1328970deSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 2fa60ce2cSMasahiro Yamada /* 3ccd979bdSMark Fasheh * journal.c 4ccd979bdSMark Fasheh * 5ccd979bdSMark Fasheh * Defines functions of journalling api 6ccd979bdSMark Fasheh * 7ccd979bdSMark Fasheh * Copyright (C) 2003, 2004 Oracle. All rights reserved. 8ccd979bdSMark Fasheh */ 9ccd979bdSMark Fasheh 10ccd979bdSMark Fasheh #include <linux/fs.h> 11ccd979bdSMark Fasheh #include <linux/types.h> 12ccd979bdSMark Fasheh #include <linux/slab.h> 13ccd979bdSMark Fasheh #include <linux/highmem.h> 14ccd979bdSMark Fasheh #include <linux/kthread.h> 1583273932SSrinivas Eeda #include <linux/time.h> 1683273932SSrinivas Eeda #include <linux/random.h> 1755b465b6SJoseph Qi #include <linux/delay.h> 18ccd979bdSMark Fasheh 19ccd979bdSMark Fasheh #include <cluster/masklog.h> 20ccd979bdSMark Fasheh 21ccd979bdSMark Fasheh #include "ocfs2.h" 22ccd979bdSMark Fasheh 23ccd979bdSMark Fasheh #include "alloc.h" 2450655ae9SJoel Becker #include "blockcheck.h" 25316f4b9fSMark Fasheh #include "dir.h" 26ccd979bdSMark Fasheh #include "dlmglue.h" 27ccd979bdSMark Fasheh #include "extent_map.h" 28ccd979bdSMark Fasheh #include "heartbeat.h" 29ccd979bdSMark Fasheh #include "inode.h" 30ccd979bdSMark Fasheh #include "journal.h" 31ccd979bdSMark Fasheh #include "localalloc.h" 32ccd979bdSMark Fasheh #include "slot_map.h" 33ccd979bdSMark Fasheh #include "super.h" 34ccd979bdSMark Fasheh #include "sysfile.h" 350cf2f763SJoel Becker #include "uptodate.h" 362205363dSJan Kara #include "quota.h" 37ed460cffSJoseph Qi #include "file.h" 38ed460cffSJoseph Qi #include "namei.h" 39ccd979bdSMark Fasheh 40ccd979bdSMark Fasheh #include "buffer_head_io.h" 41b4107950STao Ma #include "ocfs2_trace.h" 42ccd979bdSMark Fasheh 4334af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 44ccd979bdSMark Fasheh 4583273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 4683273932SSrinivas Eeda 47ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 48ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 492205363dSJan Kara int node_num, int slot_num); 50ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 51ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 5219ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 53ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 54539d8264SSunil Mushran int dirty, int replayed); 55ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 56ccd979bdSMark Fasheh int slot_num); 57ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 58ed460cffSJoseph Qi int slot, 59ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 60ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 619140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 629140db04SSrinivas Eeda int slot_num, 639140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 649140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 65ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 66ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 67ccd979bdSMark Fasheh 6819ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 6919ece546SJan Kara { 7019ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 7119ece546SJan Kara } 7219ece546SJan Kara 7319ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 7419ece546SJan Kara { 7519ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 7619ece546SJan Kara } 7719ece546SJan Kara 789140db04SSrinivas Eeda /* 799140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 809140db04SSrinivas Eeda * offline slots during recovery and mount 819140db04SSrinivas Eeda */ 829140db04SSrinivas Eeda 839140db04SSrinivas Eeda enum ocfs2_replay_state { 849140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 859140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 869140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 879140db04SSrinivas Eeda }; 889140db04SSrinivas Eeda 899140db04SSrinivas Eeda struct ocfs2_replay_map { 909140db04SSrinivas Eeda unsigned int rm_slots; 919140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 923c9210d4SGustavo A. R. Silva unsigned char rm_replay_slots[]; 939140db04SSrinivas Eeda }; 949140db04SSrinivas Eeda 95b519ea6dSJoseph Qi static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 969140db04SSrinivas Eeda { 979140db04SSrinivas Eeda if (!osb->replay_map) 989140db04SSrinivas Eeda return; 999140db04SSrinivas Eeda 1009140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1019140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1029140db04SSrinivas Eeda return; 1039140db04SSrinivas Eeda 1049140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1059140db04SSrinivas Eeda } 1069140db04SSrinivas Eeda 1079140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1089140db04SSrinivas Eeda { 1099140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1109140db04SSrinivas Eeda int i, node_num; 1119140db04SSrinivas Eeda 1129140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1139140db04SSrinivas Eeda if (osb->replay_map) 1149140db04SSrinivas Eeda return 0; 1159140db04SSrinivas Eeda 1169140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1179140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1189140db04SSrinivas Eeda 1199140db04SSrinivas Eeda if (!replay_map) { 1209140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1219140db04SSrinivas Eeda return -ENOMEM; 1229140db04SSrinivas Eeda } 1239140db04SSrinivas Eeda 1249140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1259140db04SSrinivas Eeda 1269140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1279140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1289140db04SSrinivas Eeda 1299140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1309140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1319140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1329140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1339140db04SSrinivas Eeda } 1349140db04SSrinivas Eeda 1359140db04SSrinivas Eeda osb->replay_map = replay_map; 1369140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1379140db04SSrinivas Eeda return 0; 1389140db04SSrinivas Eeda } 1399140db04SSrinivas Eeda 140b519ea6dSJoseph Qi static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, 141ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1429140db04SSrinivas Eeda { 1439140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1449140db04SSrinivas Eeda int i; 1459140db04SSrinivas Eeda 1469140db04SSrinivas Eeda if (!replay_map) 1479140db04SSrinivas Eeda return; 1489140db04SSrinivas Eeda 1499140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1509140db04SSrinivas Eeda return; 1519140db04SSrinivas Eeda 1529140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1539140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1549140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 155ed460cffSJoseph Qi NULL, NULL, 156ed460cffSJoseph Qi orphan_reco_type); 1579140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1589140db04SSrinivas Eeda } 1599140db04SSrinivas Eeda 160b519ea6dSJoseph Qi static void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1619140db04SSrinivas Eeda { 1629140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1639140db04SSrinivas Eeda 1649140db04SSrinivas Eeda if (!osb->replay_map) 1659140db04SSrinivas Eeda return; 1669140db04SSrinivas Eeda 1679140db04SSrinivas Eeda kfree(replay_map); 1689140db04SSrinivas Eeda osb->replay_map = NULL; 1699140db04SSrinivas Eeda } 1709140db04SSrinivas Eeda 171553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 172553abd04SJoel Becker { 173553abd04SJoel Becker struct ocfs2_recovery_map *rm; 174553abd04SJoel Becker 175553abd04SJoel Becker mutex_init(&osb->recovery_lock); 176553abd04SJoel Becker osb->disable_recovery = 0; 177553abd04SJoel Becker osb->recovery_thread_task = NULL; 178553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 179553abd04SJoel Becker 180553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 181553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 182553abd04SJoel Becker GFP_KERNEL); 183553abd04SJoel Becker if (!rm) { 184553abd04SJoel Becker mlog_errno(-ENOMEM); 185553abd04SJoel Becker return -ENOMEM; 186553abd04SJoel Becker } 187553abd04SJoel Becker 188553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 189553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 190553abd04SJoel Becker osb->recovery_map = rm; 191553abd04SJoel Becker 192553abd04SJoel Becker return 0; 193553abd04SJoel Becker } 194553abd04SJoel Becker 195553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 196553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 197553abd04SJoel Becker * being woken up */ 198553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 199553abd04SJoel Becker { 200553abd04SJoel Becker mb(); 201553abd04SJoel Becker return osb->recovery_thread_task != NULL; 202553abd04SJoel Becker } 203553abd04SJoel Becker 204553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 205553abd04SJoel Becker { 206553abd04SJoel Becker struct ocfs2_recovery_map *rm; 207553abd04SJoel Becker 208553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 209553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 210553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 211553abd04SJoel Becker osb->disable_recovery = 1; 212553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 213553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 214553abd04SJoel Becker 215553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 216553abd04SJoel Becker * launched, so wait for any recovery completion work to 217553abd04SJoel Becker * complete. */ 218b918c430SYi Li if (osb->ocfs2_wq) 21935ddf78eSjiangyiwen flush_workqueue(osb->ocfs2_wq); 220553abd04SJoel Becker 221553abd04SJoel Becker /* 222553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 223553abd04SJoel Becker * freed, the osb_lock is not taken here. 224553abd04SJoel Becker */ 225553abd04SJoel Becker rm = osb->recovery_map; 226553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 227553abd04SJoel Becker 228553abd04SJoel Becker kfree(rm); 229553abd04SJoel Becker } 230553abd04SJoel Becker 231553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 232553abd04SJoel Becker unsigned int node_num) 233553abd04SJoel Becker { 234553abd04SJoel Becker int i; 235553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 236553abd04SJoel Becker 237553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 238553abd04SJoel Becker 239553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 240553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 241553abd04SJoel Becker return 1; 242553abd04SJoel Becker } 243553abd04SJoel Becker 244553abd04SJoel Becker return 0; 245553abd04SJoel Becker } 246553abd04SJoel Becker 247553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 248553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 249553abd04SJoel Becker unsigned int node_num) 250553abd04SJoel Becker { 251553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 252553abd04SJoel Becker 253553abd04SJoel Becker spin_lock(&osb->osb_lock); 254553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 255553abd04SJoel Becker spin_unlock(&osb->osb_lock); 256553abd04SJoel Becker return 1; 257553abd04SJoel Becker } 258553abd04SJoel Becker 259553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 260553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 261553abd04SJoel Becker 262553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 263553abd04SJoel Becker rm->rm_used++; 264553abd04SJoel Becker spin_unlock(&osb->osb_lock); 265553abd04SJoel Becker 266553abd04SJoel Becker return 0; 267553abd04SJoel Becker } 268553abd04SJoel Becker 269553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 270553abd04SJoel Becker unsigned int node_num) 271553abd04SJoel Becker { 272553abd04SJoel Becker int i; 273553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 274553abd04SJoel Becker 275553abd04SJoel Becker spin_lock(&osb->osb_lock); 276553abd04SJoel Becker 277553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 278553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 279553abd04SJoel Becker break; 280553abd04SJoel Becker } 281553abd04SJoel Becker 282553abd04SJoel Becker if (i < rm->rm_used) { 283553abd04SJoel Becker /* XXX: be careful with the pointer math */ 284553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 285553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 286553abd04SJoel Becker rm->rm_used--; 287553abd04SJoel Becker } 288553abd04SJoel Becker 289553abd04SJoel Becker spin_unlock(&osb->osb_lock); 290553abd04SJoel Becker } 291553abd04SJoel Becker 292ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 293ccd979bdSMark Fasheh { 294ccd979bdSMark Fasheh int status = 0; 295ccd979bdSMark Fasheh unsigned int flushed; 296ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 297ccd979bdSMark Fasheh 298ccd979bdSMark Fasheh journal = osb->journal; 299ccd979bdSMark Fasheh 300ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 301ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 302ccd979bdSMark Fasheh 303b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 304b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 305b4107950STao Ma if (flushed == 0) { 306ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 