spaceman/health.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (c) 2019 Oracle.
   4  * All Rights Reserved.
   5  */
   6 #include "platform_defs.h"
   7 #include "libxfs.h"
   8 #include "command.h"
   9 #include "init.h"
  10 #include "input.h"
  11 #include "libfrog/logging.h"
  12 #include "libfrog/paths.h"
  13 #include "libfrog/fsgeom.h"
  14 #include "libfrog/bulkstat.h"
  15 #include "space.h"
  16
  17 static cmdinfo_t health_cmd;
  18 static unsigned long long reported;
  19 static bool comprehensive;
  20 static bool quiet;
  21
  22 static bool has_realtime(const struct xfs_fsop_geom *g)
  23 {
  24         return g->rtblocks > 0;
  25 }
  26
  27 static bool has_finobt(const struct xfs_fsop_geom *g)
  28 {
  29         return g->flags & XFS_FSOP_GEOM_FLAGS_FINOBT;
  30 }
  31
  32 static bool has_rmapbt(const struct xfs_fsop_geom *g)
  33 {
  34         return g->flags & XFS_FSOP_GEOM_FLAGS_RMAPBT;
  35 }
  36
  37 static bool has_reflink(const struct xfs_fsop_geom *g)
  38 {
  39         return g->flags & XFS_FSOP_GEOM_FLAGS_REFLINK;
  40 }
  41
  42 struct flag_map {
  43         unsigned int            mask;
  44         bool                    (*has_fn)(const struct xfs_fsop_geom *g);
  45         const char              *descr;
  46 };
  47
  48 static const struct flag_map fs_flags[] = {
  49         {
  50                 .mask = XFS_FSOP_GEOM_SICK_COUNTERS,
  51                 .descr = "summary counters",
  52         },
  53         {
  54                 .mask = XFS_FSOP_GEOM_SICK_UQUOTA,
  55                 .descr = "user quota",
  56         },
  57         {
  58                 .mask = XFS_FSOP_GEOM_SICK_GQUOTA,
  59                 .descr = "group quota",
  60         },
  61         {
  62                 .mask = XFS_FSOP_GEOM_SICK_PQUOTA,
  63                 .descr = "project quota",
  64         },
  65         {
  66                 .mask = XFS_FSOP_GEOM_SICK_RT_BITMAP,
  67                 .descr = "realtime bitmap",
  68                 .has_fn = has_realtime,
  69         },
  70         {
  71                 .mask = XFS_FSOP_GEOM_SICK_RT_SUMMARY,
  72                 .descr = "realtime summary",
  73                 .has_fn = has_realtime,
  74         },
  75         {0},
  76 };
  77
  78 static const struct flag_map ag_flags[] = {
  79         {
  80                 .mask = XFS_AG_GEOM_SICK_SB,
  81                 .descr = "superblock",
  82         },
  83         {
  84                 .mask = XFS_AG_GEOM_SICK_AGF,
  85                 .descr = "AGF header",
  86         },
  87         {
  88                 .mask = XFS_AG_GEOM_SICK_AGFL,
  89                 .descr = "AGFL header",
  90         },
  91         {
  92                 .mask = XFS_AG_GEOM_SICK_AGI,
  93                 .descr = "AGI header",
  94         },
  95         {
  96                 .mask = XFS_AG_GEOM_SICK_BNOBT,
  97                 .descr = "free space by block btree",
  98         },
  99         {
 100                 .mask = XFS_AG_GEOM_SICK_CNTBT,
 101                 .descr = "free space by length btree",
 102         },
 103         {
 104                 .mask = XFS_AG_GEOM_SICK_INOBT,
 105                 .descr = "inode btree",
 106         },
 107         {
 108                 .mask = XFS_AG_GEOM_SICK_FINOBT,
 109                 .descr = "free inode btree",
 110                 .has_fn = has_finobt,
 111         },
 112         {
 113                 .mask = XFS_AG_GEOM_SICK_RMAPBT,
 114                 .descr = "reverse mappings btree",
 115                 .has_fn = has_rmapbt,
 116         },
 117         {
 118                 .mask = XFS_AG_GEOM_SICK_REFCNTBT,
 119                 .descr = "reference count btree",
 120                 .has_fn = has_reflink,
 121         },
 122         {0},
 123 };
 124
 125 static const struct flag_map inode_flags[] = {
 126         {
 127                 .mask = XFS_BS_SICK_INODE,
 128                 .descr = "inode core",
 129         },
 130         {
 131                 .mask = XFS_BS_SICK_BMBTD,
 132                 .descr = "data fork",
 133         },
