1/* 2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18#include "xfs.h" 19#include "xfs_fs.h" 20#include "xfs_shared.h" 21#include "xfs_format.h" 22#include "xfs_log_format.h" 23#include "xfs_trans_resv.h" 24#include "xfs_bit.h" 25#include "xfs_sb.h" 26#include "xfs_mount.h" 27#include "xfs_da_format.h" 28#include "xfs_da_btree.h" 29#include "xfs_inode.h" 30#include "xfs_dir2.h" 31#include "xfs_ialloc.h" 32#include "xfs_alloc.h" 33#include "xfs_rtalloc.h" 34#include "xfs_bmap.h" 35#include "xfs_trans.h" 36#include "xfs_trans_priv.h" 37#include "xfs_log.h" 38#include "xfs_error.h" 39#include "xfs_quota.h" 40#include "xfs_fsops.h" 41#include "xfs_trace.h" 42#include "xfs_icache.h" 43#include "xfs_sysfs.h" 44 45 46static DEFINE_MUTEX(xfs_uuid_table_mutex); 47static int xfs_uuid_table_size; 48static uuid_t *xfs_uuid_table; 49 50/* 51 * See if the UUID is unique among mounted XFS filesystems. 52 * Mount fails if UUID is nil or a FS with the same UUID is already mounted. 53 */ 54STATIC int 55xfs_uuid_mount( 56 struct xfs_mount *mp) 57{ 58 uuid_t *uuid = &mp->m_sb.sb_uuid; 59 int hole, i; 60 61 if (mp->m_flags & XFS_MOUNT_NOUUID) 62 return 0; 63 64 if (uuid_is_nil(uuid)) { 65 xfs_warn(mp, "Filesystem has nil UUID - can't mount"); 66 return -EINVAL; 67 } 68 69 mutex_lock(&xfs_uuid_table_mutex); 70 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { 71 if (uuid_is_nil(&xfs_uuid_table[i])) { 72 hole = i; 73 continue; 74 } 75 if (uuid_equal(uuid, &xfs_uuid_table[i])) 76 goto out_duplicate; 77 } 78 79 if (hole < 0) { 80 xfs_uuid_table = kmem_realloc(xfs_uuid_table, 81 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), 82 xfs_uuid_table_size * sizeof(*xfs_uuid_table), 83 KM_SLEEP); 84 hole = xfs_uuid_table_size++; 85 } 86 xfs_uuid_table[hole] = *uuid; 87 mutex_unlock(&xfs_uuid_table_mutex); 88 89 return 0; 90 91 out_duplicate: 92 mutex_unlock(&xfs_uuid_table_mutex); 93 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); 94 return -EINVAL; 95} 96 97STATIC void 98xfs_uuid_unmount( 99 struct xfs_mount *mp) 100{ 101 uuid_t *uuid = &mp->m_sb.sb_uuid; 102 int i; 103 104 if (mp->m_flags & XFS_MOUNT_NOUUID) 105 return; 106 107 mutex_lock(&xfs_uuid_table_mutex); 108 for (i = 0; i < xfs_uuid_table_size; i++) { 109 if (uuid_is_nil(&xfs_uuid_table[i])) 110 continue; 111 if (!uuid_equal(uuid, &xfs_uuid_table[i])) 112 continue; 113 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); 114 break; 115 } 116 ASSERT(i < xfs_uuid_table_size); 117 mutex_unlock(&xfs_uuid_table_mutex); 118} 119 120 121STATIC void 122__xfs_free_perag( 123 struct rcu_head *head) 124{ 125 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); 126 127 ASSERT(atomic_read(&pag->pag_ref) == 0); 128 kmem_free(pag); 129} 130 131/* 132 * Free up the per-ag resources associated with the mount structure. 133 */ 134STATIC void 135xfs_free_perag( 136 xfs_mount_t *mp) 137{ 138 xfs_agnumber_t agno; 139 struct xfs_perag *pag; 140 141 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { 142 spin_lock(&mp->m_perag_lock); 143 pag = radix_tree_delete(&mp->m_perag_tree, agno); 144 spin_unlock(&mp->m_perag_lock); 145 ASSERT(pag); 146 ASSERT(atomic_read(&pag->pag_ref) == 0); 147 call_rcu(&pag->rcu_head, __xfs_free_perag); 148 } 149} 150 151/* 152 * Check size of device based on the (data/realtime) block count. 153 * Note: this check is used by the growfs code as well as mount. 154 */ 155int 156xfs_sb_validate_fsb_count( 157 xfs_sb_t *sbp, 158 __uint64_t nblocks) 159{ 160 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); 161 ASSERT(sbp->sb_blocklog >= BBSHIFT); 162 163 /* Limited by ULONG_MAX of page cache index */ 164 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) 165 return -EFBIG; 166 return 0; 167} 168 169int 170xfs_initialize_perag( 171 xfs_mount_t *mp, 172 xfs_agnumber_t agcount, 173 xfs_agnumber_t *maxagi) 174{ 175 xfs_agnumber_t index; 176 xfs_agnumber_t first_initialised = 0; 177 xfs_perag_t *pag; 178 xfs_agino_t agino; 179 xfs_ino_t ino; 180 xfs_sb_t *sbp = &mp->m_sb; 181 int error = -ENOMEM; 182 183 /* 184 * Walk the current per-ag tree so we don't try to initialise AGs 185 * that already exist (growfs case). Allocate and insert all the 186 * AGs we don't find ready for initialisation. 