/* hb going down releases any holds we might have had due to this node from
* conn_up, conn_err, or hb_up */
void o2quo_hb_down(u8 node)
{
struct o2quo_state *qs = &o2quo_state;
spin_lock(&qs->qs_lock);
qs->qs_heartbeating--;
mlog_bug_on_msg(qs->qs_heartbeating < 0,
"node %u, %d heartbeating\n",
node, qs->qs_heartbeating);
mlog_bug_on_msg(!test_bit(node, qs->qs_hb_bm), "node %u\n", node);
clear_bit(node, qs->qs_hb_bm);
mlog(0, "node %u, %d total\n", node, qs->qs_heartbeating);
o2quo_clear_hold(qs, node);
spin_unlock(&qs->qs_lock);
}
static void o2quo_set_hold(struct o2quo_state *qs, u8 node)
{
assert_spin_locked(&qs->qs_lock);
if (!test_and_set_bit(node, qs->qs_hold_bm)) {
qs->qs_holds++;
mlog_bug_on_msg(qs->qs_holds == O2NM_MAX_NODES,
"node %u\n", node);
mlog(0, "node %u, %d total\n", node, qs->qs_holds);
}
}
/* This is analogous to hb_up. as a node's connection comes up we delay the
* quorum decision until we see it heartbeating. the hold will be droped in
* hb_up or hb_down. it might be perpetuated by con_err until hb_down. if
* it's already heartbeating we we might be dropping a hold that conn_up got.
* */
void o2quo_conn_up(u8 node)
{
struct o2quo_state *qs = &o2quo_state;
spin_lock(&qs->qs_lock);
qs->qs_connected++;
mlog_bug_on_msg(qs->qs_connected == O2NM_MAX_NODES,
"node %u\n", node);
mlog_bug_on_msg(test_bit(node, qs->qs_conn_bm), "node %u\n", node);
set_bit(node, qs->qs_conn_bm);
mlog(0, "node %u, %d total\n", node, qs->qs_connected);
if (!test_bit(node, qs->qs_hb_bm))
o2quo_set_hold(qs, node);
else
o2quo_clear_hold(qs, node);
spin_unlock(&qs->qs_lock);
}
static void o2quo_clear_hold(struct o2quo_state *qs, u8 node)
{
assert_spin_locked(&qs->qs_lock);
if (test_and_clear_bit(node, qs->qs_hold_bm)) {
mlog(0, "node %u, %d total\n", node, qs->qs_holds - 1);
if (--qs->qs_holds == 0) {
if (qs->qs_pending) {
qs->qs_pending = 0;
schedule_work(&qs->qs_work);
}
}
mlog_bug_on_msg(qs->qs_holds < 0, "node %u, holds %d\n",
node, qs->qs_holds);
}
}
static int ocfs2_find_actor(struct inode *inode, void *opaque)
{
struct ocfs2_find_inode_args *args = NULL;
struct ocfs2_inode_info *oi = OCFS2_I(inode);
int ret = 0;
args = opaque;
mlog_bug_on_msg(!inode, "No inode in find actor!\n");
trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
if (oi->ip_blkno != args->fi_blkno)
goto bail;
ret = 1;
bail:
return ret;
}
static int ocfs2_find_actor(struct inode *inode, void *opaque)
{
struct ocfs2_find_inode_args *args = NULL;
struct ocfs2_inode_info *oi = OCFS2_I(inode);
int ret = 0;
mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
args = opaque;
mlog_bug_on_msg(!inode, "No inode in find actor!\n");
if (oi->ip_blkno != args->fi_blkno)
goto bail;
ret = 1;
bail:
mlog_exit(ret);
return ret;
}
static void user_ast(struct ocfs2_dlm_lksb *lksb)
{
struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
int status;
mlog(ML_BASTS, "AST fired for lockres %.*s, level %d => %d\n",
lockres->l_namelen, lockres->l_name, lockres->l_level,
lockres->l_requested);
spin_lock(&lockres->l_lock);
status = ocfs2_dlm_lock_status(&lockres->l_lksb);
if (status) {
mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
status, lockres->l_namelen, lockres->l_name);
spin_unlock(&lockres->l_lock);
return;
}
mlog_bug_on_msg(lockres->l_requested == DLM_LOCK_IV,
"Lockres %.*s, requested ivmode. flags 0x%x\n",
lockres->l_namelen, lockres->l_name, lockres->l_flags);
/* we're downconverting. */
if (lockres->l_requested < lockres->l_level) {
if (lockres->l_requested <=
user_highest_compat_lock_level(lockres->l_blocking)) {
lockres->l_blocking = DLM_LOCK_NL;
lockres->l_flags &= ~USER_LOCK_BLOCKED;
}
}
lockres->l_level = lockres->l_requested;
lockres->l_requested = DLM_LOCK_IV;
lockres->l_flags |= USER_LOCK_ATTACHED;
lockres->l_flags &= ~USER_LOCK_BUSY;
spin_unlock(&lockres->l_lock);
wake_up(&lockres->l_event);
}
static void user_ast(void *opaque)
{
struct user_lock_res *lockres = opaque;
struct dlm_lockstatus *lksb;
mlog(0, "AST fired for lockres %.*s\n", lockres->l_namelen,
lockres->l_name);
spin_lock(&lockres->l_lock);
lksb = &(lockres->l_lksb);
if (lksb->status != DLM_NORMAL) {
mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
