Fix: update kvm instrumentation for Ubuntu 4.15.0-92
[deliverable/lttng-modules.git] / lttng-statedump-impl.c
1 /* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
2 *
3 * lttng-statedump.c
4 *
5 * Linux Trace Toolkit Next Generation Kernel State Dump
6 *
7 * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
8 * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
9 *
10 * Changes:
11 * Eric Clement: Add listing of network IP interface
12 * 2006, 2007 Mathieu Desnoyers Fix kernel threads
13 * Various updates
14 */
15
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/netlink.h>
19 #include <linux/inet.h>
20 #include <linux/ip.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>
30 #include <linux/mm.h>
31 #include <linux/swap.h>
32 #include <linux/wait.h>
33 #include <linux/mutex.h>
34 #include <linux/device.h>
35
36 #include <lttng-events.h>
37 #include <lttng-tracer.h>
38 #include <wrapper/irqdesc.h>
39 #include <wrapper/fdtable.h>
40 #include <wrapper/namespace.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>
46
47 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
48 #include <linux/irq.h>
49 #endif
50
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>
57
58 DEFINE_TRACE(lttng_statedump_block_device);
59 DEFINE_TRACE(lttng_statedump_end);
60 DEFINE_TRACE(lttng_statedump_interrupt);
61 DEFINE_TRACE(lttng_statedump_file_descriptor);
62 DEFINE_TRACE(lttng_statedump_start);
63 DEFINE_TRACE(lttng_statedump_process_state);
64 DEFINE_TRACE(lttng_statedump_process_pid_ns);
65 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
66 DEFINE_TRACE(lttng_statedump_process_cgroup_ns);
67 #endif
68 DEFINE_TRACE(lttng_statedump_process_ipc_ns);
69 #ifndef LTTNG_MNT_NS_MISSING_HEADER
70 DEFINE_TRACE(lttng_statedump_process_mnt_ns);
71 #endif
72 DEFINE_TRACE(lttng_statedump_process_net_ns);
73 DEFINE_TRACE(lttng_statedump_process_user_ns);
74 DEFINE_TRACE(lttng_statedump_process_uts_ns);
75 DEFINE_TRACE(lttng_statedump_network_interface);
76 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
77 DEFINE_TRACE(lttng_statedump_cpu_topology);
78 #endif
79
80 struct lttng_fd_ctx {
81 char *page;
82 struct lttng_session *session;
83 struct files_struct *files;
84 };
85
86 /*
87 * Protected by the trace lock.
88 */
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;
92
93 enum lttng_thread_type {
94 LTTNG_USER_THREAD = 0,
95 LTTNG_KERNEL_THREAD = 1,
96 };
97
98 enum lttng_execution_mode {
99 LTTNG_USER_MODE = 0,
100 LTTNG_SYSCALL = 1,
101 LTTNG_TRAP = 2,
102 LTTNG_IRQ = 3,
103 LTTNG_SOFTIRQ = 4,
104 LTTNG_MODE_UNKNOWN = 5,
105 };
106
107 enum lttng_execution_submode {
108 LTTNG_NONE = 0,
109 LTTNG_UNKNOWN = 1,
110 };
111
112 enum lttng_process_status {
113 LTTNG_UNNAMED = 0,
114 LTTNG_WAIT_FORK = 1,
115 LTTNG_WAIT_CPU = 2,
116 LTTNG_EXIT = 3,
117 LTTNG_ZOMBIE = 4,
118 LTTNG_WAIT = 5,
119 LTTNG_RUN = 6,
120 LTTNG_DEAD = 7,
121 };
122
123 static
124 int lttng_enumerate_block_devices(struct lttng_session *session)
125 {
126 struct class *ptr_block_class;
127 struct device_type *ptr_disk_type;
128 struct class_dev_iter iter;
129 struct device *dev;
130
131 ptr_block_class = wrapper_get_block_class();
132 if (!ptr_block_class)
133 return -ENOSYS;
134 ptr_disk_type = wrapper_get_disk_type();
135 if (!ptr_disk_type) {
136 return -ENOSYS;
137 }
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;
143
144 /*
145 * Don't show empty devices or things that have been
146 * suppressed
147 */
148 if (get_capacity(disk) == 0 ||
149 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
150 continue;
151
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];
155 char *p;
156
157 p = wrapper_disk_name(disk, part->partno, name_buf);
158 if (!p) {
159 disk_part_iter_exit(&piter);
160 class_dev_iter_exit(&iter);
161 return -ENOSYS;
162 }
163 trace_lttng_statedump_block_device(session,
164 part_devt(part), name_buf);
165 }
166 disk_part_iter_exit(&piter);
167 }
168 class_dev_iter_exit(&iter);
169 return 0;
170 }
171
172 #ifdef CONFIG_INET
173
174 static
175 void lttng_enumerate_device(struct lttng_session *session,
176 struct net_device *dev)
177 {
178 struct in_device *in_dev;
179 struct in_ifaddr *ifa;
180
181 if (dev->flags & IFF_UP) {
182 in_dev = in_dev_get(dev);
183 if (in_dev) {
184 for (ifa = in_dev->ifa_list; ifa != NULL;
185 ifa = ifa->ifa_next) {
186 trace_lttng_statedump_network_interface(
187 session, dev, ifa);
188 }