187 */ 188 for (index = 0; index < agcount; index++) { 189 pag = xfs_perag_get(mp, index); 190 if (pag) { 191 xfs_perag_put(pag); 192 continue; 193 } 194 if (!first_initialised) 195 first_initialised = index; 196 197 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); 198 if (!pag) 199 goto out_unwind; 200 pag->pag_agno = index; 201 pag->pag_mount = mp; 202 spin_lock_init(&pag->pag_ici_lock); 203 mutex_init(&pag->pag_ici_reclaim_lock); 204 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); 205 spin_lock_init(&pag->pag_buf_lock); 206 pag->pag_buf_tree = RB_ROOT; 207 208 if (radix_tree_preload(GFP_NOFS)) 209 goto out_unwind; 210 211 spin_lock(&mp->m_perag_lock); 212 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { 213 BUG(); 214 spin_unlock(&mp->m_perag_lock); 215 radix_tree_preload_end(); 216 error = -EEXIST; 217 goto out_unwind; 218 } 219 spin_unlock(&mp->m_perag_lock); 220 radix_tree_preload_end(); 221 } 222 223 /* 224 * If we mount with the inode64 option, or no inode overflows 225 * the legacy 32-bit address space clear the inode32 option. 226 */ 227 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); 228 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); 229 230 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) 231 mp->m_flags |= XFS_MOUNT_32BITINODES; 232 else 233 mp->m_flags &= ~XFS_MOUNT_32BITINODES; 234 235 if (mp->m_flags & XFS_MOUNT_32BITINODES) 236 index = xfs_set_inode32(mp, agcount); 237 else 238 index = xfs_set_inode64(mp, agcount); 239 240 if (maxagi) 241 *maxagi = index; 242 return 0; 243 244out_unwind: 245 kmem_free(pag); 246 for (; index > first_initialised; index--) { 247 pag = radix_tree_delete(&mp->m_perag_tree, index); 248 kmem_free(pag); 249 } 250 return error; 251} 252 253/* 254 * xfs_readsb 255 * 256 * Does the initial read of the superblock. 257 */ 258int 259xfs_readsb( 260 struct xfs_mount *mp, 261 int flags) 262{ 263 unsigned int sector_size; 264 struct xfs_buf *bp; 265 struct xfs_sb *sbp = &mp->m_sb; 266 int error; 267 int loud = !(flags & XFS_MFSI_QUIET); 268 const struct xfs_buf_ops *buf_ops; 269 270 ASSERT(mp->m_sb_bp == NULL); 271 ASSERT(mp->m_ddev_targp != NULL); 272 273 /* 274 * For the initial read, we must guess at the sector 275 * size based on the block device. It's enough to 276 * get the sb_sectsize out of the superblock and 277 * then reread with the proper length. 278 * We don't verify it yet, because it may not be complete. 279 */ 280 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); 281 buf_ops = NULL; 282 283 /* 284 * Allocate a (locked) buffer to hold the superblock. 285 * This will be kept around at all times to optimize 286 * access to the superblock. 287 */ 288reread: 289 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, 290 BTOBB(sector_size), 0, &bp, buf_ops); 291 if (error) { 292 if (loud) 293 xfs_warn(mp, "SB validate failed with error %d.", error); 294 /* bad CRC means corrupted metadata */ 295 if (error == -EFSBADCRC) 296 error = -EFSCORRUPTED; 297 return error; 298 } 299 300 /* 301 * Initialize the mount structure from the superblock. 302 */ 303 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); 304 305 /* 306 * If we haven't validated the superblock, do so now before we try 307 * to check the sector size and reread the superblock appropriately. 308 */ 309 if (sbp->sb_magicnum != XFS_SB_MAGIC) { 310 if (loud) 311 xfs_warn(mp, "Invalid superblock magic number"); 312 error = -EINVAL; 313 goto release_buf; 314 } 315 316 /* 317 * We must be able to do sector-sized and sector-aligned IO. 318 */ 319 if (sector_size > sbp->sb_sectsize) { 320 if (loud) 321 xfs_warn(mp, "device supports %u byte sectors (not %u)", 322 sector_size, sbp->sb_sectsize); 323 error = -ENOSYS; 324 goto release_buf; 325 } 326 327 if (buf_ops == NULL) { 328 /* 329 * Re-read the superblock so the buffer is correctly sized, 330 * and properly verified. 331 */ 332 xfs_buf_relse(bp); 333 sector_size = sbp->sb_sectsize; 334 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; 335 goto reread; 336 } 337 338 xfs_reinit_percpu_counters(mp); 339 340 /* no need to be quiet anymore, so reset the buf ops */ 341 bp->b_ops = &xfs_sb_buf_ops; 342 343 mp->m_sb_bp = bp; 344 xfs_buf_unlock(bp); 345 return 0; 346 347release_buf: 348 xfs_buf_relse(bp); 349 return error; 350} 351 352/* 353 * Update alignment values based on mount options and sb values 354 */ 355STATIC int 356xfs_update_alignment(xfs_mount_t *mp) 357{ 358 xfs_sb_t *sbp = &(mp->m_sb); 359 360 if (mp->m_dalign) { 361 /* 362 * If stripe unit and stripe width are not multiples 363 * of the fs blocksize turn off alignment. 