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