off]] [-o | --orphan]\n"
 130                 "            [-r | --reset] [-s | --speed ops_per_sec_limit]\n"
 131                 "            [-t | --type check_type[,check_type...]]\n"
 132                 "            [-w | --window_size size]\n"
 133                 "options:\n"
 134                 "-M: device to start LFSCK/scrub on\n"
 135                 "-A: start LFSCK on all nodes via the specified MDT device "
 136                     "(see \"-M\" option) by single LFSCK command\n"
 137                 "-c: create the lost OST-object for dangling LOV EA "
 138                     "(default 'off', or 'on')\n"
 139                 "-C: create the lost MDT-object for dangling name entry "
 140                     "(default 'off', or 'on')\n"
 141                 "-e: error handle mode (default 'continue', or 'abort')\n"
 142                 "-h: this help message\n"
 143                 "-n: check with no modification (default 'off', or 'on')\n"
 144                 "-o: repair orphan OST-objects\n"
 145                 "-r: reset scanning to the start of the device\n"
 146                 "-s: maximum items to be scanned per second "
 147                     "(default '%d' = no limit)\n"
 148                 "-t: check type(s) to be performed (default all)\n"
 149                 "-w: window size for async requests pipeline\n",
 150                 LFSCK_SPEED_NO_LIMIT);
 151 }
 152
 153 static void usage_stop(void)
 154 {
 155         fprintf(stderr, "stop LFSCK\n"
 156                 "usage:\n"
 157                 "lfsck_stop <-M | --device {MDT,OST}_device>\n"
 158                 "           [-A | --all] [-h | --help]\n"
 159                 "options:\n"
 160                 "-M: device to stop LFSCK/scrub on\n"
 161                 "-A: stop LFSCK on all nodes via the specified MDT device "
 162                     "(see \"-M\" option) by single LFSCK command\n"
 163                 "-h: this help message\n");
 164 }
 165
 166 static void usage_query(void)
 167 {
 168         fprintf(stderr, "check the LFSCK global status\n"
 169                 "usage:\n"
 170                 "lfsck_query <-M | --device MDT_device> [-h | --help]\n"
 171                 "            [-t | --type check_type[,check_type...]]\n"
 172                 "            [-t | --wait]\n"
 173                 "options:\n"
 174                 "-M: device to query LFSCK on\n"
 175                 "-t: LFSCK type(s) to be queried (default is all)\n"
 176                 "-h: this help message\n"
 177                 "-w: do not return until LFSCK not running\n");
 178 }
 179
 180 static int lfsck_pack_dev(struct obd_ioctl_data *data, char *device, char *arg)
 181 {
 182         int len = strlen(arg) + 1;
 183
 184         if (len > MAX_OBD_NAME) {
 185                 fprintf(stderr, "device name is too long. "
 186                         "Valid length should be less than %d\n", MAX_OBD_NAME);
 187                 return -EINVAL;
 188         }
 189
 190         memcpy(device, arg, len);
 191         data->ioc_inlbuf4 = device;
 192         data->ioc_inllen4 = len;
 193         data->ioc_dev = OBD_DEV_BY_DEVNAME;
 194         return 0;
 195 }
 196
 197 int jt_lfsck_start(int argc, char **argv)
 198 {
 199         struct obd_ioctl_data data;
 200         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 201         char device[MAX_OBD_NAME];
 202         struct lfsck_start start;
 203         char *optstring = "Ac::C::e:hM:n::ors:t:w:";
 204         int opt, index, rc, val, i;
 205
 206         memset(&data, 0, sizeof(data));
 207         memset(&start, 0, sizeof(start));
 208         memset(device, 0, MAX_OBD_NAME);
 209         start.ls_version = LFSCK_VERSION_V1;
 210         start.ls_active = LFSCK_TYPES_ALL;
 211
 212         /* Reset the 'optind' for the case of getopt_long() called multiple
 213          * times under the same lctl. */
 214         optind = 0;
 215         while ((opt = getopt_long(argc, argv, optstring, long_opt_start,
 216                                   &index)) != EOF) {
 217                 switch (opt) {
 218                 case 'A':
 219                         start.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST;
 220                         break;
 221                 case 'c':
 222                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 223                                 start.ls_flags |= LPF_CREATE_OSTOBJ;
 224                         } else if (strcmp(optarg, "off") != 0) {
 225                                 fprintf(stderr, "invalid switch: -c '%s'. "
 226                                         "valid switches are:\n"
 227                                         "empty ('on'), or 'off' without space. "
 228                                         "For example:\n"
