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

/* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
 *
 * 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 <linux/fdtable.h>
#include <linux/irq.h>
#include <linux/genhd.h>

#include <lttng-events.h>
#include <lttng-tracer.h>

/* 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_process_pid_ns);
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
DEFINE_TRACE(lttng_statedump_process_cgroup_ns);
#endif
DEFINE_TRACE(lttng_statedump_process_ipc_ns);
#ifndef LTTNG_MNT_NS_MISSING_HEADER
DEFINE_TRACE(lttng_statedump_process_mnt_ns);
#endif
DEFINE_TRACE(lttng_statedump_process_net_ns);
DEFINE_TRACE(lttng_statedump_process_user_ns);
DEFINE_TRACE(lttng_statedump_process_uts_ns);
DEFINE_TRACE(lttng_statedump_network_interface);
#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
DEFINE_TRACE(lttng_statedump_cpu_topology);
#endif

struct lttng_fd_ctx {
	char *page;
	struct lttng_session *session;
	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_dev_iter iter;
	struct device *dev;

	class_dev_iter_init(&iter, gendisk_block_class(), NULL,
			    gendisk_device_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))) {
			char name_buf[BDEVNAME_SIZE];
			char *p;

			p = gendisk_name(disk, part->partno, 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 && 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->files, fd, dentry->d_name.name, flags,
			file->f_mode);
		spin_unlock(&dentry->d_lock);
		goto end;
	}
	trace_lttng_statedump_file_descriptor(ctx->session,
		ctx->files, fd, s, flags, file->f_mode);
end:
	return 0;
}

/* Called with task lock held. */
static
void lttng_enumerate_files(struct lttng_session *session,
		struct files_struct *files,
		char *tmp)
{
	struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };

	iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
}

#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
static
int lttng_enumerate_cpu_topology(struct lttng_session *session)
{
	int cpu;
	const cpumask_t *cpumask = cpu_possible_mask;

	for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
			cpu = cpumask_next(cpu, cpumask)) {
		trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
	}

	return 0;
}
#else
static
int lttng_enumerate_cpu_topology(struct lttng_session *session)
{
	return 0;
}
#endif

#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->f_path.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

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

/*
 * Statedump the task's namespaces using the proc filesystem inode number as
 * the unique identifier. The user and pid ns are nested and will be dumped
 * recursively.
 *
 * 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 nsproxy *proxy;
	struct pid_namespace *pid_ns;
	struct user_namespace *user_ns;

	/*
	 * The pid and user namespaces are special, they are nested and
	 * accessed with specific functions instead of the nsproxy struct
	 * like the other namespaces.
	 */
	pid_ns = task_active_pid_ns(p);
	do {
		trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
		pid_ns = pid_ns ? pid_ns->parent : NULL;
	} while (pid_ns);


	user_ns = task_cred_xxx(p, user_ns);
	do {
		trace_lttng_statedump_process_user_ns(session, p, user_ns);
		/*
		 * trace_lttng_statedump_process_user_ns() internally
		 * checks whether user_ns is NULL. While this does not
		 * appear to be a possible return value for
		 * task_cred_xxx(), err on the safe side and check
		 * for NULL here as well to be consistent with the
		 * paranoid behavior of
		 * trace_lttng_statedump_process_user_ns().
		 */
		user_ns = user_ns ? user_ns->parent : NULL;
	} while (user_ns);

	/*
	 * Back and forth on locking strategy within Linux upstream for nsproxy.
	 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
	 * "namespaces: Use task_lock and not rcu to protect nsproxy"
	 * for details.
	 */
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
		LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
		LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
		LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
	proxy = p->nsproxy;
#else
	rcu_read_lock();
	proxy = task_nsproxy(p);
#endif
	if (proxy) {
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
		trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
#endif
		trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
#ifndef LTTNG_MNT_NS_MISSING_HEADER
		trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
#endif
		trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
		trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
	}
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
		LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
		LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
		LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
	/* (nothing) */
#else
	rcu_read_unlock();
#endif
}

