1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2017 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_inode.h" 14 #include "xfs_trans.h" 15 #include "xfs_btree.h" 16 #include "xfs_rmap_btree.h" 17 #include "xfs_trace.h" 18 #include "xfs_rmap.h" 19 #include "xfs_alloc.h" 20 #include "xfs_bit.h" 21 #include <linux/fsmap.h> 22 #include "xfs_fsmap.h" 23 #include "xfs_refcount.h" 24 #include "xfs_refcount_btree.h" 25 #include "xfs_alloc_btree.h" 26 #include "xfs_rtbitmap.h" 27 #include "xfs_ag.h" 28 #include "xfs_rtgroup.h" 29 30 /* Convert an xfs_fsmap to an fsmap. */ 31 static void 32 xfs_fsmap_from_internal( 33 struct fsmap *dest, 34 struct xfs_fsmap *src) 35 { 36 dest->fmr_device = src->fmr_device; 37 dest->fmr_flags = src->fmr_flags; 38 dest->fmr_physical = BBTOB(src->fmr_physical); 39 dest->fmr_owner = src->fmr_owner; 40 dest->fmr_offset = BBTOB(src->fmr_offset); 41 dest->fmr_length = BBTOB(src->fmr_length); 42 dest->fmr_reserved[0] = 0; 43 dest->fmr_reserved[1] = 0; 44 dest->fmr_reserved[2] = 0; 45 } 46 47 /* Convert an fsmap to an xfs_fsmap. */ 48 static void 49 xfs_fsmap_to_internal( 50 struct xfs_fsmap *dest, 51 struct fsmap *src) 52 { 53 dest->fmr_device = src->fmr_device; 54 dest->fmr_flags = src->fmr_flags; 55 dest->fmr_physical = BTOBBT(src->fmr_physical); 56 dest->fmr_owner = src->fmr_owner; 57 dest->fmr_offset = BTOBBT(src->fmr_offset); 58 dest->fmr_length = BTOBBT(src->fmr_length); 59 } 60 61 /* Convert an fsmap owner into an rmapbt owner. */ 62 static int 63 xfs_fsmap_owner_to_rmap( 64 struct xfs_rmap_irec *dest, 65 const struct xfs_fsmap *src) 66 { 67 if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) { 68 dest->rm_owner = src->fmr_owner; 69 return 0; 70 } 71 72 switch (src->fmr_owner) { 73 case 0: /* "lowest owner id possible" */ 74 case -1ULL: /* "highest owner id possible" */ 75 dest->rm_owner = src->fmr_owner; 76 break; 77 case XFS_FMR_OWN_FREE: 78 dest->rm_owner = XFS_RMAP_OWN_NULL; 79 break; 80 case XFS_FMR_OWN_UNKNOWN: 81 dest->rm_owner = XFS_RMAP_OWN_UNKNOWN; 82 break; 83 case XFS_FMR_OWN_FS: 84 dest->rm_owner = XFS_RMAP_OWN_FS; 85 break; 86 case XFS_FMR_OWN_LOG: 87 dest->rm_owner = XFS_RMAP_OWN_LOG; 88 break; 89 case XFS_FMR_OWN_AG: 90 dest->rm_owner = XFS_RMAP_OWN_AG; 91 break; 92 case XFS_FMR_OWN_INOBT: 93 dest->rm_owner = XFS_RMAP_OWN_INOBT; 94 break; 95 case XFS_FMR_OWN_INODES: 96 dest->rm_owner = XFS_RMAP_OWN_INODES; 97 break; 98 case XFS_FMR_OWN_REFC: 99 dest->rm_owner = XFS_RMAP_OWN_REFC; 100 break; 101 case XFS_FMR_OWN_COW: 102 dest->rm_owner = XFS_RMAP_OWN_COW; 103 break; 104 case XFS_FMR_OWN_DEFECTIVE: /* not implemented */ 105 /* fall through */ 106 default: 107 return -EINVAL; 108 } 109 return 0; 110 } 111 112 /* Convert an rmapbt owner into an fsmap owner. */ 113 static int 114 xfs_fsmap_owner_from_frec( 115 struct xfs_fsmap *dest, 116 const struct xfs_fsmap_irec *frec) 117 { 118 dest->fmr_flags = 0; 119 if (!XFS_RMAP_NON_INODE_OWNER(frec->owner)) { 120 dest->fmr_owner = frec->owner; 121 return 0; 122 } 123 dest->fmr_flags |= FMR_OF_SPECIAL_OWNER; 124 125 switch (frec->owner) { 126 case XFS_RMAP_OWN_FS: 127 dest->fmr_owner = XFS_FMR_OWN_FS; 128 break; 129 case XFS_RMAP_OWN_LOG: 130 dest->fmr_owner = XFS_FMR_OWN_LOG; 131 break; 132 case XFS_RMAP_OWN_AG: 133 dest->fmr_owner = XFS_FMR_OWN_AG; 134 break; 135 case XFS_RMAP_OWN_INOBT: 136 dest->fmr_owner = XFS_FMR_OWN_INOBT; 137 break; 138 case XFS_RMAP_OWN_INODES: 139 dest->fmr_owner = XFS_FMR_OWN_INODES; 140 break; 141 case XFS_RMAP_OWN_REFC: 142 dest->fmr_owner = XFS_FMR_OWN_REFC; 143 break; 144 case XFS_RMAP_OWN_COW: 145 dest->fmr_owner = XFS_FMR_OWN_COW; 146 break; 147 case XFS_RMAP_OWN_NULL: /* "free" */ 148 dest->fmr_owner = XFS_FMR_OWN_FREE; 149 break; 150 default: 151 ASSERT(0); 152 return -EFSCORRUPTED; 153 } 154 return 0; 155 } 156 157 /* getfsmap query state */ 158 struct xfs_getfsmap_info { 159 struct xfs_fsmap_head *head; 160 struct fsmap *fsmap_recs; /* mapping records */ 161 struct xfs_buf *agf_bp; /* AGF, for refcount queries */ 162 struct xfs_group *group; /* group info, if applicable */ 163 xfs_daddr_t next_daddr; /* next daddr we expect */ 164 /* daddr of low fsmap key when we're using the rtbitmap */ 165 xfs_daddr_t low_daddr; 166 xfs_daddr_t end_daddr; /* daddr of high fsmap key */ 167 u64 missing_owner; /* owner of holes */ 168 u32 dev; /* device id */ 169 /* 170 * Low rmap key for the query. If low.rm_blockcount is nonzero, this 171 * is the second (or later) call to retrieve the recordset in pieces. 