| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * Copyright (c) 2022-2024 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_bit.h" |
| 12 | #include "xfs_sb.h" |
| 13 | #include "xfs_mount.h" |
| 14 | #include "xfs_btree.h" |
| 15 | #include "xfs_alloc_btree.h" |
| 16 | #include "xfs_rmap_btree.h" |
| 17 | #include "xfs_alloc.h" |
| 18 | #include "xfs_ialloc.h" |
| 19 | #include "xfs_rmap.h" |
| 20 | #include "xfs_ag.h" |
| 21 | #include "xfs_ag_resv.h" |
| 22 | #include "xfs_health.h" |
| 23 | #include "xfs_error.h" |
| 24 | #include "xfs_bmap.h" |
| 25 | #include "xfs_defer.h" |
| 26 | #include "xfs_log_format.h" |
| 27 | #include "xfs_trans.h" |
| 28 | #include "xfs_trace.h" |
| 29 | #include "xfs_inode.h" |
| 30 | #include "xfs_icache.h" |
| 31 | #include "xfs_buf_item.h" |
| 32 | #include "xfs_rtgroup.h" |
| 33 | #include "xfs_rtbitmap.h" |
| 34 | #include "xfs_metafile.h" |
| 35 | #include "xfs_metadir.h" |
| 36 | #include "xfs_rtrmap_btree.h" |
| 37 | #include "xfs_rtrefcount_btree.h" |
| 38 | |
| 39 | /* Find the first usable fsblock in this rtgroup. */ |
| 40 | static inline uint32_t |
| 41 | xfs_rtgroup_min_block( |
| 42 | struct xfs_mount *mp, |
| 43 | xfs_rgnumber_t rgno) |
| 44 | { |
| 45 | if (xfs_has_rtsb(mp) && rgno == 0) |
| 46 | return mp->m_sb.sb_rextsize; |
| 47 | |
| 48 | return 0; |
| 49 | } |
| 50 | |
| 51 | /* Compute the number of rt extents in this realtime group. */ |
| 52 | static xfs_rtxnum_t |
| 53 | __xfs_rtgroup_extents( |
| 54 | struct xfs_mount *mp, |
| 55 | xfs_rgnumber_t rgno, |
| 56 | xfs_rgnumber_t rgcount, |
| 57 | xfs_rtbxlen_t rextents) |
| 58 | { |
| 59 | ASSERT(rgno < rgcount); |
| 60 | if (rgno == rgcount - 1) |
| 61 | return rextents - ((xfs_rtxnum_t)rgno * mp->m_sb.sb_rgextents); |
| 62 | |
| 63 | ASSERT(xfs_has_rtgroups(mp)); |
| 64 | return mp->m_sb.sb_rgextents; |
| 65 | } |
| 66 | |
| 67 | xfs_rtxnum_t |
| 68 | xfs_rtgroup_extents( |
| 69 | struct xfs_mount *mp, |
| 70 | xfs_rgnumber_t rgno) |
| 71 | { |
| 72 | return __xfs_rtgroup_extents(mp, rgno, mp->m_sb.sb_rgcount, |
| 73 | mp->m_sb.sb_rextents); |
| 74 | } |
| 75 | |
| 76 | /* Precompute this group's geometry */ |
| 77 | void |
| 78 | xfs_rtgroup_calc_geometry( |
| 79 | struct xfs_mount *mp, |
| 80 | struct xfs_rtgroup *rtg, |
| 81 | xfs_rgnumber_t rgno, |
| 82 | xfs_rgnumber_t rgcount, |
| 83 | xfs_rtbxlen_t rextents) |
| 84 | { |
| 85 | rtg->rtg_extents = __xfs_rtgroup_extents(mp, rgno, rgcount, rextents); |
| 86 | rtg_group(rtg)->xg_block_count = |
| 87 | rtg->rtg_extents * mp->m_sb.sb_rextsize; |
| 88 | rtg_group(rtg)->xg_min_gbno = xfs_rtgroup_min_block(mp, rgno); |
| 89 | } |
| 90 | |
| 91 | int |
| 92 | xfs_rtgroup_alloc( |
| 93 | struct xfs_mount *mp, |
| 94 | xfs_rgnumber_t rgno, |
| 95 | xfs_rgnumber_t rgcount, |
| 96 | xfs_rtbxlen_t rextents) |
| 97 | { |
| 98 | struct xfs_rtgroup *rtg; |
| 99 | int error; |
| 100 | |
| 101 | rtg = kzalloc(sizeof(struct xfs_rtgroup), GFP_KERNEL); |
| 102 | if (!rtg) |
| 103 | return -ENOMEM; |
| 104 | |
