11da177e4SLinus Torvalds /* 27b718769SNathan Scott * Copyright (c) 2000-2005 Silicon Graphics, Inc. 37b718769SNathan Scott * All Rights Reserved. 41da177e4SLinus Torvalds * 57b718769SNathan Scott * This program is free software; you can redistribute it and/or 67b718769SNathan Scott * modify it under the terms of the GNU General Public License as 71da177e4SLinus Torvalds * published by the Free Software Foundation. 81da177e4SLinus Torvalds * 97b718769SNathan Scott * This program is distributed in the hope that it would be useful, 107b718769SNathan Scott * but WITHOUT ANY WARRANTY; without even the implied warranty of 117b718769SNathan Scott * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 127b718769SNathan Scott * GNU General Public License for more details. 131da177e4SLinus Torvalds * 147b718769SNathan Scott * You should have received a copy of the GNU General Public License 157b718769SNathan Scott * along with this program; if not, write the Free Software Foundation, 167b718769SNathan Scott * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 171da177e4SLinus Torvalds */ 181da177e4SLinus Torvalds #include "xfs.h" 19a844f451SNathan Scott #include "xfs_fs.h" 2070a9883cSDave Chinner #include "xfs_shared.h" 21239880efSDave Chinner #include "xfs_format.h" 22239880efSDave Chinner #include "xfs_log_format.h" 23239880efSDave Chinner #include "xfs_trans_resv.h" 24a844f451SNathan Scott #include "xfs_bit.h" 251da177e4SLinus Torvalds #include "xfs_sb.h" 261da177e4SLinus Torvalds #include "xfs_mount.h" 2757062787SDave Chinner #include "xfs_da_format.h" 289a2cc41cSDave Chinner #include "xfs_da_btree.h" 291da177e4SLinus Torvalds #include "xfs_inode.h" 30a4fbe6abSDave Chinner #include "xfs_dir2.h" 31a844f451SNathan Scott #include "xfs_ialloc.h" 321da177e4SLinus Torvalds #include "xfs_alloc.h" 331da177e4SLinus Torvalds #include "xfs_rtalloc.h" 341da177e4SLinus Torvalds #include "xfs_bmap.h" 35a4fbe6abSDave Chinner #include "xfs_trans.h" 36a4fbe6abSDave Chinner #include "xfs_trans_priv.h" 37a4fbe6abSDave Chinner #include "xfs_log.h" 381da177e4SLinus Torvalds #include "xfs_error.h" 391da177e4SLinus Torvalds #include "xfs_quota.h" 401da177e4SLinus Torvalds #include "xfs_fsops.h" 410b1b213fSChristoph Hellwig #include "xfs_trace.h" 426d8b79cfSDave Chinner #include "xfs_icache.h" 43a31b1d3dSBrian Foster #include "xfs_sysfs.h" 440b1b213fSChristoph Hellwig 451da177e4SLinus Torvalds 4627174203SChristoph Hellwig static DEFINE_MUTEX(xfs_uuid_table_mutex); 4727174203SChristoph Hellwig static int xfs_uuid_table_size; 4827174203SChristoph Hellwig static uuid_t *xfs_uuid_table; 4927174203SChristoph Hellwig 50af3b6382SDarrick J. Wong void 51af3b6382SDarrick J. Wong xfs_uuid_table_free(void) 52af3b6382SDarrick J. Wong { 53af3b6382SDarrick J. Wong if (xfs_uuid_table_size == 0) 54af3b6382SDarrick J. Wong return; 55af3b6382SDarrick J. Wong kmem_free(xfs_uuid_table); 56af3b6382SDarrick J. Wong xfs_uuid_table = NULL; 57af3b6382SDarrick J. Wong xfs_uuid_table_size = 0; 58af3b6382SDarrick J. Wong } 59af3b6382SDarrick J. Wong 6027174203SChristoph Hellwig /* 6127174203SChristoph Hellwig * See if the UUID is unique among mounted XFS filesystems. 6227174203SChristoph Hellwig * Mount fails if UUID is nil or a FS with the same UUID is already mounted. 6327174203SChristoph Hellwig */ 6427174203SChristoph Hellwig STATIC int 6527174203SChristoph Hellwig xfs_uuid_mount( 6627174203SChristoph Hellwig struct xfs_mount *mp) 6727174203SChristoph Hellwig { 6827174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 6927174203SChristoph Hellwig int hole, i; 7027174203SChristoph Hellwig 7127174203SChristoph Hellwig if (mp->m_flags & XFS_MOUNT_NOUUID) 7227174203SChristoph Hellwig return 0; 7327174203SChristoph Hellwig 7427174203SChristoph Hellwig if (uuid_is_nil(uuid)) { 750b932cccSDave Chinner xfs_warn(mp, "Filesystem has nil UUID - can't mount"); 762451337dSDave Chinner return -EINVAL; 7727174203SChristoph Hellwig } 7827174203SChristoph Hellwig 7927174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 8027174203SChristoph Hellwig for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { 8127174203SChristoph Hellwig if (uuid_is_nil(&xfs_uuid_table[i])) { 8227174203SChristoph Hellwig hole = i; 8327174203SChristoph Hellwig continue; 8427174203SChristoph Hellwig } 8527174203SChristoph Hellwig if (uuid_equal(uuid, &xfs_uuid_table[i])) 8627174203SChristoph Hellwig goto out_duplicate; 8727174203SChristoph Hellwig } 8827174203SChristoph Hellwig 8927174203SChristoph Hellwig if (hole < 0) { 9027174203SChristoph Hellwig xfs_uuid_table = kmem_realloc(xfs_uuid_table, 9127174203SChristoph Hellwig (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), 9227174203SChristoph Hellwig xfs_uuid_table_size * sizeof(*xfs_uuid_table), 9327174203SChristoph Hellwig KM_SLEEP); 9427174203SChristoph Hellwig hole = xfs_uuid_table_size++; 9527174203SChristoph Hellwig } 9627174203SChristoph Hellwig xfs_uuid_table[hole] = *uuid; 9727174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 9827174203SChristoph Hellwig 9927174203SChristoph Hellwig return 0; 10027174203SChristoph Hellwig 10127174203SChristoph Hellwig out_duplicate: 10227174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 103021000e5SMitsuo Hayasaka xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); 1042451337dSDave Chinner return -EINVAL; 10527174203SChristoph Hellwig } 10627174203SChristoph Hellwig 10727174203SChristoph Hellwig STATIC void 10827174203SChristoph Hellwig xfs_uuid_unmount( 10927174203SChristoph Hellwig struct xfs_mount *mp) 11027174203SChristoph Hellwig { 11127174203SChristoph Hellwig uuid_t *uuid = &mp->m_sb.sb_uuid; 11227174203SChristoph Hellwig int i; 11327174203SChristoph Hellwig 11427174203SChristoph Hellwig if (mp->m_flags & XFS_MOUNT_NOUUID) 11527174203SChristoph Hellwig return; 11627174203SChristoph Hellwig 11727174203SChristoph Hellwig mutex_lock(&xfs_uuid_table_mutex); 11827174203SChristoph Hellwig for (i = 0; i < xfs_uuid_table_size; i++) { 11927174203SChristoph Hellwig if (uuid_is_nil(&xfs_uuid_table[i])) 12027174203SChristoph Hellwig continue; 12127174203SChristoph Hellwig if (!uuid_equal(uuid, &xfs_uuid_table[i])) 12227174203SChristoph Hellwig continue; 12327174203SChristoph Hellwig memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); 12427174203SChristoph Hellwig break; 12527174203SChristoph Hellwig } 12627174203SChristoph Hellwig ASSERT(i < xfs_uuid_table_size); 12727174203SChristoph Hellwig mutex_unlock(&xfs_uuid_table_mutex); 12827174203SChristoph Hellwig } 12927174203SChristoph Hellwig 13027174203SChristoph Hellwig 131e176579eSDave Chinner STATIC void 132e176579eSDave Chinner __xfs_free_perag( 133e176579eSDave Chinner struct rcu_head *head) 134e176579eSDave Chinner { 135e176579eSDave Chinner struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); 136e176579eSDave Chinner 137e176579eSDave Chinner ASSERT(atomic_read(&pag->pag_ref) == 0); 138e176579eSDave Chinner kmem_free(pag); 139e176579eSDave Chinner } 140e176579eSDave Chinner 1410fa800fbSDave Chinner /* 142e176579eSDave Chinner * Free up the per-ag resources associated with the mount structure. 1431da177e4SLinus Torvalds */ 144c962fb79SChristoph Hellwig STATIC void 145ff4f038cSChristoph Hellwig xfs_free_perag( 146745f6919SChristoph Hellwig xfs_mount_t *mp) 1471da177e4SLinus Torvalds { 1481c1c6ebcSDave Chinner xfs_agnumber_t agno; 1491c1c6ebcSDave Chinner struct xfs_perag *pag; 1501da177e4SLinus Torvalds 1511c1c6ebcSDave Chinner for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { 1521c1c6ebcSDave Chinner spin_lock(&mp->m_perag_lock); 1531c1c6ebcSDave Chinner pag = radix_tree_delete(&mp->m_perag_tree, agno); 1541c1c6ebcSDave Chinner spin_unlock(&mp->m_perag_lock); 155e176579eSDave Chinner ASSERT(pag); 156f83282a8SDave Chinner ASSERT(atomic_read(&pag->pag_ref) == 0); 157e176579eSDave Chinner call_rcu(&pag->rcu_head, __xfs_free_perag); 1581da177e4SLinus Torvalds } 1591da177e4SLinus Torvalds } 1601da177e4SLinus Torvalds 1614cc929eeSNathan Scott /* 1624cc929eeSNathan Scott * Check size of device based on the (data/realtime) block count. 1634cc929eeSNathan Scott * Note: this check is used by the growfs code as well as mount. 