Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include "xfs.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 19 | #include "xfs_fs.h" |
Dave Chinner | 6ca1c90 | 2013-08-12 20:49:26 +1000 | [diff] [blame] | 20 | #include "xfs_format.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 21 | #include "xfs_bit.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include "xfs_log.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 23 | #include "xfs_inum.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_trans.h" |
Christoph Hellwig | 211e4d4 | 2012-04-23 15:58:34 +1000 | [diff] [blame] | 25 | #include "xfs_trans_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include "xfs_sb.h" |
| 27 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include "xfs_mount.h" |
Dave Chinner | 2b9ab5a | 2013-08-12 20:49:37 +1000 | [diff] [blame] | 29 | #include "xfs_da_btree.h" |
| 30 | #include "xfs_dir2_format.h" |
| 31 | #include "xfs_dir2.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include "xfs_bmap_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 33 | #include "xfs_alloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include "xfs_ialloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include "xfs_dinode.h" |
| 36 | #include "xfs_inode.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 37 | #include "xfs_btree.h" |
| 38 | #include "xfs_ialloc.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include "xfs_alloc.h" |
| 40 | #include "xfs_rtalloc.h" |
| 41 | #include "xfs_bmap.h" |
| 42 | #include "xfs_error.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include "xfs_quota.h" |
| 44 | #include "xfs_fsops.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 45 | #include "xfs_trace.h" |
Dave Chinner | 6d8b79c | 2012-10-08 21:56:09 +1100 | [diff] [blame] | 46 | #include "xfs_icache.h" |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 47 | #include "xfs_cksum.h" |
| 48 | #include "xfs_buf_item.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 49 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 51 | #ifdef HAVE_PERCPU_SB |
David Chinner | 20f4ebf | 2007-02-10 18:36:10 +1100 | [diff] [blame] | 52 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 53 | int); |
| 54 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, |
| 55 | int); |
David Chinner | 36fbe6e | 2008-04-10 12:19:56 +1000 | [diff] [blame] | 56 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 57 | #else |
| 58 | |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 59 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
| 60 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 61 | #endif |
| 62 | |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 63 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
| 64 | static int xfs_uuid_table_size; |
| 65 | static uuid_t *xfs_uuid_table; |
| 66 | |
| 67 | /* |
| 68 | * See if the UUID is unique among mounted XFS filesystems. |
| 69 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. |
| 70 | */ |
| 71 | STATIC int |
| 72 | xfs_uuid_mount( |
| 73 | struct xfs_mount *mp) |
| 74 | { |
| 75 | uuid_t *uuid = &mp->m_sb.sb_uuid; |
| 76 | int hole, i; |
| 77 | |
| 78 | if (mp->m_flags & XFS_MOUNT_NOUUID) |
| 79 | return 0; |
| 80 | |
| 81 | if (uuid_is_nil(uuid)) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 82 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 83 | return XFS_ERROR(EINVAL); |
| 84 | } |
| 85 | |
| 86 | mutex_lock(&xfs_uuid_table_mutex); |
| 87 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { |
| 88 | if (uuid_is_nil(&xfs_uuid_table[i])) { |
| 89 | hole = i; |
| 90 | continue; |
| 91 | } |
| 92 | if (uuid_equal(uuid, &xfs_uuid_table[i])) |
| 93 | goto out_duplicate; |
| 94 | } |
| 95 | |
| 96 | if (hole < 0) { |
| 97 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, |
| 98 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), |
| 99 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), |
| 100 | KM_SLEEP); |
| 101 | hole = xfs_uuid_table_size++; |
| 102 | } |
| 103 | xfs_uuid_table[hole] = *uuid; |
| 104 | mutex_unlock(&xfs_uuid_table_mutex); |
| 105 | |
| 106 | return 0; |
| 107 | |
| 108 | out_duplicate: |
| 109 | mutex_unlock(&xfs_uuid_table_mutex); |
Mitsuo Hayasaka | 021000e | 2012-01-13 05:58:39 +0000 | [diff] [blame] | 110 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 111 | return XFS_ERROR(EINVAL); |
| 112 | } |
| 113 | |
| 114 | STATIC void |
| 115 | xfs_uuid_unmount( |
| 116 | struct xfs_mount *mp) |
| 117 | { |
| 118 | uuid_t *uuid = &mp->m_sb.sb_uuid; |
| 119 | int i; |
| 120 | |
| 121 | if (mp->m_flags & XFS_MOUNT_NOUUID) |
| 122 | return; |
| 123 | |
| 124 | mutex_lock(&xfs_uuid_table_mutex); |
| 125 | for (i = 0; i < xfs_uuid_table_size; i++) { |
| 126 | if (uuid_is_nil(&xfs_uuid_table[i])) |
| 127 | continue; |
| 128 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) |
| 129 | continue; |
| 130 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); |
| 131 | break; |
| 132 | } |
| 133 | ASSERT(i < xfs_uuid_table_size); |
| 134 | mutex_unlock(&xfs_uuid_table_mutex); |
| 135 | } |
| 136 | |
| 137 | |
Dave Chinner | e176579 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 138 | STATIC void |
| 139 | __xfs_free_perag( |
| 140 | struct rcu_head *head) |
| 141 | { |
| 142 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); |
| 143 | |
| 144 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
| 145 | kmem_free(pag); |
| 146 | } |
| 147 | |
Dave Chinner | 0fa800f | 2010-01-11 11:47:46 +0000 | [diff] [blame] | 148 | /* |
Dave Chinner | e176579 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 149 | * Free up the per-ag resources associated with the mount structure. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | */ |
Christoph Hellwig | c962fb7 | 2008-05-20 15:10:52 +1000 | [diff] [blame] | 151 | STATIC void |
Christoph Hellwig | ff4f038 | 2008-08-13 16:50:47 +1000 | [diff] [blame] | 152 | xfs_free_perag( |
Christoph Hellwig | 745f691 | 2007-08-30 17:20:39 +1000 | [diff] [blame] | 153 | xfs_mount_t *mp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | { |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 155 | xfs_agnumber_t agno; |
| 156 | struct xfs_perag *pag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 158 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
| 159 | spin_lock(&mp->m_perag_lock); |
| 160 | pag = radix_tree_delete(&mp->m_perag_tree, agno); |
| 161 | spin_unlock(&mp->m_perag_lock); |
Dave Chinner | e176579 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 162 | ASSERT(pag); |
Dave Chinner | f83282a | 2010-11-08 08:55:04 +0000 | [diff] [blame] | 163 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
Dave Chinner | e176579 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 164 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | } |
| 167 | |
Nathan Scott | 4cc929e | 2007-05-14 18:24:02 +1000 | [diff] [blame] | 168 | /* |
| 169 | * Check size of device based on the (data/realtime) block count. |
| 170 | * Note: this check is used by the growfs code as well as mount. |
| 171 | */ |
| 172 | int |
| 173 | xfs_sb_validate_fsb_count( |
| 174 | xfs_sb_t *sbp, |
| 175 | __uint64_t nblocks) |
| 176 | { |
| 177 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); |
| 178 | ASSERT(sbp->sb_blocklog >= BBSHIFT); |
| 179 | |
| 180 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ |
| 181 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
Eric Sandeen | 657a4cf | 2010-04-30 03:42:49 +0000 | [diff] [blame] | 182 | return EFBIG; |
Nathan Scott | 4cc929e | 2007-05-14 18:24:02 +1000 | [diff] [blame] | 183 | #else /* Limited by UINT_MAX of sectors */ |
| 184 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) |
Eric Sandeen | 657a4cf | 2010-04-30 03:42:49 +0000 | [diff] [blame] | 185 | return EFBIG; |
Nathan Scott | 4cc929e | 2007-05-14 18:24:02 +1000 | [diff] [blame] | 186 | #endif |
| 187 | return 0; |
| 188 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 190 | int |
Nathan Scott | c11e2c3 | 2005-11-02 15:11:45 +1100 | [diff] [blame] | 191 | xfs_initialize_perag( |
Nathan Scott | c11e2c3 | 2005-11-02 15:11:45 +1100 | [diff] [blame] | 192 | xfs_mount_t *mp, |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 193 | xfs_agnumber_t agcount, |
| 194 | xfs_agnumber_t *maxagi) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | { |
Carlos Maiolino | 2d2194f | 2012-09-20 10:32:38 -0300 | [diff] [blame] | 196 | xfs_agnumber_t index; |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 197 | xfs_agnumber_t first_initialised = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | xfs_perag_t *pag; |
| 199 | xfs_agino_t agino; |
| 200 | xfs_ino_t ino; |
| 201 | xfs_sb_t *sbp = &mp->m_sb; |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 202 | int error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 204 | /* |
| 205 | * Walk the current per-ag tree so we don't try to initialise AGs |
| 206 | * that already exist (growfs case). Allocate and insert all the |
| 207 | * AGs we don't find ready for initialisation. |
| 208 | */ |
| 209 | for (index = 0; index < agcount; index++) { |
| 210 | pag = xfs_perag_get(mp, index); |
| 211 | if (pag) { |
| 212 | xfs_perag_put(pag); |
| 213 | continue; |
| 214 | } |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 215 | if (!first_initialised) |
| 216 | first_initialised = index; |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 217 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 218 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
| 219 | if (!pag) |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 220 | goto out_unwind; |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 221 | pag->pag_agno = index; |
| 222 | pag->pag_mount = mp; |
Dave Chinner | 1a427ab | 2010-12-16 17:08:41 +1100 | [diff] [blame] | 223 | spin_lock_init(&pag->pag_ici_lock); |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 224 | mutex_init(&pag->pag_ici_reclaim_lock); |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 225 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
Dave Chinner | 74f75a0 | 2010-09-24 19:59:04 +1000 | [diff] [blame] | 226 | spin_lock_init(&pag->pag_buf_lock); |
| 227 | pag->pag_buf_tree = RB_ROOT; |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 228 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 229 | if (radix_tree_preload(GFP_NOFS)) |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 230 | goto out_unwind; |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 231 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 232 | spin_lock(&mp->m_perag_lock); |
| 233 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { |
| 234 | BUG(); |
| 235 | spin_unlock(&mp->m_perag_lock); |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 236 | radix_tree_preload_end(); |
| 237 | error = -EEXIST; |
| 238 | goto out_unwind; |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 239 | } |
| 240 | spin_unlock(&mp->m_perag_lock); |
| 241 | radix_tree_preload_end(); |
| 242 | } |
| 243 | |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 244 | /* |
| 245 | * If we mount with the inode64 option, or no inode overflows |
| 246 | * the legacy 32-bit address space clear the inode32 option. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | */ |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 248 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
| 249 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 250 | |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 251 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) |
| 252 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
| 253 | else |
| 254 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
| 255 | |
Carlos Maiolino | 2d2194f | 2012-09-20 10:32:38 -0300 | [diff] [blame] | 256 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
| 257 | index = xfs_set_inode32(mp); |
| 258 | else |
| 259 | index = xfs_set_inode64(mp); |
Christoph Hellwig | fb3b504 | 2010-05-28 19:03:10 +0000 | [diff] [blame] | 260 | |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 261 | if (maxagi) |
| 262 | *maxagi = index; |
| 263 | return 0; |
Dave Chinner | 8b26c58 | 2010-01-11 11:47:48 +0000 | [diff] [blame] | 264 | |
| 265 | out_unwind: |
| 266 | kmem_free(pag); |
| 267 | for (; index > first_initialised; index--) { |
| 268 | pag = radix_tree_delete(&mp->m_perag_tree, index); |
| 269 | kmem_free(pag); |
| 270 | } |
| 271 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | } |
| 273 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | /* |
| 275 | * xfs_readsb |
| 276 | * |
| 277 | * Does the initial read of the superblock. |
