xfs_mount.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   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_bit.h"
  13#include "xfs_sb.h"
  14#include "xfs_mount.h"
  15#include "xfs_inode.h"
  16#include "xfs_dir2.h"
  17#include "xfs_ialloc.h"
  18#include "xfs_alloc.h"
  19#include "xfs_rtalloc.h"
  20#include "xfs_bmap.h"
  21#include "xfs_trans.h"
  22#include "xfs_trans_priv.h"
  23#include "xfs_log.h"
  24#include "xfs_log_priv.h"
  25#include "xfs_error.h"
  26#include "xfs_quota.h"
  27#include "xfs_fsops.h"
  28#include "xfs_icache.h"
  29#include "xfs_sysfs.h"
  30#include "xfs_rmap_btree.h"
  31#include "xfs_refcount_btree.h"
  32#include "xfs_reflink.h"
  33#include "xfs_extent_busy.h"
  34#include "xfs_health.h"
  35#include "xfs_trace.h"
  36#include "xfs_ag.h"
  37#include "xfs_rtbitmap.h"
  38#include "xfs_metafile.h"
  39#include "xfs_rtgroup.h"
  40#include "xfs_rtrmap_btree.h"
  41#include "xfs_rtrefcount_btree.h"
  42#include "scrub/stats.h"
  43#include "xfs_zone_alloc.h"
  44
  45static DEFINE_MUTEX(xfs_uuid_table_mutex);
  46static int xfs_uuid_table_size;
  47static uuid_t *xfs_uuid_table;
  48
  49void
  50xfs_uuid_table_free(void)
  51{
  52	if (xfs_uuid_table_size == 0)
  53		return;
  54	kfree(xfs_uuid_table);
  55	xfs_uuid_table = NULL;
  56	xfs_uuid_table_size = 0;
  57}
  58
  59/*
  60 * See if the UUID is unique among mounted XFS filesystems.
  61 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  62 */
  63STATIC int
  64xfs_uuid_mount(
  65	struct xfs_mount	*mp)
  66{
  67	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  68	int			hole, i;
  69
  70	/* Publish UUID in struct super_block */
  71	super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
  72
  73	if (xfs_has_nouuid(mp))
  74		return 0;
  75
  76	if (uuid_is_null(uuid)) {
  77		xfs_warn(mp, "Filesystem has null UUID - can't mount");
  78		return -EINVAL;
  79	}
  80
  81	mutex_lock(&xfs_uuid_table_mutex);
  82	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  83		if (uuid_is_null(&xfs_uuid_table[i])) {
  84			hole = i;
  85			continue;
  86		}
  87		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  88			goto out_duplicate;
  89	}
  90
  91	if (hole < 0) {
  92		xfs_uuid_table = krealloc(xfs_uuid_table,
  93			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
  94			GFP_KERNEL | __GFP_NOFAIL);
  95		hole = xfs_uuid_table_size++;
  96	}
  97	xfs_uuid_table[hole] = *uuid;
  98	mutex_unlock(&xfs_uuid_table_mutex);
  99
 100	return 0;
 101
 102 out_duplicate:
 103	mutex_unlock(&xfs_uuid_table_mutex);
 104	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
 105	return -EINVAL;
 106}
 107
 108STATIC void
 109xfs_uuid_unmount(
 110	struct xfs_mount	*mp)
 111{
 112	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 113	int			i;
 114
 115	if (xfs_has_nouuid(mp))
 116		return;
 117
 118	mutex_lock(&xfs_uuid_table_mutex);
 119	for (i = 0; i < xfs_uuid_table_size; i++) {
 120		if (uuid_is_null(&xfs_uuid_table[i]))
 121			continue;
 122		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 123			continue;
 124		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 125		break;
 126	}
 127	ASSERT(i < xfs_uuid_table_size);
 128	mutex_unlock(&xfs_uuid_table_mutex);
 129}
 130
 131/*
 132 * Check size of device based on the (data/realtime) block count.
 133 * Note: this check is used by the growfs code as well as mount.
 134 */
 135int
 136xfs_sb_validate_fsb_count(
 137	xfs_sb_t	*sbp,
 138	uint64_t	nblocks)
 139{
 140	uint64_t		max_bytes;
 141
 142	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 143
 144	if (check_shl_overflow(nblocks, sbp->sb_blocklog, &max_bytes))
 145		return -EFBIG;
 146
 147	/* Limited by ULONG_MAX of page cache index */
 148	if (max_bytes >> PAGE_SHIFT > ULONG_MAX)
 149		return -EFBIG;
 150	return 0;
 151}
 152
 153/*
 154 * xfs_readsb
 155 *
 156 * Does the initial read of the superblock.
 157 */
 158int
 159xfs_readsb(
 160	struct xfs_mount *mp,
 161	int		flags)
 162{
 163	unsigned int	sector_size;
 164	struct xfs_buf	*bp;
 165	struct xfs_sb	*sbp = &mp->m_sb;
 166	int		error;
 167	int		loud = !(flags & XFS_MFSI_QUIET);
 168	const struct xfs_buf_ops *buf_ops;
 169
 170	ASSERT(mp->m_sb_bp == NULL);
 171	ASSERT(mp->m_ddev_targp != NULL);
 172
 173	/*
 174	 * In the first pass, use the device sector size to just read enough
 175	 * of the superblock to extract the XFS sector size.
 176	 *
 177	 * The device sector size must be smaller than or equal to the XFS
 178	 * sector size and thus we can always read the superblock.  Once we know
 179	 * the XFS sector size, re-read it and run the buffer verifier.
 180	 */
 181	sector_size = mp->m_ddev_targp->bt_logical_sectorsize;
 182	buf_ops = NULL;
 183
 184reread:
 185	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 186				      BTOBB(sector_size), &bp, buf_ops);
 187	if (error) {
 188		if (loud)
 189			xfs_warn(mp, "SB validate failed with error %d.", error);
 190		/* bad CRC means corrupted metadata */
 191		if (error == -EFSBADCRC)
 192			error = -EFSCORRUPTED;
 193		return error;
 194	}
 195
 196	/*
 197	 * Initialize the mount structure from the superblock.
 198	 */
 199	xfs_sb_from_disk(sbp, bp->b_addr);
 200
 201	/*
 202	 * If we haven't validated the superblock, do so now before we try
 203	 * to check the sector size and reread the superblock appropriately.
 204	 */
 205	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 206		if (loud)
 207			xfs_warn(mp, "Invalid superblock magic number");
 208		error = -EINVAL;
 209		goto release_buf;
 210	}
 211
 212	/*
 213	 * We must be able to do sector-sized and sector-aligned IO.
 214	 */
 215	if (sector_size > sbp->sb_sectsize) {
 216		if (loud)
 217			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 218				sector_size, sbp->sb_sectsize);
 219		error = -ENOSYS;
 220		goto release_buf;
 221	}
 222
 223	if (buf_ops == NULL) {
 224		/*
 225		 * Re-read the superblock so the buffer is correctly sized,
 226		 * and properly verified.
 227		 */
 228		xfs_buf_relse(bp);
 229		sector_size = sbp->sb_sectsize;
 230		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 231		goto reread;
 232	}
 233
 234	mp->m_features |= xfs_sb_version_to_features(sbp);
 235	xfs_reinit_percpu_counters(mp);
 236
 237	/*
 238	 * If logged xattrs are enabled after log recovery finishes, then set
 239	 * the opstate so that log recovery will work properly.
