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

/* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
 *
 * lttng-statedump.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
 */

#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/swap.h>
#include <linux/wait.h>
#include <linux/mutex.h>
#include <linux/device.h>

#include <lttng-events.h>
#include <lttng-tracer.h>
#include <wrapper/irqdesc.h>
#include <wrapper/spinlock.h>
#include <wrapper/fdtable.h>
#include <wrapper/irq.h>
#include <wrapper/tracepoint.h>
#include <wrapper/genhd.h>
#include <wrapper/file.h>
#include <wrapper/fdtable.h>

#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
#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
#define LTTNG_INSTRUMENTATION
#include <instrumentation/events/lttng-module/lttng-statedump.h>

DEFINE_TRACE(lttng_statedump_block_device);
DEFINE_TRACE(lttng_statedump_end);
DEFINE_TRACE(lttng_statedump_interrupt);
DEFINE_TRACE(lttng_statedump_file_descriptor);
DEFINE_TRACE(lttng_statedump_start);
DEFINE_TRACE(lttng_statedump_process_state);
DEFINE_TRACE(lttng_statedump_network_interface);

struct lttng_fd_ctx {
	char *page;
	struct lttng_session *session;
	struct task_struct *p;
	struct files_struct *files;
};

/*
 * 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,
};

static
int lttng_enumerate_block_devices(struct lttng_session *session)
{
	struct class *ptr_block_class;
	struct device_type *ptr_disk_type;
	struct class_dev_iter iter;
	struct device *dev;

	ptr_block_class = wrapper_get_block_class();
	if (!ptr_block_class)
		return -ENOSYS;
	ptr_disk_type = wrapper_get_disk_type();
	if (!ptr_disk_type) {
		return -ENOSYS;
	}
	class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
	while ((dev = class_dev_iter_next(&iter))) {
		struct disk_part_iter piter;
		struct gendisk *disk = dev_to_disk(dev);
		struct hd_struct *part;

		/*
		 * Don't show empty devices or things that have been
		 * suppressed
		 */
		if (get_capacity(disk) == 0 ||
		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
			continue;

		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
		while ((part = disk_part_iter_next(&piter))) {
			struct block_device bdev;
			char name_buf[BDEVNAME_SIZE];
			const char *p;

			/*
			 * Create a partial 'struct blockdevice' to use
			 * 'bdevname()' which is a simple wrapper over
			 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
			 */
			bdev.bd_disk = disk;
			bdev.bd_part = part;

			p = bdevname(&bdev, name_buf);
			if (!p) {
				disk_part_iter_exit(&piter);
				class_dev_iter_exit(&iter);
				return -ENOSYS;
			}
			trace_lttng_statedump_block_device(session,
					part_devt(part), name_buf);
		}
		disk_part_iter_exit(&piter);
	}
	class_dev_iter_exit(&iter);
	return 0;
}

#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
int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
{
	const struct lttng_fd_ctx *ctx = p;
	const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
	unsigned int flags = file->f_flags;
	struct fdtable *fdt;

	/*
	 * We don't expose kernel internal flags, only userspace-visible
	 * flags.
	 */
	flags &= ~FMODE_NONOTIFY;
	fdt = files_fdtable(ctx->files);
	/*
	 * We need to check here again whether fd is within the fdt
	 * max_fds range, because we might be seeing a different
	 * files_fdtable() than iterate_fd(), assuming only RCU is
	 * protecting the read. In reality, iterate_fd() holds
	 * file_lock, which should ensure the fdt does not change while
	 * the lock is taken, but we are not aware whether this is
	 * guaranteed or not, so play safe.
	 */
	if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
		flags |= O_CLOEXEC;
	if (IS_ERR(s)) {
		struct dentry *dentry = file->f_path.dentry;

		/* Make sure we give at least some info */
		spin_lock(&dentry->d_lock);
		trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
			dentry->d_name.name, flags, file->f_mode);
		spin_unlock(&dentry->d_lock);
		goto end;
	}
	trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
		flags, file->f_mode);
end:
	return 0;
}

static
void lttng_enumerate_task_fd(struct lttng_session *session,
		struct task_struct *p, char *tmp)
{
	struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
	struct files_struct *files;

	task_lock(p);
	files = p->files;
	if (!files)
		goto end;
	ctx.files = files;
	lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
end:
	task_unlock(p);
}

