Commit | Line | Data |
---|---|---|
b7cdc182 | 1 | /* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only) |
9f36eaed | 2 | * |
886d51a3 MD |
3 | * lttng-statedump.c |
4 | * | |
c337ddc2 MD |
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 | |
c337ddc2 MD |
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> | |
c337ddc2 MD |
31 | #include <linux/swap.h> |
32 | #include <linux/wait.h> | |
33 | #include <linux/mutex.h> | |
f0dbdefb | 34 | #include <linux/device.h> |
997eff5e | 35 | #include <linux/fdtable.h> |
c337ddc2 | 36 | |
241ae9a8 MD |
37 | #include <lttng-events.h> |
38 | #include <lttng-tracer.h> | |
1965e6b4 | 39 | #include <wrapper/namespace.h> |
241ae9a8 MD |
40 | #include <wrapper/irq.h> |
41 | #include <wrapper/tracepoint.h> | |
42 | #include <wrapper/genhd.h> | |
43 | #include <wrapper/file.h> | |
c337ddc2 | 44 | |
29784493 | 45 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
c337ddc2 MD |
46 | #include <linux/irq.h> |
47 | #endif | |
48 | ||
49 | /* Define the tracepoints, but do not build the probes */ | |
50 | #define CREATE_TRACE_POINTS | |
241ae9a8 | 51 | #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module |
c337ddc2 | 52 | #define TRACE_INCLUDE_FILE lttng-statedump |
3bc29f0a | 53 | #define LTTNG_INSTRUMENTATION |
241ae9a8 | 54 | #include <instrumentation/events/lttng-module/lttng-statedump.h> |
c337ddc2 | 55 | |
f0dbdefb | 56 | DEFINE_TRACE(lttng_statedump_block_device); |
20591cf7 MD |
57 | DEFINE_TRACE(lttng_statedump_end); |
58 | DEFINE_TRACE(lttng_statedump_interrupt); | |
59 | DEFINE_TRACE(lttng_statedump_file_descriptor); | |
60 | DEFINE_TRACE(lttng_statedump_start); | |
61 | DEFINE_TRACE(lttng_statedump_process_state); | |
1965e6b4 MJ |
62 | DEFINE_TRACE(lttng_statedump_process_pid_ns); |
63 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0)) | |
64 | DEFINE_TRACE(lttng_statedump_process_cgroup_ns); | |
65 | #endif | |
66 | DEFINE_TRACE(lttng_statedump_process_ipc_ns); | |
67 | #ifndef LTTNG_MNT_NS_MISSING_HEADER | |
68 | DEFINE_TRACE(lttng_statedump_process_mnt_ns); | |
69 | #endif | |
70 | DEFINE_TRACE(lttng_statedump_process_net_ns); | |
71 | DEFINE_TRACE(lttng_statedump_process_user_ns); | |
72 | DEFINE_TRACE(lttng_statedump_process_uts_ns); | |
20591cf7 | 73 | DEFINE_TRACE(lttng_statedump_network_interface); |
d0b55e4c | 74 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
502e4132 JD |
75 | DEFINE_TRACE(lttng_statedump_cpu_topology); |
76 | #endif | |
20591cf7 | 77 | |
361c023a MD |
78 | struct lttng_fd_ctx { |
79 | char *page; | |
80 | struct lttng_session *session; | |
d561ecfb | 81 | struct files_struct *files; |
361c023a MD |
82 | }; |
83 | ||
c337ddc2 MD |
84 | /* |
85 | * Protected by the trace lock. | |
86 | */ | |
87 | static struct delayed_work cpu_work[NR_CPUS]; | |
88 | static DECLARE_WAIT_QUEUE_HEAD(statedump_wq); | |
89 | static atomic_t kernel_threads_to_run; | |
90 | ||
91 | enum lttng_thread_type { | |
92 | LTTNG_USER_THREAD = 0, | |
93 | LTTNG_KERNEL_THREAD = 1, | |
94 | }; | |
95 | ||
96 | enum lttng_execution_mode { | |
97 | LTTNG_USER_MODE = 0, | |
98 | LTTNG_SYSCALL = 1, | |
99 | LTTNG_TRAP = 2, | |
100 | LTTNG_IRQ = 3, | |
101 | LTTNG_SOFTIRQ = 4, | |
102 | LTTNG_MODE_UNKNOWN = 5, | |
103 | }; | |
104 | ||
105 | enum lttng_execution_submode { | |
106 | LTTNG_NONE = 0, | |
107 | LTTNG_UNKNOWN = 1, | |
108 | }; | |
109 | ||
110 | enum lttng_process_status { | |
111 | LTTNG_UNNAMED = 0, | |
112 | LTTNG_WAIT_FORK = 1, | |
113 | LTTNG_WAIT_CPU = 2, | |
114 | LTTNG_EXIT = 3, | |
115 | LTTNG_ZOMBIE = 4, | |
116 | LTTNG_WAIT = 5, | |
117 | LTTNG_RUN = 6, | |
118 | LTTNG_DEAD = 7, | |
119 | }; | |
120 | ||
f0dbdefb HD |
121 | static |
