1ccd979bdSMark Fasheh /* -*- mode: c; c-basic-offset: 8; -*- 2ccd979bdSMark Fasheh * vim: noexpandtab sw=8 ts=8 sts=0: 3ccd979bdSMark Fasheh * 4ccd979bdSMark Fasheh * journal.c 5ccd979bdSMark Fasheh * 6ccd979bdSMark Fasheh * Defines functions of journalling api 7ccd979bdSMark Fasheh * 8ccd979bdSMark Fasheh * Copyright (C) 2003, 2004 Oracle. All rights reserved. 9ccd979bdSMark Fasheh * 10ccd979bdSMark Fasheh * This program is free software; you can redistribute it and/or 11ccd979bdSMark Fasheh * modify it under the terms of the GNU General Public 12ccd979bdSMark Fasheh * License as published by the Free Software Foundation; either 13ccd979bdSMark Fasheh * version 2 of the License, or (at your option) any later version. 14ccd979bdSMark Fasheh * 15ccd979bdSMark Fasheh * This program is distributed in the hope that it will be useful, 16ccd979bdSMark Fasheh * but WITHOUT ANY WARRANTY; without even the implied warranty of 17ccd979bdSMark Fasheh * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18ccd979bdSMark Fasheh * General Public License for more details. 19ccd979bdSMark Fasheh * 20ccd979bdSMark Fasheh * You should have received a copy of the GNU General Public 21ccd979bdSMark Fasheh * License along with this program; if not, write to the 22ccd979bdSMark Fasheh * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 23ccd979bdSMark Fasheh * Boston, MA 021110-1307, USA. 24ccd979bdSMark Fasheh */ 25ccd979bdSMark Fasheh 26ccd979bdSMark Fasheh #include <linux/fs.h> 27ccd979bdSMark Fasheh #include <linux/types.h> 28ccd979bdSMark Fasheh #include <linux/slab.h> 29ccd979bdSMark Fasheh #include <linux/highmem.h> 30ccd979bdSMark Fasheh #include <linux/kthread.h> 3183273932SSrinivas Eeda #include <linux/time.h> 3283273932SSrinivas Eeda #include <linux/random.h> 3355b465b6SJoseph Qi #include <linux/delay.h> 34ccd979bdSMark Fasheh 35ccd979bdSMark Fasheh #include <cluster/masklog.h> 36ccd979bdSMark Fasheh 37ccd979bdSMark Fasheh #include "ocfs2.h" 38ccd979bdSMark Fasheh 39ccd979bdSMark Fasheh #include "alloc.h" 4050655ae9SJoel Becker #include "blockcheck.h" 41316f4b9fSMark Fasheh #include "dir.h" 42ccd979bdSMark Fasheh #include "dlmglue.h" 43ccd979bdSMark Fasheh #include "extent_map.h" 44ccd979bdSMark Fasheh #include "heartbeat.h" 45ccd979bdSMark Fasheh #include "inode.h" 46ccd979bdSMark Fasheh #include "journal.h" 47ccd979bdSMark Fasheh #include "localalloc.h" 48ccd979bdSMark Fasheh #include "slot_map.h" 49ccd979bdSMark Fasheh #include "super.h" 50ccd979bdSMark Fasheh #include "sysfile.h" 510cf2f763SJoel Becker #include "uptodate.h" 522205363dSJan Kara #include "quota.h" 53ccd979bdSMark Fasheh 54ccd979bdSMark Fasheh #include "buffer_head_io.h" 55b4107950STao Ma #include "ocfs2_trace.h" 56ccd979bdSMark Fasheh 5734af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 58ccd979bdSMark Fasheh 5983273932SSrinivas Eeda #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 6083273932SSrinivas Eeda 61ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 62ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 632205363dSJan Kara int node_num, int slot_num); 64ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 65ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 6619ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 67ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 68539d8264SSunil Mushran int dirty, int replayed); 69ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 70ccd979bdSMark Fasheh int slot_num); 71ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 72ccd979bdSMark Fasheh int slot); 73ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 749140db04SSrinivas Eeda static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 759140db04SSrinivas Eeda int slot_num, 769140db04SSrinivas Eeda struct ocfs2_dinode *la_dinode, 779140db04SSrinivas Eeda struct ocfs2_dinode *tl_dinode, 789140db04SSrinivas Eeda struct ocfs2_quota_recovery *qrec); 79ccd979bdSMark Fasheh 8019ece546SJan Kara static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 8119ece546SJan Kara { 8219ece546SJan Kara return __ocfs2_wait_on_mount(osb, 0); 8319ece546SJan Kara } 8419ece546SJan Kara 8519ece546SJan Kara static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 8619ece546SJan Kara { 8719ece546SJan Kara return __ocfs2_wait_on_mount(osb, 1); 8819ece546SJan Kara } 8919ece546SJan Kara 909140db04SSrinivas Eeda /* 919140db04SSrinivas Eeda * This replay_map is to track online/offline slots, so we could recover 929140db04SSrinivas Eeda * offline slots during recovery and mount 939140db04SSrinivas Eeda */ 949140db04SSrinivas Eeda 959140db04SSrinivas Eeda enum ocfs2_replay_state { 969140db04SSrinivas Eeda REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 979140db04SSrinivas Eeda REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 989140db04SSrinivas Eeda REPLAY_DONE /* Replay was already queued */ 999140db04SSrinivas Eeda }; 1009140db04SSrinivas Eeda 1019140db04SSrinivas Eeda struct ocfs2_replay_map { 1029140db04SSrinivas Eeda unsigned int rm_slots; 1039140db04SSrinivas Eeda enum ocfs2_replay_state rm_state; 1049140db04SSrinivas Eeda unsigned char rm_replay_slots[0]; 1059140db04SSrinivas Eeda }; 1069140db04SSrinivas Eeda 1079140db04SSrinivas Eeda void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 1089140db04SSrinivas Eeda { 1099140db04SSrinivas Eeda if (!osb->replay_map) 1109140db04SSrinivas Eeda return; 1119140db04SSrinivas Eeda 1129140db04SSrinivas Eeda /* If we've already queued the replay, we don't have any more to do */ 1139140db04SSrinivas Eeda if (osb->replay_map->rm_state == REPLAY_DONE) 1149140db04SSrinivas Eeda return; 1159140db04SSrinivas Eeda 1169140db04SSrinivas Eeda osb->replay_map->rm_state = state; 1179140db04SSrinivas Eeda } 1189140db04SSrinivas Eeda 1199140db04SSrinivas Eeda int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 1209140db04SSrinivas Eeda { 1219140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map; 1229140db04SSrinivas Eeda int i, node_num; 1239140db04SSrinivas Eeda 1249140db04SSrinivas Eeda /* If replay map is already set, we don't do it again */ 1259140db04SSrinivas Eeda if (osb->replay_map) 1269140db04SSrinivas Eeda return 0; 1279140db04SSrinivas Eeda 1289140db04SSrinivas Eeda replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 1299140db04SSrinivas Eeda (osb->max_slots * sizeof(char)), GFP_KERNEL); 1309140db04SSrinivas Eeda 1319140db04SSrinivas Eeda if (!replay_map) { 1329140db04SSrinivas Eeda mlog_errno(-ENOMEM); 1339140db04SSrinivas Eeda return -ENOMEM; 1349140db04SSrinivas Eeda } 1359140db04SSrinivas Eeda 1369140db04SSrinivas Eeda spin_lock(&osb->osb_lock); 1379140db04SSrinivas Eeda 1389140db04SSrinivas Eeda replay_map->rm_slots = osb->max_slots; 1399140db04SSrinivas Eeda replay_map->rm_state = REPLAY_UNNEEDED; 1409140db04SSrinivas Eeda 1419140db04SSrinivas Eeda /* set rm_replay_slots for offline slot(s) */ 1429140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) { 1439140db04SSrinivas Eeda if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 1449140db04SSrinivas Eeda replay_map->rm_replay_slots[i] = 1; 1459140db04SSrinivas Eeda } 1469140db04SSrinivas Eeda 1479140db04SSrinivas Eeda osb->replay_map = replay_map; 1489140db04SSrinivas Eeda spin_unlock(&osb->osb_lock); 1499140db04SSrinivas Eeda return 0; 1509140db04SSrinivas Eeda } 1519140db04SSrinivas Eeda 1529140db04SSrinivas Eeda void ocfs2_queue_replay_slots(struct ocfs2_super *osb) 1539140db04SSrinivas Eeda { 1549140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1559140db04SSrinivas Eeda int i; 1569140db04SSrinivas Eeda 1579140db04SSrinivas Eeda if (!replay_map) 1589140db04SSrinivas Eeda return; 1599140db04SSrinivas Eeda 1609140db04SSrinivas Eeda if (replay_map->rm_state != REPLAY_NEEDED) 1619140db04SSrinivas Eeda return; 1629140db04SSrinivas Eeda 1639140db04SSrinivas Eeda for (i = 0; i < replay_map->rm_slots; i++) 1649140db04SSrinivas Eeda if (replay_map->rm_replay_slots[i]) 1659140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, 1669140db04SSrinivas Eeda NULL, NULL); 1679140db04SSrinivas Eeda replay_map->rm_state = REPLAY_DONE; 1689140db04SSrinivas Eeda } 1699140db04SSrinivas Eeda 1709140db04SSrinivas Eeda void ocfs2_free_replay_slots(struct ocfs2_super *osb) 1719140db04SSrinivas Eeda { 1729140db04SSrinivas Eeda struct ocfs2_replay_map *replay_map = osb->replay_map; 1739140db04SSrinivas Eeda 1749140db04SSrinivas Eeda if (!osb->replay_map) 1759140db04SSrinivas Eeda return; 1769140db04SSrinivas Eeda 1779140db04SSrinivas Eeda kfree(replay_map); 1789140db04SSrinivas Eeda osb->replay_map = NULL; 1799140db04SSrinivas Eeda } 1809140db04SSrinivas Eeda 181553abd04SJoel Becker int ocfs2_recovery_init(struct ocfs2_super *osb) 182553abd04SJoel Becker { 183553abd04SJoel Becker struct ocfs2_recovery_map *rm; 184553abd04SJoel Becker 185553abd04SJoel Becker mutex_init(&osb->recovery_lock); 186553abd04SJoel Becker osb->disable_recovery = 0; 187553abd04SJoel Becker osb->recovery_thread_task = NULL; 188553abd04SJoel Becker init_waitqueue_head(&osb->recovery_event); 189553abd04SJoel Becker 190553abd04SJoel Becker rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 191553abd04SJoel Becker osb->max_slots * sizeof(unsigned int), 192553abd04SJoel Becker GFP_KERNEL); 193553abd04SJoel Becker if (!rm) { 194553abd04SJoel Becker mlog_errno(-ENOMEM); 195553abd04SJoel Becker return -ENOMEM; 196553abd04SJoel Becker } 197553abd04SJoel Becker 198553abd04SJoel Becker rm->rm_entries = (unsigned int *)((char *)rm + 199553abd04SJoel Becker sizeof(struct ocfs2_recovery_map)); 200553abd04SJoel Becker osb->recovery_map = rm; 201553abd04SJoel Becker 202553abd04SJoel Becker return 0; 203553abd04SJoel Becker } 204553abd04SJoel Becker 205553abd04SJoel Becker /* we can't grab the goofy sem lock from inside wait_event, so we use 206553abd04SJoel Becker * memory barriers to make sure that we'll see the null task before 207553abd04SJoel Becker * being woken up */ 208553abd04SJoel Becker static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 209553abd04SJoel Becker { 210553abd04SJoel Becker mb(); 211553abd04SJoel Becker return osb->recovery_thread_task != NULL; 212553abd04SJoel Becker } 213553abd04SJoel Becker 214553abd04SJoel Becker void ocfs2_recovery_exit(struct ocfs2_super *osb) 215553abd04SJoel Becker { 216553abd04SJoel Becker struct ocfs2_recovery_map *rm; 217553abd04SJoel Becker 218553abd04SJoel Becker /* disable any new recovery threads and wait for any currently 219553abd04SJoel Becker * running ones to exit. Do this before setting the vol_state. */ 220553abd04SJoel Becker mutex_lock(&osb->recovery_lock); 221553abd04SJoel Becker osb->disable_recovery = 1; 222553abd04SJoel Becker mutex_unlock(&osb->recovery_lock); 223553abd04SJoel Becker wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 224553abd04SJoel Becker 225553abd04SJoel Becker /* At this point, we know that no more recovery threads can be 226553abd04SJoel Becker * launched, so wait for any recovery completion work to 227553abd04SJoel Becker * complete. */ 228553abd04SJoel Becker flush_workqueue(ocfs2_wq); 229553abd04SJoel Becker 230553abd04SJoel Becker /* 231553abd04SJoel Becker * Now that recovery is shut down, and the osb is about to be 232553abd04SJoel Becker * freed, the osb_lock is not taken here. 233553abd04SJoel Becker */ 234553abd04SJoel Becker rm = osb->recovery_map; 235553abd04SJoel Becker /* XXX: Should we bug if there are dirty entries? */ 236553abd04SJoel Becker 237553abd04SJoel Becker kfree(rm); 238553abd04SJoel Becker } 239553abd04SJoel Becker 240553abd04SJoel Becker static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 241553abd04SJoel Becker unsigned int node_num) 242553abd04SJoel Becker { 243553abd04SJoel Becker int i; 244553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 245553abd04SJoel Becker 246553abd04SJoel Becker assert_spin_locked(&osb->osb_lock); 247553abd04SJoel Becker 248553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 249553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 250553abd04SJoel Becker return 1; 251553abd04SJoel Becker } 252553abd04SJoel Becker 253553abd04SJoel Becker return 0; 254553abd04SJoel Becker } 255553abd04SJoel Becker 256553abd04SJoel Becker /* Behaves like test-and-set. Returns the previous value */ 257553abd04SJoel Becker static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 258553abd04SJoel Becker unsigned int node_num) 259553abd04SJoel Becker { 260553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 261553abd04SJoel Becker 262553abd04SJoel Becker spin_lock(&osb->osb_lock); 263553abd04SJoel Becker if (__ocfs2_recovery_map_test(osb, node_num)) { 264553abd04SJoel Becker spin_unlock(&osb->osb_lock); 265553abd04SJoel Becker return 1; 266553abd04SJoel Becker } 267553abd04SJoel Becker 268553abd04SJoel Becker /* XXX: Can this be exploited? Not from o2dlm... */ 269553abd04SJoel Becker BUG_ON(rm->rm_used >= osb->max_slots); 270553abd04SJoel Becker 271553abd04SJoel Becker rm->rm_entries[rm->rm_used] = node_num; 272553abd04SJoel Becker rm->rm_used++; 273553abd04SJoel Becker spin_unlock(&osb->osb_lock); 274553abd04SJoel Becker 275553abd04SJoel Becker