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> 3355b465b6SJoseph Qi #include <linux/delay.h> 34ccd979bdSMark Fasheh 35ccd979bdSMark Fasheh #include <cluster/masklog.h> 36ccd979bdSMark Fasheh 37ccd979bdSMark Fasheh #include "ocfs2.h" 38ccd979bdSMark Fasheh 39ccd979bdSMark Fasheh #include "alloc.h" 4050655ae9SJoel Becker #include "blockcheck.h" 41316f4b9fSMark Fasheh #include "dir.h" 42ccd979bdSMark Fasheh #include "dlmglue.h" 43ccd979bdSMark Fasheh #include "extent_map.h" 44ccd979bdSMark Fasheh #include "heartbeat.h" 45ccd979bdSMark Fasheh #include "inode.h" 46ccd979bdSMark Fasheh #include "journal.h" 47ccd979bdSMark Fasheh #include "localalloc.h" 48ccd979bdSMark Fasheh #include "slot_map.h" 49ccd979bdSMark Fasheh #include "super.h" 50ccd979bdSMark Fasheh #include "sysfile.h" 510cf2f763SJoel Becker #include "uptodate.h" 522205363dSJan Kara #include "quota.h" 53ed460cffSJoseph Qi #include "file.h" 54ed460cffSJoseph Qi #include "namei.h" 55ccd979bdSMark Fasheh 56ccd979bdSMark Fasheh #include "buffer_head_io.h" 57b4107950STao Ma #include "ocfs2_trace.h" 58ccd979bdSMark Fasheh 5934af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 60ccd979bdSMark Fasheh 6183273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 6283273932SSrinivas Eeda 63ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 64ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 652205363dSJan Kara int node_num, int slot_num); 66ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 67ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 6819ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 69ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 70539d8264SSunil Mushran int dirty, int replayed); 71ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 72ccd979bdSMark Fasheh int slot_num); 73ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 74ed460cffSJoseph Qi int slot, 75ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 76ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 779140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 789140db04SSrinivas Eeda int slot_num, 799140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 809140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 81ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 82ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 83ccd979bdSMark Fasheh 8419ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 8519ece546SJan Kara { 8619ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 8719ece546SJan Kara } 8819ece546SJan Kara 8919ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 9019ece546SJan Kara { 9119ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 9219ece546SJan Kara } 9319ece546SJan Kara 949140db04SSrinivas Eeda /* 959140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 969140db04SSrinivas Eeda * offline slots during recovery and mount 979140db04SSrinivas Eeda */ 989140db04SSrinivas Eeda 999140db04SSrinivas Eeda enum ocfs2_replay_state { 1009140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 1019140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 1029140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 1039140db04SSrinivas Eeda }; 1049140db04SSrinivas Eeda 1059140db04SSrinivas Eeda struct ocfs2_replay_map { 1069140db04SSrinivas Eeda unsigned int rm_slots; 1079140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 1089140db04SSrinivas Eeda unsigned char rm_replay_slots[0]; 1099140db04SSrinivas Eeda }; 1109140db04SSrinivas Eeda 111b519ea6dSJoseph Qi static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 1129140db04SSrinivas Eeda { 1139140db04SSrinivas Eeda if (!osb->replay_map) 1149140db04SSrinivas Eeda return; 1159140db04SSrinivas Eeda 1169140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1179140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1189140db04SSrinivas Eeda return; 1199140db04SSrinivas Eeda 1209140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1219140db04SSrinivas Eeda } 1229140db04SSrinivas Eeda 1239140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1249140db04SSrinivas Eeda { 1259140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1269140db04SSrinivas Eeda int i, node_num; 1279140db04SSrinivas Eeda 1289140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1299140db04SSrinivas Eeda if (osb->replay_map) 1309140db04SSrinivas Eeda return 0; 1319140db04SSrinivas Eeda 1329140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1339140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1349140db04SSrinivas Eeda 1359140db04SSrinivas Eeda if (!replay_map) { 1369140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1379140db04SSrinivas Eeda return -ENOMEM; 1389140db04SSrinivas Eeda } 1399140db04SSrinivas Eeda 1409140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1419140db04SSrinivas Eeda 1429140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1439140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1449140db04SSrinivas Eeda 1459140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1469140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1479140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1489140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1499140db04SSrinivas Eeda } 1509140db04SSrinivas Eeda 1519140db04SSrinivas Eeda osb->replay_map = replay_map; 1529140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1539140db04SSrinivas Eeda return 0; 1549140db04SSrinivas Eeda } 1559140db04SSrinivas Eeda 156b519ea6dSJoseph Qi static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, 157ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1589140db04SSrinivas Eeda { 1599140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1609140db04SSrinivas Eeda int i; 1619140db04SSrinivas Eeda 1629140db04SSrinivas Eeda if (!replay_map) 1639140db04SSrinivas Eeda return; 1649140db04SSrinivas Eeda 1659140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1669140db04SSrinivas Eeda return; 1679140db04SSrinivas Eeda 1689140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1699140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1709140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 171ed460cffSJoseph Qi NULL, NULL, 172ed460cffSJoseph Qi orphan_reco_type); 1739140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1749140db04SSrinivas Eeda } 1759140db04SSrinivas Eeda 176b519ea6dSJoseph Qi static void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1779140db04SSrinivas Eeda { 1789140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1799140db04SSrinivas Eeda 1809140db04SSrinivas Eeda if (!osb->replay_map) 1819140db04SSrinivas Eeda return; 1829140db04SSrinivas Eeda 1839140db04SSrinivas Eeda kfree(replay_map); 1849140db04SSrinivas Eeda osb->replay_map = NULL; 1859140db04SSrinivas Eeda } 1869140db04SSrinivas Eeda 187553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 188553abd04SJoel Becker { 189553abd04SJoel Becker struct ocfs2_recovery_map *rm; 190553abd04SJoel Becker 191553abd04SJoel Becker mutex_init(&osb->recovery_lock); 192553abd04SJoel Becker osb->disable_recovery = 0; 193553abd04SJoel Becker osb->recovery_thread_task = NULL; 194553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 195553abd04SJoel Becker 196553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 197553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 198553abd04SJoel Becker GFP_KERNEL); 199553abd04SJoel Becker if (!rm) { 200553abd04SJoel Becker mlog_errno(-ENOMEM); 201553abd04SJoel Becker return -ENOMEM; 202553abd04SJoel Becker } 203553abd04SJoel Becker 204553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 205553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 206553abd04SJoel Becker osb->recovery_map = rm; 207553abd04SJoel Becker 208553abd04SJoel Becker return 0; 209553abd04SJoel Becker } 210553abd04SJoel Becker 211553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 212553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 213553abd04SJoel Becker * being woken up */ 214553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 215553abd04SJoel Becker { 216553abd04SJoel Becker mb(); 217553abd04SJoel Becker return osb->recovery_thread_task != NULL; 218553abd04SJoel Becker } 219553abd04SJoel Becker 220553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 221553abd04SJoel Becker { 222553abd04SJoel Becker struct ocfs2_recovery_map *rm; 223553abd04SJoel Becker 224553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 225553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 226553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 227553abd04SJoel Becker osb->disable_recovery = 1; 228553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 229553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 230553abd04SJoel Becker 231553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 232553abd04SJoel Becker * launched, so wait for any recovery completion work to 233553abd04SJoel Becker * complete. */ 23435ddf78eSjiangyiwen flush_workqueue(osb->ocfs2_wq); 235553abd04SJoel Becker 236553abd04SJoel Becker /* 237553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 238553abd04SJoel Becker * freed, the osb_lock is not taken here. 239553abd04SJoel Becker */ 240553abd04SJoel Becker rm = osb->recovery_map; 241553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 242553abd04SJoel Becker 243553abd04SJoel Becker kfree(rm); 244553abd04SJoel Becker } 245553abd04SJoel Becker 246553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 247553abd04SJoel Becker unsigned int node_num) 248553abd04SJoel Becker { 249553abd04SJoel Becker int i; 250553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 251553abd04SJoel Becker 252553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 253553abd04SJoel Becker 254553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 255553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 256553abd04SJoel Becker return 1; 257553abd04SJoel Becker } 258553abd04SJoel Becker 259553abd04SJoel Becker return 0; 260553abd04SJoel Becker } 261553abd04SJoel Becker 262553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 263553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 264553abd04SJoel Becker unsigned int node_num) 265553abd04SJoel Becker { 266553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 267553abd04SJoel Becker 268553abd04SJoel Becker spin_lock(&osb->osb_lock); 269553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 270553abd04SJoel Becker spin_unlock(&osb->osb_lock); 271553abd04SJoel Becker return 1; 272553abd04SJoel Becker } 273553abd04SJoel Becker 274553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 275553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 276553abd04SJoel Becker 277553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 278553abd04SJoel Becker rm->rm_used++; 279553abd04SJoel Becker spin_unlock(&osb->osb_lock); 280553abd04SJoel Becker 281553abd04SJoel Becker return 0; 282553abd04SJoel Becker } 283553abd04SJoel Becker 284553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 285553abd04SJoel Becker unsigned int node_num) 286553abd04SJoel Becker { 287553abd04SJoel Becker int i; 288553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 289553abd04SJoel Becker 290553abd04SJoel Becker spin_lock(&osb->osb_lock); 291553abd04SJoel Becker 292553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 293553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 294553abd04SJoel Becker break; 295553abd04SJoel Becker } 296553abd04SJoel Becker 297553abd04SJoel Becker if (i < rm->rm_used) { 298553abd04SJoel Becker /* XXX: be careful with the pointer math */ 299553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 300553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 301553abd04SJoel Becker rm->rm_used--; 302553abd04SJoel Becker } 303553abd04SJoel Becker 304553abd04SJoel Becker spin_unlock(&osb->osb_lock); 305553abd04SJoel Becker } 306553abd04SJoel Becker 307ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 308ccd979bdSMark Fasheh { 309ccd979bdSMark Fasheh int status = 0; 310ccd979bdSMark Fasheh unsigned int flushed; 311ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 312ccd979bdSMark Fasheh 313ccd979bdSMark Fasheh journal = osb->journal; 314ccd979bdSMark Fasheh 315ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 316ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 317ccd979bdSMark Fasheh 318b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 319b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 320b4107950STao Ma if (flushed == 0) { 321ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 322ccd979bdSMark Fasheh goto finally; 323ccd979bdSMark Fasheh } 324ccd979bdSMark Fasheh 3252b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 3262b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 3272b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 328ccd979bdSMark Fasheh if (status < 0) { 329ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 330ccd979bdSMark Fasheh mlog_errno(status); 331ccd979bdSMark Fasheh goto finally; 332ccd979bdSMark Fasheh } 333ccd979bdSMark Fasheh 