#include "lov_cl_internal.h"
-/** \addtogroup lov lov @{ */
+/** \addtogroup lov
+ * @{
+ */
/*****************************************************************************
*
EXIT;
}
-static void lovsub_lock_state_one(const struct lu_env *env,
- const struct lovsub_lock *lovsub,
- struct lov_lock *lov)
+static int lovsub_lock_state_one(const struct lu_env *env,
+ const struct lovsub_lock *lovsub,
+ struct lov_lock *lov)
{
- struct cl_lock *parent;
- const struct cl_lock *child;
+ struct cl_lock *parent;
+ struct cl_lock *child;
+ int restart = 0;
ENTRY;
parent = lov->lls_cl.cls_lock;
if (lovsub->lss_active != parent) {
lovsub_parent_lock(env, lov);
- if (child->cll_error != 0)
+ if (child->cll_error != 0 && parent->cll_error == 0) {
+ /*
+ * This is a deadlock case:
+ * cl_lock_error(for the parent lock)
+ * -> cl_lock_delete
+ * -> lov_lock_delete
+ * -> cl_lock_enclosure
+ * -> cl_lock_mutex_try(for the child lock)
+ */
+ cl_lock_mutex_put(env, child);
cl_lock_error(env, parent, child->cll_error);
- else
+ restart = 1;
+ } else {
cl_lock_signal(env, parent);
+ }
lovsub_parent_unlock(env, lov);
}
- EXIT;
+ RETURN(restart);
}
/**
{
struct lovsub_lock *sub = cl2lovsub_lock(slice);
struct lov_lock_link *scan;
- struct lov_lock_link *temp;
+ int restart = 0;
LASSERT(cl_lock_is_mutexed(slice->cls_lock));
ENTRY;
- /*
- * Use _safe() version, because
- *
- * lovsub_lock_state_one()
- * ->cl_lock_error()
- * ->cl_lock_delete()
- * ->lov_lock_delete()
- *
- * can unlink parent from the parent list.
- */
- list_for_each_entry_safe(scan, temp, &sub->lss_parents, lll_list)
- lovsub_lock_state_one(env, sub, scan->lll_super);
+ do {
+ restart = 0;
+ list_for_each_entry(scan, &sub->lss_parents, lll_list) {
+ restart = lovsub_lock_state_one(env, sub,
+ scan->lll_super);
+ if (restart) {
+ cl_lock_mutex_get(env, slice->cls_lock);
+ break;
+ }
+ }
+ } while(restart);
EXIT;
}
pd->cld_obj = parent_descr->cld_obj;
pd->cld_mode = parent_descr->cld_mode;
+ pd->cld_gid = parent_descr->cld_gid;
lovsub_lock_descr_map(d, subobj->lso_super, subobj->lso_index, pd);
lov->lls_sub[idx].sub_got = *d;
/*
}
/**
- * An implementation of cl_lock_operations::clo_delete() method. This is
- * invoked in "bottom-to-top" delete, when lock destruction starts from the
- * sub-lock (e.g, as a result of ldlm lock LRU policy).
+ * A helper function for lovsub_lock_delete() that deals with a given parent
+ * top-lock.
*/
-static void lovsub_lock_delete(const struct lu_env *env,
- const struct cl_lock_slice *slice)
+static int lovsub_lock_delete_one(const struct lu_env *env,
+ struct cl_lock *child, struct lov_lock *lov)
{
- struct lovsub_lock *sub = cl2lovsub_lock(slice);
- struct lov_lock *lov;
struct cl_lock *parent;
- struct lov_lock_link *scan;
- struct lov_lock_link *temp;
- struct lov_lock_sub *subdata;
-
- LASSERT(cl_lock_is_mutexed(slice->cls_lock));
+ int result;
ENTRY;
- list_for_each_entry_safe(scan, temp, &sub->lss_parents, lll_list) {
- lov = scan->lll_super;
- subdata = &lov->lls_sub[scan->lll_idx];
- parent = lov->lls_cl.cls_lock;
- lovsub_parent_lock(env, lov);
- subdata->sub_got = subdata->sub_descr;
- lov_lock_unlink(env, scan, sub);
- CDEBUG(D_DLMTRACE, "%p %p %i %i\n", parent, sub,
- lov->lls_nr_filled, parent->cll_state);
- switch (parent->cll_state) {
- case CLS_NEW:
- case CLS_QUEUING:
- case CLS_ENQUEUED:
- case CLS_FREEING:
- cl_lock_signal(env, parent);
- break;
- case CLS_UNLOCKING:
- /*
- * Here lies a problem: a sub-lock is canceled while
- * top-lock is being unlocked. Top-lock cannot be
- * moved into CLS_NEW state, because unlocking has to
- * succeed eventually by placing lock into CLS_CACHED
- * (or failing it), see cl_unuse_try(). Nor can
- * top-lock be left in CLS_CACHED state, because lov
- * maintains an invariant that all sub-locks exist in
- * CLS_CACHED (this allows cached top-lock to be
- * reused immediately). Nor can we wait for top-lock
- * state to change, because this can be synchronous to
- * the current thread.
