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, 60539d8264SSunil Mushran int dirty, int replayed); 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 67553abd04SJoel Becker 68553abd04SJoel Becker /* 69553abd04SJoel Becker * The recovery_list is a simple linked list of node numbers to recover. 70553abd04SJoel Becker * It is protected by the recovery_lock. 71553abd04SJoel Becker */ 72553abd04SJoel Becker 73553abd04SJoel Becker struct ocfs2_recovery_map { 74fc881fa0SJoel Becker unsigned int rm_used; 75553abd04SJoel Becker unsigned int *rm_entries; 76553abd04SJoel Becker }; 77553abd04SJoel Becker 78553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 79553abd04SJoel Becker { 80553abd04SJoel Becker struct ocfs2_recovery_map *rm; 81553abd04SJoel Becker 82553abd04SJoel Becker mutex_init(&osb->recovery_lock); 83553abd04SJoel Becker osb->disable_recovery = 0; 84553abd04SJoel Becker osb->recovery_thread_task = NULL; 85553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 86553abd04SJoel Becker 87553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 88553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 89553abd04SJoel Becker GFP_KERNEL); 90553abd04SJoel Becker if (!rm) { 91553abd04SJoel Becker mlog_errno(-ENOMEM); 92553abd04SJoel Becker return -ENOMEM; 93553abd04SJoel Becker } 94553abd04SJoel Becker 95553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 96553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 97553abd04SJoel Becker osb->recovery_map = rm; 98553abd04SJoel Becker 99553abd04SJoel Becker return 0; 100553abd04SJoel Becker } 101553abd04SJoel Becker 102553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 103553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 104553abd04SJoel Becker * being woken up */ 105553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 106553abd04SJoel Becker { 107553abd04SJoel Becker mb(); 108553abd04SJoel Becker return osb->recovery_thread_task != NULL; 109553abd04SJoel Becker } 110553abd04SJoel Becker 111553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 112553abd04SJoel Becker { 113553abd04SJoel Becker struct ocfs2_recovery_map *rm; 114553abd04SJoel Becker 115553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 116553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 117553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 118553abd04SJoel Becker osb->disable_recovery = 1; 119553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 120553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 121553abd04SJoel Becker 122553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 123553abd04SJoel Becker * launched, so wait for any recovery completion work to 124553abd04SJoel Becker * complete. */ 125553abd04SJoel Becker flush_workqueue(ocfs2_wq); 126553abd04SJoel Becker 127553abd04SJoel Becker /* 128553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 129553abd04SJoel Becker * freed, the osb_lock is not taken here. 130553abd04SJoel Becker */ 131553abd04SJoel Becker rm = osb->recovery_map; 132553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 133553abd04SJoel Becker 134553abd04SJoel Becker kfree(rm); 135553abd04SJoel Becker } 136553abd04SJoel Becker 137553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 138553abd04SJoel Becker unsigned int node_num) 139553abd04SJoel Becker { 140553abd04SJoel Becker int i; 141553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 142553abd04SJoel Becker 143553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 144553abd04SJoel Becker 145553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 146553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 147553abd04SJoel Becker return 1; 148553abd04SJoel Becker } 149553abd04SJoel Becker 150553abd04SJoel Becker return 0; 151553abd04SJoel Becker } 152553abd04SJoel Becker 153553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 154553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 155553abd04SJoel Becker unsigned int node_num) 156553abd04SJoel Becker { 157553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 158553abd04SJoel Becker 159553abd04SJoel Becker spin_lock(&osb->osb_lock); 160553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 161553abd04SJoel Becker spin_unlock(&osb->osb_lock); 162553abd04SJoel Becker return 1; 163553abd04SJoel Becker } 164553abd04SJoel Becker 165553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 166553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 167553abd04SJoel Becker 168553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 169553abd04SJoel Becker rm->rm_used++; 170553abd04SJoel Becker spin_unlock(&osb->osb_lock); 171553abd04SJoel Becker 172553abd04SJoel Becker return 0; 173553abd04SJoel Becker } 174553abd04SJoel Becker 175553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 176553abd04SJoel Becker unsigned int node_num) 177553abd04SJoel Becker { 178553abd04SJoel Becker int i; 179553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 180553abd04SJoel Becker 181553abd04SJoel Becker spin_lock(&osb->osb_lock); 182553abd04SJoel Becker 183553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 184553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 185553abd04SJoel Becker break; 186553abd04SJoel Becker } 187553abd04SJoel Becker 188553abd04SJoel Becker if (i < rm->rm_used) { 189553abd04SJoel Becker /* XXX: be careful with the pointer math */ 190553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 191553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 192553abd04SJoel Becker rm->rm_used--; 193553abd04SJoel Becker } 194553abd04SJoel Becker 195553abd04SJoel Becker spin_unlock(&osb->osb_lock); 196553abd04SJoel Becker } 197553abd04SJoel Becker 198ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 199ccd979bdSMark Fasheh { 200ccd979bdSMark Fasheh int status = 0; 201ccd979bdSMark Fasheh unsigned int flushed; 202ccd979bdSMark Fasheh unsigned long old_id; 203ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 204ccd979bdSMark Fasheh 205ccd979bdSMark Fasheh mlog_entry_void(); 206ccd979bdSMark Fasheh 207ccd979bdSMark Fasheh journal = osb->journal; 208ccd979bdSMark Fasheh 209ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 210ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 211ccd979bdSMark Fasheh 212ccd979bdSMark Fasheh if (atomic_read(&journal->j_num_trans) == 0) { 213ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 214ccd979bdSMark Fasheh mlog(0, "No transactions for me to flush!\n"); 215ccd979bdSMark Fasheh goto finally; 216ccd979bdSMark Fasheh } 217ccd979bdSMark Fasheh 218ccd979bdSMark Fasheh journal_lock_updates(journal->j_journal); 219ccd979bdSMark Fasheh status = journal_flush(journal->j_journal); 220ccd979bdSMark Fasheh journal_unlock_updates(journal->j_journal); 221ccd979bdSMark Fasheh if (status < 0) { 222ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 223ccd979bdSMark Fasheh mlog_errno(status); 224ccd979bdSMark Fasheh goto finally; 225ccd979bdSMark Fasheh } 226ccd979bdSMark Fasheh 227ccd979bdSMark Fasheh old_id = ocfs2_inc_trans_id(journal); 228ccd979bdSMark Fasheh 229ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 230ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 231ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 232ccd979bdSMark Fasheh 233ccd979bdSMark Fasheh mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 234ccd979bdSMark Fasheh journal->j_trans_id, flushed); 235ccd979bdSMark Fasheh 23634d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 237ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 238ccd979bdSMark Fasheh finally: 239ccd979bdSMark Fasheh mlog_exit(status); 240ccd979bdSMark Fasheh return status; 241ccd979bdSMark Fasheh } 242ccd979bdSMark Fasheh 243ccd979bdSMark Fasheh /* pass it NULL and it will allocate a new handle object for you. If 244ccd979bdSMark Fasheh * you pass it a handle however, it may still return error, in which 245ccd979bdSMark Fasheh * case it has free'd the passed handle for you. */ 2461fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 247ccd979bdSMark Fasheh { 248ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 2491fabe148SMark Fasheh handle_t *handle; 250ccd979bdSMark Fasheh 251ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 252ccd979bdSMark Fasheh 25365eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 25465eff9ccSMark Fasheh return ERR_PTR(-EROFS); 255ccd979bdSMark Fasheh 256ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 257ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 258ccd979bdSMark Fasheh 259ccd979bdSMark Fasheh /* JBD might support this, but our journalling code doesn't yet. */ 260ccd979bdSMark Fasheh if (journal_current_handle()) { 261ccd979bdSMark Fasheh mlog(ML_ERROR, "Recursive transaction attempted!