/* -*- mode: c; c-basic-offset: 8; -*-
* vim: noexpandtab sw=8 ts=8 sts=0:
*
* journal.c
*
* Defines functions of journalling api
*
* Copyright (C) 2003, 2004 Oracle. All rights reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this program; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 021110-1307, USA.
*/
#include <linux/fs.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/kthread.h>
#define MLOG_MASK_PREFIX ML_JOURNAL
#include <cluster/masklog.h>
#include "ocfs2.h"
#include "alloc.h"
#include "dir.h"
#include "dlmglue.h"
#include "extent_map.h"
#include "heartbeat.h"
#include "inode.h"
#include "journal.h"
#include "localalloc.h"
#include "slot_map.h"
#include "super.h"
#include "sysfile.h"
#include "buffer_head_io.h"
DEFINE_SPINLOCK(trans_inc_lock);
static int ocfs2_force_read_journal(struct inode *inode);
static int ocfs2_recover_node(struct ocfs2_super *osb,
int node_num);
static int __ocfs2_recovery_thread(void *arg);
static int ocfs2_commit_cache(struct ocfs2_super *osb);
static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
int dirty);
static int ocfs2_trylock_journal(struct ocfs2_super *osb,
int slot_num);
static int ocfs2_recover_orphans(struct ocfs2_super *osb,
int slot);
static int ocfs2_commit_thread(void *arg);
/*
* The recovery_list is a simple linked list of node numbers to recover.
* It is protected by the recovery_lock.
*/
struct ocfs2_recovery_map {
unsigned int rm_used;
unsigned int *rm_entries;
};
int ocfs2_recovery_init(struct ocfs2_super *osb)
{
struct ocfs2_recovery_map *rm;
mutex_init(&osb->recovery_lock);
osb->disable_recovery = 0;
osb->recovery_thread_task = NULL;
init_waitqueue_head(&osb->recovery_event);
rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
osb->max_slots * sizeof(unsigned int),
GFP_KERNEL);
if (!rm) {
mlog_errno(-ENOMEM);
return -ENOMEM;
}
rm->rm_entries = (unsigned int *)((char *)rm +
sizeof(struct ocfs2_recovery_map));
osb->recovery_map = rm;
return 0;
}
/* we can't grab the goofy sem lock from inside wait_event, so we use
* memory barriers to make sure that we'll see the null task before
* being woken up */
static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
{
mb();
return osb->recovery_thread_task != NULL;
}
void ocfs2_recovery_exit(struct ocfs2_super *osb)
{
struct ocfs2_recovery_map *rm;
/* disable any new recovery threads and wait for any currently
* running ones to exit. Do this before setting the vol_state. */
mutex_lock(&osb->recovery_lock);
osb->disable_recovery = 1;
mutex_unlock(&osb->recovery_lock);
wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
/* At this point, we know that no more recovery threads can be
* launched, so wait for any recovery completion work to
* complete. */
flush_workqueue(ocfs2_wq);
/*
* Now that recovery is shut down, and the osb is about to be
* freed, the osb_lock is not taken here.
*/
rm = osb->recovery_map;
/* XXX: Should we bug if there are dirty entries? */
kfree(rm);
}
static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
unsigned int node_num)
{
int i;
struct ocfs2_recovery_map *rm = osb->recovery_map;
assert_spin_locked(&osb->osb_lock);
for (i = 0; i < rm->rm_used; i++) {
if (rm->rm_entries[i] == node_num)
return 1;
}
return 0;
}
/* Behaves like test-and-set. Returns the previous value */
static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
unsigned int node_num)
{
struct ocfs2_recovery_map *rm = osb->recovery_map;
spin_lock(&osb->osb_lock);
if (__ocfs2_recovery_map_test(osb, node_num)) {
spin_unlock(&osb->osb_lock);
return 1;
}
/* XXX: Can this be exploited? Not from o2dlm... */
BUG_ON(rm->rm_used >= osb->max_slots);
rm->rm_entries[rm->rm_used] = node_num;
rm->rm_used++;
spin_unlock(&osb->osb_lock);
return 0;
}
static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
unsigned int node_num)
{
int i;
struct ocfs2_recovery_map *rm = osb->recovery_map;
spin_lock(&osb->osb_lock);
for (i = 0; i < rm->rm_used; i++) {
if (rm->rm_entries[i] == node_num)
break;
}
if (i < rm->rm_used) {
/* XXX: be careful with the pointer math */
memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
