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 <wrapper/cpu.h>
40 #include <wrapper/irqdesc.h>
41 #include <wrapper/fdtable.h>
42 #include <wrapper/tracepoint.h>
43 #include <wrapper/blkdev.h>
44 #include <wrapper/fdtable.h>
45 #include <wrapper/sched.h>
47 /* Define the tracepoints, but do not build the probes */
48 #define CREATE_TRACE_POINTS
49 #define TRACE_INCLUDE_PATH instrumentation/events
50 #define TRACE_INCLUDE_FILE lttng-statedump
51 #define LTTNG_INSTRUMENTATION
52 #include <instrumentation/events/lttng-statedump.h>
54 LTTNG_DEFINE_TRACE(lttng_statedump_block_device
,
55 TP_PROTO(struct lttng_kernel_session
*session
,
56 dev_t dev
, const char *diskname
),
57 TP_ARGS(session
, dev
, diskname
));
59 LTTNG_DEFINE_TRACE(lttng_statedump_end
,
60 TP_PROTO(struct lttng_kernel_session
*session
),
63 LTTNG_DEFINE_TRACE(lttng_statedump_interrupt
,
64 TP_PROTO(struct lttng_kernel_session
*session
,
65 unsigned int irq
, const char *chip_name
,
66 struct irqaction
*action
),
67 TP_ARGS(session
, irq
, chip_name
, action
));
69 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor
,
70 TP_PROTO(struct lttng_kernel_session
*session
,
71 struct files_struct
*files
,
72 int fd
, const char *filename
,
73 unsigned int flags
, fmode_t fmode
),
74 TP_ARGS(session
, files
, fd
, filename
, flags
, fmode
));
76 LTTNG_DEFINE_TRACE(lttng_statedump_start
,
77 TP_PROTO(struct lttng_kernel_session
*session
),
80 LTTNG_DEFINE_TRACE(lttng_statedump_process_state
,
81 TP_PROTO(struct lttng_kernel_session
*session
,
82 struct task_struct
*p
,
83 int type
, int mode
, int submode
, int status
,
84 struct files_struct
*files
),
85 TP_ARGS(session
, p
, type
, mode
, submode
, status
, files
));
87 LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns
,
88 TP_PROTO(struct lttng_kernel_session
*session
,
89 struct task_struct
*p
,
90 struct pid_namespace
*pid_ns
),
91 TP_ARGS(session
, p
, pid_ns
));
93 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
94 LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns
,
95 TP_PROTO(struct lttng_kernel_session
*session
,
96 struct task_struct
*p
,
97 struct cgroup_namespace
*cgroup_ns
),
98 TP_ARGS(session
, p
, cgroup_ns
));
101 LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns
,
102 TP_PROTO(struct lttng_kernel_session
*session
,
103 struct task_struct
*p
,
104 struct ipc_namespace
*ipc_ns
),
105 TP_ARGS(session
, p
, ipc_ns
));
107 #ifndef LTTNG_MNT_NS_MISSING_HEADER
108 LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns
,
109 TP_PROTO(struct lttng_kernel_session
*session
,
110 struct task_struct
*p
,
111 struct mnt_namespace
*mnt_ns
),
112 TP_ARGS(session
, p
, mnt_ns
));
115 LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns
,
116 TP_PROTO(struct lttng_kernel_session
*session
,
117 struct task_struct
*p
,
119 TP_ARGS(session
, p
, net_ns
));
121 LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns
,
122 TP_PROTO(struct lttng_kernel_session
*session
,
123 struct task_struct
*p
,
124 struct user_namespace
*user_ns
),
125 TP_ARGS(session
, p
, user_ns
));
127 LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns
,
128 TP_PROTO(struct lttng_kernel_session
*session
,
129 struct task_struct
*p
,
130 struct uts_namespace
*uts_ns
),
131 TP_ARGS(session
, p
, uts_ns
));
133 LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns
,
134 TP_PROTO(struct lttng_kernel_session
*session
,
135 struct task_struct
*p
,
136 struct time_namespace
*time_ns
),
137 TP_ARGS(session
, p
, time_ns
));
139 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface
,
140 TP_PROTO(struct lttng_kernel_session
*session
,
141 struct net_device
*dev
, struct in_ifaddr
*ifa
),
142 TP_ARGS(session
, dev
, ifa
));
144 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
145 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology
,
146 TP_PROTO(struct lttng_kernel_session
*session
, struct cpuinfo_x86
*c
),
147 TP_ARGS(session
, c
));
150 struct lttng_fd_ctx
{
152 struct lttng_kernel_session
*session
;
153 struct files_struct
*files
;
157 * Protected by the trace lock.
