lttng-statedump-impl.c

   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 "lttng-tracer.h"
  51 #include "wrapper/irqdesc.h"
  52 #include "wrapper/spinlock.h"
  53 #include "wrapper/fdtable.h"
  54 #include "wrapper/nsproxy.h"
  55 #include "wrapper/irq.h"
  56 #include "wrapper/tracepoint.h"
  57
  58 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  59 #include <linux/irq.h>
  60 #endif
  61
  62 /* Define the tracepoints, but do not build the probes */
  63 #define CREATE_TRACE_POINTS
  64 #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
  65 #define TRACE_INCLUDE_FILE lttng-statedump
  66 #include "instrumentation/events/lttng-module/lttng-statedump.h"
  67
  68 DEFINE_TRACE(lttng_statedump_end);
  69 DEFINE_TRACE(lttng_statedump_interrupt);
  70 DEFINE_TRACE(lttng_statedump_file_descriptor);
  71 DEFINE_TRACE(lttng_statedump_start);
  72 DEFINE_TRACE(lttng_statedump_process_state);
  73 DEFINE_TRACE(lttng_statedump_network_interface);
  74
  75 struct lttng_fd_ctx {
  76         char *page;
  77         struct lttng_session *session;
  78         struct task_struct *p;
  79 };
  80
  81 /*
  82  * Protected by the trace lock.
  83  */
  84 static struct delayed_work cpu_work[NR_CPUS];
  85 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  86 static atomic_t kernel_threads_to_run;
  87
  88 enum lttng_thread_type {
  89         LTTNG_USER_THREAD = 0,
  90         LTTNG_KERNEL_THREAD = 1,
  91 };
  92
  93 enum lttng_execution_mode {
  94         LTTNG_USER_MODE = 0,
  95         LTTNG_SYSCALL = 1,
  96         LTTNG_TRAP = 2,
  97         LTTNG_IRQ = 3,
  98         LTTNG_SOFTIRQ = 4,
  99         LTTNG_MODE_UNKNOWN = 5,
 100 };
 101
 102 enum lttng_execution_submode {
 103         LTTNG_NONE = 0,
 104         LTTNG_UNKNOWN = 1,
 105 };
 106
 107 enum lttng_process_status {
 108         LTTNG_UNNAMED = 0,
 109         LTTNG_WAIT_FORK = 1,
 110         LTTNG_WAIT_CPU = 2,
 111         LTTNG_EXIT = 3,
 112         LTTNG_ZOMBIE = 4,
 113         LTTNG_WAIT = 5,
 114         LTTNG_RUN = 6,
 115         LTTNG_DEAD = 7,
 116 };
 117
 118 #ifdef CONFIG_INET
 119 static
 120 void lttng_enumerate_device(struct lttng_session *session,
 121                 struct net_device *dev)
 122 {
 123         struct in_device *in_dev;
 124         struct in_ifaddr *ifa;
 125
 126         if (dev->flags & IFF_UP) {
 127                 in_dev = in_dev_get(dev);
 128                 if (in_dev) {
 129                         for (ifa = in_dev->ifa_list; ifa != NULL;
 130                              ifa = ifa->ifa_next) {
 131                                 trace_lttng_statedump_network_interface(
 132                                         session, dev, ifa);
 133                         }
 134                         in_dev_put(in_dev);
 135                 }
 136         } else {
 137                 trace_lttng_statedump_network_interface(
 138                         session, dev, NULL);
 139         }
 140 }
 141
 142 static
 143 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 144 {
 145         struct net_device *dev;
 146
 147         read_lock(&dev_base_lock);
 148         for_each_netdev(&init_net, dev)
 149                 lttng_enumerate_device(session, dev);
 150         read_unlock(&dev_base_lock);
 151
 152         return 0;
 153 }
 154 #else /* CONFIG_INET */
 155 static inline
 156 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 157 {
 158         return 0;
 159 }
 160 #endif /* CONFIG_INET */
 161
 162 static
 163 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 164 {
 165         const struct lttng_fd_ctx *ctx = p;
 166         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 167
 168         if (IS_ERR(s)) {
 169                 struct dentry *dentry = file->f_path.dentry;
 170
 171                 /* Make sure we give at least some info */
 172                 spin_lock(&dentry->d_lock);
 173                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 174                         dentry->d_name.name);
 175                 spin_unlock(&dentry->d_lock);
 176                 goto end;
 177         }
 178         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s);
 179 end:
 180         return 0;
 181 }
 182
 183 static
 184 void lttng_enumerate_task_fd(struct lttng_session *session,
 185                 struct task_struct *p, char *tmp)
 186 {
 187         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 188
 189         task_lock(p);
 190         lttng_iterate_fd(p->files, 0, lttng_dump_one_fd, &ctx);
 191         task_unlock(p);
 192 }
 193
 194 static
 195 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 196 {
 197         struct task_struct *p;
 198         char *tmp = (char *) __get_free_page(GFP_KERNEL);
 199
 200         /* Enumerate active file descriptors */
 201         rcu_read_lock();
 202         for_each_process(p)
 203                 lttng_enumerate_task_fd(session, p, tmp);
 204         rcu_read_unlock();
 205         free_page((unsigned long) tmp);
 206         return 0;
 207 }
 208
 209 #if 0
 210 /*
 211  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 212  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 213  * iteration, but it is not exported to modules.
