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" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include "xfs_types.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" |
| 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include "xfs_dir2.h" |
| 28 | #include "xfs_dmapi.h" |
| 29 | #include "xfs_mount.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include "xfs_bmap_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 31 | #include "xfs_alloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include "xfs_ialloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include "xfs_dir2_sf.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 34 | #include "xfs_attr_sf.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_quota.h" |
| 40 | #include "xfs_utils.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
| 42 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | * Look up an inode by number in the given file system. |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 44 | * The inode is looked up in the cache held in each AG. |
| 45 | * If the inode is found in the cache, attach it to the provided |
| 46 | * vnode. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | * |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 48 | * If it is not in core, read it in from the file system's device, |
| 49 | * add it to the cache and attach the provided vnode. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | * |
| 51 | * The inode is locked according to the value of the lock_flags parameter. |
| 52 | * This flag parameter indicates how and if the inode's IO lock and inode lock |
| 53 | * should be taken. |
| 54 | * |
| 55 | * mp -- the mount point structure for the current file system. It points |
| 56 | * to the inode hash table. |
| 57 | * tp -- a pointer to the current transaction if there is one. This is |
| 58 | * simply passed through to the xfs_iread() call. |
| 59 | * ino -- the number of the inode desired. This is the unique identifier |
| 60 | * within the file system for the inode being requested. |
| 61 | * lock_flags -- flags indicating how to lock the inode. See the comment |
| 62 | * for xfs_ilock() for a list of valid values. |
| 63 | * bno -- the block number starting the buffer containing the inode, |
| 64 | * if known (as by bulkstat), else 0. |
| 65 | */ |
| 66 | STATIC int |
| 67 | xfs_iget_core( |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 68 | bhv_vnode_t *vp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | xfs_mount_t *mp, |
| 70 | xfs_trans_t *tp, |
| 71 | xfs_ino_t ino, |
| 72 | uint flags, |
| 73 | uint lock_flags, |
| 74 | xfs_inode_t **ipp, |
| 75 | xfs_daddr_t bno) |
| 76 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | xfs_inode_t *ip; |
| 78 | xfs_inode_t *iq; |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 79 | bhv_vnode_t *inode_vp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | int error; |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 81 | xfs_icluster_t *icl, *new_icl = NULL; |
| 82 | unsigned long first_index, mask; |
| 83 | xfs_perag_t *pag; |
| 84 | xfs_agino_t agino; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 86 | /* the radix tree exists only in inode capable AGs */ |
| 87 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) |
| 88 | return EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 90 | /* get the perag structure and ensure that it's inode capable */ |
| 91 | pag = xfs_get_perag(mp, ino); |
| 92 | if (!pag->pagi_inodeok) |
| 93 | return EINVAL; |
| 94 | ASSERT(pag->pag_ici_init); |
| 95 | agino = XFS_INO_TO_AGINO(mp, ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
| 97 | again: |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 98 | read_lock(&pag->pag_ici_lock); |
| 99 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 101 | if (ip != NULL) { |
| 102 | /* |
| 103 | * If INEW is set this inode is being set up |
| 104 | * we need to pause and try again. |
| 105 | */ |
| 106 | if (xfs_iflags_test(ip, XFS_INEW)) { |
| 107 | read_unlock(&pag->pag_ici_lock); |
| 108 | delay(1); |
| 109 | XFS_STATS_INC(xs_ig_frecycle); |
| 110 | |
| 111 | goto again; |
| 112 | } |
| 113 | |
| 114 | inode_vp = XFS_ITOV_NULL(ip); |
| 115 | if (inode_vp == NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | /* |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 117 | * If IRECLAIM is set this inode is |
