Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v4.17
   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_defer.h"
  28#include "xfs_da_format.h"
  29#include "xfs_da_btree.h"
  30#include "xfs_inode.h"
  31#include "xfs_dir2.h"
  32#include "xfs_ialloc.h"
  33#include "xfs_alloc.h"
  34#include "xfs_rtalloc.h"
  35#include "xfs_bmap.h"
  36#include "xfs_trans.h"
  37#include "xfs_trans_priv.h"
  38#include "xfs_log.h"
  39#include "xfs_error.h"
  40#include "xfs_quota.h"
  41#include "xfs_fsops.h"
  42#include "xfs_trace.h"
  43#include "xfs_icache.h"
  44#include "xfs_sysfs.h"
  45#include "xfs_rmap_btree.h"
  46#include "xfs_refcount_btree.h"
  47#include "xfs_reflink.h"
  48#include "xfs_extent_busy.h"
  49
  50
  51static DEFINE_MUTEX(xfs_uuid_table_mutex);
  52static int xfs_uuid_table_size;
  53static uuid_t *xfs_uuid_table;
  54
  55void
  56xfs_uuid_table_free(void)
  57{
  58	if (xfs_uuid_table_size == 0)
  59		return;
  60	kmem_free(xfs_uuid_table);
  61	xfs_uuid_table = NULL;
  62	xfs_uuid_table_size = 0;
  63}
  64
  65/*
  66 * See if the UUID is unique among mounted XFS filesystems.
  67 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  68 */
  69STATIC int
  70xfs_uuid_mount(
  71	struct xfs_mount	*mp)
  72{
  73	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  74	int			hole, i;
  75
  76	/* Publish UUID in struct super_block */
  77	uuid_copy(&mp->m_super->s_uuid, uuid);
  78
  79	if (mp->m_flags & XFS_MOUNT_NOUUID)
  80		return 0;
  81
  82	if (uuid_is_null(uuid)) {
  83		xfs_warn(mp, "Filesystem has null UUID - can't mount");
  84		return -EINVAL;
  85	}
  86
  87	mutex_lock(&xfs_uuid_table_mutex);
  88	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  89		if (uuid_is_null(&xfs_uuid_table[i])) {
  90			hole = i;
  91			continue;
  92		}
  93		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  94			goto out_duplicate;
  95	}
  96
  97	if (hole < 0) {
  98		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
  99			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
 100			KM_SLEEP);
 101		hole = xfs_uuid_table_size++;
 102	}
 103	xfs_uuid_table[hole] = *uuid;
 104	mutex_unlock(&xfs_uuid_table_mutex);
 105
 106	return 0;
 107
 108 out_duplicate:
 109	mutex_unlock(&xfs_uuid_table_mutex);
 110	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
 111	return -EINVAL;
 112}
 113
 114STATIC void
 115xfs_uuid_unmount(
 116	struct xfs_mount	*mp)
 117{
 118	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 119	int			i;
 120
 121	if (mp->m_flags & XFS_MOUNT_NOUUID)
 122		return;
 123
 124	mutex_lock(&xfs_uuid_table_mutex);
 125	for (i = 0; i < xfs_uuid_table_size; i++) {
 126		if (uuid_is_null(&xfs_uuid_table[i]))
 127			continue;
 128		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 129			continue;
 130		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 131		break;
 132	}
 133	ASSERT(i < xfs_uuid_table_size);
 134	mutex_unlock(&xfs_uuid_table_mutex);
 135}
 136
 137
 138STATIC void
 139__xfs_free_perag(
 140	struct rcu_head	*head)
 141{
 142	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 143
 144	ASSERT(atomic_read(&pag->pag_ref) == 0);
 145	kmem_free(pag);
 146}
 147
 148/*
 149 * Free up the per-ag resources associated with the mount structure.
 150 */
 151STATIC void
 152xfs_free_perag(
 153	xfs_mount_t	*mp)
 154{
 155	xfs_agnumber_t	agno;
 156	struct xfs_perag *pag;
 157
 158	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
 159		spin_lock(&mp->m_perag_lock);
 160		pag = radix_tree_delete(&mp->m_perag_tree, agno);
 161		spin_unlock(&mp->m_perag_lock);
 162		ASSERT(pag);
 163		ASSERT(atomic_read(&pag->pag_ref) == 0);
 164		xfs_buf_hash_destroy(pag);
 165		mutex_destroy(&pag->pag_ici_reclaim_lock);
 166		call_rcu(&pag->rcu_head, __xfs_free_perag);
 167	}
 168}
 169
 170/*
 171 * Check size of device based on the (data/realtime) block count.
 172 * Note: this check is used by the growfs code as well as mount.
 173 */
 174int
 175xfs_sb_validate_fsb_count(
 176	xfs_sb_t	*sbp,
 177	uint64_t	nblocks)
 178{
 179	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 180	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 181
 182	/* Limited by ULONG_MAX of page cache index */
 183	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 184		return -EFBIG;
 185	return 0;
 186}
 187
 188int
 189xfs_initialize_perag(
 190	xfs_mount_t	*mp,
 191	xfs_agnumber_t	agcount,
 192	xfs_agnumber_t	*maxagi)
 193{
 194	xfs_agnumber_t	index;
 195	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
 196	xfs_perag_t	*pag;
 197	int		error = -ENOMEM;
 198
 199	/*
 200	 * Walk the current per-ag tree so we don't try to initialise AGs
 201	 * that already exist (growfs case). Allocate and insert all the
 202	 * AGs we don't find ready for initialisation.
 203	 */
 204	for (index = 0; index < agcount; index++) {
 205		pag = xfs_perag_get(mp, index);
 206		if (pag) {
 207			xfs_perag_put(pag);
 208			continue;
 209		}
 210
 211		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
 212		if (!pag)
 213			goto out_unwind_new_pags;
 214		pag->pag_agno = index;
 215		pag->pag_mount = mp;
 216		spin_lock_init(&pag->pag_ici_lock);
 217		mutex_init(&pag->pag_ici_reclaim_lock);
 218		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 219		if (xfs_buf_hash_init(pag))
 220			goto out_free_pag;
 221		init_waitqueue_head(&pag->pagb_wait);
 222
 223		if (radix_tree_preload(GFP_NOFS))
 224			goto out_hash_destroy;
 225
 226		spin_lock(&mp->m_perag_lock);
 227		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
 228			BUG();
 229			spin_unlock(&mp->m_perag_lock);
 230			radix_tree_preload_end();
 231			error = -EEXIST;
 232			goto out_hash_destroy;
 233		}
 234		spin_unlock(&mp->m_perag_lock);
 235		radix_tree_preload_end();
 236		/* first new pag is fully initialized */
 237		if (first_initialised == NULLAGNUMBER)
 238			first_initialised = index;
 239	}
 240
 241	index = xfs_set_inode_alloc(mp, agcount);
 242
 243	if (maxagi)
 244		*maxagi = index;
 245
 246	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
 247	return 0;
 248
 249out_hash_destroy:
 250	xfs_buf_hash_destroy(pag);
 251out_free_pag:
 252	mutex_destroy(&pag->pag_ici_reclaim_lock);
 253	kmem_free(pag);
 254out_unwind_new_pags:
 255	/* unwind any prior newly initialized pags */
 256	for (index = first_initialised; index < agcount; index++) {
 257		pag = radix_tree_delete(&mp->m_perag_tree, index);
 258		if (!pag)
 259			break;
 260		xfs_buf_hash_destroy(pag);
 261		mutex_destroy(&pag->pag_ici_reclaim_lock);
 262		kmem_free(pag);
 263	}
 264	return error;
 265}
 266
 267/*
 268 * xfs_readsb
 269 *
 270 * Does the initial read of the superblock.
 271 */
 272int
 273xfs_readsb(
 274	struct xfs_mount *mp,
 275	int		flags)
 276{
 277	unsigned int	sector_size;
 278	struct xfs_buf	*bp;
 279	struct xfs_sb	*sbp = &mp->m_sb;
 280	int		error;
 281	int		loud = !(flags & XFS_MFSI_QUIET);
 282	const struct xfs_buf_ops *buf_ops;
 283
 284	ASSERT(mp->m_sb_bp == NULL);
 285	ASSERT(mp->m_ddev_targp != NULL);
 286
 287	/*
 288	 * For the initial read, we must guess at the sector
 289	 * size based on the block device.  It's enough to
 290	 * get the sb_sectsize out of the superblock and
 291	 * then reread with the proper length.
 292	 * We don't verify it yet, because it may not be complete.
 293	 */
 294	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 295	buf_ops = NULL;
 296
 297	/*
 298	 * Allocate a (locked) buffer to hold the superblock. This will be kept
 299	 * around at all times to optimize access to the superblock. Therefore,
 300	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
 301	 * elevated.
 302	 */
 303reread:
 304	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 305				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
 306				      buf_ops);
 307	if (error) {
 308		if (loud)
 309			xfs_warn(mp, "SB validate failed with error %d.", error);
 310		/* bad CRC means corrupted metadata */
 311		if (error == -EFSBADCRC)
 312			error = -EFSCORRUPTED;
 313		return error;
 314	}
 315
 316	/*
 317	 * Initialize the mount structure from the superblock.
 318	 */
 319	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
 320
 321	/*
 322	 * If we haven't validated the superblock, do so now before we try
 323	 * to check the sector size and reread the superblock appropriately.
 324	 */
 325	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 326		if (loud)
 327			xfs_warn(mp, "Invalid superblock magic number");
 328		error = -EINVAL;
 329		goto release_buf;
 330	}
 331
 332	/*
 333	 * We must be able to do sector-sized and sector-aligned IO.
 334	 */
 335	if (sector_size > sbp->sb_sectsize) {
 336		if (loud)
 337			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 338				sector_size, sbp->sb_sectsize);
 339		error = -ENOSYS;
 340		goto release_buf;
 341	}
 342
 343	if (buf_ops == NULL) {
 344		/*
 345		 * Re-read the superblock so the buffer is correctly sized,
 346		 * and properly verified.
 347		 */
 348		xfs_buf_relse(bp);
 349		sector_size = sbp->sb_sectsize;
 350		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 351		goto reread;
 352	}
 353
 354	xfs_reinit_percpu_counters(mp);
 355
 356	/* no need to be quiet anymore, so reset the buf ops */
 357	bp->b_ops = &xfs_sb_buf_ops;
 358
 359	mp->m_sb_bp = bp;
 360	xfs_buf_unlock(bp);
 361	return 0;
 362
 363release_buf:
 364	xfs_buf_relse(bp);
 365	return error;
 366}
 367
 368/*
 369 * Update alignment values based on mount options and sb values
 370 */
 371STATIC int
 372xfs_update_alignment(xfs_mount_t *mp)
 373{
 374	xfs_sb_t	*sbp = &(mp->m_sb);
 375
 376	if (mp->m_dalign) {
 377		/*
 378		 * If stripe unit and stripe width are not multiples
 379		 * of the fs blocksize turn off alignment.
 380		 */
 381		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 382		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 383			xfs_warn(mp,
 384		"alignment check failed: sunit/swidth vs. blocksize(%d)",
 385				sbp->sb_blocksize);
 386			return -EINVAL;
 387		} else {
 388			/*
 389			 * Convert the stripe unit and width to FSBs.
 390			 */
 391			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 392			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 393				xfs_warn(mp,
 394			"alignment check failed: sunit/swidth vs. agsize(%d)",
 395					 sbp->sb_agblocks);
 396				return -EINVAL;
 397			} else if (mp->m_dalign) {
 398				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 399			} else {
 400				xfs_warn(mp,
 401			"alignment check failed: sunit(%d) less than bsize(%d)",
 402					 mp->m_dalign, sbp->sb_blocksize);
 403				return -EINVAL;
 404			}
 405		}
 406
 407		/*
 408		 * Update superblock with new values
 409		 * and log changes
 410		 */
 411		if (xfs_sb_version_hasdalign(sbp)) {
 412			if (sbp->sb_unit != mp->m_dalign) {
 413				sbp->sb_unit = mp->m_dalign;
 414				mp->m_update_sb = true;
 415			}
 416			if (sbp->sb_width != mp->m_swidth) {
 417				sbp->sb_width = mp->m_swidth;
 418				mp->m_update_sb = true;
 419			}
 420		} else {
 421			xfs_warn(mp,
 422	"cannot change alignment: superblock does not support data alignment");
 423			return -EINVAL;
 424		}
 425	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 426		    xfs_sb_version_hasdalign(&mp->m_sb)) {
 427			mp->m_dalign = sbp->sb_unit;
 428			mp->m_swidth = sbp->sb_width;
 429	}
 430
 431	return 0;
 432}
 433
 434/*
 435 * Set the maximum inode count for this filesystem
 436 */
 437STATIC void
 438xfs_set_maxicount(xfs_mount_t *mp)
 439{
 440	xfs_sb_t	*sbp = &(mp->m_sb);
 441	uint64_t	icount;
 442
 443	if (sbp->sb_imax_pct) {
 444		/*
 445		 * Make sure the maximum inode count is a multiple
 446		 * of the units we allocate inodes in.
 447		 */
 448		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 449		do_div(icount, 100);
 450		do_div(icount, mp->m_ialloc_blks);
 451		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
 452				   sbp->sb_inopblog;
 453	} else {
 454		mp->m_maxicount = 0;
 455	}
 456}
 457
 458/*
 459 * Set the default minimum read and write sizes unless
 460 * already specified in a mount option.
 461 * We use smaller I/O sizes when the file system
 462 * is being used for NFS service (wsync mount option).
 463 */
 464STATIC void
 465xfs_set_rw_sizes(xfs_mount_t *mp)
 466{
 467	xfs_sb_t	*sbp = &(mp->m_sb);
 468	int		readio_log, writeio_log;
 469
 470	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 471		if (mp->m_flags & XFS_MOUNT_WSYNC) {
 472			readio_log = XFS_WSYNC_READIO_LOG;
 473			writeio_log = XFS_WSYNC_WRITEIO_LOG;
 474		} else {
 475			readio_log = XFS_READIO_LOG_LARGE;
 476			writeio_log = XFS_WRITEIO_LOG_LARGE;
 477		}
 478	} else {
 479		readio_log = mp->m_readio_log;
 480		writeio_log = mp->m_writeio_log;
 481	}
 482
 483	if (sbp->sb_blocklog > readio_log) {
 484		mp->m_readio_log = sbp->sb_blocklog;
 485	} else {
 486		mp->m_readio_log = readio_log;
 487	}
 488	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 489	if (sbp->sb_blocklog > writeio_log) {
 490		mp->m_writeio_log = sbp->sb_blocklog;
 491	} else {
 492		mp->m_writeio_log = writeio_log;
 493	}
 494	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 495}
 496
 497/*
 498 * precalculate the low space thresholds for dynamic speculative preallocation.
 499 */
 500void
 501xfs_set_low_space_thresholds(
 502	struct xfs_mount	*mp)
 503{
 504	int i;
 505
 506	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 507		uint64_t space = mp->m_sb.sb_dblocks;
 508
 509		do_div(space, 100);
 510		mp->m_low_space[i] = space * (i + 1);
 511	}
 512}
 513
 514
 515/*
 516 * Set whether we're using inode alignment.
 517 */
 518STATIC void
 519xfs_set_inoalignment(xfs_mount_t *mp)
 520{
 521	if (xfs_sb_version_hasalign(&mp->m_sb) &&
 522		mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
 523		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
 524	else
 525		mp->m_inoalign_mask = 0;
 526	/*
 527	 * If we are using stripe alignment, check whether
 528	 * the stripe unit is a multiple of the inode alignment
 529	 */
 530	if (mp->m_dalign && mp->m_inoalign_mask &&
 531	    !(mp->m_dalign & mp->m_inoalign_mask))
 532		mp->m_sinoalign = mp->m_dalign;
 533	else
 534		mp->m_sinoalign = 0;
 535}
 536
 537/*
 538 * Check that the data (and log if separate) is an ok size.
 539 */
 540STATIC int
 541xfs_check_sizes(
 542	struct xfs_mount *mp)
 543{
 544	struct xfs_buf	*bp;
 545	xfs_daddr_t	d;
 546	int		error;
 547
 548	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 549	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 550		xfs_warn(mp, "filesystem size mismatch detected");
 551		return -EFBIG;
 552	}
 553	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 554					d - XFS_FSS_TO_BB(mp, 1),
 555					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 556	if (error) {
 557		xfs_warn(mp, "last sector read failed");
 558		return error;
 559	}
 560	xfs_buf_relse(bp);
 561
 562	if (mp->m_logdev_targp == mp->m_ddev_targp)
 563		return 0;
 564
 565	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 566	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 567		xfs_warn(mp, "log size mismatch detected");
 568		return -EFBIG;
 569	}
 570	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 571					d - XFS_FSB_TO_BB(mp, 1),
 572					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
 573	if (error) {
 574		xfs_warn(mp, "log device read failed");
 575		return error;
 576	}
 577	xfs_buf_relse(bp);
 578	return 0;
 579}
 580
 581/*
 582 * Clear the quotaflags in memory and in the superblock.
 583 */
 584int
 585xfs_mount_reset_sbqflags(
 586	struct xfs_mount	*mp)
 587{
 588	mp->m_qflags = 0;
 589
 590	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 591	if (mp->m_sb.sb_qflags == 0)
 592		return 0;
 593	spin_lock(&mp->m_sb_lock);
 594	mp->m_sb.sb_qflags = 0;
 595	spin_unlock(&mp->m_sb_lock);
 596
 597	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 598		return 0;
 599
 600	return xfs_sync_sb(mp, false);
 601}
 602
 603uint64_t
 604xfs_default_resblks(xfs_mount_t *mp)
 605{
 606	uint64_t resblks;
 607
 608	/*
 609	 * We default to 5% or 8192 fsbs of space reserved, whichever is
 610	 * smaller.  This is intended to cover concurrent allocation
 611	 * transactions when we initially hit enospc. These each require a 4
 612	 * block reservation. Hence by default we cover roughly 2000 concurrent
 613	 * allocation reservations.
 614	 */
 615	resblks = mp->m_sb.sb_dblocks;
 616	do_div(resblks, 20);
 617	resblks = min_t(uint64_t, resblks, 8192);
 618	return resblks;
 619}
 620
 621/*
 622 * This function does the following on an initial mount of a file system:
 623 *	- reads the superblock from disk and init the mount struct
 624 *	- if we're a 32-bit kernel, do a size check on the superblock
 625 *		so we don't mount terabyte filesystems
 626 *	- init mount struct realtime fields
 627 *	- allocate inode hash table for fs
 628 *	- init directory manager
 629 *	- perform recovery and init the log manager
 630 */
 631int
 632xfs_mountfs(
 633	struct xfs_mount	*mp)
 634{
 635	struct xfs_sb		*sbp = &(mp->m_sb);
 636	struct xfs_inode	*rip;
 637	uint64_t		resblks;
 638	uint			quotamount = 0;
 639	uint			quotaflags = 0;
 640	int			error = 0;
 641
 642	xfs_sb_mount_common(mp, sbp);
 643
 644	/*
 645	 * Check for a mismatched features2 values.  Older kernels read & wrote
 646	 * into the wrong sb offset for sb_features2 on some platforms due to
 647	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 648	 * which made older superblock reading/writing routines swap it as a
 649	 * 64-bit value.
 650	 *
 651	 * For backwards compatibility, we make both slots equal.
 652	 *
 653	 * If we detect a mismatched field, we OR the set bits into the existing
 654	 * features2 field in case it has already been modified; we don't want
 655	 * to lose any features.  We then update the bad location with the ORed
 656	 * value so that older kernels will see any features2 flags. The
 657	 * superblock writeback code ensures the new sb_features2 is copied to
 658	 * sb_bad_features2 before it is logged or written to disk.
 659	 */
 660	if (xfs_sb_has_mismatched_features2(sbp)) {
 661		xfs_warn(mp, "correcting sb_features alignment problem");
 662		sbp->sb_features2 |= sbp->sb_bad_features2;
 663		mp->m_update_sb = true;
 664
 665		/*
 666		 * Re-check for ATTR2 in case it was found in bad_features2
 667		 * slot.
 668		 */
 669		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 670		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
 671			mp->m_flags |= XFS_MOUNT_ATTR2;
 672	}
 673
 674	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 675	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 676		xfs_sb_version_removeattr2(&mp->m_sb);
 677		mp->m_update_sb = true;
 678
 679		/* update sb_versionnum for the clearing of the morebits */
 680		if (!sbp->sb_features2)
 681			mp->m_update_sb = true;
 682	}
 683
 684	/* always use v2 inodes by default now */
 685	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 686		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 687		mp->m_update_sb = true;
 688	}
 689
 690	/*
 691	 * Check if sb_agblocks is aligned at stripe boundary
 692	 * If sb_agblocks is NOT aligned turn off m_dalign since
 693	 * allocator alignment is within an ag, therefore ag has
 694	 * to be aligned at stripe boundary.
 695	 */
 696	error = xfs_update_alignment(mp);
 697	if (error)
 698		goto out;
 699
 700	xfs_alloc_compute_maxlevels(mp);
 701	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 702	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 703	xfs_ialloc_compute_maxlevels(mp);
 704	xfs_rmapbt_compute_maxlevels(mp);
 705	xfs_refcountbt_compute_maxlevels(mp);
 706
 707	xfs_set_maxicount(mp);
 708
 709	/* enable fail_at_unmount as default */
 710	mp->m_fail_unmount = true;
 711
 712	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
 713	if (error)
 714		goto out;
 715
 716	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
 717			       &mp->m_kobj, "stats");
 718	if (error)
 719		goto out_remove_sysfs;
 720
 721	error = xfs_error_sysfs_init(mp);
 722	if (error)
 723		goto out_del_stats;
 724
 725	error = xfs_errortag_init(mp);
 726	if (error)
 727		goto out_remove_error_sysfs;
 728
 729	error = xfs_uuid_mount(mp);
 730	if (error)
 731		goto out_remove_errortag;
 732
 733	/*
 734	 * Set the minimum read and write sizes
 735	 */
 736	xfs_set_rw_sizes(mp);
 737
 738	/* set the low space thresholds for dynamic preallocation */
 739	xfs_set_low_space_thresholds(mp);
 740
 741	/*
 742	 * Set the inode cluster size.
 743	 * This may still be overridden by the file system
 744	 * block size if it is larger than the chosen cluster size.
 745	 *
 746	 * For v5 filesystems, scale the cluster size with the inode size to
 747	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
 748	 * has set the inode alignment value appropriately for larger cluster
 749	 * sizes.
 750	 */
 751	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
 752	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 753		int	new_size = mp->m_inode_cluster_size;
 754
 755		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
 756		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
 757			mp->m_inode_cluster_size = new_size;
 758	}
 759
 760	/*
 761	 * If enabled, sparse inode chunk alignment is expected to match the
 762	 * cluster size. Full inode chunk alignment must match the chunk size,
 763	 * but that is checked on sb read verification...
 764	 */
 765	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
 766	    mp->m_sb.sb_spino_align !=
 767			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
 768		xfs_warn(mp,
 769	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 770			 mp->m_sb.sb_spino_align,
 771			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
 772		error = -EINVAL;
 773		goto out_remove_uuid;
 774	}
 775
 776	/*
 777	 * Set inode alignment fields
 778	 */
 779	xfs_set_inoalignment(mp);
 780
 781	/*
 782	 * Check that the data (and log if separate) is an ok size.
 783	 */
 784	error = xfs_check_sizes(mp);
 785	if (error)
 786		goto out_remove_uuid;
 787
 788	/*
 789	 * Initialize realtime fields in the mount structure
 790	 */
 791	error = xfs_rtmount_init(mp);
 792	if (error) {
 793		xfs_warn(mp, "RT mount failed");
 794		goto out_remove_uuid;
 795	}
 796
 797	/*
 798	 *  Copies the low order bits of the timestamp and the randomly
 799	 *  set "sequence" number out of a UUID.
 800	 */
 801	mp->m_fixedfsid[0] =
 802		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
 803		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
 804	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
 805
 806	error = xfs_da_mount(mp);
 807	if (error) {
 808		xfs_warn(mp, "Failed dir/attr init: %d", error);
 809		goto out_remove_uuid;
 810	}
 811
 812	/*
 813	 * Initialize the precomputed transaction reservations values.
 814	 */
 815	xfs_trans_init(mp);
 816
 817	/*
 818	 * Allocate and initialize the per-ag data.
 819	 */
 