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

	tmp = (char *) __get_free_page(GFP_KERNEL);
	if (!tmp)
		return -ENOMEM;

	/* 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;
}

#if 0
/*
 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 * iteration, but it is not exported to modules.
 */
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->lttng_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;
}
#endif

#ifdef CONFIG_LTTNG_HAS_LIST_IRQ

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

static
int 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);
		wrapper_desc_spin_lock(&desc->lock);
		for (action = desc->action; action; action = action->next) {
			trace_lttng_statedump_interrupt(session,
				irq, irq_chip_name, action);
		}
		wrapper_desc_spin_unlock(&desc->lock);
		local_irq_restore(flags);
	}
	return 0;
#undef irq_to_desc
}
#else
static inline
int lttng_list_interrupts(struct lttng_session *session)
{
	return 0;
}
#endif

/*
 * Called with task lock held.
 */
static
void lttng_statedump_process_ns(struct lttng_session *session,
		struct task_struct *p,
		enum lttng_thread_type type,
		enum lttng_execution_mode mode,
		enum lttng_execution_submode submode,
		enum lttng_process_status status)
{
	struct pid_namespace *pid_ns;

	pid_ns = task_active_pid_ns(p);
	do {
		trace_lttng_statedump_process_state(session,
			p, type, mode, submode, status, pid_ns);
		pid_ns = pid_ns ? pid_ns->parent : NULL;
	} while (pid_ns);
}

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;
			lttng_statedump_process_ns(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, ret;

	trace_lttng_statedump_start(session);
	ret = lttng_enumerate_process_states(session);
	if (ret)
		return ret;
	ret = lttng_enumerate_file_descriptors(session);
	if (ret)
		return ret;
	/*
	 * FIXME
	 * ret = lttng_enumerate_vm_maps(session);
	 * if (ret)
	 * 	return ret;
	 */
	ret = lttng_list_interrupts(session);
	if (ret)
		return ret;
	ret = lttng_enumerate_network_ip_interface(session);
	if (ret)
		return ret;
	ret = lttng_enumerate_block_devices(session);
	switch (ret) {
	case 0:
		break;
	case -ENOSYS:
		printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
		break;
	default:
		return ret;
	}

	/* 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 */
	trace_lttng_statedump_end(session);
	return 0;
}

/*
 * Called with session mutex held.
 */
int lttng_statedump_start(struct lttng_session *session)
{
	return do_lttng_statedump(session);
}
EXPORT_SYMBOL_GPL(lttng_statedump_start);

static
int __init lttng_statedump_init(void)
{
	/*
	 * Allow module to load even if the fixup cannot be done. This
	 * will allow seemless transition when the underlying issue fix
	 * is merged into the Linux kernel, and when tracepoint.c
	 * "tracepoint_module_notify" is turned into a static function.
	 */
	(void) wrapper_lttng_fixup_sig(THIS_MODULE);
	return 0;
}

module_init(lttng_statedump_init);

static
void __exit lttng_statedump_exit(void)
{
}

module_exit(lttng_statedump_exit);