(rm->rm_used - i - 1) * sizeof(unsigned int));
rm->rm_used--;
}
spin_unlock(&osb->osb_lock);
}
static int ocfs2_commit_cache(struct ocfs2_super *osb)
{
int status = 0;
unsigned int flushed;
unsigned long old_id;
struct ocfs2_journal *journal = NULL;
mlog_entry_void();
journal = osb->journal;
/* Flush all pending commits and checkpoint the journal. */
down_write(&journal->j_trans_barrier);
if (atomic_read(&journal->j_num_trans) == 0) {
up_write(&journal->j_trans_barrier);
mlog(0, "No transactions for me to flush!\n");
goto finally;
}
journal_lock_updates(journal->j_journal);
status = journal_flush(journal->j_journal);
journal_unlock_updates(journal->j_journal);
if (status < 0) {
up_write(&journal->j_trans_barrier);
mlog_errno(status);
goto finally;
}
old_id = ocfs2_inc_trans_id(journal);
flushed = atomic_read(&journal->j_num_trans);
atomic_set(&journal->j_num_trans, 0);
up_write(&journal->j_trans_barrier);
mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
journal->j_trans_id, flushed);
ocfs2_wake_downconvert_thread(osb);
wake_up(&journal->j_checkpointed);
finally:
mlog_exit(status);
return status;
}
/* pass it NULL and it will allocate a new handle object for you. If
* you pass it a handle however, it may still return error, in which
* case it has free'd the passed handle for you. */
handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
{
journal_t *journal = osb->journal->j_journal;
handle_t *handle;
BUG_ON(!osb || !osb->journal->j_journal);
if (ocfs2_is_hard_readonly(osb))
return ERR_PTR(-EROFS);
BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
BUG_ON(max_buffs <= 0);
/* JBD might support this, but our journalling code doesn't yet. */
if (journal_current_handle()) {
mlog(ML_ERROR, "Recursive transaction attempted!\n");
BUG();
}
down_read(&osb->journal->j_trans_barrier);
handle = journal_start(journal, max_buffs);
if (IS_ERR(handle)) {
up_read(&osb->journal->j_trans_barrier);
mlog_errno(PTR_ERR(handle));
if (is_journal_aborted(journal)) {
ocfs2_abort(osb->sb, "Detected aborted journal");
handle = ERR_PTR(-EROFS);
}
} else {
if (!ocfs2_mount_local(osb))
atomic_inc(&(osb->journal->j_num_trans));
}
return handle;
}
int ocfs2_commit_trans(struct ocfs2_super *osb,
handle_t *handle)
{
int ret;
struct ocfs2_journal *journal = osb->journal;
BUG_ON(!handle);
ret = journal_stop(handle);
if (ret < 0)
mlog_errno(ret);
up_read(&journal->j_trans_barrier);
return ret;
}
/*
* 'nblocks' is what you want to add to the current
* transaction. extend_trans will either extend the current handle by
* nblocks, or commit it and start a new one with nblocks credits.
*
* This might call journal_restart() which will commit dirty buffers
* and then restart the transaction. Before calling
* ocfs2_extend_trans(), any changed blocks should have been
* dirtied. After calling it, all blocks which need to be changed must
* go through another set of journal_access/journal_dirty calls.
*
* WARNING: This will not release any semaphores or disk locks taken
* during the transaction, so make sure they were taken *before*
* start_trans or we'll have ordering deadlocks.
*
* WARNING2: Note that we do *not* drop j_trans_barrier here. This is
* good because transaction ids haven't yet been recorded on the
* cluster locks associated with this handle.
*/
int ocfs2_extend_trans(handle_t *handle, int nblocks)
{
int status;
BUG_ON(!handle);
BUG_ON(!nblocks);
mlog_entry_void();
mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
#ifdef OCFS2_DEBUG_FS
status = 1;
#else
status = journal_extend(handle, nblocks);
if (status < 0) {
mlog_errno(status);
goto bail;
}
#endif
if (status > 0) {
mlog(0, "journal_extend failed, trying journal_restart\n");
status = journal_restart(handle, nblocks);
if (status < 0) {
mlog_errno(status);
goto bail;
}
}
status = 0;
bail:
mlog_exit(status);
return status;
}
int ocfs2_journal_access(handle_t *handle,
struct inode *inode,
struct buffer_head *bh,
int type)
{
int status;
BUG_ON(!inode);
BUG_ON(!handle);
BUG_ON(!bh);
mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
(unsigned long long)bh->b_blocknr, type,
(type == OCFS2_JOURNAL_ACCESS_CREATE) ?