lksb->status, lockres->l_namelen, lockres->l_name);
spin_unlock(&lockres->l_lock);
return;
}
mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
"Lockres %.*s, requested ivmode. flags 0x%x\n",
lockres->l_namelen, lockres->l_name, lockres->l_flags);
/* we're downconverting. */
if (lockres->l_requested < lockres->l_level) {
if (lockres->l_requested <=
user_highest_compat_lock_level(lockres->l_blocking)) {
lockres->l_blocking = LKM_NLMODE;
lockres->l_flags &= ~USER_LOCK_BLOCKED;
}
}
lockres->l_level = lockres->l_requested;
lockres->l_requested = LKM_IVMODE;
lockres->l_flags |= USER_LOCK_ATTACHED;
lockres->l_flags &= ~USER_LOCK_BUSY;
spin_unlock(&lockres->l_lock);
wake_up(&lockres->l_event);
}
/* we've decided that we won't ever be connecting to the node again. if it's
* still heartbeating we grab a hold that will delay decisions until either the
* node stops heartbeating from hb_down or the caller decides that the node is
* still up and calls still_up */
void o2quo_conn_err(u8 node)
{
struct o2quo_state *qs = &o2quo_state;
spin_lock(&qs->qs_lock);
if (test_bit(node, qs->qs_conn_bm)) {
qs->qs_connected--;
mlog_bug_on_msg(qs->qs_connected < 0,
"node %u, connected %d\n",
node, qs->qs_connected);
clear_bit(node, qs->qs_conn_bm);
}
mlog(0, "node %u, %d total\n", node, qs->qs_connected);
if (test_bit(node, qs->qs_hb_bm))
o2quo_set_hold(qs, node);
spin_unlock(&qs->qs_lock);
}
static void ocfs2_clear_inode(struct inode *inode)
{
int status;
struct ocfs2_inode_info *oi = OCFS2_I(inode);
clear_inode(inode);
trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
inode->i_nlink);
mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
"Inode=%lu\n", inode->i_ino);
dquot_drop(inode);
/* To preven remote deletes we hold open lock before, now it
* is time to unlock PR and EX open locks. */
ocfs2_open_unlock(inode);
/* Do these before all the other work so that we don't bounce
* the downconvert thread while waiting to destroy the locks. */
ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
&oi->ip_la_data_resv);
ocfs2_resv_init_once(&oi->ip_la_data_resv);
/* We very well may get a clear_inode before all an inodes
* metadata has hit disk. Of course, we can't drop any cluster
* locks until the journal has finished with it. The only
* exception here are successfully wiped inodes - their
* metadata can now be considered to be part of the system
* inodes from which it came. */
if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
ocfs2_checkpoint_inode(inode);
mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
"Clear inode of %llu, inode has io markers\n",
(unsigned long long)oi->ip_blkno);
ocfs2_extent_map_trunc(inode, 0);
status = ocfs2_drop_inode_locks(inode);
if (status < 0)
mlog_errno(status);
ocfs2_lock_res_free(&oi->ip_rw_lockres);
ocfs2_lock_res_free(&oi->ip_inode_lockres);
ocfs2_lock_res_free(&oi->ip_open_lockres);
ocfs2_metadata_cache_exit(INODE_CACHE(inode));
mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
"Clear inode of %llu, inode has %u cache items\n",
(unsigned long long)oi->ip_blkno,
INODE_CACHE(inode)->ci_num_cached);
mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
"Clear inode of %llu, inode has a bad flag\n",
(unsigned long long)oi->ip_blkno);
mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
"Clear inode of %llu, inode is locked\n",
(unsigned long long)oi->ip_blkno);
mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
"Clear inode of %llu, io_mutex is locked\n",
(unsigned long long)oi->ip_blkno);
mutex_unlock(&oi->ip_io_mutex);
/*
* down_trylock() returns 0, down_write_trylock() returns 1
* kernel 1, world 0
*/
mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
"Clear inode of %llu, alloc_sem is locked\n",
(unsigned long long)oi->ip_blkno);
up_write(&oi->ip_alloc_sem);
mlog_bug_on_msg(oi->ip_open_count,
"Clear inode of %llu has open count %d\n",
(unsigned long long)oi->ip_blkno, oi->ip_open_count);
/* Clear all other flags. */
oi->ip_flags = 0;
oi->ip_dir_start_lookup = 0;
oi->ip_blkno = 0ULL;
/*
* ip_jinode is used to track txns against this inode. We ensure that
* the journal is flushed before journal shutdown. Thus it is safe to
* have inodes get cleaned up after journal shutdown.
