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>
42 #include <wrapper/file.h>
44 /* Define the tracepoints, but do not build the probes */
45 #define CREATE_TRACE_POINTS
46 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
47 #define TRACE_INCLUDE_FILE lttng-statedump
48 #define LTTNG_INSTRUMENTATION
49 #include <instrumentation/events/lttng-module/lttng-statedump.h>
51 DEFINE_TRACE(lttng_statedump_block_device
);
52 DEFINE_TRACE(lttng_statedump_end
);
53 DEFINE_TRACE(lttng_statedump_interrupt
);
54 DEFINE_TRACE(lttng_statedump_file_descriptor
);
55 DEFINE_TRACE(lttng_statedump_start
);
56 DEFINE_TRACE(lttng_statedump_process_state
);
57 DEFINE_TRACE(lttng_statedump_process_pid_ns
);
58 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
59 DEFINE_TRACE(lttng_statedump_process_cgroup_ns
);
61 DEFINE_TRACE(lttng_statedump_process_ipc_ns
);
62 #ifndef LTTNG_MNT_NS_MISSING_HEADER
63 DEFINE_TRACE(lttng_statedump_process_mnt_ns
);
65 DEFINE_TRACE(lttng_statedump_process_net_ns
);
66 DEFINE_TRACE(lttng_statedump_process_user_ns
);
67 DEFINE_TRACE(lttng_statedump_process_uts_ns
);
68 DEFINE_TRACE(lttng_statedump_network_interface
);
69 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
70 DEFINE_TRACE(lttng_statedump_cpu_topology
);
75 struct lttng_session
*session
;
76 struct files_struct
*files
;
80 * Protected by the trace lock.
82 static struct delayed_work cpu_work
[NR_CPUS
];
83 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
84 static atomic_t kernel_threads_to_run
;
86 enum lttng_thread_type
{
87 LTTNG_USER_THREAD
= 0,
88 LTTNG_KERNEL_THREAD
= 1,
91 enum lttng_execution_mode
{
97 LTTNG_MODE_UNKNOWN
= 5,
100 enum lttng_execution_submode
{
105 enum lttng_process_status
{
117 int lttng_enumerate_block_devices(struct lttng_session
*session
)
119 struct class *ptr_block_class
;
120 struct device_type
*ptr_disk_type
;
121 struct class_dev_iter iter
;
124 ptr_block_class
= wrapper_get_block_class();
125 if (!ptr_block_class
)
127 ptr_disk_type
= wrapper_get_disk_type();
128 if (!ptr_disk_type
) {
131 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
132 while ((dev
= class_dev_iter_next(&iter
))) {
133 struct disk_part_iter piter
;
134 struct gendisk
*disk
= dev_to_disk(dev
);
135 struct hd_struct
*part
;
138 * Don't show empty devices or things that have been
141 if (get_capacity(disk
) == 0 ||
142 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
145 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
146 while ((part
= disk_part_iter_next(&piter
))) {
147 char name_buf
[BDEVNAME_SIZE
];
150 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
152 disk_part_iter_exit(&piter
);
153 class_dev_iter_exit(&iter
);
156 trace_lttng_statedump_block_device(session
,
157 part_devt(part
), name_buf
);
159 disk_part_iter_exit(&piter
);
161 class_dev_iter_exit(&iter
);
168 void lttng_enumerate_device(struct lttng_session
*session
,
169 struct net_device
*dev
)
171 struct in_device
*in_dev
;
172 struct in_ifaddr
*ifa
;
174 if (dev
->flags
& IFF_UP
) {
175 in_dev
= in_dev_get(dev
);
177 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
178 ifa
= ifa
->ifa_next
) {
179 trace_lttng_statedump_network_interface(
185 trace_lttng_statedump_network_interface(
191 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
193 struct net_device
*dev
;
195 read_lock(&dev_base_lock
);
196 for_each_netdev(&init_net
, dev
)
197 lttng_enumerate_device(session
, dev
);
198 read_unlock(&dev_base_lock
);
202 #else /* CONFIG_INET */
204 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
208 #endif /* CONFIG_INET */
211 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
213 const struct lttng_fd_ctx
*ctx
= p
;
214 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
215 unsigned int flags
= file
->f_flags
;
219 * We don't expose kernel internal flags, only userspace-visible
222 flags
&= ~FMODE_NONOTIFY
;
223 fdt
= files_fdtable(ctx
->files
);
225 * We need to check here again whether fd is within the fdt
226 * max_fds range, because we might be seeing a different
227 * files_fdtable() than iterate_fd(), assuming only RCU is
228 * protecting the read. In reality, iterate_fd() holds
229 * file_lock, which should ensure the fdt does not change while
230 * the lock is taken, but we are not aware whether this is
231 * guaranteed or not, so play safe.
