Skip to contentSkip to conversation

dlm: remove allocation parameter in msg allocation

PATCH Superseded Alexander Aring 1 participant 11 messages Oct 25, 2023 Distributed lock manager (DLM)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Alexander Aring

author maintainer
#0 lore raw

PATCH 1/10 dlm: remove allocation parameter in msg allocation

This patch removes the context parameter for message allocations and replace it by always do GFP_ATOMIC. We are preparing to process dlm message in softirq context therefore it's necessary to switch to GFP_ATOMIC allocation as we cannot sleep in this context. To simplify the code overall we just drop the allocation flag and have GFP_ATOMIC hardcoded when calling the allocation function.

---
 fs/dlm/lock.c     | 31 ++++++++++++-------------------
 fs/dlm/lowcomms.c | 16 +++++++---------
 fs/dlm/lowcomms.h |  5 ++---
 fs/dlm/memory.c   |  8 ++++----
 fs/dlm/memory.h   |  4 ++--
 fs/dlm/midcomms.c | 24 ++++++++++--------------
 fs/dlm/midcomms.h |  3 +--
 fs/dlm/rcom.c     |  7 +++----
 8 files changed, 41 insertions(+), 57 deletions(-)
fs/dlm/lock.c +12 -19
index 652c51fbbf76..44ea65a73f3b 100644
@@ -3329,8 +3329,7 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
 static int _create_message(struct dlm_ls *ls, int mb_len,
 			   int to_nodeid, int mstype,
 			   struct dlm_message **ms_ret,
-			   struct dlm_mhandle **mh_ret,
-			   gfp_t allocation)
+			   struct dlm_mhandle **mh_ret)
 {
 	struct dlm_message *ms;
 	struct dlm_mhandle *mh;
@@ -3340,7 +3339,7 @@ static int _create_message(struct dlm_ls *ls, int mb_len,
 	   pass into midcomms_commit and a message buffer (mb) that we
 	   write our data into */
 
-	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, allocation, &mb);
+	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, &mb);
 	if (!mh)
 		return -ENOBUFS;
 
@@ -3362,8 +3361,7 @@ static int _create_message(struct dlm_ls *ls, int mb_len,
 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 			  int to_nodeid, int mstype,
 			  struct dlm_message **ms_ret,
-			  struct dlm_mhandle **mh_ret,
-			  gfp_t allocation)
+			  struct dlm_mhandle **mh_ret)
 {
 	int mb_len = sizeof(struct dlm_message);
 
@@ -3384,7 +3382,7 @@ static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 	}
 
 	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
-			       ms_ret, mh_ret, allocation);
+			       ms_ret, mh_ret);
 }
 
 /* further lowcomms enhancements or alternate implementations may make
@@ -3453,7 +3451,7 @@ static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
 	if (error)
 		return error;
 
-	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
 	if (error)
 		goto fail;
 
@@ -3513,8 +3511,7 @@ static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3535,8 +3532,7 @@ static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3561,8 +3557,7 @@ static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	if (error)
 		return error;
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
 	if (error)
 		goto fail;
 
@@ -3586,8 +3581,7 @@ static int send_remove(struct dlm_rsb *r)
 
 	to_nodeid = dlm_dir_nodeid(r);
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh,
-			       GFP_ATOMIC);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3608,7 +3602,7 @@ static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3650,8 +3644,7 @@ static int send_lookup_reply(struct dlm_ls *ls,
 	struct dlm_mhandle *mh;
 	int error, nodeid = le32_to_cpu(ms_in->m_header.h_nodeid);
 
-	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -6063,7 +6056,7 @@ static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
 	int error;
 
 	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
-				DLM_MSG_PURGE, &ms, &mh, GFP_NOFS);
+				DLM_MSG_PURGE, &ms, &mh);
 	if (error)
 		return error;
 	ms->m_nodeid = cpu_to_le32(nodeid);
fs/dlm/lowcomms.c +7 -9
index 67f8dd8a05ef..db71982d709d 100644
@@ -1229,14 +1229,13 @@ static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
 };
 
 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
-						gfp_t allocation, char **ppc,
-						void (*cb)(void *data),
+						char **ppc, void (*cb)(void *data),
 						void *data)
 {
 	struct writequeue_entry *e;
 	struct dlm_msg *msg;
 
-	msg = dlm_allocate_msg(allocation);
+	msg = dlm_allocate_msg();
 	if (!msg)
 		return NULL;
 
@@ -1261,9 +1260,8 @@ static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
  * dlm_lowcomms_commit_msg which is a must call if success
  */
 #ifndef __CHECKER__
-struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
-				     char **ppc, void (*cb)(void *data),
-				     void *data)
+struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
+				     void (*cb)(void *data), void *data)
 {
 	struct connection *con;
 	struct dlm_msg *msg;
@@ -1284,7 +1282,7 @@ struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
 		return NULL;
 	}
 
-	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data);
+	msg = dlm_lowcomms_new_msg_con(con, len, ppc, cb, data);
 	if (!msg) {
 		srcu_read_unlock(&connections_srcu, idx);
 		return NULL;
@@ -1348,8 +1346,8 @@ int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
 	if (msg->retransmit)
 		return 1;
 
-	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
-					      GFP_ATOMIC, &ppc, NULL, NULL);
+	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, &ppc,
+					      NULL, NULL);
 	if (!msg_resend)
 		return -ENOMEM;
 
fs/dlm/lowcomms.h +2 -3
index 3e8dca66183b..8deb16f8f620 100644
@@ -39,9 +39,8 @@ void dlm_lowcomms_stop(void);
 void dlm_lowcomms_init(void);
 void dlm_lowcomms_exit(void);
 int dlm_lowcomms_close(int nodeid);
-struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
-				     char **ppc, void (*cb)(void *data),
-				     void *data);
+struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
+				     void (*cb)(void *data), void *data);
 void dlm_lowcomms_commit_msg(struct dlm_msg *msg);
 void dlm_lowcomms_put_msg(struct dlm_msg *msg);
 int dlm_lowcomms_resend_msg(struct dlm_msg *msg);
fs/dlm/memory.c +4 -4
index 64f212a066cf..c0c1a83f6381 100644
@@ -134,9 +134,9 @@ void dlm_free_lkb(struct dlm_lkb *lkb)
 	kmem_cache_free(lkb_cache, lkb);
 }
 
-struct dlm_mhandle *dlm_allocate_mhandle(gfp_t allocation)
+struct dlm_mhandle *dlm_allocate_mhandle(void)
 {
-	return kmem_cache_alloc(mhandle_cache, allocation);
+	return kmem_cache_alloc(mhandle_cache, GFP_ATOMIC);
 }
 
 void dlm_free_mhandle(struct dlm_mhandle *mhandle)
@@ -154,9 +154,9 @@ void dlm_free_writequeue(struct writequeue_entry *writequeue)
 	kmem_cache_free(writequeue_cache, writequeue);
 }
 
-struct dlm_msg *dlm_allocate_msg(gfp_t allocation)
+struct dlm_msg *dlm_allocate_msg(void)
 {
-	return kmem_cache_alloc(msg_cache, allocation);
+	return kmem_cache_alloc(msg_cache, GFP_ATOMIC);
 }
 
 void dlm_free_msg(struct dlm_msg *msg)
fs/dlm/memory.h +2 -2
index 6b29563d24f7..15198d46b42a 100644
@@ -20,11 +20,11 @@ struct dlm_lkb *dlm_allocate_lkb(struct dlm_ls *ls);
 void dlm_free_lkb(struct dlm_lkb *l);
 char *dlm_allocate_lvb(struct dlm_ls *ls);
 void dlm_free_lvb(char *l);
-struct dlm_mhandle *dlm_allocate_mhandle(gfp_t allocation);
+struct dlm_mhandle *dlm_allocate_mhandle(void);
 void dlm_free_mhandle(struct dlm_mhandle *mhandle);
 struct writequeue_entry *dlm_allocate_writequeue(void);
 void dlm_free_writequeue(struct writequeue_entry *writequeue);
-struct dlm_msg *dlm_allocate_msg(gfp_t allocation);
+struct dlm_msg *dlm_allocate_msg(void);
 void dlm_free_msg(struct dlm_msg *msg);
 struct dlm_callback *dlm_allocate_cb(void);
 void dlm_free_cb(struct dlm_callback *cb);
fs/dlm/midcomms.c +10 -14
index 2247ebb61be1..ea0559e2a44d 100644
@@ -380,8 +380,7 @@ static int dlm_send_ack(int nodeid, uint32_t seq)
 	struct dlm_msg *msg;
 	char *ppc;
 
-	msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_ATOMIC, &ppc,
-				   NULL, NULL);
+	msg = dlm_lowcomms_new_msg(nodeid, mb_len, &ppc, NULL, NULL);
 	if (!msg)
 		return -ENOMEM;
 
@@ -429,7 +428,7 @@ static int dlm_send_fin(struct midcomms_node *node,
 	struct dlm_mhandle *mh;
 	char *ppc;
 
-	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_ATOMIC, &ppc);
+	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, &ppc);
 	if (!mh)
 		return -ENOMEM;
 
@@ -977,13 +976,13 @@ static void midcomms_new_msg_cb(void *data)
 }
 
 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
-						int len, gfp_t allocation, char **ppc)
+						int len, char **ppc)
 {
 	struct dlm_opts *opts;
 	struct dlm_msg *msg;
 
 	msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
-				   allocation, ppc, midcomms_new_msg_cb, mh);
+				   ppc, midcomms_new_msg_cb, mh);
 	if (!msg)
 		return NULL;
 
@@ -1002,8 +1001,7 @@ static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int node
  * dlm_midcomms_commit_mhandle which is a must call if success
  */
 #ifndef __CHECKER__
-struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
-					     gfp_t allocation, char **ppc)
+struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len, char **ppc)
 {
 	struct midcomms_node *node;
 	struct dlm_mhandle *mh;
@@ -1018,7 +1016,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 	/* this is a bug, however we going on and hope it will be resolved */
 	WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
 
-	mh = dlm_allocate_mhandle(allocation);
+	mh = dlm_allocate_mhandle();
 	if (!mh)
 		goto err;
 
@@ -1029,8 +1027,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 
 	switch (node->version) {
 	case DLM_VERSION_3_1:
-		msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
-					   NULL, NULL);
+		msg = dlm_lowcomms_new_msg(nodeid, len, ppc, NULL, NULL);
 		if (!msg) {
 			dlm_free_mhandle(mh);
 			goto err;
@@ -1041,8 +1038,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 		/* send ack back if necessary */
 		dlm_send_ack_threshold(node, DLM_SEND_ACK_BACK_MSG_THRESHOLD);
 
-		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
-					       ppc);
+		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, ppc);
 		if (!msg) {
 			dlm_free_mhandle(mh);
 			goto err;
@@ -1502,8 +1498,8 @@ int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
 	rd.node = node;
 	rd.buf = buf;
 
-	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
-				   &msgbuf, midcomms_new_rawmsg_cb, &rd);
+	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, &msgbuf,
+				   midcomms_new_rawmsg_cb, &rd);
 	if (!msg)
 		return -ENOMEM;
 
fs/dlm/midcomms.h +1 -2
index e7246fb3ef57..278d26fdeb2c 100644
@@ -16,8 +16,7 @@ struct midcomms_node;
 
 int dlm_validate_incoming_buffer(int nodeid, unsigned char *buf, int len);
 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int buflen);
-struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
-					     gfp_t allocation, char **ppc);
+struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len, char **ppc);
 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh, const void *name,
 				 int namelen);
 int dlm_midcomms_addr(int nodeid, struct sockaddr_storage *addr, int len);
fs/dlm/rcom.c +3 -4
index 3b734aed26b5..2e3f529f3ff2 100644
@@ -55,7 +55,7 @@ static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
 	struct dlm_mhandle *mh;
 	char *mb;
 
-	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, GFP_NOFS, &mb);
+	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, &mb);
 	if (!mh) {
 		log_print("%s to %d type %d len %d ENOBUFS",
 			  __func__, to_nodeid, type, len);
@@ -75,8 +75,7 @@ static int create_rcom_stateless(struct dlm_ls *ls, int to_nodeid, int type,
 	struct dlm_msg *msg;
 	char *mb;
 
-	msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, GFP_NOFS, &mb,
-				   NULL, NULL);
+	msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, &mb, NULL, NULL);
 	if (!msg) {
 		log_print("create_rcom to %d type %d len %d ENOBUFS",
 			  to_nodeid, type, len);
@@ -510,7 +509,7 @@ int dlm_send_ls_not_ready(int nodeid, const struct dlm_rcom *rc_in)
 	char *mb;
 	int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
 
-	mh = dlm_midcomms_get_mhandle(nodeid, mb_len, GFP_NOFS, &mb);
+	mh = dlm_midcomms_get_mhandle(nodeid, mb_len, &mb);
 	if (!mh)
 		return -ENOBUFS;
 
Signature · 1 line
2.39.3
author maintainer
#1 lore raw

PATCH 2/10 dlm: switch to GFP_ATOMIC in dlm allocations

This patch prepares to process dlm messages in softirq context. To switch dlm to parse messages in softirq context some code parts either runs inside the softirq context or need to switch to run while a spinlock is held. This patch prepares to switch the allocation context to GFP_ATOMIC for those places. It's not possible anymore to preload idr allocations. However this is only a performance speedup and we might switch to xarray implementation with more lockless readers paradigms.

