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