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