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 | |
d82de6f3 MJ |
34 | #include <linux/blkdev.h> |
35 | ||
2df37e95 MD |
36 | #include <lttng/events.h> |
37 | #include <lttng/tracer.h> | |
8be4c8a3 | 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> | |
44 | #include <wrapper/genhd.h> | |
45 | #include <wrapper/file.h> | |
c07dca48 | 46 | #include <wrapper/fdtable.h> |
a9f6bb33 | 47 | #include <wrapper/sched.h> |
c337ddc2 | 48 | |
29784493 | 49 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
c337ddc2 MD |
50 | #include <linux/irq.h> |
51 | #endif | |
52 | ||
53 | /* Define the tracepoints, but do not build the probes */ | |
54 | #define CREATE_TRACE_POINTS | |
4f47ccf0 | 55 | #define TRACE_INCLUDE_PATH instrumentation/events |
c337ddc2 | 56 | #define TRACE_INCLUDE_FILE lttng-statedump |
3bc29f0a | 57 | #define LTTNG_INSTRUMENTATION |
4f47ccf0 | 58 | #include <instrumentation/events/lttng-statedump.h> |
c337ddc2 | 59 | |
0116245f | 60 | LTTNG_DEFINE_TRACE(lttng_statedump_block_device, |
a40e3229 | 61 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
62 | dev_t dev, const char *diskname), |
63 | TP_ARGS(session, dev, diskname)); | |
64 | ||
65 | LTTNG_DEFINE_TRACE(lttng_statedump_end, | |
a40e3229 | 66 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
67 | TP_ARGS(session)); |
68 | ||
69 | LTTNG_DEFINE_TRACE(lttng_statedump_interrupt, | |
a40e3229 | 70 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
71 | unsigned int irq, const char *chip_name, |
72 | struct irqaction *action), | |
73 | TP_ARGS(session, irq, chip_name, action)); | |
74 | ||
75 | LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor, | |
a40e3229 | 76 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
77 | struct files_struct *files, |
78 | int fd, const char *filename, | |
79 | unsigned int flags, fmode_t fmode), | |
80 | TP_ARGS(session, files, fd, filename, flags, fmode)); | |
81 | ||
82 | LTTNG_DEFINE_TRACE(lttng_statedump_start, | |
a40e3229 | 83 | TP_PROTO(struct lttng_kernel_session *session), |
0116245f MJ |
84 | TP_ARGS(session)); |
85 | ||
86 | LTTNG_DEFINE_TRACE(lttng_statedump_process_state, | |
a40e3229 | 87 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
88 | struct task_struct *p, |
89 | int type, int mode, int submode, int status, | |
90 | struct files_struct *files), | |
91 | TP_ARGS(session, p, type, mode, submode, status, files)); | |
92 | ||
93 | LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns, | |
a40e3229 | 94 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
95 | struct task_struct *p, |
96 | struct pid_namespace *pid_ns), | |
97 | TP_ARGS(session, p, pid_ns)); | |
98 | ||
5f4c791e | 99 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
0116245f | 100 | LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns, |
a40e3229 | 101 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
102 | struct task_struct *p, |
103 | struct cgroup_namespace *cgroup_ns), | |
104 | TP_ARGS(session, p, cgroup_ns)); | |
1965e6b4 | 105 | #endif |
0116245f MJ |
106 | |
107 | LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns, | |
a40e3229 | 108 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
109 | struct task_struct *p, |
110 | struct ipc_namespace *ipc_ns), | |
111 | TP_ARGS(session, p, ipc_ns)); | |
112 | ||
1965e6b4 | 113 | #ifndef LTTNG_MNT_NS_MISSING_HEADER |
0116245f | 114 | LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns, |
a40e3229 | 115 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
116 | struct task_struct *p, |
117 | struct mnt_namespace *mnt_ns), | |
118 | TP_ARGS(session, p, mnt_ns)); | |
1965e6b4 | 119 | #endif |
0116245f MJ |
120 | |
121 | LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns, | |
