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