| 1 | /* |
| 2 | * lttng-statedump.c |
| 3 | * |
| 4 | * Linux Trace Toolkit Next Generation Kernel State Dump |
| 5 | * |
| 6 | * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca> |
| 7 | * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 8 | * |
| 9 | * This library is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU Lesser General Public |
| 11 | * License as published by the Free Software Foundation; only |
| 12 | * version 2.1 of the License. |
| 13 | * |
| 14 | * This library is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | * Lesser General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU Lesser General Public |
| 20 | * License along with this library; if not, write to the Free Software |
| 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 22 | * |
| 23 | * Changes: |
| 24 | * Eric Clement: Add listing of network IP interface |
| 25 | * 2006, 2007 Mathieu Desnoyers Fix kernel threads |
| 26 | * Various updates |
| 27 | */ |
| 28 | |
| 29 | #include <linux/init.h> |
| 30 | #include <linux/module.h> |
| 31 | #include <linux/netlink.h> |
| 32 | #include <linux/inet.h> |
| 33 | #include <linux/ip.h> |
| 34 | #include <linux/kthread.h> |
| 35 | #include <linux/proc_fs.h> |
| 36 | #include <linux/file.h> |
| 37 | #include <linux/interrupt.h> |
| 38 | #include <linux/irqnr.h> |
| 39 | #include <linux/cpu.h> |
| 40 | #include <linux/netdevice.h> |
| 41 | #include <linux/inetdevice.h> |
| 42 | #include <linux/sched.h> |
| 43 | #include <linux/mm.h> |
| 44 | #include <linux/fdtable.h> |
| 45 | #include <linux/swap.h> |
| 46 | #include <linux/wait.h> |
| 47 | #include <linux/mutex.h> |
| 48 | |
| 49 | #include "lttng-events.h" |
| 50 | #include "wrapper/irqdesc.h" |
| 51 | #include "wrapper/spinlock.h" |
| 52 | #include "wrapper/fdtable.h" |
| 53 | |
| 54 | #ifdef CONFIG_GENERIC_HARDIRQS |
| 55 | #include <linux/irq.h> |
| 56 | #endif |
| 57 | |
| 58 | /* Define the tracepoints, but do not build the probes */ |
| 59 | #define CREATE_TRACE_POINTS |
| 60 | #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module |
| 61 | #define TRACE_INCLUDE_FILE lttng-statedump |
| 62 | #include "instrumentation/events/lttng-module/lttng-statedump.h" |
| 63 | |
| 64 | struct lttng_fd_ctx { |
| 65 | char *page; |
| 66 | struct lttng_session *session; |
| 67 | struct task_struct *p; |
| 68 | }; |
| 69 | |
| 70 | /* |
| 71 | * Protected by the trace lock. |
| 72 | */ |
| 73 | static struct delayed_work cpu_work[NR_CPUS]; |
| 74 | static DECLARE_WAIT_QUEUE_HEAD(statedump_wq); |
| 75 | static atomic_t kernel_threads_to_run; |
| 76 | |
| 77 | enum lttng_thread_type { |
| 78 | LTTNG_USER_THREAD = 0, |
| 79 | LTTNG_KERNEL_THREAD = 1, |
| 80 | }; |
| 81 | |
| 82 | enum lttng_execution_mode { |
| 83 | LTTNG_USER_MODE = 0, |
| 84 | LTTNG_SYSCALL = 1, |
| 85 | LTTNG_TRAP = 2, |
| 86 | LTTNG_IRQ = 3, |
| 87 | LTTNG_SOFTIRQ = 4, |
| 88 | LTTNG_MODE_UNKNOWN = 5, |
| 89 | }; |
| 90 | |
| 91 | enum lttng_execution_submode { |
| 92 | LTTNG_NONE = 0, |
| 93 | LTTNG_UNKNOWN = 1, |
| 94 | }; |
| 95 | |
| 96 | enum lttng_process_status { |
| 97 | LTTNG_UNNAMED = 0, |
| 98 | LTTNG_WAIT_FORK = 1, |
| 99 | LTTNG_WAIT_CPU = 2, |
| 100 | LTTNG_EXIT = 3, |
| 101 | LTTNG_ZOMBIE = 4, |
| 102 | LTTNG_WAIT = 5, |
| 103 | LTTNG_RUN = 6, |
| 104 | LTTNG_DEAD = 7, |
| 105 | }; |
| 106 | |
| 107 | #ifdef CONFIG_INET |
| 108 | static |
| 109 | void lttng_enumerate_device(struct lttng_session *session, |
| 110 | struct net_device *dev) |