307ccd979bdSMark Fasheh goto finally; 308ccd979bdSMark Fasheh } 309ccd979bdSMark Fasheh 3102b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 311*01d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 3122b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 313ccd979bdSMark Fasheh if (status < 0) { 314ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 315ccd979bdSMark Fasheh mlog_errno(status); 316ccd979bdSMark Fasheh goto finally; 317ccd979bdSMark Fasheh } 318ccd979bdSMark Fasheh 319f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 320ccd979bdSMark Fasheh 321ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 322ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 323ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 324ccd979bdSMark Fasheh 325b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 326ccd979bdSMark Fasheh 32734d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 328ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 329ccd979bdSMark Fasheh finally: 330ccd979bdSMark Fasheh return status; 331ccd979bdSMark Fasheh } 332ccd979bdSMark Fasheh 3331fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 334ccd979bdSMark Fasheh { 335ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3361fabe148SMark Fasheh handle_t *handle; 337ccd979bdSMark Fasheh 338ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 339ccd979bdSMark Fasheh 34065eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 34165eff9ccSMark Fasheh return ERR_PTR(-EROFS); 342ccd979bdSMark Fasheh 343ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 344ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 345ccd979bdSMark Fasheh 34690e86a63SJan Kara /* Nested transaction? Just return the handle... */ 34790e86a63SJan Kara if (journal_current_handle()) 34890e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 349ccd979bdSMark Fasheh 350fef6925cSJan Kara sb_start_intwrite(osb->sb); 351fef6925cSJan Kara 352ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 353ccd979bdSMark Fasheh 3542b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3551fabe148SMark Fasheh if (IS_ERR(handle)) { 356ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 357fef6925cSJan Kara sb_end_intwrite(osb->sb); 358ccd979bdSMark Fasheh 3591fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 360ccd979bdSMark Fasheh 361ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 3627ecef14aSJoe Perches ocfs2_abort(osb->sb, "Detected aborted journal\n"); 3631fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 364ccd979bdSMark Fasheh } 365c271c5c2SSunil Mushran } else { 366c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 367ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 368c271c5c2SSunil Mushran } 369ccd979bdSMark Fasheh 370ccd979bdSMark Fasheh return handle; 371ccd979bdSMark Fasheh } 372ccd979bdSMark Fasheh 3731fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3741fabe148SMark Fasheh handle_t *handle) 375ccd979bdSMark Fasheh { 37690e86a63SJan Kara int ret, nested; 37702dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 378ccd979bdSMark Fasheh 379ccd979bdSMark Fasheh BUG_ON(!handle); 380ccd979bdSMark Fasheh 38190e86a63SJan Kara nested = handle->h_ref > 1; 3822b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3831fabe148SMark Fasheh if (ret < 0) 3841fabe148SMark Fasheh mlog_errno(ret); 385ccd979bdSMark Fasheh 386fef6925cSJan Kara if (!nested) { 387ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 388fef6925cSJan Kara sb_end_intwrite(osb->sb); 389fef6925cSJan Kara } 390ccd979bdSMark Fasheh 3911fabe148SMark Fasheh return ret; 392ccd979bdSMark Fasheh } 393ccd979bdSMark Fasheh 394ccd979bdSMark Fasheh /* 395c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 396ccd979bdSMark Fasheh * 3972b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 398e8aed345SMark Fasheh * and then restart the transaction. Before calling 399e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 400e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 401e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 402e8aed345SMark Fasheh * 403ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 404ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 405ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 406ccd979bdSMark Fasheh * 407ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 408ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 409ccd979bdSMark Fasheh * cluster locks associated with this handle. 410ccd979bdSMark Fasheh */ 4111fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 412ccd979bdSMark Fasheh { 413c901fb00STao Ma int status, old_nblocks; 414ccd979bdSMark Fasheh 415ccd979bdSMark Fasheh BUG_ON(!handle); 416c901fb00STao Ma BUG_ON(nblocks < 0); 417ccd979bdSMark Fasheh 418c901fb00STao Ma if (!nblocks) 419c901fb00STao Ma return 0; 420c901fb00STao Ma 4219797a902SJan Kara old_nblocks = jbd2_handle_buffer_credits(handle); 422ccd979bdSMark Fasheh 423b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 424ccd979bdSMark Fasheh 425e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4260879c584SMark Fasheh status = 1; 4270879c584SMark Fasheh #else 428fdc3ef88SJan Kara status = jbd2_journal_extend(handle, nblocks, 0); 429ccd979bdSMark Fasheh if (status < 0) { 430ccd979bdSMark Fasheh mlog_errno(status); 431ccd979bdSMark Fasheh goto bail; 432ccd979bdSMark Fasheh } 4330879c584SMark Fasheh #endif 434ccd979bdSMark Fasheh 435ccd979bdSMark Fasheh if (status > 0) { 436b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 437c901fb00STao Ma status = jbd2_journal_restart(handle, 438c901fb00STao Ma old_nblocks + nblocks); 439ccd979bdSMark Fasheh if (status < 0) { 440ccd979bdSMark Fasheh mlog_errno(status); 441ccd979bdSMark Fasheh goto bail; 442ccd979bdSMark Fasheh } 44301ddf1e1SMark Fasheh } 444ccd979bdSMark Fasheh 445ccd979bdSMark Fasheh status = 0; 446ccd979bdSMark Fasheh bail: 447ccd979bdSMark Fasheh return status; 448ccd979bdSMark Fasheh } 449ccd979bdSMark Fasheh 4502b1e55c3SYounger Liu /* 4512b1e55c3SYounger Liu * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 4522b1e55c3SYounger Liu * If that fails, restart the transaction & regain write access for the 4532b1e55c3SYounger Liu * buffer head which is used for metadata modifications. 4542b1e55c3SYounger Liu * Taken from Ext4: extend_or_restart_transaction() 4552b1e55c3SYounger Liu */ 4562b1e55c3SYounger Liu int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 4572b1e55c3SYounger Liu { 4582b1e55c3SYounger Liu int status, old_nblks; 4592b1e55c3SYounger Liu 4602b1e55c3SYounger Liu BUG_ON(!handle); 4612b1e55c3SYounger Liu 4629797a902SJan Kara old_nblks = jbd2_handle_buffer_credits(handle); 4632b1e55c3SYounger Liu trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 4642b1e55c3SYounger Liu 4652b1e55c3SYounger Liu if (old_nblks < thresh) 4662b1e55c3SYounger Liu return 0; 4672b1e55c3SYounger Liu 468fdc3ef88SJan Kara status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0); 4692b1e55c3SYounger Liu if (status < 0) { 4702b1e55c3SYounger Liu mlog_errno(status); 4712b1e55c3SYounger Liu goto bail; 4722b1e55c3SYounger Liu } 4732b1e55c3SYounger Liu 4742b1e55c3SYounger Liu if (status > 0) { 4752b1e55c3SYounger Liu status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 4762b1e55c3SYounger Liu if (status < 0) 4772b1e55c3SYounger Liu mlog_errno(status); 4782b1e55c3SYounger Liu } 4792b1e55c3SYounger Liu 4802b1e55c3SYounger Liu bail: 4812b1e55c3SYounger Liu return status; 4822b1e55c3SYounger Liu } 4832b1e55c3SYounger Liu 4842b1e55c3SYounger Liu 48550655ae9SJoel Becker struct ocfs2_triggers { 48650655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 48750655ae9SJoel Becker int ot_offset; 48850655ae9SJoel Becker }; 48950655ae9SJoel Becker 49050655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 49150655ae9SJoel Becker { 49250655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 49350655ae9SJoel Becker } 49450655ae9SJoel Becker 49513ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 49650655ae9SJoel Becker struct buffer_head *bh, 49750655ae9SJoel Becker void *data, size_t size) 49850655ae9SJoel Becker { 49950655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 50050655ae9SJoel Becker 50150655ae9SJoel Becker /* 50250655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 50350655ae9SJoel Becker * must unconditionally compute the ecc data. 50450655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 50550655ae9SJoel Becker * metaecc is enabled. 50650655ae9SJoel Becker */ 50750655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 50850655ae9SJoel Becker } 50950655ae9SJoel Becker 51050655ae9SJoel Becker /* 51150655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 51250655ae9SJoel Becker * offset depends on the blocksize. 51350655ae9SJoel Becker */ 51413ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 51550655ae9SJoel Becker struct buffer_head *bh, 51650655ae9SJoel Becker void *data, size_t size) 51750655ae9SJoel Becker { 51850655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 51950655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 52050655ae9SJoel Becker 52150655ae9SJoel Becker /* 52250655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 52350655ae9SJoel Becker * must unconditionally compute the ecc data. 52450655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 52550655ae9SJoel Becker * metaecc is enabled. 52650655ae9SJoel Becker */ 52750655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 52850655ae9SJoel Becker } 52950655ae9SJoel Becker 530c175a518SJoel Becker /* 531c175a518SJoel Becker * Directory blocks also have their own trigger because the 532c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 533c175a518SJoel Becker */ 53413ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 535c175a518SJoel Becker struct buffer_head *bh, 536c175a518SJoel Becker void *data, size_t size) 537c175a518SJoel Becker { 538c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 539c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 540c175a518SJoel Becker 541c175a518SJoel Becker /* 542c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 543c175a518SJoel Becker * must unconditionally compute the ecc data. 544c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 545c175a518SJoel Becker * metaecc is enabled. 546c175a518SJoel Becker */ 547c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 548c175a518SJoel Becker } 549c175a518SJoel Becker 55050655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 55150655ae9SJoel Becker struct buffer_head *bh) 55250655ae9SJoel Becker { 55350655ae9SJoel Becker mlog(ML_ERROR, 55450655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 55550655ae9SJoel Becker "bh->b_blocknr = %llu\n", 55650655ae9SJoel Becker (unsigned long)bh, 55750655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 55850655ae9SJoel Becker 55974e364adSXue jiufei ocfs2_error(bh->b_bdev->bd_super, 56050655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 56150655ae9SJoel Becker } 56250655ae9SJoel Becker 56350655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 56450655ae9SJoel Becker .ot_triggers = { 56513ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 56650655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 56750655ae9SJoel Becker }, 56850655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 56950655ae9SJoel Becker }; 57050655ae9SJoel Becker 57150655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 57250655ae9SJoel Becker .ot_triggers = { 57313ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 57450655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 57550655ae9SJoel Becker }, 57650655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 57750655ae9SJoel Becker }; 57850655ae9SJoel Becker 57993c97087STao Ma static struct ocfs2_triggers rb_triggers = { 58093c97087STao Ma .ot_triggers = { 58113ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58293c97087STao Ma .t_abort = ocfs2_abort_trigger, 58393c97087STao Ma }, 58493c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 58593c97087STao Ma }; 58693c97087STao Ma 58750655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 58850655ae9SJoel Becker .ot_triggers = { 58913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 59050655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 59150655ae9SJoel Becker }, 59250655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 59350655ae9SJoel Becker }; 59450655ae9SJoel Becker 595c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 596c175a518SJoel Becker .ot_triggers = { 59713ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 598c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 599c175a518SJoel Becker }, 600c175a518SJoel Becker }; 601c175a518SJoel Becker 60250655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 60350655ae9SJoel Becker .ot_triggers = { 60413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 60550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 60650655ae9SJoel Becker }, 60750655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 60850655ae9SJoel Becker }; 60950655ae9SJoel Becker 61050655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 61150655ae9SJoel Becker .ot_triggers = { 61213ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 