172 * xfs_getfsmap_rec_before_start will compare all records retrieved 173 * by the rmapbt query to filter out any records that start before 174 * the last record. 175 */ 176 struct xfs_rmap_irec low; 177 struct xfs_rmap_irec high; /* high rmap key */ 178 bool last; /* last extent? */ 179 }; 180 181 /* Associate a device with a getfsmap handler. */ 182 struct xfs_getfsmap_dev { 183 u32 dev; 184 int (*fn)(struct xfs_trans *tp, 185 const struct xfs_fsmap *keys, 186 struct xfs_getfsmap_info *info); 187 sector_t nr_sectors; 188 }; 189 190 /* Compare two getfsmap device handlers. */ 191 static int 192 xfs_getfsmap_dev_compare( 193 const void *p1, 194 const void *p2) 195 { 196 const struct xfs_getfsmap_dev *d1 = p1; 197 const struct xfs_getfsmap_dev *d2 = p2; 198 199 return d1->dev - d2->dev; 200 } 201 202 /* Decide if this mapping is shared. */ 203 STATIC int 204 xfs_getfsmap_is_shared( 205 struct xfs_trans *tp, 206 struct xfs_getfsmap_info *info, 207 const struct xfs_fsmap_irec *frec, 208 bool *stat) 209 { 210 struct xfs_mount *mp = tp->t_mountp; 211 struct xfs_btree_cur *cur; 212 xfs_agblock_t fbno; 213 xfs_extlen_t flen; 214 int error; 215 216 *stat = false; 217 if (!xfs_has_reflink(mp)) 218 return 0; 219 /* rt files will have no perag structure */ 220 if (!info->group) 221 return 0; 222 223 /* Are there any shared blocks here? */ 224 flen = 0; 225 cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp, 226 to_perag(info->group)); 227 228 error = xfs_refcount_find_shared(cur, frec->rec_key, 229 XFS_BB_TO_FSBT(mp, frec->len_daddr), &fbno, &flen, 230 false); 231 232 xfs_btree_del_cursor(cur, error); 233 if (error) 234 return error; 235 236 *stat = flen > 0; 237 return 0; 238 } 239 240 static inline void 241 xfs_getfsmap_format( 242 struct xfs_mount *mp, 243 struct xfs_fsmap *xfm, 244 struct xfs_getfsmap_info *info) 245 { 246 struct fsmap *rec; 247 248 trace_xfs_getfsmap_mapping(mp, xfm); 249 250 rec = &info->fsmap_recs[info->head->fmh_entries++]; 251 xfs_fsmap_from_internal(rec, xfm); 252 } 253 254 static inline bool 255 xfs_getfsmap_frec_before_start( 256 struct xfs_getfsmap_info *info, 257 const struct xfs_fsmap_irec *frec) 258 { 259 if (info->low_daddr != XFS_BUF_DADDR_NULL) 260 return frec->start_daddr < info->low_daddr; 261 if (info->low.rm_blockcount) { 262 struct xfs_rmap_irec rec = { 263 .rm_startblock = frec->rec_key, 264 .rm_owner = frec->owner, 265 .rm_flags = frec->rm_flags, 266 }; 267 268 return xfs_rmap_compare(&rec, &info->low) < 0; 269 } 270 271 return false; 272 } 273 274 /* 275 * Format a reverse mapping for getfsmap, having translated rm_startblock 276 * into the appropriate daddr units. Pass in a nonzero @len_daddr if the 277 * length could be larger than rm_blockcount in struct xfs_rmap_irec. 278 */ 279 STATIC int 280 xfs_getfsmap_helper( 281 struct xfs_trans *tp, 282 struct xfs_getfsmap_info *info, 283 const struct xfs_fsmap_irec *frec) 284 { 285 struct xfs_fsmap fmr; 286 struct xfs_mount *mp = tp->t_mountp; 287 bool shared; 288 int error = 0; 289 290 if (fatal_signal_pending(current)) 291 return -EINTR; 292 293 /* 294 * Filter out records that start before our startpoint, if the 295 * caller requested that. 296 */ 297 if (xfs_getfsmap_frec_before_start(info, frec)) 298 goto out; 299 300 /* Are we just counting mappings? */ 301 if (info->head->fmh_count == 0) { 302 if (info->head->fmh_entries == UINT_MAX) 303 return -ECANCELED; 304 305 if (frec->start_daddr > info->next_daddr) 306 info->head->fmh_entries++; 307 308 if (info->last) 309 return 0; 310 311 info->head->fmh_entries++; 312 goto out; 313 } 314 315 /* 316 * If the record starts past the last physical block we saw, 317 * then we've found a gap. Report the gap as being owned by 318 * whatever the caller specified is the missing owner. 319 */ 320 if (frec->start_daddr > info->next_daddr) { 321 if (info->head->fmh_entries >= info->head->fmh_count) 322 return -ECANCELED; 323 324 fmr.fmr_device = info->dev; 325 fmr.fmr_physical = info->next_daddr; 326 fmr.fmr_owner = info->missing_owner; 327 fmr.fmr_offset = 0; 328 fmr.fmr_length = frec->start_daddr - info->next_daddr; 329 fmr.fmr_flags = FMR_OF_SPECIAL_OWNER; 330 xfs_getfsmap_format(mp, &fmr, info); 331 } 332 333 if (info->last) 334 goto out; 335 336 /* Fill out the extent we found */ 337 if (info->head->fmh_entries >= info->head->fmh_count) 338 return -ECANCELED; 339 340 trace_xfs_fsmap_mapping(mp, info->dev, 341 info->group ? info->group->xg_gno : NULLAGNUMBER, 342 frec); 