Linux Audio

Check our new training course

Loading...
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}
v5.4
   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_error.h"
  25#include "xfs_quota.h"
  26#include "xfs_fsops.h"
 
  27#include "xfs_icache.h"
  28#include "xfs_sysfs.h"
  29#include "xfs_rmap_btree.h"
  30#include "xfs_refcount_btree.h"
  31#include "xfs_reflink.h"
  32#include "xfs_extent_busy.h"
  33#include "xfs_health.h"
  34
  35
  36static DEFINE_MUTEX(xfs_uuid_table_mutex);
  37static int xfs_uuid_table_size;
  38static uuid_t *xfs_uuid_table;
  39
  40void
  41xfs_uuid_table_free(void)
  42{
  43	if (xfs_uuid_table_size == 0)
  44		return;
  45	kmem_free(xfs_uuid_table);
  46	xfs_uuid_table = NULL;
  47	xfs_uuid_table_size = 0;
  48}
  49
  50/*
  51 * See if the UUID is unique among mounted XFS filesystems.
  52 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  53 */
  54STATIC int
  55xfs_uuid_mount(
  56	struct xfs_mount	*mp)
  57{
  58	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  59	int			hole, i;
  60
  61	/* Publish UUID in struct super_block */
  62	uuid_copy(&mp->m_super->s_uuid, uuid);
  63
  64	if (mp->m_flags & XFS_MOUNT_NOUUID)
  65		return 0;
  66
  67	if (uuid_is_null(uuid)) {
  68		xfs_warn(mp, "Filesystem has null UUID - can't mount");
  69		return -EINVAL;
  70	}
  71
  72	mutex_lock(&xfs_uuid_table_mutex);
  73	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  74		if (uuid_is_null(&xfs_uuid_table[i])) {
  75			hole = i;
  76			continue;
  77		}
  78		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  79			goto out_duplicate;
  80	}
  81
  82	if (hole < 0) {
  83		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
  84			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
  85			0);
  86		hole = xfs_uuid_table_size++;
  87	}
  88	xfs_uuid_table[hole] = *uuid;
  89	mutex_unlock(&xfs_uuid_table_mutex);
  90
  91	return 0;
  92
  93 out_duplicate:
  94	mutex_unlock(&xfs_uuid_table_mutex);
  95	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
  96	return -EINVAL;
  97}
  98
  99STATIC void
 100xfs_uuid_unmount(
 101	struct xfs_mount	*mp)
 102{
 103	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 104	int			i;
 105
 106	if (mp->m_flags & XFS_MOUNT_NOUUID)
 107		return;
 108
 109	mutex_lock(&xfs_uuid_table_mutex);
 110	for (i = 0; i < xfs_uuid_table_size; i++) {
 111		if (uuid_is_null(&xfs_uuid_table[i]))
 112			continue;
 113		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 114			continue;
 115		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 116		break;
 117	}
 118	ASSERT(i < xfs_uuid_table_size);
 119	mutex_unlock(&xfs_uuid_table_mutex);
 120}
 121
 122
 123STATIC void
 124__xfs_free_perag(
 125	struct rcu_head	*head)
 126{
 127	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 128
 129	ASSERT(atomic_read(&pag->pag_ref) == 0);
 130	kmem_free(pag);
 131}
 132
 133/*
 134 * Free up the per-ag resources associated with the mount structure.
 135 */
 136STATIC void
 137xfs_free_perag(
 138	xfs_mount_t	*mp)
 139{
 140	xfs_agnumber_t	agno;
 141	struct xfs_perag *pag;
 142
 143	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
 144		spin_lock(&mp->m_perag_lock);
 145		pag = radix_tree_delete(&mp->m_perag_tree, agno);
 146		spin_unlock(&mp->m_perag_lock);
 147		ASSERT(pag);
 148		ASSERT(atomic_read(&pag->pag_ref) == 0);
 149		xfs_iunlink_destroy(pag);
 150		xfs_buf_hash_destroy(pag);
 151		mutex_destroy(&pag->pag_ici_reclaim_lock);
 152		call_rcu(&pag->rcu_head, __xfs_free_perag);
 153	}
 154}
 155
 156/*
 157 * Check size of device based on the (data/realtime) block count.
 158 * Note: this check is used by the growfs code as well as mount.
 159 */
 160int
 161xfs_sb_validate_fsb_count(
 162	xfs_sb_t	*sbp,
 163	uint64_t	nblocks)
 164{
 165	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 166	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 167
 168	/* Limited by ULONG_MAX of page cache index */
 169	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 170		return -EFBIG;
 171	return 0;
 172}
 173
 174int
 175xfs_initialize_perag(
 176	xfs_mount_t	*mp,
 177	xfs_agnumber_t	agcount,
 178	xfs_agnumber_t	*maxagi)
 179{
 180	xfs_agnumber_t	index;
 181	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
 182	xfs_perag_t	*pag;
 183	int		error = -ENOMEM;
 184
 185	/*
 186	 * Walk the current per-ag tree so we don't try to initialise AGs
 187	 * that already exist (growfs case). Allocate and insert all the
 188	 * AGs we don't find ready for initialisation.
 189	 */
 190	for (index = 0; index < agcount; index++) {
 191		pag = xfs_perag_get(mp, index);
 192		if (pag) {
 193			xfs_perag_put(pag);
 194			continue;
 195		}
 196
 197		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
 198		if (!pag)
 199			goto out_unwind_new_pags;
 200		pag->pag_agno = index;
 201		pag->pag_mount = mp;
 202		spin_lock_init(&pag->pag_ici_lock);
 203		mutex_init(&pag->pag_ici_reclaim_lock);
 204		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 205		if (xfs_buf_hash_init(pag))
 206			goto out_free_pag;
 207		init_waitqueue_head(&pag->pagb_wait);
 208		spin_lock_init(&pag->pagb_lock);
 209		pag->pagb_count = 0;
 210		pag->pagb_tree = RB_ROOT;
 211
 212		if (radix_tree_preload(GFP_NOFS))
 213			goto out_hash_destroy;
 214
 215		spin_lock(&mp->m_perag_lock);
 216		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
 217			WARN_ON_ONCE(1);
 218			spin_unlock(&mp->m_perag_lock);
 219			radix_tree_preload_end();
 220			error = -EEXIST;
 221			goto out_hash_destroy;
 222		}
 223		spin_unlock(&mp->m_perag_lock);
 224		radix_tree_preload_end();
 225		/* first new pag is fully initialized */
 226		if (first_initialised == NULLAGNUMBER)
 227			first_initialised = index;
 228		error = xfs_iunlink_init(pag);
 229		if (error)
 230			goto out_hash_destroy;
 231		spin_lock_init(&pag->pag_state_lock);
 232	}
 233
 234	index = xfs_set_inode_alloc(mp, agcount);
 235
 236	if (maxagi)
 237		*maxagi = index;
 238
 239	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
 240	return 0;
 241
 242out_hash_destroy:
 243	xfs_buf_hash_destroy(pag);
 244out_free_pag:
 245	mutex_destroy(&pag->pag_ici_reclaim_lock);
 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		xfs_iunlink_destroy(pag);
 255		mutex_destroy(&pag->pag_ici_reclaim_lock);
 256		kmem_free(pag);
 257	}
 258	return error;
 259}
 260
 261/*
 262 * xfs_readsb
 263 *
 264 * Does the initial read of the superblock.
 265 */
 266int
 267xfs_readsb(
 268	struct xfs_mount *mp,
 269	int		flags)
 270{
 271	unsigned int	sector_size;
 272	struct xfs_buf	*bp;
 273	struct xfs_sb	*sbp = &mp->m_sb;
 274	int		error;
 275	int		loud = !(flags & XFS_MFSI_QUIET);
 276	const struct xfs_buf_ops *buf_ops;
 277
 278	ASSERT(mp->m_sb_bp == NULL);
 279	ASSERT(mp->m_ddev_targp != NULL);
 280
