path: root/0050-tools-xenstore-limit-outstanding-requests.patch

From 49344fb86ff040bae1107e236592c2d4dc4607f3 Mon Sep 17 00:00:00 2001
From: Juergen Gross <jgross@suse.com>
Date: Tue, 13 Sep 2022 07:35:08 +0200
Subject: [PATCH 50/87] tools/xenstore: limit outstanding requests

Add another quota for limiting the number of outstanding requests of a
guest. As the way to specify quotas on the command line is becoming
rather nasty, switch to a new scheme using [--quota|-Q] <what>=<val>
allowing to add more quotas in future easily.

Set the default value to 20 (basically a random value not seeming to
be too high or too low).

A request is said to be outstanding if any message generated by this
request (the direct response plus potential watch events) is not yet
completely stored into a ring buffer. The initial watch event sent as
a result of registering a watch is an exception.

Note that across a live update the relation to buffered watch events
for other domains is lost.

Use talloc_zero() for allocating the domain structure in order to have
all per-domain quota zeroed initially.

This is part of XSA-326 / CVE-2022-42312.

Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
(cherry picked from commit 36de433a273f55d614c83b89c9a8972287a1e475)
---
 tools/xenstore/xenstored_core.c   | 88 +++++++++++++++++++++++++++++--
 tools/xenstore/xenstored_core.h   | 20 ++++++-
 tools/xenstore/xenstored_domain.c | 38 ++++++++++---
 tools/xenstore/xenstored_domain.h |  3 ++
 tools/xenstore/xenstored_watch.c  | 15 ++++--
 5 files changed, 150 insertions(+), 14 deletions(-)

diff --git a/tools/xenstore/xenstored_core.c b/tools/xenstore/xenstored_core.c
index 45244c021cd3..488d540f3a32 100644
--- a/tools/xenstore/xenstored_core.c
+++ b/tools/xenstore/xenstored_core.c
@@ -107,6 +107,7 @@ int quota_max_entry_size = 2048; /* 2K */
 int quota_max_transaction = 10;
 int quota_nb_perms_per_node = 5;
 int quota_max_path_len = XENSTORE_REL_PATH_MAX;
+int quota_req_outstanding = 20;
 
 unsigned int timeout_watch_event_msec = 20000;
 
@@ -223,12 +224,24 @@ static uint64_t get_now_msec(void)
 	return now_ts.tv_sec * 1000 + now_ts.tv_nsec / 1000000;
 }
 
+/*
+ * Remove a struct buffered_data from the list of outgoing data.
+ * A struct buffered_data related to a request having caused watch events to be
+ * sent is kept until all those events have been written out.
+ * Each watch event is referencing the related request via pend.req, while the
+ * number of watch events caused by a request is kept in pend.ref.event_cnt
+ * (those two cases are mutually exclusive, so the two fields can share memory
+ * via a union).
+ * The struct buffered_data is freed only if no related watch event is
+ * referencing it. The related return data can be freed right away.
+ */
 static void free_buffered_data(struct buffered_data *out,
 			       struct connection *conn)
 {
 	struct buffered_data *req;
 
 	list_del(&out->list);
+	out->on_out_list = false;
 
 	/*
 	 * Update conn->timeout_msec with the next found timeout value in the
@@ -244,6 +257,30 @@ static void free_buffered_data(struct buffered_data *out,
 		}
 	}
 
+	if (out->hdr.msg.type == XS_WATCH_EVENT) {
+		req = out->pend.req;
+		if (req) {
+			req->pend.ref.event_cnt--;
+			if (!req->pend.ref.event_cnt && !req->on_out_list) {
+				if (req->on_ref_list) {
+					domain_outstanding_domid_dec(
+						req->pend.ref.domid);
+					list_del(&req->list);
+				}
+				talloc_free(req);
+			}
+		}
+	} else if (out->pend.ref.event_cnt) {
+		/* Hang out off from conn. */
+		talloc_steal(NULL, out);
+		if (out->buffer != out->default_buffer)
+			talloc_free(out->buffer);
+		list_add(&out->list, &conn->ref_list);
+		out->on_ref_list = true;
+		return;
+	} else
+		domain_outstanding_dec(conn);
+
 	talloc_free(out);
 }
 
@@ -405,6 +442,7 @@ int delay_request(struct connection *conn, struct buffered_data *in,
 static int destroy_conn(void *_conn)
 {
 	struct connection *conn = _conn;
+	struct buffered_data *req;
 