---
 fs/dlm/lock.c         | 2 --
 fs/dlm/memory.c       | 6 +++---
 fs/dlm/recover.c      | 2 --
 fs/dlm/requestqueue.c | 2 +-
 4 files changed, 4 insertions(+), 8 deletions(-)
fs/dlm/lock.c +0 -2
index 44ea65a73f3b..970b8499b66f 100644
@@ -1208,13 +1208,11 @@ static int _create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret,
 	spin_lock_init(&lkb->lkb_cb_lock);
 	INIT_WORK(&lkb->lkb_cb_work, dlm_callback_work);
 
-	idr_preload(GFP_NOFS);
 	spin_lock(&ls->ls_lkbidr_spin);
 	rv = idr_alloc(&ls->ls_lkbidr, lkb, start, end, GFP_NOWAIT);
 	if (rv >= 0)
 		lkb->lkb_id = rv;
 	spin_unlock(&ls->ls_lkbidr_spin);
-	idr_preload_end();
 
 	if (rv < 0) {
 		log_error(ls, "create_lkb idr error %d", rv);
fs/dlm/memory.c +3 -3
index c0c1a83f6381..f44532d9f5c8 100644
@@ -84,7 +84,7 @@ char *dlm_allocate_lvb(struct dlm_ls *ls)
 {
 	char *p;
 
-	p = kzalloc(ls->ls_lvblen, GFP_NOFS);
+	p = kzalloc(ls->ls_lvblen, GFP_ATOMIC);
 	return p;
 }
 
@@ -97,7 +97,7 @@ struct dlm_rsb *dlm_allocate_rsb(struct dlm_ls *ls)
 {
 	struct dlm_rsb *r;
 
-	r = kmem_cache_zalloc(rsb_cache, GFP_NOFS);
+	r = kmem_cache_zalloc(rsb_cache, GFP_ATOMIC);
 	return r;
 }
 
@@ -112,7 +112,7 @@ struct dlm_lkb *dlm_allocate_lkb(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb;
 
-	lkb = kmem_cache_zalloc(lkb_cache, GFP_NOFS);
+	lkb = kmem_cache_zalloc(lkb_cache, GFP_ATOMIC);
 	return lkb;
 }
 
fs/dlm/recover.c +0 -2
index 53917c0aa3c0..ce6dc914cb86 100644
@@ -310,7 +310,6 @@ static int recover_idr_add(struct dlm_rsb *r)
 	struct dlm_ls *ls = r->res_ls;
 	int rv;
 
-	idr_preload(GFP_NOFS);
 	spin_lock(&ls->ls_recover_idr_lock);
 	if (r->res_id) {
 		rv = -1;
@@ -326,7 +325,6 @@ static int recover_idr_add(struct dlm_rsb *r)
 	rv = 0;
 out_unlock:
 	spin_unlock(&ls->ls_recover_idr_lock);
-	idr_preload_end();
 	return rv;
 }
 
fs/dlm/requestqueue.c +1 -1
index 892d6ca21e74..c05940afd063 100644
@@ -37,7 +37,7 @@ void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid,
 	int length = le16_to_cpu(ms->m_header.h_length) -
 		sizeof(struct dlm_message);
 
-	e = kmalloc(sizeof(struct rq_entry) + length, GFP_NOFS);
+	e = kmalloc(sizeof(struct rq_entry) + length, GFP_ATOMIC);
 	if (!e) {
 		log_print("dlm_add_requestqueue: out of memory len %d", length);
 		return;
Signature · 1 line
2.39.3
author maintainer
#2 lore raw

PATCH 3/10 dlm: move root_list to ls_recover() stack

In reply to Alexander Aring

This patch moves the per lockspace ls_root_list list which is mainly used for snapshoting all dlm_rsb from a hash to a list to do recovery handling into the recovery function ls_recover() as stack variable.

Doing that shows that there is no need for locking the ls_root_list which is created at the beginning of ls_recover() and destroyed at the ending of ls_recover(). In between only functionality is called doing read only access to the root_list stack variable.

A special case is assigning the per lockspace ls_recover_dir_root_list variable to the stack variable. The ls_recover_dir_root_list get accessed by another concurrent process dlm_copy_master_names() during the time between ls_recover_dir_root_list is set and set to NULL again. This is done by a special distributed barrier functionality between dlm_recover_members_wait() and dlm_recover_directory_wait(). A comment was made to mention about this handling which might be changed to a better behaviour in future. However setting the ls_recover_dir_root_list to the stack variable and set it to NULL in this specific time will show us potential issues with the recovery handling if it's breaks.

---
 fs/dlm/dir.c          | 27 +++++++++-------------
 fs/dlm/dir.h          |  3 ++-
 fs/dlm/dlm_internal.h |  4 +---
 fs/dlm/lock.c         |  6 ++---
 fs/dlm/lock.h         |  2 +-
 fs/dlm/lockspace.c    |  2 --
 fs/dlm/recover.c      | 52 +++++++++++++------------------------------
 fs/dlm/recover.h      | 12 +++++-----
 fs/dlm/recoverd.c     | 25 ++++++++++++++-------
 9 files changed, 55 insertions(+), 78 deletions(-)
fs/dlm/dir.c +10 -17
index f6acba4310a7..bed83ef6b8a5 100644
@@ -47,15 +47,13 @@ int dlm_dir_nodeid(struct dlm_rsb *r)
 	return r->res_dir_nodeid;
 }
 
-void dlm_recover_dir_nodeid(struct dlm_ls *ls)
+void dlm_recover_dir_nodeid(struct dlm_ls *ls, const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		r->res_dir_nodeid = dlm_hash2nodeid(ls, r->res_hash);
 	}
-	up_read(&ls->ls_root_sem);
 }
 
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq)
@@ -216,16 +214,14 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
 	if (!rv)
 		return r;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, ls->ls_recover_dir_root_list, res_root_list) {
 		if (len == r->res_length && !memcmp(name, r->res_name, len)) {
-			up_read(&ls->ls_root_sem);
 			log_debug(ls, "find_rsb_root revert to root_list %s",
 				  r->res_name);
 			return r;
 		}
 	}
-	up_read(&ls->ls_root_sem);
+
 	return NULL;
 }
 
@@ -236,26 +232,25 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
 void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
  			   char *outbuf, int outlen, int nodeid)
 {
+	const struct list_head *root_list = READ_ONCE(ls->ls_recover_dir_root_list);
 	struct list_head *list;
 	struct dlm_rsb *r;
 	int offset = 0, dir_nodeid;
 	__be16 be_namelen;
 
-	down_read(&ls->ls_root_sem);
-
 	if (inlen > 1) {
 		r = find_rsb_root(ls, inbuf, inlen);
 		if (!r) {
 			log_error(ls, "copy_master_names from %d start %d %.*s",
 				  nodeid, inlen, inlen, inbuf);
-			goto out;
+			return;
 		}
 		list = r->res_root_list.next;
 	} else {
-		list = ls->ls_root_list.next;
+		list = root_list->next;
 	}
 
-	for (offset = 0; list != &ls->ls_root_list; list = list->next) {
+	for (offset = 0; list != root_list; list = list->next) {
 		r = list_entry(list, struct dlm_rsb, res_root_list);
 		if (r->res_nodeid)
 			continue;
@@ -278,7 +273,7 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 			memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
 			offset += sizeof(__be16);
 			ls->ls_recover_dir_sent_msg++;
-			goto out;
+			return;
 		}
 
 		be_namelen = cpu_to_be16(r->res_length);
@@ -294,14 +289,12 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 	 * terminating record.
 	 */
 
-	if ((list == &ls->ls_root_list) &&
+	if ((list == root_list) &&
 	    (offset + sizeof(uint16_t) <= outlen)) {
 		be_namelen = cpu_to_be16(0xFFFF);
 		memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
 		offset += sizeof(__be16);
 		ls->ls_recover_dir_sent_msg++;
 	}
- out:
-	up_read(&ls->ls_root_sem);
 }
 
fs/dlm/dir.h +2 -1
index 39ecb69d7ef3..5b2a7ee3762d 100644
@@ -14,7 +14,8 @@
 
 int dlm_dir_nodeid(struct dlm_rsb *rsb);
 int dlm_hash2nodeid(struct dlm_ls *ls, uint32_t hash);
-void dlm_recover_dir_nodeid(struct dlm_ls *ls);
+void dlm_recover_dir_nodeid(struct dlm_ls *ls,
+			    const struct list_head *root_list);
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq);
 void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 			   char *outbuf, int outlen, int nodeid);
fs/dlm/dlm_internal.h +1 -3
index dfc444dad329..66c67b17d273 100644
@@ -629,6 +629,7 @@ struct dlm_ls {
 	struct mutex		ls_requestqueue_mutex;
 	struct dlm_rcom		*ls_recover_buf;
 	int			ls_recover_nodeid; /* for debugging */
+	const struct list_head	*ls_recover_dir_root_list; /* root resources */
 	unsigned int		ls_recover_dir_sent_res; /* for log info */
 	unsigned int		ls_recover_dir_sent_msg; /* for log info */
 	unsigned int		ls_recover_locks_in; /* for log info */
@@ -643,9 +644,6 @@ struct dlm_ls {
 	wait_queue_head_t	ls_recover_lock_wait;
 	spinlock_t		ls_clear_proc_locks;
 
-	struct list_head	ls_root_list;	/* root resources */
-	struct rw_semaphore	ls_root_sem;	/* protect root_list */
-
 	const struct dlm_lockspace_ops *ls_ops;
 	void			*ls_ops_arg;
 
fs/dlm/lock.c +2 -4
index 970b8499b66f..0218645e2f90 100644
@@ -5185,7 +5185,7 @@ static void purge_dead_list(struct dlm_ls *ls, struct dlm_rsb *r,
 
 /* Get rid of locks held by nodes that are gone. */
 
-void dlm_recover_purge(struct dlm_ls *ls)
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 	struct dlm_member *memb;
@@ -5204,8 +5204,7 @@ void dlm_recover_purge(struct dlm_ls *ls)
 	if (!nodes_count)
 		return;
 
-	down_write(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		hold_rsb(r);
 		lock_rsb(r);
 		if (is_master(r)) {
@@ -5220,7 +5219,6 @@ void dlm_recover_purge(struct dlm_ls *ls)
 		unhold_rsb(r);
 		cond_resched();
 	}
-	up_write(&ls->ls_root_sem);
 
 	if (lkb_count)
 		log_rinfo(ls, "dlm_recover_purge %u locks for %u nodes",
fs/dlm/lock.h +1 -1
index b54e2cbbe6e2..c8ff7780d3cc 100644
@@ -31,7 +31,7 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 int dlm_search_rsb_tree(struct rb_root *tree, const void *name, int len,
 			struct dlm_rsb **r_ret);
 
