1328970deSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 2ccd979bdSMark Fasheh /* -*- mode: c; c-basic-offset: 8; -*- 3ccd979bdSMark Fasheh * vim: noexpandtab sw=8 ts=8 sts=0: 4ccd979bdSMark Fasheh * 5ccd979bdSMark Fasheh * journal.c 6ccd979bdSMark Fasheh * 7ccd979bdSMark Fasheh * Defines functions of journalling api 8ccd979bdSMark Fasheh * 9ccd979bdSMark Fasheh * Copyright (C) 2003, 2004 Oracle. All rights reserved. 10ccd979bdSMark Fasheh */ 11ccd979bdSMark Fasheh 12ccd979bdSMark Fasheh #include <linux/fs.h> 13ccd979bdSMark Fasheh #include <linux/types.h> 14ccd979bdSMark Fasheh #include <linux/slab.h> 15ccd979bdSMark Fasheh #include <linux/highmem.h> 16ccd979bdSMark Fasheh #include <linux/kthread.h> 1783273932SSrinivas Eeda #include <linux/time.h> 1883273932SSrinivas Eeda #include <linux/random.h> 1955b465b6SJoseph Qi #include <linux/delay.h> 20ccd979bdSMark Fasheh 21ccd979bdSMark Fasheh #include <cluster/masklog.h> 22ccd979bdSMark Fasheh 23ccd979bdSMark Fasheh #include "ocfs2.h" 24ccd979bdSMark Fasheh 25ccd979bdSMark Fasheh #include "alloc.h" 2650655ae9SJoel Becker #include "blockcheck.h" 27316f4b9fSMark Fasheh #include "dir.h" 28ccd979bdSMark Fasheh #include "dlmglue.h" 29ccd979bdSMark Fasheh #include "extent_map.h" 30ccd979bdSMark Fasheh #include "heartbeat.h" 31ccd979bdSMark Fasheh #include "inode.h" 32ccd979bdSMark Fasheh #include "journal.h" 33ccd979bdSMark Fasheh #include "localalloc.h" 34ccd979bdSMark Fasheh #include "slot_map.h" 35ccd979bdSMark Fasheh #include "super.h" 36ccd979bdSMark Fasheh #include "sysfile.h" 370cf2f763SJoel Becker #include "uptodate.h" 382205363dSJan Kara #include "quota.h" 39ed460cffSJoseph Qi #include "file.h" 40ed460cffSJoseph Qi #include "namei.h" 41ccd979bdSMark Fasheh 42ccd979bdSMark Fasheh #include "buffer_head_io.h" 43b4107950STao Ma #include "ocfs2_trace.h" 44ccd979bdSMark Fasheh 4534af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 46ccd979bdSMark Fasheh 4783273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 4883273932SSrinivas Eeda 49ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 50ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 512205363dSJan Kara int node_num, int slot_num); 52ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 53ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 5419ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 55ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 56539d8264SSunil Mushran int dirty, int replayed); 57ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 58ccd979bdSMark Fasheh int slot_num); 59ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 60ed460cffSJoseph Qi int slot, 61ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 62ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 639140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 649140db04SSrinivas Eeda int slot_num, 659140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 669140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 67ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 68ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 69ccd979bdSMark Fasheh 7019ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 7119ece546SJan Kara { 7219ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 7319ece546SJan Kara } 7419ece546SJan Kara 7519ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 7619ece546SJan Kara { 7719ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 7819ece546SJan Kara } 7919ece546SJan Kara 809140db04SSrinivas Eeda /* 819140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 829140db04SSrinivas Eeda * offline slots during recovery and mount 839140db04SSrinivas Eeda */ 849140db04SSrinivas Eeda 859140db04SSrinivas Eeda enum ocfs2_replay_state { 869140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 879140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 889140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 899140db04SSrinivas Eeda }; 909140db04SSrinivas Eeda 919140db04SSrinivas Eeda struct ocfs2_replay_map { 929140db04SSrinivas Eeda unsigned int rm_slots; 939140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 943c9210d4SGustavo A. R. Silva unsigned char rm_replay_slots[]; 959140db04SSrinivas Eeda }; 969140db04SSrinivas Eeda 97b519ea6dSJoseph Qi static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 989140db04SSrinivas Eeda { 999140db04SSrinivas Eeda if (!osb->replay_map) 1009140db04SSrinivas Eeda return; 1019140db04SSrinivas Eeda 1029140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1039140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1049140db04SSrinivas Eeda return; 1059140db04SSrinivas Eeda 1069140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1079140db04SSrinivas Eeda } 1089140db04SSrinivas Eeda 1099140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1109140db04SSrinivas Eeda { 1119140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1129140db04SSrinivas Eeda int i, node_num; 1139140db04SSrinivas Eeda 1149140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1159140db04SSrinivas Eeda if (osb->replay_map) 1169140db04SSrinivas Eeda return 0; 1179140db04SSrinivas Eeda 1189140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1199140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1209140db04SSrinivas Eeda 1219140db04SSrinivas Eeda if (!replay_map) { 1229140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1239140db04SSrinivas Eeda return -ENOMEM; 1249140db04SSrinivas Eeda } 1259140db04SSrinivas Eeda 1269140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1279140db04SSrinivas Eeda 1289140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1299140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1309140db04SSrinivas Eeda 1319140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1329140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1339140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1349140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1359140db04SSrinivas Eeda } 1369140db04SSrinivas Eeda 1379140db04SSrinivas Eeda osb->replay_map = replay_map; 1389140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1399140db04SSrinivas Eeda return 0; 1409140db04SSrinivas Eeda } 1419140db04SSrinivas Eeda 142b519ea6dSJoseph Qi static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, 143ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1449140db04SSrinivas Eeda { 1459140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1469140db04SSrinivas Eeda int i; 1479140db04SSrinivas Eeda 1489140db04SSrinivas Eeda if (!replay_map) 1499140db04SSrinivas Eeda return; 1509140db04SSrinivas Eeda 1519140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1529140db04SSrinivas Eeda return; 1539140db04SSrinivas Eeda 1549140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1559140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1569140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 157ed460cffSJoseph Qi NULL, NULL, 158ed460cffSJoseph Qi orphan_reco_type); 1599140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1609140db04SSrinivas Eeda } 1619140db04SSrinivas Eeda 162b519ea6dSJoseph Qi static void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1639140db04SSrinivas Eeda { 1649140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1659140db04SSrinivas Eeda 1669140db04SSrinivas Eeda if (!osb->replay_map) 1679140db04SSrinivas Eeda return; 1689140db04SSrinivas Eeda 1699140db04SSrinivas Eeda kfree(replay_map); 1709140db04SSrinivas Eeda osb->replay_map = NULL; 1719140db04SSrinivas Eeda } 1729140db04SSrinivas Eeda 173553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 174553abd04SJoel Becker { 175553abd04SJoel Becker struct ocfs2_recovery_map *rm; 176553abd04SJoel Becker 177553abd04SJoel Becker mutex_init(&osb->recovery_lock); 178553abd04SJoel Becker osb->disable_recovery = 0; 179553abd04SJoel Becker osb->recovery_thread_task = NULL; 180553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 181553abd04SJoel Becker 182553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 183553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 184553abd04SJoel Becker GFP_KERNEL); 185553abd04SJoel Becker if (!rm) { 186553abd04SJoel Becker mlog_errno(-ENOMEM); 187553abd04SJoel Becker return -ENOMEM; 188553abd04SJoel Becker } 189553abd04SJoel Becker 190553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 191553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 192553abd04SJoel Becker osb->recovery_map = rm; 193553abd04SJoel Becker 194553abd04SJoel Becker return 0; 195553abd04SJoel Becker } 196553abd04SJoel Becker 197553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 198553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 199553abd04SJoel Becker * being woken up */ 200553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 201553abd04SJoel Becker { 202553abd04SJoel Becker mb(); 203553abd04SJoel Becker return osb->recovery_thread_task != NULL; 204553abd04SJoel Becker } 205553abd04SJoel Becker 206553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 207553abd04SJoel Becker { 208553abd04SJoel Becker struct ocfs2_recovery_map *rm; 209553abd04SJoel Becker 210553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 211553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 212553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 213553abd04SJoel Becker osb->disable_recovery = 1; 214553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 215553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 216553abd04SJoel Becker 217553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 218553abd04SJoel Becker * launched, so wait for any recovery completion work to 219553abd04SJoel Becker * complete. */ 220b918c430SYi Li if (osb->ocfs2_wq) 22135ddf78eSjiangyiwen flush_workqueue(osb->ocfs2_wq); 222553abd04SJoel Becker 223553abd04SJoel Becker /* 224553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 225553abd04SJoel Becker * freed, the osb_lock is not taken here. 226553abd04SJoel Becker */ 227553abd04SJoel Becker rm = osb->recovery_map; 228553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 229553abd04SJoel Becker 230553abd04SJoel Becker kfree(rm); 231553abd04SJoel Becker } 232553abd04SJoel Becker 233553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 234553abd04SJoel Becker unsigned int node_num) 235553abd04SJoel Becker { 236553abd04SJoel Becker int i; 237553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 238553abd04SJoel Becker 239553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 240553abd04SJoel Becker 241553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 242553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 243553abd04SJoel Becker return 1; 244553abd04SJoel Becker } 245553abd04SJoel Becker 246553abd04SJoel Becker return 0; 247553abd04SJoel Becker } 248553abd04SJoel Becker 249553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 250553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 251553abd04SJoel Becker unsigned int node_num) 252553abd04SJoel Becker { 253553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 254553abd04SJoel Becker 255553abd04SJoel Becker spin_lock(&osb->osb_lock); 256553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 257553abd04SJoel Becker spin_unlock(&osb->osb_lock); 258553abd04SJoel Becker return 1; 259553abd04SJoel Becker } 260553abd04SJoel Becker 261553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 262553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 263553abd04SJoel Becker 264553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 265553abd04SJoel Becker rm->rm_used++; 266553abd04SJoel Becker spin_unlock(&osb->osb_lock); 267553abd04SJoel Becker 268553abd04SJoel Becker return 0; 269553abd04SJoel Becker } 270553abd04SJoel Becker 271553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 272553abd04SJoel Becker unsigned int node_num) 273553abd04SJoel Becker { 274553abd04SJoel Becker int i; 275553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 276553abd04SJoel Becker 277553abd04SJoel Becker spin_lock(&osb->osb_lock); 278553abd04SJoel Becker 279553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 280553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 281553abd04SJoel Becker break; 282553abd04SJoel Becker } 283553abd04SJoel Becker 284553abd04SJoel Becker if (i < rm->rm_used) { 285553abd04SJoel Becker /* XXX: be careful with the pointer math */ 286553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 287553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 288553abd04SJoel Becker rm->rm_used--; 289553abd04SJoel Becker } 290553abd04SJoel Becker 291553abd04SJoel Becker spin_unlock(&osb->osb_lock); 292553abd04SJoel Becker } 293553abd04SJoel Becker 294ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 295ccd979bdSMark Fasheh { 296ccd979bdSMark Fasheh int status = 0; 297ccd979bdSMark Fasheh unsigned int flushed; 298ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 299ccd979bdSMark Fasheh 300ccd979bdSMark Fasheh journal = osb->journal; 301ccd979bdSMark Fasheh 302ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 303ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 304ccd979bdSMark Fasheh 305b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 306b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 