a40e3229 | 122 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
123 | struct task_struct *p, |
124 | struct net *net_ns), | |
125 | TP_ARGS(session, p, net_ns)); | |
126 | ||
127 | LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns, | |
a40e3229 | 128 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
129 | struct task_struct *p, |
130 | struct user_namespace *user_ns), | |
131 | TP_ARGS(session, p, user_ns)); | |
132 | ||
133 | LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns, | |
a40e3229 | 134 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
135 | struct task_struct *p, |
136 | struct uts_namespace *uts_ns), | |
137 | TP_ARGS(session, p, uts_ns)); | |
138 | ||
139 | LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns, | |
a40e3229 | 140 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
141 | struct task_struct *p, |
142 | struct time_namespace *time_ns), | |
143 | TP_ARGS(session, p, time_ns)); | |
144 | ||
145 | LTTNG_DEFINE_TRACE(lttng_statedump_network_interface, | |
a40e3229 | 146 | TP_PROTO(struct lttng_kernel_session *session, |
0116245f MJ |
147 | struct net_device *dev, struct in_ifaddr *ifa), |
148 | TP_ARGS(session, dev, ifa)); | |
149 | ||
d0b55e4c | 150 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
0116245f | 151 | LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology, |
a40e3229 | 152 | TP_PROTO(struct lttng_kernel_session *session, struct cpuinfo_x86 *c), |
0116245f | 153 | TP_ARGS(session, c)); |
502e4132 | 154 | #endif |
20591cf7 | 155 | |
361c023a MD |
156 | struct lttng_fd_ctx { |
157 | char *page; | |
a40e3229 | 158 | struct lttng_kernel_session *session; |
d561ecfb | 159 | struct files_struct *files; |
361c023a MD |
160 | }; |
161 | ||
c337ddc2 MD |
162 | /* |
163 | * Protected by the trace lock. | |
164 | */ | |
165 | static struct delayed_work cpu_work[NR_CPUS]; | |
166 | static DECLARE_WAIT_QUEUE_HEAD(statedump_wq); | |
167 | static atomic_t kernel_threads_to_run; | |
168 | ||
169 | enum lttng_thread_type { | |
170 | LTTNG_USER_THREAD = 0, | |
171 | LTTNG_KERNEL_THREAD = 1, | |
172 | }; | |
173 | ||
174 | enum lttng_execution_mode { | |
175 | LTTNG_USER_MODE = 0, | |
176 | LTTNG_SYSCALL = 1, | |
177 | LTTNG_TRAP = 2, | |
178 | LTTNG_IRQ = 3, | |
179 | LTTNG_SOFTIRQ = 4, | |
180 | LTTNG_MODE_UNKNOWN = 5, | |
181 | }; | |
182 | ||
183 | enum lttng_execution_submode { | |
184 | LTTNG_NONE = 0, | |
185 | LTTNG_UNKNOWN = 1, | |
186 | }; | |
187 | ||
188 | enum lttng_process_status { | |
189 | LTTNG_UNNAMED = 0, | |
190 | LTTNG_WAIT_FORK = 1, | |
191 | LTTNG_WAIT_CPU = 2, | |
192 | LTTNG_EXIT = 3, | |
193 | LTTNG_ZOMBIE = 4, | |
194 | LTTNG_WAIT = 5, | |
195 | LTTNG_RUN = 6, | |
196 | LTTNG_DEAD = 7, | |
197 | }; | |
198 | ||
fef7fe6a | 199 | |
f6800492 MH |
200 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0) || \ |
201 | LTTNG_RHEL_KERNEL_RANGE(5,14,0,163,0,0, 5,15,0,0,0,0)) | |
8c2e17ee MJ |
202 | |
203 | #define LTTNG_PART_STRUCT_TYPE struct block_device | |
204 | ||
205 | static | |
206 | int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf) | |
207 | { | |
208 | int ret; | |
209 | ||
210 | ret = snprintf(name_buf, BDEVNAME_SIZE, "%pg", part); | |
211 | if (ret < 0 || ret >= BDEVNAME_SIZE) | |
212 | return -ENOSYS; | |
213 | ||
214 | return 0; | |
215 | } | |
216 | ||
217 | static | |
218 | dev_t lttng_get_part_devt(struct block_device *part) | |
219 | { | |
220 | return part->bd_dev; | |
221 | } | |
222 | ||
223 | #elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0)) | |
fef7fe6a MJ |
224 | |
225 | #define LTTNG_PART_STRUCT_TYPE struct block_device | |
226 | ||
227 | static | |
228 | int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf) | |
229 | { | |
230 | const char *p; | |
231 | ||
