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" 38ccd979bdSMark Fasheh #include "dlmglue.h" 39ccd979bdSMark Fasheh #include "extent_map.h" 40ccd979bdSMark Fasheh #include "heartbeat.h" 41ccd979bdSMark Fasheh #include "inode.h" 42ccd979bdSMark Fasheh #include "journal.h" 43ccd979bdSMark Fasheh #include "localalloc.h" 44ccd979bdSMark Fasheh #include "namei.h" 45ccd979bdSMark Fasheh #include "slot_map.h" 46ccd979bdSMark Fasheh #include "super.h" 47ccd979bdSMark Fasheh #include "vote.h" 48ccd979bdSMark Fasheh #include "sysfile.h" 49ccd979bdSMark Fasheh 50ccd979bdSMark Fasheh #include "buffer_head_io.h" 51ccd979bdSMark Fasheh 5234af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 53ccd979bdSMark Fasheh 54ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 55ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 56ccd979bdSMark Fasheh int node_num); 57ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 58ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 59ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb); 60ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 61ccd979bdSMark Fasheh int dirty); 62ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 63ccd979bdSMark Fasheh int slot_num); 64ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 65ccd979bdSMark Fasheh int slot); 66ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 67ccd979bdSMark Fasheh 68ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 69ccd979bdSMark Fasheh { 70ccd979bdSMark Fasheh int status = 0; 71ccd979bdSMark Fasheh unsigned int flushed; 72ccd979bdSMark Fasheh unsigned long old_id; 73ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 74ccd979bdSMark Fasheh 75ccd979bdSMark Fasheh mlog_entry_void(); 76ccd979bdSMark Fasheh 77ccd979bdSMark Fasheh journal = osb->journal; 78ccd979bdSMark Fasheh 79ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 80ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 81ccd979bdSMark Fasheh 82ccd979bdSMark Fasheh if (atomic_read(&journal->j_num_trans) == 0) { 83ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 84ccd979bdSMark Fasheh mlog(0, "No transactions for me to flush!\n"); 85ccd979bdSMark Fasheh goto finally; 86ccd979bdSMark Fasheh } 87ccd979bdSMark Fasheh 88ccd979bdSMark Fasheh journal_lock_updates(journal->j_journal); 89ccd979bdSMark Fasheh status = journal_flush(journal->j_journal); 90ccd979bdSMark Fasheh journal_unlock_updates(journal->j_journal); 91ccd979bdSMark Fasheh if (status < 0) { 92ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 93ccd979bdSMark Fasheh mlog_errno(status); 94ccd979bdSMark Fasheh goto finally; 95ccd979bdSMark Fasheh } 96ccd979bdSMark Fasheh 97ccd979bdSMark Fasheh old_id = ocfs2_inc_trans_id(journal); 98ccd979bdSMark Fasheh 99ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 100ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 101ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 102ccd979bdSMark Fasheh 103ccd979bdSMark Fasheh mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 104ccd979bdSMark Fasheh journal->j_trans_id, flushed); 105ccd979bdSMark Fasheh 106ccd979bdSMark Fasheh ocfs2_kick_vote_thread(osb); 107ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 108ccd979bdSMark Fasheh finally: 109ccd979bdSMark Fasheh mlog_exit(status); 110ccd979bdSMark Fasheh return status; 111ccd979bdSMark Fasheh } 112ccd979bdSMark Fasheh 113ccd979bdSMark Fasheh /* pass it NULL and it will allocate a new handle object for you. If 114ccd979bdSMark Fasheh * you pass it a handle however, it may still return error, in which 115ccd979bdSMark Fasheh * case it has free'd the passed handle for you. */ 1161fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 117ccd979bdSMark Fasheh { 118ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 1191fabe148SMark Fasheh handle_t *handle; 120ccd979bdSMark Fasheh 121ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 122ccd979bdSMark Fasheh 12365eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 12465eff9ccSMark Fasheh return ERR_PTR(-EROFS); 125ccd979bdSMark Fasheh 126ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 127ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 128ccd979bdSMark Fasheh 129ccd979bdSMark Fasheh /* JBD might support this, but our journalling code doesn't yet. */ 130ccd979bdSMark Fasheh if (journal_current_handle()) { 131ccd979bdSMark Fasheh mlog(ML_ERROR, "Recursive transaction attempted!\n"); 132ccd979bdSMark Fasheh BUG(); 133ccd979bdSMark Fasheh } 134ccd979bdSMark Fasheh 135ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 136ccd979bdSMark Fasheh 1371fabe148SMark Fasheh handle = journal_start(journal, max_buffs); 1381fabe148SMark Fasheh if (IS_ERR(handle)) { 139ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 140ccd979bdSMark Fasheh 1411fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 142ccd979bdSMark Fasheh 143ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 144ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 1451fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 146ccd979bdSMark Fasheh } 147c271c5c2SSunil Mushran } else { 148c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 149ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 150c271c5c2SSunil Mushran } 151ccd979bdSMark Fasheh 152ccd979bdSMark Fasheh return handle; 153ccd979bdSMark Fasheh } 154ccd979bdSMark Fasheh 1551fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 1561fabe148SMark Fasheh handle_t *handle) 157ccd979bdSMark Fasheh { 1581fabe148SMark Fasheh int ret; 15902dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 160ccd979bdSMark Fasheh 161ccd979bdSMark Fasheh BUG_ON(!handle); 162ccd979bdSMark Fasheh 1631fabe148SMark Fasheh ret = journal_stop(handle); 1641fabe148SMark Fasheh if (ret < 0) 1651fabe148SMark Fasheh mlog_errno(ret); 166ccd979bdSMark Fasheh 167ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 168ccd979bdSMark Fasheh 1691fabe148SMark Fasheh return ret; 170ccd979bdSMark Fasheh } 171ccd979bdSMark Fasheh 172ccd979bdSMark Fasheh /* 173ccd979bdSMark Fasheh * 'nblocks' is what you want to add to the current 174ccd979bdSMark Fasheh * transaction. extend_trans will either extend the current handle by 175ccd979bdSMark Fasheh * nblocks, or commit it and start a new one with nblocks credits. 176ccd979bdSMark Fasheh * 177ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 178ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 179ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 180ccd979bdSMark Fasheh * 181ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 182ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 183ccd979bdSMark Fasheh * cluster locks associated with this handle. 184ccd979bdSMark Fasheh */ 1851fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 186ccd979bdSMark Fasheh { 187ccd979bdSMark Fasheh int status; 188ccd979bdSMark Fasheh 189ccd979bdSMark Fasheh BUG_ON(!handle); 190ccd979bdSMark Fasheh BUG_ON(!nblocks); 191ccd979bdSMark Fasheh 192ccd979bdSMark Fasheh mlog_entry_void(); 193ccd979bdSMark Fasheh 194ccd979bdSMark Fasheh mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 195ccd979bdSMark Fasheh 1961fc58146SMark Fasheh status = journal_extend(handle, nblocks); 197ccd979bdSMark Fasheh if (status < 0) { 198ccd979bdSMark Fasheh mlog_errno(status); 199ccd979bdSMark Fasheh goto bail; 200ccd979bdSMark Fasheh } 201ccd979bdSMark Fasheh 202ccd979bdSMark Fasheh if (status > 0) { 203ccd979bdSMark Fasheh mlog(0, "journal_extend failed, trying journal_restart\n"); 2041fc58146SMark Fasheh status = journal_restart(handle, nblocks); 205ccd979bdSMark Fasheh if (status < 0) { 206ccd979bdSMark Fasheh mlog_errno(status); 207ccd979bdSMark Fasheh goto bail; 208ccd979bdSMark Fasheh } 20901ddf1e1SMark Fasheh } 210ccd979bdSMark Fasheh 211ccd979bdSMark Fasheh status = 0; 212ccd979bdSMark Fasheh bail: 213ccd979bdSMark Fasheh 214ccd979bdSMark Fasheh mlog_exit(status); 215ccd979bdSMark Fasheh return status; 216ccd979bdSMark Fasheh } 217ccd979bdSMark Fasheh 2181fabe148SMark Fasheh int ocfs2_journal_access(handle_t *handle, 219ccd979bdSMark Fasheh struct inode *inode, 220ccd979bdSMark Fasheh struct buffer_head *bh, 221ccd979bdSMark Fasheh int type) 222ccd979bdSMark Fasheh { 223ccd979bdSMark Fasheh int status; 224ccd979bdSMark Fasheh 225ccd979bdSMark