364 */ 365 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || 366 (BBTOB(mp->m_swidth) & mp->m_blockmask)) { 367 xfs_warn(mp, 368 "alignment check failed: sunit/swidth vs. blocksize(%d)", 369 sbp->sb_blocksize); 370 return -EINVAL; 371 } else { 372 /* 373 * Convert the stripe unit and width to FSBs. 374 */ 375 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); 376 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { 377 xfs_warn(mp, 378 "alignment check failed: sunit/swidth vs. agsize(%d)", 379 sbp->sb_agblocks); 380 return -EINVAL; 381 } else if (mp->m_dalign) { 382 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); 383 } else { 384 xfs_warn(mp, 385 "alignment check failed: sunit(%d) less than bsize(%d)", 386 mp->m_dalign, sbp->sb_blocksize); 387 return -EINVAL; 388 } 389 } 390 391 /* 392 * Update superblock with new values 393 * and log changes 394 */ 395 if (xfs_sb_version_hasdalign(sbp)) { 396 if (sbp->sb_unit != mp->m_dalign) { 397 sbp->sb_unit = mp->m_dalign; 398 mp->m_update_sb = true; 399 } 400 if (sbp->sb_width != mp->m_swidth) { 401 sbp->sb_width = mp->m_swidth; 402 mp->m_update_sb = true; 403 } 404 } else { 405 xfs_warn(mp, 406 "cannot change alignment: superblock does not support data alignment"); 407 return -EINVAL; 408 } 409 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && 410 xfs_sb_version_hasdalign(&mp->m_sb)) { 411 mp->m_dalign = sbp->sb_unit; 412 mp->m_swidth = sbp->sb_width; 413 } 414 415 return 0; 416} 417 418/* 419 * Set the maximum inode count for this filesystem 420 */ 421STATIC void 422xfs_set_maxicount(xfs_mount_t *mp) 423{ 424 xfs_sb_t *sbp = &(mp->m_sb); 425 __uint64_t icount; 426 427 if (sbp->sb_imax_pct) { 428 /* 429 * Make sure the maximum inode count is a multiple 430 * of the units we allocate inodes in. 431 */ 432 icount = sbp->sb_dblocks * sbp->sb_imax_pct; 433 do_div(icount, 100); 434 do_div(icount, mp->m_ialloc_blks); 435 mp->m_maxicount = (icount * mp->m_ialloc_blks) << 436 sbp->sb_inopblog; 437 } else { 438 mp->m_maxicount = 0; 439 } 440} 441 442/* 443 * Set the default minimum read and write sizes unless 444 * already specified in a mount option. 445 * We use smaller I/O sizes when the file system 446 * is being used for NFS service (wsync mount option). 447 */ 448STATIC void 449xfs_set_rw_sizes(xfs_mount_t *mp) 450{ 451 xfs_sb_t *sbp = &(mp->m_sb); 452 int readio_log, writeio_log; 453 454 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { 455 if (mp->m_flags & XFS_MOUNT_WSYNC) { 456 readio_log = XFS_WSYNC_READIO_LOG; 457 writeio_log = XFS_WSYNC_WRITEIO_LOG; 458 } else { 459 readio_log = XFS_READIO_LOG_LARGE; 460 writeio_log = XFS_WRITEIO_LOG_LARGE; 461 } 462 } else { 463 readio_log = mp->m_readio_log; 464 writeio_log = mp->m_writeio_log; 465 } 466 467 if (sbp->sb_blocklog > readio_log) { 468 mp->m_readio_log = sbp->sb_blocklog; 469 } else { 470 mp->m_readio_log = readio_log; 471 } 472 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); 473 if (sbp->sb_blocklog > writeio_log) { 474 mp->m_writeio_log = sbp->sb_blocklog; 475 } else { 476 mp->m_writeio_log = writeio_log; 477 } 478 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); 479} 480 481/* 482 * precalculate the low space thresholds for dynamic speculative preallocation. 483 */ 484void 485xfs_set_low_space_thresholds( 486 struct xfs_mount *mp) 487{ 488 int i; 489 490 for (i = 0; i < XFS_LOWSP_MAX; i++) { 491 __uint64_t space = mp->m_sb.sb_dblocks; 492 493 do_div(space, 100); 494 mp->m_low_space[i] = space * (i + 1); 495 } 496} 497 498 499/* 500 * Set whether we're using inode alignment. 501 */ 502STATIC void 503xfs_set_inoalignment(xfs_mount_t *mp) 504{ 505 if (xfs_sb_version_hasalign(&mp->m_sb) && 506 mp->m_sb.sb_inoalignmt >= 507 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) 508 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; 509 else 510 mp->m_inoalign_mask = 0; 511 /* 512 * If we are using stripe alignment, check whether 513 * the stripe unit is a multiple of the inode alignment 514 */ 515 if (mp->m_dalign && mp->m_inoalign_mask && 516 !