lttng-statedump-impl.c

   1 /* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
   2  *
   3  * lttng-statedump.c
   4  *
   5  * Linux Trace Toolkit Next Generation Kernel State Dump
   6  *
   7  * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
   8  * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   9  *
  10  * Changes:
  11  *      Eric Clement:                   Add listing of network IP interface
  12  *      2006, 2007 Mathieu Desnoyers    Fix kernel threads
  13  *                                      Various updates
  14  */
  15
  16 #include <linux/init.h>
  17 #include <linux/module.h>
  18 #include <linux/netlink.h>
  19 #include <linux/inet.h>
  20 #include <linux/ip.h>
  21 #include <linux/kthread.h>
  22 #include <linux/proc_fs.h>
  23 #include <linux/file.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/irqnr.h>
  26 #include <linux/cpu.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/sched.h>
  30 #include <linux/mm.h>
  31 #include <linux/swap.h>
  32 #include <linux/wait.h>
  33 #include <linux/mutex.h>
  34 #include <linux/device.h>
  35
  36 #include <lttng-events.h>
  37 #include <lttng-tracer.h>
  38 #include <wrapper/irqdesc.h>
  39 #include <wrapper/spinlock.h>
  40 #include <wrapper/fdtable.h>
  41 #include <wrapper/irq.h>
  42 #include <wrapper/tracepoint.h>
  43 #include <wrapper/genhd.h>
  44 #include <wrapper/file.h>
  45 #include <wrapper/fdtable.h>
  46
  47 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  48 #include <linux/irq.h>
  49 #endif
  50
  51 /* Define the tracepoints, but do not build the probes */
  52 #define CREATE_TRACE_POINTS
  53 #define TRACE_INCLUDE_PATH instrumentation/events/lttng-module
  54 #define TRACE_INCLUDE_FILE lttng-statedump
  55 #define LTTNG_INSTRUMENTATION
  56 #include <instrumentation/events/lttng-module/lttng-statedump.h>
  57
  58 LTTNG_DEFINE_TRACE(lttng_statedump_block_device,
  59         TP_PROTO(struct lttng_session *session,
  60                 dev_t dev, const char *diskname),
  61         TP_ARGS(session, dev, diskname));
  62
  63 LTTNG_DEFINE_TRACE(lttng_statedump_end,
  64         TP_PROTO(struct lttng_session *session),
  65         TP_ARGS(session));
  66
  67 LTTNG_DEFINE_TRACE(lttng_statedump_interrupt,
  68         TP_PROTO(struct lttng_session *session,
  69                 unsigned int irq, const char *chip_name,
  70                 struct irqaction *action),
  71         TP_ARGS(session, irq, chip_name, action));
  72
  73 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor,
  74         TP_PROTO(struct lttng_session *session,
  75                 struct files_struct *files,
  76                 int fd, const char *filename,
  77                 unsigned int flags, fmode_t fmode),
  78         TP_ARGS(session, files, fd, filename, flags, fmode));
  79
  80 LTTNG_DEFINE_TRACE(lttng_statedump_start,
  81         TP_PROTO(struct lttng_session *session),
  82         TP_ARGS(session));
  83
  84 LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
  85         TP_PROTO(struct lttng_session *session,
  86                 struct task_struct *p,
  87                 int type, int mode, int submode, int status,
  88                 struct files_struct *files),
  89         TP_ARGS(session, p, type, mode, submode, status, files));
  90
  91 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
  92         TP_PROTO(struct lttng_session *session,
  93                 struct net_device *dev, struct in_ifaddr *ifa),
  94         TP_ARGS(session, dev, ifa));
  95
  96 struct lttng_fd_ctx {
  97         char *page;
  98         struct lttng_session *session;
  99         struct task_struct *p;
 100         struct files_struct *files;
 101 };
 102
 103 /*
 104  * Protected by the trace lock.
