1 /* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
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/lttng-module
54 #define TRACE_INCLUDE_FILE lttng-statedump
55 #define LTTNG_INSTRUMENTATION
56 #include <instrumentation/events/lttng-module/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_network_interface
,
138 TP_PROTO(struct lttng_session
*session
,
139 struct net_device
*dev
, struct in_ifaddr
*ifa
),
140 TP_ARGS(session
, dev
, ifa
));
142 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
143 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology
,
144 TP_PROTO(struct lttng_session
*session
, struct cpuinfo_x86
*c
),
145 TP_ARGS(session
, c
));
148 struct lttng_fd_ctx
{
150 struct lttng_session
*session
;
151 struct files_struct
*files
;
155 * Protected by the trace lock.
157 static struct delayed_work cpu_work
[NR_CPUS
];
158 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
159 static atomic_t kernel_threads_to_run
;
161 enum lttng_thread_type
{
162 LTTNG_USER_THREAD
= 0,
163 LTTNG_KERNEL_THREAD
= 1,
166 enum lttng_execution_mode
{
172 LTTNG_MODE_UNKNOWN
= 5,
175 enum lttng_execution_submode
{
180 enum lttng_process_status
{
192 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
194 #define LTTNG_PART_STRUCT_TYPE struct block_device
197 int lttng_get_part_name(struct gendisk
*disk
, struct block_device
*part
, char *name_buf
)
201 p
= bdevname(part
, name_buf
);
209 dev_t
lttng_get_part_devt(struct block_device
*part
)
216 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
219 int lttng_get_part_name(struct gendisk
*disk
, struct hd_struct
*part
, char *name_buf
)
222 struct block_device bdev
;
225 * Create a partial 'struct blockdevice' to use
226 * 'bdevname()' which is a simple wrapper over
227 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
232 p
= bdevname(&bdev
, name_buf
);
240 dev_t
lttng_get_part_devt(struct hd_struct
*part
)
242 return part_devt(part
);
246 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
248 int lttng_statedump_each_block_device(struct lttng_session
*session
, struct gendisk
*disk
)
250 struct block_device
*part
;
254 /* Include partition 0 */
258 xa_for_each(&disk
->part_tbl
, idx
, part
) {
259 char name_buf
[BDEVNAME_SIZE
];
261 /* Exclude non-partitions bdev and empty partitions. */
262 if (bdev_is_partition(part
) && !bdev_nr_sectors(part
))
265 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
269 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
278 int lttng_statedump_each_block_device(struct lttng_session
*session
, struct gendisk
*disk
)
280 struct disk_part_iter piter
;
281 LTTNG_PART_STRUCT_TYPE
*part
;
283 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
285 while ((part
= disk_part_iter_next(&piter
))) {
286 char name_buf
[BDEVNAME_SIZE
];
288 if (lttng_get_part_name(disk
, part
, name_buf
) == -ENOSYS
) {
289 disk_part_iter_exit(&piter
);
292 trace_lttng_statedump_block_device(session
, lttng_get_part_devt(part
),
295 disk_part_iter_exit(&piter
);
302 int lttng_enumerate_block_devices(struct lttng_session
*session
)
304 struct class *ptr_block_class
;
305 struct device_type
*ptr_disk_type
;
306 struct class_dev_iter iter
;
310 ptr_block_class
= wrapper_get_block_class();
311 if (!ptr_block_class
) {
315 ptr_disk_type
= wrapper_get_disk_type();
316 if (!ptr_disk_type
) {
320 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
321 while ((dev
= class_dev_iter_next(&iter
))) {
322 struct gendisk
*disk
= dev_to_disk(dev
);
325 * Don't show empty devices or things that have been
328 if (get_capacity(disk
) == 0 ||
329 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
332 ret
= lttng_statedump_each_block_device(session
, disk
);
334 class_dev_iter_exit(&iter
);
342 void lttng_enumerate_device(struct lttng_session
*session
,
343 struct net_device
*dev
)
345 struct in_device
*in_dev
;
346 struct in_ifaddr
*ifa
;
348 if (dev
->flags
& IFF_UP
) {
349 in_dev
= in_dev_get(dev
);
351 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
352 ifa
= ifa
->ifa_next
) {
353 trace_lttng_statedump_network_interface(
359 trace_lttng_statedump_network_interface(
365 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
367 struct net_device
*dev
;
369 read_lock(&dev_base_lock
);
370 for_each_netdev(&init_net
, dev
)
371 lttng_enumerate_device(session
, dev
);
372 read_unlock(&dev_base_lock
);
376 #else /* CONFIG_INET */
378 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
382 #endif /* CONFIG_INET */
385 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
387 const struct lttng_fd_ctx
*ctx
= p
;
388 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
389 unsigned int flags
= file
->f_flags
;
393 * We don't expose kernel internal flags, only userspace-visible
396 flags
&= ~FMODE_NONOTIFY
;
397 fdt
= files_fdtable(ctx
->files
);
399 * We need to check here again whether fd is within the fdt
400 * max_fds range, because we might be seeing a different
401 * files_fdtable() than iterate_fd(), assuming only RCU is
402 * protecting the read. In reality, iterate_fd() holds
403 * file_lock, which should ensure the fdt does not change while
404 * the lock is taken, but we are not aware whether this is
405 * guaranteed or not, so play safe.
