1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 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 Assessments 21 * =============================================== 22 * 23 * Online scrub and repair have the time and the ability to perform stronger 24 * checks than we can do from the metadata verifiers, because they can 25 * cross-reference records between data structures. Therefore, scrub is in a 26 * good position to update the online filesystem health assessments to reflect 27 * 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 a 32 * scrub call, set this to the default XFS_SICK_* flag(s) for the selected 33 * scrub type (call it A). Scrub and repair functions can override the default 34 * sick_mask value if they choose. 35 * 36 * 2. If the scrubber returns a runtime error code, we exit making no changes 37 * to the incore sick state. 38 * 39 * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore 40 * sick flags before exiting. 41 * 42 * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore 43 * sick flags. If the user didn't want to repair then we exit, leaving the 44 * 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 the 47 * repairer. If the repairer returns an error code, we exit with that error 48 * 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 effect 52 * that A is no longer marked sick. 53 * 54 * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and 55 * use sick_mask to set the incore sick flags. This should have no externally 56 * visible effect since we already set them in step (4). 57 * 58 * There are some complications to this story, however. For certain types of 59 * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild 60 * both structures at the same time. The following principles apply to this 61 * type of repair strategy: 62 * 63 * 8. Any repair function that rebuilds multiple structures should update 64 * sick_mask_visible to reflect whatever other structures are rebuilt, and 65 * verify that all the rebuilt structures can pass a scrub check. The outcomes 66 * of 5-7 still apply, but with a sick_mask that covers everything being 67 * 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}; 112 113/* Return the health status mask for this scrub type. */ 114unsigned int 115xchk_health_mask_for_scrub_type( 116 __u32 scrub_type) 117{ 118 return type_to_health_flag[scrub_type].sick_mask; 119} 120 121/* 122 * If the scrub state is clean, add @mask to the scrub sick mask to clear 123 * additional sick flags from the metadata object's sick state. 124 */ 125void 126xchk_mark_healthy_if_clean( 127 struct xfs_scrub *sc, 128 unsigned int mask) 129{ 130 if (!(sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT | 131 XFS_SCRUB_OFLAG_XCORRUPT))) 132 sc->sick_mask |= mask; 133} 134 135/* 136 * If we're scrubbing a piece of file metadata for the first time, does it look 137 * like it has been zapped? Skip the check if we just repaired the metadata 138 * and are revalidating it. 139 */ 140bool 141xchk_file_looks_zapped( 142 struct xfs_scrub *sc, 143 unsigned int mask) 144{ 145 ASSERT((mask & ~XFS_SICK_INO_ZAPPED) == 0); 146 147 if (sc->flags & XREP_ALREADY_FIXED) 148 return false; 149 150 return xfs_inode_has_sickness(sc->ip, mask); 151} 152 153/* 154 * Scrub gave the filesystem a clean bill of health, so clear all the indirect 155 * markers of past problems (at least for the fs and ags) so that we can be 156 * healthy again. 157 */ 158STATIC void 159xchk_mark_all_healthy( 160 struct xfs_mount *mp) 161{ 162 struct xfs_perag *pag; 163 xfs_agnumber_t agno; 164 165 xfs_fs_mark_healthy(mp, XFS_SICK_FS_INDIRECT); 166 xfs_rt_mark_healthy(mp, XFS_SICK_RT_INDIRECT); 167 for_each_perag(mp, agno, pag) 168 xfs_ag_mark_healthy(pag, XFS_SICK_AG_INDIRECT); 169} 170 171/* 172 * Update filesystem health assessments based on what we found and did. 173 * 174 * If the scrubber finds errors, we mark sick whatever's mentioned in 175 * sick_mask, no matter whether this is a first scan or an 176 * evaluation of repair effectiveness. 