334f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 335ccd979bdSMark Fasheh 336ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 337ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 338ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 339ccd979bdSMark Fasheh 340b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 341ccd979bdSMark Fasheh 34234d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 343ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 344ccd979bdSMark Fasheh finally: 345ccd979bdSMark Fasheh return status; 346ccd979bdSMark Fasheh } 347ccd979bdSMark Fasheh 3481fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 349ccd979bdSMark Fasheh { 350ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3511fabe148SMark Fasheh handle_t *handle; 352ccd979bdSMark Fasheh 353ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 354ccd979bdSMark Fasheh 35565eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 35665eff9ccSMark Fasheh return ERR_PTR(-EROFS); 357ccd979bdSMark Fasheh 358ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 359ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 360ccd979bdSMark Fasheh 36190e86a63SJan Kara /* Nested transaction? Just return the handle... */ 36290e86a63SJan Kara if (journal_current_handle()) 36390e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 364ccd979bdSMark Fasheh 365fef6925cSJan Kara sb_start_intwrite(osb->sb); 366fef6925cSJan Kara 367ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 368ccd979bdSMark Fasheh 3692b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3701fabe148SMark Fasheh if (IS_ERR(handle)) { 371ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 372fef6925cSJan Kara sb_end_intwrite(osb->sb); 373ccd979bdSMark Fasheh 3741fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 375ccd979bdSMark Fasheh 376ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 3777ecef14aSJoe Perches ocfs2_abort(osb->sb, "Detected aborted journal\n"); 3781fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 379ccd979bdSMark Fasheh } 380c271c5c2SSunil Mushran } else { 381c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 382ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 383c271c5c2SSunil Mushran } 384ccd979bdSMark Fasheh 385ccd979bdSMark Fasheh return handle; 386ccd979bdSMark Fasheh } 387ccd979bdSMark Fasheh 3881fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3891fabe148SMark Fasheh handle_t *handle) 390ccd979bdSMark Fasheh { 39190e86a63SJan Kara int ret, nested; 39202dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 393ccd979bdSMark Fasheh 394ccd979bdSMark Fasheh BUG_ON(!handle); 395ccd979bdSMark Fasheh 39690e86a63SJan Kara nested = handle->h_ref > 1; 3972b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3981fabe148SMark Fasheh if (ret < 0) 3991fabe148SMark Fasheh mlog_errno(ret); 400ccd979bdSMark Fasheh 401fef6925cSJan Kara if (!nested) { 402ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 403fef6925cSJan Kara sb_end_intwrite(osb->sb); 404fef6925cSJan Kara } 405ccd979bdSMark Fasheh 4061fabe148SMark Fasheh return ret; 407ccd979bdSMark Fasheh } 408ccd979bdSMark Fasheh 409ccd979bdSMark Fasheh /* 410c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 411ccd979bdSMark Fasheh * 4122b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 413e8aed345SMark Fasheh * and then restart the transaction. Before calling 414e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 415e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 416e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 417e8aed345SMark Fasheh * 418ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 419ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 420ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 421ccd979bdSMark Fasheh * 422ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 423ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 424ccd979bdSMark Fasheh * cluster locks associated with this handle. 425ccd979bdSMark Fasheh */ 4261fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 427ccd979bdSMark Fasheh { 428c901fb00STao Ma int status, old_nblocks; 429ccd979bdSMark Fasheh 430ccd979bdSMark Fasheh BUG_ON(!handle); 431c901fb00STao Ma BUG_ON(nblocks < 0); 432ccd979bdSMark Fasheh 433c901fb00STao Ma if (!nblocks) 434c901fb00STao Ma return 0; 435c901fb00STao Ma 436c901fb00STao Ma old_nblocks = handle->h_buffer_credits; 437ccd979bdSMark Fasheh 438b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 439ccd979bdSMark Fasheh 440e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4410879c584SMark Fasheh status = 1; 4420879c584SMark Fasheh #else 4432b4e30fbSJoel Becker status = jbd2_journal_extend(handle, nblocks); 444ccd979bdSMark Fasheh if (status < 0) { 445ccd979bdSMark Fasheh mlog_errno(status); 446ccd979bdSMark Fasheh goto bail; 447ccd979bdSMark Fasheh } 4480879c584SMark Fasheh #endif 449ccd979bdSMark Fasheh 450ccd979bdSMark Fasheh if (status > 0) { 451b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 452c901fb00STao Ma status = jbd2_journal_restart(handle, 453c901fb00STao Ma old_nblocks + nblocks); 454ccd979bdSMark Fasheh if (status < 0) { 455ccd979bdSMark Fasheh mlog_errno(status); 456ccd979bdSMark Fasheh goto bail; 457ccd979bdSMark Fasheh } 45801ddf1e1SMark Fasheh } 459ccd979bdSMark Fasheh 460ccd979bdSMark Fasheh status = 0; 461ccd979bdSMark Fasheh bail: 462ccd979bdSMark Fasheh return status; 463ccd979bdSMark Fasheh } 464ccd979bdSMark Fasheh 4652b1e55c3SYounger Liu /* 4662b1e55c3SYounger Liu * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 4672b1e55c3SYounger Liu * If that fails, restart the transaction & regain write access for the 4682b1e55c3SYounger Liu * buffer head which is used for metadata modifications. 4692b1e55c3SYounger Liu * Taken from Ext4: extend_or_restart_transaction() 4702b1e55c3SYounger Liu */ 4712b1e55c3SYounger Liu int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 4722b1e55c3SYounger Liu { 4732b1e55c3SYounger Liu int status, old_nblks; 4742b1e55c3SYounger Liu 4752b1e55c3SYounger Liu BUG_ON(!handle); 4762b1e55c3SYounger Liu 4772b1e55c3SYounger Liu old_nblks = handle->h_buffer_credits; 4782b1e55c3SYounger Liu trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 4792b1e55c3SYounger Liu 4802b1e55c3SYounger Liu if (old_nblks < thresh) 4812b1e55c3SYounger Liu return 0; 4822b1e55c3SYounger Liu 4832b1e55c3SYounger Liu status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA); 4842b1e55c3SYounger Liu if (status < 0) { 4852b1e55c3SYounger Liu mlog_errno(status); 4862b1e55c3SYounger Liu goto bail; 4872b1e55c3SYounger Liu } 4882b1e55c3SYounger Liu 4892b1e55c3SYounger Liu if (status > 0) { 4902b1e55c3SYounger Liu status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 4912b1e55c3SYounger Liu if (status < 0) 4922b1e55c3SYounger Liu mlog_errno(status); 4932b1e55c3SYounger Liu } 4942b1e55c3SYounger Liu 4952b1e55c3SYounger Liu bail: 4962b1e55c3SYounger Liu return status; 4972b1e55c3SYounger Liu } 4982b1e55c3SYounger Liu 4992b1e55c3SYounger Liu 50050655ae9SJoel Becker struct ocfs2_triggers { 50150655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 50250655ae9SJoel Becker int ot_offset; 50350655ae9SJoel Becker }; 50450655ae9SJoel Becker 50550655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 50650655ae9SJoel Becker { 50750655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 50850655ae9SJoel Becker } 50950655ae9SJoel Becker 51013ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 51150655ae9SJoel Becker struct buffer_head *bh, 51250655ae9SJoel Becker void *data, size_t size) 51350655ae9SJoel Becker { 51450655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 51550655ae9SJoel Becker 51650655ae9SJoel Becker /* 51750655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 51850655ae9SJoel Becker * must unconditionally compute the ecc data. 51950655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 52050655ae9SJoel Becker * metaecc is enabled. 52150655ae9SJoel Becker */ 52250655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 52350655ae9SJoel Becker } 52450655ae9SJoel Becker 52550655ae9SJoel Becker /* 52650655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 52750655ae9SJoel Becker * offset depends on the blocksize. 52850655ae9SJoel Becker */ 52913ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 53050655ae9SJoel Becker struct buffer_head *bh, 53150655ae9SJoel Becker void *data, size_t size) 53250655ae9SJoel Becker { 53350655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 53450655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 53550655ae9SJoel Becker 53650655ae9SJoel Becker /* 53750655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 53850655ae9SJoel Becker * must unconditionally compute the ecc data. 53950655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 54050655ae9SJoel Becker * metaecc is enabled. 54150655ae9SJoel Becker */ 54250655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 54350655ae9SJoel Becker } 54450655ae9SJoel Becker 545c175a518SJoel Becker /* 546c175a518SJoel Becker * Directory blocks also have their own trigger because the 547c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 548c175a518SJoel Becker */ 54913ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 550c175a518SJoel Becker struct buffer_head *bh, 551c175a518SJoel Becker void *data, size_t size) 552c175a518SJoel Becker { 553c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 554c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 555c175a518SJoel Becker 556c175a518SJoel Becker /* 557c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 558c175a518SJoel Becker * must unconditionally compute the ecc data. 559c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 560c175a518SJoel Becker * metaecc is enabled. 561c175a518SJoel Becker */ 562c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 563c175a518SJoel Becker } 564c175a518SJoel Becker 56550655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 56650655ae9SJoel Becker struct buffer_head *bh) 56750655ae9SJoel Becker { 56850655ae9SJoel Becker mlog(ML_ERROR, 56950655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 57050655ae9SJoel Becker "bh->b_blocknr = %llu\n", 57150655ae9SJoel Becker (unsigned long)bh, 57250655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 57350655ae9SJoel Becker 57474e364adSXue jiufei ocfs2_error(bh->b_bdev->bd_super, 57550655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 57650655ae9SJoel Becker } 57750655ae9SJoel Becker 57850655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 57950655ae9SJoel Becker .ot_triggers = { 58013ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58150655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 58250655ae9SJoel Becker }, 58350655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 58450655ae9SJoel Becker }; 58550655ae9SJoel Becker 58650655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 58750655ae9SJoel Becker .ot_triggers = { 58813ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58950655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 59050655ae9SJoel Becker }, 59150655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 59250655ae9SJoel Becker }; 59350655ae9SJoel Becker 59493c97087STao Ma static struct ocfs2_triggers rb_triggers = { 59593c97087STao Ma .ot_triggers = { 59613ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 59793c97087STao Ma .t_abort = ocfs2_abort_trigger, 59893c97087STao Ma }, 59993c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 60093c97087STao Ma }; 60193c97087STao Ma 60250655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 60350655ae9SJoel Becker .ot_triggers = { 60413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 60550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 60650655ae9SJoel Becker }, 60750655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 60850655ae9SJoel Becker }; 60950655ae9SJoel Becker 610c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 611c175a518SJoel Becker .ot_triggers = { 61213ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 613c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 614c175a518SJoel Becker }, 615c175a518SJoel Becker }; 616c175a518SJoel Becker 61750655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 61850655ae9SJoel Becker .ot_triggers = { 61913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 62050655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 62150655ae9SJoel Becker }, 62250655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 62350655ae9SJoel Becker }; 62450655ae9SJoel Becker 62550655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 62650655ae9SJoel Becker .ot_triggers = { 62713ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 62850655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 62950655ae9SJoel Becker }, 63050655ae9SJoel Becker }; 63150655ae9SJoel Becker 6329b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 6339b7895efSMark Fasheh .ot_triggers = { 63413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6359b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6369b7895efSMark