1 /* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
5 * Linux Trace Toolkit Next Generation Kernel State Dump
7 * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
8 * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
11 * Eric Clement: Add listing of network IP interface
12 * 2006, 2007 Mathieu Desnoyers Fix kernel threads
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/netlink.h>
19 #include <linux/inet.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>
31 #include <linux/swap.h>
32 #include <linux/wait.h>
33 #include <linux/mutex.h>
34 #include <linux/device.h>
36 #include <lttng/events.h>
37 #include <lttng/tracer.h>
38 #include <wrapper/irqdesc.h>
39 #include <wrapper/fdtable.h>
40 #include <wrapper/namespace.h>
41 #include <wrapper/irq.h>
42 #include <wrapper/tracepoint.h>
43 #include <wrapper/genhd.h>
44 #include <wrapper/file.h>
45 #include <wrapper/fdtable.h>
47 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
48 #include <linux/irq.h>
51 /* Define the tracepoints, but do not build the probes */
52 #define CREATE_TRACE_POINTS
53 #define TRACE_INCLUDE_PATH instrumentation/events
54 #define TRACE_INCLUDE_FILE lttng-statedump
55 #define LTTNG_INSTRUMENTATION
56 #include <instrumentation/events/lttng-statedump.h>
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
));
63 LTTNG_DEFINE_TRACE(lttng_statedump_end
,
64 TP_PROTO(struct lttng_session
*session
),
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
));
73 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor
,
74 TP_PROTO(struct lttng_session
*session
,
75 struct files_struct
*files
,
76 int fd
, const char *filename
,
77 unsigned int flags
, fmode_t fmode
),
78 TP_ARGS(session
, files
, fd
, filename
, flags
, fmode
));
80 LTTNG_DEFINE_TRACE(lttng_statedump_start
,
81 TP_PROTO(struct lttng_session
*session
),
84 LTTNG_DEFINE_TRACE(lttng_statedump_process_state
,
85 TP_PROTO(struct lttng_session
*session
,
86 struct task_struct
*p
,
87 int type
, int mode
, int submode
, int status
,
88 struct files_struct
*files
),
89 TP_ARGS(session
, p
, type
, mode
, submode
, status
, files
));
91 LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns
,
92 TP_PROTO(struct lttng_session
*session
,
93 struct task_struct
*p
,
94 struct pid_namespace
*pid_ns
),
95 TP_ARGS(session
, p
, pid_ns
));
97 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
98 LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns
,
99 TP_PROTO(struct lttng_session
*session
,
100 struct task_struct
*p
,
101 struct cgroup_namespace
*cgroup_ns
),
102 TP_ARGS(session
, p
, cgroup_ns
));
105 LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns
,
106 TP_PROTO(struct lttng_session
*session
,
107 struct task_struct
*p
,
108 struct ipc_namespace
*ipc_ns
),
109 TP_ARGS(session
, p
, ipc_ns
));
111 #ifndef LTTNG_MNT_NS_MISSING_HEADER
112 LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns
,
113 TP_PROTO(struct lttng_session
*session
,
114 struct task_struct
*p
,
115 struct mnt_namespace
*mnt_ns
),
116 TP_ARGS(session
, p
, mnt_ns
));
119 LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns
,
120 TP_PROTO(struct lttng_session
*session
,
121 struct task_struct
*p
,
123 TP_ARGS(session
, p
, net_ns
));
125 LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns
,
126 TP_PROTO(struct lttng_session
*session
,
127 struct task_struct
*p
,
128 struct user_namespace
*user_ns
),
129 TP_ARGS(session
, p
, user_ns
));
131 LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns
,
132 TP_PROTO(struct lttng_session
*session
,
133 struct task_struct
*p
,
134 struct uts_namespace
*uts_ns
),
135 TP_ARGS(session
, p
, uts_ns
));
137 LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns
,
138 TP_PROTO(struct lttng_session
*session
,
139 struct task_struct
*p
,
140 struct time_namespace
*time_ns
),
141 TP_ARGS(session
, p
, time_ns
));
143 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface
,
144 TP_PROTO(struct lttng_session
*session
,
145 struct net_device
*dev
, struct in_ifaddr
*ifa
),
146 TP_ARGS(session
, dev
, ifa
));
148 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
149 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology
,
150 TP_PROTO(struct lttng_session
*session
, struct cpuinfo_x86
*c
),
151 TP_ARGS(session
, c
));
154 struct lttng_fd_ctx
{
156 struct lttng_session
*session
;
157 struct files_struct
*files
;
161 * Protected by the trace lock.