177 * 178 * Otherwise, no direct corruption was found, so mark whatever's in 179 * sick_mask as healthy. 180 */ 181void 182xchk_update_health( 183 struct xfs_scrub *sc) 184{ 185 struct xfs_perag *pag; 186 bool bad; 187 188 /* 189 * The HEALTHY scrub type is a request from userspace to clear all the 190 * indirect flags after a clean scan of the entire filesystem. As such 191 * there's no sick flag defined for it, so we branch here ahead of the 192 * mask check. 193 */ 194 if (sc->sm->sm_type == XFS_SCRUB_TYPE_HEALTHY && 195 !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) { 196 xchk_mark_all_healthy(sc->mp); 197 return; 198 } 199 200 if (!sc->sick_mask) 201 return; 202 203 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT | 204 XFS_SCRUB_OFLAG_XCORRUPT)); 205 switch (type_to_health_flag[sc->sm->sm_type].group) { 206 case XHG_AG: 207 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno); 208 if (bad) 209 xfs_ag_mark_corrupt(pag, sc->sick_mask); 210 else 211 xfs_ag_mark_healthy(pag, sc->sick_mask); 212 xfs_perag_put(pag); 213 break; 214 case XHG_INO: 215 if (!sc->ip) 216 return; 217 if (bad) { 218 unsigned int mask = sc->sick_mask; 219 220 /* 221 * If we're coming in for repairs then we don't want 222 * sickness flags to propagate to the incore health 223 * status if the inode gets inactivated before we can 224 * fix it. 225 */ 226 if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) 227 mask |= XFS_SICK_INO_FORGET; 228 xfs_inode_mark_corrupt(sc->ip, mask); 229 } else 230 xfs_inode_mark_healthy(sc->ip, sc->sick_mask); 231 break; 232 case XHG_FS: 233 if (bad) 234 xfs_fs_mark_corrupt(sc->mp, sc->sick_mask); 235 else 236 xfs_fs_mark_healthy(sc->mp, sc->sick_mask); 237 break; 238 case XHG_RT: 239 if (bad) 240 xfs_rt_mark_corrupt(sc->mp, sc->sick_mask); 241 else 242 xfs_rt_mark_healthy(sc->mp, sc->sick_mask); 243 break; 244 default: 245 ASSERT(0); 246 break; 247 } 248} 249 250/* Is the given per-AG btree healthy enough for scanning? */ 251void 252xchk_ag_btree_del_cursor_if_sick( 253 struct xfs_scrub *sc, 254 struct xfs_btree_cur **curp, 255 unsigned int sm_type) 256{ 257 unsigned int mask = (*curp)->bc_ops->sick_mask; 258 259 /* 260 * We always want the cursor if it's the same type as whatever we're 261 * scrubbing, even if we already know the structure is corrupt. 262 * 263 * Otherwise, we're only interested in the btree for cross-referencing. 264 * If we know the btree is bad then don't bother, just set XFAIL. 265 */ 266 if (sc->sm->sm_type == sm_type) 267 return; 268 269 /* 270 * If we just repaired some AG metadata, sc->sick_mask will reflect all 271 * the per-AG metadata types that were repaired. Exclude these from 272 * the filesystem health query because we have not yet updated the 273 * health status and we want everything to be scanned. 274 */ 275 if ((sc->flags & XREP_ALREADY_FIXED) && 276 type_to_health_flag[sc->sm->sm_type].group == XHG_AG) 277 mask &= ~sc->sick_mask; 278 279 if (xfs_ag_has_sickness((*curp)->bc_ag.pag, mask)) { 280 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL; 281 xfs_btree_del_cursor(*curp, XFS_BTREE_NOERROR); 282 *curp = NULL; 283 } 284} 285 286/* 287 * Quick scan to double-check that there isn't any evidence of lingering 288 * primary health problems. If we're still clear, then the health update will 289 * take care of clearing the indirect evidence. 290 */ 291int 292xchk_health_record( 293 struct xfs_scrub *sc) 294{ 295 struct xfs_mount *mp = sc->mp; 296 struct xfs_perag *pag; 297 xfs_agnumber_t agno; 298 299 unsigned int sick; 300 unsigned int checked; 301 302 xfs_fs_measure_sickness(mp, &sick, &checked); 303 if (sick & XFS_SICK_FS_PRIMARY) 304 xchk_set_corrupt(sc); 305 306 xfs_rt_measure_sickness(mp, &sick, &checked); 307 if (sick & XFS_SICK_RT_PRIMARY) 308 xchk_set_corrupt(sc); 309 310 for_each_perag(mp, agno, pag) { 311 xfs_ag_measure_sickness(pag, &sick, &checked); 312 if (sick & XFS_SICK_AG_PRIMARY) 313 xchk_set_corrupt(sc); 314 } 315 316 return 0; 317} 318