libceph: preallocate message data items

Currently message data items are allocated with ceph_msg_data_create()
in setup_request_data() inside send_request().  send_request() has never
been allowed to fail, so each allocation is followed by a BUG_ON:

  data = ceph_msg_data_create(...);
  BUG_ON(!data);

It's been this way since support for multiple message data items was
added in commit 6644ed7b7e04 ("libceph: make message data be a pointer")
in 3.10.

There is no reason to delay the allocation of message data items until
the last possible moment and we certainly don't need a linked list of
them as they are only ever appended to the end and never erased.  Make
ceph_msg_new2() take max_data_items and adapt the rest of the code.

Reported-by: Jerry Lee <leisurelysw24@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

3 files changed, 144 insertions, 89 deletions

diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
index 76684edc43ef..88e35830198c 100644
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -156,7 +156,6 @@ static bool con_flag_test_and_set(struct ceph_connection *con,
 /* Slab caches for frequently-allocated structures */
 
 static struct kmem_cache	*ceph_msg_cache;
-static struct kmem_cache	*ceph_msg_data_cache;
 
 /* static tag bytes (protocol control messages) */
 static char tag_msg = CEPH_MSGR_TAG_MSG;
@@ -235,23 +234,11 @@ static int ceph_msgr_slab_init(void)
 	if (!ceph_msg_cache)
 		return -ENOMEM;
 
-	BUG_ON(ceph_msg_data_cache);
-	ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
-	if (ceph_msg_data_cache)
-		return 0;
-
-	kmem_cache_destroy(ceph_msg_cache);
-	ceph_msg_cache = NULL;
-
-	return -ENOMEM;
+	return 0;
 }
 
 static void ceph_msgr_slab_exit(void)
 {
-	BUG_ON(!ceph_msg_data_cache);
-	kmem_cache_destroy(ceph_msg_data_cache);
-	ceph_msg_data_cache = NULL;
-
 	BUG_ON(!ceph_msg_cache);
 	kmem_cache_destroy(ceph_msg_cache);
 	ceph_msg_cache = NULL;
@@ -1141,16 +1128,13 @@ static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
 {
 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
-	struct ceph_msg_data *data;
 
 	BUG_ON(!length);
 	BUG_ON(length > msg->data_length);
-	BUG_ON(list_empty(&msg->data));
+	BUG_ON(!msg->num_data_items);
 
-	cursor->data_head = &msg->data;
 	cursor->total_resid = length;
-	data = list_first_entry(&msg->data, struct ceph_msg_data, links);
-	cursor->data = data;
+	cursor->data = msg->data;
 
 	__ceph_msg_data_cursor_init(cursor);
 }
@@ -1231,8 +1215,7 @@ static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
 
 	if (!cursor->resid && cursor->total_resid) {
 		WARN_ON(!cursor->last_piece);
-		BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
-		cursor->data = list_next_entry(cursor->data, links);
+		cursor->data++;
 		__ceph_msg_data_cursor_init(cursor);
 		new_piece = true;
 	}
@@ -1248,9 +1231,6 @@ static size_t sizeof_footer(struct ceph_connection *con)
 
 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
 {
-	BUG_ON(!msg);
-	BUG_ON(!data_len);
-
 	/* Initialize data cursor */
 
 	ceph_msg_data_cursor_init(msg, (size_t)data_len);
@@ -1590,7 +1570,7 @@ static int write_partial_message_data(struct ceph_connection *con)
 
 	dout("%s %p msg %p\n", __func__, con, msg);
 
-	if (list_empty(&msg->data))
+	if (!msg->num_data_items)
 		return -EINVAL;
 
 	/*
@@ -2347,8 +2327,7 @@ static int read_partial_msg_data(struct ceph_connection *con)
 	u32 crc = 0;
 	int ret;
 
-	BUG_ON(!msg);
-	if (list_empty(&msg->data))
+	if (!msg->num_data_items)
 		return -EIO;
 
 	if (do_datacrc)
@@ -3256,32 +3235,16 @@ bool ceph_con_keepalive_expired(struct ceph_connection *con,
 	return false;
 }
 
-static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
+static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
 {
-	struct ceph_msg_data *data;
-
-	if (WARN_ON(!ceph_msg_data_type_valid(type)))
-		return NULL;
-
-	data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
-	if (!data)
-		return NULL;
-
-	data->type = type;
-	INIT_LIST_HEAD(&data->links);
-
-	return data;
+	BUG_ON(msg->num_data_items >= msg->max_data_items);
+	return &msg->data[msg->num_data_items++];
 }
 
