[deliverable/lttng-modules.git] / lttng-statedump-impl.c

/*
 * Linux Trace Toolkit Next Generation Kernel State Dump
 *
 * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
 * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *
 * Changes:
 *	Eric Clement:                   Add listing of network IP interface
 *	2006, 2007 Mathieu Desnoyers	Fix kernel threads
 *	                                Various updates
 *
 * Dual LGPL v2.1/GPL v2 license.
 */

#include <linux/init.h>
#include <linux/module.h>
#include <linux/netlink.h>
#include <linux/inet.h>
#include <linux/ip.h>
#include <linux/kthread.h>
#include <linux/proc_fs.h>
#include <linux/file.h>
#include <linux/interrupt.h>
#include <linux/irqnr.h>
#include <linux/cpu.h>
#include <linux/netdevice.h>
#include <linux/inetdevice.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/fdtable.h>
#include <linux/swap.h>
#include <linux/wait.h>
#include <linux/mutex.h>

#include "lttng-events.h"
#include "wrapper/irqdesc.h"

#ifdef CONFIG_GENERIC_HARDIRQS
#include <linux/irq.h>
#endif

/* Define the tracepoints, but do not build the probes */
#define CREATE_TRACE_POINTS
#define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
#define TRACE_INCLUDE_FILE lttng-statedump
#include "instrumentation/events/lttng-module/lttng-statedump.h"

/*
 * Protected by the trace lock.
 */
static struct delayed_work cpu_work[NR_CPUS];
static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
static atomic_t kernel_threads_to_run;

enum lttng_thread_type {
	LTTNG_USER_THREAD = 0,
	LTTNG_KERNEL_THREAD = 1,
};

enum lttng_execution_mode {
	LTTNG_USER_MODE = 0,
	LTTNG_SYSCALL = 1,
	LTTNG_TRAP = 2,
	LTTNG_IRQ = 3,
	LTTNG_SOFTIRQ = 4,
	LTTNG_MODE_UNKNOWN = 5,
};

enum lttng_execution_submode {
	LTTNG_NONE = 0,
	LTTNG_UNKNOWN = 1,
};

enum lttng_process_status {
	LTTNG_UNNAMED = 0,
	LTTNG_WAIT_FORK = 1,
	LTTNG_WAIT_CPU = 2,
	LTTNG_EXIT = 3,
	LTTNG_ZOMBIE = 4,
	LTTNG_WAIT = 5,
	LTTNG_RUN = 6,
	LTTNG_DEAD = 7,
};

#ifdef CONFIG_INET
static
void lttng_enumerate_device(struct lttng_session *session,
		struct net_device *dev)
{
	struct in_device *in_dev;
	struct in_ifaddr *ifa;

	if (dev->flags & IFF_UP) {
		in_dev = in_dev_get(dev);
		if (in_dev) {
			for (ifa = in_dev->ifa_list; ifa != NULL;
			     ifa = ifa->ifa_next) {
				trace_lttng_statedump_network_interface(
					session, dev, ifa);
			}
			in_dev_put(in_dev);
		}
	} else {
		trace_lttng_statedump_network_interface(
			session, dev, NULL);
	}
}

static
int lttng_enumerate_network_ip_interface(struct lttng_session *session)
{
	struct net_device *dev;

	read_lock(&dev_base_lock);
	for_each_netdev(&init_net, dev)
		lttng_enumerate_device(session, dev);
	read_unlock(&dev_base_lock);

	return 0;
}
#else /* CONFIG_INET */
static inline
int lttng_enumerate_network_ip_interface(struct lttng_session *session)
{
	return 0;
}
#endif /* CONFIG_INET */


static
void lttng_enumerate_task_fd(struct lttng_session *session,
		struct task_struct *p, char *tmp)
{
	struct fdtable *fdt;
	struct file *filp;
	unsigned int i;
	const unsigned char *path;

	task_lock(p);
	if (!p->files)
		goto unlock_task;
	spin_lock(&p->files->file_lock);
	fdt = files_fdtable(p->files);
	for (i = 0; i < fdt->max_fds; i++) {
		filp = fcheck_files(p->files, i);
		if (!filp)
			continue;
		path = d_path(&filp->f_path, tmp, PAGE_SIZE);
		/* Make sure we give at least some info */
		trace_lttng_statedump_file_descriptor(session, p, i,
			IS_ERR(path) ?
				filp->f_dentry->d_name.name :
				path);
	}
	spin_unlock(&p->files->file_lock);
unlock_task:
	task_unlock(p);
}

static
int lttng_enumerate_file_descriptors(struct lttng_session *session)
{
	struct task_struct *p;
	char *tmp = (char *) __get_free_page(GFP_KERNEL);

