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, 2016, 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 #include <libcfs/util/param.h>
  50
  51 static struct option long_opt_start[] = {
  52         {"device",              required_argument, 0, 'M'},
  53         {"all",                 no_argument,       0, 'A'},
  54         {"create_ostobj",       optional_argument, 0, 'c'},
  55         {"create-ostobj",       optional_argument, 0, 'c'},
  56         {"create_mdtobj",       optional_argument, 0, 'C'},
  57         {"create-mdtobj",       optional_argument, 0, 'C'},
  58         {"error",               required_argument, 0, 'e'},
  59         {"help",                no_argument,       0, 'h'},
  60         {"dryrun",              optional_argument, 0, 'n'},
  61         {"orphan",              no_argument,       0, 'o'},
  62         {"reset",               no_argument,       0, 'r'},
  63         {"speed",               required_argument, 0, 's'},
  64         {"type",                required_argument, 0, 't'},
  65         {"window_size",         required_argument, 0, 'w'},
  66         {"window-size",         required_argument, 0, 'w'},
  67         {0,                     0,                 0,  0 }
  68 };
  69
  70 static struct option long_opt_stop[] = {
  71         {"device",      required_argument, 0, 'M'},
  72         {"all",         no_argument,       0, 'A'},
  73         {"help",        no_argument,       0, 'h'},
  74         {0,             0,                 0,  0 }
  75 };
  76
  77 static struct option long_opt_query[] = {
  78         {"device",      required_argument, 0, 'M'},
  79         {"type",        required_argument, 0, 't'},
  80         {"help",        no_argument,       0, 'h'},
  81         {"wait",        no_argument,       0, 'w'},
  82         {0,             0,                 0,  0 }
  83 };
  84
  85 struct lfsck_type_name {
  86         char            *ltn_name;
  87         enum lfsck_type  ltn_type;
  88 };
  89
  90 static struct lfsck_type_name lfsck_types_names[] = {
  91         { "scrub",      LFSCK_TYPE_SCRUB },
  92         { "layout",     LFSCK_TYPE_LAYOUT },
  93         { "namespace",  LFSCK_TYPE_NAMESPACE },
  94         { "default",    LFSCK_TYPES_DEF },
  95         { "all",        LFSCK_TYPES_SUPPORTED },
  96         { NULL,         0 }
  97 };
  98
  99 static enum lfsck_type lfsck_name2type(const char *name)
 100 {
 101         int i;
 102
 103         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 104                 if (strcmp(lfsck_types_names[i].ltn_name, name) == 0)
 105                         return lfsck_types_names[i].ltn_type;
 106         }
 107         return -1;
 108 }
 109
 110 static const char *lfsck_type2name(__u16 type)
 111 {
 112         int i;
 113
 114         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 115                 if (type == lfsck_types_names[i].ltn_type)
 116                         return lfsck_types_names[i].ltn_name;
 117         }
 118
 119         return NULL;
 120 }
 121
 122 static void usage_start(void)
 123 {
 124         fprintf(stderr, "start LFSCK\n"
 125                 "usage:\n"
 126                 "lfsck_start [-M | --device {MDT,OST}_device]\n"
 127                 "            [-A | --all] [-c | --create_ostobj [on | off]]\n"
 128                 "            [-C | --create_mdtobj [on | off]]\n"
 129                 "            [-e | --error {continue | abort}] [-h | --help]\n"
 130                 "            [-n | --dryrun [on | off]] [-o | --orphan]\n"
 131                 "            [-r | --reset] [-s | --speed ops_per_sec_limit]\n"
 132                 "            [-t | --type check_type[,check_type...]]\n"
 133                 "            [-w | --window_size size]\n"
 134                 "options:\n"