-void dlm_recover_purge(struct dlm_ls *ls);
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list);
 void dlm_purge_mstcpy_locks(struct dlm_rsb *r);
 void dlm_recover_grant(struct dlm_ls *ls);
 int dlm_recover_waiters_post(struct dlm_ls *ls);
fs/dlm/lockspace.c +0 -2
index 0455dddb0797..cf7528144827 100644
@@ -580,8 +580,6 @@ static int new_lockspace(const char *name, const char *cluster,
 	ls->ls_recover_list_count = 0;
 	ls->ls_local_handle = ls;
 	init_waitqueue_head(&ls->ls_wait_general);
-	INIT_LIST_HEAD(&ls->ls_root_list);
-	init_rwsem(&ls->ls_root_sem);
 
 	spin_lock(&lslist_lock);
 	ls->ls_create_count = 1;
fs/dlm/recover.c +15 -37

141 lines not on this page · raw message

fs/dlm/recover.h +7 -5
index dbc51013ecad..9dffdd825fb6 100644
@@ -19,14 +19,16 @@ int dlm_recover_members_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_directory_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_locks_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_done_wait(struct dlm_ls *ls, uint64_t seq);
-int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
+			const struct list_head *root_list);
 int dlm_recover_master_reply(struct dlm_ls *ls, const struct dlm_rcom *rc);
-int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
+		      const struct list_head *root_list);
 void dlm_recovered_lock(struct dlm_rsb *r);
-int dlm_create_root_list(struct dlm_ls *ls);
-void dlm_release_root_list(struct dlm_ls *ls);
+void dlm_create_root_list(struct dlm_ls *ls, struct list_head *root_list);
+void dlm_release_root_list(struct list_head *root_list);
 void dlm_clear_toss(struct dlm_ls *ls);
-void dlm_recover_rsbs(struct dlm_ls *ls);
+void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list);
 
 #endif				/* __RECOVER_DOT_H__ */
 
fs/dlm/recoverd.c +17 -8
index 4d17491dea2f..f0f26a6a322c 100644
@@ -50,6 +50,7 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 
 static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 {
+	LIST_HEAD(root_list);
 	unsigned long start;
 	int error, neg = 0;
 
@@ -66,7 +67,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	 * routines.
 	 */
 
-	dlm_create_root_list(ls);
+	dlm_create_root_list(ls, &root_list);
 
 	/*
 	 * Add or remove nodes from the lockspace's ls_nodes list.
@@ -82,8 +83,15 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		goto fail;
 	}
 
-	dlm_recover_dir_nodeid(ls);
+	dlm_recover_dir_nodeid(ls, &root_list);
 
+	/* Those fields are set for dlm_copy_master_names() which only
+	 * access these fields during between dlm_recover_members_wait()
+	 * and dlm_recover_directory_wait(). If ls_recover_dir_root_list
+	 * get accessed outside of ls_recover() or those functions it is
+	 * a bug and a crash will signal we having problems here.
+	 */
+	WRITE_ONCE(ls->ls_recover_dir_root_list, &root_list);
 	ls->ls_recover_dir_sent_res = 0;
 	ls->ls_recover_dir_sent_msg = 0;
 	ls->ls_recover_locks_in = 0;
@@ -117,6 +125,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		goto fail;
 	}
 
+	WRITE_ONCE(ls->ls_recover_dir_root_list, NULL);
 	log_rinfo(ls, "dlm_recover_directory %u out %u messages",
 		  ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
 
@@ -138,14 +147,14 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		 * Clear lkb's for departed nodes.
 		 */
 
-		dlm_recover_purge(ls);
+		dlm_recover_purge(ls, &root_list);
 
 		/*
 		 * Get new master nodeid's for rsb's that were mastered on
 		 * departed nodes.
 		 */
 
-		error = dlm_recover_masters(ls, rv->seq);
+		error = dlm_recover_masters(ls, rv->seq, &root_list);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_masters error %d", error);
 			goto fail;
@@ -155,7 +164,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		 * Send our locks on remastered rsb's to the new masters.
 		 */
 
-		error = dlm_recover_locks(ls, rv->seq);
+		error = dlm_recover_locks(ls, rv->seq, &root_list);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_locks error %d", error);
 			goto fail;
@@ -178,7 +187,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		 * settings.
 		 */
 
-		dlm_recover_rsbs(ls);
+		dlm_recover_rsbs(ls, &root_list);
 	} else {
 		/*
 		 * Other lockspace members may be going through the "neg" steps
@@ -194,7 +203,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		}
 	}
 
-	dlm_release_root_list(ls);
+	dlm_release_root_list(&root_list);
 
 	/*
 	 * Purge directory-related requests that are saved in requestqueue.
@@ -244,7 +253,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	return 0;
 
  fail:
-	dlm_release_root_list(ls);
+	dlm_release_root_list(&root_list);
 	mutex_unlock(&ls->ls_recoverd_active);
 
 	return error;
Signature · 1 line
2.39.3
author maintainer
#3 lore raw

Re: [PATCH dlm/next 03/10] dlm: move root_list to ls_recover() stack

Hi,

Quoted 429 lines · Alexander Aring

This patch moves the per lockspace ls_root_list list which is mainly used for snapshoting all dlm_rsb from a hash to a list to do recovery handling into the recovery function ls_recover() as stack variable.

Doing that shows that there is no need for locking the ls_root_list which is created at the beginning of ls_recover() and destroyed at the ending of ls_recover(). In between only functionality is called doing read only access to the root_list stack variable.

A special case is assigning the per lockspace ls_recover_dir_root_list variable to the stack variable. The ls_recover_dir_root_list get accessed by another concurrent process dlm_copy_master_names() during the time between ls_recover_dir_root_list is set and set to NULL again. This is done by a special distributed barrier functionality between dlm_recover_members_wait() and dlm_recover_directory_wait(). A comment was made to mention about this handling which might be changed to a better behaviour in future. However setting the ls_recover_dir_root_list to the stack variable and set it to NULL in this specific time will show us potential issues with the recovery handling if it's breaks.

---
 fs/dlm/dir.c          | 27 +++++++++-------------
 fs/dlm/dir.h          |  3 ++-
 fs/dlm/dlm_internal.h |  4 +---
 fs/dlm/lock.c         |  6 ++---
 fs/dlm/lock.h         |  2 +-
 fs/dlm/lockspace.c    |  2 --
 fs/dlm/recover.c      | 52 +++++++++++++------------------------------
 fs/dlm/recover.h      | 12 +++++-----
 fs/dlm/recoverd.c     | 25 ++++++++++++++-------
 9 files changed, 55 insertions(+), 78 deletions(-)
fs/dlm/dir.c +10 -17
index f6acba4310a7..bed83ef6b8a5 100644
@@ -47,15 +47,13 @@ int dlm_dir_nodeid(struct dlm_rsb *r)
        return r->res_dir_nodeid;
 }
 
-void dlm_recover_dir_nodeid(struct dlm_ls *ls)
+void dlm_recover_dir_nodeid(struct dlm_ls *ls, const struct list_head *root_list)
 {
        struct dlm_rsb *r;
 
-       down_read(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, root_list, res_root_list) {
                r->res_dir_nodeid = dlm_hash2nodeid(ls, r->res_hash);
        }
-       up_read(&ls->ls_root_sem);
 }
 
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq)
@@ -216,16 +214,14 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
        if (!rv)
                return r;
 
-       down_read(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, ls->ls_recover_dir_root_list, res_root_list) {
                if (len == r->res_length && !memcmp(name, r->res_name, len)) {
-                       up_read(&ls->ls_root_sem);
                        log_debug(ls, "find_rsb_root revert to root_list %s",
                                  r->res_name);
                        return r;
                }
        }
-       up_read(&ls->ls_root_sem);
+
        return NULL;
 }
 
@@ -236,26 +232,25 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
 void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
                           char *outbuf, int outlen, int nodeid)
 {
+       const struct list_head *root_list = READ_ONCE(ls->ls_recover_dir_root_list);
        struct list_head *list;
        struct dlm_rsb *r;
        int offset = 0, dir_nodeid;
        __be16 be_namelen;
 
-       down_read(&ls->ls_root_sem);
-
        if (inlen > 1) {
                r = find_rsb_root(ls, inbuf, inlen);
                if (!r) {
                        log_error(ls, "copy_master_names from %d start %d %.*s",
                                  nodeid, inlen, inlen, inbuf);
-                       goto out;
+                       return;
                }
                list = r->res_root_list.next;
        } else {
-               list = ls->ls_root_list.next;
+               list = root_list->next;
        }
 
-       for (offset = 0; list != &ls->ls_root_list; list = list->next) {
+       for (offset = 0; list != root_list; list = list->next) {
                r = list_entry(list, struct dlm_rsb, res_root_list);
                if (r->res_nodeid)
                        continue;
@@ -278,7 +273,7 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
                        memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
                        offset += sizeof(__be16);
                        ls->ls_recover_dir_sent_msg++;
-                       goto out;
+                       return;
                }
 
                be_namelen = cpu_to_be16(r->res_length);
@@ -294,14 +289,12 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
         * terminating record.
         */
 
-       if ((list == &ls->ls_root_list) &&
+       if ((list == root_list) &&
            (offset + sizeof(uint16_t) <= outlen)) {
                be_namelen = cpu_to_be16(0xFFFF);
                memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
                offset += sizeof(__be16);
                ls->ls_recover_dir_sent_msg++;
        }
- out:
-       up_read(&ls->ls_root_sem);
 }
 
fs/dlm/dir.h +2 -1
index 39ecb69d7ef3..5b2a7ee3762d 100644
@@ -14,7 +14,8 @@
 
 int dlm_dir_nodeid(struct dlm_rsb *rsb);
 int dlm_hash2nodeid(struct dlm_ls *ls, uint32_t hash);
-void dlm_recover_dir_nodeid(struct dlm_ls *ls);
+void dlm_recover_dir_nodeid(struct dlm_ls *ls,
+                           const struct list_head *root_list);
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq);
 void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
                           char *outbuf, int outlen, int nodeid);
fs/dlm/dlm_internal.h +1 -3
index dfc444dad329..66c67b17d273 100644
@@ -629,6 +629,7 @@ struct dlm_ls {
        struct mutex            ls_requestqueue_mutex;
        struct dlm_rcom         *ls_recover_buf;
        int                     ls_recover_nodeid; /* for debugging */
+       const struct list_head  *ls_recover_dir_root_list; /* root resources */
        unsigned int            ls_recover_dir_sent_res; /* for log info */
        unsigned int            ls_recover_dir_sent_msg; /* for log info */
        unsigned int            ls_recover_locks_in; /* for log info */
@@ -643,9 +644,6 @@ struct dlm_ls {
        wait_queue_head_t       ls_recover_lock_wait;
        spinlock_t              ls_clear_proc_locks;
 
-       struct list_head        ls_root_list;   /* root resources */
-       struct rw_semaphore     ls_root_sem;    /* protect root_list */
-
        const struct dlm_lockspace_ops *ls_ops;
        void                    *ls_ops_arg;
 
fs/dlm/lock.c +2 -4
index 970b8499b66f..0218645e2f90 100644
@@ -5185,7 +5185,7 @@ static void purge_dead_list(struct dlm_ls *ls, struct dlm_rsb *r,
 
 /* Get rid of locks held by nodes that are gone. */
 
-void dlm_recover_purge(struct dlm_ls *ls)
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list)
 {
        struct dlm_rsb *r;
        struct dlm_member *memb;
@@ -5204,8 +5204,7 @@ void dlm_recover_purge(struct dlm_ls *ls)
        if (!nodes_count)
                return;
 
-       down_write(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, root_list, res_root_list) {
                hold_rsb(r);
                lock_rsb(r);
                if (is_master(r)) {
@@ -5220,7 +5219,6 @@ void dlm_recover_purge(struct dlm_ls *ls)
                unhold_rsb(r);
                cond_resched();
        }
-       up_write(&ls->ls_root_sem);
 
        if (lkb_count)
                log_rinfo(ls, "dlm_recover_purge %u locks for %u nodes",
fs/dlm/lock.h +1 -1
index b54e2cbbe6e2..c8ff7780d3cc 100644
@@ -31,7 +31,7 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 int dlm_search_rsb_tree(struct rb_root *tree, const void *name, int len,
                        struct dlm_rsb **r_ret);
 