1644cc929eeSNathan Scott */ 1654cc929eeSNathan Scott int 1664cc929eeSNathan Scott xfs_sb_validate_fsb_count( 1674cc929eeSNathan Scott xfs_sb_t *sbp, 1684cc929eeSNathan Scott __uint64_t nblocks) 1694cc929eeSNathan Scott { 1704cc929eeSNathan Scott ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); 1714cc929eeSNathan Scott ASSERT(sbp->sb_blocklog >= BBSHIFT); 1724cc929eeSNathan Scott 173d5cf09baSChristoph Hellwig /* Limited by ULONG_MAX of page cache index */ 1744cc929eeSNathan Scott if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) 1752451337dSDave Chinner return -EFBIG; 1764cc929eeSNathan Scott return 0; 1774cc929eeSNathan Scott } 1781da177e4SLinus Torvalds 1791c1c6ebcSDave Chinner int 180c11e2c36SNathan Scott xfs_initialize_perag( 181c11e2c36SNathan Scott xfs_mount_t *mp, 1821c1c6ebcSDave Chinner xfs_agnumber_t agcount, 1831c1c6ebcSDave Chinner xfs_agnumber_t *maxagi) 1841da177e4SLinus Torvalds { 1852d2194f6SCarlos Maiolino xfs_agnumber_t index; 1868b26c582SDave Chinner xfs_agnumber_t first_initialised = 0; 1871da177e4SLinus Torvalds xfs_perag_t *pag; 1881da177e4SLinus Torvalds xfs_agino_t agino; 1891da177e4SLinus Torvalds xfs_ino_t ino; 1901da177e4SLinus Torvalds xfs_sb_t *sbp = &mp->m_sb; 1918b26c582SDave Chinner int error = -ENOMEM; 1921da177e4SLinus Torvalds 1931c1c6ebcSDave Chinner /* 1941c1c6ebcSDave Chinner * Walk the current per-ag tree so we don't try to initialise AGs 1951c1c6ebcSDave Chinner * that already exist (growfs case). Allocate and insert all the 1961c1c6ebcSDave Chinner * AGs we don't find ready for initialisation. 1971c1c6ebcSDave Chinner */ 1981c1c6ebcSDave Chinner for (index = 0; index < agcount; index++) { 1991c1c6ebcSDave Chinner pag = xfs_perag_get(mp, index); 2001c1c6ebcSDave Chinner if (pag) { 2011c1c6ebcSDave Chinner xfs_perag_put(pag); 2021c1c6ebcSDave Chinner continue; 2031c1c6ebcSDave Chinner } 2048b26c582SDave Chinner if (!first_initialised) 2058b26c582SDave Chinner first_initialised = index; 206fb3b504aSChristoph Hellwig 2071c1c6ebcSDave Chinner pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); 2081c1c6ebcSDave Chinner if (!pag) 2098b26c582SDave Chinner goto out_unwind; 210fb3b504aSChristoph Hellwig pag->pag_agno = index; 211fb3b504aSChristoph Hellwig pag->pag_mount = mp; 2121a427ab0SDave Chinner spin_lock_init(&pag->pag_ici_lock); 21369b491c2SDave Chinner mutex_init(&pag->pag_ici_reclaim_lock); 214fb3b504aSChristoph Hellwig INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); 21574f75a0cSDave Chinner spin_lock_init(&pag->pag_buf_lock); 21674f75a0cSDave Chinner pag->pag_buf_tree = RB_ROOT; 217fb3b504aSChristoph Hellwig 2181c1c6ebcSDave Chinner if (radix_tree_preload(GFP_NOFS)) 2198b26c582SDave Chinner goto out_unwind; 220fb3b504aSChristoph Hellwig 2211c1c6ebcSDave Chinner spin_lock(&mp->m_perag_lock); 2221c1c6ebcSDave Chinner if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { 2231c1c6ebcSDave Chinner BUG(); 2241c1c6ebcSDave Chinner spin_unlock(&mp->m_perag_lock); 2258b26c582SDave Chinner radix_tree_preload_end(); 2268b26c582SDave Chinner error = -EEXIST; 2278b26c582SDave Chinner goto out_unwind; 2281c1c6ebcSDave Chinner } 2291c1c6ebcSDave Chinner spin_unlock(&mp->m_perag_lock); 2301c1c6ebcSDave Chinner radix_tree_preload_end(); 2311c1c6ebcSDave Chinner } 2321c1c6ebcSDave Chinner 233fb3b504aSChristoph Hellwig /* 234fb3b504aSChristoph Hellwig * If we mount with the inode64 option, or no inode overflows 235fb3b504aSChristoph Hellwig * the legacy 32-bit address space clear the inode32 option. 2361da177e4SLinus Torvalds */ 237fb3b504aSChristoph Hellwig agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); 238fb3b504aSChristoph Hellwig ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); 2391da177e4SLinus Torvalds 240fb3b504aSChristoph Hellwig if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) 241fb3b504aSChristoph Hellwig mp->m_flags |= XFS_MOUNT_32BITINODES; 242fb3b504aSChristoph Hellwig else 243fb3b504aSChristoph Hellwig mp->m_flags &= ~XFS_MOUNT_32BITINODES; 244fb3b504aSChristoph Hellwig 2452d2194f6SCarlos Maiolino if (mp->m_flags & XFS_MOUNT_32BITINODES) 2469de67c3bSEric Sandeen index = xfs_set_inode32(mp, agcount); 2472d2194f6SCarlos Maiolino else 2489de67c3bSEric Sandeen index = xfs_set_inode64(mp, agcount); 249fb3b504aSChristoph Hellwig 2501c1c6ebcSDave Chinner if (maxagi) 2511c1c6ebcSDave Chinner *maxagi = index; 2521c1c6ebcSDave Chinner return 0; 2538b26c582SDave Chinner 2548b26c582SDave Chinner out_unwind: 2558b26c582SDave Chinner kmem_free(pag); 2568b26c582SDave Chinner for (; index > first_initialised; index--) { 2578b26c582SDave Chinner pag = radix_tree_delete(&mp->m_perag_tree, index); 2588b26c582SDave Chinner kmem_free(pag); 2598b26c582SDave Chinner } 2608b26c582SDave Chinner return error; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds /* 2641da177e4SLinus Torvalds * xfs_readsb 2651da177e4SLinus Torvalds * 2661da177e4SLinus Torvalds * Does the initial read of the superblock. 2671da177e4SLinus Torvalds */ 2681da177e4SLinus Torvalds int 269ff55068cSDave Chinner xfs_readsb( 270ff55068cSDave Chinner struct xfs_mount *mp, 271ff55068cSDave Chinner int flags) 2721da177e4SLinus Torvalds { 2731da177e4SLinus Torvalds unsigned int sector_size; 27404a1e6c5SDave Chinner struct xfs_buf *bp; 27504a1e6c5SDave Chinner struct xfs_sb *sbp = &mp->m_sb; 2761da177e4SLinus Torvalds int error; 277af34e09dSDave Chinner int loud = !(flags & XFS_MFSI_QUIET); 278daba5427SEric Sandeen const struct xfs_buf_ops *buf_ops; 2791da177e4SLinus Torvalds 2801da177e4SLinus Torvalds ASSERT(mp->m_sb_bp == NULL); 2811da177e4SLinus Torvalds ASSERT(mp->m_ddev_targp != NULL); 2821da177e4SLinus Torvalds 2831da177e4SLinus Torvalds /* 284daba5427SEric Sandeen * For the initial read, we must guess at the sector 285daba5427SEric Sandeen * size based on the block device. It's enough to 286daba5427SEric Sandeen * get the sb_sectsize out of the superblock and 287daba5427SEric Sandeen * then reread with the proper length. 288daba5427SEric Sandeen * We don't verify it yet, because it may not be complete. 289daba5427SEric Sandeen */ 290daba5427SEric Sandeen sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); 291daba5427SEric Sandeen buf_ops = NULL; 292daba5427SEric Sandeen 293daba5427SEric Sandeen /* 2941da177e4SLinus Torvalds * Allocate a (locked) buffer to hold the superblock. 2951da177e4SLinus Torvalds * This will be kept around at all times to optimize 2961da177e4SLinus Torvalds * access to the superblock. 2971da177e4SLinus Torvalds */ 29826af6552SDave Chinner reread: 299ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, 300ba372674SDave Chinner BTOBB(sector_size), 0, &bp, buf_ops); 301ba372674SDave Chinner if (error) { 302eab4e633SDave Chinner if (loud) 303e721f504SDave Chinner xfs_warn(mp, "SB validate failed with error %d.", error); 304ac75a1f7SDave Chinner /* bad CRC means corrupted metadata */ 3052451337dSDave Chinner if (error == -EFSBADCRC) 3062451337dSDave Chinner error = -EFSCORRUPTED; 307ba372674SDave Chinner return error; 308eab4e633SDave Chinner } 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds /* 3111da177e4SLinus Torvalds * Initialize the mount structure from the superblock. 3121da177e4SLinus Torvalds */ 313556b8883SDave Chinner xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); 314556b8883SDave Chinner 315556b8883SDave Chinner /* 316556b8883SDave Chinner * If we haven't validated the superblock, do so now before we try 317556b8883SDave Chinner * to check the sector size and reread the superblock appropriately. 318556b8883SDave Chinner */ 319556b8883SDave Chinner if (sbp->sb_magicnum != XFS_SB_MAGIC) { 320556b8883SDave Chinner if (loud) 321556b8883SDave Chinner xfs_warn(mp, "Invalid superblock magic number"); 3222451337dSDave Chinner error = -EINVAL; 323556b8883SDave Chinner goto release_buf; 324556b8883SDave Chinner } 325ff55068cSDave Chinner 3261da177e4SLinus Torvalds /* 3271da177e4SLinus Torvalds * We must be able to do sector-sized and sector-aligned IO. 3281da177e4SLinus Torvalds */ 32904a1e6c5SDave Chinner if (sector_size > sbp->sb_sectsize) { 330af34e09dSDave Chinner if (loud) 331af34e09dSDave Chinner xfs_warn(mp, "device supports %u byte sectors (not %u)", 33204a1e6c5SDave Chinner sector_size, sbp->sb_sectsize); 3332451337dSDave Chinner error = -ENOSYS; 33426af6552SDave Chinner goto release_buf; 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds 337556b8883SDave Chinner if (buf_ops == NULL) { 3381da177e4SLinus Torvalds /* 339daba5427SEric Sandeen * Re-read the superblock so the buffer is correctly sized, 340daba5427SEric Sandeen * and properly verified. 