 
 820	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
 821	if (error) {
 822		xfs_warn(mp, "Failed per-ag init: %d", error);
 823		goto out_free_dir;
 824	}
 825
 826	if (!sbp->sb_logblocks) {
 827		xfs_warn(mp, "no log defined");
 828		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
 829		error = -EFSCORRUPTED;
 830		goto out_free_perag;
 831	}
 832
 833	/*
 834	 * Log's mount-time initialization. The first part of recovery can place
 835	 * some items on the AIL, to be handled when recovery is finished or
 836	 * cancelled.
 837	 */
 838	error = xfs_log_mount(mp, mp->m_logdev_targp,
 839			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 840			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 841	if (error) {
 842		xfs_warn(mp, "log mount failed");
 843		goto out_fail_wait;
 844	}
 845
 846	/*
 847	 * Now the log is mounted, we know if it was an unclean shutdown or
 848	 * not. If it was, with the first phase of recovery has completed, we
 849	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 850	 * but they are recovered transactionally in the second recovery phase
 851	 * later.
 852	 *
 853	 * Hence we can safely re-initialise incore superblock counters from
 854	 * the per-ag data. These may not be correct if the filesystem was not
 855	 * cleanly unmounted, so we need to wait for recovery to finish before
 856	 * doing this.
 857	 *
 858	 * If the filesystem was cleanly unmounted, then we can trust the
 859	 * values in the superblock to be correct and we don't need to do
 860	 * anything here.
 861	 *
 862	 * If we are currently making the filesystem, the initialisation will
 863	 * fail as the perag data is in an undefined state.
 864	 */
 865	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
 866	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
 867	     !mp->m_sb.sb_inprogress) {
 868		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
 869		if (error)
 870			goto out_log_dealloc;
 871	}
 872
 873	/*
 874	 * Get and sanity-check the root inode.
 875	 * Save the pointer to it in the mount structure.
 876	 */
 877	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
 878	if (error) {
 879		xfs_warn(mp, "failed to read root inode");
 880		goto out_log_dealloc;
 881	}
 882
 883	ASSERT(rip != NULL);
 884
 885	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
 886		xfs_warn(mp, "corrupted root inode %llu: not a directory",
 887			(unsigned long long)rip->i_ino);
 888		xfs_iunlock(rip, XFS_ILOCK_EXCL);
 889		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
 890				 mp);
 891		error = -EFSCORRUPTED;
 892		goto out_rele_rip;
 893	}
 894	mp->m_rootip = rip;	/* save it */
 895
 896	xfs_iunlock(rip, XFS_ILOCK_EXCL);
 897
 898	/*
 899	 * Initialize realtime inode pointers in the mount structure
 900	 */
 901	error = xfs_rtmount_inodes(mp);
 902	if (error) {
 903		/*
 904		 * Free up the root inode.
 905		 */
 906		xfs_warn(mp, "failed to read RT inodes");
 907		goto out_rele_rip;
 908	}
 909
 910	/*
 911	 * If this is a read-only mount defer the superblock updates until
 912	 * the next remount into writeable mode.  Otherwise we would never
 913	 * perform the update e.g. for the root filesystem.
 914	 */
 915	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 916		error = xfs_sync_sb(mp, false);
 917		if (error) {
 918			xfs_warn(mp, "failed to write sb changes");
 919			goto out_rtunmount;
 920		}
 921	}
 922
 923	/*
 924	 * Initialise the XFS quota management subsystem for this mount
 925	 */
 926	if (XFS_IS_QUOTA_RUNNING(mp)) {
 927		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
 928		if (error)
 929			goto out_rtunmount;
 930	} else {
 931		ASSERT(!XFS_IS_QUOTA_ON(mp));
 932
 933		/*
 934		 * If a file system had quotas running earlier, but decided to
 935		 * mount without -o uquota/pquota/gquota options, revoke the
 936		 * quotachecked license.
 937		 */
 938		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
 939			xfs_notice(mp, "resetting quota flags");
 940			error = xfs_mount_reset_sbqflags(mp);
 941			if (error)
 942				goto out_rtunmount;
 943		}
 944	}
 945
 946	/*
 