307b4107950STao Ma if (flushed == 0) { 308ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 309ccd979bdSMark Fasheh goto finally; 310ccd979bdSMark Fasheh } 311ccd979bdSMark Fasheh 3122b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 3132b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 3142b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 315ccd979bdSMark Fasheh if (status < 0) { 316ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 317ccd979bdSMark Fasheh mlog_errno(status); 318ccd979bdSMark Fasheh goto finally; 319ccd979bdSMark Fasheh } 320ccd979bdSMark Fasheh 321f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 322ccd979bdSMark Fasheh 323ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 324ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 325ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 326ccd979bdSMark Fasheh 327b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 328ccd979bdSMark Fasheh 32934d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 330ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 331ccd979bdSMark Fasheh finally: 332ccd979bdSMark Fasheh return status; 333ccd979bdSMark Fasheh } 334ccd979bdSMark Fasheh 3351fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 336ccd979bdSMark Fasheh { 337ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3381fabe148SMark Fasheh handle_t *handle; 339ccd979bdSMark Fasheh 340ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 341ccd979bdSMark Fasheh 34265eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 34365eff9ccSMark Fasheh return ERR_PTR(-EROFS); 344ccd979bdSMark Fasheh 345ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 346ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 347ccd979bdSMark Fasheh 34890e86a63SJan Kara /* Nested transaction? Just return the handle... */ 34990e86a63SJan Kara if (journal_current_handle()) 35090e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 351ccd979bdSMark Fasheh 352fef6925cSJan Kara sb_start_intwrite(osb->sb); 353fef6925cSJan Kara 354ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 355ccd979bdSMark Fasheh 3562b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3571fabe148SMark Fasheh if (IS_ERR(handle)) { 358ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 359fef6925cSJan Kara sb_end_intwrite(osb->sb); 360ccd979bdSMark Fasheh 3611fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 362ccd979bdSMark Fasheh 363ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 3647ecef14aSJoe Perches ocfs2_abort(osb->sb, "Detected aborted journal\n"); 3651fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 366ccd979bdSMark Fasheh } 367c271c5c2SSunil Mushran } else { 368c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 369ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 370c271c5c2SSunil Mushran } 371ccd979bdSMark Fasheh 372ccd979bdSMark Fasheh return handle; 373ccd979bdSMark Fasheh } 374ccd979bdSMark Fasheh 3751fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3761fabe148SMark Fasheh handle_t *handle) 377ccd979bdSMark Fasheh { 37890e86a63SJan Kara int ret, nested; 37902dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 380ccd979bdSMark Fasheh 381ccd979bdSMark Fasheh BUG_ON(!handle); 382ccd979bdSMark Fasheh 38390e86a63SJan Kara nested = handle->h_ref > 1; 3842b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3851fabe148SMark Fasheh if (ret < 0) 3861fabe148SMark Fasheh mlog_errno(ret); 387ccd979bdSMark Fasheh 388fef6925cSJan Kara if (!nested) { 389ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 390fef6925cSJan Kara sb_end_intwrite(osb->sb); 391fef6925cSJan Kara } 392ccd979bdSMark Fasheh 3931fabe148SMark Fasheh return ret; 394ccd979bdSMark Fasheh } 395ccd979bdSMark Fasheh 396ccd979bdSMark Fasheh /* 397c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 398ccd979bdSMark Fasheh * 3992b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 400e8aed345SMark Fasheh * and then restart the transaction. Before calling 401e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 402e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 403e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 404e8aed345SMark Fasheh * 405ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 406ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 407ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 408ccd979bdSMark Fasheh * 409ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 410ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 411ccd979bdSMark Fasheh * cluster locks associated with this handle. 412ccd979bdSMark Fasheh */ 4131fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 414ccd979bdSMark Fasheh { 415c901fb00STao Ma int status, old_nblocks; 416ccd979bdSMark Fasheh 417ccd979bdSMark Fasheh BUG_ON(!handle); 418c901fb00STao Ma BUG_ON(nblocks < 0); 419ccd979bdSMark Fasheh 420c901fb00STao Ma if (!nblocks) 421c901fb00STao Ma return 0; 422c901fb00STao Ma 4239797a902SJan Kara old_nblocks = jbd2_handle_buffer_credits(handle); 424ccd979bdSMark Fasheh 425b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 426ccd979bdSMark Fasheh 427e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4280879c584SMark Fasheh status = 1; 4290879c584SMark Fasheh #else 430fdc3ef88SJan Kara status = jbd2_journal_extend(handle, nblocks, 0); 431ccd979bdSMark Fasheh if (status < 0) { 432ccd979bdSMark Fasheh mlog_errno(status); 433ccd979bdSMark Fasheh goto bail; 434ccd979bdSMark Fasheh } 4350879c584SMark Fasheh #endif 436ccd979bdSMark Fasheh 437ccd979bdSMark Fasheh if (status > 0) { 438b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 439c901fb00STao Ma status = jbd2_journal_restart(handle, 440c901fb00STao Ma old_nblocks + nblocks); 441ccd979bdSMark Fasheh if (status < 0) { 442ccd979bdSMark Fasheh mlog_errno(status); 443ccd979bdSMark Fasheh goto bail; 444ccd979bdSMark Fasheh } 44501ddf1e1SMark Fasheh } 446ccd979bdSMark Fasheh 447ccd979bdSMark Fasheh status = 0; 448ccd979bdSMark Fasheh bail: 449ccd979bdSMark Fasheh return status; 450ccd979bdSMark Fasheh } 451ccd979bdSMark Fasheh 4522b1e55c3SYounger Liu /* 4532b1e55c3SYounger Liu * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 4542b1e55c3SYounger Liu * If that fails, restart the transaction & regain write access for the 4552b1e55c3SYounger Liu * buffer head which is used for metadata modifications. 4562b1e55c3SYounger Liu * Taken from Ext4: extend_or_restart_transaction() 4572b1e55c3SYounger Liu */ 4582b1e55c3SYounger Liu int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 4592b1e55c3SYounger Liu { 4602b1e55c3SYounger Liu int status, old_nblks; 4612b1e55c3SYounger Liu 4622b1e55c3SYounger Liu BUG_ON(!handle); 4632b1e55c3SYounger Liu 4649797a902SJan Kara old_nblks = jbd2_handle_buffer_credits(handle); 4652b1e55c3SYounger Liu trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 4662b1e55c3SYounger Liu 4672b1e55c3SYounger Liu if (old_nblks < thresh) 4682b1e55c3SYounger Liu return 0; 4692b1e55c3SYounger Liu 470fdc3ef88SJan Kara status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0); 4712b1e55c3SYounger Liu if (status < 0) { 4722b1e55c3SYounger Liu mlog_errno(status); 4732b1e55c3SYounger Liu goto bail; 4742b1e55c3SYounger Liu } 4752b1e55c3SYounger Liu 4762b1e55c3SYounger Liu if (status > 0) { 4772b1e55c3SYounger Liu status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 4782b1e55c3SYounger Liu if (status < 0) 4792b1e55c3SYounger Liu mlog_errno(status); 4802b1e55c3SYounger Liu } 4812b1e55c3SYounger Liu 4822b1e55c3SYounger Liu bail: 4832b1e55c3SYounger Liu return status; 4842b1e55c3SYounger Liu } 4852b1e55c3SYounger Liu 4862b1e55c3SYounger Liu 48750655ae9SJoel Becker struct ocfs2_triggers { 48850655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 48950655ae9SJoel Becker int ot_offset; 49050655ae9SJoel Becker }; 49150655ae9SJoel Becker 49250655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 49350655ae9SJoel Becker { 49450655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 49550655ae9SJoel Becker } 49650655ae9SJoel Becker 49713ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 49850655ae9SJoel Becker struct buffer_head *bh, 49950655ae9SJoel Becker void *data, size_t size) 50050655ae9SJoel Becker { 50150655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 50250655ae9SJoel Becker 50350655ae9SJoel Becker /* 50450655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 50550655ae9SJoel Becker * must unconditionally compute the ecc data. 50650655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 50750655ae9SJoel Becker * metaecc is enabled. 50850655ae9SJoel Becker */ 50950655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 51050655ae9SJoel Becker } 51150655ae9SJoel Becker 51250655ae9SJoel Becker /* 51350655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 51450655ae9SJoel Becker * offset depends on the blocksize. 51550655ae9SJoel Becker */ 51613ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 51750655ae9SJoel Becker struct buffer_head *bh, 51850655ae9SJoel Becker void *data, size_t size) 51950655ae9SJoel Becker { 52050655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 52150655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 52250655ae9SJoel Becker 52350655ae9SJoel Becker /* 52450655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 52550655ae9SJoel Becker * must unconditionally compute the ecc data. 52650655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 52750655ae9SJoel Becker * metaecc is enabled. 52850655ae9SJoel Becker */ 52950655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 53050655ae9SJoel Becker } 53150655ae9SJoel Becker 532c175a518SJoel Becker /* 533c175a518SJoel Becker * Directory blocks also have their own trigger because the 534c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 535c175a518SJoel Becker */ 53613ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 537c175a518SJoel Becker struct buffer_head *bh, 538c175a518SJoel Becker void *data, size_t size) 539c175a518SJoel Becker { 540c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 541c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 542c175a518SJoel Becker 543c175a518SJoel Becker /* 544c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 545c175a518SJoel Becker * must unconditionally compute the ecc data. 546c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 547c175a518SJoel Becker * metaecc is enabled. 548c175a518SJoel Becker */ 549c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 550c175a518SJoel Becker } 551c175a518SJoel Becker 55250655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 55350655ae9SJoel Becker struct buffer_head *bh) 55450655ae9SJoel Becker { 55550655ae9SJoel Becker mlog(ML_ERROR, 55650655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 55750655ae9SJoel Becker "bh->b_blocknr = %llu\n", 55850655ae9SJoel Becker (unsigned long)bh, 55950655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 56050655ae9SJoel Becker 56174e364adSXue jiufei ocfs2_error(bh->b_bdev->bd_super, 56250655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 56350655ae9SJoel Becker } 56450655ae9SJoel Becker 56550655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 56650655ae9SJoel Becker .ot_triggers = { 56713ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 56850655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 56950655ae9SJoel Becker }, 57050655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 57150655ae9SJoel Becker }; 57250655ae9SJoel Becker 57350655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 57450655ae9SJoel Becker .ot_triggers = { 57513ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 57650655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 57750655ae9SJoel Becker }, 57850655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 57950655ae9SJoel Becker }; 58050655ae9SJoel Becker 58193c97087STao Ma static struct ocfs2_triggers rb_triggers = { 58293c97087STao Ma .ot_triggers = { 58313ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58493c97087STao Ma .t_abort = ocfs2_abort_trigger, 58593c97087STao Ma }, 58693c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 58793c97087STao Ma }; 58893c97087STao Ma 58950655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 59050655ae9SJoel Becker .ot_triggers = { 59113ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 59250655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 59350655ae9SJoel Becker }, 59450655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 59550655ae9SJoel Becker }; 59650655ae9SJoel Becker 597c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 598c175a518SJoel Becker .ot_triggers = { 59913ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 600c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 601c175a518SJoel Becker }, 602c175a518SJoel Becker }; 603c175a518SJoel Becker 60450655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 60550655ae9SJoel Becker .ot_triggers = { 60613ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 60750655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 60850655ae9SJoel Becker }, 60950655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 61050655ae9SJoel Becker }; 61150655ae9SJoel Becker 61250655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 