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> 18cff61bbcSChristoph Hellwig #include <linux/writeback.h> 19ccd979bdSMark Fasheh 20ccd979bdSMark Fasheh #include <cluster/masklog.h> 21ccd979bdSMark Fasheh 22ccd979bdSMark Fasheh #include "ocfs2.h" 23ccd979bdSMark Fasheh 24ccd979bdSMark Fasheh #include "alloc.h" 2550655ae9SJoel Becker #include "blockcheck.h" 26316f4b9fSMark Fasheh #include "dir.h" 27ccd979bdSMark Fasheh #include "dlmglue.h" 28ccd979bdSMark Fasheh #include "extent_map.h" 29ccd979bdSMark Fasheh #include "heartbeat.h" 30ccd979bdSMark Fasheh #include "inode.h" 31ccd979bdSMark Fasheh #include "journal.h" 32ccd979bdSMark Fasheh #include "localalloc.h" 33ccd979bdSMark Fasheh #include "slot_map.h" 34ccd979bdSMark Fasheh #include "super.h" 35ccd979bdSMark Fasheh #include "sysfile.h" 360cf2f763SJoel Becker #include "uptodate.h" 372205363dSJan Kara #include "quota.h" 38ed460cffSJoseph Qi #include "file.h" 39ed460cffSJoseph Qi #include "namei.h" 40ccd979bdSMark Fasheh 41ccd979bdSMark Fasheh #include "buffer_head_io.h" 42b4107950STao Ma #include "ocfs2_trace.h" 43ccd979bdSMark Fasheh 4434af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 45ccd979bdSMark Fasheh 4683273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 4783273932SSrinivas Eeda 48ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 49ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 502205363dSJan Kara int node_num, int slot_num); 51ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 52ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 5319ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 54ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 55539d8264SSunil Mushran int dirty, int replayed); 56ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 57ccd979bdSMark Fasheh int slot_num); 58ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 59ed460cffSJoseph Qi int slot, 60ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 61ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 629140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 639140db04SSrinivas Eeda int slot_num, 649140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 659140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 66ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 67ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type); 68ccd979bdSMark Fasheh 6919ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 7019ece546SJan Kara { 7119ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 7219ece546SJan Kara } 7319ece546SJan Kara 7419ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 7519ece546SJan Kara { 7619ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 7719ece546SJan Kara } 7819ece546SJan Kara 799140db04SSrinivas Eeda /* 809140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 819140db04SSrinivas Eeda * offline slots during recovery and mount 829140db04SSrinivas Eeda */ 839140db04SSrinivas Eeda 849140db04SSrinivas Eeda enum ocfs2_replay_state { 859140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 869140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 879140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 889140db04SSrinivas Eeda }; 899140db04SSrinivas Eeda 909140db04SSrinivas Eeda struct ocfs2_replay_map { 919140db04SSrinivas Eeda unsigned int rm_slots; 929140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 933c9210d4SGustavo A. R. Silva unsigned char rm_replay_slots[]; 949140db04SSrinivas Eeda }; 959140db04SSrinivas Eeda 96b519ea6dSJoseph Qi static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 979140db04SSrinivas Eeda { 989140db04SSrinivas Eeda if (!osb->replay_map) 999140db04SSrinivas Eeda return; 1009140db04SSrinivas Eeda 1019140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1029140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1039140db04SSrinivas Eeda return; 1049140db04SSrinivas Eeda 1059140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1069140db04SSrinivas Eeda } 1079140db04SSrinivas Eeda 1089140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1099140db04SSrinivas Eeda { 1109140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1119140db04SSrinivas Eeda int i, node_num; 1129140db04SSrinivas Eeda 1139140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1149140db04SSrinivas Eeda if (osb->replay_map) 1159140db04SSrinivas Eeda return 0; 1169140db04SSrinivas Eeda 1179140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1189140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1199140db04SSrinivas Eeda 1209140db04SSrinivas Eeda if (!replay_map) { 1219140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1229140db04SSrinivas Eeda return -ENOMEM; 1239140db04SSrinivas Eeda } 1249140db04SSrinivas Eeda 1259140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1269140db04SSrinivas Eeda 1279140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1289140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1299140db04SSrinivas Eeda 1309140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1319140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1329140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1339140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1349140db04SSrinivas Eeda } 1359140db04SSrinivas Eeda 1369140db04SSrinivas Eeda osb->replay_map = replay_map; 1379140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1389140db04SSrinivas Eeda return 0; 1399140db04SSrinivas Eeda } 1409140db04SSrinivas Eeda 141b519ea6dSJoseph Qi static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, 142ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1439140db04SSrinivas Eeda { 1449140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1459140db04SSrinivas Eeda int i; 1469140db04SSrinivas Eeda 1479140db04SSrinivas Eeda if (!replay_map) 1489140db04SSrinivas Eeda return; 1499140db04SSrinivas Eeda 1509140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1519140db04SSrinivas Eeda return; 1529140db04SSrinivas Eeda 1539140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1549140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1559140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 156ed460cffSJoseph Qi NULL, NULL, 157ed460cffSJoseph Qi orphan_reco_type); 1589140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1599140db04SSrinivas Eeda } 1609140db04SSrinivas Eeda 161ce2fcf15SLi Zetao void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1629140db04SSrinivas Eeda { 1639140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1649140db04SSrinivas Eeda 1659140db04SSrinivas Eeda if (!osb->replay_map) 1669140db04SSrinivas Eeda return; 1679140db04SSrinivas Eeda 1689140db04SSrinivas Eeda kfree(replay_map); 1699140db04SSrinivas Eeda osb->replay_map = NULL; 1709140db04SSrinivas Eeda } 1719140db04SSrinivas Eeda 172553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 173553abd04SJoel Becker { 174553abd04SJoel Becker struct ocfs2_recovery_map *rm; 175553abd04SJoel Becker 176553abd04SJoel Becker mutex_init(&osb->recovery_lock); 177553abd04SJoel Becker osb->disable_recovery = 0; 178553abd04SJoel Becker osb->recovery_thread_task = NULL; 179553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 180553abd04SJoel Becker 181553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 182553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 183553abd04SJoel Becker GFP_KERNEL); 184553abd04SJoel Becker if (!rm) { 185553abd04SJoel Becker mlog_errno(-ENOMEM); 186553abd04SJoel Becker return -ENOMEM; 187553abd04SJoel Becker } 188553abd04SJoel Becker 189553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 190553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 191553abd04SJoel Becker osb->recovery_map = rm; 192553abd04SJoel Becker 193553abd04SJoel Becker return 0; 194553abd04SJoel Becker } 195553abd04SJoel Becker 196553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 197553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 198553abd04SJoel Becker * being woken up */ 199553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 200553abd04SJoel Becker { 201553abd04SJoel Becker mb(); 202553abd04SJoel Becker return osb->recovery_thread_task != NULL; 203553abd04SJoel Becker } 204553abd04SJoel Becker 205553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 206553abd04SJoel Becker { 207553abd04SJoel Becker struct ocfs2_recovery_map *rm; 208553abd04SJoel Becker 209553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 210553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 211553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 212553abd04SJoel Becker osb->disable_recovery = 1; 213553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 214553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 215553abd04SJoel Becker 216553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 217553abd04SJoel Becker * launched, so wait for any recovery completion work to 218553abd04SJoel Becker * complete. */ 219b918c430SYi Li if (osb->ocfs2_wq) 22035ddf78eSjiangyiwen flush_workqueue(osb->ocfs2_wq); 221553abd04SJoel Becker 222553abd04SJoel Becker /* 223553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 224553abd04SJoel Becker * freed, the osb_lock is not taken here. 225553abd04SJoel Becker */ 226553abd04SJoel Becker rm = osb->recovery_map; 227553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 228553abd04SJoel Becker 229553abd04SJoel Becker kfree(rm); 230553abd04SJoel Becker } 231553abd04SJoel Becker 232553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 233553abd04SJoel Becker unsigned int node_num) 234553abd04SJoel Becker { 235553abd04SJoel Becker int i; 236553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 237553abd04SJoel Becker 238553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 239553abd04SJoel Becker 240553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 241553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 242553abd04SJoel Becker return 1; 243553abd04SJoel Becker } 244553abd04SJoel Becker 245553abd04SJoel Becker return 0; 246553abd04SJoel Becker } 247553abd04SJoel Becker 248553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 249553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 250553abd04SJoel Becker unsigned int node_num) 251553abd04SJoel Becker { 252553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 253553abd04SJoel Becker 254553abd04SJoel Becker spin_lock(&osb->osb_lock); 255553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 256553abd04SJoel Becker spin_unlock(&osb->osb_lock); 257553abd04SJoel Becker return 1; 258553abd04SJoel Becker } 259553abd04SJoel Becker 260553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 261553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 262553abd04SJoel Becker 263553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 264553abd04SJoel Becker rm->rm_used++; 265553abd04SJoel Becker spin_unlock(&osb->osb_lock); 266553abd04SJoel Becker 267553abd04SJoel Becker return 0; 268553abd04SJoel Becker } 269553abd04SJoel Becker 270553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 271553abd04SJoel Becker unsigned int node_num) 272553abd04SJoel Becker { 273553abd04SJoel Becker int i; 274553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 275553abd04SJoel Becker 276553abd04SJoel Becker spin_lock(&osb->osb_lock); 277553abd04SJoel Becker 278553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 279553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 280553abd04SJoel Becker break; 281553abd04SJoel Becker } 282553abd04SJoel Becker 283553abd04SJoel Becker if (i < rm->rm_used) { 284553abd04SJoel Becker /* XXX: be careful with the pointer math */ 285553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 286553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 287553abd04SJoel Becker rm->rm_used--; 288553abd04SJoel Becker } 289553abd04SJoel Becker 290553abd04SJoel Becker spin_unlock(&osb->osb_lock); 291553abd04SJoel Becker } 292553abd04SJoel Becker 293ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 294ccd979bdSMark Fasheh { 295ccd979bdSMark Fasheh int status = 0; 296ccd979bdSMark Fasheh unsigned int flushed; 297ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 298ccd979bdSMark Fasheh 299ccd979bdSMark Fasheh journal = osb->journal; 300ccd979bdSMark Fasheh 301ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 302ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 303ccd979bdSMark Fasheh 304b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 305b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 306b4107950STao Ma if (flushed == 0) { 307ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 308ccd979bdSMark Fasheh goto finally; 309ccd979bdSMark Fasheh } 310ccd979bdSMark Fasheh 3112b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 31201d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 3132b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 314ccd979bdSMark Fasheh if (status < 0) { 315ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 316ccd979bdSMark Fasheh mlog_errno(status); 317ccd979bdSMark Fasheh goto finally; 318ccd979bdSMark Fasheh } 319ccd979bdSMark Fasheh 320f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 321ccd979bdSMark Fasheh 322ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 323ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 324ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 325ccd979bdSMark Fasheh 326b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 327ccd979bdSMark Fasheh 