 	/* Flush outgoing if possible, but don't block. */
 	if (!conn->domain) {
@@ -418,6 +456,11 @@ static int destroy_conn(void *_conn)
 				break;
 		close(conn->fd);
 	}
+
+	conn_free_buffered_data(conn);
+	list_for_each_entry(req, &conn->ref_list, list)
+		req->on_ref_list = false;
+
         if (conn->target)
                 talloc_unlink(conn, conn->target);
 	list_del(&conn->list);
@@ -893,6 +936,8 @@ void send_reply(struct connection *conn, enum xsd_sockmsg_type type,
 
 	/* Queue for later transmission. */
 	list_add_tail(&bdata->list, &conn->out_list);
+	bdata->on_out_list = true;
+	domain_outstanding_inc(conn);
 }
 
 /*
@@ -900,7 +945,8 @@ void send_reply(struct connection *conn, enum xsd_sockmsg_type type,
  * As this is not directly related to the current command, errors can't be
  * reported.
  */
-void send_event(struct connection *conn, const char *path, const char *token)
+void send_event(struct buffered_data *req, struct connection *conn,
+		const char *path, const char *token)
 {
 	struct buffered_data *bdata;
 	unsigned int len;
@@ -930,8 +976,13 @@ void send_event(struct connection *conn, const char *path, const char *token)
 			conn->timeout_msec = bdata->timeout_msec;
 	}
 
+	bdata->pend.req = req;
+	if (req)
+		req->pend.ref.event_cnt++;
+
 	/* Queue for later transmission. */
 	list_add_tail(&bdata->list, &conn->out_list);
+	bdata->on_out_list = true;
 }
 
 /* Some routines (write, mkdir, etc) just need a non-error return */
@@ -1740,6 +1791,7 @@ static void handle_input(struct connection *conn)
 			return;
 	}
 	in = conn->in;
+	in->pend.ref.domid = conn->id;
 
 	/* Not finished header yet? */
 	if (in->inhdr) {
@@ -1808,6 +1860,7 @@ struct connection *new_connection(const struct interface_funcs *funcs)
 	new->is_stalled = false;
 	new->transaction_started = 0;
 	INIT_LIST_HEAD(&new->out_list);
+	INIT_LIST_HEAD(&new->ref_list);
 	INIT_LIST_HEAD(&new->watches);
 	INIT_LIST_HEAD(&new->transaction_list);
 	INIT_LIST_HEAD(&new->delayed);
@@ -2286,6 +2339,9 @@ static void usage(void)
 "  -t, --transaction <nb>  limit the number of transaction allowed per domain,\n"
 "  -A, --perm-nb <nb>      limit the number of permissions per node,\n"
 "  -M, --path-max <chars>  limit the allowed Xenstore node path length,\n"
+"  -Q, --quota <what>=<nb> set the quota <what> to the value <nb>, allowed\n"
+"                          quotas are:\n"
+"                          outstanding: number of outstanding requests\n"
 "  -w, --timeout <what>=<seconds>   set the timeout in seconds for <what>,\n"
 "                          allowed timeout candidates are:\n"
 "                          watch-event: time a watch-event is kept pending\n"
@@ -2311,6 +2367,7 @@ static struct option options[] = {
 	{ "transaction", 1, NULL, 't' },
 	{ "perm-nb", 1, NULL, 'A' },
 	{ "path-max", 1, NULL, 'M' },
+	{ "quota", 1, NULL, 'Q' },
 	{ "timeout", 1, NULL, 'w' },
 	{ "no-recovery", 0, NULL, 'R' },
 	{ "internal-db", 0, NULL, 'I' },
@@ -2358,6 +2415,20 @@ static void set_timeout(const char *arg)
 		barf("unknown timeout \"%s\"\n", arg);
 }
 
+static void set_quota(const char *arg)
+{
+	const char *eq = strchr(arg, '=');
+	int val;
+
+	if (!eq)
+		barf("quotas must be specified via <what>=<nb>\n");
+	val = get_optval_int(eq + 1);
+	if (what_matches(arg, "outstanding"))
+		quota_req_outstanding = val;
+	else
+		barf("unknown quota \"%s\"\n", arg);
+}
+
 int main(int argc, char *argv[])
 {
 	int opt;
@@ -2372,8 +2443,8 @@ int main(int argc, char *argv[])
 	orig_argc = argc;
 	orig_argv = argv;
 