 
 
 
 
 
 
 
 
 947	 * Finish recovering the file system.  This part needed to be delayed
 948	 * until after the root and real-time bitmap inodes were consistently
 949	 * read in.
 950	 */
 951	error = xfs_log_mount_finish(mp);
 952	if (error) {
 953		xfs_warn(mp, "log mount finish failed");
 954		goto out_rtunmount;
 955	}
 956
 957	/*
 958	 * Now the log is fully replayed, we can transition to full read-only
 959	 * mode for read-only mounts. This will sync all the metadata and clean
 960	 * the log so that the recovery we just performed does not have to be
 961	 * replayed again on the next mount.
 962	 *
 963	 * We use the same quiesce mechanism as the rw->ro remount, as they are
 964	 * semantically identical operations.
 965	 */
 966	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
 967							XFS_MOUNT_RDONLY) {
 968		xfs_quiesce_attr(mp);
 969	}
 970
 971	/*
 972	 * Complete the quota initialisation, post-log-replay component.
 973	 */
 974	if (quotamount) {
 975		ASSERT(mp->m_qflags == 0);
 976		mp->m_qflags = quotaflags;
 977
 978		xfs_qm_mount_quotas(mp);
 979	}
 980
 981	/*
 982	 * Now we are mounted, reserve a small amount of unused space for
 983	 * privileged transactions. This is needed so that transaction
 984	 * space required for critical operations can dip into this pool
 985	 * when at ENOSPC. This is needed for operations like create with
 986	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
 987	 * are not allowed to use this reserved space.
 988	 *
 989	 * This may drive us straight to ENOSPC on mount, but that implies
 990	 * we were already there on the last unmount. Warn if this occurs.
 991	 */
 992	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 993		resblks = xfs_default_resblks(mp);
 994		error = xfs_reserve_blocks(mp, &resblks, NULL);
 995		if (error)
 996			xfs_warn(mp,
 997	"Unable to allocate reserve blocks. Continuing without reserve pool.");
 998
 999		/* Recover any CoW blocks that never got remapped. */
1000		error = xfs_reflink_recover_cow(mp);
1001		if (error) {
1002			xfs_err(mp,
1003	"Error %d recovering leftover CoW allocations.", error);
1004			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1005			goto out_quota;
1006		}
1007
1008		/* Reserve AG blocks for future btree expansion. */
1009		error = xfs_fs_reserve_ag_blocks(mp);
1010		if (error && error != -ENOSPC)
1011			goto out_agresv;
1012	}
1013
1014	return 0;
1015
1016 out_agresv:
1017	xfs_fs_unreserve_ag_blocks(mp);
1018 out_quota:
1019	xfs_qm_unmount_quotas(mp);
1020 out_rtunmount:
 
