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 LTTNG_DEFINE_TRACE(lttng_statedump_block_device,
  59         TP_PROTO(struct lttng_session *session,
  60                 dev_t dev, const char *diskname),
  61         TP_ARGS(session, dev, diskname));
  62
  63 LTTNG_DEFINE_TRACE(lttng_statedump_end,
  64         TP_PROTO(struct lttng_session *session),
  65         TP_ARGS(session));
  66
  67 LTTNG_DEFINE_TRACE(lttng_statedump_interrupt,
  68         TP_PROTO(struct lttng_session *session,
  69                 unsigned int irq, const char *chip_name,
  70                 struct irqaction *action),
  71         TP_ARGS(session, irq, chip_name, action));
  72
  73 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor,
  74         TP_PROTO(struct lttng_session *session,
  75                 struct task_struct *p, int fd, const char *filename,
  76                 unsigned int flags, fmode_t fmode),
  77         TP_ARGS(session, p, fd, filename, flags, fmode));
  78
  79 LTTNG_DEFINE_TRACE(lttng_statedump_start,
  80         TP_PROTO(struct lttng_session *session),
  81         TP_ARGS(session));
  82
  83 LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
  84         TP_PROTO(struct lttng_session *session,
  85                 struct task_struct *p,
  86                 int type, int mode, int submode, int status,
  87                 struct pid_namespace *pid_ns),
  88         TP_ARGS(session, p, type, mode, submode, status, pid_ns));
  89
  90 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
  91         TP_PROTO(struct lttng_session *session,
  92                 struct net_device *dev, struct in_ifaddr *ifa),
  93         TP_ARGS(session, dev, ifa));
  94
  95 struct lttng_fd_ctx {
  96         char *page;
  97         struct lttng_session *session;
  98         struct task_struct *p;
  99         struct files_struct *files;
 100 };
 101
 102 /*
 103  * Protected by the trace lock.
 104  */
 105 static struct delayed_work cpu_work[NR_CPUS];
 106 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 107 static atomic_t kernel_threads_to_run;
 108
 109 enum lttng_thread_type {
 110         LTTNG_USER_THREAD = 0,
 111         LTTNG_KERNEL_THREAD = 1,
 112 };
 113
 114 enum lttng_execution_mode {
 115         LTTNG_USER_MODE = 0,
 116         LTTNG_SYSCALL = 1,
 117         LTTNG_TRAP = 2,
 118         LTTNG_IRQ = 3,
 119         LTTNG_SOFTIRQ = 4,
 120         LTTNG_MODE_UNKNOWN = 5,
 121 };
 122
 123 enum lttng_execution_submode {
 124         LTTNG_NONE = 0,
 125         LTTNG_UNKNOWN = 1,
 126 };
 127
 128 enum lttng_process_status {
 129         LTTNG_UNNAMED = 0,
 130         LTTNG_WAIT_FORK = 1,
 131         LTTNG_WAIT_CPU = 2,
 132         LTTNG_EXIT = 3,
 133         LTTNG_ZOMBIE = 4,
 134         LTTNG_WAIT = 5,
 135         LTTNG_RUN = 6,
 136         LTTNG_DEAD = 7,
 137 };
 138
 139
 140 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 141
 142 #define LTTNG_PART_STRUCT_TYPE struct block_device
 143
 144 static
 145 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 146 {
 147         const char *p;
 148
 149         p = bdevname(part, name_buf);
 150         if (!p)
 151                 return -ENOSYS;
 152
 153         return 0;
 154 }
 155
 156 static
 157 dev_t lttng_get_part_devt(struct block_device *part)
 158 {
 159         return part->bd_dev;
 160 }
 161
 162 #else
 163
 164 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
 165
 166 static
 167 int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf)
 168 {
 169         const char *p;
 170         struct block_device bdev;
 171
 172         /*
 173          * Create a partial 'struct blockdevice' to use
 174          * 'bdevname()' which is a simple wrapper over
 175          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 176          */
 177         bdev.bd_disk = disk;
 178         bdev.bd_part = part;
 179
 180         p = bdevname(&bdev, name_buf);
 181         if (!p)
 182                 return -ENOSYS;
 183
 184         return 0;
 185 }
 186
 187 static
 188 dev_t lttng_get_part_devt(struct hd_struct *part)
 189 {
 190         return part_devt(part);
 191 }
 192 #endif
 193
 194 static
 195 int lttng_enumerate_block_devices(struct lttng_session *session)
 196 {
 197         struct class *ptr_block_class;
 198         struct device_type *ptr_disk_type;
 199         struct class_dev_iter iter;
 200         struct device *dev;
 201
 202         ptr_block_class = wrapper_get_block_class();
 203         if (!ptr_block_class)
 204                 return -ENOSYS;
 205         ptr_disk_type = wrapper_get_disk_type();
 206         if (!ptr_disk_type) {
 207                 return -ENOSYS;
 208         }
 209         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 210         while ((dev = class_dev_iter_next(&iter))) {
 211                 struct disk_part_iter piter;
 212                 struct gendisk *disk = dev_to_disk(dev);
 213                 LTTNG_PART_STRUCT_TYPE *part;
 214
 215                 /*
 216                  * Don't show empty devices or things that have been
 217                  * suppressed
 218                  */
 219                 if (get_capacity(disk) == 0 ||
 220                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 221                         continue;