| 278 | */ |
| 279 | int |
Dave Chinner | ff55068 | 2013-08-12 20:49:41 +1000 | [diff] [blame] | 280 | xfs_readsb( |
| 281 | struct xfs_mount *mp, |
| 282 | int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | { |
| 284 | unsigned int sector_size; |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 285 | struct xfs_buf *bp; |
| 286 | struct xfs_sb *sbp = &mp->m_sb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | int error; |
Dave Chinner | af34e09 | 2011-03-07 10:04:35 +1100 | [diff] [blame] | 288 | int loud = !(flags & XFS_MFSI_QUIET); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | |
| 290 | ASSERT(mp->m_sb_bp == NULL); |
| 291 | ASSERT(mp->m_ddev_targp != NULL); |
| 292 | |
| 293 | /* |
| 294 | * Allocate a (locked) buffer to hold the superblock. |
| 295 | * This will be kept around at all times to optimize |
| 296 | * access to the superblock. |
| 297 | */ |
| 298 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 300 | reread: |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 301 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 302 | BTOBB(sector_size), 0, |
Dave Chinner | 1813dd6 | 2012-11-14 17:54:40 +1100 | [diff] [blame] | 303 | loud ? &xfs_sb_buf_ops |
| 304 | : &xfs_sb_quiet_buf_ops); |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 305 | if (!bp) { |
Dave Chinner | af34e09 | 2011-03-07 10:04:35 +1100 | [diff] [blame] | 306 | if (loud) |
| 307 | xfs_warn(mp, "SB buffer read failed"); |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 308 | return EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | } |
Dave Chinner | eab4e63 | 2012-11-12 22:54:02 +1100 | [diff] [blame] | 310 | if (bp->b_error) { |
| 311 | error = bp->b_error; |
| 312 | if (loud) |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 313 | xfs_warn(mp, "SB validate failed with error %d.", error); |
Dave Chinner | eab4e63 | 2012-11-12 22:54:02 +1100 | [diff] [blame] | 314 | goto release_buf; |
| 315 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | /* |
| 318 | * Initialize the mount structure from the superblock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | */ |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 320 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
Chandra Seetharaman | 83e782e | 2013-06-27 17:25:10 -0500 | [diff] [blame] | 321 | xfs_sb_quota_from_disk(&mp->m_sb); |
Dave Chinner | ff55068 | 2013-08-12 20:49:41 +1000 | [diff] [blame] | 322 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | /* |
| 324 | * We must be able to do sector-sized and sector-aligned IO. |
| 325 | */ |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 326 | if (sector_size > sbp->sb_sectsize) { |
Dave Chinner | af34e09 | 2011-03-07 10:04:35 +1100 | [diff] [blame] | 327 | if (loud) |
| 328 | xfs_warn(mp, "device supports %u byte sectors (not %u)", |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 329 | sector_size, sbp->sb_sectsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | error = ENOSYS; |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 331 | goto release_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | } |
| 333 | |
| 334 | /* |
| 335 | * If device sector size is smaller than the superblock size, |
| 336 | * re-read the superblock so the buffer is correctly sized. |
| 337 | */ |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 338 | if (sector_size < sbp->sb_sectsize) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | xfs_buf_relse(bp); |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 340 | sector_size = sbp->sb_sectsize; |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 341 | goto reread; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | } |
| 343 | |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 344 | /* Initialize per-cpu counters */ |
| 345 | xfs_icsb_reinit_counters(mp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 346 | |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 347 | /* no need to be quiet anymore, so reset the buf ops */ |
| 348 | bp->b_ops = &xfs_sb_buf_ops; |
| 349 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | mp->m_sb_bp = bp; |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 351 | xfs_buf_unlock(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 352 | return 0; |
| 353 | |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 354 | release_buf: |
| 355 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | return error; |
| 357 | } |
| 358 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | /* |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 360 | * Update alignment values based on mount options and sb values |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | */ |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 362 | STATIC int |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 363 | xfs_update_alignment(xfs_mount_t *mp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | xfs_sb_t *sbp = &(mp->m_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 367 | if (mp->m_dalign) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | /* |
| 369 | * If stripe unit and stripe width are not multiples |
| 370 | * of the fs blocksize turn off alignment. |
| 371 | */ |
| 372 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || |
| 373 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { |
Jie Liu | 39a45d8 | 2013-05-02 19:27:47 +0800 | [diff] [blame] | 374 | xfs_warn(mp, |
| 375 | "alignment check failed: sunit/swidth vs. blocksize(%d)", |
| 376 | sbp->sb_blocksize); |
| 377 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | } else { |
| 379 | /* |
| 380 | * Convert the stripe unit and width to FSBs. |
| 381 | */ |
| 382 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); |
| 383 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { |
Dave Chinner | 5348778 | 2011-03-07 10:05:35 +1100 | [diff] [blame] | 384 | xfs_warn(mp, |
Jie Liu | 39a45d8 | 2013-05-02 19:27:47 +0800 | [diff] [blame] | 385 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
| 386 | sbp->sb_agblocks); |
| 387 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | } else if (mp->m_dalign) { |
| 389 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); |
| 390 | } else { |
Jie Liu | 39a45d8 | 2013-05-02 19:27:47 +0800 | [diff] [blame] | 391 | xfs_warn(mp, |
| 392 | "alignment check failed: sunit(%d) less than bsize(%d)", |
| 393 | mp->m_dalign, sbp->sb_blocksize); |
| 394 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | } |
| 396 | } |
| 397 | |
| 398 | /* |
| 399 | * Update superblock with new values |
| 400 | * and log changes |
| 401 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 402 | if (xfs_sb_version_hasdalign(sbp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | if (sbp->sb_unit != mp->m_dalign) { |
| 404 | sbp->sb_unit = mp->m_dalign; |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 405 | mp->m_update_flags |= XFS_SB_UNIT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | } |
| 407 | if (sbp->sb_width != mp->m_swidth) { |
| 408 | sbp->sb_width = mp->m_swidth; |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 409 | mp->m_update_flags |= XFS_SB_WIDTH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | } |
Jie Liu | 34d7f60 | 2013-05-02 19:27:53 +0800 | [diff] [blame] | 411 | } else { |
| 412 | xfs_warn(mp, |
| 413 | "cannot change alignment: superblock does not support data alignment"); |
| 414 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | } |
| 416 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 417 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | mp->m_dalign = sbp->sb_unit; |
| 419 | mp->m_swidth = sbp->sb_width; |
| 420 | } |
| 421 | |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 422 | return 0; |
| 423 | } |
| 424 | |
| 425 | /* |
| 426 | * Set the maximum inode count for this filesystem |
| 427 | */ |
| 428 | STATIC void |
| 429 | xfs_set_maxicount(xfs_mount_t *mp) |
| 430 | { |
| 431 | xfs_sb_t *sbp = &(mp->m_sb); |
| 432 | __uint64_t icount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | |
| 434 | if (sbp->sb_imax_pct) { |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 435 | /* |
| 436 | * Make sure the maximum inode count is a multiple |
| 437 | * of the units we allocate inodes in. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
| 440 | do_div(icount, 100); |
| 441 | do_div(icount, mp->m_ialloc_blks); |
| 442 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << |
| 443 | sbp->sb_inopblog; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 444 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | mp->m_maxicount = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | } |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 447 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 448 | |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 449 | /* |
| 450 | * Set the default minimum read and write sizes unless |
| 451 | * already specified in a mount option. |
| 452 | * We use smaller I/O sizes when the file system |
| 453 | * is being used for NFS service (wsync mount option). |
| 454 | */ |
| 455 | STATIC void |
| 456 | xfs_set_rw_sizes(xfs_mount_t *mp) |
| 457 | { |
| 458 | xfs_sb_t *sbp = &(mp->m_sb); |
| 459 | int readio_log, writeio_log; |
| 460 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
| 462 | if (mp->m_flags & XFS_MOUNT_WSYNC) { |
| 463 | readio_log = XFS_WSYNC_READIO_LOG; |
| 464 | writeio_log = XFS_WSYNC_WRITEIO_LOG; |
| 465 | } else { |
| 466 | readio_log = XFS_READIO_LOG_LARGE; |
| 467 | writeio_log = XFS_WRITEIO_LOG_LARGE; |
| 468 | } |
| 469 | } else { |
| 470 | readio_log = mp->m_readio_log; |
| 471 | writeio_log = mp->m_writeio_log; |
| 472 | } |
| 473 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | if (sbp->sb_blocklog > readio_log) { |
| 475 | mp->m_readio_log = sbp->sb_blocklog; |
| 476 | } else { |
| 477 | mp->m_readio_log = readio_log; |
| 478 | } |
| 479 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); |
| 480 | if (sbp->sb_blocklog > writeio_log) { |
| 481 | mp->m_writeio_log = sbp->sb_blocklog; |
| 482 | } else { |
| 483 | mp->m_writeio_log = writeio_log; |
| 484 | } |
| 485 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 486 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 488 | /* |
Dave Chinner | 055388a | 2011-01-04 11:35:03 +1100 | [diff] [blame] | 489 | * precalculate the low space thresholds for dynamic speculative preallocation. |
| 490 | */ |
| 491 | void |
| 492 | xfs_set_low_space_thresholds( |
| 493 | struct xfs_mount *mp) |
| 494 | { |
| 495 | int i; |
| 496 | |
| 497 | for (i = 0; i < XFS_LOWSP_MAX; i++) { |
| 498 | __uint64_t space = mp->m_sb.sb_dblocks; |
| 499 | |
| 500 | do_div(space, 100); |
| 501 | mp->m_low_space[i] = space * (i + 1); |
| 502 | } |
| 503 | } |
| 504 | |
| 505 | |
| 506 | /* |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 507 | * Set whether we're using inode alignment. |
| 508 | */ |
| 509 | STATIC void |
| 510 | xfs_set_inoalignment(xfs_mount_t *mp) |
| 511 | { |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 512 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | mp->m_sb.sb_inoalignmt >= |
| 514 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) |
| 515 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; |
| 516 | else |
| 517 | mp->m_inoalign_mask = 0; |
| 518 | /* |
| 519 | * If we are using stripe alignment, check whether |
| 520 | * the stripe unit is a multiple of the inode alignment |
| 521 | */ |
| 522 | if (mp->m_dalign && mp->m_inoalign_mask && |
| 523 | !(mp->m_dalign & mp->m_inoalign_mask)) |
| 524 | mp->m_sinoalign = mp->m_dalign; |
| 525 | else |
| 526 | mp->m_sinoalign = 0; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 527 | } |
| 528 | |
| 529 | /* |
Zhi Yong Wu | 0471f62 | 2013-08-07 10:10:58 +0000 | [diff] [blame] | 530 | * Check that the data (and log if separate) is an ok size. |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 531 | */ |
| 532 | STATIC int |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 533 | xfs_check_sizes(xfs_mount_t *mp) |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 534 | { |
| 535 | xfs_buf_t *bp; |
| 536 | xfs_daddr_t d; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 537 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
| 539 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 540 | xfs_warn(mp, "filesystem size mismatch detected"); |
