src/lttng-statedump-impl.c

   1 /* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
   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/fdtable.h>
  40 #include <wrapper/namespace.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
  54 #define TRACE_INCLUDE_FILE lttng-statedump
  55 #define LTTNG_INSTRUMENTATION
  56 #include <instrumentation/events/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 files_struct *files,
  76                 int fd, const char *filename,
  77                 unsigned int flags, fmode_t fmode),
  78         TP_ARGS(session, files, fd, filename, flags, fmode));
  79
  80 LTTNG_DEFINE_TRACE(lttng_statedump_start,
  81         TP_PROTO(struct lttng_session *session),
  82         TP_ARGS(session));
  83
  84 LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
  85         TP_PROTO(struct lttng_session *session,
  86                 struct task_struct *p,
  87                 int type, int mode, int submode, int status,
  88                 struct files_struct *files),
  89         TP_ARGS(session, p, type, mode, submode, status, files));
  90
  91 LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns,
  92         TP_PROTO(struct lttng_session *session,
  93                 struct task_struct *p,
  94                 struct pid_namespace *pid_ns),
  95         TP_ARGS(session, p, pid_ns));
  96
  97 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
  98 LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns,
  99         TP_PROTO(struct lttng_session *session,
 100                 struct task_struct *p,
 101                 struct cgroup_namespace *cgroup_ns),
 102         TP_ARGS(session, p, cgroup_ns));
 103 #endif
 104
 105 LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns,
 106         TP_PROTO(struct lttng_session *session,
 107                 struct task_struct *p,
 108                 struct ipc_namespace *ipc_ns),
 109         TP_ARGS(session, p, ipc_ns));
 110
 111 #ifndef LTTNG_MNT_NS_MISSING_HEADER
 112 LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns,
 113         TP_PROTO(struct lttng_session *session,
 114                 struct task_struct *p,
 115                 struct mnt_namespace *mnt_ns),
 116         TP_ARGS(session, p, mnt_ns));
 117 #endif
 118
 119 LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns,
 120         TP_PROTO(struct lttng_session *session,
 121                 struct task_struct *p,
 122                 struct net *net_ns),
 123         TP_ARGS(session, p, net_ns));
 124
 125 LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns,
 126         TP_PROTO(struct lttng_session *session,
 127                 struct task_struct *p,
 128                 struct user_namespace *user_ns),
 129         TP_ARGS(session, p, user_ns));
 130
 131 LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns,
 132         TP_PROTO(struct lttng_session *session,
 133                 struct task_struct *p,
 134                 struct uts_namespace *uts_ns),
 135         TP_ARGS(session, p, uts_ns));
 136
 137 LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns,
 138         TP_PROTO(struct lttng_session *session,
 139                 struct task_struct *p,
 140                 struct time_namespace *time_ns),
 141         TP_ARGS(session, p, time_ns));
 142
 143 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
 144         TP_PROTO(struct lttng_session *session,
 145                 struct net_device *dev, struct in_ifaddr *ifa),
 146         TP_ARGS(session, dev, ifa));
 147
 148 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 149 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology,
 150         TP_PROTO(struct lttng_session *session, struct cpuinfo_x86 *c),
 151         TP_ARGS(session, c));
 152 #endif
 153
 154 struct lttng_fd_ctx {
 155         char *page;
 156         struct lttng_session *session;
 157         struct files_struct *files;
 158 };
 159
 160 /*
 161  * Protected by the trace lock.