return 0; 276553abd04SJoel Becker } 277553abd04SJoel Becker 278553abd04SJoel Becker static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 279553abd04SJoel Becker unsigned int node_num) 280553abd04SJoel Becker { 281553abd04SJoel Becker int i; 282553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 283553abd04SJoel Becker 284553abd04SJoel Becker spin_lock(&osb->osb_lock); 285553abd04SJoel Becker 286553abd04SJoel Becker for (i = 0; i < rm->rm_used; i++) { 287553abd04SJoel Becker if (rm->rm_entries[i] == node_num) 288553abd04SJoel Becker break; 289553abd04SJoel Becker } 290553abd04SJoel Becker 291553abd04SJoel Becker if (i < rm->rm_used) { 292553abd04SJoel Becker /* XXX: be careful with the pointer math */ 293553abd04SJoel Becker memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 294553abd04SJoel Becker (rm->rm_used - i - 1) * sizeof(unsigned int)); 295553abd04SJoel Becker rm->rm_used--; 296553abd04SJoel Becker } 297553abd04SJoel Becker 298553abd04SJoel Becker spin_unlock(&osb->osb_lock); 299553abd04SJoel Becker } 300553abd04SJoel Becker 301ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 302ccd979bdSMark Fasheh { 303ccd979bdSMark Fasheh int status = 0; 304ccd979bdSMark Fasheh unsigned int flushed; 305ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 306ccd979bdSMark Fasheh 307ccd979bdSMark Fasheh journal = osb->journal; 308ccd979bdSMark Fasheh 309ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 310ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 311ccd979bdSMark Fasheh 312b4107950STao Ma flushed = atomic_read(&journal->j_num_trans); 313b4107950STao Ma trace_ocfs2_commit_cache_begin(flushed); 314b4107950STao Ma if (flushed == 0) { 315ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 316ccd979bdSMark Fasheh goto finally; 317ccd979bdSMark Fasheh } 318ccd979bdSMark Fasheh 3192b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 3202b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 3212b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 322ccd979bdSMark Fasheh if (status < 0) { 323ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 324ccd979bdSMark Fasheh mlog_errno(status); 325ccd979bdSMark Fasheh goto finally; 326ccd979bdSMark Fasheh } 327ccd979bdSMark Fasheh 328f9c57adaSTao Ma ocfs2_inc_trans_id(journal); 329ccd979bdSMark Fasheh 330ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 331ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 332ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 333ccd979bdSMark Fasheh 334b4107950STao Ma trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 335ccd979bdSMark Fasheh 33634d024f8SMark Fasheh ocfs2_wake_downconvert_thread(osb); 337ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 338ccd979bdSMark Fasheh finally: 339ccd979bdSMark Fasheh return status; 340ccd979bdSMark Fasheh } 341ccd979bdSMark Fasheh 3421fabe148SMark Fasheh handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 343ccd979bdSMark Fasheh { 344ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 3451fabe148SMark Fasheh handle_t *handle; 346ccd979bdSMark Fasheh 347ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 348ccd979bdSMark Fasheh 34965eff9ccSMark Fasheh if (ocfs2_is_hard_readonly(osb)) 35065eff9ccSMark Fasheh return ERR_PTR(-EROFS); 351ccd979bdSMark Fasheh 352ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 353ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 354ccd979bdSMark Fasheh 35590e86a63SJan Kara /* Nested transaction? Just return the handle... */ 35690e86a63SJan Kara if (journal_current_handle()) 35790e86a63SJan Kara return jbd2_journal_start(journal, max_buffs); 358ccd979bdSMark Fasheh 359fef6925cSJan Kara sb_start_intwrite(osb->sb); 360fef6925cSJan Kara 361ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 362ccd979bdSMark Fasheh 3632b4e30fbSJoel Becker handle = jbd2_journal_start(journal, max_buffs); 3641fabe148SMark Fasheh if (IS_ERR(handle)) { 365ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 366fef6925cSJan Kara sb_end_intwrite(osb->sb); 367ccd979bdSMark Fasheh 3681fabe148SMark Fasheh mlog_errno(PTR_ERR(handle)); 369ccd979bdSMark Fasheh 370ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 371ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 3721fabe148SMark Fasheh handle = ERR_PTR(-EROFS); 373ccd979bdSMark Fasheh } 374c271c5c2SSunil Mushran } else { 375c271c5c2SSunil Mushran if (!ocfs2_mount_local(osb)) 376ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 377c271c5c2SSunil Mushran } 378ccd979bdSMark Fasheh 379ccd979bdSMark Fasheh return handle; 380ccd979bdSMark Fasheh } 381ccd979bdSMark Fasheh 3821fabe148SMark Fasheh int ocfs2_commit_trans(struct ocfs2_super *osb, 3831fabe148SMark Fasheh handle_t *handle) 384ccd979bdSMark Fasheh { 38590e86a63SJan Kara int ret, nested; 38602dc1af4SMark Fasheh struct ocfs2_journal *journal = osb->journal; 387ccd979bdSMark Fasheh 388ccd979bdSMark Fasheh BUG_ON(!handle); 389ccd979bdSMark Fasheh 39090e86a63SJan Kara nested = handle->h_ref > 1; 3912b4e30fbSJoel Becker ret = jbd2_journal_stop(handle); 3921fabe148SMark Fasheh if (ret < 0) 3931fabe148SMark Fasheh mlog_errno(ret); 394ccd979bdSMark Fasheh 395fef6925cSJan Kara if (!nested) { 396ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 397fef6925cSJan Kara sb_end_intwrite(osb->sb); 398fef6925cSJan Kara } 399ccd979bdSMark Fasheh 4001fabe148SMark Fasheh return ret; 401ccd979bdSMark Fasheh } 402ccd979bdSMark Fasheh 403ccd979bdSMark Fasheh /* 404c901fb00STao Ma * 'nblocks' is what you want to add to the current transaction. 405ccd979bdSMark Fasheh * 4062b4e30fbSJoel Becker * This might call jbd2_journal_restart() which will commit dirty buffers 407e8aed345SMark Fasheh * and then restart the transaction. Before calling 408e8aed345SMark Fasheh * ocfs2_extend_trans(), any changed blocks should have been 409e8aed345SMark Fasheh * dirtied. After calling it, all blocks which need to be changed must 410e8aed345SMark Fasheh * go through another set of journal_access/journal_dirty calls. 411e8aed345SMark Fasheh * 412ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 413ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 414ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 415ccd979bdSMark Fasheh * 416ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 417ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 418ccd979bdSMark Fasheh * cluster locks associated with this handle. 419ccd979bdSMark Fasheh */ 4201fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 421ccd979bdSMark Fasheh { 422c901fb00STao Ma int status, old_nblocks; 423ccd979bdSMark Fasheh 424ccd979bdSMark Fasheh BUG_ON(!handle); 425c901fb00STao Ma BUG_ON(nblocks < 0); 426ccd979bdSMark Fasheh 427c901fb00STao Ma if (!nblocks) 428c901fb00STao Ma return 0; 429c901fb00STao Ma 430c901fb00STao Ma old_nblocks = handle->h_buffer_credits; 431ccd979bdSMark Fasheh 432b4107950STao Ma trace_ocfs2_extend_trans(old_nblocks, nblocks); 433ccd979bdSMark Fasheh 434e407e397SJoel Becker #ifdef CONFIG_OCFS2_DEBUG_FS 4350879c584SMark Fasheh status = 1; 4360879c584SMark Fasheh #else 4372b4e30fbSJoel Becker status = jbd2_journal_extend(handle, nblocks); 438ccd979bdSMark Fasheh if (status < 0) { 439ccd979bdSMark Fasheh mlog_errno(status); 440ccd979bdSMark Fasheh goto bail; 441ccd979bdSMark Fasheh } 4420879c584SMark Fasheh #endif 443ccd979bdSMark Fasheh 444ccd979bdSMark Fasheh if (status > 0) { 445b4107950STao Ma trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 446c901fb00STao Ma status = jbd2_journal_restart(handle, 447c901fb00STao Ma old_nblocks + nblocks); 448ccd979bdSMark Fasheh if (status < 0) { 449ccd979bdSMark Fasheh mlog_errno(status); 450ccd979bdSMark Fasheh goto bail; 451ccd979bdSMark Fasheh } 45201ddf1e1SMark Fasheh } 453ccd979bdSMark Fasheh 454ccd979bdSMark Fasheh status = 0; 455ccd979bdSMark Fasheh bail: 456ccd979bdSMark Fasheh return status; 457ccd979bdSMark Fasheh } 458ccd979bdSMark Fasheh 4592b1e55c3SYounger Liu /* 4602b1e55c3SYounger Liu * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 4612b1e55c3SYounger Liu * If that fails, restart the transaction & regain write access for the 4622b1e55c3SYounger Liu * buffer head which is used for metadata modifications. 4632b1e55c3SYounger Liu * Taken from Ext4: extend_or_restart_transaction() 4642b1e55c3SYounger Liu */ 4652b1e55c3SYounger Liu int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 4662b1e55c3SYounger Liu { 4672b1e55c3SYounger Liu int status, old_nblks; 4682b1e55c3SYounger Liu 4692b1e55c3SYounger Liu BUG_ON(!handle); 4702b1e55c3SYounger Liu 4712b1e55c3SYounger Liu old_nblks = handle->h_buffer_credits; 4722b1e55c3SYounger Liu trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 4732b1e55c3SYounger Liu 4742b1e55c3SYounger Liu if (old_nblks < thresh) 4752b1e55c3SYounger Liu return 0; 4762b1e55c3SYounger Liu 4772b1e55c3SYounger Liu status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA); 4782b1e55c3SYounger Liu if (status < 0) { 4792b1e55c3SYounger Liu mlog_errno(status); 4802b1e55c3SYounger Liu goto bail; 4812b1e55c3SYounger Liu } 4822b1e55c3SYounger Liu 4832b1e55c3SYounger Liu if (status > 0) { 4842b1e55c3SYounger Liu status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 4852b1e55c3SYounger Liu if (status < 0) 4862b1e55c3SYounger Liu mlog_errno(status); 4872b1e55c3SYounger Liu } 4882b1e55c3SYounger Liu 4892b1e55c3SYounger Liu bail: 4902b1e55c3SYounger Liu return status; 4912b1e55c3SYounger Liu } 4922b1e55c3SYounger Liu 4932b1e55c3SYounger Liu 49450655ae9SJoel Becker struct ocfs2_triggers { 49550655ae9SJoel Becker struct jbd2_buffer_trigger_type ot_triggers; 49650655ae9SJoel Becker int ot_offset; 49750655ae9SJoel Becker }; 49850655ae9SJoel Becker 49950655ae9SJoel Becker static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 50050655ae9SJoel Becker { 50150655ae9SJoel Becker return container_of(triggers, struct ocfs2_triggers, ot_triggers); 50250655ae9SJoel Becker } 50350655ae9SJoel Becker 50413ceef09SJan Kara static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 50550655ae9SJoel Becker struct buffer_head *bh, 50650655ae9SJoel Becker void *data, size_t size) 50750655ae9SJoel Becker { 50850655ae9SJoel Becker struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 50950655ae9SJoel Becker 51050655ae9SJoel Becker /* 51150655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 51250655ae9SJoel Becker * must unconditionally compute the ecc data. 51350655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 51450655ae9SJoel Becker * metaecc is enabled. 51550655ae9SJoel Becker */ 51650655ae9SJoel Becker ocfs2_block_check_compute(data, size, data + ot->ot_offset); 51750655ae9SJoel Becker } 51850655ae9SJoel Becker 51950655ae9SJoel Becker /* 52050655ae9SJoel Becker * Quota blocks have their own trigger because the struct ocfs2_block_check 52150655ae9SJoel Becker * offset depends on the blocksize. 52250655ae9SJoel Becker */ 52313ceef09SJan Kara static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 52450655ae9SJoel Becker struct buffer_head *bh, 52550655ae9SJoel Becker void *data, size_t size) 52650655ae9SJoel Becker { 52750655ae9SJoel Becker struct ocfs2_disk_dqtrailer *dqt = 52850655ae9SJoel Becker ocfs2_block_dqtrailer(size, data); 52950655ae9SJoel Becker 53050655ae9SJoel Becker /* 53150655ae9SJoel Becker * We aren't guaranteed to have the superblock here, so we 53250655ae9SJoel Becker * must unconditionally compute the ecc data. 53350655ae9SJoel Becker * __ocfs2_journal_access() will only set the triggers if 53450655ae9SJoel Becker * metaecc is enabled. 53550655ae9SJoel Becker */ 53650655ae9SJoel Becker ocfs2_block_check_compute(data, size, &dqt->dq_check); 53750655ae9SJoel Becker } 53850655ae9SJoel Becker 539c175a518SJoel Becker /* 540c175a518SJoel Becker * Directory blocks also have their own trigger because the 541c175a518SJoel Becker * struct ocfs2_block_check offset depends on the blocksize. 542c175a518SJoel Becker */ 54313ceef09SJan Kara static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 544c175a518SJoel Becker struct buffer_head *bh, 545c175a518SJoel Becker void *data, size_t size) 546c175a518SJoel Becker { 547c175a518SJoel Becker struct ocfs2_dir_block_trailer *trailer = 548c175a518SJoel Becker ocfs2_dir_trailer_from_size(size, data); 549c175a518SJoel Becker 550c175a518SJoel Becker /* 551c175a518SJoel Becker * We aren't guaranteed to have the superblock here, so we 552c175a518SJoel Becker * must unconditionally compute the ecc data. 553c175a518SJoel Becker * __ocfs2_journal_access() will only set the triggers if 554c175a518SJoel Becker * metaecc is enabled. 