scrub/phase1.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2018 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include <unistd.h>
   8 #include <sys/types.h>
   9 #include <sys/time.h>
  10 #include <sys/resource.h>
  11 #include <sys/statvfs.h>
  12 #include <fcntl.h>
  13 #include <dirent.h>
  14 #include <stdint.h>
  15 #include <pthread.h>
  16 #include "libfrog/util.h"
  17 #include "libfrog/workqueue.h"
  18 #include "input.h"
  19 #include "libfrog/paths.h"
  20 #include "handle.h"
  21 #include "bitops.h"
  22 #include "libfrog/avl64.h"
  23 #include "list.h"
  24 #include "xfs_scrub.h"
  25 #include "common.h"
  26 #include "disk.h"
  27 #include "scrub.h"
  28 #include "repair.h"
  29 #include "libfrog/fsgeom.h"
  30
  31 /* Phase 1: Find filesystem geometry (and clean up after) */
  32
  33 /* Shut down the filesystem. */
  34 void
  35 xfs_shutdown_fs(
  36         struct scrub_ctx                *ctx)
  37 {
  38         int                             flag;
  39
  40         flag = XFS_FSOP_GOING_FLAGS_LOGFLUSH;
  41         str_info(ctx, ctx->mntpoint, _("Shutting down filesystem!"));
  42         if (ioctl(ctx->mnt.fd, XFS_IOC_GOINGDOWN, &flag))
  43                 str_errno(ctx, ctx->mntpoint);
  44 }
  45
  46 /* Clean up the XFS-specific state data. */
  47 int
  48 scrub_cleanup(
  49         struct scrub_ctx        *ctx)
  50 {
  51         int                     error;
  52
  53         action_lists_free(&ctx->action_lists);
  54         if (ctx->fshandle)
  55                 free_handle(ctx->fshandle, ctx->fshandle_len);
  56         if (ctx->rtdev)
  57                 disk_close(ctx->rtdev);
  58         if (ctx->logdev)
  59                 disk_close(ctx->logdev);
  60         if (ctx->datadev)
  61                 disk_close(ctx->datadev);
  62         fshandle_destroy();
  63         error = -xfd_close(&ctx->mnt);
  64         if (error)
  65                 str_liberror(ctx, error, _("closing mountpoint fd"));
  66         fs_table_destroy();
  67
  68         return error;
  69 }
  70
  71 /*
  72  * Bind to the mountpoint, read the XFS geometry, bind to the block devices.
  73  * Anything we've already built will be cleaned up by scrub_cleanup.
  74  */
  75 int
  76 phase1_func(
  77         struct scrub_ctx                *ctx)
  78 {
  79         int                             error;
  80
  81         /*
  82          * Open the directory with O_NOATIME.  For mountpoints owned
  83          * by root, this should be sufficient to ensure that we have
  84          * CAP_SYS_ADMIN, which we probably need to do anything fancy
  85          * with the (XFS driver) kernel.
  86          */
  87         error = -xfd_open(&ctx->mnt, ctx->mntpoint,
  88                         O_RDONLY | O_NOATIME | O_DIRECTORY);
  89         if (error) {
  90                 if (error == EPERM)
  91                         str_error(ctx, ctx->mntpoint,
  92 _("Must be root to run scrub."));
  93                 else if (error == ENOTTY)
  94                         str_error(ctx, ctx->mntpoint,
  95 _("Not an XFS filesystem."));
  96                 else
  97                         str_liberror(ctx, error, ctx->mntpoint);
  98                 return error;
  99         }
 100
 101         error = fstat(ctx->mnt.fd, &ctx->mnt_sb);
 102         if (error) {
 103                 str_errno(ctx, ctx->mntpoint);
 104                 return error;
 105         }
 106         error = fstatvfs(ctx->mnt.fd, &ctx->mnt_sv);
 107         if (error) {
 108                 str_errno(ctx, ctx->mntpoint);
 109                 return error;
 110         }
 111         error = fstatfs(ctx->mnt.fd, &ctx->mnt_sf);
 112         if (error) {
 113                 str_errno(ctx, ctx->mntpoint);
 114                 return error;
 115         }
 116
 117         /*
 118          * Flush everything out to disk before we start checking.
