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 task_struct *p, int fd, const char *filename,
  76                 unsigned int flags, fmode_t fmode),
  77         TP_ARGS(session, p, fd, filename, flags, fmode));
  78
  79 LTTNG_DEFINE_TRACE(lttng_statedump_start,
  80         TP_PROTO(struct lttng_session *session),
  81         TP_ARGS(session));
  82
  83 LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
  84         TP_PROTO(struct lttng_session *session,
  85                 struct task_struct *p,
  86                 int type, int mode, int submode, int status,
  87                 struct pid_namespace *pid_ns),
  88         TP_ARGS(session, p, type, mode, submode, status, pid_ns));
  89
  90 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
  91         TP_PROTO(struct lttng_session *session,
  92                 struct net_device *dev, struct in_ifaddr *ifa),
  93         TP_ARGS(session, dev, ifa));
  94
  95 struct lttng_fd_ctx {
  96         char *page;
  97         struct lttng_session *session;
  98         struct task_struct *p;
  99         struct files_struct *files;
 100 };
 101
 102 /*
 103  * Protected by the trace lock.
 104  */
 105 static struct delayed_work cpu_work[NR_CPUS];
 106 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 107 static atomic_t kernel_threads_to_run;
 108
 109 enum lttng_thread_type {
 110         LTTNG_USER_THREAD = 0,
 111         LTTNG_KERNEL_THREAD = 1,
 112 };
 113
 114 enum lttng_execution_mode {
 115         LTTNG_USER_MODE = 0,
 116         LTTNG_SYSCALL = 1,
 117         LTTNG_TRAP = 2,
 118         LTTNG_IRQ = 3,
 119         LTTNG_SOFTIRQ = 4,
 120         LTTNG_MODE_UNKNOWN = 5,
 121 };
 122
 123 enum lttng_execution_submode {
 124         LTTNG_NONE = 0,
 125         LTTNG_UNKNOWN = 1,
 126 };
 127
 128 enum lttng_process_status {
 129         LTTNG_UNNAMED = 0,
 130         LTTNG_WAIT_FORK = 1,
 131         LTTNG_WAIT_CPU = 2,
 132         LTTNG_EXIT = 3,
 133         LTTNG_ZOMBIE = 4,
 134         LTTNG_WAIT = 5,
 135         LTTNG_RUN = 6,
 136         LTTNG_DEAD = 7,
 137 };
 138
 139
 140 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 141
 142 #define LTTNG_PART_STRUCT_TYPE struct block_device
 143
 144 static
 145 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 146 {
 147         const char *p;
 148
 149         p = bdevname(part, name_buf);
 150         if (!p)
 151                 return -ENOSYS;
 152
 153         return 0;
 154 }
 155
 156 static
 157 dev_t lttng_get_part_devt(struct block_device *part)
 158 {
 159         return part->bd_dev;
 160 }
 161
 162 #else
 163
 164 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
 165
 166 static
 167 int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf)
 168 {
 169         const char *p;
 170         struct block_device bdev;
 171
 172         /*
 173          * Create a partial 'struct blockdevice' to use
 174          * 'bdevname()' which is a simple wrapper over
 175          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 176          */
 177         bdev.bd_disk = disk;
 178         bdev.bd_part = part;
 179
 180         p = bdevname(&bdev, name_buf);
 181         if (!p)
 182                 return -ENOSYS;
 183
 184         return 0;
 185 }
 186
 187 static
 188 dev_t lttng_get_part_devt(struct hd_struct *part)
 189 {
 190         return part_devt(part);
 191 }
 192 #endif
 193
 194 static
 195 int lttng_enumerate_block_devices(struct lttng_session *session)
 196 {
 197         struct class *ptr_block_class;
 198         struct device_type *ptr_disk_type;
 199         struct class_dev_iter iter;
 200         struct device *dev;
 201         int ret = 0;
 202
 203         ptr_block_class = wrapper_get_block_class();
 204         if (!ptr_block_class) {
 205                 ret = -ENOSYS;
 206                 goto end;
 207         }
 208         ptr_disk_type = wrapper_get_disk_type();
 209         if (!ptr_disk_type) {
 210                 ret = -ENOSYS;
 211                 goto end;
 212         }
 213         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 214         while ((dev = class_dev_iter_next(&iter))) {
 215                 struct disk_part_iter piter;
 216                 struct gendisk *disk = dev_to_disk(dev);
 217                 LTTNG_PART_STRUCT_TYPE *part;
 218
 219                 /*
 220                  * Don't show empty devices or things that have been
 221                  * suppressed
 222                  */
 223                 if (get_capacity(disk) == 0 ||
 224                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 225                         continue;
 226
 227                 /*
 228                  * The original 'disk_part_iter_init' returns void, but our
 229                  * wrapper can fail to lookup the original symbol.
