From bc4cf523839420bf8778ae28819e17c92687f087 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Mon, 14 Jan 2008 18:22:05 -0800
Subject: [PATCH] reworked monmap

---
 src/kernel/client.c     | 24 ++++++++++++----------
 src/kernel/mon_client.c | 44 ++++++++++++++++++++++++-----------------
 src/kernel/mon_client.h |  8 ++++----
 src/kernel/super.c      |  5 +----
 src/kernel/super.h      |  1 -
 5 files changed, 45 insertions(+), 37 deletions(-)

diff --git a/src/kernel/client.c b/src/kernel/client.c
index 0f03a88eae07b..14953ba13eaab 100644
--- a/src/kernel/client.c
+++ b/src/kernel/client.c
@@ -82,7 +82,8 @@ static struct ceph_client *create_client(struct ceph_mount_args *args)
 	cl->msgr->prepare_pages = ceph_osdc_prepare_pages;
 	
 	cl->whoami = -1;
-	ceph_monc_init(&cl->monc, cl);
+	if ((err = ceph_monc_init(&cl->monc, cl)) < 0)
+		goto fail;
 	ceph_mdsc_init(&cl->mdsc, cl);
 	ceph_osdc_init(&cl->osdc, cl);
 
@@ -143,16 +144,19 @@ trymount:
 static void handle_monmap(struct ceph_client *client, struct ceph_msg *msg)
 {
 	int err;
-	int first = (client->monc.monmap.epoch == 0);
-
-	dout(1, "handle_monmap had epoch %d\n", client->monc.monmap.epoch);
+	int first = (client->monc.monmap->epoch == 0);
+	void *new;
 
-	/* parse */
-	err = ceph_monmap_decode(&client->monc.monmap, 
-				 msg->front.iov_base, 
+	dout(1, "handle_monmap had epoch %d\n", client->monc.monmap->epoch);
+	new = ceph_monmap_decode(msg->front.iov_base, 
 				 msg->front.iov_base + msg->front.iov_len);
-	if (err != 0)
+	if (IS_ERR(new)) {
+		err = PTR_ERR(new);
+		derr(0, "problem decoding monmap, %d\n", err);
 		return;
+	}
+	kfree(client->monc.monmap);
+	client->monc.monmap = new;
 
 	if (first) {
 		client->whoami = msg->hdr.dst.name.num;
@@ -242,9 +246,9 @@ void ceph_dispatch(void *p, struct ceph_msg *msg)
 	switch (msg->hdr.type) {
 		/* me */
 	case CEPH_MSG_MON_MAP:
-		had = client->monc.monmap.epoch ? 1:0;
+		had = client->monc.monmap->epoch ? 1:0;
 		handle_monmap(client, msg);
-		if (!had && client->monc.monmap.epoch)
+		if (!had && client->monc.monmap->epoch)
 			got_first_map(client, 0);
 		break;
 
diff --git a/src/kernel/mon_client.c b/src/kernel/mon_client.c
index e83ee0b93459c..38aa2cd5f7041 100644
--- a/src/kernel/mon_client.c
+++ b/src/kernel/mon_client.c
@@ -9,12 +9,15 @@ int ceph_mon_debug = 50;
 #include "super.h"
 
 
-int ceph_monmap_decode(struct ceph_monmap *m, void *p, void *end)
+struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
 {
-	int err;
-	void *old;
+	struct ceph_monmap *m;
+	int i, err;
 
 	dout(30, "monmap_decode %p %p\n", p, end);
+	m = kmalloc(end-p, GFP_KERNEL);
+	if (m == NULL)
+		return ERR_PTR(-ENOMEM);
 
 	if ((err = ceph_decode_32(&p, end, &m->epoch)) < 0)
 		goto bad;
@@ -23,25 +26,26 @@ int ceph_monmap_decode(struct ceph_monmap *m, void *p, void *end)
 	if ((err = ceph_decode_64(&p, end, &m->fsid.minor)) < 0)
 		goto bad;
 	if ((err = ceph_decode_32(&p, end, &m->num_mon)) < 0)
-		return err;
-
-	old = m->mon_inst;
-	m->mon_inst = kmalloc(m->num_mon*sizeof(*m->mon_inst), GFP_KERNEL);
-	if (m->mon_inst == NULL) {
-		m->mon_inst = old;
-		return -ENOMEM;
-	}
-	kfree(old);
+		goto bad;
 
-	if ((err = ceph_decode_copy(&p, end, m->mon_inst, m->num_mon*sizeof(m->mon_inst[0]))) < 0)
+	if (p + m->num_mon*sizeof(m->mon_inst[0]) != end) { 
+		err = -EINVAL;
 		goto bad;
+	}
+	memcpy(m->mon_inst, p, end-p);
+	for (i=0; i<m->num_mon; i++) {
+		ceph_decode_inst(&m->mon_inst[i]);
+		dout(30, "monmap_decode mon%d is %x:%d\n", i,
+		     ntohl(m->mon_inst[i].addr.ipaddr.sin_addr.s_addr), 
+		     ntohs(m->mon_inst[i].addr.ipaddr.sin_port));
+	}
 
 	dout(30, "monmap_decode got epoch %d, num_mon %d\n", m->epoch, m->num_mon);
-	return 0;
+	return m;
 
 bad:
 	dout(30, "monmap_decode failed with %d\n", err);
-	return err;
+	return ERR_PTR(err);
 }
 
