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 #include "instrumentation/events/lttng-module/lttng-statedump.h"
70 DEFINE_TRACE(lttng_statedump_block_device
);
71 DEFINE_TRACE(lttng_statedump_end
);
72 DEFINE_TRACE(lttng_statedump_interrupt
);
73 DEFINE_TRACE(lttng_statedump_file_descriptor
);
74 DEFINE_TRACE(lttng_statedump_start
);
75 DEFINE_TRACE(lttng_statedump_process_state
);
76 DEFINE_TRACE(lttng_statedump_network_interface
);
80 struct lttng_session
*session
;
81 struct task_struct
*p
;
82 struct files_struct
*files
;
86 * Protected by the trace lock.
88 static struct delayed_work cpu_work
[NR_CPUS
];
89 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq
);
90 static atomic_t kernel_threads_to_run
;
92 enum lttng_thread_type
{
93 LTTNG_USER_THREAD
= 0,
94 LTTNG_KERNEL_THREAD
= 1,
97 enum lttng_execution_mode
{
103 LTTNG_MODE_UNKNOWN
= 5,
106 enum lttng_execution_submode
{
111 enum lttng_process_status
{
123 int lttng_enumerate_block_devices(struct lttng_session
*session
)
125 struct class *ptr_block_class
;
126 struct device_type
*ptr_disk_type
;
127 struct class_dev_iter iter
;
130 ptr_block_class
= wrapper_get_block_class();
131 if (!ptr_block_class
)
133 ptr_disk_type
= wrapper_get_disk_type();
134 if (!ptr_disk_type
) {
137 class_dev_iter_init(&iter
, ptr_block_class
, NULL
, ptr_disk_type
);
138 while ((dev
= class_dev_iter_next(&iter
))) {
139 struct disk_part_iter piter
;
140 struct gendisk
*disk
= dev_to_disk(dev
);
141 struct hd_struct
*part
;
144 * Don't show empty devices or things that have been
147 if (get_capacity(disk
) == 0 ||
148 (disk
->flags
& GENHD_FL_SUPPRESS_PARTITION_INFO
))
151 disk_part_iter_init(&piter
, disk
, DISK_PITER_INCL_PART0
);
152 while ((part
= disk_part_iter_next(&piter
))) {
153 char name_buf
[BDEVNAME_SIZE
];
156 p
= wrapper_disk_name(disk
, part
->partno
, name_buf
);
158 disk_part_iter_exit(&piter
);
159 class_dev_iter_exit(&iter
);
162 trace_lttng_statedump_block_device(session
,
163 part_devt(part
), name_buf
);
165 disk_part_iter_exit(&piter
);
167 class_dev_iter_exit(&iter
);
174 void lttng_enumerate_device(struct lttng_session
*session
,
175 struct net_device
*dev
)
177 struct in_device
*in_dev
;
178 struct in_ifaddr
*ifa
;
180 if (dev
->flags
& IFF_UP
) {
181 in_dev
= in_dev_get(dev
);
183 for (ifa
= in_dev
->ifa_list
; ifa
!= NULL
;
184 ifa
= ifa
->ifa_next
) {
185 trace_lttng_statedump_network_interface(
191 trace_lttng_statedump_network_interface(
197 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
199 struct net_device
*dev
;
201 read_lock(&dev_base_lock
);
202 for_each_netdev(&init_net
, dev
)
203 lttng_enumerate_device(session
, dev
);
204 read_unlock(&dev_base_lock
);
208 #else /* CONFIG_INET */
210 int lttng_enumerate_network_ip_interface(struct lttng_session
*session
)
214 #endif /* CONFIG_INET */
216 #ifdef FD_ISSET /* For old kernels lacking close_on_exec() */
217 static inline bool lttng_close_on_exec(int fd
, const struct fdtable
*fdt
)
219 return FD_ISSET(fd
, fdt
->close_on_exec
);
222 static inline bool lttng_close_on_exec(int fd
, const struct fdtable
*fdt
)
224 return close_on_exec(fd
, fdt
);
229 int lttng_dump_one_fd(const void *p
, struct file
*file
, unsigned int fd
)
231 const struct lttng_fd_ctx
*ctx
= p
;
232 const char *s
= d_path(&file
->f_path
, ctx
->page
, PAGE_SIZE
);
233 unsigned int flags
= file
->f_flags
;
237 * We don't expose kernel internal flags, only userspace-visible
240 flags
&= ~FMODE_NONOTIFY
;
241 fdt
= files_fdtable(ctx
->files
);
243 * We need to check here again whether fd is within the fdt
244 * max_fds range, because we might be seeing a different
245 * files_fdtable() than iterate_fd(), assuming only RCU is
246 * protecting the read. In reality, iterate_fd() holds
247 * file_lock, which should ensure the fdt does not change while
248 * the lock is taken, but we are not aware whether this is
249 * guaranteed or not, so play safe.
