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" 241da177e4SLinus Torvalds #include "xfs_error.h" 251da177e4SLinus Torvalds #include "xfs_quota.h" 261da177e4SLinus Torvalds #include "xfs_fsops.h" 276d8b79cfSDave Chinner #include "xfs_icache.h" 28a31b1d3dSBrian Foster #include "xfs_sysfs.h" 29035e00acSDarrick J. Wong #include "xfs_rmap_btree.h" 301946b91cSDarrick J. Wong #include "xfs_refcount_btree.h" 31174edb0eSDarrick J. Wong #include "xfs_reflink.h" 32ebf55872SChristoph Hellwig #include "xfs_extent_busy.h" 3339353ff6SDarrick J. Wong #include "xfs_health.h" 3413eaec4bSDarrick J. Wong #include "xfs_trace.h" 359bbafc71SDave Chinner #include "xfs_ag.h" 361da177e4SLinus Torvalds 3727174203SChristoph Hellwig static DEFINE_MUTEX(xfs_uuid_table_mutex); 3827174203SChristoph Hellwig static int xfs_uuid_table_size; 3927174203SChristoph Hellwig static uuid_t *xfs_uuid_table; 4027174203SChristoph Hellwig 41af3b6382SDarrick J. Wong void 42af3b6382SDarrick J. Wong xfs_uuid_table_free(void) 43af3b6382SDarrick J. Wong { 44af3b6382SDarrick J. Wong if (xfs_uuid_table_size == 0) 45af3b6382SDarrick J. Wong return; 46af3b6382SDarrick J. Wong kmem_free(xfs_uuid_table); 47af3b6382SDarrick J. Wong xfs_uuid_table = NULL; 48af3b6382SDarrick J. Wong xfs_uuid_table_size = 0; 49af3b6382SDarrick J. Wong } 50af3b6382SDarrick J. Wong 5127174203SChristoph Hellwig /* 5227174203SChristoph Hellwig * See if the UUID is unique among mounted XFS filesystems. 5327174203SChristoph Hellwig * Mount fails if UUID is nil or a FS with the same UUID is already mounted. 5427174203SChristoph Hellwig */ 5527174203SChristoph Hellwig STATIC int 5627174203SChristoph Hellwig xfs_uuid_mount( 5727174203SChristoph Hellwig struct xfs_mount *mp) 5827174203SChristoph Hellwig { 5927174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 6027174203SChristoph Hellwig int hole, i; 6127174203SChristoph Hellwig 628f720d9fSAmir Goldstein /* Publish UUID in struct super_block */ 6385787090SChristoph Hellwig uuid_copy(&mp->m_super->s_uuid, uuid); 648f720d9fSAmir Goldstein 650560f31aSDave Chinner if (xfs_has_nouuid(mp)) 6627174203SChristoph Hellwig return 0; 6727174203SChristoph Hellwig 68d905fdaaSAmir Goldstein if (uuid_is_null(uuid)) { 69d905fdaaSAmir Goldstein xfs_warn(mp, "Filesystem has null UUID - can't mount"); 702451337dSDave Chinner return -EINVAL; 7127174203SChristoph Hellwig } 7227174203SChristoph Hellwig 7327174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 7427174203SChristoph Hellwig for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { 75d905fdaaSAmir Goldstein if (uuid_is_null(&xfs_uuid_table[i])) { 7627174203SChristoph Hellwig hole = i; 7727174203SChristoph Hellwig continue; 7827174203SChristoph Hellwig } 7927174203SChristoph Hellwig if (uuid_equal(uuid, &xfs_uuid_table[i])) 8027174203SChristoph Hellwig goto out_duplicate; 8127174203SChristoph Hellwig } 8227174203SChristoph Hellwig 8327174203SChristoph Hellwig if (hole < 0) { 84771915c4SCarlos Maiolino xfs_uuid_table = krealloc(xfs_uuid_table, 8527174203SChristoph Hellwig (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), 86771915c4SCarlos Maiolino GFP_KERNEL | __GFP_NOFAIL); 8727174203SChristoph Hellwig hole = xfs_uuid_table_size++; 8827174203SChristoph Hellwig } 8927174203SChristoph Hellwig xfs_uuid_table[hole] = *uuid; 9027174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 9127174203SChristoph Hellwig 9227174203SChristoph Hellwig return 0; 9327174203SChristoph Hellwig 9427174203SChristoph Hellwig out_duplicate: 9527174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 96021000e5SMitsuo Hayasaka xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); 972451337dSDave Chinner return -EINVAL; 9827174203SChristoph Hellwig } 9927174203SChristoph Hellwig 10027174203SChristoph Hellwig STATIC void 10127174203SChristoph Hellwig xfs_uuid_unmount( 10227174203SChristoph Hellwig struct xfs_mount *mp) 10327174203SChristoph Hellwig { 10427174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 10527174203SChristoph Hellwig int i; 10627174203SChristoph Hellwig 1070560f31aSDave Chinner if (xfs_has_nouuid(mp)) 10827174203SChristoph Hellwig return; 10927174203SChristoph Hellwig 11027174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 11127174203SChristoph Hellwig for (i = 0; i < xfs_uuid_table_size; i++) { 112d905fdaaSAmir Goldstein if (uuid_is_null(&xfs_uuid_table[i])) 11327174203SChristoph Hellwig continue; 11427174203SChristoph Hellwig if (!uuid_equal(uuid, &xfs_uuid_table[i])) 11527174203SChristoph Hellwig continue; 11627174203SChristoph Hellwig memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); 11727174203SChristoph Hellwig break; 11827174203SChristoph Hellwig } 11927174203SChristoph Hellwig ASSERT(i < xfs_uuid_table_size); 12027174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 12127174203SChristoph Hellwig } 12227174203SChristoph Hellwig 1234cc929eeSNathan Scott /* 1244cc929eeSNathan Scott * Check size of device based on the (data/realtime) block count. 1254cc929eeSNathan Scott * Note: this check is used by the growfs code as well as mount. 1264cc929eeSNathan Scott */ 1274cc929eeSNathan Scott int 1284cc929eeSNathan Scott xfs_sb_validate_fsb_count( 1294cc929eeSNathan Scott xfs_sb_t *sbp, 130c8ce540dSDarrick J. Wong uint64_t nblocks) 1314cc929eeSNathan Scott { 1324cc929eeSNathan Scott ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); 1334cc929eeSNathan Scott ASSERT(sbp->sb_blocklog >= BBSHIFT); 1344cc929eeSNathan Scott 135d5cf09baSChristoph Hellwig /* Limited by ULONG_MAX of page cache index */ 13609cbfeafSKirill A. Shutemov if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) 1372451337dSDave Chinner return -EFBIG; 1384cc929eeSNathan Scott return 0; 1394cc929eeSNathan Scott } 1401da177e4SLinus Torvalds 1411da177e4SLinus Torvalds /* 1421da177e4SLinus Torvalds * xfs_readsb 1431da177e4SLinus Torvalds * 1441da177e4SLinus Torvalds * Does the initial read of the superblock. 1451da177e4SLinus Torvalds */ 1461da177e4SLinus Torvalds int 147ff55068cSDave Chinner xfs_readsb( 148ff55068cSDave Chinner struct xfs_mount *mp, 149ff55068cSDave Chinner int flags) 1501da177e4SLinus Torvalds { 1511da177e4SLinus Torvalds unsigned int sector_size; 15204a1e6c5SDave Chinner struct xfs_buf *bp; 15304a1e6c5SDave Chinner struct xfs_sb *sbp = &mp->m_sb; 1541da177e4SLinus Torvalds int error; 155af34e09dSDave Chinner int loud = !(flags & XFS_MFSI_QUIET); 156daba5427SEric Sandeen const struct xfs_buf_ops *buf_ops; 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds ASSERT(mp->m_sb_bp == NULL); 1591da177e4SLinus Torvalds ASSERT(mp->m_ddev_targp != NULL); 1601da177e4SLinus Torvalds 1611da177e4SLinus Torvalds /* 162daba5427SEric Sandeen * For the initial read, we must guess at the sector 163daba5427SEric Sandeen * size based on the block device. It's enough to 164daba5427SEric Sandeen * get the sb_sectsize out of the superblock and 165daba5427SEric Sandeen * then reread with the proper length. 166daba5427SEric Sandeen * We don't verify it yet, because it may not be complete. 167daba5427SEric Sandeen */ 168daba5427SEric Sandeen sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); 169daba5427SEric Sandeen buf_ops = NULL; 170daba5427SEric Sandeen 171daba5427SEric Sandeen /* 172c891c30aSBrian Foster * Allocate a (locked) buffer to hold the superblock. This will be kept 173c891c30aSBrian Foster * around at all times to optimize access to the superblock. Therefore, 174c891c30aSBrian Foster * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count 175c891c30aSBrian Foster * elevated. 1761da177e4SLinus Torvalds */ 17726af6552SDave Chinner reread: 178ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, 179c891c30aSBrian Foster BTOBB(sector_size), XBF_NO_IOACCT, &bp, 180c891c30aSBrian Foster buf_ops); 181ba372674SDave Chinner if (error) { 182eab4e633SDave Chinner if (loud) 183e721f504SDave Chinner xfs_warn(mp, "SB validate failed with error %d.", error); 184ac75a1f7SDave Chinner /* bad CRC means corrupted metadata */ 1852451337dSDave Chinner if (error == -EFSBADCRC) 1862451337dSDave Chinner error = -EFSCORRUPTED; 187ba372674SDave Chinner return error; 188eab4e633SDave Chinner } 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds /* 1911da177e4SLinus Torvalds * Initialize the mount structure from the superblock. 1921da177e4SLinus Torvalds */ 1933e6e8afdSChristoph Hellwig xfs_sb_from_disk(sbp, bp->b_addr); 194556b8883SDave Chinner 195556b8883SDave Chinner /* 196556b8883SDave Chinner * If we haven't validated the superblock, do so now before we try 197556b8883SDave Chinner * to check the sector size and reread the superblock appropriately. 