 240	 */
 241	if (xfs_sb_version_haslogxattrs(&mp->m_sb))
 242		xfs_set_using_logged_xattrs(mp);
 243
 244	/* no need to be quiet anymore, so reset the buf ops */
 245	bp->b_ops = &xfs_sb_buf_ops;
 246
 247	/*
 248	 * Keep a pointer of the sb buffer around instead of caching it in the
 249	 * buffer cache because we access it frequently.
 250	 */
 251	mp->m_sb_bp = bp;
 252	xfs_buf_unlock(bp);
 253	return 0;
 254
 255release_buf:
 256	xfs_buf_relse(bp);
 257	return error;
 258}
 259
 260/*
 261 * If the sunit/swidth change would move the precomputed root inode value, we
 262 * must reject the ondisk change because repair will stumble over that.
 263 * However, we allow the mount to proceed because we never rejected this
 264 * combination before.  Returns true to update the sb, false otherwise.
 265 */
 266static inline int
 267xfs_check_new_dalign(
 268	struct xfs_mount	*mp,
 269	int			new_dalign,
 270	bool			*update_sb)
 271{
 272	struct xfs_sb		*sbp = &mp->m_sb;
 273	xfs_ino_t		calc_ino;
 274
 275	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
 276	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
 277
 278	if (sbp->sb_rootino == calc_ino) {
 279		*update_sb = true;
 280		return 0;
 281	}
 282
 283	xfs_warn(mp,
 284"Cannot change stripe alignment; would require moving root inode.");
 285
 286	/*
 287	 * XXX: Next time we add a new incompat feature, this should start
 288	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
 289	 * that we're ignoring the administrator's instructions.
 290	 */
 291	xfs_warn(mp, "Skipping superblock stripe alignment update.");
 292	*update_sb = false;
 293	return 0;
 294}
 295
 296/*
 297 * If we were provided with new sunit/swidth values as mount options, make sure
 298 * that they pass basic alignment and superblock feature checks, and convert
 299 * them into the same units (FSB) that everything else expects.  This step
 300 * /must/ be done before computing the inode geometry.
 301 */
 302STATIC int
 303xfs_validate_new_dalign(
 304	struct xfs_mount	*mp)
 305{
 306	if (mp->m_dalign == 0)
 307		return 0;
 308
 309	/*
 310	 * If stripe unit and stripe width are not multiples
 311	 * of the fs blocksize turn off alignment.
 312	 */
 313	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 314	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 315		xfs_warn(mp,
 316	"alignment check failed: sunit/swidth vs. blocksize(%d)",
 317			mp->m_sb.sb_blocksize);
 318		return -EINVAL;
 319	}
 320
 321	/*
 322	 * Convert the stripe unit and width to FSBs.
 323	 */
 324	mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 325	if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
 326		xfs_warn(mp,
 327	"alignment check failed: sunit/swidth vs. agsize(%d)",
 328			mp->m_sb.sb_agblocks);
 329		return -EINVAL;
 330	}
 331
 332	if (!mp->m_dalign) {
 333		xfs_warn(mp,
 334	"alignment check failed: sunit(%d) less than bsize(%d)",
 335			mp->m_dalign, mp->m_sb.sb_blocksize);
 336		return -EINVAL;
 337	}
 338
 339	mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 340
 341	if (!xfs_has_dalign(mp)) {
 342		xfs_warn(mp,
 343"cannot change alignment: superblock does not support data alignment");
 344		return -EINVAL;
 345	}
 346
 347	return 0;
 348}
 349
 350/* Update alignment values based on mount options and sb values. */
 351STATIC int
 352xfs_update_alignment(
 353	struct xfs_mount	*mp)
 354{
 355	struct xfs_sb		*sbp = &mp->m_sb;
 356
 357	if (mp->m_dalign) {
 358		bool		update_sb;
 359		int		error;
 360
 361		if (sbp->sb_unit == mp->m_dalign &&
 362		    sbp->sb_width == mp->m_swidth)
 363			return 0;
 364
 365		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
 366		if (error || !update_sb)
 367			return error;
 368
 369		sbp->sb_unit = mp->m_dalign;
 370		sbp->sb_width = mp->m_swidth;
 371		mp->m_update_sb = true;
 372	} else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
 373		mp->m_dalign = sbp->sb_unit;
 374		mp->m_swidth = sbp->sb_width;
 375	}
 376
 377	return 0;
 378}
 379
 380/*
 381 * precalculate the low space thresholds for dynamic speculative preallocation.
 382 */
 383void
 384xfs_set_low_space_thresholds(
 385	struct xfs_mount	*mp)
 386{
 387	uint64_t		dblocks = mp->m_sb.sb_dblocks;
 388	uint64_t		rtexts = mp->m_sb.sb_rextents;
 389	int			i;
 390
 391	do_div(dblocks, 100);
 392	do_div(rtexts, 100);
 393
 394	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 395		mp->m_low_space[i] = dblocks * (i + 1);
 396		mp->m_low_rtexts[i] = rtexts * (i + 1);
 397	}
 398}
 399
 400/*
 401 * Check that the data (and log if separate) is an ok size.
 402 */
 403STATIC int
 404xfs_check_sizes(
 405	struct xfs_mount *mp)
 406{
 407	struct xfs_buf	*bp;
 408	xfs_daddr_t	d;
 409	int		error;
 410
 411	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 412	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 413		xfs_warn(mp, "filesystem size mismatch detected");
 414		return -EFBIG;
 415	}
 416	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 417					d - XFS_FSS_TO_BB(mp, 1),
 418					XFS_FSS_TO_BB(mp, 1), &bp, NULL);
 419	if (error) {
 420		xfs_warn(mp, "last sector read failed");
 421		return error;
 422	}
 423	xfs_buf_relse(bp);
 424
 425	if (mp->m_logdev_targp == mp->m_ddev_targp)
 426		return 0;
 427
 428	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 429	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 430		xfs_warn(mp, "log size mismatch detected");
 431		return -EFBIG;
 432	}
 433	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 434					d - XFS_FSB_TO_BB(mp, 1),
 435					XFS_FSB_TO_BB(mp, 1), &bp, NULL);
 436	if (error) {
 437		xfs_warn(mp, "log device read failed");
 438		return error;
 439	}
 440	xfs_buf_relse(bp);
 441	return 0;
 442}
 443
 444/*
 445 * Clear the quotaflags in memory and in the superblock.
 446 */
 447int
 448xfs_mount_reset_sbqflags(
 449	struct xfs_mount	*mp)
 450{
 451	mp->m_qflags = 0;
 452
 453	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 454	if (mp->m_sb.sb_qflags == 0)
 455		return 0;
 456	spin_lock(&mp->m_sb_lock);
 457	mp->m_sb.sb_qflags = 0;
 458	spin_unlock(&mp->m_sb_lock);
 459
 460	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 461		return 0;
 462
 463	return xfs_sync_sb(mp, false);
 464}
 465
 466static const char *const xfs_free_pool_name[] = {
 467	[XC_FREE_BLOCKS]	= "free blocks",
 468	[XC_FREE_RTEXTENTS]	= "free rt extents",
 469	[XC_FREE_RTAVAILABLE]	= "available rt extents",
 470};
 471
 472uint64_t
 473xfs_default_resblks(
 474	struct xfs_mount	*mp,
 475	enum xfs_free_counter	ctr)
 476{
 477	switch (ctr) {
 478	case XC_FREE_BLOCKS:
 479		/*
 480		 * Default to 5% or 8192 FSBs of space reserved, whichever is
 481		 * smaller.