\n"); 262ccd979bdSMark Fasheh BUG(); 263ccd979bdSMark Fasheh } 264ccd979bdSMark Fasheh 265ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 266ccd979bdSMark Fasheh 2671fabe148SMark Fasheh handle = journal_start(journal, max_buffs); 2681fabe148SMark Fasheh if (IS_ERR(handle)) { 269ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 270ccd979bdSMark Fasheh 2711fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 272ccd979bdSMark Fasheh 273ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 274ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 2751fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 276ccd979bdSMark Fasheh } 277c271c5c2SSunil Mushran } else { 278c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 279ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 280c271c5c2SSunil Mushran } 281ccd979bdSMark Fasheh 282ccd979bdSMark Fasheh return handle; 283ccd979bdSMark Fasheh } 284ccd979bdSMark Fasheh 2851fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 2861fabe148SMark Fasheh handle_t *handle) 287ccd979bdSMark Fasheh { 2881fabe148SMark Fasheh int ret; 28902dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 290ccd979bdSMark Fasheh 291ccd979bdSMark Fasheh BUG_ON(!handle); 292ccd979bdSMark Fasheh 2931fabe148SMark Fasheh ret = journal_stop(handle); 2941fabe148SMark Fasheh if (ret < 0) 2951fabe148SMark Fasheh mlog_errno(ret); 296ccd979bdSMark Fasheh 297ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 298ccd979bdSMark Fasheh 2991fabe148SMark Fasheh return ret; 300ccd979bdSMark Fasheh } 301ccd979bdSMark Fasheh 302ccd979bdSMark Fasheh /* 303ccd979bdSMark Fasheh * 'nblocks' is what you want to add to the current 304ccd979bdSMark Fasheh * transaction. extend_trans will either extend the current handle by 305ccd979bdSMark Fasheh * nblocks, or commit it and start a new one with nblocks credits. 306ccd979bdSMark Fasheh * 307e8aed345SMark Fasheh * This might call journal_restart() which will commit dirty buffers 308e8aed345SMark Fasheh * and then restart the transaction. Before calling 309e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 310e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 311e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 312e8aed345SMark Fasheh * 313ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 314ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 315ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 316ccd979bdSMark Fasheh * 317ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 318ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 319ccd979bdSMark Fasheh * cluster locks associated with this handle. 320ccd979bdSMark Fasheh */ 3211fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 322ccd979bdSMark Fasheh { 323ccd979bdSMark Fasheh int status; 324ccd979bdSMark Fasheh 325ccd979bdSMark Fasheh BUG_ON(!handle); 326ccd979bdSMark Fasheh BUG_ON(!nblocks); 327ccd979bdSMark Fasheh 328ccd979bdSMark Fasheh mlog_entry_void(); 329ccd979bdSMark Fasheh 330ccd979bdSMark Fasheh mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 331ccd979bdSMark Fasheh 332e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 3330879c584SMark Fasheh status = 1; 3340879c584SMark Fasheh #else 3351fc58146SMark Fasheh status = journal_extend(handle, nblocks); 336ccd979bdSMark Fasheh if (status < 0) { 337ccd979bdSMark Fasheh mlog_errno(status); 338ccd979bdSMark Fasheh goto bail; 339ccd979bdSMark Fasheh } 3400879c584SMark Fasheh #endif 341ccd979bdSMark Fasheh 342ccd979bdSMark Fasheh if (status > 0) { 343ccd979bdSMark Fasheh mlog(0, "journal_extend failed, trying journal_restart\n"); 3441fc58146SMark Fasheh status = journal_restart(handle, nblocks); 345ccd979bdSMark Fasheh if (status < 0) { 346ccd979bdSMark Fasheh mlog_errno(status); 347ccd979bdSMark Fasheh goto bail; 348ccd979bdSMark Fasheh } 34901ddf1e1SMark Fasheh } 350ccd979bdSMark Fasheh 351ccd979bdSMark Fasheh status = 0; 352ccd979bdSMark Fasheh bail: 353ccd979bdSMark Fasheh 354ccd979bdSMark Fasheh mlog_exit(status); 355ccd979bdSMark Fasheh return status; 356ccd979bdSMark Fasheh } 357ccd979bdSMark Fasheh 3581fabe148SMark Fasheh int ocfs2_journal_access(handle_t *handle, 359ccd979bdSMark Fasheh struct inode *inode, 360ccd979bdSMark Fasheh struct buffer_head *bh, 361ccd979bdSMark Fasheh int type) 362ccd979bdSMark Fasheh { 363ccd979bdSMark Fasheh int status; 364ccd979bdSMark Fasheh 365ccd979bdSMark Fasheh BUG_ON(!inode); 366ccd979bdSMark Fasheh BUG_ON(!handle); 367ccd979bdSMark Fasheh BUG_ON(!bh); 368ccd979bdSMark Fasheh 369205f87f6SBadari Pulavarty mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 370ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, type, 371ccd979bdSMark Fasheh (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 372ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_CREATE" : 373ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_WRITE", 374ccd979bdSMark Fasheh bh->b_size); 375ccd979bdSMark Fasheh 376ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 377ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 378ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 379ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 380ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 381ccd979bdSMark Fasheh BUG(); 382ccd979bdSMark Fasheh } 383ccd979bdSMark Fasheh 384ccd979bdSMark Fasheh /* Set the current transaction information on the inode so 385ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 386ccd979bdSMark Fasheh * on this inode or not. We're protected from the commit 387ccd979bdSMark Fasheh * thread updating the current transaction id until 388ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 389ccd979bdSMark Fasheh * j_trans_barrier for us. */ 390ccd979bdSMark Fasheh ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 391ccd979bdSMark Fasheh 392251b6eccSMark Fasheh mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 393ccd979bdSMark Fasheh switch (type) { 394ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 395ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 3961fabe148SMark Fasheh status = journal_get_write_access(handle, bh); 397ccd979bdSMark Fasheh break; 398ccd979bdSMark Fasheh 399ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 4001fabe148SMark Fasheh status = journal_get_undo_access(handle, bh); 401ccd979bdSMark Fasheh break; 402ccd979bdSMark Fasheh 403ccd979bdSMark Fasheh default: 404ccd979bdSMark Fasheh status = -EINVAL; 405ccd979bdSMark Fasheh mlog(ML_ERROR, "Uknown access type!\n"); 406ccd979bdSMark Fasheh } 407251b6eccSMark Fasheh mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 408ccd979bdSMark Fasheh 409ccd979bdSMark Fasheh if (status < 0) 410ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 411ccd979bdSMark Fasheh status, type); 412ccd979bdSMark Fasheh 413ccd979bdSMark Fasheh mlog_exit(status); 414ccd979bdSMark Fasheh return status; 415ccd979bdSMark Fasheh } 416ccd979bdSMark Fasheh 4171fabe148SMark Fasheh int ocfs2_journal_dirty(handle_t *handle, 418ccd979bdSMark Fasheh struct buffer_head *bh) 419ccd979bdSMark Fasheh { 420ccd979bdSMark Fasheh int status; 421ccd979bdSMark Fasheh 422ccd979bdSMark Fasheh mlog_entry("(bh->b_blocknr=%llu)\n", 423ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 424ccd979bdSMark Fasheh 4251fabe148SMark Fasheh status = journal_dirty_metadata(handle, bh); 426ccd979bdSMark Fasheh if (status < 0) 427ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not dirty metadata buffer. " 428ccd979bdSMark Fasheh "(bh->b_blocknr=%llu)\n", 429ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 430ccd979bdSMark Fasheh 431ccd979bdSMark Fasheh mlog_exit(status); 432ccd979bdSMark Fasheh return status; 433ccd979bdSMark Fasheh } 434ccd979bdSMark Fasheh 435ccd979bdSMark Fasheh int ocfs2_journal_dirty_data(handle_t *handle, 436ccd979bdSMark Fasheh struct buffer_head *bh) 437ccd979bdSMark Fasheh { 438ccd979bdSMark Fasheh int err = journal_dirty_data(handle, bh); 439ccd979bdSMark Fasheh if (err) 440ccd979bdSMark Fasheh mlog_errno(err); 441ccd979bdSMark Fasheh /* TODO: When we can handle it, abort the handle and go RO on 442ccd979bdSMark Fasheh * error here. */ 443ccd979bdSMark Fasheh 444ccd979bdSMark Fasheh return err; 445ccd979bdSMark Fasheh } 446ccd979bdSMark Fasheh 447d147b3d6SMark Fasheh #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD_DEFAULT_MAX_COMMIT_AGE) 448ccd979bdSMark Fasheh 449ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 450ccd979bdSMark Fasheh { 451ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 452d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 453d147b3d6SMark Fasheh 454d147b3d6SMark Fasheh if (osb->osb_commit_interval) 455d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 456ccd979bdSMark Fasheh 457ccd979bdSMark Fasheh spin_lock(&journal->j_state_lock); 458d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 459ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 460ccd979bdSMark Fasheh journal->j_flags |= JFS_BARRIER; 461ccd979bdSMark Fasheh else 462ccd979bdSMark Fasheh journal->j_flags &= ~JFS_BARRIER; 463ccd979bdSMark Fasheh spin_unlock(&journal->j_state_lock); 464ccd979bdSMark Fasheh } 465ccd979bdSMark Fasheh 466ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 467ccd979bdSMark Fasheh { 468ccd979bdSMark Fasheh int status = -1; 469ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 470ccd979bdSMark Fasheh journal_t *j_journal = NULL; 471ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 472ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 473ccd979bdSMark Fasheh struct ocfs2_super *osb; 474e63aecb6SMark Fasheh int inode_lock = 0; 475ccd979bdSMark Fasheh 476ccd979bdSMark Fasheh mlog_entry_void(); 477ccd979bdSMark Fasheh 478ccd979bdSMark Fasheh BUG_ON(!journal); 479ccd979bdSMark Fasheh 480ccd979bdSMark Fasheh osb = journal->j_osb; 481ccd979bdSMark Fasheh 482ccd979bdSMark Fasheh /* already have the inode for our journal */ 483ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 484ccd979bdSMark Fasheh osb->slot_num); 485ccd979bdSMark Fasheh if (inode == NULL) { 486ccd979bdSMark Fasheh status = -EACCES; 487ccd979bdSMark Fasheh mlog_errno(status); 488ccd979bdSMark Fasheh goto done; 489ccd979bdSMark Fasheh } 490ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 491ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 492ccd979bdSMark Fasheh iput(inode); 493ccd979bdSMark Fasheh inode = NULL; 494ccd979bdSMark Fasheh status = -EACCES; 495ccd979bdSMark Fasheh goto done; 496ccd979bdSMark Fasheh } 497ccd979bdSMark Fasheh 498ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 499ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 500ccd979bdSMark Fasheh 5016eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 5026eff5790SMark Fasheh * changes in a live cluster so it can be considered an 5036eff5790SMark Fasheh * exception to the rule. */ 504e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 505ccd979bdSMark Fasheh if (status < 0) { 506ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 507ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 508ccd979bdSMark Fasheh goto done; 509ccd979bdSMark Fasheh } 510ccd979bdSMark Fasheh 511e63aecb6SMark Fasheh inode_lock = 1; 512ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 513ccd979bdSMark Fasheh 514ccd979bdSMark Fasheh if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 515ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 516ccd979bdSMark Fasheh inode->i_size); 517ccd979bdSMark Fasheh status = -EINVAL; 518ccd979bdSMark Fasheh goto done; 519ccd979bdSMark Fasheh } 520ccd979bdSMark Fasheh 521ccd979bdSMark Fasheh mlog(0, "inode->i_size = %lld\n", inode->i_size); 5225515eff8SAndrew Morton mlog(0, "inode->i_blocks = %llu\n", 5235515eff8SAndrew Morton (unsigned long long)inode->i_blocks); 524ccd979bdSMark Fasheh mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 525ccd979bdSMark Fasheh 526ccd979bdSMark Fasheh /* call the kernels journal init function now */ 527ccd979bdSMark Fasheh j_journal = journal_init_inode(inode); 528ccd979bdSMark Fasheh if (j_journal == NULL) { 529ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 530ccd979bdSMark Fasheh status = -EINVAL; 531ccd979bdSMark Fasheh goto done; 532ccd979bdSMark Fasheh } 533ccd979bdSMark Fasheh 534ccd979bdSMark Fasheh mlog(0, "Returned from journal_init_inode\n"); 535ccd979bdSMark Fasheh mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 536ccd979bdSMark Fasheh 537ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 538ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 539ccd979bdSMark Fasheh 540ccd979bdSMark Fasheh journal->j_journal = j_journal; 541ccd979bdSMark Fasheh journal->j_inode = inode; 542ccd979bdSMark Fasheh journal->j_bh = bh; 543ccd979bdSMark Fasheh 544ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 545ccd979bdSMark Fasheh 546ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 547ccd979bdSMark Fasheh 548ccd979bdSMark Fasheh status = 0; 549ccd979bdSMark Fasheh done: 550ccd979bdSMark Fasheh if (status < 0) { 551e63aecb6SMark Fasheh if (inode_lock) 552e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 553ccd979bdSMark Fasheh if (bh != NULL) 554ccd979bdSMark Fasheh brelse(bh); 555ccd979bdSMark Fasheh if (inode) { 556ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 557ccd979bdSMark Fasheh iput(inode); 558ccd979bdSMark Fasheh } 559ccd979bdSMark Fasheh } 560ccd979bdSMark Fasheh 561ccd979bdSMark Fasheh mlog_exit(status); 562ccd979bdSMark Fasheh return status; 563ccd979bdSMark Fasheh } 564ccd979bdSMark Fasheh 565539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 566539d8264SSunil Mushran { 567539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 568539d8264SSunil Mushran } 569539d8264SSunil Mushran 570539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 571539d8264SSunil Mushran { 572539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 573539d8264SSunil Mushran } 574539d8264SSunil Mushran 575ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 576539d8264SSunil Mushran int dirty, int replayed) 577ccd979bdSMark Fasheh { 578ccd979bdSMark Fasheh int status; 579ccd979bdSMark Fasheh unsigned int flags; 580ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 581ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 582ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 583ccd979bdSMark Fasheh 584ccd979bdSMark Fasheh mlog_entry_void(); 585ccd979bdSMark Fasheh 586ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 587ccd979bdSMark Fasheh if (!OCFS2_IS_VALID_DINODE(fe)) { 588ccd979bdSMark Fasheh /* This is called from startup/shutdown which will 589ccd979bdSMark Fasheh * handle the errors in a specific manner, so no need 590ccd979bdSMark Fasheh * to call ocfs2_error() here. */ 591b0697053SMark Fasheh mlog(ML_ERROR, "Journal dinode %llu has invalid " 5921ca1a111SMark Fasheh "signature: %.*s", 5931ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(fe->i_blkno), 7, 594b0697053SMark Fasheh fe->i_signature); 595ccd979bdSMark Fasheh status = -EIO; 596ccd979bdSMark Fasheh goto out; 597ccd979bdSMark Fasheh } 598ccd979bdSMark Fasheh 599ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 600ccd979bdSMark Fasheh if (dirty) 601ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 602ccd979bdSMark Fasheh else 603ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 604ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 605ccd979bdSMark Fasheh 606539d8264SSunil Mushran if (replayed) 607539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 608539d8264SSunil Mushran 609ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, journal->j_inode); 610ccd979bdSMark Fasheh if (status < 0) 611ccd979bdSMark Fasheh mlog_errno(status); 612ccd979bdSMark Fasheh 613ccd979bdSMark Fasheh out: 614ccd979bdSMark Fasheh mlog_exit(status); 615ccd979bdSMark Fasheh return status; 616ccd979bdSMark Fasheh } 617ccd979bdSMark Fasheh 618ccd979bdSMark Fasheh /* 619ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 620ccd979bdSMark Fasheh * call. 621ccd979bdSMark Fasheh */ 622ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 623ccd979bdSMark Fasheh { 624ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 625ccd979bdSMark Fasheh int status = 0; 626ccd979bdSMark Fasheh struct inode *inode = NULL; 627ccd979bdSMark Fasheh int num_running_trans = 0; 628ccd979bdSMark Fasheh 629ccd979bdSMark Fasheh mlog_entry_void(); 630ccd979bdSMark Fasheh 631ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 632ccd979bdSMark Fasheh 633ccd979bdSMark Fasheh journal = osb->journal; 634ccd979bdSMark Fasheh if (!journal) 635ccd979bdSMark Fasheh goto done; 636ccd979bdSMark Fasheh 637ccd979bdSMark Fasheh inode = journal->j_inode; 638ccd979bdSMark Fasheh 639ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 640ccd979bdSMark Fasheh goto done; 641ccd979bdSMark Fasheh 642ccd979bdSMark Fasheh /* need to inc inode use count as journal_destroy will iput. */ 643ccd979bdSMark Fasheh if (!igrab(inode)) 644ccd979bdSMark Fasheh BUG(); 645ccd979bdSMark Fasheh 646ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 647ccd979bdSMark Fasheh if (num_running_trans > 0) 648ccd979bdSMark Fasheh mlog(0, "Shutting down journal: must wait on %d " 649ccd979bdSMark Fasheh "running transactions!\n", 650ccd979bdSMark Fasheh num_running_trans); 651ccd979bdSMark Fasheh 652ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 653ccd979bdSMark Fasheh * release any locks that are still held. 654ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 655ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 656ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 657ccd979bdSMark Fasheh 658ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 659ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 660ccd979bdSMark Fasheh * completely destroy the journal. */ 661ccd979bdSMark Fasheh if (osb->commit_task) { 662ccd979bdSMark Fasheh /* Wait for the commit thread */ 663ccd979bdSMark Fasheh mlog(0, "Waiting for ocfs2commit to exit....