| 105 | xfs_rtgroup_calc_geometry(mp, rtg, rgno, rgcount, rextents); |
| 106 | |
| 107 | error = xfs_group_insert(mp, rtg_group(rtg), rgno, XG_TYPE_RTG); |
| 108 | if (error) |
| 109 | goto out_free_rtg; |
| 110 | return 0; |
| 111 | |
| 112 | out_free_rtg: |
| 113 | kfree(rtg); |
| 114 | return error; |
| 115 | } |
| 116 | |
| 117 | void |
| 118 | xfs_rtgroup_free( |
| 119 | struct xfs_mount *mp, |
| 120 | xfs_rgnumber_t rgno) |
| 121 | { |
| 122 | xfs_group_free(mp, rgno, XG_TYPE_RTG, NULL); |
| 123 | } |
| 124 | |
| 125 | /* Free a range of incore rtgroup objects. */ |
| 126 | void |
| 127 | xfs_free_rtgroups( |
| 128 | struct xfs_mount *mp, |
| 129 | xfs_rgnumber_t first_rgno, |
| 130 | xfs_rgnumber_t end_rgno) |
| 131 | { |
| 132 | xfs_rgnumber_t rgno; |
| 133 | |
| 134 | for (rgno = first_rgno; rgno < end_rgno; rgno++) |
| 135 | xfs_rtgroup_free(mp, rgno); |
| 136 | } |
| 137 | |
| 138 | /* Initialize some range of incore rtgroup objects. */ |
| 139 | int |
| 140 | xfs_initialize_rtgroups( |
| 141 | struct xfs_mount *mp, |
| 142 | xfs_rgnumber_t first_rgno, |
| 143 | xfs_rgnumber_t end_rgno, |
| 144 | xfs_rtbxlen_t rextents) |
| 145 | { |
| 146 | xfs_rgnumber_t index; |
| 147 | int error; |
| 148 | |
| 149 | if (first_rgno >= end_rgno) |
| 150 | return 0; |
| 151 | |
| 152 | for (index = first_rgno; index < end_rgno; index++) { |
| 153 | error = xfs_rtgroup_alloc(mp, index, end_rgno, rextents); |
| 154 | if (error) |
| 155 | goto out_unwind_new_rtgs; |
| 156 | } |
| 157 | |
| 158 | return 0; |
| 159 | |
| 160 | out_unwind_new_rtgs: |
| 161 | xfs_free_rtgroups(mp, first_rgno, index); |
| 162 | return error; |
| 163 | } |
| 164 | |
| 165 | /* |
| 166 | * Update the rt extent count of the previous tail rtgroup if it changed during |
| 167 | * recovery (i.e. recovery of a growfs). |
| 168 | */ |
| 169 | int |
| 170 | xfs_update_last_rtgroup_size( |
| 171 | struct xfs_mount *mp, |
| 172 | xfs_rgnumber_t prev_rgcount) |
| 173 | { |
| 174 | struct xfs_rtgroup *rtg; |
| 175 | |
| 176 | ASSERT(prev_rgcount > 0); |
| 177 | |
| 178 | rtg = xfs_rtgroup_grab(mp, prev_rgcount - 1); |
| 179 | if (!rtg) |
| 180 | return -EFSCORRUPTED; |
| 181 | rtg->rtg_extents = __xfs_rtgroup_extents(mp, prev_rgcount - 1, |
| 182 | mp->m_sb.sb_rgcount, mp->m_sb.sb_rextents); |
| 183 | rtg_group(rtg)->xg_block_count = rtg->rtg_extents * mp->m_sb.sb_rextsize; |
| 184 | xfs_rtgroup_rele(rtg); |
| 185 | return 0; |
| 186 | } |
| 187 | |
| 188 | /* Lock metadata inodes associated with this rt group. */ |
| 189 | void |
| 190 | xfs_rtgroup_lock( |
| 191 | struct xfs_rtgroup *rtg, |
| 192 | unsigned int rtglock_flags) |
| 193 | { |
| 194 | ASSERT(!(rtglock_flags & ~XFS_RTGLOCK_ALL_FLAGS)); |
| 195 | ASSERT(!(rtglock_flags & XFS_RTGLOCK_BITMAP_SHARED) || |
| 196 | !(rtglock_flags & XFS_RTGLOCK_BITMAP)); |
| 197 | |
| 198 | if (!xfs_has_zoned(rtg_mount(rtg))) { |
| 199 | if (rtglock_flags & XFS_RTGLOCK_BITMAP) { |
| 200 | /* |
| 201 | * Lock both realtime free space metadata inodes for a |
| 202 | * freespace update. |
| 203 | */ |