32834d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 329ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 330ccd979bdSMark Fasheh finally: 331ccd979bdSMark Fasheh return status; 332ccd979bdSMark Fasheh } 333ccd979bdSMark Fasheh 3341fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 335ccd979bdSMark Fasheh { 336ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3371fabe148SMark Fasheh handle_t *handle; 338ccd979bdSMark Fasheh 339ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 340ccd979bdSMark Fasheh 34165eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 34265eff9ccSMark Fasheh return ERR_PTR(-EROFS); 343ccd979bdSMark Fasheh 344ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 345ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 346ccd979bdSMark Fasheh 34790e86a63SJan Kara /* Nested transaction? Just return the handle... */ 34890e86a63SJan Kara if (journal_current_handle()) 34990e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 350ccd979bdSMark Fasheh 351fef6925cSJan Kara sb_start_intwrite(osb->sb); 352fef6925cSJan Kara 353ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 354ccd979bdSMark Fasheh 3552b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3561fabe148SMark Fasheh if (IS_ERR(handle)) { 357ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 358fef6925cSJan Kara sb_end_intwrite(osb->sb); 359ccd979bdSMark Fasheh 3601fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 361ccd979bdSMark Fasheh 362ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 3637ecef14aSJoe Perches ocfs2_abort(osb->sb, "Detected aborted journal\n"); 3641fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 365ccd979bdSMark Fasheh } 366c271c5c2SSunil Mushran } else { 367c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 368ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 369c271c5c2SSunil Mushran } 370ccd979bdSMark Fasheh 371ccd979bdSMark Fasheh return handle; 372ccd979bdSMark Fasheh } 373ccd979bdSMark Fasheh 3741fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3751fabe148SMark Fasheh handle_t *handle) 376ccd979bdSMark Fasheh { 37790e86a63SJan Kara int ret, nested; 37802dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 379ccd979bdSMark Fasheh 380ccd979bdSMark Fasheh BUG_ON(!handle); 381ccd979bdSMark Fasheh 38290e86a63SJan Kara nested = handle->h_ref > 1; 3832b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3841fabe148SMark Fasheh if (ret < 0) 3851fabe148SMark Fasheh mlog_errno(ret); 386ccd979bdSMark Fasheh 387fef6925cSJan Kara if (!nested) { 388ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 389fef6925cSJan Kara sb_end_intwrite(osb->sb); 390fef6925cSJan Kara } 391ccd979bdSMark Fasheh 3921fabe148SMark Fasheh return ret; 393ccd979bdSMark Fasheh } 394ccd979bdSMark Fasheh 395ccd979bdSMark Fasheh /* 396c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 397ccd979bdSMark Fasheh * 3982b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 399e8aed345SMark Fasheh * and then restart the transaction. Before calling 400e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 401e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 402e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 403e8aed345SMark Fasheh * 404ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 405ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 406ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 407ccd979bdSMark Fasheh * 408ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 409ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 410ccd979bdSMark Fasheh * cluster locks associated with this handle. 411ccd979bdSMark Fasheh */ 4121fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 413ccd979bdSMark Fasheh { 414c901fb00STao Ma int status, old_nblocks; 415ccd979bdSMark Fasheh 416ccd979bdSMark Fasheh BUG_ON(!handle); 417c901fb00STao Ma BUG_ON(nblocks < 0); 418ccd979bdSMark Fasheh 419c901fb00STao Ma if (!nblocks) 420c901fb00STao Ma return 0; 421c901fb00STao Ma 4229797a902SJan Kara old_nblocks = jbd2_handle_buffer_credits(handle); 423ccd979bdSMark Fasheh 424b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 425ccd979bdSMark Fasheh 426e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4270879c584SMark Fasheh status = 1; 4280879c584SMark Fasheh #else 429fdc3ef88SJan Kara status = jbd2_journal_extend(handle, nblocks, 0); 430ccd979bdSMark Fasheh if (status < 0) { 431ccd979bdSMark Fasheh mlog_errno(status); 432ccd979bdSMark Fasheh goto bail; 433ccd979bdSMark Fasheh } 4340879c584SMark Fasheh #endif 435ccd979bdSMark Fasheh 436ccd979bdSMark Fasheh if (status > 0) { 437b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 438c901fb00STao Ma status = jbd2_journal_restart(handle, 439c901fb00STao Ma old_nblocks + nblocks); 440ccd979bdSMark Fasheh if (status < 0) { 441ccd979bdSMark Fasheh mlog_errno(status); 442ccd979bdSMark Fasheh goto bail; 443ccd979bdSMark Fasheh } 44401ddf1e1SMark Fasheh } 445ccd979bdSMark Fasheh 446ccd979bdSMark Fasheh status = 0; 447ccd979bdSMark Fasheh bail: 448ccd979bdSMark Fasheh return status; 449ccd979bdSMark Fasheh } 450ccd979bdSMark Fasheh 4512b1e55c3SYounger Liu /* 4522b1e55c3SYounger Liu * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 4532b1e55c3SYounger Liu * If that fails, restart the transaction & regain write access for the 4542b1e55c3SYounger Liu * buffer head which is used for metadata modifications. 4552b1e55c3SYounger Liu * Taken from Ext4: extend_or_restart_transaction() 4562b1e55c3SYounger Liu */ 4572b1e55c3SYounger Liu int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 4582b1e55c3SYounger Liu { 4592b1e55c3SYounger Liu int status, old_nblks; 4602b1e55c3SYounger Liu 4612b1e55c3SYounger Liu BUG_ON(!handle); 4622b1e55c3SYounger Liu 4639797a902SJan Kara old_nblks = jbd2_handle_buffer_credits(handle); 4642b1e55c3SYounger Liu trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 4652b1e55c3SYounger Liu 4662b1e55c3SYounger Liu if (old_nblks < thresh) 4672b1e55c3SYounger Liu return 0; 4682b1e55c3SYounger Liu 469fdc3ef88SJan Kara status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0); 4702b1e55c3SYounger Liu if (status < 0) { 4712b1e55c3SYounger Liu mlog_errno(status); 4722b1e55c3SYounger Liu goto bail; 4732b1e55c3SYounger Liu } 4742b1e55c3SYounger Liu 4752b1e55c3SYounger Liu if (status > 0) { 4762b1e55c3SYounger Liu status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 4772b1e55c3SYounger Liu if (status < 0) 4782b1e55c3SYounger Liu mlog_errno(status); 4792b1e55c3SYounger Liu } 4802b1e55c3SYounger Liu 4812b1e55c3SYounger Liu bail: 4822b1e55c3SYounger Liu return status; 4832b1e55c3SYounger Liu } 4842b1e55c3SYounger Liu 4852b1e55c3SYounger Liu 48650655ae9SJoel Becker struct ocfs2_triggers { 48750655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 48850655ae9SJoel Becker int ot_offset; 48950655ae9SJoel Becker }; 49050655ae9SJoel Becker 49150655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 49250655ae9SJoel Becker { 49350655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 49450655ae9SJoel Becker } 49550655ae9SJoel Becker 49613ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 49750655ae9SJoel Becker struct buffer_head *bh, 49850655ae9SJoel Becker void *data, size_t size) 49950655ae9SJoel Becker { 50050655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 50150655ae9SJoel Becker 50250655ae9SJoel Becker /* 50350655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 50450655ae9SJoel Becker * must unconditionally compute the ecc data. 50550655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 50650655ae9SJoel Becker * metaecc is enabled. 50750655ae9SJoel Becker */ 50850655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 50950655ae9SJoel Becker } 51050655ae9SJoel Becker 51150655ae9SJoel Becker /* 51250655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 51350655ae9SJoel Becker * offset depends on the blocksize. 51450655ae9SJoel Becker */ 51513ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 51650655ae9SJoel Becker struct buffer_head *bh, 51750655ae9SJoel Becker void *data, size_t size) 51850655ae9SJoel Becker { 51950655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 52050655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 52150655ae9SJoel Becker 52250655ae9SJoel Becker /* 52350655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 52450655ae9SJoel Becker * must unconditionally compute the ecc data. 52550655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 52650655ae9SJoel Becker * metaecc is enabled. 52750655ae9SJoel Becker */ 52850655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 52950655ae9SJoel Becker } 53050655ae9SJoel Becker 531c175a518SJoel Becker /* 532c175a518SJoel Becker * Directory blocks also have their own trigger because the 533c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 534c175a518SJoel Becker */ 53513ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 536c175a518SJoel Becker struct buffer_head *bh, 537c175a518SJoel Becker void *data, size_t size) 538c175a518SJoel Becker { 539c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 540c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 541c175a518SJoel Becker 542c175a518SJoel Becker /* 543c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 544c175a518SJoel Becker * must unconditionally compute the ecc data. 545c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 546c175a518SJoel Becker * metaecc is enabled. 547c175a518SJoel Becker */ 548c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 549c175a518SJoel Becker } 550c175a518SJoel Becker 55150655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 55250655ae9SJoel Becker struct buffer_head *bh) 55350655ae9SJoel Becker { 55450655ae9SJoel Becker mlog(ML_ERROR, 55550655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 55650655ae9SJoel Becker "bh->b_blocknr = %llu\n", 55750655ae9SJoel Becker (unsigned long)bh, 55850655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 55950655ae9SJoel Becker 56074e364adSXue jiufei ocfs2_error(bh->b_bdev->bd_super, 56150655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 56250655ae9SJoel Becker } 56350655ae9SJoel Becker 56450655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 56550655ae9SJoel Becker .ot_triggers = { 56613ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 56750655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 56850655ae9SJoel Becker }, 56950655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 57050655ae9SJoel Becker }; 57150655ae9SJoel Becker 57250655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 57350655ae9SJoel Becker .ot_triggers = { 57413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 57550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 57650655ae9SJoel Becker }, 57750655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 57850655ae9SJoel Becker }; 57950655ae9SJoel Becker 58093c97087STao Ma static struct ocfs2_triggers rb_triggers = { 58193c97087STao Ma .ot_triggers = { 58213ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58393c97087STao Ma .t_abort = ocfs2_abort_trigger, 58493c97087STao Ma }, 58593c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 58693c97087STao Ma }; 58793c97087STao Ma 58850655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 58950655ae9SJoel Becker .ot_triggers = { 59013ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 59150655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 59250655ae9SJoel Becker }, 59350655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 59450655ae9SJoel Becker }; 59550655ae9SJoel Becker 596c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 597c175a518SJoel Becker .ot_triggers = { 59813ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 599c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 600c175a518SJoel Becker }, 601c175a518SJoel Becker }; 602c175a518SJoel Becker 60350655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 60450655ae9SJoel Becker .ot_triggers = { 60513ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 60650655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 60750655ae9SJoel Becker }, 60850655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 60950655ae9SJoel Becker }; 61050655ae9SJoel Becker 61150655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 61250655ae9SJoel Becker .ot_triggers = { 61313ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 61450655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 61550655ae9SJoel Becker }, 61650655ae9SJoel Becker }; 61750655ae9SJoel Becker 6189b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 6199b7895efSMark Fasheh .ot_triggers = { 62013ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6219b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6229b7895efSMark Fasheh }, 6239b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 6249b7895efSMark Fasheh }; 6259b7895efSMark Fasheh 6269b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 6279b7895efSMark Fasheh .ot_triggers = { 62813ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6299b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6309b7895efSMark Fasheh }, 6319b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6329b7895efSMark Fasheh }; 6339b7895efSMark Fasheh 63450655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6350cf2f763SJoel Becker struct ocfs2_caching_info *ci, 636ccd979bdSMark Fasheh struct buffer_head *bh, 63750655ae9SJoel Becker struct ocfs2_triggers *triggers, 638ccd979bdSMark Fasheh int type) 639ccd979bdSMark Fasheh { 640ccd979bdSMark Fasheh int status; 6410cf2f763SJoel Becker struct ocfs2_super *osb = 6420cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 643ccd979bdSMark Fasheh 6440cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 645ccd979bdSMark Fasheh BUG_ON(!handle); 646ccd979bdSMark Fasheh BUG_ON(!bh); 647ccd979bdSMark Fasheh 648b4107950STao Ma trace_ocfs2_journal_access( 649b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 650b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 651ccd979bdSMark Fasheh 652ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 653ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 654ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 655d984187eSpiaojun mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", 656d984187eSpiaojun (unsigned long long)bh->b_blocknr, bh->b_state); 657acf8fdbeSJoseph Qi 658acf8fdbeSJoseph Qi lock_buffer(bh); 659acf8fdbeSJoseph Qi /* 660d984187eSpiaojun * A previous transaction with a couple of buffer heads fail 661d984187eSpiaojun * to checkpoint, so all the bhs are marked as BH_Write_EIO. 662d984187eSpiaojun * For current transaction, the bh is just among those error 663d984187eSpiaojun * bhs which previous transaction handle. We can't just clear 664d984187eSpiaojun * its BH_Write_EIO and reuse directly, since other bhs are 665d984187eSpiaojun * not written to disk yet and that will cause metadata 666d984187eSpiaojun * inconsistency. So we should set fs read-only to avoid 667d984187eSpiaojun * further damage. 