Eric Sandeen | 657a4cf | 2010-04-30 03:42:49 +0000 | [diff] [blame] | 541 | return XFS_ERROR(EFBIG); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | } |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 543 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 544 | d - XFS_FSS_TO_BB(mp, 1), |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 545 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 546 | if (!bp) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 547 | xfs_warn(mp, "last sector read failed"); |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 548 | return EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | } |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 550 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 552 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
| 554 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 555 | xfs_warn(mp, "log size mismatch detected"); |
Eric Sandeen | 657a4cf | 2010-04-30 03:42:49 +0000 | [diff] [blame] | 556 | return XFS_ERROR(EFBIG); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | } |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 558 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 559 | d - XFS_FSB_TO_BB(mp, 1), |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 560 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 561 | if (!bp) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 562 | xfs_warn(mp, "log device read failed"); |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 563 | return EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | } |
Dave Chinner | 1922c94 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 565 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | } |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 567 | return 0; |
| 568 | } |
| 569 | |
| 570 | /* |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 571 | * Clear the quotaflags in memory and in the superblock. |
| 572 | */ |
| 573 | int |
| 574 | xfs_mount_reset_sbqflags( |
| 575 | struct xfs_mount *mp) |
| 576 | { |
| 577 | int error; |
| 578 | struct xfs_trans *tp; |
| 579 | |
| 580 | mp->m_qflags = 0; |
| 581 | |
| 582 | /* |
| 583 | * It is OK to look at sb_qflags here in mount path, |
| 584 | * without m_sb_lock. |
| 585 | */ |
| 586 | if (mp->m_sb.sb_qflags == 0) |
| 587 | return 0; |
| 588 | spin_lock(&mp->m_sb_lock); |
| 589 | mp->m_sb.sb_qflags = 0; |
| 590 | spin_unlock(&mp->m_sb_lock); |
| 591 | |
| 592 | /* |
| 593 | * If the fs is readonly, let the incore superblock run |
| 594 | * with quotas off but don't flush the update out to disk |
| 595 | */ |
| 596 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
| 597 | return 0; |
| 598 | |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 599 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 600 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0); |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 601 | if (error) { |
| 602 | xfs_trans_cancel(tp, 0); |
Dave Chinner | 5348778 | 2011-03-07 10:05:35 +1100 | [diff] [blame] | 603 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 604 | return error; |
| 605 | } |
| 606 | |
| 607 | xfs_mod_sb(tp, XFS_SB_QFLAGS); |
| 608 | return xfs_trans_commit(tp, 0); |
| 609 | } |
| 610 | |
Eric Sandeen | d5db0f9 | 2010-02-05 22:59:53 +0000 | [diff] [blame] | 611 | __uint64_t |
| 612 | xfs_default_resblks(xfs_mount_t *mp) |
| 613 | { |
| 614 | __uint64_t resblks; |
| 615 | |
| 616 | /* |
Dave Chinner | 8babd8a | 2010-03-04 01:46:25 +0000 | [diff] [blame] | 617 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
| 618 | * smaller. This is intended to cover concurrent allocation |
| 619 | * transactions when we initially hit enospc. These each require a 4 |
| 620 | * block reservation. Hence by default we cover roughly 2000 concurrent |
| 621 | * allocation reservations. |
Eric Sandeen | d5db0f9 | 2010-02-05 22:59:53 +0000 | [diff] [blame] | 622 | */ |
| 623 | resblks = mp->m_sb.sb_dblocks; |
| 624 | do_div(resblks, 20); |
Dave Chinner | 8babd8a | 2010-03-04 01:46:25 +0000 | [diff] [blame] | 625 | resblks = min_t(__uint64_t, resblks, 8192); |
Eric Sandeen | d5db0f9 | 2010-02-05 22:59:53 +0000 | [diff] [blame] | 626 | return resblks; |
| 627 | } |
| 628 | |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 629 | /* |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 630 | * This function does the following on an initial mount of a file system: |
| 631 | * - reads the superblock from disk and init the mount struct |
| 632 | * - if we're a 32-bit kernel, do a size check on the superblock |
| 633 | * so we don't mount terabyte filesystems |
| 634 | * - init mount struct realtime fields |
| 635 | * - allocate inode hash table for fs |
| 636 | * - init directory manager |
| 637 | * - perform recovery and init the log manager |
| 638 | */ |
| 639 | int |
| 640 | xfs_mountfs( |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 641 | xfs_mount_t *mp) |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 642 | { |
| 643 | xfs_sb_t *sbp = &(mp->m_sb); |
| 644 | xfs_inode_t *rip; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 645 | __uint64_t resblks; |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 646 | uint quotamount = 0; |
| 647 | uint quotaflags = 0; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 648 | int error = 0; |
| 649 | |
Dave Chinner | ff55068 | 2013-08-12 20:49:41 +1000 | [diff] [blame] | 650 | xfs_sb_mount_common(mp, sbp); |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 651 | |
| 652 | /* |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 653 | * Check for a mismatched features2 values. Older kernels |
| 654 | * read & wrote into the wrong sb offset for sb_features2 |
| 655 | * on some platforms due to xfs_sb_t not being 64bit size aligned |
| 656 | * when sb_features2 was added, which made older superblock |
| 657 | * reading/writing routines swap it as a 64-bit value. |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 658 | * |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 659 | * For backwards compatibility, we make both slots equal. |
| 660 | * |
| 661 | * If we detect a mismatched field, we OR the set bits into the |
| 662 | * existing features2 field in case it has already been modified; we |
| 663 | * don't want to lose any features. We then update the bad location |
| 664 | * with the ORed value so that older kernels will see any features2 |
| 665 | * flags, and mark the two fields as needing updates once the |
| 666 | * transaction subsystem is online. |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 667 | */ |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 668 | if (xfs_sb_has_mismatched_features2(sbp)) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 669 | xfs_warn(mp, "correcting sb_features alignment problem"); |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 670 | sbp->sb_features2 |= sbp->sb_bad_features2; |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 671 | sbp->sb_bad_features2 = sbp->sb_features2; |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 672 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 673 | |
| 674 | /* |
| 675 | * Re-check for ATTR2 in case it was found in bad_features2 |
| 676 | * slot. |
| 677 | */ |
Tim Shimmin | 7c12f29 | 2008-04-30 18:15:28 +1000 | [diff] [blame] | 678 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
| 679 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 680 | mp->m_flags |= XFS_MOUNT_ATTR2; |
Tim Shimmin | 7c12f29 | 2008-04-30 18:15:28 +1000 | [diff] [blame] | 681 | } |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 682 | |
Tim Shimmin | 7c12f29 | 2008-04-30 18:15:28 +1000 | [diff] [blame] | 683 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
| 684 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { |
| 685 | xfs_sb_version_removeattr2(&mp->m_sb); |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 686 | mp->m_update_flags |= XFS_SB_FEATURES2; |
Tim Shimmin | 7c12f29 | 2008-04-30 18:15:28 +1000 | [diff] [blame] | 687 | |
| 688 | /* update sb_versionnum for the clearing of the morebits */ |
| 689 | if (!sbp->sb_features2) |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 690 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 691 | } |
| 692 | |
| 693 | /* |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 694 | * Check if sb_agblocks is aligned at stripe boundary |
| 695 | * If sb_agblocks is NOT aligned turn off m_dalign since |
| 696 | * allocator alignment is within an ag, therefore ag has |
| 697 | * to be aligned at stripe boundary. |
| 698 | */ |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 699 | error = xfs_update_alignment(mp); |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 700 | if (error) |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 701 | goto out; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 702 | |
| 703 | xfs_alloc_compute_maxlevels(mp); |
| 704 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); |
| 705 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); |
| 706 | xfs_ialloc_compute_maxlevels(mp); |
| 707 | |
| 708 | xfs_set_maxicount(mp); |
| 709 | |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 710 | error = xfs_uuid_mount(mp); |
| 711 | if (error) |
| 712 | goto out; |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 713 | |
| 714 | /* |
| 715 | * Set the minimum read and write sizes |
| 716 | */ |
| 717 | xfs_set_rw_sizes(mp); |
| 718 | |
Dave Chinner | 055388a | 2011-01-04 11:35:03 +1100 | [diff] [blame] | 719 | /* set the low space thresholds for dynamic preallocation */ |
| 720 | xfs_set_low_space_thresholds(mp); |
| 721 | |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 722 | /* |
| 723 | * Set the inode cluster size. |
| 724 | * This may still be overridden by the file system |
| 725 | * block size if it is larger than the chosen cluster size. |
| 726 | */ |
| 727 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; |
| 728 | |
| 729 | /* |
| 730 | * Set inode alignment fields |
| 731 | */ |
| 732 | xfs_set_inoalignment(mp); |
| 733 | |
| 734 | /* |
| 735 | * Check that the data (and log if separate) are an ok size. |
| 736 | */ |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 737 | error = xfs_check_sizes(mp); |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 738 | if (error) |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 739 | goto out_remove_uuid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | |
| 741 | /* |
| 742 | * Initialize realtime fields in the mount structure |
| 743 | */ |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 744 | error = xfs_rtmount_init(mp); |
| 745 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 746 | xfs_warn(mp, "RT mount failed"); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 747 | goto out_remove_uuid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | } |
| 749 | |
| 750 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | * Copies the low order bits of the timestamp and the randomly |
| 752 | * set "sequence" number out of a UUID. |
| 753 | */ |
| 754 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); |
| 755 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
| 757 | |
Nathan Scott | f6c2d1f | 2006-06-20 13:04:51 +1000 | [diff] [blame] | 758 | xfs_dir_mount(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | |
| 760 | /* |
| 761 | * Initialize the attribute manager's entries. |
| 762 | */ |
| 763 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; |
| 764 | |
| 765 | /* |
| 766 | * Initialize the precomputed transaction reservations values. |
| 767 | */ |
| 768 | xfs_trans_init(mp); |
| 769 | |
| 770 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | * Allocate and initialize the per-ag data. |
| 772 | */ |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 773 | spin_lock_init(&mp->m_perag_lock); |
Dave Chinner | 9b98b6f | 2010-05-27 01:58:13 +0000 | [diff] [blame] | 774 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 775 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
| 776 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 777 | xfs_warn(mp, "Failed per-ag init: %d", error); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 778 | goto out_remove_uuid; |
Dave Chinner | 1c1c6eb | 2010-01-11 11:47:44 +0000 | [diff] [blame] | 779 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 781 | if (!sbp->sb_logblocks) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 782 | xfs_warn(mp, "no log defined"); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 783 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
| 784 | error = XFS_ERROR(EFSCORRUPTED); |
| 785 | goto out_free_perag; |
| 786 | } |
| 787 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 788 | /* |
| 789 | * log's mount-time initialization. Perform 1st part recovery if needed |