-void dlm_recover_purge(struct dlm_ls *ls);
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list);
 void dlm_purge_mstcpy_locks(struct dlm_rsb *r);
 void dlm_recover_grant(struct dlm_ls *ls);
 int dlm_recover_waiters_post(struct dlm_ls *ls);
fs/dlm/lockspace.c +0 -2
index 0455dddb0797..cf7528144827 100644
@@ -580,8 +580,6 @@ static int new_lockspace(const char *name, const char *cluster,
        ls->ls_recover_list_count = 0;
        ls->ls_local_handle = ls;
        init_waitqueue_head(&ls->ls_wait_general);
-       INIT_LIST_HEAD(&ls->ls_root_list);
-       init_rwsem(&ls->ls_root_sem);
 
        spin_lock(&lslist_lock);
        ls->ls_create_count = 1;
fs/dlm/recover.c +15 -37
index ce6dc914cb86..4a4345455d7f 100644
@@ -519,7 +519,8 @@ static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
  * the correct dir node.
  */
 
-int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
+int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
+                       const struct list_head *root_list)
 {
        struct dlm_rsb *r;
        unsigned int total = 0;
@@ -529,10 +530,8 @@ int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
 
        log_rinfo(ls, "dlm_recover_masters");
 
-       down_read(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, root_list, res_root_list) {
                if (dlm_recovery_stopped(ls)) {
-                       up_read(&ls->ls_root_sem);
                        error = -EINTR;
                        goto out;
                }
@@ -546,12 +545,9 @@ int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
                cond_resched();
                total++;
 
-               if (error) {
-                       up_read(&ls->ls_root_sem);
+               if (error)
                        goto out;
-               }
        }
-       up_read(&ls->ls_root_sem);
 
        log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
 
@@ -656,13 +652,13 @@ static int recover_locks(struct dlm_rsb *r, uint64_t seq)
        return error;
 }
 
-int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq)
+int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
+                     const struct list_head *root_list)
 {
        struct dlm_rsb *r;
        int error, count = 0;
 
-       down_read(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, root_list, res_root_list) {
                if (is_master(r)) {
                        rsb_clear_flag(r, RSB_NEW_MASTER);
                        continue;
@@ -673,19 +669,15 @@ int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq)
 
                if (dlm_recovery_stopped(ls)) {
                        error = -EINTR;
-                       up_read(&ls->ls_root_sem);
                        goto out;
                }
 
                error = recover_locks(r, seq);
-               if (error) {
-                       up_read(&ls->ls_root_sem);
+               if (error)
                        goto out;
-               }
 
                count += r->res_recover_locks_count;
        }
-       up_read(&ls->ls_root_sem);
 
        log_rinfo(ls, "dlm_recover_locks %d out", count);
 
@@ -854,13 +846,12 @@ static void recover_grant(struct dlm_rsb *r)
                rsb_set_flag(r, RSB_RECOVER_GRANT);
 }
 
-void dlm_recover_rsbs(struct dlm_ls *ls)
+void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list)
 {
        struct dlm_rsb *r;
        unsigned int count = 0;
 
-       down_read(&ls->ls_root_sem);
-       list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry(r, root_list, res_root_list) {
                lock_rsb(r);
                if (is_master(r)) {
                        if (rsb_flag(r, RSB_RECOVER_CONVERT))
@@ -881,7 +872,6 @@ void dlm_recover_rsbs(struct dlm_ls *ls)
                rsb_clear_flag(r, RSB_NEW_MASTER2);
                unlock_rsb(r);
        }
-       up_read(&ls->ls_root_sem);
 
        if (count)
                log_rinfo(ls, "dlm_recover_rsbs %d done", count);
@@ -889,24 +879,17 @@ void dlm_recover_rsbs(struct dlm_ls *ls)
 
 /* Create a single list of all root rsb's to be used during recovery */
 
-int dlm_create_root_list(struct dlm_ls *ls)
+void dlm_create_root_list(struct dlm_ls *ls, struct list_head *root_list)
 {
        struct rb_node *n;
        struct dlm_rsb *r;
-       int i, error = 0;
-
-       down_write(&ls->ls_root_sem);
-       if (!list_empty(&ls->ls_root_list)) {
-               log_error(ls, "root list not empty");
-               error = -EINVAL;
-               goto out;
-       }
+       int i;
 
        for (i = 0; i < ls->ls_rsbtbl_size; i++) {
                spin_lock(&ls->ls_rsbtbl[i].lock);
                for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
                        r = rb_entry(n, struct dlm_rsb, res_hashnode);
-                       list_add(&r->res_root_list, &ls->ls_root_list);
+                       list_add(&r->res_root_list, root_list);
                        dlm_hold_rsb(r);
                }
 
@@ -914,21 +897,16 @@ int dlm_create_root_list(struct dlm_ls *ls)
                        log_error(ls, "dlm_create_root_list toss not empty");
                spin_unlock(&ls->ls_rsbtbl[i].lock);
        }
- out:
-       up_write(&ls->ls_root_sem);
-       return error;
 }
 
-void dlm_release_root_list(struct dlm_ls *ls)
+void dlm_release_root_list(struct list_head *root_list)
 {
        struct dlm_rsb *r, *safe;
 
-       down_write(&ls->ls_root_sem);
-       list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
+       list_for_each_entry_safe(r, safe, root_list, res_root_list) {
                list_del_init(&r->res_root_list);
                dlm_put_rsb(r);
        }
-       up_write(&ls->ls_root_sem);
 }
 
 void dlm_clear_toss(struct dlm_ls *ls)
fs/dlm/recover.h +7 -5
index dbc51013ecad..9dffdd825fb6 100644
@@ -19,14 +19,16 @@ int dlm_recover_members_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_directory_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_locks_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_done_wait(struct dlm_ls *ls, uint64_t seq);
-int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
+                       const struct list_head *root_list);
 int dlm_recover_master_reply(struct dlm_ls *ls, const struct dlm_rcom *rc);
-int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
+                     const struct list_head *root_list);
 void dlm_recovered_lock(struct dlm_rsb *r);
-int dlm_create_root_list(struct dlm_ls *ls);
-void dlm_release_root_list(struct dlm_ls *ls);
+void dlm_create_root_list(struct dlm_ls *ls, struct list_head *root_list);
+void dlm_release_root_list(struct list_head *root_list);
 void dlm_clear_toss(struct dlm_ls *ls);
-void dlm_recover_rsbs(struct dlm_ls *ls);
+void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list);
 
 #endif                         /* __RECOVER_DOT_H__ */
 
fs/dlm/recoverd.c +2 -1
index 4d17491dea2f..f0f26a6a322c 100644
@@ -50,6 +50,7 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 
 static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 {
+       LIST_HEAD(root_list);
        unsigned long start;
        int error, neg = 0;
 
@@ -66,7 +67,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
         * routines.
         */
 
-       dlm_create_root_list(ls);
+       dlm_create_root_list(ls, &root_list);
 
        /*
         * Add or remove nodes from the lockspace's ls_nodes list.
@@ -82,8 +83,15 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
                goto fail;
        }
 

fs/dlm/recoverd.c static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv) · line 94

Quoted 10 lines
fs/dlm/recoverd.c +8 -1
-       dlm_recover_dir_nodeid(ls);
+       dlm_recover_dir_nodeid(ls, &root_list);
 
+       /* Those fields are set for dlm_copy_master_names() which only
+        * access these fields during between dlm_recover_members_wait()
+        * and dlm_recover_directory_wait(). If ls_recover_dir_root_list
+        * get accessed outside of ls_recover() or those functions it is
+        * a bug and a crash will signal we having problems here.
+        */
+       WRITE_ONCE(ls->ls_recover_dir_root_list, &root_list);

not sure what happens when some of the recovery code fails here and then it releases the ls_recover_dir_root_list list while there could be still some rcom messages being processed and executing dlm_copy_master_names(). It will probably crash... I need to think about a better handling for this.

- Alex

author maintainer
#4 lore raw

PATCH 4/10 dlm: drop holding waiters mutex in waiters recovery

In reply to Alexander Aring

This patch drops to hold the ls_waiters_mutex in dlm_recover_waiters_pre(). The dlm_recover_waiters_pre() function is only being called when recovery handling is being done for the specific lockspace. During this time there can't be new lock requests initiated or dlm messages being processed that manipulates the lockspace waiters list.

Only debugfs can access the lockspace waiters list when dlm_recover_waiters_pre() may manipulates it. This is not possible anymore because debugfs will hold the recovery lock when it's acccessing the waiters list.

A check was introduced in remove_from_waiters_ms() for local dlm messaging to check if really the lockspace is stopped and no new lock requests can be initiated.

---
 fs/dlm/debug_fs.c |  4 ++++
 fs/dlm/lock.c     | 17 +++++++++--------
 fs/dlm/lock.h     |  1 +
 3 files changed, 14 insertions(+), 8 deletions(-)
fs/dlm/debug_fs.c +4 -0
index 42f332f46359..1ef59f7223a6 100644
@@ -823,6 +823,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 	size_t len = DLM_DEBUG_BUF_LEN, pos = 0, ret, rv;
 
 	mutex_lock(&debug_buf_lock);
+	dlm_lock_recovery(ls);
 	mutex_lock(&ls->ls_waiters_mutex);
 	memset(debug_buf, 0, sizeof(debug_buf));
 
@@ -835,6 +836,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 		pos += ret;
 	}
 	mutex_unlock(&ls->ls_waiters_mutex);
+	dlm_unlock_recovery(ls);
 
 	rv = simple_read_from_buffer(userbuf, count, ppos, debug_buf, pos);
 	mutex_unlock(&debug_buf_lock);
@@ -858,7 +860,9 @@ static ssize_t waiters_write(struct file *file, const char __user *user_buf,
 	if (n != 3)
 		return -EINVAL;
 
+	dlm_lock_recovery(ls);
 	error = dlm_debug_add_lkb_to_waiters(ls, lkb_id, mstype, to_nodeid);
+	dlm_unlock_recovery(ls);
 	if (error)
 		return error;
 
fs/dlm/lock.c +9 -8
index 0218645e2f90..79f1f741af13 100644
@@ -201,7 +201,7 @@ void dlm_dump_rsb(struct dlm_rsb *r)
 
 /* Threads cannot use the lockspace while it's being recovered */
 
-static inline void dlm_lock_recovery(struct dlm_ls *ls)
+void dlm_lock_recovery(struct dlm_ls *ls)
 {
 	down_read(&ls->ls_in_recovery);
 }
@@ -1553,7 +1553,11 @@ static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 }
 
 /* Handles situations where we might be processing a "fake" or "local" reply in
-   which we can't try to take waiters_mutex again. */
+ * the recovery context which stops any locking activity. Only debugfs might
+ * change the lockspace waiters but they will held the recovery lock to ensure
+ * remove_from_waiters_ms() in local case will be the only user manipulating the
+ * lockspace waiters in recovery context.
+ */
 
 static int remove_from_waiters_ms(struct dlm_lkb *lkb,
 				  const struct dlm_message *ms, bool local)
@@ -1563,6 +1567,9 @@ static int remove_from_waiters_ms(struct dlm_lkb *lkb,
 
 	if (!local)
 		mutex_lock(&ls->ls_waiters_mutex);
+	else
+		WARN_ON_ONCE(!rwsem_is_locked(&ls->ls_in_recovery) ||
+			     !dlm_locking_stopped(ls));
 	error = _remove_from_waiters(lkb, le32_to_cpu(ms->m_type), ms);
 	if (!local)
 		mutex_unlock(&ls->ls_waiters_mutex);
@@ -4395,7 +4402,6 @@ static void _receive_convert_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4434,7 +4440,6 @@ static void _receive_unlock_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4486,7 +4491,6 @@ static void _receive_cancel_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4887,8 +4891,6 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 	if (!ms_local)
 		return;
 
-	mutex_lock(&ls->ls_waiters_mutex);
-
 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
 
 		dir_nodeid = dlm_dir_nodeid(lkb->lkb_resource);
@@ -4981,7 +4983,6 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 		}
 		schedule();
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
 	kfree(ms_local);
 }
 
fs/dlm/lock.h +1 -0
index c8ff7780d3cc..b2fd74a2f8eb 100644
@@ -23,6 +23,7 @@ void dlm_hold_rsb(struct dlm_rsb *r);
 int dlm_put_lkb(struct dlm_lkb *lkb);
 void dlm_scan_rsbs(struct dlm_ls *ls);
 int dlm_lock_recovery_try(struct dlm_ls *ls);
+void dlm_lock_recovery(struct dlm_ls *ls);
 void dlm_unlock_recovery(struct dlm_ls *ls);
 
 int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
Signature · 1 line
2.39.3
author maintainer
#5 lore raw

PATCH 5/10 dlm: convert ls_waiters_mutex to spinlock

In reply to Alexander Aring

This patch converts the per dlm lockspace waiters lock from a mutex to a spinlock.