3411da177e4SLinus Torvalds */ 3421da177e4SLinus Torvalds xfs_buf_relse(bp); 34304a1e6c5SDave Chinner sector_size = sbp->sb_sectsize; 344daba5427SEric Sandeen buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; 34526af6552SDave Chinner goto reread; 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds 3485681ca40SDave Chinner xfs_reinit_percpu_counters(mp); 3498d280b98SDavid Chinner 35004a1e6c5SDave Chinner /* no need to be quiet anymore, so reset the buf ops */ 35104a1e6c5SDave Chinner bp->b_ops = &xfs_sb_buf_ops; 35204a1e6c5SDave Chinner 3531da177e4SLinus Torvalds mp->m_sb_bp = bp; 35426af6552SDave Chinner xfs_buf_unlock(bp); 3551da177e4SLinus Torvalds return 0; 3561da177e4SLinus Torvalds 35726af6552SDave Chinner release_buf: 3581da177e4SLinus Torvalds xfs_buf_relse(bp); 3591da177e4SLinus Torvalds return error; 3601da177e4SLinus Torvalds } 3611da177e4SLinus Torvalds 3621da177e4SLinus Torvalds /* 3630771fb45SEric Sandeen * Update alignment values based on mount options and sb values 3641da177e4SLinus Torvalds */ 3650771fb45SEric Sandeen STATIC int 3667884bc86SChristoph Hellwig xfs_update_alignment(xfs_mount_t *mp) 3671da177e4SLinus Torvalds { 3681da177e4SLinus Torvalds xfs_sb_t *sbp = &(mp->m_sb); 3691da177e4SLinus Torvalds 3704249023aSChristoph Hellwig if (mp->m_dalign) { 3711da177e4SLinus Torvalds /* 3721da177e4SLinus Torvalds * If stripe unit and stripe width are not multiples 3731da177e4SLinus Torvalds * of the fs blocksize turn off alignment. 3741da177e4SLinus Torvalds */ 3751da177e4SLinus Torvalds if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || 3761da177e4SLinus Torvalds (BBTOB(mp->m_swidth) & mp->m_blockmask)) { 37739a45d84SJie Liu xfs_warn(mp, 37839a45d84SJie Liu "alignment check failed: sunit/swidth vs. blocksize(%d)", 37939a45d84SJie Liu sbp->sb_blocksize); 3802451337dSDave Chinner return -EINVAL; 3811da177e4SLinus Torvalds } else { 3821da177e4SLinus Torvalds /* 3831da177e4SLinus Torvalds * Convert the stripe unit and width to FSBs. 3841da177e4SLinus Torvalds */ 3851da177e4SLinus Torvalds mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); 3861da177e4SLinus Torvalds if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { 38753487786SDave Chinner xfs_warn(mp, 38839a45d84SJie Liu "alignment check failed: sunit/swidth vs. agsize(%d)", 3891da177e4SLinus Torvalds sbp->sb_agblocks); 3902451337dSDave Chinner return -EINVAL; 3911da177e4SLinus Torvalds } else if (mp->m_dalign) { 3921da177e4SLinus Torvalds mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); 3931da177e4SLinus Torvalds } else { 39439a45d84SJie Liu xfs_warn(mp, 39539a45d84SJie Liu "alignment check failed: sunit(%d) less than bsize(%d)", 39639a45d84SJie Liu mp->m_dalign, sbp->sb_blocksize); 3972451337dSDave Chinner return -EINVAL; 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds /* 4021da177e4SLinus Torvalds * Update superblock with new values 4031da177e4SLinus Torvalds * and log changes 4041da177e4SLinus Torvalds */ 40562118709SEric Sandeen if (xfs_sb_version_hasdalign(sbp)) { 4061da177e4SLinus Torvalds if (sbp->sb_unit != mp->m_dalign) { 4071da177e4SLinus Torvalds sbp->sb_unit = mp->m_dalign; 40861e63ecbSDave Chinner mp->m_update_sb = true; 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds if (sbp->sb_width != mp->m_swidth) { 4111da177e4SLinus Torvalds sbp->sb_width = mp->m_swidth; 41261e63ecbSDave Chinner mp->m_update_sb = true; 4131da177e4SLinus Torvalds } 41434d7f603SJie Liu } else { 41534d7f603SJie Liu xfs_warn(mp, 41634d7f603SJie Liu "cannot change alignment: superblock does not support data alignment"); 4172451337dSDave Chinner return -EINVAL; 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && 42062118709SEric Sandeen xfs_sb_version_hasdalign(&mp->m_sb)) { 4211da177e4SLinus Torvalds mp->m_dalign = sbp->sb_unit; 4221da177e4SLinus Torvalds mp->m_swidth = sbp->sb_width; 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds 4250771fb45SEric Sandeen return 0; 4260771fb45SEric Sandeen } 4271da177e4SLinus Torvalds 4280771fb45SEric Sandeen /* 4290771fb45SEric Sandeen * Set the maximum inode count for this filesystem 4300771fb45SEric Sandeen */ 4310771fb45SEric Sandeen STATIC void 4320771fb45SEric Sandeen xfs_set_maxicount(xfs_mount_t *mp) 4330771fb45SEric Sandeen { 4340771fb45SEric Sandeen xfs_sb_t *sbp = &(mp->m_sb); 4351da177e4SLinus Torvalds __uint64_t icount; 4361da177e4SLinus Torvalds 4370771fb45SEric Sandeen if (sbp->sb_imax_pct) { 4380771fb45SEric Sandeen /* 4390771fb45SEric Sandeen * Make sure the maximum inode count is a multiple 4400771fb45SEric Sandeen * of the units we allocate inodes in. 4411da177e4SLinus Torvalds */ 4421da177e4SLinus Torvalds icount = sbp->sb_dblocks * sbp->sb_imax_pct; 4431da177e4SLinus Torvalds do_div(icount, 100); 4441da177e4SLinus Torvalds do_div(icount, mp->m_ialloc_blks); 4451da177e4SLinus Torvalds mp->m_maxicount = (icount * mp->m_ialloc_blks) << 4461da177e4SLinus Torvalds sbp->sb_inopblog; 4470771fb45SEric Sandeen } else { 4481da177e4SLinus Torvalds mp->m_maxicount = 0; 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds } 4511da177e4SLinus Torvalds 4521da177e4SLinus Torvalds /* 4531da177e4SLinus Torvalds * Set the default minimum read and write sizes unless 4541da177e4SLinus Torvalds * already specified in a mount option. 4551da177e4SLinus Torvalds * We use smaller I/O sizes when the file system 4561da177e4SLinus Torvalds * is being used for NFS service (wsync mount option). 4571da177e4SLinus Torvalds */ 4580771fb45SEric Sandeen STATIC void 4590771fb45SEric Sandeen xfs_set_rw_sizes(xfs_mount_t *mp) 4600771fb45SEric Sandeen { 4610771fb45SEric Sandeen xfs_sb_t *sbp = &(mp->m_sb); 4620771fb45SEric Sandeen int readio_log, writeio_log; 4630771fb45SEric Sandeen 4641da177e4SLinus Torvalds if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { 4651da177e4SLinus Torvalds if (mp->m_flags & XFS_MOUNT_WSYNC) { 4661da177e4SLinus Torvalds readio_log = XFS_WSYNC_READIO_LOG; 4671da177e4SLinus Torvalds writeio_log = XFS_WSYNC_WRITEIO_LOG; 4681da177e4SLinus Torvalds } else { 4691da177e4SLinus Torvalds readio_log = XFS_READIO_LOG_LARGE; 4701da177e4SLinus Torvalds writeio_log = XFS_WRITEIO_LOG_LARGE; 4711da177e4SLinus Torvalds } 4721da177e4SLinus Torvalds } else { 4731da177e4SLinus Torvalds readio_log = mp->m_readio_log; 4741da177e4SLinus Torvalds writeio_log = mp->m_writeio_log; 4751da177e4SLinus Torvalds } 4761da177e4SLinus Torvalds 4771da177e4SLinus Torvalds if (sbp->sb_blocklog > readio_log) { 4781da177e4SLinus Torvalds mp->m_readio_log = sbp->sb_blocklog; 4791da177e4SLinus Torvalds } else { 4801da177e4SLinus Torvalds mp->m_readio_log = readio_log; 4811da177e4SLinus Torvalds } 4821da177e4SLinus Torvalds mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); 4831da177e4SLinus Torvalds if (sbp->sb_blocklog > writeio_log) { 4841da177e4SLinus Torvalds mp->m_writeio_log = sbp->sb_blocklog; 4851da177e4SLinus Torvalds } else { 4861da177e4SLinus Torvalds mp->m_writeio_log = writeio_log; 4871da177e4SLinus Torvalds } 4881da177e4SLinus Torvalds mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); 4890771fb45SEric Sandeen } 490425f9dddSEric Sandeen 4911da177e4SLinus Torvalds /* 492055388a3SDave Chinner * precalculate the low space thresholds for dynamic speculative preallocation. 493055388a3SDave Chinner */ 494055388a3SDave Chinner void 495055388a3SDave Chinner xfs_set_low_space_thresholds( 496055388a3SDave Chinner struct xfs_mount *mp) 497055388a3SDave Chinner { 498055388a3SDave Chinner int i; 499055388a3SDave Chinner 500055388a3SDave Chinner for (i = 0; i < XFS_LOWSP_MAX; i++) { 501055388a3SDave Chinner __uint64_t space = mp->m_sb.sb_dblocks; 502055388a3SDave Chinner 503055388a3SDave Chinner do_div(space, 100); 504055388a3SDave Chinner mp->m_low_space[i] = space * (i + 1); 505055388a3SDave Chinner } 506055388a3SDave Chinner } 507055388a3SDave Chinner 508055388a3SDave Chinner 509055388a3SDave Chinner /* 5101da177e4SLinus Torvalds * Set whether we're using inode alignment. 