10b61f8a4SDave Chinner // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 37b718769SNathan Scott * Copyright (c) 2000-2005 Silicon Graphics, Inc. 47b718769SNathan Scott * All Rights Reserved. 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds #include "xfs.h" 7a844f451SNathan Scott #include "xfs_fs.h" 870a9883cSDave Chinner #include "xfs_shared.h" 9239880efSDave Chinner #include "xfs_format.h" 10239880efSDave Chinner #include "xfs_log_format.h" 11239880efSDave Chinner #include "xfs_trans_resv.h" 12a844f451SNathan Scott #include "xfs_bit.h" 131da177e4SLinus Torvalds #include "xfs_sb.h" 141da177e4SLinus Torvalds #include "xfs_mount.h" 151da177e4SLinus Torvalds #include "xfs_inode.h" 16a4fbe6abSDave Chinner #include "xfs_dir2.h" 17a844f451SNathan Scott #include "xfs_ialloc.h" 181da177e4SLinus Torvalds #include "xfs_alloc.h" 191da177e4SLinus Torvalds #include "xfs_rtalloc.h" 201da177e4SLinus Torvalds #include "xfs_bmap.h" 21a4fbe6abSDave Chinner #include "xfs_trans.h" 22a4fbe6abSDave Chinner #include "xfs_trans_priv.h" 23a4fbe6abSDave Chinner #include "xfs_log.h" 24b5f17becSDave Chinner #include "xfs_log_priv.h" 251da177e4SLinus Torvalds #include "xfs_error.h" 261da177e4SLinus Torvalds #include "xfs_quota.h" 271da177e4SLinus Torvalds #include "xfs_fsops.h" 286d8b79cfSDave Chinner #include "xfs_icache.h" 29a31b1d3dSBrian Foster #include "xfs_sysfs.h" 30035e00acSDarrick J. Wong #include "xfs_rmap_btree.h" 311946b91cSDarrick J. Wong #include "xfs_refcount_btree.h" 32174edb0eSDarrick J. Wong #include "xfs_reflink.h" 33ebf55872SChristoph Hellwig #include "xfs_extent_busy.h" 3439353ff6SDarrick J. Wong #include "xfs_health.h" 3513eaec4bSDarrick J. Wong #include "xfs_trace.h" 369bbafc71SDave Chinner #include "xfs_ag.h" 371da177e4SLinus Torvalds 3827174203SChristoph Hellwig static DEFINE_MUTEX(xfs_uuid_table_mutex); 3927174203SChristoph Hellwig static int xfs_uuid_table_size; 4027174203SChristoph Hellwig static uuid_t *xfs_uuid_table; 4127174203SChristoph Hellwig 42af3b6382SDarrick J. Wong void 43af3b6382SDarrick J. Wong xfs_uuid_table_free(void) 44af3b6382SDarrick J. Wong { 45af3b6382SDarrick J. Wong if (xfs_uuid_table_size == 0) 46af3b6382SDarrick J. Wong return; 47af3b6382SDarrick J. Wong kmem_free(xfs_uuid_table); 48af3b6382SDarrick J. Wong xfs_uuid_table = NULL; 49af3b6382SDarrick J. Wong xfs_uuid_table_size = 0; 50af3b6382SDarrick J. Wong } 51af3b6382SDarrick J. Wong 5227174203SChristoph Hellwig /* 5327174203SChristoph Hellwig * See if the UUID is unique among mounted XFS filesystems. 5427174203SChristoph Hellwig * Mount fails if UUID is nil or a FS with the same UUID is already mounted. 5527174203SChristoph Hellwig */ 5627174203SChristoph Hellwig STATIC int 5727174203SChristoph Hellwig xfs_uuid_mount( 5827174203SChristoph Hellwig struct xfs_mount *mp) 5927174203SChristoph Hellwig { 6027174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 6127174203SChristoph Hellwig int hole, i; 6227174203SChristoph Hellwig 638f720d9fSAmir Goldstein /* Publish UUID in struct super_block */ 6485787090SChristoph Hellwig uuid_copy(&mp->m_super->s_uuid, uuid); 658f720d9fSAmir Goldstein 660560f31aSDave Chinner if (xfs_has_nouuid(mp)) 6727174203SChristoph Hellwig return 0; 6827174203SChristoph Hellwig 69d905fdaaSAmir Goldstein if (uuid_is_null(uuid)) { 70d905fdaaSAmir Goldstein xfs_warn(mp, "Filesystem has null UUID - can't mount"); 712451337dSDave Chinner return -EINVAL; 7227174203SChristoph Hellwig } 7327174203SChristoph Hellwig 7427174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 7527174203SChristoph Hellwig for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { 76d905fdaaSAmir Goldstein if (uuid_is_null(&xfs_uuid_table[i])) { 7727174203SChristoph Hellwig hole = i; 7827174203SChristoph Hellwig continue; 7927174203SChristoph Hellwig } 8027174203SChristoph Hellwig if (uuid_equal(uuid, &xfs_uuid_table[i])) 8127174203SChristoph Hellwig goto out_duplicate; 8227174203SChristoph Hellwig } 8327174203SChristoph Hellwig 8427174203SChristoph Hellwig if (hole < 0) { 85771915c4SCarlos Maiolino xfs_uuid_table = krealloc(xfs_uuid_table, 8627174203SChristoph Hellwig (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), 87771915c4SCarlos Maiolino GFP_KERNEL | __GFP_NOFAIL); 8827174203SChristoph Hellwig hole = xfs_uuid_table_size++; 8927174203SChristoph Hellwig } 9027174203SChristoph Hellwig xfs_uuid_table[hole] = *uuid; 9127174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 9227174203SChristoph Hellwig 9327174203SChristoph Hellwig return 0; 9427174203SChristoph Hellwig 9527174203SChristoph Hellwig out_duplicate: 9627174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 97021000e5SMitsuo Hayasaka xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); 982451337dSDave Chinner return -EINVAL; 9927174203SChristoph Hellwig } 10027174203SChristoph Hellwig 10127174203SChristoph Hellwig STATIC void 10227174203SChristoph Hellwig xfs_uuid_unmount( 10327174203SChristoph Hellwig struct xfs_mount *mp) 10427174203SChristoph Hellwig { 10527174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 10627174203SChristoph Hellwig int i; 10727174203SChristoph Hellwig 1080560f31aSDave Chinner if (xfs_has_nouuid(mp)) 10927174203SChristoph Hellwig return; 11027174203SChristoph Hellwig 11127174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 11227174203SChristoph Hellwig for (i = 0; i < xfs_uuid_table_size; i++) { 113d905fdaaSAmir Goldstein if (uuid_is_null(&xfs_uuid_table[i])) 11427174203SChristoph Hellwig continue; 11527174203SChristoph Hellwig if (!uuid_equal(uuid, &xfs_uuid_table[i])) 11627174203SChristoph Hellwig continue; 11727174203SChristoph Hellwig memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); 11827174203SChristoph Hellwig break; 11927174203SChristoph Hellwig } 12027174203SChristoph Hellwig ASSERT(i < xfs_uuid_table_size); 12127174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 12227174203SChristoph Hellwig } 12327174203SChristoph Hellwig 1244cc929eeSNathan Scott /* 1254cc929eeSNathan Scott * Check size of device based on the (data/realtime) block count. 1264cc929eeSNathan Scott * Note: this check is used by the growfs code as well as mount. 1274cc929eeSNathan Scott */ 1284cc929eeSNathan Scott int 1294cc929eeSNathan Scott xfs_sb_validate_fsb_count( 1304cc929eeSNathan Scott xfs_sb_t *sbp, 131c8ce540dSDarrick J. Wong uint64_t nblocks) 1324cc929eeSNathan Scott { 1334cc929eeSNathan Scott ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); 1344cc929eeSNathan Scott ASSERT(sbp->sb_blocklog >= BBSHIFT); 1354cc929eeSNathan Scott 136d5cf09baSChristoph Hellwig /* Limited by ULONG_MAX of page cache index */ 13709cbfeafSKirill A. Shutemov if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) 1382451337dSDave Chinner return -EFBIG; 1394cc929eeSNathan Scott return 0; 1404cc929eeSNathan Scott } 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds /* 1431da177e4SLinus Torvalds * xfs_readsb 1441da177e4SLinus Torvalds * 1451da177e4SLinus Torvalds * Does the initial read of the superblock. 1461da177e4SLinus Torvalds */ 1471da177e4SLinus Torvalds int 148ff55068cSDave Chinner xfs_readsb( 149ff55068cSDave Chinner struct xfs_mount *mp, 150ff55068cSDave Chinner int flags) 1511da177e4SLinus Torvalds { 1521da177e4SLinus Torvalds unsigned int sector_size; 15304a1e6c5SDave Chinner struct xfs_buf *bp; 15404a1e6c5SDave Chinner struct xfs_sb *sbp = &mp->m_sb; 1551da177e4SLinus Torvalds int error; 156af34e09dSDave Chinner int loud = !(flags & XFS_MFSI_QUIET); 157daba5427SEric Sandeen const struct xfs_buf_ops *buf_ops; 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds ASSERT(mp->m_sb_bp == NULL); 1601da177e4SLinus Torvalds ASSERT(mp->m_ddev_targp != NULL); 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds /* 163daba5427SEric Sandeen * For the initial read, we must guess at the sector 164daba5427SEric Sandeen * size based on the block device. It's enough to 165daba5427SEric Sandeen * get the sb_sectsize out of the superblock and 166daba5427SEric Sandeen * then reread with the proper length. 167daba5427SEric Sandeen * We don't verify it yet, because it may not be complete. 168daba5427SEric Sandeen */ 169daba5427SEric Sandeen sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); 170daba5427SEric Sandeen buf_ops = NULL; 171daba5427SEric Sandeen 172daba5427SEric Sandeen /* 173c891c30aSBrian Foster * Allocate a (locked) buffer to hold the superblock. This will be kept 174c891c30aSBrian Foster * around at all times to optimize access to the superblock. Therefore, 175c891c30aSBrian Foster * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count 176c891c30aSBrian Foster * elevated. 1771da177e4SLinus Torvalds */ 17826af6552SDave Chinner reread: 179ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, 180c891c30aSBrian Foster BTOBB(sector_size), XBF_NO_IOACCT, &bp, 181c891c30aSBrian Foster buf_ops); 182ba372674SDave Chinner if (error) { 183eab4e633SDave Chinner if (loud) 184e721f504SDave Chinner xfs_warn(mp, "SB validate failed with error %d.", error); 185ac75a1f7SDave Chinner /* bad CRC means corrupted metadata */ 1862451337dSDave Chinner if (error == -EFSBADCRC) 1872451337dSDave Chinner error = -EFSCORRUPTED; 188ba372674SDave Chinner return error; 189eab4e633SDave Chinner } 1901da177e4SLinus Torvalds 1911da177e4SLinus Torvalds /* 1921da177e4SLinus Torvalds * Initialize the mount structure from the superblock. 