 229                                         "'-c', '-con', '-coff'\n", optarg);
 230                                 return -EINVAL;
 231                         }
 232                         start.ls_valid |= LSV_CREATE_OSTOBJ;
 233                         break;
 234                 case 'C':
 235                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 236                                 start.ls_flags |= LPF_CREATE_MDTOBJ;
 237                         } else if (strcmp(optarg, "off") != 0) {
 238                                 fprintf(stderr, "invalid switch: -C '%s'. "
 239                                         "valid switches are:\n"
 240                                         "empty ('on'), or 'off' without space. "
 241                                         "For example:\n"
 242                                         "'-C', '-Con', '-Coff'\n", optarg);
 243                                 return -EINVAL;
 244                         }
 245                         start.ls_valid |= LSV_CREATE_MDTOBJ;
 246                         break;
 247                 case 'e':
 248                         if (strcmp(optarg, "abort") == 0) {
 249                                 start.ls_flags |= LPF_FAILOUT;
 250                         } else if (strcmp(optarg, "continue") != 0) {
 251                                 fprintf(stderr, "invalid error mode: -e '%s'."
 252                                         "valid modes are: "
 253                                         "'continue' or 'abort'.\n", optarg);
 254                                 return -EINVAL;
 255                         }
 256                         start.ls_valid |= LSV_ERROR_HANDLE;
 257                         break;
 258                 case 'h':
 259                         usage_start();
 260                         return 0;
 261                 case 'M':
 262                         rc = lfsck_pack_dev(&data, device, optarg);
 263                         if (rc != 0)
 264                                 return rc;
 265                         break;
 266                 case 'n':
 267                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 268                                 start.ls_flags |= LPF_DRYRUN;
 269                         } else if (strcmp(optarg, "off") != 0) {
 270                                 fprintf(stderr, "invalid switch: -n '%s'. "
 271                                         "valid switches are:\n"
 272                                         "empty ('on'), or 'off' without space. "
 273                                         "For example:\n"
 274                                         "'-n', '-non', '-noff'\n", optarg);
 275                                 return -EINVAL;
 276                         }
 277                         start.ls_valid |= LSV_DRYRUN;
 278                         break;
 279                 case 'o':
 280                         start.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST |
 281                                           LPF_OST_ORPHAN;
 282                         break;
 283                 case 'r':
 284                         start.ls_flags |= LPF_RESET;
 285                         break;
 286                 case 's':
 287                         val = atoi(optarg);
 288                         start.ls_speed_limit = val;
 289                         start.ls_valid |= LSV_SPEED_LIMIT;
 290                         break;
 291                 case 't': {
 292                         char *typename;
 293
 294                         if (start.ls_active == LFSCK_TYPES_ALL)
 295                                 start.ls_active = 0;
 296                         while ((typename = strsep(&optarg, ",")) != NULL) {
 297                                 enum lfsck_type type;
 298
 299                                 type = lfsck_name2type(typename);
 300                                 if (type == -1)
 301                                         goto bad_type;
 302                                 start.ls_active |= type;
 303                         }
 304                         break;
 305 bad_type:
 306                         fprintf(stderr, "invalid check type -t '%s'. "
 307                                 "valid types are:\n", typename);
 308                         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++)
 309                                 fprintf(stderr, "%s%s", i != 0 ? "," : "",
 310                                         lfsck_types_names[i].ltn_name);
 311                         fprintf(stderr, "\n");
 312                         return -EINVAL;
 313                 }
 314                 case 'w':
 315                         val = atoi(optarg);
 316                         if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
