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 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_process_pid_ns
);
65 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
66 DEFINE_TRACE(lttng_statedump_process_cgroup_ns
);
68 DEFINE_TRACE(lttng_statedump_process_ipc_ns
);
69 #ifndef LTTNG_MNT_NS_MISSING_HEADER
70 DEFINE_TRACE(lttng_statedump_process_mnt_ns
);
72 DEFINE_TRACE(lttng_statedump_process_net_ns
);
73 DEFINE_TRACE(lttng_statedump_process_user_ns
);
74 DEFINE_TRACE(lttng_statedump_process_uts_ns
);
75 DEFINE_TRACE(lttng_statedump_network_interface
);
76 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
77 DEFINE_TRACE(lttng_statedump_cpu_topology
);
82 struct lttng_session
*session
;
83 struct files_struct
*files
;
87 * Protected by the trace lock.
89 static struct delayed_work cpu_work
[NR_CPUS
];
90 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
91 static atomic_t kernel_threads_to_run
;
93 enum lttng_thread_type
{
94 LTTNG_USER_THREAD
= 0,
95 LTTNG_KERNEL_THREAD
= 1,
98 enum lttng_execution_mode
{
104 LTTNG_MODE_UNKNOWN
= 5,
107 enum lttng_execution_submode
{
112 enum lttng_process_status
{
124 int lttng_enumerate_block_devices(struct lttng_session
*session
)
126 struct class *ptr_block_class
;
127 struct device_type
*ptr_disk_type
;
128 struct class_dev_iter iter
;
131 ptr_block_class
= wrapper_get_block_class();
132 if (!ptr_block_class
)
134 ptr_disk_type
= wrapper_get_disk_type();
135 if (!ptr_disk_type
) {
138 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
139 while ((dev
= class_dev_iter_next(&iter
))) {
140 struct disk_part_iter piter
;
141 struct gendisk
*disk
= dev_to_disk(dev
);
142 struct hd_struct
*part
;
145 * Don't show empty devices or things that have been
148 if (get_capacity(disk
) == 0 ||
149 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
152 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
153 while ((part
= disk_part_iter_next(&piter
))) {
154 struct block_device bdev
;
155 char name_buf
[BDEVNAME_SIZE
];
159 * Create a partial 'struct blockdevice' to use
160 * 'bdevname()' which is a simple wrapper over
161 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
166 p
= bdevname(&bdev
, name_buf
);
168 disk_part_iter_exit(&piter
);
169 class_dev_iter_exit(&iter
);
172 trace_lttng_statedump_block_device(session
,
173 part_devt(part
), name_buf
);
175 disk_part_iter_exit(&piter
);
177 class_dev_iter_exit(&iter
);
184 void lttng_enumerate_device(struct lttng_session
*session
,
185 struct net_device
*dev
)
187 struct in_device
*in_dev
;
188 struct in_ifaddr
*ifa
;
190 if (dev
->flags
& IFF_UP
) {
191 in_dev
= in_dev_get(dev
);
193 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
194 ifa
= ifa
->ifa_next
) {
195 trace_lttng_statedump_network_interface(
201 trace_lttng_statedump_network_interface(
207 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
209 struct net_device
*dev
;
211 read_lock(&dev_base_lock
);
212 for_each_netdev(&init_net
, dev
)
213 lttng_enumerate_device(session
, dev
);
214 read_unlock(&dev_base_lock
);
218 #else /* CONFIG_INET */
220 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
224 #endif /* CONFIG_INET */
227 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
229 const struct lttng_fd_ctx
*ctx
= p
;
230 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
231 unsigned int flags
= file
->f_flags
;
235 * We don't expose kernel internal flags, only userspace-visible
238 flags
&= ~FMODE_NONOTIFY
;
239 fdt
= files_fdtable(ctx
->files
);
241 * We need to check here again whether fd is within the fdt
242 * max_fds range, because we might be seeing a different
243 * files_fdtable() than iterate_fd(), assuming only RCU is
244 * protecting the read. In reality, iterate_fd() holds
245 * file_lock, which should ensure the fdt does not change while
246 * the lock is taken, but we are not aware whether this is
247 * guaranteed or not, so play safe.