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

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

	rcu_read_lock();
	for_each_process(g) {
		struct files_struct *prev_files = NULL;

		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;
			struct files_struct *files;

			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;
			files = p->files;

			trace_lttng_statedump_process_state(session,
				p, type, mode, submode, status, files);
			lttng_statedump_process_ns(session,
				p, type, mode, submode, status);
			/*
			 * As an optimisation for the common case, do not
			 * repeat information for the same files_struct in
			 * two consecutive threads. This is the common case
			 * for threads sharing the same fd table. RCU guarantees
			 * that the same files_struct pointer is not re-used
			 * throughout processes/threads iteration.
			 */
			if (files && files != prev_files) {
				lttng_enumerate_files(session, files, tmp);
				prev_files = files;
			}
			task_unlock(p);
		} while_each_thread(g, p);
	}
	rcu_read_unlock();

	free_page((unsigned long) tmp);

	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;
	/*
	 * 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;
	}
	ret = lttng_enumerate_cpu_topology(session);
	if (ret)
		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)
{
	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
b7cdc182	1	/* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
9f36eaed	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>
997eff5e	35	#include <linux/fdtable.h>
88eb3d8e	36	#include <linux/irq.h>
6afd444b	37	#include <linux/genhd.h>
c337ddc2	38
241ae9a8 MD	39	#include <lttng-events.h>
241ae9a8 MD	40	#include <lttng-tracer.h>
c337ddc2	41
c337ddc2 MD	42	/* Define the tracepoints, but do not build the probes */
c337ddc2 MD	43	#define CREATE_TRACE_POINTS
241ae9a8	44	#define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
c337ddc2	45	#define TRACE_INCLUDE_FILE lttng-statedump
3bc29f0a	46	#define LTTNG_INSTRUMENTATION
241ae9a8	47	#include <instrumentation/events/lttng-module/lttng-statedump.h>
c337ddc2	48
f0dbdefb	49	DEFINE_TRACE(lttng_statedump_block_device);
20591cf7 MD	50	DEFINE_TRACE(lttng_statedump_end);
	51	DEFINE_TRACE(lttng_statedump_interrupt);
	52	DEFINE_TRACE(lttng_statedump_file_descriptor);
	53	DEFINE_TRACE(lttng_statedump_start);
	54	DEFINE_TRACE(lttng_statedump_process_state);
1965e6b4 MJ	55	DEFINE_TRACE(lttng_statedump_process_pid_ns);
	56	#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
	57	DEFINE_TRACE(lttng_statedump_process_cgroup_ns);
	58	#endif
	59	DEFINE_TRACE(lttng_statedump_process_ipc_ns);
	60	#ifndef LTTNG_MNT_NS_MISSING_HEADER
	61	DEFINE_TRACE(lttng_statedump_process_mnt_ns);
	62	#endif
	63	DEFINE_TRACE(lttng_statedump_process_net_ns);
	64	DEFINE_TRACE(lttng_statedump_process_user_ns);
	65	DEFINE_TRACE(lttng_statedump_process_uts_ns);
20591cf7	66	DEFINE_TRACE(lttng_statedump_network_interface);
d0b55e4c	67	#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
502e4132 JD	68	DEFINE_TRACE(lttng_statedump_cpu_topology);
502e4132 JD	69	#endif
20591cf7	70
361c023a MD	71	struct lttng_fd_ctx {
	72	char *page;
	73	struct lttng_session *session;
d561ecfb	74	struct files_struct *files;
361c023a MD	75	};
361c023a MD	76
c337ddc2 MD	77	/*