 134         {
 135                 .mask = XFS_BS_SICK_BMBTA,
 136                 .descr = "extended attribute fork",
 137         },
 138         {
 139                 .mask = XFS_BS_SICK_BMBTC,
 140                 .descr = "copy on write fork",
 141         },
 142         {
 143                 .mask = XFS_BS_SICK_DIR,
 144                 .descr = "directory",
 145         },
 146         {
 147                 .mask = XFS_BS_SICK_XATTR,
 148                 .descr = "extended attributes",
 149         },
 150         {
 151                 .mask = XFS_BS_SICK_SYMLINK,
 152                 .descr = "symbolic link target",
 153         },
 154         {
 155                 .mask = XFS_BS_SICK_PARENT,
 156                 .descr = "parent pointers",
 157         },
 158         {0},
 159 };
 160
 161 /* Convert a flag mask to a report. */
 162 static void
 163 report_sick(
 164         const char                      *descr,
 165         const struct flag_map           *maps,
 166         unsigned int                    sick,
 167         unsigned int                    checked)
 168 {
 169         const struct flag_map           *f;
 170         bool                            bad;
 171
 172         for (f = maps; f->mask != 0; f++) {
 173                 if (f->has_fn && !f->has_fn(&file->xfd.fsgeom))
 174                         continue;
 175                 bad = sick & f->mask;
 176                 if (!bad && !(checked & f->mask))
 177                         continue;
 178                 reported++;
 179                 if (!bad && quiet)
 180                         continue;
 181                 printf("%s %s: %s\n", descr, _(f->descr),
 182                                 bad ? _("unhealthy") : _("ok"));
 183         }
 184 }
 185
 186 /* Report on an AG's health. */
 187 static int
 188 report_ag_sick(
 189         xfs_agnumber_t          agno)
 190 {
 191         struct xfs_ag_geometry  ageo = { 0 };
 192         char                    descr[256];
 193         int                     ret;
 194
 195         ret = -xfrog_ag_geometry(file->xfd.fd, agno, &ageo);
 196         if (ret) {
 197                 xfrog_perror(ret, "ag_geometry");
 198                 return 1;
 199         }
 200         snprintf(descr, sizeof(descr) - 1, _("AG %u"), agno);
 201         report_sick(descr, ag_flags, ageo.ag_sick, ageo.ag_checked);
 202         return 0;
 203 }
 204
 205 /* Report on an inode's health. */
 206 static int
 207 report_inode_health(
 208         unsigned long long      ino,
 209         const char              *descr)
 210 {
 211         struct xfs_bulkstat     bs;
 212         char                    d[256];
 213         int                     ret;
 214
 215         if (!descr) {
 216                 snprintf(d, sizeof(d) - 1, _("inode %llu"), ino);
 217                 descr = d;
 218         }
 219
 220         ret = xfrog_bulkstat_single(&file->xfd, ino, 0, &bs);
 221         if (ret) {
 222                 xfrog_perror(ret, descr);
 223                 return 1;
 224         }
 225
 226         report_sick(descr, inode_flags, bs.bs_sick, bs.bs_checked);
 227         return 0;
 228 }
 229
 230 /* Report on a file's health. */
 231 static int
 232 report_file_health(
 233         const char      *path)
 234 {
 235         struct stat     stata, statb;
 236         int             ret;
 237
 238         ret = lstat(path, &statb);
 239         if (ret) {
 240                 perror(path);
 241                 return 1;
 242         }
 243
 244         ret = fstat(file->xfd.fd, &stata);
 245         if (ret) {
 246                 perror(file->name);
 247                 return 1;
 248         }
 249
 250         if (stata.st_dev != statb.st_dev) {
 251                 fprintf(stderr, _("%s: not on the open filesystem"), path);
 252                 return 1;
 253         }
 254
 255         return report_inode_health(statb.st_ino, path);
 256 }
 257
 258 #define BULKSTAT_NR             (128)
 259
 260 /*
 261  * Report on all files' health for a given @agno.  If @agno is NULLAGNUMBER,
 262  * report on all files in the filesystem.