	/* Enumerate active file descriptors */
	rcu_read_lock();
	for_each_process(p)
		lttng_enumerate_task_fd(session, p, tmp);
	rcu_read_unlock();
	free_page((unsigned long) tmp);
	return 0;
}

static
void lttng_enumerate_task_vm_maps(struct lttng_session *session,
		struct task_struct *p)
{
	struct mm_struct *mm;
	struct vm_area_struct *map;
	unsigned long ino;

	/* get_task_mm does a task_lock... */
	mm = get_task_mm(p);
	if (!mm)
		return;

	map = mm->mmap;
	if (map) {
		down_read(&mm->mmap_sem);
		while (map) {
			if (map->vm_file)
				ino = map->vm_file->f_dentry->d_inode->i_ino;
			else
				ino = 0;
			trace_lttng_statedump_vm_map(session, p, map, ino);
			map = map->vm_next;
		}
		up_read(&mm->mmap_sem);
	}
	mmput(mm);
}

static
int lttng_enumerate_vm_maps(struct lttng_session *session)
{
	struct task_struct *p;

	rcu_read_lock();
	for_each_process(p)
		lttng_enumerate_task_vm_maps(session, p);
	rcu_read_unlock();
	return 0;
}

#ifdef CONFIG_GENERIC_HARDIRQS

#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
#endif

static
void lttng_list_interrupts(struct lttng_session *session)
{
	unsigned int irq;
	unsigned long flags = 0;
	struct irq_desc *desc;

#define irq_to_desc	wrapper_irq_to_desc
	/* needs irq_desc */
	for_each_irq_desc(irq, desc) {
		struct irqaction *action;
		const char *irq_chip_name =
			irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";

		local_irq_save(flags);
		raw_spin_lock(&desc->lock);
		for (action = desc->action; action; action = action->next) {
			trace_lttng_statedump_interrupt(session,
				irq, irq_chip_name, action);
		}
		raw_spin_unlock(&desc->lock);
		local_irq_restore(flags);
	}
#undef irq_to_desc
}
#else
static inline
void list_interrupts(struct lttng_session *session)
{
}
#endif

static
int lttng_enumerate_process_states(struct lttng_session *session)
{
	struct task_struct *g, *p;

	rcu_read_lock();
	for_each_process(g) {
		p = g;
		do {
			enum lttng_execution_mode mode =
				LTTNG_MODE_UNKNOWN;
			enum lttng_execution_submode submode =
				LTTNG_UNKNOWN;
			enum lttng_process_status status;
			enum lttng_thread_type type;

			task_lock(p);
			if (p->exit_state == EXIT_ZOMBIE)
				status = LTTNG_ZOMBIE;
			else if (p->exit_state == EXIT_DEAD)
				status = LTTNG_DEAD;
			else if (p->state == TASK_RUNNING) {
				/* Is this a forked child that has not run yet? */
				if (list_empty(&p->rt.run_list))
					status = LTTNG_WAIT_FORK;
				else
					/*
					 * All tasks are considered as wait_cpu;
					 * the viewer will sort out if the task
					 * was really running at this time.
					 */
					status = LTTNG_WAIT_CPU;
			} else if (p->state &
				(TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
				/* Task is waiting for something to complete */
				status = LTTNG_WAIT;
			} else
				status = LTTNG_UNNAMED;
			submode = LTTNG_NONE;

			/*
			 * Verification of t->mm is to filter out kernel
			 * threads; Viewer will further filter out if a
			 * user-space thread was in syscall mode or not.
			 */
			if (p->mm)
				type = LTTNG_USER_THREAD;
			else
				type = LTTNG_KERNEL_THREAD;
			trace_lttng_statedump_process_state(session,
				p, type, mode, submode, status);
			task_unlock(p);
		} while_each_thread(g, p);
	}
	rcu_read_unlock();

	return 0;
}

static
void lttng_statedump_work_func(struct work_struct *work)
{
	if (atomic_dec_and_test(&kernel_threads_to_run))
		/* If we are the last thread, wake up do_lttng_statedump */
		wake_up(&statedump_wq);
}

static
int do_lttng_statedump(struct lttng_session *session)
{
	int cpu;

	printk(KERN_DEBUG "LTT state dump thread start\n");
	trace_lttng_statedump_start(session);
	lttng_enumerate_process_states(session);
	lttng_enumerate_file_descriptors(session);
	lttng_enumerate_vm_maps(session);
	lttng_list_interrupts(session);
	lttng_enumerate_network_ip_interface(session);

