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/spinlock.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/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_network_interface
,
92 TP_PROTO(struct lttng_session
*session
,
93 struct net_device
*dev
, struct in_ifaddr
*ifa
),
94 TP_ARGS(session
, dev
, ifa
));
98 struct lttng_session
*session
;
99 struct task_struct
*p
;
100 struct files_struct
*files
;
104 * Protected by the trace lock.
106 static struct delayed_work cpu_work
[NR_CPUS
];
107 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
108 static atomic_t kernel_threads_to_run
;
110 enum lttng_thread_type
{
111 LTTNG_USER_THREAD
= 0,
112 LTTNG_KERNEL_THREAD
= 1,
115 enum lttng_execution_mode
{
121 LTTNG_MODE_UNKNOWN
= 5,
124 enum lttng_execution_submode
{
129 enum lttng_process_status
{
141 int lttng_enumerate_block_devices(struct lttng_session
*session
)
143 struct class *ptr_block_class
;
144 struct device_type
*ptr_disk_type
;
145 struct class_dev_iter iter
;
148 ptr_block_class
= wrapper_get_block_class();
149 if (!ptr_block_class
)
151 ptr_disk_type
= wrapper_get_disk_type();
152 if (!ptr_disk_type
) {
155 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
156 while ((dev
= class_dev_iter_next(&iter
))) {
157 struct disk_part_iter piter
;
158 struct gendisk
*disk
= dev_to_disk(dev
);
159 struct hd_struct
*part
;
162 * Don't show empty devices or things that have been
165 if (get_capacity(disk
) == 0 ||
166 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
169 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
170 while ((part
= disk_part_iter_next(&piter
))) {
171 struct block_device bdev
;
172 char name_buf
[BDEVNAME_SIZE
];
176 * Create a partial 'struct blockdevice' to use
177 * 'bdevname()' which is a simple wrapper over
178 * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
183 p
= bdevname(&bdev
, name_buf
);
185 disk_part_iter_exit(&piter
);
186 class_dev_iter_exit(&iter
);
189 trace_lttng_statedump_block_device(session
,
190 part_devt(part
), name_buf
);
192 disk_part_iter_exit(&piter
);
194 class_dev_iter_exit(&iter
);
201 void lttng_enumerate_device(struct lttng_session
*session
,
202 struct net_device
*dev
)
204 struct in_device
*in_dev
;
205 struct in_ifaddr
*ifa
;
207 if (dev
->flags
& IFF_UP
) {
208 in_dev
= in_dev_get(dev
);
210 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
211 ifa
= ifa
->ifa_next
) {
212 trace_lttng_statedump_network_interface(
218 trace_lttng_statedump_network_interface(
224 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
226 struct net_device
*dev
;
228 read_lock(&dev_base_lock
);
229 for_each_netdev(&init_net
, dev
)
230 lttng_enumerate_device(session
, dev
);
231 read_unlock(&dev_base_lock
);
235 #else /* CONFIG_INET */
237 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
241 #endif /* CONFIG_INET */
244 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
246 const struct lttng_fd_ctx
*ctx
= p
;
247 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
248 unsigned int flags
= file
->f_flags
;
252 * We don't expose kernel internal flags, only userspace-visible
255 flags
&= ~FMODE_NONOTIFY
;
256 fdt
= files_fdtable(ctx
->files
);
258 * We need to check here again whether fd is within the fdt
259 * max_fds range, because we might be seeing a different
260 * files_fdtable() than iterate_fd(), assuming only RCU is
261 * protecting the read. In reality, iterate_fd() holds
262 * file_lock, which should ensure the fdt does not change while
263 * the lock is taken, but we are not aware whether this is
264 * guaranteed or not, so play safe.
266 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
269 struct dentry
*dentry
= file
->f_path
.dentry
;
271 /* Make sure we give at least some info */
272 spin_lock(&dentry
->d_lock
);
273 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
,
274 dentry
->d_name
.name
, flags
, file
->f_mode
);
275 spin_unlock(&dentry
->d_lock
);
278 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
, s
,
279 flags
, file
->f_mode
);
285 void lttng_enumerate_task_fd(struct lttng_session
*session
,
286 struct task_struct
*p
, char *tmp
)
288 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .p
= p
};
289 struct files_struct
*files
;
296 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
302 int lttng_enumerate_file_descriptors(struct lttng_session
*session
)
304 struct task_struct
*p
;
307 tmp
= (char *) __get_free_page(GFP_KERNEL
);
311 /* Enumerate active file descriptors */
314 lttng_enumerate_task_fd(session
, p
, tmp
);
316 free_page((unsigned long) tmp
);
322 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
323 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
324 * iteration, but it is not exported to modules.