122 | int lttng_enumerate_block_devices(struct lttng_session *session) | |
123 | { | |
124 | struct class *ptr_block_class; | |
125 | struct device_type *ptr_disk_type; | |
126 | struct class_dev_iter iter; | |
127 | struct device *dev; | |
128 | ||
129 | ptr_block_class = wrapper_get_block_class(); | |
130 | if (!ptr_block_class) | |
131 | return -ENOSYS; | |
132 | ptr_disk_type = wrapper_get_disk_type(); | |
133 | if (!ptr_disk_type) { | |
134 | return -ENOSYS; | |
135 | } | |
136 | class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type); | |
137 | while ((dev = class_dev_iter_next(&iter))) { | |
138 | struct disk_part_iter piter; | |
139 | struct gendisk *disk = dev_to_disk(dev); | |
140 | struct hd_struct *part; | |
141 | ||
5a91f3df MD |
142 | /* |
143 | * Don't show empty devices or things that have been | |
144 | * suppressed | |
145 | */ | |
146 | if (get_capacity(disk) == 0 || | |
147 | (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) | |
148 | continue; | |
149 | ||
f0dbdefb HD |
150 | disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0); |
151 | while ((part = disk_part_iter_next(&piter))) { | |
152 | char name_buf[BDEVNAME_SIZE]; | |
153 | char *p; | |
154 | ||
155 | p = wrapper_disk_name(disk, part->partno, name_buf); | |
156 | if (!p) { | |
157 | disk_part_iter_exit(&piter); | |
158 | class_dev_iter_exit(&iter); | |
159 | return -ENOSYS; | |
160 | } | |
161 | trace_lttng_statedump_block_device(session, | |
162 | part_devt(part), name_buf); | |
163 | } | |
164 | disk_part_iter_exit(&piter); | |
165 | } | |
166 | class_dev_iter_exit(&iter); | |
167 | return 0; | |
168 | } | |
169 | ||
c337ddc2 | 170 | #ifdef CONFIG_INET |
f0dbdefb | 171 | |
c337ddc2 MD |
172 | static |
173 | void lttng_enumerate_device(struct lttng_session *session, | |
174 | struct net_device *dev) | |
175 | { | |
176 | struct in_device *in_dev; | |
177 | struct in_ifaddr *ifa; | |
178 | ||
179 | if (dev->flags & IFF_UP) { | |
180 | in_dev = in_dev_get(dev); | |
181 | if (in_dev) { | |
182 | for (ifa = in_dev->ifa_list; ifa != NULL; | |
183 | ifa = ifa->ifa_next) { | |
184 | trace_lttng_statedump_network_interface( | |
185 | session, dev, ifa); | |
186 | } | |
187 | in_dev_put(in_dev); | |
188 | } | |
189 | } else { | |
190 | trace_lttng_statedump_network_interface( | |
191 | session, dev, NULL); | |
192 | } | |
193 | } | |
194 | ||
195 | static | |
196 | int lttng_enumerate_network_ip_interface(struct lttng_session *session) | |
197 | { | |
198 | struct net_device *dev; | |
199 | ||
200 | read_lock(&dev_base_lock); | |
201 | for_each_netdev(&init_net, dev) | |
202 | lttng_enumerate_device(session, dev); | |
203 | read_unlock(&dev_base_lock); | |
204 | ||
205 | return 0; | |
206 | } | |
207 | #else /* CONFIG_INET */ | |
208 | static inline | |
209 | int lttng_enumerate_network_ip_interface(struct lttng_session *session) | |
210 | { | |
211 | return 0; | |
212 | } | |
213 | #endif /* CONFIG_INET */ | |
214 | ||
361c023a MD |
215 | static |
216 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) | |
217 | { | |
218 | const struct lttng_fd_ctx *ctx = p; | |
219 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); | |
29021503 | 220 | unsigned int flags = file->f_flags; |
d561ecfb | 221 | struct fdtable *fdt; |
361c023a | 222 | |
29021503 MD |
223 | /* |
224 | * We don't expose kernel internal flags, only userspace-visible | |
225 | * flags. | |
226 | */ | |
227 | flags &= ~FMODE_NONOTIFY; | |
d561ecfb MD |
228 | fdt = files_fdtable(ctx->files); |
229 | /* | |
230 | * We need to check here again whether fd is within the fdt | |
231 | * max_fds range, because we might be seeing a different | |
232 | * files_fdtable() than iterate_fd(), assuming only RCU is | |