 222
 223                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 224                 while ((part = disk_part_iter_next(&piter))) {
 225                         char name_buf[BDEVNAME_SIZE];
 226
 227                         if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 228                                 disk_part_iter_exit(&piter);
 229                                 class_dev_iter_exit(&iter);
 230                                 return -ENOSYS;
 231                         }
 232                         trace_lttng_statedump_block_device(session,
 233                                         lttng_get_part_devt(part), name_buf);
 234                 }
 235                 disk_part_iter_exit(&piter);
 236         }
 237         class_dev_iter_exit(&iter);
 238         return 0;
 239 }
 240
 241 #ifdef CONFIG_INET
 242
 243 static
 244 void lttng_enumerate_device(struct lttng_session *session,
 245                 struct net_device *dev)
 246 {
 247         struct in_device *in_dev;
 248         struct in_ifaddr *ifa;
 249
 250         if (dev->flags & IFF_UP) {
 251                 in_dev = in_dev_get(dev);
 252                 if (in_dev) {
 253                         for (ifa = in_dev->ifa_list; ifa != NULL;
 254                              ifa = ifa->ifa_next) {
 255                                 trace_lttng_statedump_network_interface(
 256                                         session, dev, ifa);
 257                         }
 258                         in_dev_put(in_dev);
 259                 }
 260         } else {
 261                 trace_lttng_statedump_network_interface(
 262                         session, dev, NULL);
 263         }
 264 }
 265
 266 static
 267 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 268 {
 269         struct net_device *dev;
 270
 271         read_lock(&dev_base_lock);
 272         for_each_netdev(&init_net, dev)
 273                 lttng_enumerate_device(session, dev);
 274         read_unlock(&dev_base_lock);
 275
 276         return 0;
 277 }
 278 #else /* CONFIG_INET */
 279 static inline
 280 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 281 {
 282         return 0;
 283 }
 284 #endif /* CONFIG_INET */
 285
 286 static
 287 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 288 {
 289         const struct lttng_fd_ctx *ctx = p;
 290         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 291         unsigned int flags = file->f_flags;
 292         struct fdtable *fdt;
 293
 294         /*
 295          * We don't expose kernel internal flags, only userspace-visible
 296          * flags.
 297          */
 298         flags &= ~FMODE_NONOTIFY;
 299         fdt = files_fdtable(ctx->files);
 300         /*
 301          * We need to check here again whether fd is within the fdt
 302          * max_fds range, because we might be seeing a different
 303          * files_fdtable() than iterate_fd(), assuming only RCU is
 304          * protecting the read. In reality, iterate_fd() holds
 305          * file_lock, which should ensure the fdt does not change while
 306          * the lock is taken, but we are not aware whether this is
 307          * guaranteed or not, so play safe.
 308          */
 309         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 310                 flags |= O_CLOEXEC;
 311         if (IS_ERR(s)) {
 312                 struct dentry *dentry = file->f_path.dentry;
 313
 314                 /* Make sure we give at least some info */
 315                 spin_lock(&dentry->d_lock);
 316                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 317                         dentry->d_name.name, flags, file->f_mode);
 318                 spin_unlock(&dentry->d_lock);
 319                 goto end;
 320         }
 321         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 322                 flags, file->f_mode);
 323 end:
 324         return 0;
 325 }
 326
 327 static
 328 void lttng_enumerate_task_fd(struct lttng_session *session,
 329                 struct task_struct *p, char *tmp)
 330 {
 331         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 332         struct files_struct *files;
 333
 334         task_lock(p);
 335         files = p->files;
 336         if (!files)
 337                 goto end;
 338         ctx.files = files;
 339         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 340 end:
 341         task_unlock(p);
 342 }
 343
 344 static
 345 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 346 {
 347         struct task_struct *p;
 348         char *tmp;
 349
 350         tmp = (char *) __get_free_page(GFP_KERNEL);
 351         if (!tmp)
 352                 return -ENOMEM;
 353
 354         /* Enumerate active file descriptors */
 355         rcu_read_lock();
 356         for_each_process(p)
 357                 lttng_enumerate_task_fd(session, p, tmp);
 358         rcu_read_unlock();
 359         free_page((unsigned long) tmp);
 360         return 0;
 361 }
 362
 363 #if 0
 364 /*
 365  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 366  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 367  * iteration, but it is not exported to modules.