MODULE_LICENSE("GPL and additional rights");
MODULE_AUTHOR("Jean-Hugues Deschenes");
MODULE_DESCRIPTION("LTTng statedump provider");
MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
	__stringify(LTTNG_MODULES_MINOR_VERSION) "."
	__stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
	LTTNG_MODULES_EXTRAVERSION);
Commit	Line	Data
9f36eaed MJ	1	/* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
9f36eaed MJ	2	*
886d51a3 MD	3	* lttng-statedump.c
886d51a3 MD	4	*
c337ddc2 MD	5	* Linux Trace Toolkit Next Generation Kernel State Dump
	6	*
	7	* Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
	8	* Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
	9	*
	10	* Changes:
	11	* Eric Clement: Add listing of network IP interface
	12	* 2006, 2007 Mathieu Desnoyers Fix kernel threads
	13	* Various updates
c337ddc2 MD	14	*/
	15
	16	#include <linux/init.h>
	17	#include <linux/module.h>
	18	#include <linux/netlink.h>
	19	#include <linux/inet.h>
	20	#include <linux/ip.h>
	21	#include <linux/kthread.h>
	22	#include <linux/proc_fs.h>
	23	#include <linux/file.h>
	24	#include <linux/interrupt.h>
	25	#include <linux/irqnr.h>
	26	#include <linux/cpu.h>
	27	#include <linux/netdevice.h>
	28	#include <linux/inetdevice.h>
	29	#include <linux/sched.h>
	30	#include <linux/mm.h>
c337ddc2 MD	31	#include <linux/swap.h>
	32	#include <linux/wait.h>
	33	#include <linux/mutex.h>
f0dbdefb	34	#include <linux/device.h>
c337ddc2	35
241ae9a8 MD	36	#include <lttng-events.h>
	37	#include <lttng-tracer.h>
	38	#include <wrapper/irqdesc.h>
	39	#include <wrapper/spinlock.h>
	40	#include <wrapper/fdtable.h>
241ae9a8 MD	41	#include <wrapper/irq.h>
	42	#include <wrapper/tracepoint.h>
	43	#include <wrapper/genhd.h>
	44	#include <wrapper/file.h>
a117385f	45	#include <wrapper/fdtable.h>
c337ddc2	46
29784493	47	#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
c337ddc2 MD	48	#include <linux/irq.h>
	49	#endif
	50
	51	/* Define the tracepoints, but do not build the probes */
	52	#define CREATE_TRACE_POINTS
241ae9a8	53	#define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
c337ddc2	54	#define TRACE_INCLUDE_FILE lttng-statedump
3bc29f0a	55	#define LTTNG_INSTRUMENTATION
241ae9a8	56	#include <instrumentation/events/lttng-module/lttng-statedump.h>
c337ddc2	57
f0dbdefb	58	DEFINE_TRACE(lttng_statedump_block_device);
20591cf7 MD	59	DEFINE_TRACE(lttng_statedump_end);
	60	DEFINE_TRACE(lttng_statedump_interrupt);
	61	DEFINE_TRACE(lttng_statedump_file_descriptor);
	62	DEFINE_TRACE(lttng_statedump_start);
	63	DEFINE_TRACE(lttng_statedump_process_state);
	64	DEFINE_TRACE(lttng_statedump_network_interface);
	65
361c023a MD	66	struct lttng_fd_ctx {
	67	char *page;
	68	struct lttng_session *session;
	69	struct task_struct *p;
d561ecfb	70	struct files_struct *files;
361c023a MD	71	};
361c023a MD	72
c337ddc2 MD	73	/*
	74	* Protected by the trace lock.
	75	*/
	76	static struct delayed_work cpu_work[NR_CPUS];
	77	static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
	78	static atomic_t kernel_threads_to_run;
	79
	80	enum lttng_thread_type {
	81	LTTNG_USER_THREAD = 0,
	82	LTTNG_KERNEL_THREAD = 1,
	83	};
	84
	85	enum lttng_execution_mode {
	86	LTTNG_USER_MODE = 0,
	87	LTTNG_SYSCALL = 1,
	88	LTTNG_TRAP = 2,
	89	LTTNG_IRQ = 3,
	90	LTTNG_SOFTIRQ = 4,
	91	LTTNG_MODE_UNKNOWN = 5,
	92	};
	93
	94	enum lttng_execution_submode {
	95	LTTNG_NONE = 0,
	96	LTTNG_UNKNOWN = 1,
	97	};
	98
	99	enum lttng_process_status {