343 344 fmr.fmr_device = info->dev; 345 fmr.fmr_physical = frec->start_daddr; 346 error = xfs_fsmap_owner_from_frec(&fmr, frec); 347 if (error) 348 return error; 349 fmr.fmr_offset = XFS_FSB_TO_BB(mp, frec->offset); 350 fmr.fmr_length = frec->len_daddr; 351 if (frec->rm_flags & XFS_RMAP_UNWRITTEN) 352 fmr.fmr_flags |= FMR_OF_PREALLOC; 353 if (frec->rm_flags & XFS_RMAP_ATTR_FORK) 354 fmr.fmr_flags |= FMR_OF_ATTR_FORK; 355 if (frec->rm_flags & XFS_RMAP_BMBT_BLOCK) 356 fmr.fmr_flags |= FMR_OF_EXTENT_MAP; 357 if (fmr.fmr_flags == 0) { 358 error = xfs_getfsmap_is_shared(tp, info, frec, &shared); 359 if (error) 360 return error; 361 if (shared) 362 fmr.fmr_flags |= FMR_OF_SHARED; 363 } 364 365 xfs_getfsmap_format(mp, &fmr, info); 366 out: 367 info->next_daddr = max(info->next_daddr, 368 frec->start_daddr + frec->len_daddr); 369 return 0; 370 } 371 372 static inline int 373 xfs_getfsmap_group_helper( 374 struct xfs_getfsmap_info *info, 375 struct xfs_trans *tp, 376 struct xfs_group *xg, 377 xfs_agblock_t startblock, 378 xfs_extlen_t blockcount, 379 struct xfs_fsmap_irec *frec) 380 { 381 /* 382 * For an info->last query, we're looking for a gap between the last 383 * mapping emitted and the high key specified by userspace. If the 384 * user's query spans less than 1 fsblock, then info->high and 385 * info->low will have the same rm_startblock, which causes rec_daddr 386 * and next_daddr to be the same. Therefore, use the end_daddr that 387 * we calculated from userspace's high key to synthesize the record. 388 * Note that if the btree query found a mapping, there won't be a gap. 389 */ 390 if (info->last && info->end_daddr != XFS_BUF_DADDR_NULL) 391 frec->start_daddr = info->end_daddr; 392 else 393 frec->start_daddr = xfs_gbno_to_daddr(xg, startblock); 394 395 frec->len_daddr = XFS_FSB_TO_BB(xg->xg_mount, blockcount); 396 return xfs_getfsmap_helper(tp, info, frec); 397 } 398 399 /* Transform a rmapbt irec into a fsmap */ 400 STATIC int 401 xfs_getfsmap_rmapbt_helper( 402 struct xfs_btree_cur *cur, 403 const struct xfs_rmap_irec *rec, 404 void *priv) 405 { 406 struct xfs_fsmap_irec frec = { 407 .owner = rec->rm_owner, 408 .offset = rec->rm_offset, 409 .rm_flags = rec->rm_flags, 410 .rec_key = rec->rm_startblock, 411 }; 412 struct xfs_getfsmap_info *info = priv; 413 414 return xfs_getfsmap_group_helper(info, cur->bc_tp, cur->bc_group, 415 rec->rm_startblock, rec->rm_blockcount, &frec); 416 } 417 418 /* Transform a bnobt irec into a fsmap */ 419 STATIC int 420 xfs_getfsmap_datadev_bnobt_helper( 421 struct xfs_btree_cur *cur, 422 const struct xfs_alloc_rec_incore *rec, 423 void *priv) 424 { 425 struct xfs_fsmap_irec frec = { 426 .owner = XFS_RMAP_OWN_NULL, /* "free" */ 427 .rec_key = rec->ar_startblock, 428 }; 429 struct xfs_getfsmap_info *info = priv; 430 431 return xfs_getfsmap_group_helper(info, cur->bc_tp, cur->bc_group, 432 rec->ar_startblock, rec->ar_blockcount, &frec); 433 } 434 435 /* Set rmap flags based on the getfsmap flags */ 436 static void 437 xfs_getfsmap_set_irec_flags( 438 struct xfs_rmap_irec *irec, 439 const struct xfs_fsmap *fmr) 440 { 441 irec->rm_flags = 0; 442 if (fmr->fmr_flags & FMR_OF_ATTR_FORK) 443 irec->rm_flags |= XFS_RMAP_ATTR_FORK; 444 if (fmr->fmr_flags & FMR_OF_EXTENT_MAP) 445 irec->rm_flags |= XFS_RMAP_BMBT_BLOCK; 446 if (fmr->fmr_flags & FMR_OF_PREALLOC) 447 irec->rm_flags |= XFS_RMAP_UNWRITTEN; 448 } 449 450 static inline bool 451 rmap_not_shareable(struct xfs_mount *mp, const struct xfs_rmap_irec *r) 452 { 453 if (!xfs_has_reflink(mp)) 454 return true; 455 if (XFS_RMAP_NON_INODE_OWNER(r->rm_owner)) 456 return true; 457 if (r->rm_flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK | 458 XFS_RMAP_UNWRITTEN)) 459 return true; 460 return false; 461 } 462 463 /* Execute a getfsmap query against the regular data device. */ 464 STATIC int 465 __xfs_getfsmap_datadev( 466 struct xfs_trans *tp, 467 const struct xfs_fsmap *keys, 468 struct xfs_getfsmap_info *info, 469 int (*query_fn)(struct xfs_trans *, 470 struct xfs_getfsmap_info *, 471 struct xfs_btree_cur **, 472 void *), 473 void *priv) 474 { 475 struct xfs_mount *mp = tp->t_mountp; 476 struct xfs_perag *pag = NULL; 477 struct xfs_btree_cur *bt_cur = NULL; 478 xfs_fsblock_t start_fsb; 479 xfs_fsblock_t end_fsb; 480 xfs_agnumber_t start_ag, end_ag; 481 uint64_t eofs; 482 int error = 0; 483 484 eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 485 if (keys[0].fmr_physical >= eofs) 486 return 0; 487 start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical); 