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