 482		 *
 483		 * This is intended to cover concurrent allocation transactions
 484		 * when we initially hit ENOSPC.  These each require a 4 block
 485		 * reservation. Hence by default we cover roughly 2000
 486		 * concurrent allocation reservations.
 487		 */
 488		return min(div_u64(mp->m_sb.sb_dblocks, 20), 8192ULL);
 489	case XC_FREE_RTEXTENTS:
 490	case XC_FREE_RTAVAILABLE:
 491		if (IS_ENABLED(CONFIG_XFS_RT) && xfs_has_zoned(mp))
 492			return xfs_zoned_default_resblks(mp, ctr);
 493		return 0;
 494	default:
 495		ASSERT(0);
 496		return 0;
 497	}
 498}
 499
 500/* Ensure the summary counts are correct. */
 501STATIC int
 502xfs_check_summary_counts(
 503	struct xfs_mount	*mp)
 504{
 505	int			error = 0;
 506
 507	/*
 508	 * The AG0 superblock verifier rejects in-progress filesystems,
 509	 * so we should never see the flag set this far into mounting.
 510	 */
 511	if (mp->m_sb.sb_inprogress) {
 512		xfs_err(mp, "sb_inprogress set after log recovery??");
 513		WARN_ON(1);
 514		return -EFSCORRUPTED;
 515	}
 516
 517	/*
 518	 * Now the log is mounted, we know if it was an unclean shutdown or
 519	 * not. If it was, with the first phase of recovery has completed, we
 520	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 521	 * but they are recovered transactionally in the second recovery phase
 522	 * later.
 523	 *
 524	 * If the log was clean when we mounted, we can check the summary
 525	 * counters.  If any of them are obviously incorrect, we can recompute
 526	 * them from the AGF headers in the next step.
 527	 */
 528	if (xfs_is_clean(mp) &&
 529	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
 530	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
 531	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
 532		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
 533
 534	/*
 535	 * We can safely re-initialise incore superblock counters from the
 536	 * per-ag data. These may not be correct if the filesystem was not
 537	 * cleanly unmounted, so we waited for recovery to finish before doing
 538	 * this.
 539	 *
 540	 * If the filesystem was cleanly unmounted or the previous check did
 541	 * not flag anything weird, then we can trust the values in the
 542	 * superblock to be correct and we don't need to do anything here.
 543	 * Otherwise, recalculate the summary counters.
 544	 */
 545	if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
 546	    xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
 547		error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
 548		if (error)
 549			return error;
 550	}
 551
 552	/*
 553	 * Older kernels misused sb_frextents to reflect both incore
 554	 * reservations made by running transactions and the actual count of
 555	 * free rt extents in the ondisk metadata.  Transactions committed
 556	 * during runtime can therefore contain a superblock update that
 557	 * undercounts the number of free rt extents tracked in the rt bitmap.
 558	 * A clean unmount record will have the correct frextents value since
 559	 * there can be no other transactions running at that point.
 560	 *
 561	 * If we're mounting the rt volume after recovering the log, recompute
 562	 * frextents from the rtbitmap file to fix the inconsistency.
 563	 */
 564	if (xfs_has_realtime(mp) && !xfs_has_zoned(mp) && !xfs_is_clean(mp)) {
 565		error = xfs_rtalloc_reinit_frextents(mp);
 566		if (error)
 567			return error;
 568	}
 569
 570	return 0;
 571}
 572
 573static void
 574xfs_unmount_check(
 575	struct xfs_mount	*mp)
 576{
 577	if (xfs_is_shutdown(mp))
 578		return;
 579
 580	if (percpu_counter_sum(&mp->m_ifree) >
 581			percpu_counter_sum(&mp->m_icount)) {
 582		xfs_alert(mp, "ifree/icount mismatch at unmount");
 583		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
 584	}
 585}
 586
 587/*
 588 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
 589 * internal inode structures can be sitting in the CIL and AIL at this point,
 590 * so we need to unpin them, write them back and/or reclaim them before unmount
 591 * can proceed.  In other words, callers are required to have inactivated all
 592 * inodes.
 593 *
 594 * An inode cluster that has been freed can have its buffer still pinned in
 595 * memory because the transaction is still sitting in a iclog. The stale inodes
 596 * on that buffer will be pinned to the buffer until the transaction hits the
 597 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
 598 * may never see the pinned buffer, so nothing will push out the iclog and
 599 * unpin the buffer.
 600 *
 601 * Hence we need to force the log to unpin everything first. However, log
 602 * forces don't wait for the discards they issue to complete, so we have to
 603 * explicitly wait for them to complete here as well.
 604 *
 605 * Then we can tell the world we are unmounting so that error handling knows
 606 * that the filesystem is going away and we should error out anything that we
 607 * have been retrying in the background.  This will prevent never-ending
 608 * retries in AIL pushing from hanging the unmount.
 609 *
 610 * Stop inodegc and background reclaim before pushing the AIL so that they
 611 * are not running while the AIL is being flushed. Then push the AIL to
 612 * clean all the remaining dirty objects and reclaim the remaining inodes.
 613 */
 614static void
 615xfs_unmount_flush_inodes(
 616	struct xfs_mount	*mp)
 617{
 618	xfs_log_force(mp, XFS_LOG_SYNC);
 619	xfs_extent_busy_wait_all(mp);
 620	flush_workqueue(xfs_discard_wq);
 621
 622	xfs_set_unmounting(mp);
 623
 624	xfs_inodegc_stop(mp);
 625	cancel_delayed_work_sync(&mp->m_reclaim_work);
 626	xfs_ail_push_all_sync(mp->m_ail);
 627	xfs_reclaim_inodes(mp);
 628	xfs_health_unmount(mp);
 629}
 630
 631static void
 632xfs_mount_setup_inode_geom(
 633	struct xfs_mount	*mp)
 634{
 635	struct xfs_ino_geometry *igeo = M_IGEO(mp);
 636
 637	igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
 638	ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
 639
 640	xfs_ialloc_setup_geometry(mp);
 641}
 642
 643/* Mount the metadata directory tree root. */
 644STATIC int
 645xfs_mount_setup_metadir(
 646	struct xfs_mount	*mp)
 647{
 648	int			error;
 649
 650	/* Load the metadata directory root inode into memory. */
 651	error = xfs_metafile_iget(mp, mp->m_sb.sb_metadirino, XFS_METAFILE_DIR,
 652			&mp->m_metadirip);
 653	if (error)
 654		xfs_warn(mp, "Failed to load metadir root directory, error %d",
 655				error);
 656	return error;
 657}
 658
 659/* Compute maximum possible height for per-AG btree types for this fs. */
 660static inline void
 661xfs_agbtree_compute_maxlevels(
 662	struct xfs_mount	*mp)
 663{
 664	unsigned int		levels;
 665
 666	levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
 667	levels = max(levels, mp->m_rmap_maxlevels);
 668	mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
 669}
 670
 671/* Maximum atomic write IO size that the kernel allows. */
 672static inline xfs_extlen_t xfs_calc_atomic_write_max(struct xfs_mount *mp)
 673{
 674	return rounddown_pow_of_two(XFS_B_TO_FSB(mp, MAX_RW_COUNT));
 675}
 676
 677/*
 678 * If the underlying device advertises atomic write support, limit the size of
 679 * atomic writes to the greatest power-of-two factor of the group size so
 680 * that every atomic write unit aligns with the start of every group.  This is
 681 * required so that the allocations for an atomic write will always be
 682 * aligned compatibly with the alignment requirements of the storage.
