Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * inode.c |
| 5 | * |
| 6 | * vfs' aops, fops, dops and iops |
| 7 | * |
| 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public |
| 21 | * License along with this program; if not, write to the |
| 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 23 | * Boston, MA 021110-1307, USA. |
| 24 | */ |
| 25 | |
| 26 | #include <linux/fs.h> |
| 27 | #include <linux/types.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 28 | #include <linux/highmem.h> |
| 29 | #include <linux/pagemap.h> |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 30 | #include <linux/quotaops.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 31 | |
| 32 | #include <asm/byteorder.h> |
| 33 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 34 | #include <cluster/masklog.h> |
| 35 | |
| 36 | #include "ocfs2.h" |
| 37 | |
| 38 | #include "alloc.h" |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 39 | #include "dir.h" |
Joel Becker | d6b32bb | 2008-10-17 14:55:01 -0700 | [diff] [blame] | 40 | #include "blockcheck.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 41 | #include "dlmglue.h" |
| 42 | #include "extent_map.h" |
| 43 | #include "file.h" |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 44 | #include "heartbeat.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 45 | #include "inode.h" |
| 46 | #include "journal.h" |
| 47 | #include "namei.h" |
| 48 | #include "suballoc.h" |
| 49 | #include "super.h" |
| 50 | #include "symlink.h" |
| 51 | #include "sysfile.h" |
| 52 | #include "uptodate.h" |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 53 | #include "xattr.h" |
Tao Ma | 8b2c0db | 2009-08-18 11:43:49 +0800 | [diff] [blame] | 54 | #include "refcounttree.h" |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 55 | #include "ocfs2_trace.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 56 | |
| 57 | #include "buffer_head_io.h" |
| 58 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 59 | struct ocfs2_find_inode_args |
| 60 | { |
| 61 | u64 fi_blkno; |
| 62 | unsigned long fi_ino; |
| 63 | unsigned int fi_flags; |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 64 | unsigned int fi_sysfile_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 65 | }; |
| 66 | |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 67 | static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES]; |
| 68 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 69 | static int ocfs2_read_locked_inode(struct inode *inode, |
| 70 | struct ocfs2_find_inode_args *args); |
| 71 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); |
| 72 | static int ocfs2_find_actor(struct inode *inode, void *opaque); |
| 73 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, |
| 74 | struct inode *inode, |
| 75 | struct buffer_head *fe_bh); |
| 76 | |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 77 | void ocfs2_set_inode_flags(struct inode *inode) |
| 78 | { |
| 79 | unsigned int flags = OCFS2_I(inode)->ip_attr; |
| 80 | |
| 81 | inode->i_flags &= ~(S_IMMUTABLE | |
| 82 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); |
| 83 | |
| 84 | if (flags & OCFS2_IMMUTABLE_FL) |
| 85 | inode->i_flags |= S_IMMUTABLE; |
| 86 | |
| 87 | if (flags & OCFS2_SYNC_FL) |
| 88 | inode->i_flags |= S_SYNC; |
| 89 | if (flags & OCFS2_APPEND_FL) |
| 90 | inode->i_flags |= S_APPEND; |
| 91 | if (flags & OCFS2_NOATIME_FL) |
| 92 | inode->i_flags |= S_NOATIME; |
| 93 | if (flags & OCFS2_DIRSYNC_FL) |
| 94 | inode->i_flags |= S_DIRSYNC; |
| 95 | } |
| 96 | |
Jan Kara | 6e4b0d5 | 2007-04-27 11:08:01 -0700 | [diff] [blame] | 97 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
| 98 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) |
| 99 | { |
| 100 | unsigned int flags = oi->vfs_inode.i_flags; |
| 101 | |
| 102 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| |
| 103 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); |
| 104 | if (flags & S_SYNC) |
| 105 | oi->ip_attr |= OCFS2_SYNC_FL; |
| 106 | if (flags & S_APPEND) |
| 107 | oi->ip_attr |= OCFS2_APPEND_FL; |
| 108 | if (flags & S_IMMUTABLE) |
| 109 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; |
| 110 | if (flags & S_NOATIME) |
| 111 | oi->ip_attr |= OCFS2_NOATIME_FL; |
| 112 | if (flags & S_DIRSYNC) |
| 113 | oi->ip_attr |= OCFS2_DIRSYNC_FL; |
| 114 | } |
| 115 | |
wengang wang | 6ca497a | 2009-03-06 21:29:10 +0800 | [diff] [blame] | 116 | struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno) |
| 117 | { |
| 118 | struct ocfs2_find_inode_args args; |
| 119 | |
| 120 | args.fi_blkno = blkno; |
| 121 | args.fi_flags = 0; |
| 122 | args.fi_ino = ino_from_blkno(sb, blkno); |
| 123 | args.fi_sysfile_type = 0; |
| 124 | |
| 125 | return ilookup5(sb, blkno, ocfs2_find_actor, &args); |
| 126 | } |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 127 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, |
| 128 | int sysfile_type) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 129 | { |
| 130 | struct inode *inode = NULL; |
| 131 | struct super_block *sb = osb->sb; |
| 132 | struct ocfs2_find_inode_args args; |
Darrick J. Wong | 2931cdc | 2014-04-03 14:46:48 -0700 | [diff] [blame] | 133 | journal_t *journal = OCFS2_SB(sb)->journal->j_journal; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 134 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 135 | trace_ocfs2_iget_begin((unsigned long long)blkno, flags, |
| 136 | sysfile_type); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 137 | |
| 138 | /* Ok. By now we've either got the offsets passed to us by the |
| 139 | * caller, or we just pulled them off the bh. Lets do some |
| 140 | * sanity checks to make sure they're OK. */ |
| 141 | if (blkno == 0) { |
| 142 | inode = ERR_PTR(-EINVAL); |
| 143 | mlog_errno(PTR_ERR(inode)); |
| 144 | goto bail; |
| 145 | } |
| 146 | |
| 147 | args.fi_blkno = blkno; |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 148 | args.fi_flags = flags; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 149 | args.fi_ino = ino_from_blkno(sb, blkno); |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 150 | args.fi_sysfile_type = sysfile_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 151 | |
| 152 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, |
| 153 | ocfs2_init_locked_inode, &args); |
| 154 | /* inode was *not* in the inode cache. 2.6.x requires |
| 155 | * us to do our own read_inode call and unlock it |
| 156 | * afterwards. */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 157 | if (inode == NULL) { |
| 158 | inode = ERR_PTR(-ENOMEM); |
| 159 | mlog_errno(PTR_ERR(inode)); |
| 160 | goto bail; |
| 161 | } |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 162 | trace_ocfs2_iget5_locked(inode->i_state); |
Tao Ma | 6218b90 | 2011-02-21 11:18:30 +0800 | [diff] [blame] | 163 | if (inode->i_state & I_NEW) { |
Tao Ma | 6218b90 | 2011-02-21 11:18:30 +0800 | [diff] [blame] | 164 | ocfs2_read_locked_inode(inode, &args); |
| 165 | unlock_new_inode(inode); |
| 166 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 167 | if (is_bad_inode(inode)) { |
| 168 | iput(inode); |
| 169 | inode = ERR_PTR(-ESTALE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 170 | goto bail; |
| 171 | } |
| 172 | |
Darrick J. Wong | 2931cdc | 2014-04-03 14:46:48 -0700 | [diff] [blame] | 173 | /* |
| 174 | * Set transaction id's of transactions that have to be committed |
| 175 | * to finish f[data]sync. We set them to currently running transaction |
| 176 | * as we cannot be sure that the inode or some of its metadata isn't |
| 177 | * part of the transaction - the inode could have been reclaimed and |
| 178 | * now it is reread from disk. |
| 179 | */ |
| 180 | if (journal) { |
| 181 | transaction_t *transaction; |
| 182 | tid_t tid; |
| 183 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 184 | |
| 185 | read_lock(&journal->j_state_lock); |
| 186 | if (journal->j_running_transaction) |
| 187 | transaction = journal->j_running_transaction; |
| 188 | else |
| 189 | transaction = journal->j_committing_transaction; |
| 190 | if (transaction) |
| 191 | tid = transaction->t_tid; |
| 192 | else |
| 193 | tid = journal->j_commit_sequence; |
| 194 | read_unlock(&journal->j_state_lock); |
| 195 | oi->i_sync_tid = tid; |
| 196 | oi->i_datasync_tid = tid; |
| 197 | } |
| 198 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 199 | bail: |
| 200 | if (!IS_ERR(inode)) { |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 201 | trace_ocfs2_iget_end(inode, |
| 202 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
Mark Fasheh | 6a1bd4a | 2007-01-03 17:06:59 -0800 | [diff] [blame] | 203 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 204 | |
| 205 | return inode; |
| 206 | } |
| 207 | |
| 208 | |
| 209 | /* |
| 210 | * here's how inodes get read from disk: |
| 211 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR |
| 212 | * found? : return the in-memory inode |
| 213 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE |
| 214 | */ |
| 215 | |
| 216 | static int ocfs2_find_actor(struct inode *inode, void *opaque) |