Fasheh BUG_ON(!inode); 226ccd979bdSMark Fasheh BUG_ON(!handle); 227ccd979bdSMark Fasheh BUG_ON(!bh); 228ccd979bdSMark Fasheh 229205f87f6SBadari Pulavarty mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 230ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, type, 231ccd979bdSMark Fasheh (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 232ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_CREATE" : 233ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_WRITE", 234ccd979bdSMark Fasheh bh->b_size); 235ccd979bdSMark Fasheh 236ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 237ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 238ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 239ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 240ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 241ccd979bdSMark Fasheh BUG(); 242ccd979bdSMark Fasheh } 243ccd979bdSMark Fasheh 244ccd979bdSMark Fasheh /* Set the current transaction information on the inode so 245ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 246ccd979bdSMark Fasheh * on this inode or not. We're protected from the commit 247ccd979bdSMark Fasheh * thread updating the current transaction id until 248ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 249ccd979bdSMark Fasheh * j_trans_barrier for us. */ 250ccd979bdSMark Fasheh ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 251ccd979bdSMark Fasheh 252251b6eccSMark Fasheh mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 253ccd979bdSMark Fasheh switch (type) { 254ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 255ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 2561fabe148SMark Fasheh status = journal_get_write_access(handle, bh); 257ccd979bdSMark Fasheh break; 258ccd979bdSMark Fasheh 259ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 2601fabe148SMark Fasheh status = journal_get_undo_access(handle, bh); 261ccd979bdSMark Fasheh break; 262ccd979bdSMark Fasheh 263ccd979bdSMark Fasheh default: 264ccd979bdSMark Fasheh status = -EINVAL; 265ccd979bdSMark Fasheh mlog(ML_ERROR, "Uknown access type!\n"); 266ccd979bdSMark Fasheh } 267251b6eccSMark Fasheh mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 268ccd979bdSMark Fasheh 269ccd979bdSMark Fasheh if (status < 0) 270ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 271ccd979bdSMark Fasheh status, type); 272ccd979bdSMark Fasheh 273ccd979bdSMark Fasheh mlog_exit(status); 274ccd979bdSMark Fasheh return status; 275ccd979bdSMark Fasheh } 276ccd979bdSMark Fasheh 2771fabe148SMark Fasheh int ocfs2_journal_dirty(handle_t *handle, 278ccd979bdSMark Fasheh struct buffer_head *bh) 279ccd979bdSMark Fasheh { 280ccd979bdSMark Fasheh int status; 281ccd979bdSMark Fasheh 282ccd979bdSMark Fasheh mlog_entry("(bh->b_blocknr=%llu)\n", 283ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 284ccd979bdSMark Fasheh 2851fabe148SMark Fasheh status = journal_dirty_metadata(handle, bh); 286ccd979bdSMark Fasheh if (status < 0) 287ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not dirty metadata buffer. " 288ccd979bdSMark Fasheh "(bh->b_blocknr=%llu)\n", 289ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 290ccd979bdSMark Fasheh 291ccd979bdSMark Fasheh mlog_exit(status); 292ccd979bdSMark Fasheh return status; 293ccd979bdSMark Fasheh } 294ccd979bdSMark Fasheh 295ccd979bdSMark Fasheh int ocfs2_journal_dirty_data(handle_t *handle, 296ccd979bdSMark Fasheh struct buffer_head *bh) 297ccd979bdSMark Fasheh { 298ccd979bdSMark Fasheh int err = journal_dirty_data(handle, bh); 299ccd979bdSMark Fasheh if (err) 300ccd979bdSMark Fasheh mlog_errno(err); 301ccd979bdSMark Fasheh /* TODO: When we can handle it, abort the handle and go RO on 302ccd979bdSMark Fasheh * error here. */ 303ccd979bdSMark Fasheh 304ccd979bdSMark Fasheh return err; 305ccd979bdSMark Fasheh } 306ccd979bdSMark Fasheh 307ccd979bdSMark Fasheh #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) 308ccd979bdSMark Fasheh 309ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 310ccd979bdSMark Fasheh { 311ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 312ccd979bdSMark Fasheh 313ccd979bdSMark Fasheh spin_lock(&journal->j_state_lock); 314ccd979bdSMark Fasheh journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 315ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 316ccd979bdSMark Fasheh journal->j_flags |= JFS_BARRIER; 317ccd979bdSMark Fasheh else 318ccd979bdSMark Fasheh journal->j_flags &= ~JFS_BARRIER; 319ccd979bdSMark Fasheh spin_unlock(&journal->j_state_lock); 320ccd979bdSMark Fasheh } 321ccd979bdSMark Fasheh 322ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 323ccd979bdSMark Fasheh { 324ccd979bdSMark Fasheh int status = -1; 325ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 326ccd979bdSMark Fasheh journal_t *j_journal = NULL; 327ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 328ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 329ccd979bdSMark Fasheh struct ocfs2_super *osb; 330ccd979bdSMark Fasheh int meta_lock = 0; 331ccd979bdSMark Fasheh 332ccd979bdSMark Fasheh mlog_entry_void(); 333ccd979bdSMark Fasheh 334ccd979bdSMark Fasheh BUG_ON(!journal); 335ccd979bdSMark Fasheh 336ccd979bdSMark Fasheh osb = journal->j_osb; 337ccd979bdSMark Fasheh 338ccd979bdSMark Fasheh /* already have the inode for our journal */ 339ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 340ccd979bdSMark Fasheh osb->slot_num); 341ccd979bdSMark Fasheh if (inode == NULL) { 342ccd979bdSMark Fasheh status = -EACCES; 343ccd979bdSMark Fasheh mlog_errno(status); 344ccd979bdSMark Fasheh goto done; 345ccd979bdSMark Fasheh } 346ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 347ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 348ccd979bdSMark Fasheh iput(inode); 349ccd979bdSMark Fasheh inode = NULL; 350ccd979bdSMark Fasheh status = -EACCES; 351ccd979bdSMark Fasheh goto done; 352ccd979bdSMark Fasheh } 353ccd979bdSMark Fasheh 354ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 355ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 356ccd979bdSMark Fasheh 3576eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 3586eff5790SMark Fasheh * changes in a live cluster so it can be considered an 3596eff5790SMark Fasheh * exception to the rule. */ 3604bcec184SMark Fasheh status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 361ccd979bdSMark Fasheh if (status < 0) { 362ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 363ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 364ccd979bdSMark Fasheh goto done; 365ccd979bdSMark Fasheh } 366ccd979bdSMark Fasheh 367ccd979bdSMark Fasheh meta_lock = 1; 368ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 369ccd979bdSMark Fasheh 370ccd979bdSMark Fasheh if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 371ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 372ccd979bdSMark Fasheh inode->i_size); 373ccd979bdSMark Fasheh status = -EINVAL; 374ccd979bdSMark Fasheh goto done; 375ccd979bdSMark Fasheh } 376ccd979bdSMark Fasheh 377ccd979bdSMark Fasheh mlog(0, "inode->i_size = %lld\n", inode->i_size); 3785515eff8SAndrew Morton mlog(0, "inode->i_blocks = %llu\n", 3795515eff8SAndrew Morton (unsigned long long)inode->i_blocks); 380ccd979bdSMark Fasheh mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 381ccd979bdSMark Fasheh 382ccd979bdSMark Fasheh /* call the kernels journal init function now */ 383ccd979bdSMark Fasheh j_journal = journal_init_inode(inode); 384ccd979bdSMark Fasheh if (j_journal == NULL) { 385ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 386ccd979bdSMark Fasheh status = -EINVAL; 387ccd979bdSMark Fasheh goto done; 388ccd979bdSMark Fasheh } 389ccd979bdSMark Fasheh 390ccd979bdSMark Fasheh mlog(0, "Returned from journal_init_inode\n"); 391ccd979bdSMark Fasheh mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 392ccd979bdSMark Fasheh 393ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 394ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 395ccd979bdSMark Fasheh 396ccd979bdSMark Fasheh journal->j_journal = j_journal; 397ccd979bdSMark Fasheh journal->j_inode = inode; 398ccd979bdSMark Fasheh journal->j_bh = bh; 399ccd979bdSMark Fasheh 400ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 401ccd979bdSMark Fasheh 402ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 