| 790 | */ |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 791 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
| 792 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), |
| 793 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); |
| 794 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 795 | xfs_warn(mp, "log mount failed"); |
Dave Chinner | d4f3512 | 2012-04-23 15:59:06 +1000 | [diff] [blame] | 796 | goto out_fail_wait; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | } |
| 798 | |
| 799 | /* |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 800 | * Now the log is mounted, we know if it was an unclean shutdown or |
| 801 | * not. If it was, with the first phase of recovery has completed, we |
| 802 | * have consistent AG blocks on disk. We have not recovered EFIs yet, |
| 803 | * but they are recovered transactionally in the second recovery phase |
| 804 | * later. |
| 805 | * |
| 806 | * Hence we can safely re-initialise incore superblock counters from |
| 807 | * the per-ag data. These may not be correct if the filesystem was not |
| 808 | * cleanly unmounted, so we need to wait for recovery to finish before |
| 809 | * doing this. |
| 810 | * |
| 811 | * If the filesystem was cleanly unmounted, then we can trust the |
| 812 | * values in the superblock to be correct and we don't need to do |
| 813 | * anything here. |
| 814 | * |
| 815 | * If we are currently making the filesystem, the initialisation will |
| 816 | * fail as the perag data is in an undefined state. |
| 817 | */ |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 818 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
| 819 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && |
| 820 | !mp->m_sb.sb_inprogress) { |
| 821 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 822 | if (error) |
Dave Chinner | d4f3512 | 2012-04-23 15:59:06 +1000 | [diff] [blame] | 823 | goto out_fail_wait; |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 824 | } |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 825 | |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 826 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | * Get and sanity-check the root inode. |
| 828 | * Save the pointer to it in the mount structure. |
| 829 | */ |
Dave Chinner | 7b6259e | 2010-06-24 11:35:17 +1000 | [diff] [blame] | 830 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 832 | xfs_warn(mp, "failed to read root inode"); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 833 | goto out_log_dealloc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | } |
| 835 | |
| 836 | ASSERT(rip != NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 838 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 839 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
Nathan Scott | b657452 | 2006-06-09 15:29:40 +1000 | [diff] [blame] | 840 | (unsigned long long)rip->i_ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 841 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
| 842 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, |
| 843 | mp); |
| 844 | error = XFS_ERROR(EFSCORRUPTED); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 845 | goto out_rele_rip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | } |
| 847 | mp->m_rootip = rip; /* save it */ |
| 848 | |
| 849 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
| 850 | |
| 851 | /* |
| 852 | * Initialize realtime inode pointers in the mount structure |
| 853 | */ |
Eric Sandeen | 0771fb4 | 2007-10-12 11:03:40 +1000 | [diff] [blame] | 854 | error = xfs_rtmount_inodes(mp); |
| 855 | if (error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | /* |
| 857 | * Free up the root inode. |
| 858 | */ |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 859 | xfs_warn(mp, "failed to read RT inodes"); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 860 | goto out_rele_rip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | } |
| 862 | |
| 863 | /* |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 864 | * If this is a read-only mount defer the superblock updates until |
| 865 | * the next remount into writeable mode. Otherwise we would never |
| 866 | * perform the update e.g. for the root filesystem. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 867 | */ |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 868 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
| 869 | error = xfs_mount_log_sb(mp, mp->m_update_flags); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 870 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 871 | xfs_warn(mp, "failed to write sb changes"); |
Christoph Hellwig | b93b6e4 | 2009-02-04 09:33:58 +0100 | [diff] [blame] | 872 | goto out_rtunmount; |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 873 | } |
| 874 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | |
| 876 | /* |
| 877 | * Initialise the XFS quota management subsystem for this mount |
| 878 | */ |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 879 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
| 880 | error = xfs_qm_newmount(mp, "amount, "aflags); |
| 881 | if (error) |
| 882 | goto out_rtunmount; |
| 883 | } else { |
| 884 | ASSERT(!XFS_IS_QUOTA_ON(mp)); |
| 885 | |
| 886 | /* |
| 887 | * If a file system had quotas running earlier, but decided to |
| 888 | * mount without -o uquota/pquota/gquota options, revoke the |
| 889 | * quotachecked license. |
| 890 | */ |
| 891 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 892 | xfs_notice(mp, "resetting quota flags"); |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 893 | error = xfs_mount_reset_sbqflags(mp); |
| 894 | if (error) |
| 895 | return error; |
| 896 | } |
| 897 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | |
| 899 | /* |
| 900 | * Finish recovering the file system. This part needed to be |
| 901 | * delayed until after the root and real-time bitmap inodes |
| 902 | * were consistently read in. |
| 903 | */ |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 904 | error = xfs_log_mount_finish(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 906 | xfs_warn(mp, "log mount finish failed"); |
Christoph Hellwig | b93b6e4 | 2009-02-04 09:33:58 +0100 | [diff] [blame] | 907 | goto out_rtunmount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | } |
| 909 | |
| 910 | /* |
| 911 | * Complete the quota initialisation, post-log-replay component. |
| 912 | */ |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 913 | if (quotamount) { |
| 914 | ASSERT(mp->m_qflags == 0); |
| 915 | mp->m_qflags = quotaflags; |
| 916 | |
| 917 | xfs_qm_mount_quotas(mp); |
| 918 | } |
| 919 | |
David Chinner | 84e1e99 | 2007-06-18 16:50:27 +1000 | [diff] [blame] | 920 | /* |
| 921 | * Now we are mounted, reserve a small amount of unused space for |
| 922 | * privileged transactions. This is needed so that transaction |
| 923 | * space required for critical operations can dip into this pool |
| 924 | * when at ENOSPC. This is needed for operations like create with |
| 925 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations |
| 926 | * are not allowed to use this reserved space. |
Dave Chinner | 8babd8a | 2010-03-04 01:46:25 +0000 | [diff] [blame] | 927 | * |
| 928 | * This may drive us straight to ENOSPC on mount, but that implies |
| 929 | * we were already there on the last unmount. Warn if this occurs. |
David Chinner | 84e1e99 | 2007-06-18 16:50:27 +1000 | [diff] [blame] | 930 | */ |
Eric Sandeen | d5db0f9 | 2010-02-05 22:59:53 +0000 | [diff] [blame] | 931 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
| 932 | resblks = xfs_default_resblks(mp); |
| 933 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
| 934 | if (error) |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 935 | xfs_warn(mp, |
| 936 | "Unable to allocate reserve blocks. Continuing without reserve pool."); |
Eric Sandeen | d5db0f9 | 2010-02-05 22:59:53 +0000 | [diff] [blame] | 937 | } |
David Chinner | 84e1e99 | 2007-06-18 16:50:27 +1000 | [diff] [blame] | 938 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | return 0; |
| 940 | |
Christoph Hellwig | b93b6e4 | 2009-02-04 09:33:58 +0100 | [diff] [blame] | 941 | out_rtunmount: |
| 942 | xfs_rtunmount_inodes(mp); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 943 | out_rele_rip: |
Christoph Hellwig | 4335509 | 2008-03-27 18:01:08 +1100 | [diff] [blame] | 944 | IRELE(rip); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 945 | out_log_dealloc: |
Christoph Hellwig | 21b699c | 2009-03-16 08:19:29 +0100 | [diff] [blame] | 946 | xfs_log_unmount(mp); |
Dave Chinner | d4f3512 | 2012-04-23 15:59:06 +1000 | [diff] [blame] | 947 | out_fail_wait: |
| 948 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) |
| 949 | xfs_wait_buftarg(mp->m_logdev_targp); |
| 950 | xfs_wait_buftarg(mp->m_ddev_targp); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 951 | out_free_perag: |
Christoph Hellwig | ff4f038 | 2008-08-13 16:50:47 +1000 | [diff] [blame] | 952 | xfs_free_perag(mp); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 953 | out_remove_uuid: |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 954 | xfs_uuid_unmount(mp); |
Christoph Hellwig | f9057e3 | 2009-02-04 09:31:52 +0100 | [diff] [blame] | 955 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | return error; |
| 957 | } |
| 958 | |
| 959 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | * This flushes out the inodes,dquots and the superblock, unmounts the |
| 961 | * log and makes sure that incore structures are freed. |
| 962 | */ |
Christoph Hellwig | 41b5c2e | 2008-08-13 16:49:57 +1000 | [diff] [blame] | 963 | void |
| 964 | xfs_unmountfs( |
| 965 | struct xfs_mount *mp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | { |
Christoph Hellwig | 41b5c2e | 2008-08-13 16:49:57 +1000 | [diff] [blame] | 967 | __uint64_t resblks; |
| 968 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
Brian Foster | 579b62f | 2012-11-06 09:50:47 -0500 | [diff] [blame] | 970 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
| 971 | |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 972 | xfs_qm_unmount_quotas(mp); |
Christoph Hellwig | b93b6e4 | 2009-02-04 09:33:58 +0100 | [diff] [blame] | 973 | xfs_rtunmount_inodes(mp); |
Christoph Hellwig | 77508ec | 2008-08-13 16:49:04 +1000 | [diff] [blame] | 974 | IRELE(mp->m_rootip); |
| 975 | |
David Chinner | 641c56f | 2007-06-18 16:50:17 +1000 | [diff] [blame] | 976 | /* |
| 977 | * We can potentially deadlock here if we have an inode cluster |
Malcolm Parsons | 9da096f | 2009-03-29 09:55:42 +0200 | [diff] [blame] | 978 | * that has been freed has its buffer still pinned in memory because |
David Chinner | 641c56f | 2007-06-18 16:50:17 +1000 | [diff] [blame] | 979 | * the transaction is still sitting in a iclog. The stale inodes |
| 980 | * on that buffer will have their flush locks held until the |
| 981 | * transaction hits the disk and the callbacks run. the inode |
| 982 | * flush takes the flush lock unconditionally and with nothing to |
| 983 | * push out the iclog we will never get that unlocked. hence we |
| 984 | * need to force the log first. |
| 985 | */ |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 986 | xfs_log_force(mp, XFS_LOG_SYNC); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 987 | |
| 988 | /* |
Christoph Hellwig | 211e4d4 | 2012-04-23 15:58:34 +1000 | [diff] [blame] | 989 | * Flush all pending changes from the AIL. |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 990 | */ |
Christoph Hellwig | 211e4d4 | 2012-04-23 15:58:34 +1000 | [diff] [blame] | 991 | xfs_ail_push_all_sync(mp->m_ail); |
| 992 | |
| 993 | /* |
| 994 | * And reclaim all inodes. At this point there should be no dirty |
Dave Chinner | 7e18530 | 2012-10-08 21:56:00 +1100 | [diff] [blame] | 995 | * inodes and none should be pinned or locked, but use synchronous |
| 996 | * reclaim just to be sure. We can stop background inode reclaim |
| 997 | * here as well if it is still running. |
Christoph Hellwig | 211e4d4 | 2012-04-23 15:58:34 +1000 | [diff] [blame] | 998 | */ |
Dave Chinner | 7e18530 | 2012-10-08 21:56:00 +1100 | [diff] [blame] | 999 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 1000 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1001 | |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 1002 | xfs_qm_unmount(mp); |
Lachlan McIlroy | a357a12 | 2008-10-30 16:53:25 +1100 | [diff] [blame] | 1003 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1004 | /* |
David Chinner | 84e1e99 | 2007-06-18 16:50:27 +1000 | [diff] [blame] | 1005 | * Unreserve any blocks we have so that when we unmount we don't account |