233 | * protecting the read. In reality, iterate_fd() holds | |
234 | * file_lock, which should ensure the fdt does not change while | |
235 | * the lock is taken, but we are not aware whether this is | |
236 | * guaranteed or not, so play safe. | |
237 | */ | |
997eff5e | 238 | if (fd < fdt->max_fds && close_on_exec(fd, fdt)) |
29021503 | 239 | flags |= O_CLOEXEC; |
361c023a MD |
240 | if (IS_ERR(s)) { |
241 | struct dentry *dentry = file->f_path.dentry; | |
242 | ||
243 | /* Make sure we give at least some info */ | |
244 | spin_lock(&dentry->d_lock); | |
e7a0ca72 MD |
245 | trace_lttng_statedump_file_descriptor(ctx->session, |
246 | ctx->files, fd, dentry->d_name.name, flags, | |
247 | file->f_mode); | |
361c023a MD |
248 | spin_unlock(&dentry->d_lock); |
249 | goto end; | |
250 | } | |
e7a0ca72 MD |
251 | trace_lttng_statedump_file_descriptor(ctx->session, |
252 | ctx->files, fd, s, flags, file->f_mode); | |
361c023a MD |
253 | end: |
254 | return 0; | |
255 | } | |
c337ddc2 | 256 | |
e7a0ca72 | 257 | /* Called with task lock held. */ |
c337ddc2 | 258 | static |
e7a0ca72 MD |
259 | void lttng_enumerate_files(struct lttng_session *session, |
260 | struct files_struct *files, | |
261 | char *tmp) | |
c337ddc2 | 262 | { |
e7a0ca72 | 263 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, }; |
c337ddc2 | 264 | |
997eff5e | 265 | iterate_fd(files, 0, lttng_dump_one_fd, &ctx); |
c337ddc2 MD |
266 | } |
267 | ||
d0b55e4c | 268 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
502e4132 JD |
269 | static |
270 | int lttng_enumerate_cpu_topology(struct lttng_session *session) | |
271 | { | |
272 | int cpu; | |
273 | const cpumask_t *cpumask = cpu_possible_mask; | |
274 | ||
275 | for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids; | |
276 | cpu = cpumask_next(cpu, cpumask)) { | |
277 | trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu)); | |
278 | } | |
279 | ||
280 | return 0; | |
281 | } | |
282 | #else | |
283 | static | |
284 | int lttng_enumerate_cpu_topology(struct lttng_session *session) | |
285 | { | |
286 | return 0; | |
287 | } | |
288 | #endif | |
289 | ||
0658bdda MD |
290 | #if 0 |
291 | /* | |
292 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section | |
293 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of | |
294 | * iteration, but it is not exported to modules. | |
295 | */ | |
c337ddc2 MD |
296 | static |
297 | void lttng_enumerate_task_vm_maps(struct lttng_session *session, | |
298 | struct task_struct *p) | |
299 | { | |
300 | struct mm_struct *mm; | |
301 | struct vm_area_struct *map; | |
302 | unsigned long ino; | |
303 | ||
304 | /* get_task_mm does a task_lock... */ | |
305 | mm = get_task_mm(p); | |
306 | if (!mm) | |
307 | return; | |
308 | ||
309 | map = mm->mmap; | |
310 | if (map) { | |
311 | down_read(&mm->mmap_sem); | |
312 | while (map) { | |
313 | if (map->vm_file) | |
b06ed645 | 314 | ino = map->vm_file->lttng_f_dentry->d_inode->i_ino; |
c337ddc2 MD |
315 | else |
316 | ino = 0; | |
317 | trace_lttng_statedump_vm_map(session, p, map, ino); | |
318 | map = map->vm_next; | |
319 | } | |
320 | up_read(&mm->mmap_sem); | |
321 | } | |
322 | mmput(mm); | |
323 | } | |
324 | ||
325 | static | |
326 | int lttng_enumerate_vm_maps(struct lttng_session *session) | |
327 | { | |
328 | struct task_struct *p; | |
329 | ||
330 | rcu_read_lock(); | |
331 | for_each_process(p) | |
332 | lttng_enumerate_task_vm_maps(session, p); | |
333 | rcu_read_unlock(); | |
334 | return 0; | |
335 | } | |
0658bdda | 336 | #endif |
c337ddc2 | 337 | |
29784493 | 338 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
47faec4b | 339 | |
c337ddc2 | 340 | static |
cfcee1c7 | 341 | int lttng_list_interrupts(struct lttng_session *session) |
c337ddc2 MD |