 281	/*
 282	 * For the initial read, we must guess at the sector
 283	 * size based on the block device.  It's enough to
 284	 * get the sb_sectsize out of the superblock and
 285	 * then reread with the proper length.
 286	 * We don't verify it yet, because it may not be complete.
 287	 */
 288	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 289	buf_ops = NULL;
 290
 291	/*
 292	 * Allocate a (locked) buffer to hold the superblock. This will be kept
 293	 * around at all times to optimize access to the superblock. Therefore,
 294	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
 295	 * elevated.
 296	 */
 297reread:
 298	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 299				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
 300				      buf_ops);
 301	if (error) {
 302		if (loud)
 303			xfs_warn(mp, "SB validate failed with error %d.", error);
 304		/* bad CRC means corrupted metadata */
 305		if (error == -EFSBADCRC)
 306			error = -EFSCORRUPTED;
 307		return error;
 308	}
 309
 310	/*
 311	 * Initialize the mount structure from the superblock.
 312	 */
 313	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
 314
 315	/*
 316	 * If we haven't validated the superblock, do so now before we try
 317	 * to check the sector size and reread the superblock appropriately.
 318	 */
 319	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 320		if (loud)
 321			xfs_warn(mp, "Invalid superblock magic number");
 322		error = -EINVAL;
 323		goto release_buf;
 324	}
 325
 326	/*
 327	 * We must be able to do sector-sized and sector-aligned IO.
 328	 */
 329	if (sector_size > sbp->sb_sectsize) {
 330		if (loud)
 331			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 332				sector_size, sbp->sb_sectsize);
 333		error = -ENOSYS;
 334		goto release_buf;
 335	}
 336
 337	if (buf_ops == NULL) {
 338		/*
 339		 * Re-read the superblock so the buffer is correctly sized,
 340		 * and properly verified.
 341		 */
 342		xfs_buf_relse(bp);
 343		sector_size = sbp->sb_sectsize;
 344		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 345		goto reread;
 346	}
 347
 348	xfs_reinit_percpu_counters(mp);
 349
 350	/* no need to be quiet anymore, so reset the buf ops */
 351	bp->b_ops = &xfs_sb_buf_ops;
 352
 353	mp->m_sb_bp = bp;
 354	xfs_buf_unlock(bp);
 355	return 0;
 356
 357release_buf:
 358	xfs_buf_relse(bp);
 359	return error;
 360}
 361
 362/*
 363 * Update alignment values based on mount options and sb values
 364 */
 365STATIC int
 366xfs_update_alignment(xfs_mount_t *mp)
 367{
 368	xfs_sb_t	*sbp = &(mp->m_sb);
 369
 370	if (mp->m_dalign) {
 371		/*
 372		 * If stripe unit and stripe width are not multiples
 373		 * of the fs blocksize turn off alignment.
 374		 */
 375		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 376		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 377			xfs_warn(mp,
 378		"alignment check failed: sunit/swidth vs. blocksize(%d)",
 379				sbp->sb_blocksize);
 380			return -EINVAL;
 381		} else {
 382			/*
 383			 * Convert the stripe unit and width to FSBs.
 384			 */
 385			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 386			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 387				xfs_warn(mp,
 388			"alignment check failed: sunit/swidth vs. agsize(%d)",
 389					 sbp->sb_agblocks);
 390				return -EINVAL;
 391			} else if (mp->m_dalign) {
 392				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 393			} else {
 394				xfs_warn(mp,
 395			"alignment check failed: sunit(%d) less than bsize(%d)",
 396					 mp->m_dalign, sbp->sb_blocksize);
 397				return -EINVAL;
 398			}
 399		}
 400
 401		/*
 402		 * Update superblock with new values
 403		 * and log changes
 404		 */
 405		if (xfs_sb_version_hasdalign(sbp)) {
 406			if (sbp->sb_unit != mp->m_dalign) {
 407				sbp->sb_unit = mp->m_dalign;
 408				mp->m_update_sb = true;
 409			}
 410			if (sbp->sb_width != mp->m_swidth) {
 411				sbp->sb_width = mp->m_swidth;
 412				mp->m_update_sb = true;
 413			}
 414		} else {
 415			xfs_warn(mp,
 416	"cannot change alignment: superblock does not support data alignment");
 417			return -EINVAL;
 418		}
 419	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 420		    xfs_sb_version_hasdalign(&mp->m_sb)) {
 421			mp->m_dalign = sbp->sb_unit;
 422			mp->m_swidth = sbp->sb_width;
 423	}
 424
 425	return 0;
 426}
 427
 428/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 429 * Set the default minimum read and write sizes unless
 430 * already specified in a mount option.
 431 * We use smaller I/O sizes when the file system
 432 * is being used for NFS service (wsync mount option).
 433 */
 434STATIC void
 435xfs_set_rw_sizes(xfs_mount_t *mp)
 436{
 437	xfs_sb_t	*sbp = &(mp->m_sb);
 438	int		readio_log, writeio_log;
 439
 440	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 441		if (mp->m_flags & XFS_MOUNT_WSYNC) {
 442			readio_log = XFS_WSYNC_READIO_LOG;
 443			writeio_log = XFS_WSYNC_WRITEIO_LOG;
 444		} else {
 445			readio_log = XFS_READIO_LOG_LARGE;
 446			writeio_log = XFS_WRITEIO_LOG_LARGE;
 447		}
 448	} else {
 449		readio_log = mp->m_readio_log;
 450		writeio_log = mp->m_writeio_log;
 451	}
 452
 453	if (sbp->sb_blocklog > readio_log) {
 454		mp->m_readio_log = sbp->sb_blocklog;
 455	} else {
 456		mp->m_readio_log = readio_log;
 457	}
 458	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 459	if (sbp->sb_blocklog > writeio_log) {
 460		mp->m_writeio_log = sbp->sb_blocklog;
 461	} else {
 462		mp->m_writeio_log = writeio_log;
 463	}
 464	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 465}
 466
 467/*
 468 * precalculate the low space thresholds for dynamic speculative preallocation.
 469 */
 470void
 471xfs_set_low_space_thresholds(
 472	struct xfs_mount	*mp)
 473{
 474	int i;
 475
 476	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 477		uint64_t space = mp->m_sb.sb_dblocks;
 478
 479		do_div(space, 100);
 480		mp->m_low_space[i] = space * (i + 1);
 481	}
 482}
 483
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 484/*
 485 * Check that the data (and log if separate) is an ok size.
 486 */
 487STATIC int
 488xfs_check_sizes(
 489	struct xfs_mount *mp)
 490{
 491	struct xfs_buf	*bp;
 492	xfs_daddr_t	d;
 493	int		error;
 494
 495	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 496	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 497		xfs_warn(mp, "filesystem size mismatch detected");
 498		return -EFBIG;
 499	}
 500	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 501					d - XFS_FSS_TO_BB(mp, 1),
 502					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 503	if (error) {
 504		xfs_warn(mp, "last sector read failed");
 505		return error;
 506	}
 507	xfs_buf_relse(bp);
 508
 509	if (mp->m_logdev_targp == mp->m_ddev_targp)
 510		return 0;
 511
