cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

xfs_sb.c (40457B)


      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_ialloc.h"
     16#include "xfs_alloc.h"
     17#include "xfs_error.h"
     18#include "xfs_trans.h"
     19#include "xfs_buf_item.h"
     20#include "xfs_bmap_btree.h"
     21#include "xfs_alloc_btree.h"
     22#include "xfs_log.h"
     23#include "xfs_rmap_btree.h"
     24#include "xfs_refcount_btree.h"
     25#include "xfs_da_format.h"
     26#include "xfs_health.h"
     27#include "xfs_ag.h"
     28
     29/*
     30 * Physical superblock buffer manipulations. Shared with libxfs in userspace.
     31 */
     32
     33/*
     34 * Check that all the V4 feature bits that the V5 filesystem format requires are
     35 * correctly set.
     36 */
     37static bool
     38xfs_sb_validate_v5_features(
     39	struct xfs_sb	*sbp)
     40{
     41	/* We must not have any unknown V4 feature bits set */
     42	if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS)
     43		return false;
     44
     45	/*
     46	 * The CRC bit is considered an invalid V4 flag, so we have to add it
     47	 * manually to the OKBITS mask.
     48	 */
     49	if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS |
     50				  XFS_SB_VERSION2_CRCBIT))
     51		return false;
     52
     53	/* Now check all the required V4 feature flags are set. */
     54
     55#define V5_VERS_FLAGS	(XFS_SB_VERSION_NLINKBIT	| \
     56			XFS_SB_VERSION_ALIGNBIT		| \
     57			XFS_SB_VERSION_LOGV2BIT		| \
     58			XFS_SB_VERSION_EXTFLGBIT	| \
     59			XFS_SB_VERSION_DIRV2BIT		| \
     60			XFS_SB_VERSION_MOREBITSBIT)
     61
     62#define V5_FEAT_FLAGS	(XFS_SB_VERSION2_LAZYSBCOUNTBIT	| \
     63			XFS_SB_VERSION2_ATTR2BIT	| \
     64			XFS_SB_VERSION2_PROJID32BIT	| \
     65			XFS_SB_VERSION2_CRCBIT)
     66
     67	if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS)
     68		return false;
     69	if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS)
     70		return false;
     71	return true;
     72}
     73
     74/*
     75 * We support all XFS versions newer than a v4 superblock with V2 directories.
     76 */
     77bool
     78xfs_sb_good_version(
     79	struct xfs_sb	*sbp)
     80{
     81	/*
     82	 * All v5 filesystems are supported, but we must check that all the
     83	 * required v4 feature flags are enabled correctly as the code checks
     84	 * those flags and not for v5 support.
     85	 */
     86	if (xfs_sb_is_v5(sbp))
     87		return xfs_sb_validate_v5_features(sbp);
     88
     89	/* We must not have any unknown v4 feature bits set */
     90	if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) ||
     91	    ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) &&
     92	     (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS)))
     93		return false;
     94
     95	/* versions prior to v4 are not supported */
     96	if (XFS_SB_VERSION_NUM(sbp) < XFS_SB_VERSION_4)
     97		return false;
     98
     99	/* V4 filesystems need v2 directories and unwritten extents */
    100	if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT))
    101		return false;
    102	if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT))
    103		return false;
    104
    105	/* It's a supported v4 filesystem */
    106	return true;
    107}
    108
    109uint64_t
    110xfs_sb_version_to_features(
    111	struct xfs_sb	*sbp)
    112{
    113	uint64_t	features = 0;
    114
    115	/* optional V4 features */
    116	if (sbp->sb_rblocks > 0)
    117		features |= XFS_FEAT_REALTIME;
    118	if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT)
    119		features |= XFS_FEAT_NLINK;
    120	if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT)
    121		features |= XFS_FEAT_ATTR;
    122	if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT)
    123		features |= XFS_FEAT_QUOTA;
    124	if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT)
    125		features |= XFS_FEAT_ALIGN;
    126	if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT)
    127		features |= XFS_FEAT_LOGV2;
    128	if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT)
    129		features |= XFS_FEAT_DALIGN;
    130	if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)
    131		features |= XFS_FEAT_EXTFLG;
    132	if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT)
    133		features |= XFS_FEAT_SECTOR;
    134	if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT)
    135		features |= XFS_FEAT_ASCIICI;
    136	if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) {
    137		if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT)
    138			features |= XFS_FEAT_LAZYSBCOUNT;
    139		if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
    140			features |= XFS_FEAT_ATTR2;
    141		if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT)
    142			features |= XFS_FEAT_PROJID32;
    143		if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE)
    144			features |= XFS_FEAT_FTYPE;
    145	}
    146
    147	if (!xfs_sb_is_v5(sbp))
    148		return features;
    149
    150	/* Always on V5 features */
    151	features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG |
    152		    XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 |
    153		    XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO;
    154
    155	/* Optional V5 features */
    156	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT)
    157		features |= XFS_FEAT_FINOBT;
    158	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT)
    159		features |= XFS_FEAT_RMAPBT;
    160	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK)
    161		features |= XFS_FEAT_REFLINK;
    162	if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT)
    163		features |= XFS_FEAT_INOBTCNT;
    164	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE)
    165		features |= XFS_FEAT_FTYPE;
    166	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES)
    167		features |= XFS_FEAT_SPINODES;
    168	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
    169		features |= XFS_FEAT_META_UUID;
    170	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME)
    171		features |= XFS_FEAT_BIGTIME;
    172	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR)
    173		features |= XFS_FEAT_NEEDSREPAIR;
    174	if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64)
    175		features |= XFS_FEAT_NREXT64;
    176
    177	return features;
    178}
    179
    180/* Check all the superblock fields we care about when reading one in. */
    181STATIC int
    182xfs_validate_sb_read(
    183	struct xfs_mount	*mp,
    184	struct xfs_sb		*sbp)
    185{
    186	if (!xfs_sb_is_v5(sbp))
    187		return 0;
    188
    189	/*
    190	 * Version 5 superblock feature mask validation. Reject combinations
    191	 * the kernel cannot support up front before checking anything else.