 683 *
 684 * If the device doesn't advertise atomic writes, then there are no alignment
 685 * restrictions and the largest out-of-place write we can do ourselves is the
 686 * number of blocks that user files can allocate from any group.
 687 */
 688static xfs_extlen_t
 689xfs_calc_group_awu_max(
 690	struct xfs_mount	*mp,
 691	enum xfs_group_type	type)
 692{
 693	struct xfs_groups	*g = &mp->m_groups[type];
 694	struct xfs_buftarg	*btp = xfs_group_type_buftarg(mp, type);
 695
 696	if (g->blocks == 0)
 697		return 0;
 698	if (btp && btp->bt_awu_min > 0)
 699		return max_pow_of_two_factor(g->blocks);
 700	return rounddown_pow_of_two(g->blocks);
 701}
 702
 703/* Compute the maximum atomic write unit size for each section. */
 704static inline void
 705xfs_calc_atomic_write_unit_max(
 706	struct xfs_mount	*mp,
 707	enum xfs_group_type	type)
 708{
 709	struct xfs_groups	*g = &mp->m_groups[type];
 710
 711	const xfs_extlen_t	max_write = xfs_calc_atomic_write_max(mp);
 712	const xfs_extlen_t	max_ioend = xfs_reflink_max_atomic_cow(mp);
 713	const xfs_extlen_t	max_gsize = xfs_calc_group_awu_max(mp, type);
 714
 715	g->awu_max = min3(max_write, max_ioend, max_gsize);
 716	trace_xfs_calc_atomic_write_unit_max(mp, type, max_write, max_ioend,
 717			max_gsize, g->awu_max);
 718}
 719
 720/*
 721 * Try to set the atomic write maximum to a new value that we got from
 722 * userspace via mount option.
 723 */
 724int
 725xfs_set_max_atomic_write_opt(
 726	struct xfs_mount	*mp,
 727	unsigned long long	new_max_bytes)
 728{
 729	const xfs_filblks_t	new_max_fsbs = XFS_B_TO_FSBT(mp, new_max_bytes);
 730	const xfs_extlen_t	max_write = xfs_calc_atomic_write_max(mp);
 731	const xfs_extlen_t	max_group =
 732		max(mp->m_groups[XG_TYPE_AG].blocks,
 733		    mp->m_groups[XG_TYPE_RTG].blocks);
 734	const xfs_extlen_t	max_group_write =
 735		max(xfs_calc_group_awu_max(mp, XG_TYPE_AG),
 736		    xfs_calc_group_awu_max(mp, XG_TYPE_RTG));
 737	int			error;
 738
 739	if (new_max_bytes == 0)
 740		goto set_limit;
 741
 742	ASSERT(max_write <= U32_MAX);
 743
 744	/* generic_atomic_write_valid enforces power of two length */
 745	if (!is_power_of_2(new_max_bytes)) {
 746		xfs_warn(mp,
 747 "max atomic write size of %llu bytes is not a power of 2",
 748				new_max_bytes);
 749		return -EINVAL;
 750	}
 751
 752	if (new_max_bytes & mp->m_blockmask) {
 753		xfs_warn(mp,
 754 "max atomic write size of %llu bytes not aligned with fsblock",
 755				new_max_bytes);
 756		return -EINVAL;
 757	}
 758
 759	if (new_max_fsbs > max_write) {
 760		xfs_warn(mp,
 761 "max atomic write size of %lluk cannot be larger than max write size %lluk",
 762				new_max_bytes >> 10,
 763				XFS_FSB_TO_B(mp, max_write) >> 10);
 764		return -EINVAL;
 765	}
 766
 767	if (new_max_fsbs > max_group) {
 768		xfs_warn(mp,
 769 "max atomic write size of %lluk cannot be larger than allocation group size %lluk",
 770				new_max_bytes >> 10,
 771				XFS_FSB_TO_B(mp, max_group) >> 10);
 772		return -EINVAL;
 773	}
 774
 775	if (new_max_fsbs > max_group_write) {
 776		xfs_warn(mp,
 777 "max atomic write size of %lluk cannot be larger than max allocation group write size %lluk",
 778				new_max_bytes >> 10,
 779				XFS_FSB_TO_B(mp, max_group_write) >> 10);
 780		return -EINVAL;
 781	}
 782
 783	if (xfs_has_reflink(mp))
 784		goto set_limit;
 785
 786	if (new_max_fsbs == 1) {
 787		if (mp->m_ddev_targp->bt_awu_max ||
 788		    (mp->m_rtdev_targp && mp->m_rtdev_targp->bt_awu_max)) {
 789		} else {
 790			xfs_warn(mp,
 791 "cannot support atomic writes of size %lluk with no reflink or HW support",
 792				new_max_bytes >> 10);
 793			return -EINVAL;
 794		}
 795	} else {
 796		xfs_warn(mp,
 797 "cannot support atomic writes of size %lluk with no reflink support",
 798				new_max_bytes >> 10);
 799		return -EINVAL;
 800	}
 801
 802set_limit:
 803	error = xfs_calc_atomic_write_reservation(mp, new_max_fsbs);
 804	if (error) {
 805		xfs_warn(mp,
 806 "cannot support completing atomic writes of %lluk",
 807				new_max_bytes >> 10);
 808		return error;
 809	}
 810
 811	xfs_calc_atomic_write_unit_max(mp, XG_TYPE_AG);
 812	xfs_calc_atomic_write_unit_max(mp, XG_TYPE_RTG);
 813	mp->m_awu_max_bytes = new_max_bytes;
 814	return 0;
 815}
 816
 817/* Compute maximum possible height for realtime btree types for this fs. */
 818static inline void
 819xfs_rtbtree_compute_maxlevels(
 820	struct xfs_mount	*mp)
 821{
 822	mp->m_rtbtree_maxlevels = max(mp->m_rtrmap_maxlevels,
 823				      mp->m_rtrefc_maxlevels);
 824}
 825
 826/*
 827 * This function does the following on an initial mount of a file system:
 828 *	- reads the superblock from disk and init the mount struct
 829 *	- if we're a 32-bit kernel, do a size check on the superblock
 830 *		so we don't mount terabyte filesystems
 831 *	- init mount struct realtime fields
 832 *	- allocate inode hash table for fs
 833 *	- init directory manager
 834 *	- perform recovery and init the log manager
 835 */
 836int
 837xfs_mountfs(
 838	struct xfs_mount	*mp)
 839{
 840	struct xfs_sb		*sbp = &(mp->m_sb);
 841	struct xfs_inode	*rip;
 842	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
 843	uint			quotamount = 0;
 844	uint			quotaflags = 0;
 845	int			error = 0;
 846	int			i;
 847
 848	xfs_sb_mount_common(mp, sbp);
 849
 850	/*
 851	 * Check for a mismatched features2 values.  Older kernels read & wrote
 852	 * into the wrong sb offset for sb_features2 on some platforms due to
 853	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 854	 * which made older superblock reading/writing routines swap it as a
 855	 * 64-bit value.
 856	 *
 857	 * For backwards compatibility, we make both slots equal.
