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
  60 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  61 #include <linux/irq.h>
  62 #endif
  63
  64 /* Define the tracepoints, but do not build the probes */
  65 #define CREATE_TRACE_POINTS
  66 #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
  67 #define TRACE_INCLUDE_FILE lttng-statedump
  68 #define LTTNG_INSTRUMENTATION
  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 #ifdef FD_ISSET /* For old kernels lacking close_on_exec() */
 218 static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
 219 {
 220         return FD_ISSET(fd, fdt->close_on_exec);
 221 }
 222 #else
 223 static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
 224 {
 225         return close_on_exec(fd, fdt);
 226 }
 227 #endif
 228
 229 static
 230 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 231 {
 232         const struct lttng_fd_ctx *ctx = p;
 233         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 234         unsigned int flags = file->f_flags;
 235         struct fdtable *fdt;
 236
 237         /*
 238          * We don't expose kernel internal flags, only userspace-visible
 239          * flags.
 240          */
 241         flags &= ~FMODE_NONOTIFY;
 242         fdt = files_fdtable(ctx->files);
 243         /*
 244          * We need to check here again whether fd is within the fdt
 245          * max_fds range, because we might be seeing a different
 246          * files_fdtable() than iterate_fd(), assuming only RCU is
 247          * protecting the read. In reality, iterate_fd() holds
 248          * file_lock, which should ensure the fdt does not change while
 249          * the lock is taken, but we are not aware whether this is
 250          * guaranteed or not, so play safe.
 251          */
 252         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 253                 flags |= O_CLOEXEC;
 254         if (IS_ERR(s)) {
 255                 struct dentry *dentry = file->f_path.dentry;
 256
 257                 /* Make sure we give at least some info */
 258                 spin_lock(&dentry->d_lock);
 259                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 260                         dentry->d_name.name, flags, file->f_mode);
 261                 spin_unlock(&dentry->d_lock);
 262                 goto end;
 263         }
 264         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 265                 flags, file->f_mode);
 266 end:
 267         return 0;
 268 }
 269
 270 static
 271 void lttng_enumerate_task_fd(struct lttng_session *session,
 272                 struct task_struct *p, char *tmp)
 273 {
 274         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 275         struct files_struct *files;
 276
 277         task_lock(p);
 278         files = p->files;
 279         if (!files)
 280                 goto end;
 281         ctx.files = files;
 282         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 283 end:
 284         task_unlock(p);
 285 }
 286
 287 static
 288 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 289 {
 290         struct task_struct *p;
 291         char *tmp;
 292
 293         tmp = (char *) __get_free_page(GFP_KERNEL);
 294         if (!tmp)
 295                 return -ENOMEM;
 296
 297         /* Enumerate active file descriptors */
 298         rcu_read_lock();
 299         for_each_process(p)
 300                 lttng_enumerate_task_fd(session, p, tmp);
 301         rcu_read_unlock();
 302         free_page((unsigned long) tmp);
 303         return 0;
 304 }
 305
 306 #if 0
 307 /*
 308  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 309  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 310  * iteration, but it is not exported to modules.
