| 1 | /* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only) |
| 2 | * |
| 3 | * lttng-statedump.c |
| 4 | * |
| 5 | * Linux Trace Toolkit Next Generation Kernel State Dump |
| 6 | * |
| 7 | * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca> |
| 8 | * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 9 | * |
| 10 | * Changes: |
| 11 | * Eric Clement: Add listing of network IP interface |
| 12 | * 2006, 2007 Mathieu Desnoyers Fix kernel threads |
| 13 | * Various updates |
| 14 | */ |
| 15 | |
| 16 | #include <linux/init.h> |
| 17 | #include <linux/module.h> |
| 18 | #include <linux/netlink.h> |
| 19 | #include <linux/inet.h> |
| 20 | #include <linux/ip.h> |
| 21 | #include <linux/kthread.h> |
| 22 | #include <linux/proc_fs.h> |
| 23 | #include <linux/file.h> |
| 24 | #include <linux/interrupt.h> |
| 25 | #include <linux/irqnr.h> |
| 26 | #include <linux/cpu.h> |
| 27 | #include <linux/netdevice.h> |
| 28 | #include <linux/inetdevice.h> |
| 29 | #include <linux/sched.h> |
| 30 | #include <linux/mm.h> |
| 31 | #include <linux/swap.h> |
| 32 | #include <linux/wait.h> |
| 33 | #include <linux/mutex.h> |
| 34 | #include <linux/device.h> |
| 35 | |
| 36 | #include <lttng/events.h> |
| 37 | #include <lttng/tracer.h> |
| 38 | #include <wrapper/irqdesc.h> |
| 39 | #include <wrapper/fdtable.h> |
| 40 | #include <wrapper/namespace.h> |
| 41 | #include <wrapper/irq.h> |
| 42 | #include <wrapper/tracepoint.h> |
| 43 | #include <wrapper/genhd.h> |
| 44 | #include <wrapper/file.h> |
| 45 | #include <wrapper/fdtable.h> |
| 46 | |
| 47 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
| 48 | #include <linux/irq.h> |
| 49 | #endif |
| 50 | |
| 51 | /* Define the tracepoints, but do not build the probes */ |
| 52 | #define CREATE_TRACE_POINTS |
| 53 | #define TRACE_INCLUDE_PATH instrumentation/events |
| 54 | #define TRACE_INCLUDE_FILE lttng-statedump |
| 55 | #define LTTNG_INSTRUMENTATION |
| 56 | #include <instrumentation/events/lttng-statedump.h> |
| 57 | |
| 58 | LTTNG_DEFINE_TRACE(lttng_statedump_block_device, |
| 59 | TP_PROTO(struct lttng_kernel_session *session, |
| 60 | dev_t dev, const char *diskname), |
| 61 | TP_ARGS(session, dev, diskname)); |
| 62 | |
| 63 | LTTNG_DEFINE_TRACE(lttng_statedump_end, |
| 64 | TP_PROTO(struct lttng_kernel_session *session), |
| 65 | TP_ARGS(session)); |
| 66 | |
| 67 | LTTNG_DEFINE_TRACE(lttng_statedump_interrupt, |
| 68 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 74 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 81 | TP_PROTO(struct lttng_kernel_session *session), |
| 82 | TP_ARGS(session)); |
| 83 | |
| 84 | LTTNG_DEFINE_TRACE(lttng_statedump_process_state, |
| 85 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 92 | TP_PROTO(struct lttng_kernel_session *session, |
| 93 | struct task_struct *p, |
| 94 | struct pid_namespace *pid_ns), |
| 95 | TP_ARGS(session, p, pid_ns)); |
| 96 | |
| 97 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
| 98 | LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns, |
| 99 | TP_PROTO(struct lttng_kernel_session *session, |
| 100 | struct task_struct *p, |
| 101 | struct cgroup_namespace *cgroup_ns), |
| 102 | TP_ARGS(session, p, cgroup_ns)); |
| 103 | #endif |
| 104 | |
| 105 | LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns, |
| 106 | TP_PROTO(struct lttng_kernel_session *session, |
| 107 | struct task_struct *p, |
| 108 | struct ipc_namespace *ipc_ns), |
| 109 | TP_ARGS(session, p, ipc_ns)); |
