LTTng metadata fix

[lttng-modules.git] / ltt-events.c

/*
 * ltt-events.c
 *
 * Copyright 2010 (c) - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *
 * Holds LTTng per-session event registry.
 */

#include <linux/module.h>
#include <linux/list.h>
#include <linux/mutex.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/jiffies.h>
#include <linux/uuid.h>
#include "wrapper/vmalloc.h"	/* for wrapper_vmalloc_sync_all() */
#include "ltt-events.h"
#include "ltt-tracer.h"

static LIST_HEAD(sessions);
static LIST_HEAD(ltt_transport_list);
static DEFINE_MUTEX(sessions_mutex);
static struct kmem_cache *event_cache;

static void _ltt_event_destroy(struct ltt_event *event);
static void _ltt_channel_destroy(struct ltt_channel *chan);
static int _ltt_event_unregister(struct ltt_event *event);
static
int _ltt_event_metadata_statedump(struct ltt_session *session,
				  struct ltt_channel *chan,
				  struct ltt_event *event);
static
int _ltt_session_metadata_statedump(struct ltt_session *session);


static
void synchronize_trace(void)
{
	synchronize_sched();
#ifdef CONFIG_PREEMPT_RT
	synchronize_rcu();
#endif
}

struct ltt_session *ltt_session_create(void)
{
	struct ltt_session *session;

	mutex_lock(&sessions_mutex);
	session = kzalloc(sizeof(struct ltt_session), GFP_KERNEL);
	if (!session)
		return NULL;
	INIT_LIST_HEAD(&session->chan);
	INIT_LIST_HEAD(&session->events);
	uuid_le_gen(&session->uuid);
	list_add(&session->list, &sessions);
	mutex_unlock(&sessions_mutex);
	return session;
}

void ltt_session_destroy(struct ltt_session *session)
{
	struct ltt_channel *chan, *tmpchan;
	struct ltt_event *event, *tmpevent;
	int ret;

	mutex_lock(&sessions_mutex);
	ACCESS_ONCE(session->active) = 0;
	list_for_each_entry(event, &session->events, list) {
		ret = _ltt_event_unregister(event);
		WARN_ON(ret);
	}
	synchronize_trace();	/* Wait for in-flight events to complete */
	list_for_each_entry_safe(event, tmpevent, &session->events, list)
		_ltt_event_destroy(event);
	list_for_each_entry_safe(chan, tmpchan, &session->chan, list)
		_ltt_channel_destroy(chan);
	list_del(&session->list);
	mutex_unlock(&sessions_mutex);
	kfree(session);
}

int ltt_session_start(struct ltt_session *session)
{
	int ret = 0;
	struct ltt_channel *chan;

	mutex_lock(&sessions_mutex);
	if (session->active) {
		ret = -EBUSY;
		goto end;
	}

	/*
	 * Snapshot the number of events per channel to know the type of header
	 * we need to use.
	 */
	list_for_each_entry(chan, &session->chan, list) {
		if (chan->header_type)
			continue;		/* don't change it if session stop/restart */
		if (chan->free_event_id < 31)
			chan->header_type = 1;	/* compact */
		else
			chan->header_type = 2;	/* large */
	}

	ACCESS_ONCE(session->active) = 1;
	synchronize_trace();	/* Wait for in-flight events to complete */
	ret = _ltt_session_metadata_statedump(session);
	if (ret) {
		ACCESS_ONCE(session->active) = 0;
		synchronize_trace();	/* Wait for in-flight events to complete */
	}
end:
	mutex_unlock(&sessions_mutex);
	return ret;
}

int ltt_session_stop(struct ltt_session *session)
{
	int ret = 0;

	mutex_lock(&sessions_mutex);
	if (!session->active) {
		ret = -EBUSY;
		goto end;
	}
	ACCESS_ONCE(session->active) = 0;
	synchronize_trace();	/* Wait for in-flight events to complete */
end:
	mutex_unlock(&sessions_mutex);
	return ret;
}

static struct ltt_transport *ltt_transport_find(const char *name)
{
	struct ltt_transport *transport;