*/
jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
&oi->ip_jinode);
}
int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
void **ret_data)
{
int ret;
unsigned int locklen;
struct dlm_ctxt *dlm = data;
struct dlm_lock_resource *res = NULL;
struct dlm_lock *lock = NULL;
struct dlm_proxy_ast *past = (struct dlm_proxy_ast *) msg->buf;
char *name;
struct list_head *head = NULL;
__be64 cookie;
u32 flags;
u8 node;
if (!dlm_grab(dlm)) {
dlm_error(DLM_REJECTED);
return DLM_REJECTED;
}
mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
"Domain %s not fully joined!\n", dlm->name);
name = past->name;
locklen = past->namelen;
cookie = past->cookie;
flags = be32_to_cpu(past->flags);
node = past->node_idx;
if (locklen > DLM_LOCKID_NAME_MAX) {
ret = DLM_IVBUFLEN;
mlog(ML_ERROR, "Invalid name length (%d) in proxy ast "
"handler!\n", locklen);
goto leave;
}
if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
(LKM_PUT_LVB|LKM_GET_LVB)) {
mlog(ML_ERROR, "Both PUT and GET lvb specified, (0x%x)\n",
flags);
ret = DLM_BADARGS;
goto leave;
}
mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
(flags & LKM_GET_LVB ? "get lvb" : "none"));
mlog(0, "type=%d, blocked_type=%d\n", past->type, past->blocked_type);
if (past->type != DLM_AST &&
past->type != DLM_BAST) {
mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu"
"name=%.*s, node=%u\n", past->type,
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
locklen, name, node);
ret = DLM_IVLOCKID;
goto leave;
}
res = dlm_lookup_lockres(dlm, name, locklen);
if (!res) {
mlog(0, "Got %sast for unknown lockres! cookie=%u:%llu, "
"name=%.*s, node=%u\n", (past->type == DLM_AST ? "" : "b"),
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
locklen, name, node);
ret = DLM_IVLOCKID;
goto leave;
}
/* cannot get a proxy ast message if this node owns it */
BUG_ON(res->owner == dlm->node_num);
mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
res->lockname.name);
spin_lock(&res->spinlock);
if (res->state & DLM_LOCK_RES_RECOVERING) {
mlog(0, "Responding with DLM_RECOVERING!\n");
ret = DLM_RECOVERING;
goto unlock_out;
}
if (res->state & DLM_LOCK_RES_MIGRATING) {
mlog(0, "Responding with DLM_MIGRATING!\n");
ret = DLM_MIGRATING;
goto unlock_out;
}
/* try convert queue for both ast/bast */
head = &res->converting;
lock = NULL;
list_for_each_entry(lock, head, list) {
if (lock->ml.cookie == cookie)
goto do_ast;
}
/* if not on convert, try blocked for ast, granted for bast */
if (past->type == DLM_AST)
head = &res->blocked;
else
head = &res->granted;
list_for_each_entry(lock, head, list) {
if (lock->ml.cookie == cookie)
goto do_ast;
}
mlog(0, "Got %sast for unknown lock! cookie=%u:%llu, name=%.*s, "
"node=%u\n", past->type == DLM_AST ? "" : "b",
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
locklen, name, node);
ret = DLM_NORMAL;
unlock_out:
spin_unlock(&res->spinlock);
goto leave;
do_ast:
ret = DLM_NORMAL;
if (past->type == DLM_AST) {
/* do not alter lock refcount. switching lists. */
list_move_tail(&lock->list, &res->granted);
mlog(0, "%s: res %.*s, lock %u:%llu, Granted type %d => %d\n",
dlm->name, res->lockname.len, res->lockname.name,
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
lock->ml.type, lock->ml.convert_type);
if (lock->ml.convert_type != LKM_IVMODE) {
lock->ml.type = lock->ml.convert_type;
lock->ml.convert_type = LKM_IVMODE;
} else {
// should already be there....
