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> |
c337ddc2 | 35 | |
2df37e95 MD |
36 | #include <lttng/events.h> |
37 | #include <lttng/tracer.h> | |
241ae9a8 | 38 | #include <wrapper/irqdesc.h> |
241ae9a8 | 39 | #include <wrapper/fdtable.h> |
1965e6b4 | 40 | #include <wrapper/namespace.h> |
241ae9a8 MD |
41 | #include <wrapper/irq.h> |
42 | #include <wrapper/tracepoint.h> | |
43 | #include <wrapper/genhd.h> | |
44 | #include <wrapper/file.h> | |
c07dca48 | 45 | #include <wrapper/fdtable.h> |
c337ddc2 | 46 | |
29784493 | 47 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
c337ddc2 MD |
48 | #include <linux/irq.h> |
49 | #endif | |
50 | ||
51 | /* Define the tracepoints, but do not build the probes */ | |
52 | #define CREATE_TRACE_POINTS | |
4f47ccf0 | 53 | #define TRACE_INCLUDE_PATH instrumentation/events |
c337ddc2 | 54 | #define TRACE_INCLUDE_FILE lttng-statedump |
3bc29f0a | 55 | #define LTTNG_INSTRUMENTATION |
4f47ccf0 | 56 | #include <instrumentation/events/lttng-statedump.h> |
c337ddc2 | 57 | |
0116245f | 58 | LTTNG_DEFINE_TRACE(lttng_statedump_block_device, |
8cdc1a81 | 59 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
60 | dev_t dev, const char *diskname), |
61 | TP_ARGS(session, dev, diskname)); | |
62 | ||
63 | LTTNG_DEFINE_TRACE(lttng_statedump_end, | |
8cdc1a81 | 64 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
65 | TP_ARGS(session)); |
66 | ||
67 | LTTNG_DEFINE_TRACE(lttng_statedump_interrupt, | |
8cdc1a81 | 68 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
69 | unsigned int irq, const char *chip_name, |
70 | struct irqaction *action), | |
71 | TP_ARGS(session, irq, chip_name, action)); | |
72 | ||
73 | LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor, | |
8cdc1a81 | 74 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
75 | struct files_struct *files, |
76 | int fd, const char *filename, | |
77 | unsigned int flags, fmode_t fmode), | |
78 | TP_ARGS(session, files, fd, filename, flags, fmode)); | |
79 | ||
80 | LTTNG_DEFINE_TRACE(lttng_statedump_start, | |
8cdc1a81 | 81 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
82 | TP_ARGS(session)); |
83 | ||
84 | LTTNG_DEFINE_TRACE(lttng_statedump_process_state, | |
8cdc1a81 | 85 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
86 | struct task_struct *p, |
87 | int type, int mode, int submode, int status, | |
88 | struct files_struct *files), | |
89 | TP_ARGS(session, p, type, mode, submode, status, files)); | |
90 | ||
91 | LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns, | |
8cdc1a81 | 92 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
93 | struct task_struct *p, |
94 | struct pid_namespace *pid_ns), | |
95 | TP_ARGS(session, p, pid_ns)); | |
96 | ||
5f4c791e | 97 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
0116245f | 98 | LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns, |
8cdc1a81 | 99 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
100 | struct task_struct *p, |
101 | struct cgroup_namespace *cgroup_ns), | |
102 | TP_ARGS(session, p, cgroup_ns)); | |
1965e6b4 | 103 | #endif |
0116245f MJ |
104 | |
105 | LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns, | |
8cdc1a81 | 106 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
107 | struct task_struct *p, |
108 | struct ipc_namespace *ipc_ns), | |
109 | TP_ARGS(session, p, ipc_ns)); | |
110 | ||
1965e6b4 | 111 | #ifndef LTTNG_MNT_NS_MISSING_HEADER |
0116245f | 112 | LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns, |