	list_for_each_entry(transport, &ltt_transport_list, node) {
		if (!strcmp(transport->name, name))
			return transport;
	}
	return NULL;
}

struct ltt_channel *ltt_channel_create(struct ltt_session *session,
				       const char *transport_name,
				       void *buf_addr,
				       size_t subbuf_size, size_t num_subbuf,
				       unsigned int switch_timer_interval,
				       unsigned int read_timer_interval)
{
	struct ltt_channel *chan;
	struct ltt_transport *transport;

	mutex_lock(&sessions_mutex);
	if (session->active) {
		printk(KERN_WARNING "LTTng refusing to add channel to active session\n");
		goto active;	/* Refuse to add channel to active session */
	}
	transport = ltt_transport_find(transport_name);
	if (!transport) {
		printk(KERN_WARNING "LTTng transport %s not found\n",
		       transport_name);
		goto notransport;
	}
	chan = kzalloc(sizeof(struct ltt_channel), GFP_KERNEL);
	if (!chan)
		goto nomem;
	chan->session = session;
	init_waitqueue_head(&chan->notify_wait);
	chan->chan = transport->ops.channel_create("[lttng]", chan, buf_addr,
			subbuf_size, num_subbuf, switch_timer_interval,
			read_timer_interval);
	if (!chan->chan)
		goto create_error;
	chan->id = session->free_chan_id++;
	chan->ops = &transport->ops;
	list_add(&chan->list, &session->chan);
	mutex_unlock(&sessions_mutex);
	return chan;

create_error:
	kfree(chan);
nomem:
notransport:
active:
	mutex_unlock(&sessions_mutex);
	return NULL;
}

/*
 * Only used internally at session destruction.
 */
static
void _ltt_channel_destroy(struct ltt_channel *chan)
{
	chan->ops->channel_destroy(chan->chan);
	list_del(&chan->list);
	kfree(chan);
}

/*
 * Supports event creation while tracing session is active.
 */
struct ltt_event *ltt_event_create(struct ltt_channel *chan, char *name,
				   struct lttng_kernel_event *event_param,
				   void *filter)
{
	struct ltt_event *event;
	int ret;

	mutex_lock(&sessions_mutex);
	if (chan->free_event_id == -1UL)
		goto full;
	/*
	 * This is O(n^2) (for each event, the loop is called at event
	 * creation). Might require a hash if we have lots of events.
	 */
	list_for_each_entry(event, &chan->session->events, list)
		if (!strcmp(event->desc->name, name))
			goto exist;
	event = kmem_cache_zalloc(event_cache, GFP_KERNEL);
	if (!event)
		goto cache_error;
	event->chan = chan;
	event->filter = filter;
	event->id = chan->free_event_id++;
	event->instrumentation = event_param->instrumentation;
	/* Populate ltt_event structure before tracepoint registration. */
	smp_wmb();
	switch (event_param->instrumentation) {
	case LTTNG_KERNEL_TRACEPOINTS:
		event->desc = ltt_event_get(name);
		if (!event->desc)
			goto register_error;
		ret = tracepoint_probe_register(name,
				event->desc->probe_callback,
				event);
		if (ret)
			goto register_error;
		break;
	case LTTNG_KERNEL_KPROBES:
		ret = lttng_kprobes_register(name,
				event_param->u.kprobe.symbol_name,
				event_param->u.kprobe.offset,
				event_param->u.kprobe.addr,
				event);
		if (ret)
			goto register_error;
		break;
	case LTTNG_KERNEL_FUNCTION_TRACER:
		ret = lttng_ftrace_register(name,
				event_param->u.ftrace.symbol_name,
				event);
		if (ret)
			goto register_error;
		break;
	default:
		WARN_ON_ONCE(1);
	}
	ret = _ltt_event_metadata_statedump(chan->session, chan, event);
	if (ret)
		goto statedump_error;
	list_add(&event->list, &chan->session->events);
	mutex_unlock(&sessions_mutex);
	return event;

statedump_error:
	WARN_ON_ONCE(tracepoint_probe_unregister(name,
				event->desc->probe_callback,
				event));
	ltt_event_put(event->desc);
register_error:
	kmem_cache_free(event_cache, event);
cache_error:
exist:
full:
	mutex_unlock(&sessions_mutex);
	return NULL;
}

