[deliverable/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 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]", session, 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.
 */
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,
				   enum instrum_type itype,
				   const struct lttng_event_desc *event_desc,
				   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->desc = event_desc;
	event->filter = filter;
	event->id = chan->free_event_id++;
	event->itype = itype;
	/* Populate ltt_event structure before tracepoint registration. */
	smp_wmb();
	switch (itype) {
	case INSTRUM_TRACEPOINTS:
		ret = tracepoint_probe_register(name, event_desc->probe_callback,
						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));
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->itype) {
	case INSTRUM_TRACEPOINTS:
		ret = tracepoint_probe_unregister(event->desc->name,
						  event->desc->probe_callback,
						  event);
		if (ret)
			return ret;
		break;
	default:
		WARN_ON_ONCE(1);
	}
	return ret;
}

/*
 * Used when an event FD is released.
 */
int ltt_event_unregister(struct ltt_event *event)
{
	int ret;

	mutex_lock(&sessions_mutex);
	ret = ltt_event_unregister(event);
	mutex_unlock(&sessions_mutex);
	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;%s } %s;\n",
				field->type.u.basic.integer.size,
				field->type.u.basic.integer.alignment,
				field->type.u.basic.integer.signedness,
#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;%s } %s[%u];\n",
				elem_type->u.basic.integer.size,
				elem_type->u.basic.integer.alignment,
				elem_type->u.basic.integer.signedness,
#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;%s } %s[ integer { size = %u; align = %u; signed = %u;%s } ];\n",
				elem_type->u.basic.integer.size,
				elem_type->u.basic.integer.alignment,
				elem_type->u.basic.integer.signedness,
#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,
					length_type->u.basic.integer.size,
					length_type->u.basic.integer.alignment,
					length_type->u.basic.integer.signedness,
#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
				);
			break;
		}

		case atype_string:
			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,
		"	} aligned(%u);\n"
		"};\n\n", ltt_get_header_alignment());
	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",
		"};\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;
}

/*
 * 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 = 32; align = %u; signed = false; } := uint32_t;\n"
		"typealias integer {size = 64; align = %u; signed = false; } := uint64_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"
		"		uint64_t timestamp_begin;\n"
		"		uint64_t timestamp_end;\n"
		"		uint32_t content_size;\n"
		"		uint32_t packet_size;\n"
		"		uint32_t events_lost;\n"
		"	};\n",
		"};\n\n",
		ltt_alignof(uint8_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;

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");
Commit	Line	Data
	1	/*
	2	* ltt-events.c
	3	*
	4	* Copyright 2010 (c) - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
	5	*
	6	* Holds LTTng per-session event registry.
	7	*/
	8
	9	#include <linux/module.h>
	10	#include <linux/list.h>
	11	#include <linux/mutex.h>
	12	#include <linux/sched.h>
	13	#include <linux/slab.h>
	14	#include <linux/jiffies.h>
	15	#include <linux/uuid.h>
	16	#include "wrapper/vmalloc.h" /* for wrapper_vmalloc_sync_all() */
	17	#include "ltt-events.h"
	18	#include "ltt-tracer.h"
	19
	20	static LIST_HEAD(sessions);
	21	static LIST_HEAD(ltt_transport_list);
	22	static DEFINE_MUTEX(sessions_mutex);
	23	static struct kmem_cache *event_cache;
	24
	25	static void _ltt_event_destroy(struct ltt_event *event);
	26	static int _ltt_event_unregister(struct ltt_event *event);
	27	static
	28	int _ltt_event_metadata_statedump(struct ltt_session *session,
	29	struct ltt_channel *chan,
	30	struct ltt_event *event);
	31	static
	32	int _ltt_session_metadata_statedump(struct ltt_session *session);
	33
	34
	35	static
	36	void synchronize_trace(void)
	37	{
	38	synchronize_sched();
	39	#ifdef CONFIG_PREEMPT_RT
	40	synchronize_rcu();
	41	#endif
	42	}
	43
	44	struct ltt_session *ltt_session_create(void)
	45	{
