[lttng-modules.git] / lttng-syscalls.c

/*
 * lttng-syscalls.c
 *
 * Copyright 2010 (c) - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *
 * LTTng sched probes.
 *
 * Dual LGPL v2.1/GPL v2 license.
 */

#include <linux/module.h>
#include <linux/slab.h>
#include <asm/ptrace.h>
#include <asm/syscall.h>

#include "ltt-events.h"

static void syscall_entry_probe(void *__data, struct pt_regs *regs, long id);

/*
 * Take care of NOARGS not supported by mainline.
 */
#define DECLARE_EVENT_CLASS_NOARGS(name, tstruct, assign, print)
#define DEFINE_EVENT_NOARGS(template, name)
#define TRACE_EVENT_NOARGS(name, struct, assign, print)

/*
 * Create LTTng tracepoint probes.
 */
#define LTTNG_PACKAGE_BUILD
#define CREATE_TRACE_POINTS

/* Hijack probe callback for system calls */
#define TP_PROBE_CB(_template)		&syscall_entry_probe
#define TP_MODULE_OVERRIDE

#define TRACE_INCLUDE_PATH ../instrumentation/syscalls/headers
#include "instrumentation/syscalls/headers/syscalls_integers.h"
#include "instrumentation/syscalls/headers/syscalls_pointers.h"
#include "instrumentation/syscalls/headers/syscalls_unknown.h"

#undef TP_MODULE_OVERRIDE
#undef TP_PROBE_CB
#undef LTTNG_PACKAGE_BUILD
#undef CREATE_TRACE_POINTS

struct trace_syscall_entry {
	void *func;
	const struct lttng_event_desc *desc;
	const struct lttng_event_field *fields;
	unsigned int nrargs;
};

#define CREATE_SYSCALL_TABLE

#undef TRACE_SYSCALL_TABLE
#define TRACE_SYSCALL_TABLE(_template, _name, _nr, _nrargs)	\
	[ _nr ] = {						\
		.func = __event_probe__##_template,		\
		.nrargs = (_nrargs),				\
		.fields = __event_fields___##_template,		\
		.desc = &__event_desc___##_name,		\
	},

static struct trace_syscall_entry sc_table[] = {
#include "instrumentation/syscalls/headers/syscalls_integers.h"
#include "instrumentation/syscalls/headers/syscalls_pointers.h"
};

static int sc_table_filled;

#undef CREATE_SYSCALL_TABLE

static void syscall_entry_probe(void *__data, struct pt_regs *regs, long id)
{
	struct trace_syscall_entry *entry;
	struct ltt_channel *chan = __data;
	struct ltt_event *event;

	if (unlikely(id >= ARRAY_SIZE(sc_table)))
		return;
	entry = &sc_table[id];
	if (unlikely(!entry->func))
		return;
	event = chan->sc_table[id];
	WARN_ON_ONCE(!event);

	switch (entry->nrargs) {
	case 0:
	{
		void (*fptr)(void *__data) = entry->func;

		fptr(event);
		break;
	}
	case 1:
	{
		void (*fptr)(void *__data, unsigned long arg0) = entry->func;
		unsigned long args[1];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0]);
		break;
	}
	case 2:
	{
		void (*fptr)(void *__data,
			unsigned long arg0,
			unsigned long arg1) = entry->func;
		unsigned long args[2];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0], args[1]);
		break;
	}
	case 3:
	{
		void (*fptr)(void *__data,
			unsigned long arg0,
			unsigned long arg1,
			unsigned long arg2) = entry->func;
		unsigned long args[3];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0], args[1], args[2]);
		break;
	}
	case 4:
	{
		void (*fptr)(void *__data,
			unsigned long arg0,
			unsigned long arg1,
			unsigned long arg2,
			unsigned long arg3) = entry->func;
		unsigned long args[4];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0], args[1], args[2], args[3]);
		break;
	}
	case 5:
	{
		void (*fptr)(void *__data,
			unsigned long arg0,
			unsigned long arg1,
			unsigned long arg2,
			unsigned long arg3,
			unsigned long arg4) = entry->func;
		unsigned long args[5];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0], args[1], args[2], args[3], args[4]);
		break;
	}
	case 6:
	{
		void (*fptr)(void *__data,
			unsigned long arg0,
			unsigned long arg1,
			unsigned long arg2,
			unsigned long arg3,
			unsigned long arg4,
			unsigned long arg5) = entry->func;
		unsigned long args[6];

		syscall_get_arguments(current, regs, 0, entry->nrargs, args);
		fptr(event, args[0], args[1], args[2],
			args[3], args[4], args[5]);
		break;
	}
	default:
		break;
	}
}

