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_trace.h" 20 #include "xfs_trans.h" 21 #include "xfs_bit.h" 22 #include "xfs_rmap.h" 23 #include "xfs_ag.h" 24 25 static struct kmem_cache *xfs_refcountbt_cur_cache; 26 27 static struct xfs_btree_cur * 28 xfs_refcountbt_dup_cursor( 29 struct xfs_btree_cur *cur) 30 { 31 return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp, 32 cur->bc_ag.agbp, cur->bc_ag.pag); 33 } 34 35 STATIC void 36 xfs_refcountbt_set_root( 37 struct xfs_btree_cur *cur, 38 const union xfs_btree_ptr *ptr, 39 int inc) 40 { 41 struct xfs_buf *agbp = cur->bc_ag.agbp; 42 struct xfs_agf *agf = agbp->b_addr; 43 struct xfs_perag *pag = agbp->b_pag; 44 45 ASSERT(ptr->s != 0); 46 47 agf->agf_refcount_root = ptr->s; 48 be32_add_cpu(&agf->agf_refcount_level, inc); 49 pag->pagf_refcount_level += inc; 50 51 xfs_alloc_log_agf(cur->bc_tp, agbp, 52 XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL); 53 } 54 55 STATIC int 56 xfs_refcountbt_alloc_block( 57 struct xfs_btree_cur *cur, 58 const union xfs_btree_ptr *start, 59 union xfs_btree_ptr *new, 60 int *stat) 61 { 62 struct xfs_buf *agbp = cur->bc_ag.agbp; 63 struct xfs_agf *agf = agbp->b_addr; 64 struct xfs_alloc_arg args; /* block allocation args */ 65 int error; /* error return value */ 66 67 memset(&args, 0, sizeof(args)); 68 args.tp = cur->bc_tp; 69 args.mp = cur->bc_mp; 70 args.pag = cur->bc_ag.pag; 71 args.type = XFS_ALLOCTYPE_NEAR_BNO; 72 args.fsbno = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_ag.pag->pag_agno, 73 xfs_refc_block(args.mp)); 74 args.oinfo = XFS_RMAP_OINFO_REFC; 75 args.minlen = args.maxlen = args.prod = 1; 76 args.resv = XFS_AG_RESV_METADATA; 77 78 error = xfs_alloc_vextent_this_ag(&args); 79 if (error) 80 goto out_error; 81 trace_xfs_refcountbt_alloc_block(cur->bc_mp, cur->bc_ag.pag->pag_agno, 82 args.agbno, 1); 83 if (args.fsbno == NULLFSBLOCK) { 84 *stat = 0; 85 return 0; 86 } 87 ASSERT(args.agno == cur->bc_ag.pag->pag_agno); 88 ASSERT(args.len == 1); 89 90 new->s = cpu_to_be32(args.agbno); 91 be32_add_cpu(&agf->agf_refcount_blocks, 1); 92 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS); 93 94 *stat = 1; 95 return 0; 96 97 out_error: 98 return error; 99 } 100 101 STATIC int 102 xfs_refcountbt_free_block( 103 struct xfs_btree_cur *cur, 104 struct xfs_buf *bp) 105 { 106 struct xfs_mount *mp = cur->bc_mp; 107 struct xfs_buf *agbp = cur->bc_ag.agbp; 108 struct xfs_agf *agf = agbp->b_addr; 109 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp)); 110 int error; 111 112 trace_xfs_refcountbt_free_block(cur->bc_mp, cur->bc_ag.pag->pag_agno, 113 XFS_FSB_TO_AGBNO(cur->bc_mp, fsbno), 1); 114 be32_add_cpu(&agf->agf_refcount_blocks, -1); 115 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS); 116 error = xfs_free_extent(cur->bc_tp, fsbno, 1, &XFS_RMAP_OINFO_REFC, 117 XFS_AG_RESV_METADATA); 118 if (error) 119 return error; 120 121 return error; 122 } 123 124 STATIC int 125 xfs_refcountbt_get_minrecs( 126 struct xfs_btree_cur *cur, 127 int level) 128 { 129 return cur->bc_mp->m_refc_mnr[level != 0]; 130 } 131 132 STATIC int 133 xfs_refcountbt_get_maxrecs( 134 struct xfs_btree_cur *cur, 135 int level) 136 { 137 return cur->bc_mp->m_refc_mxr[level != 0]; 138 } 139 140 STATIC void 141 xfs_refcountbt_init_key_from_rec( 142 union xfs_btree_key *key, 143 const union xfs_btree_rec *rec) 144 { 145 key->refc.rc_startblock = rec->refc.rc_startblock; 146 } 147 148 STATIC void 149 xfs_refcountbt_init_high_key_from_rec( 150 union xfs_btree_key *key, 151 const union xfs_btree_rec *rec) 152 { 153 __u32 x; 154 155 x = be32_to_cpu(rec->refc.rc_startblock); 156 x += be32_to_cpu(rec->refc.rc_blockcount) - 1; 157 key->refc.rc_startblock = cpu_to_be32(x); 158 } 159 160 STATIC void 161 xfs_refcountbt_init_rec_from_cur( 162 struct xfs_btree_cur *cur, 163 union xfs_btree_rec *rec) 164 { 165 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc; 166 uint32_t start; 167 168 start = xfs_refcount_encode_startblock(irec->rc_startblock, 169 irec->rc_domain); 170 rec->refc.rc_startblock = cpu_to_be32(start); 171 rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount); 172 rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount); 173 } 174 175 STATIC void 176 xfs_refcountbt_init_ptr_from_cur( 177 struct xfs_btree_cur *cur, 178 union xfs_btree_ptr *ptr) 179 { 180 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr; 181 182 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno)); 183 184 ptr->s = agf->agf_refcount_root; 185 } 186 187 STATIC int64_t 188 xfs_refcountbt_key_diff( 189 struct xfs_btree_cur *cur, 190 const union xfs_btree_key *key) 191 { 192 const struct xfs_refcount_key *kp = &key->refc; 193 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc; 194 uint32_t start; 195 196 start = xfs_refcount_encode_startblock(irec->rc_startblock, 197 irec->rc_domain); 198 return (int64_t)be32_to_cpu(kp->rc_startblock) - start; 199 } 200 201 STATIC int64_t 202 xfs_refcountbt_diff_two_keys( 203 struct xfs_btree_cur *cur, 204 const union xfs_btree_key *k1, 205 const union xfs_btree_key *k2) 206 { 207 return (int64_t)be32_to_cpu(k1->refc.rc_startblock) - 208 be32_to_cpu(k2->refc.rc_startblock); 209 } 210 211 STATIC xfs_failaddr_t 212 xfs_refcountbt_verify( 213 struct xfs_buf *bp) 214 { 215 struct xfs_mount *mp = bp->b_mount; 216 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); 217 struct xfs_perag *pag = bp->b_pag; 218 xfs_failaddr_t fa; 219 unsigned int level; 220 221 if (!xfs_verify_magic(bp, block->bb_magic)) 222 return __this_address; 223 224 if (!xfs_has_reflink(mp)) 225 return __this_address; 226 fa = xfs_btree_sblock_v5hdr_verify(bp); 227 if (fa) 228 return fa; 229 230 level = be16_to_cpu(block->bb_level); 231 if (pag && xfs_perag_initialised_agf(pag)) { 232 if (level >= pag->pagf_refcount_level) 233 return __this_address; 234 } else if (level >= mp->m_refc_maxlevels) 235 return __this_address; 236 237 return xfs_btree_sblock_verify(bp, mp->m_refc_mxr[level != 0]); 238 } 239 240 STATIC void 241 xfs_refcountbt_read_verify( 242 struct xfs_buf *bp) 243 { 244 xfs_failaddr_t fa; 245 246 if (!xfs_btree_sblock_verify_crc(bp)) 247 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 248 else { 249 fa = xfs_refcountbt_verify(bp); 250 if (fa) 251 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 252 } 253 254 if (bp->b_error) 255 trace_xfs_btree_corrupt(bp, _RET_IP_); 256 } 257 258 STATIC void 259 xfs_refcountbt_write_verify( 260 struct xfs_buf *bp) 261 { 262 xfs_failaddr_t fa; 263 264 fa = xfs_refcountbt_verify(bp); 265 if (fa) { 266 trace_xfs_btree_corrupt(bp, _RET_IP_); 267 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 268 return; 269 } 270 xfs_btree_sblock_calc_crc(bp); 271 272 } 273 274 const struct xfs_buf_ops xfs_refcountbt_buf_ops = { 275 .name = "xfs_refcountbt", 276 .magic = { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) }, 277 .verify_read = xfs_refcountbt_read_verify, 278 .verify_write = xfs_refcountbt_write_verify, 279 .verify_struct = xfs_refcountbt_verify, 280 }; 281 282 STATIC int 283 xfs_refcountbt_keys_inorder( 284 struct xfs_btree_cur *cur, 285 const union xfs_btree_key *k1, 286 const union xfs_btree_key *k2) 287 { 288 return be32_to_cpu(k1->refc.rc_startblock) < 289 be32_to_cpu(k2->refc.rc_startblock); 290 } 