403ccd979bdSMark Fasheh 404ccd979bdSMark Fasheh status = 0; 405ccd979bdSMark Fasheh done: 406ccd979bdSMark Fasheh if (status < 0) { 407ccd979bdSMark Fasheh if (meta_lock) 408ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 409ccd979bdSMark Fasheh if (bh != NULL) 410ccd979bdSMark Fasheh brelse(bh); 411ccd979bdSMark Fasheh if (inode) { 412ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 413ccd979bdSMark Fasheh iput(inode); 414ccd979bdSMark Fasheh } 415ccd979bdSMark Fasheh } 416ccd979bdSMark Fasheh 417ccd979bdSMark Fasheh mlog_exit(status); 418ccd979bdSMark Fasheh return status; 419ccd979bdSMark Fasheh } 420ccd979bdSMark Fasheh 421ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 422ccd979bdSMark Fasheh int dirty) 423ccd979bdSMark Fasheh { 424ccd979bdSMark Fasheh int status; 425ccd979bdSMark Fasheh unsigned int flags; 426ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 427ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 428ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 429ccd979bdSMark Fasheh 430ccd979bdSMark Fasheh mlog_entry_void(); 431ccd979bdSMark Fasheh 432ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 433ccd979bdSMark Fasheh if (!OCFS2_IS_VALID_DINODE(fe)) { 434ccd979bdSMark Fasheh /* This is called from startup/shutdown which will 435ccd979bdSMark Fasheh * handle the errors in a specific manner, so no need 436ccd979bdSMark Fasheh * to call ocfs2_error() here. */ 437b0697053SMark Fasheh mlog(ML_ERROR, "Journal dinode %llu has invalid " 438b0697053SMark Fasheh "signature: %.*s", (unsigned long long)fe->i_blkno, 7, 439b0697053SMark Fasheh fe->i_signature); 440ccd979bdSMark Fasheh status = -EIO; 441ccd979bdSMark Fasheh goto out; 442ccd979bdSMark Fasheh } 443ccd979bdSMark Fasheh 444ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 445ccd979bdSMark Fasheh if (dirty) 446ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 447ccd979bdSMark Fasheh else 448ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 449ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 450ccd979bdSMark Fasheh 451ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, journal->j_inode); 452ccd979bdSMark Fasheh if (status < 0) 453ccd979bdSMark Fasheh mlog_errno(status); 454ccd979bdSMark Fasheh 455ccd979bdSMark Fasheh out: 456ccd979bdSMark Fasheh mlog_exit(status); 457ccd979bdSMark Fasheh return status; 458ccd979bdSMark Fasheh } 459ccd979bdSMark Fasheh 460ccd979bdSMark Fasheh /* 461ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 462ccd979bdSMark Fasheh * call. 463ccd979bdSMark Fasheh */ 464ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 465ccd979bdSMark Fasheh { 466ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 467ccd979bdSMark Fasheh int status = 0; 468ccd979bdSMark Fasheh struct inode *inode = NULL; 469ccd979bdSMark Fasheh int num_running_trans = 0; 470ccd979bdSMark Fasheh 471ccd979bdSMark Fasheh mlog_entry_void(); 472ccd979bdSMark Fasheh 473ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 474ccd979bdSMark Fasheh 475ccd979bdSMark Fasheh journal = osb->journal; 476ccd979bdSMark Fasheh if (!journal) 477ccd979bdSMark Fasheh goto done; 478ccd979bdSMark Fasheh 479ccd979bdSMark Fasheh inode = journal->j_inode; 480ccd979bdSMark Fasheh 481ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 482ccd979bdSMark Fasheh goto done; 483ccd979bdSMark Fasheh 484ccd979bdSMark Fasheh /* need to inc inode use count as journal_destroy will iput. */ 485ccd979bdSMark Fasheh if (!igrab(inode)) 486ccd979bdSMark Fasheh BUG(); 487ccd979bdSMark Fasheh 488ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 489ccd979bdSMark Fasheh if (num_running_trans > 0) 490ccd979bdSMark Fasheh mlog(0, "Shutting down journal: must wait on %d " 491ccd979bdSMark Fasheh "running transactions!\n", 492ccd979bdSMark Fasheh num_running_trans); 493ccd979bdSMark Fasheh 494ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 495ccd979bdSMark Fasheh * release any locks that are still held. 496ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 497ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 498ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 499ccd979bdSMark Fasheh 500ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 501ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 502ccd979bdSMark Fasheh * completely destroy the journal. */ 503ccd979bdSMark Fasheh if (osb->commit_task) { 504ccd979bdSMark Fasheh /* Wait for the commit thread */ 505ccd979bdSMark Fasheh mlog(0, "Waiting for ocfs2commit to exit....\n"); 506ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 507ccd979bdSMark Fasheh osb->commit_task = NULL; 508ccd979bdSMark Fasheh } 509ccd979bdSMark Fasheh 510ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 511ccd979bdSMark Fasheh 512c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 513c271c5c2SSunil Mushran journal_lock_updates(journal->j_journal); 514c271c5c2SSunil Mushran status = journal_flush(journal->j_journal); 515c271c5c2SSunil Mushran journal_unlock_updates(journal->j_journal); 516c271c5c2SSunil Mushran if (status < 0) 517c271c5c2SSunil Mushran mlog_errno(status); 518c271c5c2SSunil Mushran } 519c271c5c2SSunil Mushran 520c271c5c2SSunil Mushran if (status == 0) { 521c271c5c2SSunil Mushran /* 522c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 523c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 524c271c5c2SSunil Mushran */ 525ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 0); 526ccd979bdSMark Fasheh if (status < 0) 527ccd979bdSMark Fasheh mlog_errno(status); 528c271c5c2SSunil Mushran } 529ccd979bdSMark Fasheh 530ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 531ccd979bdSMark Fasheh journal_destroy(journal->j_journal); 532ccd979bdSMark Fasheh 533ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 534ccd979bdSMark Fasheh 535ccd979bdSMark Fasheh /* unlock our journal */ 536ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 537ccd979bdSMark Fasheh 538ccd979bdSMark Fasheh brelse(journal->j_bh); 539ccd979bdSMark Fasheh journal->j_bh = NULL; 540ccd979bdSMark Fasheh 541ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 542ccd979bdSMark Fasheh 543ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 544ccd979bdSMark Fasheh done: 545ccd979bdSMark Fasheh if (inode) 546ccd979bdSMark Fasheh iput(inode); 547ccd979bdSMark Fasheh mlog_exit_void(); 548ccd979bdSMark Fasheh } 549ccd979bdSMark Fasheh 550ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 551ccd979bdSMark Fasheh journal_t *journal, 552ccd979bdSMark Fasheh int slot) 553ccd979bdSMark Fasheh { 554ccd979bdSMark Fasheh int olderr; 555ccd979bdSMark Fasheh 556ccd979bdSMark Fasheh olderr = journal_errno(journal); 557ccd979bdSMark Fasheh if (olderr) { 558ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 559ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 560ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 561ccd979bdSMark Fasheh sb->s_id); 562ccd979bdSMark Fasheh 563ccd979bdSMark Fasheh journal_ack_err(journal); 564ccd979bdSMark Fasheh journal_clear_err(journal); 565ccd979bdSMark Fasheh } 566ccd979bdSMark Fasheh } 567ccd979bdSMark Fasheh 568c271c5c2SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local) 569ccd979bdSMark Fasheh { 570ccd979bdSMark Fasheh int status = 0; 571ccd979bdSMark Fasheh struct ocfs2_super *osb; 572ccd979bdSMark Fasheh 573ccd979bdSMark Fasheh mlog_entry_void(); 574ccd979bdSMark Fasheh 575ccd979bdSMark Fasheh if (!journal) 576ccd979bdSMark Fasheh BUG(); 577ccd979bdSMark Fasheh 578ccd979bdSMark Fasheh osb = journal->j_osb; 579ccd979bdSMark Fasheh 580ccd979bdSMark Fasheh status = journal_load(journal->j_journal); 581ccd979bdSMark Fasheh if (status < 0) { 582ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 583ccd979bdSMark Fasheh goto done; 584ccd979bdSMark Fasheh } 585ccd979bdSMark Fasheh 586ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 587ccd979bdSMark Fasheh 588ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 1); 589ccd979bdSMark Fasheh if (status < 0) { 590ccd979bdSMark Fasheh mlog_errno(status); 591ccd979bdSMark Fasheh goto done; 592ccd979bdSMark Fasheh } 593ccd979bdSMark Fasheh 594ccd979bdSMark Fasheh /* Launch the commit thread */ 595c271c5c2SSunil Mushran if (!local) { 596c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 