[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>

LTTNG_DEFINE_TRACE(lttng_statedump_block_device,
	TP_PROTO(struct lttng_session *session,
		dev_t dev, const char *diskname),
	TP_ARGS(session, dev, diskname));

LTTNG_DEFINE_TRACE(lttng_statedump_end,
	TP_PROTO(struct lttng_session *session),
	TP_ARGS(session));

LTTNG_DEFINE_TRACE(lttng_statedump_interrupt,
	TP_PROTO(struct lttng_session *session,
		unsigned int irq, const char *chip_name,
		struct irqaction *action),
	TP_ARGS(session, irq, chip_name, action));

LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor,
	TP_PROTO(struct lttng_session *session,
		struct task_struct *p, int fd, const char *filename,
		unsigned int flags, fmode_t fmode),
	TP_ARGS(session, p, fd, filename, flags, fmode));

LTTNG_DEFINE_TRACE(lttng_statedump_start,
	TP_PROTO(struct lttng_session *session),
	TP_ARGS(session));

LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
	TP_PROTO(struct lttng_session *session,
		struct task_struct *p,
		int type, int mode, int submode, int status,
		struct pid_namespace *pid_ns),
	TP_ARGS(session, p, type, mode, submode, status, pid_ns));

LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
	TP_PROTO(struct lttng_session *session,
		struct net_device *dev, struct in_ifaddr *ifa),
	TP_ARGS(session, dev, ifa));

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
c6b31b34 MJ	58	LTTNG_DEFINE_TRACE(lttng_statedump_block_device,
	59	TP_PROTO(struct lttng_session *session,
	60	dev_t dev, const char *diskname),
	61	TP_ARGS(session, dev, diskname));
	62
	63	LTTNG_DEFINE_TRACE(lttng_statedump_end,
	64	TP_PROTO(struct lttng_session *session),
	65	TP_ARGS(session));
	66
	67	LTTNG_DEFINE_TRACE(lttng_statedump_interrupt,
	68	TP_PROTO(struct lttng_session *session,
	69	unsigned int irq, const char *chip_name,
	70	struct irqaction *action),
	71	TP_ARGS(session, irq, chip_name, action));
	72
	73	LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor,
	74	TP_PROTO(struct lttng_session *session,
f8922333	75	struct task_struct p, int fd, const char filename,
c6b31b34	76	unsigned int flags, fmode_t fmode),
f8922333	77	TP_ARGS(session, p, fd, filename, flags, fmode));
c6b31b34 MJ	78
	79	LTTNG_DEFINE_TRACE(lttng_statedump_start,
	80	TP_PROTO(struct lttng_session *session),
	81	TP_ARGS(session));
	82
	83	LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
	84	TP_PROTO(struct lttng_session *session,
	85	struct task_struct *p,
	86	int type, int mode, int submode, int status,
f8922333 MJ	87	struct pid_namespace *pid_ns),
f8922333 MJ	88	TP_ARGS(session, p, type, mode, submode, status, pid_ns));
c6b31b34 MJ	89
	90	LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
	91	TP_PROTO(struct lttng_session *session,
	92	struct net_device dev, struct in_ifaddr ifa),
	93	TP_ARGS(session, dev, ifa));
20591cf7	94
361c023a MD	95	struct lttng_fd_ctx {
	96	char *page;
	97	struct lttng_session *session;
	98	struct task_struct *p;
d561ecfb	99	struct files_struct *files;
361c023a MD	100	};
361c023a MD	101
c337ddc2 MD	102	/*
	103	* Protected by the trace lock.
