Fix: statedump hang due to incorrect wait/wakeup use
[lttng-modules.git] / lttng-statedump-impl.c
CommitLineData
c337ddc2 1/*
886d51a3
MD
2 * lttng-statedump.c
3 *
c337ddc2
MD
4 * Linux Trace Toolkit Next Generation Kernel State Dump
5 *
6 * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
7 * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
8 *
886d51a3
MD
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.
13 *
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.
18 *
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
22 *
c337ddc2
MD
23 * Changes:
24 * Eric Clement: Add listing of network IP interface
25 * 2006, 2007 Mathieu Desnoyers Fix kernel threads
26 * Various updates
c337ddc2
MD
27 */
28
29#include <linux/init.h>
30#include <linux/module.h>
31#include <linux/netlink.h>
32#include <linux/inet.h>
33#include <linux/ip.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>
43#include <linux/mm.h>
44#include <linux/fdtable.h>
45#include <linux/swap.h>
46#include <linux/wait.h>
47#include <linux/mutex.h>
48
49#include "lttng-events.h"
50#include "wrapper/irqdesc.h"
3a523f5b 51#include "wrapper/spinlock.h"
c337ddc2
MD
52
53#ifdef CONFIG_GENERIC_HARDIRQS
54#include <linux/irq.h>
55#endif
56
57/* Define the tracepoints, but do not build the probes */
58#define CREATE_TRACE_POINTS
59#define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
60#define TRACE_INCLUDE_FILE lttng-statedump
61#include "instrumentation/events/lttng-module/lttng-statedump.h"
62
63/*
64 * Protected by the trace lock.
65 */
66static struct delayed_work cpu_work[NR_CPUS];
67static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
68static atomic_t kernel_threads_to_run;
69
70enum lttng_thread_type {
71 LTTNG_USER_THREAD = 0,
72 LTTNG_KERNEL_THREAD = 1,
73};
74
75enum lttng_execution_mode {
76 LTTNG_USER_MODE = 0,
77 LTTNG_SYSCALL = 1,
78 LTTNG_TRAP = 2,
79 LTTNG_IRQ = 3,
80 LTTNG_SOFTIRQ = 4,
81 LTTNG_MODE_UNKNOWN = 5,
82};
83
84enum lttng_execution_submode {
85 LTTNG_NONE = 0,
86 LTTNG_UNKNOWN = 1,
87};
88
89enum lttng_process_status {
90 LTTNG_UNNAMED = 0,
91 LTTNG_WAIT_FORK = 1,
92 LTTNG_WAIT_CPU = 2,
93 LTTNG_EXIT = 3,
94 LTTNG_ZOMBIE = 4,
95 LTTNG_WAIT = 5,
96 LTTNG_RUN = 6,
97 LTTNG_DEAD = 7,
98};
99
100#ifdef CONFIG_INET
101static
102void lttng_enumerate_device(struct lttng_session *session,
103 struct net_device *dev)
104{
105 struct in_device *in_dev;
106 struct in_ifaddr *ifa;
107
108 if (dev->flags & IFF_UP) {
109 in_dev = in_dev_get(dev);
110 if (in_dev) {
111 for (ifa = in_dev->ifa_list; ifa != NULL;
112 ifa = ifa->ifa_next) {
113 trace_lttng_statedump_network_interface(
114 session, dev, ifa);
115 }
116 in_dev_put(in_dev);
117 }
118 } else {
119 trace_lttng_statedump_network_interface(
120 session, dev, NULL);
121 }
122}
123
124static
125int lttng_enumerate_network_ip_interface(struct lttng_session *session)
126{
127 struct net_device *dev;
128
129 read_lock(&dev_base_lock);
130 for_each_netdev(&init_net, dev)
131 lttng_enumerate_device(session, dev);
132 read_unlock(&dev_base_lock);
133
134 return 0;
135}
136#else /* CONFIG_INET */
137static inline
138int lttng_enumerate_network_ip_interface(struct lttng_session *session)
139{
140 return 0;
141}
142#endif /* CONFIG_INET */
143
144
145static
146void lttng_enumerate_task_fd(struct lttng_session *session,
147 struct task_struct *p, char *tmp)
148{
149 struct fdtable *fdt;
150 struct file *filp;
151 unsigned int i;
152 const unsigned char *path;
153
154 task_lock(p);
155 if (!p->files)
156 goto unlock_task;
157 spin_lock(&p->files->file_lock);
158 fdt = files_fdtable(p->files);
159 for (i = 0; i < fdt->max_fds; i++) {
160 filp = fcheck_files(p->files, i);
161 if (!filp)
162 continue;
163 path = d_path(&filp->f_path, tmp, PAGE_SIZE);
