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> | |
c337ddc2 MD |
26 | #include <linux/netdevice.h> |
27 | #include <linux/inetdevice.h> | |
c337ddc2 | 28 | #include <linux/mm.h> |
c337ddc2 MD |
29 | #include <linux/swap.h> |
30 | #include <linux/wait.h> | |
31 | #include <linux/mutex.h> | |
f0dbdefb | 32 | #include <linux/device.h> |
c337ddc2 | 33 | |
5677d2e5 MJ |
34 | #include <linux/blkdev.h> |
35 | ||
2df37e95 MD |
36 | #include <lttng/events.h> |
37 | #include <lttng/tracer.h> | |
ffcc8734 | 38 | #include <wrapper/cpu.h> |
241ae9a8 | 39 | #include <wrapper/irqdesc.h> |
241ae9a8 | 40 | #include <wrapper/fdtable.h> |
1965e6b4 | 41 | #include <wrapper/namespace.h> |
241ae9a8 MD |
42 | #include <wrapper/irq.h> |
43 | #include <wrapper/tracepoint.h> | |
82fbf9d3 | 44 | #include <wrapper/blkdev.h> |
c07dca48 | 45 | #include <wrapper/fdtable.h> |
7259e6a5 | 46 | #include <wrapper/sched.h> |
c337ddc2 | 47 | |
29784493 | 48 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
c337ddc2 MD |
49 | #include <linux/irq.h> |
50 | #endif | |
51 | ||
52 | /* Define the tracepoints, but do not build the probes */ | |
53 | #define CREATE_TRACE_POINTS | |
4f47ccf0 | 54 | #define TRACE_INCLUDE_PATH instrumentation/events |
c337ddc2 | 55 | #define TRACE_INCLUDE_FILE lttng-statedump |
3bc29f0a | 56 | #define LTTNG_INSTRUMENTATION |
4f47ccf0 | 57 | #include <instrumentation/events/lttng-statedump.h> |
c337ddc2 | 58 | |
0116245f | 59 | LTTNG_DEFINE_TRACE(lttng_statedump_block_device, |
8cdc1a81 | 60 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
61 | dev_t dev, const char *diskname), |
62 | TP_ARGS(session, dev, diskname)); | |
63 | ||
64 | LTTNG_DEFINE_TRACE(lttng_statedump_end, | |
8cdc1a81 | 65 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
66 | TP_ARGS(session)); |
67 | ||
68 | LTTNG_DEFINE_TRACE(lttng_statedump_interrupt, | |
8cdc1a81 | 69 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
70 | unsigned int irq, const char *chip_name, |
71 | struct irqaction *action), | |
72 | TP_ARGS(session, irq, chip_name, action)); | |
73 | ||
74 | LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor, | |
8cdc1a81 | 75 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
76 | struct files_struct *files, |
77 | int fd, const char *filename, | |
78 | unsigned int flags, fmode_t fmode), | |
79 | TP_ARGS(session, files, fd, filename, flags, fmode)); | |
80 | ||
81 | LTTNG_DEFINE_TRACE(lttng_statedump_start, | |
8cdc1a81 | 82 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
83 | TP_ARGS(session)); |
84 | ||
85 | LTTNG_DEFINE_TRACE(lttng_statedump_process_state, | |
8cdc1a81 | 86 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
87 | struct task_struct *p, |
88 | int type, int mode, int submode, int status, | |
89 | struct files_struct *files), | |
90 | TP_ARGS(session, p, type, mode, submode, status, files)); | |
91 | ||
92 | LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns, | |
8cdc1a81 | 93 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
94 | struct task_struct *p, |
95 | struct pid_namespace *pid_ns), | |
96 | TP_ARGS(session, p, pid_ns)); | |
97 | ||
5f4c791e | 98 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
0116245f | 99 | LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns, |
8cdc1a81 | 100 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
101 | struct task_struct *p, |
102 | struct cgroup_namespace *cgroup_ns), | |
103 | TP_ARGS(session, p, cgroup_ns)); | |
1965e6b4 | 104 | #endif |
0116245f MJ |
105 | |
106 | LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns, | |
8cdc1a81 | 107 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
108 | struct task_struct *p, |
109 | struct ipc_namespace *ipc_ns), | |