232 | p = bdevname(part, name_buf); | |
233 | if (!p) | |
234 | return -ENOSYS; | |
235 | ||
236 | return 0; | |
237 | } | |
238 | ||
239 | static | |
240 | dev_t lttng_get_part_devt(struct block_device *part) | |
241 | { | |
242 | return part->bd_dev; | |
243 | } | |
244 | ||
245 | #else | |
246 | ||
247 | #define LTTNG_PART_STRUCT_TYPE struct hd_struct | |
248 | ||
249 | static | |
250 | int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf) | |
251 | { | |
252 | const char *p; | |
253 | struct block_device bdev; | |
254 | ||
255 | /* | |
256 | * Create a partial 'struct blockdevice' to use | |
257 | * 'bdevname()' which is a simple wrapper over | |
258 | * 'disk_name()' but has the honor to be EXPORT_SYMBOL. | |
259 | */ | |
260 | bdev.bd_disk = disk; | |
261 | bdev.bd_part = part; | |
262 | ||
263 | p = bdevname(&bdev, name_buf); | |
264 | if (!p) | |
265 | return -ENOSYS; | |
266 | ||
267 | return 0; | |
268 | } | |
269 | ||
270 | static | |
271 | dev_t lttng_get_part_devt(struct hd_struct *part) | |
272 | { | |
273 | return part_devt(part); | |
274 | } | |
275 | #endif | |
276 | ||
de874bde MJ |
277 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0)) |
278 | static | |
279 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) | |
280 | { | |
281 | struct block_device *part; | |
282 | unsigned long idx; | |
283 | int ret = 0; | |
284 | ||
285 | /* Include partition 0 */ | |
286 | idx = 0; | |
287 | ||
288 | rcu_read_lock(); | |
289 | xa_for_each(&disk->part_tbl, idx, part) { | |
290 | char name_buf[BDEVNAME_SIZE]; | |
291 | ||
292 | /* Exclude non-partitions bdev and empty partitions. */ | |
293 | if (bdev_is_partition(part) && !bdev_nr_sectors(part)) | |
294 | continue; | |
295 | ||
296 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { | |
297 | ret = -ENOSYS; | |
298 | goto end; | |
299 | } | |
300 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), | |
301 | name_buf); | |
302 | } | |
303 | end: | |
304 | rcu_read_unlock(); | |
305 | return ret; | |
306 | } | |
307 | #else | |
308 | static | |
309 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) | |
310 | { | |
311 | struct disk_part_iter piter; | |
312 | LTTNG_PART_STRUCT_TYPE *part; | |
313 | ||
314 | disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0); | |
315 | ||
316 | while ((part = disk_part_iter_next(&piter))) { | |
317 | char name_buf[BDEVNAME_SIZE]; | |
318 | ||
319 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { | |
320 | disk_part_iter_exit(&piter); | |
321 | return -ENOSYS; | |
322 | } | |
323 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), | |
324 | name_buf); | |
325 | } | |
326 | disk_part_iter_exit(&piter); | |
327 | ||
328 | return 0; | |
329 | } | |
330 | #endif | |
331 | ||
f0dbdefb | 332 | static |
a40e3229 | 333 | int lttng_enumerate_block_devices(struct lttng_kernel_session *session) |
f0dbdefb HD |
334 | { |
335 | struct class *ptr_block_class; | |
336 | struct device_type *ptr_disk_type; | |
337 | struct class_dev_iter iter; | |
338 | struct device *dev; | |
7633c773 | 339 | int ret = 0; |
f0dbdefb HD |
340 | |
341 | ptr_block_class = wrapper_get_block_class(); | |
7633c773 MJ |
342 | if (!ptr_block_class) { |
343 | ret = -ENOSYS; | |
344 | goto end; | |
345 | } | |
f0dbdefb HD |
346 | ptr_disk_type = wrapper_get_disk_type(); |
347 | if (!ptr_disk_type) { | |
7633c773 MJ |
348 | ret = -ENOSYS; |
349 | goto end; | |
f0dbdefb HD |
350 | } |
351 | class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type); | |
352 | while ((dev = class_dev_iter_next(&iter))) { | |
f0dbdefb | 353 | struct gendisk *disk = dev_to_disk(dev); |
f0dbdefb | 354 | |
5a91f3df MD |
355 | /* |
356 | * Don't show empty devices or things that have been | |