342 | { |
343 | unsigned int irq; | |
344 | unsigned long flags = 0; | |
345 | struct irq_desc *desc; | |
346 | ||
c337ddc2 MD |
347 | /* needs irq_desc */ |
348 | for_each_irq_desc(irq, desc) { | |
349 | struct irqaction *action; | |
350 | const char *irq_chip_name = | |
351 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; | |
352 | ||
353 | local_irq_save(flags); | |
fc94c945 | 354 | raw_spin_lock(&desc->lock); |
c337ddc2 MD |
355 | for (action = desc->action; action; action = action->next) { |
356 | trace_lttng_statedump_interrupt(session, | |
357 | irq, irq_chip_name, action); | |
358 | } | |
fc94c945 | 359 | raw_spin_unlock(&desc->lock); |
c337ddc2 MD |
360 | local_irq_restore(flags); |
361 | } | |
cfcee1c7 | 362 | return 0; |
c337ddc2 MD |
363 | } |
364 | #else | |
365 | static inline | |
cfcee1c7 | 366 | int lttng_list_interrupts(struct lttng_session *session) |
c337ddc2 | 367 | { |
cfcee1c7 | 368 | return 0; |
c337ddc2 MD |
369 | } |
370 | #endif | |
371 | ||
4ba1f53c | 372 | /* |
1965e6b4 MJ |
373 | * Statedump the task's namespaces using the proc filesystem inode number as |
374 | * the unique identifier. The user and pid ns are nested and will be dumped | |
375 | * recursively. | |
376 | * | |
4ba1f53c MD |
377 | * Called with task lock held. |
378 | */ | |
73e8ba37 JD |
379 | static |
380 | void lttng_statedump_process_ns(struct lttng_session *session, | |
381 | struct task_struct *p, | |
382 | enum lttng_thread_type type, | |
383 | enum lttng_execution_mode mode, | |
384 | enum lttng_execution_submode submode, | |
385 | enum lttng_process_status status) | |
386 | { | |
1965e6b4 | 387 | struct nsproxy *proxy; |
73e8ba37 | 388 | struct pid_namespace *pid_ns; |
1965e6b4 | 389 | struct user_namespace *user_ns; |
73e8ba37 | 390 | |
1965e6b4 MJ |
391 | /* |
392 | * The pid and user namespaces are special, they are nested and | |
393 | * accessed with specific functions instead of the nsproxy struct | |
394 | * like the other namespaces. | |
395 | */ | |
887bcdac MJ |
396 | pid_ns = task_active_pid_ns(p); |
397 | do { | |
1965e6b4 | 398 | trace_lttng_statedump_process_pid_ns(session, p, pid_ns); |
adcc8b5e | 399 | pid_ns = pid_ns ? pid_ns->parent : NULL; |
887bcdac | 400 | } while (pid_ns); |
1965e6b4 MJ |
401 | |
402 | ||
403 | user_ns = task_cred_xxx(p, user_ns); | |
404 | do { | |
405 | trace_lttng_statedump_process_user_ns(session, p, user_ns); | |
1964cccb MD |
406 | /* |
407 | * trace_lttng_statedump_process_user_ns() internally | |
408 | * checks whether user_ns is NULL. While this does not | |
409 | * appear to be a possible return value for | |
410 | * task_cred_xxx(), err on the safe side and check | |
411 | * for NULL here as well to be consistent with the | |
412 | * paranoid behavior of | |
413 | * trace_lttng_statedump_process_user_ns(). | |
414 | */ | |
415 | user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL; | |
1965e6b4 MJ |
416 | } while (user_ns); |
417 | ||
418 | /* | |
419 | * Back and forth on locking strategy within Linux upstream for nsproxy. | |
420 | * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3 | |
421 | * "namespaces: Use task_lock and not rcu to protect nsproxy" | |
422 | * for details. | |
423 | */ | |
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)) | |
428 | proxy = p->nsproxy; | |
429 | #else | |
430 | rcu_read_lock(); | |
431 | proxy = task_nsproxy(p); | |
432 | #endif | |
433 | if (proxy) { | |
434 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0)) | |
435 | trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns); | |
436 | #endif | |
437 | trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns); | |
438 | #ifndef LTTNG_MNT_NS_MISSING_HEADER | |