 214  */
 215 static
 216 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 217                 struct task_struct *p)
 218 {
 219         struct mm_struct *mm;
 220         struct vm_area_struct *map;
 221         unsigned long ino;
 222
 223         /* get_task_mm does a task_lock... */
 224         mm = get_task_mm(p);
 225         if (!mm)
 226                 return;
 227
 228         map = mm->mmap;
 229         if (map) {
 230                 down_read(&mm->mmap_sem);
 231                 while (map) {
 232                         if (map->vm_file)
 233                                 ino = map->vm_file->f_dentry->d_inode->i_ino;
 234                         else
 235                                 ino = 0;
 236                         trace_lttng_statedump_vm_map(session, p, map, ino);
 237                         map = map->vm_next;
 238                 }
 239                 up_read(&mm->mmap_sem);
 240         }
 241         mmput(mm);
 242 }
 243
 244 static
 245 int lttng_enumerate_vm_maps(struct lttng_session *session)
 246 {
 247         struct task_struct *p;
 248
 249         rcu_read_lock();
 250         for_each_process(p)
 251                 lttng_enumerate_task_vm_maps(session, p);
 252         rcu_read_unlock();
 253         return 0;
 254 }
 255 #endif
 256
 257 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 258
 259 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 260 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 261 #endif
 262
 263 static
 264 void lttng_list_interrupts(struct lttng_session *session)
 265 {
 266         unsigned int irq;
 267         unsigned long flags = 0;
 268         struct irq_desc *desc;
 269
 270 #define irq_to_desc     wrapper_irq_to_desc
 271         /* needs irq_desc */
 272         for_each_irq_desc(irq, desc) {
 273                 struct irqaction *action;
 274                 const char *irq_chip_name =
 275                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 276
 277                 local_irq_save(flags);
 278                 wrapper_desc_spin_lock(&desc->lock);
 279                 for (action = desc->action; action; action = action->next) {
 280                         trace_lttng_statedump_interrupt(session,
 281                                 irq, irq_chip_name, action);
 282                 }
 283                 wrapper_desc_spin_unlock(&desc->lock);
 284                 local_irq_restore(flags);
 285         }
 286 #undef irq_to_desc
 287 }
 288 #else
 289 static inline
 290 void lttng_list_interrupts(struct lttng_session *session)
 291 {
 292 }
 293 #endif
 294
 295 static
 296 void lttng_statedump_process_ns(struct lttng_session *session,
 297                 struct task_struct *p,
 298                 enum lttng_thread_type type,
 299                 enum lttng_execution_mode mode,
 300                 enum lttng_execution_submode submode,
 301                 enum lttng_process_status status)
 302 {
 303         struct nsproxy *proxy;
 304         struct pid_namespace *pid_ns;
 305
 306         rcu_read_lock();
 307         proxy = task_nsproxy(p);
 308         if (proxy) {
 309                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 310                 do {
 311                         trace_lttng_statedump_process_state(session,
 312                                 p, type, mode, submode, status, pid_ns);
 313                         pid_ns = pid_ns->parent;
 314                 } while (pid_ns);
 315         } else {
 316                 trace_lttng_statedump_process_state(session,
 317                         p, type, mode, submode, status, NULL);
 318         }
 319         rcu_read_unlock();
 320 }
 321
 322 static
 323 int lttng_enumerate_process_states(struct lttng_session *session)
 324 {
 325         struct task_struct *g, *p;
 326
 327         rcu_read_lock();
 328         for_each_process(g) {
 329                 p = g;
 330                 do {
 331                         enum lttng_execution_mode mode =
 332                                 LTTNG_MODE_UNKNOWN;
 333                         enum lttng_execution_submode submode =
 334                                 LTTNG_UNKNOWN;
 335                         enum lttng_process_status status;
 336                         enum lttng_thread_type type;
 337
 338                         task_lock(p);
 339                         if (p->exit_state == EXIT_ZOMBIE)
 340                                 status = LTTNG_ZOMBIE;
 341                         else if (p->exit_state == EXIT_DEAD)
 342                                 status = LTTNG_DEAD;
 343                         else if (p->state == TASK_RUNNING) {
 344                                 /* Is this a forked child that has not run yet? */
 345                                 if (list_empty(&p->rt.run_list))
 346                                         status = LTTNG_WAIT_FORK;
 347                                 else
 348                                         /*
 349                                          * All tasks are considered as wait_cpu;
 350                                          * the viewer will sort out if the task
 351                                          * was really running at this time.