555c175a518SJoel Becker */ 556c175a518SJoel Becker ocfs2_block_check_compute(data, size, &trailer->db_check); 557c175a518SJoel Becker } 558c175a518SJoel Becker 55950655ae9SJoel Becker static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 56050655ae9SJoel Becker struct buffer_head *bh) 56150655ae9SJoel Becker { 56250655ae9SJoel Becker mlog(ML_ERROR, 56350655ae9SJoel Becker "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 56450655ae9SJoel Becker "bh->b_blocknr = %llu\n", 56550655ae9SJoel Becker (unsigned long)bh, 56650655ae9SJoel Becker (unsigned long long)bh->b_blocknr); 56750655ae9SJoel Becker 56850655ae9SJoel Becker /* We aren't guaranteed to have the superblock here - but if we 56950655ae9SJoel Becker * don't, it'll just crash. */ 57050655ae9SJoel Becker ocfs2_error(bh->b_assoc_map->host->i_sb, 57150655ae9SJoel Becker "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 57250655ae9SJoel Becker } 57350655ae9SJoel Becker 57450655ae9SJoel Becker static struct ocfs2_triggers di_triggers = { 57550655ae9SJoel Becker .ot_triggers = { 57613ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 57750655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 57850655ae9SJoel Becker }, 57950655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_dinode, i_check), 58050655ae9SJoel Becker }; 58150655ae9SJoel Becker 58250655ae9SJoel Becker static struct ocfs2_triggers eb_triggers = { 58350655ae9SJoel Becker .ot_triggers = { 58413ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 58550655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 58650655ae9SJoel Becker }, 58750655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 58850655ae9SJoel Becker }; 58950655ae9SJoel Becker 59093c97087STao Ma static struct ocfs2_triggers rb_triggers = { 59193c97087STao Ma .ot_triggers = { 59213ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 59393c97087STao Ma .t_abort = ocfs2_abort_trigger, 59493c97087STao Ma }, 59593c97087STao Ma .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 59693c97087STao Ma }; 59793c97087STao Ma 59850655ae9SJoel Becker static struct ocfs2_triggers gd_triggers = { 59950655ae9SJoel Becker .ot_triggers = { 60013ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 60150655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 60250655ae9SJoel Becker }, 60350655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 60450655ae9SJoel Becker }; 60550655ae9SJoel Becker 606c175a518SJoel Becker static struct ocfs2_triggers db_triggers = { 607c175a518SJoel Becker .ot_triggers = { 60813ceef09SJan Kara .t_frozen = ocfs2_db_frozen_trigger, 609c175a518SJoel Becker .t_abort = ocfs2_abort_trigger, 610c175a518SJoel Becker }, 611c175a518SJoel Becker }; 612c175a518SJoel Becker 61350655ae9SJoel Becker static struct ocfs2_triggers xb_triggers = { 61450655ae9SJoel Becker .ot_triggers = { 61513ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 61650655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 61750655ae9SJoel Becker }, 61850655ae9SJoel Becker .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 61950655ae9SJoel Becker }; 62050655ae9SJoel Becker 62150655ae9SJoel Becker static struct ocfs2_triggers dq_triggers = { 62250655ae9SJoel Becker .ot_triggers = { 62313ceef09SJan Kara .t_frozen = ocfs2_dq_frozen_trigger, 62450655ae9SJoel Becker .t_abort = ocfs2_abort_trigger, 62550655ae9SJoel Becker }, 62650655ae9SJoel Becker }; 62750655ae9SJoel Becker 6289b7895efSMark Fasheh static struct ocfs2_triggers dr_triggers = { 6299b7895efSMark Fasheh .ot_triggers = { 63013ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6319b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6329b7895efSMark Fasheh }, 6339b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 6349b7895efSMark Fasheh }; 6359b7895efSMark Fasheh 6369b7895efSMark Fasheh static struct ocfs2_triggers dl_triggers = { 6379b7895efSMark Fasheh .ot_triggers = { 63813ceef09SJan Kara .t_frozen = ocfs2_frozen_trigger, 6399b7895efSMark Fasheh .t_abort = ocfs2_abort_trigger, 6409b7895efSMark Fasheh }, 6419b7895efSMark Fasheh .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 6429b7895efSMark Fasheh }; 6439b7895efSMark Fasheh 64450655ae9SJoel Becker static int __ocfs2_journal_access(handle_t *handle, 6450cf2f763SJoel Becker struct ocfs2_caching_info *ci, 646ccd979bdSMark Fasheh struct buffer_head *bh, 64750655ae9SJoel Becker struct ocfs2_triggers *triggers, 648ccd979bdSMark Fasheh int type) 649ccd979bdSMark Fasheh { 650ccd979bdSMark Fasheh int status; 6510cf2f763SJoel Becker struct ocfs2_super *osb = 6520cf2f763SJoel Becker OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 653ccd979bdSMark Fasheh 6540cf2f763SJoel Becker BUG_ON(!ci || !ci->ci_ops); 655ccd979bdSMark Fasheh BUG_ON(!handle); 656ccd979bdSMark Fasheh BUG_ON(!bh); 657ccd979bdSMark Fasheh 658b4107950STao Ma trace_ocfs2_journal_access( 659b4107950STao Ma (unsigned long long)ocfs2_metadata_cache_owner(ci), 660b4107950STao Ma (unsigned long long)bh->b_blocknr, type, bh->b_size); 661ccd979bdSMark Fasheh 662ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 663ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 664ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 665ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 666ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 667ccd979bdSMark Fasheh BUG(); 668ccd979bdSMark Fasheh } 669ccd979bdSMark Fasheh 6700cf2f763SJoel Becker /* Set the current transaction information on the ci so 671ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 6720cf2f763SJoel Becker * on this ci or not. We're protected from the commit 673ccd979bdSMark Fasheh * thread updating the current transaction id until 674ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 675ccd979bdSMark Fasheh * j_trans_barrier for us. */ 6760cf2f763SJoel Becker ocfs2_set_ci_lock_trans(osb->journal, ci); 677ccd979bdSMark Fasheh 6780cf2f763SJoel Becker ocfs2_metadata_cache_io_lock(ci); 679ccd979bdSMark Fasheh switch (type) { 680ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 681ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 6822b4e30fbSJoel Becker status = jbd2_journal_get_write_access(handle, bh); 683ccd979bdSMark Fasheh break; 684ccd979bdSMark Fasheh 685ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 6862b4e30fbSJoel Becker status = jbd2_journal_get_undo_access(handle, bh); 687ccd979bdSMark Fasheh break; 688ccd979bdSMark Fasheh 689ccd979bdSMark Fasheh default: 690ccd979bdSMark Fasheh status = -EINVAL; 691af901ca1SAndré Goddard Rosa mlog(ML_ERROR, "Unknown access type!\n"); 692ccd979bdSMark Fasheh } 6930cf2f763SJoel Becker if (!status && ocfs2_meta_ecc(osb) && triggers) 69450655ae9SJoel Becker jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 6950cf2f763SJoel Becker ocfs2_metadata_cache_io_unlock(ci); 696ccd979bdSMark Fasheh 697ccd979bdSMark Fasheh if (status < 0) 698ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 699ccd979bdSMark Fasheh status, type); 700ccd979bdSMark Fasheh 701ccd979bdSMark Fasheh return status; 702ccd979bdSMark Fasheh } 703ccd979bdSMark Fasheh 7040cf2f763SJoel Becker int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 70550655ae9SJoel Becker struct buffer_head *bh, int type) 70650655ae9SJoel Becker { 7070cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 70850655ae9SJoel Becker } 70950655ae9SJoel Becker 7100cf2f763SJoel Becker int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 71150655ae9SJoel Becker struct buffer_head *bh, int type) 71250655ae9SJoel Becker { 7130cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 71450655ae9SJoel Becker } 71550655ae9SJoel Becker 71693c97087STao Ma int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 71793c97087STao Ma struct buffer_head *bh, int type) 71893c97087STao Ma { 71993c97087STao Ma return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 72093c97087STao Ma type); 72193c97087STao Ma } 72293c97087STao Ma 7230cf2f763SJoel Becker int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 72450655ae9SJoel Becker struct buffer_head *bh, int type) 72550655ae9SJoel Becker { 7260cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 72750655ae9SJoel Becker } 72850655ae9SJoel Becker 7290cf2f763SJoel Becker int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 73050655ae9SJoel Becker struct buffer_head *bh, int type) 73150655ae9SJoel Becker { 7320cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 73350655ae9SJoel Becker } 73450655ae9SJoel Becker 7350cf2f763SJoel Becker int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 73650655ae9SJoel Becker struct buffer_head *bh, int type) 73750655ae9SJoel Becker { 7380cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 73950655ae9SJoel Becker } 74050655ae9SJoel Becker 7410cf2f763SJoel Becker int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 74250655ae9SJoel Becker struct buffer_head *bh, int type) 74350655ae9SJoel Becker { 7440cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 74550655ae9SJoel Becker } 74650655ae9SJoel Becker 7470cf2f763SJoel Becker int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 7489b7895efSMark Fasheh struct buffer_head *bh, int type) 7499b7895efSMark Fasheh { 7500cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 7519b7895efSMark Fasheh } 7529b7895efSMark Fasheh 7530cf2f763SJoel Becker int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 7549b7895efSMark Fasheh struct buffer_head *bh, int type) 7559b7895efSMark Fasheh { 7560cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 7579b7895efSMark Fasheh } 7589b7895efSMark Fasheh 7590cf2f763SJoel Becker int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 76050655ae9SJoel Becker struct buffer_head *bh, int type) 76150655ae9SJoel Becker { 7620cf2f763SJoel Becker return __ocfs2_journal_access(handle, ci, bh, NULL, type); 76350655ae9SJoel Becker } 76450655ae9SJoel Becker 765ec20cec7SJoel Becker void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 766ccd979bdSMark Fasheh { 767ccd979bdSMark Fasheh int status; 768ccd979bdSMark Fasheh 769b4107950STao Ma trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 770ccd979bdSMark Fasheh 7712b4e30fbSJoel Becker status = jbd2_journal_dirty_metadata(handle, bh); 772ec20cec7SJoel Becker BUG_ON(status); 773ccd979bdSMark Fasheh } 774ccd979bdSMark Fasheh 7752b4e30fbSJoel Becker #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 776ccd979bdSMark Fasheh 777ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 778ccd979bdSMark Fasheh { 779ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 780d147b3d6SMark Fasheh unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 781d147b3d6SMark Fasheh 782d147b3d6SMark Fasheh if (osb->osb_commit_interval) 783d147b3d6SMark Fasheh commit_interval = osb->osb_commit_interval; 784ccd979bdSMark Fasheh 785a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 786d147b3d6SMark Fasheh journal->j_commit_interval = commit_interval; 787ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 7882b4e30fbSJoel Becker journal->j_flags |= JBD2_BARRIER; 789ccd979bdSMark Fasheh else 7902b4e30fbSJoel Becker journal->j_flags &= ~JBD2_BARRIER; 791a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 792ccd979bdSMark Fasheh } 793ccd979bdSMark Fasheh 794ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 795ccd979bdSMark Fasheh { 796ccd979bdSMark Fasheh int status = -1; 797ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 798ccd979bdSMark Fasheh journal_t *j_journal = NULL; 799ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 800ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 801ccd979bdSMark Fasheh struct ocfs2_super *osb; 802e63aecb6SMark Fasheh int inode_lock = 0; 803ccd979bdSMark Fasheh 804ccd979bdSMark Fasheh BUG_ON(!journal); 805ccd979bdSMark Fasheh 806ccd979bdSMark Fasheh osb = journal->j_osb; 807ccd979bdSMark Fasheh 808ccd979bdSMark Fasheh /* already have the inode for our journal */ 809ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 810ccd979bdSMark Fasheh osb->slot_num); 811ccd979bdSMark Fasheh if (inode == NULL) { 812ccd979bdSMark Fasheh status = -EACCES; 813ccd979bdSMark Fasheh mlog_errno(status); 814ccd979bdSMark Fasheh goto done; 815ccd979bdSMark Fasheh } 816ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 817ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 818ccd979bdSMark Fasheh iput(inode); 819ccd979bdSMark Fasheh inode = NULL; 820ccd979bdSMark Fasheh status = -EACCES; 821ccd979bdSMark Fasheh goto done; 822ccd979bdSMark Fasheh } 823ccd979bdSMark Fasheh 824ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 825ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 826ccd979bdSMark Fasheh 8276eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 8286eff5790SMark Fasheh * changes in a live cluster so it can be considered an 8296eff5790SMark Fasheh * exception to the rule. */ 830e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 831ccd979bdSMark Fasheh if (status < 0) { 832ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 833ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 834ccd979bdSMark Fasheh goto done; 835ccd979bdSMark Fasheh } 836ccd979bdSMark Fasheh 837e63aecb6SMark Fasheh inode_lock = 1; 838ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 839ccd979bdSMark Fasheh 840f17c20ddSJunxiao Bi if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 841ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 842f17c20ddSJunxiao Bi i_size_read(inode)); 843ccd979bdSMark Fasheh status = -EINVAL; 844ccd979bdSMark Fasheh goto done; 845ccd979bdSMark Fasheh } 846ccd979bdSMark Fasheh 847f17c20ddSJunxiao Bi trace_ocfs2_journal_init(i_size_read(inode), 848b4107950STao Ma (unsigned long long)inode->i_blocks, 849b4107950STao