| 111 | { |
| 112 | struct in_device *in_dev; |
| 113 | struct in_ifaddr *ifa; |
| 114 | |
| 115 | if (dev->flags & IFF_UP) { |
| 116 | in_dev = in_dev_get(dev); |
| 117 | if (in_dev) { |
| 118 | for (ifa = in_dev->ifa_list; ifa != NULL; |
| 119 | ifa = ifa->ifa_next) { |
| 120 | trace_lttng_statedump_network_interface( |
| 121 | session, dev, ifa); |
| 122 | } |
| 123 | in_dev_put(in_dev); |
| 124 | } |
| 125 | } else { |
| 126 | trace_lttng_statedump_network_interface( |
| 127 | session, dev, NULL); |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | static |
| 132 | int lttng_enumerate_network_ip_interface(struct lttng_session *session) |
| 133 | { |
| 134 | struct net_device *dev; |
| 135 | |
| 136 | read_lock(&dev_base_lock); |
| 137 | for_each_netdev(&init_net, dev) |
| 138 | lttng_enumerate_device(session, dev); |
| 139 | read_unlock(&dev_base_lock); |
| 140 | |
| 141 | return 0; |
| 142 | } |
| 143 | #else /* CONFIG_INET */ |
| 144 | static inline |
| 145 | int lttng_enumerate_network_ip_interface(struct lttng_session *session) |
| 146 | { |
| 147 | return 0; |
| 148 | } |
| 149 | #endif /* CONFIG_INET */ |
| 150 | |
| 151 | static |
| 152 | int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) |
| 153 | { |
| 154 | const struct lttng_fd_ctx *ctx = p; |
| 155 | const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE); |
| 156 | |
| 157 | if (IS_ERR(s)) { |
| 158 | struct dentry *dentry = file->f_path.dentry; |
| 159 | |
| 160 | /* Make sure we give at least some info */ |
| 161 | spin_lock(&dentry->d_lock); |
| 162 | trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, |
| 163 | dentry->d_name.name); |
| 164 | spin_unlock(&dentry->d_lock); |
| 165 | goto end; |
| 166 | } |
| 167 | trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s); |
| 168 | end: |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | static |
| 173 | void lttng_enumerate_task_fd(struct lttng_session *session, |
| 174 | struct task_struct *p, char *tmp) |
| 175 | { |
| 176 | struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p }; |
| 177 | |
| 178 | task_lock(p); |
| 179 | lttng_iterate_fd(p->files, 0, lttng_dump_one_fd, &ctx); |
| 180 | task_unlock(p); |
| 181 | } |
| 182 | |
| 183 | static |
| 184 | int lttng_enumerate_file_descriptors(struct lttng_session *session) |
| 185 | { |
| 186 | struct task_struct *p; |
| 187 | char *tmp = (char *) __get_free_page(GFP_KERNEL); |
| 188 | |
| 189 | /* Enumerate active file descriptors */ |
| 190 | rcu_read_lock(); |
| 191 | for_each_process(p) |
| 192 | lttng_enumerate_task_fd(session, p, tmp); |
| 193 | rcu_read_unlock(); |
| 194 | free_page((unsigned long) tmp); |
| 195 | return 0; |
| 196 | } |
| 197 | |
| 198 | #if 0 |
| 199 | /* |
| 200 | * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section |
| 201 | * (scheduling in atomic). Normally, the tasklist lock protects this kind of |
| 202 | * iteration, but it is not exported to modules. |
| 203 | */ |
| 204 | static |
| 205 | void lttng_enumerate_task_vm_maps(struct lttng_session *session, |
| 206 | struct task_struct *p) |
| 207 | { |
| 208 | struct mm_struct *mm; |
| 209 | struct vm_area_struct *map; |
| 210 | unsigned long ino; |
| 211 | |
| 212 | /* get_task_mm does a task_lock... */ |
| 213 | mm = get_task_mm(p); |
| 214 | if (!mm) |
| 215 | return; |
| 216 | |
| 217 | map = mm->mmap; |
| 218 | if (map) { |
| 219 | down_read(&mm->mmap_sem); |
| 220 | while (map) { |
| 221 | if (map->vm_file) |
| 222 | ino = map->vm_file->f_dentry->d_inode->i_ino; |
| 223 | else |
| 224 | ino = 0; |
| 225 | trace_lttng_statedump_vm_map(session, p, map, ino); |
| 226 | map = map->vm_next; |
| 227 | } |
| 228 | up_read(&mm->mmap_sem); |
| 229 | } |
| 230 | mmput(mm); |
| 231 | } |
| 232 | |
| 233 | static |
| 234 | int lttng_enumerate_vm_maps(struct lttng_session *session) |
| 235 | { |
| 236 | struct task_struct *p; |
| 237 | |
| 238 | rcu_read_lock(); |
| 239 | for_each_process(p) |
| 240 | lttng_enumerate_task_vm_maps(session, p); |
| 241 | rcu_read_unlock(); |
| 242 | return 0; |
| 243 | } |
| 244 | #endif |
| 245 | |
| 246 | #ifdef CONFIG_GENERIC_HARDIRQS |
| 247 | |
| 248 | #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)) |
| 249 | #define irq_desc_get_chip(desc) get_irq_desc_chip(desc) |
| 250 | #endif |
| 251 | |
| 252 | static |
| 253 | void lttng_list_interrupts(struct lttng_session *session) |
| 254 | { |
| 255 | unsigned int irq; |
| 256 | unsigned long flags = 0; |
| 257 | struct irq_desc *desc; |
| 258 | |
| 259 | #define irq_to_desc wrapper_irq_to_desc |
| 260 | /* needs irq_desc */ |
| 261 | for_each_irq_desc(irq, desc) { |
| 262 | struct irqaction *action; |
| 263 | const char *irq_chip_name = |
| 264 | irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip"; |
| 265 | |
| 266 | local_irq_save(flags); |
| 267 | wrapper_desc_spin_lock(&desc->lock); |
| 268 | for (action = desc->action; action; action = action->next) { |
| 269 | trace_lttng_statedump_interrupt(session, |
| 270 | irq, irq_chip_name, action); |
| 271 | } |
| 272 | wrapper_desc_spin_unlock(&desc->lock); |
| 273 | local_irq_restore(flags); |
| 274 | } |
| 275 | #undef irq_to_desc |
| 276 | } |
| 277 | #else |
| 278 | static inline |
| 279 | void list_interrupts(struct lttng_session *session) |
| 280 | { |
| 281 | } |
| 282 | #endif |
| 283 | |
| 284 | static |
| 285 | void lttng_statedump_process_ns(struct lttng_session *session, |
| 286 | struct task_struct *p, |
| 287 | enum lttng_thread_type type, |
| 288 | enum lttng_execution_mode mode, |
| 289 | enum lttng_execution_submode submode, |
| 290 | enum lttng_process_status status) |
| 291 | { |
| 292 | struct nsproxy *proxy; |
| 293 | struct pid_namespace *pid_ns; |
| 294 | |
| 295 | rcu_read_lock(); |
| 296 | proxy = task_nsproxy(p); |
| 297 | if (proxy) { |
| 298 | pid_ns = proxy->pid_ns; |
| 299 | do { |
| 300 | trace_lttng_statedump_process_state(session, |
| 301 | p, type, mode, submode, status, pid_ns); |
| 302 | pid_ns = pid_ns->parent; |
| 303 | } while (pid_ns); |
| 304 | } else { |
| 305 | trace_lttng_statedump_process_state(session, |
| 306 | p, type, mode, submode, status, NULL); |
| 307 | } |
| 308 | rcu_read_unlock(); |
| 309 | } |
| 310 | |
| 311 | static |
| 312 | int lttng_enumerate_process_states(struct lttng_session *session) |
| 313 | { |
| 314 | struct task_struct *g, *p; |
| 315 | |
| 316 | rcu_read_lock(); |
| 317 | for_each_process(g) { |
| 318 | p = g; |
| 319 | do { |
| 320 | enum lttng_execution_mode mode = |
| 321 | LTTNG_MODE_UNKNOWN; |
| 322 | enum lttng_execution_submode submode = |
| 323 | LTTNG_UNKNOWN; |
| 324 | enum lttng_process_status status; |
| 325 | enum lttng_thread_type type; |
| 326 | |
| 327 | task_lock(p); |
| 328 | if (p->exit_state == EXIT_ZOMBIE) |
| 329 | status = LTTNG_ZOMBIE; |
| 330 | else if (p->exit_state == EXIT_DEAD) |
| 331 | status = LTTNG_DEAD; |
| 332 | else if (p->state == TASK_RUNNING) { |
| 333 | /* Is this a forked child that has not run yet? */ |