61350655ae9SJoel Becker .ot_triggers = { 61413ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 61550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 61650655ae9SJoel Becker }, 61750655ae9SJoel Becker }; 61850655ae9SJoel Becker 6199b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 6209b7895efSMark Fasheh .ot_triggers = { 62113ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6229b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6239b7895efSMark Fasheh }, 6249b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 6259b7895efSMark Fasheh }; 6269b7895efSMark Fasheh 6279b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 6289b7895efSMark Fasheh .ot_triggers = { 62913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6309b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6319b7895efSMark Fasheh }, 6329b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6339b7895efSMark Fasheh }; 6349b7895efSMark Fasheh 63550655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6360cf2f763SJoel Becker struct ocfs2_caching_info *ci, 637ccd979bdSMark Fasheh struct buffer_head *bh, 63850655ae9SJoel Becker struct ocfs2_triggers *triggers, 639ccd979bdSMark Fasheh int type) 640ccd979bdSMark Fasheh { 641ccd979bdSMark Fasheh int status; 6420cf2f763SJoel Becker struct ocfs2_super *osb = 6430cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 644ccd979bdSMark Fasheh 6450cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 646ccd979bdSMark Fasheh BUG_ON(!handle); 647ccd979bdSMark Fasheh BUG_ON(!bh); 648ccd979bdSMark Fasheh 649b4107950STao Ma trace_ocfs2_journal_access( 650b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 651b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 652ccd979bdSMark Fasheh 653ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 654ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 655ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 656d984187eSpiaojun mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", 657d984187eSpiaojun (unsigned long long)bh->b_blocknr, bh->b_state); 658acf8fdbeSJoseph Qi 659acf8fdbeSJoseph Qi lock_buffer(bh); 660acf8fdbeSJoseph Qi /* 661d984187eSpiaojun * A previous transaction with a couple of buffer heads fail 662d984187eSpiaojun * to checkpoint, so all the bhs are marked as BH_Write_EIO. 663d984187eSpiaojun * For current transaction, the bh is just among those error 664d984187eSpiaojun * bhs which previous transaction handle. We can't just clear 665d984187eSpiaojun * its BH_Write_EIO and reuse directly, since other bhs are 666d984187eSpiaojun * not written to disk yet and that will cause metadata 667d984187eSpiaojun * inconsistency. So we should set fs read-only to avoid 668d984187eSpiaojun * further damage. 669acf8fdbeSJoseph Qi */ 670acf8fdbeSJoseph Qi if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { 671acf8fdbeSJoseph Qi unlock_buffer(bh); 672d984187eSpiaojun return ocfs2_error(osb->sb, "A previous attempt to " 673d984187eSpiaojun "write this buffer head failed\n"); 674acf8fdbeSJoseph Qi } 675acf8fdbeSJoseph Qi unlock_buffer(bh); 676ccd979bdSMark Fasheh } 677ccd979bdSMark Fasheh 6780cf2f763SJoel Becker /* Set the current transaction information on the ci so 679ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6800cf2f763SJoel Becker * on this ci or not. We're protected from the commit 681ccd979bdSMark Fasheh * thread updating the current transaction id until 682ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 683ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6840cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 685ccd979bdSMark Fasheh 6860cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 687ccd979bdSMark Fasheh switch (type) { 688ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 689ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 6902b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 691ccd979bdSMark Fasheh break; 692ccd979bdSMark Fasheh 693ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 6942b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 695ccd979bdSMark Fasheh break; 696ccd979bdSMark Fasheh 697ccd979bdSMark Fasheh default: 698ccd979bdSMark Fasheh status = -EINVAL; 699af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 700ccd979bdSMark Fasheh } 7010cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 70250655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 7030cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 704ccd979bdSMark Fasheh 705ccd979bdSMark Fasheh if (status < 0) 706ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 707ccd979bdSMark Fasheh status, type); 708ccd979bdSMark Fasheh 709ccd979bdSMark Fasheh return status; 710ccd979bdSMark Fasheh } 711ccd979bdSMark Fasheh 7120cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 71350655ae9SJoel Becker struct buffer_head *bh, int type) 71450655ae9SJoel Becker { 7150cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 71650655ae9SJoel Becker } 71750655ae9SJoel Becker 7180cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 71950655ae9SJoel Becker struct buffer_head *bh, int type) 72050655ae9SJoel Becker { 7210cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 72250655ae9SJoel Becker } 72350655ae9SJoel Becker 72493c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 72593c97087STao Ma struct buffer_head *bh, int type) 72693c97087STao Ma { 72793c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 72893c97087STao Ma type); 72993c97087STao Ma } 73093c97087STao Ma 7310cf2f763SJoel Becker int ocfs2_journal_access_gd(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, &gd_triggers, type); 73550655ae9SJoel Becker } 73650655ae9SJoel Becker 7370cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 73850655ae9SJoel Becker struct buffer_head *bh, int type) 73950655ae9SJoel Becker { 7400cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 74150655ae9SJoel Becker } 74250655ae9SJoel Becker 7430cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 74450655ae9SJoel Becker struct buffer_head *bh, int type) 74550655ae9SJoel Becker { 7460cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 74750655ae9SJoel Becker } 74850655ae9SJoel Becker 7490cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 75050655ae9SJoel Becker struct buffer_head *bh, int type) 75150655ae9SJoel Becker { 7520cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 75350655ae9SJoel Becker } 75450655ae9SJoel Becker 7550cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7569b7895efSMark Fasheh struct buffer_head *bh, int type) 7579b7895efSMark Fasheh { 7580cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7599b7895efSMark Fasheh } 7609b7895efSMark Fasheh 7610cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7629b7895efSMark Fasheh struct buffer_head *bh, int type) 7639b7895efSMark Fasheh { 7640cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7659b7895efSMark Fasheh } 7669b7895efSMark Fasheh 7670cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 76850655ae9SJoel Becker struct buffer_head *bh, int type) 76950655ae9SJoel Becker { 7700cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 77150655ae9SJoel Becker } 77250655ae9SJoel Becker 773ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 774ccd979bdSMark Fasheh { 775ccd979bdSMark Fasheh int status; 776ccd979bdSMark Fasheh 777b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 778ccd979bdSMark Fasheh 7792b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 780e272e7f0SJoseph Qi if (status) { 781e272e7f0SJoseph Qi mlog_errno(status); 782e272e7f0SJoseph Qi if (!is_handle_aborted(handle)) { 783e272e7f0SJoseph Qi journal_t *journal = handle->h_transaction->t_journal; 784e272e7f0SJoseph Qi struct super_block *sb = bh->b_bdev->bd_super; 785e272e7f0SJoseph Qi 786e272e7f0SJoseph Qi mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " 787e272e7f0SJoseph Qi "Aborting transaction and journal.\n"); 788e272e7f0SJoseph Qi handle->h_err = status; 789e272e7f0SJoseph Qi jbd2_journal_abort_handle(handle); 790e272e7f0SJoseph Qi jbd2_journal_abort(journal, status); 791e272e7f0SJoseph Qi ocfs2_abort(sb, "Journal already aborted.\n"); 792e272e7f0SJoseph Qi } 793e272e7f0SJoseph Qi } 794ccd979bdSMark Fasheh } 795ccd979bdSMark Fasheh 7962b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 797ccd979bdSMark Fasheh 798ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 799ccd979bdSMark Fasheh { 800ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 801d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 802d147b3d6SMark Fasheh 803d147b3d6SMark Fasheh if (osb->osb_commit_interval) 804d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 805ccd979bdSMark Fasheh 806a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 807d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 808ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 8092b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 810ccd979bdSMark Fasheh else 8112b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 812a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 813ccd979bdSMark Fasheh } 814ccd979bdSMark Fasheh 815ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 816ccd979bdSMark Fasheh { 817ccd979bdSMark Fasheh int status = -1; 818ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 819ccd979bdSMark Fasheh journal_t *j_journal = NULL; 820ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 821ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 822ccd979bdSMark Fasheh struct ocfs2_super *osb; 823e63aecb6SMark Fasheh int inode_lock = 0; 824ccd979bdSMark Fasheh 825ccd979bdSMark Fasheh BUG_ON(!journal); 826ccd979bdSMark Fasheh 827ccd979bdSMark Fasheh osb = journal->j_osb; 828ccd979bdSMark Fasheh 829ccd979bdSMark Fasheh /* already have the inode for our journal */ 830ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 831ccd979bdSMark Fasheh osb->slot_num); 832ccd979bdSMark Fasheh if (inode == NULL) { 833ccd979bdSMark Fasheh status = -EACCES; 834ccd979bdSMark Fasheh mlog_errno(status); 835ccd979bdSMark Fasheh goto done; 836ccd979bdSMark Fasheh } 837ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 838ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 839ccd979bdSMark Fasheh iput(inode); 840ccd979bdSMark Fasheh inode = NULL; 841ccd979bdSMark Fasheh status = -EACCES; 842ccd979bdSMark Fasheh goto done; 843ccd979bdSMark Fasheh } 844ccd979bdSMark Fasheh 845ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 846ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 847ccd979bdSMark Fasheh 8486eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 8496eff5790SMark Fasheh * changes in a live cluster so it can be considered an 8506eff5790SMark Fasheh * exception to the rule. */ 851e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 852ccd979bdSMark Fasheh if (status < 0) { 853ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 854ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 855ccd979bdSMark Fasheh goto done; 856ccd979bdSMark Fasheh } 857ccd979bdSMark Fasheh 858e63aecb6SMark Fasheh inode_lock = 1; 859ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 860ccd979bdSMark Fasheh 861f17c20ddSJunxiao Bi if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 862ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 863f17c20ddSJunxiao Bi i_size_read(inode)); 864ccd979bdSMark Fasheh status = -EINVAL; 865ccd979bdSMark Fasheh goto done; 866ccd979bdSMark Fasheh } 867ccd979bdSMark Fasheh 868f17c20ddSJunxiao Bi trace_ocfs2_journal_init(i_size_read(inode), 869b4107950STao Ma (unsigned long long)inode->i_blocks, 870b4107950STao Ma OCFS2_I(inode)->ip_clusters); 871ccd979bdSMark Fasheh 872ccd979bdSMark Fasheh /* call the kernels journal init function now */ 8732b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 874ccd979bdSMark Fasheh if (j_journal == NULL) { 875ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 876ccd979bdSMark Fasheh status = -EINVAL; 877ccd979bdSMark Fasheh goto done; 878ccd979bdSMark Fasheh } 879ccd979bdSMark Fasheh 880ede7dc7fSHarshad Shirwadkar trace_ocfs2_journal_init_maxlen(j_journal->j_total_len); 881ccd979bdSMark Fasheh 882ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 883ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 884ccd979bdSMark Fasheh 885ccd979bdSMark Fasheh journal->j_journal = j_journal; 886342af94eSMauricio Faria de Oliveira journal->j_journal->j_submit_inode_data_buffers = 887342af94eSMauricio Faria de Oliveira jbd2_journal_submit_inode_data_buffers; 888342af94eSMauricio Faria de Oliveira journal->j_journal->j_finish_inode_data_buffers = 889342af94eSMauricio Faria de Oliveira jbd2_journal_finish_inode_data_buffers; 890ccd979bdSMark Fasheh journal->j_inode = inode; 891ccd979bdSMark Fasheh journal->j_bh = bh; 892ccd979bdSMark Fasheh 893ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 894ccd979bdSMark Fasheh 895ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 896ccd979bdSMark Fasheh 897ccd979bdSMark Fasheh status = 0; 898ccd979bdSMark Fasheh done: 899ccd979bdSMark Fasheh if (status < 0) { 900e63aecb6SMark Fasheh if (inode_lock) 901e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 902ccd979bdSMark Fasheh brelse(bh); 903ccd979bdSMark Fasheh if (inode) { 904ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 905ccd979bdSMark