668acf8fdbeSJoseph Qi */ 669acf8fdbeSJoseph Qi if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { 670acf8fdbeSJoseph Qi unlock_buffer(bh); 671d984187eSpiaojun return ocfs2_error(osb->sb, "A previous attempt to " 672d984187eSpiaojun "write this buffer head failed\n"); 673acf8fdbeSJoseph Qi } 674acf8fdbeSJoseph Qi unlock_buffer(bh); 675ccd979bdSMark Fasheh } 676ccd979bdSMark Fasheh 6770cf2f763SJoel Becker /* Set the current transaction information on the ci so 678ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6790cf2f763SJoel Becker * on this ci or not. We're protected from the commit 680ccd979bdSMark Fasheh * thread updating the current transaction id until 681ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 682ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6830cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 684ccd979bdSMark Fasheh 6850cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 686ccd979bdSMark Fasheh switch (type) { 687ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 688ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 6892b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 690ccd979bdSMark Fasheh break; 691ccd979bdSMark Fasheh 692ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 6932b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 694ccd979bdSMark Fasheh break; 695ccd979bdSMark Fasheh 696ccd979bdSMark Fasheh default: 697ccd979bdSMark Fasheh status = -EINVAL; 698af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 699ccd979bdSMark Fasheh } 7000cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 70150655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 7020cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 703ccd979bdSMark Fasheh 704ccd979bdSMark Fasheh if (status < 0) 705ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 706ccd979bdSMark Fasheh status, type); 707ccd979bdSMark Fasheh 708ccd979bdSMark Fasheh return status; 709ccd979bdSMark Fasheh } 710ccd979bdSMark Fasheh 7110cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 71250655ae9SJoel Becker struct buffer_head *bh, int type) 71350655ae9SJoel Becker { 7140cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 71550655ae9SJoel Becker } 71650655ae9SJoel Becker 7170cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 71850655ae9SJoel Becker struct buffer_head *bh, int type) 71950655ae9SJoel Becker { 7200cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 72150655ae9SJoel Becker } 72250655ae9SJoel Becker 72393c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 72493c97087STao Ma struct buffer_head *bh, int type) 72593c97087STao Ma { 72693c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 72793c97087STao Ma type); 72893c97087STao Ma } 72993c97087STao Ma 7300cf2f763SJoel Becker int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 73150655ae9SJoel Becker struct buffer_head *bh, int type) 73250655ae9SJoel Becker { 7330cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 73450655ae9SJoel Becker } 73550655ae9SJoel Becker 7360cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 73750655ae9SJoel Becker struct buffer_head *bh, int type) 73850655ae9SJoel Becker { 7390cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 74050655ae9SJoel Becker } 74150655ae9SJoel Becker 7420cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 74350655ae9SJoel Becker struct buffer_head *bh, int type) 74450655ae9SJoel Becker { 7450cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 74650655ae9SJoel Becker } 74750655ae9SJoel Becker 7480cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 74950655ae9SJoel Becker struct buffer_head *bh, int type) 75050655ae9SJoel Becker { 7510cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 75250655ae9SJoel Becker } 75350655ae9SJoel Becker 7540cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7559b7895efSMark Fasheh struct buffer_head *bh, int type) 7569b7895efSMark Fasheh { 7570cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7589b7895efSMark Fasheh } 7599b7895efSMark Fasheh 7600cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7619b7895efSMark Fasheh struct buffer_head *bh, int type) 7629b7895efSMark Fasheh { 7630cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7649b7895efSMark Fasheh } 7659b7895efSMark Fasheh 7660cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 76750655ae9SJoel Becker struct buffer_head *bh, int type) 76850655ae9SJoel Becker { 7690cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 77050655ae9SJoel Becker } 77150655ae9SJoel Becker 772ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 773ccd979bdSMark Fasheh { 774ccd979bdSMark Fasheh int status; 775ccd979bdSMark Fasheh 776b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 777ccd979bdSMark Fasheh 7782b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 779e272e7f0SJoseph Qi if (status) { 780e272e7f0SJoseph Qi mlog_errno(status); 781e272e7f0SJoseph Qi if (!is_handle_aborted(handle)) { 782e272e7f0SJoseph Qi journal_t *journal = handle->h_transaction->t_journal; 783e272e7f0SJoseph Qi struct super_block *sb = bh->b_bdev->bd_super; 784e272e7f0SJoseph Qi 785e272e7f0SJoseph Qi mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " 786e272e7f0SJoseph Qi "Aborting transaction and journal.\n"); 787e272e7f0SJoseph Qi handle->h_err = status; 788e272e7f0SJoseph Qi jbd2_journal_abort_handle(handle); 789e272e7f0SJoseph Qi jbd2_journal_abort(journal, status); 790e272e7f0SJoseph Qi ocfs2_abort(sb, "Journal already aborted.\n"); 791e272e7f0SJoseph Qi } 792e272e7f0SJoseph Qi } 793ccd979bdSMark Fasheh } 794ccd979bdSMark Fasheh 7952b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 796ccd979bdSMark Fasheh 797ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 798ccd979bdSMark Fasheh { 799ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 800d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 801d147b3d6SMark Fasheh 802d147b3d6SMark Fasheh if (osb->osb_commit_interval) 803d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 804ccd979bdSMark Fasheh 805a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 806d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 807ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 8082b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 809ccd979bdSMark Fasheh else 8102b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 811a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 812ccd979bdSMark Fasheh } 813ccd979bdSMark Fasheh 814bb20b31dSHeming Zhao via Ocfs2-devel /* 815bb20b31dSHeming Zhao via Ocfs2-devel * alloc & initialize skeleton for journal structure. 816bb20b31dSHeming Zhao via Ocfs2-devel * ocfs2_journal_init() will make fs have journal ability. 817bb20b31dSHeming Zhao via Ocfs2-devel */ 818bb20b31dSHeming Zhao via Ocfs2-devel int ocfs2_journal_alloc(struct ocfs2_super *osb) 819ccd979bdSMark Fasheh { 820bb20b31dSHeming Zhao via Ocfs2-devel int status = 0; 821bb20b31dSHeming Zhao via Ocfs2-devel struct ocfs2_journal *journal; 822ccd979bdSMark Fasheh 823da5e7c87SValentin Vidic journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL); 824da5e7c87SValentin Vidic if (!journal) { 825da5e7c87SValentin Vidic mlog(ML_ERROR, "unable to alloc journal\n"); 826da5e7c87SValentin Vidic status = -ENOMEM; 827bb20b31dSHeming Zhao via Ocfs2-devel goto bail; 828da5e7c87SValentin Vidic } 829da5e7c87SValentin Vidic osb->journal = journal; 830da5e7c87SValentin Vidic journal->j_osb = osb; 831ccd979bdSMark Fasheh 832da5e7c87SValentin Vidic atomic_set(&journal->j_num_trans, 0); 833da5e7c87SValentin Vidic init_rwsem(&journal->j_trans_barrier); 834da5e7c87SValentin Vidic init_waitqueue_head(&journal->j_checkpointed); 835da5e7c87SValentin Vidic spin_lock_init(&journal->j_lock); 836da5e7c87SValentin Vidic journal->j_trans_id = 1UL; 837da5e7c87SValentin Vidic INIT_LIST_HEAD(&journal->j_la_cleanups); 838da5e7c87SValentin Vidic INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery); 839da5e7c87SValentin Vidic journal->j_state = OCFS2_JOURNAL_FREE; 840ccd979bdSMark Fasheh 841bb20b31dSHeming Zhao via Ocfs2-devel bail: 842bb20b31dSHeming Zhao via Ocfs2-devel return status; 843bb20b31dSHeming Zhao via Ocfs2-devel } 844bb20b31dSHeming Zhao via Ocfs2-devel 845cff61bbcSChristoph Hellwig static int ocfs2_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 846cff61bbcSChristoph Hellwig { 847cff61bbcSChristoph Hellwig struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 848cff61bbcSChristoph Hellwig struct writeback_control wbc = { 849cff61bbcSChristoph Hellwig .sync_mode = WB_SYNC_ALL, 850cff61bbcSChristoph Hellwig .nr_to_write = mapping->nrpages * 2, 851cff61bbcSChristoph Hellwig .range_start = jinode->i_dirty_start, 852cff61bbcSChristoph Hellwig .range_end = jinode->i_dirty_end, 853cff61bbcSChristoph Hellwig }; 854cff61bbcSChristoph Hellwig 85517c30ee6SChristoph Hellwig return filemap_fdatawrite_wbc(mapping, &wbc); 856cff61bbcSChristoph Hellwig } 857cff61bbcSChristoph Hellwig 858bb20b31dSHeming Zhao via Ocfs2-devel int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty) 859bb20b31dSHeming Zhao via Ocfs2-devel { 860bb20b31dSHeming Zhao via Ocfs2-devel int status = -1; 861bb20b31dSHeming Zhao via Ocfs2-devel struct inode *inode = NULL; /* the journal inode */ 862bb20b31dSHeming Zhao via Ocfs2-devel journal_t *j_journal = NULL; 863bb20b31dSHeming Zhao via Ocfs2-devel struct ocfs2_journal *journal = osb->journal; 864bb20b31dSHeming Zhao via Ocfs2-devel struct ocfs2_dinode *di = NULL; 865bb20b31dSHeming Zhao via Ocfs2-devel struct buffer_head *bh = NULL; 866bb20b31dSHeming Zhao via Ocfs2-devel int inode_lock = 0; 867bb20b31dSHeming Zhao via Ocfs2-devel 868bb20b31dSHeming Zhao via Ocfs2-devel BUG_ON(!journal); 869ccd979bdSMark Fasheh /* already have the inode for our journal */ 870ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 871ccd979bdSMark Fasheh osb->slot_num); 872ccd979bdSMark Fasheh if (inode == NULL) { 873ccd979bdSMark Fasheh status = -EACCES; 874ccd979bdSMark Fasheh mlog_errno(status); 875ccd979bdSMark Fasheh goto done; 876ccd979bdSMark Fasheh } 877ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 878ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 879ccd979bdSMark Fasheh iput(inode); 880ccd979bdSMark Fasheh inode = NULL; 881ccd979bdSMark Fasheh status = -EACCES; 882ccd979bdSMark Fasheh goto done; 883ccd979bdSMark Fasheh } 884ccd979bdSMark Fasheh 885ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 886ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 887ccd979bdSMark Fasheh 8886eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 8896eff5790SMark Fasheh * changes in a live cluster so it can be considered an 8906eff5790SMark Fasheh * exception to the rule. */ 891e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 892ccd979bdSMark Fasheh if (status < 0) { 893ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 894ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 895ccd979bdSMark Fasheh goto done; 896ccd979bdSMark Fasheh } 897ccd979bdSMark Fasheh 898e63aecb6SMark Fasheh inode_lock = 1; 899ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 900ccd979bdSMark Fasheh 901f17c20ddSJunxiao Bi if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 902ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 903f17c20ddSJunxiao Bi i_size_read(inode)); 904ccd979bdSMark Fasheh status = -EINVAL; 905ccd979bdSMark Fasheh goto done; 906ccd979bdSMark Fasheh } 907ccd979bdSMark Fasheh 908f17c20ddSJunxiao Bi trace_ocfs2_journal_init(i_size_read(inode), 909b4107950STao Ma (unsigned long long)inode->i_blocks, 910b4107950STao Ma OCFS2_I(inode)->ip_clusters); 911ccd979bdSMark Fasheh 912ccd979bdSMark Fasheh /* call the kernels journal init function now */ 9132b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 914*8e6cf5fbSZhang Yi if (IS_ERR(j_journal)) { 915ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 916*8e6cf5fbSZhang Yi status = PTR_ERR(j_journal); 917ccd979bdSMark Fasheh goto done; 918ccd979bdSMark Fasheh } 919ccd979bdSMark Fasheh 