| 217 | { |
| 218 | struct ocfs2_find_inode_args *args = NULL; |
| 219 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 220 | int ret = 0; |
| 221 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 222 | args = opaque; |
| 223 | |
| 224 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); |
| 225 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 226 | trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno); |
| 227 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 228 | if (oi->ip_blkno != args->fi_blkno) |
| 229 | goto bail; |
| 230 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 231 | ret = 1; |
| 232 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 233 | return ret; |
| 234 | } |
| 235 | |
| 236 | /* |
| 237 | * initialize the new inode, but don't do anything that would cause |
| 238 | * us to sleep. |
| 239 | * return 0 on success, 1 on failure |
| 240 | */ |
| 241 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) |
| 242 | { |
| 243 | struct ocfs2_find_inode_args *args = opaque; |
Jan Kara | cb25797 | 2009-06-04 15:26:50 +0200 | [diff] [blame] | 244 | static struct lock_class_key ocfs2_quota_ip_alloc_sem_key, |
| 245 | ocfs2_file_ip_alloc_sem_key; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 246 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 247 | inode->i_ino = args->fi_ino; |
| 248 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 249 | if (args->fi_sysfile_type != 0) |
| 250 | lockdep_set_class(&inode->i_mutex, |
| 251 | &ocfs2_sysfile_lock_key[args->fi_sysfile_type]); |
Jan Kara | cb25797 | 2009-06-04 15:26:50 +0200 | [diff] [blame] | 252 | if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE || |
| 253 | args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE || |
| 254 | args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE || |
| 255 | args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE) |
| 256 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, |
| 257 | &ocfs2_quota_ip_alloc_sem_key); |
| 258 | else |
| 259 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, |
| 260 | &ocfs2_file_ip_alloc_sem_key); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 261 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 262 | return 0; |
| 263 | } |
| 264 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 265 | void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, |
| 266 | int create_ino) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 267 | { |
| 268 | struct super_block *sb; |
| 269 | struct ocfs2_super *osb; |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 270 | int use_plocks = 1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 271 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 272 | sb = inode->i_sb; |
| 273 | osb = OCFS2_SB(sb); |
| 274 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 275 | if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) || |
| 276 | ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks()) |
| 277 | use_plocks = 0; |
| 278 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 279 | /* |
| 280 | * These have all been checked by ocfs2_read_inode_block() or set |
| 281 | * by ocfs2_mknod_locked(), so a failure is a code bug. |
| 282 | */ |
| 283 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode |
| 284 | cannot create a superblock |
| 285 | inode today. change if |
| 286 | that is needed. */ |
| 287 | BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))); |
| 288 | BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 289 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 290 | |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 291 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
| 292 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
Mark Fasheh | 15b1e36 | 2007-09-07 13:58:15 -0700 | [diff] [blame] | 293 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 294 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 295 | inode->i_version = 1; |
| 296 | inode->i_generation = le32_to_cpu(fe->i_generation); |
| 297 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); |
| 298 | inode->i_mode = le16_to_cpu(fe->i_mode); |
Eric W. Biederman | 2c03417 | 2013-01-31 17:27:11 -0800 | [diff] [blame] | 299 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
| 300 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 301 | |
| 302 | /* Fast symlinks will have i_size but no allocated clusters. */ |
Al Viro | ea022dfb | 2012-05-03 10:14:29 -0400 | [diff] [blame] | 303 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 304 | inode->i_blocks = 0; |
Al Viro | ea022dfb | 2012-05-03 10:14:29 -0400 | [diff] [blame] | 305 | inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops; |
| 306 | } else { |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 307 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
Al Viro | ea022dfb | 2012-05-03 10:14:29 -0400 | [diff] [blame] | 308 | inode->i_mapping->a_ops = &ocfs2_aops; |
| 309 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 310 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
| 311 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); |
| 312 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); |
| 313 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); |
| 314 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); |
| 315 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); |
| 316 | |
| 317 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) |
| 318 | mlog(ML_ERROR, |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 319 | "ip_blkno %llu != i_blkno %llu!\n", |
| 320 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 321 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 322 | |
Miklos Szeredi | bfe8684 | 2011-10-28 14:13:29 +0200 | [diff] [blame] | 323 | set_nlink(inode, ocfs2_read_links_count(fe)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 324 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 325 | trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno, |
| 326 | le32_to_cpu(fe->i_flags)); |
Jan Kara | bbbd0eb | 2008-08-21 18:22:30 +0200 | [diff] [blame] | 327 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) { |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 328 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; |
Jan Kara | bbbd0eb | 2008-08-21 18:22:30 +0200 | [diff] [blame] | 329 | inode->i_flags |= S_NOQUOTA; |
| 330 | } |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 331 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 332 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
| 333 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 334 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { |
| 335 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; |
Jan Kara | 1a224ad | 2008-08-20 15:43:36 +0200 | [diff] [blame] | 336 | } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) { |
| 337 | inode->i_flags |= S_NOQUOTA; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 338 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 339 | /* we can't actually hit this as read_inode can't |
| 340 | * handle superblocks today ;-) */ |
| 341 | BUG(); |
| 342 | } |
| 343 | |
| 344 | switch (inode->i_mode & S_IFMT) { |
| 345 | case S_IFREG: |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 346 | if (use_plocks) |
| 347 | inode->i_fop = &ocfs2_fops; |
| 348 | else |
| 349 | inode->i_fop = &ocfs2_fops_no_plocks; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 350 | inode->i_op = &ocfs2_file_iops; |
| 351 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
| 352 | break; |
| 353 | case S_IFDIR: |
| 354 | inode->i_op = &ocfs2_dir_iops; |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 355 | if (use_plocks) |
| 356 | inode->i_fop = &ocfs2_dops; |
| 357 | else |
| 358 | inode->i_fop = &ocfs2_dops_no_plocks; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 359 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
Goldwyn Rodrigues | 5e98d49 | 2010-06-28 10:04:32 -0500 | [diff] [blame] | 360 | OCFS2_I(inode)->ip_dir_lock_gen = 1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 361 | break; |
| 362 | case S_IFLNK: |
Al Viro | ea022dfb | 2012-05-03 10:14:29 -0400 | [diff] [blame] | 363 | inode->i_op = &ocfs2_symlink_inode_operations; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 364 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
| 365 | break; |
| 366 | default: |
| 367 | inode->i_op = &ocfs2_special_file_iops; |
| 368 | init_special_inode(inode, inode->i_mode, |
| 369 | inode->i_rdev); |
| 370 | break; |
| 371 | } |
| 372 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 373 | if (create_ino) { |
| 374 | inode->i_ino = ino_from_blkno(inode->i_sb, |
| 375 | le64_to_cpu(fe->i_blkno)); |
| 376 | |
| 377 | /* |
| 378 | * If we ever want to create system files from kernel, |
| 379 | * the generation argument to |
| 380 | * ocfs2_inode_lock_res_init() will have to change. |