-	while ((opt = getopt_long(argc, argv, "DE:F:HNPS:t:A:M:T:RVW:w:U", options,
-				  NULL)) != -1) {
+	while ((opt = getopt_long(argc, argv, "DE:F:HNPS:t:A:M:Q:T:RVW:w:U",
+				  options, NULL)) != -1) {
 		switch (opt) {
 		case 'D':
 			no_domain_init = true;
@@ -2422,6 +2493,9 @@ int main(int argc, char *argv[])
 			quota_max_path_len = min(XENSTORE_REL_PATH_MAX,
 						 quota_max_path_len);
 			break;
+		case 'Q':
+			set_quota(optarg);
+			break;
 		case 'w':
 			set_timeout(optarg);
 			break;
@@ -2875,6 +2949,14 @@ static void add_buffered_data(struct buffered_data *bdata,
 
 	/* Queue for later transmission. */
 	list_add_tail(&bdata->list, &conn->out_list);
+	bdata->on_out_list = true;
+	/*
+	 * Watch events are never "outstanding", but the request causing them
+	 * are instead kept "outstanding" until all watch events caused by that
+	 * request have been delivered.
+	 */
+	if (bdata->hdr.msg.type != XS_WATCH_EVENT)
+		domain_outstanding_inc(conn);
 }
 
 void read_state_buffered_data(const void *ctx, struct connection *conn,
diff --git a/tools/xenstore/xenstored_core.h b/tools/xenstore/xenstored_core.h
index 8a81fc693f01..db09f463a657 100644
--- a/tools/xenstore/xenstored_core.h
+++ b/tools/xenstore/xenstored_core.h
@@ -56,6 +56,8 @@ struct xs_state_connection;
 struct buffered_data
 {
 	struct list_head list;
+	bool on_out_list;
+	bool on_ref_list;
 
 	/* Are we still doing the header? */
 	bool inhdr;
@@ -63,6 +65,17 @@ struct buffered_data
 	/* How far are we? */
 	unsigned int used;
 
+	/* Outstanding request accounting. */
+	union {
+		/* ref is being used for requests. */
+		struct {
+			unsigned int event_cnt; /* # of outstanding events. */
+			unsigned int domid;     /* domid of request. */
+		} ref;
+		/* req is being used for watch events. */
+		struct buffered_data *req;      /* request causing event. */
+	} pend;
+
 	union {
 		struct xsd_sockmsg msg;
 		char raw[sizeof(struct xsd_sockmsg)];
@@ -123,6 +136,9 @@ struct connection
 	struct list_head out_list;
 	uint64_t timeout_msec;
 
+	/* Referenced requests no longer pending. */
+	struct list_head ref_list;
+
 	/* Transaction context for current request (NULL if none). */
 	struct transaction *transaction;
 
@@ -191,7 +207,8 @@ unsigned int get_string(const struct buffered_data *data, unsigned int offset);
 
 void send_reply(struct connection *conn, enum xsd_sockmsg_type type,
 		const void *data, unsigned int len);
-void send_event(struct connection *conn, const char *path, const char *token);
+void send_event(struct buffered_data *req, struct connection *conn,
+		const char *path, const char *token);
 
 /* Some routines (write, mkdir, etc) just need a non-error return */
 void send_ack(struct connection *conn, enum xsd_sockmsg_type type);
@@ -247,6 +264,7 @@ extern int dom0_domid;
 extern int dom0_event;
 extern int priv_domid;
 extern int quota_nb_entry_per_domain;
+extern int quota_req_outstanding;
 
 extern unsigned int timeout_watch_event_msec;
 
diff --git a/tools/xenstore/xenstored_domain.c b/tools/xenstore/xenstored_domain.c
index 93c4c1edcdd1..850085a92c76 100644
--- a/tools/xenstore/xenstored_domain.c
+++ b/tools/xenstore/xenstored_domain.c
@@ -78,6 +78,9 @@ struct domain
 	/* number of watch for this domain */
 	int nbwatch;
 
+	/* Number of outstanding requests. */
+	int nboutstanding;
+
 	/* write rate limit */
 	wrl_creditt wrl_credit; /* [ -wrl_config_writecost, +_dburst ] */
 	struct wrl_timestampt wrl_timestamp;
@@ -183,8 +186,12 @@ static bool domain_can_read(struct connection *conn)
 {
 	struct xenstore_domain_interface *intf = conn->domain->interface;
 
-	if (domain_is_unprivileged(conn) && conn->domain->wrl_credit < 0)
-		return false;
+	if (domain_is_unprivileged(conn)) {
+		if (conn->domain->wrl_credit < 0)
+			return false;
+		if (conn->domain->nboutstanding >= quota_req_outstanding)
+			return false;
+	}
 
 	return (intf->req_cons != intf->req_prod);
 }
@@ -331,7 +338,7 @@ static struct domain *alloc_domain(const void *context, unsigned int domid)
 {
 	struct domain *domain;
 