233 if (fd
< fdt
->max_fds
&& close_on_exec(fd
, fdt
))
236 struct dentry
*dentry
= file
->f_path
.dentry
;
238 /* Make sure we give at least some info */
239 spin_lock(&dentry
->d_lock
);
240 trace_lttng_statedump_file_descriptor(ctx
->session
,
241 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
243 spin_unlock(&dentry
->d_lock
);
246 trace_lttng_statedump_file_descriptor(ctx
->session
,
247 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
252 /* Called with task lock held. */
254 void lttng_enumerate_files(struct lttng_session
*session
,
255 struct files_struct
*files
,
258 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
260 iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
263 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
265 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
268 const cpumask_t
*cpumask
= cpu_possible_mask
;
270 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
271 cpu
= cpumask_next(cpu
, cpumask
)) {
272 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
279 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
287 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
288 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
289 * iteration, but it is not exported to modules.
292 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
293 struct task_struct
*p
)
295 struct mm_struct
*mm
;
296 struct vm_area_struct
*map
;
299 /* get_task_mm does a task_lock... */
306 down_read(&mm
->mmap_sem
);
309 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
312 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
315 up_read(&mm
->mmap_sem
);
321 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
323 struct task_struct
*p
;
327 lttng_enumerate_task_vm_maps(session
, p
);
334 int lttng_list_interrupts(struct lttng_session
*session
)
337 unsigned long flags
= 0;
338 struct irq_desc
*desc
;
341 for_each_irq_desc(irq
, desc
) {
342 struct irqaction
*action
;
343 const char *irq_chip_name
=
344 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
346 local_irq_save(flags
);
347 raw_spin_lock(&desc
->lock
);
348 for (action
= desc
->action
; action
; action
= action
->next
) {
349 trace_lttng_statedump_interrupt(session
,
350 irq
, irq_chip_name
, action
);
352 raw_spin_unlock(&desc
->lock
);
353 local_irq_restore(flags
);
359 * Statedump the task's namespaces using the proc filesystem inode number as
360 * the unique identifier. The user and pid ns are nested and will be dumped
363 * Called with task lock held.
366 void lttng_statedump_process_ns(struct lttng_session
*session
,
367 struct task_struct
*p
,
368 enum lttng_thread_type type
,
369 enum lttng_execution_mode mode
,
370 enum lttng_execution_submode submode
,
371 enum lttng_process_status status
)
373 struct nsproxy
*proxy
;
374 struct pid_namespace
*pid_ns
;
375 struct user_namespace
*user_ns
;
378 * The pid and user namespaces are special, they are nested and
379 * accessed with specific functions instead of the nsproxy struct
380 * like the other namespaces.
382 pid_ns
= task_active_pid_ns(p
);
384 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
385 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
389 user_ns
= task_cred_xxx(p
, user_ns
);
391 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
393 * trace_lttng_statedump_process_user_ns() internally
394 * checks whether user_ns is NULL. While this does not
395 * appear to be a possible return value for
396 * task_cred_xxx(), err on the safe side and check
397 * for NULL here as well to be consistent with the
398 * paranoid behavior of
399 * trace_lttng_statedump_process_user_ns().
401 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
405 * Back and forth on locking strategy within Linux upstream for nsproxy.
