4 * Linux Trace Toolkit Next Generation Kernel State Dump
6 * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
7 * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; only
12 * version 2.1 of the License.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * Eric Clement: Add listing of network IP interface
25 * 2006, 2007 Mathieu Desnoyers Fix kernel threads
29 #include <linux/init.h>
30 #include <linux/module.h>
31 #include <linux/netlink.h>
32 #include <linux/inet.h>
34 #include <linux/kthread.h>
35 #include <linux/proc_fs.h>
36 #include <linux/file.h>
37 #include <linux/interrupt.h>
38 #include <linux/irqnr.h>
39 #include <linux/cpu.h>
40 #include <linux/netdevice.h>
41 #include <linux/inetdevice.h>
42 #include <linux/sched.h>
44 #include <linux/fdtable.h>
45 #include <linux/swap.h>
46 #include <linux/wait.h>
47 #include <linux/mutex.h>
48 #include <linux/device.h>
50 #include "lttng-events.h"
51 #include "lttng-tracer.h"
52 #include "wrapper/irqdesc.h"
53 #include "wrapper/spinlock.h"
54 #include "wrapper/fdtable.h"
55 #include "wrapper/nsproxy.h"
56 #include "wrapper/irq.h"
57 #include "wrapper/tracepoint.h"
58 #include "wrapper/genhd.h"
60 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
61 #include <linux/irq.h>
64 /* Define the tracepoints, but do not build the probes */
65 #define CREATE_TRACE_POINTS
66 #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
67 #define TRACE_INCLUDE_FILE lttng-statedump
68 #define LTTNG_INSTRUMENTATION
69 #include "instrumentation/events/lttng-module/lttng-statedump.h"
71 DEFINE_TRACE(lttng_statedump_block_device
);
72 DEFINE_TRACE(lttng_statedump_end
);
73 DEFINE_TRACE(lttng_statedump_interrupt
);
74 DEFINE_TRACE(lttng_statedump_file_descriptor
);
75 DEFINE_TRACE(lttng_statedump_start
);
76 DEFINE_TRACE(lttng_statedump_process_state
);
77 DEFINE_TRACE(lttng_statedump_network_interface
);
81 struct lttng_session
*session
;
82 struct task_struct
*p
;
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 char name_buf
[BDEVNAME_SIZE
];
157 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
159 disk_part_iter_exit(&piter
);
160 class_dev_iter_exit(&iter
);
163 trace_lttng_statedump_block_device(session
,
164 part_devt(part
), name_buf
);
166 disk_part_iter_exit(&piter
);
168 class_dev_iter_exit(&iter
);
175 void lttng_enumerate_device(struct lttng_session
*session
,
176 struct net_device
*dev
)
178 struct in_device
*in_dev
;
179 struct in_ifaddr
*ifa
;
181 if (dev
->flags
& IFF_UP
) {
182 in_dev
= in_dev_get(dev
);
184 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
185 ifa
= ifa
->ifa_next
) {
186 trace_lttng_statedump_network_interface(
192 trace_lttng_statedump_network_interface(
198 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
200 struct net_device
*dev
;
202 read_lock(&dev_base_lock
);
203 for_each_netdev(&init_net
, dev
)
204 lttng_enumerate_device(session
, dev
);
205 read_unlock(&dev_base_lock
);
209 #else /* CONFIG_INET */
211 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
215 #endif /* CONFIG_INET */
218 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
220 const struct lttng_fd_ctx
*ctx
= p
;
221 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
222 unsigned int flags
= file
->f_flags
;
226 * We don't expose kernel internal flags, only userspace-visible
229 flags
&= ~FMODE_NONOTIFY
;
230 fdt
= files_fdtable(ctx
->files
);
232 * We need to check here again whether fd is within the fdt
233 * max_fds range, because we might be seeing a different
234 * files_fdtable() than iterate_fd(), assuming only RCU is
235 * protecting the read. In reality, iterate_fd() holds
236 * file_lock, which should ensure the fdt does not change while
237 * the lock is taken, but we are not aware whether this is
238 * guaranteed or not, so play safe.
