1 /* 2 * Copyright (C) 2017 Oracle. All Rights Reserved. 3 * 4 * Author: Darrick J. Wong <darrick.wong@oracle.com> 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 2 9 * of the License, or (at your option) any later version. 10 * 11 * This program is distributed in the hope that it would be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, write the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. 19 */ 20 #include "xfs.h" 21 #include "xfs_fs.h" 22 #include "xfs_shared.h" 23 #include "xfs_format.h" 24 #include "xfs_trans_resv.h" 25 #include "xfs_mount.h" 26 #include "xfs_defer.h" 27 #include "xfs_btree.h" 28 #include "xfs_bit.h" 29 #include "xfs_log_format.h" 30 #include "xfs_trans.h" 31 #include "xfs_sb.h" 32 #include "xfs_alloc.h" 33 #include "xfs_rmap.h" 34 #include "xfs_refcount.h" 35 #include "scrub/xfs_scrub.h" 36 #include "scrub/scrub.h" 37 #include "scrub/common.h" 38 #include "scrub/btree.h" 39 #include "scrub/trace.h" 40 41 /* 42 * Set us up to scrub reference count btrees. 43 */ 44 int 45 xfs_scrub_setup_ag_refcountbt( 46 struct xfs_scrub_context *sc, 47 struct xfs_inode *ip) 48 { 49 return xfs_scrub_setup_ag_btree(sc, ip, false); 50 } 51 52 /* Reference count btree scrubber. */ 53 54 /* 55 * Confirming Reference Counts via Reverse Mappings 56 * 57 * We want to count the reverse mappings overlapping a refcount record 58 * (bno, len, refcount), allowing for the possibility that some of the 59 * overlap may come from smaller adjoining reverse mappings, while some 60 * comes from single extents which overlap the range entirely. The 61 * outer loop is as follows: 62 * 63 * 1. For all reverse mappings overlapping the refcount extent, 64 * a. If a given rmap completely overlaps, mark it as seen. 65 * b. Otherwise, record the fragment (in agbno order) for later 66 * processing. 67 * 68 * Once we've seen all the rmaps, we know that for all blocks in the 69 * refcount record we want to find $refcount owners and we've already 70 * visited $seen extents that overlap all the blocks. Therefore, we 71 * need to find ($refcount - $seen) owners for every block in the 72 * extent; call that quantity $target_nr. Proceed as follows: 73 * 74 * 2. Pull the first $target_nr fragments from the list; all of them 75 * should start at or before the start of the extent. 76 * Call this subset of fragments the working set. 77 * 3. Until there are no more unprocessed fragments, 78 * a. Find the shortest fragments in the set and remove them. 79 * b. Note the block number of the end of these fragments. 80 * c. Pull the same number of fragments from the list. All of these 81 * fragments should start at the block number recorded in the 82 * previous step. 83 * d. Put those fragments in the set. 84 * 4. Check that there are $target_nr fragments remaining in the list, 85 * and that they all end at or beyond the end of the refcount extent. 86 * 87 * If the refcount is correct, all the check conditions in the algorithm 88 * should always hold true. If not, the refcount is incorrect. 89 */ 90 struct xfs_scrub_refcnt_frag { 91 struct list_head list; 92 struct xfs_rmap_irec rm; 93 }; 94 95 struct xfs_scrub_refcnt_check { 96 struct xfs_scrub_context *sc; 97 struct list_head fragments; 98 99 /* refcount extent we're examining */ 100 xfs_agblock_t bno; 101 xfs_extlen_t len; 102 xfs_nlink_t refcount; 103 104 /* number of owners seen */ 105 xfs_nlink_t seen; 106 }; 107 108 /* 109 * Decide if the given rmap is large enough that we can redeem it 110 * towards refcount verification now, or if it's a fragment, in 111 * which case we'll hang onto it in the hopes that we'll later 112 * discover that we've collected exactly the correct number of 113 * fragments as the refcountbt says we should have. 114 */ 115 STATIC int 116 xfs_scrub_refcountbt_rmap_check( 117 struct xfs_btree_cur *cur, 118 struct xfs_rmap_irec *rec, 119 void *priv) 120 { 121 