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
  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 #ifdef FD_ISSET /* For old kernels lacking close_on_exec() */
 219 static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
 220 {
 221         return FD_ISSET(fd, fdt->close_on_exec);
 222 }
 223 #else
 224 static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
 225 {
 226         return close_on_exec(fd, fdt);
 227 }
 228 #endif
 229
 230 static
 231 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 232 {
 233         const struct lttng_fd_ctx *ctx = p;
 234         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 235         unsigned int flags = file->f_flags;
 236         struct fdtable *fdt;
 237
 238         /*
 239          * We don't expose kernel internal flags, only userspace-visible
 240          * flags.
 241          */
 242         flags &= ~FMODE_NONOTIFY;
 243         fdt = files_fdtable(ctx->files);
 244         /*
 245          * We need to check here again whether fd is within the fdt
 246          * max_fds range, because we might be seeing a different
 247          * files_fdtable() than iterate_fd(), assuming only RCU is
 248          * protecting the read. In reality, iterate_fd() holds
 249          * file_lock, which should ensure the fdt does not change while
 250          * the lock is taken, but we are not aware whether this is
 251          * guaranteed or not, so play safe.
 252          */
 253         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 254                 flags |= O_CLOEXEC;
 255         if (IS_ERR(s)) {
 256                 struct dentry *dentry = file->f_path.dentry;
 257
 258                 /* Make sure we give at least some info */
 259                 spin_lock(&dentry->d_lock);
 260                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 261                         dentry->d_name.name, flags, file->f_mode);
 262                 spin_unlock(&dentry->d_lock);
 263                 goto end;
 264         }
 265         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 266                 flags, file->f_mode);
 267 end:
 268         return 0;
 269 }
 270
 271 static
 272 void lttng_enumerate_task_fd(struct lttng_session *session,
 273                 struct task_struct *p, char *tmp)
 274 {
 275         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 276         struct files_struct *files;
 277
 278         task_lock(p);
 279         files = p->files;
 280         if (!files)
 281                 goto end;
 282         ctx.files = files;
 283         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 284 end:
 285         task_unlock(p);
 286 }
 287
 288 static
 289 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 290 {
 291         struct task_struct *p;
 292         char *tmp;
 293
 294         tmp = (char *) __get_free_page(GFP_KERNEL);
 295         if (!tmp)
 296                 return -ENOMEM;
 297
 298         /* Enumerate active file descriptors */
 299         rcu_read_lock();
 300         for_each_process(p)
 301                 lttng_enumerate_task_fd(session, p, tmp);
 302         rcu_read_unlock();
 303         free_page((unsigned long) tmp);
 304         return 0;
 305 }
 306
 307 #if 0
 308 /*
 309  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 310  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 311  * iteration, but it is not exported to modules.