| 204 | xfs_ilock(rtg_bitmap(rtg), XFS_ILOCK_EXCL); |
| 205 | xfs_ilock(rtg_summary(rtg), XFS_ILOCK_EXCL); |
| 206 | } else if (rtglock_flags & XFS_RTGLOCK_BITMAP_SHARED) { |
| 207 | xfs_ilock(rtg_bitmap(rtg), XFS_ILOCK_SHARED); |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | if ((rtglock_flags & XFS_RTGLOCK_RMAP) && rtg_rmap(rtg)) |
| 212 | xfs_ilock(rtg_rmap(rtg), XFS_ILOCK_EXCL); |
| 213 | |
| 214 | if ((rtglock_flags & XFS_RTGLOCK_REFCOUNT) && rtg_refcount(rtg)) |
| 215 | xfs_ilock(rtg_refcount(rtg), XFS_ILOCK_EXCL); |
| 216 | } |
| 217 | |
| 218 | /* Unlock metadata inodes associated with this rt group. */ |
| 219 | void |
| 220 | xfs_rtgroup_unlock( |
| 221 | struct xfs_rtgroup *rtg, |
| 222 | unsigned int rtglock_flags) |
| 223 | { |
| 224 | ASSERT(!(rtglock_flags & ~XFS_RTGLOCK_ALL_FLAGS)); |
| 225 | ASSERT(!(rtglock_flags & XFS_RTGLOCK_BITMAP_SHARED) || |
| 226 | !(rtglock_flags & XFS_RTGLOCK_BITMAP)); |
| 227 | |
| 228 | if ((rtglock_flags & XFS_RTGLOCK_REFCOUNT) && rtg_refcount(rtg)) |
| 229 | xfs_iunlock(rtg_refcount(rtg), XFS_ILOCK_EXCL); |
| 230 | |
| 231 | if ((rtglock_flags & XFS_RTGLOCK_RMAP) && rtg_rmap(rtg)) |
| 232 | xfs_iunlock(rtg_rmap(rtg), XFS_ILOCK_EXCL); |
| 233 | |
| 234 | if (!xfs_has_zoned(rtg_mount(rtg))) { |
| 235 | if (rtglock_flags & XFS_RTGLOCK_BITMAP) { |
| 236 | xfs_iunlock(rtg_summary(rtg), XFS_ILOCK_EXCL); |
| 237 | xfs_iunlock(rtg_bitmap(rtg), XFS_ILOCK_EXCL); |
| 238 | } else if (rtglock_flags & XFS_RTGLOCK_BITMAP_SHARED) { |
| 239 | xfs_iunlock(rtg_bitmap(rtg), XFS_ILOCK_SHARED); |
| 240 | } |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * Join realtime group metadata inodes to the transaction. The ILOCKs will be |
| 246 | * released on transaction commit. |
| 247 | */ |
| 248 | void |
| 249 | xfs_rtgroup_trans_join( |
| 250 | struct xfs_trans *tp, |
| 251 | struct xfs_rtgroup *rtg, |
| 252 | unsigned int rtglock_flags) |
| 253 | { |
| 254 | ASSERT(!(rtglock_flags & ~XFS_RTGLOCK_ALL_FLAGS)); |
| 255 | ASSERT(!(rtglock_flags & XFS_RTGLOCK_BITMAP_SHARED)); |
| 256 | |
| 257 | if (!xfs_has_zoned(rtg_mount(rtg)) && |
| 258 | (rtglock_flags & XFS_RTGLOCK_BITMAP)) { |
| 259 | xfs_trans_ijoin(tp, rtg_bitmap(rtg), XFS_ILOCK_EXCL); |
| 260 | xfs_trans_ijoin(tp, rtg_summary(rtg), XFS_ILOCK_EXCL); |
| 261 | } |
| 262 | |
| 263 | if ((rtglock_flags & XFS_RTGLOCK_RMAP) && rtg_rmap(rtg)) |
| 264 | xfs_trans_ijoin(tp, rtg_rmap(rtg), XFS_ILOCK_EXCL); |
| 265 | |
| 266 | if ((rtglock_flags & XFS_RTGLOCK_REFCOUNT) && rtg_refcount(rtg)) |
| 267 | xfs_trans_ijoin(tp, rtg_refcount(rtg), XFS_ILOCK_EXCL); |
| 268 | } |
| 269 | |
| 270 | /* Retrieve rt group geometry. */ |
| 271 | int |
| 272 | xfs_rtgroup_get_geometry( |
| 273 | struct xfs_rtgroup *rtg, |
| 274 | struct xfs_rtgroup_geometry *rgeo) |
| 275 | { |
| 276 | /* Fill out form. */ |
| 277 | memset(rgeo, 0, sizeof(*rgeo)); |
| 278 | rgeo->rg_number = rtg_rgno(rtg); |
| 279 | rgeo->rg_length = rtg_blocks(rtg); |
| 280 | xfs_rtgroup_geom_health(rtg, rgeo); |
| 281 | return 0; |
| 282 | } |
| 283 | |
| 284 | #ifdef CONFIG_PROVE_LOCKING |