	100	LTTNG_UNNAMED = 0,
	101	LTTNG_WAIT_FORK = 1,
	102	LTTNG_WAIT_CPU = 2,
	103	LTTNG_EXIT = 3,
	104	LTTNG_ZOMBIE = 4,
	105	LTTNG_WAIT = 5,
	106	LTTNG_RUN = 6,
	107	LTTNG_DEAD = 7,
	108	};
	109
f0dbdefb HD	110	static
	111	int lttng_enumerate_block_devices(struct lttng_session *session)
	112	{
	113	struct class *ptr_block_class;
	114	struct device_type *ptr_disk_type;
	115	struct class_dev_iter iter;
	116	struct device *dev;
	117
	118	ptr_block_class = wrapper_get_block_class();
	119	if (!ptr_block_class)
	120	return -ENOSYS;
	121	ptr_disk_type = wrapper_get_disk_type();
	122	if (!ptr_disk_type) {
	123	return -ENOSYS;
	124	}
	125	class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
	126	while ((dev = class_dev_iter_next(&iter))) {
	127	struct disk_part_iter piter;
	128	struct gendisk *disk = dev_to_disk(dev);
	129	struct hd_struct *part;
	130
5a91f3df MD	131	/*
	132	* Don't show empty devices or things that have been
	133	* suppressed
	134	*/
	135	if (get_capacity(disk) == 0 \|\|
	136	(disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
	137	continue;
	138
f0dbdefb HD	139	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
f0dbdefb HD	140	while ((part = disk_part_iter_next(&piter))) {
bf5a011d	141	struct block_device bdev;
f0dbdefb	142	char name_buf[BDEVNAME_SIZE];
bf5a011d	143	const char *p;
f0dbdefb	144
bf5a011d MJ	145	/*
	146	* Create a partial 'struct blockdevice' to use
	147	* 'bdevname()' which is a simple wrapper over
	148	* 'disk_name()' but has the honor to be EXPORT_SYMBOL.
	149	*/
	150	bdev.bd_disk = disk;
	151	bdev.bd_part = part;
	152
	153	p = bdevname(&bdev, name_buf);
f0dbdefb HD	154	if (!p) {
	155	disk_part_iter_exit(&piter);
	156	class_dev_iter_exit(&iter);
	157	return -ENOSYS;
	158	}
	159	trace_lttng_statedump_block_device(session,
	160	part_devt(part), name_buf);
	161	}
	162	disk_part_iter_exit(&piter);
	163	}
	164	class_dev_iter_exit(&iter);
	165	return 0;
	166	}
	167
c337ddc2	168	#ifdef CONFIG_INET
f0dbdefb	169
c337ddc2 MD	170	static
	171	void lttng_enumerate_device(struct lttng_session *session,
	172	struct net_device *dev)
	173	{
	174	struct in_device *in_dev;
	175	struct in_ifaddr *ifa;
	176
	177	if (dev->flags & IFF_UP) {
	178	in_dev = in_dev_get(dev);
	179	if (in_dev) {
	180	for (ifa = in_dev->ifa_list; ifa != NULL;
	181	ifa = ifa->ifa_next) {
	182	trace_lttng_statedump_network_interface(
	183	session, dev, ifa);
	184	}
	185	in_dev_put(in_dev);
	186	}
	187	} else {
	188	trace_lttng_statedump_network_interface(
	189	session, dev, NULL);
	190	}
	191	}
	192
	193	static
	194	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	195	{
	196	struct net_device *dev;
	197
	198	read_lock(&dev_base_lock);
	199	for_each_netdev(&init_net, dev)
	200	lttng_enumerate_device(session, dev);
	201	read_unlock(&dev_base_lock);
	202
	203	return 0;
	204	}
	205	#else /* CONFIG_INET */
	206	static inline
	207	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	208	{
	209	return 0;
	210	}
	211	#endif /* CONFIG_INET */
	212
361c023a MD	213	static
	214	int lttng_dump_one_fd(const void p, struct file file, unsigned int fd)
	215	{
	216	const struct lttng_fd_ctx *ctx = p;
	217	const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
29021503	218	unsigned int flags = file->f_flags;
d561ecfb	219	struct fdtable *fdt;
361c023a	220
29021503 MD	221	/*
	222	* We don't expose kernel internal flags, only userspace-visible
	223	* flags.