110 | TP_ARGS(session, p, ipc_ns)); | |
111 | ||
1965e6b4 | 112 | #ifndef LTTNG_MNT_NS_MISSING_HEADER |
0116245f | 113 | LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns, |
8cdc1a81 | 114 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
115 | struct task_struct *p, |
116 | struct mnt_namespace *mnt_ns), | |
117 | TP_ARGS(session, p, mnt_ns)); | |
1965e6b4 | 118 | #endif |
0116245f MJ |
119 | |
120 | LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns, | |
8cdc1a81 | 121 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
122 | struct task_struct *p, |
123 | struct net *net_ns), | |
124 | TP_ARGS(session, p, net_ns)); | |
125 | ||
126 | LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns, | |
8cdc1a81 | 127 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
128 | struct task_struct *p, |
129 | struct user_namespace *user_ns), | |
130 | TP_ARGS(session, p, user_ns)); | |
131 | ||
132 | LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns, | |
8cdc1a81 | 133 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
134 | struct task_struct *p, |
135 | struct uts_namespace *uts_ns), | |
136 | TP_ARGS(session, p, uts_ns)); | |
137 | ||
138 | LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns, | |
8cdc1a81 | 139 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
140 | struct task_struct *p, |
141 | struct time_namespace *time_ns), | |
142 | TP_ARGS(session, p, time_ns)); | |
143 | ||
144 | LTTNG_DEFINE_TRACE(lttng_statedump_network_interface, | |
8cdc1a81 | 145 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
146 | struct net_device *dev, struct in_ifaddr *ifa), |
147 | TP_ARGS(session, dev, ifa)); | |
148 | ||
d0b55e4c | 149 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
0116245f | 150 | LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology, |
8cdc1a81 | 151 | TP_PROTO(struct lttng_kernel_session *session, struct cpuinfo_x86 *c), |
0116245f | 152 | TP_ARGS(session, c)); |
502e4132 | 153 | #endif |
20591cf7 | 154 | |
361c023a MD |
155 | struct lttng_fd_ctx { |
156 | char *page; | |
8cdc1a81 | 157 | struct lttng_kernel_session *session; |
d561ecfb | 158 | struct files_struct *files; |
361c023a MD |
159 | }; |
160 | ||
c337ddc2 MD |
161 | /* |
162 | * Protected by the trace lock. | |
163 | */ | |
164 | static struct delayed_work cpu_work[NR_CPUS]; | |
165 | static DECLARE_WAIT_QUEUE_HEAD(statedump_wq); | |
166 | static atomic_t kernel_threads_to_run; | |
167 | ||
168 | enum lttng_thread_type { | |
169 | LTTNG_USER_THREAD = 0, | |
170 | LTTNG_KERNEL_THREAD = 1, | |
171 | }; | |
172 | ||
173 | enum lttng_execution_mode { | |
174 | LTTNG_USER_MODE = 0, | |
175 | LTTNG_SYSCALL = 1, | |
176 | LTTNG_TRAP = 2, | |
177 | LTTNG_IRQ = 3, | |
178 | LTTNG_SOFTIRQ = 4, | |
179 | LTTNG_MODE_UNKNOWN = 5, | |
180 | }; | |
181 | ||
182 | enum lttng_execution_submode { | |
183 | LTTNG_NONE = 0, | |
184 | LTTNG_UNKNOWN = 1, | |
185 | }; | |
186 | ||
187 | enum lttng_process_status { | |
188 | LTTNG_UNNAMED = 0, | |
189 | LTTNG_WAIT_FORK = 1, | |
190 | LTTNG_WAIT_CPU = 2, | |
191 | LTTNG_EXIT = 3, | |
192 | LTTNG_ZOMBIE = 4, | |
193 | LTTNG_WAIT = 5, | |
194 | LTTNG_RUN = 6, | |
195 | LTTNG_DEAD = 7, | |
196 | }; | |
197 | ||
fef7fe6a | 198 | |
cc92a211 MJ |
199 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0)) |
200 | ||
201 | #define LTTNG_PART_STRUCT_TYPE struct block_device | |
202 | ||
203 | static | |
204 | int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf) | |
205 | { | |
206 | int ret; | |
207 | ||
208 | ret = snprintf(name_buf, BDEVNAME_SIZE, "%pg", part); | |
209 | if (ret < 0 || ret >= BDEVNAME_SIZE) | |
210 | return -ENOSYS; | |
211 | ||
212 | return 0; | |
213 | } | |
214 | ||