| 1006 | * the reserved free space as used. This is really only necessary for |
| 1007 | * lazy superblock counting because it trusts the incore superblock |
Malcolm Parsons | 9da096f | 2009-03-29 09:55:42 +0200 | [diff] [blame] | 1008 | * counters to be absolutely correct on clean unmount. |
David Chinner | 84e1e99 | 2007-06-18 16:50:27 +1000 | [diff] [blame] | 1009 | * |
| 1010 | * We don't bother correcting this elsewhere for lazy superblock |
| 1011 | * counting because on mount of an unclean filesystem we reconstruct the |
| 1012 | * correct counter value and this is irrelevant. |
| 1013 | * |
| 1014 | * For non-lazy counter filesystems, this doesn't matter at all because |
| 1015 | * we only every apply deltas to the superblock and hence the incore |
| 1016 | * value does not matter.... |
| 1017 | */ |
| 1018 | resblks = 0; |
David Chinner | 714082b | 2008-04-10 12:20:03 +1000 | [diff] [blame] | 1019 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
| 1020 | if (error) |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 1021 | xfs_warn(mp, "Unable to free reserved block pool. " |
David Chinner | 714082b | 2008-04-10 12:20:03 +1000 | [diff] [blame] | 1022 | "Freespace may not be correct on next mount."); |
| 1023 | |
Chandra Seetharaman | adab0f6 | 2011-06-29 22:10:14 +0000 | [diff] [blame] | 1024 | error = xfs_log_sbcount(mp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1025 | if (error) |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 1026 | xfs_warn(mp, "Unable to update superblock counters. " |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1027 | "Freespace may not be correct on next mount."); |
Christoph Hellwig | 87c7bec | 2011-09-14 14:08:26 +0000 | [diff] [blame] | 1028 | |
Christoph Hellwig | 21b699c | 2009-03-16 08:19:29 +0100 | [diff] [blame] | 1029 | xfs_log_unmount(mp); |
Christoph Hellwig | 2717420 | 2009-03-30 10:21:31 +0200 | [diff] [blame] | 1030 | xfs_uuid_unmount(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | |
Christoph Hellwig | 1550d0b | 2008-08-13 16:17:37 +1000 | [diff] [blame] | 1032 | #if defined(DEBUG) |
Christoph Hellwig | 0ce4cfd | 2007-08-30 17:20:53 +1000 | [diff] [blame] | 1033 | xfs_errortag_clearall(mp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | #endif |
Christoph Hellwig | ff4f038 | 2008-08-13 16:50:47 +1000 | [diff] [blame] | 1035 | xfs_free_perag(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | } |
| 1037 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | int |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1039 | xfs_fs_writable(xfs_mount_t *mp) |
| 1040 | { |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 1041 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
Christoph Hellwig | bd186aa | 2007-08-30 17:21:12 +1000 | [diff] [blame] | 1042 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1043 | } |
| 1044 | |
| 1045 | /* |
Alex Elder | b2ce397 | 2011-07-11 09:51:44 -0500 | [diff] [blame] | 1046 | * xfs_log_sbcount |
| 1047 | * |
Chandra Seetharaman | adab0f6 | 2011-06-29 22:10:14 +0000 | [diff] [blame] | 1048 | * Sync the superblock counters to disk. |
Alex Elder | b2ce397 | 2011-07-11 09:51:44 -0500 | [diff] [blame] | 1049 | * |
| 1050 | * Note this code can be called during the process of freezing, so |
Chandra Seetharaman | adab0f6 | 2011-06-29 22:10:14 +0000 | [diff] [blame] | 1051 | * we may need to use the transaction allocator which does not |
Alex Elder | b2ce397 | 2011-07-11 09:51:44 -0500 | [diff] [blame] | 1052 | * block when the transaction subsystem is in its frozen state. |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1053 | */ |
| 1054 | int |
Chandra Seetharaman | adab0f6 | 2011-06-29 22:10:14 +0000 | [diff] [blame] | 1055 | xfs_log_sbcount(xfs_mount_t *mp) |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1056 | { |
| 1057 | xfs_trans_t *tp; |
| 1058 | int error; |
| 1059 | |
| 1060 | if (!xfs_fs_writable(mp)) |
| 1061 | return 0; |
| 1062 | |
Christoph Hellwig | d4d90b5 | 2008-04-22 17:34:37 +1000 | [diff] [blame] | 1063 | xfs_icsb_sync_counters(mp, 0); |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1064 | |
| 1065 | /* |
| 1066 | * we don't need to do this if we are updating the superblock |
| 1067 | * counters on every modification. |
| 1068 | */ |
| 1069 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) |
| 1070 | return 0; |
| 1071 | |
Alex Elder | b2ce397 | 2011-07-11 09:51:44 -0500 | [diff] [blame] | 1072 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1073 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1074 | if (error) { |
| 1075 | xfs_trans_cancel(tp, 0); |
| 1076 | return error; |
| 1077 | } |
| 1078 | |
| 1079 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); |
Chandra Seetharaman | adab0f6 | 2011-06-29 22:10:14 +0000 | [diff] [blame] | 1080 | xfs_trans_set_sync(tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1081 | error = xfs_trans_commit(tp, 0); |
| 1082 | return error; |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1083 | } |
| 1084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply |
| 1087 | * a delta to a specified field in the in-core superblock. Simply |
| 1088 | * switch on the field indicated and apply the delta to that field. |
| 1089 | * Fields are not allowed to dip below zero, so if the delta would |
| 1090 | * do this do not apply it and return EINVAL. |
| 1091 | * |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1092 | * The m_sb_lock must be held when this routine is called. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1093 | */ |
Eric Sandeen | d96f8f8 | 2009-07-02 00:09:33 -0500 | [diff] [blame] | 1094 | STATIC int |
David Chinner | 20f4ebf | 2007-02-10 18:36:10 +1100 | [diff] [blame] | 1095 | xfs_mod_incore_sb_unlocked( |
| 1096 | xfs_mount_t *mp, |
| 1097 | xfs_sb_field_t field, |
| 1098 | int64_t delta, |
| 1099 | int rsvd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | { |
| 1101 | int scounter; /* short counter for 32 bit fields */ |
| 1102 | long long lcounter; /* long counter for 64 bit fields */ |
| 1103 | long long res_used, rem; |
| 1104 | |
| 1105 | /* |
| 1106 | * With the in-core superblock spin lock held, switch |
| 1107 | * on the indicated field. Apply the delta to the |
| 1108 | * proper field. If the fields value would dip below |
| 1109 | * 0, then do not apply the delta and return EINVAL. |
| 1110 | */ |
| 1111 | switch (field) { |
| 1112 | case XFS_SBS_ICOUNT: |
| 1113 | lcounter = (long long)mp->m_sb.sb_icount; |
| 1114 | lcounter += delta; |
| 1115 | if (lcounter < 0) { |
| 1116 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1117 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | } |
| 1119 | mp->m_sb.sb_icount = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1120 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | case XFS_SBS_IFREE: |
| 1122 | lcounter = (long long)mp->m_sb.sb_ifree; |
| 1123 | lcounter += delta; |
| 1124 | if (lcounter < 0) { |
| 1125 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1126 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | } |
| 1128 | mp->m_sb.sb_ifree = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1129 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | case XFS_SBS_FDBLOCKS: |
David Chinner | 4be536d | 2006-09-07 14:26:50 +1000 | [diff] [blame] | 1131 | lcounter = (long long) |
| 1132 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
| 1134 | |
| 1135 | if (delta > 0) { /* Putting blocks back */ |
| 1136 | if (res_used > delta) { |
| 1137 | mp->m_resblks_avail += delta; |
| 1138 | } else { |
| 1139 | rem = delta - res_used; |
| 1140 | mp->m_resblks_avail = mp->m_resblks; |
| 1141 | lcounter += rem; |
| 1142 | } |
| 1143 | } else { /* Taking blocks away */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | lcounter += delta; |
Dave Chinner | 8babd8a | 2010-03-04 01:46:25 +0000 | [diff] [blame] | 1145 | if (lcounter >= 0) { |
| 1146 | mp->m_sb.sb_fdblocks = lcounter + |
| 1147 | XFS_ALLOC_SET_ASIDE(mp); |
| 1148 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | } |
Dave Chinner | 8babd8a | 2010-03-04 01:46:25 +0000 | [diff] [blame] | 1150 | |
| 1151 | /* |
| 1152 | * We are out of blocks, use any available reserved |
| 1153 | * blocks if were allowed to. |
| 1154 | */ |
| 1155 | if (!rsvd) |
| 1156 | return XFS_ERROR(ENOSPC); |
| 1157 | |
| 1158 | lcounter = (long long)mp->m_resblks_avail + delta; |
| 1159 | if (lcounter >= 0) { |
| 1160 | mp->m_resblks_avail = lcounter; |
| 1161 | return 0; |
| 1162 | } |
| 1163 | printk_once(KERN_WARNING |
| 1164 | "Filesystem \"%s\": reserve blocks depleted! " |
| 1165 | "Consider increasing reserve pool size.", |
| 1166 | mp->m_fsname); |
| 1167 | return XFS_ERROR(ENOSPC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1168 | } |
| 1169 | |
David Chinner | 4be536d | 2006-09-07 14:26:50 +1000 | [diff] [blame] | 1170 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1171 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | case XFS_SBS_FREXTENTS: |
| 1173 | lcounter = (long long)mp->m_sb.sb_frextents; |
| 1174 | lcounter += delta; |
| 1175 | if (lcounter < 0) { |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1176 | return XFS_ERROR(ENOSPC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | } |
| 1178 | mp->m_sb.sb_frextents = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1179 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1180 | case XFS_SBS_DBLOCKS: |
| 1181 | lcounter = (long long)mp->m_sb.sb_dblocks; |
| 1182 | lcounter += delta; |
| 1183 | if (lcounter < 0) { |
| 1184 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1185 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | } |
| 1187 | mp->m_sb.sb_dblocks = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1188 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1189 | case XFS_SBS_AGCOUNT: |
| 1190 | scounter = mp->m_sb.sb_agcount; |
| 1191 | scounter += delta; |
| 1192 | if (scounter < 0) { |
| 1193 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1194 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | } |
| 1196 | mp->m_sb.sb_agcount = scounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1197 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | case XFS_SBS_IMAX_PCT: |
| 1199 | scounter = mp->m_sb.sb_imax_pct; |
| 1200 | scounter += delta; |
| 1201 | if (scounter < 0) { |
| 1202 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1203 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1204 | } |
| 1205 | mp->m_sb.sb_imax_pct = scounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1206 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | case XFS_SBS_REXTSIZE: |
| 1208 | scounter = mp->m_sb.sb_rextsize; |
| 1209 | scounter += delta; |
| 1210 | if (scounter < 0) { |
| 1211 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1212 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1213 | } |
| 1214 | mp->m_sb.sb_rextsize = scounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1215 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | case XFS_SBS_RBMBLOCKS: |
| 1217 | scounter = mp->m_sb.sb_rbmblocks; |
| 1218 | scounter += delta; |
| 1219 | if (scounter < 0) { |
| 1220 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1221 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | } |
| 1223 | mp->m_sb.sb_rbmblocks = scounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1224 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1225 | case XFS_SBS_RBLOCKS: |
| 1226 | lcounter = (long long)mp->m_sb.sb_rblocks; |
| 1227 | lcounter += delta; |
| 1228 | if (lcounter < 0) { |
| 1229 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1230 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1231 | } |
| 1232 | mp->m_sb.sb_rblocks = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1233 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1234 | case XFS_SBS_REXTENTS: |
| 1235 | lcounter = (long long)mp->m_sb.sb_rextents; |
| 1236 | lcounter += delta; |
| 1237 | if (lcounter < 0) { |
| 1238 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1239 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1240 | } |
| 1241 | mp->m_sb.sb_rextents = lcounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1242 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | case XFS_SBS_REXTSLOG: |
| 1244 | scounter = mp->m_sb.sb_rextslog; |
| 1245 | scounter += delta; |
| 1246 | if (scounter < 0) { |
| 1247 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1248 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | } |
| 1250 | mp->m_sb.sb_rextslog = scounter; |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1251 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | default: |