"OCFS2_JOURNAL_ACCESS_CREATE" :
"OCFS2_JOURNAL_ACCESS_WRITE",
bh->b_size);
/* we can safely remove this assertion after testing. */
if (!buffer_uptodate(bh)) {
mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
mlog(ML_ERROR, "b_blocknr=%llu\n",
(unsigned long long)bh->b_blocknr);
BUG();
}
/* Set the current transaction information on the inode so
* that the locking code knows whether it can drop it's locks
* on this inode or not. We're protected from the commit
* thread updating the current transaction id until
* ocfs2_commit_trans() because ocfs2_start_trans() took
* j_trans_barrier for us. */
ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
switch (type) {
case OCFS2_JOURNAL_ACCESS_CREATE:
case OCFS2_JOURNAL_ACCESS_WRITE:
status = journal_get_write_access(handle, bh);
break;
case OCFS2_JOURNAL_ACCESS_UNDO:
status = journal_get_undo_access(handle, bh);
break;
default:
status = -EINVAL;
mlog(ML_ERROR, "Uknown access type!\n");
}
mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
if (status < 0)
mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
status, type);
mlog_exit(status);
return status;
}
int ocfs2_journal_dirty(handle_t *handle,
struct buffer_head *bh)
{
int status;
mlog_entry("(bh->b_blocknr=%llu)\n",
(unsigned long long)bh->b_blocknr);
status = journal_dirty_metadata(handle, bh);
if (status < 0)
mlog(ML_ERROR, "Could not dirty metadata buffer. "
"(bh->b_blocknr=%llu)\n",
(unsigned long long)bh->b_blocknr);
mlog_exit(status);
return status;
}
int ocfs2_journal_dirty_data(handle_t *handle,
struct buffer_head *bh)
{
int err = journal_dirty_data(handle, bh);
if (err)
mlog_errno(err);
/* TODO: When we can handle it, abort the handle and go RO on
* error here. */
return err;
}
#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD_DEFAULT_MAX_COMMIT_AGE)
void ocfs2_set_journal_params(struct ocfs2_super *osb)
{
journal_t *journal = osb->journal->j_journal;
unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
if (osb->osb_commit_interval)
commit_interval = osb->osb_commit_interval;
spin_lock(&journal->j_state_lock);
journal->j_commit_interval = commit_interval;
if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
journal->j_flags |= JFS_BARRIER;
else
journal->j_flags &= ~JFS_BARRIER;
spin_unlock(&journal->j_state_lock);
}
int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
{
int status = -1;
struct inode *inode = NULL; /* the journal inode */
journal_t *j_journal = NULL;
struct ocfs2_dinode *di = NULL;
struct buffer_head *bh = NULL;
struct ocfs2_super *osb;
int inode_lock = 0;
mlog_entry_void();
BUG_ON(!journal);
osb = journal->j_osb;
/* already have the inode for our journal */
inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
osb->slot_num);
if (inode == NULL) {
status = -EACCES;
mlog_errno(status);
goto done;
}
if (is_bad_inode(inode)) {
mlog(ML_ERROR, "access error (bad inode)\n");
iput(inode);
inode = NULL;
status = -EACCES;
goto done;
}
SET_INODE_JOURNAL(inode);
OCFS2_I(inode)->ip_open_count++;
/* Skip recovery waits here - journal inode metadata never
* changes in a live cluster so it can be considered an
* exception to the rule. */
status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
if (status < 0) {
if (status != -ERESTARTSYS)
mlog(ML_ERROR, "Could not get lock on journal!\n");
goto done;
}
inode_lock = 1;
di = (struct ocfs2_dinode *)bh->b_data;
if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
inode->i_size);
status = -EINVAL;
goto done;
}
mlog(0, "inode->i_size = %lld\n", inode->i_size);
mlog(0, "inode->i_blocks = %llu\n",
(unsigned long long)inode->i_blocks);
mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
/* call the kernels journal init function now */
j_journal = journal_init_inode(inode);
if (j_journal == NULL) {
mlog(ML_ERROR, "Linux journal layer error\n");
status = -EINVAL;
goto done;
}
mlog(0, "Returned from journal_init_inode\n");
mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
*dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
OCFS2_JOURNAL_DIRTY_FL);
journal->j_journal = j_journal;
journal->j_inode = inode;
journal->j_bh = bh;
ocfs2_set_journal_params(osb);
journal->j_state = OCFS2_JOURNAL_LOADED;
status = 0;
done:
if (status < 0) {
if (inode_lock)
ocfs2_inode_unlock(inode, 1);
if (bh != NULL)
brelse(bh);
if (inode) {
OCFS2_I(inode)->ip_open_count--;
iput(inode);
}
}
mlog_exit(status);
return status;
}
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
int dirty)
{
int status;
unsigned int flags;
struct ocfs2_journal *journal = osb->journal;
struct buffer_head *bh = journal->j_bh;
struct ocfs2_dinode *fe;
mlog_entry_void();
fe = (struct ocfs2_dinode *)bh->b_data;
if (!OCFS2_IS_VALID_DINODE(fe)) {
/* This is called from startup/shutdown which will
* handle the errors in a specific manner, so no need
* to call ocfs2_error() here. */
mlog(ML_ERROR, "Journal dinode %llu has invalid "
"signature: %.*s",
(unsigned long long)le64_to_cpu(fe->i_blkno), 7,
fe->i_signature);
status = -EIO;
goto out;
}
flags = le32_to_cpu(fe->id1.journal1.ij_flags);
if (dirty)
flags |= OCFS2_JOURNAL_DIRTY_FL;
else
flags &= ~OCFS2_JOURNAL_DIRTY_FL;
fe->id1.journal1.ij_flags = cpu_to_le32(flags);
status = ocfs2_write_block(osb, bh, journal->j_inode);
if (status < 0)
mlog_errno(status);
out:
mlog_exit(status);
return status;
}
/*
* If the journal has been kmalloc'd it needs to be freed after this
* call.
