libceph: fix ceph_msg_revoke()

There are a number of problems with revoking a "was sending" message:

(1) We never make any attempt to revoke data - only kvecs contibute to
con->out_skip.  However, once the header (envelope) is written to the
socket, our peer learns data_len and sets itself to expect at least
data_len bytes to follow front or front+middle.  If ceph_msg_revoke()
is called while the messenger is sending message's data portion,
anything we send after that call is counted by the OSD towards the now
revoked message's data portion.  The effects vary, the most common one
is the eventual hang - higher layers get stuck waiting for the reply to
the message that was sent out after ceph_msg_revoke() returned and
treated by the OSD as a bunch of data bytes.  This is what Matt ran
into.

(2) Flat out zeroing con->out_kvec_bytes worth of bytes to handle kvecs
is wrong.  If ceph_msg_revoke() is called before the tag is sent out or
while the messenger is sending the header, we will get a connection
reset, either due to a bad tag (0 is not a valid tag) or a bad header
CRC, which kind of defeats the purpose of revoke.  Currently the kernel
client refuses to work with header CRCs disabled, but that will likely
change in the future, making this even worse.

(3) con->out_skip is not reset on connection reset, leading to one or
more spurious connection resets if we happen to get a real one between
con->out_skip is set in ceph_msg_revoke() and before it's cleared in
write_partial_skip().

Fixing (1) and (3) is trivial.  The idea behind fixing (2) is to never
zero the tag or the header, i.e. send out tag+header regardless of when
ceph_msg_revoke() is called.  That way the header is always correct, no
unnecessary resets are induced and revoke stands ready for disabled
CRCs.  Since ceph_msg_revoke() rips out con->out_msg, introduce a new
"message out temp" and copy the header into it before sending.

Cc: stable@vger.kernel.org # 4.0+
Reported-by: NMatt Conner <matt.conner@keepertech.com>
Signed-off-by: NIlya Dryomov <idryomov@gmail.com>
Tested-by: NMatt Conner <matt.conner@keepertech.com>
Reviewed-by: NSage Weil <sage@redhat.com>

include/linux/ceph/messenger.h

@@ -220,6 +220,7 @@ struct ceph_connection {
 	struct ceph_entity_addr actual_peer_addr;
 
 	/* message out temps */
+	struct ceph_msg_header out_hdr;
 	struct ceph_msg *out_msg;        /* sending message (== tail of
 					    out_sent) */
 	bool out_msg_done;
@@ -229,7 +230,6 @@ struct ceph_connection {
 	int out_kvec_left;   /* kvec's left in out_kvec */
 	int out_skip;        /* skip this many bytes */
 	int out_kvec_bytes;  /* total bytes left */
-	bool out_kvec_is_msg; /* kvec refers to out_msg */
 	int out_more;        /* there is more data after the kvecs */
 	__le64 out_temp_ack; /* for writing an ack */
 	struct ceph_timespec out_temp_keepalive2; /* for writing keepalive2

net/ceph/messenger.c

@@ -669,6 +669,8 @@ static void reset_connection(struct ceph_connection *con)
 	}
 	con->in_seq = 0;
 	con->in_seq_acked = 0;
+
+	con->out_skip = 0;
 }
 
 /*
@@ -768,6 +770,8 @@ static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
 
 static void con_out_kvec_reset(struct ceph_connection *con)
 {
+	BUG_ON(con->out_skip);
+
 	con->out_kvec_left = 0;
 	con->out_kvec_bytes = 0;
 	con->out_kvec_cur = &con->out_kvec[0];
@@ -776,9 +780,9 @@ static void con_out_kvec_reset(struct ceph_connection *con)
 static void con_out_kvec_add(struct ceph_connection *con,
 				size_t size, void *data)
 {
-	int index;
+	int index = con->out_kvec_left;
 
-	index = con->out_kvec_left;
+	BUG_ON(con->out_skip);
 	BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
 
 	con->out_kvec[index].iov_len = size;
@@ -787,6 +791,27 @@ static void con_out_kvec_add(struct ceph_connection *con,
 	con->out_kvec_bytes += size;
 }
 
+/*
+ * Chop off a kvec from the end.  Return residual number of bytes for
+ * that kvec, i.e. how many bytes would have been written if the kvec
+ * hadn't been nuked.
+ */
+static int con_out_kvec_skip(struct ceph_connection *con)
+{
+	int off = con->out_kvec_cur - con->out_kvec;
+	int skip = 0;
+
+	if (con->out_kvec_bytes > 0) {
+		skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
+		BUG_ON(con->out_kvec_bytes < skip);
+		BUG_ON(!con->out_kvec_left);
+		con->out_kvec_bytes -= skip;
+		con->out_kvec_left--;
+	}
+
+	return skip;
+}
+
 #ifdef CONFIG_BLOCK
 