 /*
@@ -63,18 +67,22 @@ static int pick_mon(struct ceph_mon_client *monc, int notmon)
 	if (notmon < 0 && monc->last_mon >= 0)
 		return monc->last_mon;
 	get_random_bytes(&r, 1);
-	monc->last_mon = r % monc->monmap.num_mon;
+	monc->last_mon = r % monc->monmap->num_mon;
 	return monc->last_mon;
 }
 
 
-void ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
+int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
 {
 	dout(5, "ceph_monc_init\n");
 	memset(monc, 0, sizeof(*monc));
 	monc->client = cl;
+	monc->monmap = kzalloc(sizeof(struct ceph_monmap), GFP_KERNEL);
+	if (monc->monmap == NULL) 
+		return -ENOMEM;
 	spin_lock_init(&monc->lock);
 	INIT_RADIX_TREE(&monc->statfs_request_tree, GFP_KERNEL);
+	return 0;
 }
 
 
@@ -135,7 +143,7 @@ int send_statfs(struct ceph_mon_client *monc, u64 tid)
 	if (IS_ERR(msg))
 		return PTR_ERR(msg);
 	*(__le64*)msg->front.iov_base = cpu_to_le64(tid);
-	msg->hdr.dst = monc->monmap.mon_inst[mon];
+	msg->hdr.dst = monc->monmap->mon_inst[mon];
 	ceph_msg_send(monc->client->msgr, msg, BASE_DELAY_INTERVAL);
 	return 0;
 }
diff --git a/src/kernel/mon_client.h b/src/kernel/mon_client.h
index 16015e25758f6..fc88b4d98e4c2 100644
--- a/src/kernel/mon_client.h
+++ b/src/kernel/mon_client.h
@@ -12,7 +12,7 @@ struct ceph_monmap {
 	ceph_epoch_t epoch;
 	struct ceph_fsid fsid;
 	__u32 num_mon;
-	struct ceph_entity_inst *mon_inst;
+	struct ceph_entity_inst mon_inst[0];
 };
 
 struct ceph_mon_statfs_request {
@@ -25,17 +25,17 @@ struct ceph_mon_statfs_request {
 struct ceph_mon_client {
 	struct ceph_client *client;
 	int last_mon;  /* last monitor i contacted */
-	struct ceph_monmap monmap;
+	struct ceph_monmap *monmap;
 
 	spinlock_t lock;
 	struct radix_tree_root statfs_request_tree;  /* statfs requests */
 	u64 last_tid;
 };
 
-extern int ceph_monmap_decode(struct ceph_monmap *m, void *p, void *end);
+extern struct ceph_monmap *ceph_monmap_decode(void *p, void *end);
 extern int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr);
 
-extern void ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl);
+extern int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl);
 
 extern void ceph_monc_request_mdsmap(struct ceph_mon_client *monc, __u64 have);
 extern void ceph_monc_request_osdmap(struct ceph_mon_client *monc, __u64 have);
diff --git a/src/kernel/super.c b/src/kernel/super.c
index 23bc6daf95bbd..89464a91309a9 100644
--- a/src/kernel/super.c
+++ b/src/kernel/super.c
@@ -78,7 +78,6 @@ static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
 			   args->fsid.major, args->fsid.minor);
 	if (args->flags & CEPH_MOUNT_NOSHARE)
 		seq_puts(m, ",noshare");
-	seq_printf(m, ",monport=%d", args->mon_port);
 	return 0;
 }
 
@@ -236,7 +235,6 @@ static int parse_mount_args(int flags, char *options, const char *dev_name, stru
 	/* defaults */
 	args->mntflags = flags;
 	args->flags = 0;
-	args->mon_port = CEPH_MON_PORT;
 
 	/* ip1[,ip2...]:/server/path */
 	c = strchr(dev_name, ':');
@@ -285,7 +283,6 @@ static int parse_mount_args(int flags, char *options, const char *dev_name, stru
 			break;
 		case Opt_monport:
 			dout(25, "parse_mount_args monport=%d\n", intval);
-			args->mon_port = intval;
 			for (i=0; i<args->num_mon; i++)
 				args->mon_addr[i].ipaddr.sin_port = htons(intval);
 			break;
@@ -479,7 +476,7 @@ static int ceph_compare_super(struct super_block *sb, void *data)
 	} else {
 		/* do we share (a) monitor? */
 		for (i=0; i<args->num_mon; i++)
-			if (ceph_monmap_contains(&other->sb_client->monc.monmap, 
+			if (ceph_monmap_contains(other->sb_client->monc.monmap, 
 						 &args->mon_addr[i]))
 				break;
 		if (i == args->num_mon) {
diff --git a/src/kernel/super.h b/src/kernel/super.h
index 5e10850393df7..74eb404ead181 100644
--- a/src/kernel/super.h
+++ b/src/kernel/super.h
@@ -30,7 +30,6 @@ struct ceph_mount_args {
 	struct ceph_entity_addr my_addr;
 	int num_mon;
 	struct ceph_entity_addr mon_addr[5];
-	int mon_port;
 	char path[100];
 };
 
-- 
2.39.5