1931da177e4SLinus Torvalds */ 1943e6e8afdSChristoph Hellwig xfs_sb_from_disk(sbp, bp->b_addr); 195556b8883SDave Chinner 196556b8883SDave Chinner /* 197556b8883SDave Chinner * If we haven't validated the superblock, do so now before we try 198556b8883SDave Chinner * to check the sector size and reread the superblock appropriately. 199556b8883SDave Chinner */ 200556b8883SDave Chinner if (sbp->sb_magicnum != XFS_SB_MAGIC) { 201556b8883SDave Chinner if (loud) 202556b8883SDave Chinner xfs_warn(mp, "Invalid superblock magic number"); 2032451337dSDave Chinner error = -EINVAL; 204556b8883SDave Chinner goto release_buf; 205556b8883SDave Chinner } 206ff55068cSDave Chinner 2071da177e4SLinus Torvalds /* 2081da177e4SLinus Torvalds * We must be able to do sector-sized and sector-aligned IO. 2091da177e4SLinus Torvalds */ 21004a1e6c5SDave Chinner if (sector_size > sbp->sb_sectsize) { 211af34e09dSDave Chinner if (loud) 212af34e09dSDave Chinner xfs_warn(mp, "device supports %u byte sectors (not %u)", 21304a1e6c5SDave Chinner sector_size, sbp->sb_sectsize); 2142451337dSDave Chinner error = -ENOSYS; 21526af6552SDave Chinner goto release_buf; 2161da177e4SLinus Torvalds } 2171da177e4SLinus Torvalds 218556b8883SDave Chinner if (buf_ops == NULL) { 2191da177e4SLinus Torvalds /* 220daba5427SEric Sandeen * Re-read the superblock so the buffer is correctly sized, 221daba5427SEric Sandeen * and properly verified. 2221da177e4SLinus Torvalds */ 2231da177e4SLinus Torvalds xfs_buf_relse(bp); 22404a1e6c5SDave Chinner sector_size = sbp->sb_sectsize; 225daba5427SEric Sandeen buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; 22626af6552SDave Chinner goto reread; 2271da177e4SLinus Torvalds } 2281da177e4SLinus Torvalds 229a1d86e8dSDave Chinner mp->m_features |= xfs_sb_version_to_features(sbp); 2305681ca40SDave Chinner xfs_reinit_percpu_counters(mp); 2318d280b98SDavid Chinner 23204a1e6c5SDave Chinner /* no need to be quiet anymore, so reset the buf ops */ 23304a1e6c5SDave Chinner bp->b_ops = &xfs_sb_buf_ops; 23404a1e6c5SDave Chinner 2351da177e4SLinus Torvalds mp->m_sb_bp = bp; 23626af6552SDave Chinner xfs_buf_unlock(bp); 2371da177e4SLinus Torvalds return 0; 2381da177e4SLinus Torvalds 23926af6552SDave Chinner release_buf: 2401da177e4SLinus Torvalds xfs_buf_relse(bp); 2411da177e4SLinus Torvalds return error; 2421da177e4SLinus Torvalds } 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds /* 24513eaec4bSDarrick J. Wong * If the sunit/swidth change would move the precomputed root inode value, we 24613eaec4bSDarrick J. Wong * must reject the ondisk change because repair will stumble over that. 24713eaec4bSDarrick J. Wong * However, we allow the mount to proceed because we never rejected this 24813eaec4bSDarrick J. Wong * combination before. Returns true to update the sb, false otherwise. 24913eaec4bSDarrick J. Wong */ 25013eaec4bSDarrick J. Wong static inline int 25113eaec4bSDarrick J. Wong xfs_check_new_dalign( 25213eaec4bSDarrick J. Wong struct xfs_mount *mp, 25313eaec4bSDarrick J. Wong int new_dalign, 25413eaec4bSDarrick J. Wong bool *update_sb) 25513eaec4bSDarrick J. Wong { 25613eaec4bSDarrick J. Wong struct xfs_sb *sbp = &mp->m_sb; 25713eaec4bSDarrick J. Wong xfs_ino_t calc_ino; 25813eaec4bSDarrick J. Wong 25913eaec4bSDarrick J. Wong calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign); 26013eaec4bSDarrick J. Wong trace_xfs_check_new_dalign(mp, new_dalign, calc_ino); 26113eaec4bSDarrick J. Wong 26213eaec4bSDarrick J. Wong if (sbp->sb_rootino == calc_ino) { 26313eaec4bSDarrick J. Wong *update_sb = true; 26413eaec4bSDarrick J. Wong return 0; 26513eaec4bSDarrick J. Wong } 26613eaec4bSDarrick J. Wong 26713eaec4bSDarrick J. Wong xfs_warn(mp, 26813eaec4bSDarrick J. Wong "Cannot change stripe alignment; would require moving root inode."); 26913eaec4bSDarrick J. Wong 27013eaec4bSDarrick J. Wong /* 27113eaec4bSDarrick J. Wong * XXX: Next time we add a new incompat feature, this should start 27213eaec4bSDarrick J. Wong * returning -EINVAL to fail the mount. Until then, spit out a warning 27313eaec4bSDarrick J. Wong * that we're ignoring the administrator's instructions. 27413eaec4bSDarrick J. Wong */ 27513eaec4bSDarrick J. Wong xfs_warn(mp, "Skipping superblock stripe alignment update."); 27613eaec4bSDarrick J. Wong *update_sb = false; 27713eaec4bSDarrick J. Wong return 0; 27813eaec4bSDarrick J. Wong } 27913eaec4bSDarrick J. Wong 28013eaec4bSDarrick J. Wong /* 2814f5b1b3aSDarrick J. Wong * If we were provided with new sunit/swidth values as mount options, make sure 2824f5b1b3aSDarrick J. Wong * that they pass basic alignment and superblock feature checks, and convert 2834f5b1b3aSDarrick J. Wong * them into the same units (FSB) that everything else expects. This step 2844f5b1b3aSDarrick J. Wong * /must/ be done before computing the inode geometry. 2851da177e4SLinus Torvalds */ 2860771fb45SEric Sandeen STATIC int 2874f5b1b3aSDarrick J. Wong xfs_validate_new_dalign( 2884f5b1b3aSDarrick J. Wong struct xfs_mount *mp) 2891da177e4SLinus Torvalds { 2904f5b1b3aSDarrick J. Wong if (mp->m_dalign == 0) 2914f5b1b3aSDarrick J. Wong return 0; 2921da177e4SLinus Torvalds 2931da177e4SLinus Torvalds /* 2941da177e4SLinus Torvalds * If stripe unit and stripe width are not multiples 2951da177e4SLinus Torvalds * of the fs blocksize turn off alignment. 2961da177e4SLinus Torvalds */ 2971da177e4SLinus Torvalds if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || 2981da177e4SLinus Torvalds (BBTOB(mp->m_swidth) & mp->m_blockmask)) { 29939a45d84SJie Liu xfs_warn(mp, 30039a45d84SJie Liu "alignment check failed: sunit/swidth vs. blocksize(%d)", 3014f5b1b3aSDarrick J. Wong mp->m_sb.sb_blocksize); 3022451337dSDave Chinner return -EINVAL; 303*de94a2e1SZeng Heng } 304*de94a2e1SZeng Heng 3051da177e4SLinus Torvalds /* 3061da177e4SLinus Torvalds * Convert the stripe unit and width to FSBs. 3071da177e4SLinus Torvalds */ 3081da177e4SLinus Torvalds mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); 3094f5b1b3aSDarrick J. Wong if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) { 31053487786SDave Chinner xfs_warn(mp, 31139a45d84SJie Liu "alignment check failed: sunit/swidth vs. agsize(%d)", 3124f5b1b3aSDarrick J. Wong mp->m_sb.sb_agblocks); 3132451337dSDave Chinner return -EINVAL; 314*de94a2e1SZeng Heng } 315*de94a2e1SZeng Heng 316*de94a2e1SZeng Heng if (!mp->m_dalign) { 31739a45d84SJie Liu xfs_warn(mp, 31839a45d84SJie Liu "alignment check failed: sunit(%d) less than bsize(%d)", 3194f5b1b3aSDarrick J. Wong mp->m_dalign, mp->m_sb.sb_blocksize); 3202451337dSDave Chinner return -EINVAL; 3211da177e4SLinus Torvalds } 322*de94a2e1SZeng Heng 323*de94a2e1SZeng Heng mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); 3241da177e4SLinus Torvalds 32538c26bfdSDave Chinner if (!xfs_has_dalign(mp)) { 32634d7f603SJie Liu xfs_warn(mp, 32734d7f603SJie Liu "cannot change alignment: superblock does not support data alignment"); 3282451337dSDave Chinner return -EINVAL; 3291da177e4SLinus Torvalds } 3304f5b1b3aSDarrick J. Wong 3314f5b1b3aSDarrick J. Wong return 0; 3324f5b1b3aSDarrick J. Wong } 3334f5b1b3aSDarrick J. Wong 3344f5b1b3aSDarrick J. Wong /* Update alignment values based on mount options and sb values. */ 3354f5b1b3aSDarrick J. Wong STATIC int 3364f5b1b3aSDarrick J. Wong xfs_update_alignment( 3374f5b1b3aSDarrick J. Wong struct xfs_mount *mp) 3384f5b1b3aSDarrick J. Wong { 3394f5b1b3aSDarrick J. Wong struct xfs_sb *sbp = &mp->m_sb; 3404f5b1b3aSDarrick J. Wong 3414f5b1b3aSDarrick J. Wong if (mp->m_dalign) { 34213eaec4bSDarrick J. Wong bool update_sb; 34313eaec4bSDarrick J. Wong int error; 34413eaec4bSDarrick J. Wong 3454f5b1b3aSDarrick J. Wong if (sbp->sb_unit == mp->m_dalign && 3464f5b1b3aSDarrick J. Wong sbp->sb_width == mp->m_swidth) 3474f5b1b3aSDarrick J. Wong return 0; 3484f5b1b3aSDarrick J. Wong 34913eaec4bSDarrick J. Wong error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb); 35013eaec4bSDarrick J. Wong if (error || !update_sb) 35113eaec4bSDarrick J. Wong return error; 35213eaec4bSDarrick J. Wong 3534f5b1b3aSDarrick J. Wong sbp->sb_unit = mp->m_dalign; 3544f5b1b3aSDarrick J. Wong sbp->sb_width = mp->m_swidth; 3554f5b1b3aSDarrick J. Wong mp->m_update_sb = true; 3560560f31aSDave Chinner } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) { 3571da177e4SLinus Torvalds mp->m_dalign = sbp->sb_unit; 3581da177e4SLinus Torvalds mp->m_swidth = sbp->sb_width; 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 3610771fb45SEric Sandeen return 0; 3620771fb45SEric Sandeen } 3631da177e4SLinus Torvalds 3640771fb45SEric Sandeen /* 365055388a3SDave Chinner * precalculate the low space thresholds for dynamic speculative preallocation. 