| 110 | |
| 111 | #ifndef LTTNG_MNT_NS_MISSING_HEADER |
| 112 | LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns, |
| 113 | TP_PROTO(struct lttng_kernel_session *session, |
| 114 | struct task_struct *p, |
| 115 | struct mnt_namespace *mnt_ns), |
| 116 | TP_ARGS(session, p, mnt_ns)); |
| 117 | #endif |
| 118 | |
| 119 | LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns, |
| 120 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 126 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 132 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 138 | TP_PROTO(struct lttng_kernel_session *session, |
| 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, |
| 144 | TP_PROTO(struct lttng_kernel_session *session, |
| 145 | struct net_device *dev, struct in_ifaddr *ifa), |
| 146 | TP_ARGS(session, dev, ifa)); |
| 147 | |
| 148 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
| 149 | LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology, |
| 150 | TP_PROTO(struct lttng_kernel_session *session, struct cpuinfo_x86 *c), |
| 151 | TP_ARGS(session, c)); |
| 152 | #endif |
| 153 | |
| 154 | struct lttng_fd_ctx { |
| 155 | char *page; |
| 156 | struct lttng_kernel_session *session; |
| 157 | struct files_struct *files; |
| 158 | }; |
| 159 | |
| 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 | |
| 197 | |
| 198 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0)) |
| 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 | |
| 252 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0)) |
| 253 | static |
| 254 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) |
| 255 | { |
| 256 | struct block_device *part; |
| 257 | unsigned long idx; |
| 258 | int ret = 0; |
| 259 | |
| 260 | /* Include partition 0 */ |
| 261 | idx = 0; |
| 262 | |
| 263 | rcu_read_lock(); |
| 264 | xa_for_each(&disk->part_tbl, idx, part) { |
| 265 | char name_buf[BDEVNAME_SIZE]; |
| 266 | |
| 267 | /* Exclude non-partitions bdev and empty partitions. */ |
| 268 | if (bdev_is_partition(part) && !bdev_nr_sectors(part)) |
| 269 | continue; |
| 270 | |
| 271 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { |
| 272 | ret = -ENOSYS; |
| 273 | goto end; |
| 274 | } |
| 275 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), |
| 276 | name_buf); |
| 277 | } |
| 278 | end: |
| 279 | rcu_read_unlock(); |
| 280 | return ret; |
| 281 | } |
| 282 | #else |
| 283 | static |
| 284 | int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk) |
| 285 | { |
| 286 | struct disk_part_iter piter; |
| 287 | LTTNG_PART_STRUCT_TYPE *part; |
| 288 | |
| 289 | disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0); |
| 290 | |
| 291 | while ((part = disk_part_iter_next(&piter))) { |
| 292 | char name_buf[BDEVNAME_SIZE]; |
| 293 | |
| 294 | if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) { |
| 295 | disk_part_iter_exit(&piter); |
| 296 | return -ENOSYS; |
| 297 | } |
| 298 | trace_lttng_statedump_block_device(session, lttng_get_part_devt(part), |
| 299 | name_buf); |
| 300 | } |
| 301 | disk_part_iter_exit(&piter); |
| 302 | |
| 303 | return 0; |
| 304 | } |
| 305 | #endif |
| 306 | |
| 307 | static |
| 308 | int lttng_enumerate_block_devices(struct lttng_kernel_session *session) |
| 309 | { |
| 310 | struct class *ptr_block_class; |
| 311 | struct device_type *ptr_disk_type; |
| 312 | struct class_dev_iter iter; |
| 313 | struct device *dev; |
| 314 | int ret = 0; |
| 315 | |