| 285 | static struct lock_class_key xfs_rtginode_lock_class; |
| 286 | |
| 287 | static int |
| 288 | xfs_rtginode_ilock_cmp_fn( |
| 289 | const struct lockdep_map *m1, |
| 290 | const struct lockdep_map *m2) |
| 291 | { |
| 292 | const struct xfs_inode *ip1 = |
| 293 | container_of(m1, struct xfs_inode, i_lock.dep_map); |
| 294 | const struct xfs_inode *ip2 = |
| 295 | container_of(m2, struct xfs_inode, i_lock.dep_map); |
| 296 | |
| 297 | if (ip1->i_projid < ip2->i_projid) |
| 298 | return -1; |
| 299 | if (ip1->i_projid > ip2->i_projid) |
| 300 | return 1; |
| 301 | return 0; |
| 302 | } |
| 303 | |
| 304 | static inline void |
| 305 | xfs_rtginode_ilock_print_fn( |
| 306 | const struct lockdep_map *m) |
| 307 | { |
| 308 | const struct xfs_inode *ip = |
| 309 | container_of(m, struct xfs_inode, i_lock.dep_map); |
| 310 | |
| 311 | printk(KERN_CONT " rgno=%u metatype=%s" , ip->i_projid, |
| 312 | xfs_metafile_type_str(ip->i_metatype)); |
| 313 | } |
| 314 | |
| 315 | /* |
| 316 | * Most of the time each of the RTG inode locks are only taken one at a time. |
| 317 | * But when committing deferred ops, more than one of a kind can be taken. |
| 318 | * However, deferred rt ops will be committed in rgno order so there is no |
| 319 | * potential for deadlocks. The code here is needed to tell lockdep about this |
| 320 | * order. |
| 321 | */ |
| 322 | static inline void |
| 323 | xfs_rtginode_lockdep_setup( |
| 324 | struct xfs_inode *ip, |
| 325 | xfs_rgnumber_t rgno, |
| 326 | enum xfs_rtg_inodes type) |
| 327 | { |
| 328 | lockdep_set_class_and_subclass(&ip->i_lock, &xfs_rtginode_lock_class, |
| 329 | type); |
| 330 | lock_set_cmp_fn(&ip->i_lock, xfs_rtginode_ilock_cmp_fn, |
| 331 | xfs_rtginode_ilock_print_fn); |
| 332 | } |
| 333 | #else |
| 334 | #define xfs_rtginode_lockdep_setup(ip, rgno, type) do { } while (0) |
| 335 | #endif /* CONFIG_PROVE_LOCKING */ |
| 336 | |
| 337 | struct xfs_rtginode_ops { |
| 338 | const char *name; /* short name */ |
| 339 | |
| 340 | enum xfs_metafile_type metafile_type; |
| 341 | |
| 342 | unsigned int sick; /* rtgroup sickness flag */ |
| 343 | |
| 344 | unsigned int fmt_mask; /* all valid data fork formats */ |
| 345 | |
| 346 | /* Does the fs have this feature? */ |
| 347 | bool (*enabled)(const struct xfs_mount *mp); |
| 348 | |
| 349 | /* Create this rtgroup metadata inode and initialize it. */ |
| 350 | int (*create)(struct xfs_rtgroup *rtg, |
| 351 | struct xfs_inode *ip, |
| 352 | struct xfs_trans *tp, |
| 353 | bool init); |
| 354 | }; |
| 355 | |
| 356 | static const struct xfs_rtginode_ops xfs_rtginode_ops[XFS_RTGI_MAX] = { |
| 357 | [XFS_RTGI_BITMAP] = { |
| 358 | .name = "bitmap" , |
| 359 | .metafile_type = XFS_METAFILE_RTBITMAP, |
| 360 | .sick = XFS_SICK_RG_BITMAP, |
| 361 | .fmt_mask = (1U << XFS_DINODE_FMT_EXTENTS) | |
| 362 | (1U << XFS_DINODE_FMT_BTREE), |
| 363 | .enabled = xfs_has_nonzoned, |
| 364 | .create = xfs_rtbitmap_create, |
| 365 | }, |
| 366 | [XFS_RTGI_SUMMARY] = { |
| 367 | .name = "summary" , |
