1ccd979bdSMark Fasheh /* -*- mode: c; c-basic-offset: 8; -*- 2ccd979bdSMark Fasheh * vim: noexpandtab sw=8 ts=8 sts=0: 3ccd979bdSMark Fasheh * 4ccd979bdSMark Fasheh * journal.c 5ccd979bdSMark Fasheh * 6ccd979bdSMark Fasheh * Defines functions of journalling api 7ccd979bdSMark Fasheh * 8ccd979bdSMark Fasheh * Copyright (C) 2003, 2004 Oracle. All rights reserved. 9ccd979bdSMark Fasheh * 10ccd979bdSMark Fasheh * This program is free software; you can redistribute it and/or 11ccd979bdSMark Fasheh * modify it under the terms of the GNU General Public 12ccd979bdSMark Fasheh * License as published by the Free Software Foundation; either 13ccd979bdSMark Fasheh * version 2 of the License, or (at your option) any later version. 14ccd979bdSMark Fasheh * 15ccd979bdSMark Fasheh * This program is distributed in the hope that it will be useful, 16ccd979bdSMark Fasheh * but WITHOUT ANY WARRANTY; without even the implied warranty of 17ccd979bdSMark Fasheh * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18ccd979bdSMark Fasheh * General Public License for more details. 19ccd979bdSMark Fasheh * 20ccd979bdSMark Fasheh * You should have received a copy of the GNU General Public 21ccd979bdSMark Fasheh * License along with this program; if not, write to the 22ccd979bdSMark Fasheh * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 23ccd979bdSMark Fasheh * Boston, MA 021110-1307, USA. 24ccd979bdSMark Fasheh */ 25ccd979bdSMark Fasheh 26ccd979bdSMark Fasheh #include <linux/fs.h> 27ccd979bdSMark Fasheh #include <linux/types.h> 28ccd979bdSMark Fasheh #include <linux/slab.h> 29ccd979bdSMark Fasheh #include <linux/highmem.h> 30ccd979bdSMark Fasheh #include <linux/kthread.h> 31ccd979bdSMark Fasheh 32ccd979bdSMark Fasheh #define MLOG_MASK_PREFIX ML_JOURNAL 33ccd979bdSMark Fasheh #include <cluster/masklog.h> 34ccd979bdSMark Fasheh 35ccd979bdSMark Fasheh #include "ocfs2.h" 36ccd979bdSMark Fasheh 37ccd979bdSMark Fasheh #include "alloc.h" 38316f4b9fSMark Fasheh #include "dir.h" 39ccd979bdSMark Fasheh #include "dlmglue.h" 40ccd979bdSMark Fasheh #include "extent_map.h" 41ccd979bdSMark Fasheh #include "heartbeat.h" 42ccd979bdSMark Fasheh #include "inode.h" 43ccd979bdSMark Fasheh #include "journal.h" 44ccd979bdSMark Fasheh #include "localalloc.h" 45ccd979bdSMark Fasheh #include "slot_map.h" 46ccd979bdSMark Fasheh #include "super.h" 47ccd979bdSMark Fasheh #include "sysfile.h" 48ccd979bdSMark Fasheh 49ccd979bdSMark Fasheh #include "buffer_head_io.h" 50ccd979bdSMark Fasheh 5134af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 52ccd979bdSMark Fasheh 53ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 54ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 55ccd979bdSMark Fasheh int node_num); 56ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 57ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 58ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb); 59ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 60ccd979bdSMark Fasheh int dirty); 61ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 62ccd979bdSMark Fasheh int slot_num); 63ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 64ccd979bdSMark Fasheh int slot); 65ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 66ccd979bdSMark Fasheh 67ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 68ccd979bdSMark Fasheh { 69ccd979bdSMark Fasheh int status = 0; 70ccd979bdSMark Fasheh unsigned int flushed; 71ccd979bdSMark Fasheh unsigned long old_id; 72ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 73ccd979bdSMark Fasheh 74ccd979bdSMark Fasheh mlog_entry_void(); 75ccd979bdSMark Fasheh 76ccd979bdSMark Fasheh journal = osb->journal; 77ccd979bdSMark Fasheh 78ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 79ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 80ccd979bdSMark Fasheh 81ccd979bdSMark Fasheh if (atomic_read(&journal->j_num_trans) == 0) { 82ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 83ccd979bdSMark Fasheh mlog(0, "No transactions for me to flush!\n"); 84ccd979bdSMark Fasheh goto finally; 85ccd979bdSMark Fasheh } 86ccd979bdSMark Fasheh 87ccd979bdSMark Fasheh journal_lock_updates(journal->j_journal); 88ccd979bdSMark Fasheh status = journal_flush(journal->j_journal); 89ccd979bdSMark Fasheh journal_unlock_updates(journal->j_journal); 90ccd979bdSMark Fasheh if (status < 0) { 91ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 92ccd979bdSMark Fasheh mlog_errno(status); 93ccd979bdSMark Fasheh goto finally; 94ccd979bdSMark Fasheh } 95ccd979bdSMark Fasheh 96ccd979bdSMark Fasheh old_id = ocfs2_inc_trans_id(journal); 97ccd979bdSMark Fasheh 98ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 99ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 100ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 101ccd979bdSMark Fasheh 102ccd979bdSMark Fasheh mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 103ccd979bdSMark Fasheh journal->j_trans_id, flushed); 104ccd979bdSMark Fasheh 10534d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 106ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 107ccd979bdSMark Fasheh finally: 108ccd979bdSMark Fasheh mlog_exit(status); 109ccd979bdSMark Fasheh return status; 110ccd979bdSMark Fasheh } 111ccd979bdSMark Fasheh 112ccd979bdSMark Fasheh /* pass it NULL and it will allocate a new handle object for you. If 113ccd979bdSMark Fasheh * you pass it a handle however, it may still return error, in which 114ccd979bdSMark Fasheh * case it has free'd the passed handle for you. */ 1151fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 116ccd979bdSMark Fasheh { 117ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 1181fabe148SMark Fasheh handle_t *handle; 119ccd979bdSMark Fasheh 120ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 121ccd979bdSMark Fasheh 12265eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 12365eff9ccSMark Fasheh return ERR_PTR(-EROFS); 124ccd979bdSMark Fasheh 125ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 126ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 127ccd979bdSMark Fasheh 128ccd979bdSMark Fasheh /* JBD might support this, but our journalling code doesn't yet. */ 129ccd979bdSMark Fasheh if (journal_current_handle()) { 130ccd979bdSMark Fasheh mlog(ML_ERROR, "Recursive transaction attempted!\n"); 131ccd979bdSMark Fasheh BUG(); 132ccd979bdSMark Fasheh } 133ccd979bdSMark Fasheh 134ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 135ccd979bdSMark Fasheh 1361fabe148SMark Fasheh handle = journal_start(journal, max_buffs); 1371fabe148SMark Fasheh if (IS_ERR(handle)) { 138ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 139ccd979bdSMark Fasheh 1401fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 141ccd979bdSMark Fasheh 142ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 143ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 1441fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 145ccd979bdSMark Fasheh } 146c271c5c2SSunil Mushran } else { 147c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 148ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 149c271c5c2SSunil Mushran } 150ccd979bdSMark Fasheh 151ccd979bdSMark Fasheh return handle; 152ccd979bdSMark Fasheh } 153ccd979bdSMark Fasheh 1541fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 1551fabe148SMark Fasheh handle_t *handle) 156ccd979bdSMark Fasheh { 1571fabe148SMark Fasheh int ret; 15802dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 159ccd979bdSMark Fasheh 160ccd979bdSMark Fasheh BUG_ON(!handle); 161ccd979bdSMark Fasheh 1621fabe148SMark Fasheh ret = journal_stop(handle); 1631fabe148SMark Fasheh if (ret < 0) 1641fabe148SMark Fasheh mlog_errno(ret); 165ccd979bdSMark Fasheh 166ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 167ccd979bdSMark Fasheh 1681fabe148SMark Fasheh return ret; 169ccd979bdSMark Fasheh } 170ccd979bdSMark Fasheh 171ccd979bdSMark Fasheh /* 172ccd979bdSMark Fasheh * 'nblocks' is what you want to add to the current 173ccd979bdSMark Fasheh * transaction. extend_trans will either extend the current handle by 174ccd979bdSMark Fasheh * nblocks, or commit it and start a new one with nblocks credits. 