439 | trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns); | |
440 | #endif | |
441 | trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns); | |
442 | trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns); | |
443 | } | |
444 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \ | |
445 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ | |
446 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
447 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
448 | /* (nothing) */ | |
449 | #else | |
450 | rcu_read_unlock(); | |
451 | #endif | |
73e8ba37 JD |
452 | } |
453 | ||
c337ddc2 MD |
454 | static |
455 | int lttng_enumerate_process_states(struct lttng_session *session) | |
456 | { | |
457 | struct task_struct *g, *p; | |
e7a0ca72 MD |
458 | char *tmp; |
459 | ||
460 | tmp = (char *) __get_free_page(GFP_KERNEL); | |
461 | if (!tmp) | |
462 | return -ENOMEM; | |
c337ddc2 MD |
463 | |
464 | rcu_read_lock(); | |
465 | for_each_process(g) { | |
e7a0ca72 MD |
466 | struct files_struct *prev_files = NULL; |
467 | ||
c337ddc2 MD |
468 | p = g; |
469 | do { | |
470 | enum lttng_execution_mode mode = | |
471 | LTTNG_MODE_UNKNOWN; | |
472 | enum lttng_execution_submode submode = | |
473 | LTTNG_UNKNOWN; | |
474 | enum lttng_process_status status; | |
475 | enum lttng_thread_type type; | |
e7a0ca72 | 476 | struct files_struct *files; |
c337ddc2 MD |
477 | |
478 | task_lock(p); | |
479 | if (p->exit_state == EXIT_ZOMBIE) | |
480 | status = LTTNG_ZOMBIE; | |
481 | else if (p->exit_state == EXIT_DEAD) | |
482 | status = LTTNG_DEAD; | |
483 | else if (p->state == TASK_RUNNING) { | |
484 | /* Is this a forked child that has not run yet? */ | |
485 | if (list_empty(&p->rt.run_list)) | |
486 | status = LTTNG_WAIT_FORK; | |
487 | else | |
488 | /* | |
489 | * All tasks are considered as wait_cpu; | |
490 | * the viewer will sort out if the task | |
491 | * was really running at this time. | |
492 | */ | |
493 | status = LTTNG_WAIT_CPU; | |
494 | } else if (p->state & | |
495 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { | |
496 | /* Task is waiting for something to complete */ | |
497 | status = LTTNG_WAIT; | |
498 | } else | |
499 | status = LTTNG_UNNAMED; | |
500 | submode = LTTNG_NONE; | |
501 | ||
502 | /* | |
503 | * Verification of t->mm is to filter out kernel | |
504 | * threads; Viewer will further filter out if a | |
505 | * user-space thread was in syscall mode or not. | |
506 | */ | |
507 | if (p->mm) | |
508 | type = LTTNG_USER_THREAD; | |
509 | else | |
510 | type = LTTNG_KERNEL_THREAD; | |
e7a0ca72 | 511 | files = p->files; |
d2a927ac MJ |
512 | |
513 | trace_lttng_statedump_process_state(session, | |
e7a0ca72 | 514 | p, type, mode, submode, status, files); |
73e8ba37 | 515 | lttng_statedump_process_ns(session, |
c337ddc2 | 516 | p, type, mode, submode, status); |
e7a0ca72 MD |
517 | /* |
518 | * As an optimisation for the common case, do not | |
519 | * repeat information for the same files_struct in | |
520 | * two consecutive threads. This is the common case | |
521 | * for threads sharing the same fd table. RCU guarantees | |
522 | * that the same files_struct pointer is not re-used | |
523 | * throughout processes/threads iteration. | |
524 | */ | |
525 | if (files && files != prev_files) { | |
526 | lttng_enumerate_files(session, files, tmp); | |
527 | prev_files = files; | |
528 | } | |
c337ddc2 MD |
529 | task_unlock(p); |
530 | } while_each_thread(g, p); | |
531 | } | |
532 | rcu_read_unlock(); | |
533 | ||
e7a0ca72 MD |
534 | free_page((unsigned long) tmp); |
535 | ||
c337ddc2 MD |
536 | return 0; |
537 | } | |
538 | ||
539 | static | |
540 | void lttng_statedump_work_func(struct work_struct *work) | |
541 | { | |
