1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * Copyright (c) 2018 Red Hat, Inc. 5 * All rights reserved. 6 */ 7 8 #include "xfs.h" 9 #include "xfs_fs.h" 10 #include "xfs_shared.h" 11 #include "xfs_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_bit.h" 14 #include "xfs_sb.h" 15 #include "xfs_mount.h" 16 #include "xfs_btree.h" 17 #include "xfs_alloc_btree.h" 18 #include "xfs_rmap_btree.h" 19 #include "xfs_alloc.h" 20 #include "xfs_ialloc.h" 21 #include "xfs_rmap.h" 22 #include "xfs_ag.h" 23 #include "xfs_ag_resv.h" 24 #include "xfs_health.h" 25 #include "xfs_error.h" 26 #include "xfs_bmap.h" 27 #include "xfs_defer.h" 28 #include "xfs_log_format.h" 29 #include "xfs_trans.h" 30 #include "xfs_trace.h" 31 #include "xfs_inode.h" 32 #include "xfs_icache.h" 33 34 35 /* 36 * Passive reference counting access wrappers to the perag structures. If the 37 * per-ag structure is to be freed, the freeing code is responsible for cleaning 38 * up objects with passive references before freeing the structure. This is 39 * things like cached buffers. 40 */ 41 struct xfs_perag * 42 xfs_perag_get( 43 struct xfs_mount *mp, 44 xfs_agnumber_t agno) 45 { 46 struct xfs_perag *pag; 47 48 rcu_read_lock(); 49 pag = radix_tree_lookup(&mp->m_perag_tree, agno); 50 if (pag) { 51 trace_xfs_perag_get(pag, _RET_IP_); 52 ASSERT(atomic_read(&pag->pag_ref) >= 0); 53 atomic_inc(&pag->pag_ref); 54 } 55 rcu_read_unlock(); 56 return pag; 57 } 58 59 /* 60 * search from @first to find the next perag with the given tag set. 61 */ 62 struct xfs_perag * 63 xfs_perag_get_tag( 64 struct xfs_mount *mp, 65 xfs_agnumber_t first, 66 unsigned int tag) 67 { 68 struct xfs_perag *pag; 69 int found; 70 71 rcu_read_lock(); 72 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, 73 (void **)&pag, first, 1, tag); 74 if (found <= 0) { 75 rcu_read_unlock(); 76 return NULL; 77 } 78 trace_xfs_perag_get_tag(pag, _RET_IP_); 79 atomic_inc(&pag->pag_ref); 80 rcu_read_unlock(); 81 return pag; 82 } 83 84 /* Get a passive reference to the given perag. */ 85 struct xfs_perag * 86 xfs_perag_hold( 87 struct xfs_perag *pag) 88 { 89 ASSERT(atomic_read(&pag->pag_ref) > 0 || 90 atomic_read(&pag->pag_active_ref) > 0); 91 92 trace_xfs_perag_hold(pag, _RET_IP_); 93 atomic_inc(&pag->pag_ref); 94 return pag; 95 } 96 97 void 98 xfs_perag_put( 99 struct xfs_perag *pag) 100 { 101 trace_xfs_perag_put(pag, _RET_IP_); 102 ASSERT(atomic_read(&pag->pag_ref) > 0); 103 atomic_dec(&pag->pag_ref); 104 } 105 106 /* 107 * Active references for perag structures. This is for short term access to the 108 * per ag structures for walking trees or accessing state. If an AG is being 109 * shrunk or is offline, then this will fail to find that AG and return NULL 110 * instead. 111 */ 112 struct xfs_perag * 113 xfs_perag_grab( 114 struct xfs_mount *mp, 115 xfs_agnumber_t agno) 116 { 117 struct xfs_perag *pag; 118 119 rcu_read_lock(); 120 pag = radix_tree_lookup(&mp->m_perag_tree, agno); 121 if (pag) { 122 trace_xfs_perag_grab(pag, _RET_IP_); 123 if (!atomic_inc_not_zero(&pag->pag_active_ref)) 124 pag = NULL; 125 } 126 rcu_read_unlock(); 127 return pag; 128 } 129 130 /* 131 * search from @first to find the next perag with the given tag set. 132 */ 133 struct xfs_perag * 134 xfs_perag_grab_tag( 135 struct xfs_mount *mp, 136 xfs_agnumber_t first, 137 int tag) 138 { 139 struct xfs_perag *pag; 140 int found; 141 142 rcu_read_lock(); 143 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, 144 (void **)&pag, first, 1, tag); 145 if (found <= 0) { 146 rcu_read_unlock(); 147 return NULL; 148 } 149 trace_xfs_perag_grab_tag(pag, _RET_IP_); 150 if (!atomic_inc_not_zero(&pag->pag_active_ref)) 151 pag = NULL; 152 rcu_read_unlock(); 153 return pag; 154 } 155 156 void 157 xfs_perag_rele( 158 struct xfs_perag *pag) 159 { 160 trace_xfs_perag_rele(pag, _RET_IP_); 161 if (atomic_dec_and_test(&pag->pag_active_ref)) 162 wake_up(&pag->pag_active_wq); 163 } 164 165 /* 166 * xfs_initialize_perag_data 167 * 168 * Read in each per-ag structure so we can count up the number of 169 * allocated inodes, free inodes and used filesystem blocks as this 170 * information is no longer persistent in the superblock. Once we have 171 * this information, write it into the in-core superblock structure. 