61350655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 61450655ae9SJoel Becker }, 61550655ae9SJoel Becker }; 61650655ae9SJoel Becker 6179b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 6189b7895efSMark Fasheh .ot_triggers = { 61913ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6209b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6219b7895efSMark Fasheh }, 6229b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 6239b7895efSMark Fasheh }; 6249b7895efSMark Fasheh 6259b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 6269b7895efSMark Fasheh .ot_triggers = { 62713ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6289b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6299b7895efSMark Fasheh }, 6309b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6319b7895efSMark Fasheh }; 6329b7895efSMark Fasheh 63350655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6340cf2f763SJoel Becker struct ocfs2_caching_info *ci, 635ccd979bdSMark Fasheh struct buffer_head *bh, 63650655ae9SJoel Becker struct ocfs2_triggers *triggers, 637ccd979bdSMark Fasheh int type) 638ccd979bdSMark Fasheh { 639ccd979bdSMark Fasheh int status; 6400cf2f763SJoel Becker struct ocfs2_super *osb = 6410cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 642ccd979bdSMark Fasheh 6430cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 644ccd979bdSMark Fasheh BUG_ON(!handle); 645ccd979bdSMark Fasheh BUG_ON(!bh); 646ccd979bdSMark Fasheh 647b4107950STao Ma trace_ocfs2_journal_access( 648b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 649b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 650ccd979bdSMark Fasheh 651ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 652ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 653ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 654d984187eSpiaojun mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", 655d984187eSpiaojun (unsigned long long)bh->b_blocknr, bh->b_state); 656acf8fdbeSJoseph Qi 657acf8fdbeSJoseph Qi lock_buffer(bh); 658acf8fdbeSJoseph Qi /* 659d984187eSpiaojun * A previous transaction with a couple of buffer heads fail 660d984187eSpiaojun * to checkpoint, so all the bhs are marked as BH_Write_EIO. 661d984187eSpiaojun * For current transaction, the bh is just among those error 662d984187eSpiaojun * bhs which previous transaction handle. We can't just clear 663d984187eSpiaojun * its BH_Write_EIO and reuse directly, since other bhs are 664d984187eSpiaojun * not written to disk yet and that will cause metadata 665d984187eSpiaojun * inconsistency. So we should set fs read-only to avoid 666d984187eSpiaojun * further damage. 667acf8fdbeSJoseph Qi */ 668acf8fdbeSJoseph Qi if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { 669acf8fdbeSJoseph Qi unlock_buffer(bh); 670d984187eSpiaojun return ocfs2_error(osb->sb, "A previous attempt to " 671d984187eSpiaojun "write this buffer head failed\n"); 672acf8fdbeSJoseph Qi } 673acf8fdbeSJoseph Qi unlock_buffer(bh); 674ccd979bdSMark Fasheh } 675ccd979bdSMark Fasheh 6760cf2f763SJoel Becker /* Set the current transaction information on the ci so 677ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6780cf2f763SJoel Becker * on this ci or not. We're protected from the commit 679ccd979bdSMark Fasheh * thread updating the current transaction id until 680ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 681ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6820cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 683ccd979bdSMark Fasheh 6840cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 685ccd979bdSMark Fasheh switch (type) { 686ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 687ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 6882b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 689ccd979bdSMark Fasheh break; 690ccd979bdSMark Fasheh 691ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 6922b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 693ccd979bdSMark Fasheh break; 694ccd979bdSMark Fasheh 695ccd979bdSMark Fasheh default: 696ccd979bdSMark Fasheh status = -EINVAL; 697af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 698ccd979bdSMark Fasheh } 6990cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 70050655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 7010cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 702ccd979bdSMark Fasheh 703ccd979bdSMark Fasheh if (status < 0) 704ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 705ccd979bdSMark Fasheh status, type); 706ccd979bdSMark Fasheh 707ccd979bdSMark Fasheh return status; 708ccd979bdSMark Fasheh } 709ccd979bdSMark Fasheh 7100cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 71150655ae9SJoel Becker struct buffer_head *bh, int type) 71250655ae9SJoel Becker { 7130cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 71450655ae9SJoel Becker } 71550655ae9SJoel Becker 7160cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 71750655ae9SJoel Becker struct buffer_head *bh, int type) 71850655ae9SJoel Becker { 7190cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 72050655ae9SJoel Becker } 72150655ae9SJoel Becker 72293c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 72393c97087STao Ma struct buffer_head *bh, int type) 72493c97087STao Ma { 72593c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 72693c97087STao Ma type); 72793c97087STao Ma } 72893c97087STao Ma 7290cf2f763SJoel Becker int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 73050655ae9SJoel Becker struct buffer_head *bh, int type) 73150655ae9SJoel Becker { 7320cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 73350655ae9SJoel Becker } 73450655ae9SJoel Becker 7350cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 73650655ae9SJoel Becker struct buffer_head *bh, int type) 73750655ae9SJoel Becker { 7380cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 73950655ae9SJoel Becker } 74050655ae9SJoel Becker 7410cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 74250655ae9SJoel Becker struct buffer_head *bh, int type) 74350655ae9SJoel Becker { 7440cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 74550655ae9SJoel Becker } 74650655ae9SJoel Becker 7470cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 74850655ae9SJoel Becker struct buffer_head *bh, int type) 74950655ae9SJoel Becker { 7500cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 75150655ae9SJoel Becker } 75250655ae9SJoel Becker 7530cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7549b7895efSMark Fasheh struct buffer_head *bh, int type) 7559b7895efSMark Fasheh { 7560cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7579b7895efSMark Fasheh } 7589b7895efSMark Fasheh 7590cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7609b7895efSMark Fasheh struct buffer_head *bh, int type) 7619b7895efSMark Fasheh { 7620cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7639b7895efSMark Fasheh } 7649b7895efSMark Fasheh 7650cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 76650655ae9SJoel Becker struct buffer_head *bh, int type) 76750655ae9SJoel Becker { 7680cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 76950655ae9SJoel Becker } 77050655ae9SJoel Becker 771ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 772ccd979bdSMark Fasheh { 773ccd979bdSMark Fasheh int status; 774ccd979bdSMark Fasheh 775b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 776ccd979bdSMark Fasheh 7772b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 778e272e7f0SJoseph Qi if (status) { 779e272e7f0SJoseph Qi mlog_errno(status); 780e272e7f0SJoseph Qi if (!is_handle_aborted(handle)) { 781e272e7f0SJoseph Qi journal_t *journal = handle->h_transaction->t_journal; 782e272e7f0SJoseph Qi struct super_block *sb = bh->b_bdev->bd_super; 783e272e7f0SJoseph Qi 784e272e7f0SJoseph Qi mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " 785e272e7f0SJoseph Qi "Aborting transaction and journal.\n"); 786e272e7f0SJoseph Qi handle->h_err = status; 787e272e7f0SJoseph Qi jbd2_journal_abort_handle(handle); 788e272e7f0SJoseph Qi jbd2_journal_abort(journal, status); 789e272e7f0SJoseph Qi ocfs2_abort(sb, "Journal already aborted.\n"); 790e272e7f0SJoseph Qi } 791e272e7f0SJoseph Qi } 792ccd979bdSMark Fasheh } 793ccd979bdSMark Fasheh 7942b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 795ccd979bdSMark Fasheh 796ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 797ccd979bdSMark Fasheh { 798ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 799d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 800d147b3d6SMark Fasheh 801d147b3d6SMark Fasheh if (osb->osb_commit_interval) 802d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 803ccd979bdSMark Fasheh 804a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 805d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 806ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 8072b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 808ccd979bdSMark Fasheh else 8092b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 810a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 811ccd979bdSMark Fasheh } 812ccd979bdSMark Fasheh 813ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 814ccd979bdSMark Fasheh { 815ccd979bdSMark Fasheh int status = -1; 816ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 817ccd979bdSMark Fasheh journal_t *j_journal = NULL; 818ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 819ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 820ccd979bdSMark Fasheh struct ocfs2_super *osb; 821e63aecb6SMark Fasheh int inode_lock = 0; 822ccd979bdSMark Fasheh 823ccd979bdSMark Fasheh BUG_ON(!journal); 824ccd979bdSMark Fasheh 825ccd979bdSMark Fasheh osb = journal->j_osb; 826ccd979bdSMark Fasheh 827ccd979bdSMark Fasheh /* already have the inode for our journal */ 828ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 829ccd979bdSMark Fasheh osb->slot_num); 830ccd979bdSMark Fasheh if (inode == NULL) { 831ccd979bdSMark Fasheh status = -EACCES; 832ccd979bdSMark Fasheh mlog_errno(status); 833ccd979bdSMark Fasheh goto done; 834ccd979bdSMark Fasheh } 835ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 836ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 837ccd979bdSMark Fasheh iput(inode); 838ccd979bdSMark Fasheh inode = NULL; 839ccd979bdSMark Fasheh status = -EACCES; 840ccd979bdSMark Fasheh goto done; 841ccd979bdSMark Fasheh } 842ccd979bdSMark Fasheh 843ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 844ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 845ccd979bdSMark Fasheh 8466eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 8476eff5790SMark Fasheh * changes in a live cluster so it can be considered an 8486eff5790SMark Fasheh * exception to the rule. */ 849e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 850ccd979bdSMark Fasheh if (status < 0) { 851ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 852ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 853ccd979bdSMark Fasheh goto done; 854ccd979bdSMark Fasheh } 855ccd979bdSMark Fasheh 856e63aecb6SMark Fasheh inode_lock = 1; 857ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 858ccd979bdSMark Fasheh 859f17c20ddSJunxiao Bi if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 860ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 861f17c20ddSJunxiao Bi i_size_read(inode)); 862ccd979bdSMark Fasheh status = -EINVAL; 863ccd979bdSMark Fasheh goto done; 864ccd979bdSMark Fasheh } 865ccd979bdSMark Fasheh 866f17c20ddSJunxiao Bi trace_ocfs2_journal_init(i_size_read(inode), 867b4107950STao Ma (unsigned long long)inode->i_blocks, 868b4107950STao Ma OCFS2_I(inode)->ip_clusters); 869ccd979bdSMark Fasheh 870ccd979bdSMark Fasheh /* call the kernels journal init function now */ 8712b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 872ccd979bdSMark Fasheh if (j_journal == NULL) { 873ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 874ccd979bdSMark Fasheh status = -EINVAL; 875ccd979bdSMark Fasheh goto done; 876ccd979bdSMark Fasheh } 877ccd979bdSMark Fasheh 878ede7dc7fSHarshad Shirwadkar trace_ocfs2_journal_init_maxlen(j_journal->j_total_len); 879ccd979bdSMark Fasheh 880ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 881ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 882ccd979bdSMark Fasheh 883ccd979bdSMark Fasheh journal->j_journal = j_journal; 884342af94eSMauricio Faria de Oliveira journal->j_journal->j_submit_inode_data_buffers = 885342af94eSMauricio Faria de Oliveira jbd2_journal_submit_inode_data_buffers; 886342af94eSMauricio Faria de Oliveira journal->j_journal->j_finish_inode_data_buffers = 887342af94eSMauricio Faria de Oliveira jbd2_journal_finish_inode_data_buffers; 888ccd979bdSMark Fasheh journal->j_inode = inode; 889ccd979bdSMark Fasheh journal->j_bh = bh; 890ccd979bdSMark Fasheh 891ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 892ccd979bdSMark Fasheh 893ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 894ccd979bdSMark Fasheh 895ccd979bdSMark Fasheh status = 0; 896ccd979bdSMark Fasheh done: 897ccd979bdSMark Fasheh if (status < 0) { 898e63aecb6SMark Fasheh if (inode_lock) 899e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 900ccd979bdSMark Fasheh brelse(bh); 901ccd979bdSMark Fasheh if (inode) { 902ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 903ccd979bdSMark Fasheh