    192	 */
    193	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
    194		xfs_warn(mp,
    195"Superblock has unknown compatible features (0x%x) enabled.",
    196			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
    197		xfs_warn(mp,
    198"Using a more recent kernel is recommended.");
    199	}
    200
    201	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
    202		xfs_alert(mp,
    203"Superblock has unknown read-only compatible features (0x%x) enabled.",
    204			(sbp->sb_features_ro_compat &
    205					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
    206		if (!xfs_is_readonly(mp)) {
    207			xfs_warn(mp,
    208"Attempted to mount read-only compatible filesystem read-write.");
    209			xfs_warn(mp,
    210"Filesystem can only be safely mounted read only.");
    211
    212			return -EINVAL;
    213		}
    214	}
    215	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
    216		xfs_warn(mp,
    217"Superblock has unknown incompatible features (0x%x) enabled.",
    218			(sbp->sb_features_incompat &
    219					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
    220		xfs_warn(mp,
    221"Filesystem cannot be safely mounted by this kernel.");
    222		return -EINVAL;
    223	}
    224
    225	return 0;
    226}
    227
    228/* Check all the superblock fields we care about when writing one out. */
    229STATIC int
    230xfs_validate_sb_write(
    231	struct xfs_mount	*mp,
    232	struct xfs_buf		*bp,
    233	struct xfs_sb		*sbp)
    234{
    235	/*
    236	 * Carry out additional sb summary counter sanity checks when we write
    237	 * the superblock.  We skip this in the read validator because there
    238	 * could be newer superblocks in the log and if the values are garbage
    239	 * even after replay we'll recalculate them at the end of log mount.
    240	 *
    241	 * mkfs has traditionally written zeroed counters to inprogress and
    242	 * secondary superblocks, so allow this usage to continue because
    243	 * we never read counters from such superblocks.
    244	 */
    245	if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress &&
    246	    (sbp->sb_fdblocks > sbp->sb_dblocks ||
    247	     !xfs_verify_icount(mp, sbp->sb_icount) ||
    248	     sbp->sb_ifree > sbp->sb_icount)) {
    249		xfs_warn(mp, "SB summary counter sanity check failed");
    250		return -EFSCORRUPTED;
    251	}
    252
    253	if (!xfs_sb_is_v5(sbp))
    254		return 0;
    255
    256	/*
    257	 * Version 5 superblock feature mask validation. Reject combinations
    258	 * the kernel cannot support since we checked for unsupported bits in
    259	 * the read verifier, which means that memory is corrupt.
    260	 */
    261	if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) {
    262		xfs_warn(mp,
    263"Corruption detected in superblock compatible features (0x%x)!",
    264			(sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN));
    265		return -EFSCORRUPTED;
    266	}
    267
    268	if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
    269		xfs_alert(mp,
    270"Corruption detected in superblock read-only compatible features (0x%x)!",
    271			(sbp->sb_features_ro_compat &
    272					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
    273		return -EFSCORRUPTED;
    274	}
    275	if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) {
    276		xfs_warn(mp,
    277"Corruption detected in superblock incompatible features (0x%x)!",
    278			(sbp->sb_features_incompat &
    279					XFS_SB_FEAT_INCOMPAT_UNKNOWN));
    280		return -EFSCORRUPTED;
    281	}
    282	if (xfs_sb_has_incompat_log_feature(sbp,
    283			XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
    284		xfs_warn(mp,
    285"Corruption detected in superblock incompatible log features (0x%x)!",
    286			(sbp->sb_features_log_incompat &
    287					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
    288		return -EFSCORRUPTED;
    289	}
    290
    291	/*
    292	 * We can't read verify the sb LSN because the read verifier is called
    293	 * before the log is allocated and processed. We know the log is set up
    294	 * before write verifier calls, so check it here.
    295	 */
    296	if (!xfs_log_check_lsn(mp, sbp->sb_lsn))
    297		return -EFSCORRUPTED;
    298
    299	return 0;
    300}
    301
    302/* Check the validity of the SB. */
    303STATIC int
    304xfs_validate_sb_common(
    305	struct xfs_mount	*mp,
    306	struct xfs_buf		*bp,
    307	struct xfs_sb		*sbp)
    308{
    309	struct xfs_dsb		*dsb = bp->b_addr;
    310	uint32_t		agcount = 0;
    311	uint32_t		rem;
    312	bool			has_dalign;
    313
    314	if (!xfs_verify_magic(bp, dsb->sb_magicnum)) {
    315		xfs_warn(mp,
    316"Superblock has bad magic number 0x%x. Not an XFS filesystem?",
    317			be32_to_cpu(dsb->sb_magicnum));
    318		return -EWRONGFS;
    319	}
    320
    321	if (!xfs_sb_good_version(sbp)) {
    322		xfs_warn(mp,
    323"Superblock has unknown features enabled or corrupted feature masks.");
    324		return -EWRONGFS;
    325	}
    326
    327	/*
    328	 * Validate feature flags and state
    329	 */
    330	if (xfs_sb_is_v5(sbp)) {
    331		if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) {
    332			xfs_notice(mp,
    333"Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)",
    334				sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE);
    335			return -EFSCORRUPTED;
    336		}
    337
    338		/* V5 has a separate project quota inode */
    339		if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) {
    340			xfs_notice(mp,
    341			   "Version 5 of Super block has XFS_OQUOTA bits.");
    342			return -EFSCORRUPTED;
    343		}
    344
    345		/*
    346		 * Full inode chunks must be aligned to inode chunk size when
    347		 * sparse inodes are enabled to support the sparse chunk
    348		 * allocation algorithm and prevent overlapping inode records.
