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> 31ccd979bdSMark Fasheh 32ccd979bdSMark Fasheh #define MLOG_MASK_PREFIX ML_JOURNAL 33ccd979bdSMark Fasheh #include <cluster/masklog.h> 34ccd979bdSMark Fasheh 35ccd979bdSMark Fasheh #include "ocfs2.h" 36ccd979bdSMark Fasheh 37ccd979bdSMark Fasheh #include "alloc.h" 38ccd979bdSMark Fasheh #include "dlmglue.h" 39ccd979bdSMark Fasheh #include "extent_map.h" 40ccd979bdSMark Fasheh #include "heartbeat.h" 41ccd979bdSMark Fasheh #include "inode.h" 42ccd979bdSMark Fasheh #include "journal.h" 43ccd979bdSMark Fasheh #include "localalloc.h" 44ccd979bdSMark Fasheh #include "namei.h" 45ccd979bdSMark Fasheh #include "slot_map.h" 46ccd979bdSMark Fasheh #include "super.h" 47ccd979bdSMark Fasheh #include "vote.h" 48ccd979bdSMark Fasheh #include "sysfile.h" 49ccd979bdSMark Fasheh 50ccd979bdSMark Fasheh #include "buffer_head_io.h" 51ccd979bdSMark Fasheh 5234af946aSIngo Molnar DEFINE_SPINLOCK(trans_inc_lock); 53ccd979bdSMark Fasheh 54ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode); 55ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 56ccd979bdSMark Fasheh int node_num); 57ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg); 58ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb); 59ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb); 60ccd979bdSMark Fasheh static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, 61ccd979bdSMark Fasheh struct ocfs2_journal_handle *handle); 62ccd979bdSMark Fasheh static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle); 63ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 64ccd979bdSMark Fasheh int dirty); 65ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 66ccd979bdSMark Fasheh int slot_num); 67ccd979bdSMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 68ccd979bdSMark Fasheh int slot); 69ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg); 70ccd979bdSMark Fasheh 71ccd979bdSMark Fasheh static int ocfs2_commit_cache(struct ocfs2_super *osb) 72ccd979bdSMark Fasheh { 73ccd979bdSMark Fasheh int status = 0; 74ccd979bdSMark Fasheh unsigned int flushed; 75ccd979bdSMark Fasheh unsigned long old_id; 76ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 77ccd979bdSMark Fasheh 78ccd979bdSMark Fasheh mlog_entry_void(); 79ccd979bdSMark Fasheh 80ccd979bdSMark Fasheh journal = osb->journal; 81ccd979bdSMark Fasheh 82ccd979bdSMark Fasheh /* Flush all pending commits and checkpoint the journal. */ 83ccd979bdSMark Fasheh down_write(&journal->j_trans_barrier); 84ccd979bdSMark Fasheh 85ccd979bdSMark Fasheh if (atomic_read(&journal->j_num_trans) == 0) { 86ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 87ccd979bdSMark Fasheh mlog(0, "No transactions for me to flush!\n"); 88ccd979bdSMark Fasheh goto finally; 89ccd979bdSMark Fasheh } 90ccd979bdSMark Fasheh 91ccd979bdSMark Fasheh journal_lock_updates(journal->j_journal); 92ccd979bdSMark Fasheh status = journal_flush(journal->j_journal); 93ccd979bdSMark Fasheh journal_unlock_updates(journal->j_journal); 94ccd979bdSMark Fasheh if (status < 0) { 95ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 96ccd979bdSMark Fasheh mlog_errno(status); 97ccd979bdSMark Fasheh goto finally; 98ccd979bdSMark Fasheh } 99ccd979bdSMark Fasheh 100ccd979bdSMark Fasheh old_id = ocfs2_inc_trans_id(journal); 101ccd979bdSMark Fasheh 102ccd979bdSMark Fasheh flushed = atomic_read(&journal->j_num_trans); 103ccd979bdSMark Fasheh atomic_set(&journal->j_num_trans, 0); 104ccd979bdSMark Fasheh up_write(&journal->j_trans_barrier); 105ccd979bdSMark Fasheh 106ccd979bdSMark Fasheh mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 107ccd979bdSMark Fasheh journal->j_trans_id, flushed); 108ccd979bdSMark Fasheh 109ccd979bdSMark Fasheh ocfs2_kick_vote_thread(osb); 110ccd979bdSMark Fasheh wake_up(&journal->j_checkpointed); 111ccd979bdSMark Fasheh finally: 112ccd979bdSMark Fasheh mlog_exit(status); 113ccd979bdSMark Fasheh return status; 114ccd979bdSMark Fasheh } 115ccd979bdSMark Fasheh 116ccd979bdSMark Fasheh struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb) 117ccd979bdSMark Fasheh { 118ccd979bdSMark Fasheh struct ocfs2_journal_handle *retval = NULL; 119ccd979bdSMark Fasheh 120afae00abSSunil Mushran retval = kcalloc(1, sizeof(*retval), GFP_NOFS); 121ccd979bdSMark Fasheh if (!retval) { 122ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to allocate memory for journal " 123ccd979bdSMark Fasheh "handle!\n"); 124ccd979bdSMark Fasheh return NULL; 125ccd979bdSMark Fasheh } 126ccd979bdSMark Fasheh 127ccd979bdSMark Fasheh retval->num_locks = 0; 128ccd979bdSMark Fasheh retval->k_handle = NULL; 129ccd979bdSMark Fasheh 130ccd979bdSMark Fasheh INIT_LIST_HEAD(&retval->locks); 131ccd979bdSMark Fasheh INIT_LIST_HEAD(&retval->inode_list); 132ccd979bdSMark Fasheh retval->journal = osb->journal; 133ccd979bdSMark Fasheh 134ccd979bdSMark Fasheh return retval; 135ccd979bdSMark Fasheh } 136ccd979bdSMark Fasheh 137ccd979bdSMark Fasheh /* pass it NULL and it will allocate a new handle object for you. If 138ccd979bdSMark Fasheh * you pass it a handle however, it may still return error, in which 139ccd979bdSMark Fasheh * case it has free'd the passed handle for you. */ 140ccd979bdSMark Fasheh struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb, 141ccd979bdSMark Fasheh struct ocfs2_journal_handle *handle, 142ccd979bdSMark Fasheh int max_buffs) 143ccd979bdSMark Fasheh { 144ccd979bdSMark Fasheh int ret; 145ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 146ccd979bdSMark Fasheh 147ccd979bdSMark Fasheh mlog_entry("(max_buffs = %d)\n", max_buffs); 148ccd979bdSMark Fasheh 149ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb || !osb->journal->j_journal); 150ccd979bdSMark Fasheh 151ccd979bdSMark Fasheh if (ocfs2_is_hard_readonly(osb)) { 152ccd979bdSMark Fasheh ret = -EROFS; 153ccd979bdSMark Fasheh goto done_free; 154ccd979bdSMark Fasheh } 155ccd979bdSMark Fasheh 156ccd979bdSMark Fasheh BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 157ccd979bdSMark Fasheh BUG_ON(max_buffs <= 0); 158ccd979bdSMark Fasheh 159ccd979bdSMark Fasheh /* JBD might support this, but our journalling code doesn't yet. */ 160ccd979bdSMark Fasheh if (journal_current_handle()) { 161ccd979bdSMark Fasheh mlog(ML_ERROR, "Recursive transaction attempted!\n"); 162ccd979bdSMark Fasheh BUG(); 163ccd979bdSMark Fasheh } 164ccd979bdSMark Fasheh 165ccd979bdSMark Fasheh if (!handle) 166ccd979bdSMark Fasheh handle = ocfs2_alloc_handle(osb); 167ccd979bdSMark Fasheh if (!handle) { 168ccd979bdSMark Fasheh ret = -ENOMEM; 169ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to allocate memory for journal " 170ccd979bdSMark Fasheh "handle!\n"); 171ccd979bdSMark Fasheh goto done_free; 172ccd979bdSMark Fasheh } 173ccd979bdSMark Fasheh 174ccd979bdSMark Fasheh down_read(&osb->journal->j_trans_barrier); 175ccd979bdSMark Fasheh 176ccd979bdSMark Fasheh /* actually start the transaction now */ 177ccd979bdSMark Fasheh handle->k_handle = journal_start(journal, max_buffs); 178ccd979bdSMark Fasheh if (IS_ERR(handle->k_handle)) { 179ccd979bdSMark Fasheh up_read(&osb->journal->j_trans_barrier); 180ccd979bdSMark Fasheh 181ccd979bdSMark Fasheh ret = PTR_ERR(handle->k_handle); 182ccd979bdSMark Fasheh handle->k_handle = NULL; 183ccd979bdSMark Fasheh mlog_errno(ret); 184ccd979bdSMark Fasheh 185ccd979bdSMark Fasheh if (is_journal_aborted(journal)) { 186ccd979bdSMark Fasheh ocfs2_abort(osb->sb, "Detected aborted journal"); 187ccd979bdSMark Fasheh ret = -EROFS; 188ccd979bdSMark Fasheh } 189ccd979bdSMark Fasheh goto done_free; 190ccd979bdSMark Fasheh } 191ccd979bdSMark Fasheh 192ccd979bdSMark Fasheh atomic_inc(&(osb->journal->j_num_trans)); 193ccd979bdSMark Fasheh 194ccd979bdSMark Fasheh mlog_exit_ptr(handle); 195ccd979bdSMark Fasheh return handle; 196ccd979bdSMark Fasheh 197ccd979bdSMark Fasheh done_free: 198ccd979bdSMark Fasheh if (handle) 199ccd979bdSMark Fasheh ocfs2_commit_unstarted_handle(handle); /* will kfree handle */ 200ccd979bdSMark Fasheh 201ccd979bdSMark Fasheh mlog_exit(ret); 202ccd979bdSMark Fasheh return ERR_PTR(ret); 203ccd979bdSMark Fasheh } 204ccd979bdSMark Fasheh 205ccd979bdSMark Fasheh void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle, 206ccd979bdSMark Fasheh struct inode *inode) 207ccd979bdSMark Fasheh { 208ccd979bdSMark Fasheh BUG_ON(!handle); 209ccd979bdSMark Fasheh BUG_ON(!inode); 210ccd979bdSMark Fasheh 211ccd979bdSMark Fasheh atomic_inc(&inode->i_count); 212ccd979bdSMark Fasheh 213ccd979bdSMark Fasheh /* we're obviously changing it... */ 2141b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 215ccd979bdSMark Fasheh 216ccd979bdSMark Fasheh /* sanity check */ 217ccd979bdSMark Fasheh BUG_ON(OCFS2_I(inode)->ip_handle); 218ccd979bdSMark Fasheh BUG_ON(!list_empty(&OCFS2_I(inode)->ip_handle_list)); 219ccd979bdSMark Fasheh 220ccd979bdSMark Fasheh OCFS2_I(inode)->ip_handle = handle; 221f116629dSAkinobu Mita list_move_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list)); 222ccd979bdSMark Fasheh } 223ccd979bdSMark Fasheh 224ccd979bdSMark Fasheh static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle) 225ccd979bdSMark Fasheh { 226ccd979bdSMark Fasheh struct list_head *p, *n; 227ccd979bdSMark Fasheh struct inode *inode; 228ccd979bdSMark Fasheh struct ocfs2_inode_info *oi; 229ccd979bdSMark Fasheh 230ccd979bdSMark Fasheh list_for_each_safe(p, n, &handle->inode_list) { 231ccd979bdSMark Fasheh oi = list_entry(p, struct ocfs2_inode_info, 232ccd979bdSMark Fasheh ip_handle_list); 233ccd979bdSMark Fasheh inode = &oi->vfs_inode; 234ccd979bdSMark Fasheh 235ccd979bdSMark Fasheh OCFS2_I(inode)->ip_handle = NULL; 236ccd979bdSMark Fasheh list_del_init(&OCFS2_I(inode)->ip_handle_list); 237ccd979bdSMark Fasheh 2381b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 239ccd979bdSMark Fasheh iput(inode); 240ccd979bdSMark Fasheh } 241ccd979bdSMark Fasheh } 242ccd979bdSMark Fasheh 243ccd979bdSMark Fasheh /* This is trivial so we do it out of the main commit 