 312  */
 313 static
 314 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 315                 struct task_struct *p)
 316 {
 317         struct mm_struct *mm;
 318         struct vm_area_struct *map;
 319         unsigned long ino;
 320
 321         /* get_task_mm does a task_lock... */
 322         mm = get_task_mm(p);
 323         if (!mm)
 324                 return;
 325
 326         map = mm->mmap;
 327         if (map) {
 328                 down_read(&mm->mmap_sem);
 329                 while (map) {
 330                         if (map->vm_file)
 331                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 332                         else
 333                                 ino = 0;
 334                         trace_lttng_statedump_vm_map(session, p, map, ino);
 335                         map = map->vm_next;
 336                 }
 337                 up_read(&mm->mmap_sem);
 338         }
 339         mmput(mm);
 340 }
 341
 342 static
 343 int lttng_enumerate_vm_maps(struct lttng_session *session)
 344 {
 345         struct task_struct *p;
 346
 347         rcu_read_lock();
 348         for_each_process(p)
 349                 lttng_enumerate_task_vm_maps(session, p);
 350         rcu_read_unlock();
 351         return 0;
 352 }
 353 #endif
 354
 355 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 356
 357 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 358 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 359 #endif
 360
 361 static
 362 int lttng_list_interrupts(struct lttng_session *session)
 363 {
 364         unsigned int irq;
 365         unsigned long flags = 0;
 366         struct irq_desc *desc;
 367
 368 #define irq_to_desc     wrapper_irq_to_desc
 369         /* needs irq_desc */
 370         for_each_irq_desc(irq, desc) {
 371                 struct irqaction *action;
 372                 const char *irq_chip_name =
 373                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 374
 375                 local_irq_save(flags);
 376                 wrapper_desc_spin_lock(&desc->lock);
 377                 for (action = desc->action; action; action = action->next) {
 378                         trace_lttng_statedump_interrupt(session,
 379                                 irq, irq_chip_name, action);
 380                 }
 381                 wrapper_desc_spin_unlock(&desc->lock);
 382                 local_irq_restore(flags);
 383         }
 384         return 0;
 385 #undef irq_to_desc
 386 }
 387 #else
 388 static inline
 389 int lttng_list_interrupts(struct lttng_session *session)
 390 {
 391         return 0;
 392 }
 393 #endif
 394
 395 /*
 396  * Called with task lock held.
 397  */
 398 static
 399 void lttng_statedump_process_ns(struct lttng_session *session,
 400                 struct task_struct *p,
 401                 enum lttng_thread_type type,
 402                 enum lttng_execution_mode mode,
 403                 enum lttng_execution_submode submode,
 404                 enum lttng_process_status status)
 405 {
 406         struct nsproxy *proxy;
 407         struct pid_namespace *pid_ns;
 408
 409         /*
 410          * Back and forth on locking strategy within Linux upstream for nsproxy.
 411          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 412          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 413          * for details.
 414          */
 415 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 416                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 417                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,0,11, 3,17,0,0))
 418         proxy = p->nsproxy;
 419 #else
 420         rcu_read_lock();
 421         proxy = task_nsproxy(p);
 422 #endif
 423         if (proxy) {
 424                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 425                 do {
 426                         trace_lttng_statedump_process_state(session,
 427                                 p, type, mode, submode, status, pid_ns);
 428                         pid_ns = pid_ns->parent;
 429                 } while (pid_ns);
 430         } else {
 431                 trace_lttng_statedump_process_state(session,
 432                         p, type, mode, submode, status, NULL);
 433         }
 434 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
 435                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 436                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,0,11, 3,17,0,0))
 437         /* (nothing) */
 438 #else
 439         rcu_read_unlock();
 440 #endif
 441 }
 442
 443 static
 444 int lttng_enumerate_process_states(struct lttng_session *session)
 445 {
 446         struct task_struct *g, *p;
 447
 448         rcu_read_lock();
 449         for_each_process(g) {
 450                 p = g;
 451                 do {
 452                         enum lttng_execution_mode mode =
 453                                 LTTNG_MODE_UNKNOWN;
 454                         enum lttng_execution_submode submode =
 455                                 LTTNG_UNKNOWN;
 456                         enum lttng_process_status status;
 457                         enum lttng_thread_type type;
 458
 459                         task_lock(p);
 460                         if (p->exit_state == EXIT_ZOMBIE)
 461                                 status = LTTNG_ZOMBIE;
 462                         else if (p->exit_state == EXIT_DEAD)
 463                                 status = LTTNG_DEAD;
 464                         else if (p->state == TASK_RUNNING) {
 465                                 /* Is this a forked child that has not run yet? */
 466                                 if (list_empty(&p->rt.run_list))
 467                                         status = LTTNG_WAIT_FORK;
 468                                 else
 469                                         /*
 470                                          * All tasks are considered as wait_cpu;
 471                                          * the viewer will sort out if the task
 472                                          * was really running at this time.