| 1253 | ASSERT(0); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1254 | return XFS_ERROR(EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | } |
| 1256 | } |
| 1257 | |
| 1258 | /* |
| 1259 | * xfs_mod_incore_sb() is used to change a field in the in-core |
| 1260 | * superblock structure by the specified delta. This modification |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1261 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1262 | * routine to do the work. |
| 1263 | */ |
| 1264 | int |
David Chinner | 20f4ebf | 2007-02-10 18:36:10 +1100 | [diff] [blame] | 1265 | xfs_mod_incore_sb( |
Christoph Hellwig | 96540c7 | 2010-09-30 02:25:55 +0000 | [diff] [blame] | 1266 | struct xfs_mount *mp, |
| 1267 | xfs_sb_field_t field, |
| 1268 | int64_t delta, |
| 1269 | int rsvd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | { |
Christoph Hellwig | 96540c7 | 2010-09-30 02:25:55 +0000 | [diff] [blame] | 1271 | int status; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1272 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1273 | #ifdef HAVE_PERCPU_SB |
Christoph Hellwig | 96540c7 | 2010-09-30 02:25:55 +0000 | [diff] [blame] | 1274 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1275 | #endif |
Christoph Hellwig | 96540c7 | 2010-09-30 02:25:55 +0000 | [diff] [blame] | 1276 | spin_lock(&mp->m_sb_lock); |
| 1277 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
| 1278 | spin_unlock(&mp->m_sb_lock); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1279 | |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1280 | return status; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | } |
| 1282 | |
| 1283 | /* |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1284 | * Change more than one field in the in-core superblock structure at a time. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | * |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1286 | * The fields and changes to those fields are specified in the array of |
| 1287 | * xfs_mod_sb structures passed in. Either all of the specified deltas |
| 1288 | * will be applied or none of them will. If any modified field dips below 0, |
| 1289 | * then all modifications will be backed out and EINVAL will be returned. |
| 1290 | * |
| 1291 | * Note that this function may not be used for the superblock values that |
| 1292 | * are tracked with the in-memory per-cpu counters - a direct call to |
| 1293 | * xfs_icsb_modify_counters is required for these. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | */ |
| 1295 | int |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1296 | xfs_mod_incore_sb_batch( |
| 1297 | struct xfs_mount *mp, |
| 1298 | xfs_mod_sb_t *msb, |
| 1299 | uint nmsb, |
| 1300 | int rsvd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1301 | { |
David Sterba | 45c51b9 | 2011-04-13 22:03:28 +0000 | [diff] [blame] | 1302 | xfs_mod_sb_t *msbp; |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1303 | int error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | |
| 1305 | /* |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1306 | * Loop through the array of mod structures and apply each individually. |
| 1307 | * If any fail, then back out all those which have already been applied. |
| 1308 | * Do all of this within the scope of the m_sb_lock so that all of the |
| 1309 | * changes will be atomic. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | */ |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1311 | spin_lock(&mp->m_sb_lock); |
David Sterba | 45c51b9 | 2011-04-13 22:03:28 +0000 | [diff] [blame] | 1312 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1313 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
| 1314 | msbp->msb_field > XFS_SBS_FDBLOCKS); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1315 | |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1316 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
| 1317 | msbp->msb_delta, rsvd); |
| 1318 | if (error) |
| 1319 | goto unwind; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | } |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1321 | spin_unlock(&mp->m_sb_lock); |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1322 | return 0; |
| 1323 | |
| 1324 | unwind: |
| 1325 | while (--msbp >= msb) { |
| 1326 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
| 1327 | -msbp->msb_delta, rsvd); |
| 1328 | ASSERT(error == 0); |
| 1329 | } |
| 1330 | spin_unlock(&mp->m_sb_lock); |
| 1331 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1332 | } |
| 1333 | |
| 1334 | /* |
| 1335 | * xfs_getsb() is called to obtain the buffer for the superblock. |
| 1336 | * The buffer is returned locked and read in from disk. |
| 1337 | * The buffer should be released with a call to xfs_brelse(). |
| 1338 | * |
| 1339 | * If the flags parameter is BUF_TRYLOCK, then we'll only return |
| 1340 | * the superblock buffer if it can be locked without sleeping. |
| 1341 | * If it can't then we'll return NULL. |
| 1342 | */ |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1343 | struct xfs_buf * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | xfs_getsb( |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1345 | struct xfs_mount *mp, |
| 1346 | int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | { |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1348 | struct xfs_buf *bp = mp->m_sb_bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1350 | if (!xfs_buf_trylock(bp)) { |
| 1351 | if (flags & XBF_TRYLOCK) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | return NULL; |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1353 | xfs_buf_lock(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | } |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 1355 | |
Chandra Seetharaman | 72790aa | 2011-07-22 23:40:04 +0000 | [diff] [blame] | 1356 | xfs_buf_hold(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | ASSERT(XFS_BUF_ISDONE(bp)); |
Jesper Juhl | 014c254 | 2006-01-15 02:37:08 +0100 | [diff] [blame] | 1358 | return bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | } |
| 1360 | |
| 1361 | /* |
| 1362 | * Used to free the superblock along various error paths. |
| 1363 | */ |
| 1364 | void |
| 1365 | xfs_freesb( |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 1366 | struct xfs_mount *mp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1367 | { |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 1368 | struct xfs_buf *bp = mp->m_sb_bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 1370 | xfs_buf_lock(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1371 | mp->m_sb_bp = NULL; |
Dave Chinner | 26af655 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 1372 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | } |
| 1374 | |
| 1375 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | * Used to log changes to the superblock unit and width fields which could |
Eric Sandeen | e6957ea | 2008-04-10 12:19:34 +1000 | [diff] [blame] | 1377 | * be altered by the mount options, as well as any potential sb_features2 |
| 1378 | * fixup. Only the first superblock is updated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1379 | */ |
Christoph Hellwig | 7884bc8 | 2009-01-19 02:04:07 +0100 | [diff] [blame] | 1380 | int |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 1381 | xfs_mount_log_sb( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | xfs_mount_t *mp, |
| 1383 | __int64_t fields) |
| 1384 | { |
| 1385 | xfs_trans_t *tp; |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1386 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | |
David Chinner | ee1c090 | 2008-03-06 13:45:50 +1100 | [diff] [blame] | 1388 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
David Chinner | 4b166de | 2008-05-20 11:30:27 +1000 | [diff] [blame] | 1389 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
| 1390 | XFS_SB_VERSIONNUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | |
| 1392 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1393 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1394 | if (error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | xfs_trans_cancel(tp, 0); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1396 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | } |
| 1398 | xfs_mod_sb(tp, fields); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1399 | error = xfs_trans_commit(tp, 0); |
| 1400 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | } |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1402 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 1403 | /* |
| 1404 | * If the underlying (data/log/rt) device is readonly, there are some |
| 1405 | * operations that cannot proceed. |
| 1406 | */ |
| 1407 | int |
| 1408 | xfs_dev_is_read_only( |
| 1409 | struct xfs_mount *mp, |
| 1410 | char *message) |
| 1411 | { |
| 1412 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || |
| 1413 | xfs_readonly_buftarg(mp->m_logdev_targp) || |
| 1414 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 1415 | xfs_notice(mp, "%s required on read-only device.", message); |
| 1416 | xfs_notice(mp, "write access unavailable, cannot proceed."); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 1417 | return EROFS; |
| 1418 | } |
| 1419 | return 0; |
| 1420 | } |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1421 | |
| 1422 | #ifdef HAVE_PERCPU_SB |
| 1423 | /* |
| 1424 | * Per-cpu incore superblock counters |
| 1425 | * |
| 1426 | * Simple concept, difficult implementation |
| 1427 | * |
| 1428 | * Basically, replace the incore superblock counters with a distributed per cpu |
| 1429 | * counter for contended fields (e.g. free block count). |
| 1430 | * |
| 1431 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and |
| 1432 | * hence needs to be accurately read when we are running low on space. Hence |
| 1433 | * there is a method to enable and disable the per-cpu counters based on how |
| 1434 | * much "stuff" is available in them. |
| 1435 | * |
| 1436 | * Basically, a counter is enabled if there is enough free resource to justify |
| 1437 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local |
| 1438 | * ENOSPC), then we disable the counters to synchronise all callers and |
| 1439 | * re-distribute the available resources. |
| 1440 | * |
| 1441 | * If, once we redistributed the available resources, we still get a failure, |
| 1442 | * we disable the per-cpu counter and go through the slow path. |
| 1443 | * |
| 1444 | * The slow path is the current xfs_mod_incore_sb() function. This means that |
Malcolm Parsons | 9da096f | 2009-03-29 09:55:42 +0200 | [diff] [blame] | 1445 | * when we disable a per-cpu counter, we need to drain its resources back to |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1446 | * the global superblock. We do this after disabling the counter to prevent |
| 1447 | * more threads from queueing up on the counter. |
| 1448 | * |
| 1449 | * Essentially, this means that we still need a lock in the fast path to enable |
| 1450 | * synchronisation between the global counters and the per-cpu counters. This |
| 1451 | * is not a problem because the lock will be local to a CPU almost all the time |
| 1452 | * and have little contention except when we get to ENOSPC conditions. |
| 1453 | * |
| 1454 | * Basically, this lock becomes a barrier that enables us to lock out the fast |
| 1455 | * path while we do things like enabling and disabling counters and |
| 1456 | * synchronising the counters. |
| 1457 | * |
| 1458 | * Locking rules: |
| 1459 | * |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1460 | * 1. m_sb_lock before picking up per-cpu locks |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1461 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1462 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1463 | * 4. modifying per-cpu counters requires holding per-cpu lock |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1464 | * 5. modifying global counters requires holding m_sb_lock |
| 1465 | * 6. enabling or disabling a counter requires holding the m_sb_lock |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1466 | * and _none_ of the per-cpu locks. |
| 1467 | * |
| 1468 | * Disabled counters are only ever re-enabled by a balance operation |
| 1469 | * that results in more free resources per CPU than a given threshold. |
| 1470 | * To ensure counters don't remain disabled, they are rebalanced when |
| 1471 | * the global resource goes above a higher threshold (i.e. some hysteresis |