215 | static | |
216 | dev_t lttng_get_part_devt(struct block_device *part) | |
217 | { | |
218 | return part->bd_dev; | |
219 | } | |
220 | ||
221 | #elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0)) | |
fef7fe6a MJ |
222 | |
223 | #define LTTNG_PART_STRUCT_TYPE struct block_device | |
224 | ||
225 | static | |
226 | int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf) | |
227 | { | |
228 | const char *p; | |
229 | ||
230 | p = bdevname(part, name_buf); | |
231 | if (!p) | |
232 | return -ENOSYS; | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | static | |
238 | dev_t lttng_get_part_devt(struct block_device *part) | |
239 | { | |
240 | return part->bd_dev; | |
241 | } | |
242 | ||
243 | #else | |
244 | ||
245 | #define LTTNG_PART_STRUCT_TYPE struct hd_struct | |
246 | ||
247 | static | |
248 | int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf) | |
249 | { | |
250 | const char *p; | |
251 | struct block_device bdev; | |
252 | ||
253 | /* | |
254 | * Create a partial 'struct blockdevice' to use | |
255 | * 'bdevname()' which is a simple wrapper over | |
256 | * 'disk_name()' but has the honor to be EXPORT_SYMBOL. | |
257 | */ | |
258 | bdev.bd_disk = disk; | |
259 | bdev.bd_part = part; | |
260 | ||
261 | p = bdevname(&bdev, name_buf); | |
262 | if (!p) | |
263 | return -ENOSYS; | |
264 | ||
265 | return 0; | |
266 | } | |
267 | ||
268 | static | |
269 | dev_t lttng_get_part_devt(struct hd_struct *part) | |
270 | { | |
271 | return part_devt(part); | |
272 | } | |
273 | #endif | |
274 | ||
481fd2a2 MJ |
275 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0)) |
276 | static | |
277 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) | |
278 | { | |
279 | struct block_device *part; | |
280 | unsigned long idx; | |
281 | int ret = 0; | |
282 | ||
283 | /* Include partition 0 */ | |
284 | idx = 0; | |
285 | ||
286 | rcu_read_lock(); | |
287 | xa_for_each(&disk->part_tbl, idx, part) { | |
288 | char name_buf[BDEVNAME_SIZE]; | |
289 | ||
290 | /* Exclude non-partitions bdev and empty partitions. */ | |
291 | if (bdev_is_partition(part) && !bdev_nr_sectors(part)) | |
292 | continue; | |
293 | ||
294 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { | |
295 | ret = -ENOSYS; | |
296 | goto end; | |
297 | } | |
298 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), | |
299 | name_buf); | |
300 | } | |
301 | end: | |
302 | rcu_read_unlock(); | |
303 | return ret; | |
304 | } | |
305 | #else | |
306 | static | |
307 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) | |
308 | { | |
309 | struct disk_part_iter piter; | |
310 | LTTNG_PART_STRUCT_TYPE *part; | |
311 | ||
312 | disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0); | |
313 | ||
314 | while ((part = disk_part_iter_next(&piter))) { | |
315 | char name_buf[BDEVNAME_SIZE]; | |
316 | ||
317 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { | |
318 | disk_part_iter_exit(&piter); | |
319 | return -ENOSYS; | |
320 | } | |
321 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), | |
322 | name_buf); | |
323 | } | |
324 | disk_part_iter_exit(&piter); | |
325 | ||
326 | return 0; | |
327 | } | |
328 | #endif | |
329 | ||
f0dbdefb | 330 | static |
8cdc1a81 | 331 | int lttng_enumerate_block_devices(struct lttng_kernel_session *session) |
f0dbdefb HD |
332 | { |
333 | struct class *ptr_block_class; | |
334 | struct device_type *ptr_disk_type; | |
335 | struct class_dev_iter iter; | |
336 | struct device *dev; | |
7633c773 | 337 | int ret = 0; |
f0dbdefb HD |
338 | |
339 | ptr_block_class = wrapper_get_block_class(); | |
7633c773 MJ |
340 | if (!ptr_block_class) { |
341 | ret = -ENOSYS; | |
342 | goto end; | |
343 | } | |
f0dbdefb HD |
