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 (LINUX_VERSION_CODE >= 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_network_interface
,
120 TP_PROTO(struct lttng_session
*session
,
121 struct net_device
*dev
, struct in_ifaddr
*ifa
),
122 TP_ARGS(session
, dev
, ifa
));
124 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
125 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology
,
126 TP_PROTO(struct lttng_session
*session
, struct cpuinfo_x86
*c
),
127 TP_ARGS(session
, c
));
130 struct lttng_fd_ctx
{
132 struct lttng_session
*session
;
133 struct files_struct
*files
;
137 * Protected by the trace lock.
139 static struct delayed_work cpu_work
[NR_CPUS
];
140 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
141 static atomic_t kernel_threads_to_run
;
143 enum lttng_thread_type
{
144 LTTNG_USER_THREAD
= 0,
145 LTTNG_KERNEL_THREAD
= 1,
148 enum lttng_execution_mode
{
154 LTTNG_MODE_UNKNOWN
= 5,
157 enum lttng_execution_submode
{
162 enum lttng_process_status
{
174 int lttng_enumerate_block_devices(struct lttng_session
*session
)
176 struct class *ptr_block_class
;
177 struct device_type
*ptr_disk_type
;
178 struct class_dev_iter iter
;
181 ptr_block_class
= wrapper_get_block_class();
182 if (!ptr_block_class
)
184 ptr_disk_type
= wrapper_get_disk_type();
185 if (!ptr_disk_type
) {
188 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
189 while ((dev
= class_dev_iter_next(&iter
))) {
190 struct disk_part_iter piter
;
191 struct gendisk
*disk
= dev_to_disk(dev
);
192 struct hd_struct
*part
;
195 * Don't show empty devices or things that have been
198 if (get_capacity(disk
) == 0 ||
199 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
202 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
203 while ((part
= disk_part_iter_next(&piter
))) {
204 struct block_device bdev
;
205 char name_buf
[BDEVNAME_SIZE
];
209 * Create a partial 'struct blockdevice' to use
210 * 'bdevname()' which is a simple wrapper over
211 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
216 p
= bdevname(&bdev
, name_buf
);
218 disk_part_iter_exit(&piter
);
219 class_dev_iter_exit(&iter
);
222 trace_lttng_statedump_block_device(session
,
223 part_devt(part
), name_buf
);
225 disk_part_iter_exit(&piter
);
227 class_dev_iter_exit(&iter
);
234 void lttng_enumerate_device(struct lttng_session
*session
,
235 struct net_device
*dev
)
237 struct in_device
*in_dev
;
238 struct in_ifaddr
*ifa
;
240 if (dev
->flags
& IFF_UP
) {
241 in_dev
= in_dev_get(dev
);
243 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
244 ifa
= ifa
->ifa_next
) {
245 trace_lttng_statedump_network_interface(
251 trace_lttng_statedump_network_interface(
257 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
259 struct net_device
*dev
;
261 read_lock(&dev_base_lock
);
262 for_each_netdev(&init_net
, dev
)
263 lttng_enumerate_device(session
, dev
);
264 read_unlock(&dev_base_lock
);
268 #else /* CONFIG_INET */
270 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
274 #endif /* CONFIG_INET */
277 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
279 const struct lttng_fd_ctx
*ctx
= p
;
280 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
281 unsigned int flags
= file
->f_flags
;
285 * We don't expose kernel internal flags, only userspace-visible
288 flags
&= ~FMODE_NONOTIFY
;
289 fdt
= files_fdtable(ctx
->files
);
291 * We need to check here again whether fd is within the fdt
292 * max_fds range, because we might be seeing a different
293 * files_fdtable() than iterate_fd(), assuming only RCU is
294 * protecting the read. In reality, iterate_fd() holds
295 * file_lock, which should ensure the fdt does not change while
296 * the lock is taken, but we are not aware whether this is
297 * guaranteed or not, so play safe.