	46	struct ltt_session *session;
	47
	48	mutex_lock(&sessions_mutex);
	49	session = kzalloc(sizeof(struct ltt_session), GFP_KERNEL);
	50	if (!session)
	51	return NULL;
	52	INIT_LIST_HEAD(&session->chan);
	53	INIT_LIST_HEAD(&session->events);
	54	uuid_le_gen(&session->uuid);
	55	list_add(&session->list, &sessions);
	56	mutex_unlock(&sessions_mutex);
	57	return session;
	58	}
	59
	60	void ltt_session_destroy(struct ltt_session *session)
	61	{
	62	struct ltt_channel chan, tmpchan;
	63	struct ltt_event event, tmpevent;
	64	int ret;
	65
	66	mutex_lock(&sessions_mutex);
	67	ACCESS_ONCE(session->active) = 0;
	68	list_for_each_entry(event, &session->events, list) {
	69	ret = _ltt_event_unregister(event);
	70	WARN_ON(ret);
	71	}
	72	synchronize_trace(); /* Wait for in-flight events to complete */
	73	list_for_each_entry_safe(event, tmpevent, &session->events, list)
	74	_ltt_event_destroy(event);
	75	list_for_each_entry_safe(chan, tmpchan, &session->chan, list)
	76	_ltt_channel_destroy(chan);
	77	list_del(&session->list);
	78	mutex_unlock(&sessions_mutex);
	79	kfree(session);
	80	}
	81
	82	int ltt_session_start(struct ltt_session *session)
	83	{
	84	int ret = 0;
	85	struct ltt_channel *chan;
	86
	87	mutex_lock(&sessions_mutex);
	88	if (session->active) {
	89	ret = -EBUSY;
	90	goto end;
	91	}
	92
	93	/*
	94	* Snapshot the number of events per channel to know the type of header
	95	* we need to use.
	96	*/
	97	list_for_each_entry(chan, &session->chan, list) {
	98	if (chan->header_type)
	99	continue; /* don't change it if session stop/restart */
	100	if (chan->free_event_id < 31)
	101	chan->header_type = 1; /* compact */
	102	else
	103	chan->header_type = 2; /* large */
	104	}
	105
	106	ACCESS_ONCE(session->active) = 1;
	107	synchronize_trace(); /* Wait for in-flight events to complete */
	108	ret = _ltt_session_metadata_statedump(session);
	109	if (ret) {
	110	ACCESS_ONCE(session->active) = 0;
	111	synchronize_trace(); /* Wait for in-flight events to complete */
	112	}
	113	end:
	114	mutex_unlock(&sessions_mutex);
	115	return ret;
	116	}
	117
	118	int ltt_session_stop(struct ltt_session *session)
	119	{
	120	int ret = 0;
	121
	122	mutex_lock(&sessions_mutex);
	123	if (!session->active) {
	124	ret = -EBUSY;
	125	goto end;
	126	}
	127	ACCESS_ONCE(session->active) = 0;
	128	synchronize_trace(); /* Wait for in-flight events to complete */
	129	end:
	130	mutex_unlock(&sessions_mutex);
	131	return ret;
	132	}
	133
	134	static struct ltt_transport ltt_transport_find(const char name)
	135	{
	136	struct ltt_transport *transport;
	137
	138	list_for_each_entry(transport, &ltt_transport_list, node) {
	139	if (!strcmp(transport->name, name))
	140	return transport;
	141	}
	142	return NULL;
	143	}
	144
	145	struct ltt_channel ltt_channel_create(struct ltt_session session,
	146	const char *transport_name,
	147	void *buf_addr,
	148	size_t subbuf_size, size_t num_subbuf,
	149	unsigned int switch_timer_interval,
	150	unsigned int read_timer_interval)
	151	{
	152	struct ltt_channel *chan;
	153	struct ltt_transport *transport;
	154
	155	mutex_lock(&sessions_mutex);
	156	if (session->active) {
	157	printk(KERN_WARNING "LTTng refusing to add channel to active session\n");
	158	goto active; /* Refuse to add channel to active session */
	159	}
	160	transport = ltt_transport_find(transport_name);
	161	if (!transport) {
	162	printk(KERN_WARNING "LTTng transport %s not found\n",
	163	transport_name);
	164	goto notransport;
	165	}
	166	chan = kzalloc(sizeof(struct ltt_channel), GFP_KERNEL);
	167	if (!chan)
	168	goto nomem;
	169	chan->session = session;
	170	init_waitqueue_head(&chan->notify_wait);
	171	chan->chan = transport->ops.channel_create("[lttng]", session, buf_addr,
	172	subbuf_size, num_subbuf, switch_timer_interval,
	173	read_timer_interval);
	174	if (!chan->chan)
	175	goto create_error;
	176	chan->id = session->free_chan_id++;
	177	chan->ops = &transport->ops;
	178	list_add(&chan->list, &session->chan);
	179	mutex_unlock(&sessions_mutex);
	180	return chan;
	181
	182	create_error:
	183	kfree(chan);
	184	nomem:
	185	notransport:
	186	active:
	187	mutex_unlock(&sessions_mutex);
	188	return NULL;
	189	}
	190
	191	/*
	192	* Only used internally at session destruction.