 512	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 513	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 514		xfs_warn(mp, "log size mismatch detected");
 515		return -EFBIG;
 516	}
 517	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 518					d - XFS_FSB_TO_BB(mp, 1),
 519					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
 520	if (error) {
 521		xfs_warn(mp, "log device read failed");
 522		return error;
 523	}
 524	xfs_buf_relse(bp);
 525	return 0;
 526}
 527
 528/*
 529 * Clear the quotaflags in memory and in the superblock.
 530 */
 531int
 532xfs_mount_reset_sbqflags(
 533	struct xfs_mount	*mp)
 534{
 535	mp->m_qflags = 0;
 536
 537	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 538	if (mp->m_sb.sb_qflags == 0)
 539		return 0;
 540	spin_lock(&mp->m_sb_lock);
 541	mp->m_sb.sb_qflags = 0;
 542	spin_unlock(&mp->m_sb_lock);
 543
 544	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 545		return 0;
 546
 547	return xfs_sync_sb(mp, false);
 548}
 549
 550uint64_t
 551xfs_default_resblks(xfs_mount_t *mp)
 552{
 553	uint64_t resblks;
 554
 555	/*
 556	 * We default to 5% or 8192 fsbs of space reserved, whichever is
 557	 * smaller.  This is intended to cover concurrent allocation
 558	 * transactions when we initially hit enospc. These each require a 4
 559	 * block reservation. Hence by default we cover roughly 2000 concurrent
 560	 * allocation reservations.
 561	 */
 562	resblks = mp->m_sb.sb_dblocks;
 563	do_div(resblks, 20);
 564	resblks = min_t(uint64_t, resblks, 8192);
 565	return resblks;
 566}
 567
 568/* Ensure the summary counts are correct. */
 569STATIC int
 570xfs_check_summary_counts(
 571	struct xfs_mount	*mp)
 572{
 573	/*
 574	 * The AG0 superblock verifier rejects in-progress filesystems,
 575	 * so we should never see the flag set this far into mounting.
 576	 */
 577	if (mp->m_sb.sb_inprogress) {
 578		xfs_err(mp, "sb_inprogress set after log recovery??");
 579		WARN_ON(1);
 580		return -EFSCORRUPTED;
 581	}
 582
 583	/*
 584	 * Now the log is mounted, we know if it was an unclean shutdown or
 585	 * not. If it was, with the first phase of recovery has completed, we
 586	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 587	 * but they are recovered transactionally in the second recovery phase
 588	 * later.
 589	 *
 590	 * If the log was clean when we mounted, we can check the summary
 591	 * counters.  If any of them are obviously incorrect, we can recompute
 592	 * them from the AGF headers in the next step.
 593	 */
 594	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
 595	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
 596	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
 597	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
 598		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
 599
 600	/*
 601	 * We can safely re-initialise incore superblock counters from the
 602	 * per-ag data. These may not be correct if the filesystem was not
 603	 * cleanly unmounted, so we waited for recovery to finish before doing
 604	 * this.
 605	 *
 606	 * If the filesystem was cleanly unmounted or the previous check did
 607	 * not flag anything weird, then we can trust the values in the
 608	 * superblock to be correct and we don't need to do anything here.
 609	 * Otherwise, recalculate the summary counters.
 610	 */
 611	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 612	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
 613	    !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
 614		return 0;
 615
 616	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
 617}
 618
 619/*
 620 * This function does the following on an initial mount of a file system:
 621 *	- reads the superblock from disk and init the mount struct
 622 *	- if we're a 32-bit kernel, do a size check on the superblock
 623 *		so we don't mount terabyte filesystems
 624 *	- init mount struct realtime fields
 625 *	- allocate inode hash table for fs
 626 *	- init directory manager
 627 *	- perform recovery and init the log manager
 628 */
 629int
 630xfs_mountfs(
 631	struct xfs_mount	*mp)
 632{
 633	struct xfs_sb		*sbp = &(mp->m_sb);
 634	struct xfs_inode	*rip;
 635	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
 636	uint64_t		resblks;
 637	uint			quotamount = 0;
 638	uint			quotaflags = 0;
 639	int			error = 0;
 640
 641	xfs_sb_mount_common(mp, sbp);
 642
 643	/*
 644	 * Check for a mismatched features2 values.  Older kernels read & wrote
 645	 * into the wrong sb offset for sb_features2 on some platforms due to
 646	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 647	 * which made older superblock reading/writing routines swap it as a
 648	 * 64-bit value.
 649	 *
 650	 * For backwards compatibility, we make both slots equal.
 651	 *
 652	 * If we detect a mismatched field, we OR the set bits into the existing
 653	 * features2 field in case it has already been modified; we don't want
 654	 * to lose any features.  We then update the bad location with the ORed
 655	 * value so that older kernels will see any features2 flags. The
 656	 * superblock writeback code ensures the new sb_features2 is copied to
 657	 * sb_bad_features2 before it is logged or written to disk.
 658	 */
 659	if (xfs_sb_has_mismatched_features2(sbp)) {
 660		xfs_warn(mp, "correcting sb_features alignment problem");
 661		sbp->sb_features2 |= sbp->sb_bad_features2;
 662		mp->m_update_sb = true;
 663
 664		/*
 665		 * Re-check for ATTR2 in case it was found in bad_features2
 666		 * slot.
 667		 */
 668		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 669		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
 670			mp->m_flags |= XFS_MOUNT_ATTR2;
 671	}
 672
 673	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 674	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 675		xfs_sb_version_removeattr2(&mp->m_sb);
 676		mp->m_update_sb = true;
 677
 678		/* update sb_versionnum for the clearing of the morebits */
 679		if (!sbp->sb_features2)
 680			mp->m_update_sb = true;
 681	}
 682
 683	/* always use v2 inodes by default now */
 684	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 685		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 686		mp->m_update_sb = true;
 687	}
 688
 689	/*
 690	 * Check if sb_agblocks is aligned at stripe boundary
 691	 * If sb_agblocks is NOT aligned turn off m_dalign since
 692	 * allocator alignment is within an ag, therefore ag has
 693	 * to be aligned at stripe boundary.
 694	 */
 695	error = xfs_update_alignment(mp);
 696	if (error)
 697		goto out;
 698
 699	xfs_alloc_compute_maxlevels(mp);
 700	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 701	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 702	xfs_ialloc_setup_geometry(mp);
 703	xfs_rmapbt_compute_maxlevels(mp);
 704	xfs_refcountbt_compute_maxlevels(mp);
 705
 