| 334 | if (list_empty(&p->rt.run_list)) |
| 335 | status = LTTNG_WAIT_FORK; |
| 336 | else |
| 337 | /* |
| 338 | * All tasks are considered as wait_cpu; |
| 339 | * the viewer will sort out if the task |
| 340 | * was really running at this time. |
| 341 | */ |
| 342 | status = LTTNG_WAIT_CPU; |
| 343 | } else if (p->state & |
| 344 | (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) { |
| 345 | /* Task is waiting for something to complete */ |
| 346 | status = LTTNG_WAIT; |
| 347 | } else |
| 348 | status = LTTNG_UNNAMED; |
| 349 | submode = LTTNG_NONE; |
| 350 | |
| 351 | /* |
| 352 | * Verification of t->mm is to filter out kernel |
| 353 | * threads; Viewer will further filter out if a |
| 354 | * user-space thread was in syscall mode or not. |
| 355 | */ |
| 356 | if (p->mm) |
| 357 | type = LTTNG_USER_THREAD; |
| 358 | else |
| 359 | type = LTTNG_KERNEL_THREAD; |
| 360 | lttng_statedump_process_ns(session, |
| 361 | p, type, mode, submode, status); |
| 362 | task_unlock(p); |
| 363 | } while_each_thread(g, p); |
| 364 | } |
| 365 | rcu_read_unlock(); |
| 366 | |
| 367 | return 0; |
| 368 | } |
| 369 | |
| 370 | static |
| 371 | void lttng_statedump_work_func(struct work_struct *work) |
| 372 | { |
| 373 | if (atomic_dec_and_test(&kernel_threads_to_run)) |
| 374 | /* If we are the last thread, wake up do_lttng_statedump */ |
| 375 | wake_up(&statedump_wq); |
| 376 | } |
| 377 | |
| 378 | static |
| 379 | int do_lttng_statedump(struct lttng_session *session) |
| 380 | { |
| 381 | int cpu; |
| 382 | |
| 383 | printk(KERN_DEBUG "LTT state dump thread start\n"); |
| 384 | trace_lttng_statedump_start(session); |
| 385 | lttng_enumerate_process_states(session); |
| 386 | lttng_enumerate_file_descriptors(session); |
| 387 | /* FIXME lttng_enumerate_vm_maps(session); */ |
| 388 | lttng_list_interrupts(session); |
| 389 | lttng_enumerate_network_ip_interface(session); |
| 390 | |
| 391 | /* TODO lttng_dump_idt_table(session); */ |
| 392 | /* TODO lttng_dump_softirq_vec(session); */ |
| 393 | /* TODO lttng_list_modules(session); */ |
| 394 | /* TODO lttng_dump_swap_files(session); */ |
| 395 | |
| 396 | /* |
| 397 | * Fire off a work queue on each CPU. Their sole purpose in life |
| 398 | * is to guarantee that each CPU has been in a state where is was in |
| 399 | * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ). |
| 400 | */ |
| 401 | get_online_cpus(); |
| 402 | atomic_set(&kernel_threads_to_run, num_online_cpus()); |
| 403 | for_each_online_cpu(cpu) { |
| 404 | INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func); |
| 405 | schedule_delayed_work_on(cpu, &cpu_work[cpu], 0); |
| 406 | } |
| 407 | /* Wait for all threads to run */ |
| 408 | __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0)); |
| 409 | put_online_cpus(); |
| 410 | /* Our work is done */ |
| 411 | printk(KERN_DEBUG "LTT state dump end\n"); |
| 412 | trace_lttng_statedump_end(session); |
| 413 | return 0; |
| 414 | } |
| 415 | |
| 416 | /* |
| 417 | * Called with session mutex held. |
| 418 | */ |
| 419 | int lttng_statedump_start(struct lttng_session *session) |
| 420 | { |
| 421 | printk(KERN_DEBUG "LTTng: state dump begin\n"); |
| 422 | return do_lttng_statedump(session); |
| 423 | } |
| 424 | EXPORT_SYMBOL_GPL(lttng_statedump_start); |
| 425 | |
| 426 | MODULE_LICENSE("GPL and additional rights"); |
| 427 | MODULE_AUTHOR("Jean-Hugues Deschenes"); |
| 428 | MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump"); |