(mp->m_dalign & mp->m_inoalign_mask)) 517 mp->m_sinoalign = mp->m_dalign; 518 else 519 mp->m_sinoalign = 0; 520} 521 522/* 523 * Check that the data (and log if separate) is an ok size. 524 */ 525STATIC int 526xfs_check_sizes( 527 struct xfs_mount *mp) 528{ 529 struct xfs_buf *bp; 530 xfs_daddr_t d; 531 int error; 532 533 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 534 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { 535 xfs_warn(mp, "filesystem size mismatch detected"); 536 return -EFBIG; 537 } 538 error = xfs_buf_read_uncached(mp->m_ddev_targp, 539 d - XFS_FSS_TO_BB(mp, 1), 540 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 541 if (error) { 542 xfs_warn(mp, "last sector read failed"); 543 return error; 544 } 545 xfs_buf_relse(bp); 546 547 if (mp->m_logdev_targp == mp->m_ddev_targp) 548 return 0; 549 550 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 551 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { 552 xfs_warn(mp, "log size mismatch detected"); 553 return -EFBIG; 554 } 555 error = xfs_buf_read_uncached(mp->m_logdev_targp, 556 d - XFS_FSB_TO_BB(mp, 1), 557 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); 558 if (error) { 559 xfs_warn(mp, "log device read failed"); 560 return error; 561 } 562 xfs_buf_relse(bp); 563 return 0; 564} 565 566/* 567 * Clear the quotaflags in memory and in the superblock. 568 */ 569int 570xfs_mount_reset_sbqflags( 571 struct xfs_mount *mp) 572{ 573 mp->m_qflags = 0; 574 575 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ 576 if (mp->m_sb.sb_qflags == 0) 577 return 0; 578 spin_lock(&mp->m_sb_lock); 579 mp->m_sb.sb_qflags = 0; 580 spin_unlock(&mp->m_sb_lock); 581 582 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) 583 return 0; 584 585 return xfs_sync_sb(mp, false); 586} 587 588__uint64_t 589xfs_default_resblks(xfs_mount_t *mp) 590{ 591 __uint64_t resblks; 592 593 /* 594 * We default to 5% or 8192 fsbs of space reserved, whichever is 595 * smaller. This is intended to cover concurrent allocation 596 * transactions when we initially hit enospc. These each require a 4 597 * block reservation. Hence by default we cover roughly 2000 concurrent 598 * allocation reservations. 599 */ 600 resblks = mp->m_sb.sb_dblocks; 601 do_div(resblks, 20); 602 resblks = min_t(__uint64_t, resblks, 8192); 603 return resblks; 604} 605 606/* 607 * This function does the following on an initial mount of a file system: 608 * - reads the superblock from disk and init the mount struct 609 * - if we're a 32-bit kernel, do a size check on the superblock 610 * so we don't mount terabyte filesystems 611 * - init mount struct realtime fields 612 * - allocate inode hash table for fs 613 * - init directory manager 614 * - perform recovery and init the log manager 615 */ 616int 617xfs_mountfs( 618 xfs_mount_t *mp) 619{ 620 xfs_sb_t *sbp = &(mp->m_sb); 621 xfs_inode_t *rip; 622 __uint64_t resblks; 623 uint quotamount = 0; 624 uint quotaflags = 0; 625 int error = 0; 626 627 xfs_sb_mount_common(mp, sbp); 628 629 /* 630 * Check for a mismatched features2 values. Older kernels read & wrote 631 * into the wrong sb offset for sb_features2 on some platforms due to 632 * xfs_sb_t not being 64bit size aligned when sb_features2 was added, 633 * which made older superblock reading/writing routines swap it as a 634 * 64-bit value. 635 * 636 * For backwards compatibility, we make both slots equal. 637 * 638 * If we detect a mismatched field, we OR the set bits into the existing 639 * features2 field in case it has already been modified; we don't want 640 * to lose any features. We then update the bad location with the ORed 641 * value so that older kernels will see any features2 flags. The 642 * superblock writeback code ensures the new sb_features2 is copied to 643 * sb_bad_features2 before it is logged or written to disk. 644 */ 645 if (xfs_sb_has_mismatched_features2(sbp)) { 646 xfs_warn(mp, "correcting sb_features alignment problem"); 647 sbp->sb_features2 |= sbp->sb_bad_features2; 648 mp->m_update_sb = true; 649 650 /* 651 * Re-check for ATTR2 in case it was found in bad_features2 652 * slot. 653 */ 654 if (xfs_sb_version_hasattr2(&mp->m_sb) && 655 !(mp->m_flags & XFS_MOUNT_NOATTR2)) 656 mp->m_flags |= XFS_MOUNT_ATTR2; 657 } 658 659 if (xfs_sb_version_hasattr2(&mp->m_sb) && 660 (mp->m_flags & XFS_MOUNT_NOATTR2)) { 661 xfs_sb_version_removeattr2(&mp->m_sb); 662 mp->m_update_sb = true; 663 664 /* update sb_versionnum for the clearing of the morebits */ 665 if (!sbp->sb_features2) 666 mp->m_update_sb = true; 667 } 668 669 /* always use v2 inodes by default now */ 670 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { 671 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; 672 mp->m_update_sb = true; 673 } 674 675 /* 676 * Check if sb_agblocks is aligned at stripe boundary 677 * If sb_agblocks is NOT aligned turn off m_dalign since 678 * allocator alignment is within an ag, therefore ag has 679 * to be aligned at stripe boundary. 