198556b8883SDave Chinner */ 199556b8883SDave Chinner if (sbp->sb_magicnum != XFS_SB_MAGIC) { 200556b8883SDave Chinner if (loud) 201556b8883SDave Chinner xfs_warn(mp, "Invalid superblock magic number"); 2022451337dSDave Chinner error = -EINVAL; 203556b8883SDave Chinner goto release_buf; 204556b8883SDave Chinner } 205ff55068cSDave Chinner 2061da177e4SLinus Torvalds /* 2071da177e4SLinus Torvalds * We must be able to do sector-sized and sector-aligned IO. 2081da177e4SLinus Torvalds */ 20904a1e6c5SDave Chinner if (sector_size > sbp->sb_sectsize) { 210af34e09dSDave Chinner if (loud) 211af34e09dSDave Chinner xfs_warn(mp, "device supports %u byte sectors (not %u)", 21204a1e6c5SDave Chinner sector_size, sbp->sb_sectsize); 2132451337dSDave Chinner error = -ENOSYS; 21426af6552SDave Chinner goto release_buf; 2151da177e4SLinus Torvalds } 2161da177e4SLinus Torvalds 217556b8883SDave Chinner if (buf_ops == NULL) { 2181da177e4SLinus Torvalds /* 219daba5427SEric Sandeen * Re-read the superblock so the buffer is correctly sized, 220daba5427SEric Sandeen * and properly verified. 2211da177e4SLinus Torvalds */ 2221da177e4SLinus Torvalds xfs_buf_relse(bp); 22304a1e6c5SDave Chinner sector_size = sbp->sb_sectsize; 224daba5427SEric Sandeen buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; 22526af6552SDave Chinner goto reread; 2261da177e4SLinus Torvalds } 2271da177e4SLinus Torvalds 228a1d86e8dSDave Chinner mp->m_features |= xfs_sb_version_to_features(sbp); 2295681ca40SDave Chinner xfs_reinit_percpu_counters(mp); 2308d280b98SDavid Chinner 23104a1e6c5SDave Chinner /* no need to be quiet anymore, so reset the buf ops */ 23204a1e6c5SDave Chinner bp->b_ops = &xfs_sb_buf_ops; 23304a1e6c5SDave Chinner 2341da177e4SLinus Torvalds mp->m_sb_bp = bp; 23526af6552SDave Chinner xfs_buf_unlock(bp); 2361da177e4SLinus Torvalds return 0; 2371da177e4SLinus Torvalds 23826af6552SDave Chinner release_buf: 2391da177e4SLinus Torvalds xfs_buf_relse(bp); 2401da177e4SLinus Torvalds return error; 2411da177e4SLinus Torvalds } 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds /* 24413eaec4bSDarrick J. Wong * If the sunit/swidth change would move the precomputed root inode value, we 24513eaec4bSDarrick J. Wong * must reject the ondisk change because repair will stumble over that. 24613eaec4bSDarrick J. Wong * However, we allow the mount to proceed because we never rejected this 24713eaec4bSDarrick J. Wong * combination before. Returns true to update the sb, false otherwise. 24813eaec4bSDarrick J. Wong */ 24913eaec4bSDarrick J. Wong static inline int 25013eaec4bSDarrick J. Wong xfs_check_new_dalign( 25113eaec4bSDarrick J. Wong struct xfs_mount *mp, 25213eaec4bSDarrick J. Wong int new_dalign, 25313eaec4bSDarrick J. Wong bool *update_sb) 25413eaec4bSDarrick J. Wong { 25513eaec4bSDarrick J. Wong struct xfs_sb *sbp = &mp->m_sb; 25613eaec4bSDarrick J. Wong xfs_ino_t calc_ino; 25713eaec4bSDarrick J. Wong 25813eaec4bSDarrick J. Wong calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign); 25913eaec4bSDarrick J. Wong trace_xfs_check_new_dalign(mp, new_dalign, calc_ino); 26013eaec4bSDarrick J. Wong 26113eaec4bSDarrick J. Wong if (sbp->sb_rootino == calc_ino) { 26213eaec4bSDarrick J. Wong *update_sb = true; 26313eaec4bSDarrick J. Wong return 0; 26413eaec4bSDarrick J. Wong } 26513eaec4bSDarrick J. Wong 26613eaec4bSDarrick J. Wong xfs_warn(mp, 26713eaec4bSDarrick J. Wong "Cannot change stripe alignment; would require moving root inode."); 26813eaec4bSDarrick J. Wong 26913eaec4bSDarrick J. Wong /* 27013eaec4bSDarrick J. Wong * XXX: Next time we add a new incompat feature, this should start 27113eaec4bSDarrick J. Wong * returning -EINVAL to fail the mount. Until then, spit out a warning 27213eaec4bSDarrick J. Wong * that we're ignoring the administrator's instructions. 27313eaec4bSDarrick J. Wong */ 27413eaec4bSDarrick J. Wong xfs_warn(mp, "Skipping superblock stripe alignment update."); 27513eaec4bSDarrick J. Wong *update_sb = false; 27613eaec4bSDarrick J. Wong return 0; 27713eaec4bSDarrick J. Wong } 27813eaec4bSDarrick J. Wong 27913eaec4bSDarrick J. Wong /* 2804f5b1b3aSDarrick J. Wong * If we were provided with new sunit/swidth values as mount options, make sure 2814f5b1b3aSDarrick J. Wong * that they pass basic alignment and superblock feature checks, and convert 2824f5b1b3aSDarrick J. Wong * them into the same units (FSB) that everything else expects. This step 2834f5b1b3aSDarrick J. Wong * /must/ be done before computing the inode geometry. 2841da177e4SLinus Torvalds */ 2850771fb45SEric Sandeen STATIC int 2864f5b1b3aSDarrick J. Wong xfs_validate_new_dalign( 2874f5b1b3aSDarrick J. Wong struct xfs_mount *mp) 2881da177e4SLinus Torvalds { 2894f5b1b3aSDarrick J. Wong if (mp->m_dalign == 0) 2904f5b1b3aSDarrick J. Wong return 0; 2911da177e4SLinus Torvalds 2921da177e4SLinus Torvalds /* 2931da177e4SLinus Torvalds * If stripe unit and stripe width are not multiples 2941da177e4SLinus Torvalds * of the fs blocksize turn off alignment. 2951da177e4SLinus Torvalds */ 2961da177e4SLinus Torvalds if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || 2971da177e4SLinus Torvalds (BBTOB(mp->m_swidth) & mp->m_blockmask)) { 29839a45d84SJie Liu xfs_warn(mp, 29939a45d84SJie Liu "alignment check failed: sunit/swidth vs. blocksize(%d)", 3004f5b1b3aSDarrick J. Wong mp->m_sb.sb_blocksize); 3012451337dSDave Chinner return -EINVAL; 3021da177e4SLinus Torvalds } else { 3031da177e4SLinus Torvalds /* 3041da177e4SLinus Torvalds * Convert the stripe unit and width to FSBs. 3051da177e4SLinus Torvalds */ 3061da177e4SLinus Torvalds mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); 3074f5b1b3aSDarrick J. Wong if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) { 30853487786SDave Chinner xfs_warn(mp, 30939a45d84SJie Liu "alignment check failed: sunit/swidth vs. agsize(%d)", 3104f5b1b3aSDarrick J. Wong mp->m_sb.sb_agblocks); 3112451337dSDave Chinner return -EINVAL; 3121da177e4SLinus Torvalds } else if (mp->m_dalign) { 3131da177e4SLinus Torvalds mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); 3141da177e4SLinus Torvalds } else { 31539a45d84SJie Liu xfs_warn(mp, 31639a45d84SJie Liu "alignment check failed: sunit(%d) less than bsize(%d)", 3174f5b1b3aSDarrick J. Wong mp->m_dalign, mp->m_sb.sb_blocksize); 3182451337dSDave Chinner return -EINVAL; 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds } 3211da177e4SLinus Torvalds 32238c26bfdSDave Chinner if (!xfs_has_dalign(mp)) { 32334d7f603SJie Liu xfs_warn(mp, 32434d7f603SJie Liu "cannot change alignment: superblock does not support data alignment"); 3252451337dSDave Chinner return -EINVAL; 3261da177e4SLinus Torvalds } 3274f5b1b3aSDarrick J. Wong 3284f5b1b3aSDarrick J. Wong return 0; 3294f5b1b3aSDarrick J. Wong } 3304f5b1b3aSDarrick J. Wong 3314f5b1b3aSDarrick J. Wong /* Update alignment values based on mount options and sb values. */ 3324f5b1b3aSDarrick J. Wong STATIC int 3334f5b1b3aSDarrick J. Wong xfs_update_alignment( 3344f5b1b3aSDarrick J. Wong struct xfs_mount *mp) 3354f5b1b3aSDarrick J. Wong { 3364f5b1b3aSDarrick J. Wong struct xfs_sb *sbp = &mp->m_sb; 3374f5b1b3aSDarrick J. Wong 3384f5b1b3aSDarrick J. Wong if (mp->m_dalign) { 33913eaec4bSDarrick J. Wong bool update_sb; 34013eaec4bSDarrick J. Wong int error; 34113eaec4bSDarrick J. Wong 3424f5b1b3aSDarrick J. Wong if (sbp->sb_unit == mp->m_dalign && 3434f5b1b3aSDarrick J. Wong sbp->sb_width == mp->m_swidth) 3444f5b1b3aSDarrick J. Wong return 0; 3454f5b1b3aSDarrick J. Wong 34613eaec4bSDarrick J. Wong error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb); 34713eaec4bSDarrick J. Wong if (error || !update_sb) 34813eaec4bSDarrick J. Wong return error; 34913eaec4bSDarrick J. Wong 3504f5b1b3aSDarrick J. Wong sbp->sb_unit = mp->m_dalign; 3514f5b1b3aSDarrick J. Wong sbp->sb_width = mp->m_swidth; 3524f5b1b3aSDarrick J. Wong mp->m_update_sb = true; 3530560f31aSDave Chinner } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) { 3541da177e4SLinus Torvalds mp->m_dalign = sbp->sb_unit; 3551da177e4SLinus Torvalds mp->m_swidth = sbp->sb_width; 3561da177e4SLinus Torvalds } 3571da177e4SLinus Torvalds 3580771fb45SEric Sandeen return 0; 3590771fb45SEric Sandeen } 3601da177e4SLinus Torvalds 3610771fb45SEric Sandeen /* 362055388a3SDave Chinner * precalculate the low space thresholds for dynamic speculative preallocation. 