 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
 {
-	if (!data)
-		return;
-
-	WARN_ON(!list_empty(&data->links));
 	if (data->type == CEPH_MSG_DATA_PAGELIST)
 		ceph_pagelist_release(data->pagelist);
-	kmem_cache_free(ceph_msg_data_cache, data);
 }
 
 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
@@ -3292,13 +3255,12 @@ void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
 	BUG_ON(!pages);
 	BUG_ON(!length);
 
-	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
-	BUG_ON(!data);
+	data = ceph_msg_data_add(msg);
+	data->type = CEPH_MSG_DATA_PAGES;
 	data->pages = pages;
 	data->length = length;
 	data->alignment = alignment & ~PAGE_MASK;
 
-	list_add_tail(&data->links, &msg->data);
 	msg->data_length += length;
 }
 EXPORT_SYMBOL(ceph_msg_data_add_pages);
@@ -3311,12 +3273,11 @@ void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
 	BUG_ON(!pagelist);
 	BUG_ON(!pagelist->length);
 
-	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
-	BUG_ON(!data);
+	data = ceph_msg_data_add(msg);
+	data->type = CEPH_MSG_DATA_PAGELIST;
 	refcount_inc(&pagelist->refcnt);
 	data->pagelist = pagelist;
 
-	list_add_tail(&data->links, &msg->data);
 	msg->data_length += pagelist->length;
 }
 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
@@ -3327,12 +3288,11 @@ void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
 {
 	struct ceph_msg_data *data;
 
-	data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
-	BUG_ON(!data);
+	data = ceph_msg_data_add(msg);
+	data->type = CEPH_MSG_DATA_BIO;
 	data->bio_pos = *bio_pos;
 	data->bio_length = length;
 
-	list_add_tail(&data->links, &msg->data);
 	msg->data_length += length;
 }
 EXPORT_SYMBOL(ceph_msg_data_add_bio);
@@ -3343,11 +3303,10 @@ void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
 {
 	struct ceph_msg_data *data;
 
-	data = ceph_msg_data_create(CEPH_MSG_DATA_BVECS);
-	BUG_ON(!data);
+	data = ceph_msg_data_add(msg);
+	data->type = CEPH_MSG_DATA_BVECS;
 	data->bvec_pos = *bvec_pos;
 
-	list_add_tail(&data->links, &msg->data);
 	msg->data_length += bvec_pos->iter.bi_size;
 }
 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
@@ -3356,8 +3315,8 @@ EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
  * construct a new message with given type, size
  * the new msg has a ref count of 1.
  */
-struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
-			      bool can_fail)
+struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
+			       gfp_t flags, bool can_fail)
 {
 	struct ceph_msg *m;
 
@@ -3371,7 +3330,6 @@ struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 
 	INIT_LIST_HEAD(&m->list_head);
 	kref_init(&m->kref);
-	INIT_LIST_HEAD(&m->data);
 
 	/* front */
 	if (front_len) {
@@ -3386,6 +3344,15 @@ struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 	}
 	m->front_alloc_len = m->front.iov_len = front_len;
 
+	if (max_data_items) {
+		m->data = kmalloc_array(max_data_items, sizeof(*m->data),
+					flags);
+		if (!m->data)
+			goto out2;
+
+		m->max_data_items = max_data_items;
+	}
+
 	dout("ceph_msg_new %p front %d\n", m, front_len);
 	return m;
 
@@ -3402,6 +3369,13 @@ out:
 	}
 	return NULL;
 }
+EXPORT_SYMBOL(ceph_msg_new2);
+
+struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
+			      bool can_fail)
+{
+	return ceph_msg_new2(type, front_len, 0, flags, can_fail);
+}
 EXPORT_SYMBOL(ceph_msg_new);
 
 /*
@@ -3497,13 +3471,14 @@ static void ceph_msg_free(struct ceph_msg *m)
 {
 	dout("%s %p\n", __func__, m);
 	kvfree(m->front.iov_base);
+	kfree(m->data);
 	kmem_cache_free(ceph_msg_cache, m);
 }
 
 static void ceph_msg_release(struct kref *kref)
 {
 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
-	struct ceph_msg_data *data, *next;
+	int i;
 
 	dout("%s %p\n", __func__, m);
 	WARN_ON(!list_empty(&m->list_head));
@@ -3516,11 +3491,8 @@ static void ceph_msg_release(struct kref *kref)
 		m->middle = NULL;
 	}
 