488 end_fsb = XFS_DADDR_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical)); 489 490 /* 491 * Convert the fsmap low/high keys to AG based keys. Initialize 492 * low to the fsmap low key and max out the high key to the end 493 * of the AG. 494 */ 495 info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset); 496 error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]); 497 if (error) 498 return error; 499 info->low.rm_blockcount = XFS_BB_TO_FSBT(mp, keys[0].fmr_length); 500 xfs_getfsmap_set_irec_flags(&info->low, &keys[0]); 501 502 /* Adjust the low key if we are continuing from where we left off. */ 503 if (info->low.rm_blockcount == 0) { 504 /* No previous record from which to continue */ 505 } else if (rmap_not_shareable(mp, &info->low)) { 506 /* Last record seen was an unshareable extent */ 507 info->low.rm_owner = 0; 508 info->low.rm_offset = 0; 509 510 start_fsb += info->low.rm_blockcount; 511 if (XFS_FSB_TO_DADDR(mp, start_fsb) >= eofs) 512 return 0; 513 } else { 514 /* Last record seen was a shareable file data extent */ 515 info->low.rm_offset += info->low.rm_blockcount; 516 } 517 info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb); 518 519 info->high.rm_startblock = -1U; 520 info->high.rm_owner = ULLONG_MAX; 521 info->high.rm_offset = ULLONG_MAX; 522 info->high.rm_blockcount = 0; 523 info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS; 524 525 start_ag = XFS_FSB_TO_AGNO(mp, start_fsb); 526 end_ag = XFS_FSB_TO_AGNO(mp, end_fsb); 527 528 while ((pag = xfs_perag_next_range(mp, pag, start_ag, end_ag))) { 529 /* 530 * Set the AG high key from the fsmap high key if this 531 * is the last AG that we're querying. 532 */ 533 info->group = pag_group(pag); 534 if (pag_agno(pag) == end_ag) { 535 info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp, 536 end_fsb); 537 info->high.rm_offset = XFS_BB_TO_FSBT(mp, 538 keys[1].fmr_offset); 539 error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]); 540 if (error) 541 break; 542 xfs_getfsmap_set_irec_flags(&info->high, &keys[1]); 543 } 544 545 if (bt_cur) { 546 xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR); 547 bt_cur = NULL; 548 xfs_trans_brelse(tp, info->agf_bp); 549 info->agf_bp = NULL; 550 } 551 552 error = xfs_alloc_read_agf(pag, tp, 0, &info->agf_bp); 553 if (error) 554 break; 555 556 trace_xfs_fsmap_low_group_key(mp, info->dev, pag_agno(pag), 557 &info->low); 558 trace_xfs_fsmap_high_group_key(mp, info->dev, pag_agno(pag), 559 &info->high); 560 561 error = query_fn(tp, info, &bt_cur, priv); 562 if (error) 563 break; 564 565 /* 566 * Set the AG low key to the start of the AG prior to 567 * moving on to the next AG. 568 */ 569 if (pag_agno(pag) == start_ag) 570 memset(&info->low, 0, sizeof(info->low)); 571 572 /* 573 * If this is the last AG, report any gap at the end of it 574 * before we drop the reference to the perag when the loop 575 * terminates. 576 */ 577 if (pag_agno(pag) == end_ag) { 578 info->last = true; 579 error = query_fn(tp, info, &bt_cur, priv); 580 if (error) 581 break; 582 } 583 info->group = NULL; 584 } 585 586 if (bt_cur) 587 xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR : 588 XFS_BTREE_NOERROR); 589 if (info->agf_bp) { 590 xfs_trans_brelse(tp, info->agf_bp); 591 info->agf_bp = NULL; 592 } 593 if (info->group) { 594 xfs_perag_rele(pag); 595 info->group = NULL; 596 } else if (pag) { 597 /* loop termination case */ 598 xfs_perag_rele(pag); 599 } 600 601 return error; 602 } 603 604 /* Actually query the rmap btree. */ 605 STATIC int 606 xfs_getfsmap_datadev_rmapbt_query( 607 struct xfs_trans *tp, 608 struct xfs_getfsmap_info *info, 609 struct xfs_btree_cur **curpp, 610 void *priv) 611 { 612 /* Report any gap at the end of the last AG. */ 613 if (info->last) 614 return xfs_getfsmap_rmapbt_helper(*curpp, &info->high, info); 615 616 /* Allocate cursor for this AG and query_range it. */ 617 *curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp, 618 to_perag(info->group)); 619 return xfs_rmap_query_range(*curpp, &info->low, &info->high, 620 xfs_getfsmap_rmapbt_helper, info); 621 } 622 623 /* Execute a getfsmap query against the regular data device rmapbt. */ 624 STATIC int 625 xfs_getfsmap_datadev_rmapbt( 626 struct xfs_trans *tp, 627 const struct xfs_fsmap *keys, 628 struct xfs_getfsmap_info *info) 629 { 630 info->missing_owner = XFS_FMR_OWN_FREE; 631 return __xfs_getfsmap_datadev(tp, keys, info, 632 xfs_getfsmap_datadev_rmapbt_query, NULL); 633 } 634 635 /* Actually query the bno btree. */ 636 STATIC int 637 xfs_getfsmap_datadev_bnobt_query( 638 struct xfs_trans *tp, 639 struct xfs_getfsmap_info *info, 640 struct xfs_btree_cur **curpp, 641 void *priv) 642 { 643 struct xfs_alloc_rec_incore *key = priv; 644 645 /* Report any gap at the end of the last AG. */ 646 if (info->last) 647 return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info); 648 649 /* Allocate cursor for this AG and query_range it. */ 650 *curpp = xfs_bnobt_init_cursor(tp->t_mountp, tp, info->agf_bp, 651 to_perag(info->group)); 652 key->ar_startblock = info->low.rm_startblock; 653 key[1].ar_startblock = info->high.rm_startblock; 654 return xfs_alloc_query_range(*curpp, key, &key[1], 655 xfs_getfsmap_datadev_bnobt_helper, info); 656 } 657 658 /* Execute a getfsmap query against the regular data device's bnobt. */ 659 STATIC int 660 xfs_getfsmap_datadev_bnobt( 661 struct xfs_trans *tp, 662 const struct xfs_fsmap *keys, 663 struct xfs_getfsmap_info *info) 664 { 665 struct xfs_alloc_rec_incore akeys[2]; 666 667 memset(akeys, 0, sizeof(akeys)); 668 info->missing_owner = XFS_FMR_OWN_UNKNOWN; 669 return __xfs_getfsmap_datadev(tp, keys, info, 670 xfs_getfsmap_datadev_bnobt_query, &akeys[0]); 671 } 672 673 /* Execute a getfsmap query against the log device. */ 674 STATIC int 675 xfs_getfsmap_logdev( 676 struct xfs_trans *tp, 677 const struct xfs_fsmap *keys, 678 struct xfs_getfsmap_info *info) 679 { 680 struct xfs_fsmap_irec frec = { 681 .start_daddr = 0, 682 .rec_key = 0, 683 .owner = XFS_RMAP_OWN_LOG, 684 }; 685 struct xfs_mount *mp = tp->t_mountp; 686 xfs_fsblock_t start_fsb, end_fsb; 687 uint64_t eofs; 688 689 eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 690 if (keys[0].fmr_physical >= eofs) 691 return 0; 692 start_fsb = XFS_BB_TO_FSBT(mp, 693 keys[0].fmr_physical + keys[0].fmr_length); 694 end_fsb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical)); 695 696 /* Adjust the low key if we are continuing from where we left off. */ 697 if (keys[0].fmr_length > 0) 698 info->low_daddr = XFS_FSB_TO_BB(mp, start_fsb); 699 700 trace_xfs_fsmap_low_linear_key(mp, info->dev, start_fsb); 701 trace_xfs_fsmap_high_linear_key(mp, info->dev, end_fsb); 702 703 if (start_fsb > 0) 704 return 0; 705 706 /* Fabricate an rmap entry for the external log device. */ 707 frec.len_daddr = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 708 return xfs_getfsmap_helper(tp, info, &frec); 709 } 710 711 #ifdef CONFIG_XFS_RT 712 /* Transform a rtbitmap "record" into a fsmap */ 713 STATIC int 714 xfs_getfsmap_rtdev_rtbitmap_helper( 715 struct xfs_rtgroup *rtg, 716 struct xfs_trans *tp, 717 const struct xfs_rtalloc_rec *rec, 718 void *priv) 719 { 720 struct xfs_fsmap_irec frec = { 721 .owner = XFS_RMAP_OWN_NULL, /* "free" */ 722 }; 723 struct xfs_mount *mp = rtg_mount(rtg); 724 struct xfs_getfsmap_info *info = priv; 725 xfs_rtblock_t start_rtb = 726 xfs_rtx_to_rtb(rtg, rec->ar_startext); 727 uint64_t rtbcount = 728 xfs_rtbxlen_to_blen(mp, rec->ar_extcount); 729 730 /* 731 * For an info->last query, we're looking for a gap between the last 732 * mapping emitted and the high key specified by userspace. If the 733 * user's query spans less than 1 fsblock, then info->high and 734 * info->low will have the same rm_startblock, which causes rec_daddr 735 * and next_daddr to be the same. Therefore, use the end_daddr that 736 * we calculated from userspace's high key to synthesize the record. 737 * Note that if the btree query found a mapping, there won't be a gap. 738 */ 739 if (info->last && info->end_daddr != XFS_BUF_DADDR_NULL) { 740 frec.start_daddr = info->end_daddr; 741 } else { 742 frec.start_daddr = xfs_rtb_to_daddr(mp, start_rtb); 743 } 744 745 frec.len_daddr = XFS_FSB_TO_BB(mp, rtbcount); 746 return xfs_getfsmap_helper(tp, info, &frec); 747 } 748 749 /* Execute a getfsmap query against the realtime device rtbitmap. */ 750 STATIC int 751 xfs_getfsmap_rtdev_rtbitmap( 752 struct xfs_trans *tp, 753 const struct xfs_fsmap *keys, 754 struct xfs_getfsmap_info *info) 755 { 756 struct xfs_mount *mp = tp->t_mountp; 757 xfs_rtblock_t start_rtbno, end_rtbno; 758 xfs_rtxnum_t start_rtx, end_rtx; 759 xfs_rgnumber_t start_rgno, end_rgno; 760 struct xfs_rtgroup *rtg = NULL; 761 uint64_t eofs; 762 int error; 763 764 eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks); 765 if (keys[0].fmr_physical >= eofs) 766 return 0; 767 768 info->missing_owner = XFS_FMR_OWN_UNKNOWN; 769 770 /* Adjust the low key if we are continuing