\n"); 664ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 665ccd979bdSMark Fasheh osb->commit_task = NULL; 666ccd979bdSMark Fasheh } 667ccd979bdSMark Fasheh 668ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 669ccd979bdSMark Fasheh 670c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 671c271c5c2SSunil Mushran journal_lock_updates(journal->j_journal); 672c271c5c2SSunil Mushran status = journal_flush(journal->j_journal); 673c271c5c2SSunil Mushran journal_unlock_updates(journal->j_journal); 674c271c5c2SSunil Mushran if (status < 0) 675c271c5c2SSunil Mushran mlog_errno(status); 676c271c5c2SSunil Mushran } 677c271c5c2SSunil Mushran 678c271c5c2SSunil Mushran if (status == 0) { 679c271c5c2SSunil Mushran /* 680c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 681c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 682c271c5c2SSunil Mushran */ 683539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 684ccd979bdSMark Fasheh if (status < 0) 685ccd979bdSMark Fasheh mlog_errno(status); 686c271c5c2SSunil Mushran } 687ccd979bdSMark Fasheh 688ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 689ccd979bdSMark Fasheh journal_destroy(journal->j_journal); 690ccd979bdSMark Fasheh 691ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 692ccd979bdSMark Fasheh 693ccd979bdSMark Fasheh /* unlock our journal */ 694e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 695ccd979bdSMark Fasheh 696ccd979bdSMark Fasheh brelse(journal->j_bh); 697ccd979bdSMark Fasheh journal->j_bh = NULL; 698ccd979bdSMark Fasheh 699ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 700ccd979bdSMark Fasheh 701ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 702ccd979bdSMark Fasheh done: 703ccd979bdSMark Fasheh if (inode) 704ccd979bdSMark Fasheh iput(inode); 705ccd979bdSMark Fasheh mlog_exit_void(); 706ccd979bdSMark Fasheh } 707ccd979bdSMark Fasheh 708ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 709ccd979bdSMark Fasheh journal_t *journal, 710ccd979bdSMark Fasheh int slot) 711ccd979bdSMark Fasheh { 712ccd979bdSMark Fasheh int olderr; 713ccd979bdSMark Fasheh 714ccd979bdSMark Fasheh olderr = journal_errno(journal); 715ccd979bdSMark Fasheh if (olderr) { 716ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 717ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 718ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 719ccd979bdSMark Fasheh sb->s_id); 720ccd979bdSMark Fasheh 721ccd979bdSMark Fasheh journal_ack_err(journal); 722ccd979bdSMark Fasheh journal_clear_err(journal); 723ccd979bdSMark Fasheh } 724ccd979bdSMark Fasheh } 725ccd979bdSMark Fasheh 726539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 727ccd979bdSMark Fasheh { 728ccd979bdSMark Fasheh int status = 0; 729ccd979bdSMark Fasheh struct ocfs2_super *osb; 730ccd979bdSMark Fasheh 731ccd979bdSMark Fasheh mlog_entry_void(); 732ccd979bdSMark Fasheh 733b1f3550fSJulia Lawall BUG_ON(!journal); 734ccd979bdSMark Fasheh 735ccd979bdSMark Fasheh osb = journal->j_osb; 736ccd979bdSMark Fasheh 737ccd979bdSMark Fasheh status = journal_load(journal->j_journal); 738ccd979bdSMark Fasheh if (status < 0) { 739ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 740ccd979bdSMark Fasheh goto done; 741ccd979bdSMark Fasheh } 742ccd979bdSMark Fasheh 743ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 744ccd979bdSMark Fasheh 745539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 746ccd979bdSMark Fasheh if (status < 0) { 747ccd979bdSMark Fasheh mlog_errno(status); 748ccd979bdSMark Fasheh goto done; 749ccd979bdSMark Fasheh } 750ccd979bdSMark Fasheh 751ccd979bdSMark Fasheh /* Launch the commit thread */ 752c271c5c2SSunil Mushran if (!local) { 753c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 754c271c5c2SSunil Mushran "ocfs2cmt"); 755ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 756ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 757ccd979bdSMark Fasheh osb->commit_task = NULL; 758c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 759c271c5c2SSunil Mushran "error=%d", status); 760ccd979bdSMark Fasheh goto done; 761ccd979bdSMark Fasheh } 762c271c5c2SSunil Mushran } else 763c271c5c2SSunil Mushran osb->commit_task = NULL; 764ccd979bdSMark Fasheh 765ccd979bdSMark Fasheh done: 766ccd979bdSMark Fasheh mlog_exit(status); 767ccd979bdSMark Fasheh return status; 768ccd979bdSMark Fasheh } 769ccd979bdSMark Fasheh 770ccd979bdSMark Fasheh 771ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 772ccd979bdSMark Fasheh * mark the journal clean */ 773ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 774ccd979bdSMark Fasheh { 775ccd979bdSMark Fasheh int status; 776ccd979bdSMark Fasheh 777ccd979bdSMark Fasheh mlog_entry_void(); 778ccd979bdSMark Fasheh 779ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 780ccd979bdSMark Fasheh 781ccd979bdSMark Fasheh status = journal_wipe(journal->j_journal, full); 782ccd979bdSMark Fasheh if (status < 0) { 783ccd979bdSMark Fasheh mlog_errno(status); 784ccd979bdSMark Fasheh goto bail; 785ccd979bdSMark Fasheh } 786ccd979bdSMark Fasheh 787539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 788ccd979bdSMark Fasheh if (status < 0) 789ccd979bdSMark Fasheh mlog_errno(status); 790ccd979bdSMark Fasheh 791ccd979bdSMark Fasheh bail: 792ccd979bdSMark Fasheh mlog_exit(status); 793ccd979bdSMark Fasheh return status; 794ccd979bdSMark Fasheh } 795ccd979bdSMark Fasheh 796553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 797553abd04SJoel Becker { 798553abd04SJoel Becker int empty; 799553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 800553abd04SJoel Becker 801553abd04SJoel Becker spin_lock(&osb->osb_lock); 802553abd04SJoel Becker empty = (rm->rm_used == 0); 803553abd04SJoel Becker spin_unlock(&osb->osb_lock); 804553abd04SJoel Becker 805553abd04SJoel Becker return empty; 806553abd04SJoel Becker } 807553abd04SJoel Becker 808553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 809553abd04SJoel Becker { 810553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 811553abd04SJoel Becker } 812553abd04SJoel Becker 813ccd979bdSMark Fasheh /* 814ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 815ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 816ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 817ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 818ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 819ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 820ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 821ccd979bdSMark Fasheh * as we have the lock by now :) 822ccd979bdSMark Fasheh */ 823ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 824ccd979bdSMark Fasheh { 825ccd979bdSMark Fasheh int status = 0; 8264f902c37SMark Fasheh int i; 8278110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 8284f902c37SMark Fasheh #define CONCURRENT_JOURNAL_FILL 32ULL 829ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 830ccd979bdSMark Fasheh 831ccd979bdSMark Fasheh mlog_entry_void(); 832ccd979bdSMark Fasheh 833ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 834ccd979bdSMark Fasheh 8358110b073SMark Fasheh num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); 836ccd979bdSMark Fasheh v_blkno = 0; 8378110b073SMark Fasheh while (v_blkno < num_blocks) { 838ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 83949cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 840ccd979bdSMark Fasheh if (status < 0) { 841ccd979bdSMark Fasheh mlog_errno(status); 842ccd979bdSMark Fasheh goto bail; 843ccd979bdSMark Fasheh } 844ccd979bdSMark Fasheh 845ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 846ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 847ccd979bdSMark Fasheh 848dd4a2c2bSMark Fasheh /* We are reading journal data which should not 849dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 850ccd979bdSMark Fasheh status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 851ccd979bdSMark Fasheh p_blkno, p_blocks, bhs, 0, 852dd4a2c2bSMark Fasheh NULL); 853ccd979bdSMark Fasheh if (status < 0) { 854ccd979bdSMark Fasheh mlog_errno(status); 855ccd979bdSMark Fasheh goto bail; 856ccd979bdSMark Fasheh } 857ccd979bdSMark Fasheh 858ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 859ccd979bdSMark Fasheh brelse(bhs[i]); 860ccd979bdSMark Fasheh bhs[i] = NULL; 861ccd979bdSMark Fasheh } 862ccd979bdSMark Fasheh 863ccd979bdSMark Fasheh v_blkno += p_blocks; 864ccd979bdSMark Fasheh } 865ccd979bdSMark Fasheh 866ccd979bdSMark Fasheh bail: 867ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 868ccd979bdSMark Fasheh if (bhs[i]) 869ccd979bdSMark Fasheh brelse(bhs[i]); 870ccd979bdSMark Fasheh mlog_exit(status); 871ccd979bdSMark Fasheh return status; 872ccd979bdSMark Fasheh } 873ccd979bdSMark Fasheh 874ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 875ccd979bdSMark Fasheh struct list_head lri_list; 876ccd979bdSMark Fasheh int lri_slot; 877ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 878ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 879ccd979bdSMark Fasheh }; 880ccd979bdSMark Fasheh 881ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 882ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 883ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 884ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 885ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 886ccd979bdSMark Fasheh * recovering nodes into account. 