| 368 | .metafile_type = XFS_METAFILE_RTSUMMARY, |
| 369 | .sick = XFS_SICK_RG_SUMMARY, |
| 370 | .fmt_mask = (1U << XFS_DINODE_FMT_EXTENTS) | |
| 371 | (1U << XFS_DINODE_FMT_BTREE), |
| 372 | .enabled = xfs_has_nonzoned, |
| 373 | .create = xfs_rtsummary_create, |
| 374 | }, |
| 375 | [XFS_RTGI_RMAP] = { |
| 376 | .name = "rmap" , |
| 377 | .metafile_type = XFS_METAFILE_RTRMAP, |
| 378 | .sick = XFS_SICK_RG_RMAPBT, |
| 379 | .fmt_mask = 1U << XFS_DINODE_FMT_META_BTREE, |
| 380 | /* |
| 381 | * growfs must create the rtrmap inodes before adding a |
| 382 | * realtime volume to the filesystem, so we cannot use the |
| 383 | * rtrmapbt predicate here. |
| 384 | */ |
| 385 | .enabled = xfs_has_rmapbt, |
| 386 | .create = xfs_rtrmapbt_create, |
| 387 | }, |
| 388 | [XFS_RTGI_REFCOUNT] = { |
| 389 | .name = "refcount" , |
| 390 | .metafile_type = XFS_METAFILE_RTREFCOUNT, |
| 391 | .sick = XFS_SICK_RG_REFCNTBT, |
| 392 | .fmt_mask = 1U << XFS_DINODE_FMT_META_BTREE, |
| 393 | /* same comment about growfs and rmap inodes applies here */ |
| 394 | .enabled = xfs_has_reflink, |
| 395 | .create = xfs_rtrefcountbt_create, |
| 396 | }, |
| 397 | }; |
| 398 | |
| 399 | /* Return the shortname of this rtgroup inode. */ |
| 400 | const char * |
| 401 | xfs_rtginode_name( |
| 402 | enum xfs_rtg_inodes type) |
| 403 | { |
| 404 | return xfs_rtginode_ops[type].name; |
| 405 | } |
| 406 | |
| 407 | /* Return the metafile type of this rtgroup inode. */ |
| 408 | enum xfs_metafile_type |
| 409 | xfs_rtginode_metafile_type( |
| 410 | enum xfs_rtg_inodes type) |
| 411 | { |
| 412 | return xfs_rtginode_ops[type].metafile_type; |
| 413 | } |
| 414 | |
| 415 | /* Should this rtgroup inode be present? */ |
| 416 | bool |
| 417 | xfs_rtginode_enabled( |
| 418 | struct xfs_rtgroup *rtg, |
| 419 | enum xfs_rtg_inodes type) |
| 420 | { |
| 421 | const struct xfs_rtginode_ops *ops = &xfs_rtginode_ops[type]; |
| 422 | |
| 423 | if (!ops->enabled) |
| 424 | return true; |
| 425 | return ops->enabled(rtg_mount(rtg)); |
| 426 | } |
| 427 | |
| 428 | /* Mark an rtgroup inode sick */ |
| 429 | void |
| 430 | xfs_rtginode_mark_sick( |
| 431 | struct xfs_rtgroup *rtg, |
| 432 | enum xfs_rtg_inodes type) |
| 433 | { |
| 434 | const struct xfs_rtginode_ops *ops = &xfs_rtginode_ops[type]; |
| 435 | |
| 436 | xfs_group_mark_sick(xg: rtg_group(rtg), mask: ops->sick); |
| 437 | } |
| 438 | |
| 439 | /* Load and existing rtgroup inode into the rtgroup structure. */ |
| 440 | int |
| 441 | xfs_rtginode_load( |
| 442 | struct xfs_rtgroup *rtg, |
| 443 | enum xfs_rtg_inodes type, |
| 444 | struct xfs_trans *tp) |
| 445 | { |
| 446 | struct xfs_mount *mp = tp->t_mountp; |
| 447 | struct xfs_inode *ip; |
| 448 | const struct xfs_rtginode_ops *ops = &xfs_rtginode_ops[type]; |
| 449 | int error; |
| 450 | |
| 451 | if (!xfs_rtginode_enabled(rtg, type)) |
| 452 | return 0; |
| 453 | |
| 454 | if (!xfs_has_rtgroups(mp)) { |
| 455 | xfs_ino_t ino; |
| 456 | |
| 457 | switch (type) { |
| 458 | case XFS_RTGI_BITMAP: |
| 459 | ino = mp->m_sb.sb_rbmino; |