172 */ 173 int 174 xfs_initialize_perag_data( 175 struct xfs_mount *mp, 176 xfs_agnumber_t agcount) 177 { 178 xfs_agnumber_t index; 179 struct xfs_perag *pag; 180 struct xfs_sb *sbp = &mp->m_sb; 181 uint64_t ifree = 0; 182 uint64_t ialloc = 0; 183 uint64_t bfree = 0; 184 uint64_t bfreelst = 0; 185 uint64_t btree = 0; 186 uint64_t fdblocks; 187 int error = 0; 188 189 for (index = 0; index < agcount; index++) { 190 /* 191 * Read the AGF and AGI buffers to populate the per-ag 192 * structures for us. 193 */ 194 pag = xfs_perag_get(mp, index); 195 error = xfs_alloc_read_agf(pag, NULL, 0, NULL); 196 if (!error) 197 error = xfs_ialloc_read_agi(pag, NULL, NULL); 198 if (error) { 199 xfs_perag_put(pag); 200 return error; 201 } 202 203 ifree += pag->pagi_freecount; 204 ialloc += pag->pagi_count; 205 bfree += pag->pagf_freeblks; 206 bfreelst += pag->pagf_flcount; 207 btree += pag->pagf_btreeblks; 208 xfs_perag_put(pag); 209 } 210 fdblocks = bfree + bfreelst + btree; 211 212 /* 213 * If the new summary counts are obviously incorrect, fail the 214 * mount operation because that implies the AGFs are also corrupt. 215 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which 216 * will prevent xfs_repair from fixing anything. 217 */ 218 if (fdblocks > sbp->sb_dblocks || ifree > ialloc) { 219 xfs_alert(mp, "AGF corruption. Please run xfs_repair."); 220 error = -EFSCORRUPTED; 221 goto out; 222 } 223 224 /* Overwrite incore superblock counters with just-read data */ 225 spin_lock(&mp->m_sb_lock); 226 sbp->sb_ifree = ifree; 227 sbp->sb_icount = ialloc; 228 sbp->sb_fdblocks = fdblocks; 229 spin_unlock(&mp->m_sb_lock); 230 231 xfs_reinit_percpu_counters(mp); 232 out: 233 xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS); 234 return error; 235 } 236 237 STATIC void 238 __xfs_free_perag( 239 struct rcu_head *head) 240 { 241 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); 242 243 ASSERT(!delayed_work_pending(&pag->pag_blockgc_work)); 244 kmem_free(pag); 245 } 246 247 /* 248 * Free up the per-ag resources associated with the mount structure. 249 */ 250 void 251 xfs_free_perag( 252 struct xfs_mount *mp) 253 { 254 struct xfs_perag *pag; 255 xfs_agnumber_t agno; 256 257 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { 258 spin_lock(&mp->m_perag_lock); 259 pag = radix_tree_delete(&mp->m_perag_tree, agno); 260 spin_unlock(&mp->m_perag_lock); 261 ASSERT(pag); 262 XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0); 263 xfs_defer_drain_free(&pag->pag_intents_drain); 264 265 cancel_delayed_work_sync(&pag->pag_blockgc_work); 266 xfs_buf_hash_destroy(pag); 267 268 /* drop the mount's active reference */ 269 xfs_perag_rele(pag); 270 XFS_IS_CORRUPT(pag->pag_mount, 271 atomic_read(&pag->pag_active_ref) != 0); 272 call_rcu(&pag->rcu_head, __xfs_free_perag); 273 } 274 } 275 276 /* Find the size of the AG, in blocks. */ 277 static xfs_agblock_t 278 __xfs_ag_block_count( 279 struct xfs_mount *mp, 280 xfs_agnumber_t agno, 281 xfs_agnumber_t agcount, 282 xfs_rfsblock_t dblocks) 283 { 284 ASSERT(agno < agcount); 285 286 if (agno < agcount - 1) 287 return mp->m_sb.sb_agblocks; 288 return dblocks - (agno * mp->m_sb.sb_agblocks); 289 } 290 291 xfs_agblock_t 292 xfs_ag_block_count( 293 struct xfs_mount *mp, 294 xfs_agnumber_t agno) 295 { 296 return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount, 297 mp->m_sb.sb_dblocks); 298 } 299 300 /* Calculate the first and last possible inode number in an AG. */ 301 static void 302 __xfs_agino_range( 303 struct xfs_mount *mp, 304 xfs_agblock_t eoag, 305 xfs_agino_t *first, 306 xfs_agino_t *last) 307 { 308 xfs_agblock_t bno; 