| 118 | * on its way out of the system, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | * we need to pause and try again. |
| 120 | */ |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 121 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { |
| 122 | read_unlock(&pag->pag_ici_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | delay(1); |
| 124 | XFS_STATS_INC(xs_ig_frecycle); |
| 125 | |
| 126 | goto again; |
| 127 | } |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 128 | ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 130 | /* |
| 131 | * If lookup is racing with unlink, then we |
| 132 | * should return an error immediately so we |
| 133 | * don't remove it from the reclaim list and |
| 134 | * potentially leak the inode. |
| 135 | */ |
| 136 | if ((ip->i_d.di_mode == 0) && |
| 137 | !(flags & XFS_IGET_CREATE)) { |
| 138 | read_unlock(&pag->pag_ici_lock); |
| 139 | xfs_put_perag(mp, pag); |
| 140 | return ENOENT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } |
| 142 | |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame] | 143 | /* |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 144 | * There may be transactions sitting in the |
| 145 | * incore log buffers or being flushed to disk |
| 146 | * at this time. We can't clear the |
| 147 | * XFS_IRECLAIMABLE flag until these |
| 148 | * transactions have hit the disk, otherwise we |
| 149 | * will void the guarantee the flag provides |
| 150 | * xfs_iunpin() |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame] | 151 | */ |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 152 | if (xfs_ipincount(ip)) { |
| 153 | read_unlock(&pag->pag_ici_lock); |
| 154 | xfs_log_force(mp, 0, |
| 155 | XFS_LOG_FORCE|XFS_LOG_SYNC); |
| 156 | XFS_STATS_INC(xs_ig_frecycle); |
| 157 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | } |
Nathan Scott | 745b1f47 | 2006-09-28 11:02:23 +1000 | [diff] [blame] | 159 | |
Lachlan McIlroy | 15947f2 | 2007-09-17 13:11:58 +1000 | [diff] [blame^] | 160 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 162 | XFS_STATS_INC(xs_ig_found); |
| 163 | |
| 164 | xfs_iflags_clear(ip, XFS_IRECLAIMABLE); |
| 165 | read_unlock(&pag->pag_ici_lock); |
| 166 | |
| 167 | XFS_MOUNT_ILOCK(mp); |
| 168 | list_del_init(&ip->i_reclaim); |
| 169 | XFS_MOUNT_IUNLOCK(mp); |
| 170 | |
| 171 | goto finish_inode; |
| 172 | |
| 173 | } else if (vp != inode_vp) { |
| 174 | struct inode *inode = vn_to_inode(inode_vp); |
| 175 | |
| 176 | /* The inode is being torn down, pause and |
| 177 | * try again. |
| 178 | */ |
| 179 | if (inode->i_state & (I_FREEING | I_CLEAR)) { |
| 180 | read_unlock(&pag->pag_ici_lock); |
| 181 | delay(1); |
| 182 | XFS_STATS_INC(xs_ig_frecycle); |
| 183 | |
| 184 | goto again; |
| 185 | } |
| 186 | /* Chances are the other vnode (the one in the inode) is being torn |
| 187 | * down right now, and we landed on top of it. Question is, what do |
| 188 | * we do? Unhook the old inode and hook up the new one? |
| 189 | */ |
| 190 | cmn_err(CE_PANIC, |
| 191 | "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p", |
| 192 | inode_vp, vp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | } |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 194 | |
| 195 | /* |
| 196 | * Inode cache hit |
| 197 | */ |
| 198 | read_unlock(&pag->pag_ici_lock); |
| 199 | XFS_STATS_INC(xs_ig_found); |
| 200 | |
| 201 | finish_inode: |
| 202 | if (ip->i_d.di_mode == 0) { |
| 203 | if (!(flags & XFS_IGET_CREATE)) { |
| 204 | xfs_put_perag(mp, pag); |
| 205 | return ENOENT; |
| 206 | } |
| 207 | xfs_iocore_inode_reinit(ip); |
| 208 | } |
| 209 | |
| 210 | if (lock_flags != 0) |
| 211 | xfs_ilock(ip, lock_flags); |
| 212 | |
| 213 | xfs_iflags_clear(ip, XFS_ISTALE); |
Lachlan McIlroy | 15947f2 | 2007-09-17 13:11:58 +1000 | [diff] [blame^] | 214 | xfs_itrace_exit_tag(ip, "xfs_iget.found"); |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 215 | goto return_ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | } |
| 217 | |
| 218 | /* |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 219 | * Inode cache miss |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | */ |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 221 | read_unlock(&pag->pag_ici_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | XFS_STATS_INC(xs_ig_missed); |
| 223 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | /* |