| 381 | */ |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 382 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 383 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 384 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 385 | OCFS2_LOCK_TYPE_META, 0, inode); |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 386 | |
| 387 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
| 388 | OCFS2_LOCK_TYPE_OPEN, 0, inode); |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 389 | } |
| 390 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 391 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 392 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
| 393 | inode); |
| 394 | |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 395 | ocfs2_set_inode_flags(inode); |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 396 | |
Tao Ma | 1382115 | 2009-02-25 00:53:23 +0800 | [diff] [blame] | 397 | OCFS2_I(inode)->ip_last_used_slot = 0; |
| 398 | OCFS2_I(inode)->ip_last_used_group = 0; |
Mark Fasheh | e3b4a97 | 2009-12-07 13:16:07 -0800 | [diff] [blame] | 399 | |
| 400 | if (S_ISDIR(inode->i_mode)) |
| 401 | ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv, |
| 402 | OCFS2_RESV_FLAG_DIR); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 403 | } |
| 404 | |
| 405 | static int ocfs2_read_locked_inode(struct inode *inode, |
| 406 | struct ocfs2_find_inode_args *args) |
| 407 | { |
| 408 | struct super_block *sb; |
| 409 | struct ocfs2_super *osb; |
| 410 | struct ocfs2_dinode *fe; |
| 411 | struct buffer_head *bh = NULL; |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 412 | int status, can_lock; |
| 413 | u32 generation = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 414 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 415 | status = -EINVAL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 416 | sb = inode->i_sb; |
| 417 | osb = OCFS2_SB(sb); |
| 418 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 419 | /* |
| 420 | * To improve performance of cold-cache inode stats, we take |
| 421 | * the cluster lock here if possible. |
| 422 | * |
| 423 | * Generally, OCFS2 never trusts the contents of an inode |
| 424 | * unless it's holding a cluster lock, so taking it here isn't |
| 425 | * a correctness issue as much as it is a performance |
| 426 | * improvement. |
| 427 | * |
| 428 | * There are three times when taking the lock is not a good idea: |
| 429 | * |
| 430 | * 1) During startup, before we have initialized the DLM. |
| 431 | * |
| 432 | * 2) If we are reading certain system files which never get |
| 433 | * cluster locks (local alloc, truncate log). |
| 434 | * |
| 435 | * 3) If the process doing the iget() is responsible for |
| 436 | * orphan dir recovery. We're holding the orphan dir lock and |
| 437 | * can get into a deadlock with another process on another |
| 438 | * node in ->delete_inode(). |
| 439 | * |
| 440 | * #1 and #2 can be simply solved by never taking the lock |
| 441 | * here for system files (which are the only type we read |
| 442 | * during mount). It's a heavier approach, but our main |
Uwe Kleine-König | b595076 | 2010-11-01 15:38:34 -0400 | [diff] [blame] | 443 | * concern is user-accessible files anyway. |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 444 | * |
| 445 | * #3 works itself out because we'll eventually take the |
| 446 | * cluster lock before trusting anything anyway. |
| 447 | */ |
| 448 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 449 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 450 | && !ocfs2_mount_local(osb); |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 451 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 452 | trace_ocfs2_read_locked_inode( |
| 453 | (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock); |
| 454 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 455 | /* |
| 456 | * To maintain backwards compatibility with older versions of |
| 457 | * ocfs2-tools, we still store the generation value for system |
| 458 | * files. The only ones that actually matter to userspace are |
| 459 | * the journals, but it's easier and inexpensive to just flag |
| 460 | * all system files similarly. |
| 461 | */ |
| 462 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) |
| 463 | generation = osb->fs_generation; |
| 464 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 465 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 466 | OCFS2_LOCK_TYPE_META, |
| 467 | generation, inode); |
| 468 | |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 469 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
| 470 | OCFS2_LOCK_TYPE_OPEN, |
| 471 | 0, inode); |
| 472 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 473 | if (can_lock) { |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 474 | status = ocfs2_open_lock(inode); |
| 475 | if (status) { |
| 476 | make_bad_inode(inode); |
| 477 | mlog_errno(status); |
| 478 | return status; |
| 479 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 480 | status = ocfs2_inode_lock(inode, NULL, 0); |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 481 | if (status) { |
| 482 | make_bad_inode(inode); |
| 483 | mlog_errno(status); |
| 484 | return status; |
| 485 | } |
| 486 | } |
| 487 | |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 488 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
| 489 | status = ocfs2_try_open_lock(inode, 0); |
| 490 | if (status) { |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 491 | make_bad_inode(inode); |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 492 | return status; |
| 493 | } |
| 494 | } |
| 495 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 496 | if (can_lock) { |
| 497 | status = ocfs2_read_inode_block_full(inode, &bh, |
| 498 | OCFS2_BH_IGNORE_CACHE); |
| 499 | } else { |
Joel Becker | da1e909 | 2008-10-09 17:20:29 -0700 | [diff] [blame] | 500 | status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh); |
Sunil Mushran | f5ce5a0 | 2010-08-12 16:24:26 -0700 | [diff] [blame] | 501 | /* |
| 502 | * If buffer is in jbd, then its checksum may not have been |
| 503 | * computed as yet. |
| 504 | */ |
| 505 | if (!status && !buffer_jbd(bh)) |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 506 | status = ocfs2_validate_inode_block(osb->sb, bh); |
| 507 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 508 | if (status < 0) { |
| 509 | mlog_errno(status); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 510 | goto bail; |
| 511 | } |
| 512 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 513 | status = -EINVAL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 514 | fe = (struct ocfs2_dinode *) bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 515 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 516 | /* |
| 517 | * This is a code bug. Right now the caller needs to |
| 518 | * understand whether it is asking for a system file inode or |
| 519 | * not so the proper lock names can be built. |
| 520 | */ |
| 521 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != |
| 522 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), |
| 523 | "Inode %llu: system file state is ambigous\n", |
| 524 | (unsigned long long)args->fi_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 525 | |
| 526 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || |
| 527 | S_ISBLK(le16_to_cpu(fe->i_mode))) |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 528 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 529 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 530 | ocfs2_populate_inode(inode, fe, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 531 | |
| 532 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); |
| 533 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 534 | status = 0; |
| 535 | |
| 536 | bail: |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 537 | if (can_lock) |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 538 | ocfs2_inode_unlock(inode, 0); |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 539 | |
| 540 | if (status < 0) |
| 541 | make_bad_inode(inode); |
| 542 | |
Jan Kara | 4a635a1 | 2014-12-10 15:41:42 -0800 | [diff] [blame] | 543 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 544 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 545 | return status; |
| 546 | } |
| 547 | |
| 548 | void ocfs2_sync_blockdev(struct super_block *sb) |
| 549 | { |
| 550 | sync_blockdev(sb->s_bdev); |
| 551 | } |
| 552 | |
| 553 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, |
| 554 | struct inode *inode, |
| 555 | struct buffer_head *fe_bh) |
| 556 | { |
| 557 | int status = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 558 | struct ocfs2_dinode *fe; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 559 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 560 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 561 | fe = (struct ocfs2_dinode *) fe_bh->b_data; |
| 562 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 563 | /* |
| 564 | * This check will also skip truncate of inodes with inline |
| 565 | * data and fast symlinks. |
