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> 3183273932SSrinivas Eeda #include <linux/time.h> 3283273932SSrinivas Eeda #include <linux/random.h> 33ccd979bdSMark Fasheh 34ccd979bdSMark Fasheh #include <cluster/masklog.h> 35ccd979bdSMark Fasheh 36ccd979bdSMark Fasheh #include "ocfs2.h" 37ccd979bdSMark Fasheh 38ccd979bdSMark Fasheh #include "alloc.h" 3950655ae9SJoel Becker #include "blockcheck.h" 40316f4b9fSMark Fasheh #include "dir.h" 41ccd979bdSMark Fasheh #include "dlmglue.h" 42ccd979bdSMark Fasheh #include "extent_map.h" 43ccd979bdSMark Fasheh #include "heartbeat.h" 44ccd979bdSMark Fasheh #include "inode.h" 45ccd979bdSMark Fasheh #include "journal.h" 46ccd979bdSMark Fasheh #include "localalloc.h" 47ccd979bdSMark Fasheh #include "slot_map.h" 48ccd979bdSMark Fasheh #include "super.h" 49ccd979bdSMark Fasheh #include "sysfile.h" 500cf2f763SJoel Becker #include "uptodate.h" 512205363dSJan Kara #include "quota.h" 52ccd979bdSMark Fasheh 53ccd979bdSMark Fasheh #include "buffer_head_io.h" 54*b4107950STao Ma #include "ocfs2_trace.h" 55ccd979bdSMark Fasheh 5634af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 57ccd979bdSMark Fasheh 5883273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 5983273932SSrinivas Eeda 60ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 61ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 622205363dSJan Kara int node_num, int slot_num); 63ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 64ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 6519ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 66ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 67539d8264SSunil Mushran int dirty, int replayed); 68ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 69ccd979bdSMark Fasheh int slot_num); 70ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 71ccd979bdSMark Fasheh int slot); 72ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 739140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 749140db04SSrinivas Eeda int slot_num, 759140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 769140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 779140db04SSrinivas Eeda struct ocfs2_quota_recovery *qrec); 78ccd979bdSMark Fasheh 7919ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 8019ece546SJan Kara { 8119ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 8219ece546SJan Kara } 8319ece546SJan Kara 8419ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 8519ece546SJan Kara { 8619ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 8719ece546SJan Kara } 8819ece546SJan Kara 899140db04SSrinivas Eeda /* 909140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 919140db04SSrinivas Eeda * offline slots during recovery and mount 929140db04SSrinivas Eeda */ 939140db04SSrinivas Eeda 949140db04SSrinivas Eeda enum ocfs2_replay_state { 959140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 969140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 979140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 989140db04SSrinivas Eeda }; 999140db04SSrinivas Eeda 1009140db04SSrinivas Eeda struct ocfs2_replay_map { 1019140db04SSrinivas Eeda unsigned int rm_slots; 1029140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 1039140db04SSrinivas Eeda unsigned char rm_replay_slots[0]; 1049140db04SSrinivas Eeda }; 1059140db04SSrinivas Eeda 1069140db04SSrinivas Eeda void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 1079140db04SSrinivas Eeda { 1089140db04SSrinivas Eeda if (!osb->replay_map) 1099140db04SSrinivas Eeda return; 1109140db04SSrinivas Eeda 1119140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1129140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1139140db04SSrinivas Eeda return; 1149140db04SSrinivas Eeda 1159140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1169140db04SSrinivas Eeda } 1179140db04SSrinivas Eeda 1189140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1199140db04SSrinivas Eeda { 1209140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1219140db04SSrinivas Eeda int i, node_num; 1229140db04SSrinivas Eeda 1239140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1249140db04SSrinivas Eeda if (osb->replay_map) 1259140db04SSrinivas Eeda return 0; 1269140db04SSrinivas Eeda 1279140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1289140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1299140db04SSrinivas Eeda 1309140db04SSrinivas Eeda if (!replay_map) { 1319140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1329140db04SSrinivas Eeda return -ENOMEM; 1339140db04SSrinivas Eeda } 1349140db04SSrinivas Eeda 1359140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1369140db04SSrinivas Eeda 1379140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1389140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1399140db04SSrinivas Eeda 1409140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1419140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1429140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1439140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1449140db04SSrinivas Eeda } 1459140db04SSrinivas Eeda 1469140db04SSrinivas Eeda osb->replay_map = replay_map; 1479140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1489140db04SSrinivas Eeda return 0; 1499140db04SSrinivas Eeda } 1509140db04SSrinivas Eeda 1519140db04SSrinivas Eeda void ocfs2_queue_replay_slots(struct ocfs2_super *osb) 1529140db04SSrinivas Eeda { 1539140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1549140db04SSrinivas Eeda int i; 1559140db04SSrinivas Eeda 1569140db04SSrinivas Eeda if (!replay_map) 1579140db04SSrinivas Eeda return; 1589140db04SSrinivas Eeda 1599140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1609140db04SSrinivas Eeda return; 1619140db04SSrinivas Eeda 1629140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1639140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1649140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 1659140db04SSrinivas Eeda NULL, NULL); 1669140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1679140db04SSrinivas Eeda } 1689140db04SSrinivas Eeda 1699140db04SSrinivas Eeda void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1709140db04SSrinivas Eeda { 1719140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1729140db04SSrinivas Eeda 1739140db04SSrinivas Eeda if (!osb->replay_map) 1749140db04SSrinivas Eeda return; 1759140db04SSrinivas Eeda 1769140db04SSrinivas Eeda kfree(replay_map); 1779140db04SSrinivas Eeda osb->replay_map = NULL; 1789140db04SSrinivas Eeda } 1799140db04SSrinivas Eeda 180553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 181553abd04SJoel Becker { 182553abd04SJoel Becker struct ocfs2_recovery_map *rm; 183553abd04SJoel Becker 184553abd04SJoel Becker mutex_init(&osb->recovery_lock); 185553abd04SJoel Becker osb->disable_recovery = 0; 186553abd04SJoel Becker osb->recovery_thread_task = NULL; 187553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 188553abd04SJoel Becker 189553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 190553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 191553abd04SJoel Becker GFP_KERNEL); 192553abd04SJoel Becker if (!rm) { 193553abd04SJoel Becker mlog_errno(-ENOMEM); 194553abd04SJoel Becker return -ENOMEM; 195553abd04SJoel Becker } 196553abd04SJoel Becker 197553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 198553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 199553abd04SJoel Becker osb->recovery_map = rm; 200553abd04SJoel Becker 201553abd04SJoel Becker return 0; 202553abd04SJoel Becker } 203553abd04SJoel Becker 204553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 205553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 206553abd04SJoel Becker * being woken up */ 207553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 208553abd04SJoel Becker { 209553abd04SJoel Becker mb(); 210553abd04SJoel Becker return osb->recovery_thread_task != NULL; 211553abd04SJoel Becker } 212553abd04SJoel Becker 213553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 214553abd04SJoel Becker { 215553abd04SJoel Becker struct ocfs2_recovery_map *rm; 216553abd04SJoel Becker 217553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 218553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 219553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 220553abd04SJoel Becker osb->disable_recovery = 1; 221553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 222553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 223553abd04SJoel Becker 224553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 225553abd04SJoel Becker * launched, so wait for any recovery completion work to 226553abd04SJoel Becker * complete. */ 227553abd04SJoel Becker flush_workqueue(ocfs2_wq); 228553abd04SJoel Becker 229553abd04SJoel Becker /* 230553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 231553abd04SJoel Becker * freed, the osb_lock is not taken here. 232553abd04SJoel Becker */ 233553abd04SJoel Becker rm = osb->recovery_map; 234553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 235553abd04SJoel Becker 236553abd04SJoel Becker kfree(rm); 237553abd04SJoel Becker } 238553abd04SJoel Becker 239553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 240553abd04SJoel Becker unsigned int node_num) 241553abd04SJoel Becker { 242553abd04SJoel Becker int i; 243553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 244553abd04SJoel Becker 245553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 246553abd04SJoel Becker 247553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 248553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 249553abd04SJoel Becker return 1; 250553abd04SJoel Becker } 251553abd04SJoel Becker 252553abd04SJoel Becker return 0; 253553abd04SJoel Becker } 254553abd04SJoel Becker 255553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 256553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 257553abd04SJoel Becker unsigned int node_num) 258553abd04SJoel Becker { 259553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 260553abd04SJoel Becker 261553abd04SJoel Becker spin_lock(&osb->osb_lock); 262553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 263553abd04SJoel Becker spin_unlock(&osb->osb_lock); 264553abd04SJoel Becker return 1; 265553abd04SJoel Becker } 266553abd04SJoel Becker 267553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 268553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 269553abd04SJoel Becker 270553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 271553abd04SJoel Becker rm->rm_used++; 272553abd04SJoel Becker spin_unlock(&osb->osb_lock); 273553abd04SJoel Becker 274553abd04SJoel Becker return 0; 275553abd04SJoel Becker } 276553abd04SJoel Becker 277553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 278553abd04SJoel Becker unsigned int node_num) 279553abd04SJoel Becker { 280553abd04SJoel Becker int i; 281553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 282553abd04SJoel Becker 283553abd04SJoel Becker spin_lock(&osb->osb_lock); 284553abd04SJoel Becker 285553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 286553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 287553abd04SJoel Becker break; 288553abd04SJoel Becker } 289553abd04SJoel Becker 290553abd04SJoel Becker if (i < rm->rm_used) { 291553abd04SJoel Becker /* XXX: be careful with the pointer math */ 292553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 293553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 294553abd04SJoel Becker rm->rm_used--; 295553abd04SJoel Becker } 296553abd04SJoel Becker 297553abd04SJoel Becker spin_unlock(&osb->osb_lock); 298553abd04SJoel Becker } 299553abd04SJoel Becker 300ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 301ccd979bdSMark Fasheh { 302ccd979bdSMark Fasheh int status = 0; 303ccd979bdSMark Fasheh unsigned int flushed; 304ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 