	193	*/
	194	void _ltt_channel_destroy(struct ltt_channel *chan)
	195	{
	196	chan->ops->channel_destroy(chan->chan);
	197	list_del(&chan->list);
	198	kfree(chan);
	199	}
	200
	201	/*
	202	* Supports event creation while tracing session is active.
	203	*/
	204	struct ltt_event ltt_event_create(struct ltt_channel chan, char *name,
	205	enum instrum_type itype,
	206	const struct lttng_event_desc *event_desc,
	207	void *filter)
	208	{
	209	struct ltt_event *event;
	210	int ret;
	211
	212	mutex_lock(&sessions_mutex);
	213	if (chan->free_event_id == -1UL)
	214	goto full;
	215	/*
	216	* This is O(n^2) (for each event, the loop is called at event
	217	* creation). Might require a hash if we have lots of events.
	218	*/
	219	list_for_each_entry(event, &chan->session->events, list)
	220	if (!strcmp(event->desc->name, name))
	221	goto exist;
	222	event = kmem_cache_zalloc(event_cache, GFP_KERNEL);
	223	if (!event)
	224	goto cache_error;
	225	event->chan = chan;
	226	event->desc = event_desc;
	227	event->filter = filter;
	228	event->id = chan->free_event_id++;
	229	event->itype = itype;
	230	/* Populate ltt_event structure before tracepoint registration. */
	231	smp_wmb();
	232	switch (itype) {
	233	case INSTRUM_TRACEPOINTS:
	234	ret = tracepoint_probe_register(name, event_desc->probe_callback,
	235	event);
	236	if (ret)
	237	goto register_error;
	238	break;
	239	default:
	240	WARN_ON_ONCE(1);
	241	}
	242	ret = _ltt_event_metadata_statedump(chan->session, chan, event);
	243	if (ret)
	244	goto statedump_error;
	245	list_add(&event->list, &chan->session->events);
	246	mutex_unlock(&sessions_mutex);
	247	return event;
	248
	249	statedump_error:
	250	WARN_ON_ONCE(tracepoint_probe_unregister(name, event_desc->probe_callback,
	251	event));
	252	register_error:
	253	kmem_cache_free(event_cache, event);
	254	cache_error:
	255	exist:
	256	full:
	257	mutex_unlock(&sessions_mutex);
	258	return NULL;
	259	}
	260
	261	/*
	262	* Only used internally at session destruction.
	263	*/
	264	int _ltt_event_unregister(struct ltt_event *event)
	265	{
	266	int ret = -EINVAL;
	267
	268	switch (event->itype) {
	269	case INSTRUM_TRACEPOINTS:
	270	ret = tracepoint_probe_unregister(event->desc->name,
	271	event->desc->probe_callback,
	272	event);
	273	if (ret)
	274	return ret;
	275	break;
	276	default:
	277	WARN_ON_ONCE(1);
	278	}
	279	return ret;
	280	}
	281
	282	/*
	283	* Used when an event FD is released.
	284	*/
	285	int ltt_event_unregister(struct ltt_event *event)
	286	{
	287	int ret;
	288
	289	mutex_lock(&sessions_mutex);
	290	ret = ltt_event_unregister(event);
	291	mutex_unlock(&sessions_mutex);
	292	return ret;
	293	}
	294
	295	/*
	296	* Only used internally at session destruction.
	297	*/
	298	static
	299	void _ltt_event_destroy(struct ltt_event *event)
	300	{
	301	ltt_event_put(event->desc);
	302	list_del(&event->list);
	303	kmem_cache_free(event_cache, event);
	304	}
	305
	306	/*
	307	* We have exclusive access to our metadata buffer (protected by the
	308	* sessions_mutex), so we can do racy operations such as looking for
	309	* remaining space left in packet and write, since mutual exclusion
	310	* protects us from concurrent writes.
	311	*/
	312	int lttng_metadata_printf(struct ltt_session *session,
	313	const char *fmt, ...)