327 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
328 struct task_struct
*p
)
330 struct mm_struct
*mm
;
331 struct vm_area_struct
*map
;
334 /* get_task_mm does a task_lock... */
341 down_read(&mm
->mmap_sem
);
344 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
347 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
350 up_read(&mm
->mmap_sem
);
356 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
358 struct task_struct
*p
;
362 lttng_enumerate_task_vm_maps(session
, p
);
368 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
370 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
371 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
375 int lttng_list_interrupts(struct lttng_session
*session
)
378 unsigned long flags
= 0;
379 struct irq_desc
*desc
;
381 #define irq_to_desc wrapper_irq_to_desc
383 for_each_irq_desc(irq
, desc
) {
384 struct irqaction
*action
;
385 const char *irq_chip_name
=
386 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
388 local_irq_save(flags
);
389 wrapper_desc_spin_lock(&desc
->lock
);
390 for (action
= desc
->action
; action
; action
= action
->next
) {
391 trace_lttng_statedump_interrupt(session
,
392 irq
, irq_chip_name
, action
);
394 wrapper_desc_spin_unlock(&desc
->lock
);
395 local_irq_restore(flags
);
402 int lttng_list_interrupts(struct lttng_session
*session
)
409 * Called with task lock held.
412 void lttng_statedump_process_ns(struct lttng_session
*session
,
413 struct task_struct
*p
,
414 enum lttng_thread_type type
,
415 enum lttng_execution_mode mode
,
416 enum lttng_execution_submode submode
,
417 enum lttng_process_status status
)
419 struct pid_namespace
*pid_ns
;
421 pid_ns
= task_active_pid_ns(p
);
423 trace_lttng_statedump_process_state(session
,
424 p
, type
, mode
, submode
, status
, pid_ns
);
425 pid_ns
= pid_ns
? pid_ns
->parent
: NULL
;
430 int lttng_enumerate_process_states(struct lttng_session
*session
)
432 struct task_struct
*g
, *p
;
435 for_each_process(g
) {
438 enum lttng_execution_mode mode
=
440 enum lttng_execution_submode submode
=
442 enum lttng_process_status status
;
443 enum lttng_thread_type type
;
446 if (p
->exit_state
== EXIT_ZOMBIE
)
447 status
= LTTNG_ZOMBIE
;
448 else if (p
->exit_state
== EXIT_DEAD
)
450 else if (p
->state
== TASK_RUNNING
) {
451 /* Is this a forked child that has not run yet? */
452 if (list_empty(&p
->rt
.run_list
))
453 status
= LTTNG_WAIT_FORK
;
456 * All tasks are considered as wait_cpu;
457 * the viewer will sort out if the task
458 * was really running at this time.
460 status
= LTTNG_WAIT_CPU
;
461 } else if (p
->state
&
462 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
463 /* Task is waiting for something to complete */
466 status
= LTTNG_UNNAMED
;
467 submode
= LTTNG_NONE
;
470 * Verification of t->mm is to filter out kernel
471 * threads; Viewer will further filter out if a
472 * user-space thread was in syscall mode or not.
475 type
= LTTNG_USER_THREAD
;
477 type
= LTTNG_KERNEL_THREAD
;
478 lttng_statedump_process_ns(session
,
479 p
, type
, mode
, submode
, status
);
481 } while_each_thread(g
, p
);
489 void lttng_statedump_work_func(struct work_struct
*work
)
491 if (atomic_dec_and_test(&kernel_threads_to_run
))
492 /* If we are the last thread, wake up do_lttng_statedump */
493 wake_up(&statedump_wq
);
497 int do_lttng_statedump(struct lttng_session
*session
)
501 trace_lttng_statedump_start(session
);
502 ret
= lttng_enumerate_process_states(session
);
505 ret
= lttng_enumerate_file_descriptors(session
);
510 * ret = lttng_enumerate_vm_maps(session);
514 ret
= lttng_list_interrupts(session
);
517 ret
= lttng_enumerate_network_ip_interface(session
);
520 ret
= lttng_enumerate_block_devices(session
);
525 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
531 /* TODO lttng_dump_idt_table(session); */
532 /* TODO lttng_dump_softirq_vec(session); */
533 /* TODO lttng_list_modules(session); */
534 /* TODO lttng_dump_swap_files(session); */
537 * Fire off a work queue on each CPU. Their sole purpose in life
538 * is to guarantee that each CPU has been in a state where is was in
539 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
542 atomic_set(&kernel_threads_to_run
, num_online_cpus());
543 for_each_online_cpu(cpu
) {
544 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
545 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
547 /* Wait for all threads to run */
548 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
550 /* Our work is done */
551 trace_lttng_statedump_end(session
);
556 * Called with session mutex held.
558 int lttng_statedump_start(struct lttng_session
*session
)
560 return do_lttng_statedump(session
);
562 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
565 int __init
lttng_statedump_init(void)
568 * Allow module to load even if the fixup cannot be done. This
569 * will allow seemless transition when the underlying issue fix
570 * is merged into the Linux kernel, and when tracepoint.c
571 * "tracepoint_module_notify" is turned into a static function.
573 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
577 module_init(lttng_statedump_init
);
580 void __exit
lttng_statedump_exit(void)
584 module_exit(lttng_statedump_exit
);
586 MODULE_LICENSE("GPL and additional rights");
587 MODULE_AUTHOR("Jean-Hugues Deschenes");
588 MODULE_DESCRIPTION("LTTng statedump provider");
589 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
590 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
591 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
592 LTTNG_MODULES_EXTRAVERSION
);