309 310 /* 311 * Calculate the first inode, which will be in the first 312 * cluster-aligned block after the AGFL. 313 */ 314 bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align); 315 *first = XFS_AGB_TO_AGINO(mp, bno); 316 317 /* 318 * Calculate the last inode, which will be at the end of the 319 * last (aligned) cluster that can be allocated in the AG. 320 */ 321 bno = round_down(eoag, M_IGEO(mp)->cluster_align); 322 *last = XFS_AGB_TO_AGINO(mp, bno) - 1; 323 } 324 325 void 326 xfs_agino_range( 327 struct xfs_mount *mp, 328 xfs_agnumber_t agno, 329 xfs_agino_t *first, 330 xfs_agino_t *last) 331 { 332 return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last); 333 } 334 335 int 336 xfs_initialize_perag( 337 struct xfs_mount *mp, 338 xfs_agnumber_t agcount, 339 xfs_rfsblock_t dblocks, 340 xfs_agnumber_t *maxagi) 341 { 342 struct xfs_perag *pag; 343 xfs_agnumber_t index; 344 xfs_agnumber_t first_initialised = NULLAGNUMBER; 345 int error; 346 347 /* 348 * Walk the current per-ag tree so we don't try to initialise AGs 349 * that already exist (growfs case). Allocate and insert all the 350 * AGs we don't find ready for initialisation. 351 */ 352 for (index = 0; index < agcount; index++) { 353 pag = xfs_perag_get(mp, index); 354 if (pag) { 355 xfs_perag_put(pag); 356 continue; 357 } 358 359 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); 360 if (!pag) { 361 error = -ENOMEM; 362 goto out_unwind_new_pags; 363 } 364 pag->pag_agno = index; 365 pag->pag_mount = mp; 366 367 error = radix_tree_preload(GFP_NOFS); 368 if (error) 369 goto out_free_pag; 370 371 spin_lock(&mp->m_perag_lock); 372 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { 373 WARN_ON_ONCE(1); 374 spin_unlock(&mp->m_perag_lock); 375 radix_tree_preload_end(); 376 error = -EEXIST; 377 goto out_free_pag; 378 } 379 spin_unlock(&mp->m_perag_lock); 380 radix_tree_preload_end(); 381 382 #ifdef __KERNEL__ 383 /* Place kernel structure only init below this point. */ 384 spin_lock_init(&pag->pag_ici_lock); 385 spin_lock_init(&pag->pagb_lock); 386 spin_lock_init(&pag->pag_state_lock); 387 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker); 388 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); 389 xfs_defer_drain_init(&pag->pag_intents_drain); 390 init_waitqueue_head(&pag->pagb_wait); 391 init_waitqueue_head(&pag->pag_active_wq); 392 pag->pagb_count = 0; 393 pag->pagb_tree = RB_ROOT; 394 #endif /* __KERNEL__ */ 395 396 error = xfs_buf_hash_init(pag); 397 if (error) 398 goto out_remove_pag; 399 400 /* Active ref owned by mount indicates AG is online. */ 401 atomic_set(&pag->pag_active_ref, 1); 402 403 /* first new pag is fully initialized */ 404 if (first_initialised == NULLAGNUMBER) 405 first_initialised = index; 406 407 /* 408 * Pre-calculated geometry 409 */ 410 pag->block_count = __xfs_ag_block_count(mp, index, agcount, 411 dblocks); 412 pag->min_block = XFS_AGFL_BLOCK(mp); 413 __xfs_agino_range(mp, pag->block_count, &pag->agino_min, 414 &pag->agino_max); 415 } 416 417 index = xfs_set_inode_alloc(mp, agcount); 418 419 if (maxagi) 420 *maxagi = index; 421 422 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp); 423 return 0; 424 425 out_remove_pag: 426 xfs_defer_drain_free(&pag->pag_intents_drain); 427 radix_tree_delete(&mp->m_perag_tree, index); 428 out_free_pag: 429 kmem_free(pag); 430 out_unwind_new_pags: 431 /* unwind any prior newly initialized pags */ 432 for (index = first_initialised; index < agcount; index++) { 433 pag = radix_tree_delete(&mp->m_perag_tree, index); 434 if (!pag) 435 break; 436 xfs_buf_hash_destroy(pag); 437 xfs_defer_drain_free(&pag->pag_intents_drain); 438 kmem_free(pag); 439 } 440 return error; 441 } 442 443 static int 444 xfs_get_aghdr_buf( 445 struct xfs_mount *mp, 446 xfs_daddr_t blkno, 447 size_t numblks, 448 struct xfs_buf **bpp, 449 const struct xfs_buf_ops *ops) 450 { 451 struct xfs_buf *bp; 452 int error; 453 454 