305ccd979bdSMark Fasheh 306ccd979bdSMark Fasheh journal = osb->journal; 307ccd979bdSMark Fasheh 308ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 309ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 310ccd979bdSMark Fasheh 311*b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 312*b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 313*b4107950STao Ma if (flushed == 0) { 314ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 315ccd979bdSMark Fasheh goto finally; 316ccd979bdSMark Fasheh } 317ccd979bdSMark Fasheh 3182b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 3192b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 3202b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 321ccd979bdSMark Fasheh if (status < 0) { 322ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 323ccd979bdSMark Fasheh mlog_errno(status); 324ccd979bdSMark Fasheh goto finally; 325ccd979bdSMark Fasheh } 326ccd979bdSMark Fasheh 327f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 328ccd979bdSMark Fasheh 329ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 330ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 331ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 332ccd979bdSMark Fasheh 333*b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 334ccd979bdSMark Fasheh 33534d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 336ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 337ccd979bdSMark Fasheh finally: 338ccd979bdSMark Fasheh return status; 339ccd979bdSMark Fasheh } 340ccd979bdSMark Fasheh 3411fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 342ccd979bdSMark Fasheh { 343ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3441fabe148SMark Fasheh handle_t *handle; 345ccd979bdSMark Fasheh 346ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 347ccd979bdSMark Fasheh 34865eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 34965eff9ccSMark Fasheh return ERR_PTR(-EROFS); 350ccd979bdSMark Fasheh 351ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 352ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 353ccd979bdSMark Fasheh 35490e86a63SJan Kara /* Nested transaction? Just return the handle... */ 35590e86a63SJan Kara if (journal_current_handle()) 35690e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 357ccd979bdSMark Fasheh 358ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 359ccd979bdSMark Fasheh 3602b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3611fabe148SMark Fasheh if (IS_ERR(handle)) { 362ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 363ccd979bdSMark Fasheh 3641fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 365ccd979bdSMark Fasheh 366ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 367ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 3681fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 369ccd979bdSMark Fasheh } 370c271c5c2SSunil Mushran } else { 371c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 372ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 373c271c5c2SSunil Mushran } 374ccd979bdSMark Fasheh 375ccd979bdSMark Fasheh return handle; 376ccd979bdSMark Fasheh } 377ccd979bdSMark Fasheh 3781fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3791fabe148SMark Fasheh handle_t *handle) 380ccd979bdSMark Fasheh { 38190e86a63SJan Kara int ret, nested; 38202dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 383ccd979bdSMark Fasheh 384ccd979bdSMark Fasheh BUG_ON(!handle); 385ccd979bdSMark Fasheh 38690e86a63SJan Kara nested = handle->h_ref > 1; 3872b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3881fabe148SMark Fasheh if (ret < 0) 3891fabe148SMark Fasheh mlog_errno(ret); 390ccd979bdSMark Fasheh 39190e86a63SJan Kara if (!nested) 392ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 393ccd979bdSMark Fasheh 3941fabe148SMark Fasheh return ret; 395ccd979bdSMark Fasheh } 396ccd979bdSMark Fasheh 397ccd979bdSMark Fasheh /* 398c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 399ccd979bdSMark Fasheh * 4002b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 401e8aed345SMark Fasheh * and then restart the transaction. Before calling 402e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 403e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 404e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 405e8aed345SMark Fasheh * 406ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 407ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 408ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 409ccd979bdSMark Fasheh * 410ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 411ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 412ccd979bdSMark Fasheh * cluster locks associated with this handle. 413ccd979bdSMark Fasheh */ 4141fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 415ccd979bdSMark Fasheh { 416c901fb00STao Ma int status, old_nblocks; 417ccd979bdSMark Fasheh 418ccd979bdSMark Fasheh BUG_ON(!handle); 419c901fb00STao Ma BUG_ON(nblocks < 0); 420ccd979bdSMark Fasheh 421c901fb00STao Ma if (!nblocks) 422c901fb00STao Ma return 0; 423c901fb00STao Ma 424c901fb00STao Ma old_nblocks = handle->h_buffer_credits; 425ccd979bdSMark Fasheh 426*b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 427ccd979bdSMark Fasheh 428e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4290879c584SMark Fasheh status = 1; 4300879c584SMark Fasheh #else 4312b4e30fbSJoel Becker status = jbd2_journal_extend(handle, nblocks); 432ccd979bdSMark Fasheh if (status < 0) { 433ccd979bdSMark Fasheh mlog_errno(status); 434ccd979bdSMark Fasheh goto bail; 435ccd979bdSMark Fasheh } 4360879c584SMark Fasheh #endif 437ccd979bdSMark Fasheh 438ccd979bdSMark Fasheh if (status > 0) { 439*b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 440c901fb00STao Ma status = jbd2_journal_restart(handle, 441c901fb00STao Ma old_nblocks + nblocks); 442ccd979bdSMark Fasheh if (status < 0) { 443ccd979bdSMark Fasheh mlog_errno(status); 444ccd979bdSMark Fasheh goto bail; 445ccd979bdSMark Fasheh } 44601ddf1e1SMark Fasheh } 447ccd979bdSMark Fasheh 448ccd979bdSMark Fasheh status = 0; 449ccd979bdSMark Fasheh bail: 450ccd979bdSMark Fasheh return status; 451ccd979bdSMark Fasheh } 452ccd979bdSMark Fasheh 45350655ae9SJoel Becker struct ocfs2_triggers { 45450655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 45550655ae9SJoel Becker int ot_offset; 45650655ae9SJoel Becker }; 45750655ae9SJoel Becker 45850655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 45950655ae9SJoel Becker { 46050655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 46150655ae9SJoel Becker } 46250655ae9SJoel Becker 46313ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 46450655ae9SJoel Becker struct buffer_head *bh, 46550655ae9SJoel Becker void *data, size_t size) 46650655ae9SJoel Becker { 46750655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 46850655ae9SJoel Becker 46950655ae9SJoel Becker /* 47050655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 47150655ae9SJoel Becker * must unconditionally compute the ecc data. 47250655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 47350655ae9SJoel Becker * metaecc is enabled. 47450655ae9SJoel Becker */ 47550655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 47650655ae9SJoel Becker } 47750655ae9SJoel Becker 47850655ae9SJoel Becker /* 47950655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 48050655ae9SJoel Becker * offset depends on the blocksize. 48150655ae9SJoel Becker */ 48213ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 48350655ae9SJoel Becker struct buffer_head *bh, 48450655ae9SJoel Becker void *data, size_t size) 48550655ae9SJoel Becker { 48650655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 48750655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 48850655ae9SJoel Becker 48950655ae9SJoel Becker /* 49050655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 49150655ae9SJoel Becker * must unconditionally compute the ecc data. 49250655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 49350655ae9SJoel Becker * metaecc is enabled. 49450655ae9SJoel Becker */ 49550655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 49650655ae9SJoel Becker } 49750655ae9SJoel Becker 498c175a518SJoel Becker /* 499c175a518SJoel Becker * Directory blocks also have their own trigger because the 500c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 501c175a518SJoel Becker */ 50213ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 503c175a518SJoel Becker struct buffer_head *bh, 504c175a518SJoel Becker void *data, size_t size) 505c175a518SJoel Becker { 506c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 507c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 508c175a518SJoel Becker 509c175a518SJoel Becker /* 510c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 511c175a518SJoel Becker * must unconditionally compute the ecc data. 512c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 513c175a518SJoel Becker * metaecc is enabled. 514c175a518SJoel Becker */ 515c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 516c175a518SJoel Becker } 517c175a518SJoel Becker 51850655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 51950655ae9SJoel Becker struct buffer_head *bh) 52050655ae9SJoel Becker { 52150655ae9SJoel Becker mlog(ML_ERROR, 52250655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 52350655ae9SJoel Becker "bh->b_blocknr = %llu\n", 52450655ae9SJoel Becker (unsigned long)bh, 52550655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 52650655ae9SJoel Becker 52750655ae9SJoel Becker /* We aren't guaranteed to have the superblock here - but if we 52850655ae9SJoel Becker * don't, it'll just crash. */ 52950655ae9SJoel Becker ocfs2_error(bh->b_assoc_map->host->i_sb, 53050655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 53150655ae9SJoel Becker } 53250655ae9SJoel Becker 53350655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 53450655ae9SJoel Becker .ot_triggers = { 53513ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 53650655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 53750655ae9SJoel Becker }, 53850655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 53950655ae9SJoel Becker }; 54050655ae9SJoel Becker 54150655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 54250655ae9SJoel Becker .ot_triggers = { 54313ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 54450655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 54550655ae9SJoel Becker }, 54650655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 54750655ae9SJoel Becker }; 54850655ae9SJoel Becker 54993c97087STao Ma static struct ocfs2_triggers rb_triggers = { 55093c97087STao Ma .ot_triggers = { 55113ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 55293c97087STao Ma .t_abort = ocfs2_abort_trigger, 55393c97087STao Ma }, 55493c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 55593c97087STao Ma }; 55693c97087STao Ma 55750655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 55850655ae9SJoel Becker .ot_triggers = { 55913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 56050655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 56150655ae9SJoel Becker }, 56250655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 56350655ae9SJoel Becker }; 56450655ae9SJoel Becker 565c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 566c175a518SJoel Becker .ot_triggers = { 56713ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 568c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 569c175a518SJoel Becker }, 570c175a518SJoel Becker }; 571c175a518SJoel Becker 57250655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 57350655ae9SJoel Becker .ot_triggers = { 57413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 57550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 57650655ae9SJoel Becker }, 57750655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 57850655ae9SJoel Becker }; 57950655ae9SJoel Becker 58050655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 58150655ae9SJoel Becker .ot_triggers = { 58213ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 58350655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 58450655ae9SJoel Becker }, 58550655ae9SJoel Becker }; 58650655ae9SJoel Becker 5879b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 5889b7895efSMark Fasheh .ot_triggers = { 58913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 5909b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 5919b7895efSMark Fasheh }, 5929b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 5939b7895efSMark Fasheh }; 5949b7895efSMark Fasheh 5959b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 5969b7895efSMark Fasheh .ot_triggers = { 59713ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 5989b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 5999b7895efSMark Fasheh }, 6009b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6019b7895efSMark Fasheh }; 6029b7895efSMark Fasheh 60350655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6040cf2f763SJoel Becker struct ocfs2_caching_info *ci, 605ccd979bdSMark Fasheh struct buffer_head *bh, 60650655ae9SJoel Becker struct ocfs2_triggers *triggers, 607ccd979bdSMark Fasheh int type) 608ccd979bdSMark Fasheh { 609ccd979bdSMark Fasheh int status; 6100cf2f763SJoel Becker struct ocfs2_super *osb = 6110cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 612ccd979bdSMark Fasheh 6130cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 614ccd979bdSMark Fasheh BUG_ON(!handle); 615ccd979bdSMark Fasheh BUG_ON(!bh); 616ccd979bdSMark Fasheh 617*b4107950STao Ma trace_ocfs2_journal_access( 618*b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 619*b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 620ccd979bdSMark Fasheh 621ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 622ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 623ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 624ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 625ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 626ccd979bdSMark Fasheh BUG(); 627ccd979bdSMark Fasheh } 628ccd979bdSMark Fasheh 6290cf2f763SJoel Becker /* Set the current transaction information on the ci so 630ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6310cf2f763SJoel Becker * on this ci or not. We're protected from the commit 632ccd979bdSMark Fasheh * thread updating the current transaction id until 633ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 634ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6350cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 636ccd979bdSMark Fasheh 6370cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 638ccd979bdSMark Fasheh switch (type) { 639ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 640ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 6412b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 642ccd979bdSMark Fasheh break; 643ccd979bdSMark Fasheh 644ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 6452b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 646ccd979bdSMark Fasheh break; 647ccd979bdSMark Fasheh 648ccd979bdSMark Fasheh default: 649ccd979bdSMark Fasheh status = -EINVAL; 650af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 651ccd979bdSMark Fasheh } 6520cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 65350655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 6540cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 655ccd979bdSMark Fasheh 656ccd979bdSMark Fasheh if (status < 0) 657ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 658ccd979bdSMark Fasheh status, type); 659ccd979bdSMark Fasheh 660ccd979bdSMark Fasheh return status; 661ccd979bdSMark Fasheh } 662ccd979bdSMark Fasheh 6630cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 66450655ae9SJoel Becker struct buffer_head *bh, int type) 66550655ae9SJoel Becker { 6660cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 66750655ae9SJoel Becker } 66850655ae9SJoel Becker 6690cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 67050655ae9SJoel Becker struct buffer_head *bh, int type) 67150655ae9SJoel Becker { 6720cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 67350655ae9SJoel Becker } 67450655ae9SJoel Becker 67593c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 67693c97087STao Ma struct buffer_head *bh, int type) 67793c97087STao Ma { 67893c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 67993c97087STao Ma type); 68093c97087STao Ma } 68193c97087STao Ma 6820cf2f763SJoel Becker int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 68350655ae9SJoel Becker struct buffer_head *bh, int type) 68450655ae9SJoel Becker { 6850cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 68650655ae9SJoel Becker } 68750655ae9SJoel Becker 6880cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 68950655ae9SJoel Becker struct buffer_head *bh, int type) 69050655ae9SJoel Becker { 6910cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 69250655ae9SJoel Becker } 69350655ae9SJoel Becker 6940cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 69550655ae9SJoel Becker struct buffer_head *bh, int type) 69650655ae9SJoel Becker { 6970cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 69850655ae9SJoel Becker } 69950655ae9SJoel Becker 7000cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 70150655ae9SJoel Becker struct buffer_head *bh, int type) 70250655ae9SJoel Becker { 7030cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 70450655ae9SJoel Becker } 70550655ae9SJoel Becker 7060cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7079b7895efSMark Fasheh struct buffer_head *bh, int type) 7089b7895efSMark Fasheh { 7090cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7109b7895efSMark Fasheh } 7119b7895efSMark Fasheh 7120cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7139b7895efSMark Fasheh struct buffer_head *bh, int type) 7149b7895efSMark Fasheh { 7150cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7169b7895efSMark Fasheh } 7179b7895efSMark Fasheh 7180cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 71950655ae9SJoel Becker struct buffer_head *bh, int type) 72050655ae9SJoel Becker { 7210cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 72250655ae9SJoel Becker } 72350655ae9SJoel Becker 724ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 725ccd979bdSMark Fasheh { 726ccd979bdSMark Fasheh int status; 727ccd979bdSMark Fasheh 728*b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 729ccd979bdSMark Fasheh 7302b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 731ec20cec7SJoel Becker BUG_ON(status); 732ccd979bdSMark Fasheh } 733ccd979bdSMark Fasheh 7342b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 735ccd979bdSMark Fasheh 736ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 737ccd979bdSMark Fasheh { 738ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 739d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 740d147b3d6SMark Fasheh 741d147b3d6SMark Fasheh if (osb->osb_commit_interval) 742d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 743ccd979bdSMark Fasheh 744a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 745d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 746ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 7472b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 748ccd979bdSMark Fasheh else 7492b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 750a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 751ccd979bdSMark Fasheh } 752ccd979bdSMark Fasheh 753ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 754ccd979bdSMark Fasheh { 755ccd979bdSMark Fasheh int status = -1; 756ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 757ccd979bdSMark Fasheh journal_t *j_journal = NULL; 758ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 759ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 760ccd979bdSMark Fasheh struct ocfs2_super *osb; 761e63aecb6SMark Fasheh int inode_lock = 0; 762ccd979bdSMark Fasheh 763ccd979bdSMark Fasheh BUG_ON(!journal); 764ccd979bdSMark Fasheh 765ccd979bdSMark Fasheh osb = journal->j_osb; 766ccd979bdSMark Fasheh 767ccd979bdSMark Fasheh /* already have the inode for our journal */ 768ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 769ccd979bdSMark Fasheh osb->slot_num); 770ccd979bdSMark Fasheh if (inode == NULL) { 771ccd979bdSMark Fasheh status = -EACCES; 772ccd979bdSMark Fasheh mlog_errno(status); 773ccd979bdSMark Fasheh goto done; 774ccd979bdSMark Fasheh } 775ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 776ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 777ccd979bdSMark Fasheh iput(inode); 778ccd979bdSMark Fasheh inode = NULL; 779ccd979bdSMark Fasheh status = -EACCES; 780ccd979bdSMark Fasheh goto done; 781ccd979bdSMark Fasheh } 782ccd979bdSMark Fasheh 783ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 784ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 785ccd979bdSMark Fasheh 7866eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 7876eff5790SMark Fasheh * changes in a live cluster so it can be considered an 7886eff5790SMark Fasheh * exception to the rule. */ 789e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 790ccd979bdSMark Fasheh if (status < 0) { 791ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 792ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 793ccd979bdSMark Fasheh goto done; 794ccd979bdSMark Fasheh } 795ccd979bdSMark Fasheh 796e63aecb6SMark Fasheh inode_lock = 1; 797ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 798ccd979bdSMark Fasheh 799ccd979bdSMark Fasheh if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 800ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 801ccd979bdSMark Fasheh inode->i_size); 802ccd979bdSMark Fasheh status = -EINVAL; 803ccd979bdSMark Fasheh goto done; 804ccd979bdSMark Fasheh } 805ccd979bdSMark Fasheh 806*b4107950STao Ma trace_ocfs2_journal_init(inode->i_size, 807*b4107950STao Ma (unsigned long long)inode->i_blocks, 808*b4107950STao Ma OCFS2_I(inode)->ip_clusters); 809ccd979bdSMark Fasheh 810ccd979bdSMark Fasheh /* call the kernels journal init function now */ 8112b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 812ccd979bdSMark Fasheh if (j_journal == NULL) { 813ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 814ccd979bdSMark Fasheh status = -EINVAL; 815ccd979bdSMark Fasheh goto done; 816ccd979bdSMark Fasheh } 817ccd979bdSMark Fasheh 818*b4107950STao Ma trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); 819ccd979bdSMark Fasheh 820ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 821ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 822ccd979bdSMark Fasheh 823ccd979bdSMark Fasheh journal->j_journal = j_journal; 824ccd979bdSMark Fasheh journal->j_inode = inode; 825ccd979bdSMark Fasheh journal->j_bh = bh; 826ccd979bdSMark Fasheh 827ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 828ccd979bdSMark Fasheh 829ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 830ccd979bdSMark Fasheh 831ccd979bdSMark Fasheh status = 0; 832ccd979bdSMark Fasheh done: 833ccd979bdSMark Fasheh if (status < 0) { 834e63aecb6SMark Fasheh if (inode_lock) 835e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 836ccd979bdSMark Fasheh brelse(bh); 837ccd979bdSMark Fasheh if (inode) { 838ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 839ccd979bdSMark Fasheh iput(inode); 840ccd979bdSMark Fasheh } 841ccd979bdSMark Fasheh } 842ccd979bdSMark Fasheh 843ccd979bdSMark Fasheh return status; 844ccd979bdSMark Fasheh } 845ccd979bdSMark Fasheh 846539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 847539d8264SSunil Mushran { 848539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 849539d8264SSunil Mushran } 850539d8264SSunil Mushran 851539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 852539d8264SSunil Mushran { 