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>
35 #include <linux/fdtable.h>
36 #include <linux/irq.h>
38 #include <lttng-events.h>
39 #include <lttng-tracer.h>
40 #include <wrapper/namespace.h>
41 #include <wrapper/genhd.h>
43 /* Define the tracepoints, but do not build the probes */
44 #define CREATE_TRACE_POINTS
45 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
46 #define TRACE_INCLUDE_FILE lttng-statedump
47 #define LTTNG_INSTRUMENTATION
48 #include <instrumentation/events/lttng-module/lttng-statedump.h>
50 DEFINE_TRACE(lttng_statedump_block_device
);
51 DEFINE_TRACE(lttng_statedump_end
);
52 DEFINE_TRACE(lttng_statedump_interrupt
);
53 DEFINE_TRACE(lttng_statedump_file_descriptor
);
54 DEFINE_TRACE(lttng_statedump_start
);
55 DEFINE_TRACE(lttng_statedump_process_state
);
56 DEFINE_TRACE(lttng_statedump_process_pid_ns
);
57 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
58 DEFINE_TRACE(lttng_statedump_process_cgroup_ns
);
60 DEFINE_TRACE(lttng_statedump_process_ipc_ns
);
61 #ifndef LTTNG_MNT_NS_MISSING_HEADER
62 DEFINE_TRACE(lttng_statedump_process_mnt_ns
);
64 DEFINE_TRACE(lttng_statedump_process_net_ns
);
65 DEFINE_TRACE(lttng_statedump_process_user_ns
);
66 DEFINE_TRACE(lttng_statedump_process_uts_ns
);
67 DEFINE_TRACE(lttng_statedump_network_interface
);
68 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
69 DEFINE_TRACE(lttng_statedump_cpu_topology
);
74 struct lttng_session
*session
;
75 struct files_struct
*files
;
79 * Protected by the trace lock.
81 static struct delayed_work cpu_work
[NR_CPUS
];
82 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
83 static atomic_t kernel_threads_to_run
;
85 enum lttng_thread_type
{
86 LTTNG_USER_THREAD
= 0,
87 LTTNG_KERNEL_THREAD
= 1,
90 enum lttng_execution_mode
{
96 LTTNG_MODE_UNKNOWN
= 5,
99 enum lttng_execution_submode
{
104 enum lttng_process_status
{
116 int lttng_enumerate_block_devices(struct lttng_session
*session
)
118 struct class *ptr_block_class
;
119 struct device_type
*ptr_disk_type
;
120 struct class_dev_iter iter
;
123 ptr_block_class
= wrapper_get_block_class();
124 if (!ptr_block_class
)
126 ptr_disk_type
= wrapper_get_disk_type();
127 if (!ptr_disk_type
) {
130 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
131 while ((dev
= class_dev_iter_next(&iter
))) {
132 struct disk_part_iter piter
;
133 struct gendisk
*disk
= dev_to_disk(dev
);
134 struct hd_struct
*part
;
137 * Don't show empty devices or things that have been
140 if (get_capacity(disk
) == 0 ||
141 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
144 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
145 while ((part
= disk_part_iter_next(&piter
))) {
146 char name_buf
[BDEVNAME_SIZE
];
149 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
151 disk_part_iter_exit(&piter
);
152 class_dev_iter_exit(&iter
);
155 trace_lttng_statedump_block_device(session
,
156 part_devt(part
), name_buf
);
158 disk_part_iter_exit(&piter
);
160 class_dev_iter_exit(&iter
);
167 void lttng_enumerate_device(struct lttng_session
*session
,
168 struct net_device
*dev
)
170 struct in_device
*in_dev
;
171 struct in_ifaddr
*ifa
;
173 if (dev
->flags
& IFF_UP
) {
174 in_dev
= in_dev_get(dev
);
176 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
177 ifa
= ifa
->ifa_next
) {
178 trace_lttng_statedump_network_interface(
184 trace_lttng_statedump_network_interface(
190 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
192 struct net_device
*dev
;
194 read_lock(&dev_base_lock
);
195 for_each_netdev(&init_net
, dev
)
196 lttng_enumerate_device(session
, dev
);
197 read_unlock(&dev_base_lock
);
201 #else /* CONFIG_INET */
203 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
207 #endif /* CONFIG_INET */
210 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
212 const struct lttng_fd_ctx
*ctx
= p
;
213 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
214 unsigned int flags
= file
->f_flags
;
218 * We don't expose kernel internal flags, only userspace-visible
221 flags
&= ~FMODE_NONOTIFY
;
222 fdt
= files_fdtable(ctx
->files
);
224 * We need to check here again whether fd is within the fdt
225 * max_fds range, because we might be seeing a different
226 * files_fdtable() than iterate_fd(), assuming only RCU is
227 * protecting the read. In reality, iterate_fd() holds
228 * file_lock, which should ensure the fdt does not change while
229 * the lock is taken, but we are not aware whether this is
230 * guaranteed or not, so play safe.