887ccd979bdSMark Fasheh * 888ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 889ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 890ccd979bdSMark Fasheh */ 891c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 892ccd979bdSMark Fasheh { 893ccd979bdSMark Fasheh int ret; 894c4028958SDavid Howells struct ocfs2_journal *journal = 895c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 896c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 897ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 898800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 899ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 900ccd979bdSMark Fasheh 901ccd979bdSMark Fasheh mlog_entry_void(); 902ccd979bdSMark Fasheh 903ccd979bdSMark Fasheh mlog(0, "completing recovery from keventd\n"); 904ccd979bdSMark Fasheh 905ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 906ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 907ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 908ccd979bdSMark Fasheh 909800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 910ccd979bdSMark Fasheh list_del_init(&item->lri_list); 911ccd979bdSMark Fasheh 912ccd979bdSMark Fasheh mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 913ccd979bdSMark Fasheh 914ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 915ccd979bdSMark Fasheh if (la_dinode) { 916b0697053SMark Fasheh mlog(0, "Clean up local alloc %llu\n", 9171ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); 918ccd979bdSMark Fasheh 919ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 920ccd979bdSMark Fasheh la_dinode); 921ccd979bdSMark Fasheh if (ret < 0) 922ccd979bdSMark Fasheh mlog_errno(ret); 923ccd979bdSMark Fasheh 924ccd979bdSMark Fasheh kfree(la_dinode); 925ccd979bdSMark Fasheh } 926ccd979bdSMark Fasheh 927ccd979bdSMark Fasheh tl_dinode = item->lri_tl_dinode; 928ccd979bdSMark Fasheh if (tl_dinode) { 929b0697053SMark Fasheh mlog(0, "Clean up truncate log %llu\n", 9301ca1a111SMark Fasheh (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); 931ccd979bdSMark Fasheh 932ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 933ccd979bdSMark Fasheh tl_dinode); 934ccd979bdSMark Fasheh if (ret < 0) 935ccd979bdSMark Fasheh mlog_errno(ret); 936ccd979bdSMark Fasheh 937ccd979bdSMark Fasheh kfree(tl_dinode); 938ccd979bdSMark Fasheh } 939ccd979bdSMark Fasheh 940ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 941ccd979bdSMark Fasheh if (ret < 0) 942ccd979bdSMark Fasheh mlog_errno(ret); 943ccd979bdSMark Fasheh 944ccd979bdSMark Fasheh kfree(item); 945ccd979bdSMark Fasheh } 946ccd979bdSMark Fasheh 947ccd979bdSMark Fasheh mlog(0, "Recovery completion\n"); 948ccd979bdSMark Fasheh mlog_exit_void(); 949ccd979bdSMark Fasheh } 950ccd979bdSMark Fasheh 951ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 952ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 953ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 954ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 955ccd979bdSMark Fasheh int slot_num, 956ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 957ccd979bdSMark Fasheh struct ocfs2_dinode *tl_dinode) 958ccd979bdSMark Fasheh { 959ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 960ccd979bdSMark Fasheh 961afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 962ccd979bdSMark Fasheh if (!item) { 963ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 964ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 965ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 966ccd979bdSMark Fasheh if (la_dinode) 967ccd979bdSMark Fasheh kfree(la_dinode); 968ccd979bdSMark Fasheh 969ccd979bdSMark Fasheh if (tl_dinode) 970ccd979bdSMark Fasheh kfree(tl_dinode); 971ccd979bdSMark Fasheh 972ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 973ccd979bdSMark Fasheh return; 974ccd979bdSMark Fasheh } 975ccd979bdSMark Fasheh 976ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 977ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 978ccd979bdSMark Fasheh item->lri_slot = slot_num; 979ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 980ccd979bdSMark Fasheh 981ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 982ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 983ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 984ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 985ccd979bdSMark Fasheh } 986ccd979bdSMark Fasheh 987ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 988ccd979bdSMark Fasheh * recovery for it's own slot. */ 989ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 990ccd979bdSMark Fasheh { 991ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 992ccd979bdSMark Fasheh 993ccd979bdSMark Fasheh if (osb->dirty) { 994ccd979bdSMark Fasheh /* No need to queue up our truncate_log as regular 995ccd979bdSMark Fasheh * cleanup will catch that. */ 996ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(journal, 997ccd979bdSMark Fasheh osb->slot_num, 998ccd979bdSMark Fasheh osb->local_alloc_copy, 999ccd979bdSMark Fasheh NULL); 1000ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1001ccd979bdSMark Fasheh 1002ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 1003ccd979bdSMark Fasheh osb->dirty = 0; 1004ccd979bdSMark Fasheh } 1005ccd979bdSMark Fasheh } 1006ccd979bdSMark Fasheh 1007ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1008ccd979bdSMark Fasheh { 1009ccd979bdSMark Fasheh int status, node_num; 1010ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1011553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1012ccd979bdSMark Fasheh 1013ccd979bdSMark Fasheh mlog_entry_void(); 1014ccd979bdSMark Fasheh 1015ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1016ccd979bdSMark Fasheh if (status < 0) { 1017ccd979bdSMark Fasheh goto bail; 1018ccd979bdSMark Fasheh } 1019ccd979bdSMark Fasheh 1020ccd979bdSMark Fasheh restart: 1021ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1022ccd979bdSMark Fasheh if (status < 0) { 1023ccd979bdSMark Fasheh mlog_errno(status); 1024ccd979bdSMark Fasheh goto bail; 1025ccd979bdSMark Fasheh } 1026ccd979bdSMark Fasheh 1027553abd04SJoel Becker spin_lock(&osb->osb_lock); 1028553abd04SJoel Becker while (rm->rm_used) { 1029553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1030553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1031553abd04SJoel Becker node_num = rm->rm_entries[0]; 1032553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1033ccd979bdSMark Fasheh 1034ccd979bdSMark Fasheh status = ocfs2_recover_node(osb, node_num); 1035553abd04SJoel Becker if (!status) { 1036553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1037553abd04SJoel Becker } else { 1038ccd979bdSMark Fasheh mlog(ML_ERROR, 1039ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1040ccd979bdSMark Fasheh status, node_num, 1041ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1042ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1043ccd979bdSMark Fasheh } 1044ccd979bdSMark Fasheh 1045553abd04SJoel Becker spin_lock(&osb->osb_lock); 1046ccd979bdSMark Fasheh } 1047553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1048553abd04SJoel Becker mlog(0, "All nodes recovered\n"); 1049553abd04SJoel Becker 1050539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1051539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1052539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1053539d8264SSunil Mushran if (status < 0) 1054539d8264SSunil Mushran mlog_errno(status); 1055539d8264SSunil Mushran 1056ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1057ccd979bdSMark Fasheh 1058ccd979bdSMark Fasheh /* We always run recovery on our own orphan dir - the dead 105934d024f8SMark Fasheh * node(s) may have disallowd a previos inode delete. Re-processing 106034d024f8SMark Fasheh * is therefore required. */ 1061ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1062ccd979bdSMark Fasheh NULL); 1063ccd979bdSMark Fasheh 1064ccd979bdSMark Fasheh bail: 1065c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1066553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1067c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1068ccd979bdSMark Fasheh goto restart; 1069ccd979bdSMark Fasheh } 1070ccd979bdSMark Fasheh 1071ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1072ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1073ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1074ccd979bdSMark Fasheh 1075c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1076ccd979bdSMark Fasheh 1077ccd979bdSMark Fasheh mlog_exit(status); 1078ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1079ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1080ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1081ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1082ccd979bdSMark Fasheh return status; 1083ccd979bdSMark Fasheh } 1084ccd979bdSMark Fasheh 1085ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1086ccd979bdSMark Fasheh { 1087ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1088ccd979bdSMark Fasheh node_num, osb->node_num); 1089ccd979bdSMark Fasheh 1090c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1091ccd979bdSMark Fasheh if (osb->disable_recovery) 1092ccd979bdSMark Fasheh goto out; 1093ccd979bdSMark Fasheh 1094ccd979bdSMark Fasheh /* People waiting on recovery will wait on 1095ccd979bdSMark Fasheh * the recovery map to empty. */ 1096553abd04SJoel Becker if (ocfs2_recovery_map_set(osb, node_num)) 1097553abd04SJoel Becker mlog(0, "node %d already in recovery map.\n", node_num); 1098ccd979bdSMark Fasheh 1099ccd979bdSMark Fasheh mlog(0, "starting recovery thread...