8cdc1a81 | 113 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
114 | struct task_struct *p, |
115 | struct mnt_namespace *mnt_ns), | |
116 | TP_ARGS(session, p, mnt_ns)); | |
1965e6b4 | 117 | #endif |
0116245f MJ |
118 | |
119 | LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns, | |
8cdc1a81 | 120 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
121 | struct task_struct *p, |
122 | struct net *net_ns), | |
123 | TP_ARGS(session, p, net_ns)); | |
124 | ||
125 | LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns, | |
8cdc1a81 | 126 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
127 | struct task_struct *p, |
128 | struct user_namespace *user_ns), | |
129 | TP_ARGS(session, p, user_ns)); | |
130 | ||
131 | LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns, | |
8cdc1a81 | 132 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
133 | struct task_struct *p, |
134 | struct uts_namespace *uts_ns), | |
135 | TP_ARGS(session, p, uts_ns)); | |
136 | ||
137 | LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns, | |
8cdc1a81 | 138 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
139 | struct task_struct *p, |
140 | struct time_namespace *time_ns), | |
141 | TP_ARGS(session, p, time_ns)); | |
142 | ||
143 | LTTNG_DEFINE_TRACE(lttng_statedump_network_interface, | |
8cdc1a81 | 144 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
145 | struct net_device *dev, struct in_ifaddr *ifa), |
146 | TP_ARGS(session, dev, ifa)); | |
147 | ||
d0b55e4c | 148 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
0116245f | 149 | LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology, |
8cdc1a81 | 150 | TP_PROTO(struct lttng_kernel_session *session, struct cpuinfo_x86 *c), |
0116245f | 151 | TP_ARGS(session, c)); |
502e4132 | 152 | #endif |
20591cf7 | 153 | |
361c023a MD |
154 | struct lttng_fd_ctx { |
155 | char *page; | |
8cdc1a81 | 156 | struct lttng_kernel_session *session; |
d561ecfb | 157 | struct files_struct *files; |
361c023a MD |
158 | }; |
159 | ||
c337ddc2 MD |
160 | /* |
161 | * Protected by the trace lock. | |
162 | */ | |
163 | static struct delayed_work cpu_work[NR_CPUS]; | |
164 | static DECLARE_WAIT_QUEUE_HEAD(statedump_wq); | |
165 | static atomic_t kernel_threads_to_run; | |
166 | ||
167 | enum lttng_thread_type { | |
168 | LTTNG_USER_THREAD = 0, | |
169 | LTTNG_KERNEL_THREAD = 1, | |
170 | }; | |
171 | ||
172 | enum lttng_execution_mode { | |
173 | LTTNG_USER_MODE = 0, | |
174 | LTTNG_SYSCALL = 1, | |
175 | LTTNG_TRAP = 2, | |
176 | LTTNG_IRQ = 3, | |
177 | LTTNG_SOFTIRQ = 4, | |
178 | LTTNG_MODE_UNKNOWN = 5, | |
179 | }; | |
180 | ||
181 | enum lttng_execution_submode { | |
182 | LTTNG_NONE = 0, | |
183 | LTTNG_UNKNOWN = 1, | |
184 | }; | |
185 | ||
186 | enum lttng_process_status { | |
187 | LTTNG_UNNAMED = 0, | |
188 | LTTNG_WAIT_FORK = 1, | |
189 | LTTNG_WAIT_CPU = 2, | |
190 | LTTNG_EXIT = 3, | |
191 | LTTNG_ZOMBIE = 4, | |
192 | LTTNG_WAIT = 5, | |
193 | LTTNG_RUN = 6, | |
194 | LTTNG_DEAD = 7, | |
195 | }; | |
196 | ||
fef7fe6a | 197 | |
5f4c791e | 198 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0)) |
fef7fe6a MJ |
199 | |
200 | #define LTTNG_PART_STRUCT_TYPE struct block_device | |
201 | ||
202 | static | |
203 | int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf) | |
204 | { | |
205 | const char *p; | |
206 | ||
207 | p = bdevname(part, name_buf); | |
208 | if (!p) | |
209 | return -ENOSYS; | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