680 */ 681 error = xfs_update_alignment(mp); 682 if (error) 683 goto out; 684 685 xfs_alloc_compute_maxlevels(mp); 686 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); 687 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); 688 xfs_ialloc_compute_maxlevels(mp); 689 690 xfs_set_maxicount(mp); 691 692 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); 693 if (error) 694 goto out; 695 696 error = xfs_uuid_mount(mp); 697 if (error) 698 goto out_remove_sysfs; 699 700 /* 701 * Set the minimum read and write sizes 702 */ 703 xfs_set_rw_sizes(mp); 704 705 /* set the low space thresholds for dynamic preallocation */ 706 xfs_set_low_space_thresholds(mp); 707 708 /* 709 * Set the inode cluster size. 710 * This may still be overridden by the file system 711 * block size if it is larger than the chosen cluster size. 712 * 713 * For v5 filesystems, scale the cluster size with the inode size to 714 * keep a constant ratio of inode per cluster buffer, but only if mkfs 715 * has set the inode alignment value appropriately for larger cluster 716 * sizes. 717 */ 718 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; 719 if (xfs_sb_version_hascrc(&mp->m_sb)) { 720 int new_size = mp->m_inode_cluster_size; 721 722 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; 723 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) 724 mp->m_inode_cluster_size = new_size; 725 } 726 727 /* 728 * Set inode alignment fields 729 */ 730 xfs_set_inoalignment(mp); 731 732 /* 733 * Check that the data (and log if separate) is an ok size. 734 */ 735 error = xfs_check_sizes(mp); 736 if (error) 737 goto out_remove_uuid; 738 739 /* 740 * Initialize realtime fields in the mount structure 741 */ 742 error = xfs_rtmount_init(mp); 743 if (error) { 744 xfs_warn(mp, "RT mount failed"); 745 goto out_remove_uuid; 746 } 747 748 /* 749 * Copies the low order bits of the timestamp and the randomly 750 * set "sequence" number out of a UUID. 751 */ 752 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); 753 754 mp->m_dmevmask = 0; /* not persistent; set after each mount */ 755 756 error = xfs_da_mount(mp); 757 if (error) { 758 xfs_warn(mp, "Failed dir/attr init: %d", error); 759 goto out_remove_uuid; 760 } 761 762 /* 763 * Initialize the precomputed transaction reservations values. 764 */ 765 xfs_trans_init(mp); 766 767 /* 768 * Allocate and initialize the per-ag data. 769 */ 770 spin_lock_init(&mp->m_perag_lock); 771 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); 772 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); 773 if (error) { 774 xfs_warn(mp, "Failed per-ag init: %d", error); 775 goto out_free_dir; 776 } 777 778 if (!sbp->sb_logblocks) { 779 xfs_warn(mp, "no log defined"); 780 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); 781 error = -EFSCORRUPTED; 782 goto out_free_perag; 783 } 784 785 /* 786 * log's mount-time initialization. Perform 1st part recovery if needed 787 */ 788 error = xfs_log_mount(mp, mp->m_logdev_targp, 789 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), 790 XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); 791 if (error) { 792 xfs_warn(mp, "log mount failed"); 793 goto out_fail_wait; 794 } 795 796 /* 797 * Now the log is mounted, we know if it was an unclean shutdown or 798 * not. If it was, with the first phase of recovery has completed, we 799 * have consistent AG blocks on disk. We have not recovered EFIs yet, 800 * but they are recovered transactionally in the second recovery phase 801 * later. 802 * 803 * Hence we can safely re-initialise incore superblock counters from 804 * the per-ag data. These may not be correct if the filesystem was not 805 * cleanly unmounted, so we need to wait for recovery to finish before 806 * doing this. 807 * 808 * If the filesystem was cleanly unmounted, then we can trust the 809 * values in the superblock to be correct and we don't need to do 810 * anything here. 811 * 812 * If we are currently making the filesystem, the initialisation will 813 * fail as the perag data is in an undefined state. 814 */ 815 if (xfs_sb_version_haslazysbcount(&mp->m_sb) && 816 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && 817 !mp->m_sb.sb_inprogress) { 818 error = xfs_initialize_perag_data(mp, sbp->sb_agcount); 819 if (error) 820 goto out_log_dealloc; 821 } 822 823 /* 824 * Get and sanity-check the root inode. 