344 | ptr_disk_type = wrapper_get_disk_type(); |
345 | if (!ptr_disk_type) { | |
7633c773 MJ |
346 | ret = -ENOSYS; |
347 | goto end; | |
f0dbdefb HD |
348 | } |
349 | class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type); | |
350 | while ((dev = class_dev_iter_next(&iter))) { | |
f0dbdefb | 351 | struct gendisk *disk = dev_to_disk(dev); |
f0dbdefb | 352 | |
5a91f3df MD |
353 | /* |
354 | * Don't show empty devices or things that have been | |
355 | * suppressed | |
356 | */ | |
357 | if (get_capacity(disk) == 0 || | |
fe72ab45 | 358 | (disk->flags & LTTNG_GENHD_FL_HIDDEN)) |
5a91f3df MD |
359 | continue; |
360 | ||
481fd2a2 | 361 | ret = lttng_statedump_each_block_device(session, disk); |
f0dbdefb HD |
362 | } |
363 | class_dev_iter_exit(&iter); | |
7633c773 MJ |
364 | end: |
365 | return ret; | |
f0dbdefb HD |
366 | } |
367 | ||
c337ddc2 | 368 | #ifdef CONFIG_INET |
f0dbdefb | 369 | |
c337ddc2 | 370 | static |
8cdc1a81 | 371 | void lttng_enumerate_device(struct lttng_kernel_session *session, |
c337ddc2 MD |
372 | struct net_device *dev) |
373 | { | |
374 | struct in_device *in_dev; | |
375 | struct in_ifaddr *ifa; | |
376 | ||
377 | if (dev->flags & IFF_UP) { | |
378 | in_dev = in_dev_get(dev); | |
379 | if (in_dev) { | |
380 | for (ifa = in_dev->ifa_list; ifa != NULL; | |
381 | ifa = ifa->ifa_next) { | |
382 | trace_lttng_statedump_network_interface( | |
383 | session, dev, ifa); | |
384 | } | |
385 | in_dev_put(in_dev); | |
386 | } | |
387 | } else { | |
388 | trace_lttng_statedump_network_interface( | |
389 | session, dev, NULL); | |
390 | } | |
391 | } | |
392 | ||
393 | static | |
8cdc1a81 | 394 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
395 | { |
396 | struct net_device *dev; | |
397 | ||
398 | read_lock(&dev_base_lock); | |
399 | for_each_netdev(&init_net, dev) | |
400 | lttng_enumerate_device(session, dev); | |
401 | read_unlock(&dev_base_lock); | |
402 | ||
403 | return 0; | |
404 | } | |
405 | #else /* CONFIG_INET */ | |
406 | static inline | |
8cdc1a81 | 407 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
408 | { |
409 | return 0; | |
410 | } | |
411 | #endif /* CONFIG_INET */ | |
412 | ||
361c023a MD |
413 | static |
414 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) | |
415 | { | |
416 | const struct lttng_fd_ctx *ctx = p; | |
417 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); | |
29021503 | 418 | unsigned int flags = file->f_flags; |
d561ecfb | 419 | struct fdtable *fdt; |
361c023a | 420 | |
29021503 MD |
421 | /* |
422 | * We don't expose kernel internal flags, only userspace-visible | |
423 | * flags. | |
424 | */ | |
425 | flags &= ~FMODE_NONOTIFY; | |
d561ecfb MD |
426 | fdt = files_fdtable(ctx->files); |
427 | /* | |
428 | * We need to check here again whether fd is within the fdt | |
429 | * max_fds range, because we might be seeing a different | |
430 | * files_fdtable() than iterate_fd(), assuming only RCU is | |
431 | * protecting the read. In reality, iterate_fd() holds | |
432 | * file_lock, which should ensure the fdt does not change while | |
433 | * the lock is taken, but we are not aware whether this is | |
434 | * guaranteed or not, so play safe. | |
435 | */ | |
aa29f2d3 | 436 | if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt)) |
29021503 | 437 | flags |= O_CLOEXEC; |
361c023a MD |
438 | if (IS_ERR(s)) { |
439 | struct dentry *dentry = file->f_path.dentry; | |
440 | ||
441 | /* Make sure we give at least some info */ | |
442 | spin_lock(&dentry->d_lock); | |
e7a0ca72 MD |
443 | trace_lttng_statedump_file_descriptor(ctx->session, |
444 | ctx->files, fd, dentry->d_name.name, flags, | |
445 | file->f_mode); | |
361c023a MD |
446 | spin_unlock(&dentry->d_lock); |
447 | goto end; | |
