1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2017-2023 Oracle. All Rights Reserved. 4 * Author: Darrick J. Wong <djwong@kernel.org> 5 */ 6 #ifndef __XFS_SCRUB_SCRUB_H__ 7 #define __XFS_SCRUB_SCRUB_H__ 8 9 struct xfs_scrub; 10 11 struct xchk_relax { 12 unsigned long next_resched; 13 unsigned int resched_nr; 14 bool interruptible; 15 }; 16 17 /* Yield to the scheduler at most 10x per second. */ 18 #define XCHK_RELAX_NEXT (jiffies + (HZ / 10)) 19 20 #define INIT_XCHK_RELAX \ 21 (struct xchk_relax){ \ 22 .next_resched = XCHK_RELAX_NEXT, \ 23 .resched_nr = 0, \ 24 .interruptible = true, \ 25 } 26 27 /* 28 * Relax during a scrub operation and exit if there's a fatal signal pending. 29 * 30 * If preemption is disabled, we need to yield to the scheduler every now and 31 * then so that we don't run afoul of the soft lockup watchdog or RCU stall 32 * detector. cond_resched calls are somewhat expensive (~5ns) so we want to 33 * ratelimit this to 10x per second. Amortize the cost of the other checks by 34 * only doing it once every 100 calls. 35 */ 36 static inline int xchk_maybe_relax(struct xchk_relax *widget) 37 { 38 /* Amortize the cost of scheduling and checking signals. */ 39 if (likely(++widget->resched_nr < 100)) 40 return 0; 41 widget->resched_nr = 0; 42 43 if (unlikely(widget->next_resched <= jiffies)) { 44 cond_resched(); 45 widget->next_resched = XCHK_RELAX_NEXT; 46 } 47 48 if (widget->interruptible && fatal_signal_pending(current)) 49 return -EINTR; 50 51 return 0; 52 } 53 54 /* 55 * Standard flags for allocating memory within scrub. NOFS context is 56 * configured by the process allocation scope. Scrub and repair must be able 57 * to back out gracefully if there isn't enough memory. Force-cast to avoid 58 * complaints from static checkers. 59 */ 60 #define XCHK_GFP_FLAGS ((__force gfp_t)(GFP_KERNEL | __GFP_NOWARN | \ 61 __GFP_RETRY_MAYFAIL)) 62 63 /* 64 * For opening files by handle for fsck operations, we don't trust the inumber 65 * or the allocation state; therefore, perform an untrusted lookup. We don't 66 * want these inodes to pollute the cache, so mark them for immediate removal. 67 */ 68 #define XCHK_IGET_FLAGS (XFS_IGET_UNTRUSTED | XFS_IGET_DONTCACHE) 69 70 /* Type info and names for the scrub types. */ 71 enum xchk_type { 72 ST_NONE = 1, /* disabled */ 73 ST_PERAG, /* per-AG metadata */ 74 ST_FS, /* per-FS metadata */ 75 ST_INODE, /* per-inode metadata */ 76 }; 77 78 struct xchk_meta_ops { 79 /* Acquire whatever resources are needed for the operation. */ 80 int (*setup)(struct xfs_scrub *sc); 81 82 /* Examine metadata for errors. */ 83 int (*scrub)(struct xfs_scrub *); 84 85 /* Repair or optimize the metadata. */ 86 int (*repair)(struct xfs_scrub *); 87 88 /* 89 * Re-scrub the metadata we repaired, in case there's extra work that 90 * we need to do to check our repair work. If this is NULL, we'll use 91 * the ->scrub function pointer, assuming that the regular scrub is 92 * sufficient. 