Ma OCFS2_I(inode)->ip_clusters); 850ccd979bdSMark Fasheh 851ccd979bdSMark Fasheh /* call the kernels journal init function now */ 8522b4e30fbSJoel Becker j_journal = jbd2_journal_init_inode(inode); 853ccd979bdSMark Fasheh if (j_journal == NULL) { 854ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 855ccd979bdSMark Fasheh status = -EINVAL; 856ccd979bdSMark Fasheh goto done; 857ccd979bdSMark Fasheh } 858ccd979bdSMark Fasheh 859b4107950STao Ma trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); 860ccd979bdSMark Fasheh 861ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 862ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 863ccd979bdSMark Fasheh 864ccd979bdSMark Fasheh journal->j_journal = j_journal; 865ccd979bdSMark Fasheh journal->j_inode = inode; 866ccd979bdSMark Fasheh journal->j_bh = bh; 867ccd979bdSMark Fasheh 868ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 869ccd979bdSMark Fasheh 870ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 871ccd979bdSMark Fasheh 872ccd979bdSMark Fasheh status = 0; 873ccd979bdSMark Fasheh done: 874ccd979bdSMark Fasheh if (status < 0) { 875e63aecb6SMark Fasheh if (inode_lock) 876e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 877ccd979bdSMark Fasheh brelse(bh); 878ccd979bdSMark Fasheh if (inode) { 879ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 880ccd979bdSMark Fasheh iput(inode); 881ccd979bdSMark Fasheh } 882ccd979bdSMark Fasheh } 883ccd979bdSMark Fasheh 884ccd979bdSMark Fasheh return status; 885ccd979bdSMark Fasheh } 886ccd979bdSMark Fasheh 887539d8264SSunil Mushran static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 888539d8264SSunil Mushran { 889539d8264SSunil Mushran le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 890539d8264SSunil Mushran } 891539d8264SSunil Mushran 892539d8264SSunil Mushran static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 893539d8264SSunil Mushran { 894539d8264SSunil Mushran return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 895539d8264SSunil Mushran } 896539d8264SSunil Mushran 897ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 898539d8264SSunil Mushran int dirty, int replayed) 899ccd979bdSMark Fasheh { 900ccd979bdSMark Fasheh int status; 901ccd979bdSMark Fasheh unsigned int flags; 902ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 903ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 904ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 905ccd979bdSMark Fasheh 906ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 90710995aa2SJoel Becker 90810995aa2SJoel Becker /* The journal bh on the osb always comes from ocfs2_journal_init() 90910995aa2SJoel Becker * and was validated there inside ocfs2_inode_lock_full(). It's a 91010995aa2SJoel Becker * code bug if we mess it up. */ 91110995aa2SJoel Becker BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 912ccd979bdSMark Fasheh 913ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 914ccd979bdSMark Fasheh if (dirty) 915ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 916ccd979bdSMark Fasheh else 917ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 918ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 919ccd979bdSMark Fasheh 920539d8264SSunil Mushran if (replayed) 921539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 922539d8264SSunil Mushran 92313723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 9248cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 925ccd979bdSMark Fasheh if (status < 0) 926ccd979bdSMark Fasheh mlog_errno(status); 927ccd979bdSMark Fasheh 928ccd979bdSMark Fasheh return status; 929ccd979bdSMark Fasheh } 930ccd979bdSMark Fasheh 931ccd979bdSMark Fasheh /* 932ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 933ccd979bdSMark Fasheh * call. 934ccd979bdSMark Fasheh */ 935ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 936ccd979bdSMark Fasheh { 937ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 938ccd979bdSMark Fasheh int status = 0; 939ccd979bdSMark Fasheh struct inode *inode = NULL; 940ccd979bdSMark Fasheh int num_running_trans = 0; 941ccd979bdSMark Fasheh 942ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 943ccd979bdSMark Fasheh 944ccd979bdSMark Fasheh journal = osb->journal; 945ccd979bdSMark Fasheh if (!journal) 946ccd979bdSMark Fasheh goto done; 947ccd979bdSMark Fasheh 948ccd979bdSMark Fasheh inode = journal->j_inode; 949ccd979bdSMark Fasheh 950ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 951ccd979bdSMark Fasheh goto done; 952ccd979bdSMark Fasheh 9532b4e30fbSJoel Becker /* need to inc inode use count - jbd2_journal_destroy will iput. */ 954ccd979bdSMark Fasheh if (!igrab(inode)) 955ccd979bdSMark Fasheh BUG(); 956ccd979bdSMark Fasheh 957ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 958b4107950STao Ma trace_ocfs2_journal_shutdown(num_running_trans); 959ccd979bdSMark Fasheh 960ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 961ccd979bdSMark Fasheh * release any locks that are still held. 962ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 963ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 964ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 965ccd979bdSMark Fasheh 966ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 967ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 968ccd979bdSMark Fasheh * completely destroy the journal. */ 969ccd979bdSMark Fasheh if (osb->commit_task) { 970ccd979bdSMark Fasheh /* Wait for the commit thread */ 971b4107950STao Ma trace_ocfs2_journal_shutdown_wait(osb->commit_task); 972ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 973ccd979bdSMark Fasheh osb->commit_task = NULL; 974ccd979bdSMark Fasheh } 975ccd979bdSMark Fasheh 976ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 977ccd979bdSMark Fasheh 978c271c5c2SSunil Mushran if (ocfs2_mount_local(osb)) { 9792b4e30fbSJoel Becker jbd2_journal_lock_updates(journal->j_journal); 9802b4e30fbSJoel Becker status = jbd2_journal_flush(journal->j_journal); 9812b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal->j_journal); 982c271c5c2SSunil Mushran if (status < 0) 983c271c5c2SSunil Mushran mlog_errno(status); 984c271c5c2SSunil Mushran } 985c271c5c2SSunil Mushran 986c271c5c2SSunil Mushran if (status == 0) { 987c271c5c2SSunil Mushran /* 988c271c5c2SSunil Mushran * Do not toggle if flush was unsuccessful otherwise 989c271c5c2SSunil Mushran * will leave dirty metadata in a "clean" journal 990c271c5c2SSunil Mushran */ 991539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 0, 0); 992ccd979bdSMark Fasheh if (status < 0) 993ccd979bdSMark Fasheh mlog_errno(status); 994c271c5c2SSunil Mushran } 995ccd979bdSMark Fasheh 996ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 9972b4e30fbSJoel Becker jbd2_journal_destroy(journal->j_journal); 998ae0dff68SSunil Mushran journal->j_journal = NULL; 999ccd979bdSMark Fasheh 1000ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 1001ccd979bdSMark Fasheh 1002ccd979bdSMark Fasheh /* unlock our journal */ 1003e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1004ccd979bdSMark Fasheh 1005ccd979bdSMark Fasheh brelse(journal->j_bh); 1006ccd979bdSMark Fasheh journal->j_bh = NULL; 1007ccd979bdSMark Fasheh 1008ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 1009ccd979bdSMark Fasheh 1010ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 1011ccd979bdSMark Fasheh done: 1012ccd979bdSMark Fasheh if (inode) 1013ccd979bdSMark Fasheh iput(inode); 1014ccd979bdSMark Fasheh } 1015ccd979bdSMark Fasheh 1016ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 1017ccd979bdSMark Fasheh journal_t *journal, 1018ccd979bdSMark Fasheh int slot) 1019ccd979bdSMark Fasheh { 1020ccd979bdSMark Fasheh int olderr; 1021ccd979bdSMark Fasheh 10222b4e30fbSJoel Becker olderr = jbd2_journal_errno(journal); 1023ccd979bdSMark Fasheh if (olderr) { 1024ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 1025ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 1026ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 1027ccd979bdSMark Fasheh sb->s_id); 1028ccd979bdSMark Fasheh 10292b4e30fbSJoel Becker jbd2_journal_ack_err(journal); 10302b4e30fbSJoel Becker jbd2_journal_clear_err(journal); 1031ccd979bdSMark Fasheh } 1032ccd979bdSMark Fasheh } 1033ccd979bdSMark Fasheh 1034539d8264SSunil Mushran int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1035ccd979bdSMark Fasheh { 1036ccd979bdSMark Fasheh int status = 0; 1037ccd979bdSMark Fasheh struct ocfs2_super *osb; 1038ccd979bdSMark Fasheh 1039b1f3550fSJulia Lawall BUG_ON(!journal); 1040ccd979bdSMark Fasheh 1041ccd979bdSMark Fasheh osb = journal->j_osb; 1042ccd979bdSMark Fasheh 10432b4e30fbSJoel Becker status = jbd2_journal_load(journal->j_journal); 1044ccd979bdSMark Fasheh if (status < 0) { 1045ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 1046ccd979bdSMark Fasheh goto done; 1047ccd979bdSMark Fasheh } 1048ccd979bdSMark Fasheh 1049ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1050ccd979bdSMark Fasheh 1051539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1052ccd979bdSMark Fasheh if (status < 0) { 1053ccd979bdSMark Fasheh mlog_errno(status); 1054ccd979bdSMark Fasheh goto done; 1055ccd979bdSMark Fasheh } 1056ccd979bdSMark Fasheh 1057ccd979bdSMark Fasheh /* Launch the commit thread */ 1058c271c5c2SSunil Mushran if (!local) { 1059c271c5c2SSunil Mushran osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 1060c271c5c2SSunil Mushran "ocfs2cmt"); 1061ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 1062ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 1063ccd979bdSMark Fasheh osb->commit_task = NULL; 1064c271c5c2SSunil Mushran mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1065c271c5c2SSunil Mushran "error=%d", status); 1066ccd979bdSMark Fasheh goto done; 1067ccd979bdSMark Fasheh } 1068c271c5c2SSunil Mushran } else 1069c271c5c2SSunil Mushran osb->commit_task = NULL; 1070ccd979bdSMark Fasheh 1071ccd979bdSMark Fasheh done: 1072ccd979bdSMark Fasheh return status; 1073ccd979bdSMark Fasheh } 1074ccd979bdSMark Fasheh 1075ccd979bdSMark Fasheh 1076ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 1077ccd979bdSMark Fasheh * mark the journal clean */ 1078ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1079ccd979bdSMark Fasheh { 1080ccd979bdSMark Fasheh int status; 1081ccd979bdSMark Fasheh 1082ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 1083ccd979bdSMark Fasheh 10842b4e30fbSJoel Becker status = jbd2_journal_wipe(journal->j_journal, full); 1085ccd979bdSMark Fasheh if (status < 0) { 1086ccd979bdSMark Fasheh mlog_errno(status); 1087ccd979bdSMark Fasheh goto bail; 1088ccd979bdSMark Fasheh } 1089ccd979bdSMark Fasheh 1090539d8264SSunil Mushran status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1091ccd979bdSMark Fasheh if (status < 0) 1092ccd979bdSMark Fasheh mlog_errno(status); 1093ccd979bdSMark Fasheh 1094ccd979bdSMark Fasheh bail: 1095ccd979bdSMark Fasheh return status; 1096ccd979bdSMark Fasheh } 1097ccd979bdSMark Fasheh 1098553abd04SJoel Becker static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1099553abd04SJoel Becker { 1100553abd04SJoel Becker int empty; 1101553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 1102553abd04SJoel Becker 1103553abd04SJoel Becker spin_lock(&osb->osb_lock); 1104553abd04SJoel Becker empty = (rm->rm_used == 0); 1105553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1106553abd04SJoel Becker 1107553abd04SJoel Becker return empty; 1108553abd04SJoel Becker } 1109553abd04SJoel Becker 1110553abd04SJoel Becker void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1111553abd04SJoel Becker { 1112553abd04SJoel Becker wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1113553abd04SJoel Becker } 1114553abd04SJoel Becker 1115ccd979bdSMark Fasheh /* 1116ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 1117ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 1118ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 1119ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 1120ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 1121ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 1122ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 1123ccd979bdSMark Fasheh * as we have the lock by now :) 1124ccd979bdSMark Fasheh */ 1125ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 1126ccd979bdSMark Fasheh { 1127ccd979bdSMark Fasheh int status = 0; 11284f902c37SMark Fasheh int i; 11298110b073SMark Fasheh u64 v_blkno, p_blkno, p_blocks, num_blocks; 11304f902c37SMark Fasheh #define CONCURRENT_JOURNAL_FILL 32ULL 1131ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 1132ccd979bdSMark Fasheh 1133ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 1134ccd979bdSMark Fasheh 1135f17c20ddSJunxiao Bi num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1136ccd979bdSMark Fasheh v_blkno = 0; 11378110b073SMark Fasheh while (v_blkno < num_blocks) { 1138ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 113949cb8d2dSMark Fasheh &p_blkno, &p_blocks, NULL); 1140ccd979bdSMark Fasheh if (status < 0) { 1141ccd979bdSMark Fasheh mlog_errno(status); 1142ccd979bdSMark Fasheh goto bail; 1143ccd979bdSMark Fasheh } 1144ccd979bdSMark Fasheh 1145ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 1146ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 1147ccd979bdSMark Fasheh 1148dd4a2c2bSMark Fasheh /* We are reading journal data which should not 1149dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 1150da1e9098SJoel Becker status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), 1151da1e9098SJoel Becker p_blkno, p_blocks, bhs); 1152ccd979bdSMark Fasheh if (status < 0) { 1153ccd979bdSMark Fasheh mlog_errno(status); 1154ccd979bdSMark Fasheh goto bail; 1155ccd979bdSMark Fasheh } 1156ccd979bdSMark Fasheh 1157ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 1158ccd979bdSMark Fasheh brelse(bhs[i]); 1159ccd979bdSMark Fasheh bhs[i] = NULL; 1160ccd979bdSMark Fasheh } 1161ccd979bdSMark Fasheh 1162ccd979bdSMark Fasheh v_blkno += p_blocks; 1163ccd979bdSMark Fasheh } 1164ccd979bdSMark Fasheh 1165ccd979bdSMark Fasheh bail: 1166ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 1167ccd979bdSMark Fasheh brelse(bhs[i]); 1168ccd979bdSMark Fasheh return status; 1169ccd979bdSMark Fasheh } 1170ccd979bdSMark Fasheh 1171ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 1172ccd979bdSMark Fasheh struct list_head lri_list; 1173ccd979bdSMark Fasheh int lri_slot; 1174ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 1175ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 11762205363dSJan Kara struct ocfs2_quota_recovery *lri_qrec; 1177ccd979bdSMark Fasheh }; 1178ccd979bdSMark Fasheh 1179ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 1180ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 1181ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 1182ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 1183ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 1184ccd979bdSMark Fasheh * recovering nodes into account. 