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