214 | static | |
215 | dev_t lttng_get_part_devt(struct block_device *part) | |
216 | { | |
217 | return part->bd_dev; | |
218 | } | |
219 | ||
220 | #else | |
221 | ||
222 | #define LTTNG_PART_STRUCT_TYPE struct hd_struct | |
223 | ||
224 | static | |
225 | int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf) | |
226 | { | |
227 | const char *p; | |
228 | struct block_device bdev; | |
229 | ||
230 | /* | |
231 | * Create a partial 'struct blockdevice' to use | |
232 | * 'bdevname()' which is a simple wrapper over | |
233 | * 'disk_name()' but has the honor to be EXPORT_SYMBOL. | |
234 | */ | |
235 | bdev.bd_disk = disk; | |
236 | bdev.bd_part = part; | |
237 | ||
238 | p = bdevname(&bdev, name_buf); | |
239 | if (!p) | |
240 | return -ENOSYS; | |
241 | ||
242 | return 0; | |
243 | } | |
244 | ||
245 | static | |
246 | dev_t lttng_get_part_devt(struct hd_struct *part) | |
247 | { | |
248 | return part_devt(part); | |
249 | } | |
250 | #endif | |
251 | ||
f0dbdefb | 252 | static |
8cdc1a81 | 253 | int lttng_enumerate_block_devices(struct lttng_kernel_session *session) |
f0dbdefb HD |
254 | { |
255 | struct class *ptr_block_class; | |
256 | struct device_type *ptr_disk_type; | |
257 | struct class_dev_iter iter; | |
258 | struct device *dev; | |
7633c773 | 259 | int ret = 0; |
f0dbdefb HD |
260 | |
261 | ptr_block_class = wrapper_get_block_class(); | |
7633c773 MJ |
262 | if (!ptr_block_class) { |
263 | ret = -ENOSYS; | |
264 | goto end; | |
265 | } | |
f0dbdefb HD |
266 | ptr_disk_type = wrapper_get_disk_type(); |
267 | if (!ptr_disk_type) { | |
7633c773 MJ |
268 | ret = -ENOSYS; |
269 | goto end; | |
f0dbdefb HD |
270 | } |
271 | class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type); | |
272 | while ((dev = class_dev_iter_next(&iter))) { | |
273 | struct disk_part_iter piter; | |
274 | struct gendisk *disk = dev_to_disk(dev); | |
fef7fe6a | 275 | LTTNG_PART_STRUCT_TYPE *part; |
f0dbdefb | 276 | |
5a91f3df MD |
277 | /* |
278 | * Don't show empty devices or things that have been | |
279 | * suppressed | |
280 | */ | |
281 | if (get_capacity(disk) == 0 || | |
282 | (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) | |
283 | continue; | |
284 | ||
7633c773 MJ |
285 | /* |
286 | * The original 'disk_part_iter_init' returns void, but our | |
287 | * wrapper can fail to lookup the original symbol. | |
288 | */ | |
289 | if (wrapper_disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0) < 0) { | |
290 | ret = -ENOSYS; | |
291 | goto iter_exit; | |
292 | } | |
293 | ||
294 | while ((part = wrapper_disk_part_iter_next(&piter))) { | |
f0dbdefb | 295 | char name_buf[BDEVNAME_SIZE]; |
fef7fe6a MJ |
296 | |
297 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { | |
7633c773 MJ |
298 | wrapper_disk_part_iter_exit(&piter); |
299 | ret = -ENOSYS; | |
300 | goto iter_exit; | |
f0dbdefb HD |
301 | } |
302 | trace_lttng_statedump_block_device(session, | |
fef7fe6a | 303 | lttng_get_part_devt(part), name_buf); |
f0dbdefb | 304 | } |
7633c773 | 305 | wrapper_disk_part_iter_exit(&piter); |
f0dbdefb | 306 | } |
7633c773 | 307 | iter_exit: |
f0dbdefb | 308 | class_dev_iter_exit(&iter); |
7633c773 MJ |
309 | end: |
310 | return ret; | |
f0dbdefb HD |
311 | } |
312 | ||
c337ddc2 | 313 | #ifdef CONFIG_INET |
f0dbdefb | 314 | |
c337ddc2 | 315 | static |
8cdc1a81 | 316 | void lttng_enumerate_device(struct lttng_kernel_session *session, |
c337ddc2 MD |
317 | struct net_device *dev) |