1021	xfs_rtunmount_inodes(mp);
1022 out_rele_rip:
1023	IRELE(rip);
1024	/* Clean out dquots that might be in memory after quotacheck. */
1025	xfs_qm_unmount(mp);
1026	/*
1027	 * Cancel all delayed reclaim work and reclaim the inodes directly.
1028	 * We have to do this /after/ rtunmount and qm_unmount because those
1029	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1030	 *
1031	 * This is slightly different from the unmountfs call sequence
1032	 * because we could be tearing down a partially set up mount.  In
1033	 * particular, if log_mount_finish fails we bail out without calling
1034	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1035	 * quota inodes.
1036	 */
1037	cancel_delayed_work_sync(&mp->m_reclaim_work);
1038	xfs_reclaim_inodes(mp, SYNC_WAIT);
1039 out_log_dealloc:
1040	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1041	xfs_log_mount_cancel(mp);
1042 out_fail_wait:
1043	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1044		xfs_wait_buftarg(mp->m_logdev_targp);
1045	xfs_wait_buftarg(mp->m_ddev_targp);
1046 out_free_perag:
1047	xfs_free_perag(mp);
1048 out_free_dir:
1049	xfs_da_unmount(mp);
1050 out_remove_uuid:
1051	xfs_uuid_unmount(mp);
1052 out_remove_errortag:
1053	xfs_errortag_del(mp);
1054 out_remove_error_sysfs:
1055	xfs_error_sysfs_del(mp);
1056 out_del_stats:
1057	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1058 out_remove_sysfs:
1059	xfs_sysfs_del(&mp->m_kobj);
1060 out:
1061	return error;
1062}
1063
1064/*
1065 * This flushes out the inodes,dquots and the superblock, unmounts the
1066 * log and makes sure that incore structures are freed.
1067 */
1068void
1069xfs_unmountfs(
1070	struct xfs_mount	*mp)
1071{
1072	uint64_t		resblks;
1073	int			error;
1074
1075	cancel_delayed_work_sync(&mp->m_eofblocks_work);
1076	cancel_delayed_work_sync(&mp->m_cowblocks_work);
1077
1078	xfs_fs_unreserve_ag_blocks(mp);
1079	xfs_qm_unmount_quotas(mp);
1080	xfs_rtunmount_inodes(mp);
1081	IRELE(mp->m_rootip);
1082
1083	/*
1084	 * We can potentially deadlock here if we have an inode cluster
1085	 * that has been freed has its buffer still pinned in memory because
1086	 * the transaction is still sitting in a iclog. The stale inodes
1087	 * on that buffer will have their flush locks held until the
1088	 * transaction hits the disk and the callbacks run. the inode
1089	 * flush takes the flush lock unconditionally and with nothing to
1090	 * push out the iclog we will never get that unlocked. hence we
1091	 * need to force the log first.
1092	 */
1093	xfs_log_force(mp, XFS_LOG_SYNC);
1094
1095	/*
1096	 * Wait for all busy extents to be freed, including completion of
1097	 * any discard operation.
1098	 */
1099	xfs_extent_busy_wait_all(mp);
1100	flush_workqueue(xfs_discard_wq);
1101
1102	/*
1103	 * We now need to tell the world we are unmounting. This will allow
1104	 * us to detect that the filesystem is going away and we should error
1105	 * out anything that we have been retrying in the background. This will
1106	 * prevent neverending retries in AIL pushing from hanging the unmount.
1107	 */
1108	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1109
1110	/*
1111	 * Flush all pending changes from the AIL.
1112	 */
1113	xfs_ail_push_all_sync(mp->m_ail);
1114
1115	/*
1116	 * And reclaim all inodes.  At this point there should be no dirty
1117	 * inodes and none should be pinned or locked, but use synchronous
1118	 * reclaim just to be sure. We can stop background inode reclaim
1119	 * here as well if it is still running.
1120	 */
1121	cancel_delayed_work_sync(&mp->m_reclaim_work);
1122	xfs_reclaim_inodes(mp, SYNC_WAIT);
1123
1124	xfs_qm_unmount(mp);
1125
1126	/*
1127	 * Unreserve any blocks we have so that when we unmount we don't account
1128	 * the reserved free space as used. This is really only necessary for
1129	 * lazy superblock counting because it trusts the incore superblock
1130	 * counters to be absolutely correct on clean unmount.
1131	 *
1132	 * We don't bother correcting this elsewhere for lazy superblock
1133	 * counting because on mount of an unclean filesystem we reconstruct the
1134	 * correct counter value and this is irrelevant.
1135	 *
1136	 * For non-lazy counter filesystems, this doesn't matter at all because
1137	 * we only every apply deltas to the superblock and hence the incore
1138	 * value does not matter....
1139	 */
1140	resblks = 0;
1141	error = xfs_reserve_blocks(mp, &resblks, NULL);
1142	if (error)
1143		xfs_warn(mp, "Unable to free reserved block pool. "
1144				"Freespace may not be correct on next mount.");
1145
1146	error = xfs_log_sbcount(mp);
1147	if (error)
1148		xfs_warn(mp, "Unable to update superblock counters. "
1149				"Freespace may not be correct on next mount.");
1150
1151
1152	xfs_log_unmount(mp);
1153	xfs_da_unmount(mp);
1154	xfs_uuid_unmount(mp);
1155
1156#if defined(DEBUG)
1157	xfs_errortag_clearall(mp);
1158#endif
1159	xfs_free_perag(mp);
1160
1161	xfs_errortag_del(mp);
1162	xfs_error_sysfs_del(mp);
1163	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1164	xfs_sysfs_del(&mp->m_kobj);
1165}
1166
1167/*
1168 * Determine whether modifications can proceed. The caller specifies the minimum
1169 * freeze level for which modifications should not be allowed. This allows
1170 * certain operations to proceed while the freeze sequence is in progress, if
1171 * necessary.
1172 */
1173bool
1174xfs_fs_writable(
1175	struct xfs_mount	*mp,
1176	int			level)
1177{
1178	ASSERT(level > SB_UNFROZEN);
1179	if ((mp->m_super->s_writers.frozen >= level) ||
1180	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1181		return false;
1182
1183	return true;
1184}
1185
1186/*
1187 * xfs_log_sbcount
1188 *
1189 * Sync the superblock counters to disk.
1190 *
1191 * Note this code can be called during the process of freezing, so we use the
1192 * transaction allocator that does not block when the transaction subsystem is
1193 * in its frozen state.
1194 */
1195int
1196xfs_log_sbcount(xfs_mount_t *mp)
1197{
1198	/* allow this to proceed during the freeze sequence... */
1199	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1200		return 0;
1201
1202	/*
1203	 * we don't need to do this if we are updating the superblock
1204	 * counters on every modification.
1205	 */
1206	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1207		return 0;
1208
1209	return xfs_sync_sb(mp, true);
1210}
1211
1212/*
1213 * Deltas for the inode count are +/-64, hence we use a large batch size
1214 * of 128 so we don't need to take the counter lock on every update.
1215 */
1216#define XFS_ICOUNT_BATCH	128
1217int
1218xfs_mod_icount(
1219	struct xfs_mount	*mp,
1220	int64_t			delta)
1221{
1222	percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1223	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1224		ASSERT(0);
1225		percpu_counter_add(&mp->m_icount, -delta);
1226		return -EINVAL;
1227	}
1228	return 0;
1229}
1230
1231int
1232xfs_mod_ifree(
1233	struct xfs_mount	*mp,
1234	int64_t			delta)
1235{
1236	percpu_counter_add(&mp->m_ifree, delta);
1237	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1238		ASSERT(0);
1239		percpu_counter_add(&mp->m_ifree, -delta);
1240		return -EINVAL;
1241	}
1242	return 0;
1243}
1244
1245/*
1246 * Deltas for the block count can vary from 1 to very large, but lock contention
1247 * only occurs on frequent small block count updates such as in the delayed
1248 * allocation path for buffered writes (page a time updates). Hence we set
1249 * a large batch count (1024) to minimise global counter updates except when
1250 * we get near to ENOSPC and we have to be very accurate with our updates.
1251 */
1252#define XFS_FDBLOCKS_BATCH	1024
1253int
1254xfs_mod_fdblocks(
1255	struct xfs_mount	*mp,
1256	int64_t			delta,
1257	bool			rsvd)
1258{
1259	int64_t			lcounter;
1260	long long		res_used;
1261	s32			batch;
1262
1263	if (delta > 0) {
1264		/*
1265		 * If the reserve pool is depleted, put blocks back into it
1266		 * first. Most of the time the pool is full.
1267		 */
1268		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1269			percpu_counter_add(&mp->m_fdblocks, delta);
1270			return 0;
1271		}
1272
1273		spin_lock(&mp->m_sb_lock);
1274		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1275
1276		if (res_used > delta) {
1277			mp->m_resblks_avail += delta;
1278		} else {
1279			delta -= res_used;
1280			mp->m_resblks_avail = mp->m_resblks;
1281			percpu_counter_add(&mp->m_fdblocks, delta);
1282		}
1283		spin_unlock(&mp->m_sb_lock);
1284		return 0;
1285	}
1286
1287	/*
1288	 * Taking blocks away, need to be more accurate the closer we
1289	 * are to zero.
1290	 *
1291	 * If the counter has a value of less than 2 * max batch size,
1292	 * then make everything serialise as we are real close to
1293	 * ENOSPC.
1294	 */
1295	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1296				     XFS_FDBLOCKS_BATCH) < 0)
1297		batch = 1;
1298	else
1299		batch = XFS_FDBLOCKS_BATCH;
1300
1301	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1302	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1303				     XFS_FDBLOCKS_BATCH) >= 0) {
1304		/* we had space! */
1305		return 0;
1306	}
1307
1308	/*
1309	 * lock up the sb for dipping into reserves before releasing the space
1310	 * that took us to ENOSPC.
1311	 */
1312	spin_lock(&mp->m_sb_lock);
1313	percpu_counter_add(&mp->m_fdblocks, -delta);
1314	if (!rsvd)
1315		goto fdblocks_enospc;
1316
1317	lcounter = (long long)mp->m_resblks_avail + delta;
1318	if (lcounter >= 0) {
1319		mp->m_resblks_avail = lcounter;
1320		spin_unlock(&mp->m_sb_lock);
1321		return 0;
1322	}
1323	printk_once(KERN_WARNING
1324		"Filesystem \"%s\": reserve blocks depleted! "
1325		"Consider increasing reserve pool size.",
1326		mp->m_fsname);
1327fdblocks_enospc:
1328	spin_unlock(&mp->m_sb_lock);
1329	return -ENOSPC;
1330}
1331
1332int
1333xfs_mod_frextents(
1334	struct xfs_mount	*mp,
1335	int64_t			delta)
1336{
1337	int64_t			lcounter;
1338	int			ret = 0;
1339
1340	spin_lock(&mp->m_sb_lock);
1341	lcounter = mp->m_sb.sb_frextents + delta;
1342	if (lcounter < 0)
1343		ret = -ENOSPC;
1344	else
1345		mp->m_sb.sb_frextents = lcounter;
1346	spin_unlock(&mp->m_sb_lock);
1347	return ret;
1348}
1349
1350/*
1351 * xfs_getsb() is called to obtain the buffer for the superblock.
1352 * The buffer is returned locked and read in from disk.
1353 * The buffer should be released with a call to xfs_brelse().
1354 *
1355 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1356 * the superblock buffer if it can be locked without sleeping.
1357 * If it can't then we'll return NULL.
1358 */
1359struct xfs_buf *
1360xfs_getsb(
1361	struct xfs_mount	*mp,
1362	int			flags)
1363{
1364	struct xfs_buf		*bp = mp->m_sb_bp;
1365
1366	if (!xfs_buf_trylock(bp)) {
1367		if (flags & XBF_TRYLOCK)
1368			return NULL;
1369		xfs_buf_lock(bp);
1370	}
1371
1372	xfs_buf_hold(bp);
1373	ASSERT(bp->b_flags & XBF_DONE);
1374	return bp;
1375}
1376
1377/*
1378 * Used to free the superblock along various error paths.
1379 */
1380void
1381xfs_freesb(
1382	struct xfs_mount	*mp)
1383{
1384	struct xfs_buf		*bp = mp->m_sb_bp;
1385
1386	xfs_buf_lock(bp);
1387	mp->m_sb_bp = NULL;
1388	xfs_buf_relse(bp);
1389}
1390
1391/*
1392 * If the underlying (data/log/rt) device is readonly, there are some
1393 * operations that cannot proceed.
1394 */
1395int
1396xfs_dev_is_read_only(
1397	struct xfs_mount	*mp,
1398	char			*message)
1399{
1400	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1401	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1402	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1403		xfs_notice(mp, "%s required on read-only device.", message);
1404		xfs_notice(mp, "write access unavailable, cannot proceed.");
1405		return -EROFS;
1406	}
1407	return 0;
1408}
v4.10.11
   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_defer.h"
  28#include "xfs_da_format.h"
  29#include "xfs_da_btree.h"
  30#include "xfs_inode.h"
  31#include "xfs_dir2.h"
  32#include "xfs_ialloc.h"
  33#include "xfs_alloc.h"
  34#include "xfs_rtalloc.h"
  35#include "xfs_bmap.h"
  36#include "xfs_trans.h"
  37#include "xfs_trans_priv.h"
  38#include "xfs_log.h"
  39#include "xfs_error.h"
  40#include "xfs_quota.h"
  41#include "xfs_fsops.h"
  42#include "xfs_trace.h"
  43#include "xfs_icache.h"
  44#include "xfs_sysfs.h"
  45#include "xfs_rmap_btree.h"
  46#include "xfs_refcount_btree.h"
  47#include "xfs_reflink.h"
 