Fasheh iput(inode); 906ccd979bdSMark Fasheh } 907ccd979bdSMark Fasheh } 908ccd979bdSMark Fasheh 909ccd979bdSMark Fasheh return status; 910ccd979bdSMark Fasheh } 911ccd979bdSMark Fasheh 912539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 913539d8264SSunil Mushran { 914539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 915539d8264SSunil Mushran } 916539d8264SSunil Mushran 917539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 918539d8264SSunil Mushran { 919539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 920539d8264SSunil Mushran } 921539d8264SSunil Mushran 922ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 923539d8264SSunil Mushran int dirty, int replayed) 924ccd979bdSMark Fasheh { 925ccd979bdSMark Fasheh int status; 926ccd979bdSMark Fasheh unsigned int flags; 927ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 928ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 929ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 930ccd979bdSMark Fasheh 931ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 93210995aa2SJoel Becker 93310995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 93410995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 93510995aa2SJoel Becker * code bug if we mess it up. */ 93610995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 937ccd979bdSMark Fasheh 938ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 939ccd979bdSMark Fasheh if (dirty) 940ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 941ccd979bdSMark Fasheh else 942ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 943ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 944ccd979bdSMark Fasheh 945539d8264SSunil Mushran if (replayed) 946539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 947539d8264SSunil Mushran 94813723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 9498cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 950ccd979bdSMark Fasheh if (status < 0) 951ccd979bdSMark Fasheh mlog_errno(status); 952ccd979bdSMark Fasheh 953ccd979bdSMark Fasheh return status; 954ccd979bdSMark Fasheh } 955ccd979bdSMark Fasheh 956ccd979bdSMark Fasheh /* 957ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 958ccd979bdSMark Fasheh * call. 959ccd979bdSMark Fasheh */ 960ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 961ccd979bdSMark Fasheh { 962ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 963ccd979bdSMark Fasheh int status = 0; 964ccd979bdSMark Fasheh struct inode *inode = NULL; 965ccd979bdSMark Fasheh int num_running_trans = 0; 966ccd979bdSMark Fasheh 967ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 968ccd979bdSMark Fasheh 969ccd979bdSMark Fasheh journal = osb->journal; 970ccd979bdSMark Fasheh if (!journal) 971ccd979bdSMark Fasheh goto done; 972ccd979bdSMark Fasheh 973ccd979bdSMark Fasheh inode = journal->j_inode; 974ccd979bdSMark Fasheh 975ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 976ccd979bdSMark Fasheh goto done; 977ccd979bdSMark Fasheh 9782b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 979ccd979bdSMark Fasheh if (!igrab(inode)) 980ccd979bdSMark Fasheh BUG(); 981ccd979bdSMark Fasheh 982ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 983b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 984ccd979bdSMark Fasheh 985ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 986ccd979bdSMark Fasheh * release any locks that are still held. 987ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 988ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 989ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 990ccd979bdSMark Fasheh 991ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 992ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 993ccd979bdSMark Fasheh * completely destroy the journal. */ 994ccd979bdSMark Fasheh if (osb->commit_task) { 995ccd979bdSMark Fasheh /* Wait for the commit thread */ 996b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 997ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 998ccd979bdSMark Fasheh osb->commit_task = NULL; 999ccd979bdSMark Fasheh } 1000ccd979bdSMark Fasheh 1001ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 1002ccd979bdSMark Fasheh 1003c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 10042b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 10052b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 10062b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 1007c271c5c2SSunil Mushran if (status < 0) 1008c271c5c2SSunil Mushran mlog_errno(status); 1009c271c5c2SSunil Mushran } 1010c271c5c2SSunil Mushran 1011d85400afSJunxiao Bi /* Shutdown the kernel journal system */ 1012d85400afSJunxiao Bi if (!jbd2_journal_destroy(journal->j_journal) && !status) { 1013c271c5c2SSunil Mushran /* 1014c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 1015c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 1016c271c5c2SSunil Mushran */ 1017539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 1018ccd979bdSMark Fasheh if (status < 0) 1019ccd979bdSMark Fasheh mlog_errno(status); 1020c271c5c2SSunil Mushran } 1021ae0dff68SSunil Mushran journal->j_journal = NULL; 1022ccd979bdSMark Fasheh 1023ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 1024ccd979bdSMark Fasheh 1025ccd979bdSMark Fasheh /* unlock our journal */ 1026e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1027ccd979bdSMark Fasheh 1028ccd979bdSMark Fasheh brelse(journal->j_bh); 1029ccd979bdSMark Fasheh journal->j_bh = NULL; 1030ccd979bdSMark Fasheh 1031ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 1032ccd979bdSMark Fasheh 1033ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 1034ccd979bdSMark Fasheh done: 1035ccd979bdSMark Fasheh iput(inode); 1036ccd979bdSMark Fasheh } 1037ccd979bdSMark Fasheh 1038ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 1039ccd979bdSMark Fasheh journal_t *journal, 1040ccd979bdSMark Fasheh int slot) 1041ccd979bdSMark Fasheh { 1042ccd979bdSMark Fasheh int olderr; 1043ccd979bdSMark Fasheh 10442b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 1045ccd979bdSMark Fasheh if (olderr) { 1046ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 1047ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 1048ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 1049ccd979bdSMark Fasheh sb->s_id); 1050ccd979bdSMark Fasheh 10512b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 10522b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 1053ccd979bdSMark Fasheh } 1054ccd979bdSMark Fasheh } 1055ccd979bdSMark Fasheh 1056539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1057ccd979bdSMark Fasheh { 1058ccd979bdSMark Fasheh int status = 0; 1059ccd979bdSMark Fasheh struct ocfs2_super *osb; 1060ccd979bdSMark Fasheh 1061b1f3550fSJulia Lawall BUG_ON(!journal); 1062ccd979bdSMark Fasheh 1063ccd979bdSMark Fasheh osb = journal->j_osb; 1064ccd979bdSMark Fasheh 10652b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1066ccd979bdSMark Fasheh if (status < 0) { 1067ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1068ccd979bdSMark Fasheh goto done; 1069ccd979bdSMark Fasheh } 1070ccd979bdSMark Fasheh 1071ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1072ccd979bdSMark Fasheh 1073397eac17SKai Li if (replayed) { 1074397eac17SKai Li jbd2_journal_lock_updates(journal->j_journal); 1075397eac17SKai Li status = jbd2_journal_flush(journal->j_journal); 1076397eac17SKai Li jbd2_journal_unlock_updates(journal->j_journal); 1077397eac17SKai Li if (status < 0) 1078397eac17SKai Li mlog_errno(status); 1079397eac17SKai Li } 1080397eac17SKai Li 1081539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1082ccd979bdSMark Fasheh if (status < 0) { 1083ccd979bdSMark Fasheh mlog_errno(status); 1084ccd979bdSMark Fasheh goto done; 1085ccd979bdSMark Fasheh } 1086ccd979bdSMark Fasheh 1087ccd979bdSMark Fasheh /* Launch the commit thread */ 1088c271c5c2SSunil Mushran if (!local) { 1089c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 10905afc44e2SJoseph Qi "ocfs2cmt-%s", osb->uuid_str); 1091ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1092ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1093ccd979bdSMark Fasheh osb->commit_task = NULL; 1094c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1095c271c5c2SSunil Mushran "error=%d", status); 1096ccd979bdSMark Fasheh goto done; 1097ccd979bdSMark Fasheh } 1098c271c5c2SSunil Mushran } else 1099c271c5c2SSunil Mushran osb->commit_task = NULL; 1100ccd979bdSMark Fasheh 1101ccd979bdSMark Fasheh done: 1102ccd979bdSMark Fasheh return status; 1103ccd979bdSMark Fasheh } 1104ccd979bdSMark Fasheh 1105ccd979bdSMark Fasheh 1106ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1107ccd979bdSMark Fasheh * mark the journal clean */ 1108ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1109ccd979bdSMark Fasheh { 1110ccd979bdSMark Fasheh int status; 1111ccd979bdSMark Fasheh 1112ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1113ccd979bdSMark Fasheh 11142b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1115ccd979bdSMark Fasheh if (status < 0) { 1116ccd979bdSMark Fasheh mlog_errno(status); 1117ccd979bdSMark Fasheh goto bail; 1118ccd979bdSMark Fasheh } 1119ccd979bdSMark Fasheh 1120539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1121ccd979bdSMark Fasheh if (status < 0) 1122ccd979bdSMark Fasheh mlog_errno(status); 1123ccd979bdSMark Fasheh 1124ccd979bdSMark Fasheh bail: 1125ccd979bdSMark Fasheh return status; 1126ccd979bdSMark Fasheh } 1127ccd979bdSMark Fasheh 1128553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1129553abd04SJoel Becker { 1130553abd04SJoel Becker int empty; 1131553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1132553abd04SJoel Becker 1133553abd04SJoel Becker spin_lock(&osb->osb_lock); 1134553abd04SJoel Becker empty = (rm->rm_used == 0); 1135553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1136553abd04SJoel Becker 1137553abd04SJoel Becker return empty; 1138553abd04SJoel Becker } 1139553abd04SJoel Becker 1140553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1141553abd04SJoel Becker { 1142553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1143553abd04SJoel Becker } 1144553abd04SJoel Becker 1145ccd979bdSMark Fasheh /* 1146ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1147ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1148ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1149ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1150ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1151ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1152ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1153ccd979bdSMark Fasheh * as we have the lock by now :) 1154ccd979bdSMark Fasheh */ 1155ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1156ccd979bdSMark Fasheh { 1157ccd979bdSMark Fasheh int status = 0; 11584f902c37SMark Fasheh int i; 11598110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 11600b492f68SJunxiao Bi struct buffer_head *bh = NULL; 11610b492f68SJunxiao Bi struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1162ccd979bdSMark Fasheh 1163f17c20ddSJunxiao Bi num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1164ccd979bdSMark Fasheh v_blkno = 0; 11658110b073SMark Fasheh while (v_blkno < num_blocks) { 1166ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 116749cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1168ccd979bdSMark Fasheh if (status < 0) { 1169ccd979bdSMark Fasheh mlog_errno(status); 1170ccd979bdSMark Fasheh goto bail; 1171ccd979bdSMark Fasheh } 1172ccd979bdSMark Fasheh 11730b492f68SJunxiao Bi for (i = 0; i < p_blocks; i++, p_blkno++) { 11740b492f68SJunxiao Bi bh = __find_get_block(osb->sb->s_bdev, p_blkno, 11750b492f68SJunxiao Bi osb->sb->s_blocksize); 11760b492f68SJunxiao Bi /* block not cached. */ 11770b492f68SJunxiao Bi if (!bh) 11780b492f68SJunxiao Bi continue; 1179ccd979bdSMark Fasheh 11800b492f68SJunxiao Bi brelse(bh); 11810b492f68SJunxiao Bi bh = NULL; 1182dd4a2c2bSMark Fasheh /* We are reading journal data which should not 11830b492f68SJunxiao Bi * be put in the uptodate cache. 11840b492f68SJunxiao Bi */ 11850b492f68SJunxiao Bi status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); 1186ccd979bdSMark Fasheh if (status < 0) { 1187ccd979bdSMark Fasheh mlog_errno(status); 1188ccd979bdSMark Fasheh goto bail; 1189ccd979bdSMark Fasheh } 1190ccd979bdSMark Fasheh 11910b492f68SJunxiao Bi brelse(bh); 11920b492f68SJunxiao Bi bh = NULL; 1193ccd979bdSMark Fasheh } 1194ccd979bdSMark Fasheh 1195ccd979bdSMark Fasheh v_blkno += p_blocks; 1196ccd979bdSMark Fasheh } 1197ccd979bdSMark Fasheh 1198ccd979bdSMark Fasheh bail: 1199ccd979bdSMark Fasheh return status; 1200ccd979bdSMark Fasheh } 1201ccd979bdSMark Fasheh 1202ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1203ccd979bdSMark Fasheh struct list_head lri_list; 1204ccd979bdSMark Fasheh int lri_slot; 1205ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1206ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 12072205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1208ed460cffSJoseph Qi enum ocfs2_orphan_reco_type lri_orphan_reco_type; 1209ccd979bdSMark Fasheh }; 1210ccd979bdSMark Fasheh 1211ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1212ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1213ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1214ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1215ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1216ccd979bdSMark Fasheh * recovering nodes into account. 