175ccd979bdSMark Fasheh * 176e8aed345SMark Fasheh * This might call journal_restart() which will commit dirty buffers 177e8aed345SMark Fasheh * and then restart the transaction. Before calling 178e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 179e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 180e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 181e8aed345SMark Fasheh * 182ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 183ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 184ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 185ccd979bdSMark Fasheh * 186ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 187ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 188ccd979bdSMark Fasheh * cluster locks associated with this handle. 189ccd979bdSMark Fasheh */ 1901fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 191ccd979bdSMark Fasheh { 192ccd979bdSMark Fasheh int status; 193ccd979bdSMark Fasheh 194ccd979bdSMark Fasheh BUG_ON(!handle); 195ccd979bdSMark Fasheh BUG_ON(!nblocks); 196ccd979bdSMark Fasheh 197ccd979bdSMark Fasheh mlog_entry_void(); 198ccd979bdSMark Fasheh 199ccd979bdSMark Fasheh mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 200ccd979bdSMark Fasheh 2010879c584SMark Fasheh #ifdef OCFS2_DEBUG_FS 2020879c584SMark Fasheh status = 1; 2030879c584SMark Fasheh #else 2041fc58146SMark Fasheh status = journal_extend(handle, nblocks); 205ccd979bdSMark Fasheh if (status < 0) { 206ccd979bdSMark Fasheh mlog_errno(status); 207ccd979bdSMark Fasheh goto bail; 208ccd979bdSMark Fasheh } 2090879c584SMark Fasheh #endif 210ccd979bdSMark Fasheh 211ccd979bdSMark Fasheh if (status > 0) { 212ccd979bdSMark Fasheh mlog(0, "journal_extend failed, trying journal_restart\n"); 2131fc58146SMark Fasheh status = journal_restart(handle, nblocks); 214ccd979bdSMark Fasheh if (status < 0) { 215ccd979bdSMark Fasheh mlog_errno(status); 216ccd979bdSMark Fasheh goto bail; 217ccd979bdSMark Fasheh } 21801ddf1e1SMark Fasheh } 219ccd979bdSMark Fasheh 220ccd979bdSMark Fasheh status = 0; 221ccd979bdSMark Fasheh bail: 222ccd979bdSMark Fasheh 223ccd979bdSMark Fasheh mlog_exit(status); 224ccd979bdSMark Fasheh return status; 225ccd979bdSMark Fasheh } 226ccd979bdSMark Fasheh 2271fabe148SMark Fasheh int ocfs2_journal_access(handle_t *handle, 228ccd979bdSMark Fasheh struct inode *inode, 229ccd979bdSMark Fasheh struct buffer_head *bh, 230ccd979bdSMark Fasheh int type) 231ccd979bdSMark Fasheh { 232ccd979bdSMark Fasheh int status; 233ccd979bdSMark Fasheh 234ccd979bdSMark Fasheh BUG_ON(!inode); 235ccd979bdSMark Fasheh BUG_ON(!handle); 236ccd979bdSMark Fasheh BUG_ON(!bh); 237ccd979bdSMark Fasheh 238205f87f6SBadari Pulavarty mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 239ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, type, 240ccd979bdSMark Fasheh (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 241ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_CREATE" : 242ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_WRITE", 243ccd979bdSMark Fasheh bh->b_size); 244ccd979bdSMark Fasheh 245ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 246ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 247ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 248ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 249ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 250ccd979bdSMark Fasheh BUG(); 251ccd979bdSMark Fasheh } 252ccd979bdSMark Fasheh 253ccd979bdSMark Fasheh /* Set the current transaction information on the inode so 254ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 255ccd979bdSMark Fasheh * on this inode or not. We're protected from the commit 256ccd979bdSMark Fasheh * thread updating the current transaction id until 257ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 258ccd979bdSMark Fasheh * j_trans_barrier for us. */ 259ccd979bdSMark Fasheh ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 260ccd979bdSMark Fasheh 261251b6eccSMark Fasheh mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 262ccd979bdSMark Fasheh switch (type) { 263ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 264ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 2651fabe148SMark Fasheh status = journal_get_write_access(handle, bh); 266ccd979bdSMark Fasheh break; 267ccd979bdSMark Fasheh 268ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 2691fabe148SMark Fasheh status = journal_get_undo_access(handle, bh); 270ccd979bdSMark Fasheh break; 271ccd979bdSMark Fasheh 272ccd979bdSMark Fasheh default: 273ccd979bdSMark Fasheh status = -EINVAL; 274ccd979bdSMark Fasheh mlog(ML_ERROR, "Uknown access type!\n"); 275ccd979bdSMark Fasheh } 276251b6eccSMark Fasheh mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 277ccd979bdSMark Fasheh 278ccd979bdSMark Fasheh if (status < 0) 279ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 280ccd979bdSMark Fasheh status, type); 281ccd979bdSMark Fasheh 282ccd979bdSMark Fasheh mlog_exit(status); 283ccd979bdSMark Fasheh return status; 284ccd979bdSMark Fasheh } 285ccd979bdSMark Fasheh 2861fabe148SMark Fasheh int ocfs2_journal_dirty(handle_t *handle, 287ccd979bdSMark Fasheh struct buffer_head *bh) 288ccd979bdSMark Fasheh { 289ccd979bdSMark Fasheh int status; 290ccd979bdSMark Fasheh 291ccd979bdSMark Fasheh mlog_entry("(bh->b_blocknr=%llu)\n", 292ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 293ccd979bdSMark Fasheh 2941fabe148SMark Fasheh status = journal_dirty_metadata(handle, bh); 295ccd979bdSMark Fasheh if (status < 0) 296ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not dirty metadata buffer. " 297ccd979bdSMark Fasheh "(bh->b_blocknr=%llu)\n", 298ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 299ccd979bdSMark Fasheh 300ccd979bdSMark Fasheh mlog_exit(status); 301ccd979bdSMark Fasheh return status; 302ccd979bdSMark Fasheh } 303ccd979bdSMark Fasheh 304ccd979bdSMark Fasheh int ocfs2_journal_dirty_data(handle_t *handle, 305ccd979bdSMark Fasheh struct buffer_head *bh) 306ccd979bdSMark Fasheh { 307ccd979bdSMark Fasheh int err = journal_dirty_data(handle, bh); 308ccd979bdSMark Fasheh if (err) 309ccd979bdSMark Fasheh mlog_errno(err); 310ccd979bdSMark Fasheh /* TODO: When we can handle it, abort the handle and go RO on 311ccd979bdSMark Fasheh * error here. */ 312ccd979bdSMark Fasheh 313ccd979bdSMark Fasheh return err; 314ccd979bdSMark Fasheh } 315ccd979bdSMark Fasheh 316d147b3d6SMark Fasheh #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD_DEFAULT_MAX_COMMIT_AGE) 317ccd979bdSMark Fasheh 318ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 319ccd979bdSMark Fasheh { 320ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 321d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 322d147b3d6SMark Fasheh 323d147b3d6SMark Fasheh if (osb->osb_commit_interval) 324d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 325ccd979bdSMark Fasheh 326ccd979bdSMark Fasheh spin_lock(&journal->j_state_lock); 327d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 328ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 329ccd979bdSMark Fasheh journal->j_flags |= JFS_BARRIER; 330ccd979bdSMark Fasheh else 331ccd979bdSMark Fasheh journal->j_flags &= ~JFS_BARRIER; 332ccd979bdSMark Fasheh spin_unlock(&journal->j_state_lock); 333ccd979bdSMark Fasheh } 334ccd979bdSMark Fasheh 335ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 336ccd979bdSMark Fasheh { 337ccd979bdSMark Fasheh int status = -1; 338ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 339ccd979bdSMark Fasheh journal_t *j_journal = NULL; 340ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 341ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 342ccd979bdSMark Fasheh struct ocfs2_super *osb; 343e63aecb6SMark Fasheh int inode_lock = 0; 344ccd979bdSMark Fasheh 345ccd979bdSMark Fasheh mlog_entry_void(); 346ccd979bdSMark Fasheh 347ccd979bdSMark Fasheh BUG_ON(!journal); 348ccd979bdSMark Fasheh 349ccd979bdSMark Fasheh osb = journal->j_osb; 350ccd979bdSMark Fasheh 351ccd979bdSMark Fasheh /* already have the inode for our journal */ 352ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 353ccd979bdSMark Fasheh osb->slot_num); 354ccd979bdSMark Fasheh if (inode == NULL) { 355ccd979bdSMark Fasheh status = -EACCES; 356ccd979bdSMark Fasheh mlog_errno(status); 357ccd979bdSMark Fasheh goto done; 358ccd979bdSMark Fasheh } 359ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 360ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 361ccd979bdSMark Fasheh iput(inode); 362ccd979bdSMark Fasheh inode = NULL; 363ccd979bdSMark Fasheh status = -EACCES; 