| 1472 | * is present to prevent thrashing). |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1473 | */ |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1474 | |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1475 | #ifdef CONFIG_HOTPLUG_CPU |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1476 | /* |
| 1477 | * hot-plug CPU notifier support. |
| 1478 | * |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1479 | * We need a notifier per filesystem as we need to be able to identify |
| 1480 | * the filesystem to balance the counters out. This is achieved by |
| 1481 | * having a notifier block embedded in the xfs_mount_t and doing pointer |
| 1482 | * magic to get the mount pointer from the notifier block address. |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1483 | */ |
| 1484 | STATIC int |
| 1485 | xfs_icsb_cpu_notify( |
| 1486 | struct notifier_block *nfb, |
| 1487 | unsigned long action, |
| 1488 | void *hcpu) |
| 1489 | { |
| 1490 | xfs_icsb_cnts_t *cntp; |
| 1491 | xfs_mount_t *mp; |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1492 | |
| 1493 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); |
| 1494 | cntp = (xfs_icsb_cnts_t *) |
| 1495 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); |
| 1496 | switch (action) { |
| 1497 | case CPU_UP_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1498 | case CPU_UP_PREPARE_FROZEN: |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1499 | /* Easy Case - initialize the area and locks, and |
| 1500 | * then rebalance when online does everything else for us. */ |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1501 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1502 | break; |
| 1503 | case CPU_ONLINE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1504 | case CPU_ONLINE_FROZEN: |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1505 | xfs_icsb_lock(mp); |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1506 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
| 1507 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); |
| 1508 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1509 | xfs_icsb_unlock(mp); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1510 | break; |
| 1511 | case CPU_DEAD: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1512 | case CPU_DEAD_FROZEN: |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1513 | /* Disable all the counters, then fold the dead cpu's |
| 1514 | * count into the total on the global superblock and |
| 1515 | * re-enable the counters. */ |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1516 | xfs_icsb_lock(mp); |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1517 | spin_lock(&mp->m_sb_lock); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1518 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
| 1519 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); |
| 1520 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); |
| 1521 | |
| 1522 | mp->m_sb.sb_icount += cntp->icsb_icount; |
| 1523 | mp->m_sb.sb_ifree += cntp->icsb_ifree; |
| 1524 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; |
| 1525 | |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1526 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1527 | |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1528 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
| 1529 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); |
| 1530 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1531 | spin_unlock(&mp->m_sb_lock); |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1532 | xfs_icsb_unlock(mp); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1533 | break; |
| 1534 | } |
| 1535 | |
| 1536 | return NOTIFY_OK; |
| 1537 | } |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1538 | #endif /* CONFIG_HOTPLUG_CPU */ |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1539 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1540 | int |
| 1541 | xfs_icsb_init_counters( |
| 1542 | xfs_mount_t *mp) |
| 1543 | { |
| 1544 | xfs_icsb_cnts_t *cntp; |
| 1545 | int i; |
| 1546 | |
| 1547 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); |
| 1548 | if (mp->m_sb_cnts == NULL) |
| 1549 | return -ENOMEM; |
| 1550 | |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1551 | #ifdef CONFIG_HOTPLUG_CPU |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1552 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
| 1553 | mp->m_icsb_notifier.priority = 0; |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1554 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
| 1555 | #endif /* CONFIG_HOTPLUG_CPU */ |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1556 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1557 | for_each_online_cpu(i) { |
| 1558 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1559 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1560 | } |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1561 | |
| 1562 | mutex_init(&mp->m_icsb_mutex); |
| 1563 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1564 | /* |
| 1565 | * start with all counters disabled so that the |
| 1566 | * initial balance kicks us off correctly |
| 1567 | */ |
| 1568 | mp->m_icsb_counters = -1; |
| 1569 | return 0; |
| 1570 | } |
| 1571 | |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 1572 | void |
| 1573 | xfs_icsb_reinit_counters( |
| 1574 | xfs_mount_t *mp) |
| 1575 | { |
| 1576 | xfs_icsb_lock(mp); |
| 1577 | /* |
| 1578 | * start with all counters disabled so that the |
| 1579 | * initial balance kicks us off correctly |
| 1580 | */ |
| 1581 | mp->m_icsb_counters = -1; |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1582 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
| 1583 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); |
| 1584 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 1585 | xfs_icsb_unlock(mp); |
| 1586 | } |
| 1587 | |
Christoph Hellwig | c962fb7 | 2008-05-20 15:10:52 +1000 | [diff] [blame] | 1588 | void |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1589 | xfs_icsb_destroy_counters( |
| 1590 | xfs_mount_t *mp) |
| 1591 | { |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1592 | if (mp->m_sb_cnts) { |
Chandra Seetharaman | 5a67e4c | 2006-06-27 02:54:11 -0700 | [diff] [blame] | 1593 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1594 | free_percpu(mp->m_sb_cnts); |
David Chinner | e8234a6 | 2006-03-14 13:23:52 +1100 | [diff] [blame] | 1595 | } |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1596 | mutex_destroy(&mp->m_icsb_mutex); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1597 | } |
| 1598 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 1599 | STATIC void |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1600 | xfs_icsb_lock_cntr( |
| 1601 | xfs_icsb_cnts_t *icsbp) |
| 1602 | { |
| 1603 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { |
| 1604 | ndelay(1000); |
| 1605 | } |
| 1606 | } |
| 1607 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 1608 | STATIC void |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1609 | xfs_icsb_unlock_cntr( |
| 1610 | xfs_icsb_cnts_t *icsbp) |
| 1611 | { |
| 1612 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); |
| 1613 | } |
| 1614 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1615 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 1616 | STATIC void |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1617 | xfs_icsb_lock_all_counters( |
| 1618 | xfs_mount_t *mp) |
| 1619 | { |
| 1620 | xfs_icsb_cnts_t *cntp; |
| 1621 | int i; |
| 1622 | |
| 1623 | for_each_online_cpu(i) { |
| 1624 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1625 | xfs_icsb_lock_cntr(cntp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1626 | } |
| 1627 | } |
| 1628 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 1629 | STATIC void |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1630 | xfs_icsb_unlock_all_counters( |
| 1631 | xfs_mount_t *mp) |
| 1632 | { |
| 1633 | xfs_icsb_cnts_t *cntp; |
| 1634 | int i; |
| 1635 | |
| 1636 | for_each_online_cpu(i) { |
| 1637 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1638 | xfs_icsb_unlock_cntr(cntp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1639 | } |
| 1640 | } |
| 1641 | |
| 1642 | STATIC void |
| 1643 | xfs_icsb_count( |
| 1644 | xfs_mount_t *mp, |
| 1645 | xfs_icsb_cnts_t *cnt, |
| 1646 | int flags) |
| 1647 | { |
| 1648 | xfs_icsb_cnts_t *cntp; |
| 1649 | int i; |
| 1650 | |
| 1651 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); |
| 1652 | |
| 1653 | if (!(flags & XFS_ICSB_LAZY_COUNT)) |
| 1654 | xfs_icsb_lock_all_counters(mp); |
| 1655 | |
| 1656 | for_each_online_cpu(i) { |
| 1657 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); |
| 1658 | cnt->icsb_icount += cntp->icsb_icount; |
| 1659 | cnt->icsb_ifree += cntp->icsb_ifree; |
| 1660 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; |
| 1661 | } |
| 1662 | |
| 1663 | if (!(flags & XFS_ICSB_LAZY_COUNT)) |
| 1664 | xfs_icsb_unlock_all_counters(mp); |
| 1665 | } |
| 1666 | |
| 1667 | STATIC int |
| 1668 | xfs_icsb_counter_disabled( |
| 1669 | xfs_mount_t *mp, |
| 1670 | xfs_sb_field_t field) |
| 1671 | { |
| 1672 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); |
| 1673 | return test_bit(field, &mp->m_icsb_counters); |
| 1674 | } |
| 1675 | |
David Chinner | 36fbe6e | 2008-04-10 12:19:56 +1000 | [diff] [blame] | 1676 | STATIC void |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1677 | xfs_icsb_disable_counter( |
| 1678 | xfs_mount_t *mp, |
| 1679 | xfs_sb_field_t field) |
| 1680 | { |
| 1681 | xfs_icsb_cnts_t cnt; |
| 1682 | |
| 1683 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); |
| 1684 | |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1685 | /* |
| 1686 | * If we are already disabled, then there is nothing to do |
| 1687 | * here. We check before locking all the counters to avoid |
| 1688 | * the expensive lock operation when being called in the |
| 1689 | * slow path and the counter is already disabled. This is |
| 1690 | * safe because the only time we set or clear this state is under |
| 1691 | * the m_icsb_mutex. |
| 1692 | */ |
| 1693 | if (xfs_icsb_counter_disabled(mp, field)) |
David Chinner | 36fbe6e | 2008-04-10 12:19:56 +1000 | [diff] [blame] | 1694 | return; |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1695 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1696 | xfs_icsb_lock_all_counters(mp); |
| 1697 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { |
| 1698 | /* drain back to superblock */ |
| 1699 | |
Christoph Hellwig | ce46193 | 2008-04-22 17:34:50 +1000 | [diff] [blame] | 1700 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1701 | switch(field) { |
| 1702 | case XFS_SBS_ICOUNT: |
| 1703 | mp->m_sb.sb_icount = cnt.icsb_icount; |
| 1704 | break; |
| 1705 | case XFS_SBS_IFREE: |
| 1706 | mp->m_sb.sb_ifree = cnt.icsb_ifree; |
| 1707 | break; |
| 1708 | case XFS_SBS_FDBLOCKS: |
| 1709 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; |
| 1710 | break; |
| 1711 | default: |
| 1712 | BUG(); |
| 1713 | } |
| 1714 | } |
| 1715 | |
| 1716 | xfs_icsb_unlock_all_counters(mp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1717 | } |
| 1718 | |
| 1719 | STATIC void |
| 1720 | xfs_icsb_enable_counter( |
| 1721 | xfs_mount_t *mp, |
| 1722 | xfs_sb_field_t field, |
| 1723 | uint64_t count, |
| 1724 | uint64_t resid) |
| 1725 | { |
| 1726 | xfs_icsb_cnts_t *cntp; |
| 1727 | int i; |
| 1728 | |
| 1729 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); |
| 1730 | |
| 1731 | xfs_icsb_lock_all_counters(mp); |
| 1732 | for_each_online_cpu(i) { |
| 1733 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); |
| 1734 | switch (field) { |
| 1735 | case XFS_SBS_ICOUNT: |
| 1736 | cntp->icsb_icount = count + resid; |
| 1737 | break; |
| 1738 | case XFS_SBS_IFREE: |
| 1739 | cntp->icsb_ifree = count + resid; |
| 1740 | break; |
| 1741 | case XFS_SBS_FDBLOCKS: |
| 1742 | cntp->icsb_fdblocks = count + resid; |
| 1743 | break; |
| 1744 | default: |
| 1745 | BUG(); |
| 1746 | break; |
| 1747 | } |
| 1748 | resid = 0; |
| 1749 | } |
| 1750 | clear_bit(field, &mp->m_icsb_counters); |
| 1751 | xfs_icsb_unlock_all_counters(mp); |
| 1752 | } |
| 1753 | |
David Chinner | dbcabad | 2007-02-10 18:36:17 +1100 | [diff] [blame] | 1754 | void |
Christoph Hellwig | d4d90b5 | 2008-04-22 17:34:37 +1000 | [diff] [blame] | 1755 | xfs_icsb_sync_counters_locked( |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1756 | xfs_mount_t *mp, |
| 1757 | int flags) |
| 1758 | { |
| 1759 | xfs_icsb_cnts_t cnt; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1760 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1761 | xfs_icsb_count(mp, &cnt, flags); |