static void fill_sc_table(void)
{
	int i;

	if (sc_table_filled) {
		smp_rmb();	/* read flag before table */
		return;
	}

	for (i = 0; i < ARRAY_SIZE(sc_table); i++) {
		if (sc_table[i].func)
			continue;
		sc_table[i].func = __event_probe__sys_unknown;
		sc_table[i].nrargs = UNKNOWN_SYSCALL_NRARGS;
		sc_table[i].fields = __event_fields___sys_unknown;
		sc_table[i].desc = &__event_desc___sys_unknown;
	}
	smp_wmb();	/* Fill sc table before set flag to 1 */
	sc_table_filled = 1;
}

int lttng_syscalls_register(struct ltt_channel *chan, void *filter)
{
	unsigned int i;
	int ret;

	wrapper_vmalloc_sync_all();

	fill_sc_table();

	if (!chan->sc_table) {
		/* create syscall table mapping syscall to events */
		chan->sc_table = kzalloc(sizeof(struct ltt_event *)
					* ARRAY_SIZE(sc_table), GFP_KERNEL);
		if (!chan->sc_table)
			return -ENOMEM;
	}

	/* Allocate events for each syscall, insert into table */
	for (i = 0; i < ARRAY_SIZE(sc_table); i++) {
		struct lttng_kernel_event ev;
		const struct lttng_event_desc *desc = sc_table[i].desc;

		WARN_ON_ONCE(!desc);
		/*
		 * Skip those already populated by previous failed
		 * register for this channel.
		 */
		if (chan->sc_table[i])
			continue;
		memset(&ev, 0, sizeof(ev));
		strncpy(ev.name, desc->name, LTTNG_SYM_NAME_LEN);
		ev.name[LTTNG_SYM_NAME_LEN - 1] = '\0';
		ev.instrumentation = LTTNG_KERNEL_NOOP;
		chan->sc_table[i] = ltt_event_create(chan, &ev, filter,
						     desc);
		if (!chan->sc_table[i]) {
			/*
			 * If something goes wrong in event registration
			 * after the first one, we have no choice but to
			 * leave the previous events in there, until
			 * deleted by session teardown.
			 */
			return -EINVAL;
		}
	}
	ret = tracepoint_probe_register("sys_enter",
			(void *) syscall_entry_probe, chan);
	if (ret)
		return ret;
	/*
	 * We change the name of sys_exit tracepoint due to namespace
	 * conflict with sys_exit syscall entry.
	 */
	ret = tracepoint_probe_register("sys_exit",
			(void *) __event_probe__exit_syscall, chan);
	if (ret) {
		WARN_ON_ONCE(tracepoint_probe_unregister("sys_enter",
			(void *) syscall_entry_probe, chan));
	}
	return ret;
}

/*
 * Only called at session destruction.
 */
int lttng_syscalls_unregister(struct ltt_channel *chan)
{
	int ret;