| 566 | */ |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 567 | if (fe->i_clusters) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 568 | if (ocfs2_should_order_data(inode)) |
| 569 | ocfs2_begin_ordered_truncate(inode, 0); |
| 570 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 571 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 572 | if (IS_ERR(handle)) { |
| 573 | status = PTR_ERR(handle); |
Dan Carpenter | 0350cb0 | 2010-04-22 11:39:29 +0200 | [diff] [blame] | 574 | handle = NULL; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 575 | mlog_errno(status); |
| 576 | goto out; |
| 577 | } |
| 578 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 579 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), |
| 580 | fe_bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 581 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 582 | if (status < 0) { |
| 583 | mlog_errno(status); |
| 584 | goto out; |
| 585 | } |
| 586 | |
| 587 | i_size_write(inode, 0); |
| 588 | |
| 589 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
| 590 | if (status < 0) { |
| 591 | mlog_errno(status); |
| 592 | goto out; |
| 593 | } |
| 594 | |
| 595 | ocfs2_commit_trans(osb, handle); |
| 596 | handle = NULL; |
| 597 | |
Tristan Ye | 78f9467 | 2010-05-11 17:54:42 +0800 | [diff] [blame] | 598 | status = ocfs2_commit_truncate(osb, inode, fe_bh); |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 599 | if (status < 0) { |
| 600 | mlog_errno(status); |
| 601 | goto out; |
| 602 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 603 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 604 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 605 | out: |
| 606 | if (handle) |
| 607 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 608 | return status; |
| 609 | } |
| 610 | |
| 611 | static int ocfs2_remove_inode(struct inode *inode, |
| 612 | struct buffer_head *di_bh, |
| 613 | struct inode *orphan_dir_inode, |
| 614 | struct buffer_head *orphan_dir_bh) |
| 615 | { |
| 616 | int status; |
| 617 | struct inode *inode_alloc_inode = NULL; |
| 618 | struct buffer_head *inode_alloc_bh = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 619 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 620 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 621 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; |
| 622 | |
| 623 | inode_alloc_inode = |
| 624 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, |
| 625 | le16_to_cpu(di->i_suballoc_slot)); |
| 626 | if (!inode_alloc_inode) { |
Joseph Qi | 62f8b1f | 2015-04-14 15:43:24 -0700 | [diff] [blame] | 627 | status = -ENOENT; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 628 | mlog_errno(status); |
| 629 | goto bail; |
| 630 | } |
| 631 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 632 | mutex_lock(&inode_alloc_inode->i_mutex); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 633 | status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 634 | if (status < 0) { |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 635 | mutex_unlock(&inode_alloc_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 636 | |
| 637 | mlog_errno(status); |
| 638 | goto bail; |
| 639 | } |
| 640 | |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 641 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS + |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 642 | ocfs2_quota_trans_credits(inode->i_sb)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 643 | if (IS_ERR(handle)) { |
| 644 | status = PTR_ERR(handle); |
| 645 | mlog_errno(status); |
| 646 | goto bail_unlock; |
| 647 | } |
| 648 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 649 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
| 650 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, |
Joseph Qi | 06ee5c7 | 2015-02-16 15:59:54 -0800 | [diff] [blame] | 651 | orphan_dir_bh, false); |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 652 | if (status < 0) { |
| 653 | mlog_errno(status); |
| 654 | goto bail_commit; |
| 655 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 656 | } |
| 657 | |
| 658 | /* set the inodes dtime */ |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 659 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 660 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 661 | if (status < 0) { |
| 662 | mlog_errno(status); |
| 663 | goto bail_commit; |
| 664 | } |
| 665 | |
| 666 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); |
Marcin Slusarz | 4092d49 | 2007-12-25 15:52:59 +0100 | [diff] [blame] | 667 | di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 668 | ocfs2_journal_dirty(handle, di_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 669 | |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 670 | ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh); |
Christoph Hellwig | 63936dd | 2010-03-03 09:05:01 -0500 | [diff] [blame] | 671 | dquot_free_inode(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 672 | |
| 673 | status = ocfs2_free_dinode(handle, inode_alloc_inode, |
| 674 | inode_alloc_bh, di); |
| 675 | if (status < 0) |
| 676 | mlog_errno(status); |
| 677 | |
| 678 | bail_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 679 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 680 | bail_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 681 | ocfs2_inode_unlock(inode_alloc_inode, 1); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 682 | mutex_unlock(&inode_alloc_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 683 | brelse(inode_alloc_bh); |
| 684 | bail: |
| 685 | iput(inode_alloc_inode); |
| 686 | |
| 687 | return status; |
| 688 | } |
| 689 | |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 690 | /* |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 691 | * Serialize with orphan dir recovery. If the process doing |
| 692 | * recovery on this orphan dir does an iget() with the dir |
| 693 | * i_mutex held, we'll deadlock here. Instead we detect this |
| 694 | * and exit early - recovery will wipe this inode for us. |
| 695 | */ |
| 696 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, |
| 697 | int slot) |
| 698 | { |
| 699 | int ret = 0; |
| 700 | |
| 701 | spin_lock(&osb->osb_lock); |
| 702 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 703 | ret = -EDEADLK; |
| 704 | goto out; |
| 705 | } |
| 706 | /* This signals to the orphan recovery process that it should |
| 707 | * wait for us to handle the wipe. */ |
| 708 | osb->osb_orphan_wipes[slot]++; |
| 709 | out: |
| 710 | spin_unlock(&osb->osb_lock); |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 711 | trace_ocfs2_check_orphan_recovery_state(slot, ret); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 712 | return ret; |
| 713 | } |
| 714 | |
| 715 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, |
| 716 | int slot) |
| 717 | { |
| 718 | spin_lock(&osb->osb_lock); |
| 719 | osb->osb_orphan_wipes[slot]--; |
| 720 | spin_unlock(&osb->osb_lock); |
| 721 | |
| 722 | wake_up(&osb->osb_wipe_event); |
| 723 | } |
| 724 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 725 | static int ocfs2_wipe_inode(struct inode *inode, |
| 726 | struct buffer_head *di_bh) |
| 727 | { |
Joel Becker | d577632 | 2010-05-03 19:15:49 -0700 | [diff] [blame] | 728 | int status, orphaned_slot = -1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 729 | struct inode *orphan_dir_inode = NULL; |
| 730 | struct buffer_head *orphan_dir_bh = NULL; |
| 731 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 732 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 733 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 734 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
| 735 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 736 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 737 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); |
| 738 | if (status) |
| 739 | return status; |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 740 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 741 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
| 742 | ORPHAN_DIR_SYSTEM_INODE, |
| 743 | orphaned_slot); |
| 744 | if (!orphan_dir_inode) { |
Joseph Qi | 62f8b1f | 2015-04-14 15:43:24 -0700 | [diff] [blame] | 745 | status = -ENOENT; |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 746 | mlog_errno(status); |
| 747 | goto bail; |
| 748 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 749 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 750 | /* Lock the orphan dir. The lock will be held for the entire |
| 751 | * delete_inode operation. We do this now to avoid races with |
| 752 | * recovery completion on other nodes. */ |
| 753 | mutex_lock(&orphan_dir_inode->i_mutex); |
| 754 | status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
| 755 | if (status < 0) { |
| 756 | mutex_unlock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 757 | |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 758 | mlog_errno(status); |
| 759 | goto bail; |
| 760 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 761 | } |
| 762 | |