 858	 *
 859	 * If we detect a mismatched field, we OR the set bits into the existing
 860	 * features2 field in case it has already been modified; we don't want
 861	 * to lose any features.  We then update the bad location with the ORed
 862	 * value so that older kernels will see any features2 flags. The
 863	 * superblock writeback code ensures the new sb_features2 is copied to
 864	 * sb_bad_features2 before it is logged or written to disk.
 865	 */
 866	if (xfs_sb_has_mismatched_features2(sbp)) {
 867		xfs_warn(mp, "correcting sb_features alignment problem");
 868		sbp->sb_features2 |= sbp->sb_bad_features2;
 869		mp->m_update_sb = true;
 870	}
 871
 872
 873	/* always use v2 inodes by default now */
 874	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 875		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 876		mp->m_features |= XFS_FEAT_NLINK;
 877		mp->m_update_sb = true;
 878	}
 879
 880	/*
 881	 * If we were given new sunit/swidth options, do some basic validation
 882	 * checks and convert the incore dalign and swidth values to the
 883	 * same units (FSB) that everything else uses.  This /must/ happen
 884	 * before computing the inode geometry.
 885	 */
 886	error = xfs_validate_new_dalign(mp);
 887	if (error)
 888		goto out;
 889
 890	xfs_alloc_compute_maxlevels(mp);
 891	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 892	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 893	xfs_mount_setup_inode_geom(mp);
 894	xfs_rmapbt_compute_maxlevels(mp);
 895	xfs_rtrmapbt_compute_maxlevels(mp);
 896	xfs_refcountbt_compute_maxlevels(mp);
 897	xfs_rtrefcountbt_compute_maxlevels(mp);
 898
 899	xfs_agbtree_compute_maxlevels(mp);
 900	xfs_rtbtree_compute_maxlevels(mp);
 901
 902	/*
 903	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
 904	 * is NOT aligned turn off m_dalign since allocator alignment is within
 905	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
 906	 * we must compute the free space and rmap btree geometry before doing
 907	 * this.
 908	 */
 909	error = xfs_update_alignment(mp);
 910	if (error)
 911		goto out;
 912
 913	/* enable fail_at_unmount as default */
 914	mp->m_fail_unmount = true;
 915
 916	error = xfs_mount_sysfs_init(mp);
 917	if (error)
 918		goto out_remove_scrub_stats;
 919
 920	xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
 921
 922	error = xfs_errortag_init(mp);
 923	if (error)
 924		goto out_remove_sysfs;
 925
 926	error = xfs_uuid_mount(mp);
 927	if (error)
 928		goto out_remove_errortag;
 929
 930	/*
 931	 * Update the preferred write size based on the information from the
 932	 * on-disk superblock.
 933	 */
 934	mp->m_allocsize_log =
 935		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
 936	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
 937
 938	/* set the low space thresholds for dynamic preallocation */
 939	xfs_set_low_space_thresholds(mp);
 940
 941	/*
 942	 * If enabled, sparse inode chunk alignment is expected to match the
 943	 * cluster size. Full inode chunk alignment must match the chunk size,
 944	 * but that is checked on sb read verification...
 945	 */
 946	if (xfs_has_sparseinodes(mp) &&
 947	    mp->m_sb.sb_spino_align !=
 948			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
 949		xfs_warn(mp,
 950	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 951			 mp->m_sb.sb_spino_align,
 952			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
 953		error = -EINVAL;
 954		goto out_remove_uuid;
 955	}
 956
 957	/*
 958	 * Check that the data (and log if separate) is an ok size.
 959	 */
 960	error = xfs_check_sizes(mp);
 961	if (error)
 962		goto out_remove_uuid;
 963
 964	/*
 965	 * Initialize realtime fields in the mount structure
 966	 */
 967	error = xfs_rtmount_init(mp);
 968	if (error) {
 969		xfs_warn(mp, "RT mount failed");
 970		goto out_remove_uuid;
 971	}
 972
 973	/*
 974	 *  Copies the low order bits of the timestamp and the randomly
 975	 *  set "sequence" number out of a UUID.
 976	 */
 977	mp->m_fixedfsid[0] =
 978		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
 979		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
 980	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
 981
 982	error = xfs_da_mount(mp);
 983	if (error) {
 984		xfs_warn(mp, "Failed dir/attr init: %d", error);
 985		goto out_remove_uuid;
 986	}
 987
 988	/*
 989	 * Initialize the precomputed transaction reservations values.
 990	 */
 991	xfs_trans_init(mp);
 992
 993	/*
 994	 * Allocate and initialize the per-ag data.
 995	 */
 996	error = xfs_initialize_perag(mp, 0, sbp->sb_agcount,
 997			mp->m_sb.sb_dblocks, &mp->m_maxagi);
 998	if (error) {
 999		xfs_warn(mp, "Failed per-ag init: %d", error);
1000		goto out_free_dir;
1001	}
1002
1003	error = xfs_initialize_rtgroups(mp, 0, sbp->sb_rgcount,
1004			mp->m_sb.sb_rextents);
1005	if (error) {
1006		xfs_warn(mp, "Failed rtgroup init: %d", error);
1007		goto out_free_perag;
1008	}
1009
1010	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
1011		xfs_warn(mp, "no log defined");
1012		error = -EFSCORRUPTED;
1013		goto out_free_rtgroup;
1014	}
1015
1016	error = xfs_inodegc_register_shrinker(mp);
1017	if (error)
1018		goto out_fail_wait;
1019
1020	/*
1021	 * If we're resuming quota status, pick up the preliminary qflags from
1022	 * the ondisk superblock so that we know if we should recover dquots.
1023	 */
1024	if (xfs_is_resuming_quotaon(mp))
1025		xfs_qm_resume_quotaon(mp);
1026
1027	/*
1028	 * Log's mount-time initialization. The first part of recovery can place
1029	 * some items on the AIL, to be handled when recovery is finished or
1030	 * cancelled.
1031	 */
1032	error = xfs_log_mount(mp, mp->m_logdev_targp,
1033			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1034			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1035	if (error) {
1036		xfs_warn(mp, "log mount failed");
1037		goto out_inodegc_shrinker;
1038	}
1039
1040	/*
1041	 * If we're resuming quota status and recovered the log, re-sample the
1042	 * qflags from the ondisk superblock now that we've recovered it, just
1043	 * in case someone shut down enforcement just before a crash.
1044	 */
1045	if (xfs_clear_resuming_quotaon(mp) && xlog_recovery_needed(mp->m_log))
1046		xfs_qm_resume_quotaon(mp);
1047
1048	/*
1049	 * If logged xattrs are still enabled after log recovery finishes, then
1050	 * they'll be available until unmount.  Otherwise, turn them off.
1051	 */
1052	if (xfs_sb_version_haslogxattrs(&mp->m_sb))
1053		xfs_set_using_logged_xattrs(mp);
1054	else
1055		xfs_clear_using_logged_xattrs(mp);
1056
1057	/* Enable background inode inactivation workers. */
1058	xfs_inodegc_start(mp);
1059	xfs_blockgc_start(mp);
1060
1061	if (xfs_has_metadir(mp)) {
1062		error = xfs_mount_setup_metadir(mp);
1063		if (error)
1064			goto out_free_metadir;
1065	}
1066
1067	/*
1068	 * Get and sanity-check the root inode.
1069	 * Save the pointer to it in the mount structure.