597c271c5c2SSunil Mushran "ocfs2cmt"); 598ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 599ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 600ccd979bdSMark Fasheh osb->commit_task = NULL; 601c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 602c271c5c2SSunil Mushran "error=%d", status); 603ccd979bdSMark Fasheh goto done; 604ccd979bdSMark Fasheh } 605c271c5c2SSunil Mushran } else 606c271c5c2SSunil Mushran osb->commit_task = NULL; 607ccd979bdSMark Fasheh 608ccd979bdSMark Fasheh done: 609ccd979bdSMark Fasheh mlog_exit(status); 610ccd979bdSMark Fasheh return status; 611ccd979bdSMark Fasheh } 612ccd979bdSMark Fasheh 613ccd979bdSMark Fasheh 614ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 615ccd979bdSMark Fasheh * mark the journal clean */ 616ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 617ccd979bdSMark Fasheh { 618ccd979bdSMark Fasheh int status; 619ccd979bdSMark Fasheh 620ccd979bdSMark Fasheh mlog_entry_void(); 621ccd979bdSMark Fasheh 622ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 623ccd979bdSMark Fasheh 624ccd979bdSMark Fasheh status = journal_wipe(journal->j_journal, full); 625ccd979bdSMark Fasheh if (status < 0) { 626ccd979bdSMark Fasheh mlog_errno(status); 627ccd979bdSMark Fasheh goto bail; 628ccd979bdSMark Fasheh } 629ccd979bdSMark Fasheh 630ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); 631ccd979bdSMark Fasheh if (status < 0) 632ccd979bdSMark Fasheh mlog_errno(status); 633ccd979bdSMark Fasheh 634ccd979bdSMark Fasheh bail: 635ccd979bdSMark Fasheh mlog_exit(status); 636ccd979bdSMark Fasheh return status; 637ccd979bdSMark Fasheh } 638ccd979bdSMark Fasheh 639ccd979bdSMark Fasheh /* 640ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 641ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 642ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 643ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 644ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 645ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 646ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 647ccd979bdSMark Fasheh * as we have the lock by now :) 648ccd979bdSMark Fasheh */ 649ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 650ccd979bdSMark Fasheh { 651ccd979bdSMark Fasheh int status = 0; 652ccd979bdSMark Fasheh int i, p_blocks; 653ccd979bdSMark Fasheh u64 v_blkno, p_blkno; 654ccd979bdSMark Fasheh #define CONCURRENT_JOURNAL_FILL 32 655ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 656ccd979bdSMark Fasheh 657ccd979bdSMark Fasheh mlog_entry_void(); 658ccd979bdSMark Fasheh 659ccd979bdSMark Fasheh BUG_ON(inode->i_blocks != 660ccd979bdSMark Fasheh ocfs2_align_bytes_to_sectors(i_size_read(inode))); 661ccd979bdSMark Fasheh 662ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 663ccd979bdSMark Fasheh 6645515eff8SAndrew Morton mlog(0, "Force reading %llu blocks\n", 6655515eff8SAndrew Morton (unsigned long long)(inode->i_blocks >> 6665515eff8SAndrew Morton (inode->i_sb->s_blocksize_bits - 9))); 667ccd979bdSMark Fasheh 668ccd979bdSMark Fasheh v_blkno = 0; 669ccd979bdSMark Fasheh while (v_blkno < 670ccd979bdSMark Fasheh (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) { 671ccd979bdSMark Fasheh 672ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 67349cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 674ccd979bdSMark Fasheh if (status < 0) { 675ccd979bdSMark Fasheh mlog_errno(status); 676ccd979bdSMark Fasheh goto bail; 677ccd979bdSMark Fasheh } 678ccd979bdSMark Fasheh 679ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 680ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 681ccd979bdSMark Fasheh 682dd4a2c2bSMark Fasheh /* We are reading journal data which should not 683dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 684ccd979bdSMark Fasheh status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 685ccd979bdSMark Fasheh p_blkno, p_blocks, bhs, 0, 686dd4a2c2bSMark Fasheh NULL); 687ccd979bdSMark Fasheh if (status < 0) { 688ccd979bdSMark Fasheh mlog_errno(status); 689ccd979bdSMark Fasheh goto bail; 690ccd979bdSMark Fasheh } 691ccd979bdSMark Fasheh 692ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 693ccd979bdSMark Fasheh brelse(bhs[i]); 694ccd979bdSMark Fasheh bhs[i] = NULL; 695ccd979bdSMark Fasheh } 696ccd979bdSMark Fasheh 697ccd979bdSMark Fasheh v_blkno += p_blocks; 698ccd979bdSMark Fasheh } 699ccd979bdSMark Fasheh 700ccd979bdSMark Fasheh bail: 701ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 702ccd979bdSMark Fasheh if (bhs[i]) 703ccd979bdSMark Fasheh brelse(bhs[i]); 704ccd979bdSMark Fasheh mlog_exit(status); 705ccd979bdSMark Fasheh return status; 706ccd979bdSMark Fasheh } 707ccd979bdSMark Fasheh 708ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 709ccd979bdSMark Fasheh struct list_head lri_list; 710ccd979bdSMark Fasheh int lri_slot; 711ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 712ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 713ccd979bdSMark Fasheh }; 714ccd979bdSMark Fasheh 715ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 716ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 717ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 718ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 719ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 720ccd979bdSMark Fasheh * recovering nodes into account. 721ccd979bdSMark Fasheh * 722ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 723ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 724ccd979bdSMark Fasheh */ 725c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 726ccd979bdSMark Fasheh { 727ccd979bdSMark Fasheh int ret; 728c4028958SDavid Howells struct ocfs2_journal *journal = 729c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 730c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 731ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 732ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 733ccd979bdSMark Fasheh struct list_head *p, *n; 734ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 735ccd979bdSMark Fasheh 736ccd979bdSMark Fasheh mlog_entry_void(); 737ccd979bdSMark Fasheh 738ccd979bdSMark Fasheh mlog(0, "completing recovery from keventd\n"); 739ccd979bdSMark Fasheh 740ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 741ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 742ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 743ccd979bdSMark Fasheh 744ccd979bdSMark Fasheh list_for_each_safe(p, n, &tmp_la_list) { 745ccd979bdSMark Fasheh item = list_entry(p, struct ocfs2_la_recovery_item, lri_list); 746ccd979bdSMark Fasheh list_del_init(&item->lri_list); 747ccd979bdSMark Fasheh 748ccd979bdSMark Fasheh mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 749ccd979bdSMark Fasheh 750ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 751ccd979bdSMark Fasheh if (la_dinode) { 752b0697053SMark Fasheh mlog(0, "Clean up local alloc %llu\n", 753b0697053SMark Fasheh (unsigned long long)la_dinode->i_blkno); 754ccd979bdSMark Fasheh 755ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 756ccd979bdSMark Fasheh la_dinode); 757ccd979bdSMark Fasheh if (ret < 0) 758ccd979bdSMark Fasheh mlog_errno(ret); 759ccd979bdSMark Fasheh 760ccd979bdSMark Fasheh kfree(la_dinode); 761ccd979bdSMark Fasheh } 762ccd979bdSMark Fasheh 763ccd979bdSMark Fasheh tl_dinode = item->lri_tl_dinode; 764ccd979bdSMark Fasheh if (tl_dinode) { 765b0697053SMark Fasheh mlog(0, "Clean up truncate log %llu\n", 766b0697053SMark Fasheh (unsigned long long)tl_dinode->i_blkno); 767ccd979bdSMark Fasheh 768ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 769ccd979bdSMark Fasheh tl_dinode); 770ccd979bdSMark Fasheh if (ret < 0) 771ccd979bdSMark Fasheh mlog_errno(ret); 772ccd979bdSMark Fasheh 773ccd979bdSMark Fasheh kfree(tl_dinode); 774ccd979bdSMark Fasheh } 775ccd979bdSMark Fasheh 776ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 777ccd979bdSMark Fasheh if (ret < 0) 778ccd979bdSMark Fasheh mlog_errno(ret); 779ccd979bdSMark Fasheh 780ccd979bdSMark Fasheh kfree(item); 781ccd979bdSMark Fasheh } 782ccd979bdSMark Fasheh 783ccd979bdSMark Fasheh mlog(0, "Recovery completion\n"); 784ccd979bdSMark Fasheh mlog_exit_void(); 785ccd979bdSMark Fasheh } 786ccd979bdSMark Fasheh 787ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 788ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 789ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 790ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 791ccd979bdSMark Fasheh int slot_num, 792ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 793ccd979bdSMark Fasheh struct ocfs2_dinode *tl_dinode) 794ccd979bdSMark Fasheh { 795ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 796ccd979bdSMark Fasheh 797afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 798ccd979bdSMark Fasheh if (!item) { 799ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 800ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 801ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 802ccd979bdSMark Fasheh if (la_dinode) 803ccd979bdSMark Fasheh kfree(la_dinode); 804ccd979bdSMark Fasheh 805ccd979bdSMark Fasheh if (tl_dinode) 806ccd979bdSMark Fasheh kfree(tl_dinode); 807ccd979bdSMark Fasheh 808ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 809ccd979bdSMark Fasheh return; 810ccd979bdSMark Fasheh } 811ccd979bdSMark Fasheh 812ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 813ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 814ccd979bdSMark Fasheh item->lri_slot = slot_num; 815ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 816ccd979bdSMark Fasheh 817ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 818ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 819ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 820ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 821ccd979bdSMark Fasheh } 822ccd979bdSMark Fasheh 823ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 824ccd979bdSMark Fasheh * recovery for it's own slot. */ 825ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 826ccd979bdSMark Fasheh { 827ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 828ccd979bdSMark Fasheh 829ccd979bdSMark Fasheh if (osb->dirty) { 830ccd979bdSMark Fasheh /* No need to queue up our truncate_log as regular 831ccd979bdSMark Fasheh * cleanup will catch that. */ 832ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(journal, 833ccd979bdSMark Fasheh osb->slot_num, 834ccd979bdSMark Fasheh osb->local_alloc_copy, 835ccd979bdSMark Fasheh NULL); 836ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 837ccd979bdSMark Fasheh 838ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 839ccd979bdSMark Fasheh osb->dirty = 0; 840ccd979bdSMark Fasheh } 841ccd979bdSMark Fasheh } 842ccd979bdSMark Fasheh 843ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 844ccd979bdSMark Fasheh { 845ccd979bdSMark Fasheh int status, node_num; 846ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 847ccd979bdSMark Fasheh 848ccd979bdSMark Fasheh mlog_entry_void(); 849ccd979bdSMark Fasheh 850ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 851ccd979bdSMark Fasheh if (status < 0) { 852ccd979bdSMark Fasheh goto bail; 853ccd979bdSMark Fasheh } 854ccd979bdSMark Fasheh 855ccd979bdSMark Fasheh restart: 856ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 857ccd979bdSMark Fasheh if (status < 0) { 858ccd979bdSMark Fasheh mlog_errno(status); 859ccd979bdSMark Fasheh goto bail; 860ccd979bdSMark Fasheh } 861ccd979bdSMark Fasheh 862ccd979bdSMark Fasheh while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 863ccd979bdSMark Fasheh node_num = ocfs2_node_map_first_set_bit(osb, 864ccd979bdSMark Fasheh &osb->recovery_map); 865ccd979bdSMark Fasheh if (node_num == O2NM_INVALID_NODE_NUM) { 866ccd979bdSMark Fasheh mlog(0, "Out of nodes to recover.\n"); 867ccd979bdSMark Fasheh break; 868ccd979bdSMark Fasheh } 869ccd979bdSMark Fasheh 870ccd979bdSMark Fasheh status = ocfs2_recover_node(osb, node_num); 871ccd979bdSMark Fasheh if (status < 0) { 872ccd979bdSMark Fasheh mlog(ML_ERROR, 873ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 874ccd979bdSMark Fasheh status, node_num, 875ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 876ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 877ccd979bdSMark Fasheh continue; 878ccd979bdSMark Fasheh } 879ccd979bdSMark Fasheh 880ccd979bdSMark Fasheh ocfs2_recovery_map_clear(osb, node_num); 881ccd979bdSMark Fasheh } 882ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 883ccd979bdSMark Fasheh 884ccd979bdSMark Fasheh /* We always run recovery on our own orphan dir - the dead 885ccd979bdSMark Fasheh * node(s) may have voted "no" on an inode delete earlier. A 886ccd979bdSMark Fasheh * revote is therefore required. */ 887ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 888ccd979bdSMark Fasheh NULL); 889ccd979bdSMark Fasheh 890ccd979bdSMark Fasheh bail: 891c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 892ccd979bdSMark Fasheh if (!status && 893ccd979bdSMark Fasheh !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 894c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 895ccd979bdSMark Fasheh goto restart; 896ccd979bdSMark Fasheh } 897ccd979bdSMark Fasheh 898ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 899ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 900ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 901ccd979bdSMark Fasheh 902c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 903ccd979bdSMark Fasheh 904ccd979bdSMark Fasheh mlog_exit(status); 905ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 906ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 907ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 908ccd979bdSMark Fasheh complete_and_exit(NULL, status); 909ccd979bdSMark Fasheh return status; 910ccd979bdSMark Fasheh } 911ccd979bdSMark Fasheh 912ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 913ccd979bdSMark Fasheh { 914ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 915ccd979bdSMark Fasheh node_num, osb->node_num); 916ccd979bdSMark Fasheh 917c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 918ccd979bdSMark Fasheh if (osb->disable_recovery) 919ccd979bdSMark Fasheh goto out; 920ccd979bdSMark Fasheh 921ccd979bdSMark Fasheh /* People waiting on recovery will wait on 922ccd979bdSMark Fasheh * the recovery map to empty. */ 923ccd979bdSMark Fasheh if (!ocfs2_recovery_map_set(osb, node_num)) 924ccd979bdSMark Fasheh mlog(0, "node %d already be in recovery.\n", node_num); 925ccd979bdSMark Fasheh 926ccd979bdSMark Fasheh mlog(0, "starting recovery thread...\n"); 927ccd979bdSMark Fasheh 928ccd979bdSMark Fasheh if (osb->recovery_thread_task) 929ccd979bdSMark Fasheh goto out; 930ccd979bdSMark Fasheh 931ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 93278427043SMark Fasheh "ocfs2rec"); 933ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 934ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 935ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 936ccd979bdSMark Fasheh } 937ccd979bdSMark Fasheh 938ccd979bdSMark Fasheh out: 939c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 940ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 941ccd979bdSMark Fasheh 942ccd979bdSMark Fasheh mlog_exit_void(); 943ccd979bdSMark Fasheh } 944ccd979bdSMark Fasheh 945ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 946ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 947ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 948ccd979bdSMark Fasheh int node_num, 949ccd979bdSMark Fasheh int slot_num) 950ccd979bdSMark Fasheh { 951ccd979bdSMark Fasheh int status; 952ccd979bdSMark Fasheh int got_lock = 0; 953ccd979bdSMark Fasheh unsigned int flags; 954ccd979bdSMark Fasheh struct inode *inode = NULL; 955ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 956ccd979bdSMark Fasheh journal_t *journal = NULL; 957ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 958ccd979bdSMark Fasheh 959ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 960ccd979bdSMark Fasheh slot_num); 961ccd979bdSMark Fasheh if (inode == NULL) { 962ccd979bdSMark Fasheh status = -EACCES; 963ccd979bdSMark Fasheh mlog_errno(status); 964ccd979bdSMark Fasheh goto done; 965ccd979bdSMark Fasheh } 966ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 967ccd979bdSMark Fasheh status = -EACCES; 968ccd979bdSMark Fasheh iput(inode); 969ccd979bdSMark Fasheh inode = NULL; 970ccd979bdSMark Fasheh mlog_errno(status); 971ccd979bdSMark Fasheh goto done; 972ccd979bdSMark Fasheh } 973ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 974ccd979bdSMark Fasheh 9754bcec184SMark Fasheh