\n"); 1100ccd979bdSMark Fasheh 1101ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1102ccd979bdSMark Fasheh goto out; 1103ccd979bdSMark Fasheh 1104ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 110578427043SMark Fasheh "ocfs2rec"); 1106ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1107ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1108ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1109ccd979bdSMark Fasheh } 1110ccd979bdSMark Fasheh 1111ccd979bdSMark Fasheh out: 1112c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1113ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1114ccd979bdSMark Fasheh 1115ccd979bdSMark Fasheh mlog_exit_void(); 1116ccd979bdSMark Fasheh } 1117ccd979bdSMark Fasheh 1118539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1119539d8264SSunil Mushran int slot_num, 1120539d8264SSunil Mushran struct buffer_head **bh, 1121539d8264SSunil Mushran struct inode **ret_inode) 1122539d8264SSunil Mushran { 1123539d8264SSunil Mushran int status = -EACCES; 1124539d8264SSunil Mushran struct inode *inode = NULL; 1125539d8264SSunil Mushran 1126539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1127539d8264SSunil Mushran 1128539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1129539d8264SSunil Mushran slot_num); 1130539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1131539d8264SSunil Mushran mlog_errno(status); 1132539d8264SSunil Mushran goto bail; 1133539d8264SSunil Mushran } 1134539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1135539d8264SSunil Mushran 1136539d8264SSunil Mushran status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, bh, 0, inode); 1137539d8264SSunil Mushran if (status < 0) { 1138539d8264SSunil Mushran mlog_errno(status); 1139539d8264SSunil Mushran goto bail; 1140539d8264SSunil Mushran } 1141539d8264SSunil Mushran 1142539d8264SSunil Mushran status = 0; 1143539d8264SSunil Mushran 1144539d8264SSunil Mushran bail: 1145539d8264SSunil Mushran if (inode) { 1146539d8264SSunil Mushran if (status || !ret_inode) 1147539d8264SSunil Mushran iput(inode); 1148539d8264SSunil Mushran else 1149539d8264SSunil Mushran *ret_inode = inode; 1150539d8264SSunil Mushran } 1151539d8264SSunil Mushran return status; 1152539d8264SSunil Mushran } 1153539d8264SSunil Mushran 1154ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1155ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1156ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1157ccd979bdSMark Fasheh int node_num, 1158ccd979bdSMark Fasheh int slot_num) 1159ccd979bdSMark Fasheh { 1160ccd979bdSMark Fasheh int status; 1161ccd979bdSMark Fasheh int got_lock = 0; 1162ccd979bdSMark Fasheh unsigned int flags; 1163ccd979bdSMark Fasheh struct inode *inode = NULL; 1164ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1165ccd979bdSMark Fasheh journal_t *journal = NULL; 1166ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1167539d8264SSunil Mushran u32 slot_reco_gen; 1168ccd979bdSMark Fasheh 1169539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1170539d8264SSunil Mushran if (status) { 1171ccd979bdSMark Fasheh mlog_errno(status); 1172ccd979bdSMark Fasheh goto done; 1173ccd979bdSMark Fasheh } 1174539d8264SSunil Mushran 1175539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1176539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1177539d8264SSunil Mushran brelse(bh); 1178539d8264SSunil Mushran bh = NULL; 1179539d8264SSunil Mushran 1180539d8264SSunil Mushran /* 1181539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1182539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1183539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1184539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1185539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1186539d8264SSunil Mushran * If not, it needs recovery. 1187539d8264SSunil Mushran */ 1188539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1189539d8264SSunil Mushran mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num, 1190539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1191539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1192539d8264SSunil Mushran status = -EBUSY; 1193ccd979bdSMark Fasheh goto done; 1194ccd979bdSMark Fasheh } 1195539d8264SSunil Mushran 1196539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1197ccd979bdSMark Fasheh 1198e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1199ccd979bdSMark Fasheh if (status < 0) { 1200e63aecb6SMark Fasheh mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); 1201ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1202ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1203ccd979bdSMark Fasheh goto done; 1204ccd979bdSMark Fasheh } 1205ccd979bdSMark Fasheh got_lock = 1; 1206ccd979bdSMark Fasheh 1207ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1208ccd979bdSMark Fasheh 1209ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1210539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1211ccd979bdSMark Fasheh 1212ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1213ccd979bdSMark Fasheh mlog(0, "No recovery required for node %d\n", node_num); 1214539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1215539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1216ccd979bdSMark Fasheh goto done; 1217ccd979bdSMark Fasheh } 1218ccd979bdSMark Fasheh 1219ccd979bdSMark Fasheh mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 1220ccd979bdSMark Fasheh node_num, slot_num, 1221ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1222ccd979bdSMark Fasheh 1223ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1224ccd979bdSMark Fasheh 1225ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1226ccd979bdSMark Fasheh if (status < 0) { 1227ccd979bdSMark Fasheh mlog_errno(status); 1228ccd979bdSMark Fasheh goto done; 1229ccd979bdSMark Fasheh } 1230ccd979bdSMark Fasheh 1231ccd979bdSMark Fasheh mlog(0, "calling journal_init_inode\n"); 1232ccd979bdSMark Fasheh journal = journal_init_inode(inode); 1233ccd979bdSMark Fasheh if (journal == NULL) { 1234ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1235ccd979bdSMark Fasheh status = -EIO; 1236ccd979bdSMark Fasheh goto done; 1237ccd979bdSMark Fasheh } 1238ccd979bdSMark Fasheh 1239ccd979bdSMark Fasheh status = journal_load(journal); 1240ccd979bdSMark Fasheh if (status < 0) { 1241ccd979bdSMark Fasheh mlog_errno(status); 1242ccd979bdSMark Fasheh if (!igrab(inode)) 1243ccd979bdSMark Fasheh BUG(); 1244ccd979bdSMark Fasheh journal_destroy(journal); 1245ccd979bdSMark Fasheh goto done; 1246ccd979bdSMark Fasheh } 1247ccd979bdSMark Fasheh 1248ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1249ccd979bdSMark Fasheh 1250ccd979bdSMark Fasheh /* wipe the journal */ 1251ccd979bdSMark Fasheh mlog(0, "flushing the journal.\n"); 1252ccd979bdSMark Fasheh journal_lock_updates(journal); 1253ccd979bdSMark Fasheh status = journal_flush(journal); 1254ccd979bdSMark Fasheh journal_unlock_updates(journal); 1255ccd979bdSMark Fasheh if (status < 0) 1256ccd979bdSMark Fasheh mlog_errno(status); 1257ccd979bdSMark Fasheh 1258ccd979bdSMark Fasheh /* This will mark the node clean */ 1259ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1260ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1261ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1262ccd979bdSMark Fasheh 1263539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1264539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1265539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1266539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1267539d8264SSunil Mushran 1268ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, inode); 1269ccd979bdSMark Fasheh if (status < 0) 1270ccd979bdSMark Fasheh mlog_errno(status); 1271ccd979bdSMark Fasheh 1272ccd979bdSMark Fasheh if (!igrab(inode)) 1273ccd979bdSMark Fasheh BUG(); 1274ccd979bdSMark Fasheh 1275ccd979bdSMark Fasheh journal_destroy(journal); 1276ccd979bdSMark Fasheh 1277ccd979bdSMark Fasheh done: 1278ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1279ccd979bdSMark Fasheh if (got_lock) 1280e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1281ccd979bdSMark Fasheh 1282ccd979bdSMark Fasheh if (inode) 1283ccd979bdSMark Fasheh iput(inode); 1284ccd979bdSMark Fasheh 1285ccd979bdSMark Fasheh if (bh) 1286ccd979bdSMark Fasheh brelse(bh); 1287ccd979bdSMark Fasheh 1288ccd979bdSMark Fasheh mlog_exit(status); 1289ccd979bdSMark Fasheh return status; 1290ccd979bdSMark Fasheh } 1291ccd979bdSMark Fasheh 1292ccd979bdSMark Fasheh /* 1293ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1294ccd979bdSMark Fasheh * - Replay it's journal 1295ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1296ccd979bdSMark Fasheh * - Stamp a clean truncate log 1297ccd979bdSMark Fasheh * - Mark the node clean 1298ccd979bdSMark Fasheh * 1299ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1300ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1301ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1302ccd979bdSMark Fasheh * far less concerning. 1303ccd979bdSMark Fasheh */ 1304ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 1305ccd979bdSMark Fasheh int node_num) 1306ccd979bdSMark Fasheh { 1307ccd979bdSMark Fasheh int status = 0; 1308ccd979bdSMark Fasheh int slot_num; 1309ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1310ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1311ccd979bdSMark Fasheh 1312ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1313ccd979bdSMark Fasheh node_num, osb->node_num); 1314ccd979bdSMark Fasheh 1315ccd979bdSMark Fasheh mlog(0, "checking node %d\n", node_num); 1316ccd979bdSMark Fasheh 1317ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1318ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1319ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1320ccd979bdSMark Fasheh 1321d85b20e4SJoel Becker slot_num = ocfs2_node_num_to_slot(osb, node_num); 1322d85b20e4SJoel Becker if (slot_num == -ENOENT) { 1323ccd979bdSMark Fasheh status = 0; 1324ccd979bdSMark Fasheh mlog(0, "no slot for this node, so no recovery required.