5111da177e4SLinus Torvalds */ 5120771fb45SEric Sandeen STATIC void 5130771fb45SEric Sandeen xfs_set_inoalignment(xfs_mount_t *mp) 5140771fb45SEric Sandeen { 51562118709SEric Sandeen if (xfs_sb_version_hasalign(&mp->m_sb) && 5161da177e4SLinus Torvalds mp->m_sb.sb_inoalignmt >= 5171da177e4SLinus Torvalds XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) 5181da177e4SLinus Torvalds mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; 5191da177e4SLinus Torvalds else 5201da177e4SLinus Torvalds mp->m_inoalign_mask = 0; 5211da177e4SLinus Torvalds /* 5221da177e4SLinus Torvalds * If we are using stripe alignment, check whether 5231da177e4SLinus Torvalds * the stripe unit is a multiple of the inode alignment 5241da177e4SLinus Torvalds */ 5251da177e4SLinus Torvalds if (mp->m_dalign && mp->m_inoalign_mask && 5261da177e4SLinus Torvalds !(mp->m_dalign & mp->m_inoalign_mask)) 5271da177e4SLinus Torvalds mp->m_sinoalign = mp->m_dalign; 5281da177e4SLinus Torvalds else 5291da177e4SLinus Torvalds mp->m_sinoalign = 0; 5300771fb45SEric Sandeen } 5310771fb45SEric Sandeen 5321da177e4SLinus Torvalds /* 5330471f62eSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 5341da177e4SLinus Torvalds */ 5350771fb45SEric Sandeen STATIC int 536ba372674SDave Chinner xfs_check_sizes( 537ba372674SDave Chinner struct xfs_mount *mp) 5380771fb45SEric Sandeen { 539ba372674SDave Chinner struct xfs_buf *bp; 5400771fb45SEric Sandeen xfs_daddr_t d; 541ba372674SDave Chinner int error; 5420771fb45SEric Sandeen 5431da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); 5441da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { 5450b932cccSDave Chinner xfs_warn(mp, "filesystem size mismatch detected"); 5462451337dSDave Chinner return -EFBIG; 5471da177e4SLinus Torvalds } 548ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_ddev_targp, 5491da177e4SLinus Torvalds d - XFS_FSS_TO_BB(mp, 1), 550ba372674SDave Chinner XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 551ba372674SDave Chinner if (error) { 5520b932cccSDave Chinner xfs_warn(mp, "last sector read failed"); 553ba372674SDave Chinner return error; 5541da177e4SLinus Torvalds } 5551922c949SDave Chinner xfs_buf_relse(bp); 5561da177e4SLinus Torvalds 557ba372674SDave Chinner if (mp->m_logdev_targp == mp->m_ddev_targp) 558ba372674SDave Chinner return 0; 559ba372674SDave Chinner 5601da177e4SLinus Torvalds d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); 5611da177e4SLinus Torvalds if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { 5620b932cccSDave Chinner xfs_warn(mp, "log size mismatch detected"); 5632451337dSDave Chinner return -EFBIG; 5641da177e4SLinus Torvalds } 565ba372674SDave Chinner error = xfs_buf_read_uncached(mp->m_logdev_targp, 5661da177e4SLinus Torvalds d - XFS_FSB_TO_BB(mp, 1), 567ba372674SDave Chinner XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); 568ba372674SDave Chinner if (error) { 5690b932cccSDave Chinner xfs_warn(mp, "log device read failed"); 570ba372674SDave Chinner return error; 5711da177e4SLinus Torvalds } 5721922c949SDave Chinner xfs_buf_relse(bp); 5730771fb45SEric Sandeen return 0; 5740771fb45SEric Sandeen } 5750771fb45SEric Sandeen 5760771fb45SEric Sandeen /* 5777d095257SChristoph Hellwig * Clear the quotaflags in memory and in the superblock. 5787d095257SChristoph Hellwig */ 5797d095257SChristoph Hellwig int 5807d095257SChristoph Hellwig xfs_mount_reset_sbqflags( 5817d095257SChristoph Hellwig struct xfs_mount *mp) 5827d095257SChristoph Hellwig { 5837d095257SChristoph Hellwig mp->m_qflags = 0; 5847d095257SChristoph Hellwig 58561e63ecbSDave Chinner /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ 5867d095257SChristoph Hellwig if (mp->m_sb.sb_qflags == 0) 5877d095257SChristoph Hellwig return 0; 5887d095257SChristoph Hellwig spin_lock(&mp->m_sb_lock); 5897d095257SChristoph Hellwig mp->m_sb.sb_qflags = 0; 5907d095257SChristoph Hellwig spin_unlock(&mp->m_sb_lock); 5917d095257SChristoph Hellwig 59261e63ecbSDave Chinner if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) 5937d095257SChristoph Hellwig return 0; 5947d095257SChristoph Hellwig 59561e63ecbSDave Chinner return xfs_sync_sb(mp, false); 5967d095257SChristoph Hellwig } 5977d095257SChristoph Hellwig 598d5db0f97SEric Sandeen __uint64_t 599d5db0f97SEric Sandeen xfs_default_resblks(xfs_mount_t *mp) 600d5db0f97SEric Sandeen { 601d5db0f97SEric Sandeen __uint64_t resblks; 602d5db0f97SEric Sandeen 603d5db0f97SEric Sandeen /* 6048babd8a2SDave Chinner * We default to 5% or 8192 fsbs of space reserved, whichever is 6058babd8a2SDave Chinner * smaller. This is intended to cover concurrent allocation 6068babd8a2SDave Chinner * transactions when we initially hit enospc. These each require a 4 6078babd8a2SDave Chinner * block reservation. Hence by default we cover roughly 2000 concurrent 6088babd8a2SDave Chinner * allocation reservations. 609d5db0f97SEric Sandeen */ 610d5db0f97SEric Sandeen resblks = mp->m_sb.sb_dblocks; 611d5db0f97SEric Sandeen do_div(resblks, 20); 6128babd8a2SDave Chinner resblks = min_t(__uint64_t, resblks, 8192); 613d5db0f97SEric Sandeen return resblks; 614d5db0f97SEric Sandeen } 615d5db0f97SEric Sandeen 6167d095257SChristoph Hellwig /* 6170771fb45SEric Sandeen * This function does the following on an initial mount of a file system: 6180771fb45SEric Sandeen * - reads the superblock from disk and init the mount struct 6190771fb45SEric Sandeen * - if we're a 32-bit kernel, do a size check on the superblock 6200771fb45SEric Sandeen * so we don't mount terabyte filesystems 6210771fb45SEric Sandeen * - init mount struct realtime fields 6220771fb45SEric Sandeen * - allocate inode hash table for fs 6230771fb45SEric Sandeen * - init directory manager 6240771fb45SEric Sandeen * - perform recovery and init the log manager 6250771fb45SEric Sandeen */ 6260771fb45SEric Sandeen int 6270771fb45SEric Sandeen xfs_mountfs( 628f0b2efadSBrian Foster struct xfs_mount *mp) 6290771fb45SEric Sandeen { 630f0b2efadSBrian Foster struct xfs_sb *sbp = &(mp->m_sb); 631f0b2efadSBrian Foster struct xfs_inode *rip; 6320771fb45SEric Sandeen __uint64_t resblks; 6337d095257SChristoph Hellwig uint quotamount = 0; 6347d095257SChristoph Hellwig uint quotaflags = 0; 6350771fb45SEric Sandeen int error = 0; 6360771fb45SEric Sandeen 637ff55068cSDave Chinner xfs_sb_mount_common(mp, sbp); 6380771fb45SEric Sandeen 6390771fb45SEric Sandeen /* 640074e427bSDave Chinner * Check for a mismatched features2 values. Older kernels read & wrote 641074e427bSDave Chinner * into the wrong sb offset for sb_features2 on some platforms due to 642074e427bSDave Chinner * xfs_sb_t not being 64bit size aligned when sb_features2 was added, 643074e427bSDave Chinner * which made older superblock reading/writing routines swap it as a 644074e427bSDave Chinner * 64-bit value. 645ee1c0908SDavid Chinner * 646e6957ea4SEric Sandeen * For backwards compatibility, we make both slots equal. 647e6957ea4SEric Sandeen * 648074e427bSDave Chinner * If we detect a mismatched field, we OR the set bits into the existing 649074e427bSDave Chinner * features2 field in case it has already been modified; we don't want 650074e427bSDave Chinner * to lose any features. We then update the bad location with the ORed 651074e427bSDave Chinner * value so that older kernels will see any features2 flags. The 652074e427bSDave Chinner * superblock writeback code ensures the new sb_features2 is copied to 653074e427bSDave Chinner * sb_bad_features2 before it is logged or written to disk. 654ee1c0908SDavid Chinner */ 655e6957ea4SEric Sandeen if (xfs_sb_has_mismatched_features2(sbp)) { 6560b932cccSDave Chinner xfs_warn(mp, "correcting sb_features alignment problem"); 657ee1c0908SDavid Chinner sbp->sb_features2 |= sbp->sb_bad_features2; 65861e63ecbSDave Chinner mp->m_update_sb = true; 659e6957ea4SEric Sandeen 660e6957ea4SEric Sandeen /* 661e6957ea4SEric Sandeen * Re-check for ATTR2 in case it was found in bad_features2 662e6957ea4SEric Sandeen * slot. 663e6957ea4SEric Sandeen */ 6647c12f296STim Shimmin if (xfs_sb_version_hasattr2(&mp->m_sb) && 6657c12f296STim Shimmin !