Fasheh }, 6379b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 6389b7895efSMark Fasheh }; 6399b7895efSMark Fasheh 6409b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 6419b7895efSMark Fasheh .ot_triggers = { 64213ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6439b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6449b7895efSMark Fasheh }, 6459b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6469b7895efSMark Fasheh }; 6479b7895efSMark Fasheh 64850655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6490cf2f763SJoel Becker struct ocfs2_caching_info *ci, 650ccd979bdSMark Fasheh struct buffer_head *bh, 65150655ae9SJoel Becker struct ocfs2_triggers *triggers, 652ccd979bdSMark Fasheh int type) 653ccd979bdSMark Fasheh { 654ccd979bdSMark Fasheh int status; 6550cf2f763SJoel Becker struct ocfs2_super *osb = 6560cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 657ccd979bdSMark Fasheh 6580cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 659ccd979bdSMark Fasheh BUG_ON(!handle); 660ccd979bdSMark Fasheh BUG_ON(!bh); 661ccd979bdSMark Fasheh 662b4107950STao Ma trace_ocfs2_journal_access( 663b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 664b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 665ccd979bdSMark Fasheh 666ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 667ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 668ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 669d984187eSpiaojun mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", 670d984187eSpiaojun (unsigned long long)bh->b_blocknr, bh->b_state); 671acf8fdbeSJoseph Qi 672acf8fdbeSJoseph Qi lock_buffer(bh); 673acf8fdbeSJoseph Qi /* 674d984187eSpiaojun * A previous transaction with a couple of buffer heads fail 675d984187eSpiaojun * to checkpoint, so all the bhs are marked as BH_Write_EIO. 676d984187eSpiaojun * For current transaction, the bh is just among those error 677d984187eSpiaojun * bhs which previous transaction handle. We can't just clear 678d984187eSpiaojun * its BH_Write_EIO and reuse directly, since other bhs are 679d984187eSpiaojun * not written to disk yet and that will cause metadata 680d984187eSpiaojun * inconsistency. So we should set fs read-only to avoid 681d984187eSpiaojun * further damage. 682acf8fdbeSJoseph Qi */ 683acf8fdbeSJoseph Qi if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { 684acf8fdbeSJoseph Qi unlock_buffer(bh); 685d984187eSpiaojun return ocfs2_error(osb->sb, "A previous attempt to " 686d984187eSpiaojun "write this buffer head failed\n"); 687acf8fdbeSJoseph Qi } 688acf8fdbeSJoseph Qi unlock_buffer(bh); 689ccd979bdSMark Fasheh } 690ccd979bdSMark Fasheh 6910cf2f763SJoel Becker /* Set the current transaction information on the ci so 692ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6930cf2f763SJoel Becker * on this ci or not. We're protected from the commit 694ccd979bdSMark Fasheh * thread updating the current transaction id until 695ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 696ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6970cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 698ccd979bdSMark Fasheh 6990cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 700ccd979bdSMark Fasheh switch (type) { 701ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 702ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 7032b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 704ccd979bdSMark Fasheh break; 705ccd979bdSMark Fasheh 706ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 7072b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 708ccd979bdSMark Fasheh break; 709ccd979bdSMark Fasheh 710ccd979bdSMark Fasheh default: 711ccd979bdSMark Fasheh status = -EINVAL; 712af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 713ccd979bdSMark Fasheh } 7140cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 71550655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 7160cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 717ccd979bdSMark Fasheh 718ccd979bdSMark Fasheh if (status < 0) 719ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 720ccd979bdSMark Fasheh status, type); 721ccd979bdSMark Fasheh 722ccd979bdSMark Fasheh return status; 723ccd979bdSMark Fasheh } 724ccd979bdSMark Fasheh 7250cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 72650655ae9SJoel Becker struct buffer_head *bh, int type) 72750655ae9SJoel Becker { 7280cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 72950655ae9SJoel Becker } 73050655ae9SJoel Becker 7310cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 73250655ae9SJoel Becker struct buffer_head *bh, int type) 73350655ae9SJoel Becker { 7340cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 73550655ae9SJoel Becker } 73650655ae9SJoel Becker 73793c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 73893c97087STao Ma struct buffer_head *bh, int type) 73993c97087STao Ma { 74093c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 74193c97087STao Ma type); 74293c97087STao Ma } 74393c97087STao Ma 7440cf2f763SJoel Becker int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 74550655ae9SJoel Becker struct buffer_head *bh, int type) 74650655ae9SJoel Becker { 7470cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 74850655ae9SJoel Becker } 74950655ae9SJoel Becker 7500cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 75150655ae9SJoel Becker struct buffer_head *bh, int type) 75250655ae9SJoel Becker { 7530cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 75450655ae9SJoel Becker } 75550655ae9SJoel Becker 7560cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 75750655ae9SJoel Becker struct buffer_head *bh, int type) 75850655ae9SJoel Becker { 7590cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 76050655ae9SJoel Becker } 76150655ae9SJoel Becker 7620cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 76350655ae9SJoel Becker struct buffer_head *bh, int type) 76450655ae9SJoel Becker { 7650cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 76650655ae9SJoel Becker } 76750655ae9SJoel Becker 7680cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7699b7895efSMark Fasheh struct buffer_head *bh, int type) 7709b7895efSMark Fasheh { 7710cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7729b7895efSMark Fasheh } 7739b7895efSMark Fasheh 7740cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7759b7895efSMark Fasheh struct buffer_head *bh, int type) 7769b7895efSMark Fasheh { 7770cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7789b7895efSMark Fasheh } 7799b7895efSMark Fasheh 7800cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 78150655ae9SJoel Becker struct buffer_head *bh, int type) 78250655ae9SJoel Becker { 7830cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 78450655ae9SJoel Becker } 78550655ae9SJoel Becker 786ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 787ccd979bdSMark Fasheh { 788ccd979bdSMark Fasheh int status; 789ccd979bdSMark Fasheh 790b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 791ccd979bdSMark Fasheh 7922b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 793e272e7f0SJoseph Qi if (status) { 794e272e7f0SJoseph Qi mlog_errno(status); 795e272e7f0SJoseph Qi if (!is_handle_aborted(handle)) { 796e272e7f0SJoseph Qi journal_t *journal = handle->h_transaction->t_journal; 797e272e7f0SJoseph Qi struct super_block *sb = bh->b_bdev->bd_super; 798e272e7f0SJoseph Qi 799e272e7f0SJoseph Qi mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " 800e272e7f0SJoseph Qi "Aborting transaction and journal.\n"); 801e272e7f0SJoseph Qi handle->h_err = status; 802e272e7f0SJoseph Qi jbd2_journal_abort_handle(handle); 803e272e7f0SJoseph Qi jbd2_journal_abort(journal, status); 804e272e7f0SJoseph Qi ocfs2_abort(sb, "Journal already aborted.\n"); 805e272e7f0SJoseph Qi } 806e272e7f0SJoseph Qi } 807ccd979bdSMark Fasheh } 808ccd979bdSMark Fasheh 8092b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 810ccd979bdSMark Fasheh 811ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 812ccd979bdSMark Fasheh { 813ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 814d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 815d147b3d6SMark Fasheh 816d147b3d6SMark Fasheh if (osb->osb_commit_interval) 817d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 818ccd979bdSMark Fasheh 819a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 820d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 821ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 8222b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 823ccd979bdSMark Fasheh else 8242b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 825a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 826ccd979bdSMark Fasheh } 827ccd979bdSMark Fasheh 828ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 829ccd979bdSMark Fasheh { 830ccd979bdSMark Fasheh int status = -1; 831ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 832ccd979bdSMark Fasheh journal_t *j_journal = NULL; 833ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 834ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 835ccd979bdSMark Fasheh struct ocfs2_super *osb; 836e63aecb6SMark Fasheh int inode_lock = 0; 837ccd979bdSMark Fasheh 838ccd979bdSMark Fasheh BUG_ON(!journal); 839ccd979bdSMark Fasheh 840ccd979bdSMark Fasheh osb = journal->j_osb; 841ccd979bdSMark Fasheh 842ccd979bdSMark Fasheh /* already have the inode for our journal */ 843ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 844ccd979bdSMark Fasheh osb->slot_num); 845ccd979bdSMark Fasheh if (inode == NULL) { 846ccd979bdSMark Fasheh status = -EACCES; 847ccd979bdSMark Fasheh mlog_errno(status); 848ccd979bdSMark Fasheh goto done; 849ccd979bdSMark Fasheh } 850ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 851ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 852ccd979bdSMark Fasheh iput(inode); 853ccd979bdSMark Fasheh inode = NULL; 854ccd979bdSMark Fasheh status = -EACCES; 855ccd979bdSMark Fasheh goto done; 856ccd979bdSMark Fasheh } 857ccd979bdSMark Fasheh 858ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 859ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 860ccd979bdSMark Fasheh 8616eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 8626eff5790SMark Fasheh * changes in a live cluster so it can be considered an 8636eff5790SMark Fasheh * exception to the rule. */ 864e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 865ccd979bdSMark Fasheh if (status < 0) { 866ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 867ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 868ccd979bdSMark Fasheh goto done; 869ccd979bdSMark Fasheh } 870ccd979bdSMark Fasheh 871e63aecb6SMark Fasheh inode_lock = 1; 872ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 873ccd979bdSMark Fasheh 874f17c20ddSJunxiao Bi if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 875ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 876f17c20ddSJunxiao Bi i_size_read(inode)); 877ccd979bdSMark Fasheh status = -EINVAL; 878ccd979bdSMark Fasheh goto done; 879ccd979bdSMark Fasheh } 880ccd979bdSMark Fasheh 881f17c20ddSJunxiao Bi trace_ocfs2_journal_init(i_size_read(inode), 882b4107950STao Ma (unsigned long long)inode->i_blocks, 883b4107950STao Ma OCFS2_I(inode)->ip_clusters); 884ccd979bdSMark Fasheh 885ccd979bdSMark Fasheh /* call the kernels journal init function now */ 8862b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 887ccd979bdSMark Fasheh if (j_journal == NULL) { 888ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 889ccd979bdSMark Fasheh status = -EINVAL; 890ccd979bdSMark Fasheh goto done; 891ccd979bdSMark Fasheh } 892ccd979bdSMark Fasheh 893b4107950STao Ma trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); 894ccd979bdSMark Fasheh 895ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 896ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 897ccd979bdSMark Fasheh 898ccd979bdSMark Fasheh journal->j_journal = j_journal; 899ccd979bdSMark Fasheh journal->j_inode = inode; 900ccd979bdSMark Fasheh journal->j_bh = bh; 901ccd979bdSMark Fasheh 902ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 903ccd979bdSMark Fasheh 904ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 905ccd979bdSMark Fasheh 906ccd979bdSMark Fasheh status = 0; 907ccd979bdSMark Fasheh done: 908ccd979bdSMark Fasheh if (status < 0) { 909e63aecb6SMark Fasheh if (inode_lock) 910e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 911ccd979bdSMark Fasheh brelse(bh); 912ccd979bdSMark Fasheh if (inode) { 913ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 914ccd979bdSMark Fasheh iput(inode); 915ccd979bdSMark Fasheh } 916ccd979bdSMark Fasheh } 917ccd979bdSMark Fasheh 918ccd979bdSMark Fasheh return status; 919ccd979bdSMark Fasheh } 920ccd979bdSMark Fasheh 921539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 922539d8264SSunil Mushran { 923539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 924539d8264SSunil Mushran } 925539d8264SSunil Mushran 926539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 927539d8264SSunil Mushran { 928539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 