  48
  49
  50static DEFINE_MUTEX(xfs_uuid_table_mutex);
  51static int xfs_uuid_table_size;
  52static uuid_t *xfs_uuid_table;
  53
  54void
  55xfs_uuid_table_free(void)
  56{
  57	if (xfs_uuid_table_size == 0)
  58		return;
  59	kmem_free(xfs_uuid_table);
  60	xfs_uuid_table = NULL;
  61	xfs_uuid_table_size = 0;
  62}
  63
  64/*
  65 * See if the UUID is unique among mounted XFS filesystems.
  66 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  67 */
  68STATIC int
  69xfs_uuid_mount(
  70	struct xfs_mount	*mp)
  71{
  72	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  73	int			hole, i;
  74
 
 
 
  75	if (mp->m_flags & XFS_MOUNT_NOUUID)
  76		return 0;
  77
  78	if (uuid_is_nil(uuid)) {
  79		xfs_warn(mp, "Filesystem has nil UUID - can't mount");
  80		return -EINVAL;
  81	}
  82
  83	mutex_lock(&xfs_uuid_table_mutex);
  84	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  85		if (uuid_is_nil(&xfs_uuid_table[i])) {
  86			hole = i;
  87			continue;
  88		}
  89		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  90			goto out_duplicate;
  91	}
  92
  93	if (hole < 0) {
  94		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
  95			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
  96			KM_SLEEP);
  97		hole = xfs_uuid_table_size++;
  98	}
  99	xfs_uuid_table[hole] = *uuid;
 100	mutex_unlock(&xfs_uuid_table_mutex);
 101
 102	return 0;
 103
 104 out_duplicate:
 105	mutex_unlock(&xfs_uuid_table_mutex);
 106	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
 107	return -EINVAL;
 108}
 109
 110STATIC void
 111xfs_uuid_unmount(
 112	struct xfs_mount	*mp)
 113{
 114	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 115	int			i;
 116
 117	if (mp->m_flags & XFS_MOUNT_NOUUID)
 118		return;
 119
 120	mutex_lock(&xfs_uuid_table_mutex);
 121	for (i = 0; i < xfs_uuid_table_size; i++) {
 122		if (uuid_is_nil(&xfs_uuid_table[i]))
 123			continue;
 124		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 125			continue;
 126		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 127		break;
 128	}
 129	ASSERT(i < xfs_uuid_table_size);
 130	mutex_unlock(&xfs_uuid_table_mutex);
 131}
 132
 133
 134STATIC void
 135__xfs_free_perag(
 136	struct rcu_head	*head)
 137{
 138	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 139
 140	ASSERT(atomic_read(&pag->pag_ref) == 0);
 141	kmem_free(pag);
 142}
 143
 144/*
 145 * Free up the per-ag resources associated with the mount structure.
 146 */
 147STATIC void
 148xfs_free_perag(
 149	xfs_mount_t	*mp)
 150{
 151	xfs_agnumber_t	agno;
 152	struct xfs_perag *pag;
 153
 154	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
 155		spin_lock(&mp->m_perag_lock);
 156		pag = radix_tree_delete(&mp->m_perag_tree, agno);
 157		spin_unlock(&mp->m_perag_lock);
 158		ASSERT(pag);
 159		ASSERT(atomic_read(&pag->pag_ref) == 0);
 160		xfs_buf_hash_destroy(pag);
 
 161		call_rcu(&pag->rcu_head, __xfs_free_perag);
 162	}
 163}
 164
 165/*
 166 * Check size of device based on the (data/realtime) block count.
 167 * Note: this check is used by the growfs code as well as mount.
 168 */
 169int
 170xfs_sb_validate_fsb_count(
 171	xfs_sb_t	*sbp,
 172	__uint64_t	nblocks)
 173{
 174	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 175	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 176
 177	/* Limited by ULONG_MAX of page cache index */
 178	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 179		return -EFBIG;
 180	return 0;
 181}
 182
 183int
 184xfs_initialize_perag(
 185	xfs_mount_t	*mp,
 186	xfs_agnumber_t	agcount,
 187	xfs_agnumber_t	*maxagi)
 188{
 189	xfs_agnumber_t	index;
 190	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
 191	xfs_perag_t	*pag;
 192	int		error = -ENOMEM;
 193
 194	/*
 195	 * Walk the current per-ag tree so we don't try to initialise AGs
 196	 * that already exist (growfs case). Allocate and insert all the
 197	 * AGs we don't find ready for initialisation.
 198	 */
 199	for (index = 0; index < agcount; index++) {
 200		pag = xfs_perag_get(mp, index);
 201		if (pag) {
 202			xfs_perag_put(pag);
 203			continue;
 204		}
 205
 206		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
 207		if (!pag)
 208			goto out_unwind_new_pags;
 209		pag->pag_agno = index;
 210		pag->pag_mount = mp;
 211		spin_lock_init(&pag->pag_ici_lock);
 212		mutex_init(&pag->pag_ici_reclaim_lock);
 213		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 214		if (xfs_buf_hash_init(pag))
 215			goto out_free_pag;
 
 216
 217		if (radix_tree_preload(GFP_NOFS))
 218			goto out_hash_destroy;
 219
 220		spin_lock(&mp->m_perag_lock);
 221		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
 222			BUG();
 223			spin_unlock(&mp->m_perag_lock);
 224			radix_tree_preload_end();
 225			error = -EEXIST;
 226			goto out_hash_destroy;
 227		}
 228		spin_unlock(&mp->m_perag_lock);
 229		radix_tree_preload_end();
 230		/* first new pag is fully initialized */
 231		if (first_initialised == NULLAGNUMBER)
 232			first_initialised = index;
 233	}
 234
 235	index = xfs_set_inode_alloc(mp, agcount);
 236
 237	if (maxagi)
 238		*maxagi = index;
 239
 240	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
 241	return 0;
 242
 243out_hash_destroy:
 244	xfs_buf_hash_destroy(pag);
 245out_free_pag:
 
 246	kmem_free(pag);
 247out_unwind_new_pags:
 248	/* unwind any prior newly initialized pags */
 249	for (index = first_initialised; index < agcount; index++) {
 250		pag = radix_tree_delete(&mp->m_perag_tree, index);
 251		if (!pag)
 252			break;
 253		xfs_buf_hash_destroy(pag);
 
