1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2019 Oracle. All Rights Reserved. 4 * Author: Darrick J. Wong <darrick.wong@oracle.com> 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_defer.h" 13 #include "xfs_btree.h" 14 #include "xfs_bit.h" 15 #include "xfs_log_format.h" 16 #include "xfs_trans.h" 17 #include "xfs_sb.h" 18 #include "xfs_inode.h" 19 #include "xfs_health.h" 20 #include "scrub/scrub.h" 21 #include "scrub/health.h" 22 23 /* 24 * Scrub and In-Core Filesystem Health Assessments 25 * =============================================== 26 * 27 * Online scrub and repair have the time and the ability to perform stronger 28 * checks than we can do from the metadata verifiers, because they can 29 * cross-reference records between data structures. Therefore, scrub is in a 30 * good position to update the online filesystem health assessments to reflect 31 * the good/bad state of the data structure. 32 * 33 * We therefore extend scrub in the following ways to achieve this: 34 * 35 * 1. Create a "sick_mask" field in the scrub context. When we're setting up a 36 * scrub call, set this to the default XFS_SICK_* flag(s) for the selected 37 * scrub type (call it A). Scrub and repair functions can override the default 38 * sick_mask value if they choose. 39 * 40 * 2. If the scrubber returns a runtime error code, we exit making no changes 41 * to the incore sick state. 42 * 43 * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore 44 * sick flags before exiting. 45 * 46 * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore 47 * sick flags. If the user didn't want to repair then we exit, leaving the 48 * metadata structure unfixed and the sick flag set. 49 * 50 * 5. Now we know that A is corrupt and the user wants to repair, so run the 51 * repairer. If the repairer returns an error code, we exit with that error 52 * code, having made no further changes to the incore sick state. 53 * 54 * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean, 55 * use sick_mask to clear the incore sick flags. This should have the effect 56 * that A is no longer marked sick. 57 * 58 * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and 59 * use sick_mask to set the incore sick flags. This should have no externally 60 * visible effect since we already set them in step (4). 61 * 62 * There are some complications to this story, however. For certain types of 63 * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild 64 * both structures at the same time. The following principles apply to this 65 * type of repair strategy: 66 * 67 * 8. Any repair function that rebuilds multiple structures should update 68 * sick_mask_visible to reflect whatever other structures are rebuilt, and 69 * verify that all the rebuilt structures can pass a scrub check. The outcomes 70 * of 5-7 still apply, but with a sick_mask that covers everything being 71 * rebuilt. 72 */ 73 74 /* Map our scrub type to a sick mask and a set of health update functions. */ 75 76 enum xchk_health_group { 77 XHG_FS = 1, 78 XHG_RT, 79 XHG_AG, 80 XHG_INO, 81 }; 82 83 struct xchk_health_map { 84 enum xchk_health_group group; 85 unsigned int sick_mask; 86 }; 87 88 static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = { 89 [XFS_SCRUB_TYPE_SB] = { XHG_AG, XFS_SICK_AG_SB }, 90 [XFS_SCRUB_TYPE_AGF] = { XHG_AG, XFS_SICK_AG_AGF }, 91 [XFS_SCRUB_TYPE_AGFL] = { XHG_AG, XFS_SICK_AG_AGFL }, 92 [XFS_SCRUB_TYPE_AGI] = { XHG_AG, XFS_SICK_AG_AGI }, 93 [XFS_SCRUB_TYPE_BNOBT] = { XHG_AG, XFS_SICK_AG_BNOBT }, 94 [XFS_SCRUB_TYPE_CNTBT] = { XHG_AG, XFS_SICK_AG_CNTBT }, 95 [XFS_SCRUB_TYPE_INOBT] = { XHG_AG, XFS_SICK_AG_INOBT }, 96 [XFS_SCRUB_TYPE_FINOBT] = { XHG_AG, XFS_SICK_AG_FINOBT }, 97 [XFS_SCRUB_TYPE_RMAPBT] = { XHG_AG, XFS_SICK_AG_RMAPBT }, 98 [XFS_SCRUB_TYPE_REFCNTBT] = { XHG_AG, XFS_SICK_AG_REFCNTBT }, 99 [XFS_SCRUB_TYPE_INODE] = { XHG_INO, XFS_SICK_INO_CORE }, 100 [XFS_SCRUB_TYPE_BMBTD] = { XHG_INO, XFS_SICK_INO_BMBTD }, 101 [XFS_SCRUB_TYPE_BMBTA] = { XHG_INO, XFS_SICK_INO_BMBTA }, 102 [XFS_SCRUB_TYPE_BMBTC] = { XHG_INO, XFS_SICK_INO_BMBTC }, 103 [XFS_SCRUB_TYPE_DIR] = { XHG_INO, XFS_SICK_INO_DIR }, 104 [XFS_SCRUB_TYPE_XATTR] = { XHG_INO, XFS_SICK_INO_XATTR }, 105 [XFS_SCRUB_TYPE_SYMLINK] = { XHG_INO, XFS_SICK_INO_SYMLINK }, 106 [XFS_SCRUB_TYPE_PARENT] = { XHG_INO, XFS_SICK_INO_PARENT }, 107 [XFS_SCRUB_TYPE_RTBITMAP] = { XHG_RT, XFS_SICK_RT_BITMAP }, 108 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RT, XFS_SICK_RT_SUMMARY }, 109 [XFS_SCRUB_TYPE_UQUOTA] = { XHG_FS, XFS_SICK_FS_UQUOTA }, 110 [XFS_SCRUB_TYPE_GQUOTA] = { XHG_FS, XFS_SICK_FS_GQUOTA }, 111 [XFS_SCRUB_TYPE_PQUOTA] = { XHG_FS, XFS_SICK_FS_PQUOTA }, 112 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XHG_FS, XFS_SICK_FS_COUNTERS }, 113 }; 114 115 /* Return the health status mask for this scrub type. */ 116 unsigned int 117 xchk_health_mask_for_scrub_type( 118 __u32 scrub_type) 119 { 120 return type_to_health_flag[scrub_type].sick_mask; 121 } 122 123 /* 124 * Update filesystem health assessments based on what we found and did. 125 * 126 * If the scrubber finds errors, we mark sick whatever's mentioned in 127 * sick_mask, no matter whether this is a first scan or an 128 * evaluation of repair effectiveness. 129 * 130 * Otherwise, no direct corruption was found, so mark whatever's in 131 * sick_mask as healthy. 132 */ 133 void 134 xchk_update_health( 135 struct xfs_scrub *sc) 136 { 137 struct xfs_perag *pag; 138 bool bad; 139 140 if (!sc->sick_mask) 141 return; 142 143 bad = (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT); 144 switch (type_to_health_flag[sc->sm->sm_type].group) { 145 case XHG_AG: 146 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno); 147 if (bad) 148 xfs_ag_mark_sick(pag, sc->sick_mask); 149 else 150 xfs_ag_mark_healthy(pag, sc->sick_mask); 151 xfs_perag_put(pag); 152 break; 153 case XHG_INO: 154 if (!sc->ip) 155 return; 156 if (bad) 157 xfs_inode_mark_sick(sc->ip, sc->sick_mask); 158 else 159 xfs_inode_mark_healthy(sc->ip, sc->sick_mask); 160 break; 161 case XHG_FS: 162 if (bad) 163 xfs_fs_mark_sick(sc->mp, sc->sick_mask); 164 else 165 xfs_fs_mark_healthy(sc->mp, sc->sick_mask); 166 break; 167 case XHG_RT: 168 if (bad) 169 xfs_rt_mark_sick(sc->mp, sc->sick_mask); 170 else 171 xfs_rt_mark_healthy(sc->mp, sc->sick_mask); 172 break; 173 default: 174 ASSERT(0); 175 break; 176 } 177 } 178 179 /* Is the given per-AG btree healthy enough for scanning? */ 180 bool 181 xchk_ag_btree_healthy_enough( 182 struct xfs_scrub *sc, 183 struct xfs_perag *pag, 184 xfs_btnum_t btnum) 185 { 186 unsigned int mask = 0; 187 188 /* 189 * We always want the cursor if it's the same type as whatever we're 190 * scrubbing, even if we already know the structure is corrupt. 191 * 192 * Otherwise, we're only interested in the btree for cross-referencing. 193 * If we know the btree is bad then don't bother, just set XFAIL. 194 */ 195 switch (btnum) { 196 case XFS_BTNUM_BNO: 197 if (sc->sm->sm_type == XFS_SCRUB_TYPE_BNOBT) 198 return true; 199 mask = XFS_SICK_AG_BNOBT; 200 break; 201 case XFS_BTNUM_CNT: 202 if (sc->sm->sm_type == XFS_SCRUB_TYPE_CNTBT) 203 return true; 204 mask = XFS_SICK_AG_CNTBT; 205 break; 206 case XFS_BTNUM_INO: 207 if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT) 208 return true; 209 mask = XFS_SICK_AG_INOBT; 210 break; 211 case XFS_BTNUM_FINO: 212 if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT) 213 return true; 214 mask = XFS_SICK_AG_FINOBT; 215 break; 216 case XFS_BTNUM_RMAP: 217 if (sc->sm->sm_type == XFS_SCRUB_TYPE_RMAPBT) 218 return true; 219 mask = XFS_SICK_AG_RMAPBT; 220 break; 221 case XFS_BTNUM_REFC: 222 if (sc->sm->sm_type == XFS_SCRUB_TYPE_REFCNTBT) 223 return true; 224 mask = XFS_SICK_AG_REFCNTBT; 225 break; 226 default: 227 ASSERT(0); 228 return true; 229 } 230 231 if (xfs_ag_has_sickness(pag, mask)) { 232 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL; 233 return false; 234 } 235 236 return true; 237 } 238