| 316 | ptr_block_class = wrapper_get_block_class(); |
| 317 | if (!ptr_block_class) { |
| 318 | ret = -ENOSYS; |
| 319 | goto end; |
| 320 | } |
| 321 | ptr_disk_type = wrapper_get_disk_type(); |
| 322 | if (!ptr_disk_type) { |
| 323 | ret = -ENOSYS; |
| 324 | goto end; |
| 325 | } |
| 326 | class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type); |
| 327 | while ((dev = class_dev_iter_next(&iter))) { |
| 328 | struct gendisk *disk = dev_to_disk(dev); |
| 329 | |
| 330 | /* |
| 331 | * Don't show empty devices or things that have been |
| 332 | * suppressed |
| 333 | */ |
| 334 | if (get_capacity(disk) == 0 || |
| 335 | (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) |
| 336 | continue; |
| 337 | |
| 338 | ret = lttng_statedump_each_block_device(session, disk); |
| 339 | } |
| 340 | class_dev_iter_exit(&iter); |
| 341 | end: |
| 342 | return ret; |
| 343 | } |
| 344 | |
| 345 | #ifdef CONFIG_INET |
| 346 | |
| 347 | static |
| 348 | void lttng_enumerate_device(struct lttng_kernel_session *session, |
| 349 | struct net_device *dev) |
| 350 | { |
| 351 | struct in_device *in_dev; |
| 352 | struct in_ifaddr *ifa; |
| 353 | |
| 354 | if (dev->flags & IFF_UP) { |
| 355 | in_dev = in_dev_get(dev); |
| 356 | if (in_dev) { |
| 357 | for (ifa = in_dev->ifa_list; ifa != NULL; |
| 358 | ifa = ifa->ifa_next) { |
| 359 | trace_lttng_statedump_network_interface( |
| 360 | session, dev, ifa); |
| 361 | } |
| 362 | in_dev_put(in_dev); |
| 363 | } |
| 364 | } else { |
| 365 | trace_lttng_statedump_network_interface( |
| 366 | session, dev, NULL); |
| 367 | } |
| 368 | } |
| 369 | |
| 370 | static |
| 371 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
| 372 | { |
| 373 | struct net_device *dev; |
| 374 | |
| 375 | read_lock(&dev_base_lock); |
| 376 | for_each_netdev(&init_net, dev) |
| 377 | lttng_enumerate_device(session, dev); |
| 378 | read_unlock(&dev_base_lock); |
| 379 | |
| 380 | return 0; |
| 381 | } |
| 382 | #else /* CONFIG_INET */ |
| 383 | static inline |
| 384 | int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session) |
| 385 | { |
| 386 | return 0; |
| 387 | } |
| 388 | #endif /* CONFIG_INET */ |
| 389 | |
| 390 | static |
| 391 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) |
| 392 | { |
| 393 | const struct lttng_fd_ctx *ctx = p; |
| 394 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); |
| 395 | unsigned int flags = file->f_flags; |
| 396 | struct fdtable *fdt; |
| 397 | |
| 398 | /* |
| 399 | * We don't expose kernel internal flags, only userspace-visible |
| 400 | * flags. |
| 401 | */ |
| 402 | flags &= ~FMODE_NONOTIFY; |
| 403 | fdt = files_fdtable(ctx->files); |
| 404 | /* |
| 405 | * We need to check here again whether fd is within the fdt |
| 406 | * max_fds range, because we might be seeing a different |
| 407 | * files_fdtable() than iterate_fd(), assuming only RCU is |
| 408 | * protecting the read. In reality, iterate_fd() holds |
| 409 | * file_lock, which should ensure the fdt does not change while |
| 410 | * the lock is taken, but we are not aware whether this is |
| 411 | * guaranteed or not, so play safe. |
| 412 | */ |
| 413 | if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt)) |
| 414 | flags |= O_CLOEXEC; |
| 415 | if (IS_ERR(s)) { |
| 416 | struct dentry *dentry = file->f_path.dentry; |
| 417 | |
| 418 | /* Make sure we give at least some info */ |
| 419 | spin_lock(&dentry->d_lock); |
| 420 | trace_lttng_statedump_file_descriptor(ctx->session, |