| 763 | /* we do this while holding the orphan dir lock because we |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 764 | * don't want recovery being run from another node to try an |
| 765 | * inode delete underneath us -- this will result in two nodes |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 766 | * truncating the same file! */ |
| 767 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); |
| 768 | if (status < 0) { |
| 769 | mlog_errno(status); |
| 770 | goto bail_unlock_dir; |
| 771 | } |
| 772 | |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 773 | /* Remove any dir index tree */ |
| 774 | if (S_ISDIR(inode->i_mode)) { |
| 775 | status = ocfs2_dx_dir_truncate(inode, di_bh); |
| 776 | if (status) { |
| 777 | mlog_errno(status); |
| 778 | goto bail_unlock_dir; |
| 779 | } |
| 780 | } |
| 781 | |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 782 | /*Free extended attribute resources associated with this inode.*/ |
| 783 | status = ocfs2_xattr_remove(inode, di_bh); |
| 784 | if (status < 0) { |
| 785 | mlog_errno(status); |
| 786 | goto bail_unlock_dir; |
| 787 | } |
| 788 | |
Tao Ma | 8b2c0db | 2009-08-18 11:43:49 +0800 | [diff] [blame] | 789 | status = ocfs2_remove_refcount_tree(inode, di_bh); |
| 790 | if (status < 0) { |
| 791 | mlog_errno(status); |
| 792 | goto bail_unlock_dir; |
| 793 | } |
| 794 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 795 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, |
| 796 | orphan_dir_bh); |
| 797 | if (status < 0) |
| 798 | mlog_errno(status); |
| 799 | |
| 800 | bail_unlock_dir: |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 801 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR) |
| 802 | return status; |
| 803 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 804 | ocfs2_inode_unlock(orphan_dir_inode, 1); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 805 | mutex_unlock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 806 | brelse(orphan_dir_bh); |
| 807 | bail: |
| 808 | iput(orphan_dir_inode); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 809 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 810 | |
| 811 | return status; |
| 812 | } |
| 813 | |
| 814 | /* There is a series of simple checks that should be done before a |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 815 | * trylock is even considered. Encapsulate those in this function. */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 816 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) |
| 817 | { |
| 818 | int ret = 0; |
| 819 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 820 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 821 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 822 | trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task, |
| 823 | (unsigned long long)oi->ip_blkno, |
| 824 | oi->ip_flags); |
| 825 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 826 | /* We shouldn't be getting here for the root directory |
| 827 | * inode.. */ |
| 828 | if (inode == osb->root_inode) { |
| 829 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); |
| 830 | goto bail; |
| 831 | } |
| 832 | |
Jan Kara | bd62ad7 | 2014-04-03 14:46:54 -0700 | [diff] [blame] | 833 | /* |
| 834 | * If we're coming from downconvert_thread we can't go into our own |
| 835 | * voting [hello, deadlock city!] so we cannot delete the inode. But |
| 836 | * since we dropped last inode ref when downconverting dentry lock, |
| 837 | * we cannot have the file open and thus the node doing unlink will |
| 838 | * take care of deleting the inode. |
| 839 | */ |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 840 | if (current == osb->dc_task) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 841 | goto bail; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 842 | |
| 843 | spin_lock(&oi->ip_lock); |
| 844 | /* OCFS2 *never* deletes system files. This should technically |
| 845 | * never get here as system file inodes should always have a |
| 846 | * positive link count. */ |
| 847 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 848 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
| 849 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 850 | goto bail_unlock; |
| 851 | } |
| 852 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 853 | ret = 1; |
| 854 | bail_unlock: |
| 855 | spin_unlock(&oi->ip_lock); |
| 856 | bail: |
| 857 | return ret; |
| 858 | } |
| 859 | |
| 860 | /* Query the cluster to determine whether we should wipe an inode from |
| 861 | * disk or not. |
| 862 | * |
| 863 | * Requires the inode to have the cluster lock. */ |
| 864 | static int ocfs2_query_inode_wipe(struct inode *inode, |
| 865 | struct buffer_head *di_bh, |
| 866 | int *wipe) |
| 867 | { |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 868 | int status = 0, reason = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 869 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 870 | struct ocfs2_dinode *di; |
| 871 | |
| 872 | *wipe = 0; |
| 873 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 874 | trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno, |
| 875 | inode->i_nlink); |
| 876 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 877 | /* While we were waiting for the cluster lock in |
| 878 | * ocfs2_delete_inode, another node might have asked to delete |
| 879 | * the inode. Recheck our flags to catch this. */ |
| 880 | if (!ocfs2_inode_is_valid_to_delete(inode)) { |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 881 | reason = 1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 882 | goto bail; |
| 883 | } |
| 884 | |
| 885 | /* Now that we have an up to date inode, we can double check |
| 886 | * the link count. */ |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 887 | if (inode->i_nlink) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 888 | goto bail; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 889 | |
| 890 | /* Do some basic inode verification... */ |
| 891 | di = (struct ocfs2_dinode *) di_bh->b_data; |
Li Dongyang | d4cd187 | 2010-04-22 16:11:19 +0800 | [diff] [blame] | 892 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) && |
| 893 | !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
Tristan Ye | b54c2ca | 2010-03-19 09:21:10 +0800 | [diff] [blame] | 894 | /* |
| 895 | * Inodes in the orphan dir must have ORPHANED_FL. The only |
| 896 | * inodes that come back out of the orphan dir are reflink |
| 897 | * targets. A reflink target may be moved out of the orphan |
| 898 | * dir between the time we scan the directory and the time we |
| 899 | * process it. This would lead to HAS_REFCOUNT_FL being set but |
| 900 | * ORPHANED_FL not. |
| 901 | */ |
| 902 | if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) { |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 903 | reason = 2; |
Tristan Ye | b54c2ca | 2010-03-19 09:21:10 +0800 | [diff] [blame] | 904 | goto bail; |
| 905 | } |
| 906 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 907 | /* for lack of a better error? */ |
| 908 | status = -EEXIST; |
| 909 | mlog(ML_ERROR, |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 910 | "Inode %llu (on-disk %llu) not orphaned! " |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 911 | "Disk flags 0x%x, inode flags 0x%x\n", |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 912 | (unsigned long long)oi->ip_blkno, |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 913 | (unsigned long long)le64_to_cpu(di->i_blkno), |
| 914 | le32_to_cpu(di->i_flags), oi->ip_flags); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 915 | goto bail; |
| 916 | } |
| 917 | |
| 918 | /* has someone already deleted us?! baaad... */ |
| 919 | if (di->i_dtime) { |
| 920 | status = -EEXIST; |
| 921 | mlog_errno(status); |
| 922 | goto bail; |
| 923 | } |
| 924 | |
Mark Fasheh | 6f16bf6 | 2007-03-20 17:17:54 -0700 | [diff] [blame] | 925 | /* |
| 926 | * This is how ocfs2 determines whether an inode is still live |
| 927 | * within the cluster. Every node takes a shared read lock on |
| 928 | * the inode open lock in ocfs2_read_locked_inode(). When we |
| 929 | * get to ->delete_inode(), each node tries to convert it's |
| 930 | * lock to an exclusive. Trylocks are serialized by the inode |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 931 | * meta data lock. If the upconvert succeeds, we know the inode |
Mark Fasheh | 6f16bf6 | 2007-03-20 17:17:54 -0700 | [diff] [blame] | 932 | * is no longer live and can be deleted. |
| 933 | * |
| 934 | * Though we call this with the meta data lock held, the |
| 935 | * trylock keeps us from ABBA deadlock. |
| 936 | */ |
| 937 | status = ocfs2_try_open_lock(inode, 1); |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 938 | if (status == -EAGAIN) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 939 | status = 0; |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 940 | reason = 3; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 941 | goto bail; |
| 942 | } |
| 943 | if (status < 0) { |
| 944 | mlog_errno(status); |