 473                                          */
 474                                         status = LTTNG_WAIT_CPU;
 475                         } else if (p->state &
 476                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 477                                 /* Task is waiting for something to complete */
 478                                 status = LTTNG_WAIT;
 479                         } else
 480                                 status = LTTNG_UNNAMED;
 481                         submode = LTTNG_NONE;
 482
 483                         /*
 484                          * Verification of t->mm is to filter out kernel
 485                          * threads; Viewer will further filter out if a
 486                          * user-space thread was in syscall mode or not.
 487                          */
 488                         if (p->mm)
 489                                 type = LTTNG_USER_THREAD;
 490                         else
 491                                 type = LTTNG_KERNEL_THREAD;
 492                         lttng_statedump_process_ns(session,
 493                                 p, type, mode, submode, status);
 494                         task_unlock(p);
 495                 } while_each_thread(g, p);
 496         }
 497         rcu_read_unlock();
 498
 499         return 0;
 500 }
 501
 502 static
 503 void lttng_statedump_work_func(struct work_struct *work)
 504 {
 505         if (atomic_dec_and_test(&kernel_threads_to_run))
 506                 /* If we are the last thread, wake up do_lttng_statedump */
 507                 wake_up(&statedump_wq);
 508 }
 509
 510 static
 511 int do_lttng_statedump(struct lttng_session *session)
 512 {
 513         int cpu, ret;
 514
 515         trace_lttng_statedump_start(session);
 516         ret = lttng_enumerate_process_states(session);
 517         if (ret)
 518                 return ret;
 519         ret = lttng_enumerate_file_descriptors(session);
 520         if (ret)
 521                 return ret;
 522         /*
 523          * FIXME
 524          * ret = lttng_enumerate_vm_maps(session);
 525          * if (ret)
 526          *      return ret;
 527          */
 528         ret = lttng_list_interrupts(session);
 529         if (ret)
 530                 return ret;
 531         ret = lttng_enumerate_network_ip_interface(session);
 532         if (ret)
 533                 return ret;
 534         ret = lttng_enumerate_block_devices(session);
 535         switch (ret) {
 536         case -ENOSYS:
 537                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 538                 break;
 539         default:
 540                 return ret;
 541         }
 542
 543         /* TODO lttng_dump_idt_table(session); */
 544         /* TODO lttng_dump_softirq_vec(session); */
 545         /* TODO lttng_list_modules(session); */
 546         /* TODO lttng_dump_swap_files(session); */
 547
 548         /*
 549          * Fire off a work queue on each CPU. Their sole purpose in life
 550          * is to guarantee that each CPU has been in a state where is was in
 551          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 552          */
 553         get_online_cpus();
 554         atomic_set(&kernel_threads_to_run, num_online_cpus());
 555         for_each_online_cpu(cpu) {
 556                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 557                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 558         }
 559         /* Wait for all threads to run */
 560         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 561         put_online_cpus();
 562         /* Our work is done */
 563         trace_lttng_statedump_end(session);
 564         return 0;
 565 }
 566
 567 /*
 568  * Called with session mutex held.
 569  */
 570 int lttng_statedump_start(struct lttng_session *session)
 571 {
 572         return do_lttng_statedump(session);
 573 }
 574 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 575
 576 static
 577 int __init lttng_statedump_init(void)
 578 {
 579         /*
 580          * Allow module to load even if the fixup cannot be done. This
 581          * will allow seemless transition when the underlying issue fix
 582          * is merged into the Linux kernel, and when tracepoint.c
 583          * "tracepoint_module_notify" is turned into a static function.
 584          */
 585         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 586         return 0;
 587 }
 588
 589 module_init(lttng_statedump_init);
 590
 591 static
 592 void __exit lttng_statedump_exit(void)
 593 {
 594 }
 595
 596 module_exit(lttng_statedump_exit);
 597
 598 MODULE_LICENSE("GPL and additional rights");
 599 MODULE_AUTHOR("Jean-Hugues Deschenes");
 600 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 601 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 602         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 603         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 604         LTTNG_MODULES_EXTRAVERSION);