163 static struct delayed_work cpu_work
[NR_CPUS
];
164 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
165 static atomic_t kernel_threads_to_run
;
167 enum lttng_thread_type
{
168 LTTNG_USER_THREAD
= 0,
169 LTTNG_KERNEL_THREAD
= 1,
172 enum lttng_execution_mode
{
178 LTTNG_MODE_UNKNOWN
= 5,
181 enum lttng_execution_submode
{
186 enum lttng_process_status
{
198 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
200 #define LTTNG_PART_STRUCT_TYPE struct block_device
203 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
207 p
= bdevname(part
, name_buf
);
215 dev_t
lttng_get_part_devt(struct block_device
*part
)
222 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
225 int lttng_get_part_name(struct gendisk
*disk
, struct hd_struct
*part
, char *name_buf
)
228 struct block_device bdev
;
231 * Create a partial 'struct blockdevice' to use
232 * 'bdevname()' which is a simple wrapper over
233 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
238 p
= bdevname(&bdev
, name_buf
);
246 dev_t
lttng_get_part_devt(struct hd_struct
*part
)
248 return part_devt(part
);
253 int lttng_enumerate_block_devices(struct lttng_session
*session
)
255 struct class *ptr_block_class
;
256 struct device_type
*ptr_disk_type
;
257 struct class_dev_iter iter
;
261 ptr_block_class
= wrapper_get_block_class();
262 if (!ptr_block_class
) {
266 ptr_disk_type
= wrapper_get_disk_type();
267 if (!ptr_disk_type
) {
271 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
272 while ((dev
= class_dev_iter_next(&iter
))) {
273 struct disk_part_iter piter
;
274 struct gendisk
*disk
= dev_to_disk(dev
);
275 LTTNG_PART_STRUCT_TYPE
*part
;
278 * Don't show empty devices or things that have been
281 if (get_capacity(disk
) == 0 ||
282 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
286 * The original 'disk_part_iter_init' returns void, but our
287 * wrapper can fail to lookup the original symbol.
289 if (wrapper_disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
) < 0) {
294 while ((part
= wrapper_disk_part_iter_next(&piter
))) {
295 char name_buf
[BDEVNAME_SIZE
];
297 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
298 wrapper_disk_part_iter_exit(&piter
);
302 trace_lttng_statedump_block_device(session
,
303 lttng_get_part_devt(part
), name_buf
);
305 wrapper_disk_part_iter_exit(&piter
);
308 class_dev_iter_exit(&iter
);
316 void lttng_enumerate_device(struct lttng_session
*session
,
317 struct net_device
*dev
)
319 struct in_device
*in_dev
;
320 struct in_ifaddr
*ifa
;
322 if (dev
->flags
& IFF_UP
) {
323 in_dev
= in_dev_get(dev
);
325 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
326 ifa
= ifa
->ifa_next
) {
327 trace_lttng_statedump_network_interface(
333 trace_lttng_statedump_network_interface(
339 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
341 struct net_device
*dev
;
343 read_lock(&dev_base_lock
);
344 for_each_netdev(&init_net
, dev
)
345 lttng_enumerate_device(session
, dev
);
346 read_unlock(&dev_base_lock
);
350 #else /* CONFIG_INET */
352 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
356 #endif /* CONFIG_INET */
359 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
361 const struct lttng_fd_ctx
*ctx
= p
;
362 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
363 unsigned int flags
= file
->f_flags
;
367 * We don't expose kernel internal flags, only userspace-visible
370 flags
&= ~FMODE_NONOTIFY
;
371 fdt
= files_fdtable(ctx
->files
);
373 * We need to check here again whether fd is within the fdt
374 * max_fds range, because we might be seeing a different
375 * files_fdtable() than iterate_fd(), assuming only RCU is
376 * protecting the read. In reality, iterate_fd() holds
377 * file_lock, which should ensure the fdt does not change while
378 * the lock is taken, but we are not aware whether this is
379 * guaranteed or not, so play safe.