1185ccd979bdSMark Fasheh * 1186ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 1187ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 1188ccd979bdSMark Fasheh */ 1189c4028958SDavid Howells void ocfs2_complete_recovery(struct work_struct *work) 1190ccd979bdSMark Fasheh { 1191b4107950STao Ma int ret = 0; 1192c4028958SDavid Howells struct ocfs2_journal *journal = 1193c4028958SDavid Howells container_of(work, struct ocfs2_journal, j_recovery_work); 1194c4028958SDavid Howells struct ocfs2_super *osb = journal->j_osb; 1195ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 1196800deef3SChristoph Hellwig struct ocfs2_la_recovery_item *item, *n; 11972205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1198ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 1199ccd979bdSMark Fasheh 1200b4107950STao Ma trace_ocfs2_complete_recovery( 1201b4107950STao Ma (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1202ccd979bdSMark Fasheh 1203ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1204ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1205ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1206ccd979bdSMark Fasheh 1207800deef3SChristoph Hellwig list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1208ccd979bdSMark Fasheh list_del_init(&item->lri_list); 1209ccd979bdSMark Fasheh 121019ece546SJan Kara ocfs2_wait_on_quotas(osb); 121119ece546SJan Kara 1212ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 1213b4107950STao Ma tl_dinode = item->lri_tl_dinode; 1214b4107950STao Ma qrec = item->lri_qrec; 1215ccd979bdSMark Fasheh 1216b4107950STao Ma trace_ocfs2_complete_recovery_slot(item->lri_slot, 1217b4107950STao Ma la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1218b4107950STao Ma tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1219b4107950STao Ma qrec); 1220b4107950STao Ma 1221b4107950STao Ma if (la_dinode) { 1222ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 1223ccd979bdSMark Fasheh la_dinode); 1224ccd979bdSMark Fasheh if (ret < 0) 1225ccd979bdSMark Fasheh mlog_errno(ret); 1226ccd979bdSMark Fasheh 1227ccd979bdSMark Fasheh kfree(la_dinode); 1228ccd979bdSMark Fasheh } 1229ccd979bdSMark Fasheh 1230ccd979bdSMark Fasheh if (tl_dinode) { 1231ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 1232ccd979bdSMark Fasheh tl_dinode); 1233ccd979bdSMark Fasheh if (ret < 0) 1234ccd979bdSMark Fasheh mlog_errno(ret); 1235ccd979bdSMark Fasheh 1236ccd979bdSMark Fasheh kfree(tl_dinode); 1237ccd979bdSMark Fasheh } 1238ccd979bdSMark Fasheh 1239ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 1240ccd979bdSMark Fasheh if (ret < 0) 1241ccd979bdSMark Fasheh mlog_errno(ret); 1242ccd979bdSMark Fasheh 12432205363dSJan Kara if (qrec) { 12442205363dSJan Kara ret = ocfs2_finish_quota_recovery(osb, qrec, 12452205363dSJan Kara item->lri_slot); 12462205363dSJan Kara if (ret < 0) 12472205363dSJan Kara mlog_errno(ret); 12482205363dSJan Kara /* Recovery info is already freed now */ 12492205363dSJan Kara } 12502205363dSJan Kara 1251ccd979bdSMark Fasheh kfree(item); 1252ccd979bdSMark Fasheh } 1253ccd979bdSMark Fasheh 1254b4107950STao Ma trace_ocfs2_complete_recovery_end(ret); 1255ccd979bdSMark Fasheh } 1256ccd979bdSMark Fasheh 1257ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 1258ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 1259ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 1260ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1261ccd979bdSMark Fasheh int slot_num, 1262ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 12632205363dSJan Kara struct ocfs2_dinode *tl_dinode, 12642205363dSJan Kara struct ocfs2_quota_recovery *qrec) 1265ccd979bdSMark Fasheh { 1266ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 1267ccd979bdSMark Fasheh 1268afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1269ccd979bdSMark Fasheh if (!item) { 1270ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 1271ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 1272ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 1273ccd979bdSMark Fasheh kfree(la_dinode); 1274ccd979bdSMark Fasheh kfree(tl_dinode); 1275ccd979bdSMark Fasheh 12762205363dSJan Kara if (qrec) 12772205363dSJan Kara ocfs2_free_quota_recovery(qrec); 12782205363dSJan Kara 1279ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 1280ccd979bdSMark Fasheh return; 1281ccd979bdSMark Fasheh } 1282ccd979bdSMark Fasheh 1283ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 1284ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 1285ccd979bdSMark Fasheh item->lri_slot = slot_num; 1286ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 12872205363dSJan Kara item->lri_qrec = qrec; 1288ccd979bdSMark Fasheh 1289ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 1290ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 1291ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 1292ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 1293ccd979bdSMark Fasheh } 1294ccd979bdSMark Fasheh 1295ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 12969140db04SSrinivas Eeda * recovery for it's own and offline slot(s). */ 1297ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1298ccd979bdSMark Fasheh { 1299ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1300ccd979bdSMark Fasheh 130110b3dd76SSunil Mushran if (ocfs2_is_hard_readonly(osb)) 130210b3dd76SSunil Mushran return; 130310b3dd76SSunil Mushran 13049140db04SSrinivas Eeda /* No need to queue up our truncate_log as regular cleanup will catch 13059140db04SSrinivas Eeda * that */ 13069140db04SSrinivas Eeda ocfs2_queue_recovery_completion(journal, osb->slot_num, 13079140db04SSrinivas Eeda osb->local_alloc_copy, NULL, NULL); 1308ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1309ccd979bdSMark Fasheh 1310ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 1311ccd979bdSMark Fasheh osb->dirty = 0; 13129140db04SSrinivas Eeda 13139140db04SSrinivas Eeda /* queue to recover orphan slots for all offline slots */ 13149140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 13159140db04SSrinivas Eeda ocfs2_queue_replay_slots(osb); 13169140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1317ccd979bdSMark Fasheh } 1318ccd979bdSMark Fasheh 13192205363dSJan Kara void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 13202205363dSJan Kara { 13212205363dSJan Kara if (osb->quota_rec) { 13222205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, 13232205363dSJan Kara osb->slot_num, 13242205363dSJan Kara NULL, 13252205363dSJan Kara NULL, 13262205363dSJan Kara osb->quota_rec); 13272205363dSJan Kara osb->quota_rec = NULL; 13282205363dSJan Kara } 13292205363dSJan Kara } 13302205363dSJan Kara 1331ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1332ccd979bdSMark Fasheh { 13332205363dSJan Kara int status, node_num, slot_num; 1334ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1335553abd04SJoel Becker struct ocfs2_recovery_map *rm = osb->recovery_map; 13362205363dSJan Kara int *rm_quota = NULL; 13372205363dSJan Kara int rm_quota_used = 0, i; 13382205363dSJan Kara struct ocfs2_quota_recovery *qrec; 1339ccd979bdSMark Fasheh 1340ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1341ccd979bdSMark Fasheh if (status < 0) { 1342ccd979bdSMark Fasheh goto bail; 1343ccd979bdSMark Fasheh } 1344ccd979bdSMark Fasheh 13452205363dSJan Kara rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); 13462205363dSJan Kara if (!rm_quota) { 13472205363dSJan Kara status = -ENOMEM; 13482205363dSJan Kara goto bail; 13492205363dSJan Kara } 1350ccd979bdSMark Fasheh restart: 1351ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1352ccd979bdSMark Fasheh if (status < 0) { 1353ccd979bdSMark Fasheh mlog_errno(status); 1354ccd979bdSMark Fasheh goto bail; 1355ccd979bdSMark Fasheh } 1356ccd979bdSMark Fasheh 13579140db04SSrinivas Eeda status = ocfs2_compute_replay_slots(osb); 13589140db04SSrinivas Eeda if (status < 0) 13599140db04SSrinivas Eeda mlog_errno(status); 13609140db04SSrinivas Eeda 13619140db04SSrinivas Eeda /* queue recovery for our own slot */ 13629140db04SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 13639140db04SSrinivas Eeda NULL, NULL); 13649140db04SSrinivas Eeda 1365553abd04SJoel Becker spin_lock(&osb->osb_lock); 1366553abd04SJoel Becker while (rm->rm_used) { 1367553abd04SJoel Becker /* It's always safe to remove entry zero, as we won't 1368553abd04SJoel Becker * clear it until ocfs2_recover_node() has succeeded. */ 1369553abd04SJoel Becker node_num = rm->rm_entries[0]; 1370553abd04SJoel Becker spin_unlock(&osb->osb_lock); 13712205363dSJan Kara slot_num = ocfs2_node_num_to_slot(osb, node_num); 1372b4107950STao Ma trace_ocfs2_recovery_thread_node(node_num, slot_num); 13732205363dSJan Kara if (slot_num == -ENOENT) { 13742205363dSJan Kara status = 0; 13752205363dSJan Kara goto skip_recovery; 13762205363dSJan Kara } 1377ccd979bdSMark Fasheh 13782205363dSJan Kara /* It is a bit subtle with quota recovery. We cannot do it 13792205363dSJan Kara * immediately because we have to obtain cluster locks from 13802205363dSJan Kara * quota files and we also don't want to just skip it because 13812205363dSJan Kara * then quota usage would be out of sync until some node takes 13822205363dSJan Kara * the slot. So we remember which nodes need quota recovery 13832205363dSJan Kara * and when everything else is done, we recover quotas. */ 13842205363dSJan Kara for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); 13852205363dSJan Kara if (i == rm_quota_used) 13862205363dSJan Kara rm_quota[rm_quota_used++] = slot_num; 13872205363dSJan Kara 13882205363dSJan Kara status = ocfs2_recover_node(osb, node_num, slot_num); 13892205363dSJan Kara skip_recovery: 1390553abd04SJoel Becker if (!status) { 1391553abd04SJoel Becker ocfs2_recovery_map_clear(osb, node_num); 1392553abd04SJoel Becker } else { 1393ccd979bdSMark Fasheh mlog(ML_ERROR, 1394ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1395ccd979bdSMark Fasheh status, node_num, 1396ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1397ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1398ccd979bdSMark Fasheh } 1399ccd979bdSMark Fasheh 1400553abd04SJoel Becker spin_lock(&osb->osb_lock); 1401ccd979bdSMark Fasheh } 1402553abd04SJoel Becker spin_unlock(&osb->osb_lock); 1403b4107950STao Ma trace_ocfs2_recovery_thread_end(status); 1404553abd04SJoel Becker 1405539d8264SSunil Mushran /* Refresh all journal recovery generations from disk */ 1406539d8264SSunil Mushran status = ocfs2_check_journals_nolocks(osb); 1407539d8264SSunil Mushran status = (status == -EROFS) ? 