| 225 | * Read the disk inode attributes into a new inode structure and get |
| 226 | * a new vnode for it. This should also initialize i_ino and i_mount. |
| 227 | */ |
Nathan Scott | 745b1f47 | 2006-09-28 11:02:23 +1000 | [diff] [blame] | 228 | error = xfs_iread(mp, tp, ino, &ip, bno, |
| 229 | (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0); |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 230 | if (error) { |
| 231 | xfs_put_perag(mp, pag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | return error; |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 233 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | |
Lachlan McIlroy | 15947f2 | 2007-09-17 13:11:58 +1000 | [diff] [blame^] | 235 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | |
| 237 | xfs_inode_lock_init(ip, vp); |
| 238 | xfs_iocore_inode_init(ip); |
Nathan Scott | 745b1f47 | 2006-09-28 11:02:23 +1000 | [diff] [blame] | 239 | if (lock_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | xfs_ilock(ip, lock_flags); |
Nathan Scott | 745b1f47 | 2006-09-28 11:02:23 +1000 | [diff] [blame] | 241 | |
| 242 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 243 | xfs_idestroy(ip); |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 244 | xfs_put_perag(mp, pag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | return ENOENT; |
| 246 | } |
| 247 | |
| 248 | /* |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 249 | * This is a bit messy - we preallocate everything we _might_ |
| 250 | * need before we pick up the ici lock. That way we don't have to |
| 251 | * juggle locks and go all the way back to the start. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 252 | */ |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 253 | new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP); |
| 254 | if (radix_tree_preload(GFP_KERNEL)) { |
| 255 | delay(1); |
| 256 | goto again; |
| 257 | } |
| 258 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
| 259 | first_index = agino & mask; |
| 260 | write_lock(&pag->pag_ici_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 261 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 262 | /* |
| 263 | * Find the cluster if it exists |
| 264 | */ |
| 265 | icl = NULL; |
| 266 | if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq, |
| 267 | first_index, 1)) { |
David Chinner | a7430847 | 2007-11-23 16:30:23 +1100 | [diff] [blame] | 268 | if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) == first_index) |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 269 | icl = iq->i_cluster; |
| 270 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 272 | /* |
| 273 | * insert the new inode |
| 274 | */ |
| 275 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
| 276 | if (unlikely(error)) { |
| 277 | BUG_ON(error != -EEXIST); |
| 278 | write_unlock(&pag->pag_ici_lock); |
| 279 | radix_tree_preload_end(); |
| 280 | xfs_idestroy(ip); |
| 281 | XFS_STATS_INC(xs_ig_dup); |
| 282 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | } |
| 284 | |
| 285 | /* |
| 286 | * These values _must_ be set before releasing ihlock! |
| 287 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | ip->i_udquot = ip->i_gdquot = NULL; |
David Chinner | 7a18c38 | 2006-11-11 18:04:54 +1100 | [diff] [blame] | 289 | xfs_iflags_set(ip, XFS_INEW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 291 | ASSERT(ip->i_cluster == NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 293 | if (!icl) { |
| 294 | spin_lock_init(&new_icl->icl_lock); |
| 295 | INIT_HLIST_HEAD(&new_icl->icl_inodes); |
| 296 | icl = new_icl; |
| 297 | new_icl = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | } else { |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 299 | ASSERT(!hlist_empty(&icl->icl_inodes)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | } |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 301 | spin_lock(&icl->icl_lock); |
| 302 | hlist_add_head(&ip->i_cnode, &icl->icl_inodes); |
| 303 | ip->i_cluster = icl; |
| 304 | spin_unlock(&icl->icl_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 306 | write_unlock(&pag->pag_ici_lock); |
| 307 | radix_tree_preload_end(); |
| 308 | if (new_icl) |
| 309 | kmem_zone_free(xfs_icluster_zone, new_icl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | |
| 311 | /* |
| 312 | * Link ip to its mount and thread it on the mount's inode list. |
| 313 | */ |
| 314 | XFS_MOUNT_ILOCK(mp); |