929539d8264SSunil Mushran } 930539d8264SSunil Mushran 931ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 932539d8264SSunil Mushran int dirty, int replayed) 933ccd979bdSMark Fasheh { 934ccd979bdSMark Fasheh int status; 935ccd979bdSMark Fasheh unsigned int flags; 936ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 937ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 938ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 939ccd979bdSMark Fasheh 940ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 94110995aa2SJoel Becker 94210995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 94310995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 94410995aa2SJoel Becker * code bug if we mess it up. */ 94510995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 946ccd979bdSMark Fasheh 947ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 948ccd979bdSMark Fasheh if (dirty) 949ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 950ccd979bdSMark Fasheh else 951ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 952ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 953ccd979bdSMark Fasheh 954539d8264SSunil Mushran if (replayed) 955539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 956539d8264SSunil Mushran 95713723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 9588cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 959ccd979bdSMark Fasheh if (status < 0) 960ccd979bdSMark Fasheh mlog_errno(status); 961ccd979bdSMark Fasheh 962ccd979bdSMark Fasheh return status; 963ccd979bdSMark Fasheh } 964ccd979bdSMark Fasheh 965ccd979bdSMark Fasheh /* 966ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 967ccd979bdSMark Fasheh * call. 968ccd979bdSMark Fasheh */ 969ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 970ccd979bdSMark Fasheh { 971ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 972ccd979bdSMark Fasheh int status = 0; 973ccd979bdSMark Fasheh struct inode *inode = NULL; 974ccd979bdSMark Fasheh int num_running_trans = 0; 975ccd979bdSMark Fasheh 976ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 977ccd979bdSMark Fasheh 978ccd979bdSMark Fasheh journal = osb->journal; 979ccd979bdSMark Fasheh if (!journal) 980ccd979bdSMark Fasheh goto done; 981ccd979bdSMark Fasheh 982ccd979bdSMark Fasheh inode = journal->j_inode; 983ccd979bdSMark Fasheh 984ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 985ccd979bdSMark Fasheh goto done; 986ccd979bdSMark Fasheh 9872b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 988ccd979bdSMark Fasheh if (!igrab(inode)) 989ccd979bdSMark Fasheh BUG(); 990ccd979bdSMark Fasheh 991ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 992b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 993ccd979bdSMark Fasheh 994ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 995ccd979bdSMark Fasheh * release any locks that are still held. 996ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 997ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 998ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 999ccd979bdSMark Fasheh 1000ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 1001ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 1002ccd979bdSMark Fasheh * completely destroy the journal. */ 1003ccd979bdSMark Fasheh if (osb->commit_task) { 1004ccd979bdSMark Fasheh /* Wait for the commit thread */ 1005b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 1006ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 1007ccd979bdSMark Fasheh osb->commit_task = NULL; 1008ccd979bdSMark Fasheh } 1009ccd979bdSMark Fasheh 1010ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 1011ccd979bdSMark Fasheh 1012c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 10132b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 10142b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 10152b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 1016c271c5c2SSunil Mushran if (status < 0) 1017c271c5c2SSunil Mushran mlog_errno(status); 1018c271c5c2SSunil Mushran } 1019c271c5c2SSunil Mushran 1020d85400afSJunxiao Bi /* Shutdown the kernel journal system */ 1021d85400afSJunxiao Bi if (!jbd2_journal_destroy(journal->j_journal) && !status) { 1022c271c5c2SSunil Mushran /* 1023c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 1024c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 1025c271c5c2SSunil Mushran */ 1026539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 1027ccd979bdSMark Fasheh if (status < 0) 1028ccd979bdSMark Fasheh mlog_errno(status); 1029c271c5c2SSunil Mushran } 1030ae0dff68SSunil Mushran journal->j_journal = NULL; 1031ccd979bdSMark Fasheh 1032ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 1033ccd979bdSMark Fasheh 1034ccd979bdSMark Fasheh /* unlock our journal */ 1035e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1036ccd979bdSMark Fasheh 1037ccd979bdSMark Fasheh brelse(journal->j_bh); 1038ccd979bdSMark Fasheh journal->j_bh = NULL; 1039ccd979bdSMark Fasheh 1040ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 1041ccd979bdSMark Fasheh 1042ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 1043ccd979bdSMark Fasheh done: 1044ccd979bdSMark Fasheh iput(inode); 1045ccd979bdSMark Fasheh } 1046ccd979bdSMark Fasheh 1047ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 1048ccd979bdSMark Fasheh journal_t *journal, 1049ccd979bdSMark Fasheh int slot) 1050ccd979bdSMark Fasheh { 1051ccd979bdSMark Fasheh int olderr; 1052ccd979bdSMark Fasheh 10532b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 1054ccd979bdSMark Fasheh if (olderr) { 1055ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 1056ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 1057ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 1058ccd979bdSMark Fasheh sb->s_id); 1059ccd979bdSMark Fasheh 10602b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 10612b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 1062ccd979bdSMark Fasheh } 1063ccd979bdSMark Fasheh } 1064ccd979bdSMark Fasheh 1065539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1066ccd979bdSMark Fasheh { 1067ccd979bdSMark Fasheh int status = 0; 1068ccd979bdSMark Fasheh struct ocfs2_super *osb; 1069ccd979bdSMark Fasheh 1070b1f3550fSJulia Lawall BUG_ON(!journal); 1071ccd979bdSMark Fasheh 1072ccd979bdSMark Fasheh osb = journal->j_osb; 1073ccd979bdSMark Fasheh 10742b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1075ccd979bdSMark Fasheh if (status < 0) { 1076ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1077ccd979bdSMark Fasheh goto done; 1078ccd979bdSMark Fasheh } 1079ccd979bdSMark Fasheh 1080ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1081ccd979bdSMark Fasheh 1082539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1083ccd979bdSMark Fasheh if (status < 0) { 1084ccd979bdSMark Fasheh mlog_errno(status); 1085ccd979bdSMark Fasheh goto done; 1086ccd979bdSMark Fasheh } 1087ccd979bdSMark Fasheh 1088ccd979bdSMark Fasheh /* Launch the commit thread */ 1089c271c5c2SSunil Mushran if (!local) { 1090c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 10915afc44e2SJoseph Qi "ocfs2cmt-%s", osb->uuid_str); 1092ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1093ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1094ccd979bdSMark Fasheh osb->commit_task = NULL; 1095c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1096c271c5c2SSunil Mushran "error=%d", status); 1097ccd979bdSMark Fasheh goto done; 1098ccd979bdSMark Fasheh } 1099c271c5c2SSunil Mushran } else 1100c271c5c2SSunil Mushran osb->commit_task = NULL; 1101ccd979bdSMark Fasheh 1102ccd979bdSMark Fasheh done: 1103ccd979bdSMark Fasheh return status; 1104ccd979bdSMark Fasheh } 1105ccd979bdSMark Fasheh 1106ccd979bdSMark Fasheh 1107ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1108ccd979bdSMark Fasheh * mark the journal clean */ 1109ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1110ccd979bdSMark Fasheh { 1111ccd979bdSMark Fasheh int status; 1112ccd979bdSMark Fasheh 1113ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1114ccd979bdSMark Fasheh 11152b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1116ccd979bdSMark Fasheh if (status < 0) { 1117ccd979bdSMark Fasheh mlog_errno(status); 1118ccd979bdSMark Fasheh goto bail; 1119ccd979bdSMark Fasheh } 1120ccd979bdSMark Fasheh 1121539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1122ccd979bdSMark Fasheh if (status < 0) 1123ccd979bdSMark Fasheh mlog_errno(status); 1124ccd979bdSMark Fasheh 1125ccd979bdSMark Fasheh bail: 1126ccd979bdSMark Fasheh return status; 1127ccd979bdSMark Fasheh } 1128ccd979bdSMark Fasheh 1129553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1130553abd04SJoel Becker { 1131553abd04SJoel Becker int empty; 1132553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1133553abd04SJoel Becker 1134553abd04SJoel Becker spin_lock(&osb->osb_lock); 1135553abd04SJoel Becker empty = (rm->rm_used == 0); 1136553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1137553abd04SJoel Becker 1138553abd04SJoel Becker return empty; 1139553abd04SJoel Becker } 1140553abd04SJoel Becker 1141553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1142553abd04SJoel Becker { 1143553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1144553abd04SJoel Becker } 1145553abd04SJoel Becker 1146ccd979bdSMark Fasheh /* 1147ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1148ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1149ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1150ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1151ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1152ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1153ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1154ccd979bdSMark Fasheh * as we have the lock by now :) 1155ccd979bdSMark Fasheh */ 1156ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1157ccd979bdSMark Fasheh { 1158ccd979bdSMark Fasheh int status = 0; 11594f902c37SMark Fasheh int i; 11608110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 11610b492f68SJunxiao Bi struct buffer_head *bh = NULL; 11620b492f68SJunxiao Bi struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1163ccd979bdSMark Fasheh 1164f17c20ddSJunxiao Bi num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1165ccd979bdSMark Fasheh v_blkno = 0; 11668110b073SMark Fasheh while (v_blkno < num_blocks) { 1167ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 116849cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1169ccd979bdSMark Fasheh if (status < 0) { 1170ccd979bdSMark Fasheh mlog_errno(status); 1171ccd979bdSMark Fasheh goto bail; 1172ccd979bdSMark Fasheh } 1173ccd979bdSMark Fasheh 11740b492f68SJunxiao Bi for (i = 0; i < p_blocks; i++, p_blkno++) { 11750b492f68SJunxiao Bi bh = __find_get_block(osb->sb->s_bdev, p_blkno, 11760b492f68SJunxiao Bi osb->sb->s_blocksize); 11770b492f68SJunxiao Bi /* block not cached. */ 11780b492f68SJunxiao Bi if (!bh) 11790b492f68SJunxiao Bi continue; 1180ccd979bdSMark Fasheh 11810b492f68SJunxiao Bi brelse(bh); 11820b492f68SJunxiao Bi bh = NULL; 1183dd4a2c2bSMark Fasheh /* We are reading journal data which should not 11840b492f68SJunxiao Bi * be put in the uptodate cache. 11850b492f68SJunxiao Bi */ 11860b492f68SJunxiao Bi status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); 1187ccd979bdSMark Fasheh if (status < 0) { 1188ccd979bdSMark Fasheh mlog_errno(status); 1189ccd979bdSMark Fasheh goto bail; 1190ccd979bdSMark Fasheh } 1191ccd979bdSMark Fasheh 11920b492f68SJunxiao Bi brelse(bh); 11930b492f68SJunxiao Bi bh = NULL; 1194ccd979bdSMark Fasheh } 1195ccd979bdSMark Fasheh 1196ccd979bdSMark Fasheh v_blkno += p_blocks; 1197ccd979bdSMark Fasheh } 1198ccd979bdSMark Fasheh 1199ccd979bdSMark Fasheh bail: 1200ccd979bdSMark Fasheh return status; 1201ccd979bdSMark Fasheh } 1202ccd979bdSMark Fasheh 1203ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1204ccd979bdSMark Fasheh struct list_head lri_list; 1205ccd979bdSMark Fasheh int lri_slot; 1206ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1207ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 12082205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1209ed460cffSJoseph Qi enum ocfs2_orphan_reco_type lri_orphan_reco_type; 1210ccd979bdSMark Fasheh }; 1211ccd979bdSMark Fasheh 1212ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1213ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1214ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1215ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1216ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1217ccd979bdSMark Fasheh * recovering nodes into account. 