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