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/irq.h>
26 #include <linux/irqnr.h>
27 #include <linux/netdevice.h>
28 #include <linux/inetdevice.h>
30 #include <linux/swap.h>
31 #include <linux/wait.h>
32 #include <linux/mutex.h>
33 #include <linux/device.h>
35 #include <linux/blkdev.h>
37 #include <lttng/events.h>
38 #include <lttng/tracer.h>
39 #include <lttng/events-internal.h>
40 #include <wrapper/cpu.h>
41 #include <wrapper/irqdesc.h>
42 #include <wrapper/fdtable.h>
43 #include <wrapper/tracepoint.h>
44 #include <wrapper/blkdev.h>
45 #include <wrapper/fdtable.h>
46 #include <wrapper/sched.h>
48 /* Define the tracepoints, but do not build the probes */
49 #define CREATE_TRACE_POINTS
50 #define TRACE_INCLUDE_PATH instrumentation/events
51 #define TRACE_INCLUDE_FILE lttng-statedump
52 #define LTTNG_INSTRUMENTATION
53 #include <instrumentation/events/lttng-statedump.h>
55 LTTNG_DEFINE_TRACE(lttng_statedump_block_device
,
56 TP_PROTO(struct lttng_kernel_session
*session
,
57 dev_t dev
, const char *diskname
),
58 TP_ARGS(session
, dev
, diskname
));
60 LTTNG_DEFINE_TRACE(lttng_statedump_end
,
61 TP_PROTO(struct lttng_kernel_session
*session
),
64 LTTNG_DEFINE_TRACE(lttng_statedump_interrupt
,
65 TP_PROTO(struct lttng_kernel_session
*session
,
66 unsigned int irq
, const char *chip_name
,
67 struct irqaction
*action
),
68 TP_ARGS(session
, irq
, chip_name
, action
));
70 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor
,
71 TP_PROTO(struct lttng_kernel_session
*session
,
72 struct files_struct
*files
,
73 int fd
, const char *filename
,
74 unsigned int flags
, fmode_t fmode
),
75 TP_ARGS(session
, files
, fd
, filename
, flags
, fmode
));
77 LTTNG_DEFINE_TRACE(lttng_statedump_start
,
78 TP_PROTO(struct lttng_kernel_session
*session
),
81 LTTNG_DEFINE_TRACE(lttng_statedump_process_state
,
82 TP_PROTO(struct lttng_kernel_session
*session
,
83 struct task_struct
*p
,
84 int type
, int mode
, int submode
, int status
,
85 struct files_struct
*files
),
86 TP_ARGS(session
, p
, type
, mode
, submode
, status
, files
));
88 LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns
,
89 TP_PROTO(struct lttng_kernel_session
*session
,
90 struct task_struct
*p
,
91 struct pid_namespace
*pid_ns
),
92 TP_ARGS(session
, p
, pid_ns
));
94 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
95 LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns
,
96 TP_PROTO(struct lttng_kernel_session
*session
,
97 struct task_struct
*p
,
98 struct cgroup_namespace
*cgroup_ns
),
99 TP_ARGS(session
, p
, cgroup_ns
));
102 LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns
,
103 TP_PROTO(struct lttng_kernel_session
*session
,
104 struct task_struct
*p
,
105 struct ipc_namespace
*ipc_ns
),
106 TP_ARGS(session
, p
, ipc_ns
));
108 #ifndef LTTNG_MNT_NS_MISSING_HEADER
109 LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns
,
110 TP_PROTO(struct lttng_kernel_session
*session
,
111 struct task_struct
*p
,
112 struct mnt_namespace
*mnt_ns
),
113 TP_ARGS(session
, p
, mnt_ns
));
116 LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns
,
117 TP_PROTO(struct lttng_kernel_session
*session
,
118 struct task_struct
*p
,
120 TP_ARGS(session
, p
, net_ns
));
122 LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns
,
123 TP_PROTO(struct lttng_kernel_session
*session
,
124 struct task_struct
*p
,
125 struct user_namespace
*user_ns
),
126 TP_ARGS(session
, p
, user_ns
));
128 LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns
,
129 TP_PROTO(struct lttng_kernel_session
*session
,
130 struct task_struct
*p
,
131 struct uts_namespace
*uts_ns
),
132 TP_ARGS(session
, p
, uts_ns
));
134 LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns
,
135 TP_PROTO(struct lttng_kernel_session
*session
,
136 struct task_struct
*p
,
137 struct time_namespace
*time_ns
),
138 TP_ARGS(session
, p
, time_ns
));
140 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface
,
141 TP_PROTO(struct lttng_kernel_session
*session
,
142 struct net_device
*dev
, struct in_ifaddr
*ifa
),
143 TP_ARGS(session
, dev
, ifa
));
145 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
146 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology
,
147 TP_PROTO(struct lttng_kernel_session
*session
, struct cpuinfo_x86
*c
),
148 TP_ARGS(session
, c
));
151 struct lttng_fd_ctx
{
153 struct lttng_kernel_session
*session
;
154 struct files_struct
*files
;
158 * Protected by the trace lock.