	78	* Protected by the trace lock.
	79	*/
	80	static struct delayed_work cpu_work[NR_CPUS];
	81	static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
	82	static atomic_t kernel_threads_to_run;
	83
	84	enum lttng_thread_type {
	85	LTTNG_USER_THREAD = 0,
	86	LTTNG_KERNEL_THREAD = 1,
	87	};
	88
	89	enum lttng_execution_mode {
	90	LTTNG_USER_MODE = 0,
	91	LTTNG_SYSCALL = 1,
	92	LTTNG_TRAP = 2,
	93	LTTNG_IRQ = 3,
	94	LTTNG_SOFTIRQ = 4,
	95	LTTNG_MODE_UNKNOWN = 5,
	96	};
	97
	98	enum lttng_execution_submode {
	99	LTTNG_NONE = 0,
	100	LTTNG_UNKNOWN = 1,
	101	};
	102
	103	enum lttng_process_status {
	104	LTTNG_UNNAMED = 0,
	105	LTTNG_WAIT_FORK = 1,
	106	LTTNG_WAIT_CPU = 2,
	107	LTTNG_EXIT = 3,
	108	LTTNG_ZOMBIE = 4,
	109	LTTNG_WAIT = 5,
	110	LTTNG_RUN = 6,
	111	LTTNG_DEAD = 7,
	112	};
	113
f0dbdefb HD	114	static
	115	int lttng_enumerate_block_devices(struct lttng_session *session)
	116	{
f0dbdefb HD	117	struct class_dev_iter iter;
	118	struct device *dev;
	119
6afd444b MD	120	class_dev_iter_init(&iter, gendisk_block_class(), NULL,
6afd444b MD	121	gendisk_device_type());
f0dbdefb HD	122	while ((dev = class_dev_iter_next(&iter))) {
	123	struct disk_part_iter piter;
	124	struct gendisk *disk = dev_to_disk(dev);
	125	struct hd_struct *part;
	126
5a91f3df MD	127	/*
	128	* Don't show empty devices or things that have been
	129	* suppressed
	130	*/
	131	if (get_capacity(disk) == 0 \|\|
	132	(disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
	133	continue;
	134
f0dbdefb HD	135	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
	136	while ((part = disk_part_iter_next(&piter))) {
	137	char name_buf[BDEVNAME_SIZE];
	138	char *p;
	139
6afd444b	140	p = gendisk_name(disk, part->partno, name_buf);
f0dbdefb HD	141	if (!p) {
	142	disk_part_iter_exit(&piter);
	143	class_dev_iter_exit(&iter);
	144	return -ENOSYS;
	145	}
	146	trace_lttng_statedump_block_device(session,
	147	part_devt(part), name_buf);
	148	}
	149	disk_part_iter_exit(&piter);
	150	}
	151	class_dev_iter_exit(&iter);
	152	return 0;
	153	}
	154
c337ddc2	155	#ifdef CONFIG_INET
f0dbdefb	156
c337ddc2 MD	157	static
	158	void lttng_enumerate_device(struct lttng_session *session,
	159	struct net_device *dev)
	160	{
	161	struct in_device *in_dev;
	162	struct in_ifaddr *ifa;
	163
	164	if (dev->flags & IFF_UP) {
	165	in_dev = in_dev_get(dev);
	166	if (in_dev) {
	167	for (ifa = in_dev->ifa_list; ifa != NULL;
	168	ifa = ifa->ifa_next) {
	169	trace_lttng_statedump_network_interface(
	170	session, dev, ifa);
	171	}
	172	in_dev_put(in_dev);
	173	}
	174	} else {
	175	trace_lttng_statedump_network_interface(
	176	session, dev, NULL);
	177	}
	178	}
	179
	180	static
	181	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	182	{
	183	struct net_device *dev;
	184
	185	read_lock(&dev_base_lock);
	186	for_each_netdev(&init_net, dev)
	187	lttng_enumerate_device(session, dev);
	188	read_unlock(&dev_base_lock);
	189
	190	return 0;
	191	}
	192	#else /* CONFIG_INET */
	193	static inline
	194	int lttng_enumerate_network_ip_interface(struct lttng_session *session)
	195	{
	196	return 0;
	197	}
	198	#endif /* CONFIG_INET */
	199
361c023a MD	200	static
	201	int lttng_dump_one_fd(const void p, struct file file, unsigned int fd)
	202	{
	203	const struct lttng_fd_ctx *ctx = p;