 
 706	/* enable fail_at_unmount as default */
 707	mp->m_fail_unmount = true;
 708
 709	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
 710	if (error)
 711		goto out;
 712
 713	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
 714			       &mp->m_kobj, "stats");
 715	if (error)
 716		goto out_remove_sysfs;
 717
 718	error = xfs_error_sysfs_init(mp);
 719	if (error)
 720		goto out_del_stats;
 721
 722	error = xfs_errortag_init(mp);
 723	if (error)
 724		goto out_remove_error_sysfs;
 725
 726	error = xfs_uuid_mount(mp);
 727	if (error)
 728		goto out_remove_errortag;
 729
 730	/*
 731	 * Set the minimum read and write sizes
 732	 */
 733	xfs_set_rw_sizes(mp);
 734
 735	/* set the low space thresholds for dynamic preallocation */
 736	xfs_set_low_space_thresholds(mp);
 737
 738	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 739	 * If enabled, sparse inode chunk alignment is expected to match the
 740	 * cluster size. Full inode chunk alignment must match the chunk size,
 741	 * but that is checked on sb read verification...
 742	 */
 743	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
 744	    mp->m_sb.sb_spino_align !=
 745			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
 746		xfs_warn(mp,
 747	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 748			 mp->m_sb.sb_spino_align,
 749			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
 750		error = -EINVAL;
 751		goto out_remove_uuid;
 752	}
 753
 754	/*
 