363055388a3SDave Chinner */ 364055388a3SDave Chinner void 365055388a3SDave Chinner xfs_set_low_space_thresholds( 366055388a3SDave Chinner struct xfs_mount *mp) 367055388a3SDave Chinner { 36865f03d86SDarrick J. Wong uint64_t dblocks = mp->m_sb.sb_dblocks; 36965f03d86SDarrick J. Wong uint64_t rtexts = mp->m_sb.sb_rextents; 370055388a3SDave Chinner int i; 371055388a3SDave Chinner 37265f03d86SDarrick J. Wong do_div(dblocks, 100); 37365f03d86SDarrick J. Wong do_div(rtexts, 100); 374055388a3SDave Chinner 37565f03d86SDarrick J. Wong for (i = 0; i < XFS_LOWSP_MAX; i++) { 37665f03d86SDarrick J. Wong mp->m_low_space[i] = dblocks * (i + 1); 37765f03d86SDarrick J. Wong mp->m_low_rtexts[i] = rtexts * (i + 1); 378055388a3SDave Chinner } 379055388a3SDave Chinner } 380055388a3SDave Chinner 3811da177e4SLinus Torvalds /* 3820471f62eSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 3831da177e4SLinus Torvalds */ 3840771fb45SEric Sandeen STATIC int 385ba372674SDave Chinner xfs_check_sizes( 386ba372674SDave Chinner struct xfs_mount *mp) 3870771fb45SEric Sandeen { 388ba372674SDave Chinner struct xfs_buf *bp; 3890771fb45SEric Sandeen xfs_daddr_t d; 390ba372674SDave Chinner int error; 3910771fb45SEric Sandeen 3921da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 3931da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { 3940b932cccSDave Chinner xfs_warn(mp, "filesystem size mismatch detected"); 3952451337dSDave Chinner return -EFBIG; 3961da177e4SLinus Torvalds } 397ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, 3981da177e4SLinus Torvalds d - XFS_FSS_TO_BB(mp, 1), 399ba372674SDave Chinner XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 400ba372674SDave Chinner if (error) { 4010b932cccSDave Chinner xfs_warn(mp, "last sector read failed"); 402ba372674SDave Chinner return error; 4031da177e4SLinus Torvalds } 4041922c949SDave Chinner xfs_buf_relse(bp); 4051da177e4SLinus Torvalds 406ba372674SDave Chinner if (mp->m_logdev_targp == mp->m_ddev_targp) 407ba372674SDave Chinner return 0; 408ba372674SDave Chinner 4091da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 4101da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { 4110b932cccSDave Chinner xfs_warn(mp, "log size mismatch detected"); 4122451337dSDave Chinner return -EFBIG; 4131da177e4SLinus Torvalds } 414ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_logdev_targp, 4151da177e4SLinus Torvalds d - XFS_FSB_TO_BB(mp, 1), 416ba372674SDave Chinner XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); 417ba372674SDave Chinner if (error) { 4180b932cccSDave Chinner xfs_warn(mp, "log device read failed"); 419ba372674SDave Chinner return error; 4201da177e4SLinus Torvalds } 4211922c949SDave Chinner xfs_buf_relse(bp); 4220771fb45SEric Sandeen return 0; 4230771fb45SEric Sandeen } 4240771fb45SEric Sandeen 4250771fb45SEric Sandeen /* 4267d095257SChristoph Hellwig * Clear the quotaflags in memory and in the superblock. 4277d095257SChristoph Hellwig */ 4287d095257SChristoph Hellwig int 4297d095257SChristoph Hellwig xfs_mount_reset_sbqflags( 4307d095257SChristoph Hellwig struct xfs_mount *mp) 4317d095257SChristoph Hellwig { 4327d095257SChristoph Hellwig mp->m_qflags = 0; 4337d095257SChristoph Hellwig 43461e63ecbSDave Chinner /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ 4357d095257SChristoph Hellwig if (mp->m_sb.sb_qflags == 0) 4367d095257SChristoph Hellwig return 0; 4377d095257SChristoph Hellwig spin_lock(&mp->m_sb_lock); 4387d095257SChristoph Hellwig mp->m_sb.sb_qflags = 0; 4397d095257SChristoph Hellwig spin_unlock(&mp->m_sb_lock); 4407d095257SChristoph Hellwig 44161e63ecbSDave Chinner if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) 4427d095257SChristoph Hellwig return 0; 4437d095257SChristoph Hellwig 44461e63ecbSDave Chinner return xfs_sync_sb(mp, false); 4457d095257SChristoph Hellwig } 4467d095257SChristoph Hellwig 447c8ce540dSDarrick J. Wong uint64_t 448d5db0f97SEric Sandeen xfs_default_resblks(xfs_mount_t *mp) 449d5db0f97SEric Sandeen { 450c8ce540dSDarrick J. Wong uint64_t resblks; 451d5db0f97SEric Sandeen 452d5db0f97SEric Sandeen /* 4538babd8a2SDave Chinner * We default to 5% or 8192 fsbs of space reserved, whichever is 4548babd8a2SDave Chinner * smaller. This is intended to cover concurrent allocation 4558babd8a2SDave Chinner * transactions when we initially hit enospc. These each require a 4 4568babd8a2SDave Chinner * block reservation. Hence by default we cover roughly 2000 concurrent 4578babd8a2SDave Chinner * allocation reservations. 458d5db0f97SEric Sandeen */ 459d5db0f97SEric Sandeen resblks = mp->m_sb.sb_dblocks; 460d5db0f97SEric Sandeen do_div(resblks, 20); 461c8ce540dSDarrick J. Wong resblks = min_t(uint64_t, resblks, 8192); 462d5db0f97SEric Sandeen return resblks; 463d5db0f97SEric Sandeen } 464d5db0f97SEric Sandeen 4652e9e6481SDarrick J. Wong /* Ensure the summary counts are correct. */ 4662e9e6481SDarrick J. Wong STATIC int 4672e9e6481SDarrick J. Wong xfs_check_summary_counts( 4682e9e6481SDarrick J. Wong struct xfs_mount *mp) 4692e9e6481SDarrick J. Wong { 4702e9e6481SDarrick J. Wong /* 4712e9e6481SDarrick J. Wong * The AG0 superblock verifier rejects in-progress filesystems, 4722e9e6481SDarrick J. Wong * so we should never see the flag set this far into mounting. 4732e9e6481SDarrick J. Wong */ 4742e9e6481SDarrick J. Wong if (mp->m_sb.sb_inprogress) { 4752e9e6481SDarrick J. Wong xfs_err(mp, "sb_inprogress set after log recovery??"); 4762e9e6481SDarrick J. Wong WARN_ON(1); 4772e9e6481SDarrick J. Wong return -EFSCORRUPTED; 4782e9e6481SDarrick J. Wong } 4792e9e6481SDarrick J. Wong 4802e9e6481SDarrick J. Wong /* 4812e9e6481SDarrick J. Wong * Now the log is mounted, we know if it was an unclean shutdown or 4822e9e6481SDarrick J. Wong * not. If it was, with the first phase of recovery has completed, we 4832e9e6481SDarrick J. Wong * have consistent AG blocks on disk. We have not recovered EFIs yet, 4842e9e6481SDarrick J. Wong * but they are recovered transactionally in the second recovery phase 4852e9e6481SDarrick J. Wong * later. 4862e9e6481SDarrick J. Wong * 4872e9e6481SDarrick J. Wong * If the log was clean when we mounted, we can check the summary 4882e9e6481SDarrick J. Wong * counters. If any of them are obviously incorrect, we can recompute 4892e9e6481SDarrick J. Wong * them from the AGF headers in the next step. 4902e9e6481SDarrick J. Wong */ 4912e973b2cSDave Chinner if (xfs_is_clean(mp) && 4922e9e6481SDarrick J. Wong (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks || 49300d22a1cSDarrick J. Wong !xfs_verify_icount(mp, mp->m_sb.sb_icount) || 4942e9e6481SDarrick J. Wong mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) 49539353ff6SDarrick J. Wong xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); 4962e9e6481SDarrick J. Wong 4972e9e6481SDarrick J. Wong /* 4982e9e6481SDarrick J. Wong * We can safely re-initialise incore superblock counters from the 4992e9e6481SDarrick J. Wong * per-ag data. These may not be correct if the filesystem was not 5002e9e6481SDarrick J. Wong * cleanly unmounted, so we waited for recovery to finish before doing 5012e9e6481SDarrick J. Wong * this. 5022e9e6481SDarrick J. Wong * 5032e9e6481SDarrick J. Wong * If the filesystem was cleanly unmounted or the previous check did 5042e9e6481SDarrick J. Wong * not flag anything weird, then we can trust the values in the 5052e9e6481SDarrick J. Wong * superblock to be correct and we don't need to do anything here. 5062e9e6481SDarrick J. Wong * Otherwise, recalculate the summary counters. 5072e9e6481SDarrick J. Wong */ 5082e973b2cSDave Chinner if ((!xfs_has_lazysbcount(mp) || xfs_is_clean(mp)) && 50939353ff6SDarrick J. Wong !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) 5102e9e6481SDarrick J. Wong return 0; 5112e9e6481SDarrick J. Wong 5122e9e6481SDarrick J. Wong return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount); 5132e9e6481SDarrick J. Wong } 5142e9e6481SDarrick J. Wong 5157d095257SChristoph Hellwig /* 516d336f7ebSDarrick J. Wong * Flush and reclaim dirty inodes in preparation for unmount. Inodes and 517d336f7ebSDarrick J. Wong * internal inode structures can be sitting in the CIL and AIL at this point, 518d336f7ebSDarrick J. Wong * so we need to unpin them, write them back and/or reclaim them before unmount 519ab23a776SDave Chinner * can proceed. In other words, callers are required to have inactivated all 520ab23a776SDave Chinner * inodes. 521d336f7ebSDarrick J. Wong * 522d336f7ebSDarrick J. Wong * An inode cluster that has been freed can have its buffer still pinned in 523d336f7ebSDarrick J. Wong * memory because the transaction is still sitting in a iclog. The stale inodes 524d336f7ebSDarrick J. Wong * on that buffer will be pinned to the buffer until the transaction hits the 525d336f7ebSDarrick J. Wong * disk and the callbacks run. Pushing the AIL will skip the stale inodes and 526d336f7ebSDarrick J. Wong * may never see the pinned buffer, so nothing will push out the iclog and 527d336f7ebSDarrick J. Wong * unpin the buffer. 528d336f7ebSDarrick J. Wong * 529d336f7ebSDarrick J. Wong * Hence we need to force the log to unpin everything first. However, log 530d336f7ebSDarrick J. Wong * forces don't wait for the discards they issue to complete, so we have to 531d336f7ebSDarrick J. Wong * explicitly wait for them to complete here as well. 532d336f7ebSDarrick J. Wong * 533d336f7ebSDarrick J. Wong * Then we can tell the world we are unmounting so that error handling knows 534d336f7ebSDarrick J. Wong * that the filesystem is going away and we should error out anything that we 535d336f7ebSDarrick J. Wong * have been retrying in the background. This will prevent never-ending 536d336f7ebSDarrick J. Wong * retries in AIL pushing from hanging the unmount. 