struct xfs_scrub_refcnt_check *refchk = priv; 122 struct xfs_scrub_refcnt_frag *frag; 123 xfs_agblock_t rm_last; 124 xfs_agblock_t rc_last; 125 int error = 0; 126 127 if (xfs_scrub_should_terminate(refchk->sc, &error)) 128 return error; 129 130 rm_last = rec->rm_startblock + rec->rm_blockcount - 1; 131 rc_last = refchk->bno + refchk->len - 1; 132 133 /* Confirm that a single-owner refc extent is a CoW stage. */ 134 if (refchk->refcount == 1 && rec->rm_owner != XFS_RMAP_OWN_COW) { 135 xfs_scrub_btree_xref_set_corrupt(refchk->sc, cur, 0); 136 return 0; 137 } 138 139 if (rec->rm_startblock <= refchk->bno && rm_last >= rc_last) { 140 /* 141 * The rmap overlaps the refcount record, so we can confirm 142 * one refcount owner seen. 143 */ 144 refchk->seen++; 145 } else { 146 /* 147 * This rmap covers only part of the refcount record, so 148 * save the fragment for later processing. If the rmapbt 149 * is healthy each rmap_irec we see will be in agbno order 150 * so we don't need insertion sort here. 151 */ 152 frag = kmem_alloc(sizeof(struct xfs_scrub_refcnt_frag), 153 KM_MAYFAIL | KM_NOFS); 154 if (!frag) 155 return -ENOMEM; 156 memcpy(&frag->rm, rec, sizeof(frag->rm)); 157 list_add_tail(&frag->list, &refchk->fragments); 158 } 159 160 return 0; 161 } 162 163 /* 164 * Given a bunch of rmap fragments, iterate through them, keeping 165 * a running tally of the refcount. If this ever deviates from 166 * what we expect (which is the refcountbt's refcount minus the 167 * number of extents that totally covered the refcountbt extent), 168 * we have a refcountbt error. 169 */ 170 STATIC void 171 xfs_scrub_refcountbt_process_rmap_fragments( 172 struct xfs_scrub_refcnt_check *refchk) 173 { 174 struct list_head worklist; 175 struct xfs_scrub_refcnt_frag *frag; 176 struct xfs_scrub_refcnt_frag *n; 177 xfs_agblock_t bno; 178 xfs_agblock_t rbno; 179 xfs_agblock_t next_rbno; 180 xfs_nlink_t nr; 181 xfs_nlink_t target_nr; 182 183 target_nr = refchk->refcount - refchk->seen; 184 if (target_nr == 0) 185 return; 186 187 /* 188 * There are (refchk->rc.rc_refcount - refchk->nr refcount) 189 * references we haven't found yet. Pull that many off the 190 * fragment list and figure out where the smallest rmap ends 191 * (and therefore the next rmap should start). All the rmaps 192 * we pull off should start at or before the beginning of the 193 * refcount record's range. 194 */ 195 INIT_LIST_HEAD(&worklist); 196 rbno = NULLAGBLOCK; 197 nr = 1; 198 199 /* Make sure the fragments actually /are/ in agbno order. */ 200 bno = 0; 201 list_for_each_entry(frag, &refchk->fragments, list) { 202 if (frag->rm.rm_startblock < bno) 203 goto done; 204 bno = frag->rm.rm_startblock; 205 } 206 207 /* 208 * Find all the rmaps that start at or before the refc extent, 209 * and put them on the worklist. 210 */ 211 list_for_each_entry_safe(frag, n, &refchk->fragments, list) { 212 if (frag->rm.rm_startblock > refchk->bno) 213 goto done; 214 bno = frag->rm.rm_startblock + frag->rm.rm_blockcount; 215 if (bno < rbno) 216 rbno = bno; 217 list_move_tail(&frag->list, &worklist); 218 if (nr == target_nr) 219 break; 220 nr++; 221 } 222 223 /* 224 * We should have found exactly $target_nr rmap fragments starting 225 * at or before the refcount extent. 226 */ 227 if (nr != target_nr) 228 goto done; 229 230 while (!list_empty(&refchk->fragments)) { 231 /* Discard any fragments ending at rbno from the worklist. */ 232 nr = 0; 233 next_rbno = NULLAGBLOCK; 234 list_for_each_entry_safe(frag, n, &worklist, list) { 235 bno = frag->rm.rm_startblock + frag->rm.rm_blockcount; 236 if (bno != rbno) { 237 if (bno < next_rbno) 238 next_rbno = bno; 239 continue; 240 } 241 list_del(&frag->list); 242 kmem_free(frag); 243 nr++; 244 } 245 246 /* Try to add nr rmaps starting at rbno to the worklist. */ 247 list_for_each_entry_safe(frag, n, &refchk->fragments, list) { 248 bno = frag->rm.rm_startblock + frag->rm.rm_blockcount; 249 if (frag->rm.rm_startblock != rbno) 250 goto done; 251 list_move_tail(&frag->list, &worklist); 252 if (next_rbno > bno) 253 next_rbno = bno; 254 nr--; 255 if (nr == 0) 256 break; 257 } 258 259 /* 260 * If we get here and nr > 0, this means that we added fewer 261 * items to the worklist than we discarded because the fragment 262 * list ran out of items. Therefore, we cannot maintain the 263 * required refcount. Something is wrong, so we're done. 264 */ 265 if (nr) 266 goto done; 267 268 rbno = next_rbno; 269 } 270 271 /* 272 * Make sure the last extent we processed ends at or beyond 273 * the end of the refcount extent. 274 */ 275 if (rbno < refchk->bno + refchk->len) 276 goto done; 277 278 /* Actually record us having seen the remaining refcount. */ 279 refchk->seen = refchk->refcount; 280 done: 281 /* Delete fragments and work list. */ 282 list_for_each_entry_safe(frag, n, &worklist, list) { 283 list_del(&frag->list); 284 kmem_free(frag); 285 } 286 list_for_each_entry_safe(frag, n, &refchk->fragments, list) { 287 list_del(&frag->list); 288 kmem_free(frag); 289 } 290 } 291 292 /* Use the rmap entries covering this extent to verify the refcount. */ 293 STATIC void 294 xfs_scrub_refcountbt_xref_rmap( 295 struct xfs_scrub_context *sc, 296 xfs_agblock_t bno, 297 xfs_extlen_t len, 298 xfs_nlink_t refcount) 299 { 300 struct xfs_scrub_refcnt_check refchk = { 301 .sc = sc, 302 .bno = bno, 303 .len = len, 304 .refcount = refcount, 305 .seen = 0, 306 }; 307 struct xfs_rmap_irec low; 308 struct xfs_rmap_irec high; 309 struct xfs_scrub_refcnt_frag *frag; 310 struct xfs_scrub_refcnt_frag *n; 311 int error; 312 313 if (!sc->sa.rmap_cur) 314 return; 315 316 /* Cross-reference with the rmapbt to confirm the refcount. */ 317 memset(&low, 0, sizeof(low)); 318 low.rm_startblock = bno; 319 memset(&high, 0xFF, sizeof(high)); 320 high.rm_startblock = bno + len - 1; 321 322 INIT_LIST_HEAD(&refchk.fragments); 323 error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high, 324 &xfs_scrub_refcountbt_rmap_check, &refchk); 325 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur)) 326 goto out_free; 327 328 xfs_scrub_refcountbt_process_rmap_fragments(&refchk); 329 if (refcount != refchk.seen) 330 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0); 331 332 out_free: 333 list_for_each_entry_safe(frag, n, &refchk.fragments, list) { 334 list_del(&frag->list); 335 kmem_free(frag); 336 } 337 } 338 339 /* Cross-reference with the other btrees. */ 340 STATIC void 341 xfs_scrub_refcountbt_xref( 342 struct xfs_scrub_context *sc, 343 xfs_agblock_t agbno, 344 xfs_extlen_t len, 345 xfs_nlink_t refcount) 346 { 347 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) 348 return; 349 350 xfs_scrub_xref_is_used_space(sc, agbno, len); 351 xfs_scrub_xref_is_not_inode_chunk(sc, agbno, len); 352 xfs_scrub_refcountbt_xref_rmap(sc, agbno, len, refcount); 353 } 354 355 /* Scrub a refcountbt record. */ 356 STATIC int 357 xfs_scrub_refcountbt_rec( 358 struct xfs_scrub_btree *bs, 359 union xfs_btree_rec *rec) 360 { 361 struct xfs_mount *mp = bs->cur->bc_mp; 362 xfs_agblock_t *cow_blocks = bs->private; 363 xfs_agnumber_t agno = bs->cur->bc_private.a.agno; 364 xfs_agblock_t bno; 365 xfs_extlen_t len; 366 xfs_nlink_t refcount; 367 bool has_cowflag; 368 int error = 0; 369 370 bno = be32_to_cpu(rec->refc.rc_startblock); 371 len = be32_to_cpu(rec->refc.rc_blockcount); 372 refcount = be32_to_cpu(rec->refc.rc_refcount); 373 374 /* Only CoW records can have refcount == 1. */ 375 has_cowflag = (bno & XFS_REFC_COW_START); 376 if ((refcount == 1 && !has_cowflag) || (refcount != 1 && has_cowflag)) 377 xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0); 378 if (has_cowflag) 379 (*cow_blocks) += len; 380 381 /* Check the extent. */ 382 bno &= ~XFS_REFC_COW_START; 383 if (bno + len <= bno || 384 !xfs_verify_agbno(mp, agno, bno) || 385 !xfs_verify_agbno(mp, agno, bno + len - 1)) 386 xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0); 387 388 if (refcount == 0) 389 xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0); 390 391 xfs_scrub_refcountbt_xref(bs->sc, bno, len, refcount); 392 393 return error; 394 } 395 396 /* Make sure we have as many refc blocks as the rmap says. */ 397 STATIC void 398 xfs_scrub_refcount_xref_rmap( 399 struct xfs_scrub_context *sc, 400 struct xfs_owner_info *oinfo, 401 xfs_filblks_t cow_blocks) 402 { 403 xfs_extlen_t refcbt_blocks = 0; 404 xfs_filblks_t blocks; 405 int error; 406 407 if (!sc->sa.rmap_cur) 408 return; 409 410 /* Check that we saw as many refcbt blocks as the rmap knows about. */ 411 error = xfs_btree_count_blocks(sc->sa.refc_cur, &refcbt_blocks); 412 if (!xfs_scrub_btree_process_error(sc, sc->sa.refc_cur, 0, &error)) 413 return; 414 error = xfs_scrub_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur, oinfo, 415 &blocks); 416 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur)) 417 return; 418 if (blocks != refcbt_blocks) 419 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0); 420 421 /* Check that we saw as many cow blocks as the rmap knows about. */ 422 xfs_rmap_ag_owner(oinfo, XFS_RMAP_OWN_COW); 423 error = xfs_scrub_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur, oinfo, 424 &blocks); 425 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur)) 426 return; 427 if (blocks != cow_blocks) 428 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0); 429 } 430 431 /* Scrub the refcount btree for some AG. */ 432 int 433 xfs_scrub_refcountbt( 434 struct xfs_scrub_context *sc) 435 { 436 struct xfs_owner_info oinfo; 437 xfs_agblock_t cow_blocks = 0; 438 int error; 439 440 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_REFC); 441 error = xfs_scrub_btree(sc, sc->sa.refc_cur, xfs_scrub_refcountbt_rec, 442 &oinfo, &cow_blocks); 443 if (error) 444 return error; 445 446 xfs_scrub_refcount_xref_rmap(sc, &oinfo, cow_blocks); 447 448 return 0; 449 } 450 451 /* xref check that a cow staging extent is marked in the refcountbt. */ 452 void 453 xfs_scrub_xref_is_cow_staging( 454 struct xfs_scrub_context *sc, 455 xfs_agblock_t agbno, 456 xfs_extlen_t len) 457 { 458 struct xfs_refcount_irec rc; 459 bool has_cowflag; 460 int has_refcount; 461 int error; 462 463 if (!sc->sa.refc_cur) 464 return; 465 466 /* Find the CoW staging extent. */ 467 error = xfs_refcount_lookup_le(sc->sa.refc_cur, 468 agbno + XFS_REFC_COW_START, &has_refcount); 469 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur)) 470 return; 471 if (!has_refcount) { 472 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0); 473 return; 474 } 475 476 error = xfs_refcount_get_rec(sc->sa.refc_cur, &rc, &has_refcount); 477 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur)) 478 return; 479 if (!has_refcount) { 480 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0); 481 return; 482 } 483 484 /* CoW flag must be set, refcount must be 1. */ 485 has_cowflag = (rc.rc_startblock & XFS_REFC_COW_START); 486 if (!has_cowflag || rc.rc_refcount != 1) 487 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0); 488 489 /* Must be at least as long as what was passed in */ 490 if (rc.rc_blockcount < len) 491 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0); 492 } 493 494 /* 495 * xref check that the extent is not shared. Only file data blocks 496 * can have multiple owners. 497 */ 498 void 499 xfs_scrub_xref_is_not_shared( 500 struct xfs_scrub_context *sc, 501 xfs_agblock_t agbno, 502 xfs_extlen_t len) 503 { 504 bool shared; 505 int error; 506 507 if (!sc->sa.refc_cur) 508 return; 509 510 error = xfs_refcount_has_record(sc->sa.refc_cur, agbno, len, &shared); 511 if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur)) 512 return; 513 if (shared) 514 xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0); 515 } 516