	if (!chan->sc_table)
		return 0;
	ret = tracepoint_probe_unregister("sys_exit",
			(void *) __event_probe__exit_syscall, chan);
	if (ret)
		return ret;
	ret = tracepoint_probe_unregister("sys_enter",
			(void *) syscall_entry_probe, chan);
	if (ret)
		return ret;
	/* ltt_event destroy will be performed by ltt_session_destroy() */
	kfree(chan->sc_table);
	return 0;
}
Commit	Line	Data
259b6cb3 MD	1	/*
	2	* lttng-syscalls.c
	3	*
	4	* Copyright 2010 (c) - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
	5	*
	6	* LTTng sched probes.
	7	*
	8	* Dual LGPL v2.1/GPL v2 license.
	9	*/
	10
	11	#include <linux/module.h>
	12	#include <linux/slab.h>
	13	#include <asm/ptrace.h>
	14	#include <asm/syscall.h>
	15
	16	#include "ltt-events.h"
	17
	18	static void syscall_entry_probe(void __data, struct pt_regs regs, long id);
	19
f7bdf4db MD	20	/*
	21	* Take care of NOARGS not supported by mainline.
	22	*/
	23	#define DECLARE_EVENT_CLASS_NOARGS(name, tstruct, assign, print)
	24	#define DEFINE_EVENT_NOARGS(template, name)
	25	#define TRACE_EVENT_NOARGS(name, struct, assign, print)
	26
259b6cb3 MD	27	/*
	28	* Create LTTng tracepoint probes.
	29	*/
	30	#define LTTNG_PACKAGE_BUILD
	31	#define CREATE_TRACE_POINTS
	32
	33	/* Hijack probe callback for system calls */
	34	#define TP_PROBE_CB(_template) &syscall_entry_probe
	35	#define TP_MODULE_OVERRIDE
	36
	37	#define TRACE_INCLUDE_PATH ../instrumentation/syscalls/headers
177b3692 MD	38	#include "instrumentation/syscalls/headers/syscalls_integers.h"
177b3692 MD	39	#include "instrumentation/syscalls/headers/syscalls_pointers.h"
63728b02	40	#include "instrumentation/syscalls/headers/syscalls_unknown.h"
259b6cb3 MD	41
	42	#undef TP_MODULE_OVERRIDE
	43	#undef TP_PROBE_CB
	44	#undef LTTNG_PACKAGE_BUILD
	45	#undef CREATE_TRACE_POINTS
	46
	47	struct trace_syscall_entry {
	48	void *func;
f7bdf4db	49	const struct lttng_event_desc *desc;
259b6cb3 MD	50	const struct lttng_event_field *fields;
	51	unsigned int nrargs;
	52	};
	53
259b6cb3 MD	54	#define CREATE_SYSCALL_TABLE
	55
	56	#undef TRACE_SYSCALL_TABLE
f7bdf4db	57	#define TRACE_SYSCALL_TABLE(_template, _name, _nr, _nrargs) \
259b6cb3	58	[ _nr ] = { \
f7bdf4db	59	.func = __event_probe__##_template, \
259b6cb3	60	.nrargs = (_nrargs), \
f7bdf4db MD	61	.fields = __event_fields___##_template, \
f7bdf4db MD	62	.desc = &__event_desc___##_name, \
259b6cb3 MD	63	},
	64
	65	static struct trace_syscall_entry sc_table[] = {
177b3692 MD	66	#include "instrumentation/syscalls/headers/syscalls_integers.h"
177b3692 MD	67	#include "instrumentation/syscalls/headers/syscalls_pointers.h"
259b6cb3 MD	68	};
259b6cb3 MD	69
63728b02 MD	70	static int sc_table_filled;
63728b02 MD	71
259b6cb3 MD	72	#undef CREATE_SYSCALL_TABLE
	73
	74	static void syscall_entry_probe(void __data, struct pt_regs regs, long id)
	75	{
	76	struct trace_syscall_entry *entry;
	77	struct ltt_channel *chan = __data;
	78	struct ltt_event *event;
	79
	80	if (unlikely(id >= ARRAY_SIZE(sc_table)))
	81	return;
	82	entry = &sc_table[id];
	83	if (unlikely(!entry->func))
	84	return;
	85	event = chan->sc_table[id];
	86	WARN_ON_ONCE(!event);
	87
	88	switch (entry->nrargs) {
	89	case 0:
	90	{
	91	void (fptr)(void __data) = entry->func;
	92
	93	fptr(event);
	94	break;
	95	}
	96	case 1:
	97	{
	98	void (fptr)(void __data, unsigned long arg0) = entry->func;
	99	unsigned long args[1];
	100
	101	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
	102	fptr(event, args[0]);
	103	break;
	104	}
	105	case 2:
	106	{
	107	void (fptr)(void __data,
	108	unsigned long arg0,
	109	unsigned long arg1) = entry->func;
	110	unsigned long args[2];
	111
	112	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
	113	fptr(event, args[0], args[1]);
	114	break;
	115	}
	116	case 3:
	117	{
	118	void (fptr)(void __data,
	119	unsigned long arg0,
	120	unsigned long arg1,
	121	unsigned long arg2) = entry->func;
	122	unsigned long args[3];
	123
	124	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
	125	fptr(event, args[0], args[1], args[2]);
	126	break;
	127	}
	128	case 4:
	129	{
	130	void (fptr)(void __data,
	131	unsigned long arg0,
	132	unsigned long arg1,
