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 | * journal.c |
| 5 | * |
| 6 | * Defines functions of journalling api |
| 7 | * |
| 8 | * Copyright (C) 2003, 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> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/highmem.h> |
| 30 | #include <linux/kthread.h> |
| 31 | |
| 32 | #define MLOG_MASK_PREFIX ML_JOURNAL |
| 33 | #include <cluster/masklog.h> |
| 34 | |
| 35 | #include "ocfs2.h" |
| 36 | |
| 37 | #include "alloc.h" |
Mark Fasheh | 316f4b9 | 2007-09-07 18:21:26 -0700 | [diff] [blame] | 38 | #include "dir.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 39 | #include "dlmglue.h" |
| 40 | #include "extent_map.h" |
| 41 | #include "heartbeat.h" |
| 42 | #include "inode.h" |
| 43 | #include "journal.h" |
| 44 | #include "localalloc.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 45 | #include "slot_map.h" |
| 46 | #include "super.h" |
| 47 | #include "vote.h" |
| 48 | #include "sysfile.h" |
| 49 | |
| 50 | #include "buffer_head_io.h" |
| 51 | |
Ingo Molnar | 34af946 | 2006-06-27 02:53:55 -0700 | [diff] [blame] | 52 | DEFINE_SPINLOCK(trans_inc_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 53 | |
| 54 | static int ocfs2_force_read_journal(struct inode *inode); |
| 55 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
| 56 | int node_num); |
| 57 | static int __ocfs2_recovery_thread(void *arg); |
| 58 | static int ocfs2_commit_cache(struct ocfs2_super *osb); |
| 59 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 60 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
| 61 | int dirty); |
| 62 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 63 | int slot_num); |
| 64 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
| 65 | int slot); |
| 66 | static int ocfs2_commit_thread(void *arg); |
| 67 | |
| 68 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
| 69 | { |
| 70 | int status = 0; |
| 71 | unsigned int flushed; |
| 72 | unsigned long old_id; |
| 73 | struct ocfs2_journal *journal = NULL; |
| 74 | |
| 75 | mlog_entry_void(); |
| 76 | |
| 77 | journal = osb->journal; |
| 78 | |
| 79 | /* Flush all pending commits and checkpoint the journal. */ |
| 80 | down_write(&journal->j_trans_barrier); |
| 81 | |
| 82 | if (atomic_read(&journal->j_num_trans) == 0) { |
| 83 | up_write(&journal->j_trans_barrier); |
| 84 | mlog(0, "No transactions for me to flush!\n"); |
| 85 | goto finally; |
| 86 | } |
| 87 | |
| 88 | journal_lock_updates(journal->j_journal); |
| 89 | status = journal_flush(journal->j_journal); |
| 90 | journal_unlock_updates(journal->j_journal); |
| 91 | if (status < 0) { |
| 92 | up_write(&journal->j_trans_barrier); |
| 93 | mlog_errno(status); |
| 94 | goto finally; |
| 95 | } |
| 96 | |
| 97 | old_id = ocfs2_inc_trans_id(journal); |
| 98 | |
| 99 | flushed = atomic_read(&journal->j_num_trans); |
| 100 | atomic_set(&journal->j_num_trans, 0); |
| 101 | up_write(&journal->j_trans_barrier); |
| 102 | |
| 103 | mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", |
| 104 | journal->j_trans_id, flushed); |
| 105 | |
| 106 | ocfs2_kick_vote_thread(osb); |
| 107 | wake_up(&journal->j_checkpointed); |
| 108 | finally: |
| 109 | mlog_exit(status); |
| 110 | return status; |
| 111 | } |
| 112 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 113 | /* pass it NULL and it will allocate a new handle object for you. If |
| 114 | * you pass it a handle however, it may still return error, in which |
| 115 | * case it has free'd the passed handle for you. */ |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 116 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 117 | { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 118 | journal_t *journal = osb->journal->j_journal; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 119 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 120 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 121 | BUG_ON(!osb || !osb->journal->j_journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 122 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 123 | if (ocfs2_is_hard_readonly(osb)) |
| 124 | return ERR_PTR(-EROFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 125 | |
| 126 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); |
| 127 | BUG_ON(max_buffs <= 0); |
| 128 | |
| 129 | /* JBD might support this, but our journalling code doesn't yet. */ |
| 130 | if (journal_current_handle()) { |
| 131 | mlog(ML_ERROR, "Recursive transaction attempted!\n"); |
| 132 | BUG(); |
| 133 | } |
| 134 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 135 | down_read(&osb->journal->j_trans_barrier); |
| 136 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 137 | handle = journal_start(journal, max_buffs); |
| 138 | if (IS_ERR(handle)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 139 | up_read(&osb->journal->j_trans_barrier); |
| 140 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 141 | mlog_errno(PTR_ERR(handle)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 142 | |
| 143 | if (is_journal_aborted(journal)) { |
| 144 | ocfs2_abort(osb->sb, "Detected aborted journal"); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 145 | handle = ERR_PTR(-EROFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 146 | } |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 147 | } else { |
| 148 | if (!ocfs2_mount_local(osb)) |
| 149 | atomic_inc(&(osb->journal->j_num_trans)); |
| 150 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 151 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 152 | return handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 153 | } |
| 154 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 155 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
| 156 | handle_t *handle) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 157 | { |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 158 | int ret; |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 159 | struct ocfs2_journal *journal = osb->journal; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 160 | |
| 161 | BUG_ON(!handle); |
| 162 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 163 | ret = journal_stop(handle); |
| 164 | if (ret < 0) |
| 165 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 166 | |
| 167 | up_read(&journal->j_trans_barrier); |
| 168 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 169 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 170 | } |
| 171 | |
| 172 | /* |
| 173 | * 'nblocks' is what you want to add to the current |
| 174 | * transaction. extend_trans will either extend the current handle by |
| 175 | * nblocks, or commit it and start a new one with nblocks credits. |
| 176 | * |
| 177 | * WARNING: This will not release any semaphores or disk locks taken |
| 178 | * during the transaction, so make sure they were taken *before* |
| 179 | * start_trans or we'll have ordering deadlocks. |
| 180 | * |
| 181 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is |
| 182 | * good because transaction ids haven't yet been recorded on the |