1218ccd979bdSMark Fasheh * 1219ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1220ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1221ccd979bdSMark Fasheh */ 1222c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1223ccd979bdSMark Fasheh { 1224b4107950STao Ma int ret = 0; 1225c4028958SDavid Howells struct ocfs2_journal *journal = 1226c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1227c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1228ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1229800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 12302205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1231ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 1232ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1233ccd979bdSMark Fasheh 1234b4107950STao Ma trace_ocfs2_complete_recovery( 1235b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1236ccd979bdSMark Fasheh 1237ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1238ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1239ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1240ccd979bdSMark Fasheh 1241800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1242ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1243ccd979bdSMark Fasheh 124419ece546SJan Kara ocfs2_wait_on_quotas(osb); 124519ece546SJan Kara 1246ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1247b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1248b4107950STao Ma qrec = item->lri_qrec; 1249ed460cffSJoseph Qi orphan_reco_type = item->lri_orphan_reco_type; 1250ccd979bdSMark Fasheh 1251b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1252b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1253b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1254b4107950STao Ma qrec); 1255b4107950STao Ma 1256b4107950STao Ma if (la_dinode) { 1257ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1258ccd979bdSMark Fasheh la_dinode); 1259ccd979bdSMark Fasheh if (ret < 0) 1260ccd979bdSMark Fasheh mlog_errno(ret); 1261ccd979bdSMark Fasheh 1262ccd979bdSMark Fasheh kfree(la_dinode); 1263ccd979bdSMark Fasheh } 1264ccd979bdSMark Fasheh 1265ccd979bdSMark Fasheh if (tl_dinode) { 1266ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1267ccd979bdSMark Fasheh tl_dinode); 1268ccd979bdSMark Fasheh if (ret < 0) 1269ccd979bdSMark Fasheh mlog_errno(ret); 1270ccd979bdSMark Fasheh 1271ccd979bdSMark Fasheh kfree(tl_dinode); 1272ccd979bdSMark Fasheh } 1273ccd979bdSMark Fasheh 1274ed460cffSJoseph Qi ret = ocfs2_recover_orphans(osb, item->lri_slot, 1275ed460cffSJoseph Qi orphan_reco_type); 1276ccd979bdSMark Fasheh if (ret < 0) 1277ccd979bdSMark Fasheh mlog_errno(ret); 1278ccd979bdSMark Fasheh 12792205363dSJan Kara if (qrec) { 12802205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 12812205363dSJan Kara item->lri_slot); 12822205363dSJan Kara if (ret < 0) 12832205363dSJan Kara mlog_errno(ret); 12842205363dSJan Kara /* Recovery info is already freed now */ 12852205363dSJan Kara } 12862205363dSJan Kara 1287ccd979bdSMark Fasheh kfree(item); 1288ccd979bdSMark Fasheh } 1289ccd979bdSMark Fasheh 1290b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1291ccd979bdSMark Fasheh } 1292ccd979bdSMark Fasheh 1293ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1294ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1295ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1296ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1297ccd979bdSMark Fasheh int slot_num, 1298ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 12992205363dSJan Kara struct ocfs2_dinode *tl_dinode, 1300ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 1301ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1302ccd979bdSMark Fasheh { 1303ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1304ccd979bdSMark Fasheh 1305afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1306ccd979bdSMark Fasheh if (!item) { 1307ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1308ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1309ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1310ccd979bdSMark Fasheh kfree(la_dinode); 1311ccd979bdSMark Fasheh kfree(tl_dinode); 1312ccd979bdSMark Fasheh 13132205363dSJan Kara if (qrec) 13142205363dSJan Kara ocfs2_free_quota_recovery(qrec); 13152205363dSJan Kara 1316ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1317ccd979bdSMark Fasheh return; 1318ccd979bdSMark Fasheh } 1319ccd979bdSMark Fasheh 1320ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1321ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1322ccd979bdSMark Fasheh item->lri_slot = slot_num; 1323ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 13242205363dSJan Kara item->lri_qrec = qrec; 1325ed460cffSJoseph Qi item->lri_orphan_reco_type = orphan_reco_type; 1326ccd979bdSMark Fasheh 1327ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1328ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 132935ddf78eSjiangyiwen queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); 1330ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1331ccd979bdSMark Fasheh } 1332ccd979bdSMark Fasheh 1333ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 13349140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1335ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1336ccd979bdSMark Fasheh { 1337ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1338ccd979bdSMark Fasheh 133910b3dd76SSunil Mushran if (ocfs2_is_hard_readonly(osb)) 134010b3dd76SSunil Mushran return; 134110b3dd76SSunil Mushran 13429140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 13439140db04SSrinivas Eeda * that */ 13449140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 1345ed460cffSJoseph Qi osb->local_alloc_copy, NULL, NULL, 1346ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 1347ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1348ccd979bdSMark Fasheh 1349ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 13509140db04SSrinivas Eeda 13519140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 13529140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1353ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 13549140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1355ccd979bdSMark Fasheh } 1356ccd979bdSMark Fasheh 13572205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 13582205363dSJan Kara { 13592205363dSJan Kara if (osb->quota_rec) { 13602205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 13612205363dSJan Kara osb->slot_num, 13622205363dSJan Kara NULL, 13632205363dSJan Kara NULL, 1364ed460cffSJoseph Qi osb->quota_rec, 1365ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 13662205363dSJan Kara osb->quota_rec = NULL; 13672205363dSJan Kara } 13682205363dSJan Kara } 13692205363dSJan Kara 1370ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1371ccd979bdSMark Fasheh { 13722205363dSJan Kara int status, node_num, slot_num; 1373ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1374553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 13752205363dSJan Kara int *rm_quota = NULL; 13762205363dSJan Kara int rm_quota_used = 0, i; 13772205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1378ccd979bdSMark Fasheh 137921158ca8SGuozhonghua /* Whether the quota supported. */ 138021158ca8SGuozhonghua int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 138121158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_USRQUOTA) 138221158ca8SGuozhonghua || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 138321158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); 138421158ca8SGuozhonghua 1385ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1386ccd979bdSMark Fasheh if (status < 0) { 1387ccd979bdSMark Fasheh goto bail; 1388ccd979bdSMark Fasheh } 1389ccd979bdSMark Fasheh 139021158ca8SGuozhonghua if (quota_enabled) { 13916396bb22SKees Cook rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); 13922205363dSJan Kara if (!rm_quota) { 13932205363dSJan Kara status = -ENOMEM; 13942205363dSJan Kara goto bail; 13952205363dSJan Kara } 139621158ca8SGuozhonghua } 1397ccd979bdSMark Fasheh restart: 1398ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1399ccd979bdSMark Fasheh if (status < 0) { 1400ccd979bdSMark Fasheh mlog_errno(status); 1401ccd979bdSMark Fasheh goto bail; 1402ccd979bdSMark Fasheh } 1403ccd979bdSMark Fasheh 14049140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 14059140db04SSrinivas Eeda if (status < 0) 14069140db04SSrinivas Eeda mlog_errno(status); 14079140db04SSrinivas Eeda 14089140db04SSrinivas Eeda /* queue recovery for our own slot */ 14099140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1410ed460cffSJoseph Qi NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); 14119140db04SSrinivas Eeda 1412553abd04SJoel Becker spin_lock(&osb->osb_lock); 1413553abd04SJoel Becker while (rm->rm_used) { 1414553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1415553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1416553abd04SJoel Becker node_num = rm->rm_entries[0]; 1417553abd04SJoel Becker spin_unlock(&osb->osb_lock); 14182205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1419b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 14202205363dSJan Kara if (slot_num == -ENOENT) { 14212205363dSJan Kara status = 0; 14222205363dSJan Kara goto skip_recovery; 14232205363dSJan Kara } 1424ccd979bdSMark Fasheh 14252205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 14262205363dSJan Kara * immediately because we have to obtain cluster locks from 14272205363dSJan Kara * quota files and we also don't want to just skip it because 14282205363dSJan Kara * then quota usage would be out of sync until some node takes 14292205363dSJan Kara * the slot. So we remember which nodes need quota recovery 14302205363dSJan Kara * and when everything else is done, we recover quotas. */ 143121158ca8SGuozhonghua if (quota_enabled) { 143221158ca8SGuozhonghua for (i = 0; i < rm_quota_used 143321158ca8SGuozhonghua && rm_quota[i] != slot_num; i++) 143421158ca8SGuozhonghua ; 143521158ca8SGuozhonghua 14362205363dSJan Kara if (i == rm_quota_used) 14372205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 143821158ca8SGuozhonghua } 14392205363dSJan Kara 14402205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 14412205363dSJan Kara skip_recovery: 1442553abd04SJoel Becker if (!status) { 1443553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1444553abd04SJoel Becker } else { 1445ccd979bdSMark Fasheh mlog(ML_ERROR, 1446ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1447ccd979bdSMark Fasheh status, node_num, 1448ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1449ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1450ccd979bdSMark Fasheh } 1451ccd979bdSMark Fasheh 1452553abd04SJoel Becker spin_lock(&osb->osb_lock); 1453ccd979bdSMark Fasheh } 1454553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1455b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1456553abd04SJoel Becker 1457539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1458539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1459539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1460539d8264SSunil Mushran if (status < 0) 1461539d8264SSunil Mushran mlog_errno(status); 1462539d8264SSunil Mushran 14632205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 14642205363dSJan Kara * superblock lock so that no one can start using the slot (and crash) 14652205363dSJan Kara * before we recover it */ 146621158ca8SGuozhonghua if (quota_enabled) { 14672205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 14682205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 14692205363dSJan Kara if (IS_ERR(qrec)) { 14702205363dSJan Kara status = PTR_ERR(qrec); 14712205363dSJan Kara mlog_errno(status); 14722205363dSJan Kara continue; 14732205363dSJan Kara } 147421158ca8SGuozhonghua ocfs2_queue_recovery_completion(osb->journal, 147521158ca8SGuozhonghua rm_quota[i], 1476ed460cffSJoseph Qi NULL, NULL, qrec, 1477ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 14782205363dSJan Kara } 147921158ca8SGuozhonghua } 14802205363dSJan Kara 1481ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1482ccd979bdSMark Fasheh 14839140db04SSrinivas Eeda /* queue recovery for offline slots */ 1484ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1485ccd979bdSMark Fasheh 1486ccd979bdSMark Fasheh bail: 1487c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1488553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1489c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1490ccd979bdSMark Fasheh goto restart; 1491ccd979bdSMark Fasheh } 1492ccd979bdSMark Fasheh 14939140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1494ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1495ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1496ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1497ccd979bdSMark Fasheh 1498c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1499ccd979bdSMark Fasheh 150021158ca8SGuozhonghua if (quota_enabled) 15012205363dSJan Kara kfree(rm_quota); 15022205363dSJan Kara 1503ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1504ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1505ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1506ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1507ccd979bdSMark Fasheh } 1508ccd979bdSMark Fasheh 1509ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1510ccd979bdSMark Fasheh { 1511c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1512b4107950STao Ma 1513b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1514b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1515b4107950STao Ma osb->disable_recovery ? 