159 static struct delayed_work cpu_work
[NR_CPUS
];
160 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
161 static atomic_t kernel_threads_to_run
;
163 enum lttng_thread_type
{
164 LTTNG_USER_THREAD
= 0,
165 LTTNG_KERNEL_THREAD
= 1,
168 enum lttng_execution_mode
{
174 LTTNG_MODE_UNKNOWN
= 5,
177 enum lttng_execution_submode
{
182 enum lttng_process_status
{
194 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0))
196 #define LTTNG_PART_STRUCT_TYPE struct block_device
199 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
203 ret
= snprintf(name_buf
, BDEVNAME_SIZE
, "%pg", part
);
204 if (ret
< 0 || ret
>= BDEVNAME_SIZE
)
211 dev_t
lttng_get_part_devt(struct block_device
*part
)
216 #elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
218 #define LTTNG_PART_STRUCT_TYPE struct block_device
221 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
225 p
= bdevname(part
, name_buf
);
233 dev_t
lttng_get_part_devt(struct block_device
*part
)
240 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
243 int lttng_get_part_name(struct gendisk
*disk
, struct hd_struct
*part
, char *name_buf
)
246 struct block_device bdev
;
249 * Create a partial 'struct blockdevice' to use
250 * 'bdevname()' which is a simple wrapper over
251 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
256 p
= bdevname(&bdev
, name_buf
);
264 dev_t
lttng_get_part_devt(struct hd_struct
*part
)
266 return part_devt(part
);
270 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
272 int lttng_statedump_each_block_device(struct lttng_kernel_session
*session
, struct gendisk
*disk
)
274 struct block_device
*part
;
278 /* Include partition 0 */
282 xa_for_each(&disk
->part_tbl
, idx
, part
) {
283 char name_buf
[BDEVNAME_SIZE
];
285 /* Exclude non-partitions bdev and empty partitions. */
286 if (bdev_is_partition(part
) && !bdev_nr_sectors(part
))
289 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
293 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
302 int lttng_statedump_each_block_device(struct lttng_kernel_session
*session
, struct gendisk
*disk
)
304 struct disk_part_iter piter
;
305 LTTNG_PART_STRUCT_TYPE
*part
;
307 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
309 while ((part
= disk_part_iter_next(&piter
))) {
310 char name_buf
[BDEVNAME_SIZE
];
312 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
313 disk_part_iter_exit(&piter
);
316 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
319 disk_part_iter_exit(&piter
);
326 int lttng_enumerate_block_devices(struct lttng_kernel_session
*session
)
328 struct class *ptr_block_class
;
329 struct device_type
*ptr_disk_type
;
330 struct class_dev_iter iter
;
334 ptr_block_class
= wrapper_get_block_class();
335 if (!ptr_block_class
) {
339 ptr_disk_type
= wrapper_get_disk_type();
340 if (!ptr_disk_type
) {
344 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
345 while ((dev
= class_dev_iter_next(&iter
))) {
346 struct gendisk
*disk
= dev_to_disk(dev
);
349 * Don't show empty devices or things that have been
352 if (get_capacity(disk
) == 0 ||
353 (disk
->flags
& LTTNG_GENHD_FL_HIDDEN
))
356 ret
= lttng_statedump_each_block_device(session
, disk
);
358 class_dev_iter_exit(&iter
);
366 void lttng_enumerate_device(struct lttng_kernel_session
*session
,
367 struct net_device
*dev
)
369 struct in_device
*in_dev
;
370 struct in_ifaddr
*ifa
;
372 if (dev
->flags
& IFF_UP
) {
373 in_dev
= in_dev_get(dev
);
375 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
376 ifa
= ifa
->ifa_next
) {
377 trace_lttng_statedump_network_interface(
383 trace_lttng_statedump_network_interface(
389 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session
*session
)
391 struct net_device
*dev
;
393 read_lock(&dev_base_lock
);
394 for_each_netdev(&init_net
, dev
)
395 lttng_enumerate_device(session
, dev
);
396 read_unlock(&dev_base_lock
);
400 #else /* CONFIG_INET */
402 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session
*session
)
406 #endif /* CONFIG_INET */
409 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
411 const struct lttng_fd_ctx
*ctx
= p
;
412 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
413 unsigned int flags
= file
->f_flags
;
417 * We don't expose kernel internal flags, only userspace-visible
420 flags
&= ~FMODE_NONOTIFY
;
421 fdt
= files_fdtable(ctx
->files
);
423 * We need to check here again whether fd is within the fdt
424 * max_fds range, because we might be seeing a different
425 * files_fdtable() than iterate_fd(), assuming only RCU is
426 * protecting the read. In reality, iterate_fd() holds
427 * file_lock, which should ensure the fdt does not change while
428 * the lock is taken, but we are not aware whether this is
429 * guaranteed or not, so play safe.