| 183 | * cluster locks associated with this handle. |
| 184 | */ |
Mark Fasheh | 1fc5814 | 2006-10-05 14:15:36 -0700 | [diff] [blame] | 185 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 186 | { |
| 187 | int status; |
| 188 | |
| 189 | BUG_ON(!handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 190 | BUG_ON(!nblocks); |
| 191 | |
| 192 | mlog_entry_void(); |
| 193 | |
| 194 | mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); |
| 195 | |
Mark Fasheh | 0879c58 | 2007-12-03 16:42:19 -0800 | [diff] [blame^] | 196 | #ifdef OCFS2_DEBUG_FS |
| 197 | status = 1; |
| 198 | #else |
Mark Fasheh | 1fc5814 | 2006-10-05 14:15:36 -0700 | [diff] [blame] | 199 | status = journal_extend(handle, nblocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 200 | if (status < 0) { |
| 201 | mlog_errno(status); |
| 202 | goto bail; |
| 203 | } |
Mark Fasheh | 0879c58 | 2007-12-03 16:42:19 -0800 | [diff] [blame^] | 204 | #endif |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 205 | |
| 206 | if (status > 0) { |
| 207 | mlog(0, "journal_extend failed, trying journal_restart\n"); |
Mark Fasheh | 1fc5814 | 2006-10-05 14:15:36 -0700 | [diff] [blame] | 208 | status = journal_restart(handle, nblocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 209 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 210 | mlog_errno(status); |
| 211 | goto bail; |
| 212 | } |
Mark Fasheh | 01ddf1e | 2006-10-05 13:54:39 -0700 | [diff] [blame] | 213 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 214 | |
| 215 | status = 0; |
| 216 | bail: |
| 217 | |
| 218 | mlog_exit(status); |
| 219 | return status; |
| 220 | } |
| 221 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 222 | int ocfs2_journal_access(handle_t *handle, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 223 | struct inode *inode, |
| 224 | struct buffer_head *bh, |
| 225 | int type) |
| 226 | { |
| 227 | int status; |
| 228 | |
| 229 | BUG_ON(!inode); |
| 230 | BUG_ON(!handle); |
| 231 | BUG_ON(!bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 232 | |
Badari Pulavarty | 205f87f | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 233 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 234 | (unsigned long long)bh->b_blocknr, type, |
| 235 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? |
| 236 | "OCFS2_JOURNAL_ACCESS_CREATE" : |
| 237 | "OCFS2_JOURNAL_ACCESS_WRITE", |
| 238 | bh->b_size); |
| 239 | |
| 240 | /* we can safely remove this assertion after testing. */ |
| 241 | if (!buffer_uptodate(bh)) { |
| 242 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); |
| 243 | mlog(ML_ERROR, "b_blocknr=%llu\n", |
| 244 | (unsigned long long)bh->b_blocknr); |
| 245 | BUG(); |
| 246 | } |
| 247 | |
| 248 | /* Set the current transaction information on the inode so |
| 249 | * that the locking code knows whether it can drop it's locks |
| 250 | * on this inode or not. We're protected from the commit |
| 251 | * thread updating the current transaction id until |
| 252 | * ocfs2_commit_trans() because ocfs2_start_trans() took |
| 253 | * j_trans_barrier for us. */ |
| 254 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); |
| 255 | |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 256 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 257 | switch (type) { |
| 258 | case OCFS2_JOURNAL_ACCESS_CREATE: |
| 259 | case OCFS2_JOURNAL_ACCESS_WRITE: |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 260 | status = journal_get_write_access(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 261 | break; |
| 262 | |
| 263 | case OCFS2_JOURNAL_ACCESS_UNDO: |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 264 | status = journal_get_undo_access(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 265 | break; |
| 266 | |
| 267 | default: |
| 268 | status = -EINVAL; |
| 269 | mlog(ML_ERROR, "Uknown access type!\n"); |
| 270 | } |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 271 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 272 | |
| 273 | if (status < 0) |
| 274 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", |
| 275 | status, type); |
| 276 | |
| 277 | mlog_exit(status); |
| 278 | return status; |
| 279 | } |
| 280 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 281 | int ocfs2_journal_dirty(handle_t *handle, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 282 | struct buffer_head *bh) |
| 283 | { |
| 284 | int status; |
| 285 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 286 | mlog_entry("(bh->b_blocknr=%llu)\n", |
| 287 | (unsigned long long)bh->b_blocknr); |
| 288 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 289 | status = journal_dirty_metadata(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 290 | if (status < 0) |
| 291 | mlog(ML_ERROR, "Could not dirty metadata buffer. " |
| 292 | "(bh->b_blocknr=%llu)\n", |
| 293 | (unsigned long long)bh->b_blocknr); |
| 294 | |
| 295 | mlog_exit(status); |
| 296 | return status; |
| 297 | } |
| 298 | |
| 299 | int ocfs2_journal_dirty_data(handle_t *handle, |
| 300 | struct buffer_head *bh) |
| 301 | { |
| 302 | int err = journal_dirty_data(handle, bh); |
| 303 | if (err) |
| 304 | mlog_errno(err); |
| 305 | /* TODO: When we can handle it, abort the handle and go RO on |
| 306 | * error here. */ |
| 307 | |
| 308 | return err; |
| 309 | } |
| 310 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 311 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) |
| 312 | |
| 313 | void ocfs2_set_journal_params(struct ocfs2_super *osb) |
| 314 | { |
| 315 | journal_t *journal = osb->journal->j_journal; |
| 316 | |
| 317 | spin_lock(&journal->j_state_lock); |
| 318 | journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
| 319 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
| 320 | journal->j_flags |= JFS_BARRIER; |
| 321 | else |
| 322 | journal->j_flags &= ~JFS_BARRIER; |
| 323 | spin_unlock(&journal->j_state_lock); |
| 324 | } |
| 325 | |
| 326 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) |
| 327 | { |
| 328 | int status = -1; |
| 329 | struct inode *inode = NULL; /* the journal inode */ |
| 330 | journal_t *j_journal = NULL; |
| 331 | struct ocfs2_dinode *di = NULL; |
| 332 | struct buffer_head *bh = NULL; |
| 333 | struct ocfs2_super *osb; |
| 334 | int meta_lock = 0; |
| 335 | |
| 336 | mlog_entry_void(); |
| 337 | |
| 338 | BUG_ON(!journal); |
| 339 | |
| 340 | osb = journal->j_osb; |
| 341 | |
| 342 | /* already have the inode for our journal */ |
| 343 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 344 | osb->slot_num); |
| 345 | if (inode == NULL) { |
| 346 | status = -EACCES; |
| 347 | mlog_errno(status); |
| 348 | goto done; |
| 349 | } |
| 350 | if (is_bad_inode(inode)) { |
| 351 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 352 | iput(inode); |
| 353 | inode = NULL; |
| 354 | status = -EACCES; |
| 355 | goto done; |
| 356 | } |
| 357 | |
| 358 | SET_INODE_JOURNAL(inode); |
| 359 | OCFS2_I(inode)->ip_open_count++; |
| 360 | |
Mark Fasheh | 6eff579 | 2006-01-18 10:31:47 -0800 | [diff] [blame] | 361 | /* Skip recovery waits here - journal inode metadata never |