364ccd979bdSMark Fasheh goto done; 365ccd979bdSMark Fasheh } 366ccd979bdSMark Fasheh 367ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 368ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 369ccd979bdSMark Fasheh 3706eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 3716eff5790SMark Fasheh * changes in a live cluster so it can be considered an 3726eff5790SMark Fasheh * exception to the rule. */ 373e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 374ccd979bdSMark Fasheh if (status < 0) { 375ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 376ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 377ccd979bdSMark Fasheh goto done; 378ccd979bdSMark Fasheh } 379ccd979bdSMark Fasheh 380e63aecb6SMark Fasheh inode_lock = 1; 381ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 382ccd979bdSMark Fasheh 383ccd979bdSMark Fasheh if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 384ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 385ccd979bdSMark Fasheh inode->i_size); 386ccd979bdSMark Fasheh status = -EINVAL; 387ccd979bdSMark Fasheh goto done; 388ccd979bdSMark Fasheh } 389ccd979bdSMark Fasheh 390ccd979bdSMark Fasheh mlog(0, "inode->i_size = %lld\n", inode->i_size); 3915515eff8SAndrew Morton mlog(0, "inode->i_blocks = %llu\n", 3925515eff8SAndrew Morton (unsigned long long)inode->i_blocks); 393ccd979bdSMark Fasheh mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 394ccd979bdSMark Fasheh 395ccd979bdSMark Fasheh /* call the kernels journal init function now */ 396ccd979bdSMark Fasheh j_journal = journal_init_inode(inode); 397ccd979bdSMark Fasheh if (j_journal == NULL) { 398ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 399ccd979bdSMark Fasheh status = -EINVAL; 400ccd979bdSMark Fasheh goto done; 401ccd979bdSMark Fasheh } 402ccd979bdSMark Fasheh 403ccd979bdSMark Fasheh mlog(0, "Returned from journal_init_inode\n"); 404ccd979bdSMark Fasheh mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 405ccd979bdSMark Fasheh 406ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 407ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 408ccd979bdSMark Fasheh 409ccd979bdSMark Fasheh journal->j_journal = j_journal; 410ccd979bdSMark Fasheh journal->j_inode = inode; 411ccd979bdSMark Fasheh journal->j_bh = bh; 412ccd979bdSMark Fasheh 413ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 414ccd979bdSMark Fasheh 415ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 416ccd979bdSMark Fasheh 417ccd979bdSMark Fasheh status = 0; 418ccd979bdSMark Fasheh done: 419ccd979bdSMark Fasheh if (status < 0) { 420e63aecb6SMark Fasheh if (inode_lock) 421e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 422ccd979bdSMark Fasheh if (bh != NULL) 423ccd979bdSMark Fasheh brelse(bh); 424ccd979bdSMark Fasheh if (inode) { 425ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 426ccd979bdSMark Fasheh iput(inode); 427ccd979bdSMark Fasheh } 428ccd979bdSMark Fasheh } 429ccd979bdSMark Fasheh 430ccd979bdSMark Fasheh mlog_exit(status); 431ccd979bdSMark Fasheh return status; 432ccd979bdSMark Fasheh } 433ccd979bdSMark Fasheh 434ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 435ccd979bdSMark Fasheh int dirty) 436ccd979bdSMark Fasheh { 437ccd979bdSMark Fasheh int status; 438ccd979bdSMark Fasheh unsigned int flags; 439ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 440ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 441ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 442ccd979bdSMark Fasheh 443ccd979bdSMark Fasheh mlog_entry_void(); 444ccd979bdSMark Fasheh 445ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 446ccd979bdSMark Fasheh if (!OCFS2_IS_VALID_DINODE(fe)) { 447ccd979bdSMark Fasheh /* This is called from startup/shutdown which will 448ccd979bdSMark Fasheh * handle the errors in a specific manner, so no need 449ccd979bdSMark Fasheh * to call ocfs2_error() here. */ 450b0697053SMark Fasheh mlog(ML_ERROR, "Journal dinode %llu has invalid " 4511ca1a111SMark Fasheh "signature: %.*s", 4521ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(fe->i_blkno), 7, 453b0697053SMark Fasheh fe->i_signature); 454ccd979bdSMark Fasheh status = -EIO; 455ccd979bdSMark Fasheh goto out; 456ccd979bdSMark Fasheh } 457ccd979bdSMark Fasheh 458ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 459ccd979bdSMark Fasheh if (dirty) 460ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 461ccd979bdSMark Fasheh else 462ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 463ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 464ccd979bdSMark Fasheh 465ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, journal->j_inode); 466ccd979bdSMark Fasheh if (status < 0) 467ccd979bdSMark Fasheh mlog_errno(status); 468ccd979bdSMark Fasheh 469ccd979bdSMark Fasheh out: 470ccd979bdSMark Fasheh mlog_exit(status); 471ccd979bdSMark Fasheh return status; 472ccd979bdSMark Fasheh } 473ccd979bdSMark Fasheh 474ccd979bdSMark Fasheh /* 475ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 476ccd979bdSMark Fasheh * call. 477ccd979bdSMark Fasheh */ 478ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 479ccd979bdSMark Fasheh { 480ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 481ccd979bdSMark Fasheh int status = 0; 482ccd979bdSMark Fasheh struct inode *inode = NULL; 483ccd979bdSMark Fasheh int num_running_trans = 0; 484ccd979bdSMark Fasheh 485ccd979bdSMark Fasheh mlog_entry_void(); 486ccd979bdSMark Fasheh 487ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 488ccd979bdSMark Fasheh 489ccd979bdSMark Fasheh journal = osb->journal; 490ccd979bdSMark Fasheh if (!journal) 491ccd979bdSMark Fasheh goto done; 492ccd979bdSMark Fasheh 493ccd979bdSMark Fasheh inode = journal->j_inode; 494ccd979bdSMark Fasheh 495ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 496ccd979bdSMark Fasheh goto done; 497ccd979bdSMark Fasheh 498ccd979bdSMark Fasheh /* need to inc inode use count as journal_destroy will iput. */ 499ccd979bdSMark Fasheh if (!igrab(inode)) 500ccd979bdSMark Fasheh BUG(); 501ccd979bdSMark Fasheh 502ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 503ccd979bdSMark Fasheh if (num_running_trans > 0) 504ccd979bdSMark Fasheh mlog(0, "Shutting down journal: must wait on %d " 505ccd979bdSMark Fasheh "running transactions!\n", 506ccd979bdSMark Fasheh num_running_trans); 507ccd979bdSMark Fasheh 508ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 509ccd979bdSMark Fasheh * release any locks that are still held. 510ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 511ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 512ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 513ccd979bdSMark Fasheh 514ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 515ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 516ccd979bdSMark Fasheh * completely destroy the journal. */ 517ccd979bdSMark Fasheh if (osb->commit_task) { 518ccd979bdSMark Fasheh /* Wait for the commit thread */ 519ccd979bdSMark Fasheh mlog(0, "Waiting for ocfs2commit to exit....\n"); 520ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 521ccd979bdSMark Fasheh osb->commit_task = NULL; 522ccd979bdSMark Fasheh } 523ccd979bdSMark Fasheh 524ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 525ccd979bdSMark Fasheh 526c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 527c271c5c2SSunil Mushran journal_lock_updates(journal->j_journal); 528c271c5c2SSunil Mushran status = journal_flush(journal->j_journal); 529c271c5c2SSunil Mushran journal_unlock_updates(journal->j_journal); 530c271c5c2SSunil Mushran if (status < 0) 531c271c5c2SSunil Mushran mlog_errno(status); 532c271c5c2SSunil Mushran } 533c271c5c2SSunil Mushran 534c271c5c2SSunil Mushran if (status == 0) { 535c271c5c2SSunil Mushran /* 536c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 537c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 538c271c5c2SSunil Mushran */ 539ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 0); 540ccd979bdSMark Fasheh if (status < 0) 541ccd979bdSMark Fasheh mlog_errno(status); 542c271c5c2SSunil Mushran } 543ccd979bdSMark Fasheh 544ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 545ccd979bdSMark Fasheh journal_destroy(journal->j_journal); 546ccd979bdSMark Fasheh 547ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 548ccd979bdSMark Fasheh 549ccd979bdSMark Fasheh /* unlock our journal */ 