357 | * suppressed | |
358 | */ | |
359 | if (get_capacity(disk) == 0 || | |
a88ee460 | 360 | (disk->flags & LTTNG_GENHD_FL_HIDDEN)) |
5a91f3df MD |
361 | continue; |
362 | ||
de874bde | 363 | ret = lttng_statedump_each_block_device(session, disk); |
f0dbdefb HD |
364 | } |
365 | class_dev_iter_exit(&iter); | |
7633c773 MJ |
366 | end: |
367 | return ret; | |
f0dbdefb HD |
368 | } |
369 | ||
c337ddc2 | 370 | #ifdef CONFIG_INET |
f0dbdefb | 371 | |
c337ddc2 | 372 | static |
a40e3229 | 373 | void lttng_enumerate_device(struct lttng_kernel_session *session, |
c337ddc2 MD |
374 | struct net_device *dev) |
375 | { | |
376 | struct in_device *in_dev; | |
377 | struct in_ifaddr *ifa; | |
378 | ||
379 | if (dev->flags & IFF_UP) { | |
380 | in_dev = in_dev_get(dev); | |
381 | if (in_dev) { | |
382 | for (ifa = in_dev->ifa_list; ifa != NULL; | |
383 | ifa = ifa->ifa_next) { | |
384 | trace_lttng_statedump_network_interface( | |
385 | session, dev, ifa); | |
386 | } | |
387 | in_dev_put(in_dev); | |
388 | } | |
389 | } else { | |
390 | trace_lttng_statedump_network_interface( | |
391 | session, dev, NULL); | |
392 | } | |
393 | } | |
394 | ||
52eb2ee9 KS |
395 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0)) |
396 | static | |
397 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) | |
398 | { | |
399 | struct net_device *dev; | |
400 | ||
401 | rtnl_lock(); | |
402 | for_each_netdev(&init_net, dev) | |
403 | lttng_enumerate_device(session, dev); | |
404 | rtnl_unlock(); | |
405 | ||
406 | return 0; | |
407 | } | |
408 | #else | |
c337ddc2 | 409 | static |
a40e3229 | 410 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
411 | { |
412 | struct net_device *dev; | |
413 | ||
414 | read_lock(&dev_base_lock); | |
415 | for_each_netdev(&init_net, dev) | |
416 | lttng_enumerate_device(session, dev); | |
417 | read_unlock(&dev_base_lock); | |
418 | ||
419 | return 0; | |
420 | } | |
52eb2ee9 | 421 | #endif /* (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0)) */ |
c337ddc2 MD |
422 | #else /* CONFIG_INET */ |
423 | static inline | |
a40e3229 | 424 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
c337ddc2 MD |
425 | { |
426 | return 0; | |
427 | } | |
428 | #endif /* CONFIG_INET */ | |
429 | ||
361c023a MD |
430 | static |
431 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) | |
432 | { | |
433 | const struct lttng_fd_ctx *ctx = p; | |
434 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); | |
29021503 | 435 | unsigned int flags = file->f_flags; |
d561ecfb | 436 | struct fdtable *fdt; |
361c023a | 437 | |
29021503 MD |
438 | /* |
439 | * We don't expose kernel internal flags, only userspace-visible | |
440 | * flags. | |
441 | */ | |
442 | flags &= ~FMODE_NONOTIFY; | |
d561ecfb MD |
443 | fdt = files_fdtable(ctx->files); |
444 | /* | |
445 | * We need to check here again whether fd is within the fdt | |
446 | * max_fds range, because we might be seeing a different | |
447 | * files_fdtable() than iterate_fd(), assuming only RCU is | |
448 | * protecting the read. In reality, iterate_fd() holds | |
449 | * file_lock, which should ensure the fdt does not change while | |
450 | * the lock is taken, but we are not aware whether this is | |
451 | * guaranteed or not, so play safe. | |
452 | */ | |
aa29f2d3 | 453 | if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt)) |
29021503 | 454 | flags |= O_CLOEXEC; |
361c023a MD |
455 | if (IS_ERR(s)) { |
456 | struct dentry *dentry = file->f_path.dentry; | |
457 | ||
458 | /* Make sure we give at least some info */ | |
459 | spin_lock(&dentry->d_lock); | |
e7a0ca72 MD |
460 | trace_lttng_statedump_file_descriptor(ctx->session, |
461 | ctx->files, fd, dentry->d_name.name, flags, | |
462 | file->f_mode); | |
361c023a MD |