 254		kmem_free(pag);
 255	}
 256	return error;
 257}
 258
 259/*
 260 * xfs_readsb
 261 *
 262 * Does the initial read of the superblock.
 263 */
 264int
 265xfs_readsb(
 266	struct xfs_mount *mp,
 267	int		flags)
 268{
 269	unsigned int	sector_size;
 270	struct xfs_buf	*bp;
 271	struct xfs_sb	*sbp = &mp->m_sb;
 272	int		error;
 273	int		loud = !(flags & XFS_MFSI_QUIET);
 274	const struct xfs_buf_ops *buf_ops;
 275
 276	ASSERT(mp->m_sb_bp == NULL);
 277	ASSERT(mp->m_ddev_targp != NULL);
 278
 279	/*
 280	 * For the initial read, we must guess at the sector
 281	 * size based on the block device.  It's enough to
 282	 * get the sb_sectsize out of the superblock and
 283	 * then reread with the proper length.
 284	 * We don't verify it yet, because it may not be complete.
 285	 */
 286	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 287	buf_ops = NULL;
 288
 289	/*
 290	 * Allocate a (locked) buffer to hold the superblock. This will be kept
 291	 * around at all times to optimize access to the superblock. Therefore,
 292	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
 293	 * elevated.
 294	 */
 295reread:
 296	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 297				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
 298				      buf_ops);
 299	if (error) {
 300		if (loud)
 301			xfs_warn(mp, "SB validate failed with error %d.", error);
 302		/* bad CRC means corrupted metadata */
 303		if (error == -EFSBADCRC)
 304			error = -EFSCORRUPTED;
 305		return error;
 306	}
 307
 308	/*
 309	 * Initialize the mount structure from the superblock.
 310	 */
 311	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
 312
 313	/*
 314	 * If we haven't validated the superblock, do so now before we try
 315	 * to check the sector size and reread the superblock appropriately.
 316	 */
 317	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 318		if (loud)
 319			xfs_warn(mp, "Invalid superblock magic number");
 320		error = -EINVAL;
 321		goto release_buf;
 322	}
 323
 324	/*
 325	 * We must be able to do sector-sized and sector-aligned IO.
 326	 */
 327	if (sector_size > sbp->sb_sectsize) {
 328		if (loud)
 329			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 330				sector_size, sbp->sb_sectsize);
 331		error = -ENOSYS;
 332		goto release_buf;
 333	}
 334
 335	if (buf_ops == NULL) {
 336		/*
 337		 * Re-read the superblock so the buffer is correctly sized,
 338		 * and properly verified.
 339		 */
 340		xfs_buf_relse(bp);
 341		sector_size = sbp->sb_sectsize;
 342		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 343		goto reread;
 344	}
 345
 346	xfs_reinit_percpu_counters(mp);
 347
 348	/* no need to be quiet anymore, so reset the buf ops */
 349	bp->b_ops = &xfs_sb_buf_ops;
 350
 351	mp->m_sb_bp = bp;
 352	xfs_buf_unlock(bp);
 353	return 0;
 354
 355release_buf:
 356	xfs_buf_relse(bp);
 357	return error;
 358}
 359
 360/*
 361 * Update alignment values based on mount options and sb values
 362 */
 363STATIC int
 364xfs_update_alignment(xfs_mount_t *mp)
 365{
 366	xfs_sb_t	*sbp = &(mp->m_sb);
 367
 368	if (mp->m_dalign) {
 369		/*
 370		 * If stripe unit and stripe width are not multiples
 371		 * of the fs blocksize turn off alignment.
 372		 */
 373		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 374		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 375			xfs_warn(mp,
 376		"alignment check failed: sunit/swidth vs. blocksize(%d)",
 377				sbp->sb_blocksize);
 378			return -EINVAL;
 379		} else {
 380			/*
 381			 * Convert the stripe unit and width to FSBs.
 382			 */
 383			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 384			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 385				xfs_warn(mp,
 386			"alignment check failed: sunit/swidth vs. agsize(%d)",
 387					 sbp->sb_agblocks);
 388				return -EINVAL;
 389			} else if (mp->m_dalign) {
 390				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 391			} else {
 392				xfs_warn(mp,
 393			"alignment check failed: sunit(%d) less than bsize(%d)",
 394					 mp->m_dalign, sbp->sb_blocksize);
 395				return -EINVAL;
 396			}
 397		}
 398
 399		/*
 400		 * Update superblock with new values
 401		 * and log changes
 402		 */
 403		if (xfs_sb_version_hasdalign(sbp)) {
 404			if (sbp->sb_unit != mp->m_dalign) {
 405				sbp->sb_unit = mp->m_dalign;
 406				mp->m_update_sb = true;
 407			}
 408			if (sbp->sb_width != mp->m_swidth) {
 409				sbp->sb_width = mp->m_swidth;
 410				mp->m_update_sb = true;
 411			}
 412		} else {
 413			xfs_warn(mp,
 414	"cannot change alignment: superblock does not support data alignment");
 415			return -EINVAL;
 416		}
 417	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 418		    xfs_sb_version_hasdalign(&mp->m_sb)) {
 419			mp->m_dalign = sbp->sb_unit;
 420			mp->m_swidth = sbp->sb_width;
 421	}
 422
 423	return 0;
 424}
 425
 426/*
 427 * Set the maximum inode count for this filesystem
 428 */
 429STATIC void
 430xfs_set_maxicount(xfs_mount_t *mp)
 431{
 432	xfs_sb_t	*sbp = &(mp->m_sb);
 433	__uint64_t	icount;
 434
 435	if (sbp->sb_imax_pct) {
 436		/*
 437		 * Make sure the maximum inode count is a multiple
 438		 * of the units we allocate inodes in.
 439		 */
 440		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 441		do_div(icount, 100);
 442		do_div(icount, mp->m_ialloc_blks);
 443		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
 444				   sbp->sb_inopblog;
 445	} else {
 446		mp->m_maxicount = 0;
 447	}
 448}
 449
 450/*
 451 * Set the default minimum read and write sizes unless
 452 * already specified in a mount option.
 453 * We use smaller I/O sizes when the file system
 454 * is being used for NFS service (wsync mount option).
 455 */
 456STATIC void
 457xfs_set_rw_sizes(xfs_mount_t *mp)
 458{
 459	xfs_sb_t	*sbp = &(mp->m_sb);
 460	int		readio_log, writeio_log;
 461
 462	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 463		if (mp->m_flags & XFS_MOUNT_WSYNC) {
 464			readio_log = XFS_WSYNC_READIO_LOG;
 465			writeio_log = XFS_WSYNC_WRITEIO_LOG;
 466		} else {
 467			readio_log = XFS_READIO_LOG_LARGE;
 468			writeio_log = XFS_WRITEIO_LOG_LARGE;
 469		}
 470	} else {
 471		readio_log = mp->m_readio_log;
 472		writeio_log = mp->m_writeio_log;
 473	}
 474
 475	if (sbp->sb_blocklog > readio_log) {
 476		mp->m_readio_log = sbp->sb_blocklog;
 477	} else {
 478		mp->m_readio_log = readio_log;
 479	}
 480	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 481	if (sbp->sb_blocklog > writeio_log) {
 482		mp->m_writeio_log = sbp->sb_blocklog;
 483	} else {
 484		mp->m_writeio_log = writeio_log;
 485	}
 486	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 487}
 488
 489/*
 490 * precalculate the low space thresholds for dynamic speculative preallocation.
 491 */
 492void
 493xfs_set_low_space_thresholds(
 494	struct xfs_mount	*mp)
 495{
 496	int i;
 497
 498	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 499		__uint64_t space = mp->m_sb.sb_dblocks;
 500
 501		do_div(space, 100);
 502		mp->m_low_space[i] = space * (i + 1);
 503	}
 504}
 505
 506
 507/*
 508 * Set whether we're using inode alignment.
 509 */
 510STATIC void
 511xfs_set_inoalignment(xfs_mount_t *mp)
 512{
 513	if (xfs_sb_version_hasalign(&mp->m_sb) &&
 514		mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
 515		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
 516	else
 517		mp->m_inoalign_mask = 0;
 518	/*
 519	 * If we are using stripe alignment, check whether
 520	 * the stripe unit is a multiple of the inode alignment
 521	 */
 522	if (mp->m_dalign && mp->m_inoalign_mask &&
 523	    !(mp->m_dalign & mp->m_inoalign_mask))
 524		mp->m_sinoalign = mp->m_dalign;
 525	else
 526		mp->m_sinoalign = 0;
 527}
 528
 529/*
 530 * Check that the data (and log if separate) is an ok size.
 531 */
 532STATIC int
 533xfs_check_sizes(
 534	struct xfs_mount *mp)
 535{
 536	struct xfs_buf	*bp;
 537	xfs_daddr_t	d;
 538	int		error;
 539
 540	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 541	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 542		xfs_warn(mp, "filesystem size mismatch detected");
 543		return -EFBIG;
 544	}
 545	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 546					d - XFS_FSS_TO_BB(mp, 1),
 547					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 548	if (error) {
 549		xfs_warn(mp, "last sector read failed");
 550		return error;
 551	}
 552	xfs_buf_relse(bp);
 553
 554	if (mp->m_logdev_targp == mp->m_ddev_targp)
 555		return 0;
 556
 557	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 558	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 559		xfs_warn(mp, "log size mismatch detected");
 560		return -EFBIG;
 561	}
 562	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 563					d - XFS_FSB_TO_BB(mp, 1),
 564					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
 565	if (error) {
 566		xfs_warn(mp, "log device read failed");
 567		return error;
 568	}
 569	xfs_buf_relse(bp);
 570	return 0;
 571}
 572
 573/*
 574 * Clear the quotaflags in memory and in the superblock.
 575 */
 576int
 577xfs_mount_reset_sbqflags(
 578	struct xfs_mount	*mp)
 579{
 580	mp->m_qflags = 0;
 581
 582	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 583	if (mp->m_sb.sb_qflags == 0)
 584		return 0;
 585	spin_lock(&mp->m_sb_lock);
 586	mp->m_sb.sb_qflags = 0;
 587	spin_unlock(&mp->m_sb_lock);
 588
 589	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 590		return 0;
 591
 592	return xfs_sync_sb(mp, false);
 593}
 594
 595__uint64_t
 596xfs_default_resblks(xfs_mount_t *mp)
 597{
 598	__uint64_t resblks;
 599
 600	/*
 601	 * We default to 5% or 8192 fsbs of space reserved, whichever is
 602	 * smaller.  This is intended to cover concurrent allocation
 603	 * transactions when we initially hit enospc. These each require a 4
 604	 * block reservation. Hence by default we cover roughly 2000 concurrent
 605	 * allocation reservations.
 606	 */
 607	resblks = mp->m_sb.sb_dblocks;
 608	do_div(resblks, 20);
 609	resblks = min_t(__uint64_t, resblks, 8192);
 610	return resblks;
 611}
 612
 613/*
 614 * This function does the following on an initial mount of a file system:
 615 *	- reads the superblock from disk and init the mount struct
 616 *	- if we're a 32-bit kernel, do a size check on the superblock
 617 *		so we don't mount terabyte filesystems
 618 *	- init mount struct realtime fields
 619 *	- allocate inode hash table for fs
 620 *	- init directory manager
 621 *	- perform recovery and init the log manager
 622 */
 623int
 624xfs_mountfs(
 625	struct xfs_mount	*mp)
 626{
 627	struct xfs_sb		*sbp = &(mp->m_sb);
 628	struct xfs_inode	*rip;
 629	__uint64_t		resblks;
 630	uint			quotamount = 0;
 631	uint			quotaflags = 0;
 632	int			error = 0;
 633
 634	xfs_sb_mount_common(mp, sbp);
 635
 636	/*
 637	 * Check for a mismatched features2 values.  Older kernels read & wrote
 638	 * into the wrong sb offset for sb_features2 on some platforms due to
 639	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 640	 * which made older superblock reading/writing routines swap it as a
 641	 * 64-bit value.
 642	 *
 643	 * For backwards compatibility, we make both slots equal.
 644	 *
 645	 * If we detect a mismatched field, we OR the set bits into the existing
 646	 * features2 field in case it has already been modified; we don't want
 647	 * to lose any features.  We then update the bad location with the ORed
 648	 * value so that older kernels will see any features2 flags. The
 649	 * superblock writeback code ensures the new sb_features2 is copied to
 650	 * sb_bad_features2 before it is logged or written to disk.
 651	 */
 652	if (xfs_sb_has_mismatched_features2(sbp)) {
 653		xfs_warn(mp, "correcting sb_features alignment problem");
 654		sbp->sb_features2 |= sbp->sb_bad_features2;
 655		mp->m_update_sb = true;
 656
 657		/*
 658		 * Re-check for ATTR2 in case it was found in bad_features2
 659		 * slot.
 660		 */
 661		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 662		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
 663			mp->m_flags |= XFS_MOUNT_ATTR2;
 664	}
 665
 666	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 667	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 668		xfs_sb_version_removeattr2(&mp->m_sb);
 669		mp->m_update_sb = true;
 670
 671		/* update sb_versionnum for the clearing of the morebits */
 672		if (!sbp->sb_features2)
 673			mp->m_update_sb = true;
 674	}
 675
 676	/* always use v2 inodes by default now */
 677	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 678		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 679		mp->m_update_sb = true;
 680	}
 681
 682	/*
 683	 * Check if sb_agblocks is aligned at stripe boundary
 684	 * If sb_agblocks is NOT aligned turn off m_dalign since
 685	 * allocator alignment is within an ag, therefore ag has
 686	 * to be aligned at stripe boundary.
 687	 */
 688	error = xfs_update_alignment(mp);
 689	if (error)
 690		goto out;
 691
 692	xfs_alloc_compute_maxlevels(mp);
 693	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 694	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 695	xfs_ialloc_compute_maxlevels(mp);
 696	xfs_rmapbt_compute_maxlevels(mp);
 697	xfs_refcountbt_compute_maxlevels(mp);
 698
 699	xfs_set_maxicount(mp);
 700
 701	/* enable fail_at_unmount as default */
 702	mp->m_fail_unmount = 1;
 703
 704	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
 705	if (error)
 706		goto out;
 707
 708	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
 709			       &mp->m_kobj, "stats");
 710	if (error)
 711		goto out_remove_sysfs;
 712
 713	error = xfs_error_sysfs_init(mp);
 714	if (error)
 715		goto out_del_stats;
 716
 
 
 