 352                                          */
 353                                         status = LTTNG_WAIT_CPU;
 354                         } else if (p->state &
 355                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 356                                 /* Task is waiting for something to complete */
 357                                 status = LTTNG_WAIT;
 358                         } else
 359                                 status = LTTNG_UNNAMED;
 360                         submode = LTTNG_NONE;
 361
 362                         /*
 363                          * Verification of t->mm is to filter out kernel
 364                          * threads; Viewer will further filter out if a
 365                          * user-space thread was in syscall mode or not.
 366                          */
 367                         if (p->mm)
 368                                 type = LTTNG_USER_THREAD;
 369                         else
 370                                 type = LTTNG_KERNEL_THREAD;
 371                         lttng_statedump_process_ns(session,
 372                                 p, type, mode, submode, status);
 373                         task_unlock(p);
 374                 } while_each_thread(g, p);
 375         }
 376         rcu_read_unlock();
 377
 378         return 0;
 379 }
 380
 381 static
 382 void lttng_statedump_work_func(struct work_struct *work)
 383 {
 384         if (atomic_dec_and_test(&kernel_threads_to_run))
 385                 /* If we are the last thread, wake up do_lttng_statedump */
 386                 wake_up(&statedump_wq);
 387 }
 388
 389 static
 390 int do_lttng_statedump(struct lttng_session *session)
 391 {
 392         int cpu;
 393
 394         trace_lttng_statedump_start(session);
 395         lttng_enumerate_process_states(session);
 396         lttng_enumerate_file_descriptors(session);
 397         /* FIXME lttng_enumerate_vm_maps(session); */
 398         lttng_list_interrupts(session);
 399         lttng_enumerate_network_ip_interface(session);
 400
 401         /* TODO lttng_dump_idt_table(session); */
 402         /* TODO lttng_dump_softirq_vec(session); */
 403         /* TODO lttng_list_modules(session); */
 404         /* TODO lttng_dump_swap_files(session); */
 405
 406         /*
 407          * Fire off a work queue on each CPU. Their sole purpose in life
 408          * is to guarantee that each CPU has been in a state where is was in
 409          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 410          */
 411         get_online_cpus();
 412         atomic_set(&kernel_threads_to_run, num_online_cpus());
 413         for_each_online_cpu(cpu) {
 414                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 415                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 416         }
 417         /* Wait for all threads to run */
 418         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 419         put_online_cpus();
 420         /* Our work is done */
 421         trace_lttng_statedump_end(session);
 422         return 0;
 423 }
 424
 425 /*
 426  * Called with session mutex held.
 427  */
 428 int lttng_statedump_start(struct lttng_session *session)
 429 {
 430         return do_lttng_statedump(session);
 431 }
 432 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 433
 434 static
 435 int __init lttng_statedump_init(void)
 436 {
 437         /*
 438          * Allow module to load even if the fixup cannot be done. This
 439          * will allow seemless transition when the underlying issue fix
 440          * is merged into the Linux kernel, and when tracepoint.c
 441          * "tracepoint_module_notify" is turned into a static function.
 442          */
 443         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 444         return 0;
 445 }
 446
 447 module_init(lttng_statedump_init);
 448
 449 static
 450 void __exit lttng_statedump_exit(void)
 451 {
 452 }
 453
 454 module_exit(lttng_statedump_exit);
 455
 456 MODULE_LICENSE("GPL and additional rights");
 457 MODULE_AUTHOR("Jean-Hugues Deschenes");
 458 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 459 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 460         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 461         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 462         LTTNG_MODULES_EXTRAVERSION);