318 | { | |
319 | struct in_device *in_dev; | |
320 | struct in_ifaddr *ifa; | |
321 | ||
322 | if (dev->flags & IFF_UP) { | |
323 | in_dev = in_dev_get(dev); | |
324 | if (in_dev) { | |
325 | for (ifa = in_dev->ifa_list; ifa != NULL; | |
326 | ifa = ifa->ifa_next) { | |
327 | trace_lttng_statedump_network_interface( | |
328 | session, dev, ifa); | |
329 | } | |
330 | in_dev_put(in_dev); | |
331 | } | |
332 | } else { | |
333 | trace_lttng_statedump_network_interface( | |
334 | session, dev, NULL); | |
335 | } | |
336 | } | |
337 | ||
338 | static | |
8cdc1a81 | 339 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
340 | { |
341 | struct net_device *dev; | |
342 | ||
343 | read_lock(&dev_base_lock); | |
344 | for_each_netdev(&init_net, dev) | |
345 | lttng_enumerate_device(session, dev); | |
346 | read_unlock(&dev_base_lock); | |
347 | ||
348 | return 0; | |
349 | } | |
350 | #else /* CONFIG_INET */ | |
351 | static inline | |
8cdc1a81 | 352 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
353 | { |
354 | return 0; | |
355 | } | |
356 | #endif /* CONFIG_INET */ | |
357 | ||
361c023a MD |
358 | static |
359 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) | |
360 | { | |
361 | const struct lttng_fd_ctx *ctx = p; | |
362 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); | |
29021503 | 363 | unsigned int flags = file->f_flags; |
d561ecfb | 364 | struct fdtable *fdt; |
361c023a | 365 | |
29021503 MD |
366 | /* |
367 | * We don't expose kernel internal flags, only userspace-visible | |
368 | * flags. | |
369 | */ | |
370 | flags &= ~FMODE_NONOTIFY; | |
d561ecfb MD |
371 | fdt = files_fdtable(ctx->files); |
372 | /* | |
373 | * We need to check here again whether fd is within the fdt | |
374 | * max_fds range, because we might be seeing a different | |
375 | * files_fdtable() than iterate_fd(), assuming only RCU is | |
376 | * protecting the read. In reality, iterate_fd() holds | |
377 | * file_lock, which should ensure the fdt does not change while | |
378 | * the lock is taken, but we are not aware whether this is | |
379 | * guaranteed or not, so play safe. | |
380 | */ | |
aa29f2d3 | 381 | if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt)) |
29021503 | 382 | flags |= O_CLOEXEC; |
361c023a MD |
383 | if (IS_ERR(s)) { |
384 | struct dentry *dentry = file->f_path.dentry; | |
385 | ||
386 | /* Make sure we give at least some info */ | |
387 | spin_lock(&dentry->d_lock); | |
e7a0ca72 MD |
388 | trace_lttng_statedump_file_descriptor(ctx->session, |
389 | ctx->files, fd, dentry->d_name.name, flags, | |
390 | file->f_mode); | |
361c023a MD |
391 | spin_unlock(&dentry->d_lock); |
392 | goto end; | |
393 | } | |
e7a0ca72 MD |
394 | trace_lttng_statedump_file_descriptor(ctx->session, |
395 | ctx->files, fd, s, flags, file->f_mode); | |
361c023a MD |
396 | end: |
397 | return 0; | |
398 | } | |
c337ddc2 | 399 | |
e7a0ca72 | 400 | /* Called with task lock held. */ |
c337ddc2 | 401 | static |
8cdc1a81 | 402 | void lttng_enumerate_files(struct lttng_kernel_session *session, |
e7a0ca72 MD |
403 | struct files_struct *files, |
404 | char *tmp) | |
c337ddc2 | 405 | { |
e7a0ca72 | 406 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, }; |
c337ddc2 | 407 | |
d561ecfb | 408 | lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx); |
c337ddc2 MD |
409 | } |
410 | ||
d0b55e4c | 411 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
502e4132 | 412 | static |
