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.h"
  17 #include "workqueue.h"
  18 #include "input.h"
  19 #include "path.h"
  20 #include "handle.h"
  21 #include "bitops.h"
  22 #include "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
  30 /* Phase 1: Find filesystem geometry (and clean up after) */
  31
  32 /* Shut down the filesystem. */
  33 void
  34 xfs_shutdown_fs(
  35         struct scrub_ctx                *ctx)
  36 {
  37         int                             flag;
  38
  39         flag = XFS_FSOP_GOING_FLAGS_LOGFLUSH;
  40         str_info(ctx, ctx->mntpoint, _("Shutting down filesystem!"));
  41         if (ioctl(ctx->mnt_fd, XFS_IOC_GOINGDOWN, &flag))
  42                 str_errno(ctx, ctx->mntpoint);
  43 }
  44
  45 /* Clean up the XFS-specific state data. */
  46 bool
  47 xfs_cleanup_fs(
  48         struct scrub_ctx        *ctx)
  49 {
  50         int                     error;
  51
  52         xfs_action_lists_free(&ctx->action_lists);
  53         if (ctx->fshandle)
  54                 free_handle(ctx->fshandle, ctx->fshandle_len);
  55         if (ctx->rtdev)
  56                 disk_close(ctx->rtdev);
  57         if (ctx->logdev)
  58                 disk_close(ctx->logdev);
  59         if (ctx->datadev)
  60                 disk_close(ctx->datadev);
  61         fshandle_destroy();
  62         error = close(ctx->mnt_fd);
  63         if (error)
  64                 str_errno(ctx, _("closing mountpoint fd"));
  65         fs_table_destroy();
  66
  67         return true;
  68 }
  69
  70 /*
  71  * Bind to the mountpoint, read the XFS geometry, bind to the block devices.
  72  * Anything we've already built will be cleaned up by xfs_cleanup_fs.
  73  */
  74 bool
  75 xfs_setup_fs(
  76         struct scrub_ctx                *ctx)
  77 {
  78         int                             error;
  79
  80         /*
  81          * Open the directory with O_NOATIME.  For mountpoints owned
  82          * by root, this should be sufficient to ensure that we have
  83          * CAP_SYS_ADMIN, which we probably need to do anything fancy
  84          * with the (XFS driver) kernel.
  85          */
  86         ctx->mnt_fd = open(ctx->mntpoint, O_RDONLY | O_NOATIME | O_DIRECTORY);
  87         if (ctx->mnt_fd < 0) {
  88                 if (errno == EPERM)
  89                         str_info(ctx, ctx->mntpoint,
  90 _("Must be root to run scrub."));
  91                 else
  92                         str_errno(ctx, ctx->mntpoint);
  93                 return false;
  94         }
  95
  96         error = fstat(ctx->mnt_fd, &ctx->mnt_sb);
  97         if (error) {
  98                 str_errno(ctx, ctx->mntpoint);
  99                 return false;
 100         }
 101         error = fstatvfs(ctx->mnt_fd, &ctx->mnt_sv);
 102         if (error) {
 103                 str_errno(ctx, ctx->mntpoint);
 104                 return false;
 105         }
 106         error = fstatfs(ctx->mnt_fd, &ctx->mnt_sf);
 107         if (error) {
 108                 str_errno(ctx, ctx->mntpoint);
 109                 return false;
 110         }
 111
 112         if (!platform_test_xfs_fd(ctx->mnt_fd)) {
 113                 str_info(ctx, ctx->mntpoint,
 114 _("Does not appear to be an XFS filesystem!"));
 115                 return false;
 116         }
 117
 118         /*
 119          * Flush everything out to disk before we start checking.
 120          * This seems to reduce the incidence of stale file handle
 121          * errors when we open things by handle.
