Fix: ust-consumer: segfault on snapshot after regenerate metadata

[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c

diff --git a/src/common/kernel-consumer/kernel-consumer.c b/src/common/kernel-consumer/kernel-consumer.c
index c0d8e09720b1c3409fda2fb315f1b79d8ca98068..362cbd2dca61bde6f23a4375725b172b22f3bb70 100644 (file)
--- a/src/common/kernel-consumer/kernel-consumer.c
+++ b/src/common/kernel-consumer/kernel-consumer.c
@@ -139,7 +139,7 @@ error:
 /*
  * Take a snapshot of all the stream of a channel
  * RCU read-side lock must be held across this function to ensure existence of
- * channel. The channel lock must be held by the caller.
+ * channel.
  *
  * Returns 0 on success, < 0 on error
  */
@@ -154,6 +154,9 @@ static int lttng_kconsumer_snapshot_channel(
 
        DBG("Kernel consumer snapshot channel %" PRIu64, key);
 
+       /* Prevent channel modifications while we perform the snapshot.*/
+       pthread_mutex_lock(&channel->lock);
+
        rcu_read_lock();
 
        /* Splice is not supported yet for channel snapshot. */
@@ -347,13 +350,14 @@ end_unlock:
        pthread_mutex_unlock(&stream->lock);
 end:
        rcu_read_unlock();
+       pthread_mutex_unlock(&channel->lock);
        return ret;
 }
 
 /*
  * Read the whole metadata available for a snapshot.
  * RCU read-side lock must be held across this function to ensure existence of
- * metadata_channel. The channel lock must be held by the caller.
+ * metadata_channel.
  *
  * Returns 0 on success, < 0 on error
  */
@@ -376,7 +380,8 @@ static int lttng_kconsumer_snapshot_metadata(
        metadata_stream = metadata_channel->metadata_stream;
        assert(metadata_stream);
 
-       pthread_mutex_lock(&metadata_stream->lock);
+       /* Take all the appropriate locks hehehe.*/
+       metadata_stream->read_subbuffer_ops.lock(metadata_stream);
        assert(metadata_channel->trace_chunk);
        assert(metadata_stream->trace_chunk);
 
@@ -403,16 +408,12 @@ static int lttng_kconsumer_snapshot_metadata(
 
                ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
                if (ret_read < 0) {
-                       if (ret_read != -EAGAIN) {
-                               ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
-                                               ret_read);
-                               ret = ret_read;
-                               goto error_snapshot;
-                       }
-                       /* ret_read is negative at this point so we will exit the loop. */
-                       continue;
+                       ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
+                                       ret_read);
+                       ret = ret_read;
+                       goto error_snapshot;
                }
-       } while (ret_read >= 0);
+       } while (ret_read > 0);
 
        if (use_relayd) {
                close_relayd_stream(metadata_stream);
@@ -435,7 +436,7 @@ static int lttng_kconsumer_snapshot_metadata(
 
        ret = 0;
 error_snapshot:
-       pthread_mutex_unlock(&metadata_stream->lock);
+       metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
        cds_list_del(&metadata_stream->send_node);
        consumer_stream_destroy(metadata_stream, NULL);
        metadata_channel->metadata_stream = NULL;
@@ -939,7 +940,6 @@ error_streams_sent_nosignal:
                        ERR("Channel %" PRIu64 " not found", key);
                        ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
                } else {
-                       pthread_mutex_lock(&channel->lock);
                        if (msg.u.snapshot_channel.metadata == 1) {
                                ret = lttng_kconsumer_snapshot_metadata(channel, key,
                                                msg.u.snapshot_channel.pathname,
@@ -959,7 +959,6 @@ error_streams_sent_nosignal:
                                        ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
                                }
                        }
-                       pthread_mutex_unlock(&channel->lock);
                }
                health_code_update();
 
@@ -1531,6 +1530,17 @@ int get_subbuffer_common(struct lttng_consumer_stream *stream,
 
        ret = kernctl_get_next_subbuf(stream->wait_fd);
        if (ret) {
+               /*
+                * The caller only expects -ENODATA when there is no data to
+                * read, but the kernel tracer returns -EAGAIN when there is
+                * currently no data for a non-finalized stream, and -ENODATA
+                * when there is no data for a finalized stream. Those can be
+                * combined into a -ENODATA return value.
+                */
+               if (ret == -EAGAIN) {
+                       ret = -ENODATA;
+               }
+
                goto end;
        }
 
@@ -1612,6 +1622,16 @@ int get_next_subbuffer_metadata_check(struct lttng_consumer_stream *stream,
                        subbuffer->info.metadata.padded_subbuf_size,
                        coherent ? "true" : "false");
 end:
+       /*
+        * The caller only expects -ENODATA when there is no data to read, but
+        * the kernel tracer returns -EAGAIN when there is currently no data
+        * for a non-finalized stream, and -ENODATA when there is no data for a
+        * finalized stream. Those can be combined into a -ENODATA return value.
+        */
+       if (ret == -EAGAIN) {
+               ret = -ENODATA;
+       }
+
        return ret;
 }