	/* TODO lttng_dump_idt_table(session); */
	/* TODO lttng_dump_softirq_vec(session); */
	/* TODO lttng_list_modules(session); */
	/* TODO lttng_dump_swap_files(session); */

	/*
	 * Fire off a work queue on each CPU. Their sole purpose in life
	 * is to guarantee that each CPU has been in a state where is was in
	 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
	 */
	get_online_cpus();
	atomic_set(&kernel_threads_to_run, num_online_cpus());
	for_each_online_cpu(cpu) {
		INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
		schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
	}
	/* Wait for all threads to run */
	__wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) != 0));
	put_online_cpus();
	/* Our work is done */
	printk(KERN_DEBUG "LTT state dump end\n");
	trace_lttng_statedump_end(session);
	return 0;
}

/*
 * Called with session mutex held.
 */
int lttng_statedump_start(struct lttng_session *session)
{
	printk(KERN_DEBUG "LTTng: state dump begin\n");
	return do_lttng_statedump(session);
}
EXPORT_SYMBOL_GPL(lttng_statedump_start);

MODULE_LICENSE("GPL and additional rights");
MODULE_AUTHOR("Jean-Hugues Deschenes");
MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
Commit	Line	Data
	1	/*
	2	* Linux Trace Toolkit Next Generation Kernel State Dump
	3	*
	4	* Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
	5	* Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
	6	*
	7	* Changes:
	8	* Eric Clement: Add listing of network IP interface
	9	* 2006, 2007 Mathieu Desnoyers Fix kernel threads
	10	* Various updates
	11	*
	12	* Dual LGPL v2.1/GPL v2 license.
	13	*/
	14
	15	#include <linux/init.h>
	16	#include <linux/module.h>
	17	#include <linux/netlink.h>
	18	#include <linux/inet.h>
	19	#include <linux/ip.h>
	20	#include <linux/kthread.h>
	21	#include <linux/proc_fs.h>
	22	#include <linux/file.h>
	23	#include <linux/interrupt.h>
	24	#include <linux/irqnr.h>
	25	#include <linux/cpu.h>
	26	#include <linux/netdevice.h>
	27	#include <linux/inetdevice.h>
	28	#include <linux/sched.h>
	29	#include <linux/mm.h>
	30	#include <linux/fdtable.h>
	31	#include <linux/swap.h>
	32	#include <linux/wait.h>
	33	#include <linux/mutex.h>
	34
	35	#include "lttng-events.h"
	36	#include "wrapper/irqdesc.h"
	37
	38	#ifdef CONFIG_GENERIC_HARDIRQS
	39	#include <linux/irq.h>
	40	#endif
	41
	42	/* Define the tracepoints, but do not build the probes */
	43	#define CREATE_TRACE_POINTS
	44	#define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
	45	#define TRACE_INCLUDE_FILE lttng-statedump
	46	#include "instrumentation/events/lttng-module/lttng-statedump.h"
	47
	48	/*
	49	* Protected by the trace lock.
	50	*/
	51	static struct delayed_work cpu_work[NR_CPUS];
	52	static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
	53	static atomic_t kernel_threads_to_run;
	54
	55	enum lttng_thread_type {
	56	LTTNG_USER_THREAD = 0,
	57	LTTNG_KERNEL_THREAD = 1,
	58	};
	59
	60	enum lttng_execution_mode {
	61	LTTNG_USER_MODE = 0,
	62	LTTNG_SYSCALL = 1,
	63	LTTNG_TRAP = 2,
	64	LTTNG_IRQ = 3,
	65	LTTNG_SOFTIRQ = 4,
	66	LTTNG_MODE_UNKNOWN = 5,
	67	};
	68
	69	enum lttng_execution_submode {
	70	LTTNG_NONE = 0,
	71	LTTNG_UNKNOWN = 1,
	72	};
	73
	74	enum lttng_process_status {
	75	LTTNG_UNNAMED = 0,
	76	LTTNG_WAIT_FORK = 1,
	77	LTTNG_WAIT_CPU = 2,
	78	LTTNG_EXIT = 3,
	79	LTTNG_ZOMBIE = 4,
	80	LTTNG_WAIT = 5,
	81	LTTNG_RUN = 6,
	82	LTTNG_DEAD = 7,
	83	};
	84
	85	#ifdef CONFIG_INET
	86	static
	87	void lttng_enumerate_device(struct lttng_session *session,
	88	struct net_device *dev)
	89	{
	90	struct in_device *in_dev;
	91	struct in_ifaddr *ifa;
	92
	93	if (dev->flags & IFF_UP) {
	94	in_dev = in_dev_get(dev);
	95	if (in_dev) {
	96	for (ifa = in_dev->ifa_list; ifa != NULL;
	97	ifa = ifa->ifa_next) {