 
 
 
 
 755	 * Check that the data (and log if separate) is an ok size.
 756	 */
 757	error = xfs_check_sizes(mp);
 758	if (error)
 759		goto out_remove_uuid;
 760
 761	/*
 762	 * Initialize realtime fields in the mount structure
 763	 */
 764	error = xfs_rtmount_init(mp);
 765	if (error) {
 766		xfs_warn(mp, "RT mount failed");
 767		goto out_remove_uuid;
 768	}
 769
 770	/*
 771	 *  Copies the low order bits of the timestamp and the randomly
 772	 *  set "sequence" number out of a UUID.
 773	 */
 774	mp->m_fixedfsid[0] =
 775		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
 776		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
 777	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
 778
 779	error = xfs_da_mount(mp);
 780	if (error) {
 781		xfs_warn(mp, "Failed dir/attr init: %d", error);
 782		goto out_remove_uuid;
 783	}
 784
 785	/*
 786	 * Initialize the precomputed transaction reservations values.
 787	 */
 788	xfs_trans_init(mp);
 789
 790	/*
 791	 * Allocate and initialize the per-ag data.
 792	 */
 
 
 793	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
 794	if (error) {
 795		xfs_warn(mp, "Failed per-ag init: %d", error);
 796		goto out_free_dir;
 797	}
 798
 799	if (!sbp->sb_logblocks) {
 800		xfs_warn(mp, "no log defined");
 801		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
 802		error = -EFSCORRUPTED;
 803		goto out_free_perag;
 804	}
 805
 806	/*
 807	 * Log's mount-time initialization. The first part of recovery can place
 808	 * some items on the AIL, to be handled when recovery is finished or
 809	 * cancelled.
 810	 */
 811	error = xfs_log_mount(mp, mp->m_logdev_targp,
 812			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 813			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 814	if (error) {
 815		xfs_warn(mp, "log mount failed");
 816		goto out_fail_wait;
 817	}
 818
 819	/* Make sure the summary counts are ok. */
 820	error = xfs_check_summary_counts(mp);
 821	if (error)
 822		goto out_log_dealloc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 823
 824	/*
 825	 * Get and sanity-check the root inode.
 826	 * Save the pointer to it in the mount structure.
 827	 */
 828	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
 829			 XFS_ILOCK_EXCL, &rip);
 830	if (error) {
 831		xfs_warn(mp,
 832			"Failed to read root inode 0x%llx, error %d",
 833			sbp->sb_rootino, -error);
 834		goto out_log_dealloc;
 835	}
 836
 837	ASSERT(rip != NULL);
 838
 839	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
 840		xfs_warn(mp, "corrupted root inode %llu: not a directory",
 841			(unsigned long long)rip->i_ino);
 842		xfs_iunlock(rip, XFS_ILOCK_EXCL);
 843		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
 844				 mp);
 845		error = -EFSCORRUPTED;
 846		goto out_rele_rip;
 847	}
 848	mp->m_rootip = rip;	/* save it */
 849
 850	xfs_iunlock(rip, XFS_ILOCK_EXCL);
 851
 852	/*
 853	 * Initialize realtime inode pointers in the mount structure
 854	 */
 855	error = xfs_rtmount_inodes(mp);
 856	if (error) {
 857		/*
 858		 * Free up the root inode.
 859		 */
 860		xfs_warn(mp, "failed to read RT inodes");
 861		goto out_rele_rip;
 862	}
 863
 864	/*
 865	 * If this is a read-only mount defer the superblock updates until
 866	 * the next remount into writeable mode.  Otherwise we would never
 867	 * perform the update e.g. for the root filesystem.
 868	 */
 869	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 870		error = xfs_sync_sb(mp, false);
 871		if (error) {
 872			xfs_warn(mp, "failed to write sb changes");
 873			goto out_rtunmount;
 874		}
 875	}
 876
 877	/*
 878	 * Initialise the XFS quota management subsystem for this mount
 879	 */
 880	if (XFS_IS_QUOTA_RUNNING(mp)) {
 881		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
 882		if (error)
 883			goto out_rtunmount;
 884	} else {
 885		ASSERT(!XFS_IS_QUOTA_ON(mp));
 886
 887		/*
 888		 * If a file system had quotas running earlier, but decided to
 889		 * mount without -o uquota/pquota/gquota options, revoke the
 890		 * quotachecked license.
 891		 */
 892		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
 893			xfs_notice(mp, "resetting quota flags");
 894			error = xfs_mount_reset_sbqflags(mp);
 895			if (error)
 896				goto out_rtunmount;
 897		}
 898	}
 899
 900	/*
 