-	domain = talloc(context, struct domain);
+	domain = talloc_zero(context, struct domain);
 	if (!domain) {
 		errno = ENOMEM;
 		return NULL;
@@ -392,9 +399,6 @@ static int new_domain(struct domain *domain, int port, bool restore)
 	domain->conn->domain = domain;
 	domain->conn->id = domain->domid;
 
-	domain->nbentry = 0;
-	domain->nbwatch = 0;
-
 	return 0;
 }
 
@@ -938,6 +942,28 @@ int domain_watch(struct connection *conn)
 		: 0;
 }
 
+void domain_outstanding_inc(struct connection *conn)
+{
+	if (!conn || !conn->domain)
+		return;
+	conn->domain->nboutstanding++;
+}
+
+void domain_outstanding_dec(struct connection *conn)
+{
+	if (!conn || !conn->domain)
+		return;
+	conn->domain->nboutstanding--;
+}
+
+void domain_outstanding_domid_dec(unsigned int domid)
+{
+	struct domain *d = find_domain_by_domid(domid);
+
+	if (d)
+		d->nboutstanding--;
+}
+
 static wrl_creditt wrl_config_writecost      = WRL_FACTOR;
 static wrl_creditt wrl_config_rate           = WRL_RATE   * WRL_FACTOR;
 static wrl_creditt wrl_config_dburst         = WRL_DBURST * WRL_FACTOR;
diff --git a/tools/xenstore/xenstored_domain.h b/tools/xenstore/xenstored_domain.h
index 1e929b8f8c6f..4f51b005291a 100644
--- a/tools/xenstore/xenstored_domain.h
+++ b/tools/xenstore/xenstored_domain.h
@@ -64,6 +64,9 @@ int domain_entry(struct connection *conn);
 void domain_watch_inc(struct connection *conn);
 void domain_watch_dec(struct connection *conn);
 int domain_watch(struct connection *conn);
+void domain_outstanding_inc(struct connection *conn);
+void domain_outstanding_dec(struct connection *conn);
+void domain_outstanding_domid_dec(unsigned int domid);
 
 /* Special node permission handling. */
 int set_perms_special(struct connection *conn, const char *name,
diff --git a/tools/xenstore/xenstored_watch.c b/tools/xenstore/xenstored_watch.c
index 205d9d8ea116..0755ffa375ba 100644
--- a/tools/xenstore/xenstored_watch.c
+++ b/tools/xenstore/xenstored_watch.c
@@ -142,6 +142,7 @@ void fire_watches(struct connection *conn, const void *ctx, const char *name,
 		  struct node *node, bool exact, struct node_perms *perms)
 {
 	struct connection *i;
+	struct buffered_data *req;
 	struct watch *watch;
 
 	/* During transactions, don't fire watches, but queue them. */
@@ -150,6 +151,8 @@ void fire_watches(struct connection *conn, const void *ctx, const char *name,
 		return;
 	}
 
+	req = domain_is_unprivileged(conn) ? conn->in : NULL;
+
 	/* Create an event for each watch. */
 	list_for_each_entry(i, &connections, list) {
 		/* introduce/release domain watches */
@@ -164,12 +167,12 @@ void fire_watches(struct connection *conn, const void *ctx, const char *name,
 		list_for_each_entry(watch, &i->watches, list) {
 			if (exact) {
 				if (streq(name, watch->node))
-					send_event(i,
+					send_event(req, i,
 						   get_watch_path(watch, name),
 						   watch->token);
 			} else {
 				if (is_child(name, watch->node))
-					send_event(i,
+					send_event(req, i,
 						   get_watch_path(watch, name),
 						   watch->token);
 			}
@@ -269,8 +272,12 @@ int do_watch(struct connection *conn, struct buffered_data *in)
 	trace_create(watch, "watch");
 	send_ack(conn, XS_WATCH);
 
-	/* We fire once up front: simplifies clients and restart. */
-	send_event(conn, get_watch_path(watch, watch->node), watch->token);
+	/*
+	 * We fire once up front: simplifies clients and restart.
+	 * This event will not be linked to the XS_WATCH request.
+	 */
+	send_event(NULL, conn, get_watch_path(watch, watch->node),
+		   watch->token);
 
 	return 0;
 }
-- 
2.37.4