 135                 "-M: device to start LFSCK/scrub on\n"
 136                 "-A: start LFSCK on all nodes via the specified MDT device "
 137                     "(see \"-M\" option) by single LFSCK command\n"
 138                 "-c: create the lost OST-object for dangling LOV EA "
 139                     "(default 'off', or 'on')\n"
 140                 "-C: create the lost MDT-object for dangling name entry "
 141                     "(default 'off', or 'on')\n"
 142                 "-e: error handle mode (default 'continue', or 'abort')\n"
 143                 "-h: this help message\n"
 144                 "-n: check with no modification (default 'off', or 'on')\n"
 145                 "-o: repair orphan OST-objects\n"
 146                 "-r: reset scanning to the start of the device\n"
 147                 "-s: maximum items to be scanned per second "
 148                     "(default '%d' = no limit)\n"
 149                 "-t: check type(s) to be performed (default all)\n"
 150                 "-w: window size for async requests pipeline\n",
 151                 LFSCK_SPEED_NO_LIMIT);
 152 }
 153
 154 static void usage_stop(void)
 155 {
 156         fprintf(stderr, "stop LFSCK\n"
 157                 "usage:\n"
 158                 "lfsck_stop [-M | --device {MDT,OST}_device]\n"
 159                 "           [-A | --all] [-h | --help]\n"
 160                 "options:\n"
 161                 "-M: device to stop LFSCK/scrub on\n"
 162                 "-A: stop LFSCK on all nodes via the specified MDT device "
 163                     "(see \"-M\" option) by single LFSCK command\n"
 164                 "-h: this help message\n");
 165 }
 166
 167 static void usage_query(void)
 168 {
 169         fprintf(stderr, "check the LFSCK global status\n"
 170                 "usage:\n"
 171                 "lfsck_query [-M | --device MDT_device] [-h | --help]\n"
 172                 "            [-t | --type check_type[,check_type...]]\n"
 173                 "            [-t | --wait]\n"
 174                 "options:\n"
 175                 "-M: device to query LFSCK on\n"
 176                 "-t: LFSCK type(s) to be queried (default is all)\n"
 177                 "-h: this help message\n"
 178                 "-w: do not return until LFSCK not running\n");
 179 }
 180
 181 static int lfsck_pack_dev(struct obd_ioctl_data *data, char *device, char *arg)
 182 {
 183         int len = strlen(arg) + 1;
 184
 185         if (len > MAX_OBD_NAME) {
 186                 fprintf(stderr, "device name is too long. "
 187                         "Valid length should be less than %d\n", MAX_OBD_NAME);
 188                 return -EINVAL;
 189         }
 190
 191         memcpy(device, arg, len);
 192         data->ioc_inlbuf4 = device;
 193         data->ioc_inllen4 = len;
 194         data->ioc_dev = OBD_DEV_BY_DEVNAME;
 195         return 0;
 196 }
 197
 198 static int lfsck_get_dev_name(struct obd_ioctl_data *data, char *device,
 199                               int types, bool multipe_devices)
 200 {
 201         glob_t param = { 0 };
 202         char *ptr;
 203         int rc;
 204         int i;
 205
 206         rc = cfs_get_param_paths(&param, "mdd/*-MDT*");
 207         if (rc) {
 208                 if (multipe_devices || errno != ENOENT ||
 209                     types & LFSCK_TYPE_NAMESPACE) {
 210                         fprintf(stderr, "Fail to get device name: rc = %d\n."
 211                                 "You can specify the device explicitly "
 212                                 "via '-M' option.\n", rc);
 213                         return rc;
 214                 }
 215
 216                 rc = cfs_get_param_paths(&param, "obdfilter/*-OST*");
 217                 if (rc) {
 218                         fprintf(stderr, "Fail to get device name: rc = %d\n."