 317                                 fprintf(stderr,
 318                                         "Invalid async window size that "
 319                                         "may cause memory issues. The valid "
 320                                         "range is [1 - %u].\n",
 321                                         LFSCK_ASYNC_WIN_MAX);
 322                                 return -EINVAL;
 323                         }
 324
 325                         start.ls_async_windows = val;
 326                         start.ls_valid |= LSV_ASYNC_WINDOWS;
 327                         break;
 328                 default:
 329                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 330                         return -EINVAL;
 331                 }
 332         }
 333
 334         if (start.ls_active == LFSCK_TYPES_ALL)
 335                 start.ls_active = LFSCK_TYPES_DEF;
 336
 337         if (data.ioc_inlbuf4 == NULL) {
 338                 if (lcfg_get_devname() != NULL) {
 339                         rc = lfsck_pack_dev(&data, device, lcfg_get_devname());
 340                         if (rc != 0)
 341                                 return rc;
 342                 } else {
 343                         fprintf(stderr,
 344                                 "Must specify device to start LFSCK.\n");
 345                         return -EINVAL;
 346                 }
 347         }
 348
 349         data.ioc_inlbuf1 = (char *)&start;
 350         data.ioc_inllen1 = sizeof(start);
 351         memset(buf, 0, sizeof(rawbuf));
 352         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 353         if (rc != 0) {
 354                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 355                 return rc;
 356         }
 357
 358         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_START_LFSCK, buf);
 359         if (rc < 0) {
 360                 perror("Fail to start LFSCK");
 361                 return rc;
 362         }
 363
 364         obd_ioctl_unpack(&data, buf, sizeof(rawbuf));
 365         printf("Started LFSCK on the device %s: scrub", device);
 366         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 367                 if (start.ls_active & lfsck_types_names[i].ltn_type) {
 368                         printf(" %s", lfsck_types_names[i].ltn_name);
 369                         start.ls_active &= ~lfsck_types_names[i].ltn_type;
 370                 }
 371         }
 372         if (start.ls_active != 0)
 373                 printf(" unknown(0x%x)", start.ls_active);
 374         printf("\n");
 375
 376         return 0;
 377 }
 378
 379 int jt_lfsck_stop(int argc, char **argv)
 380 {
 381         struct obd_ioctl_data data;
 382         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 383         char device[MAX_OBD_NAME];
 384         struct lfsck_stop stop;
 385         char *optstring = "AhM:";
 386         int opt, index, rc;
 387
 388         memset(&data, 0, sizeof(data));
 389         memset(&stop, 0, sizeof(stop));
 390         memset(device, 0, MAX_OBD_NAME);
 391
 392         /* Reset the 'optind' for the case of getopt_long() called multiple
 393          * times under the same lctl. */
 394         optind = 0;
 395         while ((opt = getopt_long(argc, argv, optstring, long_opt_stop,
 396                                   &index)) != EOF) {
 397                 switch (opt) {
 398                 case 'A':
 399                         stop.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST;
 400                         break;
 401                 case 'h':
 402                         usage_stop();
 403                         return 0;
 404                 case 'M':
 405                         rc = lfsck_pack_dev(&data, device, optarg);
 406                         if (rc != 0)
 407                                 return rc;
 408                         break;
 409                 default:
 410                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 411                         return -EINVAL;
 412                 }
 413         }
 414
 415         if (data.ioc_inlbuf4 == NULL) {
 416                 if (lcfg_get_devname() != NULL) {
 417                         rc = lfsck_pack_dev(&data, device, lcfg_get_devname());
 418                         if (rc != 0)
 419                                 return rc;
 420                 } else {
 421                         fprintf(stderr,
 422                                 "Must specify device to stop LFSCK.\n");
 423                         return -EINVAL;
 424                 }
 425         }
 426
 427         data.ioc_inlbuf1 = (char *)&stop;
 428         data.ioc_inllen1 = sizeof(stop);
 429         memset(buf, 0, sizeof(rawbuf));
 430         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 431         if (rc != 0) {