164 /* Make sure we give at least some info */
165 trace_lttng_statedump_file_descriptor(session, p, i,
166 IS_ERR(path) ?
167 filp->f_dentry->d_name.name :
168 path);
169 }
170 spin_unlock(&p->files->file_lock);
171unlock_task:
172 task_unlock(p);
173}
174
175static
176int lttng_enumerate_file_descriptors(struct lttng_session *session)
177{
178 struct task_struct *p;
179 char *tmp = (char *) __get_free_page(GFP_KERNEL);
180
181 /* Enumerate active file descriptors */
182 rcu_read_lock();
183 for_each_process(p)
184 lttng_enumerate_task_fd(session, p, tmp);
185 rcu_read_unlock();
186 free_page((unsigned long) tmp);
187 return 0;
188}
189
0658bdda
MD
190#if 0
191/*
192 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
193 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
194 * iteration, but it is not exported to modules.
195 */
c337ddc2
MD
196static
197void lttng_enumerate_task_vm_maps(struct lttng_session *session,
198 struct task_struct *p)
199{
200 struct mm_struct *mm;
201 struct vm_area_struct *map;
202 unsigned long ino;
203
204 /* get_task_mm does a task_lock... */
205 mm = get_task_mm(p);
206 if (!mm)
207 return;
208
209 map = mm->mmap;
210 if (map) {
211 down_read(&mm->mmap_sem);
212 while (map) {
213 if (map->vm_file)
214 ino = map->vm_file->f_dentry->d_inode->i_ino;
215 else
216 ino = 0;
217 trace_lttng_statedump_vm_map(session, p, map, ino);
218 map = map->vm_next;
219 }
220 up_read(&mm->mmap_sem);
221 }
222 mmput(mm);
223}
224
225static
226int lttng_enumerate_vm_maps(struct lttng_session *session)
227{
228 struct task_struct *p;
229
230 rcu_read_lock();
231 for_each_process(p)
232 lttng_enumerate_task_vm_maps(session, p);
233 rcu_read_unlock();
234 return 0;
235}
0658bdda 236#endif
c337ddc2
MD
237
238#ifdef CONFIG_GENERIC_HARDIRQS
47faec4b
JN
239
240#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
241#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
242#endif
243
c337ddc2
MD
244static
245void lttng_list_interrupts(struct lttng_session *session)
246{
247 unsigned int irq;
248 unsigned long flags = 0;
249 struct irq_desc *desc;
250
251#define irq_to_desc wrapper_irq_to_desc
252 /* needs irq_desc */
253 for_each_irq_desc(irq, desc) {
254 struct irqaction *action;
255 const char *irq_chip_name =
256 irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
257
258 local_irq_save(flags);
3a523f5b 259 wrapper_desc_spin_lock(&desc->lock);
c337ddc2
MD
260 for (action = desc->action; action; action = action->next) {
261 trace_lttng_statedump_interrupt(session,
262 irq, irq_chip_name, action);
263 }
3a523f5b 264 wrapper_desc_spin_unlock(&desc->lock);
c337ddc2
MD
265 local_irq_restore(flags);
266 }
267#undef irq_to_desc
268}
269#else
270static inline
271void list_interrupts(struct lttng_session *session)
272{
273}
274#endif
275
73e8ba37
JD
276static
277void lttng_statedump_process_ns(struct lttng_session *session,
278 struct task_struct *p,
279 enum lttng_thread_type type,
280 enum lttng_execution_mode mode,
281 enum lttng_execution_submode submode,
282 enum lttng_process_status status)
283{
284 struct nsproxy *proxy;
285 struct pid_namespace *pid_ns;
286
287 rcu_read_lock();
288 proxy = task_nsproxy(p);
289 if (proxy) {
290 pid_ns = proxy->pid_ns;
291 do {
292 trace_lttng_statedump_process_state(session,
293 p, type, mode, submode, status, pid_ns);
af73f727 294 pid_ns = pid_ns->parent;
73e8ba37
JD
295 } while (pid_ns);
296 } else {
297 trace_lttng_statedump_process_state(session,
298 p, type, mode, submode, status, NULL);
299 }
300 rcu_read_unlock();
301}
302
c337ddc2
MD
303static
304int lttng_enumerate_process_states(struct lttng_session *session)
305{
306 struct task_struct *g, *p;
307
308 rcu_read_lock();
309 for_each_process(g) {
310 p = g;
311 do {
312 enum lttng_execution_mode mode =
313 LTTNG_MODE_UNKNOWN;
314 enum lttng_execution_submode submode =
315 LTTNG_UNKNOWN;
316 enum lttng_process_status status;
317 enum lttng_thread_type type;
318
319 task_lock(p);
320 if (p->exit_state == EXIT_ZOMBIE)
321 status = LTTNG_ZOMBIE;
322 else if (p->exit_state == EXIT_DEAD)
323 status = LTTNG_DEAD;
324 else if (p->state == TASK_RUNNING) {
325 /* Is this a forked child that has not run yet? */
326 if (list_empty(&p->rt.run_list))
327 status = LTTNG_WAIT_FORK;
328 else
329 /*
330 * All tasks are considered as wait_cpu;
331 * the viewer will sort out if the task
332 * was really running at this time.