/*
 * Only used internally at session destruction.
 */
int _ltt_event_unregister(struct ltt_event *event)
{
	int ret = -EINVAL;

	switch (event->instrumentation) {
	case LTTNG_KERNEL_TRACEPOINTS:
		ret = tracepoint_probe_unregister(event->desc->name,
						  event->desc->probe_callback,
						  event);
		if (ret)
			return ret;
		ltt_event_put(event->desc);
		break;
	case LTTNG_KERNEL_KPROBES:
		lttng_kprobes_unregister(event);
		ret = 0;
		break;
	case LTTNG_KERNEL_FUNCTION_TRACER:
		lttng_ftrace_unregister(event);
		ret = 0;
		break;
	default:
		WARN_ON_ONCE(1);
	}
	return ret;
}

/*
 * Only used internally at session destruction.
 */
static
void _ltt_event_destroy(struct ltt_event *event)
{
	ltt_event_put(event->desc);
	list_del(&event->list);
	kmem_cache_free(event_cache, event);
}

/*
 * We have exclusive access to our metadata buffer (protected by the
 * sessions_mutex), so we can do racy operations such as looking for
 * remaining space left in packet and write, since mutual exclusion
 * protects us from concurrent writes.
 */
int lttng_metadata_printf(struct ltt_session *session,
			  const char *fmt, ...)
{
	struct lib_ring_buffer_ctx ctx;
	struct ltt_channel *chan = session->metadata;
	char *str;
	int ret = 0, waitret;
	size_t len, reserve_len, pos;
	va_list ap;

	WARN_ON_ONCE(!ACCESS_ONCE(session->active));

	va_start(ap, fmt);
	str = kvasprintf(GFP_KERNEL, fmt, ap);
	va_end(ap);
	if (!str)
		return -ENOMEM;

	len = strlen(str);
	pos = 0;

	for (pos = 0; pos < len; pos += reserve_len) {
		reserve_len = min_t(size_t,
				chan->ops->packet_avail_size(chan->chan),
				len - pos);
		lib_ring_buffer_ctx_init(&ctx, chan->chan, NULL, reserve_len,
					 sizeof(char), -1);
		/*
		 * We don't care about metadata buffer's records lost
		 * count, because we always retry here. Report error if
		 * we need to bail out after timeout or being
		 * interrupted.
		 */
		waitret = wait_event_interruptible_timeout(*chan->ops->get_reader_wait_queue(chan),
			({
				ret = chan->ops->event_reserve(&ctx);
				ret != -ENOBUFS || !ret;
			}),
			msecs_to_jiffies(LTTNG_METADATA_TIMEOUT_MSEC));
		if (!waitret || waitret == -ERESTARTSYS || ret) {
			printk(KERN_WARNING "LTTng: Failure to write metadata to buffers (%s)\n",
				waitret == -ERESTARTSYS ? "interrupted" :
					(ret == -ENOBUFS ? "timeout" : "I/O error"));
			if (waitret == -ERESTARTSYS)
				ret = waitret;
			goto end;
		}
		chan->ops->event_write(&ctx, &str[pos], len);
		chan->ops->event_commit(&ctx);
	}
end:
	kfree(str);
	return ret;
}

static
int _ltt_fields_metadata_statedump(struct ltt_session *session,
				   struct ltt_event *event)
{
	const struct lttng_event_desc *desc = event->desc;
	int ret = 0;
	int i;

	for (i = 0; i < desc->nr_fields; i++) {
		const struct lttng_event_field *field = &desc->fields[i];