status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 976ccd979bdSMark Fasheh if (status < 0) { 977ccd979bdSMark Fasheh mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); 978ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 979ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 980ccd979bdSMark Fasheh goto done; 981ccd979bdSMark Fasheh } 982ccd979bdSMark Fasheh got_lock = 1; 983ccd979bdSMark Fasheh 984ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 985ccd979bdSMark Fasheh 986ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 987ccd979bdSMark Fasheh 988ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 989ccd979bdSMark Fasheh mlog(0, "No recovery required for node %d\n", node_num); 990ccd979bdSMark Fasheh goto done; 991ccd979bdSMark Fasheh } 992ccd979bdSMark Fasheh 993ccd979bdSMark Fasheh mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 994ccd979bdSMark Fasheh node_num, slot_num, 995ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 996ccd979bdSMark Fasheh 997ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 998ccd979bdSMark Fasheh 999ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1000ccd979bdSMark Fasheh if (status < 0) { 1001ccd979bdSMark Fasheh mlog_errno(status); 1002ccd979bdSMark Fasheh goto done; 1003ccd979bdSMark Fasheh } 1004ccd979bdSMark Fasheh 1005ccd979bdSMark Fasheh mlog(0, "calling journal_init_inode\n"); 1006ccd979bdSMark Fasheh journal = journal_init_inode(inode); 1007ccd979bdSMark Fasheh if (journal == NULL) { 1008ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1009ccd979bdSMark Fasheh status = -EIO; 1010ccd979bdSMark Fasheh goto done; 1011ccd979bdSMark Fasheh } 1012ccd979bdSMark Fasheh 1013ccd979bdSMark Fasheh status = journal_load(journal); 1014ccd979bdSMark Fasheh if (status < 0) { 1015ccd979bdSMark Fasheh mlog_errno(status); 1016ccd979bdSMark Fasheh if (!igrab(inode)) 1017ccd979bdSMark Fasheh BUG(); 1018ccd979bdSMark Fasheh journal_destroy(journal); 1019ccd979bdSMark Fasheh goto done; 1020ccd979bdSMark Fasheh } 1021ccd979bdSMark Fasheh 1022ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1023ccd979bdSMark Fasheh 1024ccd979bdSMark Fasheh /* wipe the journal */ 1025ccd979bdSMark Fasheh mlog(0, "flushing the journal.\n"); 1026ccd979bdSMark Fasheh journal_lock_updates(journal); 1027ccd979bdSMark Fasheh status = journal_flush(journal); 1028ccd979bdSMark Fasheh journal_unlock_updates(journal); 1029ccd979bdSMark Fasheh if (status < 0) 1030ccd979bdSMark Fasheh mlog_errno(status); 1031ccd979bdSMark Fasheh 1032ccd979bdSMark Fasheh /* This will mark the node clean */ 1033ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1034ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1035ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1036ccd979bdSMark Fasheh 1037ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, inode); 1038ccd979bdSMark Fasheh if (status < 0) 1039ccd979bdSMark Fasheh mlog_errno(status); 1040ccd979bdSMark Fasheh 1041ccd979bdSMark Fasheh if (!igrab(inode)) 1042ccd979bdSMark Fasheh BUG(); 1043ccd979bdSMark Fasheh 1044ccd979bdSMark Fasheh journal_destroy(journal); 1045ccd979bdSMark Fasheh 1046ccd979bdSMark Fasheh done: 1047ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1048ccd979bdSMark Fasheh if (got_lock) 1049ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 1050ccd979bdSMark Fasheh 1051ccd979bdSMark Fasheh if (inode) 1052ccd979bdSMark Fasheh iput(inode); 1053ccd979bdSMark Fasheh 1054ccd979bdSMark Fasheh if (bh) 1055ccd979bdSMark Fasheh brelse(bh); 1056ccd979bdSMark Fasheh 1057ccd979bdSMark Fasheh mlog_exit(status); 1058ccd979bdSMark Fasheh return status; 1059ccd979bdSMark Fasheh } 1060ccd979bdSMark Fasheh 1061ccd979bdSMark Fasheh /* 1062ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1063ccd979bdSMark Fasheh * - Replay it's journal 1064ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1065ccd979bdSMark Fasheh * - Stamp a clean truncate log 1066ccd979bdSMark Fasheh * - Mark the node clean 1067ccd979bdSMark Fasheh * 1068ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1069ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1070ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1071ccd979bdSMark Fasheh * far less concerning. 1072ccd979bdSMark Fasheh */ 1073ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 1074ccd979bdSMark Fasheh int node_num) 1075ccd979bdSMark Fasheh { 1076ccd979bdSMark Fasheh int status = 0; 1077ccd979bdSMark Fasheh int slot_num; 1078ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1079ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1080ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1081ccd979bdSMark Fasheh 1082ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1083ccd979bdSMark Fasheh node_num, osb->node_num); 1084ccd979bdSMark Fasheh 1085ccd979bdSMark Fasheh mlog(0, "checking node %d\n", node_num); 1086ccd979bdSMark Fasheh 1087ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1088ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1089ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1090ccd979bdSMark Fasheh 1091ccd979bdSMark Fasheh slot_num = ocfs2_node_num_to_slot(si, node_num); 1092ccd979bdSMark Fasheh if (slot_num == OCFS2_INVALID_SLOT) { 1093ccd979bdSMark Fasheh status = 0; 1094ccd979bdSMark Fasheh mlog(0, "no slot for this node, so no recovery required.\n"); 1095ccd979bdSMark Fasheh goto done; 1096ccd979bdSMark Fasheh } 1097ccd979bdSMark Fasheh 1098ccd979bdSMark Fasheh mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1099ccd979bdSMark Fasheh 1100ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1101ccd979bdSMark Fasheh if (status < 0) { 1102ccd979bdSMark Fasheh mlog_errno(status); 1103ccd979bdSMark Fasheh goto done; 1104ccd979bdSMark Fasheh } 1105ccd979bdSMark Fasheh 1106ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1107ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1108ccd979bdSMark Fasheh if (status < 0) { 1109ccd979bdSMark Fasheh mlog_errno(status); 1110ccd979bdSMark Fasheh goto done; 1111ccd979bdSMark Fasheh } 1112ccd979bdSMark Fasheh 1113ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1114ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1115ccd979bdSMark Fasheh * recovery. */ 1116ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1117ccd979bdSMark Fasheh if (status < 0) 1118ccd979bdSMark Fasheh mlog_errno(status); 1119ccd979bdSMark Fasheh 1120ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1121ccd979bdSMark Fasheh * harmful place to get an error... */ 1122ccd979bdSMark Fasheh ocfs2_clear_slot(si, slot_num); 1123ccd979bdSMark Fasheh status = ocfs2_update_disk_slots(osb, si); 1124ccd979bdSMark Fasheh if (status < 0) 1125ccd979bdSMark Fasheh mlog_errno(status); 1126ccd979bdSMark Fasheh 1127ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1128ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1129ccd979bdSMark Fasheh tl_copy); 1130ccd979bdSMark Fasheh 1131ccd979bdSMark Fasheh status = 0; 1132ccd979bdSMark Fasheh done: 1133ccd979bdSMark Fasheh 1134ccd979bdSMark Fasheh mlog_exit(status); 1135ccd979bdSMark Fasheh return status; 1136ccd979bdSMark Fasheh } 1137ccd979bdSMark Fasheh 1138ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1139ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1140ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1141ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1142ccd979bdSMark Fasheh int slot_num) 1143ccd979bdSMark Fasheh { 1144ccd979bdSMark Fasheh int status, flags; 1145ccd979bdSMark Fasheh struct inode *inode = NULL; 1146ccd979bdSMark Fasheh 1147ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1148ccd979bdSMark Fasheh slot_num); 1149ccd979bdSMark Fasheh if (inode == NULL) { 1150ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1151ccd979bdSMark Fasheh status = -EACCES; 1152ccd979bdSMark Fasheh goto bail; 1153ccd979bdSMark Fasheh } 1154ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1155ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1156ccd979bdSMark Fasheh iput(inode); 1157ccd979bdSMark Fasheh inode = NULL; 1158ccd979bdSMark Fasheh status = -EACCES; 1159ccd979bdSMark Fasheh goto bail; 1160ccd979bdSMark Fasheh } 1161ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1162ccd979bdSMark Fasheh 1163ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 