| 315 | if ((iq = mp->m_inodes)) { |
| 316 | ASSERT(iq->i_mprev->i_mnext == iq); |
| 317 | ip->i_mprev = iq->i_mprev; |
| 318 | iq->i_mprev->i_mnext = ip; |
| 319 | iq->i_mprev = ip; |
| 320 | ip->i_mnext = iq; |
| 321 | } else { |
| 322 | ip->i_mnext = ip; |
| 323 | ip->i_mprev = ip; |
| 324 | } |
| 325 | mp->m_inodes = ip; |
| 326 | |
| 327 | XFS_MOUNT_IUNLOCK(mp); |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 328 | xfs_put_perag(mp, pag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | |
| 330 | return_ip: |
| 331 | ASSERT(ip->i_df.if_ext_max == |
| 332 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); |
| 333 | |
| 334 | ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) == |
| 335 | ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0)); |
| 336 | |
Christoph Hellwig | b3aea4e | 2007-08-29 11:44:37 +1000 | [diff] [blame] | 337 | xfs_iflags_set(ip, XFS_IMODIFIED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 338 | *ipp = ip; |
| 339 | |
| 340 | /* |
| 341 | * If we have a real type for an on-disk inode, we can set ops(&unlock) |
| 342 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
| 343 | */ |
Christoph Hellwig | 745f691 | 2007-08-30 17:20:39 +1000 | [diff] [blame] | 344 | xfs_initialize_vnode(mp, vp, ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | return 0; |
| 346 | } |
| 347 | |
| 348 | |
| 349 | /* |
| 350 | * The 'normal' internal xfs_iget, if needed it will |
| 351 | * 'allocate', or 'get', the vnode. |
| 352 | */ |
| 353 | int |
| 354 | xfs_iget( |
| 355 | xfs_mount_t *mp, |
| 356 | xfs_trans_t *tp, |
| 357 | xfs_ino_t ino, |
| 358 | uint flags, |
| 359 | uint lock_flags, |
| 360 | xfs_inode_t **ipp, |
| 361 | xfs_daddr_t bno) |
| 362 | { |
| 363 | struct inode *inode; |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 364 | bhv_vnode_t *vp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | int error; |
| 366 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | XFS_STATS_INC(xs_ig_attempts); |
| 368 | |
Christoph Hellwig | ba403ab | 2005-09-05 08:33:00 +1000 | [diff] [blame] | 369 | retry: |
Christoph Hellwig | b267ce9 | 2007-08-30 17:21:30 +1000 | [diff] [blame] | 370 | inode = iget_locked(mp->m_super, ino); |
Christoph Hellwig | 745f691 | 2007-08-30 17:20:39 +1000 | [diff] [blame] | 371 | if (inode) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | xfs_inode_t *ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | |
Nathan Scott | ec86dc0 | 2006-03-17 17:25:36 +1100 | [diff] [blame] | 374 | vp = vn_from_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | if (inode->i_state & I_NEW) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | vn_initialize(inode); |
| 377 | error = xfs_iget_core(vp, mp, tp, ino, flags, |
| 378 | lock_flags, ipp, bno); |
| 379 | if (error) { |
| 380 | vn_mark_bad(vp); |
| 381 | if (inode->i_state & I_NEW) |
| 382 | unlock_new_inode(inode); |
| 383 | iput(inode); |
| 384 | } |
| 385 | } else { |
Christoph Hellwig | ba403ab | 2005-09-05 08:33:00 +1000 | [diff] [blame] | 386 | /* |
| 387 | * If the inode is not fully constructed due to |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 388 | * filehandle mismatches wait for the inode to go |
Christoph Hellwig | ba403ab | 2005-09-05 08:33:00 +1000 | [diff] [blame] | 389 | * away and try again. |
| 390 | * |
| 391 | * iget_locked will call __wait_on_freeing_inode |
| 392 | * to wait for the inode to go away. |
| 393 | */ |
| 394 | if (is_bad_inode(inode) || |
Christoph Hellwig | 75e17b3 | 2006-01-11 20:58:44 +1100 | [diff] [blame] | 395 | ((ip = xfs_vtoi(vp)) == NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | iput(inode); |
Christoph Hellwig | ba403ab | 2005-09-05 08:33:00 +1000 | [diff] [blame] | 397 | delay(1); |
| 398 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | } |
| 400 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | if (lock_flags != 0) |
| 402 | xfs_ilock(ip, lock_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | XFS_STATS_INC(xs_ig_found); |
| 404 | *ipp = ip; |
| 405 | error = 0; |
| 406 | } |
| 407 | } else |
| 408 | error = ENOMEM; /* If we got no inode we are out of memory */ |
| 409 | |
| 410 | return error; |
| 411 | } |
| 412 | |
| 413 | /* |
| 414 | * Do the setup for the various locks within the incore inode. |
| 415 | */ |
| 416 | void |
| 417 | xfs_inode_lock_init( |
| 418 | xfs_inode_t *ip, |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 419 | bhv_vnode_t *vp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | { |
| 421 | mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, |
Christoph Hellwig | 2aeaa25 | 2007-08-29 11:46:57 +1000 | [diff] [blame] | 422 | "xfsino", ip->i_ino); |
| 423 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | init_waitqueue_head(&ip->i_ipin_wait); |
| 425 | atomic_set(&ip->i_pincount, 0); |
Eric Sandeen | 7ae67d7 | 2006-09-28 11:05:46 +1000 | [diff] [blame] | 426 | initnsema(&ip->i_flock, 1, "xfsfino"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | } |
| 428 | |
| 429 | /* |
| 430 | * Look for the inode corresponding to the given ino in the hash table. |
| 431 | * If it is there and its i_transp pointer matches tp, return it. |
| 432 | * Otherwise, return NULL. |
| 433 | */ |
| 434 | xfs_inode_t * |
| 435 | xfs_inode_incore(xfs_mount_t *mp, |
| 436 | xfs_ino_t ino, |
| 437 | xfs_trans_t *tp) |
| 438 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | xfs_inode_t *ip; |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 440 | xfs_perag_t *pag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 442 | pag = xfs_get_perag(mp, ino); |
| 443 | read_lock(&pag->pag_ici_lock); |
| 444 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino)); |
| 445 | read_unlock(&pag->pag_ici_lock); |
| 446 | xfs_put_perag(mp, pag); |
| 447 | |
| 448 | /* the returned inode must match the transaction */ |
| 449 | if (ip && (ip->i_transp != tp)) |
| 450 | return NULL; |
| 451 | return ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | } |
| 453 | |
| 454 | /* |
| 455 | * Decrement reference count of an inode structure and unlock it. |
| 456 | * |
| 457 | * ip -- the inode being released |
| 458 | * lock_flags -- this parameter indicates the inode's locks to be |
| 459 | * to be released. See the comment on xfs_iunlock() for a list |
| 460 | * of valid values. |
| 461 | */ |
| 462 | void |
| 463 | xfs_iput(xfs_inode_t *ip, |
| 464 | uint lock_flags) |
| 465 | { |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 466 | bhv_vnode_t *vp = XFS_ITOV(ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | |
Lachlan McIlroy | cf441ee | 2008-02-07 16:42:19 +1100 | [diff] [blame] | 468 | xfs_itrace_entry(ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | xfs_iunlock(ip, lock_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | VN_RELE(vp); |
| 471 | } |
| 472 | |
| 473 | /* |
| 474 | * Special iput for brand-new inodes that are still locked |
| 475 | */ |
| 476 | void |
| 477 | xfs_iput_new(xfs_inode_t *ip, |
| 478 | uint lock_flags) |
| 479 | { |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 480 | bhv_vnode_t *vp = XFS_ITOV(ip); |
Nathan Scott | ec86dc0 | 2006-03-17 17:25:36 +1100 | [diff] [blame] | 481 | struct inode *inode = vn_to_inode(vp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | |
Lachlan McIlroy | cf441ee | 2008-02-07 16:42:19 +1100 | [diff] [blame] | 483 | xfs_itrace_entry(ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | |
| 485 | if ((ip->i_d.di_mode == 0)) { |
David Chinner | 7a18c38 | 2006-11-11 18:04:54 +1100 | [diff] [blame] | 486 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | vn_mark_bad(vp); |
| 488 | } |
| 489 | if (inode->i_state & I_NEW) |
| 490 | unlock_new_inode(inode); |
| 491 | if (lock_flags) |
| 492 | xfs_iunlock(ip, lock_flags); |
| 493 | VN_RELE(vp); |
| 494 | } |
| 495 | |
| 496 | |
| 497 | /* |
| 498 | * This routine embodies the part of the reclaim code that pulls |
| 499 | * the inode from the inode hash table and the mount structure's |
| 500 | * inode list. |
| 501 | * This should only be called from xfs_reclaim(). |
| 502 | */ |
| 503 | void |
| 504 | xfs_ireclaim(xfs_inode_t *ip) |
| 505 | { |
Nathan Scott | 67fcaa7 | 2006-06-09 17:00:52 +1000 | [diff] [blame] | 506 | bhv_vnode_t *vp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | |
| 508 | /* |
| 509 | * Remove from old hash list and mount list. |
| 510 | */ |
| 511 | XFS_STATS_INC(xs_ig_reclaims); |
| 512 | |
| 513 | xfs_iextract(ip); |
| 514 | |
| 515 | /* |
| 516 | * Here we do a spurious inode lock in order to coordinate with |
| 517 | * xfs_sync(). This is because xfs_sync() references the inodes |
| 518 | * in the mount list without taking references on the corresponding |
| 519 | * vnodes. We make that OK here by ensuring that we wait until |
| 520 | * the inode is unlocked in xfs_sync() before we go ahead and |
