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>
59 #include <wrapper/file.h>
60 #include <wrapper/time.h>
62 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
63 #include <linux/irq.h>
66 /* Define the tracepoints, but do not build the probes */
67 #define CREATE_TRACE_POINTS
68 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
69 #define TRACE_INCLUDE_FILE lttng-statedump
70 #define LTTNG_INSTRUMENTATION
71 #include <instrumentation/events/lttng-module/lttng-statedump.h>
73 DEFINE_TRACE(lttng_statedump_block_device
);
74 DEFINE_TRACE(lttng_statedump_end
);
75 DEFINE_TRACE(lttng_statedump_interrupt
);
76 DEFINE_TRACE(lttng_statedump_file_descriptor
);
77 DEFINE_TRACE(lttng_statedump_start
);
78 DEFINE_TRACE(lttng_statedump_process_state
);
79 DEFINE_TRACE(lttng_statedump_network_interface
);
83 struct lttng_session
*session
;
84 struct task_struct
*p
;
85 struct files_struct
*files
;
89 * Protected by the trace lock.
91 static struct delayed_work cpu_work
[NR_CPUS
];
92 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
93 static atomic_t kernel_threads_to_run
;
95 enum lttng_thread_type
{
96 LTTNG_USER_THREAD
= 0,
97 LTTNG_KERNEL_THREAD
= 1,
100 enum lttng_execution_mode
{
106 LTTNG_MODE_UNKNOWN
= 5,
109 enum lttng_execution_submode
{
114 enum lttng_process_status
{
126 int lttng_enumerate_block_devices(struct lttng_session
*session
)
128 struct class *ptr_block_class
;
129 struct device_type
*ptr_disk_type
;
130 struct class_dev_iter iter
;
133 ptr_block_class
= wrapper_get_block_class();
134 if (!ptr_block_class
)
136 ptr_disk_type
= wrapper_get_disk_type();
137 if (!ptr_disk_type
) {
140 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
141 while ((dev
= class_dev_iter_next(&iter
))) {
142 struct disk_part_iter piter
;
143 struct gendisk
*disk
= dev_to_disk(dev
);
144 struct hd_struct
*part
;
147 * Don't show empty devices or things that have been
150 if (get_capacity(disk
) == 0 ||
151 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
154 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
155 while ((part
= disk_part_iter_next(&piter
))) {
156 char name_buf
[BDEVNAME_SIZE
];
159 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
161 disk_part_iter_exit(&piter
);
162 class_dev_iter_exit(&iter
);
165 trace_lttng_statedump_block_device(session
,
166 part_devt(part
), name_buf
);
168 disk_part_iter_exit(&piter
);
170 class_dev_iter_exit(&iter
);
177 void lttng_enumerate_device(struct lttng_session
*session
,
178 struct net_device
*dev
)
180 struct in_device
*in_dev
;
181 struct in_ifaddr
*ifa
;
183 if (dev
->flags
& IFF_UP
) {
184 in_dev
= in_dev_get(dev
);
186 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
187 ifa
= ifa
->ifa_next
) {
188 trace_lttng_statedump_network_interface(
194 trace_lttng_statedump_network_interface(
200 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
202 struct net_device
*dev
;
204 read_lock(&dev_base_lock
);
205 for_each_netdev(&init_net
, dev
)
206 lttng_enumerate_device(session
, dev
);
207 read_unlock(&dev_base_lock
);
211 #else /* CONFIG_INET */
213 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
217 #endif /* CONFIG_INET */
220 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
222 const struct lttng_fd_ctx
*ctx
= p
;
223 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
224 unsigned int flags
= file
->f_flags
;
228 * We don't expose kernel internal flags, only userspace-visible
231 flags
&= ~FMODE_NONOTIFY
;
232 fdt
= files_fdtable(ctx
->files
);
234 * We need to check here again whether fd is within the fdt
235 * max_fds range, because we might be seeing a different
236 * files_fdtable() than iterate_fd(), assuming only RCU is
237 * protecting the read. In reality, iterate_fd() holds
238 * file_lock, which should ensure the fdt does not change while
239 * the lock is taken, but we are not aware whether this is
240 * guaranteed or not, so play safe.
242 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
245 struct dentry
*dentry
= file
->f_path
.dentry
;
247 /* Make sure we give at least some info */
248 spin_lock(&dentry
->d_lock
);
249 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
,
250 dentry
->d_name
.name
, flags
, file
->f_mode
);
251 spin_unlock(&dentry
->d_lock
);
254 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
, s
,
255 flags
, file
->f_mode
);
261 void lttng_enumerate_task_fd(struct lttng_session
*session
,
262 struct task_struct
*p
, char *tmp
)
264 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .p
= p
};
265 struct files_struct
*files
;
272 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
278 int lttng_enumerate_file_descriptors(struct lttng_session
*session
)
280 struct task_struct
*p
;
283 tmp
= (char *) __get_free_page(GFP_KERNEL
);
287 /* Enumerate active file descriptors */
290 lttng_enumerate_task_fd(session
, p
, tmp
);
292 free_page((unsigned long) tmp
);
298 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
299 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
300 * iteration, but it is not exported to modules.
