1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2016 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_log_format.h" 11 #include "xfs_trans_resv.h" 12 #include "xfs_mount.h" 13 #include "xfs_btree.h" 14 #include "xfs_btree_staging.h" 15 #include "xfs_refcount_btree.h" 16 #include "xfs_refcount.h" 17 #include "xfs_alloc.h" 18 #include "xfs_error.h" 19 #include "xfs_health.h" 20 #include "xfs_trace.h" 21 #include "xfs_trans.h" 22 #include "xfs_bit.h" 23 #include "xfs_rmap.h" 24 #include "xfs_ag.h" 25 26 static struct kmem_cache *xfs_refcountbt_cur_cache; 27 28 static struct xfs_btree_cur * 29 xfs_refcountbt_dup_cursor( 30 struct xfs_btree_cur *cur) 31 { 32 return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp, 33 cur->bc_ag.agbp, cur->bc_ag.pag); 34 } 35 36 STATIC void 37 xfs_refcountbt_set_root( 38 struct xfs_btree_cur *cur, 39 const union xfs_btree_ptr *ptr, 40 int inc) 41 { 42 struct xfs_buf *agbp = cur->bc_ag.agbp; 43 struct xfs_agf *agf = agbp->b_addr; 44 struct xfs_perag *pag = agbp->b_pag; 45 46 ASSERT(ptr->s != 0); 47 48 agf->agf_refcount_root = ptr->s; 49 be32_add_cpu(&agf->agf_refcount_level, inc); 50 pag->pagf_refcount_level += inc; 51 52 xfs_alloc_log_agf(cur->bc_tp, agbp, 53 XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL); 54 } 55 56 STATIC int 57 xfs_refcountbt_alloc_block( 58 struct xfs_btree_cur *cur, 59 const union xfs_btree_ptr *start, 60 union xfs_btree_ptr *new, 61 int *stat) 62 { 63 struct xfs_buf *agbp = cur->bc_ag.agbp; 64 struct xfs_agf *agf = agbp->b_addr; 65 struct xfs_alloc_arg args; /* block allocation args */ 66 int error; /* error return value */ 67 68 memset(&args, 0, sizeof(args)); 69 args.tp = cur->bc_tp; 70 args.mp = cur->bc_mp; 71 args.pag = cur->bc_ag.pag; 72 args.oinfo = XFS_RMAP_OINFO_REFC; 73 args.minlen = args.maxlen = args.prod = 1; 74 args.resv = XFS_AG_RESV_METADATA; 75 76 error = xfs_alloc_vextent_near_bno(&args, 77 XFS_AGB_TO_FSB(args.mp, args.pag->pag_agno, 78 xfs_refc_block(args.mp))); 79 if (error) 80 goto out_error; 81 if (args.fsbno == NULLFSBLOCK) { 82 *stat = 0; 83 return 0; 84 } 85 ASSERT(args.agno == cur->bc_ag.pag->pag_agno); 86 ASSERT(args.len == 1); 87 88 new->s = cpu_to_be32(args.agbno); 89 be32_add_cpu(&agf->agf_refcount_blocks, 1); 90 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS); 91 92 *stat = 1; 93 return 0; 94 95 out_error: 96 return error; 97 } 98 99 STATIC int 100 xfs_refcountbt_free_block( 101 struct xfs_btree_cur *cur, 102 struct xfs_buf *bp) 103 { 104 struct xfs_mount *mp = cur->bc_mp; 105 struct xfs_buf *agbp = cur->bc_ag.agbp; 106 struct xfs_agf *agf = agbp->b_addr; 107 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp)); 108 109 be32_add_cpu(&agf->agf_refcount_blocks, -1); 110 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS); 111 return xfs_free_extent_later(cur->bc_tp, fsbno, 1, 112 &XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, 0); 113 } 114 115 STATIC int 116 xfs_refcountbt_get_minrecs( 117 struct xfs_btree_cur *cur, 118 int level) 119 { 120 return cur->bc_mp->m_refc_mnr[level != 0]; 121 } 122 123 STATIC int 124 xfs_refcountbt_get_maxrecs( 125 struct xfs_btree_cur *cur, 126 int level) 127 { 128 return cur->bc_mp->m_refc_mxr[level != 0]; 129 } 130 131 STATIC void 132 xfs_refcountbt_init_key_from_rec( 133 union xfs_btree_key *key, 134 const union xfs_btree_rec *rec) 135 { 136 key->refc.rc_startblock = rec->refc.rc_startblock; 137 } 138 139 STATIC void 140 xfs_refcountbt_init_high_key_from_rec( 141 union xfs_btree_key *key, 142 const union xfs_btree_rec *rec) 143 { 144 __u32 x; 145 146 x = be32_to_cpu(rec->refc.rc_startblock); 147 x += be32_to_cpu(rec->refc.rc_blockcount) - 1; 148 key->refc.rc_startblock = cpu_to_be32(x); 