	204	const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
29021503	205	unsigned int flags = file->f_flags;
d561ecfb	206	struct fdtable *fdt;
361c023a	207
29021503 MD	208	/*
	209	* We don't expose kernel internal flags, only userspace-visible
	210	* flags.
	211	*/
	212	flags &= ~FMODE_NONOTIFY;
d561ecfb MD	213	fdt = files_fdtable(ctx->files);
	214	/*
	215	* We need to check here again whether fd is within the fdt
	216	* max_fds range, because we might be seeing a different
	217	* files_fdtable() than iterate_fd(), assuming only RCU is
	218	* protecting the read. In reality, iterate_fd() holds
	219	* file_lock, which should ensure the fdt does not change while
	220	* the lock is taken, but we are not aware whether this is
	221	* guaranteed or not, so play safe.
	222	*/
997eff5e	223	if (fd < fdt->max_fds && close_on_exec(fd, fdt))
29021503	224	flags \|= O_CLOEXEC;
361c023a MD	225	if (IS_ERR(s)) {
	226	struct dentry *dentry = file->f_path.dentry;
	227
	228	/* Make sure we give at least some info */
	229	spin_lock(&dentry->d_lock);
e7a0ca72 MD	230	trace_lttng_statedump_file_descriptor(ctx->session,
	231	ctx->files, fd, dentry->d_name.name, flags,
	232	file->f_mode);
361c023a MD	233	spin_unlock(&dentry->d_lock);
	234	goto end;
	235	}
e7a0ca72 MD	236	trace_lttng_statedump_file_descriptor(ctx->session,
e7a0ca72 MD	237	ctx->files, fd, s, flags, file->f_mode);
361c023a MD	238	end:
	239	return 0;
	240	}
c337ddc2	241
e7a0ca72	242	/* Called with task lock held. */
c337ddc2	243	static
e7a0ca72 MD	244	void lttng_enumerate_files(struct lttng_session *session,
	245	struct files_struct *files,
	246	char *tmp)
c337ddc2	247	{
e7a0ca72	248	struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
c337ddc2	249
997eff5e	250	iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
c337ddc2 MD	251	}
c337ddc2 MD	252
d0b55e4c	253	#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
502e4132 JD	254	static
	255	int lttng_enumerate_cpu_topology(struct lttng_session *session)
	256	{
	257	int cpu;
	258	const cpumask_t *cpumask = cpu_possible_mask;
	259
	260	for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
	261	cpu = cpumask_next(cpu, cpumask)) {
	262	trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
	263	}
	264
	265	return 0;
	266	}
	267	#else
	268	static
	269	int lttng_enumerate_cpu_topology(struct lttng_session *session)
	270	{
	271	return 0;
	272	}
	273	#endif
	274
0658bdda MD	275	#if 0
	276	/*
	277	* FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
	278	* (scheduling in atomic). Normally, the tasklist lock protects this kind of
	279	* iteration, but it is not exported to modules.
	280	*/
c337ddc2 MD	281	static
	282	void lttng_enumerate_task_vm_maps(struct lttng_session *session,
	283	struct task_struct *p)
	284	{
	285	struct mm_struct *mm;
	286	struct vm_area_struct *map;
	287	unsigned long ino;
	288
	289	/* get_task_mm does a task_lock... */
	290	mm = get_task_mm(p);
	291	if (!mm)
	292	return;
	293
	294	map = mm->mmap;
	295	if (map) {
	296	down_read(&mm->mmap_sem);
	297	while (map) {
	298	if (map->vm_file)
34522739	299	ino = map->vm_file->f_path.dentry->d_inode->i_ino;
c337ddc2 MD	300	else
	301	ino = 0;
	302	trace_lttng_statedump_vm_map(session, p, map, ino);
	303	map = map->vm_next;
	304	}
	305	up_read(&mm->mmap_sem);
	306	}
	307	mmput(mm);
	308	}
	309