	224	*/
	225	flags &= ~FMODE_NONOTIFY;
d561ecfb MD	226	fdt = files_fdtable(ctx->files);
	227	/*
	228	* We need to check here again whether fd is within the fdt
	229	* max_fds range, because we might be seeing a different
	230	* files_fdtable() than iterate_fd(), assuming only RCU is
	231	* protecting the read. In reality, iterate_fd() holds
	232	* file_lock, which should ensure the fdt does not change while
	233	* the lock is taken, but we are not aware whether this is
	234	* guaranteed or not, so play safe.
	235	*/
aa29f2d3	236	if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
29021503	237	flags \|= O_CLOEXEC;
361c023a MD	238	if (IS_ERR(s)) {
	239	struct dentry *dentry = file->f_path.dentry;
	240
	241	/* Make sure we give at least some info */
	242	spin_lock(&dentry->d_lock);
	243	trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
29021503	244	dentry->d_name.name, flags, file->f_mode);
361c023a MD	245	spin_unlock(&dentry->d_lock);
	246	goto end;
	247	}
29021503 MD	248	trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
29021503 MD	249	flags, file->f_mode);
361c023a MD	250	end:
	251	return 0;
	252	}
c337ddc2 MD	253
	254	static
	255	void lttng_enumerate_task_fd(struct lttng_session *session,
	256	struct task_struct p, char tmp)
	257	{
361c023a	258	struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
d561ecfb	259	struct files_struct *files;
c337ddc2 MD	260
c337ddc2 MD	261	task_lock(p);
d561ecfb MD	262	files = p->files;
	263	if (!files)
	264	goto end;
	265	ctx.files = files;
	266	lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
	267	end:
c337ddc2 MD	268	task_unlock(p);
	269	}
	270
	271	static
	272	int lttng_enumerate_file_descriptors(struct lttng_session *session)
	273	{
	274	struct task_struct *p;
cfcee1c7 MD	275	char *tmp;
	276
	277	tmp = (char *) __get_free_page(GFP_KERNEL);
	278	if (!tmp)
	279	return -ENOMEM;
c337ddc2 MD	280
	281	/* Enumerate active file descriptors */
	282	rcu_read_lock();
	283	for_each_process(p)
	284	lttng_enumerate_task_fd(session, p, tmp);
	285	rcu_read_unlock();
	286	free_page((unsigned long) tmp);
	287	return 0;
	288	}
	289
0658bdda MD	290	#if 0
	291	/*
	292	* FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
	293	* (scheduling in atomic). Normally, the tasklist lock protects this kind of
	294	* iteration, but it is not exported to modules.
	295	*/
c337ddc2 MD	296	static
	297	void lttng_enumerate_task_vm_maps(struct lttng_session *session,
	298	struct task_struct *p)
	299	{
	300	struct mm_struct *mm;
	301	struct vm_area_struct *map;
	302	unsigned long ino;
	303
	304	/* get_task_mm does a task_lock... */
	305	mm = get_task_mm(p);
	306	if (!mm)
	307	return;
	308
	309	map = mm->mmap;
	310	if (map) {
	311	down_read(&mm->mmap_sem);
	312	while (map) {
	313	if (map->vm_file)
b06ed645	314	ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
c337ddc2 MD	315	else
	316	ino = 0;
	317	trace_lttng_statedump_vm_map(session, p, map, ino);
	318	map = map->vm_next;
	319	}
	320	up_read(&mm->mmap_sem);
	321	}
	322	mmput(mm);
	323	}
	324
	325	static
	326	int lttng_enumerate_vm_maps(struct lttng_session *session)
	327	{
	328	struct task_struct *p;
	329
	330	rcu_read_lock();
	331	for_each_process(p)
	332	lttng_enumerate_task_vm_maps(session, p);
	333	rcu_read_unlock();
	334	return 0;
	335	}
0658bdda	336	#endif
c337ddc2	337
29784493	338	#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
47faec4b JN	339