*/
void ocfs2_journal_shutdown(struct ocfs2_super *osb)
{
struct ocfs2_journal *journal = NULL;
int status = 0;
struct inode *inode = NULL;
int num_running_trans = 0;
mlog_entry_void();
BUG_ON(!osb);
journal = osb->journal;
if (!journal)
goto done;
inode = journal->j_inode;
if (journal->j_state != OCFS2_JOURNAL_LOADED)
goto done;
/* need to inc inode use count as journal_destroy will iput. */
if (!igrab(inode))
BUG();
num_running_trans = atomic_read(&(osb->journal->j_num_trans));
if (num_running_trans > 0)
mlog(0, "Shutting down journal: must wait on %d "
"running transactions!\n",
num_running_trans);
/* Do a commit_cache here. It will flush our journal, *and*
* release any locks that are still held.
* set the SHUTDOWN flag and release the trans lock.
* the commit thread will take the trans lock for us below. */
journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
* drop the trans_lock (which we want to hold until we
* completely destroy the journal. */
if (osb->commit_task) {
/* Wait for the commit thread */
mlog(0, "Waiting for ocfs2commit to exit....\n");
kthread_stop(osb->commit_task);
osb->commit_task = NULL;
}
BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
if (ocfs2_mount_local(osb)) {
journal_lock_updates(journal->j_journal);
status = journal_flush(journal->j_journal);
journal_unlock_updates(journal->j_journal);
if (status < 0)
mlog_errno(status);
}
if (status == 0) {
/*
* Do not toggle if flush was unsuccessful otherwise
* will leave dirty metadata in a "clean" journal
*/
status = ocfs2_journal_toggle_dirty(osb, 0);
if (status < 0)
mlog_errno(status);
}
/* Shutdown the kernel journal system */
journal_destroy(journal->j_journal);
OCFS2_I(inode)->ip_open_count--;
/* unlock our journal */
ocfs2_inode_unlock(inode, 1);
brelse(journal->j_bh);
journal->j_bh = NULL;
journal->j_state = OCFS2_JOURNAL_FREE;
// up_write(&journal->j_trans_barrier);
done:
if (inode)
iput(inode);
mlog_exit_void();
}
static void ocfs2_clear_journal_error(struct super_block *sb,
journal_t *journal,
int slot)
{
int olderr;
olderr = journal_errno(journal);
if (olderr) {
mlog(ML_ERROR, "File system error %d recorded in "
"journal %u.\n", olderr, slot);
mlog(ML_ERROR, "File system on device %s needs checking.\n",
sb->s_id);
journal_ack_err(journal);
journal_clear_err(journal);
}
}
int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
{
int status = 0;
struct ocfs2_super *osb;
mlog_entry_void();
if (!journal)
BUG();
osb = journal->j_osb;
status = journal_load(journal->j_journal);
if (status < 0) {
mlog(ML_ERROR, "Failed to load journal!\n");
goto done;
}
ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
status = ocfs2_journal_toggle_dirty(osb, 1);
if (status < 0) {
mlog_errno(status);
goto done;
}
/* Launch the commit thread */
if (!local) {
osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
"ocfs2cmt");
if (IS_ERR(osb->commit_task)) {
status = PTR_ERR(osb->commit_task);
osb->commit_task = NULL;
mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
"error=%d", status);
goto done;
}
} else
osb->commit_task = NULL;
done:
mlog_exit(status);
return status;
}
/* 'full' flag tells us whether we clear out all blocks or if we just
* mark the journal clean */
int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
{
int status;
mlog_entry_void();
BUG_ON(!journal);
status = journal_wipe(journal->j_journal, full);
if (status < 0) {
mlog_errno(status);
goto bail;
}
status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
if (status < 0)
mlog_errno(status);
bail:
mlog_exit(status);
return status;
}
static int ocfs2_recovery_completed(struct ocfs2_super *osb)
{
int empty;
struct ocfs2_recovery_map *rm = osb->recovery_map;
spin_lock(&osb->osb_lock);
empty = (rm->rm_used == 0);
spin_unlock(&osb->osb_lock);
return empty;
}
void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
{
wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
}
/*
* JBD Might read a cached version of another nodes journal file. We
* don't want this as this file changes often and we get no
* notification on those changes. The only way to be sure that we've
* got the most up to date version of those blocks then is to force
* read them off disk. Just searching through the buffer cache won't
* work as there may be pages backing this file which are still marked
* up to date. We know things can't change on this file underneath us
* as we have the lock by now :)
*/
static int ocfs2_force_read_journal(struct inode *inode)
{
int status = 0;
int i;
u64 v_blkno, p_blkno, p_blocks, num_blocks;
#define CONCURRENT_JOURNAL_FILL 32ULL
struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
mlog_entry_void();
memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
v_blkno = 0;
while (v_blkno < num_blocks) {
status = ocfs2_extent_map_get_blocks(inode, v_blkno,
&p_blkno, &p_blocks, NULL);
if (status < 0) {
mlog_errno(status);
goto bail;
}
if (p_blocks > CONCURRENT_JOURNAL_FILL)
p_blocks = CONCURRENT_JOURNAL_FILL;
/* We are reading journal data which should not
* be put in the uptodate cache */
status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
p_blkno, p_blocks, bhs, 0,
NULL);
if (status < 0) {
mlog_errno(status);
goto bail;
}
for(i = 0; i < p_blocks; i++) {
brelse(bhs[i]);
bhs[i] = NULL;
}
v_blkno += p_blocks;
}
bail:
for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
if (bhs[i])
brelse(bhs[i]);
mlog_exit(status);
return status;
}
struct ocfs2_la_recovery_item {
struct list_head lri_list;
int lri_slot;
struct ocfs2_dinode *lri_la_dinode;
struct ocfs2_dinode *lri_tl_dinode;
};
/* Does the second half of the recovery process. By this point, the
* node is marked clean and can actually be considered recovered,
* hence it's no longer in the recovery map, but there's still some
* cleanup we can do which shouldn't happen within the recovery thread
* as locking in that context becomes very difficult if we are to take
* recovering nodes into account.