853539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 854539d8264SSunil Mushran } 855539d8264SSunil Mushran 856ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 857539d8264SSunil Mushran int dirty, int replayed) 858ccd979bdSMark Fasheh { 859ccd979bdSMark Fasheh int status; 860ccd979bdSMark Fasheh unsigned int flags; 861ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 862ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 863ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 864ccd979bdSMark Fasheh 865ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 86610995aa2SJoel Becker 86710995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 86810995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 86910995aa2SJoel Becker * code bug if we mess it up. */ 87010995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 871ccd979bdSMark Fasheh 872ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 873ccd979bdSMark Fasheh if (dirty) 874ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 875ccd979bdSMark Fasheh else 876ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 877ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 878ccd979bdSMark Fasheh 879539d8264SSunil Mushran if (replayed) 880539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 881539d8264SSunil Mushran 88213723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 8838cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 884ccd979bdSMark Fasheh if (status < 0) 885ccd979bdSMark Fasheh mlog_errno(status); 886ccd979bdSMark Fasheh 887ccd979bdSMark Fasheh return status; 888ccd979bdSMark Fasheh } 889ccd979bdSMark Fasheh 890ccd979bdSMark Fasheh /* 891ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 892ccd979bdSMark Fasheh * call. 893ccd979bdSMark Fasheh */ 894ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 895ccd979bdSMark Fasheh { 896ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 897ccd979bdSMark Fasheh int status = 0; 898ccd979bdSMark Fasheh struct inode *inode = NULL; 899ccd979bdSMark Fasheh int num_running_trans = 0; 900ccd979bdSMark Fasheh 901ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 902ccd979bdSMark Fasheh 903ccd979bdSMark Fasheh journal = osb->journal; 904ccd979bdSMark Fasheh if (!journal) 905ccd979bdSMark Fasheh goto done; 906ccd979bdSMark Fasheh 907ccd979bdSMark Fasheh inode = journal->j_inode; 908ccd979bdSMark Fasheh 909ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 910ccd979bdSMark Fasheh goto done; 911ccd979bdSMark Fasheh 9122b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 913ccd979bdSMark Fasheh if (!igrab(inode)) 914ccd979bdSMark Fasheh BUG(); 915ccd979bdSMark Fasheh 916ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 917*b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 918ccd979bdSMark Fasheh 919ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 920ccd979bdSMark Fasheh * release any locks that are still held. 921ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 922ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 923ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 924ccd979bdSMark Fasheh 925ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 926ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 927ccd979bdSMark Fasheh * completely destroy the journal. */ 928ccd979bdSMark Fasheh if (osb->commit_task) { 929ccd979bdSMark Fasheh /* Wait for the commit thread */ 930*b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 931ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 932ccd979bdSMark Fasheh osb->commit_task = NULL; 933ccd979bdSMark Fasheh } 934ccd979bdSMark Fasheh 935ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 936ccd979bdSMark Fasheh 937c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 9382b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 9392b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 9402b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 941c271c5c2SSunil Mushran if (status < 0) 942c271c5c2SSunil Mushran mlog_errno(status); 943c271c5c2SSunil Mushran } 944c271c5c2SSunil Mushran 945c271c5c2SSunil Mushran if (status == 0) { 946c271c5c2SSunil Mushran /* 947c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 948c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 949c271c5c2SSunil Mushran */ 950539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 951ccd979bdSMark Fasheh if (status < 0) 952ccd979bdSMark Fasheh mlog_errno(status); 953c271c5c2SSunil Mushran } 954ccd979bdSMark Fasheh 955ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 9562b4e30fbSJoel Becker jbd2_journal_destroy(journal->j_journal); 957ae0dff68SSunil Mushran journal->j_journal = NULL; 958ccd979bdSMark Fasheh 959ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 960ccd979bdSMark Fasheh 961ccd979bdSMark Fasheh /* unlock our journal */ 962e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 963ccd979bdSMark Fasheh 964ccd979bdSMark Fasheh brelse(journal->j_bh); 965ccd979bdSMark Fasheh journal->j_bh = NULL; 966ccd979bdSMark Fasheh 967ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 968ccd979bdSMark Fasheh 969ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 970ccd979bdSMark Fasheh done: 971ccd979bdSMark Fasheh if (inode) 972ccd979bdSMark Fasheh iput(inode); 973ccd979bdSMark Fasheh } 974ccd979bdSMark Fasheh 975ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 976ccd979bdSMark Fasheh journal_t *journal, 977ccd979bdSMark Fasheh int slot) 978ccd979bdSMark Fasheh { 979ccd979bdSMark Fasheh int olderr; 980ccd979bdSMark Fasheh 9812b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 982ccd979bdSMark Fasheh if (olderr) { 983ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 984ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 985ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 986ccd979bdSMark Fasheh sb->s_id); 987ccd979bdSMark Fasheh 9882b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 9892b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 990ccd979bdSMark Fasheh } 991ccd979bdSMark Fasheh } 992ccd979bdSMark Fasheh 993539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 994ccd979bdSMark Fasheh { 995ccd979bdSMark Fasheh int status = 0; 996ccd979bdSMark Fasheh struct ocfs2_super *osb; 997ccd979bdSMark Fasheh 998b1f3550fSJulia Lawall BUG_ON(!journal); 999ccd979bdSMark Fasheh 1000ccd979bdSMark Fasheh osb = journal->j_osb; 1001ccd979bdSMark Fasheh 10022b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1003ccd979bdSMark Fasheh if (status < 0) { 1004ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1005ccd979bdSMark Fasheh goto done; 1006ccd979bdSMark Fasheh } 1007ccd979bdSMark Fasheh 1008ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1009ccd979bdSMark Fasheh 1010539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1011ccd979bdSMark Fasheh if (status < 0) { 1012ccd979bdSMark Fasheh mlog_errno(status); 1013ccd979bdSMark Fasheh goto done; 1014ccd979bdSMark Fasheh } 1015ccd979bdSMark Fasheh 1016ccd979bdSMark Fasheh /* Launch the commit thread */ 1017c271c5c2SSunil Mushran if (!local) { 1018c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 1019c271c5c2SSunil Mushran "ocfs2cmt"); 1020ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1021ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1022ccd979bdSMark Fasheh osb->commit_task = NULL; 1023c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1024c271c5c2SSunil Mushran "error=%d", status); 1025ccd979bdSMark Fasheh goto done; 1026ccd979bdSMark Fasheh } 1027c271c5c2SSunil Mushran } else 1028c271c5c2SSunil Mushran osb->commit_task = NULL; 1029ccd979bdSMark Fasheh 1030ccd979bdSMark Fasheh done: 1031ccd979bdSMark Fasheh return status; 1032ccd979bdSMark Fasheh } 1033ccd979bdSMark Fasheh 1034ccd979bdSMark Fasheh 1035ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1036ccd979bdSMark Fasheh * mark the journal clean */ 1037ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1038ccd979bdSMark Fasheh { 1039ccd979bdSMark Fasheh int status; 1040ccd979bdSMark Fasheh 1041ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1042ccd979bdSMark Fasheh 10432b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1044ccd979bdSMark Fasheh if (status < 0) { 1045ccd979bdSMark Fasheh mlog_errno(status); 1046ccd979bdSMark Fasheh goto bail; 1047ccd979bdSMark Fasheh } 1048ccd979bdSMark Fasheh 1049539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1050ccd979bdSMark Fasheh if (status < 0) 1051ccd979bdSMark Fasheh mlog_errno(status); 1052ccd979bdSMark Fasheh 1053ccd979bdSMark Fasheh bail: 1054ccd979bdSMark Fasheh return status; 1055ccd979bdSMark Fasheh } 1056ccd979bdSMark Fasheh 1057553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1058553abd04SJoel Becker { 1059553abd04SJoel Becker int empty; 1060553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1061553abd04SJoel Becker 1062553abd04SJoel Becker spin_lock(&osb->osb_lock); 1063553abd04SJoel Becker empty = (rm->rm_used == 0); 1064553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1065553abd04SJoel Becker 1066553abd04SJoel Becker return empty; 1067553abd04SJoel Becker } 1068553abd04SJoel Becker 1069553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1070553abd04SJoel Becker { 1071553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1072553abd04SJoel Becker } 1073553abd04SJoel Becker 1074ccd979bdSMark Fasheh /* 1075ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1076ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1077ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1078ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1079ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1080ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1081ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1082ccd979bdSMark Fasheh * as we have the lock by now :) 1083ccd979bdSMark Fasheh */ 1084ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1085ccd979bdSMark Fasheh { 1086ccd979bdSMark Fasheh int status = 0; 10874f902c37SMark Fasheh int i; 10888110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 10894f902c37SMark Fasheh #define CONCURRENT_JOURNAL_FILL 32ULL 1090ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 1091ccd979bdSMark Fasheh 1092ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 1093ccd979bdSMark Fasheh 10948110b073SMark Fasheh num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); 1095ccd979bdSMark Fasheh v_blkno = 0; 10968110b073SMark Fasheh while (v_blkno < num_blocks) { 1097ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 109849cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1099ccd979bdSMark Fasheh if (status < 0) { 1100ccd979bdSMark Fasheh mlog_errno(status); 1101ccd979bdSMark Fasheh goto bail; 1102ccd979bdSMark Fasheh } 1103ccd979bdSMark Fasheh 1104ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 1105ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 1106ccd979bdSMark Fasheh 1107dd4a2c2bSMark Fasheh /* We are reading journal data which should not 1108dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 1109da1e9098SJoel Becker status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), 1110da1e9098SJoel Becker p_blkno, p_blocks, bhs); 1111ccd979bdSMark Fasheh if (status < 0) { 1112ccd979bdSMark Fasheh mlog_errno(status); 1113ccd979bdSMark Fasheh goto bail; 1114ccd979bdSMark Fasheh } 1115ccd979bdSMark Fasheh 1116ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 1117ccd979bdSMark Fasheh brelse(bhs[i]); 1118ccd979bdSMark Fasheh bhs[i] = NULL; 1119ccd979bdSMark Fasheh } 1120ccd979bdSMark Fasheh 1121ccd979bdSMark Fasheh v_blkno += p_blocks; 1122ccd979bdSMark Fasheh } 1123ccd979bdSMark Fasheh 1124ccd979bdSMark Fasheh bail: 1125ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 1126ccd979bdSMark Fasheh brelse(bhs[i]); 1127ccd979bdSMark Fasheh return status; 1128ccd979bdSMark Fasheh } 1129ccd979bdSMark Fasheh 1130ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1131ccd979bdSMark Fasheh struct list_head lri_list; 1132ccd979bdSMark Fasheh int lri_slot; 1133ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1134ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 11352205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1136ccd979bdSMark Fasheh }; 1137ccd979bdSMark Fasheh 1138ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1139ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1140ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1141ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1142ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1143ccd979bdSMark Fasheh * recovering nodes into account. 