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