	314	{
	315	struct lib_ring_buffer_ctx ctx;
	316	struct ltt_channel *chan = session->metadata;
	317	char *str;
	318	int ret = 0, waitret;
	319	size_t len, reserve_len, pos;
	320	va_list ap;
	321
	322	WARN_ON_ONCE(!ACCESS_ONCE(session->active));
	323
	324	va_start(ap, fmt);
	325	str = kvasprintf(GFP_KERNEL, fmt, ap);
	326	va_end(ap);
	327	if (!str)
	328	return -ENOMEM;
	329
	330	len = strlen(str);
	331	pos = 0;
	332
	333	for (pos = 0; pos < len; pos += reserve_len) {
	334	reserve_len = min_t(size_t,
	335	chan->ops->packet_avail_size(chan->chan),
	336	len - pos);
	337	lib_ring_buffer_ctx_init(&ctx, chan->chan, NULL, reserve_len,
	338	sizeof(char), -1);
	339	/*
	340	* We don't care about metadata buffer's records lost
	341	* count, because we always retry here. Report error if
	342	* we need to bail out after timeout or being
	343	* interrupted.
	344	*/
	345	waitret = wait_event_interruptible_timeout(*chan->ops->get_reader_wait_queue(chan),
	346	({
	347	ret = chan->ops->event_reserve(&ctx);
	348	ret != -ENOBUFS \|\| !ret;
	349	}),
	350	msecs_to_jiffies(LTTNG_METADATA_TIMEOUT_MSEC));
	351	if (!waitret \|\| waitret == -ERESTARTSYS \|\| ret) {
	352	printk(KERN_WARNING "LTTng: Failure to write metadata to buffers (%s)\n",
	353	waitret == -ERESTARTSYS ? "interrupted" :
	354	(ret == -ENOBUFS ? "timeout" : "I/O error"));
	355	if (waitret == -ERESTARTSYS)
	356	ret = waitret;
	357	goto end;
	358	}
	359	chan->ops->event_write(&ctx, &str[pos], len);
	360	chan->ops->event_commit(&ctx);
	361	}
	362	end:
	363	kfree(str);
	364	return ret;
	365	}
	366
	367	static
	368	int _ltt_fields_metadata_statedump(struct ltt_session *session,
	369	struct ltt_event *event)
	370	{
	371	const struct lttng_event_desc *desc = event->desc;
	372	int ret = 0;
	373	int i;
	374
	375	for (i = 0; i < desc->nr_fields; i++) {
	376	const struct lttng_event_field *field = &desc->fields[i];
	377
	378	switch (field->type.atype) {
	379	case atype_integer:
	380	ret = lttng_metadata_printf(session,
	381	" integer { size = %u; align = %u; signed = %u;%s } %s;\n",
	382	field->type.u.basic.integer.size,
	383	field->type.u.basic.integer.alignment,
	384	field->type.u.basic.integer.signedness,
	385	#ifdef __BIG_ENDIAN
	386	field->type.u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
	387	#else
	388	field->type.u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
	389	#endif
	390	field->name);
	391	break;
	392	case atype_enum:
	393	ret = lttng_metadata_printf(session,
	394	" %s %s;\n",
	395	field->type.u.basic.enumeration.name,
	396	field->name);
	397	break;
	398	case atype_array:
	399	{
	400	const struct lttng_basic_type *elem_type;
	401
	402	elem_type = &field->type.u.array.elem_type;
	403	ret = lttng_metadata_printf(session,
	404	" integer { size = %u; align = %u; signed = %u;%s } %s[%u];\n",
	405	elem_type->u.basic.integer.size,
	406	elem_type->u.basic.integer.alignment,
	407	elem_type->u.basic.integer.signedness,
	408	#ifdef __BIG_ENDIAN
	409	elem_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
	410	#else
	411	elem_type->u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
	412	#endif
	413	field->name, field->type.u.array.length);
	414	break;
	415	}
	416	case atype_sequence:
	417	{
	418	const struct lttng_basic_type *elem_type;
	419	const struct lttng_basic_type *length_type;
	420
	421	elem_type = &field->type.u.sequence.elem_type;
	422	length_type = &field->type.u.sequence.length_type;
	423	ret = lttng_metadata_printf(session,
	424	" integer { size = %u; align = %u; signed = %u;%s } %s[ integer { size = %u; align = %u; signed = %u;%s } ];\n",
	425	elem_type->u.basic.integer.size,
	426	elem_type->u.basic.integer.alignment,
	427	elem_type->u.basic.integer.signedness,
	428	#ifdef __BIG_ENDIAN
	429	elem_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
	430	#else
	431	elem_type->u.basic.integer.reverse_byte_order ? " byte_order = be;" : "",
	432	#endif
	433	field->name,
	434	length_type->u.basic.integer.size,
	435	length_type->u.basic.integer.alignment,
	436	length_type->u.basic.integer.signedness,
	437	#ifdef __BIG_ENDIAN
	438	length_type->u.basic.integer.reverse_byte_order ? " byte_order = le;" : "",
	439	#else
	440	length_type->u.basic.integer.reverse_byte_order ? " byte_order = be;" : ""
	441	#endif
	442	);
	443	break;
	444	}
	445
	446	case atype_string:
	447	ret = lttng_metadata_printf(session,
	448	" string%s %s;\n",
	449	field->type.u.basic.string.encoding == lttng_encode_ASCII ?