    349		 */
    350		if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) {
    351			uint32_t	align;
    352
    353			align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize
    354					>> sbp->sb_blocklog;
    355			if (sbp->sb_inoalignmt != align) {
    356				xfs_warn(mp,
    357"Inode block alignment (%u) must match chunk size (%u) for sparse inodes.",
    358					 sbp->sb_inoalignmt, align);
    359				return -EINVAL;
    360			}
    361		}
    362	} else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD |
    363				XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) {
    364			xfs_notice(mp,
    365"Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits.");
    366			return -EFSCORRUPTED;
    367	}
    368
    369	if (unlikely(
    370	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
    371		xfs_warn(mp,
    372		"filesystem is marked as having an external log; "
    373		"specify logdev on the mount command line.");
    374		return -EINVAL;
    375	}
    376
    377	if (unlikely(
    378	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
    379		xfs_warn(mp,
    380		"filesystem is marked as having an internal log; "
    381		"do not specify logdev on the mount command line.");
    382		return -EINVAL;
    383	}
    384
    385	/* Compute agcount for this number of dblocks and agblocks */
    386	if (sbp->sb_agblocks) {
    387		agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem);
    388		if (rem)
    389			agcount++;
    390	}
    391
    392	/*
    393	 * More sanity checking.  Most of these were stolen directly from
    394	 * xfs_repair.
    395	 */
    396	if (unlikely(
    397	    sbp->sb_agcount <= 0					||
    398	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
    399	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
    400	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
    401	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
    402	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
    403	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
    404	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
    405	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
    406	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
    407	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
    408	    sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
    409	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
    410	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
    411	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
    412	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
    413	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
    414	    sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE			||
    415	    sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) ||
    416	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES	||
    417	    XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES	||
    418	    sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1	||
    419	    agcount == 0 || agcount != sbp->sb_agcount			||
    420	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
    421	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
    422	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
    423	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */)	||
    424	    sbp->sb_dblocks == 0					||
    425	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
    426	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp)			||
    427	    sbp->sb_shared_vn != 0)) {
    428		xfs_notice(mp, "SB sanity check failed");
    429		return -EFSCORRUPTED;
    430	}
    431
    432	/* Validate the realtime geometry; stolen from xfs_repair */
    433	if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE ||
    434	    sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) {
    435		xfs_notice(mp,
    436			"realtime extent sanity check failed");
    437		return -EFSCORRUPTED;
    438	}
    439
    440	if (sbp->sb_rblocks == 0) {
    441		if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 ||
    442		    sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) {
    443			xfs_notice(mp,
    444				"realtime zeroed geometry check failed");
    445			return -EFSCORRUPTED;
    446		}
    447	} else {
    448		uint64_t	rexts;
    449		uint64_t	rbmblocks;
    450
    451		rexts = div_u64(sbp->sb_rblocks, sbp->sb_rextsize);
    452		rbmblocks = howmany_64(sbp->sb_rextents,
    453				       NBBY * sbp->sb_blocksize);
    454
    455		if (sbp->sb_rextents != rexts ||
    456		    sbp->sb_rextslog != xfs_highbit32(sbp->sb_rextents) ||
    457		    sbp->sb_rbmblocks != rbmblocks) {
    458			xfs_notice(mp,
    459				"realtime geometry sanity check failed");
    460			return -EFSCORRUPTED;
    461		}
    462	}
    463
    464	/*
    465	 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign)
    466	 * would imply the image is corrupted.
    467	 */
    468	has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT;
    469	if (!!sbp->sb_unit ^ has_dalign) {
    470		xfs_notice(mp, "SB stripe alignment sanity check failed");
    471		return -EFSCORRUPTED;
    472	}
    473
    474	if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit),
    475			XFS_FSB_TO_B(mp, sbp->sb_width), 0, false))
    476		return -EFSCORRUPTED;
    477
    478	/*
    479	 * Currently only very few inode sizes are supported.
    480	 */
    481	switch (sbp->sb_inodesize) {
    482	case 256:
    483	case 512:
    484	case 1024:
    485	case 2048:
    486		break;
    487	default:
    488		xfs_warn(mp, "inode size of %d bytes not supported",
    489				sbp->sb_inodesize);
    490		return -ENOSYS;
    491	}
    492
    493	return 0;
    494}
    495
    496void
    497xfs_sb_quota_from_disk(struct xfs_sb *sbp)
    498{
    499	/*
    500	 * older mkfs doesn't initialize quota inodes to NULLFSINO. This
    501	 * leads to in-core values having two different values for a quota
    502	 * inode to be invalid: 0 and NULLFSINO. Change it to a single value
    503	 * NULLFSINO.
    504	 *
    505	 * Note that this change affect only the in-core values. These
    506	 * values are not written back to disk unless any quota information
    507	 * is written to the disk. Even in that case, sb_pquotino field is
    508	 * not written to disk unless the superblock supports pquotino.
    509	 */
    510	if (sbp->sb_uquotino == 0)
    511		sbp->sb_uquotino = NULLFSINO;
    512	if (sbp->sb_gquotino == 0)
    513		sbp->sb_gquotino = NULLFSINO;
    514	if (sbp->sb_pquotino == 0)
    515		sbp->sb_pquotino = NULLFSINO;
    516
    517	/*
    518	 * We need to do these manipilations only if we are working
    519	 * with an older version of on-disk superblock.
    520	 */
    521	if (xfs_sb_is_v5(sbp))
    522		return;
    523
    524	if (sbp->sb_qflags & XFS_OQUOTA_ENFD)
    525		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
    526					XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD;
    527	if (sbp->sb_qflags & XFS_OQUOTA_CHKD)
    528		sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ?