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp); 455 if (error) 456 return error; 457 458 bp->b_maps[0].bm_bn = blkno; 459 bp->b_ops = ops; 460 461 *bpp = bp; 462 return 0; 463 } 464 465 /* 466 * Generic btree root block init function 467 */ 468 static void 469 xfs_btroot_init( 470 struct xfs_mount *mp, 471 struct xfs_buf *bp, 472 struct aghdr_init_data *id) 473 { 474 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno); 475 } 476 477 /* Finish initializing a free space btree. */ 478 static void 479 xfs_freesp_init_recs( 480 struct xfs_mount *mp, 481 struct xfs_buf *bp, 482 struct aghdr_init_data *id) 483 { 484 struct xfs_alloc_rec *arec; 485 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); 486 487 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1); 488 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks); 489 490 if (xfs_ag_contains_log(mp, id->agno)) { 491 struct xfs_alloc_rec *nrec; 492 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp, 493 mp->m_sb.sb_logstart); 494 495 ASSERT(start >= mp->m_ag_prealloc_blocks); 496 if (start != mp->m_ag_prealloc_blocks) { 497 /* 498 * Modify first record to pad stripe align of log 499 */ 500 arec->ar_blockcount = cpu_to_be32(start - 501 mp->m_ag_prealloc_blocks); 502 nrec = arec + 1; 503 504 /* 505 * Insert second record at start of internal log 506 * which then gets trimmed. 507 */ 508 nrec->ar_startblock = cpu_to_be32( 509 be32_to_cpu(arec->ar_startblock) + 510 be32_to_cpu(arec->ar_blockcount)); 511 arec = nrec; 512 be16_add_cpu(&block->bb_numrecs, 1); 513 } 514 /* 515 * Change record start to after the internal log 516 */ 517 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks); 518 } 519 520 /* 521 * Calculate the record block count and check for the case where 522 * the log might have consumed all available space in the AG. If 523 * so, reset the record count to 0 to avoid exposure of an invalid 524 * record start block. 525 */ 526 arec->ar_blockcount = cpu_to_be32(id->agsize - 527 be32_to_cpu(arec->ar_startblock)); 528 if (!arec->ar_blockcount) 529 block->bb_numrecs = 0; 530 } 531 532 /* 533 * Alloc btree root block init functions 534 */ 535 static void 536 xfs_bnoroot_init( 537 struct xfs_mount *mp, 538 struct xfs_buf *bp, 539 struct aghdr_init_data *id) 540 { 541 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno); 542 xfs_freesp_init_recs(mp, bp, id); 543 } 544 545 static void 546 xfs_cntroot_init( 547 struct xfs_mount *mp, 548 struct xfs_buf *bp, 549 struct aghdr_init_data *id) 550 { 551 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno); 552 xfs_freesp_init_recs(mp, bp, id); 553 } 554 555 /* 556 * Reverse map root block init 557 */ 558 static void 559 xfs_rmaproot_init( 560 struct xfs_mount *mp, 561 struct xfs_buf *bp, 562 struct aghdr_init_data *id) 563 { 564 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); 565 struct xfs_rmap_rec *rrec; 566 567 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno); 568 569 /* 570 * mark the AG header regions as static metadata The BNO 571 * btree block is the first block after the headers, so 572 * it's location defines the size of region the static 573 * metadata consumes. 574 * 575 * Note: unlike mkfs, we never have to account for log 576 * space when growing the data regions 577 */ 578 rrec = XFS_RMAP_REC_ADDR(block, 1); 579 rrec->rm_startblock = 0; 580 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp)); 581 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS); 582 rrec->rm_offset = 0; 583 584 /* account freespace btree root blocks */ 585 rrec = XFS_RMAP_REC_ADDR(block, 2); 586 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp)); 587 rrec->rm_blockcount = cpu_to_be32(2); 588 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG); 589 rrec->rm_offset = 0; 590 591 /* account inode btree root blocks */ 592 rrec = XFS_RMAP_REC_ADDR(block, 3); 593 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp)); 594 