 105  */
 106 static struct delayed_work cpu_work[NR_CPUS];
 107 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 108 static atomic_t kernel_threads_to_run;
 109
 110 enum lttng_thread_type {
 111         LTTNG_USER_THREAD = 0,
 112         LTTNG_KERNEL_THREAD = 1,
 113 };
 114
 115 enum lttng_execution_mode {
 116         LTTNG_USER_MODE = 0,
 117         LTTNG_SYSCALL = 1,
 118         LTTNG_TRAP = 2,
 119         LTTNG_IRQ = 3,
 120         LTTNG_SOFTIRQ = 4,
 121         LTTNG_MODE_UNKNOWN = 5,
 122 };
 123
 124 enum lttng_execution_submode {
 125         LTTNG_NONE = 0,
 126         LTTNG_UNKNOWN = 1,
 127 };
 128
 129 enum lttng_process_status {
 130         LTTNG_UNNAMED = 0,
 131         LTTNG_WAIT_FORK = 1,
 132         LTTNG_WAIT_CPU = 2,
 133         LTTNG_EXIT = 3,
 134         LTTNG_ZOMBIE = 4,
 135         LTTNG_WAIT = 5,
 136         LTTNG_RUN = 6,
 137         LTTNG_DEAD = 7,
 138 };
 139
 140 static
 141 int lttng_enumerate_block_devices(struct lttng_session *session)
 142 {
 143         struct class *ptr_block_class;
 144         struct device_type *ptr_disk_type;
 145         struct class_dev_iter iter;
 146         struct device *dev;
 147
 148         ptr_block_class = wrapper_get_block_class();
 149         if (!ptr_block_class)
 150                 return -ENOSYS;
 151         ptr_disk_type = wrapper_get_disk_type();
 152         if (!ptr_disk_type) {
 153                 return -ENOSYS;
 154         }
 155         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 156         while ((dev = class_dev_iter_next(&iter))) {
 157                 struct disk_part_iter piter;
 158                 struct gendisk *disk = dev_to_disk(dev);
 159                 struct hd_struct *part;
 160
 161                 /*
 162                  * Don't show empty devices or things that have been
 163                  * suppressed
 164                  */
 165                 if (get_capacity(disk) == 0 ||
 166                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 167                         continue;
 168
 169                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 170                 while ((part = disk_part_iter_next(&piter))) {
 171                         struct block_device bdev;
 172                         char name_buf[BDEVNAME_SIZE];
 173                         const char *p;
 174
 175                         /*
 176                          * Create a partial 'struct blockdevice' to use
 177                          * 'bdevname()' which is a simple wrapper over
 178                          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 179                          */
 180                         bdev.bd_disk = disk;
 181                         bdev.bd_part = part;
 182
 183                         p = bdevname(&bdev, name_buf);
 184                         if (!p) {
 185                                 disk_part_iter_exit(&piter);
 186                                 class_dev_iter_exit(&iter);
 187                                 return -ENOSYS;
 188                         }
 189                         trace_lttng_statedump_block_device(session,
 190                                         part_devt(part), name_buf);
 191                 }
 192                 disk_part_iter_exit(&piter);
 193         }
 194         class_dev_iter_exit(&iter);
 195         return 0;
 196 }
 197
 198 #ifdef CONFIG_INET
 199
 200 static
 201 void lttng_enumerate_device(struct lttng_session *session,
 202                 struct net_device *dev)
 203 {
 204         struct in_device *in_dev;
 205         struct in_ifaddr *ifa;
 206
 207         if (dev->flags & IFF_UP) {
 208                 in_dev = in_dev_get(dev);
 209                 if (in_dev) {
 210                         for (ifa = in_dev->ifa_list; ifa != NULL;
 211                              ifa = ifa->ifa_next) {
 212                                 trace_lttng_statedump_network_interface(
 213                                         session, dev, ifa);
 214                         }
 215                         in_dev_put(in_dev);
 216                 }
 217         } else {
 218                 trace_lttng_statedump_network_interface(
 219                         session, dev, NULL);
 220         }
 221 }
 222
 223 static
 224 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 225 {
 226         struct net_device *dev;
 227
 228         read_lock(&dev_base_lock);
 229         for_each_netdev(&init_net, dev)
 230                 lttng_enumerate_device(session, dev);
 231         read_unlock(&dev_base_lock);
 232
 233         return 0;
 234 }
 235 #else /* CONFIG_INET */
 236 static inline
 237 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 238 {
 239         return 0;
 240 }
 241 #endif /* CONFIG_INET */
 242
 243 static
 244 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 245 {
 246         const struct lttng_fd_ctx *ctx = p;
 247         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 248         unsigned int flags = file->f_flags;
 249         struct fdtable *fdt;
 250
 251         /*
 252          * We don't expose kernel internal flags, only userspace-visible
 253          * flags.
 254          */
 255         flags &= ~FMODE_NONOTIFY;
 256         fdt = files_fdtable(ctx->files);
 257         /*
 258          * We need to check here again whether fd is within the fdt
 259          * max_fds range, because we might be seeing a different
 260          * files_fdtable() than iterate_fd(), assuming only RCU is
 261          * protecting the read. In reality, iterate_fd() holds
 262          * file_lock, which should ensure the fdt does not change while
 263          * the lock is taken, but we are not aware whether this is
 264          * guaranteed or not, so play safe.
