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 #include <linux/device.h>
  49
  50 #include <lttng-events.h>
  51 #include <lttng-tracer.h>
  52 #include <wrapper/irqdesc.h>
  53 #include <wrapper/spinlock.h>
  54 #include <wrapper/fdtable.h>
  55 #include <wrapper/irq.h>
  56 #include <wrapper/tracepoint.h>
  57 #include <wrapper/genhd.h>
  58 #include <wrapper/file.h>
  59 #include <wrapper/time.h>
  60
  61 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  62 #include <linux/irq.h>
  63 #endif
  64
  65 /* Define the tracepoints, but do not build the probes */
  66 #define CREATE_TRACE_POINTS
  67 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
  68 #define TRACE_INCLUDE_FILE lttng-statedump
  69 #define LTTNG_INSTRUMENTATION
  70 #include <instrumentation/events/lttng-module/lttng-statedump.h>
  71
  72 DEFINE_TRACE(lttng_statedump_block_device);
  73 DEFINE_TRACE(lttng_statedump_end);
  74 DEFINE_TRACE(lttng_statedump_interrupt);
  75 DEFINE_TRACE(lttng_statedump_file_descriptor);
  76 DEFINE_TRACE(lttng_statedump_start);
  77 DEFINE_TRACE(lttng_statedump_process_state);
  78 DEFINE_TRACE(lttng_statedump_network_interface);
  79
  80 struct lttng_fd_ctx {
  81         char *page;
  82         struct lttng_session *session;
  83         struct task_struct *p;
  84         struct files_struct *files;
  85 };
  86
  87 /*
  88  * Protected by the trace lock.
  89  */
  90 static struct delayed_work cpu_work[NR_CPUS];
  91 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  92 static atomic_t kernel_threads_to_run;
  93
  94 enum lttng_thread_type {
  95         LTTNG_USER_THREAD = 0,
  96         LTTNG_KERNEL_THREAD = 1,
  97 };
  98
  99 enum lttng_execution_mode {
 100         LTTNG_USER_MODE = 0,
 101         LTTNG_SYSCALL = 1,
 102         LTTNG_TRAP = 2,
 103         LTTNG_IRQ = 3,
 104         LTTNG_SOFTIRQ = 4,
 105         LTTNG_MODE_UNKNOWN = 5,
 106 };
 107
 108 enum lttng_execution_submode {
 109         LTTNG_NONE = 0,
 110         LTTNG_UNKNOWN = 1,
 111 };
 112
 113 enum lttng_process_status {
 114         LTTNG_UNNAMED = 0,
 115         LTTNG_WAIT_FORK = 1,
 116         LTTNG_WAIT_CPU = 2,
 117         LTTNG_EXIT = 3,
 118         LTTNG_ZOMBIE = 4,
 119         LTTNG_WAIT = 5,
 120         LTTNG_RUN = 6,
 121         LTTNG_DEAD = 7,
 122 };
 123
 124 static
 125 int lttng_enumerate_block_devices(struct lttng_session *session)
 126 {
 127         struct class *ptr_block_class;
 128         struct device_type *ptr_disk_type;
 129         struct class_dev_iter iter;
 130         struct device *dev;
 131
 132         ptr_block_class = wrapper_get_block_class();
 133         if (!ptr_block_class)
 134                 return -ENOSYS;
 135         ptr_disk_type = wrapper_get_disk_type();
 136         if (!ptr_disk_type) {
 137                 return -ENOSYS;
 138         }
 139         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 140         while ((dev = class_dev_iter_next(&iter))) {
 141                 struct disk_part_iter piter;
 142                 struct gendisk *disk = dev_to_disk(dev);
 143                 struct hd_struct *part;
 144
 145                 /*
 146                  * Don't show empty devices or things that have been
 147                  * suppressed
 148                  */
 149                 if (get_capacity(disk) == 0 ||
 150                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 151                         continue;
 152
 153                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 154                 while ((part = disk_part_iter_next(&piter))) {
 155                         char name_buf[BDEVNAME_SIZE];
 156                         char *p;
 157
 158                         p = wrapper_disk_name(disk, part->partno, name_buf);
 159                         if (!p) {
 160                                 disk_part_iter_exit(&piter);
 161                                 class_dev_iter_exit(&iter);
 162                                 return -ENOSYS;
 163                         }
 164                         trace_lttng_statedump_block_device(session,
 165                                         part_devt(part), name_buf);
 166                 }
 167                 disk_part_iter_exit(&piter);
 168         }
 169         class_dev_iter_exit(&iter);
 170         return 0;
 171 }
 172
 173 #ifdef CONFIG_INET