 162  */
 163 static struct delayed_work cpu_work[NR_CPUS];
 164 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 165 static atomic_t kernel_threads_to_run;
 166
 167 enum lttng_thread_type {
 168         LTTNG_USER_THREAD = 0,
 169         LTTNG_KERNEL_THREAD = 1,
 170 };
 171
 172 enum lttng_execution_mode {
 173         LTTNG_USER_MODE = 0,
 174         LTTNG_SYSCALL = 1,
 175         LTTNG_TRAP = 2,
 176         LTTNG_IRQ = 3,
 177         LTTNG_SOFTIRQ = 4,
 178         LTTNG_MODE_UNKNOWN = 5,
 179 };
 180
 181 enum lttng_execution_submode {
 182         LTTNG_NONE = 0,
 183         LTTNG_UNKNOWN = 1,
 184 };
 185
 186 enum lttng_process_status {
 187         LTTNG_UNNAMED = 0,
 188         LTTNG_WAIT_FORK = 1,
 189         LTTNG_WAIT_CPU = 2,
 190         LTTNG_EXIT = 3,
 191         LTTNG_ZOMBIE = 4,
 192         LTTNG_WAIT = 5,
 193         LTTNG_RUN = 6,
 194         LTTNG_DEAD = 7,
 195 };
 196
 197
 198 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 199
 200 #define LTTNG_PART_STRUCT_TYPE struct block_device
 201
 202 static
 203 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 204 {
 205         const char *p;
 206
 207         p = bdevname(part, name_buf);
 208         if (!p)
 209                 return -ENOSYS;
 210
 211         return 0;
 212 }
 213
 214 static
 215 dev_t lttng_get_part_devt(struct block_device *part)
 216 {
 217         return part->bd_dev;
 218 }
 219
 220 #else
 221
 222 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
 223
 224 static
 225 int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf)
 226 {
 227         const char *p;
 228         struct block_device bdev;
 229
 230         /*
 231          * Create a partial 'struct blockdevice' to use
 232          * 'bdevname()' which is a simple wrapper over
 233          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 234          */
 235         bdev.bd_disk = disk;
 236         bdev.bd_part = part;
 237
 238         p = bdevname(&bdev, name_buf);
 239         if (!p)
 240                 return -ENOSYS;
 241
 242         return 0;
 243 }
 244
 245 static
 246 dev_t lttng_get_part_devt(struct hd_struct *part)
 247 {
 248         return part_devt(part);
 249 }
 250 #endif
 251
 252 static
 253 int lttng_enumerate_block_devices(struct lttng_session *session)
 254 {
 255         struct class *ptr_block_class;
 256         struct device_type *ptr_disk_type;
 257         struct class_dev_iter iter;
 258         struct device *dev;
 259
 260         ptr_block_class = wrapper_get_block_class();
 261         if (!ptr_block_class)
 262                 return -ENOSYS;
 263         ptr_disk_type = wrapper_get_disk_type();
 264         if (!ptr_disk_type) {
 265                 return -ENOSYS;
 266         }
 267         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 268         while ((dev = class_dev_iter_next(&iter))) {
 269                 struct disk_part_iter piter;
 270                 struct gendisk *disk = dev_to_disk(dev);
 271                 LTTNG_PART_STRUCT_TYPE *part;
 272
 273                 /*
 274                  * Don't show empty devices or things that have been
 275                  * suppressed
 276                  */
 277                 if (get_capacity(disk) == 0 ||
 278                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 279                         continue;
 280
 281                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 282                 while ((part = disk_part_iter_next(&piter))) {
 283                         char name_buf[BDEVNAME_SIZE];
 284
 285                         if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 286                                 disk_part_iter_exit(&piter);
 287                                 class_dev_iter_exit(&iter);
 288                                 return -ENOSYS;
 289                         }
 290                         trace_lttng_statedump_block_device(session,
 291                                         lttng_get_part_devt(part), name_buf);
 292                 }
 293                 disk_part_iter_exit(&piter);
 294         }
 295         class_dev_iter_exit(&iter);
 296         return 0;
 297 }
 298
 299 #ifdef CONFIG_INET
 300
 301 static
 302 void lttng_enumerate_device(struct lttng_session *session,
 303                 struct net_device *dev)
 304 {
 305         struct in_device *in_dev;
 306         struct in_ifaddr *ifa;
 307
 308         if (dev->flags & IFF_UP) {
 309                 in_dev = in_dev_get(dev);
 310                 if (in_dev) {
 311                         for (ifa = in_dev->ifa_list; ifa != NULL;
 312                              ifa = ifa->ifa_next) {
 313                                 trace_lttng_statedump_network_interface(
 314                                         session, dev, ifa);
 315                         }
 316                         in_dev_put(in_dev);
 317                 }
 318         } else {
 319                 trace_lttng_statedump_network_interface(
 320                         session, dev, NULL);
 321         }
 322 }
 323
 324 static
 325 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 326 {
 327         struct net_device *dev;
 328
 329         read_lock(&dev_base_lock);
 330         for_each_netdev(&init_net, dev)
 331                 lttng_enumerate_device(session, dev);
 332         read_unlock(&dev_base_lock);
 333
 334         return 0;
 335 }
 336 #else /* CONFIG_INET */
 337 static inline
 338 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 339 {
 340         return 0;
 341 }
 342 #endif /* CONFIG_INET */
 343
 344 static
 345 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 346 {
 347         const struct lttng_fd_ctx *ctx = p;
 348         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 349         unsigned int flags = file->f_flags;
 350         struct fdtable *fdt;
 351
 352         /*
 353          * We don't expose kernel internal flags, only userspace-visible
 354          * flags.