303 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
304 struct task_struct
*p
)
306 struct mm_struct
*mm
;
307 struct vm_area_struct
*map
;
310 /* get_task_mm does a task_lock... */
317 down_read(&mm
->mmap_sem
);
320 ino
= map
->vm_file
->lttng_f_dentry
->d_inode
->i_ino
;
323 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
326 up_read(&mm
->mmap_sem
);
332 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
334 struct task_struct
*p
;
338 lttng_enumerate_task_vm_maps(session
, p
);
344 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
346 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
347 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
351 int lttng_list_interrupts(struct lttng_session
*session
)
354 unsigned long flags
= 0;
355 struct irq_desc
*desc
;
357 #define irq_to_desc wrapper_irq_to_desc
359 for_each_irq_desc(irq
, desc
) {
360 struct irqaction
*action
;
361 const char *irq_chip_name
=
362 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
364 local_irq_save(flags
);
365 wrapper_desc_spin_lock(&desc
->lock
);
366 for (action
= desc
->action
; action
; action
= action
->next
) {
367 trace_lttng_statedump_interrupt(session
,
368 irq
, irq_chip_name
, action
);
370 wrapper_desc_spin_unlock(&desc
->lock
);
371 local_irq_restore(flags
);
378 int lttng_list_interrupts(struct lttng_session
*session
)
385 * Called with task lock held.
388 void lttng_statedump_process_ns(struct lttng_session
*session
,
389 struct task_struct
*p
,
390 enum lttng_thread_type type
,
391 enum lttng_execution_mode mode
,
392 enum lttng_execution_submode submode
,
393 enum lttng_process_status status
)
395 struct nsproxy
*proxy
;
396 struct pid_namespace
*pid_ns
;
399 * Back and forth on locking strategy within Linux upstream for nsproxy.
400 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
401 * "namespaces: Use task_lock and not rcu to protect nsproxy"
404 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
405 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
406 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
407 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
411 proxy
= task_nsproxy(p
);
414 pid_ns
= lttng_get_proxy_pid_ns(proxy
);
416 trace_lttng_statedump_process_state(session
,
417 p
, type
, mode
, submode
, status
, pid_ns
);
418 pid_ns
= pid_ns
->parent
;
421 trace_lttng_statedump_process_state(session
,
422 p
, type
, mode
, submode
, status
, NULL
);
424 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
425 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
426 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
427 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
435 int lttng_enumerate_process_states(struct lttng_session
*session
)
437 struct task_struct
*g
, *p
;
440 for_each_process(g
) {
443 enum lttng_execution_mode mode
=
445 enum lttng_execution_submode submode
=
447 enum lttng_process_status status
;
448 enum lttng_thread_type type
;
451 if (p
->exit_state
== EXIT_ZOMBIE
)
452 status
= LTTNG_ZOMBIE
;
453 else if (p
->exit_state
== EXIT_DEAD
)
455 else if (p
->state
== TASK_RUNNING
) {
456 /* Is this a forked child that has not run yet? */
457 if (list_empty(&p
->rt
.run_list
))
458 status
= LTTNG_WAIT_FORK
;
461 * All tasks are considered as wait_cpu;
462 * the viewer will sort out if the task
463 * was really running at this time.
465 status
= LTTNG_WAIT_CPU
;
466 } else if (p
->state
&
467 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
468 /* Task is waiting for something to complete */
471 status
= LTTNG_UNNAMED
;
472 submode
= LTTNG_NONE
;
475 * Verification of t->mm is to filter out kernel
476 * threads; Viewer will further filter out if a
477 * user-space thread was in syscall mode or not.
480 type
= LTTNG_USER_THREAD
;
482 type
= LTTNG_KERNEL_THREAD
;
483 lttng_statedump_process_ns(session
,
484 p
, type
, mode
, submode
, status
);
486 } while_each_thread(g
, p
);
494 void lttng_statedump_work_func(struct work_struct
*work
)
496 if (atomic_dec_and_test(&kernel_threads_to_run
))
497 /* If we are the last thread, wake up do_lttng_statedump */
498 wake_up(&statedump_wq
);
502 int do_lttng_statedump(struct lttng_session
*session
)
506 trace_lttng_statedump_start(session
);
507 ret
= lttng_enumerate_process_states(session
);
510 ret
= lttng_enumerate_file_descriptors(session
);
515 * ret = lttng_enumerate_vm_maps(session);
519 ret
= lttng_list_interrupts(session
);
522 ret
= lttng_enumerate_network_ip_interface(session
);
525 ret
= lttng_enumerate_block_devices(session
);
530 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
536 /* TODO lttng_dump_idt_table(session); */
537 /* TODO lttng_dump_softirq_vec(session); */
538 /* TODO lttng_list_modules(session); */
539 /* TODO lttng_dump_swap_files(session); */
542 * Fire off a work queue on each CPU. Their sole purpose in life
543 * is to guarantee that each CPU has been in a state where is was in
544 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
547 atomic_set(&kernel_threads_to_run
, num_online_cpus());
548 for_each_online_cpu(cpu
) {
549 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
550 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
552 /* Wait for all threads to run */
553 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
555 /* Our work is done */
556 trace_lttng_statedump_end(session
);
561 * Called with session mutex held.
563 int lttng_statedump_start(struct lttng_session
*session
)
565 return do_lttng_statedump(session
);
567 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
570 int __init
lttng_statedump_init(void)
573 * Allow module to load even if the fixup cannot be done. This
574 * will allow seemless transition when the underlying issue fix
575 * is merged into the Linux kernel, and when tracepoint.c
576 * "tracepoint_module_notify" is turned into a static function.
578 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
582 module_init(lttng_statedump_init
);
585 void __exit
lttng_statedump_exit(void)
589 module_exit(lttng_statedump_exit
);
591 MODULE_LICENSE("GPL and additional rights");
592 MODULE_AUTHOR("Jean-Hugues Deschenes");
593 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
594 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
595 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
596 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
597 LTTNG_MODULES_EXTRAVERSION
);