825 * Save the pointer to it in the mount structure. 826 */ 827 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); 828 if (error) { 829 xfs_warn(mp, "failed to read root inode"); 830 goto out_log_dealloc; 831 } 832 833 ASSERT(rip != NULL); 834 835 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { 836 xfs_warn(mp, "corrupted root inode %llu: not a directory", 837 (unsigned long long)rip->i_ino); 838 xfs_iunlock(rip, XFS_ILOCK_EXCL); 839 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, 840 mp); 841 error = -EFSCORRUPTED; 842 goto out_rele_rip; 843 } 844 mp->m_rootip = rip; /* save it */ 845 846 xfs_iunlock(rip, XFS_ILOCK_EXCL); 847 848 /* 849 * Initialize realtime inode pointers in the mount structure 850 */ 851 error = xfs_rtmount_inodes(mp); 852 if (error) { 853 /* 854 * Free up the root inode. 855 */ 856 xfs_warn(mp, "failed to read RT inodes"); 857 goto out_rele_rip; 858 } 859 860 /* 861 * If this is a read-only mount defer the superblock updates until 862 * the next remount into writeable mode. Otherwise we would never 863 * perform the update e.g. for the root filesystem. 864 */ 865 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { 866 error = xfs_sync_sb(mp, false); 867 if (error) { 868 xfs_warn(mp, "failed to write sb changes"); 869 goto out_rtunmount; 870 } 871 } 872 873 /* 874 * Initialise the XFS quota management subsystem for this mount 875 */ 876 if (XFS_IS_QUOTA_RUNNING(mp)) { 877 error = xfs_qm_newmount(mp, "amount, "aflags); 878 if (error) 879 goto out_rtunmount; 880 } else { 881 ASSERT(!XFS_IS_QUOTA_ON(mp)); 882 883 /* 884 * If a file system had quotas running earlier, but decided to 885 * mount without -o uquota/pquota/gquota options, revoke the 886 * quotachecked license. 887 */ 888 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { 889 xfs_notice(mp, "resetting quota flags"); 890 error = xfs_mount_reset_sbqflags(mp); 891 if (error) 892 goto out_rtunmount; 893 } 894 } 895 896 /* 897 * Finish recovering the file system. This part needed to be 898 * delayed until after the root and real-time bitmap inodes 899 * were consistently read in. 900 */ 901 error = xfs_log_mount_finish(mp); 902 if (error) { 903 xfs_warn(mp, "log mount finish failed"); 904 goto out_rtunmount; 905 } 906 907 /* 908 * Complete the quota initialisation, post-log-replay component. 909 */ 910 if (quotamount) { 911 ASSERT(mp->m_qflags == 0); 912 mp->m_qflags = quotaflags; 913 914 xfs_qm_mount_quotas(mp); 915 } 916 917 /* 918 * Now we are mounted, reserve a small amount of unused space for 919 * privileged transactions. This is needed so that transaction 920 * space required for critical operations can dip into this pool 921 * when at ENOSPC. This is needed for operations like create with 922 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations 923 * are not allowed to use this reserved space. 924 * 925 * This may drive us straight to ENOSPC on mount, but that implies 926 * we were already there on the last unmount. Warn if this occurs. 927 */ 928 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { 929 resblks = xfs_default_resblks(mp); 930 error = xfs_reserve_blocks(mp, &resblks, NULL); 931 if (error) 932 xfs_warn(mp, 933 "Unable to allocate reserve blocks. Continuing without reserve pool."); 934 } 935 936 return 0; 937 938 out_rtunmount: 939 xfs_rtunmount_inodes(mp); 940 out_rele_rip: 941 IRELE(rip); 942 out_log_dealloc: 943 xfs_log_unmount(mp); 944 out_fail_wait: 945 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) 946 xfs_wait_buftarg(mp->m_logdev_targp); 947 xfs_wait_buftarg(mp->m_ddev_targp); 948 out_free_perag: 949 xfs_free_perag(mp); 950 out_free_dir: 951 xfs_da_unmount(mp); 952 out_remove_uuid: 953 xfs_uuid_unmount(mp); 954 out_remove_sysfs: 955 xfs_sysfs_del(&mp->m_kobj); 956 out: 957 return error; 958} 959 960/* 961 * This flushes out the inodes,dquots and the superblock, unmounts the 962 * log and makes sure that incore structures are freed. 