	450	" { encoding = ASCII; }" : "",
	451	field->name);
	452	break;
	453	default:
	454	WARN_ON_ONCE(1);
	455	return -EINVAL;
	456	}
	457	}
	458	return ret;
	459	}
	460
	461	static
	462	int _ltt_event_metadata_statedump(struct ltt_session *session,
	463	struct ltt_channel *chan,
	464	struct ltt_event *event)
	465	{
	466	int ret = 0;
	467
	468	if (event->metadata_dumped \|\| !ACCESS_ONCE(session->active))
	469	return 0;
	470	if (chan == session->metadata)
	471	return 0;
	472
	473	ret = lttng_metadata_printf(session,
	474	"event {\n"
	475	" name = %s;\n"
	476	" id = %u;\n"
	477	" stream_id = %u;\n"
	478	" event.fields := struct {\n",
	479	event->desc->name,
	480	event->id,
	481	event->chan->id);
	482	if (ret)
	483	goto end;
	484
	485	ret = _ltt_fields_metadata_statedump(session, event);
	486	if (ret)
	487	goto end;
	488
	489	/*
	490	* LTTng space reservation can only reserve multiples of the
	491	* byte size.
	492	*/
	493	ret = lttng_metadata_printf(session,
	494	" } aligned(%u);\n"
	495	"};\n\n", ltt_get_header_alignment());
	496	if (ret)
	497	goto end;
	498
	499
	500
	501
	502	event->metadata_dumped = 1;
	503	end:
	504	return ret;
	505
	506	}
	507
	508	static
	509	int _ltt_channel_metadata_statedump(struct ltt_session *session,
	510	struct ltt_channel *chan)
	511	{
	512	int ret = 0;
	513
	514	if (chan->metadata_dumped \|\| !ACCESS_ONCE(session->active))
	515	return 0;
	516	if (chan == session->metadata)
	517	return 0;
	518
	519	WARN_ON_ONCE(!chan->header_type);
	520	ret = lttng_metadata_printf(session,
	521	"stream {\n"
	522	" id = %u;\n"
	523	" event.header := %s;\n",
	524	"};\n\n",
	525	chan->id,
	526	chan->header_type == 1 ? "struct event_header_compact" :
	527	"struct event_header_large");
	528	if (ret)
	529	goto end;
	530
	531	chan->metadata_dumped = 1;
	532	end:
	533	return ret;
	534	}
	535
	536	/*
	537	* Output metadata into this session's metadata buffers.
	538	*/
	539	static
	540	int _ltt_session_metadata_statedump(struct ltt_session *session)
	541	{
	542	char uuid_s[37];
	543	struct ltt_channel *chan;
	544	struct ltt_event *event;
	545	int ret = 0;
	546
	547	if (!ACCESS_ONCE(session->active))
	548	return 0;
	549	if (session->metadata_dumped)
	550	goto skip_session;
	551	if (!session->metadata) {
	552	printk(KERN_WARNING "LTTng: attempt to start tracing, but metadata channel is not found. Operation abort.\n");
	553	return -EPERM;
	554	}
	555
	556	snprintf(uuid_s, sizeof(uuid_s),
	557	"%x%x%x%x-%x%x-%x%x-%x%x-%x%x%x%x%x%x",
	558	uuid_s[0], uuid_s[1], uuid_s[2], uuid_s[3],
	559	uuid_s[4], uuid_s[5], uuid_s[6], uuid_s[7],
	560	uuid_s[8], uuid_s[9], uuid_s[10], uuid_s[11],
	561	uuid_s[12], uuid_s[13], uuid_s[14], uuid_s[15]);
	562
	563	ret = lttng_metadata_printf(session,
	564	"typealias integer {size = 8; align = %u; signed = false; } := uint8_t;\n"
	565	"typealias integer {size = 32; align = %u; signed = false; } := uint32_t;\n"
	566	"typealias integer {size = 64; align = %u; signed = false; } := uint64_t;\n"
	567	"\n"
	568	"trace {\n"
	569	" major = %u;\n"
	570	" minor = %u;\n"
	571	" uuid = %s;\n"
	572	" byte_order = %s;\n"
	573	" packet.header := struct {\n"
	574	" uint32_t magic;\n"
	575	" uint8_t uuid[16];\n"
	576	" uint32_t stream_id;\n"
	577	" uint64_t timestamp_begin;\n"
