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