iput(inode); 904ccd979bdSMark Fasheh } 905ccd979bdSMark Fasheh } 906ccd979bdSMark Fasheh 907ccd979bdSMark Fasheh return status; 908ccd979bdSMark Fasheh } 909ccd979bdSMark Fasheh 910539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 911539d8264SSunil Mushran { 912539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 913539d8264SSunil Mushran } 914539d8264SSunil Mushran 915539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 916539d8264SSunil Mushran { 917539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 918539d8264SSunil Mushran } 919539d8264SSunil Mushran 920ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 921539d8264SSunil Mushran int dirty, int replayed) 922ccd979bdSMark Fasheh { 923ccd979bdSMark Fasheh int status; 924ccd979bdSMark Fasheh unsigned int flags; 925ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 926ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 927ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 928ccd979bdSMark Fasheh 929ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 93010995aa2SJoel Becker 93110995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 93210995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 93310995aa2SJoel Becker * code bug if we mess it up. */ 93410995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 935ccd979bdSMark Fasheh 936ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 937ccd979bdSMark Fasheh if (dirty) 938ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 939ccd979bdSMark Fasheh else 940ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 941ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 942ccd979bdSMark Fasheh 943539d8264SSunil Mushran if (replayed) 944539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 945539d8264SSunil Mushran 94613723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 9478cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 948ccd979bdSMark Fasheh if (status < 0) 949ccd979bdSMark Fasheh mlog_errno(status); 950ccd979bdSMark Fasheh 951ccd979bdSMark Fasheh return status; 952ccd979bdSMark Fasheh } 953ccd979bdSMark Fasheh 954ccd979bdSMark Fasheh /* 955ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 956ccd979bdSMark Fasheh * call. 957ccd979bdSMark Fasheh */ 958ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 959ccd979bdSMark Fasheh { 960ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 961ccd979bdSMark Fasheh int status = 0; 962ccd979bdSMark Fasheh struct inode *inode = NULL; 963ccd979bdSMark Fasheh int num_running_trans = 0; 964ccd979bdSMark Fasheh 965ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 966ccd979bdSMark Fasheh 967ccd979bdSMark Fasheh journal = osb->journal; 968ccd979bdSMark Fasheh if (!journal) 969ccd979bdSMark Fasheh goto done; 970ccd979bdSMark Fasheh 971ccd979bdSMark Fasheh inode = journal->j_inode; 972ccd979bdSMark Fasheh 973ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 974ccd979bdSMark Fasheh goto done; 975ccd979bdSMark Fasheh 9762b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 977ccd979bdSMark Fasheh if (!igrab(inode)) 978ccd979bdSMark Fasheh BUG(); 979ccd979bdSMark Fasheh 980ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 981b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 982ccd979bdSMark Fasheh 983ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 984ccd979bdSMark Fasheh * release any locks that are still held. 985ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 986ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 987ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 988ccd979bdSMark Fasheh 989ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 990ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 991ccd979bdSMark Fasheh * completely destroy the journal. */ 992ccd979bdSMark Fasheh if (osb->commit_task) { 993ccd979bdSMark Fasheh /* Wait for the commit thread */ 994b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 995ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 996ccd979bdSMark Fasheh osb->commit_task = NULL; 997ccd979bdSMark Fasheh } 998ccd979bdSMark Fasheh 999ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 1000ccd979bdSMark Fasheh 1001c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 10022b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 1003*01d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 10042b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 1005c271c5c2SSunil Mushran if (status < 0) 1006c271c5c2SSunil Mushran mlog_errno(status); 1007c271c5c2SSunil Mushran } 1008c271c5c2SSunil Mushran 1009d85400afSJunxiao Bi /* Shutdown the kernel journal system */ 1010d85400afSJunxiao Bi if (!jbd2_journal_destroy(journal->j_journal) && !status) { 1011c271c5c2SSunil Mushran /* 1012c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 1013c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 1014c271c5c2SSunil Mushran */ 1015539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 1016ccd979bdSMark Fasheh if (status < 0) 1017ccd979bdSMark Fasheh mlog_errno(status); 1018c271c5c2SSunil Mushran } 1019ae0dff68SSunil Mushran journal->j_journal = NULL; 1020ccd979bdSMark Fasheh 1021ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 1022ccd979bdSMark Fasheh 1023ccd979bdSMark Fasheh /* unlock our journal */ 1024e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1025ccd979bdSMark Fasheh 1026ccd979bdSMark Fasheh brelse(journal->j_bh); 1027ccd979bdSMark Fasheh journal->j_bh = NULL; 1028ccd979bdSMark Fasheh 1029ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 1030ccd979bdSMark Fasheh 1031ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 1032ccd979bdSMark Fasheh done: 1033ccd979bdSMark Fasheh iput(inode); 1034ccd979bdSMark Fasheh } 1035ccd979bdSMark Fasheh 1036ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 1037ccd979bdSMark Fasheh journal_t *journal, 1038ccd979bdSMark Fasheh int slot) 1039ccd979bdSMark Fasheh { 1040ccd979bdSMark Fasheh int olderr; 1041ccd979bdSMark Fasheh 10422b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 1043ccd979bdSMark Fasheh if (olderr) { 1044ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 1045ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 1046ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 1047ccd979bdSMark Fasheh sb->s_id); 1048ccd979bdSMark Fasheh 10492b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 10502b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 1051ccd979bdSMark Fasheh } 1052ccd979bdSMark Fasheh } 1053ccd979bdSMark Fasheh 1054539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1055ccd979bdSMark Fasheh { 1056ccd979bdSMark Fasheh int status = 0; 1057ccd979bdSMark Fasheh struct ocfs2_super *osb; 1058ccd979bdSMark Fasheh 1059b1f3550fSJulia Lawall BUG_ON(!journal); 1060ccd979bdSMark Fasheh 1061ccd979bdSMark Fasheh osb = journal->j_osb; 1062ccd979bdSMark Fasheh 10632b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1064ccd979bdSMark Fasheh if (status < 0) { 1065ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1066ccd979bdSMark Fasheh goto done; 1067ccd979bdSMark Fasheh } 1068ccd979bdSMark Fasheh 1069ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1070ccd979bdSMark Fasheh 1071397eac17SKai Li if (replayed) { 1072397eac17SKai Li jbd2_journal_lock_updates(journal->j_journal); 1073*01d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 1074397eac17SKai Li jbd2_journal_unlock_updates(journal->j_journal); 1075397eac17SKai Li if (status < 0) 1076397eac17SKai Li mlog_errno(status); 1077397eac17SKai Li } 1078397eac17SKai Li 1079539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1080ccd979bdSMark Fasheh if (status < 0) { 1081ccd979bdSMark Fasheh mlog_errno(status); 1082ccd979bdSMark Fasheh goto done; 1083ccd979bdSMark Fasheh } 1084ccd979bdSMark Fasheh 1085ccd979bdSMark Fasheh /* Launch the commit thread */ 1086c271c5c2SSunil Mushran if (!local) { 1087c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 10885afc44e2SJoseph Qi "ocfs2cmt-%s", osb->uuid_str); 1089ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1090ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1091ccd979bdSMark Fasheh osb->commit_task = NULL; 1092c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1093c271c5c2SSunil Mushran "error=%d", status); 1094ccd979bdSMark Fasheh goto done; 1095ccd979bdSMark Fasheh } 1096c271c5c2SSunil Mushran } else 1097c271c5c2SSunil Mushran osb->commit_task = NULL; 1098ccd979bdSMark Fasheh 1099ccd979bdSMark Fasheh done: 1100ccd979bdSMark Fasheh return status; 1101ccd979bdSMark Fasheh } 1102ccd979bdSMark Fasheh 1103ccd979bdSMark Fasheh 1104ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1105ccd979bdSMark Fasheh * mark the journal clean */ 1106ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1107ccd979bdSMark Fasheh { 1108ccd979bdSMark Fasheh int status; 1109ccd979bdSMark Fasheh 1110ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1111ccd979bdSMark Fasheh 11122b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1113ccd979bdSMark Fasheh if (status < 0) { 1114ccd979bdSMark Fasheh mlog_errno(status); 1115ccd979bdSMark Fasheh goto bail; 1116ccd979bdSMark Fasheh } 1117ccd979bdSMark Fasheh 1118539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1119ccd979bdSMark Fasheh if (status < 0) 1120ccd979bdSMark Fasheh mlog_errno(status); 1121ccd979bdSMark Fasheh 1122ccd979bdSMark Fasheh bail: 1123ccd979bdSMark Fasheh return status; 1124ccd979bdSMark Fasheh } 1125ccd979bdSMark Fasheh 1126553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1127553abd04SJoel Becker { 1128553abd04SJoel Becker int empty; 1129553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1130553abd04SJoel Becker 1131553abd04SJoel Becker spin_lock(&osb->osb_lock); 1132553abd04SJoel Becker empty = (rm->rm_used == 0); 1133553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1134553abd04SJoel Becker 1135553abd04SJoel Becker return empty; 1136553abd04SJoel Becker } 1137553abd04SJoel Becker 1138553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1139553abd04SJoel Becker { 1140553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1141553abd04SJoel Becker } 1142553abd04SJoel Becker 1143ccd979bdSMark Fasheh /* 1144ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1145ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1146ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1147ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1148ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1149ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1150ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1151ccd979bdSMark Fasheh * as we have the lock by now :) 1152ccd979bdSMark Fasheh */ 1153ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1154ccd979bdSMark Fasheh { 1155ccd979bdSMark Fasheh int status = 0; 11564f902c37SMark Fasheh int i; 11578110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 11580b492f68SJunxiao Bi struct buffer_head *bh = NULL; 11590b492f68SJunxiao Bi struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1160ccd979bdSMark Fasheh 1161f17c20ddSJunxiao Bi num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1162ccd979bdSMark Fasheh v_blkno = 0; 11638110b073SMark Fasheh while (v_blkno < num_blocks) { 1164ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 116549cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1166ccd979bdSMark Fasheh if (status < 0) { 1167ccd979bdSMark Fasheh mlog_errno(status); 1168ccd979bdSMark Fasheh goto bail; 1169ccd979bdSMark Fasheh } 1170ccd979bdSMark Fasheh 11710b492f68SJunxiao Bi for (i = 0; i < p_blocks; i++, p_blkno++) { 11720b492f68SJunxiao Bi bh = __find_get_block(osb->sb->s_bdev, p_blkno, 11730b492f68SJunxiao Bi osb->sb->s_blocksize); 11740b492f68SJunxiao Bi /* block not cached. */ 11750b492f68SJunxiao Bi if (!bh) 11760b492f68SJunxiao Bi continue; 1177ccd979bdSMark Fasheh 11780b492f68SJunxiao Bi brelse(bh); 11790b492f68SJunxiao Bi bh = NULL; 1180dd4a2c2bSMark Fasheh /* We are reading journal data which should not 11810b492f68SJunxiao Bi * be put in the uptodate cache. 11820b492f68SJunxiao Bi */ 11830b492f68SJunxiao Bi status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); 1184ccd979bdSMark Fasheh if (status < 0) { 1185ccd979bdSMark Fasheh mlog_errno(status); 1186ccd979bdSMark Fasheh goto bail; 1187ccd979bdSMark Fasheh } 1188ccd979bdSMark Fasheh 11890b492f68SJunxiao Bi brelse(bh); 11900b492f68SJunxiao Bi bh = NULL; 1191ccd979bdSMark Fasheh } 1192ccd979bdSMark Fasheh 1193ccd979bdSMark Fasheh v_blkno += p_blocks; 1194ccd979bdSMark Fasheh } 1195ccd979bdSMark Fasheh 1196ccd979bdSMark Fasheh bail: 1197ccd979bdSMark Fasheh return status; 1198ccd979bdSMark Fasheh } 1199ccd979bdSMark Fasheh 1200ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1201ccd979bdSMark Fasheh struct list_head lri_list; 1202ccd979bdSMark Fasheh int lri_slot; 1203ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1204ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 12052205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1206ed460cffSJoseph Qi enum ocfs2_orphan_reco_type lri_orphan_reco_type; 1207ccd979bdSMark Fasheh }; 1208ccd979bdSMark Fasheh 1209ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1210ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1211ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1212ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1213ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1214ccd979bdSMark Fasheh * recovering nodes into account. 