448 | } | |
e7a0ca72 MD |
449 | trace_lttng_statedump_file_descriptor(ctx->session, |
450 | ctx->files, fd, s, flags, file->f_mode); | |
361c023a MD |
451 | end: |
452 | return 0; | |
453 | } | |
c337ddc2 | 454 | |
e7a0ca72 | 455 | /* Called with task lock held. */ |
c337ddc2 | 456 | static |
8cdc1a81 | 457 | void lttng_enumerate_files(struct lttng_kernel_session *session, |
e7a0ca72 MD |
458 | struct files_struct *files, |
459 | char *tmp) | |
c337ddc2 | 460 | { |
e7a0ca72 | 461 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, }; |
c337ddc2 | 462 | |
d561ecfb | 463 | lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx); |
c337ddc2 MD |
464 | } |
465 | ||
d0b55e4c | 466 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
502e4132 | 467 | static |
8cdc1a81 | 468 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
469 | { |
470 | int cpu; | |
471 | const cpumask_t *cpumask = cpu_possible_mask; | |
472 | ||
473 | for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids; | |
474 | cpu = cpumask_next(cpu, cpumask)) { | |
475 | trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu)); | |
476 | } | |
477 | ||
478 | return 0; | |
479 | } | |
480 | #else | |
481 | static | |
8cdc1a81 | 482 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
483 | { |
484 | return 0; | |
485 | } | |
486 | #endif | |
487 | ||
0658bdda MD |
488 | #if 0 |
489 | /* | |
490 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section | |
491 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of | |
492 | * iteration, but it is not exported to modules. | |
493 | */ | |
c337ddc2 | 494 | static |
8cdc1a81 | 495 | void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session, |
c337ddc2 MD |
496 | struct task_struct *p) |
497 | { | |
498 | struct mm_struct *mm; | |
499 | struct vm_area_struct *map; | |
500 | unsigned long ino; | |
501 | ||
502 | /* get_task_mm does a task_lock... */ | |
503 | mm = get_task_mm(p); | |
504 | if (!mm) | |
505 | return; | |
506 | ||
507 | map = mm->mmap; | |
508 | if (map) { | |
509 | down_read(&mm->mmap_sem); | |
510 | while (map) { | |
511 | if (map->vm_file) | |
2a059b14 | 512 | ino = map->vm_file->f_path.dentry->d_inode->i_ino; |
c337ddc2 MD |
513 | else |
514 | ino = 0; | |
515 | trace_lttng_statedump_vm_map(session, p, map, ino); | |
516 | map = map->vm_next; | |
517 | } | |
518 | up_read(&mm->mmap_sem); | |
519 | } | |
520 | mmput(mm); | |
521 | } | |
522 | ||
523 | static | |
8cdc1a81 | 524 | int lttng_enumerate_vm_maps(struct lttng_kernel_session *session) |
c337ddc2 MD |
525 | { |
526 | struct task_struct *p; | |
527 | ||
528 | rcu_read_lock(); | |
529 | for_each_process(p) | |
530 | lttng_enumerate_task_vm_maps(session, p); | |
531 | rcu_read_unlock(); | |
532 | return 0; | |
533 | } | |
0658bdda | 534 | #endif |
c337ddc2 | 535 | |
29784493 | 536 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
47faec4b | 537 | |
c337ddc2 | 538 | static |
8cdc1a81 | 539 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 MD |
540 | { |
541 | unsigned int irq; | |
542 | unsigned long flags = 0; | |
543 | struct irq_desc *desc; | |
544 | ||
545 | #define irq_to_desc wrapper_irq_to_desc | |
546 | /* needs irq_desc */ | |
547 | for_each_irq_desc(irq, desc) { | |
548 | struct irqaction *action; | |
549 | const char *irq_chip_name = | |
550 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; | |
551 | ||
552 | local_irq_save(flags); | |
fc94c945 | 553 | raw_spin_lock(&desc->lock); |
c337ddc2 MD |
554 | for (action = desc->action; action; action = action->next) { |
555 | trace_lttng_statedump_interrupt(session, | |
556 | irq, irq_chip_name, action); | |
557 | } | |
fc94c945 | 558 | raw_spin_unlock(&desc->lock); |
c337ddc2 MD |