 
 
 
 
 
 
 
 
 901	 * Finish recovering the file system.  This part needed to be delayed
 902	 * until after the root and real-time bitmap inodes were consistently
 903	 * read in.
 904	 */
 905	error = xfs_log_mount_finish(mp);
 906	if (error) {
 907		xfs_warn(mp, "log mount finish failed");
 908		goto out_rtunmount;
 909	}
 910
 911	/*
 912	 * Now the log is fully replayed, we can transition to full read-only
 913	 * mode for read-only mounts. This will sync all the metadata and clean
 914	 * the log so that the recovery we just performed does not have to be
 915	 * replayed again on the next mount.
 916	 *
 917	 * We use the same quiesce mechanism as the rw->ro remount, as they are
 918	 * semantically identical operations.
 919	 */
 920	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
 921							XFS_MOUNT_RDONLY) {
 922		xfs_quiesce_attr(mp);
 923	}
 924
 925	/*
 926	 * Complete the quota initialisation, post-log-replay component.
 927	 */
 928	if (quotamount) {
 929		ASSERT(mp->m_qflags == 0);
 930		mp->m_qflags = quotaflags;
 931
 932		xfs_qm_mount_quotas(mp);
 933	}
 934
 935	/*
 936	 * Now we are mounted, reserve a small amount of unused space for
 937	 * privileged transactions. This is needed so that transaction
 938	 * space required for critical operations can dip into this pool
 939	 * when at ENOSPC. This is needed for operations like create with
 940	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
 941	 * are not allowed to use this reserved space.
 942	 *
 943	 * This may drive us straight to ENOSPC on mount, but that implies
 944	 * we were already there on the last unmount. Warn if this occurs.
 945	 */
 946	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 947		resblks = xfs_default_resblks(mp);
 948		error = xfs_reserve_blocks(mp, &resblks, NULL);
 949		if (error)
 950			xfs_warn(mp,
 951	"Unable to allocate reserve blocks. Continuing without reserve pool.");
 952
 953		/* Recover any CoW blocks that never got remapped. */
 954		error = xfs_reflink_recover_cow(mp);
 955		if (error) {
 956			xfs_err(mp,
 957	"Error %d recovering leftover CoW allocations.", error);
 958			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 959			goto out_quota;
 960		}
 961
 962		/* Reserve AG blocks for future btree expansion. */
 963		error = xfs_fs_reserve_ag_blocks(mp);
 964		if (error && error != -ENOSPC)
 965			goto out_agresv;
 966	}
 967
 968	return 0;
 969
 970 out_agresv:
 971	xfs_fs_unreserve_ag_blocks(mp);
 972 out_quota:
 973	xfs_qm_unmount_quotas(mp);
 974 out_rtunmount:
 
 975	xfs_rtunmount_inodes(mp);
 976 out_rele_rip:
 977	xfs_irele(rip);
 978	/* Clean out dquots that might be in memory after quotacheck. */
 979	xfs_qm_unmount(mp);
 980	/*
 981	 * Cancel all delayed reclaim work and reclaim the inodes directly.
 982	 * We have to do this /after/ rtunmount and qm_unmount because those
 983	 * two will have scheduled delayed reclaim for the rt/quota inodes.
 984	 *
 985	 * This is slightly different from the unmountfs call sequence
 986	 * because we could be tearing down a partially set up mount.  In
 987	 * particular, if log_mount_finish fails we bail out without calling
 988	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
 989	 * quota inodes.
 990	 */
 991	cancel_delayed_work_sync(&mp->m_reclaim_work);
 992	xfs_reclaim_inodes(mp, SYNC_WAIT);
 993	xfs_health_unmount(mp);
 994 out_log_dealloc:
 995	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
 996	xfs_log_mount_cancel(mp);
 997 out_fail_wait:
 998	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
 999		xfs_wait_buftarg(mp->m_logdev_targp);
1000	xfs_wait_buftarg(mp->m_ddev_targp);
1001 out_free_perag:
1002	xfs_free_perag(mp);
1003 out_free_dir:
1004	xfs_da_unmount(mp);
1005 out_remove_uuid:
1006	xfs_uuid_unmount(mp);
1007 out_remove_errortag:
1008	xfs_errortag_del(mp);
1009 out_remove_error_sysfs:
1010	xfs_error_sysfs_del(mp);
1011 out_del_stats:
1012	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1013 out_remove_sysfs:
1014	xfs_sysfs_del(&mp->m_kobj);
1015 out:
1016	return error;
1017}
1018
1019/*
1020 * This flushes out the inodes,dquots and the superblock, unmounts the
1021 * log and makes sure that incore structures are freed.
1022 */
1023void
1024xfs_unmountfs(
1025	struct xfs_mount	*mp)
1026{
1027	uint64_t		resblks;
1028	int			error;
1029
1030	xfs_stop_block_reaping(mp);
 