from where we left off. */ 771 start_rtbno = xfs_daddr_to_rtb(mp, 772 keys[0].fmr_physical + keys[0].fmr_length); 773 if (keys[0].fmr_length > 0) { 774 info->low_daddr = xfs_rtb_to_daddr(mp, start_rtbno); 775 if (info->low_daddr >= eofs) 776 return 0; 777 } 778 start_rtx = xfs_rtb_to_rtx(mp, start_rtbno); 779 start_rgno = xfs_rtb_to_rgno(mp, start_rtbno); 780 781 end_rtbno = xfs_daddr_to_rtb(mp, min(eofs - 1, keys[1].fmr_physical)); 782 end_rgno = xfs_rtb_to_rgno(mp, end_rtbno); 783 784 trace_xfs_fsmap_low_linear_key(mp, info->dev, start_rtbno); 785 trace_xfs_fsmap_high_linear_key(mp, info->dev, end_rtbno); 786 787 end_rtx = -1ULL; 788 789 while ((rtg = xfs_rtgroup_next_range(mp, rtg, start_rgno, end_rgno))) { 790 if (rtg_rgno(rtg) == end_rgno) 791 end_rtx = xfs_rtb_to_rtx(mp, 792 end_rtbno + mp->m_sb.sb_rextsize - 1); 793 794 info->group = rtg_group(rtg); 795 xfs_rtgroup_lock(rtg, XFS_RTGLOCK_BITMAP_SHARED); 796 error = xfs_rtalloc_query_range(rtg, tp, start_rtx, end_rtx, 797 xfs_getfsmap_rtdev_rtbitmap_helper, info); 798 if (error) 799 break; 800 801 /* 802 * Report any gaps at the end of the rtbitmap by simulating a 803 * zero-length free extent starting at the rtx after the end 804 * of the query range. 805 */ 806 if (rtg_rgno(rtg) == end_rgno) { 807 struct xfs_rtalloc_rec ahigh = { 808 .ar_startext = min(end_rtx + 1, 809 rtg->rtg_extents), 810 }; 811 812 info->last = true; 813 error = xfs_getfsmap_rtdev_rtbitmap_helper(rtg, tp, 814 &ahigh, info); 815 if (error) 816 break; 817 } 818 819 xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_BITMAP_SHARED); 820 info->group = NULL; 821 start_rtx = 0; 822 } 823 824 /* loop termination case */ 825 if (rtg) { 826 if (info->group) { 827 xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_BITMAP_SHARED); 828 info->group = NULL; 829 } 830 xfs_rtgroup_rele(rtg); 831 } 832 833 return error; 834 } 835 #endif /* CONFIG_XFS_RT */ 836 837 /* Do we recognize the device? */ 838 STATIC bool 839 xfs_getfsmap_is_valid_device( 840 struct xfs_mount *mp, 841 struct xfs_fsmap *fm) 842 { 843 if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX || 844 fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev)) 845 return true; 846 if (mp->m_logdev_targp && 847 fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev)) 848 return true; 849 if (mp->m_rtdev_targp && 850 fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev)) 851 return true; 852 return false; 853 } 854 855 /* Ensure that the low key is less than the high key. */ 856 STATIC bool 857 xfs_getfsmap_check_keys( 858 struct xfs_fsmap *low_key, 859 struct xfs_fsmap *high_key) 860 { 861 if (low_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) { 862 if (low_key->fmr_offset) 863 return false; 864 } 865 if (high_key->fmr_flags != -1U && 866 (high_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | 867 FMR_OF_EXTENT_MAP))) { 868 if (high_key->fmr_offset && high_key->fmr_offset != -1ULL) 869 return false; 870 } 871 if (high_key->fmr_length && high_key->fmr_length != -1ULL) 872 return false; 873 874 if (low_key->fmr_device > high_key->fmr_device) 875 return false; 876 if (low_key->fmr_device < high_key->fmr_device) 877 return true; 878 879 if (low_key->fmr_physical > high_key->fmr_physical) 880 return false; 881 if (low_key->fmr_physical < high_key->fmr_physical) 882 return true; 883 884 if (low_key->fmr_owner > high_key->fmr_owner) 885 return false; 886 if (low_key->fmr_owner < high_key->fmr_owner) 887 return true; 888 889 if (low_key->fmr_offset > high_key->fmr_offset) 890 return false; 891 if (low_key->fmr_offset < high_key->fmr_offset) 892 return true; 893 894 return false; 895 } 896 897 /* 898 * There are only two devices if we didn't configure RT devices at build time. 899 */ 900 #ifdef CONFIG_XFS_RT 901 #define XFS_GETFSMAP_DEVS 3 902 #else 903 #define XFS_GETFSMAP_DEVS 2 904 #endif /* CONFIG_XFS_RT */ 905 906 /* 907 * Get filesystem's extents as described in head, and format for output. Fills 908 * in the supplied records array until there are no more reverse mappings to 909 * return or head.fmh_entries == head.fmh_count. In the second case, this 910 * function returns -ECANCELED to indicate that more records would have been 911 * returned. 912 * 913 * Key to Confusion 914 * ---------------- 915 * There are multiple levels of keys and counters at work here: 916 * xfs_fsmap_head.fmh_keys -- low and high fsmap keys passed in; 917 * these reflect fs-wide sector addrs. 