189 in_dev_put(in_dev);
190 }
191 } else {
192 trace_lttng_statedump_network_interface(
193 session, dev, NULL);
194 }
195 }
196
197 static
198 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
199 {
200 struct net_device *dev;
201
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);
206
207 return 0;
208 }
209 #else /* CONFIG_INET */
210 static inline
211 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
212 {
213 return 0;
214 }
215 #endif /* CONFIG_INET */
216
217 static
218 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
219 {
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;
223 struct fdtable *fdt;
224
225 /*
226 * We don't expose kernel internal flags, only userspace-visible
227 * flags.
228 */
229 flags &= ~FMODE_NONOTIFY;
230 fdt = files_fdtable(ctx->files);
231 /*
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.
239 */
240 if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
241 flags |= O_CLOEXEC;
242 if (IS_ERR(s)) {
243 struct dentry *dentry = file->f_path.dentry;
244
245 /* Make sure we give at least some info */
246 spin_lock(&dentry->d_lock);
247 trace_lttng_statedump_file_descriptor(ctx->session,
248 ctx->files, fd, dentry->d_name.name, flags,
249 file->f_mode);
250 spin_unlock(&dentry->d_lock);
251 goto end;
252 }
253 trace_lttng_statedump_file_descriptor(ctx->session,
254 ctx->files, fd, s, flags, file->f_mode);
255 end:
256 return 0;
257 }
258
259 /* Called with task lock held. */
260 static
261 void lttng_enumerate_files(struct lttng_session *session,
262 struct files_struct *files,
263 char *tmp)
264 {
265 struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
266
267 lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
268 }
269
270 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
271 static
272 int lttng_enumerate_cpu_topology(struct lttng_session *session)
273 {
274 int cpu;
275 const cpumask_t *cpumask = cpu_possible_mask;
276
277 for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
278 cpu = cpumask_next(cpu, cpumask)) {
279 trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
280 }
281
282 return 0;
283 }
284 #else
285 static
286 int lttng_enumerate_cpu_topology(struct lttng_session *session)
287 {
288 return 0;
289 }
290 #endif
291
292 #if 0
293 /*
294 * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
295 * (scheduling in atomic). Normally, the tasklist lock protects this kind of
296 * iteration, but it is not exported to modules.
297 */
298 static
299 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
300 struct task_struct *p)
301 {
302 struct mm_struct *mm;
303 struct vm_area_struct *map;
304 unsigned long ino;
305
306 /* get_task_mm does a task_lock... */
307 mm = get_task_mm(p);
308 if (!mm)
309 return;
310
311 map = mm->mmap;
312 if (map) {
313 down_read(&mm->mmap_sem);
314 while (map) {
315 if (map->vm_file)
316 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
317 else
318 ino = 0;
319 trace_lttng_statedump_vm_map(session, p, map, ino);
320 map = map->vm_next;
321 }
322 up_read(&mm->mmap_sem);
323 }
324 mmput(mm);
325 }
326
327 static
328 int lttng_enumerate_vm_maps(struct lttng_session *session)
329 {
330 struct task_struct *p;
331
332 rcu_read_lock();
333 for_each_process(p)
334 lttng_enumerate_task_vm_maps(session, p);
335 rcu_read_unlock();
336 return 0;
337 }
338 #endif
339
340 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
341
342 static
343 int lttng_list_interrupts(struct lttng_session *session)
344 {
345 unsigned int irq;
346 unsigned long flags = 0;
347 struct irq_desc *desc;
348
349 #define irq_to_desc wrapper_irq_to_desc
350 /* needs irq_desc */
351 for_each_irq_desc(irq, desc) {
352 struct irqaction *action;
353 const char *irq_chip_name =
354 irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
355
356 local_irq_save(flags);
357 raw_spin_lock(&desc->lock);
358 for (action = desc->action; action; action = action->next) {
359 trace_lttng_statedump_interrupt(session,
360 irq, irq_chip_name, action);
361 }
362 raw_spin_unlock(&desc->lock);
363 local_irq_restore(flags);
364 }
365 return 0;
366 #undef irq_to_desc
367 }
368 #else
369 static inline
370 int lttng_list_interrupts(struct lttng_session *session)
371 {
372 return 0;
373 }
374 #endif
375
376 /*
377 * Statedump the task's namespaces using the proc filesystem inode number as
378 * the unique identifier. The user and pid ns are nested and will be dumped
379 * recursively.