550e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 551ccd979bdSMark Fasheh 552ccd979bdSMark Fasheh brelse(journal->j_bh); 553ccd979bdSMark Fasheh journal->j_bh = NULL; 554ccd979bdSMark Fasheh 555ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 556ccd979bdSMark Fasheh 557ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 558ccd979bdSMark Fasheh done: 559ccd979bdSMark Fasheh if (inode) 560ccd979bdSMark Fasheh iput(inode); 561ccd979bdSMark Fasheh mlog_exit_void(); 562ccd979bdSMark Fasheh } 563ccd979bdSMark Fasheh 564ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 565ccd979bdSMark Fasheh journal_t *journal, 566ccd979bdSMark Fasheh int slot) 567ccd979bdSMark Fasheh { 568ccd979bdSMark Fasheh int olderr; 569ccd979bdSMark Fasheh 570ccd979bdSMark Fasheh olderr = journal_errno(journal); 571ccd979bdSMark Fasheh if (olderr) { 572ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 573ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 574ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 575ccd979bdSMark Fasheh sb->s_id); 576ccd979bdSMark Fasheh 577ccd979bdSMark Fasheh journal_ack_err(journal); 578ccd979bdSMark Fasheh journal_clear_err(journal); 579ccd979bdSMark Fasheh } 580ccd979bdSMark Fasheh } 581ccd979bdSMark Fasheh 582c271c5c2SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local) 583ccd979bdSMark Fasheh { 584ccd979bdSMark Fasheh int status = 0; 585ccd979bdSMark Fasheh struct ocfs2_super *osb; 586ccd979bdSMark Fasheh 587ccd979bdSMark Fasheh mlog_entry_void(); 588ccd979bdSMark Fasheh 589ccd979bdSMark Fasheh if (!journal) 590ccd979bdSMark Fasheh BUG(); 591ccd979bdSMark Fasheh 592ccd979bdSMark Fasheh osb = journal->j_osb; 593ccd979bdSMark Fasheh 594ccd979bdSMark Fasheh status = journal_load(journal->j_journal); 595ccd979bdSMark Fasheh if (status < 0) { 596ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 597ccd979bdSMark Fasheh goto done; 598ccd979bdSMark Fasheh } 599ccd979bdSMark Fasheh 600ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 601ccd979bdSMark Fasheh 602ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 1); 603ccd979bdSMark Fasheh if (status < 0) { 604ccd979bdSMark Fasheh mlog_errno(status); 605ccd979bdSMark Fasheh goto done; 606ccd979bdSMark Fasheh } 607ccd979bdSMark Fasheh 608ccd979bdSMark Fasheh /* Launch the commit thread */ 609c271c5c2SSunil Mushran if (!local) { 610c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 611c271c5c2SSunil Mushran "ocfs2cmt"); 612ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 613ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 614ccd979bdSMark Fasheh osb->commit_task = NULL; 615c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 616c271c5c2SSunil Mushran "error=%d", status); 617ccd979bdSMark Fasheh goto done; 618ccd979bdSMark Fasheh } 619c271c5c2SSunil Mushran } else 620c271c5c2SSunil Mushran osb->commit_task = NULL; 621ccd979bdSMark Fasheh 622ccd979bdSMark Fasheh done: 623ccd979bdSMark Fasheh mlog_exit(status); 624ccd979bdSMark Fasheh return status; 625ccd979bdSMark Fasheh } 626ccd979bdSMark Fasheh 627ccd979bdSMark Fasheh 628ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 629ccd979bdSMark Fasheh * mark the journal clean */ 630ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 631ccd979bdSMark Fasheh { 632ccd979bdSMark Fasheh int status; 633ccd979bdSMark Fasheh 634ccd979bdSMark Fasheh mlog_entry_void(); 635ccd979bdSMark Fasheh 636ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 637ccd979bdSMark Fasheh 638ccd979bdSMark Fasheh status = journal_wipe(journal->j_journal, full); 639ccd979bdSMark Fasheh if (status < 0) { 640ccd979bdSMark Fasheh mlog_errno(status); 641ccd979bdSMark Fasheh goto bail; 642ccd979bdSMark Fasheh } 643ccd979bdSMark Fasheh 644ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); 645ccd979bdSMark Fasheh if (status < 0) 646ccd979bdSMark Fasheh mlog_errno(status); 647ccd979bdSMark Fasheh 648ccd979bdSMark Fasheh bail: 649ccd979bdSMark Fasheh mlog_exit(status); 650ccd979bdSMark Fasheh return status; 651ccd979bdSMark Fasheh } 652ccd979bdSMark Fasheh 653ccd979bdSMark Fasheh /* 654ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 655ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 656ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 657ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 658ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 659ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 660ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 661ccd979bdSMark Fasheh * as we have the lock by now :) 662ccd979bdSMark Fasheh */ 663ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 664ccd979bdSMark Fasheh { 665ccd979bdSMark Fasheh int status = 0; 6664f902c37SMark Fasheh int i; 6678110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 6684f902c37SMark Fasheh #define CONCURRENT_JOURNAL_FILL 32ULL 669ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 670ccd979bdSMark Fasheh 671ccd979bdSMark Fasheh mlog_entry_void(); 672ccd979bdSMark Fasheh 673ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 674ccd979bdSMark Fasheh 6758110b073SMark Fasheh num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); 676ccd979bdSMark Fasheh v_blkno = 0; 6778110b073SMark Fasheh while (v_blkno < num_blocks) { 678ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 67949cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 680ccd979bdSMark Fasheh if (status < 0) { 681ccd979bdSMark Fasheh mlog_errno(status); 682ccd979bdSMark Fasheh goto bail; 683ccd979bdSMark Fasheh } 684ccd979bdSMark Fasheh 685ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 686ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 687ccd979bdSMark Fasheh 688dd4a2c2bSMark Fasheh /* We are reading journal data which should not 689dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 690ccd979bdSMark Fasheh status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 691ccd979bdSMark Fasheh p_blkno, p_blocks, bhs, 0, 692dd4a2c2bSMark Fasheh NULL); 693ccd979bdSMark Fasheh if (status < 0) { 694ccd979bdSMark Fasheh mlog_errno(status); 695ccd979bdSMark Fasheh goto bail; 696ccd979bdSMark Fasheh } 697ccd979bdSMark Fasheh 698ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 699ccd979bdSMark Fasheh brelse(bhs[i]); 700ccd979bdSMark Fasheh bhs[i] = NULL; 701ccd979bdSMark Fasheh } 702ccd979bdSMark Fasheh 703ccd979bdSMark Fasheh v_blkno += p_blocks; 704ccd979bdSMark Fasheh } 705ccd979bdSMark Fasheh 706ccd979bdSMark Fasheh bail: 707ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 708ccd979bdSMark Fasheh if (bhs[i]) 709ccd979bdSMark Fasheh brelse(bhs[i]); 710ccd979bdSMark Fasheh mlog_exit(status); 711ccd979bdSMark Fasheh return status; 712ccd979bdSMark Fasheh } 713ccd979bdSMark Fasheh 714ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 715ccd979bdSMark Fasheh struct list_head lri_list; 716ccd979bdSMark Fasheh int lri_slot; 717ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 718ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 719ccd979bdSMark Fasheh }; 720ccd979bdSMark Fasheh 721ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 722ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 723ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 724ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 725ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 726ccd979bdSMark Fasheh * recovering nodes into account. 