319 lines · c
1// SPDX-License-Identifier: GPL-2.0-or-later2/*3 * Copyright (C) 2019-2023 Oracle. All Rights Reserved.4 * Author: Darrick J. Wong <djwong@kernel.org>5 */6#include "xfs.h"7#include "xfs_fs.h"8#include "xfs_shared.h"9#include "xfs_format.h"10#include "xfs_trans_resv.h"11#include "xfs_mount.h"12#include "xfs_btree.h"13#include "xfs_ag.h"14#include "xfs_health.h"15#include "scrub/scrub.h"16#include "scrub/health.h"17#include "scrub/common.h"18 19/*20 * Scrub and In-Core Filesystem Health Assessments21 * ===============================================22 *23 * Online scrub and repair have the time and the ability to perform stronger24 * checks than we can do from the metadata verifiers, because they can25 * cross-reference records between data structures. Therefore, scrub is in a26 * good position to update the online filesystem health assessments to reflect27 * the good/bad state of the data structure.28 *29 * We therefore extend scrub in the following ways to achieve this:30 *31 * 1. Create a "sick_mask" field in the scrub context. When we're setting up a32 * scrub call, set this to the default XFS_SICK_* flag(s) for the selected33 * scrub type (call it A). Scrub and repair functions can override the default34 * sick_mask value if they choose.35 *36 * 2. If the scrubber returns a runtime error code, we exit making no changes37 * to the incore sick state.38 *39 * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore40 * sick flags before exiting.41 *42 * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore43 * sick flags. If the user didn't want to repair then we exit, leaving the44 * metadata structure unfixed and the sick flag set.45 *46 * 5. Now we know that A is corrupt and the user wants to repair, so run the47 * repairer. If the repairer returns an error code, we exit with that error48 * code, having made no further changes to the incore sick state.49 *50 * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean,51 * use sick_mask to clear the incore sick flags. This should have the effect52 * that A is no longer marked sick.53 *54 * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and55 * use sick_mask to set the incore sick flags. This should have no externally56 * visible effect since we already set them in step (4).57 *58 * There are some complications to this story, however. For certain types of59 * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild60 * both structures at the same time. The following principles apply to this61 * type of repair strategy:62 *63 * 8. Any repair function that rebuilds multiple structures should update64 * sick_mask_visible to reflect whatever other structures are rebuilt, and65 * verify that all the rebuilt structures can pass a scrub check. The outcomes66 * of 5-7 still apply, but with a sick_mask that covers everything being67 * rebuilt.68 */69 70/* Map our scrub type to a sick mask and a set of health update functions. */71 72enum xchk_health_group {73 XHG_FS = 1,74 XHG_RT,75 XHG_AG,76 XHG_INO,77};78 79struct xchk_health_map {80 enum xchk_health_group group;81 unsigned int sick_mask;82};83 84static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {85 [XFS_SCRUB_TYPE_SB] = { XHG_AG, XFS_SICK_AG_SB },86 [XFS_SCRUB_TYPE_AGF] = { XHG_AG, XFS_SICK_AG_AGF },87 [XFS_SCRUB_TYPE_AGFL] = { XHG_AG, XFS_SICK_AG_AGFL },88 [XFS_SCRUB_TYPE_AGI] = { XHG_AG, XFS_SICK_AG_AGI },89 [XFS_SCRUB_TYPE_BNOBT] = { XHG_AG, XFS_SICK_AG_BNOBT },90 [XFS_SCRUB_TYPE_CNTBT] = { XHG_AG, XFS_SICK_AG_CNTBT },91 [XFS_SCRUB_TYPE_INOBT] = { XHG_AG, XFS_SICK_AG_INOBT },92 [XFS_SCRUB_TYPE_FINOBT] = { XHG_AG, XFS_SICK_AG_FINOBT },93 [XFS_SCRUB_TYPE_RMAPBT] = { XHG_AG, XFS_SICK_AG_RMAPBT },94 [XFS_SCRUB_TYPE_REFCNTBT] = { XHG_AG, XFS_SICK_AG_REFCNTBT },95 [XFS_SCRUB_TYPE_INODE] = { XHG_INO, XFS_SICK_INO_CORE },96 [XFS_SCRUB_TYPE_BMBTD] = { XHG_INO, XFS_SICK_INO_BMBTD },97 [XFS_SCRUB_TYPE_BMBTA] = { XHG_INO, XFS_SICK_INO_BMBTA },98 [XFS_SCRUB_TYPE_BMBTC] = { XHG_INO, XFS_SICK_INO_BMBTC },99 [XFS_SCRUB_TYPE_DIR] = { XHG_INO, XFS_SICK_INO_DIR },100 [XFS_SCRUB_TYPE_XATTR] = { XHG_INO, XFS_SICK_INO_XATTR },101 [XFS_SCRUB_TYPE_SYMLINK] = { XHG_INO, XFS_SICK_INO_SYMLINK },102 [XFS_SCRUB_TYPE_PARENT] = { XHG_INO, XFS_SICK_INO_PARENT },103 [XFS_SCRUB_TYPE_RTBITMAP] = { XHG_RT, XFS_SICK_RT_BITMAP },104 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RT, XFS_SICK_RT_SUMMARY },105 [XFS_SCRUB_TYPE_UQUOTA] = { XHG_FS, XFS_SICK_FS_UQUOTA },106 [XFS_SCRUB_TYPE_GQUOTA] = { XHG_FS, XFS_SICK_FS_GQUOTA },107 [XFS_SCRUB_TYPE_PQUOTA] = { XHG_FS, XFS_SICK_FS_PQUOTA },108 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XHG_FS, XFS_SICK_FS_COUNTERS },109 [XFS_SCRUB_TYPE_QUOTACHECK] = { XHG_FS, XFS_SICK_FS_QUOTACHECK },110 [XFS_SCRUB_TYPE_NLINKS] = { XHG_FS, XFS_SICK_FS_NLINKS },111 [XFS_SCRUB_TYPE_DIRTREE] = { XHG_INO, XFS_SICK_INO_DIRTREE },112};113 114/* Return the health status mask for this scrub type. */115unsigned int116xchk_health_mask_for_scrub_type(117 __u32 scrub_type)118{119 return type_to_health_flag[scrub_type].sick_mask;120}121 122/*123 * If the scrub state is clean, add @mask to the scrub sick mask to clear124 * additional sick flags from the metadata object's sick state.125 */126void127xchk_mark_healthy_if_clean(128 struct xfs_scrub *sc,129 unsigned int mask)130{131 if (!(sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |132 XFS_SCRUB_OFLAG_XCORRUPT)))133 sc->sick_mask |= mask;134}135 136/*137 * If we're scrubbing a piece of file metadata for the first time, does it look138 * like it has been zapped? Skip the check if we just repaired the metadata139 * and are revalidating it.140 */141bool142xchk_file_looks_zapped(143 struct xfs_scrub *sc,144 unsigned int mask)145{146 ASSERT((mask & ~XFS_SICK_INO_ZAPPED) == 0);147 148 if (sc->flags & XREP_ALREADY_FIXED)149 return false;150 151 return xfs_inode_has_sickness(sc->ip, mask);152}153 154/*155 * Scrub gave the filesystem a clean bill of health, so clear all the indirect156 * markers of past problems (at least for the fs and ags) so that we can be157 * healthy again.158 */159STATIC void160xchk_mark_all_healthy(161 struct xfs_mount *mp)162{163 struct xfs_perag *pag;164 xfs_agnumber_t agno;165 166 xfs_fs_mark_healthy(mp, XFS_SICK_FS_INDIRECT);167 xfs_rt_mark_healthy(mp, XFS_SICK_RT_INDIRECT);168 for_each_perag(mp, agno, pag)169 xfs_ag_mark_healthy(pag, XFS_SICK_AG_INDIRECT);170}171 172/*173 * Update filesystem health assessments based on what we found and did.174 *175 * If the scrubber finds errors, we mark sick whatever's mentioned in176 * sick_mask, no matter whether this is a first scan or an177 * evaluation of repair effectiveness.178 *179 * Otherwise, no direct corruption was found, so mark whatever's in180 * sick_mask as healthy.181 */182void183xchk_update_health(184 struct xfs_scrub *sc)185{186 struct xfs_perag *pag;187 bool bad;188 189 /*190 * The HEALTHY scrub type is a request from userspace to clear all the191 * indirect flags after a clean scan of the entire filesystem. As such192 * there's no sick flag defined for it, so we branch here ahead of the193 * mask check.194 */195 if (sc->sm->sm_type == XFS_SCRUB_TYPE_HEALTHY &&196 !