staging: lustre: ldlm: tidy list walking in ldlm_flock()
authorNeilBrown <neilb@suse.com>
Mon, 23 Oct 2017 00:53:49 +0000 (11:53 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 3 Nov 2017 14:00:34 +0000 (15:00 +0100)
Use list_for_each_entry variants to
avoid the explicit list_entry() calls.
This allows us to use list_for_each_entry_safe_from()
instread of adding a local list-walking macro.

Also improve some comments so that it is more obvious
that the locks are sorted per-owner and that we need
to find the insertion point.

Signed-off-by: NeilBrown <neilb@suse.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/ldlm/ldlm_flock.c

index b911352c5347a2295ed9ed1431e8a530c7f4873e..7b90f8ef89d794fa356145ba04afbddd3b0cc81c 100644 (file)
 #include <linux/list.h>
 #include "ldlm_internal.h"
 
-/**
- * list_for_remaining_safe - iterate over the remaining entries in a list
- *           and safeguard against removal of a list entry.
- * \param pos   the &struct list_head to use as a loop counter. pos MUST
- *           have been initialized prior to using it in this macro.
- * \param n     another &struct list_head to use as temporary storage
- * \param head  the head for your list.
- */
-#define list_for_remaining_safe(pos, n, head) \
-       for (n = pos->next; pos != (head); pos = n, n = pos->next)
-
 static inline int
 ldlm_same_flock_owner(struct ldlm_lock *lock, struct ldlm_lock *new)
 {
@@ -125,8 +114,8 @@ static int ldlm_process_flock_lock(struct ldlm_lock *req)
 {
        struct ldlm_resource *res = req->l_resource;
        struct ldlm_namespace *ns = ldlm_res_to_ns(res);
-       struct list_head *tmp;
-       struct list_head *ownlocks = NULL;
+       struct ldlm_lock *tmp;
+       struct ldlm_lock *ownlocks = NULL;
        struct ldlm_lock *lock = NULL;
        struct ldlm_lock *new = req;
        struct ldlm_lock *new2 = NULL;
@@ -151,23 +140,23 @@ reprocess:
        /* This loop determines where this processes locks start
         * in the resource lr_granted list.
         */
-       list_for_each(tmp, &res->lr_granted) {
-               lock = list_entry(tmp, struct ldlm_lock,
-                                 l_res_link);
+       list_for_each_entry(lock, &res->lr_granted, l_res_link) {
                if (ldlm_same_flock_owner(lock, req)) {
-                       ownlocks = tmp;
+                       ownlocks = lock;
                        break;
                }
        }
 
-       /* Scan the locks owned by this process that overlap this request.
+       /* Scan the locks owned by this process to find the insertion point
+        * (as locks are ordered), and to handle overlaps.
         * We may have to merge or split existing locks.
         */
-       if (!ownlocks)
-               ownlocks = &res->lr_granted;
-
-       list_for_remaining_safe(ownlocks, tmp, &res->lr_granted) {
-               lock = list_entry(ownlocks, struct ldlm_lock, l_res_link);
+       if (ownlocks)
+               lock = ownlocks;
+       else
+               lock = list_entry(&res->lr_granted,
+                                 struct ldlm_lock, l_res_link);
+       list_for_each_entry_safe_from(lock, tmp, &res->lr_granted, l_res_link) {
 
                if (!ldlm_same_flock_owner(lock, new))
                        break;
@@ -295,7 +284,7 @@ reprocess:
                                                 lock->l_granted_mode);
 
                /* insert new2 at lock */
-               ldlm_resource_add_lock(res, ownlocks, new2);
+               ldlm_resource_add_lock(res, &lock->l_res_link, new2);
                LDLM_LOCK_RELEASE(new2);
                break;
        }
@@ -309,8 +298,12 @@ reprocess:
 
        if (!added) {
                list_del_init(&req->l_res_link);
-               /* insert new lock before ownlocks in list. */
-               ldlm_resource_add_lock(res, ownlocks, req);
+               /* insert new lock before "lock", which might be the
+                * next lock for this owner, or might be the first
+                * lock for the next owner, or might not be a lock at
+                * all, but instead points at the head of the list
+                */
+               ldlm_resource_add_lock(res, &lock->l_res_link, req);
        }
 
        /* In case we're reprocessing the requested lock we can't destroy