| 421 | ctx->files, fd, dentry->d_name.name, flags, |
| 422 | file->f_mode); |
| 423 | spin_unlock(&dentry->d_lock); |
| 424 | goto end; |
| 425 | } |
| 426 | trace_lttng_statedump_file_descriptor(ctx->session, |
| 427 | ctx->files, fd, s, flags, file->f_mode); |
| 428 | end: |
| 429 | return 0; |
| 430 | } |
| 431 | |
| 432 | /* Called with task lock held. */ |
| 433 | static |
| 434 | void lttng_enumerate_files(struct lttng_kernel_session *session, |
| 435 | struct files_struct *files, |
| 436 | char *tmp) |
| 437 | { |
| 438 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, }; |
| 439 | |
| 440 | lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx); |
| 441 | } |
| 442 | |
| 443 | #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY |
| 444 | static |
| 445 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
| 446 | { |
| 447 | int cpu; |
| 448 | const cpumask_t *cpumask = cpu_possible_mask; |
| 449 | |
| 450 | for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids; |
| 451 | cpu = cpumask_next(cpu, cpumask)) { |
| 452 | trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu)); |
| 453 | } |
| 454 | |
| 455 | return 0; |
| 456 | } |
| 457 | #else |
| 458 | static |
| 459 | int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session) |
| 460 | { |
| 461 | return 0; |
| 462 | } |
| 463 | #endif |
| 464 | |
| 465 | #if 0 |
| 466 | /* |
| 467 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section |
| 468 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of |
| 469 | * iteration, but it is not exported to modules. |
| 470 | */ |
| 471 | static |
| 472 | void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session, |
| 473 | struct task_struct *p) |
| 474 | { |
| 475 | struct mm_struct *mm; |
| 476 | struct vm_area_struct *map; |
| 477 | unsigned long ino; |
| 478 | |
| 479 | /* get_task_mm does a task_lock... */ |
| 480 | mm = get_task_mm(p); |
| 481 | if (!mm) |
| 482 | return; |
| 483 | |
| 484 | map = mm->mmap; |
| 485 | if (map) { |
| 486 | down_read(&mm->mmap_sem); |
| 487 | while (map) { |
| 488 | if (map->vm_file) |
| 489 | ino = map->vm_file->lttng_f_dentry->d_inode->i_ino; |
| 490 | else |
| 491 | ino = 0; |
| 492 | trace_lttng_statedump_vm_map(session, p, map, ino); |
| 493 | map = map->vm_next; |
| 494 | } |
| 495 | up_read(&mm->mmap_sem); |
| 496 | } |
| 497 | mmput(mm); |
| 498 | } |
| 499 | |
| 500 | static |
| 501 | int lttng_enumerate_vm_maps(struct lttng_kernel_session *session) |
| 502 | { |
| 503 | struct task_struct *p; |
| 504 | |
| 505 | rcu_read_lock(); |
| 506 | for_each_process(p) |
| 507 | lttng_enumerate_task_vm_maps(session, p); |
| 508 | rcu_read_unlock(); |
| 509 | return 0; |
| 510 | } |
| 511 | #endif |
| 512 | |
| 513 | #ifdef CONFIG_LTTNG_HAS_LIST_IRQ |
| 514 | |
| 515 | static |
| 516 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
| 517 | { |
| 518 | unsigned int irq; |
| 519 | unsigned long flags = 0; |
| 520 | struct irq_desc *desc; |
| 521 | |
| 522 | #define irq_to_desc wrapper_irq_to_desc |
| 523 | /* needs irq_desc */ |
| 524 | for_each_irq_desc(irq, desc) { |
| 525 | struct irqaction *action; |
| 526 | const char *irq_chip_name = |
| 527 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; |
| 528 | |
| 529 | local_irq_save(flags); |
| 530 | raw_spin_lock(&desc->lock); |
| 531 | for (action = desc->action; action; action = action->next) { |
| 532 | trace_lttng_statedump_interrupt(session, |
| 533 | irq, irq_chip_name, action); |