291 292 STATIC int 293 xfs_refcountbt_recs_inorder( 294 struct xfs_btree_cur *cur, 295 const union xfs_btree_rec *r1, 296 const union xfs_btree_rec *r2) 297 { 298 return be32_to_cpu(r1->refc.rc_startblock) + 299 be32_to_cpu(r1->refc.rc_blockcount) <= 300 be32_to_cpu(r2->refc.rc_startblock); 301 } 302 303 static const struct xfs_btree_ops xfs_refcountbt_ops = { 304 .rec_len = sizeof(struct xfs_refcount_rec), 305 .key_len = sizeof(struct xfs_refcount_key), 306 307 .dup_cursor = xfs_refcountbt_dup_cursor, 308 .set_root = xfs_refcountbt_set_root, 309 .alloc_block = xfs_refcountbt_alloc_block, 310 .free_block = xfs_refcountbt_free_block, 311 .get_minrecs = xfs_refcountbt_get_minrecs, 312 .get_maxrecs = xfs_refcountbt_get_maxrecs, 313 .init_key_from_rec = xfs_refcountbt_init_key_from_rec, 314 .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec, 315 .init_rec_from_cur = xfs_refcountbt_init_rec_from_cur, 316 .init_ptr_from_cur = xfs_refcountbt_init_ptr_from_cur, 317 .key_diff = xfs_refcountbt_key_diff, 318 .buf_ops = &xfs_refcountbt_buf_ops, 319 .diff_two_keys = xfs_refcountbt_diff_two_keys, 320 .keys_inorder = xfs_refcountbt_keys_inorder, 321 .recs_inorder = xfs_refcountbt_recs_inorder, 322 }; 323 324 /* 325 * Initialize a new refcount btree cursor. 326 */ 327 static struct xfs_btree_cur * 328 xfs_refcountbt_init_common( 329 struct xfs_mount *mp, 330 struct xfs_trans *tp, 331 struct xfs_perag *pag) 332 { 333 struct xfs_btree_cur *cur; 334 335 ASSERT(pag->pag_agno < mp->m_sb.sb_agcount); 336 337 cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_REFC, 338 mp->m_refc_maxlevels, xfs_refcountbt_cur_cache); 339 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2); 340 341 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS; 342 343 /* take a reference for the cursor */ 344 atomic_inc(&pag->pag_ref); 345 cur->bc_ag.pag = pag; 346 347 cur->bc_ag.refc.nr_ops = 0; 348 cur->bc_ag.refc.shape_changes = 0; 349 cur->bc_ops = &xfs_refcountbt_ops; 350 return cur; 351 } 352 353 /* Create a btree cursor. */ 354 struct xfs_btree_cur * 355 xfs_refcountbt_init_cursor( 356 struct xfs_mount *mp, 357 struct xfs_trans *tp, 358 struct xfs_buf *agbp, 359 struct xfs_perag *pag) 360 { 361 struct xfs_agf *agf = agbp->b_addr; 362 struct xfs_btree_cur *cur; 363 364 cur = xfs_refcountbt_init_common(mp, tp, pag); 365 cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level); 366 cur->bc_ag.agbp = agbp; 367 return cur; 368 } 369 370 /* Create a btree cursor with a fake root for staging. */ 371 struct xfs_btree_cur * 372 xfs_refcountbt_stage_cursor( 373 struct xfs_mount *mp, 374 struct xbtree_afakeroot *afake, 375 struct xfs_perag *pag) 376 { 377 struct xfs_btree_cur *cur; 378 379 cur = xfs_refcountbt_init_common(mp, NULL, pag); 380 xfs_btree_stage_afakeroot(cur, afake); 381 return cur; 382 } 383 384 /* 385 * Swap in the new btree root. Once we pass this point the newly rebuilt btree 386 * is in place and we have to kill off all the old btree blocks. 387 */ 388 void 389 xfs_refcountbt_commit_staged_btree( 390 struct xfs_btree_cur *cur, 391 struct xfs_trans *tp, 392 struct xfs_buf *agbp) 393 { 394 struct xfs_agf *agf = agbp->b_addr; 395 struct xbtree_afakeroot *afake = cur->bc_ag.afake; 396 397 ASSERT(cur->bc_flags & XFS_BTREE_STAGING); 398 399 agf->agf_refcount_root = cpu_to_be32(afake->af_root); 400 agf->agf_refcount_level = cpu_to_be32(afake->af_levels); 401 agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks); 402 xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS | 403 XFS_AGF_REFCOUNT_ROOT | 404 XFS_AGF_REFCOUNT_LEVEL); 405 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_refcountbt_ops); 406 } 407 408 /* Calculate number of records in a refcount btree block. */ 409 static inline unsigned int 410 xfs_refcountbt_block_maxrecs( 411 unsigned int blocklen, 412 bool leaf) 413 { 414 if (leaf) 415 return blocklen / sizeof(struct xfs_refcount_rec); 416 return blocklen / (sizeof(struct xfs_refcount_key) + 417 sizeof(xfs_refcount_ptr_t)); 418 } 419 420 /* 421 * Calculate the number of records in a refcount btree block. 