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) - 595 XFS_IBT_BLOCK(mp)); 596 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT); 597 rrec->rm_offset = 0; 598 599 /* account for rmap btree root */ 600 rrec = XFS_RMAP_REC_ADDR(block, 4); 601 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp)); 602 rrec->rm_blockcount = cpu_to_be32(1); 603 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG); 604 rrec->rm_offset = 0; 605 606 /* account for refc btree root */ 607 if (xfs_has_reflink(mp)) { 608 rrec = XFS_RMAP_REC_ADDR(block, 5); 609 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp)); 610 rrec->rm_blockcount = cpu_to_be32(1); 611 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC); 612 rrec->rm_offset = 0; 613 be16_add_cpu(&block->bb_numrecs, 1); 614 } 615 616 /* account for the log space */ 617 if (xfs_ag_contains_log(mp, id->agno)) { 618 rrec = XFS_RMAP_REC_ADDR(block, 619 be16_to_cpu(block->bb_numrecs) + 1); 620 rrec->rm_startblock = cpu_to_be32( 621 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart)); 622 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks); 623 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG); 624 rrec->rm_offset = 0; 625 be16_add_cpu(&block->bb_numrecs, 1); 626 } 627 } 628 629 /* 630 * Initialise new secondary superblocks with the pre-grow geometry, but mark 631 * them as "in progress" so we know they haven't yet been activated. This will 632 * get cleared when the update with the new geometry information is done after 633 * changes to the primary are committed. This isn't strictly necessary, but we 634 * get it for free with the delayed buffer write lists and it means we can tell 635 * if a grow operation didn't complete properly after the fact. 636 */ 637 static void 638 xfs_sbblock_init( 639 struct xfs_mount *mp, 640 struct xfs_buf *bp, 641 struct aghdr_init_data *id) 642 { 643 struct xfs_dsb *dsb = bp->b_addr; 644 645 xfs_sb_to_disk(dsb, &mp->m_sb); 646 dsb->sb_inprogress = 1; 647 } 648 649 static void 650 xfs_agfblock_init( 651 struct xfs_mount *mp, 652 struct xfs_buf *bp, 653 struct aghdr_init_data *id) 654 { 655 struct xfs_agf *agf = bp->b_addr; 656 xfs_extlen_t tmpsize; 657 658 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC); 659 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION); 660 agf->agf_seqno = cpu_to_be32(id->agno); 661 agf->agf_length = cpu_to_be32(id->agsize); 662 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp)); 663 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp)); 664 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1); 665 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1); 666 if (xfs_has_rmapbt(mp)) { 667 agf->agf_roots[XFS_BTNUM_RMAPi] = 668 cpu_to_be32(XFS_RMAP_BLOCK(mp)); 669 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1); 670 agf->agf_rmap_blocks = cpu_to_be32(1); 671 } 672 673 agf->agf_flfirst = cpu_to_be32(1); 674 agf->agf_fllast = 0; 675 agf->agf_flcount = 0; 676 tmpsize = id->agsize - mp->m_ag_prealloc_blocks; 677 agf->agf_freeblks = cpu_to_be32(tmpsize); 678 agf->agf_longest = cpu_to_be32(tmpsize); 679 if (xfs_has_crc(mp)) 680 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid); 681 if (xfs_has_reflink(mp)) { 682 agf->agf_refcount_root = cpu_to_be32( 683 xfs_refc_block(mp)); 684 agf->agf_refcount_level = cpu_to_be32(1); 685 agf->agf_refcount_blocks = cpu_to_be32(1); 686 } 687 688 if (xfs_ag_contains_log(mp, id->agno)) { 689 int64_t logblocks = mp->m_sb.sb_logblocks; 690 691 be32_add_cpu(&agf->agf_freeblks, -logblocks); 692 agf->agf_longest = cpu_to_be32(id->agsize - 693 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks); 694 } 695 } 696 697 static void 698 xfs_agflblock_init( 699 struct xfs_mount *mp, 700 struct xfs_buf *bp, 701 struct aghdr_init_data *id) 702 { 703 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp); 704 __be32 *agfl_bno; 705 int bucket; 706 707 if (xfs_has_crc(mp)) { 708 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC); 