		switch (field->type.atype) {
		case atype_integer:
			ret = lttng_metadata_printf(session,
				"		integer { size = %u; align = %u; signed = %u; encoding = %s; base = %u;%s } %s;\n",
				field->type.u.basic.integer.size,
				field->type.u.basic.integer.alignment,
				field->type.u.basic.integer.signedness,
				(field->type.u.basic.integer.encoding == lttng_encode_none)
					? "none"
					: (field->type.u.basic.integer.encoding == lttng_encode_UTF8)
						? "UTF8"
						: "ASCII",
				field->type.u.basic.integer.base,
#ifdef __BIG_ENDIAN
				field->type.u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
#else
				field->type.u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
#endif
				field->name);
			break;
		case atype_enum:
			ret = lttng_metadata_printf(session,
				"		%s %s;\n",
				field->type.u.basic.enumeration.name,
				field->name);
			break;
		case atype_array:
		{
			const struct lttng_basic_type *elem_type;

			elem_type = &field->type.u.array.elem_type;
			ret = lttng_metadata_printf(session,
				"		integer { size = %u; align = %u; signed = %u; encoding = %s; base = %u;%s } %s[%u];\n",
				elem_type->u.basic.integer.size,
				elem_type->u.basic.integer.alignment,
				elem_type->u.basic.integer.signedness,
				(elem_type->u.basic.integer.encoding == lttng_encode_none)
					? "none"
					: (elem_type->u.basic.integer.encoding == lttng_encode_UTF8)
						? "UTF8"
						: "ASCII",
				elem_type->u.basic.integer.base,
#ifdef __BIG_ENDIAN
				elem_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
#else
				elem_type->u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
#endif
				field->name, field->type.u.array.length);
			break;
		}
		case atype_sequence:
		{
			const struct lttng_basic_type *elem_type;
			const struct lttng_basic_type *length_type;

			elem_type = &field->type.u.sequence.elem_type;
			length_type = &field->type.u.sequence.length_type;
			ret = lttng_metadata_printf(session,
				"		integer { size = %u; align = %u; signed = %u; encoding = %s; base = %u;%s } __%s_length;\n",
				"		integer { size = %u; align = %u; signed = %u; encoding = %s; base = %u;%s } %s[ __%s_length ];\n",
				length_type->u.basic.integer.size,
				length_type->u.basic.integer.alignment,
				length_type->u.basic.integer.signedness,
				(length_type->u.basic.integer.encoding == lttng_encode_none)
					? "none"
					: (length_type->u.basic.integer.encoding == lttng_encode_UTF8)
						? "UTF8"
						: "ASCII",
				length_type->u.basic.integer.base,
#ifdef __BIG_ENDIAN
				length_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
#else
				length_type->u.basic.integer.reverse_byte_order
? " byte_order = be;" : "",
#endif
				field->name,
				elem_type->u.basic.integer.size,
				elem_type->u.basic.integer.alignment,
				elem_type->u.basic.integer.signedness,
				(elem_type->u.basic.integer.encoding == lttng_encode_none)
					? "none"
					: (elem_type->u.basic.integer.encoding == lttng_encode_UTF8)
						? "UTF8"
						: "ASCII",
				elem_type->u.basic.integer.base,
#ifdef __BIG_ENDIAN
				elem_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
#else
				elem_type->u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
#endif
				field->name,
				field->name
				);
			break;
		}

		case atype_string:
			/* Default encoding is UTF8 */
			ret = lttng_metadata_printf(session,
				"		string%s %s;\n",
				field->type.u.basic.string.encoding == lttng_encode_ASCII ?
					" { encoding = ASCII; }" : "",
				field->name);
			break;
		default:
			WARN_ON_ONCE(1);
			return -EINVAL;
		}
	}
	return ret;
}

static
int _ltt_event_metadata_statedump(struct ltt_session *session,
				  struct ltt_channel *chan,
				  struct ltt_event *event)
{
	int ret = 0;

	if (event->metadata_dumped || !ACCESS_ONCE(session->active))
		return 0;
	if (chan == session->metadata)
		return 0;

	ret = lttng_metadata_printf(session,
		"event {\n"
		"	name = %s;\n"
		"	id = %u;\n"
		"	stream_id = %u;\n"
		"	event.fields := struct {\n",
		event->desc->name,
		event->id,
		event->chan->id);
	if (ret)
		goto end;

	ret = _ltt_fields_metadata_statedump(session, event);
	if (ret)
		goto end;