299 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
302 struct dentry
*dentry
= file
->f_path
.dentry
;
304 /* Make sure we give at least some info */
305 spin_lock(&dentry
->d_lock
);
306 trace_lttng_statedump_file_descriptor(ctx
->session
,
307 ctx
->files
, fd
, dentry
->d_name
.name
, flags
,
309 spin_unlock(&dentry
->d_lock
);
312 trace_lttng_statedump_file_descriptor(ctx
->session
,
313 ctx
->files
, fd
, s
, flags
, file
->f_mode
);
318 /* Called with task lock held. */
320 void lttng_enumerate_files(struct lttng_session
*session
,
321 struct files_struct
*files
,
324 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .files
= files
, };
326 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
329 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
331 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
334 const cpumask_t
*cpumask
= cpu_possible_mask
;
336 for (cpu
= cpumask_first(cpumask
); cpu
< nr_cpu_ids
;
337 cpu
= cpumask_next(cpu
, cpumask
)) {
338 trace_lttng_statedump_cpu_topology(session
, &cpu_data(cpu
));
345 int lttng_enumerate_cpu_topology(struct lttng_session
*session
)
353 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
354 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
355 * iteration, but it is not exported to modules.
358 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
359 struct task_struct
*p
)
361 struct mm_struct
*mm
;
362 struct vm_area_struct
*map
;
365 /* get_task_mm does a task_lock... */
372 down_read(&mm
->mmap_sem
);
375 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
378 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
381 up_read(&mm
->mmap_sem
);
387 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
389 struct task_struct
*p
;
393 lttng_enumerate_task_vm_maps(session
, p
);
399 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
402 int lttng_list_interrupts(struct lttng_session
*session
)
405 unsigned long flags
= 0;
406 struct irq_desc
*desc
;
408 #define irq_to_desc wrapper_irq_to_desc
410 for_each_irq_desc(irq
, desc
) {
411 struct irqaction
*action
;
412 const char *irq_chip_name
=
413 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
415 local_irq_save(flags
);
416 raw_spin_lock(&desc
->lock
);
417 for (action
= desc
->action
; action
; action
= action
->next
) {
418 trace_lttng_statedump_interrupt(session
,
419 irq
, irq_chip_name
, action
);
421 raw_spin_unlock(&desc
->lock
);
422 local_irq_restore(flags
);
429 int lttng_list_interrupts(struct lttng_session
*session
)
436 * Statedump the task's namespaces using the proc filesystem inode number as
437 * the unique identifier. The user and pid ns are nested and will be dumped
440 * Called with task lock held.
443 void lttng_statedump_process_ns(struct lttng_session
*session
,
444 struct task_struct
*p
,
445 enum lttng_thread_type type
,
446 enum lttng_execution_mode mode
,
447 enum lttng_execution_submode submode
,
448 enum lttng_process_status status
)
450 struct nsproxy
*proxy
;
451 struct pid_namespace
*pid_ns
;
452 struct user_namespace
*user_ns
;
455 * The pid and user namespaces are special, they are nested and
456 * accessed with specific functions instead of the nsproxy struct
457 * like the other namespaces.
459 pid_ns
= task_active_pid_ns(p
);
461 trace_lttng_statedump_process_pid_ns(session
, p
, pid_ns
);
462 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
466 user_ns
= task_cred_xxx(p
, user_ns
);
468 trace_lttng_statedump_process_user_ns(session
, p
, user_ns
);
470 * trace_lttng_statedump_process_user_ns() internally
471 * checks whether user_ns is NULL. While this does not
472 * appear to be a possible return value for
473 * task_cred_xxx(), err on the safe side and check
474 * for NULL here as well to be consistent with the
475 * paranoid behavior of
476 * trace_lttng_statedump_process_user_ns().
478 user_ns
= user_ns
? user_ns
->lttng_user_ns_parent
: NULL
;
482 * Back and forth on locking strategy within Linux upstream for nsproxy.