 174
 175 static
 176 void lttng_enumerate_device(struct lttng_session *session,
 177                 struct net_device *dev)
 178 {
 179         struct in_device *in_dev;
 180         struct in_ifaddr *ifa;
 181
 182         if (dev->flags & IFF_UP) {
 183                 in_dev = in_dev_get(dev);
 184                 if (in_dev) {
 185                         for (ifa = in_dev->ifa_list; ifa != NULL;
 186                              ifa = ifa->ifa_next) {
 187                                 trace_lttng_statedump_network_interface(
 188                                         session, dev, ifa);
 189                         }
 190                         in_dev_put(in_dev);
 191                 }
 192         } else {
 193                 trace_lttng_statedump_network_interface(
 194                         session, dev, NULL);
 195         }
 196 }
 197
 198 static
 199 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 200 {
 201         struct net_device *dev;
 202
 203         read_lock(&dev_base_lock);
 204         for_each_netdev(&init_net, dev)
 205                 lttng_enumerate_device(session, dev);
 206         read_unlock(&dev_base_lock);
 207
 208         return 0;
 209 }
 210 #else /* CONFIG_INET */
 211 static inline
 212 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 213 {
 214         return 0;
 215 }
 216 #endif /* CONFIG_INET */
 217
 218 static
 219 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 220 {
 221         const struct lttng_fd_ctx *ctx = p;
 222         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 223         unsigned int flags = file->f_flags;
 224         struct fdtable *fdt;
 225
 226         /*
 227          * We don't expose kernel internal flags, only userspace-visible
 228          * flags.
 229          */
 230         flags &= ~FMODE_NONOTIFY;
 231         fdt = files_fdtable(ctx->files);
 232         /*
 233          * We need to check here again whether fd is within the fdt
 234          * max_fds range, because we might be seeing a different
 235          * files_fdtable() than iterate_fd(), assuming only RCU is
 236          * protecting the read. In reality, iterate_fd() holds
 237          * file_lock, which should ensure the fdt does not change while
 238          * the lock is taken, but we are not aware whether this is
 239          * guaranteed or not, so play safe.
 240          */
 241         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 242                 flags |= O_CLOEXEC;
 243         if (IS_ERR(s)) {
 244                 struct dentry *dentry = file->f_path.dentry;
 245
 246                 /* Make sure we give at least some info */
 247                 spin_lock(&dentry->d_lock);
 248                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 249                         dentry->d_name.name, flags, file->f_mode);
 250                 spin_unlock(&dentry->d_lock);
 251                 goto end;
 252         }
 253         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 254                 flags, file->f_mode);
 255 end:
 256         return 0;
 257 }
 258
 259 static
 260 void lttng_enumerate_task_fd(struct lttng_session *session,
 261                 struct task_struct *p, char *tmp)
 262 {
 263         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 264         struct files_struct *files;
 265
 266         task_lock(p);
 267         files = p->files;
 268         if (!files)
 269                 goto end;
 270         ctx.files = files;
 271         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 272 end:
 273         task_unlock(p);
 274 }
 275
 276 static
 277 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 278 {
 279         struct task_struct *p;
 280         char *tmp;
 281
 282         tmp = (char *) __get_free_page(GFP_KERNEL);
 283         if (!tmp)
 284                 return -ENOMEM;
 285
 286         /* Enumerate active file descriptors */
 287         rcu_read_lock();
 288         for_each_process(p)
 289                 lttng_enumerate_task_fd(session, p, tmp);
 290         rcu_read_unlock();
 291         free_page((unsigned long) tmp);
 292         return 0;
 293 }
 294
 295 #if 0
 296 /*
 297  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 298  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 299  * iteration, but it is not exported to modules.