537d336f7ebSDarrick J. Wong * 538d336f7ebSDarrick J. Wong * Finally, we can push the AIL to clean all the remaining dirty objects, then 539d336f7ebSDarrick J. Wong * reclaim the remaining inodes that are still in memory at this point in time. 540d336f7ebSDarrick J. Wong */ 541d336f7ebSDarrick J. Wong static void 542d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes( 543d336f7ebSDarrick J. Wong struct xfs_mount *mp) 544d336f7ebSDarrick J. Wong { 545d336f7ebSDarrick J. Wong xfs_log_force(mp, XFS_LOG_SYNC); 546d336f7ebSDarrick J. Wong xfs_extent_busy_wait_all(mp); 547d336f7ebSDarrick J. Wong flush_workqueue(xfs_discard_wq); 548d336f7ebSDarrick J. Wong 5492e973b2cSDave Chinner set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate); 550d336f7ebSDarrick J. Wong 551d336f7ebSDarrick J. Wong xfs_ail_push_all_sync(mp->m_ail); 552ab23a776SDave Chinner xfs_inodegc_stop(mp); 553d336f7ebSDarrick J. Wong cancel_delayed_work_sync(&mp->m_reclaim_work); 554d336f7ebSDarrick J. Wong xfs_reclaim_inodes(mp); 555d336f7ebSDarrick J. Wong xfs_health_unmount(mp); 556d336f7ebSDarrick J. Wong } 557d336f7ebSDarrick J. Wong 558b2941046SDave Chinner static void 559b2941046SDave Chinner xfs_mount_setup_inode_geom( 560b2941046SDave Chinner struct xfs_mount *mp) 561b2941046SDave Chinner { 562b2941046SDave Chinner struct xfs_ino_geometry *igeo = M_IGEO(mp); 563b2941046SDave Chinner 564b2941046SDave Chinner igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp); 565b2941046SDave Chinner ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp)); 566b2941046SDave Chinner 567b2941046SDave Chinner xfs_ialloc_setup_geometry(mp); 568b2941046SDave Chinner } 569b2941046SDave Chinner 570*b74e15d7SDarrick J. Wong /* Compute maximum possible height for per-AG btree types for this fs. */ 571*b74e15d7SDarrick J. Wong static inline void 572*b74e15d7SDarrick J. Wong xfs_agbtree_compute_maxlevels( 573*b74e15d7SDarrick J. Wong struct xfs_mount *mp) 574*b74e15d7SDarrick J. Wong { 575*b74e15d7SDarrick J. Wong unsigned int levels; 576*b74e15d7SDarrick J. Wong 577*b74e15d7SDarrick J. Wong levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels); 578*b74e15d7SDarrick J. Wong levels = max(levels, mp->m_rmap_maxlevels); 579*b74e15d7SDarrick J. Wong mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels); 580*b74e15d7SDarrick J. Wong } 581*b74e15d7SDarrick J. Wong 582d336f7ebSDarrick J. Wong /* 5830771fb45SEric Sandeen * This function does the following on an initial mount of a file system: 5840771fb45SEric Sandeen * - reads the superblock from disk and init the mount struct 5850771fb45SEric Sandeen * - if we're a 32-bit kernel, do a size check on the superblock 5860771fb45SEric Sandeen * so we don't mount terabyte filesystems 5870771fb45SEric Sandeen * - init mount struct realtime fields 5880771fb45SEric Sandeen * - allocate inode hash table for fs 5890771fb45SEric Sandeen * - init directory manager 5900771fb45SEric Sandeen * - perform recovery and init the log manager 5910771fb45SEric Sandeen */ 5920771fb45SEric Sandeen int 5930771fb45SEric Sandeen xfs_mountfs( 594f0b2efadSBrian Foster struct xfs_mount *mp) 5950771fb45SEric Sandeen { 596f0b2efadSBrian Foster struct xfs_sb *sbp = &(mp->m_sb); 597f0b2efadSBrian Foster struct xfs_inode *rip; 598ef325959SDarrick J. Wong struct xfs_ino_geometry *igeo = M_IGEO(mp); 599c8ce540dSDarrick J. Wong uint64_t resblks; 6007d095257SChristoph Hellwig uint quotamount = 0; 6017d095257SChristoph Hellwig uint quotaflags = 0; 6020771fb45SEric Sandeen int error = 0; 6030771fb45SEric Sandeen 604ff55068cSDave Chinner xfs_sb_mount_common(mp, sbp); 6050771fb45SEric Sandeen 6060771fb45SEric Sandeen /* 607074e427bSDave Chinner * Check for a mismatched features2 values. Older kernels read & wrote 608074e427bSDave Chinner * into the wrong sb offset for sb_features2 on some platforms due to 609074e427bSDave Chinner * xfs_sb_t not being 64bit size aligned when sb_features2 was added, 610074e427bSDave Chinner * which made older superblock reading/writing routines swap it as a 611074e427bSDave Chinner * 64-bit value. 612ee1c0908SDavid Chinner * 613e6957ea4SEric Sandeen * For backwards compatibility, we make both slots equal. 614e6957ea4SEric Sandeen * 615074e427bSDave Chinner * If we detect a mismatched field, we OR the set bits into the existing 616074e427bSDave Chinner * features2 field in case it has already been modified; we don't want 617074e427bSDave Chinner * to lose any features. We then update the bad location with the ORed 618074e427bSDave Chinner * value so that older kernels will see any features2 flags. The 619074e427bSDave Chinner * superblock writeback code ensures the new sb_features2 is copied to 620074e427bSDave Chinner * sb_bad_features2 before it is logged or written to disk. 621ee1c0908SDavid Chinner */ 622e6957ea4SEric Sandeen if (xfs_sb_has_mismatched_features2(sbp)) { 6230b932cccSDave Chinner xfs_warn(mp, "correcting sb_features alignment problem"); 624ee1c0908SDavid Chinner sbp->sb_features2 |= sbp->sb_bad_features2; 62561e63ecbSDave Chinner mp->m_update_sb = true; 6267c12f296STim Shimmin } 627e6957ea4SEric Sandeen 628ee1c0908SDavid Chinner 629263997a6SDave Chinner /* always use v2 inodes by default now */ 630263997a6SDave Chinner if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { 631263997a6SDave Chinner mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; 63238c26bfdSDave Chinner mp->m_features |= XFS_FEAT_NLINK; 63361e63ecbSDave Chinner mp->m_update_sb = true; 634263997a6SDave Chinner } 635263997a6SDave Chinner 636ee1c0908SDavid Chinner /* 6374f5b1b3aSDarrick J. Wong * If we were given new sunit/swidth options, do some basic validation 6384f5b1b3aSDarrick J. Wong * checks and convert the incore dalign and swidth values to the 6394f5b1b3aSDarrick J. Wong * same units (FSB) that everything else uses. This /must/ happen 6404f5b1b3aSDarrick J. Wong * before computing the inode geometry. 6410771fb45SEric Sandeen */ 6424f5b1b3aSDarrick J. Wong error = xfs_validate_new_dalign(mp); 6430771fb45SEric Sandeen if (error) 644f9057e3dSChristoph Hellwig goto out; 6450771fb45SEric Sandeen 6460771fb45SEric Sandeen xfs_alloc_compute_maxlevels(mp); 6470771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); 6480771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); 649b2941046SDave Chinner xfs_mount_setup_inode_geom(mp); 650035e00acSDarrick J. Wong xfs_rmapbt_compute_maxlevels(mp); 6511946b91cSDarrick J. Wong xfs_refcountbt_compute_maxlevels(mp); 6520771fb45SEric Sandeen 653*b74e15d7SDarrick J. Wong xfs_agbtree_compute_maxlevels(mp); 654*b74e15d7SDarrick J. Wong 6554f5b1b3aSDarrick J. Wong /* 6564f5b1b3aSDarrick J. Wong * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks 6574f5b1b3aSDarrick J. Wong * is NOT aligned turn off m_dalign since allocator alignment is within 6584f5b1b3aSDarrick J. Wong * an ag, therefore ag has to be aligned at stripe boundary. Note that 6594f5b1b3aSDarrick J. Wong * we must compute the free space and rmap btree geometry before doing 6604f5b1b3aSDarrick J. Wong * this. 6614f5b1b3aSDarrick J. Wong */ 6624f5b1b3aSDarrick J. Wong error = xfs_update_alignment(mp); 6634f5b1b3aSDarrick J. Wong if (error) 6644f5b1b3aSDarrick J. Wong goto out; 6654f5b1b3aSDarrick J. Wong 666e6b3bb78SCarlos Maiolino /* enable fail_at_unmount as default */ 667749f24f3SThomas Meyer mp->m_fail_unmount = true; 668e6b3bb78SCarlos Maiolino 669e1d3d218SIan Kent error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, 670e1d3d218SIan Kent NULL, mp->m_super->s_id); 67127174203SChristoph Hellwig if (error) 672f9057e3dSChristoph Hellwig goto out; 6731da177e4SLinus Torvalds 674225e4635SBill O'Donnell error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, 675225e4635SBill O'Donnell &mp->m_kobj, "stats"); 676a31b1d3dSBrian Foster if (error) 677a31b1d3dSBrian Foster goto out_remove_sysfs; 678a31b1d3dSBrian Foster 679192852beSCarlos Maiolino error = xfs_error_sysfs_init(mp); 680225e4635SBill O'Donnell if (error) 681225e4635SBill O'Donnell goto out_del_stats; 682225e4635SBill O'Donnell 68331965ef3SDarrick J. Wong error = xfs_errortag_init(mp); 68431965ef3SDarrick J. Wong if (error) 68531965ef3SDarrick J. Wong goto out_remove_error_sysfs; 686192852beSCarlos Maiolino 687192852beSCarlos Maiolino error = xfs_uuid_mount(mp); 688192852beSCarlos Maiolino if (error) 68931965ef3SDarrick J. Wong goto out_remove_errortag; 690192852beSCarlos Maiolino 6911da177e4SLinus Torvalds /* 6922fcddee8SChristoph Hellwig * Update the preferred write size based on the information from the 6932fcddee8SChristoph Hellwig * on-disk superblock. 6940771fb45SEric Sandeen */ 6952fcddee8SChristoph Hellwig mp->m_allocsize_log = 6962fcddee8SChristoph Hellwig max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log); 6972fcddee8SChristoph Hellwig mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog); 6980771fb45SEric Sandeen 699055388a3SDave Chinner /* set the low space thresholds for dynamic preallocation */ 700055388a3SDave Chinner xfs_set_low_space_thresholds(mp); 701055388a3SDave Chinner 7020771fb45SEric Sandeen /* 703e5376fc1SBrian Foster * If enabled, sparse inode chunk alignment is expected to match the 704e5376fc1SBrian Foster * cluster size. Full inode chunk alignment must match the chunk size, 705e5376fc1SBrian Foster * but that is checked on sb read verification... 