Fix: check reference counts for overflow
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tue, 19 Jan 2016 14:51:55 +0000 (09:51 -0500)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tue, 19 Jan 2016 14:51:55 +0000 (09:51 -0500)
Linux kernel CVE-2016-0728 is a use-after-free based on overflow of the
reference counting mechanism.

Implement a kref wrapper in lttng that validates overflows, and use it
instead of kref_get(). Also check explicitly for overflows on file
fcount counters.

This should not be an issue in practice in lttng-modules because the ABI
is only exposed to root, but let's err on the safe side.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
lib/ringbuffer/ring_buffer_frontend.c
lttng-abi.c
lttng-events.c
probes/lttng-kretprobes.c
wrapper/kref.h [new file with mode: 0644]

index c4b797ce8af76faf9ba5209d054b89fb42326656..dbe52c157c71be0ec560bbc6b97dc77d1b1e27ea 100644 (file)
@@ -61,6 +61,7 @@
 #include "../../wrapper/ringbuffer/iterator.h"
 #include "../../wrapper/ringbuffer/nohz.h"
 #include "../../wrapper/atomic.h"
+#include "../../wrapper/kref.h"
 #include "../../wrapper/percpu-defs.h"
 
 /*
@@ -793,7 +794,10 @@ int lib_ring_buffer_open_read(struct lib_ring_buffer *buf)
 
        if (!atomic_long_add_unless(&buf->active_readers, 1, 1))
                return -EBUSY;
-       kref_get(&chan->ref);
+       if (!lttng_kref_get(&chan->ref)) {
+               atomic_long_dec(&buf->active_readers);
+               return -EOVERFLOW;
+       }
        lttng_smp_mb__after_atomic();
        return 0;
 }
index b3de09264783cd69c7996eb53515182aa205768f..54988b79d0821da9ef71d02f845c2345b1ea730f 100644 (file)
@@ -50,6 +50,7 @@
 #include "wrapper/ringbuffer/frontend.h"
 #include "wrapper/poll.h"
 #include "wrapper/file.h"
+#include "wrapper/kref.h"
 #include "lttng-abi.h"
 #include "lttng-abi-old.h"
 #include "lttng-events.h"
@@ -427,6 +428,10 @@ int lttng_abi_create_channel(struct file *session_file,
                transport_name = "<unknown>";
                break;
        }
+       if (atomic_long_add_unless(&session_file->f_count,
+               1, INT_MAX) == INT_MAX) {
+               goto refcount_error;
+       }
        /*
         * We tolerate no failure path after channel creation. It will stay
         * invariant for the rest of the session.
@@ -444,11 +449,12 @@ int lttng_abi_create_channel(struct file *session_file,
        chan->file = chan_file;
        chan_file->private_data = chan;
        fd_install(chan_fd, chan_file);
-       atomic_long_inc(&session_file->f_count);
 
        return chan_fd;
 
 chan_error:
+       atomic_long_dec(&session_file->f_count);
+refcount_error:
        fput(chan_file);
 file_error:
        put_unused_fd(chan_fd);
@@ -951,17 +957,20 @@ int lttng_abi_open_metadata_stream(struct file *channel_file)
                goto notransport;
        }
 
+       if (!lttng_kref_get(&session->metadata_cache->refcount))
+               goto kref_error;
        ret = lttng_abi_create_stream_fd(channel_file, stream_priv,
                        &lttng_metadata_ring_buffer_file_operations);
        if (ret < 0)
                goto fd_error;
 