422 */ 423 int 424 xfs_refcountbt_maxrecs( 425 int blocklen, 426 bool leaf) 427 { 428 blocklen -= XFS_REFCOUNT_BLOCK_LEN; 429 return xfs_refcountbt_block_maxrecs(blocklen, leaf); 430 } 431 432 /* Compute the max possible height of the maximally sized refcount btree. */ 433 unsigned int 434 xfs_refcountbt_maxlevels_ondisk(void) 435 { 436 unsigned int minrecs[2]; 437 unsigned int blocklen; 438 439 blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN; 440 441 minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2; 442 minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2; 443 444 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS); 445 } 446 447 /* Compute the maximum height of a refcount btree. */ 448 void 449 xfs_refcountbt_compute_maxlevels( 450 struct xfs_mount *mp) 451 { 452 if (!xfs_has_reflink(mp)) { 453 mp->m_refc_maxlevels = 0; 454 return; 455 } 456 457 mp->m_refc_maxlevels = xfs_btree_compute_maxlevels( 458 mp->m_refc_mnr, mp->m_sb.sb_agblocks); 459 ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk()); 460 } 461 462 /* Calculate the refcount btree size for some records. */ 463 xfs_extlen_t 464 xfs_refcountbt_calc_size( 465 struct xfs_mount *mp, 466 unsigned long long len) 467 { 468 return xfs_btree_calc_size(mp->m_refc_mnr, len); 469 } 470 471 /* 472 * Calculate the maximum refcount btree size. 473 */ 474 xfs_extlen_t 475 xfs_refcountbt_max_size( 476 struct xfs_mount *mp, 477 xfs_agblock_t agblocks) 478 { 479 /* Bail out if we're uninitialized, which can happen in mkfs. */ 480 if (mp->m_refc_mxr[0] == 0) 481 return 0; 482 483 return xfs_refcountbt_calc_size(mp, agblocks); 484 } 485 486 /* 487 * Figure out how many blocks to reserve and how many are used by this btree. 488 */ 489 int 490 xfs_refcountbt_calc_reserves( 491 struct xfs_mount *mp, 492 struct xfs_trans *tp, 493 struct xfs_perag *pag, 494 xfs_extlen_t *ask, 495 xfs_extlen_t *used) 496 { 497 struct xfs_buf *agbp; 498 struct xfs_agf *agf; 499 xfs_agblock_t agblocks; 500 xfs_extlen_t tree_len; 501 int error; 502 503 if (!xfs_has_reflink(mp)) 504 return 0; 505 506 error = xfs_alloc_read_agf(pag, tp, 0, &agbp); 507 if (error) 508 return error; 509 510 agf = agbp->b_addr; 511 agblocks = be32_to_cpu(agf->agf_length); 512 tree_len = be32_to_cpu(agf->agf_refcount_blocks); 513 xfs_trans_brelse(tp, agbp); 514 515 /* 516 * The log is permanently allocated, so the space it occupies will 517 * never be available for the kinds of things that would require btree 518 * expansion. We therefore can pretend the space isn't there. 519 */ 520 if (xfs_ag_contains_log(mp, pag->pag_agno)) 521 agblocks -= mp->m_sb.sb_logblocks; 522 523 *ask += xfs_refcountbt_max_size(mp, agblocks); 524 *used += tree_len; 525 526 return error; 527 } 528 529 int __init 530 xfs_refcountbt_init_cur_cache(void) 531 { 532 xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur", 533 xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()), 534 0, 0, NULL); 535 536 if (!xfs_refcountbt_cur_cache) 537 return -ENOMEM; 538 return 0; 539 } 540 541 void 542 xfs_refcountbt_destroy_cur_cache(void) 543 { 544 kmem_cache_destroy(xfs_refcountbt_cur_cache); 545 xfs_refcountbt_cur_cache = NULL; 546 } 547