 230                  */
 231                 if (wrapper_disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0) < 0) {
 232                         ret = -ENOSYS;
 233                         goto iter_exit;
 234                 }
 235
 236                 while ((part = wrapper_disk_part_iter_next(&piter))) {
 237                         char name_buf[BDEVNAME_SIZE];
 238
 239                         if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 240                                 wrapper_disk_part_iter_exit(&piter);
 241                                 ret = -ENOSYS;
 242                                 goto iter_exit;
 243                         }
 244                         trace_lttng_statedump_block_device(session,
 245                                         lttng_get_part_devt(part), name_buf);
 246                 }
 247                 wrapper_disk_part_iter_exit(&piter);
 248         }
 249 iter_exit:
 250         class_dev_iter_exit(&iter);
 251 end:
 252         return ret;
 253 }
 254
 255 #ifdef CONFIG_INET
 256
 257 static
 258 void lttng_enumerate_device(struct lttng_session *session,
 259                 struct net_device *dev)
 260 {
 261         struct in_device *in_dev;
 262         struct in_ifaddr *ifa;
 263
 264         if (dev->flags & IFF_UP) {
 265                 in_dev = in_dev_get(dev);
 266                 if (in_dev) {
 267                         for (ifa = in_dev->ifa_list; ifa != NULL;
 268                              ifa = ifa->ifa_next) {
 269                                 trace_lttng_statedump_network_interface(
 270                                         session, dev, ifa);
 271                         }
 272                         in_dev_put(in_dev);
 273                 }
 274         } else {
 275                 trace_lttng_statedump_network_interface(
 276                         session, dev, NULL);
 277         }
 278 }
 279
 280 static
 281 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 282 {
 283         struct net_device *dev;
 284
 285         read_lock(&dev_base_lock);
 286         for_each_netdev(&init_net, dev)
 287                 lttng_enumerate_device(session, dev);
 288         read_unlock(&dev_base_lock);
 289
 290         return 0;
 291 }
 292 #else /* CONFIG_INET */
 293 static inline
 294 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 295 {
 296         return 0;
 297 }
 298 #endif /* CONFIG_INET */
 299
 300 static
 301 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 302 {
 303         const struct lttng_fd_ctx *ctx = p;
 304         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 305         unsigned int flags = file->f_flags;
 306         struct fdtable *fdt;
 307
 308         /*
 309          * We don't expose kernel internal flags, only userspace-visible
 310          * flags.
 311          */
 312         flags &= ~FMODE_NONOTIFY;
 313         fdt = files_fdtable(ctx->files);
 314         /*
 315          * We need to check here again whether fd is within the fdt
 316          * max_fds range, because we might be seeing a different
 317          * files_fdtable() than iterate_fd(), assuming only RCU is
 318          * protecting the read. In reality, iterate_fd() holds
 319          * file_lock, which should ensure the fdt does not change while
 320          * the lock is taken, but we are not aware whether this is
 321          * guaranteed or not, so play safe.
 322          */
 323         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 324                 flags |= O_CLOEXEC;
 325         if (IS_ERR(s)) {
 326                 struct dentry *dentry = file->f_path.dentry;
 327
 328                 /* Make sure we give at least some info */
 329                 spin_lock(&dentry->d_lock);
 330                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 331                         dentry->d_name.name, flags, file->f_mode);
 332                 spin_unlock(&dentry->d_lock);
 333                 goto end;
 334         }
 335         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 336                 flags, file->f_mode);
 337 end:
 338         return 0;
 339 }
 340
 341 static
 342 void lttng_enumerate_task_fd(struct lttng_session *session,
 343                 struct task_struct *p, char *tmp)
 344 {
 345         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 346         struct files_struct *files;
 347
 348         task_lock(p);
 349         files = p->files;
 350         if (!files)
 351                 goto end;
 352         ctx.files = files;
 353         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 354 end:
 355         task_unlock(p);
 356 }
 357
 358 static
 359 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 360 {
 361         struct task_struct *p;
 362         char *tmp;
 363
 364         tmp = (char *) __get_free_page(GFP_KERNEL);
 365         if (!tmp)
 366                 return -ENOMEM;
 367
 368         /* Enumerate active file descriptors */
 369         rcu_read_lock();
 370         for_each_process(p)
 371                 lttng_enumerate_task_fd(session, p, tmp);
 372         rcu_read_unlock();
 373         free_page((unsigned long) tmp);
 374         return 0;
 375 }
 376
 377 #if 0
 378 /*
 379  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 380  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 381  * iteration, but it is not exported to modules.