	340	#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
	341	#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
	342	#endif
	343
c337ddc2	344	static
cfcee1c7	345	int lttng_list_interrupts(struct lttng_session *session)
c337ddc2 MD	346	{
	347	unsigned int irq;
	348	unsigned long flags = 0;
	349	struct irq_desc *desc;
	350
	351	#define irq_to_desc wrapper_irq_to_desc
	352	/* needs irq_desc */
	353	for_each_irq_desc(irq, desc) {
	354	struct irqaction *action;
	355	const char *irq_chip_name =
	356	irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
	357
	358	local_irq_save(flags);
3a523f5b	359	wrapper_desc_spin_lock(&desc->lock);
c337ddc2 MD	360	for (action = desc->action; action; action = action->next) {
	361	trace_lttng_statedump_interrupt(session,
	362	irq, irq_chip_name, action);
	363	}
3a523f5b	364	wrapper_desc_spin_unlock(&desc->lock);
c337ddc2 MD	365	local_irq_restore(flags);
c337ddc2 MD	366	}
cfcee1c7	367	return 0;
c337ddc2 MD	368	#undef irq_to_desc
	369	}
	370	#else
	371	static inline
cfcee1c7	372	int lttng_list_interrupts(struct lttng_session *session)
c337ddc2	373	{
cfcee1c7	374	return 0;
c337ddc2 MD	375	}
	376	#endif
	377
4ba1f53c MD	378	/*
	379	* Called with task lock held.
	380	*/
73e8ba37 JD	381	static
	382	void lttng_statedump_process_ns(struct lttng_session *session,
	383	struct task_struct *p,
	384	enum lttng_thread_type type,
	385	enum lttng_execution_mode mode,
	386	enum lttng_execution_submode submode,
	387	enum lttng_process_status status)
	388	{
73e8ba37 JD	389	struct pid_namespace *pid_ns;
73e8ba37 JD	390
887bcdac MJ	391	pid_ns = task_active_pid_ns(p);
887bcdac MJ	392	do {
73e8ba37	393	trace_lttng_statedump_process_state(session,
887bcdac	394	p, type, mode, submode, status, pid_ns);
d16203a5	395	pid_ns = pid_ns ? pid_ns->parent : NULL;
887bcdac	396	} while (pid_ns);
73e8ba37 JD	397	}
73e8ba37 JD	398
c337ddc2 MD	399	static
	400	int lttng_enumerate_process_states(struct lttng_session *session)
	401	{
	402	struct task_struct g, p;
	403
	404	rcu_read_lock();
	405	for_each_process(g) {
	406	p = g;
	407	do {
	408	enum lttng_execution_mode mode =
	409	LTTNG_MODE_UNKNOWN;
	410	enum lttng_execution_submode submode =
	411	LTTNG_UNKNOWN;
	412	enum lttng_process_status status;
	413	enum lttng_thread_type type;
	414
	415	task_lock(p);
	416	if (p->exit_state == EXIT_ZOMBIE)
	417	status = LTTNG_ZOMBIE;
	418	else if (p->exit_state == EXIT_DEAD)
	419	status = LTTNG_DEAD;
	420	else if (p->state == TASK_RUNNING) {
	421	/* Is this a forked child that has not run yet? */
	422	if (list_empty(&p->rt.run_list))
	423	status = LTTNG_WAIT_FORK;
	424	else
	425	/*
	426	* All tasks are considered as wait_cpu;
	427	* the viewer will sort out if the task
	428	* was really running at this time.
	429	*/
	430	status = LTTNG_WAIT_CPU;
	431	} else if (p->state &
	432	(TASK_INTERRUPTIBLE \| TASK_UNINTERRUPTIBLE)) {
	433	/* Task is waiting for something to complete */
	434	status = LTTNG_WAIT;
	435	} else
	436	status = LTTNG_UNNAMED;
	437	submode = LTTNG_NONE;
	438
	439	/*
	440	* Verification of t->mm is to filter out kernel
	441	* threads; Viewer will further filter out if a
	442	* user-space thread was in syscall mode or not.
	443	*/
	444	if (p->mm)
	445	type = LTTNG_USER_THREAD;
	446	else
	447	type = LTTNG_KERNEL_THREAD;
73e8ba37	448	lttng_statedump_process_ns(session,
c337ddc2 MD	449	p, type, mode, submode, status);
	450	task_unlock(p);