706e5376fc1SBrian Foster */ 70738c26bfdSDave Chinner if (xfs_has_sparseinodes(mp) && 708e5376fc1SBrian Foster mp->m_sb.sb_spino_align != 709490d451fSDarrick J. Wong XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) { 710e5376fc1SBrian Foster xfs_warn(mp, 711e5376fc1SBrian Foster "Sparse inode block alignment (%u) must match cluster size (%llu).", 712e5376fc1SBrian Foster mp->m_sb.sb_spino_align, 713490d451fSDarrick J. Wong XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)); 714e5376fc1SBrian Foster error = -EINVAL; 715e5376fc1SBrian Foster goto out_remove_uuid; 716e5376fc1SBrian Foster } 717e5376fc1SBrian Foster 718e5376fc1SBrian Foster /* 719c2bfbc9bSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 7200771fb45SEric Sandeen */ 7214249023aSChristoph Hellwig error = xfs_check_sizes(mp); 7220771fb45SEric Sandeen if (error) 723f9057e3dSChristoph Hellwig goto out_remove_uuid; 7240771fb45SEric Sandeen 7250771fb45SEric Sandeen /* 7261da177e4SLinus Torvalds * Initialize realtime fields in the mount structure 7271da177e4SLinus Torvalds */ 7280771fb45SEric Sandeen error = xfs_rtmount_init(mp); 7290771fb45SEric Sandeen if (error) { 7300b932cccSDave Chinner xfs_warn(mp, "RT mount failed"); 731f9057e3dSChristoph Hellwig goto out_remove_uuid; 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds /* 7351da177e4SLinus Torvalds * Copies the low order bits of the timestamp and the randomly 7361da177e4SLinus Torvalds * set "sequence" number out of a UUID. 7371da177e4SLinus Torvalds */ 738cb0ba6ccSChristoph Hellwig mp->m_fixedfsid[0] = 739cb0ba6ccSChristoph Hellwig (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) | 740cb0ba6ccSChristoph Hellwig get_unaligned_be16(&sbp->sb_uuid.b[4]); 741cb0ba6ccSChristoph Hellwig mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]); 7421da177e4SLinus Torvalds 7430650b554SDave Chinner error = xfs_da_mount(mp); 7440650b554SDave Chinner if (error) { 7450650b554SDave Chinner xfs_warn(mp, "Failed dir/attr init: %d", error); 7460650b554SDave Chinner goto out_remove_uuid; 7470650b554SDave Chinner } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * Initialize the precomputed transaction reservations values. 7511da177e4SLinus Torvalds */ 7521da177e4SLinus Torvalds xfs_trans_init(mp); 7531da177e4SLinus Torvalds 7541da177e4SLinus Torvalds /* 7551da177e4SLinus Torvalds * Allocate and initialize the per-ag data. 7561da177e4SLinus Torvalds */ 7571c1c6ebcSDave Chinner error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); 7581c1c6ebcSDave Chinner if (error) { 7590b932cccSDave Chinner xfs_warn(mp, "Failed per-ag init: %d", error); 7600650b554SDave Chinner goto out_free_dir; 7611c1c6ebcSDave Chinner } 7621da177e4SLinus Torvalds 763a71895c5SDarrick J. Wong if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) { 7640b932cccSDave Chinner xfs_warn(mp, "no log defined"); 7652451337dSDave Chinner error = -EFSCORRUPTED; 766f9057e3dSChristoph Hellwig goto out_free_perag; 767f9057e3dSChristoph Hellwig } 768f9057e3dSChristoph Hellwig 76940b1de00SDarrick J. Wong error = xfs_inodegc_register_shrinker(mp); 77040b1de00SDarrick J. Wong if (error) 77140b1de00SDarrick J. Wong goto out_fail_wait; 77240b1de00SDarrick J. Wong 7731da177e4SLinus Torvalds /* 774f0b2efadSBrian Foster * Log's mount-time initialization. The first part of recovery can place 775f0b2efadSBrian Foster * some items on the AIL, to be handled when recovery is finished or 776f0b2efadSBrian Foster * cancelled. 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds error = xfs_log_mount(mp, mp->m_logdev_targp, 7791da177e4SLinus Torvalds XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), 7801da177e4SLinus Torvalds XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); 7811da177e4SLinus Torvalds if (error) { 7820b932cccSDave Chinner xfs_warn(mp, "log mount failed"); 78340b1de00SDarrick J. Wong goto out_inodegc_shrinker; 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds 7862e9e6481SDarrick J. Wong /* Make sure the summary counts are ok. */ 7872e9e6481SDarrick J. Wong error = xfs_check_summary_counts(mp); 788f9057e3dSChristoph Hellwig if (error) 7896eee8972Skbuild test robot goto out_log_dealloc; 790f9057e3dSChristoph Hellwig 791ab23a776SDave Chinner /* Enable background inode inactivation workers. */ 792ab23a776SDave Chinner xfs_inodegc_start(mp); 7936f649091SDarrick J. Wong xfs_blockgc_start(mp); 794ab23a776SDave Chinner 79592821e2bSDavid Chinner /* 796e23b55d5SDave Chinner * Now that we've recovered any pending superblock feature bit 797e23b55d5SDave Chinner * additions, we can finish setting up the attr2 behaviour for the 7980560f31aSDave Chinner * mount. The noattr2 option overrides the superblock flag, so only 7990560f31aSDave Chinner * check the superblock feature flag if the mount option is not set. 800e23b55d5SDave Chinner */ 8010560f31aSDave Chinner if (xfs_has_noattr2(mp)) { 8020560f31aSDave Chinner mp->m_features &= ~XFS_FEAT_ATTR2; 8030560f31aSDave Chinner } else if (!xfs_has_attr2(mp) && 8040560f31aSDave Chinner (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) { 8050560f31aSDave Chinner mp->m_features |= XFS_FEAT_ATTR2; 8060560f31aSDave Chinner } 807e23b55d5SDave Chinner 808e23b55d5SDave Chinner /* 8091da177e4SLinus Torvalds * Get and sanity-check the root inode. 8101da177e4SLinus Torvalds * Save the pointer to it in the mount structure. 8111da177e4SLinus Torvalds */ 812541b5accSDave Chinner error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED, 813541b5accSDave Chinner XFS_ILOCK_EXCL, &rip); 8141da177e4SLinus Torvalds if (error) { 815541b5accSDave Chinner xfs_warn(mp, 816541b5accSDave Chinner "Failed to read root inode 0x%llx, error %d", 817541b5accSDave Chinner sbp->sb_rootino, -error); 818f9057e3dSChristoph Hellwig goto out_log_dealloc; 8191da177e4SLinus Torvalds } 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds ASSERT(rip != NULL); 8221da177e4SLinus Torvalds 823a71895c5SDarrick J. Wong if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) { 8240b932cccSDave Chinner xfs_warn(mp, "corrupted root inode %llu: not a directory", 825b6574520SNathan Scott (unsigned long long)rip->i_ino); 8261da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8272451337dSDave Chinner error = -EFSCORRUPTED; 828f9057e3dSChristoph Hellwig goto out_rele_rip; 8291da177e4SLinus Torvalds } 8301da177e4SLinus Torvalds mp->m_rootip = rip; /* save it */ 8311da177e4SLinus Torvalds 8321da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds /* 8351da177e4SLinus Torvalds * Initialize realtime inode pointers in the mount structure 8361da177e4SLinus Torvalds */ 8370771fb45SEric Sandeen error = xfs_rtmount_inodes(mp); 8380771fb45SEric Sandeen if (error) { 8391da177e4SLinus Torvalds /* 8401da177e4SLinus Torvalds * Free up the root inode. 8411da177e4SLinus Torvalds */ 8420b932cccSDave Chinner xfs_warn(mp, "failed to read RT inodes"); 843f9057e3dSChristoph Hellwig goto out_rele_rip; 8441da177e4SLinus Torvalds } 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds /* 8477884bc86SChristoph Hellwig * If this is a read-only mount defer the superblock updates until 8487884bc86SChristoph Hellwig * the next remount into writeable mode. Otherwise we would never 8497884bc86SChristoph Hellwig * perform the update e.g. for the root filesystem. 8501da177e4SLinus Torvalds */ 8512e973b2cSDave Chinner if (mp->m_update_sb && !xfs_is_readonly(mp)) { 85261e63ecbSDave Chinner error = xfs_sync_sb(mp, false); 853e5720eecSDavid Chinner if (error) { 8540b932cccSDave Chinner xfs_warn(mp, "failed to write sb changes"); 855b93b6e43SChristoph Hellwig goto out_rtunmount; 856e5720eecSDavid Chinner } 857e5720eecSDavid Chinner } 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * Initialise the XFS quota management subsystem for this mount 8611da177e4SLinus Torvalds */ 862149e53afSChristoph Hellwig if (XFS_IS_QUOTA_ON(mp)) { 8637d095257SChristoph Hellwig error = xfs_qm_newmount(mp, "amount, "aflags); 8640771fb45SEric Sandeen if (error) 865b93b6e43SChristoph Hellwig goto out_rtunmount; 8667d095257SChristoph Hellwig } else { 8677d095257SChristoph Hellwig /* 8687d095257SChristoph Hellwig * If a file system had quotas running earlier, but decided to 8697d095257SChristoph Hellwig * mount without -o uquota/pquota/gquota options, revoke the 8707d095257SChristoph Hellwig * quotachecked license. 8717d095257SChristoph Hellwig */ 8727d095257SChristoph Hellwig if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { 8730b932cccSDave Chinner xfs_notice(mp, "resetting quota flags"); 8747d095257SChristoph Hellwig error = xfs_mount_reset_sbqflags(mp); 8757d095257SChristoph Hellwig if (error) 876a70a4fa5SBrian Foster goto out_rtunmount; 8777d095257SChristoph Hellwig } 8787d095257SChristoph Hellwig } 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds /* 881f0b2efadSBrian Foster * Finish recovering the file system. This part needed to be delayed 882f0b2efadSBrian Foster * until after the root and real-time bitmap inodes were consistently 88381ed9475SDarrick J. Wong * read in. Temporarily create per-AG space reservations for metadata 88481ed9475SDarrick J. Wong * btree shape changes because space freeing transactions (for inode 88581ed9475SDarrick J. Wong * inactivation) require the per-AG reservation in lieu of reserving 88681ed9475SDarrick J. Wong * blocks. 