463 | spin_unlock(&dentry->d_lock); |
464 | goto end; | |
465 | } | |
e7a0ca72 MD |
466 | trace_lttng_statedump_file_descriptor(ctx->session, |
467 | ctx->files, fd, s, flags, file->f_mode); | |
361c023a MD |
468 | end: |
469 | return 0; | |
470 | } | |
c337ddc2 | 471 | |
e7a0ca72 | 472 | /* Called with task lock held. */ |
c337ddc2 | 473 | static |
a40e3229 | 474 | void lttng_enumerate_files(struct lttng_kernel_session *session, |
e7a0ca72 MD |
475 | struct files_struct *files, |
476 | char *tmp) | |
c337ddc2 | 477 | { |
e7a0ca72 | 478 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, }; |
c337ddc2 | 479 | |
d561ecfb | 480 | lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx); |
c337ddc2 MD |
481 | } |
482 | ||
d0b55e4c | 483 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
502e4132 | 484 | static |
a40e3229 | 485 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
486 | { |
487 | int cpu; | |
488 | const cpumask_t *cpumask = cpu_possible_mask; | |
489 | ||
490 | for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids; | |
491 | cpu = cpumask_next(cpu, cpumask)) { | |
492 | trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu)); | |
493 | } | |
494 | ||
495 | return 0; | |
496 | } | |
497 | #else | |
498 | static | |
a40e3229 | 499 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
502e4132 JD |
500 | { |
501 | return 0; | |
502 | } | |
503 | #endif | |
504 | ||
0658bdda MD |
505 | #if 0 |
506 | /* | |
507 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section | |
508 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of | |
509 | * iteration, but it is not exported to modules. | |
510 | */ | |
c337ddc2 | 511 | static |
a40e3229 | 512 | void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session, |
c337ddc2 MD |
513 | struct task_struct *p) |
514 | { | |
515 | struct mm_struct *mm; | |
516 | struct vm_area_struct *map; | |
517 | unsigned long ino; | |
518 | ||
519 | /* get_task_mm does a task_lock... */ | |
520 | mm = get_task_mm(p); | |
521 | if (!mm) | |
522 | return; | |
523 | ||
524 | map = mm->mmap; | |
525 | if (map) { | |
526 | down_read(&mm->mmap_sem); | |
527 | while (map) { | |
528 | if (map->vm_file) | |
b06ed645 | 529 | ino = map->vm_file->lttng_f_dentry->d_inode->i_ino; |
c337ddc2 MD |
530 | else |
531 | ino = 0; | |
532 | trace_lttng_statedump_vm_map(session, p, map, ino); | |
533 | map = map->vm_next; | |
534 | } | |
535 | up_read(&mm->mmap_sem); | |
536 | } | |
537 | mmput(mm); | |
538 | } | |
539 | ||
540 | static | |
a40e3229 | 541 | int lttng_enumerate_vm_maps(struct lttng_kernel_session *session) |
c337ddc2 MD |
542 | { |
543 | struct task_struct *p; | |
544 | ||
545 | rcu_read_lock(); | |
546 | for_each_process(p) | |
547 | lttng_enumerate_task_vm_maps(session, p); | |
548 | rcu_read_unlock(); | |
549 | return 0; | |
550 | } | |
0658bdda | 551 | #endif |
c337ddc2 | 552 | |
29784493 | 553 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
47faec4b | 554 | |
c337ddc2 | 555 | static |
a40e3229 | 556 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 MD |
557 | { |
558 | unsigned int irq; | |
559 | unsigned long flags = 0; | |
560 | struct irq_desc *desc; | |
561 | ||
562 | #define irq_to_desc wrapper_irq_to_desc | |
563 | /* needs irq_desc */ | |
564 | for_each_irq_desc(irq, desc) { | |
565 | struct irqaction *action; | |
566 | const char *irq_chip_name = | |
567 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; | |
568 | ||
569 | local_irq_save(flags); | |
fc94c945 | 570 | raw_spin_lock(&desc->lock); |
c337ddc2 MD |
571 | for (action = desc->action; action; action = action->next) { |
572 | trace_lttng_statedump_interrupt(session, | |
573 | irq, irq_chip_name, action); | |
574 | } | |