8cdc1a81 | 413 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
414 | { |
415 | int cpu; | |
416 | const cpumask_t *cpumask = cpu_possible_mask; | |
417 | ||
418 | for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids; | |
419 | cpu = cpumask_next(cpu, cpumask)) { | |
420 | trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu)); | |
421 | } | |
422 | ||
423 | return 0; | |
424 | } | |
425 | #else | |
426 | static | |
8cdc1a81 | 427 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
428 | { |
429 | return 0; | |
430 | } | |
431 | #endif | |
432 | ||
0658bdda MD |
433 | #if 0 |
434 | /* | |
435 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section | |
436 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of | |
437 | * iteration, but it is not exported to modules. | |
438 | */ | |
c337ddc2 | 439 | static |
8cdc1a81 | 440 | void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session, |
c337ddc2 MD |
441 | struct task_struct *p) |
442 | { | |
443 | struct mm_struct *mm; | |
444 | struct vm_area_struct *map; | |
445 | unsigned long ino; | |
446 | ||
447 | /* get_task_mm does a task_lock... */ | |
448 | mm = get_task_mm(p); | |
449 | if (!mm) | |
450 | return; | |
451 | ||
452 | map = mm->mmap; | |
453 | if (map) { | |
454 | down_read(&mm->mmap_sem); | |
455 | while (map) { | |
456 | if (map->vm_file) | |
b06ed645 | 457 | ino = map->vm_file->lttng_f_dentry->d_inode->i_ino; |
c337ddc2 MD |
458 | else |
459 | ino = 0; | |
460 | trace_lttng_statedump_vm_map(session, p, map, ino); | |
461 | map = map->vm_next; | |
462 | } | |
463 | up_read(&mm->mmap_sem); | |
464 | } | |
465 | mmput(mm); | |
466 | } | |
467 | ||
468 | static | |
8cdc1a81 | 469 | int lttng_enumerate_vm_maps(struct lttng_kernel_session *session) |
c337ddc2 MD |
470 | { |
471 | struct task_struct *p; | |
472 | ||
473 | rcu_read_lock(); | |
474 | for_each_process(p) | |
475 | lttng_enumerate_task_vm_maps(session, p); | |
476 | rcu_read_unlock(); | |
477 | return 0; | |
478 | } | |
0658bdda | 479 | #endif |
c337ddc2 | 480 | |
29784493 | 481 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
47faec4b | 482 | |
c337ddc2 | 483 | static |
8cdc1a81 | 484 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 MD |
485 | { |
486 | unsigned int irq; | |
487 | unsigned long flags = 0; | |
488 | struct irq_desc *desc; | |
489 | ||
490 | #define irq_to_desc wrapper_irq_to_desc | |
491 | /* needs irq_desc */ | |
492 | for_each_irq_desc(irq, desc) { | |
493 | struct irqaction *action; | |
494 | const char *irq_chip_name = | |
495 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; | |
496 | ||
497 | local_irq_save(flags); | |
fc94c945 | 498 | raw_spin_lock(&desc->lock); |
c337ddc2 MD |
499 | for (action = desc->action; action; action = action->next) { |
500 | trace_lttng_statedump_interrupt(session, | |
501 | irq, irq_chip_name, action); | |
502 | } | |
fc94c945 | 503 | raw_spin_unlock(&desc->lock); |
c337ddc2 MD |
504 | local_irq_restore(flags); |
505 | } | |
cfcee1c7 | 506 | return 0; |
c337ddc2 MD |
507 | #undef irq_to_desc |
508 | } | |
509 | #else | |
510 | static inline | |
8cdc1a81 | 511 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 | 512 | { |
cfcee1c7 | 513 | return 0; |
c337ddc2 MD |
514 | } |
515 | #endif | |
516 | ||
4ba1f53c | 517 | /* |
1965e6b4 MJ |
518 | * Statedump the task's namespaces using the proc filesystem inode number as |
519 | * the unique identifier. The user and pid ns are nested and will be dumped | |
520 | * recursively. | |