920ede7dc7fSHarshad Shirwadkar trace_ocfs2_journal_init_maxlen(j_journal->j_total_len); 921ccd979bdSMark Fasheh 922ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 923ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 924ccd979bdSMark Fasheh 925ccd979bdSMark Fasheh journal->j_journal = j_journal; 926342af94eSMauricio Faria de Oliveira journal->j_journal->j_submit_inode_data_buffers = 927cff61bbcSChristoph Hellwig ocfs2_journal_submit_inode_data_buffers; 928342af94eSMauricio Faria de Oliveira journal->j_journal->j_finish_inode_data_buffers = 929342af94eSMauricio Faria de Oliveira jbd2_journal_finish_inode_data_buffers; 930ccd979bdSMark Fasheh journal->j_inode = inode; 931ccd979bdSMark Fasheh journal->j_bh = bh; 932ccd979bdSMark Fasheh 933ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 934ccd979bdSMark Fasheh 935ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 936ccd979bdSMark Fasheh 937ccd979bdSMark Fasheh status = 0; 938ccd979bdSMark Fasheh done: 939ccd979bdSMark Fasheh if (status < 0) { 940e63aecb6SMark Fasheh if (inode_lock) 941e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 942ccd979bdSMark Fasheh brelse(bh); 943ccd979bdSMark Fasheh if (inode) { 944ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 945ccd979bdSMark Fasheh iput(inode); 946ccd979bdSMark Fasheh } 947ccd979bdSMark Fasheh } 948ccd979bdSMark Fasheh 949ccd979bdSMark Fasheh return status; 950ccd979bdSMark Fasheh } 951ccd979bdSMark Fasheh 952539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 953539d8264SSunil Mushran { 954539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 955539d8264SSunil Mushran } 956539d8264SSunil Mushran 957539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 958539d8264SSunil Mushran { 959539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 960539d8264SSunil Mushran } 961539d8264SSunil Mushran 962ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 963539d8264SSunil Mushran int dirty, int replayed) 964ccd979bdSMark Fasheh { 965ccd979bdSMark Fasheh int status; 966ccd979bdSMark Fasheh unsigned int flags; 967ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 968ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 969ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 970ccd979bdSMark Fasheh 971ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 97210995aa2SJoel Becker 97310995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 97410995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 97510995aa2SJoel Becker * code bug if we mess it up. */ 97610995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 977ccd979bdSMark Fasheh 978ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 979ccd979bdSMark Fasheh if (dirty) 980ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 981ccd979bdSMark Fasheh else 982ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 983ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 984ccd979bdSMark Fasheh 985539d8264SSunil Mushran if (replayed) 986539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 987539d8264SSunil Mushran 98813723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 9898cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 990ccd979bdSMark Fasheh if (status < 0) 991ccd979bdSMark Fasheh mlog_errno(status); 992ccd979bdSMark Fasheh 993ccd979bdSMark Fasheh return status; 994ccd979bdSMark Fasheh } 995ccd979bdSMark Fasheh 996ccd979bdSMark Fasheh /* 997ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 998ccd979bdSMark Fasheh * call. 999ccd979bdSMark Fasheh */ 1000ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 1001ccd979bdSMark Fasheh { 1002ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 1003ccd979bdSMark Fasheh int status = 0; 1004ccd979bdSMark Fasheh struct inode *inode = NULL; 1005ccd979bdSMark Fasheh int num_running_trans = 0; 1006ccd979bdSMark Fasheh 1007ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 1008ccd979bdSMark Fasheh 1009ccd979bdSMark Fasheh journal = osb->journal; 1010ccd979bdSMark Fasheh if (!journal) 1011ccd979bdSMark Fasheh goto done; 1012ccd979bdSMark Fasheh 1013ccd979bdSMark Fasheh inode = journal->j_inode; 1014ccd979bdSMark Fasheh 1015ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 1016ccd979bdSMark Fasheh goto done; 1017ccd979bdSMark Fasheh 10182b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 1019ccd979bdSMark Fasheh if (!igrab(inode)) 1020ccd979bdSMark Fasheh BUG(); 1021ccd979bdSMark Fasheh 1022ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 1023b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 1024ccd979bdSMark Fasheh 1025ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 1026ccd979bdSMark Fasheh * release any locks that are still held. 1027ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 1028ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 1029ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 1030ccd979bdSMark Fasheh 1031ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 1032ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 1033ccd979bdSMark Fasheh * completely destroy the journal. */ 1034ccd979bdSMark Fasheh if (osb->commit_task) { 1035ccd979bdSMark Fasheh /* Wait for the commit thread */ 1036b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 1037ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 1038ccd979bdSMark Fasheh osb->commit_task = NULL; 1039ccd979bdSMark Fasheh } 1040ccd979bdSMark Fasheh 1041ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 1042ccd979bdSMark Fasheh 1043c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 10442b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 104501d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 10462b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 1047c271c5c2SSunil Mushran if (status < 0) 1048c271c5c2SSunil Mushran mlog_errno(status); 1049c271c5c2SSunil Mushran } 1050c271c5c2SSunil Mushran 1051d85400afSJunxiao Bi /* Shutdown the kernel journal system */ 1052d85400afSJunxiao Bi if (!jbd2_journal_destroy(journal->j_journal) && !status) { 1053c271c5c2SSunil Mushran /* 1054c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 1055c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 1056c271c5c2SSunil Mushran */ 1057539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 1058ccd979bdSMark Fasheh if (status < 0) 1059ccd979bdSMark Fasheh mlog_errno(status); 1060c271c5c2SSunil Mushran } 1061ae0dff68SSunil Mushran journal->j_journal = NULL; 1062ccd979bdSMark Fasheh 1063ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 1064ccd979bdSMark Fasheh 1065ccd979bdSMark Fasheh /* unlock our journal */ 1066e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1067ccd979bdSMark Fasheh 1068ccd979bdSMark Fasheh brelse(journal->j_bh); 1069ccd979bdSMark Fasheh journal->j_bh = NULL; 1070ccd979bdSMark Fasheh 1071ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 1072ccd979bdSMark Fasheh 1073ccd979bdSMark Fasheh done: 1074ccd979bdSMark Fasheh iput(inode); 1075da5e7c87SValentin Vidic kfree(journal); 1076da5e7c87SValentin Vidic osb->journal = NULL; 1077ccd979bdSMark Fasheh } 1078ccd979bdSMark Fasheh 1079ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 1080ccd979bdSMark Fasheh journal_t *journal, 1081ccd979bdSMark Fasheh int slot) 1082ccd979bdSMark Fasheh { 1083ccd979bdSMark Fasheh int olderr; 1084ccd979bdSMark Fasheh 10852b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 1086ccd979bdSMark Fasheh if (olderr) { 1087ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 1088ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 1089ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 1090ccd979bdSMark Fasheh sb->s_id); 1091ccd979bdSMark Fasheh 10922b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 10932b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 1094ccd979bdSMark Fasheh } 1095ccd979bdSMark Fasheh } 1096ccd979bdSMark Fasheh 1097539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1098ccd979bdSMark Fasheh { 1099ccd979bdSMark Fasheh int status = 0; 1100ccd979bdSMark Fasheh struct ocfs2_super *osb; 1101ccd979bdSMark Fasheh 1102b1f3550fSJulia Lawall BUG_ON(!journal); 1103ccd979bdSMark Fasheh 1104ccd979bdSMark Fasheh osb = journal->j_osb; 1105ccd979bdSMark Fasheh 11062b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1107ccd979bdSMark Fasheh if (status < 0) { 1108ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1109ccd979bdSMark Fasheh goto done; 1110ccd979bdSMark Fasheh } 1111ccd979bdSMark Fasheh 1112ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1113ccd979bdSMark Fasheh 1114397eac17SKai Li if (replayed) { 1115397eac17SKai Li jbd2_journal_lock_updates(journal->j_journal); 111601d5d965SLeah Rumancik status = jbd2_journal_flush(journal->j_journal, 0); 1117397eac17SKai Li jbd2_journal_unlock_updates(journal->j_journal); 1118397eac17SKai Li if (status < 0) 1119397eac17SKai Li mlog_errno(status); 1120397eac17SKai Li } 1121397eac17SKai Li 1122539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1123ccd979bdSMark Fasheh if (status < 0) { 1124ccd979bdSMark Fasheh mlog_errno(status); 1125ccd979bdSMark Fasheh goto done; 1126ccd979bdSMark Fasheh } 1127ccd979bdSMark Fasheh 1128ccd979bdSMark Fasheh /* Launch the commit thread */ 1129c271c5c2SSunil Mushran if (!local) { 1130c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 11315afc44e2SJoseph Qi "ocfs2cmt-%s", osb->uuid_str); 1132ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1133ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1134ccd979bdSMark Fasheh osb->commit_task = NULL; 1135c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1136c271c5c2SSunil Mushran "error=%d", status); 1137ccd979bdSMark Fasheh goto done; 1138ccd979bdSMark Fasheh } 1139c271c5c2SSunil Mushran } else 1140c271c5c2SSunil Mushran osb->commit_task = NULL; 1141ccd979bdSMark Fasheh 1142ccd979bdSMark Fasheh done: 1143ccd979bdSMark Fasheh return status; 1144ccd979bdSMark Fasheh } 1145ccd979bdSMark Fasheh 1146ccd979bdSMark Fasheh 1147ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1148ccd979bdSMark Fasheh * mark the journal clean */ 1149ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1150ccd979bdSMark Fasheh { 1151ccd979bdSMark Fasheh int status; 1152ccd979bdSMark Fasheh 1153ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1154ccd979bdSMark Fasheh 11552b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1156ccd979bdSMark Fasheh if (status < 0) { 1157ccd979bdSMark Fasheh mlog_errno(status); 1158ccd979bdSMark Fasheh goto bail; 1159ccd979bdSMark Fasheh } 1160ccd979bdSMark Fasheh 1161539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1162ccd979bdSMark Fasheh if (status < 0) 1163ccd979bdSMark Fasheh mlog_errno(status); 1164ccd979bdSMark Fasheh 1165ccd979bdSMark Fasheh bail: 1166ccd979bdSMark Fasheh return status; 1167ccd979bdSMark Fasheh } 1168ccd979bdSMark Fasheh 1169553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1170553abd04SJoel Becker { 1171553abd04SJoel Becker int empty; 1172553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1173553abd04SJoel Becker 1174553abd04SJoel Becker spin_lock(&osb->osb_lock); 1175553abd04SJoel Becker empty = (rm->rm_used == 0); 1176553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1177553abd04SJoel Becker 1178553abd04SJoel Becker return empty; 1179553abd04SJoel Becker } 1180553abd04SJoel Becker 1181553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1182553abd04SJoel Becker { 1183553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1184553abd04SJoel Becker } 1185553abd04SJoel Becker 1186ccd979bdSMark Fasheh /* 1187ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1188ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1189ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1190ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1191ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1192ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1193ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1194ccd979bdSMark Fasheh * as we have the lock by now :) 1195ccd979bdSMark Fasheh */ 1196ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1197ccd979bdSMark Fasheh { 1198ccd979bdSMark Fasheh int status = 0; 11994f902c37SMark Fasheh int i; 12008110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 12010b492f68SJunxiao Bi struct buffer_head *bh = NULL; 12020b492f68SJunxiao Bi struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1203ccd979bdSMark Fasheh 1204f17c20ddSJunxiao Bi num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1205ccd979bdSMark Fasheh v_blkno = 0; 12068110b073SMark Fasheh while (v_blkno < num_blocks) { 1207ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 120849cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1209ccd979bdSMark Fasheh if (status < 0) { 1210ccd979bdSMark Fasheh mlog_errno(status); 1211ccd979bdSMark Fasheh goto bail; 1212ccd979bdSMark Fasheh } 1213ccd979bdSMark Fasheh 12140b492f68SJunxiao Bi for (i = 0; i < p_blocks; i++, p_blkno++) { 12150b492f68SJunxiao Bi bh = __find_get_block(osb->sb->s_bdev, p_blkno, 12160b492f68SJunxiao Bi osb->sb->s_blocksize); 12170b492f68SJunxiao Bi /* block not cached. */ 12180b492f68SJunxiao Bi if (!bh) 12190b492f68SJunxiao Bi continue; 1220ccd979bdSMark Fasheh 12210b492f68SJunxiao Bi brelse(bh); 12220b492f68SJunxiao Bi bh = NULL; 1223dd4a2c2bSMark Fasheh /* We are reading journal data which should not 12240b492f68SJunxiao Bi * be put in the uptodate cache. 