 382  */
 383 static
 384 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 385                 struct task_struct *p)
 386 {
 387         struct mm_struct *mm;
 388         struct vm_area_struct *map;
 389         unsigned long ino;
 390
 391         /* get_task_mm does a task_lock... */
 392         mm = get_task_mm(p);
 393         if (!mm)
 394                 return;
 395
 396         map = mm->mmap;
 397         if (map) {
 398                 down_read(&mm->mmap_sem);
 399                 while (map) {
 400                         if (map->vm_file)
 401                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 402                         else
 403                                 ino = 0;
 404                         trace_lttng_statedump_vm_map(session, p, map, ino);
 405                         map = map->vm_next;
 406                 }
 407                 up_read(&mm->mmap_sem);
 408         }
 409         mmput(mm);
 410 }
 411
 412 static
 413 int lttng_enumerate_vm_maps(struct lttng_session *session)
 414 {
 415         struct task_struct *p;
 416
 417         rcu_read_lock();
 418         for_each_process(p)
 419                 lttng_enumerate_task_vm_maps(session, p);
 420         rcu_read_unlock();
 421         return 0;
 422 }
 423 #endif
 424
 425 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 426
 427 #if (LTTNG_LINUX_VERSION_CODE < LTTNG_KERNEL_VERSION(2,6,39))
 428 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 429 #endif
 430
 431 static
 432 int lttng_list_interrupts(struct lttng_session *session)
 433 {
 434         unsigned int irq;
 435         unsigned long flags = 0;
 436         struct irq_desc *desc;
 437
 438 #define irq_to_desc     wrapper_irq_to_desc
 439         /* needs irq_desc */
 440         for_each_irq_desc(irq, desc) {
 441                 struct irqaction *action;
 442                 const char *irq_chip_name =
 443                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 444
 445                 local_irq_save(flags);
 446                 wrapper_desc_spin_lock(&desc->lock);
 447                 for (action = desc->action; action; action = action->next) {
 448                         trace_lttng_statedump_interrupt(session,
 449                                 irq, irq_chip_name, action);
 450                 }
 451                 wrapper_desc_spin_unlock(&desc->lock);
 452                 local_irq_restore(flags);
 453         }
 454         return 0;
 455 #undef irq_to_desc
 456 }
 457 #else
 458 static inline
 459 int lttng_list_interrupts(struct lttng_session *session)
 460 {
 461         return 0;
 462 }
 463 #endif
 464
 465 /*
 466  * Called with task lock held.
 467  */
 468 static
 469 void lttng_statedump_process_ns(struct lttng_session *session,
 470                 struct task_struct *p,
 471                 enum lttng_thread_type type,
 472                 enum lttng_execution_mode mode,
 473                 enum lttng_execution_submode submode,
 474                 enum lttng_process_status status)
 475 {
 476         struct pid_namespace *pid_ns;
 477
 478         pid_ns = task_active_pid_ns(p);
 479         do {
 480                 trace_lttng_statedump_process_state(session,
 481                         p, type, mode, submode, status, pid_ns);
 482                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 483         } while (pid_ns);
 484 }
 485
 486 static
 487 int lttng_enumerate_process_states(struct lttng_session *session)
 488 {
 489         struct task_struct *g, *p;
 490
 491         rcu_read_lock();
 492         for_each_process(g) {
 493                 p = g;
 494                 do {
 495                         enum lttng_execution_mode mode =
 496                                 LTTNG_MODE_UNKNOWN;
 497                         enum lttng_execution_submode submode =
 498                                 LTTNG_UNKNOWN;
 499                         enum lttng_process_status status;
 500                         enum lttng_thread_type type;
 501
 502                         task_lock(p);
 503                         if (p->exit_state == EXIT_ZOMBIE)
 504                                 status = LTTNG_ZOMBIE;
 505                         else if (p->exit_state == EXIT_DEAD)
 506                                 status = LTTNG_DEAD;
 507                         else if (p->state == TASK_RUNNING) {
 508                                 /* Is this a forked child that has not run yet? */
 509                                 if (list_empty(&p->rt.run_list))
 510                                         status = LTTNG_WAIT_FORK;
 511                                 else
 512                                         /*
 513                                          * All tasks are considered as wait_cpu;
 514                                          * the viewer will sort out if the task
 515                                          * was really running at this time.