521 | * | |
4ba1f53c MD |
522 | * Called with task lock held. |
523 | */ | |
73e8ba37 | 524 | static |
8cdc1a81 | 525 | void lttng_statedump_process_ns(struct lttng_kernel_session *session, |
73e8ba37 JD |
526 | struct task_struct *p, |
527 | enum lttng_thread_type type, | |
528 | enum lttng_execution_mode mode, | |
529 | enum lttng_execution_submode submode, | |
530 | enum lttng_process_status status) | |
531 | { | |
1965e6b4 | 532 | struct nsproxy *proxy; |
73e8ba37 | 533 | struct pid_namespace *pid_ns; |
1965e6b4 | 534 | struct user_namespace *user_ns; |
73e8ba37 | 535 | |
1965e6b4 MJ |
536 | /* |
537 | * The pid and user namespaces are special, they are nested and | |
538 | * accessed with specific functions instead of the nsproxy struct | |
539 | * like the other namespaces. | |
540 | */ | |
887bcdac MJ |
541 | pid_ns = task_active_pid_ns(p); |
542 | do { | |
1965e6b4 | 543 | trace_lttng_statedump_process_pid_ns(session, p, pid_ns); |
adcc8b5e | 544 | pid_ns = pid_ns ? pid_ns->parent : NULL; |
887bcdac | 545 | } while (pid_ns); |
1965e6b4 MJ |
546 | |
547 | ||
548 | user_ns = task_cred_xxx(p, user_ns); | |
549 | do { | |
550 | trace_lttng_statedump_process_user_ns(session, p, user_ns); | |
1964cccb MD |
551 | /* |
552 | * trace_lttng_statedump_process_user_ns() internally | |
553 | * checks whether user_ns is NULL. While this does not | |
554 | * appear to be a possible return value for | |
555 | * task_cred_xxx(), err on the safe side and check | |
556 | * for NULL here as well to be consistent with the | |
557 | * paranoid behavior of | |
558 | * trace_lttng_statedump_process_user_ns(). | |
559 | */ | |
560 | user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL; | |
1965e6b4 MJ |
561 | } while (user_ns); |
562 | ||
563 | /* | |
564 | * Back and forth on locking strategy within Linux upstream for nsproxy. | |
565 | * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3 | |
566 | * "namespaces: Use task_lock and not rcu to protect nsproxy" | |
567 | * for details. | |
568 | */ | |
5f4c791e | 569 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
570 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
571 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
572 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
573 | proxy = p->nsproxy; | |
574 | #else | |
575 | rcu_read_lock(); | |
576 | proxy = task_nsproxy(p); | |
577 | #endif | |
578 | if (proxy) { | |
5f4c791e | 579 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
1965e6b4 MJ |
580 | trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns); |
581 | #endif | |
582 | trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns); | |
583 | #ifndef LTTNG_MNT_NS_MISSING_HEADER | |
584 | trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns); | |
585 | #endif | |
586 | trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns); | |
587 | trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns); | |
5f4c791e | 588 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0)) |
876e2e92 MJ |
589 | trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns); |
590 | #endif | |
1965e6b4 | 591 | } |
5f4c791e | 592 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
593 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
594 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
595 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
596 | /* (nothing) */ | |
597 | #else | |
598 | rcu_read_unlock(); | |
599 | #endif | |
73e8ba37 JD |