 300  */
 301 static
 302 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 303                 struct task_struct *p)
 304 {
 305         struct mm_struct *mm;
 306         struct vm_area_struct *map;
 307         unsigned long ino;
 308
 309         /* get_task_mm does a task_lock... */
 310         mm = get_task_mm(p);
 311         if (!mm)
 312                 return;
 313
 314         map = mm->mmap;
 315         if (map) {
 316                 down_read(&mm->mmap_sem);
 317                 while (map) {
 318                         if (map->vm_file)
 319                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 320                         else
 321                                 ino = 0;
 322                         trace_lttng_statedump_vm_map(session, p, map, ino);
 323                         map = map->vm_next;
 324                 }
 325                 up_read(&mm->mmap_sem);
 326         }
 327         mmput(mm);
 328 }
 329
 330 static
 331 int lttng_enumerate_vm_maps(struct lttng_session *session)
 332 {
 333         struct task_struct *p;
 334
 335         rcu_read_lock();
 336         for_each_process(p)
 337                 lttng_enumerate_task_vm_maps(session, p);
 338         rcu_read_unlock();
 339         return 0;
 340 }
 341 #endif
 342
 343 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 344
 345 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 346 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 347 #endif
 348
 349 static
 350 int lttng_list_interrupts(struct lttng_session *session)
 351 {
 352         unsigned int irq;
 353         unsigned long flags = 0;
 354         struct irq_desc *desc;
 355
 356 #define irq_to_desc     wrapper_irq_to_desc
 357         /* needs irq_desc */
 358         for_each_irq_desc(irq, desc) {
 359                 struct irqaction *action;
 360                 const char *irq_chip_name =
 361                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 362
 363                 local_irq_save(flags);
 364                 wrapper_desc_spin_lock(&desc->lock);
 365                 for (action = desc->action; action; action = action->next) {
 366                         trace_lttng_statedump_interrupt(session,
 367                                 irq, irq_chip_name, action);
 368                 }
 369                 wrapper_desc_spin_unlock(&desc->lock);
 370                 local_irq_restore(flags);
 371         }
 372         return 0;
 373 #undef irq_to_desc
 374 }
 375 #else
 376 static inline
 377 int lttng_list_interrupts(struct lttng_session *session)
 378 {
 379         return 0;
 380 }
 381 #endif
 382
 383 /*
 384  * Called with task lock held.
 385  */
 386 static
 387 void lttng_statedump_process_ns(struct lttng_session *session,
 388                 struct task_struct *p,
 389                 enum lttng_thread_type type,
 390                 enum lttng_execution_mode mode,
 391                 enum lttng_execution_submode submode,
 392                 enum lttng_process_status status)
 393 {
 394         struct pid_namespace *pid_ns;
 395
 396         pid_ns = task_active_pid_ns(p);
 397         do {
 398                 trace_lttng_statedump_process_state(session,
 399                         p, type, mode, submode, status, pid_ns);
 400                 pid_ns = pid_ns->parent;
 401         } while (pid_ns);
 402 }
 403
 404 static
 405 int lttng_enumerate_process_states(struct lttng_session *session)
 406 {
 407         struct task_struct *g, *p;
 408
 409         rcu_read_lock();
 410         for_each_process(g) {
 411                 p = g;
 412                 do {
 413                         enum lttng_execution_mode mode =
 414                                 LTTNG_MODE_UNKNOWN;
 415                         enum lttng_execution_submode submode =
 416                                 LTTNG_UNKNOWN;
 417                         enum lttng_process_status status;
 418                         enum lttng_thread_type type;
 419
 420                         task_lock(p);
 421                         if (p->exit_state == EXIT_ZOMBIE)
 422                                 status = LTTNG_ZOMBIE;
 423                         else if (p->exit_state == EXIT_DEAD)
 424                                 status = LTTNG_DEAD;
 425                         else if (p->state == TASK_RUNNING) {
 426                                 /* Is this a forked child that has not run yet? */
 427                                 if (list_empty(&p->rt.run_list))
 428                                         status = LTTNG_WAIT_FORK;
 429                                 else
 430                                         /*
 431                                          * All tasks are considered as wait_cpu;
 432                                          * the viewer will sort out if the task
 433                                          * was really running at this time.