727ccd979bdSMark Fasheh * 728ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 729ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 730ccd979bdSMark Fasheh */ 731c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 732ccd979bdSMark Fasheh { 733ccd979bdSMark Fasheh int ret; 734c4028958SDavid Howells struct ocfs2_journal *journal = 735c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 736c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 737ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 738800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 739ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 740ccd979bdSMark Fasheh 741ccd979bdSMark Fasheh mlog_entry_void(); 742ccd979bdSMark Fasheh 743ccd979bdSMark Fasheh mlog(0, "completing recovery from keventd\n"); 744ccd979bdSMark Fasheh 745ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 746ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 747ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 748ccd979bdSMark Fasheh 749800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 750ccd979bdSMark Fasheh list_del_init(&item->lri_list); 751ccd979bdSMark Fasheh 752ccd979bdSMark Fasheh mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 753ccd979bdSMark Fasheh 754ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 755ccd979bdSMark Fasheh if (la_dinode) { 756b0697053SMark Fasheh mlog(0, "Clean up local alloc %llu\n", 7571ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); 758ccd979bdSMark Fasheh 759ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 760ccd979bdSMark Fasheh la_dinode); 761ccd979bdSMark Fasheh if (ret < 0) 762ccd979bdSMark Fasheh mlog_errno(ret); 763ccd979bdSMark Fasheh 764ccd979bdSMark Fasheh kfree(la_dinode); 765ccd979bdSMark Fasheh } 766ccd979bdSMark Fasheh 767ccd979bdSMark Fasheh tl_dinode = item->lri_tl_dinode; 768ccd979bdSMark Fasheh if (tl_dinode) { 769b0697053SMark Fasheh mlog(0, "Clean up truncate log %llu\n", 7701ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); 771ccd979bdSMark Fasheh 772ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 773ccd979bdSMark Fasheh tl_dinode); 774ccd979bdSMark Fasheh if (ret < 0) 775ccd979bdSMark Fasheh mlog_errno(ret); 776ccd979bdSMark Fasheh 777ccd979bdSMark Fasheh kfree(tl_dinode); 778ccd979bdSMark Fasheh } 779ccd979bdSMark Fasheh 780ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 781ccd979bdSMark Fasheh if (ret < 0) 782ccd979bdSMark Fasheh mlog_errno(ret); 783ccd979bdSMark Fasheh 784ccd979bdSMark Fasheh kfree(item); 785ccd979bdSMark Fasheh } 786ccd979bdSMark Fasheh 787ccd979bdSMark Fasheh mlog(0, "Recovery completion\n"); 788ccd979bdSMark Fasheh mlog_exit_void(); 789ccd979bdSMark Fasheh } 790ccd979bdSMark Fasheh 791ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 792ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 793ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 794ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 795ccd979bdSMark Fasheh int slot_num, 796ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 797ccd979bdSMark Fasheh struct ocfs2_dinode *tl_dinode) 798ccd979bdSMark Fasheh { 799ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 800ccd979bdSMark Fasheh 801afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 802ccd979bdSMark Fasheh if (!item) { 803ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 804ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 805ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 806ccd979bdSMark Fasheh if (la_dinode) 807ccd979bdSMark Fasheh kfree(la_dinode); 808ccd979bdSMark Fasheh 809ccd979bdSMark Fasheh if (tl_dinode) 810ccd979bdSMark Fasheh kfree(tl_dinode); 811ccd979bdSMark Fasheh 812ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 813ccd979bdSMark Fasheh return; 814ccd979bdSMark Fasheh } 815ccd979bdSMark Fasheh 816ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 817ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 818ccd979bdSMark Fasheh item->lri_slot = slot_num; 819ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 820ccd979bdSMark Fasheh 821ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 822ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 823ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 824ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 825ccd979bdSMark Fasheh } 826ccd979bdSMark Fasheh 827ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 828ccd979bdSMark Fasheh * recovery for it's own slot. */ 829ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 830ccd979bdSMark Fasheh { 831ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 832ccd979bdSMark Fasheh 833ccd979bdSMark Fasheh if (osb->dirty) { 834ccd979bdSMark Fasheh /* No need to queue up our truncate_log as regular 835ccd979bdSMark Fasheh * cleanup will catch that. */ 836ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(journal, 837ccd979bdSMark Fasheh osb->slot_num, 838ccd979bdSMark Fasheh osb->local_alloc_copy, 839ccd979bdSMark Fasheh NULL); 840ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 841ccd979bdSMark Fasheh 842ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 843ccd979bdSMark Fasheh osb->dirty = 0; 844ccd979bdSMark Fasheh } 845ccd979bdSMark Fasheh } 846ccd979bdSMark Fasheh 847ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 848ccd979bdSMark Fasheh { 849ccd979bdSMark Fasheh int status, node_num; 850ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 851ccd979bdSMark Fasheh 852ccd979bdSMark Fasheh mlog_entry_void(); 853ccd979bdSMark Fasheh 854ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 855ccd979bdSMark Fasheh if (status < 0) { 856ccd979bdSMark Fasheh goto bail; 857ccd979bdSMark Fasheh } 858ccd979bdSMark Fasheh 859ccd979bdSMark Fasheh restart: 860ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 861ccd979bdSMark Fasheh if (status < 0) { 862ccd979bdSMark Fasheh mlog_errno(status); 863ccd979bdSMark Fasheh goto bail; 864ccd979bdSMark Fasheh } 865ccd979bdSMark Fasheh 866ccd979bdSMark Fasheh while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 867ccd979bdSMark Fasheh node_num = ocfs2_node_map_first_set_bit(osb, 868ccd979bdSMark Fasheh &osb->recovery_map); 869ccd979bdSMark Fasheh if (node_num == O2NM_INVALID_NODE_NUM) { 870ccd979bdSMark Fasheh mlog(0, "Out of nodes to recover.\n"); 871ccd979bdSMark Fasheh break; 872ccd979bdSMark Fasheh } 873ccd979bdSMark Fasheh 874ccd979bdSMark Fasheh status = ocfs2_recover_node(osb, node_num); 875ccd979bdSMark Fasheh if (status < 0) { 876ccd979bdSMark Fasheh mlog(ML_ERROR, 877ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 878ccd979bdSMark Fasheh status, node_num, 879ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 880ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 881ccd979bdSMark Fasheh continue; 882ccd979bdSMark Fasheh } 883ccd979bdSMark Fasheh 884ccd979bdSMark Fasheh ocfs2_recovery_map_clear(osb, node_num); 885ccd979bdSMark Fasheh } 886ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 887ccd979bdSMark Fasheh 888ccd979bdSMark Fasheh /* We always run recovery on our own orphan dir - the dead 88934d024f8SMark Fasheh * node(s) may have disallowd a previos inode delete. Re-processing 89034d024f8SMark Fasheh * is therefore required. */ 891ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 892ccd979bdSMark Fasheh NULL); 893ccd979bdSMark Fasheh 894ccd979bdSMark Fasheh bail: 895c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 896ccd979bdSMark Fasheh if (!status && 897ccd979bdSMark Fasheh !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 898c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 899ccd979bdSMark Fasheh goto restart; 900ccd979bdSMark Fasheh } 901ccd979bdSMark Fasheh 902ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 903ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 904ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 905ccd979bdSMark Fasheh 906c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 907ccd979bdSMark Fasheh 908ccd979bdSMark Fasheh mlog_exit(status); 909ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 910ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 911ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 912ccd979bdSMark Fasheh complete_and_exit(NULL, status); 913ccd979bdSMark Fasheh return status; 914ccd979bdSMark Fasheh } 915ccd979bdSMark Fasheh 916ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 917ccd979bdSMark Fasheh { 918ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 919ccd979bdSMark Fasheh node_num, osb->node_num); 920ccd979bdSMark Fasheh 921c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 922ccd979bdSMark Fasheh if (osb->disable_recovery) 923ccd979bdSMark Fasheh goto out; 924ccd979bdSMark Fasheh 925ccd979bdSMark Fasheh /* People waiting on recovery will wait on 926ccd979bdSMark Fasheh * the recovery map to empty. */ 927ccd979bdSMark Fasheh if (!ocfs2_recovery_map_set(osb, node_num)) 928ccd979bdSMark Fasheh mlog(0, "node %d already be in recovery.\n", node_num); 929ccd979bdSMark Fasheh 930ccd979bdSMark Fasheh mlog(0, "starting recovery thread...