600 | } |
601 | ||
c337ddc2 | 602 | static |
8cdc1a81 | 603 | int lttng_enumerate_process_states(struct lttng_kernel_session *session) |
c337ddc2 MD |
604 | { |
605 | struct task_struct *g, *p; | |
e7a0ca72 MD |
606 | char *tmp; |
607 | ||
608 | tmp = (char *) __get_free_page(GFP_KERNEL); | |
609 | if (!tmp) | |
610 | return -ENOMEM; | |
c337ddc2 MD |
611 | |
612 | rcu_read_lock(); | |
613 | for_each_process(g) { | |
e7a0ca72 MD |
614 | struct files_struct *prev_files = NULL; |
615 | ||
c337ddc2 MD |
616 | p = g; |
617 | do { | |
618 | enum lttng_execution_mode mode = | |
619 | LTTNG_MODE_UNKNOWN; | |
620 | enum lttng_execution_submode submode = | |
621 | LTTNG_UNKNOWN; | |
622 | enum lttng_process_status status; | |
623 | enum lttng_thread_type type; | |
e7a0ca72 | 624 | struct files_struct *files; |
c337ddc2 MD |
625 | |
626 | task_lock(p); | |
627 | if (p->exit_state == EXIT_ZOMBIE) | |
628 | status = LTTNG_ZOMBIE; | |
629 | else if (p->exit_state == EXIT_DEAD) | |
630 | status = LTTNG_DEAD; | |
631 | else if (p->state == TASK_RUNNING) { | |
632 | /* Is this a forked child that has not run yet? */ | |
633 | if (list_empty(&p->rt.run_list)) | |
634 | status = LTTNG_WAIT_FORK; | |
635 | else | |
636 | /* | |
637 | * All tasks are considered as wait_cpu; | |
638 | * the viewer will sort out if the task | |
639 | * was really running at this time. | |
640 | */ | |
641 | status = LTTNG_WAIT_CPU; | |
642 | } else if (p->state & | |
643 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { | |
644 | /* Task is waiting for something to complete */ | |
645 | status = LTTNG_WAIT; | |
646 | } else | |
647 | status = LTTNG_UNNAMED; | |
648 | submode = LTTNG_NONE; | |
649 | ||
650 | /* | |
651 | * Verification of t->mm is to filter out kernel | |
652 | * threads; Viewer will further filter out if a | |
653 | * user-space thread was in syscall mode or not. | |
654 | */ | |
655 | if (p->mm) | |
656 | type = LTTNG_USER_THREAD; | |
657 | else | |
658 | type = LTTNG_KERNEL_THREAD; | |
e7a0ca72 | 659 | files = p->files; |
d2a927ac MJ |
660 | |
661 | trace_lttng_statedump_process_state(session, | |
e7a0ca72 | 662 | p, type, mode, submode, status, files); |
73e8ba37 | 663 | lttng_statedump_process_ns(session, |
c337ddc2 | 664 | p, type, mode, submode, status); |
e7a0ca72 MD |
665 | /* |
666 | * As an optimisation for the common case, do not | |
667 | * repeat information for the same files_struct in | |
668 | * two consecutive threads. This is the common case | |
669 | * for threads sharing the same fd table. RCU guarantees | |
670 | * that the same files_struct pointer is not re-used | |
671 | * throughout processes/threads iteration. | |
672 | */ | |
673 | if (files && files != prev_files) { | |
674 | lttng_enumerate_files(session, files, tmp); | |
675 | prev_files = files; | |
676 | } | |
c337ddc2 MD |
677 | task_unlock(p); |
678 | } while_each_thread(g, p); | |
679 | } | |
680 | rcu_read_unlock(); | |
681 | ||
e7a0ca72 MD |
682 | free_page((unsigned long) tmp); |
683 | ||
c337ddc2 MD |
684 | return 0; |
685 | } | |
686 | ||
687 | static | |
688 | void lttng_statedump_work_func(struct work_struct *work) | |
689 | { | |
690 | if (atomic_dec_and_test(&kernel_threads_to_run)) | |
691 | /* If we are the last thread, wake up do_lttng_statedump */ | |
692 | wake_up(&statedump_wq); | |
693 | } | |
694 | ||
695 | static | |
8cdc1a81 | 696 | int do_lttng_statedump(struct lttng_kernel_session *session) |
c337ddc2 | 697 | { |