12250b492f68SJunxiao Bi */ 12260b492f68SJunxiao Bi status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); 1227ccd979bdSMark Fasheh if (status < 0) { 1228ccd979bdSMark Fasheh mlog_errno(status); 1229ccd979bdSMark Fasheh goto bail; 1230ccd979bdSMark Fasheh } 1231ccd979bdSMark Fasheh 12320b492f68SJunxiao Bi brelse(bh); 12330b492f68SJunxiao Bi bh = NULL; 1234ccd979bdSMark Fasheh } 1235ccd979bdSMark Fasheh 1236ccd979bdSMark Fasheh v_blkno += p_blocks; 1237ccd979bdSMark Fasheh } 1238ccd979bdSMark Fasheh 1239ccd979bdSMark Fasheh bail: 1240ccd979bdSMark Fasheh return status; 1241ccd979bdSMark Fasheh } 1242ccd979bdSMark Fasheh 1243ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1244ccd979bdSMark Fasheh struct list_head lri_list; 1245ccd979bdSMark Fasheh int lri_slot; 1246ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1247ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 12482205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1249ed460cffSJoseph Qi enum ocfs2_orphan_reco_type lri_orphan_reco_type; 1250ccd979bdSMark Fasheh }; 1251ccd979bdSMark Fasheh 1252ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1253ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1254ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1255ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1256ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1257ccd979bdSMark Fasheh * recovering nodes into account. 1258ccd979bdSMark Fasheh * 1259ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1260ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1261ccd979bdSMark Fasheh */ 1262c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1263ccd979bdSMark Fasheh { 1264b4107950STao Ma int ret = 0; 1265c4028958SDavid Howells struct ocfs2_journal *journal = 1266c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1267c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1268ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1269800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 12702205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1271ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 1272ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1273ccd979bdSMark Fasheh 1274b4107950STao Ma trace_ocfs2_complete_recovery( 1275b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1276ccd979bdSMark Fasheh 1277ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1278ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1279ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1280ccd979bdSMark Fasheh 1281800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1282ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1283ccd979bdSMark Fasheh 128419ece546SJan Kara ocfs2_wait_on_quotas(osb); 128519ece546SJan Kara 1286ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1287b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1288b4107950STao Ma qrec = item->lri_qrec; 1289ed460cffSJoseph Qi orphan_reco_type = item->lri_orphan_reco_type; 1290ccd979bdSMark Fasheh 1291b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1292b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1293b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1294b4107950STao Ma qrec); 1295b4107950STao Ma 1296b4107950STao Ma if (la_dinode) { 1297ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1298ccd979bdSMark Fasheh la_dinode); 1299ccd979bdSMark Fasheh if (ret < 0) 1300ccd979bdSMark Fasheh mlog_errno(ret); 1301ccd979bdSMark Fasheh 1302ccd979bdSMark Fasheh kfree(la_dinode); 1303ccd979bdSMark Fasheh } 1304ccd979bdSMark Fasheh 1305ccd979bdSMark Fasheh if (tl_dinode) { 1306ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1307ccd979bdSMark Fasheh tl_dinode); 1308ccd979bdSMark Fasheh if (ret < 0) 1309ccd979bdSMark Fasheh mlog_errno(ret); 1310ccd979bdSMark Fasheh 1311ccd979bdSMark Fasheh kfree(tl_dinode); 1312ccd979bdSMark Fasheh } 1313ccd979bdSMark Fasheh 1314ed460cffSJoseph Qi ret = ocfs2_recover_orphans(osb, item->lri_slot, 1315ed460cffSJoseph Qi orphan_reco_type); 1316ccd979bdSMark Fasheh if (ret < 0) 1317ccd979bdSMark Fasheh mlog_errno(ret); 1318ccd979bdSMark Fasheh 13192205363dSJan Kara if (qrec) { 13202205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 13212205363dSJan Kara item->lri_slot); 13222205363dSJan Kara if (ret < 0) 13232205363dSJan Kara mlog_errno(ret); 13242205363dSJan Kara /* Recovery info is already freed now */ 13252205363dSJan Kara } 13262205363dSJan Kara 1327ccd979bdSMark Fasheh kfree(item); 1328ccd979bdSMark Fasheh } 1329ccd979bdSMark Fasheh 1330b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1331ccd979bdSMark Fasheh } 1332ccd979bdSMark Fasheh 1333ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1334ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1335ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1336ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1337ccd979bdSMark Fasheh int slot_num, 1338ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 13392205363dSJan Kara struct ocfs2_dinode *tl_dinode, 1340ed460cffSJoseph Qi struct ocfs2_quota_recovery *qrec, 1341ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 1342ccd979bdSMark Fasheh { 1343ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1344ccd979bdSMark Fasheh 1345afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1346ccd979bdSMark Fasheh if (!item) { 1347ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1348ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1349ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1350ccd979bdSMark Fasheh kfree(la_dinode); 1351ccd979bdSMark Fasheh kfree(tl_dinode); 1352ccd979bdSMark Fasheh 13532205363dSJan Kara if (qrec) 13542205363dSJan Kara ocfs2_free_quota_recovery(qrec); 13552205363dSJan Kara 1356ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1357ccd979bdSMark Fasheh return; 1358ccd979bdSMark Fasheh } 1359ccd979bdSMark Fasheh 1360ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1361ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1362ccd979bdSMark Fasheh item->lri_slot = slot_num; 1363ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 13642205363dSJan Kara item->lri_qrec = qrec; 1365ed460cffSJoseph Qi item->lri_orphan_reco_type = orphan_reco_type; 1366ccd979bdSMark Fasheh 1367ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1368ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 136935ddf78eSjiangyiwen queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); 1370ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1371ccd979bdSMark Fasheh } 1372ccd979bdSMark Fasheh 1373ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 13749140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1375ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1376ccd979bdSMark Fasheh { 1377ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1378ccd979bdSMark Fasheh 137910b3dd76SSunil Mushran if (ocfs2_is_hard_readonly(osb)) 138010b3dd76SSunil Mushran return; 138110b3dd76SSunil Mushran 13829140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 13839140db04SSrinivas Eeda * that */ 13849140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 1385ed460cffSJoseph Qi osb->local_alloc_copy, NULL, NULL, 1386ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 1387ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1388ccd979bdSMark Fasheh 1389ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 13909140db04SSrinivas Eeda 13919140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 13929140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1393ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 13949140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1395ccd979bdSMark Fasheh } 1396ccd979bdSMark Fasheh 13972205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 13982205363dSJan Kara { 13992205363dSJan Kara if (osb->quota_rec) { 14002205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 14012205363dSJan Kara osb->slot_num, 14022205363dSJan Kara NULL, 14032205363dSJan Kara NULL, 1404ed460cffSJoseph Qi osb->quota_rec, 1405ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 14062205363dSJan Kara osb->quota_rec = NULL; 14072205363dSJan Kara } 14082205363dSJan Kara } 14092205363dSJan Kara 1410ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1411ccd979bdSMark Fasheh { 14122205363dSJan Kara int status, node_num, slot_num; 1413ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1414553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 14152205363dSJan Kara int *rm_quota = NULL; 14162205363dSJan Kara int rm_quota_used = 0, i; 14172205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1418ccd979bdSMark Fasheh 141921158ca8SGuozhonghua /* Whether the quota supported. */ 142021158ca8SGuozhonghua int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 142121158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_USRQUOTA) 142221158ca8SGuozhonghua || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 142321158ca8SGuozhonghua OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); 142421158ca8SGuozhonghua 1425ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1426ccd979bdSMark Fasheh if (status < 0) { 1427ccd979bdSMark Fasheh goto bail; 1428ccd979bdSMark Fasheh } 1429ccd979bdSMark Fasheh 143021158ca8SGuozhonghua if (quota_enabled) { 14316396bb22SKees Cook rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); 14322205363dSJan Kara if (!rm_quota) { 14332205363dSJan Kara status = -ENOMEM; 14342205363dSJan Kara goto bail; 14352205363dSJan Kara } 143621158ca8SGuozhonghua } 1437ccd979bdSMark Fasheh restart: 1438ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1439ccd979bdSMark Fasheh if (status < 0) { 1440ccd979bdSMark Fasheh mlog_errno(status); 1441ccd979bdSMark Fasheh goto bail; 1442ccd979bdSMark Fasheh } 1443ccd979bdSMark Fasheh 14449140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 14459140db04SSrinivas Eeda if (status < 0) 14469140db04SSrinivas Eeda mlog_errno(status); 14479140db04SSrinivas Eeda 14489140db04SSrinivas Eeda /* queue recovery for our own slot */ 14499140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1450ed460cffSJoseph Qi NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); 14519140db04SSrinivas Eeda 1452553abd04SJoel Becker spin_lock(&osb->osb_lock); 1453553abd04SJoel Becker while (rm->rm_used) { 1454553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1455553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1456553abd04SJoel Becker node_num = rm->rm_entries[0]; 1457553abd04SJoel Becker spin_unlock(&osb->osb_lock); 14582205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1459b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 14602205363dSJan Kara if (slot_num == -ENOENT) { 14612205363dSJan Kara status = 0; 14622205363dSJan Kara goto skip_recovery; 14632205363dSJan Kara } 1464ccd979bdSMark Fasheh 14652205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 14662205363dSJan Kara * immediately because we have to obtain cluster locks from 14672205363dSJan Kara * quota files and we also don't want to just skip it because 14682205363dSJan Kara * then quota usage would be out of sync until some node takes 14692205363dSJan Kara * the slot. So we remember which nodes need quota recovery 14702205363dSJan Kara * and when everything else is done, we recover quotas. */ 147121158ca8SGuozhonghua if (quota_enabled) { 147221158ca8SGuozhonghua for (i = 0; i < rm_quota_used 147321158ca8SGuozhonghua && rm_quota[i] != slot_num; i++) 147421158ca8SGuozhonghua ; 147521158ca8SGuozhonghua 14762205363dSJan Kara if (i == rm_quota_used) 14772205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 147821158ca8SGuozhonghua } 14792205363dSJan Kara 14802205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 14812205363dSJan Kara skip_recovery: 1482553abd04SJoel Becker if (!status) { 1483553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1484553abd04SJoel Becker } else { 1485ccd979bdSMark Fasheh mlog(ML_ERROR, 1486ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1487ccd979bdSMark Fasheh status, node_num, 1488ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1489ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1490ccd979bdSMark Fasheh } 1491ccd979bdSMark Fasheh 1492553abd04SJoel Becker spin_lock(&osb->osb_lock); 1493ccd979bdSMark Fasheh } 1494553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1495b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1496553abd04SJoel Becker 1497539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1498539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1499539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1500539d8264SSunil Mushran if (status < 0) 1501539d8264SSunil Mushran mlog_errno(status); 1502539d8264SSunil Mushran 15032205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 15042205363dSJan Kara * superblock lock so that no one can start using the slot (and crash) 15052205363dSJan Kara * before we recover it */ 150621158ca8SGuozhonghua if (quota_enabled) { 15072205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 15082205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 15092205363dSJan Kara if (IS_ERR(qrec)) { 15102205363dSJan Kara status = PTR_ERR(qrec); 15112205363dSJan Kara mlog_errno(status); 15122205363dSJan Kara continue; 15132205363dSJan Kara } 151421158ca8SGuozhonghua ocfs2_queue_recovery_completion(osb->journal, 151521158ca8SGuozhonghua rm_quota[i], 1516ed460cffSJoseph Qi NULL, NULL, qrec, 1517ed460cffSJoseph Qi ORPHAN_NEED_TRUNCATE); 15182205363dSJan Kara } 151921158ca8SGuozhonghua } 15202205363dSJan Kara 1521ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1522ccd979bdSMark Fasheh 15239140db04SSrinivas Eeda /* queue recovery for offline slots */ 1524ed460cffSJoseph Qi ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1525ccd979bdSMark Fasheh 1526ccd979bdSMark Fasheh bail: 1527c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1528553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1529c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1530ccd979bdSMark Fasheh goto restart; 1531ccd979bdSMark Fasheh } 1532ccd979bdSMark Fasheh 15339140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1534ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1535ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1536ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1537ccd979bdSMark Fasheh 1538c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1539ccd979bdSMark Fasheh 154021158ca8SGuozhonghua if (quota_enabled) 15412205363dSJan Kara kfree(rm_quota); 15422205363dSJan Kara 1543111e7049SEric W. Biederman return status; 1544ccd979bdSMark Fasheh } 1545ccd979bdSMark Fasheh 1546ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1547ccd979bdSMark Fasheh { 1548c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1549b4107950STao Ma 1550b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1551b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1552b4107950STao Ma osb->disable_recovery ? 