| 945 | goto bail; |
| 946 | } |
| 947 | |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 948 | *wipe = 1; |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 949 | trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 950 | |
| 951 | bail: |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 952 | trace_ocfs2_query_inode_wipe_end(status, reason); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 953 | return status; |
| 954 | } |
| 955 | |
| 956 | /* Support function for ocfs2_delete_inode. Will help us keep the |
| 957 | * inode data in a consistent state for clear_inode. Always truncates |
| 958 | * pages, optionally sync's them first. */ |
| 959 | static void ocfs2_cleanup_delete_inode(struct inode *inode, |
| 960 | int sync_data) |
| 961 | { |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 962 | trace_ocfs2_cleanup_delete_inode( |
| 963 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 964 | if (sync_data) |
Jan Kara | 249ec93 | 2011-11-08 00:20:39 +0100 | [diff] [blame] | 965 | filemap_write_and_wait(inode->i_mapping); |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 966 | truncate_inode_pages_final(&inode->i_data); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 967 | } |
| 968 | |
Al Viro | 066d92d | 2010-06-08 21:28:10 -0400 | [diff] [blame] | 969 | static void ocfs2_delete_inode(struct inode *inode) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 970 | { |
| 971 | int wipe, status; |
Joel Becker | e4b963f | 2009-09-02 17:17:36 -0700 | [diff] [blame] | 972 | sigset_t oldset; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 973 | struct buffer_head *di_bh = NULL; |
| 974 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 975 | trace_ocfs2_delete_inode(inode->i_ino, |
| 976 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 977 | is_bad_inode(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 978 | |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 979 | /* When we fail in read_inode() we mark inode as bad. The second test |
| 980 | * catches the case when inode allocation fails before allocating |
| 981 | * a block for inode. */ |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 982 | if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 983 | goto bail; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 984 | |
| 985 | if (!ocfs2_inode_is_valid_to_delete(inode)) { |
| 986 | /* It's probably not necessary to truncate_inode_pages |
| 987 | * here but we do it for safety anyway (it will most |
| 988 | * likely be a no-op anyway) */ |
| 989 | ocfs2_cleanup_delete_inode(inode, 0); |
| 990 | goto bail; |
| 991 | } |
| 992 | |
Jan Kara | bd62ad7 | 2014-04-03 14:46:54 -0700 | [diff] [blame] | 993 | dquot_initialize(inode); |
| 994 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 995 | /* We want to block signals in delete_inode as the lock and |
| 996 | * messaging paths may return us -ERESTARTSYS. Which would |
| 997 | * cause us to exit early, resulting in inodes being orphaned |
| 998 | * forever. */ |
Joel Becker | e4b963f | 2009-09-02 17:17:36 -0700 | [diff] [blame] | 999 | ocfs2_block_signals(&oldset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1000 | |
wengang wang | 6ca497a | 2009-03-06 21:29:10 +0800 | [diff] [blame] | 1001 | /* |
| 1002 | * Synchronize us against ocfs2_get_dentry. We take this in |
| 1003 | * shared mode so that all nodes can still concurrently |
| 1004 | * process deletes. |
| 1005 | */ |
| 1006 | status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0); |
| 1007 | if (status < 0) { |
| 1008 | mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status); |
| 1009 | ocfs2_cleanup_delete_inode(inode, 0); |
| 1010 | goto bail_unblock; |
| 1011 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1012 | /* Lock down the inode. This gives us an up to date view of |
| 1013 | * it's metadata (for verification), and allows us to |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 1014 | * serialize delete_inode on multiple nodes. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1015 | * |
| 1016 | * Even though we might be doing a truncate, we don't take the |
| 1017 | * allocation lock here as it won't be needed - nobody will |
| 1018 | * have the file open. |
| 1019 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1020 | status = ocfs2_inode_lock(inode, &di_bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1021 | if (status < 0) { |
| 1022 | if (status != -ENOENT) |
| 1023 | mlog_errno(status); |
| 1024 | ocfs2_cleanup_delete_inode(inode, 0); |
wengang wang | 6ca497a | 2009-03-06 21:29:10 +0800 | [diff] [blame] | 1025 | goto bail_unlock_nfs_sync; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1026 | } |
| 1027 | |
| 1028 | /* Query the cluster. This will be the final decision made |
| 1029 | * before we go ahead and wipe the inode. */ |
| 1030 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); |
| 1031 | if (!wipe || status < 0) { |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 1032 | /* Error and remote inode busy both mean we won't be |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1033 | * removing the inode, so they take almost the same |
| 1034 | * path. */ |
| 1035 | if (status < 0) |
| 1036 | mlog_errno(status); |
| 1037 | |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 1038 | /* Someone in the cluster has disallowed a wipe of |
| 1039 | * this inode, or it was never completely |
| 1040 | * orphaned. Write out the pages and exit now. */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1041 | ocfs2_cleanup_delete_inode(inode, 1); |
| 1042 | goto bail_unlock_inode; |
| 1043 | } |
| 1044 | |
| 1045 | ocfs2_cleanup_delete_inode(inode, 0); |
| 1046 | |
| 1047 | status = ocfs2_wipe_inode(inode, di_bh); |
| 1048 | if (status < 0) { |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1049 | if (status != -EDEADLK) |
| 1050 | mlog_errno(status); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1051 | goto bail_unlock_inode; |
| 1052 | } |
| 1053 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 1054 | /* |
| 1055 | * Mark the inode as successfully deleted. |
| 1056 | * |
| 1057 | * This is important for ocfs2_clear_inode() as it will check |
| 1058 | * this flag and skip any checkpointing work |
| 1059 | * |
| 1060 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate |
| 1061 | * the LVB for other nodes. |
| 1062 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1063 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
| 1064 | |
| 1065 | bail_unlock_inode: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1066 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1067 | brelse(di_bh); |
wengang wang | 6ca497a | 2009-03-06 21:29:10 +0800 | [diff] [blame] | 1068 | |
| 1069 | bail_unlock_nfs_sync: |
| 1070 | ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0); |
| 1071 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1072 | bail_unblock: |
Joel Becker | e4b963f | 2009-09-02 17:17:36 -0700 | [diff] [blame] | 1073 | ocfs2_unblock_signals(&oldset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1074 | bail: |
Tao Ma | c1e8d35 | 2011-03-07 16:43:21 +0800 | [diff] [blame] | 1075 | return; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1076 | } |
| 1077 | |
Al Viro | 066d92d | 2010-06-08 21:28:10 -0400 | [diff] [blame] | 1078 | static void ocfs2_clear_inode(struct inode *inode) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1079 | { |
| 1080 | int status; |
| 1081 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Jan Kara | 84d86f8 | 2014-04-03 14:46:57 -0700 | [diff] [blame] | 1082 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1083 | |
Jan Kara | dbd5768 | 2012-05-03 14:48:02 +0200 | [diff] [blame] | 1084 | clear_inode(inode); |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 1085 | trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno, |
| 1086 | inode->i_nlink); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1087 | |
| 1088 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, |
| 1089 | "Inode=%lu\n", inode->i_ino); |
| 1090 | |
Christoph Hellwig | 9f75475 | 2010-03-03 09:05:05 -0500 | [diff] [blame] | 1091 | dquot_drop(inode); |
Christoph Hellwig | 257ba15 | 2010-03-03 09:05:04 -0500 | [diff] [blame] | 1092 | |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 1093 | /* To preven remote deletes we hold open lock before, now it |
| 1094 | * is time to unlock PR and EX open locks. */ |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 1095 | ocfs2_open_unlock(inode); |
| 1096 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1097 | /* Do these before all the other work so that we don't bounce |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 1098 | * the downconvert thread while waiting to destroy the locks. */ |
Jan Kara | 84d86f8 | 2014-04-03 14:46:57 -0700 | [diff] [blame] | 1099 | ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres); |
| 1100 | ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres); |
| 1101 | ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1102 | |