cfcee1c7 | 698 | int cpu, ret; |
c337ddc2 | 699 | |
c337ddc2 | 700 | trace_lttng_statedump_start(session); |
cfcee1c7 | 701 | ret = lttng_enumerate_process_states(session); |
cfcee1c7 MD |
702 | if (ret) |
703 | return ret; | |
704 | /* | |
705 | * FIXME | |
706 | * ret = lttng_enumerate_vm_maps(session); | |
707 | * if (ret) | |
708 | * return ret; | |
709 | */ | |
710 | ret = lttng_list_interrupts(session); | |
711 | if (ret) | |
712 | return ret; | |
713 | ret = lttng_enumerate_network_ip_interface(session); | |
714 | if (ret) | |
715 | return ret; | |
716 | ret = lttng_enumerate_block_devices(session); | |
717 | switch (ret) { | |
84c7055e MD |
718 | case 0: |
719 | break; | |
cfcee1c7 MD |
720 | case -ENOSYS: |
721 | printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n"); | |
722 | break; | |
723 | default: | |
724 | return ret; | |
725 | } | |
502e4132 JD |
726 | ret = lttng_enumerate_cpu_topology(session); |
727 | if (ret) | |
728 | return ret; | |
c337ddc2 MD |
729 | |
730 | /* TODO lttng_dump_idt_table(session); */ | |
731 | /* TODO lttng_dump_softirq_vec(session); */ | |
732 | /* TODO lttng_list_modules(session); */ | |
733 | /* TODO lttng_dump_swap_files(session); */ | |
734 | ||
735 | /* | |
736 | * Fire off a work queue on each CPU. Their sole purpose in life | |
737 | * is to guarantee that each CPU has been in a state where is was in | |
738 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). | |
739 | */ | |
740 | get_online_cpus(); | |
741 | atomic_set(&kernel_threads_to_run, num_online_cpus()); | |
742 | for_each_online_cpu(cpu) { | |
743 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); | |
744 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); | |
745 | } | |
746 | /* Wait for all threads to run */ | |
7a7128e0 | 747 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
c337ddc2 MD |
748 | put_online_cpus(); |
749 | /* Our work is done */ | |
c337ddc2 MD |
750 | trace_lttng_statedump_end(session); |
751 | return 0; | |
752 | } | |
753 | ||
754 | /* | |
755 | * Called with session mutex held. | |
756 | */ | |
8cdc1a81 | 757 | int lttng_statedump_start(struct lttng_kernel_session *session) |
c337ddc2 | 758 | { |
c337ddc2 MD |
759 | return do_lttng_statedump(session); |
760 | } | |
761 | EXPORT_SYMBOL_GPL(lttng_statedump_start); | |
762 | ||
dd8d5afb MD |
763 | static |
764 | int __init lttng_statedump_init(void) | |
765 | { | |
d16aa9c9 MD |
766 | /* |
767 | * Allow module to load even if the fixup cannot be done. This | |
768 | * will allow seemless transition when the underlying issue fix | |
769 | * is merged into the Linux kernel, and when tracepoint.c | |
770 | * "tracepoint_module_notify" is turned into a static function. | |
771 | */ | |
772 | (void) wrapper_lttng_fixup_sig(THIS_MODULE); | |
773 | return 0; | |
dd8d5afb MD |
774 | } |
775 | ||
776 | module_init(lttng_statedump_init); | |
777 | ||
461277e7 MD |
778 | static |
779 | void __exit lttng_statedump_exit(void) | |
780 | { | |
781 | } | |
782 | ||
783 | module_exit(lttng_statedump_exit); | |
784 | ||
c337ddc2 MD |
785 | MODULE_LICENSE("GPL and additional rights"); |
786 | MODULE_AUTHOR("Jean-Hugues Deschenes"); | |
1c124020 | 787 | MODULE_DESCRIPTION("LTTng statedump provider"); |
13ab8b0a MD |
788 | MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "." |
789 | __stringify(LTTNG_MODULES_MINOR_VERSION) "." | |
790 | __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION) | |
791 | LTTNG_MODULES_EXTRAVERSION); |