	/*
	 * LTTng space reservation can only reserve multiples of the
	 * byte size.
	 */
	ret = lttng_metadata_printf(session,
		"	};\n"
		"};\n\n");
	if (ret)
		goto end;

	event->metadata_dumped = 1;
end:
	return ret;

}

static
int _ltt_channel_metadata_statedump(struct ltt_session *session,
				    struct ltt_channel *chan)
{
	int ret = 0;

	if (chan->metadata_dumped || !ACCESS_ONCE(session->active))
		return 0;
	if (chan == session->metadata)
		return 0;

	WARN_ON_ONCE(!chan->header_type);
	ret = lttng_metadata_printf(session,
		"stream {\n"
		"	id = %u;\n"
		"	event.header := %s;\n"
		"	packet.context := struct packet_context;\n"
		"};\n\n",
		chan->id,
		chan->header_type == 1 ? "struct event_header_compact" :
			"struct event_header_large");
	if (ret)
		goto end;

	chan->metadata_dumped = 1;
end:
	return ret;
}

static
int _ltt_stream_packet_context_declare(struct ltt_session *session)
{
	return lttng_metadata_printf(session,
		"struct packet_context {\n"
		"	uint64_t timestamp_begin;\n"
		"	uint64_t timestamp_end;\n"
		"	uint32_t events_discarded;\n"
		"	uint32_t content_size;\n"
		"	uint32_t packet_size;\n"
		"	uint32_t cpu_id;\n"
		"};\n"
		);
}

/*
 * Compact header:
 * id: range: 0 - 30.
 * id 31 is reserved to indicate an extended header.
 *
 * Large header:
 * id: range: 0 - 65534.
 * id 65535 is reserved to indicate an extended header.
 */
static
int _ltt_event_header_declare(struct ltt_session *session)
{
	return lttng_metadata_printf(session,
	"struct event_header_compact {\n"
	"	enum : uint5_t { compact = 0 ... 30, extended = 31 } id;\n"
	"	variant <id> {\n"
	"		struct {\n"
	"			uint27_t timestamp;\n"
	"		} compact;\n"
	"		struct {\n"
	"			uint32_t id;\n"
	"			uint64_t timestamp;\n"
	"		} extended;\n"
	"	} v;\n"
	"} align(%u);\n"
	"\n"
	"struct event_header_large {\n"
	"	enum : uint16_t { compact = 0 ... 65534, extended = 65535 } id;\n"
	"	variant <id> {\n"
	"		struct {\n"
	"			uint32_t timestamp;\n"
	"		} compact;\n"
	"		struct {\n"
	"			uint32_t id;\n"
	"			uint64_t timestamp;\n"
	"		} extended;\n"
	"	} v;\n"
	"} align(%u);\n\n",
	ltt_alignof(uint32_t) * CHAR_BIT,
	ltt_alignof(uint16_t) * CHAR_BIT
	);
}

/*
 * Output metadata into this session's metadata buffers.
 */
static
int _ltt_session_metadata_statedump(struct ltt_session *session)
{
	char uuid_s[37];
	struct ltt_channel *chan;
	struct ltt_event *event;
	int ret = 0;

	if (!ACCESS_ONCE(session->active))
		return 0;
	if (session->metadata_dumped)
		goto skip_session;
	if (!session->metadata) {
		printk(KERN_WARNING "LTTng: attempt to start tracing, but metadata channel is not found. Operation abort.\n");
		return -EPERM;
	}

	snprintf(uuid_s, sizeof(uuid_s),
		"%x%x%x%x-%x%x-%x%x-%x%x-%x%x%x%x%x%x",
		uuid_s[0], uuid_s[1], uuid_s[2], uuid_s[3],
		uuid_s[4], uuid_s[5], uuid_s[6], uuid_s[7],
		uuid_s[8], uuid_s[9], uuid_s[10], uuid_s[11],
		uuid_s[12], uuid_s[13], uuid_s[14], uuid_s[15]);