366055388a3SDave Chinner */ 367055388a3SDave Chinner void 368055388a3SDave Chinner xfs_set_low_space_thresholds( 369055388a3SDave Chinner struct xfs_mount *mp) 370055388a3SDave Chinner { 37165f03d86SDarrick J. Wong uint64_t dblocks = mp->m_sb.sb_dblocks; 37265f03d86SDarrick J. Wong uint64_t rtexts = mp->m_sb.sb_rextents; 373055388a3SDave Chinner int i; 374055388a3SDave Chinner 37565f03d86SDarrick J. Wong do_div(dblocks, 100); 37665f03d86SDarrick J. Wong do_div(rtexts, 100); 377055388a3SDave Chinner 37865f03d86SDarrick J. Wong for (i = 0; i < XFS_LOWSP_MAX; i++) { 37965f03d86SDarrick J. Wong mp->m_low_space[i] = dblocks * (i + 1); 38065f03d86SDarrick J. Wong mp->m_low_rtexts[i] = rtexts * (i + 1); 381055388a3SDave Chinner } 382055388a3SDave Chinner } 383055388a3SDave Chinner 3841da177e4SLinus Torvalds /* 3850471f62eSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 3861da177e4SLinus Torvalds */ 3870771fb45SEric Sandeen STATIC int 388ba372674SDave Chinner xfs_check_sizes( 389ba372674SDave Chinner struct xfs_mount *mp) 3900771fb45SEric Sandeen { 391ba372674SDave Chinner struct xfs_buf *bp; 3920771fb45SEric Sandeen xfs_daddr_t d; 393ba372674SDave Chinner int error; 3940771fb45SEric Sandeen 3951da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 3961da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { 3970b932cccSDave Chinner xfs_warn(mp, "filesystem size mismatch detected"); 3982451337dSDave Chinner return -EFBIG; 3991da177e4SLinus Torvalds } 400ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, 4011da177e4SLinus Torvalds d - XFS_FSS_TO_BB(mp, 1), 402ba372674SDave Chinner XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 403ba372674SDave Chinner if (error) { 4040b932cccSDave Chinner xfs_warn(mp, "last sector read failed"); 405ba372674SDave Chinner return error; 4061da177e4SLinus Torvalds } 4071922c949SDave Chinner xfs_buf_relse(bp); 4081da177e4SLinus Torvalds 409ba372674SDave Chinner if (mp->m_logdev_targp == mp->m_ddev_targp) 410ba372674SDave Chinner return 0; 411ba372674SDave Chinner 4121da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 4131da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { 4140b932cccSDave Chinner xfs_warn(mp, "log size mismatch detected"); 4152451337dSDave Chinner return -EFBIG; 4161da177e4SLinus Torvalds } 417ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_logdev_targp, 4181da177e4SLinus Torvalds d - XFS_FSB_TO_BB(mp, 1), 419ba372674SDave Chinner XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); 420ba372674SDave Chinner if (error) { 4210b932cccSDave Chinner xfs_warn(mp, "log device read failed"); 422ba372674SDave Chinner return error; 4231da177e4SLinus Torvalds } 4241922c949SDave Chinner xfs_buf_relse(bp); 4250771fb45SEric Sandeen return 0; 4260771fb45SEric Sandeen } 4270771fb45SEric Sandeen 4280771fb45SEric Sandeen /* 4297d095257SChristoph Hellwig * Clear the quotaflags in memory and in the superblock. 4307d095257SChristoph Hellwig */ 4317d095257SChristoph Hellwig int 4327d095257SChristoph Hellwig xfs_mount_reset_sbqflags( 4337d095257SChristoph Hellwig struct xfs_mount *mp) 4347d095257SChristoph Hellwig { 4357d095257SChristoph Hellwig mp->m_qflags = 0; 4367d095257SChristoph Hellwig 43761e63ecbSDave Chinner /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ 4387d095257SChristoph Hellwig if (mp->m_sb.sb_qflags == 0) 4397d095257SChristoph Hellwig return 0; 4407d095257SChristoph Hellwig spin_lock(&mp->m_sb_lock); 4417d095257SChristoph Hellwig mp->m_sb.sb_qflags = 0; 4427d095257SChristoph Hellwig spin_unlock(&mp->m_sb_lock); 4437d095257SChristoph Hellwig 44461e63ecbSDave Chinner if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) 4457d095257SChristoph Hellwig return 0; 4467d095257SChristoph Hellwig 44761e63ecbSDave Chinner return xfs_sync_sb(mp, false); 4487d095257SChristoph Hellwig } 4497d095257SChristoph Hellwig 450c8ce540dSDarrick J. Wong uint64_t 451d5db0f97SEric Sandeen xfs_default_resblks(xfs_mount_t *mp) 452d5db0f97SEric Sandeen { 453c8ce540dSDarrick J. Wong uint64_t resblks; 454d5db0f97SEric Sandeen 455d5db0f97SEric Sandeen /* 4568babd8a2SDave Chinner * We default to 5% or 8192 fsbs of space reserved, whichever is 4578babd8a2SDave Chinner * smaller. This is intended to cover concurrent allocation 4588babd8a2SDave Chinner * transactions when we initially hit enospc. These each require a 4 4598babd8a2SDave Chinner * block reservation. Hence by default we cover roughly 2000 concurrent 4608babd8a2SDave Chinner * allocation reservations. 461d5db0f97SEric Sandeen */ 462d5db0f97SEric Sandeen resblks = mp->m_sb.sb_dblocks; 463d5db0f97SEric Sandeen do_div(resblks, 20); 464c8ce540dSDarrick J. Wong resblks = min_t(uint64_t, resblks, 8192); 465d5db0f97SEric Sandeen return resblks; 466d5db0f97SEric Sandeen } 467d5db0f97SEric Sandeen 4682e9e6481SDarrick J. Wong /* Ensure the summary counts are correct. */ 4692e9e6481SDarrick J. Wong STATIC int 4702e9e6481SDarrick J. Wong xfs_check_summary_counts( 4712e9e6481SDarrick J. Wong struct xfs_mount *mp) 4722e9e6481SDarrick J. Wong { 4735a605fd6SDarrick J. Wong int error = 0; 4745a605fd6SDarrick J. Wong 4752e9e6481SDarrick J. Wong /* 4762e9e6481SDarrick J. Wong * The AG0 superblock verifier rejects in-progress filesystems, 4772e9e6481SDarrick J. Wong * so we should never see the flag set this far into mounting. 4782e9e6481SDarrick J. Wong */ 4792e9e6481SDarrick J. Wong if (mp->m_sb.sb_inprogress) { 4802e9e6481SDarrick J. Wong xfs_err(mp, "sb_inprogress set after log recovery??"); 4812e9e6481SDarrick J. Wong WARN_ON(1); 4822e9e6481SDarrick J. Wong return -EFSCORRUPTED; 4832e9e6481SDarrick J. Wong } 4842e9e6481SDarrick J. Wong 4852e9e6481SDarrick J. Wong /* 4862e9e6481SDarrick J. Wong * Now the log is mounted, we know if it was an unclean shutdown or 4872e9e6481SDarrick J. Wong * not. If it was, with the first phase of recovery has completed, we 4882e9e6481SDarrick J. Wong * have consistent AG blocks on disk. We have not recovered EFIs yet, 4892e9e6481SDarrick J. Wong * but they are recovered transactionally in the second recovery phase 4902e9e6481SDarrick J. Wong * later. 4912e9e6481SDarrick J. Wong * 4922e9e6481SDarrick J. Wong * If the log was clean when we mounted, we can check the summary 4932e9e6481SDarrick J. Wong * counters. If any of them are obviously incorrect, we can recompute 4942e9e6481SDarrick J. Wong * them from the AGF headers in the next step. 4952e9e6481SDarrick J. Wong */ 4962e973b2cSDave Chinner if (xfs_is_clean(mp) && 4972e9e6481SDarrick J. Wong (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks || 49800d22a1cSDarrick J. Wong !xfs_verify_icount(mp, mp->m_sb.sb_icount) || 4992e9e6481SDarrick J. Wong mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) 50039353ff6SDarrick J. Wong xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); 5012e9e6481SDarrick J. Wong 5022e9e6481SDarrick J. Wong /* 5032e9e6481SDarrick J. Wong * We can safely re-initialise incore superblock counters from the 5042e9e6481SDarrick J. Wong * per-ag data. These may not be correct if the filesystem was not 5052e9e6481SDarrick J. Wong * cleanly unmounted, so we waited for recovery to finish before doing 5062e9e6481SDarrick J. Wong * this. 5072e9e6481SDarrick J. Wong * 5082e9e6481SDarrick J. Wong * If the filesystem was cleanly unmounted or the previous check did 5092e9e6481SDarrick J. Wong * not flag anything weird, then we can trust the values in the 5102e9e6481SDarrick J. Wong * superblock to be correct and we don't need to do anything here. 5112e9e6481SDarrick J. Wong * Otherwise, recalculate the summary counters. 5122e9e6481SDarrick J. Wong */ 5135a605fd6SDarrick J. Wong if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) || 5145a605fd6SDarrick J. Wong xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) { 5155a605fd6SDarrick J. Wong error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount); 5165a605fd6SDarrick J. Wong if (error) 5175a605fd6SDarrick J. Wong return error; 5185a605fd6SDarrick J. Wong } 5192e9e6481SDarrick J. Wong 5205a605fd6SDarrick J. Wong /* 5215a605fd6SDarrick J. Wong * Older kernels misused sb_frextents to reflect both incore 5225a605fd6SDarrick J. Wong * reservations made by running transactions and the actual count of 5235a605fd6SDarrick J. Wong * free rt extents in the ondisk metadata. Transactions committed 5245a605fd6SDarrick J. Wong * during runtime can therefore contain a superblock update that 5255a605fd6SDarrick J. Wong * undercounts the number of free rt extents tracked in the rt bitmap. 5265a605fd6SDarrick J. Wong * A clean unmount record will have the correct frextents value since 5275a605fd6SDarrick J. Wong * there can be no other transactions running at that point. 5285a605fd6SDarrick J. Wong * 5295a605fd6SDarrick J. Wong * If we're mounting the rt volume after recovering the log, recompute 5305a605fd6SDarrick J. Wong * frextents from the rtbitmap file to fix the inconsistency. 5315a605fd6SDarrick J. Wong */ 5325a605fd6SDarrick J. Wong if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) { 5335a605fd6SDarrick J. Wong error = xfs_rtalloc_reinit_frextents(mp); 5345a605fd6SDarrick J. Wong if (error) 5355a605fd6SDarrick J. Wong return error; 5365a605fd6SDarrick J. Wong } 5375a605fd6SDarrick J. Wong 5385a605fd6SDarrick J. Wong return 0; 5392e9e6481SDarrick J. Wong } 5402e9e6481SDarrick J. Wong 5417d095257SChristoph Hellwig /* 542d336f7ebSDarrick J. Wong * Flush and reclaim dirty inodes in preparation for unmount. Inodes and 543d336f7ebSDarrick J. Wong * internal inode structures can be sitting in the CIL and AIL at this point, 544d336f7ebSDarrick J. Wong * so we need to unpin them, write them back and/or reclaim them before unmount 545ab23a776SDave Chinner * can proceed. In other words, callers are required to have inactivated all 546ab23a776SDave Chinner * inodes. 