 368  */
 369 static
 370 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 371                 struct task_struct *p)
 372 {
 373         struct mm_struct *mm;
 374         struct vm_area_struct *map;
 375         unsigned long ino;
 376
 377         /* get_task_mm does a task_lock... */
 378         mm = get_task_mm(p);
 379         if (!mm)
 380                 return;
 381
 382         map = mm->mmap;
 383         if (map) {
 384                 down_read(&mm->mmap_sem);
 385                 while (map) {
 386                         if (map->vm_file)
 387                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 388                         else
 389                                 ino = 0;
 390                         trace_lttng_statedump_vm_map(session, p, map, ino);
 391                         map = map->vm_next;
 392                 }
 393                 up_read(&mm->mmap_sem);
 394         }
 395         mmput(mm);
 396 }
 397
 398 static
 399 int lttng_enumerate_vm_maps(struct lttng_session *session)
 400 {
 401         struct task_struct *p;
 402
 403         rcu_read_lock();
 404         for_each_process(p)
 405                 lttng_enumerate_task_vm_maps(session, p);
 406         rcu_read_unlock();
 407         return 0;
 408 }
 409 #endif
 410
 411 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 412
 413 #if (LTTNG_LINUX_VERSION_CODE < LTTNG_KERNEL_VERSION(2,6,39))
 414 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 415 #endif
 416
 417 static
 418 int lttng_list_interrupts(struct lttng_session *session)
 419 {
 420         unsigned int irq;
 421         unsigned long flags = 0;
 422         struct irq_desc *desc;
 423
 424 #define irq_to_desc     wrapper_irq_to_desc
 425         /* needs irq_desc */
 426         for_each_irq_desc(irq, desc) {
 427                 struct irqaction *action;
 428                 const char *irq_chip_name =
 429                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 430
 431                 local_irq_save(flags);
 432                 wrapper_desc_spin_lock(&desc->lock);
 433                 for (action = desc->action; action; action = action->next) {
 434                         trace_lttng_statedump_interrupt(session,
 435                                 irq, irq_chip_name, action);
 436                 }
 437                 wrapper_desc_spin_unlock(&desc->lock);
 438                 local_irq_restore(flags);
 439         }
 440         return 0;
 441 #undef irq_to_desc
 442 }
 443 #else
 444 static inline
 445 int lttng_list_interrupts(struct lttng_session *session)
 446 {
 447         return 0;
 448 }
 449 #endif
 450
 451 /*
 452  * Called with task lock held.
 453  */
 454 static
 455 void lttng_statedump_process_ns(struct lttng_session *session,
 456                 struct task_struct *p,
 457                 enum lttng_thread_type type,
 458                 enum lttng_execution_mode mode,
 459                 enum lttng_execution_submode submode,
 460                 enum lttng_process_status status)
 461 {
 462         struct pid_namespace *pid_ns;
 463
 464         pid_ns = task_active_pid_ns(p);
 465         do {
 466                 trace_lttng_statedump_process_state(session,
 467                         p, type, mode, submode, status, pid_ns);
 468                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 469         } while (pid_ns);
 470 }
 471
 472 static
 473 int lttng_enumerate_process_states(struct lttng_session *session)
 474 {
 475         struct task_struct *g, *p;
 476
 477         rcu_read_lock();
 478         for_each_process(g) {
 479                 p = g;
 480                 do {
 481                         enum lttng_execution_mode mode =
 482                                 LTTNG_MODE_UNKNOWN;
 483                         enum lttng_execution_submode submode =
 484                                 LTTNG_UNKNOWN;
 485                         enum lttng_process_status status;
 486                         enum lttng_thread_type type;
 487
 488                         task_lock(p);
 489                         if (p->exit_state == EXIT_ZOMBIE)
 490                                 status = LTTNG_ZOMBIE;
 491                         else if (p->exit_state == EXIT_DEAD)
 492                                 status = LTTNG_DEAD;
 493                         else if (p->state == TASK_RUNNING) {
 494                                 /* Is this a forked child that has not run yet? */
 495                                 if (list_empty(&p->rt.run_list))
 496                                         status = LTTNG_WAIT_FORK;
 497                                 else
 498                                         /*
 499                                          * All tasks are considered as wait_cpu;
 500                                          * the viewer will sort out if the task
 501                                          * was really running at this time.