1553b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1554b4107950STao Ma 1555ccd979bdSMark Fasheh if (osb->disable_recovery) 1556ccd979bdSMark Fasheh goto out; 1557ccd979bdSMark Fasheh 1558ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1559ccd979bdSMark Fasheh goto out; 1560ccd979bdSMark Fasheh 1561ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 15625afc44e2SJoseph Qi "ocfs2rec-%s", osb->uuid_str); 1563ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1564ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1565ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1566ccd979bdSMark Fasheh } 1567ccd979bdSMark Fasheh 1568ccd979bdSMark Fasheh out: 1569c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1570ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1571ccd979bdSMark Fasheh } 1572ccd979bdSMark Fasheh 1573539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1574539d8264SSunil Mushran int slot_num, 1575539d8264SSunil Mushran struct buffer_head **bh, 1576539d8264SSunil Mushran struct inode **ret_inode) 1577539d8264SSunil Mushran { 1578539d8264SSunil Mushran int status = -EACCES; 1579539d8264SSunil Mushran struct inode *inode = NULL; 1580539d8264SSunil Mushran 1581539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1582539d8264SSunil Mushran 1583539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1584539d8264SSunil Mushran slot_num); 1585539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1586539d8264SSunil Mushran mlog_errno(status); 1587539d8264SSunil Mushran goto bail; 1588539d8264SSunil Mushran } 1589539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1590539d8264SSunil Mushran 1591b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1592539d8264SSunil Mushran if (status < 0) { 1593539d8264SSunil Mushran mlog_errno(status); 1594539d8264SSunil Mushran goto bail; 1595539d8264SSunil Mushran } 1596539d8264SSunil Mushran 1597539d8264SSunil Mushran status = 0; 1598539d8264SSunil Mushran 1599539d8264SSunil Mushran bail: 1600539d8264SSunil Mushran if (inode) { 1601539d8264SSunil Mushran if (status || !ret_inode) 1602539d8264SSunil Mushran iput(inode); 1603539d8264SSunil Mushran else 1604539d8264SSunil Mushran *ret_inode = inode; 1605539d8264SSunil Mushran } 1606539d8264SSunil Mushran return status; 1607539d8264SSunil Mushran } 1608539d8264SSunil Mushran 1609ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1610ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1611ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1612ccd979bdSMark Fasheh int node_num, 1613ccd979bdSMark Fasheh int slot_num) 1614ccd979bdSMark Fasheh { 1615ccd979bdSMark Fasheh int status; 1616ccd979bdSMark Fasheh int got_lock = 0; 1617ccd979bdSMark Fasheh unsigned int flags; 1618ccd979bdSMark Fasheh struct inode *inode = NULL; 1619ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1620ccd979bdSMark Fasheh journal_t *journal = NULL; 1621ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1622539d8264SSunil Mushran u32 slot_reco_gen; 1623ccd979bdSMark Fasheh 1624539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1625539d8264SSunil Mushran if (status) { 1626ccd979bdSMark Fasheh mlog_errno(status); 1627ccd979bdSMark Fasheh goto done; 1628ccd979bdSMark Fasheh } 1629539d8264SSunil Mushran 1630539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1631539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1632539d8264SSunil Mushran brelse(bh); 1633539d8264SSunil Mushran bh = NULL; 1634539d8264SSunil Mushran 1635539d8264SSunil Mushran /* 1636539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1637539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1638539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1639539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1640539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1641539d8264SSunil Mushran * If not, it needs recovery. 1642539d8264SSunil Mushran */ 1643539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1644b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1645539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1646539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1647539d8264SSunil Mushran status = -EBUSY; 1648ccd979bdSMark Fasheh goto done; 1649ccd979bdSMark Fasheh } 1650539d8264SSunil Mushran 1651539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1652ccd979bdSMark Fasheh 1653e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1654ccd979bdSMark Fasheh if (status < 0) { 1655b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1656ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1657ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1658ccd979bdSMark Fasheh goto done; 1659ccd979bdSMark Fasheh } 1660ccd979bdSMark Fasheh got_lock = 1; 1661ccd979bdSMark Fasheh 1662ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1663ccd979bdSMark Fasheh 1664ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1665539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1666ccd979bdSMark Fasheh 1667ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1668b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1669539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1670539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1671ccd979bdSMark Fasheh goto done; 1672ccd979bdSMark Fasheh } 1673ccd979bdSMark Fasheh 16749140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 16759140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 16769140db04SSrinivas Eeda 1677619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1678619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1679619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1680ccd979bdSMark Fasheh 1681ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1682ccd979bdSMark Fasheh 1683ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1684ccd979bdSMark Fasheh if (status < 0) { 1685ccd979bdSMark Fasheh mlog_errno(status); 1686ccd979bdSMark Fasheh goto done; 1687ccd979bdSMark Fasheh } 1688ccd979bdSMark Fasheh 16892b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1690*8e6cf5fbSZhang Yi if (IS_ERR(journal)) { 1691ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1692*8e6cf5fbSZhang Yi status = PTR_ERR(journal); 1693ccd979bdSMark Fasheh goto done; 1694ccd979bdSMark Fasheh } 1695ccd979bdSMark Fasheh 16962b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1697ccd979bdSMark Fasheh if (status < 0) { 1698ccd979bdSMark Fasheh mlog_errno(status); 1699783cc68dSZhang Mingyu BUG_ON(!igrab(inode)); 17002b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1701ccd979bdSMark Fasheh goto done; 1702ccd979bdSMark Fasheh } 1703ccd979bdSMark Fasheh 1704ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1705ccd979bdSMark Fasheh 1706ccd979bdSMark Fasheh /* wipe the journal */ 17072b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 170801d5d965SLeah Rumancik status = jbd2_journal_flush(journal, 0); 17092b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1710ccd979bdSMark Fasheh if (status < 0) 1711ccd979bdSMark Fasheh mlog_errno(status); 1712ccd979bdSMark Fasheh 1713ccd979bdSMark Fasheh /* This will mark the node clean */ 1714ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1715ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1716ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1717ccd979bdSMark Fasheh 1718539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1719539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1720539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1721539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1722539d8264SSunil Mushran 172313723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 17248cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1725ccd979bdSMark Fasheh if (status < 0) 1726ccd979bdSMark Fasheh mlog_errno(status); 1727ccd979bdSMark Fasheh 1728783cc68dSZhang Mingyu BUG_ON(!igrab(inode)); 1729ccd979bdSMark Fasheh 17302b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1731ccd979bdSMark Fasheh 1732619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1733619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1734619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1735ccd979bdSMark Fasheh done: 1736ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1737ccd979bdSMark Fasheh if (got_lock) 1738e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1739ccd979bdSMark Fasheh 1740ccd979bdSMark Fasheh iput(inode); 1741ccd979bdSMark Fasheh brelse(bh); 1742ccd979bdSMark Fasheh 1743ccd979bdSMark Fasheh return status; 1744ccd979bdSMark Fasheh } 1745ccd979bdSMark Fasheh 1746ccd979bdSMark Fasheh /* 1747ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1748ccd979bdSMark Fasheh * - Replay it's journal 1749ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1750ccd979bdSMark Fasheh * - Stamp a clean truncate log 1751ccd979bdSMark Fasheh * - Mark the node clean 1752ccd979bdSMark Fasheh * 1753ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1754ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1755ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1756ccd979bdSMark Fasheh * far less concerning. 1757ccd979bdSMark Fasheh */ 1758ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 17592205363dSJan Kara int node_num, int slot_num) 1760ccd979bdSMark Fasheh { 1761ccd979bdSMark Fasheh int status = 0; 1762ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1763ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1764ccd979bdSMark Fasheh 1765b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1766ccd979bdSMark Fasheh 1767ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1768ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1769ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1770ccd979bdSMark Fasheh 1771ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1772ccd979bdSMark Fasheh if (status < 0) { 1773539d8264SSunil Mushran if (status == -EBUSY) { 1774b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1775539d8264SSunil Mushran status = 0; 1776539d8264SSunil Mushran goto done; 1777539d8264SSunil Mushran } 1778ccd979bdSMark Fasheh mlog_errno(status); 1779ccd979bdSMark Fasheh goto done; 1780ccd979bdSMark Fasheh } 1781ccd979bdSMark Fasheh 1782ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1783ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1784ccd979bdSMark Fasheh if (status < 0) { 1785ccd979bdSMark Fasheh mlog_errno(status); 1786ccd979bdSMark Fasheh goto done; 1787ccd979bdSMark Fasheh } 1788ccd979bdSMark Fasheh 1789ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1790ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1791ccd979bdSMark Fasheh * recovery. */ 1792ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1793ccd979bdSMark Fasheh if (status < 0) 1794ccd979bdSMark Fasheh mlog_errno(status); 1795ccd979bdSMark Fasheh 1796ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1797ccd979bdSMark Fasheh * harmful place to get an error... */ 17988e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1799ccd979bdSMark Fasheh if (status < 0) 1800ccd979bdSMark Fasheh mlog_errno(status); 1801ccd979bdSMark Fasheh 1802ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1803ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1804ed460cffSJoseph Qi tl_copy, NULL, ORPHAN_NEED_TRUNCATE); 1805ccd979bdSMark Fasheh 1806ccd979bdSMark Fasheh status = 0; 1807ccd979bdSMark Fasheh done: 1808ccd979bdSMark Fasheh 1809ccd979bdSMark Fasheh return status; 1810ccd979bdSMark Fasheh } 1811ccd979bdSMark Fasheh 1812ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1813ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1814ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1815ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1816ccd979bdSMark Fasheh int slot_num) 1817ccd979bdSMark Fasheh { 1818ccd979bdSMark Fasheh int status, flags; 1819ccd979bdSMark Fasheh struct inode *inode = NULL; 1820ccd979bdSMark Fasheh 1821ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1822ccd979bdSMark Fasheh slot_num); 1823ccd979bdSMark Fasheh if (inode == NULL) { 1824ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1825ccd979bdSMark Fasheh status = -EACCES; 1826ccd979bdSMark Fasheh goto bail; 1827ccd979bdSMark Fasheh } 