483 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
484 * "namespaces: Use task_lock and not rcu to protect nsproxy"
487 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
488 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
489 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
490 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
494 proxy
= task_nsproxy(p
);
497 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
498 trace_lttng_statedump_process_cgroup_ns(session
, p
, proxy
->cgroup_ns
);
500 trace_lttng_statedump_process_ipc_ns(session
, p
, proxy
->ipc_ns
);
501 #ifndef LTTNG_MNT_NS_MISSING_HEADER
502 trace_lttng_statedump_process_mnt_ns(session
, p
, proxy
->mnt_ns
);
504 trace_lttng_statedump_process_net_ns(session
, p
, proxy
->net_ns
);
505 trace_lttng_statedump_process_uts_ns(session
, p
, proxy
->uts_ns
);
507 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
508 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
509 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
510 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
518 int lttng_enumerate_process_states(struct lttng_session
*session
)
520 struct task_struct
*g
, *p
;
523 tmp
= (char *) __get_free_page(GFP_KERNEL
);
528 for_each_process(g
) {
529 struct files_struct
*prev_files
= NULL
;
533 enum lttng_execution_mode mode
=
535 enum lttng_execution_submode submode
=
537 enum lttng_process_status status
;
538 enum lttng_thread_type type
;
539 struct files_struct
*files
;
542 if (p
->exit_state
== EXIT_ZOMBIE
)
543 status
= LTTNG_ZOMBIE
;
544 else if (p
->exit_state
== EXIT_DEAD
)
546 else if (p
->state
== TASK_RUNNING
) {
547 /* Is this a forked child that has not run yet? */
548 if (list_empty(&p
->rt
.run_list
))
549 status
= LTTNG_WAIT_FORK
;
552 * All tasks are considered as wait_cpu;
553 * the viewer will sort out if the task
554 * was really running at this time.
556 status
= LTTNG_WAIT_CPU
;
557 } else if (p
->state
&
558 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
559 /* Task is waiting for something to complete */
562 status
= LTTNG_UNNAMED
;
563 submode
= LTTNG_NONE
;
566 * Verification of t->mm is to filter out kernel
567 * threads; Viewer will further filter out if a
568 * user-space thread was in syscall mode or not.
571 type
= LTTNG_USER_THREAD
;
573 type
= LTTNG_KERNEL_THREAD
;
576 trace_lttng_statedump_process_state(session
,
577 p
, type
, mode
, submode
, status
, files
);
578 lttng_statedump_process_ns(session
,
579 p
, type
, mode
, submode
, status
);
581 * As an optimisation for the common case, do not
582 * repeat information for the same files_struct in
583 * two consecutive threads. This is the common case
584 * for threads sharing the same fd table. RCU guarantees
585 * that the same files_struct pointer is not re-used
586 * throughout processes/threads iteration.
588 if (files
&& files
!= prev_files
) {
589 lttng_enumerate_files(session
, files
, tmp
);
593 } while_each_thread(g
, p
);
597 free_page((unsigned long) tmp
);
603 void lttng_statedump_work_func(struct work_struct
*work
)
605 if (atomic_dec_and_test(&kernel_threads_to_run
))
606 /* If we are the last thread, wake up do_lttng_statedump */
607 wake_up(&statedump_wq
);
611 int do_lttng_statedump(struct lttng_session
*session
)
615 trace_lttng_statedump_start(session
);
616 ret
= lttng_enumerate_process_states(session
);
621 * ret = lttng_enumerate_vm_maps(session);
625 ret
= lttng_list_interrupts(session
);
628 ret
= lttng_enumerate_network_ip_interface(session
);
631 ret
= lttng_enumerate_block_devices(session
);
636 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
641 ret
= lttng_enumerate_cpu_topology(session
);
645 /* TODO lttng_dump_idt_table(session); */
646 /* TODO lttng_dump_softirq_vec(session); */
647 /* TODO lttng_list_modules(session); */
648 /* TODO lttng_dump_swap_files(session); */
651 * Fire off a work queue on each CPU. Their sole purpose in life
652 * is to guarantee that each CPU has been in a state where is was in
653 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
656 atomic_set(&kernel_threads_to_run
, num_online_cpus());
657 for_each_online_cpu(cpu
) {
658 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
659 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
661 /* Wait for all threads to run */
662 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
664 /* Our work is done */
665 trace_lttng_statedump_end(session
);
670 * Called with session mutex held.
672 int lttng_statedump_start(struct lttng_session
*session
)
674 return do_lttng_statedump(session
);
676 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
679 int __init
lttng_statedump_init(void)
682 * Allow module to load even if the fixup cannot be done. This
683 * will allow seemless transition when the underlying issue fix
684 * is merged into the Linux kernel, and when tracepoint.c
685 * "tracepoint_module_notify" is turned into a static function.
687 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
691 module_init(lttng_statedump_init
);
694 void __exit
lttng_statedump_exit(void)
698 module_exit(lttng_statedump_exit
);
700 MODULE_LICENSE("GPL and additional rights");
701 MODULE_AUTHOR("Jean-Hugues Deschenes");
702 MODULE_DESCRIPTION("LTTng statedump provider");
703 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
704 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
705 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
706 LTTNG_MODULES_EXTRAVERSION
);