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