 355          */
 356         flags &= ~FMODE_NONOTIFY;
 357         fdt = files_fdtable(ctx->files);
 358         /*
 359          * We need to check here again whether fd is within the fdt
 360          * max_fds range, because we might be seeing a different
 361          * files_fdtable() than iterate_fd(), assuming only RCU is
 362          * protecting the read. In reality, iterate_fd() holds
 363          * file_lock, which should ensure the fdt does not change while
 364          * the lock is taken, but we are not aware whether this is
 365          * guaranteed or not, so play safe.
 366          */
 367         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 368                 flags |= O_CLOEXEC;
 369         if (IS_ERR(s)) {
 370                 struct dentry *dentry = file->f_path.dentry;
 371
 372                 /* Make sure we give at least some info */
 373                 spin_lock(&dentry->d_lock);
 374                 trace_lttng_statedump_file_descriptor(ctx->session,
 375                         ctx->files, fd, dentry->d_name.name, flags,
 376                         file->f_mode);
 377                 spin_unlock(&dentry->d_lock);
 378                 goto end;
 379         }
 380         trace_lttng_statedump_file_descriptor(ctx->session,
 381                 ctx->files, fd, s, flags, file->f_mode);
 382 end:
 383         return 0;
 384 }
 385
 386 /* Called with task lock held. */
 387 static
 388 void lttng_enumerate_files(struct lttng_session *session,
 389                 struct files_struct *files,
 390                 char *tmp)
 391 {
 392         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
 393
 394         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 395 }
 396
 397 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 398 static
 399 int lttng_enumerate_cpu_topology(struct lttng_session *session)
 400 {
 401         int cpu;
 402         const cpumask_t *cpumask = cpu_possible_mask;
 403
 404         for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
 405                         cpu = cpumask_next(cpu, cpumask)) {
 406                 trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
 407         }
 408
 409         return 0;
 410 }
 411 #else
 412 static
 413 int lttng_enumerate_cpu_topology(struct lttng_session *session)
 414 {
 415         return 0;
 416 }
 417 #endif
 418
 419 #if 0
 420 /*
 421  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 422  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 423  * iteration, but it is not exported to modules.
 424  */
 425 static
 426 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 427                 struct task_struct *p)
 428 {
 429         struct mm_struct *mm;
 430         struct vm_area_struct *map;
 431         unsigned long ino;
 432
 433         /* get_task_mm does a task_lock... */
 434         mm = get_task_mm(p);
 435         if (!mm)
 436                 return;
 437
 438         map = mm->mmap;
 439         if (map) {
 440                 down_read(&mm->mmap_sem);
 441                 while (map) {
 442                         if (map->vm_file)
 443                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 444                         else
 445                                 ino = 0;
 446                         trace_lttng_statedump_vm_map(session, p, map, ino);
 447                         map = map->vm_next;
 448                 }
 449                 up_read(&mm->mmap_sem);
 450         }
 451         mmput(mm);
 452 }
 453
 454 static
 455 int lttng_enumerate_vm_maps(struct lttng_session *session)
 456 {
 457         struct task_struct *p;
 458
 459         rcu_read_lock();
 460         for_each_process(p)
 461                 lttng_enumerate_task_vm_maps(session, p);
 462         rcu_read_unlock();
 463         return 0;
 464 }
 465 #endif
 466
 467 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 468
 469 static
 470 int lttng_list_interrupts(struct lttng_session *session)
 471 {
 472         unsigned int irq;
 473         unsigned long flags = 0;
 474         struct irq_desc *desc;
 475
 476 #define irq_to_desc     wrapper_irq_to_desc
 477         /* needs irq_desc */
 478         for_each_irq_desc(irq, desc) {
 479                 struct irqaction *action;
 480                 const char *irq_chip_name =
 481                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 482
 483                 local_irq_save(flags);
 484                 raw_spin_lock(&desc->lock);
 485                 for (action = desc->action; action; action = action->next) {
 486                         trace_lttng_statedump_interrupt(session,
 487                                 irq, irq_chip_name, action);
 488                 }
 489                 raw_spin_unlock(&desc->lock);
 490                 local_irq_restore(flags);
 491         }
 492         return 0;
 493 #undef irq_to_desc
 494 }
 495 #else
 496 static inline
 497 int lttng_list_interrupts(struct lttng_session *session)
 498 {
 499         return 0;
 500 }
 501 #endif
 502
 503 /*
 504  * Statedump the task's namespaces using the proc filesystem inode number as
 505  * the unique identifier. The user and pid ns are nested and will be dumped
 506  * recursively.