1217ccd979bdSMark Fasheh * 1218ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1219ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1220ccd979bdSMark Fasheh */ 1221c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1222ccd979bdSMark Fasheh { 1223b4107950STao Ma int ret = 0; 1224c4028958SDavid Howells struct ocfs2_journal *journal = 1225c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1226c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1227ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1228800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 12292205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1230ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 1231ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1232ccd979bdSMark Fasheh 1233b4107950STao Ma trace_ocfs2_complete_recovery( 1234b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1235ccd979bdSMark Fasheh 1236ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1237ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1238ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1239ccd979bdSMark Fasheh 1240800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1241ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1242ccd979bdSMark Fasheh 124319ece546SJan Kara ocfs2_wait_on_quotas(osb); 124419ece546SJan Kara 1245ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1246b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1247b4107950STao Ma qrec = item->lri_qrec; 1248ed460cffSJoseph Qi orphan_reco_type = item->lri_orphan_reco_type; 1249ccd979bdSMark Fasheh 1250b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1251b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1252b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1253b4107950STao Ma qrec); 1254b4107950STao Ma 1255b4107950STao Ma if (la_dinode) { 1256ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1257ccd979bdSMark Fasheh la_dinode); 1258ccd979bdSMark Fasheh if (ret < 0) 1259ccd979bdSMark Fasheh mlog_errno(ret); 1260ccd979bdSMark Fasheh 1261ccd979bdSMark Fasheh kfree(la_dinode); 1262ccd979bdSMark Fasheh } 1263ccd979bdSMark Fasheh 1264ccd979bdSMark Fasheh if (tl_dinode) { 1265ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1266ccd979bdSMark Fasheh tl_dinode); 1267ccd979bdSMark Fasheh if (ret < 0) 1268ccd979bdSMark Fasheh mlog_errno(ret); 1269ccd979bdSMark Fasheh 1270ccd979bdSMark Fasheh kfree(tl_dinode); 1271ccd979bdSMark Fasheh } 1272ccd979bdSMark Fasheh 1273ed460cffSJoseph Qi ret = ocfs2_recover_orphans(osb, item->lri_slot, 1274ed460cffSJoseph Qi orphan_reco_type); 1275ccd979bdSMark Fasheh if (ret < 0) 1276ccd979bdSMark Fasheh mlog_errno(ret); 1277ccd979bdSMark Fasheh 12782205363dSJan Kara if (qrec) { 12792205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 12802205363dSJan Kara item->lri_slot); 12812205363dSJan Kara if (ret < 0) 12822205363dSJan Kara mlog_errno(ret); 12832205363dSJan Kara /* Recovery info is already freed now */ 12842205363dSJan Kara } 12852205363dSJan Kara 1286ccd979bdSMark Fasheh kfree(item); 1287ccd979bdSMark Fasheh } 1288ccd979bdSMark Fasheh 1289b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1290ccd979bdSMark Fasheh } 1291ccd979bdSMark Fasheh 1292ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1293ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1294ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1295ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1296ccd979bdSMark Fasheh int slot_num, 1297ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 12982205363dSJan Kara struct ocfs2_dinode *tl_dinode, 1299ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 1300ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1301ccd979bdSMark Fasheh { 1302ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1303ccd979bdSMark Fasheh 1304afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1305ccd979bdSMark Fasheh if (!item) { 1306ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1307ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1308ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1309ccd979bdSMark Fasheh kfree(la_dinode); 1310ccd979bdSMark Fasheh kfree(tl_dinode); 1311ccd979bdSMark Fasheh 13122205363dSJan Kara if (qrec) 13132205363dSJan Kara ocfs2_free_quota_recovery(qrec); 13142205363dSJan Kara 1315ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1316ccd979bdSMark Fasheh return; 1317ccd979bdSMark Fasheh } 1318ccd979bdSMark Fasheh 1319ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1320ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1321ccd979bdSMark Fasheh item->lri_slot = slot_num; 1322ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 13232205363dSJan Kara item->lri_qrec = qrec; 1324ed460cffSJoseph Qi item->lri_orphan_reco_type = orphan_reco_type; 1325ccd979bdSMark Fasheh 1326ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1327ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 132835ddf78eSjiangyiwen queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); 1329ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1330ccd979bdSMark Fasheh } 1331ccd979bdSMark Fasheh 1332ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 13339140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1334ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1335ccd979bdSMark Fasheh { 1336ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1337ccd979bdSMark Fasheh 133810b3dd76SSunil Mushran if (ocfs2_is_hard_readonly(osb)) 133910b3dd76SSunil Mushran return; 134010b3dd76SSunil Mushran 13419140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 13429140db04SSrinivas Eeda * that */ 13439140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 1344ed460cffSJoseph Qi osb->local_alloc_copy, NULL, NULL, 1345ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 1346ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1347ccd979bdSMark Fasheh 1348ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 13499140db04SSrinivas Eeda 13509140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 13519140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1352ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 13539140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1354ccd979bdSMark Fasheh } 1355ccd979bdSMark Fasheh 13562205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 13572205363dSJan Kara { 13582205363dSJan Kara if (osb->quota_rec) { 13592205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 13602205363dSJan Kara osb->slot_num, 13612205363dSJan Kara NULL, 13622205363dSJan Kara NULL, 1363ed460cffSJoseph Qi osb->quota_rec, 1364ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 13652205363dSJan Kara osb->quota_rec = NULL; 13662205363dSJan Kara } 13672205363dSJan Kara } 13682205363dSJan Kara 1369ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1370ccd979bdSMark Fasheh { 13712205363dSJan Kara int status, node_num, slot_num; 1372ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1373553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 13742205363dSJan Kara int *rm_quota = NULL; 13752205363dSJan Kara int rm_quota_used = 0, i; 13762205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1377ccd979bdSMark Fasheh 137821158ca8SGuozhonghua /* Whether the quota supported. */ 137921158ca8SGuozhonghua int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 138021158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_USRQUOTA) 138121158ca8SGuozhonghua || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 138221158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); 138321158ca8SGuozhonghua 1384ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1385ccd979bdSMark Fasheh if (status < 0) { 1386ccd979bdSMark Fasheh goto bail; 1387ccd979bdSMark Fasheh } 1388ccd979bdSMark Fasheh 138921158ca8SGuozhonghua if (quota_enabled) { 13906396bb22SKees Cook rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); 13912205363dSJan Kara if (!rm_quota) { 13922205363dSJan Kara status = -ENOMEM; 13932205363dSJan Kara goto bail; 13942205363dSJan Kara } 139521158ca8SGuozhonghua } 1396ccd979bdSMark Fasheh restart: 1397ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1398ccd979bdSMark Fasheh if (status < 0) { 1399ccd979bdSMark Fasheh mlog_errno(status); 1400ccd979bdSMark Fasheh goto bail; 1401ccd979bdSMark Fasheh } 1402ccd979bdSMark Fasheh 14039140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 14049140db04SSrinivas Eeda if (status < 0) 14059140db04SSrinivas Eeda mlog_errno(status); 14069140db04SSrinivas Eeda 14079140db04SSrinivas Eeda /* queue recovery for our own slot */ 14089140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1409ed460cffSJoseph Qi NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); 14109140db04SSrinivas Eeda 1411553abd04SJoel Becker spin_lock(&osb->osb_lock); 1412553abd04SJoel Becker while (rm->rm_used) { 1413553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1414553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1415553abd04SJoel Becker node_num = rm->rm_entries[0]; 1416553abd04SJoel Becker spin_unlock(&osb->osb_lock); 14172205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1418b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 14192205363dSJan Kara if (slot_num == -ENOENT) { 14202205363dSJan Kara status = 0; 14212205363dSJan Kara goto skip_recovery; 14222205363dSJan Kara } 1423ccd979bdSMark Fasheh 14242205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 14252205363dSJan Kara * immediately because we have to obtain cluster locks from 14262205363dSJan Kara * quota files and we also don't want to just skip it because 14272205363dSJan Kara * then quota usage would be out of sync until some node takes 14282205363dSJan Kara * the slot. So we remember which nodes need quota recovery 14292205363dSJan Kara * and when everything else is done, we recover quotas. */ 143021158ca8SGuozhonghua if (quota_enabled) { 143121158ca8SGuozhonghua for (i = 0; i < rm_quota_used 143221158ca8SGuozhonghua && rm_quota[i] != slot_num; i++) 143321158ca8SGuozhonghua ; 143421158ca8SGuozhonghua 14352205363dSJan Kara if (i == rm_quota_used) 14362205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 143721158ca8SGuozhonghua } 14382205363dSJan Kara 14392205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 14402205363dSJan Kara skip_recovery: 1441553abd04SJoel Becker if (!status) { 1442553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1443553abd04SJoel Becker } else { 1444ccd979bdSMark Fasheh mlog(ML_ERROR, 1445ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1446ccd979bdSMark Fasheh status, node_num, 1447ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1448ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1449ccd979bdSMark Fasheh } 1450ccd979bdSMark Fasheh 1451553abd04SJoel Becker spin_lock(&osb->osb_lock); 1452ccd979bdSMark Fasheh } 1453553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1454b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1455553abd04SJoel Becker 1456539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1457539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1458539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1459539d8264SSunil Mushran if (status < 0) 1460539d8264SSunil Mushran mlog_errno(status); 1461539d8264SSunil Mushran 14622205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 14632205363dSJan Kara * superblock lock so that no one can start using the slot (and crash) 14642205363dSJan Kara * before we recover it */ 146521158ca8SGuozhonghua if (quota_enabled) { 14662205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 14672205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 14682205363dSJan Kara if (IS_ERR(qrec)) { 14692205363dSJan Kara status = PTR_ERR(qrec); 14702205363dSJan Kara mlog_errno(status); 14712205363dSJan Kara continue; 14722205363dSJan Kara } 147321158ca8SGuozhonghua ocfs2_queue_recovery_completion(osb->journal, 147421158ca8SGuozhonghua rm_quota[i], 1475ed460cffSJoseph Qi NULL, NULL, qrec, 1476ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 14772205363dSJan Kara } 147821158ca8SGuozhonghua } 14792205363dSJan Kara 1480ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1481ccd979bdSMark Fasheh 14829140db04SSrinivas Eeda /* queue recovery for offline slots */ 1483ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1484ccd979bdSMark Fasheh 1485ccd979bdSMark Fasheh bail: 1486c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1487553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1488c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1489ccd979bdSMark Fasheh goto restart; 1490ccd979bdSMark Fasheh } 1491ccd979bdSMark Fasheh 14929140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1493ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1494ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1495ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1496ccd979bdSMark Fasheh 1497c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1498ccd979bdSMark Fasheh 149921158ca8SGuozhonghua if (quota_enabled) 15002205363dSJan Kara kfree(rm_quota); 15012205363dSJan Kara 1502ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1503ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1504ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1505ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1506ccd979bdSMark Fasheh } 1507ccd979bdSMark Fasheh 1508ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1509ccd979bdSMark Fasheh { 1510c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1511b4107950STao Ma 1512b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1513b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1514b4107950STao Ma osb->disable_recovery ? 