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/fdtable.h>
32 #include <linux/swap.h>
33 #include <linux/wait.h>
34 #include <linux/mutex.h>
35 #include <linux/device.h>
37 #include <lttng-events.h>
38 #include <lttng-tracer.h>
39 #include <wrapper/irqdesc.h>
40 #include <wrapper/fdtable.h>
41 #include <wrapper/irq.h>
42 #include <wrapper/tracepoint.h>
43 #include <wrapper/genhd.h>
44 #include <wrapper/file.h>
45 #include <wrapper/time.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 DEFINE_TRACE(lttng_statedump_block_device
);
59 DEFINE_TRACE(lttng_statedump_end
);
60 DEFINE_TRACE(lttng_statedump_interrupt
);
61 DEFINE_TRACE(lttng_statedump_file_descriptor
);
62 DEFINE_TRACE(lttng_statedump_start
);
63 DEFINE_TRACE(lttng_statedump_process_state
);
64 DEFINE_TRACE(lttng_statedump_network_interface
);
65 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
66 DEFINE_TRACE(lttng_statedump_cpu_topology
);
71 struct lttng_session
*session
;
72 struct task_struct
*p
;
73 struct files_struct
*files
;
77 * Protected by the trace lock.
79 static struct delayed_work cpu_work
[NR_CPUS
];
80 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
81 static atomic_t kernel_threads_to_run
;
83 enum lttng_thread_type
{
84 LTTNG_USER_THREAD
= 0,
85 LTTNG_KERNEL_THREAD
= 1,
88 enum lttng_execution_mode
{
94 LTTNG_MODE_UNKNOWN
= 5,
97 enum lttng_execution_submode
{
102 enum lttng_process_status
{
114 int lttng_enumerate_block_devices(struct lttng_session
*session
)
116 struct class *ptr_block_class
;
117 struct device_type
*ptr_disk_type
;
118 struct class_dev_iter iter
;
121 ptr_block_class
= wrapper_get_block_class();
122 if (!ptr_block_class
)
124 ptr_disk_type
= wrapper_get_disk_type();
125 if (!ptr_disk_type
) {
128 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
129 while ((dev
= class_dev_iter_next(&iter
))) {
130 struct disk_part_iter piter
;
131 struct gendisk
*disk
= dev_to_disk(dev
);
132 struct hd_struct
*part
;
135 * Don't show empty devices or things that have been
138 if (get_capacity(disk
) == 0 ||
139 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
142 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
143 while ((part
= disk_part_iter_next(&piter
))) {
144 char name_buf
[BDEVNAME_SIZE
];
147 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
149 disk_part_iter_exit(&piter
);
150 class_dev_iter_exit(&iter
);
153 trace_lttng_statedump_block_device(session
,
154 part_devt(part
), name_buf
);
156 disk_part_iter_exit(&piter
);
158 class_dev_iter_exit(&iter
);
165 void lttng_enumerate_device(struct lttng_session
*session
,
166 struct net_device
*dev
)
168 struct in_device
*in_dev
;
169 struct in_ifaddr
*ifa
;
171 if (dev
->flags
& IFF_UP
) {
172 in_dev
= in_dev_get(dev
);
174 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
175 ifa
= ifa
->ifa_next
) {
176 trace_lttng_statedump_network_interface(
182 trace_lttng_statedump_network_interface(
188 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
190 struct net_device
*dev
;
192 read_lock(&dev_base_lock
);
193 for_each_netdev(&init_net
, dev
)
194 lttng_enumerate_device(session
, dev
);
195 read_unlock(&dev_base_lock
);
199 #else /* CONFIG_INET */
201 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
205 #endif /* CONFIG_INET */
208 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
210 const struct lttng_fd_ctx
*ctx
= p
;
211 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
212 unsigned int flags
= file
->f_flags
;
216 * We don't expose kernel internal flags, only userspace-visible
219 flags
&= ~FMODE_NONOTIFY
;
220 fdt
= files_fdtable(ctx
->files
);
222 * We need to check here again whether fd is within the fdt
223 * max_fds range, because we might be seeing a different
224 * files_fdtable() than iterate_fd(), assuming only RCU is
225 * protecting the read. In reality, iterate_fd() holds
226 * file_lock, which should ensure the fdt does not change while
227 * the lock is taken, but we are not aware whether this is
228 * guaranteed or not, so play safe.
230 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
233 struct dentry
*dentry
= file
->f_path
.dentry
;
235 /* Make sure we give at least some info */
236 spin_lock(&dentry
->d_lock
);
237 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
,
238 dentry
->d_name
.name
, flags
, file
->f_mode
);
239 spin_unlock(&dentry
->d_lock
);
242 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
, s
,
243 flags
, file
->f_mode
);
249 void lttng_enumerate_task_fd(struct lttng_session
*session
,
250 struct task_struct
*p
, char *tmp
)
252 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .p
= p
};
253 struct files_struct
*files
;
260 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
266 int lttng_enumerate_file_descriptors(struct lttng_session
*session
)
268 struct task_struct
*p
;
271 tmp
= (char *) __get_free_page(GFP_KERNEL
);
275 /* Enumerate active file descriptors */
278 lttng_enumerate_task_fd(session
, p
, tmp
);
280 free_page((unsigned long) tmp
);
284 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
286 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
289 const cpumask_t
*cpumask
= cpu_possible_mask
;
291 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
292 cpu
= cpumask_next(cpu
, cpumask
)) {
293 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
300 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
308 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
309 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
310 * iteration, but it is not exported to modules.