}
lock->lksb->status = DLM_NORMAL;
/* if we requested the lvb, fetch it into our lksb now */
if (flags & LKM_GET_LVB) {
BUG_ON(!(lock->lksb->flags & DLM_LKSB_GET_LVB));
memcpy(lock->lksb->lvb, past->lvb, DLM_LVB_LEN);
}
}
spin_unlock(&res->spinlock);
if (past->type == DLM_AST)
dlm_do_local_ast(dlm, res, lock);
else
dlm_do_local_bast(dlm, res, lock, past->blocked_type);
leave:
if (res)
dlm_lockres_put(res);
dlm_put(dlm);
return ret;
}
/*
* locking:
* caller needs: none
* taken: takes and drops res->spinlock
* held on exit: none
* returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID,
* return value from dlmunlock_master
*/
int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_unlock_lock *unlock = (struct dlm_unlock_lock *)msg->buf;
struct dlm_lock_resource *res = NULL;
struct list_head *iter;
struct dlm_lock *lock = NULL;
enum dlm_status status = DLM_NORMAL;
int found = 0, i;
struct dlm_lockstatus *lksb = NULL;
int ignore;
u32 flags;
struct list_head *queue;
flags = be32_to_cpu(unlock->flags);
if (flags & LKM_GET_LVB) {
mlog(ML_ERROR, "bad args! GET_LVB specified on unlock!\n");
return DLM_BADARGS;
}
if ((flags & (LKM_PUT_LVB|LKM_CANCEL)) == (LKM_PUT_LVB|LKM_CANCEL)) {
mlog(ML_ERROR, "bad args! cannot modify lvb on a CANCEL "
"request!\n");
return DLM_BADARGS;
}
if (unlock->namelen > DLM_LOCKID_NAME_MAX) {
mlog(ML_ERROR, "Invalid name length in unlock handler!\n");
return DLM_IVBUFLEN;
}
if (!dlm_grab(dlm))
return DLM_REJECTED;
mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
"Domain %s not fully joined!\n", dlm->name);
mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" : "none");
res = dlm_lookup_lockres(dlm, unlock->name, unlock->namelen);
if (!res) {
/* We assume here that a no lock resource simply means
* it was migrated away and destroyed before the other
* node could detect it. */
mlog(0, "returning DLM_FORWARD -- res no longer exists\n");
status = DLM_FORWARD;
goto not_found;
}
queue=&res->granted;
found = 0;
spin_lock(&res->spinlock);
if (res->state & DLM_LOCK_RES_RECOVERING) {
spin_unlock(&res->spinlock);
mlog(0, "returning DLM_RECOVERING\n");
status = DLM_RECOVERING;
goto leave;
}
if (res->state & DLM_LOCK_RES_MIGRATING) {
spin_unlock(&res->spinlock);
mlog(0, "returning DLM_MIGRATING\n");
status = DLM_MIGRATING;
goto leave;
}
if (res->owner != dlm->node_num) {
spin_unlock(&res->spinlock);
mlog(0, "returning DLM_FORWARD -- not master\n");
status = DLM_FORWARD;
goto leave;
}
for (i=0; i<3; i++) {
list_for_each(iter, queue) {
lock = list_entry(iter, struct dlm_lock, list);
if (lock->ml.cookie == unlock->cookie &&
lock->ml.node == unlock->node_idx) {
dlm_lock_get(lock);
found = 1;
break;
}
}
if (found)
break;
/* scan granted -> converting -> blocked queues */
queue++;
}
/*
* Append this record to the tail of the extent map. It must be
* tree_depth 0. The record might be an extension of an existing
* record, and as such that needs to be handled. eg:
*
* Existing record in the extent map:
*
* cpos = 10, len = 10
* |---------|
*
* New Record:
*
* cpos = 10, len = 20
* |------------------|
*
* The passed record is the new on-disk record. The new_clusters value
* is how many clusters were added to the file. If the append is a
* contiguous append, the new_clusters has been added to
* rec->e_clusters. If the append is an entirely new extent, then
* rec->e_clusters is == new_clusters.
*/
int ocfs2_extent_map_append(struct inode *inode,
struct ocfs2_extent_rec *rec,
u32 new_clusters)
{
int ret;
struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
struct ocfs2_extent_map_entry *ent;
struct ocfs2_extent_rec *old;
BUG_ON(!new_clusters);
BUG_ON(le32_to_cpu(rec->e_clusters) < new_clusters);
if (em->em_clusters < OCFS2_I(inode)->ip_clusters) {
/*
* Size changed underneath us on disk. Drop any
* straddling records and update our idea of
* i_clusters
*/
ocfs2_extent_map_drop(inode, em->em_clusters - 1);
em->em_clusters = OCFS2_I(inode)->ip_clusters;
}
mlog_bug_on_msg((le32_to_cpu(rec->e_cpos) +
le32_to_cpu(rec->e_clusters)) !=
(em->em_clusters + new_clusters),
"Inode %llu:\n"
"rec->e_cpos = %u + rec->e_clusters = %u = %u\n"
"em->em_clusters = %u + new_clusters = %u = %u\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
le32_to_cpu(rec->e_cpos), le32_to_cpu(rec->e_clusters),
le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters),
em->em_clusters, new_clusters,
em->em_clusters + new_clusters);
em->em_clusters += new_clusters;
ret = -ENOENT;
if (le32_to_cpu(rec->e_clusters) > new_clusters) {
/* This is a contiguous append */
ent = ocfs2_extent_map_lookup(em, le32_to_cpu(rec->e_cpos), 1,
NULL, NULL);
if (ent) {
old = &ent->e_rec;
BUG_ON((le32_to_cpu(rec->e_cpos) +
le32_to_cpu(rec->e_clusters)) !=
(le32_to_cpu(old->e_cpos) +
le32_to_cpu(old->e_clusters) +
new_clusters));
if (ent->e_tree_depth == 0) {
BUG_ON(le32_to_cpu(old->e_cpos) !=
le32_to_cpu(rec->e_cpos));
BUG_ON(le64_to_cpu(old->e_blkno) !=
le64_to_cpu(rec->e_blkno));
ret = 0;
}
/*
* Let non-leafs fall through as -ENOENT to
* force insertion of the new leaf.