    529					XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD;
    530	sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD);
    531
    532	if (sbp->sb_qflags & XFS_PQUOTA_ACCT &&
    533	    sbp->sb_gquotino != NULLFSINO)  {
    534		/*
    535		 * In older version of superblock, on-disk superblock only
    536		 * has sb_gquotino, and in-core superblock has both sb_gquotino
    537		 * and sb_pquotino. But, only one of them is supported at any
    538		 * point of time. So, if PQUOTA is set in disk superblock,
    539		 * copy over sb_gquotino to sb_pquotino.  The NULLFSINO test
    540		 * above is to make sure we don't do this twice and wipe them
    541		 * both out!
    542		 */
    543		sbp->sb_pquotino = sbp->sb_gquotino;
    544		sbp->sb_gquotino = NULLFSINO;
    545	}
    546}
    547
    548static void
    549__xfs_sb_from_disk(
    550	struct xfs_sb	*to,
    551	struct xfs_dsb	*from,
    552	bool		convert_xquota)
    553{
    554	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
    555	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
    556	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
    557	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
    558	to->sb_rextents = be64_to_cpu(from->sb_rextents);
    559	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
    560	to->sb_logstart = be64_to_cpu(from->sb_logstart);
    561	to->sb_rootino = be64_to_cpu(from->sb_rootino);
    562	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
    563	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
    564	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
    565	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
    566	to->sb_agcount = be32_to_cpu(from->sb_agcount);
    567	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
    568	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
    569	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
    570	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
    571	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
    572	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
    573	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
    574	to->sb_blocklog = from->sb_blocklog;
    575	to->sb_sectlog = from->sb_sectlog;
    576	to->sb_inodelog = from->sb_inodelog;
    577	to->sb_inopblog = from->sb_inopblog;
    578	to->sb_agblklog = from->sb_agblklog;
    579	to->sb_rextslog = from->sb_rextslog;
    580	to->sb_inprogress = from->sb_inprogress;
    581	to->sb_imax_pct = from->sb_imax_pct;
    582	to->sb_icount = be64_to_cpu(from->sb_icount);
    583	to->sb_ifree = be64_to_cpu(from->sb_ifree);
    584	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
    585	to->sb_frextents = be64_to_cpu(from->sb_frextents);
    586	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
    587	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
    588	to->sb_qflags = be16_to_cpu(from->sb_qflags);
    589	to->sb_flags = from->sb_flags;
    590	to->sb_shared_vn = from->sb_shared_vn;
    591	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
    592	to->sb_unit = be32_to_cpu(from->sb_unit);
    593	to->sb_width = be32_to_cpu(from->sb_width);
    594	to->sb_dirblklog = from->sb_dirblklog;
    595	to->sb_logsectlog = from->sb_logsectlog;
    596	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
    597	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
    598	to->sb_features2 = be32_to_cpu(from->sb_features2);
    599	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
    600	to->sb_features_compat = be32_to_cpu(from->sb_features_compat);
    601	to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat);
    602	to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat);
    603	to->sb_features_log_incompat =
    604				be32_to_cpu(from->sb_features_log_incompat);
    605	/* crc is only used on disk, not in memory; just init to 0 here. */
    606	to->sb_crc = 0;
    607	to->sb_spino_align = be32_to_cpu(from->sb_spino_align);
    608	to->sb_pquotino = be64_to_cpu(from->sb_pquotino);
    609	to->sb_lsn = be64_to_cpu(from->sb_lsn);
    610	/*
    611	 * sb_meta_uuid is only on disk if it differs from sb_uuid and the
    612	 * feature flag is set; if not set we keep it only in memory.
    613	 */
    614	if (xfs_sb_is_v5(to) &&
    615	    (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID))
    616		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
    617	else
    618		uuid_copy(&to->sb_meta_uuid, &from->sb_uuid);
    619	/* Convert on-disk flags to in-memory flags? */
    620	if (convert_xquota)
    621		xfs_sb_quota_from_disk(to);
    622}
    623
    624void
    625xfs_sb_from_disk(
    626	struct xfs_sb	*to,
    627	struct xfs_dsb	*from)
    628{
    629	__xfs_sb_from_disk(to, from, true);
    630}
    631
    632static void
    633xfs_sb_quota_to_disk(
    634	struct xfs_dsb	*to,
    635	struct xfs_sb	*from)
    636{
    637	uint16_t	qflags = from->sb_qflags;
    638
    639	to->sb_uquotino = cpu_to_be64(from->sb_uquotino);
    640
    641	/*
    642	 * The in-memory superblock quota state matches the v5 on-disk format so
    643	 * just write them out and return
    644	 */
    645	if (xfs_sb_is_v5(from)) {
    646		to->sb_qflags = cpu_to_be16(from->sb_qflags);
    647		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
    648		to->sb_pquotino = cpu_to_be64(from->sb_pquotino);
    649		return;
    650	}
    651
    652	/*
    653	 * For older superblocks (v4), the in-core version of sb_qflags do not
    654	 * have XFS_OQUOTA_* flags, whereas the on-disk version does.  So,
    655	 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags.
    656	 */
    657	qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD |
    658			XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD);
    659
    660	if (from->sb_qflags &
    661			(XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD))
    662		qflags |= XFS_OQUOTA_ENFD;
    663	if (from->sb_qflags &
    664			(XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD))
    665		qflags |= XFS_OQUOTA_CHKD;
    666	to->sb_qflags = cpu_to_be16(qflags);
    667
    668	/*
    669	 * GQUOTINO and PQUOTINO cannot be used together in versions
    670	 * of superblock that do not have pquotino. from->sb_flags
    671	 * tells us which quota is active and should be copied to
    672	 * disk. If neither are active, we should NULL the inode.
    673	 *
    674	 * In all cases, the separate pquotino must remain 0 because it
    675	 * is beyond the "end" of the valid non-pquotino superblock.
    676	 */
    677	if (from->sb_qflags & XFS_GQUOTA_ACCT)
    678		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);
    679	else if (from->sb_qflags & XFS_PQUOTA_ACCT)
    680		to->sb_gquotino = cpu_to_be64(from->sb_pquotino);
    681	else {
    682		/*
    683		 * We can't rely on just the fields being logged to tell us
    684		 * that it is safe to write NULLFSINO - we should only do that
    685		 * if quotas are not actually enabled. Hence only write
    686		 * NULLFSINO if both in-core quota inodes are NULL.