1144ccd979bdSMark Fasheh * 1145ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1146ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1147ccd979bdSMark Fasheh */ 1148c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1149ccd979bdSMark Fasheh { 1150*b4107950STao Ma int ret = 0; 1151c4028958SDavid Howells struct ocfs2_journal *journal = 1152c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1153c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1154ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1155800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 11562205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1157ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1158ccd979bdSMark Fasheh 1159*b4107950STao Ma trace_ocfs2_complete_recovery( 1160*b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1161ccd979bdSMark Fasheh 1162ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1163ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1164ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1165ccd979bdSMark Fasheh 1166800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1167ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1168ccd979bdSMark Fasheh 116919ece546SJan Kara ocfs2_wait_on_quotas(osb); 117019ece546SJan Kara 1171ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1172*b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1173*b4107950STao Ma qrec = item->lri_qrec; 1174ccd979bdSMark Fasheh 1175*b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1176*b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1177*b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1178*b4107950STao Ma qrec); 1179*b4107950STao Ma 1180*b4107950STao Ma if (la_dinode) { 1181ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1182ccd979bdSMark Fasheh la_dinode); 1183ccd979bdSMark Fasheh if (ret < 0) 1184ccd979bdSMark Fasheh mlog_errno(ret); 1185ccd979bdSMark Fasheh 1186ccd979bdSMark Fasheh kfree(la_dinode); 1187ccd979bdSMark Fasheh } 1188ccd979bdSMark Fasheh 1189ccd979bdSMark Fasheh if (tl_dinode) { 1190ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1191ccd979bdSMark Fasheh tl_dinode); 1192ccd979bdSMark Fasheh if (ret < 0) 1193ccd979bdSMark Fasheh mlog_errno(ret); 1194ccd979bdSMark Fasheh 1195ccd979bdSMark Fasheh kfree(tl_dinode); 1196ccd979bdSMark Fasheh } 1197ccd979bdSMark Fasheh 1198ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 1199ccd979bdSMark Fasheh if (ret < 0) 1200ccd979bdSMark Fasheh mlog_errno(ret); 1201ccd979bdSMark Fasheh 12022205363dSJan Kara if (qrec) { 12032205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 12042205363dSJan Kara item->lri_slot); 12052205363dSJan Kara if (ret < 0) 12062205363dSJan Kara mlog_errno(ret); 12072205363dSJan Kara /* Recovery info is already freed now */ 12082205363dSJan Kara } 12092205363dSJan Kara 1210ccd979bdSMark Fasheh kfree(item); 1211ccd979bdSMark Fasheh } 1212ccd979bdSMark Fasheh 1213*b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1214ccd979bdSMark Fasheh } 1215ccd979bdSMark Fasheh 1216ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1217ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1218ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1219ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1220ccd979bdSMark Fasheh int slot_num, 1221ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 12222205363dSJan Kara struct ocfs2_dinode *tl_dinode, 12232205363dSJan Kara struct ocfs2_quota_recovery *qrec) 1224ccd979bdSMark Fasheh { 1225ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1226ccd979bdSMark Fasheh 1227afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1228ccd979bdSMark Fasheh if (!item) { 1229ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1230ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1231ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1232ccd979bdSMark Fasheh if (la_dinode) 1233ccd979bdSMark Fasheh kfree(la_dinode); 1234ccd979bdSMark Fasheh 1235ccd979bdSMark Fasheh if (tl_dinode) 1236ccd979bdSMark Fasheh kfree(tl_dinode); 1237ccd979bdSMark Fasheh 12382205363dSJan Kara if (qrec) 12392205363dSJan Kara ocfs2_free_quota_recovery(qrec); 12402205363dSJan Kara 1241ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1242ccd979bdSMark Fasheh return; 1243ccd979bdSMark Fasheh } 1244ccd979bdSMark Fasheh 1245ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1246ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1247ccd979bdSMark Fasheh item->lri_slot = slot_num; 1248ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 12492205363dSJan Kara item->lri_qrec = qrec; 1250ccd979bdSMark Fasheh 1251ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1252ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 1253ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 1254ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1255ccd979bdSMark Fasheh } 1256ccd979bdSMark Fasheh 1257ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 12589140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1259ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1260ccd979bdSMark Fasheh { 1261ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1262ccd979bdSMark Fasheh 12639140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 12649140db04SSrinivas Eeda * that */ 12659140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 12669140db04SSrinivas Eeda osb->local_alloc_copy, NULL, NULL); 1267ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1268ccd979bdSMark Fasheh 1269ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 1270ccd979bdSMark Fasheh osb->dirty = 0; 12719140db04SSrinivas Eeda 12729140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 12739140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 12749140db04SSrinivas Eeda ocfs2_queue_replay_slots(osb); 12759140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1276ccd979bdSMark Fasheh } 1277ccd979bdSMark Fasheh 12782205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 12792205363dSJan Kara { 12802205363dSJan Kara if (osb->quota_rec) { 12812205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 12822205363dSJan Kara osb->slot_num, 12832205363dSJan Kara NULL, 12842205363dSJan Kara NULL, 12852205363dSJan Kara osb->quota_rec); 12862205363dSJan Kara osb->quota_rec = NULL; 12872205363dSJan Kara } 12882205363dSJan Kara } 12892205363dSJan Kara 1290ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1291ccd979bdSMark Fasheh { 12922205363dSJan Kara int status, node_num, slot_num; 1293ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1294553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 12952205363dSJan Kara int *rm_quota = NULL; 12962205363dSJan Kara int rm_quota_used = 0, i; 12972205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1298ccd979bdSMark Fasheh 1299ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1300ccd979bdSMark Fasheh if (status < 0) { 1301ccd979bdSMark Fasheh goto bail; 1302ccd979bdSMark Fasheh } 1303ccd979bdSMark Fasheh 13042205363dSJan Kara rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); 13052205363dSJan Kara if (!rm_quota) { 13062205363dSJan Kara status = -ENOMEM; 13072205363dSJan Kara goto bail; 13082205363dSJan Kara } 1309ccd979bdSMark Fasheh restart: 1310ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1311ccd979bdSMark Fasheh if (status < 0) { 1312ccd979bdSMark Fasheh mlog_errno(status); 1313ccd979bdSMark Fasheh goto bail; 1314ccd979bdSMark Fasheh } 1315ccd979bdSMark Fasheh 13169140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 13179140db04SSrinivas Eeda if (status < 0) 13189140db04SSrinivas Eeda mlog_errno(status); 13199140db04SSrinivas Eeda 13209140db04SSrinivas Eeda /* queue recovery for our own slot */ 13219140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 13229140db04SSrinivas Eeda NULL, NULL); 13239140db04SSrinivas Eeda 1324553abd04SJoel Becker spin_lock(&osb->osb_lock); 1325553abd04SJoel Becker while (rm->rm_used) { 1326553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1327553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1328553abd04SJoel Becker node_num = rm->rm_entries[0]; 1329553abd04SJoel Becker spin_unlock(&osb->osb_lock); 13302205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1331*b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 13322205363dSJan Kara if (slot_num == -ENOENT) { 13332205363dSJan Kara status = 0; 13342205363dSJan Kara goto skip_recovery; 13352205363dSJan Kara } 1336ccd979bdSMark Fasheh 13372205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 13382205363dSJan Kara * immediately because we have to obtain cluster locks from 13392205363dSJan Kara * quota files and we also don't want to just skip it because 13402205363dSJan Kara * then quota usage would be out of sync until some node takes 13412205363dSJan Kara * the slot. So we remember which nodes need quota recovery 13422205363dSJan Kara * and when everything else is done, we recover quotas. */ 13432205363dSJan Kara for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); 13442205363dSJan Kara if (i == rm_quota_used) 13452205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 13462205363dSJan Kara 13472205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 13482205363dSJan Kara skip_recovery: 1349553abd04SJoel Becker if (!status) { 1350553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1351553abd04SJoel Becker } else { 1352ccd979bdSMark Fasheh mlog(ML_ERROR, 1353ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1354ccd979bdSMark Fasheh status, node_num, 1355ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1356ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1357ccd979bdSMark Fasheh } 1358ccd979bdSMark Fasheh 1359553abd04SJoel Becker spin_lock(&osb->osb_lock); 1360ccd979bdSMark Fasheh } 1361553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1362*b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1363553abd04SJoel Becker 1364539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1365539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1366539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1367539d8264SSunil Mushran if (status < 0) 1368539d8264SSunil Mushran mlog_errno(status); 1369539d8264SSunil Mushran 13702205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 13712205363dSJan Kara * superblock lock so that noone can start using the slot (and crash) 13722205363dSJan Kara * before we recover it */ 13732205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 13742205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 13752205363dSJan Kara if (IS_ERR(qrec)) { 13762205363dSJan Kara status = PTR_ERR(qrec); 13772205363dSJan Kara mlog_errno(status); 13782205363dSJan Kara continue; 13792205363dSJan Kara } 13802205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], 13812205363dSJan Kara NULL, NULL, qrec); 13822205363dSJan Kara } 13832205363dSJan Kara 1384ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1385ccd979bdSMark Fasheh 13869140db04SSrinivas Eeda /* queue recovery for offline slots */ 13879140db04SSrinivas Eeda ocfs2_queue_replay_slots(osb); 1388ccd979bdSMark Fasheh 1389ccd979bdSMark Fasheh bail: 1390c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1391553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1392c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1393ccd979bdSMark Fasheh goto restart; 1394ccd979bdSMark Fasheh } 1395ccd979bdSMark Fasheh 13969140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1397ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1398ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1399ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1400ccd979bdSMark Fasheh 1401c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1402ccd979bdSMark Fasheh 14032205363dSJan Kara if (rm_quota) 14042205363dSJan Kara kfree(rm_quota); 14052205363dSJan Kara 1406ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1407ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1408ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1409ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1410ccd979bdSMark Fasheh return status; 1411ccd979bdSMark Fasheh } 1412ccd979bdSMark Fasheh 1413ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1414ccd979bdSMark Fasheh { 1415c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1416*b4107950STao Ma 1417*b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1418*b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1419*b4107950STao Ma osb->disable_recovery ? 