| 362 | * changes in a live cluster so it can be considered an |
| 363 | * exception to the rule. */ |
Mark Fasheh | 4bcec18 | 2006-10-09 16:02:40 -0700 | [diff] [blame] | 364 | status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 365 | if (status < 0) { |
| 366 | if (status != -ERESTARTSYS) |
| 367 | mlog(ML_ERROR, "Could not get lock on journal!\n"); |
| 368 | goto done; |
| 369 | } |
| 370 | |
| 371 | meta_lock = 1; |
| 372 | di = (struct ocfs2_dinode *)bh->b_data; |
| 373 | |
| 374 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { |
| 375 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", |
| 376 | inode->i_size); |
| 377 | status = -EINVAL; |
| 378 | goto done; |
| 379 | } |
| 380 | |
| 381 | mlog(0, "inode->i_size = %lld\n", inode->i_size); |
Andrew Morton | 5515eff | 2006-03-26 01:37:53 -0800 | [diff] [blame] | 382 | mlog(0, "inode->i_blocks = %llu\n", |
| 383 | (unsigned long long)inode->i_blocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 384 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
| 385 | |
| 386 | /* call the kernels journal init function now */ |
| 387 | j_journal = journal_init_inode(inode); |
| 388 | if (j_journal == NULL) { |
| 389 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 390 | status = -EINVAL; |
| 391 | goto done; |
| 392 | } |
| 393 | |
| 394 | mlog(0, "Returned from journal_init_inode\n"); |
| 395 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); |
| 396 | |
| 397 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 398 | OCFS2_JOURNAL_DIRTY_FL); |
| 399 | |
| 400 | journal->j_journal = j_journal; |
| 401 | journal->j_inode = inode; |
| 402 | journal->j_bh = bh; |
| 403 | |
| 404 | ocfs2_set_journal_params(osb); |
| 405 | |
| 406 | journal->j_state = OCFS2_JOURNAL_LOADED; |
| 407 | |
| 408 | status = 0; |
| 409 | done: |
| 410 | if (status < 0) { |
| 411 | if (meta_lock) |
| 412 | ocfs2_meta_unlock(inode, 1); |
| 413 | if (bh != NULL) |
| 414 | brelse(bh); |
| 415 | if (inode) { |
| 416 | OCFS2_I(inode)->ip_open_count--; |
| 417 | iput(inode); |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | mlog_exit(status); |
| 422 | return status; |
| 423 | } |
| 424 | |
| 425 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
| 426 | int dirty) |
| 427 | { |
| 428 | int status; |
| 429 | unsigned int flags; |
| 430 | struct ocfs2_journal *journal = osb->journal; |
| 431 | struct buffer_head *bh = journal->j_bh; |
| 432 | struct ocfs2_dinode *fe; |
| 433 | |
| 434 | mlog_entry_void(); |
| 435 | |
| 436 | fe = (struct ocfs2_dinode *)bh->b_data; |
| 437 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 438 | /* This is called from startup/shutdown which will |
| 439 | * handle the errors in a specific manner, so no need |
| 440 | * to call ocfs2_error() here. */ |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 441 | mlog(ML_ERROR, "Journal dinode %llu has invalid " |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 442 | "signature: %.*s", |
| 443 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 444 | fe->i_signature); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 445 | status = -EIO; |
| 446 | goto out; |
| 447 | } |
| 448 | |
| 449 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 450 | if (dirty) |
| 451 | flags |= OCFS2_JOURNAL_DIRTY_FL; |
| 452 | else |
| 453 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 454 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 455 | |
| 456 | status = ocfs2_write_block(osb, bh, journal->j_inode); |
| 457 | if (status < 0) |
| 458 | mlog_errno(status); |
| 459 | |
| 460 | out: |
| 461 | mlog_exit(status); |
| 462 | return status; |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * If the journal has been kmalloc'd it needs to be freed after this |
| 467 | * call. |
| 468 | */ |
| 469 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) |
| 470 | { |
| 471 | struct ocfs2_journal *journal = NULL; |
| 472 | int status = 0; |
| 473 | struct inode *inode = NULL; |
| 474 | int num_running_trans = 0; |
| 475 | |
| 476 | mlog_entry_void(); |
| 477 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 478 | BUG_ON(!osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 479 | |
| 480 | journal = osb->journal; |
| 481 | if (!journal) |
| 482 | goto done; |
| 483 | |
| 484 | inode = journal->j_inode; |
| 485 | |
| 486 | if (journal->j_state != OCFS2_JOURNAL_LOADED) |
| 487 | goto done; |
| 488 | |
| 489 | /* need to inc inode use count as journal_destroy will iput. */ |
| 490 | if (!igrab(inode)) |
| 491 | BUG(); |
| 492 | |
| 493 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); |
| 494 | if (num_running_trans > 0) |
| 495 | mlog(0, "Shutting down journal: must wait on %d " |
| 496 | "running transactions!\n", |
| 497 | num_running_trans); |
| 498 | |
| 499 | /* Do a commit_cache here. It will flush our journal, *and* |
| 500 | * release any locks that are still held. |
| 501 | * set the SHUTDOWN flag and release the trans lock. |
| 502 | * the commit thread will take the trans lock for us below. */ |
| 503 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; |
| 504 | |
| 505 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not |
| 506 | * drop the trans_lock (which we want to hold until we |
| 507 | * completely destroy the journal. */ |
| 508 | if (osb->commit_task) { |
| 509 | /* Wait for the commit thread */ |
| 510 | mlog(0, "Waiting for ocfs2commit to exit....\n"); |
| 511 | kthread_stop(osb->commit_task); |
| 512 | osb->commit_task = NULL; |
| 513 | } |
| 514 | |
| 515 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); |
| 516 | |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 517 | if (ocfs2_mount_local(osb)) { |
| 518 | journal_lock_updates(journal->j_journal); |
| 519 | status = journal_flush(journal->j_journal); |
| 520 | journal_unlock_updates(journal->j_journal); |
| 521 | if (status < 0) |
| 522 | mlog_errno(status); |
| 523 | } |
| 524 | |
| 525 | if (status == 0) { |
| 526 | /* |
| 527 | * Do not toggle if flush was unsuccessful otherwise |
| 528 | * will leave dirty metadata in a "clean" journal |
| 529 | */ |
| 530 | status = ocfs2_journal_toggle_dirty(osb, 0); |
| 531 | if (status < 0) |
| 532 | mlog_errno(status); |
| 533 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 534 | |
| 535 | /* Shutdown the kernel journal system */ |
| 536 | journal_destroy(journal->j_journal); |
| 537 | |
| 538 | OCFS2_I(inode)->ip_open_count--; |
| 539 | |
| 540 | /* unlock our journal */ |
| 541 | ocfs2_meta_unlock(inode, 1); |
| 542 | |
| 543 | brelse(journal->j_bh); |
| 544 | journal->j_bh = NULL; |
| 545 | |
| 546 | journal->j_state = OCFS2_JOURNAL_FREE; |
| 547 | |
| 548 | // up_write(&journal->j_trans_barrier); |
| 549 | done: |
| 550 | if (inode) |
| 551 | iput(inode); |
| 552 | mlog_exit_void(); |
| 553 | } |
| 554 | |
| 555 | static void ocfs2_clear_journal_error(struct super_block *sb, |
| 556 | journal_t *journal, |
| 557 | int slot) |
| 558 | { |
| 559 | int olderr; |
| 560 | |