    687		 */
    688		if (from->sb_gquotino == NULLFSINO &&
    689		    from->sb_pquotino == NULLFSINO)
    690			to->sb_gquotino = cpu_to_be64(NULLFSINO);
    691	}
    692
    693	to->sb_pquotino = 0;
    694}
    695
    696void
    697xfs_sb_to_disk(
    698	struct xfs_dsb	*to,
    699	struct xfs_sb	*from)
    700{
    701	xfs_sb_quota_to_disk(to, from);
    702
    703	to->sb_magicnum = cpu_to_be32(from->sb_magicnum);
    704	to->sb_blocksize = cpu_to_be32(from->sb_blocksize);
    705	to->sb_dblocks = cpu_to_be64(from->sb_dblocks);
    706	to->sb_rblocks = cpu_to_be64(from->sb_rblocks);
    707	to->sb_rextents = cpu_to_be64(from->sb_rextents);
    708	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
    709	to->sb_logstart = cpu_to_be64(from->sb_logstart);
    710	to->sb_rootino = cpu_to_be64(from->sb_rootino);
    711	to->sb_rbmino = cpu_to_be64(from->sb_rbmino);
    712	to->sb_rsumino = cpu_to_be64(from->sb_rsumino);
    713	to->sb_rextsize = cpu_to_be32(from->sb_rextsize);
    714	to->sb_agblocks = cpu_to_be32(from->sb_agblocks);
    715	to->sb_agcount = cpu_to_be32(from->sb_agcount);
    716	to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks);
    717	to->sb_logblocks = cpu_to_be32(from->sb_logblocks);
    718	to->sb_versionnum = cpu_to_be16(from->sb_versionnum);
    719	to->sb_sectsize = cpu_to_be16(from->sb_sectsize);
    720	to->sb_inodesize = cpu_to_be16(from->sb_inodesize);
    721	to->sb_inopblock = cpu_to_be16(from->sb_inopblock);
    722	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
    723	to->sb_blocklog = from->sb_blocklog;
    724	to->sb_sectlog = from->sb_sectlog;
    725	to->sb_inodelog = from->sb_inodelog;
    726	to->sb_inopblog = from->sb_inopblog;
    727	to->sb_agblklog = from->sb_agblklog;
    728	to->sb_rextslog = from->sb_rextslog;
    729	to->sb_inprogress = from->sb_inprogress;
    730	to->sb_imax_pct = from->sb_imax_pct;
    731	to->sb_icount = cpu_to_be64(from->sb_icount);
    732	to->sb_ifree = cpu_to_be64(from->sb_ifree);
    733	to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks);
    734	to->sb_frextents = cpu_to_be64(from->sb_frextents);
    735
    736	to->sb_flags = from->sb_flags;
    737	to->sb_shared_vn = from->sb_shared_vn;
    738	to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt);
    739	to->sb_unit = cpu_to_be32(from->sb_unit);
    740	to->sb_width = cpu_to_be32(from->sb_width);
    741	to->sb_dirblklog = from->sb_dirblklog;
    742	to->sb_logsectlog = from->sb_logsectlog;
    743	to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize);
    744	to->sb_logsunit = cpu_to_be32(from->sb_logsunit);
    745
    746	/*
    747	 * We need to ensure that bad_features2 always matches features2.
    748	 * Hence we enforce that here rather than having to remember to do it
    749	 * everywhere else that updates features2.
    750	 */
    751	from->sb_bad_features2 = from->sb_features2;
    752	to->sb_features2 = cpu_to_be32(from->sb_features2);
    753	to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2);
    754
    755	if (!xfs_sb_is_v5(from))
    756		return;
    757
    758	to->sb_features_compat = cpu_to_be32(from->sb_features_compat);
    759	to->sb_features_ro_compat =
    760			cpu_to_be32(from->sb_features_ro_compat);
    761	to->sb_features_incompat =
    762			cpu_to_be32(from->sb_features_incompat);
    763	to->sb_features_log_incompat =
    764			cpu_to_be32(from->sb_features_log_incompat);
    765	to->sb_spino_align = cpu_to_be32(from->sb_spino_align);
    766	to->sb_lsn = cpu_to_be64(from->sb_lsn);
    767	if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)
    768		uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid);
    769}
    770
    771/*
    772 * If the superblock has the CRC feature bit set or the CRC field is non-null,
    773 * check that the CRC is valid.  We check the CRC field is non-null because a
    774 * single bit error could clear the feature bit and unused parts of the
    775 * superblock are supposed to be zero. Hence a non-null crc field indicates that
    776 * we've potentially lost a feature bit and we should check it anyway.
    777 *
    778 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
    779 * last field in V4 secondary superblocks.  So for secondary superblocks,
    780 * we are more forgiving, and ignore CRC failures if the primary doesn't
    781 * indicate that the fs version is V5.
    782 */
    783static void
    784xfs_sb_read_verify(
    785	struct xfs_buf		*bp)
    786{
    787	struct xfs_sb		sb;
    788	struct xfs_mount	*mp = bp->b_mount;
    789	struct xfs_dsb		*dsb = bp->b_addr;
    790	int			error;
    791
    792	/*
    793	 * open code the version check to avoid needing to convert the entire
    794	 * superblock from disk order just to check the version number
    795	 */
    796	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) &&
    797	    (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) ==
    798						XFS_SB_VERSION_5) ||
    799	     dsb->sb_crc != 0)) {
    800
    801		if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) {
    802			/* Only fail bad secondaries on a known V5 filesystem */
    803			if (xfs_buf_daddr(bp) == XFS_SB_DADDR ||
    804			    xfs_has_crc(mp)) {
    805				error = -EFSBADCRC;
    806				goto out_error;
    807			}
    808		}
    809	}
    810
    811	/*
    812	 * Check all the superblock fields.  Don't byteswap the xquota flags
    813	 * because _verify_common checks the on-disk values.