 265          */
 266         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 267                 flags |= O_CLOEXEC;
 268         if (IS_ERR(s)) {
 269                 struct dentry *dentry = file->f_path.dentry;
 270
 271                 /* Make sure we give at least some info */
 272                 spin_lock(&dentry->d_lock);
 273                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 274                         dentry->d_name.name, flags, file->f_mode);
 275                 spin_unlock(&dentry->d_lock);
 276                 goto end;
 277         }
 278         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 279                 flags, file->f_mode);
 280 end:
 281         return 0;
 282 }
 283
 284 static
 285 void lttng_enumerate_task_fd(struct lttng_session *session,
 286                 struct task_struct *p, char *tmp)
 287 {
 288         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 289         struct files_struct *files;
 290
 291         task_lock(p);
 292         files = p->files;
 293         if (!files)
 294                 goto end;
 295         ctx.files = files;
 296         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 297 end:
 298         task_unlock(p);
 299 }
 300
 301 static
 302 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 303 {
 304         struct task_struct *p;
 305         char *tmp;
 306
 307         tmp = (char *) __get_free_page(GFP_KERNEL);
 308         if (!tmp)
 309                 return -ENOMEM;
 310
 311         /* Enumerate active file descriptors */
 312         rcu_read_lock();
 313         for_each_process(p)
 314                 lttng_enumerate_task_fd(session, p, tmp);
 315         rcu_read_unlock();
 316         free_page((unsigned long) tmp);
 317         return 0;
 318 }
 319
 320 #if 0
 321 /*
 322  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 323  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 324  * iteration, but it is not exported to modules.
 325  */
 326 static
 327 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 328                 struct task_struct *p)
 329 {
 330         struct mm_struct *mm;
 331         struct vm_area_struct *map;
 332         unsigned long ino;
 333
 334         /* get_task_mm does a task_lock... */
 335         mm = get_task_mm(p);
 336         if (!mm)
 337                 return;
 338
 339         map = mm->mmap;
 340         if (map) {
 341                 down_read(&mm->mmap_sem);
 342                 while (map) {
 343                         if (map->vm_file)
 344                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 345                         else
 346                                 ino = 0;
 347                         trace_lttng_statedump_vm_map(session, p, map, ino);
 348                         map = map->vm_next;
 349                 }
 350                 up_read(&mm->mmap_sem);
 351         }
 352         mmput(mm);
 353 }
 354
 355 static
 356 int lttng_enumerate_vm_maps(struct lttng_session *session)
 357 {
 358         struct task_struct *p;
 359
 360         rcu_read_lock();
 361         for_each_process(p)
 362                 lttng_enumerate_task_vm_maps(session, p);
 363         rcu_read_unlock();
 364         return 0;
 365 }
 366 #endif
 367
 368 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 369
 370 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 371 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 372 #endif
 373
 374 static
 375 int lttng_list_interrupts(struct lttng_session *session)
 376 {
 377         unsigned int irq;
 378         unsigned long flags = 0;
 379         struct irq_desc *desc;
 380
 381 #define irq_to_desc     wrapper_irq_to_desc
 382         /* needs irq_desc */
 383         for_each_irq_desc(irq, desc) {
 384                 struct irqaction *action;
 385                 const char *irq_chip_name =
 386                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 387
 388                 local_irq_save(flags);
 389                 wrapper_desc_spin_lock(&desc->lock);
 390                 for (action = desc->action; action; action = action->next) {
 391                         trace_lttng_statedump_interrupt(session,
 392                                 irq, irq_chip_name, action);
 393                 }
 394                 wrapper_desc_spin_unlock(&desc->lock);
 395                 local_irq_restore(flags);
 396         }
 397         return 0;
 398 #undef irq_to_desc
 399 }
 400 #else
 401 static inline
 402 int lttng_list_interrupts(struct lttng_session *session)
 403 {
 404         return 0;
 405 }
 406 #endif
 407
 408 /*
 409  * Called with task lock held.
 410  */
 411 static
 412 void lttng_statedump_process_ns(struct lttng_session *session,
 413                 struct task_struct *p,
 414                 enum lttng_thread_type type,
 415                 enum lttng_execution_mode mode,
 416                 enum lttng_execution_submode submode,
 417                 enum lttng_process_status status)
 418 {
 419         struct pid_namespace *pid_ns;
 420
 421         pid_ns = task_active_pid_ns(p);
 422         do {
 423                 trace_lttng_statedump_process_state(session,
 424                         p, type, mode, submode, status, pid_ns);
 425                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 426         } while (pid_ns);
 427 }
 428
 429 static
 430 int lttng_enumerate_process_states(struct lttng_session *session)
 431 {
 432         struct task_struct *g, *p;
 433
 434         rcu_read_lock();
 435         for_each_process(g) {
 436                 p = g;
 437                 do {
 438                         enum lttng_execution_mode mode =
 439                                 LTTNG_MODE_UNKNOWN;
 440                         enum lttng_execution_submode submode =
 441                                 LTTNG_UNKNOWN;
 442                         enum lttng_process_status status;
 443                         enum lttng_thread_type type;
 444
 445                         task_lock(p);
 446                         if (p->exit_state == EXIT_ZOMBIE)
 447                                 status = LTTNG_ZOMBIE;
 448                         else if (p->exit_state == EXIT_DEAD)
 449                                 status = LTTNG_DEAD;
 450                         else if (p->state == TASK_RUNNING) {
 451                                 /* Is this a forked child that has not run yet? */
 452                                 if (list_empty(&p->rt.run_list))
 453                                         status = LTTNG_WAIT_FORK;
 454                                 else
 455                                         /*
 456                                          * All tasks are considered as wait_cpu;
 457                                          * the viewer will sort out if the task
 458                                          * was really running at this time.