547d336f7ebSDarrick J. Wong * 548d336f7ebSDarrick J. Wong * An inode cluster that has been freed can have its buffer still pinned in 549d336f7ebSDarrick J. Wong * memory because the transaction is still sitting in a iclog. The stale inodes 550d336f7ebSDarrick J. Wong * on that buffer will be pinned to the buffer until the transaction hits the 551d336f7ebSDarrick J. Wong * disk and the callbacks run. Pushing the AIL will skip the stale inodes and 552d336f7ebSDarrick J. Wong * may never see the pinned buffer, so nothing will push out the iclog and 553d336f7ebSDarrick J. Wong * unpin the buffer. 554d336f7ebSDarrick J. Wong * 555d336f7ebSDarrick J. Wong * Hence we need to force the log to unpin everything first. However, log 556d336f7ebSDarrick J. Wong * forces don't wait for the discards they issue to complete, so we have to 557d336f7ebSDarrick J. Wong * explicitly wait for them to complete here as well. 558d336f7ebSDarrick J. Wong * 559d336f7ebSDarrick J. Wong * Then we can tell the world we are unmounting so that error handling knows 560d336f7ebSDarrick J. Wong * that the filesystem is going away and we should error out anything that we 561d336f7ebSDarrick J. Wong * have been retrying in the background. This will prevent never-ending 562d336f7ebSDarrick J. Wong * retries in AIL pushing from hanging the unmount. 563d336f7ebSDarrick J. Wong * 564d336f7ebSDarrick J. Wong * Finally, we can push the AIL to clean all the remaining dirty objects, then 565d336f7ebSDarrick J. Wong * reclaim the remaining inodes that are still in memory at this point in time. 566d336f7ebSDarrick J. Wong */ 567d336f7ebSDarrick J. Wong static void 568d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes( 569d336f7ebSDarrick J. Wong struct xfs_mount *mp) 570d336f7ebSDarrick J. Wong { 571d336f7ebSDarrick J. Wong xfs_log_force(mp, XFS_LOG_SYNC); 572d336f7ebSDarrick J. Wong xfs_extent_busy_wait_all(mp); 573d336f7ebSDarrick J. Wong flush_workqueue(xfs_discard_wq); 574d336f7ebSDarrick J. Wong 5752e973b2cSDave Chinner set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate); 576d336f7ebSDarrick J. Wong 577d336f7ebSDarrick J. Wong xfs_ail_push_all_sync(mp->m_ail); 578ab23a776SDave Chinner xfs_inodegc_stop(mp); 579d336f7ebSDarrick J. Wong cancel_delayed_work_sync(&mp->m_reclaim_work); 580d336f7ebSDarrick J. Wong xfs_reclaim_inodes(mp); 581d336f7ebSDarrick J. Wong xfs_health_unmount(mp); 582d336f7ebSDarrick J. Wong } 583d336f7ebSDarrick J. Wong 584b2941046SDave Chinner static void 585b2941046SDave Chinner xfs_mount_setup_inode_geom( 586b2941046SDave Chinner struct xfs_mount *mp) 587b2941046SDave Chinner { 588b2941046SDave Chinner struct xfs_ino_geometry *igeo = M_IGEO(mp); 589b2941046SDave Chinner 590b2941046SDave Chinner igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp); 591b2941046SDave Chinner ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp)); 592b2941046SDave Chinner 593b2941046SDave Chinner xfs_ialloc_setup_geometry(mp); 594b2941046SDave Chinner } 595b2941046SDave Chinner 596b74e15d7SDarrick J. Wong /* Compute maximum possible height for per-AG btree types for this fs. */ 597b74e15d7SDarrick J. Wong static inline void 598b74e15d7SDarrick J. Wong xfs_agbtree_compute_maxlevels( 599b74e15d7SDarrick J. Wong struct xfs_mount *mp) 600b74e15d7SDarrick J. Wong { 601b74e15d7SDarrick J. Wong unsigned int levels; 602b74e15d7SDarrick J. Wong 603b74e15d7SDarrick J. Wong levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels); 604b74e15d7SDarrick J. Wong levels = max(levels, mp->m_rmap_maxlevels); 605b74e15d7SDarrick J. Wong mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels); 606b74e15d7SDarrick J. Wong } 607b74e15d7SDarrick J. Wong 608d336f7ebSDarrick J. Wong /* 6090771fb45SEric Sandeen * This function does the following on an initial mount of a file system: 6100771fb45SEric Sandeen * - reads the superblock from disk and init the mount struct 6110771fb45SEric Sandeen * - if we're a 32-bit kernel, do a size check on the superblock 6120771fb45SEric Sandeen * so we don't mount terabyte filesystems 6130771fb45SEric Sandeen * - init mount struct realtime fields 6140771fb45SEric Sandeen * - allocate inode hash table for fs 6150771fb45SEric Sandeen * - init directory manager 6160771fb45SEric Sandeen * - perform recovery and init the log manager 6170771fb45SEric Sandeen */ 6180771fb45SEric Sandeen int 6190771fb45SEric Sandeen xfs_mountfs( 620f0b2efadSBrian Foster struct xfs_mount *mp) 6210771fb45SEric Sandeen { 622f0b2efadSBrian Foster struct xfs_sb *sbp = &(mp->m_sb); 623f0b2efadSBrian Foster struct xfs_inode *rip; 624ef325959SDarrick J. Wong struct xfs_ino_geometry *igeo = M_IGEO(mp); 625c8ce540dSDarrick J. Wong uint64_t resblks; 6267d095257SChristoph Hellwig uint quotamount = 0; 6277d095257SChristoph Hellwig uint quotaflags = 0; 6280771fb45SEric Sandeen int error = 0; 6290771fb45SEric Sandeen 630ff55068cSDave Chinner xfs_sb_mount_common(mp, sbp); 6310771fb45SEric Sandeen 6320771fb45SEric Sandeen /* 633074e427bSDave Chinner * Check for a mismatched features2 values. Older kernels read & wrote 634074e427bSDave Chinner * into the wrong sb offset for sb_features2 on some platforms due to 635074e427bSDave Chinner * xfs_sb_t not being 64bit size aligned when sb_features2 was added, 636074e427bSDave Chinner * which made older superblock reading/writing routines swap it as a 637074e427bSDave Chinner * 64-bit value. 638ee1c0908SDavid Chinner * 639e6957ea4SEric Sandeen * For backwards compatibility, we make both slots equal. 640e6957ea4SEric Sandeen * 641074e427bSDave Chinner * If we detect a mismatched field, we OR the set bits into the existing 642074e427bSDave Chinner * features2 field in case it has already been modified; we don't want 643074e427bSDave Chinner * to lose any features. We then update the bad location with the ORed 644074e427bSDave Chinner * value so that older kernels will see any features2 flags. The 645074e427bSDave Chinner * superblock writeback code ensures the new sb_features2 is copied to 646074e427bSDave Chinner * sb_bad_features2 before it is logged or written to disk. 647ee1c0908SDavid Chinner */ 648e6957ea4SEric Sandeen if (xfs_sb_has_mismatched_features2(sbp)) { 6490b932cccSDave Chinner xfs_warn(mp, "correcting sb_features alignment problem"); 650ee1c0908SDavid Chinner sbp->sb_features2 |= sbp->sb_bad_features2; 65161e63ecbSDave Chinner mp->m_update_sb = true; 6527c12f296STim Shimmin } 653e6957ea4SEric Sandeen 654ee1c0908SDavid Chinner 655263997a6SDave Chinner /* always use v2 inodes by default now */ 656263997a6SDave Chinner if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { 657263997a6SDave Chinner mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; 65838c26bfdSDave Chinner mp->m_features |= XFS_FEAT_NLINK; 65961e63ecbSDave Chinner mp->m_update_sb = true; 660263997a6SDave Chinner } 661263997a6SDave Chinner 662ee1c0908SDavid Chinner /* 6634f5b1b3aSDarrick J. Wong * If we were given new sunit/swidth options, do some basic validation 6644f5b1b3aSDarrick J. Wong * checks and convert the incore dalign and swidth values to the 6654f5b1b3aSDarrick J. Wong * same units (FSB) that everything else uses. This /must/ happen 6664f5b1b3aSDarrick J. Wong * before computing the inode geometry. 6670771fb45SEric Sandeen */ 6684f5b1b3aSDarrick J. Wong error = xfs_validate_new_dalign(mp); 6690771fb45SEric Sandeen if (error) 670f9057e3dSChristoph Hellwig goto out; 6710771fb45SEric Sandeen 6720771fb45SEric Sandeen xfs_alloc_compute_maxlevels(mp); 6730771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); 6740771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); 675b2941046SDave Chinner xfs_mount_setup_inode_geom(mp); 676035e00acSDarrick J. Wong xfs_rmapbt_compute_maxlevels(mp); 6771946b91cSDarrick J. Wong xfs_refcountbt_compute_maxlevels(mp); 6780771fb45SEric Sandeen 679b74e15d7SDarrick J. Wong xfs_agbtree_compute_maxlevels(mp); 680b74e15d7SDarrick J. Wong 6814f5b1b3aSDarrick J. Wong /* 6824f5b1b3aSDarrick J. Wong * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks 6834f5b1b3aSDarrick J. Wong * is NOT aligned turn off m_dalign since allocator alignment is within 6844f5b1b3aSDarrick J. Wong * an ag, therefore ag has to be aligned at stripe boundary. Note that 6854f5b1b3aSDarrick J. Wong * we must compute the free space and rmap btree geometry before doing 6864f5b1b3aSDarrick J. Wong * this. 6874f5b1b3aSDarrick J. Wong */ 6884f5b1b3aSDarrick J. Wong error = xfs_update_alignment(mp); 6894f5b1b3aSDarrick J. Wong if (error) 6904f5b1b3aSDarrick J. Wong goto out; 6914f5b1b3aSDarrick J. Wong 692e6b3bb78SCarlos Maiolino /* enable fail_at_unmount as default */ 693749f24f3SThomas Meyer mp->m_fail_unmount = true; 694e6b3bb78SCarlos Maiolino 695e1d3d218SIan Kent error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, 696e1d3d218SIan Kent NULL, mp->m_super->s_id); 69727174203SChristoph Hellwig if (error) 698f9057e3dSChristoph Hellwig goto out; 6991da177e4SLinus Torvalds 700225e4635SBill O'Donnell error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, 701225e4635SBill O'Donnell &mp->m_kobj, "stats"); 702a31b1d3dSBrian Foster if (error) 703a31b1d3dSBrian Foster goto out_remove_sysfs; 704a31b1d3dSBrian Foster 705192852beSCarlos Maiolino error = xfs_error_sysfs_init(mp); 706225e4635SBill O'Donnell if (error) 707225e4635SBill O'Donnell goto out_del_stats; 708225e4635SBill O'Donnell 70931965ef3SDarrick J. Wong error = xfs_errortag_init(mp); 71031965ef3SDarrick J. Wong if (error) 71131965ef3SDarrick J. Wong goto out_remove_error_sysfs; 712192852beSCarlos Maiolino 713192852beSCarlos Maiolino error = xfs_uuid_mount(mp); 714192852beSCarlos Maiolino if (error) 71531965ef3SDarrick J. Wong goto out_remove_errortag; 716192852beSCarlos Maiolino 7171da177e4SLinus Torvalds /* 7182fcddee8SChristoph Hellwig * Update the preferred write size based on the information from the 7192fcddee8SChristoph Hellwig * on-disk superblock. 