8871da177e4SLinus Torvalds */ 88881ed9475SDarrick J. Wong error = xfs_fs_reserve_ag_blocks(mp); 88981ed9475SDarrick J. Wong if (error && error == -ENOSPC) 89081ed9475SDarrick J. Wong xfs_warn(mp, 89181ed9475SDarrick J. Wong "ENOSPC reserving per-AG metadata pool, log recovery may fail."); 8924249023aSChristoph Hellwig error = xfs_log_mount_finish(mp); 89381ed9475SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 8941da177e4SLinus Torvalds if (error) { 8950b932cccSDave Chinner xfs_warn(mp, "log mount finish failed"); 896b93b6e43SChristoph Hellwig goto out_rtunmount; 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds 8991da177e4SLinus Torvalds /* 900ddeb14f4SDave Chinner * Now the log is fully replayed, we can transition to full read-only 901ddeb14f4SDave Chinner * mode for read-only mounts. This will sync all the metadata and clean 902ddeb14f4SDave Chinner * the log so that the recovery we just performed does not have to be 903ddeb14f4SDave Chinner * replayed again on the next mount. 904ddeb14f4SDave Chinner * 905ddeb14f4SDave Chinner * We use the same quiesce mechanism as the rw->ro remount, as they are 906ddeb14f4SDave Chinner * semantically identical operations. 907ddeb14f4SDave Chinner */ 9082e973b2cSDave Chinner if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp)) 909ea2064daSBrian Foster xfs_log_clean(mp); 910ddeb14f4SDave Chinner 911ddeb14f4SDave Chinner /* 9121da177e4SLinus Torvalds * Complete the quota initialisation, post-log-replay component. 9131da177e4SLinus Torvalds */ 9147d095257SChristoph Hellwig if (quotamount) { 9157d095257SChristoph Hellwig ASSERT(mp->m_qflags == 0); 9167d095257SChristoph Hellwig mp->m_qflags = quotaflags; 9177d095257SChristoph Hellwig 9187d095257SChristoph Hellwig xfs_qm_mount_quotas(mp); 9197d095257SChristoph Hellwig } 9207d095257SChristoph Hellwig 92184e1e99fSDavid Chinner /* 92284e1e99fSDavid Chinner * Now we are mounted, reserve a small amount of unused space for 92384e1e99fSDavid Chinner * privileged transactions. This is needed so that transaction 92484e1e99fSDavid Chinner * space required for critical operations can dip into this pool 92584e1e99fSDavid Chinner * when at ENOSPC. This is needed for operations like create with 92684e1e99fSDavid Chinner * attr, unwritten extent conversion at ENOSPC, etc. Data allocations 92784e1e99fSDavid Chinner * are not allowed to use this reserved space. 9288babd8a2SDave Chinner * 9298babd8a2SDave Chinner * This may drive us straight to ENOSPC on mount, but that implies 9308babd8a2SDave Chinner * we were already there on the last unmount. Warn if this occurs. 93184e1e99fSDavid Chinner */ 9322e973b2cSDave Chinner if (!xfs_is_readonly(mp)) { 933d5db0f97SEric Sandeen resblks = xfs_default_resblks(mp); 934714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 935714082bcSDavid Chinner if (error) 9360b932cccSDave Chinner xfs_warn(mp, 9370b932cccSDave Chinner "Unable to allocate reserve blocks. Continuing without reserve pool."); 938174edb0eSDarrick J. Wong 939174edb0eSDarrick J. Wong /* Recover any CoW blocks that never got remapped. */ 940174edb0eSDarrick J. Wong error = xfs_reflink_recover_cow(mp); 941174edb0eSDarrick J. Wong if (error) { 942174edb0eSDarrick J. Wong xfs_err(mp, 943174edb0eSDarrick J. Wong "Error %d recovering leftover CoW allocations.", error); 944174edb0eSDarrick J. Wong xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 945174edb0eSDarrick J. Wong goto out_quota; 946174edb0eSDarrick J. Wong } 94784d69619SDarrick J. Wong 94884d69619SDarrick J. Wong /* Reserve AG blocks for future btree expansion. */ 94984d69619SDarrick J. Wong error = xfs_fs_reserve_ag_blocks(mp); 95084d69619SDarrick J. Wong if (error && error != -ENOSPC) 95184d69619SDarrick J. Wong goto out_agresv; 952d5db0f97SEric Sandeen } 95384e1e99fSDavid Chinner 9541da177e4SLinus Torvalds return 0; 9551da177e4SLinus Torvalds 95684d69619SDarrick J. Wong out_agresv: 95784d69619SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 958174edb0eSDarrick J. Wong out_quota: 959174edb0eSDarrick J. Wong xfs_qm_unmount_quotas(mp); 960b93b6e43SChristoph Hellwig out_rtunmount: 961b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 962f9057e3dSChristoph Hellwig out_rele_rip: 96344a8736bSDarrick J. Wong xfs_irele(rip); 96477aff8c7SDarrick J. Wong /* Clean out dquots that might be in memory after quotacheck. */ 96577aff8c7SDarrick J. Wong xfs_qm_unmount(mp); 966ab23a776SDave Chinner 967ab23a776SDave Chinner /* 968ab23a776SDave Chinner * Inactivate all inodes that might still be in memory after a log 969ab23a776SDave Chinner * intent recovery failure so that reclaim can free them. Metadata 970ab23a776SDave Chinner * inodes and the root directory shouldn't need inactivation, but the 971ab23a776SDave Chinner * mount failed for some reason, so pull down all the state and flee. 972ab23a776SDave Chinner */ 973ab23a776SDave Chinner xfs_inodegc_flush(mp); 974ab23a776SDave Chinner 9752d1d1da3SDarrick J. Wong /* 976d336f7ebSDarrick J. Wong * Flush all inode reclamation work and flush the log. 9772d1d1da3SDarrick J. Wong * We have to do this /after/ rtunmount and qm_unmount because those 9782d1d1da3SDarrick J. Wong * two will have scheduled delayed reclaim for the rt/quota inodes. 9792d1d1da3SDarrick J. Wong * 9802d1d1da3SDarrick J. Wong * This is slightly different from the unmountfs call sequence 9812d1d1da3SDarrick J. Wong * because we could be tearing down a partially set up mount. In 9822d1d1da3SDarrick J. Wong * particular, if log_mount_finish fails we bail out without calling 9832d1d1da3SDarrick J. Wong * qm_unmount_quotas and therefore rely on qm_unmount to release the 9842d1d1da3SDarrick J. Wong * quota inodes. 9852d1d1da3SDarrick J. Wong */ 986d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes(mp); 987f9057e3dSChristoph Hellwig out_log_dealloc: 988f0b2efadSBrian Foster xfs_log_mount_cancel(mp); 98940b1de00SDarrick J. Wong out_inodegc_shrinker: 99040b1de00SDarrick J. Wong unregister_shrinker(&mp->m_inodegc_shrinker); 991d4f3512bSDave Chinner out_fail_wait: 992d4f3512bSDave Chinner if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) 99310fb9ac1SBrian Foster xfs_buftarg_drain(mp->m_logdev_targp); 99410fb9ac1SBrian Foster xfs_buftarg_drain(mp->m_ddev_targp); 995f9057e3dSChristoph Hellwig out_free_perag: 996ff4f038cSChristoph Hellwig xfs_free_perag(mp); 9970650b554SDave Chinner out_free_dir: 9980650b554SDave Chinner xfs_da_unmount(mp); 999f9057e3dSChristoph Hellwig out_remove_uuid: 100027174203SChristoph Hellwig xfs_uuid_unmount(mp); 100131965ef3SDarrick J. Wong out_remove_errortag: 100231965ef3SDarrick J. Wong xfs_errortag_del(mp); 1003192852beSCarlos Maiolino out_remove_error_sysfs: 1004192852beSCarlos Maiolino xfs_error_sysfs_del(mp); 1005225e4635SBill O'Donnell out_del_stats: 1006225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 1007a31b1d3dSBrian Foster out_remove_sysfs: 1008a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 1009f9057e3dSChristoph Hellwig out: 10101da177e4SLinus Torvalds return error; 10111da177e4SLinus Torvalds } 10121da177e4SLinus Torvalds 10131da177e4SLinus Torvalds /* 10141da177e4SLinus Torvalds * This flushes out the inodes,dquots and the superblock, unmounts the 10151da177e4SLinus Torvalds * log and makes sure that incore structures are freed. 10161da177e4SLinus Torvalds */ 101741b5c2e7SChristoph Hellwig void 101841b5c2e7SChristoph Hellwig xfs_unmountfs( 101941b5c2e7SChristoph Hellwig struct xfs_mount *mp) 10201da177e4SLinus Torvalds { 1021c8ce540dSDarrick J. Wong uint64_t resblks; 102241b5c2e7SChristoph Hellwig int error; 10231da177e4SLinus Torvalds 1024ab23a776SDave Chinner /* 1025ab23a776SDave Chinner * Perform all on-disk metadata updates required to inactivate inodes 1026ab23a776SDave Chinner * that the VFS evicted earlier in the unmount process. Freeing inodes 1027ab23a776SDave Chinner * and discarding CoW fork preallocations can cause shape changes to 1028ab23a776SDave Chinner * the free inode and refcount btrees, respectively, so we must finish 1029ab23a776SDave Chinner * this before we discard the metadata space reservations. Metadata 1030ab23a776SDave Chinner * inodes and the root directory do not require inactivation. 