1420*b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1421*b4107950STao Ma 1422ccd979bdSMark Fasheh if (osb->disable_recovery) 1423ccd979bdSMark Fasheh goto out; 1424ccd979bdSMark Fasheh 1425ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1426ccd979bdSMark Fasheh goto out; 1427ccd979bdSMark Fasheh 1428ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 142978427043SMark Fasheh "ocfs2rec"); 1430ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1431ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1432ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1433ccd979bdSMark Fasheh } 1434ccd979bdSMark Fasheh 1435ccd979bdSMark Fasheh out: 1436c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1437ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1438ccd979bdSMark Fasheh } 1439ccd979bdSMark Fasheh 1440539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1441539d8264SSunil Mushran int slot_num, 1442539d8264SSunil Mushran struct buffer_head **bh, 1443539d8264SSunil Mushran struct inode **ret_inode) 1444539d8264SSunil Mushran { 1445539d8264SSunil Mushran int status = -EACCES; 1446539d8264SSunil Mushran struct inode *inode = NULL; 1447539d8264SSunil Mushran 1448539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1449539d8264SSunil Mushran 1450539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1451539d8264SSunil Mushran slot_num); 1452539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1453539d8264SSunil Mushran mlog_errno(status); 1454539d8264SSunil Mushran goto bail; 1455539d8264SSunil Mushran } 1456539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1457539d8264SSunil Mushran 1458b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1459539d8264SSunil Mushran if (status < 0) { 1460539d8264SSunil Mushran mlog_errno(status); 1461539d8264SSunil Mushran goto bail; 1462539d8264SSunil Mushran } 1463539d8264SSunil Mushran 1464539d8264SSunil Mushran status = 0; 1465539d8264SSunil Mushran 1466539d8264SSunil Mushran bail: 1467539d8264SSunil Mushran if (inode) { 1468539d8264SSunil Mushran if (status || !ret_inode) 1469539d8264SSunil Mushran iput(inode); 1470539d8264SSunil Mushran else 1471539d8264SSunil Mushran *ret_inode = inode; 1472539d8264SSunil Mushran } 1473539d8264SSunil Mushran return status; 1474539d8264SSunil Mushran } 1475539d8264SSunil Mushran 1476ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1477ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1478ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1479ccd979bdSMark Fasheh int node_num, 1480ccd979bdSMark Fasheh int slot_num) 1481ccd979bdSMark Fasheh { 1482ccd979bdSMark Fasheh int status; 1483ccd979bdSMark Fasheh int got_lock = 0; 1484ccd979bdSMark Fasheh unsigned int flags; 1485ccd979bdSMark Fasheh struct inode *inode = NULL; 1486ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1487ccd979bdSMark Fasheh journal_t *journal = NULL; 1488ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1489539d8264SSunil Mushran u32 slot_reco_gen; 1490ccd979bdSMark Fasheh 1491539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1492539d8264SSunil Mushran if (status) { 1493ccd979bdSMark Fasheh mlog_errno(status); 1494ccd979bdSMark Fasheh goto done; 1495ccd979bdSMark Fasheh } 1496539d8264SSunil Mushran 1497539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1498539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1499539d8264SSunil Mushran brelse(bh); 1500539d8264SSunil Mushran bh = NULL; 1501539d8264SSunil Mushran 1502539d8264SSunil Mushran /* 1503539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1504539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1505539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1506539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1507539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1508539d8264SSunil Mushran * If not, it needs recovery. 1509539d8264SSunil Mushran */ 1510539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1511*b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1512539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1513539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1514539d8264SSunil Mushran status = -EBUSY; 1515ccd979bdSMark Fasheh goto done; 1516ccd979bdSMark Fasheh } 1517539d8264SSunil Mushran 1518539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1519ccd979bdSMark Fasheh 1520e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1521ccd979bdSMark Fasheh if (status < 0) { 1522*b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1523ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1524ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1525ccd979bdSMark Fasheh goto done; 1526ccd979bdSMark Fasheh } 1527ccd979bdSMark Fasheh got_lock = 1; 1528ccd979bdSMark Fasheh 1529ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1530ccd979bdSMark Fasheh 1531ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1532539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1533ccd979bdSMark Fasheh 1534ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1535*b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1536539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1537539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1538ccd979bdSMark Fasheh goto done; 1539ccd979bdSMark Fasheh } 1540ccd979bdSMark Fasheh 15419140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 15429140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 15439140db04SSrinivas Eeda 1544ccd979bdSMark Fasheh mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 1545ccd979bdSMark Fasheh node_num, slot_num, 1546ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1547ccd979bdSMark Fasheh 1548ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1549ccd979bdSMark Fasheh 1550ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1551ccd979bdSMark Fasheh if (status < 0) { 1552ccd979bdSMark Fasheh mlog_errno(status); 1553ccd979bdSMark Fasheh goto done; 1554ccd979bdSMark Fasheh } 1555ccd979bdSMark Fasheh 15562b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1557ccd979bdSMark Fasheh if (journal == NULL) { 1558ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1559ccd979bdSMark Fasheh status = -EIO; 1560ccd979bdSMark Fasheh goto done; 1561ccd979bdSMark Fasheh } 1562ccd979bdSMark Fasheh 15632b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1564ccd979bdSMark Fasheh if (status < 0) { 1565ccd979bdSMark Fasheh mlog_errno(status); 1566ccd979bdSMark Fasheh if (!igrab(inode)) 1567ccd979bdSMark Fasheh BUG(); 15682b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1569ccd979bdSMark Fasheh goto done; 1570ccd979bdSMark Fasheh } 1571ccd979bdSMark Fasheh 1572ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1573ccd979bdSMark Fasheh 1574ccd979bdSMark Fasheh /* wipe the journal */ 15752b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 15762b4e30fbSJoel Becker status = jbd2_journal_flush(journal); 15772b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1578ccd979bdSMark Fasheh if (status < 0) 1579ccd979bdSMark Fasheh mlog_errno(status); 1580ccd979bdSMark Fasheh 1581ccd979bdSMark Fasheh /* This will mark the node clean */ 1582ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1583ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1584ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1585ccd979bdSMark Fasheh 1586539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1587539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1588539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1589539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1590539d8264SSunil Mushran 159113723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 15928cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1593ccd979bdSMark Fasheh if (status < 0) 1594ccd979bdSMark Fasheh mlog_errno(status); 1595ccd979bdSMark Fasheh 1596ccd979bdSMark Fasheh if (!igrab(inode)) 1597ccd979bdSMark Fasheh BUG(); 1598ccd979bdSMark Fasheh 15992b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1600ccd979bdSMark Fasheh 1601ccd979bdSMark Fasheh done: 1602ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1603ccd979bdSMark Fasheh if (got_lock) 1604e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1605ccd979bdSMark Fasheh 1606ccd979bdSMark Fasheh if (inode) 1607ccd979bdSMark Fasheh iput(inode); 1608ccd979bdSMark Fasheh 1609ccd979bdSMark Fasheh brelse(bh); 1610ccd979bdSMark Fasheh 1611ccd979bdSMark Fasheh return status; 1612ccd979bdSMark Fasheh } 1613ccd979bdSMark Fasheh 1614ccd979bdSMark Fasheh /* 1615ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1616ccd979bdSMark Fasheh * - Replay it's journal 1617ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1618ccd979bdSMark Fasheh * - Stamp a clean truncate log 1619ccd979bdSMark Fasheh * - Mark the node clean 1620ccd979bdSMark Fasheh * 1621ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1622ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1623ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1624ccd979bdSMark Fasheh * far less concerning. 