7200771fb45SEric Sandeen */ 7212fcddee8SChristoph Hellwig mp->m_allocsize_log = 7222fcddee8SChristoph Hellwig max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log); 7232fcddee8SChristoph Hellwig mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog); 7240771fb45SEric Sandeen 725055388a3SDave Chinner /* set the low space thresholds for dynamic preallocation */ 726055388a3SDave Chinner xfs_set_low_space_thresholds(mp); 727055388a3SDave Chinner 7280771fb45SEric Sandeen /* 729e5376fc1SBrian Foster * If enabled, sparse inode chunk alignment is expected to match the 730e5376fc1SBrian Foster * cluster size. Full inode chunk alignment must match the chunk size, 731e5376fc1SBrian Foster * but that is checked on sb read verification... 732e5376fc1SBrian Foster */ 73338c26bfdSDave Chinner if (xfs_has_sparseinodes(mp) && 734e5376fc1SBrian Foster mp->m_sb.sb_spino_align != 735490d451fSDarrick J. Wong XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) { 736e5376fc1SBrian Foster xfs_warn(mp, 737e5376fc1SBrian Foster "Sparse inode block alignment (%u) must match cluster size (%llu).", 738e5376fc1SBrian Foster mp->m_sb.sb_spino_align, 739490d451fSDarrick J. Wong XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)); 740e5376fc1SBrian Foster error = -EINVAL; 741e5376fc1SBrian Foster goto out_remove_uuid; 742e5376fc1SBrian Foster } 743e5376fc1SBrian Foster 744e5376fc1SBrian Foster /* 745c2bfbc9bSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 7460771fb45SEric Sandeen */ 7474249023aSChristoph Hellwig error = xfs_check_sizes(mp); 7480771fb45SEric Sandeen if (error) 749f9057e3dSChristoph Hellwig goto out_remove_uuid; 7500771fb45SEric Sandeen 7510771fb45SEric Sandeen /* 7521da177e4SLinus Torvalds * Initialize realtime fields in the mount structure 7531da177e4SLinus Torvalds */ 7540771fb45SEric Sandeen error = xfs_rtmount_init(mp); 7550771fb45SEric Sandeen if (error) { 7560b932cccSDave Chinner xfs_warn(mp, "RT mount failed"); 757f9057e3dSChristoph Hellwig goto out_remove_uuid; 7581da177e4SLinus Torvalds } 7591da177e4SLinus Torvalds 7601da177e4SLinus Torvalds /* 7611da177e4SLinus Torvalds * Copies the low order bits of the timestamp and the randomly 7621da177e4SLinus Torvalds * set "sequence" number out of a UUID. 7631da177e4SLinus Torvalds */ 764cb0ba6ccSChristoph Hellwig mp->m_fixedfsid[0] = 765cb0ba6ccSChristoph Hellwig (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) | 766cb0ba6ccSChristoph Hellwig get_unaligned_be16(&sbp->sb_uuid.b[4]); 767cb0ba6ccSChristoph Hellwig mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]); 7681da177e4SLinus Torvalds 7690650b554SDave Chinner error = xfs_da_mount(mp); 7700650b554SDave Chinner if (error) { 7710650b554SDave Chinner xfs_warn(mp, "Failed dir/attr init: %d", error); 7720650b554SDave Chinner goto out_remove_uuid; 7730650b554SDave Chinner } 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 7761da177e4SLinus Torvalds * Initialize the precomputed transaction reservations values. 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds xfs_trans_init(mp); 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds /* 7811da177e4SLinus Torvalds * Allocate and initialize the per-ag data. 7821da177e4SLinus Torvalds */ 7830800169eSDave Chinner error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks, 7840800169eSDave Chinner &mp->m_maxagi); 7851c1c6ebcSDave Chinner if (error) { 7860b932cccSDave Chinner xfs_warn(mp, "Failed per-ag init: %d", error); 7870650b554SDave Chinner goto out_free_dir; 7881c1c6ebcSDave Chinner } 7891da177e4SLinus Torvalds 790a71895c5SDarrick J. Wong if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) { 7910b932cccSDave Chinner xfs_warn(mp, "no log defined"); 7922451337dSDave Chinner error = -EFSCORRUPTED; 793f9057e3dSChristoph Hellwig goto out_free_perag; 794f9057e3dSChristoph Hellwig } 795f9057e3dSChristoph Hellwig 79640b1de00SDarrick J. Wong error = xfs_inodegc_register_shrinker(mp); 79740b1de00SDarrick J. Wong if (error) 79840b1de00SDarrick J. Wong goto out_fail_wait; 79940b1de00SDarrick J. Wong 8001da177e4SLinus Torvalds /* 801f0b2efadSBrian Foster * Log's mount-time initialization. The first part of recovery can place 802f0b2efadSBrian Foster * some items on the AIL, to be handled when recovery is finished or 803f0b2efadSBrian Foster * cancelled. 8041da177e4SLinus Torvalds */ 8051da177e4SLinus Torvalds error = xfs_log_mount(mp, mp->m_logdev_targp, 8061da177e4SLinus Torvalds XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), 8071da177e4SLinus Torvalds XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); 8081da177e4SLinus Torvalds if (error) { 8090b932cccSDave Chinner xfs_warn(mp, "log mount failed"); 81040b1de00SDarrick J. Wong goto out_inodegc_shrinker; 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds 813ab23a776SDave Chinner /* Enable background inode inactivation workers. */ 814ab23a776SDave Chinner xfs_inodegc_start(mp); 8156f649091SDarrick J. Wong xfs_blockgc_start(mp); 816ab23a776SDave Chinner 81792821e2bSDavid Chinner /* 818e23b55d5SDave Chinner * Now that we've recovered any pending superblock feature bit 819e23b55d5SDave Chinner * additions, we can finish setting up the attr2 behaviour for the 8200560f31aSDave Chinner * mount. The noattr2 option overrides the superblock flag, so only 8210560f31aSDave Chinner * check the superblock feature flag if the mount option is not set. 822e23b55d5SDave Chinner */ 8230560f31aSDave Chinner if (xfs_has_noattr2(mp)) { 8240560f31aSDave Chinner mp->m_features &= ~XFS_FEAT_ATTR2; 8250560f31aSDave Chinner } else if (!xfs_has_attr2(mp) && 8260560f31aSDave Chinner (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) { 8270560f31aSDave Chinner mp->m_features |= XFS_FEAT_ATTR2; 8280560f31aSDave Chinner } 829e23b55d5SDave Chinner 830e23b55d5SDave Chinner /* 8311da177e4SLinus Torvalds * Get and sanity-check the root inode. 8321da177e4SLinus Torvalds * Save the pointer to it in the mount structure. 8331da177e4SLinus Torvalds */ 834541b5accSDave Chinner error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED, 835541b5accSDave Chinner XFS_ILOCK_EXCL, &rip); 8361da177e4SLinus Torvalds if (error) { 837541b5accSDave Chinner xfs_warn(mp, 838541b5accSDave Chinner "Failed to read root inode 0x%llx, error %d", 839541b5accSDave Chinner sbp->sb_rootino, -error); 840f9057e3dSChristoph Hellwig goto out_log_dealloc; 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds 8431da177e4SLinus Torvalds ASSERT(rip != NULL); 8441da177e4SLinus Torvalds 845a71895c5SDarrick J. Wong if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) { 8460b932cccSDave Chinner xfs_warn(mp, "corrupted root inode %llu: not a directory", 847b6574520SNathan Scott (unsigned long long)rip->i_ino); 8481da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8492451337dSDave Chinner error = -EFSCORRUPTED; 850f9057e3dSChristoph Hellwig goto out_rele_rip; 8511da177e4SLinus Torvalds } 8521da177e4SLinus Torvalds mp->m_rootip = rip; /* save it */ 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds /* 8571da177e4SLinus Torvalds * Initialize realtime inode pointers in the mount structure 8581da177e4SLinus Torvalds */ 8590771fb45SEric Sandeen error = xfs_rtmount_inodes(mp); 8600771fb45SEric Sandeen if (error) { 8611da177e4SLinus Torvalds /* 8621da177e4SLinus Torvalds * Free up the root inode. 8631da177e4SLinus Torvalds */ 8640b932cccSDave Chinner xfs_warn(mp, "failed to read RT inodes"); 865f9057e3dSChristoph Hellwig goto out_rele_rip; 8661da177e4SLinus Torvalds } 8671da177e4SLinus Torvalds 8685a605fd6SDarrick J. Wong /* Make sure the summary counts are ok. */ 8695a605fd6SDarrick J. Wong error = xfs_check_summary_counts(mp); 8705a605fd6SDarrick J. Wong if (error) 8715a605fd6SDarrick J. Wong goto out_rtunmount; 8725a605fd6SDarrick J. Wong 8731da177e4SLinus Torvalds /* 8747884bc86SChristoph Hellwig * If this is a read-only mount defer the superblock updates until 8757884bc86SChristoph Hellwig * the next remount into writeable mode. Otherwise we would never 8767884bc86SChristoph Hellwig * perform the update e.g. for the root filesystem. 