559 | local_irq_restore(flags); |
560 | } | |
cfcee1c7 | 561 | return 0; |
c337ddc2 MD |
562 | #undef irq_to_desc |
563 | } | |
564 | #else | |
565 | static inline | |
8cdc1a81 | 566 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 | 567 | { |
cfcee1c7 | 568 | return 0; |
c337ddc2 MD |
569 | } |
570 | #endif | |
571 | ||
4ba1f53c | 572 | /* |
1965e6b4 MJ |
573 | * Statedump the task's namespaces using the proc filesystem inode number as |
574 | * the unique identifier. The user and pid ns are nested and will be dumped | |
575 | * recursively. | |
576 | * | |
4ba1f53c MD |
577 | * Called with task lock held. |
578 | */ | |
73e8ba37 | 579 | static |
8cdc1a81 | 580 | void lttng_statedump_process_ns(struct lttng_kernel_session *session, |
73e8ba37 JD |
581 | struct task_struct *p, |
582 | enum lttng_thread_type type, | |
583 | enum lttng_execution_mode mode, | |
584 | enum lttng_execution_submode submode, | |
585 | enum lttng_process_status status) | |
586 | { | |
1965e6b4 | 587 | struct nsproxy *proxy; |
73e8ba37 | 588 | struct pid_namespace *pid_ns; |
1965e6b4 | 589 | struct user_namespace *user_ns; |
73e8ba37 | 590 | |
1965e6b4 MJ |
591 | /* |
592 | * The pid and user namespaces are special, they are nested and | |
593 | * accessed with specific functions instead of the nsproxy struct | |
594 | * like the other namespaces. | |
595 | */ | |
887bcdac MJ |
596 | pid_ns = task_active_pid_ns(p); |
597 | do { | |
1965e6b4 | 598 | trace_lttng_statedump_process_pid_ns(session, p, pid_ns); |
adcc8b5e | 599 | pid_ns = pid_ns ? pid_ns->parent : NULL; |
887bcdac | 600 | } while (pid_ns); |
1965e6b4 MJ |
601 | |
602 | ||
603 | user_ns = task_cred_xxx(p, user_ns); | |
604 | do { | |
605 | trace_lttng_statedump_process_user_ns(session, p, user_ns); | |
1964cccb MD |
606 | /* |
607 | * trace_lttng_statedump_process_user_ns() internally | |
608 | * checks whether user_ns is NULL. While this does not | |
609 | * appear to be a possible return value for | |
610 | * task_cred_xxx(), err on the safe side and check | |
611 | * for NULL here as well to be consistent with the | |
612 | * paranoid behavior of | |
613 | * trace_lttng_statedump_process_user_ns(). | |
614 | */ | |
615 | user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL; | |
1965e6b4 MJ |
616 | } while (user_ns); |
617 | ||
618 | /* | |
619 | * Back and forth on locking strategy within Linux upstream for nsproxy. | |
620 | * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3 | |
621 | * "namespaces: Use task_lock and not rcu to protect nsproxy" | |
622 | * for details. | |
623 | */ | |
5f4c791e | 624 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
625 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
626 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
627 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
628 | proxy = p->nsproxy; | |
629 | #else | |
630 | rcu_read_lock(); | |
631 | proxy = task_nsproxy(p); | |
632 | #endif | |
633 | if (proxy) { | |
5f4c791e | 634 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
1965e6b4 MJ |
635 | trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns); |
636 | #endif | |
637 | trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns); | |
638 | #ifndef LTTNG_MNT_NS_MISSING_HEADER | |
639 | trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns); | |
640 | #endif | |
641 | trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns); | |
642 | trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns); | |
fce87a6e MJ |
643 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \ |
644 | LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0)) | |
876e2e92 MJ |
645 | trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns); |