 434                                          */
 435                                         status = LTTNG_WAIT_CPU;
 436                         } else if (p->state &
 437                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 438                                 /* Task is waiting for something to complete */
 439                                 status = LTTNG_WAIT;
 440                         } else
 441                                 status = LTTNG_UNNAMED;
 442                         submode = LTTNG_NONE;
 443
 444                         /*
 445                          * Verification of t->mm is to filter out kernel
 446                          * threads; Viewer will further filter out if a
 447                          * user-space thread was in syscall mode or not.
 448                          */
 449                         if (p->mm)
 450                                 type = LTTNG_USER_THREAD;
 451                         else
 452                                 type = LTTNG_KERNEL_THREAD;
 453                         lttng_statedump_process_ns(session,
 454                                 p, type, mode, submode, status);
 455                         task_unlock(p);
 456                 } while_each_thread(g, p);
 457         }
 458         rcu_read_unlock();
 459
 460         return 0;
 461 }
 462
 463 static
 464 void lttng_statedump_work_func(struct work_struct *work)
 465 {
 466         if (atomic_dec_and_test(&kernel_threads_to_run))
 467                 /* If we are the last thread, wake up do_lttng_statedump */
 468                 wake_up(&statedump_wq);
 469 }
 470
 471 static
 472 int do_lttng_statedump(struct lttng_session *session)
 473 {
 474         int cpu, ret;
 475
 476         trace_lttng_statedump_start(session);
 477         ret = lttng_enumerate_process_states(session);
 478         if (ret)
 479                 return ret;
 480         ret = lttng_enumerate_file_descriptors(session);
 481         if (ret)
 482                 return ret;
 483         /*
 484          * FIXME
 485          * ret = lttng_enumerate_vm_maps(session);
 486          * if (ret)
 487          *      return ret;
 488          */
 489         ret = lttng_list_interrupts(session);
 490         if (ret)
 491                 return ret;
 492         ret = lttng_enumerate_network_ip_interface(session);
 493         if (ret)
 494                 return ret;
 495         ret = lttng_enumerate_block_devices(session);
 496         switch (ret) {
 497         case 0:
 498                 break;
 499         case -ENOSYS:
 500                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 501                 break;
 502         default:
 503                 return ret;
 504         }
 505
 506         /* TODO lttng_dump_idt_table(session); */
 507         /* TODO lttng_dump_softirq_vec(session); */
 508         /* TODO lttng_list_modules(session); */
 509         /* TODO lttng_dump_swap_files(session); */
 510
 511         /*
 512          * Fire off a work queue on each CPU. Their sole purpose in life
 513          * is to guarantee that each CPU has been in a state where is was in
 514          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 515          */
 516         get_online_cpus();
 517         atomic_set(&kernel_threads_to_run, num_online_cpus());
 518         for_each_online_cpu(cpu) {
 519                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 520                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 521         }
 522         /* Wait for all threads to run */
 523         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 524         put_online_cpus();
 525         /* Our work is done */
 526         trace_lttng_statedump_end(session);
 527         return 0;
 528 }
 529
 530 /*
 531  * Called with session mutex held.
 532  */
 533 int lttng_statedump_start(struct lttng_session *session)
 534 {
 535         return do_lttng_statedump(session);
 536 }
 537 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 538
 539 static
 540 int __init lttng_statedump_init(void)
 541 {
 542         /*
 543          * Allow module to load even if the fixup cannot be done. This
 544          * will allow seemless transition when the underlying issue fix
 545          * is merged into the Linux kernel, and when tracepoint.c
 546          * "tracepoint_module_notify" is turned into a static function.
 547          */
 548         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 549         return 0;
 550 }
 551
 552 module_init(lttng_statedump_init);
 553
 554 static
 555 void __exit lttng_statedump_exit(void)
 556 {
 557 }
 558
 559 module_exit(lttng_statedump_exit);
 560
 561 MODULE_LICENSE("GPL and additional rights");
 562 MODULE_AUTHOR("Jean-Hugues Deschenes");
 563 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 564 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 565         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 566         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 567         LTTNG_MODULES_EXTRAVERSION);