1215ccd979bdSMark Fasheh * 1216ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1217ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1218ccd979bdSMark Fasheh */ 1219c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1220ccd979bdSMark Fasheh { 1221b4107950STao Ma int ret = 0; 1222c4028958SDavid Howells struct ocfs2_journal *journal = 1223c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1224c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1225ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1226800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 12272205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1228ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 1229ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1230ccd979bdSMark Fasheh 1231b4107950STao Ma trace_ocfs2_complete_recovery( 1232b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1233ccd979bdSMark Fasheh 1234ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1235ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1236ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1237ccd979bdSMark Fasheh 1238800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1239ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1240ccd979bdSMark Fasheh 124119ece546SJan Kara ocfs2_wait_on_quotas(osb); 124219ece546SJan Kara 1243ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1244b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1245b4107950STao Ma qrec = item->lri_qrec; 1246ed460cffSJoseph Qi orphan_reco_type = item->lri_orphan_reco_type; 1247ccd979bdSMark Fasheh 1248b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1249b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1250b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1251b4107950STao Ma qrec); 1252b4107950STao Ma 1253b4107950STao Ma if (la_dinode) { 1254ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1255ccd979bdSMark Fasheh la_dinode); 1256ccd979bdSMark Fasheh if (ret < 0) 1257ccd979bdSMark Fasheh mlog_errno(ret); 1258ccd979bdSMark Fasheh 1259ccd979bdSMark Fasheh kfree(la_dinode); 1260ccd979bdSMark Fasheh } 1261ccd979bdSMark Fasheh 1262ccd979bdSMark Fasheh if (tl_dinode) { 1263ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1264ccd979bdSMark Fasheh tl_dinode); 1265ccd979bdSMark Fasheh if (ret < 0) 1266ccd979bdSMark Fasheh mlog_errno(ret); 1267ccd979bdSMark Fasheh 1268ccd979bdSMark Fasheh kfree(tl_dinode); 1269ccd979bdSMark Fasheh } 1270ccd979bdSMark Fasheh 1271ed460cffSJoseph Qi ret = ocfs2_recover_orphans(osb, item->lri_slot, 1272ed460cffSJoseph Qi orphan_reco_type); 1273ccd979bdSMark Fasheh if (ret < 0) 1274ccd979bdSMark Fasheh mlog_errno(ret); 1275ccd979bdSMark Fasheh 12762205363dSJan Kara if (qrec) { 12772205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 12782205363dSJan Kara item->lri_slot); 12792205363dSJan Kara if (ret < 0) 12802205363dSJan Kara mlog_errno(ret); 12812205363dSJan Kara /* Recovery info is already freed now */ 12822205363dSJan Kara } 12832205363dSJan Kara 1284ccd979bdSMark Fasheh kfree(item); 1285ccd979bdSMark Fasheh } 1286ccd979bdSMark Fasheh 1287b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1288ccd979bdSMark Fasheh } 1289ccd979bdSMark Fasheh 1290ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1291ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1292ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1293ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1294ccd979bdSMark Fasheh int slot_num, 1295ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 12962205363dSJan Kara struct ocfs2_dinode *tl_dinode, 1297ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 1298ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1299ccd979bdSMark Fasheh { 1300ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1301ccd979bdSMark Fasheh 1302afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1303ccd979bdSMark Fasheh if (!item) { 1304ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1305ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1306ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1307ccd979bdSMark Fasheh kfree(la_dinode); 1308ccd979bdSMark Fasheh kfree(tl_dinode); 1309ccd979bdSMark Fasheh 13102205363dSJan Kara if (qrec) 13112205363dSJan Kara ocfs2_free_quota_recovery(qrec); 13122205363dSJan Kara 1313ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1314ccd979bdSMark Fasheh return; 1315ccd979bdSMark Fasheh } 1316ccd979bdSMark Fasheh 1317ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1318ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1319ccd979bdSMark Fasheh item->lri_slot = slot_num; 1320ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 13212205363dSJan Kara item->lri_qrec = qrec; 1322ed460cffSJoseph Qi item->lri_orphan_reco_type = orphan_reco_type; 1323ccd979bdSMark Fasheh 1324ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1325ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 132635ddf78eSjiangyiwen queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); 1327ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1328ccd979bdSMark Fasheh } 1329ccd979bdSMark Fasheh 1330ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 13319140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1332ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1333ccd979bdSMark Fasheh { 1334ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1335ccd979bdSMark Fasheh 133610b3dd76SSunil Mushran if (ocfs2_is_hard_readonly(osb)) 133710b3dd76SSunil Mushran return; 133810b3dd76SSunil Mushran 13399140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 13409140db04SSrinivas Eeda * that */ 13419140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 1342ed460cffSJoseph Qi osb->local_alloc_copy, NULL, NULL, 1343ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 1344ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1345ccd979bdSMark Fasheh 1346ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 13479140db04SSrinivas Eeda 13489140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 13499140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1350ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 13519140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1352ccd979bdSMark Fasheh } 1353ccd979bdSMark Fasheh 13542205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 13552205363dSJan Kara { 13562205363dSJan Kara if (osb->quota_rec) { 13572205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 13582205363dSJan Kara osb->slot_num, 13592205363dSJan Kara NULL, 13602205363dSJan Kara NULL, 1361ed460cffSJoseph Qi osb->quota_rec, 1362ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 13632205363dSJan Kara osb->quota_rec = NULL; 13642205363dSJan Kara } 13652205363dSJan Kara } 13662205363dSJan Kara 1367ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1368ccd979bdSMark Fasheh { 13692205363dSJan Kara int status, node_num, slot_num; 1370ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1371553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 13722205363dSJan Kara int *rm_quota = NULL; 13732205363dSJan Kara int rm_quota_used = 0, i; 13742205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1375ccd979bdSMark Fasheh 137621158ca8SGuozhonghua /* Whether the quota supported. */ 137721158ca8SGuozhonghua int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 137821158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_USRQUOTA) 137921158ca8SGuozhonghua || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 138021158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); 138121158ca8SGuozhonghua 1382ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1383ccd979bdSMark Fasheh if (status < 0) { 1384ccd979bdSMark Fasheh goto bail; 1385ccd979bdSMark Fasheh } 1386ccd979bdSMark Fasheh 138721158ca8SGuozhonghua if (quota_enabled) { 13886396bb22SKees Cook rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); 13892205363dSJan Kara if (!rm_quota) { 13902205363dSJan Kara status = -ENOMEM; 13912205363dSJan Kara goto bail; 13922205363dSJan Kara } 139321158ca8SGuozhonghua } 1394ccd979bdSMark Fasheh restart: 1395ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1396ccd979bdSMark Fasheh if (status < 0) { 1397ccd979bdSMark Fasheh mlog_errno(status); 1398ccd979bdSMark Fasheh goto bail; 1399ccd979bdSMark Fasheh } 1400ccd979bdSMark Fasheh 14019140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 14029140db04SSrinivas Eeda if (status < 0) 14039140db04SSrinivas Eeda mlog_errno(status); 14049140db04SSrinivas Eeda 14059140db04SSrinivas Eeda /* queue recovery for our own slot */ 14069140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1407ed460cffSJoseph Qi NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); 14089140db04SSrinivas Eeda 1409553abd04SJoel Becker spin_lock(&osb->osb_lock); 1410553abd04SJoel Becker while (rm->rm_used) { 1411553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1412553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1413553abd04SJoel Becker node_num = rm->rm_entries[0]; 1414553abd04SJoel Becker spin_unlock(&osb->osb_lock); 14152205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1416b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 14172205363dSJan Kara if (slot_num == -ENOENT) { 14182205363dSJan Kara status = 0; 14192205363dSJan Kara goto skip_recovery; 14202205363dSJan Kara } 1421ccd979bdSMark Fasheh 14222205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 14232205363dSJan Kara * immediately because we have to obtain cluster locks from 14242205363dSJan Kara * quota files and we also don't want to just skip it because 14252205363dSJan Kara * then quota usage would be out of sync until some node takes 14262205363dSJan Kara * the slot. So we remember which nodes need quota recovery 14272205363dSJan Kara * and when everything else is done, we recover quotas. */ 142821158ca8SGuozhonghua if (quota_enabled) { 142921158ca8SGuozhonghua for (i = 0; i < rm_quota_used 143021158ca8SGuozhonghua && rm_quota[i] != slot_num; i++) 143121158ca8SGuozhonghua ; 143221158ca8SGuozhonghua 14332205363dSJan Kara if (i == rm_quota_used) 14342205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 143521158ca8SGuozhonghua } 14362205363dSJan Kara 14372205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 14382205363dSJan Kara skip_recovery: 1439553abd04SJoel Becker if (!status) { 1440553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1441553abd04SJoel Becker } else { 1442ccd979bdSMark Fasheh mlog(ML_ERROR, 1443ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1444ccd979bdSMark Fasheh status, node_num, 1445ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1446ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1447ccd979bdSMark Fasheh } 1448ccd979bdSMark Fasheh 1449553abd04SJoel Becker spin_lock(&osb->osb_lock); 1450ccd979bdSMark Fasheh } 1451553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1452b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1453553abd04SJoel Becker 1454539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1455539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1456539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1457539d8264SSunil Mushran if (status < 0) 1458539d8264SSunil Mushran mlog_errno(status); 1459539d8264SSunil Mushran 14602205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 14612205363dSJan Kara * superblock lock so that no one can start using the slot (and crash) 14622205363dSJan Kara * before we recover it */ 146321158ca8SGuozhonghua if (quota_enabled) { 14642205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 14652205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 14662205363dSJan Kara if (IS_ERR(qrec)) { 14672205363dSJan Kara status = PTR_ERR(qrec); 14682205363dSJan Kara mlog_errno(status); 14692205363dSJan Kara continue; 14702205363dSJan Kara } 147121158ca8SGuozhonghua ocfs2_queue_recovery_completion(osb->journal, 147221158ca8SGuozhonghua rm_quota[i], 1473ed460cffSJoseph Qi NULL, NULL, qrec, 1474ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 14752205363dSJan Kara } 147621158ca8SGuozhonghua } 14772205363dSJan Kara 1478ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1479ccd979bdSMark Fasheh 14809140db04SSrinivas Eeda /* queue recovery for offline slots */ 1481ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1482ccd979bdSMark Fasheh 1483ccd979bdSMark Fasheh bail: 1484c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1485553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1486c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1487ccd979bdSMark Fasheh goto restart; 1488ccd979bdSMark Fasheh } 1489ccd979bdSMark Fasheh 14909140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1491ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1492ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1493ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1494ccd979bdSMark Fasheh 1495c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1496ccd979bdSMark Fasheh 149721158ca8SGuozhonghua if (quota_enabled) 14982205363dSJan Kara kfree(rm_quota); 14992205363dSJan Kara 1500ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1501ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1502ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1503ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1504ccd979bdSMark Fasheh } 1505ccd979bdSMark Fasheh 1506ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1507ccd979bdSMark Fasheh { 1508c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1509b4107950STao Ma 1510b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1511b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1512b4107950STao Ma osb->disable_recovery ? 