646 | #endif | |
1965e6b4 | 647 | } |
5f4c791e | 648 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
649 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
650 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
651 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
652 | /* (nothing) */ | |
653 | #else | |
654 | rcu_read_unlock(); | |
655 | #endif | |
73e8ba37 JD |
656 | } |
657 | ||
c337ddc2 | 658 | static |
8cdc1a81 | 659 | int lttng_enumerate_process_states(struct lttng_kernel_session *session) |
c337ddc2 MD |
660 | { |
661 | struct task_struct *g, *p; | |
e7a0ca72 MD |
662 | char *tmp; |
663 | ||
664 | tmp = (char *) __get_free_page(GFP_KERNEL); | |
665 | if (!tmp) | |
666 | return -ENOMEM; | |
c337ddc2 MD |
667 | |
668 | rcu_read_lock(); | |
669 | for_each_process(g) { | |
e7a0ca72 MD |
670 | struct files_struct *prev_files = NULL; |
671 | ||
c337ddc2 MD |
672 | p = g; |
673 | do { | |
674 | enum lttng_execution_mode mode = | |
675 | LTTNG_MODE_UNKNOWN; | |
676 | enum lttng_execution_submode submode = | |
677 | LTTNG_UNKNOWN; | |
678 | enum lttng_process_status status; | |
679 | enum lttng_thread_type type; | |
e7a0ca72 | 680 | struct files_struct *files; |
c337ddc2 MD |
681 | |
682 | task_lock(p); | |
683 | if (p->exit_state == EXIT_ZOMBIE) | |
684 | status = LTTNG_ZOMBIE; | |
685 | else if (p->exit_state == EXIT_DEAD) | |
686 | status = LTTNG_DEAD; | |
7259e6a5 | 687 | else if (lttng_task_is_running(p)) { |
c337ddc2 MD |
688 | /* Is this a forked child that has not run yet? */ |
689 | if (list_empty(&p->rt.run_list)) | |
690 | status = LTTNG_WAIT_FORK; | |
691 | else | |
692 | /* | |
693 | * All tasks are considered as wait_cpu; | |
694 | * the viewer will sort out if the task | |
695 | * was really running at this time. | |
696 | */ | |
697 | status = LTTNG_WAIT_CPU; | |
7259e6a5 | 698 | } else if (lttng_get_task_state(p) & |
c337ddc2 MD |
699 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { |
700 | /* Task is waiting for something to complete */ | |
701 | status = LTTNG_WAIT; | |
702 | } else | |
703 | status = LTTNG_UNNAMED; | |
704 | submode = LTTNG_NONE; | |
705 | ||
706 | /* | |
707 | * Verification of t->mm is to filter out kernel | |
708 | * threads; Viewer will further filter out if a | |
709 | * user-space thread was in syscall mode or not. | |
710 | */ | |
711 | if (p->mm) | |
712 | type = LTTNG_USER_THREAD; | |
713 | else | |
714 | type = LTTNG_KERNEL_THREAD; | |
e7a0ca72 | 715 | files = p->files; |
d2a927ac MJ |
716 | |
717 | trace_lttng_statedump_process_state(session, | |
e7a0ca72 | 718 | p, type, mode, submode, status, files); |
73e8ba37 | 719 | lttng_statedump_process_ns(session, |
c337ddc2 | 720 | p, type, mode, submode, status); |
e7a0ca72 MD |
721 | /* |
722 | * As an optimisation for the common case, do not | |
723 | * repeat information for the same files_struct in | |
724 | * two consecutive threads. This is the common case | |
725 | * for threads sharing the same fd table. RCU guarantees | |
726 | * that the same files_struct pointer is not re-used | |
727 | * throughout processes/threads iteration. | |
728 | */ | |
729 | if (files && files != prev_files) { | |
730 | lttng_enumerate_files(session, files, tmp); | |
731 | prev_files = files; | |
732 | } | |
c337ddc2 MD |
733 | task_unlock(p); |
734 | } while_each_thread(g, p); | |
735 | } | |
736 | rcu_read_unlock(); | |
737 | ||
e7a0ca72 MD |
738 | free_page((unsigned long) tmp); |
739 | ||
c337ddc2 MD |
740 | return 0; |
741 | } | |
742 | ||
743 | static | |
744 | void lttng_statedump_work_func(struct work_struct *work) | |
745 | { | |
746 | if (atomic_dec_and_test(&kernel_threads_to_run)) | |
747 | /* If we are the last thread, wake up do_lttng_statedump */ | |