 507  *
 508  * Called with task lock held.
 509  */
 510 static
 511 void lttng_statedump_process_ns(struct lttng_session *session,
 512                 struct task_struct *p,
 513                 enum lttng_thread_type type,
 514                 enum lttng_execution_mode mode,
 515                 enum lttng_execution_submode submode,
 516                 enum lttng_process_status status)
 517 {
 518         struct nsproxy *proxy;
 519         struct pid_namespace *pid_ns;
 520         struct user_namespace *user_ns;
 521
 522         /*
 523          * The pid and user namespaces are special, they are nested and
 524          * accessed with specific functions instead of the nsproxy struct
 525          * like the other namespaces.
 526          */
 527         pid_ns = task_active_pid_ns(p);
 528         do {
 529                 trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
 530                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 531         } while (pid_ns);
 532
 533
 534         user_ns = task_cred_xxx(p, user_ns);
 535         do {
 536                 trace_lttng_statedump_process_user_ns(session, p, user_ns);
 537                 /*
 538                  * trace_lttng_statedump_process_user_ns() internally
 539                  * checks whether user_ns is NULL. While this does not
 540                  * appear to be a possible return value for
 541                  * task_cred_xxx(), err on the safe side and check
 542                  * for NULL here as well to be consistent with the
 543                  * paranoid behavior of
 544                  * trace_lttng_statedump_process_user_ns().
 545                  */
 546                 user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
 547         } while (user_ns);
 548
 549         /*
 550          * Back and forth on locking strategy within Linux upstream for nsproxy.
 551          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 552          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 553          * for details.
 554          */
 555 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 556                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 557                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 558                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 559         proxy = p->nsproxy;
 560 #else
 561         rcu_read_lock();
 562         proxy = task_nsproxy(p);
 563 #endif
 564         if (proxy) {
 565 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
 566                 trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
 567 #endif
 568                 trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
 569 #ifndef LTTNG_MNT_NS_MISSING_HEADER
 570                 trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
 571 #endif
 572                 trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
 573                 trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
 574 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0))
 575                 trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns);
 576 #endif
 577         }
 578 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 579                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 580                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 581                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 582         /* (nothing) */
 583 #else
 584         rcu_read_unlock();
 585 #endif
 586 }
 587
 588 static
 589 int lttng_enumerate_process_states(struct lttng_session *session)
 590 {
 591         struct task_struct *g, *p;
 592         char *tmp;
 593
 594         tmp = (char *) __get_free_page(GFP_KERNEL);
 595         if (!tmp)
 596                 return -ENOMEM;
 597
 598         rcu_read_lock();
 599         for_each_process(g) {
 600                 struct files_struct *prev_files = NULL;
 601
 602                 p = g;
 603                 do {
 604                         enum lttng_execution_mode mode =
 605                                 LTTNG_MODE_UNKNOWN;
 606                         enum lttng_execution_submode submode =
 607                                 LTTNG_UNKNOWN;
 608                         enum lttng_process_status status;
 609                         enum lttng_thread_type type;
 610                         struct files_struct *files;
 611
 612                         task_lock(p);
 613                         if (p->exit_state == EXIT_ZOMBIE)
 614                                 status = LTTNG_ZOMBIE;
 615                         else if (p->exit_state == EXIT_DEAD)
 616                                 status = LTTNG_DEAD;
 617                         else if (p->state == TASK_RUNNING) {
 618                                 /* Is this a forked child that has not run yet? */
 619                                 if (list_empty(&p->rt.run_list))
 620                                         status = LTTNG_WAIT_FORK;
 621                                 else
 622                                         /*
 623                                          * All tasks are considered as wait_cpu;
 624                                          * the viewer will sort out if the task
 625                                          * was really running at this time.
 626                                          */
 627                                         status = LTTNG_WAIT_CPU;
 628                         } else if (p->state &
 629                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 630                                 /* Task is waiting for something to complete */
 631                                 status = LTTNG_WAIT;
 632                         } else
 633                                 status = LTTNG_UNNAMED;
 634                         submode = LTTNG_NONE;
 635
 636                         /*
 637                          * Verification of t->mm is to filter out kernel
 638                          * threads; Viewer will further filter out if a
 639                          * user-space thread was in syscall mode or not.