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {197 xchk_mark_all_healthy(sc->mp);198 return;199 }200 201 if (!sc->sick_mask)202 return;203 204 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |205 XFS_SCRUB_OFLAG_XCORRUPT));206 switch (type_to_health_flag[sc->sm->sm_type].group) {207 case XHG_AG:208 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);209 if (bad)210 xfs_ag_mark_corrupt(pag, sc->sick_mask);211 else212 xfs_ag_mark_healthy(pag, sc->sick_mask);213 xfs_perag_put(pag);214 break;215 case XHG_INO:216 if (!sc->ip)217 return;218 if (bad) {219 unsigned int mask = sc->sick_mask;220 221 /*222 * If we're coming in for repairs then we don't want223 * sickness flags to propagate to the incore health224 * status if the inode gets inactivated before we can225 * fix it.226 */227 if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)228 mask |= XFS_SICK_INO_FORGET;229 xfs_inode_mark_corrupt(sc->ip, mask);230 } else231 xfs_inode_mark_healthy(sc->ip, sc->sick_mask);232 break;233 case XHG_FS:234 if (bad)235 xfs_fs_mark_corrupt(sc->mp, sc->sick_mask);236 else237 xfs_fs_mark_healthy(sc->mp, sc->sick_mask);238 break;239 case XHG_RT:240 if (bad)241 xfs_rt_mark_corrupt(sc->mp, sc->sick_mask);242 else243 xfs_rt_mark_healthy(sc->mp, sc->sick_mask);244 break;245 default:246 ASSERT(0);247 break;248 }249}250 251/* Is the given per-AG btree healthy enough for scanning? */252void253xchk_ag_btree_del_cursor_if_sick(254 struct xfs_scrub *sc,255 struct xfs_btree_cur **curp,256 unsigned int sm_type)257{258 unsigned int mask = (*curp)->bc_ops->sick_mask;259 260 /*261 * We always want the cursor if it's the same type as whatever we're262 * scrubbing, even if we already know the structure is corrupt.263 *264 * Otherwise, we're only interested in the btree for cross-referencing.265 * If we know the btree is bad then don't bother, just set XFAIL.266 */267 if (sc->sm->sm_type == sm_type)268 return;269 270 /*271 * If we just repaired some AG metadata, sc->sick_mask will reflect all272 * the per-AG metadata types that were repaired. Exclude these from273 * the filesystem health query because we have not yet updated the274 * health status and we want everything to be scanned.275 */276 if ((sc->flags & XREP_ALREADY_FIXED) &&277 type_to_health_flag[sc->sm->sm_type].group == XHG_AG)278 mask &= ~sc->sick_mask;279 280 if (xfs_ag_has_sickness((*curp)->bc_ag.pag, mask)) {281 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;282 xfs_btree_del_cursor(*curp, XFS_BTREE_NOERROR);283 *curp = NULL;284 }285}286 287/*288 * Quick scan to double-check that there isn't any evidence of lingering289 * primary health problems. If we're still clear, then the health update will290 * take care of clearing the indirect evidence.291 */292int293xchk_health_record(294 struct xfs_scrub *sc)295{296 struct xfs_mount *mp = sc->mp;297 struct xfs_perag *pag;298 xfs_agnumber_t agno;299 300 unsigned int sick;301 unsigned int checked;302 303 xfs_fs_measure_sickness(mp, &sick, &checked);304 if (sick & XFS_SICK_FS_PRIMARY)305 xchk_set_corrupt(sc);306 307 xfs_rt_measure_sickness(mp, &sick, &checked);308 if (sick & XFS_SICK_RT_PRIMARY)309 xchk_set_corrupt(sc);310 311 for_each_perag(mp, agno, pag) {312 xfs_ag_measure_sickness(pag, &sick, &checked);313 if (sick & XFS_SICK_AG_PRIMARY)314 xchk_set_corrupt(sc);315 }316 317 return 0;318}319