93 */ 94 int (*repair_eval)(struct xfs_scrub *sc); 95 96 /* Decide if we even have this piece of metadata. */ 97 bool (*has)(struct xfs_mount *); 98 99 /* type describing required/allowed inputs */ 100 enum xchk_type type; 101 }; 102 103 /* Buffer pointers and btree cursors for an entire AG. */ 104 struct xchk_ag { 105 struct xfs_perag *pag; 106 107 /* AG btree roots */ 108 struct xfs_buf *agf_bp; 109 struct xfs_buf *agi_bp; 110 111 /* AG btrees */ 112 struct xfs_btree_cur *bno_cur; 113 struct xfs_btree_cur *cnt_cur; 114 struct xfs_btree_cur *ino_cur; 115 struct xfs_btree_cur *fino_cur; 116 struct xfs_btree_cur *rmap_cur; 117 struct xfs_btree_cur *refc_cur; 118 }; 119 120 struct xfs_scrub { 121 /* General scrub state. */ 122 struct xfs_mount *mp; 123 struct xfs_scrub_metadata *sm; 124 const struct xchk_meta_ops *ops; 125 struct xfs_trans *tp; 126 127 /* File that scrub was called with. */ 128 struct file *file; 129 130 /* 131 * File that is undergoing the scrub operation. This can differ from 132 * the file that scrub was called with if we're checking file-based fs 133 * metadata (e.g. rt bitmaps) or if we're doing a scrub-by-handle for 134 * something that can't be opened directly (e.g. symlinks). 135 */ 136 struct xfs_inode *ip; 137 138 /* Kernel memory buffer used by scrubbers; freed at teardown. */ 139 void *buf; 140 141 /* 142 * Clean up resources owned by whatever is in the buffer. Cleanup can 143 * be deferred with this hook as a means for scrub functions to pass 144 * data to repair functions. This function must not free the buffer 145 * itself. 146 */ 147 void (*buf_cleanup)(void *buf); 148 149 /* xfile used by the scrubbers; freed at teardown. */ 150 struct xfile *xfile; 151 152 /* buffer target for in-memory btrees; also freed at teardown. */ 153 struct xfs_buftarg *xmbtp; 154 155 /* Lock flags for @ip. */ 156 uint ilock_flags; 157 158 /* The orphanage, for stashing files that have lost their parent. */ 159 uint orphanage_ilock_flags; 160 struct xfs_inode *orphanage; 161 162 /* A temporary file on this filesystem, for staging new metadata. */ 163 struct xfs_inode *tempip; 164 uint temp_ilock_flags; 165 166 /* See the XCHK/XREP state flags below. */ 167 unsigned int flags; 168 169 /* 170 * The XFS_SICK_* flags that correspond to the metadata being scrubbed 171 * or repaired. We will use this mask to update the in-core fs health 172 * status with whatever we find. 173 */ 174 unsigned int sick_mask; 175 176 /* next time we want to cond_resched() */ 177 struct xchk_relax relax; 178 179 /* State tracking for single-AG operations. */ 180 struct xchk_ag sa; 181 }; 182 183 /* XCHK state flags grow up from zero, XREP state flags grown down from 2^31 */ 184 #define XCHK_TRY_HARDER (1U << 0) /* can't get resources, try again */ 185 #define XCHK_HAVE_FREEZE_PROT (1U << 1) /* do we have freeze protection? */ 186 #define XCHK_FSGATES_DRAIN (1U << 2) /* defer ops draining enabled */ 187 #define XCHK_NEED_DRAIN (1U << 3) /* scrub needs to drain defer ops */ 188 #define XCHK_FSGATES_QUOTA (1U << 4) /* quota live update enabled */ 189 #define XCHK_FSGATES_DIRENTS (1U << 5) /* directory live update enabled */ 190 #define XCHK_FSGATES_RMAP (1U << 6) /* rmapbt live update enabled */ 191 #define XREP_RESET_PERAG_RESV (1U << 30) /* must reset AG space reservation */ 192 #define XREP_ALREADY_FIXED (1U << 31) /* checking our repair work */ 193 194 /* 195 * The XCHK_FSGATES* flags reflect functionality in the main filesystem that 196 * are only enabled for this particular online fsck. When not in use, the 197 * features are gated off via dynamic code patching, which is why the state 198 * must be enabled during scrub setup and can only be torn down afterwards. 