| 460 | break; |
| 461 | case XFS_RTGI_SUMMARY: |
| 462 | ino = mp->m_sb.sb_rsumino; |
| 463 | break; |
| 464 | default: |
| 465 | /* None of the other types exist on !rtgroups */ |
| 466 | return 0; |
| 467 | } |
| 468 | |
| 469 | error = xfs_trans_metafile_iget(tp, ino, ops->metafile_type, |
| 470 | &ip); |
| 471 | } else { |
| 472 | const char *path; |
| 473 | |
| 474 | if (!mp->m_rtdirip) { |
| 475 | xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR); |
| 476 | return -EFSCORRUPTED; |
| 477 | } |
| 478 | |
| 479 | path = xfs_rtginode_path(rtg_rgno(rtg), type); |
| 480 | if (!path) |
| 481 | return -ENOMEM; |
| 482 | error = xfs_metadir_load(tp, dp: mp->m_rtdirip, path, |
| 483 | metafile_type: ops->metafile_type, ipp: &ip); |
| 484 | kfree(path); |
| 485 | } |
| 486 | |
| 487 | if (error) { |
| 488 | if (xfs_metadata_is_sick(error)) |
| 489 | xfs_rtginode_mark_sick(rtg, type); |
| 490 | return error; |
| 491 | } |
| 492 | |
| 493 | if (XFS_IS_CORRUPT(mp, !((1U << ip->i_df.if_format) & ops->fmt_mask))) { |
| 494 | xfs_irele(ip); |
| 495 | xfs_rtginode_mark_sick(rtg, type); |
| 496 | return -EFSCORRUPTED; |
| 497 | } |
| 498 | |
| 499 | if (XFS_IS_CORRUPT(mp, ip->i_projid != rtg_rgno(rtg))) { |
| 500 | xfs_irele(ip); |
| 501 | xfs_rtginode_mark_sick(rtg, type); |
| 502 | return -EFSCORRUPTED; |
| 503 | } |
| 504 | |
| 505 | xfs_rtginode_lockdep_setup(ip, rtg_rgno(rtg), type); |
| 506 | rtg->rtg_inodes[type] = ip; |
| 507 | return 0; |
| 508 | } |
| 509 | |
| 510 | /* Release an rtgroup metadata inode. */ |
| 511 | void |
| 512 | xfs_rtginode_irele( |
| 513 | struct xfs_inode **ipp) |
| 514 | { |
| 515 | if (*ipp) |
| 516 | xfs_irele(*ipp); |
| 517 | *ipp = NULL; |
| 518 | } |
| 519 | |
| 520 | /* Add a metadata inode for a realtime rmap btree. */ |
| 521 | int |
| 522 | xfs_rtginode_create( |
| 523 | struct xfs_rtgroup *rtg, |
| 524 | enum xfs_rtg_inodes type, |
| 525 | bool init) |
| 526 | { |
| 527 | const struct xfs_rtginode_ops *ops = &xfs_rtginode_ops[type]; |
| 528 | struct xfs_mount *mp = rtg_mount(rtg); |
| 529 | struct xfs_metadir_update upd = { |
| 530 | .dp = mp->m_rtdirip, |
| 531 | .metafile_type = ops->metafile_type, |
| 532 | }; |
| 533 | int error; |
| 534 | |
| 535 | if (!xfs_rtginode_enabled(rtg, type)) |
| 536 | return 0; |
| 537 | |
| 538 | if (!mp->m_rtdirip) { |
| 539 | xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR); |
| 540 | return -EFSCORRUPTED; |
| 541 | } |
| 542 | |
| 543 | upd.path = xfs_rtginode_path(rtg_rgno(rtg), type); |
| 544 | if (!upd.path) |
| 545 | return -ENOMEM; |
| 546 | |
| 547 | error = xfs_metadir_start_create(upd: &upd); |
| 548 | if (error) |
| 549 | goto out_path; |
| 550 | |
| 551 | error = xfs_metadir_create(&upd, S_IFREG); |
| 552 | if (error) |
| 553 | goto out_cancel; |
| 554 | |
| 555 | xfs_rtginode_lockdep_setup(upd.ip, rtg_rgno(rtg), type); |
| 556 | |
| 557 | upd.ip->i_projid = rtg_rgno(rtg); |
| 558 | error = ops->create(rtg, upd.ip, upd.tp, init); |
| 559 | if (error) |
| 560 | goto out_cancel; |
| 561 | |