1513b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1514b4107950STao Ma 1515ccd979bdSMark Fasheh if (osb->disable_recovery) 1516ccd979bdSMark Fasheh goto out; 1517ccd979bdSMark Fasheh 1518ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1519ccd979bdSMark Fasheh goto out; 1520ccd979bdSMark Fasheh 1521ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 15225afc44e2SJoseph Qi "ocfs2rec-%s", osb->uuid_str); 1523ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1524ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1525ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1526ccd979bdSMark Fasheh } 1527ccd979bdSMark Fasheh 1528ccd979bdSMark Fasheh out: 1529c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1530ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1531ccd979bdSMark Fasheh } 1532ccd979bdSMark Fasheh 1533539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1534539d8264SSunil Mushran int slot_num, 1535539d8264SSunil Mushran struct buffer_head **bh, 1536539d8264SSunil Mushran struct inode **ret_inode) 1537539d8264SSunil Mushran { 1538539d8264SSunil Mushran int status = -EACCES; 1539539d8264SSunil Mushran struct inode *inode = NULL; 1540539d8264SSunil Mushran 1541539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1542539d8264SSunil Mushran 1543539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1544539d8264SSunil Mushran slot_num); 1545539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1546539d8264SSunil Mushran mlog_errno(status); 1547539d8264SSunil Mushran goto bail; 1548539d8264SSunil Mushran } 1549539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1550539d8264SSunil Mushran 1551b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1552539d8264SSunil Mushran if (status < 0) { 1553539d8264SSunil Mushran mlog_errno(status); 1554539d8264SSunil Mushran goto bail; 1555539d8264SSunil Mushran } 1556539d8264SSunil Mushran 1557539d8264SSunil Mushran status = 0; 1558539d8264SSunil Mushran 1559539d8264SSunil Mushran bail: 1560539d8264SSunil Mushran if (inode) { 1561539d8264SSunil Mushran if (status || !ret_inode) 1562539d8264SSunil Mushran iput(inode); 1563539d8264SSunil Mushran else 1564539d8264SSunil Mushran *ret_inode = inode; 1565539d8264SSunil Mushran } 1566539d8264SSunil Mushran return status; 1567539d8264SSunil Mushran } 1568539d8264SSunil Mushran 1569ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1570ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1571ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1572ccd979bdSMark Fasheh int node_num, 1573ccd979bdSMark Fasheh int slot_num) 1574ccd979bdSMark Fasheh { 1575ccd979bdSMark Fasheh int status; 1576ccd979bdSMark Fasheh int got_lock = 0; 1577ccd979bdSMark Fasheh unsigned int flags; 1578ccd979bdSMark Fasheh struct inode *inode = NULL; 1579ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1580ccd979bdSMark Fasheh journal_t *journal = NULL; 1581ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1582539d8264SSunil Mushran u32 slot_reco_gen; 1583ccd979bdSMark Fasheh 1584539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1585539d8264SSunil Mushran if (status) { 1586ccd979bdSMark Fasheh mlog_errno(status); 1587ccd979bdSMark Fasheh goto done; 1588ccd979bdSMark Fasheh } 1589539d8264SSunil Mushran 1590539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1591539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1592539d8264SSunil Mushran brelse(bh); 1593539d8264SSunil Mushran bh = NULL; 1594539d8264SSunil Mushran 1595539d8264SSunil Mushran /* 1596539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1597539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1598539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1599539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1600539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1601539d8264SSunil Mushran * If not, it needs recovery. 1602539d8264SSunil Mushran */ 1603539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1604b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1605539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1606539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1607539d8264SSunil Mushran status = -EBUSY; 1608ccd979bdSMark Fasheh goto done; 1609ccd979bdSMark Fasheh } 1610539d8264SSunil Mushran 1611539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1612ccd979bdSMark Fasheh 1613e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1614ccd979bdSMark Fasheh if (status < 0) { 1615b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1616ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1617ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1618ccd979bdSMark Fasheh goto done; 1619ccd979bdSMark Fasheh } 1620ccd979bdSMark Fasheh got_lock = 1; 1621ccd979bdSMark Fasheh 1622ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1623ccd979bdSMark Fasheh 1624ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1625539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1626ccd979bdSMark Fasheh 1627ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1628b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1629539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1630539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1631ccd979bdSMark Fasheh goto done; 1632ccd979bdSMark Fasheh } 1633ccd979bdSMark Fasheh 16349140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 16359140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 16369140db04SSrinivas Eeda 1637619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1638619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1639619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1640ccd979bdSMark Fasheh 1641ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1642ccd979bdSMark Fasheh 1643ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1644ccd979bdSMark Fasheh if (status < 0) { 1645ccd979bdSMark Fasheh mlog_errno(status); 1646ccd979bdSMark Fasheh goto done; 1647ccd979bdSMark Fasheh } 1648ccd979bdSMark Fasheh 16492b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1650ccd979bdSMark Fasheh if (journal == NULL) { 1651ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1652ccd979bdSMark Fasheh status = -EIO; 1653ccd979bdSMark Fasheh goto done; 1654ccd979bdSMark Fasheh } 1655ccd979bdSMark Fasheh 16562b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1657ccd979bdSMark Fasheh if (status < 0) { 1658ccd979bdSMark Fasheh mlog_errno(status); 1659ccd979bdSMark Fasheh if (!igrab(inode)) 1660ccd979bdSMark Fasheh BUG(); 16612b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1662ccd979bdSMark Fasheh goto done; 1663ccd979bdSMark Fasheh } 1664ccd979bdSMark Fasheh 1665ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1666ccd979bdSMark Fasheh 1667ccd979bdSMark Fasheh /* wipe the journal */ 16682b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 1669*01d5d965SLeah Rumancik status = jbd2_journal_flush(journal, 0); 16702b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1671ccd979bdSMark Fasheh if (status < 0) 1672ccd979bdSMark Fasheh mlog_errno(status); 1673ccd979bdSMark Fasheh 1674ccd979bdSMark Fasheh /* This will mark the node clean */ 1675ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1676ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1677ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1678ccd979bdSMark Fasheh 1679539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1680539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1681539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1682539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1683539d8264SSunil Mushran 168413723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 16858cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1686ccd979bdSMark Fasheh if (status < 0) 1687ccd979bdSMark Fasheh mlog_errno(status); 1688ccd979bdSMark Fasheh 1689ccd979bdSMark Fasheh if (!igrab(inode)) 1690ccd979bdSMark Fasheh BUG(); 1691ccd979bdSMark Fasheh 16922b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1693ccd979bdSMark Fasheh 1694619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1695619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1696619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1697ccd979bdSMark Fasheh done: 1698ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1699ccd979bdSMark Fasheh if (got_lock) 1700e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1701ccd979bdSMark Fasheh 1702ccd979bdSMark Fasheh iput(inode); 1703ccd979bdSMark Fasheh brelse(bh); 1704ccd979bdSMark Fasheh 1705ccd979bdSMark Fasheh return status; 1706ccd979bdSMark Fasheh } 1707ccd979bdSMark Fasheh 1708ccd979bdSMark Fasheh /* 1709ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1710ccd979bdSMark Fasheh * - Replay it's journal 1711ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1712ccd979bdSMark Fasheh * - Stamp a clean truncate log 1713ccd979bdSMark Fasheh * - Mark the node clean 1714ccd979bdSMark Fasheh * 1715ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1716ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1717ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1718ccd979bdSMark Fasheh * far less concerning. 1719ccd979bdSMark Fasheh */ 1720ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 17212205363dSJan Kara int node_num, int slot_num) 1722ccd979bdSMark Fasheh { 1723ccd979bdSMark Fasheh int status = 0; 1724ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1725ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1726ccd979bdSMark Fasheh 1727b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1728ccd979bdSMark Fasheh 1729ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1730ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1731ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1732ccd979bdSMark Fasheh 1733ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1734ccd979bdSMark Fasheh if (status < 0) { 1735539d8264SSunil Mushran if (status == -EBUSY) { 1736b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1737539d8264SSunil Mushran status = 0; 1738539d8264SSunil Mushran goto done; 1739539d8264SSunil Mushran } 1740ccd979bdSMark Fasheh mlog_errno(status); 1741ccd979bdSMark Fasheh goto done; 1742ccd979bdSMark Fasheh } 1743ccd979bdSMark Fasheh 1744ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1745ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1746ccd979bdSMark Fasheh if (status < 0) { 1747ccd979bdSMark Fasheh mlog_errno(status); 1748ccd979bdSMark Fasheh goto done; 1749ccd979bdSMark Fasheh } 1750ccd979bdSMark Fasheh 1751ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1752ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1753ccd979bdSMark Fasheh * recovery. */ 1754ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1755ccd979bdSMark Fasheh if (status < 0) 1756ccd979bdSMark Fasheh mlog_errno(status); 1757ccd979bdSMark Fasheh 1758ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1759ccd979bdSMark Fasheh * harmful place to get an error... */ 17608e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1761ccd979bdSMark Fasheh if (status < 0) 1762ccd979bdSMark Fasheh mlog_errno(status); 1763ccd979bdSMark Fasheh 1764ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1765ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1766ed460cffSJoseph Qi tl_copy, NULL, ORPHAN_NEED_TRUNCATE); 1767ccd979bdSMark Fasheh 1768ccd979bdSMark Fasheh status = 0; 1769ccd979bdSMark Fasheh done: 1770ccd979bdSMark Fasheh 1771ccd979bdSMark Fasheh return status; 1772ccd979bdSMark Fasheh } 1773ccd979bdSMark Fasheh 1774ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1775ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1776ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1777ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1778ccd979bdSMark Fasheh int slot_num) 1779ccd979bdSMark Fasheh { 1780ccd979bdSMark Fasheh int status, flags; 1781ccd979bdSMark Fasheh struct inode *inode = NULL; 1782ccd979bdSMark Fasheh 1783ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1784ccd979bdSMark Fasheh slot_num); 1785ccd979bdSMark Fasheh if (inode == NULL) { 1786ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1787ccd979bdSMark Fasheh status = -EACCES; 1788ccd979bdSMark Fasheh goto bail; 1789ccd979bdSMark Fasheh } 1790ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1791ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1792ccd979bdSMark Fasheh iput(inode); 1793ccd979bdSMark Fasheh inode = NULL; 1794ccd979bdSMark Fasheh status = -EACCES; 1795ccd979bdSMark Fasheh goto bail; 1796ccd979bdSMark Fasheh } 1797ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1798ccd979bdSMark Fasheh 1799ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1800e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1801ccd979bdSMark