	104	*/
	105	static struct delayed_work cpu_work[NR_CPUS];
	106	static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
	107	static atomic_t kernel_threads_to_run;
	108
	109	enum lttng_thread_type {
	110	LTTNG_USER_THREAD = 0,
	111	LTTNG_KERNEL_THREAD = 1,
	112	};
	113
	114	enum lttng_execution_mode {
	115	LTTNG_USER_MODE = 0,
	116	LTTNG_SYSCALL = 1,
	117	LTTNG_TRAP = 2,
	118	LTTNG_IRQ = 3,
	119	LTTNG_SOFTIRQ = 4,
	120	LTTNG_MODE_UNKNOWN = 5,
	121	};
	122
	123	enum lttng_execution_submode {
	124	LTTNG_NONE = 0,
	125	LTTNG_UNKNOWN = 1,
	126	};
	127
	128	enum lttng_process_status {
	129	LTTNG_UNNAMED = 0,
	130	LTTNG_WAIT_FORK = 1,
	131	LTTNG_WAIT_CPU = 2,
	132	LTTNG_EXIT = 3,
	133	LTTNG_ZOMBIE = 4,
	134	LTTNG_WAIT = 5,
	135	LTTNG_RUN = 6,
	136	LTTNG_DEAD = 7,
	137	};
	138
f0dbdefb HD	139	static
	140	int lttng_enumerate_block_devices(struct lttng_session *session)
	141	{
	142	struct class *ptr_block_class;
	143	struct device_type *ptr_disk_type;
	144	struct class_dev_iter iter;
	145	struct device *dev;
	146
	147	ptr_block_class = wrapper_get_block_class();
	148	if (!ptr_block_class)
	149	return -ENOSYS;
	150	ptr_disk_type = wrapper_get_disk_type();
	151	if (!ptr_disk_type) {
	152	return -ENOSYS;
	153	}
	154	class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
	155	while ((dev = class_dev_iter_next(&iter))) {
	156	struct disk_part_iter piter;
	157	struct gendisk *disk = dev_to_disk(dev);
	158	struct hd_struct *part;
	159
5a91f3df MD	160	/*
	161	* Don't show empty devices or things that have been
	162	* suppressed
	163	*/
	164	if (get_capacity(disk) == 0 \|\|
	165	(disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
	166	continue;
	167
f0dbdefb HD	168	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
f0dbdefb HD	169	while ((part = disk_part_iter_next(&piter))) {
bf5a011d	170	struct block_device bdev;
f0dbdefb	171	char name_buf[BDEVNAME_SIZE];
bf5a011d	172	const char *p;
f0dbdefb	173
bf5a011d MJ	174	/*
	175	* Create a partial 'struct blockdevice' to use
	176	* 'bdevname()' which is a simple wrapper over
	177	* 'disk_name()' but has the honor to be EXPORT_SYMBOL.
	178	*/
	179	bdev.bd_disk = disk;
	180	bdev.bd_part = part;
	181
	182	p = bdevname(&bdev, name_buf);
f0dbdefb HD	183	if (!p) {
	184	disk_part_iter_exit(&piter);
	185	class_dev_iter_exit(&iter);
	186	return -ENOSYS;
	187	}
	188	trace_lttng_statedump_block_device(session,
	189	part_devt(part), name_buf);
	190	}
	191	disk_part_iter_exit(&piter);
	192	}
	193	class_dev_iter_exit(&iter);
	194	return 0;
	195	}
	196
c337ddc2	197	#ifdef CONFIG_INET
f0dbdefb	198
c337ddc2 MD	199	static
	200	void lttng_enumerate_device(struct lttng_session *session,
	201	struct net_device *dev)
	202	{
	203	struct in_device *in_dev;
	204	struct in_ifaddr *ifa;
	205
	206	if (dev->flags & IFF_UP) {
	207	in_dev = in_dev_get(dev);
	208	if (in_dev) {
	209	for (ifa = in_dev->ifa_list; ifa != NULL;
	210	ifa = ifa->ifa_next) {
	211	trace_lttng_statedump_network_interface(
	212	session, dev, ifa);
	213	}
	214	in_dev_put(in_dev);
	215	}
	216	} else {
	217	trace_lttng_statedump_network_interface(
	218	session, dev, NULL);
	219	}
	220	}
	221
	222	static
	223	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	224	{
	225	struct net_device *dev;
	226
	227	read_lock(&dev_base_lock);
	228	for_each_netdev(&init_net, dev)
	229	lttng_enumerate_device(session, dev);
	230	read_unlock(&dev_base_lock);
	231
	232	return 0;
	233	}
	234	#else /* CONFIG_INET */
	235	static inline
	236	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	237	{
	238	return 0;
	239	}
	240	#endif /* CONFIG_INET */
	241
361c023a MD	242	static
	243	int lttng_dump_one_fd(const void p, struct file file, unsigned int fd)
	244	{
	245	const struct lttng_fd_ctx *ctx = p;
	246	const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
29021503	247	unsigned int flags = file->f_flags;
d561ecfb	248	struct fdtable *fdt;
361c023a	249
29021503 MD	250	/*
	251	* We don't expose kernel internal flags, only userspace-visible
	252	* flags.