709 agfl->agfl_seqno = cpu_to_be32(id->agno); 710 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid); 711 } 712 713 agfl_bno = xfs_buf_to_agfl_bno(bp); 714 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++) 715 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK); 716 } 717 718 static void 719 xfs_agiblock_init( 720 struct xfs_mount *mp, 721 struct xfs_buf *bp, 722 struct aghdr_init_data *id) 723 { 724 struct xfs_agi *agi = bp->b_addr; 725 int bucket; 726 727 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC); 728 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION); 729 agi->agi_seqno = cpu_to_be32(id->agno); 730 agi->agi_length = cpu_to_be32(id->agsize); 731 agi->agi_count = 0; 732 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp)); 733 agi->agi_level = cpu_to_be32(1); 734 agi->agi_freecount = 0; 735 agi->agi_newino = cpu_to_be32(NULLAGINO); 736 agi->agi_dirino = cpu_to_be32(NULLAGINO); 737 if (xfs_has_crc(mp)) 738 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid); 739 if (xfs_has_finobt(mp)) { 740 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp)); 741 agi->agi_free_level = cpu_to_be32(1); 742 } 743 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) 744 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); 745 if (xfs_has_inobtcounts(mp)) { 746 agi->agi_iblocks = cpu_to_be32(1); 747 if (xfs_has_finobt(mp)) 748 agi->agi_fblocks = cpu_to_be32(1); 749 } 750 } 751 752 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp, 753 struct aghdr_init_data *id); 754 static int 755 xfs_ag_init_hdr( 756 struct xfs_mount *mp, 757 struct aghdr_init_data *id, 758 aghdr_init_work_f work, 759 const struct xfs_buf_ops *ops) 760 { 761 struct xfs_buf *bp; 762 int error; 763 764 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops); 765 if (error) 766 return error; 767 768 (*work)(mp, bp, id); 769 770 xfs_buf_delwri_queue(bp, &id->buffer_list); 771 xfs_buf_relse(bp); 772 return 0; 773 } 774 775 struct xfs_aghdr_grow_data { 776 xfs_daddr_t daddr; 777 size_t numblks; 778 const struct xfs_buf_ops *ops; 779 aghdr_init_work_f work; 780 xfs_btnum_t type; 781 bool need_init; 782 }; 783 784 /* 785 * Prepare new AG headers to be written to disk. We use uncached buffers here, 786 * as it is assumed these new AG headers are currently beyond the currently 787 * valid filesystem address space. Using cached buffers would trip over EOFS 788 * corruption detection alogrithms in the buffer cache lookup routines. 789 * 790 * This is a non-transactional function, but the prepared buffers are added to a 791 * delayed write buffer list supplied by the caller so they can submit them to 792 * disk and wait on them as required. 793 */ 794 int 795 xfs_ag_init_headers( 796 struct xfs_mount *mp, 797 struct aghdr_init_data *id) 798 799 { 800 struct xfs_aghdr_grow_data aghdr_data[] = { 801 { /* SB */ 802 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR), 803 .numblks = XFS_FSS_TO_BB(mp, 1), 804 .ops = &xfs_sb_buf_ops, 805 .work = &xfs_sbblock_init, 806 .need_init = true 807 }, 808 { /* AGF */ 809 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)), 810 .numblks = XFS_FSS_TO_BB(mp, 1), 811 .ops = &xfs_agf_buf_ops, 812 .work = &xfs_agfblock_init, 813 .need_init = true 814 }, 815 { /* AGFL */ 816 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)), 817 .numblks = XFS_FSS_TO_BB(mp, 1), 818 .ops = &xfs_agfl_buf_ops, 819 .work = &xfs_agflblock_init, 820 .need_init = true 821 }, 822 { /* AGI */ 823 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)), 824 .numblks = XFS_FSS_TO_BB(mp, 1), 825 .ops = &xfs_agi_buf_ops, 826 .work = &xfs_agiblock_init, 827 .need_init = true 828 }, 829 { /* BNO root block */ 830 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)), 831 .numblks = BTOBB(mp->m_sb.sb_blocksize), 832 .ops = &xfs_bnobt_buf_ops, 833 .work = &xfs_bnoroot_init, 834 .need_init = true 835 }, 836 { /* CNT root block */ 837 