---
 fs/dlm/debug_fs.c     |  4 ++--
 fs/dlm/dlm_internal.h |  2 +-
 fs/dlm/lock.c         | 20 ++++++++++----------
 fs/dlm/lockspace.c    |  2 +-
 4 files changed, 14 insertions(+), 14 deletions(-)
fs/dlm/debug_fs.c +2 -2
index 1ef59f7223a6..26abeff5b602 100644
@@ -824,7 +824,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 
 	mutex_lock(&debug_buf_lock);
 	dlm_lock_recovery(ls);
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock(&ls->ls_waiters_lock);
 	memset(debug_buf, 0, sizeof(debug_buf));
 
 	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
@@ -835,7 +835,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 			break;
 		pos += ret;
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock(&ls->ls_waiters_lock);
 	dlm_unlock_recovery(ls);
 
 	rv = simple_read_from_buffer(userbuf, count, ppos, debug_buf, pos);
fs/dlm/dlm_internal.h +1 -1
index 66c67b17d273..0f37436be075 100644
@@ -564,7 +564,7 @@ struct dlm_ls {
 	struct dlm_rsbtable	*ls_rsbtbl;
 	uint32_t		ls_rsbtbl_size;
 
-	struct mutex		ls_waiters_mutex;
+	spinlock_t		ls_waiters_lock;
 	struct list_head	ls_waiters;	/* lkbs needing a reply */
 
 	struct mutex		ls_orphans_mutex;
fs/dlm/lock.c +10 -10
index 79f1f741af13..005e2d167cbd 100644
@@ -1407,7 +1407,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
 	int error = 0;
 	int wc;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock(&ls->ls_waiters_lock);
 
 	if (is_overlap_unlock(lkb) ||
 	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
@@ -1447,7 +1447,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
 		log_error(ls, "addwait error %x %d flags %x %d %d %s",
 			  lkb->lkb_id, error, dlm_iflags_val(lkb), mstype,
 			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock(&ls->ls_waiters_lock);
 	return error;
 }
 
@@ -1546,9 +1546,9 @@ static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int error;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock(&ls->ls_waiters_lock);
 	error = _remove_from_waiters(lkb, mstype, NULL);
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock(&ls->ls_waiters_lock);
 	return error;
 }
 
@@ -1566,13 +1566,13 @@ static int remove_from_waiters_ms(struct dlm_lkb *lkb,
 	int error;
 
 	if (!local)
-		mutex_lock(&ls->ls_waiters_mutex);
+		spin_lock(&ls->ls_waiters_lock);
 	else
 		WARN_ON_ONCE(!rwsem_is_locked(&ls->ls_in_recovery) ||
 			     !dlm_locking_stopped(ls));
 	error = _remove_from_waiters(lkb, le32_to_cpu(ms->m_type), ms);
 	if (!local)
-		mutex_unlock(&ls->ls_waiters_mutex);
+		spin_unlock(&ls->ls_waiters_lock);
 	return error;
 }
 
@@ -4990,7 +4990,7 @@ static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb = NULL, *iter;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock(&ls->ls_waiters_lock);
 	list_for_each_entry(iter, &ls->ls_waiters, lkb_wait_reply) {
 		if (test_bit(DLM_IFL_RESEND_BIT, &iter->lkb_iflags)) {
 			hold_lkb(iter);
@@ -4998,7 +4998,7 @@ static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
 			break;
 		}
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock(&ls->ls_waiters_lock);
 
 	return lkb;
 }
@@ -5065,9 +5065,9 @@ int dlm_recover_waiters_post(struct dlm_ls *ls)
 		while (!atomic_dec_and_test(&lkb->lkb_wait_count))
 			unhold_lkb(lkb);
 
-		mutex_lock(&ls->ls_waiters_mutex);
+		spin_lock(&ls->ls_waiters_lock);
 		list_del_init(&lkb->lkb_wait_reply);
-		mutex_unlock(&ls->ls_waiters_mutex);
+		spin_unlock(&ls->ls_waiters_lock);
 
 		if (oc || ou) {
 			/* do an unlock or cancel instead of resending */
fs/dlm/lockspace.c +1 -1
index cf7528144827..79001cc646d8 100644
@@ -515,7 +515,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	spin_lock_init(&ls->ls_lkbidr_spin);
 
 	INIT_LIST_HEAD(&ls->ls_waiters);
-	mutex_init(&ls->ls_waiters_mutex);
+	spin_lock_init(&ls->ls_waiters_lock);
 	INIT_LIST_HEAD(&ls->ls_orphans);
 	mutex_init(&ls->ls_orphans_mutex);
 
Signature · 1 line
2.39.3
author maintainer
#6 lore raw

PATCH 6/10 dlm: convert res_lock to spinlock

In reply to Alexander Aring

This patch converts the per dlm rsb res_lock from a mutex to a spinlock.

---
 fs/dlm/dlm_internal.h | 2 +-
 fs/dlm/lock.c         | 2 +-
 fs/dlm/lock.h         | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)
fs/dlm/dlm_internal.h +1 -1
index 0f37436be075..a052b1dd8eb2 100644
@@ -290,7 +290,7 @@ struct dlm_lkb {
 struct dlm_rsb {
 	struct dlm_ls		*res_ls;	/* the lockspace */
 	struct kref		res_ref;
-	struct mutex		res_mutex;
+	spinlock_t		res_lock;
 	unsigned long		res_flags;
 	int			res_length;	/* length of rsb name */
 	int			res_nodeid;
fs/dlm/lock.c +1 -1
index 005e2d167cbd..4fc6bc3690a6 100644
@@ -415,7 +415,7 @@ static int get_rsb_struct(struct dlm_ls *ls, const void *name, int len,
 	r->res_ls = ls;
 	r->res_length = len;
 	memcpy(r->res_name, name, len);
-	mutex_init(&r->res_mutex);
+	spin_lock_init(&r->res_lock);
 
 	INIT_LIST_HEAD(&r->res_lookup);
 	INIT_LIST_HEAD(&r->res_grantqueue);
fs/dlm/lock.h +2 -2
index b2fd74a2f8eb..0f6b2700c0da 100644
@@ -69,12 +69,12 @@ static inline int is_master(struct dlm_rsb *r)
 
 static inline void lock_rsb(struct dlm_rsb *r)
 {
-	mutex_lock(&r->res_mutex);
+	spin_lock(&r->res_lock);
 }
 
 static inline void unlock_rsb(struct dlm_rsb *r)
 {
-	mutex_unlock(&r->res_mutex);
+	spin_unlock(&r->res_lock);
 }
 
 #endif
Signature · 1 line
2.39.3
author maintainer
#7 lore raw

PATCH 7/10 dlm: make requestqueue handling non sleepable

In reply to Alexander Aring

This patch removes the ls_requestqueue_wait and convert the ls_requestqueue_mutex to a rw lock. Instead of calling wait_event() in dlm processing which waits until all messages are processed and allow new message processing after recovering is done, this patch is using a bitflag to signal when a message should be saved for future or not. When recovery processes all saved messages we will clear this bit again and allow new messages to processed directly.

---
 fs/dlm/dlm_internal.h |  5 ++---
 fs/dlm/lock.c         | 16 ++++++++++++++--
 fs/dlm/lockspace.c    |  4 +---
 fs/dlm/member.c       |  5 +++++
 fs/dlm/requestqueue.c | 41 ++++++++---------------------------------
 5 files changed, 30 insertions(+), 41 deletions(-)
fs/dlm/dlm_internal.h +2 -3
index a052b1dd8eb2..02a265a70de7 100644
@@ -624,9 +624,7 @@ struct dlm_ls {
 	struct rw_semaphore	ls_in_recovery;	/* block local requests */
 	struct rw_semaphore	ls_recv_active;	/* block dlm_recv */
 	struct list_head	ls_requestqueue;/* queue remote requests */
-	atomic_t		ls_requestqueue_cnt;
-	wait_queue_head_t	ls_requestqueue_wait;
-	struct mutex		ls_requestqueue_mutex;
+	rwlock_t		ls_requestqueue_lock;
 	struct dlm_rcom		*ls_recover_buf;
 	int			ls_recover_nodeid; /* for debugging */
 	const struct list_head	*ls_recover_dir_root_list; /* root resources */
@@ -684,6 +682,7 @@ struct dlm_ls {
 #define LSFL_UEVENT_WAIT	7
 #define LSFL_CB_DELAY		9
 #define LSFL_NODIR		10
+#define LSFL_RECV_MSG_BLOCKED	11
 
 /* much of this is just saving user space pointers associated with the
    lock that we pass back to the user lib with an ast */
fs/dlm/lock.c +14 -2
index 4fc6bc3690a6..e2e98811eafa 100644
@@ -4749,20 +4749,32 @@ static void _receive_message(struct dlm_ls *ls, const struct dlm_message *ms,
 static void dlm_receive_message(struct dlm_ls *ls, const struct dlm_message *ms,
 				int nodeid)
 {
-	if (dlm_locking_stopped(ls)) {
+try_again:
+	read_lock(&ls->ls_requestqueue_lock);
+	if (test_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags)) {
 		/* If we were a member of this lockspace, left, and rejoined,
 		   other nodes may still be sending us messages from the
 		   lockspace generation before we left. */
 		if (WARN_ON_ONCE(!ls->ls_generation)) {
+			read_unlock(&ls->ls_requestqueue_lock);
 			log_limit(ls, "receive %d from %d ignore old gen",
 				  le32_to_cpu(ms->m_type), nodeid);
 			return;
 		}
 
+		read_unlock(&ls->ls_requestqueue_lock);
+		write_lock(&ls->ls_requestqueue_lock);
+		/* recheck because we hold writelock now */
+		if (!test_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags)) {
+			write_unlock_bh(&ls->ls_requestqueue_lock);
+			goto try_again;
+		}
+
 		dlm_add_requestqueue(ls, nodeid, ms);
+		write_unlock(&ls->ls_requestqueue_lock);
 	} else {
-		dlm_wait_requestqueue(ls);
 		_receive_message(ls, ms, 0);
+		read_unlock(&ls->ls_requestqueue_lock);
 	}
 }
 
fs/dlm/lockspace.c +1 -3
index 79001cc646d8..b7aa42e8fa4e 100644
@@ -554,9 +554,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	init_rwsem(&ls->ls_in_recovery);
 	init_rwsem(&ls->ls_recv_active);
 	INIT_LIST_HEAD(&ls->ls_requestqueue);
-	atomic_set(&ls->ls_requestqueue_cnt, 0);
-	init_waitqueue_head(&ls->ls_requestqueue_wait);
-	mutex_init(&ls->ls_requestqueue_mutex);
+	rwlock_init(&ls->ls_requestqueue_lock);
 	spin_lock_init(&ls->ls_clear_proc_locks);
 
 	/* Due backwards compatibility with 3.1 we need to use maximum
fs/dlm/member.c +5 -0
index be7909ead71b..707cebcdc533 100644
@@ -642,6 +642,11 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
 	ls->ls_recover_seq++;
+
+	/* activate requestqueue and stop processing */
+	write_lock(&ls->ls_requestqueue_lock);
+	set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
+	write_unlock(&ls->ls_requestqueue_lock);
 	spin_unlock(&ls->ls_recover_lock);
 
 	/*
fs/dlm/requestqueue.c +8 -33

92 lines not on this page · raw message

Signature · 1 line
2.39.3
author maintainer
#8 lore raw

PATCH 8/10 dlm: ls_recv_active semaphore to rwlock

In reply to Alexander Aring

This patch converts the ls_recv_active semaphore to a rwlock to not sleep during dlm message processing.

