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