244ccd979bdSMark Fasheh * paths. Beware, it can be called from start_trans too! */ 245ccd979bdSMark Fasheh static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle) 246ccd979bdSMark Fasheh { 247ccd979bdSMark Fasheh mlog_entry_void(); 248ccd979bdSMark Fasheh 249ccd979bdSMark Fasheh ocfs2_handle_unlock_inodes(handle); 250ccd979bdSMark Fasheh /* You are allowed to add journal locks before the transaction 251ccd979bdSMark Fasheh * has started. */ 252ccd979bdSMark Fasheh ocfs2_handle_cleanup_locks(handle->journal, handle); 253ccd979bdSMark Fasheh 254ccd979bdSMark Fasheh kfree(handle); 255ccd979bdSMark Fasheh 256ccd979bdSMark Fasheh mlog_exit_void(); 257ccd979bdSMark Fasheh } 258ccd979bdSMark Fasheh 259ccd979bdSMark Fasheh void ocfs2_commit_trans(struct ocfs2_journal_handle *handle) 260ccd979bdSMark Fasheh { 261ccd979bdSMark Fasheh handle_t *jbd_handle; 262ccd979bdSMark Fasheh int retval; 263ccd979bdSMark Fasheh struct ocfs2_journal *journal = handle->journal; 264ccd979bdSMark Fasheh 265ccd979bdSMark Fasheh mlog_entry_void(); 266ccd979bdSMark Fasheh 267ccd979bdSMark Fasheh BUG_ON(!handle); 268ccd979bdSMark Fasheh 269*c161f89bSMark Fasheh if (!handle->k_handle) { 270ccd979bdSMark Fasheh ocfs2_commit_unstarted_handle(handle); 271ccd979bdSMark Fasheh mlog_exit_void(); 272ccd979bdSMark Fasheh return; 273ccd979bdSMark Fasheh } 274ccd979bdSMark Fasheh 275ccd979bdSMark Fasheh /* release inode semaphores we took during this transaction */ 276ccd979bdSMark Fasheh ocfs2_handle_unlock_inodes(handle); 277ccd979bdSMark Fasheh 278ccd979bdSMark Fasheh /* ocfs2_extend_trans may have had to call journal_restart 279ccd979bdSMark Fasheh * which will always commit the transaction, but may return 280ccd979bdSMark Fasheh * error for any number of reasons. If this is the case, we 281ccd979bdSMark Fasheh * clear k_handle as it's not valid any more. */ 282ccd979bdSMark Fasheh if (handle->k_handle) { 283ccd979bdSMark Fasheh jbd_handle = handle->k_handle; 284ccd979bdSMark Fasheh 285ccd979bdSMark Fasheh /* actually stop the transaction. if we've set h_sync, 286ccd979bdSMark Fasheh * it'll have been committed when we return */ 287ccd979bdSMark Fasheh retval = journal_stop(jbd_handle); 288ccd979bdSMark Fasheh if (retval < 0) { 289ccd979bdSMark Fasheh mlog_errno(retval); 290ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not commit transaction\n"); 291ccd979bdSMark Fasheh BUG(); 292ccd979bdSMark Fasheh } 293ccd979bdSMark Fasheh 294ccd979bdSMark Fasheh handle->k_handle = NULL; /* it's been free'd in journal_stop */ 295ccd979bdSMark Fasheh } 296ccd979bdSMark Fasheh 297ccd979bdSMark Fasheh ocfs2_handle_cleanup_locks(journal, handle); 298ccd979bdSMark Fasheh 299ccd979bdSMark Fasheh up_read(&journal->j_trans_barrier); 300ccd979bdSMark Fasheh 301ccd979bdSMark Fasheh kfree(handle); 302ccd979bdSMark Fasheh mlog_exit_void(); 303ccd979bdSMark Fasheh } 304ccd979bdSMark Fasheh 305ccd979bdSMark Fasheh /* 306ccd979bdSMark Fasheh * 'nblocks' is what you want to add to the current 307ccd979bdSMark Fasheh * transaction. extend_trans will either extend the current handle by 308ccd979bdSMark Fasheh * nblocks, or commit it and start a new one with nblocks credits. 309ccd979bdSMark Fasheh * 310ccd979bdSMark Fasheh * WARNING: This will not release any semaphores or disk locks taken 311ccd979bdSMark Fasheh * during the transaction, so make sure they were taken *before* 312ccd979bdSMark Fasheh * start_trans or we'll have ordering deadlocks. 313ccd979bdSMark Fasheh * 314ccd979bdSMark Fasheh * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 315ccd979bdSMark Fasheh * good because transaction ids haven't yet been recorded on the 316ccd979bdSMark Fasheh * cluster locks associated with this handle. 317ccd979bdSMark Fasheh */ 3181fc58146SMark Fasheh int ocfs2_extend_trans(handle_t *handle, int nblocks) 319ccd979bdSMark Fasheh { 320ccd979bdSMark Fasheh int status; 321ccd979bdSMark Fasheh 322ccd979bdSMark Fasheh BUG_ON(!handle); 323ccd979bdSMark Fasheh BUG_ON(!nblocks); 324ccd979bdSMark Fasheh 325ccd979bdSMark Fasheh mlog_entry_void(); 326ccd979bdSMark Fasheh 327ccd979bdSMark Fasheh mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 328ccd979bdSMark Fasheh 3291fc58146SMark Fasheh status = journal_extend(handle, nblocks); 330ccd979bdSMark Fasheh if (status < 0) { 331ccd979bdSMark Fasheh mlog_errno(status); 332ccd979bdSMark Fasheh goto bail; 333ccd979bdSMark Fasheh } 334ccd979bdSMark Fasheh 335ccd979bdSMark Fasheh if (status > 0) { 336ccd979bdSMark Fasheh mlog(0, "journal_extend failed, trying journal_restart\n"); 3371fc58146SMark Fasheh status = journal_restart(handle, nblocks); 338ccd979bdSMark Fasheh if (status < 0) { 339ccd979bdSMark Fasheh mlog_errno(status); 340ccd979bdSMark Fasheh goto bail; 341ccd979bdSMark Fasheh } 34201ddf1e1SMark Fasheh } 343ccd979bdSMark Fasheh 344ccd979bdSMark Fasheh status = 0; 345ccd979bdSMark Fasheh bail: 346ccd979bdSMark Fasheh 347ccd979bdSMark Fasheh mlog_exit(status); 348ccd979bdSMark Fasheh return status; 349ccd979bdSMark Fasheh } 350ccd979bdSMark Fasheh 351ccd979bdSMark Fasheh int ocfs2_journal_access(struct ocfs2_journal_handle *handle, 352ccd979bdSMark Fasheh struct inode *inode, 353ccd979bdSMark Fasheh struct buffer_head *bh, 354ccd979bdSMark Fasheh int type) 355ccd979bdSMark Fasheh { 356ccd979bdSMark Fasheh int status; 357ccd979bdSMark Fasheh 358ccd979bdSMark Fasheh BUG_ON(!inode); 359ccd979bdSMark Fasheh BUG_ON(!handle); 360ccd979bdSMark Fasheh BUG_ON(!bh); 361ccd979bdSMark Fasheh 362205f87f6SBadari Pulavarty mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 363ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, type, 364ccd979bdSMark Fasheh (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 365ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_CREATE" : 366ccd979bdSMark Fasheh "OCFS2_JOURNAL_ACCESS_WRITE", 367ccd979bdSMark Fasheh bh->b_size); 368ccd979bdSMark Fasheh 369ccd979bdSMark Fasheh /* we can safely remove this assertion after testing. */ 370ccd979bdSMark Fasheh if (!buffer_uptodate(bh)) { 371ccd979bdSMark Fasheh mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 372ccd979bdSMark Fasheh mlog(ML_ERROR, "b_blocknr=%llu\n", 373ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 374ccd979bdSMark Fasheh BUG(); 375ccd979bdSMark Fasheh } 376ccd979bdSMark Fasheh 377ccd979bdSMark Fasheh /* Set the current transaction information on the inode so 378ccd979bdSMark Fasheh * that the locking code knows whether it can drop it's locks 379ccd979bdSMark Fasheh * on this inode or not. We're protected from the commit 380ccd979bdSMark Fasheh * thread updating the current transaction id until 381ccd979bdSMark Fasheh * ocfs2_commit_trans() because ocfs2_start_trans() took 382ccd979bdSMark Fasheh * j_trans_barrier for us. */ 383ccd979bdSMark Fasheh ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 384ccd979bdSMark Fasheh 385251b6eccSMark Fasheh mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 386ccd979bdSMark Fasheh switch (type) { 387ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_CREATE: 388ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_WRITE: 389ccd979bdSMark Fasheh status = journal_get_write_access(handle->k_handle, bh); 390ccd979bdSMark Fasheh break; 391ccd979bdSMark Fasheh 392ccd979bdSMark Fasheh case OCFS2_JOURNAL_ACCESS_UNDO: 393ccd979bdSMark Fasheh status = journal_get_undo_access(handle->k_handle, bh); 394ccd979bdSMark Fasheh break; 395ccd979bdSMark Fasheh 396ccd979bdSMark Fasheh default: 397ccd979bdSMark Fasheh status = -EINVAL; 398ccd979bdSMark Fasheh mlog(ML_ERROR, "Uknown access type!\n"); 399ccd979bdSMark Fasheh } 400251b6eccSMark Fasheh mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 401ccd979bdSMark Fasheh 402ccd979bdSMark Fasheh if (status < 0) 403ccd979bdSMark Fasheh mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 404ccd979bdSMark Fasheh status, type); 405ccd979bdSMark Fasheh 406ccd979bdSMark Fasheh mlog_exit(status); 407ccd979bdSMark Fasheh return status; 408ccd979bdSMark Fasheh } 409ccd979bdSMark Fasheh 410ccd979bdSMark Fasheh int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle, 411ccd979bdSMark Fasheh struct buffer_head *bh) 412ccd979bdSMark Fasheh { 413ccd979bdSMark Fasheh int status; 414ccd979bdSMark Fasheh 415ccd979bdSMark Fasheh mlog_entry("(bh->b_blocknr=%llu)\n", 416ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 417ccd979bdSMark Fasheh 418ccd979bdSMark Fasheh status = journal_dirty_metadata(handle->k_handle, bh); 419ccd979bdSMark Fasheh if (status < 0) 420ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not dirty metadata buffer. " 421ccd979bdSMark Fasheh "(bh->b_blocknr=%llu)\n", 422ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr); 423ccd979bdSMark Fasheh 424ccd979bdSMark Fasheh mlog_exit(status); 425ccd979bdSMark Fasheh return status; 426ccd979bdSMark Fasheh } 427ccd979bdSMark Fasheh 428ccd979bdSMark Fasheh int ocfs2_journal_dirty_data(handle_t *handle, 429ccd979bdSMark Fasheh struct buffer_head *bh) 430ccd979bdSMark Fasheh { 431ccd979bdSMark Fasheh int err = journal_dirty_data(handle, bh); 432ccd979bdSMark Fasheh if (err) 433ccd979bdSMark Fasheh mlog_errno(err); 434ccd979bdSMark Fasheh /* TODO: When we can handle it, abort the handle and go RO on 435ccd979bdSMark Fasheh * error here. */ 436ccd979bdSMark Fasheh 437ccd979bdSMark Fasheh return err; 438ccd979bdSMark Fasheh } 439ccd979bdSMark Fasheh 440ccd979bdSMark Fasheh /* We always assume you're adding a metadata lock at level 'ex' */ 441ccd979bdSMark Fasheh int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle, 442ccd979bdSMark Fasheh struct inode *inode) 443ccd979bdSMark Fasheh { 444ccd979bdSMark Fasheh int status; 445ccd979bdSMark Fasheh struct ocfs2_journal_lock *lock; 446ccd979bdSMark Fasheh 447ccd979bdSMark Fasheh BUG_ON(!inode); 448ccd979bdSMark Fasheh 449ccd979bdSMark Fasheh lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS); 450ccd979bdSMark Fasheh if (!lock) { 451ccd979bdSMark Fasheh