313 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
314 struct task_struct
*p
)
316 struct mm_struct
*mm
;
317 struct vm_area_struct
*map
;
320 /* get_task_mm does a task_lock... */
327 down_read(&mm
->mmap_sem
);
330 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
333 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
336 up_read(&mm
->mmap_sem
);
342 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
344 struct task_struct
*p
;
348 lttng_enumerate_task_vm_maps(session
, p
);
354 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
357 int lttng_list_interrupts(struct lttng_session
*session
)
360 unsigned long flags
= 0;
361 struct irq_desc
*desc
;
363 #define irq_to_desc wrapper_irq_to_desc
365 for_each_irq_desc(irq
, desc
) {
366 struct irqaction
*action
;
367 const char *irq_chip_name
=
368 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
370 local_irq_save(flags
);
371 raw_spin_lock(&desc
->lock
);
372 for (action
= desc
->action
; action
; action
= action
->next
) {
373 trace_lttng_statedump_interrupt(session
,
374 irq
, irq_chip_name
, action
);
376 raw_spin_unlock(&desc
->lock
);
377 local_irq_restore(flags
);
384 int lttng_list_interrupts(struct lttng_session
*session
)
391 * Called with task lock held.
394 void lttng_statedump_process_ns(struct lttng_session
*session
,
395 struct task_struct
*p
,
396 enum lttng_thread_type type
,
397 enum lttng_execution_mode mode
,
398 enum lttng_execution_submode submode
,
399 enum lttng_process_status status
)
401 struct pid_namespace
*pid_ns
;
403 pid_ns
= task_active_pid_ns(p
);
405 trace_lttng_statedump_process_state(session
,
406 p
, type
, mode
, submode
, status
, pid_ns
);
407 pid_ns
= pid_ns
->parent
;
412 int lttng_enumerate_process_states(struct lttng_session
*session
)
414 struct task_struct
*g
, *p
;
417 for_each_process(g
) {
420 enum lttng_execution_mode mode
=
422 enum lttng_execution_submode submode
=
424 enum lttng_process_status status
;
425 enum lttng_thread_type type
;
428 if (p
->exit_state
== EXIT_ZOMBIE
)
429 status
= LTTNG_ZOMBIE
;
430 else if (p
->exit_state
== EXIT_DEAD
)
432 else if (p
->state
== TASK_RUNNING
) {
433 /* Is this a forked child that has not run yet? */
434 if (list_empty(&p
->rt
.run_list
))
435 status
= LTTNG_WAIT_FORK
;
438 * All tasks are considered as wait_cpu;
439 * the viewer will sort out if the task
440 * was really running at this time.
442 status
= LTTNG_WAIT_CPU
;
443 } else if (p
->state
&
444 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
445 /* Task is waiting for something to complete */
448 status
= LTTNG_UNNAMED
;
449 submode
= LTTNG_NONE
;
452 * Verification of t->mm is to filter out kernel
453 * threads; Viewer will further filter out if a
454 * user-space thread was in syscall mode or not.
457 type
= LTTNG_USER_THREAD
;
459 type
= LTTNG_KERNEL_THREAD
;
460 lttng_statedump_process_ns(session
,
461 p
, type
, mode
, submode
, status
);
463 } while_each_thread(g
, p
);
471 void lttng_statedump_work_func(struct work_struct
*work
)
473 if (atomic_dec_and_test(&kernel_threads_to_run
))
474 /* If we are the last thread, wake up do_lttng_statedump */
475 wake_up(&statedump_wq
);
479 int do_lttng_statedump(struct lttng_session
*session
)
483 trace_lttng_statedump_start(session
);
484 ret
= lttng_enumerate_process_states(session
);
487 ret
= lttng_enumerate_file_descriptors(session
);
492 * ret = lttng_enumerate_vm_maps(session);
496 ret
= lttng_list_interrupts(session
);
499 ret
= lttng_enumerate_network_ip_interface(session
);
502 ret
= lttng_enumerate_block_devices(session
);
507 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
512 ret
= lttng_enumerate_cpu_topology(session
);
516 /* TODO lttng_dump_idt_table(session); */
517 /* TODO lttng_dump_softirq_vec(session); */
518 /* TODO lttng_list_modules(session); */
519 /* TODO lttng_dump_swap_files(session); */
522 * Fire off a work queue on each CPU. Their sole purpose in life
523 * is to guarantee that each CPU has been in a state where is was in
524 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
527 atomic_set(&kernel_threads_to_run
, num_online_cpus());
528 for_each_online_cpu(cpu
) {
529 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
530 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
532 /* Wait for all threads to run */
533 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
535 /* Our work is done */
536 trace_lttng_statedump_end(session
);
541 * Called with session mutex held.
543 int lttng_statedump_start(struct lttng_session
*session
)
545 return do_lttng_statedump(session
);
547 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
550 int __init
lttng_statedump_init(void)
553 * Allow module to load even if the fixup cannot be done. This
554 * will allow seemless transition when the underlying issue fix
555 * is merged into the Linux kernel, and when tracepoint.c
556 * "tracepoint_module_notify" is turned into a static function.
558 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
562 module_init(lttng_statedump_init
);
565 void __exit
lttng_statedump_exit(void)
569 module_exit(lttng_statedump_exit
);
571 MODULE_LICENSE("GPL and additional rights");
572 MODULE_AUTHOR("Jean-Hugues Deschenes");
573 MODULE_DESCRIPTION("LTTng statedump provider");
574 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
575 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
576 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
577 LTTNG_MODULES_EXTRAVERSION
);