963 */ 964void 965xfs_unmountfs( 966 struct xfs_mount *mp) 967{ 968 __uint64_t resblks; 969 int error; 970 971 cancel_delayed_work_sync(&mp->m_eofblocks_work); 972 973 xfs_qm_unmount_quotas(mp); 974 xfs_rtunmount_inodes(mp); 975 IRELE(mp->m_rootip); 976 977 /* 978 * We can potentially deadlock here if we have an inode cluster 979 * that has been freed has its buffer still pinned in memory because 980 * the transaction is still sitting in a iclog. The stale inodes 981 * on that buffer will have their flush locks held until the 982 * transaction hits the disk and the callbacks run. the inode 983 * flush takes the flush lock unconditionally and with nothing to 984 * push out the iclog we will never get that unlocked. hence we 985 * need to force the log first. 986 */ 987 xfs_log_force(mp, XFS_LOG_SYNC); 988 989 /* 990 * Flush all pending changes from the AIL. 991 */ 992 xfs_ail_push_all_sync(mp->m_ail); 993 994 /* 995 * And reclaim all inodes. At this point there should be no dirty 996 * inodes and none should be pinned or locked, but use synchronous 997 * reclaim just to be sure. We can stop background inode reclaim 998 * here as well if it is still running. 999 */ 1000 cancel_delayed_work_sync(&mp->m_reclaim_work); 1001 xfs_reclaim_inodes(mp, SYNC_WAIT); 1002 1003 xfs_qm_unmount(mp); 1004 1005 /* 1006 * Unreserve any blocks we have so that when we unmount we don't account 1007 * the reserved free space as used. This is really only necessary for 1008 * lazy superblock counting because it trusts the incore superblock 1009 * counters to be absolutely correct on clean unmount. 1010 * 1011 * We don't bother correcting this elsewhere for lazy superblock 1012 * counting because on mount of an unclean filesystem we reconstruct the 1013 * correct counter value and this is irrelevant. 1014 * 1015 * For non-lazy counter filesystems, this doesn't matter at all because 1016 * we only every apply deltas to the superblock and hence the incore 1017 * value does not matter.... 1018 */ 1019 resblks = 0; 1020 error = xfs_reserve_blocks(mp, &resblks, NULL); 1021 if (error) 1022 xfs_warn(mp, "Unable to free reserved block pool. " 1023 "Freespace may not be correct on next mount."); 1024 1025 error = xfs_log_sbcount(mp); 1026 if (error) 1027 xfs_warn(mp, "Unable to update superblock counters. " 1028 "Freespace may not be correct on next mount."); 1029 1030 xfs_log_unmount(mp); 1031 xfs_da_unmount(mp); 1032 xfs_uuid_unmount(mp); 1033 1034#if defined(DEBUG) 1035 xfs_errortag_clearall(mp, 0); 1036#endif 1037 xfs_free_perag(mp); 1038 1039 xfs_sysfs_del(&mp->m_kobj); 1040} 1041 1042/* 1043 * Determine whether modifications can proceed. The caller specifies the minimum 1044 * freeze level for which modifications should not be allowed. This allows 1045 * certain operations to proceed while the freeze sequence is in progress, if 1046 * necessary. 1047 */ 1048bool 1049xfs_fs_writable( 1050 struct xfs_mount *mp, 1051 int level) 1052{ 1053 ASSERT(level > SB_UNFROZEN); 1054 if ((mp->m_super->s_writers.frozen >= level) || 1055 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) 1056 return false; 1057 1058 return true; 1059} 1060 1061/* 1062 * xfs_log_sbcount 1063 * 1064 * Sync the superblock counters to disk. 1065 * 1066 * Note this code can be called during the process of freezing, so we use the 1067 * transaction allocator that does not block when the transaction subsystem is 1068 * in its frozen state. 1069 */ 1070int 1071xfs_log_sbcount(xfs_mount_t *mp) 1072{ 1073 /* allow this to proceed during the freeze sequence... */ 1074 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) 1075 return 0; 1076 1077 /* 1078 * we don't need to do this if we are updating the superblock 1079 * counters on every modification. 1080 */ 1081 if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) 1082 return 0; 1083 1084 return xfs_sync_sb(mp, true); 1085} 1086 1087/* 1088 * Deltas for the inode count are +/-64, hence we use a large batch size 1089 * of 128 so we don't need to take the counter lock on every update. 1090 */ 1091#define XFS_ICOUNT_BATCH 128 1092int 1093xfs_mod_icount( 1094 struct xfs_mount *mp, 1095 int64_t delta) 1096{ 1097 __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); 1098 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { 1099 ASSERT(0); 1100 percpu_counter_add(&mp->m_icount, -delta); 1101 return -EINVAL; 1102 } 1103 return 0; 1104} 1105 1106int 1107xfs_mod_ifree( 1108 struct xfs_mount *mp, 1109 int64_t delta) 1110{ 1111 percpu_counter_add(&mp->m_ifree, delta); 1112 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { 1113 ASSERT(0); 1114 percpu_counter_add(&mp->m_ifree, -delta); 1115 return -EINVAL; 1116 } 1117 return 0; 1118} 1119 1120/* 1121 * Deltas for the block count can vary from 1 to very large, but lock contention 1122 * only occurs on frequent small block count updates such as in the delayed 1123 * allocation path for buffered writes (page a time updates). Hence we set 1124 * a large batch count (1024) to minimise global counter updates except when 1125 * we get near to ENOSPC and we have to be very accurate with our updates. 