*
* NOTE: This function can and will sleep on recovery of other nodes
* during cluster locking, just like any other ocfs2 process.
*/
void ocfs2_complete_recovery(struct work_struct *work)
{
int ret;
struct ocfs2_journal *journal =
container_of(work, struct ocfs2_journal, j_recovery_work);
struct ocfs2_super *osb = journal->j_osb;
struct ocfs2_dinode *la_dinode, *tl_dinode;
struct ocfs2_la_recovery_item *item, *n;
LIST_HEAD(tmp_la_list);
mlog_entry_void();
mlog(0, "completing recovery from keventd\n");
spin_lock(&journal->j_lock);
list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
spin_unlock(&journal->j_lock);
list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
list_del_init(&item->lri_list);
mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
la_dinode = item->lri_la_dinode;
if (la_dinode) {
mlog(0, "Clean up local alloc %llu\n",
(unsigned long long)le64_to_cpu(la_dinode->i_blkno));
ret = ocfs2_complete_local_alloc_recovery(osb,
la_dinode);
if (ret < 0)
mlog_errno(ret);
kfree(la_dinode);
}
tl_dinode = item->lri_tl_dinode;
if (tl_dinode) {
mlog(0, "Clean up truncate log %llu\n",
(unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
ret = ocfs2_complete_truncate_log_recovery(osb,
tl_dinode);
if (ret < 0)
mlog_errno(ret);
kfree(tl_dinode);
}
ret = ocfs2_recover_orphans(osb, item->lri_slot);
if (ret < 0)
mlog_errno(ret);
kfree(item);
}
mlog(0, "Recovery completion\n");
mlog_exit_void();
}
/* NOTE: This function always eats your references to la_dinode and
* tl_dinode, either manually on error, or by passing them to
* ocfs2_complete_recovery */
static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
int slot_num,
struct ocfs2_dinode *la_dinode,
struct ocfs2_dinode *tl_dinode)
{
struct ocfs2_la_recovery_item *item;
item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
if (!item) {
/* Though we wish to avoid it, we are in fact safe in
* skipping local alloc cleanup as fsck.ocfs2 is more
* than capable of reclaiming unused space. */
if (la_dinode)
kfree(la_dinode);
if (tl_dinode)
kfree(tl_dinode);
mlog_errno(-ENOMEM);
return;
}
INIT_LIST_HEAD(&item->lri_list);
item->lri_la_dinode = la_dinode;
item->lri_slot = slot_num;
item->lri_tl_dinode = tl_dinode;
spin_lock(&journal->j_lock);
list_add_tail(&item->lri_list, &journal->j_la_cleanups);
queue_work(ocfs2_wq, &journal->j_recovery_work);
spin_unlock(&journal->j_lock);
}
/* Called by the mount code to queue recovery the last part of
* recovery for it's own slot. */
void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
{
struct ocfs2_journal *journal = osb->journal;
if (osb->dirty) {
/* No need to queue up our truncate_log as regular
* cleanup will catch that. */
ocfs2_queue_recovery_completion(journal,
osb->slot_num,
osb->local_alloc_copy,
NULL);
ocfs2_schedule_truncate_log_flush(osb, 0);
osb->local_alloc_copy = NULL;
osb->dirty = 0;
}
}
static int __ocfs2_recovery_thread(void *arg)
{
int status, node_num;
struct ocfs2_super *osb = arg;
struct ocfs2_recovery_map *rm = osb->recovery_map;
mlog_entry_void();
status = ocfs2_wait_on_mount(osb);
if (status < 0) {
goto bail;
}
restart:
status = ocfs2_super_lock(osb, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
spin_lock(&osb->osb_lock);
while (rm->rm_used) {
/* It's always safe to remove entry zero, as we won't
* clear it until ocfs2_recover_node() has succeeded. */
node_num = rm->rm_entries[0];
spin_unlock(&osb->osb_lock);
status = ocfs2_recover_node(osb, node_num);
if (!status) {
ocfs2_recovery_map_clear(osb, node_num);
} else {
mlog(ML_ERROR,
"Error %d recovering node %d on device (%u,%u)!\n",
status, node_num,
MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
mlog(ML_ERROR, "Volume requires unmount.\n");
}
spin_lock(&osb->osb_lock);
}
spin_unlock(&osb->osb_lock);
mlog(0, "All nodes recovered\n");
ocfs2_super_unlock(osb, 1);
/* We always run recovery on our own orphan dir - the dead
* node(s) may have disallowd a previos inode delete. Re-processing
* is therefore required. */
ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
NULL);
bail:
mutex_lock(&osb->recovery_lock);
if (!status && !ocfs2_recovery_completed(osb)) {
mutex_unlock(&osb->recovery_lock);
goto restart;
}
osb->recovery_thread_task = NULL;
mb(); /* sync with ocfs2_recovery_thread_running */
wake_up(&osb->recovery_event);
mutex_unlock(&osb->recovery_lock);
mlog_exit(status);
/* no one is callint kthread_stop() for us so the kthread() api
* requires that we call do_exit(). And it isn't exported, but
* complete_and_exit() seems to be a minimal wrapper around it. */
complete_and_exit(NULL, status);
return status;
}
void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
{
mlog_entry("(node_num=%d, osb->node_num = %d)\n",
node_num, osb->node_num);
mutex_lock(&osb->recovery_lock);
if (osb->disable_recovery)
goto out;
/* People waiting on recovery will wait on
* the recovery map to empty. */
if (ocfs2_recovery_map_set(osb, node_num))
mlog(0, "node %d already in recovery map.\n", node_num);
mlog(0, "starting recovery thread...\n");
if (osb->recovery_thread_task)
goto out;
osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
"ocfs2rec");
if (IS_ERR(osb->recovery_thread_task)) {
mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
osb->recovery_thread_task = NULL;
}
out:
mutex_unlock(&osb->recovery_lock);
wake_up(&osb->recovery_event);
mlog_exit_void();
}
/* Does the actual journal replay and marks the journal inode as
* clean. Will only replay if the journal inode is marked dirty. */
static int ocfs2_replay_journal(struct ocfs2_super *osb,
int node_num,
int slot_num)
{
int status;
int got_lock = 0;
unsigned int flags;
struct inode *inode = NULL;
struct ocfs2_dinode *fe;
journal_t *journal = NULL;
struct buffer_head *bh = NULL;
inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
slot_num);
if (inode == NULL) {
status = -EACCES;
mlog_errno(status);
goto done;
}
if (is_bad_inode(inode)) {
status = -EACCES;
iput(inode);
inode = NULL;
mlog_errno(status);
goto done;
}
SET_INODE_JOURNAL(inode);
status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
if (status < 0) {
mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
if (status != -ERESTARTSYS)
mlog(ML_ERROR, "Could not lock journal!\n");
goto done;
}
got_lock = 1;
fe = (struct ocfs2_dinode *) bh->b_data;
flags = le32_to_cpu(fe->id1.journal1.ij_flags);
if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
mlog(0, "No recovery required for node %d\n", node_num);
goto done;
}
mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
node_num, slot_num,
MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
status = ocfs2_force_read_journal(inode);
if (status < 0) {
mlog_errno(status);
goto done;
}
mlog(0, "calling journal_init_inode\n");
journal = journal_init_inode(inode);
if (journal == NULL) {
mlog(ML_ERROR, "Linux journal layer error\n");
status = -EIO;
goto done;
}
status = journal_load(journal);
if (status < 0) {
mlog_errno(status);
if (!igrab(inode))
BUG();
journal_destroy(journal);
goto done;
}
ocfs2_clear_journal_error(osb->sb, journal, slot_num);
/* wipe the journal */
mlog(0, "flushing the journal.\n");
journal_lock_updates(journal);
status = journal_flush(journal);
journal_unlock_updates(journal);
if (status < 0)
mlog_errno(status);
/* This will mark the node clean */
flags = le32_to_cpu(fe->id1.journal1.ij_flags);
flags &= ~OCFS2_JOURNAL_DIRTY_FL;
fe->id1.journal1.ij_flags = cpu_to_le32(flags);
status = ocfs2_write_block(osb, bh, inode);
if (status < 0)
mlog_errno(status);
if (!igrab(inode))
BUG();
journal_destroy(journal);
done:
/* drop the lock on this nodes journal */
if (got_lock)
ocfs2_inode_unlock(inode, 1);
if (inode)
iput(inode);
if (bh)
brelse(bh);
mlog_exit(status);
return status;
}
/*
* Do the most important parts of node recovery:
* - Replay it's journal
* - Stamp a clean local allocator file
* - Stamp a clean truncate log
* - Mark the node clean
*
* If this function completes without error, a node in OCFS2 can be
* said to have been safely recovered. As a result, failure during the
* second part of a nodes recovery process (local alloc recovery) is
* far less concerning.