 502                                          */
 503                                         status = LTTNG_WAIT_CPU;
 504                         } else if (p->state &
 505                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 506                                 /* Task is waiting for something to complete */
 507                                 status = LTTNG_WAIT;
 508                         } else
 509                                 status = LTTNG_UNNAMED;
 510                         submode = LTTNG_NONE;
 511
 512                         /*
 513                          * Verification of t->mm is to filter out kernel
 514                          * threads; Viewer will further filter out if a
 515                          * user-space thread was in syscall mode or not.
 516                          */
 517                         if (p->mm)
 518                                 type = LTTNG_USER_THREAD;
 519                         else
 520                                 type = LTTNG_KERNEL_THREAD;
 521                         lttng_statedump_process_ns(session,
 522                                 p, type, mode, submode, status);
 523                         task_unlock(p);
 524                 } while_each_thread(g, p);
 525         }
 526         rcu_read_unlock();
 527
 528         return 0;
 529 }
 530
 531 static
 532 void lttng_statedump_work_func(struct work_struct *work)
 533 {
 534         if (atomic_dec_and_test(&kernel_threads_to_run))
 535                 /* If we are the last thread, wake up do_lttng_statedump */
 536                 wake_up(&statedump_wq);
 537 }
 538
 539 static
 540 int do_lttng_statedump(struct lttng_session *session)
 541 {
 542         int cpu, ret;
 543
 544         trace_lttng_statedump_start(session);
 545         ret = lttng_enumerate_process_states(session);
 546         if (ret)
 547                 return ret;
 548         ret = lttng_enumerate_file_descriptors(session);
 549         if (ret)
 550                 return ret;
 551         /*
 552          * FIXME
 553          * ret = lttng_enumerate_vm_maps(session);
 554          * if (ret)
 555          *      return ret;
 556          */
 557         ret = lttng_list_interrupts(session);
 558         if (ret)
 559                 return ret;
 560         ret = lttng_enumerate_network_ip_interface(session);
 561         if (ret)
 562                 return ret;
 563         ret = lttng_enumerate_block_devices(session);
 564         switch (ret) {
 565         case 0:
 566                 break;
 567         case -ENOSYS:
 568                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 569                 break;
 570         default:
 571                 return ret;
 572         }
 573
 574         /* TODO lttng_dump_idt_table(session); */
 575         /* TODO lttng_dump_softirq_vec(session); */
 576         /* TODO lttng_list_modules(session); */
 577         /* TODO lttng_dump_swap_files(session); */
 578
 579         /*
 580          * Fire off a work queue on each CPU. Their sole purpose in life
 581          * is to guarantee that each CPU has been in a state where is was in
 582          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 583          */
 584         get_online_cpus();
 585         atomic_set(&kernel_threads_to_run, num_online_cpus());
 586         for_each_online_cpu(cpu) {
 587                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 588                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 589         }
 590         /* Wait for all threads to run */
 591         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 592         put_online_cpus();
 593         /* Our work is done */
 594         trace_lttng_statedump_end(session);
 595         return 0;
 596 }
 597
 598 /*
 599  * Called with session mutex held.
 600  */
 601 int lttng_statedump_start(struct lttng_session *session)
 602 {
 603         return do_lttng_statedump(session);
 604 }
 605 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 606
 607 static
 608 int __init lttng_statedump_init(void)
 609 {
 610         /*
 611          * Allow module to load even if the fixup cannot be done. This
 612          * will allow seemless transition when the underlying issue fix
 613          * is merged into the Linux kernel, and when tracepoint.c
 614          * "tracepoint_module_notify" is turned into a static function.
 615          */
 616         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 617         return 0;
 618 }
 619
 620 module_init(lttng_statedump_init);
 621
 622 static
 623 void __exit lttng_statedump_exit(void)
 624 {
 625 }
 626
 627 module_exit(lttng_statedump_exit);
 628
 629 MODULE_LICENSE("GPL and additional rights");
 630 MODULE_AUTHOR("Jean-Hugues Deschenes");
 631 MODULE_DESCRIPTION("LTTng statedump provider");
 632 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 633         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 634         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 635         LTTNG_MODULES_EXTRAVERSION);