(mp->m_flags & XFS_MOUNT_NOATTR2)) 666e6957ea4SEric Sandeen mp->m_flags |= XFS_MOUNT_ATTR2; 6677c12f296STim Shimmin } 668e6957ea4SEric Sandeen 6697c12f296STim Shimmin if (xfs_sb_version_hasattr2(&mp->m_sb) && 6707c12f296STim Shimmin (mp->m_flags & XFS_MOUNT_NOATTR2)) { 6717c12f296STim Shimmin xfs_sb_version_removeattr2(&mp->m_sb); 67261e63ecbSDave Chinner mp->m_update_sb = true; 6737c12f296STim Shimmin 6747c12f296STim Shimmin /* update sb_versionnum for the clearing of the morebits */ 6757c12f296STim Shimmin if (!sbp->sb_features2) 67661e63ecbSDave Chinner mp->m_update_sb = true; 677ee1c0908SDavid Chinner } 678ee1c0908SDavid Chinner 679263997a6SDave Chinner /* always use v2 inodes by default now */ 680263997a6SDave Chinner if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { 681263997a6SDave Chinner mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; 68261e63ecbSDave Chinner mp->m_update_sb = true; 683263997a6SDave Chinner } 684263997a6SDave Chinner 685ee1c0908SDavid Chinner /* 6860771fb45SEric Sandeen * Check if sb_agblocks is aligned at stripe boundary 6870771fb45SEric Sandeen * If sb_agblocks is NOT aligned turn off m_dalign since 6880771fb45SEric Sandeen * allocator alignment is within an ag, therefore ag has 6890771fb45SEric Sandeen * to be aligned at stripe boundary. 6900771fb45SEric Sandeen */ 6917884bc86SChristoph Hellwig error = xfs_update_alignment(mp); 6920771fb45SEric Sandeen if (error) 693f9057e3dSChristoph Hellwig goto out; 6940771fb45SEric Sandeen 6950771fb45SEric Sandeen xfs_alloc_compute_maxlevels(mp); 6960771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); 6970771fb45SEric Sandeen xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); 6980771fb45SEric Sandeen xfs_ialloc_compute_maxlevels(mp); 6990771fb45SEric Sandeen 7000771fb45SEric Sandeen xfs_set_maxicount(mp); 7010771fb45SEric Sandeen 702a31b1d3dSBrian Foster error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); 70327174203SChristoph Hellwig if (error) 704f9057e3dSChristoph Hellwig goto out; 7051da177e4SLinus Torvalds 706225e4635SBill O'Donnell error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, 707225e4635SBill O'Donnell &mp->m_kobj, "stats"); 708a31b1d3dSBrian Foster if (error) 709a31b1d3dSBrian Foster goto out_remove_sysfs; 710a31b1d3dSBrian Foster 711225e4635SBill O'Donnell error = xfs_uuid_mount(mp); 712225e4635SBill O'Donnell if (error) 713225e4635SBill O'Donnell goto out_del_stats; 714225e4635SBill O'Donnell 7151da177e4SLinus Torvalds /* 7160771fb45SEric Sandeen * Set the minimum read and write sizes 7170771fb45SEric Sandeen */ 7180771fb45SEric Sandeen xfs_set_rw_sizes(mp); 7190771fb45SEric Sandeen 720055388a3SDave Chinner /* set the low space thresholds for dynamic preallocation */ 721055388a3SDave Chinner xfs_set_low_space_thresholds(mp); 722055388a3SDave Chinner 7230771fb45SEric Sandeen /* 7240771fb45SEric Sandeen * Set the inode cluster size. 7250771fb45SEric Sandeen * This may still be overridden by the file system 7260771fb45SEric Sandeen * block size if it is larger than the chosen cluster size. 7278f80587bSDave Chinner * 7288f80587bSDave Chinner * For v5 filesystems, scale the cluster size with the inode size to 7298f80587bSDave Chinner * keep a constant ratio of inode per cluster buffer, but only if mkfs 7308f80587bSDave Chinner * has set the inode alignment value appropriately for larger cluster 7318f80587bSDave Chinner * sizes. 7320771fb45SEric Sandeen */ 7330771fb45SEric Sandeen mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; 7348f80587bSDave Chinner if (xfs_sb_version_hascrc(&mp->m_sb)) { 7358f80587bSDave Chinner int new_size = mp->m_inode_cluster_size; 7368f80587bSDave Chinner 7378f80587bSDave Chinner new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; 7388f80587bSDave Chinner if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) 7398f80587bSDave Chinner mp->m_inode_cluster_size = new_size; 7408f80587bSDave Chinner } 7410771fb45SEric Sandeen 7420771fb45SEric Sandeen /* 743e5376fc1SBrian Foster * If enabled, sparse inode chunk alignment is expected to match the 744e5376fc1SBrian Foster * cluster size. Full inode chunk alignment must match the chunk size, 745e5376fc1SBrian Foster * but that is checked on sb read verification... 746e5376fc1SBrian Foster */ 747e5376fc1SBrian Foster if (xfs_sb_version_hassparseinodes(&mp->m_sb) && 748e5376fc1SBrian Foster mp->m_sb.sb_spino_align != 749e5376fc1SBrian Foster XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { 750e5376fc1SBrian Foster xfs_warn(mp, 751e5376fc1SBrian Foster "Sparse inode block alignment (%u) must match cluster size (%llu).", 752e5376fc1SBrian Foster mp->m_sb.sb_spino_align, 753e5376fc1SBrian Foster XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); 754e5376fc1SBrian Foster error = -EINVAL; 755e5376fc1SBrian Foster goto out_remove_uuid; 756e5376fc1SBrian Foster } 757e5376fc1SBrian Foster 758e5376fc1SBrian Foster /* 7590771fb45SEric Sandeen * Set inode alignment fields 7600771fb45SEric Sandeen */ 7610771fb45SEric Sandeen xfs_set_inoalignment(mp); 7620771fb45SEric Sandeen 7630771fb45SEric Sandeen /* 764c2bfbc9bSZhi Yong Wu * Check that the data (and log if separate) is an ok size. 7650771fb45SEric Sandeen */ 7664249023aSChristoph Hellwig error = xfs_check_sizes(mp); 7670771fb45SEric Sandeen if (error) 768f9057e3dSChristoph Hellwig goto out_remove_uuid; 7690771fb45SEric Sandeen 7700771fb45SEric Sandeen /* 7711da177e4SLinus Torvalds * Initialize realtime fields in the mount structure 7721da177e4SLinus Torvalds */ 7730771fb45SEric Sandeen error = xfs_rtmount_init(mp); 7740771fb45SEric Sandeen if (error) { 7750b932cccSDave Chinner xfs_warn(mp, "RT mount failed"); 776f9057e3dSChristoph Hellwig goto out_remove_uuid; 7771da177e4SLinus Torvalds } 7781da177e4SLinus Torvalds 7791da177e4SLinus Torvalds /* 7801da177e4SLinus Torvalds * Copies the low order bits of the timestamp and the randomly 7811da177e4SLinus Torvalds * set "sequence" number out of a UUID. 7821da177e4SLinus Torvalds */ 7831da177e4SLinus Torvalds uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds mp->m_dmevmask = 0; /* not persistent; set after each mount */ 7861da177e4SLinus Torvalds 7870650b554SDave Chinner error = xfs_da_mount(mp); 7880650b554SDave Chinner if (error) { 7890650b554SDave Chinner xfs_warn(mp, "Failed dir/attr init: %d", error); 7900650b554SDave Chinner goto out_remove_uuid; 7910650b554SDave Chinner } 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds /* 7941da177e4SLinus Torvalds * Initialize the precomputed transaction reservations values. 7951da177e4SLinus Torvalds */ 7961da177e4SLinus Torvalds xfs_trans_init(mp); 7971da177e4SLinus Torvalds 7981da177e4SLinus Torvalds /* 7991da177e4SLinus Torvalds * Allocate and initialize the per-ag data. 8001da177e4SLinus Torvalds */ 8011c1c6ebcSDave Chinner spin_lock_init(&mp->m_perag_lock); 8029b98b6f3SDave Chinner INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); 8031c1c6ebcSDave Chinner error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); 8041c1c6ebcSDave Chinner if (error) { 8050b932cccSDave Chinner xfs_warn(mp, "Failed per-ag init: %d", error); 8060650b554SDave Chinner goto out_free_dir; 8071c1c6ebcSDave Chinner } 8081da177e4SLinus Torvalds 809f9057e3dSChristoph Hellwig if (!sbp->sb_logblocks) { 8100b932cccSDave Chinner xfs_warn(mp, "no log defined"); 811f9057e3dSChristoph Hellwig XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); 8122451337dSDave Chinner error = -EFSCORRUPTED; 813f9057e3dSChristoph Hellwig goto out_free_perag; 814f9057e3dSChristoph Hellwig } 815f9057e3dSChristoph Hellwig 8161da177e4SLinus Torvalds /* 817f0b2efadSBrian Foster * Log's mount-time initialization. The first part of recovery can place 818f0b2efadSBrian Foster * some items on the AIL, to be handled when recovery is finished or 819f0b2efadSBrian Foster * cancelled. 8201da177e4SLinus Torvalds */ 8211da177e4SLinus Torvalds error = xfs_log_mount(mp, mp->m_logdev_targp, 8221da177e4SLinus Torvalds XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), 8231da177e4SLinus Torvalds XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); 8241da177e4SLinus Torvalds if (error) { 8250b932cccSDave Chinner xfs_warn(mp, "log mount failed"); 826d4f3512bSDave Chinner goto out_fail_wait; 8271da177e4SLinus Torvalds } 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds /* 83092821e2bSDavid Chinner * Now the log is mounted, we know if it was an unclean shutdown or 83192821e2bSDavid Chinner * not. If it was, with the first phase of recovery has completed, we 83292821e2bSDavid Chinner * have consistent AG blocks on disk. We have not recovered EFIs yet, 83392821e2bSDavid Chinner * but they are recovered transactionally in the second recovery phase 83492821e2bSDavid Chinner * later. 