*/
static int ocfs2_recover_node(struct ocfs2_super *osb,
int node_num)
{
int status = 0;
int slot_num;
struct ocfs2_dinode *la_copy = NULL;
struct ocfs2_dinode *tl_copy = NULL;
mlog_entry("(node_num=%d, osb->node_num = %d)\n",
node_num, osb->node_num);
mlog(0, "checking node %d\n", node_num);
/* Should not ever be called to recover ourselves -- in that
* case we should've called ocfs2_journal_load instead. */
BUG_ON(osb->node_num == node_num);
slot_num = ocfs2_node_num_to_slot(osb, node_num);
if (slot_num == -ENOENT) {
status = 0;
mlog(0, "no slot for this node, so no recovery required.\n");
goto done;
}
mlog(0, "node %d was using slot %d\n", node_num, slot_num);
status = ocfs2_replay_journal(osb, node_num, slot_num);
if (status < 0) {
mlog_errno(status);
goto done;
}
/* Stamp a clean local alloc file AFTER recovering the journal... */
status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
if (status < 0) {
mlog_errno(status);
goto done;
}
/* An error from begin_truncate_log_recovery is not
* serious enough to warrant halting the rest of
* recovery. */
status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
if (status < 0)
mlog_errno(status);
/* Likewise, this would be a strange but ultimately not so
* harmful place to get an error... */
status = ocfs2_clear_slot(osb, slot_num);
if (status < 0)
mlog_errno(status);
/* This will kfree the memory pointed to by la_copy and tl_copy */
ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
tl_copy);
status = 0;
done:
mlog_exit(status);
return status;
}
/* Test node liveness by trylocking his journal. If we get the lock,
* we drop it here. Return 0 if we got the lock, -EAGAIN if node is
* still alive (we couldn't get the lock) and < 0 on error. */
static int ocfs2_trylock_journal(struct ocfs2_super *osb,
int slot_num)
{
int status, flags;
struct inode *inode = NULL;
inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
slot_num);
if (inode == NULL) {
mlog(ML_ERROR, "access error\n");
status = -EACCES;
goto bail;
}
if (is_bad_inode(inode)) {
mlog(ML_ERROR, "access error (bad inode)\n");
iput(inode);
inode = NULL;
status = -EACCES;
goto bail;
}
SET_INODE_JOURNAL(inode);
flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
if (status < 0) {
if (status != -EAGAIN)
mlog_errno(status);
goto bail;
}
ocfs2_inode_unlock(inode, 1);
bail:
if (inode)
iput(inode);
return status;
}
/* Call this underneath ocfs2_super_lock. It also assumes that the
* slot info struct has been updated from disk. */
int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
{
unsigned int node_num;
int status, i;
/* This is called with the super block cluster lock, so we
* know that the slot map can't change underneath us. */
spin_lock(&osb->osb_lock);
for (i = 0; i < osb->max_slots; i++) {
if (i == osb->slot_num)
continue;
status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
if (status == -ENOENT)
continue;
if (__ocfs2_recovery_map_test(osb, node_num))
continue;
spin_unlock(&osb->osb_lock);
/* Ok, we have a slot occupied by another node which
* is not in the recovery map. We trylock his journal
* file here to test if he's alive. */
status = ocfs2_trylock_journal(osb, i);
if (!status) {
/* Since we're called from mount, we know that
* the recovery thread can't race us on
* setting / checking the recovery bits. */
ocfs2_recovery_thread(osb, node_num);
} else if ((status < 0) && (status != -EAGAIN)) {
mlog_errno(status);
goto bail;
}
spin_lock(&osb->osb_lock);
}
spin_unlock(&osb->osb_lock);
status = 0;
bail:
mlog_exit(status);
return status;
}
struct ocfs2_orphan_filldir_priv {
struct inode *head;
struct ocfs2_super *osb;
};
static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
loff_t pos, u64 ino, unsigned type)
{
struct ocfs2_orphan_filldir_priv *p = priv;
struct inode *iter;
if (name_len == 1 && !strncmp(".", name, 1))
return 0;
if (name_len == 2 && !strncmp("..", name, 2))
return 0;
/* Skip bad inodes so that recovery can continue */
iter = ocfs2_iget(p->osb, ino,
OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
if (IS_ERR(iter))
return 0;
mlog(0, "queue orphan %llu\n",
(unsigned long long)OCFS2_I(iter)->ip_blkno);
/* No locking is required for the next_orphan queue as there
* is only ever a single process doing orphan recovery. */
OCFS2_I(iter)->ip_next_orphan = p->head;
p->head = iter;
return 0;
}
static int ocfs2_queue_orphans(struct ocfs2_super *osb,
int slot,
struct inode **head)
{
int status;
struct inode *orphan_dir_inode = NULL;
struct ocfs2_orphan_filldir_priv priv;
loff_t pos = 0;
priv.osb = osb;
priv.head = *head;
orphan_dir_inode = ocfs2_get_system_file_inode(osb,
ORPHAN_DIR_SYSTEM_INODE,
slot);
if (!orphan_dir_inode) {
status = -ENOENT;
mlog_errno(status);
return status;
}
mutex_lock(&orphan_dir_inode->i_mutex);
status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
if (status < 0) {
mlog_errno(status);
goto out;
}
status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
ocfs2_orphan_filldir);
if (status) {
mlog_errno(status);
goto out_cluster;
}
*head = priv.head;
out_cluster:
ocfs2_inode_unlock(orphan_dir_inode, 0);
out:
mutex_unlock(&orphan_dir_inode->i_mutex);
iput(orphan_dir_inode);
return status;
}
static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
int slot)
{
int ret;
spin_lock(&osb->osb_lock);
ret = !osb->osb_orphan_wipes[slot];
spin_unlock(&osb->osb_lock);
return ret;
}
static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
int slot)
{
spin_lock(&osb->osb_lock);
/* Mark ourselves such that new processes in delete_inode()
* know to quit early. */
ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
while (osb->osb_orphan_wipes[slot]) {
/* If any processes are already in the middle of an
* orphan wipe on this dir, then we need to wait for
* them. */
spin_unlock(&osb->osb_lock);
wait_event_interruptible(osb->osb_wipe_event,
ocfs2_orphan_recovery_can_continue(osb, slot));
spin_lock(&osb->osb_lock);
}
spin_unlock(&osb->osb_lock);
}
static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
int slot)
{
ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
}
/*
* Orphan recovery. Each mounted node has it's own orphan dir which we
* must run during recovery. Our strategy here is to build a list of
* the inodes in the orphan dir and iget/iput them. The VFS does
* (most) of the rest of the work.