407 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
410 struct dentry
*dentry
= file
->f_path
.dentry
;
412 /* Make sure we give at least some info */
413 spin_lock(&dentry
->d_lock
);
414 trace_lttng_statedump_file_descriptor(ctx
->session
,
415 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
417 spin_unlock(&dentry
->d_lock
);
420 trace_lttng_statedump_file_descriptor(ctx
->session
,
421 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
426 /* Called with task lock held. */
428 void lttng_enumerate_files(struct lttng_session
*session
,
429 struct files_struct
*files
,
432 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
434 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
437 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
439 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
442 const cpumask_t
*cpumask
= cpu_possible_mask
;
444 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
445 cpu
= cpumask_next(cpu
, cpumask
)) {
446 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
453 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
461 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
462 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
463 * iteration, but it is not exported to modules.
466 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
467 struct task_struct
*p
)
469 struct mm_struct
*mm
;
470 struct vm_area_struct
*map
;
473 /* get_task_mm does a task_lock... */
480 down_read(&mm
->mmap_sem
);
483 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
486 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
489 up_read(&mm
->mmap_sem
);
495 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
497 struct task_struct
*p
;
501 lttng_enumerate_task_vm_maps(session
, p
);
507 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
510 int lttng_list_interrupts(struct lttng_session
*session
)
513 unsigned long flags
= 0;
514 struct irq_desc
*desc
;
516 #define irq_to_desc wrapper_irq_to_desc
518 for_each_irq_desc(irq
, desc
) {
519 struct irqaction
*action
;
520 const char *irq_chip_name
=
521 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
523 local_irq_save(flags
);
524 raw_spin_lock(&desc
->lock
);
525 for (action
= desc
->action
; action
; action
= action
->next
) {
526 trace_lttng_statedump_interrupt(session
,
527 irq
, irq_chip_name
, action
);
529 raw_spin_unlock(&desc
->lock
);
530 local_irq_restore(flags
);
537 int lttng_list_interrupts(struct lttng_session
*session
)
544 * Statedump the task's namespaces using the proc filesystem inode number as
545 * the unique identifier. The user and pid ns are nested and will be dumped
548 * Called with task lock held.
551 void lttng_statedump_process_ns(struct lttng_session
*session
,
552 struct task_struct
*p
,
553 enum lttng_thread_type type
,
554 enum lttng_execution_mode mode
,
555 enum lttng_execution_submode submode
,
556 enum lttng_process_status status
)
558 struct nsproxy
*proxy
;
559 struct pid_namespace
*pid_ns
;
560 struct user_namespace
*user_ns
;
563 * The pid and user namespaces are special, they are nested and
564 * accessed with specific functions instead of the nsproxy struct
565 * like the other namespaces.
567 pid_ns
= task_active_pid_ns(p
);
569 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
570 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
574 user_ns
= task_cred_xxx(p
, user_ns
);
576 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
578 * trace_lttng_statedump_process_user_ns() internally
579 * checks whether user_ns is NULL. While this does not
580 * appear to be a possible return value for
581 * task_cred_xxx(), err on the safe side and check
582 * for NULL here as well to be consistent with the
583 * paranoid behavior of
584 * trace_lttng_statedump_process_user_ns().
586 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
590 * Back and forth on locking strategy within Linux upstream for nsproxy.