1515b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1516b4107950STao Ma 1517ccd979bdSMark Fasheh if (osb->disable_recovery) 1518ccd979bdSMark Fasheh goto out; 1519ccd979bdSMark Fasheh 1520ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1521ccd979bdSMark Fasheh goto out; 1522ccd979bdSMark Fasheh 1523ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 15245afc44e2SJoseph Qi "ocfs2rec-%s", osb->uuid_str); 1525ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1526ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1527ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1528ccd979bdSMark Fasheh } 1529ccd979bdSMark Fasheh 1530ccd979bdSMark Fasheh out: 1531c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1532ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1533ccd979bdSMark Fasheh } 1534ccd979bdSMark Fasheh 1535539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1536539d8264SSunil Mushran int slot_num, 1537539d8264SSunil Mushran struct buffer_head **bh, 1538539d8264SSunil Mushran struct inode **ret_inode) 1539539d8264SSunil Mushran { 1540539d8264SSunil Mushran int status = -EACCES; 1541539d8264SSunil Mushran struct inode *inode = NULL; 1542539d8264SSunil Mushran 1543539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1544539d8264SSunil Mushran 1545539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1546539d8264SSunil Mushran slot_num); 1547539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1548539d8264SSunil Mushran mlog_errno(status); 1549539d8264SSunil Mushran goto bail; 1550539d8264SSunil Mushran } 1551539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1552539d8264SSunil Mushran 1553b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1554539d8264SSunil Mushran if (status < 0) { 1555539d8264SSunil Mushran mlog_errno(status); 1556539d8264SSunil Mushran goto bail; 1557539d8264SSunil Mushran } 1558539d8264SSunil Mushran 1559539d8264SSunil Mushran status = 0; 1560539d8264SSunil Mushran 1561539d8264SSunil Mushran bail: 1562539d8264SSunil Mushran if (inode) { 1563539d8264SSunil Mushran if (status || !ret_inode) 1564539d8264SSunil Mushran iput(inode); 1565539d8264SSunil Mushran else 1566539d8264SSunil Mushran *ret_inode = inode; 1567539d8264SSunil Mushran } 1568539d8264SSunil Mushran return status; 1569539d8264SSunil Mushran } 1570539d8264SSunil Mushran 1571ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1572ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1573ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1574ccd979bdSMark Fasheh int node_num, 1575ccd979bdSMark Fasheh int slot_num) 1576ccd979bdSMark Fasheh { 1577ccd979bdSMark Fasheh int status; 1578ccd979bdSMark Fasheh int got_lock = 0; 1579ccd979bdSMark Fasheh unsigned int flags; 1580ccd979bdSMark Fasheh struct inode *inode = NULL; 1581ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1582ccd979bdSMark Fasheh journal_t *journal = NULL; 1583ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1584539d8264SSunil Mushran u32 slot_reco_gen; 1585ccd979bdSMark Fasheh 1586539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1587539d8264SSunil Mushran if (status) { 1588ccd979bdSMark Fasheh mlog_errno(status); 1589ccd979bdSMark Fasheh goto done; 1590ccd979bdSMark Fasheh } 1591539d8264SSunil Mushran 1592539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1593539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1594539d8264SSunil Mushran brelse(bh); 1595539d8264SSunil Mushran bh = NULL; 1596539d8264SSunil Mushran 1597539d8264SSunil Mushran /* 1598539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1599539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1600539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1601539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1602539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1603539d8264SSunil Mushran * If not, it needs recovery. 1604539d8264SSunil Mushran */ 1605539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1606b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1607539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1608539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1609539d8264SSunil Mushran status = -EBUSY; 1610ccd979bdSMark Fasheh goto done; 1611ccd979bdSMark Fasheh } 1612539d8264SSunil Mushran 1613539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1614ccd979bdSMark Fasheh 1615e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1616ccd979bdSMark Fasheh if (status < 0) { 1617b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1618ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1619ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1620ccd979bdSMark Fasheh goto done; 1621ccd979bdSMark Fasheh } 1622ccd979bdSMark Fasheh got_lock = 1; 1623ccd979bdSMark Fasheh 1624ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1625ccd979bdSMark Fasheh 1626ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1627539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1628ccd979bdSMark Fasheh 1629ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1630b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1631539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1632539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1633ccd979bdSMark Fasheh goto done; 1634ccd979bdSMark Fasheh } 1635ccd979bdSMark Fasheh 16369140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 16379140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 16389140db04SSrinivas Eeda 1639619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1640619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1641619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1642ccd979bdSMark Fasheh 1643ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1644ccd979bdSMark Fasheh 1645ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1646ccd979bdSMark Fasheh if (status < 0) { 1647ccd979bdSMark Fasheh mlog_errno(status); 1648ccd979bdSMark Fasheh goto done; 1649ccd979bdSMark Fasheh } 1650ccd979bdSMark Fasheh 16512b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1652ccd979bdSMark Fasheh if (journal == NULL) { 1653ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1654ccd979bdSMark Fasheh status = -EIO; 1655ccd979bdSMark Fasheh goto done; 1656ccd979bdSMark Fasheh } 1657ccd979bdSMark Fasheh 16582b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1659ccd979bdSMark Fasheh if (status < 0) { 1660ccd979bdSMark Fasheh mlog_errno(status); 1661ccd979bdSMark Fasheh if (!igrab(inode)) 1662ccd979bdSMark Fasheh BUG(); 16632b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1664ccd979bdSMark Fasheh goto done; 1665ccd979bdSMark Fasheh } 1666ccd979bdSMark Fasheh 1667ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1668ccd979bdSMark Fasheh 1669ccd979bdSMark Fasheh /* wipe the journal */ 16702b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 16712b4e30fbSJoel Becker status = jbd2_journal_flush(journal); 16722b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1673ccd979bdSMark Fasheh if (status < 0) 1674ccd979bdSMark Fasheh mlog_errno(status); 1675ccd979bdSMark Fasheh 1676ccd979bdSMark Fasheh /* This will mark the node clean */ 1677ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1678ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1679ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1680ccd979bdSMark Fasheh 1681539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1682539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1683539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1684539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1685539d8264SSunil Mushran 168613723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 16878cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1688ccd979bdSMark Fasheh if (status < 0) 1689ccd979bdSMark Fasheh mlog_errno(status); 1690ccd979bdSMark Fasheh 1691ccd979bdSMark Fasheh if (!igrab(inode)) 1692ccd979bdSMark Fasheh BUG(); 1693ccd979bdSMark Fasheh 16942b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1695ccd979bdSMark Fasheh 1696619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1697619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1698619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1699ccd979bdSMark Fasheh done: 1700ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1701ccd979bdSMark Fasheh if (got_lock) 1702e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1703ccd979bdSMark Fasheh 1704ccd979bdSMark Fasheh iput(inode); 1705ccd979bdSMark Fasheh brelse(bh); 1706ccd979bdSMark Fasheh 1707ccd979bdSMark Fasheh return status; 1708ccd979bdSMark Fasheh } 1709ccd979bdSMark Fasheh 1710ccd979bdSMark Fasheh /* 1711ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1712ccd979bdSMark Fasheh * - Replay it's journal 1713ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1714ccd979bdSMark Fasheh * - Stamp a clean truncate log 1715ccd979bdSMark Fasheh * - Mark the node clean 1716ccd979bdSMark Fasheh * 1717ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1718ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1719ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1720ccd979bdSMark Fasheh * far less concerning. 1721ccd979bdSMark Fasheh */ 1722ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 17232205363dSJan Kara int node_num, int slot_num) 1724ccd979bdSMark Fasheh { 1725ccd979bdSMark Fasheh int status = 0; 1726ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1727ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1728ccd979bdSMark Fasheh 1729b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1730ccd979bdSMark Fasheh 1731ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1732ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1733ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1734ccd979bdSMark Fasheh 1735ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1736ccd979bdSMark Fasheh if (status < 0) { 1737539d8264SSunil Mushran if (status == -EBUSY) { 1738b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1739539d8264SSunil Mushran status = 0; 1740539d8264SSunil Mushran goto done; 1741539d8264SSunil Mushran } 1742ccd979bdSMark Fasheh mlog_errno(status); 1743ccd979bdSMark Fasheh goto done; 1744ccd979bdSMark Fasheh } 1745ccd979bdSMark Fasheh 1746ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1747ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1748ccd979bdSMark Fasheh if (status < 0) { 1749ccd979bdSMark Fasheh mlog_errno(status); 1750ccd979bdSMark Fasheh goto done; 1751ccd979bdSMark Fasheh } 1752ccd979bdSMark Fasheh 1753ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1754ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1755ccd979bdSMark Fasheh * recovery. */ 1756ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1757ccd979bdSMark Fasheh if (status < 0) 1758ccd979bdSMark Fasheh mlog_errno(status); 1759ccd979bdSMark Fasheh 1760ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1761ccd979bdSMark Fasheh * harmful place to get an error... */ 17628e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1763ccd979bdSMark Fasheh if (status < 0) 1764ccd979bdSMark Fasheh mlog_errno(status); 1765ccd979bdSMark Fasheh 1766ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1767ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1768ed460cffSJoseph Qi tl_copy, NULL, ORPHAN_NEED_TRUNCATE); 1769ccd979bdSMark Fasheh 1770ccd979bdSMark Fasheh status = 0; 1771ccd979bdSMark Fasheh done: 1772ccd979bdSMark Fasheh 1773ccd979bdSMark Fasheh return status; 1774ccd979bdSMark Fasheh } 1775ccd979bdSMark Fasheh 1776ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1777ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1778ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1779ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1780ccd979bdSMark Fasheh int slot_num) 1781ccd979bdSMark Fasheh { 1782ccd979bdSMark Fasheh int status, flags; 1783ccd979bdSMark Fasheh struct inode *inode = NULL; 1784ccd979bdSMark Fasheh 1785ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1786ccd979bdSMark Fasheh slot_num); 1787ccd979bdSMark Fasheh if (inode == NULL) { 1788ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1789ccd979bdSMark Fasheh status = -EACCES; 1790ccd979bdSMark Fasheh goto bail; 1791ccd979bdSMark Fasheh } 1792ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1793ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1794ccd979bdSMark Fasheh iput(inode); 1795ccd979bdSMark Fasheh inode = NULL; 1796ccd979bdSMark Fasheh status = -EACCES; 1797ccd979bdSMark Fasheh goto bail; 1798ccd979bdSMark Fasheh } 1799ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1800ccd979bdSMark Fasheh 1801ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1802e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1803ccd979bdSMark