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