 /*
@@ -1194,7 +1219,6 @@ static void prepare_write_message_footer(struct ceph_connection *con)
 	m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
 
 	dout("prepare_write_message_footer %p\n", con);
-	con->out_kvec_is_msg = true;
 	con->out_kvec[v].iov_base = &m->footer;
 	if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
 		if (con->ops->sign_message)
@@ -1222,7 +1246,6 @@ static void prepare_write_message(struct ceph_connection *con)
 	u32 crc;
 
 	con_out_kvec_reset(con);
-	con->out_kvec_is_msg = true;
 	con->out_msg_done = false;
 
 	/* Sneak an ack in there first?  If we can get it into the same
@@ -1262,18 +1285,19 @@ static void prepare_write_message(struct ceph_connection *con)
 
 	/* tag + hdr + front + middle */
 	con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
-	con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
+	con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
 	con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
 
 	if (m->middle)
 		con_out_kvec_add(con, m->middle->vec.iov_len,
 			m->middle->vec.iov_base);
 
-	/* fill in crc (except data pages), footer */
+	/* fill in hdr crc and finalize hdr */
 	crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
 	con->out_msg->hdr.crc = cpu_to_le32(crc);
-	con->out_msg->footer.flags = 0;
+	memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
 
+	/* fill in front and middle crc, footer */
 	crc = crc32c(0, m->front.iov_base, m->front.iov_len);
 	con->out_msg->footer.front_crc = cpu_to_le32(crc);
 	if (m->middle) {
@@ -1285,6 +1309,7 @@ static void prepare_write_message(struct ceph_connection *con)
 	dout("%s front_crc %u middle_crc %u\n", __func__,
 	     le32_to_cpu(con->out_msg->footer.front_crc),
 	     le32_to_cpu(con->out_msg->footer.middle_crc));
+	con->out_msg->footer.flags = 0;
 
 	/* is there a data payload? */
 	con->out_msg->footer.data_crc = 0;
@@ -1489,7 +1514,6 @@ static int write_partial_kvec(struct ceph_connection *con)
 		}
 	}
 	con->out_kvec_left = 0;
-	con->out_kvec_is_msg = false;
 	ret = 1;
 out:
 	dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
@@ -1581,6 +1605,7 @@ static int write_partial_skip(struct ceph_connection *con)
 {
 	int ret;
 
+	dout("%s %p %d left\n", __func__, con, con->out_skip);
 	while (con->out_skip > 0) {
 		size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
 
@@ -2503,13 +2528,13 @@ static int try_write(struct ceph_connection *con)
 
 more_kvec:
 	/* kvec data queued? */
-	if (con->out_skip) {
-		ret = write_partial_skip(con);
+	if (con->out_kvec_left) {
+		ret = write_partial_kvec(con);
 		if (ret <= 0)
 			goto out;
 	}
-	if (con->out_kvec_left) {
-		ret = write_partial_kvec(con);
+	if (con->out_skip) {
+		ret = write_partial_skip(con);
 		if (ret <= 0)
 			goto out;
 	}
@@ -3047,16 +3072,31 @@ void ceph_msg_revoke(struct ceph_msg *msg)
 		ceph_msg_put(msg);
 	}
 	if (con->out_msg == msg) {
-		dout("%s %p msg %p - was sending\n", __func__, con, msg);
-		con->out_msg = NULL;
-		if (con->out_kvec_is_msg) {
-			con->out_skip = con->out_kvec_bytes;
-			con->out_kvec_is_msg = false;
+		BUG_ON(con->out_skip);
+		/* footer */
+		if (con->out_msg_done) {
+			con->out_skip += con_out_kvec_skip(con);
+		} else {
+			BUG_ON(!msg->data_length);
+			if (con->peer_features & CEPH_FEATURE_MSG_AUTH)
+				con->out_skip += sizeof(msg->footer);
+			else
+				con->out_skip += sizeof(msg->old_footer);
 		}
+		/* data, middle, front */
+		if (msg->data_length)
+			con->out_skip += msg->cursor.total_resid;
+		if (msg->middle)
+			con->out_skip += con_out_kvec_skip(con);
+		con->out_skip += con_out_kvec_skip(con);
+
+		dout("%s %p msg %p - was sending, will write %d skip %d\n",
+		     __func__, con, msg, con->out_kvec_bytes, con->out_skip);
 		msg->hdr.seq = 0;
-
+		con->out_msg = NULL;
 		ceph_msg_put(msg);
 	}
+
 	mutex_unlock(&con->mutex);
 }