+ parent = lov->lls_cl.cls_lock;
+ result = 0;
+
+ switch (parent->cll_state) {
+ case CLS_NEW:
+ case CLS_QUEUING:
+ case CLS_ENQUEUED:
+ case CLS_FREEING:
+ cl_lock_signal(env, parent);
+ break;
+ case CLS_INTRANSIT:
+ /*
+ * Here lies a problem: a sub-lock is canceled while top-lock
+ * is being unlocked. Top-lock cannot be moved into CLS_NEW
+ * state, because unlocking has to succeed eventually by
+ * placing lock into CLS_CACHED (or failing it), see
+ * cl_unuse_try(). Nor can top-lock be left in CLS_CACHED
+ * state, because lov maintains an invariant that all
+ * sub-locks exist in CLS_CACHED (this allows cached top-lock
+ * to be reused immediately). Nor can we wait for top-lock
+ * state to change, because this can be synchronous to the
+ * current thread.
+ *
+ * We know for sure that lov_lock_unuse() will be called at
+ * least one more time to finish un-using, so leave a mark on
+ * the top-lock, that will be seen by the next call to
+ * lov_lock_unuse().
+ */
+ if (cl_lock_is_intransit(parent))
+ lov->lls_cancel_race = 1;
+ break;
+ case CLS_CACHED:
+ /*
+ * if a sub-lock is canceled move its top-lock into CLS_NEW
+ * state to preserve an invariant that a top-lock in
+ * CLS_CACHED is immediately ready for re-use (i.e., has all
+ * sub-locks), and so that next attempt to re-use the top-lock
+ * enqueues missing sub-lock.
+ */
+ cl_lock_state_set(env, parent, CLS_NEW);
+ /*
+ * if last sub-lock is canceled, destroy the top-lock (which
+ * is now `empty') proactively.
+ */
+ if (lov->lls_nr_filled == 0) {
+ /* ... but unfortunately, this cannot be done easily,
+ * as cancellation of a top-lock might acquire mutices
+ * of its other sub-locks, violating lock ordering,
+ * see cl_lock_{cancel,delete}() preconditions.
+ *
+ * To work around this, the mutex of this sub-lock is
+ * released, top-lock is destroyed, and sub-lock mutex
+ * acquired again. The list of parents has to be
+ * re-scanned from the beginning after this.
+ *
+ * Only do this if no mutices other than on @child and
+ * @parent are held by the current thread.
+ *
+ * TODO: The lock modal here is too complex, because
+ * the lock may be canceled and deleted by voluntarily:
+ * cl_lock_request
+ * -> osc_lock_enqueue_wait
+ * -> osc_lock_cancel_wait
+ * -> cl_lock_delete
+ * -> lovsub_lock_delete
+ * -> cl_lock_cancel/delete
+ * -> ...
*
- * We know for sure that lov_lock_unuse() will be
- * called at least one more time to finish un-using,
- * so leave a mark on the top-lock, that will be seen
- * by the next call to lov_lock_unuse().
+ * The better choice is to spawn a kernel thread for
+ * this purpose. -jay
*/
- lov->lls_unuse_race = 1;
- break;
- case CLS_CACHED:
- cl_lock_state_set(env, parent, CLS_NEW);
- if (lov->lls_nr_filled == 0) {
+ if (cl_lock_nr_mutexed(env) == 2) {
+ cl_lock_mutex_put(env, child);
cl_lock_cancel(env, parent);
cl_lock_delete(env, parent);
- cl_lock_signal(env, parent);
+ result = 1;
}
- break;
- case CLS_HELD:
- default:
- CERROR("Impossible state: %i\n", parent->cll_state);
- LBUG();
}
- lovsub_parent_unlock(env, lov);
+ break;
+ case CLS_HELD:
+ default:
+ CERROR("Impossible state: %i\n", parent->cll_state);
+ LBUG();
}
+
+ RETURN(result);
+}
+
+/**
+ * An implementation of cl_lock_operations::clo_delete() method. This is
+ * invoked in "bottom-to-top" delete, when lock destruction starts from the
+ * sub-lock (e.g, as a result of ldlm lock LRU policy).
+ */
+static void lovsub_lock_delete(const struct lu_env *env,
+ const struct cl_lock_slice *slice)
+{
+ struct cl_lock *child = slice->cls_lock;
+ struct lovsub_lock *sub = cl2lovsub_lock(slice);
+ int restart;
+
+ LASSERT(cl_lock_is_mutexed(child));
+
+ ENTRY;
+ /*
+ * Destruction of a sub-lock might take multiple iterations, because
+ * when the last sub-lock of a given top-lock is deleted, top-lock is
+ * canceled proactively, and this requires to release sub-lock
+ * mutex. Once sub-lock mutex has been released, list of its parents
+ * has to be re-scanned from the beginning.
+ */
+ do {
+ struct lov_lock *lov;
+ struct lov_lock_link *scan;
+ struct lov_lock_link *temp;
+ struct lov_lock_sub *subdata;
+
+ restart = 0;
+ list_for_each_entry_safe(scan, temp,
+ &sub->lss_parents, lll_list) {
+ lov = scan->lll_super;
+ subdata = &lov->lls_sub[scan->lll_idx];
+ lovsub_parent_lock(env, lov);
+ subdata->sub_got = subdata->sub_descr;
+ lov_lock_unlink(env, scan, sub);
+ restart = lovsub_lock_delete_one(env, child, lov);
+ lovsub_parent_unlock(env, lov);
+
+ if (restart) {
+ cl_lock_mutex_get(env, child);
+ break;
+ }
+ }
+ } while (restart);
EXIT;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff