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/nsproxy.h>
  56 #include <wrapper/irq.h>
  57 #include <wrapper/tracepoint.h>
  58 #include <wrapper/genhd.h>
  59 #include <wrapper/file.h>
  60 #include <wrapper/time.h>
  61
  62 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  63 #include <linux/irq.h>
  64 #endif
  65
  66 /* Define the tracepoints, but do not build the probes */
  67 #define CREATE_TRACE_POINTS
  68 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
  69 #define TRACE_INCLUDE_FILE lttng-statedump
  70 #define LTTNG_INSTRUMENTATION
  71 #include <instrumentation/events/lttng-module/lttng-statedump.h>
  72
  73 DEFINE_TRACE(lttng_statedump_block_device);
  74 DEFINE_TRACE(lttng_statedump_end);
  75 DEFINE_TRACE(lttng_statedump_interrupt);
  76 DEFINE_TRACE(lttng_statedump_file_descriptor);
  77 DEFINE_TRACE(lttng_statedump_start);
  78 DEFINE_TRACE(lttng_statedump_process_state);
  79 DEFINE_TRACE(lttng_statedump_network_interface);
  80
  81 struct lttng_fd_ctx {
  82         char *page;
  83         struct lttng_session *session;
  84         struct task_struct *p;
  85         struct files_struct *files;
  86 };
  87
  88 /*
  89  * Protected by the trace lock.
  90  */
  91 static struct delayed_work cpu_work[NR_CPUS];
  92 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  93 static atomic_t kernel_threads_to_run;
  94
  95 enum lttng_thread_type {
  96         LTTNG_USER_THREAD = 0,
  97         LTTNG_KERNEL_THREAD = 1,
  98 };
  99
 100 enum lttng_execution_mode {
 101         LTTNG_USER_MODE = 0,
 102         LTTNG_SYSCALL = 1,
 103         LTTNG_TRAP = 2,
 104         LTTNG_IRQ = 3,
 105         LTTNG_SOFTIRQ = 4,
 106         LTTNG_MODE_UNKNOWN = 5,
 107 };
 108
 109 enum lttng_execution_submode {
 110         LTTNG_NONE = 0,
 111         LTTNG_UNKNOWN = 1,
 112 };
 113
 114 enum lttng_process_status {
 115         LTTNG_UNNAMED = 0,
 116         LTTNG_WAIT_FORK = 1,
 117         LTTNG_WAIT_CPU = 2,
 118         LTTNG_EXIT = 3,
 119         LTTNG_ZOMBIE = 4,
 120         LTTNG_WAIT = 5,
 121         LTTNG_RUN = 6,
 122         LTTNG_DEAD = 7,
 123 };
 124
 125 static
 126 int lttng_enumerate_block_devices(struct lttng_session *session)
 127 {
 128         struct class *ptr_block_class;
 129         struct device_type *ptr_disk_type;
 130         struct class_dev_iter iter;
 131         struct device *dev;
 132
 133         ptr_block_class = wrapper_get_block_class();
 134         if (!ptr_block_class)
 135                 return -ENOSYS;
 136         ptr_disk_type = wrapper_get_disk_type();
 137         if (!ptr_disk_type) {
 138                 return -ENOSYS;
 139         }
 140         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 141         while ((dev = class_dev_iter_next(&iter))) {
 142                 struct disk_part_iter piter;
 143                 struct gendisk *disk = dev_to_disk(dev);
 144                 struct hd_struct *part;
 145
 146                 /*
 147                  * Don't show empty devices or things that have been
 148                  * suppressed
 149                  */
 150                 if (get_capacity(disk) == 0 ||
 151                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 152                         continue;
 153
 154                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 155                 while ((part = disk_part_iter_next(&piter))) {
 156                         char name_buf[BDEVNAME_SIZE];
 157                         char *p;
 158
 159                         p = wrapper_disk_name(disk, part->partno, name_buf);
 160                         if (!p) {
 161                                 disk_part_iter_exit(&piter);
 162                                 class_dev_iter_exit(&iter);
 163                                 return -ENOSYS;
 164                         }
 165                         trace_lttng_statedump_block_device(session,
 166                                         part_devt(part), name_buf);
 167                 }
 168                 disk_part_iter_exit(&piter);
 169         }
 170         class_dev_iter_exit(&iter);
 171         return 0;
 172 }
 173
 174 #ifdef CONFIG_INET
 175
 176 static
 177 void lttng_enumerate_device(struct lttng_session *session,
 178                 struct net_device *dev)
 179 {
 180         struct in_device *in_dev;
 181         struct in_ifaddr *ifa;
 182
 183         if (dev->flags & IFF_UP) {
 184                 in_dev = in_dev_get(dev);
 185                 if (in_dev) {
 186                         for (ifa = in_dev->ifa_list; ifa != NULL;
 187                              ifa = ifa->ifa_next) {
 188                                 trace_lttng_statedump_network_interface(
 189                                         session, dev, ifa);
 190                         }
 191                         in_dev_put(in_dev);
 192                 }
 193         } else {
 194                 trace_lttng_statedump_network_interface(
 195                         session, dev, NULL);
 196         }
 197 }
 198
 199 static
 200 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 201 {
 202         struct net_device *dev;
 203
 204         read_lock(&dev_base_lock);
 205         for_each_netdev(&init_net, dev)
 206                 lttng_enumerate_device(session, dev);
 207         read_unlock(&dev_base_lock);
 208
 209         return 0;
 210 }
 211 #else /* CONFIG_INET */
 212 static inline
 213 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 214 {
 215         return 0;
 216 }
 217 #endif /* CONFIG_INET */
 218
 219 static
 220 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 221 {
 222         const struct lttng_fd_ctx *ctx = p;
 223         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 224         unsigned int flags = file->f_flags;
 225         struct fdtable *fdt;
 226
 227         /*
 228          * We don't expose kernel internal flags, only userspace-visible
 229          * flags.
 230          */
 231         flags &= ~FMODE_NONOTIFY;
 232         fdt = files_fdtable(ctx->files);
 233         /*
 234          * We need to check here again whether fd is within the fdt
 235          * max_fds range, because we might be seeing a different
 236          * files_fdtable() than iterate_fd(), assuming only RCU is
 237          * protecting the read. In reality, iterate_fd() holds
 238          * file_lock, which should ensure the fdt does not change while
 239          * the lock is taken, but we are not aware whether this is
 240          * guaranteed or not, so play safe.
 241          */
 242         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 243                 flags |= O_CLOEXEC;
 244         if (IS_ERR(s)) {
 245                 struct dentry *dentry = file->f_path.dentry;
 246
 247                 /* Make sure we give at least some info */
 248                 spin_lock(&dentry->d_lock);
 249                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 250                         dentry->d_name.name, flags, file->f_mode);
 251                 spin_unlock(&dentry->d_lock);
 252                 goto end;
 253         }
 254         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 255                 flags, file->f_mode);
 256 end:
 257         return 0;
 258 }
 259
 260 static
 261 void lttng_enumerate_task_fd(struct lttng_session *session,
 262                 struct task_struct *p, char *tmp)
 263 {
 264         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 265         struct files_struct *files;
 266
 267         task_lock(p);
 268         files = p->files;
 269         if (!files)
 270                 goto end;
 271         ctx.files = files;
 272         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 273 end:
 274         task_unlock(p);
 275 }
 276
 277 static
 278 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 279 {
 280         struct task_struct *p;
 281         char *tmp;
 282
 283         tmp = (char *) __get_free_page(GFP_KERNEL);
 284         if (!tmp)
 285                 return -ENOMEM;
 286
 287         /* Enumerate active file descriptors */
 288         rcu_read_lock();
 289         for_each_process(p)
 290                 lttng_enumerate_task_fd(session, p, tmp);
 291         rcu_read_unlock();
 292         free_page((unsigned long) tmp);
 293         return 0;
 294 }
 295
 296 #if 0
 297 /*
 298  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 299  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 300  * iteration, but it is not exported to modules.
 301  */
 302 static
 303 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 304                 struct task_struct *p)
 305 {
 306         struct mm_struct *mm;
 307         struct vm_area_struct *map;
 308         unsigned long ino;
 309
 310         /* get_task_mm does a task_lock... */
 311         mm = get_task_mm(p);
 312         if (!mm)
 313                 return;
 314
 315         map = mm->mmap;
 316         if (map) {
 317                 down_read(&mm->mmap_sem);
 318                 while (map) {
 319                         if (map->vm_file)
 320                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 321                         else
 322                                 ino = 0;
 323                         trace_lttng_statedump_vm_map(session, p, map, ino);
 324                         map = map->vm_next;
 325                 }
 326                 up_read(&mm->mmap_sem);
 327         }
 328         mmput(mm);
 329 }
 330
 331 static
 332 int lttng_enumerate_vm_maps(struct lttng_session *session)
 333 {
 334         struct task_struct *p;
 335
 336         rcu_read_lock();
 337         for_each_process(p)
 338                 lttng_enumerate_task_vm_maps(session, p);
 339         rcu_read_unlock();
 340         return 0;
 341 }
 342 #endif
 343
 344 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 345
 346 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 347 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 348 #endif
 349
 350 static
 351 int lttng_list_interrupts(struct lttng_session *session)
 352 {
 353         unsigned int irq;
 354         unsigned long flags = 0;
 355         struct irq_desc *desc;
 356
 357 #define irq_to_desc     wrapper_irq_to_desc
 358         /* needs irq_desc */
 359         for_each_irq_desc(irq, desc) {
 360                 struct irqaction *action;
 361                 const char *irq_chip_name =
 362                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 363
 364                 local_irq_save(flags);
 365                 wrapper_desc_spin_lock(&desc->lock);
 366                 for (action = desc->action; action; action = action->next) {
 367                         trace_lttng_statedump_interrupt(session,
 368                                 irq, irq_chip_name, action);
 369                 }
 370                 wrapper_desc_spin_unlock(&desc->lock);
 371                 local_irq_restore(flags);
 372         }
 373         return 0;
 374 #undef irq_to_desc
 375 }
 376 #else
 377 static inline
 378 int lttng_list_interrupts(struct lttng_session *session)
 379 {
 380         return 0;
 381 }
 382 #endif
 383
 384 /*
 385  * Called with task lock held.
 386  */
 387 static
 388 void lttng_statedump_process_ns(struct lttng_session *session,
 389                 struct task_struct *p,
 390                 enum lttng_thread_type type,
 391                 enum lttng_execution_mode mode,
 392                 enum lttng_execution_submode submode,
 393                 enum lttng_process_status status)
 394 {
 395         struct nsproxy *proxy;
 396         struct pid_namespace *pid_ns;
 397
 398         /*
 399          * Back and forth on locking strategy within Linux upstream for nsproxy.
 400          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 401          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 402          * for details.
 403          */
 404 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 405                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 406                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 407                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 408         proxy = p->nsproxy;
 409 #else
 410         rcu_read_lock();
 411         proxy = task_nsproxy(p);
 412 #endif
 413         if (proxy) {
 414                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 415                 do {
 416                         trace_lttng_statedump_process_state(session,
 417                                 p, type, mode, submode, status, pid_ns);
 418                         pid_ns = pid_ns->parent;
 419                 } while (pid_ns);
 420         } else {
 421                 trace_lttng_statedump_process_state(session,
 422                         p, type, mode, submode, status, NULL);
 423         }
 424 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 425                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 426                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 427                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 428         /* (nothing) */
 429 #else
 430         rcu_read_unlock();
 431 #endif
 432 }
 433
 434 static
 435 int lttng_enumerate_process_states(struct lttng_session *session)
 436 {
 437         struct task_struct *g, *p;
 438
 439         rcu_read_lock();
 440         for_each_process(g) {
 441                 p = g;
 442                 do {
 443                         enum lttng_execution_mode mode =
 444                                 LTTNG_MODE_UNKNOWN;
 445                         enum lttng_execution_submode submode =
 446                                 LTTNG_UNKNOWN;
 447                         enum lttng_process_status status;
 448                         enum lttng_thread_type type;
 449
 450                         task_lock(p);
 451                         if (p->exit_state == EXIT_ZOMBIE)
 452                                 status = LTTNG_ZOMBIE;
 453                         else if (p->exit_state == EXIT_DEAD)
 454                                 status = LTTNG_DEAD;
 455                         else if (p->state == TASK_RUNNING) {
 456                                 /* Is this a forked child that has not run yet? */
 457                                 if (list_empty(&p->rt.run_list))
 458                                         status = LTTNG_WAIT_FORK;
 459                                 else
 460                                         /*
 461                                          * All tasks are considered as wait_cpu;
 462                                          * the viewer will sort out if the task
 463                                          * was really running at this time.