431 if (fd
< fdt
->max_fds
&& close_on_exec(fd
, fdt
))
434 struct dentry
*dentry
= file
->f_path
.dentry
;
436 /* Make sure we give at least some info */
437 spin_lock(&dentry
->d_lock
);
438 trace_lttng_statedump_file_descriptor(ctx
->session
,
439 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
441 spin_unlock(&dentry
->d_lock
);
444 trace_lttng_statedump_file_descriptor(ctx
->session
,
445 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
450 /* Called with task lock held. */
452 void lttng_enumerate_files(struct lttng_kernel_session
*session
,
453 struct files_struct
*files
,
456 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
458 iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
461 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
463 int lttng_enumerate_cpu_topology(struct lttng_kernel_session
*session
)
466 const cpumask_t
*cpumask
= cpu_possible_mask
;
468 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
469 cpu
= cpumask_next(cpu
, cpumask
)) {
470 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
477 int lttng_enumerate_cpu_topology(struct lttng_kernel_session
*session
)
485 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
486 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
487 * iteration, but it is not exported to modules.
490 void lttng_enumerate_task_vm_maps(struct lttng_kernel_session
*session
,
491 struct task_struct
*p
)
493 struct mm_struct
*mm
;
494 struct vm_area_struct
*map
;
497 /* get_task_mm does a task_lock... */
504 down_read(&mm
->mmap_sem
);
507 ino
= map
->vm_file
->f_path
.dentry
->d_inode
->i_ino
;
510 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
513 up_read(&mm
->mmap_sem
);
519 int lttng_enumerate_vm_maps(struct lttng_kernel_session
*session
)
521 struct task_struct
*p
;
525 lttng_enumerate_task_vm_maps(session
, p
);
532 int lttng_list_interrupts(struct lttng_kernel_session
*session
)
535 unsigned long flags
= 0;
536 struct irq_desc
*desc
;
538 #define irq_to_desc wrapper_irq_to_desc
540 for_each_irq_desc(irq
, desc
) {
541 struct irqaction
*action
;
542 const char *irq_chip_name
=
543 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
545 local_irq_save(flags
);
546 raw_spin_lock(&desc
->lock
);
547 for (action
= desc
->action
; action
; action
= action
->next
) {
548 trace_lttng_statedump_interrupt(session
,
549 irq
, irq_chip_name
, action
);
551 raw_spin_unlock(&desc
->lock
);
552 local_irq_restore(flags
);
559 * Statedump the task's namespaces using the proc filesystem inode number as
560 * the unique identifier. The user and pid ns are nested and will be dumped
563 * Called with task lock held.
566 void lttng_statedump_process_ns(struct lttng_kernel_session
*session
,
567 struct task_struct
*p
,
568 enum lttng_thread_type type
,
569 enum lttng_execution_mode mode
,
570 enum lttng_execution_submode submode
,
571 enum lttng_process_status status
)
573 struct nsproxy
*proxy
;
574 struct pid_namespace
*pid_ns
;
575 struct user_namespace
*user_ns
;
578 * The pid and user namespaces are special, they are nested and
579 * accessed with specific functions instead of the nsproxy struct
580 * like the other namespaces.
582 pid_ns
= task_active_pid_ns(p
);
584 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
585 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
589 user_ns
= task_cred_xxx(p
, user_ns
);
591 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
593 * trace_lttng_statedump_process_user_ns() internally
594 * checks whether user_ns is NULL. While this does not
595 * appear to be a possible return value for
596 * task_cred_xxx(), err on the safe side and check
597 * for NULL here as well to be consistent with the
598 * paranoid behavior of
599 * trace_lttng_statedump_process_user_ns().