8771da177e4SLinus Torvalds */ 8782e973b2cSDave Chinner if (mp->m_update_sb && !xfs_is_readonly(mp)) { 87961e63ecbSDave Chinner error = xfs_sync_sb(mp, false); 880e5720eecSDavid Chinner if (error) { 8810b932cccSDave Chinner xfs_warn(mp, "failed to write sb changes"); 882b93b6e43SChristoph Hellwig goto out_rtunmount; 883e5720eecSDavid Chinner } 884e5720eecSDavid Chinner } 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * Initialise the XFS quota management subsystem for this mount 8881da177e4SLinus Torvalds */ 889149e53afSChristoph Hellwig if (XFS_IS_QUOTA_ON(mp)) { 8907d095257SChristoph Hellwig error = xfs_qm_newmount(mp, "amount, "aflags); 8910771fb45SEric Sandeen if (error) 892b93b6e43SChristoph Hellwig goto out_rtunmount; 8937d095257SChristoph Hellwig } else { 8947d095257SChristoph Hellwig /* 8957d095257SChristoph Hellwig * If a file system had quotas running earlier, but decided to 8967d095257SChristoph Hellwig * mount without -o uquota/pquota/gquota options, revoke the 8977d095257SChristoph Hellwig * quotachecked license. 8987d095257SChristoph Hellwig */ 8997d095257SChristoph Hellwig if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { 9000b932cccSDave Chinner xfs_notice(mp, "resetting quota flags"); 9017d095257SChristoph Hellwig error = xfs_mount_reset_sbqflags(mp); 9027d095257SChristoph Hellwig if (error) 903a70a4fa5SBrian Foster goto out_rtunmount; 9047d095257SChristoph Hellwig } 9057d095257SChristoph Hellwig } 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds /* 908f0b2efadSBrian Foster * Finish recovering the file system. This part needed to be delayed 909f0b2efadSBrian Foster * until after the root and real-time bitmap inodes were consistently 91081ed9475SDarrick J. Wong * read in. Temporarily create per-AG space reservations for metadata 91181ed9475SDarrick J. Wong * btree shape changes because space freeing transactions (for inode 91281ed9475SDarrick J. Wong * inactivation) require the per-AG reservation in lieu of reserving 91381ed9475SDarrick J. Wong * blocks. 9141da177e4SLinus Torvalds */ 91581ed9475SDarrick J. Wong error = xfs_fs_reserve_ag_blocks(mp); 91681ed9475SDarrick J. Wong if (error && error == -ENOSPC) 91781ed9475SDarrick J. Wong xfs_warn(mp, 91881ed9475SDarrick J. Wong "ENOSPC reserving per-AG metadata pool, log recovery may fail."); 9194249023aSChristoph Hellwig error = xfs_log_mount_finish(mp); 92081ed9475SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 9211da177e4SLinus Torvalds if (error) { 9220b932cccSDave Chinner xfs_warn(mp, "log mount finish failed"); 923b93b6e43SChristoph Hellwig goto out_rtunmount; 9241da177e4SLinus Torvalds } 9251da177e4SLinus Torvalds 9261da177e4SLinus Torvalds /* 927ddeb14f4SDave Chinner * Now the log is fully replayed, we can transition to full read-only 928ddeb14f4SDave Chinner * mode for read-only mounts. This will sync all the metadata and clean 929ddeb14f4SDave Chinner * the log so that the recovery we just performed does not have to be 930ddeb14f4SDave Chinner * replayed again on the next mount. 931ddeb14f4SDave Chinner * 932ddeb14f4SDave Chinner * We use the same quiesce mechanism as the rw->ro remount, as they are 933ddeb14f4SDave Chinner * semantically identical operations. 934ddeb14f4SDave Chinner */ 9352e973b2cSDave Chinner if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp)) 936ea2064daSBrian Foster xfs_log_clean(mp); 937ddeb14f4SDave Chinner 938ddeb14f4SDave Chinner /* 9391da177e4SLinus Torvalds * Complete the quota initialisation, post-log-replay component. 9401da177e4SLinus Torvalds */ 9417d095257SChristoph Hellwig if (quotamount) { 9427d095257SChristoph Hellwig ASSERT(mp->m_qflags == 0); 9437d095257SChristoph Hellwig mp->m_qflags = quotaflags; 9447d095257SChristoph Hellwig 9457d095257SChristoph Hellwig xfs_qm_mount_quotas(mp); 9467d095257SChristoph Hellwig } 9477d095257SChristoph Hellwig 94884e1e99fSDavid Chinner /* 94984e1e99fSDavid Chinner * Now we are mounted, reserve a small amount of unused space for 95084e1e99fSDavid Chinner * privileged transactions. This is needed so that transaction 95184e1e99fSDavid Chinner * space required for critical operations can dip into this pool 95284e1e99fSDavid Chinner * when at ENOSPC. This is needed for operations like create with 95384e1e99fSDavid Chinner * attr, unwritten extent conversion at ENOSPC, etc. Data allocations 95484e1e99fSDavid Chinner * are not allowed to use this reserved space. 9558babd8a2SDave Chinner * 9568babd8a2SDave Chinner * This may drive us straight to ENOSPC on mount, but that implies 9578babd8a2SDave Chinner * we were already there on the last unmount. Warn if this occurs. 95884e1e99fSDavid Chinner */ 9592e973b2cSDave Chinner if (!xfs_is_readonly(mp)) { 960d5db0f97SEric Sandeen resblks = xfs_default_resblks(mp); 961714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 962714082bcSDavid Chinner if (error) 9630b932cccSDave Chinner xfs_warn(mp, 9640b932cccSDave Chinner "Unable to allocate reserve blocks. Continuing without reserve pool."); 965174edb0eSDarrick J. Wong 96684d69619SDarrick J. Wong /* Reserve AG blocks for future btree expansion. */ 96784d69619SDarrick J. Wong error = xfs_fs_reserve_ag_blocks(mp); 96884d69619SDarrick J. Wong if (error && error != -ENOSPC) 96984d69619SDarrick J. Wong goto out_agresv; 970d5db0f97SEric Sandeen } 97184e1e99fSDavid Chinner 9721da177e4SLinus Torvalds return 0; 9731da177e4SLinus Torvalds 97484d69619SDarrick J. Wong out_agresv: 97584d69619SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 976174edb0eSDarrick J. Wong xfs_qm_unmount_quotas(mp); 977b93b6e43SChristoph Hellwig out_rtunmount: 978b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 979f9057e3dSChristoph Hellwig out_rele_rip: 98044a8736bSDarrick J. Wong xfs_irele(rip); 98177aff8c7SDarrick J. Wong /* Clean out dquots that might be in memory after quotacheck. */ 98277aff8c7SDarrick J. Wong xfs_qm_unmount(mp); 983ab23a776SDave Chinner 984ab23a776SDave Chinner /* 985ab23a776SDave Chinner * Inactivate all inodes that might still be in memory after a log 986ab23a776SDave Chinner * intent recovery failure so that reclaim can free them. Metadata 987ab23a776SDave Chinner * inodes and the root directory shouldn't need inactivation, but the 988ab23a776SDave Chinner * mount failed for some reason, so pull down all the state and flee. 989ab23a776SDave Chinner */ 990ab23a776SDave Chinner xfs_inodegc_flush(mp); 991ab23a776SDave Chinner 9922d1d1da3SDarrick J. Wong /* 993d336f7ebSDarrick J. Wong * Flush all inode reclamation work and flush the log. 9942d1d1da3SDarrick J. Wong * We have to do this /after/ rtunmount and qm_unmount because those 9952d1d1da3SDarrick J. Wong * two will have scheduled delayed reclaim for the rt/quota inodes. 9962d1d1da3SDarrick J. Wong * 9972d1d1da3SDarrick J. Wong * This is slightly different from the unmountfs call sequence 9982d1d1da3SDarrick J. Wong * because we could be tearing down a partially set up mount. In 9992d1d1da3SDarrick J. Wong * particular, if log_mount_finish fails we bail out without calling 10002d1d1da3SDarrick J. Wong * qm_unmount_quotas and therefore rely on qm_unmount to release the 10012d1d1da3SDarrick J. Wong * quota inodes. 10022d1d1da3SDarrick J. Wong */ 1003d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes(mp); 1004f9057e3dSChristoph Hellwig out_log_dealloc: 1005f0b2efadSBrian Foster xfs_log_mount_cancel(mp); 100640b1de00SDarrick J. Wong out_inodegc_shrinker: 100740b1de00SDarrick J. Wong unregister_shrinker(&mp->m_inodegc_shrinker); 1008d4f3512bSDave Chinner out_fail_wait: 1009d4f3512bSDave Chinner if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) 101010fb9ac1SBrian Foster xfs_buftarg_drain(mp->m_logdev_targp); 101110fb9ac1SBrian Foster xfs_buftarg_drain(mp->m_ddev_targp); 1012f9057e3dSChristoph Hellwig out_free_perag: 1013ff4f038cSChristoph Hellwig xfs_free_perag(mp); 10140650b554SDave Chinner out_free_dir: 10150650b554SDave Chinner xfs_da_unmount(mp); 1016f9057e3dSChristoph Hellwig out_remove_uuid: 101727174203SChristoph Hellwig xfs_uuid_unmount(mp); 101831965ef3SDarrick J. Wong out_remove_errortag: 101931965ef3SDarrick J. Wong xfs_errortag_del(mp); 1020192852beSCarlos Maiolino out_remove_error_sysfs: 1021192852beSCarlos Maiolino xfs_error_sysfs_del(mp); 1022225e4635SBill O'Donnell out_del_stats: 1023225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 1024a31b1d3dSBrian Foster out_remove_sysfs: 1025a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 1026f9057e3dSChristoph Hellwig out: 10271da177e4SLinus Torvalds return error; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10301da177e4SLinus Torvalds /* 10311da177e4SLinus Torvalds * This flushes out the inodes,dquots and the superblock, unmounts the 10321da177e4SLinus Torvalds * log and makes sure that incore structures are freed. 10331da177e4SLinus Torvalds */ 103441b5c2e7SChristoph Hellwig void 103541b5c2e7SChristoph Hellwig xfs_unmountfs( 103641b5c2e7SChristoph Hellwig struct xfs_mount *mp) 10371da177e4SLinus Torvalds { 1038c8ce540dSDarrick J. Wong uint64_t resblks; 103941b5c2e7SChristoph Hellwig int error; 10401da177e4SLinus Torvalds 1041ab23a776SDave Chinner /* 1042ab23a776SDave Chinner * Perform all on-disk metadata updates required to inactivate inodes 1043ab23a776SDave Chinner * that the VFS evicted earlier in the unmount process. Freeing inodes 1044ab23a776SDave Chinner * and discarding CoW fork preallocations can cause shape changes to 1045ab23a776SDave Chinner * the free inode and refcount btrees, respectively, so we must finish 1046ab23a776SDave Chinner * this before we discard the metadata space reservations. Metadata 1047ab23a776SDave Chinner * inodes and the root directory do not require inactivation. 