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)), 838 .numblks = BTOBB(mp->m_sb.sb_blocksize), 839 .ops = &xfs_cntbt_buf_ops, 840 .work = &xfs_cntroot_init, 841 .need_init = true 842 }, 843 { /* INO root block */ 844 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)), 845 .numblks = BTOBB(mp->m_sb.sb_blocksize), 846 .ops = &xfs_inobt_buf_ops, 847 .work = &xfs_btroot_init, 848 .type = XFS_BTNUM_INO, 849 .need_init = true 850 }, 851 { /* FINO root block */ 852 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)), 853 .numblks = BTOBB(mp->m_sb.sb_blocksize), 854 .ops = &xfs_finobt_buf_ops, 855 .work = &xfs_btroot_init, 856 .type = XFS_BTNUM_FINO, 857 .need_init = xfs_has_finobt(mp) 858 }, 859 { /* RMAP root block */ 860 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)), 861 .numblks = BTOBB(mp->m_sb.sb_blocksize), 862 .ops = &xfs_rmapbt_buf_ops, 863 .work = &xfs_rmaproot_init, 864 .need_init = xfs_has_rmapbt(mp) 865 }, 866 { /* REFC root block */ 867 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)), 868 .numblks = BTOBB(mp->m_sb.sb_blocksize), 869 .ops = &xfs_refcountbt_buf_ops, 870 .work = &xfs_btroot_init, 871 .type = XFS_BTNUM_REFC, 872 .need_init = xfs_has_reflink(mp) 873 }, 874 { /* NULL terminating block */ 875 .daddr = XFS_BUF_DADDR_NULL, 876 } 877 }; 878 struct xfs_aghdr_grow_data *dp; 879 int error = 0; 880 881 /* Account for AG free space in new AG */ 882 id->nfree += id->agsize - mp->m_ag_prealloc_blocks; 883 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) { 884 if (!dp->need_init) 885 continue; 886 887 id->daddr = dp->daddr; 888 id->numblks = dp->numblks; 889 id->type = dp->type; 890 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops); 891 if (error) 892 break; 893 } 894 return error; 895 } 896 897 int 898 xfs_ag_shrink_space( 899 struct xfs_perag *pag, 900 struct xfs_trans **tpp, 901 xfs_extlen_t delta) 902 { 903 struct xfs_mount *mp = pag->pag_mount; 904 struct xfs_alloc_arg args = { 905 .tp = *tpp, 906 .mp = mp, 907 .pag = pag, 908 .minlen = delta, 909 .maxlen = delta, 910 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE, 911 .resv = XFS_AG_RESV_NONE, 912 .prod = 1 913 }; 914 struct xfs_buf *agibp, *agfbp; 915 struct xfs_agi *agi; 916 struct xfs_agf *agf; 917 xfs_agblock_t aglen; 918 int error, err2; 919 920 ASSERT(pag->pag_agno == mp->m_sb.sb_agcount - 1); 921 error = xfs_ialloc_read_agi(pag, *tpp, &agibp); 922 if (error) 923 return error; 924 925 agi = agibp->b_addr; 926 927 error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp); 928 if (error) 929 return error; 930 931 agf = agfbp->b_addr; 932 aglen = be32_to_cpu(agi->agi_length); 933 /* some extra paranoid checks before we shrink the ag */ 934 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length)) 935 return -EFSCORRUPTED; 936 if (delta >= aglen) 937 return -EINVAL; 938 939 /* 940 * Make sure that the last inode cluster cannot overlap with the new 941 * end of the AG, even if it's sparse. 942 */ 943 error = xfs_ialloc_check_shrink(pag, *tpp, agibp, aglen - delta); 944 if (error) 945 return error; 946 947 /* 948 * Disable perag reservations so it doesn't cause the allocation request 949 * to fail. We'll reestablish reservation before we return. 950 */ 951 error = xfs_ag_resv_free(pag); 952 if (error) 953 return error; 954 955 /* internal log shouldn't also show up in the free space btrees */ 956 error = xfs_alloc_vextent_exact_bno(&args, 957 XFS_AGB_TO_FSB(mp, pag->pag_agno, aglen - delta)); 958 if (!error && args.agbno == NULLAGBLOCK) 959 error = -ENOSPC; 960 961 if (error) { 962 /* 963 * if extent allocation fails, need to roll the transaction to 964 * ensure that the AGFL fixup has been committed anyway. 965 */ 966 xfs_trans_bhold(*tpp, agfbp); 967 err2 = xfs_trans_roll(tpp); 968 if (err2) 969 return err2; 970 xfs_trans_bjoin(*tpp, agfbp); 971 goto resv_init_out; 972 } 973 974 /* 975 * if successfully deleted from freespace btrees, need to confirm 976 * per-AG reservation works as expected. 