251 if (fd
< fdt
->max_fds
&& lttng_close_on_exec(fd
, fdt
))
254 struct dentry
*dentry
= file
->f_path
.dentry
;
256 /* Make sure we give at least some info */
257 spin_lock(&dentry
->d_lock
);
258 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
,
259 dentry
->d_name
.name
, flags
, file
->f_mode
);
260 spin_unlock(&dentry
->d_lock
);
263 trace_lttng_statedump_file_descriptor(ctx
->session
, ctx
->p
, fd
, s
,
264 flags
, file
->f_mode
);
270 void lttng_enumerate_task_fd(struct lttng_session
*session
,
271 struct task_struct
*p
, char *tmp
)
273 struct lttng_fd_ctx ctx
= { .page
= tmp
, .session
= session
, .p
= p
};
274 struct files_struct
*files
;
281 lttng_iterate_fd(files
, 0, lttng_dump_one_fd
, &ctx
);
287 int lttng_enumerate_file_descriptors(struct lttng_session
*session
)
289 struct task_struct
*p
;
292 tmp
= (char *) __get_free_page(GFP_KERNEL
);
296 /* Enumerate active file descriptors */
299 lttng_enumerate_task_fd(session
, p
, tmp
);
301 free_page((unsigned long) tmp
);
307 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
308 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
309 * iteration, but it is not exported to modules.
312 void lttng_enumerate_task_vm_maps(struct lttng_session
*session
,
313 struct task_struct
*p
)
315 struct mm_struct
*mm
;
316 struct vm_area_struct
*map
;
319 /* get_task_mm does a task_lock... */
326 down_read(&mm
->mmap_sem
);
329 ino
= map
->vm_file
->f_dentry
->d_inode
->i_ino
;
332 trace_lttng_statedump_vm_map(session
, p
, map
, ino
);
335 up_read(&mm
->mmap_sem
);
341 int lttng_enumerate_vm_maps(struct lttng_session
*session
)
343 struct task_struct
*p
;
347 lttng_enumerate_task_vm_maps(session
, p
);
353 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
355 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
356 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
360 int lttng_list_interrupts(struct lttng_session
*session
)
363 unsigned long flags
= 0;
364 struct irq_desc
*desc
;
366 #define irq_to_desc wrapper_irq_to_desc
368 for_each_irq_desc(irq
, desc
) {
369 struct irqaction
*action
;
370 const char *irq_chip_name
=
371 irq_desc_get_chip(desc
)->name
? : "unnamed_irq_chip";
373 local_irq_save(flags
);
374 wrapper_desc_spin_lock(&desc
->lock
);
375 for (action
= desc
->action
; action
; action
= action
->next
) {
376 trace_lttng_statedump_interrupt(session
,
377 irq
, irq_chip_name
, action
);
379 wrapper_desc_spin_unlock(&desc
->lock
);
380 local_irq_restore(flags
);
387 int lttng_list_interrupts(struct lttng_session
*session
)
394 * Called with task lock held.
397 void lttng_statedump_process_ns(struct lttng_session
*session
,
398 struct task_struct
*p
,
399 enum lttng_thread_type type
,
400 enum lttng_execution_mode mode
,
401 enum lttng_execution_submode submode
,
402 enum lttng_process_status status
)
404 struct nsproxy
*proxy
;
405 struct pid_namespace
*pid_ns
;
408 * Back and forth on locking strategy within Linux upstream for nsproxy.