| 561 | olderr = journal_errno(journal); |
| 562 | if (olderr) { |
| 563 | mlog(ML_ERROR, "File system error %d recorded in " |
| 564 | "journal %u.\n", olderr, slot); |
| 565 | mlog(ML_ERROR, "File system on device %s needs checking.\n", |
| 566 | sb->s_id); |
| 567 | |
| 568 | journal_ack_err(journal); |
| 569 | journal_clear_err(journal); |
| 570 | } |
| 571 | } |
| 572 | |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 573 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 574 | { |
| 575 | int status = 0; |
| 576 | struct ocfs2_super *osb; |
| 577 | |
| 578 | mlog_entry_void(); |
| 579 | |
| 580 | if (!journal) |
| 581 | BUG(); |
| 582 | |
| 583 | osb = journal->j_osb; |
| 584 | |
| 585 | status = journal_load(journal->j_journal); |
| 586 | if (status < 0) { |
| 587 | mlog(ML_ERROR, "Failed to load journal!\n"); |
| 588 | goto done; |
| 589 | } |
| 590 | |
| 591 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); |
| 592 | |
| 593 | status = ocfs2_journal_toggle_dirty(osb, 1); |
| 594 | if (status < 0) { |
| 595 | mlog_errno(status); |
| 596 | goto done; |
| 597 | } |
| 598 | |
| 599 | /* Launch the commit thread */ |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 600 | if (!local) { |
| 601 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, |
| 602 | "ocfs2cmt"); |
| 603 | if (IS_ERR(osb->commit_task)) { |
| 604 | status = PTR_ERR(osb->commit_task); |
| 605 | osb->commit_task = NULL; |
| 606 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " |
| 607 | "error=%d", status); |
| 608 | goto done; |
| 609 | } |
| 610 | } else |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 611 | osb->commit_task = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 612 | |
| 613 | done: |
| 614 | mlog_exit(status); |
| 615 | return status; |
| 616 | } |
| 617 | |
| 618 | |
| 619 | /* 'full' flag tells us whether we clear out all blocks or if we just |
| 620 | * mark the journal clean */ |
| 621 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) |
| 622 | { |
| 623 | int status; |
| 624 | |
| 625 | mlog_entry_void(); |
| 626 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 627 | BUG_ON(!journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 628 | |
| 629 | status = journal_wipe(journal->j_journal, full); |
| 630 | if (status < 0) { |
| 631 | mlog_errno(status); |
| 632 | goto bail; |
| 633 | } |
| 634 | |
| 635 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); |
| 636 | if (status < 0) |
| 637 | mlog_errno(status); |
| 638 | |
| 639 | bail: |
| 640 | mlog_exit(status); |
| 641 | return status; |
| 642 | } |
| 643 | |
| 644 | /* |
| 645 | * JBD Might read a cached version of another nodes journal file. We |
| 646 | * don't want this as this file changes often and we get no |
| 647 | * notification on those changes. The only way to be sure that we've |
| 648 | * got the most up to date version of those blocks then is to force |
| 649 | * read them off disk. Just searching through the buffer cache won't |
| 650 | * work as there may be pages backing this file which are still marked |
| 651 | * up to date. We know things can't change on this file underneath us |
| 652 | * as we have the lock by now :) |
| 653 | */ |
| 654 | static int ocfs2_force_read_journal(struct inode *inode) |
| 655 | { |
| 656 | int status = 0; |
Mark Fasheh | 4f902c3 | 2007-03-09 16:26:50 -0800 | [diff] [blame] | 657 | int i; |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 658 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
Mark Fasheh | 4f902c3 | 2007-03-09 16:26:50 -0800 | [diff] [blame] | 659 | #define CONCURRENT_JOURNAL_FILL 32ULL |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 660 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
| 661 | |
| 662 | mlog_entry_void(); |
| 663 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 664 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
| 665 | |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 666 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 667 | v_blkno = 0; |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 668 | while (v_blkno < num_blocks) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 669 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 670 | &p_blkno, &p_blocks, NULL); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 671 | if (status < 0) { |
| 672 | mlog_errno(status); |
| 673 | goto bail; |
| 674 | } |
| 675 | |
| 676 | if (p_blocks > CONCURRENT_JOURNAL_FILL) |
| 677 | p_blocks = CONCURRENT_JOURNAL_FILL; |
| 678 | |
Mark Fasheh | dd4a2c2 | 2006-04-12 14:24:05 -0700 | [diff] [blame] | 679 | /* We are reading journal data which should not |
| 680 | * be put in the uptodate cache */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 681 | status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), |
| 682 | p_blkno, p_blocks, bhs, 0, |
Mark Fasheh | dd4a2c2 | 2006-04-12 14:24:05 -0700 | [diff] [blame] | 683 | NULL); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 684 | if (status < 0) { |
| 685 | mlog_errno(status); |
| 686 | goto bail; |
| 687 | } |
| 688 | |
| 689 | for(i = 0; i < p_blocks; i++) { |
| 690 | brelse(bhs[i]); |
| 691 | bhs[i] = NULL; |
| 692 | } |
| 693 | |
| 694 | v_blkno += p_blocks; |
| 695 | } |
| 696 | |
| 697 | bail: |
| 698 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) |
| 699 | if (bhs[i]) |
| 700 | brelse(bhs[i]); |
| 701 | mlog_exit(status); |
| 702 | return status; |
| 703 | } |
| 704 | |
| 705 | struct ocfs2_la_recovery_item { |
| 706 | struct list_head lri_list; |
| 707 | int lri_slot; |
| 708 | struct ocfs2_dinode *lri_la_dinode; |
| 709 | struct ocfs2_dinode *lri_tl_dinode; |
| 710 | }; |
| 711 | |
| 712 | /* Does the second half of the recovery process. By this point, the |
| 713 | * node is marked clean and can actually be considered recovered, |
| 714 | * hence it's no longer in the recovery map, but there's still some |
| 715 | * cleanup we can do which shouldn't happen within the recovery thread |
| 716 | * as locking in that context becomes very difficult if we are to take |
| 717 | * recovering nodes into account. |
| 718 | * |
| 719 | * NOTE: This function can and will sleep on recovery of other nodes |
| 720 | * during cluster locking, just like any other ocfs2 process. |
| 721 | */ |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 722 | void ocfs2_complete_recovery(struct work_struct *work) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 723 | { |
| 724 | int ret; |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 725 | struct ocfs2_journal *journal = |
| 726 | container_of(work, struct ocfs2_journal, j_recovery_work); |
| 727 | struct ocfs2_super *osb = journal->j_osb; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 728 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
Christoph Hellwig | 800deef | 2007-05-17 16:03:13 +0200 | [diff] [blame] | 729 | struct ocfs2_la_recovery_item *item, *n; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 730 | LIST_HEAD(tmp_la_list); |
| 731 | |
| 732 | mlog_entry_void(); |
| 733 | |