	ret = lttng_metadata_printf(session,
		"typealias integer { size = 8; align = %u; signed = false; } := uint8_t;\n"
		"typealias integer { size = 16; align = %u; signed = false; } := uint16_t;\n"
		"typealias integer { size = 32; align = %u; signed = false; } := uint32_t;\n"
		"typealias integer { size = 64; align = %u; signed = false; } := uint64_t;\n"
		"typealias integer { size = 5; align = 1; signed = false; } := uint5_t;\n"
		"typealias integer { size = 27; align = 1; signed = false; } := uint27_t;\n"
		"\n"
		"trace {\n"
		"	major = %u;\n"
		"	minor = %u;\n"
		"	uuid = %s;\n"
		"	byte_order = %s;\n"
		"	packet.header := struct {\n"
		"		uint32_t magic;\n"
		"		uint8_t  uuid[16];\n"
		"		uint32_t stream_id;\n"
		"	};\n"
		"};\n\n",
		ltt_alignof(uint8_t) * CHAR_BIT,
		ltt_alignof(uint16_t) * CHAR_BIT,
		ltt_alignof(uint32_t) * CHAR_BIT,
		ltt_alignof(uint64_t) * CHAR_BIT,
		CTF_VERSION_MAJOR,
		CTF_VERSION_MINOR,
		uuid_s,
#ifdef __BIG_ENDIAN
		"be"
#else
		"le"
#endif
		);
	if (ret)
		goto end;

	ret = _ltt_stream_packet_context_declare(session);
	if (ret)
		goto end;

	ret = _ltt_event_header_declare(session);
	if (ret)
		goto end;

skip_session:
	list_for_each_entry(chan, &session->chan, list) {
		ret = _ltt_channel_metadata_statedump(session, chan);
		if (ret)
			goto end;
	}

	list_for_each_entry(event, &session->events, list) {
		ret = _ltt_event_metadata_statedump(session, chan, event);
		if (ret)
			goto end;
	}
	session->metadata_dumped = 1;
end:
	return ret;
}

/**
 * ltt_transport_register - LTT transport registration
 * @transport: transport structure
 *
 * Registers a transport which can be used as output to extract the data out of
 * LTTng. The module calling this registration function must ensure that no
 * trap-inducing code will be executed by the transport functions. E.g.
 * vmalloc_sync_all() must be called between a vmalloc and the moment the memory
 * is made visible to the transport function. This registration acts as a
 * vmalloc_sync_all. Therefore, only if the module allocates virtual memory
 * after its registration must it synchronize the TLBs.
 */
void ltt_transport_register(struct ltt_transport *transport)
{
	/*
	 * Make sure no page fault can be triggered by the module about to be
	 * registered. We deal with this here so we don't have to call
	 * vmalloc_sync_all() in each module's init.
	 */
	wrapper_vmalloc_sync_all();

	mutex_lock(&sessions_mutex);
	list_add_tail(&transport->node, &ltt_transport_list);
	mutex_unlock(&sessions_mutex);
}
EXPORT_SYMBOL_GPL(ltt_transport_register);

/**
 * ltt_transport_unregister - LTT transport unregistration
 * @transport: transport structure
 */
void ltt_transport_unregister(struct ltt_transport *transport)
{
	mutex_lock(&sessions_mutex);
	list_del(&transport->node);
	mutex_unlock(&sessions_mutex);
}
EXPORT_SYMBOL_GPL(ltt_transport_unregister);

static int __init ltt_events_init(void)
{
	int ret;

	event_cache = KMEM_CACHE(ltt_event, 0);
	if (!event_cache)
		return -ENOMEM;
	ret = ltt_debugfs_abi_init();
	if (ret)
		goto error_abi;
	return 0;
error_abi:
	kmem_cache_destroy(event_cache);
	return ret;
}

module_init(ltt_events_init);

static void __exit ltt_events_exit(void)
{
	struct ltt_session *session, *tmpsession;

	ltt_debugfs_abi_exit();
	list_for_each_entry_safe(session, tmpsession, &sessions, list)
		ltt_session_destroy(session);
	kmem_cache_destroy(event_cache);
}

module_exit(ltt_events_exit);

MODULE_LICENSE("GPL and additional rights");
MODULE_AUTHOR("Mathieu Desnoyers <mathieu.desnoyers@efficios.com>");
MODULE_DESCRIPTION("LTTng Events");