160 static struct delayed_work cpu_work
[NR_CPUS
];
161 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
162 static atomic_t kernel_threads_to_run
;
164 enum lttng_thread_type
{
165 LTTNG_USER_THREAD
= 0,
166 LTTNG_KERNEL_THREAD
= 1,
169 enum lttng_execution_mode
{
175 LTTNG_MODE_UNKNOWN
= 5,
178 enum lttng_execution_submode
{
183 enum lttng_process_status
{
195 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0) || \
196 LTTNG_RHEL_KERNEL_RANGE(5,14,0,163,0,0, 5,15,0,0,0,0))
198 #define LTTNG_PART_STRUCT_TYPE struct block_device
201 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
205 ret
= snprintf(name_buf
, BDEVNAME_SIZE
, "%pg", part
);
206 if (ret
< 0 || ret
>= BDEVNAME_SIZE
)
213 dev_t
lttng_get_part_devt(struct block_device
*part
)
218 #elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
220 #define LTTNG_PART_STRUCT_TYPE struct block_device
223 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
227 p
= bdevname(part
, name_buf
);
235 dev_t
lttng_get_part_devt(struct block_device
*part
)
242 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
245 int lttng_get_part_name(struct gendisk
*disk
, struct hd_struct
*part
, char *name_buf
)
248 struct block_device bdev
;
251 * Create a partial 'struct blockdevice' to use
252 * 'bdevname()' which is a simple wrapper over
253 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
258 p
= bdevname(&bdev
, name_buf
);
266 dev_t
lttng_get_part_devt(struct hd_struct
*part
)
268 return part_devt(part
);
272 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
274 int lttng_statedump_each_block_device(struct lttng_kernel_session
*session
, struct gendisk
*disk
)
276 struct block_device
*part
;
280 /* Include partition 0 */
284 xa_for_each(&disk
->part_tbl
, idx
, part
) {
285 char name_buf
[BDEVNAME_SIZE
];
287 /* Exclude non-partitions bdev and empty partitions. */
288 if (bdev_is_partition(part
) && !bdev_nr_sectors(part
))
291 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
295 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
304 int lttng_statedump_each_block_device(struct lttng_kernel_session
*session
, struct gendisk
*disk
)
306 struct disk_part_iter piter
;
307 LTTNG_PART_STRUCT_TYPE
*part
;
309 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
311 while ((part
= disk_part_iter_next(&piter
))) {
312 char name_buf
[BDEVNAME_SIZE
];
314 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
315 disk_part_iter_exit(&piter
);
318 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
321 disk_part_iter_exit(&piter
);
328 int lttng_enumerate_block_devices(struct lttng_kernel_session
*session
)
330 struct class *ptr_block_class
;
331 struct device_type
*ptr_disk_type
;
332 struct class_dev_iter iter
;
336 ptr_block_class
= wrapper_get_block_class();
337 if (!ptr_block_class
) {
341 ptr_disk_type
= wrapper_get_disk_type();
342 if (!ptr_disk_type
) {
346 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
347 while ((dev
= class_dev_iter_next(&iter
))) {
348 struct gendisk
*disk
= dev_to_disk(dev
);
351 * Don't show empty devices or things that have been
354 if (get_capacity(disk
) == 0 ||
355 (disk
->flags
& LTTNG_GENHD_FL_HIDDEN
))
358 ret
= lttng_statedump_each_block_device(session
, disk
);
360 class_dev_iter_exit(&iter
);
368 void lttng_enumerate_device(struct lttng_kernel_session
*session
,
369 struct net_device
*dev
)
371 struct in_device
*in_dev
;
372 struct in_ifaddr
*ifa
;
374 if (dev
->flags
& IFF_UP
) {
375 in_dev
= in_dev_get(dev
);
377 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
378 ifa
= ifa
->ifa_next
) {
379 trace_lttng_statedump_network_interface(
385 trace_lttng_statedump_network_interface(
390 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0))
392 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session
*session
)
394 struct net_device
*dev
;
397 for_each_netdev(&init_net
, dev
)
398 lttng_enumerate_device(session
, dev
);
405 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session
*session
)
407 struct net_device
*dev
;
409 read_lock(&dev_base_lock
);
410 for_each_netdev(&init_net
, dev
)
411 lttng_enumerate_device(session
, dev
);
412 read_unlock(&dev_base_lock
);
416 #endif /* (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0)) */
417 #else /* CONFIG_INET */
419 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session
*session
)
423 #endif /* CONFIG_INET */
426 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
428 const struct lttng_fd_ctx
*ctx
= p
;
429 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
430 unsigned int flags
= file
->f_flags
;
434 * We don't expose kernel internal flags, only userspace-visible
437 flags
&= ~FMODE_NONOTIFY
;
438 fdt
= files_fdtable(ctx
->files
);
440 * We need to check here again whether fd is within the fdt
441 * max_fds range, because we might be seeing a different
442 * files_fdtable() than iterate_fd(), assuming only RCU is
443 * protecting the read. In reality, iterate_fd() holds
444 * file_lock, which should ensure the fdt does not change while
445 * the lock is taken, but we are not aware whether this is
446 * guaranteed or not, so play safe.