83592821e2bSDavid Chinner * 83692821e2bSDavid Chinner * Hence we can safely re-initialise incore superblock counters from 83792821e2bSDavid Chinner * the per-ag data. These may not be correct if the filesystem was not 83892821e2bSDavid Chinner * cleanly unmounted, so we need to wait for recovery to finish before 83992821e2bSDavid Chinner * doing this. 84092821e2bSDavid Chinner * 84192821e2bSDavid Chinner * If the filesystem was cleanly unmounted, then we can trust the 84292821e2bSDavid Chinner * values in the superblock to be correct and we don't need to do 84392821e2bSDavid Chinner * anything here. 84492821e2bSDavid Chinner * 84592821e2bSDavid Chinner * If we are currently making the filesystem, the initialisation will 84692821e2bSDavid Chinner * fail as the perag data is in an undefined state. 84792821e2bSDavid Chinner */ 84892821e2bSDavid Chinner if (xfs_sb_version_haslazysbcount(&mp->m_sb) && 84992821e2bSDavid Chinner !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && 85092821e2bSDavid Chinner !mp->m_sb.sb_inprogress) { 85192821e2bSDavid Chinner error = xfs_initialize_perag_data(mp, sbp->sb_agcount); 852f9057e3dSChristoph Hellwig if (error) 8536eee8972Skbuild test robot goto out_log_dealloc; 85492821e2bSDavid Chinner } 855f9057e3dSChristoph Hellwig 85692821e2bSDavid Chinner /* 8571da177e4SLinus Torvalds * Get and sanity-check the root inode. 8581da177e4SLinus Torvalds * Save the pointer to it in the mount structure. 8591da177e4SLinus Torvalds */ 8607b6259e7SDave Chinner error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); 8611da177e4SLinus Torvalds if (error) { 8620b932cccSDave Chinner xfs_warn(mp, "failed to read root inode"); 863f9057e3dSChristoph Hellwig goto out_log_dealloc; 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds ASSERT(rip != NULL); 8671da177e4SLinus Torvalds 868*c19b3b05SDave Chinner if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { 8690b932cccSDave Chinner xfs_warn(mp, "corrupted root inode %llu: not a directory", 870b6574520SNathan Scott (unsigned long long)rip->i_ino); 8711da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8721da177e4SLinus Torvalds XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, 8731da177e4SLinus Torvalds mp); 8742451337dSDave Chinner error = -EFSCORRUPTED; 875f9057e3dSChristoph Hellwig goto out_rele_rip; 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds mp->m_rootip = rip; /* save it */ 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds xfs_iunlock(rip, XFS_ILOCK_EXCL); 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds /* 8821da177e4SLinus Torvalds * Initialize realtime inode pointers in the mount structure 8831da177e4SLinus Torvalds */ 8840771fb45SEric Sandeen error = xfs_rtmount_inodes(mp); 8850771fb45SEric Sandeen if (error) { 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * Free up the root inode. 8881da177e4SLinus Torvalds */ 8890b932cccSDave Chinner xfs_warn(mp, "failed to read RT inodes"); 890f9057e3dSChristoph Hellwig goto out_rele_rip; 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds /* 8947884bc86SChristoph Hellwig * If this is a read-only mount defer the superblock updates until 8957884bc86SChristoph Hellwig * the next remount into writeable mode. Otherwise we would never 8967884bc86SChristoph Hellwig * perform the update e.g. for the root filesystem. 8971da177e4SLinus Torvalds */ 89861e63ecbSDave Chinner if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { 89961e63ecbSDave Chinner error = xfs_sync_sb(mp, false); 900e5720eecSDavid Chinner if (error) { 9010b932cccSDave Chinner xfs_warn(mp, "failed to write sb changes"); 902b93b6e43SChristoph Hellwig goto out_rtunmount; 903e5720eecSDavid Chinner } 904e5720eecSDavid Chinner } 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * Initialise the XFS quota management subsystem for this mount 9081da177e4SLinus Torvalds */ 9097d095257SChristoph Hellwig if (XFS_IS_QUOTA_RUNNING(mp)) { 9107d095257SChristoph Hellwig error = xfs_qm_newmount(mp, "amount, "aflags); 9110771fb45SEric Sandeen if (error) 912b93b6e43SChristoph Hellwig goto out_rtunmount; 9137d095257SChristoph Hellwig } else { 9147d095257SChristoph Hellwig ASSERT(!XFS_IS_QUOTA_ON(mp)); 9157d095257SChristoph Hellwig 9167d095257SChristoph Hellwig /* 9177d095257SChristoph Hellwig * If a file system had quotas running earlier, but decided to 9187d095257SChristoph Hellwig * mount without -o uquota/pquota/gquota options, revoke the 9197d095257SChristoph Hellwig * quotachecked license. 9207d095257SChristoph Hellwig */ 9217d095257SChristoph Hellwig if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { 9220b932cccSDave Chinner xfs_notice(mp, "resetting quota flags"); 9237d095257SChristoph Hellwig error = xfs_mount_reset_sbqflags(mp); 9247d095257SChristoph Hellwig if (error) 925a70a4fa5SBrian Foster goto out_rtunmount; 9267d095257SChristoph Hellwig } 9277d095257SChristoph Hellwig } 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds /* 930f0b2efadSBrian Foster * Finish recovering the file system. This part needed to be delayed 931f0b2efadSBrian Foster * until after the root and real-time bitmap inodes were consistently 932f0b2efadSBrian Foster * read in. 9331da177e4SLinus Torvalds */ 9344249023aSChristoph Hellwig error = xfs_log_mount_finish(mp); 9351da177e4SLinus Torvalds if (error) { 9360b932cccSDave Chinner xfs_warn(mp, "log mount finish failed"); 937b93b6e43SChristoph Hellwig goto out_rtunmount; 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds 9401da177e4SLinus Torvalds /* 9411da177e4SLinus Torvalds * Complete the quota initialisation, post-log-replay component. 9421da177e4SLinus Torvalds */ 9437d095257SChristoph Hellwig if (quotamount) { 9447d095257SChristoph Hellwig ASSERT(mp->m_qflags == 0); 9457d095257SChristoph Hellwig mp->m_qflags = quotaflags; 9467d095257SChristoph Hellwig 9477d095257SChristoph Hellwig xfs_qm_mount_quotas(mp); 9487d095257SChristoph Hellwig } 9497d095257SChristoph Hellwig 95084e1e99fSDavid Chinner /* 95184e1e99fSDavid Chinner * Now we are mounted, reserve a small amount of unused space for 95284e1e99fSDavid Chinner * privileged transactions. This is needed so that transaction 95384e1e99fSDavid Chinner * space required for critical operations can dip into this pool 95484e1e99fSDavid Chinner * when at ENOSPC. This is needed for operations like create with 95584e1e99fSDavid Chinner * attr, unwritten extent conversion at ENOSPC, etc. Data allocations 95684e1e99fSDavid Chinner * are not allowed to use this reserved space. 9578babd8a2SDave Chinner * 9588babd8a2SDave Chinner * This may drive us straight to ENOSPC on mount, but that implies 9598babd8a2SDave Chinner * we were already there on the last unmount. Warn if this occurs. 96084e1e99fSDavid Chinner */ 961d5db0f97SEric Sandeen if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { 962d5db0f97SEric Sandeen resblks = xfs_default_resblks(mp); 963714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 964714082bcSDavid Chinner if (error) 9650b932cccSDave Chinner xfs_warn(mp, 9660b932cccSDave Chinner "Unable to allocate reserve blocks. Continuing without reserve pool."); 967d5db0f97SEric Sandeen } 96884e1e99fSDavid Chinner 9691da177e4SLinus Torvalds return 0; 9701da177e4SLinus Torvalds 971b93b6e43SChristoph Hellwig out_rtunmount: 972b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 973f9057e3dSChristoph Hellwig out_rele_rip: 97443355099SChristoph Hellwig IRELE(rip); 9750ae120f8SBrian Foster cancel_delayed_work_sync(&mp->m_reclaim_work); 9760ae120f8SBrian Foster xfs_reclaim_inodes(mp, SYNC_WAIT); 977f9057e3dSChristoph Hellwig out_log_dealloc: 978f0b2efadSBrian Foster xfs_log_mount_cancel(mp); 979d4f3512bSDave Chinner out_fail_wait: 980d4f3512bSDave Chinner if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) 981d4f3512bSDave Chinner xfs_wait_buftarg(mp->m_logdev_targp); 982d4f3512bSDave Chinner xfs_wait_buftarg(mp->m_ddev_targp); 983f9057e3dSChristoph Hellwig out_free_perag: 984ff4f038cSChristoph Hellwig xfs_free_perag(mp); 9850650b554SDave Chinner out_free_dir: 9860650b554SDave Chinner xfs_da_unmount(mp); 987f9057e3dSChristoph Hellwig out_remove_uuid: 98827174203SChristoph Hellwig xfs_uuid_unmount(mp); 989225e4635SBill O'Donnell out_del_stats: 990225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 991a31b1d3dSBrian Foster out_remove_sysfs: 992a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 993f9057e3dSChristoph Hellwig out: 9941da177e4SLinus Torvalds return error; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds /* 9981da177e4SLinus Torvalds * This flushes out the inodes,dquots and the superblock, unmounts the 9991da177e4SLinus Torvalds * log and makes sure that incore structures are freed. 