 516                                          */
 517                                         status = LTTNG_WAIT_CPU;
 518                         } else if (p->state &
 519                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 520                                 /* Task is waiting for something to complete */
 521                                 status = LTTNG_WAIT;
 522                         } else
 523                                 status = LTTNG_UNNAMED;
 524                         submode = LTTNG_NONE;
 525
 526                         /*
 527                          * Verification of t->mm is to filter out kernel
 528                          * threads; Viewer will further filter out if a
 529                          * user-space thread was in syscall mode or not.
 530                          */
 531                         if (p->mm)
 532                                 type = LTTNG_USER_THREAD;
 533                         else
 534                                 type = LTTNG_KERNEL_THREAD;
 535                         lttng_statedump_process_ns(session,
 536                                 p, type, mode, submode, status);
 537                         task_unlock(p);
 538                 } while_each_thread(g, p);
 539         }
 540         rcu_read_unlock();
 541
 542         return 0;
 543 }
 544
 545 static
 546 void lttng_statedump_work_func(struct work_struct *work)
 547 {
 548         if (atomic_dec_and_test(&kernel_threads_to_run))
 549                 /* If we are the last thread, wake up do_lttng_statedump */
 550                 wake_up(&statedump_wq);
 551 }
 552
 553 static
 554 int do_lttng_statedump(struct lttng_session *session)
 555 {
 556         int cpu, ret;
 557
 558         trace_lttng_statedump_start(session);
 559         ret = lttng_enumerate_process_states(session);
 560         if (ret)
 561                 return ret;
 562         ret = lttng_enumerate_file_descriptors(session);
 563         if (ret)
 564                 return ret;
 565         /*
 566          * FIXME
 567          * ret = lttng_enumerate_vm_maps(session);
 568          * if (ret)
 569          *      return ret;
 570          */
 571         ret = lttng_list_interrupts(session);
 572         if (ret)
 573                 return ret;
 574         ret = lttng_enumerate_network_ip_interface(session);
 575         if (ret)
 576                 return ret;
 577         ret = lttng_enumerate_block_devices(session);
 578         switch (ret) {
 579         case 0:
 580                 break;
 581         case -ENOSYS:
 582                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 583                 break;
 584         default:
 585                 return ret;
 586         }
 587
 588         /* TODO lttng_dump_idt_table(session); */
 589         /* TODO lttng_dump_softirq_vec(session); */
 590         /* TODO lttng_list_modules(session); */
 591         /* TODO lttng_dump_swap_files(session); */
 592
 593         /*
 594          * Fire off a work queue on each CPU. Their sole purpose in life
 595          * is to guarantee that each CPU has been in a state where is was in
 596          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 597          */
 598         get_online_cpus();
 599         atomic_set(&kernel_threads_to_run, num_online_cpus());
 600         for_each_online_cpu(cpu) {
 601                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 602                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 603         }
 604         /* Wait for all threads to run */
 605         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 606         put_online_cpus();
 607         /* Our work is done */
 608         trace_lttng_statedump_end(session);
 609         return 0;
 610 }
 611
 612 /*
 613  * Called with session mutex held.
 614  */
 615 int lttng_statedump_start(struct lttng_session *session)
 616 {
 617         return do_lttng_statedump(session);
 618 }
 619 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 620
 621 static
 622 int __init lttng_statedump_init(void)
 623 {
 624         /*
 625          * Allow module to load even if the fixup cannot be done. This
 626          * will allow seemless transition when the underlying issue fix
 627          * is merged into the Linux kernel, and when tracepoint.c
 628          * "tracepoint_module_notify" is turned into a static function.
 629          */
 630         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 631         return 0;
 632 }
 633
 634 module_init(lttng_statedump_init);
 635
 636 static
 637 void __exit lttng_statedump_exit(void)
 638 {
 639 }
 640
 641 module_exit(lttng_statedump_exit);
 642
 643 MODULE_LICENSE("GPL and additional rights");
 644 MODULE_AUTHOR("Jean-Hugues Deschenes");
 645 MODULE_DESCRIPTION("LTTng statedump provider");
 646 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 647         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 648         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 649         LTTNG_MODULES_EXTRAVERSION);