11644bcec184SMark Fasheh status = ocfs2_meta_lock_full(inode, NULL, 1, flags); 1165ccd979bdSMark Fasheh if (status < 0) { 1166ccd979bdSMark Fasheh if (status != -EAGAIN) 1167ccd979bdSMark Fasheh mlog_errno(status); 1168ccd979bdSMark Fasheh goto bail; 1169ccd979bdSMark Fasheh } 1170ccd979bdSMark Fasheh 1171ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 1172ccd979bdSMark Fasheh bail: 1173ccd979bdSMark Fasheh if (inode) 1174ccd979bdSMark Fasheh iput(inode); 1175ccd979bdSMark Fasheh 1176ccd979bdSMark Fasheh return status; 1177ccd979bdSMark Fasheh } 1178ccd979bdSMark Fasheh 1179ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1180ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1181ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1182ccd979bdSMark Fasheh { 1183ccd979bdSMark Fasheh int status, i, node_num; 1184ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1185ccd979bdSMark Fasheh 1186ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1187ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1188ccd979bdSMark Fasheh 1189ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1190ccd979bdSMark Fasheh for(i = 0; i < si->si_num_slots; i++) { 1191ccd979bdSMark Fasheh if (i == osb->slot_num) 1192ccd979bdSMark Fasheh continue; 1193ccd979bdSMark Fasheh if (ocfs2_is_empty_slot(si, i)) 1194ccd979bdSMark Fasheh continue; 1195ccd979bdSMark Fasheh 1196ccd979bdSMark Fasheh node_num = si->si_global_node_nums[i]; 1197ccd979bdSMark Fasheh if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) 1198ccd979bdSMark Fasheh continue; 1199ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1200ccd979bdSMark Fasheh 1201ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1202ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1203ccd979bdSMark Fasheh * file here to test if he's alive. */ 1204ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1205ccd979bdSMark Fasheh if (!status) { 1206ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1207ccd979bdSMark Fasheh * the recovery thread can't race us on 1208ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1209ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1210ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1211ccd979bdSMark Fasheh mlog_errno(status); 1212ccd979bdSMark Fasheh goto bail; 1213ccd979bdSMark Fasheh } 1214ccd979bdSMark Fasheh 1215ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1216ccd979bdSMark Fasheh } 1217ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1218ccd979bdSMark Fasheh 1219ccd979bdSMark Fasheh status = 0; 1220ccd979bdSMark Fasheh bail: 1221ccd979bdSMark Fasheh mlog_exit(status); 1222ccd979bdSMark Fasheh return status; 1223ccd979bdSMark Fasheh } 1224ccd979bdSMark Fasheh 1225b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1226b4df6ed8SMark Fasheh int slot, 1227b4df6ed8SMark Fasheh struct inode **head) 1228ccd979bdSMark Fasheh { 1229b4df6ed8SMark Fasheh int status; 1230ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 1231b4df6ed8SMark Fasheh struct inode *iter; 1232ccd979bdSMark Fasheh unsigned long offset, blk, local; 1233ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1234ccd979bdSMark Fasheh struct ocfs2_dir_entry *de; 1235ccd979bdSMark Fasheh struct super_block *sb = osb->sb; 1236ccd979bdSMark Fasheh 1237ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1238ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 1239ccd979bdSMark Fasheh slot); 1240ccd979bdSMark Fasheh if (!orphan_dir_inode) { 1241ccd979bdSMark Fasheh status = -ENOENT; 1242ccd979bdSMark Fasheh mlog_errno(status); 1243b4df6ed8SMark Fasheh return status; 1244ccd979bdSMark Fasheh } 1245ccd979bdSMark Fasheh 12461b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 12474bcec184SMark Fasheh status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0); 1248ccd979bdSMark Fasheh if (status < 0) { 1249ccd979bdSMark Fasheh mlog_errno(status); 1250ccd979bdSMark Fasheh goto out; 1251ccd979bdSMark Fasheh } 1252ccd979bdSMark Fasheh 1253ccd979bdSMark Fasheh offset = 0; 1254ccd979bdSMark Fasheh iter = NULL; 1255ccd979bdSMark Fasheh while(offset < i_size_read(orphan_dir_inode)) { 1256ccd979bdSMark Fasheh blk = offset >> sb->s_blocksize_bits; 1257ccd979bdSMark Fasheh 1258ccd979bdSMark Fasheh bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0); 1259ccd979bdSMark Fasheh if (!bh) 1260ccd979bdSMark Fasheh status = -EINVAL; 1261ccd979bdSMark Fasheh if (status < 0) { 1262ccd979bdSMark Fasheh if (bh) 1263ccd979bdSMark Fasheh brelse(bh); 1264ccd979bdSMark Fasheh mlog_errno(status); 1265b4df6ed8SMark Fasheh goto out_unlock; 1266ccd979bdSMark Fasheh } 1267ccd979bdSMark Fasheh 1268ccd979bdSMark Fasheh local = 0; 1269ccd979bdSMark Fasheh while(offset < i_size_read(orphan_dir_inode) 1270ccd979bdSMark Fasheh && local < sb->s_blocksize) { 1271ccd979bdSMark Fasheh de = (struct ocfs2_dir_entry *) (bh->b_data + local); 1272ccd979bdSMark Fasheh 1273ccd979bdSMark Fasheh if (!ocfs2_check_dir_entry(orphan_dir_inode, 1274ccd979bdSMark Fasheh de, bh, local)) { 1275ccd979bdSMark Fasheh status = -EINVAL; 1276ccd979bdSMark Fasheh mlog_errno(status); 1277ccd979bdSMark Fasheh brelse(bh); 1278b4df6ed8SMark Fasheh goto out_unlock; 1279ccd979bdSMark Fasheh } 1280ccd979bdSMark Fasheh 1281ccd979bdSMark Fasheh local += le16_to_cpu(de->rec_len); 1282ccd979bdSMark Fasheh offset += le16_to_cpu(de->rec_len); 1283ccd979bdSMark Fasheh 1284ccd979bdSMark Fasheh /* I guess we silently fail on no inode? */ 1285ccd979bdSMark Fasheh if (!le64_to_cpu(de->inode)) 1286ccd979bdSMark Fasheh continue; 1287ccd979bdSMark Fasheh if (de->file_type > OCFS2_FT_MAX) { 1288ccd979bdSMark Fasheh mlog(ML_ERROR, 1289ccd979bdSMark Fasheh "block %llu contains invalid de: " 1290b0697053SMark Fasheh "inode = %llu, rec_len = %u, " 1291ccd979bdSMark Fasheh "name_len = %u, file_type = %u, " 1292ccd979bdSMark Fasheh "name='%.*s'\n", 1293ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, 1294b0697053SMark Fasheh (unsigned long long)le64_to_cpu(de->inode), 1295ccd979bdSMark Fasheh le16_to_cpu(de->rec_len), 1296ccd979bdSMark Fasheh de->name_len, 1297ccd979bdSMark Fasheh de->file_type, 1298ccd979bdSMark Fasheh de->name_len, 1299ccd979bdSMark Fasheh de->name); 1300ccd979bdSMark Fasheh continue; 1301ccd979bdSMark Fasheh } 1302ccd979bdSMark Fasheh if (de->name_len == 1 && !strncmp(".", de->name, 1)) 1303ccd979bdSMark Fasheh continue; 1304ccd979bdSMark Fasheh if (de->name_len == 2 && !strncmp("..", de->name, 2)) 1305ccd979bdSMark Fasheh continue; 1306ccd979bdSMark Fasheh 130724c19ef4SMark Fasheh iter = ocfs2_iget(osb, le64_to_cpu(de->inode), 130850008630STiger Yang OCFS2_FI_FLAG_ORPHAN_RECOVERY); 1309ccd979bdSMark Fasheh if (IS_ERR(iter)) 1310ccd979bdSMark Fasheh continue; 1311ccd979bdSMark Fasheh 1312b0697053SMark Fasheh mlog(0, "queue orphan %llu\n", 1313b0697053SMark Fasheh (unsigned long long)OCFS2_I(iter)->ip_blkno); 1314b4df6ed8SMark Fasheh /* No locking is required for the next_orphan 1315b4df6ed8SMark Fasheh * queue as there is only ever a single 1316b4df6ed8SMark Fasheh * process doing orphan recovery. */ 1317b4df6ed8SMark Fasheh OCFS2_I(iter)->ip_next_orphan = *head; 1318b4df6ed8SMark Fasheh *head = iter; 1319ccd979bdSMark Fasheh } 1320ccd979bdSMark Fasheh brelse(bh); 1321ccd979bdSMark Fasheh } 1322ccd979bdSMark Fasheh 1323b4df6ed8SMark Fasheh out_unlock: 1324ccd979bdSMark Fasheh ocfs2_meta_unlock(orphan_dir_inode, 0); 1325b4df6ed8SMark Fasheh out: 1326b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 1327ccd979bdSMark Fasheh iput(orphan_dir_inode); 1328b4df6ed8SMark Fasheh return status; 1329b4df6ed8SMark Fasheh } 1330b4df6ed8SMark Fasheh 1331b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1332b4df6ed8SMark Fasheh int slot) 1333b4df6ed8SMark Fasheh { 1334b4df6ed8SMark Fasheh int ret; 1335b4df6ed8SMark Fasheh 1336b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1337b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 1338b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1339b4df6ed8SMark Fasheh return ret; 1340b4df6ed8SMark Fasheh } 1341b4df6ed8SMark Fasheh 1342b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1343b4df6ed8SMark Fasheh int slot) 1344b4df6ed8SMark Fasheh { 1345b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1346b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 