-	list_for_each_entry_safe(data, next, &m->data, links) {
-		list_del_init(&data->links);
-		ceph_msg_data_destroy(data);
-	}
-	m->data_length = 0;
+	for (i = 0; i < m->num_data_items; i++)
+		ceph_msg_data_destroy(&m->data[i]);
 
 	if (m->pool)
 		ceph_msgpool_put(m->pool, m);
diff --git a/net/ceph/msgpool.c b/net/ceph/msgpool.c
index 3dddc074f0d7..e3ecb80cd182 100644
--- a/net/ceph/msgpool.c
+++ b/net/ceph/msgpool.c
@@ -14,7 +14,8 @@ static void *msgpool_alloc(gfp_t gfp_mask, void *arg)
 	struct ceph_msgpool *pool = arg;
 	struct ceph_msg *msg;
 
-	msg = ceph_msg_new(pool->type, pool->front_len, gfp_mask, true);
+	msg = ceph_msg_new2(pool->type, pool->front_len, pool->max_data_items,
+			    gfp_mask, true);
 	if (!msg) {
 		dout("msgpool_alloc %s failed\n", pool->name);
 	} else {
@@ -35,11 +36,13 @@ static void msgpool_free(void *element, void *arg)
 }
 
 int ceph_msgpool_init(struct ceph_msgpool *pool, int type,
-		      int front_len, int size, bool blocking, const char *name)
+		      int front_len, int max_data_items, int size,
+		      const char *name)
 {
 	dout("msgpool %s init\n", name);
 	pool->type = type;
 	pool->front_len = front_len;
+	pool->max_data_items = max_data_items;
 	pool->pool = mempool_create(size, msgpool_alloc, msgpool_free, pool);
 	if (!pool->pool)
 		return -ENOMEM;
@@ -53,18 +56,21 @@ void ceph_msgpool_destroy(struct ceph_msgpool *pool)
 	mempool_destroy(pool->pool);
 }
 
-struct ceph_msg *ceph_msgpool_get(struct ceph_msgpool *pool,
-				  int front_len)
+struct ceph_msg *ceph_msgpool_get(struct ceph_msgpool *pool, int front_len,
+				  int max_data_items)
 {
 	struct ceph_msg *msg;
 
-	if (front_len > pool->front_len) {
-		dout("msgpool_get %s need front %d, pool size is %d\n",
-		       pool->name, front_len, pool->front_len);
+	if (front_len > pool->front_len ||
+	    max_data_items > pool->max_data_items) {
+		pr_warn_ratelimited("%s need %d/%d, pool %s has %d/%d\n",
+		    __func__, front_len, max_data_items, pool->name,
+		    pool->front_len, pool->max_data_items);
 		WARN_ON_ONCE(1);
 
 		/* try to alloc a fresh message */
-		return ceph_msg_new(pool->type, front_len, GFP_NOFS, false);
+		return ceph_msg_new2(pool->type, front_len, max_data_items,
+				     GFP_NOFS, false);
 	}
 
 	msg = mempool_alloc(pool->pool, GFP_NOFS);
@@ -80,6 +86,9 @@ void ceph_msgpool_put(struct ceph_msgpool *pool, struct ceph_msg *msg)
 	msg->front.iov_len = pool->front_len;
 	msg->hdr.front_len = cpu_to_le32(pool->front_len);
 
+	msg->data_length = 0;
+	msg->num_data_items = 0;
+
 	kref_init(&msg->kref);  /* retake single ref */
 	mempool_free(msg, pool->pool);
 }
diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c
index 7ac7f21ff317..cf0bd2cce848 100644
--- a/net/ceph/osd_client.c
+++ b/net/ceph/osd_client.c
@@ -614,12 +614,15 @@ static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
 	return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
 }
 
-int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
+static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
+				      int num_request_data_items,
+				      int num_reply_data_items)
 {
 	struct ceph_osd_client *osdc = req->r_osdc;
 	struct ceph_msg *msg;
 	int msg_size;
 
+	WARN_ON(req->r_request || req->r_reply);
 	WARN_ON(ceph_oid_empty(&req->r_base_oid));
 	WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
 
@@ -641,9 +644,11 @@ int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 	msg_size += 4 + 8; /* retry_attempt, features */
 
 	if (req->r_mempool)
-		msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size);
+		msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
+				       num_request_data_items);
 	else
-		msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true);
+		msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
+				    num_request_data_items, gfp, true);
 	if (!msg)
 		return -ENOMEM;
 
@@ -656,9 +661,11 @@ int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 	msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
 
 	if (req->r_mempool)
-		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size);
+		msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
+				       num_reply_data_items);
 	else
-		msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true);
+		msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
+				    num_reply_data_items, gfp, true);
 	if (!msg)
 		return -ENOMEM;
 