status = -ENOMEM; 452ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 453ccd979bdSMark Fasheh goto bail; 454ccd979bdSMark Fasheh } 455ccd979bdSMark Fasheh 456ccd979bdSMark Fasheh if (!igrab(inode)) 457ccd979bdSMark Fasheh BUG(); 458ccd979bdSMark Fasheh lock->jl_inode = inode; 459ccd979bdSMark Fasheh 460ccd979bdSMark Fasheh list_add_tail(&(lock->jl_lock_list), &(handle->locks)); 461ccd979bdSMark Fasheh handle->num_locks++; 462ccd979bdSMark Fasheh 463ccd979bdSMark Fasheh status = 0; 464ccd979bdSMark Fasheh bail: 465ccd979bdSMark Fasheh mlog_exit(status); 466ccd979bdSMark Fasheh return status; 467ccd979bdSMark Fasheh } 468ccd979bdSMark Fasheh 469ccd979bdSMark Fasheh static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, 470ccd979bdSMark Fasheh struct ocfs2_journal_handle *handle) 471ccd979bdSMark Fasheh { 472ccd979bdSMark Fasheh struct list_head *p, *n; 473ccd979bdSMark Fasheh struct ocfs2_journal_lock *lock; 474ccd979bdSMark Fasheh struct inode *inode; 475ccd979bdSMark Fasheh 476ccd979bdSMark Fasheh list_for_each_safe(p, n, &(handle->locks)) { 477ccd979bdSMark Fasheh lock = list_entry(p, struct ocfs2_journal_lock, 478ccd979bdSMark Fasheh jl_lock_list); 479ccd979bdSMark Fasheh list_del(&lock->jl_lock_list); 480ccd979bdSMark Fasheh handle->num_locks--; 481ccd979bdSMark Fasheh 482ccd979bdSMark Fasheh inode = lock->jl_inode; 483ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 484ccd979bdSMark Fasheh if (atomic_read(&inode->i_count) == 1) 485ccd979bdSMark Fasheh mlog(ML_ERROR, 486b0697053SMark Fasheh "Inode %llu, I'm doing a last iput for!", 487b0697053SMark Fasheh (unsigned long long)OCFS2_I(inode)->ip_blkno); 488ccd979bdSMark Fasheh iput(inode); 489ccd979bdSMark Fasheh kmem_cache_free(ocfs2_lock_cache, lock); 490ccd979bdSMark Fasheh } 491ccd979bdSMark Fasheh } 492ccd979bdSMark Fasheh 493ccd979bdSMark Fasheh #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) 494ccd979bdSMark Fasheh 495ccd979bdSMark Fasheh void ocfs2_set_journal_params(struct ocfs2_super *osb) 496ccd979bdSMark Fasheh { 497ccd979bdSMark Fasheh journal_t *journal = osb->journal->j_journal; 498ccd979bdSMark Fasheh 499ccd979bdSMark Fasheh spin_lock(&journal->j_state_lock); 500ccd979bdSMark Fasheh journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 501ccd979bdSMark Fasheh if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 502ccd979bdSMark Fasheh journal->j_flags |= JFS_BARRIER; 503ccd979bdSMark Fasheh else 504ccd979bdSMark Fasheh journal->j_flags &= ~JFS_BARRIER; 505ccd979bdSMark Fasheh spin_unlock(&journal->j_state_lock); 506ccd979bdSMark Fasheh } 507ccd979bdSMark Fasheh 508ccd979bdSMark Fasheh int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 509ccd979bdSMark Fasheh { 510ccd979bdSMark Fasheh int status = -1; 511ccd979bdSMark Fasheh struct inode *inode = NULL; /* the journal inode */ 512ccd979bdSMark Fasheh journal_t *j_journal = NULL; 513ccd979bdSMark Fasheh struct ocfs2_dinode *di = NULL; 514ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 515ccd979bdSMark Fasheh struct ocfs2_super *osb; 516ccd979bdSMark Fasheh int meta_lock = 0; 517ccd979bdSMark Fasheh 518ccd979bdSMark Fasheh mlog_entry_void(); 519ccd979bdSMark Fasheh 520ccd979bdSMark Fasheh BUG_ON(!journal); 521ccd979bdSMark Fasheh 522ccd979bdSMark Fasheh osb = journal->j_osb; 523ccd979bdSMark Fasheh 524ccd979bdSMark Fasheh /* already have the inode for our journal */ 525ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 526ccd979bdSMark Fasheh osb->slot_num); 527ccd979bdSMark Fasheh if (inode == NULL) { 528ccd979bdSMark Fasheh status = -EACCES; 529ccd979bdSMark Fasheh mlog_errno(status); 530ccd979bdSMark Fasheh goto done; 531ccd979bdSMark Fasheh } 532ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 533ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 534ccd979bdSMark Fasheh iput(inode); 535ccd979bdSMark Fasheh inode = NULL; 536ccd979bdSMark Fasheh status = -EACCES; 537ccd979bdSMark Fasheh goto done; 538ccd979bdSMark Fasheh } 539ccd979bdSMark Fasheh 540ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 541ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count++; 542ccd979bdSMark Fasheh 5436eff5790SMark Fasheh /* Skip recovery waits here - journal inode metadata never 5446eff5790SMark Fasheh * changes in a live cluster so it can be considered an 5456eff5790SMark Fasheh * exception to the rule. */ 5466eff5790SMark Fasheh status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, 5476eff5790SMark Fasheh OCFS2_META_LOCK_RECOVERY); 548ccd979bdSMark Fasheh if (status < 0) { 549ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 550ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not get lock on journal!\n"); 551ccd979bdSMark Fasheh goto done; 552ccd979bdSMark Fasheh } 553ccd979bdSMark Fasheh 554ccd979bdSMark Fasheh meta_lock = 1; 555ccd979bdSMark Fasheh di = (struct ocfs2_dinode *)bh->b_data; 556ccd979bdSMark Fasheh 557ccd979bdSMark Fasheh if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 558ccd979bdSMark Fasheh mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 559ccd979bdSMark Fasheh inode->i_size); 560ccd979bdSMark Fasheh status = -EINVAL; 561ccd979bdSMark Fasheh goto done; 562ccd979bdSMark Fasheh } 563ccd979bdSMark Fasheh 564ccd979bdSMark Fasheh mlog(0, "inode->i_size = %lld\n", inode->i_size); 5655515eff8SAndrew Morton mlog(0, "inode->i_blocks = %llu\n", 5665515eff8SAndrew Morton (unsigned long long)inode->i_blocks); 567ccd979bdSMark Fasheh mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 568ccd979bdSMark Fasheh 569ccd979bdSMark Fasheh /* call the kernels journal init function now */ 570ccd979bdSMark Fasheh j_journal = journal_init_inode(inode); 571ccd979bdSMark Fasheh if (j_journal == NULL) { 572ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 573ccd979bdSMark Fasheh status = -EINVAL; 574ccd979bdSMark Fasheh goto done; 575ccd979bdSMark Fasheh } 576ccd979bdSMark Fasheh 577ccd979bdSMark Fasheh mlog(0, "Returned from journal_init_inode\n"); 578ccd979bdSMark Fasheh mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 579ccd979bdSMark Fasheh 580ccd979bdSMark Fasheh *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 581ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL); 582ccd979bdSMark Fasheh 583ccd979bdSMark Fasheh journal->j_journal = j_journal; 584ccd979bdSMark Fasheh journal->j_inode = inode; 585ccd979bdSMark Fasheh journal->j_bh = bh; 586ccd979bdSMark Fasheh 587ccd979bdSMark Fasheh ocfs2_set_journal_params(osb); 588ccd979bdSMark Fasheh 589ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_LOADED; 590ccd979bdSMark Fasheh 591ccd979bdSMark Fasheh status = 0; 592ccd979bdSMark Fasheh done: 593ccd979bdSMark Fasheh if (status < 0) { 594ccd979bdSMark Fasheh if (meta_lock) 595ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 596ccd979bdSMark Fasheh if (bh != NULL) 597ccd979bdSMark Fasheh brelse(bh); 598ccd979bdSMark Fasheh if (inode) { 599ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 600ccd979bdSMark Fasheh iput(inode); 601ccd979bdSMark Fasheh } 602ccd979bdSMark Fasheh } 603ccd979bdSMark Fasheh 604ccd979bdSMark Fasheh mlog_exit(status); 605ccd979bdSMark Fasheh return status; 606ccd979bdSMark Fasheh } 607ccd979bdSMark Fasheh 608ccd979bdSMark Fasheh static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 609ccd979bdSMark Fasheh int dirty) 610ccd979bdSMark Fasheh { 611ccd979bdSMark Fasheh int status; 612ccd979bdSMark Fasheh unsigned int flags; 613ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 614ccd979bdSMark Fasheh struct buffer_head *bh = journal->j_bh; 615ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 616ccd979bdSMark Fasheh 617ccd979bdSMark Fasheh mlog_entry_void(); 618ccd979bdSMark Fasheh 619ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *)bh->b_data; 620ccd979bdSMark Fasheh if (!OCFS2_IS_VALID_DINODE(fe)) { 621ccd979bdSMark Fasheh /* This is called from startup/shutdown which will 622ccd979bdSMark Fasheh * handle the errors in a specific manner, so no need 623ccd979bdSMark Fasheh * to call ocfs2_error() here. */ 624b0697053SMark Fasheh mlog(ML_ERROR, "Journal dinode %llu has invalid " 625b0697053SMark Fasheh "signature: %.*s", (unsigned long long)fe->i_blkno, 7, 626b0697053SMark Fasheh fe->i_signature); 627ccd979bdSMark Fasheh status = -EIO; 628ccd979bdSMark Fasheh goto out; 629ccd979bdSMark Fasheh } 630ccd979bdSMark Fasheh 631ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 632ccd979bdSMark Fasheh if (dirty) 633ccd979bdSMark Fasheh flags |= OCFS2_JOURNAL_DIRTY_FL; 634ccd979bdSMark Fasheh else 635ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 636ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 637ccd979bdSMark Fasheh 638ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, journal->j_inode); 639ccd979bdSMark Fasheh if (status < 0) 640ccd979bdSMark Fasheh mlog_errno(status); 641ccd979bdSMark Fasheh 642ccd979bdSMark Fasheh out: 643ccd979bdSMark Fasheh mlog_exit(status); 644ccd979bdSMark Fasheh return status; 645ccd979bdSMark Fasheh } 646ccd979bdSMark Fasheh 647ccd979bdSMark Fasheh /* 648ccd979bdSMark Fasheh * If the journal has been kmalloc'd it needs to be freed after this 649ccd979bdSMark Fasheh * call. 650ccd979bdSMark Fasheh */ 651ccd979bdSMark Fasheh void ocfs2_journal_shutdown(struct ocfs2_super *osb) 652ccd979bdSMark Fasheh { 653ccd979bdSMark Fasheh struct ocfs2_journal *journal = NULL; 654ccd979bdSMark Fasheh int status = 0; 655ccd979bdSMark Fasheh struct inode *inode = NULL; 656ccd979bdSMark Fasheh int num_running_trans = 0; 657ccd979bdSMark Fasheh 658ccd979bdSMark Fasheh mlog_entry_void(); 659ccd979bdSMark Fasheh 660ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!osb); 661ccd979bdSMark Fasheh 662ccd979bdSMark Fasheh journal = osb->journal; 663ccd979bdSMark Fasheh if (!journal) 664ccd979bdSMark Fasheh goto done; 665ccd979bdSMark Fasheh 666ccd979bdSMark Fasheh inode = journal->j_inode; 667ccd979bdSMark Fasheh 668ccd979bdSMark Fasheh if (journal->j_state != OCFS2_JOURNAL_LOADED) 669ccd979bdSMark Fasheh goto done; 670ccd979bdSMark Fasheh 671ccd979bdSMark Fasheh /* need to inc inode use count as journal_destroy will iput. */ 672ccd979bdSMark Fasheh if (!igrab(inode)) 673ccd979bdSMark Fasheh BUG(); 674ccd979bdSMark Fasheh 675ccd979bdSMark Fasheh num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 676ccd979bdSMark Fasheh if (num_running_trans > 0) 677ccd979bdSMark Fasheh mlog(0, "Shutting down journal: must wait on %d " 678ccd979bdSMark Fasheh "running transactions!