---
 fs/dlm/dlm_internal.h | 2 +-
 fs/dlm/lock.c         | 4 ++--
 fs/dlm/lockspace.c    | 2 +-
 fs/dlm/member.c       | 4 ++--
 fs/dlm/recoverd.c     | 4 ++--
 5 files changed, 8 insertions(+), 8 deletions(-)
fs/dlm/dlm_internal.h +1 -1
index 02a265a70de7..ea206764e7f0 100644
@@ -622,7 +622,7 @@ struct dlm_ls {
 	uint64_t		ls_recover_seq;
 	struct dlm_recover	*ls_recover_args;
 	struct rw_semaphore	ls_in_recovery;	/* block local requests */
-	struct rw_semaphore	ls_recv_active;	/* block dlm_recv */
+	rwlock_t		ls_recv_active;	/* block dlm_recv */
 	struct list_head	ls_requestqueue;/* queue remote requests */
 	rwlock_t		ls_requestqueue_lock;
 	struct dlm_rcom		*ls_recover_buf;
fs/dlm/lock.c +2 -2
index e2e98811eafa..091a8e58c06c 100644
@@ -4834,7 +4834,7 @@ void dlm_receive_buffer(const union dlm_packet *p, int nodeid)
 	/* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
 	   be inactive (in this ls) before transitioning to recovery mode */
 
-	down_read(&ls->ls_recv_active);
+	read_lock(&ls->ls_recv_active);
 	if (hd->h_cmd == DLM_MSG)
 		dlm_receive_message(ls, &p->message, nodeid);
 	else if (hd->h_cmd == DLM_RCOM)
@@ -4842,7 +4842,7 @@ void dlm_receive_buffer(const union dlm_packet *p, int nodeid)
 	else
 		log_error(ls, "invalid h_cmd %d from %d lockspace %x",
 			  hd->h_cmd, nodeid, le32_to_cpu(hd->u.h_lockspace));
-	up_read(&ls->ls_recv_active);
+	read_unlock(&ls->ls_recv_active);
 
 	dlm_put_lockspace(ls);
 }
fs/dlm/lockspace.c +1 -1
index b7aa42e8fa4e..823a91fa9a74 100644
@@ -552,7 +552,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	ls->ls_recover_seq = get_random_u64();
 	ls->ls_recover_args = NULL;
 	init_rwsem(&ls->ls_in_recovery);
-	init_rwsem(&ls->ls_recv_active);
+	rwlock_init(&ls->ls_recv_active);
 	INIT_LIST_HEAD(&ls->ls_requestqueue);
 	rwlock_init(&ls->ls_requestqueue_lock);
 	spin_lock_init(&ls->ls_clear_proc_locks);
fs/dlm/member.c +2 -2
index 707cebcdc533..ac1b555af9d6 100644
@@ -630,7 +630,7 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * message to the requestqueue without races.
 	 */
 
-	down_write(&ls->ls_recv_active);
+	write_lock(&ls->ls_recv_active);
 
 	/*
 	 * Abort any recovery that's in progress (see RECOVER_STOP,
@@ -654,7 +654,7 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * requestqueue for later.
 	 */
 
-	up_write(&ls->ls_recv_active);
+	write_unlock(&ls->ls_recv_active);
 
 	/*
 	 * This in_recovery lock does two things:
fs/dlm/recoverd.c +2 -2
index f0f26a6a322c..b3c9b704883c 100644
@@ -32,7 +32,7 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 {
 	int error = -EINTR;
 
-	down_write(&ls->ls_recv_active);
+	write_lock(&ls->ls_recv_active);
 
 	spin_lock(&ls->ls_recover_lock);
 	if (ls->ls_recover_seq == seq) {
@@ -44,7 +44,7 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 	}
 	spin_unlock(&ls->ls_recover_lock);
 
-	up_write(&ls->ls_recv_active);
+	write_unlock(&ls->ls_recv_active);
 	return error;
 }
 
Signature · 1 line
2.39.3
author maintainer
#9 lore raw

PATCH 9/10 dlm: convert message parsing locks to disable bh

In reply to Alexander Aring

This patch converts all spinlocks involved in message parsing to it's _bh version. The reason to do that is to convert the message parsing into softirq context and we need to prevent that those locks can be interrupted by a softirq if those are held.

---
 fs/dlm/ast.c          |  28 +++----
 fs/dlm/debug_fs.c     |  32 ++++----
 fs/dlm/lock.c         | 174 ++++++++++++++++++++++--------------------
 fs/dlm/lock.h         |   4 +-
 fs/dlm/lockspace.c    |  51 ++++++-------
 fs/dlm/lowcomms.c     |  16 ++--
 fs/dlm/member.c       |  22 +++---
 fs/dlm/midcomms.c     |  40 +++++-----
 fs/dlm/rcom.c         |  26 +++----
 fs/dlm/recover.c      |  56 +++++++-------
 fs/dlm/recoverd.c     |  12 +--
 fs/dlm/requestqueue.c |  12 +--
 fs/dlm/user.c         |  34 ++++-----
 13 files changed, 258 insertions(+), 249 deletions(-)
fs/dlm/ast.c +14 -14
index 1f2f70a1b824..e3c0903aaa6f 100644
@@ -127,19 +127,19 @@ void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status,
 		return;
 	}
 
-	spin_lock(&lkb->lkb_cb_lock);
+	spin_lock_bh(&lkb->lkb_cb_lock);
 	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
 	switch (rv) {
 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
 		kref_get(&lkb->lkb_ref);
 
-		spin_lock(&ls->ls_cb_lock);
+		spin_lock_bh(&ls->ls_cb_lock);
 		if (test_bit(LSFL_CB_DELAY, &ls->ls_flags)) {
 			list_add(&lkb->lkb_cb_list, &ls->ls_cb_delay);
 		} else {
 			queue_work(ls->ls_callback_wq, &lkb->lkb_cb_work);
 		}
-		spin_unlock(&ls->ls_cb_lock);
+		spin_unlock_bh(&ls->ls_cb_lock);
 		break;
 	case DLM_ENQUEUE_CALLBACK_FAILURE:
 		WARN_ON_ONCE(1);
@@ -150,7 +150,7 @@ void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status,
 		WARN_ON_ONCE(1);
 		break;
 	}
-	spin_unlock(&lkb->lkb_cb_lock);
+	spin_unlock_bh(&lkb->lkb_cb_lock);
 }
 
 void dlm_callback_work(struct work_struct *work)
@@ -162,14 +162,14 @@ void dlm_callback_work(struct work_struct *work)
 	struct dlm_callback *cb;
 	int rv;
 
-	spin_lock(&lkb->lkb_cb_lock);
+	spin_lock_bh(&lkb->lkb_cb_lock);
 	rv = dlm_dequeue_lkb_callback(lkb, &cb);
 	if (WARN_ON_ONCE(rv == DLM_DEQUEUE_CALLBACK_EMPTY)) {
 		clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-		spin_unlock(&lkb->lkb_cb_lock);
+		spin_unlock_bh(&lkb->lkb_cb_lock);
 		goto out;
 	}
-	spin_unlock(&lkb->lkb_cb_lock);
+	spin_unlock_bh(&lkb->lkb_cb_lock);
 
 	for (;;) {
 		castfn = lkb->lkb_astfn;
@@ -190,14 +190,14 @@ void dlm_callback_work(struct work_struct *work)
 
 		kref_put(&cb->ref, dlm_release_callback);
 
-		spin_lock(&lkb->lkb_cb_lock);
+		spin_lock_bh(&lkb->lkb_cb_lock);
 		rv = dlm_dequeue_lkb_callback(lkb, &cb);
 		if (rv == DLM_DEQUEUE_CALLBACK_EMPTY) {
 			clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-			spin_unlock(&lkb->lkb_cb_lock);
+			spin_unlock_bh(&lkb->lkb_cb_lock);
 			break;
 		}
-		spin_unlock(&lkb->lkb_cb_lock);
+		spin_unlock_bh(&lkb->lkb_cb_lock);
 	}
 
 out:
@@ -225,9 +225,9 @@ void dlm_callback_stop(struct dlm_ls *ls)
 void dlm_callback_suspend(struct dlm_ls *ls)
 {
 	if (ls->ls_callback_wq) {
-		spin_lock(&ls->ls_cb_lock);
+		spin_lock_bh(&ls->ls_cb_lock);
 		set_bit(LSFL_CB_DELAY, &ls->ls_flags);
-		spin_unlock(&ls->ls_cb_lock);
+		spin_unlock_bh(&ls->ls_cb_lock);
 
 		flush_workqueue(ls->ls_callback_wq);
 	}
@@ -245,7 +245,7 @@ void dlm_callback_resume(struct dlm_ls *ls)
 		return;
 
 more:
-	spin_lock(&ls->ls_cb_lock);
+	spin_lock_bh(&ls->ls_cb_lock);
 	list_for_each_entry_safe(lkb, safe, &ls->ls_cb_delay, lkb_cb_list) {
 		list_del_init(&lkb->lkb_cb_list);
 		queue_work(ls->ls_callback_wq, &lkb->lkb_cb_work);
@@ -256,7 +256,7 @@ void dlm_callback_resume(struct dlm_ls *ls)
 	empty = list_empty(&ls->ls_cb_delay);
 	if (empty)
 		clear_bit(LSFL_CB_DELAY, &ls->ls_flags);
-	spin_unlock(&ls->ls_cb_lock);
+	spin_unlock_bh(&ls->ls_cb_lock);
 
 	sum += count;
 	if (!empty) {
fs/dlm/debug_fs.c +16 -16
index 26abeff5b602..ca65481df70d 100644
@@ -372,7 +372,7 @@ static void print_format5_lock(struct seq_file *s, struct dlm_lkb *lkb)
 
 	/* lkb_id lkb_flags mode flags sb_status sb_flags */
 
-	spin_lock(&lkb->lkb_cb_lock);
+	spin_lock_bh(&lkb->lkb_cb_lock);
 	list_for_each_entry(cb, &lkb->lkb_callbacks, list) {
 		seq_printf(s, "%x %x %d %x %d %x\n",
 			   lkb->lkb_id,
@@ -382,7 +382,7 @@ static void print_format5_lock(struct seq_file *s, struct dlm_lkb *lkb)
 			   cb->sb_status,
 			   cb->sb_flags);
 	}
-	spin_unlock(&lkb->lkb_cb_lock);
+	spin_unlock_bh(&lkb->lkb_cb_lock);
 }
 
 static void print_format5(struct dlm_rsb *r, struct seq_file *s)
@@ -508,7 +508,7 @@ static void *table_seq_start(struct seq_file *seq, loff_t *pos)
 
 	tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
 
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
+	spin_lock_bh(&ls->ls_rsbtbl[bucket].lock);
 	if (!RB_EMPTY_ROOT(tree)) {
 		for (node = rb_first(tree); node; node = rb_next(node)) {
 			r = rb_entry(node, struct dlm_rsb, res_hashnode);
@@ -516,12 +516,12 @@ static void *table_seq_start(struct seq_file *seq, loff_t *pos)
 				dlm_hold_rsb(r);
 				ri->rsb = r;
 				ri->bucket = bucket;
-				spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+				spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 				return ri;
 			}
 		}
 	}
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+	spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 
 	/*
 	 * move to the first rsb in the next non-empty bucket
@@ -540,18 +540,18 @@ static void *table_seq_start(struct seq_file *seq, loff_t *pos)
 		}
 		tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
 
-		spin_lock(&ls->ls_rsbtbl[bucket].lock);
+		spin_lock_bh(&ls->ls_rsbtbl[bucket].lock);
 		if (!RB_EMPTY_ROOT(tree)) {
 			node = rb_first(tree);
 			r = rb_entry(node, struct dlm_rsb, res_hashnode);
 			dlm_hold_rsb(r);
 			ri->rsb = r;
 			ri->bucket = bucket;
-			spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+			spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 			*pos = n;
 			return ri;
 		}
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+		spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 	}
 }
 
@@ -572,7 +572,7 @@ static void *table_seq_next(struct seq_file *seq, void *iter_ptr, loff_t *pos)
 	 * move to the next rsb in the same bucket
 	 */
 