1828ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1829ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1830ccd979bdSMark Fasheh iput(inode); 1831ccd979bdSMark Fasheh inode = NULL; 1832ccd979bdSMark Fasheh status = -EACCES; 1833ccd979bdSMark Fasheh goto bail; 1834ccd979bdSMark Fasheh } 1835ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1836ccd979bdSMark Fasheh 1837ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1838e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1839ccd979bdSMark Fasheh if (status < 0) { 1840ccd979bdSMark Fasheh if (status != -EAGAIN) 1841ccd979bdSMark Fasheh mlog_errno(status); 1842ccd979bdSMark Fasheh goto bail; 1843ccd979bdSMark Fasheh } 1844ccd979bdSMark Fasheh 1845e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1846ccd979bdSMark Fasheh bail: 1847ccd979bdSMark Fasheh iput(inode); 1848ccd979bdSMark Fasheh 1849ccd979bdSMark Fasheh return status; 1850ccd979bdSMark Fasheh } 1851ccd979bdSMark Fasheh 1852ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1853ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1854ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1855ccd979bdSMark Fasheh { 1856d85b20e4SJoel Becker unsigned int node_num; 1857d85b20e4SJoel Becker int status, i; 1858a1af7d15SMark Fasheh u32 gen; 1859539d8264SSunil Mushran struct buffer_head *bh = NULL; 1860539d8264SSunil Mushran struct ocfs2_dinode *di; 1861ccd979bdSMark Fasheh 1862ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1863ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1864ccd979bdSMark Fasheh 1865d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1866539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1867539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1868539d8264SSunil Mushran if (status) { 1869539d8264SSunil Mushran mlog_errno(status); 1870539d8264SSunil Mushran goto bail; 1871539d8264SSunil Mushran } 1872539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1873a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1874539d8264SSunil Mushran brelse(bh); 1875539d8264SSunil Mushran bh = NULL; 1876539d8264SSunil Mushran 1877a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1878a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1879a1af7d15SMark Fasheh 1880b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1881539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1882539d8264SSunil Mushran 1883a1af7d15SMark Fasheh if (i == osb->slot_num) { 1884a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1885ccd979bdSMark Fasheh continue; 1886a1af7d15SMark Fasheh } 1887d85b20e4SJoel Becker 1888d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1889a1af7d15SMark Fasheh if (status == -ENOENT) { 1890a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1891ccd979bdSMark Fasheh continue; 1892a1af7d15SMark Fasheh } 1893ccd979bdSMark Fasheh 1894a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1895a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1896ccd979bdSMark Fasheh continue; 1897a1af7d15SMark Fasheh } 1898d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1899ccd979bdSMark Fasheh 1900ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1901ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1902ccd979bdSMark Fasheh * file here to test if he's alive. */ 1903ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1904ccd979bdSMark Fasheh if (!status) { 1905ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1906ccd979bdSMark Fasheh * the recovery thread can't race us on 1907ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1908ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1909ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1910ccd979bdSMark Fasheh mlog_errno(status); 1911ccd979bdSMark Fasheh goto bail; 1912ccd979bdSMark Fasheh } 1913ccd979bdSMark Fasheh } 1914ccd979bdSMark Fasheh 1915ccd979bdSMark Fasheh status = 0; 1916ccd979bdSMark Fasheh bail: 1917ccd979bdSMark Fasheh return status; 1918ccd979bdSMark Fasheh } 1919ccd979bdSMark Fasheh 192083273932SSrinivas Eeda /* 192183273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 192283273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 192383273932SSrinivas Eeda * same time. 192483273932SSrinivas Eeda */ 192583273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 192683273932SSrinivas Eeda { 192783273932SSrinivas Eeda unsigned long time; 192883273932SSrinivas Eeda 192983273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 193083273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 193183273932SSrinivas Eeda return msecs_to_jiffies(time); 193283273932SSrinivas Eeda } 193383273932SSrinivas Eeda 193483273932SSrinivas Eeda /* 193583273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 193683273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 193783273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 193883273932SSrinivas Eeda * 1939a035bff6SSunil Mushran * It scans all slots, even ones that are in use. It does so to handle the 1940a035bff6SSunil Mushran * case described below: 1941a035bff6SSunil Mushran * 1942a035bff6SSunil Mushran * Node 1 has an inode it was using. The dentry went away due to memory 1943a035bff6SSunil Mushran * pressure. Node 1 closes the inode, but it's on the free list. The node 1944a035bff6SSunil Mushran * has the open lock. 1945a035bff6SSunil Mushran * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1946a035bff6SSunil Mushran * but node 1 has no dentry and doesn't get the message. It trylocks the 1947a035bff6SSunil Mushran * open lock, sees that another node has a PR, and does nothing. 1948a035bff6SSunil Mushran * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1949a035bff6SSunil Mushran * open lock, sees the PR still, and does nothing. 1950a035bff6SSunil Mushran * Basically, we have to trigger an orphan iput on node 1. The only way 1951a035bff6SSunil Mushran * for this to happen is if node 1 runs node 2's orphan dir. 1952a035bff6SSunil Mushran * 195383273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 195483273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 195583273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 195683273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 195783273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 195883273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 195983273932SSrinivas Eeda * sequence number in the LVB. 196083273932SSrinivas Eeda */ 1961b519ea6dSJoseph Qi static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 196283273932SSrinivas Eeda { 196383273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 196483273932SSrinivas Eeda int status, i; 196583273932SSrinivas Eeda u32 seqno = 0; 196683273932SSrinivas Eeda 196783273932SSrinivas Eeda os = &osb->osb_orphan_scan; 196883273932SSrinivas Eeda 1969692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1970692684e1SSunil Mushran goto out; 1971692684e1SSunil Mushran 1972b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1973b4107950STao Ma atomic_read(&os->os_state)); 1974b4107950STao Ma 1975df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 197683273932SSrinivas Eeda if (status < 0) { 197783273932SSrinivas Eeda if (status != -EAGAIN) 197883273932SSrinivas Eeda mlog_errno(status); 197983273932SSrinivas Eeda goto out; 198083273932SSrinivas Eeda } 198183273932SSrinivas Eeda 1982692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1983692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1984692684e1SSunil Mushran goto unlock; 1985692684e1SSunil Mushran 198683273932SSrinivas Eeda if (os->os_seqno != seqno) { 198783273932SSrinivas Eeda os->os_seqno = seqno; 198883273932SSrinivas Eeda goto unlock; 198983273932SSrinivas Eeda } 199083273932SSrinivas Eeda 199183273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 199283273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1993ed460cffSJoseph Qi NULL, ORPHAN_NO_NEED_TRUNCATE); 199483273932SSrinivas Eeda /* 199583273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 199683273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 199783273932SSrinivas Eeda */ 199883273932SSrinivas Eeda seqno++; 199915633a22SSrinivas Eeda os->os_count++; 2000395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 200183273932SSrinivas Eeda unlock: 2002df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 200383273932SSrinivas Eeda out: 2004b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 2005b4107950STao Ma atomic_read(&os->os_state)); 200683273932SSrinivas Eeda return; 200783273932SSrinivas Eeda } 200883273932SSrinivas Eeda 200983273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 2010b519ea6dSJoseph Qi static void ocfs2_orphan_scan_work(struct work_struct *work) 201183273932SSrinivas Eeda { 201283273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 201383273932SSrinivas Eeda struct ocfs2_super *osb; 201483273932SSrinivas Eeda 201583273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 201683273932SSrinivas Eeda os_orphan_scan_work.work); 201783273932SSrinivas Eeda osb = os->os_osb; 201883273932SSrinivas Eeda 201983273932SSrinivas Eeda mutex_lock(&os->os_lock); 202083273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 2021692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 202235ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 202383273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 202483273932SSrinivas Eeda mutex_unlock(&os->os_lock); 202583273932SSrinivas Eeda } 202683273932SSrinivas Eeda 202783273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 202883273932SSrinivas Eeda { 202983273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 203083273932SSrinivas Eeda 203183273932SSrinivas Eeda os = &osb->osb_orphan_scan; 2032df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 2033692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 203483273932SSrinivas Eeda mutex_lock(&os->os_lock); 203583273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 203683273932SSrinivas Eeda mutex_unlock(&os->os_lock); 203783273932SSrinivas Eeda } 2038df152c24SSunil Mushran } 203983273932SSrinivas Eeda 2040df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 204183273932SSrinivas Eeda { 204283273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 204383273932SSrinivas Eeda 204483273932SSrinivas Eeda os = &osb->osb_orphan_scan; 204583273932SSrinivas Eeda os->os_osb = osb; 204615633a22SSrinivas Eeda os->os_count = 0; 20473211949fSSunil Mushran os->os_seqno = 0; 204883273932SSrinivas Eeda mutex_init(&os->os_lock); 2049df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 20508b712cd5SJeff Mahoney } 205183273932SSrinivas Eeda 20528b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 20538b712cd5SJeff Mahoney { 20548b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 20558b712cd5SJeff Mahoney 20568b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 2057395627b0SDeepa Dinamani os->os_scantime = ktime_get_seconds(); 2058df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 2059df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 2060df152c24SSunil Mushran else { 2061df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 206235ddf78eSjiangyiwen queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 206383273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 2064df152c24SSunil Mushran } 206583273932SSrinivas Eeda } 206683273932SSrinivas Eeda 20675eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 20683704412bSAl Viro struct dir_context ctx; 20695eae5b96SMark Fasheh struct inode *head; 20705eae5b96SMark Fasheh struct ocfs2_super *osb; 207130edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type; 20725eae5b96SMark Fasheh }; 20735eae5b96SMark Fasheh 207425885a35SAl Viro static bool ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, 2075ac7576f4SMiklos Szeredi int name_len, loff_t pos, u64 ino, 2076ac7576f4SMiklos Szeredi unsigned type) 20775eae5b96SMark Fasheh { 2078ac7576f4SMiklos Szeredi struct ocfs2_orphan_filldir_priv *p = 2079ac7576f4SMiklos Szeredi container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); 20805eae5b96SMark Fasheh struct inode *iter; 20815eae5b96SMark Fasheh 20825eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 208325885a35SAl Viro return true; 20845eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 208525885a35SAl Viro return true; 20865eae5b96SMark Fasheh 208730edc43cSJoseph Qi /* do not include dio entry in case of orphan scan */ 208830edc43cSJoseph Qi if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && 208930edc43cSJoseph Qi (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 209030edc43cSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN))) 209125885a35SAl Viro return true; 209230edc43cSJoseph Qi 20935eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 20945eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 20955fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 20965eae5b96SMark Fasheh if (IS_ERR(iter)) 209725885a35SAl Viro return true; 20985eae5b96SMark Fasheh 209993d911fcSJoseph Qi if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 210093d911fcSJoseph Qi OCFS2_DIO_ORPHAN_PREFIX_LEN)) 210193d911fcSJoseph Qi OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; 210293d911fcSJoseph Qi 2103ed460cffSJoseph Qi /* Skip inodes which are already added to recover list, since dio may 2104ed460cffSJoseph Qi * happen concurrently with unlink/rename */ 2105ed460cffSJoseph Qi if (OCFS2_I(iter)->ip_next_orphan) { 2106ed460cffSJoseph Qi iput(iter); 210725885a35SAl Viro return true; 