| 521 | * free it. We get both the regular lock and the io lock because |
| 522 | * the xfs_sync() code may need to drop the regular one but will |
| 523 | * still hold the io lock. |
| 524 | */ |
| 525 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 526 | |
| 527 | /* |
| 528 | * Release dquots (and their references) if any. An inode may escape |
| 529 | * xfs_inactive and get here via vn_alloc->vn_reclaim path. |
| 530 | */ |
| 531 | XFS_QM_DQDETACH(ip->i_mount, ip); |
| 532 | |
| 533 | /* |
| 534 | * Pull our behavior descriptor from the vnode chain. |
| 535 | */ |
| 536 | vp = XFS_ITOV_NULL(ip); |
| 537 | if (vp) { |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 538 | vn_to_inode(vp)->i_private = NULL; |
| 539 | ip->i_vnode = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | } |
| 541 | |
| 542 | /* |
| 543 | * Free all memory associated with the inode. |
| 544 | */ |
Tim Shimmin | 439b843 | 2006-11-11 18:04:34 +1100 | [diff] [blame] | 545 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | xfs_idestroy(ip); |
| 547 | } |
| 548 | |
| 549 | /* |
| 550 | * This routine removes an about-to-be-destroyed inode from |
| 551 | * all of the lists in which it is located with the exception |
| 552 | * of the behavior chain. |
| 553 | */ |
| 554 | void |
| 555 | xfs_iextract( |
| 556 | xfs_inode_t *ip) |
| 557 | { |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 558 | xfs_mount_t *mp = ip->i_mount; |
| 559 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | xfs_inode_t *iq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 562 | write_lock(&pag->pag_ici_lock); |
| 563 | radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino)); |
| 564 | write_unlock(&pag->pag_ici_lock); |
| 565 | xfs_put_perag(mp, pag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | |
| 567 | /* |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 568 | * Remove from cluster list |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | */ |
| 570 | mp = ip->i_mount; |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 571 | spin_lock(&ip->i_cluster->icl_lock); |
| 572 | hlist_del(&ip->i_cnode); |
| 573 | spin_unlock(&ip->i_cluster->icl_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 575 | /* was last inode in cluster? */ |
| 576 | if (hlist_empty(&ip->i_cluster->icl_inodes)) |
| 577 | kmem_zone_free(xfs_icluster_zone, ip->i_cluster); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | |
| 579 | /* |
| 580 | * Remove from mount's inode list. |
| 581 | */ |
| 582 | XFS_MOUNT_ILOCK(mp); |
| 583 | ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL)); |
| 584 | iq = ip->i_mnext; |
| 585 | iq->i_mprev = ip->i_mprev; |
| 586 | ip->i_mprev->i_mnext = iq; |
| 587 | |
| 588 | /* |
| 589 | * Fix up the head pointer if it points to the inode being deleted. |
| 590 | */ |
| 591 | if (mp->m_inodes == ip) { |
| 592 | if (ip == iq) { |
| 593 | mp->m_inodes = NULL; |
| 594 | } else { |
| 595 | mp->m_inodes = iq; |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | /* Deal with the deleted inodes list */ |
| 600 | list_del_init(&ip->i_reclaim); |
| 601 | |
| 602 | mp->m_ireclaims++; |
| 603 | XFS_MOUNT_IUNLOCK(mp); |
| 604 | } |
| 605 | |
| 606 | /* |
| 607 | * This is a wrapper routine around the xfs_ilock() routine |
| 608 | * used to centralize some grungy code. It is used in places |
| 609 | * that wish to lock the inode solely for reading the extents. |
| 610 | * The reason these places can't just call xfs_ilock(SHARED) |
| 611 | * is that the inode lock also guards to bringing in of the |
| 612 | * extents from disk for a file in b-tree format. If the inode |
| 613 | * is in b-tree format, then we need to lock the inode exclusively |
| 614 | * until the extents are read in. Locking it exclusively all |
| 615 | * the time would limit our parallelism unnecessarily, though. |
| 616 | * What we do instead is check to see if the extents have been |
| 617 | * read in yet, and only lock the inode exclusively if they |
| 618 | * have not. |
| 619 | * |
| 620 | * The function returns a value which should be given to the |
| 621 | * corresponding xfs_iunlock_map_shared(). This value is |
| 622 | * the mode in which the lock was actually taken. |
| 623 | */ |
| 624 | uint |
| 625 | xfs_ilock_map_shared( |
| 626 | xfs_inode_t *ip) |
| 627 | { |
| 628 | uint lock_mode; |
| 629 | |
| 630 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && |
| 631 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { |
| 632 | lock_mode = XFS_ILOCK_EXCL; |
| 633 | } else { |
| 634 | lock_mode = XFS_ILOCK_SHARED; |
| 635 | } |
| 636 | |
| 637 | xfs_ilock(ip, lock_mode); |
| 638 | |
| 639 | return lock_mode; |
| 640 | } |
| 641 | |
| 642 | /* |
| 643 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). |
| 644 | * All it does is call xfs_iunlock() with the given lock_mode. |
| 645 | */ |
| 646 | void |
| 647 | xfs_iunlock_map_shared( |
| 648 | xfs_inode_t *ip, |
| 649 | unsigned int lock_mode) |
| 650 | { |
| 651 | xfs_iunlock(ip, lock_mode); |
| 652 | } |
| 653 | |
| 654 | /* |
| 655 | * The xfs inode contains 2 locks: a multi-reader lock called the |
| 656 | * i_iolock and a multi-reader lock called the i_lock. This routine |
| 657 | * allows either or both of the locks to be obtained. |
| 658 | * |
| 659 | * The 2 locks should always be ordered so that the IO lock is |
| 660 | * obtained first in order to prevent deadlock. |
| 661 | * |
| 662 | * ip -- the inode being locked |
| 663 | * lock_flags -- this parameter indicates the inode's locks |
| 664 | * to be locked. It can be: |
| 665 | * XFS_IOLOCK_SHARED, |
| 666 | * XFS_IOLOCK_EXCL, |
| 667 | * XFS_ILOCK_SHARED, |
| 668 | * XFS_ILOCK_EXCL, |
| 669 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, |
| 670 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, |
| 671 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, |
| 672 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
| 673 | */ |
| 674 | void |
| 675 | xfs_ilock(xfs_inode_t *ip, |
| 676 | uint lock_flags) |
| 677 | { |
| 678 | /* |
| 679 | * You can't set both SHARED and EXCL for the same lock, |
| 680 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 681 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 682 | */ |
| 683 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 684 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 685 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 686 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 687 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | |
| 689 | if (lock_flags & XFS_IOLOCK_EXCL) { |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 690 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 692 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | } |
| 694 | if (lock_flags & XFS_ILOCK_EXCL) { |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 695 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 697 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | } |
| 699 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); |
| 700 | } |
| 701 | |
| 702 | /* |
| 703 | * This is just like xfs_ilock(), except that the caller |
| 704 | * is guaranteed not to sleep. It returns 1 if it gets |
| 705 | * the requested locks and 0 otherwise. If the IO lock is |
| 706 | * obtained but the inode lock cannot be, then the IO lock |
| 707 | * is dropped before returning. |
| 708 | * |
| 709 | * ip -- the inode being locked |
| 710 | * lock_flags -- this parameter indicates the inode's locks to be |
| 711 | * to be locked. See the comment for xfs_ilock() for a list |
| 712 | * of valid values. |
| 713 | * |
| 714 | */ |
| 715 | int |
| 716 | xfs_ilock_nowait(xfs_inode_t *ip, |
| 717 | uint lock_flags) |
| 718 | { |
| 719 | int iolocked; |
| 720 | int ilocked; |
| 721 | |
| 722 | /* |
| 723 | * You can't set both SHARED and EXCL for the same lock, |
| 724 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 725 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 726 | */ |
| 727 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 728 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 729 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 730 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 731 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 732 | |
| 733 | iolocked = 0; |
| 734 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 735 | iolocked = mrtryupdate(&ip->i_iolock); |
| 736 | if (!iolocked) { |
| 737 | return 0; |
| 738 | } |
| 739 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
| 740 | iolocked = mrtryaccess(&ip->i_iolock); |
| 741 | if (!iolocked) { |
| 742 | return 0; |
| 743 | } |
| 744 | } |
| 745 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 746 | ilocked = mrtryupdate(&ip->i_lock); |
| 747 | if (!ilocked) { |
| 748 | if (iolocked) { |
| 749 | mrunlock(&ip->i_iolock); |