10001da177e4SLinus Torvalds */ 100141b5c2e7SChristoph Hellwig void 100241b5c2e7SChristoph Hellwig xfs_unmountfs( 100341b5c2e7SChristoph Hellwig struct xfs_mount *mp) 10041da177e4SLinus Torvalds { 100584e1e99fSDavid Chinner __uint64_t resblks; 100641b5c2e7SChristoph Hellwig int error; 10071da177e4SLinus Torvalds 1008579b62faSBrian Foster cancel_delayed_work_sync(&mp->m_eofblocks_work); 1009579b62faSBrian Foster 10107d095257SChristoph Hellwig xfs_qm_unmount_quotas(mp); 1011b93b6e43SChristoph Hellwig xfs_rtunmount_inodes(mp); 101277508ec8SChristoph Hellwig IRELE(mp->m_rootip); 101377508ec8SChristoph Hellwig 1014641c56fbSDavid Chinner /* 1015641c56fbSDavid Chinner * We can potentially deadlock here if we have an inode cluster 10169da096fdSMalcolm Parsons * that has been freed has its buffer still pinned in memory because 1017641c56fbSDavid Chinner * the transaction is still sitting in a iclog. The stale inodes 1018641c56fbSDavid Chinner * on that buffer will have their flush locks held until the 1019641c56fbSDavid Chinner * transaction hits the disk and the callbacks run. the inode 1020641c56fbSDavid Chinner * flush takes the flush lock unconditionally and with nothing to 1021641c56fbSDavid Chinner * push out the iclog we will never get that unlocked. hence we 1022641c56fbSDavid Chinner * need to force the log first. 1023641c56fbSDavid Chinner */ 1024a14a348bSChristoph Hellwig xfs_log_force(mp, XFS_LOG_SYNC); 1025c854363eSDave Chinner 1026c854363eSDave Chinner /* 1027211e4d43SChristoph Hellwig * Flush all pending changes from the AIL. 1028c854363eSDave Chinner */ 1029211e4d43SChristoph Hellwig xfs_ail_push_all_sync(mp->m_ail); 1030211e4d43SChristoph Hellwig 1031211e4d43SChristoph Hellwig /* 1032211e4d43SChristoph Hellwig * And reclaim all inodes. At this point there should be no dirty 10337e18530bSDave Chinner * inodes and none should be pinned or locked, but use synchronous 10347e18530bSDave Chinner * reclaim just to be sure. We can stop background inode reclaim 10357e18530bSDave Chinner * here as well if it is still running. 1036211e4d43SChristoph Hellwig */ 10377e18530bSDave Chinner cancel_delayed_work_sync(&mp->m_reclaim_work); 1038c854363eSDave Chinner xfs_reclaim_inodes(mp, SYNC_WAIT); 10391da177e4SLinus Torvalds 10407d095257SChristoph Hellwig xfs_qm_unmount(mp); 1041a357a121SLachlan McIlroy 10421da177e4SLinus Torvalds /* 104384e1e99fSDavid Chinner * Unreserve any blocks we have so that when we unmount we don't account 104484e1e99fSDavid Chinner * the reserved free space as used. This is really only necessary for 104584e1e99fSDavid Chinner * lazy superblock counting because it trusts the incore superblock 10469da096fdSMalcolm Parsons * counters to be absolutely correct on clean unmount. 104784e1e99fSDavid Chinner * 104884e1e99fSDavid Chinner * We don't bother correcting this elsewhere for lazy superblock 104984e1e99fSDavid Chinner * counting because on mount of an unclean filesystem we reconstruct the 105084e1e99fSDavid Chinner * correct counter value and this is irrelevant. 105184e1e99fSDavid Chinner * 105284e1e99fSDavid Chinner * For non-lazy counter filesystems, this doesn't matter at all because 105384e1e99fSDavid Chinner * we only every apply deltas to the superblock and hence the incore 105484e1e99fSDavid Chinner * value does not matter.... 105584e1e99fSDavid Chinner */ 105684e1e99fSDavid Chinner resblks = 0; 1057714082bcSDavid Chinner error = xfs_reserve_blocks(mp, &resblks, NULL); 1058714082bcSDavid Chinner if (error) 10590b932cccSDave Chinner xfs_warn(mp, "Unable to free reserved block pool. " 1060714082bcSDavid Chinner "Freespace may not be correct on next mount."); 1061714082bcSDavid Chinner 1062adab0f67SChandra Seetharaman error = xfs_log_sbcount(mp); 1063e5720eecSDavid Chinner if (error) 10640b932cccSDave Chinner xfs_warn(mp, "Unable to update superblock counters. " 1065e5720eecSDavid Chinner "Freespace may not be correct on next mount."); 106687c7bec7SChristoph Hellwig 1067225e4635SBill O'Donnell 106821b699c8SChristoph Hellwig xfs_log_unmount(mp); 10690650b554SDave Chinner xfs_da_unmount(mp); 107027174203SChristoph Hellwig xfs_uuid_unmount(mp); 10711da177e4SLinus Torvalds 10721550d0b0SChristoph Hellwig #if defined(DEBUG) 10730ce4cfd4SChristoph Hellwig xfs_errortag_clearall(mp, 0); 10741da177e4SLinus Torvalds #endif 1075ff4f038cSChristoph Hellwig xfs_free_perag(mp); 1076a31b1d3dSBrian Foster 1077225e4635SBill O'Donnell xfs_sysfs_del(&mp->m_stats.xs_kobj); 1078a31b1d3dSBrian Foster xfs_sysfs_del(&mp->m_kobj); 10791da177e4SLinus Torvalds } 10801da177e4SLinus Torvalds 108191ee575fSBrian Foster /* 108291ee575fSBrian Foster * Determine whether modifications can proceed. The caller specifies the minimum 108391ee575fSBrian Foster * freeze level for which modifications should not be allowed. This allows 108491ee575fSBrian Foster * certain operations to proceed while the freeze sequence is in progress, if 108591ee575fSBrian Foster * necessary. 108691ee575fSBrian Foster */ 108791ee575fSBrian Foster bool 108891ee575fSBrian Foster xfs_fs_writable( 108991ee575fSBrian Foster struct xfs_mount *mp, 109091ee575fSBrian Foster int level) 109192821e2bSDavid Chinner { 109291ee575fSBrian Foster ASSERT(level > SB_UNFROZEN); 109391ee575fSBrian Foster if ((mp->m_super->s_writers.frozen >= level) || 109491ee575fSBrian Foster XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) 109591ee575fSBrian Foster return false; 109691ee575fSBrian Foster 109791ee575fSBrian Foster return true; 109892821e2bSDavid Chinner } 109992821e2bSDavid Chinner 110092821e2bSDavid Chinner /* 1101b2ce3974SAlex Elder * xfs_log_sbcount 1102b2ce3974SAlex Elder * 1103adab0f67SChandra Seetharaman * Sync the superblock counters to disk. 1104b2ce3974SAlex Elder * 110591ee575fSBrian Foster * Note this code can be called during the process of freezing, so we use the 110691ee575fSBrian Foster * transaction allocator that does not block when the transaction subsystem is 110791ee575fSBrian Foster * in its frozen state. 110892821e2bSDavid Chinner */ 110992821e2bSDavid Chinner int 1110adab0f67SChandra Seetharaman xfs_log_sbcount(xfs_mount_t *mp) 111192821e2bSDavid Chinner { 111291ee575fSBrian Foster /* allow this to proceed during the freeze sequence... */ 111391ee575fSBrian Foster if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) 111492821e2bSDavid Chinner return 0; 111592821e2bSDavid Chinner 111692821e2bSDavid Chinner /* 111792821e2bSDavid Chinner * we don't need to do this if we are updating the superblock 111892821e2bSDavid Chinner * counters on every modification. 111992821e2bSDavid Chinner */ 112092821e2bSDavid Chinner if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) 112192821e2bSDavid Chinner return 0; 112292821e2bSDavid Chinner 112361e63ecbSDave Chinner return xfs_sync_sb(mp, true); 112492821e2bSDavid Chinner } 112592821e2bSDavid Chinner 11268c1903d3SDave Chinner /* 11278c1903d3SDave Chinner * Deltas for the inode count are +/-64, hence we use a large batch size 11288c1903d3SDave Chinner * of 128 so we don't need to take the counter lock on every update. 11298c1903d3SDave Chinner */ 11308c1903d3SDave Chinner #define XFS_ICOUNT_BATCH 128 1131501ab323SDave Chinner int 1132501ab323SDave Chinner xfs_mod_icount( 1133501ab323SDave Chinner struct xfs_mount *mp, 1134501ab323SDave Chinner int64_t delta) 1135501ab323SDave Chinner { 11368c1903d3SDave Chinner __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); 11378c1903d3SDave Chinner if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { 1138501ab323SDave Chinner ASSERT(0); 1139501ab323SDave Chinner percpu_counter_add(&mp->m_icount, -delta); 1140501ab323SDave Chinner return -EINVAL; 1141501ab323SDave Chinner } 1142501ab323SDave Chinner return 0; 1143501ab323SDave Chinner } 1144501ab323SDave Chinner 1145e88b64eaSDave Chinner int 1146e88b64eaSDave Chinner xfs_mod_ifree( 1147e88b64eaSDave Chinner struct xfs_mount *mp, 1148e88b64eaSDave Chinner int64_t delta) 1149e88b64eaSDave Chinner { 1150e88b64eaSDave Chinner percpu_counter_add(&mp->m_ifree, delta); 1151e88b64eaSDave Chinner if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { 1152e88b64eaSDave Chinner ASSERT(0); 1153e88b64eaSDave Chinner percpu_counter_add(&mp->m_ifree, -delta); 1154e88b64eaSDave Chinner return -EINVAL; 1155e88b64eaSDave Chinner } 1156e88b64eaSDave Chinner return 0; 1157e88b64eaSDave Chinner } 11580d485adaSDave Chinner 11598c1903d3SDave Chinner /* 11608c1903d3SDave Chinner * Deltas for the block count can vary from 1 to very large, but lock contention 11618c1903d3SDave Chinner * only occurs on frequent small block count updates such as in the delayed 11628c1903d3SDave Chinner * allocation path for buffered writes (page a time updates). Hence we set 11638c1903d3SDave Chinner * a large batch count (1024) to minimise global counter updates except when 11648c1903d3SDave Chinner * we get near to ENOSPC and we have to be very accurate with our updates. 