249 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
252 struct dentry
*dentry
= file
->f_path
.dentry
;
254 /* Make sure we give at least some info */
255 spin_lock(&dentry
->d_lock
);
256 trace_lttng_statedump_file_descriptor(ctx
->session
,
257 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
259 spin_unlock(&dentry
->d_lock
);
262 trace_lttng_statedump_file_descriptor(ctx
->session
,
263 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
268 /* Called with task lock held. */
270 void lttng_enumerate_files(struct lttng_session
*session
,
271 struct files_struct
*files
,
274 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
276 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
279 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
281 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
284 const cpumask_t
*cpumask
= cpu_possible_mask
;
286 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
287 cpu
= cpumask_next(cpu
, cpumask
)) {
288 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
295 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
303 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
304 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
305 * iteration, but it is not exported to modules.
308 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
309 struct task_struct
*p
)
311 struct mm_struct
*mm
;
312 struct vm_area_struct
*map
;
315 /* get_task_mm does a task_lock... */
322 down_read(&mm
->mmap_sem
);
325 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
328 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
331 up_read(&mm
->mmap_sem
);
337 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
339 struct task_struct
*p
;
343 lttng_enumerate_task_vm_maps(session
, p
);
349 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
352 int lttng_list_interrupts(struct lttng_session
*session
)
355 unsigned long flags
= 0;
356 struct irq_desc
*desc
;
358 #define irq_to_desc wrapper_irq_to_desc
360 for_each_irq_desc(irq
, desc
) {
361 struct irqaction
*action
;
362 const char *irq_chip_name
=
363 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
365 local_irq_save(flags
);
366 raw_spin_lock(&desc
->lock
);
367 for (action
= desc
->action
; action
; action
= action
->next
) {
368 trace_lttng_statedump_interrupt(session
,
369 irq
, irq_chip_name
, action
);
371 raw_spin_unlock(&desc
->lock
);
372 local_irq_restore(flags
);
379 int lttng_list_interrupts(struct lttng_session
*session
)
386 * Statedump the task's namespaces using the proc filesystem inode number as
387 * the unique identifier. The user and pid ns are nested and will be dumped
390 * Called with task lock held.
393 void lttng_statedump_process_ns(struct lttng_session
*session
,
394 struct task_struct
*p
,
395 enum lttng_thread_type type
,
396 enum lttng_execution_mode mode
,
397 enum lttng_execution_submode submode
,
398 enum lttng_process_status status
)
400 struct nsproxy
*proxy
;
401 struct pid_namespace
*pid_ns
;
402 struct user_namespace
*user_ns
;
405 * The pid and user namespaces are special, they are nested and
406 * accessed with specific functions instead of the nsproxy struct
407 * like the other namespaces.
409 pid_ns
= task_active_pid_ns(p
);
411 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
412 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
416 user_ns
= task_cred_xxx(p
, user_ns
);
418 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
420 * trace_lttng_statedump_process_user_ns() internally
421 * checks whether user_ns is NULL. While this does not
422 * appear to be a possible return value for
423 * task_cred_xxx(), err on the safe side and check
424 * for NULL here as well to be consistent with the
425 * paranoid behavior of
426 * trace_lttng_statedump_process_user_ns().
428 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
432 * Back and forth on locking strategy within Linux upstream for nsproxy.
433 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
434 * "namespaces: Use task_lock and not rcu to protect nsproxy"
437 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
438 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
439 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
440 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
444 proxy
= task_nsproxy(p
);
447 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
448 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
450 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
451 #ifndef LTTNG_MNT_NS_MISSING_HEADER
452 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
454 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
455 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
457 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
458 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
459 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
460 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
468 int lttng_enumerate_process_states(struct lttng_session
*session
)
470 struct task_struct
*g
, *p
;
473 tmp
= (char *) __get_free_page(GFP_KERNEL
);
478 for_each_process(g
) {
479 struct files_struct
*prev_files
= NULL
;
483 enum lttng_execution_mode mode
=
485 enum lttng_execution_submode submode
=
487 enum lttng_process_status status
;
488 enum lttng_thread_type type
;
489 struct files_struct
*files
;
492 if (p
->exit_state
== EXIT_ZOMBIE
)
493 status
= LTTNG_ZOMBIE
;
494 else if (p
->exit_state
== EXIT_DEAD
)
496 else if (p
->state
== TASK_RUNNING
) {
497 /* Is this a forked child that has not run yet? */
498 if (list_empty(&p
->rt
.run_list
))
499 status
= LTTNG_WAIT_FORK
;
502 * All tasks are considered as wait_cpu;
503 * the viewer will sort out if the task
504 * was really running at this time.