1031ab23a776SDave Chinner */ 1032ab23a776SDave Chinner xfs_inodegc_flush(mp); 1033ab23a776SDave Chinner 1034c9a6526fSDarrick J. Wong xfs_blockgc_stop(mp); 103584d69619SDarrick J. Wong xfs_fs_unreserve_ag_blocks(mp); 10367d095257SChristoph Hellwig xfs_qm_unmount_quotas(mp); 1037b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 103844a8736bSDarrick J. Wong xfs_irele(mp->m_rootip); 103977508ec8SChristoph Hellwig 1040d336f7ebSDarrick J. Wong xfs_unmount_flush_inodes(mp); 10411da177e4SLinus Torvalds 10427d095257SChristoph Hellwig xfs_qm_unmount(mp); 1043a357a121SLachlan McIlroy 10441da177e4SLinus Torvalds /* 104584e1e99fSDavid Chinner * Unreserve any blocks we have so that when we unmount we don't account 104684e1e99fSDavid Chinner * the reserved free space as used. This is really only necessary for 104784e1e99fSDavid Chinner * lazy superblock counting because it trusts the incore superblock 10489da096fdSMalcolm Parsons * counters to be absolutely correct on clean unmount. 104984e1e99fSDavid Chinner * 105084e1e99fSDavid Chinner * We don't bother correcting this elsewhere for lazy superblock 105184e1e99fSDavid Chinner * counting because on mount of an unclean filesystem we reconstruct the 105284e1e99fSDavid Chinner * correct counter value and this is irrelevant. 105384e1e99fSDavid Chinner * 105484e1e99fSDavid Chinner * For non-lazy counter filesystems, this doesn't matter at all because 105584e1e99fSDavid Chinner * we only every apply deltas to the superblock and hence the incore 105684e1e99fSDavid Chinner * value does not matter.... 105784e1e99fSDavid Chinner */ 105884e1e99fSDavid Chinner resblks = 0; 1059714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 1060714082bcSDavid Chinner if (error) 10610b932cccSDave Chinner xfs_warn(mp, "Unable to free reserved block pool. " 1062714082bcSDavid Chinner "Freespace may not be correct on next mount."); 1063714082bcSDavid Chinner 106421b699c8SChristoph Hellwig xfs_log_unmount(mp); 10650650b554SDave Chinner xfs_da_unmount(mp); 106627174203SChristoph Hellwig xfs_uuid_unmount(mp); 10671da177e4SLinus Torvalds 10681550d0b0SChristoph Hellwig #if defined(DEBUG) 106931965ef3SDarrick J. Wong xfs_errortag_clearall(mp); 10701da177e4SLinus Torvalds #endif 107140b1de00SDarrick J. Wong unregister_shrinker(&mp->m_inodegc_shrinker); 1072ff4f038cSChristoph Hellwig xfs_free_perag(mp); 1073a31b1d3dSBrian Foster 107431965ef3SDarrick J. Wong xfs_errortag_del(mp); 1075192852beSCarlos Maiolino xfs_error_sysfs_del(mp); 1076225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 1077a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds 108091ee575fSBrian Foster /* 108191ee575fSBrian Foster * Determine whether modifications can proceed. The caller specifies the minimum 108291ee575fSBrian Foster * freeze level for which modifications should not be allowed. This allows 108391ee575fSBrian Foster * certain operations to proceed while the freeze sequence is in progress, if 108491ee575fSBrian Foster * necessary. 108591ee575fSBrian Foster */ 108691ee575fSBrian Foster bool 108791ee575fSBrian Foster xfs_fs_writable( 108891ee575fSBrian Foster struct xfs_mount *mp, 108991ee575fSBrian Foster int level) 109092821e2bSDavid Chinner { 109191ee575fSBrian Foster ASSERT(level > SB_UNFROZEN); 109291ee575fSBrian Foster if ((mp->m_super->s_writers.frozen >= level) || 109375c8c50fSDave Chinner xfs_is_shutdown(mp) || xfs_is_readonly(mp)) 109491ee575fSBrian Foster return false; 109591ee575fSBrian Foster 109691ee575fSBrian Foster return true; 109792821e2bSDavid Chinner } 109892821e2bSDavid Chinner 10990d485adaSDave Chinner int 11000d485adaSDave Chinner xfs_mod_fdblocks( 11010d485adaSDave Chinner struct xfs_mount *mp, 11020d485adaSDave Chinner int64_t delta, 11030d485adaSDave Chinner bool rsvd) 11040d485adaSDave Chinner { 11050d485adaSDave Chinner int64_t lcounter; 11060d485adaSDave Chinner long long res_used; 11070d485adaSDave Chinner s32 batch; 1108fd43cf60SBrian Foster uint64_t set_aside; 11090d485adaSDave Chinner 11100d485adaSDave Chinner if (delta > 0) { 11110d485adaSDave Chinner /* 11120d485adaSDave Chinner * If the reserve pool is depleted, put blocks back into it 11130d485adaSDave Chinner * first. Most of the time the pool is full. 11140d485adaSDave Chinner */ 11150d485adaSDave Chinner if (likely(mp->m_resblks == mp->m_resblks_avail)) { 11160d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, delta); 11170d485adaSDave Chinner return 0; 11180d485adaSDave Chinner } 11190d485adaSDave Chinner 11200d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 11210d485adaSDave Chinner res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); 11220d485adaSDave Chinner 11230d485adaSDave Chinner if (res_used > delta) { 11240d485adaSDave Chinner mp->m_resblks_avail += delta; 11250d485adaSDave Chinner } else { 11260d485adaSDave Chinner delta -= res_used; 11270d485adaSDave Chinner mp->m_resblks_avail = mp->m_resblks; 11280d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, delta); 11290d485adaSDave Chinner } 11300d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 11310d485adaSDave Chinner return 0; 11320d485adaSDave Chinner } 11330d485adaSDave Chinner 11340d485adaSDave Chinner /* 11350d485adaSDave Chinner * Taking blocks away, need to be more accurate the closer we 11360d485adaSDave Chinner * are to zero. 11370d485adaSDave Chinner * 11380d485adaSDave Chinner * If the counter has a value of less than 2 * max batch size, 11390d485adaSDave Chinner * then make everything serialise as we are real close to 11400d485adaSDave Chinner * ENOSPC. 11410d485adaSDave Chinner */ 11428c1903d3SDave Chinner if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, 11438c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) < 0) 11440d485adaSDave Chinner batch = 1; 11450d485adaSDave Chinner else 11468c1903d3SDave Chinner batch = XFS_FDBLOCKS_BATCH; 11470d485adaSDave Chinner 1148fd43cf60SBrian Foster /* 1149fd43cf60SBrian Foster * Set aside allocbt blocks because these blocks are tracked as free 1150fd43cf60SBrian Foster * space but not available for allocation. Technically this means that a 1151fd43cf60SBrian Foster * single reservation cannot consume all remaining free space, but the 1152fd43cf60SBrian Foster * ratio of allocbt blocks to usable free blocks should be rather small. 1153fd43cf60SBrian Foster * The tradeoff without this is that filesystems that maintain high 1154fd43cf60SBrian Foster * perag block reservations can over reserve physical block availability 1155fd43cf60SBrian Foster * and fail physical allocation, which leads to much more serious 1156fd43cf60SBrian Foster * problems (i.e. transaction abort, pagecache discards, etc.) than 1157fd43cf60SBrian Foster * slightly premature -ENOSPC. 1158fd43cf60SBrian Foster */ 1159fd43cf60SBrian Foster set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks); 1160104b4e51SNikolay Borisov percpu_counter_add_batch(&mp->m_fdblocks, delta, batch); 1161fd43cf60SBrian Foster if (__percpu_counter_compare(&mp->m_fdblocks, set_aside, 11628c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) >= 0) { 11630d485adaSDave Chinner /* we had space! */ 11640d485adaSDave Chinner return 0; 11650d485adaSDave Chinner } 11660d485adaSDave Chinner 11670d485adaSDave Chinner /* 11680d485adaSDave Chinner * lock up the sb for dipping into reserves before releasing the space 11690d485adaSDave Chinner * that took us to ENOSPC. 11700d485adaSDave Chinner */ 11710d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 11720d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, -delta); 11730d485adaSDave Chinner if (!rsvd) 11740d485adaSDave Chinner goto fdblocks_enospc; 11750d485adaSDave Chinner 11760d485adaSDave Chinner lcounter = (long long)mp->m_resblks_avail + delta; 11770d485adaSDave Chinner if (lcounter >= 0) { 11780d485adaSDave Chinner mp->m_resblks_avail = lcounter; 11790d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 11800d485adaSDave Chinner return 0; 11810d485adaSDave Chinner } 1182ec43f6daSEric Sandeen xfs_warn_once(mp, 1183ec43f6daSEric Sandeen "Reserve blocks depleted! Consider increasing reserve pool size."); 1184ec43f6daSEric Sandeen 11850d485adaSDave Chinner fdblocks_enospc: 11860d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 11870d485adaSDave Chinner return -ENOSPC; 11880d485adaSDave Chinner } 11890d485adaSDave Chinner 1190bab98bbeSDave Chinner int 1191bab98bbeSDave Chinner xfs_mod_frextents( 1192bab98bbeSDave Chinner struct xfs_mount *mp, 1193bab98bbeSDave Chinner int64_t delta) 1194bab98bbeSDave Chinner { 1195bab98bbeSDave Chinner int64_t lcounter; 1196bab98bbeSDave Chinner int ret = 0; 1197bab98bbeSDave Chinner 1198bab98bbeSDave Chinner spin_lock(&mp->m_sb_lock); 1199bab98bbeSDave Chinner lcounter = mp->m_sb.sb_frextents + delta; 1200bab98bbeSDave Chinner if (lcounter < 0) 1201bab98bbeSDave Chinner ret = -ENOSPC; 1202bab98bbeSDave Chinner else 1203bab98bbeSDave Chinner mp->m_sb.sb_frextents = lcounter; 1204bab98bbeSDave Chinner spin_unlock(&mp->m_sb_lock); 1205bab98bbeSDave Chinner return ret; 1206bab98bbeSDave Chinner } 1207bab98bbeSDave Chinner 12081da177e4SLinus Torvalds /* 12091da177e4SLinus Torvalds * Used to free the superblock along various error paths. 