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