| 734 | mlog(0, "completing recovery from keventd\n"); |
| 735 | |
| 736 | spin_lock(&journal->j_lock); |
| 737 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); |
| 738 | spin_unlock(&journal->j_lock); |
| 739 | |
Christoph Hellwig | 800deef | 2007-05-17 16:03:13 +0200 | [diff] [blame] | 740 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 741 | list_del_init(&item->lri_list); |
| 742 | |
| 743 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); |
| 744 | |
| 745 | la_dinode = item->lri_la_dinode; |
| 746 | if (la_dinode) { |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 747 | mlog(0, "Clean up local alloc %llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 748 | (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 749 | |
| 750 | ret = ocfs2_complete_local_alloc_recovery(osb, |
| 751 | la_dinode); |
| 752 | if (ret < 0) |
| 753 | mlog_errno(ret); |
| 754 | |
| 755 | kfree(la_dinode); |
| 756 | } |
| 757 | |
| 758 | tl_dinode = item->lri_tl_dinode; |
| 759 | if (tl_dinode) { |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 760 | mlog(0, "Clean up truncate log %llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 761 | (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 762 | |
| 763 | ret = ocfs2_complete_truncate_log_recovery(osb, |
| 764 | tl_dinode); |
| 765 | if (ret < 0) |
| 766 | mlog_errno(ret); |
| 767 | |
| 768 | kfree(tl_dinode); |
| 769 | } |
| 770 | |
| 771 | ret = ocfs2_recover_orphans(osb, item->lri_slot); |
| 772 | if (ret < 0) |
| 773 | mlog_errno(ret); |
| 774 | |
| 775 | kfree(item); |
| 776 | } |
| 777 | |
| 778 | mlog(0, "Recovery completion\n"); |
| 779 | mlog_exit_void(); |
| 780 | } |
| 781 | |
| 782 | /* NOTE: This function always eats your references to la_dinode and |
| 783 | * tl_dinode, either manually on error, or by passing them to |
| 784 | * ocfs2_complete_recovery */ |
| 785 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
| 786 | int slot_num, |
| 787 | struct ocfs2_dinode *la_dinode, |
| 788 | struct ocfs2_dinode *tl_dinode) |
| 789 | { |
| 790 | struct ocfs2_la_recovery_item *item; |
| 791 | |
Sunil Mushran | afae00ab | 2006-04-12 14:37:00 -0700 | [diff] [blame] | 792 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 793 | if (!item) { |
| 794 | /* Though we wish to avoid it, we are in fact safe in |
| 795 | * skipping local alloc cleanup as fsck.ocfs2 is more |
| 796 | * than capable of reclaiming unused space. */ |
| 797 | if (la_dinode) |
| 798 | kfree(la_dinode); |
| 799 | |
| 800 | if (tl_dinode) |
| 801 | kfree(tl_dinode); |
| 802 | |
| 803 | mlog_errno(-ENOMEM); |
| 804 | return; |
| 805 | } |
| 806 | |
| 807 | INIT_LIST_HEAD(&item->lri_list); |
| 808 | item->lri_la_dinode = la_dinode; |
| 809 | item->lri_slot = slot_num; |
| 810 | item->lri_tl_dinode = tl_dinode; |
| 811 | |
| 812 | spin_lock(&journal->j_lock); |
| 813 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); |
| 814 | queue_work(ocfs2_wq, &journal->j_recovery_work); |
| 815 | spin_unlock(&journal->j_lock); |
| 816 | } |
| 817 | |
| 818 | /* Called by the mount code to queue recovery the last part of |
| 819 | * recovery for it's own slot. */ |
| 820 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
| 821 | { |
| 822 | struct ocfs2_journal *journal = osb->journal; |
| 823 | |
| 824 | if (osb->dirty) { |
| 825 | /* No need to queue up our truncate_log as regular |
| 826 | * cleanup will catch that. */ |
| 827 | ocfs2_queue_recovery_completion(journal, |
| 828 | osb->slot_num, |
| 829 | osb->local_alloc_copy, |
| 830 | NULL); |
| 831 | ocfs2_schedule_truncate_log_flush(osb, 0); |
| 832 | |
| 833 | osb->local_alloc_copy = NULL; |
| 834 | osb->dirty = 0; |
| 835 | } |
| 836 | } |
| 837 | |
| 838 | static int __ocfs2_recovery_thread(void *arg) |
| 839 | { |
| 840 | int status, node_num; |
| 841 | struct ocfs2_super *osb = arg; |
| 842 | |
| 843 | mlog_entry_void(); |
| 844 | |
| 845 | status = ocfs2_wait_on_mount(osb); |
| 846 | if (status < 0) { |
| 847 | goto bail; |
| 848 | } |
| 849 | |
| 850 | restart: |
| 851 | status = ocfs2_super_lock(osb, 1); |
| 852 | if (status < 0) { |
| 853 | mlog_errno(status); |
| 854 | goto bail; |
| 855 | } |
| 856 | |
| 857 | while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { |
| 858 | node_num = ocfs2_node_map_first_set_bit(osb, |
| 859 | &osb->recovery_map); |
| 860 | if (node_num == O2NM_INVALID_NODE_NUM) { |
| 861 | mlog(0, "Out of nodes to recover.\n"); |
| 862 | break; |
| 863 | } |
| 864 | |
| 865 | status = ocfs2_recover_node(osb, node_num); |
| 866 | if (status < 0) { |
| 867 | mlog(ML_ERROR, |
| 868 | "Error %d recovering node %d on device (%u,%u)!\n", |
| 869 | status, node_num, |
| 870 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); |
| 871 | mlog(ML_ERROR, "Volume requires unmount.\n"); |
| 872 | continue; |
| 873 | } |
| 874 | |
| 875 | ocfs2_recovery_map_clear(osb, node_num); |
| 876 | } |
| 877 | ocfs2_super_unlock(osb, 1); |
| 878 | |
| 879 | /* We always run recovery on our own orphan dir - the dead |
| 880 | * node(s) may have voted "no" on an inode delete earlier. A |
| 881 | * revote is therefore required. */ |
| 882 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, |
| 883 | NULL); |
| 884 | |
| 885 | bail: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 886 | mutex_lock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 887 | if (!status && |
| 888 | !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 889 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 890 | goto restart; |
| 891 | } |
| 892 | |
| 893 | osb->recovery_thread_task = NULL; |
| 894 | mb(); /* sync with ocfs2_recovery_thread_running */ |
| 895 | wake_up(&osb->recovery_event); |
| 896 | |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 897 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 898 | |
| 899 | mlog_exit(status); |
| 900 | /* no one is callint kthread_stop() for us so the kthread() api |
| 901 | * requires that we call do_exit(). And it isn't exported, but |
| 902 | * complete_and_exit() seems to be a minimal wrapper around it. */ |
| 903 | complete_and_exit(NULL, status); |
| 904 | return status; |
| 905 | } |
| 906 | |
| 907 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) |
| 908 | { |
| 909 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", |
| 910 | node_num, osb->node_num); |
| 911 | |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 912 | mutex_lock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 913 | if (osb->disable_recovery) |
| 914 | goto out; |
| 915 | |
| 916 | /* People waiting on recovery will wait on |
| 917 | * the recovery map to empty. */ |
| 918 | if (!ocfs2_recovery_map_set(osb, node_num)) |
| 919 | mlog(0, "node %d already be in recovery.\n", node_num); |
| 920 | |
| 921 | mlog(0, "starting recovery thread...\n"); |
| 922 | |
| 923 | if (osb->recovery_thread_task) |
| 924 | goto out; |
| 925 | |
| 926 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, |
Mark Fasheh | 7842704 | 2006-05-04 12:03:26 -0700 | [diff] [blame] | 927 | "ocfs2rec"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 928 | if (IS_ERR(osb->recovery_thread_task)) { |
