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 struct lttng_fd_ctx {
  69         char *page;
  70         struct lttng_session *session;
  71         struct task_struct *p;
  72 };
  73
  74 /*
  75  * Protected by the trace lock.
  76  */
  77 static struct delayed_work cpu_work[NR_CPUS];
  78 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  79 static atomic_t kernel_threads_to_run;
  80
  81 enum lttng_thread_type {
  82         LTTNG_USER_THREAD = 0,
  83         LTTNG_KERNEL_THREAD = 1,
  84 };
  85
  86 enum lttng_execution_mode {
  87         LTTNG_USER_MODE = 0,
  88         LTTNG_SYSCALL = 1,
  89         LTTNG_TRAP = 2,
  90         LTTNG_IRQ = 3,
  91         LTTNG_SOFTIRQ = 4,
  92         LTTNG_MODE_UNKNOWN = 5,
  93 };
  94
  95 enum lttng_execution_submode {
  96         LTTNG_NONE = 0,
  97         LTTNG_UNKNOWN = 1,
  98 };
  99
 100 enum lttng_process_status {
 101         LTTNG_UNNAMED = 0,
 102         LTTNG_WAIT_FORK = 1,
 103         LTTNG_WAIT_CPU = 2,
 104         LTTNG_EXIT = 3,
 105         LTTNG_ZOMBIE = 4,
 106         LTTNG_WAIT = 5,
 107         LTTNG_RUN = 6,
 108         LTTNG_DEAD = 7,
 109 };
 110
 111 #ifdef CONFIG_INET
 112 static
 113 void lttng_enumerate_device(struct lttng_session *session,
 114                 struct net_device *dev)
 115 {
 116         struct in_device *in_dev;
 117         struct in_ifaddr *ifa;
 118
 119         if (dev->flags & IFF_UP) {
 120                 in_dev = in_dev_get(dev);
 121                 if (in_dev) {
 122                         for (ifa = in_dev->ifa_list; ifa != NULL;
 123                              ifa = ifa->ifa_next) {
 124                                 trace_lttng_statedump_network_interface(
 125                                         session, dev, ifa);
 126                         }
 127                         in_dev_put(in_dev);
 128                 }
 129         } else {
 130                 trace_lttng_statedump_network_interface(
 131                         session, dev, NULL);
 132         }
 133 }
 134
 135 static
 136 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 137 {
 138         struct net_device *dev;
 139
 140         read_lock(&dev_base_lock);
 141         for_each_netdev(&init_net, dev)
 142                 lttng_enumerate_device(session, dev);
 143         read_unlock(&dev_base_lock);
 144
 145         return 0;
 146 }
 147 #else /* CONFIG_INET */
 148 static inline
 149 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 150 {
 151         return 0;
 152 }
 153 #endif /* CONFIG_INET */
 154
 155 static
 156 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 157 {
 158         const struct lttng_fd_ctx *ctx = p;
 159         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 160
 161         if (IS_ERR(s)) {
 162                 struct dentry *dentry = file->f_path.dentry;
 163
 164                 /* Make sure we give at least some info */
 165                 spin_lock(&dentry->d_lock);
 166                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 167                         dentry->d_name.name);
 168                 spin_unlock(&dentry->d_lock);
 169                 goto end;
 170         }
 171         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s);
 172 end:
 173         return 0;
 174 }
 175
 176 static
 177 void lttng_enumerate_task_fd(struct lttng_session *session,
 178                 struct task_struct *p, char *tmp)
 179 {
 180         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 181
 182         task_lock(p);
 183         lttng_iterate_fd(p->files, 0, lttng_dump_one_fd, &ctx);
 184         task_unlock(p);
 185 }
 186
 187 static
 188 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 189 {
 190         struct task_struct *p;
 191         char *tmp;
 192
 193         tmp = (char *) __get_free_page(GFP_KERNEL);
 194         if (!tmp)
 195                 return -ENOMEM;
 196
 197         /* Enumerate active file descriptors */
 198         rcu_read_lock();
 199         for_each_process(p)
 200                 lttng_enumerate_task_fd(session, p, tmp);
 201         rcu_read_unlock();
 202         free_page((unsigned long) tmp);
 203         return 0;
 204 }
 205
 206 #if 0
 207 /*
 208  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 209  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 210  * iteration, but it is not exported to modules.