 119          * This seems to reduce the incidence of stale file handle
 120          * errors when we open things by handle.
 121          */
 122         error = syncfs(ctx->mnt.fd);
 123         if (error) {
 124                 str_errno(ctx, ctx->mntpoint);
 125                 return error;
 126         }
 127
 128         error = action_lists_alloc(ctx->mnt.fsgeom.agcount,
 129                         &ctx->action_lists);
 130         if (error) {
 131                 str_liberror(ctx, error, _("allocating action lists"));
 132                 return error;
 133         }
 134
 135         error = path_to_fshandle(ctx->mntpoint, &ctx->fshandle,
 136                         &ctx->fshandle_len);
 137         if (error) {
 138                 str_errno(ctx, _("getting fshandle"));
 139                 return error;
 140         }
 141
 142         /* Do we have kernel-assisted metadata scrubbing? */
 143         if (!can_scrub_fs_metadata(ctx) || !can_scrub_inode(ctx) ||
 144             !can_scrub_bmap(ctx) || !can_scrub_dir(ctx) ||
 145             !can_scrub_attr(ctx) || !can_scrub_symlink(ctx) ||
 146             !can_scrub_parent(ctx)) {
 147                 str_error(ctx, ctx->mntpoint,
 148 _("Kernel metadata scrubbing facility is not available."));
 149                 return ECANCELED;
 150         }
 151
 152         /* Do we need kernel-assisted metadata repair? */
 153         if (ctx->mode != SCRUB_MODE_DRY_RUN && !xfs_can_repair(ctx)) {
 154                 str_error(ctx, ctx->mntpoint,
 155 _("Kernel metadata repair facility is not available.  Use -n to scrub."));
 156                 return ECANCELED;
 157         }
 158
 159         /* Did we find the log and rt devices, if they're present? */
 160         if (ctx->mnt.fsgeom.logstart == 0 && ctx->fsinfo.fs_log == NULL) {
 161                 str_error(ctx, ctx->mntpoint,
 162 _("Unable to find log device path."));
 163                 return ECANCELED;
 164         }
 165         if (ctx->mnt.fsgeom.rtblocks && ctx->fsinfo.fs_rt == NULL) {
 166                 str_error(ctx, ctx->mntpoint,
 167 _("Unable to find realtime device path."));
 168                 return ECANCELED;
 169         }
 170
 171         /* Open the raw devices. */
 172         ctx->datadev = disk_open(ctx->fsinfo.fs_name);
 173         if (!ctx->datadev) {
 174                 str_error(ctx, ctx->mntpoint, _("Unable to open data device."));
 175                 return ECANCELED;
 176         }
 177
 178         ctx->nr_io_threads = disk_heads(ctx->datadev);
 179         if (verbose) {
 180                 fprintf(stdout, _("%s: using %d threads to scrub.\n"),
 181                                 ctx->mntpoint, scrub_nproc(ctx));
 182                 fflush(stdout);
 183         }
 184
 185         if (ctx->fsinfo.fs_log) {
 186                 ctx->logdev = disk_open(ctx->fsinfo.fs_log);
 187                 if (!ctx->logdev) {
 188                         str_error(ctx, ctx->mntpoint,
 189                                 _("Unable to open external log device."));
 190                         return ECANCELED;
 191                 }
 192         }
 193         if (ctx->fsinfo.fs_rt) {
 194                 ctx->rtdev = disk_open(ctx->fsinfo.fs_rt);
 195                 if (!ctx->rtdev) {
 196                         str_error(ctx, ctx->mntpoint,
 197                                 _("Unable to open realtime device."));
 198                         return ECANCELED;
 199                 }
 200         }
 201
 202         /*
 203          * Everything's set up, which means any failures recorded after
 204          * this point are most probably corruption errors (as opposed to
 205          * purely setup errors).
 206          */
 207         log_info(ctx, _("Invoking online scrub."), ctx);
 208         ctx->scrub_setup_succeeded = true;
 209         return 0;
 210 }