 
1031	xfs_fs_unreserve_ag_blocks(mp);
1032	xfs_qm_unmount_quotas(mp);
1033	xfs_rtunmount_inodes(mp);
1034	xfs_irele(mp->m_rootip);
1035
1036	/*
1037	 * We can potentially deadlock here if we have an inode cluster
1038	 * that has been freed has its buffer still pinned in memory because
1039	 * the transaction is still sitting in a iclog. The stale inodes
1040	 * on that buffer will have their flush locks held until the
1041	 * transaction hits the disk and the callbacks run. the inode
1042	 * flush takes the flush lock unconditionally and with nothing to
1043	 * push out the iclog we will never get that unlocked. hence we
1044	 * need to force the log first.
1045	 */
1046	xfs_log_force(mp, XFS_LOG_SYNC);
1047
1048	/*
1049	 * Wait for all busy extents to be freed, including completion of
1050	 * any discard operation.
1051	 */
1052	xfs_extent_busy_wait_all(mp);
1053	flush_workqueue(xfs_discard_wq);
1054
1055	/*
1056	 * We now need to tell the world we are unmounting. This will allow
1057	 * us to detect that the filesystem is going away and we should error
1058	 * out anything that we have been retrying in the background. This will
1059	 * prevent neverending retries in AIL pushing from hanging the unmount.
1060	 */
1061	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1062
1063	/*
1064	 * Flush all pending changes from the AIL.
1065	 */
1066	xfs_ail_push_all_sync(mp->m_ail);
1067
1068	/*
1069	 * And reclaim all inodes.  At this point there should be no dirty
1070	 * inodes and none should be pinned or locked, but use synchronous
1071	 * reclaim just to be sure. We can stop background inode reclaim
1072	 * here as well if it is still running.
1073	 */
1074	cancel_delayed_work_sync(&mp->m_reclaim_work);
1075	xfs_reclaim_inodes(mp, SYNC_WAIT);
1076	xfs_health_unmount(mp);
1077
1078	xfs_qm_unmount(mp);
1079
1080	/*
1081	 * Unreserve any blocks we have so that when we unmount we don't account
1082	 * the reserved free space as used. This is really only necessary for
1083	 * lazy superblock counting because it trusts the incore superblock
1084	 * counters to be absolutely correct on clean unmount.
1085	 *
1086	 * We don't bother correcting this elsewhere for lazy superblock
1087	 * counting because on mount of an unclean filesystem we reconstruct the
1088	 * correct counter value and this is irrelevant.
1089	 *
1090	 * For non-lazy counter filesystems, this doesn't matter at all because
1091	 * we only every apply deltas to the superblock and hence the incore
1092	 * value does not matter....
1093	 */
1094	resblks = 0;
1095	error = xfs_reserve_blocks(mp, &resblks, NULL);
1096	if (error)
1097		xfs_warn(mp, "Unable to free reserved block pool. "
1098				"Freespace may not be correct on next mount.");
1099
1100	error = xfs_log_sbcount(mp);
1101	if (error)
1102		xfs_warn(mp, "Unable to update superblock counters. "
1103				"Freespace may not be correct on next mount.");
1104
1105
1106	xfs_log_unmount(mp);
1107	xfs_da_unmount(mp);
1108	xfs_uuid_unmount(mp);
1109
1110#if defined(DEBUG)
1111	xfs_errortag_clearall(mp);
1112#endif
1113	xfs_free_perag(mp);
1114
1115	xfs_errortag_del(mp);
1116	xfs_error_sysfs_del(mp);
1117	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1118	xfs_sysfs_del(&mp->m_kobj);
1119}
1120
1121/*
1122 * Determine whether modifications can proceed. The caller specifies the minimum
1123 * freeze level for which modifications should not be allowed. This allows
1124 * certain operations to proceed while the freeze sequence is in progress, if
1125 * necessary.
1126 */
1127bool
1128xfs_fs_writable(
1129	struct xfs_mount	*mp,
1130	int			level)
1131{
1132	ASSERT(level > SB_UNFROZEN);
1133	if ((mp->m_super->s_writers.frozen >= level) ||
1134	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1135		return false;
1136
1137	return true;
1138}
1139
1140/*
1141 * xfs_log_sbcount
1142 *
1143 * Sync the superblock counters to disk.
1144 *
1145 * Note this code can be called during the process of freezing, so we use the
1146 * transaction allocator that does not block when the transaction subsystem is
1147 * in its frozen state.
1148 */
1149int
1150xfs_log_sbcount(xfs_mount_t *mp)
1151{
1152	/* allow this to proceed during the freeze sequence... */
1153	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1154		return 0;
1155
1156	/*
1157	 * we don't need to do this if we are updating the superblock
1158	 * counters on every modification.
1159	 */
1160	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1161		return 0;
1162
1163	return xfs_sync_sb(mp, true);
1164}
1165
1166/*
1167 * Deltas for the inode count are +/-64, hence we use a large batch size
1168 * of 128 so we don't need to take the counter lock on every update.
1169 */
1170#define XFS_ICOUNT_BATCH	128
1171int
1172xfs_mod_icount(
1173	struct xfs_mount	*mp,
1174	int64_t			delta)
1175{
1176	percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1177	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1178		ASSERT(0);
1179		percpu_counter_add(&mp->m_icount, -delta);
1180		return -EINVAL;
1181	}
1182	return 0;
1183}
1184
1185int
1186xfs_mod_ifree(
1187	struct xfs_mount	*mp,
1188	int64_t			delta)
1189{
1190	percpu_counter_add(&mp->m_ifree, delta);
1191	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1192		ASSERT(0);
1193		percpu_counter_add(&mp->m_ifree, -delta);
1194		return -EINVAL;
1195	}
1196	return 0;
1197}
1198
1199/*
1200 * Deltas for the block count can vary from 1 to very large, but lock contention
1201 * only occurs on frequent small block count updates such as in the delayed
1202 * allocation path for buffered writes (page a time updates). Hence we set
1203 * a large batch count (1024) to minimise global counter updates except when
1204 * we get near to ENOSPC and we have to be very accurate with our updates.
1205 */
1206#define XFS_FDBLOCKS_BATCH	1024
1207int
1208xfs_mod_fdblocks(
1209	struct xfs_mount	*mp,
1210	int64_t			delta,
1211	bool			rsvd)
1212{
1213	int64_t			lcounter;
1214	long long		res_used;
1215	s32			batch;
1216
1217	if (delta > 0) {
1218		/*
1219		 * If the reserve pool is depleted, put blocks back into it
1220		 * first. Most of the time the pool is full.
1221		 */
1222		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1223			percpu_counter_add(&mp->m_fdblocks, delta);
1224			return 0;
1225		}
1226
1227		spin_lock(&mp->m_sb_lock);
1228		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1229
1230		if (res_used > delta) {
1231			mp->m_resblks_avail += delta;
1232		} else {
1233			delta -= res_used;
1234			mp->m_resblks_avail = mp->m_resblks;
1235			percpu_counter_add(&mp->m_fdblocks, delta);
1236		}
1237		spin_unlock(&mp->m_sb_lock);
1238		return 0;
1239	}
1240
1241	/*
1242	 * Taking blocks away, need to be more accurate the closer we
1243	 * are to zero.
1244	 *
1245	 * If the counter has a value of less than 2 * max batch size,
1246	 * then make everything serialise as we are real close to
1247	 * ENOSPC.
1248	 */
1249	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1250				     XFS_FDBLOCKS_BATCH) < 0)
1251		batch = 1;
1252	else
1253		batch = XFS_FDBLOCKS_BATCH;
1254
1255	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1256	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1257				     XFS_FDBLOCKS_BATCH) >= 0) {
1258		/* we had space! */
1259		return 0;
1260	}
1261
1262	/*
1263	 * lock up the sb for dipping into reserves before releasing the space
1264	 * that took us to ENOSPC.
1265	 */
1266	spin_lock(&mp->m_sb_lock);
1267	percpu_counter_add(&mp->m_fdblocks, -delta);
1268	if (!rsvd)
1269		goto fdblocks_enospc;
1270
1271	lcounter = (long long)mp->m_resblks_avail + delta;
1272	if (lcounter >= 0) {
1273		mp->m_resblks_avail = lcounter;
1274		spin_unlock(&mp->m_sb_lock);
1275		return 0;
1276	}
1277	printk_once(KERN_WARNING
1278		"Filesystem \"%s\": reserve blocks depleted! "
1279		"Consider increasing reserve pool size.",
1280		mp->m_fsname);
1281fdblocks_enospc:
1282	spin_unlock(&mp->m_sb_lock);
1283	return -ENOSPC;
1284}
1285
1286int
1287xfs_mod_frextents(
1288	struct xfs_mount	*mp,
1289	int64_t			delta)
1290{
1291	int64_t			lcounter;
1292	int			ret = 0;
1293
1294	spin_lock(&mp->m_sb_lock);
1295	lcounter = mp->m_sb.sb_frextents + delta;
1296	if (lcounter < 0)
1297		ret = -ENOSPC;
1298	else
1299		mp->m_sb.sb_frextents = lcounter;
1300	spin_unlock(&mp->m_sb_lock);
1301	return ret;
1302}
1303
1304/*
1305 * xfs_getsb() is called to obtain the buffer for the superblock.
1306 * The buffer is returned locked and read in from disk.
1307 * The buffer should be released with a call to xfs_brelse().
 