	98	trace_lttng_statedump_network_interface(
	99	session, dev, ifa);
	100	}
	101	in_dev_put(in_dev);
	102	}
	103	} else {
	104	trace_lttng_statedump_network_interface(
	105	session, dev, NULL);
	106	}
	107	}
	108
	109	static
	110	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	111	{
	112	struct net_device *dev;
	113
	114	read_lock(&dev_base_lock);
	115	for_each_netdev(&init_net, dev)
	116	lttng_enumerate_device(session, dev);
	117	read_unlock(&dev_base_lock);
	118
	119	return 0;
	120	}
	121	#else /* CONFIG_INET */
	122	static inline
	123	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	124	{
	125	return 0;
	126	}
	127	#endif /* CONFIG_INET */
	128
	129
	130	static
	131	void lttng_enumerate_task_fd(struct lttng_session *session,
	132	struct task_struct p, char tmp)
	133	{
	134	struct fdtable *fdt;
	135	struct file *filp;
	136	unsigned int i;
	137	const unsigned char *path;
	138
	139	task_lock(p);
	140	if (!p->files)
	141	goto unlock_task;
	142	spin_lock(&p->files->file_lock);
	143	fdt = files_fdtable(p->files);
	144	for (i = 0; i < fdt->max_fds; i++) {
	145	filp = fcheck_files(p->files, i);
	146	if (!filp)
	147	continue;
	148	path = d_path(&filp->f_path, tmp, PAGE_SIZE);
	149	/* Make sure we give at least some info */
	150	trace_lttng_statedump_file_descriptor(session, p, i,
	151	IS_ERR(path) ?
	152	filp->f_dentry->d_name.name :
	153	path);
	154	}
	155	spin_unlock(&p->files->file_lock);
	156	unlock_task:
	157	task_unlock(p);
	158	}
	159
	160	static
	161	int lttng_enumerate_file_descriptors(struct lttng_session *session)
	162	{
	163	struct task_struct *p;
	164	char tmp = (char ) __get_free_page(GFP_KERNEL);
	165
	166	/* Enumerate active file descriptors */
	167	rcu_read_lock();
	168	for_each_process(p)
	169	lttng_enumerate_task_fd(session, p, tmp);
	170	rcu_read_unlock();
	171	free_page((unsigned long) tmp);
	172	return 0;
	173	}
	174
	175	static
	176	void lttng_enumerate_task_vm_maps(struct lttng_session *session,
	177	struct task_struct *p)
	178	{
	179	struct mm_struct *mm;
	180	struct vm_area_struct *map;
	181	unsigned long ino;
	182
	183	/* get_task_mm does a task_lock... */
	184	mm = get_task_mm(p);
	185	if (!mm)
	186	return;
	187
	188	map = mm->mmap;
	189	if (map) {
	190	down_read(&mm->mmap_sem);
	191	while (map) {
	192	if (map->vm_file)
	193	ino = map->vm_file->f_dentry->d_inode->i_ino;
	194	else
	195	ino = 0;
	196	trace_lttng_statedump_vm_map(session, p, map, ino);
	197	map = map->vm_next;
	198	}
	199	up_read(&mm->mmap_sem);
	200	}
	201	mmput(mm);
	202	}
	203
	204	static
	205	int lttng_enumerate_vm_maps(struct lttng_session *session)
	206	{
	207	struct task_struct *p;
	208
	209	rcu_read_lock();
	210	for_each_process(p)
	211	lttng_enumerate_task_vm_maps(session, p);
	212	rcu_read_unlock();
	213	return 0;
	214	}
	215
	216	#ifdef CONFIG_GENERIC_HARDIRQS
	217
	218	#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
	219	#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
	220	#endif
	221
	222	static
	223	void lttng_list_interrupts(struct lttng_session *session)
	224	{
	225	unsigned int irq;
	226	unsigned long flags = 0;
	227	struct irq_desc *desc;
	228
	229	#define irq_to_desc wrapper_irq_to_desc
	230	/* needs irq_desc */
	231	for_each_irq_desc(irq, desc) {
	232	struct irqaction *action;
	233	const char *irq_chip_name =
	234	irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
	235
	236	local_irq_save(flags);
	237	raw_spin_lock(&desc->lock);
	238	for (action = desc->action; action; action = action->next) {
	239	trace_lttng_statedump_interrupt(session,
	240	irq, irq_chip_name, action);
	241	}
	242	raw_spin_unlock(&desc->lock);
	243	local_irq_restore(flags);
	244	}
	245	#undef irq_to_desc
	246	}
	247	#else
	248	static inline