506 status
= LTTNG_WAIT_CPU
;
507 } else if (p
->state
&
508 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
509 /* Task is waiting for something to complete */
512 status
= LTTNG_UNNAMED
;
513 submode
= LTTNG_NONE
;
516 * Verification of t->mm is to filter out kernel
517 * threads; Viewer will further filter out if a
518 * user-space thread was in syscall mode or not.
521 type
= LTTNG_USER_THREAD
;
523 type
= LTTNG_KERNEL_THREAD
;
526 trace_lttng_statedump_process_state(session
,
527 p
, type
, mode
, submode
, status
, files
);
528 lttng_statedump_process_ns(session
,
529 p
, type
, mode
, submode
, status
);
531 * As an optimisation for the common case, do not
532 * repeat information for the same files_struct in
533 * two consecutive threads. This is the common case
534 * for threads sharing the same fd table. RCU guarantees
535 * that the same files_struct pointer is not re-used
536 * throughout processes/threads iteration.
538 if (files
&& files
!= prev_files
) {
539 lttng_enumerate_files(session
, files
, tmp
);
543 } while_each_thread(g
, p
);
547 free_page((unsigned long) tmp
);
553 void lttng_statedump_work_func(struct work_struct
*work
)
555 if (atomic_dec_and_test(&kernel_threads_to_run
))
556 /* If we are the last thread, wake up do_lttng_statedump */
557 wake_up(&statedump_wq
);
561 int do_lttng_statedump(struct lttng_session
*session
)
565 trace_lttng_statedump_start(session
);
566 ret
= lttng_enumerate_process_states(session
);
571 * ret = lttng_enumerate_vm_maps(session);
575 ret
= lttng_list_interrupts(session
);
578 ret
= lttng_enumerate_network_ip_interface(session
);
581 ret
= lttng_enumerate_block_devices(session
);
586 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
591 ret
= lttng_enumerate_cpu_topology(session
);
595 /* TODO lttng_dump_idt_table(session); */
596 /* TODO lttng_dump_softirq_vec(session); */
597 /* TODO lttng_list_modules(session); */
598 /* TODO lttng_dump_swap_files(session); */
601 * Fire off a work queue on each CPU. Their sole purpose in life
602 * is to guarantee that each CPU has been in a state where is was in
603 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
606 atomic_set(&kernel_threads_to_run
, num_online_cpus());
607 for_each_online_cpu(cpu
) {
608 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
609 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
611 /* Wait for all threads to run */
612 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
614 /* Our work is done */
615 trace_lttng_statedump_end(session
);
620 * Called with session mutex held.
622 int lttng_statedump_start(struct lttng_session
*session
)
624 return do_lttng_statedump(session
);
626 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
629 int __init
lttng_statedump_init(void)
632 * Allow module to load even if the fixup cannot be done. This
633 * will allow seemless transition when the underlying issue fix
634 * is merged into the Linux kernel, and when tracepoint.c
635 * "tracepoint_module_notify" is turned into a static function.
637 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
641 module_init(lttng_statedump_init
);
644 void __exit
lttng_statedump_exit(void)
648 module_exit(lttng_statedump_exit
);
650 MODULE_LICENSE("GPL and additional rights");
651 MODULE_AUTHOR("Jean-Hugues Deschenes");
652 MODULE_DESCRIPTION("LTTng statedump provider");
653 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
654 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
655 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
656 LTTNG_MODULES_EXTRAVERSION
);