1048ab23a776SDave Chinner */ 1049ab23a776SDave Chinner xfs_inodegc_flush(mp); 1050ab23a776SDave Chinner 1051c9a6526fSDarrick J. Wong xfs_blockgc_stop(mp); 105284d69619SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 10537d095257SChristoph Hellwig xfs_qm_unmount_quotas(mp); 1054b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 105544a8736bSDarrick J. Wong xfs_irele(mp->m_rootip); 105677508ec8SChristoph Hellwig 1057d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes(mp); 10581da177e4SLinus Torvalds 10597d095257SChristoph Hellwig xfs_qm_unmount(mp); 1060a357a121SLachlan McIlroy 10611da177e4SLinus Torvalds /* 106284e1e99fSDavid Chinner * Unreserve any blocks we have so that when we unmount we don't account 106384e1e99fSDavid Chinner * the reserved free space as used. This is really only necessary for 106484e1e99fSDavid Chinner * lazy superblock counting because it trusts the incore superblock 10659da096fdSMalcolm Parsons * counters to be absolutely correct on clean unmount. 106684e1e99fSDavid Chinner * 106784e1e99fSDavid Chinner * We don't bother correcting this elsewhere for lazy superblock 106884e1e99fSDavid Chinner * counting because on mount of an unclean filesystem we reconstruct the 106984e1e99fSDavid Chinner * correct counter value and this is irrelevant. 107084e1e99fSDavid Chinner * 107184e1e99fSDavid Chinner * For non-lazy counter filesystems, this doesn't matter at all because 107284e1e99fSDavid Chinner * we only every apply deltas to the superblock and hence the incore 107384e1e99fSDavid Chinner * value does not matter.... 107484e1e99fSDavid Chinner */ 107584e1e99fSDavid Chinner resblks = 0; 1076714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 1077714082bcSDavid Chinner if (error) 10780b932cccSDave Chinner xfs_warn(mp, "Unable to free reserved block pool. " 1079714082bcSDavid Chinner "Freespace may not be correct on next mount."); 1080714082bcSDavid Chinner 108121b699c8SChristoph Hellwig xfs_log_unmount(mp); 10820650b554SDave Chinner xfs_da_unmount(mp); 108327174203SChristoph Hellwig xfs_uuid_unmount(mp); 10841da177e4SLinus Torvalds 10851550d0b0SChristoph Hellwig #if defined(DEBUG) 108631965ef3SDarrick J. Wong xfs_errortag_clearall(mp); 10871da177e4SLinus Torvalds #endif 108840b1de00SDarrick J. Wong unregister_shrinker(&mp->m_inodegc_shrinker); 1089ff4f038cSChristoph Hellwig xfs_free_perag(mp); 1090a31b1d3dSBrian Foster 109131965ef3SDarrick J. Wong xfs_errortag_del(mp); 1092192852beSCarlos Maiolino xfs_error_sysfs_del(mp); 1093225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 1094a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 10951da177e4SLinus Torvalds } 10961da177e4SLinus Torvalds 109791ee575fSBrian Foster /* 109891ee575fSBrian Foster * Determine whether modifications can proceed. The caller specifies the minimum 109991ee575fSBrian Foster * freeze level for which modifications should not be allowed. This allows 110091ee575fSBrian Foster * certain operations to proceed while the freeze sequence is in progress, if 110191ee575fSBrian Foster * necessary. 110291ee575fSBrian Foster */ 110391ee575fSBrian Foster bool 110491ee575fSBrian Foster xfs_fs_writable( 110591ee575fSBrian Foster struct xfs_mount *mp, 110691ee575fSBrian Foster int level) 110792821e2bSDavid Chinner { 110891ee575fSBrian Foster ASSERT(level > SB_UNFROZEN); 110991ee575fSBrian Foster if ((mp->m_super->s_writers.frozen >= level) || 111075c8c50fSDave Chinner xfs_is_shutdown(mp) || xfs_is_readonly(mp)) 111191ee575fSBrian Foster return false; 111291ee575fSBrian Foster 111391ee575fSBrian Foster return true; 111492821e2bSDavid Chinner } 111592821e2bSDavid Chinner 11162229276cSDarrick J. Wong /* Adjust m_fdblocks or m_frextents. */ 11170d485adaSDave Chinner int 11182229276cSDarrick J. Wong xfs_mod_freecounter( 11190d485adaSDave Chinner struct xfs_mount *mp, 11202229276cSDarrick J. Wong struct percpu_counter *counter, 11210d485adaSDave Chinner int64_t delta, 11220d485adaSDave Chinner bool rsvd) 11230d485adaSDave Chinner { 11240d485adaSDave Chinner int64_t lcounter; 11250d485adaSDave Chinner long long res_used; 11262229276cSDarrick J. Wong uint64_t set_aside = 0; 11270d485adaSDave Chinner s32 batch; 11282229276cSDarrick J. Wong bool has_resv_pool; 11292229276cSDarrick J. Wong 11302229276cSDarrick J. Wong ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents); 11312229276cSDarrick J. Wong has_resv_pool = (counter == &mp->m_fdblocks); 11322229276cSDarrick J. Wong if (rsvd) 11332229276cSDarrick J. Wong ASSERT(has_resv_pool); 11340d485adaSDave Chinner 11350d485adaSDave Chinner if (delta > 0) { 11360d485adaSDave Chinner /* 11370d485adaSDave Chinner * If the reserve pool is depleted, put blocks back into it 11380d485adaSDave Chinner * first. Most of the time the pool is full. 11390d485adaSDave Chinner */ 11402229276cSDarrick J. Wong if (likely(!has_resv_pool || 11412229276cSDarrick J. Wong mp->m_resblks == mp->m_resblks_avail)) { 11422229276cSDarrick J. Wong percpu_counter_add(counter, delta); 11430d485adaSDave Chinner return 0; 11440d485adaSDave Chinner } 11450d485adaSDave Chinner 11460d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 11470d485adaSDave Chinner res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); 11480d485adaSDave Chinner 11490d485adaSDave Chinner if (res_used > delta) { 11500d485adaSDave Chinner mp->m_resblks_avail += delta; 11510d485adaSDave Chinner } else { 11520d485adaSDave Chinner delta -= res_used; 11530d485adaSDave Chinner mp->m_resblks_avail = mp->m_resblks; 11542229276cSDarrick J. Wong percpu_counter_add(counter, delta); 11550d485adaSDave Chinner } 11560d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 11570d485adaSDave Chinner return 0; 11580d485adaSDave Chinner } 11590d485adaSDave Chinner 11600d485adaSDave Chinner /* 11610d485adaSDave Chinner * Taking blocks away, need to be more accurate the closer we 11620d485adaSDave Chinner * are to zero. 11630d485adaSDave Chinner * 11640d485adaSDave Chinner * If the counter has a value of less than 2 * max batch size, 11650d485adaSDave Chinner * then make everything serialise as we are real close to 11660d485adaSDave Chinner * ENOSPC. 11670d485adaSDave Chinner */ 11682229276cSDarrick J. Wong if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH, 11698c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) < 0) 11700d485adaSDave Chinner batch = 1; 11710d485adaSDave Chinner else 11728c1903d3SDave Chinner batch = XFS_FDBLOCKS_BATCH; 11730d485adaSDave Chinner 1174fd43cf60SBrian Foster /* 1175fd43cf60SBrian Foster * Set aside allocbt blocks because these blocks are tracked as free 1176fd43cf60SBrian Foster * space but not available for allocation. Technically this means that a 1177fd43cf60SBrian Foster * single reservation cannot consume all remaining free space, but the 1178fd43cf60SBrian Foster * ratio of allocbt blocks to usable free blocks should be rather small. 1179fd43cf60SBrian Foster * The tradeoff without this is that filesystems that maintain high 1180fd43cf60SBrian Foster * perag block reservations can over reserve physical block availability 1181fd43cf60SBrian Foster * and fail physical allocation, which leads to much more serious 1182fd43cf60SBrian Foster * problems (i.e. transaction abort, pagecache discards, etc.) than 1183fd43cf60SBrian Foster * slightly premature -ENOSPC. 1184fd43cf60SBrian Foster */ 11852229276cSDarrick J. Wong if (has_resv_pool) 1186c8c56825SDarrick J. Wong set_aside = xfs_fdblocks_unavailable(mp); 11872229276cSDarrick J. Wong percpu_counter_add_batch(counter, delta, batch); 11882229276cSDarrick J. Wong if (__percpu_counter_compare(counter, set_aside, 11898c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) >= 0) { 11900d485adaSDave Chinner /* we had space! */ 11910d485adaSDave Chinner return 0; 11920d485adaSDave Chinner } 11930d485adaSDave Chinner 11940d485adaSDave Chinner /* 11950d485adaSDave Chinner * lock up the sb for dipping into reserves before releasing the space 11960d485adaSDave Chinner * that took us to ENOSPC. 11970d485adaSDave Chinner */ 11980d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 11992229276cSDarrick J. Wong percpu_counter_add(counter, -delta); 12002229276cSDarrick J. Wong if (!has_resv_pool || !rsvd) 12010d485adaSDave Chinner goto fdblocks_enospc; 12020d485adaSDave Chinner 12030d485adaSDave Chinner lcounter = (long long)mp->m_resblks_avail + delta; 12040d485adaSDave Chinner if (lcounter >= 0) { 12050d485adaSDave Chinner mp->m_resblks_avail = lcounter; 12060d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 12070d485adaSDave Chinner return 0; 12080d485adaSDave Chinner } 1209ec43f6daSEric Sandeen xfs_warn_once(mp, 1210ec43f6daSEric Sandeen "Reserve blocks depleted! Consider increasing reserve pool size."); 1211ec43f6daSEric Sandeen 12120d485adaSDave Chinner fdblocks_enospc: 12130d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 12140d485adaSDave Chinner return -ENOSPC; 12150d485adaSDave Chinner } 12160d485adaSDave Chinner 12171da177e4SLinus Torvalds /* 12181da177e4SLinus Torvalds * Used to free the superblock along various error paths. 