\n", 679ccd979bdSMark Fasheh num_running_trans); 680ccd979bdSMark Fasheh 681ccd979bdSMark Fasheh /* Do a commit_cache here. It will flush our journal, *and* 682ccd979bdSMark Fasheh * release any locks that are still held. 683ccd979bdSMark Fasheh * set the SHUTDOWN flag and release the trans lock. 684ccd979bdSMark Fasheh * the commit thread will take the trans lock for us below. */ 685ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 686ccd979bdSMark Fasheh 687ccd979bdSMark Fasheh /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 688ccd979bdSMark Fasheh * drop the trans_lock (which we want to hold until we 689ccd979bdSMark Fasheh * completely destroy the journal. */ 690ccd979bdSMark Fasheh if (osb->commit_task) { 691ccd979bdSMark Fasheh /* Wait for the commit thread */ 692ccd979bdSMark Fasheh mlog(0, "Waiting for ocfs2commit to exit....\n"); 693ccd979bdSMark Fasheh kthread_stop(osb->commit_task); 694ccd979bdSMark Fasheh osb->commit_task = NULL; 695ccd979bdSMark Fasheh } 696ccd979bdSMark Fasheh 697ccd979bdSMark Fasheh BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 698ccd979bdSMark Fasheh 699ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 0); 700ccd979bdSMark Fasheh if (status < 0) 701ccd979bdSMark Fasheh mlog_errno(status); 702ccd979bdSMark Fasheh 703ccd979bdSMark Fasheh /* Shutdown the kernel journal system */ 704ccd979bdSMark Fasheh journal_destroy(journal->j_journal); 705ccd979bdSMark Fasheh 706ccd979bdSMark Fasheh OCFS2_I(inode)->ip_open_count--; 707ccd979bdSMark Fasheh 708ccd979bdSMark Fasheh /* unlock our journal */ 709ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 710ccd979bdSMark Fasheh 711ccd979bdSMark Fasheh brelse(journal->j_bh); 712ccd979bdSMark Fasheh journal->j_bh = NULL; 713ccd979bdSMark Fasheh 714ccd979bdSMark Fasheh journal->j_state = OCFS2_JOURNAL_FREE; 715ccd979bdSMark Fasheh 716ccd979bdSMark Fasheh // up_write(&journal->j_trans_barrier); 717ccd979bdSMark Fasheh done: 718ccd979bdSMark Fasheh if (inode) 719ccd979bdSMark Fasheh iput(inode); 720ccd979bdSMark Fasheh mlog_exit_void(); 721ccd979bdSMark Fasheh } 722ccd979bdSMark Fasheh 723ccd979bdSMark Fasheh static void ocfs2_clear_journal_error(struct super_block *sb, 724ccd979bdSMark Fasheh journal_t *journal, 725ccd979bdSMark Fasheh int slot) 726ccd979bdSMark Fasheh { 727ccd979bdSMark Fasheh int olderr; 728ccd979bdSMark Fasheh 729ccd979bdSMark Fasheh olderr = journal_errno(journal); 730ccd979bdSMark Fasheh if (olderr) { 731ccd979bdSMark Fasheh mlog(ML_ERROR, "File system error %d recorded in " 732ccd979bdSMark Fasheh "journal %u.\n", olderr, slot); 733ccd979bdSMark Fasheh mlog(ML_ERROR, "File system on device %s needs checking.\n", 734ccd979bdSMark Fasheh sb->s_id); 735ccd979bdSMark Fasheh 736ccd979bdSMark Fasheh journal_ack_err(journal); 737ccd979bdSMark Fasheh journal_clear_err(journal); 738ccd979bdSMark Fasheh } 739ccd979bdSMark Fasheh } 740ccd979bdSMark Fasheh 741ccd979bdSMark Fasheh int ocfs2_journal_load(struct ocfs2_journal *journal) 742ccd979bdSMark Fasheh { 743ccd979bdSMark Fasheh int status = 0; 744ccd979bdSMark Fasheh struct ocfs2_super *osb; 745ccd979bdSMark Fasheh 746ccd979bdSMark Fasheh mlog_entry_void(); 747ccd979bdSMark Fasheh 748ccd979bdSMark Fasheh if (!journal) 749ccd979bdSMark Fasheh BUG(); 750ccd979bdSMark Fasheh 751ccd979bdSMark Fasheh osb = journal->j_osb; 752ccd979bdSMark Fasheh 753ccd979bdSMark Fasheh status = journal_load(journal->j_journal); 754ccd979bdSMark Fasheh if (status < 0) { 755ccd979bdSMark Fasheh mlog(ML_ERROR, "Failed to load journal!\n"); 756ccd979bdSMark Fasheh goto done; 757ccd979bdSMark Fasheh } 758ccd979bdSMark Fasheh 759ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 760ccd979bdSMark Fasheh 761ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(osb, 1); 762ccd979bdSMark Fasheh if (status < 0) { 763ccd979bdSMark Fasheh mlog_errno(status); 764ccd979bdSMark Fasheh goto done; 765ccd979bdSMark Fasheh } 766ccd979bdSMark Fasheh 767ccd979bdSMark Fasheh /* Launch the commit thread */ 76878427043SMark Fasheh osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt"); 769ccd979bdSMark Fasheh if (IS_ERR(osb->commit_task)) { 770ccd979bdSMark Fasheh status = PTR_ERR(osb->commit_task); 771ccd979bdSMark Fasheh osb->commit_task = NULL; 772ccd979bdSMark Fasheh mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d", 773ccd979bdSMark Fasheh status); 774ccd979bdSMark Fasheh goto done; 775ccd979bdSMark Fasheh } 776ccd979bdSMark Fasheh 777ccd979bdSMark Fasheh done: 778ccd979bdSMark Fasheh mlog_exit(status); 779ccd979bdSMark Fasheh return status; 780ccd979bdSMark Fasheh } 781ccd979bdSMark Fasheh 782ccd979bdSMark Fasheh 783ccd979bdSMark Fasheh /* 'full' flag tells us whether we clear out all blocks or if we just 784ccd979bdSMark Fasheh * mark the journal clean */ 785ccd979bdSMark Fasheh int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 786ccd979bdSMark Fasheh { 787ccd979bdSMark Fasheh int status; 788ccd979bdSMark Fasheh 789ccd979bdSMark Fasheh mlog_entry_void(); 790ccd979bdSMark Fasheh 791ebdec83bSEric Sesterhenn / snakebyte BUG_ON(!journal); 792ccd979bdSMark Fasheh 793ccd979bdSMark Fasheh status = journal_wipe(journal->j_journal, full); 794ccd979bdSMark Fasheh if (status < 0) { 795ccd979bdSMark Fasheh mlog_errno(status); 796ccd979bdSMark Fasheh goto bail; 797ccd979bdSMark Fasheh } 798ccd979bdSMark Fasheh 799ccd979bdSMark Fasheh status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); 800ccd979bdSMark Fasheh if (status < 0) 801ccd979bdSMark Fasheh mlog_errno(status); 802ccd979bdSMark Fasheh 803ccd979bdSMark Fasheh bail: 804ccd979bdSMark Fasheh mlog_exit(status); 805ccd979bdSMark Fasheh return status; 806ccd979bdSMark Fasheh } 807ccd979bdSMark Fasheh 808ccd979bdSMark Fasheh /* 809ccd979bdSMark Fasheh * JBD Might read a cached version of another nodes journal file. We 810ccd979bdSMark Fasheh * don't want this as this file changes often and we get no 811ccd979bdSMark Fasheh * notification on those changes. The only way to be sure that we've 812ccd979bdSMark Fasheh * got the most up to date version of those blocks then is to force 813ccd979bdSMark Fasheh * read them off disk. Just searching through the buffer cache won't 814ccd979bdSMark Fasheh * work as there may be pages backing this file which are still marked 815ccd979bdSMark Fasheh * up to date. We know things can't change on this file underneath us 816ccd979bdSMark Fasheh * as we have the lock by now :) 817ccd979bdSMark Fasheh */ 818ccd979bdSMark Fasheh static int ocfs2_force_read_journal(struct inode *inode) 819ccd979bdSMark Fasheh { 820ccd979bdSMark Fasheh int status = 0; 821ccd979bdSMark Fasheh int i, p_blocks; 822ccd979bdSMark Fasheh u64 v_blkno, p_blkno; 823ccd979bdSMark Fasheh #define CONCURRENT_JOURNAL_FILL 32 824ccd979bdSMark Fasheh struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 825ccd979bdSMark Fasheh 826ccd979bdSMark Fasheh mlog_entry_void(); 827ccd979bdSMark Fasheh 828ccd979bdSMark Fasheh BUG_ON(inode->i_blocks != 829ccd979bdSMark Fasheh ocfs2_align_bytes_to_sectors(i_size_read(inode))); 830ccd979bdSMark Fasheh 831ccd979bdSMark Fasheh memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 832ccd979bdSMark Fasheh 8335515eff8SAndrew Morton mlog(0, "Force reading %llu blocks\n", 8345515eff8SAndrew Morton (unsigned long long)(inode->i_blocks >> 8355515eff8SAndrew Morton (inode->i_sb->s_blocksize_bits - 9))); 836ccd979bdSMark Fasheh 837ccd979bdSMark Fasheh v_blkno = 0; 838ccd979bdSMark Fasheh while (v_blkno < 839ccd979bdSMark Fasheh (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) { 840ccd979bdSMark Fasheh 841ccd979bdSMark Fasheh status = ocfs2_extent_map_get_blocks(inode, v_blkno, 842ccd979bdSMark Fasheh 1, &p_blkno, 843ccd979bdSMark Fasheh &p_blocks); 844ccd979bdSMark Fasheh if (status < 0) { 845ccd979bdSMark Fasheh mlog_errno(status); 846ccd979bdSMark Fasheh goto bail; 847ccd979bdSMark Fasheh } 848ccd979bdSMark Fasheh 849ccd979bdSMark Fasheh if (p_blocks > CONCURRENT_JOURNAL_FILL) 850ccd979bdSMark Fasheh p_blocks = CONCURRENT_JOURNAL_FILL; 851ccd979bdSMark Fasheh 852dd4a2c2bSMark Fasheh /* We are reading journal data which should not 853dd4a2c2bSMark Fasheh * be put in the uptodate cache */ 854ccd979bdSMark Fasheh status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 855ccd979bdSMark Fasheh p_blkno, p_blocks, bhs, 0, 856dd4a2c2bSMark Fasheh NULL); 857ccd979bdSMark Fasheh if (status < 0) { 858ccd979bdSMark Fasheh mlog_errno(status); 859ccd979bdSMark Fasheh goto bail; 860ccd979bdSMark Fasheh } 861ccd979bdSMark Fasheh 862ccd979bdSMark Fasheh for(i = 0; i < p_blocks; i++) { 863ccd979bdSMark Fasheh brelse(bhs[i]); 864ccd979bdSMark Fasheh bhs[i] = NULL; 865ccd979bdSMark Fasheh } 866ccd979bdSMark Fasheh 867ccd979bdSMark Fasheh v_blkno += p_blocks; 868ccd979bdSMark Fasheh } 869ccd979bdSMark Fasheh 870ccd979bdSMark Fasheh bail: 871ccd979bdSMark Fasheh for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 872ccd979bdSMark Fasheh if (bhs[i]) 873ccd979bdSMark Fasheh brelse(bhs[i]); 874ccd979bdSMark Fasheh mlog_exit(status); 875ccd979bdSMark Fasheh return status; 876ccd979bdSMark Fasheh } 877ccd979bdSMark Fasheh 878ccd979bdSMark Fasheh struct ocfs2_la_recovery_item { 879ccd979bdSMark Fasheh struct list_head lri_list; 880ccd979bdSMark Fasheh int lri_slot; 881ccd979bdSMark Fasheh struct ocfs2_dinode *lri_la_dinode; 882ccd979bdSMark Fasheh struct ocfs2_dinode *lri_tl_dinode; 883ccd979bdSMark Fasheh }; 884ccd979bdSMark Fasheh 885ccd979bdSMark Fasheh /* Does the second half of the recovery process. By this point, the 886ccd979bdSMark Fasheh * node is marked clean and can actually be considered recovered, 887ccd979bdSMark Fasheh * hence it's no longer in the recovery map, but there's still some 888ccd979bdSMark Fasheh * cleanup we can do which shouldn't happen within the recovery thread 889ccd979bdSMark Fasheh * as locking in that context becomes very difficult if we are to take 890ccd979bdSMark Fasheh * recovering nodes into account. 