	249	void list_interrupts(struct lttng_session *session)
	250	{
	251	}
	252	#endif
	253
	254	static
	255	int lttng_enumerate_process_states(struct lttng_session *session)
	256	{
	257	struct task_struct g, p;
	258
	259	rcu_read_lock();
	260	for_each_process(g) {
	261	p = g;
	262	do {
	263	enum lttng_execution_mode mode =
	264	LTTNG_MODE_UNKNOWN;
	265	enum lttng_execution_submode submode =
	266	LTTNG_UNKNOWN;
	267	enum lttng_process_status status;
	268	enum lttng_thread_type type;
	269
	270	task_lock(p);
	271	if (p->exit_state == EXIT_ZOMBIE)
	272	status = LTTNG_ZOMBIE;
	273	else if (p->exit_state == EXIT_DEAD)
	274	status = LTTNG_DEAD;
	275	else if (p->state == TASK_RUNNING) {
	276	/* Is this a forked child that has not run yet? */
	277	if (list_empty(&p->rt.run_list))
	278	status = LTTNG_WAIT_FORK;
	279	else
	280	/*
	281	* All tasks are considered as wait_cpu;
	282	* the viewer will sort out if the task
	283	* was really running at this time.
	284	*/
	285	status = LTTNG_WAIT_CPU;
	286	} else if (p->state &
	287	(TASK_INTERRUPTIBLE \| TASK_UNINTERRUPTIBLE)) {
	288	/* Task is waiting for something to complete */
	289	status = LTTNG_WAIT;
	290	} else
	291	status = LTTNG_UNNAMED;
	292	submode = LTTNG_NONE;
	293
	294	/*
	295	* Verification of t->mm is to filter out kernel
	296	* threads; Viewer will further filter out if a
	297	* user-space thread was in syscall mode or not.
	298	*/
	299	if (p->mm)
	300	type = LTTNG_USER_THREAD;
	301	else
	302	type = LTTNG_KERNEL_THREAD;
	303	trace_lttng_statedump_process_state(session,
	304	p, type, mode, submode, status);
	305	task_unlock(p);
	306	} while_each_thread(g, p);
	307	}
	308	rcu_read_unlock();
	309
	310	return 0;
	311	}
	312
	313	static
	314	void lttng_statedump_work_func(struct work_struct *work)
	315	{
	316	if (atomic_dec_and_test(&kernel_threads_to_run))
	317	/* If we are the last thread, wake up do_lttng_statedump */
	318	wake_up(&statedump_wq);
	319	}
	320
	321	static
	322	int do_lttng_statedump(struct lttng_session *session)
	323	{
	324	int cpu;
	325
	326	printk(KERN_DEBUG "LTT state dump thread start\n");
	327	trace_lttng_statedump_start(session);
	328	lttng_enumerate_process_states(session);
	329	lttng_enumerate_file_descriptors(session);
	330	lttng_enumerate_vm_maps(session);
	331	lttng_list_interrupts(session);
	332	lttng_enumerate_network_ip_interface(session);
	333
	334	/* TODO lttng_dump_idt_table(session); */
	335	/* TODO lttng_dump_softirq_vec(session); */
	336	/* TODO lttng_list_modules(session); */
	337	/* TODO lttng_dump_swap_files(session); */
	338
	339	/*
	340	* Fire off a work queue on each CPU. Their sole purpose in life
	341	* is to guarantee that each CPU has been in a state where is was in
	342	* syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
	343	*/
	344	get_online_cpus();
	345	atomic_set(&kernel_threads_to_run, num_online_cpus());
	346	for_each_online_cpu(cpu) {
	347	INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
	348	schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
	349	}
	350	/* Wait for all threads to run */
	351	__wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) != 0));
	352	put_online_cpus();
	353	/* Our work is done */
	354	printk(KERN_DEBUG "LTT state dump end\n");
	355	trace_lttng_statedump_end(session);
	356	return 0;
	357	}
	358
	359	/*
	360	* Called with session mutex held.
	361	*/
	362	int lttng_statedump_start(struct lttng_session *session)
	363	{
	364	printk(KERN_DEBUG "LTTng: state dump begin\n");
	365	return do_lttng_statedump(session);
	366	}
	367	EXPORT_SYMBOL_GPL(lttng_statedump_start);
	368
	369	MODULE_LICENSE("GPL and additional rights");
	370	MODULE_AUTHOR("Jean-Hugues Deschenes");
	371	MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");