542 | if (atomic_dec_and_test(&kernel_threads_to_run)) | |
543 | /* If we are the last thread, wake up do_lttng_statedump */ | |
544 | wake_up(&statedump_wq); | |
545 | } | |
546 | ||
547 | static | |
548 | int do_lttng_statedump(struct lttng_session *session) | |
549 | { | |
cfcee1c7 | 550 | int cpu, ret; |
c337ddc2 | 551 | |
c337ddc2 | 552 | trace_lttng_statedump_start(session); |
cfcee1c7 | 553 | ret = lttng_enumerate_process_states(session); |
cfcee1c7 MD |
554 | if (ret) |
555 | return ret; | |
556 | /* | |
557 | * FIXME | |
558 | * ret = lttng_enumerate_vm_maps(session); | |
559 | * if (ret) | |
560 | * return ret; | |
561 | */ | |
562 | ret = lttng_list_interrupts(session); | |
563 | if (ret) | |
564 | return ret; | |
565 | ret = lttng_enumerate_network_ip_interface(session); | |
566 | if (ret) | |
567 | return ret; | |
568 | ret = lttng_enumerate_block_devices(session); | |
569 | switch (ret) { | |
84c7055e MD |
570 | case 0: |
571 | break; | |
cfcee1c7 MD |
572 | case -ENOSYS: |
573 | printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n"); | |
574 | break; | |
575 | default: | |
576 | return ret; | |
577 | } | |
502e4132 JD |
578 | ret = lttng_enumerate_cpu_topology(session); |
579 | if (ret) | |
580 | return ret; | |
c337ddc2 MD |
581 | |
582 | /* TODO lttng_dump_idt_table(session); */ | |
583 | /* TODO lttng_dump_softirq_vec(session); */ | |
584 | /* TODO lttng_list_modules(session); */ | |
585 | /* TODO lttng_dump_swap_files(session); */ | |
586 | ||
587 | /* | |
588 | * Fire off a work queue on each CPU. Their sole purpose in life | |
589 | * is to guarantee that each CPU has been in a state where is was in | |
590 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). | |
591 | */ | |
592 | get_online_cpus(); | |
593 | atomic_set(&kernel_threads_to_run, num_online_cpus()); | |
594 | for_each_online_cpu(cpu) { | |
595 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); | |
596 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); | |
597 | } | |
598 | /* Wait for all threads to run */ | |
7a7128e0 | 599 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
c337ddc2 MD |
600 | put_online_cpus(); |
601 | /* Our work is done */ | |
c337ddc2 MD |
602 | trace_lttng_statedump_end(session); |
603 | return 0; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Called with session mutex held. | |
608 | */ | |
609 | int lttng_statedump_start(struct lttng_session *session) | |
610 | { | |
c337ddc2 MD |
611 | return do_lttng_statedump(session); |
612 | } | |
613 | EXPORT_SYMBOL_GPL(lttng_statedump_start); | |
614 | ||
dd8d5afb MD |
615 | static |
616 | int __init lttng_statedump_init(void) | |
617 | { | |
d16aa9c9 MD |
618 | /* |
619 | * Allow module to load even if the fixup cannot be done. This | |
620 | * will allow seemless transition when the underlying issue fix | |
621 | * is merged into the Linux kernel, and when tracepoint.c | |
622 | * "tracepoint_module_notify" is turned into a static function. | |
623 | */ | |
624 | (void) wrapper_lttng_fixup_sig(THIS_MODULE); | |
625 | return 0; | |
dd8d5afb MD |
626 | } |
627 | ||
628 | module_init(lttng_statedump_init); | |
629 | ||
461277e7 MD |
630 | static |
631 | void __exit lttng_statedump_exit(void) | |
632 | { | |
633 | } | |
634 | ||
635 | module_exit(lttng_statedump_exit); | |
636 | ||
c337ddc2 MD |
637 | MODULE_LICENSE("GPL and additional rights"); |
638 | MODULE_AUTHOR("Jean-Hugues Deschenes"); | |
1c124020 | 639 | MODULE_DESCRIPTION("LTTng statedump provider"); |
13ab8b0a MD |
640 | MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "." |
641 | __stringify(LTTNG_MODULES_MINOR_VERSION) "." | |
642 | __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION) | |
643 | LTTNG_MODULES_EXTRAVERSION); |