 
 
 
1308 */
1309struct xfs_buf *
1310xfs_getsb(
1311	struct xfs_mount	*mp)
 
1312{
1313	struct xfs_buf		*bp = mp->m_sb_bp;
1314
1315	xfs_buf_lock(bp);
 
 
 
 
 
1316	xfs_buf_hold(bp);
1317	ASSERT(bp->b_flags & XBF_DONE);
1318	return bp;
1319}
1320
1321/*
1322 * Used to free the superblock along various error paths.
1323 */
1324void
1325xfs_freesb(
1326	struct xfs_mount	*mp)
1327{
1328	struct xfs_buf		*bp = mp->m_sb_bp;
1329
1330	xfs_buf_lock(bp);
1331	mp->m_sb_bp = NULL;
1332	xfs_buf_relse(bp);
1333}
1334
1335/*
1336 * If the underlying (data/log/rt) device is readonly, there are some
1337 * operations that cannot proceed.
1338 */
1339int
1340xfs_dev_is_read_only(
1341	struct xfs_mount	*mp,
1342	char			*message)
1343{
1344	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1345	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1346	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1347		xfs_notice(mp, "%s required on read-only device.", message);
1348		xfs_notice(mp, "write access unavailable, cannot proceed.");
1349		return -EROFS;
1350	}
1351	return 0;
1352}
1353
1354/* Force the summary counters to be recalculated at next mount. */
1355void
1356xfs_force_summary_recalc(
1357	struct xfs_mount	*mp)
1358{
1359	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1360		return;
1361
1362	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1363}
1364
1365/*
1366 * Update the in-core delayed block counter.
1367 *
1368 * We prefer to update the counter without having to take a spinlock for every
1369 * counter update (i.e. batching).  Each change to delayed allocation
1370 * reservations can change can easily exceed the default percpu counter
1371 * batching, so we use a larger batch factor here.
1372 *
1373 * Note that we don't currently have any callers requiring fast summation
1374 * (e.g. percpu_counter_read) so we can use a big batch value here.
1375 */
1376#define XFS_DELALLOC_BATCH	(4096)
1377void
1378xfs_mod_delalloc(
1379	struct xfs_mount	*mp,
1380	int64_t			delta)
1381{
1382	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1383			XFS_DELALLOC_BATCH);
1384}