 219                                 "You can specify the device explicitly "
 220                                 "via '-M' option.\n", rc);
 221                         return rc;
 222                 }
 223         }
 224
 225         if (param.gl_pathc == 1)
 226                 goto pack;
 227
 228         if (!multipe_devices) {
 229                 fprintf(stderr,
 230                         "Detect multiple devices on current node. "
 231                         "Please specify the device explicitly "
 232                         "via '-M' option or '-A' option for all.\n");
 233                 rc = -EINVAL;
 234                 goto out;
 235         }
 236
 237         ptr = strrchr(param.gl_pathv[0], '-');
 238         LASSERT(ptr != NULL);
 239
 240         for (i = 1; i < param.gl_pathc; i++) {
 241                 char *ptr2 = strrchr(param.gl_pathv[i], '-');
 242
 243                 LASSERT(ptr2 != NULL);
 244
 245                 if ((ptr - param.gl_pathv[0]) != (ptr2 - param.gl_pathv[i]) ||
 246                     strncmp(param.gl_pathv[0], param.gl_pathv[i],
 247                             (ptr - param.gl_pathv[0])) != 0) {
 248                         fprintf(stderr,
 249                                 "Detect multiple filesystems on current node. "
 250                                 "Please specify the device explicitly "
 251                                 "via '-M' option.\n");
 252                         rc = -EINVAL;
 253                         goto out;
 254                 }
 255         }
 256
 257 pack:
 258         rc = lfsck_pack_dev(data, device, basename(param.gl_pathv[0]));
 259
 260 out:
 261         cfs_free_param_data(&param);
 262
 263         return rc;
 264 }
 265
 266 int jt_lfsck_start(int argc, char **argv)
 267 {
 268         struct obd_ioctl_data data;
 269         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 270         char device[MAX_OBD_NAME];
 271         struct lfsck_start start;
 272         char *optstring = "Ac::C::e:hM:n::ors:t:w:";
 273         int opt, index, rc, val, i;
 274
 275         memset(&data, 0, sizeof(data));
 276         memset(&start, 0, sizeof(start));
 277         memset(device, 0, MAX_OBD_NAME);
 278         start.ls_version = LFSCK_VERSION_V1;
 279         start.ls_active = LFSCK_TYPES_ALL;
 280
 281         /* Reset the 'optind' for the case of getopt_long() called multiple
 282          * times under the same lctl. */
 283         optind = 0;
 284         while ((opt = getopt_long(argc, argv, optstring, long_opt_start,
 285                                   &index)) != EOF) {
 286                 switch (opt) {
 287                 case 'A':
 288                         start.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST;
 289                         break;
 290                 case 'c':
 291                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 292                                 start.ls_flags |= LPF_CREATE_OSTOBJ;
 293                         } else if (strcmp(optarg, "off") != 0) {
 294                                 fprintf(stderr, "invalid switch: -c '%s'. "
 295                                         "valid switches are:\n"
 296                                         "empty ('on'), or 'off' without space. "
 297                                         "For example:\n"
 298                                         "'-c', '-con', '-coff'\n", optarg);
 299                                 return -EINVAL;
 300                         }
 301                         start.ls_valid |= LSV_CREATE_OSTOBJ;
 302                         break;
 303                 case 'C':
 304                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 305                                 start.ls_flags |= LPF_CREATE_MDTOBJ;
 306                         } else if (strcmp(optarg, "off") != 0) {
 307                                 fprintf(stderr, "invalid switch: -C '%s'. "
 308                                         "valid switches are:\n"
 309                                         "empty ('on'), or 'off' without space. "
 310                                         "For example:\n"
 311                                         "'-C', '-Con', '-Coff'\n", optarg);
 312                                 return -EINVAL;
 313                         }
 314                         start.ls_valid |= LSV_CREATE_MDTOBJ;
 315                         break;
 316                 case 'e':
 317                         if (strcmp(optarg, "abort") == 0) {
 318                                 start.ls_flags |= LPF_FAILOUT;
 319                         } else if (strcmp(optarg, "continue") != 0) {
 320                                 fprintf(stderr, "invalid error mode: -e '%s'."
 321                                         "valid modes are: "
 322                                         "'continue' or 'abort'.\n", optarg);
 323                                 return -EINVAL;
 324                         }
 325                         start.ls_valid |= LSV_ERROR_HANDLE;
 326                         break;
 327                 case 'h':
 328                         usage_start();
 329                         return 0;
 330                 case 'M':
 331                         rc = lfsck_pack_dev(&data, device, optarg);
 332                         if (rc != 0)
 333                                 return rc;
 334                         break;
 335                 case 'n':
 336                         if (optarg == NULL || strcmp(optarg, "on") == 0) {
 337                                 start.ls_flags |= LPF_DRYRUN;
 338                         } else if (strcmp(optarg, "off") != 0) {