232 if (fd
< fdt
->max_fds
&& close_on_exec(fd
, fdt
))
235 struct dentry
*dentry
= file
->f_path
.dentry
;
237 /* Make sure we give at least some info */
238 spin_lock(&dentry
->d_lock
);
239 trace_lttng_statedump_file_descriptor(ctx
->session
,
240 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
242 spin_unlock(&dentry
->d_lock
);
245 trace_lttng_statedump_file_descriptor(ctx
->session
,
246 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
251 /* Called with task lock held. */
253 void lttng_enumerate_files(struct lttng_session
*session
,
254 struct files_struct
*files
,
257 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
259 iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
262 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
264 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
267 const cpumask_t
*cpumask
= cpu_possible_mask
;
269 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
270 cpu
= cpumask_next(cpu
, cpumask
)) {
271 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
278 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
286 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
287 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
288 * iteration, but it is not exported to modules.
291 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
292 struct task_struct
*p
)
294 struct mm_struct
*mm
;
295 struct vm_area_struct
*map
;
298 /* get_task_mm does a task_lock... */
305 down_read(&mm
->mmap_sem
);
308 ino
= map
->vm_file
->f_path
.dentry
->d_inode
->i_ino
;
311 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
314 up_read(&mm
->mmap_sem
);
320 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
322 struct task_struct
*p
;
326 lttng_enumerate_task_vm_maps(session
, p
);
333 int lttng_list_interrupts(struct lttng_session
*session
)
336 unsigned long flags
= 0;
337 struct irq_desc
*desc
;
340 for_each_irq_desc(irq
, desc
) {
341 struct irqaction
*action
;
342 const char *irq_chip_name
=
343 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
345 local_irq_save(flags
);
346 raw_spin_lock(&desc
->lock
);
347 for (action
= desc
->action
; action
; action
= action
->next
) {
348 trace_lttng_statedump_interrupt(session
,
349 irq
, irq_chip_name
, action
);
351 raw_spin_unlock(&desc
->lock
);
352 local_irq_restore(flags
);
358 * Statedump the task's namespaces using the proc filesystem inode number as
359 * the unique identifier. The user and pid ns are nested and will be dumped
362 * Called with task lock held.
365 void lttng_statedump_process_ns(struct lttng_session
*session
,
366 struct task_struct
*p
,
367 enum lttng_thread_type type
,
368 enum lttng_execution_mode mode
,
369 enum lttng_execution_submode submode
,
370 enum lttng_process_status status
)
372 struct nsproxy
*proxy
;
373 struct pid_namespace
*pid_ns
;
374 struct user_namespace
*user_ns
;
377 * The pid and user namespaces are special, they are nested and
378 * accessed with specific functions instead of the nsproxy struct
379 * like the other namespaces.
381 pid_ns
= task_active_pid_ns(p
);
383 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
384 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
388 user_ns
= task_cred_xxx(p
, user_ns
);
390 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
392 * trace_lttng_statedump_process_user_ns() internally
393 * checks whether user_ns is NULL. While this does not
394 * appear to be a possible return value for
395 * task_cred_xxx(), err on the safe side and check
396 * for NULL here as well to be consistent with the
397 * paranoid behavior of
398 * trace_lttng_statedump_process_user_ns().
400 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
404 * Back and forth on locking strategy within Linux upstream for nsproxy.