 211  */
 212 static
 213 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 214                 struct task_struct *p)
 215 {
 216         struct mm_struct *mm;
 217         struct vm_area_struct *map;
 218         unsigned long ino;
 219
 220         /* get_task_mm does a task_lock... */
 221         mm = get_task_mm(p);
 222         if (!mm)
 223                 return;
 224
 225         map = mm->mmap;
 226         if (map) {
 227                 down_read(&mm->mmap_sem);
 228                 while (map) {
 229                         if (map->vm_file)
 230                                 ino = map->vm_file->f_dentry->d_inode->i_ino;
 231                         else
 232                                 ino = 0;
 233                         trace_lttng_statedump_vm_map(session, p, map, ino);
 234                         map = map->vm_next;
 235                 }
 236                 up_read(&mm->mmap_sem);
 237         }
 238         mmput(mm);
 239 }
 240
 241 static
 242 int lttng_enumerate_vm_maps(struct lttng_session *session)
 243 {
 244         struct task_struct *p;
 245
 246         rcu_read_lock();
 247         for_each_process(p)
 248                 lttng_enumerate_task_vm_maps(session, p);
 249         rcu_read_unlock();
 250         return 0;
 251 }
 252 #endif
 253
 254 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 255
 256 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 257 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 258 #endif
 259
 260 static
 261 int lttng_list_interrupts(struct lttng_session *session)
 262 {
 263         unsigned int irq;
 264         unsigned long flags = 0;
 265         struct irq_desc *desc;
 266
 267 #define irq_to_desc     wrapper_irq_to_desc
 268         /* needs irq_desc */
 269         for_each_irq_desc(irq, desc) {
 270                 struct irqaction *action;
 271                 const char *irq_chip_name =
 272                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 273
 274                 local_irq_save(flags);
 275                 wrapper_desc_spin_lock(&desc->lock);
 276                 for (action = desc->action; action; action = action->next) {
 277                         trace_lttng_statedump_interrupt(session,
 278                                 irq, irq_chip_name, action);
 279                 }
 280                 wrapper_desc_spin_unlock(&desc->lock);
 281                 local_irq_restore(flags);
 282         }
 283         return 0;
 284 #undef irq_to_desc
 285 }
 286 #else
 287 static inline
 288 int lttng_list_interrupts(struct lttng_session *session)
 289 {
 290         return 0;
 291 }
 292 #endif
 293
 294 static
 295 void lttng_statedump_process_ns(struct lttng_session *session,
 296                 struct task_struct *p,
 297                 enum lttng_thread_type type,
 298                 enum lttng_execution_mode mode,
 299                 enum lttng_execution_submode submode,
 300                 enum lttng_process_status status)
 301 {
 302         struct nsproxy *proxy;
 303         struct pid_namespace *pid_ns;
 304
 305         rcu_read_lock();
 306         proxy = task_nsproxy(p);
 307         if (proxy) {
 308                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 309                 do {
 310                         trace_lttng_statedump_process_state(session,
 311                                 p, type, mode, submode, status, pid_ns);
 312                         pid_ns = pid_ns->parent;
 313                 } while (pid_ns);
 314         } else {
 315                 trace_lttng_statedump_process_state(session,
 316                         p, type, mode, submode, status, NULL);
 317         }
 318         rcu_read_unlock();
 319 }
 320
 321 static
 322 int lttng_enumerate_process_states(struct lttng_session *session)
 323 {
 324         struct task_struct *g, *p;
 325
 326         rcu_read_lock();
 327         for_each_process(g) {
 328                 p = g;
 329                 do {
 330                         enum lttng_execution_mode mode =
 331                                 LTTNG_MODE_UNKNOWN;
 332                         enum lttng_execution_submode submode =
 333                                 LTTNG_UNKNOWN;
 334                         enum lttng_process_status status;
 335                         enum lttng_thread_type type;
 336
 337                         task_lock(p);
 338                         if (p->exit_state == EXIT_ZOMBIE)
 339                                 status = LTTNG_ZOMBIE;
 340                         else if (p->exit_state == EXIT_DEAD)
 341                                 status = LTTNG_DEAD;
 342                         else if (p->state == TASK_RUNNING) {
 343                                 /* Is this a forked child that has not run yet? */
 344                                 if (list_empty(&p->rt.run_list))
 345                                         status = LTTNG_WAIT_FORK;
 346                                 else
 347                                         /*
 348                                          * All tasks are considered as wait_cpu;
 349                                          * the viewer will sort out if the task
 350                                          * was really running at this time.