601 user_ns
= user_ns
? user_ns
->parent
: NULL
;
607 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
608 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
610 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
611 #ifndef LTTNG_MNT_NS_MISSING_HEADER
612 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
614 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
615 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
616 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \
617 LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0))
618 trace_lttng_statedump_process_time_ns(session
, p
, proxy
->time_ns
);
624 int lttng_enumerate_process_states(struct lttng_kernel_session
*session
)
626 struct task_struct
*g
, *p
;
629 tmp
= (char *) __get_free_page(GFP_KERNEL
);
634 for_each_process(g
) {
635 struct files_struct
*prev_files
= NULL
;
639 enum lttng_execution_mode mode
=
641 enum lttng_execution_submode submode
=
643 enum lttng_process_status status
;
644 enum lttng_thread_type type
;
645 struct files_struct
*files
;
648 if (p
->exit_state
== EXIT_ZOMBIE
)
649 status
= LTTNG_ZOMBIE
;
650 else if (p
->exit_state
== EXIT_DEAD
)
652 else if (lttng_task_is_running(p
)) {
653 /* Is this a forked child that has not run yet? */
654 if (list_empty(&p
->rt
.run_list
))
655 status
= LTTNG_WAIT_FORK
;
658 * All tasks are considered as wait_cpu;
659 * the viewer will sort out if the task
660 * was really running at this time.
662 status
= LTTNG_WAIT_CPU
;
663 } else if (lttng_get_task_state(p
) &
664 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
665 /* Task is waiting for something to complete */
668 status
= LTTNG_UNNAMED
;
669 submode
= LTTNG_NONE
;
672 * Verification of t->mm is to filter out kernel
673 * threads; Viewer will further filter out if a
674 * user-space thread was in syscall mode or not.
677 type
= LTTNG_USER_THREAD
;
679 type
= LTTNG_KERNEL_THREAD
;
682 trace_lttng_statedump_process_state(session
,
683 p
, type
, mode
, submode
, status
, files
);
684 lttng_statedump_process_ns(session
,
685 p
, type
, mode
, submode
, status
);
687 * As an optimisation for the common case, do not
688 * repeat information for the same files_struct in
689 * two consecutive threads. This is the common case
690 * for threads sharing the same fd table. RCU guarantees
691 * that the same files_struct pointer is not re-used
692 * throughout processes/threads iteration.
694 if (files
&& files
!= prev_files
) {
695 lttng_enumerate_files(session
, files
, tmp
);
699 } while_each_thread(g
, p
);
703 free_page((unsigned long) tmp
);
709 void lttng_statedump_work_func(struct work_struct
*work
)
711 if (atomic_dec_and_test(&kernel_threads_to_run
))
712 /* If we are the last thread, wake up do_lttng_statedump */
713 wake_up(&statedump_wq
);
717 int do_lttng_statedump(struct lttng_kernel_session
*session
)
721 trace_lttng_statedump_start(session
);
722 ret
= lttng_enumerate_process_states(session
);
727 * ret = lttng_enumerate_vm_maps(session);
731 ret
= lttng_list_interrupts(session
);
734 ret
= lttng_enumerate_network_ip_interface(session
);
737 ret
= lttng_enumerate_block_devices(session
);
742 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
747 ret
= lttng_enumerate_cpu_topology(session
);
751 /* TODO lttng_dump_idt_table(session); */
752 /* TODO lttng_dump_softirq_vec(session); */
753 /* TODO lttng_list_modules(session); */
754 /* TODO lttng_dump_swap_files(session); */
757 * Fire off a work queue on each CPU. Their sole purpose in life
758 * is to guarantee that each CPU has been in a state where is was in
759 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
761 lttng_cpus_read_lock();
762 atomic_set(&kernel_threads_to_run
, num_online_cpus());
763 for_each_online_cpu(cpu
) {
764 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
765 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
767 /* Wait for all threads to run */
768 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
769 lttng_cpus_read_unlock();
770 /* Our work is done */
771 trace_lttng_statedump_end(session
);
776 * Called with session mutex held.
778 int lttng_statedump_start(struct lttng_kernel_session
*session
)
780 return do_lttng_statedump(session
);
782 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
785 int __init
lttng_statedump_init(void)
790 module_init(lttng_statedump_init
);
793 void __exit
lttng_statedump_exit(void)
797 module_exit(lttng_statedump_exit
);
799 MODULE_LICENSE("GPL and additional rights");
800 MODULE_AUTHOR("Jean-Hugues Deschenes");
801 MODULE_DESCRIPTION("LTTng statedump provider");
802 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
803 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
804 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
805 LTTNG_MODULES_EXTRAVERSION
);