405 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
406 * "namespaces: Use task_lock and not rcu to protect nsproxy"
409 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
410 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
411 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
412 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
416 proxy
= task_nsproxy(p
);
419 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
420 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
422 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
423 #ifndef LTTNG_MNT_NS_MISSING_HEADER
424 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
426 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
427 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
429 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
430 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
431 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
432 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
440 int lttng_enumerate_process_states(struct lttng_session
*session
)
442 struct task_struct
*g
, *p
;
445 tmp
= (char *) __get_free_page(GFP_KERNEL
);
450 for_each_process(g
) {
451 struct files_struct
*prev_files
= NULL
;
455 enum lttng_execution_mode mode
=
457 enum lttng_execution_submode submode
=
459 enum lttng_process_status status
;
460 enum lttng_thread_type type
;
461 struct files_struct
*files
;
464 if (p
->exit_state
== EXIT_ZOMBIE
)
465 status
= LTTNG_ZOMBIE
;
466 else if (p
->exit_state
== EXIT_DEAD
)
468 else if (p
->state
== TASK_RUNNING
) {
469 /* Is this a forked child that has not run yet? */
470 if (list_empty(&p
->rt
.run_list
))
471 status
= LTTNG_WAIT_FORK
;
474 * All tasks are considered as wait_cpu;
475 * the viewer will sort out if the task
476 * was really running at this time.
478 status
= LTTNG_WAIT_CPU
;
479 } else if (p
->state
&
480 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
481 /* Task is waiting for something to complete */
484 status
= LTTNG_UNNAMED
;
485 submode
= LTTNG_NONE
;
488 * Verification of t->mm is to filter out kernel
489 * threads; Viewer will further filter out if a
490 * user-space thread was in syscall mode or not.
493 type
= LTTNG_USER_THREAD
;
495 type
= LTTNG_KERNEL_THREAD
;
498 trace_lttng_statedump_process_state(session
,
499 p
, type
, mode
, submode
, status
, files
);
500 lttng_statedump_process_ns(session
,
501 p
, type
, mode
, submode
, status
);
503 * As an optimisation for the common case, do not
504 * repeat information for the same files_struct in
505 * two consecutive threads. This is the common case
506 * for threads sharing the same fd table. RCU guarantees
507 * that the same files_struct pointer is not re-used
508 * throughout processes/threads iteration.
510 if (files
&& files
!= prev_files
) {
511 lttng_enumerate_files(session
, files
, tmp
);
515 } while_each_thread(g
, p
);
519 free_page((unsigned long) tmp
);
525 void lttng_statedump_work_func(struct work_struct
*work
)
527 if (atomic_dec_and_test(&kernel_threads_to_run
))
528 /* If we are the last thread, wake up do_lttng_statedump */
529 wake_up(&statedump_wq
);
533 int do_lttng_statedump(struct lttng_session
*session
)
537 trace_lttng_statedump_start(session
);
538 ret
= lttng_enumerate_process_states(session
);
543 * ret = lttng_enumerate_vm_maps(session);
547 ret
= lttng_list_interrupts(session
);
550 ret
= lttng_enumerate_network_ip_interface(session
);
553 ret
= lttng_enumerate_block_devices(session
);
558 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
563 ret
= lttng_enumerate_cpu_topology(session
);
567 /* TODO lttng_dump_idt_table(session); */
568 /* TODO lttng_dump_softirq_vec(session); */
569 /* TODO lttng_list_modules(session); */
570 /* TODO lttng_dump_swap_files(session); */
573 * Fire off a work queue on each CPU. Their sole purpose in life
574 * is to guarantee that each CPU has been in a state where is was in
575 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
578 atomic_set(&kernel_threads_to_run
, num_online_cpus());
579 for_each_online_cpu(cpu
) {
580 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
581 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
583 /* Wait for all threads to run */
584 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
586 /* Our work is done */
587 trace_lttng_statedump_end(session
);
592 * Called with session mutex held.
594 int lttng_statedump_start(struct lttng_session
*session
)
596 return do_lttng_statedump(session
);
598 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
601 int __init
lttng_statedump_init(void)
606 module_init(lttng_statedump_init
);
609 void __exit
lttng_statedump_exit(void)
613 module_exit(lttng_statedump_exit
);
615 MODULE_LICENSE("GPL and additional rights");
616 MODULE_AUTHOR("Jean-Hugues Deschenes");
617 MODULE_DESCRIPTION("LTTng statedump provider");
618 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
619 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
620 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
621 LTTNG_MODULES_EXTRAVERSION
);