448 if (fd
< fdt
->max_fds
&& close_on_exec(fd
, fdt
))
451 struct dentry
*dentry
= file
->f_path
.dentry
;
453 /* Make sure we give at least some info */
454 spin_lock(&dentry
->d_lock
);
455 trace_lttng_statedump_file_descriptor(ctx
->session
,
456 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
458 spin_unlock(&dentry
->d_lock
);
461 trace_lttng_statedump_file_descriptor(ctx
->session
,
462 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
467 /* Called with task lock held. */
469 void lttng_enumerate_files(struct lttng_kernel_session
*session
,
470 struct files_struct
*files
,
473 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
475 iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
478 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
480 int lttng_enumerate_cpu_topology(struct lttng_kernel_session
*session
)
483 const cpumask_t
*cpumask
= cpu_possible_mask
;
485 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
486 cpu
= cpumask_next(cpu
, cpumask
)) {
487 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
494 int lttng_enumerate_cpu_topology(struct lttng_kernel_session
*session
)
502 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
503 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
504 * iteration, but it is not exported to modules.
507 void lttng_enumerate_task_vm_maps(struct lttng_kernel_session
*session
,
508 struct task_struct
*p
)
510 struct mm_struct
*mm
;
511 struct vm_area_struct
*map
;
514 /* get_task_mm does a task_lock... */
521 down_read(&mm
->mmap_sem
);
524 ino
= map
->vm_file
->f_path
.dentry
->d_inode
->i_ino
;
527 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
530 up_read(&mm
->mmap_sem
);
536 int lttng_enumerate_vm_maps(struct lttng_kernel_session
*session
)
538 struct task_struct
*p
;
542 lttng_enumerate_task_vm_maps(session
, p
);
549 int lttng_list_interrupts(struct lttng_kernel_session
*session
)
552 unsigned long flags
= 0;
553 struct irq_desc
*desc
;
555 #define irq_to_desc wrapper_irq_to_desc
557 for_each_irq_desc(irq
, desc
) {
558 struct irqaction
*action
;
559 const char *irq_chip_name
=
560 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
562 local_irq_save(flags
);
563 raw_spin_lock(&desc
->lock
);
564 for (action
= desc
->action
; action
; action
= action
->next
) {
565 trace_lttng_statedump_interrupt(session
,
566 irq
, irq_chip_name
, action
);
568 raw_spin_unlock(&desc
->lock
);
569 local_irq_restore(flags
);
576 * Statedump the task's namespaces using the proc filesystem inode number as
577 * the unique identifier. The user and pid ns are nested and will be dumped
580 * Called with task lock held.
583 void lttng_statedump_process_ns(struct lttng_kernel_session
*session
,
584 struct task_struct
*p
,
585 enum lttng_thread_type type
,
586 enum lttng_execution_mode mode
,
587 enum lttng_execution_submode submode
,
588 enum lttng_process_status status
)
590 struct nsproxy
*proxy
;
591 struct pid_namespace
*pid_ns
;
592 struct user_namespace
*user_ns
;
595 * The pid and user namespaces are special, they are nested and
596 * accessed with specific functions instead of the nsproxy struct
597 * like the other namespaces.
599 pid_ns
= task_active_pid_ns(p
);
601 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
602 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
606 user_ns
= task_cred_xxx(p
, user_ns
);
608 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
610 * trace_lttng_statedump_process_user_ns() internally
611 * checks whether user_ns is NULL. While this does not
612 * appear to be a possible return value for
613 * task_cred_xxx(), err on the safe side and check
614 * for NULL here as well to be consistent with the
615 * paranoid behavior of
616 * trace_lttng_statedump_process_user_ns().