| 562 | error = xfs_metadir_commit(upd: &upd); |
| 563 | if (error) |
| 564 | goto out_path; |
| 565 | |
| 566 | kfree(upd.path); |
| 567 | xfs_finish_inode_setup(upd.ip); |
| 568 | rtg->rtg_inodes[type] = upd.ip; |
| 569 | return 0; |
| 570 | |
| 571 | out_cancel: |
| 572 | xfs_metadir_cancel(upd: &upd, error); |
| 573 | /* Have to finish setting up the inode to ensure it's deleted. */ |
| 574 | if (upd.ip) { |
| 575 | xfs_finish_inode_setup(upd.ip); |
| 576 | xfs_irele(upd.ip); |
| 577 | } |
| 578 | out_path: |
| 579 | kfree(upd.path); |
| 580 | return error; |
| 581 | } |
| 582 | |
| 583 | /* Create the parent directory for all rtgroup inodes and load it. */ |
| 584 | int |
| 585 | xfs_rtginode_mkdir_parent( |
| 586 | struct xfs_mount *mp) |
| 587 | { |
| 588 | if (!mp->m_metadirip) { |
| 589 | xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR); |
| 590 | return -EFSCORRUPTED; |
| 591 | } |
| 592 | |
| 593 | return xfs_metadir_mkdir(dp: mp->m_metadirip, path: "rtgroups" , ipp: &mp->m_rtdirip); |
| 594 | } |
| 595 | |
| 596 | /* Load the parent directory of all rtgroup inodes. */ |
| 597 | int |
| 598 | xfs_rtginode_load_parent( |
| 599 | struct xfs_trans *tp) |
| 600 | { |
| 601 | struct xfs_mount *mp = tp->t_mountp; |
| 602 | |
| 603 | if (!mp->m_metadirip) { |
| 604 | xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR); |
| 605 | return -EFSCORRUPTED; |
| 606 | } |
| 607 | |
| 608 | return xfs_metadir_load(tp, dp: mp->m_metadirip, path: "rtgroups" , |
| 609 | metafile_type: XFS_METAFILE_DIR, ipp: &mp->m_rtdirip); |
| 610 | } |
| 611 | |
| 612 | /* Check superblock fields for a read or a write. */ |
| 613 | static xfs_failaddr_t |
| 614 | xfs_rtsb_verify_common( |
| 615 | struct xfs_buf *bp) |
| 616 | { |
| 617 | struct xfs_rtsb *rsb = bp->b_addr; |
| 618 | |
| 619 | if (!xfs_verify_magic(bp, rsb->rsb_magicnum)) |
| 620 | return __this_address; |
| 621 | if (rsb->rsb_pad) |
| 622 | return __this_address; |
| 623 | |
| 624 | /* Everything to the end of the fs block must be zero */ |
| 625 | if (memchr_inv(rsb + 1, 0, BBTOB(bp->b_length) - sizeof(*rsb))) |
| 626 | return __this_address; |
| 627 | |
| 628 | return NULL; |
| 629 | } |
| 630 | |
| 631 | /* Check superblock fields for a read or revalidation. */ |
| 632 | static inline xfs_failaddr_t |
| 633 | xfs_rtsb_verify_all( |
| 634 | struct xfs_buf *bp) |
| 635 | { |
| 636 | struct xfs_rtsb *rsb = bp->b_addr; |
| 637 | struct xfs_mount *mp = bp->b_mount; |
| 638 | xfs_failaddr_t fa; |
| 639 | |
| 640 | fa = xfs_rtsb_verify_common(bp); |
| 641 | if (fa) |
| 642 | return fa; |
| 643 | |
| 644 | if (memcmp(&rsb->rsb_fname, &mp->m_sb.sb_fname, XFSLABEL_MAX)) |
| 645 | return __this_address; |
| 646 | if (!uuid_equal(&rsb->rsb_uuid, &mp->m_sb.sb_uuid)) |
| 647 | return __this_address; |
| 648 | if (!uuid_equal(&rsb->rsb_meta_uuid, &mp->m_sb.sb_meta_uuid)) |
| 649 | return __this_address; |
| 650 | |
| 651 | return NULL; |
| 652 | } |
| 653 | |
| 654 | static void |
| 655 | xfs_rtsb_read_verify( |
| 656 | struct xfs_buf *bp) |
| 657 | { |