	133	unsigned long arg2,
	134	unsigned long arg3) = entry->func;
	135	unsigned long args[4];
136
137	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
138	fptr(event, args[0], args[1], args[2], args[3]);
139	break;
140	}
141	case 5:
142	{
143	void (fptr)(void __data,
144	unsigned long arg0,
145	unsigned long arg1,
146	unsigned long arg2,
147	unsigned long arg3,
148	unsigned long arg4) = entry->func;
149	unsigned long args[5];
150
151	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
152	fptr(event, args[0], args[1], args[2], args[3], args[4]);
153	break;
154	}
155	case 6:
156	{
157	void (fptr)(void __data,
158	unsigned long arg0,
159	unsigned long arg1,
160	unsigned long arg2,
161	unsigned long arg3,
162	unsigned long arg4,
163	unsigned long arg5) = entry->func;
164	unsigned long args[6];
165
166	syscall_get_arguments(current, regs, 0, entry->nrargs, args);
167	fptr(event, args[0], args[1], args[2],
168	args[3], args[4], args[5]);
169	break;
170	}
171	default:
172	break;
173	}
174	}
175
63728b02 MD	176	static void fill_sc_table(void)
	177	{
	178	int i;
	179
	180	if (sc_table_filled) {
	181	smp_rmb(); /* read flag before table */
	182	return;
	183	}
	184
	185	for (i = 0; i < ARRAY_SIZE(sc_table); i++) {
	186	if (sc_table[i].func)
	187	continue;
	188	sc_table[i].func = __event_probe__sys_unknown;
	189	sc_table[i].nrargs = UNKNOWN_SYSCALL_NRARGS;
	190	sc_table[i].fields = __event_fields___sys_unknown;
	191	sc_table[i].desc = &__event_desc___sys_unknown;
	192	}
	193	smp_wmb(); /* Fill sc table before set flag to 1 */
	194	sc_table_filled = 1;
	195	}
	196
259b6cb3 MD	197	int lttng_syscalls_register(struct ltt_channel chan, void filter)
	198	{
	199	unsigned int i;
	200	int ret;
	201
	202	wrapper_vmalloc_sync_all();
259b6cb3	203
63728b02 MD	204	fill_sc_table();
63728b02 MD	205
259b6cb3 MD	206	if (!chan->sc_table) {
	207	/* create syscall table mapping syscall to events */
	208	chan->sc_table = kzalloc(sizeof(struct ltt_event *)
	209	* ARRAY_SIZE(sc_table), GFP_KERNEL);
	210	if (!chan->sc_table)
	211	return -ENOMEM;
	212	}
	213
	214	/* Allocate events for each syscall, insert into table */
	215	for (i = 0; i < ARRAY_SIZE(sc_table); i++) {
	216	struct lttng_kernel_event ev;
	217	const struct lttng_event_desc *desc = sc_table[i].desc;
	218
63728b02	219	WARN_ON_ONCE(!desc);
259b6cb3 MD	220	/*
	221	* Skip those already populated by previous failed
	222	* register for this channel.
	223	*/
	224	if (chan->sc_table[i])
	225	continue;
	226	memset(&ev, 0, sizeof(ev));
	227	strncpy(ev.name, desc->name, LTTNG_SYM_NAME_LEN);
	228	ev.name[LTTNG_SYM_NAME_LEN - 1] = '\0';
	229	ev.instrumentation = LTTNG_KERNEL_NOOP;
	230	chan->sc_table[i] = ltt_event_create(chan, &ev, filter,
	231	desc);
	232	if (!chan->sc_table[i]) {
	233	/*
	234	* If something goes wrong in event registration
	235	* after the first one, we have no choice but to
	236	* leave the previous events in there, until
	237	* deleted by session teardown.
	238	*/
	239	return -EINVAL;
	240	}
	241	}
	242	ret = tracepoint_probe_register("sys_enter",
	243	(void *) syscall_entry_probe, chan);
63728b02 MD	244	if (ret)
	245	return ret;
	246	/*
	247	* We change the name of sys_exit tracepoint due to namespace
	248	* conflict with sys_exit syscall entry.
	249	*/
	250	ret = tracepoint_probe_register("sys_exit",
	251	(void *) __event_probe__exit_syscall, chan);
	252	if (ret) {
	253	WARN_ON_ONCE(tracepoint_probe_unregister("sys_enter",
	254	(void *) syscall_entry_probe, chan));
	255	}
259b6cb3 MD	256	return ret;
	257	}
	258
	259	/*
	260	* Only called at session destruction.
	261	*/
	262	int lttng_syscalls_unregister(struct ltt_channel *chan)
	263	{
	264	int ret;
	265
	266	if (!chan->sc_table)
	267	return 0;
63728b02 MD	268	ret = tracepoint_probe_unregister("sys_exit",
	269	(void *) __event_probe__exit_syscall, chan);
	270	if (ret)
	271	return ret;
259b6cb3 MD	272	ret = tracepoint_probe_unregister("sys_enter",
	273	(void *) syscall_entry_probe, chan);
	274	if (ret)
	275	return ret;
	276	/* ltt_event destroy will be performed by ltt_session_destroy() */
	277	kfree(chan->sc_table);
	278	return 0;
	279	}