    814	 */
    815	__xfs_sb_from_disk(&sb, dsb, false);
    816	error = xfs_validate_sb_common(mp, bp, &sb);
    817	if (error)
    818		goto out_error;
    819	error = xfs_validate_sb_read(mp, &sb);
    820
    821out_error:
    822	if (error == -EFSCORRUPTED || error == -EFSBADCRC)
    823		xfs_verifier_error(bp, error, __this_address);
    824	else if (error)
    825		xfs_buf_ioerror(bp, error);
    826}
    827
    828/*
    829 * We may be probed for a filesystem match, so we may not want to emit
    830 * messages when the superblock buffer is not actually an XFS superblock.
    831 * If we find an XFS superblock, then run a normal, noisy mount because we are
    832 * really going to mount it and want to know about errors.
    833 */
    834static void
    835xfs_sb_quiet_read_verify(
    836	struct xfs_buf	*bp)
    837{
    838	struct xfs_dsb	*dsb = bp->b_addr;
    839
    840	if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) {
    841		/* XFS filesystem, verify noisily! */
    842		xfs_sb_read_verify(bp);
    843		return;
    844	}
    845	/* quietly fail */
    846	xfs_buf_ioerror(bp, -EWRONGFS);
    847}
    848
    849static void
    850xfs_sb_write_verify(
    851	struct xfs_buf		*bp)
    852{
    853	struct xfs_sb		sb;
    854	struct xfs_mount	*mp = bp->b_mount;
    855	struct xfs_buf_log_item	*bip = bp->b_log_item;
    856	struct xfs_dsb		*dsb = bp->b_addr;
    857	int			error;
    858
    859	/*
    860	 * Check all the superblock fields.  Don't byteswap the xquota flags
    861	 * because _verify_common checks the on-disk values.
    862	 */
    863	__xfs_sb_from_disk(&sb, dsb, false);
    864	error = xfs_validate_sb_common(mp, bp, &sb);
    865	if (error)
    866		goto out_error;
    867	error = xfs_validate_sb_write(mp, bp, &sb);
    868	if (error)
    869		goto out_error;
    870
    871	if (!xfs_sb_is_v5(&sb))
    872		return;
    873
    874	if (bip)
    875		dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
    876
    877	xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF);
    878	return;
    879
    880out_error:
    881	xfs_verifier_error(bp, error, __this_address);
    882}
    883
    884const struct xfs_buf_ops xfs_sb_buf_ops = {
    885	.name = "xfs_sb",
    886	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
    887	.verify_read = xfs_sb_read_verify,
    888	.verify_write = xfs_sb_write_verify,
    889};
    890
    891const struct xfs_buf_ops xfs_sb_quiet_buf_ops = {
    892	.name = "xfs_sb_quiet",
    893	.magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) },
    894	.verify_read = xfs_sb_quiet_read_verify,
    895	.verify_write = xfs_sb_write_verify,
    896};
    897
    898/*
    899 * xfs_mount_common
    900 *
    901 * Mount initialization code establishing various mount
    902 * fields from the superblock associated with the given
    903 * mount structure.
    904 *
    905 * Inode geometry are calculated in xfs_ialloc_setup_geometry.
    906 */
    907void
    908xfs_sb_mount_common(
    909	struct xfs_mount	*mp,
    910	struct xfs_sb		*sbp)
    911{
    912	mp->m_agfrotor = mp->m_agirotor = 0;
    913	mp->m_maxagi = mp->m_sb.sb_agcount;
    914	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
    915	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
    916	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
    917	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
    918	mp->m_blockmask = sbp->sb_blocksize - 1;
    919	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
    920	mp->m_blockwmask = mp->m_blockwsize - 1;
    921
    922	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
    923	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
    924	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
    925	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
    926
    927	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
    928	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
    929	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
    930	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
    931
    932	mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 1);
    933	mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(sbp->sb_blocksize, 0);
    934	mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2;
    935	mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2;
    936
    937	mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, true);
    938	mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(sbp->sb_blocksize, false);
    939	mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2;
    940	mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2;
    941
    942	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
    943	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
    944	mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp);
    945}
    946
    947/*
    948 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock
    949 * into the superblock buffer to be logged.  It does not provide the higher
    950 * level of locking that is needed to protect the in-core superblock from
    951 * concurrent access.
    952 */
    953void
    954xfs_log_sb(
    955	struct xfs_trans	*tp)
    956{
    957	struct xfs_mount	*mp = tp->t_mountp;
    958	struct xfs_buf		*bp = xfs_trans_getsb(tp);
    959
    960	/*
    961	 * Lazy sb counters don't update the in-core superblock so do that now.
    962	 * If this is at unmount, the counters will be exactly correct, but at
    963	 * any other time they will only be ballpark correct because of
    964	 * reservations that have been taken out percpu counters. If we have an
    965	 * unclean shutdown, this will be corrected by log recovery rebuilding
    966	 * the counters from the AGF block counts.
    967	 *
    968	 * Do not update sb_frextents here because it is not part of the lazy
    969	 * sb counters, despite having a percpu counter. It is always kept
    970	 * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas()
    971	 * and hence we don't need have to update it here.
    972	 */
    973	if (xfs_has_lazysbcount(mp)) {
    974		mp->m_sb.sb_icount = percpu_counter_sum(&mp->m_icount);
    975		mp->m_sb.sb_ifree = percpu_counter_sum(&mp->m_ifree);
    976		mp->m_sb.sb_fdblocks = percpu_counter_sum(&mp->m_fdblocks);
    977	}
    978
    979	xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
    980	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
    981	xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
    982}
    983
    984/*
    985 * xfs_sync_sb
    986 *
    987 * Sync the superblock to disk.