fc94c945 | 575 | raw_spin_unlock(&desc->lock); |
c337ddc2 MD |
576 | local_irq_restore(flags); |
577 | } | |
cfcee1c7 | 578 | return 0; |
c337ddc2 MD |
579 | #undef irq_to_desc |
580 | } | |
581 | #else | |
582 | static inline | |
a40e3229 | 583 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
c337ddc2 | 584 | { |
cfcee1c7 | 585 | return 0; |
c337ddc2 MD |
586 | } |
587 | #endif | |
588 | ||
4ba1f53c | 589 | /* |
1965e6b4 MJ |
590 | * Statedump the task's namespaces using the proc filesystem inode number as |
591 | * the unique identifier. The user and pid ns are nested and will be dumped | |
592 | * recursively. | |
593 | * | |
4ba1f53c MD |
594 | * Called with task lock held. |
595 | */ | |
73e8ba37 | 596 | static |
a40e3229 | 597 | void lttng_statedump_process_ns(struct lttng_kernel_session *session, |
73e8ba37 JD |
598 | struct task_struct *p, |
599 | enum lttng_thread_type type, | |
600 | enum lttng_execution_mode mode, | |
601 | enum lttng_execution_submode submode, | |
602 | enum lttng_process_status status) | |
603 | { | |
1965e6b4 | 604 | struct nsproxy *proxy; |
73e8ba37 | 605 | struct pid_namespace *pid_ns; |
1965e6b4 | 606 | struct user_namespace *user_ns; |
73e8ba37 | 607 | |
1965e6b4 MJ |
608 | /* |
609 | * The pid and user namespaces are special, they are nested and | |
610 | * accessed with specific functions instead of the nsproxy struct | |
611 | * like the other namespaces. | |
612 | */ | |
887bcdac MJ |
613 | pid_ns = task_active_pid_ns(p); |
614 | do { | |
1965e6b4 | 615 | trace_lttng_statedump_process_pid_ns(session, p, pid_ns); |
adcc8b5e | 616 | pid_ns = pid_ns ? pid_ns->parent : NULL; |
887bcdac | 617 | } while (pid_ns); |
1965e6b4 MJ |
618 | |
619 | ||
620 | user_ns = task_cred_xxx(p, user_ns); | |
621 | do { | |
622 | trace_lttng_statedump_process_user_ns(session, p, user_ns); | |
1964cccb MD |
623 | /* |
624 | * trace_lttng_statedump_process_user_ns() internally | |
625 | * checks whether user_ns is NULL. While this does not | |
626 | * appear to be a possible return value for | |
627 | * task_cred_xxx(), err on the safe side and check | |
628 | * for NULL here as well to be consistent with the | |
629 | * paranoid behavior of | |
630 | * trace_lttng_statedump_process_user_ns(). | |
631 | */ | |
632 | user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL; | |
1965e6b4 MJ |
633 | } while (user_ns); |
634 | ||
635 | /* | |
636 | * Back and forth on locking strategy within Linux upstream for nsproxy. | |
637 | * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3 | |
638 | * "namespaces: Use task_lock and not rcu to protect nsproxy" | |
639 | * for details. | |
640 | */ | |
5f4c791e | 641 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
642 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
643 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
644 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
645 | proxy = p->nsproxy; | |
646 | #else | |
647 | rcu_read_lock(); | |
648 | proxy = task_nsproxy(p); | |
649 | #endif | |
650 | if (proxy) { | |
5f4c791e | 651 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
1965e6b4 MJ |
652 | trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns); |
653 | #endif | |
654 | trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns); | |
655 | #ifndef LTTNG_MNT_NS_MISSING_HEADER | |
656 | trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns); | |
657 | #endif | |
658 | trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns); | |
659 | trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns); | |
f3a2b1ec MJ |
660 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \ |
661 | LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0)) | |
876e2e92 MJ |
662 | trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns); |