	451	} while_each_thread(g, p);
	452	}
	453	rcu_read_unlock();
	454
	455	return 0;
	456	}
	457
	458	static
	459	void lttng_statedump_work_func(struct work_struct *work)
	460	{
	461	if (atomic_dec_and_test(&kernel_threads_to_run))
	462	/* If we are the last thread, wake up do_lttng_statedump */
	463	wake_up(&statedump_wq);
	464	}
	465
	466	static
	467	int do_lttng_statedump(struct lttng_session *session)
	468	{
cfcee1c7	469	int cpu, ret;
c337ddc2	470
c337ddc2	471	trace_lttng_statedump_start(session);
cfcee1c7 MD	472	ret = lttng_enumerate_process_states(session);
	473	if (ret)
	474	return ret;
	475	ret = lttng_enumerate_file_descriptors(session);
	476	if (ret)
	477	return ret;
	478	/*
	479	* FIXME
	480	* ret = lttng_enumerate_vm_maps(session);
	481	* if (ret)
	482	* return ret;
	483	*/
	484	ret = lttng_list_interrupts(session);
	485	if (ret)
	486	return ret;
	487	ret = lttng_enumerate_network_ip_interface(session);
	488	if (ret)
	489	return ret;
	490	ret = lttng_enumerate_block_devices(session);
	491	switch (ret) {
84c7055e MD	492	case 0:
84c7055e MD	493	break;
cfcee1c7 MD	494	case -ENOSYS:
	495	printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
	496	break;
	497	default:
	498	return ret;
	499	}
c337ddc2 MD	500
	501	/* TODO lttng_dump_idt_table(session); */
	502	/* TODO lttng_dump_softirq_vec(session); */
	503	/* TODO lttng_list_modules(session); */
	504	/* TODO lttng_dump_swap_files(session); */
	505
	506	/*
	507	* Fire off a work queue on each CPU. Their sole purpose in life
	508	* is to guarantee that each CPU has been in a state where is was in
	509	* syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
	510	*/
	511	get_online_cpus();
	512	atomic_set(&kernel_threads_to_run, num_online_cpus());
	513	for_each_online_cpu(cpu) {
	514	INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
	515	schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
	516	}
	517	/* Wait for all threads to run */
7a7128e0	518	__wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
c337ddc2 MD	519	put_online_cpus();
c337ddc2 MD	520	/* Our work is done */
c337ddc2 MD	521	trace_lttng_statedump_end(session);
	522	return 0;
	523	}
	524
	525	/*
	526	* Called with session mutex held.
	527	*/
	528	int lttng_statedump_start(struct lttng_session *session)
	529	{
c337ddc2 MD	530	return do_lttng_statedump(session);
	531	}
	532	EXPORT_SYMBOL_GPL(lttng_statedump_start);
	533
dd8d5afb MD	534	static
	535	int __init lttng_statedump_init(void)
	536	{
d16aa9c9 MD	537	/*
	538	* Allow module to load even if the fixup cannot be done. This
	539	* will allow seemless transition when the underlying issue fix
	540	* is merged into the Linux kernel, and when tracepoint.c
	541	* "tracepoint_module_notify" is turned into a static function.
	542	*/
	543	(void) wrapper_lttng_fixup_sig(THIS_MODULE);
	544	return 0;
dd8d5afb MD	545	}
	546
	547	module_init(lttng_statedump_init);
	548
461277e7 MD	549	static
	550	void __exit lttng_statedump_exit(void)
	551	{
	552	}
	553
	554	module_exit(lttng_statedump_exit);
	555
c337ddc2 MD	556	MODULE_LICENSE("GPL and additional rights");
c337ddc2 MD	557	MODULE_AUTHOR("Jean-Hugues Deschenes");
1c124020	558	MODULE_DESCRIPTION("LTTng statedump provider");
13ab8b0a MD	559	MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
	560	__stringify(LTTNG_MODULES_MINOR_VERSION) "."
	561	__stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
	562	LTTNG_MODULES_EXTRAVERSION);