*/
le32_add_cpu(&old->e_clusters, new_clusters);
}
}
if (ret == -ENOENT)
ret = ocfs2_extent_map_insert(inode, rec, 0);
if (ret < 0)
mlog_errno(ret);
return ret;
}
static void user_dlm_unblock_lock(struct work_struct *work)
{
int new_level, status;
struct user_lock_res *lockres =
container_of(work, struct user_lock_res, l_work);
struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
mlog(0, "processing lockres %.*s\n", lockres->l_namelen,
lockres->l_name);
spin_lock(&lockres->l_lock);
mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
"Lockres %.*s, flags 0x%x\n",
lockres->l_namelen, lockres->l_name, lockres->l_flags);
/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
* set, we want user_ast clear it. */
lockres->l_flags &= ~USER_LOCK_QUEUED;
/* It's valid to get here and no longer be blocked - if we get
* several basts in a row, we might be queued by the first
* one, the unblock thread might run and clear the queued
* flag, and finally we might get another bast which re-queues
* us before our ast for the downconvert is called. */
if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
if (lockres->l_flags & USER_LOCK_BUSY) {
if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
lockres->l_flags |= USER_LOCK_IN_CANCEL;
spin_unlock(&lockres->l_lock);
status = dlmunlock(dlm,
&lockres->l_lksb,
LKM_CANCEL,
user_unlock_ast,
lockres);
if (status != DLM_NORMAL)
user_log_dlm_error("dlmunlock", status, lockres);
goto drop_ref;
}
/* If there are still incompat holders, we can exit safely
* without worrying about re-queueing this lock as that will
* happen on the last call to user_cluster_unlock. */
if ((lockres->l_blocking == LKM_EXMODE)
&& (lockres->l_ex_holders || lockres->l_ro_holders)) {
spin_unlock(&lockres->l_lock);
mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
lockres->l_ro_holders, lockres->l_ex_holders);
goto drop_ref;
}
if ((lockres->l_blocking == LKM_PRMODE)
&& lockres->l_ex_holders) {
spin_unlock(&lockres->l_lock);
mlog(0, "can't downconvert for pr: ex = %u\n",
lockres->l_ex_holders);
goto drop_ref;
}
/* yay, we can downconvert now. */
new_level = user_highest_compat_lock_level(lockres->l_blocking);
lockres->l_requested = new_level;
lockres->l_flags |= USER_LOCK_BUSY;
mlog(0, "Downconvert lock from %d to %d\n",
lockres->l_level, new_level);
spin_unlock(&lockres->l_lock);
/* need lock downconvert request now... */
status = dlmlock(dlm,
new_level,
&lockres->l_lksb,
LKM_CONVERT|LKM_VALBLK,
lockres->l_name,
lockres->l_namelen,
user_ast,
lockres,
user_bast);
if (status != DLM_NORMAL) {
user_log_dlm_error("dlmlock", status, lockres);
user_recover_from_dlm_error(lockres);
}
drop_ref:
user_dlm_drop_inode_ref(lockres);
}
static int ocfs2_read_locked_inode(struct inode *inode,
struct ocfs2_find_inode_args *args)
{
struct super_block *sb;
struct ocfs2_super *osb;
struct ocfs2_dinode *fe;
struct buffer_head *bh = NULL;
int status, can_lock;
u32 generation = 0;
status = -EINVAL;
if (inode == NULL || inode->i_sb == NULL) {
mlog(ML_ERROR, "bad inode\n");
return status;
}
sb = inode->i_sb;
osb = OCFS2_SB(sb);
if (!args) {
mlog(ML_ERROR, "bad inode args\n");
make_bad_inode(inode);
return status;
}
/*
* To improve performance of cold-cache inode stats, we take
* the cluster lock here if possible.
*
* Generally, OCFS2 never trusts the contents of an inode
* unless it's holding a cluster lock, so taking it here isn't
* a correctness issue as much as it is a performance
* improvement.
*
* There are three times when taking the lock is not a good idea:
*
* 1) During startup, before we have initialized the DLM.
*
* 2) If we are reading certain system files which never get
* cluster locks (local alloc, truncate log).
*
* 3) If the process doing the iget() is responsible for
* orphan dir recovery. We're holding the orphan dir lock and
* can get into a deadlock with another process on another
* node in ->delete_inode().