	253	*/
	254	flags &= ~FMODE_NONOTIFY;
d561ecfb MD	255	fdt = files_fdtable(ctx->files);
	256	/*
	257	* We need to check here again whether fd is within the fdt
	258	* max_fds range, because we might be seeing a different
	259	* files_fdtable() than iterate_fd(), assuming only RCU is
	260	* protecting the read. In reality, iterate_fd() holds
	261	* file_lock, which should ensure the fdt does not change while
	262	* the lock is taken, but we are not aware whether this is
	263	* guaranteed or not, so play safe.
	264	*/
aa29f2d3	265	if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
29021503	266	flags \|= O_CLOEXEC;
361c023a MD	267	if (IS_ERR(s)) {
	268	struct dentry *dentry = file->f_path.dentry;
	269
	270	/* Make sure we give at least some info */
	271	spin_lock(&dentry->d_lock);
	272	trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
29021503	273	dentry->d_name.name, flags, file->f_mode);
361c023a MD	274	spin_unlock(&dentry->d_lock);
	275	goto end;
	276	}
29021503 MD	277	trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
29021503 MD	278	flags, file->f_mode);
361c023a MD	279	end:
	280	return 0;
	281	}
c337ddc2 MD	282
	283	static
	284	void lttng_enumerate_task_fd(struct lttng_session *session,
	285	struct task_struct p, char tmp)
	286	{
361c023a	287	struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
d561ecfb	288	struct files_struct *files;
c337ddc2 MD	289
c337ddc2 MD	290	task_lock(p);
d561ecfb MD	291	files = p->files;
	292	if (!files)
	293	goto end;
	294	ctx.files = files;
	295	lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
	296	end:
c337ddc2 MD	297	task_unlock(p);
	298	}
	299
	300	static
	301	int lttng_enumerate_file_descriptors(struct lttng_session *session)
	302	{
	303	struct task_struct *p;
cfcee1c7 MD	304	char *tmp;
	305
	306	tmp = (char *) __get_free_page(GFP_KERNEL);
	307	if (!tmp)
	308	return -ENOMEM;
c337ddc2 MD	309
	310	/* Enumerate active file descriptors */
	311	rcu_read_lock();
	312	for_each_process(p)
	313	lttng_enumerate_task_fd(session, p, tmp);
	314	rcu_read_unlock();
	315	free_page((unsigned long) tmp);
	316	return 0;
	317	}
	318
0658bdda MD	319	#if 0
	320	/*
	321	* FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
	322	* (scheduling in atomic). Normally, the tasklist lock protects this kind of
	323	* iteration, but it is not exported to modules.
	324	*/
c337ddc2 MD	325	static
	326	void lttng_enumerate_task_vm_maps(struct lttng_session *session,
	327	struct task_struct *p)
	328	{
	329	struct mm_struct *mm;
	330	struct vm_area_struct *map;
	331	unsigned long ino;
	332
	333	/* get_task_mm does a task_lock... */
	334	mm = get_task_mm(p);
	335	if (!mm)
	336	return;
	337
	338	map = mm->mmap;
	339	if (map) {
	340	down_read(&mm->mmap_sem);
	341	while (map) {
	342	if (map->vm_file)
b06ed645	343	ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
c337ddc2 MD	344	else
	345	ino = 0;
	346	trace_lttng_statedump_vm_map(session, p, map, ino);
	347	map = map->vm_next;
	348	}
	349	up_read(&mm->mmap_sem);
	350	}
	351	mmput(mm);
	352	}
	353
	354	static
	355	int lttng_enumerate_vm_maps(struct lttng_session *session)
	356	{
	357	struct task_struct *p;
	358
	359	rcu_read_lock();
	360	for_each_process(p)
	361	lttng_enumerate_task_vm_maps(session, p);
	362	rcu_read_unlock();
	363	return 0;
	364	}