891ccd979bdSMark Fasheh * 892ccd979bdSMark Fasheh * NOTE: This function can and will sleep on recovery of other nodes 893ccd979bdSMark Fasheh * during cluster locking, just like any other ocfs2 process. 894ccd979bdSMark Fasheh */ 895ccd979bdSMark Fasheh void ocfs2_complete_recovery(void *data) 896ccd979bdSMark Fasheh { 897ccd979bdSMark Fasheh int ret; 898ccd979bdSMark Fasheh struct ocfs2_super *osb = data; 899ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 900ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, *tl_dinode; 901ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 902ccd979bdSMark Fasheh struct list_head *p, *n; 903ccd979bdSMark Fasheh LIST_HEAD(tmp_la_list); 904ccd979bdSMark Fasheh 905ccd979bdSMark Fasheh mlog_entry_void(); 906ccd979bdSMark Fasheh 907ccd979bdSMark Fasheh mlog(0, "completing recovery from keventd\n"); 908ccd979bdSMark Fasheh 909ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 910ccd979bdSMark Fasheh list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 911ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 912ccd979bdSMark Fasheh 913ccd979bdSMark Fasheh list_for_each_safe(p, n, &tmp_la_list) { 914ccd979bdSMark Fasheh item = list_entry(p, struct ocfs2_la_recovery_item, lri_list); 915ccd979bdSMark Fasheh list_del_init(&item->lri_list); 916ccd979bdSMark Fasheh 917ccd979bdSMark Fasheh mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 918ccd979bdSMark Fasheh 919ccd979bdSMark Fasheh la_dinode = item->lri_la_dinode; 920ccd979bdSMark Fasheh if (la_dinode) { 921b0697053SMark Fasheh mlog(0, "Clean up local alloc %llu\n", 922b0697053SMark Fasheh (unsigned long long)la_dinode->i_blkno); 923ccd979bdSMark Fasheh 924ccd979bdSMark Fasheh ret = ocfs2_complete_local_alloc_recovery(osb, 925ccd979bdSMark Fasheh la_dinode); 926ccd979bdSMark Fasheh if (ret < 0) 927ccd979bdSMark Fasheh mlog_errno(ret); 928ccd979bdSMark Fasheh 929ccd979bdSMark Fasheh kfree(la_dinode); 930ccd979bdSMark Fasheh } 931ccd979bdSMark Fasheh 932ccd979bdSMark Fasheh tl_dinode = item->lri_tl_dinode; 933ccd979bdSMark Fasheh if (tl_dinode) { 934b0697053SMark Fasheh mlog(0, "Clean up truncate log %llu\n", 935b0697053SMark Fasheh (unsigned long long)tl_dinode->i_blkno); 936ccd979bdSMark Fasheh 937ccd979bdSMark Fasheh ret = ocfs2_complete_truncate_log_recovery(osb, 938ccd979bdSMark Fasheh tl_dinode); 939ccd979bdSMark Fasheh if (ret < 0) 940ccd979bdSMark Fasheh mlog_errno(ret); 941ccd979bdSMark Fasheh 942ccd979bdSMark Fasheh kfree(tl_dinode); 943ccd979bdSMark Fasheh } 944ccd979bdSMark Fasheh 945ccd979bdSMark Fasheh ret = ocfs2_recover_orphans(osb, item->lri_slot); 946ccd979bdSMark Fasheh if (ret < 0) 947ccd979bdSMark Fasheh mlog_errno(ret); 948ccd979bdSMark Fasheh 949ccd979bdSMark Fasheh kfree(item); 950ccd979bdSMark Fasheh } 951ccd979bdSMark Fasheh 952ccd979bdSMark Fasheh mlog(0, "Recovery completion\n"); 953ccd979bdSMark Fasheh mlog_exit_void(); 954ccd979bdSMark Fasheh } 955ccd979bdSMark Fasheh 956ccd979bdSMark Fasheh /* NOTE: This function always eats your references to la_dinode and 957ccd979bdSMark Fasheh * tl_dinode, either manually on error, or by passing them to 958ccd979bdSMark Fasheh * ocfs2_complete_recovery */ 959ccd979bdSMark Fasheh static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 960ccd979bdSMark Fasheh int slot_num, 961ccd979bdSMark Fasheh struct ocfs2_dinode *la_dinode, 962ccd979bdSMark Fasheh struct ocfs2_dinode *tl_dinode) 963ccd979bdSMark Fasheh { 964ccd979bdSMark Fasheh struct ocfs2_la_recovery_item *item; 965ccd979bdSMark Fasheh 966afae00abSSunil Mushran item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 967ccd979bdSMark Fasheh if (!item) { 968ccd979bdSMark Fasheh /* Though we wish to avoid it, we are in fact safe in 969ccd979bdSMark Fasheh * skipping local alloc cleanup as fsck.ocfs2 is more 970ccd979bdSMark Fasheh * than capable of reclaiming unused space. */ 971ccd979bdSMark Fasheh if (la_dinode) 972ccd979bdSMark Fasheh kfree(la_dinode); 973ccd979bdSMark Fasheh 974ccd979bdSMark Fasheh if (tl_dinode) 975ccd979bdSMark Fasheh kfree(tl_dinode); 976ccd979bdSMark Fasheh 977ccd979bdSMark Fasheh mlog_errno(-ENOMEM); 978ccd979bdSMark Fasheh return; 979ccd979bdSMark Fasheh } 980ccd979bdSMark Fasheh 981ccd979bdSMark Fasheh INIT_LIST_HEAD(&item->lri_list); 982ccd979bdSMark Fasheh item->lri_la_dinode = la_dinode; 983ccd979bdSMark Fasheh item->lri_slot = slot_num; 984ccd979bdSMark Fasheh item->lri_tl_dinode = tl_dinode; 985ccd979bdSMark Fasheh 986ccd979bdSMark Fasheh spin_lock(&journal->j_lock); 987ccd979bdSMark Fasheh list_add_tail(&item->lri_list, &journal->j_la_cleanups); 988ccd979bdSMark Fasheh queue_work(ocfs2_wq, &journal->j_recovery_work); 989ccd979bdSMark Fasheh spin_unlock(&journal->j_lock); 990ccd979bdSMark Fasheh } 991ccd979bdSMark Fasheh 992ccd979bdSMark Fasheh /* Called by the mount code to queue recovery the last part of 993ccd979bdSMark Fasheh * recovery for it's own slot. */ 994ccd979bdSMark Fasheh void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 995ccd979bdSMark Fasheh { 996ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 997ccd979bdSMark Fasheh 998ccd979bdSMark Fasheh if (osb->dirty) { 999ccd979bdSMark Fasheh /* No need to queue up our truncate_log as regular 1000ccd979bdSMark Fasheh * cleanup will catch that. */ 1001ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(journal, 1002ccd979bdSMark Fasheh osb->slot_num, 1003ccd979bdSMark Fasheh osb->local_alloc_copy, 1004ccd979bdSMark Fasheh NULL); 1005ccd979bdSMark Fasheh ocfs2_schedule_truncate_log_flush(osb, 0); 1006ccd979bdSMark Fasheh 1007ccd979bdSMark Fasheh osb->local_alloc_copy = NULL; 1008ccd979bdSMark Fasheh osb->dirty = 0; 1009ccd979bdSMark Fasheh } 1010ccd979bdSMark Fasheh } 1011ccd979bdSMark Fasheh 1012ccd979bdSMark Fasheh static int __ocfs2_recovery_thread(void *arg) 1013ccd979bdSMark Fasheh { 1014ccd979bdSMark Fasheh int status, node_num; 1015ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1016ccd979bdSMark Fasheh 1017ccd979bdSMark Fasheh mlog_entry_void(); 1018ccd979bdSMark Fasheh 1019ccd979bdSMark Fasheh status = ocfs2_wait_on_mount(osb); 1020ccd979bdSMark Fasheh if (status < 0) { 1021ccd979bdSMark Fasheh goto bail; 1022ccd979bdSMark Fasheh } 1023ccd979bdSMark Fasheh 1024ccd979bdSMark Fasheh restart: 1025ccd979bdSMark Fasheh status = ocfs2_super_lock(osb, 1); 1026ccd979bdSMark Fasheh if (status < 0) { 1027ccd979bdSMark Fasheh mlog_errno(status); 1028ccd979bdSMark Fasheh goto bail; 1029ccd979bdSMark Fasheh } 1030ccd979bdSMark Fasheh 1031ccd979bdSMark Fasheh while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 1032ccd979bdSMark Fasheh node_num = ocfs2_node_map_first_set_bit(osb, 1033ccd979bdSMark Fasheh &osb->recovery_map); 1034ccd979bdSMark Fasheh if (node_num == O2NM_INVALID_NODE_NUM) { 1035ccd979bdSMark Fasheh mlog(0, "Out of nodes to recover.\n"); 1036ccd979bdSMark Fasheh break; 1037ccd979bdSMark Fasheh } 1038ccd979bdSMark Fasheh 1039ccd979bdSMark Fasheh status = ocfs2_recover_node(osb, node_num); 1040ccd979bdSMark Fasheh if (status < 0) { 1041ccd979bdSMark Fasheh mlog(ML_ERROR, 1042ccd979bdSMark Fasheh "Error %d recovering node %d on device (%u,%u)!\n", 1043ccd979bdSMark Fasheh status, node_num, 1044ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1045ccd979bdSMark Fasheh mlog(ML_ERROR, "Volume requires unmount.\n"); 1046ccd979bdSMark Fasheh continue; 1047ccd979bdSMark Fasheh } 1048ccd979bdSMark Fasheh 1049ccd979bdSMark Fasheh ocfs2_recovery_map_clear(osb, node_num); 1050ccd979bdSMark Fasheh } 1051ccd979bdSMark Fasheh ocfs2_super_unlock(osb, 1); 1052ccd979bdSMark Fasheh 1053ccd979bdSMark Fasheh /* We always run recovery on our own orphan dir - the dead 1054ccd979bdSMark Fasheh * node(s) may have voted "no" on an inode delete earlier. A 1055ccd979bdSMark Fasheh * revote is therefore required. */ 1056ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1057ccd979bdSMark Fasheh NULL); 1058ccd979bdSMark Fasheh 1059ccd979bdSMark Fasheh bail: 1060c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1061ccd979bdSMark Fasheh if (!status && 1062ccd979bdSMark Fasheh !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 1063c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1064ccd979bdSMark Fasheh goto restart; 1065ccd979bdSMark Fasheh } 1066ccd979bdSMark Fasheh 1067ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1068ccd979bdSMark Fasheh mb(); /* sync with ocfs2_recovery_thread_running */ 1069ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1070ccd979bdSMark Fasheh 1071c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1072ccd979bdSMark Fasheh 1073ccd979bdSMark Fasheh mlog_exit(status); 1074ccd979bdSMark Fasheh /* no one is callint kthread_stop() for us so the kthread() api 1075ccd979bdSMark Fasheh * requires that we call do_exit(). And it isn't exported, but 1076ccd979bdSMark Fasheh * complete_and_exit() seems to be a minimal wrapper around it. */ 1077ccd979bdSMark Fasheh complete_and_exit(NULL, status); 1078ccd979bdSMark Fasheh return status; 1079ccd979bdSMark Fasheh } 1080ccd979bdSMark Fasheh 1081ccd979bdSMark Fasheh void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1082ccd979bdSMark Fasheh { 1083ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1084ccd979bdSMark Fasheh node_num, osb->node_num); 1085ccd979bdSMark Fasheh 1086c74ec2f7SArjan van de Ven mutex_lock(&osb->recovery_lock); 1087ccd979bdSMark Fasheh if (osb->disable_recovery) 1088ccd979bdSMark Fasheh goto out; 1089ccd979bdSMark Fasheh 1090ccd979bdSMark Fasheh /* People waiting on recovery will wait on 1091ccd979bdSMark Fasheh * the recovery map to empty. */ 1092ccd979bdSMark Fasheh if (!ocfs2_recovery_map_set(osb, node_num)) 1093ccd979bdSMark Fasheh mlog(0, "node %d already be in recovery.