-       kref_get(&session->metadata_cache->refcount);
        list_add(&metadata_stream->list,
                &session->metadata_cache->metadata_stream);
        return ret;
 
 fd_error:
+       kref_put(&session->metadata_cache->refcount, metadata_cache_destroy);
+kref_error:
        module_put(metadata_stream->transport->owner);
 notransport:
        kfree(metadata_stream);
@@ -1005,6 +1014,11 @@ int lttng_abi_create_event(struct file *channel_file,
                ret = PTR_ERR(event_file);
                goto file_error;
        }
+       /* The event holds a reference on the channel */
+       if (atomic_long_add_unless(&channel_file->f_count,
+               1, INT_MAX) == INT_MAX) {
+               goto refcount_error;
+       }
        if (event_param->instrumentation == LTTNG_KERNEL_TRACEPOINT
                        || event_param->instrumentation == LTTNG_KERNEL_SYSCALL) {
                struct lttng_enabler *enabler;
@@ -1036,11 +1050,11 @@ int lttng_abi_create_event(struct file *channel_file,
        }
        event_file->private_data = priv;
        fd_install(event_fd, event_file);
-       /* The event holds a reference on the channel */
-       atomic_long_inc(&channel_file->f_count);
        return event_fd;
 
 event_error:
+       atomic_long_dec(&channel_file->f_count);
+refcount_error:
        fput(event_file);
 file_error:
        put_unused_fd(event_fd);
index 1f34fbbc3341afd6ce8094c02fba4ed4b818bb70..f0f5d0f80dbe340bf1e4663681ff621acd25f504 100644 (file)
@@ -1085,17 +1085,22 @@ int lttng_session_list_tracker_pids(struct lttng_session *session)
                ret = PTR_ERR(tracker_pids_list_file);
                goto file_error;
        }
+       if (atomic_long_add_unless(&session->file->f_count,
+               1, INT_MAX) == INT_MAX) {
+               goto refcount_error;
+       }
        ret = lttng_tracker_pids_list_fops.open(NULL, tracker_pids_list_file);
        if (ret < 0)
                goto open_error;
        m = tracker_pids_list_file->private_data;
        m->private = session;
        fd_install(file_fd, tracker_pids_list_file);
-       atomic_long_inc(&session->file->f_count);
 
        return file_fd;
 
 open_error:
+       atomic_long_dec(&session->file->f_count);
+refcount_error:
        fput(tracker_pids_list_file);
 file_error:
        put_unused_fd(file_fd);
index 73f2656553e1d98df8d8d5e0ce497f21290c3767..f4911f99f8fefd708dad1ebc50e14b457a6974ed 100644 (file)
@@ -224,9 +224,9 @@ int lttng_kretprobes_register(const char *name,
         * unregistered. Same for memory allocation.
         */
        kref_init(&lttng_krp->kref_alloc);
-       kref_get(&lttng_krp->kref_alloc);       /* inc refcount to 2 */
+       kref_get(&lttng_krp->kref_alloc);       /* inc refcount to 2, no overflow. */
        kref_init(&lttng_krp->kref_register);
-       kref_get(&lttng_krp->kref_register);    /* inc refcount to 2 */
+       kref_get(&lttng_krp->kref_register);    /* inc refcount to 2, no overflow. */
 
        /*
         * Ensure the memory we just allocated don't trigger page faults.
diff --git a/wrapper/kref.h b/wrapper/kref.h
new file mode 100644 (file)
index 0000000..eedefbf
--- /dev/null
@@ -0,0 +1,46 @@
+#ifndef _LTTNG_WRAPPER_KREF_H
+#define _LTTNG_WRAPPER_KREF_H
+
+/*
+ * wrapper/kref.h
+ *
+ * wrapper around linux/kref.h.
+ *
+ * Copyright (C) 2016 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; only version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * This wrapper code is derived from Linux 3.19.2 include/linux/list.h
+ * and include/linux/rculist.h, hence the GPLv2 license applied to this
+ * file.
+ */
+
+#include <linux/kref.h>
+#include <linux/rculist.h>
+
+/*
+ * lttng_kref_get: get reference count, checking for overflow.
+ *
+ * Return 1 if reference is taken, 0 otherwise (overflow).
+ */
+static inline int lttng_kref_get(struct kref *kref)
+{
+       if (atomic_add_unless(&kref->refcount, 1, INT_MAX) != INT_MAX) {
+               return 1;
+       } else {
+               return 0;
+       }
+}
+
+#endif /* _LTTNG_WRAPPER_KREF_H */
This page took 0.030838 seconds and 4 git commands to generate.