0658bdda	365	#endif
c337ddc2	366
29784493	367	#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
47faec4b JN	368
	369	#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
	370	#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
	371	#endif
	372
c337ddc2	373	static
cfcee1c7	374	int lttng_list_interrupts(struct lttng_session *session)
c337ddc2 MD	375	{
	376	unsigned int irq;
	377	unsigned long flags = 0;
	378	struct irq_desc *desc;
	379
	380	#define irq_to_desc wrapper_irq_to_desc
	381	/* needs irq_desc */
	382	for_each_irq_desc(irq, desc) {
	383	struct irqaction *action;
	384	const char *irq_chip_name =
	385	irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
	386
	387	local_irq_save(flags);
3a523f5b	388	wrapper_desc_spin_lock(&desc->lock);
c337ddc2 MD	389	for (action = desc->action; action; action = action->next) {
	390	trace_lttng_statedump_interrupt(session,
	391	irq, irq_chip_name, action);
	392	}
3a523f5b	393	wrapper_desc_spin_unlock(&desc->lock);
c337ddc2 MD	394	local_irq_restore(flags);
c337ddc2 MD	395	}
cfcee1c7	396	return 0;
c337ddc2 MD	397	#undef irq_to_desc
	398	}
	399	#else
	400	static inline
cfcee1c7	401	int lttng_list_interrupts(struct lttng_session *session)
c337ddc2	402	{
cfcee1c7	403	return 0;
c337ddc2 MD	404	}
	405	#endif
	406
4ba1f53c MD	407	/*
	408	* Called with task lock held.
	409	*/
73e8ba37 JD	410	static
	411	void lttng_statedump_process_ns(struct lttng_session *session,
	412	struct task_struct *p,
	413	enum lttng_thread_type type,
	414	enum lttng_execution_mode mode,
	415	enum lttng_execution_submode submode,
	416	enum lttng_process_status status)
	417	{
73e8ba37 JD	418	struct pid_namespace *pid_ns;
73e8ba37 JD	419
887bcdac MJ	420	pid_ns = task_active_pid_ns(p);
887bcdac MJ	421	do {
73e8ba37	422	trace_lttng_statedump_process_state(session,
887bcdac	423	p, type, mode, submode, status, pid_ns);
d16203a5	424	pid_ns = pid_ns ? pid_ns->parent : NULL;
887bcdac	425	} while (pid_ns);
73e8ba37 JD	426	}
73e8ba37 JD	427
c337ddc2 MD	428	static
	429	int lttng_enumerate_process_states(struct lttng_session *session)
	430	{
	431	struct task_struct g, p;
	432
	433	rcu_read_lock();
	434	for_each_process(g) {
	435	p = g;
	436	do {
	437	enum lttng_execution_mode mode =
	438	LTTNG_MODE_UNKNOWN;
	439	enum lttng_execution_submode submode =
	440	LTTNG_UNKNOWN;
	441	enum lttng_process_status status;
	442	enum lttng_thread_type type;
	443
	444	task_lock(p);
	445	if (p->exit_state == EXIT_ZOMBIE)
	446	status = LTTNG_ZOMBIE;
	447	else if (p->exit_state == EXIT_DEAD)
	448	status = LTTNG_DEAD;
	449	else if (p->state == TASK_RUNNING) {
	450	/* Is this a forked child that has not run yet? */
	451	if (list_empty(&p->rt.run_list))
	452	status = LTTNG_WAIT_FORK;
	453	else
	454	/*
	455	* All tasks are considered as wait_cpu;
	456	* the viewer will sort out if the task
	457	* was really running at this time.
	458	*/
	459	status = LTTNG_WAIT_CPU;
	460	} else if (p->state &
	461	(TASK_INTERRUPTIBLE \| TASK_UNINTERRUPTIBLE)) {
	462	/* Task is waiting for something to complete */
	463	status = LTTNG_WAIT;
	464	} else
	465	status = LTTNG_UNNAMED;
	466	submode = LTTNG_NONE;
	467
	468	/*
	469	* Verification of t->mm is to filter out kernel
	470	* threads; Viewer will further filter out if a
	471	* user-space thread was in syscall mode or not.