 432                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 433                 return rc;
 434         }
 435
 436         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_STOP_LFSCK, buf);
 437         if (rc < 0) {
 438                 perror("Fail to stop LFSCK");
 439                 return rc;
 440         }
 441
 442         printf("Stopped LFSCK on the device %s.\n", device);
 443         return 0;
 444 }
 445
 446 int jt_lfsck_query(int argc, char **argv)
 447 {
 448         struct obd_ioctl_data data = { 0 };
 449         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 450         char device[MAX_OBD_NAME] = "";
 451         struct lfsck_query query = { .lu_types = LFSCK_TYPES_ALL };
 452         int opt, index, rc, i;
 453         enum lfsck_type type;
 454
 455         while ((opt = getopt_long(argc, argv, "hM:t:w", long_opt_query,
 456                                   &index)) != EOF) {
 457                 switch (opt) {
 458                 case 'h':
 459                         usage_query();
 460                         return 0;
 461                 case 'M':
 462                         rc = lfsck_pack_dev(&data, device, optarg);
 463                         if (rc != 0)
 464                                 return rc;
 465                         break;
 466                 case 't': {
 467                         char *typename;
 468
 469                         if (query.lu_types == LFSCK_TYPES_ALL)
 470                                 query.lu_types = 0;
 471                         while ((typename = strsep(&optarg, ",")) != NULL) {
 472                                 type = lfsck_name2type(typename);
 473                                 if (type == -1)
 474                                         goto bad_type;
 475                                 query.lu_types |= type;
 476                         }
 477                         break;
 478
 479 bad_type:
 480                         fprintf(stderr, "invalid LFSCK type -t '%s'. "
 481                                 "valid types are:\n", typename);
 482                         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++)
 483                                 fprintf(stderr, "%s%s", i != 0 ? "," : "",
 484                                         lfsck_types_names[i].ltn_name);
 485                         fprintf(stderr, "\n");
 486                         return -EINVAL;
 487                 }
 488                 case 'w':
 489                         query.lu_flags |= LPF_WAIT;
 490                         break;
 491                 default:
 492                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 493                         usage_query();
 494                         return -EINVAL;
 495                 }
 496         }
 497
 498         if (data.ioc_inlbuf4 == NULL) {
 499                 if (lcfg_get_devname() != NULL) {
 500                         rc = lfsck_pack_dev(&data, device, lcfg_get_devname());
 501                         if (rc != 0)
 502                                 return rc;
 503                 } else {
 504                         fprintf(stderr,
 505                                 "Must specify device to query LFSCK.\n");
 506                         return -EINVAL;
 507                 }
 508         }
 509
 510         data.ioc_inlbuf1 = (char *)&query;
 511         data.ioc_inllen1 = sizeof(query);
 512         memset(buf, 0, sizeof(rawbuf));
 513         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 514         if (rc != 0) {
 515                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 516                 return rc;
 517         }
 518
 519         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_QUERY_LFSCK, buf);
 520         if (rc < 0) {
 521                 perror("Fail to query LFSCK");
 522                 return rc;
 523         }
 524
 525         obd_ioctl_unpack(&data, buf, sizeof(rawbuf));
 526         for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS; i++, type = 1 << i) {
 527                 const char *name;
 528                 int j;
 529
 530                 if (!(query.lu_types & type))
 531                         continue;
 532
 533                 name = lfsck_type2name(type);
 534                 for (j = 0; j <= LS_MAX; j++)
 535                         printf("%s_mdts_%s: %d\n", name,
 536                                lfsck_status2name(j), query.lu_mdts_count[i][j]);
 537
 538                 for (j = 0; j <= LS_MAX; j++)
 539                         printf("%s_osts_%s: %d\n", name,
 540                                lfsck_status2name(j), query.lu_osts_count[i][j]);
 541
 542                 printf("%s_repaired: %llu\n", name, query.lu_repaired[i]);
 543         }
 544
 545         return 0;
 546 }