 717
 718	error = xfs_uuid_mount(mp);
 719	if (error)
 720		goto out_remove_error_sysfs;
 721
 722	/*
 723	 * Set the minimum read and write sizes
 724	 */
 725	xfs_set_rw_sizes(mp);
 726
 727	/* set the low space thresholds for dynamic preallocation */
 728	xfs_set_low_space_thresholds(mp);
 729
 730	/*
 731	 * Set the inode cluster size.
 732	 * This may still be overridden by the file system
 733	 * block size if it is larger than the chosen cluster size.
 734	 *
 735	 * For v5 filesystems, scale the cluster size with the inode size to
 736	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
 737	 * has set the inode alignment value appropriately for larger cluster
 738	 * sizes.
 739	 */
 740	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
 741	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 742		int	new_size = mp->m_inode_cluster_size;
 743
 744		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
 745		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
 746			mp->m_inode_cluster_size = new_size;
 747	}
 748
 749	/*
 750	 * If enabled, sparse inode chunk alignment is expected to match the
 751	 * cluster size. Full inode chunk alignment must match the chunk size,
 752	 * but that is checked on sb read verification...
 753	 */
 754	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
 755	    mp->m_sb.sb_spino_align !=
 756			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
 757		xfs_warn(mp,
 758	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 759			 mp->m_sb.sb_spino_align,
 760			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
 761		error = -EINVAL;
 762		goto out_remove_uuid;
 763	}
 764
 765	/*
 766	 * Set inode alignment fields
 767	 */
 768	xfs_set_inoalignment(mp);
 769
 770	/*
 771	 * Check that the data (and log if separate) is an ok size.
 772	 */
 773	error = xfs_check_sizes(mp);
 774	if (error)
 775		goto out_remove_uuid;
 776
 777	/*
 778	 * Initialize realtime fields in the mount structure
 779	 */
 780	error = xfs_rtmount_init(mp);
 781	if (error) {
 782		xfs_warn(mp, "RT mount failed");
 783		goto out_remove_uuid;
 784	}
 785
 786	/*
 787	 *  Copies the low order bits of the timestamp and the randomly
 788	 *  set "sequence" number out of a UUID.
 789	 */
 790	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
 791
 792	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
 
 793
 794	error = xfs_da_mount(mp);
 795	if (error) {
 796		xfs_warn(mp, "Failed dir/attr init: %d", error);
 797		goto out_remove_uuid;
 798	}
 799
 800	/*
 801	 * Initialize the precomputed transaction reservations values.
 802	 */
 803	xfs_trans_init(mp);
 804
 805	/*
 806	 * Allocate and initialize the per-ag data.
 807	 */
 808	spin_lock_init(&mp->m_perag_lock);
 809	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
 810	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
 811	if (error) {
 812		xfs_warn(mp, "Failed per-ag init: %d", error);
 813		goto out_free_dir;
 814	}
 815
 816	if (!sbp->sb_logblocks) {
 817		xfs_warn(mp, "no log defined");
 818		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
 819		error = -EFSCORRUPTED;
 820		goto out_free_perag;
 821	}
 822
 823	/*
 824	 * Log's mount-time initialization. The first part of recovery can place
 825	 * some items on the AIL, to be handled when recovery is finished or
 826	 * cancelled.
 827	 */
 828	error = xfs_log_mount(mp, mp->m_logdev_targp,
 829			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 830			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 831	if (error) {
 832		xfs_warn(mp, "log mount failed");
 833		goto out_fail_wait;
 834	}
 835
 836	/*
 837	 * Now the log is mounted, we know if it was an unclean shutdown or
 838	 * not. If it was, with the first phase of recovery has completed, we
 839	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 840	 * but they are recovered transactionally in the second recovery phase
 841	 * later.
 842	 *
 843	 * Hence we can safely re-initialise incore superblock counters from
 844	 * the per-ag data. These may not be correct if the filesystem was not
 845	 * cleanly unmounted, so we need to wait for recovery to finish before
 846	 * doing this.
 847	 *
 848	 * If the filesystem was cleanly unmounted, then we can trust the
 849	 * values in the superblock to be correct and we don't need to do
 850	 * anything here.
 851	 *
 852	 * If we are currently making the filesystem, the initialisation will
 853	 * fail as the perag data is in an undefined state.
 854	 */
 855	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
 856	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
 857	     !mp->m_sb.sb_inprogress) {
 858		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
 859		if (error)
 860			goto out_log_dealloc;
 861	}
 862
 863	/*
 864	 * Get and sanity-check the root inode.
 865	 * Save the pointer to it in the mount structure.
 866	 */
 867	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
 868	if (error) {
 869		xfs_warn(mp, "failed to read root inode");
 870		goto out_log_dealloc;
 871	}
 872
 873	ASSERT(rip != NULL);
 874
 875	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
 876		xfs_warn(mp, "corrupted root inode %llu: not a directory",
 877			(unsigned long long)rip->i_ino);
 878		xfs_iunlock(rip, XFS_ILOCK_EXCL);
 879		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
 880				 mp);
 881		error = -EFSCORRUPTED;
 882		goto out_rele_rip;
 883	}
 884	mp->m_rootip = rip;	/* save it */
 885
 886	xfs_iunlock(rip, XFS_ILOCK_EXCL);
 887
 888	/*
 889	 * Initialize realtime inode pointers in the mount structure
 890	 */
 891	error = xfs_rtmount_inodes(mp);
 892	if (error) {
 893		/*
 894		 * Free up the root inode.
 895		 */
 896		xfs_warn(mp, "failed to read RT inodes");
 897		goto out_rele_rip;
 898	}
 899
 900	/*
 901	 * If this is a read-only mount defer the superblock updates until
 902	 * the next remount into writeable mode.  Otherwise we would never
 903	 * perform the update e.g. for the root filesystem.
 904	 */
 905	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 906		error = xfs_sync_sb(mp, false);
 907		if (error) {
 908			xfs_warn(mp, "failed to write sb changes");
 909			goto out_rtunmount;
 910		}
 911	}
 912
 913	/*
 914	 * Initialise the XFS quota management subsystem for this mount
 915	 */
 916	if (XFS_IS_QUOTA_RUNNING(mp)) {
 917		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
 918		if (error)
 919			goto out_rtunmount;
 920	} else {
 921		ASSERT(!XFS_IS_QUOTA_ON(mp));
 922
 923		/*
 924		 * If a file system had quotas running earlier, but decided to
 925		 * mount without -o uquota/pquota/gquota options, revoke the
 926		 * quotachecked license.
 927		 */
 928		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
 929			xfs_notice(mp, "resetting quota flags");
 930			error = xfs_mount_reset_sbqflags(mp);
 931			if (error)
 932				goto out_rtunmount;
 933		}
 934	}
 935
 936	/*
 937	 * During the second phase of log recovery, we need iget and
 938	 * iput to behave like they do for an active filesystem.
 939	 * xfs_fs_drop_inode needs to be able to prevent the deletion
 940	 * of inodes before we're done replaying log items on those
 941	 * inodes.
 942	 */
 943	mp->m_super->s_flags |= MS_ACTIVE;
 944
 945	/*
 946	 * Finish recovering the file system.  This part needed to be delayed
 947	 * until after the root and real-time bitmap inodes were consistently
 948	 * read in.
 949	 */
 950	error = xfs_log_mount_finish(mp);
 951	if (error) {
 952		xfs_warn(mp, "log mount finish failed");
 953		goto out_rtunmount;
 954	}
 955
 956	/*
 957	 * Now the log is fully replayed, we can transition to full read-only
 958	 * mode for read-only mounts. This will sync all the metadata and clean
 959	 * the log so that the recovery we just performed does not have to be
 960	 * replayed again on the next mount.
 961	 *
 962	 * We use the same quiesce mechanism as the rw->ro remount, as they are
 963	 * semantically identical operations.
 964	 */
 965	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
 966							XFS_MOUNT_RDONLY) {
 967		xfs_quiesce_attr(mp);
 968	}
 969
 970	/*
 971	 * Complete the quota initialisation, post-log-replay component.
 972	 */
 973	if (quotamount) {
 974		ASSERT(mp->m_qflags == 0);
 975		mp->m_qflags = quotaflags;
 976
 977		xfs_qm_mount_quotas(mp);
 978	}
 979
 980	/*
 981	 * Now we are mounted, reserve a small amount of unused space for
 982	 * privileged transactions. This is needed so that transaction
 983	 * space required for critical operations can dip into this pool
 984	 * when at ENOSPC. This is needed for operations like create with
 985	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
 986	 * are not allowed to use this reserved space.
 987	 *
 988	 * This may drive us straight to ENOSPC on mount, but that implies
 989	 * we were already there on the last unmount. Warn if this occurs.
 990	 */
 991	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 992		resblks = xfs_default_resblks(mp);
 993		error = xfs_reserve_blocks(mp, &resblks, NULL);
 994		if (error)
 995			xfs_warn(mp,
 996	"Unable to allocate reserve blocks. Continuing without reserve pool.");
 997
 998		/* Recover any CoW blocks that never got remapped. */
 999		error = xfs_reflink_recover_cow(mp);
1000		if (error) {
1001			xfs_err(mp,
1002	"Error %d recovering leftover CoW allocations.", error);
1003			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1004			goto out_quota;
1005		}
1006
1007		/* Reserve AG blocks for future btree expansion. */
1008		error = xfs_fs_reserve_ag_blocks(mp);
1009		if (error && error != -ENOSPC)
1010			goto out_agresv;
1011	}
1012
1013	return 0;
1014
1015 out_agresv:
1016	xfs_fs_unreserve_ag_blocks(mp);
1017 out_quota:
1018	xfs_qm_unmount_quotas(mp);
1019 out_rtunmount:
1020	mp->m_super->s_flags &= ~MS_ACTIVE;
1021	xfs_rtunmount_inodes(mp);
1022 out_rele_rip:
1023	IRELE(rip);
 
 
 
 
 
 
 
 
 
 
 
 
 
1024	cancel_delayed_work_sync(&mp->m_reclaim_work);
1025	xfs_reclaim_inodes(mp, SYNC_WAIT);
1026 out_log_dealloc:
1027	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1028	xfs_log_mount_cancel(mp);
1029 out_fail_wait:
1030	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1031		xfs_wait_buftarg(mp->m_logdev_targp);
1032	xfs_wait_buftarg(mp->m_ddev_targp);
1033 out_free_perag:
1034	xfs_free_perag(mp);
1035 out_free_dir:
1036	xfs_da_unmount(mp);
1037 out_remove_uuid:
1038	xfs_uuid_unmount(mp);
 