Fasheh if (status < 0) { 1802ccd979bdSMark Fasheh if (status != -EAGAIN) 1803ccd979bdSMark Fasheh mlog_errno(status); 1804ccd979bdSMark Fasheh goto bail; 1805ccd979bdSMark Fasheh } 1806ccd979bdSMark Fasheh 1807e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1808ccd979bdSMark Fasheh bail: 1809ccd979bdSMark Fasheh iput(inode); 1810ccd979bdSMark Fasheh 1811ccd979bdSMark Fasheh return status; 1812ccd979bdSMark Fasheh } 1813ccd979bdSMark Fasheh 1814ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1815ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1816ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1817ccd979bdSMark Fasheh { 1818d85b20e4SJoel Becker unsigned int node_num; 1819d85b20e4SJoel Becker int status, i; 1820a1af7d15SMark Fasheh u32 gen; 1821539d8264SSunil Mushran struct buffer_head *bh = NULL; 1822539d8264SSunil Mushran struct ocfs2_dinode *di; 1823ccd979bdSMark Fasheh 1824ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1825ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1826ccd979bdSMark Fasheh 1827d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1828539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1829539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1830539d8264SSunil Mushran if (status) { 1831539d8264SSunil Mushran mlog_errno(status); 1832539d8264SSunil Mushran goto bail; 1833539d8264SSunil Mushran } 1834539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1835a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1836539d8264SSunil Mushran brelse(bh); 1837539d8264SSunil Mushran bh = NULL; 1838539d8264SSunil Mushran 1839a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1840a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1841a1af7d15SMark Fasheh 1842b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1843539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1844539d8264SSunil Mushran 1845a1af7d15SMark Fasheh if (i == osb->slot_num) { 1846a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1847ccd979bdSMark Fasheh continue; 1848a1af7d15SMark Fasheh } 1849d85b20e4SJoel Becker 1850d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1851a1af7d15SMark Fasheh if (status == -ENOENT) { 1852a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1853ccd979bdSMark Fasheh continue; 1854a1af7d15SMark Fasheh } 1855ccd979bdSMark Fasheh 1856a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1857a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1858ccd979bdSMark Fasheh continue; 1859a1af7d15SMark Fasheh } 1860d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1861ccd979bdSMark Fasheh 1862ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1863ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1864ccd979bdSMark Fasheh * file here to test if he's alive. */ 1865ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1866ccd979bdSMark Fasheh if (!status) { 1867ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1868ccd979bdSMark Fasheh * the recovery thread can't race us on 1869ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1870ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1871ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1872ccd979bdSMark Fasheh mlog_errno(status); 1873ccd979bdSMark Fasheh goto bail; 1874ccd979bdSMark Fasheh } 1875ccd979bdSMark Fasheh } 1876ccd979bdSMark Fasheh 1877ccd979bdSMark Fasheh status = 0; 1878ccd979bdSMark Fasheh bail: 1879ccd979bdSMark Fasheh return status; 1880ccd979bdSMark Fasheh } 1881ccd979bdSMark Fasheh 188283273932SSrinivas Eeda /* 188383273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 188483273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 188583273932SSrinivas Eeda * same time. 188683273932SSrinivas Eeda */ 188783273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 188883273932SSrinivas Eeda { 188983273932SSrinivas Eeda unsigned long time; 189083273932SSrinivas Eeda 189183273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 189283273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 189383273932SSrinivas Eeda return msecs_to_jiffies(time); 189483273932SSrinivas Eeda } 189583273932SSrinivas Eeda 189683273932SSrinivas Eeda /* 189783273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 189883273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 189983273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 190083273932SSrinivas Eeda * 1901a035bff6SSunil Mushran * It scans all slots, even ones that are in use. It does so to handle the 1902a035bff6SSunil Mushran * case described below: 1903a035bff6SSunil Mushran * 1904a035bff6SSunil Mushran * Node 1 has an inode it was using. The dentry went away due to memory 1905a035bff6SSunil Mushran * pressure. Node 1 closes the inode, but it's on the free list. The node 1906a035bff6SSunil Mushran * has the open lock. 1907a035bff6SSunil Mushran * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1908a035bff6SSunil Mushran * but node 1 has no dentry and doesn't get the message. It trylocks the 1909a035bff6SSunil Mushran * open lock, sees that another node has a PR, and does nothing. 1910a035bff6SSunil Mushran * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1911a035bff6SSunil Mushran * open lock, sees the PR still, and does nothing. 1912a035bff6SSunil Mushran * Basically, we have to trigger an orphan iput on node 1. The only way 1913a035bff6SSunil Mushran * for this to happen is if node 1 runs node 2's orphan dir. 1914a035bff6SSunil Mushran * 191583273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 191683273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 191783273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 191883273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 191983273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 192083273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 192183273932SSrinivas Eeda * sequence number in the LVB. 192283273932SSrinivas Eeda */ 1923b519ea6dSJoseph Qi static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 192483273932SSrinivas Eeda { 192583273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 192683273932SSrinivas Eeda int status, i; 192783273932SSrinivas Eeda u32 seqno = 0; 192883273932SSrinivas Eeda 192983273932SSrinivas Eeda os = &osb->osb_orphan_scan; 193083273932SSrinivas Eeda 1931692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1932692684e1SSunil Mushran goto out; 1933692684e1SSunil Mushran 1934b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1935b4107950STao Ma atomic_read(&os->os_state)); 1936b4107950STao Ma 1937df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 193883273932SSrinivas Eeda if (status < 0) { 193983273932SSrinivas Eeda if (status != -EAGAIN) 194083273932SSrinivas Eeda mlog_errno(status); 194183273932SSrinivas Eeda goto out; 194283273932SSrinivas Eeda } 194383273932SSrinivas Eeda 1944692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1945692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1946692684e1SSunil Mushran goto unlock; 1947692684e1SSunil Mushran 194883273932SSrinivas Eeda if (os->os_seqno != seqno) { 194983273932SSrinivas Eeda os->os_seqno = seqno; 195083273932SSrinivas Eeda goto unlock; 195183273932SSrinivas Eeda } 195283273932SSrinivas Eeda 195383273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 195483273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1955ed460cffSJoseph Qi NULL, ORPHAN_NO_NEED_TRUNCATE); 195683273932SSrinivas Eeda /* 195783273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 195883273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 195983273932SSrinivas Eeda */ 196083273932SSrinivas Eeda seqno++; 196115633a22SSrinivas Eeda os->os_count++; 1962395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 196383273932SSrinivas Eeda unlock: 1964df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 196583273932SSrinivas Eeda out: 1966b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1967b4107950STao Ma atomic_read(&os->os_state)); 196883273932SSrinivas Eeda return; 196983273932SSrinivas Eeda } 197083273932SSrinivas Eeda 197183273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 1972b519ea6dSJoseph Qi static void ocfs2_orphan_scan_work(struct work_struct *work) 197383273932SSrinivas Eeda { 197483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 197583273932SSrinivas Eeda struct ocfs2_super *osb; 197683273932SSrinivas Eeda 197783273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 197883273932SSrinivas Eeda os_orphan_scan_work.work); 197983273932SSrinivas Eeda osb = os->os_osb; 198083273932SSrinivas Eeda 198183273932SSrinivas Eeda mutex_lock(&os->os_lock); 198283273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 1983692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 198435ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 198583273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 198683273932SSrinivas Eeda mutex_unlock(&os->os_lock); 198783273932SSrinivas Eeda } 198883273932SSrinivas Eeda 198983273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 199083273932SSrinivas Eeda { 199183273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 199283273932SSrinivas Eeda 199383273932SSrinivas Eeda os = &osb->osb_orphan_scan; 1994df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1995692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 199683273932SSrinivas Eeda mutex_lock(&os->os_lock); 199783273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 199883273932SSrinivas Eeda mutex_unlock(&os->os_lock); 199983273932SSrinivas Eeda } 2000df152c24SSunil Mushran } 200183273932SSrinivas Eeda 2002df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 200383273932SSrinivas Eeda { 200483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 200583273932SSrinivas Eeda 200683273932SSrinivas Eeda os = &osb->osb_orphan_scan; 200783273932SSrinivas Eeda os->os_osb = osb; 200815633a22SSrinivas Eeda os->os_count = 0; 20093211949fSSunil Mushran os->os_seqno = 0; 201083273932SSrinivas Eeda mutex_init(&os->os_lock); 2011df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 20128b712cd5SJeff Mahoney } 201383273932SSrinivas Eeda 20148b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 20158b712cd5SJeff Mahoney { 20168b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 20178b712cd5SJeff Mahoney 20188b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 2019395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 2020df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 2021df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 2022df152c24SSunil Mushran else { 2023df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 202435ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 202583273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 2026df152c24SSunil Mushran } 202783273932SSrinivas Eeda } 202883273932SSrinivas Eeda 20295eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 20303704412bSAl Viro struct dir_context ctx; 20315eae5b96SMark Fasheh struct inode *head; 20325eae5b96SMark Fasheh struct ocfs2_super *osb; 203330edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 20345eae5b96SMark Fasheh }; 20355eae5b96SMark Fasheh 2036ac7576f4SMiklos Szeredi static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, 2037ac7576f4SMiklos Szeredi int name_len, loff_t pos, u64 ino, 2038ac7576f4SMiklos Szeredi unsigned type) 20395eae5b96SMark Fasheh { 2040ac7576f4SMiklos Szeredi struct ocfs2_orphan_filldir_priv *p = 2041ac7576f4SMiklos Szeredi container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); 20425eae5b96SMark Fasheh struct inode *iter; 20435eae5b96SMark Fasheh 20445eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 20455eae5b96SMark Fasheh return 0; 20465eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 20475eae5b96SMark Fasheh return 0; 20485eae5b96SMark Fasheh 204930edc43cSJoseph Qi /* do not include dio entry in case of orphan scan */ 205030edc43cSJoseph Qi if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && 205130edc43cSJoseph Qi (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 205230edc43cSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN))) 205330edc43cSJoseph Qi return 0; 205430edc43cSJoseph Qi 20555eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 20565eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 20575fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 20585eae5b96SMark Fasheh if (IS_ERR(iter)) 20595eae5b96SMark Fasheh return 0; 20605eae5b96SMark Fasheh 206193d911fcSJoseph Qi if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 206293d911fcSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN)) 206393d911fcSJoseph Qi OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; 206493d911fcSJoseph Qi 2065ed460cffSJoseph Qi /* Skip inodes which are already added to recover list, since dio may 2066ed460cffSJoseph Qi * happen concurrently with unlink/rename */ 2067ed460cffSJoseph Qi if (OCFS2_I(iter)->ip_next_orphan) { 2068ed460cffSJoseph Qi iput(iter); 2069ed460cffSJoseph Qi return 0; 2070ed460cffSJoseph Qi } 2071ed460cffSJoseph Qi 2072b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 20735eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 20745eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 