409 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
410 * "namespaces: Use task_lock and not rcu to protect nsproxy"
413 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
414 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
415 LTTNG_UBUNTU_KERNEL_RANGE(3,16,0,11, 3,17,0,0))
419 proxy
= task_nsproxy(p
);
422 pid_ns
= lttng_get_proxy_pid_ns(proxy
);
424 trace_lttng_statedump_process_state(session
,
425 p
, type
, mode
, submode
, status
, pid_ns
);
426 pid_ns
= pid_ns
->parent
;
429 trace_lttng_statedump_process_state(session
,
430 p
, type
, mode
, submode
, status
, NULL
);
432 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
433 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
434 LTTNG_UBUNTU_KERNEL_RANGE(3,16,0,11, 3,17,0,0))
442 int lttng_enumerate_process_states(struct lttng_session
*session
)
444 struct task_struct
*g
, *p
;
447 for_each_process(g
) {
450 enum lttng_execution_mode mode
=
452 enum lttng_execution_submode submode
=
454 enum lttng_process_status status
;
455 enum lttng_thread_type type
;
458 if (p
->exit_state
== EXIT_ZOMBIE
)
459 status
= LTTNG_ZOMBIE
;
460 else if (p
->exit_state
== EXIT_DEAD
)
462 else if (p
->state
== TASK_RUNNING
) {
463 /* Is this a forked child that has not run yet? */
464 if (list_empty(&p
->rt
.run_list
))
465 status
= LTTNG_WAIT_FORK
;
468 * All tasks are considered as wait_cpu;
469 * the viewer will sort out if the task
470 * was really running at this time.
472 status
= LTTNG_WAIT_CPU
;
473 } else if (p
->state
&
474 (TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
)) {
475 /* Task is waiting for something to complete */
478 status
= LTTNG_UNNAMED
;
479 submode
= LTTNG_NONE
;
482 * Verification of t->mm is to filter out kernel
483 * threads; Viewer will further filter out if a
484 * user-space thread was in syscall mode or not.
487 type
= LTTNG_USER_THREAD
;
489 type
= LTTNG_KERNEL_THREAD
;
490 lttng_statedump_process_ns(session
,
491 p
, type
, mode
, submode
, status
);
493 } while_each_thread(g
, p
);
501 void lttng_statedump_work_func(struct work_struct
*work
)
503 if (atomic_dec_and_test(&kernel_threads_to_run
))
504 /* If we are the last thread, wake up do_lttng_statedump */
505 wake_up(&statedump_wq
);
509 int do_lttng_statedump(struct lttng_session
*session
)
513 trace_lttng_statedump_start(session
);
514 ret
= lttng_enumerate_process_states(session
);
517 ret
= lttng_enumerate_file_descriptors(session
);
522 * ret = lttng_enumerate_vm_maps(session);
526 ret
= lttng_list_interrupts(session
);
529 ret
= lttng_enumerate_network_ip_interface(session
);
532 ret
= lttng_enumerate_block_devices(session
);
535 printk(KERN_WARNING
"LTTng: block device enumeration is not supported by kernel\n");
541 /* TODO lttng_dump_idt_table(session); */
542 /* TODO lttng_dump_softirq_vec(session); */
543 /* TODO lttng_list_modules(session); */
544 /* TODO lttng_dump_swap_files(session); */
547 * Fire off a work queue on each CPU. Their sole purpose in life
548 * is to guarantee that each CPU has been in a state where is was in
549 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
552 atomic_set(&kernel_threads_to_run
, num_online_cpus());
553 for_each_online_cpu(cpu
) {
554 INIT_DELAYED_WORK(&cpu_work
[cpu
], lttng_statedump_work_func
);
555 schedule_delayed_work_on(cpu
, &cpu_work
[cpu
], 0);
557 /* Wait for all threads to run */
558 __wait_event(statedump_wq
, (atomic_read(&kernel_threads_to_run
) == 0));
560 /* Our work is done */
561 trace_lttng_statedump_end(session
);
566 * Called with session mutex held.
568 int lttng_statedump_start(struct lttng_session
*session
)
570 return do_lttng_statedump(session
);
572 EXPORT_SYMBOL_GPL(lttng_statedump_start
);
575 int __init
lttng_statedump_init(void)
578 * Allow module to load even if the fixup cannot be done. This
579 * will allow seemless transition when the underlying issue fix
580 * is merged into the Linux kernel, and when tracepoint.c
581 * "tracepoint_module_notify" is turned into a static function.
583 (void) wrapper_lttng_fixup_sig(THIS_MODULE
);
587 module_init(lttng_statedump_init
);
590 void __exit
lttng_statedump_exit(void)
594 module_exit(lttng_statedump_exit
);
596 MODULE_LICENSE("GPL and additional rights");
597 MODULE_AUTHOR("Jean-Hugues Deschenes");
598 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
599 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION
) "."
600 __stringify(LTTNG_MODULES_MINOR_VERSION
) "."
601 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION
)
602 LTTNG_MODULES_EXTRAVERSION
);