*
* #1 and #2 can be simply solved by never taking the lock
* here for system files (which are the only type we read
* during mount). It's a heavier approach, but our main
* concern is user-accessible files anyway.
*
* #3 works itself out because we'll eventually take the
* cluster lock before trusting anything anyway.
*/
can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
&& !ocfs2_mount_local(osb);
trace_ocfs2_read_locked_inode(
(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
/*
* To maintain backwards compatibility with older versions of
* ocfs2-tools, we still store the generation value for system
* files. The only ones that actually matter to userspace are
* the journals, but it's easier and inexpensive to just flag
* all system files similarly.
*/
if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
generation = osb->fs_generation;
ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
OCFS2_LOCK_TYPE_META,
generation, inode);
ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
OCFS2_LOCK_TYPE_OPEN,
0, inode);
if (can_lock) {
status = ocfs2_open_lock(inode);
if (status) {
make_bad_inode(inode);
mlog_errno(status);
return status;
}
status = ocfs2_inode_lock(inode, NULL, 0);
if (status) {
make_bad_inode(inode);
mlog_errno(status);
return status;
}
}
if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
status = ocfs2_try_open_lock(inode, 0);
if (status) {
make_bad_inode(inode);
return status;
}
}
if (can_lock) {
status = ocfs2_read_inode_block_full(inode, &bh,
OCFS2_BH_IGNORE_CACHE);
} else {
status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
/*
* If buffer is in jbd, then its checksum may not have been
* computed as yet.
*/
if (!status && !buffer_jbd(bh))
status = ocfs2_validate_inode_block(osb->sb, bh);
}
if (status < 0) {
mlog_errno(status);
goto bail;
}
status = -EINVAL;
fe = (struct ocfs2_dinode *) bh->b_data;
/*
* This is a code bug. Right now the caller needs to
* understand whether it is asking for a system file inode or
* not so the proper lock names can be built.
*/
mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
"Inode %llu: system file state is ambigous\n",
(unsigned long long)args->fi_blkno);
if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
S_ISBLK(le16_to_cpu(fe->i_mode)))
inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
ocfs2_populate_inode(inode, fe, 0);
BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
status = 0;
bail:
if (can_lock)
ocfs2_inode_unlock(inode, 0);
if (status < 0)
make_bad_inode(inode);
if (args && bh)
brelse(bh);
return status;
}
void ocfs2_clear_inode(struct inode *inode)
{
int status;
struct ocfs2_inode_info *oi = OCFS2_I(inode);
mlog_entry_void();
if (!inode)
goto bail;
mlog(0, "Clearing inode: %llu, nlink = %u\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
"Inode=%lu\n", inode->i_ino);
/* For remove delete_inode vote, we hold open lock before,
* now it is time to unlock PR and EX open locks. */
ocfs2_open_unlock(inode);
/* Do these before all the other work so that we don't bounce
* the vote thread while waiting to destroy the locks. */
ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
/* We very well may get a clear_inode before all an inodes
* metadata has hit disk. Of course, we can't drop any cluster
* locks until the journal has finished with it. The only
* exception here are successfully wiped inodes - their
* metadata can now be considered to be part of the system
* inodes from which it came. */
if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
ocfs2_checkpoint_inode(inode);
mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
"Clear inode of %llu, inode has io markers\n",
(unsigned long long)oi->ip_blkno);
ocfs2_extent_map_trunc(inode, 0);
status = ocfs2_drop_inode_locks(inode);
if (status < 0)
mlog_errno(status);
ocfs2_lock_res_free(&oi->ip_rw_lockres);
ocfs2_lock_res_free(&oi->ip_meta_lockres);
ocfs2_lock_res_free(&oi->ip_data_lockres);
ocfs2_lock_res_free(&oi->ip_open_lockres);
ocfs2_metadata_cache_purge(inode);
mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
"Clear inode of %llu, inode has %u cache items\n",
(unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
"Clear inode of %llu, inode has a bad flag\n",
