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