381 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
384 struct dentry
*dentry
= file
->f_path
.dentry
;
386 /* Make sure we give at least some info */
387 spin_lock(&dentry
->d_lock
);
388 trace_lttng_statedump_file_descriptor(ctx
->session
,
389 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
391 spin_unlock(&dentry
->d_lock
);
394 trace_lttng_statedump_file_descriptor(ctx
->session
,
395 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
400 /* Called with task lock held. */
402 void lttng_enumerate_files(struct lttng_session
*session
,
403 struct files_struct
*files
,
406 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
408 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
411 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
413 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
416 const cpumask_t
*cpumask
= cpu_possible_mask
;
418 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
419 cpu
= cpumask_next(cpu
, cpumask
)) {
420 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
427 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
435 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
436 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
437 * iteration, but it is not exported to modules.
440 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
441 struct task_struct
*p
)
443 struct mm_struct
*mm
;
444 struct vm_area_struct
*map
;
447 /* get_task_mm does a task_lock... */
454 down_read(&mm
->mmap_sem
);
457 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
460 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
463 up_read(&mm
->mmap_sem
);
469 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
471 struct task_struct
*p
;
475 lttng_enumerate_task_vm_maps(session
, p
);
481 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
484 int lttng_list_interrupts(struct lttng_session
*session
)
487 unsigned long flags
= 0;
488 struct irq_desc
*desc
;
490 #define irq_to_desc wrapper_irq_to_desc
492 for_each_irq_desc(irq
, desc
) {
493 struct irqaction
*action
;
494 const char *irq_chip_name
=
495 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
497 local_irq_save(flags
);
498 raw_spin_lock(&desc
->lock
);
499 for (action
= desc
->action
; action
; action
= action
->next
) {
500 trace_lttng_statedump_interrupt(session
,
501 irq
, irq_chip_name
, action
);
503 raw_spin_unlock(&desc
->lock
);
504 local_irq_restore(flags
);
511 int lttng_list_interrupts(struct lttng_session
*session
)
518 * Statedump the task's namespaces using the proc filesystem inode number as
519 * the unique identifier. The user and pid ns are nested and will be dumped
522 * Called with task lock held.
525 void lttng_statedump_process_ns(struct lttng_session
*session
,
526 struct task_struct
*p
,
527 enum lttng_thread_type type
,
528 enum lttng_execution_mode mode
,
529 enum lttng_execution_submode submode
,
530 enum lttng_process_status status
)
532 struct nsproxy
*proxy
;
533 struct pid_namespace
*pid_ns
;
534 struct user_namespace
*user_ns
;
537 * The pid and user namespaces are special, they are nested and
538 * accessed with specific functions instead of the nsproxy struct
539 * like the other namespaces.
541 pid_ns
= task_active_pid_ns(p
);
543 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
544 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
548 user_ns
= task_cred_xxx(p
, user_ns
);
550 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
552 * trace_lttng_statedump_process_user_ns() internally
553 * checks whether user_ns is NULL. While this does not
554 * appear to be a possible return value for
555 * task_cred_xxx(), err on the safe side and check
556 * for NULL here as well to be consistent with the
557 * paranoid behavior of
558 * trace_lttng_statedump_process_user_ns().
560 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
564 * Back and forth on locking strategy within Linux upstream for nsproxy.