 640                          */
 641                         if (p->mm)
 642                                 type = LTTNG_USER_THREAD;
 643                         else
 644                                 type = LTTNG_KERNEL_THREAD;
 645                         files = p->files;
 646
 647                         trace_lttng_statedump_process_state(session,
 648                                 p, type, mode, submode, status, files);
 649                         lttng_statedump_process_ns(session,
 650                                 p, type, mode, submode, status);
 651                         /*
 652                          * As an optimisation for the common case, do not
 653                          * repeat information for the same files_struct in
 654                          * two consecutive threads. This is the common case
 655                          * for threads sharing the same fd table. RCU guarantees
 656                          * that the same files_struct pointer is not re-used
 657                          * throughout processes/threads iteration.
 658                          */
 659                         if (files && files != prev_files) {
 660                                 lttng_enumerate_files(session, files, tmp);
 661                                 prev_files = files;
 662                         }
 663                         task_unlock(p);
 664                 } while_each_thread(g, p);
 665         }
 666         rcu_read_unlock();
 667
 668         free_page((unsigned long) tmp);
 669
 670         return 0;
 671 }
 672
 673 static
 674 void lttng_statedump_work_func(struct work_struct *work)
 675 {
 676         if (atomic_dec_and_test(&kernel_threads_to_run))
 677                 /* If we are the last thread, wake up do_lttng_statedump */
 678                 wake_up(&statedump_wq);
 679 }
 680
 681 static
 682 int do_lttng_statedump(struct lttng_session *session)
 683 {
 684         int cpu, ret;
 685
 686         trace_lttng_statedump_start(session);
 687         ret = lttng_enumerate_process_states(session);
 688         if (ret)
 689                 return ret;
 690         /*
 691          * FIXME
 692          * ret = lttng_enumerate_vm_maps(session);
 693          * if (ret)
 694          *      return ret;
 695          */
 696         ret = lttng_list_interrupts(session);
 697         if (ret)
 698                 return ret;
 699         ret = lttng_enumerate_network_ip_interface(session);
 700         if (ret)
 701                 return ret;
 702         ret = lttng_enumerate_block_devices(session);
 703         switch (ret) {
 704         case 0:
 705                 break;
 706         case -ENOSYS:
 707                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 708                 break;
 709         default:
 710                 return ret;
 711         }
 712         ret = lttng_enumerate_cpu_topology(session);
 713         if (ret)
 714                 return ret;
 715
 716         /* TODO lttng_dump_idt_table(session); */
 717         /* TODO lttng_dump_softirq_vec(session); */
 718         /* TODO lttng_list_modules(session); */
 719         /* TODO lttng_dump_swap_files(session); */
 720
 721         /*
 722          * Fire off a work queue on each CPU. Their sole purpose in life
 723          * is to guarantee that each CPU has been in a state where is was in
 724          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 725          */
 726         get_online_cpus();
 727         atomic_set(&kernel_threads_to_run, num_online_cpus());
 728         for_each_online_cpu(cpu) {
 729                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 730                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 731         }
 732         /* Wait for all threads to run */
 733         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 734         put_online_cpus();
 735         /* Our work is done */
 736         trace_lttng_statedump_end(session);
 737         return 0;
 738 }
 739
 740 /*
 741  * Called with session mutex held.
 742  */
 743 int lttng_statedump_start(struct lttng_session *session)
 744 {
 745         return do_lttng_statedump(session);
 746 }
 747 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 748
 749 static
 750 int __init lttng_statedump_init(void)
 751 {
 752         /*
 753          * Allow module to load even if the fixup cannot be done. This
 754          * will allow seemless transition when the underlying issue fix
 755          * is merged into the Linux kernel, and when tracepoint.c
 756          * "tracepoint_module_notify" is turned into a static function.
 757          */
 758         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 759         return 0;
 760 }
 761
 762 module_init(lttng_statedump_init);
 763
 764 static
 765 void __exit lttng_statedump_exit(void)
 766 {
 767 }
 768
 769 module_exit(lttng_statedump_exit);
 770
 771 MODULE_LICENSE("GPL and additional rights");
 772 MODULE_AUTHOR("Jean-Hugues Deschenes");
 773 MODULE_DESCRIPTION("LTTng statedump provider");
 774 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 775         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 776         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 777         LTTNG_MODULES_EXTRAVERSION);