0 : status; 1408539d8264SSunil Mushran if (status < 0) 1409539d8264SSunil Mushran mlog_errno(status); 1410539d8264SSunil Mushran 14112205363dSJan Kara /* Now it is right time to recover quotas... We have to do this under 14122205363dSJan Kara * superblock lock so that no one can start using the slot (and crash) 14132205363dSJan Kara * before we recover it */ 14142205363dSJan Kara for (i = 0; i < rm_quota_used; i++) { 14152205363dSJan Kara qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 14162205363dSJan Kara if (IS_ERR(qrec)) { 14172205363dSJan Kara status = PTR_ERR(qrec); 14182205363dSJan Kara mlog_errno(status); 14192205363dSJan Kara continue; 14202205363dSJan Kara } 14212205363dSJan Kara ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], 14222205363dSJan Kara NULL, NULL, qrec); 14232205363dSJan Kara } 14242205363dSJan Kara 1425ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1426ccd979bdSMark Fasheh 14279140db04SSrinivas Eeda /* queue recovery for offline slots */ 14289140db04SSrinivas Eeda ocfs2_queue_replay_slots(osb); 1429ccd979bdSMark Fasheh 1430ccd979bdSMark Fasheh bail: 1431c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1432553abd04SJoel Becker if (!status && !ocfs2_recovery_completed(osb)) { 1433c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1434ccd979bdSMark Fasheh goto restart; 1435ccd979bdSMark Fasheh } 1436ccd979bdSMark Fasheh 14379140db04SSrinivas Eeda ocfs2_free_replay_slots(osb); 1438ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1439ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1440ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1441ccd979bdSMark Fasheh 1442c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1443ccd979bdSMark Fasheh 14442205363dSJan Kara kfree(rm_quota); 14452205363dSJan Kara 1446ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1447ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1448ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1449ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1450ccd979bdSMark Fasheh return status; 1451ccd979bdSMark Fasheh } 1452ccd979bdSMark Fasheh 1453ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1454ccd979bdSMark Fasheh { 1455c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1456b4107950STao Ma 1457b4107950STao Ma trace_ocfs2_recovery_thread(node_num, osb->node_num, 1458b4107950STao Ma osb->disable_recovery, osb->recovery_thread_task, 1459b4107950STao Ma osb->disable_recovery ? 1460b4107950STao Ma -1 : ocfs2_recovery_map_set(osb, node_num)); 1461b4107950STao Ma 1462ccd979bdSMark Fasheh if (osb->disable_recovery) 1463ccd979bdSMark Fasheh goto out; 1464ccd979bdSMark Fasheh 1465ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1466ccd979bdSMark Fasheh goto out; 1467ccd979bdSMark Fasheh 1468ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 146978427043SMark Fasheh "ocfs2rec"); 1470ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1471ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1472ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1473ccd979bdSMark Fasheh } 1474ccd979bdSMark Fasheh 1475ccd979bdSMark Fasheh out: 1476c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1477ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1478ccd979bdSMark Fasheh } 1479ccd979bdSMark Fasheh 1480539d8264SSunil Mushran static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1481539d8264SSunil Mushran int slot_num, 1482539d8264SSunil Mushran struct buffer_head **bh, 1483539d8264SSunil Mushran struct inode **ret_inode) 1484539d8264SSunil Mushran { 1485539d8264SSunil Mushran int status = -EACCES; 1486539d8264SSunil Mushran struct inode *inode = NULL; 1487539d8264SSunil Mushran 1488539d8264SSunil Mushran BUG_ON(slot_num >= osb->max_slots); 1489539d8264SSunil Mushran 1490539d8264SSunil Mushran inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1491539d8264SSunil Mushran slot_num); 1492539d8264SSunil Mushran if (!inode || is_bad_inode(inode)) { 1493539d8264SSunil Mushran mlog_errno(status); 1494539d8264SSunil Mushran goto bail; 1495539d8264SSunil Mushran } 1496539d8264SSunil Mushran SET_INODE_JOURNAL(inode); 1497539d8264SSunil Mushran 1498b657c95cSJoel Becker status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1499539d8264SSunil Mushran if (status < 0) { 1500539d8264SSunil Mushran mlog_errno(status); 1501539d8264SSunil Mushran goto bail; 1502539d8264SSunil Mushran } 1503539d8264SSunil Mushran 1504539d8264SSunil Mushran status = 0; 1505539d8264SSunil Mushran 1506539d8264SSunil Mushran bail: 1507539d8264SSunil Mushran if (inode) { 1508539d8264SSunil Mushran if (status || !ret_inode) 1509539d8264SSunil Mushran iput(inode); 1510539d8264SSunil Mushran else 1511539d8264SSunil Mushran *ret_inode = inode; 1512539d8264SSunil Mushran } 1513539d8264SSunil Mushran return status; 1514539d8264SSunil Mushran } 1515539d8264SSunil Mushran 1516ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1517ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1518ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1519ccd979bdSMark Fasheh int node_num, 1520ccd979bdSMark Fasheh int slot_num) 1521ccd979bdSMark Fasheh { 1522ccd979bdSMark Fasheh int status; 1523ccd979bdSMark Fasheh int got_lock = 0; 1524ccd979bdSMark Fasheh unsigned int flags; 1525ccd979bdSMark Fasheh struct inode *inode = NULL; 1526ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1527ccd979bdSMark Fasheh journal_t *journal = NULL; 1528ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1529539d8264SSunil Mushran u32 slot_reco_gen; 1530ccd979bdSMark Fasheh 1531539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1532539d8264SSunil Mushran if (status) { 1533ccd979bdSMark Fasheh mlog_errno(status); 1534ccd979bdSMark Fasheh goto done; 1535ccd979bdSMark Fasheh } 1536539d8264SSunil Mushran 1537539d8264SSunil Mushran fe = (struct ocfs2_dinode *)bh->b_data; 1538539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1539539d8264SSunil Mushran brelse(bh); 1540539d8264SSunil Mushran bh = NULL; 1541539d8264SSunil Mushran 1542539d8264SSunil Mushran /* 1543539d8264SSunil Mushran * As the fs recovery is asynchronous, there is a small chance that 1544539d8264SSunil Mushran * another node mounted (and recovered) the slot before the recovery 1545539d8264SSunil Mushran * thread could get the lock. To handle that, we dirty read the journal 1546539d8264SSunil Mushran * inode for that slot to get the recovery generation. If it is 1547539d8264SSunil Mushran * different than what we expected, the slot has been recovered. 1548539d8264SSunil Mushran * If not, it needs recovery. 1549539d8264SSunil Mushran */ 1550539d8264SSunil Mushran if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1551b4107950STao Ma trace_ocfs2_replay_journal_recovered(slot_num, 1552539d8264SSunil Mushran osb->slot_recovery_generations[slot_num], slot_reco_gen); 1553539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1554539d8264SSunil Mushran status = -EBUSY; 1555ccd979bdSMark Fasheh goto done; 1556ccd979bdSMark Fasheh } 1557539d8264SSunil Mushran 1558539d8264SSunil Mushran /* Continue with recovery as the journal has not yet been recovered */ 1559ccd979bdSMark Fasheh 1560e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1561ccd979bdSMark Fasheh if (status < 0) { 1562b4107950STao Ma trace_ocfs2_replay_journal_lock_err(status); 1563ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1564ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1565ccd979bdSMark Fasheh goto done; 1566ccd979bdSMark Fasheh } 1567ccd979bdSMark Fasheh got_lock = 1; 1568ccd979bdSMark Fasheh 1569ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1570ccd979bdSMark Fasheh 1571ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1572539d8264SSunil Mushran slot_reco_gen = ocfs2_get_recovery_generation(fe); 1573ccd979bdSMark Fasheh 1574ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1575b4107950STao Ma trace_ocfs2_replay_journal_skip(node_num); 1576539d8264SSunil Mushran /* Refresh recovery generation for the slot */ 1577539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1578ccd979bdSMark Fasheh goto done; 1579ccd979bdSMark Fasheh } 1580ccd979bdSMark Fasheh 15819140db04SSrinivas Eeda /* we need to run complete recovery for offline orphan slots */ 15829140db04SSrinivas Eeda ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 15839140db04SSrinivas Eeda 1584619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1585619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1586619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1587ccd979bdSMark Fasheh 1588ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1589ccd979bdSMark Fasheh 1590ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1591ccd979bdSMark Fasheh if (status < 0) { 1592ccd979bdSMark Fasheh mlog_errno(status); 1593ccd979bdSMark Fasheh goto done; 1594ccd979bdSMark Fasheh } 1595ccd979bdSMark Fasheh 15962b4e30fbSJoel Becker journal = jbd2_journal_init_inode(inode); 1597ccd979bdSMark Fasheh if (journal == NULL) { 1598ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1599ccd979bdSMark Fasheh status = -EIO; 1600ccd979bdSMark Fasheh goto done; 1601ccd979bdSMark Fasheh } 1602ccd979bdSMark Fasheh 16032b4e30fbSJoel Becker status = jbd2_journal_load(journal); 1604ccd979bdSMark Fasheh if (status < 0) { 1605ccd979bdSMark Fasheh mlog_errno(status); 1606ccd979bdSMark Fasheh if (!igrab(inode)) 1607ccd979bdSMark Fasheh BUG(); 16082b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1609ccd979bdSMark Fasheh goto done; 1610ccd979bdSMark Fasheh } 1611ccd979bdSMark Fasheh 1612ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1613ccd979bdSMark Fasheh 1614ccd979bdSMark Fasheh /* wipe the journal */ 16152b4e30fbSJoel Becker jbd2_journal_lock_updates(journal); 16162b4e30fbSJoel Becker status = jbd2_journal_flush(journal); 16172b4e30fbSJoel Becker jbd2_journal_unlock_updates(journal); 1618ccd979bdSMark Fasheh if (status < 0) 1619ccd979bdSMark Fasheh mlog_errno(status); 1620ccd979bdSMark Fasheh 1621ccd979bdSMark Fasheh /* This will mark the node clean */ 1622ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1623ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1624ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1625ccd979bdSMark Fasheh 1626539d8264SSunil Mushran /* Increment recovery generation to indicate successful recovery */ 1627539d8264SSunil Mushran ocfs2_bump_recovery_generation(fe); 1628539d8264SSunil Mushran osb->slot_recovery_generations[slot_num] = 1629539d8264SSunil Mushran ocfs2_get_recovery_generation(fe); 1630539d8264SSunil Mushran 163113723d00SJoel Becker ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 16328cb471e8SJoel Becker status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1633ccd979bdSMark Fasheh if (status < 0) 1634ccd979bdSMark Fasheh mlog_errno(status); 1635ccd979bdSMark Fasheh 1636ccd979bdSMark Fasheh if (!igrab(inode)) 1637ccd979bdSMark Fasheh BUG(); 1638ccd979bdSMark Fasheh 16392b4e30fbSJoel Becker jbd2_journal_destroy(journal); 1640ccd979bdSMark Fasheh 1641619c200dSSunil Mushran printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1642619c200dSSunil Mushran "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1643619c200dSSunil Mushran MINOR(osb->sb->s_dev)); 1644ccd979bdSMark Fasheh done: 1645ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1646ccd979bdSMark Fasheh if (got_lock) 1647e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1648ccd979bdSMark Fasheh 1649ccd979bdSMark Fasheh if (inode) 1650ccd979bdSMark Fasheh iput(inode); 1651ccd979bdSMark Fasheh 1652ccd979bdSMark Fasheh brelse(bh); 1653ccd979bdSMark Fasheh 1654ccd979bdSMark Fasheh return status; 1655ccd979bdSMark Fasheh } 1656ccd979bdSMark Fasheh 1657ccd979bdSMark Fasheh /* 1658ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1659ccd979bdSMark Fasheh * - Replay it's journal 1660ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1661ccd979bdSMark Fasheh * - Stamp a clean truncate log 1662ccd979bdSMark Fasheh * - Mark the node clean 1663ccd979bdSMark Fasheh * 1664ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1665ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1666ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1667ccd979bdSMark Fasheh * far less concerning. 