1070	 */
1071	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
1072			 XFS_ILOCK_EXCL, &rip);
1073	if (error) {
1074		xfs_warn(mp,
1075			"Failed to read root inode 0x%llx, error %d",
1076			sbp->sb_rootino, -error);
1077		goto out_free_metadir;
1078	}
1079
1080	ASSERT(rip != NULL);
1081
1082	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
1083		xfs_warn(mp, "corrupted root inode %llu: not a directory",
1084			(unsigned long long)rip->i_ino);
1085		xfs_iunlock(rip, XFS_ILOCK_EXCL);
1086		error = -EFSCORRUPTED;
1087		goto out_rele_rip;
1088	}
1089	mp->m_rootip = rip;	/* save it */
1090
1091	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1092
1093	/*
1094	 * Initialize realtime inode pointers in the mount structure
1095	 */
1096	error = xfs_rtmount_inodes(mp);
1097	if (error) {
1098		/*
1099		 * Free up the root inode.
1100		 */
1101		xfs_warn(mp, "failed to read RT inodes");
1102		goto out_rele_rip;
1103	}
1104
1105	/* Make sure the summary counts are ok. */
1106	error = xfs_check_summary_counts(mp);
1107	if (error)
1108		goto out_rtunmount;
1109
1110	/*
1111	 * If this is a read-only mount defer the superblock updates until
1112	 * the next remount into writeable mode.  Otherwise we would never
1113	 * perform the update e.g. for the root filesystem.
1114	 */
1115	if (mp->m_update_sb && !xfs_is_readonly(mp)) {
1116		error = xfs_sync_sb(mp, false);
1117		if (error) {
1118			xfs_warn(mp, "failed to write sb changes");
1119			goto out_rtunmount;
1120		}
1121	}
1122
1123	/*
1124	 * Initialise the XFS quota management subsystem for this mount
1125	 */
1126	if (XFS_IS_QUOTA_ON(mp)) {
1127		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1128		if (error)
1129			goto out_rtunmount;
1130	} else {
1131		/*
1132		 * If a file system had quotas running earlier, but decided to
1133		 * mount without -o uquota/pquota/gquota options, revoke the
1134		 * quotachecked license.
1135		 */
1136		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1137			xfs_notice(mp, "resetting quota flags");
1138			error = xfs_mount_reset_sbqflags(mp);
1139			if (error)
1140				goto out_rtunmount;
1141		}
1142	}
1143
1144	/*
1145	 * Finish recovering the file system.  This part needed to be delayed
1146	 * until after the root and real-time bitmap inodes were consistently
1147	 * read in.  Temporarily create per-AG space reservations for metadata
1148	 * btree shape changes because space freeing transactions (for inode
1149	 * inactivation) require the per-AG reservation in lieu of reserving
1150	 * blocks.
1151	 */
1152	error = xfs_fs_reserve_ag_blocks(mp);
1153	if (error && error == -ENOSPC)
1154		xfs_warn(mp,
1155	"ENOSPC reserving per-AG metadata pool, log recovery may fail.");
1156	error = xfs_log_mount_finish(mp);
1157	xfs_fs_unreserve_ag_blocks(mp);
1158	if (error) {
1159		xfs_warn(mp, "log mount finish failed");
1160		goto out_rtunmount;
1161	}
1162
1163	/*
1164	 * Now the log is fully replayed, we can transition to full read-only
1165	 * mode for read-only mounts. This will sync all the metadata and clean
1166	 * the log so that the recovery we just performed does not have to be
1167	 * replayed again on the next mount.
1168	 *
1169	 * We use the same quiesce mechanism as the rw->ro remount, as they are
1170	 * semantically identical operations.
1171	 */
1172	if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
1173		xfs_log_clean(mp);
1174
1175	if (xfs_has_zoned(mp)) {
1176		error = xfs_mount_zones(mp);
1177		if (error)
1178			goto out_rtunmount;
1179	}
1180
1181	/*
1182	 * Complete the quota initialisation, post-log-replay component.
1183	 */
1184	if (quotamount) {
1185		ASSERT(mp->m_qflags == 0);
1186		mp->m_qflags = quotaflags;
1187
1188		xfs_qm_mount_quotas(mp);
1189	}
1190
1191	/*
1192	 * Now we are mounted, reserve a small amount of unused space for
1193	 * privileged transactions. This is needed so that transaction
1194	 * space required for critical operations can dip into this pool
1195	 * when at ENOSPC. This is needed for operations like create with
1196	 * attr, unwritten extent conversion at ENOSPC, garbage collection
1197	 * etc. Data allocations are not allowed to use this reserved space.
1198	 *
1199	 * This may drive us straight to ENOSPC on mount, but that implies
1200	 * we were already there on the last unmount. Warn if this occurs.
1201	 */
1202	if (!xfs_is_readonly(mp)) {
1203		for (i = 0; i < XC_FREE_NR; i++) {
1204			error = xfs_reserve_blocks(mp, i,
1205					xfs_default_resblks(mp, i));
1206			if (error)
1207				xfs_warn(mp,
1208"Unable to allocate reserve blocks. Continuing without reserve pool for %s.",
1209					xfs_free_pool_name[i]);
1210		}
1211
1212		/* Reserve AG blocks for future btree expansion. */
1213		error = xfs_fs_reserve_ag_blocks(mp);
1214		if (error && error != -ENOSPC)
1215			goto out_agresv;
1216
1217		xfs_zone_gc_start(mp);
1218	}
1219
1220	/*
1221	 * Pre-calculate atomic write unit max.  This involves computations
1222	 * derived from transaction reservations, so we must do this after the
1223	 * log is fully initialized.
1224	 */
1225	error = xfs_set_max_atomic_write_opt(mp, mp->m_awu_max_bytes);
1226	if (error)
1227		goto out_agresv;
1228
1229	return 0;
1230
1231 out_agresv:
1232	xfs_fs_unreserve_ag_blocks(mp);
1233	xfs_qm_unmount_quotas(mp);
1234	if (xfs_has_zoned(mp))
1235		xfs_unmount_zones(mp);
1236 out_rtunmount:
1237	xfs_rtunmount_inodes(mp);
1238 out_rele_rip:
1239	xfs_irele(rip);
1240	/* Clean out dquots that might be in memory after quotacheck. */
1241	xfs_qm_unmount(mp);
1242 out_free_metadir:
1243	if (mp->m_metadirip)
1244		xfs_irele(mp->m_metadirip);
1245
1246	/*
1247	 * Inactivate all inodes that might still be in memory after a log
1248	 * intent recovery failure so that reclaim can free them.  Metadata
1249	 * inodes and the root directory shouldn't need inactivation, but the
1250	 * mount failed for some reason, so pull down all the state and flee.
1251	 */
1252	xfs_inodegc_flush(mp);
1253
1254	/*
1255	 * Flush all inode reclamation work and flush the log.
1256	 * We have to do this /after/ rtunmount and qm_unmount because those
1257	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1258	 *
1259	 * This is slightly different from the unmountfs call sequence
1260	 * because we could be tearing down a partially set up mount.  In
1261	 * particular, if log_mount_finish fails we bail out without calling
1262	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1263	 * quota inodes.