380 *
381 * Called with task lock held.
382 */
383 static
384 void lttng_statedump_process_ns(struct lttng_session *session,
385 struct task_struct *p,
386 enum lttng_thread_type type,
387 enum lttng_execution_mode mode,
388 enum lttng_execution_submode submode,
389 enum lttng_process_status status)
390 {
391 struct nsproxy *proxy;
392 struct pid_namespace *pid_ns;
393 struct user_namespace *user_ns;
394
395 /*
396 * The pid and user namespaces are special, they are nested and
397 * accessed with specific functions instead of the nsproxy struct
398 * like the other namespaces.
399 */
400 pid_ns = task_active_pid_ns(p);
401 do {
402 trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
403 pid_ns = pid_ns ? pid_ns->parent : NULL;
404 } while (pid_ns);
405
406
407 user_ns = task_cred_xxx(p, user_ns);
408 do {
409 trace_lttng_statedump_process_user_ns(session, p, user_ns);
410 /*
411 * trace_lttng_statedump_process_user_ns() internally
412 * checks whether user_ns is NULL. While this does not
413 * appear to be a possible return value for
414 * task_cred_xxx(), err on the safe side and check
415 * for NULL here as well to be consistent with the
416 * paranoid behavior of
417 * trace_lttng_statedump_process_user_ns().
418 */
419 user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
420 } while (user_ns);
421
422 /*
423 * Back and forth on locking strategy within Linux upstream for nsproxy.
424 * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
425 * "namespaces: Use task_lock and not rcu to protect nsproxy"
426 * for details.
427 */
428 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
429 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
430 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
431 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
432 proxy = p->nsproxy;
433 #else
434 rcu_read_lock();
435 proxy = task_nsproxy(p);
436 #endif
437 if (proxy) {
438 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
439 trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
440 #endif
441 trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
442 #ifndef LTTNG_MNT_NS_MISSING_HEADER
443 trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
444 #endif
445 trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
446 trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
447 }
448 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
449 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
450 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
451 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
452 /* (nothing) */
453 #else
454 rcu_read_unlock();
455 #endif
456 }
457
458 static
459 int lttng_enumerate_process_states(struct lttng_session *session)
460 {
461 struct task_struct *g, *p;
462 char *tmp;
463
464 tmp = (char *) __get_free_page(GFP_KERNEL);
465 if (!tmp)
466 return -ENOMEM;
467
468 rcu_read_lock();
469 for_each_process(g) {
470 struct files_struct *prev_files = NULL;
471
472 p = g;
473 do {
474 enum lttng_execution_mode mode =
475 LTTNG_MODE_UNKNOWN;
476 enum lttng_execution_submode submode =
477 LTTNG_UNKNOWN;
478 enum lttng_process_status status;
479 enum lttng_thread_type type;
480 struct files_struct *files;
481
482 task_lock(p);
483 if (p->exit_state == EXIT_ZOMBIE)
484 status = LTTNG_ZOMBIE;
485 else if (p->exit_state == EXIT_DEAD)
486 status = LTTNG_DEAD;
487 else if (p->state == TASK_RUNNING) {
488 /* Is this a forked child that has not run yet? */
489 if (list_empty(&p->rt.run_list))
490 status = LTTNG_WAIT_FORK;
491 else
492 /*
493 * All tasks are considered as wait_cpu;
494 * the viewer will sort out if the task
495 * was really running at this time.