Mark Fasheh | 4fe370a | 2009-12-07 13:15:40 -0800 | [diff] [blame] | 1103 | ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap, |
| 1104 | &oi->ip_la_data_resv); |
| 1105 | ocfs2_resv_init_once(&oi->ip_la_data_resv); |
| 1106 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1107 | /* We very well may get a clear_inode before all an inodes |
| 1108 | * metadata has hit disk. Of course, we can't drop any cluster |
| 1109 | * locks until the journal has finished with it. The only |
| 1110 | * exception here are successfully wiped inodes - their |
| 1111 | * metadata can now be considered to be part of the system |
| 1112 | * inodes from which it came. */ |
| 1113 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) |
| 1114 | ocfs2_checkpoint_inode(inode); |
| 1115 | |
| 1116 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1117 | "Clear inode of %llu, inode has io markers\n", |
| 1118 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1119 | |
Mark Fasheh | 8341897 | 2007-04-23 18:53:12 -0700 | [diff] [blame] | 1120 | ocfs2_extent_map_trunc(inode, 0); |
| 1121 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1122 | status = ocfs2_drop_inode_locks(inode); |
| 1123 | if (status < 0) |
| 1124 | mlog_errno(status); |
| 1125 | |
| 1126 | ocfs2_lock_res_free(&oi->ip_rw_lockres); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1127 | ocfs2_lock_res_free(&oi->ip_inode_lockres); |
Tiger Yang | 5000863 | 2007-03-20 16:01:38 -0700 | [diff] [blame] | 1128 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1129 | |
Joel Becker | 66fb345 | 2009-02-12 15:24:40 -0800 | [diff] [blame] | 1130 | ocfs2_metadata_cache_exit(INODE_CACHE(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1131 | |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1132 | mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached, |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1133 | "Clear inode of %llu, inode has %u cache items\n", |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1134 | (unsigned long long)oi->ip_blkno, |
| 1135 | INODE_CACHE(inode)->ci_num_cached); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1136 | |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1137 | mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE), |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1138 | "Clear inode of %llu, inode has a bad flag\n", |
| 1139 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1140 | |
| 1141 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1142 | "Clear inode of %llu, inode is locked\n", |
| 1143 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1144 | |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 1145 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1146 | "Clear inode of %llu, io_mutex is locked\n", |
| 1147 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 1148 | mutex_unlock(&oi->ip_io_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1149 | |
| 1150 | /* |
| 1151 | * down_trylock() returns 0, down_write_trylock() returns 1 |
| 1152 | * kernel 1, world 0 |
| 1153 | */ |
| 1154 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1155 | "Clear inode of %llu, alloc_sem is locked\n", |
| 1156 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1157 | up_write(&oi->ip_alloc_sem); |
| 1158 | |
| 1159 | mlog_bug_on_msg(oi->ip_open_count, |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1160 | "Clear inode of %llu has open count %d\n", |
| 1161 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1162 | |
| 1163 | /* Clear all other flags. */ |
Joel Becker | 47460d6 | 2009-02-10 16:05:07 -0800 | [diff] [blame] | 1164 | oi->ip_flags = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1165 | oi->ip_dir_start_lookup = 0; |
| 1166 | oi->ip_blkno = 0ULL; |
Sunil Mushran | ae0dff6 | 2008-10-22 13:24:29 -0700 | [diff] [blame] | 1167 | |
| 1168 | /* |
| 1169 | * ip_jinode is used to track txns against this inode. We ensure that |
| 1170 | * the journal is flushed before journal shutdown. Thus it is safe to |
| 1171 | * have inodes get cleaned up after journal shutdown. |
| 1172 | */ |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1173 | jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal, |
| 1174 | &oi->ip_jinode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1175 | } |
| 1176 | |
Al Viro | 066d92d | 2010-06-08 21:28:10 -0400 | [diff] [blame] | 1177 | void ocfs2_evict_inode(struct inode *inode) |
| 1178 | { |
| 1179 | if (!inode->i_nlink || |
| 1180 | (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) { |
| 1181 | ocfs2_delete_inode(inode); |
| 1182 | } else { |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 1183 | truncate_inode_pages_final(&inode->i_data); |
Al Viro | 066d92d | 2010-06-08 21:28:10 -0400 | [diff] [blame] | 1184 | } |
| 1185 | ocfs2_clear_inode(inode); |
| 1186 | } |
| 1187 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1188 | /* Called under inode_lock, with no more references on the |
| 1189 | * struct inode, so it's safe here to check the flags field |
| 1190 | * and to manipulate i_nlink without any other locks. */ |
Al Viro | 45321ac | 2010-06-07 13:43:19 -0400 | [diff] [blame] | 1191 | int ocfs2_drop_inode(struct inode *inode) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1192 | { |
| 1193 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Al Viro | 45321ac | 2010-06-07 13:43:19 -0400 | [diff] [blame] | 1194 | int res; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1195 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 1196 | trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno, |
| 1197 | inode->i_nlink, oi->ip_flags); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1198 | |
Mark Fasheh | 379dfe9 | 2006-09-08 14:21:03 -0700 | [diff] [blame] | 1199 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
Al Viro | 45321ac | 2010-06-07 13:43:19 -0400 | [diff] [blame] | 1200 | res = 1; |
Mark Fasheh | 379dfe9 | 2006-09-08 14:21:03 -0700 | [diff] [blame] | 1201 | else |
Al Viro | 45321ac | 2010-06-07 13:43:19 -0400 | [diff] [blame] | 1202 | res = generic_drop_inode(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1203 | |
Al Viro | 45321ac | 2010-06-07 13:43:19 -0400 | [diff] [blame] | 1204 | return res; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1205 | } |
| 1206 | |
| 1207 | /* |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1208 | * This is called from our getattr. |
| 1209 | */ |
| 1210 | int ocfs2_inode_revalidate(struct dentry *dentry) |
| 1211 | { |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 1212 | struct inode *inode = d_inode(dentry); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1213 | int status = 0; |
| 1214 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 1215 | trace_ocfs2_inode_revalidate(inode, |
| 1216 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL, |
| 1217 | inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1218 | |
| 1219 | if (!inode) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1220 | status = -ENOENT; |
| 1221 | goto bail; |
| 1222 | } |
| 1223 | |
| 1224 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 1225 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 1226 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1227 | status = -ENOENT; |
| 1228 | goto bail; |
| 1229 | } |
| 1230 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 1231 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1232 | /* Let ocfs2_inode_lock do the work of updating our struct |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1233 | * inode for us. */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1234 | status = ocfs2_inode_lock(inode, NULL, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1235 | if (status < 0) { |
| 1236 | if (status != -ENOENT) |
| 1237 | mlog_errno(status); |
| 1238 | goto bail; |
| 1239 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1240 | ocfs2_inode_unlock(inode, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1241 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1242 | return status; |
| 1243 | } |
| 1244 | |
| 1245 | /* |
| 1246 | * Updates a disk inode from a |
| 1247 | * struct inode. |
| 1248 | * Only takes ip_lock. |
| 1249 | */ |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 1250 | int ocfs2_mark_inode_dirty(handle_t *handle, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1251 | struct inode *inode, |
| 1252 | struct buffer_head *bh) |
| 1253 | { |
| 1254 | int status; |
| 1255 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; |
| 1256 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 1257 | trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1258 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 1259 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1260 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1261 | if (status < 0) { |
| 1262 | mlog_errno(status); |
| 1263 | goto leave; |
| 1264 | } |
| 1265 | |
| 1266 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 1267 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); |