591 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
592 * "namespaces: Use task_lock and not rcu to protect nsproxy"
595 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
596 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
597 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
598 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
602 proxy
= task_nsproxy(p
);
605 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
606 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
608 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
609 #ifndef LTTNG_MNT_NS_MISSING_HEADER
610 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
612 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
613 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
615 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
616 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
617 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
618 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
626 int lttng_enumerate_process_states(struct lttng_session
*session
)
628 struct task_struct
*g
, *p
;
631 tmp
= (char *) __get_free_page(GFP_KERNEL
);
636 for_each_process(g
) {
637 struct files_struct
*prev_files
= NULL
;
641 enum lttng_execution_mode mode
=
643 enum lttng_execution_submode submode
=
645 enum lttng_process_status status
;
646 enum lttng_thread_type type
;
647 struct files_struct
*files
;
650 if (p
->exit_state
== EXIT_ZOMBIE
)
651 status
= LTTNG_ZOMBIE
;
652 else if (p
->exit_state
== EXIT_DEAD
)
654 else if (p
->state
== TASK_RUNNING
) {
655 /* Is this a forked child that has not run yet? */
656 if (list_empty(&p
->rt
.run_list
))
657 status
= LTTNG_WAIT_FORK
;
660 * All tasks are considered as wait_cpu;
661 * the viewer will sort out if the task
662 * was really running at this time.
664 status
= LTTNG_WAIT_CPU
;
665 } else if (p
->state
&
666 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
667 /* Task is waiting for something to complete */
670 status
= LTTNG_UNNAMED
;
671 submode
= LTTNG_NONE
;
674 * Verification of t->mm is to filter out kernel
675 * threads; Viewer will further filter out if a
676 * user-space thread was in syscall mode or not.
679 type
= LTTNG_USER_THREAD
;
681 type
= LTTNG_KERNEL_THREAD
;
684 trace_lttng_statedump_process_state(session
,
685 p
, type
, mode
, submode
, status
, files
);
686 lttng_statedump_process_ns(session
,
687 p
, type
, mode
, submode
, status
);
689 * As an optimisation for the common case, do not
690 * repeat information for the same files_struct in
691 * two consecutive threads. This is the common case
692 * for threads sharing the same fd table. RCU guarantees
693 * that the same files_struct pointer is not re-used
694 * throughout processes/threads iteration.
696 if (files
&& files
!= prev_files
) {
697 lttng_enumerate_files(session
, files
, tmp
);
701 } while_each_thread(g
, p
);
705 free_page((unsigned long) tmp
);
711 void lttng_statedump_work_func(struct work_struct
*work
)
713 if (atomic_dec_and_test(&kernel_threads_to_run
))
714 /* If we are the last thread, wake up do_lttng_statedump */
715 wake_up(&statedump_wq
);
719 int do_lttng_statedump(struct lttng_session
*session
)
723 trace_lttng_statedump_start(session
);
724 ret
= lttng_enumerate_process_states(session
);
729 * ret = lttng_enumerate_vm_maps(session);
733 ret
= lttng_list_interrupts(session
);
736 ret
= lttng_enumerate_network_ip_interface(session
);
739 ret
= lttng_enumerate_block_devices(session
);
744 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
749 ret
= lttng_enumerate_cpu_topology(session
);
753 /* TODO lttng_dump_idt_table(session); */
754 /* TODO lttng_dump_softirq_vec(session); */
755 /* TODO lttng_list_modules(session); */
756 /* TODO lttng_dump_swap_files(session); */
759 * Fire off a work queue on each CPU. Their sole purpose in life
760 * is to guarantee that each CPU has been in a state where is was in
761 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
764 atomic_set(&kernel_threads_to_run
, num_online_cpus());
765 for_each_online_cpu(cpu
) {
766 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
767 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
769 /* Wait for all threads to run */
770 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
772 /* Our work is done */
773 trace_lttng_statedump_end(session
);
778 * Called with session mutex held.
780 int lttng_statedump_start(struct lttng_session
*session
)
782 return do_lttng_statedump(session
);
784 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
787 int __init
lttng_statedump_init(void)
790 * Allow module to load even if the fixup cannot be done. This
791 * will allow seemless transition when the underlying issue fix
792 * is merged into the Linux kernel, and when tracepoint.c
793 * "tracepoint_module_notify" is turned into a static function.
795 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
799 module_init(lttng_statedump_init
);
802 void __exit
lttng_statedump_exit(void)
806 module_exit(lttng_statedump_exit
);
808 MODULE_LICENSE("GPL and additional rights");
809 MODULE_AUTHOR("Jean-Hugues Deschenes");
810 MODULE_DESCRIPTION("LTTng statedump provider");
811 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
812 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
813 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
814 LTTNG_MODULES_EXTRAVERSION
);