 311  */
 312 static
 313 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 314                 struct task_struct *p)
 315 {
 316         struct mm_struct *mm;
 317         struct vm_area_struct *map;
 318         unsigned long ino;
 319
 320         /* get_task_mm does a task_lock... */
 321         mm = get_task_mm(p);
 322         if (!mm)
 323                 return;
 324
 325         map = mm->mmap;
 326         if (map) {
 327                 down_read(&mm->mmap_sem);
 328                 while (map) {
 329                         if (map->vm_file)
 330                                 ino = map->vm_file->f_dentry->d_inode->i_ino;
 331                         else
 332                                 ino = 0;
 333                         trace_lttng_statedump_vm_map(session, p, map, ino);
 334                         map = map->vm_next;
 335                 }
 336                 up_read(&mm->mmap_sem);
 337         }
 338         mmput(mm);
 339 }
 340
 341 static
 342 int lttng_enumerate_vm_maps(struct lttng_session *session)
 343 {
 344         struct task_struct *p;
 345
 346         rcu_read_lock();
 347         for_each_process(p)
 348                 lttng_enumerate_task_vm_maps(session, p);
 349         rcu_read_unlock();
 350         return 0;
 351 }
 352 #endif
 353
 354 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 355
 356 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 357 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 358 #endif
 359
 360 static
 361 int lttng_list_interrupts(struct lttng_session *session)
 362 {
 363         unsigned int irq;
 364         unsigned long flags = 0;
 365         struct irq_desc *desc;
 366
 367 #define irq_to_desc     wrapper_irq_to_desc
 368         /* needs irq_desc */
 369         for_each_irq_desc(irq, desc) {
 370                 struct irqaction *action;
 371                 const char *irq_chip_name =
 372                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 373
 374                 local_irq_save(flags);
 375                 wrapper_desc_spin_lock(&desc->lock);
 376                 for (action = desc->action; action; action = action->next) {
 377                         trace_lttng_statedump_interrupt(session,
 378                                 irq, irq_chip_name, action);
 379                 }
 380                 wrapper_desc_spin_unlock(&desc->lock);
 381                 local_irq_restore(flags);
 382         }
 383         return 0;
 384 #undef irq_to_desc
 385 }
 386 #else
 387 static inline
 388 int lttng_list_interrupts(struct lttng_session *session)
 389 {
 390         return 0;
 391 }
 392 #endif
 393
 394 /*
 395  * Called with task lock held.
 396  */
 397 static
 398 void lttng_statedump_process_ns(struct lttng_session *session,
 399                 struct task_struct *p,
 400                 enum lttng_thread_type type,
 401                 enum lttng_execution_mode mode,
 402                 enum lttng_execution_submode submode,
 403                 enum lttng_process_status status)
 404 {
 405         struct nsproxy *proxy;
 406         struct pid_namespace *pid_ns;
 407
 408         /*
 409          * Back and forth on locking strategy within Linux upstream for nsproxy.
 410          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 411          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 412          * for details.
 413          */
 414 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 415                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0))
 416         proxy = p->nsproxy;
 417 #else
 418         rcu_read_lock();
 419         proxy = task_nsproxy(p);
 420 #endif
 421         if (proxy) {
 422                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 423                 do {
 424                         trace_lttng_statedump_process_state(session,
 425                                 p, type, mode, submode, status, pid_ns);
 426                         pid_ns = pid_ns->parent;
 427                 } while (pid_ns);
 428         } else {
 429                 trace_lttng_statedump_process_state(session,
 430                         p, type, mode, submode, status, NULL);
 431         }
 432 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 433                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0))
 434         /* (nothing) */
 435 #else
 436         rcu_read_unlock();
 437 #endif
 438 }
 439
 440 static
 441 int lttng_enumerate_process_states(struct lttng_session *session)
 442 {
 443         struct task_struct *g, *p;
 444
 445         rcu_read_lock();
 446         for_each_process(g) {
 447                 p = g;
 448                 do {
 449                         enum lttng_execution_mode mode =
 450                                 LTTNG_MODE_UNKNOWN;
 451                         enum lttng_execution_submode submode =
 452                                 LTTNG_UNKNOWN;
 453                         enum lttng_process_status status;
 454                         enum lttng_thread_type type;
 455
 456                         task_lock(p);
 457                         if (p->exit_state == EXIT_ZOMBIE)
 458                                 status = LTTNG_ZOMBIE;
 459                         else if (p->exit_state == EXIT_DEAD)
 460                                 status = LTTNG_DEAD;
 461                         else if (p->state == TASK_RUNNING) {
 462                                 /* Is this a forked child that has not run yet? */
 463                                 if (list_empty(&p->rt.run_list))
 464                                         status = LTTNG_WAIT_FORK;
 465                                 else
 466                                         /*
 467                                          * All tasks are considered as wait_cpu;
 468                                          * the viewer will sort out if the task
 469                                          * was really running at this time.