11658c1903d3SDave Chinner */ 11668c1903d3SDave Chinner #define XFS_FDBLOCKS_BATCH 1024 11670d485adaSDave Chinner int 11680d485adaSDave Chinner xfs_mod_fdblocks( 11690d485adaSDave Chinner struct xfs_mount *mp, 11700d485adaSDave Chinner int64_t delta, 11710d485adaSDave Chinner bool rsvd) 11720d485adaSDave Chinner { 11730d485adaSDave Chinner int64_t lcounter; 11740d485adaSDave Chinner long long res_used; 11750d485adaSDave Chinner s32 batch; 11760d485adaSDave Chinner 11770d485adaSDave Chinner if (delta > 0) { 11780d485adaSDave Chinner /* 11790d485adaSDave Chinner * If the reserve pool is depleted, put blocks back into it 11800d485adaSDave Chinner * first. Most of the time the pool is full. 11810d485adaSDave Chinner */ 11820d485adaSDave Chinner if (likely(mp->m_resblks == mp->m_resblks_avail)) { 11830d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, delta); 11840d485adaSDave Chinner return 0; 11850d485adaSDave Chinner } 11860d485adaSDave Chinner 11870d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 11880d485adaSDave Chinner res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); 11890d485adaSDave Chinner 11900d485adaSDave Chinner if (res_used > delta) { 11910d485adaSDave Chinner mp->m_resblks_avail += delta; 11920d485adaSDave Chinner } else { 11930d485adaSDave Chinner delta -= res_used; 11940d485adaSDave Chinner mp->m_resblks_avail = mp->m_resblks; 11950d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, delta); 11960d485adaSDave Chinner } 11970d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 11980d485adaSDave Chinner return 0; 11990d485adaSDave Chinner } 12000d485adaSDave Chinner 12010d485adaSDave Chinner /* 12020d485adaSDave Chinner * Taking blocks away, need to be more accurate the closer we 12030d485adaSDave Chinner * are to zero. 12040d485adaSDave Chinner * 12050d485adaSDave Chinner * If the counter has a value of less than 2 * max batch size, 12060d485adaSDave Chinner * then make everything serialise as we are real close to 12070d485adaSDave Chinner * ENOSPC. 12080d485adaSDave Chinner */ 12098c1903d3SDave Chinner if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, 12108c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) < 0) 12110d485adaSDave Chinner batch = 1; 12120d485adaSDave Chinner else 12138c1903d3SDave Chinner batch = XFS_FDBLOCKS_BATCH; 12140d485adaSDave Chinner 12150d485adaSDave Chinner __percpu_counter_add(&mp->m_fdblocks, delta, batch); 12168c1903d3SDave Chinner if (__percpu_counter_compare(&mp->m_fdblocks, XFS_ALLOC_SET_ASIDE(mp), 12178c1903d3SDave Chinner XFS_FDBLOCKS_BATCH) >= 0) { 12180d485adaSDave Chinner /* we had space! */ 12190d485adaSDave Chinner return 0; 12200d485adaSDave Chinner } 12210d485adaSDave Chinner 12220d485adaSDave Chinner /* 12230d485adaSDave Chinner * lock up the sb for dipping into reserves before releasing the space 12240d485adaSDave Chinner * that took us to ENOSPC. 12250d485adaSDave Chinner */ 12260d485adaSDave Chinner spin_lock(&mp->m_sb_lock); 12270d485adaSDave Chinner percpu_counter_add(&mp->m_fdblocks, -delta); 12280d485adaSDave Chinner if (!rsvd) 12290d485adaSDave Chinner goto fdblocks_enospc; 12300d485adaSDave Chinner 12310d485adaSDave Chinner lcounter = (long long)mp->m_resblks_avail + delta; 12320d485adaSDave Chinner if (lcounter >= 0) { 12330d485adaSDave Chinner mp->m_resblks_avail = lcounter; 12340d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 12350d485adaSDave Chinner return 0; 12360d485adaSDave Chinner } 12370d485adaSDave Chinner printk_once(KERN_WARNING 12380d485adaSDave Chinner "Filesystem \"%s\": reserve blocks depleted! " 12390d485adaSDave Chinner "Consider increasing reserve pool size.", 12400d485adaSDave Chinner mp->m_fsname); 12410d485adaSDave Chinner fdblocks_enospc: 12420d485adaSDave Chinner spin_unlock(&mp->m_sb_lock); 12430d485adaSDave Chinner return -ENOSPC; 12440d485adaSDave Chinner } 12450d485adaSDave Chinner 1246bab98bbeSDave Chinner int 1247bab98bbeSDave Chinner xfs_mod_frextents( 1248bab98bbeSDave Chinner struct xfs_mount *mp, 1249bab98bbeSDave Chinner int64_t delta) 1250bab98bbeSDave Chinner { 1251bab98bbeSDave Chinner int64_t lcounter; 1252bab98bbeSDave Chinner int ret = 0; 1253bab98bbeSDave Chinner 1254bab98bbeSDave Chinner spin_lock(&mp->m_sb_lock); 1255bab98bbeSDave Chinner lcounter = mp->m_sb.sb_frextents + delta; 1256bab98bbeSDave Chinner if (lcounter < 0) 1257bab98bbeSDave Chinner ret = -ENOSPC; 1258bab98bbeSDave Chinner else 1259bab98bbeSDave Chinner mp->m_sb.sb_frextents = lcounter; 1260bab98bbeSDave Chinner spin_unlock(&mp->m_sb_lock); 1261bab98bbeSDave Chinner return ret; 1262bab98bbeSDave Chinner } 1263bab98bbeSDave Chinner 12641da177e4SLinus Torvalds /* 12651da177e4SLinus Torvalds * xfs_getsb() is called to obtain the buffer for the superblock. 12661da177e4SLinus Torvalds * The buffer is returned locked and read in from disk. 12671da177e4SLinus Torvalds * The buffer should be released with a call to xfs_brelse(). 12681da177e4SLinus Torvalds * 12691da177e4SLinus Torvalds * If the flags parameter is BUF_TRYLOCK, then we'll only return 12701da177e4SLinus Torvalds * the superblock buffer if it can be locked without sleeping. 12711da177e4SLinus Torvalds * If it can't then we'll return NULL. 12721da177e4SLinus Torvalds */ 12730c842ad4SChristoph Hellwig struct xfs_buf * 12741da177e4SLinus Torvalds xfs_getsb( 12750c842ad4SChristoph Hellwig struct xfs_mount *mp, 12761da177e4SLinus Torvalds int flags) 12771da177e4SLinus Torvalds { 12780c842ad4SChristoph Hellwig struct xfs_buf *bp = mp->m_sb_bp; 12791da177e4SLinus Torvalds 12800c842ad4SChristoph Hellwig if (!xfs_buf_trylock(bp)) { 12810c842ad4SChristoph Hellwig if (flags & XBF_TRYLOCK) 12821da177e4SLinus Torvalds return NULL; 12830c842ad4SChristoph Hellwig xfs_buf_lock(bp); 12841da177e4SLinus Torvalds } 12850c842ad4SChristoph Hellwig 128672790aa1SChandra Seetharaman xfs_buf_hold(bp); 12871da177e4SLinus Torvalds ASSERT(XFS_BUF_ISDONE(bp)); 1288014c2544SJesper Juhl return bp; 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* 12921da177e4SLinus Torvalds * Used to free the superblock along various error paths. 12931da177e4SLinus Torvalds */ 12941da177e4SLinus Torvalds void 12951da177e4SLinus Torvalds xfs_freesb( 129626af6552SDave Chinner struct xfs_mount *mp) 12971da177e4SLinus Torvalds { 129826af6552SDave Chinner struct xfs_buf *bp = mp->m_sb_bp; 12991da177e4SLinus Torvalds 130026af6552SDave Chinner xfs_buf_lock(bp); 13011da177e4SLinus Torvalds mp->m_sb_bp = NULL; 130226af6552SDave Chinner xfs_buf_relse(bp); 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* 1306dda35b8fSChristoph Hellwig * If the underlying (data/log/rt) device is readonly, there are some 1307dda35b8fSChristoph Hellwig * operations that cannot proceed. 1308dda35b8fSChristoph Hellwig */ 1309dda35b8fSChristoph Hellwig int 1310dda35b8fSChristoph Hellwig xfs_dev_is_read_only( 1311dda35b8fSChristoph Hellwig struct xfs_mount *mp, 1312dda35b8fSChristoph Hellwig char *message) 1313dda35b8fSChristoph Hellwig { 1314dda35b8fSChristoph Hellwig if (xfs_readonly_buftarg(mp->m_ddev_targp) || 1315dda35b8fSChristoph Hellwig xfs_readonly_buftarg(mp->m_logdev_targp) || 1316dda35b8fSChristoph Hellwig (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { 13170b932cccSDave Chinner xfs_notice(mp, "%s required on read-only device.", message); 13180b932cccSDave Chinner xfs_notice(mp, "write access unavailable, cannot proceed."); 13192451337dSDave Chinner return -EROFS; 1320dda35b8fSChristoph Hellwig } 1321dda35b8fSChristoph Hellwig return 0; 1322dda35b8fSChristoph Hellwig } 1323