dlm: fix sleep in atomic context
authorAlexander Aring <aahringo@redhat.com>
Wed, 17 Apr 2024 19:13:22 +0000 (15:13 -0400)
committerDavid Teigland <teigland@redhat.com>
Wed, 17 Apr 2024 19:20:04 +0000 (14:20 -0500)
This patch changes the orphans mutex to a spinlock since commit
c288745f1d4a ("dlm: avoid blocking receive at the end of recovery") is
using a rwlock_t to lock the DLM message receive path and do_purge() can
be called while this lock is held that forbids to sleep.

We need to use spin_lock_bh() because also a user context that calls
dlm_user_purge() can call do_purge() and since commit 92d59adfaf71
("dlm: do message processing in softirq context") the DLM message
receive path is done under softirq context.

Fixes: c288745f1d4a ("dlm: avoid blocking receive at the end of recovery")
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/gfs2/9ad928eb-2ece-4ad9-a79c-d2bce228e4bc@moroto.mountain/
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
fs/dlm/dlm_internal.h
fs/dlm/lock.c
fs/dlm/lockspace.c

index 19e57cbd5b13470e811faf4f8e935aab7ad3e855..9085ba3b2f20152aa7d2502d3794771437f9becf 100644 (file)
@@ -602,7 +602,7 @@ struct dlm_ls {
        spinlock_t              ls_waiters_lock;
        struct list_head        ls_waiters;     /* lkbs needing a reply */
 
-       struct mutex            ls_orphans_mutex;
+       spinlock_t              ls_orphans_lock;
        struct list_head        ls_orphans;
 
        spinlock_t              ls_new_rsb_spin;
index bbbc9593a64e5603e07dba4372b355cff20c66bd..f103b8c3059289dd1880a86eece65bc206586cc2 100644 (file)
@@ -5880,7 +5880,7 @@ int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
        int found_other_mode = 0;
        int rv = 0;
 
-       mutex_lock(&ls->ls_orphans_mutex);
+       spin_lock_bh(&ls->ls_orphans_lock);
        list_for_each_entry(iter, &ls->ls_orphans, lkb_ownqueue) {
                if (iter->lkb_resource->res_length != namelen)
                        continue;
@@ -5897,7 +5897,7 @@ int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
                *lkid = iter->lkb_id;
                break;
        }
-       mutex_unlock(&ls->ls_orphans_mutex);
+       spin_unlock_bh(&ls->ls_orphans_lock);
 
        if (!lkb && found_other_mode) {
                rv = -EAGAIN;
@@ -6089,9 +6089,9 @@ static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
        int error;
 
        hold_lkb(lkb); /* reference for the ls_orphans list */
-       mutex_lock(&ls->ls_orphans_mutex);
+       spin_lock_bh(&ls->ls_orphans_lock);
        list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
-       mutex_unlock(&ls->ls_orphans_mutex);
+       spin_unlock_bh(&ls->ls_orphans_lock);
 
        set_unlock_args(0, lkb->lkb_ua, &args);
 
@@ -6241,7 +6241,7 @@ static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
 {
        struct dlm_lkb *lkb, *safe;
 
-       mutex_lock(&ls->ls_orphans_mutex);
+       spin_lock_bh(&ls->ls_orphans_lock);
        list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
                if (pid && lkb->lkb_ownpid != pid)
                        continue;
@@ -6249,7 +6249,7 @@ static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
                list_del_init(&lkb->lkb_ownqueue);
                dlm_put_lkb(lkb);
        }
-       mutex_unlock(&ls->ls_orphans_mutex);
+       spin_unlock_bh(&ls->ls_orphans_lock);
 }
 
 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
index 5ce26882159e53b9d30b7283b1d58998c59a41a3..ed23787271b11af935202bbbfb9ab25d86ee2db0 100644 (file)
@@ -436,7 +436,7 @@ static int new_lockspace(const char *name, const char *cluster,
        INIT_LIST_HEAD(&ls->ls_waiters);
        spin_lock_init(&ls->ls_waiters_lock);
        INIT_LIST_HEAD(&ls->ls_orphans);
-       mutex_init(&ls->ls_orphans_mutex);
+       spin_lock_init(&ls->ls_orphans_lock);
 
        INIT_LIST_HEAD(&ls->ls_new_rsb);
        spin_lock_init(&ls->ls_new_rsb_spin);