| 929 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); |
| 930 | osb->recovery_thread_task = NULL; |
| 931 | } |
| 932 | |
| 933 | out: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 934 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 935 | wake_up(&osb->recovery_event); |
| 936 | |
| 937 | mlog_exit_void(); |
| 938 | } |
| 939 | |
| 940 | /* Does the actual journal replay and marks the journal inode as |
| 941 | * clean. Will only replay if the journal inode is marked dirty. */ |
| 942 | static int ocfs2_replay_journal(struct ocfs2_super *osb, |
| 943 | int node_num, |
| 944 | int slot_num) |
| 945 | { |
| 946 | int status; |
| 947 | int got_lock = 0; |
| 948 | unsigned int flags; |
| 949 | struct inode *inode = NULL; |
| 950 | struct ocfs2_dinode *fe; |
| 951 | journal_t *journal = NULL; |
| 952 | struct buffer_head *bh = NULL; |
| 953 | |
| 954 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 955 | slot_num); |
| 956 | if (inode == NULL) { |
| 957 | status = -EACCES; |
| 958 | mlog_errno(status); |
| 959 | goto done; |
| 960 | } |
| 961 | if (is_bad_inode(inode)) { |
| 962 | status = -EACCES; |
| 963 | iput(inode); |
| 964 | inode = NULL; |
| 965 | mlog_errno(status); |
| 966 | goto done; |
| 967 | } |
| 968 | SET_INODE_JOURNAL(inode); |
| 969 | |
Mark Fasheh | 4bcec18 | 2006-10-09 16:02:40 -0700 | [diff] [blame] | 970 | status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 971 | if (status < 0) { |
| 972 | mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); |
| 973 | if (status != -ERESTARTSYS) |
| 974 | mlog(ML_ERROR, "Could not lock journal!\n"); |
| 975 | goto done; |
| 976 | } |
| 977 | got_lock = 1; |
| 978 | |
| 979 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 980 | |
| 981 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 982 | |
| 983 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { |
| 984 | mlog(0, "No recovery required for node %d\n", node_num); |
| 985 | goto done; |
| 986 | } |
| 987 | |
| 988 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", |
| 989 | node_num, slot_num, |
| 990 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); |
| 991 | |
| 992 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
| 993 | |
| 994 | status = ocfs2_force_read_journal(inode); |
| 995 | if (status < 0) { |
| 996 | mlog_errno(status); |
| 997 | goto done; |
| 998 | } |
| 999 | |
| 1000 | mlog(0, "calling journal_init_inode\n"); |
| 1001 | journal = journal_init_inode(inode); |
| 1002 | if (journal == NULL) { |
| 1003 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 1004 | status = -EIO; |
| 1005 | goto done; |
| 1006 | } |
| 1007 | |
| 1008 | status = journal_load(journal); |
| 1009 | if (status < 0) { |
| 1010 | mlog_errno(status); |
| 1011 | if (!igrab(inode)) |
| 1012 | BUG(); |
| 1013 | journal_destroy(journal); |
| 1014 | goto done; |
| 1015 | } |
| 1016 | |
| 1017 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); |
| 1018 | |
| 1019 | /* wipe the journal */ |
| 1020 | mlog(0, "flushing the journal.\n"); |
| 1021 | journal_lock_updates(journal); |
| 1022 | status = journal_flush(journal); |
| 1023 | journal_unlock_updates(journal); |
| 1024 | if (status < 0) |
| 1025 | mlog_errno(status); |
| 1026 | |
| 1027 | /* This will mark the node clean */ |
| 1028 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 1029 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 1030 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 1031 | |
| 1032 | status = ocfs2_write_block(osb, bh, inode); |
| 1033 | if (status < 0) |
| 1034 | mlog_errno(status); |
| 1035 | |
| 1036 | if (!igrab(inode)) |
| 1037 | BUG(); |
| 1038 | |
| 1039 | journal_destroy(journal); |
| 1040 | |
| 1041 | done: |
| 1042 | /* drop the lock on this nodes journal */ |
| 1043 | if (got_lock) |
| 1044 | ocfs2_meta_unlock(inode, 1); |
| 1045 | |
| 1046 | if (inode) |
| 1047 | iput(inode); |
| 1048 | |
| 1049 | if (bh) |
| 1050 | brelse(bh); |
| 1051 | |
| 1052 | mlog_exit(status); |
| 1053 | return status; |
| 1054 | } |
| 1055 | |
| 1056 | /* |
| 1057 | * Do the most important parts of node recovery: |
| 1058 | * - Replay it's journal |
| 1059 | * - Stamp a clean local allocator file |
| 1060 | * - Stamp a clean truncate log |
| 1061 | * - Mark the node clean |
| 1062 | * |
| 1063 | * If this function completes without error, a node in OCFS2 can be |
| 1064 | * said to have been safely recovered. As a result, failure during the |
| 1065 | * second part of a nodes recovery process (local alloc recovery) is |
| 1066 | * far less concerning. |
| 1067 | */ |
| 1068 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
| 1069 | int node_num) |
| 1070 | { |
| 1071 | int status = 0; |
| 1072 | int slot_num; |
| 1073 | struct ocfs2_slot_info *si = osb->slot_info; |
| 1074 | struct ocfs2_dinode *la_copy = NULL; |
| 1075 | struct ocfs2_dinode *tl_copy = NULL; |
| 1076 | |
| 1077 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", |
| 1078 | node_num, osb->node_num); |
| 1079 | |
| 1080 | mlog(0, "checking node %d\n", node_num); |
| 1081 | |
| 1082 | /* Should not ever be called to recover ourselves -- in that |
| 1083 | * case we should've called ocfs2_journal_load instead. */ |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 1084 | BUG_ON(osb->node_num == node_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1085 | |
| 1086 | slot_num = ocfs2_node_num_to_slot(si, node_num); |
| 1087 | if (slot_num == OCFS2_INVALID_SLOT) { |
| 1088 | status = 0; |
| 1089 | mlog(0, "no slot for this node, so no recovery required.\n"); |
| 1090 | goto done; |
| 1091 | } |
| 1092 | |
| 1093 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); |
| 1094 | |
| 1095 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
| 1096 | if (status < 0) { |
| 1097 | mlog_errno(status); |
| 1098 | goto done; |
| 1099 | } |
| 1100 | |
| 1101 | /* Stamp a clean local alloc file AFTER recovering the journal... */ |
| 1102 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); |
| 1103 | if (status < 0) { |
| 1104 | mlog_errno(status); |
| 1105 | goto done; |
| 1106 | } |
| 1107 | |
| 1108 | /* An error from begin_truncate_log_recovery is not |
| 1109 | * serious enough to warrant halting the rest of |
| 1110 | * recovery. */ |
| 1111 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); |
| 1112 | if (status < 0) |
| 1113 | mlog_errno(status); |
| 1114 | |
| 1115 | /* Likewise, this would be a strange but ultimately not so |
| 1116 | * harmful place to get an error... */ |
| 1117 | ocfs2_clear_slot(si, slot_num); |
| 1118 | status = ocfs2_update_disk_slots(osb, si); |
| 1119 | if (status < 0) |
| 1120 | mlog_errno(status); |
| 1121 | |
| 1122 | /* This will kfree the memory pointed to by la_copy and tl_copy */ |
| 1123 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, |
| 1124 | tl_copy); |
| 1125 | |
| 1126 | status = 0; |
| 1127 | done: |
| 1128 | |
| 1129 | mlog_exit(status); |
| 1130 | return status; |
| 1131 | } |
| 1132 | |
| 1133 | /* Test node liveness by trylocking his journal. If we get the lock, |
| 1134 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is |
| 1135 | * still alive (we couldn't get the lock) and < 0 on error. */ |
| 1136 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 1137 | int slot_num) |
| 1138 | { |
| 1139 | int status, flags; |
| 1140 | struct inode *inode = NULL; |
| 1141 | |
| 1142 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 1143 | slot_num); |
| 1144 | if (inode == NULL) { |
| 1145 | mlog(ML_ERROR, "access error\n"); |
| 1146 | status = -EACCES; |
| 1147 | goto bail; |
| 1148 | } |
| 1149 | if (is_bad_inode(inode)) { |
| 1150 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 1151 | iput(inode); |
| 1152 | inode = NULL; |
| 1153 | status = -EACCES; |
| 1154 | goto bail; |
| 1155 | } |
| 1156 | SET_INODE_JOURNAL(inode); |
| 1157 | |
| 1158 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; |
Mark Fasheh | 4bcec18 | 2006-10-09 16:02:40 -0700 | [diff] [blame] | 1159 | status = ocfs2_meta_lock_full(inode, NULL, 1, flags); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1160 | if (status < 0) { |
| 1161 | if (status != -EAGAIN) |
| 1162 | mlog_errno(status); |
| 1163 | goto bail; |
| 1164 | } |
| 1165 | |
| 1166 | ocfs2_meta_unlock(inode, 1); |
| 1167 | bail: |
| 1168 | if (inode) |
| 1169 | iput(inode); |
| 1170 | |
| 1171 | return status; |
| 1172 | } |
| 1173 | |
| 1174 | /* Call this underneath ocfs2_super_lock. It also assumes that the |
| 1175 | * slot info struct has been updated from disk. */ |
| 1176 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) |
| 1177 | { |
| 1178 | int status, i, node_num; |
| 1179 | struct ocfs2_slot_info *si = osb->slot_info; |
| 1180 | |
| 1181 | /* This is called with the super block cluster lock, so we |
| 1182 | * know that the slot map can't change underneath us. */ |
| 1183 | |
| 1184 | spin_lock(&si->si_lock); |
| 1185 | for(i = 0; i < si->si_num_slots; i++) { |
| 1186 | if (i == osb->slot_num) |
| 1187 | continue; |
| 1188 | if (ocfs2_is_empty_slot(si, i)) |
| 1189 | continue; |
| 1190 | |
| 1191 | node_num = si->si_global_node_nums[i]; |
| 1192 | if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) |
| 1193 | continue; |
| 1194 | spin_unlock(&si->si_lock); |
| 1195 | |
| 1196 | /* Ok, we have a slot occupied by another node which |
| 1197 | * is not in the recovery map. We trylock his journal |
| 1198 | * file here to test if he's alive. */ |
| 1199 | status = ocfs2_trylock_journal(osb, i); |
| 1200 | if (!status) { |
| 1201 | /* Since we're called from mount, we know that |
| 1202 | * the recovery thread can't race us on |
| 1203 | * setting / checking the recovery bits. */ |
| 1204 | ocfs2_recovery_thread(osb, node_num); |
| 1205 | } else if ((status < 0) && (status != -EAGAIN)) { |
| 1206 | mlog_errno(status); |
| 1207 | goto bail; |
| 1208 | } |
| 1209 | |
| 1210 | spin_lock(&si->si_lock); |
| 1211 | } |
| 1212 | spin_unlock(&si->si_lock); |
| 1213 | |
| 1214 | status = 0; |
| 1215 | bail: |
| 1216 | mlog_exit(status); |
| 1217 | return status; |
| 1218 | } |
| 1219 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 1220 | struct ocfs2_orphan_filldir_priv { |
| 1221 | struct inode *head; |
| 1222 | struct ocfs2_super *osb; |
| 1223 | }; |
| 1224 | |
| 1225 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, |
| 1226 | loff_t pos, u64 ino, unsigned type) |
| 1227 | { |
| 1228 | struct ocfs2_orphan_filldir_priv *p = priv; |
| 1229 | struct inode *iter; |
| 1230 | |
| 1231 | if (name_len == 1 && !strncmp(".", name, 1)) |
| 1232 | return 0; |
| 1233 | if (name_len == 2 && !strncmp("..", name, 2)) |
| 1234 | return 0; |
| 1235 | |
| 1236 | /* Skip bad inodes so that recovery can continue */ |
| 1237 | iter = ocfs2_iget(p->osb, ino, |
| 1238 | OCFS2_FI_FLAG_ORPHAN_RECOVERY); |
| 1239 | if (IS_ERR(iter)) |
| 1240 | return 0; |
| 1241 | |
| 1242 | mlog(0, "queue orphan %llu\n", |
| 1243 | (unsigned long long)OCFS2_I(iter)->ip_blkno); |
| 1244 | /* No locking is required for the next_orphan queue as there |
| 1245 | * is only ever a single process doing orphan recovery. */ |
| 1246 | OCFS2_I(iter)->ip_next_orphan = p->head; |
| 1247 | p->head = iter; |
| 1248 | |
| 1249 | return 0; |
| 1250 | } |
| 1251 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1252 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
| 1253 | int slot, |
| 1254 | struct inode **head) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1255 | { |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1256 | int status; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1257 | struct inode *orphan_dir_inode = NULL; |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 1258 | struct ocfs2_orphan_filldir_priv priv; |
| 1259 | loff_t pos = 0; |
| 1260 | |
| 1261 | priv.osb = osb; |
| 1262 | priv.head = *head; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1263 | |
| 1264 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
| 1265 | ORPHAN_DIR_SYSTEM_INODE, |
| 1266 | slot); |
| 1267 | if (!orphan_dir_inode) { |
| 1268 | status = -ENOENT; |
| 1269 | mlog_errno(status); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1270 | return status; |
| 1271 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1272 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1273 | mutex_lock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | 4bcec18 | 2006-10-09 16:02:40 -0700 | [diff] [blame] | 1274 | status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1275 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1276 | mlog_errno(status); |
| 1277 | goto out; |
| 1278 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1279 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 1280 | status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, |
| 1281 | ocfs2_orphan_filldir); |
| 1282 | if (status) { |
| 1283 | mlog_errno(status); |
Mark Fasheh | a86370f | 2007-12-03 14:06:23 -0800 | [diff] [blame] | 1284 | goto out_cluster; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1285 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1286 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 1287 | *head = priv.head; |
| 1288 | |
Mark Fasheh | a86370f | 2007-12-03 14:06:23 -0800 | [diff] [blame] | 1289 | out_cluster: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1290 | ocfs2_meta_unlock(orphan_dir_inode, 0); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1291 | out: |
| 1292 | mutex_unlock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1293 | iput(orphan_dir_inode); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1294 | return status; |
| 1295 | } |
| 1296 | |
| 1297 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, |
| 1298 | int slot) |
| 1299 | { |
| 1300 | int ret; |
| 1301 | |
| 1302 | spin_lock(&osb->osb_lock); |
| 1303 | ret = !osb->osb_orphan_wipes[slot]; |
| 1304 | spin_unlock(&osb->osb_lock); |
| 1305 | return ret; |
| 1306 | } |
| 1307 | |
| 1308 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, |
| 1309 | int slot) |
| 1310 | { |