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
+	spin_lock_bh(&ls->ls_rsbtbl[bucket].lock);
 	rp = ri->rsb;
 	next = rb_next(&rp->res_hashnode);
 
@@ -580,12 +580,12 @@ static void *table_seq_next(struct seq_file *seq, void *iter_ptr, loff_t *pos)
 		r = rb_entry(next, struct dlm_rsb, res_hashnode);
 		dlm_hold_rsb(r);
 		ri->rsb = r;
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+		spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 		dlm_put_rsb(rp);
 		++*pos;
 		return ri;
 	}
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+	spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 	dlm_put_rsb(rp);
 
 	/*
@@ -606,18 +606,18 @@ static void *table_seq_next(struct seq_file *seq, void *iter_ptr, loff_t *pos)
 		}
 		tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
 
-		spin_lock(&ls->ls_rsbtbl[bucket].lock);
+		spin_lock_bh(&ls->ls_rsbtbl[bucket].lock);
 		if (!RB_EMPTY_ROOT(tree)) {
 			next = rb_first(tree);
 			r = rb_entry(next, struct dlm_rsb, res_hashnode);
 			dlm_hold_rsb(r);
 			ri->rsb = r;
 			ri->bucket = bucket;
-			spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+			spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 			*pos = n;
 			return ri;
 		}
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+		spin_unlock_bh(&ls->ls_rsbtbl[bucket].lock);
 	}
 }
 
@@ -824,7 +824,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 
 	mutex_lock(&debug_buf_lock);
 	dlm_lock_recovery(ls);
-	spin_lock(&ls->ls_waiters_lock);
+	spin_lock_bh(&ls->ls_waiters_lock);
 	memset(debug_buf, 0, sizeof(debug_buf));
 
 	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
@@ -835,7 +835,7 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 			break;
 		pos += ret;
 	}
-	spin_unlock(&ls->ls_waiters_lock);
+	spin_unlock_bh(&ls->ls_waiters_lock);
 	dlm_unlock_recovery(ls);
 
 	rv = simple_read_from_buffer(userbuf, count, ppos, debug_buf, pos);
fs/dlm/lock.c +92 -82

608 lines not on this page · raw message

fs/dlm/lock.h +2 -2
index 0f6b2700c0da..45a74869810a 100644
@@ -69,12 +69,12 @@ static inline int is_master(struct dlm_rsb *r)
 
 static inline void lock_rsb(struct dlm_rsb *r)
 {
-	spin_lock(&r->res_lock);
+	spin_lock_bh(&r->res_lock);
 }
 
 static inline void unlock_rsb(struct dlm_rsb *r)
 {
-	spin_unlock(&r->res_lock);
+	spin_unlock_bh(&r->res_lock);
 }
 
 #endif
fs/dlm/lockspace.c +25 -26

178 lines not on this page · raw message

fs/dlm/lowcomms.c +8 -8
index db71982d709d..28dd74aebc84 100644
@@ -867,36 +867,36 @@ static void process_dlm_messages(struct work_struct *work)
 {
 	struct processqueue_entry *pentry;
 
-	spin_lock(&processqueue_lock);
+	spin_lock_bh(&processqueue_lock);
 	pentry = list_first_entry_or_null(&processqueue,
 					  struct processqueue_entry, list);
 	if (WARN_ON_ONCE(!pentry)) {
 		process_dlm_messages_pending = false;
-		spin_unlock(&processqueue_lock);
+		spin_unlock_bh(&processqueue_lock);
 		return;
 	}
 
 	list_del(&pentry->list);
 	atomic_dec(&processqueue_count);
-	spin_unlock(&processqueue_lock);
+	spin_unlock_bh(&processqueue_lock);
 
 	for (;;) {
 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
 					    pentry->buflen);
 		free_processqueue_entry(pentry);
 
-		spin_lock(&processqueue_lock);
+		spin_lock_bh(&processqueue_lock);
 		pentry = list_first_entry_or_null(&processqueue,
 						  struct processqueue_entry, list);
 		if (!pentry) {
 			process_dlm_messages_pending = false;
-			spin_unlock(&processqueue_lock);
+			spin_unlock_bh(&processqueue_lock);
 			break;
 		}
 
 		list_del(&pentry->list);
 		atomic_dec(&processqueue_count);
-		spin_unlock(&processqueue_lock);
+		spin_unlock_bh(&processqueue_lock);
 	}
 }
 
@@ -966,14 +966,14 @@ static int receive_from_sock(struct connection *con, int buflen)
 	memmove(con->rx_leftover_buf, pentry->buf + ret,
 		con->rx_leftover);
 
-	spin_lock(&processqueue_lock);
+	spin_lock_bh(&processqueue_lock);
 	ret = atomic_inc_return(&processqueue_count);
 	list_add_tail(&pentry->list, &processqueue);
 	if (!process_dlm_messages_pending) {
 		process_dlm_messages_pending = true;
 		queue_work(process_workqueue, &process_work);
 	}
-	spin_unlock(&processqueue_lock);
+	spin_unlock_bh(&processqueue_lock);
 
 	if (ret > DLM_MAX_PROCESS_BUFFERS)
 		return DLM_IO_FLUSH;
fs/dlm/member.c +11 -11
index ac1b555af9d6..6401916a97ef 100644
@@ -630,7 +630,7 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * message to the requestqueue without races.
 	 */
 
-	write_lock(&ls->ls_recv_active);
+	write_lock_bh(&ls->ls_recv_active);
 
 	/*
 	 * Abort any recovery that's in progress (see RECOVER_STOP,
@@ -638,23 +638,23 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
 	 */
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
 	ls->ls_recover_seq++;
 
 	/* activate requestqueue and stop processing */
-	write_lock(&ls->ls_requestqueue_lock);
+	write_lock_bh(&ls->ls_requestqueue_lock);
 	set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
-	write_unlock(&ls->ls_requestqueue_lock);
-	spin_unlock(&ls->ls_recover_lock);
+	write_unlock_bh(&ls->ls_requestqueue_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	/*
 	 * Let dlm_recv run again, now any normal messages will be saved on the
 	 * requestqueue for later.
 	 */
 
-	write_unlock(&ls->ls_recv_active);
+	write_unlock_bh(&ls->ls_recv_active);
 
 	/*
 	 * This in_recovery lock does two things:
@@ -679,13 +679,13 @@ int dlm_ls_stop(struct dlm_ls *ls)
 
 	dlm_recoverd_suspend(ls);
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	kfree(ls->ls_slots);
 	ls->ls_slots = NULL;
 	ls->ls_num_slots = 0;
 	ls->ls_slots_size = 0;
 	ls->ls_recover_status = 0;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	dlm_recoverd_resume(ls);
 
@@ -719,12 +719,12 @@ int dlm_ls_start(struct dlm_ls *ls)
 	if (error < 0)
 		goto fail_rv;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 
 	/* the lockspace needs to be stopped before it can be started */
 
 	if (!dlm_locking_stopped(ls)) {
-		spin_unlock(&ls->ls_recover_lock);
+		spin_unlock_bh(&ls->ls_recover_lock);
 		log_error(ls, "start ignored: lockspace running");
 		error = -EINVAL;
 		goto fail;
@@ -735,7 +735,7 @@ int dlm_ls_start(struct dlm_ls *ls)
 	rv->seq = ++ls->ls_recover_seq;
 	rv_old = ls->ls_recover_args;
 	ls->ls_recover_args = rv;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (rv_old) {
 		log_error(ls, "unused recovery %llx %d",
fs/dlm/midcomms.c +20 -20
index ea0559e2a44d..7d75f86450d6 100644
@@ -365,9 +365,9 @@ int dlm_midcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 	node->users = 0;
 	midcomms_node_reset(node);
 
-	spin_lock(&nodes_lock);
+	spin_lock_bh(&nodes_lock);
 	hlist_add_head_rcu(&node->hlist, &node_hash[r]);
-	spin_unlock(&nodes_lock);
+	spin_unlock_bh(&nodes_lock);
 
 	node->debugfs = dlm_create_debug_comms_file(nodeid, node);
 	return 0;
@@ -478,7 +478,7 @@ static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
 
 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
 {
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive passive fin ack from node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 
@@ -492,13 +492,13 @@ static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
 		wake_up(&node->shutdown_wait);
 		break;
 	default:
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		log_print("%s: unexpected state: %d",
 			  __func__, node->state);
 		WARN_ON_ONCE(1);
 		return;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 }
 
 static void dlm_receive_buffer_3_2_trace(uint32_t seq,
@@ -535,7 +535,7 @@ static void dlm_midcomms_receive_buffer(const union dlm_packet *p,
 	if (is_expected_seq) {
 		switch (p->header.h_cmd) {
 		case DLM_FIN:
-			spin_lock(&node->state_lock);
+			spin_lock_bh(&node->state_lock);
 			pr_debug("receive fin msg from node %d with state %s\n",
 				 node->nodeid, dlm_state_str(node->state));
 
@@ -576,13 +576,13 @@ static void dlm_midcomms_receive_buffer(const union dlm_packet *p,
 				/* probably remove_member caught it, do nothing */
 				break;
 			default:
-				spin_unlock(&node->state_lock);
+				spin_unlock_bh(&node->state_lock);
 				log_print("%s: unexpected state: %d",
 					  __func__, node->state);
 				WARN_ON_ONCE(1);
 				return;
 			}
-			spin_unlock(&node->state_lock);
+			spin_unlock_bh(&node->state_lock);
 			break;
 		default:
 			WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
@@ -1183,7 +1183,7 @@ void dlm_midcomms_exit(void)
 
 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
 {
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive active fin ack from node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 
@@ -1203,13 +1203,13 @@ static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
 		wake_up(&node->shutdown_wait);
 		break;
 	default:
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		log_print("%s: unexpected state: %d",
 			  __func__, node->state);
 		WARN_ON_ONCE(1);
 		return;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 }
 
 void dlm_midcomms_add_member(int nodeid)
@@ -1224,7 +1224,7 @@ void dlm_midcomms_add_member(int nodeid)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	if (!node->users) {
 		pr_debug("receive add member from node %d with state %s\n",
 			 node->nodeid, dlm_state_str(node->state));
@@ -1252,7 +1252,7 @@ void dlm_midcomms_add_member(int nodeid)
 
 	node->users++;
 	pr_debug("node %d users inc count %d\n", nodeid, node->users);
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	srcu_read_unlock(&nodes_srcu, idx);
 }
@@ -1270,13 +1270,13 @@ void dlm_midcomms_remove_member(int nodeid)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	/* case of dlm_midcomms_addr() created node but
 	 * was not added before because dlm_midcomms_close()
 	 * removed the node
 	 */
 	if (!node->users) {
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		srcu_read_unlock(&nodes_srcu, idx);
 		return;
 	}
@@ -1314,7 +1314,7 @@ void dlm_midcomms_remove_member(int nodeid)
 			break;
 		}
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	srcu_read_unlock(&nodes_srcu, idx);
 }
@@ -1352,7 +1352,7 @@ static void midcomms_shutdown(struct midcomms_node *node)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive active shutdown for node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 	switch (node->state) {
@@ -1371,7 +1371,7 @@ static void midcomms_shutdown(struct midcomms_node *node)
 		 */
 		break;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	if (DLM_DEBUG_FENCE_TERMINATION)
 		msleep(5000);
@@ -1442,9 +1442,9 @@ int dlm_midcomms_close(int nodeid)
 	ret = dlm_lowcomms_close(nodeid);
 	dlm_delete_debug_comms_file(node->debugfs);
 
-	spin_lock(&nodes_lock);
+	spin_lock_bh(&nodes_lock);
 	hlist_del_rcu(&node->hlist);
-	spin_unlock(&nodes_lock);
+	spin_unlock_bh(&nodes_lock);
 	srcu_read_unlock(&nodes_srcu, idx);
 