406 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
407 * "namespaces: Use task_lock and not rcu to protect nsproxy"
410 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
411 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
412 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
413 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
417 proxy
= task_nsproxy(p
);
420 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
421 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
423 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
424 #ifndef LTTNG_MNT_NS_MISSING_HEADER
425 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
427 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
428 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
430 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
431 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
432 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
433 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
441 int lttng_enumerate_process_states(struct lttng_session
*session
)
443 struct task_struct
*g
, *p
;
446 tmp
= (char *) __get_free_page(GFP_KERNEL
);
451 for_each_process(g
) {
452 struct files_struct
*prev_files
= NULL
;
456 enum lttng_execution_mode mode
=
458 enum lttng_execution_submode submode
=
460 enum lttng_process_status status
;
461 enum lttng_thread_type type
;
462 struct files_struct
*files
;
465 if (p
->exit_state
== EXIT_ZOMBIE
)
466 status
= LTTNG_ZOMBIE
;
467 else if (p
->exit_state
== EXIT_DEAD
)
469 else if (p
->state
== TASK_RUNNING
) {
470 /* Is this a forked child that has not run yet? */
471 if (list_empty(&p
->rt
.run_list
))
472 status
= LTTNG_WAIT_FORK
;
475 * All tasks are considered as wait_cpu;
476 * the viewer will sort out if the task
477 * was really running at this time.
479 status
= LTTNG_WAIT_CPU
;
480 } else if (p
->state
&
481 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
482 /* Task is waiting for something to complete */
485 status
= LTTNG_UNNAMED
;
486 submode
= LTTNG_NONE
;
489 * Verification of t->mm is to filter out kernel
490 * threads; Viewer will further filter out if a
491 * user-space thread was in syscall mode or not.
494 type
= LTTNG_USER_THREAD
;
496 type
= LTTNG_KERNEL_THREAD
;
499 trace_lttng_statedump_process_state(session
,
500 p
, type
, mode
, submode
, status
, files
);
501 lttng_statedump_process_ns(session
,
502 p
, type
, mode
, submode
, status
);
504 * As an optimisation for the common case, do not
505 * repeat information for the same files_struct in
506 * two consecutive threads. This is the common case
507 * for threads sharing the same fd table. RCU guarantees
508 * that the same files_struct pointer is not re-used
509 * throughout processes/threads iteration.
511 if (files
&& files
!= prev_files
) {
512 lttng_enumerate_files(session
, files
, tmp
);
516 } while_each_thread(g
, p
);
520 free_page((unsigned long) tmp
);
526 void lttng_statedump_work_func(struct work_struct
*work
)
528 if (atomic_dec_and_test(&kernel_threads_to_run
))
529 /* If we are the last thread, wake up do_lttng_statedump */
530 wake_up(&statedump_wq
);
534 int do_lttng_statedump(struct lttng_session
*session
)
538 trace_lttng_statedump_start(session
);
539 ret
= lttng_enumerate_process_states(session
);
544 * ret = lttng_enumerate_vm_maps(session);
548 ret
= lttng_list_interrupts(session
);
551 ret
= lttng_enumerate_network_ip_interface(session
);
554 ret
= lttng_enumerate_block_devices(session
);
559 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
564 ret
= lttng_enumerate_cpu_topology(session
);
568 /* TODO lttng_dump_idt_table(session); */
569 /* TODO lttng_dump_softirq_vec(session); */
570 /* TODO lttng_list_modules(session); */
571 /* TODO lttng_dump_swap_files(session); */
574 * Fire off a work queue on each CPU. Their sole purpose in life
575 * is to guarantee that each CPU has been in a state where is was in
576 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
579 atomic_set(&kernel_threads_to_run
, num_online_cpus());
580 for_each_online_cpu(cpu
) {
581 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
582 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
584 /* Wait for all threads to run */
585 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
587 /* Our work is done */
588 trace_lttng_statedump_end(session
);
593 * Called with session mutex held.
595 int lttng_statedump_start(struct lttng_session
*session
)
597 return do_lttng_statedump(session
);
599 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
602 int __init
lttng_statedump_init(void)
607 module_init(lttng_statedump_init
);
610 void __exit
lttng_statedump_exit(void)
614 module_exit(lttng_statedump_exit
);
616 MODULE_LICENSE("GPL and additional rights");
617 MODULE_AUTHOR("Jean-Hugues Deschenes");
618 MODULE_DESCRIPTION("LTTng statedump provider");
619 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
620 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
621 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
622 LTTNG_MODULES_EXTRAVERSION
);