748 | wake_up(&statedump_wq); | |
749 | } | |
750 | ||
751 | static | |
8cdc1a81 | 752 | int do_lttng_statedump(struct lttng_kernel_session *session) |
c337ddc2 | 753 | { |
cfcee1c7 | 754 | int cpu, ret; |
c337ddc2 | 755 | |
c337ddc2 | 756 | trace_lttng_statedump_start(session); |
cfcee1c7 | 757 | ret = lttng_enumerate_process_states(session); |
cfcee1c7 MD |
758 | if (ret) |
759 | return ret; | |
760 | /* | |
761 | * FIXME | |
762 | * ret = lttng_enumerate_vm_maps(session); | |
763 | * if (ret) | |
764 | * return ret; | |
765 | */ | |
766 | ret = lttng_list_interrupts(session); | |
767 | if (ret) | |
768 | return ret; | |
769 | ret = lttng_enumerate_network_ip_interface(session); | |
770 | if (ret) | |
771 | return ret; | |
772 | ret = lttng_enumerate_block_devices(session); | |
773 | switch (ret) { | |
84c7055e MD |
774 | case 0: |
775 | break; | |
cfcee1c7 MD |
776 | case -ENOSYS: |
777 | printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n"); | |
778 | break; | |
779 | default: | |
780 | return ret; | |
781 | } | |
502e4132 JD |
782 | ret = lttng_enumerate_cpu_topology(session); |
783 | if (ret) | |
784 | return ret; | |
c337ddc2 MD |
785 | |
786 | /* TODO lttng_dump_idt_table(session); */ | |
787 | /* TODO lttng_dump_softirq_vec(session); */ | |
788 | /* TODO lttng_list_modules(session); */ | |
789 | /* TODO lttng_dump_swap_files(session); */ | |
790 | ||
791 | /* | |
792 | * Fire off a work queue on each CPU. Their sole purpose in life | |
793 | * is to guarantee that each CPU has been in a state where is was in | |
794 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). | |
795 | */ | |
ffcc8734 | 796 | lttng_cpus_read_lock(); |
c337ddc2 MD |
797 | atomic_set(&kernel_threads_to_run, num_online_cpus()); |
798 | for_each_online_cpu(cpu) { | |
799 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); | |
800 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); | |
801 | } | |
802 | /* Wait for all threads to run */ | |
7a7128e0 | 803 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
ffcc8734 | 804 | lttng_cpus_read_unlock(); |
c337ddc2 | 805 | /* Our work is done */ |
c337ddc2 MD |
806 | trace_lttng_statedump_end(session); |
807 | return 0; | |
808 | } | |
809 | ||
810 | /* | |
811 | * Called with session mutex held. | |
812 | */ | |
8cdc1a81 | 813 | int lttng_statedump_start(struct lttng_kernel_session *session) |
c337ddc2 | 814 | { |
c337ddc2 MD |
815 | return do_lttng_statedump(session); |
816 | } | |
817 | EXPORT_SYMBOL_GPL(lttng_statedump_start); | |
818 | ||
dd8d5afb MD |
819 | static |
820 | int __init lttng_statedump_init(void) | |
821 | { | |
d16aa9c9 MD |
822 | /* |
823 | * Allow module to load even if the fixup cannot be done. This | |
824 | * will allow seemless transition when the underlying issue fix | |
825 | * is merged into the Linux kernel, and when tracepoint.c | |
826 | * "tracepoint_module_notify" is turned into a static function. | |
827 | */ | |
828 | (void) wrapper_lttng_fixup_sig(THIS_MODULE); | |
829 | return 0; | |
dd8d5afb MD |
830 | } |
831 | ||
832 | module_init(lttng_statedump_init); | |
833 | ||
461277e7 MD |
834 | static |
835 | void __exit lttng_statedump_exit(void) | |
836 | { | |
837 | } | |
838 | ||
839 | module_exit(lttng_statedump_exit); | |
840 | ||
c337ddc2 MD |
841 | MODULE_LICENSE("GPL and additional rights"); |
842 | MODULE_AUTHOR("Jean-Hugues Deschenes"); | |
1c124020 | 843 | MODULE_DESCRIPTION("LTTng statedump provider"); |
13ab8b0a MD |
844 | MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "." |
845 | __stringify(LTTNG_MODULES_MINOR_VERSION) "." | |
846 | __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION) | |
847 | LTTNG_MODULES_EXTRAVERSION); |