    988 *
    989 * Note that the caller is responsible for checking the frozen state of the
    990 * filesystem. This procedure uses the non-blocking transaction allocator and
    991 * thus will allow modifications to a frozen fs. This is required because this
    992 * code can be called during the process of freezing where use of the high-level
    993 * allocator would deadlock.
    994 */
    995int
    996xfs_sync_sb(
    997	struct xfs_mount	*mp,
    998	bool			wait)
    999{
   1000	struct xfs_trans	*tp;
   1001	int			error;
   1002
   1003	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0,
   1004			XFS_TRANS_NO_WRITECOUNT, &tp);
   1005	if (error)
   1006		return error;
   1007
   1008	xfs_log_sb(tp);
   1009	if (wait)
   1010		xfs_trans_set_sync(tp);
   1011	return xfs_trans_commit(tp);
   1012}
   1013
   1014/*
   1015 * Update all the secondary superblocks to match the new state of the primary.
   1016 * Because we are completely overwriting all the existing fields in the
   1017 * secondary superblock buffers, there is no need to read them in from disk.
   1018 * Just get a new buffer, stamp it and write it.
   1019 *
   1020 * The sb buffers need to be cached here so that we serialise against other
   1021 * operations that access the secondary superblocks, but we don't want to keep
   1022 * them in memory once it is written so we mark it as a one-shot buffer.
   1023 */
   1024int
   1025xfs_update_secondary_sbs(
   1026	struct xfs_mount	*mp)
   1027{
   1028	struct xfs_perag	*pag;
   1029	xfs_agnumber_t		agno = 1;
   1030	int			saved_error = 0;
   1031	int			error = 0;
   1032	LIST_HEAD		(buffer_list);
   1033
   1034	/* update secondary superblocks. */
   1035	for_each_perag_from(mp, agno, pag) {
   1036		struct xfs_buf		*bp;
   1037
   1038		error = xfs_buf_get(mp->m_ddev_targp,
   1039				 XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR),
   1040				 XFS_FSS_TO_BB(mp, 1), &bp);
   1041		/*
   1042		 * If we get an error reading or writing alternate superblocks,
   1043		 * continue.  xfs_repair chooses the "best" superblock based
   1044		 * on most matches; if we break early, we'll leave more
   1045		 * superblocks un-updated than updated, and xfs_repair may
   1046		 * pick them over the properly-updated primary.
   1047		 */
   1048		if (error) {
   1049			xfs_warn(mp,
   1050		"error allocating secondary superblock for ag %d",
   1051				pag->pag_agno);
   1052			if (!saved_error)
   1053				saved_error = error;
   1054			continue;
   1055		}
   1056
   1057		bp->b_ops = &xfs_sb_buf_ops;
   1058		xfs_buf_oneshot(bp);
   1059		xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
   1060		xfs_sb_to_disk(bp->b_addr, &mp->m_sb);
   1061		xfs_buf_delwri_queue(bp, &buffer_list);
   1062		xfs_buf_relse(bp);
   1063
   1064		/* don't hold too many buffers at once */
   1065		if (agno % 16)
   1066			continue;
   1067
   1068		error = xfs_buf_delwri_submit(&buffer_list);
   1069		if (error) {
   1070			xfs_warn(mp,
   1071		"write error %d updating a secondary superblock near ag %d",
   1072				error, pag->pag_agno);
   1073			if (!saved_error)
   1074				saved_error = error;
   1075			continue;
   1076		}
   1077	}
   1078	error = xfs_buf_delwri_submit(&buffer_list);
   1079	if (error) {
   1080		xfs_warn(mp,
   1081		"write error %d updating a secondary superblock near ag %d",
   1082			error, agno);
   1083	}
   1084
   1085	return saved_error ? saved_error : error;
   1086}
   1087
   1088/*
   1089 * Same behavior as xfs_sync_sb, except that it is always synchronous and it
   1090 * also writes the superblock buffer to disk sector 0 immediately.
   1091 */
   1092int
   1093xfs_sync_sb_buf(
   1094	struct xfs_mount	*mp)
   1095{
   1096	struct xfs_trans	*tp;
   1097	struct xfs_buf		*bp;
   1098	int			error;
   1099
   1100	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp);
   1101	if (error)
   1102		return error;
   1103
   1104	bp = xfs_trans_getsb(tp);
   1105	xfs_log_sb(tp);
   1106	xfs_trans_bhold(tp, bp);
   1107	xfs_trans_set_sync(tp);
   1108	error = xfs_trans_commit(tp);
   1109	if (error)
   1110		goto out;
   1111	/*
   1112	 * write out the sb buffer to get the changes to disk
   1113	 */
   1114	error = xfs_bwrite(bp);
   1115out:
   1116	xfs_buf_relse(bp);
   1117	return error;
   1118}
   1119
   1120void
   1121xfs_fs_geometry(
   1122	struct xfs_mount	*mp,
   1123	struct xfs_fsop_geom	*geo,
   1124	int			struct_version)
   1125{
   1126	struct xfs_sb		*sbp = &mp->m_sb;
   1127
   1128	memset(geo, 0, sizeof(struct xfs_fsop_geom));
   1129
   1130	geo->blocksize = sbp->sb_blocksize;
   1131	geo->rtextsize = sbp->sb_rextsize;
   1132	geo->agblocks = sbp->sb_agblocks;
   1133	geo->agcount = sbp->sb_agcount;
   1134	geo->logblocks = sbp->sb_logblocks;
   1135	geo->sectsize = sbp->sb_sectsize;
   1136	geo->inodesize = sbp->sb_inodesize;
   1137	geo->imaxpct = sbp->sb_imax_pct;
   1138	geo->datablocks = sbp->sb_dblocks;
   1139	geo->rtblocks = sbp->sb_rblocks;
   1140	geo->rtextents = sbp->sb_rextents;
   1141	geo->logstart = sbp->sb_logstart;
   1142	BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid));
   1143	memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid));
   1144
   1145	if (struct_version < 2)
   1146		return;
   1147
   1148	geo->sunit = sbp->sb_unit;
   1149	geo->swidth = sbp->sb_width;
   1150
   1151	if (struct_version < 3)
   1152		return;
   1153
   1154	geo->version = XFS_FSOP_GEOM_VERSION;