\n"); 931ccd979bdSMark Fasheh 932ccd979bdSMark Fasheh if (osb->recovery_thread_task) 933ccd979bdSMark Fasheh goto out; 934ccd979bdSMark Fasheh 935ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 93678427043SMark Fasheh "ocfs2rec"); 937ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 938ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 939ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 940ccd979bdSMark Fasheh } 941ccd979bdSMark Fasheh 942ccd979bdSMark Fasheh out: 943c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 944ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 945ccd979bdSMark Fasheh 946ccd979bdSMark Fasheh mlog_exit_void(); 947ccd979bdSMark Fasheh } 948ccd979bdSMark Fasheh 949ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 950ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 951ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 952ccd979bdSMark Fasheh int node_num, 953ccd979bdSMark Fasheh int slot_num) 954ccd979bdSMark Fasheh { 955ccd979bdSMark Fasheh int status; 956ccd979bdSMark Fasheh int got_lock = 0; 957ccd979bdSMark Fasheh unsigned int flags; 958ccd979bdSMark Fasheh struct inode *inode = NULL; 959ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 960ccd979bdSMark Fasheh journal_t *journal = NULL; 961ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 962ccd979bdSMark Fasheh 963ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 964ccd979bdSMark Fasheh slot_num); 965ccd979bdSMark Fasheh if (inode == NULL) { 966ccd979bdSMark Fasheh status = -EACCES; 967ccd979bdSMark Fasheh mlog_errno(status); 968ccd979bdSMark Fasheh goto done; 969ccd979bdSMark Fasheh } 970ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 971ccd979bdSMark Fasheh status = -EACCES; 972ccd979bdSMark Fasheh iput(inode); 973ccd979bdSMark Fasheh inode = NULL; 974ccd979bdSMark Fasheh mlog_errno(status); 975ccd979bdSMark Fasheh goto done; 976ccd979bdSMark Fasheh } 977ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 978ccd979bdSMark Fasheh 979e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 980ccd979bdSMark Fasheh if (status < 0) { 981e63aecb6SMark Fasheh mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); 982ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 983ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 984ccd979bdSMark Fasheh goto done; 985ccd979bdSMark Fasheh } 986ccd979bdSMark Fasheh got_lock = 1; 987ccd979bdSMark Fasheh 988ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 989ccd979bdSMark Fasheh 990ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 991ccd979bdSMark Fasheh 992ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 993ccd979bdSMark Fasheh mlog(0, "No recovery required for node %d\n", node_num); 994ccd979bdSMark Fasheh goto done; 995ccd979bdSMark Fasheh } 996ccd979bdSMark Fasheh 997ccd979bdSMark Fasheh mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 998ccd979bdSMark Fasheh node_num, slot_num, 999ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1000ccd979bdSMark Fasheh 1001ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1002ccd979bdSMark Fasheh 1003ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1004ccd979bdSMark Fasheh if (status < 0) { 1005ccd979bdSMark Fasheh mlog_errno(status); 1006ccd979bdSMark Fasheh goto done; 1007ccd979bdSMark Fasheh } 1008ccd979bdSMark Fasheh 1009ccd979bdSMark Fasheh mlog(0, "calling journal_init_inode\n"); 1010ccd979bdSMark Fasheh journal = journal_init_inode(inode); 1011ccd979bdSMark Fasheh if (journal == NULL) { 1012ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1013ccd979bdSMark Fasheh status = -EIO; 1014ccd979bdSMark Fasheh goto done; 1015ccd979bdSMark Fasheh } 1016ccd979bdSMark Fasheh 1017ccd979bdSMark Fasheh status = journal_load(journal); 1018ccd979bdSMark Fasheh if (status < 0) { 1019ccd979bdSMark Fasheh mlog_errno(status); 1020ccd979bdSMark Fasheh if (!igrab(inode)) 1021ccd979bdSMark Fasheh BUG(); 1022ccd979bdSMark Fasheh journal_destroy(journal); 1023ccd979bdSMark Fasheh goto done; 1024ccd979bdSMark Fasheh } 1025ccd979bdSMark Fasheh 1026ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1027ccd979bdSMark Fasheh 1028ccd979bdSMark Fasheh /* wipe the journal */ 1029ccd979bdSMark Fasheh mlog(0, "flushing the journal.\n"); 1030ccd979bdSMark Fasheh journal_lock_updates(journal); 1031ccd979bdSMark Fasheh status = journal_flush(journal); 1032ccd979bdSMark Fasheh journal_unlock_updates(journal); 1033ccd979bdSMark Fasheh if (status < 0) 1034ccd979bdSMark Fasheh mlog_errno(status); 1035ccd979bdSMark Fasheh 1036ccd979bdSMark Fasheh /* This will mark the node clean */ 1037ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1038ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1039ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1040ccd979bdSMark Fasheh 1041ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, inode); 1042ccd979bdSMark Fasheh if (status < 0) 1043ccd979bdSMark Fasheh mlog_errno(status); 1044ccd979bdSMark Fasheh 1045ccd979bdSMark Fasheh if (!igrab(inode)) 1046ccd979bdSMark Fasheh BUG(); 1047ccd979bdSMark Fasheh 1048ccd979bdSMark Fasheh journal_destroy(journal); 1049ccd979bdSMark Fasheh 1050ccd979bdSMark Fasheh done: 1051ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1052ccd979bdSMark Fasheh if (got_lock) 1053e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1054ccd979bdSMark Fasheh 1055ccd979bdSMark Fasheh if (inode) 1056ccd979bdSMark Fasheh iput(inode); 1057ccd979bdSMark Fasheh 1058ccd979bdSMark Fasheh if (bh) 1059ccd979bdSMark Fasheh brelse(bh); 1060ccd979bdSMark Fasheh 1061ccd979bdSMark Fasheh mlog_exit(status); 1062ccd979bdSMark Fasheh return status; 1063ccd979bdSMark Fasheh } 1064ccd979bdSMark Fasheh 1065ccd979bdSMark Fasheh /* 1066ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1067ccd979bdSMark Fasheh * - Replay it's journal 1068ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1069ccd979bdSMark Fasheh * - Stamp a clean truncate log 1070ccd979bdSMark Fasheh * - Mark the node clean 1071ccd979bdSMark Fasheh * 1072ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1073ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1074ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1075ccd979bdSMark Fasheh * far less concerning. 1076ccd979bdSMark Fasheh */ 1077ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 1078ccd979bdSMark Fasheh int node_num) 1079ccd979bdSMark Fasheh { 1080ccd979bdSMark Fasheh int status = 0; 1081ccd979bdSMark Fasheh int slot_num; 1082ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1083ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1084ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1085ccd979bdSMark Fasheh 1086ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1087ccd979bdSMark Fasheh node_num, osb->node_num); 1088ccd979bdSMark Fasheh 1089ccd979bdSMark Fasheh mlog(0, "checking node %d\n", node_num); 1090ccd979bdSMark Fasheh 1091ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1092ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1093ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1094ccd979bdSMark Fasheh 1095ccd979bdSMark Fasheh slot_num = ocfs2_node_num_to_slot(si, node_num); 1096ccd979bdSMark Fasheh if (slot_num == OCFS2_INVALID_SLOT) { 1097ccd979bdSMark Fasheh status = 0; 1098ccd979bdSMark Fasheh mlog(0, "no slot for this node, so no recovery required.\n"); 1099ccd979bdSMark Fasheh goto done; 1100ccd979bdSMark Fasheh } 1101ccd979bdSMark Fasheh 1102ccd979bdSMark Fasheh mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1103ccd979bdSMark Fasheh 1104ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1105ccd979bdSMark Fasheh if (status < 0) { 1106ccd979bdSMark Fasheh mlog_errno(status); 1107ccd979bdSMark Fasheh goto done; 1108ccd979bdSMark Fasheh } 1109ccd979bdSMark Fasheh 1110ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1111ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1112ccd979bdSMark Fasheh if (status < 0) { 1113ccd979bdSMark Fasheh mlog_errno(status); 1114ccd979bdSMark Fasheh goto done; 1115ccd979bdSMark Fasheh } 1116ccd979bdSMark Fasheh 1117ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1118ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1119ccd979bdSMark Fasheh * recovery. */ 1120ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1121ccd979bdSMark Fasheh if (status < 0) 1122ccd979bdSMark Fasheh mlog_errno(status); 1123ccd979bdSMark Fasheh 1124ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1125ccd979bdSMark Fasheh * harmful place to get an error... */ 11268e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1127ccd979bdSMark Fasheh