| 1311 | spin_lock(&osb->osb_lock); |
| 1312 | /* Mark ourselves such that new processes in delete_inode() |
| 1313 | * know to quit early. */ |
| 1314 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 1315 | while (osb->osb_orphan_wipes[slot]) { |
| 1316 | /* If any processes are already in the middle of an |
| 1317 | * orphan wipe on this dir, then we need to wait for |
| 1318 | * them. */ |
| 1319 | spin_unlock(&osb->osb_lock); |
| 1320 | wait_event_interruptible(osb->osb_wipe_event, |
| 1321 | ocfs2_orphan_recovery_can_continue(osb, slot)); |
| 1322 | spin_lock(&osb->osb_lock); |
| 1323 | } |
| 1324 | spin_unlock(&osb->osb_lock); |
| 1325 | } |
| 1326 | |
| 1327 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, |
| 1328 | int slot) |
| 1329 | { |
| 1330 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 1331 | } |
| 1332 | |
| 1333 | /* |
| 1334 | * Orphan recovery. Each mounted node has it's own orphan dir which we |
| 1335 | * must run during recovery. Our strategy here is to build a list of |
| 1336 | * the inodes in the orphan dir and iget/iput them. The VFS does |
| 1337 | * (most) of the rest of the work. |
| 1338 | * |
| 1339 | * Orphan recovery can happen at any time, not just mount so we have a |
| 1340 | * couple of extra considerations. |
| 1341 | * |
| 1342 | * - We grab as many inodes as we can under the orphan dir lock - |
| 1343 | * doing iget() outside the orphan dir risks getting a reference on |
| 1344 | * an invalid inode. |
| 1345 | * - We must be sure not to deadlock with other processes on the |
| 1346 | * system wanting to run delete_inode(). This can happen when they go |
| 1347 | * to lock the orphan dir and the orphan recovery process attempts to |
| 1348 | * iget() inside the orphan dir lock. This can be avoided by |
| 1349 | * advertising our state to ocfs2_delete_inode(). |
| 1350 | */ |
| 1351 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
| 1352 | int slot) |
| 1353 | { |
| 1354 | int ret = 0; |
| 1355 | struct inode *inode = NULL; |
| 1356 | struct inode *iter; |
| 1357 | struct ocfs2_inode_info *oi; |
| 1358 | |
| 1359 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); |
| 1360 | |
| 1361 | ocfs2_mark_recovering_orphan_dir(osb, slot); |
| 1362 | ret = ocfs2_queue_orphans(osb, slot, &inode); |
| 1363 | ocfs2_clear_recovering_orphan_dir(osb, slot); |
| 1364 | |
| 1365 | /* Error here should be noted, but we want to continue with as |
| 1366 | * many queued inodes as we've got. */ |
| 1367 | if (ret) |
| 1368 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1369 | |
| 1370 | while (inode) { |
| 1371 | oi = OCFS2_I(inode); |
Mark Fasheh | b0697053 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1372 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1373 | |
| 1374 | iter = oi->ip_next_orphan; |
| 1375 | |
| 1376 | spin_lock(&oi->ip_lock); |
| 1377 | /* Delete voting may have set these on the assumption |
| 1378 | * that the other node would wipe them successfully. |
| 1379 | * If they are still in the node's orphan dir, we need |
| 1380 | * to reset that state. */ |
| 1381 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); |
| 1382 | |
| 1383 | /* Set the proper information to get us going into |
| 1384 | * ocfs2_delete_inode. */ |
| 1385 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1386 | spin_unlock(&oi->ip_lock); |
| 1387 | |
| 1388 | iput(inode); |
| 1389 | |
| 1390 | inode = iter; |
| 1391 | } |
| 1392 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1393 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1394 | } |
| 1395 | |
| 1396 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
| 1397 | { |
| 1398 | /* This check is good because ocfs2 will wait on our recovery |
| 1399 | * thread before changing it to something other than MOUNTED |
| 1400 | * or DISABLED. */ |
| 1401 | wait_event(osb->osb_mount_event, |
| 1402 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED || |
| 1403 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
| 1404 | |
| 1405 | /* If there's an error on mount, then we may never get to the |
| 1406 | * MOUNTED flag, but this is set right before |
| 1407 | * dismount_volume() so we can trust it. */ |
| 1408 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { |
| 1409 | mlog(0, "mount error, exiting!\n"); |
| 1410 | return -EBUSY; |
| 1411 | } |
| 1412 | |
| 1413 | return 0; |
| 1414 | } |
| 1415 | |
| 1416 | static int ocfs2_commit_thread(void *arg) |
| 1417 | { |
| 1418 | int status; |
| 1419 | struct ocfs2_super *osb = arg; |
| 1420 | struct ocfs2_journal *journal = osb->journal; |
| 1421 | |
| 1422 | /* we can trust j_num_trans here because _should_stop() is only set in |
| 1423 | * shutdown and nobody other than ourselves should be able to start |
| 1424 | * transactions. committing on shutdown might take a few iterations |
| 1425 | * as final transactions put deleted inodes on the list */ |
| 1426 | while (!(kthread_should_stop() && |
| 1427 | atomic_read(&journal->j_num_trans) == 0)) { |
| 1428 | |
Mark Fasheh | 745ae8ba | 2006-02-09 13:23:39 -0800 | [diff] [blame] | 1429 | wait_event_interruptible(osb->checkpoint_event, |
| 1430 | atomic_read(&journal->j_num_trans) |
| 1431 | || kthread_should_stop()); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1432 | |
| 1433 | status = ocfs2_commit_cache(osb); |
| 1434 | if (status < 0) |
| 1435 | mlog_errno(status); |
| 1436 | |
| 1437 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ |
| 1438 | mlog(ML_KTHREAD, |
| 1439 | "commit_thread: %u transactions pending on " |
| 1440 | "shutdown\n", |
| 1441 | atomic_read(&journal->j_num_trans)); |
| 1442 | } |
| 1443 | } |
| 1444 | |
| 1445 | return 0; |
| 1446 | } |
| 1447 | |
| 1448 | /* Look for a dirty journal without taking any cluster locks. Used for |
| 1449 | * hard readonly access to determine whether the file system journals |
| 1450 | * require recovery. */ |
| 1451 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
| 1452 | { |
| 1453 | int ret = 0; |
| 1454 | unsigned int slot; |
| 1455 | struct buffer_head *di_bh; |
| 1456 | struct ocfs2_dinode *di; |
| 1457 | struct inode *journal = NULL; |
| 1458 | |
| 1459 | for(slot = 0; slot < osb->max_slots; slot++) { |
| 1460 | journal = ocfs2_get_system_file_inode(osb, |
| 1461 | JOURNAL_SYSTEM_INODE, |
| 1462 | slot); |
| 1463 | if (!journal || is_bad_inode(journal)) { |
| 1464 | ret = -EACCES; |
| 1465 | mlog_errno(ret); |
| 1466 | goto out; |
| 1467 | } |
| 1468 | |
| 1469 | di_bh = NULL; |
| 1470 | ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, |
| 1471 | 0, journal); |
| 1472 | if (ret < 0) { |
| 1473 | mlog_errno(ret); |
| 1474 | goto out; |
| 1475 | } |
| 1476 | |
| 1477 | di = (struct ocfs2_dinode *) di_bh->b_data; |
| 1478 | |
| 1479 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 1480 | OCFS2_JOURNAL_DIRTY_FL) |
| 1481 | ret = -EROFS; |
| 1482 | |
| 1483 | brelse(di_bh); |
| 1484 | if (ret) |
| 1485 | break; |
| 1486 | } |
| 1487 | |
| 1488 | out: |
| 1489 | if (journal) |
| 1490 | iput(journal); |
| 1491 | |
| 1492 | return ret; |
| 1493 | } |