(unsigned long long)oi->ip_blkno);
mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
"Clear inode of %llu, inode is locked\n",
(unsigned long long)oi->ip_blkno);
mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
"Clear inode of %llu, io_mutex is locked\n",
(unsigned long long)oi->ip_blkno);
mutex_unlock(&oi->ip_io_mutex);
/*
* down_trylock() returns 0, down_write_trylock() returns 1
* kernel 1, world 0
*/
mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
"Clear inode of %llu, alloc_sem is locked\n",
(unsigned long long)oi->ip_blkno);
up_write(&oi->ip_alloc_sem);
mlog_bug_on_msg(oi->ip_open_count,
"Clear inode of %llu has open count %d\n",
(unsigned long long)oi->ip_blkno, oi->ip_open_count);
/* Clear all other flags. */
oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
oi->ip_created_trans = 0;
oi->ip_last_trans = 0;
oi->ip_dir_start_lookup = 0;
oi->ip_blkno = 0ULL;
bail:
mlog_exit_void();
}
int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
void **ret_data)
{
int ret;
unsigned int locklen;
struct dlm_ctxt *dlm = data;
struct dlm_lock_resource *res = NULL;
struct dlm_lock *lock = NULL;
struct dlm_proxy_ast *past = (struct dlm_proxy_ast *) msg->buf;
char *name;
struct list_head *iter, *head=NULL;
u64 cookie;
u32 flags;
u8 node;
if (!dlm_grab(dlm)) {
dlm_error(DLM_REJECTED);
return DLM_REJECTED;
}
mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
"Domain %s not fully joined!\n", dlm->name);
name = past->name;
locklen = past->namelen;
cookie = past->cookie;
flags = be32_to_cpu(past->flags);
node = past->node_idx;
if (locklen > DLM_LOCKID_NAME_MAX) {
ret = DLM_IVBUFLEN;
mlog(ML_ERROR, "Invalid name length (%d) in proxy ast "
"handler!\n", locklen);
goto leave;
}
if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
(LKM_PUT_LVB|LKM_GET_LVB)) {
mlog(ML_ERROR, "Both PUT and GET lvb specified, (0x%x)\n",
flags);
ret = DLM_BADARGS;
goto leave;
}
mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
(flags & LKM_GET_LVB ? "get lvb" : "none"));
mlog(0, "type=%d, blocked_type=%d\n", past->type, past->blocked_type);
if (past->type != DLM_AST &&
past->type != DLM_BAST) {
mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu"
"name=%.*s, node=%u\n", past->type,
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
locklen, name, node);
ret = DLM_IVLOCKID;
goto leave;
}
res = dlm_lookup_lockres(dlm, name, locklen);
if (!res) {
mlog(0, "Got %sast for unknown lockres! cookie=%u:%llu, "
"name=%.*s, node=%u\n", (past->type == DLM_AST ? "" : "b"),
dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
locklen, name, node);
ret = DLM_IVLOCKID;
goto leave;
}
/* cannot get a proxy ast message if this node owns it */
BUG_ON(res->owner == dlm->node_num);
mlog(0, "lockres %.*s\n", res->lockname.len, res->lockname.name);
spin_lock(&res->spinlock);
if (res->state & DLM_LOCK_RES_RECOVERING) {
mlog(0, "Responding with DLM_RECOVERING!\n");
ret = DLM_RECOVERING;
goto unlock_out;
}
if (res->state & DLM_LOCK_RES_MIGRATING) {
mlog(0, "Responding with DLM_MIGRATING!\n");
ret = DLM_MIGRATING;
goto unlock_out;
}
/* try convert queue for both ast/bast */
head = &res->converting;
lock = NULL;
list_for_each(iter, head) {
lock = list_entry (iter, struct dlm_lock, list);
if (lock->ml.cookie == cookie)
goto do_ast;
}
static int ocfs2_get_block(struct inode *inode, sector_t iblock,
struct buffer_head *bh_result, int create)
{
int err = 0;
unsigned int ext_flags;
u64 p_blkno, past_eof;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
(unsigned long long)iblock, bh_result, create);
if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
inode, inode->i_ino);
if (S_ISLNK(inode->i_mode)) {
/* this always does I/O for some reason. */
err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
goto bail;
}
err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL,
&ext_flags);
if (err) {
mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
"%llu, NULL)\n", err, inode, (unsigned long long)iblock,
(unsigned long long)p_blkno);
goto bail;
}
/*
* ocfs2 never allocates in this function - the only time we
* need to use BH_New is when we're extending i_size on a file
* system which doesn't support holes, in which case BH_New
* allows block_prepare_write() to zero.