 263  */
 264 static int
 265 report_bulkstat_health(
 266         xfs_agnumber_t          agno)
 267 {
 268         struct xfs_bulkstat_req *breq;
 269         char                    descr[256];
 270         uint32_t                i;
 271         int                     error;
 272
 273         breq = xfrog_bulkstat_alloc_req(BULKSTAT_NR, 0);
 274         if (!breq) {
 275                 perror("bulk alloc req");
 276                 exitcode = 1;
 277                 return 1;
 278         }
 279
 280         if (agno != NULLAGNUMBER)
 281                 xfrog_bulkstat_set_ag(breq, agno);
 282
 283         do {
 284                 error = xfrog_bulkstat(&file->xfd, breq);
 285                 if (error)
 286                         break;
 287                 for (i = 0; i < breq->hdr.ocount; i++) {
 288                         snprintf(descr, sizeof(descr) - 1, _("inode %"PRIu64),
 289                                         breq->bulkstat[i].bs_ino);
 290                         report_sick(descr, inode_flags,
 291                                         breq->bulkstat[i].bs_sick,
 292                                         breq->bulkstat[i].bs_checked);
 293                 }
 294         } while (breq->hdr.ocount > 0);
 295
 296         if (error)
 297                 xfrog_perror(error, "bulkstat");
 298
 299         free(breq);
 300         return error;
 301 }
 302
 303 #define OPT_STRING ("a:cfi:q")
 304
 305 /* Report on health problems in XFS filesystem. */
 306 static int
 307 health_f(
 308         int                     argc,
 309         char                    **argv)
 310 {
 311         unsigned long long      x;
 312         xfs_agnumber_t          agno;
 313         bool                    default_report = true;
 314         int                     c;
 315         int                     ret;
 316
 317         reported = 0;
 318
 319         if (file->xfd.fsgeom.version != XFS_FSOP_GEOM_VERSION_V5) {
 320                 perror("health");
 321                 return 1;
 322         }
 323
 324         /* Set our reporting options appropriately in the first pass. */
 325         while ((c = getopt(argc, argv, OPT_STRING)) != EOF) {
 326                 switch (c) {
 327                 case 'a':
 328                         default_report = false;
 329                         errno = 0;
 330                         x = strtoll(optarg, NULL, 10);
 331                         if (!errno && x >= NULLAGNUMBER)
 332                                 errno = ERANGE;
 333                         if (errno) {
 334                                 perror("ag health");
 335                                 return 1;
 336                         }
 337                         break;
 338                 case 'c':
 339                         comprehensive = true;
 340                         break;
 341                 case 'f':
 342                         default_report = false;
 343                         break;
 344                 case 'i':
 345                         default_report = false;
 346                         errno = 0;
 347                         x = strtoll(optarg, NULL, 10);
 348                         if (errno) {
 349                                 perror("inode health");
 350                                 return 1;
 351                         }
 352                         break;
 353                 case 'q':
 354                         quiet = true;
 355                         break;
 356                 default:
 357                         return command_usage(&health_cmd);
 358                 }
 359         }
 360         if (optind < argc)
 361                 default_report = false;
 362
 363         /* Reparse arguments, this time for reporting actions. */
 364         optind = 1;
 365         while ((c = getopt(argc, argv, OPT_STRING)) != EOF) {
 366                 switch (c) {
 367                 case 'a':
 368                         agno = strtoll(optarg, NULL, 10);
 369                         ret = report_ag_sick(agno);
 370                         if (!ret && comprehensive)
 371                                 ret = report_bulkstat_health(agno);
 372                         if (ret)
 373                                 return 1;
 374                         break;
 375                 case 'f':
 376                         report_sick(_("filesystem"), fs_flags,
 377                                         file->xfd.fsgeom.sick,
 378                                         file->xfd.fsgeom.checked);
 379                         if (comprehensive) {
 380                                 ret = report_bulkstat_health(NULLAGNUMBER);
 381                                 if (ret)
 382                                         return 1;
 383                         }
 384                         break;
 385                 case 'i':
 386                         x = strtoll(optarg, NULL, 10);
 387                         ret = report_inode_health(x, NULL);
 388                         if (ret)
 389                                 return 1;
 390                         break;
 391                 default:
 392                         break;
 393                 }
 394         }
 395
 396         for (c = optind; c < argc; c++) {
 397                 ret = report_file_health(argv[c]);
 398                 if (ret)
 399                         return 1;
 400         }
 401
 402         /* No arguments gets us a summary of fs state. */
 403         if (default_report) {
 404                 report_sick(_("filesystem"), fs_flags, file->xfd.fsgeom.sick,
 405                                 file->xfd.fsgeom.checked);
 406
 407                 for (agno = 0; agno < file->xfd.fsgeom.agcount; agno++) {
 408                         ret = report_ag_sick(agno);
 409                         if (ret)
 410                                 return 1;
 411                 }
 412                 if (comprehensive) {
 413                         ret = report_bulkstat_health(NULLAGNUMBER);
 414                         if (ret)
 415                                 return 1;
 416                 }
 417         }
 418
 419         if (!reported) {
 420                 fprintf(stderr,
 421 _("Health status has not been collected for this filesystem.\n"));
 422                 fprintf(stderr,
 423 _("Please run xfs_scrub(8) to remedy this situation.\n"));
 424         }
 425
 426         return 0;
 427 }
 428
 429 static void
 430 health_help(void)
 431 {
 432         printf(_(
 433 "\n"
 434 "Report all observed filesystem health problems.\n"
 435 "\n"
 436 " -a agno  -- Report health of the given allocation group.\n"
 437 " -c       -- Report on the health of all inodes.\n"
 438 " -f       -- Report health of the overall filesystem.\n"
 439 " -i inum  -- Report health of a given inode number.\n"
 440 " -q       -- Only report unhealthy metadata.\n"
 441 " paths    -- Report health of the given file path.\n"
 442 "\n"));
 443
 444 }
 445
 446 static cmdinfo_t health_cmd = {
 447         .name = "health",
 448         .cfunc = health_f,
 449         .argmin = 0,
 450         .argmax = -1,
 451         .args = "[-a agno] [-c] [-f] [-i inum] [-q] [paths]",
 452         .flags = CMD_FLAG_ONESHOT,
 453         .help = health_help,
 454 };
 455
 456 void
 457 health_init(void)
 458 {
 459         health_cmd.oneline = _("Report observed XFS health problems."),
 460         add_command(&health_cmd);
 461 }