| 750 | } |
| 751 | return 0; |
| 752 | } |
| 753 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
| 754 | ilocked = mrtryaccess(&ip->i_lock); |
| 755 | if (!ilocked) { |
| 756 | if (iolocked) { |
| 757 | mrunlock(&ip->i_iolock); |
| 758 | } |
| 759 | return 0; |
| 760 | } |
| 761 | } |
| 762 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); |
| 763 | return 1; |
| 764 | } |
| 765 | |
| 766 | /* |
| 767 | * xfs_iunlock() is used to drop the inode locks acquired with |
| 768 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass |
| 769 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so |
| 770 | * that we know which locks to drop. |
| 771 | * |
| 772 | * ip -- the inode being unlocked |
| 773 | * lock_flags -- this parameter indicates the inode's locks to be |
| 774 | * to be unlocked. See the comment for xfs_ilock() for a list |
| 775 | * of valid values for this parameter. |
| 776 | * |
| 777 | */ |
| 778 | void |
| 779 | xfs_iunlock(xfs_inode_t *ip, |
| 780 | uint lock_flags) |
| 781 | { |
| 782 | /* |
| 783 | * You can't set both SHARED and EXCL for the same lock, |
| 784 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 785 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 786 | */ |
| 787 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 788 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 789 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 790 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
Lachlan McIlroy | f7c66ce | 2007-05-08 13:50:19 +1000 | [diff] [blame] | 791 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
| 792 | XFS_LOCK_DEP_MASK)) == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | ASSERT(lock_flags != 0); |
| 794 | |
| 795 | if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) { |
| 796 | ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) || |
| 797 | (ismrlocked(&ip->i_iolock, MR_ACCESS))); |
| 798 | ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) || |
| 799 | (ismrlocked(&ip->i_iolock, MR_UPDATE))); |
| 800 | mrunlock(&ip->i_iolock); |
| 801 | } |
| 802 | |
| 803 | if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) { |
| 804 | ASSERT(!(lock_flags & XFS_ILOCK_SHARED) || |
| 805 | (ismrlocked(&ip->i_lock, MR_ACCESS))); |
| 806 | ASSERT(!(lock_flags & XFS_ILOCK_EXCL) || |
| 807 | (ismrlocked(&ip->i_lock, MR_UPDATE))); |
| 808 | mrunlock(&ip->i_lock); |
| 809 | |
| 810 | /* |
| 811 | * Let the AIL know that this item has been unlocked in case |
| 812 | * it is in the AIL and anyone is waiting on it. Don't do |
| 813 | * this if the caller has asked us not to. |
| 814 | */ |
| 815 | if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) && |
| 816 | ip->i_itemp != NULL) { |
| 817 | xfs_trans_unlocked_item(ip->i_mount, |
| 818 | (xfs_log_item_t*)(ip->i_itemp)); |
| 819 | } |
| 820 | } |
| 821 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); |
| 822 | } |
| 823 | |
| 824 | /* |
| 825 | * give up write locks. the i/o lock cannot be held nested |
| 826 | * if it is being demoted. |
| 827 | */ |
| 828 | void |
| 829 | xfs_ilock_demote(xfs_inode_t *ip, |
| 830 | uint lock_flags) |
| 831 | { |
| 832 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); |
| 833 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); |
| 834 | |
| 835 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 836 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); |
| 837 | mrdemote(&ip->i_lock); |
| 838 | } |
| 839 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 840 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); |
| 841 | mrdemote(&ip->i_iolock); |
| 842 | } |
| 843 | } |
| 844 | |
| 845 | /* |
| 846 | * The following three routines simply manage the i_flock |
| 847 | * semaphore embedded in the inode. This semaphore synchronizes |
| 848 | * processes attempting to flush the in-core inode back to disk. |
| 849 | */ |
| 850 | void |
| 851 | xfs_iflock(xfs_inode_t *ip) |
| 852 | { |
| 853 | psema(&(ip->i_flock), PINOD|PLTWAIT); |
| 854 | } |
| 855 | |
| 856 | int |
| 857 | xfs_iflock_nowait(xfs_inode_t *ip) |
| 858 | { |
| 859 | return (cpsema(&(ip->i_flock))); |
| 860 | } |
| 861 | |
| 862 | void |
| 863 | xfs_ifunlock(xfs_inode_t *ip) |
| 864 | { |
Al Viro | 0d8fee3 | 2006-06-19 08:41:30 +1000 | [diff] [blame] | 865 | ASSERT(issemalocked(&(ip->i_flock))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 866 | vsema(&(ip->i_flock)); |
| 867 | } |