*/
mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb),
"ino %lu, iblock %llu\n", inode->i_ino,
(unsigned long long)iblock);
/* Treat the unwritten extent as a hole for zeroing purposes. */
if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
map_bh(bh_result, inode->i_sb, p_blkno);
if (!ocfs2_sparse_alloc(osb)) {
if (p_blkno == 0) {
err = -EIO;
mlog(ML_ERROR,
"iblock = %llu p_blkno = %llu blkno=(%llu)\n",
(unsigned long long)iblock,
(unsigned long long)p_blkno,
(unsigned long long)OCFS2_I(inode)->ip_blkno);
mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
dump_stack();
}
past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
(unsigned long long)past_eof);
if (create && (iblock >= past_eof))
set_buffer_new(bh_result);
}
bail:
if (err < 0)
err = -EIO;
mlog_exit(err);
return err;
}
static void user_dlm_unblock_lock(struct work_struct *work)
{
int new_level, status;
struct user_lock_res *lockres =
container_of(work, struct user_lock_res, l_work);
struct ocfs2_cluster_connection *conn =
cluster_connection_from_user_lockres(lockres);
mlog(0, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
spin_lock(&lockres->l_lock);
mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
"Lockres %.*s, flags 0x%x\n",
lockres->l_namelen, lockres->l_name, lockres->l_flags);
/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
* set, we want user_ast clear it. */
lockres->l_flags &= ~USER_LOCK_QUEUED;
/* It's valid to get here and no longer be blocked - if we get
* several basts in a row, we might be queued by the first
* one, the unblock thread might run and clear the queued
* flag, and finally we might get another bast which re-queues
* us before our ast for the downconvert is called. */
if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
mlog(ML_BASTS, "lockres %.*s USER_LOCK_BLOCKED\n",
lockres->l_namelen, lockres->l_name);
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_TEARDOWN\n",
lockres->l_namelen, lockres->l_name);
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
if (lockres->l_flags & USER_LOCK_BUSY) {
if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_CANCEL\n",
lockres->l_namelen, lockres->l_name);
spin_unlock(&lockres->l_lock);
goto drop_ref;
}
lockres->l_flags |= USER_LOCK_IN_CANCEL;
spin_unlock(&lockres->l_lock);
status = ocfs2_dlm_unlock(conn, &lockres->l_lksb,
DLM_LKF_CANCEL);
if (status)
user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
goto drop_ref;
}
/* If there are still incompat holders, we can exit safely
* without worrying about re-queueing this lock as that will
* happen on the last call to user_cluster_unlock. */
if ((lockres->l_blocking == DLM_LOCK_EX)
&& (lockres->l_ex_holders || lockres->l_ro_holders)) {
spin_unlock(&lockres->l_lock);
mlog(ML_BASTS, "lockres %.*s, EX/PR Holders %u,%u\n",
lockres->l_namelen, lockres->l_name,
lockres->l_ex_holders, lockres->l_ro_holders);
goto drop_ref;
}
if ((lockres->l_blocking == DLM_LOCK_PR)
&& lockres->l_ex_holders) {
spin_unlock(&lockres->l_lock);
mlog(ML_BASTS, "lockres %.*s, EX Holders %u\n",
lockres->l_namelen, lockres->l_name,
lockres->l_ex_holders);
goto drop_ref;
}
/* yay, we can downconvert now. */
new_level = user_highest_compat_lock_level(lockres->l_blocking);
lockres->l_requested = new_level;
lockres->l_flags |= USER_LOCK_BUSY;
mlog(ML_BASTS, "lockres %.*s, downconvert %d => %d\n",
lockres->l_namelen, lockres->l_name, lockres->l_level, new_level);
spin_unlock(&lockres->l_lock);
/* need lock downconvert request now... */
status = ocfs2_dlm_lock(conn, new_level, &lockres->l_lksb,
DLM_LKF_CONVERT|DLM_LKF_VALBLK,
lockres->l_name,
lockres->l_namelen);
if (status) {
user_log_dlm_error("ocfs2_dlm_lock", status, lockres);
user_recover_from_dlm_error(lockres);
}
drop_ref:
user_dlm_drop_inode_ref(lockres);
}
static int ocfs2_truncate_file(struct inode *inode,
struct buffer_head *di_bh,
u64 new_i_size)
{
int status = 0;
struct ocfs2_dinode *fe = NULL;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct ocfs2_truncate_context *tc = NULL;
mlog_entry("(inode = %llu, new_i_size = %llu\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
(unsigned long long)new_i_size);
unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
truncate_inode_pages(inode->i_mapping, new_i_size);
fe = (struct ocfs2_dinode *) di_bh->b_data;
if (!OCFS2_IS_VALID_DINODE(fe)) {
OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
status = -EIO;
goto bail;
}
mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
"Inode %llu, inode i_size = %lld != di "
"i_size = %llu, i_flags = 0x%x\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
i_size_read(inode),
(unsigned long long)le64_to_cpu(fe->i_size),
le32_to_cpu(fe->i_flags));
if (new_i_size > le64_to_cpu(fe->i_size)) {
mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
(unsigned long long)le64_to_cpu(fe->i_size),
(unsigned long long)new_i_size);
status = -EINVAL;
mlog_errno(status);
goto bail;
}
mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
(unsigned long long)le64_to_cpu(fe->i_blkno),
(unsigned long long)le64_to_cpu(fe->i_size),
(unsigned long long)new_i_size);
/* lets handle the simple truncate cases before doing any more
* cluster locking. */
if (new_i_size == le64_to_cpu(fe->i_size))
goto bail;
/* This forces other nodes to sync and drop their pages. Do
* this even if we have a truncate without allocation change -
* ocfs2 cluster sizes can be much greater than page size, so
* we have to truncate them anyway. */
status = ocfs2_data_lock(inode, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
/* alright, we're going to need to do a full blown alloc size
* change. Orphan the inode so that recovery can complete the
* truncate if necessary. This does the task of marking
* i_size. */
status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
/* TODO: orphan dir cleanup here. */
bail_unlock_data:
ocfs2_data_unlock(inode, 1);
bail:
mlog_exit(status);
return status;
}