1516b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1517b4107950STao Ma 1518ccd979bdSMark Fasheh if (osb->disable_recovery) 1519ccd979bdSMark Fasheh goto out; 1520ccd979bdSMark Fasheh 1521ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1522ccd979bdSMark Fasheh goto out; 1523ccd979bdSMark Fasheh 1524ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 15255afc44e2SJoseph Qi "ocfs2rec-%s", osb->uuid_str); 1526ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1527ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1528ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1529ccd979bdSMark Fasheh } 1530ccd979bdSMark Fasheh 1531ccd979bdSMark Fasheh out: 1532c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1533ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1534ccd979bdSMark Fasheh } 1535ccd979bdSMark Fasheh 1536539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1537539d8264SSunil Mushran int slot_num, 1538539d8264SSunil Mushran struct buffer_head **bh, 1539539d8264SSunil Mushran struct inode **ret_inode) 1540539d8264SSunil Mushran { 1541539d8264SSunil Mushran int status = -EACCES; 1542539d8264SSunil Mushran struct inode *inode = NULL; 1543539d8264SSunil Mushran 1544539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1545539d8264SSunil Mushran 1546539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1547539d8264SSunil Mushran slot_num); 1548539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1549539d8264SSunil Mushran mlog_errno(status); 1550539d8264SSunil Mushran goto bail; 1551539d8264SSunil Mushran } 1552539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1553539d8264SSunil Mushran 1554b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1555539d8264SSunil Mushran if (status < 0) { 1556539d8264SSunil Mushran mlog_errno(status); 1557539d8264SSunil Mushran goto bail; 1558539d8264SSunil Mushran } 1559539d8264SSunil Mushran 1560539d8264SSunil Mushran status = 0; 1561539d8264SSunil Mushran 1562539d8264SSunil Mushran bail: 1563539d8264SSunil Mushran if (inode) { 1564539d8264SSunil Mushran if (status || !ret_inode) 1565539d8264SSunil Mushran iput(inode); 1566539d8264SSunil Mushran else 1567539d8264SSunil Mushran *ret_inode = inode; 1568539d8264SSunil Mushran } 1569539d8264SSunil Mushran return status; 1570539d8264SSunil Mushran } 1571539d8264SSunil Mushran 1572ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1573ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1574ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1575ccd979bdSMark Fasheh int node_num, 1576ccd979bdSMark Fasheh int slot_num) 1577ccd979bdSMark Fasheh { 1578ccd979bdSMark Fasheh int status; 1579ccd979bdSMark Fasheh int got_lock = 0; 1580ccd979bdSMark Fasheh unsigned int flags; 1581ccd979bdSMark Fasheh struct inode *inode = NULL; 1582ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1583ccd979bdSMark Fasheh journal_t *journal = NULL; 1584ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1585539d8264SSunil Mushran u32 slot_reco_gen; 1586ccd979bdSMark Fasheh 1587539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1588539d8264SSunil Mushran if (status) { 1589ccd979bdSMark Fasheh mlog_errno(status); 1590ccd979bdSMark Fasheh goto done; 1591ccd979bdSMark Fasheh } 1592539d8264SSunil Mushran 1593539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1594539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1595539d8264SSunil Mushran brelse(bh); 1596539d8264SSunil Mushran bh = NULL; 1597539d8264SSunil Mushran 1598539d8264SSunil Mushran /* 1599539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1600539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1601539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1602539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1603539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1604539d8264SSunil Mushran * If not, it needs recovery. 1605539d8264SSunil Mushran */ 1606539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1607b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1608539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1609539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1610539d8264SSunil Mushran status = -EBUSY; 1611ccd979bdSMark Fasheh goto done; 1612ccd979bdSMark Fasheh } 1613539d8264SSunil Mushran 1614539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1615ccd979bdSMark Fasheh 1616e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1617ccd979bdSMark Fasheh if (status < 0) { 1618b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1619ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1620ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1621ccd979bdSMark Fasheh goto done; 1622ccd979bdSMark Fasheh } 1623ccd979bdSMark Fasheh got_lock = 1; 1624ccd979bdSMark Fasheh 1625ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1626ccd979bdSMark Fasheh 1627ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1628539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1629ccd979bdSMark Fasheh 1630ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1631b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1632539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1633539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1634ccd979bdSMark Fasheh goto done; 1635ccd979bdSMark Fasheh } 1636ccd979bdSMark Fasheh 16379140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 16389140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 16399140db04SSrinivas Eeda 1640619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1641619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1642619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1643ccd979bdSMark Fasheh 1644ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1645ccd979bdSMark Fasheh 1646ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1647ccd979bdSMark Fasheh if (status < 0) { 1648ccd979bdSMark Fasheh mlog_errno(status); 1649ccd979bdSMark Fasheh goto done; 1650ccd979bdSMark Fasheh } 1651ccd979bdSMark Fasheh 16522b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1653ccd979bdSMark Fasheh if (journal == NULL) { 1654ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1655ccd979bdSMark Fasheh status = -EIO; 1656ccd979bdSMark Fasheh goto done; 1657ccd979bdSMark Fasheh } 1658ccd979bdSMark Fasheh 16592b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1660ccd979bdSMark Fasheh if (status < 0) { 1661ccd979bdSMark Fasheh mlog_errno(status); 1662ccd979bdSMark Fasheh if (!igrab(inode)) 1663ccd979bdSMark Fasheh BUG(); 16642b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1665ccd979bdSMark Fasheh goto done; 1666ccd979bdSMark Fasheh } 1667ccd979bdSMark Fasheh 1668ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1669ccd979bdSMark Fasheh 1670ccd979bdSMark Fasheh /* wipe the journal */ 16712b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 16722b4e30fbSJoel Becker status = jbd2_journal_flush(journal); 16732b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1674ccd979bdSMark Fasheh if (status < 0) 1675ccd979bdSMark Fasheh mlog_errno(status); 1676ccd979bdSMark Fasheh 1677ccd979bdSMark Fasheh /* This will mark the node clean */ 1678ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1679ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1680ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1681ccd979bdSMark Fasheh 1682539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1683539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1684539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1685539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1686539d8264SSunil Mushran 168713723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 16888cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1689ccd979bdSMark Fasheh if (status < 0) 1690ccd979bdSMark Fasheh mlog_errno(status); 1691ccd979bdSMark Fasheh 1692ccd979bdSMark Fasheh if (!igrab(inode)) 1693ccd979bdSMark Fasheh BUG(); 1694ccd979bdSMark Fasheh 16952b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1696ccd979bdSMark Fasheh 1697619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1698619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1699619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1700ccd979bdSMark Fasheh done: 1701ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1702ccd979bdSMark Fasheh if (got_lock) 1703e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1704ccd979bdSMark Fasheh 1705ccd979bdSMark Fasheh iput(inode); 1706ccd979bdSMark Fasheh brelse(bh); 1707ccd979bdSMark Fasheh 1708ccd979bdSMark Fasheh return status; 1709ccd979bdSMark Fasheh } 1710ccd979bdSMark Fasheh 1711ccd979bdSMark Fasheh /* 1712ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1713ccd979bdSMark Fasheh * - Replay it's journal 1714ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1715ccd979bdSMark Fasheh * - Stamp a clean truncate log 1716ccd979bdSMark Fasheh * - Mark the node clean 1717ccd979bdSMark Fasheh * 1718ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1719ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1720ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1721ccd979bdSMark Fasheh * far less concerning. 1722ccd979bdSMark Fasheh */ 1723ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 17242205363dSJan Kara int node_num, int slot_num) 1725ccd979bdSMark Fasheh { 1726ccd979bdSMark Fasheh int status = 0; 1727ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1728ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1729ccd979bdSMark Fasheh 1730b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1731ccd979bdSMark Fasheh 1732ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1733ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1734ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1735ccd979bdSMark Fasheh 1736ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1737ccd979bdSMark Fasheh if (status < 0) { 1738539d8264SSunil Mushran if (status == -EBUSY) { 1739b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1740539d8264SSunil Mushran status = 0; 1741539d8264SSunil Mushran goto done; 1742539d8264SSunil Mushran } 1743ccd979bdSMark Fasheh mlog_errno(status); 1744ccd979bdSMark Fasheh goto done; 1745ccd979bdSMark Fasheh } 1746ccd979bdSMark Fasheh 1747ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1748ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1749ccd979bdSMark Fasheh if (status < 0) { 1750ccd979bdSMark Fasheh mlog_errno(status); 1751ccd979bdSMark Fasheh goto done; 1752ccd979bdSMark Fasheh } 1753ccd979bdSMark Fasheh 1754ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1755ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1756ccd979bdSMark Fasheh * recovery. */ 1757ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1758ccd979bdSMark Fasheh if (status < 0) 1759ccd979bdSMark Fasheh mlog_errno(status); 1760ccd979bdSMark Fasheh 1761ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1762ccd979bdSMark Fasheh * harmful place to get an error... */ 17638e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1764ccd979bdSMark Fasheh if (status < 0) 1765ccd979bdSMark Fasheh mlog_errno(status); 1766ccd979bdSMark Fasheh 1767ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1768ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1769ed460cffSJoseph Qi tl_copy, NULL, ORPHAN_NEED_TRUNCATE); 1770ccd979bdSMark Fasheh 1771ccd979bdSMark Fasheh status = 0; 1772ccd979bdSMark Fasheh done: 1773ccd979bdSMark Fasheh 1774ccd979bdSMark Fasheh return status; 1775ccd979bdSMark Fasheh } 1776ccd979bdSMark Fasheh 1777ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1778ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1779ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1780ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1781ccd979bdSMark Fasheh int slot_num) 1782ccd979bdSMark Fasheh { 1783ccd979bdSMark Fasheh int status, flags; 1784ccd979bdSMark Fasheh struct inode *inode = NULL; 1785ccd979bdSMark Fasheh 1786ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1787ccd979bdSMark Fasheh slot_num); 1788ccd979bdSMark Fasheh if (inode == NULL) { 1789ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1790ccd979bdSMark Fasheh status = -EACCES; 1791ccd979bdSMark Fasheh goto bail; 1792ccd979bdSMark Fasheh } 1793ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1794ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1795ccd979bdSMark Fasheh iput(inode); 1796ccd979bdSMark Fasheh inode = NULL; 1797ccd979bdSMark Fasheh status = -EACCES; 1798ccd979bdSMark Fasheh goto bail; 1799ccd979bdSMark Fasheh } 1800ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1801ccd979bdSMark Fasheh 1802ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1803e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1804ccd979bdSMark Fasheh if (status < 0) { 1805ccd979bdSMark Fasheh if (status != -EAGAIN) 1806ccd979bdSMark Fasheh mlog_errno(status); 1807ccd979bdSMark Fasheh goto bail; 1808ccd979bdSMark Fasheh } 1809ccd979bdSMark Fasheh 1810e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1811ccd979bdSMark Fasheh bail: 1812ccd979bdSMark