149 } 150 151 STATIC void 152 xfs_refcountbt_init_rec_from_cur( 153 struct xfs_btree_cur *cur, 154 union xfs_btree_rec *rec) 155 { 156 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc; 157 uint32_t start; 158 159 start = xfs_refcount_encode_startblock(irec->rc_startblock, 160 irec->rc_domain); 161 rec->refc.rc_startblock = cpu_to_be32(start); 162 rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount); 163 rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount); 164 } 165 166 STATIC void 167 xfs_refcountbt_init_ptr_from_cur( 168 struct xfs_btree_cur *cur, 169 union xfs_btree_ptr *ptr) 170 { 171 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr; 172 173 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno)); 174 175 ptr->s = agf->agf_refcount_root; 176 } 177 178 STATIC int64_t 179 xfs_refcountbt_key_diff( 180 struct xfs_btree_cur *cur, 181 const union xfs_btree_key *key) 182 { 183 const struct xfs_refcount_key *kp = &key->refc; 184 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc; 185 uint32_t start; 186 187 start = xfs_refcount_encode_startblock(irec->rc_startblock, 188 irec->rc_domain); 189 return (int64_t)be32_to_cpu(kp->rc_startblock) - start; 190 } 191 192 STATIC int64_t 193 xfs_refcountbt_diff_two_keys( 194 struct xfs_btree_cur *cur, 195 const union xfs_btree_key *k1, 196 const union xfs_btree_key *k2, 197 const union xfs_btree_key *mask) 198 { 199 ASSERT(!mask || mask->refc.rc_startblock); 200 201 return (int64_t)be32_to_cpu(k1->refc.rc_startblock) - 202 be32_to_cpu(k2->refc.rc_startblock); 203 } 204 205 STATIC xfs_failaddr_t 206 xfs_refcountbt_verify( 207 struct xfs_buf *bp) 208 { 209 struct xfs_mount *mp = bp->b_mount; 210 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); 211 struct xfs_perag *pag = bp->b_pag; 212 xfs_failaddr_t fa; 213 unsigned int level; 214 215 if (!xfs_verify_magic(bp, block->bb_magic)) 216 return __this_address; 217 218 if (!xfs_has_reflink(mp)) 219 return __this_address; 220 fa = xfs_btree_agblock_v5hdr_verify(bp); 221 if (fa) 222 return fa; 223 224 level = be16_to_cpu(block->bb_level); 225 if (pag && xfs_perag_initialised_agf(pag)) { 226 unsigned int maxlevel = pag->pagf_refcount_level; 227 228 #ifdef CONFIG_XFS_ONLINE_REPAIR 229 /* 230 * Online repair could be rewriting the refcount btree, so 231 * we'll validate against the larger of either tree while this 232 * is going on. 233 */ 234 maxlevel = max_t(unsigned int, maxlevel, 235 pag->pagf_repair_refcount_level); 236 #endif 237 if (level >= maxlevel) 238 return __this_address; 239 } else if (level >= mp->m_refc_maxlevels) 240 return __this_address; 241 242 return xfs_btree_agblock_verify(bp, mp->m_refc_mxr[level != 0]); 243 } 244 245 STATIC void 246 xfs_refcountbt_read_verify( 247 struct xfs_buf *bp) 248 { 249 xfs_failaddr_t fa; 250 251 if (!xfs_btree_agblock_verify_crc(bp)) 252 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 253 else { 254 fa = xfs_refcountbt_verify(bp); 255 if (fa) 256 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 257 } 258 259 if (bp->b_error) 260 trace_xfs_btree_corrupt(bp, _RET_IP_); 261 } 262 263 STATIC void 264 xfs_refcountbt_write_verify( 265 struct xfs_buf *bp) 266 { 267 xfs_failaddr_t fa; 268 269 fa = xfs_refcountbt_verify(bp); 270 if (fa) { 271 trace_xfs_btree_corrupt(bp, _RET_IP_); 272 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 273 return; 274 } 275 xfs_btree_agblock_calc_crc(bp); 276 277 } 278 279 const struct xfs_buf_ops xfs_refcountbt_buf_ops = { 280 .name = "xfs_refcountbt", 281 .magic = { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) }, 282 .verify_read = xfs_refcountbt_read_verify, 283 .verify_write = xfs_refcountbt_write_verify, 284 .verify_struct = xfs_refcountbt_verify, 285 }; 286 287 STATIC int 288 xfs_refcountbt_keys_inorder( 289 struct xfs_btree_cur *cur, 290 const union xfs_btree_key *k1, 291 const union xfs_btree_key *k2) 292 { 293 return be32_to_cpu(k1->refc.rc_startblock) < 294 be32_to_cpu(k2->refc.rc_startblock); 295 } 296 297 STATIC int 298 xfs_refcountbt_recs_inorder( 299 struct xfs_btree_cur *cur, 300 const union xfs_btree_rec *r1, 301 const union xfs_btree_rec *r2) 302 { 303 return be32_to_cpu(r1->refc.rc_startblock) + 304 be32_to_cpu(r1->refc.rc_blockcount) <= 305 be32_to_cpu(r2->refc.rc_startblock); 306 } 307 308 STATIC enum xbtree_key_contig 309 xfs_refcountbt_keys_contiguous( 310 struct xfs_btree_cur *cur, 311 const union xfs_btree_key *key1, 312 const union xfs_btree_key *key2, 313 const union xfs_btree_key *mask) 314 { 315 ASSERT(!mask || mask->refc.rc_startblock); 316 317 return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock), 318 be32_to_cpu(key2->refc.rc_startblock)); 319 } 320 321 const struct xfs_btree_ops xfs_refcountbt_ops = { 322 .name = "refcount", 323 .type = XFS_BTREE_TYPE_AG, 324 325 .rec_len = sizeof(struct xfs_refcount_rec), 326 .key_len = sizeof(struct xfs_refcount_key), 327 .ptr_len = XFS_BTREE_SHORT_PTR_LEN, 328 329 .lru_refs = XFS_REFC_BTREE_REF, 330 .statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2), 331 .sick_mask = XFS_SICK_AG_REFCNTBT, 332 333 .dup_cursor = xfs_refcountbt_dup_cursor, 334 .set_root = xfs_refcountbt_set_root, 335 .alloc_block = xfs_refcountbt_alloc_block, 336 .free_block = xfs_refcountbt_free_block, 337 .get_minrecs = xfs_refcountbt_get_minrecs, 338 .get_maxrecs = xfs_refcountbt_get_maxrecs, 339 .init_key_from_rec = xfs_refcountbt_init_key_from_rec, 340 .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec, 341 .init_rec_from_cur = xfs_refcountbt_init_rec_from_cur, 342 .init_ptr_from_cur = xfs_refcountbt_init_ptr_from_cur, 343 .key_diff = xfs_refcountbt_key_diff, 344 .buf_ops = &xfs_refcountbt_buf_ops, 345 .diff_two_keys = xfs_refcountbt_diff_two_keys, 346 .keys_inorder = xfs_refcountbt_keys_inorder, 347 .recs_inorder = xfs_refcountbt_recs_inorder, 348 .keys_contiguous = xfs_refcountbt_keys_contiguous, 349 }; 350 351 /* 352 * Create a new refcount btree cursor. 353 * 354 * For staging cursors tp and agbp are NULL. 355 */ 356 struct xfs_btree_cur * 357 xfs_refcountbt_init_cursor( 358 struct xfs_mount *mp, 359 struct xfs_trans *tp, 360 struct xfs_buf *agbp, 361 struct xfs_perag *pag) 362 { 363 struct xfs_btree_cur *cur; 364 365 ASSERT(pag->pag_agno < mp->m_sb.sb_agcount); 366 367 cur = xfs_btree_alloc_cursor(mp, tp, &xfs_refcountbt_ops, 368 mp->m_refc_maxlevels, xfs_refcountbt_cur_cache); 369 cur->bc_ag.pag = xfs_perag_hold(pag); 370 cur->bc_refc.nr_ops = 0; 371 cur->bc_refc.shape_changes = 0; 372 cur->bc_ag.agbp = agbp; 373 if (agbp) { 374 struct xfs_agf *agf = agbp->b_addr; 375 376 cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level); 377 } 378 return cur; 379 } 380 381 /* 382 * Swap in the new btree root. Once we pass this point the newly rebuilt btree 383 * is in place and we have to kill off all the old btree blocks. 384 */ 385 void 386 xfs_refcountbt_commit_staged_btree( 387 struct xfs_btree_cur *cur, 388 struct xfs_trans *tp, 389 struct xfs_buf *agbp) 390 { 391 struct xfs_agf *agf = agbp->b_addr; 392 struct xbtree_afakeroot *afake = cur->bc_ag.afake; 393 394 ASSERT(cur->bc_flags & XFS_BTREE_STAGING); 395 396 agf->agf_refcount_root = cpu_to_be32(afake->af_root); 397 agf->agf_refcount_level = cpu_to_be32(afake->af_levels); 398 agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks); 399 xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS | 400 XFS_AGF_REFCOUNT_ROOT | 401 XFS_AGF_REFCOUNT_LEVEL); 402 xfs_btree_commit_afakeroot(cur, tp, agbp); 403 } 404 405 /* Calculate number of records in a refcount btree block. */ 406 static inline unsigned int 407 xfs_refcountbt_block_maxrecs( 408 unsigned int blocklen, 409 bool leaf) 410 { 411 if (leaf) 412 return blocklen / sizeof(struct xfs_refcount_rec); 413 return blocklen / (sizeof(struct xfs_refcount_key) + 414 sizeof(xfs_refcount_ptr_t)); 415 } 416 417 /* 418 * Calculate the number of records in a refcount btree block. 