 
1039 out_remove_error_sysfs:
1040	xfs_error_sysfs_del(mp);
1041 out_del_stats:
1042	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1043 out_remove_sysfs:
1044	xfs_sysfs_del(&mp->m_kobj);
1045 out:
1046	return error;
1047}
1048
1049/*
1050 * This flushes out the inodes,dquots and the superblock, unmounts the
1051 * log and makes sure that incore structures are freed.
1052 */
1053void
1054xfs_unmountfs(
1055	struct xfs_mount	*mp)
1056{
1057	__uint64_t		resblks;
1058	int			error;
1059
1060	cancel_delayed_work_sync(&mp->m_eofblocks_work);
1061	cancel_delayed_work_sync(&mp->m_cowblocks_work);
1062
1063	xfs_fs_unreserve_ag_blocks(mp);
1064	xfs_qm_unmount_quotas(mp);
1065	xfs_rtunmount_inodes(mp);
1066	IRELE(mp->m_rootip);
1067
1068	/*
1069	 * We can potentially deadlock here if we have an inode cluster
1070	 * that has been freed has its buffer still pinned in memory because
1071	 * the transaction is still sitting in a iclog. The stale inodes
1072	 * on that buffer will have their flush locks held until the
1073	 * transaction hits the disk and the callbacks run. the inode
1074	 * flush takes the flush lock unconditionally and with nothing to
1075	 * push out the iclog we will never get that unlocked. hence we
1076	 * need to force the log first.
1077	 */
1078	xfs_log_force(mp, XFS_LOG_SYNC);
1079
1080	/*
 
 
 
 
 
 
 
1081	 * We now need to tell the world we are unmounting. This will allow
1082	 * us to detect that the filesystem is going away and we should error
1083	 * out anything that we have been retrying in the background. This will
1084	 * prevent neverending retries in AIL pushing from hanging the unmount.
1085	 */
1086	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1087
1088	/*
1089	 * Flush all pending changes from the AIL.
1090	 */
1091	xfs_ail_push_all_sync(mp->m_ail);
1092
1093	/*
1094	 * And reclaim all inodes.  At this point there should be no dirty
1095	 * inodes and none should be pinned or locked, but use synchronous
1096	 * reclaim just to be sure. We can stop background inode reclaim
1097	 * here as well if it is still running.
1098	 */
1099	cancel_delayed_work_sync(&mp->m_reclaim_work);
1100	xfs_reclaim_inodes(mp, SYNC_WAIT);
1101
1102	xfs_qm_unmount(mp);
1103
1104	/*
1105	 * Unreserve any blocks we have so that when we unmount we don't account
1106	 * the reserved free space as used. This is really only necessary for
1107	 * lazy superblock counting because it trusts the incore superblock
1108	 * counters to be absolutely correct on clean unmount.
1109	 *
1110	 * We don't bother correcting this elsewhere for lazy superblock
1111	 * counting because on mount of an unclean filesystem we reconstruct the
1112	 * correct counter value and this is irrelevant.
1113	 *
1114	 * For non-lazy counter filesystems, this doesn't matter at all because
1115	 * we only every apply deltas to the superblock and hence the incore
1116	 * value does not matter....
1117	 */
1118	resblks = 0;
1119	error = xfs_reserve_blocks(mp, &resblks, NULL);
1120	if (error)
1121		xfs_warn(mp, "Unable to free reserved block pool. "
1122				"Freespace may not be correct on next mount.");
1123
1124	error = xfs_log_sbcount(mp);
1125	if (error)
1126		xfs_warn(mp, "Unable to update superblock counters. "
1127				"Freespace may not be correct on next mount.");
1128
1129
1130	xfs_log_unmount(mp);
1131	xfs_da_unmount(mp);
1132	xfs_uuid_unmount(mp);
1133
1134#if defined(DEBUG)
1135	xfs_errortag_clearall(mp, 0);
1136#endif
1137	xfs_free_perag(mp);
1138
 
1139	xfs_error_sysfs_del(mp);
1140	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1141	xfs_sysfs_del(&mp->m_kobj);
1142}
1143
1144/*
1145 * Determine whether modifications can proceed. The caller specifies the minimum
1146 * freeze level for which modifications should not be allowed. This allows
1147 * certain operations to proceed while the freeze sequence is in progress, if
1148 * necessary.
1149 */
1150bool
1151xfs_fs_writable(
1152	struct xfs_mount	*mp,
1153	int			level)
1154{
1155	ASSERT(level > SB_UNFROZEN);
1156	if ((mp->m_super->s_writers.frozen >= level) ||
1157	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1158		return false;
1159
1160	return true;
1161}
1162
1163/*
1164 * xfs_log_sbcount
1165 *
1166 * Sync the superblock counters to disk.
1167 *
1168 * Note this code can be called during the process of freezing, so we use the
1169 * transaction allocator that does not block when the transaction subsystem is
1170 * in its frozen state.
1171 */
1172int
1173xfs_log_sbcount(xfs_mount_t *mp)
1174{
1175	/* allow this to proceed during the freeze sequence... */
1176	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1177		return 0;
1178
1179	/*
1180	 * we don't need to do this if we are updating the superblock
1181	 * counters on every modification.
1182	 */
1183	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1184		return 0;
1185
1186	return xfs_sync_sb(mp, true);
1187}
1188
1189/*
1190 * Deltas for the inode count are +/-64, hence we use a large batch size
1191 * of 128 so we don't need to take the counter lock on every update.
1192 */
1193#define XFS_ICOUNT_BATCH	128
1194int
1195xfs_mod_icount(
1196	struct xfs_mount	*mp,
1197	int64_t			delta)
1198{
1199	__percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1200	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1201		ASSERT(0);
1202		percpu_counter_add(&mp->m_icount, -delta);
1203		return -EINVAL;
1204	}
1205	return 0;
1206}
1207
1208int
1209xfs_mod_ifree(
1210	struct xfs_mount	*mp,
1211	int64_t			delta)
1212{
1213	percpu_counter_add(&mp->m_ifree, delta);
1214	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1215		ASSERT(0);
1216		percpu_counter_add(&mp->m_ifree, -delta);
1217		return -EINVAL;
1218	}
1219	return 0;
1220}
1221
1222/*
1223 * Deltas for the block count can vary from 1 to very large, but lock contention
1224 * only occurs on frequent small block count updates such as in the delayed
1225 * allocation path for buffered writes (page a time updates). Hence we set
1226 * a large batch count (1024) to minimise global counter updates except when
1227 * we get near to ENOSPC and we have to be very accurate with our updates.
1228 */
1229#define XFS_FDBLOCKS_BATCH	1024
1230int
1231xfs_mod_fdblocks(
1232	struct xfs_mount	*mp,
1233	int64_t			delta,
1234	bool			rsvd)
1235{
1236	int64_t			lcounter;
1237	long long		res_used;
1238	s32			batch;
1239
1240	if (delta > 0) {
1241		/*
1242		 * If the reserve pool is depleted, put blocks back into it
1243		 * first. Most of the time the pool is full.
1244		 */
1245		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1246			percpu_counter_add(&mp->m_fdblocks, delta);
1247			return 0;
1248		}
1249
1250		spin_lock(&mp->m_sb_lock);
1251		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1252
1253		if (res_used > delta) {
1254			mp->m_resblks_avail += delta;
1255		} else {
1256			delta -= res_used;
1257			mp->m_resblks_avail = mp->m_resblks;
1258			percpu_counter_add(&mp->m_fdblocks, delta);
1259		}
1260		spin_unlock(&mp->m_sb_lock);
1261		return 0;
1262	}
1263
1264	/*
1265	 * Taking blocks away, need to be more accurate the closer we
1266	 * are to zero.
1267	 *
1268	 * If the counter has a value of less than 2 * max batch size,
1269	 * then make everything serialise as we are real close to
1270	 * ENOSPC.
1271	 */
1272	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1273				     XFS_FDBLOCKS_BATCH) < 0)
1274		batch = 1;
1275	else
1276		batch = XFS_FDBLOCKS_BATCH;
1277
1278	__percpu_counter_add(&mp->m_fdblocks, delta, batch);
1279	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1280				     XFS_FDBLOCKS_BATCH) >= 0) {
1281		/* we had space! */
1282		return 0;
1283	}
1284
1285	/*
1286	 * lock up the sb for dipping into reserves before releasing the space
1287	 * that took us to ENOSPC.
1288	 */
1289	spin_lock(&mp->m_sb_lock);
1290	percpu_counter_add(&mp->m_fdblocks, -delta);
1291	if (!rsvd)
1292		goto fdblocks_enospc;
1293
1294	lcounter = (long long)mp->m_resblks_avail + delta;
1295	if (lcounter >= 0) {
1296		mp->m_resblks_avail = lcounter;
1297		spin_unlock(&mp->m_sb_lock);
1298		return 0;
1299	}
1300	printk_once(KERN_WARNING
1301		"Filesystem \"%s\": reserve blocks depleted! "
1302		"Consider increasing reserve pool size.",
1303		mp->m_fsname);
1304fdblocks_enospc:
1305	spin_unlock(&mp->m_sb_lock);
1306	return -ENOSPC;
1307}
1308
1309int
1310xfs_mod_frextents(
1311	struct xfs_mount	*mp,
1312	int64_t			delta)
1313{
1314	int64_t			lcounter;
1315	int			ret = 0;
1316
1317	spin_lock(&mp->m_sb_lock);
1318	lcounter = mp->m_sb.sb_frextents + delta;
1319	if (lcounter < 0)
1320		ret = -ENOSPC;
1321	else
1322		mp->m_sb.sb_frextents = lcounter;
1323	spin_unlock(&mp->m_sb_lock);
1324	return ret;
1325}
1326
1327/*
1328 * xfs_getsb() is called to obtain the buffer for the superblock.
1329 * The buffer is returned locked and read in from disk.
1330 * The buffer should be released with a call to xfs_brelse().
1331 *
1332 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1333 * the superblock buffer if it can be locked without sleeping.
1334 * If it can't then we'll return NULL.
1335 */
1336struct xfs_buf *
1337xfs_getsb(
1338	struct xfs_mount	*mp,
1339	int			flags)
1340{
1341	struct xfs_buf		*bp = mp->m_sb_bp;
1342
1343	if (!xfs_buf_trylock(bp)) {
1344		if (flags & XBF_TRYLOCK)
1345			return NULL;
1346		xfs_buf_lock(bp);
1347	}
1348
1349	xfs_buf_hold(bp);
1350	ASSERT(bp->b_flags & XBF_DONE);
1351	return bp;
1352}
1353
1354/*
1355 * Used to free the superblock along various error paths.
1356 */
1357void
1358xfs_freesb(
1359	struct xfs_mount	*mp)
1360{
1361	struct xfs_buf		*bp = mp->m_sb_bp;
1362
1363	xfs_buf_lock(bp);
1364	mp->m_sb_bp = NULL;
1365	xfs_buf_relse(bp);
1366}
1367
1368/*
1369 * If the underlying (data/log/rt) device is readonly, there are some
1370 * operations that cannot proceed.
1371 */
1372int
1373xfs_dev_is_read_only(
1374	struct xfs_mount	*mp,
1375	char			*message)
1376{
1377	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1378	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1379	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1380		xfs_notice(mp, "%s required on read-only device.", message);
1381		xfs_notice(mp, "write access unavailable, cannot proceed.");
1382		return -EROFS;
1383	}
1384	return 0;
1385}