Jan Kara | 6e4b0d5 | 2007-04-27 11:08:01 -0700 | [diff] [blame] | 1268 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 1269 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
Mark Fasheh | 15b1e36 | 2007-09-07 13:58:15 -0700 | [diff] [blame] | 1270 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1271 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 1272 | |
| 1273 | fe->i_size = cpu_to_le64(i_size_read(inode)); |
Mark Fasheh | 198a1ca | 2008-11-20 17:54:57 -0800 | [diff] [blame] | 1274 | ocfs2_set_links_count(fe, inode->i_nlink); |
Eric W. Biederman | 2c03417 | 2013-01-31 17:27:11 -0800 | [diff] [blame] | 1275 | fe->i_uid = cpu_to_le32(i_uid_read(inode)); |
| 1276 | fe->i_gid = cpu_to_le32(i_gid_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1277 | fe->i_mode = cpu_to_le16(inode->i_mode); |
| 1278 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 1279 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
| 1280 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 1281 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 1282 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); |
| 1283 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); |
| 1284 | |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 1285 | ocfs2_journal_dirty(handle, bh); |
Darrick J. Wong | 2931cdc | 2014-04-03 14:46:48 -0700 | [diff] [blame] | 1286 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1287 | leave: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1288 | return status; |
| 1289 | } |
| 1290 | |
| 1291 | /* |
| 1292 | * |
| 1293 | * Updates a struct inode from a disk inode. |
| 1294 | * does no i/o, only takes ip_lock. |
| 1295 | */ |
| 1296 | void ocfs2_refresh_inode(struct inode *inode, |
| 1297 | struct ocfs2_dinode *fe) |
| 1298 | { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1299 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 1300 | |
| 1301 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 1302 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
Mark Fasheh | 15b1e36 | 2007-09-07 13:58:15 -0700 | [diff] [blame] | 1303 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 1304 | ocfs2_set_inode_flags(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1305 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
Miklos Szeredi | bfe8684 | 2011-10-28 14:13:29 +0200 | [diff] [blame] | 1306 | set_nlink(inode, ocfs2_read_links_count(fe)); |
Eric W. Biederman | 2c03417 | 2013-01-31 17:27:11 -0800 | [diff] [blame] | 1307 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
| 1308 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1309 | inode->i_mode = le16_to_cpu(fe->i_mode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1310 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
| 1311 | inode->i_blocks = 0; |
| 1312 | else |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 1313 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1314 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
| 1315 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); |
| 1316 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); |
| 1317 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); |
| 1318 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); |
| 1319 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); |
| 1320 | |
| 1321 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 1322 | } |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1323 | |
| 1324 | int ocfs2_validate_inode_block(struct super_block *sb, |
| 1325 | struct buffer_head *bh) |
| 1326 | { |
Joel Becker | d6b32bb | 2008-10-17 14:55:01 -0700 | [diff] [blame] | 1327 | int rc; |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1328 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
| 1329 | |
Tao Ma | 64f3b26 | 2011-02-22 22:24:57 +0800 | [diff] [blame] | 1330 | trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr); |
Joel Becker | 970e493 | 2008-11-13 14:49:19 -0800 | [diff] [blame] | 1331 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1332 | BUG_ON(!buffer_uptodate(bh)); |
| 1333 | |
Joel Becker | d6b32bb | 2008-10-17 14:55:01 -0700 | [diff] [blame] | 1334 | /* |
| 1335 | * If the ecc fails, we return the error but otherwise |
| 1336 | * leave the filesystem running. We know any error is |
| 1337 | * local to this block. |
| 1338 | */ |
| 1339 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1340 | if (rc) { |
| 1341 | mlog(ML_ERROR, "Checksum failed for dinode %llu\n", |
| 1342 | (unsigned long long)bh->b_blocknr); |
Joel Becker | d6b32bb | 2008-10-17 14:55:01 -0700 | [diff] [blame] | 1343 | goto bail; |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1344 | } |
Joel Becker | d6b32bb | 2008-10-17 14:55:01 -0700 | [diff] [blame] | 1345 | |
| 1346 | /* |
| 1347 | * Errors after here are fatal. |
| 1348 | */ |
| 1349 | |
| 1350 | rc = -EINVAL; |
| 1351 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1352 | if (!OCFS2_IS_VALID_DINODE(di)) { |
| 1353 | ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n", |
| 1354 | (unsigned long long)bh->b_blocknr, 7, |
| 1355 | di->i_signature); |
| 1356 | goto bail; |
| 1357 | } |
| 1358 | |
| 1359 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { |
| 1360 | ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n", |
| 1361 | (unsigned long long)bh->b_blocknr, |
| 1362 | (unsigned long long)le64_to_cpu(di->i_blkno)); |
| 1363 | goto bail; |
| 1364 | } |
| 1365 | |
| 1366 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { |
| 1367 | ocfs2_error(sb, |
| 1368 | "Invalid dinode #%llu: OCFS2_VALID_FL not set\n", |
| 1369 | (unsigned long long)bh->b_blocknr); |
| 1370 | goto bail; |
| 1371 | } |
| 1372 | |
| 1373 | if (le32_to_cpu(di->i_fs_generation) != |
| 1374 | OCFS2_SB(sb)->fs_generation) { |
| 1375 | ocfs2_error(sb, |
| 1376 | "Invalid dinode #%llu: fs_generation is %u\n", |
| 1377 | (unsigned long long)bh->b_blocknr, |
| 1378 | le32_to_cpu(di->i_fs_generation)); |
| 1379 | goto bail; |
| 1380 | } |
| 1381 | |
| 1382 | rc = 0; |
| 1383 | |
| 1384 | bail: |
| 1385 | return rc; |
| 1386 | } |
| 1387 | |
| 1388 | int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh, |
| 1389 | int flags) |
| 1390 | { |
| 1391 | int rc; |
| 1392 | struct buffer_head *tmp = *bh; |
| 1393 | |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1394 | rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno, |
| 1395 | 1, &tmp, flags, ocfs2_validate_inode_block); |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1396 | |
| 1397 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ |
Joel Becker | 970e493 | 2008-11-13 14:49:19 -0800 | [diff] [blame] | 1398 | if (!rc && !*bh) |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1399 | *bh = tmp; |
| 1400 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1401 | return rc; |
| 1402 | } |
| 1403 | |
| 1404 | int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh) |
| 1405 | { |
| 1406 | return ocfs2_read_inode_block_full(inode, bh, 0); |
| 1407 | } |
Joel Becker | 6e5a3d7 | 2009-02-10 19:00:37 -0800 | [diff] [blame] | 1408 | |
Joel Becker | 6e5a3d7 | 2009-02-10 19:00:37 -0800 | [diff] [blame] | 1409 | |
| 1410 | static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci) |
| 1411 | { |
| 1412 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1413 | |
| 1414 | return oi->ip_blkno; |
| 1415 | } |
| 1416 | |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1417 | static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci) |
| 1418 | { |
| 1419 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1420 | |
| 1421 | return oi->vfs_inode.i_sb; |
| 1422 | } |
| 1423 | |
Joel Becker | 6e5a3d7 | 2009-02-10 19:00:37 -0800 | [diff] [blame] | 1424 | static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci) |
| 1425 | { |
| 1426 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1427 | |
| 1428 | spin_lock(&oi->ip_lock); |
| 1429 | } |
| 1430 | |
| 1431 | static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci) |
| 1432 | { |
| 1433 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1434 | |
| 1435 | spin_unlock(&oi->ip_lock); |
| 1436 | } |
| 1437 | |
| 1438 | static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci) |
| 1439 | { |
| 1440 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1441 | |
| 1442 | mutex_lock(&oi->ip_io_mutex); |
| 1443 | } |
| 1444 | |
| 1445 | static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci) |
| 1446 | { |
| 1447 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); |
| 1448 | |
| 1449 | mutex_unlock(&oi->ip_io_mutex); |
| 1450 | } |
| 1451 | |
| 1452 | const struct ocfs2_caching_operations ocfs2_inode_caching_ops = { |
| 1453 | .co_owner = ocfs2_inode_cache_owner, |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1454 | .co_get_super = ocfs2_inode_cache_get_super, |
Joel Becker | 6e5a3d7 | 2009-02-10 19:00:37 -0800 | [diff] [blame] | 1455 | .co_cache_lock = ocfs2_inode_cache_lock, |
| 1456 | .co_cache_unlock = ocfs2_inode_cache_unlock, |
| 1457 | .co_io_lock = ocfs2_inode_cache_io_lock, |
| 1458 | .co_io_unlock = ocfs2_inode_cache_io_unlock, |
| 1459 | }; |
| 1460 | |