918 * dkeys -- fmh_keys used to query each device; 919 * these are fmh_keys but w/ the low key 920 * bumped up by fmr_length. 921 * xfs_getfsmap_info.next_daddr -- next disk addr we expect to see; this 922 * is how we detect gaps in the fsmap 923 records and report them. 924 * xfs_getfsmap_info.low/high -- per-AG low/high keys computed from 925 * dkeys; used to query the metadata. 926 */ 927 STATIC int 928 xfs_getfsmap( 929 struct xfs_mount *mp, 930 struct xfs_fsmap_head *head, 931 struct fsmap *fsmap_recs) 932 { 933 struct xfs_trans *tp = NULL; 934 struct xfs_fsmap dkeys[2]; /* per-dev keys */ 935 struct xfs_getfsmap_dev handlers[XFS_GETFSMAP_DEVS]; 936 struct xfs_getfsmap_info info = { NULL }; 937 bool use_rmap; 938 int i; 939 int error = 0; 940 941 if (head->fmh_iflags & ~FMH_IF_VALID) 942 return -EINVAL; 943 if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) || 944 !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1])) 945 return -EINVAL; 946 if (!xfs_getfsmap_check_keys(&head->fmh_keys[0], &head->fmh_keys[1])) 947 return -EINVAL; 948 949 use_rmap = xfs_has_rmapbt(mp) && 950 has_capability_noaudit(current, CAP_SYS_ADMIN); 951 head->fmh_entries = 0; 952 953 /* Set up our device handlers. */ 954 memset(handlers, 0, sizeof(handlers)); 955 handlers[0].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 956 handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev); 957 if (use_rmap) 958 handlers[0].fn = xfs_getfsmap_datadev_rmapbt; 959 else 960 handlers[0].fn = xfs_getfsmap_datadev_bnobt; 961 if (mp->m_logdev_targp != mp->m_ddev_targp) { 962 handlers[1].nr_sectors = XFS_FSB_TO_BB(mp, 963 mp->m_sb.sb_logblocks); 964 handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev); 965 handlers[1].fn = xfs_getfsmap_logdev; 966 } 967 #ifdef CONFIG_XFS_RT 968 if (mp->m_rtdev_targp) { 969 handlers[2].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks); 970 handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev); 971 handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap; 972 } 973 #endif /* CONFIG_XFS_RT */ 974 975 xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev), 976 xfs_getfsmap_dev_compare); 977 978 /* 979 * To continue where we left off, we allow userspace to use the 980 * last mapping from a previous call as the low key of the next. 981 * This is identified by a non-zero length in the low key. We 982 * have to increment the low key in this scenario to ensure we 983 * don't return the same mapping again, and instead return the 984 * very next mapping. 985 * 986 * If the low key mapping refers to file data, the same physical 987 * blocks could be mapped to several other files/offsets. 988 * According to rmapbt record ordering, the minimal next 989 * possible record for the block range is the next starting 990 * offset in the same inode. Therefore, each fsmap backend bumps 991 * the file offset to continue the search appropriately. For 992 * all other low key mapping types (attr blocks, metadata), each 993 * fsmap backend bumps the physical offset as there can be no 994 * other mapping for the same physical block range. 995 */ 996 dkeys[0] = head->fmh_keys[0]; 997 memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap)); 998 999 info.next_daddr = head->fmh_keys[0].fmr_physical + 1000 head->fmh_keys[0].fmr_length; 1001 info.end_daddr = XFS_BUF_DADDR_NULL; 1002 info.fsmap_recs = fsmap_recs; 1003 info.head = head; 1004 1005 /* For each device we support... */ 1006 for (i = 0; i < XFS_GETFSMAP_DEVS; i++) { 1007 /* Is this device within the range the user asked for? */ 1008 if (!handlers[i].fn) 1009 continue; 1010 if (head->fmh_keys[0].fmr_device > handlers[i].dev) 1011 continue; 1012 if (head->fmh_keys[1].fmr_device < handlers[i].dev) 1013 break; 1014 1015 /* 1016 * If this device number matches the high key, we have 1017 * to pass the high key to the handler to limit the 1018 * query results. If the device number exceeds the 1019 * low key, zero out the low key so that we get 1020 * everything from the beginning. 1021 */ 1022 if (handlers[i].dev == head->fmh_keys[1].fmr_device) { 1023 dkeys[1] = head->fmh_keys[1]; 1024 info.end_daddr = min(handlers[i].nr_sectors - 1, 1025 dkeys[1].fmr_physical); 1026 } 1027 if (handlers[i].dev > head->fmh_keys[0].fmr_device) 1028 memset(&dkeys[0], 0, sizeof(struct xfs_fsmap)); 1029 1030 /* 1031 * Grab an empty transaction so that we can use its recursive 1032 * buffer locking abilities to detect cycles in the rmapbt 1033 * without deadlocking. 