1625ccd979bdSMark Fasheh */ 1626ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 16272205363dSJan Kara int node_num, int slot_num) 1628ccd979bdSMark Fasheh { 1629ccd979bdSMark Fasheh int status = 0; 1630ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1631ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1632ccd979bdSMark Fasheh 1633*b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1634ccd979bdSMark Fasheh 1635ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1636ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1637ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1638ccd979bdSMark Fasheh 1639ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1640ccd979bdSMark Fasheh if (status < 0) { 1641539d8264SSunil Mushran if (status == -EBUSY) { 1642*b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1643539d8264SSunil Mushran status = 0; 1644539d8264SSunil Mushran goto done; 1645539d8264SSunil Mushran } 1646ccd979bdSMark Fasheh mlog_errno(status); 1647ccd979bdSMark Fasheh goto done; 1648ccd979bdSMark Fasheh } 1649ccd979bdSMark Fasheh 1650ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1651ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1652ccd979bdSMark Fasheh if (status < 0) { 1653ccd979bdSMark Fasheh mlog_errno(status); 1654ccd979bdSMark Fasheh goto done; 1655ccd979bdSMark Fasheh } 1656ccd979bdSMark Fasheh 1657ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1658ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1659ccd979bdSMark Fasheh * recovery. */ 1660ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1661ccd979bdSMark Fasheh if (status < 0) 1662ccd979bdSMark Fasheh mlog_errno(status); 1663ccd979bdSMark Fasheh 1664ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1665ccd979bdSMark Fasheh * harmful place to get an error... */ 16668e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1667ccd979bdSMark Fasheh if (status < 0) 1668ccd979bdSMark Fasheh mlog_errno(status); 1669ccd979bdSMark Fasheh 1670ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1671ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 16722205363dSJan Kara tl_copy, NULL); 1673ccd979bdSMark Fasheh 1674ccd979bdSMark Fasheh status = 0; 1675ccd979bdSMark Fasheh done: 1676ccd979bdSMark Fasheh 1677ccd979bdSMark Fasheh return status; 1678ccd979bdSMark Fasheh } 1679ccd979bdSMark Fasheh 1680ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1681ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1682ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1683ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1684ccd979bdSMark Fasheh int slot_num) 1685ccd979bdSMark Fasheh { 1686ccd979bdSMark Fasheh int status, flags; 1687ccd979bdSMark Fasheh struct inode *inode = NULL; 1688ccd979bdSMark Fasheh 1689ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1690ccd979bdSMark Fasheh slot_num); 1691ccd979bdSMark Fasheh if (inode == NULL) { 1692ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1693ccd979bdSMark Fasheh status = -EACCES; 1694ccd979bdSMark Fasheh goto bail; 1695ccd979bdSMark Fasheh } 1696ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1697ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1698ccd979bdSMark Fasheh iput(inode); 1699ccd979bdSMark Fasheh inode = NULL; 1700ccd979bdSMark Fasheh status = -EACCES; 1701ccd979bdSMark Fasheh goto bail; 1702ccd979bdSMark Fasheh } 1703ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1704ccd979bdSMark Fasheh 1705ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1706e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1707ccd979bdSMark Fasheh if (status < 0) { 1708ccd979bdSMark Fasheh if (status != -EAGAIN) 1709ccd979bdSMark Fasheh mlog_errno(status); 1710ccd979bdSMark Fasheh goto bail; 1711ccd979bdSMark Fasheh } 1712ccd979bdSMark Fasheh 1713e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1714ccd979bdSMark Fasheh bail: 1715ccd979bdSMark Fasheh if (inode) 1716ccd979bdSMark Fasheh iput(inode); 1717ccd979bdSMark Fasheh 1718ccd979bdSMark Fasheh return status; 1719ccd979bdSMark Fasheh } 1720ccd979bdSMark Fasheh 1721ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1722ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1723ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1724ccd979bdSMark Fasheh { 1725d85b20e4SJoel Becker unsigned int node_num; 1726d85b20e4SJoel Becker int status, i; 1727a1af7d15SMark Fasheh u32 gen; 1728539d8264SSunil Mushran struct buffer_head *bh = NULL; 1729539d8264SSunil Mushran struct ocfs2_dinode *di; 1730ccd979bdSMark Fasheh 1731ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1732ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1733ccd979bdSMark Fasheh 1734d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1735539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1736539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1737539d8264SSunil Mushran if (status) { 1738539d8264SSunil Mushran mlog_errno(status); 1739539d8264SSunil Mushran goto bail; 1740539d8264SSunil Mushran } 1741539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1742a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1743539d8264SSunil Mushran brelse(bh); 1744539d8264SSunil Mushran bh = NULL; 1745539d8264SSunil Mushran 1746a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1747a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1748a1af7d15SMark Fasheh 1749*b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1750539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1751539d8264SSunil Mushran 1752a1af7d15SMark Fasheh if (i == osb->slot_num) { 1753a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1754ccd979bdSMark Fasheh continue; 1755a1af7d15SMark Fasheh } 1756d85b20e4SJoel Becker 1757d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1758a1af7d15SMark Fasheh if (status == -ENOENT) { 1759a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1760ccd979bdSMark Fasheh continue; 1761a1af7d15SMark Fasheh } 1762ccd979bdSMark Fasheh 1763a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1764a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1765ccd979bdSMark Fasheh continue; 1766a1af7d15SMark Fasheh } 1767d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1768ccd979bdSMark Fasheh 1769ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1770ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1771ccd979bdSMark Fasheh * file here to test if he's alive. */ 1772ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1773ccd979bdSMark Fasheh if (!status) { 1774ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1775ccd979bdSMark Fasheh * the recovery thread can't race us on 1776ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1777ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1778ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1779ccd979bdSMark Fasheh mlog_errno(status); 1780ccd979bdSMark Fasheh goto bail; 1781ccd979bdSMark Fasheh } 1782ccd979bdSMark Fasheh } 1783ccd979bdSMark Fasheh 1784ccd979bdSMark Fasheh status = 0; 1785ccd979bdSMark Fasheh bail: 1786ccd979bdSMark Fasheh return status; 1787ccd979bdSMark Fasheh } 1788ccd979bdSMark Fasheh 178983273932SSrinivas Eeda /* 179083273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 179183273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 179283273932SSrinivas Eeda * same time. 179383273932SSrinivas Eeda */ 179483273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 179583273932SSrinivas Eeda { 179683273932SSrinivas Eeda unsigned long time; 179783273932SSrinivas Eeda 179883273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 179983273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 180083273932SSrinivas Eeda return msecs_to_jiffies(time); 180183273932SSrinivas Eeda } 180283273932SSrinivas Eeda 180383273932SSrinivas Eeda /* 180483273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 180583273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 180683273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 180783273932SSrinivas Eeda * 180883273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 180983273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 181083273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 181183273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 181283273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 181383273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 181483273932SSrinivas Eeda * sequence number in the LVB. 181583273932SSrinivas Eeda */ 181683273932SSrinivas Eeda void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 181783273932SSrinivas Eeda { 181883273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 181983273932SSrinivas Eeda int status, i; 182083273932SSrinivas Eeda u32 seqno = 0; 182183273932SSrinivas Eeda 182283273932SSrinivas Eeda os = &osb->osb_orphan_scan; 182383273932SSrinivas Eeda 1824692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1825692684e1SSunil Mushran goto out; 1826692684e1SSunil Mushran 1827*b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1828*b4107950STao Ma atomic_read(&os->os_state)); 1829*b4107950STao Ma 1830df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 183183273932SSrinivas Eeda if (status < 0) { 183283273932SSrinivas Eeda if (status != -EAGAIN) 183383273932SSrinivas Eeda mlog_errno(status); 183483273932SSrinivas Eeda goto out; 183583273932SSrinivas Eeda } 183683273932SSrinivas Eeda 1837692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1838692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1839692684e1SSunil Mushran goto unlock; 1840692684e1SSunil Mushran 184183273932SSrinivas Eeda if (os->os_seqno != seqno) { 184283273932SSrinivas Eeda os->os_seqno = seqno; 184383273932SSrinivas Eeda goto unlock; 184483273932SSrinivas Eeda } 184583273932SSrinivas Eeda 184683273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 184783273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 184883273932SSrinivas Eeda NULL); 184983273932SSrinivas Eeda /* 185083273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 185183273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 185283273932SSrinivas Eeda */ 185383273932SSrinivas Eeda seqno++; 185415633a22SSrinivas Eeda os->os_count++; 185515633a22SSrinivas Eeda os->os_scantime = CURRENT_TIME; 185683273932SSrinivas Eeda unlock: 1857df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 185883273932SSrinivas Eeda out: 1859*b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1860*b4107950STao Ma atomic_read(&os->os_state)); 186183273932SSrinivas Eeda return; 186283273932SSrinivas Eeda } 186383273932SSrinivas Eeda 186483273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 186583273932SSrinivas Eeda void ocfs2_orphan_scan_work(struct work_struct *work) 186683273932SSrinivas Eeda { 186783273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 186883273932SSrinivas Eeda struct ocfs2_super *osb; 186983273932SSrinivas Eeda 187083273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 187183273932SSrinivas Eeda os_orphan_scan_work.work); 187283273932SSrinivas Eeda osb = os->os_osb; 187383273932SSrinivas Eeda 187483273932SSrinivas Eeda mutex_lock(&os->os_lock); 187583273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 1876692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 187740f165f4STao Ma queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 187883273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 187983273932SSrinivas Eeda mutex_unlock(&os->os_lock); 188083273932SSrinivas Eeda } 188183273932SSrinivas Eeda 188283273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 188383273932SSrinivas Eeda { 188483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 188583273932SSrinivas Eeda 188683273932SSrinivas Eeda os = &osb->osb_orphan_scan; 1887df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1888692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 188983273932SSrinivas Eeda mutex_lock(&os->os_lock); 189083273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 189183273932SSrinivas Eeda mutex_unlock(&os->os_lock); 189283273932SSrinivas Eeda } 1893df152c24SSunil Mushran } 189483273932SSrinivas Eeda 1895df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 189683273932SSrinivas Eeda { 189783273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 189883273932SSrinivas Eeda 189983273932SSrinivas Eeda os = &osb->osb_orphan_scan; 190083273932SSrinivas Eeda os->os_osb = osb; 190115633a22SSrinivas Eeda os->os_count = 0; 19023211949fSSunil Mushran os->os_seqno = 0; 190383273932SSrinivas Eeda mutex_init(&os->os_lock); 1904df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 19058b712cd5SJeff Mahoney } 190683273932SSrinivas Eeda 19078b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 19088b712cd5SJeff Mahoney { 19098b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 19108b712cd5SJeff Mahoney 19118b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 19128b712cd5SJeff Mahoney os->os_scantime = CURRENT_TIME; 1913df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 1914df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 1915df152c24SSunil Mushran else { 1916df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 191740f165f4STao Ma queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 191883273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 1919df152c24SSunil Mushran } 192083273932SSrinivas Eeda } 192183273932SSrinivas Eeda 19225eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 19235eae5b96SMark Fasheh struct inode *head; 19245eae5b96SMark Fasheh struct ocfs2_super *osb; 19255eae5b96SMark