	310	static
	311	int lttng_enumerate_vm_maps(struct lttng_session *session)
	312	{
	313	struct task_struct *p;
	314
	315	rcu_read_lock();
	316	for_each_process(p)
	317	lttng_enumerate_task_vm_maps(session, p);
	318	rcu_read_unlock();
	319	return 0;
	320	}
0658bdda	321	#endif
c337ddc2	322
c337ddc2	323	static
cfcee1c7	324	int lttng_list_interrupts(struct lttng_session *session)
c337ddc2 MD	325	{
	326	unsigned int irq;
	327	unsigned long flags = 0;
	328	struct irq_desc *desc;
	329
c337ddc2 MD	330	/* needs irq_desc */
	331	for_each_irq_desc(irq, desc) {
	332	struct irqaction *action;
	333	const char *irq_chip_name =
	334	irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
	335
	336	local_irq_save(flags);
fc94c945	337	raw_spin_lock(&desc->lock);
c337ddc2 MD	338	for (action = desc->action; action; action = action->next) {
	339	trace_lttng_statedump_interrupt(session,
	340	irq, irq_chip_name, action);
	341	}
fc94c945	342	raw_spin_unlock(&desc->lock);
c337ddc2 MD	343	local_irq_restore(flags);
c337ddc2 MD	344	}
cfcee1c7	345	return 0;
c337ddc2	346	}
c337ddc2	347
4ba1f53c	348	/*
1965e6b4 MJ	349	* Statedump the task's namespaces using the proc filesystem inode number as
	350	* the unique identifier. The user and pid ns are nested and will be dumped
	351	* recursively.
	352	*
4ba1f53c MD	353	* Called with task lock held.
4ba1f53c MD	354	*/
73e8ba37 JD	355	static
	356	void lttng_statedump_process_ns(struct lttng_session *session,
	357	struct task_struct *p,
	358	enum lttng_thread_type type,
	359	enum lttng_execution_mode mode,
	360	enum lttng_execution_submode submode,
	361	enum lttng_process_status status)
	362	{
1965e6b4	363	struct nsproxy *proxy;
73e8ba37	364	struct pid_namespace *pid_ns;
1965e6b4	365	struct user_namespace *user_ns;
73e8ba37	366
1965e6b4 MJ	367	/*
	368	* The pid and user namespaces are special, they are nested and
	369	* accessed with specific functions instead of the nsproxy struct
	370	* like the other namespaces.
	371	*/
887bcdac MJ	372	pid_ns = task_active_pid_ns(p);
887bcdac MJ	373	do {
1965e6b4	374	trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
adcc8b5e	375	pid_ns = pid_ns ? pid_ns->parent : NULL;
887bcdac	376	} while (pid_ns);
1965e6b4 MJ	377
	378
	379	user_ns = task_cred_xxx(p, user_ns);
	380	do {
	381	trace_lttng_statedump_process_user_ns(session, p, user_ns);
1964cccb MD	382	/*
	383	* trace_lttng_statedump_process_user_ns() internally
	384	* checks whether user_ns is NULL. While this does not
	385	* appear to be a possible return value for
	386	* task_cred_xxx(), err on the safe side and check
	387	* for NULL here as well to be consistent with the
	388	* paranoid behavior of
	389	* trace_lttng_statedump_process_user_ns().
	390	*/
6388029c	391	user_ns = user_ns ? user_ns->parent : NULL;
1965e6b4 MJ	392	} while (user_ns);
	393
	394	/*
	395	* Back and forth on locking strategy within Linux upstream for nsproxy.
	396	* See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
	397	* "namespaces: Use task_lock and not rcu to protect nsproxy"
	398	* for details.
	399	*/
	400	#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) \|\| \
	401	LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) \|\| \
	402	LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) \|\| \
	403	LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