565 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
566 * "namespaces: Use task_lock and not rcu to protect nsproxy"
569 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
570 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
571 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
572 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
576 proxy
= task_nsproxy(p
);
579 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
580 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
582 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
583 #ifndef LTTNG_MNT_NS_MISSING_HEADER
584 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
586 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
587 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
588 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0))
589 trace_lttng_statedump_process_time_ns(session
, p
, proxy
->time_ns
);
592 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
593 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
594 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
595 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
603 int lttng_enumerate_process_states(struct lttng_session
*session
)
605 struct task_struct
*g
, *p
;
608 tmp
= (char *) __get_free_page(GFP_KERNEL
);
613 for_each_process(g
) {
614 struct files_struct
*prev_files
= NULL
;
618 enum lttng_execution_mode mode
=
620 enum lttng_execution_submode submode
=
622 enum lttng_process_status status
;
623 enum lttng_thread_type type
;
624 struct files_struct
*files
;
627 if (p
->exit_state
== EXIT_ZOMBIE
)
628 status
= LTTNG_ZOMBIE
;
629 else if (p
->exit_state
== EXIT_DEAD
)
631 else if (p
->state
== TASK_RUNNING
) {
632 /* Is this a forked child that has not run yet? */
633 if (list_empty(&p
->rt
.run_list
))
634 status
= LTTNG_WAIT_FORK
;
637 * All tasks are considered as wait_cpu;
638 * the viewer will sort out if the task
639 * was really running at this time.
641 status
= LTTNG_WAIT_CPU
;
642 } else if (p
->state
&
643 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
644 /* Task is waiting for something to complete */
647 status
= LTTNG_UNNAMED
;
648 submode
= LTTNG_NONE
;
651 * Verification of t->mm is to filter out kernel
652 * threads; Viewer will further filter out if a
653 * user-space thread was in syscall mode or not.
656 type
= LTTNG_USER_THREAD
;
658 type
= LTTNG_KERNEL_THREAD
;
661 trace_lttng_statedump_process_state(session
,
662 p
, type
, mode
, submode
, status
, files
);
663 lttng_statedump_process_ns(session
,
664 p
, type
, mode
, submode
, status
);
666 * As an optimisation for the common case, do not
667 * repeat information for the same files_struct in
668 * two consecutive threads. This is the common case
669 * for threads sharing the same fd table. RCU guarantees
670 * that the same files_struct pointer is not re-used
671 * throughout processes/threads iteration.
673 if (files
&& files
!= prev_files
) {
674 lttng_enumerate_files(session
, files
, tmp
);
678 } while_each_thread(g
, p
);
682 free_page((unsigned long) tmp
);
688 void lttng_statedump_work_func(struct work_struct
*work
)
690 if (atomic_dec_and_test(&kernel_threads_to_run
))
691 /* If we are the last thread, wake up do_lttng_statedump */
692 wake_up(&statedump_wq
);
696 int do_lttng_statedump(struct lttng_session
*session
)
700 trace_lttng_statedump_start(session
);
701 ret
= lttng_enumerate_process_states(session
);
706 * ret = lttng_enumerate_vm_maps(session);
710 ret
= lttng_list_interrupts(session
);
713 ret
= lttng_enumerate_network_ip_interface(session
);
716 ret
= lttng_enumerate_block_devices(session
);
721 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
726 ret
= lttng_enumerate_cpu_topology(session
);
730 /* TODO lttng_dump_idt_table(session); */
731 /* TODO lttng_dump_softirq_vec(session); */
732 /* TODO lttng_list_modules(session); */
733 /* TODO lttng_dump_swap_files(session); */
736 * Fire off a work queue on each CPU. Their sole purpose in life
737 * is to guarantee that each CPU has been in a state where is was in
738 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
741 atomic_set(&kernel_threads_to_run
, num_online_cpus());
742 for_each_online_cpu(cpu
) {
743 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
744 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
746 /* Wait for all threads to run */
747 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
749 /* Our work is done */
750 trace_lttng_statedump_end(session
);
755 * Called with session mutex held.
757 int lttng_statedump_start(struct lttng_session
*session
)
759 return do_lttng_statedump(session
);
761 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
764 int __init
lttng_statedump_init(void)
767 * Allow module to load even if the fixup cannot be done. This
768 * will allow seemless transition when the underlying issue fix
769 * is merged into the Linux kernel, and when tracepoint.c
770 * "tracepoint_module_notify" is turned into a static function.
772 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
776 module_init(lttng_statedump_init
);
779 void __exit
lttng_statedump_exit(void)
783 module_exit(lttng_statedump_exit
);
785 MODULE_LICENSE("GPL and additional rights");
786 MODULE_AUTHOR("Jean-Hugues Deschenes");
787 MODULE_DESCRIPTION("LTTng statedump provider");
788 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
789 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
790 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
791 LTTNG_MODULES_EXTRAVERSION
);