 	/* wait that all readers left until flush send queue */
fs/dlm/rcom.c +13 -13
index 2e3f529f3ff2..be1a71a6303a 100644
@@ -143,18 +143,18 @@ static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
 
 static void allow_sync_reply(struct dlm_ls *ls, __le64 *new_seq)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	*new_seq = cpu_to_le64(++ls->ls_rcom_seq);
 	set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 static void disallow_sync_reply(struct dlm_ls *ls)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 	clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 /*
@@ -245,10 +245,10 @@ static void receive_rcom_status(struct dlm_ls *ls,
 		goto do_create;
 	}
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	status = ls->ls_recover_status;
 	num_slots = ls->ls_num_slots;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 	len += num_slots * sizeof(struct rcom_slot);
 
  do_create:
@@ -266,9 +266,9 @@ static void receive_rcom_status(struct dlm_ls *ls,
 	if (!num_slots)
 		goto do_send;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	if (ls->ls_num_slots != num_slots) {
-		spin_unlock(&ls->ls_recover_lock);
+		spin_unlock_bh(&ls->ls_recover_lock);
 		log_debug(ls, "receive_rcom_status num_slots %d to %d",
 			  num_slots, ls->ls_num_slots);
 		rc->rc_result = 0;
@@ -277,7 +277,7 @@ static void receive_rcom_status(struct dlm_ls *ls,
 	}
 
 	dlm_slots_copy_out(ls, rc);
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
  do_send:
 	send_rcom_stateless(msg, rc);
@@ -285,7 +285,7 @@ static void receive_rcom_status(struct dlm_ls *ls,
 
 static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
 	    le64_to_cpu(rc_in->rc_id) != ls->ls_rcom_seq) {
 		log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
@@ -301,7 +301,7 @@ static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in)
 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 	wake_up(&ls->ls_wait_general);
  out:
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name,
@@ -613,11 +613,11 @@ void dlm_receive_rcom(struct dlm_ls *ls, const struct dlm_rcom *rc, int nodeid)
 		break;
 	}
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	status = ls->ls_recover_status;
 	stop = dlm_recovery_stopped(ls);
 	seq = ls->ls_recover_seq;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (stop && (rc->rc_type != cpu_to_le32(DLM_RCOM_STATUS)))
 		goto ignore;
fs/dlm/recover.c +28 -28

191 lines not on this page · raw message

fs/dlm/recoverd.c +6 -6
index b3c9b704883c..67d8e84c2db2 100644
@@ -32,9 +32,9 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 {
 	int error = -EINTR;
 
-	write_lock(&ls->ls_recv_active);
+	write_lock_bh(&ls->ls_recv_active);
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	if (ls->ls_recover_seq == seq) {
 		set_bit(LSFL_RUNNING, &ls->ls_flags);
 		/* unblocks processes waiting to enter the dlm */
@@ -42,9 +42,9 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 		clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
 		error = 0;
 	}
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
-	write_unlock(&ls->ls_recv_active);
+	write_unlock_bh(&ls->ls_recv_active);
 	return error;
 }
 
@@ -268,12 +268,12 @@ static void do_ls_recovery(struct dlm_ls *ls)
 	struct dlm_recover *rv = NULL;
 	int error;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	rv = ls->ls_recover_args;
 	ls->ls_recover_args = NULL;
 	if (rv && ls->ls_recover_seq == rv->seq)
 		clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (rv) {
 		error = ls_recover(ls, rv);
fs/dlm/requestqueue.c +6 -6
index 9b646026df46..719a5243a069 100644
@@ -68,7 +68,7 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 	struct dlm_message *ms;
 	int error = 0;
 
-	write_lock(&ls->ls_requestqueue_lock);
+	write_lock_bh(&ls->ls_requestqueue_lock);
 	for (;;) {
 		if (list_empty(&ls->ls_requestqueue)) {
 			clear_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
@@ -96,11 +96,11 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 			error = -EINTR;
 			break;
 		}
-		write_unlock(&ls->ls_requestqueue_lock);
+		write_unlock_bh(&ls->ls_requestqueue_lock);
 		schedule();
-		write_lock(&ls->ls_requestqueue_lock);
+		write_lock_bh(&ls->ls_requestqueue_lock);
 	}
-	write_unlock(&ls->ls_requestqueue_lock);
+	write_unlock_bh(&ls->ls_requestqueue_lock);
 
 	return error;
 }
@@ -135,7 +135,7 @@ void dlm_purge_requestqueue(struct dlm_ls *ls)
 	struct dlm_message *ms;
 	struct rq_entry *e, *safe;
 
-	write_lock(&ls->ls_requestqueue_lock);
+	write_lock_bh(&ls->ls_requestqueue_lock);
 	list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
 		ms =  &e->request;
 
@@ -144,6 +144,6 @@ void dlm_purge_requestqueue(struct dlm_ls *ls)
 			kfree(e);
 		}
 	}
-	write_unlock(&ls->ls_requestqueue_lock);
+	write_unlock_bh(&ls->ls_requestqueue_lock);
 }
 
fs/dlm/user.c +17 -17
index 695e691b38b3..c0d35678ee54 100644
@@ -206,7 +206,7 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 		return;
 
 	ls = lkb->lkb_resource->res_ls;
-	spin_lock(&ls->ls_clear_proc_locks);
+	spin_lock_bh(&ls->ls_clear_proc_locks);
 
 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
@@ -228,12 +228,12 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
 		set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags);
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 
 	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
 	switch (rv) {
 	case DLM_ENQUEUE_CALLBACK_FAILURE:
-		spin_unlock(&proc->asts_spin);
+		spin_unlock_bh(&proc->asts_spin);
 		WARN_ON_ONCE(1);
 		goto out;
 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
@@ -247,19 +247,19 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 		WARN_ON_ONCE(1);
 		break;
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 
 	if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
 		/* N.B. spin_lock locks_spin, not asts_spin */
-		spin_lock(&proc->locks_spin);
+		spin_lock_bh(&proc->locks_spin);
 		if (!list_empty(&lkb->lkb_ownqueue)) {
 			list_del_init(&lkb->lkb_ownqueue);
 			dlm_put_lkb(lkb);
 		}
-		spin_unlock(&proc->locks_spin);
+		spin_unlock_bh(&proc->locks_spin);
 	}
  out:
-	spin_unlock(&ls->ls_clear_proc_locks);
+	spin_unlock_bh(&ls->ls_clear_proc_locks);
 }
 
 static int device_user_lock(struct dlm_user_proc *proc,
@@ -832,10 +832,10 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
 		return -EINVAL;
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 	if (list_empty(&proc->asts)) {
 		if (file->f_flags & O_NONBLOCK) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			return -EAGAIN;
 		}
 
@@ -844,16 +844,16 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 	repeat:
 		set_current_state(TASK_INTERRUPTIBLE);
 		if (list_empty(&proc->asts) && !signal_pending(current)) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			schedule();
-			spin_lock(&proc->asts_spin);
+			spin_lock_bh(&proc->asts_spin);
 			goto repeat;
 		}
 		set_current_state(TASK_RUNNING);
 		remove_wait_queue(&proc->wait, &wait);
 
 		if (signal_pending(current)) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			return -ERESTARTSYS;
 		}
 	}
@@ -875,7 +875,7 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 		 */
 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
 		list_del_init(&lkb->lkb_cb_list);
-		spin_unlock(&proc->asts_spin);
+		spin_unlock_bh(&proc->asts_spin);
 		/* removes ref for proc->asts, may cause lkb to be freed */
 		dlm_put_lkb(lkb);
 		WARN_ON_ONCE(1);
@@ -890,7 +890,7 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 		WARN_ON_ONCE(1);
 		break;
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 
 	if (cb->flags & DLM_CB_BAST) {
 		trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb->mode);
@@ -925,12 +925,12 @@ static __poll_t device_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &proc->wait, wait);
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 	if (!list_empty(&proc->asts)) {
-		spin_unlock(&proc->asts_spin);
+		spin_unlock_bh(&proc->asts_spin);
 		return EPOLLIN | EPOLLRDNORM;
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 	return 0;
 }
 
Signature · 1 line
2.39.3
author maintainer
#10 lore raw

PATCH 10/10 dlm: do dlm message processing in softirq context

In reply to Alexander Aring

This patch moves the dlm message processing from a ordered workqueue context to a ordered softirq context. Later we want to call the user defined ast/bast callbacks directly inside the dlm message processing context instead of doing an additional context switch to the exisiting callback workqueue. This should slightly improve the dlm message parsing behaviour. There are two main reasons why to change to this behaviour:

1. Bringing the ast/callback callback to softirq context that the use is aware it should not block into this context. Later patches will introduce a per lockspace flag to signal that the user is capable to handling these callbacks in softirq context to solve backwards compatibility.

2. We can easily switch to a per dlm instance concurrent dlm message parsing when DLM is ready to handle it. This instance could be in e.g. per lockspace instance or more fine granulatiry instance such as per lock instance.

Futher patches will unveil more improvements to switch to a per message softirq parsing context. Especially if we getting DLM in a state that we can allow concurrent message parsing.

---
 fs/dlm/lowcomms.c | 34 ++++++++++++----------------------
 1 file changed, 12 insertions(+), 22 deletions(-)
fs/dlm/lowcomms.c +12 -22
index 28dd74aebc84..b9de8d5b61b7 100644
@@ -183,7 +183,6 @@ static int dlm_local_count;
 
 /* Work queues */
 static struct workqueue_struct *io_workqueue;
-static struct workqueue_struct *process_workqueue;
 
 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 static DEFINE_SPINLOCK(connections_lock);
@@ -199,9 +198,9 @@ static const struct dlm_proto_ops *dlm_proto_ops;
 
 static void process_recv_sockets(struct work_struct *work);
 static void process_send_sockets(struct work_struct *work);
-static void process_dlm_messages(struct work_struct *work);
+static void process_dlm_messages(struct tasklet_struct *tasklet);
 
-static DECLARE_WORK(process_work, process_dlm_messages);
+static DECLARE_TASKLET_DISABLED(process_tasklet, process_dlm_messages);
 static DEFINE_SPINLOCK(processqueue_lock);
 static bool process_dlm_messages_pending;
 static atomic_t processqueue_count;
@@ -863,7 +862,7 @@ struct dlm_processed_nodes {
 	struct list_head list;
 };
 
-static void process_dlm_messages(struct work_struct *work)
+static void process_dlm_messages(struct tasklet_struct *tasklet)
 {
 	struct processqueue_entry *pentry;
 
@@ -971,7 +970,7 @@ static int receive_from_sock(struct connection *con, int buflen)
 	list_add_tail(&pentry->list, &processqueue);
 	if (!process_dlm_messages_pending) {
 		process_dlm_messages_pending = true;
-		queue_work(process_workqueue, &process_work);
+		tasklet_schedule(&process_tasklet);
 	}
 	spin_unlock_bh(&processqueue_lock);
 
@@ -1511,7 +1510,8 @@ static void process_recv_sockets(struct work_struct *work)
 		/* CF_RECV_PENDING cleared */
 		break;
 	case DLM_IO_FLUSH:
-		flush_workqueue(process_workqueue);
+		tasklet_disable(&process_tasklet);
+		tasklet_enable(&process_tasklet);
 		fallthrough;
 	case DLM_IO_RESCHED:
 		cond_resched();
@@ -1686,10 +1686,7 @@ static void work_stop(void)
 		io_workqueue = NULL;
 	}
 
-	if (process_workqueue) {
-		destroy_workqueue(process_workqueue);
-		process_workqueue = NULL;
-	}
+	tasklet_disable(&process_tasklet);
 }
 
 static int work_start(void)
@@ -1701,17 +1698,7 @@ static int work_start(void)
 		return -ENOMEM;
 	}
 
-	/* ordered dlm message process queue,
-	 * should be converted to a tasklet
-	 */
-	process_workqueue = alloc_ordered_workqueue("dlm_process",
-						    WQ_HIGHPRI | WQ_MEM_RECLAIM);
-	if (!process_workqueue) {
-		log_print("can't start dlm_process");
-		destroy_workqueue(io_workqueue);
-		io_workqueue = NULL;
-		return -ENOMEM;
-	}
+	tasklet_enable(&process_tasklet);
 
 	return 0;
 }
@@ -1734,7 +1721,10 @@ void dlm_lowcomms_shutdown(void)
 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
 			shutdown_connection(con, true);
 			stop_connection_io(con);
-			flush_workqueue(process_workqueue);
+
+			tasklet_disable(&process_tasklet);
+			tasklet_enable(&process_tasklet);
+
 			close_connection(con, true);
 
 			clean_one_writequeue(con);
Signature · 1 line
2.39.3