240 if (fd
< fdt
->max_fds
&& test_bit(fd
, fdt
->close_on_exec
))
243 struct dentry
*dentry
= file
->f_path
.dentry
;
245 /* Make sure we give at least some info */
246 spin_lock(&dentry
->d_lock
);
247 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
,
248 dentry
->d_name
.name
, flags
, file
->f_mode
);
249 spin_unlock(&dentry
->d_lock
);
252 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
, s
,
253 flags
, file
->f_mode
);
259 void lttng_enumerate_task_fd(struct lttng_session
*session
,
260 struct task_struct
*p
, char *tmp
)
262 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .p
= p
};
263 struct files_struct
*files
;
270 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
276 int lttng_enumerate_file_descriptors(struct lttng_session
*session
)
278 struct task_struct
*p
;
281 tmp
= (char *) __get_free_page(GFP_KERNEL
);
285 /* Enumerate active file descriptors */
288 lttng_enumerate_task_fd(session
, p
, tmp
);
290 free_page((unsigned long) tmp
);
296 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
297 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
298 * iteration, but it is not exported to modules.
301 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
302 struct task_struct
*p
)
304 struct mm_struct
*mm
;
305 struct vm_area_struct
*map
;
308 /* get_task_mm does a task_lock... */
315 down_read(&mm
->mmap_sem
);
318 ino
= map
->vm_file
->f_dentry
->d_inode
->i_ino
;
321 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
324 up_read(&mm
->mmap_sem
);
330 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
332 struct task_struct
*p
;
336 lttng_enumerate_task_vm_maps(session
, p
);
342 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
344 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
345 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
349 int lttng_list_interrupts(struct lttng_session
*session
)
352 unsigned long flags
= 0;
353 struct irq_desc
*desc
;
355 #define irq_to_desc wrapper_irq_to_desc
357 for_each_irq_desc(irq
, desc
) {
358 struct irqaction
*action
;
359 const char *irq_chip_name
=
360 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
362 local_irq_save(flags
);
363 wrapper_desc_spin_lock(&desc
->lock
);
364 for (action
= desc
->action
; action
; action
= action
->next
) {
365 trace_lttng_statedump_interrupt(session
,
366 irq
, irq_chip_name
, action
);
368 wrapper_desc_spin_unlock(&desc
->lock
);
369 local_irq_restore(flags
);
376 int lttng_list_interrupts(struct lttng_session
*session
)
383 * Called with task lock held.
386 void lttng_statedump_process_ns(struct lttng_session
*session
,
387 struct task_struct
*p
,
388 enum lttng_thread_type type
,
389 enum lttng_execution_mode mode
,
390 enum lttng_execution_submode submode
,
391 enum lttng_process_status status
)
393 struct nsproxy
*proxy
;
394 struct pid_namespace
*pid_ns
;
397 * Back and forth on locking strategy within Linux upstream for nsproxy.