1347b4df6ed8SMark Fasheh * know to quit early. */ 1348b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1349b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 1350b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 1351b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 1352b4df6ed8SMark Fasheh * them. */ 1353b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1354b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 1355b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 1356b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1357b4df6ed8SMark Fasheh } 1358b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1359b4df6ed8SMark Fasheh } 1360b4df6ed8SMark Fasheh 1361b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1362b4df6ed8SMark Fasheh int slot) 1363b4df6ed8SMark Fasheh { 1364b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1365b4df6ed8SMark Fasheh } 1366b4df6ed8SMark Fasheh 1367b4df6ed8SMark Fasheh /* 1368b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 1369b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 1370b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 1371b4df6ed8SMark Fasheh * (most) of the rest of the work. 1372b4df6ed8SMark Fasheh * 1373b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 1374b4df6ed8SMark Fasheh * couple of extra considerations. 1375b4df6ed8SMark Fasheh * 1376b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 1377b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 1378b4df6ed8SMark Fasheh * an invalid inode. 1379b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 1380b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 1381b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 1382b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 1383b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 1384b4df6ed8SMark Fasheh */ 1385b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1386b4df6ed8SMark Fasheh int slot) 1387b4df6ed8SMark Fasheh { 1388b4df6ed8SMark Fasheh int ret = 0; 1389b4df6ed8SMark Fasheh struct inode *inode = NULL; 1390b4df6ed8SMark Fasheh struct inode *iter; 1391b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 1392b4df6ed8SMark Fasheh 1393b4df6ed8SMark Fasheh mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1394b4df6ed8SMark Fasheh 1395b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 1396b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 1397b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 1398b4df6ed8SMark Fasheh 1399b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 1400b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 1401b4df6ed8SMark Fasheh if (ret) 1402b4df6ed8SMark Fasheh mlog_errno(ret); 1403ccd979bdSMark Fasheh 1404ccd979bdSMark Fasheh while (inode) { 1405ccd979bdSMark Fasheh oi = OCFS2_I(inode); 1406b0697053SMark Fasheh mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1407ccd979bdSMark Fasheh 1408ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 1409ccd979bdSMark Fasheh 1410ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 1411ccd979bdSMark Fasheh /* Delete voting may have set these on the assumption 1412ccd979bdSMark Fasheh * that the other node would wipe them successfully. 1413ccd979bdSMark Fasheh * If they are still in the node's orphan dir, we need 1414ccd979bdSMark Fasheh * to reset that state. */ 1415ccd979bdSMark Fasheh oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1416ccd979bdSMark Fasheh 1417ccd979bdSMark Fasheh /* Set the proper information to get us going into 1418ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 1419ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1420ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 1421ccd979bdSMark Fasheh 1422ccd979bdSMark Fasheh iput(inode); 1423ccd979bdSMark Fasheh 1424ccd979bdSMark Fasheh inode = iter; 1425ccd979bdSMark Fasheh } 1426ccd979bdSMark Fasheh 1427b4df6ed8SMark Fasheh return ret; 1428ccd979bdSMark Fasheh } 1429ccd979bdSMark Fasheh 1430ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1431ccd979bdSMark Fasheh { 1432ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 1433ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 1434ccd979bdSMark Fasheh * or DISABLED. */ 1435ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 1436ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1437ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1438ccd979bdSMark Fasheh 1439ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 1440ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 1441ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 1442ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1443ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 1444ccd979bdSMark Fasheh return -EBUSY; 1445ccd979bdSMark Fasheh } 1446ccd979bdSMark Fasheh 1447ccd979bdSMark Fasheh return 0; 1448ccd979bdSMark Fasheh } 1449ccd979bdSMark Fasheh 1450ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 1451ccd979bdSMark Fasheh { 1452ccd979bdSMark Fasheh int status; 1453ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1454ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1455ccd979bdSMark Fasheh 1456ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 1457ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 1458ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 1459ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 1460ccd979bdSMark Fasheh while (!(kthread_should_stop() && 1461ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 1462ccd979bdSMark Fasheh 1463745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 1464ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 1465745ae8baSMark Fasheh || kthread_should_stop()); 1466ccd979bdSMark Fasheh 1467ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 1468ccd979bdSMark Fasheh if (status < 0) 1469ccd979bdSMark Fasheh mlog_errno(status); 1470ccd979bdSMark Fasheh 1471ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1472ccd979bdSMark Fasheh mlog(ML_KTHREAD, 1473ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 1474ccd979bdSMark Fasheh "shutdown\n", 1475ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 1476ccd979bdSMark Fasheh } 1477ccd979bdSMark Fasheh } 1478ccd979bdSMark Fasheh 1479ccd979bdSMark Fasheh return 0; 1480ccd979bdSMark Fasheh } 1481ccd979bdSMark Fasheh 1482ccd979bdSMark Fasheh /* Look for a dirty journal without taking any cluster locks. Used for 1483ccd979bdSMark Fasheh * hard readonly access to determine whether the file system journals 1484ccd979bdSMark Fasheh * require recovery. */ 1485ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1486ccd979bdSMark Fasheh { 1487ccd979bdSMark Fasheh int ret = 0; 1488ccd979bdSMark Fasheh unsigned int slot; 1489ccd979bdSMark Fasheh struct buffer_head *di_bh; 1490ccd979bdSMark Fasheh struct ocfs2_dinode *di; 1491ccd979bdSMark Fasheh struct inode *journal = NULL; 1492ccd979bdSMark Fasheh 1493ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 1494ccd979bdSMark Fasheh journal = ocfs2_get_system_file_inode(osb, 1495ccd979bdSMark Fasheh JOURNAL_SYSTEM_INODE, 1496ccd979bdSMark Fasheh slot); 1497ccd979bdSMark Fasheh if (!journal || is_bad_inode(journal)) { 1498ccd979bdSMark Fasheh ret = -EACCES; 1499ccd979bdSMark Fasheh mlog_errno(ret); 1500ccd979bdSMark Fasheh goto out; 1501ccd979bdSMark Fasheh } 1502ccd979bdSMark Fasheh 1503ccd979bdSMark Fasheh di_bh = NULL; 1504ccd979bdSMark Fasheh ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, 1505ccd979bdSMark Fasheh 0, journal); 1506ccd979bdSMark Fasheh if (ret < 0) { 1507ccd979bdSMark Fasheh mlog_errno(ret); 1508ccd979bdSMark Fasheh goto out; 1509ccd979bdSMark Fasheh } 1510ccd979bdSMark Fasheh 1511ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 1512ccd979bdSMark Fasheh 1513ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 1514ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 1515ccd979bdSMark Fasheh ret = -EROFS; 1516ccd979bdSMark Fasheh 1517ccd979bdSMark Fasheh brelse(di_bh); 1518ccd979bdSMark Fasheh if (ret) 1519ccd979bdSMark Fasheh break; 1520ccd979bdSMark Fasheh } 1521ccd979bdSMark Fasheh 1522ccd979bdSMark Fasheh out: 1523ccd979bdSMark Fasheh if (journal) 1524ccd979bdSMark Fasheh iput(journal); 1525ccd979bdSMark Fasheh 1526ccd979bdSMark Fasheh return ret; 1527ccd979bdSMark Fasheh } 1528