	578	" uint64_t timestamp_end;\n"
	579	" uint32_t content_size;\n"
	580	" uint32_t packet_size;\n"
	581	" uint32_t events_lost;\n"
	582	" };\n",
	583	"};\n\n",
	584	ltt_alignof(uint8_t) * CHAR_BIT,
	585	ltt_alignof(uint32_t) * CHAR_BIT,
	586	ltt_alignof(uint64_t) * CHAR_BIT,
	587	CTF_VERSION_MAJOR,
	588	CTF_VERSION_MINOR,
	589	uuid_s,
	590	#ifdef __BIG_ENDIAN
	591	"be"
	592	#else
	593	"le"
	594	#endif
	595	);
	596	if (ret)
	597	goto end;
	598
	599	skip_session:
	600	list_for_each_entry(chan, &session->chan, list) {
	601	ret = _ltt_channel_metadata_statedump(session, chan);
	602	if (ret)
	603	goto end;
	604	}
	605
	606	list_for_each_entry(event, &session->events, list) {
	607	ret = _ltt_event_metadata_statedump(session, chan, event);
	608	if (ret)
	609	goto end;
	610	}
	611	session->metadata_dumped = 1;
	612	end:
	613	return ret;
	614	}
	615
	616	/**
	617	* ltt_transport_register - LTT transport registration
	618	* @transport: transport structure
	619	*
	620	* Registers a transport which can be used as output to extract the data out of
	621	* LTTng. The module calling this registration function must ensure that no
	622	* trap-inducing code will be executed by the transport functions. E.g.
	623	* vmalloc_sync_all() must be called between a vmalloc and the moment the memory
	624	* is made visible to the transport function. This registration acts as a
	625	* vmalloc_sync_all. Therefore, only if the module allocates virtual memory
	626	* after its registration must it synchronize the TLBs.
	627	*/
	628	void ltt_transport_register(struct ltt_transport *transport)
	629	{
	630	/*
	631	* Make sure no page fault can be triggered by the module about to be
	632	* registered. We deal with this here so we don't have to call
	633	* vmalloc_sync_all() in each module's init.
	634	*/
	635	wrapper_vmalloc_sync_all();
	636
	637	mutex_lock(&sessions_mutex);
	638	list_add_tail(&transport->node, &ltt_transport_list);
	639	mutex_unlock(&sessions_mutex);
	640	}
	641	EXPORT_SYMBOL_GPL(ltt_transport_register);
	642
	643	/**
	644	* ltt_transport_unregister - LTT transport unregistration
	645	* @transport: transport structure
	646	*/
	647	void ltt_transport_unregister(struct ltt_transport *transport)
	648	{
	649	mutex_lock(&sessions_mutex);
	650	list_del(&transport->node);
	651	mutex_unlock(&sessions_mutex);
	652	}
	653	EXPORT_SYMBOL_GPL(ltt_transport_unregister);
	654
	655	static int __init ltt_events_init(void)
	656	{
	657	int ret;
	658
	659	event_cache = KMEM_CACHE(ltt_event, 0);
	660	if (!event_cache)
	661	return -ENOMEM;
	662	ret = ltt_debugfs_abi_init();
	663	if (ret)
	664	goto error_abi;
	665	return 0;
	666	error_abi:
	667	kmem_cache_destroy(event_cache);
	668	return ret;
	669	}
	670
	671	module_init(ltt_events_init);
	672
	673	static void __exit ltt_events_exit(void)
	674	{
	675	struct ltt_session session, tmpsession;
	676
	677	ltt_debugfs_abi_exit();
	678	list_for_each_entry_safe(session, tmpsession, &sessions, list)
	679	ltt_session_destroy(session);
	680	kmem_cache_destroy(event_cache);
	681	}
	682
	683	module_exit(ltt_events_exit);
	684
	685	MODULE_LICENSE("GPL and additional rights");
	686	MODULE_AUTHOR("Mathieu Desnoyers <mathieu.desnoyers@efficios.com>");
	687	MODULE_DESCRIPTION("LTTng Events");