	404	proxy = p->nsproxy;
	405	#else
	406	rcu_read_lock();
	407	proxy = task_nsproxy(p);
	408	#endif
	409	if (proxy) {
	410	#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
	411	trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
	412	#endif
	413	trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
	414	#ifndef LTTNG_MNT_NS_MISSING_HEADER
	415	trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
	416	#endif
	417	trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
	418	trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
	419	}
	420	#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) \|\| \
	421	LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) \|\| \
	422	LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) \|\| \
	423	LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
	424	/* (nothing) */
	425	#else
	426	rcu_read_unlock();
	427	#endif
73e8ba37 JD	428	}
73e8ba37 JD	429
c337ddc2 MD	430	static
	431	int lttng_enumerate_process_states(struct lttng_session *session)
	432	{
	433	struct task_struct g, p;
e7a0ca72 MD	434	char *tmp;
	435
	436	tmp = (char *) __get_free_page(GFP_KERNEL);
	437	if (!tmp)
	438	return -ENOMEM;
c337ddc2 MD	439
	440	rcu_read_lock();
	441	for_each_process(g) {
e7a0ca72 MD	442	struct files_struct *prev_files = NULL;
e7a0ca72 MD	443
c337ddc2 MD	444	p = g;
	445	do {
	446	enum lttng_execution_mode mode =
	447	LTTNG_MODE_UNKNOWN;
	448	enum lttng_execution_submode submode =
	449	LTTNG_UNKNOWN;
	450	enum lttng_process_status status;
	451	enum lttng_thread_type type;
e7a0ca72	452	struct files_struct *files;
c337ddc2 MD	453
	454	task_lock(p);
	455	if (p->exit_state == EXIT_ZOMBIE)
	456	status = LTTNG_ZOMBIE;
	457	else if (p->exit_state == EXIT_DEAD)
	458	status = LTTNG_DEAD;
	459	else if (p->state == TASK_RUNNING) {
	460	/* Is this a forked child that has not run yet? */
	461	if (list_empty(&p->rt.run_list))
	462	status = LTTNG_WAIT_FORK;
	463	else
	464	/*
	465	* All tasks are considered as wait_cpu;
	466	* the viewer will sort out if the task
	467	* was really running at this time.
	468	*/
	469	status = LTTNG_WAIT_CPU;
	470	} else if (p->state &
	471	(TASK_INTERRUPTIBLE \| TASK_UNINTERRUPTIBLE)) {
	472	/* Task is waiting for something to complete */
	473	status = LTTNG_WAIT;
	474	} else
	475	status = LTTNG_UNNAMED;
	476	submode = LTTNG_NONE;
	477
	478	/*
	479	* Verification of t->mm is to filter out kernel
	480	* threads; Viewer will further filter out if a
	481	* user-space thread was in syscall mode or not.
	482	*/
	483	if (p->mm)
	484	type = LTTNG_USER_THREAD;
	485	else
	486	type = LTTNG_KERNEL_THREAD;
e7a0ca72	487	files = p->files;
d2a927ac MJ	488
d2a927ac MJ	489	trace_lttng_statedump_process_state(session,
e7a0ca72	490	p, type, mode, submode, status, files);
73e8ba37	491	lttng_statedump_process_ns(session,
c337ddc2	492	p, type, mode, submode, status);
e7a0ca72 MD	493	/*
	494	* As an optimisation for the common case, do not
	495	* repeat information for the same files_struct in
	496	* two consecutive threads. This is the common case
	497	* for threads sharing the same fd table. RCU guarantees
	498	* that the same files_struct pointer is not re-used
	499	* throughout processes/threads iteration.
	500	*/
	501	if (files && files != prev_files) {