12101da177e4SLinus Torvalds */ 12111da177e4SLinus Torvalds void 12121da177e4SLinus Torvalds xfs_freesb( 121326af6552SDave Chinner struct xfs_mount *mp) 12141da177e4SLinus Torvalds { 121526af6552SDave Chinner struct xfs_buf *bp = mp->m_sb_bp; 12161da177e4SLinus Torvalds 121726af6552SDave Chinner xfs_buf_lock(bp); 12181da177e4SLinus Torvalds mp->m_sb_bp = NULL; 121926af6552SDave Chinner xfs_buf_relse(bp); 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 1223dda35b8fSChristoph Hellwig * If the underlying (data/log/rt) device is readonly, there are some 1224dda35b8fSChristoph Hellwig * operations that cannot proceed. 1225dda35b8fSChristoph Hellwig */ 1226dda35b8fSChristoph Hellwig int 1227dda35b8fSChristoph Hellwig xfs_dev_is_read_only( 1228dda35b8fSChristoph Hellwig struct xfs_mount *mp, 1229dda35b8fSChristoph Hellwig char *message) 1230dda35b8fSChristoph Hellwig { 1231dda35b8fSChristoph Hellwig if (xfs_readonly_buftarg(mp->m_ddev_targp) || 1232dda35b8fSChristoph Hellwig xfs_readonly_buftarg(mp->m_logdev_targp) || 1233dda35b8fSChristoph Hellwig (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { 12340b932cccSDave Chinner xfs_notice(mp, "%s required on read-only device.", message); 12350b932cccSDave Chinner xfs_notice(mp, "write access unavailable, cannot proceed."); 12362451337dSDave Chinner return -EROFS; 1237dda35b8fSChristoph Hellwig } 1238dda35b8fSChristoph Hellwig return 0; 1239dda35b8fSChristoph Hellwig } 1240f467cad9SDarrick J. Wong 1241f467cad9SDarrick J. Wong /* Force the summary counters to be recalculated at next mount. */ 1242f467cad9SDarrick J. Wong void 1243f467cad9SDarrick J. Wong xfs_force_summary_recalc( 1244f467cad9SDarrick J. Wong struct xfs_mount *mp) 1245f467cad9SDarrick J. Wong { 124638c26bfdSDave Chinner if (!xfs_has_lazysbcount(mp)) 1247f467cad9SDarrick J. Wong return; 1248f467cad9SDarrick J. Wong 124939353ff6SDarrick J. Wong xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); 1250f467cad9SDarrick J. Wong } 12519fe82b8cSDarrick J. Wong 12529fe82b8cSDarrick J. Wong /* 1253908ce71eSDarrick J. Wong * Enable a log incompat feature flag in the primary superblock. The caller 1254908ce71eSDarrick J. Wong * cannot have any other transactions in progress. 1255908ce71eSDarrick J. Wong */ 1256908ce71eSDarrick J. Wong int 1257908ce71eSDarrick J. Wong xfs_add_incompat_log_feature( 1258908ce71eSDarrick J. Wong struct xfs_mount *mp, 1259908ce71eSDarrick J. Wong uint32_t feature) 1260908ce71eSDarrick J. Wong { 1261908ce71eSDarrick J. Wong struct xfs_dsb *dsb; 1262908ce71eSDarrick J. Wong int error; 1263908ce71eSDarrick J. Wong 1264908ce71eSDarrick J. Wong ASSERT(hweight32(feature) == 1); 1265908ce71eSDarrick J. Wong ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); 1266908ce71eSDarrick J. Wong 1267908ce71eSDarrick J. Wong /* 1268908ce71eSDarrick J. Wong * Force the log to disk and kick the background AIL thread to reduce 1269908ce71eSDarrick J. Wong * the chances that the bwrite will stall waiting for the AIL to unpin 1270908ce71eSDarrick J. Wong * the primary superblock buffer. This isn't a data integrity 1271908ce71eSDarrick J. Wong * operation, so we don't need a synchronous push. 1272908ce71eSDarrick J. Wong */ 1273908ce71eSDarrick J. Wong error = xfs_log_force(mp, XFS_LOG_SYNC); 1274908ce71eSDarrick J. Wong if (error) 1275908ce71eSDarrick J. Wong return error; 1276908ce71eSDarrick J. Wong xfs_ail_push_all(mp->m_ail); 1277908ce71eSDarrick J. Wong 1278908ce71eSDarrick J. Wong /* 1279908ce71eSDarrick J. Wong * Lock the primary superblock buffer to serialize all callers that 1280908ce71eSDarrick J. Wong * are trying to set feature bits. 1281908ce71eSDarrick J. Wong */ 1282908ce71eSDarrick J. Wong xfs_buf_lock(mp->m_sb_bp); 1283908ce71eSDarrick J. Wong xfs_buf_hold(mp->m_sb_bp); 1284908ce71eSDarrick J. Wong 128575c8c50fSDave Chinner if (xfs_is_shutdown(mp)) { 1286908ce71eSDarrick J. Wong error = -EIO; 1287908ce71eSDarrick J. Wong goto rele; 1288908ce71eSDarrick J. Wong } 1289908ce71eSDarrick J. Wong 1290908ce71eSDarrick J. Wong if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature)) 1291908ce71eSDarrick J. Wong goto rele; 1292908ce71eSDarrick J. Wong 1293908ce71eSDarrick J. Wong /* 1294908ce71eSDarrick J. Wong * Write the primary superblock to disk immediately, because we need 1295908ce71eSDarrick J. Wong * the log_incompat bit to be set in the primary super now to protect 1296908ce71eSDarrick J. Wong * the log items that we're going to commit later. 1297908ce71eSDarrick J. Wong */ 1298908ce71eSDarrick J. Wong dsb = mp->m_sb_bp->b_addr; 1299908ce71eSDarrick J. Wong xfs_sb_to_disk(dsb, &mp->m_sb); 1300908ce71eSDarrick J. Wong dsb->sb_features_log_incompat |= cpu_to_be32(feature); 1301908ce71eSDarrick J. Wong error = xfs_bwrite(mp->m_sb_bp); 1302908ce71eSDarrick J. Wong if (error) 1303908ce71eSDarrick J. Wong goto shutdown; 1304908ce71eSDarrick J. Wong 1305908ce71eSDarrick J. Wong /* 1306908ce71eSDarrick J. Wong * Add the feature bits to the incore superblock before we unlock the 1307908ce71eSDarrick J. Wong * buffer. 1308908ce71eSDarrick J. Wong */ 1309908ce71eSDarrick J. Wong xfs_sb_add_incompat_log_features(&mp->m_sb, feature); 1310908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1311908ce71eSDarrick J. Wong 1312908ce71eSDarrick J. Wong /* Log the superblock to disk. */ 1313908ce71eSDarrick J. Wong return xfs_sync_sb(mp, false); 1314908ce71eSDarrick J. Wong shutdown: 1315908ce71eSDarrick J. Wong xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1316908ce71eSDarrick J. Wong rele: 1317908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1318908ce71eSDarrick J. Wong return error; 1319908ce71eSDarrick J. Wong } 1320908ce71eSDarrick J. Wong 1321908ce71eSDarrick J. Wong /* 1322908ce71eSDarrick J. Wong * Clear all the log incompat flags from the superblock. 1323908ce71eSDarrick J. Wong * 1324908ce71eSDarrick J. Wong * The caller cannot be in a transaction, must ensure that the log does not 1325908ce71eSDarrick J. Wong * contain any log items protected by any log incompat bit, and must ensure 1326908ce71eSDarrick J. Wong * that there are no other threads that depend on the state of the log incompat 1327908ce71eSDarrick J. Wong * feature flags in the primary super. 1328908ce71eSDarrick J. Wong * 1329908ce71eSDarrick J. Wong * Returns true if the superblock is dirty. 1330908ce71eSDarrick J. Wong */ 1331908ce71eSDarrick J. Wong bool 1332908ce71eSDarrick J. Wong xfs_clear_incompat_log_features( 1333908ce71eSDarrick J. Wong struct xfs_mount *mp) 1334908ce71eSDarrick J. Wong { 1335908ce71eSDarrick J. Wong bool ret = false; 1336908ce71eSDarrick J. Wong 1337ebd9027dSDave Chinner if (!xfs_has_crc(mp) || 1338908ce71eSDarrick J. Wong !xfs_sb_has_incompat_log_feature(&mp->m_sb, 1339908ce71eSDarrick J. Wong XFS_SB_FEAT_INCOMPAT_LOG_ALL) || 134075c8c50fSDave Chinner xfs_is_shutdown(mp)) 1341908ce71eSDarrick J. Wong return false; 1342908ce71eSDarrick J. Wong 1343908ce71eSDarrick J. Wong /* 1344908ce71eSDarrick J. Wong * Update the incore superblock. We synchronize on the primary super 1345908ce71eSDarrick J. Wong * buffer lock to be consistent with the add function, though at least 1346908ce71eSDarrick J. Wong * in theory this shouldn't be necessary. 1347908ce71eSDarrick J. Wong */ 1348908ce71eSDarrick J. Wong xfs_buf_lock(mp->m_sb_bp); 1349908ce71eSDarrick J. Wong xfs_buf_hold(mp->m_sb_bp); 1350908ce71eSDarrick J. Wong 1351908ce71eSDarrick J. Wong if (xfs_sb_has_incompat_log_feature(&mp->m_sb, 1352908ce71eSDarrick J. Wong XFS_SB_FEAT_INCOMPAT_LOG_ALL)) { 1353908ce71eSDarrick J. Wong xfs_info(mp, "Clearing log incompat feature flags."); 1354908ce71eSDarrick J. Wong xfs_sb_remove_incompat_log_features(&mp->m_sb); 1355908ce71eSDarrick J. Wong ret = true; 1356908ce71eSDarrick J. Wong } 1357908ce71eSDarrick J. Wong 1358908ce71eSDarrick J. Wong xfs_buf_relse(mp->m_sb_bp); 1359908ce71eSDarrick J. Wong return ret; 1360908ce71eSDarrick J. Wong } 1361908ce71eSDarrick J. Wong 1362908ce71eSDarrick J. Wong /* 13639fe82b8cSDarrick J. Wong * Update the in-core delayed block counter. 13649fe82b8cSDarrick J. Wong * 13659fe82b8cSDarrick J. Wong * We prefer to update the counter without having to take a spinlock for every 13669fe82b8cSDarrick J. Wong * counter update (i.e. batching). Each change to delayed allocation 13679fe82b8cSDarrick J. Wong * reservations can change can easily exceed the default percpu counter 13689fe82b8cSDarrick J. Wong * batching, so we use a larger batch factor here. 13699fe82b8cSDarrick J. Wong * 13709fe82b8cSDarrick J. Wong * Note that we don't currently have any callers requiring fast summation 13719fe82b8cSDarrick J. Wong * (e.g. percpu_counter_read) so we can use a big batch value here. 13729fe82b8cSDarrick J. Wong */ 13739fe82b8cSDarrick J. Wong #define XFS_DELALLOC_BATCH (4096) 13749fe82b8cSDarrick J. Wong void 13759fe82b8cSDarrick J. Wong xfs_mod_delalloc( 13769fe82b8cSDarrick J. Wong struct xfs_mount *mp, 13779fe82b8cSDarrick J. Wong int64_t delta) 13789fe82b8cSDarrick J. Wong { 13799fe82b8cSDarrick J. Wong percpu_counter_add_batch(&mp->m_delalloc_blks, delta, 13809fe82b8cSDarrick J. Wong XFS_DELALLOC_BATCH); 13819fe82b8cSDarrick J. Wong } 1382