if (status < 0) 1128ccd979bdSMark Fasheh mlog_errno(status); 1129ccd979bdSMark Fasheh 1130ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1131ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1132ccd979bdSMark Fasheh tl_copy); 1133ccd979bdSMark Fasheh 1134ccd979bdSMark Fasheh status = 0; 1135ccd979bdSMark Fasheh done: 1136ccd979bdSMark Fasheh 1137ccd979bdSMark Fasheh mlog_exit(status); 1138ccd979bdSMark Fasheh return status; 1139ccd979bdSMark Fasheh } 1140ccd979bdSMark Fasheh 1141ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1142ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1143ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1144ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1145ccd979bdSMark Fasheh int slot_num) 1146ccd979bdSMark Fasheh { 1147ccd979bdSMark Fasheh int status, flags; 1148ccd979bdSMark Fasheh struct inode *inode = NULL; 1149ccd979bdSMark Fasheh 1150ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1151ccd979bdSMark Fasheh slot_num); 1152ccd979bdSMark Fasheh if (inode == NULL) { 1153ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1154ccd979bdSMark Fasheh status = -EACCES; 1155ccd979bdSMark Fasheh goto bail; 1156ccd979bdSMark Fasheh } 1157ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1158ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1159ccd979bdSMark Fasheh iput(inode); 1160ccd979bdSMark Fasheh inode = NULL; 1161ccd979bdSMark Fasheh status = -EACCES; 1162ccd979bdSMark Fasheh goto bail; 1163ccd979bdSMark Fasheh } 1164ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1165ccd979bdSMark Fasheh 1166ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1167e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1168ccd979bdSMark Fasheh if (status < 0) { 1169ccd979bdSMark Fasheh if (status != -EAGAIN) 1170ccd979bdSMark Fasheh mlog_errno(status); 1171ccd979bdSMark Fasheh goto bail; 1172ccd979bdSMark Fasheh } 1173ccd979bdSMark Fasheh 1174e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1175ccd979bdSMark Fasheh bail: 1176ccd979bdSMark Fasheh if (inode) 1177ccd979bdSMark Fasheh iput(inode); 1178ccd979bdSMark Fasheh 1179ccd979bdSMark Fasheh return status; 1180ccd979bdSMark Fasheh } 1181ccd979bdSMark Fasheh 1182ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1183ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1184ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1185ccd979bdSMark Fasheh { 1186ccd979bdSMark Fasheh int status, i, node_num; 1187ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1188ccd979bdSMark Fasheh 1189ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1190ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1191ccd979bdSMark Fasheh 1192ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1193ccd979bdSMark Fasheh for(i = 0; i < si->si_num_slots; i++) { 1194ccd979bdSMark Fasheh if (i == osb->slot_num) 1195ccd979bdSMark Fasheh continue; 1196ccd979bdSMark Fasheh if (ocfs2_is_empty_slot(si, i)) 1197ccd979bdSMark Fasheh continue; 1198ccd979bdSMark Fasheh 1199ccd979bdSMark Fasheh node_num = si->si_global_node_nums[i]; 1200ccd979bdSMark Fasheh if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) 1201ccd979bdSMark Fasheh continue; 1202ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1203ccd979bdSMark Fasheh 1204ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1205ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1206ccd979bdSMark Fasheh * file here to test if he's alive. */ 1207ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1208ccd979bdSMark Fasheh if (!status) { 1209ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1210ccd979bdSMark Fasheh * the recovery thread can't race us on 1211ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1212ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1213ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1214ccd979bdSMark Fasheh mlog_errno(status); 1215ccd979bdSMark Fasheh goto bail; 1216ccd979bdSMark Fasheh } 1217ccd979bdSMark Fasheh 1218ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1219ccd979bdSMark Fasheh } 1220ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1221ccd979bdSMark Fasheh 1222ccd979bdSMark Fasheh status = 0; 1223ccd979bdSMark Fasheh bail: 1224ccd979bdSMark Fasheh mlog_exit(status); 1225ccd979bdSMark Fasheh return status; 1226ccd979bdSMark Fasheh } 1227ccd979bdSMark Fasheh 12285eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 12295eae5b96SMark Fasheh struct inode *head; 12305eae5b96SMark Fasheh struct ocfs2_super *osb; 12315eae5b96SMark Fasheh }; 12325eae5b96SMark Fasheh 12335eae5b96SMark Fasheh static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, 12345eae5b96SMark Fasheh loff_t pos, u64 ino, unsigned type) 12355eae5b96SMark Fasheh { 12365eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv *p = priv; 12375eae5b96SMark Fasheh struct inode *iter; 12385eae5b96SMark Fasheh 12395eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 12405eae5b96SMark Fasheh return 0; 12415eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 12425eae5b96SMark Fasheh return 0; 12435eae5b96SMark Fasheh 12445eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 12455eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 12465fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 12475eae5b96SMark Fasheh if (IS_ERR(iter)) 12485eae5b96SMark Fasheh return 0; 12495eae5b96SMark Fasheh 12505eae5b96SMark Fasheh mlog(0, "queue orphan %llu\n", 12515eae5b96SMark Fasheh (unsigned long long)OCFS2_I(iter)->ip_blkno); 12525eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 12535eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 12545eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 12555eae5b96SMark Fasheh p->head = iter; 12565eae5b96SMark Fasheh 12575eae5b96SMark Fasheh return 0; 12585eae5b96SMark Fasheh } 12595eae5b96SMark Fasheh 1260b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1261b4df6ed8SMark Fasheh int slot, 1262b4df6ed8SMark Fasheh struct inode **head) 1263ccd979bdSMark Fasheh { 1264b4df6ed8SMark Fasheh int status; 1265ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 12665eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv priv; 12675eae5b96SMark Fasheh loff_t pos = 0; 12685eae5b96SMark Fasheh 12695eae5b96SMark Fasheh priv.osb = osb; 12705eae5b96SMark Fasheh priv.head = *head; 1271ccd979bdSMark Fasheh 1272ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1273ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 1274ccd979bdSMark Fasheh slot); 1275ccd979bdSMark Fasheh if (!orphan_dir_inode) { 1276ccd979bdSMark Fasheh status = -ENOENT; 1277ccd979bdSMark Fasheh mlog_errno(status); 1278b4df6ed8SMark Fasheh return status; 1279ccd979bdSMark Fasheh } 1280ccd979bdSMark Fasheh 12811b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 1282e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 1283ccd979bdSMark Fasheh if (status < 0) { 1284ccd979bdSMark Fasheh mlog_errno(status); 1285ccd979bdSMark Fasheh goto out; 1286ccd979bdSMark Fasheh } 1287ccd979bdSMark Fasheh 12885eae5b96SMark Fasheh status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, 12895eae5b96SMark Fasheh ocfs2_orphan_filldir); 12905eae5b96SMark Fasheh if (status) { 1291ccd979bdSMark Fasheh mlog_errno(status); 1292a86370fbSMark Fasheh goto out_cluster; 1293ccd979bdSMark Fasheh } 1294ccd979bdSMark Fasheh 12955eae5b96SMark Fasheh *head = priv.head; 1296ccd979bdSMark Fasheh 1297a86370fbSMark Fasheh out_cluster: 1298e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 1299b4df6ed8SMark Fasheh out: 1300b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 1301ccd979bdSMark Fasheh iput(orphan_dir_inode); 1302b4df6ed8SMark Fasheh return status; 1303b4df6ed8SMark Fasheh } 1304b4df6ed8SMark Fasheh 1305b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1306b4df6ed8SMark Fasheh int slot) 1307b4df6ed8SMark Fasheh { 1308b4df6ed8SMark Fasheh int ret; 1309b4df6ed8SMark Fasheh 1310b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1311b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 1312b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1313b4df6ed8SMark Fasheh return ret; 1314b4df6ed8SMark Fasheh } 1315b4df6ed8SMark Fasheh 1316b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1317b4df6ed8SMark Fasheh int slot) 1318b4df6ed8SMark Fasheh { 1319b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1320b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 