 351                                          */
 352                                         status = LTTNG_WAIT_CPU;
 353                         } else if (p->state &
 354                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 355                                 /* Task is waiting for something to complete */
 356                                 status = LTTNG_WAIT;
 357                         } else
 358                                 status = LTTNG_UNNAMED;
 359                         submode = LTTNG_NONE;
 360
 361                         /*
 362                          * Verification of t->mm is to filter out kernel
 363                          * threads; Viewer will further filter out if a
 364                          * user-space thread was in syscall mode or not.
 365                          */
 366                         if (p->mm)
 367                                 type = LTTNG_USER_THREAD;
 368                         else
 369                                 type = LTTNG_KERNEL_THREAD;
 370                         lttng_statedump_process_ns(session,
 371                                 p, type, mode, submode, status);
 372                         task_unlock(p);
 373                 } while_each_thread(g, p);
 374         }
 375         rcu_read_unlock();
 376
 377         return 0;
 378 }
 379
 380 static
 381 void lttng_statedump_work_func(struct work_struct *work)
 382 {
 383         if (atomic_dec_and_test(&kernel_threads_to_run))
 384                 /* If we are the last thread, wake up do_lttng_statedump */
 385                 wake_up(&statedump_wq);
 386 }
 387
 388 static
 389 int do_lttng_statedump(struct lttng_session *session)
 390 {
 391         int cpu, ret;
 392
 393         trace_lttng_statedump_start(session);
 394         ret = lttng_enumerate_process_states(session);
 395         if (ret)
 396                 return ret;
 397         ret = lttng_enumerate_file_descriptors(session);
 398         if (ret)
 399                 return ret;
 400         /*
 401          * FIXME
 402          * ret = lttng_enumerate_vm_maps(session);
 403          * if (ret)
 404          *      return ret;
 405          */
 406         ret = lttng_list_interrupts(session);
 407         if (ret)
 408                 return ret;
 409         ret = lttng_enumerate_network_ip_interface(session);
 410         if (ret)
 411                 return ret;
 412
 413         /* TODO lttng_dump_idt_table(session); */
 414         /* TODO lttng_dump_softirq_vec(session); */
 415         /* TODO lttng_list_modules(session); */
 416         /* TODO lttng_dump_swap_files(session); */
 417
 418         /*
 419          * Fire off a work queue on each CPU. Their sole purpose in life
 420          * is to guarantee that each CPU has been in a state where is was in
 421          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 422          */
 423         get_online_cpus();
 424         atomic_set(&kernel_threads_to_run, num_online_cpus());
 425         for_each_online_cpu(cpu) {
 426                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 427                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 428         }
 429         /* Wait for all threads to run */
 430         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 431         put_online_cpus();
 432         /* Our work is done */
 433         trace_lttng_statedump_end(session);
 434         return 0;
 435 }
 436
 437 /*
 438  * Called with session mutex held.
 439  */
 440 int lttng_statedump_start(struct lttng_session *session)
 441 {
 442         return do_lttng_statedump(session);
 443 }
 444 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 445
 446 static
 447 int __init lttng_statedump_init(void)
 448 {
 449         /*
 450          * Allow module to load even if the fixup cannot be done. This
 451          * will allow seemless transition when the underlying issue fix
 452          * is merged into the Linux kernel, and when tracepoint.c
 453          * "tracepoint_module_notify" is turned into a static function.
 454          */
 455         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 456         return 0;
 457 }
 458
 459 module_init(lttng_statedump_init);
 460
 461 static
 462 void __exit lttng_statedump_exit(void)
 463 {
 464 }
 465
 466 module_exit(lttng_statedump_exit);
 467
 468 MODULE_LICENSE("GPL and additional rights");
 469 MODULE_AUTHOR("Jean-Hugues Deschenes");
 470 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 471 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 472         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 473         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 474         LTTNG_MODULES_EXTRAVERSION);