 122          */
 123         error = syncfs(ctx->mnt_fd);
 124         if (error) {
 125                 str_errno(ctx, ctx->mntpoint);
 126                 return false;
 127         }
 128
 129         /* Retrieve XFS geometry. */
 130         error = ioctl(ctx->mnt_fd, XFS_IOC_FSGEOMETRY, &ctx->geo);
 131         if (error) {
 132                 str_errno(ctx, ctx->mntpoint);
 133                 return false;
 134         }
 135
 136         if (!xfs_action_lists_alloc(ctx->geo.agcount, &ctx->action_lists)) {
 137                 str_error(ctx, ctx->mntpoint, _("Not enough memory."));
 138                 return false;
 139         }
 140
 141         ctx->agblklog = log2_roundup(ctx->geo.agblocks);
 142         ctx->blocklog = highbit32(ctx->geo.blocksize);
 143         ctx->inodelog = highbit32(ctx->geo.inodesize);
 144         ctx->inopblog = ctx->blocklog - ctx->inodelog;
 145
 146         error = path_to_fshandle(ctx->mntpoint, &ctx->fshandle,
 147                         &ctx->fshandle_len);
 148         if (error) {
 149                 str_errno(ctx, _("getting fshandle"));
 150                 return false;
 151         }
 152
 153         /* Do we have kernel-assisted metadata scrubbing? */
 154         if (!xfs_can_scrub_fs_metadata(ctx) || !xfs_can_scrub_inode(ctx) ||
 155             !xfs_can_scrub_bmap(ctx) || !xfs_can_scrub_dir(ctx) ||
 156             !xfs_can_scrub_attr(ctx) || !xfs_can_scrub_symlink(ctx) ||
 157             !xfs_can_scrub_parent(ctx)) {
 158                 str_info(ctx, ctx->mntpoint,
 159 _("Kernel metadata scrubbing facility is not available."));
 160                 return false;
 161         }
 162
 163         /* Do we need kernel-assisted metadata repair? */
 164         if (ctx->mode != SCRUB_MODE_DRY_RUN && !xfs_can_repair(ctx)) {
 165                 str_info(ctx, ctx->mntpoint,
 166 _("Kernel metadata repair facility is not available.  Use -n to scrub."));
 167                 return false;
 168         }
 169
 170         /* Did we find the log and rt devices, if they're present? */
 171         if (ctx->geo.logstart == 0 && ctx->fsinfo.fs_log == NULL) {
 172                 str_info(ctx, ctx->mntpoint,
 173 _("Unable to find log device path."));
 174                 return false;
 175         }
 176         if (ctx->geo.rtblocks && ctx->fsinfo.fs_rt == NULL) {
 177                 str_info(ctx, ctx->mntpoint,
 178 _("Unable to find realtime device path."));
 179                 return false;
 180         }
 181
 182         /* Open the raw devices. */
 183         ctx->datadev = disk_open(ctx->fsinfo.fs_name);
 184         if (error) {
 185                 str_errno(ctx, ctx->fsinfo.fs_name);
 186                 return false;
 187         }
 188
 189         ctx->nr_io_threads = disk_heads(ctx->datadev);
 190         if (verbose) {
 191                 fprintf(stdout, _("%s: using %d threads to scrub.\n"),
 192                                 ctx->mntpoint, scrub_nproc(ctx));
 193                 fflush(stdout);
 194         }
 195
 196         if (ctx->fsinfo.fs_log) {
 197                 ctx->logdev = disk_open(ctx->fsinfo.fs_log);
 198                 if (error) {
 199                         str_errno(ctx, ctx->fsinfo.fs_name);
 200                         return false;
 201                 }
 202         }
 203         if (ctx->fsinfo.fs_rt) {
 204                 ctx->rtdev = disk_open(ctx->fsinfo.fs_rt);
 205                 if (error) {
 206                         str_errno(ctx, ctx->fsinfo.fs_name);
 207                         return false;
 208                 }
 209         }
 210
 211         /*
 212          * Everything's set up, which means any failures recorded after
 213          * this point are most probably corruption errors (as opposed to
 214          * purely setup errors).
 215          */
 216         log_info(ctx, _("Invoking online scrub."), ctx);
 217         ctx->scrub_setup_succeeded = true;
 218         return true;
 219 }