496 */
497 status = LTTNG_WAIT_CPU;
498 } else if (p->state &
499 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
500 /* Task is waiting for something to complete */
501 status = LTTNG_WAIT;
502 } else
503 status = LTTNG_UNNAMED;
504 submode = LTTNG_NONE;
505
506 /*
507 * Verification of t->mm is to filter out kernel
508 * threads; Viewer will further filter out if a
509 * user-space thread was in syscall mode or not.
510 */
511 if (p->mm)
512 type = LTTNG_USER_THREAD;
513 else
514 type = LTTNG_KERNEL_THREAD;
515 files = p->files;
516
517 trace_lttng_statedump_process_state(session,
518 p, type, mode, submode, status, files);
519 lttng_statedump_process_ns(session,
520 p, type, mode, submode, status);
521 /*
522 * As an optimisation for the common case, do not
523 * repeat information for the same files_struct in
524 * two consecutive threads. This is the common case
525 * for threads sharing the same fd table. RCU guarantees
526 * that the same files_struct pointer is not re-used
527 * throughout processes/threads iteration.
528 */
529 if (files && files != prev_files) {
530 lttng_enumerate_files(session, files, tmp);
531 prev_files = files;
532 }
533 task_unlock(p);
534 } while_each_thread(g, p);
535 }
536 rcu_read_unlock();
537
538 free_page((unsigned long) tmp);
539
540 return 0;
541 }
542
543 static
544 void lttng_statedump_work_func(struct work_struct *work)
545 {
546 if (atomic_dec_and_test(&kernel_threads_to_run))
547 /* If we are the last thread, wake up do_lttng_statedump */
548 wake_up(&statedump_wq);
549 }
550
551 static
552 int do_lttng_statedump(struct lttng_session *session)
553 {
554 int cpu, ret;
555
556 trace_lttng_statedump_start(session);
557 ret = lttng_enumerate_process_states(session);
558 if (ret)
559 return ret;
560 /*
561 * FIXME
562 * ret = lttng_enumerate_vm_maps(session);
563 * if (ret)
564 * return ret;
565 */
566 ret = lttng_list_interrupts(session);
567 if (ret)
568 return ret;
569 ret = lttng_enumerate_network_ip_interface(session);
570 if (ret)
571 return ret;
572 ret = lttng_enumerate_block_devices(session);
573 switch (ret) {
574 case 0:
575 break;
576 case -ENOSYS:
577 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
578 break;
579 default:
580 return ret;
581 }
582 ret = lttng_enumerate_cpu_topology(session);
583 if (ret)
584 return ret;
585
586 /* TODO lttng_dump_idt_table(session); */
587 /* TODO lttng_dump_softirq_vec(session); */
588 /* TODO lttng_list_modules(session); */
589 /* TODO lttng_dump_swap_files(session); */
590
591 /*
592 * Fire off a work queue on each CPU. Their sole purpose in life
593 * is to guarantee that each CPU has been in a state where is was in
594 * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
595 */
596 get_online_cpus();
597 atomic_set(&kernel_threads_to_run, num_online_cpus());
598 for_each_online_cpu(cpu) {
599 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
600 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
601 }
602 /* Wait for all threads to run */
603 __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
604 put_online_cpus();
605 /* Our work is done */
606 trace_lttng_statedump_end(session);
607 return 0;
608 }
609
610 /*
611 * Called with session mutex held.
612 */
613 int lttng_statedump_start(struct lttng_session *session)
614 {
615 return do_lttng_statedump(session);
616 }
617 EXPORT_SYMBOL_GPL(lttng_statedump_start);
618
619 static
620 int __init lttng_statedump_init(void)
621 {
622 /*
623 * Allow module to load even if the fixup cannot be done. This
624 * will allow seemless transition when the underlying issue fix
625 * is merged into the Linux kernel, and when tracepoint.c
626 * "tracepoint_module_notify" is turned into a static function.
627 */
628 (void) wrapper_lttng_fixup_sig(THIS_MODULE);
629 return 0;
630 }
631
632 module_init(lttng_statedump_init);
633
634 static
635 void __exit lttng_statedump_exit(void)
636 {
637 }
638
639 module_exit(lttng_statedump_exit);
640
641 MODULE_LICENSE("GPL and additional rights");
642 MODULE_AUTHOR("Jean-Hugues Deschenes");
643 MODULE_DESCRIPTION("LTTng statedump provider");
644 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
645 __stringify(LTTNG_MODULES_MINOR_VERSION) "."
646 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
647 LTTNG_MODULES_EXTRAVERSION);
This page took 0.04864 seconds and 5 git commands to generate.