| 534 | } |
| 535 | raw_spin_unlock(&desc->lock); |
| 536 | local_irq_restore(flags); |
| 537 | } |
| 538 | return 0; |
| 539 | #undef irq_to_desc |
| 540 | } |
| 541 | #else |
| 542 | static inline |
| 543 | int lttng_list_interrupts(struct lttng_kernel_session *session) |
| 544 | { |
| 545 | return 0; |
| 546 | } |
| 547 | #endif |
| 548 | |
| 549 | /* |
| 550 | * Statedump the task's namespaces using the proc filesystem inode number as |
| 551 | * the unique identifier. The user and pid ns are nested and will be dumped |
| 552 | * recursively. |
| 553 | * |
| 554 | * Called with task lock held. |
| 555 | */ |
| 556 | static |
| 557 | void lttng_statedump_process_ns(struct lttng_kernel_session *session, |
| 558 | struct task_struct *p, |
| 559 | enum lttng_thread_type type, |
| 560 | enum lttng_execution_mode mode, |
| 561 | enum lttng_execution_submode submode, |
| 562 | enum lttng_process_status status) |
| 563 | { |
| 564 | struct nsproxy *proxy; |
| 565 | struct pid_namespace *pid_ns; |
| 566 | struct user_namespace *user_ns; |
| 567 | |
| 568 | /* |
| 569 | * The pid and user namespaces are special, they are nested and |
| 570 | * accessed with specific functions instead of the nsproxy struct |
| 571 | * like the other namespaces. |
| 572 | */ |
| 573 | pid_ns = task_active_pid_ns(p); |
| 574 | do { |
| 575 | trace_lttng_statedump_process_pid_ns(session, p, pid_ns); |
| 576 | pid_ns = pid_ns ? pid_ns->parent : NULL; |
| 577 | } while (pid_ns); |
| 578 | |
| 579 | |
| 580 | user_ns = task_cred_xxx(p, user_ns); |
| 581 | do { |
| 582 | trace_lttng_statedump_process_user_ns(session, p, user_ns); |
| 583 | /* |
| 584 | * trace_lttng_statedump_process_user_ns() internally |
| 585 | * checks whether user_ns is NULL. While this does not |
| 586 | * appear to be a possible return value for |
| 587 | * task_cred_xxx(), err on the safe side and check |
| 588 | * for NULL here as well to be consistent with the |
| 589 | * paranoid behavior of |
| 590 | * trace_lttng_statedump_process_user_ns(). |
| 591 | */ |
| 592 | user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL; |
| 593 | } while (user_ns); |
| 594 | |
| 595 | /* |
| 596 | * Back and forth on locking strategy within Linux upstream for nsproxy. |
| 597 | * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3 |
| 598 | * "namespaces: Use task_lock and not rcu to protect nsproxy" |
| 599 | * for details. |
| 600 | */ |
| 601 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
| 602 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
| 603 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ |
| 604 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) |
| 605 | proxy = p->nsproxy; |
| 606 | #else |
| 607 | rcu_read_lock(); |
| 608 | proxy = task_nsproxy(p); |
| 609 | #endif |
| 610 | if (proxy) { |
| 611 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0)) |
| 612 | trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns); |
| 613 | #endif |
| 614 | trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns); |
| 615 | #ifndef LTTNG_MNT_NS_MISSING_HEADER |
| 616 | trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns); |
| 617 | #endif |
| 618 | trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns); |
| 619 | trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns); |
| 620 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \ |
| 621 | LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0)) |