1264	 */
1265	xfs_unmount_flush_inodes(mp);
1266	xfs_log_mount_cancel(mp);
1267 out_inodegc_shrinker:
1268	shrinker_free(mp->m_inodegc_shrinker);
1269 out_fail_wait:
1270	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1271		xfs_buftarg_drain(mp->m_logdev_targp);
1272	xfs_buftarg_drain(mp->m_ddev_targp);
1273 out_free_rtgroup:
1274	xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1275 out_free_perag:
1276	xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1277 out_free_dir:
1278	xfs_da_unmount(mp);
1279 out_remove_uuid:
1280	xfs_uuid_unmount(mp);
1281 out_remove_errortag:
1282	xfs_errortag_del(mp);
1283 out_remove_sysfs:
1284	xfs_mount_sysfs_del(mp);
1285 out_remove_scrub_stats:
1286	xchk_stats_unregister(mp->m_scrub_stats);
1287 out:
1288	return error;
1289}
1290
1291/*
1292 * This flushes out the inodes,dquots and the superblock, unmounts the
1293 * log and makes sure that incore structures are freed.
1294 */
1295void
1296xfs_unmountfs(
1297	struct xfs_mount	*mp)
1298{
1299	int			error;
1300
1301	/*
1302	 * Perform all on-disk metadata updates required to inactivate inodes
1303	 * that the VFS evicted earlier in the unmount process.  Freeing inodes
1304	 * and discarding CoW fork preallocations can cause shape changes to
1305	 * the free inode and refcount btrees, respectively, so we must finish
1306	 * this before we discard the metadata space reservations.  Metadata
1307	 * inodes and the root directory do not require inactivation.
1308	 */
1309	xfs_inodegc_flush(mp);
1310
1311	xfs_blockgc_stop(mp);
1312	if (!test_bit(XFS_OPSTATE_READONLY, &mp->m_opstate))
1313		xfs_zone_gc_stop(mp);
1314	xfs_fs_unreserve_ag_blocks(mp);
1315	xfs_qm_unmount_quotas(mp);
1316	if (xfs_has_zoned(mp))
1317		xfs_unmount_zones(mp);
1318	xfs_rtunmount_inodes(mp);
1319	xfs_irele(mp->m_rootip);
1320	if (mp->m_metadirip)
1321		xfs_irele(mp->m_metadirip);
1322
1323	xfs_unmount_flush_inodes(mp);
1324
1325	xfs_qm_unmount(mp);
1326
1327	/*
1328	 * Unreserve any blocks we have so that when we unmount we don't account
1329	 * the reserved free space as used. This is really only necessary for
1330	 * lazy superblock counting because it trusts the incore superblock
1331	 * counters to be absolutely correct on clean unmount.
1332	 *
1333	 * We don't bother correcting this elsewhere for lazy superblock
1334	 * counting because on mount of an unclean filesystem we reconstruct the
1335	 * correct counter value and this is irrelevant.
1336	 *
1337	 * For non-lazy counter filesystems, this doesn't matter at all because
1338	 * we only every apply deltas to the superblock and hence the incore
1339	 * value does not matter....
1340	 */
1341	error = xfs_reserve_blocks(mp, XC_FREE_BLOCKS, 0);
1342	if (error)
1343		xfs_warn(mp, "Unable to free reserved block pool. "
1344				"Freespace may not be correct on next mount.");
1345	xfs_unmount_check(mp);
1346
1347	/*
1348	 * Indicate that it's ok to clear log incompat bits before cleaning
1349	 * the log and writing the unmount record.
1350	 */
1351	xfs_set_done_with_log_incompat(mp);
1352	xfs_log_unmount(mp);
1353	xfs_da_unmount(mp);
1354	xfs_uuid_unmount(mp);
1355
1356#if defined(DEBUG)
1357	xfs_errortag_clearall(mp);
1358#endif
1359	shrinker_free(mp->m_inodegc_shrinker);
1360	xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1361	xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1362	xfs_errortag_del(mp);
1363	xchk_stats_unregister(mp->m_scrub_stats);
1364	xfs_mount_sysfs_del(mp);
1365}
1366
1367/*
1368 * Determine whether modifications can proceed. The caller specifies the minimum
1369 * freeze level for which modifications should not be allowed. This allows
1370 * certain operations to proceed while the freeze sequence is in progress, if
1371 * necessary.
1372 */
1373bool
1374xfs_fs_writable(
1375	struct xfs_mount	*mp,
1376	int			level)
1377{
1378	ASSERT(level > SB_UNFROZEN);
1379	if ((mp->m_super->s_writers.frozen >= level) ||
1380	    xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1381		return false;
1382
1383	return true;
1384}
1385
1386/*
1387 * Estimate the amount of free space that is not available to userspace and is
1388 * not explicitly reserved from the incore fdblocks.  This includes:
1389 *
1390 * - The minimum number of blocks needed to support splitting a bmap btree
1391 * - The blocks currently in use by the freespace btrees because they record
1392 *   the actual blocks that will fill per-AG metadata space reservations
1393 */
1394uint64_t
1395xfs_freecounter_unavailable(
1396	struct xfs_mount	*mp,
1397	enum xfs_free_counter	ctr)
1398{
1399	if (ctr != XC_FREE_BLOCKS)
1400		return 0;
1401	return mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks);
1402}
1403
1404void
1405xfs_add_freecounter(
1406	struct xfs_mount	*mp,
1407	enum xfs_free_counter	ctr,
1408	uint64_t		delta)
1409{
1410	struct xfs_freecounter	*counter = &mp->m_free[ctr];
1411	uint64_t		res_used;
1412
1413	/*
1414	 * If the reserve pool is depleted, put blocks back into it first.
1415	 * Most of the time the pool is full.
1416	 */
1417	if (likely(counter->res_avail == counter->res_total)) {
1418		percpu_counter_add(&counter->count, delta);
1419		return;
1420	}
1421
1422	spin_lock(&mp->m_sb_lock);
1423	res_used = counter->res_total - counter->res_avail;
1424	if (res_used > delta) {
1425		counter->res_avail += delta;
1426	} else {
1427		delta -= res_used;
1428		counter->res_avail = counter->res_total;
1429		percpu_counter_add(&counter->count, delta);
1430	}
1431	spin_unlock(&mp->m_sb_lock);
1432}
1433
1434
1435/* Adjust in-core free blocks or RT extents. */
1436int
1437xfs_dec_freecounter(
1438	struct xfs_mount	*mp,
1439	enum xfs_free_counter	ctr,
1440	uint64_t		delta,
1441	bool			rsvd)
1442{
1443	struct xfs_freecounter	*counter = &mp->m_free[ctr];
1444	s32			batch;
1445
1446	ASSERT(ctr < XC_FREE_NR);
1447
1448	/*
1449	 * Taking blocks away, need to be more accurate the closer we
1450	 * are to zero.
1451	 *
1452	 * If the counter has a value of less than 2 * max batch size,
1453	 * then make everything serialise as we are real close to
1454	 * ENOSPC.
1455	 */
1456	if (__percpu_counter_compare(&counter->count, 2 * XFS_FDBLOCKS_BATCH,
1457				     XFS_FDBLOCKS_BATCH) < 0)
1458		batch = 1;
1459	else
1460		batch = XFS_FDBLOCKS_BATCH;
1461
1462	/*
1463	 * Set aside allocbt blocks because these blocks are tracked as free
1464	 * space but not available for allocation. Technically this means that a
1465	 * single reservation cannot consume all remaining free space, but the
1466	 * ratio of allocbt blocks to usable free blocks should be rather small.
1467	 * The tradeoff without this is that filesystems that maintain high
1468	 * perag block reservations can over reserve physical block availability
1469	 * and fail physical allocation, which leads to much more serious
1470	 * problems (i.e. transaction abort, pagecache discards, etc.) than
1471	 * slightly premature -ENOSPC.