	472	*/
	473	if (p->mm)
	474	type = LTTNG_USER_THREAD;
	475	else
	476	type = LTTNG_KERNEL_THREAD;
73e8ba37	477	lttng_statedump_process_ns(session,
c337ddc2 MD	478	p, type, mode, submode, status);
	479	task_unlock(p);
	480	} while_each_thread(g, p);
	481	}
	482	rcu_read_unlock();
	483
	484	return 0;
	485	}
	486
	487	static
	488	void lttng_statedump_work_func(struct work_struct *work)
	489	{
	490	if (atomic_dec_and_test(&kernel_threads_to_run))
	491	/* If we are the last thread, wake up do_lttng_statedump */
	492	wake_up(&statedump_wq);
	493	}
	494
	495	static
	496	int do_lttng_statedump(struct lttng_session *session)
	497	{
cfcee1c7	498	int cpu, ret;
c337ddc2	499
c337ddc2	500	trace_lttng_statedump_start(session);
cfcee1c7 MD	501	ret = lttng_enumerate_process_states(session);
	502	if (ret)
	503	return ret;
	504	ret = lttng_enumerate_file_descriptors(session);
	505	if (ret)
	506	return ret;
	507	/*
	508	* FIXME
	509	* ret = lttng_enumerate_vm_maps(session);
	510	* if (ret)
	511	* return ret;
	512	*/
	513	ret = lttng_list_interrupts(session);
	514	if (ret)
	515	return ret;
	516	ret = lttng_enumerate_network_ip_interface(session);
	517	if (ret)
	518	return ret;
	519	ret = lttng_enumerate_block_devices(session);
	520	switch (ret) {
84c7055e MD	521	case 0:
84c7055e MD	522	break;
cfcee1c7 MD	523	case -ENOSYS:
	524	printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
	525	break;
	526	default:
	527	return ret;
	528	}
c337ddc2 MD	529
	530	/* TODO lttng_dump_idt_table(session); */
	531	/* TODO lttng_dump_softirq_vec(session); */
	532	/* TODO lttng_list_modules(session); */
	533	/* TODO lttng_dump_swap_files(session); */
	534
	535	/*
	536	* Fire off a work queue on each CPU. Their sole purpose in life
	537	* is to guarantee that each CPU has been in a state where is was in
	538	* syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
	539	*/
	540	get_online_cpus();
	541	atomic_set(&kernel_threads_to_run, num_online_cpus());
	542	for_each_online_cpu(cpu) {
	543	INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
	544	schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
	545	}
	546	/* Wait for all threads to run */
7a7128e0	547	__wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
c337ddc2 MD	548	put_online_cpus();
c337ddc2 MD	549	/* Our work is done */
c337ddc2 MD	550	trace_lttng_statedump_end(session);
	551	return 0;
	552	}
	553
	554	/*
	555	* Called with session mutex held.
	556	*/
	557	int lttng_statedump_start(struct lttng_session *session)
	558	{
c337ddc2 MD	559	return do_lttng_statedump(session);
	560	}
	561	EXPORT_SYMBOL_GPL(lttng_statedump_start);
	562
dd8d5afb MD	563	static
	564	int __init lttng_statedump_init(void)
	565	{
d16aa9c9 MD	566	/*
	567	* Allow module to load even if the fixup cannot be done. This
	568	* will allow seemless transition when the underlying issue fix
	569	* is merged into the Linux kernel, and when tracepoint.c
	570	* "tracepoint_module_notify" is turned into a static function.
	571	*/
	572	(void) wrapper_lttng_fixup_sig(THIS_MODULE);
	573	return 0;
dd8d5afb MD	574	}
	575
	576	module_init(lttng_statedump_init);
	577
461277e7 MD	578	static
	579	void __exit lttng_statedump_exit(void)
	580	{
	581	}
	582
	583	module_exit(lttng_statedump_exit);
	584
c337ddc2 MD	585	MODULE_LICENSE("GPL and additional rights");
c337ddc2 MD	586	MODULE_AUTHOR("Jean-Hugues Deschenes");
1c124020	587	MODULE_DESCRIPTION("LTTng statedump provider");
13ab8b0a MD	588	MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
	589	__stringify(LTTNG_MODULES_MINOR_VERSION) "."
	590	__stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
	591	LTTNG_MODULES_EXTRAVERSION);