Fasheh if (status < 0) { 1804ccd979bdSMark Fasheh if (status != -EAGAIN) 1805ccd979bdSMark Fasheh mlog_errno(status); 1806ccd979bdSMark Fasheh goto bail; 1807ccd979bdSMark Fasheh } 1808ccd979bdSMark Fasheh 1809e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1810ccd979bdSMark Fasheh bail: 1811ccd979bdSMark Fasheh iput(inode); 1812ccd979bdSMark Fasheh 1813ccd979bdSMark Fasheh return status; 1814ccd979bdSMark Fasheh } 1815ccd979bdSMark Fasheh 1816ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1817ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1818ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1819ccd979bdSMark Fasheh { 1820d85b20e4SJoel Becker unsigned int node_num; 1821d85b20e4SJoel Becker int status, i; 1822a1af7d15SMark Fasheh u32 gen; 1823539d8264SSunil Mushran struct buffer_head *bh = NULL; 1824539d8264SSunil Mushran struct ocfs2_dinode *di; 1825ccd979bdSMark Fasheh 1826ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1827ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1828ccd979bdSMark Fasheh 1829d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1830539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1831539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1832539d8264SSunil Mushran if (status) { 1833539d8264SSunil Mushran mlog_errno(status); 1834539d8264SSunil Mushran goto bail; 1835539d8264SSunil Mushran } 1836539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1837a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1838539d8264SSunil Mushran brelse(bh); 1839539d8264SSunil Mushran bh = NULL; 1840539d8264SSunil Mushran 1841a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1842a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1843a1af7d15SMark Fasheh 1844b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1845539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1846539d8264SSunil Mushran 1847a1af7d15SMark Fasheh if (i == osb->slot_num) { 1848a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1849ccd979bdSMark Fasheh continue; 1850a1af7d15SMark Fasheh } 1851d85b20e4SJoel Becker 1852d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1853a1af7d15SMark Fasheh if (status == -ENOENT) { 1854a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1855ccd979bdSMark Fasheh continue; 1856a1af7d15SMark Fasheh } 1857ccd979bdSMark Fasheh 1858a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1859a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1860ccd979bdSMark Fasheh continue; 1861a1af7d15SMark Fasheh } 1862d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1863ccd979bdSMark Fasheh 1864ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1865ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1866ccd979bdSMark Fasheh * file here to test if he's alive. */ 1867ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1868ccd979bdSMark Fasheh if (!status) { 1869ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1870ccd979bdSMark Fasheh * the recovery thread can't race us on 1871ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1872ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1873ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1874ccd979bdSMark Fasheh mlog_errno(status); 1875ccd979bdSMark Fasheh goto bail; 1876ccd979bdSMark Fasheh } 1877ccd979bdSMark Fasheh } 1878ccd979bdSMark Fasheh 1879ccd979bdSMark Fasheh status = 0; 1880ccd979bdSMark Fasheh bail: 1881ccd979bdSMark Fasheh return status; 1882ccd979bdSMark Fasheh } 1883ccd979bdSMark Fasheh 188483273932SSrinivas Eeda /* 188583273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 188683273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 188783273932SSrinivas Eeda * same time. 188883273932SSrinivas Eeda */ 188983273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 189083273932SSrinivas Eeda { 189183273932SSrinivas Eeda unsigned long time; 189283273932SSrinivas Eeda 189383273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 189483273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 189583273932SSrinivas Eeda return msecs_to_jiffies(time); 189683273932SSrinivas Eeda } 189783273932SSrinivas Eeda 189883273932SSrinivas Eeda /* 189983273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 190083273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 190183273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 190283273932SSrinivas Eeda * 1903a035bff6SSunil Mushran * It scans all slots, even ones that are in use. It does so to handle the 1904a035bff6SSunil Mushran * case described below: 1905a035bff6SSunil Mushran * 1906a035bff6SSunil Mushran * Node 1 has an inode it was using. The dentry went away due to memory 1907a035bff6SSunil Mushran * pressure. Node 1 closes the inode, but it's on the free list. The node 1908a035bff6SSunil Mushran * has the open lock. 1909a035bff6SSunil Mushran * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1910a035bff6SSunil Mushran * but node 1 has no dentry and doesn't get the message. It trylocks the 1911a035bff6SSunil Mushran * open lock, sees that another node has a PR, and does nothing. 1912a035bff6SSunil Mushran * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1913a035bff6SSunil Mushran * open lock, sees the PR still, and does nothing. 1914a035bff6SSunil Mushran * Basically, we have to trigger an orphan iput on node 1. The only way 1915a035bff6SSunil Mushran * for this to happen is if node 1 runs node 2's orphan dir. 1916a035bff6SSunil Mushran * 191783273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 191883273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 191983273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 192083273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 192183273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 192283273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 192383273932SSrinivas Eeda * sequence number in the LVB. 192483273932SSrinivas Eeda */ 1925b519ea6dSJoseph Qi static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 192683273932SSrinivas Eeda { 192783273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 192883273932SSrinivas Eeda int status, i; 192983273932SSrinivas Eeda u32 seqno = 0; 193083273932SSrinivas Eeda 193183273932SSrinivas Eeda os = &osb->osb_orphan_scan; 193283273932SSrinivas Eeda 1933692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1934692684e1SSunil Mushran goto out; 1935692684e1SSunil Mushran 1936b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1937b4107950STao Ma atomic_read(&os->os_state)); 1938b4107950STao Ma 1939df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 194083273932SSrinivas Eeda if (status < 0) { 194183273932SSrinivas Eeda if (status != -EAGAIN) 194283273932SSrinivas Eeda mlog_errno(status); 194383273932SSrinivas Eeda goto out; 194483273932SSrinivas Eeda } 194583273932SSrinivas Eeda 1946692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1947692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1948692684e1SSunil Mushran goto unlock; 1949692684e1SSunil Mushran 195083273932SSrinivas Eeda if (os->os_seqno != seqno) { 195183273932SSrinivas Eeda os->os_seqno = seqno; 195283273932SSrinivas Eeda goto unlock; 195383273932SSrinivas Eeda } 195483273932SSrinivas Eeda 195583273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 195683273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1957ed460cffSJoseph Qi NULL, ORPHAN_NO_NEED_TRUNCATE); 195883273932SSrinivas Eeda /* 195983273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 196083273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 196183273932SSrinivas Eeda */ 196283273932SSrinivas Eeda seqno++; 196315633a22SSrinivas Eeda os->os_count++; 1964395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 196583273932SSrinivas Eeda unlock: 1966df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 196783273932SSrinivas Eeda out: 1968b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1969b4107950STao Ma atomic_read(&os->os_state)); 197083273932SSrinivas Eeda return; 197183273932SSrinivas Eeda } 197283273932SSrinivas Eeda 197383273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 1974b519ea6dSJoseph Qi static void ocfs2_orphan_scan_work(struct work_struct *work) 197583273932SSrinivas Eeda { 197683273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 197783273932SSrinivas Eeda struct ocfs2_super *osb; 197883273932SSrinivas Eeda 197983273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 198083273932SSrinivas Eeda os_orphan_scan_work.work); 198183273932SSrinivas Eeda osb = os->os_osb; 198283273932SSrinivas Eeda 198383273932SSrinivas Eeda mutex_lock(&os->os_lock); 198483273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 1985692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 198635ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 198783273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 198883273932SSrinivas Eeda mutex_unlock(&os->os_lock); 198983273932SSrinivas Eeda } 199083273932SSrinivas Eeda 199183273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 199283273932SSrinivas Eeda { 199383273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 199483273932SSrinivas Eeda 199583273932SSrinivas Eeda os = &osb->osb_orphan_scan; 1996df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1997692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 199883273932SSrinivas Eeda mutex_lock(&os->os_lock); 199983273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 200083273932SSrinivas Eeda mutex_unlock(&os->os_lock); 200183273932SSrinivas Eeda } 2002df152c24SSunil Mushran } 200383273932SSrinivas Eeda 2004df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 200583273932SSrinivas Eeda { 200683273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 200783273932SSrinivas Eeda 200883273932SSrinivas Eeda os = &osb->osb_orphan_scan; 200983273932SSrinivas Eeda os->os_osb = osb; 201015633a22SSrinivas Eeda os->os_count = 0; 20113211949fSSunil Mushran os->os_seqno = 0; 201283273932SSrinivas Eeda mutex_init(&os->os_lock); 2013df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 20148b712cd5SJeff Mahoney } 201583273932SSrinivas Eeda 20168b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 20178b712cd5SJeff Mahoney { 20188b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 20198b712cd5SJeff Mahoney 20208b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 2021395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 2022df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 2023df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 2024df152c24SSunil Mushran else { 2025df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 202635ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 202783273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 2028df152c24SSunil Mushran } 202983273932SSrinivas Eeda } 203083273932SSrinivas Eeda 20315eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 20323704412bSAl Viro struct dir_context ctx; 20335eae5b96SMark Fasheh struct inode *head; 20345eae5b96SMark Fasheh struct ocfs2_super *osb; 203530edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 20365eae5b96SMark Fasheh }; 20375eae5b96SMark Fasheh 2038ac7576f4SMiklos Szeredi static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, 2039ac7576f4SMiklos Szeredi int name_len, loff_t pos, u64 ino, 2040ac7576f4SMiklos Szeredi unsigned type) 20415eae5b96SMark Fasheh { 2042ac7576f4SMiklos Szeredi struct ocfs2_orphan_filldir_priv *p = 2043ac7576f4SMiklos Szeredi container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); 20445eae5b96SMark Fasheh struct inode *iter; 20455eae5b96SMark Fasheh 20465eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 20475eae5b96SMark Fasheh return 0; 20485eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 20495eae5b96SMark Fasheh return 0; 20505eae5b96SMark Fasheh 205130edc43cSJoseph Qi /* do not include dio entry in case of orphan scan */ 205230edc43cSJoseph Qi if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && 205330edc43cSJoseph Qi (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 205430edc43cSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN))) 205530edc43cSJoseph Qi return 0; 205630edc43cSJoseph Qi 20575eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 20585eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 20595fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 20605eae5b96SMark Fasheh if (IS_ERR(iter)) 20615eae5b96SMark Fasheh return 0; 20625eae5b96SMark Fasheh 206393d911fcSJoseph Qi if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 206493d911fcSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN)) 206593d911fcSJoseph Qi OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; 206693d911fcSJoseph Qi 2067ed460cffSJoseph Qi /* Skip inodes which are already added to recover list, since dio may 2068ed460cffSJoseph Qi * happen concurrently with unlink/rename */ 2069ed460cffSJoseph Qi if (OCFS2_I(iter)->ip_next_orphan) { 2070ed460cffSJoseph Qi iput(iter); 2071ed460cffSJoseph Qi return 0; 2072ed460cffSJoseph Qi } 2073ed460cffSJoseph Qi 2074b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 20755eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 20765eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 