   1155	geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
   1156		     XFS_FSOP_GEOM_FLAGS_DIRV2 |
   1157		     XFS_FSOP_GEOM_FLAGS_EXTFLG;
   1158	if (xfs_has_attr(mp))
   1159		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR;
   1160	if (xfs_has_quota(mp))
   1161		geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA;
   1162	if (xfs_has_align(mp))
   1163		geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN;
   1164	if (xfs_has_dalign(mp))
   1165		geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN;
   1166	if (xfs_has_asciici(mp))
   1167		geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI;
   1168	if (xfs_has_lazysbcount(mp))
   1169		geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB;
   1170	if (xfs_has_attr2(mp))
   1171		geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2;
   1172	if (xfs_has_projid32(mp))
   1173		geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32;
   1174	if (xfs_has_crc(mp))
   1175		geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB;
   1176	if (xfs_has_ftype(mp))
   1177		geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE;
   1178	if (xfs_has_finobt(mp))
   1179		geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT;
   1180	if (xfs_has_sparseinodes(mp))
   1181		geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES;
   1182	if (xfs_has_rmapbt(mp))
   1183		geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT;
   1184	if (xfs_has_reflink(mp))
   1185		geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK;
   1186	if (xfs_has_bigtime(mp))
   1187		geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME;
   1188	if (xfs_has_inobtcounts(mp))
   1189		geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT;
   1190	if (xfs_has_sector(mp)) {
   1191		geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR;
   1192		geo->logsectsize = sbp->sb_logsectsize;
   1193	} else {
   1194		geo->logsectsize = BBSIZE;
   1195	}
   1196	if (xfs_has_large_extent_counts(mp))
   1197		geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64;
   1198	geo->rtsectsize = sbp->sb_blocksize;
   1199	geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp);
   1200
   1201	if (struct_version < 4)
   1202		return;
   1203
   1204	if (xfs_has_logv2(mp))
   1205		geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2;
   1206
   1207	geo->logsunit = sbp->sb_logsunit;
   1208
   1209	if (struct_version < 5)
   1210		return;
   1211
   1212	geo->version = XFS_FSOP_GEOM_VERSION_V5;
   1213}
   1214
   1215/* Read a secondary superblock. */
   1216int
   1217xfs_sb_read_secondary(
   1218	struct xfs_mount	*mp,
   1219	struct xfs_trans	*tp,
   1220	xfs_agnumber_t		agno,
   1221	struct xfs_buf		**bpp)
   1222{
   1223	struct xfs_buf		*bp;
   1224	int			error;
   1225
   1226	ASSERT(agno != 0 && agno != NULLAGNUMBER);
   1227	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
   1228			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
   1229			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops);
   1230	if (error)
   1231		return error;
   1232	xfs_buf_set_ref(bp, XFS_SSB_REF);
   1233	*bpp = bp;
   1234	return 0;
   1235}
   1236
   1237/* Get an uninitialised secondary superblock buffer. */
   1238int
   1239xfs_sb_get_secondary(
   1240	struct xfs_mount	*mp,
   1241	struct xfs_trans	*tp,
   1242	xfs_agnumber_t		agno,
   1243	struct xfs_buf		**bpp)
   1244{
   1245	struct xfs_buf		*bp;
   1246	int			error;
   1247
   1248	ASSERT(agno != 0 && agno != NULLAGNUMBER);
   1249	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
   1250			XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
   1251			XFS_FSS_TO_BB(mp, 1), 0, &bp);
   1252	if (error)
   1253		return error;
   1254	bp->b_ops = &xfs_sb_buf_ops;
   1255	xfs_buf_oneshot(bp);
   1256	*bpp = bp;
   1257	return 0;
   1258}
   1259
   1260/*
   1261 * sunit, swidth, sectorsize(optional with 0) should be all in bytes,
   1262 * so users won't be confused by values in error messages.
   1263 */
   1264bool
   1265xfs_validate_stripe_geometry(
   1266	struct xfs_mount	*mp,
   1267	__s64			sunit,
   1268	__s64			swidth,
   1269	int			sectorsize,
   1270	bool			silent)
   1271{
   1272	if (swidth > INT_MAX) {
   1273		if (!silent)
   1274			xfs_notice(mp,
   1275"stripe width (%lld) is too large", swidth);
   1276		return false;
   1277	}
   1278
   1279	if (sunit > swidth) {
   1280		if (!silent)
   1281			xfs_notice(mp,
   1282"stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth);
   1283		return false;
   1284	}
   1285
   1286	if (sectorsize && (int)sunit % sectorsize) {
   1287		if (!silent)
   1288			xfs_notice(mp,
   1289"stripe unit (%lld) must be a multiple of the sector size (%d)",
   1290				   sunit, sectorsize);
   1291		return false;
   1292	}
   1293
   1294	if (sunit && !swidth) {
   1295		if (!silent)
   1296			xfs_notice(mp,
   1297"invalid stripe unit (%lld) and stripe width of 0", sunit);
   1298		return false;
   1299	}
   1300
   1301	if (!sunit && swidth) {
   1302		if (!silent)
   1303			xfs_notice(mp,
   1304"invalid stripe width (%lld) and stripe unit of 0", swidth);
   1305		return false;
   1306	}
   1307
   1308	if (sunit && (int)swidth % (int)sunit) {
   1309		if (!silent)
   1310			xfs_notice(mp,
   1311"stripe width (%lld) must be a multiple of the stripe unit (%lld)",
   1312				   swidth, sunit);
   1313		return false;
   1314	}
   1315	return true;
   1316}