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds void 12211da177e4SLinus Torvalds xfs_freesb( 122226af6552SDave Chinner struct xfs_mount *mp) 12231da177e4SLinus Torvalds { 122426af6552SDave Chinner struct xfs_buf *bp = mp->m_sb_bp; 12251da177e4SLinus Torvalds 122626af6552SDave Chinner xfs_buf_lock(bp); 12271da177e4SLinus Torvalds mp->m_sb_bp = NULL; 122826af6552SDave Chinner xfs_buf_relse(bp); 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds /* 1232dda35b8fSChristoph Hellwig * If the underlying (data/log/rt) device is readonly, there are some 1233dda35b8fSChristoph Hellwig * operations that cannot proceed. 1234dda35b8fSChristoph Hellwig */ 1235dda35b8fSChristoph Hellwig int 1236dda35b8fSChristoph Hellwig xfs_dev_is_read_only( 1237dda35b8fSChristoph Hellwig struct xfs_mount *mp, 1238dda35b8fSChristoph Hellwig char *message) 1239dda35b8fSChristoph Hellwig { 1240dda35b8fSChristoph Hellwig if (xfs_readonly_buftarg(mp->m_ddev_targp) || 1241dda35b8fSChristoph Hellwig xfs_readonly_buftarg(mp->m_logdev_targp) || 1242dda35b8fSChristoph Hellwig (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { 12430b932cccSDave Chinner xfs_notice(mp, "%s required on read-only device.", message); 12440b932cccSDave Chinner xfs_notice(mp, "write access unavailable, cannot proceed."); 12452451337dSDave Chinner return -EROFS; 1246dda35b8fSChristoph Hellwig } 1247dda35b8fSChristoph Hellwig return 0; 1248dda35b8fSChristoph Hellwig } 1249f467cad9SDarrick J. Wong 1250f467cad9SDarrick J. Wong /* Force the summary counters to be recalculated at next mount. */ 1251f467cad9SDarrick J. Wong void 1252f467cad9SDarrick J. Wong xfs_force_summary_recalc( 1253f467cad9SDarrick J. Wong struct xfs_mount *mp) 1254f467cad9SDarrick J. Wong { 125538c26bfdSDave Chinner if (!xfs_has_lazysbcount(mp)) 1256f467cad9SDarrick J. Wong return; 1257f467cad9SDarrick J. Wong 125839353ff6SDarrick J. Wong xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); 1259f467cad9SDarrick J. Wong } 12609fe82b8cSDarrick J. Wong 12619fe82b8cSDarrick J. Wong /* 1262908ce71eSDarrick J. Wong * Enable a log incompat feature flag in the primary superblock. The caller 1263908ce71eSDarrick J. Wong * cannot have any other transactions in progress. 1264908ce71eSDarrick J. Wong */ 1265908ce71eSDarrick J. Wong int 1266908ce71eSDarrick J. Wong xfs_add_incompat_log_feature( 1267908ce71eSDarrick J. Wong struct xfs_mount *mp, 1268908ce71eSDarrick J. Wong uint32_t feature) 1269908ce71eSDarrick J. Wong { 1270908ce71eSDarrick J. Wong struct xfs_dsb *dsb; 1271908ce71eSDarrick J. Wong int error; 1272908ce71eSDarrick J. Wong 1273908ce71eSDarrick J. Wong ASSERT(hweight32(feature) == 1); 1274908ce71eSDarrick J. Wong ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); 1275908ce71eSDarrick J. Wong 1276908ce71eSDarrick J. Wong /* 1277908ce71eSDarrick J. Wong * Force the log to disk and kick the background AIL thread to reduce 1278908ce71eSDarrick J. Wong * the chances that the bwrite will stall waiting for the AIL to unpin 1279908ce71eSDarrick J. Wong * the primary superblock buffer. This isn't a data integrity 1280908ce71eSDarrick J. Wong * operation, so we don't need a synchronous push. 1281908ce71eSDarrick J. Wong */ 1282908ce71eSDarrick J. Wong error = xfs_log_force(mp, XFS_LOG_SYNC); 1283908ce71eSDarrick J. Wong if (error) 1284908ce71eSDarrick J. Wong return error; 1285908ce71eSDarrick J. Wong xfs_ail_push_all(mp->m_ail); 1286908ce71eSDarrick J. Wong 1287908ce71eSDarrick J. Wong /* 1288908ce71eSDarrick J. Wong * Lock the primary superblock buffer to serialize all callers that 1289908ce71eSDarrick J. Wong * are trying to set feature bits. 1290908ce71eSDarrick J. Wong */ 1291908ce71eSDarrick J. Wong xfs_buf_lock(mp->m_sb_bp); 1292908ce71eSDarrick J. Wong xfs_buf_hold(mp->m_sb_bp); 1293908ce71eSDarrick J. Wong 129475c8c50fSDave Chinner if (xfs_is_shutdown(mp)) { 1295908ce71eSDarrick J. Wong error = -EIO; 1296908ce71eSDarrick J. Wong goto rele; 1297908ce71eSDarrick J. Wong } 1298908ce71eSDarrick J. Wong 1299908ce71eSDarrick J. Wong if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature)) 1300908ce71eSDarrick J. Wong goto rele; 1301908ce71eSDarrick J. Wong 1302908ce71eSDarrick J. Wong /* 1303908ce71eSDarrick J. Wong * Write the primary superblock to disk immediately, because we need 1304908ce71eSDarrick J. Wong * the log_incompat bit to be set in the primary super now to protect 1305908ce71eSDarrick J. Wong * the log items that we're going to commit later. 1306908ce71eSDarrick J. Wong */ 1307908ce71eSDarrick J. Wong dsb = mp->m_sb_bp->b_addr; 1308908ce71eSDarrick J. Wong xfs_sb_to_disk(dsb, &mp->m_sb); 1309908ce71eSDarrick J. Wong dsb->sb_features_log_incompat |= cpu_to_be32(feature); 1310908ce71eSDarrick J. Wong error = xfs_bwrite(mp->m_sb_bp); 1311908ce71eSDarrick J. Wong if (error) 1312908ce71eSDarrick J. Wong goto shutdown; 1313908ce71eSDarrick J. Wong 1314908ce71eSDarrick J. Wong /* 1315908ce71eSDarrick J. Wong * Add the feature bits to the incore superblock before we unlock the 1316908ce71eSDarrick J. Wong * buffer. 1317908ce71eSDarrick J. Wong */ 1318908ce71eSDarrick J. Wong xfs_sb_add_incompat_log_features(&mp->m_sb, feature); 1319908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1320908ce71eSDarrick J. Wong 1321908ce71eSDarrick J. Wong /* Log the superblock to disk. */ 1322908ce71eSDarrick J. Wong return xfs_sync_sb(mp, false); 1323908ce71eSDarrick J. Wong shutdown: 1324908ce71eSDarrick J. Wong xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1325908ce71eSDarrick J. Wong rele: 1326908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1327908ce71eSDarrick J. Wong return error; 1328908ce71eSDarrick J. Wong } 1329908ce71eSDarrick J. Wong 1330908ce71eSDarrick J. Wong /* 1331908ce71eSDarrick J. Wong * Clear all the log incompat flags from the superblock. 1332908ce71eSDarrick J. Wong * 1333908ce71eSDarrick J. Wong * The caller cannot be in a transaction, must ensure that the log does not 1334908ce71eSDarrick J. Wong * contain any log items protected by any log incompat bit, and must ensure 1335908ce71eSDarrick J. Wong * that there are no other threads that depend on the state of the log incompat 1336908ce71eSDarrick J. Wong * feature flags in the primary super. 1337908ce71eSDarrick J. Wong * 1338908ce71eSDarrick J. Wong * Returns true if the superblock is dirty. 1339908ce71eSDarrick J. Wong */ 1340908ce71eSDarrick J. Wong bool 1341908ce71eSDarrick J. Wong xfs_clear_incompat_log_features( 1342908ce71eSDarrick J. Wong struct xfs_mount *mp) 1343908ce71eSDarrick J. Wong { 1344908ce71eSDarrick J. Wong bool ret = false; 1345908ce71eSDarrick J. Wong 1346ebd9027dSDave Chinner if (!xfs_has_crc(mp) || 1347908ce71eSDarrick J. Wong !xfs_sb_has_incompat_log_feature(&mp->m_sb, 1348908ce71eSDarrick J. Wong XFS_SB_FEAT_INCOMPAT_LOG_ALL) || 134975c8c50fSDave Chinner xfs_is_shutdown(mp)) 1350908ce71eSDarrick J. Wong return false; 1351908ce71eSDarrick J. Wong 1352908ce71eSDarrick J. Wong /* 1353908ce71eSDarrick J. Wong * Update the incore superblock. We synchronize on the primary super 1354908ce71eSDarrick J. Wong * buffer lock to be consistent with the add function, though at least 1355908ce71eSDarrick J. Wong * in theory this shouldn't be necessary. 1356908ce71eSDarrick J. Wong */ 1357908ce71eSDarrick J. Wong xfs_buf_lock(mp->m_sb_bp); 1358908ce71eSDarrick J. Wong xfs_buf_hold(mp->m_sb_bp); 1359908ce71eSDarrick J. Wong 1360908ce71eSDarrick J. Wong if (xfs_sb_has_incompat_log_feature(&mp->m_sb, 1361908ce71eSDarrick J. Wong XFS_SB_FEAT_INCOMPAT_LOG_ALL)) { 1362908ce71eSDarrick J. Wong xfs_sb_remove_incompat_log_features(&mp->m_sb); 1363908ce71eSDarrick J. Wong ret = true; 1364908ce71eSDarrick J. Wong } 1365908ce71eSDarrick J. Wong 1366908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1367908ce71eSDarrick J. Wong return ret; 1368908ce71eSDarrick J. Wong } 1369908ce71eSDarrick J. Wong 1370908ce71eSDarrick J. Wong /* 13719fe82b8cSDarrick J. Wong * Update the in-core delayed block counter. 13729fe82b8cSDarrick J. Wong * 13739fe82b8cSDarrick J. Wong * We prefer to update the counter without having to take a spinlock for every 13749fe82b8cSDarrick J. Wong * counter update (i.e. batching). Each change to delayed allocation 13759fe82b8cSDarrick J. Wong * reservations can change can easily exceed the default percpu counter 13769fe82b8cSDarrick J. Wong * batching, so we use a larger batch factor here. 13779fe82b8cSDarrick J. Wong * 13789fe82b8cSDarrick J. Wong * Note that we don't currently have any callers requiring fast summation 13799fe82b8cSDarrick J. Wong * (e.g. percpu_counter_read) so we can use a big batch value here. 13809fe82b8cSDarrick J. Wong */ 13819fe82b8cSDarrick J. Wong #define XFS_DELALLOC_BATCH (4096) 13829fe82b8cSDarrick J. Wong void 13839fe82b8cSDarrick J. Wong xfs_mod_delalloc( 13849fe82b8cSDarrick J. Wong struct xfs_mount *mp, 13859fe82b8cSDarrick J. Wong int64_t delta) 13869fe82b8cSDarrick J. Wong { 13879fe82b8cSDarrick J. Wong percpu_counter_add_batch(&mp->m_delalloc_blks, delta, 13889fe82b8cSDarrick J. Wong XFS_DELALLOC_BATCH); 13899fe82b8cSDarrick J. Wong } 1390