663 | #endif | |
1965e6b4 | 664 | } |
5f4c791e | 665 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
1965e6b4 MJ |
666 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
667 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ | |
668 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) | |
669 | /* (nothing) */ | |
670 | #else | |
671 | rcu_read_unlock(); | |
672 | #endif | |
73e8ba37 JD |
673 | } |
674 | ||
c337ddc2 | 675 | static |
a40e3229 | 676 | int lttng_enumerate_process_states(struct lttng_kernel_session *session) |
c337ddc2 MD |
677 | { |
678 | struct task_struct *g, *p; | |
e7a0ca72 MD |
679 | char *tmp; |
680 | ||
681 | tmp = (char *) __get_free_page(GFP_KERNEL); | |
682 | if (!tmp) | |
683 | return -ENOMEM; | |
c337ddc2 MD |
684 | |
685 | rcu_read_lock(); | |
686 | for_each_process(g) { | |
e7a0ca72 MD |
687 | struct files_struct *prev_files = NULL; |
688 | ||
c337ddc2 MD |
689 | p = g; |
690 | do { | |
691 | enum lttng_execution_mode mode = | |
692 | LTTNG_MODE_UNKNOWN; | |
693 | enum lttng_execution_submode submode = | |
694 | LTTNG_UNKNOWN; | |
695 | enum lttng_process_status status; | |
696 | enum lttng_thread_type type; | |
e7a0ca72 | 697 | struct files_struct *files; |
c337ddc2 MD |
698 | |
699 | task_lock(p); | |
700 | if (p->exit_state == EXIT_ZOMBIE) | |
701 | status = LTTNG_ZOMBIE; | |
702 | else if (p->exit_state == EXIT_DEAD) | |
703 | status = LTTNG_DEAD; | |
a9f6bb33 | 704 | else if (lttng_task_is_running(p)) { |
c337ddc2 MD |
705 | /* Is this a forked child that has not run yet? */ |
706 | if (list_empty(&p->rt.run_list)) | |
707 | status = LTTNG_WAIT_FORK; | |
708 | else | |
709 | /* | |
710 | * All tasks are considered as wait_cpu; | |
711 | * the viewer will sort out if the task | |
712 | * was really running at this time. | |
713 | */ | |
714 | status = LTTNG_WAIT_CPU; | |
a9f6bb33 | 715 | } else if (lttng_get_task_state(p) & |
c337ddc2 MD |
716 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { |
717 | /* Task is waiting for something to complete */ | |
718 | status = LTTNG_WAIT; | |
719 | } else | |
720 | status = LTTNG_UNNAMED; | |
721 | submode = LTTNG_NONE; | |
722 | ||
723 | /* | |
724 | * Verification of t->mm is to filter out kernel | |
725 | * threads; Viewer will further filter out if a | |
726 | * user-space thread was in syscall mode or not. | |
727 | */ | |
728 | if (p->mm) | |
729 | type = LTTNG_USER_THREAD; | |
730 | else | |
731 | type = LTTNG_KERNEL_THREAD; | |
e7a0ca72 | 732 | files = p->files; |
d2a927ac MJ |
733 | |
734 | trace_lttng_statedump_process_state(session, | |
e7a0ca72 | 735 | p, type, mode, submode, status, files); |
73e8ba37 | 736 | lttng_statedump_process_ns(session, |
c337ddc2 | 737 | p, type, mode, submode, status); |
e7a0ca72 MD |
738 | /* |
739 | * As an optimisation for the common case, do not | |
740 | * repeat information for the same files_struct in | |
741 | * two consecutive threads. This is the common case | |
742 | * for threads sharing the same fd table. RCU guarantees | |
743 | * that the same files_struct pointer is not re-used | |
744 | * throughout processes/threads iteration. | |
745 | */ | |
746 | if (files && files != prev_files) { | |
747 | lttng_enumerate_files(session, files, tmp); | |
748 | prev_files = files; | |
749 | } | |
c337ddc2 MD |
750 | task_unlock(p); |
751 | } while_each_thread(g, p); | |
752 | } | |
753 | rcu_read_unlock(); | |
754 | ||
e7a0ca72 MD |
755 | free_page((unsigned long) tmp); |
756 | ||
c337ddc2 MD |
757 | return 0; |
758 | } | |
759 | ||
760 | static | |
761 | void lttng_statedump_work_func(struct work_struct *work) | |
762 | { | |
763 | if (atomic_dec_and_test(&kernel_threads_to_run)) | |