 464                                          */
 465                                         status = LTTNG_WAIT_CPU;
 466                         } else if (p->state &
 467                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 468                                 /* Task is waiting for something to complete */
 469                                 status = LTTNG_WAIT;
 470                         } else
 471                                 status = LTTNG_UNNAMED;
 472                         submode = LTTNG_NONE;
 473
 474                         /*
 475                          * Verification of t->mm is to filter out kernel
 476                          * threads; Viewer will further filter out if a
 477                          * user-space thread was in syscall mode or not.
 478                          */
 479                         if (p->mm)
 480                                 type = LTTNG_USER_THREAD;
 481                         else
 482                                 type = LTTNG_KERNEL_THREAD;
 483                         lttng_statedump_process_ns(session,
 484                                 p, type, mode, submode, status);
 485                         task_unlock(p);
 486                 } while_each_thread(g, p);
 487         }
 488         rcu_read_unlock();
 489
 490         return 0;
 491 }
 492
 493 static
 494 void lttng_statedump_work_func(struct work_struct *work)
 495 {
 496         if (atomic_dec_and_test(&kernel_threads_to_run))
 497                 /* If we are the last thread, wake up do_lttng_statedump */
 498                 wake_up(&statedump_wq);
 499 }
 500
 501 static
 502 int do_lttng_statedump(struct lttng_session *session)
 503 {
 504         int cpu, ret;
 505
 506         trace_lttng_statedump_start(session);
 507         ret = lttng_enumerate_process_states(session);
 508         if (ret)
 509                 return ret;
 510         ret = lttng_enumerate_file_descriptors(session);
 511         if (ret)
 512                 return ret;
 513         /*
 514          * FIXME
 515          * ret = lttng_enumerate_vm_maps(session);
 516          * if (ret)
 517          *      return ret;
 518          */
 519         ret = lttng_list_interrupts(session);
 520         if (ret)
 521                 return ret;
 522         ret = lttng_enumerate_network_ip_interface(session);
 523         if (ret)
 524                 return ret;
 525         ret = lttng_enumerate_block_devices(session);
 526         switch (ret) {
 527         case 0:
 528                 break;
 529         case -ENOSYS:
 530                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 531                 break;
 532         default:
 533                 return ret;
 534         }
 535
 536         /* TODO lttng_dump_idt_table(session); */
 537         /* TODO lttng_dump_softirq_vec(session); */
 538         /* TODO lttng_list_modules(session); */
 539         /* TODO lttng_dump_swap_files(session); */
 540
 541         /*
 542          * Fire off a work queue on each CPU. Their sole purpose in life
 543          * is to guarantee that each CPU has been in a state where is was in
 544          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 545          */
 546         get_online_cpus();
 547         atomic_set(&kernel_threads_to_run, num_online_cpus());
 548         for_each_online_cpu(cpu) {
 549                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 550                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 551         }
 552         /* Wait for all threads to run */
 553         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 554         put_online_cpus();
 555         /* Our work is done */
 556         trace_lttng_statedump_end(session);
 557         return 0;
 558 }
 559
 560 /*
 561  * Called with session mutex held.
 562  */
 563 int lttng_statedump_start(struct lttng_session *session)
 564 {
 565         return do_lttng_statedump(session);
 566 }
 567 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 568
 569 static
 570 int __init lttng_statedump_init(void)
 571 {
 572         /*
 573          * Allow module to load even if the fixup cannot be done. This
 574          * will allow seemless transition when the underlying issue fix
 575          * is merged into the Linux kernel, and when tracepoint.c
 576          * "tracepoint_module_notify" is turned into a static function.
 577          */
 578         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 579         return 0;
 580 }
 581
 582 module_init(lttng_statedump_init);
 583
 584 static
 585 void __exit lttng_statedump_exit(void)
 586 {
 587 }
 588
 589 module_exit(lttng_statedump_exit);
 590
 591 MODULE_LICENSE("GPL and additional rights");
 592 MODULE_AUTHOR("Jean-Hugues Deschenes");
 593 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 594 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 595         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 596         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 597         LTTNG_MODULES_EXTRAVERSION);