 339                                 fprintf(stderr, "invalid switch: -n '%s'. "
 340                                         "valid switches are:\n"
 341                                         "empty ('on'), or 'off' without space. "
 342                                         "For example:\n"
 343                                         "'-n', '-non', '-noff'\n", optarg);
 344                                 return -EINVAL;
 345                         }
 346                         start.ls_valid |= LSV_DRYRUN;
 347                         break;
 348                 case 'o':
 349                         start.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST |
 350                                           LPF_OST_ORPHAN;
 351                         break;
 352                 case 'r':
 353                         start.ls_flags |= LPF_RESET;
 354                         break;
 355                 case 's':
 356                         val = atoi(optarg);
 357                         start.ls_speed_limit = val;
 358                         start.ls_valid |= LSV_SPEED_LIMIT;
 359                         break;
 360                 case 't': {
 361                         char *typename;
 362
 363                         if (start.ls_active == LFSCK_TYPES_ALL)
 364                                 start.ls_active = 0;
 365                         while ((typename = strsep(&optarg, ",")) != NULL) {
 366                                 enum lfsck_type type;
 367
 368                                 type = lfsck_name2type(typename);
 369                                 if (type == -1)
 370                                         goto bad_type;
 371                                 start.ls_active |= type;
 372                         }
 373                         break;
 374 bad_type:
 375                         fprintf(stderr, "invalid check type -t '%s'. "
 376                                 "valid types are:\n", typename);
 377                         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++)
 378                                 fprintf(stderr, "%s%s", i != 0 ? "," : "",
 379                                         lfsck_types_names[i].ltn_name);
 380                         fprintf(stderr, "\n");
 381                         return -EINVAL;
 382                 }
 383                 case 'w':
 384                         val = atoi(optarg);
 385                         if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
 386                                 fprintf(stderr,
 387                                         "Invalid async window size that "
 388                                         "may cause memory issues. The valid "
 389                                         "range is [1 - %u].\n",
 390                                         LFSCK_ASYNC_WIN_MAX);
 391                                 return -EINVAL;
 392                         }
 393
 394                         start.ls_async_windows = val;
 395                         start.ls_valid |= LSV_ASYNC_WINDOWS;
 396                         break;
 397                 default:
 398                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 399                         return -EINVAL;
 400                 }
 401         }
 402
 403         if (start.ls_active == LFSCK_TYPES_ALL)
 404                 start.ls_active = LFSCK_TYPES_DEF;
 405
 406         if (data.ioc_inlbuf4 == NULL) {
 407                 rc = lfsck_get_dev_name(&data, device, start.ls_active,
 408                                         start.ls_flags & LPF_ALL_TGT);
 409                 if (rc != 0)
 410                         return rc;
 411         }
 412
 413         data.ioc_inlbuf1 = (char *)&start;
 414         data.ioc_inllen1 = sizeof(start);
 415         memset(buf, 0, sizeof(rawbuf));
 416         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 417         if (rc != 0) {
 418                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 419                 return rc;
 420         }
 421
 422         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_START_LFSCK, buf);
 423         if (rc < 0) {
 424                 perror("Fail to start LFSCK");
 425                 return rc;
 426         }
 427
 428         obd_ioctl_unpack(&data, buf, sizeof(rawbuf));
 429         printf("Started LFSCK on the device %s: scrub", device);
 430         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++) {
 431                 if (start.ls_active & lfsck_types_names[i].ltn_type) {
 432                         printf(" %s", lfsck_types_names[i].ltn_name);
 433                         start.ls_active &= ~lfsck_types_names[i].ltn_type;
 434                 }
 435         }
 436         if (start.ls_active != 0)
 437                 printf(" unknown(0x%x)", start.ls_active);
 438         printf("\n");
 439
 440         return 0;
 441 }
 442
 443 int jt_lfsck_stop(int argc, char **argv)
 444 {
 445         struct obd_ioctl_data data;
 446         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 447         char device[MAX_OBD_NAME];
 448         struct lfsck_stop stop;
 449         char *optstring = "AhM:";
 450         int opt, index, rc;
 451
 452         memset(&data, 0, sizeof(data));
 453         memset(&stop, 0, sizeof(stop));
 454         memset(device, 0, MAX_OBD_NAME);
 455
 456         /* Reset the 'optind' for the case of getopt_long() called multiple
 457          * times under the same lctl. */
 458         optind = 0;
 459         while ((opt = getopt_long(argc, argv, optstring, long_opt_stop,
 460                                   &index)) != EOF) {
 461                 switch (opt) {