| 658 | xfs_failaddr_t fa; |
| 659 | |
| 660 | if (!xfs_buf_verify_cksum(bp, XFS_RTSB_CRC_OFF)) { |
| 661 | xfs_verifier_error(bp, -EFSBADCRC, __this_address); |
| 662 | return; |
| 663 | } |
| 664 | |
| 665 | fa = xfs_rtsb_verify_all(bp); |
| 666 | if (fa) |
| 667 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); |
| 668 | } |
| 669 | |
| 670 | static void |
| 671 | xfs_rtsb_write_verify( |
| 672 | struct xfs_buf *bp) |
| 673 | { |
| 674 | xfs_failaddr_t fa; |
| 675 | |
| 676 | fa = xfs_rtsb_verify_common(bp); |
| 677 | if (fa) { |
| 678 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); |
| 679 | return; |
| 680 | } |
| 681 | |
| 682 | xfs_buf_update_cksum(bp, XFS_RTSB_CRC_OFF); |
| 683 | } |
| 684 | |
| 685 | const struct xfs_buf_ops xfs_rtsb_buf_ops = { |
| 686 | .name = "xfs_rtsb" , |
| 687 | .magic = { 0, cpu_to_be32(XFS_RTSB_MAGIC) }, |
| 688 | .verify_read = xfs_rtsb_read_verify, |
| 689 | .verify_write = xfs_rtsb_write_verify, |
| 690 | .verify_struct = xfs_rtsb_verify_all, |
| 691 | }; |
| 692 | |
| 693 | /* Update a realtime superblock from the primary fs super */ |
| 694 | void |
| 695 | xfs_update_rtsb( |
| 696 | struct xfs_buf *rtsb_bp, |
| 697 | const struct xfs_buf *sb_bp) |
| 698 | { |
| 699 | const struct xfs_dsb *dsb = sb_bp->b_addr; |
| 700 | struct xfs_rtsb *rsb = rtsb_bp->b_addr; |
| 701 | const uuid_t *meta_uuid; |
| 702 | |
| 703 | rsb->rsb_magicnum = cpu_to_be32(XFS_RTSB_MAGIC); |
| 704 | |
| 705 | rsb->rsb_pad = 0; |
| 706 | memcpy(&rsb->rsb_fname, &dsb->sb_fname, XFSLABEL_MAX); |
| 707 | |
| 708 | memcpy(&rsb->rsb_uuid, &dsb->sb_uuid, sizeof(rsb->rsb_uuid)); |
| 709 | |
| 710 | /* |
| 711 | * The metadata uuid is the fs uuid if the metauuid feature is not |
| 712 | * enabled. |
| 713 | */ |
| 714 | if (dsb->sb_features_incompat & |
| 715 | cpu_to_be32(XFS_SB_FEAT_INCOMPAT_META_UUID)) |
| 716 | meta_uuid = &dsb->sb_meta_uuid; |
| 717 | else |
| 718 | meta_uuid = &dsb->sb_uuid; |
| 719 | memcpy(&rsb->rsb_meta_uuid, meta_uuid, sizeof(rsb->rsb_meta_uuid)); |
| 720 | } |
| 721 | |
| 722 | /* |
| 723 | * Update the realtime superblock from a filesystem superblock and log it to |
| 724 | * the given transaction. |
| 725 | */ |
| 726 | struct xfs_buf * |
| 727 | xfs_log_rtsb( |
| 728 | struct xfs_trans *tp, |
| 729 | const struct xfs_buf *sb_bp) |
| 730 | { |
| 731 | struct xfs_buf *rtsb_bp; |
| 732 | |
| 733 | if (!xfs_has_rtsb(tp->t_mountp)) |
| 734 | return NULL; |
| 735 | |
| 736 | rtsb_bp = xfs_trans_getrtsb(tp); |
| 737 | if (!rtsb_bp) { |
| 738 | /* |
| 739 | * It's possible for the rtgroups feature to be enabled but |
| 740 | * there is no incore rt superblock buffer if the rt geometry |
| 741 | * was specified at mkfs time but the rt section has not yet |
| 742 | * been attached. In this case, rblocks must be zero. |
| 743 | */ |
| 744 | ASSERT(tp->t_mountp->m_sb.sb_rblocks == 0); |
| 745 | return NULL; |
| 746 | } |
| 747 | |
| 748 | xfs_update_rtsb(rtsb_bp, sb_bp); |
| 749 | xfs_trans_ordered_buf(tp, rtsb_bp); |
| 750 | return rtsb_bp; |
| 751 | } |
| 752 | |