\n"); 1325ccd979bdSMark Fasheh goto done; 1326ccd979bdSMark Fasheh } 1327ccd979bdSMark Fasheh 1328ccd979bdSMark Fasheh mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1329ccd979bdSMark Fasheh 1330ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1331ccd979bdSMark Fasheh if (status < 0) { 1332539d8264SSunil Mushran if (status == -EBUSY) { 1333539d8264SSunil Mushran mlog(0, "Skipping recovery for slot %u (node %u) " 1334539d8264SSunil Mushran "as another node has recovered it\n", slot_num, 1335539d8264SSunil Mushran node_num); 1336539d8264SSunil Mushran status = 0; 1337539d8264SSunil Mushran goto done; 1338539d8264SSunil Mushran } 1339ccd979bdSMark Fasheh mlog_errno(status); 1340ccd979bdSMark Fasheh goto done; 1341ccd979bdSMark Fasheh } 1342ccd979bdSMark Fasheh 1343ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1344ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1345ccd979bdSMark Fasheh if (status < 0) { 1346ccd979bdSMark Fasheh mlog_errno(status); 1347ccd979bdSMark Fasheh goto done; 1348ccd979bdSMark Fasheh } 1349ccd979bdSMark Fasheh 1350ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1351ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1352ccd979bdSMark Fasheh * recovery. */ 1353ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1354ccd979bdSMark Fasheh if (status < 0) 1355ccd979bdSMark Fasheh mlog_errno(status); 1356ccd979bdSMark Fasheh 1357ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1358ccd979bdSMark Fasheh * harmful place to get an error... */ 13598e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1360ccd979bdSMark Fasheh if (status < 0) 1361ccd979bdSMark Fasheh mlog_errno(status); 1362ccd979bdSMark Fasheh 1363ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1364ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1365ccd979bdSMark Fasheh tl_copy); 1366ccd979bdSMark Fasheh 1367ccd979bdSMark Fasheh status = 0; 1368ccd979bdSMark Fasheh done: 1369ccd979bdSMark Fasheh 1370ccd979bdSMark Fasheh mlog_exit(status); 1371ccd979bdSMark Fasheh return status; 1372ccd979bdSMark Fasheh } 1373ccd979bdSMark Fasheh 1374ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1375ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1376ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1377ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1378ccd979bdSMark Fasheh int slot_num) 1379ccd979bdSMark Fasheh { 1380ccd979bdSMark Fasheh int status, flags; 1381ccd979bdSMark Fasheh struct inode *inode = NULL; 1382ccd979bdSMark Fasheh 1383ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1384ccd979bdSMark Fasheh slot_num); 1385ccd979bdSMark Fasheh if (inode == NULL) { 1386ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1387ccd979bdSMark Fasheh status = -EACCES; 1388ccd979bdSMark Fasheh goto bail; 1389ccd979bdSMark Fasheh } 1390ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1391ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1392ccd979bdSMark Fasheh iput(inode); 1393ccd979bdSMark Fasheh inode = NULL; 1394ccd979bdSMark Fasheh status = -EACCES; 1395ccd979bdSMark Fasheh goto bail; 1396ccd979bdSMark Fasheh } 1397ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1398ccd979bdSMark Fasheh 1399ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1400e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1401ccd979bdSMark Fasheh if (status < 0) { 1402ccd979bdSMark Fasheh if (status != -EAGAIN) 1403ccd979bdSMark Fasheh mlog_errno(status); 1404ccd979bdSMark Fasheh goto bail; 1405ccd979bdSMark Fasheh } 1406ccd979bdSMark Fasheh 1407e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1408ccd979bdSMark Fasheh bail: 1409ccd979bdSMark Fasheh if (inode) 1410ccd979bdSMark Fasheh iput(inode); 1411ccd979bdSMark Fasheh 1412ccd979bdSMark Fasheh return status; 1413ccd979bdSMark Fasheh } 1414ccd979bdSMark Fasheh 1415ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1416ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1417ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1418ccd979bdSMark Fasheh { 1419d85b20e4SJoel Becker unsigned int node_num; 1420d85b20e4SJoel Becker int status, i; 1421*a1af7d15SMark Fasheh u32 gen; 1422539d8264SSunil Mushran struct buffer_head *bh = NULL; 1423539d8264SSunil Mushran struct ocfs2_dinode *di; 1424ccd979bdSMark Fasheh 1425ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1426ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1427ccd979bdSMark Fasheh 1428d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1429539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1430539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1431539d8264SSunil Mushran if (status) { 1432539d8264SSunil Mushran mlog_errno(status); 1433539d8264SSunil Mushran goto bail; 1434539d8264SSunil Mushran } 1435539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1436*a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1437539d8264SSunil Mushran brelse(bh); 1438539d8264SSunil Mushran bh = NULL; 1439539d8264SSunil Mushran 1440*a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1441*a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1442*a1af7d15SMark Fasheh 1443539d8264SSunil Mushran mlog(0, "Slot %u recovery generation is %u\n", i, 1444539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1445539d8264SSunil Mushran 1446*a1af7d15SMark Fasheh if (i == osb->slot_num) { 1447*a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1448ccd979bdSMark Fasheh continue; 1449*a1af7d15SMark Fasheh } 1450d85b20e4SJoel Becker 1451d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1452*a1af7d15SMark Fasheh if (status == -ENOENT) { 1453*a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1454ccd979bdSMark Fasheh continue; 1455*a1af7d15SMark Fasheh } 1456ccd979bdSMark Fasheh 1457*a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1458*a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1459ccd979bdSMark Fasheh continue; 1460*a1af7d15SMark Fasheh } 1461d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1462ccd979bdSMark Fasheh 1463ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1464ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1465ccd979bdSMark Fasheh * file here to test if he's alive. */ 1466ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1467ccd979bdSMark Fasheh if (!status) { 1468ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1469ccd979bdSMark Fasheh * the recovery thread can't race us on 1470ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1471ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1472ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1473ccd979bdSMark Fasheh mlog_errno(status); 1474ccd979bdSMark Fasheh goto bail; 1475ccd979bdSMark Fasheh } 1476ccd979bdSMark Fasheh } 1477ccd979bdSMark Fasheh 1478ccd979bdSMark Fasheh status = 0; 1479ccd979bdSMark Fasheh bail: 1480ccd979bdSMark Fasheh mlog_exit(status); 1481ccd979bdSMark Fasheh return status; 1482ccd979bdSMark Fasheh } 1483ccd979bdSMark Fasheh 14845eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 14855eae5b96SMark Fasheh struct inode *head; 14865eae5b96SMark Fasheh struct ocfs2_super *osb; 14875eae5b96SMark Fasheh }; 14885eae5b96SMark Fasheh 14895eae5b96SMark Fasheh static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, 14905eae5b96SMark Fasheh loff_t pos, u64 ino, unsigned type) 14915eae5b96SMark Fasheh { 14925eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv *p = priv; 14935eae5b96SMark Fasheh struct inode *iter; 14945eae5b96SMark Fasheh 14955eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 14965eae5b96SMark Fasheh return 0; 14975eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 14985eae5b96SMark Fasheh return 0; 14995eae5b96SMark Fasheh 15005eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 15015eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 15025fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 15035eae5b96SMark Fasheh if (IS_ERR(iter)) 15045eae5b96SMark Fasheh return 0; 15055eae5b96SMark Fasheh 15065eae5b96SMark Fasheh mlog(0, "queue orphan %llu\n", 15075eae5b96SMark Fasheh (unsigned long long)OCFS2_I(iter)->ip_blkno); 15085eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 15095eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 15105eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 15115eae5b96SMark Fasheh p->head = iter; 15125eae5b96SMark Fasheh 15135eae5b96SMark Fasheh return 0; 15145eae5b96SMark Fasheh } 15155eae5b96SMark Fasheh 1516b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1517b4df6ed8SMark Fasheh int slot, 1518b4df6ed8SMark Fasheh struct inode **head) 1519ccd979bdSMark Fasheh { 1520b4df6ed8SMark Fasheh int status; 1521ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 15225eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv priv; 15235eae5b96SMark Fasheh loff_t pos = 0; 15245eae5b96SMark Fasheh 15255eae5b96SMark Fasheh priv.osb = osb; 15265eae5b96SMark Fasheh priv.head = *head; 1527ccd979bdSMark Fasheh 1528ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1529ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 1530ccd979bdSMark Fasheh slot); 1531ccd979bdSMark Fasheh if (!orphan_dir_inode) { 1532ccd979bdSMark Fasheh status = -ENOENT; 1533ccd979bdSMark Fasheh mlog_errno(status); 1534b4df6ed8SMark Fasheh return status; 1535ccd979bdSMark Fasheh } 1536ccd979bdSMark Fasheh 15371b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 1538e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 1539ccd979bdSMark Fasheh if (status < 0) { 1540ccd979bdSMark Fasheh mlog_errno(status); 1541ccd979bdSMark Fasheh goto out; 1542ccd979bdSMark Fasheh } 1543ccd979bdSMark Fasheh 15445eae5b96SMark Fasheh