333 */
334 status = LTTNG_WAIT_CPU;
335 } else if (p->state &
336 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
337 /* Task is waiting for something to complete */
338 status = LTTNG_WAIT;
339 } else
340 status = LTTNG_UNNAMED;
341 submode = LTTNG_NONE;
342
343 /*
344 * Verification of t->mm is to filter out kernel
345 * threads; Viewer will further filter out if a
346 * user-space thread was in syscall mode or not.
347 */
348 if (p->mm)
349 type = LTTNG_USER_THREAD;
350 else
351 type = LTTNG_KERNEL_THREAD;
73e8ba37 352 lttng_statedump_process_ns(session,
c337ddc2
MD
353 p, type, mode, submode, status);
354 task_unlock(p);
355 } while_each_thread(g, p);
356 }
357 rcu_read_unlock();
358
359 return 0;
360}
361
362static
363void lttng_statedump_work_func(struct work_struct *work)
364{
365 if (atomic_dec_and_test(&kernel_threads_to_run))
366 /* If we are the last thread, wake up do_lttng_statedump */
367 wake_up(&statedump_wq);
368}
369
370static
371int do_lttng_statedump(struct lttng_session *session)
372{
373 int cpu;
374
375 printk(KERN_DEBUG "LTT state dump thread start\n");
376 trace_lttng_statedump_start(session);
377 lttng_enumerate_process_states(session);
378 lttng_enumerate_file_descriptors(session);
0658bdda 379 /* FIXME lttng_enumerate_vm_maps(session); */
c337ddc2
MD
380 lttng_list_interrupts(session);
381 lttng_enumerate_network_ip_interface(session);
382
383 /* TODO lttng_dump_idt_table(session); */
384 /* TODO lttng_dump_softirq_vec(session); */
385 /* TODO lttng_list_modules(session); */
386 /* TODO lttng_dump_swap_files(session); */
387
388 /*
389 * Fire off a work queue on each CPU. Their sole purpose in life
390 * is to guarantee that each CPU has been in a state where is was in
391 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
392 */
393 get_online_cpus();
394 atomic_set(&kernel_threads_to_run, num_online_cpus());
395 for_each_online_cpu(cpu) {
396 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
397 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
398 }
399 /* Wait for all threads to run */
8240b2e2 400 wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) != 0));
c337ddc2
MD
401 put_online_cpus();
402 /* Our work is done */
403 printk(KERN_DEBUG "LTT state dump end\n");
404 trace_lttng_statedump_end(session);
405 return 0;
406}
407
408/*
409 * Called with session mutex held.
410 */
411int lttng_statedump_start(struct lttng_session *session)
412{
413 printk(KERN_DEBUG "LTTng: state dump begin\n");
414 return do_lttng_statedump(session);
415}
416EXPORT_SYMBOL_GPL(lttng_statedump_start);
417
418MODULE_LICENSE("GPL and additional rights");
419MODULE_AUTHOR("Jean-Hugues Deschenes");
420MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
This page took 0.037892 seconds and 4 git commands to generate.