*
* Orphan recovery can happen at any time, not just mount so we have a
* couple of extra considerations.
*
* - We grab as many inodes as we can under the orphan dir lock -
* doing iget() outside the orphan dir risks getting a reference on
* an invalid inode.
* - We must be sure not to deadlock with other processes on the
* system wanting to run delete_inode(). This can happen when they go
* to lock the orphan dir and the orphan recovery process attempts to
* iget() inside the orphan dir lock. This can be avoided by
* advertising our state to ocfs2_delete_inode().
*/
static int ocfs2_recover_orphans(struct ocfs2_super *osb,
int slot)
{
int ret = 0;
struct inode *inode = NULL;
struct inode *iter;
struct ocfs2_inode_info *oi;
mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
ocfs2_mark_recovering_orphan_dir(osb, slot);
ret = ocfs2_queue_orphans(osb, slot, &inode);
ocfs2_clear_recovering_orphan_dir(osb, slot);
/* Error here should be noted, but we want to continue with as
* many queued inodes as we've got. */
if (ret)
mlog_errno(ret);
while (inode) {
oi = OCFS2_I(inode);
mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
iter = oi->ip_next_orphan;
spin_lock(&oi->ip_lock);
/* The remote delete code may have set these on the
* assumption that the other node would wipe them
* successfully. If they are still in the node's
* orphan dir, we need to reset that state. */
oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
/* Set the proper information to get us going into
* ocfs2_delete_inode. */
oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
spin_unlock(&oi->ip_lock);
iput(inode);
inode = iter;
}
return ret;
}
static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
{
/* This check is good because ocfs2 will wait on our recovery
* thread before changing it to something other than MOUNTED
* or DISABLED. */
wait_event(osb->osb_mount_event,
atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
atomic_read(&osb->vol_state) == VOLUME_DISABLED);
/* If there's an error on mount, then we may never get to the
* MOUNTED flag, but this is set right before
* dismount_volume() so we can trust it. */
if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
mlog(0, "mount error, exiting!\n");
return -EBUSY;
}
return 0;
}
static int ocfs2_commit_thread(void *arg)
{
int status;
struct ocfs2_super *osb = arg;
struct ocfs2_journal *journal = osb->journal;
/* we can trust j_num_trans here because _should_stop() is only set in
* shutdown and nobody other than ourselves should be able to start
* transactions. committing on shutdown might take a few iterations
* as final transactions put deleted inodes on the list */
while (!(kthread_should_stop() &&
atomic_read(&journal->j_num_trans) == 0)) {
wait_event_interruptible(osb->checkpoint_event,
atomic_read(&journal->j_num_trans)
|| kthread_should_stop());
status = ocfs2_commit_cache(osb);
if (status < 0)
mlog_errno(status);
if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
mlog(ML_KTHREAD,
"commit_thread: %u transactions pending on "
"shutdown\n",
atomic_read(&journal->j_num_trans));
}
}
return 0;
}
/* Look for a dirty journal without taking any cluster locks. Used for
* hard readonly access to determine whether the file system journals
* require recovery. */
int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
{
int ret = 0;
unsigned int slot;
struct buffer_head *di_bh;
struct ocfs2_dinode *di;
struct inode *journal = NULL;
for(slot = 0; slot < osb->max_slots; slot++) {
journal = ocfs2_get_system_file_inode(osb,
JOURNAL_SYSTEM_INODE,
slot);
if (!journal || is_bad_inode(journal)) {
ret = -EACCES;
mlog_errno(ret);
goto out;
}
di_bh = NULL;
ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
0, journal);
if (ret < 0) {
mlog_errno(ret);
goto out;
}
di = (struct ocfs2_dinode *) di_bh->b_data;
if (le32_to_cpu(di->id1.journal1.ij_flags) &
OCFS2_JOURNAL_DIRTY_FL)
ret = -EROFS;
brelse(di_bh);
if (ret)
break;
}
out:
if (journal)
iput(journal);
return ret;
}