1321b4df6ed8SMark Fasheh * know to quit early. */ 1322b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1323b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 1324b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 1325b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 1326b4df6ed8SMark Fasheh * them. */ 1327b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1328b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 1329b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 1330b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1331b4df6ed8SMark Fasheh } 1332b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1333b4df6ed8SMark Fasheh } 1334b4df6ed8SMark Fasheh 1335b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1336b4df6ed8SMark Fasheh int slot) 1337b4df6ed8SMark Fasheh { 1338b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1339b4df6ed8SMark Fasheh } 1340b4df6ed8SMark Fasheh 1341b4df6ed8SMark Fasheh /* 1342b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 1343b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 1344b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 1345b4df6ed8SMark Fasheh * (most) of the rest of the work. 1346b4df6ed8SMark Fasheh * 1347b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 1348b4df6ed8SMark Fasheh * couple of extra considerations. 1349b4df6ed8SMark Fasheh * 1350b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 1351b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 1352b4df6ed8SMark Fasheh * an invalid inode. 1353b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 1354b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 1355b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 1356b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 1357b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 1358b4df6ed8SMark Fasheh */ 1359b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1360b4df6ed8SMark Fasheh int slot) 1361b4df6ed8SMark Fasheh { 1362b4df6ed8SMark Fasheh int ret = 0; 1363b4df6ed8SMark Fasheh struct inode *inode = NULL; 1364b4df6ed8SMark Fasheh struct inode *iter; 1365b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 1366b4df6ed8SMark Fasheh 1367b4df6ed8SMark Fasheh mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1368b4df6ed8SMark Fasheh 1369b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 1370b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 1371b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 1372b4df6ed8SMark Fasheh 1373b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 1374b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 1375b4df6ed8SMark Fasheh if (ret) 1376b4df6ed8SMark Fasheh mlog_errno(ret); 1377ccd979bdSMark Fasheh 1378ccd979bdSMark Fasheh while (inode) { 1379ccd979bdSMark Fasheh oi = OCFS2_I(inode); 1380b0697053SMark Fasheh mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1381ccd979bdSMark Fasheh 1382ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 1383ccd979bdSMark Fasheh 1384ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 138534d024f8SMark Fasheh /* The remote delete code may have set these on the 138634d024f8SMark Fasheh * assumption that the other node would wipe them 138734d024f8SMark Fasheh * successfully. If they are still in the node's 138834d024f8SMark Fasheh * orphan dir, we need to reset that state. */ 1389ccd979bdSMark Fasheh oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1390ccd979bdSMark Fasheh 1391ccd979bdSMark Fasheh /* Set the proper information to get us going into 1392ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 1393ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1394ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 1395ccd979bdSMark Fasheh 1396ccd979bdSMark Fasheh iput(inode); 1397ccd979bdSMark Fasheh 1398ccd979bdSMark Fasheh inode = iter; 1399ccd979bdSMark Fasheh } 1400ccd979bdSMark Fasheh 1401b4df6ed8SMark Fasheh return ret; 1402ccd979bdSMark Fasheh } 1403ccd979bdSMark Fasheh 1404ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1405ccd979bdSMark Fasheh { 1406ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 1407ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 1408ccd979bdSMark Fasheh * or DISABLED. */ 1409ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 1410ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1411ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1412ccd979bdSMark Fasheh 1413ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 1414ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 1415ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 1416ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1417ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 1418ccd979bdSMark Fasheh return -EBUSY; 1419ccd979bdSMark Fasheh } 1420ccd979bdSMark Fasheh 1421ccd979bdSMark Fasheh return 0; 1422ccd979bdSMark Fasheh } 1423ccd979bdSMark Fasheh 1424ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 1425ccd979bdSMark Fasheh { 1426ccd979bdSMark Fasheh int status; 1427ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1428ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1429ccd979bdSMark Fasheh 1430ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 1431ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 1432ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 1433ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 1434ccd979bdSMark Fasheh while (!(kthread_should_stop() && 1435ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 1436ccd979bdSMark Fasheh 1437745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 1438ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 1439745ae8baSMark Fasheh || kthread_should_stop()); 1440ccd979bdSMark Fasheh 1441ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 1442ccd979bdSMark Fasheh if (status < 0) 1443ccd979bdSMark Fasheh mlog_errno(status); 1444ccd979bdSMark Fasheh 1445ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1446ccd979bdSMark Fasheh mlog(ML_KTHREAD, 1447ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 1448ccd979bdSMark Fasheh "shutdown\n", 1449ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 1450ccd979bdSMark Fasheh } 1451ccd979bdSMark Fasheh } 1452ccd979bdSMark Fasheh 1453ccd979bdSMark Fasheh return 0; 1454ccd979bdSMark Fasheh } 1455ccd979bdSMark Fasheh 1456ccd979bdSMark Fasheh /* Look for a dirty journal without taking any cluster locks. Used for 1457ccd979bdSMark Fasheh * hard readonly access to determine whether the file system journals 1458ccd979bdSMark Fasheh * require recovery. */ 1459ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1460ccd979bdSMark Fasheh { 1461ccd979bdSMark Fasheh int ret = 0; 1462ccd979bdSMark Fasheh unsigned int slot; 1463ccd979bdSMark Fasheh struct buffer_head *di_bh; 1464ccd979bdSMark Fasheh struct ocfs2_dinode *di; 1465ccd979bdSMark Fasheh struct inode *journal = NULL; 1466ccd979bdSMark Fasheh 1467ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 1468ccd979bdSMark Fasheh journal = ocfs2_get_system_file_inode(osb, 1469ccd979bdSMark Fasheh JOURNAL_SYSTEM_INODE, 1470ccd979bdSMark Fasheh slot); 1471ccd979bdSMark Fasheh if (!journal || is_bad_inode(journal)) { 1472ccd979bdSMark Fasheh ret = -EACCES; 1473ccd979bdSMark Fasheh mlog_errno(ret); 1474ccd979bdSMark Fasheh goto out; 1475ccd979bdSMark Fasheh } 1476ccd979bdSMark Fasheh 1477ccd979bdSMark Fasheh di_bh = NULL; 1478ccd979bdSMark Fasheh ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, 1479ccd979bdSMark Fasheh 0, journal); 1480ccd979bdSMark Fasheh if (ret < 0) { 1481ccd979bdSMark Fasheh mlog_errno(ret); 1482ccd979bdSMark Fasheh goto out; 1483ccd979bdSMark Fasheh } 1484ccd979bdSMark Fasheh 1485ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 1486ccd979bdSMark Fasheh 1487ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 1488ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 1489ccd979bdSMark Fasheh ret = -EROFS; 1490ccd979bdSMark Fasheh 1491ccd979bdSMark Fasheh brelse(di_bh); 1492ccd979bdSMark Fasheh if (ret) 1493ccd979bdSMark Fasheh break; 1494ccd979bdSMark Fasheh } 1495ccd979bdSMark Fasheh 1496ccd979bdSMark Fasheh out: 1497ccd979bdSMark Fasheh if (journal) 1498ccd979bdSMark Fasheh iput(journal); 1499ccd979bdSMark Fasheh 1500ccd979bdSMark Fasheh return ret; 1501ccd979bdSMark Fasheh } 1502