Fasheh }; 19265eae5b96SMark Fasheh 19275eae5b96SMark Fasheh static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, 19285eae5b96SMark Fasheh loff_t pos, u64 ino, unsigned type) 19295eae5b96SMark Fasheh { 19305eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv *p = priv; 19315eae5b96SMark Fasheh struct inode *iter; 19325eae5b96SMark Fasheh 19335eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 19345eae5b96SMark Fasheh return 0; 19355eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 19365eae5b96SMark Fasheh return 0; 19375eae5b96SMark Fasheh 19385eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 19395eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 19405fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 19415eae5b96SMark Fasheh if (IS_ERR(iter)) 19425eae5b96SMark Fasheh return 0; 19435eae5b96SMark Fasheh 1944*b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 19455eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 19465eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 19475eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 19485eae5b96SMark Fasheh p->head = iter; 19495eae5b96SMark Fasheh 19505eae5b96SMark Fasheh return 0; 19515eae5b96SMark Fasheh } 19525eae5b96SMark Fasheh 1953b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1954b4df6ed8SMark Fasheh int slot, 1955b4df6ed8SMark Fasheh struct inode **head) 1956ccd979bdSMark Fasheh { 1957b4df6ed8SMark Fasheh int status; 1958ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 19595eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv priv; 19605eae5b96SMark Fasheh loff_t pos = 0; 19615eae5b96SMark Fasheh 19625eae5b96SMark Fasheh priv.osb = osb; 19635eae5b96SMark Fasheh priv.head = *head; 1964ccd979bdSMark Fasheh 1965ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1966ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 1967ccd979bdSMark Fasheh slot); 1968ccd979bdSMark Fasheh if (!orphan_dir_inode) { 1969ccd979bdSMark Fasheh status = -ENOENT; 1970ccd979bdSMark Fasheh mlog_errno(status); 1971b4df6ed8SMark Fasheh return status; 1972ccd979bdSMark Fasheh } 1973ccd979bdSMark Fasheh 19741b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 1975e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 1976ccd979bdSMark Fasheh if (status < 0) { 1977ccd979bdSMark Fasheh mlog_errno(status); 1978ccd979bdSMark Fasheh goto out; 1979ccd979bdSMark Fasheh } 1980ccd979bdSMark Fasheh 19815eae5b96SMark Fasheh status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, 19825eae5b96SMark Fasheh ocfs2_orphan_filldir); 19835eae5b96SMark Fasheh if (status) { 1984ccd979bdSMark Fasheh mlog_errno(status); 1985a86370fbSMark Fasheh goto out_cluster; 1986ccd979bdSMark Fasheh } 1987ccd979bdSMark Fasheh 19885eae5b96SMark Fasheh *head = priv.head; 1989ccd979bdSMark Fasheh 1990a86370fbSMark Fasheh out_cluster: 1991e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 1992b4df6ed8SMark Fasheh out: 1993b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 1994ccd979bdSMark Fasheh iput(orphan_dir_inode); 1995b4df6ed8SMark Fasheh return status; 1996b4df6ed8SMark Fasheh } 1997b4df6ed8SMark Fasheh 1998b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1999b4df6ed8SMark Fasheh int slot) 2000b4df6ed8SMark Fasheh { 2001b4df6ed8SMark Fasheh int ret; 2002b4df6ed8SMark Fasheh 2003b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2004b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2005b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2006b4df6ed8SMark Fasheh return ret; 2007b4df6ed8SMark Fasheh } 2008b4df6ed8SMark Fasheh 2009b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2010b4df6ed8SMark Fasheh int slot) 2011b4df6ed8SMark Fasheh { 2012b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2013b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2014b4df6ed8SMark Fasheh * know to quit early. */ 2015b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2016b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2017b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2018b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2019b4df6ed8SMark Fasheh * them. */ 2020b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2021b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2022b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2023b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2024b4df6ed8SMark Fasheh } 2025b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2026b4df6ed8SMark Fasheh } 2027b4df6ed8SMark Fasheh 2028b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2029b4df6ed8SMark Fasheh int slot) 2030b4df6ed8SMark Fasheh { 2031b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2032b4df6ed8SMark Fasheh } 2033b4df6ed8SMark Fasheh 2034b4df6ed8SMark Fasheh /* 2035b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2036b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2037b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2038b4df6ed8SMark Fasheh * (most) of the rest of the work. 2039b4df6ed8SMark Fasheh * 2040b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2041b4df6ed8SMark Fasheh * couple of extra considerations. 2042b4df6ed8SMark Fasheh * 2043b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2044b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2045b4df6ed8SMark Fasheh * an invalid inode. 2046b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2047b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2048b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2049b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2050b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2051b4df6ed8SMark Fasheh */ 2052b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2053b4df6ed8SMark Fasheh int slot) 2054b4df6ed8SMark Fasheh { 2055b4df6ed8SMark Fasheh int ret = 0; 2056b4df6ed8SMark Fasheh struct inode *inode = NULL; 2057b4df6ed8SMark Fasheh struct inode *iter; 2058b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2059b4df6ed8SMark Fasheh 2060*b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2061b4df6ed8SMark Fasheh 2062b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 2063b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 2064b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2065b4df6ed8SMark Fasheh 2066b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2067b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2068b4df6ed8SMark Fasheh if (ret) 2069b4df6ed8SMark Fasheh mlog_errno(ret); 2070ccd979bdSMark Fasheh 2071ccd979bdSMark Fasheh while (inode) { 2072ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2073*b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2074*b4107950STao Ma (unsigned long long)oi->ip_blkno); 2075ccd979bdSMark Fasheh 2076ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2077ccd979bdSMark Fasheh 2078ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 207934d024f8SMark Fasheh /* The remote delete code may have set these on the 208034d024f8SMark Fasheh * assumption that the other node would wipe them 208134d024f8SMark Fasheh * successfully. If they are still in the node's 208234d024f8SMark Fasheh * orphan dir, we need to reset that state. */ 2083ccd979bdSMark Fasheh oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 2084ccd979bdSMark Fasheh 2085ccd979bdSMark Fasheh /* Set the proper information to get us going into 2086ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 2087ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 2088ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 2089ccd979bdSMark Fasheh 2090ccd979bdSMark Fasheh iput(inode); 2091ccd979bdSMark Fasheh 2092ccd979bdSMark Fasheh inode = iter; 2093ccd979bdSMark Fasheh } 2094ccd979bdSMark Fasheh 2095b4df6ed8SMark Fasheh return ret; 2096ccd979bdSMark Fasheh } 2097ccd979bdSMark Fasheh 209819ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2099ccd979bdSMark Fasheh { 2100ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2101ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2102ccd979bdSMark Fasheh * or DISABLED. */ 2103ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 210419ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 210519ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2106ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2107ccd979bdSMark Fasheh 2108ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2109ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2110ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2111ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2112*b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2113ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2114ccd979bdSMark Fasheh return -EBUSY; 2115ccd979bdSMark Fasheh } 2116ccd979bdSMark Fasheh 2117ccd979bdSMark Fasheh return 0; 2118ccd979bdSMark Fasheh } 2119ccd979bdSMark Fasheh 2120ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2121ccd979bdSMark Fasheh { 2122ccd979bdSMark Fasheh int status; 2123ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2124ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2125ccd979bdSMark Fasheh 2126ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2127ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2128ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2129ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2130ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2131ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2132ccd979bdSMark Fasheh 2133745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2134ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2135745ae8baSMark Fasheh || kthread_should_stop()); 2136ccd979bdSMark Fasheh 2137ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 2138ccd979bdSMark Fasheh if (status < 0) 2139ccd979bdSMark Fasheh mlog_errno(status); 2140ccd979bdSMark Fasheh 2141ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2142ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2143ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2144ccd979bdSMark Fasheh "shutdown\n", 2145ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2146ccd979bdSMark Fasheh } 2147ccd979bdSMark Fasheh } 2148ccd979bdSMark Fasheh 2149ccd979bdSMark Fasheh return 0; 2150ccd979bdSMark Fasheh } 2151ccd979bdSMark Fasheh 2152539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2153539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2154539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2155539d8264SSunil Mushran * a journal has been recovered by another node. 2156539d8264SSunil Mushran */ 2157ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2158ccd979bdSMark Fasheh { 2159ccd979bdSMark Fasheh int ret = 0; 2160ccd979bdSMark Fasheh unsigned int slot; 2161539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2162ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2163539d8264SSunil Mushran int journal_dirty = 0; 2164ccd979bdSMark Fasheh 2165ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2166539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2167539d8264SSunil Mushran if (ret) { 2168ccd979bdSMark Fasheh mlog_errno(ret); 2169ccd979bdSMark Fasheh goto out; 2170ccd979bdSMark Fasheh } 2171ccd979bdSMark Fasheh 2172ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2173ccd979bdSMark Fasheh 2174539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2175539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2176539d8264SSunil Mushran 2177ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2178ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2179539d8264SSunil Mushran journal_dirty = 1; 2180ccd979bdSMark Fasheh 2181ccd979bdSMark Fasheh brelse(di_bh); 2182539d8264SSunil Mushran di_bh = NULL; 2183ccd979bdSMark Fasheh } 2184ccd979bdSMark Fasheh 2185ccd979bdSMark Fasheh out: 2186539d8264SSunil Mushran if (journal_dirty) 2187539d8264SSunil Mushran ret = -EROFS; 2188ccd979bdSMark Fasheh return ret; 2189ccd979bdSMark Fasheh } 2190