419 */ 420 unsigned int 421 xfs_refcountbt_maxrecs( 422 struct xfs_mount *mp, 423 unsigned int blocklen, 424 bool leaf) 425 { 426 blocklen -= XFS_REFCOUNT_BLOCK_LEN; 427 return xfs_refcountbt_block_maxrecs(blocklen, leaf); 428 } 429 430 /* Compute the max possible height of the maximally sized refcount btree. */ 431 unsigned int 432 xfs_refcountbt_maxlevels_ondisk(void) 433 { 434 unsigned int minrecs[2]; 435 unsigned int blocklen; 436 437 blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN; 438 439 minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2; 440 minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2; 441 442 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS); 443 } 444 445 /* Compute the maximum height of a refcount btree. */ 446 void 447 xfs_refcountbt_compute_maxlevels( 448 struct xfs_mount *mp) 449 { 450 if (!xfs_has_reflink(mp)) { 451 mp->m_refc_maxlevels = 0; 452 return; 453 } 454 455 mp->m_refc_maxlevels = xfs_btree_compute_maxlevels( 456 mp->m_refc_mnr, mp->m_sb.sb_agblocks); 457 ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk()); 458 } 459 460 /* Calculate the refcount btree size for some records. */ 461 xfs_extlen_t 462 xfs_refcountbt_calc_size( 463 struct xfs_mount *mp, 464 unsigned long long len) 465 { 466 return xfs_btree_calc_size(mp->m_refc_mnr, len); 467 } 468 469 /* 470 * Calculate the maximum refcount btree size. 471 */ 472 xfs_extlen_t 473 xfs_refcountbt_max_size( 474 struct xfs_mount *mp, 475 xfs_agblock_t agblocks) 476 { 477 /* Bail out if we're uninitialized, which can happen in mkfs. */ 478 if (mp->m_refc_mxr[0] == 0) 479 return 0; 480 481 return xfs_refcountbt_calc_size(mp, agblocks); 482 } 483 484 /* 485 * Figure out how many blocks to reserve and how many are used by this btree. 486 */ 487 int 488 xfs_refcountbt_calc_reserves( 489 struct xfs_mount *mp, 490 struct xfs_trans *tp, 491 struct xfs_perag *pag, 492 xfs_extlen_t *ask, 493 xfs_extlen_t *used) 494 { 495 struct xfs_buf *agbp; 496 struct xfs_agf *agf; 497 xfs_agblock_t agblocks; 498 xfs_extlen_t tree_len; 499 int error; 500 501 if (!xfs_has_reflink(mp)) 502 return 0; 503 504 error = xfs_alloc_read_agf(pag, tp, 0, &agbp); 505 if (error) 506 return error; 507 508 agf = agbp->b_addr; 509 agblocks = be32_to_cpu(agf->agf_length); 510 tree_len = be32_to_cpu(agf->agf_refcount_blocks); 511 xfs_trans_brelse(tp, agbp); 512 513 /* 514 * The log is permanently allocated, so the space it occupies will 515 * never be available for the kinds of things that would require btree 516 * expansion. We therefore can pretend the space isn't there. 517 */ 518 if (xfs_ag_contains_log(mp, pag->pag_agno)) 519 agblocks -= mp->m_sb.sb_logblocks; 520 521 *ask += xfs_refcountbt_max_size(mp, agblocks); 522 *used += tree_len; 523 524 return error; 525 } 526 527 int __init 528 xfs_refcountbt_init_cur_cache(void) 529 { 530 xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur", 531 xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()), 532 0, 0, NULL); 533 534 if (!xfs_refcountbt_cur_cache) 535 return -ENOMEM; 536 return 0; 537 } 538 539 void 540 xfs_refcountbt_destroy_cur_cache(void) 541 { 542 kmem_cache_destroy(xfs_refcountbt_cur_cache); 543 xfs_refcountbt_cur_cache = NULL; 544 } 545