20755eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 20765eae5b96SMark Fasheh p->head = iter; 20775eae5b96SMark Fasheh 20785eae5b96SMark Fasheh return 0; 20795eae5b96SMark Fasheh } 20805eae5b96SMark Fasheh 2081b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2082b4df6ed8SMark Fasheh int slot, 208330edc43cSJoseph Qi struct inode **head, 208430edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2085ccd979bdSMark Fasheh { 2086b4df6ed8SMark Fasheh int status; 2087ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 20883704412bSAl Viro struct ocfs2_orphan_filldir_priv priv = { 20893704412bSAl Viro .ctx.actor = ocfs2_orphan_filldir, 20903704412bSAl Viro .osb = osb, 209130edc43cSJoseph Qi .head = *head, 209230edc43cSJoseph Qi .orphan_reco_type = orphan_reco_type 20933704412bSAl Viro }; 2094ccd979bdSMark Fasheh 2095ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2096ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 2097ccd979bdSMark Fasheh slot); 2098ccd979bdSMark Fasheh if (!orphan_dir_inode) { 2099ccd979bdSMark Fasheh status = -ENOENT; 2100ccd979bdSMark Fasheh mlog_errno(status); 2101b4df6ed8SMark Fasheh return status; 2102ccd979bdSMark Fasheh } 2103ccd979bdSMark Fasheh 21045955102cSAl Viro inode_lock(orphan_dir_inode); 2105e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2106ccd979bdSMark Fasheh if (status < 0) { 2107ccd979bdSMark Fasheh mlog_errno(status); 2108ccd979bdSMark Fasheh goto out; 2109ccd979bdSMark Fasheh } 2110ccd979bdSMark Fasheh 21113704412bSAl Viro status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 21125eae5b96SMark Fasheh if (status) { 2113ccd979bdSMark Fasheh mlog_errno(status); 2114a86370fbSMark Fasheh goto out_cluster; 2115ccd979bdSMark Fasheh } 2116ccd979bdSMark Fasheh 21175eae5b96SMark Fasheh *head = priv.head; 2118ccd979bdSMark Fasheh 2119a86370fbSMark Fasheh out_cluster: 2120e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 2121b4df6ed8SMark Fasheh out: 21225955102cSAl Viro inode_unlock(orphan_dir_inode); 2123ccd979bdSMark Fasheh iput(orphan_dir_inode); 2124b4df6ed8SMark Fasheh return status; 2125b4df6ed8SMark Fasheh } 2126b4df6ed8SMark Fasheh 2127b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2128b4df6ed8SMark Fasheh int slot) 2129b4df6ed8SMark Fasheh { 2130b4df6ed8SMark Fasheh int ret; 2131b4df6ed8SMark Fasheh 2132b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2133b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2134b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2135b4df6ed8SMark Fasheh return ret; 2136b4df6ed8SMark Fasheh } 2137b4df6ed8SMark Fasheh 2138b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2139b4df6ed8SMark Fasheh int slot) 2140b4df6ed8SMark Fasheh { 2141b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2142b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2143b4df6ed8SMark Fasheh * know to quit early. */ 2144b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2145b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2146b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2147b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2148b4df6ed8SMark Fasheh * them. */ 2149b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2150b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2151b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2152b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2153b4df6ed8SMark Fasheh } 2154b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2155b4df6ed8SMark Fasheh } 2156b4df6ed8SMark Fasheh 2157b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2158b4df6ed8SMark Fasheh int slot) 2159b4df6ed8SMark Fasheh { 2160b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2161b4df6ed8SMark Fasheh } 2162b4df6ed8SMark Fasheh 2163b4df6ed8SMark Fasheh /* 2164b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2165b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2166b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2167b4df6ed8SMark Fasheh * (most) of the rest of the work. 2168b4df6ed8SMark Fasheh * 2169b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2170b4df6ed8SMark Fasheh * couple of extra considerations. 2171b4df6ed8SMark Fasheh * 2172b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2173b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2174b4df6ed8SMark Fasheh * an invalid inode. 2175b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2176b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2177b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2178b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2179b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2180b4df6ed8SMark Fasheh */ 2181b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2182ed460cffSJoseph Qi int slot, 2183ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2184b4df6ed8SMark Fasheh { 2185b4df6ed8SMark Fasheh int ret = 0; 2186b4df6ed8SMark Fasheh struct inode *inode = NULL; 2187b4df6ed8SMark Fasheh struct inode *iter; 2188b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2189cf1776a9SJoseph Qi struct buffer_head *di_bh = NULL; 2190cf1776a9SJoseph Qi struct ocfs2_dinode *di = NULL; 2191b4df6ed8SMark Fasheh 2192b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2193b4df6ed8SMark Fasheh 2194b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 219530edc43cSJoseph Qi ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); 2196b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2197b4df6ed8SMark Fasheh 2198b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2199b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2200b4df6ed8SMark Fasheh if (ret) 2201b4df6ed8SMark Fasheh mlog_errno(ret); 2202ccd979bdSMark Fasheh 2203ccd979bdSMark Fasheh while (inode) { 2204ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2205b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2206b4107950STao Ma (unsigned long long)oi->ip_blkno); 2207ccd979bdSMark Fasheh 2208ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2209ed460cffSJoseph Qi oi->ip_next_orphan = NULL; 2210ccd979bdSMark Fasheh 221193d911fcSJoseph Qi if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { 22125955102cSAl Viro inode_lock(inode); 2213cf1776a9SJoseph Qi ret = ocfs2_rw_lock(inode, 1); 2214cf1776a9SJoseph Qi if (ret < 0) { 2215cf1776a9SJoseph Qi mlog_errno(ret); 221693d911fcSJoseph Qi goto unlock_mutex; 2217cf1776a9SJoseph Qi } 2218ed460cffSJoseph Qi /* 2219ed460cffSJoseph Qi * We need to take and drop the inode lock to 2220ed460cffSJoseph Qi * force read inode from disk. 2221ed460cffSJoseph Qi */ 2222cf1776a9SJoseph Qi ret = ocfs2_inode_lock(inode, &di_bh, 1); 2223ed460cffSJoseph Qi if (ret) { 2224ed460cffSJoseph Qi mlog_errno(ret); 2225cf1776a9SJoseph Qi goto unlock_rw; 2226ed460cffSJoseph Qi } 2227cf1776a9SJoseph Qi 2228cf1776a9SJoseph Qi di = (struct ocfs2_dinode *)di_bh->b_data; 2229ed460cffSJoseph Qi 223093d911fcSJoseph Qi if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { 2231ed460cffSJoseph Qi ret = ocfs2_truncate_file(inode, di_bh, 2232ed460cffSJoseph Qi i_size_read(inode)); 2233ed460cffSJoseph Qi if (ret < 0) { 2234ed460cffSJoseph Qi if (ret != -ENOSPC) 2235ed460cffSJoseph Qi mlog_errno(ret); 2236cf1776a9SJoseph Qi goto unlock_inode; 2237ed460cffSJoseph Qi } 2238ed460cffSJoseph Qi 223993d911fcSJoseph Qi ret = ocfs2_del_inode_from_orphan(osb, inode, 224093d911fcSJoseph Qi di_bh, 0, 0); 2241ed460cffSJoseph Qi if (ret) 2242ed460cffSJoseph Qi mlog_errno(ret); 224393d911fcSJoseph Qi } 2244cf1776a9SJoseph Qi unlock_inode: 2245cf1776a9SJoseph Qi ocfs2_inode_unlock(inode, 1); 2246512f62acSJoseph Qi brelse(di_bh); 2247512f62acSJoseph Qi di_bh = NULL; 2248cf1776a9SJoseph Qi unlock_rw: 2249cf1776a9SJoseph Qi ocfs2_rw_unlock(inode, 1); 225093d911fcSJoseph Qi unlock_mutex: 22515955102cSAl Viro inode_unlock(inode); 225293d911fcSJoseph Qi 225393d911fcSJoseph Qi /* clear dio flag in ocfs2_inode_info */ 225493d911fcSJoseph Qi oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; 225593d911fcSJoseph Qi } else { 225693d911fcSJoseph Qi spin_lock(&oi->ip_lock); 225793d911fcSJoseph Qi /* Set the proper information to get us going into 225893d911fcSJoseph Qi * ocfs2_delete_inode. */ 225993d911fcSJoseph Qi oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 226093d911fcSJoseph Qi spin_unlock(&oi->ip_lock); 226193d911fcSJoseph Qi } 226293d911fcSJoseph Qi 2263ccd979bdSMark Fasheh iput(inode); 2264ccd979bdSMark Fasheh inode = iter; 2265ccd979bdSMark Fasheh } 2266ccd979bdSMark Fasheh 2267b4df6ed8SMark Fasheh return ret; 2268ccd979bdSMark Fasheh } 2269ccd979bdSMark Fasheh 227019ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2271ccd979bdSMark Fasheh { 2272ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2273ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2274ccd979bdSMark Fasheh * or DISABLED. */ 2275ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 227619ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 227719ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2278ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2279ccd979bdSMark Fasheh 2280ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2281ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2282ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2283ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2284b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2285ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2286ccd979bdSMark Fasheh return -EBUSY; 2287ccd979bdSMark Fasheh } 2288ccd979bdSMark Fasheh 2289ccd979bdSMark Fasheh return 0; 2290ccd979bdSMark Fasheh } 2291ccd979bdSMark Fasheh 2292ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2293ccd979bdSMark Fasheh { 2294ccd979bdSMark Fasheh int status; 2295ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2296ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2297ccd979bdSMark Fasheh 2298ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2299ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2300ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2301ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2302ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2303ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2304ccd979bdSMark Fasheh 2305745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2306ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2307745ae8baSMark Fasheh || kthread_should_stop()); 2308ccd979bdSMark Fasheh 2309ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 231055b465b6SJoseph Qi if (status < 0) { 231155b465b6SJoseph Qi static unsigned long abort_warn_time; 231255b465b6SJoseph Qi 231355b465b6SJoseph Qi /* Warn about this once per minute */ 231455b465b6SJoseph Qi if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 231555b465b6SJoseph Qi mlog(ML_ERROR, "status = %d, journal is " 231655b465b6SJoseph Qi "already aborted.\n", status); 231755b465b6SJoseph Qi /* 231855b465b6SJoseph Qi * After ocfs2_commit_cache() fails, j_num_trans has a 231955b465b6SJoseph Qi * non-zero value. Sleep here to avoid a busy-wait 232055b465b6SJoseph Qi * loop. 232155b465b6SJoseph Qi */ 232255b465b6SJoseph Qi msleep_interruptible(1000); 232355b465b6SJoseph Qi } 2324ccd979bdSMark Fasheh 2325ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2326ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2327ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2328ccd979bdSMark Fasheh "shutdown\n", 2329ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2330ccd979bdSMark Fasheh } 2331ccd979bdSMark Fasheh } 2332ccd979bdSMark Fasheh 2333ccd979bdSMark Fasheh return 0; 2334ccd979bdSMark Fasheh } 2335ccd979bdSMark Fasheh 2336539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2337539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2338539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2339539d8264SSunil Mushran * a journal has been recovered by another node. 2340539d8264SSunil Mushran */ 2341ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2342ccd979bdSMark Fasheh { 2343ccd979bdSMark Fasheh int ret = 0; 2344ccd979bdSMark Fasheh unsigned int slot; 2345539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2346ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2347539d8264SSunil Mushran int journal_dirty = 0; 2348ccd979bdSMark Fasheh 2349ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2350539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2351539d8264SSunil Mushran if (ret) { 2352ccd979bdSMark Fasheh mlog_errno(ret); 2353ccd979bdSMark Fasheh goto out; 2354ccd979bdSMark Fasheh } 2355ccd979bdSMark Fasheh 2356ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2357ccd979bdSMark Fasheh 2358539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2359539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2360539d8264SSunil Mushran 2361ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2362ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2363539d8264SSunil Mushran journal_dirty = 1; 2364ccd979bdSMark Fasheh 2365ccd979bdSMark Fasheh brelse(di_bh); 2366539d8264SSunil Mushran di_bh = NULL; 2367ccd979bdSMark Fasheh } 2368ccd979bdSMark Fasheh 2369ccd979bdSMark Fasheh out: 2370539d8264SSunil Mushran if (journal_dirty) 2371539d8264SSunil Mushran ret = -EROFS; 2372ccd979bdSMark Fasheh return ret; 2373ccd979bdSMark Fasheh } 2374