\n", node_num); 1094ccd979bdSMark Fasheh 1095ccd979bdSMark Fasheh mlog(0, "starting recovery thread...\n"); 1096ccd979bdSMark Fasheh 1097ccd979bdSMark Fasheh if (osb->recovery_thread_task) 1098ccd979bdSMark Fasheh goto out; 1099ccd979bdSMark Fasheh 1100ccd979bdSMark Fasheh osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 110178427043SMark Fasheh "ocfs2rec"); 1102ccd979bdSMark Fasheh if (IS_ERR(osb->recovery_thread_task)) { 1103ccd979bdSMark Fasheh mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1104ccd979bdSMark Fasheh osb->recovery_thread_task = NULL; 1105ccd979bdSMark Fasheh } 1106ccd979bdSMark Fasheh 1107ccd979bdSMark Fasheh out: 1108c74ec2f7SArjan van de Ven mutex_unlock(&osb->recovery_lock); 1109ccd979bdSMark Fasheh wake_up(&osb->recovery_event); 1110ccd979bdSMark Fasheh 1111ccd979bdSMark Fasheh mlog_exit_void(); 1112ccd979bdSMark Fasheh } 1113ccd979bdSMark Fasheh 1114ccd979bdSMark Fasheh /* Does the actual journal replay and marks the journal inode as 1115ccd979bdSMark Fasheh * clean. Will only replay if the journal inode is marked dirty. */ 1116ccd979bdSMark Fasheh static int ocfs2_replay_journal(struct ocfs2_super *osb, 1117ccd979bdSMark Fasheh int node_num, 1118ccd979bdSMark Fasheh int slot_num) 1119ccd979bdSMark Fasheh { 1120ccd979bdSMark Fasheh int status; 1121ccd979bdSMark Fasheh int got_lock = 0; 1122ccd979bdSMark Fasheh unsigned int flags; 1123ccd979bdSMark Fasheh struct inode *inode = NULL; 1124ccd979bdSMark Fasheh struct ocfs2_dinode *fe; 1125ccd979bdSMark Fasheh journal_t *journal = NULL; 1126ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1127ccd979bdSMark Fasheh 1128ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1129ccd979bdSMark Fasheh slot_num); 1130ccd979bdSMark Fasheh if (inode == NULL) { 1131ccd979bdSMark Fasheh status = -EACCES; 1132ccd979bdSMark Fasheh mlog_errno(status); 1133ccd979bdSMark Fasheh goto done; 1134ccd979bdSMark Fasheh } 1135ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1136ccd979bdSMark Fasheh status = -EACCES; 1137ccd979bdSMark Fasheh iput(inode); 1138ccd979bdSMark Fasheh inode = NULL; 1139ccd979bdSMark Fasheh mlog_errno(status); 1140ccd979bdSMark Fasheh goto done; 1141ccd979bdSMark Fasheh } 1142ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1143ccd979bdSMark Fasheh 1144ccd979bdSMark Fasheh status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, 1145ccd979bdSMark Fasheh OCFS2_META_LOCK_RECOVERY); 1146ccd979bdSMark Fasheh if (status < 0) { 1147ccd979bdSMark Fasheh mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); 1148ccd979bdSMark Fasheh if (status != -ERESTARTSYS) 1149ccd979bdSMark Fasheh mlog(ML_ERROR, "Could not lock journal!\n"); 1150ccd979bdSMark Fasheh goto done; 1151ccd979bdSMark Fasheh } 1152ccd979bdSMark Fasheh got_lock = 1; 1153ccd979bdSMark Fasheh 1154ccd979bdSMark Fasheh fe = (struct ocfs2_dinode *) bh->b_data; 1155ccd979bdSMark Fasheh 1156ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1157ccd979bdSMark Fasheh 1158ccd979bdSMark Fasheh if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1159ccd979bdSMark Fasheh mlog(0, "No recovery required for node %d\n", node_num); 1160ccd979bdSMark Fasheh goto done; 1161ccd979bdSMark Fasheh } 1162ccd979bdSMark Fasheh 1163ccd979bdSMark Fasheh mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 1164ccd979bdSMark Fasheh node_num, slot_num, 1165ccd979bdSMark Fasheh MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1166ccd979bdSMark Fasheh 1167ccd979bdSMark Fasheh OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1168ccd979bdSMark Fasheh 1169ccd979bdSMark Fasheh status = ocfs2_force_read_journal(inode); 1170ccd979bdSMark Fasheh if (status < 0) { 1171ccd979bdSMark Fasheh mlog_errno(status); 1172ccd979bdSMark Fasheh goto done; 1173ccd979bdSMark Fasheh } 1174ccd979bdSMark Fasheh 1175ccd979bdSMark Fasheh mlog(0, "calling journal_init_inode\n"); 1176ccd979bdSMark Fasheh journal = journal_init_inode(inode); 1177ccd979bdSMark Fasheh if (journal == NULL) { 1178ccd979bdSMark Fasheh mlog(ML_ERROR, "Linux journal layer error\n"); 1179ccd979bdSMark Fasheh status = -EIO; 1180ccd979bdSMark Fasheh goto done; 1181ccd979bdSMark Fasheh } 1182ccd979bdSMark Fasheh 1183ccd979bdSMark Fasheh status = journal_load(journal); 1184ccd979bdSMark Fasheh if (status < 0) { 1185ccd979bdSMark Fasheh mlog_errno(status); 1186ccd979bdSMark Fasheh if (!igrab(inode)) 1187ccd979bdSMark Fasheh BUG(); 1188ccd979bdSMark Fasheh journal_destroy(journal); 1189ccd979bdSMark Fasheh goto done; 1190ccd979bdSMark Fasheh } 1191ccd979bdSMark Fasheh 1192ccd979bdSMark Fasheh ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1193ccd979bdSMark Fasheh 1194ccd979bdSMark Fasheh /* wipe the journal */ 1195ccd979bdSMark Fasheh mlog(0, "flushing the journal.\n"); 1196ccd979bdSMark Fasheh journal_lock_updates(journal); 1197ccd979bdSMark Fasheh status = journal_flush(journal); 1198ccd979bdSMark Fasheh journal_unlock_updates(journal); 1199ccd979bdSMark Fasheh if (status < 0) 1200ccd979bdSMark Fasheh mlog_errno(status); 1201ccd979bdSMark Fasheh 1202ccd979bdSMark Fasheh /* This will mark the node clean */ 1203ccd979bdSMark Fasheh flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1204ccd979bdSMark Fasheh flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1205ccd979bdSMark Fasheh fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1206ccd979bdSMark Fasheh 1207ccd979bdSMark Fasheh status = ocfs2_write_block(osb, bh, inode); 1208ccd979bdSMark Fasheh if (status < 0) 1209ccd979bdSMark Fasheh mlog_errno(status); 1210ccd979bdSMark Fasheh 1211ccd979bdSMark Fasheh if (!igrab(inode)) 1212ccd979bdSMark Fasheh BUG(); 1213ccd979bdSMark Fasheh 1214ccd979bdSMark Fasheh journal_destroy(journal); 1215ccd979bdSMark Fasheh 1216ccd979bdSMark Fasheh done: 1217ccd979bdSMark Fasheh /* drop the lock on this nodes journal */ 1218ccd979bdSMark Fasheh if (got_lock) 1219ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 1220ccd979bdSMark Fasheh 1221ccd979bdSMark Fasheh if (inode) 1222ccd979bdSMark Fasheh iput(inode); 1223ccd979bdSMark Fasheh 1224ccd979bdSMark Fasheh if (bh) 1225ccd979bdSMark Fasheh brelse(bh); 1226ccd979bdSMark Fasheh 1227ccd979bdSMark Fasheh mlog_exit(status); 1228ccd979bdSMark Fasheh return status; 1229ccd979bdSMark Fasheh } 1230ccd979bdSMark Fasheh 1231ccd979bdSMark Fasheh /* 1232ccd979bdSMark Fasheh * Do the most important parts of node recovery: 1233ccd979bdSMark Fasheh * - Replay it's journal 1234ccd979bdSMark Fasheh * - Stamp a clean local allocator file 1235ccd979bdSMark Fasheh * - Stamp a clean truncate log 1236ccd979bdSMark Fasheh * - Mark the node clean 1237ccd979bdSMark Fasheh * 1238ccd979bdSMark Fasheh * If this function completes without error, a node in OCFS2 can be 1239ccd979bdSMark Fasheh * said to have been safely recovered. As a result, failure during the 1240ccd979bdSMark Fasheh * second part of a nodes recovery process (local alloc recovery) is 1241ccd979bdSMark Fasheh * far less concerning. 1242ccd979bdSMark Fasheh */ 1243ccd979bdSMark Fasheh static int ocfs2_recover_node(struct ocfs2_super *osb, 1244ccd979bdSMark Fasheh int node_num) 1245ccd979bdSMark Fasheh { 1246ccd979bdSMark Fasheh int status = 0; 1247ccd979bdSMark Fasheh int slot_num; 1248ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1249ccd979bdSMark Fasheh struct ocfs2_dinode *la_copy = NULL; 1250ccd979bdSMark Fasheh struct ocfs2_dinode *tl_copy = NULL; 1251ccd979bdSMark Fasheh 1252ccd979bdSMark Fasheh mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1253ccd979bdSMark Fasheh node_num, osb->node_num); 1254ccd979bdSMark Fasheh 1255ccd979bdSMark Fasheh mlog(0, "checking node %d\n", node_num); 1256ccd979bdSMark Fasheh 1257ccd979bdSMark Fasheh /* Should not ever be called to recover ourselves -- in that 1258ccd979bdSMark Fasheh * case we should've called ocfs2_journal_load instead. */ 1259ebdec83bSEric Sesterhenn / snakebyte BUG_ON(osb->node_num == node_num); 1260ccd979bdSMark Fasheh 1261ccd979bdSMark Fasheh slot_num = ocfs2_node_num_to_slot(si, node_num); 1262ccd979bdSMark Fasheh if (slot_num == OCFS2_INVALID_SLOT) { 1263ccd979bdSMark Fasheh status = 0; 1264ccd979bdSMark Fasheh mlog(0, "no slot for this node, so no recovery required.