 470                                          */
 471                                         status = LTTNG_WAIT_CPU;
 472                         } else if (p->state &
 473                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 474                                 /* Task is waiting for something to complete */
 475                                 status = LTTNG_WAIT;
 476                         } else
 477                                 status = LTTNG_UNNAMED;
 478                         submode = LTTNG_NONE;
 479
 480                         /*
 481                          * Verification of t->mm is to filter out kernel
 482                          * threads; Viewer will further filter out if a
 483                          * user-space thread was in syscall mode or not.
 484                          */
 485                         if (p->mm)
 486                                 type = LTTNG_USER_THREAD;
 487                         else
 488                                 type = LTTNG_KERNEL_THREAD;
 489                         lttng_statedump_process_ns(session,
 490                                 p, type, mode, submode, status);
 491                         task_unlock(p);
 492                 } while_each_thread(g, p);
 493         }
 494         rcu_read_unlock();
 495
 496         return 0;
 497 }
 498
 499 static
 500 void lttng_statedump_work_func(struct work_struct *work)
 501 {
 502         if (atomic_dec_and_test(&kernel_threads_to_run))
 503                 /* If we are the last thread, wake up do_lttng_statedump */
 504                 wake_up(&statedump_wq);
 505 }
 506
 507 static
 508 int do_lttng_statedump(struct lttng_session *session)
 509 {
 510         int cpu, ret;
 511
 512         trace_lttng_statedump_start(session);
 513         ret = lttng_enumerate_process_states(session);
 514         if (ret)
 515                 return ret;
 516         ret = lttng_enumerate_file_descriptors(session);
 517         if (ret)
 518                 return ret;
 519         /*
 520          * FIXME
 521          * ret = lttng_enumerate_vm_maps(session);
 522          * if (ret)
 523          *      return ret;
 524          */
 525         ret = lttng_list_interrupts(session);
 526         if (ret)
 527                 return ret;
 528         ret = lttng_enumerate_network_ip_interface(session);
 529         if (ret)
 530                 return ret;
 531         ret = lttng_enumerate_block_devices(session);
 532         switch (ret) {
 533         case -ENOSYS:
 534                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 535                 break;
 536         default:
 537                 return ret;
 538         }
 539
 540         /* TODO lttng_dump_idt_table(session); */
 541         /* TODO lttng_dump_softirq_vec(session); */
 542         /* TODO lttng_list_modules(session); */
 543         /* TODO lttng_dump_swap_files(session); */
 544
 545         /*
 546          * Fire off a work queue on each CPU. Their sole purpose in life
 547          * is to guarantee that each CPU has been in a state where is was in
 548          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 549          */
 550         get_online_cpus();
 551         atomic_set(&kernel_threads_to_run, num_online_cpus());
 552         for_each_online_cpu(cpu) {
 553                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 554                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 555         }
 556         /* Wait for all threads to run */
 557         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 558         put_online_cpus();
 559         /* Our work is done */
 560         trace_lttng_statedump_end(session);
 561         return 0;
 562 }
 563
 564 /*
 565  * Called with session mutex held.
 566  */
 567 int lttng_statedump_start(struct lttng_session *session)
 568 {
 569         return do_lttng_statedump(session);
 570 }
 571 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 572
 573 static
 574 int __init lttng_statedump_init(void)
 575 {
 576         /*
 577          * Allow module to load even if the fixup cannot be done. This
 578          * will allow seemless transition when the underlying issue fix
 579          * is merged into the Linux kernel, and when tracepoint.c
 580          * "tracepoint_module_notify" is turned into a static function.
 581          */
 582         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 583         return 0;
 584 }
 585
 586 module_init(lttng_statedump_init);
 587
 588 static
 589 void __exit lttng_statedump_exit(void)
 590 {
 591 }
 592
 593 module_exit(lttng_statedump_exit);
 594
 595 MODULE_LICENSE("GPL and additional rights");
 596 MODULE_AUTHOR("Jean-Hugues Deschenes");
 597 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 598 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 599         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 600         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 601         LTTNG_MODULES_EXTRAVERSION);