| 1762 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1763 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
| 1764 | mp->m_sb.sb_icount = cnt.icsb_icount; |
| 1765 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) |
| 1766 | mp->m_sb.sb_ifree = cnt.icsb_ifree; |
| 1767 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) |
| 1768 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1769 | } |
| 1770 | |
| 1771 | /* |
| 1772 | * Accurate update of per-cpu counters to incore superblock |
| 1773 | */ |
Christoph Hellwig | d4d90b5 | 2008-04-22 17:34:37 +1000 | [diff] [blame] | 1774 | void |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1775 | xfs_icsb_sync_counters( |
Christoph Hellwig | d4d90b5 | 2008-04-22 17:34:37 +1000 | [diff] [blame] | 1776 | xfs_mount_t *mp, |
| 1777 | int flags) |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1778 | { |
Christoph Hellwig | d4d90b5 | 2008-04-22 17:34:37 +1000 | [diff] [blame] | 1779 | spin_lock(&mp->m_sb_lock); |
| 1780 | xfs_icsb_sync_counters_locked(mp, flags); |
| 1781 | spin_unlock(&mp->m_sb_lock); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1782 | } |
| 1783 | |
| 1784 | /* |
| 1785 | * Balance and enable/disable counters as necessary. |
| 1786 | * |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1787 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
| 1788 | * chosen to be the same number as single on disk allocation chunk per CPU, and |
| 1789 | * free blocks is something far enough zero that we aren't going thrash when we |
| 1790 | * get near ENOSPC. We also need to supply a minimum we require per cpu to |
| 1791 | * prevent looping endlessly when xfs_alloc_space asks for more than will |
| 1792 | * be distributed to a single CPU but each CPU has enough blocks to be |
| 1793 | * reenabled. |
| 1794 | * |
| 1795 | * Note that we can be called when counters are already disabled. |
| 1796 | * xfs_icsb_disable_counter() optimises the counter locking in this case to |
| 1797 | * prevent locking every per-cpu counter needlessly. |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1798 | */ |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1799 | |
| 1800 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 |
David Chinner | 4be536d | 2006-09-07 14:26:50 +1000 | [diff] [blame] | 1801 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1802 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1803 | STATIC void |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1804 | xfs_icsb_balance_counter_locked( |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1805 | xfs_mount_t *mp, |
| 1806 | xfs_sb_field_t field, |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1807 | int min_per_cpu) |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1808 | { |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1809 | uint64_t count, resid; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1810 | int weight = num_online_cpus(); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1811 | uint64_t min = (uint64_t)min_per_cpu; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1812 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1813 | /* disable counter and sync counter */ |
| 1814 | xfs_icsb_disable_counter(mp, field); |
| 1815 | |
| 1816 | /* update counters - first CPU gets residual*/ |
| 1817 | switch (field) { |
| 1818 | case XFS_SBS_ICOUNT: |
| 1819 | count = mp->m_sb.sb_icount; |
| 1820 | resid = do_div(count, weight); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1821 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1822 | return; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1823 | break; |
| 1824 | case XFS_SBS_IFREE: |
| 1825 | count = mp->m_sb.sb_ifree; |
| 1826 | resid = do_div(count, weight); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1827 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1828 | return; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1829 | break; |
| 1830 | case XFS_SBS_FDBLOCKS: |
| 1831 | count = mp->m_sb.sb_fdblocks; |
| 1832 | resid = do_div(count, weight); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1833 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1834 | return; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1835 | break; |
| 1836 | default: |
| 1837 | BUG(); |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1838 | count = resid = 0; /* quiet, gcc */ |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1839 | break; |
| 1840 | } |
| 1841 | |
| 1842 | xfs_icsb_enable_counter(mp, field, count, resid); |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1843 | } |
| 1844 | |
| 1845 | STATIC void |
| 1846 | xfs_icsb_balance_counter( |
| 1847 | xfs_mount_t *mp, |
| 1848 | xfs_sb_field_t fields, |
| 1849 | int min_per_cpu) |
| 1850 | { |
| 1851 | spin_lock(&mp->m_sb_lock); |
| 1852 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); |
| 1853 | spin_unlock(&mp->m_sb_lock); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1854 | } |
| 1855 | |
Christoph Hellwig | 1b04071 | 2010-09-30 02:25:56 +0000 | [diff] [blame] | 1856 | int |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1857 | xfs_icsb_modify_counters( |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1858 | xfs_mount_t *mp, |
| 1859 | xfs_sb_field_t field, |
David Chinner | 20f4ebf | 2007-02-10 18:36:10 +1100 | [diff] [blame] | 1860 | int64_t delta, |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1861 | int rsvd) |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1862 | { |
| 1863 | xfs_icsb_cnts_t *icsbp; |
| 1864 | long long lcounter; /* long counter for 64 bit fields */ |
Christoph Lameter | 7a9e02d | 2009-10-03 19:48:23 +0900 | [diff] [blame] | 1865 | int ret = 0; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1866 | |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1867 | might_sleep(); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1868 | again: |
Christoph Lameter | 7a9e02d | 2009-10-03 19:48:23 +0900 | [diff] [blame] | 1869 | preempt_disable(); |
| 1870 | icsbp = this_cpu_ptr(mp->m_sb_cnts); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1871 | |
| 1872 | /* |
| 1873 | * if the counter is disabled, go to slow path |
| 1874 | */ |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1875 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
| 1876 | goto slow_path; |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1877 | xfs_icsb_lock_cntr(icsbp); |
| 1878 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { |
| 1879 | xfs_icsb_unlock_cntr(icsbp); |
| 1880 | goto slow_path; |
| 1881 | } |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1882 | |
| 1883 | switch (field) { |
| 1884 | case XFS_SBS_ICOUNT: |
| 1885 | lcounter = icsbp->icsb_icount; |
| 1886 | lcounter += delta; |
| 1887 | if (unlikely(lcounter < 0)) |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1888 | goto balance_counter; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1889 | icsbp->icsb_icount = lcounter; |
| 1890 | break; |
| 1891 | |
| 1892 | case XFS_SBS_IFREE: |
| 1893 | lcounter = icsbp->icsb_ifree; |
| 1894 | lcounter += delta; |
| 1895 | if (unlikely(lcounter < 0)) |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1896 | goto balance_counter; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1897 | icsbp->icsb_ifree = lcounter; |
| 1898 | break; |
| 1899 | |
| 1900 | case XFS_SBS_FDBLOCKS: |
| 1901 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); |
| 1902 | |
David Chinner | 4be536d | 2006-09-07 14:26:50 +1000 | [diff] [blame] | 1903 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1904 | lcounter += delta; |
| 1905 | if (unlikely(lcounter < 0)) |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1906 | goto balance_counter; |
David Chinner | 4be536d | 2006-09-07 14:26:50 +1000 | [diff] [blame] | 1907 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1908 | break; |
| 1909 | default: |
| 1910 | BUG(); |
| 1911 | break; |
| 1912 | } |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1913 | xfs_icsb_unlock_cntr(icsbp); |
Christoph Lameter | 7a9e02d | 2009-10-03 19:48:23 +0900 | [diff] [blame] | 1914 | preempt_enable(); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1915 | return 0; |
| 1916 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1917 | slow_path: |
Christoph Lameter | 7a9e02d | 2009-10-03 19:48:23 +0900 | [diff] [blame] | 1918 | preempt_enable(); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1919 | |
| 1920 | /* |
| 1921 | * serialise with a mutex so we don't burn lots of cpu on |
| 1922 | * the superblock lock. We still need to hold the superblock |
| 1923 | * lock, however, when we modify the global structures. |
| 1924 | */ |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1925 | xfs_icsb_lock(mp); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1926 | |
| 1927 | /* |
| 1928 | * Now running atomically. |
| 1929 | * |
| 1930 | * If the counter is enabled, someone has beaten us to rebalancing. |
| 1931 | * Drop the lock and try again in the fast path.... |
| 1932 | */ |
| 1933 | if (!(xfs_icsb_counter_disabled(mp, field))) { |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1934 | xfs_icsb_unlock(mp); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1935 | goto again; |
| 1936 | } |
| 1937 | |
| 1938 | /* |
| 1939 | * The counter is currently disabled. Because we are |
| 1940 | * running atomically here, we know a rebalance cannot |
| 1941 | * be in progress. Hence we can go straight to operating |
| 1942 | * on the global superblock. We do not call xfs_mod_incore_sb() |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1943 | * here even though we need to get the m_sb_lock. Doing so |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1944 | * will cause us to re-enter this function and deadlock. |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1945 | * Hence we get the m_sb_lock ourselves and then call |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1946 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
| 1947 | * directly on the global counters. |
| 1948 | */ |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1949 | spin_lock(&mp->m_sb_lock); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1950 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
Eric Sandeen | 3685c2a | 2007-10-11 17:42:32 +1000 | [diff] [blame] | 1951 | spin_unlock(&mp->m_sb_lock); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1952 | |
| 1953 | /* |
| 1954 | * Now that we've modified the global superblock, we |
| 1955 | * may be able to re-enable the distributed counters |
| 1956 | * (e.g. lots of space just got freed). After that |
| 1957 | * we are done. |
| 1958 | */ |
| 1959 | if (ret != ENOSPC) |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1960 | xfs_icsb_balance_counter(mp, field, 0); |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1961 | xfs_icsb_unlock(mp); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1962 | return ret; |
| 1963 | |
| 1964 | balance_counter: |
David Chinner | 01e1b69 | 2006-03-14 13:29:16 +1100 | [diff] [blame] | 1965 | xfs_icsb_unlock_cntr(icsbp); |
Christoph Lameter | 7a9e02d | 2009-10-03 19:48:23 +0900 | [diff] [blame] | 1966 | preempt_enable(); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1967 | |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1968 | /* |
| 1969 | * We may have multiple threads here if multiple per-cpu |
| 1970 | * counters run dry at the same time. This will mean we can |
| 1971 | * do more balances than strictly necessary but it is not |
| 1972 | * the common slowpath case. |
| 1973 | */ |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1974 | xfs_icsb_lock(mp); |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1975 | |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1976 | /* |
| 1977 | * running atomically. |
| 1978 | * |
| 1979 | * This will leave the counter in the correct state for future |
| 1980 | * accesses. After the rebalance, we simply try again and our retry |
| 1981 | * will either succeed through the fast path or slow path without |
| 1982 | * another balance operation being required. |
| 1983 | */ |
Christoph Hellwig | 45af6c6 | 2008-04-22 17:34:44 +1000 | [diff] [blame] | 1984 | xfs_icsb_balance_counter(mp, field, delta); |
David Chinner | 03135cf | 2007-02-10 18:35:15 +1100 | [diff] [blame] | 1985 | xfs_icsb_unlock(mp); |
David Chinner | 20b6428 | 2007-02-10 18:35:09 +1100 | [diff] [blame] | 1986 | goto again; |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1987 | } |
| 1988 | |
David Chinner | 8d280b9 | 2006-03-14 13:13:09 +1100 | [diff] [blame] | 1989 | #endif |