\n"); 1265ccd979bdSMark Fasheh goto done; 1266ccd979bdSMark Fasheh } 1267ccd979bdSMark Fasheh 1268ccd979bdSMark Fasheh mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1269ccd979bdSMark Fasheh 1270ccd979bdSMark Fasheh status = ocfs2_replay_journal(osb, node_num, slot_num); 1271ccd979bdSMark Fasheh if (status < 0) { 1272ccd979bdSMark Fasheh mlog_errno(status); 1273ccd979bdSMark Fasheh goto done; 1274ccd979bdSMark Fasheh } 1275ccd979bdSMark Fasheh 1276ccd979bdSMark Fasheh /* Stamp a clean local alloc file AFTER recovering the journal... */ 1277ccd979bdSMark Fasheh status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1278ccd979bdSMark Fasheh if (status < 0) { 1279ccd979bdSMark Fasheh mlog_errno(status); 1280ccd979bdSMark Fasheh goto done; 1281ccd979bdSMark Fasheh } 1282ccd979bdSMark Fasheh 1283ccd979bdSMark Fasheh /* An error from begin_truncate_log_recovery is not 1284ccd979bdSMark Fasheh * serious enough to warrant halting the rest of 1285ccd979bdSMark Fasheh * recovery. */ 1286ccd979bdSMark Fasheh status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1287ccd979bdSMark Fasheh if (status < 0) 1288ccd979bdSMark Fasheh mlog_errno(status); 1289ccd979bdSMark Fasheh 1290ccd979bdSMark Fasheh /* Likewise, this would be a strange but ultimately not so 1291ccd979bdSMark Fasheh * harmful place to get an error... */ 1292ccd979bdSMark Fasheh ocfs2_clear_slot(si, slot_num); 1293ccd979bdSMark Fasheh status = ocfs2_update_disk_slots(osb, si); 1294ccd979bdSMark Fasheh if (status < 0) 1295ccd979bdSMark Fasheh mlog_errno(status); 1296ccd979bdSMark Fasheh 1297ccd979bdSMark Fasheh /* This will kfree the memory pointed to by la_copy and tl_copy */ 1298ccd979bdSMark Fasheh ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1299ccd979bdSMark Fasheh tl_copy); 1300ccd979bdSMark Fasheh 1301ccd979bdSMark Fasheh status = 0; 1302ccd979bdSMark Fasheh done: 1303ccd979bdSMark Fasheh 1304ccd979bdSMark Fasheh mlog_exit(status); 1305ccd979bdSMark Fasheh return status; 1306ccd979bdSMark Fasheh } 1307ccd979bdSMark Fasheh 1308ccd979bdSMark Fasheh /* Test node liveness by trylocking his journal. If we get the lock, 1309ccd979bdSMark Fasheh * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1310ccd979bdSMark Fasheh * still alive (we couldn't get the lock) and < 0 on error. */ 1311ccd979bdSMark Fasheh static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1312ccd979bdSMark Fasheh int slot_num) 1313ccd979bdSMark Fasheh { 1314ccd979bdSMark Fasheh int status, flags; 1315ccd979bdSMark Fasheh struct inode *inode = NULL; 1316ccd979bdSMark Fasheh 1317ccd979bdSMark Fasheh inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1318ccd979bdSMark Fasheh slot_num); 1319ccd979bdSMark Fasheh if (inode == NULL) { 1320ccd979bdSMark Fasheh mlog(ML_ERROR, "access error\n"); 1321ccd979bdSMark Fasheh status = -EACCES; 1322ccd979bdSMark Fasheh goto bail; 1323ccd979bdSMark Fasheh } 1324ccd979bdSMark Fasheh if (is_bad_inode(inode)) { 1325ccd979bdSMark Fasheh mlog(ML_ERROR, "access error (bad inode)\n"); 1326ccd979bdSMark Fasheh iput(inode); 1327ccd979bdSMark Fasheh inode = NULL; 1328ccd979bdSMark Fasheh status = -EACCES; 1329ccd979bdSMark Fasheh goto bail; 1330ccd979bdSMark Fasheh } 1331ccd979bdSMark Fasheh SET_INODE_JOURNAL(inode); 1332ccd979bdSMark Fasheh 1333ccd979bdSMark Fasheh flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1334ccd979bdSMark Fasheh status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags); 1335ccd979bdSMark Fasheh if (status < 0) { 1336ccd979bdSMark Fasheh if (status != -EAGAIN) 1337ccd979bdSMark Fasheh mlog_errno(status); 1338ccd979bdSMark Fasheh goto bail; 1339ccd979bdSMark Fasheh } 1340ccd979bdSMark Fasheh 1341ccd979bdSMark Fasheh ocfs2_meta_unlock(inode, 1); 1342ccd979bdSMark Fasheh bail: 1343ccd979bdSMark Fasheh if (inode) 1344ccd979bdSMark Fasheh iput(inode); 1345ccd979bdSMark Fasheh 1346ccd979bdSMark Fasheh return status; 1347ccd979bdSMark Fasheh } 1348ccd979bdSMark Fasheh 1349ccd979bdSMark Fasheh /* Call this underneath ocfs2_super_lock. It also assumes that the 1350ccd979bdSMark Fasheh * slot info struct has been updated from disk. */ 1351ccd979bdSMark Fasheh int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1352ccd979bdSMark Fasheh { 1353ccd979bdSMark Fasheh int status, i, node_num; 1354ccd979bdSMark Fasheh struct ocfs2_slot_info *si = osb->slot_info; 1355ccd979bdSMark Fasheh 1356ccd979bdSMark Fasheh /* This is called with the super block cluster lock, so we 1357ccd979bdSMark Fasheh * know that the slot map can't change underneath us. */ 1358ccd979bdSMark Fasheh 1359ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1360ccd979bdSMark Fasheh for(i = 0; i < si->si_num_slots; i++) { 1361ccd979bdSMark Fasheh if (i == osb->slot_num) 1362ccd979bdSMark Fasheh continue; 1363ccd979bdSMark Fasheh if (ocfs2_is_empty_slot(si, i)) 1364ccd979bdSMark Fasheh continue; 1365ccd979bdSMark Fasheh 1366ccd979bdSMark Fasheh node_num = si->si_global_node_nums[i]; 1367ccd979bdSMark Fasheh if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) 1368ccd979bdSMark Fasheh continue; 1369ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1370ccd979bdSMark Fasheh 1371ccd979bdSMark Fasheh /* Ok, we have a slot occupied by another node which 1372ccd979bdSMark Fasheh * is not in the recovery map. We trylock his journal 1373ccd979bdSMark Fasheh * file here to test if he's alive. */ 1374ccd979bdSMark Fasheh status = ocfs2_trylock_journal(osb, i); 1375ccd979bdSMark Fasheh if (!status) { 1376ccd979bdSMark Fasheh /* Since we're called from mount, we know that 1377ccd979bdSMark Fasheh * the recovery thread can't race us on 1378ccd979bdSMark Fasheh * setting / checking the recovery bits. */ 1379ccd979bdSMark Fasheh ocfs2_recovery_thread(osb, node_num); 1380ccd979bdSMark Fasheh } else if ((status < 0) && (status != -EAGAIN)) { 1381ccd979bdSMark Fasheh mlog_errno(status); 1382ccd979bdSMark Fasheh goto bail; 1383ccd979bdSMark Fasheh } 1384ccd979bdSMark Fasheh 1385ccd979bdSMark Fasheh spin_lock(&si->si_lock); 1386ccd979bdSMark Fasheh } 1387ccd979bdSMark Fasheh spin_unlock(&si->si_lock); 1388ccd979bdSMark Fasheh 1389ccd979bdSMark Fasheh status = 0; 1390ccd979bdSMark Fasheh bail: 1391ccd979bdSMark Fasheh mlog_exit(status); 1392ccd979bdSMark Fasheh return status; 1393ccd979bdSMark Fasheh } 1394ccd979bdSMark Fasheh 1395b4df6ed8SMark Fasheh static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1396b4df6ed8SMark Fasheh int slot, 1397b4df6ed8SMark Fasheh struct inode **head) 1398ccd979bdSMark Fasheh { 1399b4df6ed8SMark Fasheh int status; 1400ccd979bdSMark Fasheh struct inode *orphan_dir_inode = NULL; 1401b4df6ed8SMark Fasheh struct inode *iter; 1402ccd979bdSMark Fasheh unsigned long offset, blk, local; 1403ccd979bdSMark Fasheh struct buffer_head *bh = NULL; 1404ccd979bdSMark Fasheh struct ocfs2_dir_entry *de; 1405ccd979bdSMark Fasheh struct super_block *sb = osb->sb; 1406ccd979bdSMark Fasheh 1407ccd979bdSMark Fasheh orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1408ccd979bdSMark Fasheh ORPHAN_DIR_SYSTEM_INODE, 1409ccd979bdSMark Fasheh slot); 1410ccd979bdSMark Fasheh if (!orphan_dir_inode) { 1411ccd979bdSMark Fasheh status = -ENOENT; 1412ccd979bdSMark Fasheh mlog_errno(status); 1413b4df6ed8SMark Fasheh return status; 1414ccd979bdSMark Fasheh } 1415ccd979bdSMark Fasheh 14161b1dcc1bSJes Sorensen mutex_lock(&orphan_dir_inode->i_mutex); 1417ccd979bdSMark Fasheh status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0); 1418ccd979bdSMark Fasheh if (status < 0) { 1419ccd979bdSMark Fasheh mlog_errno(status); 1420ccd979bdSMark Fasheh goto out; 1421ccd979bdSMark Fasheh } 1422ccd979bdSMark Fasheh 1423ccd979bdSMark Fasheh offset = 0; 1424ccd979bdSMark Fasheh iter = NULL; 1425ccd979bdSMark Fasheh while(offset < i_size_read(orphan_dir_inode)) { 1426ccd979bdSMark Fasheh blk = offset >> sb->s_blocksize_bits; 1427ccd979bdSMark Fasheh 1428ccd979bdSMark Fasheh bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0); 1429ccd979bdSMark Fasheh if (!bh) 1430ccd979bdSMark Fasheh status = -EINVAL; 1431ccd979bdSMark Fasheh if (status < 0) { 1432ccd979bdSMark Fasheh if (bh) 1433ccd979bdSMark Fasheh brelse(bh); 1434ccd979bdSMark Fasheh mlog_errno(status); 1435b4df6ed8SMark Fasheh goto out_unlock; 1436ccd979bdSMark Fasheh } 1437ccd979bdSMark Fasheh 1438ccd979bdSMark Fasheh local = 0; 1439ccd979bdSMark Fasheh while(offset < i_size_read(orphan_dir_inode) 1440ccd979bdSMark Fasheh && local < sb->s_blocksize) { 1441ccd979bdSMark Fasheh de = (struct ocfs2_dir_entry *) (bh->b_data + local); 1442ccd979bdSMark Fasheh 1443ccd979bdSMark Fasheh if (!ocfs2_check_dir_entry(orphan_dir_inode, 1444ccd979bdSMark Fasheh de, bh, local)) { 1445ccd979bdSMark Fasheh status = -EINVAL; 1446ccd979bdSMark Fasheh mlog_errno(status); 1447ccd979bdSMark Fasheh brelse(bh); 1448b4df6ed8SMark Fasheh goto out_unlock; 1449ccd979bdSMark Fasheh } 1450ccd979bdSMark Fasheh 1451ccd979bdSMark Fasheh local += le16_to_cpu(de->rec_len); 1452ccd979bdSMark Fasheh offset += le16_to_cpu(de->rec_len); 1453ccd979bdSMark Fasheh 1454ccd979bdSMark Fasheh /* I guess we silently fail on no inode? */ 1455ccd979bdSMark Fasheh if (!le64_to_cpu(de->inode)) 1456ccd979bdSMark Fasheh continue; 1457ccd979bdSMark Fasheh if (de->file_type > OCFS2_FT_MAX) { 1458ccd979bdSMark Fasheh mlog(ML_ERROR, 1459ccd979bdSMark Fasheh "block %llu contains invalid de: " 1460b0697053SMark Fasheh "inode = %llu, rec_len = %u, " 1461ccd979bdSMark Fasheh "name_len = %u, file_type = %u, " 1462ccd979bdSMark Fasheh "name='%.*s'\n", 1463ccd979bdSMark Fasheh (unsigned long long)bh->b_blocknr, 1464b0697053SMark Fasheh (unsigned long long)le64_to_cpu(de->inode), 1465ccd979bdSMark Fasheh le16_to_cpu(de->rec_len), 1466ccd979bdSMark Fasheh de->name_len, 1467ccd979bdSMark Fasheh de->file_type, 1468ccd979bdSMark Fasheh de->name_len, 1469ccd979bdSMark Fasheh de->name); 1470ccd979bdSMark Fasheh continue; 1471ccd979bdSMark Fasheh } 1472ccd979bdSMark Fasheh if (de->name_len == 1 && !strncmp(".", de->name, 1)) 1473ccd979bdSMark Fasheh continue; 1474ccd979bdSMark Fasheh if (de->name_len == 2 && !strncmp("..", de->name, 2)) 1475ccd979bdSMark Fasheh continue; 1476ccd979bdSMark Fasheh 147724c19ef4SMark Fasheh iter = ocfs2_iget(osb, le64_to_cpu(de->inode), 147824c19ef4SMark Fasheh OCFS2_FI_FLAG_NOLOCK); 1479ccd979bdSMark Fasheh if (IS_ERR(iter)) 1480ccd979bdSMark Fasheh continue; 1481ccd979bdSMark Fasheh 1482b0697053SMark Fasheh mlog(0, "queue orphan %llu\n", 1483b0697053SMark Fasheh (unsigned long long)OCFS2_I(iter)->ip_blkno); 1484b4df6ed8SMark Fasheh /* No locking is required for the next_orphan 1485b4df6ed8SMark Fasheh * queue as there is only ever a single 1486b4df6ed8SMark Fasheh * process doing orphan recovery. */ 1487b4df6ed8SMark Fasheh OCFS2_I(iter)->ip_next_orphan = *head; 1488b4df6ed8SMark Fasheh *head = iter; 1489ccd979bdSMark Fasheh } 1490ccd979bdSMark Fasheh brelse(bh); 1491ccd979bdSMark Fasheh } 1492ccd979bdSMark Fasheh 1493b4df6ed8SMark Fasheh out_unlock: 1494ccd979bdSMark Fasheh ocfs2_meta_unlock(orphan_dir_inode, 0); 1495b4df6ed8SMark Fasheh out: 1496b4df6ed8SMark Fasheh mutex_unlock(&orphan_dir_inode->i_mutex); 1497ccd979bdSMark Fasheh iput(orphan_dir_inode); 1498b4df6ed8SMark Fasheh return status; 1499b4df6ed8SMark Fasheh } 1500b4df6ed8SMark Fasheh 1501b4df6ed8SMark Fasheh static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1502b4df6ed8SMark Fasheh int slot) 1503b4df6ed8SMark Fasheh { 1504b4df6ed8SMark Fasheh int ret; 1505b4df6ed8SMark Fasheh 1506b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1507b4df6ed8SMark Fasheh ret = !osb->osb_orphan_wipes[slot]; 1508b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1509b4df6ed8SMark Fasheh return ret; 1510b4df6ed8SMark Fasheh } 1511b4df6ed8SMark Fasheh 1512b4df6ed8SMark Fasheh static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1513b4df6ed8SMark Fasheh int slot) 1514b4df6ed8SMark Fasheh { 1515b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1516b4df6ed8SMark Fasheh /* Mark ourselves such that new processes in delete_inode() 1517b4df6ed8SMark Fasheh * know to quit early. */ 1518b4df6ed8SMark Fasheh ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1519b4df6ed8SMark Fasheh while (osb->osb_orphan_wipes[slot]) { 1520b4df6ed8SMark Fasheh /* If any processes are already in the middle of an 1521b4df6ed8SMark Fasheh * orphan wipe on this dir, then we need to wait for 1522b4df6ed8SMark Fasheh * them. */ 1523b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1524b4df6ed8SMark Fasheh wait_event_interruptible(osb->osb_wipe_event, 1525b4df6ed8SMark Fasheh ocfs2_orphan_recovery_can_continue(osb, slot)); 1526b4df6ed8SMark Fasheh spin_lock(&osb->osb_lock); 1527b4df6ed8SMark Fasheh } 1528b4df6ed8SMark Fasheh spin_unlock(&osb->osb_lock); 1529b4df6ed8SMark Fasheh } 1530b4df6ed8SMark Fasheh 1531b4df6ed8SMark Fasheh static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1532b4df6ed8SMark Fasheh int slot) 1533b4df6ed8SMark Fasheh { 1534b4df6ed8SMark Fasheh ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1535b4df6ed8SMark Fasheh } 1536b4df6ed8SMark Fasheh 1537b4df6ed8SMark Fasheh /* 1538b4df6ed8SMark Fasheh * Orphan recovery. Each mounted node has it's own orphan dir which we 1539b4df6ed8SMark Fasheh * must run during recovery. Our strategy here is to build a list of 1540b4df6ed8SMark Fasheh * the inodes in the orphan dir and iget/iput them. The VFS does 1541b4df6ed8SMark Fasheh * (most) of the rest of the work. 1542b4df6ed8SMark Fasheh * 1543b4df6ed8SMark Fasheh * Orphan recovery can happen at any time, not just mount so we have a 1544b4df6ed8SMark Fasheh * couple of extra considerations. 1545b4df6ed8SMark Fasheh * 1546b4df6ed8SMark Fasheh * - We grab as many inodes as we can under the orphan dir lock - 1547b4df6ed8SMark Fasheh * doing iget() outside the orphan dir risks getting a reference on 1548b4df6ed8SMark Fasheh * an invalid inode. 1549b4df6ed8SMark Fasheh * - We must be sure not to deadlock with other processes on the 1550b4df6ed8SMark Fasheh * system wanting to run delete_inode(). This can happen when they go 1551b4df6ed8SMark Fasheh * to lock the orphan dir and the orphan recovery process attempts to 1552b4df6ed8SMark Fasheh * iget() inside the orphan dir lock. This can be avoided by 1553b4df6ed8SMark Fasheh * advertising our state to ocfs2_delete_inode(). 1554b4df6ed8SMark Fasheh */ 1555b4df6ed8SMark Fasheh static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1556b4df6ed8SMark Fasheh int slot) 1557b4df6ed8SMark Fasheh { 1558b4df6ed8SMark Fasheh int ret = 0; 1559b4df6ed8SMark Fasheh struct inode *inode = NULL; 1560b4df6ed8SMark Fasheh struct inode *iter; 1561b4df6ed8SMark Fasheh struct ocfs2_inode_info *oi; 1562b4df6ed8SMark Fasheh 1563b4df6ed8SMark Fasheh mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1564b4df6ed8SMark Fasheh 1565b4df6ed8SMark Fasheh ocfs2_mark_recovering_orphan_dir(osb, slot); 1566b4df6ed8SMark Fasheh ret = ocfs2_queue_orphans(osb, slot, &inode); 1567b4df6ed8SMark Fasheh ocfs2_clear_recovering_orphan_dir(osb, slot); 1568b4df6ed8SMark Fasheh 1569b4df6ed8SMark Fasheh /* Error here should be noted, but we want to continue with as 1570b4df6ed8SMark Fasheh * many queued inodes as we've got. */ 1571b4df6ed8SMark Fasheh if (ret) 1572b4df6ed8SMark Fasheh mlog_errno(ret); 1573ccd979bdSMark Fasheh 1574ccd979bdSMark Fasheh while (inode) { 1575ccd979bdSMark Fasheh oi = OCFS2_I(inode); 1576b0697053SMark Fasheh mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1577ccd979bdSMark Fasheh 1578ccd979bdSMark Fasheh iter = oi->ip_next_orphan; 1579ccd979bdSMark Fasheh 1580ccd979bdSMark Fasheh spin_lock(&oi->ip_lock); 1581ccd979bdSMark Fasheh /* Delete voting may have set these on the assumption 1582ccd979bdSMark Fasheh * that the other node would wipe them successfully. 1583ccd979bdSMark Fasheh * If they are still in the node's orphan dir, we need 1584ccd979bdSMark Fasheh * to reset that state. */ 1585ccd979bdSMark Fasheh oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1586ccd979bdSMark Fasheh 1587ccd979bdSMark Fasheh /* Set the proper information to get us going into 1588ccd979bdSMark Fasheh * ocfs2_delete_inode. */ 1589ccd979bdSMark Fasheh oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1590ccd979bdSMark Fasheh oi->ip_orphaned_slot = slot; 1591ccd979bdSMark Fasheh spin_unlock(&oi->ip_lock); 1592ccd979bdSMark Fasheh 1593ccd979bdSMark Fasheh iput(inode); 1594ccd979bdSMark Fasheh 1595ccd979bdSMark Fasheh inode = iter; 1596ccd979bdSMark Fasheh } 1597ccd979bdSMark Fasheh 1598b4df6ed8SMark Fasheh return ret; 1599ccd979bdSMark Fasheh } 1600ccd979bdSMark Fasheh 1601ccd979bdSMark Fasheh static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1602ccd979bdSMark Fasheh { 1603ccd979bdSMark Fasheh /* This check is good because ocfs2 will wait on our recovery 1604ccd979bdSMark Fasheh * thread before changing it to something other than MOUNTED 1605ccd979bdSMark Fasheh * or DISABLED. */ 1606ccd979bdSMark Fasheh wait_event(osb->osb_mount_event, 1607ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1608ccd979bdSMark Fasheh atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1609ccd979bdSMark Fasheh 1610ccd979bdSMark Fasheh /* If there's an error on mount, then we may never get to the 1611ccd979bdSMark Fasheh * MOUNTED flag, but this is set right before 1612ccd979bdSMark Fasheh * dismount_volume() so we can trust it. */ 1613ccd979bdSMark Fasheh if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1614ccd979bdSMark Fasheh mlog(0, "mount error, exiting!\n"); 1615ccd979bdSMark Fasheh return -EBUSY; 1616ccd979bdSMark Fasheh } 1617ccd979bdSMark Fasheh 1618ccd979bdSMark Fasheh return 0; 1619ccd979bdSMark Fasheh } 1620ccd979bdSMark Fasheh 1621ccd979bdSMark Fasheh static int ocfs2_commit_thread(void *arg) 1622ccd979bdSMark Fasheh { 1623ccd979bdSMark Fasheh int status; 1624ccd979bdSMark Fasheh struct ocfs2_super *osb = arg; 1625ccd979bdSMark Fasheh struct ocfs2_journal *journal = osb->journal; 1626ccd979bdSMark Fasheh 1627ccd979bdSMark Fasheh /* we can trust j_num_trans here because _should_stop() is only set in 1628ccd979bdSMark Fasheh * shutdown and nobody other than ourselves should be able to start 1629ccd979bdSMark Fasheh * transactions. committing on shutdown might take a few iterations 1630ccd979bdSMark Fasheh * as final transactions put deleted inodes on the list */ 1631ccd979bdSMark Fasheh while (!(kthread_should_stop() && 1632ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) == 0)) { 1633ccd979bdSMark Fasheh 1634745ae8baSMark Fasheh wait_event_interruptible(osb->checkpoint_event, 1635ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans) 1636745ae8baSMark Fasheh || kthread_should_stop()); 1637ccd979bdSMark Fasheh 1638ccd979bdSMark Fasheh status = ocfs2_commit_cache(osb); 1639ccd979bdSMark Fasheh if (status < 0) 1640ccd979bdSMark Fasheh mlog_errno(status); 1641ccd979bdSMark Fasheh 1642ccd979bdSMark Fasheh if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1643ccd979bdSMark Fasheh mlog(ML_KTHREAD, 1644ccd979bdSMark Fasheh "commit_thread: %u transactions pending on " 1645ccd979bdSMark Fasheh "shutdown\n", 1646ccd979bdSMark Fasheh atomic_read(&journal->j_num_trans)); 1647ccd979bdSMark Fasheh } 1648ccd979bdSMark Fasheh } 1649ccd979bdSMark Fasheh 1650ccd979bdSMark Fasheh return 0; 1651ccd979bdSMark Fasheh } 1652ccd979bdSMark Fasheh 1653ccd979bdSMark Fasheh /* Look for a dirty journal without taking any cluster locks. Used for 1654ccd979bdSMark Fasheh * hard readonly access to determine whether the file system journals 1655ccd979bdSMark Fasheh * require recovery. */ 1656ccd979bdSMark Fasheh int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1657ccd979bdSMark Fasheh { 1658ccd979bdSMark Fasheh int ret = 0; 1659ccd979bdSMark Fasheh unsigned int slot; 1660ccd979bdSMark Fasheh struct buffer_head *di_bh; 1661ccd979bdSMark Fasheh struct ocfs2_dinode *di; 1662ccd979bdSMark Fasheh struct inode *journal = NULL; 1663ccd979bdSMark Fasheh 1664ccd979bdSMark Fasheh for(slot = 0; slot < osb->max_slots; slot++) { 1665ccd979bdSMark Fasheh journal = ocfs2_get_system_file_inode(osb, 1666ccd979bdSMark Fasheh JOURNAL_SYSTEM_INODE, 1667ccd979bdSMark Fasheh slot); 1668ccd979bdSMark Fasheh if (!journal || is_bad_inode(journal)) { 1669ccd979bdSMark Fasheh ret = -EACCES; 1670ccd979bdSMark Fasheh mlog_errno(ret); 1671ccd979bdSMark Fasheh goto out; 1672ccd979bdSMark Fasheh } 1673ccd979bdSMark Fasheh 1674ccd979bdSMark Fasheh di_bh = NULL; 1675ccd979bdSMark Fasheh ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, 1676ccd979bdSMark Fasheh 0, journal); 1677ccd979bdSMark Fasheh if (ret < 0) { 1678ccd979bdSMark Fasheh mlog_errno(ret); 1679ccd979bdSMark Fasheh goto out; 1680ccd979bdSMark Fasheh } 1681ccd979bdSMark Fasheh 1682ccd979bdSMark Fasheh di = (struct ocfs2_dinode *) di_bh->b_data; 1683ccd979bdSMark Fasheh 1684ccd979bdSMark Fasheh if (le32_to_cpu(di->id1.journal1.ij_flags) & 1685ccd979bdSMark Fasheh OCFS2_JOURNAL_DIRTY_FL) 1686ccd979bdSMark Fasheh ret = -EROFS; 1687ccd979bdSMark Fasheh 1688ccd979bdSMark Fasheh brelse(di_bh); 1689ccd979bdSMark Fasheh if (ret) 1690ccd979bdSMark Fasheh break; 1691ccd979bdSMark Fasheh } 1692ccd979bdSMark Fasheh 1693ccd979bdSMark Fasheh out: 1694ccd979bdSMark Fasheh if (journal) 1695ccd979bdSMark Fasheh iput(journal); 1696ccd979bdSMark Fasheh 1697ccd979bdSMark Fasheh return ret; 1698ccd979bdSMark Fasheh } 1699