@@ -666,7 +673,6 @@ int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
 
 	return 0;
 }
-EXPORT_SYMBOL(ceph_osdc_alloc_messages);
 
 static bool osd_req_opcode_valid(u16 opcode)
 {
@@ -679,6 +685,64 @@ __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 	}
 }
 
+static void get_num_data_items(struct ceph_osd_request *req,
+			       int *num_request_data_items,
+			       int *num_reply_data_items)
+{
+	struct ceph_osd_req_op *op;
+
+	*num_request_data_items = 0;
+	*num_reply_data_items = 0;
+
+	for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
+		switch (op->op) {
+		/* request */
+		case CEPH_OSD_OP_WRITE:
+		case CEPH_OSD_OP_WRITEFULL:
+		case CEPH_OSD_OP_SETXATTR:
+		case CEPH_OSD_OP_CMPXATTR:
+		case CEPH_OSD_OP_NOTIFY_ACK:
+			*num_request_data_items += 1;
+			break;
+
+		/* reply */
+		case CEPH_OSD_OP_STAT:
+		case CEPH_OSD_OP_READ:
+		case CEPH_OSD_OP_LIST_WATCHERS:
+			*num_reply_data_items += 1;
+			break;
+
+		/* both */
+		case CEPH_OSD_OP_NOTIFY:
+			*num_request_data_items += 1;
+			*num_reply_data_items += 1;
+			break;
+		case CEPH_OSD_OP_CALL:
+			*num_request_data_items += 2;
+			*num_reply_data_items += 1;
+			break;
+
+		default:
+			WARN_ON(!osd_req_opcode_valid(op->op));
+			break;
+		}
+	}
+}
+
+/*
+ * oid, oloc and OSD op opcode(s) must be filled in before this function
+ * is called.
+ */
+int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
+{
+	int num_request_data_items, num_reply_data_items;
+
+	get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
+	return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
+					  num_reply_data_items);
+}
+EXPORT_SYMBOL(ceph_osdc_alloc_messages);
+
 /*
  * This is an osd op init function for opcodes that have no data or
  * other information associated with them.  It also serves as a
@@ -1035,7 +1099,15 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
 	if (flags & CEPH_OSD_FLAG_WRITE)
 		req->r_data_offset = off;
 
-	r = ceph_osdc_alloc_messages(req, GFP_NOFS);
+	if (num_ops > 1)
+		/*
+		 * This is a special case for ceph_writepages_start(), but it
+		 * also covers ceph_uninline_data().  If more multi-op request
+		 * use cases emerge, we will need a separate helper.
+		 */
+		r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
+	else
+		r = ceph_osdc_alloc_messages(req, GFP_NOFS);
 	if (r)
 		goto fail;
 
@@ -1842,13 +1914,16 @@ static bool should_plug_request(struct ceph_osd_request *req)
 	return true;
 }
 
+/*
+ * Keep get_num_data_items() in sync with this function.
+ */
 static void setup_request_data(struct ceph_osd_request *req,
 			       struct ceph_msg *msg)
 {
 	u32 data_len = 0;
 	int i;
 
-	if (!list_empty(&msg->data))
+	if (msg->num_data_items)
 		return;
 
 	WARN_ON(msg->data_length);
@@ -4325,9 +4400,7 @@ static void handle_watch_notify(struct ceph_osd_client *osdc,
 			     lreq->notify_id, notify_id);
 		} else if (!completion_done(&lreq->notify_finish_wait)) {
 			struct ceph_msg_data *data =
-			    list_first_entry_or_null(&msg->data,
-						     struct ceph_msg_data,
-						     links);
+			    msg->num_data_items ? &msg->data[0] : NULL;
 
 			if (data) {
 				if (lreq->preply_pages) {
@@ -5036,11 +5109,12 @@ int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
 		goto out_map;
 
 	err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
-				PAGE_SIZE, 10, true, "osd_op");
+				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
 	if (err < 0)
 		goto out_mempool;
 	err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
-				PAGE_SIZE, 10, true, "osd_op_reply");
+				PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
+				"osd_op_reply");
 	if (err < 0)
 		goto out_msgpool;
 
@@ -5310,7 +5384,7 @@ static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
 	u32 front_len = le32_to_cpu(hdr->front_len);
 	u32 data_len = le32_to_cpu(hdr->data_len);
 
-	m = ceph_msg_new(type, front_len, GFP_NOIO, false);
+	m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
 	if (!m)
 		return NULL;