| 622 | trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns); |
| 623 | #endif |
| 624 | } |
| 625 | #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \ |
| 626 | LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \ |
| 627 | LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \ |
| 628 | LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0)) |
| 629 | /* (nothing) */ |
| 630 | #else |
| 631 | rcu_read_unlock(); |
| 632 | #endif |
| 633 | } |
| 634 | |
| 635 | static |
| 636 | int lttng_enumerate_process_states(struct lttng_kernel_session *session) |
| 637 | { |
| 638 | struct task_struct *g, *p; |
| 639 | char *tmp; |
| 640 | |
| 641 | tmp = (char *) __get_free_page(GFP_KERNEL); |
| 642 | if (!tmp) |
| 643 | return -ENOMEM; |
| 644 | |
| 645 | rcu_read_lock(); |
| 646 | for_each_process(g) { |
| 647 | struct files_struct *prev_files = NULL; |
| 648 | |
| 649 | p = g; |
| 650 | do { |
| 651 | enum lttng_execution_mode mode = |
| 652 | LTTNG_MODE_UNKNOWN; |
| 653 | enum lttng_execution_submode submode = |
| 654 | LTTNG_UNKNOWN; |
| 655 | enum lttng_process_status status; |
| 656 | enum lttng_thread_type type; |
| 657 | struct files_struct *files; |
| 658 | |
| 659 | task_lock(p); |
| 660 | if (p->exit_state == EXIT_ZOMBIE) |
| 661 | status = LTTNG_ZOMBIE; |
| 662 | else if (p->exit_state == EXIT_DEAD) |
| 663 | status = LTTNG_DEAD; |
| 664 | else if (p->state == TASK_RUNNING) { |
| 665 | /* Is this a forked child that has not run yet? */ |
| 666 | if (list_empty(&p->rt.run_list)) |
| 667 | status = LTTNG_WAIT_FORK; |
| 668 | else |
| 669 | /* |
| 670 | * All tasks are considered as wait_cpu; |
| 671 | * the viewer will sort out if the task |
| 672 | * was really running at this time. |
| 673 | */ |
| 674 | status = LTTNG_WAIT_CPU; |
| 675 | } else if (p->state & |
| 676 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { |
| 677 | /* Task is waiting for something to complete */ |
| 678 | status = LTTNG_WAIT; |
| 679 | } else |
| 680 | status = LTTNG_UNNAMED; |
| 681 | submode = LTTNG_NONE; |
| 682 | |
| 683 | /* |
| 684 | * Verification of t->mm is to filter out kernel |
| 685 | * threads; Viewer will further filter out if a |
| 686 | * user-space thread was in syscall mode or not. |
| 687 | */ |
| 688 | if (p->mm) |
| 689 | type = LTTNG_USER_THREAD; |
| 690 | else |
| 691 | type = LTTNG_KERNEL_THREAD; |
| 692 | files = p->files; |
| 693 | |
| 694 | trace_lttng_statedump_process_state(session, |
| 695 | p, type, mode, submode, status, files); |
| 696 | lttng_statedump_process_ns(session, |
| 697 | p, type, mode, submode, status); |
| 698 | /* |
| 699 | * As an optimisation for the common case, do not |
| 700 | * repeat information for the same files_struct in |
| 701 | * two consecutive threads. This is the common case |
| 702 | * for threads sharing the same fd table. RCU guarantees |
| 703 | * that the same files_struct pointer is not re-used |
| 704 | * throughout processes/threads iteration. |
| 705 | */ |
| 706 | if (files && files != prev_files) { |
| 707 | lttng_enumerate_files(session, files, tmp); |
| 708 | prev_files = files; |
| 709 | } |
| 710 | task_unlock(p); |
| 711 | } while_each_thread(g, p); |
| 712 | } |
| 713 | rcu_read_unlock(); |
| 714 | |
| 715 | free_page((unsigned long) tmp); |
| 716 | |
| 717 | return 0; |
| 718 | } |
| 719 | |
| 720 | static |
| 721 | void lttng_statedump_work_func(struct work_struct *work) |
| 722 | { |
| 723 | if (atomic_dec_and_test(&kernel_threads_to_run)) |