618 user_ns
= user_ns
? user_ns
->parent
: NULL
;
624 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
625 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
627 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
628 #ifndef LTTNG_MNT_NS_MISSING_HEADER
629 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
631 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
632 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
633 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \
634 LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0))
635 trace_lttng_statedump_process_time_ns(session
, p
, proxy
->time_ns
);
641 int lttng_enumerate_process_states(struct lttng_kernel_session
*session
)
643 struct task_struct
*g
, *p
;
646 tmp
= (char *) __get_free_page(GFP_KERNEL
);
651 for_each_process(g
) {
652 struct files_struct
*prev_files
= NULL
;
656 enum lttng_execution_mode mode
=
658 enum lttng_execution_submode submode
=
660 enum lttng_process_status status
;
661 enum lttng_thread_type type
;
662 struct files_struct
*files
;
665 if (p
->exit_state
== EXIT_ZOMBIE
)
666 status
= LTTNG_ZOMBIE
;
667 else if (p
->exit_state
== EXIT_DEAD
)
669 else if (lttng_task_is_running(p
)) {
670 /* Is this a forked child that has not run yet? */
671 if (list_empty(&p
->rt
.run_list
))
672 status
= LTTNG_WAIT_FORK
;
675 * All tasks are considered as wait_cpu;
676 * the viewer will sort out if the task
677 * was really running at this time.
679 status
= LTTNG_WAIT_CPU
;
680 } else if (lttng_get_task_state(p
) &
681 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
682 /* Task is waiting for something to complete */
685 status
= LTTNG_UNNAMED
;
686 submode
= LTTNG_NONE
;
689 * Verification of t->mm is to filter out kernel
690 * threads; Viewer will further filter out if a
691 * user-space thread was in syscall mode or not.
694 type
= LTTNG_USER_THREAD
;
696 type
= LTTNG_KERNEL_THREAD
;
699 trace_lttng_statedump_process_state(session
,
700 p
, type
, mode
, submode
, status
, files
);
701 lttng_statedump_process_ns(session
,
702 p
, type
, mode
, submode
, status
);
704 * As an optimisation for the common case, do not
705 * repeat information for the same files_struct in
706 * two consecutive threads. This is the common case
707 * for threads sharing the same fd table. RCU guarantees
708 * that the same files_struct pointer is not re-used
709 * throughout processes/threads iteration.
711 if (files
&& files
!= prev_files
) {
712 lttng_enumerate_files(session
, files
, tmp
);
716 } while_each_thread(g
, p
);
720 free_page((unsigned long) tmp
);
726 void lttng_statedump_work_func(struct work_struct
*work
)
728 if (atomic_dec_and_test(&kernel_threads_to_run
))
729 /* If we are the last thread, wake up do_lttng_statedump */
730 wake_up(&statedump_wq
);
734 int do_lttng_statedump(struct lttng_kernel_session
*session
)
738 trace_lttng_statedump_start(session
);
739 ret
= lttng_enumerate_process_states(session
);
744 * ret = lttng_enumerate_vm_maps(session);
748 ret
= lttng_list_interrupts(session
);
751 ret
= lttng_enumerate_network_ip_interface(session
);
754 ret
= lttng_enumerate_block_devices(session
);
759 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
764 ret
= lttng_enumerate_cpu_topology(session
);
768 /* TODO lttng_dump_idt_table(session); */
769 /* TODO lttng_dump_softirq_vec(session); */
770 /* TODO lttng_list_modules(session); */
771 /* TODO lttng_dump_swap_files(session); */
774 * Fire off a work queue on each CPU. Their sole purpose in life
775 * is to guarantee that each CPU has been in a state where is was in
776 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
778 lttng_cpus_read_lock();
779 atomic_set(&kernel_threads_to_run
, num_online_cpus());
780 for_each_online_cpu(cpu
) {
781 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
782 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
784 /* Wait for all threads to run */
785 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
786 lttng_cpus_read_unlock();
787 /* Our work is done */
788 trace_lttng_statedump_end(session
);
793 * Called with session mutex held.
795 int lttng_statedump_start(struct lttng_kernel_session
*session
)
797 return do_lttng_statedump(session
);
799 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
802 int __init
lttng_statedump_init(void)
807 module_init(lttng_statedump_init
);
810 void __exit
lttng_statedump_exit(void)
814 module_exit(lttng_statedump_exit
);
816 MODULE_LICENSE("GPL and additional rights");
817 MODULE_AUTHOR("Jean-Hugues Deschenes");
818 MODULE_DESCRIPTION("LTTng statedump provider");
819 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
820 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
821 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
822 LTTNG_MODULES_EXTRAVERSION
);