764 | /* If we are the last thread, wake up do_lttng_statedump */ | |
765 | wake_up(&statedump_wq); | |
766 | } | |
767 | ||
768 | static | |
a40e3229 | 769 | int do_lttng_statedump(struct lttng_kernel_session *session) |
c337ddc2 | 770 | { |
cfcee1c7 | 771 | int cpu, ret; |
c337ddc2 | 772 | |
c337ddc2 | 773 | trace_lttng_statedump_start(session); |
cfcee1c7 | 774 | ret = lttng_enumerate_process_states(session); |
cfcee1c7 MD |
775 | if (ret) |
776 | return ret; | |
777 | /* | |
778 | * FIXME | |
779 | * ret = lttng_enumerate_vm_maps(session); | |
780 | * if (ret) | |
781 | * return ret; | |
782 | */ | |
783 | ret = lttng_list_interrupts(session); | |
784 | if (ret) | |
785 | return ret; | |
786 | ret = lttng_enumerate_network_ip_interface(session); | |
787 | if (ret) | |
788 | return ret; | |
789 | ret = lttng_enumerate_block_devices(session); | |
790 | switch (ret) { | |
84c7055e MD |
791 | case 0: |
792 | break; | |
cfcee1c7 MD |
793 | case -ENOSYS: |
794 | printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n"); | |
795 | break; | |
796 | default: | |
797 | return ret; | |
798 | } | |
502e4132 JD |
799 | ret = lttng_enumerate_cpu_topology(session); |
800 | if (ret) | |
801 | return ret; | |
c337ddc2 MD |
802 | |
803 | /* TODO lttng_dump_idt_table(session); */ | |
804 | /* TODO lttng_dump_softirq_vec(session); */ | |
805 | /* TODO lttng_list_modules(session); */ | |
806 | /* TODO lttng_dump_swap_files(session); */ | |
807 | ||
808 | /* | |
809 | * Fire off a work queue on each CPU. Their sole purpose in life | |
810 | * is to guarantee that each CPU has been in a state where is was in | |
811 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). | |
812 | */ | |
8be4c8a3 | 813 | lttng_cpus_read_lock(); |
c337ddc2 MD |
814 | atomic_set(&kernel_threads_to_run, num_online_cpus()); |
815 | for_each_online_cpu(cpu) { | |
816 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); | |
817 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); | |
818 | } | |
819 | /* Wait for all threads to run */ | |
7a7128e0 | 820 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
8be4c8a3 | 821 | lttng_cpus_read_unlock(); |
c337ddc2 | 822 | /* Our work is done */ |
c337ddc2 MD |
823 | trace_lttng_statedump_end(session); |
824 | return 0; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Called with session mutex held. | |
829 | */ | |
a40e3229 | 830 | int lttng_statedump_start(struct lttng_kernel_session *session) |
c337ddc2 | 831 | { |
c337ddc2 MD |
832 | return do_lttng_statedump(session); |
833 | } | |
834 | EXPORT_SYMBOL_GPL(lttng_statedump_start); | |
835 | ||
dd8d5afb MD |
836 | static |
837 | int __init lttng_statedump_init(void) | |
838 | { | |
d16aa9c9 MD |
839 | /* |
840 | * Allow module to load even if the fixup cannot be done. This | |
841 | * will allow seemless transition when the underlying issue fix | |
842 | * is merged into the Linux kernel, and when tracepoint.c | |
843 | * "tracepoint_module_notify" is turned into a static function. | |
844 | */ | |
845 | (void) wrapper_lttng_fixup_sig(THIS_MODULE); | |
846 | return 0; | |
dd8d5afb MD |
847 | } |
848 | ||
849 | module_init(lttng_statedump_init); | |
850 | ||
461277e7 MD |
851 | static |
852 | void __exit lttng_statedump_exit(void) | |
853 | { | |
854 | } | |
855 | ||
856 | module_exit(lttng_statedump_exit); | |
857 | ||
c337ddc2 MD |
858 | MODULE_LICENSE("GPL and additional rights"); |
859 | MODULE_AUTHOR("Jean-Hugues Deschenes"); | |
1c124020 | 860 | MODULE_DESCRIPTION("LTTng statedump provider"); |
13ab8b0a MD |
861 | MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "." |
862 | __stringify(LTTNG_MODULES_MINOR_VERSION) "." | |
863 | __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION) | |
864 | LTTNG_MODULES_EXTRAVERSION); |