199 */ 200 #define XCHK_FSGATES_ALL (XCHK_FSGATES_DRAIN | \ 201 XCHK_FSGATES_QUOTA | \ 202 XCHK_FSGATES_DIRENTS | \ 203 XCHK_FSGATES_RMAP) 204 205 struct xfs_scrub_subord { 206 struct xfs_scrub sc; 207 struct xfs_scrub *parent_sc; 208 unsigned int old_smtype; 209 unsigned int old_smflags; 210 }; 211 212 struct xfs_scrub_subord *xchk_scrub_create_subord(struct xfs_scrub *sc, 213 unsigned int subtype); 214 void xchk_scrub_free_subord(struct xfs_scrub_subord *sub); 215 216 /* 217 * We /could/ terminate a scrub/repair operation early. If we're not 218 * in a good place to continue (fatal signal, etc.) then bail out. 219 * Note that we're careful not to make any judgements about *error. 220 */ 221 static inline bool 222 xchk_should_terminate( 223 struct xfs_scrub *sc, 224 int *error) 225 { 226 if (xchk_maybe_relax(&sc->relax)) { 227 if (*error == 0) 228 *error = -EINTR; 229 return true; 230 } 231 return false; 232 } 233 234 static inline int xchk_nothing(struct xfs_scrub *sc) 235 { 236 return -ENOENT; 237 } 238 239 /* Metadata scrubbers */ 240 int xchk_tester(struct xfs_scrub *sc); 241 int xchk_superblock(struct xfs_scrub *sc); 242 int xchk_agf(struct xfs_scrub *sc); 243 int xchk_agfl(struct xfs_scrub *sc); 244 int xchk_agi(struct xfs_scrub *sc); 245 int xchk_allocbt(struct xfs_scrub *sc); 246 int xchk_iallocbt(struct xfs_scrub *sc); 247 int xchk_rmapbt(struct xfs_scrub *sc); 248 int xchk_refcountbt(struct xfs_scrub *sc); 249 int xchk_inode(struct xfs_scrub *sc); 250 int xchk_bmap_data(struct xfs_scrub *sc); 251 int xchk_bmap_attr(struct xfs_scrub *sc); 252 int xchk_bmap_cow(struct xfs_scrub *sc); 253 int xchk_directory(struct xfs_scrub *sc); 254 int xchk_xattr(struct xfs_scrub *sc); 255 int xchk_symlink(struct xfs_scrub *sc); 256 int xchk_parent(struct xfs_scrub *sc); 257 int xchk_dirtree(struct xfs_scrub *sc); 258 #ifdef CONFIG_XFS_RT 259 int xchk_rtbitmap(struct xfs_scrub *sc); 260 int xchk_rtsummary(struct xfs_scrub *sc); 261 #else 262 # define xchk_rtbitmap xchk_nothing 263 # define xchk_rtsummary xchk_nothing 264 #endif 265 #ifdef CONFIG_XFS_QUOTA 266 int xchk_quota(struct xfs_scrub *sc); 267 int xchk_quotacheck(struct xfs_scrub *sc); 268 #else 269 # define xchk_quota xchk_nothing 270 # define xchk_quotacheck xchk_nothing 271 #endif 272 int xchk_fscounters(struct xfs_scrub *sc); 273 int xchk_nlinks(struct xfs_scrub *sc); 274 275 /* cross-referencing helpers */ 276 void xchk_xref_is_used_space(struct xfs_scrub *sc, xfs_agblock_t agbno, 277 xfs_extlen_t len); 278 void xchk_xref_is_not_inode_chunk(struct xfs_scrub *sc, xfs_agblock_t agbno, 279 xfs_extlen_t len); 280 void xchk_xref_is_inode_chunk(struct xfs_scrub *sc, xfs_agblock_t agbno, 281 xfs_extlen_t len); 282 void xchk_xref_is_only_owned_by(struct xfs_scrub *sc, xfs_agblock_t agbno, 283 xfs_extlen_t len, const struct xfs_owner_info *oinfo); 284 void xchk_xref_is_not_owned_by(struct xfs_scrub *sc, xfs_agblock_t agbno, 285 xfs_extlen_t len, const struct xfs_owner_info *oinfo); 286 void xchk_xref_has_no_owner(struct xfs_scrub *sc, xfs_agblock_t agbno, 287 xfs_extlen_t len); 288 void xchk_xref_is_cow_staging(struct xfs_scrub *sc, xfs_agblock_t bno, 289 xfs_extlen_t len); 290 void xchk_xref_is_not_shared(struct xfs_scrub *sc, xfs_agblock_t bno, 291 xfs_extlen_t len); 292 void xchk_xref_is_not_cow_staging(struct xfs_scrub *sc, xfs_agblock_t bno, 293 xfs_extlen_t len); 294 #ifdef CONFIG_XFS_RT 295 void xchk_xref_is_used_rt_space(struct xfs_scrub *sc, xfs_rtblock_t rtbno, 296 xfs_extlen_t len); 297 #else 298 # define xchk_xref_is_used_rt_space(sc, rtbno, len) do { } while (0) 299 #endif 300 301 #endif /* __XFS_SCRUB_SCRUB_H__ */ 302