 462                 case 'A':
 463                         stop.ls_flags |= LPF_ALL_TGT | LPF_BROADCAST;
 464                         break;
 465                 case 'h':
 466                         usage_stop();
 467                         return 0;
 468                 case 'M':
 469                         rc = lfsck_pack_dev(&data, device, optarg);
 470                         if (rc != 0)
 471                                 return rc;
 472                         break;
 473                 default:
 474                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 475                         return -EINVAL;
 476                 }
 477         }
 478
 479         if (data.ioc_inlbuf4 == NULL) {
 480                 rc = lfsck_get_dev_name(&data, device, 0,
 481                                         stop.ls_flags & LPF_ALL_TGT);
 482                 if (rc != 0)
 483                         return rc;
 484         }
 485
 486         data.ioc_inlbuf1 = (char *)&stop;
 487         data.ioc_inllen1 = sizeof(stop);
 488         memset(buf, 0, sizeof(rawbuf));
 489         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 490         if (rc != 0) {
 491                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 492                 return rc;
 493         }
 494
 495         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_STOP_LFSCK, buf);
 496         if (rc < 0) {
 497                 perror("Fail to stop LFSCK");
 498                 return rc;
 499         }
 500
 501         printf("Stopped LFSCK on the device %s.\n", device);
 502         return 0;
 503 }
 504
 505 int jt_lfsck_query(int argc, char **argv)
 506 {
 507         struct obd_ioctl_data data = { 0 };
 508         char rawbuf[MAX_IOC_BUFLEN], *buf = rawbuf;
 509         char device[MAX_OBD_NAME] = "";
 510         struct lfsck_query query = { .lu_types = LFSCK_TYPES_ALL };
 511         int opt, index, rc, i;
 512         enum lfsck_type type;
 513
 514         while ((opt = getopt_long(argc, argv, "hM:t:w", long_opt_query,
 515                                   &index)) != EOF) {
 516                 switch (opt) {
 517                 case 'h':
 518                         usage_query();
 519                         return 0;
 520                 case 'M':
 521                         rc = lfsck_pack_dev(&data, device, optarg);
 522                         if (rc != 0)
 523                                 return rc;
 524                         break;
 525                 case 't': {
 526                         char *typename;
 527
 528                         if (query.lu_types == LFSCK_TYPES_ALL)
 529                                 query.lu_types = 0;
 530                         while ((typename = strsep(&optarg, ",")) != NULL) {
 531                                 type = lfsck_name2type(typename);
 532                                 if (type == -1)
 533                                         goto bad_type;
 534                                 query.lu_types |= type;
 535                         }
 536                         break;
 537
 538 bad_type:
 539                         fprintf(stderr, "invalid LFSCK type -t '%s'. "
 540                                 "valid types are:\n", typename);
 541                         for (i = 0; lfsck_types_names[i].ltn_name != NULL; i++)
 542                                 fprintf(stderr, "%s%s", i != 0 ? "," : "",
 543                                         lfsck_types_names[i].ltn_name);
 544                         fprintf(stderr, "\n");
 545                         return -EINVAL;
 546                 }
 547                 case 'w':
 548                         query.lu_flags |= LPF_WAIT;
 549                         break;
 550                 default:
 551                         fprintf(stderr, "Invalid option, '-h' for help.\n");
 552                         usage_query();
 553                         return -EINVAL;
 554                 }
 555         }
 556
 557         if (data.ioc_inlbuf4 == NULL) {
 558                 rc = lfsck_get_dev_name(&data, device, 0, true);
 559                 if (rc != 0)
 560                         return rc;
 561         }
 562
 563         data.ioc_inlbuf1 = (char *)&query;
 564         data.ioc_inllen1 = sizeof(query);
 565         memset(buf, 0, sizeof(rawbuf));
 566         rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
 567         if (rc != 0) {
 568                 fprintf(stderr, "Fail to pack ioctl data: rc = %d.\n", rc);
 569                 return rc;
 570         }
 571
 572         rc = l_ioctl(OBD_DEV_ID, OBD_IOC_QUERY_LFSCK, buf);
 573         if (rc < 0) {
 574                 perror("Fail to query LFSCK");
 575                 return rc;
 576         }
 577
 578         obd_ioctl_unpack(&data, buf, sizeof(rawbuf));
 579         for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS; i++, type = 1 << i) {
 580                 const char *name;
 581                 int j;
 582
 583                 if (!(query.lu_types & type))
 584                         continue;
 585
 586                 name = lfsck_type2name(type);
 587                 for (j = 0; j <= LS_MAX; j++)
 588                         printf("%s_mdts_%s: %d\n", name,
 589                                lfsck_status2name(j), query.lu_mdts_count[i][j]);
 590
 591                 for (j = 0; j <= LS_MAX; j++)
 592                         printf("%s_osts_%s: %d\n", name,
 593                                lfsck_status2name(j), query.lu_osts_count[i][j]);
 594
 595                 printf("%s_repaired: %llu\n", name, query.lu_repaired[i]);
 596         }
 597
 598         return 0;
 599 }