398 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
399 * "namespaces: Use task_lock and not rcu to protect nsproxy"
402 #if (LINUX_VERSION_CODE < KERNEL_VERSION(3,17,0))
404 proxy
= task_nsproxy(p
);
405 #else /* #if (LINUX_VERSION_CODE < KERNEL_VERSION(3,17,0)) */
407 #endif /* #else #if (LINUX_VERSION_CODE < KERNEL_VERSION(3,17,0)) */
409 pid_ns
= lttng_get_proxy_pid_ns(proxy
);
411 trace_lttng_statedump_process_state(session
,
412 p
, type
, mode
, submode
, status
, pid_ns
);
413 pid_ns
= pid_ns
->parent
;
416 trace_lttng_statedump_process_state(session
,
417 p
, type
, mode
, submode
, status
, NULL
);
419 #if (LINUX_VERSION_CODE < KERNEL_VERSION(3,17,0))
421 #endif /* #if (LINUX_VERSION_CODE < KERNEL_VERSION(3,17,0)) */
425 int lttng_enumerate_process_states(struct lttng_session
*session
)
427 struct task_struct
*g
, *p
;
430 for_each_process(g
) {
433 enum lttng_execution_mode mode
=
435 enum lttng_execution_submode submode
=
437 enum lttng_process_status status
;
438 enum lttng_thread_type type
;
441 if (p
->exit_state
== EXIT_ZOMBIE
)
442 status
= LTTNG_ZOMBIE
;
443 else if (p
->exit_state
== EXIT_DEAD
)
445 else if (p
->state
== TASK_RUNNING
) {
446 /* Is this a forked child that has not run yet? */
447 if (list_empty(&p
->rt
.run_list
))
448 status
= LTTNG_WAIT_FORK
;
451 * All tasks are considered as wait_cpu;
452 * the viewer will sort out if the task
453 * was really running at this time.
455 status
= LTTNG_WAIT_CPU
;
456 } else if (p
->state
&
457 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
458 /* Task is waiting for something to complete */
461 status
= LTTNG_UNNAMED
;
462 submode
= LTTNG_NONE
;
465 * Verification of t->mm is to filter out kernel
466 * threads; Viewer will further filter out if a
467 * user-space thread was in syscall mode or not.
470 type
= LTTNG_USER_THREAD
;
472 type
= LTTNG_KERNEL_THREAD
;
473 lttng_statedump_process_ns(session
,
474 p
, type
, mode
, submode
, status
);
476 } while_each_thread(g
, p
);
484 void lttng_statedump_work_func(struct work_struct
*work
)
486 if (atomic_dec_and_test(&kernel_threads_to_run
))
487 /* If we are the last thread, wake up do_lttng_statedump */
488 wake_up(&statedump_wq
);
492 int do_lttng_statedump(struct lttng_session
*session
)
496 trace_lttng_statedump_start(session
);
497 ret
= lttng_enumerate_process_states(session
);
500 ret
= lttng_enumerate_file_descriptors(session
);
505 * ret = lttng_enumerate_vm_maps(session);
509 ret
= lttng_list_interrupts(session
);
512 ret
= lttng_enumerate_network_ip_interface(session
);
515 ret
= lttng_enumerate_block_devices(session
);
518 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
524 /* TODO lttng_dump_idt_table(session); */
525 /* TODO lttng_dump_softirq_vec(session); */
526 /* TODO lttng_list_modules(session); */
527 /* TODO lttng_dump_swap_files(session); */
530 * Fire off a work queue on each CPU. Their sole purpose in life
531 * is to guarantee that each CPU has been in a state where is was in
532 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
535 atomic_set(&kernel_threads_to_run
, num_online_cpus());
536 for_each_online_cpu(cpu
) {
537 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
538 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
540 /* Wait for all threads to run */
541 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
543 /* Our work is done */
544 trace_lttng_statedump_end(session
);
549 * Called with session mutex held.
551 int lttng_statedump_start(struct lttng_session
*session
)
553 return do_lttng_statedump(session
);
555 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
558 int __init
lttng_statedump_init(void)
561 * Allow module to load even if the fixup cannot be done. This
562 * will allow seemless transition when the underlying issue fix
563 * is merged into the Linux kernel, and when tracepoint.c
564 * "tracepoint_module_notify" is turned into a static function.
566 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
570 module_init(lttng_statedump_init
);
573 void __exit
lttng_statedump_exit(void)
577 module_exit(lttng_statedump_exit
);
579 MODULE_LICENSE("GPL and additional rights");
580 MODULE_AUTHOR("Jean-Hugues Deschenes");
581 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
582 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
583 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
584 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
585 LTTNG_MODULES_EXTRAVERSION
);