1668ccd979bdSMark Fasheh */ 1669ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 16702205363dSJan Kara int node_num, int slot_num) 1671ccd979bdSMark Fasheh { 1672ccd979bdSMark Fasheh int status = 0; 1673ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1674ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1675ccd979bdSMark Fasheh 1676b4107950STao Ma trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1677ccd979bdSMark Fasheh 1678ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1679ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1680ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1681ccd979bdSMark Fasheh 1682ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1683ccd979bdSMark Fasheh if (status < 0) { 1684539d8264SSunil Mushran if (status == -EBUSY) { 1685b4107950STao Ma trace_ocfs2_recover_node_skip(slot_num, node_num); 1686539d8264SSunil Mushran status = 0; 1687539d8264SSunil Mushran goto done; 1688539d8264SSunil Mushran } 1689ccd979bdSMark Fasheh mlog_errno(status); 1690ccd979bdSMark Fasheh goto done; 1691ccd979bdSMark Fasheh } 1692ccd979bdSMark Fasheh 1693ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1694ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1695ccd979bdSMark Fasheh if (status < 0) { 1696ccd979bdSMark Fasheh mlog_errno(status); 1697ccd979bdSMark Fasheh goto done; 1698ccd979bdSMark Fasheh } 1699ccd979bdSMark Fasheh 1700ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1701ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1702ccd979bdSMark Fasheh * recovery. */ 1703ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1704ccd979bdSMark Fasheh if (status < 0) 1705ccd979bdSMark Fasheh mlog_errno(status); 1706ccd979bdSMark Fasheh 1707ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1708ccd979bdSMark Fasheh * harmful place to get an error... */ 17098e8a4603SMark Fasheh status = ocfs2_clear_slot(osb, slot_num); 1710ccd979bdSMark Fasheh if (status < 0) 1711ccd979bdSMark Fasheh mlog_errno(status); 1712ccd979bdSMark Fasheh 1713ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1714ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 17152205363dSJan Kara tl_copy, NULL); 1716ccd979bdSMark Fasheh 1717ccd979bdSMark Fasheh status = 0; 1718ccd979bdSMark Fasheh done: 1719ccd979bdSMark Fasheh 1720ccd979bdSMark Fasheh return status; 1721ccd979bdSMark Fasheh } 1722ccd979bdSMark Fasheh 1723ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1724ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1725ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1726ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1727ccd979bdSMark Fasheh int slot_num) 1728ccd979bdSMark Fasheh { 1729ccd979bdSMark Fasheh int status, flags; 1730ccd979bdSMark Fasheh struct inode *inode = NULL; 1731ccd979bdSMark Fasheh 1732ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1733ccd979bdSMark Fasheh slot_num); 1734ccd979bdSMark Fasheh if (inode == NULL) { 1735ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1736ccd979bdSMark Fasheh status = -EACCES; 1737ccd979bdSMark Fasheh goto bail; 1738ccd979bdSMark Fasheh } 1739ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1740ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1741ccd979bdSMark Fasheh iput(inode); 1742ccd979bdSMark Fasheh inode = NULL; 1743ccd979bdSMark Fasheh status = -EACCES; 1744ccd979bdSMark Fasheh goto bail; 1745ccd979bdSMark Fasheh } 1746ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1747ccd979bdSMark Fasheh 1748ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1749e63aecb6SMark Fasheh status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1750ccd979bdSMark Fasheh if (status < 0) { 1751ccd979bdSMark Fasheh if (status != -EAGAIN) 1752ccd979bdSMark Fasheh mlog_errno(status); 1753ccd979bdSMark Fasheh goto bail; 1754ccd979bdSMark Fasheh } 1755ccd979bdSMark Fasheh 1756e63aecb6SMark Fasheh ocfs2_inode_unlock(inode, 1); 1757ccd979bdSMark Fasheh bail: 1758ccd979bdSMark Fasheh if (inode) 1759ccd979bdSMark Fasheh iput(inode); 1760ccd979bdSMark Fasheh 1761ccd979bdSMark Fasheh return status; 1762ccd979bdSMark Fasheh } 1763ccd979bdSMark Fasheh 1764ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1765ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1766ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1767ccd979bdSMark Fasheh { 1768d85b20e4SJoel Becker unsigned int node_num; 1769d85b20e4SJoel Becker int status, i; 1770a1af7d15SMark Fasheh u32 gen; 1771539d8264SSunil Mushran struct buffer_head *bh = NULL; 1772539d8264SSunil Mushran struct ocfs2_dinode *di; 1773ccd979bdSMark Fasheh 1774ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1775ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1776ccd979bdSMark Fasheh 1777d85b20e4SJoel Becker for (i = 0; i < osb->max_slots; i++) { 1778539d8264SSunil Mushran /* Read journal inode to get the recovery generation */ 1779539d8264SSunil Mushran status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1780539d8264SSunil Mushran if (status) { 1781539d8264SSunil Mushran mlog_errno(status); 1782539d8264SSunil Mushran goto bail; 1783539d8264SSunil Mushran } 1784539d8264SSunil Mushran di = (struct ocfs2_dinode *)bh->b_data; 1785a1af7d15SMark Fasheh gen = ocfs2_get_recovery_generation(di); 1786539d8264SSunil Mushran brelse(bh); 1787539d8264SSunil Mushran bh = NULL; 1788539d8264SSunil Mushran 1789a1af7d15SMark Fasheh spin_lock(&osb->osb_lock); 1790a1af7d15SMark Fasheh osb->slot_recovery_generations[i] = gen; 1791a1af7d15SMark Fasheh 1792b4107950STao Ma trace_ocfs2_mark_dead_nodes(i, 1793539d8264SSunil Mushran osb->slot_recovery_generations[i]); 1794539d8264SSunil Mushran 1795a1af7d15SMark Fasheh if (i == osb->slot_num) { 1796a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1797ccd979bdSMark Fasheh continue; 1798a1af7d15SMark Fasheh } 1799d85b20e4SJoel Becker 1800d85b20e4SJoel Becker status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1801a1af7d15SMark Fasheh if (status == -ENOENT) { 1802a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1803ccd979bdSMark Fasheh continue; 1804a1af7d15SMark Fasheh } 1805ccd979bdSMark Fasheh 1806a1af7d15SMark Fasheh if (__ocfs2_recovery_map_test(osb, node_num)) { 1807a1af7d15SMark Fasheh spin_unlock(&osb->osb_lock); 1808ccd979bdSMark Fasheh continue; 1809a1af7d15SMark Fasheh } 1810d85b20e4SJoel Becker spin_unlock(&osb->osb_lock); 1811ccd979bdSMark Fasheh 1812ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1813ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1814ccd979bdSMark Fasheh * file here to test if he's alive. */ 1815ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1816ccd979bdSMark Fasheh if (!status) { 1817ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1818ccd979bdSMark Fasheh * the recovery thread can't race us on 1819ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1820ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1821ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1822ccd979bdSMark Fasheh mlog_errno(status); 1823ccd979bdSMark Fasheh goto bail; 1824ccd979bdSMark Fasheh } 1825ccd979bdSMark Fasheh } 1826ccd979bdSMark Fasheh 1827ccd979bdSMark Fasheh status = 0; 1828ccd979bdSMark Fasheh bail: 1829ccd979bdSMark Fasheh return status; 1830ccd979bdSMark Fasheh } 1831ccd979bdSMark Fasheh 183283273932SSrinivas Eeda /* 183383273932SSrinivas Eeda * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 183483273932SSrinivas Eeda * randomness to the timeout to minimize multple nodes firing the timer at the 183583273932SSrinivas Eeda * same time. 183683273932SSrinivas Eeda */ 183783273932SSrinivas Eeda static inline unsigned long ocfs2_orphan_scan_timeout(void) 183883273932SSrinivas Eeda { 183983273932SSrinivas Eeda unsigned long time; 184083273932SSrinivas Eeda 184183273932SSrinivas Eeda get_random_bytes(&time, sizeof(time)); 184283273932SSrinivas Eeda time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 184383273932SSrinivas Eeda return msecs_to_jiffies(time); 184483273932SSrinivas Eeda } 184583273932SSrinivas Eeda 184683273932SSrinivas Eeda /* 184783273932SSrinivas Eeda * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 184883273932SSrinivas Eeda * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 184983273932SSrinivas Eeda * is done to catch any orphans that are left over in orphan directories. 185083273932SSrinivas Eeda * 1851a035bff6SSunil Mushran * It scans all slots, even ones that are in use. It does so to handle the 1852a035bff6SSunil Mushran * case described below: 1853a035bff6SSunil Mushran * 1854a035bff6SSunil Mushran * Node 1 has an inode it was using. The dentry went away due to memory 1855a035bff6SSunil Mushran * pressure. Node 1 closes the inode, but it's on the free list. The node 1856a035bff6SSunil Mushran * has the open lock. 1857a035bff6SSunil Mushran * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1858a035bff6SSunil Mushran * but node 1 has no dentry and doesn't get the message. It trylocks the 1859a035bff6SSunil Mushran * open lock, sees that another node has a PR, and does nothing. 1860a035bff6SSunil Mushran * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1861a035bff6SSunil Mushran * open lock, sees the PR still, and does nothing. 1862a035bff6SSunil Mushran * Basically, we have to trigger an orphan iput on node 1. The only way 1863a035bff6SSunil Mushran * for this to happen is if node 1 runs node 2's orphan dir. 1864a035bff6SSunil Mushran * 186583273932SSrinivas Eeda * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 186683273932SSrinivas Eeda * seconds. It gets an EX lock on os_lockres and checks sequence number 186783273932SSrinivas Eeda * stored in LVB. If the sequence number has changed, it means some other 186883273932SSrinivas Eeda * node has done the scan. This node skips the scan and tracks the 186983273932SSrinivas Eeda * sequence number. If the sequence number didn't change, it means a scan 187083273932SSrinivas Eeda * hasn't happened. The node queues a scan and increments the 187183273932SSrinivas Eeda * sequence number in the LVB. 187283273932SSrinivas Eeda */ 187383273932SSrinivas Eeda void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 187483273932SSrinivas Eeda { 187583273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 187683273932SSrinivas Eeda int status, i; 187783273932SSrinivas Eeda u32 seqno = 0; 187883273932SSrinivas Eeda 187983273932SSrinivas Eeda os = &osb->osb_orphan_scan; 188083273932SSrinivas Eeda 1881692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1882692684e1SSunil Mushran goto out; 1883692684e1SSunil Mushran 1884b4107950STao Ma trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1885b4107950STao Ma atomic_read(&os->os_state)); 1886b4107950STao Ma 1887df152c24SSunil Mushran status = ocfs2_orphan_scan_lock(osb, &seqno); 188883273932SSrinivas Eeda if (status < 0) { 188983273932SSrinivas Eeda if (status != -EAGAIN) 189083273932SSrinivas Eeda mlog_errno(status); 189183273932SSrinivas Eeda goto out; 189283273932SSrinivas Eeda } 189383273932SSrinivas Eeda 1894692684e1SSunil Mushran /* Do no queue the tasks if the volume is being umounted */ 1895692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1896692684e1SSunil Mushran goto unlock; 1897692684e1SSunil Mushran 189883273932SSrinivas Eeda if (os->os_seqno != seqno) { 189983273932SSrinivas Eeda os->os_seqno = seqno; 190083273932SSrinivas Eeda goto unlock; 190183273932SSrinivas Eeda } 190283273932SSrinivas Eeda 190383273932SSrinivas Eeda for (i = 0; i < osb->max_slots; i++) 190483273932SSrinivas Eeda ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 190583273932SSrinivas Eeda NULL); 190683273932SSrinivas Eeda /* 190783273932SSrinivas Eeda * We queued a recovery on orphan slots, increment the sequence 190883273932SSrinivas Eeda * number and update LVB so other node will skip the scan for a while 190983273932SSrinivas Eeda */ 191083273932SSrinivas Eeda seqno++; 191115633a22SSrinivas Eeda os->os_count++; 191215633a22SSrinivas Eeda os->os_scantime = CURRENT_TIME; 191383273932SSrinivas Eeda unlock: 1914df152c24SSunil Mushran ocfs2_orphan_scan_unlock(osb, seqno); 191583273932SSrinivas Eeda out: 1916b4107950STao Ma trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1917b4107950STao Ma atomic_read(&os->os_state)); 191883273932SSrinivas Eeda return; 191983273932SSrinivas Eeda } 192083273932SSrinivas Eeda 192183273932SSrinivas Eeda /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 192283273932SSrinivas Eeda void ocfs2_orphan_scan_work(struct work_struct *work) 192383273932SSrinivas Eeda { 192483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 192583273932SSrinivas Eeda struct ocfs2_super *osb; 192683273932SSrinivas Eeda 192783273932SSrinivas Eeda os = container_of(work, struct ocfs2_orphan_scan, 192883273932SSrinivas Eeda os_orphan_scan_work.work); 192983273932SSrinivas Eeda osb = os->os_osb; 193083273932SSrinivas Eeda 193183273932SSrinivas Eeda mutex_lock(&os->os_lock); 193283273932SSrinivas Eeda ocfs2_queue_orphan_scan(osb); 1933692684e1SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 193440f165f4STao Ma queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 193583273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 193683273932SSrinivas Eeda mutex_unlock(&os->os_lock); 193783273932SSrinivas Eeda } 193883273932SSrinivas Eeda 193983273932SSrinivas Eeda void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 194083273932SSrinivas Eeda { 194183273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 194283273932SSrinivas Eeda 194383273932SSrinivas Eeda os = &osb->osb_orphan_scan; 1944df152c24SSunil Mushran if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1945692684e1SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 194683273932SSrinivas Eeda mutex_lock(&os->os_lock); 194783273932SSrinivas Eeda cancel_delayed_work(&os->os_orphan_scan_work); 194883273932SSrinivas Eeda mutex_unlock(&os->os_lock); 194983273932SSrinivas Eeda } 1950df152c24SSunil Mushran } 195183273932SSrinivas Eeda 1952df152c24SSunil Mushran void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 195383273932SSrinivas Eeda { 195483273932SSrinivas Eeda struct ocfs2_orphan_scan *os; 195583273932SSrinivas Eeda 195683273932SSrinivas Eeda os = &osb->osb_orphan_scan; 195783273932SSrinivas Eeda os->os_osb = osb; 195815633a22SSrinivas Eeda os->os_count = 0; 19593211949fSSunil Mushran os->os_seqno = 0; 196083273932SSrinivas Eeda mutex_init(&os->os_lock); 1961df152c24SSunil Mushran INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 19628b712cd5SJeff Mahoney } 196383273932SSrinivas Eeda 19648b712cd5SJeff Mahoney void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 19658b712cd5SJeff Mahoney { 19668b712cd5SJeff Mahoney struct ocfs2_orphan_scan *os; 19678b712cd5SJeff Mahoney 19688b712cd5SJeff Mahoney os = &osb->osb_orphan_scan; 19698b712cd5SJeff Mahoney os->os_scantime = CURRENT_TIME; 1970df152c24SSunil Mushran if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 1971df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 1972df152c24SSunil Mushran else { 1973df152c24SSunil Mushran atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 197440f165f4STao Ma queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 197583273932SSrinivas Eeda ocfs2_orphan_scan_timeout()); 1976df152c24SSunil Mushran } 197783273932SSrinivas Eeda } 197883273932SSrinivas Eeda 19795eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv { 19803704412bSAl Viro struct dir_context ctx; 19815eae5b96SMark Fasheh struct inode *head; 19825eae5b96SMark Fasheh struct ocfs2_super *osb; 19835eae5b96SMark Fasheh }; 19845eae5b96SMark Fasheh 19855eae5b96SMark Fasheh static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, 19865eae5b96SMark Fasheh loff_t pos, u64 ino, unsigned type) 19875eae5b96SMark Fasheh { 19885eae5b96SMark Fasheh struct ocfs2_orphan_filldir_priv *p = priv; 19895eae5b96SMark Fasheh struct inode *iter; 19905eae5b96SMark Fasheh 19915eae5b96SMark Fasheh if (name_len == 1 && !strncmp(".", name, 1)) 19925eae5b96SMark Fasheh return 0; 19935eae5b96SMark Fasheh if (name_len == 2 && !strncmp("..", name, 2)) 19945eae5b96SMark Fasheh return 0; 19955eae5b96SMark Fasheh 19965eae5b96SMark Fasheh /* Skip bad inodes so that recovery can continue */ 19975eae5b96SMark Fasheh iter = ocfs2_iget(p->osb, ino, 19985fa0613eSJan Kara OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 19995eae5b96SMark Fasheh if (IS_ERR(iter)) 20005eae5b96SMark Fasheh return 0; 20015eae5b96SMark Fasheh 2002b4107950STao Ma trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 20035eae5b96SMark Fasheh /* No locking is required for the next_orphan queue as there 20045eae5b96SMark Fasheh * is only ever a single process doing orphan recovery. */ 20055eae5b96SMark Fasheh OCFS2_I(iter)->ip_next_orphan = p->head; 20065eae5b96SMark Fasheh p->head = iter; 20075eae5b96SMark Fasheh 20085eae5b96SMark Fasheh return 0; 20095eae5b96SMark Fasheh } 20105eae5b96SMark Fasheh 2011b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2012b4df6ed8SMark Fasheh int slot, 2013b4df6ed8SMark Fasheh struct inode **head) 2014ccd979bdSMark Fasheh { 2015b4df6ed8SMark Fasheh int status; 2016ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 20173704412bSAl Viro struct ocfs2_orphan_filldir_priv priv = { 20183704412bSAl Viro .ctx.actor = ocfs2_orphan_filldir, 20193704412bSAl Viro .osb = osb, 20203704412bSAl Viro .head = *head 20213704412bSAl Viro }; 2022ccd979bdSMark Fasheh 2023ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2024ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 2025ccd979bdSMark Fasheh slot); 2026ccd979bdSMark Fasheh if (!orphan_dir_inode) { 2027ccd979bdSMark Fasheh status = -ENOENT; 2028ccd979bdSMark Fasheh mlog_errno(status); 2029b4df6ed8SMark Fasheh return status; 2030ccd979bdSMark Fasheh } 2031ccd979bdSMark Fasheh 20321b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 2033e63aecb6SMark Fasheh status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2034ccd979bdSMark Fasheh if (status < 0) { 2035ccd979bdSMark Fasheh mlog_errno(status); 2036ccd979bdSMark Fasheh goto out; 2037ccd979bdSMark Fasheh } 2038ccd979bdSMark Fasheh 20393704412bSAl Viro status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 20405eae5b96SMark Fasheh if (status) { 2041ccd979bdSMark Fasheh mlog_errno(status); 2042a86370fbSMark Fasheh goto out_cluster; 2043ccd979bdSMark Fasheh } 2044ccd979bdSMark Fasheh 20455eae5b96SMark Fasheh *head = priv.head; 2046ccd979bdSMark Fasheh 2047a86370fbSMark Fasheh out_cluster: 2048e63aecb6SMark Fasheh ocfs2_inode_unlock(orphan_dir_inode, 0); 2049b4df6ed8SMark Fasheh out: 2050b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 2051ccd979bdSMark Fasheh iput(orphan_dir_inode); 2052b4df6ed8SMark Fasheh return status; 2053b4df6ed8SMark Fasheh } 2054b4df6ed8SMark Fasheh 2055b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2056b4df6ed8SMark Fasheh int slot) 2057b4df6ed8SMark Fasheh { 2058b4df6ed8SMark Fasheh int ret; 2059b4df6ed8SMark Fasheh 2060b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2061b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 2062b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2063b4df6ed8SMark Fasheh return ret; 2064b4df6ed8SMark Fasheh } 2065b4df6ed8SMark Fasheh 2066b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2067b4df6ed8SMark Fasheh int slot) 2068b4df6ed8SMark Fasheh { 2069b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2070b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 2071b4df6ed8SMark Fasheh * know to quit early. */ 2072b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2073b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 2074b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 2075b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 2076b4df6ed8SMark Fasheh * them. */ 2077b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2078b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 2079b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 2080b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 2081b4df6ed8SMark Fasheh } 2082b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 2083b4df6ed8SMark Fasheh } 2084b4df6ed8SMark Fasheh 2085b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2086b4df6ed8SMark Fasheh int slot) 2087b4df6ed8SMark Fasheh { 2088b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2089b4df6ed8SMark Fasheh } 2090b4df6ed8SMark Fasheh 2091b4df6ed8SMark Fasheh /* 2092b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 2093b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 2094b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 2095b4df6ed8SMark Fasheh * (most) of the rest of the work. 2096b4df6ed8SMark Fasheh * 2097b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 2098b4df6ed8SMark Fasheh * couple of extra considerations. 2099b4df6ed8SMark Fasheh * 2100b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 2101b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 2102b4df6ed8SMark Fasheh * an invalid inode. 2103b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 2104b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 2105b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 2106b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 2107b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 2108b4df6ed8SMark Fasheh */ 2109b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2110b4df6ed8SMark Fasheh int slot) 2111b4df6ed8SMark Fasheh { 2112b4df6ed8SMark Fasheh int ret = 0; 2113b4df6ed8SMark Fasheh struct inode *inode = NULL; 2114b4df6ed8SMark Fasheh struct inode *iter; 2115b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 2116b4df6ed8SMark Fasheh 2117b4107950STao Ma trace_ocfs2_recover_orphans(slot); 2118b4df6ed8SMark Fasheh 2119b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 2120b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 2121b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 2122b4df6ed8SMark Fasheh 2123b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 2124b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 2125b4df6ed8SMark Fasheh if (ret) 2126b4df6ed8SMark Fasheh mlog_errno(ret); 2127ccd979bdSMark Fasheh 2128ccd979bdSMark Fasheh while (inode) { 2129ccd979bdSMark Fasheh oi = OCFS2_I(inode); 2130b4107950STao Ma trace_ocfs2_recover_orphans_iput( 2131b4107950STao Ma (unsigned long long)oi->ip_blkno); 2132ccd979bdSMark Fasheh 2133ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 2134ccd979bdSMark Fasheh 2135ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 2136ccd979bdSMark Fasheh /* Set the proper information to get us going into 2137ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 2138ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 2139ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 2140ccd979bdSMark Fasheh 2141ccd979bdSMark Fasheh iput(inode); 2142ccd979bdSMark Fasheh 2143ccd979bdSMark Fasheh inode = iter; 2144ccd979bdSMark Fasheh } 2145ccd979bdSMark Fasheh 2146b4df6ed8SMark Fasheh return ret; 2147ccd979bdSMark Fasheh } 2148ccd979bdSMark Fasheh 214919ece546SJan Kara static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2150ccd979bdSMark Fasheh { 2151ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 2152ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 2153ccd979bdSMark Fasheh * or DISABLED. */ 2154ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 215519ece546SJan Kara (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 215619ece546SJan Kara atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2157ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2158ccd979bdSMark Fasheh 2159ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 2160ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 2161ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 2162ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2163b4107950STao Ma trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2164ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 2165ccd979bdSMark Fasheh return -EBUSY; 2166ccd979bdSMark Fasheh } 2167ccd979bdSMark Fasheh 2168ccd979bdSMark Fasheh return 0; 2169ccd979bdSMark Fasheh } 2170ccd979bdSMark Fasheh 2171ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 2172ccd979bdSMark Fasheh { 2173ccd979bdSMark Fasheh int status; 2174ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 2175ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 2176ccd979bdSMark Fasheh 2177ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 2178ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 2179ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 2180ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 2181ccd979bdSMark Fasheh while (!(kthread_should_stop() && 2182ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 2183ccd979bdSMark Fasheh 2184745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 2185ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 2186745ae8baSMark Fasheh || kthread_should_stop()); 2187ccd979bdSMark Fasheh 2188ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 218955b465b6SJoseph Qi if (status < 0) { 219055b465b6SJoseph Qi static unsigned long abort_warn_time; 219155b465b6SJoseph Qi 219255b465b6SJoseph Qi /* Warn about this once per minute */ 219355b465b6SJoseph Qi if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 219455b465b6SJoseph Qi mlog(ML_ERROR, "status = %d, journal is " 219555b465b6SJoseph Qi "already aborted.\n", status); 219655b465b6SJoseph Qi /* 219755b465b6SJoseph Qi * After ocfs2_commit_cache() fails, j_num_trans has a 219855b465b6SJoseph Qi * non-zero value. Sleep here to avoid a busy-wait 219955b465b6SJoseph Qi * loop. 220055b465b6SJoseph Qi */ 220155b465b6SJoseph Qi msleep_interruptible(1000); 220255b465b6SJoseph Qi } 2203ccd979bdSMark Fasheh 2204ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2205ccd979bdSMark Fasheh mlog(ML_KTHREAD, 2206ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 2207ccd979bdSMark Fasheh "shutdown\n", 2208ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 2209ccd979bdSMark Fasheh } 2210ccd979bdSMark Fasheh } 2211ccd979bdSMark Fasheh 2212ccd979bdSMark Fasheh return 0; 2213ccd979bdSMark Fasheh } 2214ccd979bdSMark Fasheh 2215539d8264SSunil Mushran /* Reads all the journal inodes without taking any cluster locks. Used 2216539d8264SSunil Mushran * for hard readonly access to determine whether any journal requires 2217539d8264SSunil Mushran * recovery. Also used to refresh the recovery generation numbers after 2218539d8264SSunil Mushran * a journal has been recovered by another node. 2219539d8264SSunil Mushran */ 2220ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2221ccd979bdSMark Fasheh { 2222ccd979bdSMark Fasheh int ret = 0; 2223ccd979bdSMark Fasheh unsigned int slot; 2224539d8264SSunil Mushran struct buffer_head *di_bh = NULL; 2225ccd979bdSMark Fasheh struct ocfs2_dinode *di; 2226539d8264SSunil Mushran int journal_dirty = 0; 2227ccd979bdSMark Fasheh 2228ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 2229539d8264SSunil Mushran ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2230539d8264SSunil Mushran if (ret) { 2231ccd979bdSMark Fasheh mlog_errno(ret); 2232ccd979bdSMark Fasheh goto out; 2233ccd979bdSMark Fasheh } 2234ccd979bdSMark Fasheh 2235ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 2236ccd979bdSMark Fasheh 2237539d8264SSunil Mushran osb->slot_recovery_generations[slot] = 2238539d8264SSunil Mushran ocfs2_get_recovery_generation(di); 2239539d8264SSunil Mushran 2240ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 2241ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 2242539d8264SSunil Mushran journal_dirty = 1; 2243ccd979bdSMark Fasheh 2244ccd979bdSMark Fasheh brelse(di_bh); 2245539d8264SSunil Mushran di_bh = NULL; 2246ccd979bdSMark Fasheh } 2247ccd979bdSMark Fasheh 2248ccd979bdSMark Fasheh out: 2249539d8264SSunil Mushran if (journal_dirty) 2250539d8264SSunil Mushran ret = -EROFS; 2251ccd979bdSMark Fasheh return ret; 2252ccd979bdSMark Fasheh } 2253