	502	lttng_enumerate_files(session, files, tmp);
	503	prev_files = files;
	504	}
c337ddc2 MD	505	task_unlock(p);
	506	} while_each_thread(g, p);
	507	}
	508	rcu_read_unlock();
	509
e7a0ca72 MD	510	free_page((unsigned long) tmp);
e7a0ca72 MD	511
c337ddc2 MD	512	return 0;
	513	}
	514
	515	static
	516	void lttng_statedump_work_func(struct work_struct *work)
	517	{
	518	if (atomic_dec_and_test(&kernel_threads_to_run))
	519	/* If we are the last thread, wake up do_lttng_statedump */
	520	wake_up(&statedump_wq);
	521	}
	522
	523	static
	524	int do_lttng_statedump(struct lttng_session *session)
	525	{
cfcee1c7	526	int cpu, ret;
c337ddc2	527
c337ddc2	528	trace_lttng_statedump_start(session);
cfcee1c7	529	ret = lttng_enumerate_process_states(session);
cfcee1c7 MD	530	if (ret)
	531	return ret;
	532	/*
	533	* FIXME
	534	* ret = lttng_enumerate_vm_maps(session);
	535	* if (ret)
	536	* return ret;
	537	*/
	538	ret = lttng_list_interrupts(session);
	539	if (ret)
	540	return ret;
	541	ret = lttng_enumerate_network_ip_interface(session);
	542	if (ret)
	543	return ret;
	544	ret = lttng_enumerate_block_devices(session);
	545	switch (ret) {
84c7055e MD	546	case 0:
84c7055e MD	547	break;
cfcee1c7 MD	548	case -ENOSYS:
	549	printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
	550	break;
	551	default:
	552	return ret;
	553	}
502e4132 JD	554	ret = lttng_enumerate_cpu_topology(session);
	555	if (ret)
	556	return ret;
c337ddc2 MD	557
	558	/* TODO lttng_dump_idt_table(session); */
	559	/* TODO lttng_dump_softirq_vec(session); */
	560	/* TODO lttng_list_modules(session); */
	561	/* TODO lttng_dump_swap_files(session); */
	562
	563	/*
	564	* Fire off a work queue on each CPU. Their sole purpose in life
	565	* is to guarantee that each CPU has been in a state where is was in
	566	* syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
	567	*/
	568	get_online_cpus();
	569	atomic_set(&kernel_threads_to_run, num_online_cpus());
	570	for_each_online_cpu(cpu) {
	571	INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
	572	schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
	573	}
	574	/* Wait for all threads to run */
7a7128e0	575	__wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
c337ddc2 MD	576	put_online_cpus();
c337ddc2 MD	577	/* Our work is done */
c337ddc2 MD	578	trace_lttng_statedump_end(session);
	579	return 0;
	580	}
	581
	582	/*
	583	* Called with session mutex held.
	584	*/
	585	int lttng_statedump_start(struct lttng_session *session)
	586	{
c337ddc2 MD	587	return do_lttng_statedump(session);
	588	}
	589	EXPORT_SYMBOL_GPL(lttng_statedump_start);
	590
dd8d5afb MD	591	static
	592	int __init lttng_statedump_init(void)
	593	{
d16aa9c9	594	return 0;
dd8d5afb MD	595	}
	596
	597	module_init(lttng_statedump_init);
	598
461277e7 MD	599	static
	600	void __exit lttng_statedump_exit(void)
	601	{
	602	}
	603
	604	module_exit(lttng_statedump_exit);
	605
c337ddc2 MD	606	MODULE_LICENSE("GPL and additional rights");
c337ddc2 MD	607	MODULE_AUTHOR("Jean-Hugues Deschenes");
1c124020	608	MODULE_DESCRIPTION("LTTng statedump provider");
13ab8b0a MD	609	MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
	610	__stringify(LTTNG_MODULES_MINOR_VERSION) "."
	611	__stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
	612	LTTNG_MODULES_EXTRAVERSION);