2108ed460cffSJoseph Qi } 2109ed460cffSJoseph Qi 2110b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 21115eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 21125eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 21135eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 21145eae5b96SMark Fasheh p->head = iter; 21155eae5b96SMark Fasheh 211625885a35SAl Viro return true; 21175eae5b96SMark Fasheh } 21185eae5b96SMark Fasheh 2119b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2120b4df6ed8SMark Fasheh int slot, 212130edc43cSJoseph Qi struct inode **head, 212230edc43cSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2123ccd979bdSMark Fasheh { 2124b4df6ed8SMark Fasheh int status; 2125ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 21263704412bSAl Viro struct ocfs2_orphan_filldir_priv priv = { 21273704412bSAl Viro .ctx.actor = ocfs2_orphan_filldir, 21283704412bSAl Viro .osb = osb, 212930edc43cSJoseph Qi .head = *head, 213030edc43cSJoseph Qi .orphan_reco_type = orphan_reco_type 21313704412bSAl Viro }; 2132ccd979bdSMark Fasheh 2133ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2134ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 2135ccd979bdSMark Fasheh slot); 2136ccd979bdSMark Fasheh if (!orphan_dir_inode) { 2137ccd979bdSMark Fasheh status = -ENOENT; 2138ccd979bdSMark Fasheh mlog_errno(status); 2139b4df6ed8SMark Fasheh return status; 2140ccd979bdSMark Fasheh } 2141ccd979bdSMark Fasheh 21425955102cSAl Viro inode_lock(orphan_dir_inode); 2143e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2144ccd979bdSMark Fasheh if (status < 0) { 2145ccd979bdSMark Fasheh mlog_errno(status); 2146ccd979bdSMark Fasheh goto out; 2147ccd979bdSMark Fasheh } 2148ccd979bdSMark Fasheh 21493704412bSAl Viro status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 21505eae5b96SMark Fasheh if (status) { 2151ccd979bdSMark Fasheh mlog_errno(status); 2152a86370fbSMark Fasheh goto out_cluster; 2153ccd979bdSMark Fasheh } 2154ccd979bdSMark Fasheh 21555eae5b96SMark Fasheh *head = priv.head; 2156ccd979bdSMark Fasheh 2157a86370fbSMark Fasheh out_cluster: 2158e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 2159b4df6ed8SMark Fasheh out: 21605955102cSAl Viro inode_unlock(orphan_dir_inode); 2161ccd979bdSMark Fasheh iput(orphan_dir_inode); 2162b4df6ed8SMark Fasheh return status; 2163b4df6ed8SMark Fasheh } 2164b4df6ed8SMark Fasheh 2165b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2166b4df6ed8SMark Fasheh int slot) 2167b4df6ed8SMark Fasheh { 2168b4df6ed8SMark Fasheh int ret; 2169b4df6ed8SMark Fasheh 2170b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2171b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2172b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2173b4df6ed8SMark Fasheh return ret; 2174b4df6ed8SMark Fasheh } 2175b4df6ed8SMark Fasheh 2176b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2177b4df6ed8SMark Fasheh int slot) 2178b4df6ed8SMark Fasheh { 2179b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2180b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2181b4df6ed8SMark Fasheh * know to quit early. */ 2182b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2183b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2184b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2185b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2186b4df6ed8SMark Fasheh * them. */ 2187b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2188b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2189b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2190b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2191b4df6ed8SMark Fasheh } 2192b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2193b4df6ed8SMark Fasheh } 2194b4df6ed8SMark Fasheh 2195b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2196b4df6ed8SMark Fasheh int slot) 2197b4df6ed8SMark Fasheh { 2198b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2199b4df6ed8SMark Fasheh } 2200b4df6ed8SMark Fasheh 2201b4df6ed8SMark Fasheh /* 2202b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2203b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2204b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2205b4df6ed8SMark Fasheh * (most) of the rest of the work. 2206b4df6ed8SMark Fasheh * 2207b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2208b4df6ed8SMark Fasheh * couple of extra considerations. 2209b4df6ed8SMark Fasheh * 2210b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2211b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2212b4df6ed8SMark Fasheh * an invalid inode. 2213b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2214b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2215b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2216b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2217b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2218b4df6ed8SMark Fasheh */ 2219b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2220ed460cffSJoseph Qi int slot, 2221ed460cffSJoseph Qi enum ocfs2_orphan_reco_type orphan_reco_type) 2222b4df6ed8SMark Fasheh { 2223b4df6ed8SMark Fasheh int ret = 0; 2224b4df6ed8SMark Fasheh struct inode *inode = NULL; 2225b4df6ed8SMark Fasheh struct inode *iter; 2226b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2227cf1776a9SJoseph Qi struct buffer_head *di_bh = NULL; 2228cf1776a9SJoseph Qi struct ocfs2_dinode *di = NULL; 2229b4df6ed8SMark Fasheh 2230b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2231b4df6ed8SMark Fasheh 2232b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 223330edc43cSJoseph Qi ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); 2234b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2235b4df6ed8SMark Fasheh 2236b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2237b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2238b4df6ed8SMark Fasheh if (ret) 2239b4df6ed8SMark Fasheh mlog_errno(ret); 2240ccd979bdSMark Fasheh 2241ccd979bdSMark Fasheh while (inode) { 2242ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2243b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2244b4107950STao Ma (unsigned long long)oi->ip_blkno); 2245ccd979bdSMark Fasheh 2246ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2247ed460cffSJoseph Qi oi->ip_next_orphan = NULL; 2248ccd979bdSMark Fasheh 224993d911fcSJoseph Qi if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { 22505955102cSAl Viro inode_lock(inode); 2251cf1776a9SJoseph Qi ret = ocfs2_rw_lock(inode, 1); 2252cf1776a9SJoseph Qi if (ret < 0) { 2253cf1776a9SJoseph Qi mlog_errno(ret); 225493d911fcSJoseph Qi goto unlock_mutex; 2255cf1776a9SJoseph Qi } 2256ed460cffSJoseph Qi /* 2257ed460cffSJoseph Qi * We need to take and drop the inode lock to 2258ed460cffSJoseph Qi * force read inode from disk. 2259ed460cffSJoseph Qi */ 2260cf1776a9SJoseph Qi ret = ocfs2_inode_lock(inode, &di_bh, 1); 2261ed460cffSJoseph Qi if (ret) { 2262ed460cffSJoseph Qi mlog_errno(ret); 2263cf1776a9SJoseph Qi goto unlock_rw; 2264ed460cffSJoseph Qi } 2265cf1776a9SJoseph Qi 2266cf1776a9SJoseph Qi di = (struct ocfs2_dinode *)di_bh->b_data; 2267ed460cffSJoseph Qi 226893d911fcSJoseph Qi if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { 2269ed460cffSJoseph Qi ret = ocfs2_truncate_file(inode, di_bh, 2270ed460cffSJoseph Qi i_size_read(inode)); 2271ed460cffSJoseph Qi if (ret < 0) { 2272ed460cffSJoseph Qi if (ret != -ENOSPC) 2273ed460cffSJoseph Qi mlog_errno(ret); 2274cf1776a9SJoseph Qi goto unlock_inode; 2275ed460cffSJoseph Qi } 2276ed460cffSJoseph Qi 227793d911fcSJoseph Qi ret = ocfs2_del_inode_from_orphan(osb, inode, 227893d911fcSJoseph Qi di_bh, 0, 0); 2279ed460cffSJoseph Qi if (ret) 2280ed460cffSJoseph Qi mlog_errno(ret); 228193d911fcSJoseph Qi } 2282cf1776a9SJoseph Qi unlock_inode: 2283cf1776a9SJoseph Qi ocfs2_inode_unlock(inode, 1); 2284512f62acSJoseph Qi brelse(di_bh); 2285512f62acSJoseph Qi di_bh = NULL; 2286cf1776a9SJoseph Qi unlock_rw: 2287cf1776a9SJoseph Qi ocfs2_rw_unlock(inode, 1); 228893d911fcSJoseph Qi unlock_mutex: 22895955102cSAl Viro inode_unlock(inode); 229093d911fcSJoseph Qi 229193d911fcSJoseph Qi /* clear dio flag in ocfs2_inode_info */ 229293d911fcSJoseph Qi oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; 229393d911fcSJoseph Qi } else { 229493d911fcSJoseph Qi spin_lock(&oi->ip_lock); 229593d911fcSJoseph Qi /* Set the proper information to get us going into 229693d911fcSJoseph Qi * ocfs2_delete_inode. */ 229793d911fcSJoseph Qi oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 229893d911fcSJoseph Qi spin_unlock(&oi->ip_lock); 229993d911fcSJoseph Qi } 230093d911fcSJoseph Qi 2301ccd979bdSMark Fasheh iput(inode); 2302ccd979bdSMark Fasheh inode = iter; 2303ccd979bdSMark Fasheh } 2304ccd979bdSMark Fasheh 2305b4df6ed8SMark Fasheh return ret; 2306ccd979bdSMark Fasheh } 2307ccd979bdSMark Fasheh 230819ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2309ccd979bdSMark Fasheh { 2310ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2311ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2312ccd979bdSMark Fasheh * or DISABLED. */ 2313ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 231419ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 231519ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2316ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2317ccd979bdSMark Fasheh 2318ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2319ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2320ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2321ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2322b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2323ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2324ccd979bdSMark Fasheh return -EBUSY; 2325ccd979bdSMark Fasheh } 2326ccd979bdSMark Fasheh 2327ccd979bdSMark Fasheh return 0; 2328ccd979bdSMark Fasheh } 2329ccd979bdSMark Fasheh 2330ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2331ccd979bdSMark Fasheh { 2332ccd979bdSMark Fasheh int status; 2333ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2334ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2335ccd979bdSMark Fasheh 2336ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2337ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2338ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2339ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2340ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2341ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2342ccd979bdSMark Fasheh 2343745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2344ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2345745ae8baSMark Fasheh || kthread_should_stop()); 2346ccd979bdSMark Fasheh 2347ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 234855b465b6SJoseph Qi if (status < 0) { 234955b465b6SJoseph Qi static unsigned long abort_warn_time; 235055b465b6SJoseph Qi 235155b465b6SJoseph Qi /* Warn about this once per minute */ 235255b465b6SJoseph Qi if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 235355b465b6SJoseph Qi mlog(ML_ERROR, "status = %d, journal is " 235455b465b6SJoseph Qi "already aborted.\n", status); 235555b465b6SJoseph Qi /* 235655b465b6SJoseph Qi * After ocfs2_commit_cache() fails, j_num_trans has a 235755b465b6SJoseph Qi * non-zero value. Sleep here to avoid a busy-wait 235855b465b6SJoseph Qi * loop. 235955b465b6SJoseph Qi */ 236055b465b6SJoseph Qi msleep_interruptible(1000); 236155b465b6SJoseph Qi } 2362ccd979bdSMark Fasheh 2363ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2364ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2365ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2366ccd979bdSMark Fasheh "shutdown\n", 2367ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2368ccd979bdSMark Fasheh } 2369ccd979bdSMark Fasheh } 2370ccd979bdSMark Fasheh 2371ccd979bdSMark Fasheh return 0; 2372ccd979bdSMark Fasheh } 2373ccd979bdSMark Fasheh 2374539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2375539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2376539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2377539d8264SSunil Mushran * a journal has been recovered by another node. 2378539d8264SSunil Mushran */ 2379ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2380ccd979bdSMark Fasheh { 2381ccd979bdSMark Fasheh int ret = 0; 2382ccd979bdSMark Fasheh unsigned int slot; 2383539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2384ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2385539d8264SSunil Mushran int journal_dirty = 0; 2386ccd979bdSMark Fasheh 2387ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2388539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2389539d8264SSunil Mushran if (ret) { 2390ccd979bdSMark Fasheh mlog_errno(ret); 2391ccd979bdSMark Fasheh goto out; 2392ccd979bdSMark Fasheh } 2393ccd979bdSMark Fasheh 2394ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2395ccd979bdSMark Fasheh 2396539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2397539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2398539d8264SSunil Mushran 2399ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2400ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2401539d8264SSunil Mushran journal_dirty = 1; 2402ccd979bdSMark Fasheh 2403ccd979bdSMark Fasheh brelse(di_bh); 2404539d8264SSunil Mushran di_bh = NULL; 2405ccd979bdSMark Fasheh } 2406ccd979bdSMark Fasheh 2407ccd979bdSMark Fasheh out: 2408539d8264SSunil Mushran if (journal_dirty) 2409539d8264SSunil Mushran ret = -EROFS; 2410ccd979bdSMark Fasheh return ret; 2411ccd979bdSMark Fasheh } 2412