status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, 15455eae5b96SMark Fasheh ocfs2_orphan_filldir); 15465eae5b96SMark Fasheh if (status) { 1547ccd979bdSMark Fasheh mlog_errno(status); 1548a86370fbSMark Fasheh goto out_cluster; 1549ccd979bdSMark Fasheh } 1550ccd979bdSMark Fasheh 15515eae5b96SMark Fasheh *head = priv.head; 1552ccd979bdSMark Fasheh 1553a86370fbSMark Fasheh out_cluster: 1554e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 1555b4df6ed8SMark Fasheh out: 1556b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 1557ccd979bdSMark Fasheh iput(orphan_dir_inode); 1558b4df6ed8SMark Fasheh return status; 1559b4df6ed8SMark Fasheh } 1560b4df6ed8SMark Fasheh 1561b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1562b4df6ed8SMark Fasheh int slot) 1563b4df6ed8SMark Fasheh { 1564b4df6ed8SMark Fasheh int ret; 1565b4df6ed8SMark Fasheh 1566b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1567b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 1568b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1569b4df6ed8SMark Fasheh return ret; 1570b4df6ed8SMark Fasheh } 1571b4df6ed8SMark Fasheh 1572b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1573b4df6ed8SMark Fasheh int slot) 1574b4df6ed8SMark Fasheh { 1575b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1576b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 1577b4df6ed8SMark Fasheh * know to quit early. */ 1578b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1579b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 1580b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 1581b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 1582b4df6ed8SMark Fasheh * them. */ 1583b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1584b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 1585b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 1586b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1587b4df6ed8SMark Fasheh } 1588b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1589b4df6ed8SMark Fasheh } 1590b4df6ed8SMark Fasheh 1591b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1592b4df6ed8SMark Fasheh int slot) 1593b4df6ed8SMark Fasheh { 1594b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1595b4df6ed8SMark Fasheh } 1596b4df6ed8SMark Fasheh 1597b4df6ed8SMark Fasheh /* 1598b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 1599b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 1600b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 1601b4df6ed8SMark Fasheh * (most) of the rest of the work. 1602b4df6ed8SMark Fasheh * 1603b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 1604b4df6ed8SMark Fasheh * couple of extra considerations. 1605b4df6ed8SMark Fasheh * 1606b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 1607b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 1608b4df6ed8SMark Fasheh * an invalid inode. 1609b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 1610b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 1611b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 1612b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 1613b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 1614b4df6ed8SMark Fasheh */ 1615b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1616b4df6ed8SMark Fasheh int slot) 1617b4df6ed8SMark Fasheh { 1618b4df6ed8SMark Fasheh int ret = 0; 1619b4df6ed8SMark Fasheh struct inode *inode = NULL; 1620b4df6ed8SMark Fasheh struct inode *iter; 1621b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 1622b4df6ed8SMark Fasheh 1623b4df6ed8SMark Fasheh mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1624b4df6ed8SMark Fasheh 1625b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 1626b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 1627b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 1628b4df6ed8SMark Fasheh 1629b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 1630b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 1631b4df6ed8SMark Fasheh if (ret) 1632b4df6ed8SMark Fasheh mlog_errno(ret); 1633ccd979bdSMark Fasheh 1634ccd979bdSMark Fasheh while (inode) { 1635ccd979bdSMark Fasheh oi = OCFS2_I(inode); 1636b0697053SMark Fasheh mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1637ccd979bdSMark Fasheh 1638ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 1639ccd979bdSMark Fasheh 1640ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 164134d024f8SMark Fasheh /* The remote delete code may have set these on the 164234d024f8SMark Fasheh * assumption that the other node would wipe them 164334d024f8SMark Fasheh * successfully. If they are still in the node's 164434d024f8SMark Fasheh * orphan dir, we need to reset that state. */ 1645ccd979bdSMark Fasheh oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1646ccd979bdSMark Fasheh 1647ccd979bdSMark Fasheh /* Set the proper information to get us going into 1648ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 1649ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1650ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 1651ccd979bdSMark Fasheh 1652ccd979bdSMark Fasheh iput(inode); 1653ccd979bdSMark Fasheh 1654ccd979bdSMark Fasheh inode = iter; 1655ccd979bdSMark Fasheh } 1656ccd979bdSMark Fasheh 1657b4df6ed8SMark Fasheh return ret; 1658ccd979bdSMark Fasheh } 1659ccd979bdSMark Fasheh 1660ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1661ccd979bdSMark Fasheh { 1662ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 1663ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 1664ccd979bdSMark Fasheh * or DISABLED. */ 1665ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 1666ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1667ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1668ccd979bdSMark Fasheh 1669ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 1670ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 1671ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 1672ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1673ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 1674ccd979bdSMark Fasheh return -EBUSY; 1675ccd979bdSMark Fasheh } 1676ccd979bdSMark Fasheh 1677ccd979bdSMark Fasheh return 0; 1678ccd979bdSMark Fasheh } 1679ccd979bdSMark Fasheh 1680ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 1681ccd979bdSMark Fasheh { 1682ccd979bdSMark Fasheh int status; 1683ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1684ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1685ccd979bdSMark Fasheh 1686ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 1687ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 1688ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 1689ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 1690ccd979bdSMark Fasheh while (!(kthread_should_stop() && 1691ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 1692ccd979bdSMark Fasheh 1693745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 1694ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 1695745ae8baSMark Fasheh || kthread_should_stop()); 1696ccd979bdSMark Fasheh 1697ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 1698ccd979bdSMark Fasheh if (status < 0) 1699ccd979bdSMark Fasheh mlog_errno(status); 1700ccd979bdSMark Fasheh 1701ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1702ccd979bdSMark Fasheh mlog(ML_KTHREAD, 1703ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 1704ccd979bdSMark Fasheh "shutdown\n", 1705ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 1706ccd979bdSMark Fasheh } 1707ccd979bdSMark Fasheh } 1708ccd979bdSMark Fasheh 1709ccd979bdSMark Fasheh return 0; 1710ccd979bdSMark Fasheh } 1711ccd979bdSMark Fasheh 1712539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 1713539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 1714539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 1715539d8264SSunil Mushran * a journal has been recovered by another node. 1716539d8264SSunil Mushran */ 1717ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1718ccd979bdSMark Fasheh { 1719ccd979bdSMark Fasheh int ret = 0; 1720ccd979bdSMark Fasheh unsigned int slot; 1721539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 1722ccd979bdSMark Fasheh struct ocfs2_dinode *di; 1723539d8264SSunil Mushran int journal_dirty = 0; 1724ccd979bdSMark Fasheh 1725ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 1726539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 1727539d8264SSunil Mushran if (ret) { 1728ccd979bdSMark Fasheh mlog_errno(ret); 1729ccd979bdSMark Fasheh goto out; 1730ccd979bdSMark Fasheh } 1731ccd979bdSMark Fasheh 1732ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 1733ccd979bdSMark Fasheh 1734539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 1735539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 1736539d8264SSunil Mushran 1737ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 1738ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 1739539d8264SSunil Mushran journal_dirty = 1; 1740ccd979bdSMark Fasheh 1741ccd979bdSMark Fasheh brelse(di_bh); 1742539d8264SSunil Mushran di_bh = NULL; 1743ccd979bdSMark Fasheh } 1744ccd979bdSMark Fasheh 1745ccd979bdSMark Fasheh out: 1746539d8264SSunil Mushran if (journal_dirty) 1747539d8264SSunil Mushran ret = -EROFS; 1748ccd979bdSMark Fasheh return ret; 1749ccd979bdSMark Fasheh } 1750