1472	 */
1473	percpu_counter_add_batch(&counter->count, -((int64_t)delta), batch);
1474	if (__percpu_counter_compare(&counter->count,
1475			xfs_freecounter_unavailable(mp, ctr),
1476			XFS_FDBLOCKS_BATCH) < 0) {
1477		/*
1478		 * Lock up the sb for dipping into reserves before releasing the
1479		 * space that took us to ENOSPC.
1480		 */
1481		spin_lock(&mp->m_sb_lock);
1482		percpu_counter_add(&counter->count, delta);
1483		if (!rsvd)
1484			goto fdblocks_enospc;
1485		if (delta > counter->res_avail) {
1486			if (ctr == XC_FREE_BLOCKS)
1487				xfs_warn_once(mp,
1488"Reserve blocks depleted! Consider increasing reserve pool size.");
1489			goto fdblocks_enospc;
1490		}
1491		counter->res_avail -= delta;
1492		trace_xfs_freecounter_reserved(mp, ctr, delta, _RET_IP_);
1493		spin_unlock(&mp->m_sb_lock);
1494	}
1495
1496	/* we had space! */
1497	return 0;
1498
1499fdblocks_enospc:
1500	trace_xfs_freecounter_enospc(mp, ctr, delta, _RET_IP_);
1501	spin_unlock(&mp->m_sb_lock);
1502	return -ENOSPC;
1503}
1504
1505/*
1506 * Used to free the superblock along various error paths.
1507 */
1508void
1509xfs_freesb(
1510	struct xfs_mount	*mp)
1511{
1512	struct xfs_buf		*bp = mp->m_sb_bp;
1513
1514	xfs_buf_lock(bp);
1515	mp->m_sb_bp = NULL;
1516	xfs_buf_relse(bp);
1517}
1518
1519/*
1520 * If the underlying (data/log/rt) device is readonly, there are some
1521 * operations that cannot proceed.
1522 */
1523int
1524xfs_dev_is_read_only(
1525	struct xfs_mount	*mp,
1526	char			*message)
1527{
1528	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1529	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1530	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1531		xfs_notice(mp, "%s required on read-only device.", message);
1532		xfs_notice(mp, "write access unavailable, cannot proceed.");
1533		return -EROFS;
1534	}
1535	return 0;
1536}
1537
1538/* Force the summary counters to be recalculated at next mount. */
1539void
1540xfs_force_summary_recalc(
1541	struct xfs_mount	*mp)
1542{
1543	if (!xfs_has_lazysbcount(mp))
1544		return;
1545
1546	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1547}
1548
1549/*
1550 * Enable a log incompat feature flag in the primary superblock.  The caller
1551 * cannot have any other transactions in progress.
1552 */
1553int
1554xfs_add_incompat_log_feature(
1555	struct xfs_mount	*mp,
1556	uint32_t		feature)
1557{
1558	struct xfs_dsb		*dsb;
1559	int			error;
1560
1561	ASSERT(hweight32(feature) == 1);
1562	ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1563
1564	/*
1565	 * Force the log to disk and kick the background AIL thread to reduce
1566	 * the chances that the bwrite will stall waiting for the AIL to unpin
1567	 * the primary superblock buffer.  This isn't a data integrity
1568	 * operation, so we don't need a synchronous push.
1569	 */
1570	error = xfs_log_force(mp, XFS_LOG_SYNC);
1571	if (error)
1572		return error;
1573	xfs_ail_push_all(mp->m_ail);
1574
1575	/*
1576	 * Lock the primary superblock buffer to serialize all callers that
1577	 * are trying to set feature bits.
1578	 */
1579	xfs_buf_lock(mp->m_sb_bp);
1580	xfs_buf_hold(mp->m_sb_bp);
1581
1582	if (xfs_is_shutdown(mp)) {
1583		error = -EIO;
1584		goto rele;
1585	}
1586
1587	if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1588		goto rele;
1589
1590	/*
1591	 * Write the primary superblock to disk immediately, because we need
1592	 * the log_incompat bit to be set in the primary super now to protect
1593	 * the log items that we're going to commit later.
1594	 */
1595	dsb = mp->m_sb_bp->b_addr;
1596	xfs_sb_to_disk(dsb, &mp->m_sb);
1597	dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1598	error = xfs_bwrite(mp->m_sb_bp);
1599	if (error)
1600		goto shutdown;
1601
1602	/*
1603	 * Add the feature bits to the incore superblock before we unlock the
1604	 * buffer.
1605	 */
1606	xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1607	xfs_buf_relse(mp->m_sb_bp);
1608
1609	/* Log the superblock to disk. */
1610	return xfs_sync_sb(mp, false);
1611shutdown:
1612	xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1613rele:
1614	xfs_buf_relse(mp->m_sb_bp);
1615	return error;
1616}
1617
1618/*
1619 * Clear all the log incompat flags from the superblock.
1620 *
1621 * The caller cannot be in a transaction, must ensure that the log does not
1622 * contain any log items protected by any log incompat bit, and must ensure
1623 * that there are no other threads that depend on the state of the log incompat
1624 * feature flags in the primary super.
1625 *
1626 * Returns true if the superblock is dirty.
1627 */
1628bool
1629xfs_clear_incompat_log_features(
1630	struct xfs_mount	*mp)
1631{
1632	bool			ret = false;
1633
1634	if (!xfs_has_crc(mp) ||
1635	    !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1636				XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1637	    xfs_is_shutdown(mp) ||
1638	    !xfs_is_done_with_log_incompat(mp))
1639		return false;
1640
1641	/*
1642	 * Update the incore superblock.  We synchronize on the primary super
1643	 * buffer lock to be consistent with the add function, though at least
1644	 * in theory this shouldn't be necessary.
1645	 */
1646	xfs_buf_lock(mp->m_sb_bp);
1647	xfs_buf_hold(mp->m_sb_bp);
1648
1649	if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1650				XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1651		xfs_sb_remove_incompat_log_features(&mp->m_sb);
1652		ret = true;
1653	}
1654
1655	xfs_buf_relse(mp->m_sb_bp);
1656	return ret;
1657}
1658
1659/*
1660 * Update the in-core delayed block counter.
1661 *
1662 * We prefer to update the counter without having to take a spinlock for every
1663 * counter update (i.e. batching).  Each change to delayed allocation
1664 * reservations can change can easily exceed the default percpu counter
1665 * batching, so we use a larger batch factor here.
1666 *
1667 * Note that we don't currently have any callers requiring fast summation
1668 * (e.g. percpu_counter_read) so we can use a big batch value here.
1669 */
1670#define XFS_DELALLOC_BATCH	(4096)
1671void
1672xfs_mod_delalloc(
1673	struct xfs_inode	*ip,
1674	int64_t			data_delta,
1675	int64_t			ind_delta)
1676{
1677	struct xfs_mount	*mp = ip->i_mount;
1678
1679	if (XFS_IS_REALTIME_INODE(ip)) {
1680		percpu_counter_add_batch(&mp->m_delalloc_rtextents,
1681				xfs_blen_to_rtbxlen(mp, data_delta),
1682				XFS_DELALLOC_BATCH);
1683		if (!ind_delta)
1684			return;
1685		data_delta = 0;
1686	}
1687	percpu_counter_add_batch(&mp->m_delalloc_blks, data_delta + ind_delta,
1688			XFS_DELALLOC_BATCH);
1689}