Fasheh iput(inode); 1813ccd979bdSMark Fasheh 1814ccd979bdSMark Fasheh return status; 1815ccd979bdSMark Fasheh } 1816ccd979bdSMark Fasheh 1817ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1818ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1819ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1820ccd979bdSMark Fasheh { 1821d85b20e4SJoel Becker unsigned int node_num; 1822d85b20e4SJoel Becker int status, i; 1823a1af7d15SMark Fasheh u32 gen; 1824539d8264SSunil Mushran struct buffer_head *bh = NULL; 1825539d8264SSunil Mushran struct ocfs2_dinode *di; 1826ccd979bdSMark Fasheh 1827ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1828ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1829ccd979bdSMark Fasheh 1830d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1831539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1832539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1833539d8264SSunil Mushran if (status) { 1834539d8264SSunil Mushran mlog_errno(status); 1835539d8264SSunil Mushran goto bail; 1836539d8264SSunil Mushran } 1837539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1838a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1839539d8264SSunil Mushran brelse(bh); 1840539d8264SSunil Mushran bh = NULL; 1841539d8264SSunil Mushran 1842a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1843a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1844a1af7d15SMark Fasheh 1845b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1846539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1847539d8264SSunil Mushran 1848a1af7d15SMark Fasheh if (i == osb->slot_num) { 1849a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1850ccd979bdSMark Fasheh continue; 1851a1af7d15SMark Fasheh } 1852d85b20e4SJoel Becker 1853d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1854a1af7d15SMark Fasheh if (status == -ENOENT) { 1855a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1856ccd979bdSMark Fasheh continue; 1857a1af7d15SMark Fasheh } 1858ccd979bdSMark Fasheh 1859a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1860a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1861ccd979bdSMark Fasheh continue; 1862a1af7d15SMark Fasheh } 1863d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1864ccd979bdSMark Fasheh 1865ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1866ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1867ccd979bdSMark Fasheh * file here to test if he's alive. */ 1868ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1869ccd979bdSMark Fasheh if (!status) { 1870ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1871ccd979bdSMark Fasheh * the recovery thread can't race us on 1872ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1873ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1874ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1875ccd979bdSMark Fasheh mlog_errno(status); 1876ccd979bdSMark Fasheh goto bail; 1877ccd979bdSMark Fasheh } 1878ccd979bdSMark Fasheh } 1879ccd979bdSMark Fasheh 1880ccd979bdSMark Fasheh status = 0; 1881ccd979bdSMark Fasheh bail: 1882ccd979bdSMark Fasheh return status; 1883ccd979bdSMark Fasheh } 1884ccd979bdSMark Fasheh 188583273932SSrinivas Eeda /* 188683273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 188783273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 188883273932SSrinivas Eeda * same time. 188983273932SSrinivas Eeda */ 189083273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 189183273932SSrinivas Eeda { 189283273932SSrinivas Eeda unsigned long time; 189383273932SSrinivas Eeda 189483273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 189583273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 189683273932SSrinivas Eeda return msecs_to_jiffies(time); 189783273932SSrinivas Eeda } 189883273932SSrinivas Eeda 189983273932SSrinivas Eeda /* 190083273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 190183273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 190283273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 190383273932SSrinivas Eeda * 1904a035bff6SSunil Mushran * It scans all slots, even ones that are in use. It does so to handle the 1905a035bff6SSunil Mushran * case described below: 1906a035bff6SSunil Mushran * 1907a035bff6SSunil Mushran * Node 1 has an inode it was using. The dentry went away due to memory 1908a035bff6SSunil Mushran * pressure. Node 1 closes the inode, but it's on the free list. The node 1909a035bff6SSunil Mushran * has the open lock. 1910a035bff6SSunil Mushran * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1911a035bff6SSunil Mushran * but node 1 has no dentry and doesn't get the message. It trylocks the 1912a035bff6SSunil Mushran * open lock, sees that another node has a PR, and does nothing. 1913a035bff6SSunil Mushran * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1914a035bff6SSunil Mushran * open lock, sees the PR still, and does nothing. 1915a035bff6SSunil Mushran * Basically, we have to trigger an orphan iput on node 1. The only way 1916a035bff6SSunil Mushran * for this to happen is if node 1 runs node 2's orphan dir. 1917a035bff6SSunil Mushran * 191883273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 191983273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 192083273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 192183273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 192283273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 192383273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 192483273932SSrinivas Eeda * sequence number in the LVB. 192583273932SSrinivas Eeda */ 1926b519ea6dSJoseph Qi static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 192783273932SSrinivas Eeda { 192883273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 192983273932SSrinivas Eeda int status, i; 193083273932SSrinivas Eeda u32 seqno = 0; 193183273932SSrinivas Eeda 193283273932SSrinivas Eeda os = &osb->osb_orphan_scan; 193383273932SSrinivas Eeda 1934692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1935692684e1SSunil Mushran goto out; 1936692684e1SSunil Mushran 1937b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1938b4107950STao Ma atomic_read(&os->os_state)); 1939b4107950STao Ma 1940df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 194183273932SSrinivas Eeda if (status < 0) { 194283273932SSrinivas Eeda if (status != -EAGAIN) 194383273932SSrinivas Eeda mlog_errno(status); 194483273932SSrinivas Eeda goto out; 194583273932SSrinivas Eeda } 194683273932SSrinivas Eeda 1947692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1948692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1949692684e1SSunil Mushran goto unlock; 1950692684e1SSunil Mushran 195183273932SSrinivas Eeda if (os->os_seqno != seqno) { 195283273932SSrinivas Eeda os->os_seqno = seqno; 195383273932SSrinivas Eeda goto unlock; 195483273932SSrinivas Eeda } 195583273932SSrinivas Eeda 195683273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 195783273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1958ed460cffSJoseph Qi NULL, ORPHAN_NO_NEED_TRUNCATE); 195983273932SSrinivas Eeda /* 196083273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 196183273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 196283273932SSrinivas Eeda */ 196383273932SSrinivas Eeda seqno++; 196415633a22SSrinivas Eeda os->os_count++; 1965395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 196683273932SSrinivas Eeda unlock: 1967df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 196883273932SSrinivas Eeda out: 1969b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1970b4107950STao Ma atomic_read(&os->os_state)); 197183273932SSrinivas Eeda return; 197283273932SSrinivas Eeda } 197383273932SSrinivas Eeda 197483273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 1975b519ea6dSJoseph Qi static void ocfs2_orphan_scan_work(struct work_struct *work) 197683273932SSrinivas Eeda { 197783273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 197883273932SSrinivas Eeda struct ocfs2_super *osb; 197983273932SSrinivas Eeda 198083273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 198183273932SSrinivas Eeda os_orphan_scan_work.work); 198283273932SSrinivas Eeda osb = os->os_osb; 198383273932SSrinivas Eeda 198483273932SSrinivas Eeda mutex_lock(&os->os_lock); 198583273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 1986692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 198735ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 198883273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 198983273932SSrinivas Eeda mutex_unlock(&os->os_lock); 199083273932SSrinivas Eeda } 199183273932SSrinivas Eeda 199283273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 199383273932SSrinivas Eeda { 199483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 199583273932SSrinivas Eeda 199683273932SSrinivas Eeda os = &osb->osb_orphan_scan; 1997df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1998692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 199983273932SSrinivas Eeda mutex_lock(&os->os_lock); 200083273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 200183273932SSrinivas Eeda mutex_unlock(&os->os_lock); 200283273932SSrinivas Eeda } 2003df152c24SSunil Mushran } 200483273932SSrinivas Eeda 2005df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 200683273932SSrinivas Eeda { 200783273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 200883273932SSrinivas Eeda 200983273932SSrinivas Eeda os = &osb->osb_orphan_scan; 201083273932SSrinivas Eeda os->os_osb = osb; 201115633a22SSrinivas Eeda os->os_count = 0; 20123211949fSSunil Mushran os->os_seqno = 0; 201383273932SSrinivas Eeda mutex_init(&os->os_lock); 2014df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 20158b712cd5SJeff Mahoney } 201683273932SSrinivas Eeda 20178b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 20188b712cd5SJeff Mahoney { 20198b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 20208b712cd5SJeff Mahoney 20218b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 2022395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 2023df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 2024df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 2025df152c24SSunil Mushran else { 2026df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 202735ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 202883273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 2029df152c24SSunil Mushran } 203083273932SSrinivas Eeda } 203183273932SSrinivas Eeda 20325eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 20333704412bSAl Viro struct dir_context ctx; 20345eae5b96SMark Fasheh struct inode *head; 20355eae5b96SMark Fasheh struct ocfs2_super *osb; 203630edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 20375eae5b96SMark Fasheh }; 20385eae5b96SMark Fasheh 2039ac7576f4SMiklos Szeredi static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, 2040ac7576f4SMiklos Szeredi int name_len, loff_t pos, u64 ino, 2041ac7576f4SMiklos Szeredi unsigned type) 20425eae5b96SMark Fasheh { 2043ac7576f4SMiklos Szeredi struct ocfs2_orphan_filldir_priv *p = 2044ac7576f4SMiklos Szeredi container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); 20455eae5b96SMark Fasheh struct inode *iter; 20465eae5b96SMark Fasheh 20475eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 20485eae5b96SMark Fasheh return 0; 20495eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 20505eae5b96SMark Fasheh return 0; 20515eae5b96SMark Fasheh 205230edc43cSJoseph Qi /* do not include dio entry in case of orphan scan */ 205330edc43cSJoseph Qi if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && 205430edc43cSJoseph Qi (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 205530edc43cSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN))) 205630edc43cSJoseph Qi return 0; 205730edc43cSJoseph Qi 20585eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 20595eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 20605fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 20615eae5b96SMark Fasheh if (IS_ERR(iter)) 20625eae5b96SMark Fasheh return 0; 20635eae5b96SMark Fasheh 206493d911fcSJoseph Qi if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 206593d911fcSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN)) 206693d911fcSJoseph Qi OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; 206793d911fcSJoseph Qi 2068ed460cffSJoseph Qi /* Skip inodes which are already added to recover list, since dio may 2069ed460cffSJoseph Qi * happen concurrently with unlink/rename */ 2070ed460cffSJoseph Qi if (OCFS2_I(iter)->ip_next_orphan) { 2071ed460cffSJoseph Qi iput(iter); 2072ed460cffSJoseph Qi return 0; 2073ed460cffSJoseph Qi } 2074ed460cffSJoseph Qi 2075b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 20765eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 20775eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 20785eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 20795eae5b96SMark Fasheh p->head = iter; 20805eae5b96SMark Fasheh 20815eae5b96SMark Fasheh return 0; 20825eae5b96SMark