| 724 | /* If we are the last thread, wake up do_lttng_statedump */ |
| 725 | wake_up(&statedump_wq); |
| 726 | } |
| 727 | |
| 728 | static |
| 729 | int do_lttng_statedump(struct lttng_kernel_session *session) |
| 730 | { |
| 731 | int cpu, ret; |
| 732 | |
| 733 | trace_lttng_statedump_start(session); |
| 734 | ret = lttng_enumerate_process_states(session); |
| 735 | if (ret) |
| 736 | return ret; |
| 737 | /* |
| 738 | * FIXME |
| 739 | * ret = lttng_enumerate_vm_maps(session); |
| 740 | * if (ret) |
| 741 | * return ret; |
| 742 | */ |
| 743 | ret = lttng_list_interrupts(session); |
| 744 | if (ret) |
| 745 | return ret; |
| 746 | ret = lttng_enumerate_network_ip_interface(session); |
| 747 | if (ret) |
| 748 | return ret; |
| 749 | ret = lttng_enumerate_block_devices(session); |
| 750 | switch (ret) { |
| 751 | case 0: |
| 752 | break; |
| 753 | case -ENOSYS: |
| 754 | printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n"); |
| 755 | break; |
| 756 | default: |
| 757 | return ret; |
| 758 | } |
| 759 | ret = lttng_enumerate_cpu_topology(session); |
| 760 | if (ret) |
| 761 | return ret; |
| 762 | |
| 763 | /* TODO lttng_dump_idt_table(session); */ |
| 764 | /* TODO lttng_dump_softirq_vec(session); */ |
| 765 | /* TODO lttng_list_modules(session); */ |
| 766 | /* TODO lttng_dump_swap_files(session); */ |
| 767 | |
| 768 | /* |
| 769 | * Fire off a work queue on each CPU. Their sole purpose in life |
| 770 | * is to guarantee that each CPU has been in a state where is was in |
| 771 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). |
| 772 | */ |
| 773 | get_online_cpus(); |
| 774 | atomic_set(&kernel_threads_to_run, num_online_cpus()); |
| 775 | for_each_online_cpu(cpu) { |
| 776 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); |
| 777 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); |
| 778 | } |
| 779 | /* Wait for all threads to run */ |
| 780 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
| 781 | put_online_cpus(); |
| 782 | /* Our work is done */ |
| 783 | trace_lttng_statedump_end(session); |
| 784 | return 0; |
| 785 | } |
| 786 | |
| 787 | /* |
| 788 | * Called with session mutex held. |
| 789 | */ |
| 790 | int lttng_statedump_start(struct lttng_kernel_session *session) |
| 791 | { |
| 792 | return do_lttng_statedump(session); |
| 793 | } |
| 794 | EXPORT_SYMBOL_GPL(lttng_statedump_start); |
| 795 | |
| 796 | static |
| 797 | int __init lttng_statedump_init(void) |
| 798 | { |
| 799 | /* |
| 800 | * Allow module to load even if the fixup cannot be done. This |
| 801 | * will allow seemless transition when the underlying issue fix |
| 802 | * is merged into the Linux kernel, and when tracepoint.c |
| 803 | * "tracepoint_module_notify" is turned into a static function. |
| 804 | */ |
| 805 | (void) wrapper_lttng_fixup_sig(THIS_MODULE); |
| 806 | return 0; |
| 807 | } |
| 808 | |
| 809 | module_init(lttng_statedump_init); |
| 810 | |
| 811 | static |
| 812 | void __exit lttng_statedump_exit(void) |
| 813 | { |
| 814 | } |
| 815 | |
| 816 | module_exit(lttng_statedump_exit); |
| 817 | |
| 818 | MODULE_LICENSE("GPL and additional rights"); |
| 819 | MODULE_AUTHOR("Jean-Hugues Deschenes"); |
| 820 | MODULE_DESCRIPTION("LTTng statedump provider"); |
| 821 | MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "." |
| 822 | __stringify(LTTNG_MODULES_MINOR_VERSION) "." |
| 823 | __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION) |
| 824 | LTTNG_MODULES_EXTRAVERSION); |