977 */ 978 be32_add_cpu(&agi->agi_length, -delta); 979 be32_add_cpu(&agf->agf_length, -delta); 980 981 err2 = xfs_ag_resv_init(pag, *tpp); 982 if (err2) { 983 be32_add_cpu(&agi->agi_length, delta); 984 be32_add_cpu(&agf->agf_length, delta); 985 if (err2 != -ENOSPC) 986 goto resv_err; 987 988 __xfs_free_extent_later(*tpp, args.fsbno, delta, NULL, true); 989 990 /* 991 * Roll the transaction before trying to re-init the per-ag 992 * reservation. The new transaction is clean so it will cancel 993 * without any side effects. 994 */ 995 error = xfs_defer_finish(tpp); 996 if (error) 997 return error; 998 999 error = -ENOSPC; 1000 goto resv_init_out; 1001 } 1002 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH); 1003 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH); 1004 return 0; 1005 1006 resv_init_out: 1007 err2 = xfs_ag_resv_init(pag, *tpp); 1008 if (!err2) 1009 return error; 1010 resv_err: 1011 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2); 1012 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 1013 return err2; 1014 } 1015 1016 /* 1017 * Extent the AG indicated by the @id by the length passed in 1018 */ 1019 int 1020 xfs_ag_extend_space( 1021 struct xfs_perag *pag, 1022 struct xfs_trans *tp, 1023 xfs_extlen_t len) 1024 { 1025 struct xfs_buf *bp; 1026 struct xfs_agi *agi; 1027 struct xfs_agf *agf; 1028 int error; 1029 1030 ASSERT(pag->pag_agno == pag->pag_mount->m_sb.sb_agcount - 1); 1031 1032 error = xfs_ialloc_read_agi(pag, tp, &bp); 1033 if (error) 1034 return error; 1035 1036 agi = bp->b_addr; 1037 be32_add_cpu(&agi->agi_length, len); 1038 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH); 1039 1040 /* 1041 * Change agf length. 1042 */ 1043 error = xfs_alloc_read_agf(pag, tp, 0, &bp); 1044 if (error) 1045 return error; 1046 1047 agf = bp->b_addr; 1048 be32_add_cpu(&agf->agf_length, len); 1049 ASSERT(agf->agf_length == agi->agi_length); 1050 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH); 1051 1052 /* 1053 * Free the new space. 1054 * 1055 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that 1056 * this doesn't actually exist in the rmap btree. 1057 */ 1058 error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len, 1059 len, &XFS_RMAP_OINFO_SKIP_UPDATE); 1060 if (error) 1061 return error; 1062 1063 error = xfs_free_extent(tp, pag, be32_to_cpu(agf->agf_length) - len, 1064 len, &XFS_RMAP_OINFO_SKIP_UPDATE, XFS_AG_RESV_NONE); 1065 if (error) 1066 return error; 1067 1068 /* Update perag geometry */ 1069 pag->block_count = be32_to_cpu(agf->agf_length); 1070 __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min, 1071 &pag->agino_max); 1072 return 0; 1073 } 1074 1075 /* Retrieve AG geometry. */ 1076 int 1077 xfs_ag_get_geometry( 1078 struct xfs_perag *pag, 1079 struct xfs_ag_geometry *ageo) 1080 { 1081 struct xfs_buf *agi_bp; 1082 struct xfs_buf *agf_bp; 1083 struct xfs_agi *agi; 1084 struct xfs_agf *agf; 1085 unsigned int freeblks; 1086 int error; 1087 1088 /* Lock the AG headers. */ 1089 error = xfs_ialloc_read_agi(pag, NULL, &agi_bp); 1090 if (error) 1091 return error; 1092 error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp); 1093 if (error) 1094 goto out_agi; 1095 1096 /* Fill out form. */ 1097 memset(ageo, 0, sizeof(*ageo)); 1098 ageo->ag_number = pag->pag_agno; 1099 1100 agi = agi_bp->b_addr; 1101 ageo->ag_icount = be32_to_cpu(agi->agi_count); 1102 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount); 1103 1104 agf = agf_bp->b_addr; 1105 ageo->ag_length = be32_to_cpu(agf->agf_length); 1106 freeblks = pag->pagf_freeblks + 1107 pag->pagf_flcount + 1108 pag->pagf_btreeblks - 1109 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE); 1110 ageo->ag_freeblks = freeblks; 1111 xfs_ag_geom_health(pag, ageo); 1112 1113 /* Release resources. */ 1114 xfs_buf_relse(agf_bp); 1115 out_agi: 1116 xfs_buf_relse(agi_bp); 1117 return error; 1118 } 1119