1126 */ 1127#define XFS_FDBLOCKS_BATCH 1024 1128int 1129xfs_mod_fdblocks( 1130 struct xfs_mount *mp, 1131 int64_t delta, 1132 bool rsvd) 1133{ 1134 int64_t lcounter; 1135 long long res_used; 1136 s32 batch; 1137 1138 if (delta > 0) { 1139 /* 1140 * If the reserve pool is depleted, put blocks back into it 1141 * first. Most of the time the pool is full. 1142 */ 1143 if (likely(mp->m_resblks == mp->m_resblks_avail)) { 1144 percpu_counter_add(&mp->m_fdblocks, delta); 1145 return 0; 1146 } 1147 1148 spin_lock(&mp->m_sb_lock); 1149 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); 1150 1151 if (res_used > delta) { 1152 mp->m_resblks_avail += delta; 1153 } else { 1154 delta -= res_used; 1155 mp->m_resblks_avail = mp->m_resblks; 1156 percpu_counter_add(&mp->m_fdblocks, delta); 1157 } 1158 spin_unlock(&mp->m_sb_lock); 1159 return 0; 1160 } 1161 1162 /* 1163 * Taking blocks away, need to be more accurate the closer we 1164 * are to zero. 1165 * 1166 * If the counter has a value of less than 2 * max batch size, 1167 * then make everything serialise as we are real close to 1168 * ENOSPC. 1169 */ 1170 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, 1171 XFS_FDBLOCKS_BATCH) < 0) 1172 batch = 1; 1173 else 1174 batch = XFS_FDBLOCKS_BATCH; 1175 1176 __percpu_counter_add(&mp->m_fdblocks, delta, batch); 1177 if (__percpu_counter_compare(&mp->m_fdblocks, XFS_ALLOC_SET_ASIDE(mp), 1178 XFS_FDBLOCKS_BATCH) >= 0) { 1179 /* we had space! */ 1180 return 0; 1181 } 1182 1183 /* 1184 * lock up the sb for dipping into reserves before releasing the space 1185 * that took us to ENOSPC. 1186 */ 1187 spin_lock(&mp->m_sb_lock); 1188 percpu_counter_add(&mp->m_fdblocks, -delta); 1189 if (!rsvd) 1190 goto fdblocks_enospc; 1191 1192 lcounter = (long long)mp->m_resblks_avail + delta; 1193 if (lcounter >= 0) { 1194 mp->m_resblks_avail = lcounter; 1195 spin_unlock(&mp->m_sb_lock); 1196 return 0; 1197 } 1198 printk_once(KERN_WARNING 1199 "Filesystem \"%s\": reserve blocks depleted! " 1200 "Consider increasing reserve pool size.", 1201 mp->m_fsname); 1202fdblocks_enospc: 1203 spin_unlock(&mp->m_sb_lock); 1204 return -ENOSPC; 1205} 1206 1207int 1208xfs_mod_frextents( 1209 struct xfs_mount *mp, 1210 int64_t delta) 1211{ 1212 int64_t lcounter; 1213 int ret = 0; 1214 1215 spin_lock(&mp->m_sb_lock); 1216 lcounter = mp->m_sb.sb_frextents + delta; 1217 if (lcounter < 0) 1218 ret = -ENOSPC; 1219 else 1220 mp->m_sb.sb_frextents = lcounter; 1221 spin_unlock(&mp->m_sb_lock); 1222 return ret; 1223} 1224 1225/* 1226 * xfs_getsb() is called to obtain the buffer for the superblock. 1227 * The buffer is returned locked and read in from disk. 1228 * The buffer should be released with a call to xfs_brelse(). 1229 * 1230 * If the flags parameter is BUF_TRYLOCK, then we'll only return 1231 * the superblock buffer if it can be locked without sleeping. 1232 * If it can't then we'll return NULL. 1233 */ 1234struct xfs_buf * 1235xfs_getsb( 1236 struct xfs_mount *mp, 1237 int flags) 1238{ 1239 struct xfs_buf *bp = mp->m_sb_bp; 1240 1241 if (!xfs_buf_trylock(bp)) { 1242 if (flags & XBF_TRYLOCK) 1243 return NULL; 1244 xfs_buf_lock(bp); 1245 } 1246 1247 xfs_buf_hold(bp); 1248 ASSERT(XFS_BUF_ISDONE(bp)); 1249 return bp; 1250} 1251 1252/* 1253 * Used to free the superblock along various error paths. 1254 */ 1255void 1256xfs_freesb( 1257 struct xfs_mount *mp) 1258{ 1259 struct xfs_buf *bp = mp->m_sb_bp; 1260 1261 xfs_buf_lock(bp); 1262 mp->m_sb_bp = NULL; 1263 xfs_buf_relse(bp); 1264} 1265 1266/* 1267 * If the underlying (data/log/rt) device is readonly, there are some 1268 * operations that cannot proceed. 1269 */ 1270int 1271xfs_dev_is_read_only( 1272 struct xfs_mount *mp, 1273 char *message) 1274{ 1275 if (xfs_readonly_buftarg(mp->m_ddev_targp) || 1276 xfs_readonly_buftarg(mp->m_logdev_targp) || 1277 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { 1278 xfs_notice(mp, "%s required on read-only device.", message); 1279 xfs_notice(mp, "write access unavailable, cannot proceed."); 1280 return -EROFS; 1281 } 1282 return 0; 1283} 1284