 459                                          */
 460                                         status = LTTNG_WAIT_CPU;
 461                         } else if (p->state &
 462                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 463                                 /* Task is waiting for something to complete */
 464                                 status = LTTNG_WAIT;
 465                         } else
 466                                 status = LTTNG_UNNAMED;
 467                         submode = LTTNG_NONE;
 468
 469                         /*
 470                          * Verification of t->mm is to filter out kernel
 471                          * threads; Viewer will further filter out if a
 472                          * user-space thread was in syscall mode or not.
 473                          */
 474                         if (p->mm)
 475                                 type = LTTNG_USER_THREAD;
 476                         else
 477                                 type = LTTNG_KERNEL_THREAD;
 478                         lttng_statedump_process_ns(session,
 479                                 p, type, mode, submode, status);
 480                         task_unlock(p);
 481                 } while_each_thread(g, p);
 482         }
 483         rcu_read_unlock();
 484
 485         return 0;
 486 }
 487
 488 static
 489 void lttng_statedump_work_func(struct work_struct *work)
 490 {
 491         if (atomic_dec_and_test(&kernel_threads_to_run))
 492                 /* If we are the last thread, wake up do_lttng_statedump */
 493                 wake_up(&statedump_wq);
 494 }
 495
 496 static
 497 int do_lttng_statedump(struct lttng_session *session)
 498 {
 499         int cpu, ret;
 500
 501         trace_lttng_statedump_start(session);
 502         ret = lttng_enumerate_process_states(session);
 503         if (ret)
 504                 return ret;
 505         ret = lttng_enumerate_file_descriptors(session);
 506         if (ret)
 507                 return ret;
 508         /*
 509          * FIXME
 510          * ret = lttng_enumerate_vm_maps(session);
 511          * if (ret)
 512          *      return ret;
 513          */
 514         ret = lttng_list_interrupts(session);
 515         if (ret)
 516                 return ret;
 517         ret = lttng_enumerate_network_ip_interface(session);
 518         if (ret)
 519                 return ret;
 520         ret = lttng_enumerate_block_devices(session);
 521         switch (ret) {
 522         case 0:
 523                 break;
 524         case -ENOSYS:
 525                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 526                 break;
 527         default:
 528                 return ret;
 529         }
 530
 531         /* TODO lttng_dump_idt_table(session); */
 532         /* TODO lttng_dump_softirq_vec(session); */
 533         /* TODO lttng_list_modules(session); */
 534         /* TODO lttng_dump_swap_files(session); */
 535
 536         /*
 537          * Fire off a work queue on each CPU. Their sole purpose in life
 538          * is to guarantee that each CPU has been in a state where is was in
 539          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 540          */
 541         get_online_cpus();
 542         atomic_set(&kernel_threads_to_run, num_online_cpus());
 543         for_each_online_cpu(cpu) {
 544                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 545                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 546         }
 547         /* Wait for all threads to run */
 548         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 549         put_online_cpus();
 550         /* Our work is done */
 551         trace_lttng_statedump_end(session);
 552         return 0;
 553 }
 554
 555 /*
 556  * Called with session mutex held.
 557  */
 558 int lttng_statedump_start(struct lttng_session *session)
 559 {
 560         return do_lttng_statedump(session);
 561 }
 562 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 563
 564 static
 565 int __init lttng_statedump_init(void)
 566 {
 567         /*
 568          * Allow module to load even if the fixup cannot be done. This
 569          * will allow seemless transition when the underlying issue fix
 570          * is merged into the Linux kernel, and when tracepoint.c
 571          * "tracepoint_module_notify" is turned into a static function.
 572          */
 573         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 574         return 0;
 575 }
 576
 577 module_init(lttng_statedump_init);
 578
 579 static
 580 void __exit lttng_statedump_exit(void)
 581 {
 582 }
 583
 584 module_exit(lttng_statedump_exit);
 585
 586 MODULE_LICENSE("GPL and additional rights");
 587 MODULE_AUTHOR("Jean-Hugues Deschenes");
 588 MODULE_DESCRIPTION("LTTng statedump provider");
 589 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 590         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 591         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 592         LTTNG_MODULES_EXTRAVERSION);