1034 */ 1035 error = xfs_trans_alloc_empty(mp, &tp); 1036 if (error) 1037 break; 1038 1039 info.dev = handlers[i].dev; 1040 info.last = false; 1041 info.group = NULL; 1042 info.low_daddr = XFS_BUF_DADDR_NULL; 1043 info.low.rm_blockcount = 0; 1044 error = handlers[i].fn(tp, dkeys, &info); 1045 if (error) 1046 break; 1047 xfs_trans_cancel(tp); 1048 tp = NULL; 1049 info.next_daddr = 0; 1050 } 1051 1052 if (tp) 1053 xfs_trans_cancel(tp); 1054 head->fmh_oflags = FMH_OF_DEV_T; 1055 return error; 1056 } 1057 1058 int 1059 xfs_ioc_getfsmap( 1060 struct xfs_inode *ip, 1061 struct fsmap_head __user *arg) 1062 { 1063 struct xfs_fsmap_head xhead = {0}; 1064 struct fsmap_head head; 1065 struct fsmap *recs; 1066 unsigned int count; 1067 __u32 last_flags = 0; 1068 bool done = false; 1069 int error; 1070 1071 if (copy_from_user(&head, arg, sizeof(struct fsmap_head))) 1072 return -EFAULT; 1073 if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) || 1074 memchr_inv(head.fmh_keys[0].fmr_reserved, 0, 1075 sizeof(head.fmh_keys[0].fmr_reserved)) || 1076 memchr_inv(head.fmh_keys[1].fmr_reserved, 0, 1077 sizeof(head.fmh_keys[1].fmr_reserved))) 1078 return -EINVAL; 1079 1080 /* 1081 * Use an internal memory buffer so that we don't have to copy fsmap 1082 * data to userspace while holding locks. Start by trying to allocate 1083 * up to 128k for the buffer, but fall back to a single page if needed. 1084 */ 1085 count = min_t(unsigned int, head.fmh_count, 1086 131072 / sizeof(struct fsmap)); 1087 recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL); 1088 if (!recs) { 1089 count = min_t(unsigned int, head.fmh_count, 1090 PAGE_SIZE / sizeof(struct fsmap)); 1091 recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL); 1092 if (!recs) 1093 return -ENOMEM; 1094 } 1095 1096 xhead.fmh_iflags = head.fmh_iflags; 1097 xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]); 1098 xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]); 1099 1100 trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]); 1101 trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]); 1102 1103 head.fmh_entries = 0; 1104 do { 1105 struct fsmap __user *user_recs; 1106 struct fsmap *last_rec; 1107 1108 user_recs = &arg->fmh_recs[head.fmh_entries]; 1109 xhead.fmh_entries = 0; 1110 xhead.fmh_count = min_t(unsigned int, count, 1111 head.fmh_count - head.fmh_entries); 1112 1113 /* Run query, record how many entries we got. */ 1114 error = xfs_getfsmap(ip->i_mount, &xhead, recs); 1115 switch (error) { 1116 case 0: 1117 /* 1118 * There are no more records in the result set. Copy 1119 * whatever we got to userspace and break out. 1120 */ 1121 done = true; 1122 break; 1123 case -ECANCELED: 1124 /* 1125 * The internal memory buffer is full. Copy whatever 1126 * records we got to userspace and go again if we have 1127 * not yet filled the userspace buffer. 1128 */ 1129 error = 0; 1130 break; 1131 default: 1132 goto out_free; 1133 } 1134 head.fmh_entries += xhead.fmh_entries; 1135 head.fmh_oflags = xhead.fmh_oflags; 1136 1137 /* 1138 * If the caller wanted a record count or there aren't any 1139 * new records to return, we're done. 1140 */ 1141 if (head.fmh_count == 0 || xhead.fmh_entries == 0) 1142 break; 1143 1144 /* Copy all the records we got out to userspace. */ 1145 if (copy_to_user(user_recs, recs, 1146 xhead.fmh_entries * sizeof(struct fsmap))) { 1147 error = -EFAULT; 1148 goto out_free; 1149 } 1150 1151 /* Remember the last record flags we copied to userspace. */ 1152 last_rec = &recs[xhead.fmh_entries - 1]; 1153 last_flags = last_rec->fmr_flags; 1154 1155 /* Set up the low key for the next iteration. */ 1156 xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec); 1157 trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]); 1158 } while (!done && head.fmh_entries < head.fmh_count); 1159 1160 /* 1161 * If there are no more records in the query result set and we're not 1162 * in counting mode, mark the last record returned with the LAST flag. 1163 */ 1164 if (done && head.fmh_count > 0 && head.fmh_entries > 0) { 1165 struct fsmap __user *user_rec; 1166 1167 last_flags |= FMR_OF_LAST; 1168 user_rec = &arg->fmh_recs[head.fmh_entries - 1]; 1169 1170 if (copy_to_user(&user_rec->fmr_flags, &last_flags, 1171 sizeof(last_flags))) { 1172 error = -EFAULT; 1173 goto out_free; 1174 } 1175 } 1176 1177 /* copy back header */ 1178 if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) { 1179 error = -EFAULT; 1180 goto out_free; 1181 } 1182 1183 out_free: 1184 kvfree(recs); 1185 return error; 1186 } 1187