20775eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 20785eae5b96SMark Fasheh p->head = iter; 20795eae5b96SMark Fasheh 20805eae5b96SMark Fasheh return 0; 20815eae5b96SMark Fasheh } 20825eae5b96SMark Fasheh 2083b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2084b4df6ed8SMark Fasheh int slot, 208530edc43cSJoseph Qi struct inode **head, 208630edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2087ccd979bdSMark Fasheh { 2088b4df6ed8SMark Fasheh int status; 2089ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 20903704412bSAl Viro struct ocfs2_orphan_filldir_priv priv = { 20913704412bSAl Viro .ctx.actor = ocfs2_orphan_filldir, 20923704412bSAl Viro .osb = osb, 209330edc43cSJoseph Qi .head = *head, 209430edc43cSJoseph Qi .orphan_reco_type = orphan_reco_type 20953704412bSAl Viro }; 2096ccd979bdSMark Fasheh 2097ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2098ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 2099ccd979bdSMark Fasheh slot); 2100ccd979bdSMark Fasheh if (!orphan_dir_inode) { 2101ccd979bdSMark Fasheh status = -ENOENT; 2102ccd979bdSMark Fasheh mlog_errno(status); 2103b4df6ed8SMark Fasheh return status; 2104ccd979bdSMark Fasheh } 2105ccd979bdSMark Fasheh 21065955102cSAl Viro inode_lock(orphan_dir_inode); 2107e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2108ccd979bdSMark Fasheh if (status < 0) { 2109ccd979bdSMark Fasheh mlog_errno(status); 2110ccd979bdSMark Fasheh goto out; 2111ccd979bdSMark Fasheh } 2112ccd979bdSMark Fasheh 21133704412bSAl Viro status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 21145eae5b96SMark Fasheh if (status) { 2115ccd979bdSMark Fasheh mlog_errno(status); 2116a86370fbSMark Fasheh goto out_cluster; 2117ccd979bdSMark Fasheh } 2118ccd979bdSMark Fasheh 21195eae5b96SMark Fasheh *head = priv.head; 2120ccd979bdSMark Fasheh 2121a86370fbSMark Fasheh out_cluster: 2122e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 2123b4df6ed8SMark Fasheh out: 21245955102cSAl Viro inode_unlock(orphan_dir_inode); 2125ccd979bdSMark Fasheh iput(orphan_dir_inode); 2126b4df6ed8SMark Fasheh return status; 2127b4df6ed8SMark Fasheh } 2128b4df6ed8SMark Fasheh 2129b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2130b4df6ed8SMark Fasheh int slot) 2131b4df6ed8SMark Fasheh { 2132b4df6ed8SMark Fasheh int ret; 2133b4df6ed8SMark Fasheh 2134b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2135b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2136b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2137b4df6ed8SMark Fasheh return ret; 2138b4df6ed8SMark Fasheh } 2139b4df6ed8SMark Fasheh 2140b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2141b4df6ed8SMark Fasheh int slot) 2142b4df6ed8SMark Fasheh { 2143b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2144b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2145b4df6ed8SMark Fasheh * know to quit early. */ 2146b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2147b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2148b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2149b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2150b4df6ed8SMark Fasheh * them. */ 2151b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2152b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2153b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2154b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2155b4df6ed8SMark Fasheh } 2156b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2157b4df6ed8SMark Fasheh } 2158b4df6ed8SMark Fasheh 2159b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2160b4df6ed8SMark Fasheh int slot) 2161b4df6ed8SMark Fasheh { 2162b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2163b4df6ed8SMark Fasheh } 2164b4df6ed8SMark Fasheh 2165b4df6ed8SMark Fasheh /* 2166b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2167b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2168b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2169b4df6ed8SMark Fasheh * (most) of the rest of the work. 2170b4df6ed8SMark Fasheh * 2171b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2172b4df6ed8SMark Fasheh * couple of extra considerations. 2173b4df6ed8SMark Fasheh * 2174b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2175b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2176b4df6ed8SMark Fasheh * an invalid inode. 2177b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2178b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2179b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2180b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2181b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2182b4df6ed8SMark Fasheh */ 2183b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2184ed460cffSJoseph Qi int slot, 2185ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2186b4df6ed8SMark Fasheh { 2187b4df6ed8SMark Fasheh int ret = 0; 2188b4df6ed8SMark Fasheh struct inode *inode = NULL; 2189b4df6ed8SMark Fasheh struct inode *iter; 2190b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2191cf1776a9SJoseph Qi struct buffer_head *di_bh = NULL; 2192cf1776a9SJoseph Qi struct ocfs2_dinode *di = NULL; 2193b4df6ed8SMark Fasheh 2194b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2195b4df6ed8SMark Fasheh 2196b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 219730edc43cSJoseph Qi ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); 2198b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2199b4df6ed8SMark Fasheh 2200b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2201b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2202b4df6ed8SMark Fasheh if (ret) 2203b4df6ed8SMark Fasheh mlog_errno(ret); 2204ccd979bdSMark Fasheh 2205ccd979bdSMark Fasheh while (inode) { 2206ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2207b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2208b4107950STao Ma (unsigned long long)oi->ip_blkno); 2209ccd979bdSMark Fasheh 2210ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2211ed460cffSJoseph Qi oi->ip_next_orphan = NULL; 2212ccd979bdSMark Fasheh 221393d911fcSJoseph Qi if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { 22145955102cSAl Viro inode_lock(inode); 2215cf1776a9SJoseph Qi ret = ocfs2_rw_lock(inode, 1); 2216cf1776a9SJoseph Qi if (ret < 0) { 2217cf1776a9SJoseph Qi mlog_errno(ret); 221893d911fcSJoseph Qi goto unlock_mutex; 2219cf1776a9SJoseph Qi } 2220ed460cffSJoseph Qi /* 2221ed460cffSJoseph Qi * We need to take and drop the inode lock to 2222ed460cffSJoseph Qi * force read inode from disk. 2223ed460cffSJoseph Qi */ 2224cf1776a9SJoseph Qi ret = ocfs2_inode_lock(inode, &di_bh, 1); 2225ed460cffSJoseph Qi if (ret) { 2226ed460cffSJoseph Qi mlog_errno(ret); 2227cf1776a9SJoseph Qi goto unlock_rw; 2228ed460cffSJoseph Qi } 2229cf1776a9SJoseph Qi 2230cf1776a9SJoseph Qi di = (struct ocfs2_dinode *)di_bh->b_data; 2231ed460cffSJoseph Qi 223293d911fcSJoseph Qi if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { 2233ed460cffSJoseph Qi ret = ocfs2_truncate_file(inode, di_bh, 2234ed460cffSJoseph Qi i_size_read(inode)); 2235ed460cffSJoseph Qi if (ret < 0) { 2236ed460cffSJoseph Qi if (ret != -ENOSPC) 2237ed460cffSJoseph Qi mlog_errno(ret); 2238cf1776a9SJoseph Qi goto unlock_inode; 2239ed460cffSJoseph Qi } 2240ed460cffSJoseph Qi 224193d911fcSJoseph Qi ret = ocfs2_del_inode_from_orphan(osb, inode, 224293d911fcSJoseph Qi di_bh, 0, 0); 2243ed460cffSJoseph Qi if (ret) 2244ed460cffSJoseph Qi mlog_errno(ret); 224593d911fcSJoseph Qi } 2246cf1776a9SJoseph Qi unlock_inode: 2247cf1776a9SJoseph Qi ocfs2_inode_unlock(inode, 1); 2248512f62acSJoseph Qi brelse(di_bh); 2249512f62acSJoseph Qi di_bh = NULL; 2250cf1776a9SJoseph Qi unlock_rw: 2251cf1776a9SJoseph Qi ocfs2_rw_unlock(inode, 1); 225293d911fcSJoseph Qi unlock_mutex: 22535955102cSAl Viro inode_unlock(inode); 225493d911fcSJoseph Qi 225593d911fcSJoseph Qi /* clear dio flag in ocfs2_inode_info */ 225693d911fcSJoseph Qi oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; 225793d911fcSJoseph Qi } else { 225893d911fcSJoseph Qi spin_lock(&oi->ip_lock); 225993d911fcSJoseph Qi /* Set the proper information to get us going into 226093d911fcSJoseph Qi * ocfs2_delete_inode. */ 226193d911fcSJoseph Qi oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 226293d911fcSJoseph Qi spin_unlock(&oi->ip_lock); 226393d911fcSJoseph Qi } 226493d911fcSJoseph Qi 2265ccd979bdSMark Fasheh iput(inode); 2266ccd979bdSMark Fasheh inode = iter; 2267ccd979bdSMark Fasheh } 2268ccd979bdSMark Fasheh 2269b4df6ed8SMark Fasheh return ret; 2270ccd979bdSMark Fasheh } 2271ccd979bdSMark Fasheh 227219ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2273ccd979bdSMark Fasheh { 2274ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2275ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2276ccd979bdSMark Fasheh * or DISABLED. */ 2277ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 227819ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 227919ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2280ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2281ccd979bdSMark Fasheh 2282ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2283ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2284ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2285ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2286b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2287ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2288ccd979bdSMark Fasheh return -EBUSY; 2289ccd979bdSMark Fasheh } 2290ccd979bdSMark Fasheh 2291ccd979bdSMark Fasheh return 0; 2292ccd979bdSMark Fasheh } 2293ccd979bdSMark Fasheh 2294ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2295ccd979bdSMark Fasheh { 2296ccd979bdSMark Fasheh int status; 2297ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2298ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2299ccd979bdSMark Fasheh 2300ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2301ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2302ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2303ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2304ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2305ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2306ccd979bdSMark Fasheh 2307745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2308ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2309745ae8baSMark Fasheh || kthread_should_stop()); 2310ccd979bdSMark Fasheh 2311ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 231255b465b6SJoseph Qi if (status < 0) { 231355b465b6SJoseph Qi static unsigned long abort_warn_time; 231455b465b6SJoseph Qi 231555b465b6SJoseph Qi /* Warn about this once per minute */ 231655b465b6SJoseph Qi if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 231755b465b6SJoseph Qi mlog(ML_ERROR, "status = %d, journal is " 231855b465b6SJoseph Qi "already aborted.\n", status); 231955b465b6SJoseph Qi /* 232055b465b6SJoseph Qi * After ocfs2_commit_cache() fails, j_num_trans has a 232155b465b6SJoseph Qi * non-zero value. Sleep here to avoid a busy-wait 232255b465b6SJoseph Qi * loop. 232355b465b6SJoseph Qi */ 232455b465b6SJoseph Qi msleep_interruptible(1000); 232555b465b6SJoseph Qi } 2326ccd979bdSMark Fasheh 2327ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2328ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2329ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2330ccd979bdSMark Fasheh "shutdown\n", 2331ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2332ccd979bdSMark Fasheh } 2333ccd979bdSMark Fasheh } 2334ccd979bdSMark Fasheh 2335ccd979bdSMark Fasheh return 0; 2336ccd979bdSMark Fasheh } 2337ccd979bdSMark Fasheh 2338539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2339539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2340539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2341539d8264SSunil Mushran * a journal has been recovered by another node. 2342539d8264SSunil Mushran */ 2343ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2344ccd979bdSMark Fasheh { 2345ccd979bdSMark Fasheh int ret = 0; 2346ccd979bdSMark Fasheh unsigned int slot; 2347539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2348ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2349539d8264SSunil Mushran int journal_dirty = 0; 2350ccd979bdSMark Fasheh 2351ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2352539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2353539d8264SSunil Mushran if (ret) { 2354ccd979bdSMark Fasheh mlog_errno(ret); 2355ccd979bdSMark Fasheh goto out; 2356ccd979bdSMark Fasheh } 2357ccd979bdSMark Fasheh 2358ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2359ccd979bdSMark Fasheh 2360539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2361539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2362539d8264SSunil Mushran 2363ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2364ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2365539d8264SSunil Mushran journal_dirty = 1; 2366ccd979bdSMark Fasheh 2367ccd979bdSMark Fasheh brelse(di_bh); 2368539d8264SSunil Mushran di_bh = NULL; 2369ccd979bdSMark Fasheh } 2370ccd979bdSMark Fasheh 2371ccd979bdSMark Fasheh out: 2372539d8264SSunil Mushran if (journal_dirty) 2373539d8264SSunil Mushran ret = -EROFS; 2374ccd979bdSMark Fasheh return ret; 2375ccd979bdSMark Fasheh } 2376