Fasheh } 20835eae5b96SMark Fasheh 2084b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2085b4df6ed8SMark Fasheh int slot, 208630edc43cSJoseph Qi struct inode **head, 208730edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2088ccd979bdSMark Fasheh { 2089b4df6ed8SMark Fasheh int status; 2090ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 20913704412bSAl Viro struct ocfs2_orphan_filldir_priv priv = { 20923704412bSAl Viro .ctx.actor = ocfs2_orphan_filldir, 20933704412bSAl Viro .osb = osb, 209430edc43cSJoseph Qi .head = *head, 209530edc43cSJoseph Qi .orphan_reco_type = orphan_reco_type 20963704412bSAl Viro }; 2097ccd979bdSMark Fasheh 2098ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2099ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 2100ccd979bdSMark Fasheh slot); 2101ccd979bdSMark Fasheh if (!orphan_dir_inode) { 2102ccd979bdSMark Fasheh status = -ENOENT; 2103ccd979bdSMark Fasheh mlog_errno(status); 2104b4df6ed8SMark Fasheh return status; 2105ccd979bdSMark Fasheh } 2106ccd979bdSMark Fasheh 21075955102cSAl Viro inode_lock(orphan_dir_inode); 2108e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2109ccd979bdSMark Fasheh if (status < 0) { 2110ccd979bdSMark Fasheh mlog_errno(status); 2111ccd979bdSMark Fasheh goto out; 2112ccd979bdSMark Fasheh } 2113ccd979bdSMark Fasheh 21143704412bSAl Viro status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 21155eae5b96SMark Fasheh if (status) { 2116ccd979bdSMark Fasheh mlog_errno(status); 2117a86370fbSMark Fasheh goto out_cluster; 2118ccd979bdSMark Fasheh } 2119ccd979bdSMark Fasheh 21205eae5b96SMark Fasheh *head = priv.head; 2121ccd979bdSMark Fasheh 2122a86370fbSMark Fasheh out_cluster: 2123e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 2124b4df6ed8SMark Fasheh out: 21255955102cSAl Viro inode_unlock(orphan_dir_inode); 2126ccd979bdSMark Fasheh iput(orphan_dir_inode); 2127b4df6ed8SMark Fasheh return status; 2128b4df6ed8SMark Fasheh } 2129b4df6ed8SMark Fasheh 2130b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2131b4df6ed8SMark Fasheh int slot) 2132b4df6ed8SMark Fasheh { 2133b4df6ed8SMark Fasheh int ret; 2134b4df6ed8SMark Fasheh 2135b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2136b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2137b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2138b4df6ed8SMark Fasheh return ret; 2139b4df6ed8SMark Fasheh } 2140b4df6ed8SMark Fasheh 2141b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2142b4df6ed8SMark Fasheh int slot) 2143b4df6ed8SMark Fasheh { 2144b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2145b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2146b4df6ed8SMark Fasheh * know to quit early. */ 2147b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2148b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2149b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2150b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2151b4df6ed8SMark Fasheh * them. */ 2152b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2153b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2154b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2155b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2156b4df6ed8SMark Fasheh } 2157b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2158b4df6ed8SMark Fasheh } 2159b4df6ed8SMark Fasheh 2160b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2161b4df6ed8SMark Fasheh int slot) 2162b4df6ed8SMark Fasheh { 2163b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2164b4df6ed8SMark Fasheh } 2165b4df6ed8SMark Fasheh 2166b4df6ed8SMark Fasheh /* 2167b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2168b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2169b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2170b4df6ed8SMark Fasheh * (most) of the rest of the work. 2171b4df6ed8SMark Fasheh * 2172b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2173b4df6ed8SMark Fasheh * couple of extra considerations. 2174b4df6ed8SMark Fasheh * 2175b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2176b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2177b4df6ed8SMark Fasheh * an invalid inode. 2178b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2179b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2180b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2181b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2182b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2183b4df6ed8SMark Fasheh */ 2184b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2185ed460cffSJoseph Qi int slot, 2186ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2187b4df6ed8SMark Fasheh { 2188b4df6ed8SMark Fasheh int ret = 0; 2189b4df6ed8SMark Fasheh struct inode *inode = NULL; 2190b4df6ed8SMark Fasheh struct inode *iter; 2191b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2192cf1776a9SJoseph Qi struct buffer_head *di_bh = NULL; 2193cf1776a9SJoseph Qi struct ocfs2_dinode *di = NULL; 2194b4df6ed8SMark Fasheh 2195b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2196b4df6ed8SMark Fasheh 2197b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 219830edc43cSJoseph Qi ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); 2199b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2200b4df6ed8SMark Fasheh 2201b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2202b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2203b4df6ed8SMark Fasheh if (ret) 2204b4df6ed8SMark Fasheh mlog_errno(ret); 2205ccd979bdSMark Fasheh 2206ccd979bdSMark Fasheh while (inode) { 2207ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2208b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2209b4107950STao Ma (unsigned long long)oi->ip_blkno); 2210ccd979bdSMark Fasheh 2211ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2212ed460cffSJoseph Qi oi->ip_next_orphan = NULL; 2213ccd979bdSMark Fasheh 221493d911fcSJoseph Qi if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { 22155955102cSAl Viro inode_lock(inode); 2216cf1776a9SJoseph Qi ret = ocfs2_rw_lock(inode, 1); 2217cf1776a9SJoseph Qi if (ret < 0) { 2218cf1776a9SJoseph Qi mlog_errno(ret); 221993d911fcSJoseph Qi goto unlock_mutex; 2220cf1776a9SJoseph Qi } 2221ed460cffSJoseph Qi /* 2222ed460cffSJoseph Qi * We need to take and drop the inode lock to 2223ed460cffSJoseph Qi * force read inode from disk. 2224ed460cffSJoseph Qi */ 2225cf1776a9SJoseph Qi ret = ocfs2_inode_lock(inode, &di_bh, 1); 2226ed460cffSJoseph Qi if (ret) { 2227ed460cffSJoseph Qi mlog_errno(ret); 2228cf1776a9SJoseph Qi goto unlock_rw; 2229ed460cffSJoseph Qi } 2230cf1776a9SJoseph Qi 2231cf1776a9SJoseph Qi di = (struct ocfs2_dinode *)di_bh->b_data; 2232ed460cffSJoseph Qi 223393d911fcSJoseph Qi if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { 2234ed460cffSJoseph Qi ret = ocfs2_truncate_file(inode, di_bh, 2235ed460cffSJoseph Qi i_size_read(inode)); 2236ed460cffSJoseph Qi if (ret < 0) { 2237ed460cffSJoseph Qi if (ret != -ENOSPC) 2238ed460cffSJoseph Qi mlog_errno(ret); 2239cf1776a9SJoseph Qi goto unlock_inode; 2240ed460cffSJoseph Qi } 2241ed460cffSJoseph Qi 224293d911fcSJoseph Qi ret = ocfs2_del_inode_from_orphan(osb, inode, 224393d911fcSJoseph Qi di_bh, 0, 0); 2244ed460cffSJoseph Qi if (ret) 2245ed460cffSJoseph Qi mlog_errno(ret); 224693d911fcSJoseph Qi } 2247cf1776a9SJoseph Qi unlock_inode: 2248cf1776a9SJoseph Qi ocfs2_inode_unlock(inode, 1); 2249512f62acSJoseph Qi brelse(di_bh); 2250512f62acSJoseph Qi di_bh = NULL; 2251cf1776a9SJoseph Qi unlock_rw: 2252cf1776a9SJoseph Qi ocfs2_rw_unlock(inode, 1); 225393d911fcSJoseph Qi unlock_mutex: 22545955102cSAl Viro inode_unlock(inode); 225593d911fcSJoseph Qi 225693d911fcSJoseph Qi /* clear dio flag in ocfs2_inode_info */ 225793d911fcSJoseph Qi oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; 225893d911fcSJoseph Qi } else { 225993d911fcSJoseph Qi spin_lock(&oi->ip_lock); 226093d911fcSJoseph Qi /* Set the proper information to get us going into 226193d911fcSJoseph Qi * ocfs2_delete_inode. */ 226293d911fcSJoseph Qi oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 226393d911fcSJoseph Qi spin_unlock(&oi->ip_lock); 226493d911fcSJoseph Qi } 226593d911fcSJoseph Qi 2266ccd979bdSMark Fasheh iput(inode); 2267ccd979bdSMark Fasheh inode = iter; 2268ccd979bdSMark Fasheh } 2269ccd979bdSMark Fasheh 2270b4df6ed8SMark Fasheh return ret; 2271ccd979bdSMark Fasheh } 2272ccd979bdSMark Fasheh 227319ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2274ccd979bdSMark Fasheh { 2275ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2276ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2277ccd979bdSMark Fasheh * or DISABLED. */ 2278ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 227919ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 228019ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2281ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2282ccd979bdSMark Fasheh 2283ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2284ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2285ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2286ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2287b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2288ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2289ccd979bdSMark Fasheh return -EBUSY; 2290ccd979bdSMark Fasheh } 2291ccd979bdSMark Fasheh 2292ccd979bdSMark Fasheh return 0; 2293ccd979bdSMark Fasheh } 2294ccd979bdSMark Fasheh 2295ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2296ccd979bdSMark Fasheh { 2297ccd979bdSMark Fasheh int status; 2298ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2299ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2300ccd979bdSMark Fasheh 2301ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2302ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2303ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2304ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2305ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2306ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2307ccd979bdSMark Fasheh 2308745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2309ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2310745ae8baSMark Fasheh || kthread_should_stop()); 2311ccd979bdSMark Fasheh 2312ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 231355b465b6SJoseph Qi if (status < 0) { 231455b465b6SJoseph Qi static unsigned long abort_warn_time; 231555b465b6SJoseph Qi 231655b465b6SJoseph Qi /* Warn about this once per minute */ 231755b465b6SJoseph Qi if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 231855b465b6SJoseph Qi mlog(ML_ERROR, "status = %d, journal is " 231955b465b6SJoseph Qi "already aborted.\n", status); 232055b465b6SJoseph Qi /* 232155b465b6SJoseph Qi * After ocfs2_commit_cache() fails, j_num_trans has a 232255b465b6SJoseph Qi * non-zero value. Sleep here to avoid a busy-wait 232355b465b6SJoseph Qi * loop. 232455b465b6SJoseph Qi */ 232555b465b6SJoseph Qi msleep_interruptible(1000); 232655b465b6SJoseph Qi } 2327ccd979bdSMark Fasheh 2328ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2329ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2330ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2331ccd979bdSMark Fasheh "shutdown\n", 2332ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2333ccd979bdSMark Fasheh } 2334ccd979bdSMark Fasheh } 2335ccd979bdSMark Fasheh 2336ccd979bdSMark Fasheh return 0; 2337ccd979bdSMark Fasheh } 2338ccd979bdSMark Fasheh 2339539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2340539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2341539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2342539d8264SSunil Mushran * a journal has been recovered by another node. 2343539d8264SSunil Mushran */ 2344ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2345ccd979bdSMark Fasheh { 2346ccd979bdSMark Fasheh int ret = 0; 2347ccd979bdSMark Fasheh unsigned int slot; 2348539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2349ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2350539d8264SSunil Mushran int journal_dirty = 0; 2351ccd979bdSMark Fasheh 2352ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2353539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2354539d8264SSunil Mushran if (ret) { 2355ccd979bdSMark Fasheh mlog_errno(ret); 2356ccd979bdSMark Fasheh goto out; 2357ccd979bdSMark Fasheh } 2358ccd979bdSMark Fasheh 2359ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2360ccd979bdSMark Fasheh 2361539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2362539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2363539d8264SSunil Mushran 2364ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2365ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2366539d8264SSunil Mushran journal_dirty = 1; 2367ccd979bdSMark Fasheh 2368ccd979bdSMark Fasheh brelse(di_bh); 2369539d8264SSunil Mushran di_bh = NULL; 2370ccd979bdSMark Fasheh } 2371ccd979bdSMark Fasheh 2372ccd979bdSMark Fasheh out: 2373539d8264SSunil Mushran if (journal_dirty) 2374539d8264SSunil Mushran ret = -EROFS; 2375ccd979bdSMark Fasheh return ret; 2376ccd979bdSMark Fasheh } 2377